Thermo-Dynamics of Plates and Shells

Foundations of Engineering Mechanics Series Editors: V.I. Babitsky, J. Wittenburg Jan Awrejcewicz . Vadim A. Krysko . ...

Author: Jan Awrejcewicz | Vadim Anatolevich Krys'ko | Anton V. Krys'ko

36 downloads 592 Views 18MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Foundations of Engineering Mechanics Series Editors: V.I. Babitsky, J. Wittenburg

Jan Awrejcewicz . Vadim A. Krysko . Anton V. Krysko

Thermo-Dynamics of Plates and Shells

Series Editors: V.I. Babitsky Department of Mechanical Engineering Loughborough University Loughborough LE11 3TU, Leicestershire United Kingdom

J. Wittenburg Institut f u¨ r Technische Mechanik Universität Karlsruhe (TH) Kaiserstraße 12 76128 Karlsruhe Germany

Authors: Jan Awrejcewicz Department of Automatics and Biomechanics Faculty of Mechanical Engineering Technical University of Lodz 1/15 Stefanowskiego St., 90-924 Lodz Poland

Anton V. Krysko Department of Mathematics Saratov State University 410054 Saratov, Russia

Vadim A. Krysko Department of Mathematics Saratov State University 410054 Saratov, Russia

ISSN print edition: 1612-1384 ISBN-10: 3-540-34261-3 ISBN-13: 978-3-540-34261-8

Springer Berlin Heidelberg New York

Library of Congress Control Number: 2006926221 This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in other ways, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag. Violations are liable to prosecution under German Copyright Law. Springer is a part of Springer Science+Business Media springer.com © Springer-Verlag Berlin Heidelberg 2007 Printed in The Netherlands The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. A EX package Typesetting:Data conversion by the author\and SPi using Springer LT Cover-Design: deblik, Berlin Printed on acid-free paper SPIN: 11603313 89/3100/SPI - 5 4 3 2 1 0

Preface

The present monograph is devoted to nonlinear dynamics of thin plates and shells with termosensitive excitation. Since the investigated mathematical models are of different sizes (two- and three-dimensional differential equation) and different types (differential equations of hyperbolic and parabolic types with respect to spatial coordinates), there is no hope to solve them analytically. On the other hand, the proposed mathematical models and the proposed methods of their solutions allow to achieve more accurate approximation to the real processes exhibited by dynamics of shell (plate) - type structures with thermosensitive excitation. Furthermore, in this monograph an emphasis is put into a rigorous mathematical treatment of the obtained differential equations, since it helps efficiently in further developing of various suitable numerical algorithms to solve the stated problems. It is well known that designing and constructing high technology electronic devices, industrial facilities, flying objects, embedded into a temperature field is of particular importance. Engineers working in various industrial branches, and particularly in civil, electronic and electrotechnic engineering are focused on a study of stress-strain states of plates and shells with various (sometimes hybrid types) damping along their contour, with both mechanical and temperature excitations, with a simultaneous account of heat sources influence and various temperature conditions. Very often heat processes decide on stability and durability of the mentioned objects. Since numerous empirical measurement of heat processes are rather expensive, therefore the advanced precise and economical numerical approaches are highly required. A brief monograph description follows. Chapter 1 of this monograph is devoted to a study of three-dimensional problems of theory of plates in a temperature field. First, a brief historical outline as well as a state-of-art of the mentioned problems is described in introductional section. In Section 1.2, the system of differential equations governing a broad class of problems in the coupled dynamic theory of thermoelasticity in three-dimensional formulation is derived. A difference variational approximation is given and the difference scheme error is derived. Also stability of an explicit difference scheme is rigorously studied. Section 1.3 includes a comparison of solving systems governed by either hyperbolic or elliptic equations through various iterative methods.

VI

Preface

In section 1.4 numerous results of solutions of broad class of elasticity and thermoelasticity problems including coupling of temperature and deformations, are illustrated and discussed. In Chapter 2, after a brief historical research review, the variational equations for shallow anisotropic shells embedded into a temperature are derived. Coupling conditions and stress-strain state of shallow shells are formulated. In section 2.2 universality and efficiency of finite difference method devoted to boundary value problems for elliptic equations if outlined. Difference schemes for series of multidimensional stationary heat transfer equations are proposed in both sections 2.2 and 2.3. In the last section 2.4, influence of heat sources on a shell stress-strain and its stability is studied. Chapter 3 is devoted to analysis of dynamical behaviour and stability of closed cylindrical shells subject to continuous thermal load. A brief historical background is followed by variational formulation of the coupled dynamical problem of thermoelasticity. Hybrid-type variational equations of thin conical composite orthotropic thermosensitive shells are derived, and a problem of their solution is rigorously discussed. Furthermore, a solution to the biharmonic equation in relation to forcing function, as well as reliability of the obtained results, are addressed. Dynamical stability loss and non-uniform thermal loading are also studied. Dynamical behaviour and stability of rectangular shells is addressed in Chapter 4. In section 4.1, the computational algorithm to analyse differential equations with the associated boundary conditions is derived. The associated finite difference equations are given, and reliability of the results are verified. Stationary state method to analyse statical and dynamical problems is illustrated in section 4.1.4. Various vibrational phenomena and stability loss are studied. Stability of thin shallow shells with both transversal and heat loads are examined in section 4.2. Section 4.3 is devoted to stability of thin conical shells subject to both longitudinal load and heat flow. Finally, dynamical stability of flexurable conical shells with convection is studied in section 4.4. In Chapter 5 dynamics and stability of flexurable sectorial shells with thermal loads are addressed. First, theory of flexurable sectorial shells is introduced. The fundamental relations are assumed, differential equations are derived and initial conditions are given. After introduction of a thermal field the numerical “set-up” technique is illustrated and discussed, and numerical results reliability is outlined. Then various examples of stability of sectorial shells with finite deflections are studied. In addition, chaotic dynamics of sectorial shells and its control is addressed. Chapter 6 is devoted to a study of coupled problems of thin shallow shells in temperature field within the Kirchhoff-Love kinematic model. Fundamental assumptions and relations are introduced, and the differential equations are derived. The finite difference model of a solution to three dimensional heat conductivity equation is formulated. Numerical algorithm to solve the obtained equations is proposed, and then numerous examples of investigation of stability loss of shallow rectangular shells follow. Additional original method to solve a coupled thermoelastic problem is also proposed.

Preface

VII

In chapter 7 a novel optimal and exact method of solving large systems of linear algebraic equations. In the approach under consideration the solution of a system of algebraic linear equations is found as a point of intersection of hyperplanes, which needs a minimal amount of computer operating storage. The proposed method makes it possible to benefit from the essential advantages of both the direct method (universality, finitness of a computational process, exactness) and the iterational one (minimal amount of operational storage). Two examples are given. In the first example, the boundary value problem for a three-dimensional stationary heat transfer equation in a parallelepiped in R3 is considered, where boundary value problems of the 1st, 2nd or 3rd order, or their combinations are taken into account. The governing differential equations are reduced to algebraic ones with the help of the finite element and the boundary element methods for different meshes applied. The obtained results are compared with known analytical solutions. The second example concerns computation of a non-homogeneous shallow physically and geometrically non-linear shell subject to transversal uniformly distributed load. The partial differential equations are reduced to a system of non-linear algebraic equations with the error of O(h2x1 + h2x2 ). The linearization process is realized through either Newton method or differentiation with respect to a parameter. In consequence, the relations of the boundary condition variations along the shell side and the conditions for the solution matching are reported. In the last Chapter 8, some rigorous mathematical treatments of a coupled thermomechanical problems are addressed. First, the sufficient conditions of existence, uniqueness and continuity dependence on initial data of the Cauchy problem solutions for differential-operational equation of hybrid type (a part of the equation is of hyperbolic type, and another part is of parabolic type) are given. It is shown that if the operational coefficients are suitably chosen, the investigated equation can model a differential equations governing vibrations of a plate, i.e. the modified GermainLagrange equation of thermal conductivity (a parabolic equation). Second, a coupled thermo-mechanical of non-homogeneous shells with variable thickness and variable Young modulus (a so-called Timoshenko type model) is studied. The investigated problem is reduced to uniformly correct problem in the first form of a first order difference equation. Third, boundary conditions for a non-homogeneous first order operator – differential equation possessing a unique solution are derived. Two important theorems are formulated. Lodz, Saratov October 2003

J. Awrejcewicz V.A. Krysko A.V. Krysko

Table of Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

V

1

Three–Dimensional Problems of Theory of Plates in Temperature Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Coupled 3D Thermoelasticity Problem for a Cubicoid . . . . . . . . . . . . 11 1.2.1 Variational equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.2.2 Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.2.3 Difference approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.2.4 Difference approximation Error . . . . . . . . . . . . . . . . . . . . . . . . 26 1.2.5 Difference approximation Stability . . . . . . . . . . . . . . . . . . . . . . 29 1.3 Methods of Solving Difference Equations . . . . . . . . . . . . . . . . . . . . . . 39 1.3.1 Dimensionless Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 1.3.2 Systems of Elliptic Difference Equations . . . . . . . . . . . . . . . . 41 1.3.3 Systems of Parabolic and Hyperbolic Difference Equations . 47 1.3.4 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 1.3.5 Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 1.3.6 Numerical Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 1.4 Linear Problems in the Theory of Plates in 3D Space . . . . . . . . . . . . . 59 1.4.1 Static Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 1.4.2 Dynamic problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 1.4.3 Non-stationary temperature field . . . . . . . . . . . . . . . . . . . . . . . 82 1.4.4 Comparison of Solutions – non-isothermal Processes in Static Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 1.4.5 Inner heat sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 1.4.6 Deformation and Temperature . . . . . . . . . . . . . . . . . . . . . . . . . 114 1.5 3D Physically Non-Linear Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 1.5.1 Differential equations and difference approximation . . . . . . . 130 1.5.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 1.5.3 Estimation of Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 1.5.4 Temperature and Deformation Coupling . . . . . . . . . . . . . . . . . 136

2

Stability of Rectangular Shells within Temperature Field . . . . . . . . . . . 149 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 2.2 Flexible Anisotropic Shallow Shells in Temperature Fields . . . . . . . . 152

X

Table of Contents

2.2.1 Problem formulation and assumptions . . . . . . . . . . . . . . . . . . . 152 2.2.2 Fundamental relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 2.2.3 Variational and differential equations . . . . . . . . . . . . . . . . . . . . 159 2.2.4 Boundary and compatibility conditions . . . . . . . . . . . . . . . . . . 167 2.2.5 Compatibility conditions for shallow shells equations . . . . . . 177 2.2.6 Temperature field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 2.3 Solution of 3D Stationary Heat Transfer Equation . . . . . . . . . . . . . . . 186 2.3.1 The method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 2.3.2 Construction of difference schemes . . . . . . . . . . . . . . . . . . . . . 194 2.3.3 A priori convergence estimation . . . . . . . . . . . . . . . . . . . . . . . . 206 2.3.4 Algorithm of computation and compatibility conditions . . . . 209 2.3.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 2.4 Algorithm for Difference Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 2.4.1 Construction of difference equations . . . . . . . . . . . . . . . . . . . . 227 2.4.2 Stability problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 2.4.3 Reliability of obtained results . . . . . . . . . . . . . . . . . . . . . . . . . . 234 2.4.4 Transversal load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 2.4.5 Different boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 238 2.5 Computations of Plates and Shells in a Temperature Field . . . . . . . . . 252 2.5.1 Stress-strain state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 2.5.2 Stress-strain state and shells stability . . . . . . . . . . . . . . . . . . . . 264 3

Dynamical Behaviour and Stability of Closed Cylindrical Shells . . . . . 267 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 3.2 Thermoelastic Thin Thermosensitive Cylindrical Shells . . . . . . . . . . . 276 3.2.1 General Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276 3.2.2 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 3.2.3 Hybrid-Type Variational Equations . . . . . . . . . . . . . . . . . . . . . 283 3.2.4 Solution Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297 3.2.5 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 3.3 Computational Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310 3.3.1 Finite Difference Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310 3.3.2 Solution to Biharmonic Equation . . . . . . . . . . . . . . . . . . . . . . . 315 3.3.3 Reliability of the Obtained Results . . . . . . . . . . . . . . . . . . . . . . 320 3.3.4 Modified Relaxation Method . . . . . . . . . . . . . . . . . . . . . . . . . . 328 3.4 Dynamical Stability Loss with Ununiform Force Excitation . . . . . . . 334 3.4.1 Criteria of Dynamical Stability Loss (A Review) . . . . . . . . . . 334 3.4.2 Nonuniform Impulse External Pressure . . . . . . . . . . . . . . . . . . 342 3.5 Dynamical Stability Loss and Non-uniform Thermal Load . . . . . . . . 366 3.5.1 Thermal Field Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . 366 3.5.2 Influence of Time, Shell Geometry and Load . . . . . . . . . . . . . 373 3.5.3 Combined Static and Thermal Loads . . . . . . . . . . . . . . . . . . . . 382

Table of Contents

XI

4

Dynamical Behaviour and Stability of Rectangular Shells with Thermal Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 4.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402 4.2.1 Differential Equations, Boundary and Initial Thermoelastic Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402 4.2.2 Finite Difference Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 4.2.3 Reliability of the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412 4.2.4 Stationary State Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 4.3 Stability of Thin Shallow Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432 4.3.1 Influence of Heat Stream Intensity . . . . . . . . . . . . . . . . . . . . . . 432 4.3.2 Shells with Transversal Load and Heat Flow . . . . . . . . . . . . . 436 4.3.3 Influence of Thermal and Mechanical Characteristics . . . . . . 446 4.4 Stability of Thin Conical Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458 4.4.1 Boundary Conditions and Surrounding Medium . . . . . . . . . . 458 4.4.2 Constant Compressing Load and Heat Flow . . . . . . . . . . . . . . 464 4.4.3 Harmonic Longitudinal Load and Heat Flow . . . . . . . . . . . . . 466 4.5 Stability of Flexurable Conical Shells with Convection . . . . . . . . . . . 479 4.5.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479 4.5.2 Boundary and Thermal Fields Conditions . . . . . . . . . . . . . . . . 480 4.5.3 Critical Temperature Versus Heat Transfer Coefficient . . . . . 483

5

Dynamical Behaviour and Stability of Flexurable Sectorial Shells . . . . 493 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493 5.2 Flexurable Conical Sectorial Shells Computations . . . . . . . . . . . . . . . 498 5.2.1 Fundamental Relations, Differential Equations, Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499 5.2.2 Thermal Field and Set-Up Method . . . . . . . . . . . . . . . . . . . . . . 509 5.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515 5.3 Stability of Sectorial Shells with Finite Deflections . . . . . . . . . . . . . . 520 5.3.1 Influence of the Sector’s Angle . . . . . . . . . . . . . . . . . . . . . . . . . 522 5.3.2 Set-Up Method and Determination of Critical Loads . . . . . . . 560 5.3.3 Heat Impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595 5.3.4 Local Surface Load With Infinite Duration . . . . . . . . . . . . . . . 603 5.4 Chaotic Dynamics of Sectorial Shells . . . . . . . . . . . . . . . . . . . . . . . . . . 614 5.4.1 Statement of the problem and computational algorithm . . . . . 614 5.4.2 Static problems and reliability of results . . . . . . . . . . . . . . . . . 617 5.4.3 Convergence of a finite difference method along spatial coordinates for non-stationary problems . . . . . . . . . . . . . . . . . 618 5.4.4 Investigation of chaotic vibrations of spherical sector-type shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 625 5.4.5 Transitions from harmonic to chaotic vibrations . . . . . . . . . . . 627 5.4.6 Control of chaotic vibrations of flexible spherical sector-type shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630

XII

Table of Contents

6

Coupled Problems of Thin Shallow Shells in a Temperature Field . . . . 633 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633 6.2 Fundamental Assumptions and Relations . . . . . . . . . . . . . . . . . . . . . . . 634 6.3 Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 636 6.4 Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639 6.5 Solution to 3D heat conductivity equation . . . . . . . . . . . . . . . . . . . . . . 640 6.6 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642 6.7 Stability Rectangular Shells with Coupled Deformation and temperature fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645 6.8 Additional Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 666

7

Novel Solution Method for a System of Linear Algebraic Equations . . 671 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671 7.2 Elimination method for equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673 7.3 Numerical solution of a three-dimensional equation of elliptic type . 684 7.4 Computation of geometrically non-linear non-homogenous shallow shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696

8

Mathematical Approaches to Coupled Termomechanical Problems . . 705 8.1 Existence and Uniqueness of Solution of One Coupled Plate Thermomechanics Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705 8.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705 8.1.2 Basic assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707 8.1.3 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 708 8.2 On the Solution of a Coupled Thermo-mechanical Problem . . . . . . . 713 8.2.1 Introduction and Statement of the Problem . . . . . . . . . . . . . . . 713 8.2.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715 8.3 On the Solvable Operators Generated by Uniformly Correct Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 719 8.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 719 8.3.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 719

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767

1 Three–Dimensional Problems of Theory of Plates in Temperature Field

In section 1.1 historical outline putting emphasis on not solved problems in threedimensional formulation of plates thermoelastic theory is given. Section 1.2 presents a system of differential equations describing a broad class of problems of the coupled dynamic theory of thermoelasticity in a complete, threedimensional formulation including material’s non-homogeneity. The investigated system of equations has been supplemented with an equation at singular points of the examined space (a cubicoid), such as ribs, corners and simple points where various boundary conditions meet. A difference approximation of the initial differential system has been formulated with the use of the variational-difference method (the method of integral identity). The margin of the difference scheme error has been estimated. A theorem concerning stability of an explicit difference scheme has been proven and the condition of stability that guarantees weak convergence of the difference scheme’s solution towards the solution of a differential system has been obtained. Section 1.3 contains a comparison of solving systems of hyperbolic equations (using an explicit difference scheme based on applying Runge-Kutta’s method with automatic choice of an integration step and Runge-Kutta’s method with a constant step). Additionally, the section presents a comparison of applied iterative methods of solving systems of elliptic equations (Seidel’s method, the upper relaxation, the explicit and implicit methods of variable directions, and the explicit method of variable directions with the so-called Chebyshev’s acceleration). Several model problems have been used to draw the comparisons and the most economical methods have been applied as far as accuracy of solutions and computation time are concerned. Algorithms of the described methods have been formulated and a package of programs for solving problems of statics, quasistatics and elasticity and thermoelasticity dynamics has been created. An optimum choice of a spatial mesh step and an integration step within a time interval has been made and legitimacy of the theoretically obtained (in the first section) stability condition has been numerically confirmed. Feasibility of the obtained results has also been proven by means of comparison with real processes. Section 1.4 presents numerous results of solutions to a broad class of elasticity and thermoelasticity problems within the range of static, quasistatic and dynamic problems. There is also an analysis of the influence of the temperature and deformations’ coupling’s effect using some examples of thermal and mechanical impacts.

2


Finally, section 1.5 contains formulation of the equations of coupled dynamic three-dimensional problems with physical non-linearities. Moreover, the finite difference methods, Runge-Kutta’s method and the method of additional loads have been combined to form a numerical algorithm of solutions. Convergence of an approximate solution to the real one (the one searched for) has been analysed. The results of problems concerning thermal and mechanical impacts beyond the elasticity fields have been presented and the effects of the influence of reciprocal temperature and deformation fields’ coupling on the analysed processes have also been investigated in this chapter.

1.1 Introduction While designing and constructing electronic devices, industrial facilities, flying objects or technological instrumentation, the problems related to heat processes are particularly important. They appear due to the use of new materials, more complex loads affecting every single element of analysed objects, and also due to an increase of permissible heat loads in devices of smaller and smaller dimensions. As it is generally known, heat processes determine stability of functioning and durability of analysed objects. On the other hand though, numerous empirical measurements of heat processes are extremely complex and expensive. Therefore, exact computational analyses (numerical, as well as analytical) ought to be conducted in order to obtain constructions of optimum characteristics. In fact, non-stationary temperature reactions in surrounding environment require more accurate calculations than classic modelling of thermomechanical phenomena. In 1845, Duhamel [188] was the first to formulate the theory of elasticity regarding thermal stresses. However it was not until 1956, that Biot [107] introduced a dissipation function into a thermal conduction equation to account for the heat caused by the material’s deformation. Thus, the problem of thermoelasticity and the variational principle of coupled theory of thermoplasticity were first formulated. Since then there has been a great interest in that sort of problems. Earlier works on the theory of thermoelasticity [188] presented a dominating view that a change of temperature within a time interval is small, and therefore it was possible to apply a simplified (quasistatic) method, that is to neglect inertial terms in equations of motion, without the risk of major errors. The next step, introduced by means of the theory of thermoelasticity to simplify the problem, was neglecting dilatation terms in heat conduction equations. Sometimes, when both of the above mentioned terms are neglected in differential equations [598], the solution of a static problem is found. It turns out though, that due to the significance of the problems such simplifications ought not to be made. Among such problems are: the problem of investigating stress waves in deformable bodies; the problems related to determining thermoelastic vibrations; the problems related to investigating stability of conservative elastic systems [119, 164, 267, 316, 356, 466]. In their works, Danilovskoya [160, 161, 162, 163, 164], Kartashova and Shefter [316] analysed the influence of inertial terms on bodies’ behaviour considering the inertia forces. They

1.1 Introduction

3

also proved that neglecting a dilatation term does not ensure qualitatively satisfactory results due to inefficient examination of the coupling coefficient’s influence on the phenomenon. All the factors mentioned above caused a growth of interest in complete (i.e. not simplified) problems which fruited in numerous analytical works. Works of Karlsoy and Eger [315], Lykov [451], Kovalenko [355] and Nowacki [512] contain analyses and generalisation of two, so far independent disciplines, i.e. the theory of elasticity and the theory of heat conduction, and also a definition of so called coupled problem. A full formulation of the principles of variational theories of thermoelasticity is to be found in works [107, 265]. Betti’s theorem on reciprocity of virtual works is discussed in monograph [516], and a generalisation of Maizel’s method may be found in work [453]. Formulation of flat and space problems of coupled quasistatic theory of thermoelasticity is described in the works of Podstrigach, Schvetz, and Nowacki [512, 516, 545, 546, 547, 548]. Nowacki’s monograph [513] introduces equations of the coupled theory of thermoelasticity into wave equations and a method of solving linear and non-linear variants of the problems listed above. Many popular methods of solving the equations of Galerkin’s [215] or Papkovich’s [528] classic theories of elasticity are generalized in Podstrigach’s or Nowacki’s works and applied into the theory of coupled thermoelasticity. The method of solving problems of the coupled theory of thermoelasticity in case of a boundless space was proposed by Zorski [727], who used Green’s function to solve a heat conduction equation and considered dilatation to be a heat source. Chadwick’s work [145] takes up generalized problems of solving boundary problems of the coupled theory of thermoelasticity with the use of integral methods, whereas Souler and Brul use the small parameter method [632]. The problems related to accuracy of formulated boundary problems of the coupled theory of thermoelasticity were described first in book [119], which investigates an initial boundary problem for an isotropic body, later extended also onto an anisotropic body in Ionescu work [277]. Numerous dynamical problems of mathematical physics apply various integral transformations, including Laplace’s transformation [294], the solution of which is related to the use of Fourier’s series. In their work, Kupradze and others [398] propose their theory of multidimensional singular integral equations that makes it possible to investigate the static and dynamic problems of stabilised continuous systems’ vibrations. Hybrid problems, investigated by Magnaradze [452], Kupradze and Burchuadze [397] may be solved with generalized integrals that correspond to differential equations with the use of harmonic and analytical functions. Defermos’ work [175] contains many theorems concerning basic problems of the theory of thermoelasticity, including their proofs. Work [101] investigates the so-called second and third boundary and initial boundary problems of the coupled theory of thermoelasticity with the use of the method of potential and Laplace’s transformation. Work [397] analyses four basic three-dimensional boundary problems of the theory of thermoelasticity in case of harmonic vibrations of a homogeneous isotropic medium with the following conditions set in its boundaries:

4


1) displacement and distribution of temperature; 2) thermal stress and thermal flux; 3) displacement and thermal flux; 4) thermal stress and distribution of temperature. In addition, the authors formulate and prove many theorems concerning the existence and uniqueness of the above mentioned problems. The solutions to all of the four types of boundary conditions, presented in the form of generalized Fourier’s series, are to be found in Burchuladze’s work [135]. Fundamental results referring to the initial boundary problems of the theory of thermoelasticity have been obtained in the work of Kachnashviliev [294]. Nevertheless, fundamental solutions are still being perceived as classic. The conditions of smoothness appear to be too difficult to achieve for solutions of a wave equation describing impact processes. Due to the fact that such solutions do not have derivatives of the first order, they need to be examined from a generalized perspective. Integral relations contain information about solutions and emphasise physical phenomena because information on solution’s smoothness is partially lost in differential equations. The generalized mathematical theory on differential equations of the coupled theory of thermoelasticity described by means of both hyperbolic and parabolic equation has been formulated relatively recently. The works of Ladyzhenskaya [405] and Ilyisn [276] that were published in early fifties, contain numerous vital results referring to the theory of boundary problems for one hyperbolic or parabolic equation of a general type. In order to prove the existence and uniqueness of a generalized equation, it is necessary to make an entirely new a priori estimation that would take into account the right parts of equations in the form of the weakest norm and thus would accurately emphasise the physical aspect of the problem. Qualitatively most adequate examinations of general solutions seem to be the ones that apply the finite difference method. The method definitely stands out among many other approximate methods. Owing to continuing research of Samarskiy, Gulin, Nikolaev [591, 593, 594, 595], a large number of problems concerning stability of difference schemes for all types of one-dimensional equations in mathematical physics have been solved. This also started the research on difference schemes in the theory of elasticity. Let us list only a few examples of important results obtained with the use of the theory of difference schemes. Work [419] describes an a priori estimation of a solution in spaces W22,2 , W22,1 made by means of energy inequalities for dynamic problems of the theory of thermoelasticity using Dirichlet’s homogeneous boundary conditions. The authors have also constructed and examined a non-overt difference scheme and proved its convergence. In his work [483], Moskalkov presents a method of constructing difference schemes for the coupled theory of thermoelasticity boundary problems that is also useful for the equations of variable or discontinuous coefficients. Work [541] proposes a variational-difference formulation of the difference scheme of the coupled theory of thermoelasticity problems. Work [341] proves convergence of the difference solution towards the solution of a general hybrid problem for a hyperbolic equation with variable coefficients. It also shows how to improve the accuracy of presently applied difference schemes. In works [419, 694], the relation between the smoothness of a solution to the coupled theory of thermoelasticity one-dimensional dynamic problems and the smoothness

1.1 Introduction

5

of input data is examined. Smoothness is examined with the use of terminology applied for Hilbert and Sobolev’s spaces. Two difference scheme families have been constructed and their stability and convergence have been studied. Works [419, 693] extend the investigated problems by taking into account two-dimensionality or many so-called layer problems. It is worth noticing that at present, many finite differential problems modelling the flat problem of the dynamic theory of elasticity and the theory of thermoelasticity have already been solved. A large number of schemes described by displacements of high accuracy, stability and short computation time have also been presented [79, 96, 97, 345, 484, 591, 592, 664]. Among the less thoroughly examined problems are the ones that refer to the differential method of solving initial-boundary problems of the three-dimensional theory of elasticity and the theory of thermoelasticity. A review work by Suslova [643] contains a broad bibliography of works on research focused on solving boundary problems of the three-dimensional theory of elasticity. It also lists several works concerning the theory of thermoelasticity [142, 293, 643]. In works [198, 199] Ermolenko describes constructing the solution of a hybrid problem for a cubicoid by cutting the finite space out and he proves stability and convergence of the cubic difference process by applying the transformation of Lamé’s equations. He compares the result obtained in this way to the accurate one. In works [339, 340] Konovalov describes stability conditions for difference schemes for two-dimensional dynamic and static hybrid problems. The development of computational methods using computers and special algorithms has led to a sudden progress in the discussed field of science. A major contribution in the development of computational methods in the research on the dynamics of continuous media has been brought by the works of Godunov [224], Kukudzanov [393], Neuman [500], Rachmatulin [561, 562], Richtmyer [572], Wilkins [703] and Janenko [287]. Numerous examples of computations regarding the mechanics of a continuous medium are included in monographs [225, 287, 394, 573]. The problem of the coupled theory of thermoelasticity still remains a live issue due to its potential application and the numerical methods allow drawing a great deal of conclusions of a general nature. The examples of these may be the research and solutions of coupled thermoelasticity problems with the use of numerical methods for a number of particular issues: in work [546], Galerkin’s method is applied for solving a coupled problem in a finitely dimensional space with the use of a three-dimensional model; in work [616], the same method is applied to solve a two-dimensional problem; in work [430], a half-space finite difference method is applied for a one-dimensional problem, and in works [220, 721] – for a three-dimensional problem. In work [266], Huang and Shich compare solutions of free vibration problems regarding thermal processes in plates and spherical shells by applying dynamic and quasistatic theories. Non-stationary thermoelasticity problems for an infinite two-layered and initially heated plate consisting of various materials and thermally processed through interaction with fluids within Newton’s laws, have been examined in work [646]. Work [649] analyses stress-strain states of thick two-layered spheres with regard to axially symmetrical heat sources (the problem has been solved with

6


the use of the quasistatic theory). Work [648] investigates a system of coupled thermoelasticity differential equations with the use of a cylindrical coordinate system. Fourier’s method has been used to examine stress-strain states in a long circular cylinder with inserted rigid rings in work [504]. The finite difference method has been used to solve the problem of thermoelasticity for a rectangular orthotropic plate with regard to the dependence of its certain characteristics on temperature in work [641]. Work [663] investigates a non-stationary coupled thermoelasticity problem for an infinitely long, thick plate. The plate’s surfaces have been subjected to intensive heating and the coupling between the temperature field and the deformation has been analysed. The distribution of the temperature field in time has also been examined, as well as concentration of the stresses depending on the size of the stress field and the material’s thermodynamic properties. Dynamic loss of stability of thin plates has been analysed with the use of finite difference method in work [191], taking into account the effect of reciprocal coupling of the temperature field and the deformation field. Work [324] presents a solution to the coupled thermoelasticity problem for a thin rectangular shell affected by a three-dimensional temperature field. It also mathematically proves the convergence of the obtained approximate solution. All of the above mentioned works point out the differences which appear in solutions if the coupling of the deformation (strain) fields and the temperature fields are not taken into account. An increase of the coupling coefficient leads to an increase of interactions, which consequently leads to damping of the produced thermoelastic waves. Works by Karnauchov [312] and Pobedria [541] are focused on the problem of coupling in the theory of thermoelasticity. The influence of coupling on the stressstrain state of elastic and elastoplastic constructions has been investigated in work [359]. Several works of Day [169, 170, 171, 172, 173, 174] are also worth attention since the author investigates the conditions of legitimacy of applying approximations of unbounded theory of thermoelasticity and also the conditions of applying the properties of the solutions of heat conductivity equations to the solutions of a onedimensional dynamic coupled thermoelasticity problem’s equations. Research on thermal processes with regard to finite velocity of heat transfer is another direction in the development of the theory of thermoelasticity, since an entire class of physical processes (highly intensive thermal processes, laser rays) should be presented from the perspective of generalized Fourier’s law [451]. Works [323, 429, 495, 496, 558, 627] have been dedicated to the research on dynamic processes in solid bodies with regard to the heat transfer finite speed. In the works of Engelbrecht and Ivanov [285], an analysis of one- and two-dimensional models of wave processes have been made. In Kolyano and Shter’s work [337], a variational principle of reciprocal coupling of thermoelasticity for non-homogeneous media has been investigated using a cantilever beam as an example. Coupling of the deformation field and the temperature field significantly affects the solution’s character, especially in the problems of spreading impact fields in thermoelastic bodies. Therefore, the research on the dynamic coupling effects occurring in thermoelastic bodies subjected to simultaneous thermal, impulse, impact and mechanic treatment is one

1.1 Introduction

7

of the most important issues these days. Danilovskaya [163, 164] was the first to examine the dynamic effect in the “impact” problem along a half-space. The research was consequently carried on by Mura [489]. If the temperature on the surface of a body changes at a limited speed instead of sudden leaps, then the problem may be solved with a small parameter method [494]. In Pobrushin’s work [544], an analysis of some one-dimensional initial-boundary problems with thermal and mechanical impacts along the symmetry axis of an infinite rod has been made. The dynamic coupled thermoelasticity problem for a half-infinite plate at a simultaneous increase of temperature on its edge and with the use of Laplace’s integral transformation including the small parameter method has been solved in Sidlar’s work [617]. Dynamic behaviour of thin cylindrical shells subjected to impetuous thermal treatment has been investigated in work [632]. A coupled system of differential equations is derived with the use of Bubnov-Galerkin method and variational theorems, and also a simple-supported infinite cylindrical shell is investigated. Work [359] investigates dynamic thermoelastic processes during heat impacts in such construction elements as plates or spherical and cylindrical shells. The research has been conducted with the use of dynamic coupled thermoelastic equations and dynamic non-coupled equations of thermoelastoplasticity, and with the method of reduction to a series of non-coupled quasistatic problems, which in turn have been solved with KrylovBogolubov method. In Kuvyrkin’s work [402], a heat impact in the surface layer of a body limited by a curvilinear surface has been investigated. Shatalov’s work [608] shows that a decrease of equations’ couplings leads to a decrease of strain in the front of a thermoelastic wave. A method of expansion into power series in regard to a small parameter being the thermomechanical coupling has been applied in that case. Gayvas’ work [221] presents an analytical solution to a thermoelasticity problem for a plate with discontinuity caused by heat impact. The behaviours of plates subjected to steady mechanical load and rapid thermal transients on their both surfaces have been investigated in work [231]. Few of the solved problems that are related to impacts belong to the class of problems with aperiodic excitations. In this respect the theory of thermoelasticity seems to be a little underdeveloped and it faces some significant mathematical problems. Due to simultaneous mechanical and thermal impacts in constructions some small plastic deformations are ignored. The first work focused on investigation of elastoplastic stress states was published by Iliushin [272], and later by Rogoshinskov, who took non-uniformity of heating into account. Many works analyse also particular problems. Ionov’s works [278, 498] based on the theory of small elastoplastic deformations are among them. Work [148] describes a stress-strain state of an infinitely long cylindrical shell subjected to heating. In a series of works by Piskun [538, 539], cylindrical shells subjected to non-uniform heating and internal pressure have been examined. Work [307] contains some computations of thermoplastic deformations based on the variationaldifference method, and work [109] describes a stress-strain state of rotational shells in conditions of axially symmetrical heating. Monographs [609, 610] present a theory and computational methods concerning many problems of thermoplasticity at variable loads including also the history of loading (the objects of study included

8


cylinders, disks and low lift rotational shells). Work [126] applies Iliushin’s theory of plasticity to deal with heating an isotropic sphere with heat impacts of various shape and length (the problem was solved as a non-coupled one). Analytical description of thermoelastoplastic deformations is published in work [583]. In work [242], Birger’s method is applied to solve non-linear elasticity problems. Many interesting conclusions concerning dependence of physical and material parameters on temperature and work regime related to cooling shells and plates have been drawn in work [417]. Work [399] formulates a functional in order to find a variational solution to a plasticity theory problem at changing temperature for an elastoplastic material. Work [261] investigates the influence of the temperature load history, and work [150] analyses unique and continuous dependence on initial conditions in dynamic problems of non-linear thermoelasticity. A theory and a method of solving problems of thin-walled constructions heated by stationary and non-stationary heat sources are described in work [336], in which the dependence of physical and mechanical characteristics on temperature has been taken into account. A combination of the method applied for the theory of thermoelasticity with Vlasov’s variational method has been used to solve a three-dimensional problem of non-linear thermoelasticity in work [357]. It needs to be emphasised that coupling of the temperature and deformation fields (also in a quasistatic case) for problems of non-elastic material characteristics is taken into account only in selected works [180, 217, 259, 350, 584]. A recent Polish publication edited by Wo´zniak [708] contains a synthetic and abundant presentation of the level of modern knowledge of the theory of elastic plates and shells with specific reference to the contribution of Polish scientists in its development. In contrast to that approach this monograph puts more light to the contribution of scientists from the former eastern bloc into the development of the theory of plates in the temperature field. It is worth emphasising that names of the two first authors of this book are connected with a series of monographs on the theory of plates and shells published in Polish [37, 38, 39, 48, 50, 51, 53]. The latest theoretical achievements in non-classic analyses of the thermoelastic shell theory problems are described in monograph [39]. Numerous aspects of non-linear dynamics of shells and plates, including bifurcations, chaos and solitons, have been analysed in other works of the two first authors of this monograph [41, 45, 46, 47, 49, 52, 55, 56, 57, 58, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 389, 390], which also seem to be worth recommendation for readers who wish to broaden their knowledge in the field of shells and plates. At this point, several conclusions need to be drawn. (i) All of the above mentioned works investigate classic initially-boundary problems, while a typical (combined) boundary conditions are the most important in the theory of elasticity and thermoelasticity. There is a noticeable lack of solutions of that type in both linear and non-linear problems. (ii) There is no evidence for stability of difference schemes of the coupled theory of thermoelasticity in three-dimensional formulation for a cubicoid. (iii) Complexity of a physically non-linear system of differential equations limits the number of examples of solutions to thermoelastoplastic problems to only a few.

1.1 Introduction

9

The authors of this chapter focused their attention on solving the following problems: 1) construct a system of differential equations of the coupled dynamic theory of thermoelasticity taking into account a three-dimensional model and singularities of all kinds; 2) apply the variational-difference method for solving the coupled thermoelasticity theory problems; 3) prove stability of the difference approximation for the examined class of problems; 4) solve a typical problems of the theory of elasticity and the theory of thermoelasticity; 5) formulate a method and solve physically non-linear, initially-boundary problems for a three-dimensional plate in the dynamic coupled approach, and examine the influence of temperature and deformation fields’ coupling. The following notation is used: xi , i = 1, ..., 3 W(x) t Q(x, t) hα n U(u1 , u2 , u3 ) T = T0 + θ T0 θ ατ λq λ c ei j σi j e

- coordinate of a point in space; - examined field; - time; - {x ∈ Ω(x), τ ∈ (τ0 , τ1 )}; lα - step in a mesh: hα = ; Nα - normal unit vector directed outside the field: ni, j+m = cos(ni, j+m , xi ); - displacement vector; - absolute temperature; - absolute temperature in a stress-free state; - temperature increase; - linear coefficient of thermal expansion; - heat conduction coefficient; - heat emission coefficient; - thermal capacity; - strain tensor coefficient; - stress tensor coefficient; 3 - volumetric strains: e = eii ; i=1

λ, µ - Lamé’s coefficients: λ = E ρ ν P4 ∂Ωi 1 2 P(P , P , P3 ) f ( f1 , f2 , f3 )

- Young’s modulus; - material’s density; - Poisson’s ratio; - heat sources’ unit power; - plate’s wall; - volume (mass) force; - surface force;

E Eν ,µ= ; (1 + ν)(1 − 2ν) 2(1 + ν)

10


lα - plate’s dimension along xα axis; Nα - set of points of division towards xα axis; ω(ω1 × ω2 × ω3 ) - mesh surface: ω = {x(x1 , x2 , x3 ), xα ∈ ωα , α = 1, ..., 3}, ω ¯α = {xαiα , iα = 0, 1, ..., Nα−1 , Nα }, ωτ = ω1 × ω2 × ω3 × ω4 = ω × ω4 = {x(x1 , x2 , x3 , x4 ), xα ∈ ωα , α = 1, ..., 4}; 1 2 3 S (S , S , S ) - entropy vector; s - entropy flux; L2 (Ω) - Banach functional space of the following properties:

u2,Ω

⎛ ⎞ 12 ⎜⎜⎜ ⎟⎟⎟ 1 ⎜⎜⎜ ⎟ = ⎜⎜ |u|2 dx⎟⎟⎟⎟ ∼ u2,ω¯ = (u, u)ω¯2 , ⎝ ⎠

u x 2,Ω

Ω

⎛ ⎞ 12 ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟ = ⎜⎜ u2x dx⎟⎟⎟⎟ ; ⎝ ⎠ Ω

W21 (Ω) - space of elements L2 (Ω) with generalized derivatives of the first order due to Ω and of the following properties: (u, υ)(1) 2,Ω =

(uυ + u x υ x )dx ∼ (u, υ)(1) 2,ω ¯ =

N

υ(x)u(x)h,

i=0

Ω

1

1 (1) 2 (1) 2 (u, (u, = u) ∼ u) ; u(1) 2,ω ¯ 2,Ω 2,Ω W21,0 (Ω) - Hilbert’s space composed of elements u(x, τ) belonging to space L2 (Qτ ), which have generalized derivatives of the first order due to Qτ of the following properties: (u, υ)(1,0) 2,Qτ =

(uυ + u x υ x )dxdτ, Qτ

12 (u, u)(1,0) ; u(1,0) 2,Qτ = 2,Qτ 2 β = 3(λ + µ)ατ , 3 υx = υ x¯ x =

υi+1 − υi , h

⎧ ⎪ ⎪ ⎨ hα , xα ∈ ωα α = ⎪ , ⎪ ⎩ h α , xα ∈ 0, lα 2 υ x¯ =

υi+1 − 2υi + υi+1 , h2

υi − υi−1 , h υ xy =

The following markings are applied:

υ0x =

h4 =

τ1 − τ0 , M

υi+1 − υi−1 , 2h

υi+1, j+1 − υi, j+1 − υi+1, j + υi j . hi h j

1.2 Coupled 3D Thermoelasticity Problem for a Cubicoid

11

- free edge - simple support - clampededge - mechanicalimpact - thermalisolation - temperaturedistribution - thermalimpact

1.2 Coupled 3D Thermoelasticity Problem for a Cubicoid This chapter presents a variational method-based derivation of a system of coupled thermoelasticity differential equations for a three-dimensional plate, taking into account material’s non-homogeneity. The system includes equations within the plate’s field, at its edges, ribs in its corners and at simple contact points of numerous boundary conditions, which allows solving a substantial number of problems. A difference system is derived with the use of the variational-difference method by approximating the initial differential system with accuracy of such small values as O(h2 ). The obtained difference scheme’s stability theorem has been proven. 1.2.1 Variational equations We shall consider interaction between an elastic non-homogeneous body Ω and a medium that surrounds it in conditions in which thermal and mechanical processes are taken into account. Let us assume that at time instant τ = τ0 the body does not remain in the state of stress, i.e. the thermodynamic quantities that characterise the body such as absolute temperature T = T 0 , strain and stress tensor components and displacement vector components are equal to zero. Mechanical interaction makes displacement fields appear in the body. In every general case they accompany the change of the temperature field. Heating the body also causes perturbations in the investigated fields. Heat conductivity involves producing entropy, and strains cause a decrease of it, which in result leads to producing heat. Although thermoelastic damping is usually weak and for a short time interval it may be neglected (the noncoupled thermoelasticity theory), the relatively long-lasting processes require taking energy dissipation into account (the combined theory of thermoelasticity). Dissipation energy can be described by the following relation [63]: T 0 ∂S 2 1 dτ, (1.1) D= 2 λq ∂τ Ω

12


S =

cθ + βe, T0

(1.2)

where: S (S 1 , S 2 , S 3 ) is an entropy vector. Body Ω remains in motion, therefore according to Hamilton-Ostrogradski’s principle, integral τ1 (Π + K − A − B) dτ, (1.3) τ0

that describes the work utilised for the system’s movement within time interval (τ0 , τ1 ), assumes an extremal form in the movement’s trajectory: τ1 (Π + K − A − B) dτ = 0,

δ τ0

K=

ρ 2

3 Ω

2

ui dΩ,

(1.4)

(1.5)

i=1

where: K is the kinetic energy, Π = W + P denotes the elasticity potential, and in addition λ µei j ei j + e2 dΩ (1.6) W= 2 Ω

is the isothermal energy of strain. Moreover 1 c 2 1 4 θ dΩ + P θdΩ P0 = 2 T0 T0 Ω

(1.7)

Ω

denotes the thermal energy, B=

3 Ω

Pi ui dΩ

is the inertia forces’ work, whereas 3 i i f u d∂Ω + nθS n d∂Ω A= ∂Ω

i=1

(1.8)

i=1

(1.9)

∂Ω

denotes the external forces’ work. In spite of the fact that principle (1.3) does not take dissipation energy into account, it is essential to do it in energetic conditions of the coupled theory of thermoelasticity. That is why equation (1.4) takes the following form: τ1 (Π + K + D − A − B) dτ = 0.

δ τ0

(1.10)


13

Using Cauchy’s dependences ∂ui , ∂xi

eii =

ei j =

∂ui ∂u j + ∂x j ∂xi

(1.11)

and Duhamel-Neuman’s dependences σii = 2µei j + λe − βθ,

σi j = µei j ,

(1.12)

and expression (1.2), which is equivalent to the following dependence: θ=−

T0 div (S − βu) , c

(1.13)

we can transform expression (1.10). Thus we obtain a functional of the coupled thermoelasticity energy expressed by the displacement and entropy flux components. 1.2.2 Differential equations On basis of the energetic investigations discussed in 1.2.1 we shall construct a system of differential equations, the solution of which will be minimised by functional (1.10). A cubicoidal plate will serve as the object of investigation (Fig. 1.1). The edges and corners are characteristic for the surface of a cubicoid and they are sets of singular points. Also the points where the types of boundary conditions change and the points of application of concentrated central forces and heat sources

x2 l2

j W2 j W6 j W4

l1

j W5

jWm j W3

l3

j W1

x1

jWm

x3

Figure 1.1. The investigated cube-shaped plate.

14


can be called singular points. An analytical solution of the coupled thermoelasticity theory problems in the described field requires taking the field’s singularities into account. For instance, when applying the method of mesh it is necessary to thicken the mesh during the approach to the singular points. In this way the computation time needed to solve the problem will suddenly prolong. In order to avoid undesirable effects it is necessary to create additional equations in the singular points, from now on called the consistency conditions, that will constitute a part of the differential equations which function as Euler’s system for functional (1.10), used by the authors of work [429]. In addition, surface integrals will be included to describe the whole of the additional conditions imposed on the plate at its edges, in its corners and places where boundary conditions meet. k ∂u ∂ui ∂uk ∂u j Dkk+m λ + (λ + 2µ) +λ − βθ nk,k+m d∂Ωk+m + R= ∂xi ∂xk ∂x j ∂xi ∂Ωk+m

k ∂ui ∂uk ∂u j ∂u λ + (λ + 2µ) +λ − βθ nk,k+m d∂Ωk+m + ∂xi ∂xk ∂x j ∂x j

Dkk+m ∂Ωk+m

∂uk ∂u j ∂u j µ + nk,k+m d∂Ωk+m + ∂x j ∂xk ∂xi

∂uk ∂ui ∂ui µ nk,k+m d∂Ωk+m + + ∂xi ∂xk ∂xi

∂uk ∂u j ∂ui µ + nk,k+m d∂Ωk+m + ∂x j ∂xk ∂x j

∂uk ∂ui ∂ui µ + nk,k+m d∂Ωk+m + ∂xi ∂xk ∂x j

j Dk+m ∂Ωk+m

Dik+m ∂Ωk+m

Dik+m ∂Ωk+m

Dik+m ∂Ωk+m

li Dik+m

i l j ∂ui ∂uk ∂ui ∂u ∂u j ∂ui i µ nk,k+m dxi + µ + D j+m + n j, j+m d∂x j + ∂xk ∂xi ∂xi ∂x j ∂xi ∂x j

0

0

lk Dii+m

i ∂ui ∂uk ∂u j ∂u λ + (λ + 2µ) +λ − βθ ni,i+m dxk + ∂xi ∂xk ∂x j ∂xk

0

Ai+m θ + Bi+m T i+m −

∂Ωi+m

Ci+m

0 Ai+m T i+m

2 λq ∂S i 1 ∂S i ∂S i ni,i+m + + Ci+m + S + ∂x j ∂xk 2 λ ∂τ

i

λq ∂S i ∂S i λq ∂S i ∂S i ni,i+m + Ci+m ni,i+m d∂Ωi+m + λ ∂τ ∂x j λ ∂τ ∂xk


∂Ωi+mn

li

15

2 λq ∂S i ∂S i ∂S i 1 0 ni,i+m + Ai+mn θ + Bi+mn T i+mn − Ai+mn T i+mn + Ci+mn + Si + ∂x j ∂xk 2 λ ∂τ λq ∂S i ∂S i λq ∂S i ∂S i 1 Ci+mn ni,i+m + Ci+mn ni,i+m ∂Ωi+mn + 2 λ ∂τ ∂x j λ ∂τ ∂xk

0 0 Ai+mn θ + Ai+mn θ + Bi+m T i+m + Bi+mn T i+mn − Ai+mn T i+m − Ai+mn T i+mn

0

∂S i ∂x j

λq ∂S i ∂S i λq ∂S i ∂S i Ci+mn ni,i+m + Ci+mn ni,i+m dxi , λ ∂τ ∂x j λ ∂τ ∂xk

+

∂S i + ∂xk

(1.14)

0 where: i → j → k, i, j, k = 1, ..., 3, m = 0, ..., 3, T i+mn (x, t), T i+mn (x, t) – are set respectively at the limit of the function of the heat flux and the medium’ temperature; j , Ai+m , Bi+mΛ , Ci+mΠ – are constants that assume values 0 or 1 depending on the Di+m type of the boundary conditions; indeces Λ and Π define parts of the plate’s wall (left or right), where the function is set. Additionally: l j lk li m=3 νd∂Ωi+m = ν dxk dx j 0 m=0 , ∂Ωi+m

0

0

l j lk/2 νd∂Ωi+mΠ =

m=3

i

∂Ωi+mΠ

0

0

l j

lk

νd∂Ωi+m = ∂Ωi+mn

ν dx j dxk l

ν dx j dxk 0

m=0

,

.

(1.15)

0 lk/2

By setting independent variations to the displacement vector u and to the entropy flux s and making integration by parts we obtain: c 2 ∂θ i 1 i θ dΩ = − θδS d∂Ωi+m + δS dΩ− δP = − δ 2 T0 ∂xi Ω

Ω

∂Ωi+m

βθδui d∂Ωi+m +

β Ω

∂Ωi+m

∂θ i δu dΩ + ∂xi

Ω

P4 i δS dΩ, T0

(1.16)

τ1 i 2 τ1 2 i 1 ∂u ∂ui i ∂u i δu dΩτ − ρ δK = ρ δ dΩdτ = ρ δu dΩdτ, 0 2 ∂τ ∂τ ∂τ2 τ0

Ω

Ω

j j Di+m fi+m δui d∂Ωi+m , ∂Ωi+m

Ω

(1.17)

δA =

τ0

(1.18)

16


δB =

Pi δui dΩ,

(1.19)

Ω

τ1 T0 T 0 ∂S δS dΩdτ, δD = S δS dΩ − λq λq ∂τ Ω

δW = − Ω

τ0

τ0

(1.20)

Ω

∂ ∂u i ∂ui i ∂ ∂ ∂u j i (λ + 2µ) µ δu + δu + µ δu + ∂x j ∂x j ∂xi ∂xi ∂x j ∂xi i

∂ ∂uk i ∂ ∂u j i ∂ ∂uk i ∂ ∂ui i µ δu + µ δu + λ δu + λ δu dΩ+ ∂xk ∂xk ∂xi ∂xk ∂x j ∂xi ∂xk ∂xi i j k ∂u ∂u ∂u ∂u j ∂ui i (λ + 2µ) δu + µ δui + +λ +λ +µ ∂xi ∂xi ∂xk ∂xi ∂x j ∂Ωi+m

∂ui ∂uk k δu d∂Ωi+m , +µ µ ∂xi ∂xk

∂uk ∂ui ∂u j Dk+m k λ + (λ + 2µ) +λ − βθ nk,k+m δuk dxi − ∂xi ∂xk ∂x j

li δR =

(1.21)

0 ∗ Dk k+m

∂Ωi+m

∂uk ∂ ∂ui ∂u j λ + (λ + 2µ) +λ − βθ nk,k+m δui d∂Ωk+m + ∂xi ∂xi ∂xk ∂x j

li Dik+m 0

∗ Di k+m ∂Ωk+m

∂uk ∂ui µ ni,i+m δui dx− +µ ∂xi ∂xk

∂ui ∂uk ∂ µ nk,k+m δui d∂Ωk+m + ...+ +µ ∂x j ∂xk ∂xi

0 Ai+m θ + Bi+m T i+m − Ai+m T i+mn δS i d∂Ωi+mn +

∂Ωi+m,n

lk

Bi+mn T i+mn −

0 Ai+mn T i+mn

δS dxk − i

∂Ωi+m

0

lk

∂Ωi+m

∗∗

∂

0 Bi+mn T i+mn − Ai+mn T i+mn δS i d∂Ωi+mn + ∂x j

B j+mn T j+mn − A j+mn T 0j+mn δS i dxk −

0

∂

B j+mn T j+mn − A j+mn T 0j+mn δS i d∂Ω j+m + ∂xk


∂Ωi+mn

∂Ωi+mn

λq ∂S i Ci+m ni,i+m δS i d∂Ωi+m + λ ∂τ

lk Ci+mn

17

λq ∂S i ni,i+m δS i dxk − λ ∂τ

0

lk λ q ∂2 S i λq ∂S i i i ni,i+m δS d∂Ωi+m + ... + Ci+mn ni,i+m δS + Ci+mn 0 λ ∂xk ∂τ λ ∂τ l ∗∗ C i+mn

λq ∂ ∂S i ni,i+m dxi . λ ∂x j ∂τ

(1.22)

0

Considering the fact that some integrals are equal to zero, expression δR can be reduced. The integrals marked with one star equal zero because integrands are stress derivatives, which in turn occur to be constant in relation to the variable, for which a derivative is calculated. The integrals marked with two stars are also equal to zero because integrands are constant derivatives in relation to the variable, according to which differentiation is made. Substituting expressions (1.16)–(1.22) into (1.10) and assuming ui , S i , θ as independent variables (their variations are arbitrary), we obtain the following system of differential equations: 3 3 m ∂θ ∂2 u s ∂ sm ∂u kαβ −β + Ps = ρ 2 , ∂xα ∂xβ ∂x s ∂τ α,β=1 m=1 ⎞ ⎛ 3 3 1 4 ∂ ⎜⎜⎜⎜ θ ∂u s ⎟⎟⎟⎟ ∂2 θ ⎟⎟ , ⎜⎜ + + P = β ∂τ ⎝ α s=1 ∂x s ⎠ ∂x2s T 0 s=1 ⎡ 3 3 ⎤ ⎢⎢⎢ ⎥⎥⎥ ∂um s sm s s ⎢ nα,i+m kαβ − βθn s,i+m ⎥⎥⎥⎦ + Ei+m u s = fi+m , Di+m ⎢⎢⎣ ∂xβ

(1.23)

(1.24)

(1.25)

α,β=1 m=1

Ci+m

λq ∂θ 0 ni,i+m + Ai+m θ − T i+m + Bi+m T i+m = 0, λ ∂xi ⎡ 3 3 ⎤ m ⎢⎢⎢ ⎥⎥⎥ ∂u s sm Di+m ⎢⎢⎣⎢ nα,i+m kαβ − βθn s,i+m ⎥⎥⎥⎦ + ∂xβ

(1.26)

α,β=1 m=1

⎡ 3 3 ⎤ ⎢⎢⎢ ⎥⎥⎥ ∂um s sm s s ⎢ nα, j+m kαβ − βθn s, j+m ⎥⎥⎥⎦ + Ei+m u s + E sj+m u s = fi+m, D j+m ⎢⎢⎣ j+m , ∂xβ

(1.27)

α,β=1 m=1

Ci+m

λq ∂θ λq ∂θ 0 ni,i+m + C j+m n j, j+m + Ai+m θ − T i+m + A j+m θ − T 0j+m = λ ∂xi λ ∂x j Bi+m T i+m + B j+m T j+m , ⎡ 3 3 ⎤ ⎢⎢⎢ ⎥⎥⎥ ∂um s sm ⎢ nα,i+m kαβ − βθn s,i+m ⎥⎥⎥⎦ + Di+m ⎢⎢⎣ ∂xβ α,β=1 m=1

(1.28)

18


⎡ 3 3 ⎤ m ⎢⎢⎢ ⎥⎥⎥ ∂u sm D sj+m ⎢⎢⎢⎣ nα, j+m kαβ − βθn s, j+m ⎥⎥⎥⎦ + ∂xβ α,β=1 m=1

⎡ 3 3 ⎤ ⎢⎢⎢ ⎥⎥⎥ ∂um s sm ⎢ nα,k+m kαβ − βθn s,k+m ⎥⎥⎥⎦ + Dk+m ⎢⎢⎣ ∂xβ α,β=1 m=1

s us Ei+m

s s + E sj+m u s + Ek+m u s = fi+m, j+m,k+m ,

(1.29)

λq ∂θ λq ∂θ λq ∂θ 0 ni,i+m + C j+m n j, j+m + Ck+m nk,k+m + Ai+m θ − T i+m + λ ∂xi λ ∂x j λ ∂xk

0 A j+m θ − T 0j+m + Ak+m θ − T k+m + Bi+m T i+m + B j+m T j+m + Bk+m T k+m = 0, (1.30) ⎡ 3 3 ⎤ m ⎢⎢⎢ ⎥⎥⎥ ∂u s sm ⎢⎢⎣⎢ Di+mn nα,i+m kαβ − βθn s,i+m ⎥⎥⎦⎥ + ∂xβ α,β=1 m=1 ⎡ 3 3 ⎤ ⎢⎢⎢ ⎥⎥⎥ ∂um s sm ⎢ nα,i+m kαβ − βθn s,i+m ⎥⎥⎦⎥ + Di+mn ⎢⎣⎢ ∂xβ α,β=1 m=1 Ci+m

s s s s u s + Ei+mn u s + Ek+m u s = fi+mπ , Ei+mπ λq ∂θ λq ∂θ 0 Ci+mn ni,i+m + Ai+mn (θ − T i+mn + Ci+mn )+ λ ∂xi λ ∂x j

0 Ai+mn θ − T i+mn + Bi+mn T i+mn + Bi+mn T i+mn = 0 .

(1.31)

(1.32)

The initial conditions are as follows [431]: ∂u s = q2s (x) , u s |τ=τ0 = q1s (x) , ∂τ τ=τ0 θ|τ=τ0 = q (x) , s = 1, 3, i −→ j −→ k, i, j, k = 1 . . . 3, ←

←

(1.33)

sm where: kαβ = µδαs δβm + (λ + µ)δαβ δ sm , δαβ denotes Kronecker’s symbol, α = λq /cρ, S S S S fi+m , ..., fi+m, j+m = fi+m + f j+m , f j+mΛ , are set functions corresponding to the surs face forces, whereas q(x), q1 (x), q2s (x) are set functions at an initial instant of time (coordinates). Thus obtained system contains: a) three expressions (1.23) describing dynamic behaviour of a three-dimensional plate including temperature stresses within the field, b) a generalized heat conduction equation (1.24), c) three equations (1.25) and (1.26) on the cubicoid’s walls, d) three equations (1.27) and (1.28) on the parallelepiped’s edges, e) three equations (1.29) and (1.30) in the parallelepiped’s corners, f) three equations (1.31) and (1.32) in the contact points of numerous boundary s s , Ak+m , E sj+m , Ai+m , Bi+m , Ci+m are equal conditions. Assuming that coefficients Di+m to 0 or 1, we shall obtain boundary conditions well-known in the theory of elasticity and thermoelasticity [6, 198]:


19

1. Rigid fixing (the first boundary problem) j j a) Di+m = 0, Ei+m = 1, j = 1, ..., 3, or j j k i = Dii+m = Ei+m = 0, Ei+m = Ei+m = Dki+m = 1; b) Di+m

2. Jointedly supported edge (the third boundary problem) j j k i a) Dii+m = Ei+m = Ei+m = 1, Ei+m = Di+m = Dki+m = 0, or j j k i = Di+m = 1, Ei+m = Ei+m = Dki+m = 0; b) Dii+m = Ei+m

3. Free edge (the second boundary problem) j j Di+m = 1, Ei+m = 0, j = 1, ..., 3;

4. Temperature distribution (the first boundary problem) Ai+m = 1, Bi+m = Ci+m = 0; 5. Density of a heat flux’ normal component (the second boundary problem) Ai+m = 0, Ci+m = 1, Bi+m = 1; 6. Convectional heat transfer (the third boundary problem) Ci+m = Ai+m = Bk+m = 1 . Combined conditions on the plate’s surface will serve as an example. The boundary conditions related to the following walls of the cubicoid and corresponding equations (1.25), (1.26) will be considered: – wall ∂Ω1 is free and insulated from heat sources, thus (D11 = D31 = D21 = 1, E11 = E13 = E12 = 0, C1 = 1, A1 = B1 = 0): λ

∂u1 ∂u2 ∂u3 + (λ + 2µ) +λ − βθ = f11 (0, x2 , x3 , τ) , ∂x2 ∂x1 ∂x3 µ

∂u2 ∂u1 +µ = f12 (0, x2 , x3 , τ) , ∂x1 ∂x2

µ

∂u1 ∂u3 +µ = f13 (0, x2 , x3 , τ) , ∂x1 ∂x3 ∂θ = 0; ∂x1

– wall ∂Ω2 is fixed and affected by heat impact (D12 = D32 = D22 = 1, E21 = E23 = E22 = 0, C2 = 1, A2 = 0, B2 = 1): u1 = 0,

20


u2 = 0, u3 = 0, ∂θ = T 2 (x1 , 0, x3 , τ); ∂x2 – wall ∂Ω3 is jointedly supported and the temperature distribution is as follows (D33 = E32 = E33 = 1, D23 = D13 = E31 = 0, C3 = 0, A3 = 1, B3 = 0): u1 = 0, u2 = 0, λ

∂u1 ∂u2 ∂u3 +λ + (λ + 2µ) − βθ = f33 (x1 , x2 , 0, τ) , ∂x1 ∂x2 ∂x3 θ = T 30 (x1 , x2 , 0, τ) .

