LNG Risk Based Safety: Modeling and Consequence Analysis

LNG RISK BASED SAFETY Modeling and Consequence Analysis

John L. Woodward and Robin M. Pitblado

®

A JOHN WILEY & SONS, INC., PUBLICATION

LNG RISK BASED SAFETY

LNG RISK BASED SAFETY Modeling and Consequence Analysis

John L. Woodward and Robin M. Pitblado

®

A JOHN WILEY & SONS, INC., PUBLICATION

Copyright © 2010 by John Wiley & Sons, Inc. All rights reserved Published by John Wiley & Sons, Inc., Hoboken, New Jersey Published simultaneously in Canada No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/ permission. Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002. Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com. Library of Congress Cataloging-in-Publication Data Woodward, John Lowell. LNG risk based safety : modeling and consequence analysis / John L Woodward, Robin Pitblado. p. cm. Includes index. ISBN 978-0-470-31764-8 (cloth) 1. Liquefied natural gas--Safety measures. 2. Flammable gases--Accidents--Risk assessment. 3. Flammable liquids–Accidents--Risk assesment. 4. Chemicals–Fires and fire prevention. 5. Chemical plants–Accidents–Simulation methods. I. Pitblado, Robin. II. Title. TH9446.I475W66 2010 665.7′730289–dc22 2009036487 Printed in the United States of America 10 9 8 7 6 5 4 3 2 1

We dedicate this book to our wives Gerri Woodward and Mary Lynes, who defined the meaning of support with their patience and assistance throughout.

CONTENTS Preface 1

LNG Properties and Overview of Hazards 1.1 1.2

1.3 1.4 1.5

1.6 1.7 2

xv

LNG Properties / 2 Hazards of LNG with Respect to Public Risk / 4 1.2.1 Flash Fire, Pool Fire, or Jet Fire / 7 1.2.2 Outdoor Vapor Cloud Explosions / 8 1.2.3 Enclosed Vapor Cloud Explosions / 9 1.2.4 Asphyxiation / 9 1.2.5 Freeze Burns / 9 1.2.6 RPT Explosions / 10 1.2.7 Roll Over / 10 Risk Analysis Requires Adequate Modeling / 10 Flammability / 11 Regulations in Siting Onshore LNG Import Terminals / 13 1.5.1 U.S. Marine LNG Risk and Security Regulation / 13 1.5.2 U.S. Land-Based LNG Risk and Security Regulation / 14 1.5.3 European and International Regulations / 15 Regulation for Siting Offshore LNG Import Terminals / 16 Controversial Claims of LNG Opponents / 16

LNG Incidents and Marine History 2.1

1

20

LNG Ship Design History / 20 2.1.1 Initial Design Attempts / 21 2.1.2 Tank Materials / 21 2.1.3 Insulation Materials / 21 2.1.4 Tank Design / 21 vii

viii

CONTENTS

2.2

2.3

2.4 2.5

3

Current LNG Carriers 3.1 3.2

3.3

4

Designs and Issues—First Commercial LNG Ships / 22 2.2.1 Membrane Technology / 23 2.2.2 Gaztransport Solution / 24 2.2.3 Spheres / 25 2.2.4 LNG Carriers for the Asian Trade / 26 2.2.5 Current State of LNG Tankers / 27 LNG Trade History / 27 2.3.1 European Trade / 27 2.3.2 Asian Trade / 28 2.3.3 Temporary Setbacks / 28 2.3.4 Revival of LNG with Worldwide Supply–Demand Pinch of Petroleum / 28 2.3.5 Supply History / 29 2.3.6 Some Economic Factors / 30 LNG Accident History / 32 Summary of LNG History and Relevant Technical Developments / 35

Design Requirements / 39 Membrane Tanks / 39 3.2.1 Tank Design and Insulation / 39 3.2.2 Dimensions and Capacity / 41 3.2.3 Tank Materials and Insulation / 42 3.2.4 Pressure and Vacuum Relief / 44 3.2.5 Design Issues / 44 Moss Spheres / 46 3.3.1 Typical Dimensions and Capacity / 47 3.3.2 Insulation and Tank Materials / 48 3.3.3 Pressure and Vacuum Relief / 48 3.3.4 Design Issues / 48

Risk Analysis and Risk Reduction 4.1 4.2 4.3 4.4

37

Background / 51 Risk Analysis Process / 52 4.2.1 Hazard Identification / 54 Frequency: Data Sources and Analysis / 57 4.3.1 Generic Data Approach / 57 Frequency: Predictive Methods / 58 4.4.1 FTA / 59 4.4.2 Event Tree Analysis / 60

50

ix

CONTENTS

4.5 4.6 4.7

4.8 4.9 5

LNG Discharge on Water 5.1

5.2

5.3 5.4 5.5

5.6 5.7

5.8 6

Consequence Modeling / 64 Ignition Probability / 64 Risk Results / 68 4.7.1 Risk Presentation / 68 4.7.2 Risk Decision Making / 70 Special Issues—Terrorism / 70 Risk Reduction and Mitigation Measures for LNG / 71

Type 1—Above Water Breaches at Sea / 76 5.1.1 Ship-to-Ship Collisions / 76 5.1.2 Weapons Attack / 80 Type 2—At Waterline Breaches at Sea / 81 5.2.1 Grounding or Collision / 81 5.2.2 Explosive-Laden Boat Attack / 81 Type 3—Below Waterline Breaches at Sea / 84 Discharges from Ship’s Pipework / 85 Cascading Failures at Sea / 86 5.5.1 Sloshing Forces / 86 5.5.2 Explosion in Hull Chambers / 87 5.5.3 RPT in Hull Chambers / 87 5.5.4 Cryogenic Temperature Stresses on Decks and Hull / 87 5.5.5 Cascading Events Caused by Fire / 88 Initial Discharge Rate / 88 Time-Dependent Discharge (Blowdown) / 90 5.7.1 Blowdown for Type 2 Breach (at Waterline) / 90 5.7.2 Blowdown for Type 1 Breach (above Waterline) / 92 5.7.3 Blowdown of Type 3 Breach (Underwater Level) / 94 Vacuum Breaking and Glug-Glug Effects / 103

Risk Analysis for Onshore Terminals and Transport 6.1 6.2 6.3 6.4

74

104

Typical Basis for LNG Receiving Terminal / 104 Features of LNG Receiving Terminals / 105 Standards for Receiving Terminal Design / 110 U.S. Guidelines and Regulations for Receiving Terminals / 112 6.4.1 LNG Transport Administered by the Department of Transportation (DOT) and the U.S. Coast Guard / 113 6.4.2 LNG Terminal Permitting by Federal Energy Regulatory Commission (FERC) / 113 6.4.3 Pool Fire Radiation Exclusion Zone / 114 6.4.4 Vapor Dispersion Exclusion Zone / 116

x

CONTENTS

6.5

6.6 6.7

6.8 6.9 6.10 7

European Regulations for LNG Receiving Terminals / 119 6.5.1 Features of EN 1473 / 119 6.5.2 Comparison of Prescriptive and Risk-Based Approaches / 120 Empirical Formula for Required Land Area of Terminal / 121 Leak in Loading Arm or in Storage Tank / 123 6.7.1 Modeling Effects of Substrate on Evaporation Rate / 124 6.7.2 Vapor Hold-Up Effect on Dispersion Zone Calculation / 126 Rollover / 129 LNG Land Transport Risk / 132 Offshore LNG Terminals / 132

LNG Pool Modeling 7.1 7.2

7.3

7.4

7.5

134

Flashing and Droplet Evaporation in Jet Flow / 135 Pool Spread and Evaporation Modeling / 136 7.2.1 Spread Rate on Smooth Surface / 138 7.2.2 Pool Spread on Land / 144 7.2.3 Pool Evaporation on Smooth Water Surface, Test Data / 144 7.2.4 Pool Evaporation, Heat Transfer Regimes / 145 7.2.5 Heat Conduction on Shallow Water with Ice Formation / 150 7.2.6 Composition Changes with Evaporation / 151 7.2.7 Type 1 Breach—LNG Penetration into Water, Turbulent Heat Transfer / 153 7.2.8 Time-Dependent Pool Spread / 156 Rapid Phase Transition Explosions / 159 7.3.1 Historical Experience with LNG RPTs / 160 7.3.2 Similar Phenomena More Thoroughly Investigated / 161 7.3.3 Explosion Energy of an RPT / 162 7.3.4 Models of RPT Explosions / 162 7.3.5 Superheat Limits / 165 7.3.6 TNT Equivalence / 166 Aerosol Drop Size / 166 7.4.1 Drop Size Distribution / 167 7.4.2 Droplet Breakup Mechanisms / 168 Heat Balance Terms to LNG Pool / 169

CONTENTS

7.6

8

7.5.1 Heat Conduction from Solid Substrate / 169 7.5.2 Heat Convection from Wind / 170 7.5.3 Radiation to/from Pool / 170 7.5.4 Evaporative Cooling on Water / 171 7.5.5 Bubble Flow in Vaporizing LNG / 171 Nomenclature / 172

Vapor Cloud Dispersion Modeling 8.1

8.2

8.3 8.4

8.5 8.6 8.7

8.8

8.9

xi

175

Atmospheric Transport Processes / 175 8.1.1 Wind Speed, Stability, and Surface Roughness / 176 8.1.2 Effect of Obstructions / 181 Model Types / 181 8.2.1 Gaussian Models / 182 8.2.2 Integral or Similarity Models / 183 8.2.3 CFD / 185 LNG Dispersion Test Series / 188 Factors Affecting Plume Length / 193 8.4.1 Heavy Gas Properties Increase Hazard Area / 193 8.4.2 Models Predict Average Conditions of Fluctuating Plume / 197 8.4.3 Wind Speed for Longest Plume / 201 8.4.4 LNG Vapor Cloud Lift-Off Limits Hazardous Plume Length / 202 8.4.5 Scooping of Confined Vapors / 202 Effect of Wind, Currents, and Waves on LNG Plume / 204 Comparison of Dispersion Model Predictions / 205 Descriptions of Dispersion Test Series / 209 8.7.1 Matagorda Bay Tests / 209 8.7.2 Shell Jettison Tests / 209 8.7.3 Avocet, Burro, and Coyote Test Series / 210 8.7.4 Maplin Sands Test Series / 210 8.7.5 Falcon Test Series / 211 Vapor Intrusion Indoors / 212 8.8.1 Basic Response for Indoor Concentration Buildup / 212 8.8.2 Experimental Observations Show Low Indoor Concentrations / 214 8.8.3 Concentration Reduction by Plume Impinging on Buildings / 214 8.8.4 Models of Infiltration into Buildings / 215 Theoretical Basis for Suppression of Turbulence / 220

xii

9

CONTENTS

LNG Pool Fire Modeling 9.1 9.2 9.3

9.4 9.5

9.6 9.7

9.8 9.9 9.10

9.11 9.12 9.13 9.14 9.15 9.16

Types of Fires from LNG Facilities / 222 The Challenge for Pool Fire Modeling / 223 Pool Fire Characteristics / 223 9.3.1 Fires Are Low-Momentum Phenomena / 223 9.3.2 Fire Structure / 225 9.3.3 Simplifying Pool Fire Structure / 228 Summary of LNG Fire Experiments / 230 Burning Rate Data and Correlations From Fire Tests / 230 9.5.1 Consistency Checks between Evaporation Rate and Burning Rate / 236 9.5.2 Stopping Point for Pool Fire / 236 Point Source Fire Model / 237 Solid Flame Models: Flame Length Correlations / 239 9.7.1 Small-Scale Pool Fire Tests and Flame Length Correlations / 240 9.7.2 Medium-Scale Pool Fire Tests and Flame Length Correlations / 245 9.7.3 Large-Scale Pool Fire Tests and Flame Length Correlations / 248 Flame Tilt Correlations / 249 Flame Drag Near Pools / 252 Sep Correlations and Smoke Shielding / 253 9.10.1 SEP from Tests / 253 9.10.2 Smoke Shielding and Theoretical SEP Values / 254 9.10.3 Validation Comparison of a Three-Zone SEP Model / 259 Atmospheric Transmissivity / 259 Trench Fires / 262 View Factors / 264 CFD Modeling / 266 Comparison of Model Predictions / 268 Fire Engulfment of LNG Carrier / 271

10 Other LNG Hazards 10.1 10.2 10.3 10.4

222

275

Fire and Explosion Scenarios / 275 Jet Fires / 276 Flash Fires / 286 BLEVEs, Fireballs / 291 10.4.1 BLEVEs and Applicability to LNG / 292 10.4.2 Applicability of BLEVEs to LNG Marine Vessels / 294 10.4.3 Fireballs from Released Vapor / 297

xiii

CONTENTS

10.5

10.6

LNG Vapor Cloud Explosions / 302 10.5.1 Characteristics of Detonations and Deflagrations / 303 10.5.2 Fuel Reactivity Effects / 306 10.5.3 Modeling VCEs / 308 10.5.4 CFD Modeling of VCEs / 311 Asphyxiation and Cryogenic Hazard from LNG Spills / 313

11 Fire Effects 11.1

11.2

12

Fire Radiation Effects on Individuals / 318 11.1.1 Injuries to People—Definition of Burn Degrees / 318 11.1.2 Measured Effect Levels from Radiation Exposure / 319 11.1.3 Thresholds of Injury on Thermal Dose Basis / 322 11.1.4 Radiation Dosage from Transient Events / 324 Effects of Thermal Radiation on Property / 324 11.2.1 Equipment Degradation by Thermal Radiation / 324 11.2.2 Thermal Weakening of Steel and Concrete / 325 11.2.3 Bursting Pressure Vessels, Rail Tank Cars / 327

Research Needs 12.1 12.2 12.3 12.4

12.5

318

329

Uncertainties / 329 Recommendations of GAO Survey / 330 LNG Model Evaluation Protocols (MEPs) / 333 Special Topics / 335 12.4.1 LNG Pool Spill and Fire Tests / 335 12.4.2 Limitation of Boussinesq Approximation / 337 12.4.3 LNG Plumes Not Modeled Well for Calm Winds / 337 12.4.4 The Use of ½ LFL as an End Point / 338 Conclusions / 339

References

341

Index

369

PREFACE

The development of the liquefied natural gas (LNG) industry and technology has been viewed as a major improvement in the utilization of the world’s energy resources. It has also been the subject of controversy, misunderstanding, and misleading information. The quality of the final product, clean natural gas, is widely acclaimed and accepted. It is the scale of LNG tanker ships that is a source of astonishment and apprehension. Scaling is at the heart of the technical issues regarding the risk/benefit decision to site LNG facilities and to fully utilize LNG’s potential. Much has been learned from experimental LNG spills of up to 66 m3 of LNG. Models have been developed based on established fundamentals along with the objective of extrapolation to spills up to 25,000 and even 75,000 m3. This book reviews current scientific understanding of the predicted behavior of such large accidental spills. In this book, the full cycle of possible hazards and consequence mechanisms associated with loss of containment accidents or deliberate breaches is reviewed. The underlying science governing discharge, pool formation and evaporation, dispersion, ignition and flash fire, and resulting pool fire is presented. Also presented are special hazards such as rapid phase transition, boiling liquid expanding vapor explosion (BLEVE), and vapor cloud explosions that are only possible under certain special circumstances. We recognize that some of the more important issues have just begun to be addressed. This includes possible escalation of an event to affect more than one of the five or so tanks in an LNG carrier from an engulfing fire. It is widely accepted that model development should precede and guide the design of larger and more costly spill tests. Already it is known which physical processes scale well and which do not. An objective of this book is also to describe speculation as to whether some mechanisms of observed pool fire burn rate, flame length, and explosion flame speeds merit extrapolation to large spills. It is important to understand the uncertainty levels of predictions and to focus on where to best invest in reducing uncertainty levels. At many recent public inquires and meetings, very conservative estimates for consequences of releases of LNG have been presented, sometimes completely impossible physically. Conversely, some LNG advocates tend to understate the potential hazards, and this leaves the public and the interested xv

xvi

PREFACE

professional unsure of what true hazards exist and the current means of technology to assess these hazards accurately. If communities are to be asked to host such facilities, it is helpful that there be a scientifically valid statement of the hazards and the best current means to assess their consequences, and this is the purpose of this book. This book seeks to review the technologies in use, particularly those relevant to marine transportation and reception terminals where the greatest public exposure exists. The full cycle of possible hazards and consequence mechanisms associated with loss of containment accidents or deliberate breaches is reviewed. The underlying science governing discharge, pool formation and evaporation, dispersion, ignition and flash fire, and resulting pool fire are presented. Also presented are special hazards such as rapid phase transition, BLEVE and vapor cloud explosions that are only possible under certain special circumstances. This book is addressed to the needs of the public and the legal profession as well as to the needs of management and engineers. The public can suffer from both overprediction and underprediction of risk by increasing the cost and availability of clean fuels on the one hand and by incurring losses by fire and explosion on the other hand. Management and engineers are charged with implementing the proper risk reduction measures and also keeping costs reasonable. Our objective is to address a wide range of these varying needs. While it is unlikely that readers will manually calculate LNG spillage consequences using the equations and graphs in this text, the presentation of the theory will allow them to assess with better confidence the results of computer consequence codes or other predictions made by proponents or objectors to LNG developments. We gratefully acknowledge the assistance of Dr. Sam Mannan and Donna Startz of the Mary Kay O’Connor Process Safety Center at the Texas A&M University for allowing us access to their considerable collection of references on LNG and for the use of their library for our collaboration meetings. We also wish to acknowledge the support of our companies, Baker Engineering and Risk Consultants, Inc., and Det Norske Veritas, in providing materials, encouragement, and time for our efforts on this book. John L. Woodward Robin M. Pitblado

1 LNG PROPERTIES AND OVERVIEW OF HAZARDS Liquefied natural gas (LNG) is simply a convenient form of natural gas, a cryogenic liquid condensed in volume to make storage and shipping economically feasible. Natural gas consists primarily of methane with smaller amounts of other light hydrocarbons such as ethane, propane, and butane. Natural gas occurs naturally throughout the world and has long been captured and transported to residences and industries by pipeline. Some large pipelines deliver natural gas along the ocean bottom from offshore wells and across continents. But there are a number of large natural gas fields too remote from consumers for economic transport by pipelines. Liquefying natural gas provides an economical way to extend pipeline networks from gas fields to consumers almost anywhere in the world. The primary uses of LNG are •

•

•

•

transporting natural gas by ocean transport to a market pipeline terminal; transporting natural gas by truck to local distribution systems (e.g., in China and in the United States); peak shaving storage at distribution points along natural gas pipelines; and power generation or home use with vaporized LNG—as natural gas.

LNG is made at a liquefaction plant and is restored to a gas at a regasification plant. Thus, the possibility of contact between LNG and the public is typically LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

1

2

LNG PROPERTIES AND OVERVIEW OF HAZARDS

very limited except in the immediate vicinity of the plant. Truck transport of LNG provides an exception to this generalization. In a liquefaction plant, there are several steps. The main steps, starting at the natural gas feed, include CO2 removal, dehydration and mercury removal, initial chilling, liquefaction (including heavier hydrocarbon fractionation), nitrogen rejection, and, finally, product LNG storage. Dehydration is usually achieved by molecular sieves, and mercury removal (which is necessary to avoid subsequent aluminum corrosion) is achieved either with mole sieves or with sulfur-impregnated carbon or alumina. Chilling and liquefaction is achieved with large multistage centrifugal compressors and expanders combined with cold boxes of complex internal design. Hydrocarbon fractionation is achieved with standard distillation columns—often in the sequence deethanizer, depropanizer, and debutanizer depending on the inlet gas concentration). Nitrogen removal can be achieved in several flash stages or by stripping and reboiling. The overall heat exchange is very important, and heat transfer optimization using pinch technology approaches is now common. A regasification plant is inherently endothermic (absorbs energy) since the LNG must be warmed to the temperature and pressure of the delivery pipeline. Since it is much more efficient to pump a liquid than to compress a gas, the LNG is pumped to pipeline pressure and then vaporized. The heat for vaporization can be provided by circulating seawater and air fin/fans or by burning part of the natural gas in heaters submersed in a water bath around LNG tubes. Again, there is inefficiency to this process, meaning some energy or, equivalently, some LNG is used for pumping and heating.

1.1

LNG PROPERTIES

The properties of LNG vary with composition, which depends on the location of the original gas as shown in Table 1.1 (U.S. Department of Energy [DOE], 2008). The original natural gas may contain many other materials including water vapor, carbon dioxide, nitrogen, and helium, some of which must be removed for liquefaction. The lightest composition is from Trinidad, which in 2005 accounted for 80% of the LNG imports to the United States. LNG with higher proportions of hydrocarbons with two and more carbon atoms is termed “rich” LNG because it has a higher specific heat of combustion than “lean” (Trinidad) LNG. The largest amount of LNG imported in 2005 was 58.6 million tons to Japan, or 30% of the world trade in LNG. A large portion of imports to Japan, as well as to South Korea and Taiwan, have been from Indonesia and the Middle East. The critical point of methane is 190.4 K, meaning methane cannot be liquefied by pressure alone at ambient temperature. Rather, it must be cooled to liquefy. At atmospheric pressure, it must be cooled to the boiling point in Table 1.2. This is quite different from liquefied petroleum gas (LPG, largely

LNG PROPERTIES

3

Table 1.1 Typical hydrocarbon composition of LNG from various locations

Component

Mole %, Source Location

Methane (C1H4) Ethane (C2H6) Propane (C3H8) iso-Butane n-Butane (C4H10) C5+ Total

Trinidad

Algeriaa

Nigeria

Oman

96.9 2.7 0.3 0.1

87.93 7.73 2.51 0.50 0.72 0.61 100.00

91.692 4.605 2.402 1.301 — — 100.00

87.876 7.515 3.006 1.603 — — 100.00

— — 100.0

a

Skidka composition after removing nitrogen and helium.

Table 1.2 Some properties of LNG

Property

Methane

Trinidad

Nigeria

Molecular weight Boiling point, K (°C) (bubble point) Liquid density (kg/m3 at boiling point) Vapor density (kg/m3 at boiling point) Vapor density (kg/m3 at 20°C) Temperature at liftoff, K (293 K air) Heat of combustion (higher, MJ/kg) Carbon footprint (g CO2/MJ) Flammable range for vapor (mole %) Vapor reactivity classification for explosions

16.043

16.55

17.91

111.67 (−161.5)

112.1 (−161.05)

422.5

Oman

Algeria

Gasoline

18.615

18.77

100–110

112.7 (−160.4)

113.3 (−159.9)

113.25 (−159.9)

321.1 (48.0)

430.9

452.8

463.6

452.9

627.3

1.810

1.799

1.776

1.763

1.783

2.927

0.6685

0.6894

0.7459

0.7751

0.7829

3.114

170.1

175.1

185.9

192.4

199.0

>Ambient

50.04

49.86

49.43

49.24

49.20

44.75

54.8

55.3

56.3

56.8

56.9

68.4

5–15

4.9–14.9

4.6–14.6

4.4–14.4

4.4–14.4

1.1–7.6

Low

Low

Low

Low

Low

Medium

4

LNG PROPERTIES AND OVERVIEW OF HAZARDS

propane and butane) that is liquefied at ambient temperature with several bars of pressure. The safety and environmental implications of the properties of LNG are illustrated in Table 1.3. 1.2

HAZARDS OF LNG WITH RESPECT TO PUBLIC RISK

The sources of LNG hazards occur by • •

• • •

liquid leaks under pressure (pump and pipe leaks), liquid leaks from storage tanks (the head pressure is usually atmospheric), rollover of an LNG storage tank, liquid pools evaporating to form a flammable vapor plume, and liquid leaks injected into water under pressure or from a moderately high elevation giving rise to a rapid phase transition (RPT) explosion.

Leaks under pressure are hazards inside processing plants (liquefaction or regasification) and from LNG transfers from storage to carriers and vice versa. Liquid leaks can occur from land-based storage tanks and from LNG carriers. Penetrations can occur by ship collision, allision (striking a fixed object), or grounding. Corrosion is a lower-risk cause of leaks since LNG typically has low corrosivity to materials used for its handling. An accidental release of LNG can pose the following hazards: •

•

• •

• • •

radiation burns and structural weakening from flash fire, pool fire, or jet fire; overpressure and impulse from partially confined vapor cloud explosion; overpressure and impulse from confined vapor cloud explosion; rapid spreading, evaporation, and possibly overpressures from an RPT explosion; asphyxiation; freeze burns; and rollover

These events usually occur in a sequence as illustrated in Figure 1.1 (Pitblado et al., 2006). The event sequence is in chronological order from the leak to pool formation with evaporation to form a vapor cloud, vapor cloud dispersion, delayed ignition, then burn back as a flash fire to a pool fire. Modeling of these events is treated in detail in Chapters 5, 7, 8, 9, and 10. The event consequences are briefly introduced below.

HAZARDS OF LNG WITH RESPECT TO PUBLIC RISK

5

Table 1.3 Safety and environmental implications of LNG properties

Property LNG is a cryogenic liquid.

LNG evaporates completely and cleanly without a residue. LNG evaporates rapidly from ground or water contact. The liquid density of LNG is low, less than half of that of water.

The expansion factor in going from liquid at the boiling point to vapor at standard ambient temperature is around 600 (594–625). The molecular weight of natural gas is less than that of air (specific gravity of 0.60–0.68).

A boiling pool produces cold vapors (at the normal boiling point). Water condensation in plume creates a visible cloud. The LFL (Lower Flammable Limit) concentration is always within the visible cloud for relative humidity above 55%. LNG vapors will quite quickly warm to ambient temperatures by conduction and/or by dilution with air. LNG vapors will ultimately warm enough to become buoyant and lift off, reducing the chance of ignition.

LNG has slightly higher energy density than gasoline (10–11% higher)

Consequence Direct contact with skin causes freezer burns. Exposure of sufficient duration can embrittle carbon steel. An LNG spill leaves minimal environmental impact (freezing effects only). Vapor plume is the main hazard from spills. It can ignite, then fire is the main hazard. LNG tankers float high in the water. A large tank of LNG, say 30 m high, would have a liquid head of around 1.3 atmospheres. This is a comparatively low pressure to pump against. This density difference provides for the economical transport and storage of natural gas as a liquid.

The low molecular weight of LNG vapor makes it lighter than air at ambient temperature. Natural gas rises and poses a lower threat than most hydrocarbon vapors, including gasoline, that are heavier than air. LNG vapors at their boiling point are significantly heavier than air, by about a factor of 1.5. Visibility helps in taking avoidance and escape measures. Photographs of LNG visible plumes are useful approximations of the flammable cloud.

By air mixing alone, the specific gravity of an evaporated LNG vapor plume approaches unity asymptotically from above by temperature warming and from below by increasing molecular weight. Temperature and molecular weight have opposite effects on the vapor-specific gravity. The molecular weight effect always drives an ultimate specific gravity less than 1.0. As warming occurs by dilution and conduction, then a vapor plume from an LNG spill is likely to rise (lift off) at some point downwind of the spill. LNG develops relatively high flame temperatures for small fires that are not oxygen starved.

6

LNG PROPERTIES AND OVERVIEW OF HAZARDS

Table 1.3

Continued

Property LNG has a strong advantage over burning liquid hydrocarbons or coal in generating less CO2 per unit of energy (81–83% as much). LNG liquid does not burn or explode. The vapor above LNG must mix with air to below 15% and above 5% of natural gas concentration to be flammable. Methane and light composition natural gas have a relatively high lower flammability limit (LFL, 5% compared to 1% for gasoline or 0.7% for crude oil). The burn rate of an LNG pool fire on land is “above the curve” for other paraffin hydrocarbons. LNG pool fires produce relatively little smoke

Applying dry chemical powder is the only way to extinguish an LNG fire. The fire will continue until all the fuel is burned.

LNG spills at a regasification terminal are directed to a sump, so ignition results in a pool fire at a safe location. Unconfined or partially confined LNG vapor/air mixtures do not detonate (form a sonic velocity explosion that self-propagates as discussed later). LNG vapor has low reactivity for explosion propagation.

Consequence LNG is preferred over liquid hydrocarbons or coal for environmental impact.

As for all hydrocarbon liquids, only the vapor above the liquid burns and can explode if sufficiently confined or congested. Much of the vapor cloud above an LNG spill is not in the flammable range. Only a fraction of the plume will ignite. An LNG vapor plume contour to the LFL does not cover as large an area as an otherwise equivalent gasoline spill.

The higher burn rate contributes to a tall fire of shorter duration, than a corresponding amount of higher-chain hydrocarbon. Bright nonsmoky flames generate higher emitted radiation, and thus LNG fires radiate more heat than heavier hydrocarbons. Larger pool fires produce more smoke, so the emissive power drops off with pool size, and this is believed true for the largest LNG pool fires as well. Water will not extinguish an LNG fire. Preinstalled fire fighting foams may slow the fire. However, extinguishment does not stop liquid boiloff and hence vapor cloud formation; thus, controlled burning can be safer than extinguishment. Complete burning avoids late ignition flash fire. The terminal design can provide adequate insulation of nearby structures. Water spray systems are being evaluated to reduce radiant energy at important locations from a sump pool fire. Considerable congestion and/or a high-energy ignition source is required to explode as a deflagration (a subsonic explosion that decays upon burning outside of a high-congestion zone). The flame speed of a natural gas deflagration is lower than other hydrocarbons because of its low reactivity.

HAZARDS OF LNG WITH RESPECT TO PUBLIC RISK

1

2

Initial gas cloud formation

Leak Pool formation

Ignition

Dense gas dispersion & ignitor

7

Scenario sequence: 1. Leak 2. Pool formation 3. Cloud dispersion 4. Flash fire back 5. Pool fire

Dispersion and flash fire Pool fire

3

4

Burnt gases

5

Combustion zone

Ongoing LNG pool

Uncombusted gas

Initial pool fire diameter Longer-term pool fire diameter

Figure 1.1 Scenario sequence for leak of LNG at sea (Pitblado et al., 2006) (reproduced by permission from Elsevier Science Publishers).

1.2.1

Flash Fire, Pool Fire, or Jet Fire

The main threat from LNG spills is a fire. Indeed, risk analyses for LNG primarily focus on the hazard of a pool fire. A jet fire requires a pressurized release that can occur in process plants but is not typically a threat to the public. An LNG spill on land or on water would result in a rapidly evaporating pool that produces a vapor cloud driven by the wind. If any point of a vapor cloud (with dimension defined to flammable concentrations) reaches an ignition source and ignites, a flash fire would burn downwind and possibly also upwind from the ignition point. A flash fire will burn faster along the premixed (diluted by air) edges. This can create a more enveloping fire as illustrated in Chapter 10. A flash fire is inherently transient, and exposure normally lasts no more than a few tens of seconds. While fatal to people inside the fire, the total radiation reaching an object near a flash fire is substantially lower than that from a longer-lasting pool or jet fire the same distance away. A flash fire likely does not produce secondary ignition or burns to people outside of the flaming region. After a flash fire burns back to the LNG pool, or if ignition begins at the pool, the result is a pool fire. An example is seen in Figure 1.2 (Sandia, 2009). Figure 1.2 is a bright fire with no smoke. Larger fires on land (e.g., 35-m diameter) become oxygen limited and smoky. Larger fires on water are expected

8

LNG PROPERTIES AND OVERVIEW OF HAZARDS

Figure 1.2

Example of a large 23-m diameter LNG pool fire on water (Sandia, 2009).

to exhibit similar smokiness, and Sandia is carrying out larger scale experiments than in the figure in 2009 to confirm this. While the LNG outflow continues, an unconfined burning LNG pool tends to either increase or decrease in size toward achieving a final steady-state size. This is the size for which the burn-off rate equals the discharge rate. The steady-state pool size is smaller for a burning pool than for a nonburning pool. So, if ignition is not immediately after a spill begins, the burning pool will retreat significantly compared to its original size. Jet fires and pool fires are treated in detail in Chapter 9. Flash fire and fire balls or boiling liquid expanding vapor explosions (BLEVEs) are discussed in Chapter 10. 1.2.2

Outdoor Vapor Cloud Explosions

Experiments have confirmed that an outdoor vapor cloud explodes only under conditions of partial confinement and/or in congested regions. Congested regions are defined by a high density of obstacles such as piping, pumps, and other such equipment. Congested regions can be found in LNG liquefaction plants and terminals. LNG spills at sea, even if caused by colliding ships, are not in a confined or congested environment. The upper decks of modern LNG vessels may offer limited congestion with reliquefaction equipment, but this will be well above any dense cloud on the sea surface. LNG spills from a docked tanker can occur beside the side of a tanker, but this is considered a 3-D expansion zone and congestion is limited to the presence of posts supporting the dock. Another factor that mitigates against a possible outdoor explosion of LNG vapors is the low reactivity of natural gas. Detonation explosions are virtually ruled out by low reactivity. A deflagration

HAZARDS OF LNG WITH RESPECT TO PUBLIC RISK

9

explosion from an outdoor spill of LNG in an LNG terminal is a low probability event. 1.2.3

Enclosed Vapor Cloud Explosions

Explosions occur with noticeable frequency from a buildup of natural gas vapors indoors or inside any enclosed space. Commonly, such explosions result from leaking natural gas lines in a building. LNG is held at a temperature within a few degrees of the normal boiling point. The atmosphere inside an LNG storage tank, truck, or marine carrier is 100% boil-off vapor with no oxygen content. Even a worst case vacuum breaker valve opening would not allow sufficient air ingress for the vapor space to become flammable. Vapor from a passing LNG cloud could leak into or be induced into a building. LNG delivery lines at regasification or liquefaction plants are not allowed inside buildings. Air intakes into buildings are usually elevated above most LNG dense vapor clouds, and the circumstances for vapor induction into a building are rare. For these reasons, a confined LNG vapor cloud explosion is a very unlikely threat. Chapter 10 further discusses unconfined LNG explosions and vapor intrusion into buildings. 1.2.4

Asphyxiation

For asphyxiation, the LNG vapors must dilute the oxygen concentration in the breathing zone of people below 15% oxygen for impaired behavior, below 10% for nausea and vomiting, or below 6% oxygen for death. The concentration of LNG vapor required to reach these end points is 28.2%, 52.2%, and 71.3%, respectively, and the higher concentrations would also be associated with freeze burns. These concentrations exist only near the spill for an outdoor release. The normal variations in the wind direction and evasive measures by any individual so near a vapor plume make it very unlikely that asphyxiation will occur outdoors. The public is extremely unlikely to be near the point where LNG vapor concentrations are above 28.2%. A spill into an occupied confined space is also very unlikely because of industrial safety practices regarding confined space entry. These rules require the presence of a second person, the use of a rescue harness, air testing, and such precautions that mitigate any potential for an asphyxiation event, and the presence of LNG operations make it even less likely that confined space work would be ordered. 1.2.5

Freeze Burns

A single incident occurred in which LNG accidentally leaked under pressure near enough to a person to cause a freezer burn. This was in 1977 at Arzew, Algeria during a ship-loading operation when a large-diameter valve ruptured and the worker was sprayed with part of 1500–2000 m3 released LNG

10

LNG PROPERTIES AND OVERVIEW OF HAZARDS

(CHIV, 2003). The valve was made of aluminum. Current practice requires valves to be made of stainless steel. This is a recognized hazard for industrial workers, but not for the public. Further details of asphyxiation and freeze burns are discussed in Chapter 10. 1.2.6

RPT Explosions

An RPT explosion is a physical explosion and is due to the sudden boiling or phase change from liquid to vapor that has occurred upon occasion when LNG is spilled onto water, usually in a way that the LNG penetrates into and mixes well with water. No injuries have occurred from an RPT of LNG, but equipment has been damaged. The overpressures developed by an RPT have not been measured well enough yet, but observations indicate that the overpressures have not been high enough to cause personnel injury. RPTs are discussed in Chapter 7 and are included in issues that require further research in Chapter 12. 1.2.7

Roll Over

Early in the development of LNG, the importance of mixing LNG stored in tanks was not realized. It is now understood that LNG tanks can stratify upon standing. The bottom layers always exist under the pressure of hydraulic head and can, therefore, be at pressure equilibrium at a temperature quite a few degrees higher than the top layers. Since liquid density of the upper layer can increase over time due to boiloff of methane increasing the percentage of heavier components, at some point the layers can invert. This would bring the lower layer to the surface, and without the hydrostatic pressure above it, a small fraction would immediately flash. Since the expansion ratio of liquid to vapor is 600:1, even a small flash can generate a large volume of gas. The sudden increase in tank pressure can exceed the capacity of pressure relief valves that are designed for fire exposure and threaten roof or even wall failure. This is primarily a hazard to personnel at an LNG export or import terminal, although a complete tank failure would be a large event that could extend beyond plant boundaries. Rollover is treated in Chapter 6.

1.3

RISK ANALYSIS REQUIRES ADEQUATE MODELING

Experience with transporting and using LNG so far has been highly favorable, as is discussed in Chapter 2. No incidents, such as groundings or ship collisions, have resulted in spills of LNG cargo. Following the terrorist attacks of September 11, 2001, however, experts recognized that an attack on an LNG carrier could result in a large spill, that is, a volume up to 100 times greater than studied in past experiments. Because a major LNG spill has never occurred, studies evaluating LNG hazards must rely on computer models to

FLAMMABILITY

11

predict the effects of potential accidents and attacks. This approach sometimes requires extrapolation of experiments into the range where the underlying mechanisms may change. An example is discussed in Chapter 9 concerning the extrapolation of the flame height from pool fires. Pool fire experiments so far have ranged to pool diameters up to 35 m for LNG on land. The resulting flame for small LNG fires is usually bright, indicating that adequate air reaches the burning fuel. Large pool fires, though, become smoky, indicating the onset of oxygenlimited burning. Extrapolating to LNG pools of possibly 100+ m diameter poses the question of whether the fire will break up into smaller segments, fed by cells of alternating updraft and downdraft. If so, the flame height might be much shorter than extrapolations for a single united flame indicate. A shorter flame height would decrease the exposure angle for radiation flux and would produce a much lower radiation hazardous zone than is predicted from a single, very high (up to 350 m high) flame. Other examples of the limitations inherent in projecting beyond our current testing experience are covered in this book. Errors in overpredicting catastrophic effects can be as costly to the public’s best interest as can errors in underprediction. The objectives of this book are to clearly state what test data establish, what models predict, and what uncertainties remain.

1.4

FLAMMABILITY

Pure methane has flammability limits of 5–15% (volume or mole) in air, but as LNG is composed of multiple light-ends including noncombustible nitrogen, its actual flammable range can vary somewhat from the range quoted for pure methane. The ignition likelihood is also affected by the ignition energy as shown in Figure 1.3 from Zabetakis (1965). While this might imply narrower flammability limits, in practice, many common ignition sources found in LNG terminals and in surrounding urban or suburban locations can be strong ignition sources such as fired heaters, open flames, or motor vehicles. Even area classification rules (e.g., API 500, IP 15) include a probability aspect and weak ignition sources can be sited at normally nonflammable locations (Class 1, Division 2), which could be reached by a rare major spill. Strong ignition sources can be located beyond the Class 1, Division 2 zone. Typical flammable limit ranges for common LNG components are readily available from many sources (Lees, SFPE Handbook, etc.), and these are shown in Table 1.4. Rules are available in these references for estimating the flammable concentrations of mixtures. Smaller LNG spills will flash off sufficiently quickly that the cloud concentration will be close to that of the total LNG composition. However, large spills of LNG will boil off progressively with the lighter ends preferentially boiling first followed by heavier materials. Large spills may take many minutes to hours to entirely boil off and significant

12

LNG PROPERTIES AND OVERVIEW OF HAZARDS

Spark energy (MJ)

Limits of flammability 4.0 lgnitibility limits

2.0 1.0 0.8 0.6 0.4 0.2 2

4

6

8

10

12

14

16

18

methane (vol %) Figure 1.3

Table 1.4

Flammable limits and ignition energies for methane (Zabetakis, 1965).

Flammable limits for common LNG components

Material

Specific Gravity (Air = 1)

Lower Flammable Limit (Vol %)

Upper Flammable Limit (Vol %)

Methane Ethane Propane n-Butane

0.55 1.04 1.52 2.01

5.0 3.0 2.1 1.8

15.0 12.4 9.5 8.4

concentration variations would be expected with time. The earliest boiloff will tend to be at the highest rate on land (as the LNG has not yet fully cooled the soil beneath) and concentration will be richer in methane; thus, the largest cloud distance will be methane rich, and the conventional 5–15% flammability range is the most relevant, even if a subsequent boiloff may have heavier components that might reduce the lower flammable limits. Spills on the sea typically do not reduce in boil-off rate as the cool-down effect on land does not occur on sea, as chilled seawater sinks and is replaced by fresh warm seawater, but lighter ends will still preferentially boil sooner. Further details are in Section 7.2.6. The initial flash will primarily be pure methane. Reid (1980) provides a graph showing the evaporation sequence (“trajectory”) for a mixture of 85% methane, 10% ethane, and 5% propane in Figure 1.4. While the graph shows the residual liquid concentration, the vapor concentration can be inferred by the straight line decline at uniform ethane–propane residual concentration in

REGULATIONS IN SITING ONSHORE LNG IMPORT TERMINALS

13

Propane Tw = Tw / Tsl = 1.0 298K 288K 278K

Evaporation trajectory

Ethane

Initial LNG 85% CH4 10% C2H6 5% C3H8

Methane

Variation in LNG composition during evaporation Figure 1.4 LNG boil-off sequence showing residual liquid concentration (Reid, 1983) (reproduced by permission of Elsevier Science Publishers).

the liquid that the boiloff is essentially pure methane until almost all the methane is evaporated. The authors have confirmed the initial boiloff is essentially 100% methane using the PHAST consequence model. 1.5 REGULATIONS IN SITING ONSHORE LNG IMPORT TERMINALS Regulations in the United States, Europe, and Asia establish separation distances based on consequence analysis or risk assessment and other requirements for LNG import terminals, as summarily covered below. These regulations affecting risk analysis are covered in Chapter 6. 1.5.1

U.S. Marine LNG Risk and Security Regulation

The United States Coast Guard (USCG) is the lead federal agency for maritime security in the United States. It derives its risk and security responsibilities under the Ports and Waterways Safety Act of 1972 (P.L. 92-340) and the Maritime Security Act of 2002 (P.L. 107-295). Under the latter act, the USCG also has siting approval authority for offshore LNG terminals. USCG regulations cover waterways, the jetty, and the pipework up to the first valve at the receiving storage tank. The USCG determines the suitability of waterways to transport LNG safely and requires a waterway suitability assessment (WSA) for operations on proposed waterways. The WSA describes the LNG carrier escort plans, local emergency response capabilities, ship speed limits, and the like. The USCG requires operations and emergency manuals be submitted for each port where ships will operate. It creates safety rules for specific ports to minimize the chance for accidents (IELE, 2003b).

14

LNG PROPERTIES AND OVERVIEW OF HAZARDS

“The most heavily secured LNG shipments are those bound for the Everett terminal because they pass through Boston harbor” (Parfomak, 2004). For these shipments, the USCG has had numerous security provisions, including (Greenway, 2003) • • • •

•

• •

• •

inspecting tanker loading at the port of origin for Trinidad shipments; occasional on-board escort by Coast Guard “sea marshalls”; advanced notice of arrival of an LNG tanker by 96 hours; advanced notification of local police, fire, and emergency agencies as well as the Federal Aviation Administration and the U.S. Navy; boarding of LNG tankers for inspection prior to entering Boston harbor; harbor escort by armed patrol boats; enforcement of a security zone closed to other vessels two miles ahead and one mile to each side of an LNG tanker; suspension of overflights by commercial aircraft; and additional security measures that cannot be disclosed publicly.

Parfomak (2004) cites the USCG saying that many of these security provisions are in place for other U.S. LNG terminals as well and would likely be put in place for new on-shore LNG terminals. On October 22, 2003, the USCG issued final rules for security requirements mandated by P.L. 107-925 in Title 33 of the Code of Federal Regulations, Chapter 1, Subchapter H. “The rules require certain owners or operators of marine assets to designate security officers, perform security assessments, develop and implement security plans, and comply with maritime security alert levels” (ibid.).

1.5.2

U.S. Land-Based LNG Risk and Security Regulation

Federal Energy Regulatory Commission (FERC) Oversight The U.S. FERC is responsible for permitting new land-based LNG import terminals and for ensuring safe operation through subsequent inspections (18 CFR 157, 49-CFR-193). FERC requirements include security cameras, hazard detectors, and zones to protect residents and businesses surrounding an LNG terminal from 1. thermal radiation from a fire in the LNG impoundment area that holds any LNG accidentally leaked from the carrier unloading lines or the LNG storage tank and 2. flammable vapors from the LNG impoundment area that could ignite beyond the LNG terminal boundaries.

REGULATIONS IN SITING ONSHORE LNG IMPORT TERMINALS

15

The FERC derives its siting authority under the Natural Gas Act of 1938 (15 USC 717). It has jurisdiction over all existing LNG marine terminals in the United States and, in 2004, over 15 peak shaving plants involved in interstate gas trade (Parfomak, 2004). To meet these objectives, the FERC regulations cite the National Fire Protection Association (NFPA) standard NFPA 59a and establish a thermal radiation exclusion distance and a vapor cloud exclusion distance. These distances, from the impoundment area to the nearest fence line, establish the required minimum area for an LNG import terminal in the United States. Federal Pipeline Safety and Security Agencies The Office of Pipeline Safety (OPS) within the Department of Transportation has authority to regulate the safety and security of LNG peak shaving plants under the Natural Gas Pipeline Safety Act of 1968 (P.L. 90-481). The OPS regulations for peak shaving plants are found in 49 CFR 193, Liquefied Natural Gas Facilities: Federal Safety Standards (Subpart J—Security). The OPS regulations govern protective enclosures, communications, monitoring, lighting, power sources, warning signs, and security procedures. Transportation Security Administration (TSA) The Pipeline Branch of the TSA is the lead U.S. federal authority for the security of the interstate gas pipeline network under the Natural Gas Pipeline Safety Act of 1968 (P.L. 90481). This security authority was transferred to TSA from the Transportation Department’s OPS under the Aviation and Transportation Security Act of 2001 (P.L. 107-71). The TSA has visited the largest pipeline operators including some with LNG plants to review their security plans based on the OPS/ industry guidance circulated in 2002. However, TSA does not plan to inspect all plants because all land-based LNG plants may not be considered “nationally critical” (Parfomak, 2004). 1.5.3 European and International Regulations The European Standard EN 1473 (2005) stipulates the requirements for the design, construction and operation of on-shore LNG facilities. Unlike the U.S. regulations and standards, which are prescriptive, the EN 1473 regulation is based on a different philosophy. It requires a risk analysis to satisfy an acceptable level of risk for “for life and property outside and inside the plant boundaries.” This approach requires hazard assessment, consequence assessment, and assessment of frequencies of occurrence for events from small to large releases of LNG. This approach allows consideration of mitigation factors to reduce the magnitudes or frequency of potential events. That is, the analyst is allowed consideration of topography, shielding by trees and buildings, full or partial holdup of dispersion of vapors, thermal radiation absorption in the atmosphere, and the probability of early or late ignition.

16

LNG PROPERTIES AND OVERVIEW OF HAZARDS

The International Maritime Organization (IMO) followed the USCG in developing maritime security standards outside U.S. jurisdiction. These standards, the International Ship and Port Facility Security (ISPS) Code, contain detailed mandatory security requirements for shipping companies, port authorities, and governments. The code is intended to provide a standardized, consistent framework for governments to evaluate risk and “offset changes in threat with changes in vulnerability” (IMO, 2002).

1.6 REGULATION FOR SITING OFFSHORE LNG IMPORT TERMINALS In 2004, four U.S. offshore terminals were being considered: three in the Gulf of Mexico and one offshore of Oxnard, California (Parfomak, 2004). These would be connected to land only by underwater pipelines. According to one report, they may need to overcome technical challenges with their floating designs (Shook, 2003). The USCG reviews applications for offshore LNG import terminals, which must provide risk analysis, vapor dispersion modeling, and fire radiation exclusion zones. Further discussion is in Section 6.9 of Chapter 6.

1.7

CONTROVERSIAL CLAIMS OF LNG OPPONENTS

A number of claims made by opponents of LNG import terminals cite obsolete studies or are unsubstantiated by reliable data and facts. Some claims found on Internet websites in 2009 are listed in Table 1.5 and are compared with current substantiated technical material.

Claim #4. (Van der Linde and Hintze, 1978) In a totally fictional prologue, the authors postulate the effects of a 50,000-t crude oil tanker being piloted up the Arthur Kill between Staten Island and the New Jersey shore. The tanker collides at 14 knots at right angle into the midsection of a docked LNG carrier (unspecified type). The authors speculate that

Claim #2. Each gallon on LNG has several hundred times the energy potential of a gallon of gasoline. Claim #3. An LNG tanker holds 3,000,000 gal (11,360 m3) of LNG, the energy equivalent to 55 Hiroshima atomic bombs (Riley website and the film The Risks and Danger of LNG)

Claim #1. An LNG vapor cloud will be 127 mi long from an attack on an LNG carrier (Riley website and the film The Risks and Danger of LNG)

Claim

1000 lb of wood equals 3530 lb of TNT explosive 1000 lb of coal equals 4470 lb of TNT explosive A 24-gal automobile gasoline tank equals 1225 lb of TNT explosive.

1. A single tank of LNG contains 25,000 m3 or about 10,625 t. The authors, thus, speculate that nearly the entire LNG double hull is cut open by the collision. This deep a penetration is unlikely by ship collision modeling discussed in Chapter 5. 2. The spilling LNG will flow downward from the split hole, not onto the carrier deck. In any event, cracks in an embrittled deck would be irrelevant. However, the Sandia report acknowledges that “Both the ship itself and other LNG cargo tanks could be damaged from a large spill.” (Hightower et al., 2004, p. 38).

Hazard potential depends upon both the amount of energy and the rate at which it is released. Energy released by burning LNG is relatively slow (Melhem et al., 2006). Rebuttal: Much of the technology cited below and detailed later in this book was not available in 1978.

•

•

•

Using the same flawed reasoning, one can conclude that:

Obsolete models made erroneous predictions early in the history of LNG risk analysis. The errors include using an incorrect end point for the flammable limits, using incorrect air entrainment modeling, and using passive dispersion models. Subsequently, better models have been developed using information from large-scale LNG spill experiments. Modern models have “converged” and now provide much more consistent predictions as discussed in Chapter 8. The Sandia report (Hightower et al., 2004) predicts the maximum distance for an unignited LNG cloud to be 2500 m (1.55 mi). For a fire radiation of 5 kW/ m2, maximum distances are 500–1600 m (0.31–1.0 mi) (see also FERC-ABS Consulting, 2004, ABS, 2004). See Table 1.1. The heat of combustion of (e.g., Algerian) LNG is 49.2 MJ/kg or 84.35 MJ/gal compared with gasoline at 44.75 MJ/kg or 106.3 MJ/gal.

Current Substantiated Information

Table 1.5 Current public claims of LNG hazards vs. substantiated information

CONTROVERSIAL CLAIMS OF LNG OPPONENTS

17

1. 10,000-t spills from a single tank. 2. The spilling LNG fractures the deck of the carrier. 3. The spilling LNG instantly freezes the surrounding waters of the Arthur Kill. 4. The spilled LNG ignites and burns back to the source and causes the remaining four LNG tanks on the carrier to explode. 5. As the four LNG tanks explode they project fragments of steel that penetrates the on-shore storage tanks at the LNG import terminal. 6. Shock waves from the explosions flatten oil and gasoline storage tank farms. 7. Buildings are toppled by the explosions. 8. Blocks of stone and girders rain from the sky. (This must presumably be from LNG vapors entering buildings and exploding inside a building since there would be no oil and gasoline from flattened storage tank farms.) 9. The authors city an early paper by J. Fay postulating that a freeze burn area from a full tank loss of LNG would extend 12 miles.

Claim

Table 1.5 Continued

3. It is not possible to instantly freeze a flowing body of water as large as the Arthur Kill. See discussion on heat transfer from LNG to water in Chapter 7. 4. The remaining four LNG tanks are well insulated, so heat from the fire would have little influence except to increase the boiloff of vapors that would be vented by the pressure relief valve. The vented vapor would burn as a vertical plume. The concentration of LNG vapor inside the tanks would remain well above the upper flammable limit and would not ignite even as the vented vapors burn. The conditions for an explosion are absent (see Chapter 10), so the tanks would not explode. There is an issue with cryogenic damage to the carrier structural elements and Sandia National Laboratories is investigating for the DOE the potential for cascading failures from this mechanism in 2009. 5. LNG storage tanks are built with an outer wall capable of withstanding fragments from an explosion, especially as far away as the docks. 6. There is considerable evidence from large-scale explosions at refineries and chemical plants that storage tanks have too much bulk to have more than minor damage from explosion blast waves as far away as from a neighboring plant (or dock in this case). Dents and small penetrations have been observed in the vapor space of tanks, but these do not lead to more than a minor venting of vapor. 7. Buildings would not be toppled since there is too little congestion and confinement in the harbor area or on the water to generate an outdoor vapor cloud explosion. Modeling studies are done extensively around industrial areas and can be invoked to obtain a more realistic prediction of the size of a flammable vapor cloud from a spill of 10,000 t of LNG. The blast waves above 3 psig from a large methane release do not usually extend beyond 500 m. 8. There is a considerable delay before LNG vapors passing or even surrounding a building to build up to concentrations above the lower flammable limit before an indoor explosion could be possible. This delay time is readily enhanced by turning off the heating, ventilation, and air conditioning (HVAC) systems. See Chapter 8 discussion of vapor intrusion indoors. The vapor cloud dissipates rapidly if the wind is not nearly absolutely calm.

Current Substantiated Information

18 LNG PROPERTIES AND OVERVIEW OF HAZARDS

This accident did not involve LNG, but rather was determined to be a construction accident in an empty tank. Between the mid-1960s and mid-1970s, more than 60 LNG facilities were built in the United States as peak shaving plants. Over that period, these plants have had an excellent safety record.

Claim #5. The February 1973 accident at a Staten Island, New York LNG terminal storage tank was out of service and was being repaired. A fire developed and 37 people were killed (for details, see Chapter 2.) The claim is that “after all 37 people lost their lives at an LNG facility.” Claim #6. Fire burning back to an LNG tank would cause it to explode (Quillen, 2002).

Since an LNG tank is well insulated, an external fire would have a small influence on increasing LNG vaporization and release through pressure relief valves. The emitted vapors would likely ignite and generate a nearly vertical fire plume similar to a plant flare. There would be no congestion above the LNG tank to promote flame acceleration, which is a prerequisite for an explosion.

9. Chapter 10 lists the conditions for freeze burn and asphyxiation by LNG vapors. The vapors warm from −162°C by mixing with air at 20°C. Modern dispersion models put the distance for a plume of 10.8-t spill warming to 0°C as 1.4 mi. (wind speed 2.5 m/s) 10. This is a ridiculously high, supersonic wind speed (Mach 1.3) and is not physically possible in a fire.

Current Substantiated Information

10. Concerning air induced into a large pool fire of LNG: “Estimates of wind speeds resulting from a large LNG fire range as high as 1,000 miles per hour.”

Claim

CONTROVERSIAL CLAIMS OF LNG OPPONENTS

19

2 LNG INCIDENTS AND MARINE HISTORY

The liquefied natural gas (LNG) carrier industry, from its inception in the 1950s, has been distinguished in commercial shipping as a separate entity characterized by long contracts, large capital investment, and a high level of technology. The entire LNG chain has been heavily studied and has been subject to extensive governmental oversight and reviews. The focus of this chapter is toward marine historical developments and a summary of all LNG incidents. 2.1

LNG SHIP DESIGN HISTORY

Godfrey Cabot patented an LNG river barge in 1915. Natural gas pipelines were not yet widely developed in the United States, so LNG transport was proposed up the Mississippi River from Lake Charles, Louisiana, to Chicago, Illinois. As the project was first proposed, a novel application was suggested to add to the project economics; perhaps LNG could be used to freeze meat at the important Chicago stockyards. This concept did not prove feasible, and 40 years elapsed before there was interest in the international transport of LNG. Shipping LNG by sea began in the 1950s as a result of significant pressure to make use of newly discovered gas reserves and to discontinue flaring of what might otherwise be a useful product. Other significant pressures existed as well, as stated by Roger Ffooks (1993): LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

20

LNG SHIP DESIGN HISTORY

21

There was no single or major technological development that resulted in the conception of shipping natural gas by sea. The impetus to develop this capability was the result of a gradual build-up of pressures, from different directions and for different reasons—most often a combination of commercial and political factors.

2.1.1

Initial Design Attempts

In the early 1950s, several different groups began studying the feasibility of carrying LNG on rivers by barge. These included William Wood Prince in Chicago, Dr. Oivind Lorentzen (Det Norske Veritas [DNV], Norway), the Shell Group coordinated in London, and Gaz de France. While the initial attempts to develop designs for water transport of LNG were for barges, it was soon realized that the concept had much more potential for seagoing ships. For this reason, approvals of ship classification societies were essential. 2.1.2

Tank Materials

Potential materials of construction consisted of stainless steel, aluminum, aluminum bronze, copper, and some precious metals such as silver. Of these, only the first two were feasible from an availability and cost perspective. Steel was known to have improved low-temperature properties with increasing nickel content. (The phenomenon of brittle fracture was not yet well understood.) As a result, designers were uncomfortable with using any but 18.8 stainless steel—an 8% nickel steel (ibid.) Aluminum had not yet been used in larger structures and the welding of aluminum presented problems. 2.1.3

Insulation Materials

Balsa wood was identified as an appropriate material that possessed the necessary combination of insulating properties with the ability to contain methane, thus acting as a secondary barrier. All other economic, technical, and physical property questions regarding its use had been answered. The challenge was to develop a balsa design that would contain liquid methane in an emergency. 2.1.4

Tank Design

The issues that complicated tank design included ship motions, independent expansion and contraction, tank remaining in position despite ship motion, ship hull deflections (which had not been quantified), and severe temperature gradients during filling and emptying. While design work was ongoing, a long-term study was begun by the American Petroleum Institute (API) and the U.S. Bureau of Mines concern-

22

LNG INCIDENTS AND MARINE HISTORY

ing the behavior of a large burning pool of LNG. This program was to satisfy the authorities including the British Government that liquefied methane presented no special hazard provided proper precautions were taken and proper materials were used in handling and storing it. The British Government’s “Petroleum (Liquid Methane) Order, 1957, No. 859” classified liquid methane as a petroleum spirit, no more dangerous than other volatile petroleum derivatives (ibid.). A number of other studies were also carried out during this period. A summary of key findings is listed below. Balsa Wood Conceptual ship design studies were conducted in 1955. A materials survey found that balsa wood was the only known insulating material that could perform as required. Burness Corlett Work The Consulting Naval Architects were commissioned by two groups to study the economics of LNG ship transport and to develop a design for a ship of around 14,000-t methane carrying capacity. This work focused on a cylindrical form of containment, which offered reduced cost of materials and an intrinsically reduced-stress shape. Dr. Lorentzen’s Sphere Oivind Lorentzen designed and patented a methane tanker by mid-1955, with DNV’s approval in principle. His design comprised six aluminum spheres, four being 24 m in diameter and two of 20-m diameter, each supported by a circumscribing annular ring allowing fluid flow through the annular duct to maintain the temperature at the outermost surface of the aluminum ring at or about ambient. As a result, the steel supporting ring, which was a part of the ship’s structure, would not be subject to low temperature. First Prototype—Methane Pioneer The Comstock Group in the United States contracted with British Gas in the United Kingdom, and on January 25, 1959, the LNG Princess sailed from Lake Charles, LA with the first cargo of LNG, arriving at Canvey Island on February 20. By June 1959, the U.S. group had successfully shipped several cargoes of LNG across the Atlantic in the prototype ship Methane Pioneer. Land storage tank design was in a state of flux after a 1941 catastrophic failure in Cleveland, Ohio, of a newly built LNG storage tank that had walls of 3.5% nickel, low carbon steel. Considerable vital design information on materials and equipment and also the characteristics of the LNG cargo itself were amassed in these first design attempts (Ffooks, 1993). 2.2

DESIGNS AND ISSUES—FIRST COMMERCIAL LNG SHIPS

Events moved quickly after the successful trial shipments by the Methane Pioneer. By 1960, negotiations were well advanced toward the conclusion of a 15-year contract with Algeria to ship 100 million cubic feet of gas per day

DESIGNS AND ISSUES—FIRST COMMERCIAL LNG SHIPS

23

to Great Britain and half that amount to France. Different ship designs were being developed concurrently. By 1964, the Methane Progress and the Methane Princess each with a cargo capacity of 27,400 m3 were making weekly trips from Algeria to Canvey Island. Their size and schedule were set to meet the U.K. demand. In France, the Jules Verne was being designed to transport 50 million scf (standard cubic feet) per day of natural gas from Arzew to Le Havre. Soon, a fleet of seven carriers served France. They were equipped with tanks made of low-carbon 9% nickel steel and had a total capacity of 25,840 m3. The tanks were a combination of different geometrical shapes: the bottoms were part spherical, part conical, part elliptical-toroid; the sides were cylindrical; the tops were ellipsoidal. In an emergency, the tanks could be lifted right out of the ship without compromising structural integrity. Insulation was a combination of materials consisting of perlite, polyvinyl chloride (PVC), and glass cloth. The secondary barrier was also a combination of materials, depending on the location—the bottom was 9% nickel steel and was in direct contact with the tanks; the sides were impregnated glass cloth, attached to the face of the PVC and designed with built-in flexibility; stainless steel was fitted around the keys to create a liquid-tight barrier in these critical areas. This ship entered service in 1965 and has traded between Algeria and Le Havre since that date (Vaudolon, 2000). A key factor in the development of marine transport was the publication of the first International Gas Code (IGC) by the International Maritime Organization (IMO) in 1975. 2.2.1

Membrane Technology

The membrane system technology brought a cost breakthrough in LNG technology and introduced a new design and construction concept to the shipbuilding industry. Sponsors Oivind Lorentzen, shipowner in Oslo, and Bennett Group Joint Venture of Dallas demonstrated “a low-temperature integrated trial tank for sea transport of LNG.” Collectively, they had commissioned the DNV research and development department to investigate LNG shipping techniques, primarily a spherical design. DNV had already developed an idea for a biaxially flexible liner, the unique characteristics of this liner being that it embodied a method of folding the liner without stretching it so that on contraction, it could unfold elastically. The technology involved in this design would allow a construction cost decrease of about 25% and would improve the overall safety of LNG marine transport. The main thrust of their work was now to develop a membrane design that could be reliably produced on a commercial basis without creating undesirable stress concentrations at the points where the corrugations crossed, and to design a similar detail for tank corners. Control of the atmosphere in the spaces is important in a membrane system—it must exclude air and moisture, and in addition, pressure differentials

24

LNG INCIDENTS AND MARINE HISTORY

Figure 2.1 Membrane-type LNG carrier moored offshore (reproduced by permission from Gaztransport & Technigaz).

must be carefully maintained to avoid excess back pressure, which can force the membrane off its attachments to the insulation. By 1967, Conch and Technigaz (TGZ) formed a partnership, Conch Ocean; Conch was recognized to have the best insulation/secondary barrier system, and TGZ had the best membrane designs. Conch Ocean successfully married the two systems in a series of tests and trials and received required approvals from the classification societies. One of the problems inherent in this design, now the TGZ membrane, was the cost of the balsa panel system. The TGZ Mk II sought to replace balsa with PVC/ply sandwich panels and a simpler joint design, reducing cost without sacrificing safety. A later design (Mk III) used polyurethane foam blocks with a reinforced aluminum foil secondary barrier facing (Vaudolon, 2000). Figure 2.1 shows a typical membrane carrier. 2.2.2

Gaztransport Solution

Abandoning the cylindrical tank design used on the Jules Verne, Gaztransport developed a membrane design based on 36% nickel steel (Invar), which has a very low coefficient of expansion. This solved the temperature variation problems and the need for folds. Hull deflections and insulation supports and attachments remained as challenges. The secondary barrier would be identical to the primary, providing 100% secondary containment. The insulation method adopted for the support of the membrane was a series of subdivided plywood boxes filled with perlite granular/powder insulation—inexpensive but effective. Air could be purged from the boxes and nitrogen circulated with provision for sampling to detect cargo leakage.

DESIGNS AND ISSUES—FIRST COMMERCIAL LNG SHIPS

25

The maximum width of commercially available 0.5-mm Invar was about 0.4 m. This determined the width of both the boxes and the strips of membrane. It was possible for each strip to extend the complete length or width of the cargo holds. The first membrane carriers, the Polar Alaska and the Arctic Tokyo, entered service for Phillips Marathon from Kenai, Alaska to Negishi, Japan in November 1969 and March 1970, respectively. By now, the size was up to 71,650 m3 using 9% nickel steel, built in Sweden. Vaudolon (2000) reports that two incidents occurred with these ships: •

•

During Polar Alaska’s first return trip from Tokyo to Alaska, sloshing of a 15–20% LNG heel in the No. 1 tank (to provide cooldown prior to loading) caused part of the cargo pump electric cable support tray to break loose and to become washed back and forth across the aft end of the tank. This perforated the membrane in several places, causing liquid leakage into the secondary barrier space, which remained intact. Repairs took less than a week. Although carrying a heel in the tanks was discontinued briefly (until later proved safe), the Arctic Tokyo hit bad weather resulting in leakage again being detected in the No. 1 tank insulation space. It was subsequently found that local deformation of the membrane and supporting insulation boxes had occurred in the aft tank corners at the liquid surface (20% full) level. Leakage occurred at one of these points. A maximum 5% fill level was then imposed, and no further problems of this kind have been experienced on these two ships (DNV, 2006).

For 125,000-m3 capacity ships, problems with fatigue occurred at the points where the primary barrier tie rods passed through the secondary barrier at the tank corners. Existing ships were modified to provide a more flexible connection, but point penetrations have been eliminated in the new design. 2.2.3

Spheres

The sphere was one of the favored shapes for containers in the early conceptual design days but was met with disfavor by naval architects since they used enclosed hull space less efficiently, and if they were allowed to project through the deck, they destroyed the integrity of an accepted flat deck design (Vaudolon, 2000). Supporting a sphere inside a ship was a significant obstacle. The Norwegian Kvaerner Group began to explore the design possibilities of a spherical system. They had satisfactory experience with this design and also had very significant support from the Maritime Advisory Group of DNV, which provided in-depth structural design and analysis essential to the sphere. An 88,000-m3 LNG carrier design with spherical cargo tanks and no secondary barrier was introduced at the LNG 2 Conference in Paris in October 1970.

26

LNG INCIDENTS AND MARINE HISTORY

Figure 2.2 Moss sphere-type LNG carrier (SIGTTO website, www.sigtto.org; reproduced by permission from SIGTTO).

The method adopted for supporting the spheres was a continuous cylindrical, stiffened skirt attached to the equator by a special extrusion such that the sphere could freely expand and contract with minimum loads being applied to the shell. The skirt is welded integrally to the hull’s supporting structure and is therefore subjected to the ship’s deflections and is designed to absorb them. A Moss sphere LNG carrier is shown in Figure 2.2. 2.2.4

LNG Carriers for the Asian Trade

The first Asian producer, Brunei, began LNG shipments to Japan in 1972 in the Gadinia with a 75,000-m3 capacity. By 1977, Abu Dhabi began shipments to Japan with four ships of 125,000 m3 by Gotaas Larsen. Indonesia also started shipping LNG to Japan, first from Bontang, Borneo, in 1978 and then from Arun, Sumatra, in October 1978. At that point, the U.S. shipyards began building Moss-type ships, and by 1979, there were 23 carriers built or on order, mainly for the Indonesia to Japan trade. Japanese shipyards entered the LNG carrier business in 1982 when Kawasaki built the Golar Spirit. The LNG plant in Bintulu, Malaysia, began shipments to Japan in 1982 with five membranetype vessels of 130,000-m3 capacity. Australia entered the Japan LNG trade in 1987 from the Northwest Shelf project that required eight new carriers of 127,500-m3 capacity. The next wave of LNG expansion was evident in 1995 when Kvaerner delivered four new carriers and four more were built in Japan for the expansion in Bintulu, Malaysia. Qatargas built up a fleet of 10 ships starting in 1997.

LNG TRADE HISTORY

27

The world LNG fleet grew slowly at first, taking 33 years to reach 100 vessels by 1997 due more to market challenges than to technical problems, but reaching 255 vessels at the start of 2008 with an average cargo capacity of 127,000 m3. Data from E.A. Gibson Shipbrokers showed major construction activity added a further 60 vessels over the following 3 years to end 2010. The newer vessel sizes are also larger. In the period 2000–2007, 143 vessels entered the fleet all at or below 140,000 m3, whereas the 35 vessels entering service in 2007 averaged 153,000 m3 with the largest being a Q-Flex vessel of 216,000 m3 (the Al Gattara). Larger Q-Max designs are on order. In the older fleet, there were broadly similar numbers of spherical and membrane vessels, but the newer vessels are predominantly membrane, about 80% of the membrane design (by DNV). Construction times for LNG vessels can require up to 3 years, and due to the special materials and design requirements, there are only a few yards in Asia and in Europe able to construct such vessels, and order backlogs are currently significant. 2.2.5

Current State of LNG Tankers

LNG tankers carry radar, global positioning systems, automatic distress systems, and beacons to signal if they are in trouble. Cargo safety systems include instruments that can shut operations if they deviate from normal, and also gas and fire detection systems (Petroplus International N.V., 2008).

2.3 2.3.1

LNG TRADE HISTORY European Trade

As noted in Section 2.2 after an earlier start importing LNG from the U.S. gulf coast, commercial trade began in earnest from Algeria to British Gas’s Canvey Island terminal on October 12, 1964. The following year, France also began importing from Algeria, then Spain started in 1970 as did Italy in 1971. Then, the United Kingdom gained access to cheap pipeline natural gas with the first landing of North Sea gas from the West Sole field in 1967. As a result, LNG imports never grew substantially in the United Kingdom. Imports have rarely exceeded 1 billion cubic meters (bcm) per year. By 1983, France was importing nearly 9 bcm of LNG per annum, but then imports largely plateaued at that level with electricity production based predominantly on nuclear power. France now receives the majority of its gas via pipeline from Russia, the Netherlands, and Norway. Germany, the largest gas market in Europe, also relies on these three sources of natural gas and has made virtually no use of LNG. Similarly, Spain built up LNG import levels to a volume of 7 bcm in 1995, but since then, the Mahgreb gas pipeline from Algeria was commissioned, and LNG imports have dropped off. Italy, after early growth, has not taken significant quantities of LNG since 1980. Italy

28

LNG INCIDENTS AND MARINE HISTORY

receives cheaper pipeline gas from Russia, the Netherlands, and, most significantly, from Algeria via the TransMed pipeline. Altogether, largely because of pipelines, the LNG market in Europe has been less than that in the Far East (Greenwald, 1998). In 2006, the percentage distribution of LNG in Europe was the following: Spain, 46%; France, 27%; Belgium, 8%; United Kingdom, 8%; Italy, 6%; Portugal, 4%; and Greece, 1% (American Gas Foundation, 2008). 2.3.2

Asian Trade

The first cargo of LNG imported by Japan was from Kenai, Alaska, in 1969. As a consequence of the 1973 oil crisis, Japan set a goal of achieving 25% of its electricity output generated from gas. By 1977, LNG sales from Brunei to Japan reached over 7 million tons per annum (MMtpa). In that year, Abu Dhabi and Indonesia also became Japan’s suppliers. Malaysia entered the trade in 1983 and Australia in 1989, so Japan has successfully diversified its fuel suppliers. In 2007, Japan, Korea, and Taiwan accounted for 55% of world LNG demand, and the Pacific Basin accounted for 62% of world LNG trade (exceeding 110 MMtpa). Japan has largely reached its target for LNG penetration, and so the mantel has passed to South Korea and Taiwan as the new dynamic players in the field, providing the world’s fastest-growing LNG market (Greenwald, 1998). 2.3.3

Temporary Setbacks

The LNG industry suffered setbacks in the late 1970s. First, a potential project with Iran was canceled after its revolution. Prices collapsed in the United States in the 1980s after the deregulation of natural gas. Nine LNG carriers shipping from Algeria to the East Coast of the United States were withdrawn; three were scrapped, and six were converted to bulk shippers. The import terminals in Cove Point, Maryland and in Savannah, Georgia were also shut down until the late 1990s. Growth for LNG began in earnest in the Atlantic Basin area in 1999 when Trinidad began exporting to Spain and the United States. 2.3.4 Revival of LNG with Worldwide Supply–Demand Pinch of Petroleum By the mid-2000s, it became clear that LNG must play a larger role throughout the world. The industrialization of China and India generated a large increase in worldwide demand for oil, and oil production had failed to increase sufficient capacity in response. According to BP compilations, in 2006, natural gas accounted for 26% of world energy consumption and LNG accounted for 7%, compared to 63%, for oil. Even in Asia in 2005, natural gas accounted for only 11% of primary energy use, suggesting tremendous room for growth (http://

LNG TRADE HISTORY

29

60.0

Annual average (bcf/day)

50.0

40.0

30.0

North America Europe Asia

20.0

10.0

0.0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Figure 2.3 LNG global demand by region by 2016 (American Gas Foundation website; American Gas Foundation, 2009).

Globalgastrade.com/, 2008). Figure 2.3 plots historical and projected LNG demand by region, showing that as steady growth is expected in Asia, rapid increase is expected in North America, and Europe will continue to play a substantial role. Prior to the worldwide economic recession in 2008–2009, LNG imports to the United States were projected to grow at the high rate of 10% per annum over the decade 2010–2020. To satisfy a shortfall of as much as 6–7 teracubic feet (TCF) per year, imports may have to rise to more than 50 million tons per year (MMT/Y) by 2020. Figure 2.4 shows that the breakdown of LNG suppliers to the U.S. market is diversifying as new suppliers come online. Recent developments with onshore shale gas may reduce LNG import growth. 2.3.5

Supply History

Demand has historically driven LNG supply since projects are typically constructed only after nearly all of the LNG is sold under long-term contracts. Table 2.1 lists existing worldwide LNG liquefaction capacity as of 2008, projects under construction, projects in advanced stages of planning, and potential additional LNG projects through 2020. The areas are defined as •

•

Atlantic Basin: Algeria, Egypt, Equatorial Guinea, Libya, Nigeria, Norway, Trinidad (+potentially Angola, Venezuela); Middle East: Abu Dhabi, Oman, Qatar (+potentially Iran); and

30

LNG INCIDENTS AND MARINE HISTORY

14.0 Other Qatar Norway Nigeria Equatorial Guinea Egypt Angola Algeria Trinidad

Import volume (bcf/day)

12.0 10.0 8.0 6.0 4.0 2.0 0.0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Figure 2.4 Historical and projected distribution of U.S. LNG suppliers, 2007–2016 (American Gas Foundation website; American Gas Foundation, 2009, citing Altos World Gas Trade Model 2/2008 Base Case).

Table 2.1

Current and projected global LNG supply capacity

Area

Current, 2008 Under Planned Capacity Potential MMtpa Construction MMtpa (bcf/day) Additional (bcf/day) MMtpa (bcf/day) MMtpa (bcf/day)

Atlantic Basin Middle East Pacific Basin Total

75.5 (9.7) 46.5 (6.0) 66.9 (8.6) 188.9 (24.3)

21.6 (2.6) 53.5 (6.9) 30.8 (4.0) 101.2 (13.0)

40.5 (5.2) — 24.0 (3.1) 64.5 (8.3)

53.4 (6.9) 27.1 (3.4) 40.6 (5.2) 126.1 (16.1)

Source: Poten & Partners, Inc. (American Gas Foundation website; American Gas Foundation, 2008).

•

Pacific Basin: Australia, Brunei, Indonesia, Malaysia, United States, Alaska (+potentially Russia, New Guinea).

The data in Table 2.1 indicate that over a 50% increase in LNG supply will occur over the period 2008–2011 when projects under construction are completed. Almost another 100% increase will occur over the following decade, counting both projects in advanced stages of planning and some future potential projects. Figure 2.5 plots historical and projected LNG supply by country. 2.3.6 Some Economic Factors In 2005, local natural gas provided to the United States 60% as much energy as oil. Yet, in 2002, LNG supplied only 1% of U.S. natural gas (Park, 2003).

LNG TRADE HISTORY

50.0

Other Russia Malaysia Indonesia Brunei Australia Nigeria Angola UAE Oman Qatar Egypt Algeria

45.0 40.0 bcf/day, average

31

35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 07

08

09

10

11

12

13

14

15

16

Figure 2.5 Historical and projected LNG supply by country (American Gas Foundation website; American Gas Foundation, 2009, citing Altos World Gas Trade Model 2/2008 Base Case).

By 2006, LNG usage had grown to 2% of U.S. natural gas. The limited LNG import capability of the United States restricted the country’s access to abundant natural gas supplies worldwide and contributed to large spikes in domestic natural gas prices. Former Federal Reserve Chairman Alan Greenspan (2007) writes that in the United States, natural gas prices, seasonally adjusted, have historically been far more volatile than petroleum prices. He attributes this to “… in part, the relatively primitive state of global trade in natural gas –oil’s broader and more diverse markets tend to damp down wild swings in price.” Greenspan also states “Clearly, the gas industry has a long way to go before trading on a world market will be able to supply unexpected needs through a quick diversion of product from one country to another, thereby checking the swings in prices. In the end, such international price damping for natural gas will require a yet-to-be-developed broad spot market in LNG.” Important developments are changing the economics of the LNG industry. By 2010, Qatar will have 14 LNG trains (10 bcf/day) online. With their central location, supplying the Asia Pacific basin, India, Europe, and the United States, they may be able to link prices between the Atlantic and Pacific basins with unprecedented flexibility to exchange Atlantic for Pacific LNG supplies and vice versa. This would enable competition on a broader scale than heretofore. As cited in a Congressional Research Service report by Parfomak (2004), “If current natural gas trends continue, industry analysts predict that LNG imports [to the US] could increase 20% of total US gas supply by 2020” (Yergin and Stoppard, 2003).

32

LNG INCIDENTS AND MARINE HISTORY

2.4

LNG ACCIDENT HISTORY

LNG Marine Accidents Since 1959, tankers have carried over 33,000 LNG cargoes without a serious accident at sea or in port (Delano et al., 2003). Restated by Bainbridge (2003) and LNG World Shipping (2006) of LNG carriers, “There has never been a single incident involving the breach of containment systems that would result in cargo spillage from collision, grounding, fire, explosion, or hull failure.” This is confirmed in four grounding incidents that have not resulted in cargo loss: 1. 1974: The 27,400-m3 Methane Progress ran aground at Arzew, Algeria. The rudder was damaged. No LNG was released (CHIV, 2003). 2. August 14, 1978: The 123,890-m3 Khannur collided with the cargo ship Hong Hwa in the Strait of Singapore with minor damage. No LNG was released (ibid.). 3. June 29, 1979: The 125,000-m3 El Paso Paul Kayser ran aground at 19 knots under full load while maneuvering to avoid another vessel in the Strait of Gibraltar, and this may be regarded as a worst-case grounding. The bottom was damaged extensively and an LNG tank was deformed. The vessel was refloated and cargo was transferred to a sister ship, the El Paso Sonatrach. No LNG was released (ibid.). 4. December 12, 1980: The 125,000-m3 LNG Taurus ran aground in heavy weather at Mutsure Anchorage off Tobata, Japan. The bottom was extensively damaged. The vessel was refloated and proceeded under its own power to the Kita Kyushu LNG Terminal and discharged cargo. No LNG was released (ibid.). 5. 1985: The 126,000-m3 Ramdane Abane experienced a collision while loaded. The port bow was affected, but no LNG was released (ibid.). 6. May 21, 1997: The 125,000-m3 Northwest Swift collided with a fishing vessel about 400 km from Japan. There was some damage to the hull, but no ingress of water and no LNG was released (ibid.). In addition, there were incidents with LNG carriers at dock: 1. June 12, 1974: The Methane Princess was rammed by the freighter Tower Princess while moored at Canvey Island LNG Terminal, creating a 3-ft gash in the outer hull (ibid.). 2. December 1983: The 87,600-m3 Norman Lady was cooling down the cargo transfer arms prior to unloading at Sodegaura, Japan, when the ship suddenly moved astern under its own power. All cargo transfer arms sheared and LNG spilled, but did not ignite (ibid.). 3. February 1989: The 40,000-m3 Tellier at Skikda, Algeria, when wind blew the ship from its berth and sheared the cargo transfer arms. Piping on the ship was heavily damaged. Cargo transfer had been stopped. Some

LNG ACCIDENT HISTORY

33

LNG was released from the cargo transfer arms according to some verbal accounts (ibid.). 4. October 31, 1997: The 126,000-m3 LNG Capricorn struck a mooring dolphin at a pier near Senboku LNG Terminal in Japan. There was some damage to the hull but no ingress of water and no loss of LNG (ibid.). 5. September 6, 1999: The 71,500-m3 Methane Polar experienced engine failure during its approach to Atlantic LNG jetty at Trinidad and Tobago. Petrotrin pier was struck and damaged, but there were no injuries and no spill (ibid.). 6. 2002: The Norman Lady, running in ballast condition at sea east of the Straits of Gibraltar, collided with the U.S. nuclear submarine the USS Oklahoma City. The LNG ship suffered a leakage of seawater into the double bottom area below dry tanks (IELE, 2003b). Accident events at LNG facilities include 1. October 20, 1944: Cleveland, OH, USA. “In 1939, the first commercial LNG peakshaving plant was built in West Virginia. In 1941, the East Ohio Gas Company built a second such plant in Cleveland. The plant operated without incident until 1944, when the facility was expanded to include a larger tank” (IELE, 2003b). Three new tanks were built during World War II when metal rationing did not allow 9% nickel construction, so the tanks were constructed of 3.5% nickel. The tanks used cork and crushed peanut shell insulation. They did not have a retaining wall. The tanks became brittle when exposed to the extreme cold of LNG and failed shortly after they were placed in service. After one tank ruptured, the other two did as well and spilled LNG into the city sewer system. The LNG vaporized, ignited, exploded, and burned, killing 124 people and injuring 200–400 (reports by the U.S. Bureau of Mines, 1946; and NASFM, 2005). 2. February 1973: Texas Eastern Co., Staten Island, New York. At an LNG peak shaving import terminal with membrane-lined tanks, a tank leaked natural gas between the membrane and the tank wall. The tank was removed from service for entry of repair workers. Tears were found behind the mylar lining. During the repairs, nonexplosion-proof irons and vacuum cleaners were used in sealing the liner in exception to the operating procedures that called for explosion-proof equipment. Gas trapped behind the membrane ignited. The resulting fire raised the temperature in the tank, generating enough pressure to lift the 6-in. thick concrete roof dome, which then fell on workers in the tank, killing 37 people. The Fire Department of New York City (1973) report determined the event was clearly a construction accident and not an “LNG accident.” The report stated, “Although the exact causes may never be known, It is certain that LNG was not involved in the accident and the

34

LNG INCIDENTS AND MARINE HISTORY

3.

4.

5.