Consistency conditions (1.27) and (1.28) in the investigated case are as follows: – at the edge (0, 0, x3 ): (λ + 2µ)

∂u1 ∂u2 ∂u3 +λ +λ − βθ + u1 = f11 (0, 0, x3 , τ) , ∂x1 ∂x2 ∂x3 µ

∂u1 ∂u2 +µ + u2 = f12 (0, 0, x3 , τ) , ∂x1 ∂x2

µ

∂u1 ∂u3 +µ + u3 = f13 (0, 0, x3 , τ) , ∂x1 ∂x3 ∂θ ∂θ + = T 2 (0, 0, x3 , τ) ; ∂x1 ∂x2

– at the edge (x1 , 0, 0):

u1 = 0, u2 = 0,

λ

∂u1 ∂u3 ∂u2 + (λ + 2µ) +λ − βθ + u3 = f33 (x1 , 0, x3 , τ) , ∂x1 ∂x3 ∂x2

∂θ + θ = T 2 (x1 , 0, 0, τ) + T 30 (x1 , 0, 0, τ) ; ∂x2 – at the edge (0, x2 , 0): (λ + 2µ)

∂u1 ∂u2 ∂u3 ∂u3 ∂u1 +λ +λ − βθ + µ +µ = ∂x1 ∂x2 ∂x3 ∂x1 ∂x3 f13 (0, x2 , 0, τ) + f13 (0, x2 , 0, τ) , µ

∂u2 ∂u1 +µ + u2 = f12 (0, x2 , 0, τ) , ∂x1 ∂x2


λ

21

∂u1 ∂u3 ∂u2 ∂u3 ∂u1 + (λ + 2µ) +λ − βθ + µ +µ = ∂x1 ∂x3 ∂x2 ∂x1 ∂x3 f33 (0, x2 , 0, τ) + f13 (0, x2 , 0, τ) , λq ∂θ + θ = T 30 (0, x2 , 0, τ) . λ ∂x1

Compatibility conditions (1.29), (1.30) for corner (0,0,0) in the investigated case are as follows: (λ + 2µ)

∂u2 ∂u3 ∂u1 +λ +λ − βθ + 2u1 = f11 (0, 0, 0, τ) , ∂x1 ∂x2 ∂x3 µ

µ

∂u1 ∂u2 +µ + 2u2 = f12 (0, 0, 0, τ) , ∂x1 ∂x2

∂u3 ∂u1 ∂u1 ∂u2 ∂u3 +µ +λ +λ + (λ + 2µ) + u3 = ∂x1 ∂x3 ∂x1 ∂x2 ∂x3 f33 (0, 0, 0, τ) + f13 (0, 0, 0, τ) ,

λq ∂θ λq ∂θ + + θ = T 2 (0, 0, 0, τ) + T 30 (0, 0, 0, τ) . λ ∂x1 λ ∂x2 1.2.3 Difference approximation The problems described by the system of equations (1.23)–(1.33) are going to be solved with the method of mesh. The method makes it possible to bring a system of partial differential equations to a system of algebraic equations. There are a lot of approximations of the same problem. Among them there is one that provides a required approximation order and is stable (the so-called convergent approximation). If it is possible, a difference scheme should model the primary differential problem. Variational-difference methods are the most appropriate for analysing the problems discussed in this chapter and they retain the properties of a differential system. In order to build a difference scheme we are going to use the method of integral identity [429], which is based on an assumption that the energy functional is expressed in the form of (1.10). The solutions of problems (1.23)–(1.33) are generalized when there are such functions u, θ ∈ W21,0 (Qτ ) that for arbitrary functions vi (x, τ) ∈ W21,0 (Qτ ) fulfil the following integral identity: I (u, θ) = I +

τ1 3 τ0

3

i=1

∂Ωi+m

Ω

i=1

1 4 P vi + P vi dΩdτ− T0 i

0 vi + T i+m vi = 0, fi+m vi + T i+m

(1.34)

22


τ1 ⎡⎢ 3 3 ⎢⎢⎢ 3 ∂θ ∂u s ∂v s ∂un ∂v s sm ⎢⎢⎣ + kαβ I= −β vs − ρ ∂xβ ∂xα ∂x s ∂τ ∂τ s=1 α,β=1 n=1 τ0

Ω

⎤ 3 ∂θ ∂v4 ∂v4 ∂u s 1 ∂v4 ⎥⎥⎥⎥ ⎥⎥ dΩdτ. +β − θ ∂x s ∂x s ∂τ ∂x s α ∂τ ⎦ s=1

(1.35)

The field of constant change of arguments Qτ is going to be replaced with a finite set of points (nodes) in the form of ωτ = ω × ω4 {xα = ihα , x4 = jh4 , i = 0, ..., Nα , j = 0, ..., M}. For every continuous function f (x, τ) set in field Qτ , functions fi jk (x, x4 ) = f (xi , x j , xk , x4 ) are going to be constructed and defined within ωτ . Identity (1.34) consists not only of equations (1.23), (1.24), but also of conditions (1.25)–(1.33). We are going to find out about it when making a transformation of dependence (1.34). The integral identity is approximated with a summing identity replacing the integrals with quadratic and derivative forms - difference quotients. Integrals (1.34) will be replaced with quadratic forms describing trapezoids, and integrals (1.35) will be replaced with linear combinations of various relations of left and right rectangles. Strictly speaking: ⎧ 1 τ ⎪ ⎪ 1 ⎪ ⎨ [A]dΩdτ ∼ [A] [A] hi h j hk h4 + h h h + h I1 = ⎪ i j k 4 ⎪ 24 ⎪ ⎩+ ωi ×+ ω j ×+ ωk ×ω¯ 4 ω+ ×ω+ ×+ ωk ×ω ¯4 τ0

Ω

i

[A] hi h j hk h4 +

ω+i ×+ ω j ×+ ωk ×ω ¯4

+

[A] hi h j hk h4 +

ω+i ×ω+j ×ω+k ×ω ¯4

[A] hi h j hk h4 +

I2 =

ρ τ0

Ω

τ1 τ0

Ω

I4 =

β Ω

[A] hi h j hk h4 ,

ω+i ×+ ω j ×ω+k ×ω ¯4

∂u s ∂ν s ∂u s ∂v s dΩdτ ∼ h1 h2 h3 h4 , ρ ∂τ ∂τ ∂τ ∂τ ω ×ω ×ω ×ω 1

I3 =

∂θ s ∂θ ∂v4 ∂un ∂v s −β v + , i −→ j −→ k, ← ← ∂xβ ∂xα ∂x s ∂x s ∂x s

τ1

τ0

[A] hi h j hk h4 +

+ ω ×+ ω ×ω+ ×ω i j k ¯4

+ ω ×ω+ ×ω+ ×ω i j k ¯4

sn A = kαβ

τ1

[A] hi h j hk h4 +

+ ω ×ω+ ×+ ω ×ω i k ¯4 j

j

2

3

4

1 ∂v4 1 ∂v4 θ dΩdτ ∼ θ h1 h2 h3 h4 , α ∂τ α ∂τ ω ×ω ×ω ×ω 1

2

3

4

∂u s ∂v4 ∂u s ∂v4 dΩdτ ∼ h1 h2 h3 h4 . β ∂x s ∂τ ∂x s ∂τ ω ×ω ×ω ×ω 1

2

3

(1.36)

4

The derivatives in formulas (1.36) are approximated with the following relations: ∂um ∂um ∼ ymxα on net + ωα , ∼ ymx¯β on net ω+β , ∂xα ∂xβ


∂v s ∂v s ∼ ηmxα on net + ωα , ∼ ηmx¯β on net ω+α , ∂xα ∂xα ∂θ ∂θ ∂u s ∼ y sxˆ4 , ∼ y4xˆα on net ωα , ∼ y4x¯α on net ω+α . ∂τ ∂xα ∂xα For instance:

11 k11

+ ω ×+ ω ×ω+ ×ω 1 2 3 ¯4

∂u1 ∂v1 h1 h2 h3 h¯ 4 ∼ ∂x1 ∂x1 +ω

1×

+ ω ×ω+ ×ω 2 3 ¯4

23

(1.37)

µy1x1 η1x1 h1 h2 h3 h¯ 4 .

An approximate solution of problem (1.23)–(1.33) will be such two mesh func¯ τ , which for arbitrary net functions ηi , i = 1, ..., 4, set in tions yi , i = 1, ..., 4, set on ω the same net field ω ¯ τ , fulfil the following identity: 3 3 3

I1h

+

s=1 α,β=1 m=1 3

I2h

+

I4h

+

I3h

+

3

¯ 1 ×ω ¯ 2 ×ω ¯ 3 ×ω ¯4 s=1 ω

1 4 s P η + P η h1 h2 h3 h4 + T0 s s

0 η s + T s+m η s h¯ i h j = 0 i, j = 1, 3, i j. f s+m η s + T s+m

¯ i ×ω ¯j s=1 ω

Assuming that mesh functions ηi , i = 1, ..., 4 are equal units in an arbitrary chosen single point, we obtain difference equations in that node, in which ηi does not equal 0. Passing through all the nodes in net ω ¯ τ , we obtain a net equivalent that approximates differential system (1.23)–(1.33). Eventually we obtain the following system of equations:

1 ! j (λ + 2µ) yix¯i + µyix¯ j + µyix¯k + λy x j + λy x¯j j + µy xj i + xi xj xk x xj x ¯ i i 2 "

+ Pi − βy4xî = ρyix¯4 x4 , (1.38) + µy x¯j i + λykxk + λykx¯k + µykxi + µykx¯i x¯ j xi x¯i xk x¯k ⎛ ⎞ 3 3 ⎟⎟⎟ 1 4 ⎜⎜⎜⎜ cρ 4 4 α y x¯α xα + P = ⎜⎜⎝ y + β y xˆα ⎟⎟⎟⎠ , (1.39) T0 λq α=1 α=1 x4

hi ! j i + µy x¯j i xi + λy x¯j j + Di+m ni,i+m (λ + 2µ) yix¯i xi + λy xˆj j + λykxˆk + λy x j x¯i xi xj x¯i xi 2 " #

j i yi = gii+m , µy x¯i xi + µyix¯ j + µyix¯k + Pi − βy4x¯i xi − βy4 + Ei+m xj xj xk "

hi ! j j Di+m ni,i+m µy x¯j i xi + µyixˆ j + µy x¯k + (λ + 2µ) y x¯j j + λyix¯i xi + x xj k 2 xj #

j j + µykx j + P j − βy4xˆ j + Ei+m y j = gi+m , λykxk + µyix j xj x¯i xi xk "

hi ! k k Di+m ni,i+m µykx¯i xi + µyixˆk + µy x¯ j + (λ + 2µ) y x¯j k + λyix¯i xi + xj xk 2 xk

24


µyixk

x¯i xi

#

j + λy xj j + µy xj k + Pk − βy4xˆk + Ei+m yk = gki+m , xk

(1.40)

xj

λq hi 4 0 ni,i+m y4x¯i xi + Ai+m y4 − T i+m + Bi+m T i+m + y x¯ j x j + y4x¯k xk + P4 = 0, (1.41) λ 2 $ % $ % j i i k 4 Di+m ni,i+m Hi (λ + 2µ) y x¯i xi + λy x¯i xi + λy xˆk − βy + Dij+m n j, j+m H j µyix¯ j x j + µy x¯j i xi + Ci+m

j !

Dii+m + Di+m H µyix¯k + λy x¯j j x j + µy x¯j i xi + λykxk + µykx¯i xi + xk x¯ j x j x¯i xi xk x¯i xi 2 % i 4 i i i i i Di+m − βy x¯i xi + Ei+m Hi y + E j+m H j y = gi+m, j+m , $ % $ % j ni,i+m Hi µy x¯j i xi + µyix¯ j x j + D jj+m n j, j+m H j (λ + 2µ) y x¯j j x j + λyix¯i xi + λykx¯k xk − βy4 + Di+m j + Dij+m Di+m

2

!

H µy x¯j k + λy x¯i xi xk

x¯ j x j

+ λykxˆk

x¯ j x j

"

+ µyix¯ j x j − βy4x¯ j x j + P j + xˆk

j j j Hi y j + E j+m H j y j = gi+m, Ei+m j+m , $ % $ % Dki+m ni,i+m Hi µykx¯i xi + µyixˆk + Dkj+m n j, j+m H j µykx¯ j x j + µy xˆj k +

Dki+m + Dkj+m 2

!

H (λ + 2µ) ykx¯k + λyix¯i xi + µyixk xk

k Ei+m Hi yk

xk

+

E kj+m H j yk

=

x¯i xi

(1.42)

"

+ λy x¯j j x j − βy4xˆk + Pk + xk

gki+m, j+m ,

Ci+m Hi ni,i+m y4x¯i xi + C j+m H j n j, j+m y4x¯ j x j + Bi+m Hi T i+m + B j+m H j T j+m +

Ci+m + C j+m

0 H y4x¯k xk + P4 = 0, (1.43) + A j+m H j y4 − T 0j+m + Ai+m Hi y4 − T i+m 2 $ % j i i Di+m ni,i+m H¯ i (λ + 2µ) y x¯i xi + λy x¯ j x j + λykx¯k xk − βy4 + $ % $ % Dij+m n j, j+m H¯ j µy x¯j i xi + µyix¯ j x j + Dik+m nk,k+m H¯ k µykx¯i xi + µyix¯k xk + Dii+m + Dij+m + Dik+m 3

!

H¯ λykx¯k xk

x¯i xi

+ µykx¯i xi

x¯k xk

+ λy x¯j j x j

x¯i xi

+ µy x¯j i xi

" x¯ j x j

+

i i H¯ i yi + E ij+m H¯ j yi + Ek+m H¯ k yi = gki+m, j+m,k+m , Ei+m $ % $ % j ni,i+m H¯ i µy x¯j i xi + µyix¯ j x j + D jj+m n j, j+m H¯ j (λ + 2µ) y x¯j j x j + λyix¯i xi + λykx¯k xk − βy4 + Di+m $ % j Dk+m nk,k+m H¯ k µy x¯j k xk + µykx¯ j x j + j j + D jj+m + Dk+m Di+m

3

!

H¯ λyix¯i xi

x¯ j x j

+ µyix¯ j x j

x¯i xi

+ λykx¯k xk

x¯i xi

+ µykx¯ j x j

j j j j H¯ i y j + E j+m H¯ j y j + Ek+m H¯ k y j = gi+m, Ei+m j+m,k+m , $ % $ % Dki+m ni,i+m H¯ i µykx¯i xi + µyix¯k xk + Dkj+m n j, j+m H¯ j µykx¯ j x j + µykx¯k xk +

" x¯k xk

+

(1.44)


$

25

%

Dkk+m nk,k+m H¯ k (λ + 2µ) ykx¯k xk + λyix¯i xi + λy x¯j j x j − βy4x¯k xk + Dki+m + Dkj+m + Dkk+m 3

!

H¯ λyix¯i xi

x¯k xk

+ µyix¯k xk

x¯i xi

+ λy x¯j j x j

x¯k xk

+ µy x¯j k xk

" x¯ j x j

+

k k Hi yk + E kj+m H j yk + Ek+m Hk yk = gki+m, j+m,k+m , Ei+m

λq 4 λq λq y H¯ i + C j+m n j, j+m H¯ j y4x¯ j x j + Ck+m nk,k+m H¯ k y4x¯k xk + λ x¯i xi λ λ

0 + Bi+m H¯ i T i+m + B j+m H¯ j T j+m + Bk+m H¯ k T k+m + Ai+m H¯ i y4 − T i+m

0 (1.45) A j+m H¯ j y4 − T 0j+m + Ak+m H¯ k y4 − T k+m + P4 H¯ = 0,

Ci+m ni,i+m

Hi = H¯ i =

hj hi h j hi , Hj = , H= , hi + h j hi + h j hi + h j

h j hk hi h j hi hk , H¯ j = , H¯ k = , hi h j + h j hk + hk hi hi h j + h j hk + hk hi hi h j + h j hk + hk hi H¯ =

hi h j hk , m = 0, ..., 3, i −→ j −→ k . ← ← hi h j + h j hk + hk hi

The difference derivatives in equations (1.40)–(1.45) have been notated in a twolevelled form because of their reciprocal dependence on the location of the node in which the equation has been written. The upper notation corresponds to such case of the node’s location, in which the direction of the coordinate axis coincides with possibility of notating the derivative within the field in the same direction (the right derivative). The lower notation corresponds to such a location, in which both the direction of the axis and the direction of the derivative notation are opposite each other (the left derivative). The note that belongs to edge (l1 , 0, 0), (l1 , 0, l3 ) will serve as an example of our investigation. Equations (1.42), (1.43) have the following form: $ % $ % D14 H1 (λ + 2µ) y1x¯1 + λy2x¯2 + λy2xˆ3 − βy4 − D12 H2 µy1x2 + µy2x¯1 + "

D14 + D12 ! 1 H µy x¯3 + λy2x2 + µy2x¯1 + λy3xˆ3 + µy3x¯1 − βy4x¯1 + P1 + x3 x¯1 x2 x1 xˆ3 2 E41 H1 y1 + E21 H2 y1 = g142 , $ % $ % D24 H1 µy1x2 + µy2x¯1 − D22 H2 (λ + 2µ) y2x¯2 + λy1x¯1 + λy3x3 − βy4 + "

D24 + D22 ! 2 H µy x¯3 + λy1x¯1 + λy3xˆ3 + µy1x2 + µy3x2 − βy4x2 + P2 + x3 x2 x2 x¯1 xˆ3 2 E42 H1 y2 + E22 H2 y2 = g242 , $ % $ % D3 + D32 !

H (λ + 2µ) y3x¯3 + D34 H1 µy1xˆ3 + µy3x¯1 − D32 H2 µy3x2 + µy2xˆ3 + 4 x3 2 "

λy1x¯1 + µy1xˆ3 + λy2x2 + µy3xˆ3 − βy4xˆ3 + P3 + xˆ3

x¯1

xˆ3

x2

26


E43 H1 y3 + E23 H2 y3 = g342 ,

(1.46)

λq 4 λq y − C2 H2 y4x2 + B4 H1 T 4 + B2 H2 T 2 + λ x¯1 λ

C4 + C2 P4 4 0 4 0 4 H y x¯3 x3 + = 0. A4 H1 y − T 4 + A2 H2 y − T 2 + 2 T0 C4 H1

The coefficients that perform as net functions in equations (1.40)–(1.45), are selected in the following way: kh (x) = moreover:

k (x1 , x2 , x3 ) − k (x1 − h1 , x2 , x3 ) , 2

j j (x1 , x2 , x3 ) , = fi+m Ph (x) = P (x1 , x2 , x3 ) , gi+m

k k i i i + H j f j+m , gii+m j+mk+m = H¯ i fi+m + H¯ j f j+m + H¯ k fk+m . gki+m j+m = Hi fi+m

1.2.4 Difference approximation Error To start an investigation of the a priori characteristics in the form of a difference system it is necessary to begin with determining the order of the difference approximation error, since the difference scheme’s accuracy depends on it. Moreover, the function’s decomposition into Taylor’s series is used in this case. Let uih be a projection of solution ui , θ onto the mesh field ωτ , step h – a vector with norm |h| > 0, and let ui have a sufficient number of generalized derivatives. We shall investigate deviation Ψ = Lh uh − Ph − Lu − P, where Lh is a difference operator, and L is a differential operator. We say that Lh approximates L with order n on mesh ωτ , if |Ψ | < Mhn , where M − const > 0 does not depend on h. Let us make a separate estimation of the deviation of each of equations (1.36)–(1.38). We shall assume that index n, the smallest for all the equations, will be the system’s approximation order’s error. Let us analyse the error of equations (1.38) in detail:

Ψ 1 (ωτ ) = (λ + 2µ) uix¯i + µuix¯ j + µuix¯k + xi

xj

xk

1 ! j λu x j + λu x¯j j + µuixi + µuix¯i + λukxk + λukx¯k + xi xj xj xi x¯i x¯i 2 " i

∂ ∂u (λ + 2µ) + + Pi − βθ xî − ρuix¯4 x4 = µukxi + µukx¯i xk x¯k ∂xi ∂xi 2 h2i ∂2 ∂ ∂ui h j ∂2 ∂ui ∂ ∂ui ∂ui (λ + µ + µ + µ + + 2µ) 12 ∂xi2 ∂xi ∂x j ∂x j 12 ∂x2j ∂x j ∂xk ∂xk

h2k ∂2 ∂ui ∂ ∂u j h2i ∂2 ∂u j ∂ ∂ui µ + λ + λ + µ + 12 ∂xk2 ∂xk ∂xi ∂xi 12 ∂xi2 ∂xi ∂x j ∂xi


27

h2j ∂2 ∂ui ∂ ∂uk h2i ∂2 ∂uk ∂ ∂uk µ + λ + λ + µ + 2 2 12 ∂x j ∂xi ∂xi ∂xk 12 ∂xk ∂xk ∂xk ∂xi h2k ∂2 ∂uk ∂θ h2i ∂ ∂θ ∂2 ui i µ + P − β − β −ρ 2 − 2 12 ∂xk ∂xi ∂xi 2 ∂xi ∂xi ∂τ

h24 ∂2 ∂2 ui ρ + O h3 . 12 ∂τ2 ∂τ2

(1.47)

Taking (1.23) into account, we obtain Ψ 1 (ωτ ) = O(h2 ). It can be proven analogously that there is a second order approximation in equation (1.39): Ψ (ωτ ) = 2

3

θ x¯i xi + P − αθ xˆ4 − β 4

i=1

3 i=1

uixî x4

3 ⎡ 2 ⎢⎢⎢ ∂ θ h2i ∂2 ⎢⎣ 2 + = 12 ∂xi2 ∂xi i=1

⎛ 2 ⎞ ⎜⎜⎜ ∂ θ ⎟⎟⎟ ⎜⎝ 2 ⎟⎠ + ∂xi

⎤ h24 ∂2 ∂2 ui ⎥⎥⎥ ∂2 ui ∂θ h24 ∂ ∂θ ⎥⎦ + −β − β P −α − ∂τ 2 ∂τ ∂τ ∂xi ∂τ 12 ∂τ2 ∂xi ∂τ

O h2 + h24 . 4

(1.48)

Taking equations (1.24) into account, we obtain Ψ 2 (ωτ ) = O(h2 ). Equations (1.40) are of the same type for all nodes and they refer to any arbitrary wall of the cubicoid. Therefore, it seems useful to determine the error of the approximation of one of the wall’s equation. To illustrate that, for ∂Ω1 we obtain: Ψ 3 (0, x2 , x3 ) = (λ + 2µ) u1x¯1 + λu2xˆ2 + λu3xˆ3 + "

h1 ! 2 λu xˆ2 + µu2x1 + λu2xˆ2 + µu2x1 + µu1x¯2 + µu1x¯3 + P1 − βθ x1 −βθ−g11 = x1 xˆ2 x1 xˆ2 x2 x3 2 (λ + 2µ)

∂u1 ∂u1 h ∂ ∂u2 (λ + 2µ) + +λ + ∂x1 2 ∂xi ∂x1 ∂x2

∂u3 h ∂u3 ∂u3 h ∂ ∂u2 λ +λ + λ + 2 ∂x2 ∂x2 ∂x3 2 ∂x3 ∂x3

h ∂ ∂u2 ∂ ∂u2 ∂ ∂u1 ∂θ 1 − βθ − f11 + O h2 . λ + µ + µ +P −β 2 ∂x1 ∂x2 ∂x2 ∂x1 ∂x3 ∂x3 ∂x1

(1.49)

Taking equations (1.25), (1.23) into account, we obtain Ψ 3 (0, x2 , x3 ) = O(h2 ). The last three equations of the investigated node on wall ∂Ω1 also contain second order approximation errors. We are going to prove that the edge equations contain approximation errors of the O(h2 ) order. Let us consider a deviation expression of one in 48 similar equations (1.42), (1.43): Ψ 4 (0, 0, x3 ) =

% % h1 $ 1 h2 $ (λ + 2µ) u1x1 + +λu2x2 + λu3xˆ3 − βθ + µu x2 + µu2x1 + h1 + h2 h1 + h2

28


"

h1 h2 ! 1 µu x¯3 + λu2x2 + µu2x1 + λu3xˆ3 + µu3x1 + P1 − βθ x1 − g112 = x3 x1 x2 x1 xˆ3 h1 + h2 1 1

∂u h1 ∂ ∂u ∂u2 h2 (λ + 2µ) (λ + 2µ) + + O h21 + λ + h1 + h2 ∂x1 2 ∂xi ∂x1 ∂x2 ⎤

⎥⎥ ∂u3 h23 ∂ ∂u3 h2 ∂ ∂u2 2 λ + O h2 + λ + λ − βθ⎥⎦⎥ + 2 ∂x2 ∂x2 ∂x3 12 ∂x3 ∂x3 1 2

∂u h2 ∂ ∂u ∂u2 h1 ∂ ∂u2 h1 2 µ + µ + O h2 + µ + µ + O h21 + h1 + h2 ∂x2 2 ∂x2 ∂x2 ∂x1 2 ∂x1 ∂x1 & ' & ' ∂ ∂u1 h1 h2 ∂ ∂u2 ∂ ∂u2 µ + O (h3 ) + λ + O h2 + µ + O h2 + h1 + h2 ∂x3 ∂x3 ∂x1 ∂x2 ∂x2 ∂x1

& ' & ' ∂ ∂u3 ∂θ ∂ ∂u3 1 λ + O h1 + µ + O h3 + P − β − O h1 − ∂x1 ∂x3 ∂x3 ∂x1 ∂x1

h21 h22 h2 ∂ ∂u2 ∂ ∂u2 h1 f11 − f21 = O h2 − λ − λ , h1 + h2 h1 + h2 2 (h1 + h2 ) ∂x2 ∂x2 2 (h1 + h2 ) ∂x1 ∂x1 h

h

h ∂ ∂u2 λ ∼ − λu2x2 , λu2x2 ∼ λu2x2 − λu2x2 = λy2x2 , (1.50) − x x2 2 2 ∂x2 ∂x2 2 2 3y2j − 4y2j−1 + y2j−2

. 2h According to equation (1.23) and boundary conditions (1.25) we obtain Ψ 4 (0, 0, x3 ) = O(h2 ). Let us investigate the approximation error of 32 equations in the field’s corners. The following investigates a deviation for one equation in corner (0, 0, 0): y2x2 =

Ψ 5 (0, 0, 0) =

% $ h2 h3 (λ + 2µ) u1x1 + λu2x2 + λu3x3 − βθ + h1 h2 + h2 h3 + h3 h1

$ % $ % h1 h3 h1 h2 µu1x2 + µu2x1 + µu1x3 + µu3x1 + h1 h2 + h2 h3 + h3 h1 h1 h2 + h2 h3 + h3 h1 !

"

h1 h2 h3 µu2x1 + λu2x2 + µu3x1 + λu1x3 + P1 − βθ x1 − g1123 = x2 x1 x3 x1 h1 h2 + h2 h3 + h3 h1

∂u1 h1 ∂ ∂u1 ∂u2 h2 h3 (λ + 2µ) (λ + 2µ) + + O h21 + λ + h1 h2 + h2 h3 + h3 h1 ∂x1 2 ∂x1 ∂x1 ∂x2

h2 ∂ ∂u2 ∂u3 h3 ∂ ∂u3 λ + O h22 + λ + λ + O h23 − βθ + 2 ∂x2 ∂x2 ∂x3 2 ∂x3 ∂x3 1

∂u h1 h2 h2 ∂ ∂u1 ∂u2 h1 ∂ ∂u2 µ + O h22 + O h21 + + + + h1 h2 + h2 h3 + h3 h1 ∂x2 2 ∂x2 ∂x2 ∂x1 2 ∂x2 ∂x1 1 1 3

∂u h1 h3 h3 ∂ ∂u ∂u h1 ∂ ∂u3 2 µ + O h3 + O h21 + + + + h1 h2 + h2 h3 + h3 h1 ∂x3 2 ∂x3 ∂x3 ∂x1 2 ∂x1 ∂x1


29

∂ ∂u2 ∂ ∂u2 ∂θ h1 h2 h3 ∂ ∂u3 ∂ ∂u1 µ +λ −β + + + P1 = h1 h2 + h2 h3 + h3 h1 ∂x2 ∂x1 ∂x3 ∂x1 ∂x1 ∂x2 ∂x1 ∂x3 ∂x1

O h2 −

h2 h23 h1 h22 ∂ ∂u3 ∂ ∂u2 λ − µ − 2 (h1 h2 + h2 h3 + h3 h1 ) ∂x3 ∂x3 2 (h1 h2 + h2 h3 + h3 h1 ) ∂x2 ∂x2 h3 h21 ∂ ∂u3 µ . 2 (h1 h2 + h2 h3 + h3 h1 ) ∂x1 ∂x1

(1.51)

Taking equations (1.23), (1.25), (1.51) into account, we obtain Ψ 5 (0, 0, 0) = O(h2 ). Every equation of system (1.38)–(1.45) belongs to one of the five investigated forms of equations. This being so, the approximation error does not exceed O(h2 ), and the general error’s order in each node of field ωτ is not smaller than the other. 1.2.5 Difference approximation Stability It may turn out that the knowledge of a difference scheme’s approximation error’s order is insufficient to estimate the scheme’s quality. That is why, having determined the approximation order, it is necessary to analyse the scheme’s stability. The a priori estimation for yi is an essential part of the analysis of a difference scheme. If the system approximates the problem and it is stable, then its solution leads to the solution of a differential problem. The difference problem obtained with variational-difference methods are stable [384]. However, derivation of estimations imposes bounds upon the right parts and the coefficients of equations and also upon the steps of the mesh field hi , which is particularly important while making calculations. Our reasoning will be based on mesh space W21 (ω), W21,0 (ωτ ) and the following energy estimation:

2 (1)2 2 2 2 2 (1.52) y(1,0) 2,ωτ ≤ C 1 g2,1∂Ω×ω4 + P2,1ωτ + C 2 q1 2,ω + q2,ω + q2 2,ω . While deriving the proof we are going to use the most popular net relations, such as: – differential transformation (a uni-dimensional case): (ω, v) x,i = ω x,i vi + ωi+1 v x,i = ω x,i vi + ωi+1 v x¯,i+1 , (ω, v) x¯,i = ω x¯,i vi + ωi−1 v x¯,i = ω x¯,i vi + ωi−1 v x,i−1 ;

(1.53)

– summation [429] (p.225) (a one-dimensional case): n

vi ω x¯,i h = −

i=m+1 n i=m+1

v x,i ωi h = −

n−1

v x,i ωi h + vn ωm − vm ωm ,

i=m n i=m+1

vi ω x¯,i h + vn ωn − vm+1 ωn ;

(1.54)

30


– multi-dimensional summation with zero boundary values of mesh functions

hω x,i vi = −

ω ¯

hωv x¯,i .

(1.55)

ω ¯

The one-dimensional relations have been shown due to the fact that the initial transformations are derived only for one variable. Additionally we apply the following equalities: 2hω x,i ωi = ω2i+1 − ω2i − h2 ω x,i , 2hω x¯,i ωi = ω2i − ω2i−1 + h2 ω x¯,i , 2h

N

ω x¯,i ωi = ω2N − ω20 + h

i=1

2h

N−1

(1.56)

N & ' h ω x¯,i 2 ,

(1.57)

i=1

ω x,i ωi = ω2N − ω20 − h

i=0

N−1 & ' h ω x,i 2 , i=0

2h4 y x¯4 (k) y (k) = y (k) − y (k − 1) + h24 y x¯4 (k) ,

(1.58)

2h4 y xi (k) y x¯4 xi (k) = y2x (k) − y2x (k − 1) + (δy x (k − 1))2 ,

(1.59)

2

2h4 y xi (k) y x4 xi (k) =

2

y2x

(k + 1) −

y2x

(k) − (δy x (k)) ,

δv (k) = v (k + 1) − v (k) ,

2

(1.60)

and also well known Cauchy’s algebraic inequalities: ⎛ ⎞ 12 ⎛ n ⎞ 12 n n ⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟⎟⎟ ⎜ αi j ai bi ≤ ⎜⎜⎜⎝ αi j ai a j ⎟⎟⎟⎠ ⎜⎜⎜⎝ αi j bi b j ⎟⎟⎟⎠ , i, j=1 i, j=0 i, j=0 1

1

|(u, v)ω¯ | ≤ (u, u)ω¯2 (v, v)ω¯2 = u2,ω v2,ω ,

(1.61)

in which ε is a small number |ab| ≤ εa2 +

1 2 b ∀ε > 0 . 4ε

(1.62)

The ideas that refer to the proof of the difference processes’ stability have been derived from a number of scientific works. Andreev and Samarskiy’s works present a priori estimations of approximate solutions of the linear elasticity problems in the static approach taking into account mixed boundary conditions for a two-dimensional case. Ladyzhenskaya’s work contains derivations of the first initially-boundary problem for a parabolic and a hyperbolic equation in a general form. Works [231, 241, 492] address extended research into hybrid types of problems. Treating those references as basis we are going to prove a theorem that refers to stability of approximate solutions to the coupled thermoelasticity problems for three-dimensional plates.


31

To make things simpler let us assume that hi = h. Relations (1.38)–(1.44) are universal because if we assume proper form of indeces it is possible to formulate a problem in a very general way. As the multitude of symbols makes it rather difficult to comprehend the problem, we are going to investigate it within defined combined boundary conditions. The first and the second initially-boundary problems will serve as mechanical conditions. System (1.38)–(1.44) that describes thus formulated combination of conditions takes the following form:

1 ! 2 (λ + 2µ) y1x¯1 + λy1x¯2 + λy1x¯3 + λy x2 + µy2x1 + λy2x¯2 + x1 x2 x3 x x x¯1 1 2 2 "

µy2x¯1 + λy3x3 + µy3x1 + λy3x¯3 + µy3x¯1 − βy4xˆ1 + P1 = ρy1x¯4 x4 , (1.63) x¯2

x1

x3

x¯1

x¯3

1 ! 3 (λ + 2µ) y2x¯2 + λy2x¯1 + λy2x¯3 + λy x¯3 + µy2x¯2 + λy3x3 + x2 x1 x3 x ¯ x ¯ x2 2 3 2

" µy3x2 + λy1x¯1 + λy1x1 + µy1x2 + µy1x¯2 − βy4xˆ2 + P2 = ρy2x¯4 x4 ,

x3

(λ + 2µ) y3x¯3

µy1x3

x1

x¯2

x2

+ λy3x¯1

x1

+ λy3x¯2

+

x1

1 !

λy1x¯1

+ µy1x¯3 + λy1x1 +

x¯3 x¯1 x3 2

" + λy2x¯2 + λy2x2 + µy2x¯3 + µy2x¯3 − βy4xˆ3 + P3 = ρy3x¯4 x4 , (1.64) x¯3 x3 x¯2 x2

y4x¯1 x1 + y4x¯2 x2 + y4x¯3 x3 + P4 − β y1xˆ1 x1 + y2xˆ2 x4 + y3xˆ3 x4 = αy4x4

x3

x1

x2

inside plate ω:

h ! 3 µy x¯1 + µy3x¯2 + λy1x¯1 + µy1x3 + x1 x2 x3 x¯1 2 "

λy2x2 + µy2x3 + P3 − βy4x3 − βy4 + g33 = 0,

(λ + 2µ) y3x3 + λy2xˆ2 + λy1xˆ1 + x3

x2

h $

(λ + 2µ) y2x¯2 + µy2x¯1 + λy1x1 + λy3x3 + µy2x3 + µy3xˆ2 + x2 x1 x2 x2 2

%

µy1x2 + µy3x2 + P2 − βy4x2 + g23 = 0, x1 x3 !

h (λ + 2µ) y1x¯1 + µy1x¯2 + λy2x2 + λy3x3 + µy1x3 + µy3xˆ1 + x x x x1 1 2 1 2 "

µy2x1 + µy3x1 + P1 − βy4x1 + g1x3 = 0, x2

x3

(1.65)

λq 4 h

y x3 + y4x¯1 x1 + y4x¯2 x2 + P4 = g3 (1.66) λ 2 on the free wall ∂Ω3 affected by a thermal impact. The other walls are fixed and their temperature distribution is as follows: y s = gis , s = 1, ..., 3 ,

(1.67)

32


y4 = gi , i = 1, 2, 3, 4, 5, 6 .

(1.68)

The consistency conditions at the edges (0, x2 , 0), (l1 , x2 , 0), (x1 , 0, 0), (x1 , l2 , 0) are of the same kind and they take the following form:

h ! 3 1 (λ + 2µ) y3x3 + λy2xˆ2 + λy1x1 + µy x1 + µy3x¯2 + λy1x1 + µy1x3 + λy2x2 + x x x x x3 1 2 3 1 2 2 " # 1 1 1 − βy4x3 + P3 − βy4 + y3 + g33 + g31 = 0, (1.69) x2 2 2 2

h !

1 (λ + 2µ) y2x¯2 + µy2x1 + λy1x1 + λy3x3 + µy2x3 + µy3xˆ2 + x2 x1 x2 x2 2 2 "#

1 1 1 µy1x2 + µy3x2 + P2 − βy4x2 + y2 + g23 + g21 = 0, x1 x3 2 2 2 !

1 h (λ + 2µ) y1x1 µy1x3 + µy3x1 + + µy1x¯2 + λy2x2 + λy3x3 + x1 x2 x1 x1 2 2

%# 1 1 1 µy2x1 + µy3x1 + P1 − βy4x1 + y1 + g13 + g11 = 0, x2 x3 2 2 2 1 4 h

1 1 1 y x3 + y4x¯2 x2 + P4 + y4 + g3 + g1 = 0 (1.70) 2 2 2 2 2

µy2x3

at the edge (0, x2 , 0). The consistency conditions in corners (0, 0, 0), (0, l2 , 0), (l1 , 0, 0), (l1 , l2 , 0) are also of the same kind and they take the following form:

h ! 3 1 (λ + 2µ) y3x3 + λy2x2 + λy1x1 + µy x1 + µy3x2 + λy1x1 + µy1x3 + λy2x2 + x1 x2 x3 x1 x3 3 3 " # 2 1 1 1 − βy4x3 + P3 − βy4 + y3 + g33 + g31 + g32 = 0, x2 3 3 3 3 !

1 h (λ + 2µ) y2x2 + µy2x1 + λy1x1 + λy3x3 + µy1x2 + µy2x3 + µy3x2 + x2 x1 x2 x2 x1 3 2 "# 2

1 1 1 µy3x2 + P2 − βy4x2 + y2 + g23 + g21 + g23 = 0, x3 3 3 3 3 !

1 h

(λ + 2µ) y1x1 + µy1x2 + λy2x2 + λy3x3 + µy2x1 + µy1x3 + µy3x1 + x x x x x2 1 2 1 1 3 3 "#

1 1 1 2 µy3x1 + P1 − βy4x1 + y1 + g13 + g11 + g12 = 0, (1.71) x3 3 3 3 3 h 2 1 1 1 1 λq 4 y + P4 + y4 + g3 + g1 + g2 = 0 (1.72) 3 λ x3 3 3 3 3 3 in corner (0, 0, 0). The initial conditions are as follows:

µy2x3


yi x =τ = qi1 , yix4 x =τ = qi2 , y4 x =τ = q . 4

0

4

0

4

0

33

(1.73)

To make it easier, we shall denote the coefficients of the mechanical part of system (1.63)–(1.73) as ai, j,k , and the thermal part as bi, j,k . We shall assume that the net functions beyond the field’s boundaries are zero, i = 1, ..., N1 , j = 1, ..., N2 , k = 1, ..., N3 . THEOREM 1.1 Let the coefficients of system (1.23)–(1.33) fulfil the following inequalities: 3 3 3 3 2 ξ2jl ≤ ai jk (x, τ) ξil ξ2jn ≤ µ2 ξin , (1.74) µ1 j=1 l=1

∂ai jk < µ3 , ∂τ

i=1 n=1

µ4 ξ2 ≤ bi jk (x, τ) ξi ξ j ≤ µ5 ξ2 ,

(1.75) (1.76)

where ξi, j is an arbitrary number and: ∀s, i P s (x, τ) ∈ L2,1 (Qτ ) , fis (x, τ) ∈ L2,1 (∂Ω × (τ0 , τ1 )) ,

(1.77)

q1s (x) ∈ W21 (Ω) , q2s (x) ∈ L2 (Ω) , q (x) ∈ L2 (Ω) ,

(1.78)

and if the following inequalities are fulfilled: h4 µ2 √ h4 3 = 1 − ε, 1 − 12µ5 2 = ε, ε ∈ (0, 1), √ µ1 h h

(1.79)

then difference scheme (1.38)–(1.45) uniquely determines mesh functions y s and their interpolations at hi → 0, which are weakly convergent in L2,1 (Qτ ) towards the generalized solution u s (x, τ) ∈ Ω21,0 (Qτ ), s = 1, ..., 4 of problem (1.23)–(1.33). In order to find a solution to system y s , s = 1, ..., 4 we are going to derive the energy estimations. Therefore, we multiply both sides of equations (1.63) by i h3 h4 [yix4 (m) + y x4 (m)], while both sides of equations (1.64) – by 2h1 h3 y4 (m + 1), s where y (m) denotes a net function on mh4 upper layer. The results are summed along all of the nodes ω × ω4 . In the same way we transform relations into boundary conditions. Both sides of equations (1.65), (1.67) are multiplied by h2 h4 [yix4 (x1 , x2 , 0, m) + yix4 (x1 , x2 , 0, m)], and both sides of equations (1.66), (1.68) – by 2h2 h4 y4 (x1 , x2 , 0, m + 1) and they are summed in field ω × ω2 × ω4 . Equations (1.69) are multiplied by h2 h4 [yix4 (0, x2 , 0, m)+yix4 (0, x2 , 0, m)], while equation (1.70) is multiplied by 2h2 h4 y4 (0, x2 , 0, m + 1) and summation is made on account of ω2 and ω4 . Both sides of equations (1.71) are multiplied by h2 h4 [yix4 (0, m) + yix4 (0, m)], and equations (1.72) are multiplied by 2h2 h4 y4 (0, m + 1) and summed on account of ω4 . The obtained relations are grouped and as a result we obtain the following expression (1.80): !

h4 h3 − ai jk y1x1 (m) y1x4 (m) + y1x4 (m) − ... ω1 ×ω2 ×ω3 ×ω4

x1

34


1 !

× ai jk y3x¯1 y1x4 (m) + y1x¯4 (m) + ai jk y4x1 (m) y1x4 (m) + y1x¯4 (m) − x3 2 %

1 1 P y x4 (m) +y1x¯4 (m) + y1x¯4 x4 y1x4 (m) + y1x¯4 (m) + ... + $ 2h3 h4 −bi jk y1x¯1 x1 (m) y4 (m + 1) − ... − P4 y4 (m + 1) + bi jk y4x4 (m) y4 (m + 1) + % bi jk y1xˆ1 x4 y4 (m + 1) + ... + $

h4 h2 ai j0 y3x3 y3x4 (x1 , x2 , 0, m) + y3x¯4 (x1 , x2 , 0, m) + ...+ ω1 ×ω2 ×ω4

h !

y3x4 (x1 , x2 , 0, m) + y3x¯4 (x1 , x2 , 0, m) + ...+ ai j0 y3x¯1 x 1 2

ai j0 y4x3 (m) y3x¯4 (x1 , x2 , 0, m) + y3x4 (x1 , x2 , 0, m) +

P3 y3x¯4 (x1 , x2 , 0, m) + y3x4 (x1 , x2 , 0, m) −ai j0 y4 y3x4 (x1 , x2 , 0, m) + y3x¯4 (x1 , x2 , 0, m) +

g33 y3x4 (x1 , x2 , 0, m) + y3x¯4 (x1 , x2 , 0, m) + ... + 2bi j0 y4x3 y4 (x1 , x2 , 0, m + 1) + h

bi j0 2y4x¯4 x1 y4 (x1 , x2 , 0, m + 1) + ... + P4 2y4 (x1 , x2 , 0, m + 1) + ...+ 2 2g1 h4 h2 y4 (0, x2 , x3 , m + 1) + ω2 ×ω3 ×ω4

ω2 ×ω4

h4 h2

1$ a0 j0 y3x3 y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) + ...+ 2

h

y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) − a0 j0 y2x3 x2 2

a0 j0 y4x3 y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) + P3 y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) −

a0 j0 y4 y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) + y3 y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) +

g33 y3x4 (0, x2 , 0, m) + y3x¯4 (0, x2 , 0, m) + ... + 2b0 j0 y4 (0, x2 , 0, m + 1) y4x3 + h

4 4 4 4 2b0 j0 y x¯1 x1 y (0, x2 , 0, m + 1) + ... + 2y y (0, x2 , 0, m + 1) + ... + 2 1

h

h4 a0 y3x3 y3x4 (0, m) + y3x¯4 (0, m) + ... + a0 y2x3 y3x4 (0, m) + y3x¯4 (0, m) + ... − x2 3 3 ω4

a0 y4x3 y3x4 (0, m) + y3x¯4 (0, m) +P3 y3x4 (0, m) + y3x¯4 (0, m) −a0 y4 y3x4 (0, m) + y3x¯4 (0, m) +

2y3 y3x4 (0, m) + y3x¯4 (0, m) + g33 y3x4 (0, m) + y3x¯4 (0, m) + ... + 2b0 y4 y4 (0, m + 1) + h 4 (1.80) b0 y x1 x1 2y4 (0, m + 1) + ... + 2y4 y4 (0, m + 1) = 0. 3


35

It seems necessary to emphasise that the obtained relation (1.80) represents the energy of thermoelasticity of a three-dimensional plate in the form of a mesh iteration. Let us make the following transformations:

h3 ai jk (m) y1x¯1 x1 (m) y1 (m + 1) − y1 (m − 1) = I1 (M) = − ω×ω4

=

ω×ω ¯ 4

h3 ai jk (m) y1x1 (m + 1)y1x1 (m) −

2

h

ω1 ×ω2 ×ω4

ω1 ×ω2 ×ω4

ai j0 (m) y1x1

ω×ω ¯ 4

y1x1

h3 ai jk (m) y1x1 (m)y1x1 (m − 1) −

(m + 1) − y1x1 (m − 1) ±

h2 yi (m) y1x1 (m + 1) − y1 (m − 1) = h3 ai jk (M + 1)y1x1 (M + 1) y1x1 (M) − ω

ω

I2 (M) = ω×ω ¯ 4

ω×ω ¯ 4

δai jk (m) 1 y x1 (m + 1) y1x1 (m) − h 4 ω×ω ¯ 4

h2 ai j0 y1x1 (m) y1 (m + 1) − y1 (m − 1) ±

ai jk (1)y1x1 (1) y1x1 (0) h3 − h4

ω1 ×ω2 ×ω4

2

(m)

h2 yi (m) y1 (m + 1) − y1 (m − 1) ,

ω1 ×ω2 ×ω4

h3 ai jk (1) y3x1 (1) δy1x3 (0) +

ai jk (m) y3x1

(m) δy1x1

(m) −

I3 (M) = −

ω1 ×ω2 ×ω4

ω×ω ¯ 4

ω×ω ¯ 4

ω1 ×ω2 ×ω4

h ai j0 y3x1 y1 (m + 1) − y1 (m − 1) ± 2

h3 h4 ai jk y4 (m) y1x4 x1 (m) + y1x¯4 x1 (m) ±

h3 ai jk y4 (m) y1x1 (m + 1) + y1x1 (m − 1) ,

ω×ω4

ω ¯

h3 ai jk (M) y3x1 (1) Mδy1x3 (M) +

ai jk h2 h4 y4 (m) y3x1 (1) y1x4 (m) +y1x¯4 (m) =

ωi ×ω j ×ω4

I3 (M) = −

h3 ai jk (m) y3 y1 (m + 1) − y1 (m − 1) ,

ω1 ×ω2 ×ω4

−

h3

h3 ai jk y4 (m) y3x3 (m + 1) + y3x3 (m − 1) +

h2 ai jk h4 y4 (m) y4 (m) y3x4 (m) + y3x¯4 (m) ±

ω1 ×ω2 ×ω4

h2 ai jk h4 y4 (m) y3x4 (m) + y3x¯4 (m) ,

36


I4 (M) = −

ω×ω4

I5 (M) =

(1.53)

I6 (M) =

(1.54)

I6 (M) =

3

2h4 h

ω×ω ¯ 4

ω×ω4

bi jk y4x1

ω

h4 h3 P y1x4 (m) + y1x¯4 (m) ,

h3 y1x¯4 (M + 1)2 + y1x4 (1)2 ,

(m) y4x1 (m + 1)±

2bi jk h4 h2 y4 (m)y4x3 (m) y4 (m + 1) =

ωi ×ω j ×ω4

ω×ω ¯ 4

2bi jk h4 h2 y4 (m)y4x1 (m) y4 (m + 1) ,

ωi ×ω j ×ω4

2h4 h3 bi jk y4x3 (m) y4x3 (m + 1)−

h3 bi jk −y4x3 (m)2 + y4x3 (m + 1)2 − δy4x3 (m)2 h4 − ωi ×ω j ×ω4

2bi jk h4 h2 y4x3 (m) y4 (m + 1) ,

I7 (M) = − I8 (M) =

ω×ω4

2h3 h4 P4 (m) y4 (m + 1),

ω×ω4

2h4 h3 bi jk y4x4 (m) y4 (m + 1) =

(1.59)

h2 bi jk (y4 (m + 1)2 − y4 (m)2 − δy4 (m)2 )h4 ,

ω×ω4

I9 (M) =

ω×ω4

ω

2bi jk y4x1 x4 (m) y4 (m + 1) = − (1.60)

2h3 y1x1 (M + 1) y4 (M + 1) bi jk −

I10 (M) = −

ω×ω4

ω×ω ¯ 4

3

h4 h

ai jk y4x¯2 x2

ω×ω4

ω

2bi jk h3 δy4 (m)y4x1 (m) +

2h3 y1x1 (0) y4 (0) bi jk ,

(m) y1x¯4 (m) + y1x4 (m)

h3 ai jk δy1x2 (m) + δy1x2 (m − 1) y2x2 (m)±

ωi ×ω j ×ω4

=

(1.54) (1.60)

ai jk h4 h2 y2x2 (m) y1x4 (m) + y1x¯4 (m) .

The other components of expression (1.80), which have already occurred in the previously examined solutions, appear during the process of summation on account of all nodes or they are similar to the already mentioned ones. Taking the signs into account, we estimate expression (1.80), and without introducing any additional theories we shall use the estimations published in works [285, 429]: ( ( (2 (( ( I1 (M)α > µ1 ((y1x (M + 1)(( − µ2 ((y1x1 (M + 1)(( ((δy1x1 (M)(( ≥ ( (2 (( ( √ h4 ( µ1 ((y1x (M + 1)(( − 2 3 ((y1x1 (M + 1)(( ((y1x¯1 (M + 1)(( , h inequality [285] is taken into account at this point:


37

⎛ ⎞1 % ⎟⎟⎟ 2 h4 (( s (( ⎜⎜⎜ 3 $ s s (δy x (M)( = ⎜⎜⎜⎝h ≤ y x¯4 (i + 1, m + 1) − y x¯4 (i, m + 1) 2⎟⎟⎟⎠ h ω i

( √ h4 ( 2 3 ((y sx¯4 (M + 1)(( , h

( ( (( 1 ( ( ( ( (( (( 1 ( I1 (M)b ≤ µ2 (y x1 (1)( (y x1 (0)(( ≤ µ2 ((q x1 (( + h4 ((q x1 (( ((q x1 (( , ( ( ((y1 (m + 1)((( (((y1 (m)((( ≤ µ h ((y1 (m)(((2 , I1 (M)c ≤ µ3 h4 3 4 x x x ω4

(1.81)

ω4

( (( 1 (( ((P1 (m)((( (((y1 (m + 1)((( + (((y1 (m)((( ≤ 2P I4 (M) ≤ h4 x4 x¯4 2,1,ωτ max (y x¯4 (m)( , 1≤m≤M

ω4

( (2 ( (2 I5 (M) = ((y1x¯4 (M + 1)(( − ((y1x¯4 (1)(( , ( (2 ( (2 ( (2 I6 (M) = h4 bi jk ((y4x (m)(( + ((y4x (m + 1)(( − ((δy4x (m)(( , ω4

I6 (M)a ≥ µ4

( (2 ( (2 h4 ((y4x (m)(( + ((y4x (m + 1)(( ,

ω4

(2 µ5 (( 4 (δy (m)(( . 2 h In the investigated case, the following relation is applied: I6 (M)b ≤ 4 × 3

s 1 1 δy xi (x, m) = |δy s (i + 1, m) − δy s (i, m)| ≤ |δy s (i + 1, m) − δy s (i, m)| → h h (( s ((2 4 × 3 (δy x ( ≤ 2 δy s (m)2 , h I11 (M) =

3

α=1 ω4 ×ω j ×ωi

h4 h2 gαk yαx4 (x1 , x2 , 0, m) + yαx¯4 (x1 , x2 , 0, m) ,

I12 (M) = I7 (M) ≤ 2

ω4

ωi ×ω j

( (( ( ( ( h4 ((P4 (m)(( ((y4 (m + 1)(( ≤ ((P4 ((2,1,ω ,

I11 (M) ≤ 2 g (M)2,1,∂ωτ (1.61)

2h4 h2 gk y4 (x1 , x2 , 0, m) ,

τ

( ( max ((y x4 (m)(( ≤ ε g (M)22,1∂ω +

1≤m≤M

(1.62)

( (2 1 max ((y x¯4 (m)(( , 4ε 1≤m≤M ( (2 1 max ((y4 (m + 1)(( , I12 (M) ≤ ε g4 (M)22,1,∂ωτ + 4ε 1≤m≤M

38


⎛ ⎞ ⎜⎜⎜( ((2 (( 4 ((2 (( 4 ((⎟⎟⎟ ( 4 (δy (m)( − (y (0)(⎟⎟⎟⎠ . I8 (M) ≥ µ4 ⎜⎜⎝⎜(y (M + 1)( + ω4

Let us estimate the other components of formula (1.80): ⎛ ⎞ ⎟⎟⎟ ⎜⎜⎜ 3 3 1 3 1 3 3 1 h y x1 (m) δy x3 (m)⎟⎟⎟⎠ ≥ 0, |I2 (M)| ≥ µ1 ⎜⎜⎝⎜ h y x1 (1) δy x1 (0) + y x1 (M) δy x3 + 2 ω ¯ ω×ω ¯ 4 − |I3 (M)| ≤ −2µ2 (1.61)

( ((y4 (m)((( (((y3 (m + 1)((( , x3 ω4

⎞ ⎛ ⎜⎜⎜ ( (( (( 4 (( (( 1 (( (( 4 (( (( 1 ((⎟⎟⎟ ( 1 ⎜ −I9 (M) ≤ −2µ2 ⎜⎜⎝ (y x1 (m)( (δy (m)( + (y x1 (0)( (y (0)( − (y x1 (M + 1)(⎟⎟⎟⎠ , (1.61) ω4

⎛ ⎞ (( (( (( 4 (( (( 1 (( (( 4 ((⎟⎟⎟ (( 4 ⎜⎜⎜⎜ (( 1 (y (M + 1)( ≤ −2µ5 ⎜⎜⎝ (y x1 (m)( (δy (m)( + (y x1 (0)( (y (0)(⎟⎟⎟⎠ , ω4

1 2 3 δy x2 (m) y x2 (m) h ≥ 0. |I10 | ≥ 2µ1 ω×ω4 As a result, we obtain the following estimation (1.82): (( 1 ( ( (2 ( (2 ( (2 (2 (2 (y x¯4 (M+1)(( + ((y2x¯4 (M+1)(( + ((y3x¯4 (M+1)(( + µ1 ((y1x (M+1)(( + µ1 ((y2x (M+1)(( + ( (2 (( ( ( (( ( √ h4 ( µ1 ((y3x (M+1)(( − 2µ2 3 ((y1x (M+1)(( ((y1x¯4 (M+1)(( + ((y2x (M+1)(( ((y2x¯4 (M+1)(( + h ( (( (( 3 (( (2 ( ( ( (2 (( 3 (y (M+1)( (y (M+1)( + µ h ((y4 (m)(( + ((y4 (M+1)(( + µ ((y4 (M+1)(( + x

4

x¯4

4

x

x

ω4

( ( ((y1 (m)(((2 + (((y2 (m)(((2 + (((y3 (m)(((2 + ((δy4 (m)(((2 ≤µ (((q1 ((( + h (((q2 ((( q +µ h 2 4 1x 3 4 x x x x x ω4

ω4

( ( ( (( 1 (( ( ( ( ( ( ( ( (P (2,1ω max ((y1x¯4 (m)(( + ((P2 ((2,1,ω max ((y2x¯4 (m)(( + ((P3 ((2,1,ω max ((y3x¯4 (m)(( + τ

τ

1≤m≤M

τ

1≤m≤M

1≤m≤M

( (( 1 ((2 ( (2 ( (2 h ((δy4 (m)(((2 + (((P4 ((( (q2 (2,1ω + ((q22 ((2,1ω + ((q32 ((2,1ω + µ5 42 12 + 2,1ωτ τ τ τ h ω4 ( (( 4 ((2 ((δy4 (m)((( (((y3 (M+1)((( + (((y1 (M+1)((( + (((y2 (M+1)((( + (y (0)( + 2µ2 x3 x1 x2 ω4

2µ5

3 ( (( i (( ((δy4 (m)((( (((y1 (m)((( + (((y2 (m)((( + (((y3 (m)(((+ 2 (((qi ((( x1 x2 x3 i 2,1∂ω max (y x4 (m)( + ω4

τ

i=1

( ( 2 q4 2,1∂ωτ max ((y4 (m)(( . 1≤m≤M

1≤m≤M

(1.82)

We introduce similar terms and reinforce the inequality with relation (1.62) and impose condition (1.79) onto the steps of the net field ωτ . The result is as follows:

1.3 Methods of Solving Difference Equations

39

(( s (2 ( (2 ( (2 (y x¯4 (M + 1)(( + ((y sx (M + 1)(( + ((y4 (M + 1)(( + y s (M + 1)2 + (( 4 (2 ( (2 (y x4 (M + 1)(( + ((y4x (M + 1)(( ≤ ( (2 ( ( ( (2 C (τ, ε) P s 22,1ωτ + ((P4 ((2,1ω + g s 22,1∂ωτ + g4 22,1∂ωτ + g4 22,ω + ((g1s ((2,ω + ((g2s ((2,ω . τ (1.83) Inequality (1.83) occurs for every M and provides stability of the difference scheme (1.38)–(1.45). If the inequality is satisfied, the solutions (1.38)–(1.45) become stable and convergent towards the solution of problem (1.23)–(1.33). Further considerations are no different from the classic ones, described in work [285], that is why we shall not quote them here. First, a uniform bound of norms in W21,0 (Qτ ) is derived from estimation (1.83), then weak compactness of continuous couplings us (1.23)–(1.24) in W21,0 (Qτ ) is derived to prove consequently that only if the subsequence weakly converges in W21,0 (Qτ ) upon a function, then the function is the generalized solution (1.23)–(1.33) that belongs to W22,1 (Qτ ).

1.3 Methods of Solving Difference Equations The theoretical foundations presented in the first section may appear useful in formulation of methods, algorithms or writing computation programs that would make it possible to automatically solve problems in the field of statics, quasistatics, dynamics and coupled thermoplasticity taking into account the broad class of typically classic and non-classic boundary conditions at a plate’s edges. The use of the variational-difference method for solving the system of equations (1.23)–(1.33) allows obtaining the system of difference equations (1.38)–(1.45). The system is characterised by the following specific features: a) it has a high order equal to the order of the mesh nodes multiplied by four; b) the matrix is dissected, which means that in every line there are several elements that are not equal to zero; c) the matrix’ non-zero elements are distributed according to a certain scheme - the matrix becomes a cellular matrix. Many different methods may be applied to solve problems of such a type. The methods used in this chapter to solve difference systems have the best method characteristics for each of the specific problems (statics, quasistatics and dynamics). The term ‘method characteristic’ may be applied to numerical stability, accuracy, computation time economy or the amount of computer memory involved in the process of computation. This chapter describes the algorithms that carry optimum methods into effect and verifies correctness of the results. Finally, it presents the investigation of the influence of continuity conditions on the solution’s behaviour. 1.3.1 Dimensionless Equations Many problems of mechanics that involve deformable bodies require bringing the equations to their dimensionless forms, which in consequence allows estimating

40


physical phenomena in many other similar objects. Dimensionless parameters in thermoelasticity problems ui , τ, xi , θ, are related to dimensional quantities through the folowing dependences [79, 217, 384]: xi = i xi , τ = ℵ=

τ32 ET 0 α2T , θ = θT 0 , ui = ui αT T 0 , β = , α (1 − 2ν)cρ

λq T 0 τT α2 ρ ET 0 αT = , P4 = P4 2 , Pi = Pi , f = f EαT T 0 , τM i ET 0 32 3

T i0 =

T 0i − T 0 T λq T 0 αi 1 2 , BI = , Ti = i , λ1 = , λ2 = , T0 λq i 2 3 ⎧ −2 ⎪ λ , i=1 ⎪ ⎪ ⎪ 3 3 ⎨ λ3 = , λ = ⎪ λ22 , i=2 , ⎪ ⎪ 1 ⎪ ⎩ 1, i=3

(1.84)

where ℵ is the system’s inertia coefficient within the range of · 106 text− − − · 108 [217] at measured in centimetres. The system of differential equations for a homogeneous isotropic material of three-dimensional plate (1.23)–(1.24) takes the following form (the unknowns are dimensionless quantities): λ−2 λ2i ∂2 ui ∂2 u j ∂2 uk 1−ν k + + + 2 2 (1 + ν)(1 − 2ν) ∂xi 2(1 + ν) ∂x j 2(1 + ν) ∂xk2 2 j ∂u ∂θ 1 1 ∂2 uk ∂2 ui − + + Pi = λℵ 2 , 2(1 + ν)(1 − 2ν) ∂xi ∂x j ∂xk ∂xi 2(1 − 2ν) ∂xi ∂τ 2 1 2 2 2 3 2 2 2 ∂u ∂ u ∂ u ∂θ ∂ θ ∂θ ∂ θ + P4 = . (1.85) λ23 2 + λ−2 + 2 −β 2 2 ∂x1 ∂τ ∂x2 ∂τ ∂x3 ∂τ ∂τ ∂x1 ∂x2 ∂x3 The dimensionless conditions on the plate’s walls have the following form: j i (1 − 2ν) (1 + ν) i ∂u ∂u 1 ν ∂uk i θ ni,i+m + Ei+m fi+m , − + + i ui = Dii+m ∂xi 1 − ν ∂x j ∂xk 1 − 2ν 1−ν j i ∂u j j j 2 ∂u ni,i+m + Ei+m + λi j u j = 2λi (1 + ν) fi+m , Di+m ∂xi ∂x j k i ∂u k −2 ∂u k k ni,i+m + Ei+m + λk k uk = 2λ−1 Di+m k (1 + ν) fi+m , ∂xi ∂xk Ci+m

∂θ = BI(T i0 − θ)Ai+m + Bi+m T i , ∂xi

(1.86)

where α is a heat exchange coefficient. The dimensionless coefficients of the equations related with the parallel walls are equal.


41

1.3.2 Systems of Elliptic Difference Equations The problems concerning elliptic difference equations are solved with iterative methods. Certain interesting characteristics constituted of several unknowns [50, 52, 431] are the objects of study in this case. A Dirichlet model for Poisson equation will serve as a model problem in our considerations:

y x¯1 x1 + y x¯2 x2 = − fi j , fi j = f x1i , x2 j , yi j |γ = ϕi j .

(1.87)

The investigation will be carried on for the following case: a) yi j |γ1 +γ2 = ϕ0 , yi j |γ4 = ϕ1 , yi j |γ6 +γ7 = ϕ2 , 0 ≤ x1 ≤ m2 , x − m1 , 0 ≤ x2 ≤ l3 , m2 − m1 ⎧ ⎪ x2 − l1 ⎨ ϕ0 , γ3 yi j |γ8 = ϕ0 + (ϕ2 − ϕ1 ) , k=⎪ ⎩ϕ ,γ ; l2 − l 1 2 5 yi j |γ3 ,γ5 = k (ϕ1 − k)

(1.88)

b) on a square-shaped surface (Fig. 1.2b) 0 ≤ xi ≤ 1: yi j | x1 =0 = e3x2 , yi j | x2 =0 = cos 3x1 , yi j | x1 =1 = e3x2 cos 3, yi j | x2 =1 = e3 cos 3x1 , y = e3x2 cos 3x1 .

(1.89)

Seidel’s method is analysed and so is the method of upper relaxation, the explicit and implicit method of variable directions and the direct method of variable directions with Chebyshev’s acceleration. The methods are compared in consideration of the velocity of convergence and the problem of choosing parameters that makes it possible to accelerate the iterative processes is investigated. All the mentioned methods are characterised by simplicity of realisation and sufficiently good convergence. Let us consider the following system of linear difference equations: Ay = f .

(1.90)

The equation that describes the method of upper relaxation takes the following form:

(k) (k−1) + 1 − ωopt y(k−1) + ωopt fi j , (1.91) y(k) i j = ωopt L1 yi j + L2 yi j ij where: A = E+D+F, L1 = E/D, L2 = F/D, whereas D, E and F denote respectively diagonal, upper and lower triangular components of the A passage. At ωopt = 1 method (1.91) transforms into Seidel’s method ωopt =

2 , ) 1 + 1 − λ21

(1.92)

42


a)

x1 m2

g4 g5

g3 m1

g6

g2 g1

g8

l1 b)

g7

l2

l3

x2

x1 l1

l2

x2

Figure 1.2. Dirichlet’s problem for Poisson’s equation: a) on a T-shaped surface, b) on a square-shaped surface.

where λ1 denotes the spectral radius of the A passage matrix corresponding to Seidel’s method. The theoretical value of λ1 (quoted from appropriate literary sources) in a rectangular field is equal to [431]: λ1 =

h21 2(h21

+

h22 )

cos

h22 πh2 πh1 cos + , 2 l2 l1 2(h1 + h22 )

cos

h21 h23 πh1 πh2 + 2 2 cos + l1 l2 h1 h2 + h22 h23 + h23 h21

yet, for a cubicoid it equals: λ1 =

h22 h23 h21 h22 + h22 h23 + h23 h21

(1.93)


h21 h22 h21 h22

+

h22 h23

+

h23 h21

cos

πh3 . l3

43

(1.94)

In order to determine ωopt within an arbitrary field, an approximate value of λ1 is used in equation (1.92). There is a fundamental relationship binding the spectral radius of Seidel’s matrix and the spectral radius of the m1 and ωopt upper relaxation matrix. (µ1 + ω − 1)2 = λ21 . (1.95) ω2 µ1 By substituting formula (1.95) into (1.92) is possible to estimate approximately = µ1Mm can be calculated according the value of ωopt . The approximate value of µm−1 1 to the following formula based on Seidel’s iterative process [52]: (( (( y(k+1) − y(k) ij ij (y(k+1) − y(k) ( ij ( (( = , (1.96) µm 1 = lim ( (k) k→∞ (y y(k) − y(k−1) − y(k−1) ( ij ij ij

where µm is a minimum number, for which the following condition is fulfilled: Mm µ1 (1.97) Mm−1 − 1 ≤ εµ , µ 1

where: ω0 , εm u are set quantities (we may also assume that ω0 equals one). It appears that the velocity of convergence may be significantly increased, if ω0 is chosen exact to one hundredth. It can be achieved by applying the equivalent rectangle principle that consists in searching for the spectral radius of Seidel’s method passage matrix for the field of an equivalent rectangle, according to formula (1.93). The equivalent rectangle is such a rectangle, the surface of which is equal to the investigated surface and which is constructed according to the following principles: its width is the measurement of the largest circle that can be drawn within the surface’s boundaries (the circle should not cross the boundary) and its length corresponds to the surface of the investigated field divided by the field’s width. The dependence of the number of iterations necessary for obtaining solution of the set accuracy εit = 10−6 on parameter ωopt can be found in Table 1.1. Table 1.1. The dependence of the number of iterations N on ω (the upper relaxation method). ω N

1 247

1.5847 100

1.6073 96

1.65 89

1.6752 94

The formulas that describe the overt method of variable directions have the following form [163]: ! " (k+ 12 ) (k+ 12 ) (k+ 1 ) + A2 y(k) yi j 2 = y(k) |∂r = ϕi j , i j − τk A1 yi j i j , yi j

44


! " (k+ 12 ) (k+ 12 ) (k+1) ∂r = ϕi j , y(k+1) = y − τ y + A y A , y(k+1) k 1 2 ij ij ij ij ij

(1.98)

where: A1 yi j = a1i j yi−1 j −

a0i j 2

yi j + a2i j yi j−1 , A2 yi j = a3i j yi+1 j −

a0i j 2

yi j + a4i j yi j+1 .

(1.99)

Parameter τk is chosen to be equal to 1/hav [170], and numerical experiments show that changes of τk have little influence on the velocity of convergence (Table 1.2). To increase the convergence velocity of the overt method of variable directions, applying Chebyshev acceleration of convergence [52] is recommended. In result, the iterative process takes the following form: (k+1) = y(k) − y(k) y(k+1) ij i j + αk (y i j ),

(1.100)

ij

where y(k+1) is calculated according to formulas (1.98), (1.99), αk are the coefficients ij used for increasing the convergence velocity. In case of Chebyshev’s cyclic method, αk is determined according to formula [294, 318] −1

2k − 1 π αk = 2 M + m − (M − m) cos N

,

(1.101)

where m and M are respectively the minimum and the maximum proper values of the passage matrix in Chebyshev cyclic method, which is based not on formulas (1.98) and (1.99), but on the implicit method of variable directions. For the iterative process being discussed here, αk is expressed by means of M in the following way [52]:

M 2k−1 2 1 + cos 2N π .

(1.102) αk = 1 − M2 1 + cos 2k−1 2N π M is determined approximately with the use of Lusternik’s algorithm [316]. The iterative process is realised by making N1 iterations according to formulas (1.98) and (1.99) with (( ( ((y(k) − y(k−1) ((( i j i j ( ( , at λ(k) (1.103) λ(k) 1 = ( (k−1) 1 → M, when k → ∞ . (k−2) ( ((y − yi j (( ij

Table 1.2. The dependence of the number of iterations N on τ (the overt method of variable directions). τ N

15 77

16 75

20 64

26 70

32 81


45

(m) If λ(k) 1 begins to differ slightly between subsequent iterations then λ1 may be assumed as approximate to M = λ(m) 1 . Next, αk is calculated for k = 1, ..., N and further operations are conducted with acceleration (the parameter changes periodically with the period equal to the number of parameters - N). The method is characterised by greater velocity of convergence than the overt method of variable directions and the above mentioned upper relaxation method. The results of the theoretical investigations of the convergence velocity [109, 431] have also been practically proven (Table 1.3).

Table 1.3. Comparison of iterative methods for a T -shaped field.

Seidel’s method Upper relaxation method Explicit method of variable directions Chebyshev’s acceleration method

Theoretical velocity of convergence π2 h2 2πh 2πh √ 2π h

N

εit

800 57 55 30

10−4 10−4 10−4 10−4

In order to choose the most effective solution method, a comparison of a numerical solution and an exact solution (1.89) [50, 52] has been made on different nets, with various parameters accelerating the iteration process, in relation to exact location of spectral characteristics of passage matrixes, with various approximation errors, and also in relation to the velocity of convergence (Table 1.4). Table 1.4. Comparison of iterative methods for a square-shaped field. Seidel’s method Scheme O(h2 ) O(h4 ) Parameter 1 1 N 205 196 εit 10−6 10−6 Error 7 · 10−3 6 · 10−4 Theoretical velocity of π2 h2 Λ0 2π2 h2 convergence

Upper relaxation Variable directions method method Implicit Explicit O(h2 ) O(h4 ) O(h2 ) O(h4 ) 1.8225 1.792 – 20 92 88 12 88 10−6 10−6 10−6 10−6 6 · 10−4 1 · 10−4 3 · 10−5 1 · 10−4 2πh

2.09πh

–

πh

This work does not include any iterative formulas of the implicit method of variable directions because of their complexity (see [431]). Numerical experiments have shown that among all the considered methods the most effective one for a square is the upper relaxation method of an increased order of accuracy. The number of iterations necessary to achieve the set accuracy with the use of the implicit method of variable directions is smaller. However, it is not economical as far as such factors as machine computation time, sophistication of software and memory capacity (it

46


requires twice as much memory as the other methods) are concerned. The explicit method of variable directions works well in a T -shaped field (Table 1.3). The results obtained with the use of the upper (successive and block) relaxation method, the explicit and implicit methods of variable directions and the method of variable triangles with Chebyshev’s acceleration have been compared within a ring-shaped field [61] for the first and the third boundary condition in the polar coordinates system. Despite the fact that the theoretical and the numerical velocities of convergence in the method with Chebyshev’s acceleration are higher than in all the other methods, the method’s poor accuracy makes it practically useless (Table 1.5).

Table 1.5. Comparison of iterative methods for a ring-shaped field. εit

N

Error

10−3

7

0.01703

1 4 4 ln ln π2 εit η

10−3

15

0.00073

1 1 √ ln 4 η ε

Implicit method of variable directions Explicit method of variable directions

Number of iterations necessary to achieve set accuracy

Chebyshev’s acceleration method

2 ln ε √ √ 2 2 4η

10−3

37

0.06503

10−3

142

0.00015

ln

0.00015

* 1 ln ∆1 ln ε

Succesive upper relaxation method Block upper relaxation method

−3

10

47

* 1 ln ∆1 ε

The results of the investigation testing applicability of a number of methods have shown that the most useful method of analysing solutions for a cubicoid (also for static problems of thermoelasticity) is the upper relaxation method with error O(h2 ). An increased-order scheme within a rectangle-shaped field brings better results in case of a more complicated system of differential equations and a more complicated field, in which a slight difference in the number of iterations leads to significant complications of the difference scheme due to a larger number of approximation nodes. Another important problem is the choice of a digitisation step of the investigated field’s (cubicoid’s) mesh. Runge’s law [37], used so often, makes it possible to choose optimum steps h1 = h2 = h3 = 0.125. Further decrease of the net’s steps does not result in serious changes in obtained results.


47

1.3.3 Systems of Parabolic and Hyperbolic Difference Equations Works [97, 399, 419, 494] present a generalized method of constructing difference schemes for numerical solutions of non-stationary problems of heat conductivity and continuous media dynamics. A large number of publications devoted to that issue has been listed in book [399]. The system of difference equations (1.38)–(1.45) is presented in an explicit form - the most convenient for numerical integration. However, explicit schemes are not always stable. Conditions (1.74) are imposed on the difference scheme’s steps. It turns out that limitations concerning the practical use of explicit schemes may be very significant and that is the reason for applying implicit schemes. Implicit schemes are absolutely stable and the only limitation they impose on the time step is the condition of accuracy. Nevertheless, in contrast to explicit schemes, simple implicit schemes for multidimensional problems turn out to be uneconomical [97, 108]. That is why from now on we are going to consider only explicit schemes. We are going to present numerical experiments concerning a model problem applied for comparison of solutions with the use of explicit finite difference schemes (1.38)–(1.45) with accuracy O(h44 + h2 ), then with Runge-Kutta’s method with accuracy O(h44 + h2 ) and constant step and with Runge-Kutta’s method with accuracy O(h44 +h2 ) with automatic step choice. The model problem is going to be represented by Cauchy’s problem [51, 60] for the following system: d2 x1 = f1 (x1 , x2 ) , dτ2 d2 x2 = f2 (x1 , x2 ) , dτ2 x s (0) = x0s ,

(1.104)

describing movement of charged particles with initial velocities and motion start coordinates. Change of the variables allows decrease of the system’s order: dϕ s dx s = ϕ s (x1 , x2 ), = f s (x1 , x2 ), dτ dτ x s (0) = x0s , ϕ s (0) = x˙0s , s = 1, 2 .

(1.105)

Explicit difference scheme (1.38)–(1.45) takes the following form: yτs = ϕisj , y s (0) = x0si j , ϕτs = fisj , ϕ s (0) = x˙0si j , s = 1, 2 .

(1.106)

Formulas that describe Runge-Kutta’s method are commonly known [397, 474]. The most widely applied is the method with order O(h44 ). On one hand, the investigated methods reveal very good characteristics: in order to determine values in the next point they only require information about the previous point. This makes them

48


economical as far as the use of computer memory is concerned. On the other hand, the necessity to repeat calculation of the right sides at every integration step is a serious fault, since it takes significantly longer calculation time. The finite difference method (1.105) allows for computing the right sides only once but it considerably affects accuracy of the results. The trajectory that corresponds to Runge-Kutta’s computation method is the closest to the real solution, which confirms theoretical considerations concerning the error’s order. Therefore, all further calculations have been made with the use of this method. It is necessary to point out that RungeKutta’s method has been “accepted” as stable. In case of the automatic integration step choice based on the approach described in publication [37], the calculation time increases two or four times depending on how far the first step is from the optimum step. That is the reason why such modification is not applied. Stability conditions that definitely impose limitations in regard to the integration step in time and to the integration step in space are presented in Section 2.2.3. They have been applied in order to obtain an optimum solution with the use of Runge-Kutta’s method. Numerical experiments show correctness of this approach. The following problems have been solved: a) thin plate’s vibrations, b) the problem of non-stationary heat transfer (problems no. 10 and 12 formulated in Section 2.3) with the use of various integration steps. Graphs presented in Figures 1.3 and 1.4 illustrate: a) deflections in the centre of the plate in time, b) temperature distribution in the centre of the plate in time. At steps h4 > h 4,opt and h4 > h

4,opt the process is unstable and it does not reflect the real physical phenomenon. Through decreasing the step and thus approaching the optimum value, that may be calculated from inequality (1.79) for h = 0.125, the processes start becoming stable, and at h4 ≤ h 4,opt and h4 ≤ h

4,opt they describe real vibrations and real changes of temperature. If the dynamic problems and the non-stationary heat conductivity problems were solved independently, then in every case the stability condition should be applied to quicken the calculation process, since the optimum steps h 4,opt > h

4,opt are different from each other. All calculations in this work have been made with the following integration steps: h 4,opt = 0.05 and h

4,opt = 0.00125. In case of Runge-Kutta’s method, the theoretical basis for assuming stability conditions (1.79) is the fact that calculations are characterised with high accuracy [37] and are described by the following formula: yi+1 = yi + h f (x, y) ,

(1.107)

which refers to the finitely dimensional approximation (1.106) obtained in Section 2.1. The conducted consideration leads to a conclusion that Runge-Kutta’s method may be applied together with the conditions of the theorem presented in Section 2.1. 1.3.4 Algorithm Mathematical descriptions of all kinds of problems (static, quasistatic, dynamic and coupled) are characterised by many types of systems of differential equations (elliptic, parabolic, hyperbolic or parabolic-hyperbolic). Solutions to such equations can

1.3 Methods of Solving Difference Equations u

49

3

0.05

x2

x1 x3

Dt= Dt = 0.05 0.025

Dt= Dt = 0.2

0

4

6

8

t

10

k=25

-0.025

-0.05

Figure 1.3. The graph of the deflection in the plate’s centre in time (the problem of the nonstationary heat transfer during a thin plate’s vibrations).

be obtained by applying various methods or combinations of methods. For this reason, the algorithms that realise numerical calculations will be described separately. Algorithm 1. Solving static problems. Static problems in three-dimensional problems and stationary heat conduction problems are described mathematically by a system of equations or one elliptic equation formed as a result of “breaking” system (1.38)–(1.39) without taking

50

1 Three–Dimensional Problems of Theory of Plates in Temperature Field 3

u ,µ 0.9

u

0.8

3

0.7 0.6 0.5

µ

0.4 0.3 0.2 0.1 0

0.01

0.02

0.03

0.04

t

-0.1 -0.2

h4'=0,1

-0.3

h4''=0,01

-0.4 -0.5 -0.6

Figure 1.4. Fig. 4. The temperature distribution in a plate’s centre in time (the problem of the non-stationary heat transfer during a thin plate’s vibrations).

inertia forces into account. In order to solve this problem we shall use the upper relaxation method. Each of the system’s four equations takes the following form:

Λy(i) = Λi xi , y xj j xi , ykxi xk , y4 , i → j → k,

(1.108)

where Λ denotes a difference analogue of Laplace’s differential operator. Formulation of boundary conditions at the field’s boundary coincides with conditions (1.40)–(1.45), thus the algorithms realising calculation methods are different. The iteration formula, both for the system and one equation, can be notated in the following way: ai yis(k+1) = (1 − ωopt )ai yis(k) + ωopt

i−1 j=1

a j y s(k+1) − ωopt j

M

a j y s(k+1) + ωopt Λis , (1.109) j

j=1

where i denotes transitional numeration of the three-dimensional field of the vector’s elements, y s (y1s , ..., y sM ), M = N1 N2 N3 , k - number of iterations, ai - coefficients at the unknowns. Parameter ωopt is chosen according to the relations made in Section 2.2. First, initial approximations are set in the entire field of the mesh and boundary conditions are set on its boundaries, where Dirichlet’s problem is considered. Then cyclic calculations are made according to formulas (1.96), (1.97), (1.95), (1.92),


51

(1.109). While solving Neuman’s problem or the third boundary problem in the boundary nodes for every iteration within the field, the calculations are made according to the following formulas (since it is impossible to present dimensionless conditions in a boundary node by means of an index, we shall use formulas for an edge and a corner as an example): −1 h2 h2 1−ν (1 − ν) h1 λ1 1 5 D1 + λ1 D1 · = h1 (1 − 2ν)(1 + ν) h2 2(1 + ν) h1 (1 − 2ν)(1 + ν) h1 1(k−1) v 2(k) 1(k) 1(k) 1 1(k) 3(k) D1 yi−N1 − − 3yi−N1 + 4yi−2N1 + y − y x3 − y 1 − ν x2 2 i h2 2(k−1) h2 1(k−1) h1 1(k) 1(k) D15 λ1 y1(k) y + + − 3y + 4y − y i−1 i+1 i+2 2h2 (1 + ν) λ 1 x1 2 i 1 4(k−1) 1 1 y , (1.110) f5 h1 + f1 h2 − 1 − 2ν i ⎞ ⎛ D22 ⎟⎟⎟ h1 ⎜⎜⎜ 1 (1 − ν) (1 − ν) 2(k) 2 + D5 · yi = ⎝⎜ ⎠⎟ h2 h1 2(1 + ν) (1 − 2ν)(1 + ν) h2 (1 − 2ν)(1 + ν) h2 2(k−1) h2 1(k) 2(k) 2(k) 2 2(k) 3(k−1) D5 yi+1 + ν − 3yi+1 + 4yi+2 + − y − y x3 y 1 − ν x1 2 i h1 2(k−1) h2 λ 1 2(k) 2(k) 2 2(k) −1 1(k) D y λ − h1 λ1 y x2 − − 3yi−N1 + 4yi−2N1 + y h1 2(1 + ν) 1 i−N1 1 2 i 1 y4 , f12 h2 − h1 f52 + 1 − 2ν −1 h2 3 2(k) h2 3 h1 3 h1 3(k−1) 1(k) y3(k) = D1 + D5 λ1 D1 yi−N1 + h1 λ−2 − y 3 y x3 − i h1 h2 h1 2 i % h1 3(k) h2 3(k−1) 3(k) 3 −1 2(k) 3y3(k) + 4y − i−N1 i−2N1 + h D5 yi+1 λ1 − h2 λ3 λ2 y x3 − 2 yi 2 %

3(k) 3 −1 −3y3(k) + 2 h2 f13 λ−1 3 − h1 f5 λ3 (1 + ν) , i+1 + 4yi+2

y1(k) i

= y4(k) i

1 $ 4(k) + B5 T 5 h2 + B1 T 1 h1 − BI C1 h2 y4(k−1) + +C5 h1 y4(k−1) yi−N1 + y4(k) + i i i+1 c1 + c5 h1

h2 4(k−1) 4(k) 4(k) 4 4 4 − 3yi+1 + 4yi+2 + (1.111) y y − 3yi−N1 + 4yi−2N1 . 2 i 2 i

The dimensionless relations of calculations in corner ∂Ω4 ∂Ω2 ∂Ω3 at h1 = h2 = h3 take the following form: 3(k−1) h1 1(k−1) $ 2(k) ν 1(k) 1 y1(k) = D − + y1+N1 N2 − y3(k) − y2 − y2(k−1) y 4 y1+N1 − 1 1 1 2(1 − ν) 2 1

52


1 2(1 − 2ν)(1 + ν) 4(k−1) 1 (1 − ν) + h1 f1 + y + − 1−ν 1 − 2ν 1

3(k) $ % h1 1(k−1) 1(k−1) − 3y1(k) + 4y1(k) λ23 − + D13 y1(k) y1 1+N1 N2 − y1+N1 − y1 1+N 1+2N N N 1 2 1 2 2

h1 1(k−1) 2(k−1) − y2(k) λ−2 2(1 + ν)h f31 λ3 (1 − 2ν) + D12 y1(k) y 1 − 2 1+N1 − y1 2 1

% 1 − 2ν 1(k) 1 −1 1 1 1 (1 (1 D 3y1(k) + 4y f λ + − 2ν) D + D + 2 + ν) h , 2 2 1 4 2 3 2 3 2 (1 − ν)

1(k) h2 2(k−1) = D24 y2(k) − y1(k) − 3y2(k) + 4y2(k) y2(k) λ21 − − y1 1 1+N1 + y2 1 1+N 1+2N 1 1 2

1(k) % ν y 2 (1 + ν) h1 λ1 f42 (1 − 2ν) + D22 y2(k) + − y1(k) + y3(k) 2 1 1+N1 N2 − 2 (1 − ν) 1+N1 h2 1(k) (1 − 2v) (1 + ν) f22 + y3(k−1) + 4y1(k) − + y1 − 3y1(k) 1 2 3 2 h2 (1 − ν)

$ 1 y4(k) + 2 (1 + ν) + D23 y2(k) + y32 − y31 λ−2 2 − 1 1+N N 1 2 1 − 2ν % h2 2(k−1) 2(k) 2 −1 (1 (1 − 2ν) · − 3y2(k) + 4y f λ + 2 + ν) h y1 1 3 2 1+N 1+2N N N 1 2 1 2 2

% $ (1.112) 2 (1 − ν) D22 + (1 − 2ν) D24 + D23 , h2 3(k−1) 1(k) 1(k) = D34 y3(k) − − 3y3(k) + 4y3(k) − y3(k) y1 1 1+N1 + y1+N1 N2 − y1 1+N 1+2N 1 1 2

% $ 2(k) + y2(k) λ22 − 2 (1 + ν) h3 f43 (1 − 2ν) + D32 y3(k) 2 1+N1 N2 − y1 ν h1 3(k−1) 3(k) 3 3(k) (1 · − 3y3(k) + 4y − 2ν) + D y1 3 y1+N1 N2 + 2 3 2 2 (1 − v) h2 3(k−1)

1(k) 1(k) 2(k) 2(k) 3(k) 3(k) − 3y1+N1 N2 + 4y1+2N1 N2 + y1+N1 − y1 + y2 − y1 − y 2 1 $

%−1 1+ν 1 4(k−1) 3 (1 − 2ν) h1 f3 − y1 2 (1 − ν) 2(1 − ν)D33 + (1 − 2v) D32 + D34 , 1−ν 1 − 2ν C4 h 4(k−1) 4(k−1) 4(k) y4(k) = C4 y4(k) + − 3y4(k) + 4y + y1 1 1+N1 + B4 hT 4 − BIA4 y1 1+N 1+2N 1 1 2 C2 h 4(k−1) + B2 hT 2 − BIA2 y4(k−1) + − 3y4(k) + 4y4(k) + y1 C2 y4(k) 2 1 2 3 2 C3 h 4(k−1) 4(k−1) 4(k) 4(k) (C2 + C4 )−1 . + B hT − BIA y + − 3y + 4y C3 y1(k) y 3 3 3 1+N1 N2 1 1 1+N1 N2 1+2N1 N2 2 3y1(k) 1+N1

4y1(k) 1+2N1

%

Calculations are finished as soon as the following criterion is satisfied:


s(k) yi − yis(k+1) s(k) < εit . max i,s y

53

(1.113)

i

In case of a stationary heat transfer problem the applied algorithm is assumed for one equation in order to calculate y4 . When a static elasticity problem is solved the algorithm is applied in the system with unknowns y s , s = 1, ..., 3. For the problems that refer to thermal stress according to the algorithm, the temperature distribution is investigated first and then the algorithm is used again to determine the displacement distribution y s , s = 1, ..., 3. Algorithm 2. Solving dynamic problems, non-stationary heat transfer problems and coupled thermoelasticity problems. Similarly to the case of model problem (1.106)–(1.107), through changing the variables systems (1.38)–(1.45) lead to the system of seven equations. Apart from displacements y s , s = 1, ..., 3 and temperature y4 , the velocities of displacements are also unknown. At the first stage of solving the problem, the initial conditions are set for all unknown nodes. Then, depending on the type of the initial-boundary conditions, the known values of surface functions are set (in case of the first boundary problem). Next, a step in time is made using Runge-Kutta’s method formulas for every unknown node within the net’s field, and the analogous right sides of equations (1.38)–(1.39) are calculated. After completion of the calculations for all the field’s inner nodes, the analyses of the types of boundary problems are made at every time step. If the second or the third initial-boundary conditions are solved, then values y4 , that correspond to a given type of a problem, are calculated in the nodes according to relations (1.110), (1.111), (1.112). Calculations are made in the same manner within the whole time range. While solving uncoupled thermoelasticity problems, the discussed algorithm is applied to system (1.38)–(1.45), and the coefficient placed at the dilatation term in the heat conductivity equation is assumed to be equal to zero. Algorithm 2 may be used to calculate the dynamic behaviour of plates affected by all kinds of volume and surface forces, with zero coefficients placed at the thermal gradient expression. The system of equations is reduced to six equations. The number of boundary conditions is also reduced. If algorithm 2 was to be applied solely to the heat conduction equation (1.39) with thermal boundary conditions (1.41), (1.43), (1.45), then the solution would be the distribution of the temperature field of a three-dimensional plate subjected to various types of thermal load, including inner heat sources. One particular feature (free form the influence of thermal loads) of solving dynamic problems for thick plates with the use of algorithm 2 is worth emphasising. The thicker the plate, the more unstable the solution is. An analysis of that process has shown that instability occurs in proximity of the field’s boundary. Since the boundary conditions are calculated according to formulas (1.110), (1.111), (1.112), i.e. while solving the third boundary problem, the instability penetrates into the field. The instability disappears if the free edge condition is replaced with another condition for any other type of support. The instability disappears also along with the decrease of the plate’s thickness (for λ1 = 1/4, 1/10, 1/50, 1/100 - the calculations

54


are stable; for λ1 = 1/2 - slight instability occurs; for λ1 = λ2 = λ3 = 1 - the calculations are unstable). Running the iteration process within the field’s boundaries prevents instability from occurring in case of thick plates. In case of thin plates though, the process still remains unstable (for λ1 = 1/4 - instability disappears after 16 time steps; for λ1 = 1/10, 1/100 - instability disappears after the first time step). Due to the conclusions drawn above, there may be two cases of algorithm 2, depending on the investigated plate’s thickness: a) the solutions for thin walls are obtained according to the described algorithm 2; b) the solutions for thick walls’ boundaries are obtained through multiple application of formulas (1.110), (1.111), (1.112). Case b) requires a more thorough study since it consists of the following stages: 1. Calculating y4i according to formulas (1.41), (1.43), (1.45). 2. Successive calculation of yi(m) , y j(m) , yk(m) is made according to formulas (1.110) for all nodes of the wall. 3. Successive calculation of yi(m) , y j(m) , yk(m) is made according to formulas (1.111) for all nodes of the edges. 4. Successive calculation of yi(m) , y j(m) , yk(m) is made according to formulas (1.112) for the corner. The initial values in the nodes are the values calculated in the previous time step. The formulas’ values of the field’s inner nodes do not change during the entire iteration process. Moreover, an assumption is made that the displacement has an index of an axis perpendicular to the investigated wall. 5. Assumption of condition (z) yi − y(z+1) i (z) < ε1it max i yi and return to stage 2 if the condition is not fulfilled. Algorithm 3. Solving quasistatic problems. Combinations of algorithms 1 and 2 presented above produce an algorithm of solutions to a quasistatic problem. In this case, a non-stationary heat conduction equation (1.39) is solved, and the plate’s stress-strain state is additionally determined at every time step, i.e. in system (1.38) inertial terms are assumed and a static problem is solved. Algorithm 2 for solving problems of non-stationary quasistatic of heat conductivity is interrupted at every time step in order to fulfil algorithm 1 for the mechanical part with the temperature gradient, then it is repeated to carry on the calculation. 1.3.5 Reliability The algorithms described in Section 2.4 have been written using FORTRAN software. The programs have been tested on model systems, thus proving their correctness. In case of statics, the following forms have been assumed for the investigated functions:


55

u1 = (x1 − 0.5)2 + (x2 − 0.5)2 x3 , u2 = (x2 − 0.5)2 + (x3 − 0.5)2 x1 ,

u3 = (x3 − 0.5)2 + (x1 − 0.5)2 x2 , θ = (x1 − 0.5)2 + (x2 − 0.5)2 x33 .

(1.114)

In case of dynamics, the functions have assumed the forms of:

u1 = (x1 − 0.5)2 + (x2 − 0.5)2 x3 τ2 , u2 = (x2 − 0.5)2 + (x3 − 0.5)2 x1 τ2 ,

u3 = (x3 − 0.5)2 + (x1 − 0.5)2 x2 τ2 , θ = (x1 − 0.5)2 + (x2 − 0.5)2 x33 τ. (1.115) There have been combined boundary conditions set upon the walls – free edge and fixing, moving support and free edge, fixing and moving support, temperature distribution and thermal insulation, thermal insulation and convectional heat circulation with surrounding medium, temperature distribution and heat circulation. The approximate solution has been compared to the exact one (reliability of calculations has been confirmed on basis of the results yielded by other authors). Consistency of the results has been achieved exact to 5-6 digits after comma (Tab. 1.6, Tab. 1.7). A comparison has been made in points (1/2, 1/2, 1/2), (7/8, 1/4, 1/2) with the following types of support: one free, thermally insulated wall and the others - fixed with set temperature distribution. The stress-strain state of an isotropic one-dimensional plate subjected to a uniformly distributed load has been investigated in statics. The results of the comparison (see Section 3.1, problem 2) are consistent with the results obtained by Kornishin [345]. As far as non-stationary problems of heat transfer are concerned, an investigation of the temperature field distribution subjected to a heat impact (∂θ/∂x1 = 13.3) onto one of a cubicoid’s walls (with the other walls thermally insulated: ∂θ/∂n = 0) has been conducted. The coordinates ‘temperature in the plate’s centre-time’ of the graph presented in figure 1.5a and the temperature distribution along the plate’s thickness at various time instants presented by the graph in figure 1.5b represent the results consistent with the ones obtained in work [198], in which the temperature field of an infinite plate affected by heat impact has been investigated (the point curve represents Kovalenko’s results). Comparison of numerical and analytical solutions is also a vital confirmation of the reliability of calculations. Fourier’s method has been applied for a non-stationary problem of heat conductivity without the set temperature distribution at the field’s boundary and with a single, uniformly distributed inner heat source, in order to obtain a solution [49] of the following form: τ1 √ ∞ 10 √ −3π2 (τ −τ) 2 2 −14π2 τm 1 2 2 e dτ ≈ e ≈ 0.054 . θ (0.5, 0.5, 0.5, 0.01) = 14π2 m=1 m=1 τ0

The approximate numerical solution obtained in this point of the field equals y4 = 0.0544. The results of the thin plate’s dynamic behaviour obtained in this work have been compared to the results presented in work [261].

56

1 Three–Dimensional Problems of Theory of Plates in Temperature Field a)

µ

x1=1

100 90 80

x1=0.5

70 60

x1=0

x2

50 40

x1 x3

30 20 10 0

0.1

0.2

0.3

0.4

0.5

0.6

µ

b) t=0.6

120 110 100 90 80 70 60

t=0.1 t=0.05

50 40 30 20 10 0

Figure 1.5. The distribution of the temperature field affected by a heat impact: a) in coordinates ‘temperature in the plate’s centre-time’, b) the temperature distribution along the plate’s thickness at various time instants.


57

An investigation of a plate with mechanical characteristics for aluminium: ν = to a uniformly distributed load P = 1, at initial 0.3, E = 7.2 · 108 g/cm2 , subjected conditions ui τ=τ = 0, ∂ui /∂ττ=τ = 0 has been conducted. 0 0 The following support conditions have been assumed: joint (D11 = E12 = E13 = 1, E31 = D23 = D33 = 0, D12 = E22 = E23 = 1, E51 = D25 = D35 = 0), free edge (D13 = D23 = D33 = 1, E3i = 0, i = 1, ..., 3, Di6 = 1, E6i = 0), null surface forces, and dimensionless parameters: λ1 = 1, λ2 = 50, λ3 = 1/50. The results obtained for a thin plate described according to Kirchhoff-Love’s hypothesis with the use of combined Runge-Kutta’s method and the finite difference method have become the object of comparison [261]. Table 1.6. Comparison of exact and approximate solutions to static and stationary heat conduction problems.

u1 u2 u3 θ

Approximate solution Exact solution (1/2,1/2,1/2) (7/8,1/4,1/2) (1/2,1/2,1/2) (7/8,1/4,1/2) −0.35 · 10−5 0.1719 0 0.17188 −0.1 · 10−4 0.6249 · 10−1 0 0.625 · 10−1 −0.53 · 10−5 0.3151 · 10−1 0 0.3156 · 10−1 −6 −1 −0.408 · 10 0.4259 · 10 0 0.4297 · 10−1

Error (1/2,1/2,1/2) (7/8,1/4,1/2) 0.35 · 10−5 0.2 · 10−4 −4 0.1 · 10 0.1 · 10−4 −5 0.53 · 10 0.5 · 10−4 −6 0.41 · 10 0.4 · 10−4

Table 1.7. Comparison of exact and approximate solutions to dynamic and stationary heat conduction problems for τ = 0.1.

u1 u2 u3 θ

Approximate solution Exact solution (1/2,1/2,1/2) (7/8,1/4,1/2) (1/2,1/2,1/2) (7/8,1/4,1/2) 0.364 · 10−11 0.48 · 10−6 0 0 0.637 · 10−3 0.313 · 10−3 0 0.3125 · 10−3 −0.909 · 10−3 0.637 · 10−3 0 0.625 · 10−3 −5 −5 0.75 · 10 0.12 · 10 0 0

Error (1/2,1/2,1/2) (7/8,1/4,1/2) 0.36 · 10−11 0.48 · 10−6 0.641 · 10−11 0.5 · 10−6 0.91 · 10−11 0.1 · 10−5 −10 0.75 · 10 0.12 · 10−5

It is worth emphasising how precisely the curves ‘normal stress in the medium time’ overlap for the heat impact problem (which is described in Section 3.5 as Danilovskya’s problem). This phenomenon also confirms correct functioning of the program designed to solve dynamic problems of the theory of thermoelasticity. 1.3.6 Numerical Experiments Certain experiments concerning model problems of statics and dynamics (1.114)– (1.115) have been conducted in order to examine the influence of the compatibility conditions. While calculating the stress-strain state of a three-dimensional elastic plate (a cube) without taking mesh conditions into account, it can be observed that a decrease of the spatial mesh is accompanied by an increase of the calculation

58


time. Calculations that include the compatibility conditions in the singular points cause a decrease of the error’s value by 70% compared to exact values. The results presented in Table 1.8 suggest the necessity of taking the compatibility conditions into account. Such a conclusion has also been confirmed by calculations made for stationary heat transfer problems. In case of dynamic problems, considering the compatibility conditions in the calculations brings less satisfactory effects - the error’s value at every time step is about 3%. This can be explained by the fact that the pattern of a difference scheme with the error of O(h2 ) does not include the singular points (edges, corners). Therefore, they appear to be extremely significant during the calculations of the approximate values within the singularity field; they occur in the calculations of approximate solutions on the walls in the zone adjoining to the boundary. However, the errors overlap with every new integration step, therefore during long-term analyses it is possible to recognise the results as correct without decreasing the spatial net and without taking the compatibility conditions into account.

Table 1.8. Comparison of results obtained with the compatibility conditions at decreasing steps of the net.

u1 u2 u3 Calculation time

Without compatibility 1/8 1/12 1/16 0.36920 0.25860 0.24808 0.33750 0.21093 0.19816 0.19015 0.18925 0.18585 15 min

1h

4h

With compatibility 1/8 0.24608 0.19530 0.18555

Exact solution

17 min

–

0.24609 0.19531 0.18655

Table 1.9. Relation of the number of iterations for plates of various thicknesses, including the compatibility conditions. Plate’s thickness λ1 = λ2 = λ3 = 1 λ1 = 1, λ2 = 1/10 λ1 = 1, λ2 = 1/50

Number of iterations 101 85 70

Error 10−3 10−3 10−3

Eventually, the influence of the compatibility conditions on the results of the analysis of plates of various thicknesses has been examined. Table 1.9 presents the results of numerical calculations, in which the dependence of the approximate solution’s accuracy (set in advance) on the plate’s thickness has been emphasised. The importance of the compatibility conditions decreases along with the decrease of the plate’s thickness.

1.4 Linear Problems in the Theory of Plates in 3D Space

59

1.4 Linear Problems in the Theory of Plates in 3D Space This section presents the examined results of linear problems obtained with the use of the methods described in the previous section. The problems become more complex as the physical models grow more sophisticated. 1.4.1 Static Problems The reaction of an elastic plate subjected to mechanical (surface and volume) excitations for the plate’s various relative thicknesses is investigated. Through reduction of inertial terms and temperature gradients we are going to examine the stress-strain states obtained exclusively on basis of mechanical relations. The system that describes the investigated group of problems is elliptic and it should be solved with the use of the iterative method of upper relaxation. 1. Let us examine a cube-shaped plate, the walls ∂Ω2 , ∂Ω3 , ∂Ω5 , ∂Ω6 of which are fixed (E2i = 1, Di2 = 0, E5i = 1, Di5 = 0, E3i = 1, Di3 = 0, E6i = 1, Di6 = 0, i = 1, ..., 3), and walls ∂Ω1 and ∂Ω4 are free (E1i = 0, Di1 = 1, E4i = 0, Di4 = 1, i = 1, ..., 3). Problem 1. The plate is affected by surface forces f11 = f41 = 0.001, f12 = f13 = f42 = f43 = 0, and volume forces do not occur. Such a choice of mechanical reaction corresponds to a uniform distribution of an external transverse load, perpendicular to plate 0x2 x3 , with x1 = 0.5 (Fig. 1.6a) and x1 = 0.25 (Fig. 1.6b). Problem 2. The plate is affected by volume force P1 = 0.001, P2 = P3 = 0, surface forces do not occur, which corresponds to a uniform inner load (Fig. 1.7.) Qualitative representations of the stress-strain states in both problems for thick plates are different. The difference is that when the volume forces operate, displacement u1 within the entire volume is the same on plane x1 = 0.5 (Fig. 1.7a) and on plane x1 = 0.25 (Fig. 1.7b). When the surface forces operate though, displacement u1 near the active walls (Fig. 1.6b) is larger than in the plate’s centre (Fig. 1.6a). Along with the decrease of the plate’s thickness (the relative thickness is 0.5) one can observe similarity of qualitative representations of displacement distribution for problems 1 and 2 (Fig. 1.8 plane x2 = 0.5, x1 = 0.5). In the further stage of the experiment the plate’s thickness is reduced (Fig. 1.9 - the relative thickness is 0.02). The reaction to the surface load starts to coincide with the reaction to the volume force. Nevertheless, one can observe that the value of displacement u1 increases at equivalent loads along with the reduction of the plate’s size (Fig. 1.10 presents the relative thickness for corresponding curves) and it is identical along its thickness. In all the conducted experiments, displacements u2 and u3 are insignificantly small compared to displacement u1 . Solutions to problems 1 and 2 for plates of the relative thicknesses 0.1 and 0.02 have been compared to the results obtained by means of the method of mesh for a

60

1 Three–Dimensional Problems of Theory of Plates in Temperature Field b) x2

a) 1

u ·10

x1

-4

x3

4 3 2 1

1

x2

5 4 3 2 1

x3

u ·10

-4

x2

2

u ·10

x3

-4

4 3 2 1

x2

x3

3

u ·10

x2

2

u ·10

x2

-4

3

x3

x2

4 3 2 1

x3

u ·10

4 3 2 1

-4

-4

4 3 2 1

x3

Figure 1.6. The representation of the stress-strain state for problem 1 in case of: a) x1 = 0.5, b) x1 = 0.25.

plate described according to Kirchhoff-Love’s displacement hypothesis [261] (Fig. 1.11), as well as with Kornishin’s results [345] obtained according to a difference scheme of an increased order of accuracy (see Tab. 1.10). The relative error for a 0.1-thick plate is between 5% in the centre and 10% at the plate’s edges. For a 0.02-thick plate it is between 3% and 11%, respectively. Different relative errors along the middle plate and the increase of the errors as the thickness is reduced can be explained by means of Kirchhoff-Love’s method,

1.4 Linear Problems in the Theory of Plates in 3D Space x2

a)

b)

x1 x3 1

8 7 6 5 4 3 2 1

u ·10

1

u ·10

-4

-4

8 7 6 5 4 3 2 1

x2

x2

x3

u2·10-4

2

x2

x3

3

u ·10

x3

u ·10

4 3 2 1

-4

x3

3

u ·10

-4

x3

4 3 2 1

x2

3 2 1

x2

61

x2

-4

4 3 2 1

x3

Figure 1.7. The representation of the stress-strain state for problem 2 in case of: a) x1 = 0.5, b) x1 = 0.25.

usually applied for analyses of thin plates. The problems regarding relative thickness larger than 0.125 cannot be solved with the use of this method. A following conclusion may be drawn from the already made investigation. When a plate’s thickness is reduced, the reaction to a uniformly distributed surface force coincides with the reaction to the volume force (λ2 = 1/8). It is also worth noticing that a 10−6 volume load (e.g. for aluminium alloys) leads to a situation in which the displacement is also slight (Fig. 1.12) and is equal

62


u ·10

-4

2

u ·10

16 14 12 10 8 6 4 2

-4

8 6 4 2

x3

x3

x1

x1 3

u ·10

1

u ·10

x3

-4

-4

x2

10 8 6 4 2

x1 x1

18 16 14 12 10 8 6 4 2

x3 2

u ·10

-4

10 8 6 4 2

x2

x2

x3

x3 u3·10-4 8 6 4 2

x2

x3

Figure 1.8. The representation of the displacement distribution for problems 1 and 2 on planes x2 = 0.5 and x1 = 0.5 (the relative thickness is 0.5).

to 10−7 ÷ 10−10 . It means that at small loads the volume force may be neglected, which has actually been carried out in work [595].


1

u ·10

2

u ·10

-2

63

-2

4 3 2 1

5 4 3 2 1

x3

x3

x1

x1

3

u ·10

-2

4 3 2 1

x3

x2

x1

x1 x3

1

6 5 4 3 2 1

u ·10

2

u ·10

-2

-2

4 3 2 1

x2

x2

x3

x3 3

u ·10

x2

-2

4 3 2 1

x3

Figure 1.9. The representation of the displacement distribution for problems 1 and 2 on planes x2 = 0.5 and x1 = 0.5 (the relative thickness is 0.02).

64


1

u ·10

-3

18 16 14 12 10 8 6 4 2

2

u ·10

6 4 2

x3

x3

3

u ·10

-3

x1

x1

-3

8 6 4 2

x2

x3

x1

x1

x3

2

u ·10 1

u ·10

-3

6 4 2

18 16 14 12 10 8 6 4 2

x3

-3

x3

x1

3

u ·10 x1

x3

-3

8 6 4 2

xi

Figure 1.10. The quantity of the relative thickness for corresponding curves at the transverse loading.


3

65

x2

-2

u ·10 5 4 3 2 1

x1

x3

x1 x3 1

3

u ·10

5 4 3 2 1

4 3 2 1

x3

-4

u ·10

-4

x1

x3

x1

Figure 1.11. Solution to problems 1 and 2 for plates of the relative thicknesses 0.1 and 0.02.

Table 1.10. Comparison of the results obtained by different authors for a uniformly loaded plate (the value of displacement u3 in the plate’s centre is presented). Plate’s thickness Kirchhoff-Love’s [242] hypothesis [261] The three-dimensional theory Relative [242] error [261]

λ2 = 1/50 0.063 0.077 0.068 8% 11%

λ2 = 1/10 0.089 0.085 0.095 5.5% 10.5%

λ1 = λ 2 = λ 3 = 1 – – 0.021 – –

66


u ·10

-8

36 34 32 30 28 26 24 22 20 18 16 14 12 10 8 6 4 2

x1

x3

2

u ·10

-8

8 6 4 2

x3

x3

x1 1

u ·10

x1 3

u ·10

-3

8 6 4 2

-8

28 26 24 22 20 18 16 14 12 10 8 6 4 2

x1

x2

x3

x1

2

u ·10 8 6 4 2

x3

x1

x2

Figure 1.12. The quantity of displacements for a volumetric loading of 10−6 .

2. To illustrate the possibility of a three-dimensional modelling we are going to investigate solutions within the two-dimensional theories. Problem 3. Figure 1.13 presents displacements on plane x2 = 0.25 that have been obtained as a result of the operation of different surface forces f11 = 0.01 and f41 = 0.001 (problem 1).

1.4 Linear Problems in the Theory of Plates in 3D Space a)

1

u ·10

b)

-2

10 9 8 7 6 5 4 3 2 1

2

u ·10

67

-4

6 4 2

x3

x1

3

u ·10

x3

-4

6 5 4 3 2 1

x1

2

u ·10

-3

4 3 2 1

x3

x3

x1

x1 x2 3

u ·10

-3

6 5 4 3 2 1

x3

x1 x3

3

u ·10

-4

18 16 14 12 10 8 6 4 2

x1

x3

x1

Figure 1.13. The representation of displacements caused during the operation of surface forces of various values ( f11 = 0.01 and f41 = 0.001): a) an increase of displacements at a decrease of the plate’s thickness; b and c) the asymmetrical distribution of displacements u2 and u3 in respect to each other.

In this case, similarly to the previous point, reducing the plate’s thickness causes a tendency to increase displacements (Fig. 1.13a). Due to the significant influence of the different values of the surface forces that result from reducing the plate’s thickness, the displacements u2 and u3 are more asymmetrical to each other (Fig. 1.13b and 1.13c).

68


Problem 4. Let us analyse a cube-shaped plate with fixed walls ∂Ω3 , ∂Ω6 (E3i = 1, E6i = 1, Di3 = 0, Di6 = 0, i = 1, ..., 3), and free walls ∂Ω1 , ∂Ω2 , ∂Ω4 , ∂Ω5 (E1i = E4i = E2i = E5i = 0, Di1 = Di4 == Di2 = Di5 = 1, i = 1, ..., 3), which is subjected to volume force P3 = 0.01 (the surface forces do not occur). Figure 1.14 presents displacement distribution on plane x3 = 0.5 for: a) fixed edges, b) free edges. Fixing of the edges decreases the displacements’ values within the entire investigated plane. It can be observed particularly near the edge and it corresponds to a physical interpretation of the phenomenon.

a)

b) 1

u ·10

1

u ·10

-3

4 3 2 1

3 2 1

x2

x2

x1

2

u ·10

x2

3

z3

-3 2

u ·10

4 3 2 1

-3

x2 x3

x3

x1

6 5 4 3 2 1

x1

-3

4 3 2 1

x2

x1

u ·10

x2

-3

3

u ·10

x2

-3

6 5 4 3 2 1

x3

Figure 1.14. The displacement distribution (planes x3 = 0.5 and x1 = 0.25) in case of: a) fixed edges, b) free edges.


69

Problem 5. Let us investigate a cube-shaped plate with free walls ∂Ω4 , ∂Ω5 , ∂Ω6 (E4i = E5i = E6i = 0, Di4 = Di5 = Di6 = 1, i = 1, ..., 3) and fixed walls ∂Ω1 , ∂Ω2 , ∂Ω3 (E1i = E3i = E2i = 1, Di1 = Di3 = Di2 = 0, i = 1, ..., 3), which is subjected to surface force f41 = 0.001, f42 = f43 = 0 (the volume force is not taken into account). Figure 1.15 presents displacement distribution on plane x3 = 7/8 for: a) fixed edges, b) free edges. Figure 1.16a illustrates displacement distribution in case of a)

1

u ·10

14 12 10 8 6 4 2

-4

x2 2

u ·10

x2

x1

3

u ·10

8 6 4 2

x1

x3

-4

x2

x1

-4

6 4 2

x2

x1

1

b)

16 14 12 10 8 6 4

u ·10

2

u ·10

x2

x1

3

8 6 4 2

x2

-4

u ·10

-4

8 6 4 2

x2

x1

-4

x1

Figure 1.15. The displacement distribution (plane x3 = 7/8) in case of: a) fixed edges, b) free edges.

70


fixed corner (0, 0, 0) (which means that ui (0, 0, 0) = 0), whereas Figure 1.16b illustrates an increased surface of fixing. The surface of fixing and the way the edge is fixed significantly influence the plate’s stress-strain state.

a) 1

u ·10

10 8 6 4 2

2

u ·10

-4

-4

8 6 4 2

x2 x2

x1

x1

3

u ·10

-4

x2

8 6 4 2

x2

b)

x1

1

u ·10

x3

-4

10 8 6 4 2

3

u ·10

x2

x1

x2 2

u ·10

x2

x1

-4

10 8 6 4 2

x1

-4

8 6 4 2

x1

Figure 1.16. The displacement distribution for problem 5 in case of: a) fixed corner (0, 0, 0), b) an increased fixing surface.


71

Problem 6. Thin plate λ1 = 1/50, λ2 = 1, λ3 = 50 with free walls ∂Ω1 , ∂Ω4 (E1i = E4i = 0, Di1 = Di4 = 1, i = 1, ..., 3) and fixed walls ∂Ω2 , ∂Ω3 , ∂O5 , ∂O6 (E2i = E3i = E5i = E6i = 1, Di2 =Di3 =Di5 = Di6 = 0, i = 1, ..., 3) are affected by shearing forces f12 = −0.001, f42 = 0.001, f11 = f13 = f41 = f43 = 0 and volume forces, while other surface forces are equal to zero. The displacement distribution presented in Figure 1.17a on plane x3 = 0.5 and in Figure 1.17b on plane x1 = 0.5 is considerably different from the displacement

a)

b)

u1·10-4

1

u ·10

4

-4

1

3 2

u2

1

u2

u1

2

-5

u ·10

u1

-1

2

-2

1

-3

x2

-4

-1

x1

-5 2

u ·10

-5 3

u ·10

2 1

x2 -1

-5

1

x1

x2

x1

u3·10-5

x2

1

x2

x1

x3

x1

Figure 1.17. The displacement distribution for problem 6 on plane: a) x3 = 0.5, b) x1 = 0.5.

72


distribution obtained at the transverse load (Fig. 1.10). The reaction to the shearing force is much weaker than the reaction to the transverse load of the same value. 3. Our further considerations are going to focus on a heated fixing-ring system, which can be static or dynamic. In this case we are going to present solutions to a number of static problems concerning the ring with fixing for which the following modelling conditions are set on the surface: ui (1, x2 , 7/8) = 0, ui (0, x2 , 7/8) = 0, ui (x1 , 1, 7/8) = 0, ui (x1 , 0, 7/8) = 0, i = 1, ..., 3 – will be called a ring with fixing in the first row of points, ui (1, x2 , 3/4) = 0, ui (0, x2 , 3/4) = 0, ui (x1 , 1, 3/4) = 0, ui (x1 , 0, 3/4) = 0, i = 1, ..., 3 – will be called a ring with fixing in the second row of points, ui (1, x2 , 5/8) = 0, ui (0, x2 , 5/8) = 0, ui (x1 , 1, 5/8) = 0, ui (x1 , 0, 5/8) = 0, i = 1, ..., 3 – will be called a ring with fixing in the third row of points.

Problem 7. We investigate a cubic plate with fixed walls ∂Ω3 , ∂Ω6 (E3i = E6i = 1, Di3 = Di6 = 0, i = 1, ..., 3) and free walls ∂Ω1 , ∂Ω2 , ∂Ω4 , ∂Ω5 (E ij = 0, Dij = 1, i = 1, ..., 3, j = 1, 2, 4, 5) subjected to volume force P3 = 0.01, while the surface forces are neglected. A ring is placed on the surface of the cube. The distribution of displacement is presented on planes a) x3 = 0.25 and b) x2 = 0.5 for the ring in the first row of points (Fig. 1.18), in the second row of points (Fig. 1.19) and in the third row of points (Fig. 1.20). The influence of the fixingring system’s reaction is clearly visible on plane x2 = 0.5. The stress-strain graphs are recognisably different, especially for displacements u2 and u3 . The displacement distribution of the plate with the ring coincides with the displacement distribution of the plate without the ring on plate x3 = 0.25 (Fig. 1.14b). However, the values of displacements u1 , u2 , u3 tends to decrease as the ring changes its position along the axis towards plane x1 = 0.25, which also corresponds to the physics of the phenomenon.


a)

1

u ·10

73

-3 2

u ·10

4 3 2 1

x2

x1

3

u ·10

-3

4 3 2 1

x2

x1

-3

6 5 4 3 2

x2

1

x3

x2

x1

x1

b) 1

4 3 2 1

u ·10

2

u ·10

-3

-3

4 3 2 1

x3

x1

3

u ·10

x3

x1

-3

5 4 3 2 1

x3

x1

Figure 1.18. The displacement distribution for the ring in the first row of points on plane: a) x3 = 0.25, b) x2 = 0.5.

74


a)

1

3 2 1

u ·10

-3

2

u ·10

x2

x1 x2 3

u ·10

4 3 2 1

x1

-3

5 4 3 2 1

x2

x2

x1 x3

b)

-3

4 3 2 1

1

u ·10

x1

-3

2

u ·10

x3

4 3 2 1

x1 X3 3

u ·10

-3

x1

-3

6 5 4 3 2 1

x3

x1

Figure 1.19. The displacement distribution for the ring in the second row of points on plane: a) x3 = 0.25, b) x2 = 0.5.


a)

75

u ·10 1

-3

3 2 1

u ·10 2

x2

-3

x1

3 2 1

x2

x1

u ·10 3

-3

6 5 4 3 2 1

x2

x2

x1 x3

x1

b) u ·10 1

4 3 2 1

-3

u ·10 2

x2

-3

3 2 1

x1 x2

x1

u ·10 3

-3

5 4 3 2 1

x2

x1

Figure 1.20. The displacement distribution for the ring in the third row of points on plane: a) x3 = 0.25, b) x2 = 0.5.

76


Problem 8. Figures 1.21 and 1.22 illustrate the solution of the previously formulated problem 7. This time with surface force f41 = 0.01 taken into account and the volume forces neglected.

a)

1

u ·10

-3

6 5 4 3 2 2

u ·10 x2

-3

3 2 1

x3 x2 3

x3

-3

u ·10 6

5 4 3 2 1

x2

x2

x3 x3

b)

1

u ·10

x1

-3

2

u ·10

6 5 4 3 2

-3

3 2 1

x2

x2

x3

x3

3

u ·10

-3

4 3 2 1

x2

x3

Figure 1.21. The displacement distribution taking the surface force into account on plane x1 = 0: a) without a ring, b) with a ring in the third row of points.

1.4 Linear Problems in the Theory of Plates in 3D Space a)

1

u ·10

77

-3

7 6 5 4 3

2

x ·10

-3

3 2 1

x3 x3

x1

x1

3

u ·10

-3

x2

5 4 3 2 1

x3

x3

x1

x1

b)

1

u ·10

-3

2

u ·10

6 5 4 3 2 1

-3

4 3 2 1

x3

x3

x1

x1

3

u ·10

-3

4 3 2 1

x3

x1

Figure 1.22. The displacement distribution taking the surface force into account on plane x2 = 0.5: a) without a ring, b) with a ring in the third row of points.

78


Figure 1.21 presents a displacement graph on plane x1 = 0 and Figure 1.22 – on plane x2 = 0.5: a) without a ring, b) with a ring in the third row of points. The ring in case b) affects the character of the stress-strain state in all the investigated planes, in which the values of displacements change in the ring’s proximity. It is clearly visible in the place where the ring is located (Fig. 1.21b). Problem 9. Figure 1.23 shows plane x2 = 7/8 with the graphs of displacement distribution of the plate with a ring (problem 8) affected by a pair of surface forces f11 = 0.001 and f41 = 0.001. The solutions to problems 8 and 9 are used in further investigations to compare the results.

1

u ·10

-4

9 8 7 6 5 4 3 2

x2

x1

2

u ·10

-4

4 3 2 1

x2

x1

x2 3

u ·10

x1

x3

x2

-4

5 4 3 2 1

x1

Figure 1.23. The displacement graphs for a plate with a ring affected by a pair of surface forces.


79

Comparison of the stress-strain state of a cube with a ring affected by surface and volume forces may lead to a conclusion that the ring’s presence significantly influences the values of displacements. Moreover, for less accurate calculations, the surface force may be equivalently replaced with the volume force, and vice versa (the relative calculation error decreases along with reducing the plate’s thickness). Three-dimensional problems make it possible to investigate stress-strain states for such types of fixings and external forces that cannot be precisely modelled by means of the two-dimensional approach and the use of the two-dimensional theory may result in major errors in consequence. 1.4.2 Dynamic problems The process of solving dynamic problems allows investigating various kinds of changes that occur in time in the examined object. The following section presents the results of a number of problems regarding the influence of mechanical loads on plates in its dynamic aspect (the hyperbolic system of equations (1.23) does not include only the temperature gradient). Runge-Kutta’s method is applied to find the solution. A plate of variable relative thickness subjected to surface and volume forces is the object of the following analysis. Problem 10. The object of investigation is a plate with free edges ∂Ω3 , ∂Ω6 (E3i = E6i = 0, Di3 = Di6 = 1, i = 1, ..., 3) and jointedly supported walls ∂Ω1 , ∂Ω2 , ∂Ω4 , ∂Ω5 (Ei1 = 0, D1i = 1, Ei2 = Ei3 = 1, D2i = D3i = 0, i = 1, 2, 4, 5) affected by volume force P3 = 0.01 that causes free vibration. The initial conditions (1.33) are assumed as zero. Figure 1.24 illustrates movement of points (0.5, 0.5, 0.5) with respect to axis x3 , that is displacement u3 . For the sake of experiment, the plate’s thickness has been reduced. This has led to an increase of the vibration’s frequency and amplitude. Changing the dimensionless coefficient at the inertial terms also causes changes of the plate’s vibrations’ amplitude and frequency. As the inertia coefficient ℵ decreases, the vibrations’ amplitude and frequency increase. The broken line in Figure 1.24 marks the solution of the analysed problem (the relative thickness is 0.1) with coefficient ℵ = 0.01 and the full line marks the solution with coefficient ℵ = 1. The explanation for it may be a decrease of the characteristic mechanical time τ M = l3 /C, which is used to express coefficient ℵ = τ2M , c = ((λ + 2µ)/ρ)0.5 .

80


u (0.5;0.5;0.5) 0.2 x2 =0.01

0.015 x3

0.02

x1

0.1

0.05

0

4

8

12

16

20

32

0

2

4

-0.05

-0.1

Figure 1.24. The change of displacement u3 of a plate with free edges and jointedly supported walls.

Problem 11. The object of investigation is a cube-shaped plate with a fixing-ring system in the second row of points, affected by a pair of surface forces f11 = 0.001 and f41 = 0.001. Walls ∂Ω1 , ∂Ω2 , ∂Ω4 , ∂Ω5 are free (E1i = E2i = E4i = E5i = 0, Di1 = Di2 = Di4 = Di5 = 1, i = 1, ..., 3), and walls ∂Ω3 , ∂Ω6 are fixed (E3i = E6i = 1, Di3 = Di6 = 0, i = 1, ..., 3). Null boundary conditions are assumed for calculations. The problem has been solved using a static formulation (problem 9). On plane x2 = 7/8 in Figures 1.25 and 1.26 one can observe changes that occur in the plate’s stress-strain state at different time instants: Fig. 1.25a τ = 0.005, Fig. 1.25b τ = 0.01, Fig. 1.26a τ = 0.0125 and Fig. 1.26b τ = 0.0175. At the initial stage of the analysed process the changes of the ring’s behaviour become clearly visible. Besides, in course of time the stress-strain state approaches the static solution (Fig. 1.26 and Fig. 1.23). The investigated time intervals are small (dτ = 0.1 ÷ 0.5).


a)

1

12 10 8 6 4 2

u ·10

81

-5

2

u ·10

-5

4 3 2 1

x3

x1 3

u ·10

x3

x1

-5

8 6 4 2

x3

x1

b) 1

u ·10

-5

10 8 6 4 2

2

u ·10

x1

x3

3

u ·10

x3

4 3 2 1

x1

-5

x2

6 4 2

x3

-5

x1

x1

Figure 1.25. The displacement distribution of a cube-shaped plate with a fixing-ring system in the second row of points at various time instants: a) τ = 0.005, b) τ = 0.01.

82


a)

1

u ·10

-5

12 10 8 6

2

u ·10

-5

6 4 2

x3

x1

x3 3

u ·10

x1

-5

3 2 1

x3

x2

x1 x3

b)

1

u ·10

x1

-5

10 8 6 4 2

2

u ·10

-5

6 4 2

x3

x1

x3 3

u ·10

x1

-5

8 6 4 2

x3

x1

Figure 1.26. The displacement distribution of a cube-shaped plate with a fixing-ring system in the second row of points at various time instants: a) τ = 0.0125, b) τ = 0.0175.

1.4.3 Non-stationary temperature field One of the most important practical subjects of investigation is the influence of the occurrence of heat sources and heat fluxes inside an elastic body. Both continuous and discrete systems can become heat sources. This section analyses the temperature distribution modelled by general heat conductivity equation (1.24) (the term that corresponds to dilatation is assumed as equal to zero). A parabolic equation is solved by means of Runge-Kutta’s method. A three-dimensional body’s temperature field with and without a heat source is the object of the following investigation.


83

Problem 12. The temperature distribution in a cubicoid-shaped field is the object of research. The initial-boundary conditions are following: thermal insulation (Ci = 1, Ai = 0, Bi = 0, i = 2, 3) on the entire surface of the plate except for the heat impact (B1 = 1, T 1 = 133); in the centre of wall ∂Ω1 (x1 = 1, 0 ≤ x2 ≤ 1, 0.5 ≤ x3 ≤ 1, θ|τ0 = 0).

a)

µ·10 ·10

x2

3 2 1

x1

x3

x1

x3

x1

x3

µ·10 ·10 3 2 1

x2

x1

µ·10 ·10

x3

x1

µ·10 ·10

4 3 2 1

x2

x1

x1

4 3 2 1

3 2 1

x2

Figure 1.27. The temperature distribution in a cubicoid (problem 12) at time instant: a) τ = 0.002, b) τ = 0.04, c) τ = 0.07.

84


b)

µ·10 ·10 3 2 1

x2 x1

x3

x3

x1

µ·10 ·10 4

x1

x3

3 2 1

x1

x3

µ·10 ·10 3 2 1

x2

x1

3 2 1

µ·10 ·10 5 4 3 2 1

x1

µ·10 ·10

x2

x1

x2

Figure 1.27. cont.

Figure 1.27a illustrates the temperature distribution at τ = 0.002, Fig. 1.27b at τ = 0.04, and Fig. 1.27c at τ = 0.07. The heat transfer is easily observable from the side of the active wall and such is the temperature increase within the entire plate in course of time.


85

µ·10

c)

4 3 2 1

x2

X1

X3

x3

x1

X1

X3

µ·10 7 6 5 4 3 2 1

X1

X3

X2

X1

µ·10 6 5 4

µ·10

2 1

X1

X2

3 2 1

X1

X2

Figure 1.27. cont.

86


Problem 13. Let us assume that the plate’s centre (0.5, 0.5, 0.5) contains a heat source of power P4 = 300 dimensionless units, which corresponds to the power of 150 cal/(cm3 · s) = 636 W/cm3 .

µ·10 ·10 µ·10 ·10

6 5 4 3 2 1

x2 5 4 3 2 1

x3

x1

x3 x1

x1

x2

µ·10 ·10 µ·10 ·10

7 6 5 4 3 2 1

x3

x2

7 6 5 4 3 2

x1

x3

x1 x1

x2

µ·10 ·10 7

x2

6 5 4 3 2 1

x3

µ·10 ·10 8 7 6 5 4 3

x1

x1

x3

1

x1

x2

Figure 1.28. The temperature distribution in a cubicoid (problem 13) at time instant: a) τ = 0.002, b) τ = 0.04, c) τ = 0.07.