6.

surrounding areas outside the facility were not exposed to risk” (Office of Pipeline Safety cited by Wikipedia [2006] and IELE [2003a]). October 1979: Lusby, MD, USA. At the Cove Point LNG import terminal, LNG leaked through an inadequately tightened LNG pump electrical penetration seal, vaporized, and flowed through 200 ft of electrical conduit to the substation. Since natural gas vapors were not expected in this building, there were no gas detectors installed. A worker switched off a circuit breaker, igniting the gas vapors by the normal arcing contacts of the circuit breaker. An explosion killed one worker, seriously injuring a second, and caused about $3 million in damage to the building. The National Fire Transportation Safety Board found that the Cove Point terminal was designed in conformance with all appropriate regulations and codes. However, as a result of this accident, three major changes were made to National fire codes (NTSB, 1980; CHIV, 2003; Wikipedia, 2006). August 1985: Pinson, AL, USA. “At a peak shaving facility, the welds on an 8 ¼ inch by 12 inch ‘patch plate’ on a small aluminum vessel (3 ft diameter by 7 ft tall) failed as the vessel was receiving LNG being drained from the liquefaction cold box. The plate was propelled into a building that contained the control room, boiler room, and offices. Some of the windows in the control room were blown inward and natural gas escaping from the failed vessel entered the building and ignited. Six employees were injured” (CH-IV, 2003). August 29, 1987: Mercury, NV, USA. During the Falcon series Test #5 at the U.S. Department of Energy (DOE) test site (see Chapter 8, Table 8.1) after a sequence of relatively strong rapid phase transitions (RPTs), the vapor cloud ignited. The fire burned for about 30 s with a flame length 20 ft above the ground. The official explanation was that a spark generated by static electricity approximately 76 s after the spill was the most likely source of ignition. An independent investigation on behalf of Gas Research Institute showed that a more likely source of ignition was oxygen enrichment between the surface of the LNG pipe and the combustible polyurethane foam insulation. Oxygen enrichment occurred during the long cooldown period with liquid nitrogen that preceded the LNG test. Such enrichment had been previously observed during tests carried out by an LNG tank design and manufacturing company. Impacts during the RPTs may have ignited the insulation but not the nearby fuel-rich vapor cloud. However, when a smoldering insulation fragment was propelled outside the fence by an RPT, it ignited the portion of the cloud that was within the flammable limits” (CHIV, 2003). 1989: Thurley, United Kingdom, peak shaving facility. “While cooling down the vaporizers in preparation for sending out natural gas, lowpoint drain valves were opened. One of these drain valves had not been

SUMMARY OF LNG HISTORY AND RELEVANT TECHNICAL DEVELOPMENTS

35

closed when the pumps were started and LNG entered the vaporizers. As a result, LNG was released as a high pressure jet.” The vapor cloud ignited about 30 s after the release began. The flash fire covered an area approximately 40 × 25 m and burned the face and hands of two operators (ibid.) 7. January 19, 2004. Skikda, Algeria, at the Sonatrach LNG liquefaction plant. A refrigerant line leaked and vapors were drawn into the combustion air intake for a high-pressure steam boiler. The boiler fire box exploded and damaged other lines that precipitated a larger explosion outside the boiler and caused extensive damage to the facility. There were 27 workers killed, 80 injured; three LNG trains were destroyed, bringing production for 2004 down to 76% of normal for the year (NASFM, 2005; Wikipedia, 2006). LNG Truck Accident Planas-Cuchi et al. (2004) describe a boiling liquid expanding vapor explosion (BLEVE)-like event that followed an LNG truck accident in Catalonia, Spain, in June 22, 2002. An LNG truck believed to be speeding overturned and immediately caught fire. The fire became larger, fed probably by both diesel fuel and leaking LNG, and after 20 min, the tank exploded and a large fireball resulted. The driver died and two people suffered burn injuries 200 m away. Overpressure damage was evident, and large truck and tank fragments were thrown up to 260 m. Terrorist Attacks on Shipping No LNG vessel has been attacked by terrorists. However since this is possible two other marine terrorist events are described. 1. October 12, 2000: USS Cole was attacked in port of Aden, Yemen, by an explosive-laden boat. A shaped charge against the port side of the hull resulted in a 12 × 18 m (40 × 60 ft) gash at the galley. Sailors were lined up for lunch and there were 17 fatalities (Wikipedia, 2006). 2. October 6, 2002: the French double-hull crude oil carrier Limburg was attacked by an explosive-laden dinghy off Mukalla, Yemen, in the Gulf of Aden. The vessel was carrying 400,000 bbls and it lost 90,000 bbls of crude oil, one crew member killed, 12 injured, repaired for U.S.$8.5 million. Renamed the Maritime Jewel (BBC World News, 2009). 2.5 SUMMARY OF LNG HISTORY AND RELEVANT TECHNICAL DEVELOPMENTS The following synopsis puts the history of LNG developments and LNG accidents in perspective (after Coers, 2003):

36

LNG INCIDENTS AND MARINE HISTORY

• •

•

•

• • • •

•

• •

• •

• •

• • • • • • • • •

1944: Cleveland tank failure and fire National Aeronautics and Space Administration (NASA) (and its predecessors) develops cryogenic materials. 1950s: Refrigerated liquefied petroleum gas (LPG) and anhydrous ammonia come into wide use. 1959: LNG was transported from Lake Charles, Los Angeles to Canvey Island, United Kingdom Frozen hole LNG storage was developed. 1960s: NFPA 59A and API 620 were developed. 1967: Perlite insulation and Invar were used in membrane tanks. 1969: LNG shipments began, from Kenai, Alaska to Negishi, Japan, in membrane tanks. 1970: Sloshing damage was experienced on a Japan to Alaska voyage; maximum of 5% full set for heel in tanks to keep tanks cool on return voyage 1970: Moss sphere tanks were developed. 1970s: LNG terminals experience thermal overfill, rollover, and foundation frost heave; measures were added to avoid these problems. 1973: tank insulation fire in Staten Island, New York 1979: control room fire at Cove Point, Maryland; national fire codes changed. 1980s: LNG market collapsed in the United States. 1980s: British Standard 7777 was developed. In the United States, 49CFR Part 193 was adopted. 1987: Australia began LNG shipments to Japan. 1990s: LNG truck accidents occurred in the United States. 1990s: Gas detector development was strongly improved. 1990s: LNG articulated loading arms were developed. 1990s: European Union standard EN 1473 was developed. 1995: Wave of expansion began for LNG in Japan. 2000: In the United States, FERC and DoT accept NFPA 59A. 2004: explosion at a liquefaction plant in Skikda, Algeria Mid-2000s: China and India ramp up imports of crude oil. Large new LNG liquefaction plants built in Qatar, Nigeria, and Australia.

3 CURRENT LNG CARRIERS

As outlined in the previous chapter on liquefied natural gas (LNG) history, LNG carriers have been in operation since the Methane Princess and the Methane Progress entered service in 1964. The basic design options were standardized by the 1970s, and these have continued with few major changes other than to increase capacity. The majority of designs are based on the standard spherical and membrane designs illustrated in Figure 3.1. However, there are a small number of designs based on the Japanese self-supporting prismatic type B (SPB) design. Design requirements are established in the IGC code for several types of tank (types 1, 2, etc.). See section 3.1 for details regarding the IGC Code. While there is a clear trend toward increasing LNG vessel size to serve new fields in Qatar and larger consumer terminals, there is also a developing market for smaller-scale LNG vessels to deliver LNG as a transportation fuel due to its low carbon footprint. In addition to traditional, fully refrigerated LNG designs, there are many proposals for partially refrigerated and partially pressurized LNG and for compressed natural gas (CNG) carriers. These are potentially attractive for intermediate distance routes or for smaller field sizes—not justifying a pipeline or a full refrigeration plant. Most are designed to operate at 130–250 bar. Many design concepts have been presented including those by EnerSea (vertical steel pipes), Trans Ocean Gas (vertical composite pipes), CETech (horizontal steel pipes), Knutsen PNG (vertical steel pipes), and Coselle (horizontal steel coils), but none have been constructed as of 2008. LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

37

38

CURRENT LNG CARRIERS

Figure 3.1 Comparison of LNG spherical and membrane carrier profiles (from DNV Maritime Calendar illustrations).

All LNG carriers are full double-hull vessels because of ship structural requirements. This also provides significant additional strength and protection against collision and grounding events. On average, the interhull spacing is around 2 m and is used either for ballast or for void space. The structural members of LNG carriers consist of a mixture of mild steel and some high-tensile steels, just as in many other types of ships. These steels are fit for purpose but are not designed for direct exposure to LNG. They meet normal specifications for structural strength and fatigue. Should contact with LNG occur, the steel would be subject to embrittlement and possible failure. The unique portions of an LNG ship are its tanks, and these are designed to contain large quantities of cryogenic cargo in close proximity to the hull with minimal boiloff of cargo and no transfer of cryogenic temperatures to the hull. LNG carriers have special requirements for the cryogenic LNG cargo storage and pump-out, but they must also address normal issues of vessel structural integrity for dynamic sea loads and some special issues of tank integrity due to sloshing, primarily affecting membrane carriers. Newer requirements for improvements in atmospheric emissions, both for traditional emissions (SOx and NOx) and newer requirements for CO2 limits are also growing design factors. Some significant developments in technology occurred in the mid 2000s. Up until then, LNG cargo was transferred to and from vessels using fixed loading arms at terminals. In August 2006, the first ship-to-ship (STS) transfer occurred between the Excelsior to the Excalibur vessels in a demonstration test in the Gulf of Mexico. Flexible loading hoses for LNG also came into

MEMBRANE TANKS

39

production. The annual Offshore Technology Conference in Houston and the biannual Gastech International Conference provide regular updates on novel LNG developments. A further topic is floating marine terminals as these share many similarities to carriers but are permanently moored. There are projects for both LNG export and import terminals. Operationally, LNG vessels have an excellent safety record with no loss of cargo from the tank containment system in the full operational history since 1959. While accidents have occurred, the accident record for LNG carriers is 30–80% lower than for average shipping operations (Valsgard et al., 2006). 3.1

DESIGN REQUIREMENTS

The International Code for the Construction and Equipment of Ships Carrying Liquefied Gases in Bulk (IGC, 1975) is published by the International Maritime Organization (IMO). It applies to all carriers constructed after 1986. It is based on the original code issued by IMCO in 1975. The purpose of the code is to provide an international standard for the safe transport of liquefied gases in bulk by sea. The IGC design requirements are supplemented by international classification society rules for design basis and ongoing operational inspections. National regulators have also established requirements, but these are aligned with the IGC. The IGC identifies five types of gas carrier containment systems: (1) independent tanks, (2) membrane tanks, (3) semi-membrane tanks, (4) integral tanks, and (5) internal insulation tanks. Independent tanks may be of types A, B or C. The majority of vessels are either of Moss spherical tank design (independent tanks), or membrane tanks of either Technigaz (TGZ) or Gaztransport design. A few vessels use an SPB (self-supporting prismatic shape IMO type B tank) design built by IHI in Japan. Independent tanks do not require secondary containment except for a small catchment plate, whereas membrane tanks all include a full secondary containment membrane. Details of the two main design types follow. 3.2 3.2.1

MEMBRANE TANKS Tank Design and Insulation

Membrane carriers are normally of two main types: the Mark III or GT No. 96 designs, with a small number of the newer combined system (CS 1) alternative. These designs are currently owned by one company, Saipem. The original membrane concept was invented by DNV in 1962 and was later taken up by TGZ in France. The Mark III TGZ design consists of an 18/8 stainless steel (18% chrome/8% nickel) membrane bifolded in

40

CURRENT LNG CARRIERS

Figure 3.2 TGZ membrane design (spacing approximately 300–400 mm) (reproduced by permission of Gaztransport & Technigaz).

Mark III

Figure 3.3 Mark III tank internal arrangement (reproduced by permission from Gaztransport & Technigaz).

both longitudinal dimensions to allow thermal expansion in all directions (see Fig. 3.2). These panels are welded together in a large prismatic shape as shown in Figure 3.3. Details of the insulation and additional containment membrane are provided in the next section.

MEMBRANE TANKS

41

Figure 3.4 GT No. 96 membrane tank internals (reproduced by permission of Gaz Transport & Technigaz).

The Gaztransport membrane employs Invar, a very low expansion high nickel content (36%) steel alloy. This does not require folding, and this design employs long parallel sheets (Fig. 3.4). The newer Combined System (CS 1) uses reinforced polyurethane foam insulation and two membranes, the first one 0.7 mm thick made of Invar and the second made of a composite aluminum glass fiber called triplex. The system has been rationalized to make assembly easier and is prefabricated allowing quick assembly on board. This design has suffered from secondary membrane leakage problems, and the number of ships built to this design is small. 3.2.2

Dimensions and Capacity

Until the new large developments in LNG capacity came onstream starting in the late 2000s, the standard size of an LNG membrane carrier was 125,000– 140,000 m3 of cargo capacity. More recently, especially related to the very large gas discoveries in Qatar, new, larger capacity membrane vessels called Q-Flex and Q-Max were developed by Qatar Petroleum and ExxonMobil. Q-Flex stands for Qatar and flexibility—a design enabling delivery to about 65% of existing LNG ports—and Q-Max stands for Qatar and the maximum vessel size that could dock at the Qatar port facility, and these may have access to approximately 50% of current LNG ports. The Q-Flex and Q-Max designs have a capacity of around 217,000 and 266,000 m3 of LNG cargo, respectively.

42

CURRENT LNG CARRIERS

Table 3.1 Membrane vessel dimensions

Type

Typical Conventional Membrane

Reference Cargo capacity (m3) Cargo tanks Overall length (m) Breadth (m) Draft (m) Engines

Lee et al. (2008) 138–173,000 4 277–290 43.3–45.8 11.5 Single or twin

Q-Flex

Q-Max

Lee et al. (2008) 210,000 5 315 50.0 — Twin

ExxonMobil 250,000 5 345 55.0 12.0 Twin

The Q-Max design is the current largest design, and in dimensions, it approaches conventional very large crude carriers (VLCCs) (Table 3.1). They operate at significantly higher efficiency compared with older designs as LNG boil-off gas (BOG) is reliquefied rather than used in the ship’s propulsion system, thus delivering nearly the full initial cargo load. On a typical voyage from Qatar to the United States, up to 5% of the cargo could otherwise boil off. Rather than using natural gas fuel, the new vessels use two low-speed marine diesel main engines, each of 21,770-kW power running at 91 rpm. The largest operator will be Nakilat, the Qatar Gas Transport Company, and it will have 54 modern vessels including 9 conventional LNG carriers, 31 Q-Flex and 14 Q-Max vessels, with the capacity to deliver 77 million tons annually from the Qatar gas fields. The Qatar fields are estimated to contain 15% of proven global gas reserves in 2008. The combined effect of larger vessels and higher natural gas prices has led ship owners to specify different propulsion systems. Lee et al. (2008) notes several alternatives with a switch away from purely LNG BOG to higherefficiency dual-fuel diesel electric and two-stroke diesel engines. Larger LNG carriers above 210 km3 are using two-stroke slow-speed diesel engines with on-board BOG reliquefaction. Gas turbine systems are novel for LNG vessels but will be widely used for floating storage and regasification units (FSRUs) and LNG floating production storage and offloading (FPSO). Typical dimensions for the cross section of a membrane tank are given in Chapter 5, Section 5.7.2 and are used in subsequent example cases.

3.2.3

Tank Materials and Insulation

The membrane tank has two main design alternatives, with flat or folder membranes as presented earlier. The membrane wall thickness is usually low, of the order of 1 mm or less, and thus the primary strength for the containment comes from the insulation and hull structures. While wall thickness is relatively thin, it is still physically strong and rigid due to the stainless steel material. Steels get stronger as they are cooled, and the tank material is selected

MEMBRANE TANKS

43

Top bridge pad Corner panel Primary barrier

Secondary barrier scab Flat joint Plugs Flat panel

Double hull

Secondary barrier Invar membrane strake

Standard flat area

A

B C

D

Coupler Insulation box Double hull

Figure 3.5 Insulation system designs for (a) Technigaz and (b) Gaztransport membrane vessels (Vaudolon, 2000) (reproduced by permission from Gaztransport & Technigaz).

to avoid embrittlement issues associated with ordinary or lower-nickel-content steels. The insulation systems used have been in use for many years and the design is now well established. This is a system of plywood boxes filled with insulating material such as small perlite balls. A comparison of the insulation designs for the two membrane types is shown in Figure 3.5. Both of these designs incorporate two insulating layers and an intermediate secondary barrier between the inner membrane wall and the inner hull of the double-hull vessel. While intricate to construct, the design is very effective, and boiloff in newer vessels is under 0.2% per day.

44

CURRENT LNG CARRIERS

Tanks are fitted with an internal tower (e.g., Fig. 3.4) where submerged LNG pumps are located for pump-out. This design precludes the need for any liquid connections below the top of the tank where a leak could empty the tank under gravity flow. 3.2.4

Pressure and Vacuum Relief

The IGC and marine society classification rules specify that tanks must be provided with pressure and vacuum relief systems. The pressure relief is usually determined by fire exposure. While a fire scenario directly impacting the inner membrane is hard to conceive, the pressure relief system is designed based on an assumed 10–20% direct impingement. This is standard in the oil and gas industry for fire exposure of insulated tanks. However, in order to have direct fire exposure, the outer and inner hulls must be breached; the two insulation layers and secondary containment must be removed or torn away. With this degree of exposure, so much structural strength must have been removed that the tank might be in a direct structural failure mode—not due to fire exposure. This issue is discussed further in the fire consequence section of this text. As well as the relief valves, a secondary route for pressure relief in an emergency is provided by the BOG collection system. All tanks are interconnected for BOG collection for delivery to the engines or to a reliquefaction system. In the event of an overpressure in one tank, this egress route is also available for vapors, and they can be relieved out of other tank pressure relief systems. Vacuum relief primarily protects the tank against collapse due to LNG pump-out with a lack of natural gas return during unloading. These open when the pressure falls below a predetermined level, sufficient to protect the membrane structure. 3.2.5

Design Issues

The larger operational size of newer LNG vessels means historical data may be an uncertain guide to their structural and containment performance. This has led to greater reliance on modern structural design computational codes for ultimate life and fatigue. Design challenges include • • • • •

sloshing in LNG membrane tanks, alternative propulsion systems, vibrations, hull fatigue, and operations in cold climates (arctic operations).

Sloshing refers to cargo fluid motions in sea states leading to damaging inside tank wave loads particularly in prismatic tanks at corners and other loca-

MEMBRANE TANKS

Figure 3.6

45

Example of sloshing loads (Lee et al., 2008).

tions. Some historical membrane damage has occurred due to part loaded cargo operations. Most ocean-going membrane vessels have fill restrictions of less than 10% or more than 70% to avoid sloshing damage. An example numerical simulation of sloshing loads is given by Lee et al. (2008) in Figure 3.6. Sloshing experiments have been carried out using test rigs using water and air media in small and large scales (1:10) compared to the prototype. Valsgard et al. (2006) describes this work, the most recent of which uses six degrees of freedom platforms to simulate real ocean motions. Computational fluid dynamics (CFD) codes have been used to predict loads, benchmarking off the test rig data. Modeling challenges arise from the continuous boiloff in LNG tanks, meaning small bubbles are present and these affect the gas liquid interface, which generates the sloshing loads. It is possible that a cushioning effect occurs when the liquid strikes sharp knuckles and corners in the tank as the liquid phase is more compressible because of bubbles. Simulating this in CFD poses mesh definition problems due to the small scale of bubbles relative to tank dimensions. Valsgard notes that scaling from test rigs to vessel dimensions is difficult. Froude number scaling is common for inertia-dominated motions, but for compressibility, surface tension and viscous effects are also important. Starting in 2008, a full-scale trial is underway on board an LNG vessel with highly instrumented fiber optic pressure sensors on a tank. Additional environmental data, ship motions, and stresses are being collected to allow full calibration of CFD codes. Until this is complete, rule development is based on trial data. Currently, the only definitive guide for sloshing loads is a DNV class notation (DNV, 2007). Given proper definition of loads, the membrane response also needs to be evaluated. The IGC requires that this address both the ultimate strength (ultimate limit state [ULS]) and the fatigue life. Special analysis of the pump tower is also required. The hull design must also be analyzed in substantial detail for LNG carriers. The DNV CSA-2 classification notation specifies a range of design calculations that must be implemented to address the yield, buckling, and fatigue strength of the hull (Lindemark et al., 2006). This involves specialist codes for each

46

CURRENT LNG CARRIERS

aspect, and each of the main classification societies has its own approach. However, all now involve sophisticated finite element analysis and can involve both deterministic and stochastic calculations. Lindemark concludes that these analyses showed additional strengthening of the hull was required compared to conventional design rules, both with respect to yield and buckling strength and for fatigue.

3.3

MOSS SPHERES

The Moss sphere design originated in Norway between 1969–1972 involving close collaboration between the Kvaerner group and the DNV classification society. It used four to six free-standing spheres made of thick alloy aluminum corresponding to the IGC type B rules (see Section 3.1). Support for the sphere is from a central “equatorial ring,” which transmits its load to the hull through a full circular skirt. The skirt is made of steel, and a special connection between the aluminum equator and the steel limits transfer of cold to the skirt. A special connection is required to avoid welding problems between the aluminum tank and the steel skirt. A diagram showing the design is provided in Figure 3.7. More recent spherical vessels have been constructed in Japan, Korea, the United States, and Finland and typically employ four or five spherical tanks with 125,000+ m3 total capacity. Because calculations can show that it is impossible for a crack in any circumstances to propagate into a catastrophic failure, it was admitted by clas-

Tank cover of steel Insulation Aluminium tank shell Pipe tower: dome on top, foundation at bottom Structural transition joint (aluminum-stainless steel) Thermal brake of stainless steel Support skirt of high-tensile steel Ship’s double steel hull Water ballast tank

Figure 3.7 LNG spherical tank in cross section (reproduced by permission of Gaztransport & Technigaz).

MOSS SPHERES

47

Figure 3.8 LNG carrier with Moss sphere (IELE, 2003a).

sification societies and was confirmed by the IMO IGC that the spherical tank type B does not require a full secondary barrier. However, a partial secondary barrier, a drip tray, is constructed at the bottom of the hold to prevent damage to the ship’s structure should cargo drip from a limited crack. An example illustration of a Moss ship design is shown in Figure 3.8. As the LNG sphere is self-supporting, it is very strong. While the tank operates at −161°C and atmospheric pressure, it actually has a pressure capability of close to 8 bars. The hull configuration follows the tank shape and allows only a small gap all around the tank; there is no large void space as might be imagined if a spherical tank were inserted into a cubical space.

3.3.1

Typical Dimensions and Capacity

For many years, the common Moss vessel size was 125,000 m3 with five equal spherical tanks, but there are variants with fewer or greater numbers of tanks. More recently, larger Moss type carriers have been constructed, up to 140,000 m3. These designs are not yet as large as the Q-Flex and Q-Max designs described before for membrane carrier designs. However, proposals include designs that insert a cylindrical element into the center of the sphere, thereby increasing capacity. Typical Moss tank dimensions are given in Table 3.2. Spherical design vessels are higher than equivalent membrane vessels, and this can affect windage drag in strong winds. The windage effect can be assessed using marine simulations, and speed and tug requirements, for example, can be established for safe transits of narrow channels. While modern navigation systems are not dependent on visual sightings, the high upper deck does reduce the visibility from the bridge. A facet of the spherical tank design is that while the tank surface area is minimized for a spherical design, the cargo utilization of the hull volume is a

48

CURRENT LNG CARRIERS

Table 3.2 Typical dimensions for Moss spherical carriers

Dimension Cargo capacity (m3) Cargo tanks Overall length (m) Width (m) Draft (m)

DNV Study

DNV Study

Mitsui Press Release

125,000 5 293 41.6 11.7

138,000 4 270 45.0 11.5

140,000 4 290 48.5 11.3

little less efficient than the prismatic membrane design. This leads to an anomaly that the gross registered tonnage (GRT) tonnage of Moss spherical vessels is greater than for equivalent capacity membrane vessels. Under current Suez Canal rules, this means that Moss design vessels are charged higher fees for transit than a similar membrane vessel. 3.3.2

Insulation and Tank Materials

Insulation is provided by polyurethane or phenol resin foams attached by bolts to the tank surface. An exterior aluminum sheet protects the insulation. As for the membrane design, insulation reduces boiloff, and this has been reduced from around 0.2% per day to closer to 0.1% in recent designs. Typical wall thickness for a sphere varies from 57 mm at the equator (where the skirt joins) to 29 mm at its thinnest. In comparison, a membrane tank might have a wall thickness of up to 1.2 mm. A single aluminum LNG spherical tank of 36,000 m3 weighs about 700 t. The cargo contents at a specific gravity of 0.45 would add about 16,000 t. 3.3.3

Pressure and Vacuum Relief

Relief requirements for spherical tanks are similar to those for membrane tanks and are specified in the IGC Code and in classification rules. These conform to normal oil industry practice with a set point of 0.25 barg. Pressure relief for LNG tanks is based on near to atmospheric pressure. However, the spherical tank is physically very strong and is capable of withstanding up to 8-bar internal pressure. Thus, while the pressure relief is designed to limit internal pressure to under 0.3 barg in a fire situation, in fact, the tank has a far greater margin for internal pressure. It is in practice hard to conceive of major direct fire exposure to a spherical tank due to the tight constrictions of the hull space in which the tank is located. The fire would have to destroy the insulation as well to create significant heat input to the tank cargo and its shell. 3.3.4

Design Issues

As for membrane vessels, the spherical carriers must be designed to cope with full sea states. Lindemark et al. (2004) notes that vessels with closed cross

MOSS SPHERES

Loads

ULS

49

FLS

Drawings, loading manual, applicable scatter diagram, etc.

Modeling

Hydrodynamic model

Identical mass model

Global structural model

Submodel

Hydrodynamic analysis Analysis Hydrodynamic/static design loads

postprocessing

Reporting

Figure 3.9

Global structural analysis

Ultimate strength analysis (yield, buckling)

Transfer of displacements/ forces

Local analysis using Submodel technique

Fatigue analysis

Reporting

Computational ship analysis flow chart (Lindemark et al., 2004).

sections (e.g., tankers and membrane LNG vessels) can, in most cases, be analyzed using midship models only. However, open-type vessels (e.g., container vessels and spherical LNG carriers) require a full numerical model to establish torsional and warping response. Lindemark provides a block diagram in Figure 3.9 showing the typical computational analysis required for an LNG carrier design showing aspects for loads, ultimate load strength (ULS), and fatigue limit state (FLS). A general design objective for spherical tanks is that fatigue cracks should never initiate, any crack that does initiate should not penetrate, and if the crack does penetrate, that it will not grow and the leak will be small. On this basis, there is no requirement for a secondary containment system, only a catch plate for any small leak to avoid cryogenic fluid touching the inner hull.

4 RISK ANALYSIS AND RISK REDUCTION

Risk analysis, in simple terms, is analyzing the potential for loss. Managing risk is a prime responsibility of managing a project or managing a company. Quantitative risk analysis (QRA) is the process of providing management with event scenarios and the quantification of the likelihood and magnitude of potential losses. The emphasis here is upon losses by accidents, but upon beginning a new venture, those who provide the capital for the venture want to know the likelihood of a wide variety of possible losses, including • •

•

•

•

•

Market Risk Losses that occur if the demand declines for the product. Currency Risk Losses from foreign purchases and sales could occur due to an unfavorable shift in the exchange rate. Property Rights Risk Will a foreign government allow our investment to deteriorate by high taxation, low protection against fraud, and upholding of contracts? Natural Losses due to Hurricanes, Earthquakes, Tornadoes, and so on Called acts of God, but natural events often have considerable history on record. Business Interruption Risk Loss of production due to serious mechanical failure, labor instability, political instability, and so on. Losses by Accidents Ship collisions and grounding, process equipment loss of containment by corrosion, operating mistakes, random failures of components, and so on.

LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

50

BACKGROUND

•

51

Deliberate Sabotage and Acts of Terrorism These fall outside of random event distribution.

Besides direct losses to the investors and owners, the enterprise can pose a safety risk to the community. The issues in this case include •

•

Community Benefit/Risk Ratio Is this ratio favorable to the community? Litigation Potential Will unreasonable demands be placed on the company and will injured parties be fairly compensated?

Risk management has a primary role in risk reduction, which can involve a variety of strategies, including •

•

•

4.1

Measures to Reduce the Frequency of Accidents Add speed limits, add barriers, increase equipment inspection frequency, and so on. Measures to Reduce the Consequence of Accidents Provide personal protective equipment (gloves, safety glasses, fire-resistant clothing), reduce the inventory of hazardous materials, offset the plant from residents, add water spray deluge systems, and so on. Insurance How much insurance should we buy and what is a fair price for insurance? BACKGROUND

Risk assessment is now considered an integral part of engineering design, either required by local regulations (e.g., Europe, Australia, Hong Kong, parts of the Middle East) or carried out voluntarily by proponents to assure a reasonable balance between facility cost, risk, and mitigation alternatives. Several textbooks describe the risk assessment process in general (Turney and Pitblado, 1996; Center for Chemical Process Safety [CCPS], 2000a; Mannan, 2005). Conventional engineering approaches, based on design codes drafted by committees working in consensus mode, are very well suited to capture lessons from more frequent accidents, or rare major accidents if they have actually occurred (e.g., the Cleveland liquefied natural gas [LNG] tank failure accident). Risk assessment is used to address the gap for less likely major accidents that may not have occurred and to eliminate, reduce, or mitigate their outcomes by enhanced design or operation. Risk assessment methods have been developed as an engineering discipline since the mid-1970s particularly to address facilities that pose major accident risks, such as LNG facilities, major refineries, offshore oil installations, or nuclear plants. Risk is also being used to a greater extent in transportation activities such as rail (European Committee for Electrotechnical Standardization [CENELEC] standards 50126, 50128, and 50129), maritime (International Maritime Organization

52

RISK ANALYSIS AND RISK REDUCTION

[IMO] formal safety assessment approach), and aviation (European safety case approach). The earliest risk studies were carried out in the United States in the mid1970s for the nuclear industry (Rasmussen, 1975) and for LNG developments in California (Eisenberg et al., 1975). But the United States did not push forward due to perceived legal problems; notably, a liability case involving the Ford Pinto car rejected a risk-based decision approach that used cost–benefit analysis to assess gasoline tank design and location. Consequently, European efforts have driven risk technology developments for the past 25 years. Major European risk demonstration projects included the Canvey Island risk assessments in the United Kingdom (Health and Safety Executive [HSE], 1978, 1981b) and the Rhine harbor area assessment in the Netherlands (Rijnmond Public Authority, 1982). Both the Canvey Island and Rijnmond studies included LNG facilities as well as other hazardous process items. Other early studies that occurred a little later in Hong Kong and in Australia applied to large liquefied gas storages and entire petrochemical complexes, respectively. Several of these early studies are summarized in Kolluru et al. (1996). Risk assessment employs common language terms but with specific meaning, and some key terminologies are defined below (after Institution of Chemical Engineers [IChemE] and CCPS): Hazard A physical situation with the potential for human injury, damage to property, the environment, or some combination of these. Major Hazard An imprecise term for a large-scale hazard, usually with the potential for significant human impact. Often characterized as a highconsequence low-frequency event. Risk The combination of frequency and the consequence of a specified hazardous event. Risk Analysis The systematic use of available information to identify hazards and to estimate the risk to individuals or populations, property, or the environment. Risk Assessment The overall process of risk analysis and risk evaluation, usually comparing the risk analysis estimate against a target risk criteria and applying mitigations until the risk is sufficiently low.

4.2

RISK ANALYSIS PROCESS

Risk analysis is normally carried out using a formal process. The main steps are hazard identification, frequency analysis, consequence assessment, risk summation, risk assessment, and risk management. Figure 4.1 summarizes those steps. These steps may be carried out qualitatively or quantitatively, and there is a place for both approaches as well as semiquantitative approaches falling between (e.g., bow tie barrier methods and layers of protection analysis [LOPA]). General methodologies for risk in the process industry are provided

RISK ANALYSIS PROCESS

53

Study objective acceptance criteria Design facilities safety systems procedure assumptions Hazard identification

Cause and frequency analysis

Consequence evaluation

Risk summation

Options to decrease frequencies

No

Risk assessment risk controlled?

No

Options to mitigate consequences

Yes Optimize options to manage risk Reporting Figure 4.1 Steps in risk analysis.

in CCPS (2001a), Turney and Pitblado (1996), IChemE (1994), and more specifically in engineering standards (e.g., EN 1473, 2007, also designated as BS EN 1473, 2007) for LNG, Australia AS/NZS 4360:2004 for general risk, and AS 3961:2005 for LNG (Australian Standard AS/NZS 4360, 2004; Australian Standard AS 3961, 2005, among several). Skramstad et al. (2000) outline risk analysis approaches specific to LNG carriers. An important subset of risk assessment is the more limited consequence assessment, which omits the detailed frequency assessment and replaces this instead with nominated events, that in the opinion of the authors is representative of realistic cases. NFPA 59A for the siting of LNG facilities uses this approach. It considers some very serious events—10-min release of the largest line, but not, for example, catastrophic failure of the LNG cryogenic tank. Both these types of events have occurred for cryogenic liquefied gases as well as for more common process fluids. Thus, the consequence approach embod-

54

RISK ANALYSIS AND RISK REDUCTION

ies judgment from standards committee and regulatory agencies to a greater degree than risk assessment, which in principle should consider all events but rely on demonstrated low likelihood to make the largest events fall below some tolerability threshold. A possible unintended outcome of this is that locations subject to NFPA 59A still can use single-walled LNG cryogenic storage tanks, whereas, for example, European or Australian locations, which employ risk assessment, invariably require double-containment designs as single-containment designs do not meet the risk criterion (Australian Standard AS/NZS 4360, 2004; Australian Standard AS 3961, 2005; BS EN 1473, 2007). There are different ways to execute a QRA—either in a stepwise approach with an ability to vary the depth of treatment typically based on interim consequence results or in a more specified approach that applies a predetermined depth of treatment, matching local regulatory requirements or a company’s own procedures. The stepwise approach is documented in CCPS (2000a). It starts with a formal identification of potential events followed by consequence assessment. If the consequences are of sufficient magnitude to impact vulnerable locations (either on-site or off-site), then the depth of treatment is expanded to include a full likelihood analysis; otherwise, the study may stop with a consequence only analysis. The consequence approach would not meet requirements for a full QRA, but it is close to the NFPSA59a approach. The reason that risk assessment approaches have been gaining ground over the period from 1980 is that they provide more insight to owners and regulators as to the potential events that might occur along with their consequences and likelihoods, and how the various safeguards implemented either prevent or mitigate each event. Safeguards can include facility spacing (both internally and toward external vulnerabilities), LNG process facilities and cryogenic tank design specifications, design and construction standards, operation and safety policies, and emergency response. Risk criteria (e.g., CCPS, 2009a) allow determination on whether these safeguards are sufficient or whether modifications are necessary to meet some risk target or to reduce risks below that to a point as low as reasonably practicable (ALARP). 4.2.1

Hazard Identification

This is normally carried out for a risk assessment using a formal process such as a Hazard and Operability Study (HAZOP) or a what-if checklist and which builds on knowledge of historical accidents and the team’s operating experience. These methods and others are described in detail in several references (Mannan, 2005; CCPS, 2008; etc.) These methods apply equally well to onshore LNG production facilities, ship transfer, and LNG reception and regasification terminals. At their simplest, these methods challenge the detailed design with the whole range of possible deviations from normal operation and confirm how the design responds and whether the current safeguards are sufficient. Deviations may be a single mechanical deviation (e.g., more pressure, less

RISK ANALYSIS PROCESS

55

temperature, corrosion), human error (causing or responding to a deviation), or combinations of events (e.g., control system fault during power failure). An example of hazard identification applied to LNG is shown in Table 4.1. It is a truism to note that most major accidents have human cause elements and often involve multiple failures. For this reason, failure identification methods more focused on single mechanical or electrical failures such as failure modes and effects analysis (FMEA) are less useful for major accident identification. If the hazard identification process does not address these more complex causes, it is likely to miss important accident events. Extensions to this basic identification assign best-estimate likelihood and probability estimates to determine a rough risk ranking for each deviation. Beyond this, layer of protection analysis (CCPS, 2001a), which is a simplified QRA method, establishes whether sufficient safeguards are in place to meet a predetermined risk target or whether additional layers (e.g., new safeguards or improved safety reliability by incrementing the specified safety integrity level (SIL) of safety instrumented systems as per standards: IEC 61511 or ANSI S84.01). The HAZOP on its own is insufficient for a QRA. A further step to identify specific events for risk modeling is necessary. Most often, this is done by a methodical process—conceptually breaking every line and vessel into a range of characteristic hole sizes. While there is no specification of what these sizes must be, there are common sizes specified in various regulations (e.g., Environmental Protection Agency [EPA] risk management plan [RMP] regulations suggest 50 mm and 10-min worst-case release). A fuller assessment for an LNG facility, which has many large pipes, would include both smaller and larger sizes. Thus, this might include •

• •

•

10 or 25 mm—typical of smaller leaks highly likely to occur in a plant lifetime; 50 mm—serious leak event, may not occur in a single plant lifetime; 100–250 mm—very serious leak, unlikely to occur in a single plant lifetime, but will have been observed in historical records of LNG facilities; and Full bore rupture (FBR)—catastrophic FBR failure, extremely low frequency event, but more likely for pipework than for the cryogenic tank. Cryogenic tank failure is still conceivable if multiple safeguards fail simultaneously.

The HAZOP would be used to better define these events, to understand how they might occur, and whether specific defenses exist that might reduce or increase the likelihood of occurrence. Some further specification of events is necessary to complete the enumeration of potential hazard events. The most important is the duration of the leak. This will be determined initially by the flammable gas detection system, which will likely be linked to an isolation or an emergency shutdown (ESD) system

Collision of a non-LNG vessel onto the LNG jetty/trestle

Causes

Interruption of business due to damage of the jetty or trestle

Consequences

Likelihood Medium

Severity High

Table 4.1 Extract from typical HAZOP of LNG facility

Medium

Risk Ranking

3. Tugs available to assist ships to minimize collisions, as needed (if LNG vessel is present) 4. Emergency shutdown of the LNG pipelines

—

A tug is the considered one of the more likely vessels to strike the trestle. Consider a load calculation for a trestle strike from a tug, and design the structure to handle this load. Consider conducting a trestle/vessel collision analysis. — 1. Jetty located within a regulated/monitored water way

2. Navigation lights on the jetty and dolphins

Recommendations

Safeguards

56 RISK ANALYSIS AND RISK REDUCTION

FREQUENCY: DATA SOURCES AND ANALYSIS

57

to rapidly shut in the section with the leak and thus to limit its pressure and inventory. Such systems can operate reliably in 3–5 min depending on the number and distribution of detectors, weather conditions, and so on. Manual response systems with operators initiating isolation in response to gas alarms are likely to require 10 min or more depending on the location of the leak, its size, and the training of the operator. While this may seem slow, historical data show operators are very cautious before shutting down a process facility. For serious leak events, a risk assessment might split such an event into further events with different response times and probabilities assigned to each event (e.g., 50-m leak 2-min manual response, p = 0.1; 5-min response, p = 0.2; 10min response, p = 0.5; very delayed response, p = 0.1). The leak event may need to be further defined if the LNG system has a blowdown capability so that the inventory in the isolated section is both depressured and deinventorized. This would significantly affect the release rate of LNG and hence its potential consequences. 4.3

FREQUENCY: DATA SOURCES AND ANALYSIS

A risk assessment requires frequency information to match the event specification and consequence calculation. Most often, leak frequencies in risk assessments are estimated from generic historical leak statistics. This approach works well for common items (pipes, vessels, pumps, etc.) that have a high population in the process industry and hence sufficient leak experience to allow meaningful statistical estimates. Some items do not have statistically valid data. This might be for items with extremely low failure frequencies for which there is no historical failure event (e.g., double-walled LNG cryogenic storage tank failure, LNG vessel major leak) or items with unusual failure modes not represented in generic statistics. For these, a modeling approach such as fault tree analysis (FTA) is generally recommended. 4.3.1 Generic Data Approach The LNG industry is insufficiently large to have its own specific failure frequency statistics, and risk assessments commonly use the rigorously compiled hydrocarbon release database (HCRD) data set developed by the U.K. Health and Safety Executive (HSE) in cooperation with the U.K. North Sea oil and gas industry (Pitblado et al., 2009). This data set includes leaks of all sizes recorded in a standardized way and normalized by an equipment count to allow data to be reported as leaks per year for process equipment or leaks per meter-year for pipework. Spouge (2006) identifies means to screen this data to remove zero pressure or limited leaks (e.g., leaks during maintenance) and to obtain the leak frequency for a full-pressure hydrocarbon release from the nominated hole size. The formula for pipework derived from data collected between 1992 and 2003 is Ffull = 8.0 × 10 −6 (1 + 1000 D−1.3 ) d −1.42

(4.1)

58

RISK ANALYSIS AND RISK REDUCTION

Table 4.2 Frequencies of full leaks (per equipment item year) for process equipment

Equipment Type

Frequency of Full Leaks ≥1 mm Diameter

Frequency of Full Leaks ≥50 mm Diameter

Steel pipes (2″)—1-m length Steel pipes (6″)—1-m length Steel pipes (18″)—1-m length Flanged joints (2″) Flanged joints (6″) Flanged joints (18″) Manual valves (2″) Manual valves (6″) Manual valves (18″) Actuated valves (6″) (non-pipeline) Instrument (0.5″) Process vessel Centrifugal pump Reciprocating pump Centrifugal compressor Reciprocating compressor Heat exchanger (h/c in shell) Heat exchanger (h/c in tube) Heat exchanger (plate) Heat exchanger (air cooled) Filter

5.7E − 05 2.0E − 05 1.1E − 05 3.2E − 05 4.3E − 05 1.2E − 04 1.4E − 05 4.8E − 05 2.2E − 04 2.6E − 04

0.0E 7.7E 4.2E 0.0E 3.6E 1.1E 0.0E 4.9E 2.3E 1.9E

2.3E − 04 5.0E − 04 1.8E − 03 3.7E − 03 2.0E − 03 2.7E − 02 1.4E − 03 1.0E − 03 6.0E − 03 1.2E − 03 8.9E − 04

0.0E + 00 1.1E − 04 2.4E − 05 5.2E − 04 2.0E − 06 1.1E − 05 1.3E − 04 4.9E − 05 3.6E − 04 6.9E − 05 6.4E − 06

+ 00 − 08 − 08 + 00 − 07 − 06 + 00 − 07 − 06 − 06

h/c, hydrocarbon fluid.

where Ffull is the cumulative frequency of full flow rate leaks ≥ d (per meter per year), D is the pipe diameter (mm), and d is the hole diameter (mm). Applying this formula to a sample 150-mm pipe would give a frequency for a 10+ mm hole of 7.7 × 10−7/m-year and for a 50+ mm hole of 7.7 × 10−8/m-year. While these appear as small numbers, actual leak rates are based on multiplying these by the total length of relevant pipework. A tabulation covering different equipment types is provided in Table 4.2. Since this is cumulative data for all leaks, the frequency of holes in a specific size range (e.g., 50 mm represented by leaks between 30 and 70 mm would be obtained by subtracting the cumulative frequency of the 70+ mm value from the 30+ mm value). Commercial software (the LEAK program from DNV) that uses this data set to estimate leak frequencies from any item in any size range is available. 4.4

FREQUENCY: PREDICTIVE METHODS

Where generic data are not available or are considered inappropriate, the predictive techniques such as FTA and event tree analysis are used.

FREQUENCY: PREDICTIVE METHODS

4.4.1

59

FTA

The FTA method was developed initially for the nuclear industry and was summarized by Vesely (1981), and this method underpins the probabilistic risk assessment (PRA) method required to be applied in the U.S. nuclear industry after the Three Mile Island accident in 1979. The National Aeronautics and Space Administration (NASA) decided to apply a similar style of analysis after the Challenger Accident in 1986 indicated a lack of sufficient insight from the more qualitative FMEA approach. The method is described in detail in the NASA Fault Tree Handbook (Vesely et al., 2002) and in simpler terms with process industry examples in CCPS (2000b) and TNO (1997a,b). The particular advantage of the FTA approach is that it can provide a quantitative estimate of top event likelihood/frequency even for events that have never occurred. It does this by combining all the possible basic failure modes (e.g., corrosion, collision, dropped object, and human error) and by tracking how these initiating events can propagate up to the top event by progressive failures of safeguards or by combining with other events. This is done with logical AND and OR gates. Boolean algebra is used to determine the quantitative top event frequency. A further benefit of the FTA approach and CCPS (2000b) suggests this is the more important output and is the qualitative structure of how major accidents can occur and how various safeguards intervene to prevent failure. This qualitative analysis is termed cut set analysis, and the minimal cut set is the smallest number of events that must occur to lead to the top event. Usually, the cut set is a combination of an initiating event and the failure of intervening safeguards. In safety assessment, minimal cut sets with only one or two events are serious as this implies, without any quantification, a poorly safeguarded system (in this case an initiating event and zero or one failed safeguards), whereas a three- or four-event cut set implies better protection (an initiating event and two or three failed safeguards). This qualitative analysis can still yield useful safety information, even where data on basic event causes and safeguard reliability are not well known. FTA can also show the potential for common cause failures. If some of the events modeled in an FTA are dependent, then the top event frequency can be much higher (i.e., worse) than expected. For example, if for an LNG ESD system if there are two gas detection systems but they use the same hardware and are maintained by the same person using the same equipment, if one detector fails on demand, then there may be a good chance the other will also fail on demand. For this reason, a better design might use both fixed gas detectors and open-path beam detectors as these are two independent technologies and are less likely to fail from a single common cause. Normally, FaultTree software is used for the calculation of top event frequencies, cut set development, and common cause identification. There are many commercial codes, including FaultTree+, CAFTA, RiskSpectrum, Item, and Relex.

60

RISK ANALYSIS AND RISK REDUCTION

An example application for a 100,000-m3 LNG membrane storage tank is provided by Kim et al. (2005). They provide a top event of a major LNG release from the storage tank and associated pipework. One example mechanism for tank overfilling is provided in Figure 4.2, along with an explanation of fault tree symbols. The authors analyzed a total of six mechanisms to establish the total leak frequency (Fig. 4.3). Using this approach, Kim et al. finds that the total frequency of serious leak from the membrane storage tank is made up of six contributions (Table 4.3). The fault tree approach provides not only a top event frequency of 5.2 × 10−5 per year, but it shows by observation that a rupture of the outlet line (event 6 here) is the dominant mechanism. Further examination of the tree shows how the safeguards deployed can fail and thus where extra safeguarding may be required, either by enhancing currently installed safeguards or by adding new ones. This provides much greater insight than is available from generic data. However, if the fault tree cannot obtain realistic data for its base events, then the generic data may be more accurate as the data set should include all observed failure modes, even very unusual ones. 4.4.2

Event Tree Analysis

The event tree approach is used to track outcomes from a specified initiating event and thus it complements FTA. Event trees start with a specified initiating event (e.g., major leak of LNG) and branch outward based often on binomial choices on possible outcomes (e.g., ignition yes/no). Event trees are simpler to construct and are less subject to error than fault trees. An example application to LNG is provided by Vanem et al. (2008), who considered LNG carrier risks under the EU 6th Framework Project titled SAFEDOR. They presented a generic, high-level risk assessment of the global operation of ocean-going LNG carriers. The major contributions to the risk associated with LNG shipping were found to arise from five generic accident categories, that is, collision, grounding, contact, fire and explosion, and events occurring while loading or unloading LNG at the terminal; and of these, collision risk was found to be the highest. An example event tree for fire and explosion is provided in Figure 4.4. The event tree clearly shows the routes by which consequences can occur and how various factors affect the outcomes. In this event tree, the initiating event is fire or explosion somewhere on the vessel. The source for this number might be from historical records or from a more detailed FTA. Each branch thereafter examines the event as it progresses toward a consequence affecting some of the 30 crew assumed—in this case, between 0 and 16 fatalities. Each branch follows the standard convention of yes upside and no downside, and the event tree shows the probability determined. The sum of both branches must total to 1. If a particular branch is nonapplicable, it is shown in the diagram as a straight line and is assigned a value of 1.

TKLSHH13102A

OF-HL

TKLI131012H

5.00E−02

TKLSH131013H

Operator fails to respond to high level alarm.

Intermediate

Transfer symbols

External or house event

Basic Event

Intermediate Event

AND Gate

OR Gate

The transfer symbol indicates that the fault tree is developed further on another page. The symbols are labeled using numbers or a code to ensure that they can be differentiated. Transfer symbols are often used to avoid repeating identical logic in several places in a fault tree model.

A condition or an event that is assumed to exist as a boundary condition for the fault tree.

A component failure that requires no further development. A basic envent is the lowest level of resolution in a fault tree.

A fault event that results from the interactions of other fault events that are developed through logic gates such as those defined above.

The output event occurs only when all the input events exist simultaneously.

The output event occurs if any of the input events occur.

Figure 4.2 LNG tank overfilling event mechanism (Kim et al., 2005) (reproduced with kind permission of Springer Science and Business Media).

1.00E−03

TKLI131012F

8.76E−03 1.00E−06/h 8.760 h

Operator fails to correctly observe level indicator.

1.00E−03

TKLSH131013A

LSH-131.01 and 131.03 fail to actuate.

L1-131.01 and 131.02 fail to indicate true level to operator.

OF-IL

TK-FL

2.40E+01

Operator fails to respond to increasing level.

OF-OHL

OF-IL

Tank is being filled.

Operator fails to respond to high level.

Tank level increased during filling operation.

1.00E–03

LSHH-131.02 fails to actuate to close unloading lines.

Tank level reaches high level.

OF

Major LNG release from storage tank by overfilling.

FREQUENCY: PREDICTIVE METHODS

61

62

RISK ANALYSIS AND RISK REDUCTION

Major LNG release from storage tank and associated pipe TK

Major LNG release from storage tank by overfilling

OF

Major LNG Major LNG release from release from storage tank storage tank by by overpressuriz- underpressurization ation OP

Major LNG release by failure of inlet lines

Major LNG release by failure of outlet lines

Major LNG release by loss of mechanical integrity of tank

IN

OUT

TKB

VA

8.76E–06 1.00E–09/h 8760 h Figure 4.3 LNG failure top event frequency (Kim et al., 2005) (reproduced with kind permission of Springer Science and Business Media). (Note: Abbreviations are shorthand labels used by the authors for the longer text immediately above.) Table 4.3

LNG leak frequency contributions

1. Mechanical integrity failure 2. Tank overfilling 3. Tank overpressurization 4. Tank underpressurization 5. Rupture of inlet line 6. Rupture of outlet line Total frequency, all causes

8.8 × 10−6/year 1.2 × 10−5/year 6.5 × 10−7/year 2.9 × 10−10/year 2.6 × 10−6/year 2.8 × 10−5/year 5.2 × 10−5/year

The frequency of each arm is the product of all the conditional probabilities between the initiating event and the specific consequence. The worst consequence example here is 16 fatalities and its risk is the product of Initiating frequency = 0.0019 events year Cumulative condition probabilities product = 0.03 × 0.5 × 0.9 × 0.15 × 0.1 × 1 × 1 = 2.03 × 10 −4 Consequence = 16 fatalities This gives a risk from this specific route of 5.8 × 10−6 fatalities/year. The figure shows the total risk from fire and explosion onboard the LNG carrier to be 6.7 × 10−4 fatalities/year or, as it is commonly known, the potential loss of life (PLL) statistic. This must be summed with other event contributions as shown in Table 4.4.

0

0.1

1 0

0.1

0.15

1

0

1

1

1

0

1

1

1

1

1

0.37

0.989

0.37

0.37

0.989

0.37

0.37

1

1

16

1

1

8

1

5.76785E–06

1.34865E–06

7.64235E–06

3.20436E–07

1.4985E–07

8.4915E–07

1.34865E–06

7.64235E–06

1.4985E–07

8.4915E–07

Risk contribution

Sum risk

0.000672088

1 0.00010656 0.00010656

1 0.00053946 0.00053946

3.6049E–07

1.3487E–06

7.6424E–06

4.0055E–08

1.4985E–07

8.4915E–07

1.3487E–06

1 1

7.6424E–06

1.4985E–07

8.4915E–07

Frequency

1

1

1

Consequence

Fatalities per ship year

No

Yes

LNG carrier fire and explosion event tree (Vanem et al., 2008) (reproduced by permission from Elsevier Science Publishers, Inc.).

16

1

1

8

1

1

1

1

1

0.37 0.37

1

0.37

0.37

1

1

1 0.9

0.85

1

1

0.9

0.15

1 1

1

1

0.15

1

1

1

0.85

1

0.85

0.15

Probability of # of fatalities fatalities among crew

Evacuation model

0.37

0.9

0.1

0.9

1 1

0.85

1

LNG hazard Survivability model model

Accommodation area 0.16

0.5

0.5

Cargo leakage model

Machinery space 0.81

Figure 4.4

0.0018

Compressor room 0.03

Fire protection model

Firefighting No leakage no pool Surviving systems fire of LNG successful

0.1

Loading condition model

Fire and explosion In ballast at port Fire/ distribution (no LNG) explosion

Fire/explosion model

A crew of 30 is assumed

FREQUENCY: PREDICTIVE METHODS

63

64

RISK ANALYSIS AND RISK REDUCTION

Table 4.4 Potential loss of lives from LNG carrier operations per ship year

Accident Category Collision Grounding Contact Fire and explosion Loading/unloading events Total

PLL (Fatalities/Year) 4.42 2.93 1.46 6.72 2.64 9.74

× × × × × ×

10−3 10−3 10−3 10−4 10−4 10−3

The event tree clearly shows the routes by which consequences can occur and how various factors might mitigate the event.

4.5

CONSEQUENCE MODELING

Consequence assessment is the main part of this book and models for different outcome types—dispersion, flash fire, pool fire, jet fire, and boiling liquid expanding vapor explosions (BLEVEs) are presented in separate chapters and are not repeated here. A good overview of potential consequences is given in Turney and Pitblado (1996) and is reproduced in Figure 4.5. This shows in block diagram form the many potential consequences that are possible from a loss of containment event. An event tree of the type described above would need to be employed to determine which of the models was relevant. For LNG, in most cases spills will be for unpressurized releases with no toxics present, but some releases will be pressurized (e.g., in LNG transfer systems). Scenarios are further developed in Chapter 10, Section 10.2.

4.6

IGNITION PROBABILITY

A factor influencing consequence is ignition. Flammable materials are usually of minimal hazard if not ignited (in the case of LNG, only asphyxiation and freeze burns would be relevant and these apply only locally to the release). If the release is immediately ignited at the release point, then outcomes will be jet fires or pool fires. If the ignition is remote (and hence delayed in time), then flash fire and vapor cloud explosion (VCE) are possible, as well as jet fire and pool fire when the flame eventually propagates back to the source. A flammable cloud has a probability of ignition that depends both on its flammable extent and the number, strength, and duration of coverage of local ignition sources. There is no good theoretical model for this, but several empirical formulae have been developed to estimate ignition likelihood. At the simplest, these are single-value estimates and, at the other extreme, they

IGNITION PROBABILITY

65

Hazardous substance

Unpressurized leak

Pressurized leak

Other incident

Discharge Gas

Two-phase

Liquid pool

Liquid

Explosion • Condensed phase

Flash and rainout

Dispersion Jet

Fireball

Jet fire

Dense

Flash fire

Toxic effect

• Runaway Rxn • Dust • Physical

Evaporation

External fire

Neutral

VCE

Pool fire

Thermal effect

BLEVE

Explosion effect Blast and fragment

Outcome Figure 4.5 Potential consequences from release of hazardous substance (Turney and Pitblado, 1996) (reproduced by permission of IChemE).

are predictive equations. It is often observed that in many flash fire or vapor cloud events, the specific ignition source could never be identified, either because there are so many possible candidates or because no obvious candidate exists but ignition did occur. In these latter cases, often ignition is assigned to static electricity as this can exist on poorly grounded equipment or can be generated by the friction of the release itself.

66

RISK ANALYSIS AND RISK REDUCTION

Moosemiller (2009) developed a list of “hazardous characteristics” that correlate with historical explosion frequencies in various types of plants. These hazardous characteristics are the following: • •

• • • •

Process is corrosive or erosive. Process operates at temperatures that are close to the operating limits of the metallurgy used—either very high (but below the autoignition temperature) or very low. Potential exists for runaway reaction in reactors. Process is cyclic (batch/semibatch) or requires frequent shutdowns. Process contains a significant amount of volatile materials. Process contains a large number of equipment items.

LNG processing plants are listed in the lowest category of ignition potential: very low frequency (zero to two hazardous characteristics, 3 × 10−5 explosions/year) along with only storage units and out-of-service units. Moosemiller lists many possible sources of ignition. Some sources of ignition are perhaps surprising, such as “Non-sparking tools which contact oxygen-bearing materials, such as rust, can produce a spark which burns hotter and lasts longer than an ‘ordinary’ spark” (Canadian Centre for Occupational Health and Safety [CCOHS] website; CCOHS, 2008). With such findings, the use of “non-sparking tools has largely been discontinued in industry since they are not entirely safe (European Community [EC] website; EC, 2008). Ignition probability is a key parameter used in risk assessments. In consequence assessments, such as those mandated by the EPA RMP rules or in NFPA 59A, ignition is assumed and thus has a probability of 1. The difference between these two rules is that the EPA RMP regulations require a dispersion end point of the lower flammable limit (LFL), while NFPA 59A requires ½ LFL. The latter is an attempt to address dispersion model uncertainty as even the best dispersion models are thought to have an uncertainty of a factor of two either way (Britter, 1991). It is common to differentiate immediate ignition from delayed ignition. Immediate ignition precludes formation of a vapor cloud as the event will be a fire event at the source, and it therefore reduces the likelihood of a remote ignition. Typical values for immediate ignition used in the oil industry are shown below: • • •

Release caused by high-energy mechanical impact = 0.3 Average value = 0.2 Other causes = 0.1

The Netherlands uses the DNV SAFETINL risk package, and this uses a timeof-exposure ignition model. The methodology is described in an government specification (RIVM, 2009). The basic formula is given by

IGNITION PROBABILITY

67

Table 4.5 Probability of ignition of a flammable cloud during a time window of 1 min for a number of sources (RIVM, 2009)

Source Type

Ignition Source

Probability of Ignition

Point source

Adjacent process installation Flare Oven (outside) Oven (inside) Boiler (outside) Boiler (inside) High-voltage cable (per 100 m) Motor vehicle, train Ship Households (per person) Offices (per person)

0.5 1.0 0.9 0.45 0.45 0.23 0.2

Line source

Population source

a

0.5 0.01 0.01

a

Some example ignition effectiveness values from RIVM (2009) include the following: Effectiveness of the ignition for one person is 0.168 × 10−3/s. • Effectiveness of the ignition for one vehicle or a nonelectrified railway is 8.51 × 10−3/s. • Effectiveness of the ignition for electrified railway line is 26.8 × 10−3/s. •

P ( t ) = Ppresent × (1 − e − ωt )

(4.2)

where P(t) the probability of an ignition during the time window 0 to t; Ppresent the probability that the source is present when the flammable cloud passes by; ω the effectiveness of the ignition (s−1); and t time (s). The probability of ignition for a number of categories is provided in Table 4.5. The exact meaning of probability of ignition can vary. Spencer and Rew (1997) and Spencer et al. (1998) provide correlations of ignition with the area of a vapor cloud. In this sense, their correlation is based on the probability density of different types of ignition source. As discussed in Chapter 12, there is controversy as to whether the LFL or the ½ LFL contour should be used in regulations such as NFPA 59A that set the required area of an LNG terminal. The debate in this case is whether inhomogeneous pockets of gas above the LFL exist in areas where the average concentration is ½ LFL. In this regard, Fay and Lewis (1975) analyzed instrument trace records from the Esso Matagorda Bay tests spilling LNG on seawater. They plotted the probability non-ignition against peak-to-mean concentration ratios. A best-fit line through their data provides an estimate that at a peak-to-mean value of two (where the peak would be at the LFL and the mean at the ½ LFL), the probability of non-ignition is 0.8. Since by definition, the LFL measured in laboratory tests is the concentration at which on average half of the

68

RISK ANALYSIS AND RISK REDUCTION

tests (those above the LFL point) ignite and half (those just below the LFL point) do not, the LFL represents a 50% probability of ignition. In early tests (AGA, 1974), LNG was spilled on land in a diked area. The report concludes that peak-to-mean concentrations were in the range of 2–3 for neutral stability and close to 1 for stable conditions. To be complete, ignition can depend on the speed of flowing gas as well as on concentration. This is shown by Birch et al. (1989) for a gas jet in a cross wind. High velocity at the centerline of a jet, for example, prevents ignition. In the lee edge of a jet bending in a crosswind, ignition occurs well below the LFL.

4.7 4.7.1

RISK RESULTS Risk Presentation

Risk may be reported as in the event tree example above as a PLL statistic. However, there are other common formats. The Rijnmond Public Authority (1982) study recommended two graphic forms to convey risk result—individual risk contours and societal risk F-N curves. The definition of these and their calculation is discussed more fully in CCPS (2000a, 2000b). The CCPS definitions are the following: Individual Risk The risk to a person in the vicinity of a hazard. This includes the nature of the injury to the individual, the likelihood of the injury occurring, and the time period over which the injury might occur. Societal Risk A measure of risk to a group of people. It is most often expressed in terms of the frequency distribution of multiple casualty events. Thus, the two forms of risk are different and express different aspects about a given hazard, although they are calculated from the same underlying risk results. Individual risk represents the risk to a specific person at a specific place by summing all events (e.g., an individual in a nearby house or a staff member in a control building), whereas societal risk represents the risk to all the people who might be affected in every possible event and is plotted in cumulative form. An example applied to the LNG industry is taken from a public environmental impact statement, published on the Hong Kong Environmental Protection Department Environmental Impact Assessment (EIA) website (DNV, 2007). This project, although probably not proceeding for commercial reasons, shows the application of risk assessment to an LNG terminal and provides risk diagrams for both individual risk (Fig. 4.6) and societal risk (Fig. 4.7). It is important to note that risk contours and F-N curves are highly dependent on the specific project proposal and nearby population,

RISK RESULTS

69

Individual risk contours Project facilities shown in white and risk contour for 10–5 pa individual risk shown in gray. The contours surround the LNG jetty and storage tanks but do not extend to populated areas. The Hong Kong government risk criteria requires that this level of risk (10–5 pa) not intrude into residential areas.

1 × 10–5 per year Figure 4.6 Example of individual risk contour for LNG jetty and terminal (Hong Kong Environmental Protection Department EIA website).

1.00E–02 145,000 m3 LNG carriers 215,000 m3 LNG carriers

Frequency of N or more fatalities (per year)

1.00E–03

Unacceptable (per HK EIAO) 1.00E–04

1.00E–05 ALARP (per HK EIAO) 1.00E–06

Societal risk F-N cumulative plot The FN plot shows risk to groups of people and compares this to Hong Kong government societal risk criteria—also shown on the plot. The risk is very low and is primarily due here to catastrophic failure of the LNG storage tanks. The ALARP zone refers to as low as reasonably practicable—a tolerated zone of risk, only if all reasonable measures have been taken to reduce or mitigate the risk.

1.00E–07 Acceptable (per HK EIAO) 1.00E–08

1.00E–09

1

10

100 1000 Number of fatalities (N)

10,000

Figure 4.7 Example of societal risk curve for LNG jetty and terminal (Hong Kong Environmental Protection Department EIA website). EIAO, Environmental Impact Assessment Ordinance.

70

RISK ANALYSIS AND RISK REDUCTION

and the results below are not applicable to other sites without recalculation. This is of value of risk assessment—it shows project- and site-specific impacts and allows trying various mitigation measures to achieve a specified level of risk. Figures 4.6 and 4.7 show the quite different information available for risk decision making. Individual risk is plotted over a map, and it is obvious if specific houses or staff-occupied buildings are impacted at the risk level shown by the contour. Based on individual risk criteria, it can be judged if the risk is sufficiently low. Similarly, using societal risk, if the risk curves fall below the criteria lines, the risk is considered tolerable. Hong Kong has its own criteria, and CCPS (2009a) has provided a detailed survey of global risk criteria and their background. There are differences in the way risk assessment is applied in the United States. The focus is more on consequences and frequencies separately. 4.7.2

Risk Decision Making

The purpose of risk assessment is to reduce risks. The starting point would be engineering codes or codes of practice and standard good engineering practice. The risk assessment applies this already good design and generates specific hazards from the design and its operation. Frequency and consequence modeling are applied to establish risks to staff and the public taking account of local conditions (e.g., weather, ignition sources, population exposures). Where criteria are breached then options may be taken to reduce the risk (e.g., by increasing spacing or by reducing tank inventories) or mitigated (e.g., enhanced flammable gas leak detection and ESD). Many countries use an ALARP concept in addition to a maximum tolerated risk. This was developed by the U.K. HSE and was discussed in HSE (1981a, 2001). This allows a cost– benefit approach to assist in the decision making. ALARP balances the costs of reducing risks in terms of time, money, or effort against the risk reduction obtained, and when the costs become disproportionate to the benefit, then the risk is said to be ALARP. Sometimes, a value of a life approach is used to apply the test, and values of $2 million to $5 million per life are typical. The HSE triangle (Fig. 4.8) is used to explain the concept. Risk is represented as the width of the triangle and toward the top—the risk is judged unacceptable and measures must be taken to reduce the risk. Toward the bottom, the risk is judged broadly acceptable and no further risk reduction is necessary. A broad central region, often two orders of magnitude of risk, is termed tolerable, provided risks have been reduced ALARP.

4.8

SPECIAL ISSUES—TERRORISM

Terrorism is a recent, increasing important concern (Krane, 2000). A deliberate act of sabotage or terrorism is a planned event rather than a random event.

Increasing individual risks and societal concerns

RISK REDUCTION AND MITIGATION MEASURES FOR LNG

Figure 4.8

71

Unacceptable region

Tolerable region

Broadly acceptable region

HSE framework for the tolerability of risk (HSE, 2001).

Even so, some degree of threat analysis can be made. Threat analysis involves estimating the target desirability from the point of view of the terrorists, coupled with an estimate of the likelihood of success (level of defenses). Sandia (Hightower et al., 2004) explicitly addressed terrorism risks in its LNG marine risks report. They assessed more the consequence aspect directly, but commented in general terms on the terrorism likelihood.

4.9

RISK REDUCTION AND MITIGATION MEASURES FOR LNG

Mitigation measures to reduce risk can apply to reducing consequence or to reducing the event frequency. Typical measures discussed in following chapters for consequence reduction are fire barriers, fast detection and response, moving the source away from sensitive areas, and so on. Typical measures to reduce event frequency are LNG ship traffic control, escort vessels, speed limits, stronger storage vessels, and so on. For each of the hazards in Section 1.2, the LNG industry has developed a number of mitigation measures that improve safety. Risk assessment allows which of these and how many might be applied in specific circumstances. These are briefly listed below, and some are developed in subsequent chapters. This type of information is captured in industry standards and government regulations for LNG. Mitigation to Avoid Ship Collision and Allision • • •

•

Provide escorts and pilots in harbors. Maintain separation distances with other vessels in harbors. Use navigation systems with sophisticated radar systems that display other vessels and obstacles. Enforce speed limits in harbor areas.

72

RISK ANALYSIS AND RISK REDUCTION

•

•

Apply plans required by marine authorities including exclusion zones for approaching craft. Develop emergency response procedures and provide training.

Mitigation after Ship Collision or Weapons Attack at Sea •

• • • •

Use double-hull vessels (required for structural reasons but provides safety benefit) Keep moving to limit buildup of the LNG pool and spread it out. Move the carrier away from sensitive areas. Design pilot houses to survive flame engulfment. Design pressure relief and vacuum relief valves for large-scale events.

Mitigation for Pump and Pipe Leaks under Pressure • •

• • • •

Provide automatic cutoff valves for line leaks. In particular, fast-acting double-valve powered emergency release connection (PERC) for loading arms. Provide LNG impoundment basins in safe locations. Insulate drainage troughs under transfer lines. Drain troughs to an insulated sump to reduce evaporation rates. Provide shielding around flanges, valve stems, and pump axles.

Mitigation for Process Leaks • • • • • •

Minimize flanged connections. Select appropriate materials not affected by brittle fracture. Install a flammable gas leak detection system. Install sufficient isolation points to avoid large inventory losses. Install an ESD system. Install an LNG drain and impoundment basin at a safe location.

Mitigation for Storage Tank Leaks • • • •

Employ a full- or double-containment design. Install a flammable gas detection system around the tank. Install remote isolation systems to shut valves at a safe location. Install an LNG drain and impoundment basin at a safe location.

Mitigation for Storage Tank Rollover • • •

Maintain circulation with pump loop to prevent stratification. Provide sensors to detect stratification. Sample incoming LNG to test for possible stratification tendencies.

RISK REDUCTION AND MITIGATION MEASURES FOR LNG

73

Mitigation to Avoid Freeze Burns • •

Install remotely operated valves. Require adequate clothing and personal protective equipment for workers.

Mitigation to Avoid Asphyxiation • •

Keep pumps and transfer lines at terminals in open areas. Do not place personnel in low areas where vapors could flow and accumulate.

Mitigation from Radiation Burns from Flash Fire and Pool Fires •

•

•

Provision for adequate separation distances to public areas around LNG terminals and docks. Require adequate clothing and personal protective equipment for workers. Provide radiation shields for areas with personnel.

Mitigation for Rapid Phase Transitions •

Build strong vessels and transfer equipment capable of withstanding small overpressure.

Mitigation for Sloshing Cargo Damage in Transit •

For membrane tanks, run only with LNG levels near empty or full.

5 LNG DISCHARGE ON WATER

The main liquefied natural gas (LNG) vessel designs of Moss spheres and membrane tanks are all double hulled and very strong. These vessels are treated in Chapter 3. This is reflected in zero accidents resulting in loss of LNG cargo in 57 years (Chapter 2). Some small pipe leak events have occurred during transfer operations and these have led to localized hull cracking events. This highlights one of the issues of cryogenic materials—spillages can cause embrittlement of normal marine steels and allow residual stresses to cause cracking (see Section 5.5.4). Both large and small discharges are discussed here, first describing the events that can lead to each type of discharge (Sections 5.1–5.3). These three sections indicate qualitatively that large discharge events are justifiably rare. Much design and operational effort goes into assuring this rarity. Subsequently then, in Sections 5.5–5.7, we quantify the consequences over a wide range of postulated breach sizes and conditions. This is to bracket the range of potential consequences, not implying that their occurrence is remotely likely. LNG discharges can be envisaged from three distinct types of events: 1. accidental collision, grounding, or allision damaging the vessel structure and puncturing the containment; in this context, allision is a marine term referring to collision events with solid objects, such as bridges, or wharves; 2. pipework or loading arm failure allowing up to full pumping rate discharge; and LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

74

LNG DISCHARGE ON WATER

75

Category I Category II Category III

Figure 5.1

Types of LNG leak location (Luketa-Hanlin et al., 2008).

3. intentional events due to terrorism often involving weapons, but potentially deliberate events of types 1 or 2. Concerning discharge rate, if the tank is punctured, then the outcomes from similarly sized holes will be similar from accidental or intentional causes. Sandia (Luketa-Hanlin et al., 2008) terms three puncture zones, types 1, 2, and 3, respectively, as illustrated in Figure 5.1. In each case, the LNG outflow and water inflow are determined primarily by the size of the hole and the liquid head of LNG or water above the hole. •

•

•

Type 1 LNG spills from holes above the waterline penetrate to some depth in the sea. There could be cascading damage to the vessel interhull gap and structures due to exposure to cryogenic LNG. Type 2 LNG from holes at the waterline would initially be pulled, along with seawater into the double-hull area, and subsequently when the hull space fills, LNG flows over the surface of the sea. Type 3 Punctures below the waterline develop back pressure by compressing the remaining air space. Compression acts to limit further inflow rates of seawater and LNG. Ice formation in the double-hull space is possible. Water inflow into the LNG tank is possible but would contribute to pressure rise in the tank, opposing further inflow.

A variant of type 1 is a pumped discharge from the unloading pipework (on the top deck). This can be up to full flow, either accidentally or intentionally (Chapter 6, Section 6.3). Each case would be expected to generate significant LNG/water interactions with the potential for rapid phase transition (RPT) effects as LNG flashes to vapor. In each case, LNG spilled into the interhull gap could poten-

76

LNG DISCHARGE ON WATER

Table 5.1 Expected range of breach sizes for LNG carriers

Reference

Event Type

Equivalent Diameter or Area

Sandia (Hightower et al., 2004)

Accidental event Intentional events

FERC-ABS Consulting (2004) Fay (2003) Pitblado et al. (2004)

Not specified — Accidental event Intentional event —

0.5–1.5 m2 (net after partial blockage by initial event) 5.0–7.0 m2 (range 2–12 m2) up to three tanks damaged Range: 1.0–5.0 m 5.0 m 0.7 m 1.5 m 5.0 m

Quest Consultants, Inc. (2003)

tially lead to metal embrittlement, but in type 2 and type 3 spills, the lower structures would be protected against cryogenic damage by water ingress. These possibilities raise the issue of the potential for cascading events. Cascading events include sloshing forces, ignition explosions in hull chambers, RPTs, fire stresses on hull metal, metal embrittlement failures, flame envelopment of the pilot house, and boiling liquid expanding vapor explosions (BLEVEs). These are treated in Section 5.4. Issues of ignition are left to the discussion on fire outcomes in Chapter 9. Both collision and intentional events have large energy dissipation, and ignition is likely in both cases when air is available. Grounding events are below the waterline, and ignition there is less likely but still possible due to air available in the double hull. An indication of the variability of predicted breach sizes is provided in Table 5.1. 5.1 5.1.1

TYPE 1—ABOVE WATER BREACHES AT SEA Ship-to-Ship Collisions

Several recent LNG studies have used arbitrary hole sizes (Fay, 2003; Quest Consultants, Inc., 2003; ABS Consulting, 2004; Hightower et al., 2004; Pitblado et al., 2005) as a starting point for the assessment. These are typically in the range of 0.75- to 5.0-m diameter. There have been a series of international conferences on collision and grounding of ships (starting in San Francisco in 1996 up to the 4th in Hamburg in 2007) and several texts devoted to the topic. A European Union (EU) research project (known as HARDER) has provided the most complete assessment of actual marine collision data, and this is used by vessel designers and marine classification societies. Collision events cover a very wide range of scenarios: different speeds and angles of approach, different vessel sizes, and different striking bow shapes. Marine designers have long considered collision resistance, but this applies

TYPE 1—ABOVE WATER BREACHES AT SEA

77

primarily to the striking vessel, which has a partially collapsing bow shape, designed to absorb some of the energy of collision. The struck vessel has some design protection due to the use of double hulls. Double hulls do not mean double strength; in fact, the cost of a double-hull vessel (reflecting the extra steel) is typically 20% more than that of a similar capacity single-hull vessel. However, there is additional strength and the interhull gap of 2–3 m assures that many smaller collisions, groundings, or allisions do not penetrate this distance and therefore afford protection to the LNG tanks. Following a number of serious marine accidents, the EU sponsored the HARDER project to establish design criteria for ship stability for vessels involved in collisions (Laubenstein et al., 2001). Collision statistics were used for this purpose, but these had not been updated for 40 years. The project reviewed existing collision data sets (mainly from the International Maritime Organization [IMO]) and added new data (mainly from classification societies), and this provided a data set of almost 3000 collision cases. The aim was to establish a probabilistic external hole size for ship stability. HARDER statistics include open water collisions (i.e., unlike the restricted speed and additional port controls for LNG vessels making terminal approaches) and cover all types of vessels (some double-hull vessels, but no LNG carrier incidents). Figure 5.2 shows that HARDER data first provide good representation of the total world fleet, and second gives the distribution of vessel striking speeds. The data set is quite noisy, reflecting the wide variety of collisions. A finding of HARDER, not necessarily intuitive, is that the hole size caused on the struck vessel is not strongly correlated to vessel dimension, and that a normalized hole dimension (hole size divided by the ship’s length) is a reasonable basis for analysis. This is because many real holes are long tears along the vessel length, not circular holes—demonstrated in Figure 5.3 with normalized hole sizes close to 1.0. This correlation is poorer for the larger vessel sizes above 250 m. It is often assumed that after a collision occurs, the ships then part, leaving a gap equal to the total dimension of the hole. In practice, many high-energy collision events may leave the colliding vessels connected by tangled steel, and the resultant outflow path is likely to be smaller than the actual hole dimension. The data in HARDER refer to shipyard measured dimensions during repairs, not the hole size in the immediate aftermath of the event. Paik et al. (2001) have assessed a range of possible collisions of LNG carriers by other LNG carriers and by very large crude carriers (VLCCs) at multiple load levels and orientations. An indication of the wide range of impacts is provided in Figure 5.4. The figure shows a 90° impact (also known as a “T-bone” collision), but other angles are also important. The authors looked at three different loading situations for VLCCs and LNG carriers. They identified critical speeds for collision for an LNG carrier onto an LNG carrier of 6.6–7.4 kts leading to tank spillage, and for VLCCs of 1.7–7.7 kts, for heavy and light colliding loadings, respectively.

78

LNG DISCHARGE ON WATER

55.86

60 (%) 50

Ship-type representation in the database and the worldwide fleet

41.60

40

20 14.20

6.30

4.40 3.70

6.19 5.60

10

24.30 23.99

Database

Worldwide fleet

30

9.80 3.96

ke r Ta n

o R

oR

r ss en Pa

ca r al er G

en

ge

go

er in ta on C

Bu

lk er

0

Distribution of striking ship speeds HARDER data, 344 colision cases 120

No. of collision cases

100 80 60 40 20 0 0–4

4–6

6–8 8–10 10–12 12–14 14–16 16–18 Ship speed intervals (knots)

Figure 5.2 Comparison of HARDER data to actual world fleet and distribution of striking speeds (Laubenstein et al., 2001). Note: Vessel count is less than total HARDER data set as many events did not record speeds.

Pitblado et al. (2008) updated this analysis to consider a wider range of striking vessel dimensions and impacts of both membrane and spherical tank carriers. That analysis examined three striking vessel sizes (90, 140, and 230 m) striking at 45° and 90° angles and with a raked and bulbous bow and at speeds of up to 19.5 kts (i.e., 10 m/s). The two types of LNG carrier were both of 138,000-m3 capacity. The dynamic finite element code, ABAQUS Explicit, was used with the grid illustrated in Figure 5.5. Average hull plate thicknesses were around 17 mm for the outer hull and 15 mm for the inner hull. The critical failure strain was calculated to lie between 17% and 19%, and hull elements will be expected to tear when this critical strain value is exceeded. HARDER

79

I/LPP

TYPE 1—ABOVE WATER BREACHES AT SEA

HARDER 575 records Regression line

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0

100

200 LPP

300

400

Collisions, nondimensional damage length versus ship length (575 records) Figure 5.3 Correlation between normalized hole size and struck vessel length (Laubenstein et al., 2001). I, damage length; LPP, ship length (m).

Table Identification notation for the ship collision scenarios

Z X

Y

Figure 5.4

Load condition of the striking ship Light Full Medium (ballast)

Striking ship

Struck ship

LNG carrier

LNG carrier

SS-L

SS-M

SS-F

VLCC

LNG carrier

VS-L

VS-M

VS-F

SS-F∗

SS-M

SS-L

VS-F

VS-M

VS-L

Collision possibilities and impact heights (Paik et al., 2001).

data were used to identify observed external hull damage dimensions onto the inner hull and thereby to establish inner hull damages. The LNG tank was not modeled, and tears to the inner hull were assumed to cause equivalent tears to the LNG membrane tank. The same basic assumption applied to spherical tanks if the collision location was at the closest approach of the tank, but the total probability was discounted due to the curved shape of the tank, and many inner hull intrusions would not strike the tank where this curves away from the inner hull. Membrane tanks achieve their strength from the hull, whereas

80

LNG DISCHARGE ON WATER

Figure 5.5 Finite element method (FEM) grid mesh used for membrane LNG carriers (Pitblado et al., 2008). Note: This figure zooms on the transition zone between the coarse and fine mesh (with the upper deck and outer shell removed for visual clarity).

Cumulative probability

Probability versus hole size membrane tank–combined bulb and raked bows 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Ref. Indentation Ref. Dam. Length Ref. Dam. Height

0

1

2 3 4 Inner side hole (m2)

5

6

Figure 5.6 FEM prediction of bow intrusion into a membrane carrier and resultant hole size (Pitblado et al., 2008).

spherical tanks are free-standing vessels with substantial wall thickness and internal strength. A typical damage impact is shown in Figure 5.6. The results of that work show that LNG vessels are very strong, but when sufficient collision energy exists to defeat all the structural resistance of the outer hull, the internal stringers and plates, and the inner hull, then there is little residual resistance that would differentiate a small hole or a large one, and thus defining a specific characteristic hole size is difficult. This is shown in Figure 5.6 where the hole size prediction is relatively flat versus probability. 5.1.2

Weapons Attack

A breach above the waterline can be sustained by a weapons attack, primarily from a missile or artillery shell. The likely effects of such a breach are covered in Section 5.2.2.

TYPE 2—AT WATERLINE BREACHES AT SEA

5.2 5.2.1

81

TYPE 2—AT WATERLINE BREACHES AT SEA Grounding or Collision

Grounding or collision of vessels is discussed in Section 5.1.1 and is applicable also to type 2 breaches. 5.2.2

Explosive-Laden Boat Attack

Intentional events can occur above the waterline such as from a shoulder launched or another type of missile. However, experts regard the more likely intentional event to be from a small boat, as occurred to the USS Cole in Yemen in 2000. Since the attack on the World Trade Center in 2001, security reviews have generally been carried out in secrecy (e.g., Lloyd’s Register review of LNG carrier terrorism risks). The public Sandia studies (Hightower et al., 2004; Luketa-Hanlin et al., 2008) are good examples of the compromise necessary between assessing the risk, defining the possible outcomes, but not specifying so much detail that the document becomes a training manual as to exactly which mode of attack will have some specific outcome. A security study prior to the 9/11 event was carried out by Kjøk and Lia (2001) that suggests a definition for terrorism that still appears valid: the use, or threat of use, of anxiety-inducing, extra-normal violence for political purposes by any individual or group, whether acting for or in opposition to established governmental authority, when such action is intended to influence the attitudes and behavior of a target group wider than the immediate victims and when, through the nationality or foreign ties of its perpetrators, its location, the nature of its institutional or human victims, or the mechanics of its resolution, its ramifications transcend national boundaries

This applies potentially to targets such as LNG carriers. Fuel carriers are symbols of energy dependence and attacking these would cause fear and anxiety to importing nations. LNG appears to offer more dramatic consequences than attacking other types of vessels (e.g., oil tankers). While this is partly true, there are other more serious targets that are beyond the scope of a book on LNG. According to Sandia (Luketa-Hanlin et al., 2008, p. 102), a wide range of attacks against ships has been documented, including hijackings, attacks with small missiles and rockets, and attacks with bulk explosives. Wesevich et al. (2004) note these attacks are known as explosive-laden craft (ELC). These were involved in the attack on the USS Cole in Yemen in 2000 and on the French oil tanker, the Limburg, in October 2002 (Anonymous, 2002). The results are shown in Figures 5.7 and 5.8. For a single-hull vessel, damage is the combination of external hole dimension and compartmentalization (see Fig. 5.7). The USS Cole was a single-hull vessel, but typical of warships, she was highly compartmentalized. This may

82

LNG DISCHARGE ON WATER

Figure 5.7

Damage to the USS Cole by ELC Attack.

Figure 5.8 Damage to the Limburg by ELC attack.

be contrasted with a double-hull vessel such as the Limburg. Wesevich et al. (2004) noted that the Limburg was a modern double-hulled membrane crude carrier. The ship was at sea when it was struck by a small craft carrying high explosives. The explosion produced a large hole in the outer hull and numerous small holes due to perforations from outer plate fragments projected into the inner hull plate. Crude oil leaked out from these perforations, which resulted in a significant pool fire adjacent to the carrier (see Fig. 5.8). Sandia (Hightower et al., 2004) has made some similar statements in limited descriptions of its damage modeling as illustrated in Figure 5.9. Holes to the inner hull are due to a combination of propagating blast wave, steel fragments from the outer hull, and even some contribution from high-velocity water.

TYPE 2—AT WATERLINE BREACHES AT SEA

83

Figure 5.9 Illustration of the type of penetrations to the inner hull expected from explosives on the outer hull (Blanchat et al., 2007).

Sandia (Hightower et al., 2004) nominated hole area between 2 and 12 m2 (equivalent to circular holes of 1.6- to 3.9-m diameter). These were due to detailed terrorism risk assessments of potential threats. Fay (2003) discusses releases of LNG from breeches at the waterline and below the waterline. These are essentially idealized calculations based mainly on density forces as flash evaporation effects due to the contact of cryogenic LNG and warm seawater are not explicitly modeled. Key issues for spills at the waterline are likely to be RPT flash explosions and large generation of flash vapors. Initially, there will be a large head of LNG as tanks are typically near to full on arrival in ports. A typical head will be17 m (see Section 3.3.2). Rapid flash effects will occur outside the tank, on the sea, or perhaps in the interhull gap. As the head declines, however, some LNG will enter the outer hull and also flash to vapor inside the LNG tank. At atmospheric pressure, this is a 500 to 1 flash fraction (1 vol of liquid flashes to 500 vol of vapor at the LNG boiling point), and this will quickly create a positive pressure in the tank. This will force the LNG out of the tank faster and will inhibit fresh ingress of water. The pressure generated inside the tank is likely to be small, insufficient to burst the tank or significantly to increase the hole size.

84

5.3

LNG DISCHARGE ON WATER

TYPE 3—BELOW WATERLINE BREACHES AT SEA

Holes below the waterline of LNG carriers can be caused by intentional or accidental events. Collisions may cause below waterline holes due to the bulbous bow of the striking vessel (Fig. 5.6), and groundings by definition are below the waterline. Unlike above and at waterline events, where immediate ignition is a high likelihood due to the energy released in the event, underwater events are much more likely not to ignite. There is some air in the interhull space and ignition is possible there, but might quickly extinguish if replenishment air cannot ingress. As detailed in Chapter 2, while there have been no serious collisions involving LNG carriers resulting in loss of cargo, four grounding events have occurred during the 57 years that LNG ships have successfully operated, but even loaded groundings resulted in no loss of LNG cargo. The cargo tanks were not penetrated in any case. Figure 5.10 illustrates one of the groundings. The El Paso Paul Kayser was an LNG membrane tanker carrying 95,000 m3 of cargo (125,000-m3 capacity),

The external hull damage is extensive, including penetration of the external hull.

However, the internal cargo tank damage is minimal, with slight buckling of the inner membrane liner. There was no loss of containment even though damage occurred at a vulnerable location.

Figure 5.10 Damage sustained by the El Paso Paul Kayser (Vaudolon, 2000) (publicity release by El Paso Marine Co.).

DISCHARGES FROM SHIP’S PIPEWORK

85

en route from Algeria to the United States on June 29, 1979. The vessel grounded running at 17 kts (Vaudolon, 2000) or 14 kts (CHIV, 2003) off Gibraltar and sustained substantial hull damage, including damage to tank 1 and 2 secondary membranes and deformation of the primary membrane. However, the LNG tanks were not penetrated. A quick salvage operation refloated the vessel and towed it to safe anchorage 5 days later. The LNG was pumped out on July 9 and 10 into a sister vessel and no LNG cargo was lost, which, according to Vaudolon, was feasible due to prior existence of a suitable emergency plan listing all required equipment. As LNG vessels spend half their travel time in an empty state, the one loaded event and two empty events are within statistical probability for a small number of events and no special grounding vulnerability while empty. LNG vessels travel under ballast when empty and the handling characteristics are similar to fully loaded. If an LNG vessel were holed below the waterline, the discharge would in principle be much more complex (Hightower et al., 2004). Fay (2003) discussed underwater releases, but did not model the complex interactions. Pitblado et al. (2005) and Woodward (2008) both addressed these events in greater detail with respect to thermodynamic effects, but not structural response. Section 5.7.3 provides model predictions. The various analyses show that there are many possible interactions and theory can only take this so far. Ultimately, some experiments are necessary (GAO, 2007) to establish the effect of cryogenic spills on hull members, both exposed and partially protected by water, and whether pressure effects are sufficient to significantly increase the scale of event due to internal pressure rise or expansion of the initial hole. Intentional events below the waterline might be due to mines or other explosive devices, either attached to the hull or in the path of an approaching LNG carrier. No specific discussion of these is provided by Sandia, but this potential must exist. In terms of discharge, the outflow may be broadly similar to the previous cases, but with a higher initial LNG head of liquid and the potential for water ingress to increase the discharge rate. Ignition is still possible below the waterline, as is evidenced by many fires associated with torpedo attacks of oil carriers in World War II, and is more likely than with grounding.

5.4

DISCHARGES FROM SHIP’S PIPEWORK

A further class of intentional or accident event is a discharge from the export pipework of the vessel. The ship’s pumps are designed to be able to discharge the entire cargo in less than 24 h. Pitblado et al. (2005) defined two events of 7000 m3/h for an accidental spillage lasting for 10 min and 10,000 m3/h for an intentional terrorist event lasting 60 min. An interesting experimental intentional event at sea to discharge cargo, should this ever be required, is discussed

86

LNG DISCHARGE ON WATER

Figure 5.11 Intentional discharge experiment on Tenaga Dua (Vaudolon, 2000) (reproduced by permission from A. Vaudolon).

in Section 8.7.2. Shell’s test with the SS Gadila (Kneebone and Prew, 1974; Vaudolon, 2000) was with 2- and 4-in. (51 and 102 mm) lines and, as is apparent from Figure 5.11, from tests on the Tenaga Dua, most of the LNG vaporized before reaching the sea surface. The dispersing cloud was affected by turbulent eddies around the vessel itself and did not disperse completely away from the vessel. The experiment was not considered successful, and discharge of cargo over the side is not considered a safe means to deinventorize. Pump-out to another LNG carrier is preferred (as was accomplished for the El Paso Paul Kayser). Accidental discharges at sea are almost inconceivable since the pump-out system is designed only for operation at a receiving terminal with full connection to shore-based loading arms. However, accidents in port are possible, at the full export flow rate, either from the vessel pipework, the loading arm, or the terminal pipework. This is discussed in Section 6.8 in Chapter 6. 5.5

CASCADING FAILURES AT SEA

Some other failure cases potentially affecting LNG carriers are as follows. 5.5.1

Sloshing Forces

Sloshing arises where the liquid surface of the LNG cargo under the ship’s motion interacts with the tank shape in a manner that very high local impact pressures can occur. This topic was addressed in Chapter 3 on vessel design. Pressures are sufficient to damage the membrane and small leakages have occurred, releasing LNG into the insulation. Membrane tanks are double walled, to protect against such events, and no loss of cargo has occurred, but

CASCADING FAILURES AT SEA

87

substantial costs have been incurred. Experimental work is under way to better assess these risks, and in the interim, strict load limitations are in place for sailing vessels. Spherical tanks are generally not considered at much risk from sloshing due to the absence of corners to act as pressure impact points and due to the strong structure of the tank itself. Recorded sloshing events are reported by CHIV (2003). 5.5.2

Explosion in Hull Chambers

There are confined spaces within the hull, between the inner and outer hull, and between other bulkheads. Some of these act as ballast chambers; others are closed void spaces or have open access ways. It is conceivable that LNG liquid or methane vapors could reach these spaces in the event of an accident. Since these spaces are often large, the spill size would need to be significant to reach flammable concentrations. The spaces are subject to electrical area classification and either have no powered equipment or, if present, have suitable designs to prevent ignition. Furthermore, there are no LNG and vapor return lines passing through these spaces as all connections pass through the roof and onto the deck. However, in an accident, both gas and ignition sources may be present, and a local explosion cannot be ruled out. While this might be serious, it is likely to be less than the initiating event that resulted in gas and ignition in these spaces. 5.5.3

RPT in Hull Chambers

RPTs require both a spill of LNG and the presence of a large amount of water. This is only likely in the type of events discussed above—collision, grounding, or intentional explosive attack. An RPT is often associated with small local overpressures. These typically have destroyed unstrengthened apparatus (as occurred in some LNG dispersion experiments such as the Maplin Sands tests (Puttock et al., 1982a, 1982b; Colenbrander and Puttock, 1983). They have caused window breakage and metal cracks. The pressure is due to the 500× volume expansion from liquid to gas when LNG is rapidly heated by water contact. Since water and LNG enter the interhull space, there will be an egress route also for the generated vapors. RPTs are discussed in more detail in Section 7.3. 5.5.4

Cryogenic Temperature Stresses on Decks and Hull

In the early operations of LNG vessels, there were several instances of small spills of LNG from connecting pipework leaks during loading or unloading activities. These very often caused localized cracking. Marine steels are malleable at ambient temperatures but become brittle below −40°C (as measured by the Charpey impact test). Since LNG is at −161°C, this is well below the threshold. Residual stresses from the original construction welding may remain in the hull or deck plates, or the temperature shrinkage due to chilling may

88

LNG DISCHARGE ON WATER

Figure 5.12 Cracking after 30- to 40-m3 LNG spill on deck (Vaudolon, 2000) (reproduced by permission from A. Vaudolon).

be sufficient to cause stresses, which the now brittle steel cannot withstand, and cracking results. Vaudolon (2000) provides several photographs showing cracked deck and outer hull plates. While some of these cracks were extensive (several meters in length), the overall structure was not threatened and repairs were easy to achieve since the steel reverts to normal malleability once it warms. Understanding of the issue and better integrity of connecting pipework now has greatly reduced the instances of cryogenic cracking. In addition, load redistribution is built in with ship design. In the event of an accident, large spills of LNG might occur and extensive localized cracking would be expected, as, for example, shown in Figure 5.12. However, water contact would partially protect the steel. The planned U.S. government experiments on cascading failures, as recommended by GAO (2007), should resolve the potential escalation associated with this mechanism. 5.5.5

Cascading Events Caused by Fire

Additional potential cascading events are discussed subsequently: • • •

5.6

fire stresses on hull (Section 9.16 of Chapter 9), fire enveloping pilot house (Section 9.16 of Chapter 9), and BLEVE possibility (Section 10.4.1 of Chapter 10)

INITIAL DISCHARGE RATE

In general for liquids, vapors, or a two-phase discharge, the discharge rate, w (in kg/s), is given in terms of the mass flux, G (in kg/[m2s]), the

INITIAL DISCHARGE RATE

89

discharge coefficient, CD, and the effective cross-sectional area of the breach, A (in m2): w = CD AG.

(5.1)

The simplest discharge formula has proved to be adequate to predict LNG discharge rates, the Bernoulli equation for subcooled liquids: G = uρ0 = [2ρL ( P0 − Pback )]1 2 ,

(5.2)

where ρL is the liquid density of LNG in the tank in units of kg/m3, usually at the boiling point or one or two degrees subcooled below boiling. The pressures are absolute (not gauge) in pascal with Pback, the back pressure, equal to atmospheric pressure if the discharge is not underwater or into an obstructed hull space, and P0 is the pressure of liquid at the breach. This pressure accounts for the head pressure above the LNG level, Phead, and the hydraulic head pressure of a column of liquid of length h between the liquid level and the breach: P0 = Phead + gρL hL ,

(5.3)

where g is the gravitational constant and the density is constant over the liquid column. These equations apply as well for initial discharge rate and for timevarying discharge, the only difference being the value of h, which is time varying, h(t). Liquid flow does not choke, so the back pressure directly affects the discharge rate. This is not true for compressible fluids, where after the back pressure drops below a choke pressure, further decreases do not affect the discharge rate. The discharge coefficient for sharp-edged orifices has a theoretical value of 0.61, whereas values of 0.975 or so are more typical for gases and vapors. For two-phase flow (vapor plus liquid), CD increases from 0.61 to 0.975 as the vapor quality (mass fraction) increases from zero to unity (Jobson, 1955; Bragg, 1960). For vertical or horizontal slot-type tears with “petals jagged inward,” values of CD are reported varying between 0.609 and 0.629 (Dodge et al., 1980). The discharge velocity at the breach, u, is from Equations 5.1 and 5.2: u=

G w = , ρL CD AρL

(5.4)

so the discharge coefficient can be thought of as reducing the effective crosssectional area of the stream, as is literally true at the choke point of choked, compressible flow.

90

5.7

LNG DISCHARGE ON WATER

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

A time-dependent discharge is called a blowdown. The discharge rate decreases with time as the LNG level drops, thereby decreasing the liquid head and corresponding hydraulic head at the breach. As example cases, consider a double-hull membrane-type LNG carrier with five tanks averaging 35,000 m3 of LNG each. Typically, 25,000 m3 is above the water level, giving a liquid level of about 17 m above the water level. The dimensions of a typical double-hull LNG carrier are shown in Figure 5.13 and are listed in Table 5.2. The LNG composition modeled is Algerian LNG. We assume here that the capacity of the vacuum breaker valve is adequate; that is, a vacuum breaker valve may open intermittently during the blowdown and may restore the head pressure to atmospheric, causing slight increases in discharge rate. In the example case below, with a 3-m equivalent diameter puncture, the vacuum breaker valve is predicted to open essentially every 20 s. 5.7.1 Blowdown for Type 2 Breach (at Waterline) For a puncture of both inner and outer hull at the water level, it is possible that some LNG could mix with water and be carried lower into the hull space. It is readily shown, though, that the hull space would quickly fill with water, so a large proportion of LNG would simply flow onto the sea. The discharge for a type 2 release is essentially parallel to the water surface, so there is minimal injection into the water. The rainout is predicted to be 98 mass %. For various size breaches of the inner hull (assuming an even larger breach of the outer hull), Figure 5.14 plots the liquid level on a logarithmic timescale (Woodward, 2007). The discharge rate curves tail off, and this tail has been

LNG level

h d H LT

H TAN

HT

H LB W

Water level

b B

b

DW

45°

d Figure 5.13 Dimensions for example LNG membrane carrier (Woodward, 2008).

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

Table 5.2

91

Assumed dimensions for LNG membrane carrier tank

Dimension

Symbol

Value (m)

Ship length Tank length Tank height

LSHIP L HT

270 50.0 30.58

Interior tank width

W

30.0

Interior bevel height

B

6.60

Double-hull separation Outer hull hole diameter Height of top of outer hull hole

d

2.2

DOH

various

HOH

various

Dimension

Symbol

Initial Value (m)

Number of tanks Ship draft LNG level above water level LNG level from inner tank bottom Water level from bottom of inner tank Web frame spacing LNG level in upper bevel

NTANKS DW HLT

5 10.32 17.05

HLB

25.084

B

8.12

LWEB

2.8

h

3.34

Leaked water level in hull

hW

Varies

18 Level above water (m)

16 14 12 10 8 6 4 2 0 0.1

1

10

100

1000

Time (min) 5 m hole 1 m hole Figure 5.14 2007).

3 m hole 0.5 m hole

Time-dependent tank level for various hole sizes at water level (Woodward,

omitted in Figure 5.14 and for the drain time. The predicted discharge rate is shown decaying in time in Figure 5.15. Table 5.3 summarizes the initial and average discharge rate and drain time predicted for the example case. The entire 25,000 m3 above the water level is drained in each case. The discharge rate calculation includes the discharge

92

LNG DISCHARGE ON WATER

Discharge rate (kg/s)

1.E + 05

1.E + 04 1.E + 03

1.E + 02 0

60

120

180

240

Tiem (min)

Figure 5.15 2007).

Table 5.3

3 m hole

1 m hole

0.5 m hole

Time-dependent discharge rate for various hole sizes at water level (Woodward,

Initial and average conditions for various size type 2 breach with Algeria LNG

Hole Diameter (m) 0.1 0.5 1.0 3.0 5.0

5 m hole

Initial Discharge Rate (kg/s)

Initial Discharge Velocity (m/s)

Average Discharge Rate (kg/s)

Average Discharge Velocity (m/s)

Drain Time (min)

41.2 1,020 4,060 35,430 95,210

17.6 17.5 17.3 16.8 16.3

24.5 600 2,250 12,790 27,850

10.4 10.0 9.2 6.2 6.0

7,340 330 88. 14.3 6.6

coefficient, CD of 0.61, so discharge rates are lower than in the FERC-ABS report where CD is 1.0. Blowdown calculations show the discharge rate and discharge velocity decreasing essentially linearly as the level drops. An exception to this linear behavior occurs when the level drops below the top of the hole and the area of discharge decreases. Figure 5.16 illustrates that the decrease in discharge rate at the end of the blowdown can be quite substantial for a 5-m breach at the waterline. 5.7.2

Blowdown for Type 1 Breach (above Waterline)

For a type 1 breach above the waterline, the issue of what fraction of the LNG leaks into the double hull is important. The velocity of the jetting liquid stream (assumed initially horizontal) could take the jet past the outer hull or not. Consider, for example, a breach of the inner hull of an equivalent diameter of 0.10 m, with an outer hull breach of 2.0 m. Using the initial pressures of the

93

Discharge rate (kg/s)

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

100,000 90,000 80,000 70,000 60,000 50,000 40,000 30,000 20,000 10,000 0

Discharge rate/time

0

50

04/22/2009

100

150

200 250 300 Time (s) SS3G V2.03.0027 DLL:18/04/09

350

400 01:46 p.m.

Figure 5.16 Blowdown curve for 5-m breach at waterline, 25,000 m3 lost.

Table 5.4 Predictions for blowdown of double hull with type 1 breach of equivalent diameter 0.1 m

No.

1 2 3 4

Breach Elevation, m (a)

Initial Discharge Rate (kg/s)

11.05, 8.05 5.05 2.05

24.4 29.9 34.5 38.6

Initial Average Discharge udis (m/s) Rate (kg/s) 12.5 16.9 20.0 22.6

10.4 12.8 14.7 16.5

Average udis (m/s)

Average Mean (dp, μm)

Drain Time (h)

5.2 7.0 8.4 9.3

10,000 6,440 4,530 3,640

67.8 87.6 104.0 120

hydraulic head for Algerian LNG gives the initial discharge rates and discharge velocities in Table 5.4. Even at the highest breach elevation evaluated, the initial jet velocity is sufficient to project the initial jet past the outer hull breach. This is illustrated in Figure 5.17. The jet profile shows that an outer hull breach of 2-m diameter or larger would allow the jet to flow without impinging on the outer hull. Table 5.5 gives the predictions for rainout and for the velocity at which the falling jet penetrates the water surface, both at the initial pressure and at the average pressure over the blowdown. These velocities are needed to predict the depth of LNG penetration into the water, discussed in Section 7.2.7. A similar plot for the initial jet discharge at the lower breach elevation of 2.05 m is shown in Figure 5.18. This indicates that a larger pool and plume develop with the lower breach since the discharge rate is higher and more LNG is spilled. As the discharge velocity decreases, the jet droops more, as illustrated in Figure 5.19 for the blowdown average pressure for an 11-m high breach. The jet would impinge and partly drain into the inner hull.

94

LNG DISCHARGE ON WATER

Vertical height (m)

Cloud centerline

UFL

12 10 8 6 4 2 0 0

50

0.5 fraction LFL

250

100 150 200 Downwind distance (m) SS3G V2.03.0027 DLL:08/04/09

04/17/2009

Cloud centerline

UFL

05:33 p.m.

LFL

0.5 fraction LFL

Vertical height (m)

12 10 8 6 4 2 0

LFL

0

1

2

3

04/17/2009

4

5 6 7 8 Downwind distance (m) SS3G V2.03.0027 DLL:08/04/09

9

10

11 05:33 p.m.

Figure 5.17 Side-view plots of initial LNG jet from 0.10-m breach in the inner hull at 11.05 m above water with D2.5 weather. UFL, LFL, upper and lower flammable limits.

Table 5.5 Predictions for rainout and penetration velocity at water surface for type 1 breach of equivalent diameter 0.1 m

No.

1 2 3 4

5.7.3

Breach Elevation, m (a) 11.05 8.05 5.05 2.05

Initial Rainout (Mass %)

Rainout at Average Discharge (Mass %)

Initial Vertical Velocity at Water (m/s)

Vertical Velocity to Water at Average Discharge (m/s)

73 69 68 76

90 88 87 89

5.0 4.8 5.0 7

4.6 4.4 4.6 6.4

Blowdown of Type 3 Breach (Underwater Level)

A breach under the waterline can occur from grounding or collisions. In this case, the intrusion of water and possible LNG into the double hull increases the pressure inside the hull space and importantly affects the inflow rates. In the absence of experimental data, we present theoretical predictions, recognizing that unexpected phenomena may develop. Pitblado (2007) and Woodward (2008) analyzed both the flow of water into the interhull gap and the potential for water to flow into the LNG tank. Both authors found that underwater holes initially have a gravity head from the

95

Vertical height (m)

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

UFL

Cloud centerline

8

LFL

0.5 fraction LFL

6 4 2 0 0

50

100

150 200 250 Downwind distance (m) SS3G V2.03.0027 DLL:08/04/09

Vertical height (m)

04/18/2009

12 10 8 6 4 2 0

Cloud centerline

0

1

2

04/18/2009

UFL

3

300

350 10:32 a.m.

LFL

0.5 fraction LFL

4 5 6 7 8 Downwind distance (m) SS3G V2.03.0027 DLL:08/04/09

9

10

11 10:32 a.m.

Figure 5.18 Side-view plots of initial LNG jet from 0.10-m breach in inner hull at 2.05 m above water with D2.5 weather.

Vertical height (m)

Cloud centerline

UFL

LFL

0.5 fraction LFL

10 8 6 4 2 0 0

04/22/2009

2

4

6

8 10 12 Downwind distance (m)

SS3G V2.03.0027 DLL:18/04/09

14

16

18

20

05:50 p.m.

Figure 5.19 Side-view plots of LNG jet after 32 h (average conditions) from 0.10-m breach in the inner hull at 11.05 m above water with D2.5 weather.

height of LNG above the hole, which is greater than the water hydrostatic pressure from the water depth. This means that LNG will flow out and initially flash to vapor either externally or into the interhull gap. Hydraulic head differences between the seawater, Pw, and the LNG, PL, affect inflow rates. For water (f = w), and LNG (f = L) at water depth z and LNG level above the water, HL, with respective densities ρw and ρL:

96

LNG DISCHARGE ON WATER

PW = gzρw

(5.5)

PL = gH LρL .

(5.6)

The total pressure at depth z, PTf, depends also on the head pressure above the liquid, Ph: PTf = Ph + Pf .

(5.7)

Puncture of Double-Hull Tank—Outer Hull Only Breached In the event of underwater penetration of only the outer wall of a double-hull tanker, water inflow will displace the air in the double-hull space and will raise the pressure in this space by compression. The density of gas in the compressed volume is the initial mass of air, mair0, divided by the air volume in the hull space, Vhair. This volume is decreased by the volume of water intrusion, VW: ρV 2 =

mair 0 mair 0 = . Vair Vh − Vw

(5.8)

The head pressure in the hull, Ph, assuming there is no air leakage, can be found by assuming isothermal compression: Ph =

ρ V 2 RT2 , MV 2

(5.9)

where MV2 is the mole weight of the air and LNG vapor, R is the gas constant, and T2 is the temperature after compression (ambient temperature). The hull space volume is made up of the cross-sectional areas of the hull times the tank length, subtracting the web spacer volume. The web spacers have large holes to enable personnel passage and to allow the hull space to serve as a ballast tank. There is a depth of intruded water inside the hull that achieves hydraulic pressure equilibrium. This is listed in Table 5.6 along with the required volume of intruding water. If the hull water level is below the top of the hull penetration, there will be an unstable interface of air on one side and water on the other. Air in the inner hull will escape by buoyancy. An equal volume exchange of air and water can occur at constant hull pressure. Eventually, the water level in the hull space covers the hole. The final water level and water intrusion volume are also listed in Table 5.6. Figure 5.20 plots the initial and final water level along with an illustrative hole location. The 7-m deep hole is the only case where the final water level ends well above the top of the hole.

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

97

Table 5.6 Water depth in double hull at initial and final pressure equilibrium

Top of Hole Below Water Level (m)

Water Level in Hydraulic Hull at Initial Pressure Pressure of Sea at Hole (kPa) Equilibrium (m)

−1 −3 −5 −7

Water level in hull (m)

9.8 29.4 49.1 68.7

Final Water Level in Hull (m)

Initial Water Intrusion Volume (m3)

Final Water Intrusion Volume (m3)

9.32 7.32 5.32 4.26

187 525 818 1070

1812 1594 1376 1260

0.40 1.12 1.74 2.27

10 9 8 7 6 5 4 3 2 1 0 1 m deep

3 m deep

Initial P equilibrium Figure 5.20

5 m deep

7 m deep

Final depth

Water depth in hull with outer hull penetration (Woodward, 2008).

The pressure buildup in the hull affects the rate of water inflow up the point of initial pressure equilibrium as shown in Figure 5.21. Longer term, the predicted water inflow rate is shown in Figure 5.22 for a hole diameter of 0.15 m at a 1-m depth. Upon reaching the initial point of hydraulic pressure equilibrium, the water inflow rate drops to a nearly constant rate by equal volume exchange with air. This period lasts until the water volume in the hull builds to 1812 m3, and the water level covers the outer breach at 470 min (Woodward, 2008). Puncture of Double-Hull Tank—Both Outer Hull and LNG Tank Breached If the inner hull is also breached at the same time as the outer hull, then LNG and water enter the hull space. Generally, all of the LNG would very quickly vaporize. Water in the hull space will be chilled and some fraction could freeze.

98

LNG DISCHARGE ON WATER

Discharge rate (kg/s)

140 120 100 80 60 40 20 0 0

50

100

150

200

250

300

Time (min) Disch 1 m Disch 5 m

Disch 3 m Disch 7 m

2000

50 45 40 35 30 25 20 15 10 5 0

1600 1200 800 400

0

100

200

300

400

Water in hull (m3)

Discharge rate (kg/s)

Figure 5.21 Time-dependent water inflow rates for outer hull penetration; 0.15-m hole at depths of 1, 3, 5, and 7 m (Woodward, 2008).

0 500

Time (min) Disch 1 m

Vol water

Figure 5.22 Water inflow rate for 0.15-m hole at 1-m depth in the outer hull of the tanker (Woodward, 2008).

Water Chilling and Fraction of Water Frozen Assuming unhindered complete mixing, an enthalpy balance reveals the effect of LNG boiling on the incoming water. The enthalpy balance for a mass of water, mw, and a mass of LNG, mC, is mw = [ HwL (Tamb ) − HwL (Teq ) + fw HwFus ] mC = ΔHCvap (TNBP ) + HCV (Teq ) − HCV (TNBP ) ,

(5.10)

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

99

Kg LNG vapor/kg water

1.2 1.0 0.8 0.6 0.4 0.2 0 0

0.2

0.4

0.6

0.8

1.0

kg ice/kg water kg LNG vapor/kg water Figure 5.23

Mass fraction of LNG vaporized by freezing water (Woodward, 2008).

Kg LNG vapor/kg water

1.2 1.0 0.8 0.6 0.4 0.2 0 0

40

80 120 Ice subcool (°C)

160

kg LNG vapor/kg water Figure 5.24 Mass fraction of LNG vaporized by subcooling ice (Woodward, 2008).

where HwL = enthalpy of liquid water (kJ/kg) HwFus = heat of fusion of water (333.68 kJ/kg) fw = mass fraction of the mass mw that freezes ΔHCV = heat of vaporization of LNG (516.2 kJ/kg at TNBP of LNG) HCV = enthalpy of LNG vapor (kJ/kg) Tamb = ambient temperature, initial temperature of liquid water (283.15 K) TNBP = normal boiling point of LNG (K) Teq = equilibrium temperature of mixture (273.2 K if ice is present). Figure 5.23 plots the ratio of the mass ratio of LNG vaporized to water against the mass ratio of water that freezes. Figure 5.24 plots the

100

LNG DISCHARGE ON WATER

Table 5.7 Initial LNG to water inflow ratios in the hull and mass fraction of ice formed

Hole Diameter Ratio (DOH/DIH) ⎧(0.15 0.05 m m ) ⎪ 3 : 1 = ⎨(0.30 0.10 m m ) ⎪(0.45 0.15 m m ) ⎩ 2 : 1 = (0.60/0.30 m/m)

Mass Ratio Mixing (mLNG/mW) 1-m Hole Depth

3-m Hole Depth

5-m Hole Depth

7-m Hole Depth

0.222

0.135

0.110

0.097

0.495

0.312

0.250

0.218

0.175 0.536

0.142 0.455

Mass Ratio Ice/Water 3 : 1 (as above) 2 : 1 (as above)

0.46 1.0

0.24 0.695

same LNG vapor/water ratio against the degrees of ice subcooling. These plots indicate that 0.045 kg of LNG is vaporized by cooling 1 kg of liquid water from 10°C to the freezing point. A maximum value 1.085-kg LNG vapor/kg ice is achieved by subcooling the resulting ice to the normal boiling point of LNG. LNG/Water Develops Ice/Water Slush in Initial Response Period The initial inflow rates of water and LNG are predicted using the Bernoulli equation, Equation 5.2, as a function of hole size and penetration depth. These rates are plotted in Figure 5.25a,b. For the same hole size, the inflow rates for LNG are less by a factor of two to five than the inflow rate of water and are less sensitive to penetration depth. With these inflow rates at various hole size ratios (outer hull to inner hull), Table 5.7 gives the mass ratio of LNG to water. With a 3:1 hole size ratio, the mass ratio of LNG to water in Figure 5.25 is in the low range. Consequently, a slush, or a low ratio of ice to water, is expected. With 2:1 hole diameter ratios, the LNG/water ratio is higher and ice/water ratios range up to 100%; that is, under some conditions, all of the water is estimated to freeze. This does not account for salt water composition or for heat conduction through the hull. LNG Evaporation Adds to Hull Pressure An LNG leak contributes additional vapor and cooling and strongly changes the dynamics. In the early response period, there is likely to be sufficient water to vaporize all LNG in the hull space. The vapor density in the hull space changes by the added mass of LNG vapor, mCV, which modifies Equation 5.7: ρV =

mair + mCV . Vh − VW

(5.11)

LNG discharge rate (kg/s)

Water inflow rate (kg/s)

TIME-DEPENDENT DISCHARGE (BLOWDOWN)

2000 1600 1200 800 400 0 0

(a)

400 300 200 100 0 0

0.15 0.3 0.45 0.6 Equivalent hole diameter (m) Ztop = 1 m

Ztop = 3 m

Ztop = 5 m

Ztop = 7 m

101

0.05 0.1 0.15 0.2 0.25 0.3 Equivalent hole diameter (m) Ztop = 1 m

Ztop = 3 m

(b)

Inflow rates (kg/s)

60

2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0

50 40 30 20 10 0 0

10

20

Hull head pressure (bar [a])

Figure 5.25 (a) Water inflow rate to double hull (Woodward, 2008). (b) LNG inflow rate to double hull (Woodward, 2008).

30

Time (min) LNG rate

Water rate

Hull head P

Figure 5.26 Response of penetration of outer and inner hull, 1 m deep, 0.15-m outer hole, 0.05-m inner hole; leak rates, and back pressure (Woodward, 2008).

Using an adiabatic mixing curve developed by solving the mass and enthalpy balance gives the change in temperature, and Equation 5.9 gives the new pressure. Figure 5.26 illustrates predicted water and LNG inflow blowdown curves for an example 1-m deep hole. The water inflow rate drops to 5 kg/s at 3.7 min as the period of equal volume exchange begins for water inflow and vapor outflow. The LNG inflow rate begins at 10 kg/s and goes to near zero after 12 min as the equal volume exchange begins for LNG. There is little effect on the LNG tank head pressure during this period.

102

LNG DISCHARGE ON WATER

LNG/water (kg/kg); Mass fr. C in vapor

0.6 0.5 0.4 0.3 0.2 0.1 0 0

5

10

15

20

25

30

Time (min) Vapor Wc Figure 5.27 Vapor composition for example case of Figure 5.26 (Woodward, 2008). Table 5.8

DOH/DIH 3 :1

2 :1

Conditions at end of period of equal volume interchange

Conditions Time (min) Maximum head pressure (bar[g]) Volume of water (m3) Water level in hull (m) LNG losses (%) Time (min) Maximum head pressure (bar[g]) Volume of water (m3) Water level in hull (m) (Values should be) LNG losses (%)

1-m Hole Depth

3-m Hole Depth

5-m Hole Depth

7-m Hole Depth

1450 2.01

720 1.96

438 1.89

284 1.69

1810 9.3 2.45 40.4 2.53

1600 7.4 1.32 25.4 2.20

1380 5.4 0.92 15.5 2.19

1165 3.4 0.69 10.4 1.79

1950 10.6 (9.32) 5.88

1870 9.9 (7.32) 3.55

1539 6.5 (5.32) 2.34

1190 3.6 (3.32) 1.58

Figure 5.27 provides the change in vapor composition over time for the same example case. The composition reaches concentrations with about 50% natural gas, well above the UFL, so nonflammable. When the LNG outflow is essentially choked off, these values remain nearly constant. Predicted conditions for double penetrations to a double hull are summarized in Table 5.8. The pressures and water levels in the hull at the end of the period of equal volume interchange are listed along with LNG losses. The LNG losses are much smaller with a double-hull tank than with a similar single-hull tank (speaking of higher-boiling cargo). Possible Water Inflow to LNG Tank If the hole is of sufficient dimension, water will counterflow into the LNG tank and provide heat to the cargo, flashing LNG to vapor inside the tank. This will raise the tank pressure quickly

VACUUM BREAKING AND GLUG-GLUG EFFECTS

103

and will cause several simultaneous outcomes—the tank will experience higher internal pressure; the pressure relief valves will likely lift; and discharge of LNG will accelerate due to the combination of static liquid head and now pressure head as well. This will reduce or stop further water ingress. It is possible that the higher pressure may enlarge the hole, especially for membrane tanks that derive their strength from the hull and have little inherent strength in the region of the hole. However, it is unlikely that the pressure would fail the tank above the hole as spherical tanks are very strong (normally capable of withstanding 8-bar pressure) and membrane tanks are protected by the upper hull, and this would not be damaged in a grounding event. From Figure 5.23, 1.085-kg LNG vapor is generated for each kilogram of water entering the LNG tank. Taking a pressure relief valve with a 1-ft equivalent diameter, use the Bernoulli equation (Eq. 5.2) to calculate the head overpressure in the LNG tank needed to discharge the generated vapor for any rate of water inflow. These overpressures (around 0.35 barg) are far higher than the pressure driving force for water inflow (around 0.015 barg), indicating that the pressure increase caused by water intrusion into the LNG tank would quickly choke off further water inflow. Further dynamic analysis is needed, since a cyclic inflow of water to the LNG tank could develop. However, ice formation could also choke small holes.

5.8

VACUUM BREAKING AND GLUG-GLUG EFFECTS

The pressure in the vapor space decreases by the piston effect as the liquid level drops. This is partly mitigated by LNG vaporization inside the tank and by vacuum breakers that allow air intrusion. If the vacuum breaker capacity is inadequate, the interior head pressure could drop below the ambient plus hydraulic head pressure, so the pressure at the breach would be below ambient; that is, when the pressure at the breach is below ambient, air can be gulped at the breach, similar to the glug-glug effect of a water cooler. A breach below the water level could admit seawater, depending on the pressure balance at the breach, as is discussed in Chapter 7.

6 RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

This chapter considers issues related to risk analysis of onshore liquefied natural gas (LNG) operations, including receiving terminals and land transport of LNG. Figure 6.1 is an example of a planned import terminal with three storage tanks and with the capability of docking two LNG carriers at once.

6.1

TYPICAL BASIS FOR LNG RECEIVING TERMINAL

Typical operating conditions for an LNG receiving terminal are listed below (Durr, 2004): Ship • Size: 125,000 m3 • Draft: 10 m • Boil-off rate: 0.18–0.25% per day based on full ship Unloading System • Unloading time: 12 h • Unloading rate: 10,000 m3/h • Insulated line heating: 25–38 W/m2

LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

104

FEATURES OF LNG RECEIVING TERMINALS

105

Figure 6.1 Receiving terminal artist’s rendering of docks, transfer lines, and storage tanks (Cheniere, 2008) (reproduced by permission from Cheniere Energy, Inc.).

LNG Storage at Terminal • Capacity: 225,000-m3 maximum (two to three times ship) • Heat leakage: 0.05% per day (double wall) to 0.1% per day (membrane) • Steel thickness: 2-in. maximum Sendout • Temperature: 5°C • Pressure: pipeline pressure (11–17 barg) • Sendout rate: demand driven, variable • Some specifications: heating value (composition), dew point

6.2

FEATURES OF LNG RECEIVING TERMINALS

An LNG receiving terminal consists of four areas: 1. 2. 3. 4.

the dock and storage tank area, connected by the LNG transfer line loop; the LNG process area for regasification; the utilities area; and the supporting area.

The overall flow diagram is illustrated in Figure 6.2. A transfer line loop delivers liquid from the docked LNG carrier to the storage tank and returns displaced vapor to the carrier tanks to avoid drawing a vacuum in the carrier or building pressure in the terminal tank. The transfer line loop recirculates at other times. A boil-off compressor recovers vapors

106

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Vapor return line

Boil-off gas compressor

LNG unloading arm Recondenser LNG tanker

First-stage sendout pumps LNG storage tanks

Fuel gas Vaporizer

Second-stage sendout pumps

To pipeline

Figure 6.2 Flow diagram of an LNG regasification terminal (Durr, 2004) (reproduced by permission from KBR, Inc.).

Figure 6.3 Dock and transfer line from LNG carrier at LNG terminal (SIGGTO, 1986) (reproduced by permission from SIGTTO).

evaporated during the transfer. Liquid is pumped to the pipeline pressure and is then vaporized. Figure 6.3 shows a delivery line loop over water. Over land, the area beneath the transfer line is usually lined, sloped, and possibly insulated to direct any LNG spill to an impoundment area to contain the spill and to reduce the evaporation rate and subsequent vapor cloud size. The LNG process area primarily vaporizes LNG to natural gas with a heat source. The LNG process area may have a distillation column to separate and

FEATURES OF LNG RECEIVING TERMINALS

107

recover heavier components called natural gas liquids (NGLs). NGL includes propane and butane and may have a higher price than the LNG and can be sold separately. It may also be necessary to adjust the heat of combustion of LNG to deliver a consistent product to pipeline customers. If the heat of combustion or more specifically the “Wobbe Index” is too high, this requires either diluting the natural gas with nitrogen or air or extracting ethane and heavier components (C2+ extraction), so end users do not have to adjust their equipment. The utilities area provides services required by the plant including instrument air, nitrogen, fuel gas, power generation, emergency power, flare and blowdown system, drain systems, waste water and effluent treatment, demineralized water, fire water, and backup diesel-driven fire water pumps. The supporting area includes maintenance shops, parts storage, offices, and the like. There are several major types of LNG vaporizers: 1. Submerged Combustion Vaporizers (SCVs) A fired heater is placed below the LNG liquid level. 2. Submerged Combustion Vaporizers with Selective Catalyst Reduction Vaporization (SCRV) A fuel is oxidized using a catalyst to generate a lower temperature. This is also physically below the LNG liquid level. 3. Ambient Heating Vaporizers (AHVs) A large heat exchanger that uses seawater or air to supply the heat of vaporization to the LNG. 4. Indirect Fluid Vaporizers (IFVs) Heat is transferred from a fired heater or ambient fluid heat exchanger to an intermediate fluid that vaporizes the LNG by heat exchange. A schematic showing internal piping in an LNG storage tank for both top and bottom loading is illustrated in Figure 6.4 (Durr, 2004). Actual tanks have either top or bottom loading. Figure 6.4 shows a double-containment tank with two containment walls separated by insulation with all connections through the roof. There are several types of LNG storage tanks. Examples of these types of storage tanks are shown in Figures 6.5 through 6.10: • •

•

•

•

single containment (steel or prestressed concrete shell; Fig. 6.5 and 6.6); double containment (an outer concrete shell and an inner steel shell with insulation between (Fig. 6.7); double-containment concrete (both outer and inner walls are of prestressed concrete); full containment (an outer concrete shell and a steel or prestressed concrete roof; Fig. 6.8); and inground membrane (tank sunk into the ground or with earthen embankment; Fig. 6.10).

108

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Bottom fill

Top fill

Pump isolation and lifting mechanism Liquid outlet

Purge

Vapor outlet

Cooldown spray ring

Annular purge

Resilient blanket

Splash plate

Conical distributor

Perlite insulation in annular space

Standpipe Annular space purge ring

Pump well

Base insulation Purge Inner purge

Tank ring

Slots LNG tank internals

Figure 6.4 Typical LNG storage tank internals (Durr, 2004) (reproduced by permission of KBR, Inc.).

Figure 6.5 CBI).

Single-containment steel LNG tank (Coers, 2003) (reproduced by permission of

Single-containment LNG tank Carbon steel outer tank roof, shell and bottom Sidewall/deck insulation system

9% Ni inner tank

Load bearing bottom insulation system

Concrete foundation

Foundation heaters

Dike

Figure 6.6

Suspended insulation deck

Dike

Single-containment LNG tank (Coers, 2003) (reproduced by permission of CBI).

Carbon steel outer tank roof, shell and bottom

Double-containment LNG tank

Suspended deck and insulation

Cover if required Loose fill perlite insulation system Prestressed concrete outer tank wall 9% Ni outer bottom and liner

9% Ni inner tank Outer shell not able to contain liquid

Load bearing bottom insulation system

Foundation heaters

Concrete foundation

Figure 6.7

Double-containment LNG tank (Coers, 2003) (reproduced by permission of CBI).

110

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Suspended insulation deck

Carbon steel dome roof

9% Ni inner tank Sidewall/deck insulation system

Prestressed concrete wall

Vapor barrier

9% Ni bottom and corner protection Concrete foundation

Load bearing bottom insulation system

Foundation heaters

Full containment (steel roof) LNG tank

Figure 6.8 Full-containment LNG tank with steel roof (Coers, 2003) (reproduced by permission of CBI).

Figure 6.5 shows the basic properties of a single containment tank with a suspended roof, 9% nickel steel containment wall, with top penetrations, and top or bottom loading. Earlier storage tanks were built with penetrations below the LNG level, such as shown in Figure 6.9 for the Elba Island, GA, USA terminal. More recent tanks use top entry as seen in Figure 6.5. Inground storage tanks (shown in Fig. 6.10) are buried completely and have concrete caps. This minimizes several kinds of risk including spillage, exposure to domino events, or earthquake collapse. Japan has the largest inground LNG tank with 200,000-m3 capacity, in service since 1996. There were 61 such tanks in Japan in 2003 (IELE, 2003a). Korea also has inground LNG tanks. The distribution of types of LNG storage tanks worldwide in 2003 is listed in Table 6.1 (Coers, 2003). Double-wall tanks predominate by far. 6.3

STANDARDS FOR RECEIVING TERMINAL DESIGN

The basis for the design of LNG receiving terminals is to “ensure the terminal is safe, easy to construct, operate, maintain, and to meet all relevant local laws and regulations as well as relevant insurance codes” (Kuo, 2007).

STANDARDS FOR RECEIVING TERMINAL DESIGN

111

Figure 6.9 Elba Island LNG tanks with bottom liquid penetrations (Coers, 2003) (reproduced by permission of CBI).

Figure 6.10

Inground LNG storage tank (IELE, 2003b) (source: www.takenaka.co.jp).

Table 6.1 Distribution of LNG storage tanks in 2003

Type of Tank Double-wall steel Composite steel and concrete Inground External supported membrane Double-wall concrete Offshore storage a

Update by IELE (2003b).

Number 300 85 61a 30 6 0

112

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Local statutory laws, code standards, recommended practices, and regulations include • • • • • • • • • •

Australian Standards (AS), Canadian Standards (CS), Chinese Standards (GB, GBJ, and TJ), European Union Standards (European Norm [EN]), British Standards (BS), Bureau of Indian Standards (IS), Japanese Standards (JGA), Mexican Standards (NOM), United States of America Standards (see next section), and Society of International Gas Tanker and Terminal (SIGTTO, 1986).

Operators

Covering all these standards is beyond the scope of this book, but fortunately, some common concepts occur in the standards listed above. Some important concepts are treated below. In fact, some U.S. Standards are essentially copied or accepted completely by several other countries.

6.4 U.S. GUIDELINES AND REGULATIONS FOR RECEIVING TERMINALS The following organizations and sources provide guidelines for fire protection and safety in processing plants, including LNG receiving terminals: • • • • • • • • •

American National Standards Institute (ANSI) American Petroleum Institute (API) Code of Federal Regulations (CFR) (U.S. law) Engineering Equipment and Material User Association (EEMUA) Institute of Petroleum (IP) Industrial Risk Insurers (IRI) National Fire Protection Association (NFPA) Oil Insurance Association (OIA) Occupational Safety and Health Administration (OSHA)

Some important documents listed below are commonly referred to while designing LNG receiving terminals (Kuo, 2007): •

API RP 500 and 505 Recommended Practice for Classification of Locations for Electrical Installations at Petroleum Facilities

U.S. GUIDELINES AND REGULATIONS FOR RECEIVING TERMINALS

•

•

•

•

• •

• •

113

API RP 521/ISO23251 Guide for Pressure Relieving and Depressuring Systems API RP 752/753 Management of Hazards Associated with Location of Process Plant Buildings/Portable Buildings API RP 2510 Fire Protection Considerations for the Design and Construction of LPG Installations API RP 2610 Design, Construction, Operation, Maintenance and Inspection of Terminal and Tank Facilities 49 CFR, Part 193 (2005) Liquefied Natural Gas Facilities 33CFR Part 126 Waterfront Facilities Handling Liquefied Natural Gas and Liquefied Hazardous Gas NFPA 30 Flammable and Combustible Liquids Code NFPA 59A Standard for the Production, Storage and Handling of Liquefied Natural Gas (LNG)

6.4.1 LNG Transport Administered by the Department of Transportation (DOT) and the U.S. Coast Guard In the United States, LNG transport by pipelines and trucks is administered by the DOT under Section 193.2051 of the Federal Regulations 49 CFR Part 193 (2005) and as amended from time to time. This regulation cites, and thereby incorporates, NFPA 59A “Standard for the Production, Storage, and Handling of Liquefied Natural Gas (LNG).” The regulation states that each LNG container and LNG transfer system must have a thermal exclusion zone and a dispersion exclusion zone. These are discussed further below, since LNG terminals are covered under the same regulation as administered by a different commission. Transportation of LNG by sea is administered by the U.S. Coast Guard under the Marine Transportation Safety Act of 2002 as introduced in Chapter 1. 6.4.2 LNG Terminal Permitting by Federal Energy Regulatory Commission (FERC) In the United States, The U.S. FERC is responsible for approving applications for new LNG import terminals. The governing U.S. federal law is also 49 CFR Part 193 (2005), which (Coers, 2003) •

• • • •

covers LNG safety for design and construction, operations and maintenance, and personnel and security; provides requirements for terminal siting; requires design details for permit approval; refers to many sections in NFPA 59A; and is applicable to U.S. territories only.

114

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

NFPA 59A is a voluntary industry standard for the production, storage, and handling of LNG, but some parts of which are called out in the CFR, and which • • • •

• • •

originated for LNG peak shaving plants; is based on prescriptive rules; provides detailed formulas for LNG siting; refers to API 620 for steel tank design and construction (not for siting or operation); includes operations and maintenance standards; defines safety distances (exclusion zones) for fire and dispersion; and is used widely internationally, except in Japan.

Studies Required by FERC for LNG Receiving Terminals The FERC requires a number of studies for the approval of a new LNG receiving terminal, including 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 6.4.3

project description; water use and quality; fish, wildlife, and vegetation; cultural resources; socioeconomic impact; geological impact; soil impact; land use and recreational impact; air and noise impacts; alternatives analysis; reliability and safety; soil contamination; and engineering and design data. Pool Fire Radiation Exclusion Zone

The U.S. regulation 49CFR193.2057 Subpart B entitled “Thermal Radiation Protection” specifies that thermal exclusion zones must be calculated for impounding areas (diked areas, sumps, and associated trenches) around all LNG containers and LNG transfer systems. These exclusion zones are based on defined design spill scenarios and must be large enough so that the thermal radiation from an LNG fire does not exceed a specified limit for protecting offsite people and property. Those limits are listed in Table 6.2.

U.S. GUIDELINES AND REGULATIONS FOR RECEIVING TERMINALS

115

Table 6.2 Thermal hazard criteria in NFPA 59A Standard

Thermal Radiation Flux Limits to Property Lines and Occupancies Thermal Radiation Flux kW/m

2

Exposure 2

BTU/(hr·ft )

5

1,600

5

1,600

9

3,000

30

10,000

For ignition of a design spill, a property line limit for building upon.a For a fire in an impounding area, the nearest point outside the property line that, at the time of the plant siting, is used for outdoor assembly by groups of 0 or more persons. For a fire in an impounding area, the nearest point of a structure outside the property line that is in existence at the time of plant siting that is used for assembly, education, health care, detention and correction, or residential occupancies. For a fire over an impoundment area, the nearest property line that can be built upon.b,c

a

See NFPA 59A Section 5.2.3.5 for the definition of a design spill. See NFPA 59A Section 5.2.2.1 for the requirements for impoundment areas. c See NFPA 101, Life Safety Code, or NFPA 5000, Building Construction and Safety Code, for definitions of occupancies. b

For full-containment tanks, FERC requires the outer concrete shell to be considered as an impoundment area, and consequently, a full surface fire surrounding the tank must be evaluated. Fire models used to determine the required distances must be approved by the FERC Administrator. A preapproved model is the LNGFIRE3 model described in a Gas Research Institute report (Atallah and Shah, 1990). The wind speed, ambient temperature, and relative humidity are specified as those that produce the maximum distance to the above limits. An exception is that weather conditions that occur less than 5% of the time need not be used in the modeling. Some limitations of the regulation include that true hazards depend not only on the heat flux but also on the duration of exposure, the properties of the irradiated object, and the absorption of radiant energy. Neither the Federal Regulation nor NFPA 59A allows credits for mitigating phenomena such as clothing, the characteristics of skin that absorbs heat only in specific wavelength bands, movement of people into shelters, trees, and shadowing structures, cooling by wind, and water curtains. The criteria for exposure are based on very small, laboratory size tests, and the variation in susceptibility of different populations (children, elderly, etc.) has not been considered.

116

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

6.4.4

Vapor Dispersion Exclusion Zone

The U.S. regulation 49CFR193.2059 Subpart B (“Flammable Vapor-Gas Dispersion Protection”) requires calculating a dispersion exclusion zone for each LNG container and transfer system. The dispersion exclusion zone is specified as the distance to the ½ lower flammable limit (LFL) from the downwind edge of an impoundment space. The ½ LFL is defined as 2.5 mol % flammable vapors in air. Dispersion models to determine the required distance must be approved by the FERC Administrator. Preapproved models are the DEGADIS and the FEM3 models described in reports GRI-89/0242 and GRI-96/0396.5. Some modeling parameters are prescribed: • • • • •

atmospheric stability class F, wind speed 2.01 m/s (4.5 mi/hr) measured at 10-m elevation, 50% relative humidity, receptor height of 0.5 m, and surface roughness length of 0.03 m.

A higher roughness length may be used if it can be shown that the terrain both upwind and downwind has dense vegetation and that the height of the plume is over ten times the height of nearby obstacles. Different weather conditions can be used if it can be shown that they produce larger distances than any other conditions 90% of the time. Model Evaluation Protocol and Broader Range of Models for Siting As described in a report developed with a panel of LNG modeling experts for the National Association of State Fire Marshals (NASFM, 2009) (AcuTech Group, 2008, 2009): “The technical committee for NFPA 59A has worked to meet requests to allow new LNG hazard models (source term and dispersion) to help determine the exclusion zone distances to ensure adequate facility siting. Other stakeholders, including the fire service and State Fire Marshals, seek assurance that as additional LNG hazard models are allowed under NFPA 59A that these models are ‘fit for purpose.’ ” To enable the selection and evaluation of current and future LNG hazard models, the NFPA code committee enlisted NFPA’s Fire Protection Research Foundation (FPRF) to develop protocols for evaluating vapor dispersion and source term models, known as model evaluation protocols (MEP). The option to use the MEP was incorporated into NFPA 59A 2009 edition, which was adopted in 2008. The language for the 2009 update of NFPA 59A includes the following: 5.3.3.6* The spacing of an LNG tank impoundment to the property line that can be built upon shall be such that, in the event of an LNG spill as specified

U.S. GUIDELINES AND REGULATIONS FOR RECEIVING TERMINALS

117

in 5.3.3.7, an average concentration of methane in air of 50 percent of the lower flammability limit (LFL) does not extend beyond the property line that can be built upon, in accordance with a model that is acceptable for use by the authority having jurisdiction that has been evaluated by an independent body using the Model Evaluation Protocol facilities published by the NFPA Research Foundation report Evaluating Vapor Dispersion Models for Safety Analysis of LNG (emphasis added). This change is very significant for a number of reasons: •

Those involved in evaluating the siting of LNG facilities are accustomed to the use of a single commonly accepted model (DEGADIS) that has been used since the 1970s. In the future, there is the possibility that new or different models could be applied. While this allowance is promising and good for future progress in developing or enhancing models and the industry’s understanding of LNG hazards, it potentially makes the decision process more difficult. In future situations, models could be proposed for siting that are: 䊊

䊊

䊊

䊊

䊊

䊊

unfamiliar to the authorities; different from past situations even for the same facility; different between sites within the same jurisdiction; more complex and difficult to understand; less documented; used less frequently than other models.

A further discussion of the NASFM project as it relates to reducing uncertainties is provided in Chapter 12. Characteristics of Impoundment Areas Cited in NFPA 59A The NFPA 59A (2006) standard specifies for an LNG tank(s) both a dike and an impounding sump. A diked area around the tank must have a volume equal to the full tank of liquid. For example, a large LNG storage tank can be 30-m diameter filled with 30 m of liquid, a volume of 21,206 m3. A 2-m-high dike surrounding this tank would be 103 m square. This provision is standard for noncryogenic liquids where it makes sense to contain all of the liquid spilled. However, a cryogenic liquid such as LNG would not be so simply contained. Rather, it rapidly evaporates, and the resulting vapor cloud is the hazard of greatest concern. Figure 6.11 illustrates that the diked area can have some congested piping that contributes to an explosion hazard. The NFPA 59A standard also allows “provisions for detaining vapors or otherwise mitigating hazards.” Specifically mentioned are “impounding surface insulation and water curtains.” The former refers to an insulated sump that has a surface insulation liner, usually of low-density, low heat capacity “insulating concrete.” These insulating properties decrease the evaporation rate from an LNG spill, and hence the source strength for the resulting vapor plume.

118

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Figure 6.11 LNG storage tank showing dike and impoundment areas with piping (reproduced by permission of CBI).

Two situations are covered in the design specification by NFPA 59A: 1. LNG process, vaporization, or transfer areas and 2. LNG tank or tanks. For the first category, the volume of an insulated sump must be the largest volume of flammable liquid from any single source that could be discharged in a 10-min period. A shorter period is allowed upon demonstrating acceptable surveillance and shutdown features. For example, a 10-in.-diameter transfer line at 10-psig pumping pressure would discharge in 10 min approximately 158 m3. A sump holding this volume could be 2 m deep by 8.9 m2, considerably smaller than the dike for a storage tank. For the second category, the volume on an insulated sump depends upon the design of the LNG storage tank. For a tank with no penetrations below the LNG level (the usual case), the required sump volume is set by the full design pump-out rate for 10 min, provided the surveillance and shutdown systems are acceptable to the authorizing agency. The penalty for not having an acceptable system is that the insulated sump must be sized to accept the full volume of the tank over the time it takes to pump the tank empty.

EUROPEAN REGULATIONS FOR LNG RECEIVING TERMINALS

119

The insulated sump volume is similarly specified for two other LNG tank design options (penetrations below the LNG level with and without internal shutoff valves).

6.5 EUROPEAN REGULATIONS FOR LNG RECEIVING TERMINALS Two alternative approaches to regulation are applied to specifying LNG receiving terminals: •

•

Prescriptive The prescriptive approach defines required separation distances and plant area using explicit formulas, as exemplified by the U.S. 49 CFR Part 193 and NFPA 59A. Risk-Based Risk-based requirements define a limit for the risk imposed by a terminal and leave it up to the terminal designer to adjust separation distances and other design parameters, as exemplified by the European Union regulation EN 1473/1160—Installation and Equipment for Liquefied Natural Gas

6.5.1

Features of EN 1473

The European Union Standard EN 1473 (2005) applies a largely risk-based approach, although it is based on NFPA 59A and the British Standard BS 7777. Features of EN 1473 include the following: • • • • •

It It It It It

covers tanks in detail. defines single-, double-, and full-confinement tanks. includes design, materials, construction, and testing specifications. includes site requirements. does not include operations and maintenance issues.

Scenarios of potential LNG releases include not only facility spills but also releases from the LNG carrier, associated pipelines, and highway trucks. The specified thermal flux levels are listed in Table 6.3. The allowable maximum radiation flux excludes the local maximum solar flux. The flux levels can be reduced by separation distance, water sprays, fire-proofed radiation screens, or similar mitigation systems. The regulation allows consideration of the topography, size and elevation of the dike walls, and other obstructions to dispersion, atmospheric absorption of thermal radiation, and so on, in the calculation of exclusion distances and thermal flux levels. This standard is applicable to most projects except in the United States and in Japan.

120

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Table 6.3

Thermal hazard criteria in European Standard EN 1473

Maximum Allowable Thermal Radiation Flux kW/m2

Exposed Objects

BTU/(hr·ft2) Inside the Plant Boundary

32.0 15.0 15.0

10.2 4.8 4.8

8.0

2.5

5.0

1.6

Concrete outer surface of adjacent storage tanks Metal outer surface of adjacent storage tanks Outer surfaces of adjacent pressure storage vessels and process facilities Control rooms, maintenance shops, laboratories, warehouses, and so on Administrative buildings Outside the Plant Boundary

8.0 1.5 5.0

2.5 0.5 —

Remote areaa Critical areab Other areasc

a

An area infrequently occupied by small numbers of persons, for example, moor land, farmland, desert. b This is an unshielded area of critical importance where people without protective clothing can be required at all times including during emergencies, or an urban area (defined as an area with more than 20 persons per square kilometer), or a place difficult or dangerous to evacuate at short notice (e.g., hospital, retirement house, sports stadium, school, outdoor theater). c Other areas typically include industrial areas not under the control of the operator of the LNG facilities.

6.5.2 Comparison of Prescriptive and Risk-Based Approaches The advantages of a risk-based approach are that it provides the designer broad flexibility in creating mitigation measures to meet performance standards for the protection of the population. When a proposal establishes that the LNG plant risk is below the acceptability threshold, it is permitted. If the risk is in the “gray area” between the upper and lower threshold of tolerable risk, additional mitigation measures may be enforced to reduce the risk to the population. A limitation of the risk-based approach is in deciding the principal question what is a tolerable risk? Should this be based on level of injury or of fatalities? What level of detail should be considered for local topography, population density, and population diversity? The assessments are based on both the consequence estimates and the probabilities for quantity of release, probability of ignition as a function of time after the release, location of release, and environmental factors. The uncertainty level is increased by having to estimate so many difficult-to-estimate factors. Notwithstanding all of these difficulties, the risk-based approach provides a more rational basis for making informed decisions than an approach based on single, large event scenarios as in the prescriptive approach. The risk-based

EMPIRICAL FORMULA FOR REQUIRED LAND AREA OF TERMINAL

121

approach has been recommended for adoption in the United States (Raj, 2008b). A key feature of the risk-based approach is its explicit consideration of the likelihood of an event. For this reason, rarer events are often considered in a risk-based approach, albeit paired with an appropriate low frequency. Thus, major storage tank failure cases would be considered, whereas these are excluded from the NFPA 59A approach. This leads almost always to a doublewall tank design (full or double containment) as these have much less consequence if the primary container fails. The limitations of a prescriptive approach include that requirements are “geographically independent” in that they apply irrespective of whether the proposed facility is in a densely populated area or in a sparsely populated area. In addition, the prescriptive standards do not allow alternative safety mitigation considerations without obtaining a special permission from the regulatory authority. The main advantage of a prescriptive approach is that it defines clearly the methods to be used and the results to be obtained, lending an element of simplification and fairness to the approval process.

6.6 EMPIRICAL FORMULA FOR REQUIRED LAND AREA OF TERMINAL After applying the FERC requirements and designing a number of LNG receiving terminals, Kuo (2007) suggests some rule-of-thumb formulas to approximate the area required for a receiving terminal as a function of throughput capacity. Throughput is typically stated in the following alternative units: •

•

• •

U.S. unit: BCFD = billion cubic feet per day of natural gas sendout to a pipeline U.S. unit: MMSCFD = million standard cubic feet per day (1 BCFD = 1000 MMSCFD) Metric unit: MTPA = million metric tons per annum (1 BCFD ≅ 7.0 MTPA) Metric unit: 1 ha = 100 m × 100 m area

Kuo provides rules of thumb applicable up to a plant capacity of 5.2 BCFD. For multiple tanks, the minimum spacing between tanks, by tank type as tank diameters, D, is Single Containment 2.5D

Double Containment

Full Containment

Inground Membrane

Prestressed Concrete

1.6–1.75D

1.5D

1.5D

1.5D

122

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

With this spacing rule, and after using dispersion and fire models to establish the thermal exclusion zone and dispersion exclusion zone, the rules of thumb are LNG Storage Tank Area, A (Full-Containment Tanks, V in m3) 0.6 A( hectares ) = 0.0093Vtan k − 8.12

(6.1)

NGL Condensate Storage Tank Area (V in m3) 0.6 A( hectares ) = 0.0082Vtan k − 2.39

(6.2)

Double-Containment LNG Storage Tank Area Use 1.15 times area based on a full-containment tank. LNG Process Area, A, in Terms of Throughput, F, in MMSCFD A( hectares ) = 0.131F 0.6 − 0.80

(6.3)

To convert hectares to acres, multiply by 2.471. The process area formula applies to a basic process area that includes LNG unloading, recondenser, boil-off gas (BOG) compressor, sendout pump, LNG vaporization, odorizing natural gas, gas metering, electrical substation, and/or the control room and access roads. Additional plant space is needed for receiving terminals having more than the basic process area, such as the following: •

• • •

Add 6% for BTU content management by inert injection or C2+ extraction. Add 12% for NGL recovery by C2+ extraction. Add 5% for SCVs or SCRV (selective catalyst reduction vaporizers). Add 12% for AHVs.

For the utilities and support areas, •

Add 150% of process area for each.

Equations 6.3 and 6.1 are plotted, respectively, in Figures 6.12 and 6.13. The units for storage tank volume range from 50,000 to 250,000 m3. The area required for a storage tank is seen to be a substantial fraction of the area required for the processing section of a receiving terminal. The entire terminal would require the area for all four of the subareas listed at the beginning of this chapter, or • • •

the area for the storage tank (from Fig. 6.13) plus the area for the process section (from Fig. 6.12) plus the area for the utilities section (1.5 times the area from Fig. 6.12)

LEAK IN LOADING ARM OR IN STORAGE TANK

123

25

Area (ha)

20 15 10 5 0 0

1

2

3

4

5

6

Capacity (BCFD) Figure 6.12 Required area predicted by Equation 6.3 for the process section of the receiving terminal.

8 7

Area (ha)

6 5 4 3 2 1 0 50

100

150

200

250

3

Storage tank volume m /1000 Figure 6.13 Required area predicted by Equation 6.1 for the storage tank of the receiving terminal.

• •

6.7

plus the area for the support section (1.5 times the area from Fig. 6.12) plus area adjustments for other features such as BTU content management.

LEAK IN LOADING ARM OR IN STORAGE TANK

Loading arm leak events are listed in the LNG accident history in Chapter 2. These occurred by large movements by the LNG ship. Preventative measures focus on avoiding such movement as well as detecting movement and closing isolation valves. Figure 6.14 depicts a loading arm, complete with balancing weights, quick couplings, and built-in flexibility.

124

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Pantograph sheave (S/40 joint) Style 40 swivel joint

Inboard arm Transfer sheave Secondary counter weight

Pantograph sheave (S/50 joint) Outboard arm Style 50 swivel joint Style 80 swivel joint Base/riser

Manual camlock quick coupler 16”

Main counter weights Base plate

Electrical insulating flg Schematic of loading arm Figure 6.14 Example of LNG loading arm from ship to shore (Durr, 2004) (reproduced by permission from FMC, Inc.).

SIGTTO (1986) has developed useful guidance for safety systems, and most LNG loading arms now include a fast-acting double-valve powered emergency release connection (PERC), which operates in the event of excess movement of the loading arm. An extensive set of safety features is shown in Figure 6.15. 6.7.1

Modeling Effects of Substrate on Evaporation Rate

Three alternative substrates are illustrated for the “design rate” spill described above: (1) concrete, (2) dry soil, and (3) insulating concrete. The thermal properties of these materials are compared in Table 6.4. Concrete is seen to have a higher thermal conductivity than dry soil, and insulating concrete achieves a factor of four reductions from the thermal conductivity of concrete.

125

LEAK IN LOADING ARM OR IN STORAGE TANK

Detection of overmoving ESDS

ESDS

ESV Unloading arm LNG tank

ESV Cargo pump

Optical transmission unit

Optical transmission Shore Ship unit side Optical side fiber cable Tel (6-core) Cable reef Optical plug in connector Hotline

Tel Hotline Modem

Modem

Mooring tension monitoring system

Wind direction/velocity Wave height tide level Strain gauage

Mooring hook

Mooring tension monitoring system Emergency shutdown system schematic–LNG

Figure 6.15 Typical emergency shutdown systems associated with LNG unloading (SIGTTO, 1986). ESDS, emergency shutdown system; ESV, emergency shutdown valve (reproduced by permission of SIGTTO). Table 6.4

Thermal properties of three substrates

Material Concrete Dry soil Insulating concrete

Thermal Conductivity, kT (W/[m2K])

Thermal Diffusivity, α (m2/s)

Assumed Porosity

0.880 0.320 0.220

5.86 × 10−7 2.44 × 10−7 8.27 × 10−7

0.1 0 0

In Section 6.4.4, the vapor dispersion zone for an LNG import terminal is defined by U.S. regulations as the distance to the ½ LFL of a vapor cloud from a 10-min release of a complete break of a transfer line. From Section 6.1, the nominal design unloading rate is 10,000 m3/h. This is 2.78 m3/s or between 1197 kg/s (Trinidad LNG) and 1288 kg/s (Oman LNG). In 10 min, 10,000/6 = 1667 m3 would be spilled. This spill would be impounded first by a collection trench, possibly lined with insulating concrete. The trench would lead to the impoundment basin, possibly with a sloping floor of insulating concrete to minimize the pool area and the heat transfer rate to evaporate LNG. The SafeSite3G® model predicts that 4.6% of the LNG evaporates before reaching the impoundment sump, or for Trinidad LNG, 718,000 kg is

126

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

spilled and 685,000 kg reaches the sump. An impoundment sump with dimensions to hold the entire 1667-m3 spilled volume (not allowing for evaporation before reaching the sump) would, for example, be a sump 20.4 m2 × 4 m deep. This is likely to be inside the diked area sized for the contents of an entire storage tank of 120,000 m3 (half of the terminal capacity), or a dike impoundment 173 m2 and 4 m deep. Taking only the smaller impoundment sump and the “design spill” from the full break of a loading arm for 10 min produces the predicted evaporation rate and cumulative evaporated vapor plotted in Figures 6.16 and 6.17. In Figures 6.16 and 6.17, the evaporation rate curve decreases with time by cooling both the substrate and the liquid by evaporation. The evaporation rate is highest for concrete, followed by dry soil and insulating concrete, although initially, concrete and dry soil have about the same evaporation rate. As a quantitative comparison of the effectiveness of insulating concrete, from Figure 6.17 at 2400 min (40 h), the predicted mass of cumulative evaporated vapor is shown, respectively, in units of 1000 kg (t): Concrete 246

Dry Soil

Insulating Concrete

223

194

or as a percent of the total liquid reaching the sump, 36.0%, 32.6%, and 28.3%, respectively. 6.7.2

Vapor Hold-Up Effect on Dispersion Zone Calculation

Evaporation rate (kg/s)

As discussed in Section 6.4.4, U.S. regulations require a vapor dispersion zone (as well as a thermal dispersion zone) to safely separate the public from

14 12 10 8 6 4 2 0 0

10

20

30 40 Time (min)

Insulating concrete evaporation rate Concrete evaporation Figure 6.16

50

60

Dry soil evaporation

Evaporation rate in impoundment sump with three substrate materials.

14 12 10 8 6 4 2 0 0

600

1200

1800

280 240 200 160 120 80 40 0 2400

127

Cumulative vapor (kg/1000)

Evaporation rate (kg/s)

LEAK IN LOADING ARM OR IN STORAGE TANK

Time (min) Insulating concrete evaporation rate Concrete evaporation

Dry soil evaporation

Dry soil cumulative vapor

Insulating concrete cumulative vapor Concrete cumulative vapor

Figure 6.17 Evaporation rate and cumulative evaporated vapor for design loading arm break into impoundment sump with three different substrates.

onshore LNG terminals. Havens and Spicer (2006a, 2007a) have raised an issue regarding the procedure to set the vapor dispersion zone. The issue for setting the vapor dispersion zone is that the prescribed modeling does not account properly for the physics of dispersion of vapors over a dike wall. Specifically, air entrainment dilutes vapors in the sump, and the “scooping” effect of wind over the sump begins the overflow of vapor out of the sump earlier than is erroneously assumed by ignoring these effects. Explaining further, the evaporation rate from an impoundment sump is the source rate for the dispersion plume. Since the evaporation rate is time varying, what point in time should be used for the evaporation rate? As discussed in Section 7.2, the evaporation rate of LNG, Ev, depends on the pool area, Apool (in m2), the total heat reaching the pool, QTotal (in W/m2), and the heat of vaporization of LNG, Hevap (in J/kg): Ev =

ApoolQTotal H evap

(6.4)

There are several paths for heat transfer to the evaporating pool, but the main source is heat conduction from the substrate. Evaporation cools the pool and the substrate, so the substrate temperature decreases in time. It is important to examine the curve of time-varying evaporation rate to see how important the error under dispute is. To this end, continue the example case based on a design event specified by U.S. regulations, a full-bore rupture of a transfer line from an LNG carrier to a storage tank, leaking for 10 min (introduced in Section 6.7.1.). For this case, we used Trinidad LNG pumped at the transfer pipe conditions in Table 6.5. This table also gives the dimensions for a sump inside a dike with the

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

10

200

Entrained air

8

160

6

120

4

80

2

40

0 0

600

1200

1800

Cumulative vapor (kg/1000)

Evaporation rate (kg/s)

128

0 2400

Time (min) Insulating concrete evaporation rate

Figure 6.18 sump.

At T1, 100%

At T2, 62%

Insulating concrete cumulative vapor

At T1, 100%

At T2, 62%

Example of LNG evaporation rate from evaporation sump with 10-min leak to

volume to hold the contents of a 120,000-m3 storage tank. Pool evaporation modeling predicts the evaporation rate versus time and the cumulative mass of evaporated vapor for a sump lined with insulating concrete as plotted in Figure 6.18. As the FERC regulation currently reads, the hold-up time t1 is taken to be the time when the accumulated evaporated vapor volume, without allowing for air dilution, Vvapor, equals the dike hold-up volume, Vdike, or t1

Vvapor = ∫0 Ev dt = Vdike

(6.5)

In this case, the dike volume is 120,000 m3; t1 is 1967 min, and at t1, the evaporation rate is 0.815 kg/s. This rate is allowed by the current regulations as the source strength for dispersion modeling. However, allowing for air entrainment, the actual time for the plume to overflow the dike is at an earlier time, t2. In Figure 6.18 and in Table 6.5, t2 is arbitrarily indicated at 595 min, where the cumulative evaporated LNG vapor is only 96,000 kg, filling 66,640 m3. The remaining volume of 53,360 m3 must be made up of entrained air and expansion by warming. With air at 293 K (68°F) and 75% relative humidity, the mass of entrained air with ambient air density of 1.196 kg/m3 is 63,830 kg. This air dilutes the LNG vapors to 60.0 mass % or 72.3 mole %. With these mass and volume values, the mixture would have a density of 1.33 kg/m3 and a temperature of 162.8 K. At the earlier hold-up time of t2, the evaporation rate is higher than at t1. Moderating this argument, the evaporation rate curve is relatively flat after 600 min, so the evaporation rate at t2 is only about 60% higher than at t1. However, accounting for air entrainment also produces an initial dilution by air entrainment, with a warmer plume, and a lower initial plume density, so the effect on the distance to ½ LFL is considerably moderated.

ROLLOVER

129

Table 6.5 Dimensions for example case sump and storage tank inside dike

Transfer Pipe from Carrier to Storage Diameter, m (inch)

Temperature (K)

0.356 (14)

112.2

Pressure, kPa (psig)

Discharge Rate (kg/s)

517 (75)

1,197

Loss in 10 min (m3) 1,666.7

Mass Loss, (t) 718 spilled 33.2 vapor 685 liquid

Dike Length (m)

173.2

Width (m)

173.2

2

Area (m ) Net Area

Height (m)

30,000

4.0

Volume (m3)

Average Vapor Temperature (K)

120,000

140

Volume (m3)

Level (m) at 10 min

Sump Length (m) 20.4

2

Width (m)

Area (m )

Depth (m)

20.4

416.7

4

1,666.7

3.73

Pool Evaporation Model Predictions Time Point (see Fig. 6.18) 1 2—LNG Vapor 2—Air

Time (min)

1967 595

2—Mixture

Cumulative Vapor Volume (m3)

ρvapor (kg/m3)

173,000 96,000

120,000 66,640

1.441 1.441

0.815 1.31

63,830

53,360

1.196 air

Tair = 293 K

Mass fraction 0.601

Mole fraction from LNG 0.723

1.332 mix

Tmix = 182 K

Cumulative Vapor (kg)

Evaporation Rate (kg/s)

The justified conclusion is that better modeling is needed to accurately apply the prescriptive requirements of the FERC requirements to calculate the vapor dispersion zone. 6.8

ROLLOVER

The first documented LNG rollover incident occurred in 1971 at La Spezia, Italy Import Terminal. “The LNG carrier Esso Brega had been in the harbor for about a month before unloading its ‘heavy’ LNG into the storage tank.

130

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Thirty one hours after the tank was filled, the tank developed a sudden increase in pressure causing LNG vapor to discharge from the tank safety valves and vents over a period of a few hours” (Heestand and Meader, 1983; CHIV, 2003). The roof of the tank was slightly damaged and about 100 million ft3 of LNG vapor was released but did not ignite. This accident was caused by two layers of LNG having different density and heat capacity. When the heavy layer is laid down on top or the bottom layer warms and becomes the lighter layer, this inherently unstable situation allows the sudden mixing of these two layers. The heavy layer in mixing warms the lighter layer, resulting in the release of a large volume of vapor. The La Spezia incident was well documented by Sarsten (1972), who provided the conditions listed in Table 6.6. The density of the Esso Brega cargo was heavier than that of the initial heel by 0.7%, and the vapor pressure of the cargo was much higher than that of the heel. Such subtle difference was hardly expected to produce a hazard. The rollover event was modeled by Heestand and Meader (1983) using conventional heat and mass transfer correlations and a four-component model of methane, ethane, propane, n-butane, and nitrogen. Inclusion of nitrogen was deemed vital to the success of the modeling for its effect on vapor pressure and liquid density. Natural convection processes “weathered” the top layer of LNG over time, driven by mild heat transfer processes that set up circulating cells indicated in the left-hand side of Figure 6.19. At 31 h after delivery, an upflowing element from the bottom cell penetrated all the way to the top of the liquid and produced the sudden change in temperature difference between the top film and the bulk liquid indicated in the right-hand side of Figure 6.19. Spot trade in LNG adds diversity to sources and composition, increasing the possibility of a sequence that could lead to stratification in the tank. In addition, LNG peak shaving tanks could have rollover problems because tanks are allowed to age.

Table 6.6 Initial conditions for La Spezia rollover

Condition Density (kg/m3) Temperature (K) Vapor pressure (Pa) Liquid depth (m)

Initial Heel 541.74 114.36 3,923 5.03

Esso Brega Cargo 545.59 118.99 16,280 17.83 Composition (mol %)

Methane Ethane Propane Butanes Pentanes Nitrogen Total

63.62 24.16 9.36 2.35 0.16 0.35 100.00

62.26 21.85 12.66 3.14 0.07 0.02 100.00

131

ROLLOVER

qt

Mv

qwv

Cell N

Cell i

qWL

Cell l qb

Temperature difference TBULK – TFILM (K)

Mout

1.2 1.0 0.8 0.6 0.4 Rollover 0.2 0

0

20

40 Time (h)

60

80

Figure 6.19 Modeled form and prediction for La Spezia rollover incident (Heestand and Meader, 1983) (reproduced by permission from AIChE).

An experimental study of rollover by Morrison and Richardson (1990) verified the above understanding of the behavior. Rollover has very largely been eliminated by designing tanks with the following features (Durr, 2004): • • •

Monitor temperature to avoid excess heat in bottom liquid layers. Provide mixing loop to avoid stratification. Use tank fill methods to promote mixing: • jet mixing, • bottom loading via standpipe, or • top loading via splash plate.

Techniques are used to promote mixing of light material when bottom loading: • • • • •

Flash the liquid to dissipate excess heat. Break vacuum in the loading line. Dissipate free-fall energy. Avoid vapor entrainment. Use slots at bottom on standpipe to promote mixing.

Similarly, methods are used when top loading heavy LNG: • • • • •

Flash the liquid to dissipate excess heat. Avoid a potential vacuum in the loading line. Dissipate free-fall energy. Spread a thin film of liquid. Do not splash onto wall or near vapor outlet.

132

RISK ANALYSIS FOR ONSHORE TERMINALS AND TRANSPORT

Figure 6.20 Example of a satellite LNG storage and LNG truck (IELE, 2003b). Source: CH-IV International.

6.9

LNG LAND TRANSPORT RISK

Small-scale LNG tanks are important in adding “peak shaving” flexibility to natural gas pipeline grids. As of 2003, there were 240 LNG facilities worldwide, with 113 in the United States, including 96 connected to the U.S. pipeline grid. Satellite LNG storage capacity collectively totaled 80% (2.44 billion cubic meters [bcm]) of the U.S. total, while marine import terminals accounted for 18% (0.53 bcm) and the export terminal in Kenai, Alaska, accounted for 2% (IELE, 2003a). The largest facilities usually liquefy natural gas drawn from the interstate pipeline grid, but many smaller facilities without liquefaction capabilities receive LNG by truck (Parfomak, 2004). Figure 6.20 shows an example of a “satellite” LNG storage tank and also a photo of an LNG delivery truck serving these tanks. LNG trucks are of more robust construction than typical fuel trucks. LNG is notably extensively transported by trucks and barges in China where pipeline networks are not well developed. Satellite LNG tanks are generally surrounded by containment impoundments that limit the spread of an LNG spill and the potential size of a vapor cloud (Gaul and Young, 2003). The risk in land transport systems is that of highway collisions, truck rollover, spills upon loading the storage tanks, and storage tank leaks. The scale is smaller, but the event frequency is higher than for LNG import terminals and regasification systems. 6.10

OFFSHORE LNG TERMINALS

Offshore LNG terminals have been proposed and designed primarily to overcome objections by the public to onshore terminals near areas of high popula-

OFFSHORE LNG TERMINALS

133

tion density. It is recognized that offshore terminals may have security benefits for seacoast communities but may increase the vulnerability of the terminals themselves. Aldwinkle and Slater (1983) discuss risk and reliability analysis methods appropriate for a particular type of offshore LNG terminal, a floating offshore liquefaction and storage ship secured by a single-point mooring to an underwater pipeline. Conventional risk assessment and reliability methods are recommended with the caveat that the estimation of failure frequencies is difficult. The authors point out: “The choices are either to use a data base of known incidents (or destructive tests) or to extrapolate from ‘relevant’ data.” Both are, by definition, difficult for novel concepts. Fortunately, fault tree analysis techniques are appropriate to build up predicted failure rates from component events for which relevant data are available. This is coupled with failure modes and effects analysis (FMEA) to relate the consequences to the postulated failures. The following major systems were identified for fault tree analysis leading to top events of gas leakage or LNG spillage: 1. 2. 3. 4. 5.

single-point mooring failures, liquefaction process plant failures, containment system failures, LNG piping system failures, and LNG transfer arm failures.

These methods are used to evaluate the cost/benefit ratio for proposed mitigation measures. In the case of single-point mooring failures, mitigation consisted of measures to protect from collisions with shuttle ships that load the LNG for delivery to markets. The options evaluated were 1. a collision barrier and 2. a sacrificial structure The evaluation requires calculation of the energy-absorbing capability of structures. The approach illustrates the important role risk analysis can play in the design of novel concept structures. Since that time, much work has been done on offshore floating concepts. Some important contributions have been made for the FLEX LNG concept, the Hoegh LNG concept, and others (Pastoor et al., 2009; Festen and Leo, 2009; Iversen and Hellekleiv 2009). Also important have been updates to classification rules by Det Norske Veritas (DNV), American Bureau of Shipping (ABS), and Lloyd’s Register (LR) to address the special risks of floating LNG liquefaction facilities and to approve novel technologies such as floating LNG pipes using technology qualification.

7 LNG POOL MODELING

This chapter discusses the source term modeling that ultimately affects hazard zones from LNG vapors and fire. Topics addressed include: jet flow, aerosol formation, penetration into water, pool spread, and evaporation rate possibly enhanced by rapid phase transition (RPT). These phenomena are discussed for spills on both land and water. A moderate to large Type 1 LNG release (elevated above the water level) was shown in Chapter 5 to spill predominantly into the sea. LNG jet spills into water penetrate to some depth and mix, thereby increasing the area of LNG/water contact and the evaporation rate. Type 2 and 3 spills (at the water level or below the water level) will mix LNG and water in the double hull space. Consequently, with all three spill types, there is a potential for RPTs, a physical explosion without combustion. This is developed in Section 7.3. The dominant mechanism for heat transfer to an LNG pool, both on land and on water, is from conduction from the substrate. The heat gain from air convection or by solar and long-wave radiation accounts on average less than 5% of the total heat transferred to LNG pools (Webber, 1990; Cook and Woodward, 1993; Cavanaugh et al., 1994). Section 7.5 treats the equations for the heat transfer mechanisms of air convection, solar heating, long-wave radiation, and nonboiling evaporation by mass transfer. In this chapter, we show that there is good science and data underpinning most aspects of source term modeling and that reasonable estimates are possible, either from first principles or from computer codes that use these approaches. Scale is an issue, as most experiments are carried out at LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

134

FLASHING AND DROPLET EVAPORATION IN JET FLOW

135

reasonably small scale compared with the very large spills that are possible from marine carriers or storage tanks. Some differences in modeling remain to be resolved by definitive experiments. Some interesting physical phenomena are elucidated in this chapter. For a Type 2 or a Type 3 breach, water inflow or water and LNG inflow into the double hull space builds pressure in this space and hinders further inflow of water and outflow of LNG. Some water may freeze and even obstruct the outflow of LNG. There is the potential that seawater could intrude into the LNG tank. This is shown to lead to a tank pressure increase by water freezing and transferring heat to evaporate LNG in the tank. In addition, Type 1 releases of LNG outflow tend to produce a partial vacuum in the LNG tank, and vacuum relief may not be adequate for larger outflow rates. A partial vacuum in the headspace would decrease LNG outflow and a glug-glug effect could develop, inducing air or water into the tank.

7.1

FLASHING AND DROPLET EVAPORATION IN JET FLOW

LNG in storage or in transport below the surface level is always under some hydrostatic pressure. With this hydraulic pressure, punctures of the tank or of pipes below the surface level will produce a pressurized discharge. An accidental release of LNG under pressure develops a jet flow. The liquid jet breaks into an aerosol. Droplets partly or fully evaporate, or some liquid rains out to form a pool. Jet flow, droplet evaporation, and pool evaporation are treated in detail in a Center for Chemical Process Safety (CCPS) Concept Series book by Johnson and Woodward (1999), and in Lees (Mannan, 2005, pp. 15/63–68). The pumped discharge example presented in Section 5.4 shows a jet flow converting into an aerosol with cloud formation and no rainout reaching the sea surface. The sequence is illustrated in Figure 7.1, showing that the aerosol droplets evaporate and decrease in size along a trajectory. The process is usually modeled with a monodisperse mixture, letting the mean drop size represent the distribution of actual drop sizes. When the mean-sized drop reaches the ground, the liquid fraction in the aerosol at that point is taken as the rainout fraction. That is, the rainout rate is the liquid source rate to the pool. The evaporating vapor from the drops and from the pool combine to develop the plume until the discharge stops. After that, only evaporation from the pool sustains the plume. Examples of LNG aerosol jet plumes with rainout are provided in Section 5.7.2. Discharged LNG is a boiling liquid at atmospheric pressure, so there is normally no flash—that is, excess internal enthalpy due to a temperature above atmospheric liquid temperature and converted into latent heat of vaporization. Thus, droplet formation is physical and not due to flash effects and evaporation in air is solely due to heat extracted from ambient air.