The previous problem’s heat conductivity equation (1.24) did not take the investigated field’s inner heat sources into account. Figure 1.28a–c illustrates the temperature distribution at the same time instants as in problem 12 (on planes


87

x2 = x3 = 0.5). The influence of heat sources at initial time instants is characterised by an increase of temperature in the place where the heat source is located (τ = 0.002). Then, the source’s reaction starts to exceed the external heat impact’s reaction (τ = 0.1). The highest temperature can be observed in the proximity of the active wall. Solving the non-stationary three-dimensional equation of heat conductivity demonstrates the temperature field’s non-linearity along the plate’s thickness. 1.4.4 Comparison of Solutions – non-isothermal Processes in Static Problems The results of the investigations of the plate’s stress-strain state have so far been obtained without taking temperature into account. The same temperature has been assumed in every point of the field, and even the occurrence of deformations has never changed it. In reality, the plate’s deformations cause changes of temperature and the temperature’s change causes deformations of the plate due to the material’s thermal expansion. Introducing heat loading into the processes analysed in Sections 3.1 and 3.2 will additionally expand and complicate them. The mathematical model that describes those processes is different from the previously discussed models because it requires solving the system of equations (1.23) (with or without the inertial terms) and the equation of heat conductivity (1.24) (stationary or non-stationary) simultaneously. The following thermal conditions have been assumed for all of the problems presented in this section: temperature distribution (Bi = 0, Ai = 1, Ci = 0) on the 0 = 0 and constant temperature distribution within the plate’s entire surface at T i+m 0 ring, T i+m = 1. Problem 14. The object of analysis is a cube-shaped plate with fixed walls ∂Ω3 , ∂Ω6 , edges (E3i = E6i = 1, Di3 = Di6 = 0, i = 1, ..., 3) and free walls ∂Ω1 , ∂Ω2 , ∂Ω4 , ∂Ω5 (E1i = E2i = E4i = E5i = 0, Di1 = Di2 = Di4 = Di5 = 1, i = 1, ..., 3). In such a case a stationary problem is solved, which means that in system (1.23), (1.24) time-dependent inertial terms and temperature derivatives are neglected. Next, the problems of stationary heat conductivity and static elasticity are successively solved. The iterative upper relaxation method is used to solve the problem. Figure 1.29 illustrates the influence of thermal excitations on plane x2 = 7/8 (Fig. 1.29a), the influence of displacement u1 on plane x2 = 0.5 (Fig. 1.29b–c), the influence of displacement u2 (Fig. 1.29d) and the influence of displacement u3 (Fig. 1.29e). The volume and surface forces are neglected. The fixing-thermal ring system is placed in the first row (see sections 3.1, 3.3 of this chapter). Let us pay attention to the fact that the scale of the graphs illustrating displacement u3 is two times smaller than that of the graphs showing displacements u1 and u2 . The changes of displacement u3 are most visible because the ring is placed perpendicularly to axis x3 . Besides, plane x2 = 7/8, which is parallel to axis x3 , is investigated and it is the plane and along which displacement u3 occurs. Due to the fact that wall ∂Ω3 and the

88


-1 µ·10 ·10

8 7 6 5 4 3 2 1

1

u ·10-3

x3

x1

x3

x3

x3

4 3 2 1 -1 x 1 -2 -3 -4 -5

1

u ·10-3

5 4 3 2 1

x2 -1

-2 -3

x2

x2

x1

x3

x1

x3

x1

2

u ·10-3 4 3 2 1

x3 -1

-2 -3 -4 -5

x1

3

u ·10-3

-6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46 -48

Figure 1.29. The analysis of a cube-shaped plate (problem 14): a) the influence of thermal excitations on plane x2 = 7/8, b–c) the influence of displacement u1 on plane x2 = 0.5, d) the influence of displacement u2 , e) the influence of displacement u3 .


89

ring are rigidly fixed in the first row of the points in the plane’s proximity, displacement u3 that disappears during stresses may freely spread out only in the opposite direction to axis x3 (which Fig. 1.29d illustrates). The influence of the fixing-ring system located on the ring’s plane (x1 , x2 , 7/8) makes displacements occur only inside the plate. The temperature’s influence is symmetrical to plane x1 = 0.5, and as a result, the plate’s reaction to plane x1 = 0.5 is also symmetrical (Fig. 1.29e and 1.29c). The stress-strain state changes and if the influence of mechanical forces is taken into account next to the temperature, then the displacement distribution starts to become asymmetrical.

Problem 15. The plate with parameters described in problem 14 is furthermore affected by surface forces f11 = f41 = 0.001, f12 = f13 = f41 = f43 = 0 and volume forces P3 = 4 · 10−7 (Fig. 1.30). Symmetry of the distribution of displacements u1 and u2 disappears because the surface forces affect the parallel walls ∂Ω1 and ∂Ω4 in one direction along axis x1 . Thus mechanical loads strengthen the temperature-caused deformations along axis x1 and lessen them along axis x2 , which is particularly visible in case of displacement u1 . The value of the volume force is low – about 10−5 .

Problem 16. The plate investigated in problem 15 is encircled with a ring in the second row of points (Fig. 1.31). In this case, the first row of points becomes free and displacement distribution changes. While being heated, the body expands in various directions starting from the fixing-ring. Displacement u3 reveals the most relevant changes since its positive values appear between the ring and the rigid fixing of wall ∂Ω3 . Problem 17. In this very interesting case the thermal ring and the fixing-ring are located next to each other along axis x3 . The thermal ring is in the first row and the fixing-ring is in the second row. Figures 1.32a (without surface forces) and 1.32b (with surface forces f11 = f41 = 0.001, f12 = f13 = f41 = f43 = 0) show plane x2 = 7/8 on which the fixing-ring ”dashes” the expansion incited by the thermal impact. It is worth noticing that the graphic representations of all displacement distributions in this problem are in identical scale.

90

1 Three–Dimensional Problems of Theory of Plates in Temperature Field -1 µ·10 ·10

1

u ·10-3

9 8 7 6 5 4 3 2 1

x3

-1 x 1 -2 -3 -4 -5

x1

x3

4 3 2 1

2

x2

x3

5 4 3 2 1

x2

x1

x3

x1

x3 -1

-2 -3 -4 -5

x3

u ·10-3

x1

x1 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40

Figure 1.30. The analysis of a cube-shaped plate as in problem 14, additionally considering surface and volume forces.

1.4 Linear Problems in the Theory of Plates in 3D Space -1 µ·10 ·10

91

1

u ·10-3

10 9 8 7 6 5 4 3 2 1

4 3 2

x3

x3

-1 x1 -2 -3 -4 -5

x1 3

u ·10-3

x2

x2

x1

x3

26 24 22 20 18 16 14 12 10 8 6 4 2

x3

x1

2

u ·10-3 6 5 4 3 2 1

x3

x3

x1

x1 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -24 -26 -28 -40 -42 -44 -46 -48 -50

Figure 1.31. The analysis of a cube-shaped plate as in problem 15 with an additional ring in the second row of points.

92


u1·10-2 9 6 5 4 3 2 1

9 8 7 6 5 4 3 2 1

x3

x3

x1

x1

1

8 u ·10 7 6 5 4 3 2 1

-2

x2

x3

x2

x2

x1

x3

x1

x3

3

u ·10-2

x3

4 3 2 1

x1

7 6 5 4 3 2 1

x3

u2·10-2

x1

3

u ·10-2 4 3 2 1

x2

x1

Figure 1.32. The distribution of the changes of temperature and displacements in case when the thermal ring and the fixing-ring are place one after another along axis x3 : a) without the surface forces, b) with the surface forces.


93

1

u ·10-2

9 8 7 6 5 4 3 2

-1 -2 -3 -4 -5 -6 -7 -8 -9

x3

x3

x1

x2

6 5 4 3 2 1

x2

x1

3

x3

x1

x3

u ·10-2

x1

5 4 3 2 1

u2·10-2 6 5 4 3 2 1

x3

x1

x2

3

u ·10-2 x1

3 2 1

x3

Figure 1.32. cont.

x1

94


Problem 18. In this problem the boundary temperature conditions are slightly altered and plate (∂θ/∂n = 0) is thermally insulated by heating it with the fixingthermal ring in the first row of points.

-1

·10 9 8 7 6 5 4 3 2 1

x1

x3

-u ·10-2 1

4 3 2 1

x3

x2

-1 x1 -2 -3 -4 -5

x2

-u ·10-2 3

x3

-u ·10-2

x1

x1

x3

3 2 1

2

4 3 2 1

x3 -1

x3

x1

x1

-2 -3 -4

Figure 1.33. The displacement fields’ distribution in a thermally insulated plate and heated by a fixing-thermal ring in the first row of points.

Figure 1.33 illustrates non-uniform and asymmetrical displacement fields’ distributions at symmetrical temperature distribution inside the plate, which are caused by surface forces ( f11 = f41 = 0.001, f12 = f13 = f42 = f43 = 0) operating in one direction.


95

Problem 19. A plate with the same type of fixing as previously, described according to a quasistatic problem is heated by the ring in the first row of points (the boundary conditions of problem 15). Null initial conditions are assumed.

1

-1 µ·10 ·10

8 7 6 5 4 3 2 1

u ·10-3

5 4 3 2 1

x3

-1 x1 -2 -3 -4 -5 -6

x1

x3

u1·10-3

x2

4 3 2 1

x3 x3 -1 -2 -3

6 5 4 3 2 1 -1 x3 -2 -3 -4 -5 -5

x2

x1

x1

x3

x1 3

u ·10-3 6 4 2 2

u ·10-3

x3

x1

-2 -4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40

x1

Figure 1.34. A plate with the same type of fixing as in problem 15 for null initial conditions at: a) τ = 0.005, b) τ = 0.0175, c) τ = 0.03625, d) τ = 0.1125.

96


-1 µ·10 ·10

u ·10-3

9 8 7 6 5 4 3 2 1

x1

x2

u3·10-3

x2

x1

x3

x3

8 6 4 2

x1

x3

2

u ·10-3 5 4 3 2 1

x3

-1 x 1 -2 -3 -4 -5 -6 -7

x3

x3

5 4 3 2 1

x1

Figure 1.34. cont.

x1 -4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44


97

1

u ·10-3

9 8 7 6 5 4 3 2 1

3 2 1

x2

x2

-1x 1 -2 -3 -4

x1

3

x2

x1

x3

u ·10-3

x2

x3

8 6 4 2

x1

x3

2

6 5 4 3 2 1

x3

u ·10-3

x1

Figure 1.34. cont.

x1 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42 -44 -46

98


1

u ·10-3

9 8 7 6 5 4 3 2 1

x3

x2

x2

x1

x3 2

x3

x1

x1

x3

6 5 4 3 2 1

4 3 2 1

u ·10

x3

3

u ·10-3

8 6 4 2

x1 x3

-3

x1

-4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40

x1

Figure 1.34. cont.

In such a case ∂2 ui /∂τ2 = 0 is assumed in the system of equations (1.23)–(1.33), and the value of dilatation in the heat conductivity equation is assumed to be zero. Two methods are combined during calculations: the heat conductivity equation is solved with the use of Runge-Kutta’s method, and the upper relaxation method is applied to construct a graph illustrating the stress-strain state at every time step. Figure 1.34a–d presents the distribution of the temperature and displacements changes in time.


99

Problem 20. Problem 16 is investigated again, however its formulation has been changed. Figure 1.35a–d presents the plate’s stress-strain state distribution in time.

-1 µ·10 ·10

9 8 7 6

1

u ·10-3

5 4 3 2 1

x1

x3

x1

x3

x2

x3

3

u ·10-3 24 22 20 18 16 14 12 10 8 6 4 2

x2

x1

x3

x1

2

u ·10-3 4 3 2 1

x3

4 3 2

x3 x1

-4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40

x1

Figure 1.35. The distribution of a plate’s stress-strain state at: a) τ = 0.005, b) τ = 0.0175, c) τ = 0.03625, d) τ = 0.625.

100


1

u ·10-3

9 8 7 6 5 4 3 2 1

4 3 2 1 -1x1 -2 -3

x3

x1

x3

3

u ·10-3 x2

x2

x1

x3

28 26 24 22 20 18 16 14 12 10 8 6 4 2

x3

x1

2

u ·10-3 4 3 2 1

x3 -1

-2 -3 -4

x3

x1

x1 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42

Figure 1.35. cont.

The following conclusion may be drawn for problems 19 and 20: in course of time the stress-strain state “grows stable” and approaches the distribution obtained with taking static problems into account. For instance: at τ = 0.0365 the results


101

-1 µ·10 ·10

1

u ·10-3

9 8 7 6 5 4 3 2 1

3 2 1 -1 x 1 -2 -3

x3

x1

x3

3

u ·10-3 x2

20

x2

18 16 14

x3

x1

x3

12 10

x1

8 6 4

2

u ·10-3 4 3 2 1

x3

-2 -3 -4

2

x1

x3

x1

-6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42

Figure 1.35. cont.

coincide with convergence between 1% and 12% (Fig. 1.30 and 1.35a, and Fig. 1.31 and 1.35d for problem 20). Further calculations in time do not bring any changes to

102

1 Three–Dimensional Problems of Theory of Plates in Temperature Field u ·10-1 u1·10-3 9 8 7 6 5 4 3 2 1

5 4 3 2 1

x3

-1x 1 -2 -3

x1

x3

u ·10-3 3

x2

x3

22

x2

x1

x3

20 18 16 14

x1

12 10 8 6 4 2

u2·10-3 6 5 4 3 2 1

x3 -1

-2

x3

x1

-2 x 1 -4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42

Figure 1.35. cont.

the stress-strain state. Therefore, observation of the moments of direct influence of the heat impact (dτ = 0.01) becomes most interesting in this case. The thermal ring


103

with heat distribution constant in time is one of the forms of the heat impact [101]. The problems concerning stationary thermal boundary conditions may be analysed within quasistatic considerations. Problem 21. Let us analyse two last cases of solving problems with a movable thermal ring. Initially, the fixing-ring is moved towards the second row of points. The thermal ring is moved towards the first row and then to the second row of points at τ = 0.0125. 1

u ·10-3

6 5 4 3 2 1

x3

9 8 7 6 5 4 3 2 1

x3

-1 x1 -2 -3 -4 -5 -6 -7 -8 -9

-1 µ·10 ·10

x2

x2

x1 2

u ·10-2

7 6 5 4 3 2 1

x3

x3

x1

x3

3

u ·10-2 x1

x3

x1

3 2 1

x1

Figure 1.36. The distribution of a plate’s stress-strain state with a movable thermal ring (the static approach) at: a) τ = 0.05, b) τ = 0.01, c) τ = 0.0125, d) τ = 0.01625, e) τ = 0.0175, f) τ = 0.03.

104


u ·10-2 6 5 4 3 2 1 -1 x1 -2 -3 -4 -5 -6 -7 -8

x3

-1 µ·10 ·10

8 7 6 5 4 3 2 1

x3

x2

x1 x3 2

u ·10-2

x3

7 6 5 4 3 2 1

x2

x1

x1

x3

3

u ·10-2 x1

x3

3 2 1

x1

Figure 1.36. cont.

Figure 1.36a–f illustrates all changes of the stress-strain state related to the temperature field’s change. Until the thermal ring is displaced at τ = 0.0125, from the very first time instant the graph of the stress-strain state approaches the solution obtained in the analogous problem 17 (Fig. 1.32b) in the static approach. As the thermal ring is displaced, the stress-strain state also approaches the solution obtained in the analogous problem (Fig. 1.31), but at τ = 0.0125 (the moment when the thermal ring is displaced) the representation of the temperature distribution does not resemble the previously investigated ones. A similar conclusion may be drawn for the displacement distribution. It is also important to notice that the scale of the graphs at time instant τ = 0.0125 suddenly changes by one order of magnitude. With the quasistatic approach it has been possible to discover the stress-strain state - an achievement impossible to make with the use of any static methods. By integrating the system of equations (1.23) at the same time with the equation of heat


105

1

u ·10-3

-1 µ·10 ·10

5 4 3 2 1

10 9 8 7 6 5 4 3 2 1

-1 x 1 -2 -3 -4

x3

x3

x1 3

u ·10-3 x2

x1

x3

20 18 16 14 12 10 8 6 4 2

x2

x3

x1

3

u ·10-3 4 3 2 1 -1 x3 -2 -3 -4

x3

x1

-4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40

x1

Figure 1.36. cont.

conductivity (1.24), assuming null dilatation and using Runge-Kutta’s method only, we obtain a complete dynamic problem of the theory of thermoelasticity.

106

1 Three–Dimensional Problems of Theory of Plates in Temperature Field u ·10-1

u ·10-3 1

9 8 7 6

4 3 2 1

5 4 3 2 1

x3

x3

-1 x 1 -2 -3 -4

x1

x2

x3

x3

x1

u ·10-3 2

4 3 2 1

x3 -1

-2 -3 -4

3

26 24 22 20 18 16 14 12 10 8 6 4 2

x2

x1

u ·10-3

x3

x1

x1

Figure 1.36. cont.

-4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40


107

-1 µ·10 ·10

11 10 9 8 7 6 5 4 3 2 1

1

u ·10-3 4 3 2 1 -1 x 1 -2 -3

x3

x3

x1 3

u ·10-3 x2

x3

24 22 20 18 16 14 12 10 8 6 4 2

x2

x1

x3

x1

2

u ·10-3 4 3 2 1

x3 -1

-2 -3 -4 -5

x3

x1

Figure 1.36. cont.

x1 -4 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42

108


1

u ·10-3

9 8 7 6 5 4 3 2 1

3 2 1

x3

x1

-1 -2 -3

x3

x1 3

u ·10-3

x2

x2

x1

x3

x3

x1

24 22 20 18 16 14 12 10 8 6 4 2

x3 2

u ·10

-3

4 3 2 1

x3

x1

Figure 1.36. cont.

x1 -6 -8 -10 -12 -14 -16 -18 -20 -22 -24 -26 -28 -30 -32 -34 -36 -38 -40 -42


109

Problem 22. A problem formulated in the same way as problem 21 with a movable thermal ring is solved according to the dynamic approach with supplementary mechanical initial conditions (1.33). An assumption that qS1 , qS2 , q, s = 1, ..., 3 are equal to zero has been made in this problem. Figure 1.37a–f presents temperature and displacement graphs plotted on plane x2 = 7/8 at the same time instant as in case of the quasistatic analysis. Comparison of those results with the results obtained in an analogous quasistatic problem (Fig. 1.36a–f) leads to a conclusion that the temperature distributions coincide, the scale of the displacement graphs at τ = 0.0125 changes by one order of magnitude, but the similarity of the stress-strain state remains unchanged (it does not “keep up” with assuming the form it has in statics and quasistatics). That is why the time of calculation is prolonged until τ = 0.055 (Fig. 1.37f), which still seems to be not enough to reach stability of the desired stress-strain state portrait in statics (Fig. 1.31). Time-variable heat impacts belong to the most interesting phenomena investigated in the theory of thermoelasticity. They should be analysed with methods which are most related to real dynamic processes. The principle condition for obtaining reliable results is taking dynamic effects into account. Finally, it is important to notice that the parts of the thermal ring that are perpendicular to plane x2 = 7/8, as well as that part of the ring that belongs to wall ∂Ω2 , parallel to plane x2 = 7/8, reveal their influence upon the investigated plane x2 = 7/8. 1.4.5 Inner heat sources It seems trivial to claim that including heat sources into investigations makes the mathematical relations concerning solving problems more complicated. In fact, the right sides of heat conductivity equations solved through the analytical approach equal zero in very few problems of the theory of elasticity. Mostly axially symmetrical or two-dimensional problems belong to this group. The methodology presented in this chapter enables scientists to solve problems of the theory of thermoelasticity with stationary or movable heat sources or heat fluxes in three-dimensional plates. If the influence of heat sources causes a change of temperature, then the plate undergoes deformations even without mechanical excitations. Dynamic problem (1.23)– (1.33) must be formulated and Runge-Kutta’s method must be applied to find the solution in this case.

110


u ·10-2

5 4 3 2 1

-1 µ·10 ·10

x3

9 8 7 6 5 4 3 2 1

x2

x1

x3

x2

x2

X1 x1

x3

3

u ·10

x1 2

4 3 2 1

-1 x1 -2 -3 -4 -5

u ·10

-2

5 3 2 1

-3

x3

x3

x1

x1

1

u ·10-2

4 3 2 1

x3

-1 x1 -2 -3 -4 -5

µ·10 ·10-1 10 9 8 7 6 5 4 3 2 1

x2

x1

x3

x3

x1

x2

x1

x3

3

u ·10-2 3 2 1

2

u ·10-2

x3

x1

3 2 1

x3

x1

Figure 1.37. The distribution of a plate’s stress-strain state with a movable thermal ring (the dynamic approach) at: a) τ = 0.005, b) τ = 0.01, c) τ = 0.0125, d) τ = 0.01625, e) τ = 0.03, f) τ = 0.055.

1.4 Linear Problems in the Theory of Plates in 3D Space µ·10 ·10

-1

1

u ·10-2

9 8 7 6 5 4 3 2 1

3 2 1

x3

x3

x1

x2

x1

x2

x1

x3

x1

x3 3

u ·10-2

3 2 1

2

4 3 2 1

u ·10-2

x1

x3

x3

x1

1

u ·10-3

µ·10 ·10

3 2 1

-1

x3

9 8 7 6 5 4 3 2 1

-1 x 1 -2

x2

x3

x3

x1

2

x3

x1

x1

x3

3

u ·10-3 14 12 10 8 6 4 2

u ·10-3 3 2 1

x2

x3

x1

Figure 1.37. cont.

-2 x1 -4 -6 -8 -10 -12 -14 -16 -18

111

112


u ·10-3 3 2 1 -1 µ·10 ·10

x3

-1 x1 -2

9 8 7 6 5 4 3 2 1

x2

x2

x1

x3

x3

x1

x3

x1 3

u ·10-3

14 12 10 8 6 4 2

2

u ·10-3 3 2 1

x3

x3

-2 x1 -4 -6 -8 -10 -12 -14 -16

x1

1

u ·10-3 µ·10 ·10

3 2 1

-1

-1 x1 -2 -3

x3

9 8 7 6 5 4 3 2 1

x2

x3

x3

x1

u ·10-3

-1 x3 -2 -3 -4

x1

x1

x3 3

u ·10

-3

16 14 12 10 8 6 4 2

2

4 3 2 1

x2

x3

x1

Figure 1.37. cont.

-2 x1 -4 -6 -8 -10 -12 -14


113

Problem 23. Let us investigate a cube-shaped plate with free walls ∂Ω1 , ∂Ω2 , ∂Ω4 , ∂Ω5 (E1i = E2i = E4i = E5i = 0, Di1 = Di2 = Di4 = Di5 = 1, i = 1, ..., 3) and rigidly fixed walls ∂Ω3 , ∂Ω6 (E3i = E6i = 1, Di3 = Di6 = 0, i = 1, ..., 3). The temperature distribution on the entire surface of the plate is equal to (Ci = Bi = 0, Ai = 1). On plane (x3 = 0.5, 0.125 ≤ x1 ≤ 0.5, 0.125 ≤ x2 ≤ 0.875) inside the plate there is a uniformly distributed heat source of dimensionless unite power.

-4 µ·10 ·10

6 5 4 3 2 1

1

u ·10-3

x3

5 4 3 2 1

x1

x2

x3

x3

x2

x1

x3

x1

x1

3

u ·10-3 5 4 3 2 1

u2·10-3 4 3 2 1

x3

x3

x1

x1

Figure 1.38. The distribution of the changes of temperature and displacements (plane x2 = 0.5) for the plate investigated in problem 23.

Figure 1.38 presents temperature and displacement distribution on plane x2 = 0.5. The displacement graphs clearly illustrate the plate’s expansion in all directions away from the heat source: symmetrically towards x3 and asymmetrically towards x1 and x2 . The behaviour of displacement u1 appears to be the most interesting: the layers located closer to the heat source’s plane move towards x1 , whereas the further layers move in the opposite direction to axis x1 . As a result, an increase of

114


temperature causes the strongest stresses between those layers. Further numerical calculations are possible when the heat source’s surface and volume are reduced or enlarged. Problem 24. A volumetric heat source move inside a plate (the source’s form is presented in Figure 1.29 on plane x2 = 7/8, but the problem is not described analytically). The following conditions are assumed on the plate’s boundary: null temperature distribution, surface forces f11 = f41 = 0.001, f12 = f13 = f42 = f43 = 0, and volume force P3 = 4 · 10−6 , P1 = P2 = 0. The boundary conditions are the same as in problem 16. Figure 1.39a–c illustrates temperature distribution and displacement graphs. Initially (τ = 0.00875), the plate reveals stronger reaction to mechanical excitations. Figures 1.39b and 1.39c display asymmetrical range of displacement u1 , caused by the surface forces. However, already at τ = 0.03 all three displacements increase by one order of magnitude due to the reaction to heat impact which absorbs the reaction incited by surface forces. Symmetry of the heat source makes the displacements symmetrical to axis (1/2, 1/2, 1/3) grow in importance. 1.4.6 Deformation and Temperature Heat impacts have been investigated in works [98, 101, 200, 350]. The results obtained by Danilovskaya [163, 164] have become classic and hence often referred to. She has pointed out a possibility of occurrence of compressive and tensile stresses in an infinitely elastic space during heating, which can be observed only through a complete dynamic formulation of the problem. This section contains an analysis of Danilovskaya’s problem for a three-dimensional plate. System of equations (1.23)–(1.33) has been solved with the use of Runge-Kutta’s method Problem 25. The object of investigation is a plate with fixed walls ∂Ω2 , ∂Ω3 , ∂Ω5 , ∂Ω6 , and two opposite walls Ω1 and Ω4 . Wall Ω1 is affected by heat impact ∂θ/∂x1 = 13.3. The other walls are thermally insulated (∂θ/∂n = 0). The initial conditions are null (1.33). Comparison of the results described in work [98] with the results obtained by the authors of this work (Fig. 1.40) allows investigating the changes of the normal stress in point (0.5, 0.5, 0.5) (point c in the graph) in time. Figure 1.40 presents illustration of the results of calculations made for several sections of the plate in points: a) (1/8, x2 , x3 ), b) (1/4, x2 , x3 ), c) (1/2, x2 , x3 ), d) (7/8, x2 , x3 ). A configuration of curves a, b, c, d may be calculated with the use of analytical relations [98]. The compressive stresses increase until τ = x1 /a, in which a = 1/(1 + ν)(1 − 2ν) is the coefficient at the main derivatives in the movement equations, x1 is the first coordinate of the examined object. The values of the characteristic time interval obtained with this formula coincide with the obtained approximate values τa ≈ 0.065, τb ≈ 0.13, τc ≈ 0.26, td ≈ 0.42.

1.4 Linear Problems in the Theory of Plates in 3D Space µ·10 ·10

5 4 3 2 1

115

-1

1

u ·10-5

x3

6 5 4 3 2 1

x1

x2

x2 x3

x1

x3 2

u ·10

4 3 2 1

u3·10-5

x1

x3

x1

-5

5 4 3 2 1

x3

x3

x1

x1

1

u ·10-5 5 4 3 2 1

-1 µ·10 ·10

4 3 2 1

x1

x3 x1

x3

x2

x3 2

u ·10-5

x2

x1

x3

3

u ·10-5 4 3 2 1

x1

x1

x3

4 3 2 1

x3

x1

Figure 1.39. The temperature distribution and the displacement graphs in case of a volumetric (movable) heat source at: a) τ = 0.00875, b) τ = 0.01525, c) τ = 0.03.

116


U ·10-4

5 4 3 2 1

4 3 2 1

µ·10-1 X3

X1

X3

X1

x2

x2

x1

x3

3

U ·10-4 4 3 2 1

x1

x3

2

U ·10-4

X3

X1

3 2 1

X3

X1

Figure 1.39. cont. ¾x x

1 1

a

x2

20

b c

x1

x3 0.1

0.2

t

t

0.4

0.5

-20 d

x2

x3

x1

Figure 1.40. The stress distribution for the plate investigated in problem 25.


117

The scatter of values between the maximum compressive stress and the maximum tensile stress for every section is identical, S is the coefficient at the temperature derivatives of the movement equations and T 0 = 1. The results presented in Figure 1.40 coincide with the classic ones until the time instant in which the stress for the half-infinite space starts to converge towards zero. In case of a thick plate, the normal stresses regain their compressive nature in course of time, yet it is more intense than at the beginning. It can be best observed for section c). The full curve illustrates the results of calculations without taking deformation and temperature fields’ coupling (β = 0) into account. The broken curve represents the results including the coupling, which means that the generalized heat conductivity equation (1.24) has been included with all terms in the calculations. Coupling coefficient β = 0.03 [595] for aluminium alloys may reach arbitrarily high values. The coupling effect can be obtained at τ = 0.3125 – 10%. Problem 26. In order to investigate this problem, the distribution of normal stress along the axis of plane (x1 , 0.5, 0.5) has been presented in several time instants in Figure 1.41a. The same refers to Figure 1.41b, although the considerations include the second boundary conditions: identical heat impact, but the plate’s walls ∂Ωi , i = 2, ..., 6 have null temperature distribution. The stresses in this problem have very large quantities and they undergo significant changes. The thermal reaction of all six walls and adequately quicker stabilisation of the temperature distribution can serve as an explanation of this phenomenon. The coupling effect occurs at τ = 0.3125 – 12%. Problem 27. The plate examined in the previous problem is affected by simultaneous thermal and mechanical impacts, which cause a displacement on wall ∂Ω1 , which consequently approaches a stationary state of the following value: ⎧ ⎪ ⎨ 0, τ ≤ τ∗ 1 u x =l = u0 f (τ) , f (τ) = ⎪ ⎩ 1, τ > τ∗ , 1 1 where: u0 = 0.01, and f (τ) is Heaviside’s function. The full line in Figure 1.42a illustrates the normal stress distribution. Figure 1.42b presents the static stress on axis (x1 , 0.5, 0.5). Problem 28. This case refers to the following numerical experiment concerning the plate described in problem 27: sudden thermal impact has been neglected, i.e. the analysis takes only the mechanical impact into account.

118


¾x x

1 1

0.5 t=0.03 0

-0.5

-1

x1

t=0.1 t=0.2

t=0.3

-1.5 t=0.5

b)

¾x x

1 1

0.5

0

-0.5

0.03 0.1

x1

0.2 0.4 0.3

-1

Figure 1.41. The normal stress distribution along the axis of plane (x1 , 0.5, 0.5): a) the first thermal boundary conditions, b) the second thermal boundary conditions.


119

a)

¾x x

1 1

x2

3

2

x1

x3

1

0 t=0.01

x1

t=0.03

-1

-2 x2 t=0.0125

-3

-4

x3

x1

t=0.05

-5

-6

-7

t=0.03

Figure 1.42. The stress distribution in time for the plate investigated in problem 26 regarding additional heat and mechanical impacts: a) the normal stress distribution, b) stresses tangential to axis (x1 , 0.5, 0.5).

120


b)

¾x x

1 1

1

0

t=0.001 t=0.03

-1 t=0.05 -2

-3

t=0.0125

-4

-5

-6

-7

t=0.03 t=0.04

Figure 1.42. cont.

x1


121

The results of calculation have been illustrated with the broken line in Figure 1.43a–b. The dash-dot curve illustrates the results obtained taking the temperature and the deformation fields’ couplings into account. The difference in case of this effect does not exceed 1% at τ = 0.05. Figure 1.43a presents the behaviour of the normal stress in the plate’s centre within a time interval. The mechanical impact weakens the influence of the thermal impact. The impact wave becomes smooth (Fig. 1.43a–b). a)

¾x x

1 1

2 1 0.02

0.03

0.04

0.05

-1

t

-2

b)

¾x x

1 1

0.5 0.05 0.1 0.15 0.2 0.25 0.3 0.35

0.4 0.45 0.5 0.55 0.6 0.65

-0.5

t

-1

Figure 1.43. The stress distribution in time for the plate investigated in problem 27: a) the normal stress in the plate’s centre, b) the tangential stresses.

Problem 29. A mechanical impact of value u0 = −0.01 increases the influence of the heat impact investigated in this section. Figure 1.44a–d illustrates the development of the heat impact process on plane x2 = 0.5. Displacements u1 and u3 in the investigated plane are characterised by significant changes. Displacements u2 and u3 are symmetrical to each other. In order to observe this, it is necessary to present the distribution of displacement u3 on plane x2 = 0.5, and displacement u2 on plane x3 = 0.5 (Fig. 1.44d). Similarly to the previously investigated thermoelasticity problems, a strong influence of the mechanical impact on the stress-strain state reveals first (Fig. 1.44a–b), and then thermal effects occur (Fig. 1.44c–d). In Figure 1.45, the distribution of the normal

122


1

-u ·10-2 4 3 2

0 µ·10 ·10

x3

4 3 2 1

x3

x1

x2

x1

x2

2

u ·10-3 4 3 2 1

x3

x1

x1

x3 3

u ·10-3 4 3 2 1

x1

x3

x3

x1

1

-u ·10-2 4 3 2 1

b)

6 5 4 3 2 1

x3

µ

x1

3

u ·10-3

x3

x1

x2

x3

x2

x1

x3

x3

4 3 2 1

x1

x1

Figure 1.44. The development of the heat impact process on plane x2 = 0.5 at: a) τ = 0.0125, b) τ = 0.1725, c) τ = 0.3325, d) τ = 0.5.

1.4 Linear Problems in the Theory of Plates in 3D Space µ 9 8 7 6 5 4 3 2 1

c)

1

u ·10-2 5 4 3 2 1

x2

x2

x1

x3

3

x1

x3

u ·10-3

4 3 2 1

x1

x3

u2·10-3 5 4

-1 x3 -2 -3 -4

3 2 1

x1

2

u ·10-2 d)

x1

x3

x1

x3

4 3 2 1

1

u ·10-2

-1 x3 -2 -3 -4 -5 -6 -7

x1

4 3 2

x3

x1

µ 10 9 8 7 6 5 4 3 2 1

x2

x3

x1

x1

x3

3

x3

x1

u2·10-3 3 2 1

x3

x2

x3

x1

Figure 1.44. cont.

u ·10-3 4 3 2 1 -1 x1 -2 -3 -4 -5 -6 -7

123

124

1 Three–Dimensional Problems of Theory of Plates in Temperature Field ¾x x

1 1

0.5

t=0.2

0.375 0.25 0.125 0 -0.125

t=0.1 t=0.4 X1 t=0.01 t=0.03 t=0.05

-0.25 -0.375 0.5

t=0.5

-0.625 -0.75 -0.875 -1 -1.125 -1.25

Figure 1.45. The distribution of the normal stress along axis (x1 , 0.5, 0.5) for the plate investigated in problem 29.

stress along axis (x1 , 0.5, 0.5) is demonstrated by a thermoelastic wave generated by the active wall towards a parallel wall. The reciprocal coupling effect in the plate’s centre equals 1% at τ = 0.5. Problem 30. A complete three-dimensional description makes it possible to spot and observe heat impacts operating along a part of the plate’s wall. Let us investigate the influence of the heat impact formulated according to problem 29, yet limited to only a half of wall ∂Ω1 (x1 = 1, 0 ≤ x1 ≤ 1, 0.5 ≤ x3 ≤ 1), while the field of the mechanical impact’s operation is limited to a quarter of wall ∂Ω1 (0.5 ≤ x1 ≤ 1). Figure 1.46a–c presents the plate’s reaction to impacts at various time instants on plane x2 = 0.5. Reducing the active impact surface leads to substantial reduction of the plate’s stresses. It is particularly visible in Figure 1.43b, in which the full line


125

1

-u ·10-2

a)

3 2 1

x3

x1

µ 2 1

x3

2

u ·10-3

x1

x3

4 3 2 1 -1 x1

2

u ·10-3

x3

4 3 2 1

x2

-1 x1

x3

b)

x2

x1

x1

x3

µ·10 ·10 4 3 2 1

1

-u ·10-2 3 2 1

x1

x3 x3 x2

x1

x2 3

u ·10-3 x1

x3

x3

4 3 2 1

x1

2

u ·10-3 4 3 2 1 -1 x3 -2 -3 -4 -5

x3

x1

-1 x1 -2 -3 -4 -5 -6 -7 -8 -9 -10 -111 -1 -12 -13 -14 -15 -16 -17 -18 -19 -20 -21

Figure 1.46. The plate’s reaction to impacts (plane x2 = 0.5) at various time instants: a) τ = 0.1725, b) τ = 0.3325, c) τ = 0.5.

126

1 Three–Dimensional Problems of Theory of Plates in Temperature Field c) 3

u ·10-3 µ·10

4 4 3 2 1

x3

2

x3 -2

x1

x1

-4

x2

x2

-6 -8 -10

x1

x3

x3 1

x1

-12

-2

-14

-u ·10

-16

3 2 1

x3

-18

x1

-20 2

u ·10-3

-22

4 3 2 1

x3

-24 -26

x1

-28 -30 -32 -34

Figure 1.46. The plate’s reaction to impacts (plane x2 = 0.5) at various time instants: a) τ = 0.1725, b) τ = 0.3325, c) τ = 0.5.

marks the stress in the centre of the plate affected by thermal or mechanical impacts on the entire surface of wall ∂Ω1 , while the broken line marks the same impacts along a part of wall ∂Ω1 . The coupling effect for problem 30 is weaker than it is in problems 25 and 26. Its value at τ = 0.4 equals 1%. The thermoelastic wave (Fig. 1.47) is clearly visible and in course of time its front moves away from the active wall towards the opposite wall. The graph of the normal stress with respect to axis (x1 , 0.5, 0.5) takes the general value for every investigated case. Problem 31. In order to analyse the occurrence of the reciprocal coupling effect caused by inner heat conductivity processes incited by a heat source, the following investigation is conducted: a cube-shaped plate with fixed walls ∂Ω2 , ∂Ω3 , ∂Ω5 , ∂Ω6 and free walls ∂Ω1 , ∂Ω4 is affected by a uniformly distributed heat source (P4 = 1) located on plane x1 = 0.5; the temperature distribution equals zero (T i0 = 0). The energy source is non-stationary, it operates within time interval dτ = 0 ÷ 0.4 dimensionless units and then it is switched off.


127

¾x x

1 1

t=0.2 t=0.3

0.05

0

0.1

t=0.4 t=0.1 t=0.01 x t=0.02 1 t=0.03

0.2 0.3

-0.25

t=0.5 t=0.6

0.4

-0.5 0.5 0.6

Figure 1.47. The graph of the normal stress in respect to axis (x1 , 0.5, 0.5).

The graph in Figure 1.48a presents normal stress σ x1 x1 on straight line (x1 , 0.5, 0.5) at various instants of time. The full curves represent the source’s operation at τ = 0.4. The source creates a tension-related heat wave that spreads from the source to the plate’s edge, then it reflects and returns to the source’s centre. At the same time the heat source is switched off and the compressive stresses turn into tensile. Initially, until τ = 0.5, the distribution of the normal stress is not identical to the one at τ = 0.4 and its tension configuration is displaced into the positive field. However, already at τ = 0.6 the normal stresses have positive values only – it is characteristic for displacements that approach the initial state of equilibrium. The strongest stresses occur on the edges of the investigated plate. Figure 1.48b illustrates changes of temperature in the plate’s centre within a time interval. Figure 1.48c illustrates changes of the normal stress both in the plate’s centre and in point (1/8, 1/2, 1/2) within a time interval. The broken curve marks the results obtained taking the reciprocal coupling into account. The quantities of the reciprocal influence exerted on the investigated time interval differ from each other. Table 1.11 presents the relative percentage difference in calculations in which the reciprocal coupling has and has not been taken into account. At the initial stage of heating, the coupling does not occur until the heat regime become stable (unchangeable). After the heat source is switched off, the meaning of the coupling

128


¾x x

t=0.72

1 1

0.5

t=0.7

0.4 0.3 0.2 0.1 t=0.5 x1 t=0.21

0 -0.1

t=0.1

-0.2 -0.3

t=0.5 t=0.2

-0.4

t=0.42 t=0.4

x2

x2

x1 x3

x1 x3

b) µ·10-1 0.1 0.1

µ

¾x x

1 1

0.2

0.3

0.4

0.5

0.6

t

0%

0%

0%

5%

0.7%

0.3%

0.1%

0%

0%

0.3%

6%

0.7%

0.4%

0.05%

Figure 1.48. The analyses of the occurrence of the reciprocal coupling effect (problem 31): a) the graph of the normal stress on the straight (x1 , 0.5, 0.5) at various time instants, b) the temperature change in the plate’s centre in time, c) the changes of the normal stress in time in the plate’s centre and in point (1/8, 1/2, 1/2).

1.5 3D Physically Non-Linear Problems

129

c)

¾x x

1 1

0.8

0.6

(1/8,1/2,1/2)

0.4

0.2

0.1

0.2

0.3

(1/2,1/2,1/2)

0.5

t

0.6

-0.2

-0.4

Figure 1.48. cont.

Table 1.11. The relative difference in calculations with and without considering the reciprocal coupling. τ θ σ x1 x1

0.1 0% 0%

0.2 0% 0%

0.3 0% 0.3%

0.4 5% 6%

0.5 0.7% 0.7%

0.6 0.3% 0.4%

0.7 0.1% 0.05%

effect significantly increases since it reaches 6% during dτ = 0.01 for the normal stress and 5% for the temperature. In course of time the heat regime becomes stable again and the effect disappears. The process described here confirms the fact that couplings must be investigated and taken into account especially at clearly distinguished non-stationary heat regimes.

1.5 3D Physically Non-Linear Problems Relations between couplings and deformations are non-linear if the body’s load exceeds the yield point. Variety of different kinds of elastoplastic bodies leads to creating a number of various mathematical models that describe physical states of the

130


investigated bodies. There are several theories of plasticity that describe physical relations and precisely explain physical phenomena that characterise elastoplastic deformations. Therefore it is extremely important to answer the question about usefulness and applicability of those theories. Surprisingly enough, the problem may be solved only for one class of loading, namely the simple loading. The theory of small elastoplastic deformations [89, 171, 357] proves to be handy while describing phenomena of that kind. Its simplicity of relations and possibility to utilise general solving methods have made the theory of small elastoplastic deformations widely applied. This chapter uses small elastoplastic deformations to present a method of solving coupled problems of heat and mechanical loading of a three-dimensional plate. 1.5.1 Differential equations and difference approximation Let us investigate the stress-strain state at an initial time instant of an isotropic homogeneous body (plate) that is non-uniformly heated and affected by surface and volume forces. The principal conditions of the applied theory have been presented in work [360]. The theory is based on the hypothesis of proportionality of the stress tensor and the strain tensor’s components:

(1.116) ei j − δi j e = ψ σi j − δi j σ , ψ=

3εi . 2σi

(1.117)

For small elastoplastic deformations the following function relation is effective: σi = f (εi ),

(1.118)

which is similar to the relations that between stresses and deformations during tension σ = f (e) , (1.119) i.e.: the volumetric strain is elastic σ = ke. Moreover, during uniformly distributed heating we obtain σ e = + αT θ. k The components of the strain tensor take the form of superposition of the strain (p) components’ elasticity e(e) i j and plasticity ei j , i.e.: (p)

ei j = e(e) i j + ei j ,

(1.120)

where: e(e) i j is determined according to Duhamel-Neuman law, considering the fact that the coefficients depend on temperature. Equation (1.116) is true, if the body’s loading is simple. In this case, without taking temperature relations into account,


131

the external loadings increase until they are proportionally imposed to one general parameter [171]: (1.121) Pi = λ (τ) P0i , i 0i = λ(τ) f s+m , i, s = 1...3, m = 0, 3 . f s+m

(1.122)

During non-uniform heating in thermal quasistatic regime (slow gradual increase of temperature) the temperature loadings change in proportion to parameter [357] 0 = χ(τ)T 00 . (1.123) T i+m The external forces increase together with the increase of the temperature (exact to reliable multipliers) to such extent that the stress intensity increases along with the increase of the strain intensity (1.118), i.e.: Pi = BT γ , where: A=

0i f s+m

T 00γ

i f s+m = AT γ ,

,

B=

P0i T 00γ

,

(1.124)

(1.125)

and γ < 1 is a material constant. In case of a non-stationary heat regime, the simple stress occurs when 0 = T 00 eατ . T i+m

(1.126)

The external forces change according to formula (1.124), but the material characteristics do not depend on temperature. Otherwise, the stress is simple [145] if we investigate an incompressible material in plastic and elastic zones, where: e = 2τθ, ν = 0.5, k → ∞, E = 3G,

(1.127)

σi j = f (θ, εi ) .

(1.128)

The inner forces and the thermal deformations increase proportionally to some parameters and fulfil conditions (1.121), (1.122) and equation ατ (θ)θ = χ(τ)αT (T 0 )T 0 . In a general case, when all thermal and physical characteristics of the material depend on temperature, we obtain [198] T αT (T ) =

αT (ξ)dξ. T0

Function f (εi ) is determined from an experiment. Functional relations for f (εi ) that require meticulous calculations have been collected and analytically described in Krysko’s work [172]. Function f (θ, εi ) is also experimentally determined at various temperatures and presented in space εi , σi , T for certain surfaces (see [483]). In an elastic deformation field σi ≤ 1/ (3σ0 (T ))0.5 where σ0 (T ) is the flow interval, the surface turns non-linear and the equation takes the following form:

132


σi = 2µ (T ) ξi ,

(1.129)

where: µ(T ) denotes the time-dependable elasticity modulus. We may assume that at relatively high temperatures there a slight change of αT , which consequently leads to the assumption that αT = const. Therefore, a heat impact at such temperatures may result in plastic deformations, i.e.: E(θ), ν(θ). The equations of the theory of small elastoplastic deformations take the following form [356] σi j = 2 f (εi , T ) ei j +

3K − f (εi , T ) eδi j − 3KαT θδi j , 3

(1.130)

and the material characteristics depend on temperature: σi j = 2 f (εi ) ei j +

3K − f (εi ) eδi j − 3KαT θδi j , 3

(1.131)

or the material characteristics do not depend on temperature: + 3εi j εi j , εi = 2 eδi j . (1.132) 3 The system of differential equations of the non-linear theory of elasticity differs from system (1.23)–(1.33), the non-linearity of which results from additional expressions that characterise plastic deformations ε(p) i j [484], included within volumetric forces Pi : 2 j ∂2 ui ∂2 u j ∂2 uk ∂ u ∂2 uk (λ + 2µ) 2 + λ 2 + λ 2 + µ − + ∂xi ∂x j ∂xi ∂xk ∂xi ∂x j ∂xk εi j = ei j −

3Kατ

∂2 ui ∂θ − Pi − Pi∗ = ρ 2 , ∂xi ∂τ

3 ∂2 θ ∂ ∂ui 1 ∂θ 1 4 − P , −β = 2 ∂τ ∂x α ∂τ β ∂x i α α=1

(1.133) (1.134)

i ∂u 1 ∂u j G− + + 2Ψ ∂x j ∂xi k ∂u 1 ∂ui 1 ∂ui ∂u j ∂uk ∂ 1 ∂ui ∂ 2G − + G− . − ( + + + ∂xi Ψ ∂xi 3 ∂xi ∂x j ∂xk ∂xk 2Ψ ∂xi ∂xk (1.135) The boundary mechanical conditions are analogous to (1.25), (1.27), (1.29), (1.31) considering iP i i∗ = f s+m + f s+m , (1.136) f s+m where:

∂ P = ∂x j i∗


where:

133

1 ∂ui 1 ∂ui ∂u j ∂uk ni,s+m + = 2G − − + + Ψ ∂xi 3 ∂xi ∂x j ∂xk i k ∂u ∂u 1 ∂u j 1 ∂ui G− n j,s+m + G − nk,s+m , + + 2Ψ ∂x j ∂xi 2Ψ ∂xi ∂xk i∗ f s+m

i −→ j −→ k, m = 0, 3 . ←

←

(1.137)

The thermal boundary conditions and the initial conditions are identical with those presented in Section 1.1 (1.26), (1.28), (1.30), (1.32), (1.33). Difference approximation (1.133)–(1.137) is assumed analogously to approximation (1.38)–(1.45), excluding the approximations of the right sides of the equai i∗ are approximated in a usual way, while functions Pi∗ , f s+m tions. Functions Pi , f s+m take the following form: 1 1

Pi∗ ∼ 2Gi − yix¯i xi − yix¯i xi + y xj i x j + ykxi xk + Ψi 3 ⎛ ⎞ ⎜⎜⎜ Ψ ⎟ 1

j k ⎜⎝2G xi + x2i ⎟⎟⎟⎠ yi0,xi − yi0,xi + y0,x + y 0,xk + j 3 Ψi ⎞ ⎛ ⎜⎜ Ψ x j ⎟⎟ i 1 i j ⎟⎟⎠ y + y j + ⎜ ⎜ y x¯ j x j + y xi x j + ⎝G x j + Gi − 0,x 0,x j i 2Ψi 2Ψi2 ⎞ ⎛

⎜⎜ Ψ xk ⎟⎟⎟ i 1 k Gi − yix¯k xk + ykxk xi + ⎜⎝⎜G xk + (1.138) ⎠⎟ y0,xk + y0,xi , 2 2Ψi 2Ψi 1 1

j i∗ k yi0,xi − yi0,xi + y0,x ∼ 2Gi − + y f s+m 0,xk ni,s+m + j Ψi 3

% 1 $ i j k i Gi − y0,x j + y0,x + y + y n n . (1.139) j,s+m k,s+m 0,xi 0,xk i 2Ψi 1.5.2 Algorithm The non-linear system of differential equations (1.133)–(1.137) does not have an analytical solution, therefore numerical methods that are variants of the successive approximations method [64, 65, 172] are applied. Solving the theory of plasticity problems usually leads to solving a number of linear problems that can be interpreted as the theory of elasticity problems (the method of elastic solutions). This work focuses on the use of the variational method of elastic solutions – the method of additional loading. i∗ , that occur as a result of taking additional terms in If quantities Pi∗ and f s+m (1.133)–(1.137) into account, are assumed to be known, then the following system of reciprocally coupled theory of thermoelasticity is obtained. The investigated system is integrated in consideration of time by means of Runge-Kutta’s method. Initially,

134


s∗ at τ = τ0 P s∗ and f s+m = 0 are assumed. Next, functions u s are corrected at every time step with the use of the additional loading method, i.e.: successive approximations of the solution of a properly chosen deformation graph are made according to the following scheme: 1) at the first approximation, solution u s(0) = u s is assumed at previous time step; 2) the obtained displacements u s(0) make it possible to find (0) (0) deformation ε(0) i j (1.11) and strain intensity εi (1.132); 3) the stress intensity σi is determined according to the assumed diagram of deformations [259] or on basis of an appropriate thermomechanical surface [489] from charts or according to a graph, (0) s(0) , Ψ (1) the then Ψ (1) = 1.5ε(0) i /σi is calculated; 4) with known displacements u s∗ can be calculated according to relation (1.135), (1.137); 5) values of P s∗ and fi+m Seidel’s iterative method (1.84) is applied for system (1.133)–(1.137) with unknown right sides without inertial terms. The condition of completing the iterative process is the following estimation: y s(k) − y s(k−1) j j < ε1it , max 1≤ j≤M y s(k) j

where: ε1it is the set value, M = N1 N2 N3 . The result is approximation u s(1) = u s(k) (1) (1) (1) (2) and other values ε(1) = 1.5ε(1) i j , εi , σi , Ψ i /σi , etc. The solving process with the use of the additional loading method should be carried on until the difference between approximation u s(n) and the preceding approximation u s(n−1) is sufficiently small and equal to the set value ε2 . Thus obtained corrected solution u s(n) serves as basis for making the next time step with the use of Runge-Kutta’s method, which is next substituted into system (1.133)–(1.137). The combination of the successive approximations method and Runge-Kutta’s method has been applied in work [466] in order to investigate the dynamic behaviour of thin plates and shells (described by means of Kirchhoff’s hypothesis) affected by local mechanical loading. The difference between the approaches applied in work [466] and the above-described modification lies in using a combination of Seidel’s method, the additional loading method and Runge-Kutta’s method. 1.5.3 Estimation of Convergence The previous section presents the approach that utilises a combination of three numerical methods: solving a linear hyperbolic-parabolic system of differential equations by means of Runge-Kutta’s method; solving a non-linear elliptic system by means of the additional loading method; solving Lamé’s linear elliptic difference system by means of Seidel’s method. The notion of convergence of the general approximate approach to the solution of a non-linear problem should be applied to investigate the convergence of one method on basis of the solution obtained with the use of another method, starting from the inner process. Let us verify the conditions imposed on the coefficients and the right sides of the difference equations system that approximates the differential system (1.133)–(1.137). The convergence of the external overt process is provided by the conditions of the theorem quoted in the first chapter. The coefficients of system (1.133)–(1.137)


135

are assumed to be positive constants and they are bounded throughout the entire calculation process. Such a measure is made due to the fact that the plate is initially investigated as isotropic and homogeneous, therefore according to the additional loading method the following conditions are fulfilled (1.74), (1.75). Conditions (1.77) that refer to an elastoplastic case take the following form: sp s s∗ = fi+m + fi+m ∈ L2 (∂Ω × (τ0 , τ1 )). P sp = P s + P s∗ ∈ L2 (Qτ ) , fi+m

(1.140)

s ∈ They are fulfilled in consideration of the following: a) P s ∈ L2 (Qτ ), fi+m L2 (∂Ω × (τ0 , τ1 )) according to the condition of the theorem quoted in section 1 s∗ – and they do not change throughout the entire calculation process; b) P s∗ and fi+m superpositions of derivatives after the spatial variables of the solution obtained at the previous integration stage, i.e.: functions u s |τ=τ=0 ∈ W21,0 (Qτ ). The compactness of inclusion of W21 (Qτ ) within L2 (Ω) in [285] is also fulfilled for the solution field (a cubicoid), and besides W21,0 (Qτ ) is a Hilbert’s space, which means that it fulfils such an equality that if a, b ∈ W21,0 (Qτ ), then α(a + b) = αa + αb and P s∗ ∈ L2 (Qτ ), s∗ ∈ L2 (∂Ω × (τ01 , τk1 )) respectively. fi+m Boundaries on the right sides of the initial conditions (1.33) at the first time step depend on the plate’s stress at the initial time instant and they coincide with the conditions determined by formula (1.75). The solutions obtained with the use of the fulfilled approximations inside the field are assumed to be the initial conditions at the successive stages of calculations. For the convergence of the external process, the following conditions must be fulfilled:

q1s = u s | τ = τ0l+1 = u s(n) ∈ L2 (Ω) , ∂u s ∂u s τ=τ0 = ∈ W21 (Ω) , q2s = ∂τ ∂τ τ=τkl l+1 q2s ∈ W21 (Ω) ,

u s |τ=τl ∈ W21,0 (Qτ ) ,

(1.141)

q1s ∈ L2 (Ω) according to definition, which is conditioned by relation θ|τk1 = θ|τ0l+1 of

class W21,0 (Qτ ). According to the convergence theorem, which has been proven for solution u s(m) = u s of an elliptic difference scheme and published in works [429, 430], the inner process is stable and it converges towards the solution of Lamé’s differential system. In this case, the following conditions are additionally imposed on the surface and the volume forces: P1s = P s + P s∗ − 3KαT

∂θ ∈ L2 (Ω) , ∂x s

s s s∗ = f1i=m + f1i=m − 3KαT θ ∈ L2 (∂Ω), f1i=m

(1.142) (1.143)

which are consistent with relation (1.140) and concern the solution of an equation of heat conductivity θ in class W21,0 (Qτ ). As conditions (1.143) are fulfilled, the convergence of the external process towards the desired solution and at the same time the convergence of the inner process of solving Lamé’s difference system occur.

136


The proof of Seidel’s method’s convergence is presented in work [431]. It shows that self-coupling and the transition operator’s positivity are sufficient conditions for convergence. Work [431] also discusses certain properties of the operator that may be utilised in some problems of the theory of elasticity. It is important to mention that further stages of investigation utilise the simple loading, thus the external mechanical stresses and the inner volumetric stresses remain constant in time, i.e.: they fulfil conditions (1.121) and (1.122) when λ(τ) = 1. Moreover, the thermal conditions that describe insulation and heat impact occur on a part of one wall, which is provided by conditions (1.128) at αT (θ) = const. The investigated type of loading renders it possible to use the relations between the theory of small elastoplastic deformations and the solutions obtained by means of the additional loading method. The convergence of the additional loading method has been discussed in work [171], and the practical application of the presented iterative process has confirmed its good convergence. 1.5.4 Temperature and Deformation Coupling Investigation of reciprocally coupled temperature and deformation fields beyond the range of elasticity for three-dimensional plates has not yet been discussed in literature. The method, the numerical algorithm and the computation program presented in this work may be applied in order to conduct such an analysis in Section 3.6 for a physically non-linear plate. Problem 32. Let us analyse a thick plate described as a physically non-linear body, which is rigidly fixed along walls ∂Ω2 , ∂Ω3 , ∂Ω4 , ∂Ω5 , ∂Ω6 . Initial displacements u1 = 0.01, u2 = u3 = 0 are assumed on a quarter of wall ∂Ω1 , (x1 = 1, 0.5 ≤ x2 ≤ 1, 0.5 ≤ x3 ≤ 1) and the remaining part of wall ∂Ω1 is rigidly fixed. The problem is solved in the static approach, which means that inertial terms are neglected in system (1.133). The algorithm described in 4.2 is reduced due to excluding Runge-Kutta’s method: the additional loading method is combined with Seidel’s method. The following graph of aluminium strain is used: εi , σi = σ s 1 − exp − εs ε s = 0.98 · 10−3 , σ s = 3e s .

(1.144)

The results are presented in Figure 1.49a on plane x2 = 0.5. Table 1.12 shows the qualitative representation of the stress-strain state in both linear and non-linear approaches. Figure 1.49b presents the plastic deformation zone (the shaded area). Almost entire area is included within the plastic part of the deformation diagram, and only the angular area located away from the active area wall remains elastic. The plasticity field is asymmetrical, which corresponds to the asymmetrically operating loading.


137

2

u ·10-3

a)

4 3 2 1

1

-u ·10-2 5 4 3 2 1

x3

x3

x1

x1

3

u ·10-3

x2

3 2 1

x3

x1

x3

x1

b) x2

x1 x3

Figure 1.49. The analysis of the thick plate described in problem 32: a) the displacement distribution on plane x2 = 0.5, b) the representation of the plastic deformation zone (the shadowed area).

Problem 33. We shall determine the stress-strain state of a plate affected by a mechanical impact (see: problem 28) on a quarter of wall ∂Ω1 and a heat impact (∂θ/∂x1 = 13.3) imposed along a half of wall ∂Ω1 . Location of the coordinates system and the initial conditions (1.33) are identical to those in problem 31. Deformation graph (1.144) for an aluminium plate without considering the dependence on temperature is assumed. Such a step is fully justifiable, since at the

138


Table 1.12. Comparison of the values of displacements caused by heat impacts in linear and physically-non-linear approaches at τ = 0.3225.

u1 u2 u3

(1/4, 1/2, 1/2) (1/2, 1/2, 1/2) (3/4, 1/2, 1/2) Elastic Nonelastic Elastic Nonelastic Elastic Nonelastic -0.001610 -0.002562 -0.008710 -0.008879 -0.01510 -0.009788 -0.000001 -0.000036 -0.000193 -0.000465 -0.00165 -0.001215 -0.000892 -0.001508 -0.003450 -0.005207 -0.01630 -0.006735

initial deformation stages, which are not the object of our interest now, the temperature distribution is arbitrarily small and in any point it does not exceed the investigated field of certain temperature corresponding to aluminium flow range (T ≈ 300◦ ). Figure 1.50a–d presents the distribution of temperature and displacements u1 , u2 and u3 on plane x2 = 0.5 at various time instants τ = 0.0225, 0.1, 0.175, 0.3325. The results compared with the ones obtained in analogous problem 31 (Fig. 1.46c, 1.47, 1.48a) in the static approach reveal aluminium’s soft, non-elastic reaction. Some static approach impact-related phenomena can be clearly observed, the waves of thermoelastoplasticity do not disappear though (Fig. 1.51). Table 1.12 shows comparison of the displacement values for an elastic and a non-elastic problem in the central point (1/2, 1/2, 1/2), in a point close to the active wall ∂Ω1 (3/4, 1/2, 1/2), and in a point located away from the active wall (1/4, 1/2, 1/2). It can be observed that the points located further from the active wall and the central point are the places where the material’s reaction increases in case of the non-linear problem. Dislocations are much stronger in the points located in proximity of the active wall where the problem of elasticity is investigated. The distribution of the normal stress is non-uniform along the plate’s thickness. The displacement distribution and the dynamic behaviour in the less loaded area resemble the normal stress distribution for a linear problem (Fig. 1.48b–c). The stresses increase in the area close to the active wall, but the thermoplastic wave does not appear. The coupling effect in the investigated time interval has not been detected. The coupling coefficient β = 0.03. Problem 34. The boundary and the initial conditions in this problem are identical with those in problem 33. A steel plate’s reaction to mechanical and thermal (∂θ/∂x1 = 200) impacts are investigated. The deformation graph includes the temperature phenomena. Based on Table 1.13 [489], an instantaneous thermomechanical steel surface has been assumed. An increase of the heat impact does not bring any qualitative changes into the obtained results (Fig. 1.52a). Problem 35. The assumptions made in problem 34 remain the same. The heat impact’s force changes (∂θ/∂x1 = 850). The value of the temperature field suddenly increases (Fig. 1.52b, τ = 0.2), the values of displacements correspond to the physically non-linear problem’s values

1.5 3D Physically Non-Linear Problems a)

0 µ·10 ·10

4 3 2 1

1

-u ·10-2 4 3 2 1

x1

x3

x3 x2

139

x1

x2

2

u ·10-3 x1

x3

x3

4 3 2 1

x1

x3

x1

b)

1

-u ·10-2 µ·10 ·10

4 3 2 1

0

5 4 3 2 1

x1

x3 x1

x3

x2

x2

2

u ·10-3 3 2 1

x3

x1

x1

x3 3

u ·10 x1

x3

x3

-3

4 3 2 1

x1

Figure 1.50. The distribution of temperature and displacements u1 , u2 and u3 (plane x2 = 0.5) at various time instants: a) τ = 0.0225, b) τ = 0.1, c) τ = 0.175, d) τ = 0.3325.

and are placed within naturally physical intervals. In contrast to the investigated problems 33 and 34 with low temperatures in which displacement u1 is much bigger, there are also significant changes of the value of displacement u3 in the direction of both impacts’ operation in this problem. It is caused by an immense influence of

140


-u ·10-2 c)

4 3 2 1

0 µ·10 ·10

4 3 2 1

x3

x3

x1

x1

x2

x2

2

u ·10-3 4 3 2 1

x3

x1

x1

x3 3

u ·10-3

x3

x1

4 3 2 1

x3

x1

1

-u ·10-2 d)

3 2 1

x3

x1

0 µ·10 ·10

5 4 3 2 1

x1

x3

x3 x2

x3

x2

x1

x3

x1

Figure 1.50. cont.

3

u ·10-3 3 2 1

x1


141

¾x x

1 1

0.25

t=0.35 t=0.3

0.125

t=0.2

(1/4,1/2,1/2)

t=0.1 x1 t=0.01 t=0.02 t=0.03

-0.125 -0.25

¾x x

1 1

0.5 0.25 0.1

0.2

0.3

0.4

t

(1/2,1/2,1/2)

-0.25 -0.5

(3/4,1/2,1/2)

Figure 1.51. The impact-related phenomena in the elastic approach (problem 33). Table 1.13. Values of instantaneous thermomechanical surface for steel. εi ∗ 102 0 0.1 0.2 0.4 0.6 0.8 1.0 1.2 3728

0 0 1950 3900 5735 6295 6690 6945 7060 8820

100 0 1925 3850 5630 6175 6500 6690 6790 8550

200 0 1880 3760 5520 5935 6180 6340 6420 8160

σi , N/cm2 300 400 0 0 1800 1700 3600 3390 5175 5000 5650 5375 5925 5585 6075 5725 6150 5795 7700 7335

500 0 1560 3083 4580 4910 5020 5080 5112 6520

600 0 1400 2760 3885 4130 4295 4380 4410 5730

700 0 1200 2340 3225 3465 3625 3705 4410 4730

temperature which absorbs the reaction to mechanical operation and at the same time expands the plate’s material in all directions starting from the middle of wall ∂Ω1 (x1 = 1, 0 ≤ x2 ≤ 1, 0.5 ≤ x3 ≤ 1). Due to the fact that wall ∂Ω3 is rigidly fixed the expansion proceeds much easier opposite axis x3 inside the plate. The broken line in Figure 1.53 marks the normal stress distribution along axis (x1 , 0.5, 0.5) at

142

1 Three–Dimensional Problems of Theory of Plates in Temperature Field x2

µ·10 ·10

x2

0

4 3 2 1

x3

x1

x3

x1

x3

1

-u ·10

-2

3 2 1

x1

x3 2

u ·10

x1

-3

4 3 2 1

x3

3

u ·10-3

x1

3 2 1

x3

x1

1

-u ·10-2 3 2 1

x3

µ·10 ·10

x1

2

3 2 1

2

u ·10-2 4 3 2 1

x1

x3

x1

x3 3

u ·10

-3

4 3 2 1

x3

x2

x2

x1 x3

x1

x3

x1

Figure 1.52. The displacement distribution for problem 34 at an increased thermal impact: a) τ = 0.1, b) τ = 0.2.


143

¾x x

1 1

1.5

t=0.5

1.0

t=0.4

0.5

t=0.3 t=0.2 t=0.01 X1 t=0.02 t=0.03 t=0.1

0 t=0.2 -0.5

-1.0

-1.5

-2.0

-2.5

0

0.1

0.2

0.3

0.4

0.5

0.6

t

-1.0 (3/4,1/2,1/2)

-2.0

(1/2,1/2,1/2)

¾x x

1 1

Figure 1.53. The normal stress distribution along axis (x1 , 0.5, 0.5) for problem 34 (the broken line) and for problem 35 (the full line).

144


various time instants for problem 34, whereas the full line - for problem 35. In both cases the stresses are identical. The thermoelastoplasticity wave does not occur within the investigated time interval. This problem’s normal stress does not exceed the stress in problem 34, which corresponds to a sudden increase of the temperature field. The coupling effect (β = 0.01) is not observed. Based on the physical investigation equivalent, the assumption has been made of Duhamel-Neuman principle, that can be applied in case of small deformations and when the following condition is met: θ − T 0 1 . (1.145) T0 The additional loading method is also based on Duhamel-Neuman principle and on condition (1.145). Non-linear equation of heat conductivity should be applied at higher relative temperatures (problems 33 and 35). However the calculations can also be used as sort of an approximation of solution. Problem 36. The displacement distribution graphs change, if the impacts - not the temperature - are assumed to be the reason for the occurrence of the body’s plastic deformation. Work [356] discusses theses and describes experiments that confirm the fact that plastic deformations may occur caused by thermal impacts at arbitrarily low temperatures, which makes it possible to formulate a physically non-linear description of a model. The thermal impact in further experiments takes the form of ∂θ/∂x1 = 0.1. The boundary and the initial conditions are identical as those set in problems 34 and 35. The analytic form of the deformation diagram for aluminium (1.144) is used. The results of calculations are presented in Figure 1.54a–c on plane x2 = 0.5. Although the temperature distributions have the same configuration as in the previously investigated problems, they are significantly smaller. The displacement distribution at various time instants (Fig. 1.54a–c) indicates small displacements u3 in comparison to displacements u3 investigated in problems 34 and 35 at high temperature and at an arbitrarily rapid change in time (until τ = 0.1 value u3 decreases from 0 to −0.85 · 10−2 , and then increases up to −0.25 · 10−2 at τ = 0.43 in point (7/8, 0.5, 0.5)). The graph in Figure 1.55a illustrates the normal stress along axis (x1 , 0.5, 0.5) compared to the analogous graph plotted in the similar problem 33. A sudden increase of the stress can be explained by the fact that the most remote points of the field are neglected in this problem and the thermoelastoplastic wave starts to appear in point (3/4, 0.5, 0.5) at τ = 0.45. The comparison of figures illustrating the normal stress in the plate’s centre and in point (3/4, 0.5, 0.5) in time (Fig. 1.51, 1.55a–b) also shows difference between the investigated phenomena due to comparable mechanical and thermal effects that occur in the last problem. Problem 36 has been solved with the use of the coupled approach with coupling parameter β = 0.03, without taking the fields of temperature and deformation into account. The differences between the temperature distributions at τ = 0.45 are 1.5% in the plate’s centre and 0.5% at


145

1

-u ·10-2 a)

4 3 2 1

-2 µ·10 ·10

5 4 3 2 1

x3

x1

x1

x3

3

-u ·10-2

2

u ·10-3 3 2 1

4 3 2 1

x3

x1

x1

x3

x2

x2

x1

x3

x1

x3 1

b)

-u ·10-2 4 3 2 1

-2 µ·10 ·10

5 4 3 2 1

x3

x3

x1

x1

x2

x2

2

u ·10-3 4 3 2 1

x3

x3

x1

x1

x3 3

u ·10-4 x1

x3

3 2 1

x1

Figure 1.54. The distribution of temperature and displacements (plane x2 = 0.5) for the plate investigated in problem 36 at time instant: a) τ = 0.1, b) τ = 0.3, c) τ = 0.43.

146


-u ·10-2

c)

4 3 2 1

-1 µ·10 ·10

x3

4 3 2 1

x3

x1

3

x1

2

·10

-4

3 2 1

-3

x3

x1

3 2 1

x3

·10

x1 x2

x3

x2

x1

x3

x1

Figure 1.54. cont.

the active wall. The coupling effect for displacements is 1% in the plate’s centre at τ = 0.45. The effect disappears towards the edges like in the elastic problem. The changes of displacement u3 in time in points (7/8, 0.5, 0.5) and (7/8, 0.5, 7/8) are illustrated in Figure 1.56a for problem 33 and in Figure 1.56b for problem 36. Displacements u1 and u2 increase uniformly in time like in the problem with high temperature, and the order of the values of displacements u1 and u2 is the same as in problem 35 (u1 is 10−1 , whereas u2 is 10−2 ). In course of time, the coupled problem displays the tendency to relatively increase the temperature field and to reduce the deformations, i.e.: the plate’s reaction is damped. Comparison of the results of the temperature and the deformation fields’ coupling’s influence in the linear and the physically non-linear approaches, and also the solutions of problems with various boundary conditions in the linear approach, leads to various conclusions as for the effects of the coupling’s influence. The type of thermal boundary conditions (within the investigated class of problems) gives evidence of little influence on the coupling effect (Fig. 1.41a – 12%, Fig. 1.41b – 10%, problems 25 and 26 at τ = 0.4). The character of the problem’s formulation also weakly depends on the coupling effect (in problem 33’s non-linear

1.5 3D Physically Non-Linear Problems a)

147 0.47 0.45 0.4

¾x x

1 1

0.3

0.1 0.2

0.05

0.1 0.3 0.1

0

x1

-0.05

b)

¾x x

1 1

0.1

0.05

0

0.1

0.2

0.3

0.4

t

0.5

Figure 1.55. The graph of the normal stress: a) along axis (x1 , 0.5, 0.5) - σ x1 x2 (x1), b) σ x1 x1 (τ).

approach it is not visible within the investigated time interval, whereas in problem 30 with the linear approach there is a small coupling effect – 1% in the plate’s centre). The type of mechanical boundary conditions exerts the greatest influence on the coupling effect (in problem 30 with rigidly fixed walls – 1%, in problem 25 with free walls – 12%–10%). The results of the calculations confirm the theoretical assumptions that the larger the fixing surface and the thermal insulation surface are, the more the body’s inner state reflects the converse adiabatic state. An isoentropic process occurs in the form of the converse adiabatic process. If a body is thermally insulated but it can be affected by a force (a mechanical impact in the form of Heaviside’s function: problems 29, 30, 33–36), then it may also participate in the converse adiabatic process and it

148


a) 3

u ·10-3 4 2 0

0.2

0.3

0.4

0.5

t

0.5

t

-2 (7/8,0.5,7/8)

-4 -6

(7/8,0.5,0.5)

-8

b)

3

u ·10-3 4 2 0

(7/8,0.5,7/8)

0.1

0.2

0.3

0.4

-2 -4

(7/8,0.5,7/8)

-6 -8

Figure 1.56. The changes of displacement u3 in time in points (7/8, 0.5, 0.5) and (7/8, 0.5, 7/8): a) problem 33, b) problem 36.

is characterised by constant entropy. In a general case, investigation is carried out into the processes of heat conduction and into the inner irreversible processes characterised by the increasing entropy and the increasing entropy flux, which serves as the basis for the theory of the reciprocal coupling of the temperature and the deformation fields. The observation of such processes may help explain the increase of the effect of reciprocal coupling at the mechanical (the second and the third boundary problem) and the thermal (the first and the third boundary problem) “liberation” of the plate’s edges.

2 Stability of Rectangular Shells within Temperature Field

A brief historical research review is given in section 2.1. In section 2.2 variational equations in a hybrid form in curvilinear coordinates are derived for shallow anisotropic shells, as well as the variational and differential equations in rectangular coordinates for shallow homogeneous anisotropic shell within temperature field are reported. Compatibility relations of boundary conditions for homogeneous anisotropic rectangular shallow shells in a corner point and in the points, where the boundary conditions are changed, are derived. Coupling conditions for isotropic homogeneous shallow shells are given. Finally, the problem of stress-strain state of shallow shells in temperature field is formulated. In Section 2.3 universality and efficiency of the finite difference method devoted to boundary value problems for elliptic equations is discussed and illustrated. It is shown that for multi-dimensional stationary heat transfer problems an application of a 4th order finite difference method is sufficient. In particular, it is outlined that upper relaxation method possesses a relatively high convergence velocity, is simple in realization, and requires small amount of storage memory. Difference schemes with approximation error o(|h|4 ) for series of multi-dimensional stationary heat transfer equations governing temperature field distribution in isotropic, orthotropic and anisotropic homogeneous and non-homogeneous media are constructed. Theorem on convergence of the proposed difference scheme to solution of an initial differential system with velocity of o(|h|4 ) is formulated and proved. In addition, algorithm of 3D heat transfer stationary equation is proposed. Compatibility conditions for difference boundary value problems with approximation error p(|h|4 ) are obtained. Efficiency of the algorithm is illustrated using model problem with various boundary conditions. In section 2.4 the difference equations approximating the system of nonlinear differential equations of shallow shell with approximation error o(|h|4 ) are constructed. It is illustrated by comparing computational results of shells with approximation o(|h|4 ) and o(|h|2 ), that the approximation o(|h|4 ) is more efficient. Intervals of iterational parameter variations applied in the nonlinear relaxation method, are defined experimentally. Owing to computations of flexible anisotropic homogeneous shells with planes of stiff symmetry orthogonal to axis z, fibres orientation in stiff symmetry plane have essential influence on shell stability. Among other results, it is detected that a change of fixation type along shell contour essentially influences both a value of critical loads and shell stress-strain state. Owing to increase

150


of geometrical parameters k1 , k2 , the shell becomes more sensitive to fixation type along its contour side. In the section 2.5 we show that heat sources occurrence influences a shell stressstrain state and its stability. An essential influence of temperature field type defined by the corresponding boundary conditions on the stress-strain shell state is illustrated. It is also demonstrated and discussed, how both fixation type along shell contour side and transversal load action modify shell stress-strain state and its stability within a temperature field.

2.1 Introduction Shells are members of many structures and machines in many timeline branches of technics. A wide spectrum of shells application is motivated by design of simultaneously strength and light constructions. It is clear that to achieve this requirement a real stress-strain state estimation is highly required. This is a reason for development of precise and economical computational techniques devoted to analysis of various constructions including shells and plates being their members. Nowadays an investigation of plates and shells in condition of high temperatures is very challenging, since the thermal stresses can lead to stability loss or collapse of constructions. Thermal stresses have been investigated for a long time. Duhamel (1837-1838) and Neuman (1841) derived equations governing thermoelastic stresses behaviour. In 1879 Hopkinson, and in 1900 Aliband constructed equations of thermoelastic equilibrium state in the form used nowadays. Famous scientists of last century have been involved in creation of mathematical fundamentals of heat transfer theory like Ostrogradskiy, Kelvin, Duhamel, Kirchhoff, Maxwell, Stokes, Lamé, Bossinesque, Rayleigh, Lamb, and others. Many fundamental results of heat transfer theory are included in monographs [292, 451]. In the book [678] practical methods devoted to solutions of heat transfer problems are reported. Some achievements of nonlinear heat transfer problems up to 1975 are given in the monograph [358]. It includes approximate analytical and numerical methods of nonlinear problems of energy pumping or investigation of physical media possessing mathematical model analogous to heat transfer processes. However, only 2D (two dimensional) heat transfer problems are solved and the finite difference method of higher order is not applied. Various aspect of heat transfer theory are discussed in monographs [193, 401]. It is worth noticing that in this period an essential contribution to theory of thermal stresses, and in particular into solution of stationary problems have been carried out by Central an Eastern Europe scientists like Galerkin, Dinnik, Lebedev, Maslov, Muschelishvili, Papkovitch, and others. Lebedev’s monograph [420] gives state-of-art of earlier period of thermal stresses problems development. Long time ago Maizel [453] applied Betty’s theorem to investigate heat transfer phenomena. Further development in this field can be traced through monographs [342, 471, 515].

2.1 Introduction

151

Analysis of 3D (three dimensional stationary) heat transfer equation is complicated. Majority of authors introduces series of simplified hypotheses of this equation. Problem devoted to thermomechanical shell state is often reduced to consideration of the corresponded 2D problem via representation of a being sought quantities in the form of series with respect to powers of the coordinate x3 . Temperature representation as a series of infinite length is proposed first by Malkin [456]. Analogous approach to solve the problems of theory of elasticity is proposed by Lurie [446]. In the reference [460], assuming a linear temperature distribution along the plate thickness, the corresponding heat transfer equations are obtained through temperature averaging along thickness. This method has been further extended into the case of unstationary regime for thin shells. Applying similar like assumptions with respect to temperature distribution along thickness, the analogous equations are obtained using the variational principle by Bolotin [124]. Polynomial representation of temperature distribution along plate and shell thickness is also applied by Danilovskaya [162]. Monograph [555] is devoted to fundamental problems of heat transfer and thermal elasticity of thin shells. It includes methods of solutions for statical, quasistatical and dynamical problems of thermoelasticity of plates and shells for different cases of heating. Dynamical problems of linear theory of shells are considered in reference [552]. Computations of statical problems of theory of isotropic plates and shells with linear temperature distribution along thickness accounting of temperature dependence of linear expansion coefficient and Young modulus are carried out in references [374, 535]. Axially symmetric geometrical nonlinear problems using both Bubnov and Ritz methods with a linear temperature distribution along a thickness are studied in references [247, 306]. Let us briefly analyse geometrically nonlinear problems of statics of plates and shells. Equations of plates and shell finite deflections behaviour using Kirchhoff-Love hypothesis are derived in references [211, 308] and they are generalized into the case of curvilinear surfaces in references [461, 682]. Finite difference method with approximation O(h2 ) has been first applied to analyse geometrically nonlinear plates in reference [296]. Then a wide class of flexible plates problems are solved in references [345, 346, 347]. In the cited Kornishin’s works a higher order finite difference method to solve equations governing behaviour of flexible plates is used for the first time. Mathematical problems devoted to solvability of boundary value problems of nonlinear plates and shells theory have been initiated in references [187, 205, 329, 482, 688, 718]. Nowadays engineers working in various industrial branches, and particularly in civil, and electronic and electrotechnic engineering are focused on analysis of stressstrain states of plates and shells with various (sometimes hybrid types) clamping

152


along their contours with both mechanical and temperature excitations and accounting influence of heat sources and various temperature conditions. Both actions of transversal load and temperature field on the flexible rectangular shells are rarely investigated. Among others, the following unsolved questions are addressed in this chapter: (i) temperature field investigation directly from three dimensional stationary heat transfer equations; (ii) influence of various shell contour clamping; (iii) influence of heat sources and temperature conditions on stress-strain shell and its stability; (iv) derivations of compatibility conditions for orthotropic shallow shells.

2.2 Flexible Anisotropic Shallow Shells in Temperature Fields 2.2.1 Problem formulation and assumptions Consider a shell with its middle surface bounded by a closed curve Γ. Let us attach to its middle surface an orthogonal system of curvilinear coordinates α, β (note that the coordinates do not need to overlap with main shell curvatures). It is assumed that Lamé parameters A, B and curvature radii R 1 , R 12 , R 2 of the middle surface are continuous together with their first derivatives with respect to directions α, β. Contrary to main radiuses R1 , R2 (solid curves) notation, radiuses of a curvature in directions α, β are denoted by dashed curves. It is assumed that the function H(α, β) does not have first order discontinuities, and Hmax ≡ H0 is essentially smaller than the smallest main curvature radius Rmin . Furthermore, it is assumed that H0 /Rmin can be neglected in comparison to 1 (shells which satisfy this assumption are called thin shells). Let us denote displacements in direction α, β, γ by u, v, w, respectively. Since a shell has low stiffness in direction γ, deflection w is of the same order as H, and hence the inequality w/H 1 is not satisfied. Owing to this observation, a geometrically nonlinear theory is introduced, and instead of nonlinear terms w/H, derivatives of deflection with respect to coordinates appear. Therefore, the relation w ∼ H defines shell stiffness in various directions in an indirect way. All displacement components are taken significantly less than a characteristic shell dimension of middle surface. Let us introduce the initial deflection w0 (α, β). This function gives imperfections of initial form of the middle surface in prebuckling state. It is assumed that it is continuous together with its first and second derivatives with respect to α and β, and it is of thickness order. Deformations ε11 , ε12 , ε22 in middle surface are assumed to be small in comparison to one. However, it does not mean that a coupling between stresses and deformations should be linear. Variations of curvatures are characterized by the parameters κ11 , κ12 , κ22 . In this chapter technical theory of shells is used assuming that an influence of displacements u, v on the curvature variation parameters and first derivatives with

2.2 Flexible Anisotropic Shallow Shells in Temperature Fields

153

respect to deflection stress function with the multiplier AB/R1 R2 can be neglected in comparison to high order derivatives. It is assumed that shell material is non-homogeneous and anisotropic. The introduced theory is based on normal hypothesis. In typical formulation of straight normals, it is assumed that a length of normal elements is conserved. It means that deformations εγγ are neglected in comparison to one. In the introduced theory εγγ is approximately defined through condition of in-plane strain state δγγ = 0. A shell can be loaded through distributed (along its edges) perpendicular Q0n , normal T n0 and tangential T 0 forces in middle plane, and the bending moment Mn0 . Continuously distributed surface load Z is assumed to be normal to middle surface. In the case H const tangential pressure components normal to surfaces 1 ∂H γ = ±H(α, β) are neglected. In addition, derivatives A1 ∂H ∂d , B ∂β are assumed to be small in comparison to one. Volume forces in middle surface with potential U and temperature field T depending on three coordinates are accounted. 2.2.2 Fundamental relations Owing to straight normals hypothesis, shell deformation in its arbitrary point has the form ([555], p.73): eαα = ε11 + γκ11 ,

eββ = ε22 + γκ22 ,

eαβ = ε12 + 2γκ12 .

(2.1)

Deformations in middle surface [306, 685] read ε11 =

ε12

2 2 1 ∂A w 1 1 ∂w1 1 ∂u 1 1 ∂w + v− + − , A ∂α AB ∂β R1 2 A ∂α 2 A ∂α

2 2 1 ∂B w 1 1 ∂w1 1 ∂v 1 1 ∂w0 + u− + − , ε22 = B ∂β AB ∂α R2 2 B ∂β 2 B ∂β A ∂ u B ∂ v 2w 1 ∂w1 1 ∂w1 1 ∂w0 1 ∂w0 · − · , = + + + B ∂β A A ∂α B R12 A ∂α B ∂β A ∂α B ∂β

where: w1 = w + w0 . Parameters of middle surface curvature variation have te form [306]: 1 ∂A 1 ∂w 1 ∂ 1 ∂w − ≡ −w

αα , κ11 = − A ∂α A ∂α AB ∂β B ∂β 1 ∂B 1 ∂w 1 ∂ 1 ∂w − ≡ −w

ββ , κ22 = − B ∂β B ∂β AB ∂α A ∂α 1 ∂A 1 ∂w 1 ∂ 1 ∂w + ≡ −w

αβ . κ12 = − A ∂α B ∂β AB ∂β A ∂α

(2.2)

(2.3)

Note that these expressions are approximated even within a linear theory. In exact formulas also terms u/R1 , v/R1 are accounted. Neglection of these terms belongs

154


to one of the technical shell theory assumptions. In relations (2.3) nonlinear terms do not appear, since angles of rotation are neglected. Relation between stresses and deformations is presented in the form of generalized Hook’s law [14] eαα = a11 σαα + a12 σββ + a13 σγγ + a16 σαβ + αT11 T, eββ = a12 σαα + a22 σββ + a23 σγγ + a26 σαβ + αT22 T, eγγ = a13 σαα + a23 σββ + a33 σγγ + a36 σαβ + αT33 T, eαβ = a16 σαα + a26 σββ + a36 σγγ + a66 σαβ + αT12 T, eαγ = a45 σβγ + a55 σαγ + αT13 T,

eβγ = a44 σβγ + a45 σαγ + αT23 T,

(2.4)

where: αTii , αTij are temperature coefficients of linear extension and shears of an anisotropic body. Solving first four equations of (2.4) with respect to σαα , σββ , σγγ , σαβ , one gets σαα = c11 eαα + c12 eββ + c13 eγγ + c16 eαβ − βo11 T, σββ = c12 eαα + c22 eββ + c23 eγγ + c26 eαβ − βo22 T, σγγ = c13 eαα + c23 eββ + c33 eγγ + c36 eαβ − βo33 T, σαβ = c16 eαα + c26 eββ + c36 εγγ + c66 eαβ − βo12 T, where:

c11 = m−1 a22 a33 a66 − a22 a236 − a223 a66 + 2a23 a26 a36 − a226 a33 ,

c22 = m−1 a11 a33 a66 − a11 a236 − a213 a66 + 2a13 a16 a36 − a216 a33 ,

c33 = m−1 a11 a22 a66 − a11 a226 − a212 a66 + 2a12 a16 a26 − a216 a22 ,

c66 = m−1 a11 a22 a33 − a11 a223 − a212 a33 + 2a12 a13 a23 − a213 a22 ,

c12 = m−1 a12 a236 − a12 a33 a66 + a13 a23 a66 − a13 a26 a36 + a16 a26 a33 − a16 a23 a36 ) , c13 = m−1 (a12 a23 a66 − a12 a26 a36 − a13 a22 a66 + a13 a226 + a16 a22 a36 − a16 a26 a23 , c26 = m−1 (a11 a23 a36 + a11 a33 a26 − a13 a12 a36 + a213 a26 + a16 a12 a33 − a16 a13 a23 , c16 = m−1 (a12 a26 a33 − a12 a23 a36 + a13 a22 a36 − a13 a23 a26 − a16 a22 a33 + a16 a223 ,

c36 = m−1 a11 a26 a23 − a11 a22 a36 + a212 a23 +

(2.5)


155

a16 a13 a22 − a12 a13 a26 − a16 a12 a23 ) , c23 = m−1 (a11 a36 a26 − a11 a23 a66 + a13 a12 a66 − a13 a16 a26 − a16 a12 a36 + a216 a23 , a11 a12 a13 a16 a a a a m = 12 22 23 26 , a13 a23 a33 a36 a16 a26 a36 a66 βo11 = αT11 c11 + αT22 c12 + αT33 c13 + αT12 c16 , βo22 = αT11 c12 + αT22 c22 + αT33 c23 + αT12 c26 , βo33 = αT11 c13 + αT22 c23 + αT33 c33 + αT12 c36 , βo12 = αT11 c16 + αT22 c26 + αT33 c36 + αT12 c66 .

(2.6)

Note that in relation (2.4) ai j = ai j (α, β, γ). In theory of anisotropic shells a transformation of material coefficients into a new coordinate system α , β , γ is required, assuming that their values are known in another coordinates system α, β, γ. Let a structure of material in each point body point has only one plane of elastic symmetry, parallel to the middle surface, which coincides with the shifted surfaces αβ and α β . Let the coordinates α, β, γ and α , β , γ be identical after rotation in amount of angle ϕ around general axis γ = γ . The following transformation formulas are obtained [424]: a 11 = a11 cos4 ϕ + (2a12 + a66 ) sin2 ϕ cos2 ϕ + a22 sin4 ϕ+ (a16 cos2 ϕ + a26 sin2 ϕ)sin2ϕ, a 22 = a11 sin4 ϕ + (2a12 + a66 ) sin2 ϕ cos2 ϕ + a22 cos4 ϕ− (a16 sin2 ϕ + a26 cos2 ϕ) sin 2ϕ, a 12 = a12 + (a11 + a22 − 2a12 − a66 ) sin2 ϕ cos2 ϕ+ 1 (a16 − a26 ) cos 2ϕsin2ϕ, 2 a 66 = a66 + 4(a11 + a22 − 2a12 − a66 ) sin2 ϕ cos2 ϕ−

a 16

2(a16 − a26 ) cos 2ϕsin2ϕ, 1 2 2 = a22 sin ϕ − a11 cos ϕ + a12 + a66 cos 2ϕ sin 2ϕ+ 2

a16 cos2 ϕ(cos2 ϕ − 3 sin2 ϕ) + a26 sin2 ϕ(3 cos2 ϕ − sin2 ϕ), 1

2 2 a26 = a22 cos ϕ − a11 sin ϕ − a12 + a66 cos 2ϕ sin 2ϕ+ 2 a16 sin2 ϕ(3 cos2 ϕ − sin2 ϕ) + a26 cos2 ϕ(cos2 ϕ − 3 sin2 ϕ),

156


a 13 = a13 cos2 ϕ + a36 sin ϕ cos ϕ + a23 sin2 ϕ, a 23 = a13 sin2 ϕ − a36 sin ϕ cos ϕ + a23 cos2 ϕ, a 36 = (a23 − a13 ) sin 2ϕ + a36 cos 2ϕ,

a 33 = a33 .

(2.7)

In a particular case, if a body is orthotropic and old axes α, β, γ are the main axes of elasticity, i.e. they are orthogonal to planes of elastic symmetry, then formulas describing elastic constants (2.7) are more simplified owing to introduction of the technical elasticity constants Ei , Gi j , νi j [424]: a 11 =

1 cos4 ϕ 2ν12 sin4 ϕ sin2 ϕ cos2 ϕ + + − , E1 G12 E1 E2

1 sin4 ϕ 2ν12 cos4 ϕ sin2 ϕ cos2 ϕ + + − , E1 G12 E1 E2 1 1 2ν12 1 ν12 a 12 = sin2 ϕ cos2 ϕ − + + − , E1 E2 E1 G12 E1 1 1 2ν12 1 1

a66 = 4 sin2 ϕ cos2 ϕ + + + − , E1 E2 E1 G12 G12 2 1 sin ϕ cos2 ϕ 2ν12

+ cos 2ϕ sin ϕcosϕ, − − a16 = 2 E2 E1 G12 E1 2 1 cos ϕ sin2 ϕ 2ν12

− cos 2ϕ sin ϕ cos ϕ, − − a26 = 2 E2 E1 G12 E1 ν23 ν23 ν13 ν13 2 2

2 2 =− sin ϕ + cos ϕ , a23 = − cos ϕ + sin ϕ , E2 E1 E2 E1 1 ν13 ν23 a 33 = sin 2ϕ. (2.8) , a 36 = − E3 E1 E2 a 22 =

a 13

Owing to σγγ = 0, the third equation of (2.5) yields σαα = B11 eαα + B12 eββ + B16 eαβ − β11 T, σββ = B12 eαα + B22 eββ + B26 eαβ − β22 T, σαβ = B16 eαα + B26 eββ + B66 eαβ − β12 T,

(2.9)

where: B11 = c11 −

c213 , c33

B26 = c26 −

B12 = c12 −

c23 c36 , c33

c13 c29 , c33

B22 = c22 −

c223 , c33

B16 = c16 −

c13 c36 , c33

B66 = c66 −

c236 , c33


β11 = βo11 −

c13 o β , c33 33

β22 = βo22 −

c23 o β , c33 33

β12 = βo12 −

157

c36 o β . c33 33

Substituting (2.1) into (2.9) gives σαα = B11 ε11 + B12 ε22 + B16 ε12 + γ(B11 κ11 + B12 κ22 + 2B16 κ12 ) − β11 T, σββ = B12 ε11 + B22 ε22 + B26 ε12 + γ(B12 κ11 + B22 κ22 + 2B26 κ12 ) − β22 T, σαβ = B16 ε11 + B26 ε22 + B66 ε12 + γ(B16 κ11 + B26 κ22 + 2B66 κ12 ) − β12 T.

(2.10)

Integrating stresses with respect to γ, the following middle surface forces are obtained T 11 = D11,0 ε11 + D12,0 ε22 + D16,0 ε12 + D11,1 κ11 + D12,1 κ22 + 2D16,1 κ12 − t11,0 , T 22 = D12,0 ε11 + D22,0 ε22 + D26,0 ε12 + D12,1 κ11 + D22,1 κ22 + 2D26,1 κ12 − t22,0 , T 12 = D16,0 ε11 + D26,0 ε22 + D66,0 ε12 + D16,1 κ11 + D26,1 κ22 + 2D66,1 κ12 − t12,0 . (2.11) Multiplying stresses by γ and integrating over shell thickness, the following banding and rotation moments are obtained M11 = D11,1 ε11 + D12,1 ε22 + D16,1 ε12 + D11,2 κ11 + D12,2 κ22 + 2D16,2 κ12 − t11,1 , M22 = D12,1 ε11 + D22,1 ε22 + D26,1 ε12 + D12,2 κ11 + D12,1 κ22 + 2D26,2 κ12 − t22,1 , M12 = D16,1 ε11 + D26,1 ε22 + D66,1 ε12 + D16,2 κ11 + D26,2 κ22 + 2D66,2 κ12 − t12,1 . (2.12) Positive direction of forces and moments are shown in monograph [686]. Coefficients occurred in relations (2.11), (2.12) are the functions of α, β, and they are coupled through integrals H Di j,k (α, β) =

Bi j γk dγ,

i, j = 1, 2, 6,

k = 0, 1, 2.

(2.13)

−H

Temperature components of forces and moments read H ti j,k =

βi j T γk dγ,

i, j = 1, 2,

k = 0, 1.

(2.14)

−H

Formulas (2.11) yield middle surface deformations ε11 = A11 T 11 + A12 T 22 + A16 T 12 − d11 κ11 − d12 κ22 − d16 κ12 + T 1 , ε22 = A12 T 11 + A22 T 22 + A26 T 12 − d21 κ11 − d22 κ22 − d26 κ12 + T 2 , ε12 = A16 T 11 + A26 T 22 + A66 T 12 − d61 κ11 − d62 κ22 − d66 κ12 + T 12 , where coefficients Ai j and di j have the form

(2.15)

158

2 Stability of Rectangular Shells within Temperature Field 2 A11 = m−1 1 (D22,0 D66,0 − D26,0 ),

A12 = m−1 1 (D16,0 D26,0 − D12,0 D66,0 ),

A16 = m−1 1 (D12,0 D26,0 − D16,0 D22,0 ),

2 A22 = m−1 1 (D11,0 D66,0 − D16,0 ),

A26 = m−1 1 (D12,0 D16,0 − D11,0 D26,0 ),

2 A66 = m−1 1 (D11,0 D22,0 − D12,0 ),

d11 = A11 D11,1 + A12 D12,1 + A16 D16,1 , d12 = A11 D12,1 + A12 D22,1 + A16 D16,1 , & ' d16 = 2 A11 D16,1 + A12 D26,1 + A16 D66,1 , d21 = A12 D11,1 + A22 D12,1 + A26 D16,1 , d22 = A12 D12,1 + A22 D22,1 + A26 D26,1 , & ' d26 = 2 A12 D16,1 + A22 D26,1 + A26 D66,1 , d61 = A16 D11,1 + A26 D12,1 + A66 D16,1 , d62 = A16 D12,1 + A26 D22,1 + A66 D26,1 , & ' d66 = 2 A16 D16,1 + A26 D26,1 + A66 D66,1 , T 1 = A11 t11,0 + A12 t22,0 + A16 t12,0 , T 2 = A12 t11,0 + A22 t22,0 + A26 t12,0 , T 12 = A16 t11,0 + A26 t22,0 + A66 t12,0 , D11,0 D12,0 D16,0 m1 = D12,0 D22,0 D26,0 . D16,0 D26,0 D66,0

(2.16)

Owing to the following relations, the stress function in the middle surface [306]

T 11 = Fββ + U,

T 22 = Fαα + U,

T 12 = −Fαβ ,

(2.17)

are introduced. Function U represents a potential of the volume forces having the following projections X=−

1 ∂ U, A ∂α

Y=−

1 ∂ U, B ∂β

into the axes α, β. Recall known operators in curvilinear coordinates [306]: ∂ A ∂ 1 ∂ B ∂ + , ∆= AB ∂α A ∂α ∂β B ∂β ∂ 1 ∂ ∂ 1 ∂ ∂ 1 A ∂ 1 ∂ 1 B ∂ + + + . ∆k = AB ∂α R 2 A ∂α ∂α R12 ∂β ∂β R12 ∂α ∂β R 1 B ∂β

(2.18)

(2.19)

(2.20)


159

2.2.3 Variational and differential equations Owing to a principle of virtual displacements, the variation of full energy δ of a deformed shell is equal to zero in its equilibrium state δ≡ δV + δu +δc = 0,

(2.21)

where: δV - external forces work variation; δu - potential energy variation, occurred in result of bending deformation; δc - energy variation yielded by the deformation in middle surface. Variation of external forces work reads [306]

(X · u + Y · v + Z · w) − δ T no un + T lo ul − Mno w n + Qon w dl , (2.22) δV = −δ where: un , ul - normal and tangent edge displacement components in the middle surface; w u - derivative with respect to deflection in normal direction to shell edge. It is assumed that the loads T no and T lo do not appear on edge intervals, where w and w n are not defined. Potential energy variation (deflection [306]) has the form (M11 δκ11 + M22 δκ22 + 2M12 δκ12 ) dS , δu = (2.23) and variation of middle surface deformation energy is given by (T 11 δε11 + T 22 δε22 + T 12 δε12 ) dS = δc ≡ δ

(T 11 ε11 + T 22 ε22 + T 12 ε12 ) dS− (ε11 δT 11 + ε22 δT 22 + ε12 δT 12 ) dS . (2.24)

This transformation allows for obtaining variational hybrid type equation (both deflection and stress function are variated). Substituting (2.22), (2.23) and (2.24) into (2.21) we get (T 11 ε11 + T 22 ε22 + T 12 ε12 − X · u − Y · v − Z · w) dS + δ≡ δ (M11 δκ11 + M22 δκ22 + 2M12 δκ12 − ε11 δT 11 − ε22 δT 22 − ε12 δT 12 ) dS − δ

T no un + T lo ul − Mno w n + Qon w dl = 0.

(2.25)

Consider the expression (2.25) as the sum of two terms δ= δ1 +δ2 = 0, where:

(2.26)

160


δ1 = δ

o o (T 11 ε11 + T 22 ε22 + T 12 ε12 − Xu − Yv) dS − T n un + T l ul dl ,

δ2 =

(M11 δκ11 + M22 δκ22 + 2M12 δκ12 − ε11 δT 11 −

ε22 δT 22 − ε12 δT 12 ) dS − δ

ZwdS −

&

Mno w n

−

' Qon w dl

.

(2.27)

Substituting expression (2.2) instead of deformations into δ1 and integrating by parts we get [306] 1 ∂ 1 ∂ 2 ∂B (BT 11 ) − T 22 + δ1 = −δ A T 12 + X udS − AB ∂α ∂α A ∂β 1 ∂ 2 1 ∂ ∂A (AT 22 ) − T 11 + δ B T 12 + Y vdS + AB ∂β ∂β B ∂α ⎤ ⎡ ⎡ ⎧ ⎪ ⎢⎢⎢ 1 ∂w 2 2 ∂w0 1 ∂w ⎥⎥⎥ ⎢⎢⎢ 1 ∂w 2 ⎪ 1 ⎨ ⎥ ⎢ ⎢ ∂ + + T ⎢ ⎥ + T 22 ⎢⎣ ⎪ ⎪ ⎩ 11 ⎣ A ∂α 2 A ∂α A ∂α ⎦ B ∂β

1 ∂w 1 ∂w0 1 ∂w 1 ∂w 1 ∂w 1 ∂w0 2 + 2T 12 + + B ∂β B ∂β A ∂α B ∂β A ∂α B ∂β T 11 1 ∂w 1 ∂w0 T 12 T 22 dS − δ −2 + wdS + B ∂β A ∂α R 1 R12 R2 $

% ' & δ T 11 − T no un + T 12 − T l0 ul dl.

(2.28)

∂F ∂F , RAB are neThe forces are governed by formulas (2.17), the terms RAB 1 R2 ∂α 1 R2 ∂β AB glected in comparison to higher derivatives of F (except of the multiplier l1 l2 ) and the following boundary conditions are attached

T 11 = T no ,

T 12 = T lo .

(2.29)

One may observe that first and second integrals over space and contour in δ1 are equal to zero. Note that two first underintegral expressions in (2.28) represent multiplication of left hand sides of the equilibrium equations in a middle surface within the technical theory of displacements. In these equations the transversal forces do not appear, owing to neglection of terms u/R1 and v/R1 standing in expression describing a curvature [306]. Recall that the neglection of first derivatives of F in comparison with the multiplier AB/R1 R2 is the second assumption of the technical shell theory. If at least one of main curvature radiuses is infinite, than an error does not appear. Applying integrations by parts to the first term (2.28), and the Gauss formula to the fourth term, the following is obtained [306]:


1 δ1 = − δ 2

161

1 1 2∆k F + L2 (w2 , F) − Φ(w2 ) + 2 + U wdS + R1 R2 1 ∂w ∂w δ T 11 + T 12 wdl, (2.30) 2 ∂n ∂l

where: w2 = w + 2w0 ,

Φ(ϕ) = X

1 ∂ϕ 1 ∂ϕ +Y − U∆ϕ, A ∂α B ∂β

L2 (ϕ, ψ) = ϕ

αα ψ

ββ − 2ϕ

αβ ψ

αβ + ϕ

ββ ψ

αα . Integral along a contour in (2.30) is equal to zero, since in the case of clamping w = 0, and on the free side T n = T e = 0 owing to (2.29). It should be emphasized that the physical material parameters do not appear in (2.30). Let us substitute into relations for moments (2.12) the deformation values in the middle surface (2.15), expressed via stress function and curvatures, in the form M11 = D11 T 11 + D12 T 22 + D16 T 12 − d11 κ11 − d12 κ22 − d16 κ12 + T 1∗ , M22 = D21 T 11 + D22 T 22 + D26 T 12 − d21 κ11 − d22 κ22 − d26 κ12 + T 2∗ , ∗ M12 = D16 T 11 + D62 T 22 + D66 T !2 − d61 κ11 − d62 κ22 − d66 κ12 + T 12 ,

where: D11 = D11,1 A11 + D12,1 A12 + D16,1 A16 , D12 = D11,1 A12 + D12,1 A22 + D16,1 A26 , D16 = D11,1 A16 + D12,1 A26 + D16,1 A66 , D21 = D12,1 A11 + D22,1 A12 + D26,1 A16 , D22 = D12,1 A12 + D22,1 A22 + D26,1 A26 , D26 = D12,1 A16 + D22,1 A26 + D26,1 A66 , d11 = D11,1 d11 + D12,1 d21 + D16,1 d61 − D11,2 , d12 = D11,1 d12 + D12,1 d22 + D16,1 d62 − D12,2 , d16 = D11,1 d16 + D12,1 d26 + D16,1 d66 − 2D16,2 , d21 = D12,1 d11 + D22,1 d21 + D26,1 d61 − D12,2 , d22 = D12,1 d12 + D22,1 d22 + D26,1 d62 + D22,2 , d26 = D12,1 d16 + D22,1 d26 + D26,1 d66 − D26,2 , T 1∗ = D11,1 T 1 + D12,1 T 2 + D16,1 T 12 − t11,1 , T 2∗ = D12,1 T 1 + D22,1 T 2 + D26,1 T 12 − t22,1 , ∗ = D16,1 T 1 + D26,1 T 2 + D66,1 T 12 − t12,1 , T 12

D61 = D16,1 A11 + D26,1 A12 + D66,1 A16 ,

(2.31)

162


D62 = D16,1 A12 + D26,1 A22 + D66,1 A26 , D66 = D16,1 A16 + D26,1 A26 + D66,1 A66 , d61 = D16,1 d11 + D26,1 d21 + D66,1 d61 − D16,2 , d62 = D16,1 d12 + D26,1 d22 + D66,1 d62 − D26,2 , d66 = D16,1 d16 + D26,1 d26 + D66,1 d66 − 2D66,2 .

(2.32)

Moments (2.31) and deformations in the middle surface (2.15) are substituted into δ2 in the form (2.27) to yield [R(F, w) − R(w, w) − R(F, F)] dS − δ δ2 = ZwdS + δ

∂w Mno ∂n

−

Qon w

dl +

N T − M T dS ,

(2.33)

where: R(F, w) = (D11 T 11 + D12 T 22 + D16 T 12 )δk11 + (D21 T 11 + D22 T 22 + D26 T 12 )δk22 + 2(D61 T 11 − D62 T 22 + D66 T 22 )δk12 − (d11 k11 + d12 k22 + d16 k12 )δT 11 − (d21 k11 + d22 k22 + d26 k12 )δT 22 − (d61 k11 + d62 k22 + d66 k12 )δT 12 , R(w, w) = (d11 k11 + d12 k22 + d16 k12 )δk11 + (d21 k11 + d22 k22 + d26 k12 )δk22 + 2(d61 k11 + d62 k22 + d66 k12 )δk12 , R(F, F) = (A11 T 11 + A12 T 22 + A16 T 12 )δT 11 + (A12 T 11 + A22 T 22 + A26 T 12 )δT 22 + (A16 T 11 + A26 T 22 + A66 T 12 )δT 12 ,

(2.34)

M = T 1 δT 11 + T 2 δT 22 + T 12 δT 12 , T

∗ N T = T 1∗ δk11 + T 2∗ δk22 + 2T 12 δk12 ,

(2.35)

Owing to substitution of δ1 and δ2 into (2.26), the being sought variational equation is obtained. Observe that operation of function variations having physical parameters is not used during variations of the functions. The being sought variational equation is of a hybrid type, since both varied and being sought functions w and F appear independently. The obtained variational equation generalizes equations obtained in references [212, 686]. In the case of homogeneous and physically linear shells, the hybrid types variational equations are reported by Alumiae [13]. In the case of isotropic non-homogeneous isotropic shells with physical and geometrical non-linearities the hybrid type variational equation has been obtained by Kantor [306], whereas for the case of elastic-plastic material including loading and second order plastic deformations it has been derived by Krys’ko [369]. Variational background of theory of thin plates and shells is given in references [212, 533].


163

It is worth noticing that the obtained variational equation can be applied during calculations of composite shells, since the first order discontinuities of Young modulus and Poisson’s coefficients do not violate existence of integrals Di j,k . In the case of rectangular coordinates x, y, z the Lamé coefficients A = B = 1. For a given homogeneous anisotropic shell material with heat and elastic symmetry plane, perpendicular to middle shell surface and orthogonal to axis z, the coefficients in (2.4) are constant. Owing to the Kirchhoff-Love hypothesis and relation ezz = 0, the generalized Hook’s law can be rewritten in the form e xx = a11 σ xx + a12 σyy + a16 σ xy + αT11 T (x, y, z), eyy = a12 σ xx + a22 σyy + a26 σ xy + αT22 T (x, y, z), e xy = a16 σ xx + a26 σyy + a66 σ xy + αT12 T (x, y, z).

(2.36)

Solving (2.36) with respect to the stresses σ xx , σyy , σ xy , the following equations are obtained σ xx = B11 e xx + B12 eyy + B16 e xy − β11 T, σyy = B12 e xx + B22 eyy + B26 e xy − β22 T, σ xy = B16 e xx + B26 eyy + B66 e xy − β12 T, where:

2 B11 = m−1 2 (a22 a66 − a26 ),

(2.37)

B12 = m−1 2 (a26 a16 − a12 a66 ),

B16 = m−1 2 (a12 a26 − a22 a16 ),

2 B22 = m−1 2 (a11 a66 − a16 ),

B26 = m−1 2 (a12 a16 − a11 a26 ),

2 B66 = m−1 2 (a11 a22 − a12 ),

β11 = αT11 B11 + αT22 T B12 + αT12 B16 , β12 = αT11 B16 + αT22 B26 + αT12 B66 , β22 = αT11 B12 + αT22 B22 + αT12 B26 , a11 a12 a16 m2 = a12 a22 a26 . a16 a26 a66 Owing to hypothesis of straight normals, the relations similar to (2.1) occur, namely e xx = ε11 + zκ11 , eyy = ε22 + zκ22 , e xy = ε12 + 2zκ12 , where: ε11 =

2 ∂u 1 ∂w − κ1 w + , ∂x 2 ∂x ε12 =

ε22 =

2 ∂v 1 ∂w − κ2 w + , ∂y 2 ∂y

∂u ∂v ∂w ∂w + + . ∂y ∂x ∂x ∂y

Substituting these relations into multiplied by z equation (2.37), and into (2.36), and integrating them with respect to z from −H to H, the following relations governing deformations, stresses and moments in the shell middle surface are obtained

164


1

a11 T 11 + a12 T 22 + a16 T 12 + αT11 T N , 2H 1

= a12 T 11 + a22 T 22 + a26 T 12 + αT22 T N , 2H 1

= a16 T 11 + a26 T 22 + a66 T 12 + αT12 T N , 2H

ε11 = ε22 ε12 H M11 ≡

σ xx zdz =

2H 3 (B11 κ11 + B12 κ22 + 2B16 κ12 ) − β11 T M , 3

σyy zdz =

2H 3 (B12 κ11 + B22 κ22 + 2B26 κ12 ) − β22 T M , 3

σ xy zdz =

2H 3 (B16 κ11 + B26 κ22 + 2B66 κ12 ) − β12 T M , 3

−H

H M22 ≡ −H

H M12 ≡ −H

where: κ11 = − H T 11 ≡ −H

∂2 w , ∂x2

κ22 = −

∂2 F σ xx dz = 2 , ∂y

∂2 w , ∂y2

κ12 = − H

T 22 ≡

σ xy dx = − −H

H TN ≡

∂2 F , ∂x2

∂2 F , ∂x∂y H

T (x, y, z)dz, −H

σyy dz = −H

H T 12 ≡

∂2 w , ∂x∂y

TM ≡

zT (x, y, z)dz. −H

The obtained relations are substituted into (2.26) instead of deformations, stresses and moments, and the following variational equation in rectangular coordinates system for homogeneous anisotropic shell subjected to an action of transversal load q and temperature field T (x, y, z) is obtained ∂4 w ∂4 w ∂4 w (2H)3 B11 4 + 2(B12 + 2B66 ) 2 2 + B22 4 + δ= 12 ∂x ∂x ∂y ∂y ∂4 w ∂4 w ∂2 T M ∂2 T M ∂2 T M − + β11 4B16 3 + 4B26 + β22 + 2β12 3 2 2 ∂x∂y ∂x ∂y ∂x∂y ∂x ∂y ∂2 F ∂2 F ∂ 2 w ∂2 F ∂ 2 w ∂2 F ∂ 2 w ∂2 F − q δwdxdy− k1 2 − k2 2 − 2 2 − 2 + 2 ∂x∂y ∂x∂y ∂y ∂x ∂x ∂y ∂y ∂x2


∂4 F 1 ∂4 F ∂4 F ∂4 F a11 4 + (2a12 + a66 ) 2 2 + a22 4 − 2a16 − 2H ∂y ∂x ∂y ∂x ∂x∂y3 2 2 2 2 2 ∂ w∂ w ∂4 F T ∂ TN T ∂ TN T ∂ TN + 2H − α12 − 2a26 3 + α11 2 + α22 ∂x∂y ∂x ∂y ∂y ∂x2 ∂x2 ∂y2 ⎞⎤ 2 2 2 ∂w ∂2 w ∂2 w ⎟⎟⎟⎟⎥⎥⎥⎥ ∂ F ∂u ∂2 F ∂v + − + k1 2 2 ⎟⎠⎥⎦ δFdxdy + δ ∂x∂y ∂y ∂x ∂y2 ∂x ∂x2 ∂y l2 ∂ (2H)3 ∂2 F ∂u ∂v ∂2 w ∂2 w dxdy + − B11 2 + B12 2 + ∂x∂y ∂y ∂x 12 ∂x ∂x ∂y

0

∂ w ∂2 w ∂2 w ∂2 w ∂ δw + B11 2 + B12 2 + 2B16 δw− ∂x∂y ∂x∂y ∂x ∂x ∂y ∂ ∂2 w ∂2 w ∂2 w ∂ B16 2 + B26 2 + 2B66 δw + β11 T M δw− 2 ∂y ∂x∂y ∂x ∂x ∂y l1 ∂T M ∂T M ∂w ∂2 F ∂w ∂2 F β11 δw − 2β12 δw + δw δw − dy+ ∂x ∂y ∂x ∂y2 ∂y ∂x∂y x=0 2

2B16

l1

(2H)3 ∂2 w ∂2 w ∂2 w ∂ − B12 2 + B22 2 + 2B26 δw+ 12 ∂y ∂x∂y ∂x ∂y

0

∂ ∂2 w ∂2 w ∂2 w ∂ ∂2 w δw − 2 B16 2 + B12 2 + B22 2 + 2B26 ∂x∂y ∂y ∂x ∂x ∂y ∂x ∂2 w ∂2 w ∂T M ∂ δw − β22 δw + β22 T M δw− B26 2 + 2B66 ∂x∂y ∂y ∂y ∂y l2 ∂w ∂2 F ∂2 F ∂w ∂T M 2β12 δw + δw δw − dx+ ∂x ∂y ∂x2 ∂x∂y ∂x y=0 l1 l2 (2H)3 ∂2 w ∂2 w ∂2 w + B16 2 + B26 2 + 2B66 δw + 2β12 T M δw 6 ∂x∂y ∂x ∂y x=0 y=0

l2 0

1 ∂ ∂2 F ∂2 F ∂2 F T a12 2 + a22 2 − a26 + α22 T N δF− 2H ∂x ∂x∂y ∂y ∂x

∂ ∂ ∂2 F ∂2 F ∂2 F ∂2 F T + α δF − a16 2 + + a − a T 22 26 22 N 2 2 ∂x∂y ∂x ∂y ∂y ∂x ∂y 2 1 ∂w ∂ ∂2 F ∂2 F ∂w + αT12 T N δF + δF + k2 δF− a26 2 − a66 ∂x∂y 2 ∂y ∂x ∂x ∂x

a12

165

166


∂2 w ∂w ∂ δF k2 w δF + 2 ∂x ∂y ∂x

l1

l1 dy +

x=0

0

∂ F ∂ F ∂2 F ∂2 F ∂ T T +α T N δF − a16 2 +a26 −a66 +α T N δF− a16 ∂x∂y 11 ∂x ∂x∂y ∂x∂y 12 ∂y 2 ∂ ∂2 F ∂2 F ∂2 F 1 ∂w ∂ T a11 2 + a12 2 − a16 + α11 T N δF + δF+ ∂x∂y ∂y 2 ∂x ∂y ∂y ∂x l2 ∂2 w ∂w ∂w ∂ 1 ∂w ∂2 F k1 δF − k1 w δF − k1 w δF + 2 δF a16 2 + dx + ∂y ∂y ∂y 2H ∂x ∂y ∂y y=0 l1 l2 ∂w ∂w ∂2 F ∂2 F = 0. + αT12 T n δF − δF (2.38) a26 2 − α66 ∂x∂y ∂x ∂y ∂x x=0 y=0 2

1 ∂ ∂2 F ∂2 F a11 2 + a12 2 − 2H ∂y ∂y ∂x

2

Comparing the coefficients by δw and δF for x, y ∈ G {0 < x < l1 , 0 < y < l2 } in the variational equations (2.38), the following system of nonlinear differential equations is yielded ∂4 w ∂4 w (2H)3 ∂4 w B11 4 + 2(B16 + 2B66 ) 2 2 + B22 4 + −∇2κ F − L(w, F) − q + 12 ∂x ∂x ∂y ∂y ∂4 w ∂4 w − Ψ1 (x, y) = 0, 4B16 3 + 4B26 ∂x ∂y ∂x∂y3 1 1 ∂4 F ∂4 F ∂4 F 2 a22 4 + a11 4 + (2a12 + a66 ) 2 2 − ∇κ w + L(w, w) + 2 2H ∂x ∂y ∂x ∂y ∂4 F ∂4 F − 2a26 3 + Ψ2 (x, y) = 0, (2.39) 2a16 3 ∂x∂y ∂x ∂y where:

H ∂2 T ∂2 T ∂2 T Ψ1 (x, y) = dz, z β11 2 + β22 2 + 2β12 ∂x∂y ∂x ∂y −H

H Ψ2 (x, y) = −H

∂2 T αT11 2 ∂y

+

∂2 T αT22 2 ∂x

−

αT12

∂2 T dz, ∂x∂y

∂ 2 w ∂2 F ∂ 2 w ∂2 F ∂ 2 w ∂2 F , + − 2 ∂x∂y ∂x∂y ∂x2 ∂y2 ∂y2 ∂x2 2 2 ∂w 1 ∂ 2 w ∂2 w L(w, w) = 2 2 − , 2 ∂x∂y ∂x ∂y

L(w, F) =

∇2κ F = κ1

∂2 F ∂2 F + κ , 2 ∂y2 ∂x2

∇2κ w = κ1

∂2 w ∂2 w + κ . 2 ∂y2 ∂x2

(2.40)


167

2.2.4 Boundary and compatibility conditions Owing to various types of supports in real shell structures, a rich set of mathematical models of boundary value problems appear. A choice of boundary conditions has essential influence on the choice of initial differential equations (either in hybrid form or with respect to displacements). First, some of the boundary conditions for differential equations (2.39) are briefly recalled. 1. Free support of shell edges

2. Sliding clamping a)

b)

w = M11 = T 11 = T 12 = 0

for x = 0, l1 ;

w = M22 = T 22 = T 12 = 0

for y = 0, l2 .

w=

∂w = 0, ∂x

T 11 = ε22 = 0,

for x = 0, l1 ;

w=

∂w = 0, ∂y

T 22 = ε11 = 0,

for y = 0, l2 .

w=

∂w = 0, ∂x

T 11 = T 12 = 0,

for x = 0, l1 ;

w=

∂w = 0, ∂y

T 11 = T 12 = 0,

for y = 0, l2 .

(2.41)

(2.42)

(2.43)

3. Free support on flexible non-extended (non-compressed) in tangential plane ribs w = M11 = T 11 = ε22 = 0 w = M22 = T 22 = ε11 = 0

for x = 0, l1 ; for y = 0, l2 .

(2.44)

4. Hybrid type boundary condition for for

x = 0, l1 x = 0, l2

- free support; - sliding clamping.

(2.45)

Note that various combination of boundary conditions (2.41)–(2.44) are possible not only along whole contour, but also along each of shell sides. In reference [592] on an example of bi-harmonic equation governing behaviour of thin homogeneous isotropic plate for small deflections, coupling conditions for fourth order equations and all possible boundary conditions are formulated, and the variational technique is applied. A similar like approach to the system of nonlinear differential equations (2.39) of flexible shallow homogeneous anisotropic shells is applied now. So far, in order to get the variational equation (2.38), a fixation type of a shell boundary has been not accounted. Accounting of one the fixation ways for x = y = 0, it is assumed that its energy is

168


1 = + 2

⎫ 2 2 2 2 2 l2 ⎧ ⎪ ⎪ ⎪ ⎪ ∂ w ∂w ∂w ⎨ 2⎬ + C1 + α1 + β1 w ⎪ B1 ⎪ ⎪ ⎪ 2 ⎩ ⎭ ∂x∂y ∂x ∂y 0

⎫ 2 2 2 2 2 l1 ⎧ ⎪ ⎪ ⎪ ⎪ ∂ w ∂ w ∂w ⎨ 2⎬ + C2 + α2 + β2 w ⎪ B2 ⎪ ⎪ ⎪ 2 ⎩ ⎭ ∂x∂y ∂y ∂x

1 2

dy+ x=0

dx+

y=0

0

1 2

⎫ 2 2 2 l2 ⎧ ⎪ ⎪ ⎪ ∂F ∂ F ∂F ∂F ⎪ ⎨ ⎬ + K1 + 2L1 A ⎪ ⎪ ⎪ 2 ⎩ 1 ∂y ⎭ ∂y ∂x ⎪ ∂y 0

1 2

dy+ x=0

⎫ 2 2 2 l1 ⎧ ⎪ ⎪ ⎪ ∂F ∂ F ∂F ∂F ⎪ ⎨ ⎬ + K2 + 2L2 A ⎪ ⎪ ⎪ 2 ⎩ 2 ∂x ⎭ ∂x ∂y ⎪ ∂x 0

dx,

(2.46)

y=0

where the coefficients Bi , Ci , αi , βi , Ai , Ki , Li (i = 1, 2) have fully defined values depending on boundary conditions type, and corresponds to variational equation (2.38) δ= 0. Owing to an account of external forces work on fixation curve, the functional corresponding to the problem (2.39) with non-homogeneous boundary conditions takes the form (2.47) ˜ = +1 , where: l2 1 = −

∂w M1o (y) ∂x

+

M2o (x) M1o (y),

dy − x=0

0

e1

l2

Qo1 (y)w

∂w + Qo2 (x)w ∂y

y=0

∂F + qo1 (y)F ∂x

0

e1 dx −

mo1 (y) mo2 (x)

0 0 o o o o o o M2 (x), Q1 (y), Q2 (x), m1 (y), m2 (x), q1 (y), qo2 (x)

∂F + qo2 (x)F ∂y

dy− x=0

dx, y=0

are external forces acting on the boundary, and is defined with the help of (2.46). Let us compute first variation of the functional (2.47) with respect to w and compare it to zero. Applying variational equation (2.38) and carrying out integration by parts, only terms related to the edge x = 0 are given (2H)3 12

l2

∂ ∂2 w ∂2 w ∂2 w B11 2 + B12 2 + 2B16 δw− ∂x ∂x∂y ∂x ∂y

0

∂ w ∂2 w ∂2 w ∂ ∂2 w ∂2 w ∂ δw + 2 B + B + 2B + B + 12 16 16 26 ∂x∂y ∂x ∂y ∂x2 ∂y2 ∂x2 ∂y2 2

B11

l2 ∂2 w ∂T M ∂ ∂T M δw δw + β11 T M δw − 2β12 δw+ β11 dy − 2B66 ∂x∂y ∂x ∂x ∂y x=0 0


l2

∂2 F ∂w ∂w ∂2 F δw − δw ∂x∂y ∂y ∂x ∂y2

β1 wδw + α1

∂ ∂w δw ∂x ∂x

dy − x=0

B1

169

∂4 w ∂3 w ∂ δw+ δw − C1 4 ∂y ∂x∂y2 ∂x

0

l2 x=0

−M1o (y)

dy +

∂ δw − Qo1 (y)δw ∂x

0

dy. x=0

∂ ∂x δw

and δw, the being sought boundary conComparing the terms standing by ditions for x = 0 are obtained (2H)3 ∂w ∂2 w ∂2 w ∂2 w − B11 2 + B12 2 + 2B16 + α1 ∂x 12 ∂x∂y ∂x ∂y ∂3 w C1 + β11 T M = M1o (y), ∂x∂y2 ∂2 w ∂2 w ∂2 w ∂ (2H)3 ∂ B11 2 + B12 2 + 2B16 + β11 T M + ∂x 12 ∂x ∂x∂y ∂x ∂y ∂ (2H)3 ∂2 w ∂2 w ∂2 w B16 2 + B26 2 + 2B66 + β12 T M = 2 ∂y 12 ∂x∂y ∂x ∂y 2 2 2 ∂2 F ∂w ∂ ∂ F ∂w + 2 . (2.48) −β1 w − 2 B1 2 + Qo1 (y) − ∂x∂y ∂y ∂x ∂y ∂y Furthermore, relations for moments in the middle surface in the case of orthotropic material are as follows (2H)3 ∂2 w ∂2 w ∂2 w B11 2 + B12 2 + 2B16 − β11 T M , M11 = − 12 ∂x∂y ∂x ∂y (2H)3 ∂2 w ∂2 w ∂2 w B12 2 + B22 2 + 2B26 − β22 T M , M22 = − 12 ∂x∂y ∂x ∂y (2H)3 ∂2 w ∂2 w ∂2 w B16 2 + B26 2 + 2B66 − β12 T M , (2.49) M12 = − 12 ∂x∂y ∂x ∂y where:

H TM =

zT (x, y, z)dz. −H

Then boundary conditions (2.48) take the form ∂ ∂w ∂2 w + M11 = C1 + M1o (y), α1 ∂x ∂y ∂x∂y ∂ ∂2 w ∂2 F ∂w ∂2 F ∂w ∂ ∂2 M11 + 2 M12 = β1 w + 2 B1 2 + − 2 − Qo1 (y). (2.50) ∂x ∂y ∂x∂y ∂y ∂y ∂y ∂y ∂x

170


Let us compute first variation of the functional˜ with respect to F and let derive the terms with respect to the edge x = 0: ⎫ 2 l2 ⎧ ⎪ ⎪ ⎪ ⎪ 1 ∂w ∂ ∂ ∂w ⎬ ⎨ ∂w ∂2 w δF + κ2 δF − κ2 w δF ⎪ δF + δ˜ = ⎪ ⎪ 2 ⎭ ⎩ ∂x ∂y 2 ∂y ∂x ∂x ∂x ⎪ 0

1 2H

l2

dy+ x=0

∂ ∂2 F ∂2 F ∂2 F a12 2 + a22 2 − a26 + αT22 T N δF− ∂x ∂x∂y ∂y ∂x

0

∂ ∂2 F ∂2 F ∂2 F T + α22 T N δF− a12 2 + a22 2 − a26 ∂x∂y ∂x ∂y ∂x ∂ ∂ ∂2 F ∂2 F ∂2 F T a16 2 + a26 2 − a66 + a12 T N δF dy+ ∂y ∂x∂y ∂y ∂y ∂x x=0 l2 −A1

∂2 F ∂2 F ∂4 F ∂F ∂ + K δF− δF + L1 − L 1 1 2 4 ∂x∂y ∂y ∂x ∂y ∂y

0

mo1 (y)

∂ δF − qo1 (y)δF ∂x

dy + . . . = 0. x=0

∂ δF and δF to zero, the being sought Comparing coefficients standing by ∂x boundary conditions for x = 0 for F are obtained

2 1 ∂2 F ∂2 F ∂2 F 1 ∂w a12 2 + a22 2 − a26 + αT22 T N − − κ2 w − 2 ∂y 2H ∂x∂y ∂y ∂x L1 κ2

∂F = mo1 (y), ∂y

1 ∂ ∂w ∂w ∂2 w ∂2 F ∂2 F ∂2 F T + a + α + + a − a T 12 22 26 22 N − ∂x ∂x ∂y2 2H ∂x ∂x∂y ∂y2 ∂x2 ∂2 F ∂2 F ∂2 F ∂2 F ∂2 F ∂ a16 2 + a26 2 − a66 + αT12 T N − A1 2 − L1 + ∂y ∂x∂y ∂x∂y ∂y ∂x ∂y ∂2 ∂2 F K = qo1 (y). (2.51) 1 ∂y2 ∂y2

Owing to relations for deformation in middle surface ∂2 F ∂2 F ∂2 F 1 a11 2 + a12 2 − a16 + αT11 T N , ε11 = 2H ∂x∂y ∂y ∂x ∂2 F ∂2 F ∂2 F 1 T a12 2 + a22 2 − a26 + α22 T N , ε22 = 2H ∂x∂y ∂y ∂x


ε12 =

171

1 ∂2 F ∂2 F ∂2 F a16 2 + a26 2 − a66 + αT12 T N , 2H ∂x∂y ∂y ∂x

the boundary conditions (2.51) are transformed to the form 2 1 ∂w ∂F = mo1 (y), − κ2 w − ε22 − L1 2 ∂y ∂y ∂2 ∂ ∂w ∂w ∂2 w ∂ ∂2 F ∂2 F ∂2 F + + K1 2 = q◦1 (y). (2.52) + ε22 − ε12 − A1 2 − L1 κ2 ∂x ∂x ∂y2 ∂x ∂y ∂x∂y ∂y2 ∂y ∂y The most important particular cases of boundary conditions (2.50)–(2.52) for the edge x = 0 are as follows: (a) α1 = C1 = 0, β1 = ∞, K1 = A1 = L1 = 0; this case corresponds to ball type unmovable support w = 0, M11 = M1o (y), ε22 = mo1 (y),

κ2

∂ ∂ ∂w + ε22 − ε12 = qo1 (y). ∂x ∂x ∂y

(2.53)

(b) α1 = β1 = ∞, A1 = ∞, L1 = 0; this case corresponds to sliding clamping ∂w = 0, ∂x

w=

ε22 = mo1 (y),

∂2 F = 0. ∂y2

(2.54)

(c) α1 = C1 = 0, β1 = ∞, A1 = ∞, L1 = 0; in this case the edge is supported on flexible in tangential plane unstretched rib w = 0,

M11 = M1o (y),

∂2 F = 0, ∂y2

ε22 = mo1 (y).