136

LNG POOL MODELING

(Flashing) two-phase discharge from pipe/vessel

Vapor-plume centerline

Droplet trajectory

Point of rainout

Spreading evaporating liquid pool

Figure 7.1 Representation of modeling of jet flow with monodisperse aerosol raining out to form evaporating pool (Witlox et al., 2007) (reproduced by permission of Elsevier Science Publishing, Inc.).

7.2

POOL SPREAD AND EVAPORATION MODELING

A steady-state pool will form from a constant spill rate of LNG if the spill duration is long when the evaporation rate from the pool equals the discharge rate into the pool. In this section, a number of detailed modeling equations are presented and terms are defined on their first occurrence, and for convenience, a nomenclature section is provided at the end. Alternate methods are available to treat the spread and evaporation of LNG on water or land. The simplest numerical approach is to integrate two equations simultaneously, an equation for spread rate and an equation for evaporation. The numerical procedure is outlined below with details for the heat balance terms provided in Sections 7.2.4 and 7.5. Step 1. The mass evaporated is determined from an energy balance to the pool and the heat of evaporation, ΔHVL. The heat balance terms, in Watts, are inputs from conduction through the substrate, Qcond, convection from wind, Qconv, and solar radiation, losses from Qrad, long-wave radiation from the pool to the surroundings, Qlong, and from evaporation, Qevap. If the pool is boiling, the net balance to the pool, Qpool, equals the heat of evaporation, Qevap,. Qpool = Qcond + Qconv + Qrad + Qlong Qevap

Qpool if TL = Tsat ⎧ ⎪ =⎨ ⎛ 1 − Pvap Pa ⎞ ⎟ ⎪ΔH VL km ApρL ln ⎜⎝ 1 − P ⎩ back Pa ⎠ Qevap Fevap,i = ΔH VL

if TL < Tsat

(7.1) (7.2)

(7.3)

POOL SPREAD AND EVAPORATION MODELING

137

For a nonboiling pool, the net balance, Qpool, establishes the pool temperature, and the evaporation rate is found from the mass transfer equation involving the mass transfer coefficient, km, the vapor pressure of the pool, Pvap, and the back pressure of the vapor above the pool, Pback, and the atmospheric pressure, Pa. The mass evaporation rate, Fevap, is in kg/s; the evaporation flux, Gevap, is Fevap divided by the pool area, in kg/(m2s) and equals uvρv, the evaporation velocity, uv, times the vapor density, ρv: Step 2. At each step in time, add an increment of liquid from the discharge or aerosol rainout and subtract current evaporated liquid. mLi = mLi −1 + FL iΔt − Fevapi Δt

(7.4)

Step 3. Update the pool volume, VL, and average pool thickness, h: mLi ρL

(7.5)

VLi πRi2−1

(7.6)

VLi = hi =

Step 4. Use pool spread correlation to update pool radius, Ri. dR ⎞ Ri = Ri −1 + ⎛ (t − t ) ⎝ dt ⎠ i i −1

(7.7)

For cryogenic liquids, including LNG, evaporating on water, the evaporation rate flux, Gevap is, as a first approximation, constant: Gevap =

Fevap Fevap = Apool πR 2

(7.8)

If the discharge rate is constant, onto deep water, a steady-state pool area is developed in a reasonably short time. At steady-state, the discharge rate, Fdis, equals the evaporation rate, Fevap, So, the observed pool area and the known discharge rate can be used to infer the evaporation flux, or inversely, the equilibrium pool radius, Req, can be inferred from the known evaporation flux: ⎛ F ⎞ Req = ⎜ dis ⎟ ⎝ πGevap ⎠

12

(7.9)

The equilibrium pool radius is a useful scaling factor as seen below. Qiao et al. (2006) plot the steady-state pool radius against the hole size and show that this curve saturates above hole sizes of about 5 m, approaching the hole size of a semi-instantaneous spill.

138

LNG POOL MODELING

7.2.1

Spread Rate on Smooth Surface

First, we treat the more general case of spreading on water and then extend to spreading on a smooth solid surface. It has been shown for “box” models originally developed for oil spills (Fannelop and Waldman, 1972; Hoult, 1972a) that the radial spread rate of a cryogen spreading on water, R(t), is proportional to the potential energy, represented by the pool height, h, (or depth) of leading edge. The radial spread velocity, dR/dt or uL can be written in terms of the volume of liquid in the pool, V, or the mass and density of liquid in the pool, mL and ρL: dR gVΔ 12 = kS ( hg ′ ) = kS ⎛ 2 ⎞ ⎝ R ⎠ dt

12

⎛ m g′ ⎞ = kS ⎜ L ⎟ ⎝ ρL Ap ⎠

12

⎛ m g′ ⎞ = kS ⎜ L ⎟ ⎝ ρL π ⎠

12

1 R

(7.10)

where the effective gravitational acceleration, g′, accounts for part of the LNG sinking into water: g ′ = gΔ where Δ=

(ρw − ρL )

(7.11)

ρw

This assumes that the source is just above the water surface, and that its impact in disturbing the surface is minimal, so the water surface can be considered as essentially flat through the spreading process. The pool spreads as a perfect cylinder, although for an instantaneous spill, the surface is convex, being thinnest at the center (Weber and Brighton, 1986, 1987). Furthermore, the original derivation assumes that the release duration is very much shorter than the spreading time, thus applying to an instantaneous release. Extending the derivation to continuous releases, the average pool height is used instead of the leading edge height, and the concave surface becomes convex, thicker in the center, or more nearly a flat surface. This is illustrated in Figure 7.2. The average height is found by dividing the liquid

Source Continuous Instantaneous

h

2R

Figure 7.2 Illustration of pool thickness profiles with instantaneous and continuous discharge.

POOL SPREAD AND EVAPORATION MODELING

139

volume (mass, mL over liquid density, ρL, by the pool area, Ap. When the pool area term is applied, (πR2), the spread rate is seen to be inversely proportional to the pool radius. There is a theoretical value for the spreading constant kS of 1.16, but using the average pool depth, h, implies that the spreading constant must be adjusted upward to match the experimental data. The most commonly suggested value is 1.41 (i.e., the 2 ) (Brisco and Shaw, 1980), although 1.69 is suggested by several developers as summarized by Conrado and Vesovic (2000). Analytically integrating Equation 7.10 gives (Fay, 2007): R = km ( gVΔ ) t 1 2 14

⎛ k 2 ⎞ gVΔ uL = ⎜ m ⎟ ⎛ 2 ⎞ ⎝ 2 ⎠⎝ R ⎠

12

gVΔ ⎞ = kL ⎛ ⎝ πR 2 ⎠

12

(7.12)

The Hoult parameter, km, is a dimensionless value of 1.26, inferred from laboratory measurements of spreading in a one-dimensional channel (Fay, 2003). This parameter is related to the Froude number, Fr, of the pool front edge by: Fr =

uL 1⎞ ⎛ 4 =⎜ 2 + ⎟ ⎝ πkm 4 ⎠ gHΔ

−1 2

(7.13)

where H is the thickness of the pool at the leading edge. For Hoult’s solution, Fr = 1.16. The ratio of the kinetic to the potential energy in such flow is 1 proportional to (Fr)2. The value of kL is km2 π = 1.41 . 2 The above spreading relationships are adequate until the pool thickness reaches a minimum stable value that depends on the surface on which the LNG is spilled. In two tests on open water conducted by Esso, the minimum stable thicknesses were about 4.4 mm and 6.7 mm (May et al., 1973). The pool volume decreases in time by evaporation, according to: dVp Gevap = ( πR2 ) dt ρL

(7.14)

where, as discussed later, the evaporative flux, Gevap is the evaporation rate, Fevap in kg/s, divided by the pool area, Apool, in m2, and also equals an evaporation velocity, uv in m/s, times the vapor density at the evaporating temperature, ρv: Gevap =

Fevap = uvρv Apool

(7.15)

140

LNG POOL MODELING

Equations 7.10 and 7.14 can be integrated simultaneously for an instantaneous spill, where the pool volume, Vp at t = 0 equals the spill volume, V. The maximum pool radius, Rm, is reached at the evaporation time tevap (Fay, 2007). Fay’s example implies that the evaporation velocity uv in m/s should be divided by a factor of 1000. ⎛ 64kS2 ⎞ Rm = ⎜ ⎝ 9π 3 ⎟⎠ ⎛ 3 ⎞ tevap = ⎜ ⎝ 8kS ⎟⎠

12

18

⎛ gVΔ 3 ⎞ ⎜⎝ 2 ⎟⎠ uv

18

⎛ gVΔ 3 ⎞ = 0.906 ⎜ 2 ⎟ ⎝ uv ⎠

18

⎛ Γ ( 5 3) ⎞ ⎛ V ⎞ ⎛ V ⎞ ⎜⎝ Γ ( 7 6 ) ⎟⎠ ⎜⎝ gu2 Δ ⎟⎠ = 0.502 ⎜⎝ gu2 Δ ⎟⎠ v v

(7.16) (7.17)

where Γ(x) is the gamma function of argument x. The coefficients have been evaluated for ks = 1.41. Interestingly, the product Rm2tevap does not depend on kS. The value of uv is evidently meant to be in mm/s, as evidenced in Fay’s example case. 7.2.1.1 Multiple-Regime Models for Spread on Water Most generally, the spread of liquids that dissolve, evaporate, and spread on water should account for different flow regimes in which the dominant forces vary as follows: • • •

Gravity-inertia regime: Equate gravity spreading and inertial spreading Gravity-viscous regime: Equate gravity spreading and viscous resistance Surface tension regime: Equate viscous drag and surface tension

Equation 7.10 applies to the gravity-inertia regime. In using Equation 7.10, it is assumed that most of the cryogen will evaporate before the viscous and surface tension forces become important. A more generalized solution is provided by Dodge et al. (1983) for a continuous release of rate Fdis(t) by equating the dominant forces for each regime and integrating to obtain an analytical solution. The results are given below in terms of the mass of liquid in the pool, mL, the average pool depth, h, the density of spilled liquid, ρL, the viscosity of the liquid, μL, water, μw, and air, μa, the pool radius, R, and time since the spill, t. We retain the definition of Δ in Equation 7.11. Gravity-inertia regime applicable for 0 < t < t3 R ( t ) = 1.24 ( gΔ ) Fdis ( t ) t 3 4 14

14

(7.18)

Gravity-viscous regime applicable when t > t3 16

⎡ ⎤ ⎢ ⎥ gΔ ⎥ Fdis ( t )1 3 t 7 12 R ( t ) = 1.09 ⎢ 12 ⎢ ⎛ μw ⎞ ⎥ ρL ⎥ ⎢⎜ ⎟ ⎝ ⎠ ⎣ ρw ⎦

(7.19)

POOL SPREAD AND EVAPORATION MODELING

141

The transition between these regimes occurs at the time t3 given by: 6

1.09 ⎞ ⎛ 1 ⎞ t3 = ⎛ F (t ) ⎝ 1.24 ⎠ ⎜⎝ μ wρL gΔ dis ⎟⎠

12

(7.20)

A solution is given also for an instantaneous spill on water, in similar terms. Approximations are involved in applying these integral solutions with timevarying source rate and rapid evaporation since evaporation reduces the average liquid height, h, at each time step. 7.2.1.2 Alternate Spread Rate Formulas on Water, Variations on Standard Model Several other pool spread models are in use, including early models by Fay (1973), and Shaw and Brisco (1978). Weber and Brighton (1986, 1987) utilize a different procedure based on self-similar solutions to the shallow layer inviscid flow equations and obtain: 4 ( s − 1) ⎞ ⎛ gVΔ ⎞ du L = −⎛ ⎝ dt π ⎠ ⎝ R3 ⎠

(7.21)

Their shape parameter is s = 1.50. The FERC-ABS Consulting (2004) report evaluates two pool spread models in addition to Equation 7.10: Raj and Kalelkar (1974) developed a gravitational-inertia regime model by equating a gravitational force, FG, with an inertial force, FI, defined by: FG = πrh2ρL g ′ and FI = −C ( πr 2 hρL )

d2r dt 2

(7.22)

The constant C is a fraction accounting for the fact that only part of the inertial in the entire pool is manifest in the acceleration of the leading edge. Equating these forces gives: d 2r g ′h =− 2 dt CrρL

(7.23)

Upon numerically integrating Equation 7.23, the pool spread rate is about the same as that obtained from Equation 7.10 as shown by Otterman (1975). None of the above formulas introduced frictional resistance. Webber and Brighton (1987) as described by TNO (1997b) introduced a frictional resistance term, Cf, as follows: d 2 r 4 g ′Φh = − Cf dt 2 r

(7.24)

where Φ is a coefficient that is a function of hf/h where hf is the leading edge height and h is the average pool height. Webber and Brighton provide methods for estimating Cf and Φ for turbulent and laminar friction regimes.

142

LNG POOL MODELING

7.2.1.3 Supercritical Pool Spread Model The above models are collectively termed by Fay (2007) variations of the standard model. Fay (ibid.) points out that there are good physical reasons inherent in Equation 7.10 why the above models of Hoult (1972a, b) and Webber and Brighton (1987) may not always apply for LNG spills. The standard models are based upon laboratory experiments and analyses of oil pool spreading, and there are substantial differences between oil and LNG. For an oil thickness of h, the upper surface of an oil pool is elevated above the sea by hΔ, the “tip of the iceberg.” The submerged portion of the oil layer is of depth (1 − Δ)h, and displaces seawater as the pool moves radially, much like a ship parts the sea as it moves ahead. This motion of the sea provides resistance to the pool spread, and decelerates the radial pool motion. The ratio of the emerged and submerged thicknesses, Δ/(1 − Δ), is a measure of the relative vertical displacements of the top and bottom of the oil pool. Especially near the front edge of the pool, the flow is strongly dependent on this ratio. For the oil used in Hoult’s experiments, Δ = 0.1 and Δ/(1 − Δ) = 0.11. For LNG, Δ = 0.58 and Δ/(1 − Δ) = 1.38. This geometry alone makes the flow at the pool edge quite different for LNG. Fay (2007) also argues that the rapid boiling of LNG on seawater means that vapor bubbles will occupy a significant portion of the pool volume, reducing the average density of LNG and thereby increasing Δ above that for the pure liquid LNG. In Section 7.5.5, Fay’s arguments for bubble diameter and velocity are summarized, leading to his conclusion that bubbles would increase Δ to about 0.8. For a pool with this relative density, about 80% of the pool volume would lie above the sea-surface level. Concerning Weber and Brighton’s introduction of viscous effects into the spreading model, as discussed subsequently, Fay argues from heat transfer considerations that a boiling LNG pool is predicted to ride on a thin film of LNG vapor with much lower viscosity than that of water or LNG. When this is true, an evaporating LNG pool can be regarded as essentially inviscid flow. “This nearly frictionless motion is related to the Liedenfrost effect, where small droplets of liquid skitter about on a solid surface heated to well above the liquid boiling point” (ibid.). The supercritical model follows the approach of Weber and Brighton (1987), so the spreading velocity, uL, and liquid height, h, averaged over the pool radius, are given by: R t

(7.25)

V V = πR 2 π (uL t )2

(7.26)

uL = h=

where the pool spread velocity is set equal to the liquid jet discharge velocity, udis, given by the square root of the potential energy of the liquid head in the tank, hL:

POOL SPREAD AND EVAPORATION MODELING

uL = udis = 2 ghL

143

(7.27)

During the blowdown of liquid, discussed in Chapter 5, the LNG level in the tank drops, so hL is a function of time, hL(t). As a result, the general relationship is that the ratio to the initial values, uL0, R0, and level hL0 retains the similarity relationships as the decreasing ratios: uL2 R 2 hL = = uL2 0 R02 hL0

(7.28)

where R is the maximum pool radius if steady state is attained. As a consequence, the pool kinetic energy stays equal to the initial potential energy in the tank level (varying in time as the tank drains), and the average kinetic energy equals the average potential energy. The spread model essentially predicts the spread across a solid horizontal surface and g is not a parameter in the solution. Combining pool spread with evaporation, the rate of pool volume reduction is: dVp Gevap G = ( πR2 ) = evap ( πuL2 t 2 ) dt ρL ρL

(7.29)

This equation is integrated to give the evaporation time, tevap, when the pool volume is reduced to zero. At this time, the pool radius is predicted by the supercritical model to reach a maximum. By Fay’s example, there must be a factor of 1000 in the denominators: ⎛ 3uLV0 ⎞ Rm = ⎜ ⎝ πuv 1000 ⎟⎠

13

3V ⎛ ⎞ tevap = ⎜ ⎝ πuv uL2 1000 ⎟⎠

(7.30) 13

(7.31)

Table 7.1 compares predictions of the standard model using Equations 7.16 and 7.17 with those of the supercritical model using Equations 7.30 and 7.31, recognizing that Equations 7.16 and 7.17 apply to an instantaneous spill. This

Table 7.1

Example predictions by alternative analytic pool spread formulas

Model Supercritical Standard Ratio Supercritical/standard

Maximum Pool Radius (m)

Evaporation Time (s)

1190 350 3.4

65 324 0.20

144

LNG POOL MODELING

2

1 R/Req 10

20 Ut/Req

Figure 7.3 Comparison of China Lake experiments with LNG pool spread models (Fay, 2007) (reproduced by permission of Elsevier Science Publishing, Inc.).

gives, for Algerian LNG with properties in Table 1.1, V = 10,000 m3, Δ = 0.547, and Gevap = 0.185 kg/(m2s), uv = 0.104 m/s, initial LNG level of 17 m, and uL = 18.3 m/s from Equation 7.27 applied to both models for comparison: There is a large factor between the two models’ predictions. Upon comparison with test data from the China Lake series, the data seem to support both the standard model and the supercritical model depending upon the spill rate. Figure 7.3 plots the dashed and dot-dashed lines from China Lake tests 5 and 12. The solid line and the dotted line are predictions of the supercritical and the standard models, respectively. The supercritical model predicts overshoot of the pool radius (plotted as the ratio of radius to the steady-state or equilibrium radius Req), as is supported by test data for a rapid release. The standard model also predicts some overshoot, but later in the response at a distance uLt (or labeled Ut) normalized to Req. The predicted times for R to first reach Req are 3 and 30 for the supercritical and standard models, respectively. 7.2.2 Pool Spread on Land Equation 7.10 applies for pool spreading on flat ground with g′ replaced by g, the gravitational constant. A porosity factor, ε, represents the volume fraction of pores in the substrate. The volume of spreading liquid is decreased by ε, so the pool area is decreased by a factor of 1/(1 − ε). That is, the pool radius, R, after filling the pore fraction is decreased from the pool radius without accounting for porosity, R0, by: R=

7.2.3

R0 1− ε

(7.32)

Pool Evaporation on Smooth Water Surface, Test Data

There are two approaches to finding LNG pool evaporation rates. The first is to assume that the evaporation rate flux, Gevap, is constant. The evaporation

POOL SPREAD AND EVAPORATION MODELING

145

flux is defined as the evaporation rate Fevap divided by the pool area, Apool (in kg/[m2s]). This is an appealing assumption in dealing with experimental data. It bypasses the difficulty of the second method, the more fundamental approach of predicting the heat transfer coefficients for heat conduction to the LNG. In this section, we treat the first approach. Evaporation rates are cited here for nonburning pools. Burning pools, which have substantial additional heat radiation back into the pool from the flame, are treated in Chapter 9. LNG spills on water differ from those on land since: 1. They are generally unconfined or at least minimally confined. 2. The heat transfer rate does not decrease in time because water circulation convection provides a nearly constant temperature difference. In spills on open, unconfined water, no ice has been observed (Felbauer et al., 1972; Otterman, 1975). It seems that ice formation is observed only for spills of LNG on confined, shallow water, for example, in laboratory experiments (Valencia-Chavez and Reid, 1979; Chang and Reid, 1982; Boe, 1998). As discussed in Sections 7.2.4.1 and 7.2.7, the evaporation flux is not actually constant but depends upon the degree of turbulence generated during the spill. Spill from a height penetrates the water level and mixes, so the evaporation rate would be higher near the spill and decreases farther away. In the absence of more detailed experimental data, though, the assumption of a constant evaporation flux is useful. This approach neglects RPT discussed later in Section 7.3. LNG evaporation rate experiments have been summarized, including reviews by Prince (1985), Thyer (2003), FERC-ABS Consulting (2004), and Clever et al. (2007). Table 7.2 summarizes the results of experiments with LNG and other cryogens evaporating on water. Data from Table 7.2 on the evaporation flux of cryogenic liquids on water are plotted in Figure 7.4 against carbon number. The rate for liquid nitrogen is plotted against a carbon number of zero. The average value for LNG is indicated with a line as 0.180 kg/(m2s). LNG composition is not usually reported and is likely at least partly responsible for the spread in data. The heat flux is calculated as the heat of vaporization at the boiling point times the evaporation flux, Gevap. The heat transfer coefficient is the heat flux divided by the temperature difference between water (assumed at 10°C) and the normal boiling point. 7.2.4

Pool Evaporation, Heat Transfer Regimes

Pool boiling is classified into three different heat transfer regimes as a function of the superheat temperature (temperature above the boiling point) ΔT: •

Nucleate boiling—two liquids are in direct contact and bubbles form at intervals

146

LNG POOL MODELING

Table 7.2 Evaporation rate data for cryogenic liquids on water

Spilled Liquid

Nitrogen

Gevap (kg/[m2s])

Heat Flux (kW/m2)

Heat Transfer Coefficient (W/[m2 oK])

Reference, Comments Burgess et al., 1970a

0.201 0.127 0.165 0.23 0.05 0.180

40.0 25.2 32.8 117 25.5 91.7

194 123 159 683 149 535

0.250

127

744

0.200 0.195

102 99

595 581

“

0.195

99

581

“

0.181

92

539

“

0.181

92

539

0.168

86.5

509

“

0.156 0.155 0.120 0.124 0.126 0.153 0.085

80.3 79.8 61.8 63.8 64.9 78.7 43.7

472 469 363 376 382 463 257

“

0.024

12.4

73

Ethylene

0.129

62.0

545

Ethane Propane

0.112 0.0718 0.270

54.7 30.5 115

555 586 2205

n-Butane

0.0234 0.0293 0.0290

9.0 11.3 11.2

863 1081 1070

“ Methane “ LNG (take as Trinidad LNG) “ “

LNG (take as Algerian LNG) “ “ “

“

Burgess et al., 1972 Valencia-Chavez, 1979 Drake, 1975; Dincer et al., 1976 Paranouskas et al., 1980 Chang et al., 1983 Felbauer et al., 1972; Matagorda Bay Boyle and Kneebone, 1973 Burgess et al., 1970b; Spill on pond Reid and Smith, 1978b Colenbrander and Puttock, 1983 Drake et al., 1975 Burgess et al., 1972 Koopman et al., 1979 Frenchman’s Flats, NV Puttock et al. (1982a, b) Maplin Sands Boyle and Kneebone, 1973 Reid and Smith, 1978a Ibid. Ibid. Paranouskis et al., 1980 Ibid. Reid and Smith, 1978a

POOL SPREAD AND EVAPORATION MODELING

147

Evaporation flux (kg/[m2.s])

0.3 0.25 0.2 0.15 0.1 0.05 0 0

1

2

3

4

Carbon number Figure 7.4

•

•

Measured evaporation flux for cryogenic liquids on water.

Transitional boiling—part of the contact surface is in nucleate and part in film boiling Film boiling—two liquids are separated by a vapor film

As discussed later, pure methane boils in the film boiling regime, whereas, LNG with higher content of heavier components tends to boil in the transitional boiling regime. Nucleate boiling normally gives the highest heat transfer. As illustrated in Figure 7.5, at low values of ΔT, the boiling rate increases as ΔT increases. The frequency and surface density of boiling bubbles increases as the boiling rate increases. It is generally assumed that nucleate boiling will cease at a sufficiently high superheat temperature when the interaction of liquid and vapor prevents unrestricted supply of liquid to the heating surface. At this point, a maximum, or critical heat flux, qcr, occurs at a critical temperature difference, ΔTcr. With further increases of superheat, the boiling becomes metastable and enters the transitional regime. Figure 7.5 shows the boiling rate decreasing with increasing superheat in this regime. At a point termed the minimum point temperature, ΔTmin, the heat flux goes through a minimum, qmin, which corresponds to the formation of a thin vapor film over the entire heating surface. For a saturated liquid, it is equal to the liquid Leidenfrost temperature, TLF (Baumeister and Simon, 1973). Measured heat fluxes for LNG in two heat transfer regimes are plotted in Figure 7.6. Correlations allow estimating the transition from one boiling regime to another (Collier and Thome, 1994; Conrado and Vesovic, 2000). The maximum heat flux for nucleate boiling is given by: qmax = qcr = 0.16 ΔH VLρ1v 2 [ σg (ρL − ρv )]

14

(7.33)

The minimum heat flux, qmin, corresponding to the onset of film boiling is estimated by a correlation (Kalinin et al., 1975; Opschoor, 1980a) in terms of

148

LNG POOL MODELING

Nucleate

Transition

Heat flux, q/A

Film

Convective

Peak nucleate flux

Minimum film boiling flux Temperature difference, ΔT Figure 7.5 Typical boiling heat flux curve (Baumeister and Simon, 1973) (modified from Baumeister & Simon, 1973).

Heat flux (kW m–2)

Boiling curve of methane 1000

100

10 1

10

100

1000

Tw–TSat Nucleate boiling

Film boiling

Figure 7.6 Measured heat flux for boiling methane (Wűrsig et al., 2009) (reproduced by permission of SIGTTO).

the minimum ΔT, vapor thermal conductivity, kv, vapor thermal diffusivity, αv, specific volume of vapor, vv, and the liquid and vapor densities, ρL and ρv: ⎞⎤ ⎡ g ⎛ ρL qmin = 0.18kv ΔTmin ⎢ ⎜⎝ ρ − 1⎟⎠ ⎥ v α ⎣ v v v ⎦

13

(7.34)

Correlations for the corresponding critical and minimum superheat ΔT are provided by Kalinin et al., 1975):

ΔTcr = 0.625 ( qcr σTL )

13

⎡ 10 + ⎢ 12 ⎣ (ρwCpw kw )

23

13

12 ⎡ ⎛ ρLCpL kL ⎞ ⎤ υ kL ⎢1 + 10 ⎜ ⎥ ⎝ ρwCpw kw ⎟⎠ ⎥⎦ ⎢⎣ 23 ⎛ ρv,s ⎞ 1 + 10 ⎜ ⎝ ρL − ρv,s ⎟⎠ (7.35)

(

)

⎤ ⎥ ⎦

POOL SPREAD AND EVAPORATION MODELING

149

The minimum temperature difference required for film boiling is given in terms of the pseudo-critical temperature of LNG, TC, the liquid temperature, TL, the density, ρ, heat capacity, Cp, and thermal conductivity, k, of the LNG (subscript L) and water (subscript w): 14 ⎡ ⎛ ρLCpL kL ⎞ ⎤ ΔTmin = (TC − TL ) ⎢0.16 + 0.24 ⎜ ⎥ ⎝ ρwCpw kw ⎟⎠ ⎥⎦ ⎢⎣

(7.36)

Both experimental and theoretical considerations indicate that liquid methane boiling takes place in the film boiling regime. The superheat temperature for boiling on 10°C water is (283–111)°K or 172°C, well above the the Leidenfrost temperature of methane at 161°K (Spiegler, et al., 1963). This is substantiated by observations of Drake et al. (1975) and Valencia-Chavez and Reid (1979) from experiments when no ice was formed. 7.2.4.1 Heat Transfer Correlations for Film Boiling Regime The film boiling regime is well studied and a number of correlations exist (Berenson, 1961; Sauer and Ragsdell, 1971). The correlations given by Klimenko (1981) are given below. Without ice formation, heat conduction from the water is given in terms of a heat transfer coefficient, hS, the pool area, πR2, the air temperature, Ta, and pool liquid temperature, TL, at time increment i: Qcond = hS πRi2 (Ta − TLi )

(7.37)

The heat transfer coefficient for film boiling is given in terms of the Nusselt number for film boiling, NuS, the thermal conductivity of vapor at the film conditions, kvf, and a length-scale factor, LC: hS =

NuS kvf Lc

(7.38)

σ g (ρL − ρV )

(7.39)

The length-scale factor is: LC = 2π

where σ is the interfacial tension between LNG liquid and vapor (N/m), g is the acceleration of gravity, ρL is the LNG liquid density, and ρv is the LNG vapor density. The Nusselt number is given by (Conrado and Vesovic, 2000): Laminar region:

Ar < 108 , Nu S = 0.19 ( Ar Pr )

13

f1

Turbulent region: Ar ≥ 10 8 , Nu S = 0.0086Ar 1 2 Pr 1 3 f2

(7.40) (7.41)

150

LNG POOL MODELING

where: Ar = Archimedes number = ( 2 π )

σ 1.5ρv

3

μ 2v g (ρL − ρv )

Pr = Prandtl number of vapor =

Cpv μ v kv

(7.42) (7.43)

The dimensionless functions f1 and f2 given by Klimenko (1981) are in terms of the heat of vaporization, ΔHLV, the heat capacity of liquid, CpL, and the temperature difference between water and LNG, ΔT: ⎧ 1 ⎪ ⎪ Laminar region: ⎨ 13 ⎛ ΔH LV ⎞ ⎪ ⎪0.89 ⎜⎝ C ΔT ⎟⎠ pL ⎩

⎛ ΔH LV ⎞ for ⎜ ≤ 1.4 ⎝ CpL ΔT ⎟⎠ ⎛ ΔH LV ⎞ > 1.4 for ⎜ ⎝ CpL ΔT ⎟⎠ ⎛ ΔH LV ⎞ for ⎜ ≤2 ⎝ CpL ΔT ⎟⎠

⎧ 1 ⎪ ⎪ Turbulent region: ⎨ 12 ⎛ ΔH LV ⎞ ⎪ ⎪0.71 ⎜⎝ C ΔT ⎟⎠ pL ⎩

⎛ ΔH LV ⎞ >2 for ⎜ ⎝ CpL ΔT ⎟⎠

(7.44)

(7.45)

7.2.4.2 Heat Transfer Correlations for Transitional Regime The transitional boiling regime is the least studied of the three boiling modes. Opschoor (1980b) modified some work on Kalinin and coworkers (1975) and obtained reasonable agreement concerning experiments with cryogens boiling on solid surfaces. The heat transfer coefficient in the transitional regime is an interpolation between the critical heat flux, qcr, and the minimum heat flux, qmin: hS =

[ fqcr + (1 − f ) qmin ] ΔT

(7.46)

The filter factor, f, is given by: ΔT − ΔTcr ⎤ ⎡ f = ⎢1 − ⎣ ΔTmin − ΔTcr ⎥⎦

7

(7.47)

7.2.5 Heat Conduction on Shallow Water with Ice Formation In general, spills on a large expanse of water do not form an ice layer since there is a steady convection of heat from the water into the pool as freshwater circulates in replaced chilled water that sinks. Spills of liquids with a boiling point below that of water onto small volumes of water may, however, result in the formation of an ice layer.

POOL SPREAD AND EVAPORATION MODELING

151

The method of Reid and Smith (1978a) treats heat conduction into the pool when a layer of ice forms on the surface of the water. The method involves finding the conduction through an ice layer that increases in thickness. This involves a convergence loop involving the term: p α ice = 1.0907 ⎤ ⎡1 erf ⎢ f (κ ) α ice ⎥⎦ ⎣2 kave

QCice

(7.48)

in which kave and αice are the thermal conductivity and thermal diffusivity of ice evaluated at an average temperature between the water and the freezing point of ice, and f(κ) is a function of the density and the heat of fusion of ice. 7.2.6

Composition Changes with Evaporation

The evaporation rate of an LNG mixture is expected to be different from that of pure methane due to changes in the composition of the liquid over time, leading in the later stages of pool evaporation to: 1. The boiling temperature increases and consequent the temperature driving force for heat transfer decreases 2. The heat of vaporization increases 3. The boiling regime changes from film boiling to transitional mode 4. The vapor composition becomes heavier and affects the heat transfer coefficient, hS Conrado and Vesovic (2000) investigated these effects by treating LNG as a binary mixture of 90 mol % methane, 10% ethane. A key insight is provided by first plotting the vapor-liquid equilibrium phase envelope for the binary LNG in Figure 7.7 (ibid.). Figure 7.7 illustrates that the generated vapor and remaining liquid follow the dew and bubble point curves, respectively. The dew point curve is very steep near the pure methane end, while the bubble point curve is essentially flat. So, in the initial stages of a spill, the vapor consists primarily of methane and the liquid boiling point temperature hardly changes. Conversely, near the right-hand side of Figure 7.7, the bubble point curve becomes steep while the dew point curve is a straight line with a gradual upslope. This implies that late in the life of a pool, the vapor composition will hardly change while the liquid boiling temperature will increase sharply. This matches the composition diagram shown before in Figure 1.4. These expectations are realized as seen by plotting the boiling temperature and mole fraction of ethane in the vapor over time in Figure 7.8 (ibid.). This shows that the boiling temperature of LNG changes little over the first 80 s,

152

LNG POOL MODELING

Temperature, T/(K)

200 180 160 140 120 100 80

0

0.2

0.4

0.6

0.8

1

Xethane

160

1

150

0.8

140

0.6

130

0.4

120

Xmethane

Boiling temperature, T/(K)

Figure 7.7 Methane–ethane VLE phase envelope (Conrado and Vesovic, 2000) (reproduced by permission of Elsevier Science Publishers, Inc.).

0.2

110 100 0

20

40

60

80

0 100 120 140

Time/(s) Figure 7.8 Boiling temperature and vapor composition of binary LNG (Conrado and Vesovic, 2000) (reproduced by permission of Elsevier Science Publishers, Inc.).

and not surprisingly the vaporization rates of LNG and pure methane are nearly the same over that same period. At about 100 s, the LNG liquid becomes ethane-rich and the boiling temperature increases, reducing the ΔT driving force. Even so, this is not the primary reason for a steep drop in the vaporization rate. Rather, it is because the latent heat of vaporization exhibits a rapid increase and more energy goes into heating up the remaining liquid where the slope of the bubble point curve increases rapidly. The overall result on the evaporation rate for the mixture compared with that of pure methane is shown in Figure 7.9 (ibid.). With the assumption of a calm wind, the vapor above the pool remains in thermal equilibrium with the liquid, so vaporization is described along a vertical line in the phase diagram that is at constant liquid composition. Alternately, assuming a strong wind, the vapor is removed as soon as it is formed, and only the bubbles at the liquid surface are in equilibrium with the liquid. The vaporization process in this case would be described as taking place along the bubble-point line. In both cases,

Vaporisation rate, q/(kg/s)

POOL SPREAD AND EVAPORATION MODELING

153

1000 800 600 400 200 0

0

20

40

60

80 100 120 140 160

Time/(s)

Vaporisation rate, q/(kg/s)

Figure 7.9 Evaporation rate of binary LNG with calm wind or strong wind (Conrado and Vesovic, 2000) (reproduced by permission of Elsevier Science Publishing, Inc.).

1000 800 600 400 200 0

0

20

40

60 80 Time/(s)

100

120

140

Figure 7.10 Evaporation rate of binary LNG with and without change of boiling regime (Conrado and Vesovic, 2000) (reproduced by permission of Elsevier Science Publishers, Inc.).

the evaporation rate of LNG is close to that of pure methane for the first 80 s. Then, after 100 s, the evaporation rate drops, reaching almost two-thirds of its maximum by 116 s. This is when the liquid composition becomes enriched in ethane, the heat of vaporization increases, and the temperature driving force decreases. As the temperature driving force decreases, the heat transfer mode changes from film boiling to transition boiling. Figure 7.10 (ibid.) illustrates that the boiling regime is predicted to change from film boiling to the transitional boiling regime when the ethane in the liquid phase exceeds 70% at 108 s. The square points indicate the increase in boiling rate that results from the transitional regime compared with the solid line predicted for the film boiling regime. The difference is marginal. 7.2.7 Type 1 Breach—LNG Penetration into Water, Turbulent Heat Transfer The discharge rate, blowdown, and the jetting liquid discharge are treated in Sections 5.7.1 and 5.7.2. It is shown that horizontal liquid jets project, at least

154

LNG POOL MODELING

initially, past the double hull space for typical hole sizes where the outer-hull penetration is larger than the inner-hull penetration. LNG spilling from a significant elevation, such as a Type 1 breach of an LNG tanker above the water level will penetrate some distance into the water. Dahlsveen et al. (2001) found that a jet of LNG falling at a velocity of ∼15 m/s will go at least 13 m underwater. The two liquids will swirl around, mixing extensively, which greatly increases the heat transfer between phases and the evaporation rate. Hissong (2007) points out that the behavior of LNG released below the water surface is much different from tests in which LNG is released carefully above the water surface. •

•

In one of the Maplin Sands tests, the LNG was released below the water surface. Shell reported that for this test “so much heat was absorbed at the source that the cloud rapidly became buoyant and lifted off from the surface” (Jenkins and Timmers, 1983). In two tests by the Bureau of Mines (Burgess et al., 1972), LNG was suddenly released 3–4.5 m below the water surface. The LNG vaporized completely before reaching the water surface, and no liquid pool was observed.

As suggested by Hissong (2007), there is a need to develop a correlation for the turbulence factor, defined as the turbulent heat transfer coefficient between water and LNG to the value based on quiescent boiling: FT =

hSt hSq

(7.49)

The quiescent boiling coefficients, hSq, are those predicted by Equations 7.36– 7.44. The turbulent heat transfer coefficient is transient, as illustrated for two spill tests on water in which the pool diameter and evaporation rate were reported as functions of time. The tests were conducted by Esso in a joint industry project for the American Gas Association (May et al., 1973a). The tests were on open water in the Gulf of Mexico. For these tests, there is enough information to calculate the turbulent heat transfer rate by the following steps: 1. From the pool diameter and evaporation rate, find the evaporation flux, Gevap. 2. Multiply the evaporation flux by the heat of vaporization to find the total heat flux to the LNG. 3. Subtract the heat flux from air conduction and radiation flux to obtain the heat flux from the water to the LNG. 4. Divide by the temperature difference between water and LNG, ΔT, to obtain the turbulent heat transfer coefficient, hSt.

POOL SPREAD AND EVAPORATION MODELING

155

Turbulence factor at 14.6 m/s velocity

13 Esso-11 during spill Esso-11 after spill

11

Esso-12 during spill Esso-12 after spill

9 7 5 3 0

10

20

30 Time (s)

40

50

Figure 7.11 Turbulence heat transfer factors calculated for two tests (Hissong, 2007) (reproduced by permission of Elsevier Science Publishing, Inc.).

The results of this analysis are shown in Figure 7.11. In Test 11, LNG was spilled for 35 s and the turbulence factor, FT, decreased from 11 to about 4 in 25 s. In Test 12, LNG spilled for 6.2 s and FT decayed from about 12 to 4 in 10 s. An important result from Figure 7.11 is that there appears to be a long-term constant value for FT between 3 and 4 that is not accounted for in tests with deliberately gentle spill conditions. In addition, a correlation for the initial transient period may prove important, as discussed next. The initial, time-dependent turbulence is expected to be proportional to the velocity of LNG upon reaching the surface, uS, and possibly also on the total mass rate of the liquid spilled, FdisL. For a discharge at an angle to the horizontal, θ, at an initial velocity uD at height HD, the velocity at the surface is: uS = (uD sinθ ) + 2 gH D 2

(7.50)

The turbulence factor is expected to vary with uS in a power form, from a basis value at a specific surface velocity u0: ⎛u ⎞ FT = ( FT )0 ⎜ S ⎟ ⎝ u0 ⎠

n

(7.51)

For the Esso Test 11, the velocity us was 14.6 m/s, and this value is used as u0 in a first-pass correlation based on the Esso Tests 11 and 12. This uses a powerlaw relationship in terms of the dimensionless Fourier number, Fo, as plotted in Figure 7.12.

( FT )0 = 10.0 ( Fo × 10 3 )

−0.207

(7.52)

LNG POOL MODELING

Turbulence factor at 14.6 m/s velocity

156

100 Data Correlation

10

1 0.1

1

10 Fourier number × 10

100

1000

3

Figure 7.12 Correlation of turbulence factor with Fourier number (Hissong, 2007) (reproduced by permission of Elsevier Science Publishing, Inc.).

The Fourier number is defined in terms of time, t, the thermal diffusivity of liquid LNG, α, and the average pool thickness, h: Fo =

αt h2

(7.53)

7.2.8 Time-Dependent Pool Spread The time-dependent discharge rates calculated in Section 5.7.1 are used as inputs to a pool spread and evaporation model. Call this Model 1. The model integrates the pool spread and evaporation equations in Sections 7.2.1 and 7.2.2, and gives the predictions in Figure 7.13. For cryogenic liquid spills, such as LNG, the pool area rapidly reaches its steady-state size where the evaporation rate equals the spill rate as long as the spill rate is constant for longenough time. This assumption gives rise to Equation 7.9. Even if the spill rate is changing, for a comparison, steady-state pool radius is calculated for each discharge rate to solve for pool area. For each hole size, the predicted pool radius overshoots the steady-state pool radius and then settles to the steadystate value. The same type of response is predicted for a burning LNG pool with a burn flux rate of 0.45 kg/(m2s) as shown in Figure 7.14. The steady-state pool radius is smaller for a burning pool than for an unignited pool for the same discharge rate. This is because the burning pool absorbs radiation from the flame above, and overall, it evaporates at 2–2.5 times the rate due to water heat transfer alone—once evaporated of course the natural gas immediately burns. This implies that if ignition occurs at some time after the discharge begins, the response would fall along the upper curve before ignition and drop gradually to the lower curve upon ignition.

POOL SPREAD AND EVAPORATION MODELING

157

250

Pool radius (m)

200 150 100 50 0 0

300

600

900

1200

Time (s) 5 m hole 1 m hole

3 m hole

Figure 7.13 Pool radius for spreading LNG on water using constant average discharge rate for leak at water level (Woodward, 2007).

160

Pool radius (m)

140 120 100 80 60 40 20 0 0

300

600

900

1200

Time (s) Gevap = 0.45

Unignited

Figure 7.14 Comparison of ignited and unignited LNG pool for constant average discharge rate from 3-m hole (Woodward, 2007).

With a slight reduction in the pool spread rate, accounting for viscous flow, the pool does not spread fast enough to maintain an equilibrium between evaporation and discharge. In this case, the effect of a 3-m breach at the waterline taking the discharge rate as constant for 320 s is to produce a pool that reaches a peak radius and a peak in evaporation rate as shown in Figures 7.15 and 7.16. The evaporation flux is a constant 0.180 kg/(m2s). Early in the

158

LNG POOL MODELING

Pool radius (m)

Continuous pool radius function

Analysis approximating pool radius function

300 200 100 0 0

50

100

150

200

250

300 350 Time (s)

400

450

500

550

600

Figure 7.15 Pool radius prediction for constant discharge rate at initial rate for 320 s for 3-m hole, 25,000 m3 spill, Algeria LNG.

Evaporation rate (kg/s)

Continuous evap. function

Analysis approximating evap. function

60,000 40,000 20,000 0 0

50

100

150

200

250

300 350 Time (s)

400

450

500

550

600

Figure 7.16 Evaporation rate prediction for constant discharge rate at initial rate for 320 s for 3-m hole, 25,000 m3 spill, Algeria LNG.

response, the mass of liquid discharged to the pool exceeds the mass evaporated, so the pool continues to enlarge. The maximum pool radius reaches just over 300 m to achieve an evaporation rate higher than the discharge rate, and slightly after that point, the discharge stops, so the pool area decreases rapidly. A staircase approximation to the time-varying pool radius and evaporation rate can be used more readily as the input to a dispersion model. As a further improvement, apply the time-dependent discharge rate for a 3-m breach shown in Figure 5.12 in Chapter 5 as input to the viscous pool spread and evaporation model. This provides the predictions illustrated in Figures 7.17 and 7.18. Again, the steady-state Equation 7.9 can be used with each values of the time-varying discharge rate to predict a corresponding value of pool radius to provide a comparison curve. The predicted pool radius overshoots the corresponding “steady-state” pool size before settling down later to the “steady-state” values. Again, the maximum pool radius is predicted as a momentary peak value at 300 m for a 3-m diameter hole. A hole with a 5-m diameter breach has a shorter duration that limits the maximum spread even though the initial discharge rate starts out at a higher value than the 3-m hole discharge rate.

RAPID PHASE TRANSITION EXPLOSIONS

159

Stdy-st pool radius (m)

300 250 200 150 100 50 0 0

120

240

360

480

600

Time (s) 1 m Rstdy

3 m Rstdy

1 m Rpool

3 m Rpool

300

300

250

250

200

200

150

150

100

100

50

50

0

Pool radius (m)

Stdy-st pool radius (m)

Figure 7.17 Pool radius with time-varying source rate from blow down and an unignited LNG pool for a 1-m and 3-m hole, 25,000 m3 spill (Woodward, 2007).

0 0

120

240

360

480

600

720

Time (s) 3 m Rstdy 3 m Rpool

5 m Rstdy 5 m Rpool

Figure 7.18 Pool radius with time-varying source rate from blow down and an unignited LNG pool for a 3-m and a 5-m hole (Woodward, 2007).

7.3

RAPID PHASE TRANSITION EXPLOSIONS

RPT refers to an explosively fast evaporation of LNG to vapor when LNG is suddenly contacted with a warm fluid, usually water. These are physical expansions and do not involve combustion. The main concern with RPTs is from an event in which an LNG carrier is breached and LNG mixes with seawater, either by direct injection of spilling LNG into water, or by mixing

160

LNG POOL MODELING

in the double hull area. If LNG is carried downward in a double hull by inrushing seawater, there may be some effective confinement. Of interest is predicting the energy that could be developed by an RPT in these circumstances, and could the RPT be damaging enough to cause escalating damage to other tanks in the carrier. 7.3.1

Historical Experience with LNG RPTs

LNG experiments on RPT have been mainly at laboratory scale (Anderson and Armstrong, 1972; Katz and Sliepcevich, 1973; MIT LNG Research Center, 1977). Large-scale tests explicitly for the study of RPTs were obtained from the Coyote test series performed by Lawrence Livermore National Laboratories (Ermak et al., 1982b; Goldwire et al., 1983; McRae, et al., 1984; Morgan et al., 1984; Rodean, et al., 1984;). For the RPT study, 13 spills of 3–15 m3 were performed with flow rates of 6–19 m3/min, with LNG methane concentrations between 75 mol % and 92 mol %. Six tests produced RPTs. Delayed RPTs developed in three tests; the others were immediate with the spill. They were located near the spill point and appeared to be primarily underwater. Nedelka reported at the Offshore Technology Conference (OTC) in 2003 the results of 29 spills onto water with up to 9 m3 of LNG carried out jointly by Gaz de France, Shell, and British Gas. They identified that the maximum explosion was 4.15 kg trinitrotoluene (TNT) equivalence, that heavy hydrocarbons were required for RPT, and that the risk was limited to LNG/water mixing zones. The occurrence of RPTs appears to correlate with water temperature and the depth of penetration of LNG into the water, as well as composition which is discussed later. When LNG penetration was limited by a spill plate, RPTs occurred with water temperatures above 290°K (63°F). When the plate was removed, RPTs occurred with water temperatures of 285°K (52.9°F). The strength of RPTs in the Coyote tests did not correlate with impact pressure into the water. This contrasts with findings for laboratory scale tests by Jazayeri (1975) who developed a correlation between RPT strength and impact pressure for various cryogens. However, the strength of RPTs was found in the Coyote tests to correlate with spill rate as shown in Figure 7.19. An abrupt increase in the RPT strength was found at around 15 m3/min, and at 18 m3/min, the strength increased by five orders of magnitude. The maximum equivalent free air-blast strength was equated to a point-source blast of 6.3 kg TNT for the 18 m3/min spill rate. The majority of the blast energy was directed underwater, and was not measured. These rates were at the capacity of the tests. The concern might be whether even higher overpressures develop at higher discharge rates. On one Coyote series experiment, an RPT produced damage to the instrument tower, the spill plate, and some piping. Test equipment was damaged also in the Shell Oil Co. Maplin Sands tests (Puttock, 1982a; Puttock et al.,

RAPID PHASE TRANSITION EXPLOSIONS

161

RPT source yield (kg of TNT)

7 6 5 4 3 2 1 0

5

10

15

20

Spill rate (m3/min) Figure 7.19

RPT strength from Coyote tests (Ermak et al., 1982a).

1984). An LNG tank designed to be submerged by flooding ballast tanks was sunk by an RPT and was not returned to service. An LNG fire developed on Falcon Test 5 as described in Chapter 2. Enger and Hartman (1972) of Shell Oil reported from small-scale (0.1 m3) tests that the methane content of LNG had to be below 40 mol % for RPTs to occur. However, the Coyote tests found RPTs occurring with LNG compositions up to 88% methane, indicating that other mechanisms become dominant for large-scale spills. 7.3.2

Similar Phenomena More Thoroughly Investigated

Similar and very damaging evaporating liquid explosions occur when molten metal or slag is inadvertently poured into a vessel containing water, an event called a steam explosion. There is a significant history of major steam explosions in steel mills and foundries, such as the Appleby-Frodingham Steelworks in 1975 (Health and Safety Executive, 1976) where up to 90 tons of molten metal was thrown over a wide area after the contact of 320, 000

(7.72)

where the Prandtl number, Pr, and Reynolds number, Re, are evaluated at the film conditions in terms of the heat capacity of air, Cpa, the viscosity, μa, the thermal conductivity, ka, of air and the pool diameter, Dpi: Pr =

Cpa μ a ka

Re =

Dpi ua ρa μa

(7.73)

The film conditions are at the average temperature of the LNG and the air. 7.5.3

Radiation to/from Pool

The rate of heat input in W from solar radiation for a pool of radius r is calculated as Qsolar = πRp2001S

(7.74)

HEAT BALANCE TERMS TO LNG POOL

171

where S is the solar flux, a little less than 1000 W/m2 in cloudless daytime sky. The rate of heat input/loss in W from long-wave radiation is calculated in terms of the Stefan–Boltzman constant, σ(5.669 × 10−8 W/[m2K4]), the emissivity (about 0.95), the ambient temperature, Tamb, and the pool temperature, Tpool as: 4 4 Qlong = πRi2 σε (Tamb − Tpool )

(7.75)

In general, solar radiation to the pool is negligible compared with other mechanisms, but might become important with insulated impoundment basins. 7.5.4

Evaporative Cooling on Water

Dodge et al. (1983) developed a model for heat conduction to a pool on water. This model takes into account the roughness of the water and the height of waves on the water, both of which depend upon the wind speed. The form of the correlation is: 2 Qevap = πRpool ua*Da*H vL

Mc Pamb ⎛ Pamb ⎞ ln ⎜ ⎟ RTpool ⎝ Pamb − Ps ⎠

(7.76)

The boundary layer Dalton number (the ratio of turbulent mass transfer rate to the driving force for mass transfer by film theory) is Da*. There is no evaporative cooling with LNG pools on sea or on land as the pool is a boiling liquid and all the heat transfer is converted into latent heat. Evaporative cooling would imply subcooling and this does not occur. 7.5.5 Bubble Flow in Vaporizing LNG During evaporation of LNG on the sea surface, the flow of bubbles reduces the average LNG pool density, ρp, below the liquid LNG density, ρL, in proportion to the volume fraction η of bubbles in the pool. ρp = (1 − η) ρL

(7.77)

As summarized by Fay (2007), bubble theory obtains the upward velocity of vapor bubbles, ub, of diameter d by equating the buoyant force on a bubble to its aerodynamic drag, giving: ρL ub2 d 2 ∼ ρp gd 3 ub = gd (1 − η) ∼ gd

(7.78)

The bubble diameter is determined by the balance between the surface tension force and the buoyant force on the bubble, giving:

172

LNG POOL MODELING

d∼

σ ⎛ σg ⎞ ; ub ∼ ⎜ ⎟ ⎝ ρL ⎠ gρL

14

(7.79)

The bubble fraction η then is found from the ratio of the superficial velocity to the evaporative flux, Gevap, divided by the vapor density, ρv, to the bubble velocity, ub: η=

Gevap ρv Gevapρ1L 4 = 14 ub ρ v ( σg )

(7.80)

Equation 7.80 gives a value of η on the order of unity for an LNG pool, but the maximum value of the void fraction for bubbly flow would be about ½, near the value of closely packed spherical bubbles in a continuous liquid phase. For LNG, Fay (ibid.) estimates that the value of Δ from Equation 7.11 would be about 0.8.

7.6

NOMENCLATURE

Ah AT Apool b CD Da Dp d E Fdis FdisL Fevap FD Fg FI Fo FS Gdis G1,G2 Gevap g h hL

= Effective cross-sectional area of puncture, m2 = Cross-sectional area of tank, m2 = Pool area, m2 = Burning regression rate, m/s = Discharge coefficient, dimensionless = Dalton number, = Pool diameter, m = Bubble diameter, m = Energy in shockwave, J = Discharge rate, kg/s = Liquid discharge rate to pool, kg/s = Evaporation rate, kg/s = Viscous drag force, N = Gravity spreading force, N = Inertial spreading force, N αt = Fourier number, 2 h = Surface tension force, N = Discharge flux, kg/(m2 s) = Discharge flux at planes 1, 2 etc., kg/(m2 s) = Evaporation flux, kg/(m2 s) = Gravitational constant, m/s2 = Average height of liquid layer (partly submerged), m or height of leading edge of pool, m = Level of LNG above the water level or above the breach, m

NOMENCLATURE

= Heat transfer coefficient, W/(m2°K) = Thermal conductivity, W/(m°K) = Mole weight of component c =contaminant, kg/kgmole = Mole weight of component i, kg/kgmole = Molecular weight of vapor, kg/kgmole = Molecular weight of humid air, kg/kgmole = Mass of air in tank vapor, kg = Mass of liquid in tank, kg = Mass of vapor in tank, kg = Nusselt number, h/(k.Dpool) = Pressure at plane 1, 2, and so on, Pa(a) = Ambient or back pressure, Pa(a) = Pressure at midpoint of rupture, including hydraulic head, Pa(a) Cpaμ a Pr = Prandtl number, ka = Saturation pressure, Pa(a) Psat = Head pressure in tank, Pa(a) PT = Vapor pressure, Pa Pvap = Gas constant, J/(kgmole.°K) R Dpuaρa Re = Reynolds number, μa R, Rpool = Radius of pool at time t, m = Equilibrium pool radius for steady-state discharge, m Req = Maximum pool radius, m Rm Qconv = Heat transfer to pool by convection from air, W Qevap = Heat of evaporation, W Qlong = Long wave radiation heat loss from pool, W Qsolar = Heat from solar radiation to pool, W S = Solar flux, ∼1000 W/m2 = Critical temperature of mixture or component, °K TC TNBP = Normal boiling point temperature, °K = Temperature of tank contents, °K TT = Time, s t = Time to completely evaporate a pool of spilled material, s tevap = Velocity of discharging fluid, m/s udis = Final or expansion velocity of jet, m/s uexp = Sonic velocity, m/s ua = Bubble velocity, m/s ub = Velocity of liquid spill upon the water surface, m/s uL uv = Evaporation velocity, Gevap/ρv, m/s = Volume of vapor space in tank, m3 VV = Specific volume of vapor, m3/kg v,vV = Expansion work against atmosphere, J W = Mass discharge rate, kg/s w hS kS MC Mi MV Mwair mair mL mV Nu P1,P2 Pamb Phole

173

174

LNG POOL MODELING

wair wTNT x yair yI

= Mass fraction of air in vapor = Equivalent mass of TNT, kg = Mass fraction of vapor in two-phase mixture = Mole fraction of air in vapor = Mole fraction of inerts in vapor (inerts could be initial air in tank) = Non-frothy liquid level in tank, m = Interior height of tank, m

zL zT Greek α ε Δ ΔH ΔHLV ΔPVset Δρ Δz η γ κ ρ σ σG μa μw

= Thermal diffusivity of liquid, kL/(ρLCpL) = Emissivity, dimensionless = (ρw − ρL)/ρw = Change in specific enthalpy, J/kg = Heat of vaporization, J/kg = Set point for vacuum breaker valve, Pa(g) = Normalized buoyancy = Change in elevation or depth, m = Efficiency of explosion or volume fraction = Adiabatic compression exponent, Cp/Cv = Parameter defined by Reid and Smith (1978a) = Density, kg/m3 = Net surface tension on spreading liquid, N/m, or Stefan Boltzmann constant, 5.669 × 10−8 W/(m2K4) = Standard deviation for drop size distribution, m = Viscosity of air, Pa.s = Viscosity of water, Pa.s

Subscripts cr = Critical dis = Discharge eq = Equilibrium between evaporation and discharge evap = Evaporation i = Current time increment L = Liquid m = Maximum mn = Mean drop size s = Surface sat = Saturation v = Vapor vf = Vapor film w = Water

8 VAPOR CLOUD DISPERSION MODELING Liquefied natural gas (LNG) spills rapidly evaporate on land or water as discussed in Chapter 7, and form initially a cold, heavy vapor plume. The plume is blown downwind, eventually warms, and may become positively buoyant at concentrations of interest (above ½ lower flammable limit [LFL]), and may lift off to become a dilute elevated vapor cloud. The plume or vapor cloud can ignite and burn as a flash fire, treated in Chapter 10, and as a pool fire, treated in Chapter 9. Vapor cloud explosions are possible in congested liquefaction or regasification plants but less likely in other situations. This chapter reviews several topics related to dispersion. Section 8.1 reviews some basic atmospheric transport mechanisms particularly related to wind and stability. Section 8.2 introduces common modeling approaches. Section 8.3 reviews relevant dispersion tests used for validation of models. Section 8.4 discusses several detailed issues related to modeling of dispersion. Section 8.5 addresses uncertainties and means to accommodate these in dispersion estimates. The nature of the source term (discharge and evaporation) and whether these are near steady state or near instantaneous, also have significant impacts on the dispersion.

8.1

ATMOSPHERIC TRANSPORT PROCESSES

Dispersion is greatly affected by local atmospheric conditions, primarily wind speed, atmospheric stability, and ground roughness. Depending on the local LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

175

176

VAPOR CLOUD DISPERSION MODELING

terrain, obstructions and ground slope/elevation can also be important. These factors all affect atmospheric turbulence, and hence, mixing and dilution. Turbulence interacts with dense plumes in an often complex manner; simple models ignoring these interactions do not predict dispersion distances well. Good descriptions of atmospheric processes relevant to releases of process fluids, which are often initially denser than air, are provided in Lees (Mannan, 2005), Center for Chemical Process Safety ([CCPS], 2000c), Woodward (1998a), Johnson and Woodward (1999), and Britter and Griffeths (1982). 8.1.1

Wind Speed, Stability, and Surface Roughness

8.1.1.1 Wind Speed Wind speed is defined as the speed of the wind at 10-m elevation. Because of ground drag, the velocity profile drops from its nominal value u10 down to u0 = 0 m/s at 0-m elevation. Wind speed felt by individuals at 1.5- to 2-m elevation is often only 50–70% of the u10 value. The atmosphere is rarely completely still, although meteorological data often refer to periods of calm. In order to carry out dispersion studies, it is necessary to obtain meteorological data as near as possible to the location of interest. This can be obtained from a dedicated met tower or from the nearest airport. Wind and stability data are readily available at airport locations globally. However, hills or significant topography can have a major effect on wind patterns and this must be considered when using data from a location other than the actual site. Wind data are often obtained as a wind rose as shown in Figure 8.1. By convention, wind direction is always represented as wind from and not wind toward, thus a north wind is wind that comes from the north. Since no dispersion model is valid for zero wind speed, it is common to take any reported calm fraction and allocate this over all directions so the total of all winds used for analysis adds to 1. 8.1.1.2 Atmospheric Stability Stability is a measure of atmospheric turbulence, which may be thought of as large turbulent eddies. Pasquill and Gifford define several classes of atmospheric stability, most commonly from A to F (Pasquill, 1974; Pasquill and Smith, 1983). The normal, neutral atmosphere is category D. This corresponds to an atmosphere with a standard adiabatic lapse rate α which is −0.98°C per 100 m of elevation. So: Unstable:

∂Tair ∂T ∂T < α; Neutral: air = α; Stable: air > α ∂z ∂z ∂z

(8.1)

Unstable atmospheres are categories A–C. These have greater lapse rates and strong large eddies caused by solar insolation, warming the ground and causing large thermal updrafts. Conversely, stable atmospheres E–F have lower or inverse lapse rates (i.e., temperature warms with height) and are often associ-

177

ATMOSPHERIC TRANSPORT PROCESSES

N

Calm 11.0%

W

E

5 10 15

S 1–3 4–7 8–12 >13 Figure 8.1

Table 8.1 stability

Example wind rose.

Effect of wind speed and insolation (solar strength) on atmospheric

Surface Wind Speed (m/s)

6

Speed m/s

Day

Night

Strong

Moderate

Slight

Thinly Overcast or >=4/8 Low Cloud

b

Sz (x)

y

z

1+a , IyI ≤ b

Sz (x)

Cu

x (L ,O,O) 2

1+a

z

C (x,y,z) = CC (x) exp –

(O,O,O)

L

IyI – b(x)

–

S1 (L/2)

Cu

CC (L/2) (X0,O,O)

b Sy (x0)

uX = uO

Z

a

ZO

Sy (x0) Sz (x0)

b CC (x0)

Sy (x1) Sz (x1)

(X1,O,O) CC (x1)

ISO concentration contours for C = Cu

c Sy (x1) L

Figure 8.5 Illustration of HEGADAS and DEGADAS model concentration profile transitions as plume moves downwind (Spicer and Havens, 1986).

m = 50, and gradually decreases to m = 2 (Gaussian) and even below (m = 1.5). As m changes, the concentration profiles change from a very nearly top hat form to have more rounded edges as illustrated in Figure 8.6. This has the advantage of applying to elevated releases as well. Other integral dispersion models, mainly Similarity type, are discussed in the CCPS series, notably 22 programs are treated in Hanna et al. (1996) (see also CCPS 1995, 1998, 1999, 2000c, 2001b, 2002). These include the proprietary models CANARY (by Quest, Inc), FRED (by Shell Global Solutions), and TRACE or SAFER (by SAFER Systems, Inc.) and LSMS (LNG Spread by Cambridge Environmental Research Consultants, now in Advantica suite of models. The Breeze model suite by Trinity Systems adds a low-cost proprietary user interface to the DEGADIS model. Free publicly-available models include ALOHA (by the National Oceanic and Atmospheric Agency, NOAA), SLAB (Ermak, 1980 by Lawrence Livermore National Laboratory [LLNL]), as well as HGSYSTEM (Shell Oil) and DEGADIS. Natural gas production and distribution companies have developed models such as CIRRUS by BP, EXPLOJET by INERIS, RHODIA by Rhoditech and models by Archema and Snamprogetti, but history suggests that such companies do not find it economical to support model development in the long run.

MODEL TYPES

185

1.0

yc/yCL

0.8 0.6 0.4 0.2 0 –3

–2

–1

0

1

2

3

y/ay

m = 1.5

m=2

m=8

m = 50

m=4

Figure 8.6 Illustration of DRIFT, PHAST, and SafeSite model concentration profile transitions as plume moves downwind.

A feature of some of the Integral models (e.g., PHAST, SafeSite) is a direct model linkage between the source term and the dispersion and full integration of chemical component physical properties. These models are also commercial codes designed for ease of use by nonspecialists. Some well-known codes for LNG (e.g., DEGADIS) are purely dispersion codes and users must initialize the model with their own source term. This has led to serious errors when the models do not match (Havens and Spicer, 2006b). 8.2.3

CFD

This is the most fundamental approach for modeling dispersion as it solves the 3-dimensional Navier–Stokes flow equations without predetermined geometric limitations. In principle, this allows real geometries to be simulated with terrain effects, obstructions, and transitions (e.g., from flat sea to sloped and obstructed LNG facilities). Most CFD codes for LNG use the Reynolds Number Averaged Navier–Stokes (RANS) equations combined with a k–ε turbulence model. The CFD code requires specification of the domain, the computational grid, the turbulence model, and the boundary conditions. The grid must be sufficiently fine to ensure numerical stability, and for large problems, this can result in a large mesh. Obstacles can be inserted into the mesh and flow must pass around these. Luketa-Hanlin et al. (2007) survey many issues related to running CFD codes satisfactorily for LNG and they believe all issues can be addressed. Cormier et al. (2006) notes four public CFD codes have published recent validations for LNG. These include FEM3A (LLNL, Chan et al., 1981a, 1981b;

186

VAPOR CLOUD DISPERSION MODELING

Chan 1990), FLACS (Gexcon, Bakke and Hjertager, 1986), FLUENT (ANSYS) and CFX—where the first two are specific consequence codes and the latter two are general purpose CFD codes. Some CFD codes used for LNG modeling are not publicly available; these include FEM3, FEM3A (Chan, 1990), and the Sandia (Luketa-Hanlin et al., 2007; Luketa-Hanlin et al., 2008) CFD code. Moullieau and Champassith (2009) describe using the free public fire dynamics simulator (FDS) code from the National Institute of Standards Technology (NIST) for LNG dispersion. They note that better results for CFD codes are obtained if atmospheric variations are explicitly defined as described by Hanna et al. (2004). This is not simple for nonspecialists. The situation is not as clear that CFD is always superior as some suggest. While the advantages of the CFD approach are significant, they come with a cost. CFD analysis is still an experts’ domain, with many numerical issues to address and currently fewer of the built-in features that integral models have developed over 20 years (e.g., jet discharge modeling, 2-phase flashing, droplet modeling, pool evaporation, external validations). While these features can be added, they do add to complexity and cost. CFD codes require extensive computer resources. For example, Sandia uses a super-computer parallel array of CPU processors for LNG dispersion, essentially only found in National Laboratories. Furthermore, CFD codes, as applied, do not have extensive commercial deployment. This means that there are often only a handful of users globally, whereas some integral codes have thousands of users. Widespread users results in extensive feedback to modelers, bug-logs and rectification programs, and an active user community. Even so, there is a very active literature for CFD codes currently and many different ideas are being explored. A survey was carried out in Europe by Olav Hansen and reported by Hanna Consultants (2005) on models used and developed by major research organizations for various safety-related applications. Several of the models mentioned in the survey and listed in Table 8.3 have been or could well be applied to LNG. The organizations cited for developing CFD models are: • • •

• •

•

•

•

AEA Technology Engineering Software, Harwell, UK, London, UK. ANSYS, London, UK; Canonsburg, PA, USA; Software developer. Baker Eng. and Risk Consultants, San Antonio, TX, USA, Risk consultancy. Brookhaven National Laboratory, Upton, Long Island, NY, USA. Cham, Ltd, Wembleton, UK; Software developer (acquired by ANSYS). CEA (Commissaraiat a’l’ Energie Atomique), Bruyeres-le-Chatel, France, the French Atomic Energy Authority. GexCon, Bergen, Norway; Skelmersdale Lancashire, UK; acquired FLACS from Christian Michelsen Institute HSE/HSL, Health and Safety Executive/ Health and Safety Laboratory, Buxton, Derbyshire, UK.

MODEL TYPES

187

Table 8.3 Summary of dispersion models and submodules

Research Organization

Code

AEA Technology ANSYS BakerRisk

CFX4 FLUENT BWTI

Brookhaven CEA

HGSPRAY CAST3M

GexCon

FLACS

HSE/HLS JRC

See reports REACFLOW

Europlexus NCSRD

ANDREA-HF

NIST NUST

FDS KAMELEON FIREX (KFX)

Risknology TNO and Century Dynamics LLNL Sandia Von Karman Institute Warsaw University of Technology

CEBAM AutoReaGas

• •

•

•

• •

•

FEM3A MELCOR CASIMIRE2 ZND DETON, DL KIVA-3V

Applicable Area S3: explosions S2: dispersion, general purpose S3: combustion; S4: building response S5: mitigation (water curtains) S1: distribution/mixing; S3: combustion; S5: mitigation S2: ventilation, dispersion; S3: explosion; S5: mitigation methods, general purpose Review and validate CFD models S3: H2 explosion, pressure within confined/semi-confined geometries S2: fluid/structure interaction; S3: explosions, shock; S2:dispersion in complex terrain; S5: mitigation S3: fire S1: high-pressure discharge; S2: jet impingement, dispersion; S3: fire and explosions S3: explosion; S4: building response S2: dispersion; S3: explosion and pressure; S5: mitigation S2: dispersion, nuclear applications; S3: fire S5: mitigation (water curtains) S1: detailed chemistry; S3: combustion S1: release, distribution/mixing

Von Karman Institute, Chaussee’ de Waterloo, Belgium. JRC, the Joint Research Centre sponsored by the European Commission with centers in Ispra, Italy and Petten, the Netherlands. NCSRD, National Center for Scientific Research DEMOKRITAS, Athens, Greece. National Institute of Standards Technology (NIST), Boulder, Colorado, USA. LLNL, Lawrence Livermore National Lab. NUST, Norwegian University of Science and Technology, Trondheim, Norway. Risknology, Houston, Texas, USA, risk analysis and software development.

188 •

8.3

VAPOR CLOUD DISPERSION MODELING

TNO, The Hague, the Netherlands; leading research organization sponsored by Dutch government. LNG DISPERSION TEST SERIES

The need for dispersion tests is exemplified in Figure 8.7 (McQuaid, 1983). Plotted are predictions of various modelers made prior to the Thorney Island tests that released 2000 m3 of gas with a density twice that of air (Health and Safety Executive [HSE], 1984). The predictions were assembled by the British HSE and represent the state of the art at that time. The predictions varied by two orders of magnitude, even for this relatively small, isothermal release onto flat terrain. Current model predictions differ far less, primarily from the advantage of large-scale tests (e.g., Hartwig, 1984). Figure 8.8 is an example of a large-scale LNG test, a plan-view photograph of the Burro 8 test described in Table 8.3 and in Section 8.7. This shows a dense, low cloud with some discontinuities, taken 80 s after the discharge began onto a pond (Ermak et al., 1981). No dispersion tests have been carried out at the scale possible from the largest spills of LNG plants and carriers. However, the physics involved are not expected to vary as much as for largescale fires, and thus, no major new LNG dispersion trials are under planning, whereas large-scale fire trials are planned by the US Department of Energy (DoE) by Sandia National Laboratory in 2009. Compilations of test data for dispersion model validation have been made by Ermak et al. (1988), Carissimo et al. (2001) for the SMEDIS (Scientific MEthods of DISpersion modeling) project, and by Nielsen and Ott (1996) as the REDIPHEM database. The Modeler’s Data Archive (MDA) of Hanna

Cmax %

10.0

1.0

0.1 0

100

200

300

400

500

600

700

800

X (m) Figure 8.7 Pretest predictions by various modelers before Thorney Island tests (McQuaid, 1983).

LNG DISPERSION TEST SERIES

Figure 8.8 1981).

189

Aerial photograph of Burro 8 plume at 80 s after discharge began (Ermak et al.,

et al. (1991) is in convenient spreadsheet form and has been widely used for model validation. The U.S. National Association of State Fire Marshals (NASFM) commissioned its contractor, UK Health and Safety Laboratory, to develop a complete compilation of LNG dispersion data to be used for model validation. NASFM plans to release a public list late in 2009. Table 8.4 summarizes large-scale experimental test series for LNG that includes measurements of LNG plume concentrations. These experiments are necessarily large-scale. LNG experimental programs often combine evaluation of evaporation rate, dispersion, and fire characteristics, so there is considerable overlap in categorizing these programs (see Section 7.2.3, Table 7.1 for evaporation rate data and Section 9.4, Table 9.1 for fire radiation data). In fact, one of the largest sources of uncertainty in using dispersion test data to validate models is in the “source” terms—discharge rate, pool size, pool evaporation rate. Webber et al. (2009) address this issue with a compilation of source terms. The variable of most interest in LNG dispersion tests is the maximum downwind extent to the LFL, shown in Table 8.4 with measured values out to a maximum of 442 m. The maximum cloud width to the LFL is also of interest, as is whether the plume is elevated at these maximum extents. The most important variables influencing the maximum downwind extent is the evaporation rate, set primarily by the spill rate and pool size limitations, and secondarily by the wind speed and atmospheric stability. The total amount spilled influences primarily the duration of the plume. When the duration is long enough, the plume reaches steady state. Details of specific tests are summarized in Table 8.5. These and other features of LNG dispersion are discussed in the following sections. The setup for the cited test series is described in Section 8.7.

Matagorda Bay, TX Esso SS Gadilla, Shell

Avocet, LLNL Burro, LLNL

Maplin Sands, Shell

Coyote, LLNL

Falcon, LLNL

1972

1978 1980

1980

1981

1987

a

Visual estimate.

1974

Tests, Sponsor

Year

Partially confined in dike

On pond in desert

On Thames Estuary

On pond in desert

Elevated spill on seawater Ship jettison to ocean

Type of Tests

8.7–30.3

5

20.6–66.4

8–28

5–20

1.5–4

14–19

4.2–4.52 24–39

27–193

0.73–10.2

Spill Vol (m3)

4 11.3–18.4

2.7–19.3

18.9

Spill Rate (m3/min)

10 (5 burn) (5 RPT

20 LNG+ 14 LPG

4 8

6

17

No. of Tests

Table 8.4 Summary of large-scale dispersion test data

n/a

380

310

190 ± 20

∼20

n/a

220 420

2250a

442

Max. x to LFL (m)

13.6–14.4 ∼10

n/a

14–39

Pool Diam (m) Felbauer et al. (1972); May et al. (1973b) Humbert-Basset and Montet (1972); Kneebone and Prew (1974) Koopman et al. (1979, 1980) Koopman et al. (1982a,b); REDIPHEM (Nielsen and Ott, 1996); MDA (Hanna et al., 1991); Ermak et al. (1981, 1988); Rodean and Cederwall (1982) Blackmore et al. (1982a,b) Puttock et al. (1982a,b) MDA (Hanna et al., 1991). Also data reports (Colenbrander et al., 1983, 1984a, 1984b); Ermak et al. (1988) Ermak et al. (1982a, 1982b, 1985, 1988); Coyote data report (Goldwire et al., 1983); McRae et al. (1982, 1984); Morgan et al. (1987); Rodean et al. (1983, 1984); REDIPHEM (Nielsen and Ott, 1996); MDA (Hanna et al., 1991) Brown et al. (1990)

Reference 190 VAPOR CLOUD DISPERSION MODELING

30 65 98 82 114

16 13.5 17.1 16.6 14

Coyote 2 Coyote 3 Coyote 5 Coyote 6 Coyote 7

70 79 75 82 99.5

600

32 23

19.3

18.9 18.9

67 59 77 53

Spill Duration (s)

600 600

85 94

3.93 4.60 3.50 4.75

Spill Rate (m3/min)

2.7

1 2 3 4

% Methane in LNG

Shell Gadila min Shell Gadila Max volume spilled Shell Gadila max

Esso 11 Esso 17

Avocet Avocet Avocet Avocet

Test

5.9 5.8 9.7 4.6 6.0

a

5.1

1.9 3.9

8.0 4.0

2.38–5.76 1.98–4.95 5.77–10.6 1.91–4.12

Atmospheric Stability, uw, m/s

Table 8.5 Summary of LNG dispersion test conditions and results

195c 205 250

2250 visible

442

380 320

Max. x to LFL (m)

126 99.7 132.7 114 144

“

“ “

None “

None “ “ “

Ignition Time (s)

Ermak et al. (1982b); Morgan (1987)

Kneebone and Prew (1974)

Felbauer et al. (1972) May (1979)

Koopman et al. (1979, 1980) “ “ “

Reference

LNG DISPERSION TEST SERIES

191

95.2

93.3

94.7 95.6 91 91 88

Maplin S 39

Maplin S 56

Falcon 1 Falcon 2 Falcon 3 Falcon 4 Falcon 5

b

Series at 10 m elevation. Series at 3 m elevation. c Log–log least squares interpolation. d Series at 2 m elevation.

a

93.2 98.5 95.9 97.8

S 27 S 29 S 34 S 35

Maplin Maplin Maplin Maplin

28.7 15.9 18.9 8.7 30.3

2.5

4.7

3.2 4.1 3.0 3.9

2.9

84.9

Spill Rate (m3/min)

Maplin S 15

% Methane in LNG

11.9 12.2 12.1 11.3 12.8 13.6 16.0 18.4

continued

Burro 2 Burro 3 Burro 4 Burro 5 Burro 6 Burro 7 Burro 8 Burro 9

Test

Table 8.5

100 78 161 310 78

80

60

160 225 95 135

285

173 167 175 190 129 174 107 79

Spill Duration (s)

[email protected]% [email protected]% [email protected]% [email protected]% [email protected]% [email protected]% [email protected]% [email protected]% [email protected]% [email protected]%

3.6a

G1.7 D4.7 D4.1 D/E5.2 E1.8 (p B179)

D5.1

D4.1

d

230c 180 190 220 215 240 445 270

D5.59b D5.58 D9.35 D7.79 D9.35 D8.75 E2.45 D5.94

5.5 7.4 D8.6 D9.8

Max. x to LFL (m)

Atmospheric Stability, uw, m/s

81

“

“

“ “ “ “

None

none “ “ “ “ “ “ “

Ignition Time (s)

Brown et al. (1990)

Puttock et al. (1982a, 1982b, 1984)

Morgan et al. (1984); Morgan (1987)

Reference

192 VAPOR CLOUD DISPERSION MODELING

FACTORS AFFECTING PLUME LENGTH

193

Wind tunnel experiments can be useful for the evaluation of LNG dispersion models that account for the effect of obstacles. These wind tunnel tests include work at the University of Arkansas (Havens and Spicer, 2005, 2006a, 2006b, 2007a, 2007b), and two series of tests carried out as part of European Commission-funded projects, referred to as BA-Hamburg and BA-TNO (Schatzmann et al., 1991; Nielsen and Ott, 1996). 8.4

FACTORS AFFECTING PLUME LENGTH

Some factors do not vary significantly between tests in a given test series, including • • •

•

surface roughness and other substrate properties; terrain features (slope, obstacles); source temperature (saturation temperature is limited by LNG composition); and source density (set by LNG composition and saturation temperature).

To obtain the effect of these parameters, different test locations need to be used. Some factors vary significantly with each test in a given series, some under control and some not, including • •

•

wind speed (it is difficult to obtain low wind speed tests); atmospheric stability (tests at night and in the early morning hours are needed to obtain high stability); and source rate (this is sometimes difficult to maintain constant).

Some factors are uncontrollably too constant or vary seasonally, including •

•

ambient temperature (some diurnal variation occurs, but is often too limited to obtain its effect on dispersion); and relative humidity (tests in desert conditions are at low humidity and do not bring out effects of high humidity found on the ocean).

8.4.1

Heavy Gas Properties Increase Hazard Area

In atmospheric dispersion modeling, there are three distinct phases of dispersing plumes: 1. Jet release (under vapor pressure or pump pressure)—dominated by jet turbulence dilution. 2. Dense gas phase—dilution dominated by gravity forces. 3. Passive gas phase—dilution dominated by atmospheric mixing processes.

194

VAPOR CLOUD DISPERSION MODELING

For LNG spills onto land or water, jet release may not be important. However, releases in liquefaction plants will have jet releases. Heavy gas plumes originating at high concentration and density exhibit two distinctive properties: • •

Suppressed turbulence. Gravity spreading (driven by their high specific gravity).

As the plume disperses downwind, the heavy gas characteristics diminish, and the plume makes a smooth transition to Gaussian or passive dispersion. Gaussian dispersion represents a mode of air entrainment where the plume is dispersed primarily by the normal turbulence in the atmosphere. This mode is so named because the crosswind profile of concentration is usually well represented by the familiar “normal” or Gaussian distribution (see Figure 7.21). Some heavy gas dispersion properties are responsible for LNG plumes being wider and longer than predicted by using a model that incorporates only passive dispersion. Even so, early models applied to dispersion predictions for LNG plumes did not recognize the need for the dual mode of air entrainment, and applied only the Gaussian mode. There were even errors in the LFL value used. Adequate peer review and thorough validation of models are necessary to avoid promulgating erroneous predictions in this field. 8.4.1.1 Heavy Gas Plume Characteristics Suppressed turbulence was seen most clearly in the Burro 8 test. This is considered the best available test to illustrate heavy gas dispersion characteristics, likely because it was a large release (about 30 m3) with a low wind speed (1.8 m/s at 2 m high). In this test, the vapor cloud caused displacement of the atmospheric flow so the turbulence and velocity within the cloud decreased to almost zero. The stable stratification of the dense cloud dampened turbulent mixing, so the wind flowed over the cloud as if it were a solid object. In some other tests, the suppression of turbulence is less pronounced, but is still seen to a measurable extent. A theoretical basis for the suppression phenomena is briefly described in Section 8.9. Gravity spreading essentially means that a heavy, dense plume spreads of its own weight, in addition to being driven by the wind. It may even spread upwind to a small degree if wind speeds are low. In the gravity-spreading zone, the mass of evaporated vapor from a continuous release of LNG tends to be higher at the front and edges of the plume than in the middle. This is illustrated in Figures 8.8 and 8.9 for the Burro 8 test. In Figure 8.9, the measured concentrations at 1% form “shoulders” and dip in the middle. Higher concentrations appear only in the edges of the plume. The CFD model FEM3 matches the observed concentrations quite well, and fills in a continuous contour for all four concentrations plotted.

FACTORS AFFECTING PLUME LENGTH

195

10

Height (m)

(a) EXPT.

1

5

5 15

0

10 (B) FEM 3 1 5 15 10

–120

–80

–40 0 40 Crosswind distance (m)

80

120

Figure 8.9 Cross-section contours of concentration for Burro 8 test: (a) measured; (b) modeled by FEM3 (Ermak et al., 1982b). Downwind distance (m).

Crosswind distance (m)

200

100

5 15

10

1%

0 25

15

10

5

–100

(a) Experiment –200 –100

0

100

200

300

400

Downwind distance (m) Figure 8.10 1985).

Plan-view contours for Burro 8 test showing bifurcating plume (Ermak and Chan,

As a consequence of the high-mass “shoulder ridges” in the Burro 8 test, the plume also bifurcates as shown in Figure 8.10. The shoulders become essentially two plumes after a certain distance downwind. This is unusual, and most LNG plumes remain as a single plume. 8.4.1.2 Comparison of Heavy Gas and Passive Dispersion Properties A side-by-side comparison of the properties of the two phases of dispersion for nonpressurized releases is summarized in Table 8.6. Morgan et al. (1984) and Luketa-Hanlin (2006) contributed to these comparison points.

196

VAPOR CLOUD DISPERSION MODELING

Table 8.6

Comparison of heavy gas and Gaussian model properties

Heavy Gas Properties

Passive Dispersion Properties

Air flow, velocity, temperature, and density are disturbed by the release.

Air flow, velocity, temperature, and density are unperturbed by the release. Plume height and width are about equal, with greater height, lesser width than heavy gas plume. Plume concentration is proportional to source rate of release at all downwind distances. Plume width is independent of wind speed.

Plume height is much lower, width is much higher than passive dispersion plume. Plume concentration is not proportional to source rate because air entrainment rates are suppressed. Gravity-induced spread is inversely dependent on wind speed; eg plume width of Coyote 6 test (wind speed 4.6 m/s) is wider than that of Coyote 5 test (wind speed 9.7 m/s). Plume width is wider than passive dispersion plumes and with more stable atmospheres. There is strong heat flux from the substrate (ground or water) to the LNG pool and cold NG vapors that increases turbulence and air entrainment.

Plume width is inversely dependent on atmospheric stability. There is no or minimal heat flux from the substrate for passive dispersion.