(2.55)

(d) α1 = C1 = 0, β1 = ∞, L1 = A1 = ∞; in this case the edge is freely supported w = 0,

M11 = M1o (y),

∂2 F = 0, ∂y2

∂2 F = 0. ∂x∂y

(2.56)

The relations (2.53)–(2.56) yield the homogeneous boundary conditions for M1o (y) ≡ 0, Qo1 (y) ≡ 0, mo1 (y) ≡ 0, qo1 (y) ≡ 0. Our next aim is focused on obtaining the compatibility conditions in the cusps of an anisotropic shell clamped on sides x = 0, y = 0. First, variation of the functional (2.47) with respect to w is obtained. Introducing typical transformations and equaling to zero the variational terms related to the cusp corner point x = 0, y = 0, one gets

172


(2H)3 2 12

l1 l2 ∂2 w ∂2 w ∂2 w + B16 2 + B26 2 + 2B66 δw ∂x∂y ∂x ∂y x=0 y=0

l ∂2 w ∂ ∂ ∂2 w ∂2 w ∂ 1 2 y=0 − B1 2 δw + B1 δw− 2 {β12 T M δw}lx=0 δw − C 1 ∂y ∂y2 ∂x∂y ∂x ∂y ∂y B2

∂2 w ∂ ∂ ∂2 w ∂62w ∂ δw + B2 δw = 0. δw − C2 2 ∂x ∂x2 ∂x∂y ∂y ∂x ∂x

(2.57)

∂ ∂ Comparing to zero the coefficients standing by δw, ∂x δw, ∂y δw, the compatibility conditions x = y = 0 for the function w are obtained ∂2 w ∂2 w ∂2 w (2H)3 2 B16 2 + B26 2 + 2B66 + 2β12 T M + 12 ∂x∂y ∂x ∂y ∂ ∂2 w ∂2 w ∂ B1 2 + B2 2 = 0, ∂y ∂x ∂y ∂x

C1

∂2 w ∂2 w + B2 2 = 0, ∂x∂y ∂x

B1

∂2 w ∂2 w = 0. + C 2 ∂x∂y ∂y2

(2.58)

The obtained compatibility conditions (2.58) correspond to anisotropic shell governed by nonlinear equations (2.39) with boundary conditions (2.50). Note that ∂ ∂ δw, ∂y δw are differthe compatibility conditions (2.58) are required only if δw, ∂x ent from zero. For Bi = Ci = 0 (i = 1, 2) only one compatibility condition exists M12 = 0. Problems devoted to compatibility conditions in the corners during searching for solutions of boundary value problems of elasticity are addressed in monograph [592]. Now we are going to derive continuation conditions in anisotropic shell corner for the function F. Let us compute first variation of the functional (2.47) with respect to F. Owing only to the terms related to the corner point x = y = 0 and comparing ∂ ∂ δF, ∂y δF, the following compatibility successively to zero coefficients by δF, ∂x conditions hold ∂w ∂w ∂2 F ∂2 F ∂2 F ∂F 1 a16 2 + a26 2 − a66 + αT12 T N − + L2 + 2H ∂x∂y ∂x ∂y ∂y ∂y ∂x ∂ ∂ ∂F ∂F ∂2 F ∂F ∂2 F + L1 − K1 2 + A 2 − K2 2 = 0, A1 ∂y ∂x ∂y ∂x ∂x ∂y ∂x K1

∂2 F = 0, ∂y2

K2 =

∂2 F = 0. ∂x2

(2.59)

Using relations for deformations in middle surface, the following compatibility conditions for the function F in the point x = y = 0 are obtained ∂w ∂w ∂F ∂F ∂F ∂F 1 ε12 − + A1 + L1 + A2 + L2 − 2H ∂x ∂y ∂y ∂x ∂y ∂y


173

∂ ∂ ∂2 F ∂2 F K2 2 − K1 2 = 0, ∂x ∂y ∂x ∂y K1

∂2 F = 0, ∂y2

K2

∂2 F = 0. ∂x2

(2.60)

Owing to the special case of boundary conditions, the following compatibility conditions for the function F in the anisotropic shell corners are obtained. For K1 = Ai = Li = 0 (i = 1, 2), what corresponds to the boundary condition (2.53), (2.54), one gets ∂w ∂w = 0. (2.61) ε12 − ∂x ∂y For Ai = ∞, Li = 0 (i = 1, 2), we obtain ∂F ∂F = = 0. ∂y ∂x

(2.62)

This compatibility condition corresponds to the boundary condition (2.55). For Ai = L1 = ∞ (i = 1, 2) we get ∂F ∂F = = 0. ∂y ∂x

(2.63)

This compatibility condition corresponds to the boundary condition (2.56). One may conclude, owing to (2.62), (2.63), that for the boundary condition (2.55), (2.56) the function F is constant in the corner points. Applying compatibility conditions (2.62), (2.63) into boundary conditions (2.55), (2.56), correspondingly, the function F can be defined in the considered space boundary. Hence, the boundary conditions (2.55) for the functions F read 1 ∂2 F ∂2 F a22 2 − a26 + αT22 T N = mo1 (y) for x = 0, F = C0 , 2H ∂x∂y ∂x ∂2 F ∂2 F 1 T a11 2 − a16 + α11 T N = mo2 (x) for y = 0. (2.64) F = C0 , 2H ∂x∂y ∂y Proceeding in a similar way, the function F can be defined for the boundary condition (2.56), which takes the following form ∂2 F = 0, ∂x∂y

∂2 F =0 ∂y2

for x = 0,

∂2 F = 0, ∂x∂y

∂2 F =0 ∂x2

for y = 0.

(2.65)

Using compatibility condition (2.63) and boundary conditions (2.65), the function F = 0 can be defined on the space boundary. The boundary conditions (2.65) for F read

174


∂F =0 ∂x ∂F =0 ∂y

F = 0, F = 0,

for x = 0, for y = 0.

(2.66)

In what follows, the compatibility conditions satisfied by function w and F in the point y = yc lying on a rectangular side (i.e. for 0 ≤ y ≤ yc one type boundary condition is given, whereas for yc ≤ y ≤ l2 another type of boundary condition is given) are derived. The functional, corresponding to this problem, has the form 1 ˜ = + 2

⎫ 2 2 2 2 2 yl ⎧ ⎪ ⎪ ⎪ ⎪ ∂w ∂ w ∂w ⎨ 2⎬ + C1 + α1 + β1 w ⎪ B1 ⎪ ⎪ ⎪ 2 ⎩ ⎭ ∂x∂y ∂x ∂y 0

1 2

⎫ 2 2 2 2 2 l2 ⎧ ⎪ ⎪ ⎪ ⎪ ∂w ∂ w ∂w ⎨ 2⎬ + C + α + β w B2 ⎪ ⎪ 2 2 2 ⎪ ⎪ ⎩ ⎭ ∂x∂y ∂x ∂y2 yl

yl

∂w Mio (y) ∂x

+

dy − x=0

0

1 2

l2

Qo1 (y)w

∂w M2o (y) ∂x

+

⎫ 2 2 2 l2 ⎧ ⎪ ⎪ ⎪ ∂F ∂ F ∂F ∂F ⎪ ⎨ ⎬ + K2 + 2L2 A ⎪ ⎪ ⎪ ⎩ 2 ∂y ⎭ ∂y ∂x ⎪ ∂y2 yl

∂F mo1 (y) ∂x

0

+

dy+ x=0

dy− x=0

l2

∂F mo2 (y)

dy −

qo1 (y)F x=0

dy+ x=0

yl

⎫ 2 2 2 yc ⎧ ⎪ ⎪ ⎪ ∂F ∂ F ∂F ∂F ⎪ ⎨ ⎬ + K1 + 2L1 A ⎪ ⎪ ⎪ ⎩ 1 ∂y ⎭ ∂y ∂x ⎪ ∂y2

yc

dy− x=0

Q02 (y)w

0

1 2

dy+ x=0

∂x

yc

+

qo2 (y)F

dy.

(2.67)

x=0

Let us derive first variation of the functional (2.67) with respect to w and compare it to zero. Integrating by parts and remaining only terms related to edge x = 0, one gets l2 −

∂2 w ∂2 w ∂2 w ∂ (2H)3 B11 2 + B12 2 + 2B16 + β11 T M δw+ ∂x 12 ∂x∂y ∂x ∂y

0

∂ (2H)3 ∂2 w ∂2 w ∂2 w B11 2 + B12 2 + 2B16 + β11 T M δw− 12 ∂x∂y ∂x ∂x ∂y ∂2 w ∂2 w ∂2 w ∂ (2H)3 B16 2 + B26 2 + 2B66 + β12 T M δw− 2 ∂y 12 ∂x∂y ∂x ∂y


yl

∂w ∂2 F ∂2 F ∂w δw + δw ∂x∂y ∂y ∂x ∂y2

dy + x=0

∂w ∂ β1 wδw + α1 δw ∂x ∂x l2

∂4 w ∂3 w ∂ δw+ δw − C1 4 ∂y ∂x∂y2 ∂x

0

yl dy + x=0

−M1o (y)


0

l2

−M2o (y) yl

B1


175

dy + x=0

B2 yl

dy+ x=0

∂4 w ∂3 w ∂ δw+ δw − C 2 ∂y4 ∂x∂y2 ∂x

∂w ∂ β2 wδw + α2 δw ∂x ∂x

dy = 0. x=0

∂ δw and δw, the boundary Now, comparing to zero the coefficients standing by ∂x conditions for x = 0 for the function w are obtained ∂ ∂w ∂2 w + M11 − C1 = M1o (y), α1 ∂x ∂y ∂x∂y ∂2 F ∂w ∂2 F ∂w ∂ ∂2 ∂ 2 w ∂ M11 + 2 M12 = β1 w + B1 2 − 2 − Qo1 (y). + 2 ∂x ∂y ∂x∂y ∂y ∂y ∂y ∂y ∂x

Recall that this boundary condition holds for x = 0 for w (0 ≤ y ≤ yc ). Similarly, one obtains ∂ ∂w ∂2 w + M11 − C2 = M2o (y), α2 ∂x ∂y ∂x∂y ∂2 F ∂w ∂2 F ∂w ∂ ∂2 ∂2 w ∂ M11 + 2 M12 = β2 w + 2 B2 2 + − 2 − Qo2 (y). ∂x ∂y ∂x∂y ∂y ∂y ∂y ∂y ∂x for x = 0, yl ≤ y ≤ l2 . We are going to calculate the first functional variation (2.67) with respect to F and compare it to zero. Integrating by parts and carrying out some transformations, the following boundary conditions for the function F for x = 0, 0 ≤ y ≤ yc are obtained ∂ ∂2 ∂ 2 F ∂2 F ∂2 F ∂w ∂2 w ∂w ∂ ε22 − ε12 + K1 2 + k2 + 2 = qo1 (y), − A 1 2 − L1 2 ∂x ∂y ∂x∂y ∂x ∂y ∂y ∂y ∂y ∂x 2 ∂F 1 ∂w = mo1 (y). −ε22 − k2 w + − L1 2 ∂y ∂y

(2.68)

Similarly, for x = 0, yl ≤ y ≤ l2 , one gets ∂ ∂ ∂2 ∂ 2 F ∂2 F ∂2 F ∂w ∂2 w ∂w ε22 − ε12 + K2 2 + κ2 + = qo2 (y), − A 2 2 − L2 2 ∂x ∂y ∂x∂y ∂x ∂y ∂x ∂y ∂y ∂y

176


−ε22 − κ2 w +

2 ∂F 1 ∂w = mo2 (y). − L2 2 ∂y ∂y

(2.69)

For example, the boundary conditions (2.68), (2.69) have the form α1 = C1 = 0, β1 = ∞, A1 = ∞, L1 = 0 w = 0, M11 = M1o (y), ∂2 F = 0, ∂y2

for x = 0, 0 ≤ y ≤ yc ,

ε22 = mo1 (y).

(2.70)

α2 = β2 = ∞, L2 = A2 = ∞, ∂w = 0, w = 0, ∂x ∂2 F ∂2 F = 0, = 0. ∂x∂y ∂y2

for x = 0, 0 ≤ y ≤ l2 ,

(2.71)

i.e. in the point y = yc a change of boundary conditions occurs. Comparing to zero the terms of functional (2.67) variation with respect to w and the point y = yc , the following equations are obtained (B1 − B2 )

∂2 w = 0, ∂y2

(C1 − C2 )

∂2 w = 0, ∂x∂y

(B1 − B2 )

∂3 w = 0. ∂y3

(2.72)

For the function F in the point y = yc the compatibility conditions for x = 0 are as follows ∂F ∂F ∂3 F + (L2 − L1 ) + (K1 − K2 ) 3 = 0, (A1 − A2 ) ∂y ∂x ∂y (K2 − K1 )

∂2 F = 0. ∂y2

(2.73)

Owing to compatibility conditions (2.72), (2.73), for the same boundary conditions to the left and to the right of the point yc , there are not any compatibility conditions in the point yc , whereas for different boundary conditions the compatibility conditions should be satisfied. These conditions will be given for the boundary ∂2 w = 0 in the point y = yc ; conditions (2.70), (2.71). Since C1 = 0, C2 0, then ∂x∂y ∂F and since α1 = 0, α2 = ∞, then ∂x = 0 in the point y = yc , x = 0. In other words, for the boundary conditions (2.70), (2.71) given to left and to right from the point y = yc , in the point yc the following compatibility conditions ∂2 w = 0, ∂x∂y should be satisfied.

∂F = 0. ∂x

(2.74)


177

2.2.5 Compatibility conditions for shallow shells equations Let us consider a rectangular isotropic shell occupying the space G+ {0 ≤ x ≤ l1 , 0 ≤ l ≤ l2 } in the plane oxy. Energy of this shell, without account of boundary conditions, has the form [212]: ⎡ ⎛ ⎞⎤ l1l2 ⎧ ⎪ ⎜⎜⎜ ∂2 w 2 ∂2 w ∂2 w ⎟⎟⎟⎥⎥⎥ ⎪ ⎨ D ⎢⎢⎢⎢ ∂2 w ∂2 w ⎜ ⎟⎟⎥⎥ − + 2(1 − ν) + − = ⎢ ⎜⎝ ⎪ ⎪ ⎩ 2 ⎣ ∂x2 ∂x∂y ∂y2 ∂x2 ∂y2 ⎠⎦ +

0 0

⎡ ⎞⎤ ⎛ 2 ⎜⎜⎜ ∂2 F 2 ∂2 F ∂2 F ⎟⎟⎟⎥⎥⎥ ∂2 F 1 ⎢⎢⎢⎢ ∂2 F ∂2 F + 2 + 2(1 + ν) ⎜⎜⎝ − 2 2 ⎟⎟⎠⎥⎥⎦ − κ1 2 w− ⎢⎣ 2 2EH ∂x ∂x∂y ∂y ∂x ∂y ∂y ⎡ ⎤ 2 2 ∂2 F 1 ⎢⎢ ∂2 F ∂w ∂2 F ∂w ∂2 F ∂w ∂w ⎥⎥⎥⎥ κ2 w 2 + ⎢⎢⎢⎣ 2 + 2 −2 ⎥− 2 ∂x ∂y ∂x∂y ∂x ∂y ⎦ ∂x ∂y ∂x l1l2 Φ(x, y)dxdy,

q(x, y)w} dxdy = 0 0

where Φ(x, y) denotes the underintegral expression. Let the considered shell (with respect to edge x = 0) is stiffly linked with a supporting rod. The rod possesses bending stiffness coefficient B+ , and rotational stiffness coefficient C + . Energy of the shell stiffened by a rib along the edge x = 0 is given by the equation 1 ¯ + =+ + 2

2 2 2 ⎫ l2 ⎧ ⎪ ⎪ ⎪ ⎪ ⎨ + ∂2 w ⎬ + ∂ w + C B ⎪ ⎪ ⎪ ⎩ ⎭ ∂x∂y ⎪ ∂y2 0

1 2

dy+ x=0

⎫ 2 l2 ⎧ ⎪ ⎪ 2 ⎪ ⎪ ⎨ + ∂F +∂ F + ∂F ∂F ⎬ + K1 2 + 2L A ⎪ ⎪ ⎪ ⎩ ⎭ ∂y ∂y ∂x ⎪ ∂y 0

dy. x=0

Consider one more shell occupying the space G− {−l1 ≤ x ≤ 0, 0 ≤ y ≤ l2 } and also stiffened along x = 0 by a rod with stiffnesses B− , C − . Its energy reads −

0 l2

=

Φ(x, y)dxdy, −l1 0

whereas the energy accounting supporting rod has the form 1 ¯ − =− + 2

2 2 ⎤ l2 ⎡⎢ 2 2 ⎥⎥⎥ ⎢⎢⎢ − ∂ w − ∂ w ⎥⎥ dy+ + C ⎢⎣ B ∂x∂y ⎦ x=0 ∂y2 0

178


⎤ 2 2 l2 ⎡⎢ 2 ⎥⎥ ⎢⎢⎢ − ∂F − ∂ F − ∂F ∂F ⎥ ⎣⎢A ∂y + K ∂y2 + 2L ∂y ∂x ⎥⎥⎦ dy, x=0

1 2

0

˜ ¯ + + ¯ − + 1 = 2

⎫ 2 l2 ⎧ ⎪ ⎪ ⎪ ⎪ ⎨ ∂w 2⎬ + βw α ⎪ ⎪ ⎪ ⎪ ⎩ ∂x ⎭ 0

dy.

(2.75)

x=0

In the above α denotes joint stiffness coefficient; β is the support stiffness co˜ F). We are efficient under the joint. Let us construct the functional J(w, F) = (w, focused on searching its first variation with respect to w, which is then compared to zero. Obtained relation is further transformed through integration by parts to yield

l2 δw J =

D 0

l1 l2 2 ∂2 w ∂ ∂ w ∂2 w ∂ ∂2 w δw +ν 2 dy − D +ν 2 + ∂x ∂x2 ∂x2 ∂y ∂x ∂y x=0 0

l1 l1 l2 2 ∂ F ∂w ∂2 F ∂w ∂2 w ∂ δw dy + − δw 2 (1 − ν) dy+ x=0 ∂y ∂x∂y ∂y2 ∂x ∂x∂y ∂y x=0 0

l2 α

+∂

3 w + ∂ w ∂ δw + β1 wδw B δw − C ∂y4 ∂x∂y2 ∂x 4

0

∂ ∂w ∂w ∂ δw δw dy+ − − ∂x x=+0 ∂x x=−0 ∂x ∂x x=+0 x=−0

l2 D 0

l2 0

+ x=0

0 ∂2 w ∂2 w ∂ δw +ν 2 dy− ∂x2 ∂y ∂x x=−l1

0 ∂2 w ∂ ∂ ∂2 w ∂2 w δw dy+ D + ν 2 + 2 (1 − ν) 2 ∂x ∂x ∂y ∂x∂y ∂y

−l1

l2 0

l2 0

0 ∂2 F ∂w ∂2 F ∂w − δw dy+ ∂y2 ∂x ∂x∂y ∂y x=−l1

−∂

3 w − ∂ w ∂ δw B δw − C ∂y4 ∂x∂y2 ∂x 4

dy + . . . x=0

where terms not related to the joint are denoted by dots. Owing to (2.75), the function w(x, y) should be continuous for x = 0, and hence also its variation δw should be continuous. Introducing the notation


179

∂2 w ∂2 w , + ν ∂x2 ∂y2 2 ∂w ∂2 w ∂ ∂ ∂2 w (1 − ν)D , − 2 + ν Q1 (x, y) = − D ∂x ∂y ∂x∂y ∂x2 ∂y2 M1 (x, y) = −D

(2.76)

one obtains l2 δw J =

∂ ∂ M1 δw − Q1 δw| x=+0 − M1 δw + Q1 δw| x=−0 + ∂x ∂x x=+0 x=−0

0

∂2 F ∂w ∂2 F ∂w − δw − 2 x=−0 ∂y ∂x ∂x∂y ∂y 2 ∂2 +∂ w B δw + 2 2 x=+0 ∂y ∂y 2 ∂ ∂ + ∂ w C δw − ∂y ∂x∂y ∂x x=+0 α

∂2 F ∂w ∂2 F ∂w − δw + 2 x=+0 ∂y ∂x ∂x∂y ∂y 2 ∂2 −∂ w B δw − 2 2 x=−0 ∂y ∂y 2 ∂ ∂ − ∂ w C δw + ∂y ∂x∂y ∂x x=−0

∂ ∂w ∂w ∂ δw δw + − − ∂x x=+0 ∂x x=−0 ∂x ∂x x=+0 x=−0 / βwδw| x=0 dy + . . . = 0.

Comparing successively to zero the coefficients standing by δw, ∂ ∂x δw x=−0 , the following relations are obtained [592]: ' ∂2 w ∂2 & [Q1 ]| x=0 = βw + 2 B+ + B− − ∂y ∂y2 x=0

∂ ∂x δw x=+0 ,

∂2 F ∂w ∂2 F ∂w + , ∂y2 ∂x x=0 ∂x∂y ∂y x=0 2 ∂w ∂ + ∂ w C α , = −M1 + ∂x x=0 ∂y ∂x∂y x=+0

∂w α ∂x

2 ∂ − ∂ w C . = −M1 − ∂y ∂x∂y x=0 x=−0

Conjugation conditions for the function w(x, y) can be given in more symmetric form, if instead of two last relations their half-sum and difference are taken. Owing to attachment of continuity condition of w for x = 0, one obtains that [w] = 0, 2 2 ∂w ∂ w 1 ∂w ∂2 w ∂2 w = D +ν 2 +D +ν 2 + α ∂x 2 ∂x2 ∂y x=+0 ∂x2 ∂y x=−0

180


2 2 ∂ ∂ + ∂ w − ∂ w C C , − ∂y ∂x∂y x=+0 ∂y ∂x∂y x=−0 2 2 2 ∂ ∂w ∂2 w ∂ + ∂ w − ∂ w C C = − , D + ν + ∂y ∂x∂y x=+0 ∂y ∂x∂y x=−0 ∂x2 ∂y2 ∂ ∂2 w ∂2 w ∂ ∂2 w D = + ν 2 + 2 (1 − ν)D ∂x ∂x2 ∂y ∂x∂y ∂y ' ∂2 w ∂2 F ∂w ∂2 F ∂w ∂2 & − (2.77) −βw − 2 B+ + B− + ∂x∂y ∂y ∂y ∂y2 ∂y2 ∂x for x = 0. Owing to notation (2.76), most interesting particular cases of conditions (2.77) are further analysed (a) α = ∞, β = B± + C ± = 0 ([592]): ∂w [w] = = [M1 ] = [Q1 ] = 0 ∂x

for x = 0.

(2.78)

Note that these conditions exhibit stiff coupling between shells. They can be rewritten also to the form 2 ∂w ∂ ∂2 w ∂w = D 2 = D 2 =0 for x = 0. [w] = ∂x ∂x ∂x ∂x (b) α = β = 0, B+ = B− = B, C + = C − = C ([592]): ∂2 w ∂ [w] = 0, C M1 − = 0, ∂y ∂x∂y x=+0 ∂2 w ∂ ∂2 ∂2 w ∂2 F ∂w C . M1 + = 0, [Q1 ] = 2 2 B 2 − 2 ∂y ∂x∂y x=−0 ∂y ∂y ∂y ∂x Note that now we deal with two same shells coupled through an ideal joint. Shell edges, in the joint neighbourhood, are strengthened by ribs, which are exhibited to both bending and rotation. (c) α = ∞, β = 0, B+ = B− = B, C + = C − = C ([592]): 2 ∂w ∂2 w ∂ w ∂ ∂2 [w] = = 0, [M1 ] = 2 C , [Q1 ] = 2 2 B 2 . ∂x ∂y ∂x∂y ∂y ply These conditions correspond to whole shell reinforced by a rib. (d) α = β = C + = C − = 0, B+ = B− = B ([592]): [w] = 0, M1 | x=±0 = 0, 2 ∂ w ∂2 F ∂w ∂ [Q1 ] = 2 2 B 2 − 2 for x = 0. ∂y ∂y ∂y ∂x Observe that these conditions differ from the (b) case. Namely, in this case shells can be considered not as stiffly coupled with reinforced rods, but us lying on them. 2


181

(e) α = ∞, β = C + = C − = 0, B+ = B− = B ([592]): 2 2 ∂w ∂ w ∂w ∂2 = D 2 = 0, [Q1 ] = 2 2 B 2 . [w] = ∂x ∂x ∂y ∂y In this case the rod plays a role of support for stiffly coupled shells (whole shell). In what follows we are going to derive conditions for F along linking line x = 0 of two shells. Again, we start with computation of first variation of the functional J with respect to F. Integrating by parts and remaining the terms related to the rod x = 0, one gets l2 δF J = −

1 EH

0

l1 2 l2 ∂2 F ∂ ∂ F ∂2 F ∂ 1 δF −ν 2 dy + + EH ∂x ∂x2 ∂y ∂x ∂x2 x=0 0

⎫l1 ⎛ 2 ⎞ l1 l2 ⎧ ⎪ ⎪ ⎪ ⎪ ∂2 F 1 ∂w ⎟⎟⎟⎟ ∂ ⎬ ⎨⎜⎜⎜⎜ (2 + ν) 2 δF dy + ⎪ dy+ ⎟⎠ δF ⎪ ⎜⎝−k2 w + ⎪ ⎪ ⎭ ⎩ 2 ∂y ∂x ∂y x=0 x=0

0

l2 0

l1 l2 2 2 ∂w ∂w ∂2 w +∂ F + ∂ F + δF dy+ k2 δF −A dy + − L ∂x ∂x ∂y2 ∂x∂y ∂y2 x=0 x=0 0

l2

+ ∂F

∂ δF L ∂y ∂x

0

ν

l2

+∂

F K δF ∂y4

dy + x=0

0 ∂2 F ∂ δF + ∂y2 ∂x x=−l1

4

0

l1

1 EH

l2 dy − x=0

1 EH

∂2 F − ∂x2

0

0 ∂ ∂2 F ∂2 F δF + (2 + ν) dy+ ∂x ∂x2 ∂y2 x=−l1

0

⎫0 2⎞ 0 l2 ⎪ ⎪ ⎪ ⎪ ∂w ∂w ∂2 w 1 ∂w ⎟⎟⎟⎟ ∂ ⎬ ⎨⎜⎜⎜⎜ + k2 δF dy+ ⎟ δF ⎪ dy + ⎜−k w + ⎪ ⎪ ⎭ ⎩⎝ 2 2 ∂y ⎠∂x ⎪ ∂x ∂x ∂y2 x=−l1

l2⎧⎛ 0

x=−l1

l2

∂2 F ∂2 F δF −A1 2 − L− ∂x∂y ∂y

0

l2 0

K−

0

l2 dy + x=0

∂4 F δF ∂y4

− ∂F

∂ δF L ∂y ∂x

0

dy+ x=0

dy + . . . , x=0

where dots denote terms not referred to the rod x = 0. Assuming F as continuous function, also δF is continuous one. ∂ ∂ δF x=+0 , ∂x δF x=−0 to zero, the followComparing the terms standing by δF, ∂x ing equations are obtained

182


∂w ∂2 w 1 ∂ ∂2 F ∂2 + ∂w + − + (2 + ν) 2 − k2 + EH ∂x ∂x2 ∂x ∂x ∂y2 x=+0 ∂y x=+0 ∂w ∂2 w ∂2 F 1 ∂ ∂2 F − ∂w + + (2 + ν) 2 + k2 + EH ∂x ∂x2 ∂x ∂x ∂y2 x=−0 ∂y x=−0 2 2 4 +∂ F + ∂ F + ∂ F −A −L + K + ∂x∂y x=+0 ∂y2 ∂y4 x=+0 2 2 4 −∂ F − ∂ F − ∂ F −A −L + K = 0, ∂x∂y x=−0 ∂y2 ∂y4 x=−0 1 EH 1 − EH

∂2 F ∂2 F −ν ∂x2 ∂y2

+ x=+0

⎧ 2 ⎫ ⎪ ⎪ ⎪ 1 ∂w ⎪ ⎨ + ⎬ w− k ⎪ ⎪ 2 ⎪ ⎩ ⎭ 2 ∂y ⎪

⎧ 2 ⎫ ⎪ ⎪ ⎪ ∂2 F ∂2 F 1 ∂w ⎪ ⎨ − ⎬ −ν 2 + ⎪ −k2 w + ⎪ ⎪ 2 ⎩ ⎭ 2 ∂y ⎪ ∂x ∂y x=−0

+ L = 0, ∂y x=+0

+ ∂F

x=+0

+ L x=−0

− ∂F

∂y

= 0.

(2.79)

x=−0

Assuming that k2+ = k2− = k2 (i.e. both shells have the same curvature k2 ) the first relation can be rewritten to the following one 4 ∂w ∂w ∂w ∂2 F 1 ∂2 F +∂ F − ν − κ − + K − − 2 1 EH ∂x2 ∂x x=0 ∂x ∂y x=0 ∂y2 x=0 ∂y4 2 2 4 2 2 +∂ F + ∂ F −∂ F −∂ F − ∂ F A −L + K −A −L = 0. ∂x∂y x=+0 ∂x∂y x=−0 ∂y2 ∂y4 ∂y2 Two last relations in (2.79) are combined, and for k2+ = k2− = k2 one gets ⎡ 2 ⎤ 1 ∂2 F ∂2 F 1 ⎢⎢⎢⎢ ∂w ⎥⎥⎥⎥ − ν 2 + k2 [w]| x=0 − ⎣⎢ ⎥ = EH ∂x2 2 ∂y ⎦ ∂y x=0 x=0 − ∂F − L − L , ∂y x=+0 ∂y x=−0 2 ∂ F 1 ∂2 F 1 ∂2 F ∂2 F − ν + − ν + EH ∂x2 ∂y2 x=+0 EH ∂x2 ∂y2 x=−0 ⎧ ⎧ 2 ⎫ 2 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ 1 ∂w ⎪ 1 ∂w ⎪ ⎨ ⎬ ⎬ ⎨ +⎪ + k2 w − k w− ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ ⎭ ⎭ ⎩ 2 2 ∂y 2 ∂y ⎪ x=+0 x=−0 ∂F ∂F L+ + L− = 0. ∂y x=+0 ∂y x=−0

+ ∂F

(2.80)

Computing various particular cases of conditions (2.79), (2.80), the following coupling conditions for the function F are obtained


183

(a) K + = K − = A+ = A− = L+ = L− = 0 2 ∂ ∂ F 1 ∂2 F ∂w + k + (2 + ν) = 0, 2 EH ∂x ∂x2 ∂x x=±0 ∂y2 1 ∂2 F ∂2 F − ν = 0. EH ∂x2 ∂y2 x=±0 (b) A+ = A− = ∞, L+ = L− = 0 ∂2 F = 0, ∂y2 x=±0

∂2 F ∂2 F −ν 2 2 ∂x ∂y

(c) A+ = A− = ∞, L+ = L− = ∞ ∂2 F = 0, ∂y2 x=±0

= 0. x=±0

∂2 F = 0. ∂x∂y x=±0

Observe that during derivation of coupling conditions for shallow shells, an action of external forces is no accounted, and the obtained coupling conditions are homogeneous ones. Account of external forces will give non-homogeneous coupling conditions. 2.2.6 Temperature field In order to study a stress-strain state and stability loss of thin plates working in conditions of non-uniform heating, a temperature field should be defined. Both theoretical background and methods devoted to solution of the heat transfer problems of thin walled elements are required. In majority of references various methods are used to reduce 3D heat transfer equation into 2D one for plates and shells, since then the problem is essentially simplified. In this section we are going to solve directly 3D heat transfer equation to define temperature filed occurred in plates and shells. For shells made from isotropic material the stationary heat transfer equation follows W0 ∂T ∂2 T =− , (2.81) ∆T + 2 + 2k ∂z λ ∂z where k = 12 (k1 + k2 ) - average curvature of the shell surface; λ - heat transfer coefficient of an isotropic body; ∆ - Laplace operator (2.19). In the case of a plate, the equation (2.81) reads W0 ∂2 T ∂2 T ∂2 T + 2 + 2 =− . 2 λ ∂x ∂y ∂z

(2.82)

It should be emphasized that many technical materials have a heat transfer coefficient depended on heat stream direction. Recall that since the equations (2.81),

184


(2.82) are obtained for scalar heat transfer coefficient, they are not valid for such materials. For an anisotropic material, the heat transfer coefficient represents the second order tensor. In this case ∂T ∂T ∂T + λ xy + λ xz i+ div(λgradT ) = div λ xx ∂x ∂y ∂z ∂T ∂T ∂T ∂T ∂T ∂T + λyy + λyz + λzy + λzz j + λzx k = −W0 , (2.83) λyx ∂x ∂y ∂z ∂x ∂y ∂z where: i, j, k are eigenvectors of a rectangular coordinates. If components of a heat transfer tensor do not depend on coordinates, the expression (2.83) is reduced to the form div(λgradT ) = λ xx (λ xy + λyx )

∂2 T ∂2 T ∂2 T + λyy 2 + λzz 2 + 2 ∂x ∂y ∂z

∂2 T ∂2 T ∂2 T + (λ xz + λzx ) + (λzy + λyz ) = −W0 . ∂x∂y ∂x∂z ∂z∂y

(2.84)

In order to solve the stationary heat transfer equations, the boundary conditions should be attached. First (I), second (II), third (III) and fourth (IV) boundary conditions are mainly applied [315]: First order boundary conditions. A temperature distribution on the body surface S as the coordinates function is applied T s = g(x, y, z),

x, y, z ∈ S .

(2.85)

Heating and cooling processes of the body for a given temperature variation on its boundary or for intensive heat exchange on its surface, when the surface temperature, are representive examples. However, these conditions are rather rarely to be met in practice, and (2.85) are applied for purely mathematical purposes and errors estimating only. Second order boundary conditions. Heat stream distribution is assumed on a body surface as the function of coordinates W s = θ(x, y, z),

x, y, z ∈ S .

Owing to the Fourier rule, the condition (2.86) reads ∂T = θ(x, y, z), x, y, z ∈ S . −λ ∂n S

(2.86)

(2.87)

In the particular case, when a density of a heat stream on the body surface is constant, then W s = W0 = const. Such heat transfer conditions can be realized during bodies heating through high temperature sources, and when heat transfer occurs through radiation within the


185

Stefan-Boltzman rule, assuming that a body temperature is significantly less than a temperature of a radiating surface. In the case ∂T = 0, Ws = − ∂n S a so called heat isolation takes place. Third order boundary condition. On the space body boundaries a dependence of heat stream density, caused by heat transfer between surface body temperature T, and surrounding medium temperature T 0 , is given. In the case of a body cooling (T S > T 0 ), one gets W s = ξ(T S − T 0 ),

(2.88)

where ξ is proportionality coefficient, known as heat transfer coefficient and measured as W/(m2 grad). Equivalently ∂T = ξ(T S − T 0 ). (2.89) ∂n The equation (2.89) is the analytical expression governing third order boundary condition, which is widely used in heat transfer investigation in solid bodies surrounded by a fluid stream. Observe that third order boundary conditions yields also first and second order boundary conditions as its particular case. If ξ/a → ∞ (ξ → ∞ for λ = const or λ → 0 for ξ = const), then the first order boundary conditions are yielded 1 ∂T = 0, T S − T 0 = lim ξ/λ→∞ ξ/λ ∂n S −λ

i.e. T S = T 0 . If ξ → 0, than a particular case of the second order boundary conditions is obtained ∂T = 0. −λ ∂n s Fourth order boundary conditions. They govern heat transfer between a body surface and surrounding medium (convection between a body and a fluid) or between two contacting solid bodies, where a temperature of contacting surfaces is the same T 1S = T 2S , −λ1

∂T 1 ∂n

= −λ2 S

(2.90) ∂T 2 ∂n

.

(2.91)

S

Equation (2.90) exhibits continuity condition of the temperature filed, whereas equation (2.91) governs energy conservation rule on contacting bodies surface.

186


2.3 Solution of 3D Stationary Heat Transfer Equation 2.3.1 The method In order to solve a stationary heat transfer equation, the finite difference method is applied. Process of solution of PDEs through finite difference method includes two fundamental steps: (i) a transformation of differential equations and boundary conditions into difference (mesh) system of equations; (ii) solution of the obtained difference equations. On the first step, a question how to estimate an accuracy of applied approximation appears. In this chapter the 4th order approximation is applied. Owing to such mesh, the difference equations order is decreased, which plays an important role during solution to 3D equation. The second step is focused on the following problem: which method (direct or iterational) should be used to solve the system of difference equations. Here an iterational method is applied, since the system of difference equation is of high order, and a direct approach can not be used. Owing to occurrence of various iterational methods, an important question arises: how to choose the most suitable solution method for a given problem. In practice, a method requiring less computation time with simultaneous conservation of a given accuracy, occupying less amount of computer memory, and simple in realization is recommended. Owing to experimental comparison of various methods, the mostly suitable one (with respect to arithmetic calculus members or required computer time to solve the whole problem) is chosen (see for example [367]). Let us reconsider briefly the problem of efficiency of various iterational methods in application to Dirichlet problem for Laplace and Poisson’s equations in the following spaces: square and T-shape space (Figure 2.1). Iterational methods of successive upper Seidel’s relaxation, triangle method, triangle method with Chebyshev’s acceleration, implicit method of variable directions are considered. Two ways of partial derivatives approximation are studied: fifth and ninth order schemes, with the corresponding approximation o(h2 ) and o(h4 ). A comparison of the mentioned methods with respect to convergence velocity, computer memory amount and to their complexity is carried out, and the problem of parameters accelerating iterational methods convergence is also addressed. For the Laplace equation ∆T = 0, the Dirichlet problem with boundary conditions on the unit square sides ⎧ 3x1 ⎪ e , ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ cos 3x2 , T |Γ = ⎪ ⎪ ⎪ e3x1 cos 3, ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ e3 cos 3x , 2

x2 = 0, 0 ≤ x1 ≤ 1 x1 = 0, 0 ≤ x2 ≤ 1 x2 = 1, 0 ≤ x1 ≤ 1

.

(2.92)

x1 = 1, 0 ≤ x2 ≤ 1

is solved. Recall that the exact solution to this problem is T (x1 , x2 ) = e3x1 cos 3x2 .

2.3 Solution of 3D Stationary Heat Transfer Equation x2

x2

1.0

b

Γ4 Γ5

Γ3

(1)

x2

Γ2

Γ6 Γ7

Γ1

1.0

0

187

Γ8

x1

(1) x1

0

(2)

x1

a

x1

Figure 2.1. Square and T-shirt spaces.

For the Poisson equation ∆T = − f , the Dirichlet problem within the T-shape space and with respect to the boundary Γ is solved (Γ is composed of straight intervals ΓK ). The boundary conditions have the form T |Γ1 ,Γ2 = ϕ0 ,

T |Γ4 = ϕ1 ,

T |Γ3 ,Γ5 = K + (ϕ1 − K)

x2 − x2(1)

, (1)

x2(2) − x2

T |Γ8 = ϕ0 + ϕ2

T |Γ6 ,Γ7 = ϕ2 , K=

x1 − x1(1) x2(2) − x2(1)

,

ϕ0 for Γ3 ϕ2 for Γ5 (2.93)

where: ϕ0 , ϕ1 , ϕ2 are constant quantities. Let us derive a system of difference equations, approximating the equation ∆T = − f with the boundary conditions T |Γ = ϕ(x, y) in the form AT = F,

(2.94)

where: A = K + D + N; D - diagonal; K - upper triangle; N - lower triangle matrices. The method of top relaxation applied to (2.94) yields [463]: $ % (2.95) DT (n) = BT (n−1) − ω KT (n) + (D + N)T (n−1) − F . For ω = 1 this method is transformed to Seidel one. Owing to five-points approximation, the difference equations have the form

(1) (n) (2) (n) (3) (n−1) (4) (n−1) (n−1) , (2.96) T i(n) j = ω ai j T i−1 j +ai j T i j−1 +ai j T i+1 j +ai j T i j+1 + f + (1 − ω)T i j h2

h2

(3) (2) (4) 2 1 where: a(1) i j = ai j = 2(h21 +h22 ) , ai j = ai j = 2(h21 +h22 ) , h1 , h2 - are mesh cell length in direction of x1 and x2 , correspondingly. For nine-points approximation, one gets

(1) (n) (n) (n) (3) (n−1) (4) (n−1) (5) (n−1) T i(n) j = ω ai j T i−1 j + ai j T i j−1 + ai j T i+1 j + ai j T i j+1 + ai j T i−1 j+1 +

188


(n) (7) (n) (8) (n−1) (n−1) a(6) , i j T i−1 j−1 + ai j T i+1 j−1 + ai j T i+1 j+1 + f + (1 − ω)T i j where: (3) a(1) i j = ai j =

1 5h22 − h21 , 10 h21 + h22

(4) a(2) i j = ai j =

(2.97)

1 5h21 − h22 , 10 h21 + h22

1 . 20 Iterational multiplier is obtained through well known formula [463]: (6) (7) (8) a(5) i j = ai j = ai j = ai j =

ωopt =

2 , ) 1 + 1 − λ21

(2.98)

where: λ1 is a spectral matrix radius, corresponding to Seidel method. Recall that λ is known for both five- and nine-points approximation, i.e. λ(5) 1 = λ(9) 1

h22 2(h21 + h22 )

cos

h21 πh1 πh2 + , cos a b 2(h21 + h22 )

⎛ ⎞ πh2 5h22 − h21 1 ⎜⎜⎜ πh1 5h21 − h22 πh2 ⎟⎟⎟ πh1 ⎟⎠ , cos + 2 + 2 = ⎜⎝cos cos cos 5 a b a b h1 + h22 h1 + h22

(2.99)

(2.100)

where: a, b are length of rectangular sides (0 ≤ x1 ≤ a, 0 ≤ x2 ≤ b). In order to get ωopt for an arbitrary space, and approximation value λ1 is used in (2.98). There exist a relation [463], which couples spectral radiuses of both matrix and iteration process (2.95) via µ1 , λ1 and ω of the form (µ1 + ω − 1)2 = λ21 . ω2 µ1

(2.101)

Therefore, one may define an approximated relaxation multiplier ωopt via the formula 2 . (2.102) ωm+1 = + 1+

1−

µ(m) 1 +ωm −1 2 µ(m) 1 ωm

The approximated value µ(m) 1 is yielded by the Lusternik method [463]: (( ( (m+1) (m+1) (m) ( Ti j − T i(m) ( ( i j T − T j ( ( , ≈ = lim µ(m) 1 m→∞ ( (T (m) − T (m−1) (( i j T (m) − T (m−1) ij ij where: m denotes the minimal number, for which the inequality holds (m) µ1 (m−1) − 1 ≤ εµ , µ1

(2.103)

(2.104)

2.3 Solution of 3D Stationary Heat Transfer Equation

189

and εµ is a small given quantity, ω0 is a given value (for example, ω0 = 1). Carrying is defined. However, the velocity out m1 iterations through Seidel method, µ(m−1) 1 convergence essentially increases, if one takes ω0 either as 0.1 or as 0.2. The latter observation can be achieved applying the so called equivalent rectangular method. Owing to this method, the spectral radius of a matrix in Seidel method is sought for the equivalent rectangular, i.e. with respect to surface equal to given space. It can be obtained through diameter of the largest circle, which can be drawn within the space boundaries (a circle should not either intersect a boundary curve nor include an external space); its length is yielded by dividing the space area via width. Besides, ωopt can be taken as the following asymptotic value ωopt

2πh1 h2 =2− ) h21 + h22

+

1 1 + a2 b2

(2.105)

(for five-points scheme) 0 ωopt = 2 −

6πh1 h22

1 1 , + (h21 + 10h22 )(h21 + h22 )(20h22 − h21 ) a2 b2 h21 + 25h22

(2.106)

(for nine-points scheme), where: h1 , h2 , a, b are the sam as in (2.99). It is worth noticing that the iteration number will be smaller, if ωopt is given either directly through the formulas (2.105), (2.106) or through equivalent rectangular method. It follows form (2.104) that iteration number h depends on εµ . Numerical experiment shows, that for a square the optimal interval for (i) εµ is [0.001 ÷ 0.005] for h1 = h2 = 0.625 for five-points approximation; (ii) εµ ∈ [0.005 ÷ 0.01] for nine-points approximation; (iii) εµ ∈ [0.001; 0.01] for five-points approximation of T-shape space. Notice that εν decreases with a step decrease. Among all explicit methods of variable directions, good estimation of convergence speed and economical realization of spatial step exhibit triangle methods, since the triangular matrices are simply transformed. In addition, each spatial step is realized via explicit formulas. For example, if the initial matrix (2.94) is divided into two triangular matrices 1 (1) (2) (A1 T )i j = ai j T i−1 j − T i j + ai j T i j−1 , 2 1 (4) (A2 T )i j = a(3) (2.107) i j T i+1 j − T i j + ai j T i j+1 , 2 then the corresponding explicit method of variable directions is yielded by formulas ! 1 1 1 (n) (n+ 1 ) (n+ 12 ) (2) (n+ 2 ) (3) (n) + τ a(1) Ti j 2 = T τ ij i j T i−1 j + ai j T i j−1 + ai j T i+1 j + 1+ 2

190

2 Stability of Rectangular Shells within Temperature Field (n) a(4) i j T i j+1

1 − T i(n) 2 j

2(n+1) = T ij (n+ 12 ) a(2) i j T i j−1

1 1+

,

i = 1, 2, . . . , N1 − 1, j = 1, 2, . . . , N2 − 1,

1

(n+ 12 )

Ti j

τ 2

1 (n+ 1 ) − Ti j 2 2

(2.108)

! 1 (1) (n+ 2 ) (4) 2(n+1) 2(n+1) + τ a(3) i j T i+1 j + ai j T i−1 j + ai j T i j+1 +

,

i = N1 − 1, . . . , 1; j = N2 − 1, . . . , 1.

(2.109)

Numerical experiment shows, that multiplier τ increasing a convergence in (2.108), (2.109) is equal to 2 . τcp = h1 + h2 In order to increase a speed of triangular method convergence the Chebyshew convergence acceleration is applied. Iterational process is formulated in the following manner: iterational formulas of triangular method (2.108), (2.109) are supplemented by the following formula

(n+1) 2 = T i(n) − T i(n) (2.110) T i(n+1) j j + λn T i j j , where: λn are certain parameters improving the iterational process convergence αn =

−1 2n − 1 M 2n − 1 M 1 + cos π 1− 1 + cos π , n = 1, . . . , N. (2.111) 2 2N 2 2N

Quantity M (maximal matrix eigenvalue of iterational process) is usually found approximately. For this aim, for instance, the Lusternik algorithm (2.103) can be used. Owing to n1 iterations through formulas (2.108), (2.109), i.e. without Chebyshew acceleration, the following formula is obtained λ1(n1 )

⎞⎛ ⎞ ⎛ ⎟ ⎜⎜ ⎟−1 ⎜⎜⎜ ⎟ ⎟ ⎜ (n1 +1) (n1 ) ⎟ (n1 ) (n1 −1) ⎟ ⎟ ⎜ ⎜ = ⎜⎜⎝ T i j − T i j ⎟⎟⎠ ⎜⎜⎝ T i j − T i j ⎟⎟⎟⎠ , ij

(2.112)

ij

which tends to λ1 = M for n1 → ∞. If it changes from iteration to iteration slightly, this quantity can be taken as approximating value λ1 , i.e. as M. The λn is calculated, and iterations are realized through Chebyshew acceleration. Further, let us focus on consideration of the Pisman-Pakford method. In this approach a transition from one iteration to another one is realized via two steps [594]:

1 (n+ 12 ) (n) (n+ 12 ) T + A T A , T T (n+ 2 ) = T (n) − τ(1) Γ = ϕ, 1 2 n

1 1 (2.113) A1 T (n+ 2 ) + A2 T (n+1) , T (n+1) Γ = ϕ, T (n+1) = T (n+ 2 ) − τ(2) n (2) where: τ(1) n and τn are the parameters accelerating an iterational process. The first formula in (2.113) is implicit one with respect to a horizontal direction, whereas the second one is implicit with respect to vertical direction. The following formulas hold


−Aα T = Λα T = T xα xα ,

191

α = 1, 2,

1 (0) (3) (−A1 T )i j = a(1) i j T i−1 j − ai j T i j + ai j T i+1 j , 2 1 (0) (4) (−A2 T )i j = a(2) i j T i j−1 − ai j T i j + ai j T i j+1 , 2 where: a(k) i j are defined in (2.97). In the case of high order accuracy scheme A = A1 + A2 −

h21 + h22 A1 A2 . 12

Fundamental idea of the method of variable directions consists of reduction of transition from one to another iteration yielding a solution along rows and columns of one-dimensional problems, which are solved via iterational method. Algorithm of solution of algebraic equations (2.113) using this method is reduced to successive solution along rows of the following equations 1

1

(n+ 2 ) = Fn , T (n+ 2 ) = τ(1) n A1 T

where: (n) Fn = T (n) − τ(1) n A2 T ,

(2.114)

1 T (n+ 2 ) = ϕ, Γ

and along columns of the following equations (n+1) T (n+1) + τ(2) = Fn+ 12 , n A2 T

where:

1

(2.115)

1

(n+ 2 ) Fn+ 12 = T (n+ 2 ) − τ(2) . n A1 T

For the scheme of higher order approximation formulas analogous to (2.114), (2.115), can be rewritten to the following form [591]: ⎞ ⎞ ⎛ ⎛ ⎜⎜⎜ ⎜⎜⎜ (1) h21 ⎟⎟⎟ ⎟⎟⎟ (n+ 1 ) (2.116) ⎝⎜E + ⎝⎜τn − ⎠⎟ A1 ⎠⎟ T 2 = Φn , 12 where:

⎞ ⎞ ⎛ ⎛ h2 + h22 ⎜⎜⎜ ⎜⎜⎜ (1) h22 ⎟⎟⎟ ⎟⎟⎟ (n) 1 A2 ϕ, Φn = ⎝⎜E − ⎜⎝τn + ⎟⎠ A2 ⎟⎠ T , T (n+ 2 ) = ϕ − 1 Γ 12 12 ⎞ ⎞ ⎛ ⎛ ⎜⎜⎜ ⎜⎜⎜ (2) h22 ⎟⎟⎟ ⎟⎟⎟ (n+1) = Φn+ 12 , ⎝⎜E + ⎝⎜τn − ⎠⎟ A2 ⎟⎠ T 12 ⎞ ⎞ ⎛ ⎛ h21 ⎟⎟⎟ ⎟⎟⎟ (n+ 1 ) ⎜⎜ ⎜⎜ (n+ 12 ) 2 , ⎟ ⎟ Φn+ 12 = ⎜⎜⎝E − ⎜⎜⎝τ(2) + T A T Γ = ϕ. ⎠ 1⎠ n 12

(2.117)

Acceleration of iterational process convergence is achieved via the appropriate (2) choice of the parameters τ(1) n and τn . Following [591], a computation of optimal

192


(2) parameters τ(1) n and τn for the problem (2.93), (2.115) is reduced to the following formulas: S ωn + τ S ωn − τ , τ(2) , (2.118) τ(1) n = n = 1 + ωn p 1 − ωn p

where: S =τ+ τ= ωn =

1− p ∆

,

p=

∆1 − ∆2 + (∆1 + ∆2 )p

κ−ξ , κ+ξ ,

ξ=

(1 + 2θ)(1 + θσ ) , 2θσ/2 (1 + θ1−σ + θ1+σ )

σ=

ν≈

2∆1 ∆2

1 4 4 ln ln , π2 ε η

κ= 0

(∆1 − δ1 )∆2 (∆2 + δ1 )∆1

,

(∆1 − δ1 )(∆2 − δ2 ) (∆1 + δ2 )(∆2 + δ1 )

2n − 1 , 2n 1 2 1 θ= η 1 + η2 , 16 2

,

n = 1, 2, . . . , ν, η=

1−ξ . 1+ξ

In the above δ1 , δ2 are the minimal eigenvalues of the operators A1 and A2 , respectively; A1 , A2 are the maximal eigenvalues of these operators. They are assumed to be known. Finally, ε is the required accuracy of the iterational process. For a scheme of higher order accuracy (2.116), (2.117), computations of iterational parameters can be carried out through formulas (2.118), substituting δα , ∆α by δ˜ α , ∆˜ α . The latter are coupled by relations δ˜ α =

δα , 1 − κα δα

∆˜ α =

∆α 1 − κα ∆α

,

(2.119)

where:

h2α . 12 For a rectangular space, a higher order accuracy scheme with an optimal choice of parameter series is realized through formulas (2.118), (2.119). Results are given in Table 2.1. They show, that the method is fastly convergent, since after 12 iteration the required accuracy 10−6 is achieved. However, owing to comparison to explicit methods, i.e. top relaxation, triangular with Chebyshev’s acceleration, machine time required for one iteration computation using the scheme (2.116)–(2.119) is two times larger. Besides, for the scheme (2.116)–(2.119), the memory volume required for storα = 1, 2,

(n+ 1 )

κα =

2 is equal to 2N, where N denotes number of points in the space. age of T i(n) j , Ti j In all considered iterational methods the computation is continued until the following inequality is achieved T (n) − T (n−1) ij ij < εum , (2.120) maxi, j (n) T ij

where: εum is given small quantity.


193

In Table 2.2 the discussed methods for five-points approximation in application to the problem (2.93) are compared. In Table 2.3 dependencies of iteration number for upper relaxation and Seidel methods using five- and nine-points approximations reported. Nine-points approximation has higher convergence velocity than five-points one. Besides, nine-points approximation decreases an order of difference equations, and hence a shorter machine time is required to achieve a given accuracy in comparison to five-points scheme. Table 2.1. Comparison of computational results using various computational schemes and applying formulas (2.147) and (2.119) (h = h1 = h2 = 0.0625, εum = 10−6 ; squared space). Method Scheme

Variable directions Seidel Upper relaxation triangular implicit fivenine- fiveninefiveninepoints points points points points points

Iterational (2) parameters 1 1 1.65 1.65 20 τ(1) n , τn Memory storage N N N N N 2N Iterations number 247 225 89 49 80 12 Error in point (0.5; 0.5) 0.001 0.0006 0.0005 0.00009 0.0001 0.00003 Convergence π2 h2 speed 1.2π2 h2 2πh 2.09πh wπh 2

Table 2.2. Computational efficiency of various methods applied to problem (2.93) (εum = 10−4 , h1 = h2 = h = 0.25, T-shape space). Method Seidel Upper relaxation Triangular With Chebyshev’s acceleration

Convergence Iterations Iterational velocity number Storage parameters π2 h2 2

300

N

ω=1

2πh

46

N

ω0 = 1.5

2πh

55

N

τcp

√ 2π h

30

N

λn

In Table 2.1 Seidel, upper relaxation and variable directions methods with fiveand nine-points approximation are compared on example of the problem (2.92) for a squared space.

194


Table 2.3. Dependence of iteration number for upper relaxation and Seidel’s methods using five (n(5) ) and nine (n(9) ) points approximations (h1 = h2 = h = 0.0625, εum = 10−6 , squared space). ω 1 1.58 n(9) 225 68 n(5) 247 100

1.61 60 96

1.65 49 89

1.675 50 94

Owing to Table 2.1, for the squared space the method of variable directions with the scheme of higher order accuracy is mostly effective. Then upper relaxation with nine-points approximation follows. The latter one can be applied for an arbitrary space, since an optimal iterational parameter is computed in the computational process. Since a priori knowledge of maximal eigenvalue is not required, this method seems to be the most universal. Besides, the method of over relaxation is distinguished by its simplicity and requires a minimal memory storage (one working field N). To conclude, on a basis of carried out numerical experiments top relaxation method is the mostly economical one, and hence it is further applied to solve the 3D heat transfer equation. 2.3.2 Construction of difference schemes As it has been mentioned already in section 2.1.1, difference scheme of fourth order accuracy allows to take more larger mesh in comparison to the schemes o(h2 ). This yields essential decrease of an order of difference equations system, which is essentially important for solutions of multidimensional problems. In the reference [592] the schemes at 4th and 6th accuracy order for the 2D Poisson equation are constructed, as well as the 4th order scheme for the equation ∂2 T ∂2 T ∂2 T + 2a + = − f (x1 , x2 ), ∂x1 ∂x2 ∂x22 ∂x12 where |a| < 1 is constant number, is given. In what follows the 4th order accuracy differential scheme for the multidimensional equation n n ∂2 T ∂2 T LT ≡ + 2 K = − f (x), i j ∂xi ∂x j ∂xi2 i=1 i, j=1 x = (x1 , x2 , . . . , xn ) ∈ Gn {0 ≤ xα ≤ lα , α = 1, 2, . . . , n} .

(2.121)

is constructed. Consider the difference operator [592] Λv ≡

n i=1

v xi xi +

n

$ % Ki j v xi x j + v xi x j .

(2.122)

i, j=1

Difference operators in (2.122) are splitted into series with power h (h - mesh step) using the Taylor series


∂2 v h2 ∂4 v + + o(h4 ), ∂xi2 12 ∂xi4

v xi xi = v xi x j =

195

∂2 v h ∂3 v h2 ∂4 v h2 ∂4 v − + + + ∂xi ∂x j 2 ∂xi2 ∂x j 6 ∂xi3 ∂x j 6 ∂xi ∂x3j h ∂3 v h2 ∂4 v − + o(h3 ), 2 ∂xi ∂x2j 4 ∂xi2 ∂x2j

v xi x j =

∂2 v h ∂3 v h ∂3 v h2 ∂4 v + − + + 2 3 ∂xi ∂x j 2 ∂xi ∂x j 2 ∂xi ∂x j 6 ∂xi3 ∂x j h2 ∂4 v h2 ∂4 v − + o(h3 ). 6 ∂xi ∂x3j 4 ∂xi2 ∂x2j

Substituting these series in (2.122), one gets ⎞ ⎡ n ⎛ 2 n ⎜⎜⎜ ∂ v h2 ∂4 v ⎟⎟⎟ ⎢⎢⎢ ∂2 v h2 ∂4 v ⎜ ⎟ ⎢ Λv ≡ K + + + 2 ⎝ 2+ ⎠ ⎣ i j 12 ∂xi4 ∂xi ∂x j 3 ∂xi3 ∂x j ∂xi i=1 i, j=1 ⎤ n n h2 ∂4 v ⎥⎥⎥⎥ h2 ∂4 v ∂2 v ∂2 v 4 − ) = + 2 K + + o(h ⎥ i j 3 ∂xi ∂x3j 2 ∂xi2 ∂x2j ⎦ ∂xi ∂x j ∂xi2 i=1 i=1 i j

⎛ n ⎡ ⎤⎞ n 4 4 ⎢⎢⎢ ∂4 v ⎥⎥⎟⎟⎟ v v h2 ⎜⎜⎜⎜ ∂4 v ∂ ∂ ⎜⎜⎝ + Ki j ⎢⎢⎣4 3 +4 − 6 2 2 ⎥⎥⎥⎦⎟⎟⎟⎠ + o(h4 ) = 4 3 12 i=1 ∂xi i, j=1 ∂xi ∂x j ∂xi ∂x j ∂xi ∂x j Lv +

n ∂4 v h2 2 h2

L v− + o(h4 ) = 1 + 3Ki j + 2Ki2j 12 6 i=1 ∂xi2 ∂x2j i j

−f −

n ∂4 v h2

h2 Lf − + o(h4 ). 1 + 3Ki j + 2Ki2j 12 6 i=1 ∂xi2 ∂x2j i j

In what follows, the difference equation Λ v ≡

n i=1

v xi xi +

n $

Ki j v xi x j + v xi x j +

i=1 i j

h2

2 1 + 3Ki j + 2Ki j v xi xi x j x j = −ϕ , 6 where ϕ = f +

h2 L f, 12

(2.123)

(2.124)

196


approximates the equation (2.121) with error o(h4 ). If ki j ≡ 0 for i j, the the equation (2.121) is transformed into the multidimensional Poisson’s equation, and the difference equation n n h2 h2

(2.125) v xi xi + vx x x x = − f + L f Λv≡ 6 i, j=1 i i j j 12 i=1 approximates multidimensional Poisson’s equation. This difference equation (2.125) coincides on the squared mesh with a difference scheme given in reference [592] for n-dimensional Poisson’s equation, i.e. it will approximate the equation (2.82). Observe that the general type equation n

Kii

i=1

∂2 T ∂x2i

+2

n i, j=1

Ki j

∂2 T = − f (x), ∂xi x j

(2.126)

where ki j is constant, is transformed to the form(2.121) via the following variables transformation 3 xi = Kii xi . Therefore, the difference equation (2.124) is approximating the equation (2.84) with the error o(h4 ) after the variables transformation 3 3 3 x1 = K11 x1 , x2 = K22 x2 , x3 = K33 x3 . Let us construct a difference scheme for the equation LT ≡

n

Li T = − f (x),

i=1

x = (x1 , x2 , . . . , xn ) ∈ Gn {0 ≤ xα ≤ lα , α = 1, 2, . . . , n}, Li T =

∂2 T ∂T + Ki (x) . ∂xi ∂xi2

(2.127)

It is assumed that ki (x), f (x) ∈ C (4) (Gn ), T (x) ∈ C (6) (Gn ). Consider the difference operator Λv =

n

◦

Λi v,

◦

where Λi v = v xi xi + ai x x◦i , ai = Ki (x).

i=1

Applying the Taylor series with respect to step h of the form v xi xi =

∂2 v h2 ∂4 v + + o(h4 ), ∂xi2 12 ∂xi4

v x◦i =

∂v h2 ∂3 v + + o(h4 ). ∂xi 6 ∂xi3

(2.128)


Then

⎤ n n ⎡ h2 ⎢⎢⎢ ∂4 v ∂2 v ∂v ∂3 v ⎥⎥⎥ ⎢⎣ 4 + 2ai 3 ⎥⎦ + o(h4 ) = Λv = + ai + 2 ∂x 12 ∂x ∂xi ∂xi i i i=1 i=1 ⎤ n ⎡ ∂3 v ⎥⎥⎥ h2 ⎢⎢⎢ ∂4 v ⎢⎣ + 2ai (x) 3 ⎥⎦ + o(h4 ). Lv + 12 i=1 ∂xi4 ∂xi

Let us calculate

⎞⎛ n ⎞ ⎛ n ⎜⎜⎜ ∂2 ∂ ⎟⎟⎟⎟ ⎜⎜⎜⎜ ∂2 v ∂v ⎟⎟⎟⎟ ⎜ L v = ⎜⎝ + ai (x) + ai (x) ⎟⎠ = ⎟⎜ ∂x1 ⎠ ⎝ i=1 ∂xi2 ∂xi ∂xi2 i=1 2

n ⎛ 4 ⎜⎜⎜ ∂ v ∂2 ai ∂v ∂3 v ∂2 v ∂ai ∂2 v ⎜⎝ 4 + 2 +2 + ai 3 + a2i 2 + 2 ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi i=1 ⎞ ⎛ ⎞ n n ⎜⎜⎜ ∂4 v ∂ai ∂v ∂3 v ⎟⎟⎟ ∂3 v ⎟⎟⎟ ⎜⎝ 4 + 2ai 3 ⎟⎠ + + ai 3 ⎠⎟ + Li L j v = ai ∂xi ∂xi ∂xi ∂xi ∂xi i=1 i=1 i j

⎞ ⎤ n ⎡⎛ 2 n ⎢⎢⎢⎜⎜⎜ ∂ ai ∂ai ⎟⎟⎟ ∂v ∂a1 ∂2 v ⎥⎥⎥ 2 ⎢⎣⎜⎝ 2 + ai ⎟⎠ ⎥ + ai + 2 Li L j v. + ⎦ ∂xi ∂xi ∂xi ∂xi2 ∂xi i=1 i=1 i j h2 2 12 L v

Now, the expression is added and removed from Λv: ⎡ ⎤ ⎢⎢ n ⎛⎛ ⎥⎥⎥ ⎞ 2 ⎞ n 2 ⎢ 2 ⎢ ⎥⎥⎥ ⎟ ∂ ∂a ∂ai ⎟⎟⎟ ∂v v h ⎢⎢⎢ ⎜⎜⎜⎜⎜⎜ ∂ ai ⎟ i ⎟⎟⎠ + ⎥⎥⎥ ⎟⎠ ⎢⎢⎢ ⎜⎝⎜⎝ 2 + ai + a2i + 2 L L v i j ⎥⎥⎥ 12 ⎢⎢⎣ i=1 ∂xi ∂xi ∂xi ∂xi ∂xi2 ⎦ i=1 i j

Therefore one gets ⎞ ⎤ ⎛ n ⎡ h2 ⎢⎢⎢ 2 ∂ai ∂2 v ⎜⎜⎜ ∂2 ai ∂ai ⎟⎟⎟ ∂v ⎥⎥⎥ h2 2 ⎢⎣ ai + 2 ⎟⎠ ⎥⎦ − + ⎜⎝ 2 + ai Λv = Lv + L v − 12 12 i=1 ∂xi ∂xi2 ∂xi ∂xi ∂xi n n ⎡ h2 ⎢⎢⎢ 2 h2 ∂ai ∂2 v h2 4 ⎢ Li L j v + o(h ) = − f L f − + ⎣ ai + 2 12 i=1 12 12 i=1 ∂xi ∂xi2 i j

⎛ 2 ⎞ ⎤ n ⎜⎜⎜ ∂ ai ∂a ⎟ ∂v ⎥⎥⎥ h2 ⎜⎝ 2 + ai i ⎟⎟⎟⎠ ⎥⎦ − Li L j v + o(h4 ). ∂xi ∂xi 12 i=1 ∂xi i j

The following approximation is introduced ⎞ ⎞⎛ ⎛ 2 ⎜⎜⎜ ∂ ∂ ⎟⎟⎟ ⎜⎜⎜⎜ ∂2 v ∂v ⎟⎟⎟⎟ ◦ ◦ ⎟⎠ ⎜ Li L j v = ⎜⎝ 2 + ai + aj ⎟ = Λi Λ j v + o(h2 ), ∂xi ⎝ ∂x2j ∂x j ⎠ ∂xi

197

198


and hence the difference equation ⎞ ⎤ ⎛ 2 n ⎡ ⎜⎜⎜ ∂ ai h2 ⎢⎢⎢ 2 ∂ai ∂ai ⎟⎟⎟ ⎥⎥⎥

⎟⎠ v ◦ ⎥⎦ + Λ v ≡ Λv + v xi xi + ⎜⎝ 2 + ai ⎣⎢ ai + 2 12 i=1 ∂xi ∂xi xi ∂xi n i=1 i j

h2 Λi Λ j v = − f + L f 12 ◦

◦

(2.129)

approximates equation (2.127) with the error o(h4 ). The difference equation (2.129) can be rewritten to the form ⎤ ⎫ ⎡ n ⎧ ⎪ ⎢⎢⎢ ∂ai ∂2 ai ⎥⎥⎥ ⎪ ∂ai h2 2 ⎨ ⎬

Λv≡ ⎪ ⎩ 1 + 12 ai + 2 ∂xi v xi xi + ⎣⎢ai + ai ∂xi + ∂x2 ⎥⎦ v x◦i ⎪ ⎭+ i i=1 n h2 h2 ◦ ◦ Λi Λ j v = − f + L f , 12 i=1 12

(2.130)

i j

where:

◦

◦

Λi Λ j v = v xi xi x j x j + ai v x j x j x◦i + ⎛ 2 ⎞ ⎜⎜⎜ ∂ a j ∂a j ⎟⎟⎟ ∂a j ⎜⎝ 2 + ai + ai a j v x◦i x◦j . ⎠⎟ v ◦ + 2 ∂xi x j ∂xi ∂xi ◦

Note that the form of Λi Λ j v follows from approximation Li L j v with the error o(h2 ) ⎞ ⎞⎛ ⎛ 2 ⎜⎜⎜ ∂ ∂ ⎟⎟⎟ ⎜⎜⎜⎜ ∂2 v ∂v ⎟⎟⎟⎟ ∂3 v ∂4 v ⎟⎠ ⎜⎝ 2 + a j Li L j v = ⎜⎝ 2 + ai + ⎟⎠ = 2 2 + ai ∂xi ∂x j ∂x j ∂xi ∂xi ∂x j ∂xi ∂x2j ∂4 v ∂v ∂ ∂v ∂3 v ∂3 v ∂2 a + a a = + a + a + j i j i j ∂x j ∂xi ∂x j ∂xi2 ∂xi2 ∂x2j ∂xi ∂x2j ∂xi2 ∂x j ∂a j ∂v ∂a j ∂2 v ∂2 a j ∂v ∂2 v + 2 + a a + ai = i j ∂xi ∂xi ∂x j ∂xi ∂x j ∂xi ∂x j ∂xi2 ∂x j ⎞ ⎛ 2 ⎜⎜⎜ ∂ a j ∂a j ⎟⎟⎟ ⎟⎠ v ◦ + v xi xi x j x j + ai v x j x j x◦i + a j v xi xi x◦j + ⎝⎜ 2 + ai ∂xi x j ∂xi ∂a j 2 + ai a j v x◦i x◦j + o(h2 ). ∂xi If ai (x) = ai = const, then the scheme (2.130) is given by n n h2 2 h2 ◦ ◦ Λv≡ 1 + ai v xi xi + ai v x◦i + Λi Λ j v = 12 12 i=1 i=1

i j


− f+ where:

◦

199

h2 Lf , 12

(2.131)

◦

Λi Λ j v = v xi xi x j x j + ai v x j x j x◦i + a j v xi xi x◦j + ai a j v x◦i x◦j . Let us construct a difference scheme for the equation (2.81), which is rewritten in the abbreviated form LT ≡ A1

LT ≡

3

∂2 T ∂2 ∂2 T ∂T + A2 2 + A3 2 + a3 = − f (x), 2 ∂x3 ∂x1 ∂x2 ∂x3

Ai Li T + a3 L3 T = − f,

x ∈ G3 {0 ≤ xα ≤ lα , α = 1, 2, 3} .

(2.132)

i=1

Consider the difference operator Λv = A1 v x1 x1 + A2 v x2 x2 + A3 v x3 x3 + a3 v x◦3 . First, we apply the series development with respect to h of difference operators (2.128) ⎛ ⎞⎤ 3 ⎡ h23 ∂3 v ⎢⎢⎢ ∂2 v h2i ∂4 v ⎜⎜⎜ ∂v ⎟⎟⎥⎥ 4 4 ⎢⎣Ai 2 + + o(hi ) + a3 ⎜⎝ + + o(h3 )⎟⎟⎠⎥⎥⎦ = Λv = 4 3 12 ∂xi ∂x3 6 ∂x3 ∂xi i=1 3

(Ai Li v) + a3 L3 v +

i=1

Lv +

3 i=1

Ai

⎤ 3 ⎡ h2 ⎢⎢⎢ h21 2 ⎥ ⎢⎣Ai Li v + 3 a3 L3 L3 v + o(|h|4 )⎥⎥⎥⎦ = 12 6 i=1

h2 h2i 2 Li v + 3 2a3 L3 L3 v + o(|h|4 ), 12 12

|h|4 = h21 + h42 + . . . + h4n .

Second, observe that v is a solution of equation (2.132) Lv ≡ A1 L1 v + A2 L2 v + A3 L3 v + a3 L3 v = − f. Then, one finds

h23 h2i 2 12 Ai Li v, 12 a3 L3 L3 v,

h21 h2 h2 h2 h2 A1 L12 v = − 1 A2 L1 L2 v − 1 A3 L1 L3 v − 1 a3 L1 L3 v − 1 L1 f, 12 12 12 12 12 h22 h2 h2 h2 h2 A2 L22 v = − 2 A1 L2 L1 v − 2 A3 L2 L3 v − 2 L2 L3 v − 2 L2 f, 12 12 12 12 12 h23 h2 h2 h2 h2 A3 L32 v = − 3 A1 L3 L1 v − 3 A2 L3 L2 v − 3 a3 L3 L3 v − 3 L3 f, 12 12 12 12 12 h2 a3 h2 a3 h2 a3 h2 a3 h23 a3 L3 L3 v = − 3 A1 L3 L1 v − 3 A2 L3 L2 v − 3 a3 L3 L3 v − 3 L3 f. 12 12 A3 12 A3 12 A3 12 A3

200


The found relations are substituted to difference operator ⎛ 2 h2 A3 + h23 A1 ⎜⎜ h A2 + h22 A1 L1 L2 v + 1 L1 L3 v+ Λv = Lv − ⎜⎝⎜ 1 12 12 h2 A3 + h23 A1 h22 A3 + h23 A2 a3 (h22 A3 + h23 A2 ) L2 L3 v + a3 1 L3 L1 v + L3 L2 v+ 12 12A3 12A3 ⎞ h23 h23 h23 a3 h21 h22 ⎟⎟ 2 a3 L3 v + L1 f + L2 f + L3 f + L3 f ⎟⎠⎟ + o(|h|4 ). 12A3 12 12 12 12 A3 Requiring to be approximated by difference equation (2.132) with error of o(|h|4 ), it is sufficient to take it in the form ⎞ ⎛ h2 a2 ⎟⎟ ⎜⎜ Λ v ≡ A1 v x1 x1 + A2 v x2 x2 + ⎜⎜⎝A3 + 3 3 ⎟⎟⎠ v x3 x3 + a3 v x◦3 + 12 A3 h2 A3 + h23 A1 h2 A3 + h23 A2 h21 A2 + h22 A1 v x1 x1 x2 x2 + 1 v x1 x1 x3 x3 + 2 v x2 x2 x3 x3 + 12 12 12 h21 A3 + h23 A1 h2 A3 + h23 A2 v x1 x1 x◦3 + a3 2 v x2 x2 x◦3 = 12A3 12A3 ⎞ ⎛ h23 ∂2 f h23 ∂ f ⎟⎟⎟ h22 ∂2 f h21 ∂2 f ⎜⎜⎜ − ⎝⎜ f + + + + ⎠⎟ . 12 ∂x12 12 ∂x22 12 ∂x32 12A3 ∂x3 a3

(2.133)

If h1 = h2 = h3 = h, then the difference equation (2.133) takes the form ⎞ ⎛ ⎜⎜ h2 a23 ⎟⎟⎟ ⎟⎠ v x3 x3 + a3 v x◦ + Λ v ≡ A1 v x1 x1 + A2 v x2 x2 + ⎜⎜⎝A3 + 3 12 A3 h2 h2 h2 (A2 + A1 )v x1 x1 x2 x2 + (A3 + A1 )v x1 x1 x3 x3 + (A2 + A3 )v x2 x2 x3 x3 + 12 12 12 2 $ % a3 h (A3 + A1 )v x1 x1 x◦3 + (A3 + A2 )v x2 x2 x◦3 = 12A3 ⎡ ⎞⎤ ⎛ 3 ⎢⎢⎢ a3 ∂ f ⎟⎟⎟⎟⎥⎥⎥⎥ h2 ⎜⎜⎜⎜ ∂2 f ⎢ ⎟⎟⎥⎥ . ⎜⎜ + − ⎢⎢⎣ f + (2.134) 12 ⎝ i=1 ∂a21 A3 ∂x3 ⎠⎦ Let the following equation is given ⎞ n ⎛ ⎜⎜⎜ ∂2 T ∂T ⎟⎟⎟ ⎟⎠ = − f (x), LT ≡ ⎝⎜A1 2 + ai ∂x1 ∂xi i=1 x ∈ Gn {0 ≤ xα ≤ lα , α = 1, 2, . . . , n} . Proceeding in analogical way, one find that the difference equation

(2.135)


Λ v ≡

⎞ n ⎡⎛ h2 a2 ⎟ ⎢⎢⎢⎜⎜⎜ ⎢⎣⎜⎝Ai + i i ⎟⎟⎟⎠ v xi xi 12 Ai i=1

201

⎡ ⎤ n ⎢ h2 A + h2 A ⎥⎥⎥ ⎢⎢⎢ i j j i v xi xi x j x j + + ai v x◦i ⎥⎦ + ⎢⎣ 12 i, j=1 i j

a1

h2j Ai + h2i A j 12Ai

⎤ ⎥⎥ h2i v x j x j x◦i + a j v xi xi x◦j + ai a j v x◦i x◦j ⎥⎥⎥⎦ = 12A j 12Ai ⎞ ⎛ ⎛ ⎞ n ⎜⎜⎜ h2i ai ∂ f ⎟⎟⎟⎟⎟⎟ ⎜⎜⎜ h2i ∂2 f ⎜ ⎜⎝ + − ⎜⎝ f + ⎠⎟⎟⎟⎠ 2 12 12 A ∂x ∂x i i i i=1 h2i A j + h2j Ai

(2.136)

approximates (2.135) with error of o(|h|4 ). Let us construct a difference scheme for equation with variable coefficients ⎞ n ⎛ ⎜⎜⎜ ∂2 T ∂T ⎟⎟⎟ LT ≡ ⎝⎜Ai 2 + ai (x) ⎠⎟ = − f (x), ∂xi ∂xi i=1 LT ≡

n

Ai Li + ai (x)Li T

i=1

x=

(x1j , . . . , xnj )

∈ Gn {0 ≤ xα ≤ lα , α = 1, . . . , n} .

(2.137)

Consider the difference operator

Λv ≡ i = 1n Ai v xi xi + ai (x)v x◦i , ei h . x = (x1j , . . . , xnj ) ∈ Gn xij = jhi , i = 1, . . . , n, j = 0, 1, . . . , N1 , Ni = hi Applying series development of difference operators with respect to hi , we get ⎞ ⎞⎤ ⎛ n ⎡ ⎛ 2 ⎢⎢⎢ ⎜⎜⎜ ∂ v h2i ∂4 v ⎟⎟⎟ ⎜⎜⎜ ∂v h2i ∂3 v ⎟⎟⎟⎥⎥⎥ ⎢⎣Ai ⎜⎝ 2 + ⎟ ⎟⎥ + o(|h|4 ) = ⎜ Λv ≡ (x) + + a ⎝ i 4⎠ 3 ⎠⎦ 12 ∂x 6 ∂x ∂x ∂x i i i i i=1 Lv +

⎤ n ⎡ h2i ∂3 v ⎥⎥⎥ ⎢⎢⎢ h2i ∂4 v ⎢⎣Ai ⎥ + o(|h|4 ). + 2a (x) i 4 3⎦ 12 12 ∂x ∂x i i i=1

The following difference operator is obtained ⎞ n ⎛ h2i ⎜⎜⎜ h2i 2 ⎟⎟ 4 Λv = − f + ⎝⎜Ai Li v + 2ai (x) Li Li v⎟⎟⎠ + o(|h| ). 12 12 i=1 Ai

n n h2 h2i 2 h2i

h2i

Li v = − Li a j L j v − i Li f, A j Li L j v − 12 12 12 12 j=1 j=1 ji

202


⎤ n ⎡ 2

⎥⎥

h2i h2i 2 ⎢⎢⎢ hi ⎢⎣ Ai Li v + ai Li Li v = − A j Li L j v + Li a j L j v ⎥⎥⎦ − 12 12 12 j=1 ji

⎛ ⎞ h2i ⎜⎜⎜ ∂2 ai ∂v ∂ai ∂2 v ⎟⎟⎟ h2i ⎜⎝ 2 ⎟⎠ − Li f, +2 12 ∂xi ∂xi ∂xi ∂xi2 12 ai

n

h2i h2i ai

Li L i v = − A j Li L j v + Li a j L j v − 12 12 Ai j=1 ji

⎛ ⎞ h2i ai ⎜⎜⎜ ∂ai ∂v ∂2 v ⎟⎟ h2 ai ⎜⎝ + ai 2 ⎟⎟⎠ − i Li f. 12 Ai ∂xi ∂xi 12 Ai ∂xi One may conclude that the equation (2.137) will be approximated by a difference equation with error o(|h|4 ), if it is taken in the form ⎛ ⎛ ⎞⎤ ⎞⎤ ⎫ ⎡ n ⎧⎡ ⎪ h2i ⎜⎜⎜ a2i h2i ⎜⎜⎜ ∂2 ai ai ∂ai ⎟⎟⎟⎥⎥⎥ ⎪ ⎢⎢⎢ ∂ai ⎟⎟⎟⎥⎥⎥ ⎨⎢⎢⎢ ⎬

Λ v≡ ⎪ ⎩⎣⎢Ai + 12 ⎝⎜ Ai +2 ∂xi ⎠⎟⎦⎥ v xi xi + ⎣⎢ai + 12 ⎝⎜ ∂x2 + Ai ∂xi ⎟⎠⎥⎦ v x◦i ⎪ ⎭+ i i=1 ⎤ ⎡ 2 ⎧⎡ n h2i ⎪ ⎢⎢⎢ ∂ a j ai ∂ak ⎥⎥⎥ ai ⎨⎢⎢⎢ ⎥⎦ v ◦ + ⎪⎢⎣A j v xi xi x j x j + a j v xi xi x◦j + A j v x j x j x◦i + ⎢⎣ 2 + 12 ⎩ Ai Ai ∂xi xk ∂xi i, j=1 i j

⎛ ⎛ ⎞⎞ n ⎜⎜⎜ h2i ⎜⎜⎜ ∂2 f ⎟⎟⎟⎟⎟⎟ ∂a j ai a ∂ f i ⎜⎝ 2 + ⎟⎠⎟⎟ . + a j v x◦i x◦j = − ⎜⎜⎝ f + 2 ∂xi Ai 12 ∂xi Ai ∂xi ⎠ i=1

(2.138)

Finally, difference schemes with errors o(h4 ) are constructed for the equation n ∂ ∂T LT ≡ = − f (x), K(x) ∂xi ∂xi i=1 x = (x1 , . . . , xn ) ∈ Gn {0 ≤ xα ≤ lα , α = 1, . . . , n} ,

(2.139)

where LT ≡

n i=1

Li T, Li T =

∂ ∂T , K(x) = K(x1 , x2 , . . . , xn ). K(x) ∂xi ∂xi

This equation governs stationary temperature distribution of a non-homogeneous medium. 1 , and consider the difference operator Denote p(x) = k(x) Λv =

n 1 i=1

a

.

v xi xi


203

Developing the difference operator into the series with respect to h [592], one gets ∂ 1 ∂v h2 1 v xi = + Li (pLi v) + o(h4 ), a ∂xi p ∂xi 12 and the difference operator reads Λv ≡

n 1 i=1

a

=

v xi xi

n

Li v +

i=1

n h2 Li (pLi v) + o(h4 ). 12 i=1

(2.140)

Using the fact, that v is solution of equation (2.140), one obtains ⎛ n ⎞ ⎜⎜⎜ ⎟⎟⎟ Lv = − f, pLv = −p f, Li p ⎜⎜⎝ Li v⎟⎟⎠ = −Li p f, i=1

Li (pLi v) = −Li

n

pL j v − Li p f,

j=1 i j n

Li (pLi v) = −

i=1

n

Li (pL j v) −

i, j=1

n

Li p f.

i=1

Substituting the obtained relation into the difference operator (2.140), one gets ⎛ ⎞ ⎜⎜ ⎟⎟⎟ n n n n ⎜ ⎟⎟⎟ 2 ⎜ 1 h ⎜⎜⎜⎜ Λv = v xi = Li v − Li (pL j v) + Li p f ⎟⎟⎟⎟⎟ + o(h4 ) = ⎜⎜⎜ a 12 ⎜⎜⎝i, j=1 ⎟⎟⎠ xi i=1 i=1 i=1 i j

−

n n h2 Li (pL j v) − f − Li p f + o(h4 ). 12 i, j=1 i=1 i j

It is sufficient for difference equation to approximate the equation (2.139) with the error o(h4 ), if it is taken in the form n n 1 h2 h2

vx + (2.141) Λi (pΛ j v) = − f + L(p f ) , Λv≡ a i xi 12 i, j=1 12 i=1 i j

where: Λi v =

1 a v xi xi ,

and the coefficient

⎛ ⎞ 1 ⎜⎜ 1 1

4 1 ⎟⎟⎟⎟ ⎜ ⎜ + + ⎟. ai = pi−1 + 4pi− 12 + pi = ⎜⎝ 6 6 ki−1 ki− 12 ki ⎠ Note that the coefficients ai are obtained from the following condition

(2.142)

204


1 Ψ = Λi v − Li v = v xi a

xi

∂ 1 ∂v = o(h2 ). − ∂xi p ∂xi

In reference [592] it is illustrated, that it takes place when the following condition is satisfied

1 1 1 1 1 1 1 1 + = + o(h2 ), − = + o(h2 ). 2 a(x + h) a(x) p(x) h a(x + h) a(x) p This condition is satisfied, if ai are taken in the form (2.142). Consequently, the difference equation (2.141) approximates the initial equation (2.139) with the error of o(h4 ), where ai are given by (2.142). Let us now construct a difference system with an error o(h4 ) for the equation with variable coefficients Lu ≡

n

Ai (x)

i=1

∂2 u = − f (x), ∂xi2

(2.143)

x = (x1 , . . . , xn ) ∈ Gn {0 ≤ xα ≤ lα , α = 1, . . . , n} , n n Λv = Ai (x)Λi v = Ai (x)v xi xi . i=1

Λi v = v xi xi =

i=1

∂ v + L2 v + o(h4 ), ∂xi2 12 i 2

h2i

where Li v =

∂2 v , ∂xi2

where Ai (x), f (x) are sufficiently smooth functions. Developing the difference operator into the series with respect to h of of the form Λv =

n

Ai (x)Li v +

i=1

n h2i Ai (x)Li2 v + o(h4 ), 12 i=1

(2.144)

and using the fact that v is a solution to equation (2.143), the following relations are obtained Lv =

n

Ai (x)Li v = − f (x),

i=1

Li v = −

n Aj 1 L j v − f (x), A A i i j=1 i j

and therefore Li2 v

=−

n j=1 i j

Aj f . Li L j v − Li Ai Ai

Substituting the obtained result into the difference operator (2.144), one gets


Λv =

n

205

Ai (x)Λi v =

i=1 n

Ai (x)Li v −

i=1

n i=1

⎡ ⎤ ⎢⎢⎢ ⎥⎥⎥⎥ n ⎢⎢⎢ Aj f ⎥⎥⎥⎥ Ai ⎢⎢⎢⎢⎢ Li L j v + Li ⎥ + o(|h|4 ) = 12 ⎢⎢⎣ j=1 Ai Ai ⎥⎥⎥⎥⎦ h2i

i j

n n n h2i h2i Aj f − f (x) − Ai Li Ai − Li L j v + o(|h|4 ). 12 Ai 12 j=1 Ai i=1 i=1 i j

The following difference equation

Λv≡

n j=1

n h2i Aj Ai Λi Ai (x)Λi v + Λ jv = 12 Ai i, j=1 i j

⎛ ⎞ n ⎜⎜⎜ h2i f ⎟⎟⎟⎟ Ai Li − ⎜⎜⎝ f (x) + ⎟ 12 Ai ⎠ i=1 approximates (2.143) with the error of o(|h|4 ). This difference equation is transformed to the following form ⎛ ⎞ ⎜⎜⎜ ⎟⎟⎟ ⎛ ⎞ n n n ⎜⎜⎜⎜ ⎜⎜⎜ h2i h2i A j ⎟⎟⎟⎟ f ⎟⎟⎟⎟

Ai Li Ai (x)Λi ⎜⎜⎜⎜E + Λ j ⎟⎟⎟⎟ v = − ⎜⎜⎝ f (x) + Λv≡ ⎟ . (2.145) 12 j=1 Ai ⎟⎟⎠ 12 Ai ⎠ ⎜⎜⎝ i=1 i=1 i j

If Ai (x) = Ai = const, then from (2.145) the following difference equation is obtained ⎛ ⎞ n n n ⎜⎜⎜ ⎟⎟ h2i h2i

A j Λi Λ j v = − ⎜⎜⎝ f (x) + Li f ⎟⎟⎟⎠ , Ai Λi v + Λv≡ (2.146) 12 12 i=1 i, j=1 i=1 i j

which for n = 3 approximates the equation (2.147) with the error of o(|h|4 ): LT ≡

3 i=1

Ai

∂2 T = − f (x). ∂xi2

(2.147)

In what follows, the Dirichlet problem for this equation in the space G3 {0 ≤ xα ≤ lα }, {α = 1, 2, 3} with the boundary Γ T |Γ = g(x), is analysed.

x ∈ Γ,

(2.148)

206


2.3.3 A priori convergence estimation In order to prove a convergence of difference scheme (2.146) a background of difference scheme theory [592] is briefly described. Let ω is the finite set of nodes (mesh) in a certain bounded space of n-dimensional Euclidean space and pεω is the mesh point. Let us consider the equation B(P, Q)v(Q) + F(P), P ∈ ω (2.149) A(P)v(P) = Q∈Ξ (P)

for the function v(P) given on the mesh ω. Let A(P) and B(P, Q) are the equation coefficients; F(P) (denotes the right hand side of equation) are given mesh functions; Ξ (P) is the set of nodes of mesh ω except of the node R (or a neighbourhood of a node P). A pattern of mesh equation (2.149) in the node P consists of the node P and its neighbourhood Ξ (P). It is assumed that the coefficients A(P) and B(P, Q) satisfy the conditions A(P) > 0, B(P, Q) ≥ 0 for P ∈ ω and Q ∈ Ξ (P), B(P, Q) ≥ 0. D(P) = A(P) −

(2.150)

Q∈Ξ (P)

The point P is said to be a boundary node of the mesh ω, if in this point the function v(P) value is given, i.e. v(P) = g(P)

for P ∈ Γ,

(2.151)

where Γ is a set of boundary nodes. Comparing (2.151) with (2.149) one may conclude, that on the boundary Γ the following formal relations can be introduced A(P) ≡ 1,

B(P, Q) = 0,

F(P) = g(P).

Owing to the notation Zv(P) = A(P)v(P) −

B(P, Q)v(Q),

Q∈Ξ (P)

the equation (2.149) takes the form Zv(P) = F(P).

(2.152)

THEOREM 2.1 If D(x) > 0 on ω, then the problem (2.149)–(2.152) with ϕ(x) = 0 can be a priori estimated via the following inequality (( ( F(x) (( ( , vl ≤ ((( D(x) (l where vl = max x∈ω |v(x)| .


207

THEOREM 2.2 Let v(x) is the solution to the problem (2.149)–(2.152), and v(x) is the solution obtained via change in (2.152), (2.151) the functions F(x), ϕ(x) by the functions F(x), g(x), and there exist at least one node x0 of the mesh ω where D(x0 ) > 0. Then, if the following conditions are satisfied |F(x)| ≤ F(x), x ∈ ω; |g(x)| ≤ g(x), x ∈ Γ, the following inequality |v(x)| ≤ v(x) holds on ω. Proof of these two theorems is given in reference [592]. Consider now the difference Dirichlet problem approximating (2.149), (2.150) with the error of o(|h|4 ) with a help of difference scheme (2.126), which can be presented in the following form Λ v ≡

3 i=1

Ai v xi xi +

3 h2i A j v xi xi x j x j = −ϕ(x), x ∈ ω 12 i, j=1

(2.153)

i j

v = g(x) for x ∈ Γ,

ϕ(x) = f +

3 h2i ∂2 f , 12 ∂xi2 i=1

where ω is the mesh space G3 {0 ≤ xα ≤ lα , α = 1, 2, 3}. THEOREM 2.3 Difference scheme (2.153) is uniformly convergent with the velocity o(|h|4 ) (it possesses fourth accuracy order), if a solution to initial differential problem (2.148) T (x) ∈ C (6) (G3 ), function f (x) ∈ C (4) (G3 ), and the following conditions are satisfied 4A2 A1 A3 4A3 A2 A1 4A1 A2 A3 − 2 − 2 ≥ 0, − 2 − 2 ≥ 0, − 2 − 2 ≥ 0. 2 2 h1 h2 h3 h2 h1 h3 h23 h2 h1

(2.154)

Proof. For the error τ = v − T the following problem is considered Λ τ = −Ψ, x ∈ ω;

τ = 0, x ∈ Γ,

(2.155)

where 3

Ψ = Λ v + ϕ = o(|h|4 ) for x ∈ ω, if v ∈ C (6) (G ), f ∈ C (4) (G3 ). Let us check the condition (2.150). For this purpose the scheme (2.153) is rewritten in the form (2.149): B(P, Q)v(Q) + F(P). A(P)v(P) = Q∈Ξ (P)

Scheme (2.153) has the following form

Av = B1 v(+11 ) + v(−11 ) + B2 v(+12 ) + v(−12 ) + B3 v(+13 ) + v(−13 ) +

208


B4 v(+11 ,+12 ) + v(+11 ,−12 ) + v(−11 ,+12 ) + v(−11 ,+12 ) +

B5 v(+11 ,+13 ) + v(+11 ,−13 ) + v(−11 ,+13 ) + v(−11 ,−13 ) +

B6 v(+12 ,+13 ) + v(+13 ,−13 ) + v(−12 ,+13 ) + v(−12 ,−13 ) + ϕ,

(2.156)

where A and Bi are expressed via coefficients of the equation (2.147) and hi : ⎡ ⎡ ⎤ ⎤ 1 ⎢⎢⎢ 4A1 A2 A3 ⎥⎥⎥ 4 ⎢⎢⎢ A1 A2 A3 ⎥⎥⎥ A = ⎣⎢ 3 + 2 + 2 ⎦⎥ > 0, B1 = ⎣⎢ 2 − 2 − 2 ⎦⎥ , 3 h1 6 h1 h2 h3 h2 h3 ⎡ ⎡ ⎤ ⎤ 1 ⎢⎢⎢ 4A2 A1 A3 ⎥⎥⎥ 1 ⎢⎢⎢ 4A3 A1 A2 ⎥⎥⎥ B2 = ⎣⎢ 2 − 2 − 2 ⎦⎥ , B3 = ⎣⎢ 2 − 2 − 2 ⎦⎥ , 6 h2 6 h3 h1 h3 h1 h2 ⎡ ⎡ ⎡ ⎤ ⎤ ⎤ 1 ⎢⎢⎢ A1 A2 ⎥⎥⎥ 1 ⎢⎢⎢ A1 A3 ⎥⎥⎥ 1 ⎢⎢⎢ A2 A3 ⎥⎥⎥ ⎢⎣ 2 + 2 ⎥⎦ , B5 = ⎢⎣ 2 + 2 ⎥⎦ , B6 = ⎢⎣ 2 + 2 ⎥⎦ . B4 = 12 h1 12 h1 12 h2 h2 h3 h3 The following relation holds ⎛ ⎛ ⎞ ⎞ 4 ⎜⎜ A1 A2 A3 ⎟⎟ 1 ⎜⎜ 4A1 A2 A3 ⎟⎟ D(P) = A(P) − B(P, Q) = ⎜⎜⎝ 2 + 2 + 2 ⎟⎟⎠ − ⎜⎜⎝ 2 − 2 − 2 ⎟⎟⎠ − 3 h1 3 h1 h2 h3 h2 h3 Q∈Ξ (P) ⎛ ⎛ ⎛ ⎞ ⎞ ⎞ 1 ⎜⎜⎜ 4A2 A1 A3 ⎟⎟⎟ 1 ⎜⎜⎜ 4A3 A1 A2 ⎟⎟⎟ 1 ⎜⎜⎜ A1 A2 ⎟⎟⎟ ⎜⎝ 2 − 2 − 2 ⎟⎠ − ⎜⎝ 2 − 2 − 2 ⎟⎠ − ⎜⎝ 2 + 2 ⎟⎠ − 3 h2 3 h3 3 h1 h1 h3 h1 h2 h2 ⎛ ⎛ ⎞ ⎞ 1 ⎜⎜⎜ A1 A3 ⎟⎟⎟ 1 ⎜⎜⎜ A2 A3 ⎟⎟⎟ ⎜⎝ + 2 ⎟⎠ − ⎜⎝ 2 + 2 ⎟⎠ = 0, 3 h21 3 h2 h3 h3 for the boundary points D(P) = 1 > 0. In other words, conditions (2.150) are satisfied if 4A2 A1 A3 4A3 A1 A2 4A1 A2 A3 − 2 − 2 ≥ 0, − 2 − 2 ≥ 0, − 2 − 2 ≥ 0. h21 h2 h3 h22 h1 h3 h23 h1 h2 In order to estimate a solution to the problem (2.155) the following majorant function is constructed ⎛ ⎞ l2 x2 ⎟ l2 x2 x2 C0 ⎜⎜⎜ l12 ⎜⎝ + 2 + 3 − 1 − 2 + 3 ⎟⎟⎟⎠ . τ(x) = 6 A1 A2 A3 A1 A2 A3 Observe that Λ τ =

3 i=1

Ai τ xi xi +

3 h2i A j τ xi xi x j x j = −C0 . 12 i, j=1 i j

For the function τ(x) in the space G3 {0 ≤ xα ≤ lα , α = 1, 2, 3} the following estimation holds


209

⎛ ⎞ l32 ⎟⎟⎟ l22 C0 ⎜⎜⎜ l12 0 ≤ τ(x) ≤ + ⎟⎠ . ⎝⎜ + 6 A1 A2 A3 In order to solve the problem (2.155), assuming C0 = max x∈ω |Ψ (x)| = ||Ψ ||e and applying Theorem 3, the following estimation is obtained ⎛ ⎞ l2 ⎟⎟ l2 1 ⎜⎜ l2 max x∈ω |τ(x)| = max x∈ω |v − T | ≤ τ(x) ≤ ⎜⎜⎝ 1 + 2 + 3 ⎟⎟⎠ max x∈ω |Ψ (x)| 6 A1 A2 A3 under condition (2.154), i.e. ⎛ ⎞ l32 ⎟⎟⎟ l22 1 ⎜⎜⎜ l12 ⎜ + ⎟⎠ Ψ l . τl ≤ ⎝ + 6 A1 A2 A3 It means, that a solution of difference problem (2.153) is uniformly convergent with velocity o(|h|4 ) to solution of differential problem, what finishes the proof. Remark. For a mesh h1 = h2 = h h3 the condition (2.154) is satisfied, if h2 4A3 A3 ≤ 32 ≤ , 4A2 − A1 h A1 + A2

h2 A3 ≤ 32 . 4A1 − A2 h

For a squared mesh h1 = h2 = h3 = h the condition (2.154) is satisfied, if 4A1 ≥ A2 + A3 , 4A2 ≥ A1 + A3 , 4A3 ≥ A1 + A2 . 2.3.4 Algorithm of computation and compatibility conditions In this section, a stationary three dimensional heat transfer equation for shallow shells [555] is analysed LT ≡ a1

∂T ∂2 T ∂2 T ∂2 T + a2 2 + a3 2 + k = − f (x1 , x2 , x3 ), 2 ∂x3 ∂x1 ∂x2 ∂x3

where: k=

(2.157)

∂l12 l22 l1 k1 + k , λ = , f = f (l1 x1 , l2 x2 , 2Hx3 ), 2 l2 λ2 (2H)2 2 2H 1 2H T, ki = 2 ki , (i = 1, 2), T = α11 l1 li

xi = li xi , (i = 1, 2),

x3 = 2hx3 ,

2 l2 1 2 a1 = 2 , a2 = 1, a3 = λ1 = . 2H λ

Equation (2.157) is already in non-dimensional form, and bars over nondimensional parameters are omitted. Applying approximation (2.133), where a3 = k, Ai = ai , the difference equation approximating (2.157) with the error o(|h|4 ) has the form

210


ΛT ≡

3 i=1

3 k2 h23 a j h2i Λi Λ j T + kΛ˜ 3 T + Λi T + Λ3 T + 12 12a3 i, j=1 i j

⎛ 2 ⎞ h22 a3 + h23 a2 ⎜⎜⎜ h1 a3 + h23 a1 ⎟⎟ ˜ ˜ k ⎝⎜ Λ3 Λ! T + Λ3 Λ2 T ⎟⎟⎠ = −ϕ, 12a3 12a3 where: ϕ = f (x) +

(2.158)

3 kh23 ∂ f h2i ∂2 f + , 12 ∂xi2 12a3 ∂x3 i=1

and Λi T , Λ˜ 3 T (i = 1, 2, 3) are approximated by central differences with the error of o(|h|2 ). Substituting in (2.158) the difference operators by central differences relaxations, the following difference equation is obtained

AT i jk = B1 T i+1 jk + T i−1 jk + B2 T i j+1k + T i j−1k + B3 T i jk+1 + T i jk−1 +

B5 T i+1 jk+1 + T i+1 jk−1 + T i−1 jk+1 + T i−1 jk−1 + B6 T i j+1k+1 + T ik+1k−1 +

T i j−1k+1 + T i j−1k−1 + B4 T i+1 j+1k + T i+1 j−1k + T i−1 j+1k + T i−1 j−1k +

B7 T i jk+1 + T i jk−1 + B8 T i+1 jk+1 + T i−1 jk+1 − T i+1 jk−1 − T i−1 jk−1 +

B9 T i j+1k−1 + T i j−1k+1 − T i j+1k−1 − T i j−1k−1 + ϕ, (2.159) where:

⎛ ⎞ 4 ⎜⎜⎜ a1 a2 a3 ⎟⎟⎟ A = ⎜⎝ 2 + 2 + 2 ⎟⎠ + 3 h1 h2 h3 ⎛ ⎞ a1 1 ⎜⎜ a12 a13 ⎟⎟ B1 = 2 − ⎜⎜⎝ 2 2 + 2 2 ⎟⎟⎠ , a12 h1 6 h1 h2 h1 h3 ⎛ ⎞ a23 ⎟⎟ a2 1 ⎜⎜ a12 B2 = 2 − ⎜⎜⎝ 2 2 + 2 2 ⎟⎟⎠ , a13 h2 6 h1 h2 h2 h3 ⎛ ⎞ a23 ⎟⎟ a3 1 ⎜⎜ a13 k2 B3 = 2 − ⎜⎜⎝ 2 2 + 2 2 ⎟⎟⎠ + , 12a3 h3 6 h1 h3 h2 h3 B4 =

k2 , 6a3 = a1 h22 + a2 h21 , = a1 h23 + a3 h21 , a23 = a2 h23 + a3 h22 ,

a12 a13 a23 k a13 , B5 = , B6 = , B8 = , 2 2 2 2 2 2 2h3 12h21 a3 12h1 h2 12h1 h3 12h2 h3 ⎛ ⎞ a23 k ⎜⎜⎜ a23 ⎟⎟⎟ k a13 B7 = . (2.160) ⎝⎜1 − 2 − 2 ⎟⎠ , B9 = 2h3 2h3 12h22 a3 6h1 a3 6h2 a3

We are going to find approximation of boundary conditions and compatibility conditions for the equation (2.157) with the error o(|h|4 ). The boundary condition takes the form


kα −kα

∂T = κ−α − g−α (xβ , xγ ), ∂xα

211

xα = 0,

∂T = κ+α − g+α (xβ , xγ ), xα = 1, α, β, γ = 1, 2, 3; α β γ, ∂xα

(2.161)

in the space G3 {0 ≤ xα ≤ 1, α = 1, 2, 3}. One may get from the boundary condition (2.161) the first, second and third order boundary conditions by giving the coefficients kα , κα , κ+α values of 0 or 1. In the case kα = 0, i.e. for the first boundary value problem, the boundary conditions are approximated exactly. Consider approximation of second order boundary conditions and compatibility conditions, i.e. when κ−α = κ+α = 0 ∂T = −g−α (xβ , xγ ), xα = 0; ∂xα

−

∂T = −g+α (xβ , xγ ), xα = 1. ∂xα

(2.162)

Introducing notations ⎧ 2 ⎪ v x , xα = 0 ⎪ ⎪ ⎪ ⎨ hα α v , xα 0; 1 α = 1, 2, 3 , Λα v = ⎪ x α xα ⎪ ⎪ ⎪ ⎩ − 2 v xα , xα = 1 hα

(2.163)

the boundary conditions for x1 = 0 take the form ∂T = −g−1 (x2 , x3 ). ∂x1

(2.164)

Let us develop T (x) into Taylor series in neighbourhood of the node x1 = 0. Then the approximation of the boundary condition (2.164) with the error o(h2 ) has the form (recall, that T (x) is solution of equation (2.157)) 2a1 2a1 T x1 + a2 T x2 x2 + a3 T x3 x3 + kT x◦3 = − f (x) + g−1 (x2 , x3 ) . (2.165) h1 h1 Assume that x1 = 0, x2 = 0, x3 0, 1. Similarly to the previous case one may observe that the expression (2.166) approximates a compatibility condition along the rib x1 = x2 = 0 with the second order accuracy 2a2 2a1 2a2 2a1 (2.166) Tx + T x + a3 T x3 x3 + kT x◦3 = − f (x) + g−1 + g−2 . h1 1 h2 2 h1 h2 Difference equation (2.158) approximates the equation (2.157) with the error o(|h|4 ), if the point (x1 , x2 , x3 ) ∈ ω. Let us construct a difference scheme, approximating the problem (2.157) with boundary conditions (2.161) with the error amount of o(|h|4 ). A difference scheme is sought in the form analogous to (2.158)

Λ T ≡ a1 Λ1 T + a2 Λ2 T + a3 Λ3 T +

a2 h21 + a1 h22 Λ1 Λ2 T + 12

212


a3 h21 + a1 h23 a3 h22 + a2 h23 k2 h23 Λ1 Λ3 T + Λ2 Λ3 T + Λ3 T + 12 12 12a3 ⎛ 2 ⎞ h22 a3 + a2 h23 ⎜⎜⎜ h1 a3 + a1 h23 ⎟⎟ ˜ ˜ Λ3 Λ1 T + Λ3 Λ2 T ⎟⎟⎠ = −Φ . k ⎝⎜ 12a3 12a3

(2.167)

Right hand side Φ for x ∈ ω should be defined in order to obtain error of approximation o(|h|4 ). If for x ∈ ω one assumes Φ = Φ, then the equation (2.167) coincides with the equation (2.158), and hence it approximates (2.157) with the error of o(|h|4 ). Finally, we need to give Φ on a boundary of the space G3 {0 ≤ xα ≤ 1, (α = 1, 2, 3)}. Consider first the equation (2.167) for x1 = 0, x2 0, 1, x3 0, 1. Owing to (2.163), we rewrite the equation (2.167), multiplying it by h1 /2, in the form a1 T x1 +

a2 h21 + a1 h22 h1 h1 h1 a2 T x2 x2 + a3 T x3 x3 + kT x◦3 + T x1 x2 x2 + 2 2 2 12a1

a3 h21 + a1 h23 a3 h22 + a2 h23 h1 k2 h23 h1 T x2 x2 x3 x3 + Tx x + T x1 x3 x3 + 12a1 12 2 12a3 2 3 3 ⎛ ⎞ a3 h22 + a2 h23 ⎟⎟ h1 kh1 ⎜⎜⎜ a3 h21 + a2 h23 T x◦3 x1 x1 + T x◦3 x2 x2 ⎟⎟⎠ + Φ = 0. ⎝⎜ 2 12a3 12a3 2 In order to compute an error of approximation Ψ of the equation, T is developed into Taylor series ⎛ ⎜⎜ ∂T h1 ∂2 T h21 ∂3 T h31 ∂4 T + + + + o(h41 )+ Ψ (0, x2 , x3 ) = ⎜⎝⎜ ∂x1 2 ∂x12 6 ∂x13 24 ∂x14 ⎛ ⎛ ⎞ ⎞ ⎟⎟⎟ h1 ⎜⎜⎜ ∂2 T h23 ∂4 T ⎟⎟ h1 ⎜⎜⎜ ∂2 T h22 ∂4 T 4 4 a2 ⎜⎝ 2 + + o(h2 )⎟⎠ + a3 ⎜⎝ 2 + + o(h3 )⎟⎟⎠ + 4 4 2 2 ∂x2 12 ∂x2 ∂x3 12 ∂x3 ⎛ ⎛ a3 h21 + a1 h23 ⎜⎜⎜ ∂ ∂2 T a2 h21 + a1 h22 ⎜⎜⎜ ∂ ∂2 T h1 ∂2 ∂2 T 2 ⎜⎝ ⎜⎝ + + o(h ) + + 1 12 ∂x1 ∂x22 2 ∂x12 ∂x22 12 ∂x1 ∂x32 ⎛ ⎞ ⎞ ⎟⎟ a3 h22 + a2 h23 h1 ⎜⎜⎜ ∂2 ∂2 T ⎟⎟ h1 ∂2 ∂2 T 2 ⎟ 2 ⎟ ⎜ ⎟ ⎟ + o(h ) + o(h ) + ⎝ 2 2 1 ⎠ 2 ⎠+ 2 2 2 ∂x1 ∂x3 12 2 ∂x2 ∂x3 ⎛ ⎞ h2 h1 ⎜⎜⎜ ∂2 T h23 ∂4 T ⎟⎟ 4 ⎟ ⎜⎝ 2 + ⎟ + o(h ) k2 3 3 ⎠+ 4 12a3 2 ∂x3 12 ∂x3 ⎡ ⎛ ⎞ ⎟⎟ h1 ⎢⎢ a3 h21 + a1 h23 ⎜⎜⎜ ∂ ∂2 T h3 ∂2 ∂2 T 2 ⎟ ⎜⎝ ⎟ + + o(h ) k ⎢⎢⎣ 3 ⎠ + 2 2 2 2 12a3 ∂x3 ∂x1 2 ∂x3 ∂x1 ⎛ ⎞⎤ a3 h22 + a2 h23 ⎜⎜⎜ ∂ ∂2 T h3 ∂2 ∂2 T ⎟⎟⎥⎥⎥ h1

2 ⎟ ⎜⎝ ⎟⎥ Φ. + + o(h ) 3 ⎠⎦ + 2 2 2 12a3 ∂x3 ∂x2 2 ∂x3 ∂x2 2


213

Since T is a solution of the equation (2.157) with attached boundary conditions (2.161), then in order to achieve Ψ by approximation for x1 = 0 with error o(|h|4 ), it is sufficient to take ⎛ h2 ∂2 f h2 ∂2 f h2 ∂2 f h2 ∂ f 2 ⎜⎜⎜ ⎜⎝g−1 + 1 + Φ = f + 1 2 + 2 2 + 3 2 + 12 ∂x1 12 ∂x2 12 ∂x3 h1 6 ∂x1 ⎞ a1 h22 − a2 h21 ∂2 g−1 a1 h23 − a3 h21 ∂2 g−1 ⎟⎟⎟ ⎟⎠ = + 12a1 12a1 ∂x22 ∂x32

⎛ ⎞ a1 h22 + a2 h21 ∂2 g−1 a1 h23 − h21 a3 ∂2 g−1 ⎟⎟⎟ h21 ∂ f 2 ⎜⎜⎜ ⎜⎝g−1 + ⎟⎠ . + + ϕ + h1 6 ∂x1 12a1 12a1 ∂x22 ∂x32

(2.168)

Approximation of a boundary condition with the error of o(|h|4 ) for x1 = 0 has the following form 2 2 a2 h21 + a1 h22 a1 T x1 + a2 T x2 x2 + a3 T x3 x3 + T x1 x2 x2 + kT x◦3 + h1 h1 12a1 a3 h22 + a2 h23 h2 2 a3 h21 + a1 h23 T x2 x2 x3 x3 + k2 3 T x3 x3 + T x1 x3 x3 + h1 12a1 12 12a3 ⎛ ⎞ 2 2 2 2 a3 h2 + a2 h3 ⎜⎜ a3 h1 + a1 h3 ⎟⎟ k ⎜⎜⎝ T x◦3 x1 x1 + T x◦3 x2 x2 ⎟⎟⎠ = −Φ , 12a3 12a3

(2.169)

where Φ is defined by (2.168). Let us define an approximation of a compatibility condition along the rib x1 = 0, x2 = 0, x3 0, 1. Owing to (2.163), the equation (2.167) takes the form a3 h21 + a1 h22 2 2 2 2 a1 T x1 + a2 T x2 + a3 T x3 x3 + kT x◦3 + Tx x + h1 h2 12 h1 h2 1 2 a3 h21 + a1 h23 2 a3 h22 + a2 h23 2 k2 h23 T x1 x3 x3 + T x3 x3 x2 + Tx x + 12 h1 12 h2 12a3 3 3 k

h21 a3 + a1 h23 2 h2 a3 + a2 h23 2 T x1 x◦3 + k 2 T ◦ + Φ = 0. 12a3 h1 12a3 h2 x2 x3

Error of approximation of this equation reads ⎛ ⎞ ⎟ 2 ⎜⎜⎜ ∂T h1 ∂2 T h21 ∂3 T h31 ∂4 T 4 ⎟ ⎜ Ψ (0, 0, x3 ) = a1 ⎝ + + + + o(h1 )⎟⎟⎠ + h1 ∂x1 2 ∂x12 6 ∂x13 24 ∂x14 ⎛ h2 ∂2 T h22 ∂3 T h32 ∂4 T 2 ⎜⎜⎜ ∂T a2 ⎜⎝ + + + + o(h4 )+ h2 ∂x2 2 ∂x22 6 ∂x23 24 ∂x24 ⎞⎛ 2 ⎛ 2 2 ⎞ ⎞ ⎛ h2 ∂ 3 T ⎟⎟ ⎜⎜ ∂ T h2 ∂4 T ⎜⎜⎜ k h3 ⎟⎟ ⎟⎟ ⎜⎜ ∂T ⎜⎝ + a3 ⎟⎟⎠ ⎜⎜⎝ 2 + 3 4 + o(h43 )⎟⎟⎠ + k ⎜⎜⎝ + 3 3 + o(h43 )⎟⎟⎠ + 12a3 ∂x3 6 ∂x3 ∂x3 12 ∂x3

214


⎛ a2 h21 + a1 h22 2 2 ⎜⎜⎜ ∂ ∂T ∂ ∂T h2 ∂2 ∂T ⎜⎝σ + (1 − σ) + + 12 h1 h2 ∂x2 ∂x1 ∂x1 ∂x2 2 ∂x22 ∂x1 ⎞ h22 ∂3 ∂T h21 ∂3 ∂T ⎟ h1 h2 ∂4 T h1 ∂2 ∂T 3 ⎟ + + + + o(|h| )⎟⎟⎠ + 2 ∂x12 ∂x2 6 ∂x23 ∂x1 6 ∂x13 ∂x2 4 ∂x12 ∂x22 ⎛ ⎛ ⎞ a3 h21 + a1 h23 ⎜⎜⎜ ∂2 ∂T ⎟⎟⎟ a3 h22 + a2 h23 ⎜⎜⎜ ∂2 ∂T h1 ∂2 ∂2 T 2 ⎜⎝ 2 ⎜⎝ 2 + + o(|h| )⎟⎠ + + 6h1 2 ∂x32 ∂x12 6h2 ∂x3 ∂x1 ∂x3 ∂x2 ⎛ ⎞ h21 a3 + a1 h23 ⎜⎜⎜ ∂ ∂T ⎟⎟ h2 ∂2 ∂2 T h1 ∂ ∂2 T 2 ⎟ ⎜ ⎟ + o(|h| ) + + + k ⎝ ⎠ 2 ∂x32 ∂x22 6h1 a3 ∂x3 ∂x1 2 ∂x3 ∂x12 ⎞ h22 a3 + a2 h23 ∂ ∂T h21 ∂ ∂2 T h31 ∂ ∂4 T ⎟⎟ 4 ⎟ ⎟ + + o(h ) + + k 1 ⎠ 6 ∂x3 ∂x13 24 ∂x3 ∂x14 6h2 a3 ∂x3 ∂x2 ⎞ ⎟ h2 ∂ ∂2 T h22 ∂ ∂3 T h32 ∂ ∂4 T 4 ⎟ + + + o(h2 )⎟⎠⎟ + Φ . 2 ∂x3 ∂x22 6 ∂x3 ∂x23 24 ∂x3 ∂x24 Using (2.157) and boundary conditions (2.161), the expression for error of approximation takes the form ⎛ 2 2 h2 ∂2 h2 ∂2 ⎜⎜ h ∂ 2 ∂T 2 ∂T + a2 + LT + ⎜⎜⎝ 1 2 + 2 2 + 3 2 + Ψ (0, 0, x3 ) = a1 h1 ∂x1 h2 ∂x2 12 ∂x1 12 ∂x2 12 ∂x3 ⎞ h23 ∂ ⎟⎟⎟ a1 h22 − a2 h21 ∂2 ∂T h2 ∂ h ∂ ⎟⎠ LT + 1 k LT + + LT + 2 12a3 ∂x3 3 ∂x1 6h1 3 ∂x2 ∂x2 ∂x1 a1 h23 − a3 h21 ∂2 ∂T a2 h23 − a3 h2 ∂2 ∂T a2 h21 − a1 h22 ∂2 ∂T + + + 2 2 6h1 6h2 6h2 ∂x3 ∂x1 ∂x3 ∂x2 ∂x12 ∂x2 h1 (a2 h21 + a1 h22 ) ∂3 ∂T h2 (a2 h21 + a1 h22 ) ∂3 ∂T + + 18h1 18h2 ∂x23 ∂x1 ∂x13 ∂x2 a2 h21 + a1 h22 ∂ ∂T ∂ ∂T σ + Φ + o(|h|4 ) = + (1 − σ) 3h1 h2 ∂x2 ∂x1 ∂x1 ∂x2 −f − ⎛ 2 ⎜⎜⎜ ⎜⎝g−1 + h1 ⎛ 2 ⎜⎜⎜ ⎜⎝g−2 + h2

h23 ∂2 f h23 ∂ f h22 ∂2 f h21 ∂2 f − − − k − 12 ∂x12 12 ∂x22 12 ∂x32 12a3 ∂x3

⎞ a1 h22 − a2 h21 ∂2 g−1 a1 h23 − a3 h21 ∂2 g−1 ⎟⎟⎟ h21 ∂ f ⎟⎠ − + + 6 ∂x1 12a1 12a1 ∂x22 ∂x32 ⎞ a2 h21 − a1 h22 ∂2 g−2 a2 h23 − a3 h22 ∂2 g−2 ⎟⎟⎟ h22 ∂ f ⎟⎠ − + + 6 ∂x2 12a2 12a2 ∂x12 ∂x32 ⎛ 4 ⎜⎜⎜ a2 h21 + a1 h22 ∂g−1 h22 (a2 h21 + a1 h22 ) ∂3 g−1 ⎜⎝ σ + + h1 h2 12a1 ∂x2 72a1 ∂x23


215

⎞ a2 h21 + a1 h22 ∂g−2 h21 (a2 h21 + a1 h22 ) ∂3 g−2 ⎟⎟⎟

4 (1 − σ) + ⎠⎟ + Φ + o(|h| ). 12a2 ∂x1 72a2 ∂x13 Therefore, to achieve error of compatibility condition approximation of amount of o(h41 + h42 + h43 ) for x1 = x2 = 0, it is sufficient to take Φ = f +

h23 ∂2 f h23 ∂ f h21 ∂2 f h22 ∂2 f + + + k + 12 ∂x12 12 ∂x22 12 ∂x32 12a3 ∂x3

⎡ 2 ⎢⎢⎢ ⎢⎣g−1 + h1 ⎡ 2 ⎢⎢⎢ ⎢⎣g−2 + h2

⎤ a1 h22 − a2 h21 ∂2 g−1 a1 h23 − a3 h21 ∂2 g−1 ⎥⎥⎥ h21 ∂ f ⎥⎦ + + + 6 ∂x1 12a1 12a1 ∂x22 ∂x32 ⎤ a2 h21 − a1 h22 ∂2 g−1 a2 h23 − a3 h22 ∂2 g−2 ⎥⎥⎥ h22 ∂ f ⎥⎦ + + + 6 ∂x2 12a2 12a2 ∂x12 ∂x32 ⎡ 4 ⎢⎢⎢ a2 h21 + a1 h22 ∂g−1 h22 (a2 h21 + a1 h22 ) ∂3 g−1 σ + + ⎣⎢ h1 h2 12a1 ∂x2 72a1 ∂x23 ⎤ a2 h21 + a1 h22 ∂g−2 h21 (a2 h21 + a1 h22 ) ∂3 g−2 ⎥⎥⎥ ⎥⎦ . (1 − σ) + 12a2 ∂x1 72a2 ∂x13

(2.170)

In what follows, approximation of compatibility condition for x1 = x2 = 0 with the error of o(h41 + h42 + h43 ) has the following form a2 h21 + a1 h22 2 2 a1 T x1 + a2 T x2 + a3 T x3 x3 + T x1 x2 + h1 h2 3h1 h2 a3 h21 + a1 h23 a3 h22 + a2 h23 k2 h23 T x1 x3 x3 + T x2 x3 x3 + kT x◦3 + Tx x + 6h1 6h2 12a3 3 3 k

h21 a3 + a1 h23 h2 a3 + a2 h23 T x1 x◦3 + k 2 T x2 x◦3 = −Φ , 6a3 h1 6a3 h2

(2.171)

where Φ is defined in (2.170). One may obtain in an analogous way also compatibility conditions for x1 = x3 = 0 and x2 = x3 = 0. 2.3.5 Problems In order to verify efficiency of the constructed approximation using the difference scheme (2.134), a model-type problem for the equation (2.157) with attached all possible boundary conditions will be solved in the space G {0 ≤ x ≤ 1 , 0 ≤ y ≤ 1, −0.5 ≤ z ≤ 0.5}. Below some of the boundary conditions are given. $ % 1. T | x=0;1 = 0.25 + (y − 0.5)2 2Hz4 , x, y , $ % (2.172) T |z=±0.5 = ± (x − 0.5)2 + (y − 0.5)2 H0.25 .

216


2.

$ % = −0.75 + (y − 0.5)2 2Hz3 , x, y , x=0;1 $ % T |z=±0.5 = ± (x − 0.5)2 + (y − 0.5)2 H0.25 .

∂T +T ∂x

3.

T |z=±0.5

∂T = −2Hz3 , ∂x x=0;1

4.

∂T = −2Hz3 , x, y , ∂x x=0;1 $ % = ± (x − 0.5)2 + (y − 0.5)2 H0.25 .

(2.173)

(2.174)

x, y ,

$ % = (x − 0.5)2 + (y − 0.5)2 (0.75 + H0.25), z=0.5 $ % T |z=−0.5 = − (x − 0.5)2 + (y − 0.5)2 H0.25 . (2.175)

∂T + 2HT ∂z

∂T = −2Hz3 , x, y , ∂x x=0.1 $ % = − (x − 0.5)2 + (y − 0.5)2 H0.25;

5.

T | x=−0.5 $ % ∂T = (x − 0.5)2 + (y − 0.5)2 1.5H . ∂z z=0.5

(2.176)

In all problems the following function serves as the right hand sides of equation (2.157) 3

3 2 2 f = − 8Hz + z (x − 0.5) + (y − 0.5) . (2.177) H The function

% $ T = (x − 0.5)2 + (y − 0.5)2 2Hz3

(2.178)

is the exact solution of the mentioned boundary value problems. They are solved for 2H = 0.1, 0.02, 0.01. For the mesh steps h1 = h2 = h3 = 18 the obtained solutions coincide with exact solution (2.178). It means that the proposed difference scheme (2.134) with the error o(|h|4 ) gives a good approximation to a being sought solution. On example of this model problem, an investigation of error computation of the functions Ψ1 (x, y), Ψ2 (x, y), δ1 (y), δ2 (x) in dependence of 2H is carried out. The given function are transformed to non-dimensional form using the following relations 2 l1 1 2H T , x = l, x, y = l2 y, z = 2Hz, λ = , T= T l2 α11 l1


217

αTij βi j , α = , (i, j = 1, 2), ij B11 αT11 αT11

βi j =

H ∂2 T ∂2 T ∂2 T B11 (2H)4 dz = Ψ (x, y) = z β11 2 + β22 2 + 2β12 × ∂x∂y ∂x ∂y l12 l22 −H

0.5 −0.5

⎞ ⎛ ⎜⎜⎜ ∂2 T ∂2 T ∂2 T ⎟⎟⎟ B11 (2H)4 −2 ⎟ z ⎝⎜β11 λ + β + 2β λ Ψ1 (x, y), dz = ⎠ 22 12 ∂x∂y l12 l22 ∂x2 ∂y2

⎞ 0.5⎛ 2 (2H)2 ⎜⎜⎜ ∂2 T ∂2 T ⎟⎟⎟ (2H)2 T −2 ∂ T T ⎟ Ψ2 (x, y) = 2 2 ⎜⎝ 2 + α22 λ −α λ = Ψ 2 (x, y). dz ⎠ 12 ∂x∂y l1 l2 l12 l22 ∂y ∂x2 −0.5

H δ1 (y) = 12β11 −H

H δ2 (x) = 12β22 −H

σ1 (y) =

σ2 (x) =

αT22

αT11

(2H)4 B11 zT dz = 12β11 l12

0.5 zT dz =

−0.5

(2H)4 B1 zT dz = 12β22 λ−2 l22

(2H)4 B11 δ1 (y), l12

0.5 zT dz =

−0.5

1 2H

1 2H

H −H

H −H

2H T dz = l1

2H T dz = l2

2

0.5 α22 −0.5

2 λ

−2

0.5

−0.5

(2H)4 B11 δ2 (x), l22

2H T dz = l1

2H T dz = l2

2

2

σ1 (y),

λ−2 σ2 (x).