To quantify the observations in Table 8.3, Morgan et al. (1984) regressed the eight tests from the Burro series and three tests from the Coyote series to obtain an equation for the maximum extent to the LFL, xmax (in m at any elevation and any crosswind location). The variables influencing xmax were taken as the volumetric discharge rate, q, in m3/min, wind speed, uw, in m/s, and atmospheric stability. Various measures of atmospheric stability are applied as discussed in Section 8.6. In this case, Morgan uses the ambient Prandtl number, NPr, reasoning that this is related to the Richardson number and hence to the Pasquill–Gifford stability class. Prandtl number is defined in terms of the heat capacity of air, CPair, the viscosity of air, μair, and the thermal conductivity of air, kair: Prandtl number = N Pr = Cp,air μ air kair

(8.9)

In the 11 tests evaluated, q varies by a factor of 1.6, uw by 5.4, and NPr by 2.5. This variation is sufficiently substantial to ascertain the effect of each variable. Morgan finds a correlation equation for these tests in the form where each variable is divided by its mean value for the 11 tests: α

xmax

β

q ⎛ ⎞ ⎛ uw ⎞ ⎛ N Pr ⎞ = A⎜ ⎝ 14.0 m 3 min ⎟⎠ ⎜⎝ 6.1 m s ⎟⎠ ⎝ 0.88 ⎠

γ

(8.10)

197

FACTORS AFFECTING PLUME LENGTH

Table 8.7 Regression coefficients compared from test data and Gaussian model predictions of the same tests

Modeled Test data (heavy gas properties) Gaussian model predictions

A

α

β

γ

230 118

0.07 ± 0.22 0.53

–0.21 ± 0.09 –0.63

0.57 ± 0.18 1.29

The test data were then fitted to a Gaussian model and the model results were regressed in the same manner as the test data. The results are compared in Table 8.7. The coefficients for the test data are considerably different from those of a Gaussian plume. The α coefficient indicates that xmax is affected only mildly by the volumetric discharge rate whereas the Gaussian dependence is nearly a square root (0.50). The β coefficient indicates xmax is inversely affected by wind speed by a mild −0.21 power compared with a strong −0.63 power for passive dispersion. The γ coefficient also shows a much lower dependence on atmospheric stability for heavy gas dispersion than for passive dispersion. 8.4.2

Models Predict Average Conditions of Fluctuating Plume

Concentration measurements made in field tests invariably detect a high degree of fluctuations. These are caused by such low-frequency events as wind direction change, or wind meander, as well as high-frequency turbulence generated by flow past obstacles, normal wind shear from ground contact, and LNG boiling rate fluctuations. Such fluctuations require averaging or “filtering” to obtain a smooth prediction that is more useful for risk analysis purposes. In fact, upon comparing analyzer sensor plots, no two dispersion tests produce the same concentration plots. If the same experimental initial conditions could be applied (ignoring the fact that wind conditions will essentially never be exactly the same for any two tests) and the experiment were repeated several times, the resulting “ensemble” of analyzer sensor plots could be averaged together to produce an ensemble average of concentration in space and time. Hence, model predictions aspire to predict this ensemble average. An example of responses detected by two sensors at different horizontal crosswind locations is shown in Figure 8.11 (dashed lines) for the Coyote Test 5. The solid line illustrates a hypothetical ensemble-average of the dashed-line data. The right-most peak in Figure 8.11 labeled for the sensor G06 is reportedly associated with an rapid phase transition (RPT). This illustrates one more rather distinct feature of LNG spills: RPTs act as an additional source of turbulence. 8.4.2.1 Averaging Time The gas sensor traces shown in Figure 8.11 are already filtered signals. High-frequency fluctuations are averaged out in the

198

VAPOR CLOUD DISPERSION MODELING

12 Coyote 5.

row: 140 m height: 1 m

Gas concentration (vol %)

10 G06 8 T03 6

4

2

0

0

50

100 Time (s)

150

200

Figure 8.11 Example measured and averaged concentrations for Coyote Test 5 at 140 m from source, 1 m high (Morgan et al., 1984).

electronics or in the data processing, to typically produce a 1- to 10-s average signal. A short averaging time of 1 s to 10 s is justified for dealing with flammable vapors because the period required for ignition with a strong ignition source can be less than 1 s. Figure 8.12 shows the trace from the same gas sensors for the Coyote 5 test with averaging time of 2 s and 10 s. Averaging reduces and broadens the peaks to the point that several peaks are at least partially merged. For dealing with the exposure to a toxic vapor, such as hydrogen sulfide (a component in natural gas from field wells, always removed in making LNG), a larger averaging time is appropriate, typically between 15 min and 60 min. These larger averaging times should correspond to the experimental dosage time used to obtain the toxic end-point for the vapor. Commonly-used toxic end-points are listed in Table 8.8. The averaging time should not be larger than the duration of the plume. For a finite duration release, the trailing edge begins moving forward when the source stops. The leading edge reaches its maximum length independent of the source duration. So the plume behavior for a constant discharge rate and constant wind speed is as illustrated in Figure 8.13. Defining the plume as concentrations within the LFL, the plume disappears when the centerline concentration drops below the LFL over the entire plume path. This occurs when the trailing edge catches up with the leading edge. The leading edge of the LFL contour reaches a maximum distance, xmax, at time t1 = xmax/uave where uave is the average plume velocity. If the release is under pressure (a jet release), the plume velocity starts out higher than the

FACTORS AFFECTING PLUME LENGTH

199

12 2-s averaging time

10 8 6

Gas concentration (%vol.)

4 2 0

0

20

40

60

80

100

120

140

160

180

120

140

160

180

Time (s) 12 10-s averaging time

10 8 6 4 2 0

0

20

40

60

80

100

Time (s) Coyote 5. Height 1. M. Downwind distance: 200 M. Figure 8.12 1984). Table 8.8

Gas sensor traces for Coyote Test 5 at 2 s and 10 s averaging time (Morgan,

Dosage time associated with toxic endpoints

Toxic Endpoint STEL (short term exposure limit) IDLH (immediately dangerous to life and health) ERPG (emergency response planning guidelines) ARPG (alternate response planning guidelines)

Dosage Time (min) 15 30 60 60

wind speed, uw. Ultimately, though, the plume speed approaches the wind speed. The top part of the plume actually moves faster than the bottom (plume shear) because wind speed increases with height. Models usually do not try to describe plume shear and use the wind speed at the height of the plume’s

200

VAPOR CLOUD DISPERSION MODELING

(a)

(b)

15 Time (min)

Time (min)

15

10

5

10 5 0 0

50

100

150

200

250

Distance (m)

0 0

1

2

3

4

5

Source rate (kg/s)

Lead Edge

Trail Edge

Steady-St

At 4

At 6

At 8

At 10

At 12

Plume Length at Various Times for Finite Duration Constant Source Rate (d) 12 10

12 10 8 6 4 2 0

Time (min)

Time (min)

(c)

8 6 4 2 0

0

2

4

6

0

50

100

150

200

Distance (m)

Source (kg/s) Start 0

Start 2

Start 6

Start 8

Start 4

Plume Trajectories with Decreasing Source Rate Figure 8.13

Illustration of behavior in time of leading and trailing edge LFL contour.

centroid of mass to ultimately move the plume at the same speed. There is also a slight diffusion at the leading and trailing edges that rounds the nose and tail of the plume, but this can be justifiably neglected. Similarly, the trailing edge of the plume begins to move forward when the source stops. This time is at the duration of the release, tdis. If the release rate and wind speed are constant, then the trailing edge will reach xmax at time t2 = t1 + tdis. Thus, after time tdis, the plume nose is or will soon become stationary and the plume length decreases as the trailing edge moves forward until it reaches the leading edge and the plume disappears. The top section of Figure 8.13 illustrates the plume length at various times as the horizontal lines between the leading and trailing edge trajectories. For an evaporating pool or a pressured jet with decreasing pressure, the source usually trails off gradually until all of the liquid is evaporated or the pressure bleeds off. When the source is decreasing, the leading edge of

250

FACTORS AFFECTING PLUME LENGTH

201

the plume reaches a maximum and then retreats in time. To see this, approximate the source rate as a stair-step function. Consider each successive source step as beginning a new leading edge. The leading edge generated by successively weaker source rates will reach shorter values of xmax. The trailing edge will not start until the last, shortest leading edge begins, so the overall result are successively shorter trajectories as depicted in the lower section of Figure 8.13. Again, horizontal lines between the leading and trailing trajectory at each successive start time depict plumes that reaches a peak extent and then retreat and disappears as the source stops. 8.4.3

Wind Speed for Longest Plume

As wind speed increases, there is a trade-off between increased air entrainment and increased advection (physical displacement) of the plume. The former produces a shorter plume to the LFL by decreasing the concentration more rapidly. The latter “stretches” a longer plume. The trade-off between these forces can result in an intermediate wind speed that produces the longest plume extent to the LFL, xmax. This “worst case” wind speed for maximizing xmax also depends upon the atmospheric stability, and upon the spill rate and spill size. Figure 8.14 plots xmax predicted by the SafeSite model against wind speed for the initial spill rate of Algerian LNG, taken as constant, for a Type 2 breach at the waterline with an initial LNG level above the waterline of 17.0 m. The weather is at neutral stability (Pasquill–Gifford D) with 20°C ambient temperature and 70% relative humidity. A maximum value of xmax occurs with a wind speed of 5 m/s for a 1-m diameter breach. For small spill sizes, the wind that produces a peak on Figure 8.14 is at very low wind speeds. Model predic-

3000

xmax to LFL (m)

2500 2000 1500 1000 500 0

0

1

2

3

4

5

6

7

8

9

10

Wind speed (m/s) Dhole = 1 m Figure 8.14

Dhole = 0.5 m

Dhole = 0.1 m

Plume extent as a function of wind speed.

202

VAPOR CLOUD DISPERSION MODELING

tions are often limited to wind speeds above a certain minimum (1 to 2 m/s). This is because the uncertainty in model predictions increases at low wind speeds as shown for the TŨV LPG (liquefied petroleum gas) test series by Woodward (1998b). 8.4.4

LNG Vapor Cloud Lift-Off Limits Hazardous Plume Length

For a range of wind speeds, some models (e.g., SafeSite) predict a point at which LNG spills on water will have enough heat transfer to warm the plume and change plume density from negative to positive buoyancy. Some modelers dispute whether lift-off can occur at concentrations above the LFL. Neither the Sandia nor the DNV study of large marine spills reported lift-off of LNG plumes, as those studies did not track plumes beyond ½ LFL and any lift-off after that was not assessed. If the plume “lifts off,” it becomes less of a hazard. After a plume lifts off, the hazard of ignition or freeze burn effects is rapidly reduced. Thus, if lift-off occurs, the hazard distance is actually the lower of xmax (the extent to the LFL) and the lift-off distance. Lift-off at concentrations above the LFL is not usually predicted for high wind speeds or small spills at any wind speed. Some experimental evidence is discussed in the next section (Falcon-1 test) that LNG vapors were superheated by water contact in an impoundment zone. 8.4.5

Scooping of Confined Vapors

As briefly discussed in Chapter 6, export or import terminals usually design a mitigation for spills of LNG, such as from a loading-arm break or a storagetank leak. The mitigation is an insulated trough that leads to an insulated impoundment sump. Vapors accumulating from evaporated LNG will gradually fill the sump or any other such impoundment zone. At issue is when and at what rate are the cold vapors entrained by the wind flowing over such partially confined volumes. This affects the predictions used to determine the required area of an LNG import terminal (see Chapter 6, Section 6.7.2). If the time for vapor overflow can be delayed long enough, the effective source rate of the vapors decreases by evaporative chilling of the impoundment zone. The Falcon series tests were conducted in one such impoundment zone. Chan (1992) modeled the Falcon tests using CFD model, FEMA3. Figure 8.15 illustrates how the FEMA3 model predicts the plume overflow of the impoundment zone for Falcon test 1. This test was conducted under stable atmospheric conditions at low wind speeds, and was notable for superheating the vapors by prolonged contact with the water in the walled-in area. Chan’s paper did not emphasize whether vapor scooping was observed (overflow of the impoundment walls before the impoundment filled with vapor), but the data are available and upon closer inspection may answer this point. Figure 8.16 compares measured concentrations at the center of the spill inside the fenced impoundment zone at 1-m high with predictions modeled

FACTORS AFFECTING PLUME LENGTH

100

C

D

F

B

C D

AB

E

A D

50

B E

203

0 I

0

Cmax = 54.1% at (–0.5 m, 16.5 m) –100

–50

0

50

100

150

200

250

Figure 8.15 Concentration contours above vapor fence predicted for Falcon-1 test. The contour levels are (in mole %): A = 0.5, B = 1, C = 2, D = 5, E = 10, F = 15, G = 25, H = 35, I = 50 (Chan, 1992) (reproduced by permission of Elsevier Science Publishing Inc.).

Experiment

Concentration (% vol)

40 30

FEM3A w/o vapor fence

20

FEM3A w/ vapor fence

10 0 0

100

200

300 Time (s)

400

500

600

Figure 8.16 Predicted and measured concentration at 1 M above water surface and center of spill for Falcon-4 test (Chan, 1992) (reproduced by permission of Elsevier Science Publishing, Inc.).

with and without the impoundment walls. The predicted concentrations remain above 2.5% for about 530 s with the walls as opposed to 330 s without the walls present. Chan concludes that vapor impoundment zone provides the following advantages: it significantly reduces concentrations in the near field, it delays cloud arrival time at downwind locations, and it significantly shortens downwind distance of the hazardous plume. Castro et al. (1993) conducted a wind tunnel study to analyze the problem of heavy vapors in a valley with winds blowing across the valley and entraining the heavy vapor. They produced a correlation for the steady-state concentration in the valley as a function of the generation rate of vapor and the wind speed. They also produced an equation for the transient concentration in the valley after the source stops. Of some significance, they found a low depth of heavy vapor (expressed as a Richardson number) and a minimum wind speed (or Reynolds number) at which vapor pooling occurs. With vapor pooling, the

204

VAPOR CLOUD DISPERSION MODELING

entrainment of vapor by the wind is slow to negligible. This seems to be confirmed by the Falcon tests since vapor overflow began only when the vapor impoundment volume was almost filled with LNG vapors.

8.5

EFFECT OF WIND, CURRENTS, AND WAVES ON LNG PLUME

Water current effects and the motion of an LNG tanker have been analyzed by Spaulding et al. (2007). The developed model is the LNGMAP model that predicts the behavior of an LNG spill from a moving LNG carrier both inside a harbor area or in the open sea. In the open sea, the hazard of an LNG spill is only to the LNG carrier, so the more important case is in a harbor area. The LNGMAP model approach is to model individually discrete masses of LNG, called spillets. Several hundred spillets are used; the number and size depend on the release rate and time. When the edges of spillets touch or overlap by a prescribed percentage, they are combined. Time steps range from 0.01 s to 0.2 s. Spatial scales range from 100 m to 5 km. National and international databases were applied for currents, wind conditions, and waves. Surface transport calculations used a wind-/wave-induced drift algorithm. The location of each spillet is tracked by first-order integration of the location vector: X i +1 = X i + U i Δt

(8.11)

The Ui vector is a combination of observations and predictions based on wind direction, tides, currents, and residual flows. Estimates of wave-induced transport use a drift factor (usually 2–3.5% of the 10-m high wind speed) and drift angle, θ, usually 0% to 20% of the resultant of the current and wind direction (to the right of the wind in the northern hemisphere). In addition, spillets are dispersed horizontally at the sea surface by the turbulence generated by the wind, waves, or shear forces via a random-walk model. Discharge rates are time-dependent blowdown from a vessel. Liquid spread rate is calculated by integrating Equation 7.10 (in Section 7.2.1 by Weber and Brighton, 1987). Evaporation rate is calculated using the film boiling heat transfer correlations of Klimenko (1981), Equations 7.37 to 7.45 in Section 7.2.4.1. Vapor dispersion used numerous DEGADIS model runs under different atmospheric stability classes and other meteorological conditions to predict the time for lift-off (buoyant vapor), which reportedly ranged on the order of 15–20 s. For fire, a constant burning flux was used. The solidsurface flame model with the flame height correlation of Thomas (1963), the tilt angle correlation of Rew and Hulbert (1996), the view factor defined in the TNO Green Book (1992), and the surface emissive power correlation of Mudan (1984). An example of predictions by the LNGMAP model is shown in Figure 8.17. Thermal contours from a burning pool of LNG are plotted at a sequence of

COMPARISON OF DISPERSION MODEL PREDICTIONS

Thermal Radiation Area — Tanker Continues on Original Path Thermal radiation kW/m2

Thermal Radiation Area — Tanker Changes Path and Grounds

10 knots

Thermal radiation kW/m2

4,000 8,000

10 knots

5–12 12–25 25–38 >38 Continues Grounds Stops

5–12 12–25 25–38 >38 Continues Grounds Stops

0

205

16,000 Meters

0

4,000 8,000

16,000 Meters

Figure 8.17 Example fire radiation contours for spill from moving LNG carrier (Spaulding et al., 2007) (reproduced by permission of Elsevier Publishing, Inc.)

time for two alternative paths of the leaking LNG carrier. This illustrates that a moving carrier could sometimes avoid imperiling sensitive areas.

8.6

COMPARISON OF DISPERSION MODEL PREDICTIONS

It is widely accepted that acceptable agreement between model predictions and measured values is when predicted concentrations at the same location as measured is within a factor of two (e.g., Britter, 2002). A few example comparisons of predictions between models and comparison between models and experimental data are provided here. To illustrate how test data are used for model validation, Figure 8.18 compares predictions by the SafeSite3G® model with measurements from the Coyote 5 and Burro 5 tests. The Coyote 5 test shows one of the best results found, and the Burro 5 test represents a more commonly found comparison. Both a maximum value and an average concentration are plotted for the Burro 5 case. Standard metrics or Statistical Performance Measures (SPM) have been developed for model validation studies (Hanna et al., 1991). SPMs as recommended by Ivings et al. (2007a) are reproduced in Table 8.9 with measured concentrations, Cm, and predicted concentrations, Cp, and the brackets indicating averaging over all pairs of Cm and Cp. Hanna explains that the mean relative bias (MRB) can be overwhelmed by major outlying data points because the denominator contains the observed (Cm) and predicted (Cp) concentration for each point. Thus, if both Cm and Cp approach zero, that term approaches infinity. This is why Hanna defines the denominator as the average of 0.5 (Cm + Cp) (HSL, 2008, p. 17). Comparisons have been made of model performance, using the SPM in Table 8.9, including Chang and Hanna (2004) updating the earlier evaluation

VAPOR CLOUD DISPERSION MODELING

100

100 Centerline Conc., mole %

Centerline mole % methane

206

10

1

0.1 10

1000

100 x, m SS3G

10

1

0.1 10

100

Data

SS3G

Coyote 5 Test Figure 8.18

100

Distance (m)

Obs.,%

Ave, %

Burro 5 Test ® 3G

Example plots for validation of SafeSite

model with test data.

Table 8.9 Statistical performance measures for evaluating dispersion model performance

SPM

Definition

Mean relative bias MRB = Mean relative square error MSRE =

Cm − C p 1 (Cm + Cp ) 2

(Cm − Cp )2 1 (Cm + Cp )2 4

FAC2: Fraction of predictions within factor of 2 of measurements

⎛ Cp ⎞ 0.5 ≤ ⎜ ≤ 2.0 ⎝ Cm ⎟⎠

Geometric mean bias

⎛C ⎞ MG = exp ln ⎜ m ⎟ ⎝ Cp ⎠

Geometric variance

⎡ ⎛C ⎞⎤ VG = exp ⎢ ln ⎜ m ⎟ ⎥ ⎣ ⎝ Cp ⎠ ⎦

2

study of Hanna et al. (1993). Other validation studies include Djuim et al. (1996), Ivings et al. (2007a), Nielsen and Ott (1996), and early works by Woodward et al. (1982) and by Britter (1991). Individual model validations are also reported (Britter, 2002 for the Universal Dispersion Model, UDM used in PHAST). A committee of experts was convened by the U.S. NASFM to review the Model Evaluation Protocol developed by the UK HSL for the NFPA. A conclusion provided by the HSL, and accepted by the review committee is (HSL, 2008):

COMPARISON OF DISPERSION MODEL PREDICTIONS

207

There is inherent or natural uncertainty in the atmosphere and this cannot be accounted for or eliminated by even the best scientific model. Thus, we cannot expect models to perform better and better. Biases and RMSE’s [root mean square error] will not approach zero as time goes on. There is a certain minimum error that will always be present.

A recent smaller study by Ferrario et al. (2007) provided some of the data plotted in Figure 18.19. Selected experiments were evaluated to obtain the SPM plotted in Figure 18.19: • • •

Burro 3, 7, 8, 9 Coyote 3, 5, 6 Maplin Sands 27, 29, 35, 39, 56

The model versions used are: • • • •

DEGADIS 2.1 (Havens and Spicer, 1990); SLAB (Ermak, 1980); PHAST 6.51 (Witlox, 2000); and SafeSite 2.3.27 (Baker Eng. and Risk Consultants, Inc., 2003, 2005

The SafeSite model predictions are not in Ferrario et al. (2007) but are added here using the specified input values.

Geometric variance (V/G)

10

1 0.1

1

10

Geometric mean bias (MG) DEGADIS

SLAB

PHAST

SafeSite

Figure 8.19 Performance of several models with LNG tests.

208

VAPOR CLOUD DISPERSION MODELING

The parabolic curve plotted in Figure 8.19 is an acceptable error bound given by: ln VG = ( ln MG )

2

(8.12)

All four of the models tested fall within this acceptable error-bound curve. A comparison of model predictions for light LNG on water with five hole sizes was made by Pitblado et al. (2006). The case is a Type 2 breach of a single tank of a carrier with an inventory above the water level of 25,000 m3 at a level of 17 m. The hole sizes selected are labeled: 1. 2. 3. 4. 5.

10,000 m3/h rate: Maximum credible sabotage event (60 min) 0.75 m: Maximum credible hole from accidental operational events 1.0 m: Used for comparison with FERC-ABS Consulting (2004) report 1.50 m: Maximum credible hole from terrorist events 5.0 m: Hole size chosen to approximate instantaneous release

Table 8.10 provides the comparison of predictions by the following models: • •

PHAST 6.4 of 2004 by DNV SafeSite 2.3.27 of 2008 by Baker Risk (not in original paper, added here)

Table 8.10 Comparison of predictions for LNG spill on water

Dispersion Predictions for D stability 5 m/s Wind Speed Case

PHAST

SafeSite

DEGADIS

10,000 m3/h 0.75-m hole 1.0-m hole

790 m 920 m

810 m 1060 m 1430 m

1.5-m hole 5.0-m hole

2000 m

1380 m 1765 m 2000 m ABS D3 3150 m

1830 m 4870 m

HGSYSTEM

SLAB

1250 m 1740 m

1635 m 2205 m

CANARY

1000 m 3440 m

4500 m 4080 m

Dispersion Predictions for F Stability 2 m/s Wind Speed 10,000 m3/hr 0.75 m hole 1.0 m hole

1600 m 1400 m

1.5 m hole 5.0 m hole

3100 m

1080 m 2970 m 3200 m 7450 m

2420 m 3250 m 3300–3400 m ABS 6700 m 4100 m ABS

Notes: ABS reference is ABS (2004). CANARY reference is Cornwell (2001).

1690 m 2460 m

4500 m 7180 m 780 m

4265 m

7910 m 3730 m

DESCRIPTIONS OF DISPERSION TEST SERIES

• • • •

209

DEGADIS 2.1 of 1985 by Havens and Spicer HGSYSTEM 3.0 by Shell SLAB of 1988 by Lawrence Livermore Laboratories CANARY by Quest Consultants

The results listed are reported by the authors unless otherwise indicated. The ambient temperature is taken as 20°C, with wind speed in m/s, and stability class of D5 and F2 unless otherwise noted. The specified roughness length is 0.3 mm except for DEGADIS runs reported in the ABS (2004) as 10 mm. The distance to the LFL is reported to a plume centerline concentration of 4.4%. Pool radius is calculated on the boiling flux of 0.085 kg/(m2s) for DEGADIS, HGSYSTEM, and SLAB runs. SafeSite used a boiling flux of 0.20 kg/(m2s). The difference in predictions between PHAST and SafeSite is small, as would be expected since they use the same basic modeling equations. The variation in predictions between the other models is much larger than the variation in validation runs illustrated in Table 8.10. Plume length predictions are highly nonlinear with respect to discharge rate, wind speed, and atmospheric stability.

8.7

DESCRIPTIONS OF DISPERSION TEST SERIES

Descriptions of the LNG spill test series cited in Table 8.1 are provided below, partly from Ermak et al. (1988). 8.7.1

Matagorda Bay Tests

Tests were conducted by the Esso Research and Engineering Company sponsored by the American Petroleum Institute in Matagorda Bay, Texas through pressurizing an LNG tank with nitrogen and projecting it out an elevated pipe. A considerable amount evaporated in the arched plume before it reached the water. Volumes ranged from 0.73 m3 to 10.2 m3 for 18.9 m3/min. The plume was photographed sequentially from above and measured with sensors on buoys. For some tests, the plume exceeded the array of sensors so the distance to the LFL was not detected. The longest distance to the last visible fog from the plume was 1372 m for a spill size of 8.37 m3 and a wind speed of 4 m/s. 8.7.2

Shell Jettison Tests

Shell performed a series of six tests in which LNG was jettisoned for the SS Gadila, a 75,000-m3 LNG carrier at a location about 70 mi west of St. Nazaire, France. The primary objective was to determine the feasibility of emergency jettison of cargo with high discharge rates while the ship is stationary, as well

210

VAPOR CLOUD DISPERSION MODELING

as at low discharge rates while the ship is moving. Four tests were conducted while the ship moved from 3 knots to 10.5 knots (1.5–5.4 m/s). Two tests were done with two different nozzle sizes, 51 mm and 102 mm (2 in. and 4 in.) discharging roughly horizontally, located 18 m above the water with wind speeds from 1.9 m/s to 5.1 m/s. Infrared cameras showed that with the 51-mm nozzle, LNG pools did not form on the sea, and only isolated patches formed with the 102-mm nozzle. Ice formation and RPTs were not observed and no static charges that could ignite the LNG were generated in the jet. The test conclusion is that LNG can be jettisoned without the LNG contacting the ship and causing cryogenic damage or the vapor plume engulfing the ship. 8.7.3

Avocet, Burro, and Coyote Test Series

The Avocet tests were four 5-m3 spills onto a pond conducted by Lawrence Livermore National Laboratories (LLNL) at the Naval Weapons Center (NWC) in China Lake, California from August 31 to November 20, 1978. They were performed to gain insights into the measurements necessary for the larger Burro test series. The wind speed varied by as much as a factor of three and the wind direction varied by as much as 60° during each test. LLNL developed a model of varying wind speed and direction that proved useful for interpreting field data and understanding effects of wind variations on the vapor plume. Two series of field experiments with LNG were conducted jointly by LLNL and the NWC at the NWC facility. In the Burro series, nine tests, nominally 40-m3 spills, were completed between June 6 and September 17, 1980, designed to test dispersion of spills on water. The Coyote series consisted of 10 experiments from September 3 to November 24, 1981, designed to spill onto a 1-m deep pool to specifically investigate the phenomenology of RPTs, which had been seen during two of the Burro tests, as well as the characteristics of fires from igniting the LNG vapor clouds. The LNG spilled from a pipe 25 cm in diameter flowing straight down from about 1 m above the pool. However, direct water penetration was specifically avoided. A splash plate was installed below the spill pipe outlet at a shallow depth beneath the water surface to limit the LNG penetration into the water. The water basin was 58 m in diameter. The ground level immediately surrounding the water basin was about 1.5 m above the water level. Downwind (northeast) of the spill pond the terrain sloped upward at about a 7° angle until it reached a height of 7 m above the water at a downwind distance of 80 m. Thereafter, it remained relatively flat, rising less than 1° to the left and dropping to the right. 8.7.4

Maplin Sands Test Series

The Maplin Sands tests were performed by Shell Research Ltd in the summer of 1980 to study the dispersion and combustion of releases of LNG and LPG

DESCRIPTIONS OF DISPERSION TEST SERIES

211

onto the sea. Some 20 spills of up to 20 m3 of LNG were made at a point 350 m offshore on an area of tidal sands on the Thames estuary in England. A 300-m dike was constructed to retain seawater at low tide, but whenever possible, spills were made at high tide when the winds were blowing offshore. The maximum change in level at the offshore edge of the dike was 0.75 m. Behind the 5-m high seawall was a flat farmland. Instruments were deployed on 71 floating pontoons along nearly semi-circular patterns at three radii inside the dike and six radii outside the dike. Gas sensors were at elevations of 0.5–0.9 m, 1.4 m, and 2.4 m. Ten instantaneous releases were made by rapidly sinking a barge with a tank loaded with LNG or LPG. The spill volumes tested were 5–20 m3. An RPT occurred on one instantaneous LNG spill, developing an overpressure of 0.26 psi (1.8 kPa), and damaging the barge, ending the barge tests. In the early continuous experiments, liquid was released from a vertical 0.15-m diameter pipe. In four tests, the pipe was 2 m to 3 m above the water surface directed downward. A high percentage of the LNG evaporated before reaching the water. In subsequent tests it was lower, and in three tests the pipe was below the water surface, resulting in a more buoyant cloud. Later, the end of the spill pipe was flared into a vertical-axis cone with a horizontal plate below it at the waterline. The liquid emerged from a slot between the cone and the plate with negligible vertical momentum. The spatial resolution, especially in the lateral direction, of the gas concentration data is insufficient to determine the cloud widths and the average properties corresponding to that width. Cloud widths are available from overhead images of the visible cloud for all tests except for Test 15. However, information is insufficient to determine the average properties within the visible cloud because the relationship among gas concentration, humidity, and temperature is uncertain. … These problems were identified, but not necessarily solved in the Shell papers. (Ermak et al., 1988)

The distance to the LFL was found to be within the visible vapor cloud for the humidity range of 50% to 100%. 8.7.5

Falcon Test Series

The U.S. Department of Transportation and the Gas Research Institute initiated a research program to evaluate methods for predicting LNG dispersion distances for situations involving obstacles and partial confinement barriers. As part of this program, the LLNL conducted a series of large-scale field experiments called the “LNG Vapor Barrier Verification Field Tests,” also referred to as the Falcon Series at the Gaseous Fuels Spill Test Facility, Frenchmen’s Flats, Nevada in the summer of 1987. Five tests were conducted on water in a 0.76-m deep, 40 m by 60 m pond in order to generate vapor at liquid spill rates of 9 m3/min to 30 m3/min with spill volumes from 21 m3 to

212

VAPOR CLOUD DISPERSION MODELING

66 m3. The pond was enclosed in a fiberglass vapor fence 88 m long, 44 m wide, and 8.7 m high. In addition, a 17.1 m wide-by-13.3 m high billboard-like barrier was erected upwind of the pond to simulate an obstruction such as a storage tank. The LNG was fed through a 0.3-m diameter spill line that split into four 0.15-m diameter distribution arms to approximate a broad spill. The complexity of the setup complicates model replication. Several RPTs occurred, causing some of the instruments inside the vapor fence to malfunction. The Falcon Test 1 generated superheated vapors by prolonged contact with the water under calm, stable wind conditions, which has implications for plume lift-off.

8.8

VAPOR INTRUSION INDOORS

One of the only events that could develop into a damaging explosion is for vapors from a passing plume from an LNG spill, to enter buildings through open doors, windows, and cracks, but mainly through the air intake to a ventilation system. This possibility is recognized by regulations such as the U.S. NFPA 496 Section 7.2.3 “Standard for Purged and Pressurized Enclosures.” By this code, the air intake for pressurized enclosures is required to be outside of a hazardous classified area (division 1 or division 2), that is outside of areas likely to have flammable gases. Thus, the code provides a high economic incentive to place the air intake duct in a safe location, which, in practice means it is usually at least 25 ft above the roof for control rooms. The concentration inside a building would have to reach the LFL to pose an explosion hazard inside the building.

8.8.1

Basic Response for Indoor Concentration Buildup

The indoor concentration buildup from a passing vapor cloud is basically a first-order process in time t, with the rate of change of the indoor mass fraction concentration, wC(t), proportional to the outdoor concentration, wCout(t). The equations that represent this process developed later, but it suffices to show the form of the solution for a simple case as Equation 8.13: t − t0 ⎞ t − t0 ⎞ ⎤ ⎡ −⎛ −⎛ ⎢ ⎝ τ ⎠ ⎝ τ ⎠⎥ wC ( t − t0 ) = wC ( t0 ) e + wCmax ⎢1 − e ⎥ ⎢ ⎥ ⎣ ⎦

(8.13)

The outdoor concentration curve, which is the “driving force,” has two parts, which for the simplest example for the driving force is a box curve, with the initial value, wCout(t0), of zero and:

213

VAPOR INTRUSION INDOORS

⎧50 for 0 < t ≤ 10 min. wCout = ⎨ ⎩0 at t > 10 min.

(8.14)

The indoor concentration also has two terms that apply in the same two time frames. ⎧0 at t0 = 0 for 0 < t ≤ 10 min. wC ( t0 ) = ⎨ ⎩wC (10 min ) for t > 10 min.

(8.15a)

⎧50 at t0 = 0 for 0 < t ≤ 10 min. wCmax ( t0 ) = ⎨ for t > 10 min. ⎩0

(8.15b)

The indoor concentration first rises from zero toward an ultimate maximum, wCmax, which equals, in this case, the outdoor driving force, wCout. Then, when the source concentration drops to zero at 10 min, the second term drops out, the indoor concentration decays toward a wCmax of zero. Figure 8.20 plots this solution for a single room building with a single leak point and a single exhaust point to illustrate that a first-order response depends on the indoor mixing time constant, τ. With larger time constants, the peak indoor concentration decreases, and the response time is elongated. In Figure 8.20, the time constants of 11 and 20 minutes represent relatively high internal mixing with the ventilation system still running and pulling in external vapor from the passing cloud. The large time constant of 133 min represents the ventilation system turned off, so only limited leakage admits vapors from the passing cloud. This is quantified later. 60

wC (mass fr)

50 40 30 20 10 0 –5

0

5

10

15

20

25

30

35

40

Time (min) Wcout

Tau = 11.1 min

Tau = 20 min

Tau = 133 min

Figure 8.20 Concentration of LNG vapors inside a building for a passing outdoor concentration of 50 mass % for 10 minutes with various time constants, τ.

214

VAPOR CLOUD DISPERSION MODELING

8.8.2

Experimental Observations Show Low Indoor Concentrations

The problem of developing damaging indoor concentrations is much greater for toxic gases than for flammable vapors. So, we necessarily turn to the technical literature for toxic gases, where far more experiments have been conducted and models developed. Buschmann (1975) conducted experiments with 1000-kg releases of Dichloro-difluoro-methane (Refrigerant 12) vapor in the Netherlands from October 1973 to April 1974. A continuous release was made for 10 min on the windward side of a house, and concentrations were measured at several rooms in the house on the windward and the leeward sides. The performance measures are the peak concentration ratio indoors to outdoors and the ratio of indoor to outdoor dosage (integral of concentrations over time). The results are summarized in Table 8.11 for Experiment 1: doors and windows closed normally; and Experiment 2: doors and windows closed and then sealed with tape. An important finding was that it was important to not only warn people that the cloud was coming, but also to alert them that the cloud has passed. After the cloud has passed, it is necessary to open doors and windows to reduce the inside concentrations and decrease the total dosage. Again, dosage is not so important if the vapors are not toxic. The measured ratios in Table 8.11 indicate that a substantial reduction is to be expected between outdoor and indoor concentrations. To interpret the ratios in Table 8.11 for an LNG cloud passing a plant building such as a maintenance shop, as a rough estimate for the peak indoor concentration to reach the LFL of 5.0 mol %, the outdoor concentration would have to be on the order of 10 times as much or 50 mol %. 8.8.3

Concentration Reduction by Plume Impinging on Buildings

Complications arise in predicting the outdoor concentration of a plume flowing around a building. A gas plume lower than the height of a building and impact-

Table 8.11 Reduction in dosage to indoor residents from passing vapor cloud

Performance Criterion

Peak concentration Inside/outside Dosage Inside/Outside

Location or Action

Experiment 1 Close Windows and Doors Normally

Experiment 2 Close and Tape Seal Windows and Doors

Windward side Leeward side With advising of cloud passing (reopen at 15 min) Without advising of cloud passing (reopen at 60 min)

1/10 1/20 1/5–1/20

1/30 1/50 1/5–1/50

1/2–1/3

1/3–1/8

VAPOR INTRUSION INDOORS

215

Figure 8.21 Downwash and ground level vortex rollup on the upwind side of a building and well-mixed wake on the downwind side (Wilson et al., 1993).

ing upon a building develops patterns shown schematically in Figure 8.21 from Wilson et al. (1993) cited in Wilson (1996a). Using water channel simulations around a wide range of building sizes, downwind locations, and orientations, Wilson (1996b) found that a large building will dilute an impinging plume by a factor of two or more if the release location is less than 10 building heights upwind. A well-mixed zone occurs on the leeward side of the building. Flow patterns on the windward side of the building enhance gas dilution and mixes upper-level air down into a ground-level plume. The effect of a vapor plume passing a single building is indicated in Figure 8.22 and for passing multiple buildings in Figure 8.23. Figure 8.22 by Wilson (1996a) shows that the concentration around a building located 5.4 building heights downwind of a ground-level release is reduced by a factor of five by building-induced mixing. Here, X is the downwind distance from the release source to the building model, R is the building width, and θ is the direction of the water flow (representing wind flow directly onto the building). For Figure 8.23, Deaves (1987) used the model of Brighton (1986) to predict the change in concentration as a chlorine release (at the listed rate) encounters a boiler house (40 m wide × 18 m high at 50 m from the source), offices (50 m wide × 10 m high at 100 m), and a warehouse (120 m wide × 8 m high at 220 m). A substantial decrease in centerline concentration is predicted for each building in series with the plume. 8.8.4

Models of Infiltration into Buildings

The basic model described by Equation 8.13 has been modified variously to account for a number of features, including:

216

VAPOR CLOUD DISPERSION MODELING

1.5 1.4 +

Plume concetration ratio with/without building

1.3

Short building θ = 0.0° X = 270.0 mm R = 63.0 mm

Experiment Theory

1.2 1.1 1.0 0.9 0.8 0.7

+ +++

0.6 0.5

+ ++

0.4 0.3

+ ++ +++ + ++ ++++++ ++

Flow direction

Cwl --Cwe

0.2 0.1 0.0 –10.0 –8.0 –6.0 –4.0 –2.0 X+ R Figure 8.22

0.0

2.0

4.0

6.0

8.0

10.0

Distance from building front face

Concentration profile of plume impinging upon face of a building (Wilson, 1996b).

100,000

Concentration (ppm)

5 kg s–1 10,000

0.33 kg s–1

1000

100 Buildings 10

0

100

200

300

400

500

Distance (m) Figure 8.23 Effects of passing buildings on plume centreline concentration (Deaves, 1987) (reproduced by by permission of AIChE).

VAPOR INTRUSION INDOORS

217

(1 – fa) Fin

fa Fin

(1 – η) V

Fin Fin or ηFin

Fin + Fdis

Fdis + fa Fdis

ηV Fdis

Figure 8.24 2000).

•

•

•

Schematic for infiltration of gas to a building with recycling ventilation (Woodward,

Multiple rooms or chambers inside a building and consequently multiple input and exhaust points as developed by Kakko (1990). Adsorption and desorption of vapors from internal surfaces (important for toxic gases) as developed by Karlsson and Huber (1996). How indoor toxic dosage is strongly reduced for vapors that have a nonlinear dosage response to concentration (Wilson and Zelt, 1990).

Nearly all process plant buildings have a heating-/air-conditioning system with a recycle system to economize and heat or cool only a fraction of the fresh air intake. An analytical model accounting for recycled air is the WELMIX model (Woodward, 2000) schematically outlined in Figure 8.24. The model accounts for indoor discharges of rate Fdis, which can be set to zero and neglected, and for bypassing a fraction (1 − η) of the building volume, which can also be neglected by setting η to zero. The mass balance (in kg/s) on a building of volume V(in m3) with interior air density ρ (kg/m3) and total air in-flow, Fin, total out-flow from the building Fout, and a discharge in the room of Fdis is: V

dρ = Fin + Fdis − Fout dt

(8.16)

The volumetric ventilation rate is defined as the number of air changes per hour, NCH. For areas such as kitchens and rest rooms, the design is usually specified as 6 air changes/hour (ACH); office areas can be 2 ACH. Thus, Fin =

VρN CH 3600

(8.17)

With a fresh air rate (including contaminated air intrusion rate) of FSH, the fresh air make-up ratio, fa, is: fa =

FSH Fin

(8.18)

218

VAPOR CLOUD DISPERSION MODELING

with no contamination by a passing plume, the mass fraction wSH is zero. If the ventilation air intake is closed and FSH represents only intrusion of the passing plume, wSH is finite equal to the outside plume concentration. Assuming steady state in the ducts, pressures and flows will be constant. The flow at the nodes (mixing or branching points) adds to zero, so the overall mass balance is as shown in Figure 8.24. Defining the exhaust flow rate as Fex and the recirculation rate as Fcir, at the right-hand node (separation of exhaust and circulating flow): Fcir = (1 − Fa ) Fin

(8.19a)

Fcir = (1 − Fa ) Fin

(8.19b)

Fout = Fin + Fdis

(8.19c)

Thus, at the extremes: When fa = 1 (all fresh air ) Fex = Fdis + Fin and Fcir = 0

(8.20a)

When fa = 0 (total recycle) Fex = Fdis and Fcir = Fin , FSH = 0

(8.20b)

The component mass balance on the contaminant C is defined in terms of the merged input stream (fresh air, leaked contaminant, and recirculation flow), the internal discharge at concentration wC,dis and the exhaust flow: ηV

dwcρ = Fin wc ,in + Fdiswc ,dis − ( Fin + Fdis ) wc dt

(8.21)

At the left-side node point (merge of fresh air and circulation flow), the component balance is: fa Fin wcsh + (1 − fa ) Fin wcout = Fin wcin

(8.22)

This gives wCin since Fin can be eliminated. At the right-hand node point (split of exhaust and recycle), the component balance is:

( Fin + Fdis ) wc = Foutwcout

(8.23)

which with Equation 8.19c establishes that: wc = wcout

(8.24)

Substituting Equation 8.22 and 8.24 into Equation 8.21 gives the differential equation in standard form:

219

VAPOR INTRUSION INDOORS

dwc wc + =q τ dt

(8.25)

dwc w f F w + Fdiswcdis + [ Fdis + fa Fin ] c = a in sh ηVρ ηVρ dt

(8.26)

as

The analytic solution for this equation is Equation 8.13 with: ⎧ qτ wC max = min ⎨ ⎩1

(8.27)

with qτ =

fa Fin wsh +Fdiswcdis Fdis + fa Fin

τ=

ηρV Fdis + fa Fin

(8.28)

(8.29)

which reduce when Fdis is zero to: qτ = wSH

(8.30)

and with Equation 8.17: τ=

ρV 3600 = fa Fin fa N CH

(8.31)

With Equation 8.31, the example time constants used in Figure 8.20 can be interpreted in terms of the number of air changes per hour and the fresh air makeup ratio, as in Table 8.12.

Table 8.12

Example time constants for Figure 8.21

NCH. Air Changes/Hour 6 0.5

fa Mass Fr Fresh Air

τ, Time Constant (s[min])

0.5 0.9 0.5 0.9

1200 (20) 666.7 (11.1) 14,400 (240) 8,000 (133.3)

220

8.9

VAPOR CLOUD DISPERSION MODELING

THEORETICAL BASIS FOR SUPPRESSION OF TURBULENCE

A theoretical basis has been developed that justifies the observation that heavy gas plumes tend to suppress turbulence within a plume and to decrease air entrainment below that of normal ambient air. Without elaborating rigorously, we introduce the concept that air entrainment to a plume can be defined by vertical and horizontal entrainment velocities, otherwise designated as entrainments to the top of a plume, utop, and to the sides of a plume, uside. The top surface entrainment generally dominates over the side entrainment except very near the source. The top surface entrainment velocity is formulated to have the same functionality as the vertical diffusion coefficient, Kz with a functional form given by: Kz =

ku*z φ(Ri* )

(8.32)

where k is the Von Karman constant (value 0.4), z is vertical height, and u* is the “friction velocity,” a constant in the formula for vertical wind profiles. To retain this form, utop is defined by: utop =

k3u*

φ(Ri* )

(8.33)

where k3 is a constant. The layer Richardson number is intended to define the ratio of buoyancy forces to turbulent forces. It is defined in terms of the densities of the cloud, ρcld, and the ambient air, ρair, both evaluated at the centroid height of the plume, and also the effective height of a plume, Heff and a fraction hd that represents whether the plume cross-section is a hemi-ellipse (grounded) or a full ellipse (elevated) or in between. Ri* =

g (ρair − ρcld ) Heff (1 + hd ) ρcld u*2

(8.34)

The entrainment function ϕ(Ri*) represents the effect of cloud stratification or buoyancy on the entrainment velocity. On the one hand, positively buoyant clouds lifting off are known to have enhanced turbulence. For negative buoyancy (negative Richardson number), the form given by Havens and Spicer (1990) is: φ=

1 0.6 1 + 0.65 Ri*

For positive buoyancy, the function recommended by Witlox (1993) is:

(8.35)

THEORETICAL BASIS FOR SUPPRESSION OF TURBULENCE

221

Utop/U* 1.0

McQuaid

0.1

Kranenburg Kantha et al. 0.01

Scranton et al.

0.001 0.1

1

10

100

1000

Ri* Figure 8.25 Entrainment velocities used in several dispersion models compared with experimental water channel data (Spicer, 1985).

φ = (1 + 0.8Ri* )1 2

(8.36)

Entrainment velocities predicted by Equation 8.33 with ϕ given by Equation 8.36 are plotted in Figure 8.25 (Spicer, 1985) in comparison with data by McQuaid (1976), Kantha et al. (1977), Scranton and Lindberg (1983), and Kranenburg (1984). Kranenburg’s data were measured using a straight water channel with wind-induced flow of water over a salt solution. These data fall near the curve in Figure 8.25 given by Equation 8.35, indicating that as stability increases with sinking buoyancy, the top entrainment velocity decreases.

9 LNG POOL FIRE MODELING Chapter 9 reviews important characteristics of liquefied natural gas (LNG) pool fires, test data, fire models, and radiant energy received at vulnerable locations. Such predictions currently require an order of magnitude extrapolation from test data. The complex physics involved are not yet proven, yet public safety is involved, so it is necessary to extrapolate to obtain some estimates, but it is important to also state uncertainty levels and key assumptions. The literature for fire data and modeling is extensive. Zabetakis (1965) assembled a vast collection of fire data, and two very useful handbooks are the SFPE Handbook (SFPE, 1995) and the current edition of Lees (Mannan, 2005). 9.1

TYPES OF FIRES FROM LNG FACILITIES

As introduced in Chapter 1, there are four types of fires in general and of LNG fires in particular: 1. Pool Fire and Trench Fire (Circular and Rectangular Sections) A fire burning over a pool of liquid fuel. 2. Jet Fire A narrow, long turbulent jet, like a blowtorch. 3. Flash Fire A flame front that moves through a vapor cloud. 4. Fireball A nearly spherical, rising, burning mass of vapor or flashing liquid. LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

222

POOL FIRE CHARACTERISTICS

223

Pool fires and trench fires and their mathematical modeling are treated in this chapter. Jet fire, flash fire, and fireball behavior and modeling are treated in Chapter 10. 9.2

THE CHALLENGE FOR POOL FIRE MODELING

As established in Chapter 1, LNG pool fires pose the greatest hazard to people and structures from loss of containment incidents. The largest LNG pool fires are expected to be (1) from LNG impoundment sumps or diked areas around an LNG storage tank or (2) an unconfined spill from a tank of an LNG carrier at sea. LNG impoundment dikes at import terminals are normally rectangular and can vary from 10 to 30 m in equivalent diameter. Low dikes surrounding a single LNG storage tank at an import terminal can be as large as a 100-m equivalent diameter. Postulated unconfined spills of LNG on water from a single tank of an LNG carrier range from about 330- to 512-m diameters (Hightower et al., 2004). As shown below, the largest published LNG pool fire experiment to date has been a 35-m diameter test on land. Clearly, the challenge facing scientists and engineers is to establish correct principles from relatively small-scale fire tests that will enable accurate extrapolation to predict the properties of large pool fires. Unfortunately, this requires speculation on how physical mechanisms change as the LNG spill size becomes very large. This chapter addresses many important aspects of pool fires. General characteristics are addressed in the Section 9.3 and available experimental data are summarized in Section 9.4. After this, modeling aspects are addressed including burning rates (Section 9.5), the simplest modeling approach—the point source model (Section 9.6), then several sections addressing solid flame models: flame length (Section 9.7), flame tilt (Section 9.8), flame drag (Section 9.9), surface emissive power (SEP) (Section 9.10), atmospheric transmissivity (Section 9.11), trench fires (Section 9.12), and view factors (Section 9.13); next, computational fluid dynamics (CFD) approaches are discussed (Section 9.14), then model comparisons (Section 9.15) are provided and fire engulfment issues (Section 9.16) are addressed. 9.3

POOL FIRE CHARACTERISTICS

There is a definite stochastic element to fire structure, but also, upon closer analysis, systematic elements appear as described here. 9.3.1

Fires Are Low-Momentum Phenomena

Fires in general develop upward velocity by a buoyancy pressure, Fb, generated by the density difference between ambient air, ρair, and the density of the

224

LNG POOL FIRE MODELING

fire plume, ρcld, over the vertical height of the fire, Δz, times the gravitational acceleration, g: Fb = gΔz(ρair − ρcld ) .

(9.1)

That is, pressure, like the pressure that floats a ship, results from displacing a volume of cool air with hot combustion products. The difference between the relatively large buoyancy force that floats a ship weighing hundreds to thousands of tons and the buoyancy of a fire plume is that the density of the fire plume is very low. By the ideal gas law ρcld =

MWcld Patm MWair Patm and ρair = , RTcld RTair

(9.2)

where R is the gas constant (8314.32 J/kgmole K), Patm is atmospheric pressure, MW is the molecular weight, and T is temperature of air and cloud, respectively. The molecular weights are about equal (≈28 kg/kmole). At an average temperature for the fire plume of 800°C (1073 K) in air at 20°C (293 K), the respective densities are 0.318 and 1.165 kg/m3 compared to the density of water of about 1000 kg/m3. The low-momentum characteristic of fire results in the familiar flicker and random variation on flame fronts, readily affected by slight variations in local wind speed and air turbulence. Thus, spatial and time averaging is important in measuring essentially all fire properties. In addition, flame fronts readily bend in the wind and respond to drag forces of the wind flowing around a fire, producing a flame dragged into the leeward wake of the fire as shown in Figure 9.1.

Figure 9.1 Example of pool fire and flame drag (Sandia Laboratories, 2009).

POOL FIRE CHARACTERISTICS

Sandia National Labs. Water Impact Facility, 10-m pool diameter, 300-gpm spill rate

225

Montoir test, 35-m pool diameter on insulating concrete (Raj, 2007)

Figure 9.2 Examples of LNG pool fires: 10-m smokeless pool and 35-m smokey pool (Blanchat et al., 2007).

Fire characteristics differ with the material, the pool size, and the substrate. Figure 9.2 contrasts two fires with a nearly calm wind. A 10-m diameter unconfined LNG fire on water is bright with no smoke and a columnar structure. A 35-m diameter LNG fire in a dike of insulating concrete (the test fire at Montoir (1987)) is smoky with a considerably turbulent structure. 9.3.2

Fire Structure

Based on observations and measurements by investigators such as Palmer (1981), McGrattan et al. (2000), Raj (2005a), and Tieszen (2007), a large LNG pool fire develops the following characteristics as illustrated in Figures 9.3 and 9.4: •

•

• •

•

A pool fire naturally develops a swirl or “vorticity” as illustrated in Figure 9.3. This is because there is an inherent likelihood that air entrained into the fire cannot flow with perfect symmetry, but rather with some offcenter preferential flow. Near the fire base, the diameter of the fire typically contracts with height, reaching a “pinch point” then expanding (Fig. 9.4). There is a slight downdraft outside the fire and below the pinch point. Turbulence within a fire plume is considerably higher than outside the plume. Turbulence increases as the firewind velocity (induced air entrainment) increases. The base of the fire is usually bright, with smoke developing higher up.

226

LNG POOL FIRE MODELING

Figure 9.3 Fire swirl developed by induced air entrainment, 50-m diameter crude oil fire (Palmer, 1981) (reproduced by permission from Elsevier Science Publishing, Inc.).

Top boundary of the visible plume Mean visible height of the fire plume Visible flame sheets in a real fire

Air entrainment U(z) m(z) °

LF

Entrainment of ambient air Air mass entrainment rate = ma Mathematical representation of a real fire by a vertical cylinder

dz °v m D

Evaporating flammable liquid pool

Figure 9.4 Defined zones for a large LNG pool fire (Raj, 2007) (reproduced by permission from Elsevier Science Publishing, Inc.).

POOL FIRE CHARACTERISTICS

•

•

•

•

•

227

Intermittent black smoke begins to partially obscure the hot interior flame. The radiant energy emitted decreases with the amount of smoke “shrouding.” In the upper regions, a pool fire acts somewhat like a lighthouse. A bright spot occurs at a roughly constant frequency. This frequency starts low near the base and increases going up the plume, so it is high toward the top. The entrainment of air to a fire plume is far in excess of stoichiometric burning requirements. Oxygen concentrations inside the plume drop only to a minimum of around 40% of the ambient. Consistent with excess air entrainment, temperatures inside the plume typically reach a maximum of over 1200°C. This is illustrated in Figure 9.5 for a 50-m diameter JP4 fire where the temperatures reach about 1250°C but with an average closer to 1000°C.

With these characteristics in mind, some structures can be found in a large fire by defining three zones:

1500 1400 1300 1200 1100 Temperature (°c)

1000 900 TE-12

800 700 600 500 400 300 200 100

TE-13

0 0

TE-14

60 120 180 240 300 360 420 480 540 600 660 720

Figure 9.5 Temperatures inside a fire plume for a 50-m pool fire of JP4 (Japan Institute of Safety Engineering, 1982).

228

LNG POOL FIRE MODELING

1. Zone 1, at the base of the fire, is termed the “luminous region.” Entrained air is strongly induced by a buoyant uplift and combustion is very efficient. The flame sheet, visible in this region, is the outer layer of burning vapors. This part of the fire radiates at a high temperature and the flame is practically “optically thick” or nearly an ideal blackbody radiator as discussed later. Zone 1 extends to a height, LC. Based on measurements for fires of gasoline, heptane, kerosene, and crude oil, McGrattan et al. (2000) conclude that the height of the luminous region is constant for fires larger than 20-m diameters, and this constant depends only on the combustion heat release per unit area (radiant flux). 2. In Zone 2, the flame sheets are anchored to the base, but less efficient combustion occurs because of a deficiency of oxygen in the central core. 3. In the top region, Zone 3, burning is in clumps, and substantial to complete shrouding of the interior burning region occurs. Visible bright spots appear.

9.3.3

Simplifying Pool Fire Structure

A common idealization of a pool fire plume is the tilted cylinder model illustrated in Figure 9.6. The diameter of the plume is taken equal to the diameter of the pool, D. The length of the plume, L, is the visible length, where the last

Wind direction

Z

q L

E

t q''

D S

Equivalent cylindrical fire

Receiving object Atmospheric absorption of radiant heat

Figure 9.6 Illustration of the idealization of a pool fire as a tilted cylinder (Raj, 2005a) (reproduced by permission from the American Institute of Chemical Engineering (AIChE)).

POOL FIRE CHARACTERISTICS

229

bright spots are seen at an appreciable frequency. The visible plume length fluctuates, so it is generally defined as a mean value. The tilt angle bent by the wind, θ, is the angle in degrees to the vertical. The radiant energy flux from the fire, E, is measured at spots on the plume by narrow-angle radiometers (NARs) to the provide local value of E, also called the surface emissive power (SEP), in units of W/m2. The local SEP varies strongly as smoke obscures and then bright spots appear. By using wide-angle radiometers (WAR), a broader area of the plume is averaged to give a surface energy flux representative of the entire plume or of fire zones discussed above. This is called the mean surface emissive power (MSEP), or simply SEP. The energy flux that reaches a target, Qrad, in W/m2, is a small fraction of the radiant flux at the surface of the fire, E. Some radiant energy is absorbed at certain radiation wave length bands by gases and droplets in the atmosphere. The fraction not absorbed is transmitted and is calculated from the transmissivity of the atmosphere, with the symbol τatm. The transmissivity is a function of the path distance, S, between the fire surface and a target object. It also depends on the relative humidity (RH) or the concentration of absorbing constituents in the atmosphere. The energy reaching a target depends also on the “view factor” or a fraction between the surface area of a target object presented toward the fire, Aobj, and the total area of the fire surface, Afire. The symbol for the view factor is a rather complex one attempting to explain that these two areas are involved: FAfire → Aobj. The view factor is simplest for a thermal source that can be taken as a point and is the reciprocal of the spherical area at that distance. Qrad = EFAfire → Aobj τatm ( s, RH ) in W m 2 at a target.

(9.3)

The formula is generalized by recognizing that the emissive power varies with height z, up the plume, represented as E(z). The view factor also varies with incremental surface areas as we move up the plume dA(z), represented as FdA(z)→ Aobj. Finally, the distance through the atmosphere to the object also varies for each fire surface area, so transmissivity is represented as τatm(s(z), RH). Thus, the generalized formula for radiant energy reaching an object adds all of the contributions from incremental areas along the fire plume: N

Qrad = ∑ E ( zi ) FdAi → Aobj τatm ( s ( zi ) , RH ) in W m 2.

(9.4)

i =1

Presently, there are one-, two-, and three-zone radiation models (N = 1, 2, or 3) as well as full CFD representations that solve the full Navier–Stokes flow equations and link these to fire equations. These concepts are explained further as we describe here experimental measurements and correlations to predict the above values. The details for each parameter are presented in later sections of this chapter.

230

9.4

LNG POOL FIRE MODELING

SUMMARY OF LNG FIRE EXPERIMENTS

A number of LNG fire experiments have been conducted as reviewed by Raj (2005a, 2007), Luketa-Hanlin (2006), and Moorhouse and Pritchard (1982). Additional information on LNG fire models is available in the TNO Green, Red, and Yellow books (TNO, 1977, 1992, 1997a, 1997b, 2005) and in Lees’ Loss Prevention Handbook (1996; Mannan, 2005). Table 9.1 summarizes LNG pool fire experiments with some test conditions and results. Nearly all of these tests have been on land or insulating concrete in dike diameters from 1.8 to 6.0 m in the period 1962–1974. By 1982, tests were made in insulating concrete dikes of 20-m diameter, and some tests on water were unconfined. The largest LNG pool fire test to date has been in a 35-m-diameter insulated concrete dike, the Montoir tests (Nedelka et al., 1979, 1989). Larger pool fires were conducted by the Japan Institute of Safety Engineering (1979, 1982) with 15-, 30-, 50-, and 80-m-diameter pools of JP4 (jet fuel). Larger accidental pool fires have occurred at refineries, but these are not instrumented and do not generate useful experimental data. 9.5

BURNING RATE DATA AND CORRELATIONS FROM FIRE TESTS

The burning rate from fire tests can be expressed as either the burning flux, Gb, in kg/(m2s), or as the burning velocity, also called the regression rate, ub, in m/s. In physical terms, this is the downward velocity of a burning surface (e.g., LNG burning in a tank). The burning flux is the product of the regression rate and the liquid density, ρL, in kg/m3: Gb = ubρL.

(9.5)

Regression rate data from hydrocarbon pool fires on land are summarized by Mudan et al. (1984), Mudan (1989), and Mudan and Croce (1995) in Table 9.2 for fires both on land and on water. For comparison, the regression rate for fresh gasoline and diesel varies from 0.3 × 10−4 to 1.0 × 10−4 m/s according to Chatris et al. (2001) cited by Lehr and Simecek-Beatty (2004). The burning flux of LNG on insulating concrete was found in the Montoir 35-m-diameter tests (Nedelka et al., 1989) to be 0.14 kg/(m2s) or a 3.1 × 10−4 m/s regression rate. Insulating concrete minimized the contribution of heat conduction from the substrate and was used to force the main contribution to be heat from the fire. However, as discussed by Raj (2005b), the measured regression rate is lower by about a factor of 2.5 from the value calculated using measurements on heat flux from the fire, highlighting the difficulties in the interpretation of large-scale fire trials. Regression rate values from Table 9.2 are plotted in Figure 9.7 against the carbon number of the components. Only hydrocarbons from methane to butane have normal boiling points below typical ambient water temperatures, so only these have a larger regression rate on water than on land.

Location

Lake Charles, LA

—

Libya

San Clemente, CA

China Lake, CA

Japan

United Kingdom

Year

1962

1962

1969

1973

1974– 1976

1976

1980

No. of Tests

92

100 160 ± 17 185–224 (NAR)a 220 ± 47 (WAR)b

58 n/a

1.6 × 10−4

1.5 × 10−4 2.2 × 10−4 3.4 × 10−4 to 9.6 × 10−4 —

6.1-m D 8.5- to 15.0-m D 25- to 55-m L Steady state 2 m × 2 m square n/a Square and rectangular (2.5 : 1.0) dikes. Eq D 6.9–15.4 m

7 8 5

3

6

—

—

WAR Mean Emissive Power (kW/m2)

70 m L × 25 m widest × 5 m D (avg) Eq D = 18 m 1.8-m D

—

Burn Regression Rate (m/s)

—

—

Flame Length, L; Pool Diameter, D

—

—

Dike on ground 29

Continuous spill on unconfined water Dike on ground

Dike on ground

Continuous LNG feed to trench

—

Dike on ground n/a

Type of Tests

Table 9.1 Summary of LNG fire experiments

U.S. Bureau of Mines Burgess; Zabetakis (1965) Conch Methane Services Ltd. (1962) Esso; May and McQueen (1973)

Sponsor, Reference

AGA (1974); Welker (1979) 25 Raj and Atallah (1974) 12–32; depends USCG; Raj et al. on spill rate (1979) and Lind and Whitson (1977, 1979) 13 Japan Gas Association (1976) n/a British Gas; Moorhouse (1982)

20

12–16

—

—

Percent of Comb. Energy Radiated

Thornton Research Center Maplin Sands, England Japan

1980

Montoir, France

Sante Fe, NM, USA

1987

2009

Plastic lined dike with water

2

—

35-m D 77-m L

290–320 (NAR) 257–273c 165 ± 10d (WAR) —

3.1 × 10−4

—

n/a

n/a

6 Spill on pond; ignition of vapor cloud on land Insulated 3 concrete dike

Percent of Comb. Energy Radiated

—

n/a

n/a

n/a 153 ± 16 (average), 219 (maximum) 178–248 n/a 203 ± 35 (average) n/a n/a

WAR Mean Emissive Power (kW/m2)

n/a

30-m D (effective) n/a 80-m L 2.5 × 2.5 m square n/a

2.37 × 10−4

Burn Regression Rate (m/s)

Unconfined on 3 water Dike on ground 8

Flame Length, L; Pool Diameter, D 20-m D 43-m L

No. of Tests

Insulated 1 concrete dike

Type of Tests

b

Based on the average of narrow-angle radiometer (NAR) data, corrected for atmospheric transmissivity, from seven tests. Based on the data from a single working wide-angle radiometer (WAR) in one test (#12). c Surface emissive power based on actual radiating surfaces from the fire as measured from video films. d Mean surface emissive power based on the assumption of flame surface being a tilted cylinder of length by Thomas’ correlation. n/a, not applicable.

a

China Lake “Coyote Series”

1984

1981

Location

Continued

Year

Table 9.1

Sandia Laboratories (2009)

Gaz de France; Nedelka et al. (1989, 1990)

Shell Research; Mizner and Eyre (1982) Mizner and Eyre (1983) Tokyo Gas; Kataoka (1982) USCG; Rodean et al. (1984)

Sponsor, Reference

BURNING RATE DATA AND CORRELATIONS FROM FIRE TESTS

Table 9.2

233

Burn regression rate of hydrocarbon pool fires (Mudan, 1989)

Material

Burning Velocity (m/s × 104)

Boiling Point (K)

On Land Hydrogen Methane Ethane Propane Iso-Butane n-Butane Iso-Pentane n-Pentane Iso-Hexane n-Hexane n-Heptane n-Octane n-Nonane Ethylene Propylene Butylene Cyclopentane Cyclohexane Benzene Toluene o-Xylene

20.3 111.7 164.6 231.1 261.4 272.7 301.1 309.7 333.5 341.9 371.9 398.9 424.0 169.5 225.0 266.9 322.5 353.9 353.3 383.8 417.6

On Water

23.0 2.08 1.22 1.37 1.55 1.32 1.23 1.43 1.37 1.22 1.13 1.05 0.97 1.23 1.33 1.47 1.32 1.15 1.00 0.95 0.97

45.0 5.84 2.72 2.63 2.11 1.62 1.23 1.43 1.37 1.22 1.13 1.05 0.87 2.93 2.60 1.92 1.32 1.15 1.00 0.95 0.97

Burn velocity (mm/s)

1 0.8 0.6 0.4 0.2 0 1

2

3

4

5

6

7

8

9

Number of C in chain

Land

Water

China Lake

Figure 9.7 Burn regression rate for hydrocarbons on land and on water from Table 9.2.

Figure 9.8 shows that the measured regression rate correlates well with the heat of combustion, ΔHC (in J/kg), divided by a modified heat of vaporization, ΔHV*. The modified heat of vaporization represents the latent heat to warm the liquid to the boiling point plus the heat of evaporation. Specifically, this is the heat of vaporization, ΔHV, and the average heat capacity of the liquid,

LNG POOL FIRE MODELING

Burning rate, y × 104 (m/s)

234

2.0

LPG

LNG

1.5

LEG Butane Hexane

1

Xylene 1.27 × 10

–6

0.5 Methanol

ΔHc ΔHv*

UDMH

JP4

Benzene Gasoline

Acetone DETA

20

40

60 ΔHc /Hv*

80

100

Figure 9.8 Burn regression rate for various liquids on land (Mudan, 1989) (reproduced by permission of the Society of Fire Protection Engineers).

¯ pL, evaluated over the temperature range between the cool spilled liquid, Ta, C and the boiling point, Tb, times this temperature difference: ΔH v* = ΔH v + C pL(Tb ) max {0,Tb − Ta } .

(9.6)

Equation 9.6 represents the concept that some fraction of the heat from the fire (the heat of combustion) radiates to the liquid in the pool and warms the surface to the boiling point and then evaporates some liquid. From Figure 9.8, the slope of the line is 0.5/40 × 10−4 (in m/s) so the correlation for regression rate or burning velocity is ub = 0.0125 × 10 −4

ΔH C ΔH v*

.

(9.7)

The LNG and liquefied petroleum gas (LPG) points are above the curve in Figure 9.8. Since both have significantly lower liquid density compared with the other materials in the correlation, multiplying the burning velocity by the liquid density to obtain burning flux brings the LNG and LPG points more in line with the other hydrocarbons as seen in Figure 9.9.

BURNING RATE DATA AND CORRELATIONS FROM FIRE TESTS

Mass burning rate (kg/m2s)

0.12

235

LPG

0.10 Benzene

LNG

0.08 Xylene JP4 UDMH

0.06 ΔHc 10–3 ΔHv*

LEG Butane Hexane

Gasoline

0.04 Acetone Methanol

DETA

0.02

20

40

60 ΔHc /Hv*

80

100

Figure 9.9 Burning flux for various liquids on land (Mudan, 1989) (reproduced by permission of the Society of Fire Protection Engineers).

The equation that represents Figure 9.9 is Gb = 10 −3

ΔH c ΔH v*

in kg ( m 2 s ) .

(9.8)

Mudan (1989) attributes to Grumer et al. (1961) a generalization of Equation 9.8 to multicomponent mixtures with N components of mass fraction xi each, and corresponding values of ΔHCi, ΔHVi, and CPLi: N

Gb = 1.27 × 10 −6

ρLiq ∑ xi ΔHCi i =1

N

N

i =1

i =1

∑ xi ΔHVi + (Tb − Ta ) ∑ xi ∫T C pi(T ) Tb

.

(9.9)

a

Only a few experiments have been documented of burning LNG on water. Mizner and Eyre (1983) spilled LNG at Maplin Sands in the United Kingdom and measured fire radiation but did not measure the evaporation rate. Experiments at China Lake, CA, reported by Raj et al. (1979) found values listed in Table 9.3. Liquid LNG density taken as 449 kg/m3 to translate between Gb and ub.

236

LNG POOL FIRE MODELING

Table 9.3 Burning and evaporation flux for LNG on Water from China Lake tests (Raj et al., 1979)

Experiment 3 5 6 12 1 7 4

Burning, Gb (kg/[m2s])

Burning Velocity, ub (m/s)

0.431 0.401 0.361 0.221 0.199 0.154 0.152

9.60E-04 8.93E-04 8.04E-04 4.92E-04 4.43E-04 3.43E-04 3.39E-04

9.5.1 Consistency Checks between Evaporation Rate and Burning Rate The burn rate was found to increase with the LNG spill rate for the China Lake tests. The spill conditions allowed considerable penetration of the water by LNG that is known to increase the LNG evaporation rate (hence also the burn rate). The reported correlation for ub (in m/s) with the volumetric discharge rate, Vdis, (in m3/s) is ub = ( 2.73 + 58Vdis ) × 10 −4

(9.10)

This correlation would not apply beyond the spill rates and pool diameters used in the China Lake tests, so it should not be extended to large-scale tests or accidental spills. Until more definitive tests are available, it is necessary, for quality assurance purposes, to resort to comparisons to find consistency with burning tests of various materials, between burning tests on water and on land and between nonburning and burning tests. Table 9.4 calculates the ratio of burning to nonburning flux rates for LNG on water from observations. Burning flux values are from Table 9.3 and the burning velocity for methane on water from Table 9.2 converted to burning flux using a value for the liquid density of 449 kg/m3. Nonburning evaporation flux rates are taken from Chapter 7, Table 7.2. The ratio of burning to nonburning flux on the same substrate should be greater than unity as the flame radiates additional heat back into the pool. Values of this ratio in Table 9.4 largely satisfy this expectation. The ratios that exceed unity range up to 2.57. This illustrates that the remaining uncertainty in knowing the burning flux of LNG on water is about a factor of 2.57. 9.5.2

Stopping Point for Pool Fire

For nearly all materials below a minimum pool thickness of less than 1 mm, a pool fire stops burning (Fingas, 1998). With heavy fuels such as fuel oil or crude oil, a residue is left when the fire burns out, and this residue can sink

POINT SOURCE FIRE MODEL

237

Table 9.4 Ratio of burning to nonburning LNG tests on water

Reference

Burning Flux, Gb (kg/[m2s])

Nonburning Flux, Gevap (kg/[m2s]) Colenbrander and Puttock (1983)

Reid and Smith (1978a)

Chang and Reid (1983)

Paranouskis et al. (1980)

0.168

0.181

0.20

0.25

Ratio Burning/Nonburning Flux China Lakea Test 1 Test 12 Test 6 Test 5 Test 3 Table 9.2b

0.200

1.19

1.10

1.00

0.80

0.222 0.362 0.403 0.432 0.262

1.32 2.15 2.40 2.57 1.56

1.23 2.00 2.23 2.39 1.45

1.11 1.81 2.01 2.16 1.31

0.89 1.45 1.61 1.73 1.05

a

Raj et al., (1979). Mudan (1989).

b

or can form globs. LNG is volatile enough to completely evaporate after the fire burns out, even though a layer of about 1 mm is believed to remain at this point (Lehr and Simecek-Beatty, 2004). 9.6

POINT SOURCE FIRE MODEL

A point source fire model is designed as a limiting case, far-field model. As such, it is only valid in the far field and is considered too approximate for most situations of interest. However, it has a long historical use in NFPA 59A. There are three ways to estimate the energy radiated from a fire: • •

•

Assume it is a given fraction of the heat released by combustion, χR. Assume it is a given value for the heat emitted from the flame surface (SEP, correlated from radiometer measurements). Estimate it from flame temperature, emissivity, and transmissivity within the flame.

The point source fire model equations below illustrate the first approach. The sections following the present section discuss solid surface models and illustrate the second and third approach. This model assumes the fire is a point source of radiation, and the emitted heat flux in any radial direction, E(s), is the combustion energy emitted divided by the area of a hemisphere of radius s, multiplied by the fraction χR, as illustrated in Figure 9.10. Since the area of a hemisphere increases by the square of s, the above assumption is called the inverse square law, as is represented by

238

LNG POOL FIRE MODELING

Imaginary hemisphere with uniform radiation flux

Point source in LNG fire

Q

Radiant heat-absorbing object

S Perfect absorber ground

Figure 9.10 Schematic representation of point source fire model.

E ( s) = χ R

ApoolGb ΔHC , 4 πs 2

(9.11)

where E(s) = radiant energy flux from the point source (kW/m2); χR = fraction of energy of combustion emitted as radiant energy; Apool = area of pool (0.25 πD2), or it can be generalized to any other type of surface (m2); Gb = burn flux (kg/[m2s]); ΔHC = lower heat of combustion (J/kg) (i.e., water vapor produced remains as vapor); s = radial distance from source (m); and τatm = transmissivity of the atmosphere as a function of path distance and RH. The radiant flux received on an object, Qrad, is Equation 9.12 (in kW/m2): Qrad ( s ) = E ( s ) FAfire → Aobj τatm ( s, RH ) = E ( s ) τatm ( s, RH ) ,

(9.12)

that is, the same as Equation 9.3, with a view factor of unity. In effect, the ratio of the area of the object that receives radiation to the area of the hemisphere is the view factor for a point source model, so it does not appear explicitly. Implicit in Equations 9.11 and 9.12 are the following assumptions: 1. The ground is a perfect absorber of heat, so there is no reflection of heat at ground level. 2. The observer is oriented normal to the point source. 3. The observer is far from the fire (i.e., s/D >> 1, at least 5 or larger) 4. The fraction of combustion energy released as radiant heat is independent of the size of the fire. Based on measurements with a limited number of LNG fires of diameters less than 15 m, the fraction of combustion energy radiated (χR) as listed in Table 9.1 varies as 0.12 ≤ χR ≤ 0.32.

SOLID FLAME MODELS: FLAME LENGTH CORRELATIONS

239

Lees (Mannan, 2005) gives the range for natural gas fires to be 0.19 ≤ χR ≤ 0.23. It would be expected that the fraction of energy radiated for larger fires decreases because of greater smoke production and obscuration. McGrattan et al. (2000) provide a correlation for the fraction radiated from heavier hydrocarbon fuel fires (other than LNG) as decreasing with pool diameter D by χ R = 0.35e

D − ⎛⎜ ⎞⎟ ⎝ 20 ⎠

D ≤ 25 m.

(9.13)

However, this correlation predicts that the 35-m-diameter Montoir fire would radiate only 6% of the combustion energy. This is not supported by radiometer measurements. Equation 9.13 drops below 10% for pool diameters larger than 25 m, so this is taken as a likely range of applicability. Altogether, the use of the radiated fraction χR in fire models is a weak point, and correlations have not been adequately established for this parameter. The point source model suffers from this limitation. However, it provides reasonably good estimates of human exposure hazard distance provided the following conditions apply (Raj, 2007): • • •

9.7

The fraction of radiated heat is known a priori. The radiation level of interest is low (on the order of 1–5 kW/m2). The fire size is small (D < 5 m).

SOLID FLAME MODELS: FLAME LENGTH CORRELATIONS

An important problem for LNG risk analysis is to determine how high flames would reach from the largest postulated pool fires of, say, 300- to 500-m diameters. Theoretical and experimental investigations with small-scale flames predict flame length-to-source diameter ratios, L/D of 3–10. Models based on medium-scale experiments for source diameters around 35–70 m predict L/D ratios around 1–3. Large-scale tests are planned but are not yet conducted in the range of 100-m diameter. Arguments are reviewed here that a change of burning mechanism is likely with large-scale tests. The L/D ratio for largediameter fires may well fall below unity. A discussion of mass fires that have this characteristic is presented later. The term solid flame model refers to ignoring the complex and ragged shape of turbulent flames and assuming radiant energy is emitted from a smooth, idealized fire surface. The shape of a pool fire is idealized as a tilted cylinder or some variation thereof (elliptical cross section, conical top, etc.). Flame length is defined as the longest visible flame tip at which flame elements are seen with a significant frequency (Becker and Liang, 1978). This is inexactly defined, but different observers generally agree on it within limits of ±5% to ±15%. The flame length is found to coincide very nearly with the point

240

LNG POOL FIRE MODELING

when the centerline fuel mass fraction, wC, equals the stoichiometric mass fraction, wST. There is also a thermal measure of flame length, LT, at the radial ¯ , which occurs at (∂T ¯ /∂r) = 0. In reality, the maximum of mean temperature T visible length of turbulent flames is 10–15% longer than the stoichiometric concentration limit because of turbulent eddy billowing and undulation, producing flame flicker. 9.7.1

Small-Scale Pool Fire Tests and Flame Length Correlations

Flame length correlation work began with measurements of small-scale flames, with source diameters on the order of 1–10 mm. With such small flames, it is quite evident that for non-premixed flames, the combustion process is rate determined by diffusion and is locally complete to the extent that oxygen is available. This implies a locally sharp flame front in which combustion reactions are concentrated in a narrow front. For this situation modelers introduced the theory of turbulent jets where the supply of oxygen is determined by the entrainment of ambient air. Observations of small jet flames revealed the pattern shown in Figure 9.11 (Becker and Liang, 1978). As the fuel pressure and thus fuel flow rate increase, initially, the flame is smooth and laminar and increases in length up to point A, which marks the end of the laminar flow regime. Point B is the effective starting point of turbulent flow. Increasing pressure and flow rate produces turbulent flames. At one extreme, forced convection jet flame length becomes very short, as in burners in furnaces. Of interest here is the other extreme, natural convection flame length. Over the entire range between forced and natural convection flames, the governing characteristic parameters are primarily the Reynolds number and the Richardson number. The jet Reynolds number represents the ratio of viscous to inertia forces. The Richardson number represents the ratio of buoyancy forces to inertial forces. It turns out that buoyancy forces dominate for natural convection.

Flame length

A

B

Fuel rate Figure 9.11 Relation between flame length and fuel flow rate for small jets (plotted from Becker and Liang, 1978).

SOLID FLAME MODELS: FLAME LENGTH CORRELATIONS

241

The Richardson number, Ri, is defined in terms of the gravitational acceleration, g (in m/s2), the density of combustion products cooled to ambient temperature; ρ∞, the flame length; L, the source diameter; D0, the initial mass flux of fuel; G0 (in kg/[m2s]) (or correspondingly the burning rate from a pool); and the initial velocity of fuel, u0 (in m/s): Ri =

gρ∞ L3 . D02G0 u0

(9.14)

To interpret the Richardson number as the ratio of buoyancy forces to inertial forces, begin with the definition of buoyancy as the volume of the plume, V, times the density difference between the fire plume, ρplume, and ambient air, ρa, times gravitational acceleration: Buoyancy = gV (ρa − ρplume ) ≅ gLDj2ρa. The density of combustion products is usually about one-eighth the density of ambient air and can be neglected. The volume of the fire plume is estimated using L to obtain a simpler definition for the Richardson number. The jet Reynolds number is defined in terms of the jet diameter, Dj, the centerline jet speed, uj, the jet density, ρj, and the dynamic viscosity of the jet gases, μj: Re j =

Dj u j ρ j . μj

(9.15)

The similarity between unburning jet dilution by air entrainment and burning jet fires allows the use of a Reynolds number evaluated at the flame tip, ReL.. When the Reynolds number at the flame tip is sufficiently large, the turbulence is sufficiently developed so the flame length becomes effectively independent of the Reynolds number. To show this, Becker and Liang (1978) plot experimental results from small pool fires of liquid hydrocarbons by Blinov and Khudiakov (1959) and also Hottell (1958) in Figures 9.12 and 9.13. These data are plotted as Y=

ψ , ξL

(9.16)

where ψ is the scaled fire pool diameter, D0, essentially to flame length, L: ⎡ β D0 ⎤ ψ=⎢ ⎣ wST L ⎥⎦

23

(9.17)

and ζL is the cube root of the Richardson number, a dimensionless scaled flame length:

242

LNG POOL FIRE MODELING

1.0

b

Y

a

c

d

0.1

0.01 1.E + 02

1.E + 03

1.E + 04

1.E + 05

1.E + 06

ReL Heavy oil Diesel fuel c, all pulsating

Gasoline a, conical d, fully turbulent

Kerosene b, tip pulse

Figure 9.12 Results of pan fire tests with liquid hydrocarbon (Blinov and Khudiakov, 1959) for gasoline (䊊), kerosene (䊐), diesel fuel (䉱), and solar oil (䉫) (data from Becker and Liang, 1978).

Y

1.0

0.1

0.01 0.1

1

10

100

1000

10,000

Do (cm) Heavy oil

Gasoline

Diesel fuel

Fit

Kerosene

Figure 9.13 Results of pan fire tests with liquid hydrocarbon (Blinov and Khudiakov, 1959) for gasoline (䊊), kerosene (ⵧ), diesel fuel (䉱) and solar oil (䉫) (data from Becker and Liang, 1978).

13

⎛ gρ ⎞ ξ L = ⎜ 2 ∞ ⎟ L. ⎝ D0 G0 u0 ⎠

(9.18)

The scaling length in ψ is the ratio β/wST, where wST is stoichiometric mass fraction of fuel and β2 is the density ratio of the expanded combustion products at T∞ and M∞ to the hot combustion products at T1 and M1: 12

⎛M T ⎞ β=⎜ ∞ 1⎟ , ⎝ M1T∞ ⎠

(9.19)

SOLID FLAME MODELS: FLAME LENGTH CORRELATIONS

243

where T1 and M1 are the adiabatic flame temperature and mole weight of combustion products at the stoichiometric concentration, and T∞ and M∞ are the temperature and mole weight after infinite dilution by air (ambient air properties). The left-hand side of the plots represents the laminar flow region where the D/L ratio decreases as L increases. A change of mechanism to turbulent flow occurs at higher Reynolds numbers as instabilities set in. Flame behavior is indicated by the letter notation at the top of Figure 9.12 as follows: a) b) c) d)

conical, steady flame pulsating tip pulsation over entire flame fully turbulent

Above a Reynolds number of about 2 × 105, the flame is fully turbulent and the term ψ/ζL reaches a constant value between 0.2 and 0.3. This is the region for natural convection. Exploring further, the data for the high Reynolds number region are reformulated by Becker and Liang into Figure 9.14, which can be fairly well fitted by ψ D = 0.48 ⎛ 0 ⎞ . ⎝ L⎠ ξL 32

(9.20)

Y

1.0

0.1

0.01 1

10

ReL Heavy oil

Gasoline

Kerosene

Diesel fuel

Figure 9.14 Data on pan fires of liquid hydrocarbons at a high Reynolds number (data from Becker and Liang, 1978).

244

LNG POOL FIRE MODELING

Unfortunately, Equation 9.21 essentially represents a highly redundant cor13 relation between ( D02 L5 ) with (D0/L)3/2. Becker and Liang admit the result may be meaningless. Along an alternate path, though, Becker and Liang invoke another jet modeling concept, the universal structure model, which provides the following correlation: ⎡ β 2G02 A02 ⎤ L = Cn ⎢ 2 ⎥ ⎣ gρmρ∞ wST ⎦

15

(9.21)

or, substituting a circular source area and factoring the source diameter, D0, 15

⎡ β 2G02 π ⎤ L = Cn ⎢ D02 5, 2 ⎥ ⎣ 4 gρmρ∞ wST ⎦

(9.22)

where m0 is the mass rate of fuel (or the burn rate in kg/s), and ρm is the minimum value at the jet centerline (at the source for a light fuel, or at the tip for a heavy vapor). The value of the coefficient Cn is 3.0 using the data of Steward (1970) and Vienneau (1964) or 6.6 from the test data of Blinov and Khudiakov (1959). The physical properties needed for the above correlations of flame length are listed in Table 9.5. The stoichiometric concentrations are listed in mole and mass fractions, along with the adiabatic flame temperature, T1, the mole weight of combustion products, M1, and the density ratio of cool to hot combustion products, β. Equation 9.22 can be evaluated with the values in Table 9.5 for methane; using burning rates Gb values for the mass flux, G0, in Equation 9.21 gives the following predictions for L/D0.8: Go (kg/[m2s])

L/D0.8

0.0934 (on land) 0.262 (on water) Table 9.5

Fuel

Hydrogen Methane Ethane Ethylene Acetylene Propane Butane

Cn = 3.0

Cn = 6.6

3.0 4.6

6.7 10

Flame properties needed in correlations for flame length

Mole Weight (kg/kmole)

yST Mole Fraction

wST Mass Fraction

T1 (K)

M1 (kg/ kmole)

β

2.016 16.04 30.07 28.05 26.04 44.09 58.12

0.295 0.0948 0.0565 0.0653 0.0772 0.0402 0.0312

0.0283 0.0548 0.0585 0.0634 0.0699 0.0599 0.0607

2415 2223 2257 2367 2537 2265 2275

24.53 27.51 27.96 28.47 29.08 28.14 28.42

3.10 2.80 2.80 2.84 2.91 2.80 2.79

SOLID FLAME MODELS: FLAME LENGTH CORRELATIONS

9.7.2

245

Medium-Scale Pool Fire Tests and Flame Length Correlations

An average visible flame height for pool fires, L, has been found to be proportional to the pool diameter, D, in terms of the correlation parameters, A, p, and q found from test data and two dimensionless variables, FC and u*: L q = AFCp( u*) . D

(9.23)

Here, FC is the combustion Froude number, a dimensionless burn flux in terms of the burn flux, Gb, the atmospheric density, ρa, the gravitational acceleration, g, and the pool diameter: FC =

Gb , ρa gD

(9.24)

where the dimensionless wind speed, u*, is defined in terms of the wind speed at the standard reference height of 10 m, uw, and a characteristic wind speed, uC: u* =

uw uC

(9.25)

and 13

⎡ G gD ⎤ uC = ⎢ b ⎥ . ⎣ ρa ⎦

(9.26)

This correlation was first proposed by Thomas (1963, 1965) based on wood crib fires. Moorehouse (1982) conducted several moderate-scale LNG pool fire tests and adjusted the coefficients A, p, and q listed in Table 9.6 to give the predictions plotted in Figure 9.15. The same two-thirds value for p is used by Murgai (1976) from the analysis of forest fire data. There is a theoretical basis for p = 2/3 as shown by Raj (2007). Numerous models use the parameter values of Thomas (A = 42, p = 0.61, q = 0) including the LNGFIRE3 model required by U.S. regulations (see Chapter 6). With this model, since q = 0, the plume length is taken as independent of the wind speed, but the tilt angle depends on u*. Figure 9.15 provides a comparison of the above model predictions with test data for a number of LNG experiments (Raj, 2007). The combustion Froude number, FC, is defined with the pool diameter in the denominator, so the test data, based largely on small-diameter tests, are toward the right-hand side of Figure 9.15 and the large pool diameter predictions are toward the left. For example, with the two largest pool diameters (330 and 512 m) cited in the Sandia Report (Hightower et al., 2004), with the burn flux of 0.135 kg/(m2s)

246

LNG POOL FIRE MODELING

Flame length/fire diameter (L/D)

10.0 Moorhouse (1982)

L = 6.2 F 0.254 D Fire base area (m2) LEGEND 1767 China Lake tests, Raj et al. 1000 Montoir tests, Nedelka et al. 185.8 148.6 111.5 92.9 Moorehouse et al. 74.3 55.7 37.2

1.0 Thomas (1965)

2 L = 55 F /3 D

Cox and Chitty (1985) L = 430.4 F D

0.1 0.001

0.01 F=

· m** r * gD

0.1

= Dimensionless burning rate or combustion Froude number

a

Figure 9.15 2007). Table 9.6

Comparison of predictions of pool fire visible length-to-diameter ratio (Raj

Model coefficients for flame length-to-diameter ratio

Reference

A

p

Thomas (1963)

42

0.4

0

F > 10−1

Thomas (1963)

42

0.61

0

10−2 < F < 10−1

Thomas (1965) Moorhouse (1982) Heskestad (1983) Cox and Chitty (1985)

55 6.2

2/3 0.254

28

2/5

0

10−3 < F < 1

430.4

1

0

10−4 < F < 10−3

430.4

2

0

10−5 < F < 10−4

Cox and Chitty (1985)

q

−0.21 −0.044

Range of applicability

F < 10−3

Remarks No wind; laboratory-size wood crib fire Same wood crib tests with wind For u* > 1 Curve fit for one LNG fire test series Provides values of p that vary with F Based on laboratory-scale diffusion fires of very low burning rates Based on laboratory-scale diffusion fires of very low burning rates

also in that report, the combustion Froude numbers are 1.98E-3 and 1.59E-3, respectively, just about where the Thomas (1965) and Cox and Chitty (1985) curves cross. Experimental and modeling values are in agreement that for pool diameters less than 35 m, L/D = 3.0 ± 0.5. Extrapolation provides estimates that for pool fire diameters between 35 and 100 m, L/D = 2.3 ± 0.7. Further extrapolation for pool fires larger than 100 m estimate L/D = 1.0 ± 0.5 (TAMU, 2008). If

SOLID FLAME MODELS: FLAME LENGTH CORRELATIONS

247

100 Fire length (m)

80 60 40 20 0 0

10

20

30

40

Pool D (m) Thomas Gb = 0.18 Moorhouse Gb = 0.18 Moorhouse Gb = 0.135 MaplinSands China Lake

Figure 9.16

Thomas Gb = 0.135 Cox and Chitty Gb = 0.135 MaplinSands Montoir

Predictions for pool fire flame length compared to test data.

600 Fire length (m)

500 400 300 200 100 0 0

100

200

300

400

500

Pool D (m) Thomas Gb = 0.18 Possible Cox and Chitty Gb = 0.18

Figure 9.17

Thomas Gb = 0.135 Moorhouse Gb = 0.18

Predictions for pool fire length extrapolated beyond test data.

there is a significant change in the mechanism of burning as pool fire diameter exceeds 100 m, it is possible that L might become bounded and the L/D ratio could fall below the extrapolated value of 1.0. It is inherently dangerous to extrapolate correlations beyond the range of test data upon which they are based. This is illustrated in Figures 9.16 and 9.17. These figures plot the predictions of three models from Table 9.6—the models of Thomas (1965), Moorhouse (1982), and Cox and Chitty (1985).

248

LNG POOL FIRE MODELING

Another important uncertainty is the value of the burn flux. Sandia (Hightower et al., 2004) used a value of Gb = 0.135 and 0.180 kg/(m2s). The lower value of the burn flux was used by Sandia (Hightower et al., 2004), and the higher value is the average nonburning evaporation rate found in Figure 8.4 in Chapter 8. The range of measured burn flux for the test data from Table 9.1 has a wider spread. Figure 9.16 also plots experimental data for fire length and diameter from Table 9.1. The spread of these test data allow all three model predictions with both burn flux values to be acceptable. 9.7.3

Large-Scale Pool Fire Tests and Flame Length Correlations

Figure 9.17 plots the extrapolated predictions for the models and parameter values derived from medium-scale fire tests (plotted also in Figs. 9.15 and 9.16). The extrapolations are made out to the largest LNG pool diameters postulated by Sandia (Hightower et al., 2004) of 330–512 m. The differences in predictions attributed to Gb are about as large as the differences attributed to the various models. The predictions of the Thomas and Moorhouse models agree well to a pool diameter of 100 m. Beyond that, the predicted pool fire lengths are very high and, in fact, may be overpredictions. With a change of burning mechanism, the fire length/diameter could well be under 1/2 as illustrated in Figure 9.17. In addition, the form of the correlation, Equation 9.23, would have to change to accommodate a change of burning mechanism. There may be a physical limit to flame height. If, in fact, the burning mechanisms changes, it is likely that a large-diameter fire breaks into a number of smaller flames to allow air penetration into the interior of the fire. This concept as developed by Blanchat et al. (2007) is called a mass fire as illustrated in Figure 9.18.

Mass fire Coherent fire

Shorter distance to 5 kW/m2

Longer distance to 5 kW/m2

Figure 9.18 Illustration of mass fire concept for change of burning mechanism with large-scale fires (Blanchat et al., 2007).

FLAME TILT CORRELATIONS

249

Table 9.7 Instrumented mass burn forest fires

Date

Heat Flux (kW/m2)

Burn Flux (kg/m2s)

Flame Height (m)

Reference

Hill Township, Ontario, Canada

August 10, 1989

130

0.01 to 0.06a

8–12

Wicksteed Township, Ontario, Canada Hardiman, Ontario, Canada Battersby, Ontario, Canada

August 12, 1989

74.4

0.004b

10

Quintiere (1990); Heikes and Small (1990b) Heikes et al. (1990a)

August 28, 1987 August 12, 1988

5.8

—

—

22

—

—

Fire Test

Mollenkamp and Bradley (1991) ibid.

a Average heat of flaming combustion of 13 ± 2 E6 J/kg; minimum oxygen concentration in flame = 8%, minimum flame temperature = 500°C. b Average heat of flaming combustion of 18.6 E6 J/kg.

The term mass burning was first applied with forest fires. Controlled burns in forests were conducted with instrumentation, as, for example, the tests listed in Table 9.7. Other such tests are reported for the Flambeau series by the U.S. Forest Service (Countryman et al., 1969) and for Operation Eureka at Langley, Central Queensland, Australia (Adams et al., 1972; Small and Hiekes, 1988; Heikes et al., 1990a). According to Heikes et al. (1990b), “The large fire demonstrates a behavior not observed for the small fire, namely an oscillatory flow … A bubble of warm air [lifts] off,.. is swept into the core of the plume, [and] high updraft speeds result. Large fires have been hypothesized to ‘breathe’. That is, when the fire exhausts most available oxygen, it slows until resupplied by the inflow firewinds.”

9.8 FLAME TILT CORRELATIONS The tilt angle of a pool fire is taken from the vertical as illustrated in Figure 9.19. Two alternative correlations are in use to predict the mean angle of tilt of the fire plume axis with respect to the vertical, θ, namely, Equation 9.27 by the AGA (1974) and correlation in Equation 9.19 by Welker and Sliepcevich (1966). The AGA correlation uses u* defined by Equation 9.25. AGA Flame Tilt Correlation ⎧ 1 ⎪ cos θ = ⎨ u* ⎪⎩0

for u* > 1 for u* ≤ 1

(9.27)

250

LNG POOL FIRE MODELING

Wind direction Tilt angle

Figure 9.19 Illustration of LNG pool fire and definition of flame tilt angle (Raj, 2005a) (reproduced by permission of AIChE).

Table 9.8

Coefficient values for pool fire tilt correlation

Researchers Welker and Sliepcevich (1966) Moorhouse (1982)—conical flame Moorhouse (1982)—cylindrical flame Johnson (1992)

a

b

c

d

3.3 3.0 1.9 0.7

0.8 0.422 0.399 0.428

0.07 0.011 0.05 0.109

−0.6 0 0 0

Welker and Sliepcevich Flame Tilt Correlation d

tan θ ⎛ρ ⎞ = aFr bRec ⎜ v ⎟ . ⎝ ρa ⎠ cos θ

(9.28)

Equation 9.28 is in terms of the dimensionless Froude number, Fr, Reynolds number, Re, density ratio of fuel vapor (at the normal boiling point of the fuel) to air, and four parameters, a, b, c, and d. The Froude number is defined in terms of the wind speed at a reference height of 10 m, uw, the gravitational acceleration constant, g, and the pool diameter, D: Fr =

uw2 gD

(9.29)

The Reynolds number of the wind is defined in terms of the pool diameter, wind speed, the density of ambient air, ρa, and the dynamic viscosity of air, μa: Re =

Duwρa μa

(9.30)

The fuel vapor density, ρv, is evaluated at the normal boiling point temperature of the fuel (or the temperature at which the multicomponent fuel has a vapor pressure of 1 atm). Table 9.8 provides values for the correlation of Welker and Sliepcevich obtained by different researchers from fire tests. The Welker and Sliepcevich

FLAME TILT CORRELATIONS

251

values in Table 9.8 are based on laboratory-scale tests (maximum of 0.6-m diameter) with hydrocarbons. Moorhouse used a data set with 6.9- to 15.4-m LNG pool fires on land. Johnson (1992) used 6.1- to 35-m fire tests with other hydrocarbons. Two field tests by Mizner and Eyre (1982) are used to compare observed tilt angles with the Welker and Sliepcvich model (with their coefficients in Table 9.8) and the AGA model. Table 9.9 (from Lautkaski, 1992) compares model predictions for the tilt angle, θ, with observations for the observed burning flux, Gb, wind speed, uw, fuel vapor density (at the boiling point), ρv, and pool diameter, D. The predictions all fall within the range of observations, but the range of test conditions is very limited. Figure 9.20 plots flame tilt angle predictions against pool fire diameter for a constant wind speed of 5 m/s. Three equations are plotted: the AGA formula (Eq. 9.18), the Welker and Sliepcevich formula (Eq. 9.19), and Equation 9.19 with the Moorhouse cylindrical flame coefficients from Table 9.8. The burn flux rates, Gb, used to calculate uC and hence u* are 0.135 and 0.18 kg/(m2s). Clearly, the tilt angle predicted by the Moorhouse and Welker and Sliepcevich equations is not sensitive to pool diameter. The AGA formulas may be too sensitive to pool diameter. Table 9.9

Comparison on predicted flame tilt with observations

Fuel

Gb (kg/ [m2s])

uw (m/s)

ρv (kg/m3)

D (m)

θ by Equation 9.18

θ by Equation 9.19

θ° Observed

0.106 0.13 0.05 0.05

6.15 6.6 8 9

2.0 2.3 3.2 3.2

20 20 52 52

53.5 64.5 60.0 —

51.0 52.0 — 45.5

54 53 52 (43–63) 52 (43–63)

LNG LPG Isohexane Isohexane

Tilt angle (degree)

60 50 40 30 20 10 0 0

20

40

60

80

100

Pool D (m) AGA Gb = 0.18 Welker Gb = 0.135

Figure 9.20

AGA Gb = 0.135 Moorhouse

Welker Gb = 0.18

Fire tilt correlation predictions versus pool diameter.

252

LNG POOL FIRE MODELING

Tilt angle (degree)

70 60 50 40 30 20 10 0 2

4

6

8

10

12

Wind speed (m/s)

Figure 9.21

AGA D = 15 m

AGA D = 100 m

Welker D = 15 m

Welker D = 100 m

Fire tilt correlation predictions versus wind speed, Gb = 0.135 kg/(m2s).

Figure 9.21 plots the flame tilt angle predictions against wind speed with the burn flux, Gb, of 0.135 kg/(m2s) and for two values of pool diameter. The AGA formula predicts that tilt angle is highly sensitive to wind speed. The Welker and Sliepcevich formula predictions are insensitive to wind speed, since uw appears only through the Reynolds number, and this is taken to a low power (0.07). The AGA predictions are preferred with this comparison.

9.9

FLAME DRAG NEAR POOLS

When calculating the impact of pool fires on nearby structures, it is essential to model the effects of wind on the flames. As illustrated in Figure 9.1, the flame tilts and spills over the edge of the pool. That is, the downwind edge of a pool fire on land is dragged by wake effects of the wind around the fire plume. Consequently, the flame approaches objects on the lee side of the pool, increasing the thermal radiation on these objects. If the wind is high enough, the flame may impinge on the object. A flame impinging on an adjacent storage tank heats it much more intensively than non-impinging flames, as convective heat transfer exceeds radiant transfer, and it is not affected by smoke obscuration. In addition, an impinging flame typically coats part of the surface with soot, increasing its absorbance, making it more difficult to cool the tank with water sprays. Moorhouse (1982) provides a correlation for the elongation of the pool fire diameter this drag generates. The elongated dimension of the fire base in the downwind direction is D′. The ratio of D′ to the diameter of the circular fire base, D, is predicted in Equation 9.31 as a function of the Froude number Fr defined by Equation 9.29:

SEP CORRELATIONS AND SMOKE SHIELDING

Table 9.10

Fuel LNG LPG Isohexane

253

Comparison of flame drag predictions with observations

uw (m/s)

ρv (kg/m3)

D′/D, Equation 9.31

D′/D, Equation 9.32

D′/D, Observed

6.15 6.6 9.0

2.0 2.3 3.2

1.34 1.35 1.32

1.34 1.45 1.63

1.25–1.50 1.25–1.55 1.20–1.44

D′ = 1.5Fr 0.069 D 1. ≤

(9.31)

D′ ≤ 1.28 when 0.003 ≤ Fr ≤ 0.1 D

The D′ correction provides no change for very large fires and for small fires, an extension to D of as much as 20%. Theoretically, the flame drag should increase with higher fuel vapor density. A density ratio term of fuel vapor density to air density is added to Equation 9.31 by Welker and Sliepcevich (1966) and by Lautkaski (1992) to give an alternative for consideration: D′ ⎛ρ ⎞ = 1.2 FR0.069 ⎜ v ⎟ ⎝ ρa ⎠ D

0.48

(9.32)

Models were compared by Lautkaski (1992) in Table 9.10 with experimental data of Mizner and Eyre (1982). The wind speed, uw, and fuel vapor density, ρv, are given along with D′/D. The expected effect of fuel vapor is not confirmed by the data. Predictions by Equation 9.31 all fall within the range observed, so this is the preferred correlation. 9.10 SEP CORRELATIONS AND SMOKE SHIELDING SEP, a measure of the radiant flux from fire surfaces, and the reduction in SEP caused by smoke are treated here, first from experiments, then from theory. 9.10.1

SEP from Tests

Important results from the fire tests listed in Table 9.1 are data on the MSEP, or radiant flux from fires in W/m2. In order to define an average SEP, it is necessary to obtain average radiation measured from WARs and to apply some adjustments to these measurements. The dimensions used in the tilted cylinder are Thomas’s (1965) correlation for L, the length of the visible fire, D, the diameter of the pool, and θ, the tilt angle from the Welker and Sliepcevich (1966) correlation.

Mean surface emissive power (MSEP) (kW/m2)

254

LNG POOL FIRE MODELING

300

China Lake (1979) 200 Shell (1982)

Legend Using WAR data, cylindrical fire shape and flame height from Thomas’ equation Average spot emissive power (SpEP) calculated from WAR data and visible emission spots from video films

100

0 0

10

20

30

40

50

Pool fire base diameter (m) Figure 9.22 Mean surface emissive power calculated from wide-angle radiometer data and visible emission spots (Raj, 2007).

The MSEP values from Table 9.1 are plotted in Figure 9.22 for LNG. The legend box explains that the higher points in Figure 9.19 are taken from visible emission spots. The lower points represent an average emission over the entire surface of the fire idealized as a tilted cylinder. Figure 9.22 indicates a likely change in mechanism for large fires since it indicates a leveling off of MSEP between 20- and 35-m-diameter fires. The maximum spot values of E may also be reached with 35-m-diameter fires in the range 247–273 kW/m2 at a 35-m diameter. With larger fires, it is argued that smoke will absorb a larger proportion of the radiation. Tieszen (2007) and Luketa-Hanlin (2007) postulate that the average SEP will drop below 200 kW/m2 for pool diameters of 100 m and greater. However, no one offers an informed opinion as to how much it might decrease. Tieszen and LuketaHanlin suggest that until additional large-scale test data are available, a conservative value for SEP should be used when applying a solid flame model in the range of 220 ± 50 kW/m2 based on existing data for LNG pool fires on water.

9.10.2

Smoke Shielding and Theoretical SEP Values

The experimental values cited above for SEP are considered adequate for most modeling issues. However, it is useful to also consider the theoretical aspects of radiant energy emission to find the fundamentals behind these measured values and to provide a basis for extrapolating measured data. A key concept to understand is the formation and shielding effects of smoke. Notorianni et al. (1993) measured smoke concentration and smoke production rate in experiments with crude oil fires of diameters from 0.085 to 17.2 m.

SEP CORRELATIONS AND SMOKE SHIELDING

255

These are expected to apply also to LNG. The mass of soot produced per mass of fuel burned, Y, can be correlated with pool diameter, D, in meter, as Y = 9.412 + 2.758 log10 D in k kg .

(9.33)

The smoke mass concentration, Cs, is given in terms of the soot production rate, Y, air density, ρair, and other terms as follows: CS = ρairY

1 ⎡ r ΔH C ⎤ ⎢1 + β + C T ⎥ pa a ⎦ ⎣

in kg m 3,

(9.34)

where r = the stoichiometric air-to-fuel mass ratio for complete combustion (kg/kg); β = combustion efficiency factor, the fraction of mass of air entrained at any location that burns with its stoichiometric equivalent mass of fuel, assumed as a constant through the combustion zone; ΔHC = lower heat of combustion of the fuel (J/kg); Cpa = heat capacity of air, (J/[kg K]); and Ta = ambient air temperature, (K). The emissive power for the brightest part of the fire (near the base), Eb, is found to depend on the ratio of the pool diameter, D, to the optical thickness of the fire, Dopt, discussed below: E ( z) = Eb = Emax (1 − e − D Dopt ) for 0 ≤

z ≤ ψ, L

(9.35)

where Emax = the maximum emissive power of an optically thick fire (kW/m2) (usually a measured value); ψ = the fraction of the plume length that is bright (Zone 1), or LC = axial length of lower “bright” zone without smoke, m (=ψL); z = distance along the plume length (m) L = length of visible plume (m) The length of the lower, bright zone can be modeled as ψ = 0.7 + 0.25 log10 FC ,

(9.36)

that is, as a function of the combustion Froude number, Equation 9.24, repeated here: FC =

Gb . ρa gD

(9.37)

256

LNG POOL FIRE MODELING

For LNG fires on water, in the range 15 < D < 350 m, Equation 9.36 produces values in the range 0.1 < ψ < 0.3 (Raj, 2007, p. 57). The emissive power above the bright zone is given as the average of the bright plume emissivity, Eb, and a smoke-obscured emissivity, ES, that is diminished by the transmissivity of smoke, τS: Es = Eb τ S.

(9.38)

The transmissivity of smoke is related to the smoke concentration CS, and terms representing the optical characteristics of soot: τ S = e −(kmCS Lb ),

(9.39)

where km = specific soot extinction area, assumed value of 130 m2/kg of soot; Lb = beam length (m) (for cylindrical fires = 0.63 D) CS = soot concentration, kg/m3 (see Table 9.12) The emissive power above the bright zone is the proportional average of Eb and Enb, and the proportionality is p, the fraction of time the bright zone appears; that is, E ( z) = Eb[ p + (1 − p) τ S ] for ψ ≤

z ≤ 1. L

(9.40)

The fraction p can be expressed as a probability that the inner fire is visible. It should range between p = 1 at z/L = ψ (the boundary with the lower Zone 1) and p = 0 at z/L = 1 (the top of the fire). One expression that has these boundary conditions and behaves reasonably in between is a power law form: ⎡ (1 − ξ ) ⎤ p (ξ) = ⎢ ⎥ for ψ ≤ ξ ≤ 1, ⎣ (1 − ψ ) ⎦ 3

(9.41)

defining z the fractional plume length above the bright length, ψ = L /L. C ξ= L Equations 9.35 and 9.40 provide a complete vertical profile for E(z). This can be used to define any number of fire zones. As cited in Section 9.1, Raj (2005b) developed a three-zone model using this concept. Furthermore, integrating Equations 9.35 and 9.40, we obtain a mean emissive power over the entire – length of the fire, E, as

{

E = Eb ψ + (1 − ψ )

( 1 + 3τ S ) 4

}

.

(9.42)

SEP CORRELATIONS AND SMOKE SHIELDING

257

1

z/L

0.8 0.6 0.4 0.2 0 0

100

200

300

400

2

E(z) (kW/m ) E(z)

Data1

Data2

Data3

Eave

Figure 9.23 Comparison of predictions for E(z) with NAR data of 35-m Montoir fire tests (data from Raj, 2007).

– Figure 9.23 illustrates the above equations for E(z), p, and E applied to NAR data from the Montoir tests (Raj, 2007). The cubic power law in Equation 9.41 fits the observed local or NAR data better than a linear or square power. The mean emissive power from Equation 9.42 gives a value of 177 kW/m2 for the E(z) curve in Figure 9.23. To obtain an estimate of the term Emax in Equation 9.35, the theoretical radiant flux from an ideal blackbody emitter, EB, is EB = σTF4 in kW m 2,

(9.43)

where σ = Stefan–Boltzmann constant = 56.697 × 10−12 (kW/[m2 K]); TF(z) = radiative flame temperature of the fire (K); and κ = extinction coefficient (used in the next equation). The ideal emission is diminished by an extinction coefficient, similar to that in Equation 9.39, to give the effective emissive power of the fire, E(z), in a theoretical version of Equation 9.35: E ( z) = EB(1 − e − κD ) .

(9.44)

The inverse of the extinction coefficient, κ, is defined as the “optical depth,” Dopt. Extinction coefficients and optical depth values are summarized for available fire tests in Table 9.11. This shows optical depth approximately equal to a pool diameter for small diameters, and a decreasing fraction of a pool diameter for large diameters.

258

LNG POOL FIRE MODELING

Table 9.11

Experimental values of LNG fire extinction coefficient

Experiment

AGA tests

China Lake LNG fires on water Montoir test

110

Extinction Coefficient, κ (m−1)

Optical Depth, Dopt (m)

Reference

1.8

0.492

2.03

6.1 15

0.18 0.072

5.54 13.89

Raj and Atallah (1974) AGA (1974) Raj et al. (1979)

35

0.05

20.0

Nominal Fire Diameter (m)

Montoir LNG fire test # 2 Spectrometer at 20 m from dike rim Spectrometer aim = 11.5° wrt horizontal Air temperature = 21°C RH = 54%

100 Spectral radiance (kW/m2 μm)

Malvos and Raj (2006)

90 80 70

Spectrum of a black body at 1547 K and emissivity = 0.92 (Area below the curve) = 281.6 (kW/m2)

60 50 40

IR spectrometer data (Area below the curve) = 198.7 (kW/m2)

30 20 Near IR and visible spectrometer data

10 0

0

1

2

3

4

5

6

7

8

Wavelength (μm) Figure 9.24 Comparison of measured emission spectrum from a 35-m diameter fire with blackbody emission spectrum (Malvos and Raj, 2006).

For large fires (35-m diameter), spectral data from the Montoir fire test #2 (Nedelka et al., 1989) in Figure 9.24 clearly indicate that the fire is radiating nearly as a blackbody emitter. As indicated in the figure, the area under the measured radiation curve is 198.7 kW/m2 and that under the blackbody spectrum is 281.6 kW/m2, so 70.6% of the blackbody radiation is actually received at 20 m. Large decreases from the ideal radiation occur at the absorption bands for water vapor and CO2, as, for example, near 2.66–2.69 μm. The flame temperature, TF, for a China Lake test at a 13-m diameter was calculated at close to 1500 K (1230°C). The fire emissivity, temperature, and optical depth were found with a single set of spectral data taken at 236 m from

ATMOSPHERIC TRANSMISSIVITY

259

the fire. These data showed the fire to be primarily a band emitter rather than a blackbody emitter; that is, the emissions from water vapor and CO2 were at specific wavelengths, although emissions from luminous soot were a continuous function of wavelength (like a blackbody emitter). The emissivities of luminous soot, water vapor, and CO2 were found to be 0.14, 0.19, and 0.35, respectively. 9.10.3

Validation Comparison of a Three-Zone SEP Model

Predictions of the model equations in this section are compared to experimental measurements in Table 9.12. These results are reported by Raj (2007) using the coding called the PoFMISE model. Extrapolations are made to a large possible future test and to a large pool arising from a puncture of one tank in an LNG ship. 9.11

ATMOSPHERIC TRANSMISSIVITY

The atmosphere absorbs a substantial fraction of the infrared (IR) energy radiated from a fire in specific bands of wavelengths by CO2 and, primarily, by water vapor. (Fig. 9.24 shows the main absorption bands.) The atmospheric transmissivity, τatm, introduced in Equations 9.3 and 9.4 at the beginning of this chapter, is an important correction and should not be omitted. Simpson (1984) states that “The inclusion of a value for atmospheric transmissivity in the calculation of hazard ranges from pool fires and fire balls will give [for typical weather conditions] a reduction in hazard range of 10–40%.” Transmissivity values are provided by Raj et al. (1979) using a method of “windows” or regions of high transmission between absorption bands. Simpson (1984) allows not only for absorption but also for scattering. Simpson uses transmission windows over eight ranges of wavelength; an absorption coefficient, τai, and a scattering coefficient, τsi, are calculated over each window range, i. Absorption depends on the amount of precipitable water in the atmosphere, and scattering uses visibility values, V, as a correlator. The transmissivity for each window range is found by the product, τai and τsi. The curves by both authors are roughly similar (i.e., ±10%) for distances up to 1000 m, after which there is greater variation. The Raj curves give higher values of τatm than those of Simpson as distance increases, probably because Simpson considers atmospheric scattering, which becomes significant at longer distances. Figure 9.25 plots the atmospheric transmissivity, τatm, from Simpson (1984) for a specific set of parameter values against the path length, s, with RH as a parameter. Simpson provides a number of similar plots for the following parameter range: • • •

equivalent blackbody temperature of a fire: 1000–1600 K; ambient air temperature: 0, 15, and 30°C (273, 288, and 303 K); and visual range: 2, 5, 10, and 20 km.

Substrate

Water

Land

Land

Land

Water

Fire Diameter

15

20

35

100

300

90

113

177

183

172

PoFMISE Model Predictions

—

0.033

0.093

0.150

175 ± 30

—

0.180

0.196

Fractional Length of Bright Zone (ψ)

140–180

185–224

Field Test Measurement

MSEP over Visible Plume Height (kW/m2)

16.2

14.9

13.7

13.0

12.7

Soot Mass Yield, Y (%)

4.272E-4

3.926E-4

3.595E-4

3.419E-4

3.328E-4

Soot Conc., CS (kg/m3)

Table 9.12 Comparison of model predictions of MSEP with experimental data and extrapolations

0.00277

4.00

35.7

57.12

66.40

Soot Transmissivity, τS × 10−2

China Lake Tests (Raj et al., 1979) Mizner and Eyre (1983) Montoir tests (Nedelka et al., 1989) Potential future pool fire tests Estimated from one tank spill of LNG ship

Comments

260 LNG POOL FIRE MODELING

261

ATMOSPHERIC TRANSMISSIVITY

Transmissivity

Blackbody T = 1150 K, ambient T = 15 C, visual = 20 km 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 1 10 100 1000 10,000

Distance (m) RH = 10% Figure 9.25

Table 9.13

RH = 30

RH = 70

RH = 100%

Atmospheric transmissivity (data from Simpson, 1984, p. 26).

Relationship between atmospheric conditions and absorption coefficient

Atmospheric Condition Clear to very clear Clear Light haze Haze Foggy, smoky

k

Visibility, V (km)

0.1–0.003 0.2–0.4 0.4 1.0 0.4–2.0

40–80 20 10 4–5 88 kW m2

Figure 9.34 Time for weather cover to reach 1023 K (750°C) for various fire flux levels (Wuersig et al., 2009) (reproduced by permission from SIGTTO).

segments) of 6.51 min to reach the buckling temperature is required for a fire with an initial heat flux of 200 kW m−2. This burning time is available for LNG spills caused by a breach size smaller than 6 m2. For hole sizes above 3 m2 the weather cover cannot reach the buckling temperature due to a fire with an initial heat flux of 108 kW m−2 [lowest three segments in Figure 9.34]. Due to heat fluxes lower than 108 kW m−2 [lowest four segments in Figure 9.34] caused by breach sizes less than 3 m2 the weather cover can reach the melting temperature. The heat flux into the spherical LNG tank system will increase if the insulation is nearly completely deteriorated. Not till then will the LNG start boiling and consequently the pressure inside the tank will increase and cause the pressure relief valves on the tank to open. But, when the melting of the insulation starts, the period for heating up the rest of the insulation and melting has to be considered. For an initial heat flux of 108 kW m−2 the complete deterioration of the insulation will last at least 47.5 min. [Even] in the case of an initial heat flux of 300 kW m−2 the complete deterioration of the insulation will last 29.5 min.

The analysis shows that the existing systems provide sufficient thermal protection for the duration of fires defined by Sandia. The longest-lasting fires are from the smallest holes, and even a 1-m2 hole generating 108 kW/m−2 is insufficient to destroy the thermal insulation and hence to cause the relief system to lift. And even if this were to occur, there is excess capacity in the relief system as the relief valves are set at 0.25 barg and the tank strength is sufficient for over 4 barg. In conclusion, these two studies offer quite different analyses leading to two different results. While the SIGTTO analysis appears more realistic, greater insights will become available shortly from the Sandia research on pool fire, and the cryogenic response of LNG carrier structural members is completed.

10 OTHER LNG HAZARDS

The focus of liquefied natural gas (LNG) consequence assessment to this point has been on source term, dispersion, and pool fire. There are, however, other consequences, and these are addressed in this chapter. These include jet fire, flash fire, boiling liquid expanding vapor explosion (BLEVE), and vapor cloud explosion (VCE). If LNG vapor clouds do not ignite, then their primary hazard is cold exposure and potentially asphyxiation affecting only those very close to the source. 10.1

FIRE AND EXPLOSION SCENARIOS

LNG is handled in two types of plants and in two types of transportation: • • • •

LNG liquefaction plants, LNG vaporization plants, marine transportation (LNG carriers), and land transportation (LNG trucks).

Two main event trees can be identified—those for pressurized and nonpressurized releases. In each case, consequences are affected by several factors including the duration and size of the release, the temperature, pressure, and phase of the discharge, and whether there is immediate or delayed ignition. LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

275

276

OTHER LNG HAZARDS

Non-ignited cases may lead to cold burns or, conceivably, asphyxiation and are treated in Section 10.6. Key factors affecting the release duration are whether there is a gas detection array installed, adequate to detect spills in any wind direction. Early detection and shutdown primarily affects the scale of the loss of containment event. Detectors would normally alarm to a control room, and operators there should be able to initiate an emergency shutdown (ESD) system, often with multiple levels of shutdown (from limited area to total facility shutdown). Only some parts of an ESD will have automatic trips. For example, should a vessel move at berth, the loading arm powered emergency release connection (PERC) described in Chapter 6 will shut down pumps, resulting in essentially no spillage of LNG. Other locations will normally involve manual initiation of ESD functions. Large leaks on main discharge LNG lines will be detected by the gas detectors or by low-pressure alarms. Emergency isolation valves will shut on command, but these cannot be shut too quickly or a “water hammer” effect in long lines could lead to a more serious rupture. Design calculations establish the minimum closing duration. This can often be 30–60 s. LNG liquefaction plants may have depressurization and blowdown systems installed to deinventory the zone where the leakage is occurring, Such zones may apply to refrigerant fluids such as ethane, ethylene, and propane as well as to the LNG circuit. Such systems are rare at import terminals. Early and late ignition affect the outcome. Immediate ignition will result in either a jet fire alone if the release is gas or a jet fire and pool fire combined if the release is liquid—lower pressure or larger jets will rain out burning liquid and will form a pool fire. Higher pressure or smaller jets are likely to consume all the liquid with no significant rainout and pool formation. Table 10.1 lists a set of scenarios involving loss of containment in one of the above type of plant or mode of transportation. The type of fire or explosion that is likely to develop is listed along with the type of damage that is likely. (A similar table is developed by Raj [2006].) NFPA 59A also sets out a number of events that should be modeled for onshore import terminals. Taylor (2007) summarizes current consequence approaches to satisfy the NFPA 59A standard in the United States.

10.2

JET FIRES

An LNG jet fire occurs when the liquid is released under pressure through an orifice or a crack and is ignited. Jet fires can impinge upon buildings and process equipment and act, literally, like a blowtorch. They generate very high heat fluxes and can burn through most objects given enough exposure time. However, the reach of jet fires is very limited, especially for liquid jets, typically less than 50 m. So the hazard of jet fires is primarily confined to an LNG processing plant.

External fire

Liquid

6. LNG tank leakage

Liquid jet

Liquid leak

5. Refrigeration circuit (C2/C3/ C4 refrigerant)

Dense gas release

Gas leak

3. Refrigeration circuit (C2/C3/ C4 refrigerant) 4. Refrigeration circuit (C2/C3/ C4 refrigerant)

Liquid jet

BLEVE

Two-phase jet

Two-phase vapor liquid leak

2. LNG cold box area

Buoyant gas release

Outcome

Gas leak

Event

LNG liquefaction plant events

1. Natural gas feed

Zone

Table 10.1

JF or PF

Yes

JF or PF

JF

JF

JF

Immediate Ignition

FF followed by PF

FF or VCE if directed into congested plant FF and JF if free field FF or VCE if directed into congested plant FF followed by JF and PF if free field Unlikely

FF or VCE if directed into congested plant JF if free field FF or VCE if directed into congested plant FF and JF if free field

Delayed Ignition

BLEVE is possible as an escalation event from another leak and external fire impinging on refrigerant tank. Leak should flow into the impoundment, which is a designed safe location. NFPA 59A specifies design for 10-min release at full flow.

As for LNG, but greater likelihood of VCE if congested

Most releases will rise up quickly and ignition is of lower probability. Dense gas can drift into the congested area. Methane is less likely to explode than higher hydrocarbons but VCE is possible. As for LNG, but greater likelihood of VCE if congested

Comment

JET FIRES

277

Liquid spill

Liquid leak

Liquid leak

8. LNG tank instantaneous failure

9. Jetty line failure

10. Loading arm connection failure

Liquid jet under pump pressure

Liquid jet under pump pressure

Liquid overtop most dikes

Dense gas

Outcome

JF

JF followed by PF

PF

FF

Immediate Ignition

Nil, vapors disperse

FF followed by PF

FF followed by PF VCE possible in congested areas

FF

Delayed Ignition

Comment Rollover is not a design case for NFPA 59A as easily prevented by routine mixing. Not a design case for NFPA 59A but would be considered in a full risk assessment, leading possibly to a requirement for a double-wall tank design, not required by NFPA 59A Spill into impoundment and possibly sea if large—long lines require slow-acting shutdown valves to prevent water hammer rupture escalation. PERC connection should limit discharge to amount below any significant consequence.

FF, flash fire; BLEVE, boiling liquid expanding vapor explosion; JF, jet fire; PF, pool fire; VCE, vapor cloud explosion.

Gas evolution

Event

7. LNG tank rollover

Zone

Table 10.1 Continued

278 OTHER LNG HAZARDS

JET FIRES

Table 10.2

279

LNG regasification terminal events

Zone

Event

Outcome

Immediate Ignition

Delayed Ignition

1. LNG tank leakage 2. LNG tank rollover 3. LNG tank instantaneous failure 4. Jetty line failure 5. Loading arm connection failure

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

Comment Same as event 6 in Table 10.1 Same as event 7 in Table 10.1 Same as event 8 in Table 10.1 Same as event 9 in Table 10.1 Same as event 10 in Table 10.1

Two issues are important with jet fires: radiation from the jet and direct impingement upon an object. For people, the former is more important to model as the latter always results in death. For structures, both aspects are important. For impingement, protection is usually based on experimentally determined heat flux rather than on calculations. Hydrocarbon jet fire-rated firewalls are usually specified, with some duration of survival (e.g., 30-min firewalls). Since impingement is not usually modeled, no discussion is provided here and the focus is on radiant heat that can affect people or structures. The Dutch research organization TNO (2005) identifies three types of models used to predict thermal radiation from jet fires: semiempirical, integral, and field models. The simplest model for jet fires is the semiempirical type. As provided in API 521, this is a single point source model that is based on an arbitrary estimate of combustion energy radiated combined with a simple jet flame shape that allows a central impingement point to be determined. Typically, the fraction of thermal radiation energy emitted is in the range of 0.15–0.4 with methane toward the low values and ethylene toward the high values. The semiempirical models may be too simple for accurate assessment of thermal hazards. The American Petroleum Institute (API) model is used for estimating thermal radiation at ground level from ignited relief valve discharges, which are always assumed as pressurized jets of vapor. On the contrary, pressurized releases associated with LNG are either liquid or vapor and are often near ground level. For these, the API radiation method is too simple, and some method considering the full geometry of the jet and radiation along the whole length is important. A more detailed semiempirical approach of the point source approach divides the jet flame into a number of segments and develops point source radiation estimates for each segment. Then, the contributions are added to determine thermal impact at any specified point. A number of such models

— Liquid pool on sea surface Possible damage to LNG tank

— Liquid leak

Cascading event

—

—

5. LNG tank rollover

6. Export line failure

—

—

—

Outcome

—

Event

LNG carrier transport events

1. Loading arm connection failure 2. Tank fill line failure 3. Penetration of tank by collision, grounding, or terrorism 4. Prolonged external pool fire

Zone

Table 10.3

—

—

To be determined

— PF

—

Immediate Ignition

—

—

To be determined

— FF followed by PF

—

Delayed Ignition

This event is discussed in detail later in this chapter and in Chapter 9, Section 9.15. There are two divergent published views, and Sandia is carrying out experiments to better define the outcome. A BLEVE is very unlikely, but damage to the tank is conceivable if the exposure is prolonged. Not conceivable for a sailing vessel due to wave motions stirring up the tank Same as event 10 in Table 10.1

Same as event 9 in Table 10.1 Analyzed in detail in Chapter 3

Same as event 10 in Table 10.1

Comment

280 OTHER LNG HAZARDS

Liquid spill

Liquid spill

Total liquid spill

External fire

2. Truck tank leak

3. Truck tank instantaneous failure

4. Truck tank

Event

Outcome

BLEVE

Liquid spill under low pressure Major pool

Liquid jet under pump pressure

LNG truck and rail transport events

1. Loading or unloading arm connection failure

Zone

Table 10.4

PF

PF

JF followed by PF

Immediate Ignition

FF followed by PF VCE conceivable if vapor drifts into congestion

FF followed by PF

FF followed by JF VCE conceivable if vapor drifts into congestion

Delayed Ignition

Truck tank may fail in traffic accident. VCE unlikely but possible depending on the location of the accident. Immediate ignition in accident event more likely than delayed ignition. BLEVE is a possible outcome— see accident summary and later discussion in this chapter.

VCE requires a large spill event. Filling station likely to have greater congestion potential than unloading, but VCE is possible in either case. VCE very unlikely

Comment

JET FIRES

281

282

OTHER LNG HAZARDS

have been developed and are listed in TNO (ibid.), but these have for the most part fallen into disuse as better integral models have been developed and made available in commercial consequence software. More versatile computational fluid dynamics (CFD) models solve the timeaveraged partial differential Navier–Stokes equations and thus, in principle, are more fundamental than the other approaches. However, additional CFD submodels need to be developed to solve key mechanisms of turbulence and gas combustion (including soot formation). With such development, TNO notes that the CFD approach is not necessarily any more fundamental than the simpler approaches. However, CFD models can account for impingement and where this is important, they may be the preferred approach. Kameleon is one well-known CFD code for jet fires used extensively in the offshore oil industry and where necessary for LNG jet fires (Velde et al., 1998). Integral models start with a similar mathematical description as the CFD models, but apply simplifying assumptions that permit the equations to be converted to ordinary differential equations amenable to integration. Most jet fire analyses in free-field situations (i.e., not impinged) now use this approach. Even impinged cases on important structures or vessels can be analyzed. For these cases, a passive fire protection solution can be developed without the need for detailed modeling. Lees (Mannan, 2005, Sections 16.18 and 16.19) provides a more complete review of jet fires and related flares. A commonly accepted jet fire model is due to Chamberlain (1987). This model is incorporated in a model widely used by industry. A simpler model of Cook et al. (1990) described also by Lees (Mannan, 2005, p. 16/220) is described below. This is based on the API RP 521 flare model. The relationships for jet fire length L and radius Rj, in meter, are L = 0.00326 [ Fdis ΔHC ]

0.478

and

(10.1)

⎛ Rj ⎞ ⎛ s⎞ ⎛ L⎞ ⎜⎝ ⎟⎠ = 0.0365 ⎝ ⎠ ln ⎝ ⎠ , L L s

(10.2)

2

2

where Fdis = mass discharge rate (kg/s); ΔHc = absolute magnitude of heat of combustion (J/kg); and s s/L = fractional distance along the plume length, 0 ≤ ≤ 1 L For methane or natural gas jet fires, the peak flame temperature inside the fire can be as high as 1252°C (1525 K) (Mannan, 2005, Section 16.19) with an equivalent blackbody emissive power of 307 kW/m2. For a point source jet model, Clay et al. (1988) proposed for a similar model; this same emissive power should be used at a point 4/5 of the flame length from the origin. Chamberlain provides a simple relation to predict the fraction of combustion energy radiation. This is given by the equation below. For Vj = 200 m/s, the radiant fraction is 0.22.

JET FIRES

283

35

Predicted heat flux (kW/m2)

0% deviation 30

–15% deviation 15% deviation

25

–40% deviation 40% deviation Johnson (1994)-simulated data

20

PHAST 6.53

15 10 5 0 0

5

10

15

20

Measured heat flux (kW/m2) Figure 10.1 Validation of jet fire model results (Witlox and Oke, 2008).

Fs = 0.21 × exp ( −0.00323Vj) + 0.11,

(10.3)

where Fs = radiant fraction and Vj = post-throat expansion jet velocity (m/s) Overall, jet fire models based on Chamberlain’s formulation can lead to good predictions of jet fire thermal radiation intensity. Witlox and Oke (2008) present validation data for the PHAST package jet fire model (Fig. 10.1). This shows that in free field situations, commercial models can deliver accurate predictions for the thermal intensities important for impacts to people and structures. It is important to realize there are theoretical limitations to both the Cook et al. and Chamberlain models. Negligible premixing with air is possible, so jet fires are primarily diffusion fires. Air is entrained by the shear forces developed by the jet velocity, so an unignited jet plume has similar characteristics to an ignited jet. The concentration contours to the stoichiometric and lower flammable limit (LFL) of an unignited jet can be taken as bracketing approximations for the burning jet dimensions. However, jet fires are buoyant by virtue of generating heat. Horizontal gas jet fires have an upward bending shape. Horizontal liquid fires are less buoyant and remain more horizontal. These points are illustrated in Table 10.5 and in Figures 10.2–10.4 using the SafeSite model. First, from Table 10.5, for the same composition, pressure, and hole size, the discharge rate is an order of magnitude higher for a liquid.

284

OTHER LNG HAZARDS

Table 10.5 Predictions of dimensions of ignited and unignited jets of vapor and liquid Nigerian LNG

Phase

Vapor

Liquid

Hole Diameter, inch (m)

Discharge Rate (kg/s)

0.5 (0.0127) 1.0 (0.0254) 2.0 (0.0508) 0.5 (0.0127) 1.0 (0.0254) 2.0 (0.0508)

0.10 0.40 1.6 1.1 4.5 18.5

Cloud centerline

4.0

UFL

Unignited Jet Length to LFL (m) 2.9 5.5 11 13 21 85

Length to ½ LFL (m)

Jet Fire Length (m)

5.8 10.7 21 17 68 176

4.9 9.5 18.6 15.7 30.6 59.8

LFL

0.5 fraction LFL

3.8

Vertical height (m)

3.6 3.4 3.2 3.0 2.8 2.6 2.4 2.2 2.0 0

2

4

6 8 Downwind distance (m)

10

12

14

Figure 10.2 Unignited jet of Nigerian LNG vapor, 0.5-in. hole at 35 psig.

8

Cloud centerline

UFL

LFL

0.5 fraction LFL

Vertical height (m)

7 6 5 4 3 2 1 0 0

2

4

6

8 10 12 Downwind distance (m)

14

16

18

20

Figure 10.3 Unignited jet of Nigerian LNG liquid, 0.5-in. hole at 35 psig, 45° release direction.

JET FIRES

Cloud centerline

UFL

LFL

285

0.5 fraction LFL

9

Vertical height (m)

8 7 6 5 4 3 2 1 0 0

Figure 10.4

10

20

30 40 Downwind distance (m)

50

60

70

Unignited jet of Nigerian LNG liquid, 1-in. hole at 35 psig, 45° release direction.

Second, comparing Figure 10.2 (a vapor jet) with Figures 10.3 and 10.4 (liquid jets), a liquid jet bends much more because of a density that is two orders of magnitude higher than the vapor. In these figures, the middle profiles are for the LFL and the longest profile is for ½ LFL. Third, a liquid jet discharged from a relatively low elevation will shortly touch the ground and rain out. Figure 10.4 illustrates that this is about to happen just below the LFL concentration. The dispersion (including pool spread and evaporation) after touchdown and rain out is bounded by the ground. The jet length and radius predicted by Equations 10.1 and 10.2 would not be expected to apply after touchdown and rain out since the jet plume no longer exists, but rather an evaporating pool describes the situation. Figure 10.5 plots the jet fire radius predicted by Equation 10.2 for the vapor and liquid discharge rates in Table 10.5 with 0.5- and 1.0-in. (0.0127 and 0.0254 m) holes. The radius profiles are roughly proportional to the discharge rate. This proportionality illustrates the assumption that all of the liquid burns as a vapor, so phase density differences disappear in the modeled fire crosssectional area. The simple jet model applies to either horizontal jets in the downwind direction or vertical jets bending in a wind. The latter uses the following for the plume bending: dz 1.6πDj u j ⎛ 1 1 ⎞ = − , ⎝ s L⎠ dx uw

(10.4)

where Dj is the diameter of the jet (2Rj from Equation 10.2), uj is the initial jet velocity, uw is the wind speed, L is the jet flame length, and s is the distance along the jet centerline. The centerline trajectory of the jet fire is found in a spreadsheet by numerically integrating

286

OTHER LNG HAZARDS

3 2 Rs (m)

1 0 –1 –2 –3 0

5

10

15

20

25

30

35

X (m)

Figure 10.5

Vapor, 0.5 inch

Vapor, 1.0 inch

Liquid, 0.5 inch

Liquid, 1.0 inch

Jet fire radius predictions by the Cook et al. model for vapor and liquid LNG.

x ( s) =

L

∫

s=0

z( s ) =

ds ⎡ ⎛ dz ⎞ 2 ⎤ ⎢1 + ⎝ dx ⎠ ⎥ ⎦ ⎣ L

∫

s=0

12

ds ⎡ ⎛ dx ⎞ 2 ⎤ ⎢1 + ⎝ dz ⎠ ⎥ ⎦ ⎣

12

and

(10.5)

.

(10.6)

An example of the jet centerline bending in a cross-flow wind is shown in Figure 10.6 for the 1-in. hole liquid discharge in Table 10.5 with an initial jet velocity of 31.5 m/s and a wind speed of 7.5 m/s.

10.3

FLASH FIRES

Once a flammable cloud is ignited, a combustion wave moves through the cloud. A typical dispersing LNG cloud will be pancake shaped, and its flammable phase will be a dense gas (see Chapter 8). Ignition will normally be at the edge, but if it occurs in the middle, and if the cloud is narrow, then the combustion front has been observed to be almost flat like a vertical wall moving both back toward and away from the source as illustrated in Figure 10.7. The velocity of a flame front is slower when burning against the wind and faster with the wind. The idealized diagram in Figure 10.7 does not account for concentration profiles through a plume and flame speed variability with concentration. Flame speed is faster for premixed air and fuel, so as illustrated in Figure 10.8, the

FLASH FIRES

287

30

z, elevation (m)

25 20 15 10 5 0 0

2

4

6

8

10

12

14

x, downwind (m) z(s) (m)

Figure 10.6 Centerline of vertical jet fire from a 1-in. hole in 7.5 m/s wind speed.

Wind

Release point Flammable cloud

Upwind flame

× Ignition point

Flammable cloud

Downwind flame

Figure 10.7 Illustration of idealized flame fronts for a flash fire (Cracknell and Carsley, 1997) (reproduced by permission from IChemE).

flame front velocity can be faster on the edges and tends to develop an enveloping flame. Figure 10.8 is from a test performed by the Health and Safety Laboratory (HSL, 2001) for an LPG vapor cloud with a discharge rate of 2.6 kg/s for 21 s, wind speed of 2.0 m/s, and ignition source 25 m downwind of the source. In this case, the flame front did not burn into the upwind plume. Flash fires expand upon combustion, like explosions, typically by a volume factor of around 8. In a 3-D expansion, the linear expansion in each direction is a factor 81/3 or about two. This is why if the flame front reaches a rich portion of the cloud (above the stoichiometric limit and below the upper flammable limit [UFL]), then the expanding combusted gas forces some unburned fuel upward, and there is a rising flame along the flame front as idealized for the model by Raj and Emmons (1975) in Figure 10.9. If, however, the flame front

288

OTHER LNG HAZARDS

Figure 10.8

Flame front progression in LPG vapor cloud (HSL, 2001).

is in a lean portion of the cloud, there is no noticeable rising at the flame front. The flame height H in flash fires is not well studied, but the Coyote trials (Table 9.1, Chapter 9) suggest values of height to width ratio 5–10 for LNG. The model of Raj and Emmons (ibid.) is still commonly used. This considers the flash fire as a two-dimensional, turbulent flame propagating at a constant speed. Details are provided in Lees (Mannan, 2005) and in the Center for Chemical Process Safety (CCPS, 2009b). Combustion velocities measured in the laboratory are usually in laminar flow conditions. Lees (Mannan, 2005) reports that, in general, laminar burning velocities for paraffinic hydrocarbons range from a few centimeters per second near the flammability limits to about 45 cm/s near the intermediate stoichiometric concentration. Methane has a reported maximum laminar flame velocity under laboratory conditions of 36 cm/s. Real-world combustion is affected to a significant degree by turbulence, especially turbulence generated by obstacles. The flame through a flash fire is turbulent, burning at flame speeds in the range of a few meters per second to tens of meters per second. Propane experiments (Mizner and Eyre, 1982) gave average flame speeds of up to about 12 m/s. Higher transient flame speeds up

FLASH FIRES

289

H

D

Unburnt vapor

S

W

Figure 10.9

Flash fire event progression idealized for model of Raj and Emmons (1975).

to 28 m/s were observed in one instance. In completely open environments, in a report of one Maplin Sands test, with LNG, a wind speed (measured at a 10-m elevation) of 4.5 m/s was able to hold the flame stationary at a point 65 m from the source for almost 1 min (ibid.). Daish et al. (2001) compiled flame speed data and plotted it against wind speed in Figure 10.10 from Raj (2007). Experimental work reported for Maplin Sands tests (Table 9.1 in Chapter 9) shows that dispersing flammable clouds can be uneven in shape and concentration depending on release conditions, ground roughness, wind speed, and atmospheric stability. Local pockets of gas can have higher or lower concentration than the local mixture average. Inhomogeneous concentrations can lead to variations of combustion velocity. To account for such pockets and for uncertainties in modeling and averaging of concentrations, a value of 0.5 LFL is typically considered as the border for engulfing personnel. Lees (Mannan, 2005, p. 15/286) attributes Feldbauer et al. (1972) for first substituting 0.5 LFL for the LFL to account for the cloud area susceptible to ignition. This is discussed in more detail in Sections 4.6 and 12.4.4. In a flash fire event, the major hazard is to people either in the flame envelope or located above it on elevated structures. Combustion temperatures are up to 1200°C, but as the flame envelope passes over specific locations in only a few seconds, flame duration and intensity, while sufficient to cause fatalities or to damage cabling and external surfaces, are usually insufficient to cause significant damage to structures or heavy equipment items. The literature provides little information on the effects of thermal radiation from flash fires,

290

OTHER LNG HAZARDS

UF = flame speed wind respect to unburnt gas (m/s)

15 14

TRW (1968) Gaz de France (1972) LLNL (1984) AGA (1974)

12 10

Least square linear UF = 0.8 + 1.6 UW

8 6 4 Data point not included in the least square fit

2 0

0

1

2

3 4 5 6 7 8 UW = average wind speed (m/s)

9

10

Figure 10.10 Correlation of flash fire flame speed with wind speed (Raj, 2007) (reproduced by permission from Elsevier Science Publishing, Inc.).

Figure 10.11 Coyote 3 burn trial (Koopman and Ermak, 2007) (reproduced by permission from Elsevier Science Publishing, Inc.).

probably because thermal radiation hazards from burning vapor clouds are considered less significant than possible blast effects. Furthermore, flash combustion of a vapor cloud normally lasts no more than a few tens of seconds. Koopman and Ermak (2007) provide a photograph from the Coyote 3 burn trial carried out in 1981, shown in Figure 10.11. This is for an LNG pool

BLEVES, FIREBALLS

291

dispersion under unstable weather (B–C stability) and moderate wind speed (6 m/s). The spill volume was 14.6 m3, and this was spilled over about 1 min. The resulting cloud extended 195 m to LFL concentration. The cloud was ignited after 100 s at a central location, and the flame front burned in both directions as illustrated in Figure 10.7. In Figure 10.11, the right side is stationary over the pool and the left side is continuing to burn leftward with a heightto-width ratio consistent with other trials in the range of 5–10. The flame velocity was about 20 m/s in this test. The authors (ibid.) note that the heat-affected zone as measured by refractometers was significantly larger than the visible flame. They report an average maximum heat flux of about 260 kW/m2, obtained from calorimeters in the fire-affected area. The measured thermal impulse (the heat transferred) varied from 348 to 507 kJ/m2. This corresponds to a 1.5- to 2.0-s exposure at the average heat flux. These levels exceed those necessary for third-degree burns by a large factor and are, with larger exposure time, great enough to ignite most flammable materials. The Coyote 3 area burned was about 3200 m2, about half of what would be expected from the pre-ignition 5% contour. For contrast, Coyote 6, which was a 22-m3 spill, caused a burn area of 16,200 m2. The experiments also assessed the effect of ignition strength, and when a jet igniter was used, it caused a faster burn rate in both directions for the first 50 m, but thereafter, the effect was not noticeable. CCPS (2009b) specifically addresses flash fires in its text on VCE, flash fire, and BLEVE events. There is some limited discussion in Lees (Mannan, 2005) and CCPS (2000a). Altogether, flash-fire modeling is largely underdeveloped in the literature; there are large gaps in the information base. Hardly any information is available concerning flash-fire radiation; the only data available have resulted from experiments conducted to meet other objectives. Many items have not yet received sufficient attention. The fact, for instance, that burning speed is taken as proportional to wind speed implies that, under calm atmospheric conditions, burning velocities become improbably small, and flash-fire duration proportionately long.

10.4

BLEVES, FIREBALLS

Fire balls are known as events that can develop thermal radiation hazard distances larger than pool fires. There are two types of fire balls that rise as a near sphere while burning: 1. fire balls resulting from the rupture of a bursting pressure vessel and 2. fire balls from the ignition of a large accumulation of flammable vapors at atmospheric pressure.

292

10.4.1

OTHER LNG HAZARDS

BLEVEs and Applicability to LNG

The first situation usually involves flashing and expanding overheated liquids, referred to in the literature as BLEVEs. A BLEVE is a high-hazard thermal event well known in the LPG industry. Its application to LNG has been largely discounted as either impossible or highly improbable. None of the three major studies on marine transport hazards in 2004 by Sandia, American Bureau of Shipping (ABS), and Det Norske Veritas (DNV) identified BLEVE as a significant threat; however, Venart, with considerable expertise in this area (Venart et al., 1989, 1993), has challenged this position and maintains that massive LNG BLEVE events are possible (Venart 2005). The U.S. Government Accountability Office (GAO, 2007) survey, discussed in Section 12.1 of Chapter 12, identified this as an issue, but only about half the experts agreed that this was a potential hazard, and most of these believed it was very unlikely. Several definitions of BLEVE are available. A simple definition is given by CCPS (2009b): BLEVE: “any sudden loss of containment of a liquid above its normal boiling point at the moment of its failure. It can be accompanied by vessel fragmentation and, if a flammable liquid is involved, fireball, flash fire, or vapor cloud explosion.”

A fuller definition given in Lees (Mannan, 2005) explains the mechanism: When a vessel containing liquid under pressure is exposed to fire, the liquid heats up and the vapor pressure rises, increasing the pressure in the vessel. When this pressure reaches the set pressure of the relief valve, the valve operates. The liquid level falls as the vapor is released to the atmosphere. The liquid is effective in cooling that part of the vessel wall which is in contact with it, but the vapor is not. The temperature in the proportion of the vessel wall which has the benefit of liquid cooling falls as the liquid vaporizes. After a time, metal which is not cooled by liquid becomes exposed to the fire; the metal becomes hot and weakens and may then rupture. This can happen even if the relief valve is operating correctly. A pressure vessel is designed to withstand the relief valve set pressure, but only at the design temperature conditions. If the metal has its temperature raised, it may lose it strength sufficiently to rupture.

Several well-known BLEVE accidents have occurred historically, and the hazard is well understood by firefighters who have developed firefighting strategies to protect firefighters who are often close to tanks on fire and hence to potential BLEVE events. One of the largest BLEVE events was the Pemex fire in Mexico City in 1984, when several adjacent LPG spheres suffered in a cascade of BLEVEs (Pietersen, 1985; Berenblut et al., 1985; Pettitt et al., 1994). The primary impact of large BLEVE or fireball events is thermal radiation, although as the containment vessel fails, there are usually several larger fragments thrown for up to several hundred meters. The fragments are large unlike the multiple small fragments associated with a detonation, and the

BLEVES, FIREBALLS

293

damage is localized to the impact location. The thermal radiation from LPG BLEVEs is intense since the entire hydrocarbon in the containment combusts in under 60 s at relatively low elevation immediately above the tank. The combustion is associated with a large increase in volume as a fraction of the liquid almost instantaneously turns to vapor, which occurs at sonic velocity, and there is a significant pressure wave created, hence the term explosion is used. The situation is different for cryogenic LNG as not all the LNG will expand into the fireball. This is discussed later. The vast majority of BLEVE events have been to pressurized LPG vessels subject to external fire. BLEVE events occur when the container above the liquid level heats sufficiently due to external fire that it thermally weakens and fails. Construction steels as reported by the Federal Emergency Management Agency (FEMA) in its analysis of the World Trade Center fire lose 10% of ambient yield stress at 200°C and 50% at 550°C. Many pressure vessels have a safety factor of 2. Thus, when, due to external pool fire or jet fire, the temperature of the steel shell above the liquid level rises to 550°C, then the steel yield stress equals the internal pressure hoop stress and the shell will fail along its weakest line. This is usually essentially instantaneous and catastrophic. Hydrocarbon tanks, which remain at atmospheric pressure throughout the impacting fire event, will not BLEVE (e.g., fuel oil tanks). These will give a pool fire if the containment fails, but there is no internal pressure energy to dissipate. This is the argument to discount BLEVE events for LNG; it is valid if the pressure in the LNG tank remains at atmospheric pressure. Only one LNG BLEVE event has been identified. Planas-Cuchi et al. (2004) describe an LNG road tanker accident that occurred in June 2002 in Spain. Since this challenges common assumptions, it is useful to review key facts and tanker design details in Table 10.6. The truck started with the cryogenic LNG at atmospheric pressure, but due to the high relief valve set point, the LNG was heated sufficiently by the fire to reach near 8-bar pressure and no longer had the properties and hazards of an atmospheric pressure fluid. The event had three distinct impacts—all of which match classical BLEVE outcomes for LPG tanks: overpressure, thermal radiation, and fragments. Planas-Cuchi et al reviewed pressure damage impacts and based on windows remaining intact at 125 m, back calculated a TNT equivalence of 75 kg and an internal pressure in the tank of 8 bar. The trailer tank disintegrated into a small number of large fragments mostly thrown in alignment with the cylindrical tank. The vehicle motor was ejected the greatest distance of 257 m; tank fragments were ejected about 125 m. Four internal tank baffles were ejected 50–125 m sideways. This small number of fragments and orientation mostly along the cylindrical axis is typical for BLEVE events. Thermal effects were estimated by Planas-Cuchi et al. based on the total initial contents of the tank. The method of CCPS (2009b)calculated a fireball diameter of 150 m, centered at a height of 113 m, lasting 12 s. The authors

294

OTHER LNG HAZARDS

Table 10.6

Spanish LNG tanker incident

Accident Key Features •

•

•

•

•

•

•

•

•

•

An LNG road tanker rolled over onto its side. The accident dislodged thermal insulation over a space then occupied by vapor. Flames appeared immediately between the driver cab and the trailer tank. Initially, the flames were reported as smokeless. Soon after, the vehicle tires became involved and both black and white smoke evolved. The tank exploded in two steps 20 min after the initial accident. Initially, there was a small explosion followed by a hissing sound. Subsequently, there was a large explosion which created a large white cloud. This white cloud ignited immediately and a fireball ensued. The driver was killed and two persons were burned approximately 200 m from the truck.

Tanker Design Matters •

•

•

•

•

•

•

•

• •

Tank dimensions: 13.5 m long, 2.33-m diameter, volume of 56 m3 Normal fill was 85% liquid (47.6 m3) and 15% vapor (8.4 m3). Tank walls: ANSI-304 stainless steel, cylindrical wall thickness of 4 mm, ends at 6 mm Insulation: 130-mm expanded polyurethane foam, self-extinguishing Tank and insulation cover: 2-mm aluminum Design pressure: 7 bar, hydraulic test pressure: 9.1 bar LNG cargo conditions (before accident): temperature under −160°C, pressure just under 1 bar Five safety valves—vapor space: 2 × 1 in. at 7 bar, 1 × 0.75 in. at 9 bar; liquid pipe: 2 × 0.5 in. at 10 bar No manhole Heat sources (other than LNG cargo): diesel fuel tank capacity of 0.5 m3, tanker tires, aluminum cover, and cab materials

estimated a surface emissive power (SEP) of 260 kW/m2, and this correlated well with the first- and second-degree burns suffered by two bystanders located 200 m away. 10.4.2

Applicability of BLEVEs to LNG Marine Vessels

Pitblado et al. (2007) show that of the two main designs for LNG vessels, membrane and spherical tanks both incorporate many additional barriers to BLEVE compared to an LNG truck. A simple count of barriers in Table 10.7 shows some comparisons. In summary, an LNG tanker truck has three barriers, while both LNG marine vessel types have seven barriers. Every one of these barriers contributes to reducing the risk. A further protection is the limit on the internal tank pressure to 0.28–0.30 barg—it will be shown later this is a major limitation preventing LNG cargo flash and hence BLEVE potential. The initiating events and durations are different and the quality of the barriers is not identical, but they do give an indication that the transport truck compared with the marine LNG vessel is quite different.

BLEVES, FIREBALLS

Table 10.7

295

Comparison of protective barriers: LNG truck and LNG marine vessels

LNG truck

LNG Membrane Vessel

LNG Spherical Vessel

Threat: external fire

Two threats can cause BLEVE: pool fire or jet fire at base, or jet fire from relief valve

Same as membrane vessel

Barrier 1

Insulation cover: 2-mm Al (Europe) or steel (United States) Insulation (usually combustible)

One threat: pool fire on water Jet fire from relief valve cannot lead to BLEVE on its own. Hull plating: 20+ mm steel

Air gap: 2000 mm (maybe water if ballast tank) Hull plating: 15- to 20-mm steel, also steel plate above top of tank Insulation: 0.3-m plywood box filled with perlite beads Stainless steel membrane about 1-mm thick Insulation: 0.23-m plywood box filled with perlite beads Stainless steel membrane: about 1 mm 0.28–0.30 barg

Sphere is insulated

Barrier 2

Barrier 3

Steel wall: 4–6 mm

Barrier 4

—

Barrier 5a

—

Barrier 6

—

Barrier 7

—

Maximum pressure at failureb

Up to 8-barg pressure

a

Same

Same, but different geometry

Air gap due to tank curvature: 500–5000 mm Skirt plating: steel 10+ mma

Insulation: combustible

Tank wall: aluminum 30–65 mm 0.28–0.30 barg

The situation differs a little for a Moss LNG tank as barrier 5 applies only for heat entering through the top, but this heat is less (radiant only, not in contact with flames). b This assumes operational pressure relief, but relief valves are of standard design used in the industry where relief valves are extremely reliable, and all LNG tanks have multiple reliefs.

296

OTHER LNG HAZARDS

Pitblado et al. (2007) summarized fire threats for typical marine events. These are of a similar order to the Sandia report (Hightower et al., 2004) but are a little smaller in hole dimensions. The duration of exposure is at least as important for thermal escalation as dimensions. Large hole spill events are likely only to be short lasting—a 750-mm-diameter hole can empty a tank to the level of the hole above the waterline in 2.2 h, whereas a 5-m-diameter hole might discharge its contents in only 2–3 min. Also, the larger fires will most likely be smoky and will burn as a mass flame, with shorter flames than traditional pool fire flame height assumptions. Both of these would significantly reduce thermal escalation potentials. Can Fire Impact to Ship Lead to BLEVE? Assuming that the fire can breach all the ship’s structural barriers and that emergency response is not effective (e.g., continuing to move the vessel would greatly reduce the impact of the burning pool), calculations can demonstrate whether BLEVE is a possible outcome. The calculation here assumes that in a BLEVE, there is an instantaneous failure of the LNG tank shell. This analysis assumes all barriers fail somehow—this is unrealistic, but it allows this scoping calculation. The hull structure prevents 100% envelopment, and the pool fire will only be active on one side of the vessel. The exposed portion of the tank would be well below 10% of the tank area—this is considered in the relief system design as per the International Gas Carriers (IGC) Code (IGC, 1975; see section 3.1). Heat from the pool fire, once in contact with the unprotected LNG tank wall, would enter the tank and cause the LNG to boil locally. Over a short period of time, this will be sufficient to lift the LNG tank pressure relief valves set for 0.25 barg. Allowing a tank maximum pressure of 0.28–0.30 barg, the calculated flash amount on relieving to atmospheric pressure would be 2.4% (i.e., 2.4% of the slightly pressurized LNG liquid would flash to vapor). In practice, much less than the whole tank would flash due to the hydraulic head of the LNG liquid. Commercial LNG has a range of densities depending on the quantity of ethane and heavier components (see Table 1.1 in Chapter 1 for density values). For conservatism, a lower-range density for nearly pure methane is used. For this, each 1 m of liquid exerts a static pressure of 0.045 barg. Thus, at a depth of about 6 m, the pressure rises to 0.28 barg and no flash will occur. Thus, the flashing is restricted to the top 6 m and averages half the flash at the surface, or 1.2% (i.e., 2.4% at the surface and 0% at a 6-m depth). In mass terms, 1.2% of the top 6 m corresponds to 38 t of LNG and might create a sphere of 100% methane vapor about 34 m in diameter. This is similar in scale to a transport accident and would create a hazard zone of 200 m, less than the hazard zone from the preexisting pool fire. Figure 10.12 shows the process. This is about the same size as the tank itself and is the same order of size as a propane tanker truck, and its consequences might be expected to be similar in terms of thermal radiation from the immediately ignited gas puff.

297

BLEVES, FIREBALLS

Relief valve set point 0.25 barg

LNG vents to atmosphere

1. Roof fails 2. Large “puff” of methane vapor 3. Immediate ignition Radiant heat

Vapor space

35-m LNG depth LNG liquid column In tank Initially—161°C Finally—158.5°C

Radiant LNG flash: Top—flash vapor 2.4% Radiant heat in through damaged insulation

6m

Mid—flash to vapor 1.2% Limit of flashing to vapor

LNG liquid column In tank Initially—161°C Finally—158.5°C

Below: No flash to vapor due to hydrostatic pressure of liquid column

Undamaged Insulation / tank bottom Figure 10.12

LNG heating scenario (assuming all vessel barriers are defeated).

There should be virtually no overpressure as there is no massive tank contents flash expanding at near sonic velocity. There is no confinement at the top of the vessel that could lead to what is known as a partially confined VCE. Spherical tanks might in theory be subject to a large BLEVE event if their relief system failed as the spherical tank has greater internal strength to be self-supporting, but this would require multiple simultaneous failures of the double or triple relief system, conceivable but virtually unheard of in the process industry. Also, tanks are interconnected in the vapor space for boil-off gas collection, and this further protects against a relief failure by allowing venting though other tanks. Thus, BLEVE might be considered possible, but in risk terms, because it requires multiple simultaneous failures or errors, it would be considered negligible. Thus, LNG fire balls are discussed here for the second situation. 10.4.3

Fireballs from Released Vapor

The situation leading to a the second type of fire ball is quite different from those leading to a BLEVE, since for a bursting vessel, BLEVE momentum forces dominate and the duration of the event, td, is proportional to the mass of fuel, M, according to td αM 1 3, whereas for a vapor cloud fire ball, buoyancy forces dominate, and

(10.7)

298

OTHER LNG HAZARDS

Table 10.8

Some correlations for hydrocarbon fireball diameter and durationa

Duration, tp (s)

Material

Reference

5.35 M0.333 6.36 M0.325

— 2.57 M0.167

Propane Hydrocarbons

5.25 M0.314 5.8 M0.333 5.88 M0.333

1.07 M0.181 0.45 M0.333 1.09 M0.167

n-Pentane Hydrocarbons Propane

5.72 M0.303

0.45 M0.333

Butane

5.33 M0.327

0.923 M0.303

Hydrocarbons

6.48 M0.325 5.5 M0.333

0.852 M0.26 0.38 M0.033

LPG Hydrocarbons

Hardee and Lee (1973) Fay and Lewis (1977) Lab scale test Hasegawa and Satob (1977) Roberts (1981/1982, 1982) Williamson and Mann (1981) [W&M] Lihou and Maund (1982) [L&M] Moorhouse and Pritchard (1982) [M&P] Pietersen (1985) Marshall (1987)

Diameter, Dp (m)

a

Sources: Lihou and Maund (1982), Bagster and Pitblado (1989), Satyanarayana et al. (1991) and original papers. b The authors’ earlier correlation (Hasegawa and Sato, 1977) was D = 5.28 M0.277.

td αM 1 6,

(10.8)

giving considerably lower durations. Because of the importance of fireballs, a number of models have been developed, as assembled by Lees (Mannan, 2005, Section 16.15). A summary of these models is included in Table 10.8. A model that applies to LNG spills is that of Fay and Lewis (1977), who give the following correlations for a fireball starting from an assumed spherical volume of unburned pure fuel vapor, Vf, of radius r. The fuel volume is enlarged by air entrainment by a factor, β. The volumetric rate of growth of the fireball with radius r is taken as proportional to the product of the local surface area and the local rise velocity, dz/dt: d ⎛ 4 3⎞ dz πr = β 4 πr 2 . ⎝ ⎠ dt 3 dt

(10.9)

Integrating this equation with the boundary condition that r = 0 at z = 0 gives the elevation of the center of the fire ball, z, with time as the fireball volume increases with time: z (t ) =

1 ⎛ 3V ( t ) ⎞ . β ⎝ 4π ⎠ 13

(10.10)

Equating the forces of buoyancy driven by the difference between ambient air density, ρa, and the density of combustion products, ρb, and the rate of change of vertical momentum gives

BLEVES, FIREBALLS

d dt

⎡ 4 πr 3ρ ⎛ dz ⎞ ⎤ = 4 πr 3 g (ρ − ρ ) . a b b ⎢⎣ 3 ⎝ dt ⎠ ⎥⎦ 3

299

(10.11)

Momentum is the rise velocity, dz/dt, time the mass of the fireball, or volume, V(t), times the density, ρb. Equation 10.11 can be integrated to obtain the model for fireball duration, t, at which the fuel is all consumed: 12

16

⎡ 14 ⎤ ⎛ 3Vf ⎞ t=⎢ ⎟ . ⎥ ⎜ ⎣ βg ′ ⎦ ⎝ 4 π ⎠

(10.12)

The term g′ is the gravitational acceleration modified by the fireball buoyancy (or temperature): ⎛ρ ⎞ ⎛T ⎞ g ′ = g ⎜ a − 1⎟ = g ⎜ b − 1⎟ . ⎝ Ta ⎠ ⎝ ρb ⎠

(10.13)

Any combustion is accompanied by expansion. The expansion ratio, E, is the ratio of the final fireball volume, Vb, to the initial volume of unburned fuel, Vf. This is also the ratio of densities of unburned to burned fuel. This is also the ratio of the temperature of the burned to unburned fuel, Tb/Tf, and the ratio moles of combustion products per mole of fuel, mb/mf: E=

Vb ρ f Tb mb = = . Vf ρb Tf m f

(10.14)

The ratio mb/mf is given from the stoichiometric equation for combustion of hydrocarbon fuel with the general formula CnHm. The combustion equation relates the moles (or volume) of product to fuel for a wide range of air-to-fuel ratios by letting each molecular formula represent the moles of each component: ast ( aO 2O2 + (1 − aO 2 ) N 2 ) = aO 2 ⎧ nCO + m H O + ψ ast (1 − a ) N + ( ψ − 1) a O o2 st 2 2 2 2 ⎪⎪ 2 aO 2 ⎨ ⎪(1 − ψ ) Cn H m + ψ ⎡ nCO2 + m H 2O⎤ + ψ ast (1 − ao 2 ) N 2 ⎢⎣ ⎥⎦ ⎪⎩ 2 aO 2

Cn H m + ψ

if ψ ≥ 1 if ψ < 1. (10.15)

The combustion equation is simplified by writing it for dry air consisting of two components, oxygen and nitrogen. The proportion of oxygen in air is 0.20968 or roughly 21 mol %, denoted aO2, and the mole fraction of nitrogen is then (1 – aO2) or 0.79032. This later value is slightly higher than the actual nitrogen concentration in air because it accounts also for minor constituents such as carbon dioxide and argon.

300

OTHER LNG HAZARDS

The stoichiometric moles of oxygen is that value needed to completely burn a mole of fuel: ast = n +

m . 4

(10.16)

The stoichiometric concentration in mole fraction of fuel, yst, is given by yst =

1 , 1 + 4.773ast

(10.17)

which for methane is 0.0948 mole fraction. The fuel-to-air equivalence ratio, ϕ, is defined as the ratio of the stoichiometric moles of air, ast, to the actual moles of air. The reciprocal equivalence ratio, ψ, or the actual moles of air to the stoichiometric, is often preferable. When ψ is less than unity, all of the oxygen is consumed, but there is (1 – ψ) fraction of the fuel left unburned. Conversely, when ψ is larger than unity, there is excess air, so some oxygen remains unconsumed in the combustion products. The mole ratio mb/mf is readily found from the combustion equation, Equation 10.15, by adding the moles on the right-hand side in the numerator and on the left-hand side in the denominator giving

mp = mf

n+

mp = mf

m (1 − aO 2 ) + ( ψ − 1) ast + ψast 2 aO 2 a 1 + ψ st aO 2

if ψ ≥ 1, and

m⎞ (1 − aO 2 ) + ψast 2⎠ aO 2 ast 1+ ψ aO 2

(1 − ψ ) + ψ ⎛⎝ n +

if ψ ≤ 1.

(10.18)

(10.19)

For pure methane, n = 1, m = 4, ast = 2, and both Equations 10.18 and 10.19 reduce to mb/mf = 1. From soap bubble experiments, Fay and Lewis find an entrainment coefficient of β = 0.285, which is in reasonable agreement with values given by Morton et al. (1956). For the equivalence ratio, the average value is ϕ = 0.271 or ψ = 4.61. The temperature of combustion products in Equation 10.14 can be found using an energy balance. Each term in Equation 10.15 is replaced by the corresponding enthalpy term and is solved by trial and error for the flame temperature (neglecting radiant energy). The results of this approach is Figure 10.13. A computer program produced for the National Aeronautics and Space Administration (NASA) by Gordon and McBride (1996) calculated the curve,

BLEVES, FIREBALLS

301

2300

Flame T (K)

2100 1900 1700 1500 5

6

7

8

9

10

11

12

13

14

15

Mole % fuel GordonMcB Figure 10.13

Lewis and von Elbe

Adiabatic flame temperature for methane.

and a measured value at 10% methane (near the stoichiometric value of 10.48) is provided by Lewis and von Elbe (1987, p. 717) as a value of 2148 K. With an ambient temperature of 300 K (27°C), the maximum temperature ratio, Tp/Ta, found is 7.16. Since the mole ratio of products to fuel is unity in this case, the maximum volume expansion upon combustion in Equation 10.14 is also 7.16. Considering that about 30% of the energy of combustion is radiated, the actual plume temperature and expansion ratio will be lower. For a pure methane cloud at standard ambient temperature and pressure, using the values of β = 0.285 and ϕ = 0.271, the model equations above reduce to the following dimensional formulas for the time of complete combustion, tp, in seconds, the height of the fireball as it consumes the last fuel, zp in meter, and the final fireball diameter, Dp, in meter (Raj, 2006, p. 48): t p = 3.02 M 1f 6,

(10.20)

zp = 1.05M , and

(10.21)

Dp = 7.37M .

(10.22)

13 f

13 f

Different investigators have obtained different coefficients for the basic equation form t p, zp, Dp = aM n,

(10.23)

as summarized in Table 10.8 from Lees (Mannan, 2005, p. 16/180). The predictions from some formulas in Table 10.8, along with Equations 10.20 and 10.22, are shown in Figures 10.14 and 10.15. The abbreviated legends in Figures 10.14 and 10.15 refer to corresponding references in Table 10.8.

302

OTHER LNG HAZARDS

140

Fireball diameter (m)

120 100 80 60 40 20 0 0

2000

4000

6000

8000

10,000

Mass of fuel (kg) H&L,73

F&L,77

H&S,78

Rob,81

L&M,82

M&P,82

Piet,85

Mrshl,87

Burn duration (s)

Figure 10.14

30 25 20 15 10 5 0 100

W&M,88

Predicted fire ball diameter at the end of burn.

1000

10,000

100,000

Fuel mass (kg) T&MS, LNG W&M, C3

M&P, HC Roberts,HC

L&H, C4

Figure 10.15 Predicted fire ball burn duration.

Predictions by Equations 10.20 and 10.22 are labeled T&M-LNG. The model predictions diverge for large values of the fuel mass, M. The fireball duration predictions are seen to be quite short, which limits the potential hazard.

10.5 LNG VAPOR CLOUD EXPLOSIONS This section discusses one type of explosion, a vapor cloud explosion (VCE), formed from the release of an evaporating liquid or gas that disperses and is ignited. A VCE explosion is characterized by occurring outdoors with widely distributed fuel so a reacting wave passes through the fuel and often increases

LNG VAPOR CLOUD EXPLOSIONS

303

in speed over a “run-up” distance. A vapor explosion inside a building or a tank is considered an enclosed explosion. Other types of explosion, such as those of high explosives, or condensed-phase explosions, provide more of a point source for overpressure and energy release. 10.5.1

Characteristics of Detonations and Deflagrations

VCEs became widely recognized as a process hazard following the Flixborough explosion in 1974. Currently, two main approaches are adopted, either an energy based approach using correlations or a CFD approach. In general terms, the onshore process industry tends to use the correlation models and the offshore industry uses the CFD approach, especially for deflagrations. Detonation modeling is rarely carried out. Of the two types of VCEs, detonations and deflagrations, detonations are by far more damaging. Detonations occur at supersonic to sonic speed relative to the unburned fuel and the pressure pulse produced is a shock wave (near instantaneous pressure rise). Shock wave velocities are often 1500 m/s or higher. Consequently, the reacting wave front perpetuates the reaction and burns through the entire vapor cloud, not just the concentrations between flammable limits. Generally though, detonations require a strong source of ignition. Once the composition of a material is defined, detonation properties are uniquely determined. Detonations normally occur with highly reactive materials such as acetylene, hydrogen, and ethylene. Methane and LNG mixtures are considered of low to medium reactivity and do not detonate. The pressure pulse from a deflagration has a finite rise time before reaching a peak (on the order of 10 ms) and travels at subsonic speeds (typically under 250 m/s), and the pressure wave precedes the reaction wave by a small interval. Deflagrations are more complex because the flame speed and overpressure are not determined by composition alone but by factors that lead to flame acceleration, namely, congestion, confinement, and fuel reactivity. Early experiments performed by igniting stoichiometric concentrations of completely unconfined gas developed only low flame speeds with no flame acceleration and only minor overpressures (Harris and Wickens, 1989). It is now recognized that a deflagration develops only when the LNG vapors are either partially confined (e.g., inside a module or by a concrete roof overhead) or are within a highly congested region where flame speeds are accelerated by flowing around numerous obstacles (Bjerketvedt et al., 1997). The burning of gas, liquid, or solid in which fuel is oxidized involves heat release. Combustion of methane (CH4) in air can be described by the overall chemical equation CH4 + 2 (O2 + 3.76 N 2 ) → CO2 + 2H 2O + 2 ( 3.76 N 2 ) + energy The chemical products from complete combustion of a hydrocarbon fuel are mainly CO2 and H2O (vapor) plus unburned nitrogen. A count of moles shows

304

OTHER LNG HAZARDS

10.5 before and 10.5 after. The combustion process will result in substantial increased temperature due to the transformation of chemically bound energy into heat, and this results in a major volume increase (up to eight times). It should be emphasized that the above equation constitutes a major simplification of the real combustion process where free radical reactions play a dominant role and soot formation and combustion generates much of the radiation. Based on the above, it can be seen that in a completely confined space, even if no flame acceleration occurs, combustion heat alone can generate a pressure of 8 bar. A parameter of deflagrations is the dynamic pressure, resulting from the induced air movement, or the “blast wind.” Its magnitude is proportional to the square of the air velocity and to the density of the air behind the blast wave (HSE, 2005). Flame speeds vary with fuel composition and with flame accelerating turbulence. Detonation velocities, and hence overpressures, tend to decay when the flame passes outside of a congested region. They can reaccelerate upon entering a new congested zone. Deflagrations can be ignited by a relatively weak ignition source. The vast majority of VCEs are deflagrations. In general terms, deflagrations cause lower overpressures than detonations. However, the recent Buncefield explosion in the United Kingdom in 2005 is now regarded as a deflagration-to-detonation transition (DDT) event. The congestion in this case was, surprisingly, a row of trees. An exception must be noted to the generalized capsule description above. Combustion can be considered a large-scale free radical reaction. That is, fuel is oxidized in steps that involve removing portions of the fuel molecule generating charged fragments called free radicals. A jet of flame contains relatively large concentrations of free radicals and serves as a strong ignition source because of this fact. A jet of flame injected into a mildly confined or congested vapor cloud of even low reactivity material can develop a deflagration. The effects of a deflagration include •

• • •

a radiating overpressure wave that can damage walls, structures, and displace objects; a sizeable fireball; a fire updraft that can displace light material such as insulation; and secondary fires and releases by displacing pipes, pumps, and other structures.

Confinement refers to surfaces such as walls and ceilings that limit the expansion of a reacting wave to two or one dimension, referred to as 2-D or 1-D expansion. A solid surface such as a trench wall or a platform floor is considered 2-D confinement. A pipe has 1-D expansion. A vapor cloud ignited in a tank farm is not considered confined by the tank walls, so this is 3-D expansion. By extension, a vapor cloud of LNG on the sea beside the tall side of an LNG carrier would be considered in a 3-D expansion environment. An LNG

LNG VAPOR CLOUD EXPLOSIONS

305

Leading flame front

Vu

· Rf

Combustion products

Vb ~ 0

Flame front in trapped pocket

Standing eddy Shear layer

Figure 10.16 Schematic illustrating flame propagation over obstacles (Moen et al., 1980) (reproduced by permission from Elsevier Science Publishing, Inc.).

vapor cloud within an LNG terminal (either a liquefaction plant or a regasification terminal) could possibly envelop a 2-D confinement area. Congestion refers to obstacles in the path of a reaction wave that generates turbulence. Turbulence creates a folding and wrinkling of the flame surface that accelerates the flame speed. A schematic showing how repeated obstacles generate turbulence is provided in Figure 10.16 by Moen et al. (1980). Basically, a flow field is separated into a standing eddy by an obstacle. The flow field reattaches at a distance downstream of the obstacle. But if the obstacles are placed close enough together, they interfere before there is any reattachment. As a flame encounters this flow field, it becomes “stretched” with a large-scale flame fold consisting of a leading flame front in the outer region with a trailing flame in the trapped pockets between obstacles. Methods to define the degree of congestion and confinement, correlated with flame speed in explosion models, have been developed by Baker et al. (1994), van den Berg et al. (1993), Eggen (1998), Mercx et al. (1998), and Puttock (1995, 1999). To be precise, there are several defined flame speeds. The burning velocity, Su, is the speed of the flame with respect to unburned gas. Values of Su measured in the laboratory are reported in handbooks. The laminar burning velocity, Sb, is the product of the burning velocity and the expansion ratio (see Eq. 10.14): Sb = Su

ρf . ρb

(10.24)

Turbulent flame speed is the velocity of a flame front with respect to a fixed observer and is usually many times the laminar burning velocity. Historically, deflagrations occur commonly from a pressurized leak, where turbulence is developed during the release. On the other hand, there are no cases of “refrigerated atmospheric pressure storage of liquid flammables pro-

306

OTHER LNG HAZARDS

ducing any VCE incidents, which can be attributed to poor mixing of cold released vapors with air” (HSE, 2005). 10.5.2

Fuel Reactivity Effects

The reactivity of a fuel is defined by its tendency to attain high flame speeds. For example, an experiment by Mercx (1992) compared flame speeds of methane, propane, and ethylene for various types of congestion to obtain the results in Figure 10.17. In this experiment, the obstacle array was vertical cylinders set at a pitch of 3 : 1. Ethylene is considered a high reactivity fuel, as born out by the high flame speed shown. The dip in the ethylene flame speed occurred when the flame broke through the confining volume. Methane is considered a low reactivity fuel, and for all other tests in this series, methane showed lower flame speeds than propane. Various correlations have been proposed to identify the reactivity of components for which flame speed measurements are lacking. Baker et al. (1994) classify reactivity by the fundamental burning velocity, Su, as tabulated in Table 10.9. According to U.S. NFPA 497, the U.S. National Electrical Code (NEC) began defining the reactivity of gases in 1935 based on the level of explosion pressures and the likelihood that the effects of an explosion could be transmitted outside the enclosure. Four gas groups, groups A, B, C, and D, were originally defined using acetylene, hydrogen, ethyl ether, and gasoline as the “border” materials. In the 1950s, a test apparatus was developed to measure how various materials behave with respect to explosion pressure when ignited in the test vessel. The apparatus was the Westerberg Explosion Test Vessel, and it measured and documented a factor called the “maximum experimental safe gap” (MESG). In 1971, the International Electrotechnical Commission (IEC) published IEC 79-1A, defining a different apparatus for obtaining MESG results (Lunn, 1982). Table 10.9

Flammable material reactivity by fundamental burning velocity, Su

Material High Reactivity: Su > 75 cm/s Acetylene Ethylene oxide Hydrogen 1,3 Propylene oxide Medium Reactivity: 45 ≤ Su ≤ 75 cm/s Acetone 1-3 Butadiene Propane Low Reactivity: Su ≤ 45 cm/s Methane Carbon monoxide

Formula

Su (cm/s)

C2H2 C2H4O H2 C3H6O

157.0–173.0 89.5–108.0 265.0–320.0 67.0–82.0

C3HhH6O C4H6 C3H8

45.7–54.0 52.0–64.0 39.0–46.0

CH4 CO

33.8–40.0 39.6–46.0

Flame speed (m/s)

LNG VAPOR CLOUD EXPLOSIONS

307

P = 60 BR = 50%

600 500

Ethylene

400 300

Methane 200 Propane

100 0

0

2

4

6

8 10 Distance (m)

12

Figure 10.17 Flame speed measured for three fuels in an array of cylinders at a 3-D pitch (Mercx, 1992).

As a practical application, the U.S. Coast Guard uses MESG to specify flame arrestors for ship cargoes (specified by law in Appendices A and B to Part 154 of Title 33 CFR). Flame transmission is prevented for gap widths ≤MESG. The MESG approach has become so widely used that flame arresters are described in terms of the NEC groups. The current demarcation has become • • • • •

Group A—acetylene, Group B—hydrogen, Group C—ethylene, Group D—propane, and Flame arrester not required if MESG > 0.9 mm

Table 10.10 lists the MESG values that define the border materials for the gas groups. A material is placed in the group for which its MESG value is lower than the border values listed. Britton (2000) compiled the values listed in Table 10.10 citing the National Academy of Sciences (NAS, 1975) and NFPA 497. Clearly, methane qualifies by a wide margin as one of the least flamereactive materials. In addition, the National Materials Advisory Board (NMAB) (1982) tabulated a comprehensive set of materials used industrially and assigned a group

308

OTHER LNG HAZARDS

Table 10.10

MESG values by hazardous gas group

Group

Material at Lower Border of Group

A B C D Unclassified For comparison

Acetylene Hydrogen Ethylene Propanea MESG = 0.90 Methane

MESG (mm) Westerberg NAS (1975)

Westerberg British

NFPA 497

0.08 0.08 0.69 0.94 — 1.12

0.25 0.28 0.69 0.96 — 1.17

0.25 0.28 0.65 0.97 — 1.12

a

Commercial.

classification to them based on their measured or estimated MESG values. Each of the components of LNG is listed in the NMAB compilation as Class D. This includes methane, ethane, propane, isobutene, n-butane, isopentane, and pentane. 10.5.3

Modeling VCEs

The earliest approach to explosion modeling, the TNT equivalence model, is based solely on the chemical energy available, translates this into an equivalent amount of TNT, and uses the TNT pressure–distance curves as the basis for prediction. It does predict reasonably well for far-field effects of gas explosions. However, the TNT model uses incorrect physics for VCEs as it is based on a detonation of a condensed-phase solid material, and it takes no account of local congestion or confinement. Since local overpressure impact effects are important, the TNT method has tended to fall out of use. In terms of modeling a VCE event, currently, the two main approaches are the Baker–Strehlow–Tang (BST) model (Baker et al., 1994, 1997) and the TNO (2005) multi-energy model. The former is more commonly used in the United States and the latter in Europe. Both share many features. CCPS (1999, pp. 116–120) provides a short summary of these methods, and more detail is available in CCPS (2009b). The TNO model originally developed by Wiekema and later updated by van den Berg (1985) assumes • • •

•

no detonation; overpressure is greatly affected by degree of congestion; overpressure is not much dependent on the material in the cloud except for higher energy/flame speed materials such as hydrogen, ethylene, and acetylene; the blast originates only in areas of congestion, and flash fires in other parts of the cloud can be ignored as contributing to overpressure;

Dimensionless maximum “side on” overpressure (ΔPs)

10

10 9 8 7 6

1

5 4 3

0.1

2 1

0.01

0.001 0.1

Dimensionless positive phase duration (t+)

LNG VAPOR CLOUD EXPLOSIONS

309

10 5

1 2 3 4

1

5

0.5

6 7 8 9 10

0.1 0.1

0.5

1.0

RO P

5.0 10.0 50.0 100.0 Combustion energy-scaled distance (R)

ΔPS t c R ; t = + o ; R= PO + (E/PO)1/3 (EPO)1/3 PO = atmospheric pressure cO = atmospheric sound speed Time E = amount of combustion energy RO = charge radius ΔPS =

1.0 10.0 100.0 RO Combustion energy-scaled distance (R) t+

Figure 10.18 TNO multi-energy model overpressure curves (van den Berg, 1985) (reproduced by permission from Elsevier Science Publishers, Inc.).

• •

the model uses a family of possible overpressure curves; and blast at a distance follows scaled distance from one-third power law.

Detailed rule sets are available in the TNO Yellow Book (TNO, 2005) and in CCPS (2009b). The model uses a standard set of curves, labeled 1–10, to represent greater energy. TNO initially suggested using curve 10 for congested plants, but this was seen as too high. In most congested process plants (e.g., LNG liquefaction plants), TNO 7 would be used. The standard set of TNO curves for normalized distance is provided in Figure 10.18 as scaled overpressure and scaled impulse against normalized distance, defined in the figure; SI (System International) units are used. As can be seen from Figure 10.18, beyond the near field, all the higher TNO curves (6–10) combine to a single curve that approximates the TNT one-third power law equation. TNO 7 has a peak pressure of 1 bar. The figure predicts positive phase duration, which allows an impulse to be determined and hence a structural load. The BST curves in Figure 10.19a,b use as a parameter on the curves the flame speed, Mw, in the units of Mach number (uf/a0), where uf is the flame speed in meters per second. Suggested values of the Mach number are tabulated as a function of congestion and confinement. Judgment comes into assigning degrees of congestion and confinement to process areas. Detonations are represented by a Mach number above unity (supersonic).

310

OTHER LNG HAZARDS

100.0 Mw = 5.2 Mw = 4.0 Mw = 2.0 Mw = 1.0 Mw = 0.50 Mw = 0.25 Mw = 0.125 Mw = 0.074 Mw = 0.037

10.0

Ps /po

1.0

0.1

0.01

0.001

0.0001 0.01

1.0

0.1 1/3

Rpo /E

(a)

10.0

1/3

1.0 Mw = 5.2 Mw = 4.0 Mw = 2.0 Mw = 1.0

(iao)/(po2/3/E1/3)

Mw = 0.50

0.01

0.001 0.01 (b)

Mw = 0.037–0.25

0.1

1.0

0.1 1/3

10.0

1/3

Rpo /E

Figure 10.19 (a) Scaled overpressure curves by the BST method (Baker et al., 1994, 1997). (b) Scaled impulse curves by the BST method (Baker et al., 1994, 1997).

In both Figures 10.18 and 10.19a,b the scaled overpressure is the maximum side-on peak overpressure, ΔPs, normalized to atmospheric pressure P0. Impulse, i+, is defined as the integral under the positive overpressure curve. Assuming the positive overpressure curve is triangular, the integral under the overpressure curve of duration td is approximated by

LNG VAPOR CLOUD EXPLOSIONS

i+ =

1 ΔPs td . 2

311

(10.25)

Impulse has units of Pa·s or psi·ms, which, in SI units, is equivalent to joule per square meter or the energy per area imparted by an explosion wave upon a surface such as the wall of a building. Building damage is found as a function of both overpressure and impulse by modeling the deflection response of walls to an explosion wave. The BST method defines a dimensionless impulse in terms of the sonic velocity, a0, and an energy scaling term, E1/3P02/3, where E is the fuel mass within the flammable limits and also within the congested zone, mf, times the combustion energy, so the dimensionless impulse is i+ =

ia (E P

+ 0 13 23 0

)

.

(10.26)

The TNO method defines instead a dimensionless duration for the positive overpressure pulse. This is defined by a slightly different energy scaling (E/P0)1/3, in meter: td =

t + a0 . ( E P0 )1 3

(10.27)

In both models, scaled distance is the ratio of radial distance R, in meters, by the energy scaling term R=

10.5.4

R ⎛E⎞ ⎜⎝ ⎟⎠ P0

13

.

(10.28)

CFD Modeling of VCEs

The use of CFD models for VCEs is increasing in both onshore liquefaction plants and in offshore LNG applications. CFD tools are able to address specific geometries, whereas the BST and TNO models described previously are closer to free-field models correlating congestion to relate with overpressure. While in principle any CFD code could be used for explosion simulation, there are so many special issues to address that the common approach is to use explosion-optimized codes that have been validated for the application. Lea (2002) summarizes several CFD codes used for VCEs including Exsim, FLACS, CFX, and COBRA. These all are used in offshore applications in the North Sea where large quantities of methane and other hydrocarbon gases can occur in severely congested or confined spaces.

312

OTHER LNG HAZARDS

All CFD approaches solve the fundamental fluid flow Navier–Stokes equations. Lea notes that Exsim and FLACS are simpler CFD formulations with first-order formulations in some of their dimensions, whereas CFX and COBRA are higher order in both temporal and special dimensions. The implications of this are numerical—the simpler models are affected by numerical dispersion that can exceed true dispersion. All models have been verified against medium- and large-scale explosion trials. FLACS did best in the Steel Construction Institute blind trials (tests were computed before the explosion experiments were run) and for this reason, it has wide usage in offshore applications. The combustion models are different—FLACS and COBRA use empirical correlations; Exsim and CFX use an eddy break-up model. CFX also employs a thin flame model. The grid size used for modeling the flow regime, typically of the order of 0.5–2.0 m, is an order of magnitude larger than the turbulence generating elements (e.g., small-diameter pipework of 0.05–0.1 m). Since flame acceleration occurs by turbulence, different empirical models must be used for the explosion—Exsim and FLACS use the porosity/distributed resistance (PDR) approach model. The FLACS code is widely used and is a good example of a commercial CFD code that is steadily increasing its areas of application, including direct validated applications to LNG and hydrogen. FLACS computes initially the dispersion of flammable gas from a release location and determines a time-dependent flow field. The subsequent explosion code can be ignited at different locations, and this will lead to differences in predicted maximum overpressure. Thus, several runs should be made to account for possible ignition locations. A typical visual representation of the explosion results is provided in Figure 10.20 from Hansen et al. (2005). Note that in this figure, predicted pressures decline in sequence: darkest, lightest, medium dark, medium light. Because CFD codes generate so much output data and are affected by so many parameters, it is common to present CFD results in probabilistic formulations rather than as a single discrete overpressure prediction. The Norwegian offshore safety standard NORSOK Z-013 (2001) specifies a cumulative

Figure 10.20 Flammable gas dispersion (left) and overpressure prediction (right)—from FLACS (Hansen et al., 2005).

ASPHYXIATION AND CRYOGENIC HAZARD FROM LNG SPILLS

313

frequency versus overpressure exceedance curve for which the safety function (e.g., refuge, escapeway) must be demonstrated at a frequency of 10−4 pa (1 in 10,000 years). It is likely that more use will be made of CFD codes in the future as they can account specifically for geometry and can predict local zones of the highest pressure that require engineering to reduce the congestion or to increase the structural strength. The former is a prevention strategy and is preferred over a mitigation solution. Lea (2002) notes several areas where CFD codes need improvement and he speculates this will take more than a decade.

10.6 ASPHYXIATION AND CRYOGENIC HAZARD FROM LNG SPILLS LNG is extremely cold and expands upon vaporizing into a large volume of vapor. If an LNG spill does not ignite, it poses an asphyxiation or freezer-burn risk to personnel very near the spill, for example, on unloading platforms or to emergency response personnel. Table 10.11 (Shaikh and Porter, 2008) summarizes the physiological effects at various minimum oxygen concentration (MOC) levels. Table 10.12 lists suggested risk end points and an estimated probability of impairment from diminished oxygen levels. A mole balance used to relate MOC (the mole fraction yO2) to the mole fraction of LNG vapor, yLNG, is yO2 = 0.209 (1 − yLNG )

(10.29)

LNG liquid and evaporated vapor immediately above the pool are at a constant temperature at the normal boiling point listed in Chapter 1, Table 1.2 (typically −160°C, −260°F). Upon mixing with air, the temperature increases, and above approximately −113°C (−171°F), LNG vapor is lighter than air. This is considered a cutoff point for vapor affecting impairment. Direct contact with metal at cryogenic temperatures can damage skin tissue more rapidly than exposure to cold vapor. This can be a factor with platform ladders and handrails used in access and escape. Physiological criteria for the onset of cold pain, numbness, and frostbite are a function of the contact surface properties and time of exposure. Geng et al. (2006) provide the following for bare skin. For nonmetallic surfaces at −15°C, the onset of cold pain occurs within 5 s, and for surfaces at −35°C, the onset of numbness occurs within 15–65 s. For aluminum and steel at −15°C, skin temperatures reach 0°C within 2–6 s. Appropriate protective clothing mitigates the exposure risk to cryogenic burns, but this does not mitigate against a cryogenic vapor inhalation hazard. Recommendations are available for personal protective equipment for particular tasks, including breathing air apparatus.

314

OTHER LNG HAZARDS

Table 10.11

Summary of asphyxiation effects

MOC % by Volume in Air (at Sea Level)

LNG Vapor (Mol %)

20.9 20.9–19.5

0 0–6.7

19.5

6.7

19.5–16.0

6.7–23.4

16.0

23.4

16.0–15.0

23.4–28.2

15.0–10.0

28.3–52.2

10.0–6.0

52.2–71.3

71.3

Physiological Symptoms or Effects Normal Some adverse physiological effects, but they are unnoticeable U.S. Occupational Safety and Health Administration (OSHA) lower limit for confined space entry (29 CFR 1915.12) Increased pulse and increased breathing rate with disturbed muscular coordination Considered potentially life threatening by U.S. Department of Labor’s Mine Safety and Health Administration (MSHA, 2006) Impaired thinking and attention, reduced coordination Faulty judgment, rapid fatigue, insensitivity to pain; impaired respiration that might cause permanent heart damage; nausea and vomiting Inability to perform vigorous movement; nausea, vomiting, collapse, and permanent brain damage Convulsions, cessation of breathing, death

ASPHYXIATION AND CRYOGENIC HAZARD FROM LNG SPILLS

Table 10.12

315

LNG vapor end points for asphyxiation hazard

MOC (Mol %)

LNG Vapor (Mol %)

Escape Impairment

Impairment Probability (%)

Physiological Effects

14.0

33.3

Yes

70–99

16.0

23.8

Yes

10–50

19.5

7.1

No

1–10

Faulty judgment, rapid fatigue Impaired thinking and attention may result in faulty impairment decision. Some adverse physiological effects, but they are unnoticeable.

Table 10.13

Affected Body Personnel

Support structures, escape handrails / ladders

Suggested threshold temperatures for cryogenic spill risk criteria

Exposure Temperature (°C)

Phase

Escape Impaired?

Impairment Probability

−160

Liquid pool

Yes

99%

−160

Vapor

Yes

50% to 70%

−113

Vapor

No

Less than 10%

−73

Vapor

No

−160 to 0

Liquid pool, vapor

Yes No

Less than 1% 99% Less than 10%

−15

Vapor

No

Less than 10%

Contact Effects

Immediate liquid cryogenic burns; permanent damage to skin Cryogenic inhalation and damage to skin possible in short duration exposure Cryogenic inhalation may be possible with long exposure duration Some cryogenic inhalation Steel structure brittle failure possible with long exposure and without cryogenic passive protection; onset of bare skin numbness on nonmetallic surfaces Onset of bare skin numbness on aluminum or steel surfaces

Liquid spills based on a specified containment radius.

34 89 48

7.0 7.8.0 28.0

a

70

6.1

Low-pressure LNG from a loading or unloading arm Low-pressure boil-off LNG vessel Low-pressure LNG pump High-pressure LNG pump 8 31 17

16

MOC 16%

6 14 11

12

MOC 14%

Asphyxiation (m) MOC 19.5%

Discharge Rate (kg/s)

25-mm Diameter Leaks of LNG

Table 10.14 Asphyxiation and cryogenic hazard distances

4 5 8

4

8 25 22

23

18 38 34

37

−113°C

−160°C

−160°C

Vapor

Liquid

Cryogenica (m)

26 53 49

51

−73°C

316 OTHER LNG HAZARDS

ASPHYXIATION AND CRYOGENIC HAZARD FROM LNG SPILLS

317

In addition, structural failure associated with metal embrittlement, accidental loads, and metal fatigue can impair escape as well. Shaikh and Porter (2008) provide suggested temperatures and escape impairment probabilities for use in quantitative risk assessment in Table 10.13. To quantify typical hazardous zones with low wind speed conditions, LNG spill scenarios are modeled for small accidental releases using the PHAST 6.53® model. Table 10.14 lists the predicted distance to asphyxiation and cryogenic hazard end points. The listed distances are likely to occur well within the boundaries of LNG terminals.

11 FIRE EFFECTS 11.1

FIRE RADIATION EFFECTS ON INDIVIDUALS

The effect of thermal radiation from any fire is measured in terms of thermal radiation intensity and exposure duration, or a combination of the two, radiation dosage. Intensity is the radiant energy flux, I (in kW/m2). Exposure duration, t, is in seconds. Radiation dosage, in kilojoules (kJ/m2), is the product of intensity and duration. As a reference point, solar radiation on a clear day is about 1.0 kW/m2. Vapor cloud explosions are discussed in Chapter 10, where we indicate that from liquefied natural gas (LNG) vapor, they are unlikely. While the effect of overpressure loads on people and structures is not discussed in this text, useful guidance is available from Lees, Loss Prevention in the Process Industries, and the Center for Chemical Process Safety (CCPS) referenced below. 11.1.1

Injuries to People—Definition of Burn Degrees

There are several sources of reference materials for defining burn injury based on thermal exposure. The Lees textbook (Mannan, 2005 pp. 16/239–252) and CCPS (2000a) both provide useful summaries. Several classifications of skin burn severity have been proposed. The most familiar is to classify skin burns into three degrees, as follows: •

First Degree A mild level of skin burn affecting the epidermis, with persistent redness but no formation of blisters. More severe first-degree

LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

318

319

FIRE RADIATION EFFECTS ON INDIVIDUALS

Thermal radiation impact threshold - pain and blistering (bare skin) 120

Time (s)

100 80 60 40 20 0 0

5

10

15

20

Thermal flux (kW/m2 )

Figure 11.1

•

•

25% probability first-degree burn

50% probability first degree burn

75% probability first-degree burn

Pain threshold

Blister threshold 1.0 kW/m2

1.4 kW/m2

Thermal radiation pain and blistering thresholds for bare skin.

burns will produce some pain, but no permanent damage. Flaking or scaling of the outer skin layer will occur several days after exposure. Second Degree An intermediate level of skin burn characterized by the formation of blisters. The blister depth may be shallow (epidermis), with only the surface layers of the skin damaged, or more severe with nearly the full depth of the skin destroyed (epidermis and dermis). Third Degree Deep burns characterized by the destruction of all skin layers and by charring. The underlying tissue may also be damaged.

Modern medical treatment has tended toward describing skins burns by depth. 11.1.2

Measured Effect Levels from Radiation Exposure

Figures 11.1 and 11.2 show various effect levels from the onset of pain to thirddegree burns for the exposure of bare skin. The lowest data points in Figure 11.1 plot the time humans feel pain at various levels of thermal flux (reported by Henriques [1947], Buettner [1951, 1957], Hinshaw [1957], Stoll and Greene [1958, 1959], and Cassidy and Pantony [1988]). The blistering threshold curve plots data from Raj (1977), Health and Safety Executive (HSE, 1981b), Bennett et al. (1982), and Croce and Mudan (1986a). Several experiments have also been performed on rats and pigs to determine the thresholds of second- and third-degree burns. Data for flux thresholds that cause blistering or first-degree burns based on the above references are plotted in Figure 11.1. The three upper curves in Figure 11.1 show the 25%, 50%, and 75% probability of a first-degree burn

320

FIRE EFFECTS

Thermal radiation impact threshold—second- and third-degree burns (bare skin) 120

Time (s)

100 80 60 40 20 0 0

5

10

15

20

25

30

35

40

Thermal flux (kW/m2 ) 25% probability second-degree burn 50% probability second-degree burn 75% probability second-degree burn 2nd deg. burn threshold (0.1-mm depth) 3rd deg. burn threshold (2.0-mm depth) Figure 11.2 Thermal radiation second- and third-degree burn thresholds for bare skin.

Thermal radiation impact thresholds

Time (s)

100

10

1 10

1

100

Thermal flux (kW/m2 ) Pain threshold

Blister threshold

Second-degree burn threshold

Third-degree burn threshold

Figure 11.3

Log–log plot of burn thresholds for bare skin.

based on combined thermal radiation intensity and duration (i.e., thermal radiation dose). Likewise, the three lower curves in Figure 11.2 show the 25%, 50%, and 75% probability of the onset of second-degree burns. Figure 11.3 puts four burn thresholds on a single log–log plot emphasizing the essential similarity of responses on a logarithmic basis. As indicated by the spread in the probability curves, the data vary because of varying responses from indi-

FIRE RADIATION EFFECTS ON INDIVIDUALS

321

viduals tested (i.e., people have different pain tolerances and skin pigmentations). The probability curves are based on probit equations developed by Eisenberg et al. (1975) and by Tsao and Perry (1979). The highest heat flux that the skin can absorb during a long time without feeling pain is about 1.0 kW/m2 (317 BTU/h·ft2), which falls within the range of solar radiation (Green Book; TNO, 1992). Based on work done by Buettner (1951), unbearable pain is taken as the point where a layer 0.1 mm below the skin’s surface exceeds a temperature of 44.8°C (113°F). An exposure to a thermal flux level of 1.4 kW/m2 (444 BTU/h·ft2), no matter how long the exposure is, will not result in pain because an increase in peripheral blood flow prevents the localized temperature from reaching 44.8°C. Both the 1.0- and 1.4-kW/m2 flux levels are shown in Figure 11.1 (vertical lines). The data in Figure 11.1 for the threshold of pain are well represented by the following equation. Threshold of Pain t = [ 35 I ]

1.33

,

(11.1)

where t is the time to feel pain in seconds (Raj, 1977). Eisenberg et al. (1975) provide an equation for the threshold to first-degree burns that represents well the data in Figure 11.2 and in Table 11.1 with radiation intensity, I (in W/m2). Blisters and First-Degree Burns t = 550, 000 I 1.15

(11.2)

The 50% probability threshold for second-degree burns is approximated (for exposure time, t in s ≥ 10 s, and I in kW/m2) (Hymes, 1983, 1984; Mudan et al., 1984) by the following equation.

Table 11.1

Onset of second-degree burns (FEMA, 1990)

Radiation Flux (kW/m2) 1 2 3 4 5 6 8 10 12

Time for Pain (s)

Time for Second-Degree Burns (s)

115 45 27 18 13 11 7 5 4

663 187 92 57 40 30 20 14 11

322

FIRE EFFECTS

Second-Degree Burns I = exp ( −0.76307 ln ( t ) + 4.9545) .

(11.3)

Federal Emergency Management Agency (FEMA) (1990) provides values in Table 11.1 for the threshold onset of second-degree burns that are consistent with Figure 11.3 but are more conservative than the values in Figure 11.2. However, since 1 kW/m2 is close to summertime peak solar flux, the FEMA data appear to overestimate pain and second-degree burns for most individuals. NFPA 59A, API521, and the U.S. Environmental Protection Agency (EPA) Risk Management Plan required consequence calculations all nominate a fixed thermal flux of 5 kW/m2 as the basis for determining exclusion zones. A lesser value of 1.5 kW/m2 is specified as tolerable for emergency working. Some people have interpreted the 5 kw/m2 value as a fatal flux, but as can be seen from all the above data, it is more of an injury criterion. Raj (2008b) reports on LNG fire experiments where he exposed test mannequins and himself to direct thermal radiation mostly in the range 3.5–7.0 kW/m2. Wearing normal clothing, he reports that he could withstand the 5 kW/m2 flux level without any pain for 30 s and that the skin beneath the clothing could be exposed to 4 kW/m2 for 200 s with no measureable rise in skin temperature. 11.1.3 Thresholds of Injury on Thermal Dose Basis The concept of thermal dose has been developed to better account for the harm caused by intense, short-duration radiation. A modified thermal dose is defined by the British HSE as V = I 4 3t ,

(11.4)

with I in kilowatts per square meter, t in seconds, V is in thermal dose units (TDUs) or (kW/m2)4/3 s. In a Health and Safety Laboratory report, O’Sullivan and Jagger (2004) defined thresholds of harm in terms of thermal dose as listed in Table 11.2. This table also includes estimates for the likely percent of fatalities for people constrained in escaping a fire, as on offshore oil-producing platforms. Acceptance criteria are provided in modified TDUs under a discussion of the European standard for risk analysis in Chapter 6. The threshold for lethal thermal radiation, based on nuclear explosion tests, is given by Eisenberg et al. (1975) as Lethal Effect Threshold Y = −14.9 + 2.56 ln (V ) ,

(11.5)

where Y is the probit function for fatal response. This equation is unit dependent and the units are defined in Equation 11.4 above. The probit function

FIRE RADIATION EFFECTS ON INDIVIDUALS

Table 11.2

323

Thresholds of harm in thermal dose units

Harm Caused

Thermal Radiation Dose Units (TDU) (kW/m2)4/3 s

Pain Threshold first-degree burn Threshold second-degree burn Escape impeded Threshold third-degree burn 1–5% Fatality offshore 50% Fatality offshore 100% Fatality offshore

Table 11.3

Range

92 105 290

86–103 80–130 240–350

1000

870–2600

2000 3500

Burn area for 50% fatality

Age Group (years) 0–4 5–34 35–49 50–59 60–74 Over 75

Mean

Third-Degree Burn Area (%) 60.0 71.2 61.8 52.1 33.7 19.6

transforms the normal probability distribution curve to a linear function, with the central value arbitrarily shifted from 0 to 5 (i.e., Y = 5 corresponds to 50% probability of specified impact); further details are provided in CCPS (2000a) and in Lees (Mannan, 2005, pp. 18/55–57). Equation 11.5 gives a curve that agrees reasonably well with the values in Table 11.3. A much more conservative probit equation for the lethal effect threshold is provided by Tsao and Perry (1979). Lethal Effect Threshold Y = −12.8 + 2.56 ln (V ) .

(11.6)

The 1% lethality threshold of thermal radiation flux is correlated with exposure time by Equation 11.7 (based on data given by Hymes [1983, 1984], Mudan [1984], and Battelle Risk Management Services [1990]). 1% Lethality Threshold I = exp ( −0.7186 ln ( t ) + 5.2448 )

(11.7)

324

FIRE EFFECTS

Table 11.4

Effects of thermal dose from fireballs (Prugh, 1994)

I (kJ/m2) 40 100 150 250 500 1200

I (BTU/ft2)

Effect on Bare Skin

3.5 8.8 13.0 22.0 44.0 106.0

Threshold of pain Sunburn (first-degree burn) Blisters (second-degree burn) 1% fatal (third-degree burn) 50% fatal (third-degree burn) 99% fatal (third-degree burn)

Equation 11.3 assumes the exposed people are wearing average clothing. Lower radiation fluxes may cause lethal effects at longer exposure times if severe burns cover a high percentage of the body area. Table 11.3 reports data from the National Burn Information Exchange for patients in hospitals over the period 1976–1979 (O’Sullivan and Jagger, 2004). 11.1.4

Radiation Dosage from Transient Events

Some thermal events are transient—short jet fires and fireballs. For these, it is common practice to assess the impact in terms of thermal dose rather than of thermal intensity. The previous probit equations, which include both intensity and duration of exposure, may also be used. Prugh (1994) assembled radiation dosage levels applicable to fireballs as listed in Table 11.4. Prugh provides a correlation for the probability of fatality for a thermal dose (expressed as E in J/m2):

{

Probability = 1 + [1 − exp {−5.3[ln(E 53)]2 }]

11.2

0.5

} × [1 − 2 (E 53)].

(11.8)

EFFECTS OF THERMAL RADIATION ON PROPERTY

There are myriad types of property that can be affected by fire radiation. Only a few are cited here. 11.2.1

Equipment Degradation by Thermal Radiation

Thermal radiation effects on structures and equipment are summarized in Table 11.5 for a long duration exposure (on the order of 30–90 min). Data from original sources such as in Table 11.5 form the basis for recommended thresholds such as those listed in Table 11.6 by the World Bank (1988). The values in Table 11.6 and other similar thresholds are cited in the Federal Energy Regulatory Commission (FERC)-ABS report (FERC-ABS Consulting 2004).

EFFECTS OF THERMAL RADIATION ON PROPERTY

325

Table 11.5 Threshold effects of thermal radiation on equipment

Equipment and Effect Storage tank damage Steel, insulated on side away from fire Melting of plastics Wiring damage Degradation of cable insulation Uninsulated steel, loaded Ignition of cellulosic materials Equipment failure; failure time (min) = 20 × steel thickness (in.) Piloted ignition of wood in 1 min

Thermal Radiation Threshold (kW/m2)

Reference

11.7 12.0

Bennett et al. (1982) Croce and Mudan (1986a)

12.5 12.6 20.0

Bryan (1986) Bennett et al. (1982) Bryan (1986)

23.0–60.0 30.0–50.0 35.0

Croce and Mudan (1986a) Bryan (1986) and Croce and Mudan (1986a) Bryan (1986)

37.8

Bennett et al. (1982)

Table 11.6 Structural damage criteria cited by World Bank (1988)

Thermal Radiation (kW/m2)

Thermal Radiation (BTU/[h·ft2])

Type of Damage

37.5

11,890

25.0

7,930

12.5

3,960

Steel structures, process equipment Minimum to ignite wood for long exposure without a flame Minimum to ignite wood with a flame or to melt plastic tubing

11.2.2

Thermal Weakening of Steel and Concrete

The melting point of steel is in the range 1400–1500°C; however, steel weakens at temperatures well below this, and these lower temperatures affect the ability of steel structures and vessels to resist fire loads. Hydrocarbon flame temperatures are typically about 1100°C, and the temperature of a steel element directly impinged by a flame without cooling reaches 800°C and above. FEMA (2002) in its study of the World Trade Center collapse provides a good summary of structural steel properties exposed to fires. A temperature versus time curve is defined for a hydrocarbon fire with some differences between American Society of Testing Materials (ASTM) 1529, which specifies a maximum of 1093°C based on a source flux of 158 kW/m2 and Underwriters

326

FIRE EFFECTS

Temperature (°C)

1000 Fire temperature

800

Unprotected steel

600

Protected steel

400 200 0 0

20

40

60 Time (min)

80

100

120

Figure 11.4 Standard fire temperature curve for hydrocarbons compared to temperature rise time of unprotected and insulated steel (Buchanan, 2001).

Laboratories (UL) 1709, which uses 1143°C based on 200 kW/m2. The rise time is quick, about 5 min. A lower curve is defined for cellulosic (wood) fires with a longer rise time (ISO, 1975). These curves are used in designing insulation and similar mitigation measures. Figure 11.4 shows the ISO curve for a hydrocarbon fire and the curves for an unprotected and insulated steel member. The temperature at which structural elements weaken and sag or fail depends on the load on each element. So the effect of a fire on any given structural element can be evaluated once the yield stress for that element is known (see ASCI, 1986; EC3, 1995). An assumption used in ASTM E119 for fire assessment is that there is a critical temperature where the yield strength of steel falls to 50% of its ambient temperature strength. For many structural steels, this is close to 550°C. The yield strength temperature curve from the FEMA report is shown in Figure 11.5. Birk et al. (2006) provide a similar curve for pressure vessels using SA 455 steel with strength dropping to 50% of ambient around 550°C, except this steel retains ambient strength at temperatures up to 300°C. As many structural components are loaded with a safety factor of two, once the temperature reaches close to 550°C, it would start to fail. Different steels and load states can significantly change the failure temperature. Wuersig et al. (2009) predict the weather deck of a Moss LNG vessel starts to buckle (loaded only under its own weight) at 750°C. Without detailed structural load calculations, safety assessments should use the lower temperature. In practice, under hydrocarbon fire loads (e.g., 1100°C, 150 kW/m2), there is often only a short difference in time between steel rising from ambient temperature to 550°C or higher values so long as these are well less than 1100°C. Methods for predicting the exposure time to a fire before structural elements fail are provided in the Society of Fire Protection Engineering (SFPE, 1992, 1995).

EFFECTS OF THERMAL RADIATION ON PROPERTY

Temperature (°F) 752

Strength reduction factor FYT/FY

32 1.0

327

1472

0.8

0.6

0.4

0.2

0

0

200

Relative MOE or comp. strength

Figure 11.5

400 Temperature (°C)

600

800

Weakening of steel with increased temperature (FEMA, 2002).

1.0 Compressive strength

0.8 0.6 0.4 Modulus of elasticity 0.2 0 0

Figure 11.6

200

400 600 Temperature (°C)

800

1000

Weakening of concrete with increased temperature (BSI, 1975).

Figure 11.6 is a similar strength curve for concrete (BSI, 1975). Concrete fails by dehydration and spalling, that is, a different mechanism than steel, but the temperature range for degradation is very similar for both materials. 11.2.3 Bursting Pressure Vessels, Rail Tank Cars Much work has been carried out on the potential thermal failure of pressure vessels for liquefied petroleum gas (LPG) in fire events. An onset temperature

328

FIRE EFFECTS

for which tank car steel begins to lose strength by thermal radiation in a significant way is cited by Birk (2000) to be 427°C. At this temperature, the microstructure of steel has not yet started to change. As reported by Anderson (1982), at 427°C, a 112-type tank car has a burst strength of 5.6 MPa compared to a pressure relief valve popping pressure of 2 MPa, for a factor of safety of 5.6/2 = 2.8. The burst pressure, Pb, for a tank is usually taken from the “hoop” equation in terms of the wall thickness, t (in meter), the tank diameter, D (in meter), and the ultimate strength of the wall material at its temperature, σ, in newton per square meter (Pascal): Pb =

2σt D

(11.9)

Birk (ibid.) cites an example with a low alloy steel-type TC128 for which σ is 560 (MPa) (81,000 psi). A 3-m-diameter tank with a 16-mm wall thickness would have a burst pressure of 6 MPa (866 psia). Burk’s Analysis of Fire Effects on Tank Cars (AFFTAC; see www.tc.gc.ca for Transportation Canada) bursting tank model accounts for creep failure and assumes that all low alloy steels have the same thermal degradation for strength given in terms of Γ, defined as the ratio of steel strength at the elevated wall temperature, Tw, to that at ambient temperature, Tamb, or Γ= by

σ(Tw ) σ(TAMB )

4 ⎧ ⎡⎛ Tw ⎞ − 0.46 ⎤ . . 1 0 − 0 54 ⎪ ⎢⎣⎝ 555.555 ⎠ ⎥⎦ ⎪ Tw ⎞ ⎪ − 0.46 ⎤⎥ Γ = ⎨ 1.0 − 0.54 ⎡⎢⎛ ⎝ ⎠ . 555 555 ⎣ ⎦ ⎪ ⎪ 0 ⎪ ⎩

(11.10)

T < 700 K 700 < T < 1082 K

(11.11)

T > 1082 K.

The strength of the material accounting for creep can then be determined by multiplying Γ times the material ultimate strength at ambient temperature. The tank burst pressure is also determined by multiplying the Γ times the ambient temperature burst pressure. At 630°C, the tank burst pressure for the previous example would drop from 866 to 326 psi. However, if the tank pressure were, say, 99% or even 90% of the burst pressure, then Equation 11.11 would not predict failure. However, according to Birk (ibid.), the tank may fail by a creep mechanism because of the stress–temperature–time rupture characteristics of steel.

12 RESEARCH NEEDS

In the preceding chapters, all of the major consequence models have been introduced and some indication of uncertainties discussed. At various points in this book, the limits of current understanding were highlighted. This leads to a need to identify and discuss uncertainties and to identify the need for further developments. Some major projects addressing these areas of uncertainty are under way or are proposed so that the community can have some confidence in consequence predictions used for liquefied natural gas (LNG) hazard assessments. In this chapter, we elaborate on previous suggestions and elucidate some remaining issues. In particular, this chapter summarizes the findings of a U.S. Government Accountability Office (GAO) survey of LNG experts into areas of greatest current uncertainty.

12.1

UNCERTAINTIES

Uncertainties in LNG consequence results can arise from several sources: •

•

incomplete or inaccurate hazardous event specification (e.g., are LNG boiling liquid expanding vapor explosions (BLEVEs) possible and under what circumstances); incomplete knowledge of the consequence mechanisms at the scale being modeled (e.g., very large scale LNG pool fires);

LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

329

330 •

•

•

•

RESEARCH NEEDS

model prediction uncertainties due to omission of important mechanisms, numerical approximations, lack of validation, or coding implementation errors; incorrect “handover” of intermediate model results between different model tools in a consequence calculation sequence (e.g. discharge, evaporation, dispersion, flash fire); model result interpretation errors due to lack of complete model result enumeration (e.g., reporting maximum downwind distance after the plume has lifted off); and inherent uncertainties due to variation in the ambient environment (e.g., wind speed, direction, and stability).

12.2 RECOMMENDATIONS OF GAO SURVEY Nineteen specialists in LNG, including both authors on this book, participated in a survey by the U.S. GAO intended to obtain expert opinion on the priorities for research in the field of LNG safety (GAO, 2007). Each respondent selected a level of his/her agreement with statements about what is known concerning LNG behavior. The results of the survey are expressed by the number of opinions at the mean and on either side of the mean for each statement. An example outcome is the response histogram in Figure 12.1. The statement evaluated was “The Sandia report [Hightower et al., 2004] concluded that the most significant impacts to public safety exist within 500 m of a spill, with much lower impacts at distances beyond 1600 m even for very large spills.” The balanced distribution of responses indicates generally good consensus, but with considerable remaining uncertainty on this point. The GAO report listed three main conclusions for which there was broad agreement: 8

Number of experts

7 6 5 4 3 2 1 0 Too conservative

About right

Not conservative

Figure 12.1 Example histogram of responses to GAO statement on impact distances predicted by the Sandia report.

RECOMMENDATIONS OF GAO SURVEY

331

1. The most likely public safety impact of an LNG spill is the heat impact from a fire. 2. Explosions are not likely to occur in the wake of an LNG spill unless the LNG vapors are in confined spaces. 3. Some hazards, such as freeze burns and asphyxiation, do not pose a hazard to the public. There was general agreement on the broad-brush conclusions: •

•

The heat impact distance found by the Sandia report is correct as discussed above. Better models and confirming test data are needed for the components leading to this distance (hole size, blowdown, time-dependent pool spread and evaporation, fire surface emissive power (SEP), and smoke shielding of radiation).

The areas where there was substantial uncertainty or disagreement represent areas where research is needed. In priority order, as judged by the 19 experts: 1. Large-scale pool fire tests are needed to see if a single flame breaks up into shorter flamelets, the mass fire concept discussed in Chapter 9, Section 9.7.3. This issue was also listed as item 4. The median response of experts was for pool fire tests of 100 m diameter. 2. Is cascading failure of tanks on an LNG carrier likely or even possible? This includes the following: Evaluate possible fire engulfment of the carrier leading to overloading of the pressure relief system, and failure of tank walls. • Rapid phase transitions (RPTs) in the hull space as LNG and water mix below the water line. • Cryogenic embrittlement of structural members. 3. Large-scale LNG spills on water. To distinguish this item from item 1, the objectives would presumably be to verify evaporation flux, pool spread rate, and dispersion (whether plume liftoff occurs). The median response of experts is for experiments with pool diameters of 100 m. 4. Comprehensive modeling to evaluate the interaction of the interacting physical properties involved. Experts recognized that various phenomena must necessarily be evaluated separately, by experiments and modeling modules, for example, for •

•

•

•

turbulent mixing of LNG and water and consequent heat transfer rates; mixing of LNG and water inside the double-hull space: pressures, freezing of water, embrittlement of tank structural members, RPTs; pool spreading and evaporation with time-varying source rates;

332

RESEARCH NEEDS

glub-glub effect of LNG discharge and effect on vacuum relief of all (four or five) tanks; • pool fire with changing pool diameter and shielding of carrier hull; and • fire radiation impact on LNG carrier; stresses in structural members, heat flux reaching LNG, pressure relief venting required, protection of ship’s cabin. Risk tolerability assessment was placed at a high priority. Presumably, this means clarifying the risks of higher-cost fuel with more remotely located LNG terminals compared with the risks of injury to the public from locating terminals closer to market. Evaluate the vulnerability of LNG containment systems (carriers, onshore piping, and storage tanks). This includes validation tests for predictions of hole size to double-hull ship structure from ship collision/ allision or from explosive charges. Computational fluid dynamics (CFD) modeling has predicted that an explosive penetration of the outer hull can result in a number of small holes to the inner hull from fragments. Security concerns limit publishing work that would establish a correlation between explosive charge and inner hull penetration area. Evaluate the effectiveness of various mitigation techniques. This could include measures to keep possibly explosive-laden boats at a certain distance away from LNG carriers, better ship collision avoidance measures, firefighting foams, etc. Evaluate the effects of sea water and LNG together pouring into the double hull of a carrier. Of particular interest is whether the conditions might contain the LNG enough to develop a large overpressure from an RPT. Evaluate the effects of wind, waves, and water current on pool spread size and evaporation rate. Some experts postulate that waves inhibit pool spread, while others expect they enhance pool spread. Since the hazard to the public is considered greatest from an LNG spill in a harbor, where wave effects are less of a factor, the need for experiments with large waves is diminished. Improve 3-D CFD pool spread and dispersion models. No experts ranked this item as a very great need, but four saw it as a great need. Evaluate the effects of LNG composition on vaporization rate, thermal radiation, and explosive behavior. Previous studies on the effect of LNG composition on RPTs have been contradictory. Would an explosive attack result in immediate vapor cloud ignition? Currently, immediate ignition is expected by nearly all experts. However, there could be a sufficient time delay between an initial explosion at the outer hull and the cooldown of hot surfaces before LNG vapors in the flammable range reach these surfaces. Obtain a better understanding of the likelihood and impact of RPTs. Find what is the largest explosion energy likely. •

5.

6.

7.

8.

9.

10. 11.

12.

13.

LNG MODEL EVALUATION PROTOCOLS (MEPS)

333

14. Evaluate effects of igniting LNG vapors in containment or ballast tanks. There are existing models to predict these effects, and there is no doubt that the effects can be large. At issue is whether ignition sources can reach vapors inside the tanks. Flame arresters can protect vacuum relief and pressure relief lines, if a case can be made that they are needed. 15. Can LNG carrier tanks BLEVE? The capacity of pressure relief valves may not be designed for an engulfing fire that conducts heat to the LNG at a fast rate. The research should focus on calculating heat transfer rates from fires with scenarios involving loss of tank insulation. 16. Under what conditions could LNG vapors explode as a deflagration or a detonation? Since considerable research has been done on these issues for a variety of fuels, 18 of the experts voted this issue as a moderate need, some need, or little to no need. 17. Evaluate the effects of a large, unignited vapor cloud drifting from the incident site. None of the experts voted this to be a very great or great need. It is largely addressed by improving models using test data addressed in items 1–3. 18. Evaluate the effect of clothing and obstructions on the radiant heat level received by the public from an LNG fire. This item was ranked low since considerable research is already available on the subject. 12.3

LNG MODEL EVALUATION PROTOCOLS (MEPS)

In view of the uncertainties in source term definition and dispersion modeling, the U.S. National Association of State Fire Marshals (NASFM, 2009) has assessed two MEPs developed by the Fire Protection Research Association (FPRA), a subsidiary of the National Fire Protection Association (NFPA), and its contractor, Health and Safety Laboratory (HSL). The U.K. HSL has prepared initially a dispersion protocol and subsequently a source term protocol. The protocol (Ivings et al., 2007a) defines a MEP for a review of the dispersion modeling of LNG. A key aim of the MEP is to develop a formal system to evaluate models proposed for LNG source term and dispersion analysis. The latest revision of NFPA 59A in 2009 allows for this use of models in addition to the two models already approved for LNG vapor dispersion—DEGADIS (DEnse GAs DISpersion) and FEM3A (Finite Element Analysis Model 3A). The former is a well-established dispersion model but has no integrated source term and has not been updated for many years. FEM3A is a CFD code that is not available publicly. For this reason, a variety of newer models have been used for assessments of proposed LNG facilities, either to initialize DEGADIS model or to cover the entire consequence sequence, but with only the modeler’s assurance of validation. The NASFM system would document all available LNG or potentially other relevant experimental trials and would provide these in a form that

334

RESEARCH NEEDS

candidate models could be run to allow validation to be demonstrated. The HSL team used the European Union (EU) Scientific Method of Evaluating DISpersion (SMEDIS) protocol as the basis for the LNG protocol. The SMEDIS project was a more general dispersion protocol to support the EU Seveso Directive for process plant safety. Daish et al. (2000) identified three important aspects of model validation: •

•

•

Scientific Assessment Objective examination of the scientific and useroriented aspects of a model. Verification Confirmation that the computer implementation is an accurate translation of the mathematical model. Validation Comparison of the predictions of the model against experimental observations.

It is apparent that there is much more to model evaluation than simply matching experimental data, and the NASFM initiative is welcomed. While the full details of the protocol remain to be determined, HSL has collected a wide range of data in a form suitable for validation, and this may be obtained from NASFM. NASFM convened a panel of experts for a peer review of the HSL report. A review resulted in a number of changes and a final assessment was published by AcuTech (2009) with responses by the HSL (2008). This identified some areas for the MEP to be resolved, briefly discussed below: •

•

•

•

•

The MEP should be more concise and objective with regard to the goal of defining the criteria and the “how to” steps for evaluating models—the current protocol is lengthy and technically complex, and this can lead to interpretation errors and can increase the cost of compliance. Better advice is required for application to LNG terminal siting decisions. The MEP needs specific guidance on exactly how it shall be applied and what constitutes success. The separate MEPs for source term and dispersion need to be integrated—separation could allow uncertainties due to model handover of data sets. The implications of this formal MEP in terms of U.S. legislation need to be addressed.

In principle, the development of a practical MEP for LNG and for other hazard zone evaluations by regulatory agencies deserves support as there are many examples of misuse of models by both proponents and objectors. The scientific discipline of an MEP should at least ensure that models of appropriate quality are employed. A secondary benefit is that modelers of codes that do not perform well may be encouraged to enhance their codes with the latest science or numerical techniques. Over time, this is how better modeling results.

SPECIAL TOPICS

335

This was a clear benefit of a prior model validation exercise (Hanna et al., 1991, 1993). While it is easier, it is not healthy for a single model to be used for over 30 years with no updates. A quick perusal of the references in this textbook shows publication dates in the past 15 years and the majority in the past 5–10 years. A regulatory process that does not address these changes does not, in fact, protect the public or encourage beneficial investments.

12.4

SPECIAL TOPICS

12.4.1

LNG Pool Spill and Fire Tests

Field experiments form the basis for scientific understanding of the various phenomena discussed in previous chapters, including •

• •

•

evaporation and spreading on land and water (see Section 7.2.3, Table 7.2); dispersion of unignited vapor clouds (see Section 8.3, Tables 8.4 and 8.5); pool fire flame height, SEP, transmissivity (see Section 9.4, Table 9.1); and burning rate (see Section 9.5, Tables 9.1 and 9.3).

Models have been derived based on field experiments, some with a more fundamental basis than others. At issue is whether the model predictions accurately extrapolate from small-scale tests to the scale of large possible events. As pointed out by Fay (1978), “most of the important physical phenomena of interest vary as a small fractional power of the spill size.” Fay (1981) cites a document by the U.S. Department of Energy (DOE, 1978) to categorize phenomena with a weak or a strong expected dependence on spill size, as shown in Table 12.1. Assuming the models are valid, for these weak dependence properties, extrapolation to spills larger by an order of magnitude is likely to be valid. This assessment is somewhat dated. As discussed in Chapter 9 for flame height, a change of mechanism is likely to occur at higher spill scales. The Table 12.1 Dependence of LNG spill phenomena on spill size (DOE, 1978)

Weak Dependence Evaporation flux, burning flux of unconstrained spills Vapor cloud burn-up time Critical wind speed (for maximum cloud extent) Flame height

Strong Dependence Maximum vapor cloud extent to flammable limits Overall radiant energy emission rates

336

RESEARCH NEEDS

Table 12.2 Expected size of large LNG spills

Source Tank truck Import terminal transfer piping failure Import terminal tank failure Marine tanker accident

Largest Spill (m3) 40 1,400a 1,667b 75,000 25,000

Capacity (m3) 40 Varies 112,00 × 2c 35,000 × 5c

a

Fay (1978). Section 6.1, 10-min spill at full rate of 10,000 m3/h. c The second number is the multiples that could, conceivably, be involved. b

uncertainty in dispersion modeling predictions of cloud extent is much less now than was thought in 1978. The issue remains that larger field experiments are desirable and should be designed to maximize the reduction in the areas of greatest uncertainty for model extrapolation. Table 12.2 provides a summary of the expected largest size of accidental LNG spills estimated by Fay (1978), along with the capacity of typical marine and storage terminal tanks. From Table 8.4 in Chapter 8, the largest LNG spill tests conducted prior to 2009 are the Burro tests at 39 m3 and the Falcon tests at 66 m3. Fay’s recommendations are for tests at two higher volumes than previously used: 1. 300-m3 average spill volume 2. 3000-m3 average spill volume The latter would provide tests within a factor of 10 for marine tanker accidents. Before reaching this point, though, the recommendation for running 300-m3 tests is that model comparison studies (round robin model trials) should be conducted to reduce the variance in model predictions. Using the most sophisticated models should increase confidence to the point that the 3000-m3 field tests may not be needed. Larger LNG spill tests are under development at Sandia Laboratories (2009). Plans are for 70- and 100-m-diameter pool fire tests with tests to be completed in 2009 and reported in 2010. The first tests, using 99.5% LNG, were expected to produce a 33-m pool, but realized a 22.9-m pool. The flame height was a maximum of 60 m, which was in line with the expected 2.5:1.0 height/pool diameter ratio. There was very little smoke. Measured average surface emissive power were • • •

212 kW/m2 at a 8.5-m flame height, 173 kW/m2 at a 22.3-m flame height, and 95 kW/m2 at a 36.1-m flame height.

SPECIAL TOPICS

337

These are in line with previous measurements for a bright fire with little smoke absorption. Larger fires are expected to be more smoky.

12.4.2

Limitation of Boussinesq Approximation

Equation 12.1 defines an effective gravitational acceleration that is used in modeling the spread rate of liquids (see Eq. 7.11) and, in some models, vapors. Lohmeyer et al. (1980) call Equation 12.1 the Boussinesq approximation, although Fay (1980) indicates that Equation 1 is not necessarily valid only for small values of g′ when ρv − ρa g whenever ρv > 2ρa, which is possible with high-molecular-weight gases. Since it is unlikely that fluid accelerations of dense gas will exceed g, it would seem better to use g″ whenever the Boussinesq approximation is not valid, since always, g″ ≤ g. The distinction between g′ and g″ has been emphasized by Fannelop and Jacobsen (1982). They apply g″ in the shallow wave equations, but apply g′ in the boundary condition at the leading edge, so the spread rate, u = dR/dt, varies with (g′)1/2. Fannelop and Jacobsen also point out that there are limiting values of ρv/ρa above which a gap appears in the interior of the cloud. Fay (1982) suggests this might determine the onset of the bifurcation of a cloud, as was seen for the Burro 8 test in Section 8.4.1.1, Figure 8.10.

12.4.3

LNG Plumes Not Modeled Well for Calm Winds

The similarity type plume dispersion models are generally not designed to predict plume spread and dilution with a calm wind. In fact, modelers place a lower limit for applicable wind speeds, usually around 2–3 m/s (at the standard reference height of 10 m). The variance in model predictions compared to data increases at lower and lower wind speeds, both because of model limitations and also because the variability in experimental measurements increases at low wind speed. This is shown by Woodward (1998b) analyzing a set of propane releases by the German research organization TUV. Fay (1982) provides a theoretical argument that dispersion models should add some form of turbulent kinetic energy dissipation to properly treat dispersion with calm winds. However, since the maximum extent for a flammable plume usually occurs at finite wind speeds (see Section 8.2.3), there is no important need to model dispersion with calm winds.

338

12.4.4

RESEARCH NEEDS

The Use of ½ LFL as an End Point

The ½ LFL contour has been used extensively to account for inhomogeneous concentrations, that is, the presence of puffs of concentrations higher than the average. The justification for this practice is that puffs of concentration above the LFL are expected within the ½ LFL contour, so ignition could occur at a reasonable probability out to the ½ LFL average concentration. A proposal to the NFPA 59A committee has challenged the use of the ½ LFL contour for use in defining the pool fire radiation exclusion zone for LNG terminals (Raj, 2008c). The proposal recommends using the LFL contour instead and that the LFL predictions should be based on short-term averaging time. A counter view is developed by Iving and Webber (2007b). The arguments by Raj point to experiments indicating that the burn area after ignition is not larger than the pre-burn area to the LFL: 1. The Coyote tests reported by Rodean et al. (1984, pp. 34, 37, 83) placed ignition sources at various points in the plume. Specifically for Coyote Test 3, the LFL was 195 m. The burn distance was 160 m, 82% of the LFL distance. “With the 2-s [averaging time] data we found the burn region closely coincided with the instantaneous [no further averaging] 5% contour.” Rodean et al. concluded, “The propagating flames appeared to go out when they reached the 5% gas concentration limit (except for Coyote 5). However the location of this limit changes significantly during the burn because of gas advection and meander as well as the effects of fire growth and buoyancy.” An RPT in the Coyote 5 experiment lowered the LFL below 5% by ethane enrichment. 2. Propane release and ignition tests by Shell (Evans and Puttock, 1986). The cloud was ignited by a pilot ignition source on a trolley that was dragged across the plume where pre-ignition concentration measurements had been made. The averaging time for the gas sensors was a relatively high 2–3 s. Concentrations were measured at 10 Hz and then smoothed with a 1-s average. Evans and Puttock report that no flash backfire could be sustained where the local gas concentration was below 0.9 LFL (2.14% for propane) for large plumes or below 0.6 LFL for smaller plumes. The pre-burn area, possibly measured but more usually predicted, depends upon the averaging time used in obtaining the average concentration. Obviously, fluctuating concentrations should be averaged over a long enough period to be roughly independent of time, that is, at steady state. Ivings and Webber (2007b) point out that it is also necessary for burn-back that a pocket of gas above the LFL in a field at, say, ½ LFL must not be isolated from the portion of cloud that is above the LFL, or ignition will simply flash and extinguish. A larger averaging time predicts lower average concentrations and a smaller area for the average LFL contour. Averaging time depends not only on sample

CONCLUSIONS

339

duration but also on the instrument response time. For flammable gases, response time and averaging times of 1–2 s is desired. Older test data used slower instruments and averaging times up to 10–20 s. All dispersion model predictions depend on the averaging time of the data used for “tuning” or validation. Raj (2008a) cites from the DEGADIS documentation that this model uses a 10-s averaging time and infers that DEGADIS predictions to the 50% LFL represent “double dipping” in conservatism. Ivings and Webber (2007b) conclude the following of what is known and not known concerning the issue of regulations specifying the use of ½ LFL versus LFL: 1. For ignition purposes, short averaging times are appropriate. Much longer (say, 10 min) averaging times are not relevant. Cloud average concentrations may indicate whether a flame will propagate through the cloud, but for purposes of ignition, look at concentration at a single point. 2. Good models predict within a factor of 2, for releases that approximate well to the idealizations upon which the model is based. 3. Repeated tests in a wind tunnel result in measured concentrations with a range width of about two. 4. Even cloud averages with an averaging time of 1 s exhibit fluctuations up to a factor of 2. 5. Flame has been observed to propagate through the entire cloud when measurements near the ignition point reveal a concentration below the LFL. Flames have also been seen to billow out and then to extinguish without propagating through the cloud when measurements near the ignition point show a concentration below ½ LFL. Altogether, Ivings and Webber (ibid.) recommend keeping ½ LFL contours in regulations primarily because the uncertainty in modeling is a factor of 2.

12.5

CONCLUSIONS

The LNG consequence topic has had a major rejuvenation of activity in the past 10 years, and some long-standing problems are starting to be addressed in a scientific manner. This work challenges some long-standing approaches used by modelers and regulators. Newer methods such as CFD allow more detailed modeling of near-field effects, significantly affected by local structures and terrain. The answer to uncertainties is often not just larger experiments. Large-scale experiments have their own uncertainties due to variations in ambient

340

RESEARCH NEEDS

conditions during the experiment and data collection. The answer is a combination of understanding the consequence mechanism at the relevant scale through the experiment and integration of these with other important mechanisms in a well-conceived and executed model. The MEP is novel for the United States and should allow for the development of better codes, acceptable to the authority having jurisdiction, which can resolve issues that are now the subject of large-scale experiments.

REFERENCES

29 CFR 1915.12 (2002). Precautions and the order of testing before entering confined and enclosed spaces and other dangerous atmospheres. USA Code of federal Regulations. 49 CFR, Part 193 (2005). Liquefied natural gas facilities: Federal safety standards, USA Code of Federal Regulations, October 1, 2005. ABS Consulting (2004). Comments in response to public comments. www.abs-group. com (accessed June, 2009) AcuTech Group (2008). Technical Panel review of the LNG vapor dispersion model evaluation protocol. Report of the technical panel, June 25, 2008. Report to the National Association of State Fire Marshals, November 24, 2008. AcuTech (2009). Review of the LNG vapor dispersion model evaluation protocol. Report of the technical panel, National Association of State Fire Marshals, Alexandria, VA, January. Adams JS, Williams DW, Tregallis-Williams J (1972). Air velocity, temperature, and radiant heat measurements within and around a large free-burning fire, 14th International Combustion Symposium, pp. 1045–1052. Aldwinkle DS, Slater DH (1983). Risk and reliability methods used in the analysis of an offshore LNG liquefaction and storage ship. I Chem E Symposium Series No. 81, Rugby, U.K. American Gas Association (AGA) (1974). LNG safety program interim report on phase II work. Report IS-3-1, Battelle Columbus Laboratories, Arlington, VA, July 1974. American Gas Foundation (2008). Ability of the US to compete in the global LNG marketplace: An assessment of the challenges and opportunities, October, 2008. pp. 14, 24, 32. http:/www.gasfoundation.org/ResearchStudies/ (accessed June 15, 2009). American Institute of Steel Construction (AISC) (1986). Manual of steel construction, load and resistance factor design, 1st Edition, pp. I-18 to I-161. AISC, Chicago. Anderson CE (1982). Rail tank-car safety by fire protection. 6th International Fire Protection Seminar, Karlsruhe, Germany, September 1982. LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

341

342

REFERENCES

Anderson RP, Armstrong DR (1972). Experimental study of vapor explosions. In Proceedings of the Third International Conference on Liquefied Natural Gas. Washington, DC. Anonymous (2002). Ships as terrorist targets. American Shipper, November 2002, p. 59. Atallah S, Shah NJ (1990). LNGFIRE3- A thermal radiation model for LNG fire. Gas Research Institute Report #GRI-89/0176, June 1990. Auer AH Jr (1978). Correlation of land use and cover with meteorological anomalies. J. Appl. Meteorol. 17, 636–643. Australian Standard AS 3961 (2005). The storage and handling of liquefied natural gas. SAI Global Publisher, Sydney, Australia. Australian Standard AS/NZS 4360 (2004). Risk management and companion handbook HB 436 risk management guidelines. SAI Global Publisher. Sydney, Australia. Bagster DF, Pitblado RM (1989). Thermal hazards in the process industry. Chem. Eng. Prog. 85(7), 69. Bainbridge K (2003). An overview of the LNG shipping industry. Fundamentals of base load LNG. Gas Technology Institute Training Course, Linthicum, MD, September 22–25, 2003. Baker Engineering and Risk Consultants, Inc. (2003). SafeSite3G® Validation Manual. Woodward JL. Baker Engineering and Risk Consultants, Inc. (2005). SafeSite3G® Theory Manual Version 1.04. Woodward JL. Baker WE, Tang M (1991). Gas, dust and hybrid explosions. Elsevier, Amsterdam. Baker QA, Tang MJ, Scheier EA, Silva GJ (1994). Vapor cloud explosion analysis. AIChE 28th Loss Prevention Symposium, Houston, TX, April 17–24, 1994. Baker QA, Doolittle CM, Fitzgerald GA, Tang, MJ (1997). Recent developments in the Baker-Strehlow VCE analysis methodology. 31st Loss Prevention Symposium, Houston, TX, AIChE, Paper 42f, March 9–13, 1997. Bakke JR, Hjertager BH (1986). Quasi-laminar/turbulent combustion modeling, real cloud generation & boundary conditions in the FLACS-ICE code. Report CMI 865402-2, Christian Michelsen Institute; also in Bakke JR (1986). Numerical simulation of gas explosions in two-dimensional geometries. PhD thesis. University of Bergen, Bergen. Battelle Risk Management Services (1990). Thermal radiation hazard levels for Exxon SYU Expansion Project. Columbus, Ohio. Baumeister KJ, Simon FF (1973). Leidenfrost temperature—Its correlation for liquid metals, cryogens, hydrocarbons and water. Trans. ASME, J. Heat Trans. pp. 166–173. BBC World News (2009). http://www.bbc.com/ (accessed July 1, 2009). Becker HA, Liang D (1978). Visible length of vertical free turbulent diffusion flames, Combustion and Flame, 32, 115–137. Bennett GF, Feates ES, Wilder I (1982). Hazardous materials spills handbook. McGrawHill, New York. Berenblut BJ, Whitehouse SM, Phillips HCD, Salter RE (1985). Pemex in retrospect: Lessons in risk assessment and loss prevention. Chem. Eng. 418, 17.

REFERENCES

343

Berenson PV (1961). Film-boiling heat transfer from a horizontal surface. Trans. ASME, J. Heat Transf. pp. 166–173. Berthoud G (2000). Vapor explosions. Annu. Rev. Fluid Mech. 32, 573–611. Birch AD, Brown DR, Fairweather M, Hargraves GK (1989). An experimental study of a turbulent natural gas jet in a cross-flow, Combust. Sci. Tech. 66, 217–232. Birk AM (2000). Review of AFFTAC thermal model. Report for Transport Canada, Transportation Development Centre, TP13539E, January, 2000. Birk AM, Poirier D, Davison C (2006). On the thermal rupture of 1.9 m3 propane pressure vessels with defects in their thermal protection system. J. Loss Prev. Process Ind. 19, 582–597. Bjerketvedt D, Bakke JR, van Wingerden K (1997). Gas explosion handbook. J. Hazard. Mater. 52, 1–150 and report CMR-93-A25003, Version 1.1, Christian Michelsen Institute, Jan, 1993. Blackmore DR, Eyre JA, Summers GG (1982a). Dispersion and combustion behavior of gas clouds resulting from large spillages of LNG and LPG on to the sea. Trans I. Mar. E. (TM) 94, Paper 29. Blackmore DR et al. (1982b). An updated view of LNG safety. In American Gas Association Transmission Conference Operation Section Proceedings, Chicago, American Gas Association, Arlington, VA, pp. T226–T232. May 17–19, 1982. Blanchat T, Luketa-Hanlin A, Hightower M, Tieszen S (2007). DOE/Sandia National Laboratories coordinated approach for LNG safety and security research. Briefing to NARUC Staff Subcommittee on Gas, Fire Science & Technology, Energy Systems Analysis, Sandia National Laboratories, July 15, 2007. Blinov VI, Khudiakov GN (1957). Certain laws governing diffusive burning of liquids, in Doklady Akad. Nauk SSSR, 113, pp. 1094–1098) Reviewed by Hottell HC, Fire Res. Abstracts (1959) 1, p 41. Board SJ, Hall RW (1976). Recent advances in understanding large scale vapor explosions. Proceedings of Committee on Safety of Nuclear Installations Meeting for Sodium-Fuel Interactions of Fast Reactors, Tokyo, pp. 249–293. Boe R (1998). Pool boiling of hydrocarbon mixtures on water. Int. J Heat Mass Transf. 41, (8–9), 1003–1011. Boyle GJ, Kneebone A (1973). Laboratory investigations into the characteristics of LNG spills on water. Evaporation, spreading, and vapor dispersion. API Report 6732, Shell Res. Ltd., Thornton, England, March. Bragg SL (1960). Effect of compressibility on the discharge coefficient of orifices and convergent nozzles. J. Mech. Eng. Sci. 2, 35–44. Briggs AJ (1976). Experimental studies of thermal interactions at AEE Winfrith. Proceedings of Committee on Safety of Nuclear Installations Meeting for SodiumFuel Interactions of Fast Reactors, Tokyo, pp. 75–96. Brighton PWM (1986). Heavy gas dispersion from sources inside buildings or in their wakes. Proc I Chem E. Symposium Series, Refinement of estimates on the consequences of heavy toxic vapor releases, Manchester, U.K. Brisco F, Shaw P (1980). Spread and evaporation of liquid. Prog. Energy Combust. Sci. 6, 127–140.

344

REFERENCES

Britter RE (1991). The evaluation of technical models used for major accident hazard installations. Report to the Commission of the European Communities Directorate General XII, August 1991. Britter RE, Griffeths RF (1982). Dense Gas Dispersion, Chem. Eng. Monograph 16, Elsevier, NY. Britter RE (2002). Model evaluation report on UDM Version 6.0, Ref No. SMEDIS/00/9/E, Cambridge Environmental Research Consultants, Cambridge, U.K. Britter RE, McQuaid J (1988). Workbook on the dispersion of dense gases. HSE Contract Research Report No. 17/1988. Health and Safety Executive, Sheffield, U.K. Britton LG (2000). Using maximum experimental safe gap to select flame arresters. Process Saf. Prog. 19(3), 140–145. Brown TC, Cederwall RT, Chan ST, Ermak DL, Koopman RP, Lamson JW, McClure JW, Morris LK (1990). Falcon series data report: 1987 LNG vapor barrier verification field trials, Gas Research Institute, GRI-89/0138, June 1990. Bryan JL (1986). Damageability of buildings, contents, and personnel from exposure to fire, Fire Safety J. 11(1), 15. BS EN 1473 (2007). Installation and equipment for liquefied natural gas. Design of onshore installations British-Adopted European Standard, February 28, 2007, 134 pp., ISBN 9780580502293. BSI (1975). Structural use of concrete, BS8110. British Standards Institution, London, U.K. Buchanan AH (2001). Structural design for fire safety. John Wiley & Sons, New York. Buettner K (1951). Effects of extreme heat and cold on human skin; I. Analysis of temperature changes caused by different kinds of heat application; II. Surface temperature, pain and heat conductivity experiments with radiant heat; III. Numerical analysis and pilot experiments in penetrating flash radiation effects. J. Appl. Physiol. 3, 691–703. Buettner K (1957). Heat transfer and safe exposure times for man in extreme thermal environments. ASME paper 57-SA-20, American Society of Mechanical Engineers, New York. Burgess D, Zabetakis MG (1962). Fire and explosion hazards of LNG. US Bureau of Mines Investigation Report 6099, U.S. Department of Commerce, Washington, DC. Burgess DS, Biordi J, Murphy J (1972). Hazards of spillage of LNG into water. PMSRC Report No. 4177, U.S. Department of Interior, Bureau of Mines, Pittsburgh, PA, September. Burgess DS, Murphy JN, Zabetakis MG (1970a). Hazards of LNG spillage in marine transportation. Report S4105, U.S. Department of Interior, Bureau of Mines, Pittsburgh, PA, February. Burgess DS, Murphy JN, Zabetakis MG (1970b). Hazards of LNG spillage of natural gas on water. Report 7448, U.S. Department of Interior, Bureau of Mines, Pittsburgh, PA, November. Buschmann CH (1975). Experiments in the dispersion of heavy gases and abatement of chlorine clouds. Transp. Hazard. Cargoes 4, 475–488. Canadian Centre for Occupational Health and Safety (CCOHS) (2008). http://www. ccohs.ca/oshanswers/safetyhaz/hand tools/nonsparking.html (accessed July 7, 2009).

REFERENCES

345

Carissimo B, Jagger SF, Daish NC et al. (2001). The SMEDIS database and validation exercise. Int. J. Environ. Pollut. 16(1–6), 614–629. Cassidy K, Pantony MF (1988). Major industrial risks: A technical and predictive basis for on and offsite emergency planning in the context of UK legislation, IChemE Symposium Series No. 110, pp. 75–95, Institute of Chemical Engineers, Rugby, UK, Hemisphere Publishing Co. Castro IP, Kumar A, Snyder WH, Arya SPS (1993). Removal of slightly heavy gases from a valley by crosswinds. J. Hazard. Mater. 34, 271–293. Cavanaugh TA, Siegell JH, Steinberg KW (1994). Simulation of vapor emissions from the liquid spills. J. Hazard. Mater. 38, 41–63. Center for Chemical Process Safety (CCPS) (1995). Guidelines for chemical transport risk analysis. AIChE, New York, 382 pp., ISBN No. 0-8169-0626-2, Pub No: G-28. Center for Chemical Process Safety (CCPS) (1998). RELEASE: A model with data to predict aerosol rainout in accidental releases. AIChE, New York, 184 pp., ISBN No. 0-8169-0745-5. Center for Chemical Process Safety (CCPS) (1999). Estimating the flammable mass of a vapor cloud, 289 pp. AIChE, New York, ISBN No. 0-8169-0778-1, Pub. No.: G-60. Center for Chemical Process Safety (CCPS) (2000a). Guidelines for chemical process quantitative risk analysis, 2nd ed. AIChE, New York, ISBN 0-8169-0720-X. Center for Chemical Process Safety (CCPS) (2000b). Evaluating process safety in the chemical industry: A user’s guide to quantitative risk analysis, 102 pp. AIChE, New York, ISBN No: 0-8169-0746-3, Pub No: G-51. Center for Chemical Process Safety (CCPS) (2000c). Guidelines for use of vapor cloud dispersion models, 2nd ed., p. 271. John Wiley, New York. Center for Chemical Process Safety (CCPS) (2001a). Layer of protection analysis: Simplified process risk assessment. AIChE, New York. Center for Chemical Process Safety (CCPS) (2001b). Vapor cloud dispersion modeling CD-ROM, AIChE, NY, ISBN No. 0-8169-0729-3, Pub No: Y-45. Center for Chemical Process Safety (CCPS) (2002). Wind flow and vapor cloud dispersion at industrial and urban sites—With CD-ROM for calculating a surface roughness parameter, 140 pp. AIChE, New York, ISBN No. 0-8169-0863-X, Pub No.:G-75. Center for Chemical Process Safety (CCPS) (2008). Guidelines for hazard evaluation procedures with examples, 3rd ed. AIChE, New York. Center for Chemical Process Safety (CCPS) (2009a). Guidelines for risk criteria. AIChE, New York, in press. Center for Chemical Process Safety (CCPS) (2009b). Guidelines for evaluating the characteristics of vapor cloud explosions, flash fires and BLEVEs. AIChE, New York; new edition pending 2010. Chamberlain GA (1987). Developments in design methods for predicting thermal radiation from flares. Chem. Eng. Res. Des. 65, 299. Chan ST, Gresho PM, Lee RL, Upson CD (1981a). Simulation of three-dimensional, time-dependent, incompressible flows by finite element method, Proceedings of the American Institute of Aeronautics and Astronautics (AIAA) 5th Computational Fluid Dynamics Conference, Palo Alto, CA, June.

346

REFERENCES

Chan ST, Gresho PM, Ermak DL (1981b). A three-dimensional conservation equation model for simulating LNG dispersion in the atmosphere, Lawrence Livermore National Laboratory report UCID-19210, Livermore, CA, Sept. Chan ST (1990). FEMA3A-A finite element model for the simulation of gas transport and dispersion: User’s Manual. Lawrence Livermore Laboratory, UCRL-CR-105184 (also GRI 90/0189). Livermore, CA. Chan ST (1992). Numerical simulations of LNG vapor dispersion from a fenced storage area. J. Hazard. Mater. 30, 195–224. Chang JC, Hanna SR (2004). Air quality model performance evaluation. Meteorol. Atmos. Physics 87, 167–196. Chang HR, Reid RC (1982). Spreading-boiling model for instantaneous spills of liquefied petroleum gas (LPG) on water. J. Hazard. Mater. 7, 19–35. Chang HR, Reid RC, Fay JA, (1983). Boiling and spreading of liquid nitrogen and liquid methane on water. Int. Comm. Heat Mass Transf. 10, 253. Chatris JM, Quintela J, Folch J, Planas E, Arnaldos J, Casa J (2001). Experimental study of burning rate in hydrocarbon pool fires. Combust. Flame 126, 1373–1383. Cheniere web site (2008). www.cheniere.com, accessed July 2, 2009. CHIV (2003). Safety History of International Operations. CHIV International, Houston, TX. Cho DH, Ivins RO, Wright RW (1971). Pressure generation by molten FCI under LMFBR accidents conditions. American Nuclear Society Topical Meeting, New Developments in Reactor Mathematics Applications, Idaho Falls, ID. Clay GA, Fitzpatrik RRD, Hurst NW, Carter DA, Crossthwaite PJ (1988). Risk assessment for installation where liquefied petroleum gas (LPG) is stored in bulk vessels above ground. J. Hazard. Mater. 20, 357. Clever P, Johnson M, Ho B (2007). A summary of some experimental data on LNG safety. J. Hazard. Mater. 140, 429–438. Coers DH (2003). LNG storage, fundamentals of base load LNG. Gas Technology Institute Training Course, Linthicum Heights, MD, Sept. 22–25, 2003. Colenbrander GW (1980). A Mathematical Model for the transient behavior of dense vapor clouds, 3rd International Symposium on Loss Prevention and Safety Promotion in the Process Industries, Harrogate, England, Sept. Colenbrander GW, Puttock JS (1983). Dense gas dispersion behavior: Experimental observations and model developments. IChemE Symposium Series No. 80, Series 33, Rugby, U.K., F66-F75; also 4th International Symposium on Loss Prevention and Safety Promotion in the Process Industries, September 12–16, 1983. Colenbrander GW, Evans A, Puttock JS (1984a). Spill tests of LNG and refrigerated liquid propane on the sea, Maplin Sands 1980: Dispersion Data Digests. Thornton Research Centre, Shell Research Limited, Thornton, U.K. Colenbrander GW, Puttock JS (1984b). Maplin Sands Experiments 1980: Interpretation and modeling of liquefied gas spills on sea. In Atmospheric dispersion of heavy gases and small particles (eds. G Ooms, H Tennessee), Springer-Verlag, Berlin, pp. 277–295. Colenbrander GW, Puttock JS (1988). Dispersion of releases of dense gas: Development of the HEGAGAS model. HEGADAS Public Release Document. U.S. Environmental Protection Agency, Distributed by the National Technical Information Center (NTIS).

REFERENCES

347

Collier JG, Thome JR (1994). Convective boiling and condensation. Clarendon Press, Oxford. Computit (2005). Kameleon FireEx: An advanced computational system for calculation of fire structure interaction. Fabig Technical Newsletter, January 2005, 41, www. Computit.No Conch Methane Services Ltd. (1962). LNG, characteristics and burning behavior. London. Conrado C, Vesovic V (2000). The influence of chemical composition on vaporization of LNG and LPG on unconfined water surfaces. Chem. Eng. Sci. 55, 4549–4562. Cook J, Bahrami Z, Whitehouse RJ (1990). A comprehensive program for calculation of flame radiation levels. J. Loss Prev. Process Ind. 3(1) 150. Cook J, Woodward JL (1993). A new unified model for pool spreading, evaporation and solution on land and water. In Health, Safety and Loss Prevention in the Oil, Chemical and Process Industries, pp. 359–370. Butterworth, Oxford. Cook J, Woodward JL, (1995). Further development of the unified dispersion model. Symposium on Loss Prevention and Safety Promotion in the Process Industries, Antwerp, Belgium, June 19–23. Cook MA (1985). The science of high explosives. Rheinholt, New York. Cormier B, Wang Y et al. (2006). LNG mitigation experiments- Preliminary results. International Safety Symposium Mary Kay O’Connor Process Safety Center, Texas A&M University, College Station, TX, Aug. 5, 2008. Cornwell JB (2001). Letters to Don Juckett, Department of Energy, October 2–3, 2001. Countryman CM, Palmer TY, Storey TG (1969). Project Flambeau—an investigation of mass fire (1964–1967), Final Report, Pacific Southwest Forrest and Range Experimental Station, US Dept. of Agriculture, Forrest Service. Cox G, Chitty R (1985). Some source-dependent effects of unbounded fires. Combust. Flame 60, 219–232. Cox RA, Carpenter RJ (1980). Further development of a dense vapor cloud model for hazard analysis. In Heavy Gas and Risk Assessment (ed. S Hartwig), pp. 55–88. D. Reidel Publishing Co, Dordrecht, Holland. Craknell RF, Carsley AJ (1997). Cloud fires: A methodology for hazard consequence modeling. I Chem E Symp. Ser. 141, 139–150. Croce PA, Mudan KS (1986a). Calculating impacts for large hydrocarbon fires. Fire Safety J, 11(1), 99. Croce PA, Mudan KS (1986b). Thermal radiation from LNG trench fires, Gastech 85, p 158. Cronenberg AW, Grolmes MA (1975). Fragmentation modeling relative to the breakup of molten UO2 in sodium. Nucl. Saf. 16, 683–700. Dahlsveen J, Kristofferson R, Saetran L (2001). Jet mixing of cryogen and water, turbulence and shear flow phenomena. In Proceedings of the Second International Symposium, Vol. 2, pp. 329–334. Stockholm, Sweden. Daish NC, Britter RE, Linden PF, Jagger SF, Carissimo B (2000). SMEDIS—Scientific model evaluation of dense gas dispersion models, Int. J. Environmental Pollution 14, 39–51. Daish NC, Linden PF, Vieillard V, Nedelka D, Roberts TA, Butler CJ (2001). A new unified investigation into vapor cloud fires. In Proceedings of the 13th IGT Intl.

348

REFERENCES

Conference and Exhibition on Liquefied Natural Gas, Seoul, Korea 2, pp. Ps6–7.1 to Ps6–7.23, May 14–17, 2001. Davis B, Bagster D, (1989). The computation of view factors of fire models. 1. Differential targets. J. Loss Prev. Process Ind. 2, 224. Davis B, Bagster D (1990). The computation of view factors of fire models. 2. Finite targets. J. Loss Prev. Process Ind. 3, 327. Deaves DM (1987). Incorporating the effect of buildings and obstructions on gas cloud consequence analysis. In International Conference on Vapor Cloud Modeling, Boston (ed. JL Woodward), pp. 844–868. AIChE, CCPS, NY. Delano F et al. (2003). Introduction to LNG, p. 23. University of Houston Law Center, Institute for Energy, Law and Enterprise, Houston, TX. Dincer AK, Drake EM, Reid RC (1976). Boiling of liquid nitrogen and methane on water. The effect of initial water temperature. Intl. J. Heat Transf. 20, 177–180. Djuim NJ, Ott S, Nielsen M (1996). An evaluation of validation procedures and test parameters for dense gas dispersion models. J. Loss Prev. Process Ind. 9(5), 323. DNV (Det Norske Veritas) (2006). Sloshing analysis of LNG membrane tanks. Classification Note 30.9, June. DNV (Det Norske Veritas) (2007). Liquefied natural gas (LNG) receiving terminal and associated facilities, part 2—South Soko Island EIA, Section 13—Quantitative risk assessment. EIA Report by the Environmental Resources Management, Hong Kong (available on the EPD EIA website). DNV, Inc (Det Norske Veritas) (1997). User’s manual for PHAST, Version 5.1, software solutions for risk, safety and reliability. DNV, London. Dodge FT, Bowels E, White R (1980). Release rate of hazardous chemicals from damaged cargo vessels. Proceedings of the 1980 National Conference on Control of Hazardous Materials Spills. Dodge FT, Park JT, Buckingham JC, Maggott RJ (1983). Revision and experimental verification of the hazard assessment computer system models for spreading, movement, dissolution and dissipation of soluble chemicals spilled onto water. US Coast Guard Report CG-D-35-83, June, 1983. Drake EM, Jeje AA, Reid RC (1975). Transient boiling of liquefied cryogens on a water surface, I. Nitrogen, methane and ethane. Int. J. Heat Mass Transf. 18, 1361–1368, 1369–1375. Durr C (2004). The fundamentals of base load-LNG: Design and construction of LNG terminals. Training presentation, Feb 14, 2004. EC (2008). http://ec.europa.eu/enterprise/atex/guide/atexguidelines_august2008,pdf (accessed June 14, 2009). EC3 (1995). Eurocode 3: Design of steel structures, ENV1993-1-2: General rules for structural fire design. European Committee for Standardization, Brussels. Eggen JBMM (1998). GAME: Development of guidance for the application of the multi-energy method. HSE Contract Research Report 202/1998. Eisenberg, NA, Lynch CJ, Breeding RJ (1975). Vulnerability model, a simulation system for assessing damage resulting from marine spills. Report CG-D-136-75, Enviro Control Inc., Rockville, MD, NTIS AD-A015-245, Springfield, VA. EN 1473 (2005). Installation and equipment for liquefied natural gas—Design of onshore installations, prEN 1473:2005 (E) CEN/TC 282, Std Version 2.1c2. Meeting of September 2005, AFNOR Secretariat, Brussels, Belgium.

REFERENCES

349

Enger T, Harman DE (1972). LNG spillage on water, II. Final report on rapid phase transitions, Shell Pipeline Corp. Research and Development Laboratory, Houston, TX, Technical Progress Report No. 1-72. Also in Third Conference on LNG, Washington, DC, Sept 24–28, 1972. Ermak DL (1980). User’s manual for SLAB: An atmospheric dispersion model for denser-than-air releases, ACRL-MA-10-5607. Lawrence Livermore Laboratory, Livermore, CA, June. Ermak DL, Chan ST, Morgan DL, Morris LK (1981). A comparison of dense gas dispersion model simulations with Burro series LNG spill tests results, UCRL-86713. Lawrence Livermore National Laboratory, Livermore, CA. Ermak DL, Chan DL, Morris LK (1982c). A comparison of dense gas dispersion model simulations with Burro series LNG spill test results, J. Haz. Materials 6, 129–160. Ermak DL, Koopman RP (1982b). Results of 40-m3 LNG spills onto water. In Heavy gas assessment (ed. S Hartwig), pp. 163–179. Reidel, II, Battelle-Institute e.V., Frankfurt am Main, Germany. Ermak DL, Koopman RP, McRae TG, Hogan WJ (1982a). LNG spill experiments: Dispersion, RPT, and vapor burn analysis. Proceedings of the American Gas Association Section, Papers at Distribution Conference and Transmission Conference, Washington, DC, pp. T203–T209. Ermak DL, Chan ST (1985). A study of heavy gas effects Application V, on the atmospheric dispersion of dense gases, UCRL-92494, Rev. 1, 86713. Lawrence Livermore National Laboratory, Livermore, CA and Air Pollution Modeling and its Application V, Plenum Press, 723–742, 1985. Ermak DL, Chapman R, Goldwire HC Jr, Gouveia FJ, Rodean HC (1988). Heavy gas dispersion test summary report. Lawrence Livermore National Laboratory Report No. UCRL-21210, October 1988, Livermore, CA. Evans A, Puttock JS (1986). Experiments on the ignition of dense flammable gas clouds. 5th International Symposium on Loss Prevention and Safety Promotion in the Process Industries, Cannes, France, September 1986. Fannelop T, Jacobsen O (1982). Gravitational spreading of heavy gas clouds. Part I, instantaneous source in the absence of wind. Division of Aero- and Gas Dynamics, Norwegian Institute of Technology, Trondheim. Fannelop TK, Waldman GD (1972). The dynamics of oil slicks. Am. Inst. Aeronaut. Astronaut. J. 10, 506–510. Fay JA (1973). Unusual fire hazard of LNG tanker spills. Combust. Sci Technol. 7, 47–49. Fay JA, Lewis DH Jr. (1975). The inflammability and dispersion of LNG vapor clouds. Proceedings of the 4th International Symposium on Transport of Hazardous Cargoes by Sea and Inland Waterway, U.S. Coast Guard, Washington, DC, pp. 489–498. Fay JA, Lewis DH (1977). Unsteady burning of unconfined fuel vapor clouds. Combustion 16, 1397–1405. Fay JA (1978). On determining the size of LNG field tests. MIT Paper, September 15, 1978, 12 pp. Fay JA (1980). Gravitational spread and dilution of heavy vapor clouds. In SoftwarePractice and Experience. Vol. 1, pp. 471–494. Tapir, Trondheim, ISSN 00380644, June 24–27, 1980.

350

REFERENCES

Fay JA (1981). Scale effects in liquefied fuel vapor dispersion. Report DOE/EP-0032, U.S. Department of Energy, Washington, DC. Fay JA (1982). Some unsolved problems of LNG vapor dispersion. MIT Paper, March 24, 1982, 14 pp. Fay JA (2003). Model of spills and fires from LNG and oil tankers. J. Hazard. Mater. B96, 171–183. Fay JA (2007). Spread of large LNG pools on the sea. J. Hazard. Mater. 140, 541–551. Felbauer GF, Heigl JH, McQueen W, Whipp RH, May WG (1972). Spills of LNG on water- vaporization and downwind drift of combustible mixtures. Report No. EE61E-72, Esso Res. and Development Lab, Florham Park, New Jersey, November 24, 1972. FEMA (1990). Handbook of chemical hazard analysis procedures. Federal Emergency Management Agency, Washington, DC. FEMA (2002). World Trade Center Building Performance Study. Federal Emergency Management Agency, Report 403, May. FERC-ABS Consulting (2004). Consequence assessment methods for incidents involving releases from liquefied natural gas carriers. Report 131-04 GEMS 1288209, Contract # FERC04C40196, p. 33, Federal Energy Regulatory Commission, Washington, May 13, 2004. Ferrario C, Ditali S, Fiore R, Rota R (2007). LNG releases onto water—A comparison of the results from several models for dispersion analysis. I Chem E Symp. Ser. 153, 1–5. Festen L, Leo J (2009). Choices for LNG FPSO’s. Offshore Technology Conference, OTC 19884, Houston, TX, May 4–7, 2009. Ffooks R (1993). Natural gas by sea, 2nd Ed., 245 pp, Witherby, London. Fingas M (1998). In situ burning of oil spills: A historical perspective. In Proceedings of the In Situ Burning of Oil Spills Workshop (eds. W Walton, N Jason), pp. 55–61. National Institute of Standards and Technology, Gaithersberg, MD. Fire Department of New York City (1973). Report of Texas Eastern LNG tank fire and roof collapse, February 10, 1973, July 1973. Fleisher MT (1980). SPILLS: An evaporation/air dispersion model for chemical spills on land. Shell Development Company Report, Westhollow Research Center, Houston, TX. Fujii N (1995). Various styles of intense multiphase interactions viewed from volcanic explosions. Proceedings of the U.S. National Science Foundation Japan (Japan Society of the Promotion of Science) Joint Seminar, Santa Barbara, CA, June 9–13, 1995. Gaul D, Young L (2003). US LNG markets and uses, pp. 1, 11. Energy Information Administration (EIA), Washington, DC, Jan. 2003. Geng Q, Holmer I, Hartog DEA, Havenith G, Jay O, Malchaire J, Piette A, Rimtamaki H, Rissanen S (2006). Limit values for touching cold surfaces with the fingertip. Occup. Hyg. Annals of Occupational Hygiene 50(8), 851–862, Nov. 2006. Glasstone S, Dolan PJ (1977). The effects of nuclear weapons, 3rd ed. Tonbridge Wells, Castle House. Glenco JM, Lemmon AW (1970). Trans. Am Foundrymen’s Sci Direct 70, 317– 323.

REFERENCES

351

Global Gas Trade (2008). http://www.globalgastrade.com/globallng.asp/ (accessed May 1, 2008). Goldwire HC et al. (1983). Coyote series data report, LLNL/NWC 1981 LNG spill tests dispersion, vapor burn and rapid phase transition, Vols. 1 and 2, UCID-19953. Lawrence Livermore National Laboratories, Livermore, CA. Gordon S, McBride B (1996). NASA-Lewis Chemical Equilibrium Program CEA, Maritime Security. NASA Report 1311, Part 1, 1994 and Part 2, December 12, 1996. Government Accountability Office (GAO) (2007). Report to Congressional Requesters, public safety consequences of a terrorist attack on a tanker carrying liquefied natural gas need clarification. Report GAO-07-316, Washington, DC, February, 2007. Greenspan A (2007). Age of turbulence, pp. 450–451. Penguin Press, New York. Greenwald GB (1998). Liquefied natural gas: Developing and financing international energy projects, pp. 14–22, 71. Kluwer Law International, Cambridge, MA. Greenway HDS (2003). Is it safe? Boston Globe Magazine, July 27, 2003. Grumer J, Strasser A, Kubala TA, Burgess DS (1961). Uncontrolled diffusive burning of some new liquid propellants. Fire Res. Abs. Rev. 3, 159. Hankin RKS (2003). Heavy gas dispersion: integral models and shallow layer models. J. Hazard. Mater. 103, 1–10. Hankin RKS, Britter RE (1999). TWODEE: The Health and Safety Laboratory’s shallow layer model for heavy gas dispersion, Part 1. Mathematical basis and physical assumptions. J. Hazard. Mater. 66, 211–226. Hanna Consultants (2005). Source term estimation methods for releases of hazardous chemicals to the atmosphere due to accidental and terrorist incidents at industrial facilities and during transportation, Task 1 Report: Matrix of toxic industrial chemical (TIC) emissions scenarios and needs. Draft report number P081-1, U.S. Defense Advanced Research Projects Agency (DARPA), Arlington, VA, July 27. Hanna SR, Briggs GA, Hosker RP (1982). Handbook on atmospheric diffusion, pp. 6–8, DOE/TIC 11223. Department of Energy, 6–8. Hanna SR, Strimaitis DG, Chang JC (1991). Hazard response modeling uncertainty (a quantitative method), Vol II: Evaluation of commonly-used hazardous gas dispersion models. Final report, Sigma Research Corp., April 1989–1991. Hanna SR, Chang JC, Strimaitis DG (1993). Hazardous gas model evaluation with field observations. Atmos. Environ. 27A(15), 2265–2285. Hanna SR, Drivas PJ, Chang JC (1996). Guidelines for use of vapor cloud dispersion models. Center for Chemical Process Safety (CCPS), AIChE, New York. Hanna SR, Britter RE (2001). Wind flow and vapor cloud dispersion at industrial and urban sites. Center for Chemical Process Safety (CCPS), AIChE, New York. Hanna SR, Hansen OR, Dharmavaram S (2004). FLACS CFD air quality model performance evaluation with Kit Fox, MUST, Prairie Grass and EMU observations. Atmos. Environ. 38, 4675–4687. Hansen OR, Renoult J, Sherman MP, Tieszen SR (2005). Validation of FLACShydrogen CFD consequence prediction model against large scale H2 explosion experiments in the flame facility. International Conference on Hydrogen Safety, Pisa, Italy. Hardee HC, Lee DO (1973). Thermal hazard from propane fireballs. Transport Planning Technol. 2, 121.

352

REFERENCES

Harris RJ, Wickens MJ (1989). Understanding vapor cloud explosions – An experimental study. Institution of Gas Engineers and Managers, Communication 1408, Kegworth, Derbyshire, UK. Hartwig S (1984). Improved understanding of heavy gas dispersion due to the analysis of the Thorney Island trials data. J. Hazard. Mater. 11, 417–423. Hasegawa K, Sato K (1977). Study on the fireball following steam explosion of n-pentane, Loss Prevention and Safety Promotion in the Process Industries, 2, p. 297. Hasegawa K, Sato K (1978). Tech. Memo. 12, Fire Research Institute of Japan, Tokyo. Havens J, Venart J (2008). Fire performance of LNG carriers insulated with polystyrene foam. J. Hazard. Mater. 158(2–3):273–279. Havens JA, Spicer TO (1985). Development of an atmospheric dispersion model for heavier-than-air gas mixtures. U.S. Coast Guard Report No. CG-D-23-85,Washington, DC. Havens JA, Spicer TO (1988). A dispersion model for elevated dense gas jet chemical releases, Environmental Protection Agency. http://www.epa.gov/scram001/ dispersion_alt.htm (accessed July, 2009) Havens JA, Spicer TO (1990). LNG vapor dispersion prediction with the DEGADIS dense gas dispersion model. Gas Research Institute Report No. 89-0242. Havens JA, Spicer TO (2005). LNG vapor cloud exclusion zones for spills into impoundments, Process Safety Progress, 24(3) 181-186. Havens JA, Spicer TO (2006a). Error in FERC environmental impact statement determinations of LNG vapor cloud exclusion zones: Failure to account for air mixing in vapor impoundments. 9th Annual Symposium, Mary Kay O’Connor Process Safety Center, Texas A&M U niversity, College Station, TX, October 24–25, 2006. Havens JA, Spicer TO (2006b). Vapor dispersion and thermal hazard modeling. Final Topical Report to Gas Technology Institute under Sub-Contract K100029184, October 2006. Havens JA, Spicer TO (2007a). United States regulations for siting LNG terminals: Problems and potential. J. Hazard. Mater. 140, 439–443. Havens JA, Spicer TO (2007b). Wind tunnel studies of LNG vapor dispersion from impoundments. AIChE National Spring Meeting, Houston, TX, April. Health and Safety Executive (HSE) (1976). Executive Report. The explosion at the Appleby-Frodingham steelworks, Scunthorpe, U.K., HM Stationery Office, London, November 4, 1976. Health and Safety Executive (HSE) (1978). Canvey—An investigation of potential hazards from operations in the Canvey Island/Thurrock area. U.K. Health and Safety Executive, HMSO, London. Health and Safety Executive (HSE) (1981a). Managing safety. Her Majesty’s Stationery Office (HMSO), London. Health and Safety Executive (HSE) (1981b). Canvey—A second report—An investigation of potential hazards from operations in the Canvey Island/Thurrock area 3 years after publication of the Canvey Report. Health and Safety Executive, HMSO, London. Health and Safety Executive (HSE) (1984). Heavy gas dispersion trials Thorney Island 1982-3: Data for trials. Research and Laboratory Services Division, Sheffield.

REFERENCES

353

Health and Safety Executive (HSE) (2001). Reducing risks, protecting people—HSE’s decision-making process. HSE Sheffield, UK, HM Stationery Office, London. Health and Safety Executive (HSE) (2005). Hazardous installations directorate, vapor cloud explosions. Semi Permanent Circular/Technical (LD1-4)/05, File 278, August 2005. Health and Safety Laboratory (HSL) (2008). Response to second set of comments/ questions (10-1-2008) on the LNG vapor dispersion MEP received from the technical panel of the National Association of State Fire Marshals (10-8-2008), October 22, 2008, to AcuTek, Washington, DC, p. 17. Heestand J, Meader JM (1983). A predictive model for rollover in stratified LNG tanks. AIChE J., 29(2), 199–207. Heikes KE, Gay RA, Small RD (1990a). Firewinds, turbulence and plume rise: Hill and Wickstead Township area fires. PRS Report 2064 for U.S. Department of Commerce, Natiojnal Institure of Standards and Technology, Gaithersburg, MD, by PacificSierra Research Corporation, Los Angeles, CA, April 7, 1990, pp. 4, 13, 23. Heikes KE, Small RD (1990b). Area fire and plume behavior. Report by Pacific-Sierra Research Corporation, Los Angeles, CA, 1990. Henriques FC (1947) Studies of thermal injury. V, The predictability and the significance of thermally induce rate processes leading to irreversible epidermal injury. Arch. Pathol. Lab. Med. 43, 489. Heskestad G (1983). Luminous heights of turbulent diffusion flames. Fire Saf. J. 5, 103–108. Hicks E, Menzies DC (1965). in Proceedings of conference on safety, fuels and core design in large fast power reactors, Argonne National Laboratory Report 7120, pp 654–670. Hightower MM, Gritzo L, Luketa-Hanlin A, Covan J, Tieszen S, Wellman G, Irwin M, Kaneshige M, Melof B, Morrow C, Ragland D (2004) Guidance on risk analysis and safety implications of a large liquefied natural gas (LNG) spill over water, Report SAND2004-6258, Sandia National Laboratories, Albuquerque, NM and Livermore, CA, for US Dept of Energy, Dec 2004, Hinshaw JR (1957). Histologic studies of some reactions of skin to radiant thermal energy. ASME Paper 57-SA.21, American Society of Mechanical Engineers, New York. Hissong DW (2007). Keys to modeling LNG spills on water. J. Hazard. Mater. 140, 465–477. Hottell HC (1958). Certain laws governing diffusive burning of liquids. Fire Res. Abstr. Rev. 1, 41; Review of Blinov VI, Khudiakov GN (1957). Kokl. Akad. Nauk. SSSR 113, 1094. Hoult D (1972a). Oil spreading on the sea, Annual Review of Fluid Mechanics, 4, 341–368, Annual Reviews, Palo Alto, CA. Hoult D (1972b). The fire hazard of LNG spilled on water. In Proceedings of the Conference on LNG Importation and Terminal Safety, Terminal Safety, National Academy of Sciences, Boston, Mass., June 13–14, p. 87. National Academy of Sciences, Boston. Humbert-Basset R, Montet A (1972). Dispersion dans 1’atmosphere d’un nuage gazeux forme par epandage de GNL sur le sol’. Proceedings of the 3rd International Conference on LNG, Session VI, Paper No 4.

354

REFERENCES

Hymes I (1983). The physiological and pathological effects of thermal radiation, SRD R275. HM Stationery Office, London. Hymes I (1984). In The effects on people and structures of explosion, blast, thermal radiation and toxicity (ed. JI Petts), Planning, 593, Nov 2, 1984. IGC (1975). Code for the construction and equipment of ships carrying liquefied gases in bulk. Intergovernmental Maritime Consultative Organization (IMCO) (now in updated amendments and published by International Maritime Organization [IMO]), London, UK. IChemE (1994). Explosions in the process industry. Rugby, Warwickshire, UK. Institute for Energy, Law & Enterprise (IELE) (2003a). Commercial frameworks for LNG in North America. University of Houston, Houston, TX, p. 24. Institute for Energy, Law & Enterprise (IELE) (2003b). LNG safety and security. University of Houston, Houston, TX. International Maritime Organization (IMO) (2002). IMO adopts comprehensive maritime security measures. Press release, London, December 17, 2002. Irwin JS (1979). A Theoretical variation of the wind profile power law exponent as a function of surface roughness and stability. Atmos. Environ. 13(1), 191–194. ISO (1975). Fire resistance tests- Elements of building construction. ISO 834-1975. International Organization for Standardization, Geneva, Switzerland. Iversen I, Hellekleiv V (2009). LNG FPSO—Solving the supply shortfall. Offshore Technology Conference, OTC 20102, Houston, TX, May 4–7, 2009. Ivings MJ, Jagger SF, Lea CJ, Weber DM (2007a) Evaluating vapor dispersion models for safety analysis of LNG facilities. HSL (Health and Safety Laboratories) Harpur Hill, Buxton. Report MSU/2007/04 for Fire Protection Research Association, May 9, 2007. Ivings MJ, Webber D (2007b). On defining safety criterion for flammable clouds. Health and Safety Laboratory (HSL) Report HSL/2007/30, Harpur Hill, Buxton. Japan Gas Association (JGA) (1976). A study of the dispersion of evaporated gas and ignition of LNG pool resulted from continuous spillage of LNG. Japan Gas Association, Tokyo. Japan Institute of Safety Engineering (1979). Report of oil tank fire experiment (in Japanese). Japan Institute of Safety Engineering, Tokyo. Japan Institute of Safety Engineering (1982). Report on burning of petroleum fires (in Japanese). Japan Institute of Safety Engineering, Tokyo. Jazayeri B1 (1975). Impact cryogenic vapor explosions. MS thesis. MIT, Cambridge, MA. Jenkins DR, Timmers AC (1983). LNG plant safety in the light of recent experimental trials and theories. In Proceedings of the Seventh International Conference on Liquefied Natural Gas p. 6. Jakarta, Indonesia. Jensen NO (1983). On cryogenic liquid pool evaporation. J. Hazard. Mater. 3, 157–163. Jobson DA (1955). On the flow of compressible fluids through orifices. Proc. Inst. Mech. Eng. 169(37), 767–776. Johnson AD (1992). A model for predicting thermal radiation hazards from large scale LNG pool fires. In Major hazards onshore and offshore, Manchester, UK, Oct 20–22, 1992. In IChemE Symposium Series No 130, Rugby, p. 507.

REFERENCES

355

Johnson DW, Cornwell JB (2007). Modeling the release, spreading, and burning of LNG, LPG, and gasoline on water. J. Hazard. Mater. 140, 535–540. Johnson DW, Woodward JL (1999). RELEASE, a model with data to predict aerosol rainout in accidental releases, pp. 58–59, 80–82, 103. AIChE, Center for Chem Process Safety, New York. Kakko R (1990). Effects of the ventilation system and chemical reactions on the concentration of individual risk estimates of toxic materials. J. Loss Prev. Process Ind. 5(3), 145–152. Kalinin EK, Berlin II, Kostyuk VV, Nosova EM (1975). Heat transfer in transition boililng of cryogenic liquids. Adv. Cryog. Eng. 21, 273–277. Kantha LH, Phillips OM, Azad RS (1977). On turbulent entrainment at a stable density interface. J. Fluid Mech. 79, 753–768. Karlsson E, Huber U (1996). Influence of desorption on the indoor concentration of toxic gases. J. Hazard. Mater. 49, 15–27. Kataoka H (1981). Report on LNG anti-disaster experiment test, Report by Tokyo Gas Co. Ltd, Tokyo, Japan. Katz DL, Sliepcevich DM (1973). Water explosions. Report No. CG-D60-74, National Academy of Sciences. Khalil A, Fouad MA, Kamel MM (1988). Modeling of confined cryogenic fuels over water. 5th International Symposium on Multi-phase Transport and Particulate, Miami, FL, pp. 187–199. Kim H, Koh JS, Kim Y, Theofanous TG (2005). Risk assessment of membrane type LNG storage tanks in Korea-based on fault tree analysis. Korean J. Chem. Eng. 22(1), 1–8. Kinney GF, Graham KJ (1985). Explosive shocks in air, 2nd ed. Springer, Berlin. Kjøk and Lia (2001). Terrorism And Oil—An Explosive Mixture? A survey of terrorist and rebel attacks on petroleum infrastructure 1968–1999. FFI/RAPPORT2001/04031. Klimenko VV (1981). Film boiling on a horizontal plate- New correlation. Int. J. Heat Mass Transf. 24, 69–79. Kneebone A, Prew LR (1974). Shipboard jettison tests on LNG on to the sea. Proceedings of the 4th International Conference on LNG, Gastech ’74, Algiers, November 13–14, 1974. Kolluru R, Bartell S, Pitblado R, Stricoff S (1996). Risk assessment and management handbook for environmental, health and safety professionals. McGraw-Hill, New York. Koopman RP, Baker J, Cederwall RT, Goldwire HC Jr., Hogan WJ, Kamppinen LM, Kiefer RD, McClure JW, McCrae TG, Morgan DL, Morris LK, Spann MW Jr., Lind CD (1982a). Burro Series Date Report, LLNL/NWC 1980 LNG Spill Tests, UCID-19075, Lawrence Livermore Laboratory, Livermore, CA, Dec. https://e-reports-ext.llnl.gov/pdf/194414.pdf and https://e-reportsext.llnl.gov/pdf/ 194606.pdf. Koopman RP, Cederwall RT, Ermak DL, Goldwire HC Jr., Hogan WJ, McClure JW, McCrae TG, Morgan DL, Rodean HC, Shinn JH (1982b). Analysis of Burro series 40 m3 LNG Spill experiments. J. Hazard. Mater. 6, 43–83.

356

REFERENCES

Koopman RP, Bowman BR, Ermak DL (1979). Data and calculations on 5 m3 LNG spill tests. UCRL-52876, Lawrence Livermore Laboratory, Livermore, CA, October 1, 1979. Koopman RP, Bowman BR, Ermak DL (1980). Data and calculations on 5 m3 LNG spill tests. Liquefied Gaseous Fuels Safety and Environmental Assessment Program, Second Status Report, DOE/EV-0085, U.S. Department of Energy, October 1980. Koopman RP, Ermak DL (2007). Review: Lessons learned from LNG safety research. J. Hazard. Mater. 140, 412–428. Krane (2000). Piracy makes a comeback. APBnews.com, June 9, 2000. Kranenburg C (1984). Wind-induced entrainment in a stably stratified fluid. J. Fluid Mech. 145, 253–273. Kuo JC (2007). How big is the area required for LNG receiving terminals and their design considerations. 7th Topical Conference on Natural Gas Utilization, AIChE Spring Meeting, Houston, TX, April 2007. Laubenstein L, Mains C, Jost A, Tagg R, Bjørneboe NK (2001). Updated probabilistic extents of damage based on actual collision data. 2nd International Conference on Collision and Grounding of Ships, Copenhagen. Lautkaski R (1992). Validation of flame drag correlations with data from large pool fires. J Loss Prev. Process Ind. 5(3), 175–180. Lea CJ (2002) A review of the state of the art in gas explosion modeling. U.K. Health and Safety Laboratory (HSL), Her Majesty’s Stationary Office, London. Lee YM, Cho TI, Lee JH, Kwon O (2008). Trends and technologies in LNG carriers and offshore LNG facilities. Paper 19339, Offshore Technology Conference, Houston, TX, May. Lees FP (ed) (1996). Loss Prevention in the Process Industries, 2nd Ed., Elsevier Butterworth Heineman, Amsterdam. Lehr W, Simecek-Beatty D (2004). Comparison of hypothetical LNG and fuel oil fires on water. J. Hazard. Mater. 107, 3–9. Levich VG (1962). Pysicochemical hydrodynamics. Prentice-Hall, Eaglewood Cliffs, NJ. Lewis B, von Elbe G (1987). Combustion, flames, and explosions of gases, 3rd ed. Academic Press, London, UK. Lienhard JH, Day JB (1970). The breakup of superheated liquid jets. J. Basic Eng, Trans AIME 515, 168. Lihou DA, Maund JK (1982). Thermal radiation hazards from fireballs. The Assessment of major hazards, p. 191. Institution of Chemical Engineers, Rugby, U.K. Lind CD, Whitson JC (1977). Explosion hazards associated with spills of large quantities of hazardous materials, Phase II, Report No CG-D-85-77, July, 1977. Lind CD, Whitson JC (1979). China Lake spill tests. Report L., Liquefied Gaseous Fuels Safety and Environmental Control Assessment Program: A Status Report, DOE/ EV-0036, Department of Energy, Washington, DC. Lindemark T, Austefjord HN, Sele HO, Urm HS, Lee KJ, Ha TM (2006) CSA-2 analysis of a 216K LNGc membrane carrier, Proceedings of the ISCOT/RINA: International Conference on Design, Construction and Operation of Natural Gas Carriers and Offshore Systems, Publisher: Royal Institution of Naval Architects, Busan, Korea, September 14–15, 2006.

REFERENCES

357

Lindemark T, Kamsvag F, Valsgard S (2004) Fatigue analysis of gas carriers, Proceedings of the ISCOT/RINA: International Conference on Design, Construction and Operation of Natural Gas Carriers and Offshore Systems, Publisher: Royal Institution of Naval Architects, London, September. Lipsett SG (1966). Explosion from molten materials and water. Fire Technol. 2, 118. LNG World Shipping (2006). LNG spill tests, Lawrence Livermore National Lab. Report UCRL 52876, Livermore, CA, December 22, 2006. Lohmeyer A, Meroney R, Plate E (1980). Model investigations of the spreading of heavy gases released from an instantaneous volume source at the ground. 11th NATO/CCMS International Technical Meeting on Air Pollution Modeling and its Applications, Amsterdam. Luketa-Hanlin A (2006). A review of large-scale LNG spills: Experiments and modeling. J. Hazard. Mater. A132, 119–140. Luketa-Hanlin A (2007). Pool fire modeling: Radiation. CLNG Workshop on LNG pool fire modeling, Washington, DC, June 12–13, 2007. Luketa-Hanlin A, Koopman RP, Ermak DL (2007). On the application of computational fluid dynamics code for liquefied natural gas dispersion. J. Hazard. Mater. 140(3), 504–517. Luketa-Hanlin A, Hightower MM, Attaway S (2008). Breach and safety analysis of spills over water from large liquefied natural gas carriers, Sandia National Laboratories report SAND2008-3153, Albuquerque, New Mexico. Malvos H, Raj PK (2006). Details of 35 m diameter LNG fire tests conducted in Montoir, France in 1987, and analysis of fire spectral and other data. AIChE Winter Conference, Orlando, FL, April 23–27, 2006. Mannan S (ed.) (2005). Lees loss prevention in the process industries, Elsevier Butterworth Heineman 3rd ed., pp. 17/122, 17/133, 17/167–17/174, v1 Sections 16.13.11 view Factors, 16.15 Fireballs, 16.18 Flares, 16.19 Jet Flames, 16/168 Atmospheric Transmissivity. Elsevier Butterworth Heinemann, Oxford. Marshall VC (1987). Major chemical hazards. Ellis Horwood, Chichester, U.K. May MG, McQueen W, Whipp RH (1973a). AGA 1973 Operating Section Proceedings, Paper 73-D-9, p. D-143. May WG, McQueen W, Whipp RG (1973b). Dispersion of LNG spills. Hydrocarbon Process, pp. 105–109. May WG, McQueen W (1973). Radiation from large liquefied natural gas fires. Combustion Sci. Technol. 7, 51–56. McGrattan KB, Baum HR, Hamins A (2000). Thermal radiation from large pool fires. National Institute of Standards and Technology, NISTIR 6546, U.S. Department of Commerce, Washington, DC, November 2000. McGrattan K, Baum H, Rehm R, Mell W, McDermott R (2009). Fire Dynamics Simulator (Version 5) technical reference guide volume 1: Mathematical model. NIST Special Publication 1018-5. US Dept. of Commerce, Washington DC. McQuaid J (1976). Some experiments on the structure of stably-stratified shear flows. Tech Paper p. 21, Safety in Mines Research Establishment, Sheffield. McQuaid J (1983). Large scale experiments on the dispersion of heavy gas clouds, IUTAM Symposium, Delft, in Ooms G, Tennekes H (Eds) (1984). Atmospheric dispersion of heavy gases and small particles, Springer, Berlin, p. 129.

358

REFERENCES

McRae TG (1982). Preliminary analysis of RPT explosions observed in the LLNL/ NWC LNG spill tests. UCRL-87564, Lawrence Livermore National Laboratory, Livermore, CA. McRae TG, Goldwire HC Jr., Koopman RP (1984). Analysis of large-scale LNG/water RPT explosions. Lawrence Livermore National Lab. Report, UCRL 91832, October 1984, Livermore, CA. Melhem GA, Kalelkar AS, Ozog H, Saraf S (2006). Managing LNG risks: separating the facts from the myths, ioMosaic Corp. white paper. Accessed June 14, 2009. Mercx WPM (1992). Large scale experiments investigation into vapor cloud explosions, comparison with the small scale DISCOE tests. 7th International Symposium on Loss Prevention and Safety Promotion in the Process Industries, Italy, May 1992. Mercx WPM, van den Berg AC, van Leeuwen D (1998). Application of correlations to quantify the source strength of vapor cloud explosions in realistic situations, Final report for the project “GAMES”, TNO Report PML 1998-C53, Rjsuijk, The Netherlands, October, 1998, 156 p. Mine Safety and Health Administration (MSHA) (2006). Mine Improvement and New Emergency Response Act of 2006. U.S. Department of Labor, Washington DC. MIT LNG Research Center (1977). Flameless vapor explosions. DOE-OSTI No. 6775451, Cambridge, MA. Mizner GA, Eyre JA (1982). Large scale LNG and LPG pool fires. IChemE Symp. Ser. 71, 147–163. Mizner GA, Eyre JA (1983). Radiation from liquefied gas fires on water. Combustion Sci. Technol. 35, 33–57. Moen IO, Donato M, Knystautas R, Lee JH (1980). Flame acceleration due to turbulence produced by obstacles. Combust. Flame 39, 21–32. Mollenkamp CR, Bradley MM (1991). Numerical simulation of the dynamics and microphysics of prescribed forest burns. 22nd Conference on Fire and Forest Meteorology, Missoula, Montana, April 16–19, 1991. Montoir Tests (1987). See Nedelka et al. (1989) and Malvos and Raj (2006). Moorhouse DJ (1982). Scaling criteria for pool fires derived from large-scale experiments. IChemE Symp. Ser. 71, 165–179. Moorhouse DJ, Pritchard MJ (1982). Thermal radiation hazards from large pool fires and fireballs—A literature review. IChemE Symp. Ser. 71, 397–428. Moosemiller M (2009). Development of algorithms for predicting ignition probabilities and explosion frequencies, Paper 5A. AIChE Spring Meeting, Tampa, FL, April 26–30, 2009. Morgan DL Jr. (1987). Dispersion phenomenology of LNG vapor in the Burro and Coyote LNG spill experiments. UCRL-91741. Lawrence Livermore National Laboratory, Livermore, CA, Feb. Morgan DL Jr., Morris LK, Chan ST, Ermak DL, McRae TG, Caldwall RT, Koopman RP, Goldwire HC Jr., McClure JW, Hogan WJ (1984). Phenomenology and modeling of liquefied natural gas vapor dispersion. Lawrence Livermore National Lab Report UCRL 53581, Livermore, CA. Morrison DS, Richardson A (1990). An experimental study of stratified layers and rollover in LNG, Proc. of Conf. on Low Temperatures and Cryogenics, LTEC90, Southampton, UK.

REFERENCES

359

Morton BR, Taylor GI, Turner JS (1956). Turbulent gravitational convection from maintained and instantaneous sources. Proceedings of Royal Society of London, Series A (Math. and Physical Sci) 234, pp 1–23, Jan. 24, 1956. Moullieau Y, Champassith A (2009). CFD simulations of atmospheric gas dispersion using the fire dynamics simulator (FDS). J. Loss Prev. Process Ind. 22, 316–323. Mudan KS (1984). Thermal radiation hazards from hydrocarbon pool fires. Prog. Energy Combust. Sci. 10, 59–81. Mudan KS (1987). Geometric view factors for thermal radiation hazard assessment. Fire Saf. J. 12, 89–96. Mudan KS (1989). Evaluation of fire and flammability hazards, Chapter 14. In Encyclopedia of environmental control technology, Thermal Treatment of Hazardous Wastes, Vol. 1. Gulf Publishing Company, Houston, TX, pp. 366–427. Mudan KS, Croce PA (1984). Thermal radiation model for LNG trench fires. ASME Winter Annual Meeting, New Orleans, LA. Mudan KS, Croce PA (1995). Fire hazard calculations for large open hydrocarbon fires. In The SFPE Handbook of Fire Protection Engineering 2nd Edition (ed. D. Drysdale), pp. 2-45 to 2-87. National Fire Protection Association, Quincy, MA. Murgai MP (1976).Natural convection from combustion sources. Oxford and IBH Publishing, New Delhi. NASFM (2005). Liquefied natural gas: An overview of the LNG industry for fire marshals and emergency responders. National Association of State Fire Marshals, Washington, DC. NASFM (2009). Review of the LNG vapor dispersion model evaluation protocol. Report of the Technical Panel, National Association of State Fire Marshals, Washington, DC by AcuTech, Alexandria, VA, 2009. National Academy of Sciences (NAS) (1975). Matrix of electrical and fire hazard properties and classification of chemicals. Report to the U.S. Coast Guard, Contract No. DOT-CG-41680-A. National Materials Advisory Board (NMAB) (1982). Classification of gases, liquids and volatile solids relative to explosion-proof electrical equipment. US Dept. of Commerce, Washington, DC, Aug, 1982. Nedelka D (1990). Calculation of radiation effects from LNG fires. EUROGAS 90 Proceedings from the European Applied Research Conference on Natural Gas, Trondheim, May 28–30, 1990. Nedelka D, Moorhouse DJ, Tucker RF (1989). The Montoir 35 m diameter LNG pool fire experiments. Proceedings of the 9th International Conference on LNG, Vol 2, Nice, France, October 17–20, 1989, Section III, Paper 3, 1–23. NFPA 59A (National Fire Protection Agency) (2006). Standard for the production, storage, and handling of liquefied natural gas (LNG). National Fire Protection Association, Quincy, MA. Nielsen M, Ott S (1996). A collection of data from dense gas experiments, Riso-R-845 (EN), Riso National Laboratory, Denmark, March 1996. http://www.risoe.dtu.dk/ research/sustainable_energy/energy_systems/projects/rediphem.aspx (accessed July, 2009). NORSOK Standard (2001). Risk and emergency preparedness analysis, Standard Z-013, Rev 2. Sept 2, 2001 Norwegian Technology Center, Olso, Norway.

360

REFERENCES

Notorianni KA, Evans DD, Walton WD, Madrzykowski D, Lawson JR (1993). Smoke production from large oil pool fires. In Interflame ’93, Fire Safety, 6th International Fire Conference, Oxford, U.K. (ed. CA Franks), pp. 111–119. Interscience Communications Ltd, London. NTSB (National Transportation Safety Board) (1980). Report on Columbia LNG Corporation Explosion and Fire, Cove Point MD, October 6, 1979, NTSB-PAR-80-2, April 16, 1980. Opschoor G (1980a). The spreading and evaporation of LNG and burning LNG spills on water. J. Hazard. Mater. 3, 249. Opschoor G (1980b). Investigations into the evaporation of liquefied gases spreading on land. TNO Report No. 80-010884, Apledoorn, Holland. O’Sullivan S, Jagger S (2004). Human vulnerability to thermal radiation offshore. Health and Safety Laboratory, HSL/2004/04, HM Stationery, London. Otterman B (1975). Analysis of large LNG spills on water, part 1: Liquid spread and evaporation. Cryogenics 15, 455–460. Paik JK, Choe IH, Thayamballi AK (2001). An accidental limit state design of spherical type LNG carrier structures against ship collisions, 2nd International Conf. on Collision and Grounding of Ships, pp. 221–232. Copenhagen. Palmer TY (1981). Large fire winds, gases and smoke. Atmos. Environ. 15(10/11), 2079–2090. Paranouskis MC, Lind CD, Raj PK, Cece JM (1980). Vapor cloud explosion study. In 5th International Conference of Liquefied Natural Gas, Vol. 2, Session 3, Paper 12. Parfomak PW (2004). Liquefied natural gas (LNG) infrastructure security: Background and issues for Congress, CRS (Congressional Research Service). Report for Congress, pp. 14–17, Library of Congress, Washington, DC, February 2, 2004. Park G (2003). List of planned LNG ventures grows. Petroleum News, Haines, AK, June 29. Pasquill F (1974). Atmospheric diffusion, 2nd ed. Ellis Horwood Ltd, Chichester, UK. Pasquill F, Smith FB (1983). Atmospheric diffusion, 3rd ed., p. 339. John Wiley & Sons, New York. Pastoor W, Lund K, Tveitnes T (2009). The LNG producer- A generic design with great adaptability. Offshore Technology Conference, OTC 1984/5, Houston, TX. Petroplus International N.V. web site. Energy for Wales: LNG frequently asked questions. Amsterdam. http://www.petroplusinternational.com. Pettitt GN, Harms JD, Woodward JL (1994). Post-mortem risk modeling of the Mexico City disaster. Pacific Soc. Applied Math. Conference, San Diego, CA, March 20–25, 1994. Pietersen CM (1985). Analysis of the LPG incident in San Juan Ixhuatepec, Mexico City. Rep. 85-0222, TNO, Apeldoorn, November 19, 1984. Pitblado R, Baik J, Raghunathan V (2004). LNG decision making approaches compared. Mary Kay O’Connor Process Safety Symposium, Texas A&M University, College Station, TX, October 26–27, 2004. Pitblado RM, Baik J, Hughes GJ, Ferro C, Shaw SJ (2005). Consequences of Liquefied Natural Gas Marine Incidents, Process Safety Progress, 24(2), 108–114, June. Pitblado R, Baik J, Raghunathan V (2006). LNG decision making approaches compared. J. Hazard. Mater. 130, 148–154.

REFERENCES

361

Pitblado RM (2007). Potential for BLEVE associated with marine LNG vessel fires. J. Hazard. Mater. 140, 527–534. Pitblado RM, Skramstad E, Danielsen HK, Hysing T (2008). Energy based risk methodology for collision protection of LNG carriers. 8th Topical Conference on Natural Gas Utilization, AIChE Spring Conference, New Orleans, LA, April 6–10. Pitblado RM, Bain B, Falc KA, Litland K, Spitzenberger C (2009). Frequency data and modification factors used in international QRA studies, 8th World Congress of Chemical Engineering, Montreal, August. Planas-Cuchi E, Gasulla N, Ventosa A, Casal J (2004). Explosion of a road tanker containing liquefied natural gas, J. Loss Prevention in Proc. Industries, 17, 315–321. Prince AJ (1985). Details and results of spill experiments of cryogenic liquids onto land and water. Joint HSE and UKAEA, SRD Report, HSE/SRD/PD058/WP4, April. Prugh RW (1994). Quantitative evaluation of fireball hazards. Process Saf. Prog. 13(2), 83–91. Puttock JS, Blackmore DR, Colenbrander GW (1982a). Field experiments on dense gas dispersion. J. Hazard. Mater. 6(1–2), pp. 13–41; also in Dense gas dispersion (eds. RE Britter, RF Griffiths), p. 190. Elsevier, Amsterdam. Puttock JS, Colenbrander GW, Blackmore DR (1982b). Maplin sands experiments 1980: Dispersion results from continuous releases of refrigerated liquid propane and LNG, NATO/CCMS. 13th International Technical Meeting on Air Pollution Modeling and Its Application, September; also in Heavy gas and risk assessment II (ed. S Hartwig), Reidel, May, 1982, Dordrecht; also in Air pollution modeling and its application III (ed. C DeWispelaere), Plenum Press, New York, 1983. Puttock JS, Blackmore DR, Colenbrander GW, Davis PT, Evans A, Homer JB, Redfern JJ, Van’t Sant WC, Wilson RP (1984). Spill tests of LNG and refrigerated liquid propane on the sea, Maplin sands, experimental details of the dispersion tests. Shell International Research Report TNER.84.046, May. Puttock JS (1995). Fuel gas explosion guidelines—The congestion assessment method. 2nd European Conference on Major Hazards On- and Off-shore, September 24–26, 1995, Manchester; also IChemE Symposium Series No. 139. Puttock JS (1999). Improvements in guidelines for prediction of vapor-cloud explosions. International Conference and Workshop on Modeling the Consequences of Accidental Releases of Hazardous Materials, San Francisco, CA, September/ October 1999. Qiao Y, West HM, Mannan MS, Johnson DW, Cornwell JB (2006). Assessment of the effects of release variables on the consequences on LNG spillage onto water using FERC model. J. Hazard. Mater. 130, 155–162. Quest Consultants, Inc. (2001, 2003). Modeling LNG spills in Boston Harbor, Letter from John Cornwell to Don Juckett of the U.S. Department of Energy, October 2 and 3, 2001; Letter from John Cornwell to Clifford Tomaszewski of the Office of Natural Gas & Petroleum Import & Export Activities, November 17, 2003. Quillen D (2002). ChevronTexaco Corp., LNG safety myths and legends. Presentation to the Natural Gas Technology Conference, Houston, TX, May 14–15, 2002, p. 18. Quintiere JG (1990). Canadian mass fire experiments, 1989. Report NISTIR 4444, National Institute of Standards Technology, November 1990. Raj PK, Atallah S (1974). Thermal radiation from LNG fires. Adv. Cryog. Eng. 20, 143.

362

REFERENCES

Raj PK, Kalelkar AS (1974). Assessment models in support of the hazard assessment handbook. Technical Report CG-D-65-74, U.S. Coast Guard, Springfield, VA, NTIS AD776617, January 1974. Raj PK, Emmons HW (1975). On the burning of a large flammable vapor cloud. Western & Central States Section of the Combustion Institute, San Antonio, TX, April 1975. Raj PK (1977). Calculation of thermal radiation hazards from LNG fires- A review of the state-of-the-art. American Gas Association Transmission Conference, St Louis, Missouri, May 16, 1977, p. T135. Raj PK, Moussa NA, Aravamudan KS (1979). Experiments involving pool and vapor fires from spills of LNG on water. USCG Report CG-D-55-79, 20590, NTIS ADA077073, Washington, DC, March. Raj PK (2005a). LNG fire thermal radiation—Modeling issues and hazard criteria revisited. Process Saf. Prog., AIChE 24(3). Raj PK (2005b). LNG fire spectral data and calculation of emissive power. International Symposium on Process Safety, Mary Kay O’Connor Center, Texas A&M University, College Station, TX, October 25–26, 2005. Raj PK, Technology and Management Systems (2006). Spectrum of Fires in an LNG Facility, Assessments, Models and Consideration in Risk Evaluations, Report for US Dept. of Transportation, Pipeline and Hazardous Materials Safety Admin., Washington, DC, Contract DTRS56-04-T-005, Dec. 5, 2006, pp. 3–8. Raj PK (2007). LNG fires: A review of experimental results, models and hazard prediction challenges. J. Hazard. Mater. 140, 444–464 (plus errata J. Hazard. Mater. 143, 603). Raj PK (2008a). Substantiation provided for comment #59A-29, 2008 Annual Revision Cycle. Report on Comments, NFPA59A-8 Committee, February. Raj PK (2008b). Use of risk analysis for LNG facility siting in the US: Changes in NFPA 59A and in future regulations, a road map and required actions, 11th annual symposium, Mary Kay O’Connor Process Safety Center, Oct. 28–29, 2008; College Station, TX, Texas A&M University. Raj PK (2008c) White paper attachment to Temporary Interim Amendment (TIA) to NFPA 59A proposed by KJ Katulak, November 11, 2008. Rasmussen NC (1975). Reactor safety study: An assessment of accident risks in US commercial nuclear powerplants, Appendix III, Failure Data, NUREG 75/014. Atomic Energy Commission, Report WASH-1400, Washington, DC, October, 1975. Reid RC (1980). Boiling of LNG on typical dike floor materials. GRI 79/0026, MIT LNG Research Center, Cambridge, MA, March 1980. Reid RC (1983). Rapid phase transitions from liquid to vapor. Adv. Chem. Eng. 12, 105–203. Reid RC, Smith KA (1978a ). Behavior of LPG on water. Hydrocarbon Process. April, 117–121. Reid RC, Smith KA (1978b). Confined boiling rates of liquefied petroleum gas on water. EE-77-S-02-4558. MIT, Boston. Rein RG Jr., Sliepcevich CM, Welker JR (1970). Radiation view factors for tilted cylinders. J. Fire Flammability 1, 140–153.

REFERENCES

363

Rew PJ, Hulbert WG (1996). Development of pool fire thermal radiation model. HSE Contract Research Report 96/1996, Health and Safety Executive, U.K., ISBN 0-7176-1084-5. Rijnmond Public Authority (1982). Risk assessment of six potentially hazardous industrial objects in the Rijnmond area. D. Reidel Publishing Company, Dordrecht. Riley T, Riley H. http://www.TimRileyLaw.com (accessed Jan, 2009). RIVM (2009). Reference manual BEVI risk assessments version 3.1—Module B. (accessed June, 2009). Roberts AE (1981/1982). Thermal radiation hazards from releases of LPG from pressurized storage. Fire Saf. J. 4(3), 197. Roberts AE (1982). The effect of conditions prior to loss of containment on fireball behavior. The assessment of major hazards, p. 181. Institution of Chemical Engineers, Rugby, U.K.. Rodean HC, Cederwall RT (1982). Analysis of turbulent wind velocity and gas concentration fluctuations during the Burro series 40 m3 LNG spill experiments, Lawrence Livermore National Labs, report UCRL 53353, Livermore, CA, Nov. 4, 1982. Rodian HC (1983). Effects of a spill of LNG on mean flow and turbulence under low wind speed, slightly stable atmospheric conditions, Lawrence Livermore National Laboratory, report UCRL88420, Livermore, CA, Aug., 1983. Rodean HC, Hogan WJ, Urtiew HC, Goldwire HC Jr, McRae TG, Morgan EL (1984). Vapor burn analysis for the Coyote Series LNG spill experiments. UCRL-53530, Lawrence Livermore National Laboratory, Livermore, CA, April 1984. Sandia Laboratories (2009). Information Day Report, April 23, 2009; Sandia Lab News, March 13, 2009, pp. 1–5. Sarsten JA (1972). LNG stratification and rollover. Pipeline Gas J. 199, 37. Sauer HJ, Ragsdell KM (1971). Film boiling of nitrogen from flat surfaces. Adv. Cryog. Eng. 16, 412–415. Satyanarayna R, Borah M, Rao PG (1991). Prediction of thermal hazards from fireballs, J. Loss Prevention Process Industries, 4, 344. Schatzmann M, Marotzke K, Donat J (1991). Research on continuous and instantaneous heavy gas clouds. Contribution of sub-project EV 4T-0021-D to the final report of the joint CEC-project, University of Hamburg, February 1991. Schubach SA, Bagster DF (1996). Some aspects of modeling steam explosions. J. Loss Prev. Process Ind. 9, 193–197. Scranton DR, Lindberg WR (1983). An experimental study of entraining, stress-driven, stratified flow in an annulus. Physics Fluids 26(5), 1198–1205. SFPE (1992). SFPE Handbook of Fire Protection Engineering, 1st Ed., Bethesda, MD. SFPE (1995). SFPE Handbook of Fire Protection Engineering, 2nd Ed., Drysdal D (Ed), Society of Fire Protection Engineers, Bethesda, MD, ISBN 0-87765-354-2, pp 3/102, 3/197–240. See also 4th Edition, June 30, 2008. Shaikh I, Porter L (2008). Incorporating asphyxiation and cryogenic hazards into risk analysis. 11th Annual Symposium, Mary Kay O’Connor Process Safety Center, Texas A&M U., College Station, TX, October 28–29, 2008. Shaw P, Brisco F. Evaporation from spills of hazardous liquids on land and water. UKAEA Report SRD R100, May 1978.

364

REFERENCES

Shook B (2003). Environmental, safety, and security issues add to cost of offshore LNG. Oil Daily, Energy Intelligence Group, August 5. SIGTTO web site http://www.SIGTTO.org/ (accessed June 1, 2009). Simpson IC (1984). Atmospheric transmissivity—The effects of atmospheric attenuation on thermal radiation, SRD (Safety and Reliability Directorate). Report R304, U.K. Atomic Energy Authority, Culcheth, Warrington, U.K., Oct, 1984. Skramstad, Musaeus E, Melbeo E (2000). Use of risk analysis for emergency planning of LNG carriers. Gastech Conference, Am. Soc. Mech. Eng (ASME), NY, Houston, TX. Slade DH (ed.) (1968). Meteorology and atomic energy 1968. Air Resources Laboratory, U.S. Dept. of Commerce, Washington, DC. Small RD, Hiekes KE (1988). Early cloud formation by large area fires. J. Appl. Meteorol. 27(5), 654–663. Society of International Gas Tanker and Terminal Operators (SIGTTO) (1986). Transportation and storage of refrigerated liquefied fuel gases. London. Society of International Gas Tanker and Terminal Operators (SIGTTO) (2003). Safe havens for disabled gas carriers. London. Spaulding ML, Swanson JC, Jayko K, Whittier N (2007). An LNG release, transport, and fate model for marine spills. J. Hazard. Mater. 140, 488–503. Spencer H, Daycock J, Rew PJ (1998). A model for the ignition probability of flammable gases. WS Atkins Research Report 203 to the HSE (Health and Safety Executive), Her Majesty’s Stationers Office, London. Spencer H, Rew PJ (1997). Ignition probability of flammable gases. WS Atkins Research Report 46 to the HSE (Health and Safety Executive), Her Majesty’s Stationers Office, London. Spicer TO (1985). Mathematical modeling and experimental investigation of heavierthan-air gas dispersion in the atmosphere. PhD dissertation. University of Arkansas, p. 17. Spicer TO, Havens JA, Tebeau PA, Key LE (1986). DEGADIS—A heavier-than-air gas atmospheric dispersion model developed for the U.S. Coast Guard, Air Pollution Control Association Meeting, Minneapolis, MN, Paper 86-42.2, June 22–27. Spiegler P, Hopenfeld J, Silberberg M, Bumpus CF, Norman A (1963). Onset of stable film boiling and the foam limit. Int. J. Heat Mass Transf. 6, 987–989. Spouge J (2006). Leak frequencies from the hydrocarbon release database. IChemE Symposium Series 151, Institution of Chemical Engineers, Rugby, UK. Steward, FR (1970). Prediction of the height of turbulent diffusion buoyant flames. Combustion Sci. Technol., 2, 203–212. Stoll AM, Greene LC (1958). The production of burns by thermal radiation of medium intensity. ASME Paper No. 58-A-219, Am. Society of Mechanical Engineers, NY. Stoll AM, Greene LC (1959). Relationship between pain and tissue damage due to thermal radiation. J. Appl. Physiol. 14, 373–382. Strehlow RA, Baker WE (1976). The characterization and evaluation of accidental explosions. Prog. Energy Combust. Sci. 2, 27–60. Taylor DW (2007).The role of consequence modeling in LNG facility siting. J. Hazard. Mater. 142, 776–785.

REFERENCES

365

Thomas, PH (1963). The size of flames from natural fires. In 9th International Symposium of Combustion, Pittsburgh, PA, pp. 844–859. Academic Press, New York. Thomas, PH (1965). Fire spread in wooden cribs: Part III, the effect of wind. Fire Research Note Nr 600, Fire Research Station, Boreham Woods, U.K., June, 1965. Thyer AM (2003). A review of data on spreading and vaporization of cryogenic liquid spills. J. Hazard. Mater. A99, 31–40. Tieszen S (2007). Pool fire description. CLNG Workshop on LNG Pool Fire Modeling, Washington, DC, June 12–13, 2007. TNO (1977, 1992). Methods for the determination of possible damage to the people and objects resulting from release of hazardous materials (The Green Book), Chapter 16: Damage caused by heat radiation, CPR 16E. In Report to the Committee for the Prevention of Disasters Caused by Disastrous Substances, 1st and 2nd ed. (eds. X. van den Bosch et al.). TNO, Voorburg. TNO (1997a). Methods for determining and processing probabilities – The Red Book. Netherlands Committee for Prevention of Disasters, CPR 12E, 2nd ed. Sdu Uitgevers, the Hague. TNO (1997b, 2005). Heat flux from fires. In Methods for the calculation of physical effects (Yellow Book), 3rd ed. (eds. CJH van den Bosch, RAPM Weterings, WFJM Englehard), p. 141. TNO for the Committee for the Prevention of Disasters Caused by Dangerous Substances, Director General of Labor, the Hague. Trinity Consultants (2009). BREEZE suite of models, www.trinityconsultants.com, Nov. Tsao CK, Perry WW (1979). Modifications to the vulnerability model: A simulation system for assessing damage resulting from marine spills. ADA 075 231, U.S. Coast Guard NTIS Report No. CG-D-38-79, Environmental Control Inc, Rockville, MD. Turney R, Pitblado RM (1996). Risk assessment in the process industries, 2nd ed. Institution of Chemical Engineers, Rugby. U.S. Bureau of Mines (1946). Report on the investigation of the fire at the liquefaction, storage, and regasification plant of the East Ohio Gas Co., Cleveland, OH, October 20, 1944, February, 1946. U.S. Coast Guard (USCG). http://www.uscg.mil/uscg.shtm. (accessed June 10, 2009) U.S. Department of Energy (DOE) (1978). An approach to liquefied natural gas (LNG) safety and environmental control research, Appendix F. Division of Environmental Control Technology, Washington, DC. U.S. Department of Energy (DOE) (2008). http://www.netl.doe.gov/technologies/ oil-gas/publications/LNG/LNG_BasicFacts.pdf (accessed May 2008). U.S. Department of Transporation (DoT) (2006). Spectrum of fires in an LNG facility; assessments, models and consideration in risk evaluations. Report to Pipeline and Hazardous Materials Safety Administration, Washington, DC, Contract #: DTRS5604-T-005, Technology and Management Systems, Inc, Burlington, MA, December 5, 2006. U.S. Federal Energy Regulatory Commission (FERC). http://www.ferc.fed.us/(accessed June 2009). Valencia-Chavez JA, Reid RC (1979). The effect of composition on the boiling rates of liquefied natural gas for confined spills on water. Int. J. Heat Mass Transf. 22, 831–838.

366

REFERENCES

Valsgard S, Østvold TK, Rognebakke O, Byklum E, Sele HO (2006). Gas carrier development for an expanding market. Proceedings of the ISCOT/RINA Conference on Design, Construction and Operation of Natural Gas Carriers and Offshore Systems, pp. 13–28, Busan, Korea, Published by Royal Institution of Naval Architects, London. Sept. 14–25, 2006. van den Berg AC (1985). The multienergy method—A framework for vapor cloud explosion blast prediction. J. Hazard. Mater. 12, 1–10. van den Berg AC, Lannoy A (1993). Methods for vapor cloud explosion blast modeling. J. Hazard. Mater. 34, 151–171. Van der Linde P, Hintze NA (1978). Time Bomb: LNG: The truth about our newest and most dangerous energy source. Doubleday, p. 183. Vanem E, Antao P, Østvik I, de Comas FC (2008). Analyzing the risk of LNG carrier operations. Reliability Eng. Sys. Saf. 93, 1328–1344. Vaudolon A (2000). Liquefied gases—Marine transportation and storage. Witherby Seamanship Institute, London. Velde B, Linke G, Genillon P (1998). KAMELEON FIREX A simulator for gas dispersion and fires. International Gas Research Conference, pp. 429–237. Venart JES, Sumathipala K, Hadjisophocleous G, Sousa ACM (1989). Pressure relief of externally heated pressure liquefied gas vessels. Loss Prev. Saf. Promotion 6, 116.1. Venart JES, Rutledge GA, Sumathipala K, Sollows K (1993). To BLEVE or not to BLEVE: Anatomy of a boiling liquid expanding vapor explosion. Process Safety Prog 12, 67. Venart JES (2005). Letter to the editor, Process Safety Progress, 24(4), 226. Vesely W (1981). Fault Tree Handbook, NUREG-0492. Nuclear Regulatory Commission, Washington, DC. Vesely W, Dugan J, Fragola J, Minarick J, Railsback J (2002). Fault Tree Handbook with Aerospace Applications, Version 1.1. NASA, Houston, TX. Vienneau H (1964). Controlled flame heights for circular jets. BSc thesis. Department of Chemical Engineering, University of New Brunswick, Canada. Webber DM (1990). A model for pool spreading and vaporization and its implementation in the computer code GASP. UKAEA Report SRD R507. Webber DM, Brighton PWM (1986). Similarity solutions for the spreading of liquid pools. UKAEA Report SRD R371. Webber DM, Brighton PWM (1987). An integral model for spreading vaporizing pools, derived from shallow-layer equations. UKAEA Report SRD R390. Webber DM, Jones SJ, Tickle GA, Wren T (1992). A model of a dispersing dense gas cloud and the computer implementation D*R*I*F*T. I: Near instantaneous release and II: Steady continuous releases. SRD Report SRD/HSE R587, April 1992. Webber DM, Gant SE, Ivings MJ, Jagger SF (2009). LNG source term models for hazard analysis. Fire Protection Research Foundation, Quincy, Mass. Welker JR (1979). Radiant heating from LNG fire. AGA, LNG Safety Research Program, Report IS 3-1, p F1 to F25. Welker JR, Sliepcevich CM (1966). Bending of wind-blown flames from liquid pools. Fire Technol. 2(2), 127–135.

REFERENCES

367

Wesevich JW, Bennett R, Hu W (2004). Response of an offshore platform leg to terrorist attack. 14th Annual Offshore Symposium SNAME Texas Section, Nov. 1, 2004. Wikipedia (2006). LNG Safety History. http://www.wikipedia (accessed Jan 10, 2009); Cove Point, etc. USS Cole, 2009. Williamson BR, Mann LRB (1981). Thermal hazards from propane (LPG) fireballs. Combust. Sci. Technol. 25, 141. Wilson DJ (1996a). Do leaky buildings provide effective shelter? Proceedings of the 10th Annual Conference on Major Industrial Accidents, Council of Canada, Edmonton, Alberta, November 4–7, 1996. Wilson DJ (1996b). Exposure hazards from dense toxic vapors dispersing near buildings. Technical Report for Alberta Occupational Health and Safety and the Canadian Gas Processors Association, University of Alberta, Department of Mechanical Engineering, 95 pages. Wilson DJ, Zelt BW (1990). The influence of non-linear human response to toxic gases on the protection afforded by sheltering-in-place. OECD/UNEP Workshop on Emergency Preparedness and Response, Boston, May 7–10, 1990. Wilson DJ, Aynsley R, Vicery B, Peterson R (1993). Air flow around buildings, Chapter 14, ASHRAE Handbook of Fundamentals, American Society of Heating, Refrigerating, and Air Conditioning Engineers, Atlanta, GA, WS Comstock Publisher (ed. MS Owens), pp. 18, cited by Wilson (1996). Witlox HWM (1993). Technical description of the heavy-gas-dispersion program HEGADAS. External Report RKER.93.032, Thornton Research Center, Shell Research, Chester, England. Witlox HWM (2000). Unified dispersion model (UDM) Theory Manual. Det Norske Veritas, London. Witlox H, Harper M, Bowen P, Cleary V (2007). Flashing liquid jets and two-phase droplet dispersion II. Comparison and validation of droplet size and rainout formulations. J. Hazard. Mater. 142, 797–809. Witlox HWM, Oke A (2008). Verification and validation of consequence models for accidental releases of hazardous chemicals to the atmosphere. Hazards XX Conference, Institution of Chemical Engineers (NW Branch), Manchester, April 14–17, 2008. Woodward JL, Havens JA, Taft JR, McBride WC (1982). A comparison with experimental data of several models for dispersion of heavy vapor clouds. J. Hazard. Mater. 6(1–2), 161–180. Woodward JL (1998a). Estimating the flammable mass of a vapor cloud, concept-series book, 289 p. AIChE, Center for Chemical Process Safety, New York, pp 60–71, 99–102, 132–158. Woodward JL (1998b). Improving the effect of atmospheric stability class for dispersion modeling. Process Saf. Prog. 17(1), 1–8; also AIChE Loss Prevention Symposium, Paper 43-F, Houston, TX, March 9–13, 1997. Woodward JL (2000). A model of indoor releases with recirculating ventilation. Process Saf. Prog. 19, (3), 160-165. Woodward JL (2007). Coupling dynamic blow down and pool evaporation model for LNG. J. Hazard. Mater. 140, 478–487.

368

REFERENCES

Woodward JL (2008). Modeling underwater penetration of LNG carrier. AIChE Spring Meeting, New Orleans, LA, April 6–10, 2008 and Process Safety Progress. 27(4), 336–344. Woodward JL, Cook J (1993). Modeling of Dispersion with seamless transitions in entrainment mechanism, edge profile, and touchdown with the unified dispersion model. AIChE Meeting, Seattle, WA, August 14–19. World Bank (1988). Techniques for assessing industrial hazards: A manual. World Bank Technical Paper 55, ISBN 0253-7454, International Bank for Reconstruction and Development, Washington, DC. W rsig GM, Gaughan J, Scholz B, Sannes L, Kabelac S and Leder A (2009). Effect of enveloping pool fires on LNG tank containment systems. GasTech Conference, Abu Dhabi, May 2009; Germanischer Lloyd Presentation to SIGTTO, May 25, 2007. Yergin D, Stoppard M (2003). The next prize. Foreign Affairs, November/December, Zabetakis MG (1965). Flammability characteristics combustible of gases and vapors. Bureau of Mines Bulletin Report, 627, Pittsburgh, PA, 120 pp.

INDEX

accident history, 32–35, 39, 50, 161 aerosols, 134–135, 137, 166 AGA, 68, 231, 249, 251–252, 258 allision, 4, 71, 74, 77, 332 ALOHA model, 184 asphyxiation, 4, 9, 10, 64, 73, 313–317, 331 averaging time, 181, 197, 198, 224, 338–339 AVOCET tests, 190–191, 210 balsa wood, 21, 22, 24 barriers, 51, 71, 211, 294–295 BLEVE, 8, 35, 64, 76, 88, 215, 222, 277, 280–281, 291–294, 296–298, 329 Spain BLEVE, 35, 293 blow down, 90, 92–94, 107, 143, 153, 204, 269, 331 boiling flux, 204, 209 boiloff, 13, 38, 42–43, 44–45, 48, 104 boundary (plant), 120 boundary conditions, 185, 256, 298, 337 boundary layer, 170–171 Boussinesq approximation, 337 box models, 138 breaches, 44, 70, 74, 76, 80–84, 89–90, 92–94, 96–97, 135, 153, 208 BREEZE model, 184 brittleness (see also embrittlement), 33

buildings, 9, 15, 18, 34, 68, 70, 113, 115, 120, 181, 187, 212–218, 268, 276, 304, 311 buoyancy, 96, 174, 183, 202, 220–221, 223–224, 240–241, 297–299, 338 burn rate, 270 burn time, 269 burns/burning (see also freeze burns), 4–8, 11, 17–19, 21, 33–35, 64, 66, 73, 136, 145, 156, 172, 175, 190, 204, 222, 224, 228, 230–231, 233–241, 244–246, 249, 251, 255 duration, 115, 296, 318, 320, 322, 324 experiments, 319, 322 flux, 156, 204, 230, 234–238, 245, 248–249, 251–252, 319, 335 Burro tests, 189, 195, 210 bursting pressure vessels, 327 calm winds, 18, 337 CANARY model, 184, 209 carriers (LNG), 4, 23, 25–26, 28, 32, 37–39, 42, 45, 47–49, 60, 77–78, 81, 84–86, 104, 135, 188, 272, 275, 332 cracking, 80, 88 CFD models, 186, 187, 311 CFX model, 186 CIRRUS model, 184 claims, 16–19

LNG Risk Based Safety: Modeling and Consequence Analysis, by John L. Woodward and Robin M. Pitblado Copyright © 2010 by John Wiley & Sons, Inc.

369

370

INDEX

Cleveland, 22, 33, 36, 51 collisions, 4, 10, 17, 32, 38, 50, 56, 59–60, 71–72, 74, 76–77, 79–80, 84, 87, 280, 332 composition, 2, 3, 6, 11, 90, 102, 105, 130, 145, 160–161, 193, 283, 303–304, 322 Coyote tests, 190, 199, 210 cryogenic exposure, duration, 315 DEGADIS model, 116–117, 183–185, 204, 207–209, 333, 339 deflagration, 6, 8, 302–306, 333 design spills, 115–117, 124, 126 detonations, 303, 333 dimensions, 90–91, 284 discharge, flux, 8, 74–75, 85–86, 88–93, 103, 118, 129, 135–137, 142, 153–155, 158–160, 175, 187, 189, 196–198, 204, 209–210, 217–218, 236, 269, 276–279, 283–294, 288, 296, 316 dispersion characteristics, 193–194 exclusion zone, 113–114, 116–117, 119, 125, 127, 129 experimental data, 87, 188–192, 197, 207, 209–212 model comparison, 205–206, 208 modeling, 17, 19, 66, 128, 158, 175, 178–185, 294, 332–333, 337 double containment tanks, 54, 107, 109–111, 119, 121–122 double hulls, 17, 33, 35, 38, 43, 54, 57, 72, 74–77, 82, 86, 90–94, 96–97, 102. 105, 134–135, 154, 160, 278, 331–332 dosage, chemical, 199, 214, 217 dosage, radiation, 318, 324, 325 drop size, 135, 166–169 DRIFT model, 184–185 El Paso Kaiser, 32, 84, 86 embrittlement, 38, 43, 74, 76, 317, 331 emergency response, 13, 54, 72, 199, 296, 313 end points, 9, 17, 66, 199, 317, 338 engulfment, 72, 223, 267, 271, 331 entrainment, 17, 127–129, 131, 194, 196, 201, 204, 220–221, 225, 227–228, 240–241, 298, 300

environment, 4–5, 8, 45, 52, 68, 120, 182, 184, 330 escalation, 88, 160, 221, 277–278, 296 Europe (LNG supply, trade), 27–28, 29, 31 evaporation (pool), 12, 128 evaporation, composition effect, 151–153 evaporation flux, 137, 139, 144, 146, 154, 157, 172, 236, 241, 244, 332, 335 evaporation modeling, 83, 124, 128–129, 134, 136–139, 140–153, 156–158 evaporation rate, 117, 126–128, 139 evaporation rate, maximum, 170 evaporation tests, 146, 154 experiments burns, 319, 322 explosions, 312 general scale of, 8, 10–11, 85, 87–88, 245, 271, 331, 335–336, 339–340 spill, source, 17, 134–135, 142, 145, 149, 332 EXPLOJET model, 184 explosions (see also BLEVE; deflagration; detonation; rapid phase transition), 3, 8–9, 18–19, 66, 76, 87, 175, 187, 262, 287, 303, 308, 312, 318, 331 duration, 309 efficiency, 163, 166 energy, 332 experiments, 312 indoor, 18 Falcon tests, 211 fatalities, 7, 35, 60, 62, 64, 289, 322–323. See also injuries FEM3A model, 184, 202, 333 film boiling, 147–149, 151, 153, 170, 204, 272 fireballs, 35, 222, 265, 291–294, 298–299, 301–302, 304, 324 fire characteristics, 225, 227, 267 duration, 6, 115, 272–276, 279, 289, 291, 294, 296, 302, 318–320, 322, 324 experiments, 230–231, 235–236, 239, 245, 254, 262–263, 271, 288, 291, 300, 303, 338

INDEX

flash, 4, 6, 7, 35, 65, 73, 222, 287–291, 308 forest, 249 mass fire, 239, 248, 331 temperature, 271, 326 FLACS model, 186–187, 312 flame drag, 224, 252–253 flame height, length, 11, 34, 204, 231, 239–248, 262–263, 269–270, 282, 285, 288, 296, 335–336 flame tilt, 204, 223, 229, 232, 239, 245, 249–254, 263 flammability limits (LFL), 5–6, 11–12, 66–68, 116–117, 125, 129, 175, 189, 192, 194, 196, 198, 201–202, 209, 211–212, 214, 271, 283–289, 291, 338–339 flash (fire), 4, 6, 7, 35, 65, 73, 222, 287–291, 308 flash (liquids), 2, 10–12, 75, 83, 95, 102, 131, 135, 168, 186, 271, 294, 296–297 flow regimes, 140 fluctuations (of wind), 179, 197, 339 FLUENT model, 186–187 friction, 141–142 velocity, 179, 220 FRED model, 184 freeze burns, 5, 9, 18, 19, 73, 202, 275–276, 317 frequency, 9, 15, 51–55, 57–60, 62, 68, 70–71, 121, 132, 178, 197, 227, 229, 239, 313 GAO (Government Accountability Office), 192, 271, 329–330 gas, indoor intrusion, 9, 18, 79, 212, 217–218 Gaussian distribution, 166–168, 183, 194 Gaussian models, 181–183, 193–194, 196–197 Gaztransport, 39, 41 global LNG demand, 29 global LNG supply, 29, 30, 42 glub-glub effect, 332 grounding, 4, 10, 32, 38, 50, 60, 64, 74, 76–77, 81, 84–85, 94, 103, 271, 280

371

HARDER study, 76–78 hazards/hazardous, 1, 4–5, 7, 10–11, 14–15, 22, 51–52, 64, 68, 70, 204, 223, 275–276, 292, 303, 313, 317, 331–332 characteristics, 17, 66, 193, 293 gas group, 308 management of, 113, 212 models, 116–117, 134, 239, 267, 329 plume, 203 thermal, 115, 120, 259, 279, 289–291 zone, 296, 302, 316, 324 HAZOP, 54–56 heat balance, 136, 169 heat convection, 130, 134, 136, 145, 150, 170, 240, 243, 303 heat flux, 147, 150, 154, 196, 249, 273–274, 276, 279, 291, 319, 321–324 heavy gas properties, 193–197 HGSYSTEM model, 184, 208–209 hole size. See breach hydrostatic pressure, 10, 95, 135 ice, and LNG, 75, 98–100, 103, 145, 149–151, 210 ignition, 4–8, 11, 15, 34, 60, 64–68, 70, 76, 84–85, 87, 115, 120, 156, 190–192, 198, 202, 232, 271, 276–279, impingement, 44, 93, 187, 252, 272, 276, 279, 282, 325 impoundment areas, 72, 106, 115–117, 126–127, 132, 171, 202–204, 223, 277–278 incidents. See accident history indoor dispersion, 9, 18, 212–213, 217 experiments, 214 indoor explosion, 18 in-ground tanks, 108, 110–111 injuries, 10, 33, 35, 318. See also accident history; fatalities instantaneous spills, 137–138, 140–141, 143, 163, 165, 175, 183, 208, 211, 278–279, 281, 293, 296, 303, 338 integral models. See similarity models intrusion gas, indoors, 9, 18, 79, 212, 217–218 water, 94, 96–97, 103 invar, 24–25, 36, 41

372

INDEX

jet fire, 4, 7–8, 64, 240–241, 276–286, 293, 295, 304 jet plume, 34, 68, 92–93, 131, 134–135, 142, 153–154, 165, 168–169, 193–194, 200, 210, 284–285 jettison tests, 190, 209, 210 latent heat of vaporization, 233 leak, duration, 5, 55, 85, 136, 139, 158, 189, 191, 198, 200, 269, 339 Leidenfrost, 147, 149 length (plume), 193, 201 lethal effects. See fatalities lift off, 3, 5, 154, 15, 183, 202, 204, 212, 220, 249 (fire), 330, 332 Limburg, 81–82 log normal distribution, 166–168 long wave radiation, 173 Lorentzen spheres, 21–23 LNG prices, 31, 42, 107 properties, 1–6, 185 regulations (Asia), 26–29, 31 regulations (Europe), 13, 15–16, 36, 51–52, 66, 112, 119–120 regulations (US), 13–15, 34, 55, 66–67, 112–116, 125, 127–128, 212, 245, 268, 339 LNGMAP model, 204, 269 LSMS model, 184 magma, 161 Maplin sands tests, 87, 146, 154, 160, 190, 192, 207, 210, 232, 235, 289 marine accidents. See accident history mass fire, 239, 248, 331 mass flux, 88, 241, 244 Matagorda Bay tests, 67, 146, 190, 209 mechanism, 71–72, 134, 161, 164, 168, 171, 175, 223, 239, 243, 246, 248, 254, 282, 292, 327–330, 335, 340 membrane tankers, 25, 39–40, 48 Middle East, 2, 29–30, 51 mitigation, 15, 51–52, 70–73, 119–121, 133, 187, 202, 313, 326, 332 Model Evaluation Protocol (MEP), 116, 206, 333–334 modeling, uncertainties in, 236, 329

Montoir test, 225, 230, 232, 239, 257–258, 260, 270 Moss spheres, 26, 36, 39, 46–49, 74, 271–272, 245, 326 NFPA59A, 53–54, 66–67, 121, 262, 276–278, 322, 333 normal distribution. See Gaussian distribution nucleate boiling, 145–146, 272 numerical models, 45, 49, 136, 141, 185–187, 265, 285, 312, 330, 334 obstructions, 119, 176, 178, 181, 183, 185, 212, 333 offshore locations, 1, 16, 111, 132–133, 267, 282, 303, 311–313, 323 pain threshold, 319, 321–324 passive dispersion. See Gaussian models penetration, 4, 10, 17–18, 25, 28, 34, 94, 96, 101–102, 110, 118–119, 152–154, 160, 210, 236, 248, 280, 332 PHAST model, 13, 183, 185, 206–209, 270, 283, 317 pipework, 12, 55, 57–59, 74–75, 86–88, 312 plumes characteristics, 194 length, 193, 201 pools evaporation, 12, 128 fire characteristics, 223 spread, 136, 138–144 predictions, 17, 85, 93–94, 129, 143–144, 156–158, 188–194, 197, 202, 204–205, 207–209, 222, 244–245, 247–248, 251–253, 259–260, 268–269, 283–284, 301–302, 330–332, 334–336 probits, 321–323 public risk, 4, 7, 10, 70, 72, 132, 183, 222, 330–331, 333, 335 Q-flex, Q-max, 27, 41–46, 47 radiation, intensity, 283, 289, 318, 320–321, 324 rain out, 94, 136, 276, 285

INDEX

rapid phase transition (RPT), 4, 10, 34, 75–76, 83, 87, 133, 134, 159–166, 190, 197, 210–212, 331–333, 338 experiments, 160, 162, 165 regimes (flow), 140 regulations, LNG. See LNG, regulations response time, 57, 71, 165, 197, 213, 339 RHODIA model, 184 Richardson number, 196, 203, 220, 240–241 risk, 4, 14, 16, 17, 33, 60, 81, 84, 87, 132, 160, 266, 294, 297 analysis, 7, 10, 13, 15–17, 50, 52–59, 62, 66, 66, 70–71, 104–110, 120–121, 133, 168, 197, 239, 278, 317, 322 criteria (acceptability), 15, 68, 70, 119, 315, 332 management, 322 public: see public risk roll over, 4, 72 129–131 roughness (surface), 178–179 RPT. See rapid phase transition safety valves (pressure relief valves), 130 SAFESITE model, 184, 202, 207–209 scooping, 126, 202 sea water, 2, 33, 67, 75, 95, 107, 142, 159–160, 190, 211, 269, 332 ship design, 20–25, 37–38, 40–49, 74, 77, 85, 87–88 similarity (or integral) models, 183–186, 214, 282 siting, 13 SLAB model, 184, 207–209 sloshing, 25, 36, 38, 44–45, 73, 76, 86–87 SMEDIS, 188, 334 smoke, 6–8, 225, 227, 229, 239, 252–256, 267, 294, 331, 336–337 solar radiation, 119, 134, 136, 170–171, 176–178, 318, 331–332 Spain BLEVE, 35, 293 spills duration: see leak, duration instantaneous, 137–138, 140–141, 143, 163, 165, 175, 183, 208, 211, 278–279, 281, 293, 296, 303, 338 location, 8–9, 11 rate, 125, 136, 156, 160, 189–192, 236

373

size, 18, 126, 129, 135, 269, 335–336 test details, 209–211 type, 75, 133–134, 145 spillets, 204–205 spread rate, 138, 140–141, 156–157 stability (atmospheric), 68, 116, 175–180, 182, 189–193, 196–197, 201, 204, 208–209, 221, 262, 289–290, 330 Staten Island, 17, 19, 33, 36 statistical performance (validity), 57, 87, 205–206 storage tanks, 4, 9, 13–14, 18, 22, 54, 57, 60, 72, 104–105, 107–111, 117–118, 120, 123, 128–132, 202, 212, 223, 252, 262, 325, 332 structures, 324, 332 styrene, flammability, 271 super heat limit, 165 surface emissive power (SEP), 229, 237, 253–254, 259–260, 254, 269–270, 294, 331, 335 surface roughness, 178–179 surface tension effects, 45, 140, 168–169, 171 tank design, 21, 47, 52, 54, 114, 119, 121, 160. See also double containment tank; in-ground tanks; storage tanks terminals, 1, 6, 8–11, 13–19, 27–28, 32–39, 54, 60, 67–68, 73, 86, 104–129, 132–133, 202, 223, 262, 276, 279, 305, 317, 332, 336, 338 terrain, 116, 176, 178, 185, 187–188, 193, 210, 339 terrorism, 51, 79–71, 75, 81–83, 85, 271, 280 test data dispersion, 188–192, 197 fire properties, 231–232, 244–246 pool evaporation and spread, 144 thermal dose units (TDU), 322 thermal exclusion zone, 113–114, 122 thermal flux. See heat flux thermal properties, 125, 193 thermal weakening (steel and concrete), 4, 325, 327 tilt angle. See flame tilt

374

INDEX

TNT equivalence, 166 TRACE model, 184 trade (LNG), 2, 15, 23, 26–28, 31, 130 transmissivity, 223, 229, 232, 237–238, 251, 256, 259–262, 335 transitional boiling, 147, 150–151, 153 turbulence, 155, 337 TWODEE model, 182 uncertainties (in modeling), 236, 329 USS Cole, 81–82 vacuum, 9, 33, 44, 48, 72, 90, 105, 131, 135, 332 vapor cloud explosions (VCE), 3, 8, 9, 18, 64, 175, 275, 297, 302–305, 308, 311 vapor dispersion exclusion zone. See dispersion exclusion zone

vapor pressure, 130, 137, 193, 250, 261, 292 viscosity, 45, 140, 142, 157–158, 240 Von Karman constant, 178, 220 wake effects, 181, 224, 252, 331 water (see also ice) freezing, 98 intrusion, 94, 96–97, 103 sea water, 2, 33, 67, 75, 95, 107, 142, 159–160, 190, 211, 269, 332 waterways, 13, 14, 266 wave effects, 332 weapons attack, 80, 81 wind speed, 176–177, 180, 201, 204, 208 windows, 215 World Trade Center, 81, 293, 325 zone (3-zone fire model), 229, 255–256, 259–260