The bars over non-dimensional quantities are omitted. In the case of boundary value problem (2.172) for equation (2.157) for 2H = 0.1, 2H = 0.02 the maximal error for the function Ψ1 (x, y) is in amount of 0.4% (0.5%). For the function Ψ2 (x, y), having its exact value equal to zero, the maximal absolute error is equal to 10−14 (10−15 ) for 2H = 0.1 (2H = 0.01). For boundary value problem (2.173)–(2.176) and equation (2.157) for 2H = 0.1, the maximal error of Ψ1 (x, y) achieved 0.1%, whereas for 2H = 0.01 it achieved 0.2%. The investigations have shown that the difference scheme (2.134) approximating (2.157) with various boundary conditions yields good approximation either for temperature field T (x, y, z) and the function Ψ1 (x, y), Ψ2 (x, y), δ1 (y), δ2 (x), σ1 (y), σ2 (x). The following boundary value problems are considered for the equation (2.157): 1.

$ % ∂T = 0 x, y , T |z=±0.5 = ± (x − 0.5)2 + (y − 0.5)2 H · 0.25 . (2.179) ∂x x=0.1 Heat source function f is defined via (2.177).

218

2.

3.


∂T ∂T = −2Hz3 , x, y ; = 0, ∂x x=0.1 ∂z z=−0.5 $ % T |z=0.5 = (x − 0.5)2 + (y − 0.5)2 H · 0.25,

(2.180)

f is defined in the form of (2.177). ∂T ∂T = 0 x, y , = 0, ∂x x=0.1 ∂z z=−0.5 $ % T |z=0.5 = (x − 0.5)2 + (y + 0.5)2 H · 0.25, f = 0.

(2.181)

In figures 2.2–2.9 temperature fields in cross sections y = 0.25 and y = 0.5, obtained during solutions of the boundary value problems (2.172) for f , (2.177), (2.172) for f = 0, (2.179), (2.177), (2.179) for f = 0, (2.180), (2.181) are reported. In all mentioned figures by (a) curves characterizing behaviour of temperature function T (z) in the cross section y = 0.25: x = 0 (curve 1), x = 0.125 (curve 2), x = 0.25 (curve 3) are denoted. Comparison of the Figures 2.2 and 2.3 shows, that for f = 0 temperature field inside G3 {0 < x < 1, 0 < y < 1, −0.5 < z < 0.5} is almost linear with respect to z (see curves 2, 3, 2 , 3 in Figure 2.3a), although on the space boundary for x = 0 temperature field is given in the form of z3 , i.e. inside of the space without the heat sources ( f = 0) and with boundary conditions (2.172) the temperature field is linear. Lack of heat sources ( f = 0) is exhibited by Ψ1 (x, y) behaviour. In Figure 2.4a the function Φ1 (x, y) is presented, corresponding to boundary value problem (2.172) for f = 0, and also the curves of Ψ1 (x, y) function in the cross sections y = 0.5 (curve 1), y = 0.25 (curve 2) y = 0.125 (curve 3) are reported. Curve 4 corresponds to the function Ψ1 (x, y) = −0.00065 being a solution of the problem (2.157), (2.172) for f (2.177). Owing to comparison of the curves 1,2,3 and 3, it is observed that for f = 0 the function Ψ1 (x, y) changes qualitatively and becomes nonlinear with respect to x and y. For the problem (2.157), (2.172) for f (2.177), the temperature field is nonlinear (Fig. 2.2) and Ψ1 (x, y) = const, whereas for f = 0 a temperature field is close to linear one (Fig. 2.3) inside of the space with respect to z, and Ψ1 (x, y) in nonlinear (Fig. 2.4a). Temperature field, corresponding to the problem (2.157), (2.179) for f (2.177), is reported in Fig. 2.5. Comparing the results in Fig. 2.5 and Fig. 2.2, one may conclude that heat isolation of edge shell surface does not influence a qualitative behaviour of temperature field with respect to z (it remains nonlinear along z). However, Ψ1 (x, y) becomes nonlinear with respect to x, y (see Fig. 2.4b) contrary to Ψ1 (x, y) = const for the problems (2.172), (2.157) for f (2.177). Temperature field of the problem (2.157), (2.179) for f = 0 is reported in Fig. 2.6. It is linear with respect to z inside of the space. Temperature field for the problem (2.180), (2.157) for f (2.177) is shown in Fig. 2.7. It is nonlinear with respect to z. Both Ψ1 (x, y) and Ψ2 (x, y) are nonlinear with respect to x, y (see Fig. 2.8).


4·10-4 T(Z) 2 1 3

Z 0.5

0 1` 2` 3`

a) -4 4·10 T(X;0.25;Z)

-0.5

0.5

Z

1 X -4 3·10 T(X;0.25;Z)

-0.5

0

0.5

X

Z

1

b) Figure 2.2. Temperature field distributions (see text for more details).

219

220


-4

4·10 T(Z)

1 2

3 -0.5 3`

Z 0.5

0 2`

3` 4·10

-4

a)

T(X;0.5;Z) 3·10-4

-0.5

0 0.5 Z 1 X

-4

4·10

-0.5

T(X;0.25;Z)

0 0.5 Z 1 X

b)

Figure 2.3. Temperature field distributions (see text for more details).

2.3 Solution of 3D Stationary Heat Transfer Equation y1(x) -4

6·10

3 3·10

1/8

x

0

y1(x,y) 6.5·10

-4

4

-4

1

2

-3·10-4

0.5

1/8

7/8

7/8

x

y

a) 2·10

-3

y1(x)

0

y1(x,y) -2·10 1/8

2

-3

0 1/8

7/8

7/8

x

x 0.5 1

3

1

1

y

b) Figure 2.4. Function Ψ1 (x, y) and its cross sections (see text for more details).

221

222

2 Stability of Rectangular Shells within Temperature Field -4

4·10 T(Z) 2 1 3

-0.5 3`

Z 0.5

0

2` 1`

a) 4·10-4

-0.5

T(X;0.25;Z)

0

0.5 X

Z

1

3·10-4 T(X;0.5;Z)

-0.5

0 0.5

X

Z

1


2.3 Solution of 3D Stationary Heat Transfer Equation 4·10-4 T(Z) 3

-0.5

2

1

Z 0.5

0 3` 2` 1`

a) -0.5

3.3·10-4

T(X;0.25;Z)

0

0.5

Z

1 X T(X;0.5;Z) -0.5

-4

3·10

0

0.5

Z

1 X

b)

Figure 2.6. Temperature field distributions (see text for more details).

223

224


1

2

2·10 T(Z)

1` 2`

3 3` 0.5

0

-0.5

a)

Z

T(X;0.5;Z)

1·10-3 1.5·10-3 0.5

Z

1 X

T(X;0.5;Z)

0.75·10-3 1.2·10

-3

0.5 Z X

1


2.3 Solution of 3D Stationary Heat Transfer Equation -3

4·10

1

(x)

225

2 3

-3

2·10

(x,y)

x 0.5

0

1

1 -3

-2·10

1/8

0 1/8

-3

-4·10

7/8

7/8

y x

a) 2

1/8

(x,y)

1/8

-3

7/8

7/8

2·10

(x)

2

2

y

x

3 0 1

0.5

x

-3

-2·10

b) Figure 2.8. Function Ψ1 (x, y) (a) and Ψ2 (x, y) (b) and their cross sections (see text for more details).

226

2 Stability of Rectangular Shells within Temperature Field 4·10-4 T(Z)

1

1`

2 2`

3

3` 0

0.5

a)

Z

T(X;0.25;Z) 3.3·10-4 4·10

-4

0.5

X

Z

1

T(X;0.5;Z)

-4

2.5·10

3·10

-4

0.5 Z 1 X


2.4 Algorithm for Difference Equations

227

In Fig. 2.9 the temperature field corresponding to the problem (2.157), (2.181) for f = 0 is given. Owing to this figure, during heat shell isolation from edge surfaces and the surface z = −0.5 and with a lack of heat sources ( f = 0) the temperature field in linear with respect to z inside the space. Owing to analysis of boundary problems (2.172), (2.179)–(2.181) for the equation (2.157), a lack of heat sources ( f = 0) is associated with almost linear temperature field with respect to z. Functions Ψ1 (x, y) and Ψ2 (x, y) are nonlinear with respect to x, y. An occurrence of heat sources of the type (2.177) is associated with nonlinear (with respect to z) temperature field occurrence.

2.4 Algorithm for Difference Equations 2.4.1 Construction of difference equations We begin with transformation of the system (2.39) into non-dimensional form using the following relations w = 2Hw,

F = B11 (2H)3 F,

q=

Bi j = Bi j B−1 11 , ai j = ai j B11 , αi j = βi j = βi j (B11 α11 )−1 , i, j = 1, 2, λ =

B11 (2H)4 q, l12 l22

αi j , i, j = 1, 2, α11

l1 2H , k1 = 2 ki , i = 1, 2. l2 li

(2.182)

The system (2.39) possesses the following non-dimensional form (bars are omitted): ∂ 2 w ∂2 F ∂2 F ∂2 F ∂ 2 w ∂2 F ∂ 2 w ∂2 F − q+ + 2 −k1 2 − k2 2 − 2 2 − 2 ∂x∂y ∂x∂y ∂y ∂x ∂x ∂y ∂y ∂x2 ∂4 w 1 −2 ∂4 w ∂4 w λ + λ2 B22 4 + 2(B12 + 2B66 ) 2 2 + 4 12 ∂x ∂y ∂x ∂y ∂4 w ∂4 w − Ψ1 (x, y) = 0, 2B16 λ−1 3 + 2B26 λ ∂x ∂y ∂x∂y3 2 2 ∂w ∂2 w ∂2 w ∂ 2 w ∂2 w ∂4 F + a22 λ−2 4 + k1 2 + k2 2 + 2 2 − ∂x∂y ∂y ∂x ∂x ∂y ∂x a11 λ2

∂4 F ∂4 F ∂4 F + (2a + a ) − 2a λ − 12 66 16 ∂y4 ∂x2 ∂y2 ∂x∂y3 2a26 λ−1

where x, y ∈ G {0 ≤ x, y ≤ 1},

∂4 F + Ψ2 (x, y) = 0, ∂x3 ∂y

(2.183)

228


0.5

λ−2 β11

Ψ1 (x, y) = −0.5

0.5 Ψ2 (x, y) = −0.5

∂2 T ∂2 T ∂2 T −1 zdz, + β + 2λ β 22 12 ∂x∂y ∂x2 ∂y2

2 2 ∂2 T −2 ∂ T −1 ∂ T dz. + α λ − α λ 22 12 ∂x∂y ∂y2 ∂x2

Boundary conditions for the system (2.183) have the following generalized form ∂w T T li w, , Mnn , εnn , T n , T 12 , M , N = 0, (2.184) ∂n where:

0.5 M = T

0.5 zT (x, y, z)dz,

−0.5

N = T

T (x, y, z)dz. −0.5

Recall that in (2.183) and (2.184) bars are omitted. The system (2.183) is rewritten into the form ∂2 w ∂2 F ∂2 w ∂2 F ∂4 w ∂4 w ∂4 w + k + − A x 4 + A xy 2 2 + Ay 4 = k1 + 2 2 ∂x ∂x ∂y ∂y ∂x ∂y2 ∂y2 ∂x2 ∂2 F ∂2 w ∂4 w ∂4 w + q + A16 3 + A26 + Ψ1 (x, y), ∂x∂y ∂x∂y ∂x ∂y ∂x∂y3 2 2 ∂w ∂4 F ∂4 F ∂4 F ∂2 w ∂2 w a x 4 + a xy 2 2 + ay 4 = − k1 + 2 − ∂x∂y ∂x ∂x ∂y ∂y ∂x ∂y2 2

k2 where: Ax =

∂4 F ∂2 w ∂4 F + 2a16 λ + 2a26 λ−1 3 − Ψ2 (x, y), 2 3 ∂x ∂x∂y ∂x ∂y

(2.185)

1 −2 1 2 1 (2B12 + 4B66 ) , λ , Ay = λ B22 , A xy = 12 12 12 a x = a22 λ−2 , ay = a11 λ2 , a xy = (2a12 + a66 ) , A16 = −2B16 λ−1 , A26 = −2B26 λ.

The partial derivatives occurred in (2.185) are approximated by difference relations with the error o(|h|4 ) applying Taylor series development with respect to h powers (h is the mesh step of the space Gh {0 ≤ xi , y j ≤ 1, xi = ih, y j = jh, i, j = 0, 1, . . . , N; N = 1h ). h2 ∂4 u ∂2 u ≈ u − + o(h4 ) = xx 12 ∂x4 ∂x2

% 1 $ −ui+2 j + 16ui+1 j − 30ui j + 16ui−1 j − ui−2 j + o(h4 ), 2 12h


229

∂4 u h2 ∂6 u 1 $ ≈ u − + o(h4 ) = 4 −ui j+3 + 12ui j+2 − 39ui j+1 + yyyy 4 6 6 ∂y ∂y 6h % 56ui j − 39ui j−1 + 12ui j−2 − ui j−3 + o(h4 ). The difference relations approximating derivatives with o(h4 ) error are denoted by

∂k l ∂xl ∂yk−l

ij

∂k u ∂xk

ij

= lhxk ui j o(h4 ),

= lhxl yk−l ui j + o(h4 ), k = 2, 4; l = 1, 2, 3,

where:

1 $ 72ui j + 20 ui+1 j+1 + ui+1 j−1 + ui−1 j+1 + ui−1 j−1 − 4 12h

38 ui+1 j + ui j+1 + ui−1 j + ui j−1 + 2 ui+2 j + ui−2 j + ui j+2 + ui j−2 −

ui+1 j+2 + ui+1 j−2 + ui+2 j+1 + ui−2 j+1 + ui−1 j−2 + % ui−2 j−1 + ui−1 j+2 + ui+2 j−1 , l x2 y2 ui j =

1 $ ui+1 j−2 − ui−1 j−2 + ui−1 j+2 + ui−2 j+1 − ui+1 j+2 + 24h2

% ui+2 j−1 − ui−2 j−1 + 10 ui+1 j+1 − ui−1 j+1 − ui+1 j−1 + ui−1 j−1

ehxy ui j =

Substituting the obtained difference relations into the system (2.185), the following system of nonlinear algebraic equation is obtained (its order is equal to nodes number in space Gn )

1 A x 56 + A xy 36 + Ay 56 wi j = A x l˜hx4 wi j + Ay l˜yh4 wi j + A xy l˜hx2 y2 wi j + 2 $ % % $ 1 k2 12h2 + ly2 wi j lhx2 Fi j + k1 12h2 + l x2 wi j ly2 Fi j − 24 1 h h l wi j l xy Fi j + (q + Ψ1 (x, y) 6h4 , 2 yx

1 (a x 56 + a xy 36 + ay 56)Fi j = a x l˜x4 Fi j + ay l˜y4 Fi j + a xy l˜x2 y2 Fi j + 2 2 %

$ 2 1 1 h (2.186) l xy wi j − k1 12h2 + l x2 wi j ly2 wi j − k2 l x2 wi j − Ψ2 (x, y)6h4 , 96 24 2 where: l˜x4 u, l˜x2 y2 u, l˜y4 u means that in this difference relation there is a lack of a term in central point i, j. Denoting by f1 (w, F), ϕ1 (w, F) the right hand sides of the system (2.186), depending on the function w, F values in mesh nodes, the system (2.186) can be recast into the form

230


Awi j = f1 (w, F) + (q + Ψ1 )6h4 , aFi j = ϕ1 (w, F) − Ψ2 6h4 ,

(2.187)

where: A = A x 56 + A xy 36 + Ay 56, a = a x 56 + a xy 36 + ay 56. Owing to difference relations of the operators lhxk u, for the system (2.185), approximation 25-points pattern is applied and two series out-contour nodes are required. The values of w and F in out-contour nodes are defined via boundary and compatibility conditions. The boundary value condition (2.53)–(2.56) accounting of compatibility conditions for the functions w and F (2.57)–(2.61) for x = 0 have the following non-dimensional form w = 0,

∂F ∂w = 0, F = 0, = 0, ∂x ∂x

∂2 w w = 0, = 12β11 ∂x2

0.5

∂F = 0, ∂x

(2.189)

T dz, F = 0,

(2.190)

zT dz, F = 0,

−0.5

∂w ∂2 F = 0, a22 2 = −α22 w = 0, ∂x ∂x

(2.188)

0.5

−0.5

∂2 w = 12β11 w = 0, ∂x2

0.5

−0.5

∂2 F zT dz, F = 0, a22 2 = −α22 ∂x

0.5 T dz.

(2.191)

−0.5

The derivatives, occurred in boundary conditions (2.188)–(2.191) are approximated by an error o(h4 ). The following two types of approximations are used to define w and F values of the out-contour points 2 ∂u 1

= −ui+2 j + 16ui+1 j − 30ui j + 16ui−1 j − ui−2 j + o(h4 ), 2 2 ∂x i j 12h

∂2 u ∂x2

= ij

∂u ∂x

∂u ∂x

= ij

1

− 20u + 6u + 4u − u 11u + o(h4 ), i+1 j i j i−1 j i−2 j i−3 j 12h2 = ij

1

−ui+2 j + 8ui+1 j − 8ui−1 j + ui−2 j + o(h4 ), 12h

1

3ui+1 j + 10ui j − 18ui−1 j + 6ui−2 j − ui−3 j + o(h4 ), 12h

where a point with index (i, j) lies on a space boundary, and (i+1, j), (i+2, j) denote out-contour nodes. For example, boundary value conditions (2.191) have the following difference representation for x = 0


231

wi j = 0, 0.5 1

zT dz = δ1 (y j ), wi+2 j + 16wi+1 j + 16wi−1 j − wi−2 j = 12β11 12h2 −0.5

0.5 1

zT dz = δ1 (y j ). 11wi+1 j + 6wi−1 j + 4wi−2 j − wi−3 j = 12β11 12h2 −0.5

Observe that to find w values in out-contour nodes, there are two equations with two unknowns wi+2, j and wi+1, j . The second equation yields wi+1, j , assuming wi j = 0 on the boundary x = 0: % 1 $ −6wi−1 j − 4wi−2 j + wi−3 j + 12h2 δ1 (y j ) , 11 % 1 $ = 80wi−1 j − 75wi−2 j + 16wi−3 j + 204h2 δ1 (y j ) , 11 0.5 δ1 (y j ) = 12β11 zT (x, y, z)dz, x = 0.

wi+1 j = wi+2 j

−0.5

The boundary condition w = 0,

∂w = 0, ∂x

can be approximated analogously 1

18wi−1 j − 6wi−2 j + wi−3 j , 3 1

wi+2 j = 120wi−1 j − 45wi−2 j + δwi−3 j . 3 These boundary conditions can be rewritten in the following generalized form wi+1 j =

wi+1 j = wi+2 j =

(c) 2 a(c) 2 wi−1 j + a3 wi−2 j + wi−3 j + c12h f (y) ,

1

a(c) 1

(c) (c) 2 a(c) 4 wi−1 j + a5 wi−2 j + a6 wi−3 j + c204h f (y) ,

1

a(c) 1

(2.192)

where the coefficients di(c) and c take the fully defined values depended on boundary condition type (Table 2.4). Similarly, boundary conditions with respect to function F are defined by (2.192), where F is given for (2.190), (2.191) in the following way α22 f = σ1 (y) = − a22

0.5 T (x, y, z)dz, x = 0, −0.5

232

2 Stability of Rectangular Shells within Temperature Field Table 2.4. Coefficients a(c) i and c with the corresponding boundary conditions. a(c) 1 11 3

∂2 u ∂n2 ∂u ∂n

a(c) 2 -6 18

a(c) 3 -4 -6

a(c) 4 80 120

1 f = σ2 (x) = − λ−2 a11

a(c) 5 -75 -45

a(c) 6 16 8

c 1 0

f (y) δ1 (y)

0.5 T (x, y, z)dz, y = 0.

−0.5

It is worth noticing that various boundary conditions (2.188)–(2.191) are obtained when in initial data the coefficients a(c) i , c, f are variated. This observation exhibits universality of the finite difference method in comparison with other methods, where owing to variation of boundary condition type the computation algorithm must be changed. 2.4.2 Stability problems In order to investigate a stability of shallow shells within a stationary temperature field, a method of direct solution of nonlinear equations governing their stress-strain state is applied. Critical loads (pure temperature, force, force and temperature) are defined via limiting points of the problem (load-deflection diagram). Both top and larger critical loads are defined. Process of solution of a statical problem of thermoelasticity contains of two parts. The first one includes temperature field T (x, y, z) definition, and it is reduced to solution of a boundary value problem (2.81) (see the Section 2.4). In result, the h temperature field T in 3D mesh space G3 {0 ≤ xi ≤ 1 , 0 ≤ y j ≤ 1, −0.5 ≤ zk ≤ 0.5} is obtained. Next, the following quantities are defined 0.5 ∂2 T ∂2 T ∂2 T zdz, 0 < x1 , y j < 1, Ψ1 (xi , yi ) = β11 λ−2 2 + β22 2 + 2β12 λ−1 ∂x∂y ∂x ∂y −0.5

0.5 Ψ2 (xi , y j ) = −0.5

α22 λ−2

2 ∂2 T ∂ 2 T −1 ∂ T dz, 0 < xi , y j < 1, + − α λ 12 ∂x∂y ∂x2 ∂y2 0.5

δ1 (y j ) = 12β11

zT (x, y, z)dz, x = 0, 0 ≤ y j ≤ 1,

−0.5

12β22 −2 δ2 (xi ) = λ B22

0.5

−0.5

zT (x, y, z)dz, y = 0, 0 ≤ xi ≤ 1,


σ1 (y j ) = −

α22 a22

0.5 T (x, y, z)dz, x = 0, 0 ≤ y j ≤ 1, −0.5

1 σ1 (xi ) = − λ−2 a11

233

0.5 T (x, y, z)dz, y = 0, 0 ≤ xi ≤ 1,

(2.193)

−0.5

in nodes of mesh space Gh 0 ≤ x1 ≤ 1, 0 ≤ y j ≤ 1 with application of Simpson’s formula. Now second part begins, which is focused on solution to nonlinear algebraic equations system (2.186), corresponding to the problem (2.183) with boundary equation (2.188)–(2.191). Nonlinear system of algebraic equations, obtained through finite difference method with approximation o(h4 ) is solved via nonlinear relaxation method. The system (2.187) is cast in the form wi j =

1 6h4

f1 (w, F) + K q0 + Ψ1 (xi , y j ) , A A

Fi j =

1 6h4 ϕ1 (w, F) − KΨ2 (xi , y j ), a a

(2.194)

where: Kq0 = q. In addition, equations approximating boundary conditions (2.188)–(2.191) are attached, in the following form wi+1 j = wi+2 j =

(c) 2 w + a w + w + cK12h δ (y ) , a(c) i−1 j i−2 j i−3 j 1 j 2 3

1

a(c) 1

(c) (c) 2 a(c) 4 wi−1 j + a5 wi−2 j + a6 wi−3 j + cK204h δ1 (y j ) ,

1

a(c) 1

wi j+1 =

1

(c) 2 w + a w + w + cK12h δ (x ) , a(c) i j−1 i j−2 i j−3 2 i 2 3

a(c) 1 1 (c) (c) 2 wi j+1 = (c) a4 wi j−1 + a(c) w + a w + cK204h δ (x ) , i j−2 i j−3 2 i 5 6 a1 1

Fi+1 j = (c) a2(c) Fi−1 j + a3(c) Fi−2 j + Fi−3 j + cK12h2 σ1 (y j ) , a1 1

Fi+2 j = (c) a4(c) Fi−1 j + a5(c) Fi−2 j + a6(c) Fi−3 j + cK204h2 σ1 (y1 ) , a1 1

2 Fi j+1 = (c) a2(c) Fi j−1 + a(c) 3 F i j−2 + F i j−3 + cK12h σ2 (xi ) , a1 1

Fi j+2 = (c) a4(c) Fi j−1 + a5(c) Fi j−2 + a6(c) Fi j−3 + cK204h2 σ2 (xi ) , (2.195) a1

234


where the coefficients c and c can take the value of 1 or 0. The values c = 0, c = 0 correspond to boundary conditions (2.188); c = 1, c = 0 correspond to (2.189); c = 0, c = 1 – (2.190); c = c = 1 – (2.191). Owing to the boundary conditions form (2.195), either boundary value problems of (2.188)–(2.191) type are solved or their combination on different or the same shell side(s) can be used. In this case the corresponding compatibility conditions in the points of a boundary condition change are added. The values i = j = 0 correspond to the central shell point x = y = 0.5, where the computational process starts. In the equation corresponding to this point (n−1) + hw (n - step number and contrary to all other equations, the value w(n) 00 = w00 with respect to w; hw - deflection step in the centre point) is given. Then the quantity K is defined, which is used in all other remaining equations of the system (2.194) solvable with respect to wi j , Fi j . Index i = w correspond to boundary nodes along straight line x = 0; j = N boundary nodes along contour y = 0; i = N + 1, N + 2 - out contour nodes; j = N + 2, N + 2 - out-contour nodes. Parameter q0 should be given, and it characterizes a load amount in the quantity defined by the critical load K(q0 +Ψ1 (x, y)). For q0 = 0 purely temperature-like problem is obtained. For q0 > 0 one deals with force and temperature loads, and for δ1 = δ2 = Ψ1 (x, y) = Ψ2 (x, y) = σ1 = σ2 = 0 purely force load occurs. Note that a computational process is carried out through iterational scheme until the values of wi j , Fi j on the previous and next iteration coincide within the assumed accuracy. Then a next step with respect to w is realized, i.e. the value = w(n) w(n+1) 00 0 + hw , in the central point is computed, an a calculus is repeated. In result of solution computation within M steps with respect to w, the dependence load-deflection is obtained, and upper and lower critical loads are obtained. 2.4.3 Reliability of obtained results The problem (2.183), (2.184) is going to be solved applying finite difference method with approximation error o(h4 ). It is well known that occurrence of high order of algebraic equations belongs to main drawback of finite difference method o(h2 ). However, this problem can be omitted, if higher order approximations are used. The latter allow to take less density mesh and to get a solution with assumed accuracy of 1.5 – 2 times faster then applying approximation o(h2 ). Let us cut the space G {0 ≤ x ≤ 1, 0 ≤ y ≤ 1} by squared mesh with a step h and let substitute all partial derivatives in (2.183), (2.188), (2.191) by the corresponding difference relation with errors o(h2 ) and o(h4 ). Owing to a change of differential problem into difference problem, a system of nonlinear algebraic equations is obtained, which has the following generalized form fi (x1 , x2 , . . . , xn ) = 0,

i = 1, 2, . . . , n,

(2.196)

where unknown variables x1 , x2 , . . . , xn represent unknown values of the functions wi j , Fi j in the mesh nodes. It is assumed that (2.196) includes also boundary conditions.


235

To solve the system of nonlinear algebraic equations (2.196) and iterational method of nonlinear relaxation is applied, which is governed by the following equation:

(k+1) (k) , xi , . . . , xn(k) , i = 1, 2, . . . , n . (2.197) xi(k+1) = xi(k) + ω fi xi(k+1) , . . . , xi−1 It is well recognized that for a given constraint into the initial nonlinear system (2.196) the iterational process (2.197) is convergent only if accelerating parameter ω ∈ (0, 1]. Experiment shows, that for both approximation (i.e. o(h2 ) and o(h4 )) the optimal interval of ω is the following one [0.25; 0.75] for the problem associated with boundary condition (2.188), and the interval [0.1; 0.5] for the problem associated with the boundary condition (2.191), where ω essentially depends on k1 and k2 . Owing to increase of k1 and k2 , an optimal parameter ω decreases. The Runge principle is used for the step h estimation, whereas for the difference scheme o(h2 ) for the space G {0 ≤ x ≤ 1, 0 ≤ y ≤ 1} the step h = 1/8 is used (for approximation o(h4 ) the step h = 1/16 is applied). Results for different steps are reported in Table 2.5 for the point x = y = 0.5. Also results of the Bubnov-Galerkin method are attached. A comparison of computation computer time are given for the boundary conditions (2.188), (2.191) in Tables 2.6, 2.7, correspondingly. As it is seen in Table 2.6, computation time for the problem (2.183), (2.188) with approximation o(h4 ) is less comparing with o(h2 ) of amount of 1.5–2 times, and for the problem (2.183)– (2.191) (see Table 2.7) the computational time amount of o(h4 ) scheme is two time less than that of o(h2 ) approximation. Hence, for rectangular shell computations the approximation o(h4 ) is recommended (see also [344]).

Table 2.5. Comparison of the finite difference methods with approximations o(h4 ) and o(h2 ) and the Bubnov-Galerkin approach. k1 = k2 = 10, o(h4 ) k1 = k2 = 10, o(h2 ) h q F h q F 1/6 38.8 0.32 1/8 38.06 0.32 1/12 38.8 0.32 1/16 38.12 0.32

Bubnov-Galerkin method q F 38.8 0.32

Table 2.6. Computational time intervals distribution for the boundary condition (2.188). wchw = 0.1 0.1–3.5 0.1–5.0 0.1–7.5

k1 = k2 10 18 24

(2.188) o(h4 ), 9 15 38

o(h2 ), 11 18 71

236


Table 2.7. Computational time intervals distribution for the boundary condition (2.191). wchw = 0.1 0.1–2.0 0.1–3.0 0.1–2.5

k1 = k2 0 24 36

(2.191) o(h4 ), 3 8 13

o(h2 ), 6 12 33

2.4.4 Transversal load Owing to stability computations of flexible anisotropic shells during rotation of the coordinate system of amount of angle ϕ, the relations (2.8)–(2.10) are used to compute coefficients of equation (2.39). For a shell with geometrical parameters k1 = k2 = 24, subjected to an action of uniformly distributed load, the boundary value problems (2.54), (2.55) are solved. In Fig. 2.10 dependencies “deflection-load” in the shell centre q (w(0.5; 0.5)), obtained for the (2.54) problem (solid curve), (2.55) (dashed curve) for ϕ = 0◦ , 30◦ , 45◦ , 60◦ , 90◦ are reported. Curves, corresponding to the mentioned ϕ values, are denoted by digits 1, 2, 3, 4, 5. For ϕ = 30◦ and 90◦ , ϕ = 45◦ and 60◦ the curves load-deflection coincide (in Fig. 2.10 curve 3 coincides with curve 4, curve 2 coincides with curve 5) for the problem (2.54), (2.55). Increasing value from 0◦ to 53◦ , a value of top critical load increases for the problem (2.54), (2.55), whereas increasing ϕ form 53◦ to 90◦ , it decreases. Shells with k1 = k2 = 24 with boundary (2.55) loose their stability for smaller amount of top critical load than those with boundary condition (2.54). Here the upper critical load dependence on angle ϕ is displayed. Solid curve corresponds to problem (2.54), whereas dashed curve corresponds to problem (2.55). Maximal top critical load value corresponds to the angle ϕ = 53◦ . Development of shells stress-strain state, using an example of shell k1 = k2 = 24, in dependence of boundary conditions and angle ϕ can be traced in Fig. 2.11a for the points x = y, and in Fig. 2.11b for the points y = 0.5 for boundary conditions (2.54), (2.55). Diagrams of fundamental functions w, w

xx , F, F

xx corresponding to the problem (2.54) for w(0.5; 0.5) = 1.5 are represented by solid curves. Similarly, diagrams of these functions for w(0.5; 0.5) = 1.5 corresponding to problem (2.55) are marked by dashed curves. Solution of the problems (2.55), (2.54) for angles ϕ = 0◦ , 30◦ , 45◦ are marked by digits 1, 2, 3, correspondingly. Comparing solutions with the same boundary conditions but for different angle ϕ, one may conclude that diagrams of the functions w, w

xx related to different angle ϕ coincide, but the diagrams of the function F, F

xx are different. In the latter case, larger absolute value of ϕ corresponds to larger value of F, F

xx in the shell centre. Owing to analysis of shell solutions with boundary conditions (2.54), (2.55) for the same values of angle ϕ, the diagrams of the functions w

xx essentially differ either along x = y or y = 0.5. In the latter case, a difference along y = 0.5 is essential not only in the shell centre, but in its contour also. Diagrams F, F

xx for the problems (2.54), (2.55) computed in the shell centre and corresponding to the problem (2.55) are in absolute value larger of amount of 25% for the function F, and on amount of 10% for the function

237 W

X

1 0

0

100

200 100

200

300 q

3,4

300

q

1

1

2,5

2,5

30

3,4

60

2

90

j

3

Y`

0

Y

5

2

j

X`

4


Figure 2.10. “Load-deflection” dependence in the shell centre.

F

xx than in the problem (2.54). The following data are used during computations: kg 5 kg 5 kg G = 0.42 · 105 cm 2 , ν2 = 0.07, E 1 = 2.1 · 10 cm2 , E 2 = 1.6 · 10 cm2 (glass-plastic material). Note that reported calculations can be used for optimal construction design in the sense of maximal critical load in dependence on angle ϕ.

238

2 Stability of Rectangular Shells within Temperature Field -30 1.2

1,2,3

-12

F

2

3 1.5

-15

W 1,2,3

3

0,6

x=y 0.5

0

3

2

1

1,2,3

1,2,3

3 2

1

1

1

0

W``xx a)

x=y 0.5

0

F``xx

-30 1.2

F

2

1,2,3 3

-15

1.5 1,2,3

-6

2

1,2,3 1,2,3

0,6

1

1 2

3 2 3

1

1 2

x 0.5 0

x 0.5

F``xx

W``xx b)

Figure 2.11. Fundamental functions w, w

xx , F, F xx for x = y (a) and y = 0.5 (b) in the shell centre (boundary conditions (2.54), (2.55)).

2.4.5 Different boundary conditions Consider isotropic plates and shells (ν = 0.3) for different boundary conditions (2.188), (2.191) along their sides under action of uniformly distributed transversal load. Digit 9 denotes ball type conditions (2.191), digit 5 denotes clamping edge of


239

the type (2.188), digits 1–4, 6–8 denote their combinations along a shell side. The mentioned notation is used in Figures 2.12, 2.13. In Figures 2.12a, 2.13a the graphs “load-deflection” in the plate centre (0.5; 0.5) for all clamping types, denoted by digits 1–3, are reported. The curve number is the same as boundary type number. Comparing the curves in Fig. 2.12a one may conclude that ball type fixation on the corner points (curve 4) practically does not influence the plate stress-strain state (curves 4 and 5 coincide). This conclusion is supported by results reported in Figure 2.13b, where the diagrams of fundamental functions w, w

xx , F, F

xx in the point (0.5; 0.5) along the diagonal x = y are given. Now the curves are denoted in the same way as in Fig. 2.12. Decreasing length of ball supported contour, in the neighbourhood of corner points (curves 3, 4) the graph “load-deflection” approaches curve 5, corresponding to clamped contour. In Fig. 2.12b the graphs for clamping 4 and 9 differ in shell centre and in neighbourhood of the contour. Owing to decrease of clamped contour length in direction 1–4, the w

xx value in the plates centre increases with respect to absolute value and achieves maximum for fully clamped contour (curve 5). Comparing the curves in Fig. 2.13a, it is seen, that clamping of corner points has influence on the stress-strain plate state (≈ 20%, compare Figures band 9). Owing to decrease of the length of clamped contour, in the neighbourhood of corner points (curves 6, 7), the graph “load-deflection” tends to curve 9, i.e. it describes the plate supported by balls along whole contour. Increasing length of clamped contour along a side in vicinity of corner points on amount of 50% (curve 7), a plate strength is increased on amount of 50% comparing with ball supported plate. In Fig. 2.13b, curves of the fundamental functions w, w

xx , F, F

xx in the point w(0.5; 0.5) = 1.5 along the diagonal x = y are shown. A curvature changes its sign. Curvature values w

xx depends on the length of clamped contour along a side. A plate with large interval of camped contour (curve 8) is associated with large curvature w

xx absolute value in the plate centre. Absolute value of the function F decreases with increase of clamped contour length along a shell side. Function F

xx changes its sign. Increasing length of shell clamped contour, the graphs F

xx approach curve 5, governing behaviour of F

xx for a plate with clamped contour. In Fig. 2.14 the graphs “load-deflection” in the central shell point for boundaries denoted by digits 1-5, 9 are reported. Solid curves correspond to a shell with geometrical parameters k1 = k2 = 24, whereas dashed curves correspond to k1 = k2 = 28. Let us denote shell side lengths by l0 , a length clamped interval along a side by l2 , a length of ball-supported part along a side by l1 = l0 − l2 . Owing to behaviour of curves reported in Fig. 2.14, clamping in the middle of shell side essentially influences the upper critical load, since it is decreased on amount of 30% in comparison with ball-supported shell (compare curves 2 and 9, 3 and 9). Further increase of clamped contour length l2 ≥ 34 l0 in the middle of a shell side does not influence a value of the upper critical load (curves 4 and 5 fully coincide; curve 3 differs from 5 amount of 1%). For a shell with geometrical parameters k1 = k2 = 18 (dashed curves) an increase of clamped contour length in the middle of a side l2 ≥ 14 l0 does not influence stress-strain shell state, governed by “load-deflection” dependence in

1.5 b)

xy 0.5

5,4 1

F 0

45

90

135

q

0.5

9 1 2 3 4 5

a)

1.0

9

1

2

4,5

3

W

1.5

0

-0.1

-0.2

-0.3

0

0.6

1.2

1.8 W

2

1

9

9

3

5,4,3

2

x=y 0.5

2

9

4,5

5,4

3

3

2

9

1

1

2

F``xx

4,5

0

1.5

3

6

W``xx

0

-6

-12


-18

240

Figure 2.12. “Load-deflection” dependence in the shell centre (a) and fundamental function for x = y and y = 0.5 (b) (see the text for more details).

1.5

0

1.5

241

F 0

45

90

135

q

0.5

9 6 7 8 5

a)

1

9

6

5

7

8

W

1.5

0

-0.1

-0.2

-0.3

0

1.2

0.6

6

1.8 W

9

7

6

7

9

5,8

5,8

b)

x=y 0.5

x=y 0.5

5,8

7

5

7

6

9

F``xx

3

W``xx 6

9

6

8

0

-6

-12

-18


Figure 2.13. “Load-deflection” dependence in the shell centre (a) and fundamental function for x = y and y = 0.5 (b) (see the text for more details).

242


the central point (0.5; 0.5) (curves 3, 4, 5 coincide with each other). For l2 ≤ 14 l0 a shell looses its stability, and a value of upper critical load is decreased on amount of 15% in comparison to the ball-supported shell (compare curves 1 and 9). In Fig. 2.14–2.18 graphs of fundamental functions w, w

xx , F, F

xx for the shell k1 = k2 = 24, corresponding to boundary condition 1-4 along lines x = y and y = 5 are shown. In Figures 2.19, 2.20 graphs of fundamental function w, w

xx , F, F

xx for the shell k1 = k2 = 24 corresponding to clamped contour 5 and ball-supported shell 9 are shown. Curves on Figures 2.15–2.20 are denoted by the same number as a type of fixation. Index identifies a deflection in the shell centre with a corresponding curve. For example, 52 corresponds to a shell with clamped contour, and deflection value in the central point w = 1.7. In Fig. 2.21 graphs “load-deflection” in the central shell point for fixations denoted in the figure by points 5-9 are shown. Solid curves correspond to a shell with geometric parameters k1 = k2 = 24, whereas dashed zones correspond to parameters k1 = k2 = 18. Owing to analysis of Figure 2.21, clamping of corner points have an important influence on upper critical load value, decreasing it in comparison with ball supported shell on 13b (compare curves 6 and 9). An increase of clamped contour length l2 , beginning from corner points, up to 34 l0 decreases upper critical load value up to 40% for the shell k1 = k2 = 24 (compare the curves 7, 8, 9). Further increase of the clamped contour length l2 ≥ 34 l0 does not influence upper critical load value. Curves 5 and 8 up to deflection w(0.5; 0.5) = 1.7, coincide. In Fig. 2.21a the dependencies of top (curve ) and below (curve 1) critical loads on clamped contour length in the shell middle side are reported. A development of shell stress-strain state with k1 = k2 = 24 dependence on fixation type along a shell side can be traced in Figures 2.22–2.24, where graphs of fundamental functions w, w

xx , F, F

xx , corresponding to fixation 6-8 along both diagonals x = y and y = 0.5 are reported. Again, curves notation is linked with their fixation types and indeces notation is similar to the previous discussed case. In Figure 2.25 curves of equal deflections for the shell k1 = k2 = 24 for all fixation types 1-9 for w(0.5; 0.5) = 1.5 are shown. Figure 2.25 displays, how a shape of deformable shell surface is changed owing to fixation contour type. The curves differ from those occurred in problems 1, 2, 5, 6, 8, 9, and the same observation holds for deflection level values. For the problems 4, 5, the curves of equal deflection coincide either in a shape or in a magnitude. For the problems 6, 7, 9, the curves are closed in shape, but are different in magnitude. Comparing curves 4 an 5 in Fig. 2.14, 9 and 6 in Fig. 2.19 one may conclude that fixation type of corner points essentially influences a value of top critical shell load (≈ 15%), whereas a ball-type support of the shell corner points has no influence on stress-strain shell state (curves 4 and 5 coincide). In Figure 2.26, the dependencies “load-deflection” in the central point q(w(0.5; 0.5)) for the shell with k1 = k2 = 36 for all considered type of fixation along a shell side 1-9 are reported. The curves, similarly to those in Figs. 2.14, 2.21, have the same number as fixation type. It should be emphasized that the shell k1 = k2 = 36 is

4 5 l0 0

100

200

300

q

1.1

2

1

l0

5,4,3

9

9 1 2 3

3,4,5

2.2

1

2

3.3

9

4.4

243

W


Figure 2.14. Load-deflection function in the shell centre for the boundary condition 1-5, 9 (solid curves - k1 = k2 = 24; dashed curves k1 = k2 = 28; see text for more details).

more sensitive to variation of fixation along a shell side, than a shell with k1 = k2 = 24.

244

2 Stability of Rectangular Shells within Temperature Field 1.0 -50 F

-10 13

13 25

0

W

13

12

12 11

12

0.5 -25

11

x=y 11

W``xx

11

x=y

00

0.5

13

12

0.5

F``xx

a) 1.0 -50

-10

F

13

0.5 -25

13 11 0

x

12

11

00

0.5

-5

11

11

12

12

13

12

13 25

0

x

0

0.5 11

W``xx

5

25

13 10

11(W=0.9), 12(W=1.6), 13(W=2.5) b)

F``xx

15


xx , F, F xx for the shell k1 = k2 = 24 with boundary conditions 1-4 for x = y (a) and y = 0.5 (b) (see text for more details).

A ball-type support in the middle of a shell side yields an increase of top critical load already for l1 = 14 l0 in comparison to fully clamped shell in amount of 5%. Further increase of the ball-type support length in the middle shell side increases shell stability, and for l1 = 34 l0 an increase of top critical load (in comparison to clamped shell) is in amount of 60%. The ball-type clamping in contour points for l1 = 14 l0 practically does not influence a shell stress-strain state. Curves 4 and 5 coincide, and for l1 = 12 l0 the value of upper critical load increases on amount of 1% in comparison to the clamped contour 5. Further increase of ball-type support

2.4 Algorithm for Difference Equations 0.6

0.4

23

23 22

W

22

24

F

-50

24

25

245

24

22

-25 0.2

23 24

-8

23

21

-4

22 21

21 0

x=y

21 0

0.5

x=y

0

0.5

F``xx

0

W``xx a) 0.6

W

23

22,24 22

-9 23 24

21

-25

21

22

22 21 0

0.3

23 24

23

-50

24

3

F

21

x

0

0.5

0

x

0

0.5

25

W``xx

21 9

50

21(W=1), 22(W=1.7), 23(W=2.2), 24(W=2.7) b)

23 24

F``xx

18



interval l1 ≥ 12 l0 increases the value of the top critical shell load, and for l1 = 34 l0 and it is in amount of 10% in comparison to clamped contour.

246

2 Stability of Rectangular Shells within Temperature Field -50 0.7

34 33 25

32

33

34

33

31 0

x=y

31

31

32

32

31

x=y

0

0.5

0

0.5

34

32,34

-25

W

-7

33

F

F``xx

0

W``xx a)

-50 0.7

34 33

F 33

33,,34 2.5

W

33

34

31

31

31

32 0

-25

32

32

-7

34 32

31

x

0 0

0.5

x

0

0.5

25

W``xx

7

50

31(W=0.9), 32(W=1.8), 33(W=2.2), 34(W=2.6) b)

F``xx

14



In Figs. 2.27a, b the graphs dependence of top critical load on fixation type for shells k1 = k2 = 36 (curve 1), k1 = k2 = 24 (curve 3) are reported. Figure 2.27a corresponds to fixation 1-5, 9 (a length of ball-type supports l1 is variated beginning from corner points), whereas Fig. 2.27b corresponds to fixation 5-9 (a length of ball-type support l1 is variated beginning from the middle of a side). Owing to behaviour of curves 1 and 2 in Fig. 2.27a, an increase of ball-support length up to l1 ≤ 12 l0 beginning from corner points, slightly influences a shell stress-strain


247

-50

-10

44 43 25

W

43

44

42

42 41

0

x=y

0.5

F

41

W``xx

44 -5

42

43,44

41

41

0

0.5

42,43

43

-25

x=y

0

0

0.5

F``xx

a)

-50

-10

44 43 25

W

43

-25

42

44

0.5

42 41 0

42

F

43,44

0

0.5

0

-5

42

41

x

43 44

41

41

x

0

0.5

-25 5

W``xx

-50 10

41(W=0.9), 42(W=1.8), 43(W=2.2), 44(W=2.6) b)

15

F``xx



state. For l1 ≥ 34 l0 an upper critical load value increases in comparison with fully clamped shell, and for ball-supported shell an increase of upper critical load value is in amount of 68% in comparison to clamped shell (for k1 = k2 = 24). For the shell k1 = k2 = 36 its stability increases for l1 ≥ 14 l0 , and for ball-supported shell an

248

2 Stability of Rectangular Shells within Temperature Field 0.6 -60

54

W

-9

53,54

-45

53 3

F

53

0.3

53 54

52

51

1.5

52

52 51

51

0

x=y

0

0

0.5

0

0.5

-3

51

-15

x=y

-6

54

52

-30

W``xx

W``xx

3

a) -50

3

W

53 52 51 0

0.4

53

54

F

53,54 -8

52 52

-25

52

0.2

51

51

x

0 0

0.5

54

51

x

0

0.5

25 54

W``xx

8

50

51(W=1), 52(W=1.7), 53(W=2.2), 54(W=2.6) b)

53 54

F``xx

16


xx , F, F xx for the shell k1 = k2 = 24 with boundary conditions 5, 9 for x = y (a) and y = 0.5 (b) (see text for more details).

increase of upper critical load is on amount of 70% comparing with a clamped shell. In Fig. 2.27b it is seen, that an increase of a length of ball-support in the middle of a side with l1 = 14 l0 (for shell k1 = k2 = 36), l1 = 12 l0 (for k1 = k2 = 24), l1 = 34 l0 (for k1 = k2 = 18) influences the value of upper critical shell load. Therefore, ball-support in the middle of a side more essentially influences on shell stability,


249

-80 1.4

6

-14

F

-60

W

4

92

93

94

93

91

91

-40

94

92

0.7

-7

94

93

93

-20

2 92 91 0

x=y

92

94 91

x=y

00

0.5

W``xx

F``xx

0.5

0

20

a)

1.4

6

W

93

94

-14

F 92

92 93 94 91

93

0.7 -30

3 92 91 0

-7

91

94 93

91

92

94

x

00

0.5

x 0.5

F``xx

0

91(W=0.8), 92(W=1.3), 93(W=3.5), 94(W=5.8)

W``xx b)


xx , F, F xx for the shell k1 = k2 = 24 with boundary conditions 5, 9 for x = y (a) and y = 0.5 (b) (see text for more details).

than ball-type support on a shell side ends. In addition, owing to increase of the parameters k1 , k2 , the value of l1 decreases, for which the value of top critical shell load begins to increase.

W 0

100

200

300

q

5

8

5,8

10

6

6

9

9

7

5 6 7 8 9

5

20

5

8

30

6

8

40

0

150

300

q

1

0.5

1

2

7

1.0

50


2

250

Figure 2.21. Load-deflection function for the boundary condition 5-9 and the dependence qcr (R1 /R2 ) for k1 = k2 = 24 (solid curves) and k1 = k2 = 18 (dashed curves).

2.4 Algorithm for Difference Equations 75

-75

W 64

1.2

64

50

F

-12 62

63

61 0.6

62 61 0

-6

63 64

62 61

x=y

0 0

x=y 0.5

W``xx

0

0.5

F``xx

25

6

-75

W

1.2

5.0

-50

64

63

62

0.6

63

62 61

64

-25

63 62 61

F

61

64 63

0

61

64

-25

25

2.5

62

63

-50

63

7.5

251

64

62

x

61

0 0

0.5

x 0.5

F``xx

0

61(W=0.9), 62(W=1.4), 63(W=3.6), 64(W=5.8) 25

W``xx


xx , F, F xx for the boundary conditions 6-8 for x = y and y = 0.5 (see text for more details).

In Figure 2.27c, the curves of top critical load value dependence on magnitudes of geometrical shell parameters k1 , k2 , for the following fixation types: 1, 6, 7, 9 (a curves number corresponds to fixation case) are shown. The results displayed in Figure 2.27c show that the shell with k1 = k2 = 28 possesses the same value of top

252


4.5

W

1.0 -45

74

74 73

3.0

71 0

72

x=y

00

0.5

-5 71

71

71

x=y

73

72 74

-15

72

72

73 74

-30 0.5

73 1.5

-10

F

0

0.5

15

W``xx

F``xx

-75

5

74 5.0

W

74

-50

73

1.0

F 72

73 2.5

72

72

-25

0.5

-10

72 73

73

74

-5

74

71

71 71 0

x

0.5

00

W``xx

x 0.5

F``xx

0

25

71(W=0.9), 72(W=1.6), 73(W=3.0), 74(W=4.7)



critical load for fixation 1 and 6, whereas the shell with k1 = k2 = 32 - for fixations 1 and 9.

2.5 Computations of Plates and Shells in a Temperature Field 2.5.1 Stress-strain state In this section, investigation of influence of geometric parameters k1 , k2 on the temperature field T (x, y, z) and functions Ψ1 (x, y), Ψ2 (x, y) is carried out. Stationary 3D

2.5 Computations of Plates and Shells in a Temperature Field

253

-60

84 0.8

30

W

81

0

x=y

81

x=y

0

0.5

-4

81

81

82 84

84 82,84

0.4

82

-8

83

82

83

-30

83

83

F

0

0.5

W``xx

F``xx

4

-60

84

30

W

0

F

-30

83

84 83 82 81

0.8

0.4

82

82

83

84 83

82

-4 81

81

x

81 00

0.5

W``xx

30

-8

84

x

0

0.5

F``xx

4

81(W=0.9), 82(W=1.4), 83(W=2.2), 84(W=3.0)



heat transfer equation for shallow shells (2.157) is solved through difference scheme (2.133) with an error o(|h|4 ) and with occurrence of the form 2k ∂T ∂z , where either 1 k = 2 (k1 + k2 ) or k = 0. As numerical parameters for k1 , k2 ≤ 24 on the temperature field is weak (less than 1%), however, an increase of these parameters increases this influence. The functions Ψ1 (x, y), Ψ2 (x, y) also differ slightly for variations of k1 , k2 ≤ 24. The graphs “load-deflection” obtained accounting or not the term 2k ∂T ∂z , practically coincide.

254

2 Stability of Rectangular Shells within Temperature Field 9

1

2 0.3

6

8

0.4

6

8 0.6 7 1.1 43 1.

6 0.3 0 0.9 34 1.

6 0.3 0 0.9 34 1.

1.5

1.5

1.5

3

7

8

5

0.4

1.5

6 0.9 35 1.

0 0.6 9 1.0 40 1.

6 0.3 1 0.9 34 1.

1.5

3

1.5

4

5

6

4

1.3

1.3

1.5

09

1.0

7 0.5 3 1.0 38 1.

1.5

1.5

Figure 2.25. Curve of equal deflections for the shell k1 = k2 = 24 for the boundary conditions 1-9 and w(0.5; 0.5) = 1.5.

Consider a plate under action of uniformly distributed transversal load and temperature field. In Fig. 2.28a, the curves “load-deflection” in the centre of the plate clamped along contour (2.190) and subjected to an action of purely force load (curve 1), load and temperature field (2.179) (curve 2), load and temperature field (2.181) for f (2.177) (curve 3), load and temperature field (2.181) for f = 0 (curve 4),

2.5 Computations of Plates and Shells in a Temperature Field 840

255

q 1

6

630

9 8

7

420 2 4

8

5

210

W 1.5

3

4.5

6

Figure 2.26. Load-deflection function in the shell (k1 = k2 = 36) centre for the boundary condition 1-9.

and also for ball-supported plates along contour (2.189) under action of load and temperature field (2.181) for f (2.177) (curve 5), load and temperature field (2.181) for f = 0 (curve 6), and load (curve 7) are reported. Comparing the curves 2, 3, 4, with 1, one observes that for clamped plate along its contour, the temperature field

256


830

800

q`kpu

q`kpu 1

1

415

400 2 2

3

l1

l1 0

0.5

1

0

0.5

l1

l1

a) 850

1

b) q`kpu

1

9

425

6

7

0

18

36 Kx

c) Figure 2.27. Top critical load for the boundary conditions 1-5, 9 (a), 5-9 (b), and qncr (k x ) (c) for different boundary conditions (see text for more details).


257

(2.179) (corresponding to heat isolation of plate sides surfaces) yields an increase of shell strengthness on amount of 7%, and temperature field (2.181) for f = 0 corresponding to heat isolations of plate sides surfaces and the surface z = −0.5 (without heat sources, the plate strength is decreased on amount of 8% in comparison with pure force). The temperature field (2.177), decreases strength of a shell with clamped contour in amount of 40% in comparison with an action of purely forcing load. Owing to the curves 3 and 4, an occurrence of heat sources of the form (2.177) for the temperature field (2.181) decreases shell strength in amount of 35%. Occurrence of heat sources (2.177), for the temperature field (2.181), increases strength of a ball-supported plate along surface in amount of 40% (see curves 5 and 6). An action of temperature field (2.181) without heat sources ( f = 0) on a ballsupported plate is rather weak, since curves 6 and 7 coincide. In Figure 2.28b graphs of functions w

xx , F

xx corresponding to w(0.5; 0.5) = 1.5 along the axis y = 0.5 are displayed. Curves have the same notations as in Figure 2.28a. An action of temperature moment on ball-supported plate contour, corresponding to temperature field (2.181) for f (2.177) is presented also in Figure 2.28b (see curves 5-7). All results discussed in the above are obtained for q0 = 0.01. The results given in Figure 2.28a exhibit an influence on fixation type on plate thermal-stress state (compare curves 3 and 5, 4 and 6). For the plate, being under action of temperature filed (2.181) for f (2.177), fixation of the contour (2.190) increases plate strength ability in amount of 55% in comparison to ball-support (2.189). On the other hand, for the plate being under action of temperature field (2.181) for f = 0, fixation of the contour (2.190) increases plate strength in amount of 230%, whereas purely forcing load action increases plate strength in amount of 260% in comparison to ball-support (curves 1, 7). Let us analyse now shells k1 = k2 = 18 and k1 = k2 = 24, ball-supported along their contours (2.189), and subjected to an action of uniformly distributed load (q0 = 0.01) and temperature field, corresponding to the problem (2.181) for f (2.177), (2.181) for f = 0, and also to the problem (2.179) for f (2.177). In Figure 2.29 graphs “load-deflection” in the centre shell point q(w(0.5; 0.5)), being under action of purely forcing load (curve 1), load and temperature field (2.179) (curve 4), load and temperature field (2.181) for f (2.177) (curve 2), load and temperature field (2.181) for f = 0 (curve 3), are reported. The curves related to shell k1 = k2 = 24 (k1 = k2 = 18) are solid (dashed). Owing to results shown in Figure 2.29, a temperature field (2.179) weakly influences upper critical load value in comparison to an action of only forcing load, decreasing it in amount of 1.5%. For the shell k1 = k2 = 24, the temperature field (2.181) for (2.177), decreases the value of upper critical load in amount of 13%, whereas the value of lower critical load is increased in amount of 13% in comparison to an action of only forcing load. The temperature field (2.181) with a lack of heat sources ( f = 0) decreases upper critical load value in amount of 8%, whereas lower critical load value is not changed comparing to an action of purely forcing load. In what follows, comparing the curves 2, 3, one may conclude that occurrence of heat sources f (2.177) for the temperature field (2.181) decreases shell k1 = k2 = 24 stability: top (low) critical load value decreases on

258

2 Stability of Rectangular Shells within Temperature Field 170 q

2 1 4

85

3

5

W

0

1

a) 80

6,7

1.5

W``xx 1,2

60

40

F``xx

1,2

4

2

20 6,7

5

x=y 0.5

0

0

6,7

-20

-2 5

-4

b)

-6

Figure 2.28. Load-deflection graphs (a) and the functions w

xx , F xx (b) for different plate excitations (see text for more details).


259

amount of 5% (14%). A similar behaviour occurs for the shell k1 = k2 = 18. Graphs of the fundamental function w, w

xx , F, F

xx along the axis y = 0.5 for the shell k1 = k2 = 24 under an action of the temperature field (2.181) for f (2.177) and for f = 0, and forcing load, are shown in Figure 2.30a. Description of the curves is the same as in Figure 2.29. Index associated with a number denotes deflection value in the central point corresponding to this curve, for instance, 21 (w = 1), 22 (w = 2), 23 (w = 3). Consider now the shell k1 = k2 = 18, ball-supported along its contour (2.189), within a temperature field. In Figure 2.31, the dependencies “load-deflection” in the central shell point (0.5; 0.5) under an action of temperature field and forcing load are reported. Curve 1 corresponds to the shell under an action of only forcing load, curve 2 corresponds to the shell under actions of forcing load and temperature field (2.178), curve 3 - load and temperature field (2.179), curves 4, 5 - load and temperature field (2.180). The corresponding temperature field are shown in Figures 2, 5, 7. The curves 2, 3, 5 are obtained for the value of q0 = 0.01, curve 4 corresponds to q0 = 0.02. Comparing the curves given in Figure 2.31a, one may conclude that heat isolation of the shell surfaces sides slightly influences thermo-stress state (curves 2 and 3 graphically coincide). Heat isolation of upper stability (upper (lower) critical load value increases in amount of 10% (50%) - see curves 4 and 2). An increase of the load amount from 0.01 to 0.02 yields a decrease of upper (lower) critical load in amount of 7% (20%) (compare Figures 2.4 and 2.5). Comparing the curves 2-5 with the curve 1, corresponding to the shell subjected to pure load action, one may conclude that the thermal field (2.178), (2.179) increases the top critical load in amount of 7%, whereas the temperature filed (2.180) causes an increase of the top critical load in amount of 16% in comparison to the action of one load. A development of stress strain state of ball-supported shell can be traced in Figures 2.30b, 2.31b, where the functions w

xx , F

xx along the axis y = 0.5 are displayed. The curves in these figures have the same number as the curves “load-deflection” in Figure 2.31. The centre deflection values, for which graphs are constructed, are reported in the figure and denoted by a corresponding index associated with a number. Comparing the curves 21 with 3, 22 with 32 , one may trace an influence of shell sides heat isolation on its thermo-stress state. Influence of curvature increases with respect to absolute value in the shell centre in amount of 60%, whereas in zone of fixed contour increases in amount of a few times. Comparing the curves 4i and 5i (i = 1-5) the conclusions follow: increasing the load amount from 0.01 to 0.02, a curvature value decreases in a zone of fixed contour. A reason is that a temperature torque decreases on the stability boundary. The function F

xx values also are decreased in a zone of fixed contour, since the temperature stress decreases on the space boundary. Consider now the shell k1 = k2 = 24, clamped along its contour (2.188), subjected to an action of a load and temperature field (2.180) shown in Figure 2.7. The curves “load-deflection” in the central shell point q(w(0.05; 0.05)) are displayed in Figure 2.32a. The curves 3, 4 correspond to a shell being within the temperature

2.2 0

45

90

4

1.1

1

3

3

2

1

4

Kx=Ky=18

2

Kx=Ky=24

3.3

4

1,3

W

4.4


135 q

260

Figure 2.29. Load-deflection curves differently excited (see text for more details).


261

-9

32,42 33 4

3

23

23

22

33

W

32,42

33,43

x

42 32

21

0.5

0

21

23

W``xx 0.8 F 33,43

32

42

0.4 31,41

W``xx 2

2

-45

F``xx

21 X 0.5

18

a)

F``xx

25

W``xx

31

0

0.5

23

-30

9

22

q=0.01 x

0

-15

30

23

0

15

0

0.5

41

31

x

0

31,41

21

22

22 31,41

-30

43

23

0 -3

22 21

q=0.01

32 33

F``xx

x

34

0

0.5 33

-6

23

-25

21

-9

21(W=0.5), 22(W=1.2), 23(W=2) b)

32 34

31

-50

31(W=0.5), 32(W=1.2), 33(W=2.0), 34(W=3.5)


xx , F, F xx for the shell k1 = k2 = 24 (y = 0.5) with different excitations (see text for more details).

262


180

q Kx=Ky=18 4

1

5

2,3

90

0

1.5

3.0

4.5

a) 60

50

W``xx 41

42 43

51

0.5

45 44

45

44

43

0

55

F``xx 55

54

52 5 3

0

52

42

41

q=0.01 53

F``xx 0

x

0

W``xx

q=0.01

51

-50

54

-10

51(W=0.5), 52(W=1.3), 53(W=2.6), 54(W=4.2), 55(W=4.5)

-60

-12 41(W=0.5), 42(W=1.3), 43(W=2.6), 44(W=4.2), 45(W=4.5)

b)

Figure 2.31. Load-deflection graphs (a) and the functions w

xx , F xx (b) for different shell excitation (see text for more details).

field (2.180). This field is characterized in the following way: heat isolation of upper = 0, whereas its lower surface is attacked by the temperature shell surface ∂T $ ∂z z=−0.5 % 2 T |z=0.5 = (x − 0.5) + (y − 0.5)2 0.25H. The curves 1, 2 correspond to the heat isolated shell from below, i.e. the temperature field satisfies the following boundary conditions $ % ∂T = 0, T |z=−0.5 = − (x − 0.5)2 + (y − 0.5)2 0.25H, ∂x z=0.5 ∂T = 0 x, y . (2.198) ∂x x=0;1

2.5 Computations of Plates and Shells in a Temperature Field 200

q

263

2 1

3

100

4

W 3.2

0 a) 200

q 1

2

100

3,4,5

W 0

b)

3.2

Figure 2.32. Load-deflection graphs for different excitations (see text for more details).

The source f is governed by the equation (2.177). Curves 1, 3 are obtained for q0 = 0.0; curve 4 - for q0 = 0.004, curve 2 - for q0 = 0.008. Comparing the curves 1 and 3 one may conclude that an action of temperature field (2.198) increases a value of upper critical shell load in comparison to temperature field (2.180). Decreasing the load amount from 0.01 to 0.008 an action of temperature field (2.198) increases

264


yielding the shell stability increase (compare the curves 1 and 2), and a value of the upper critical load increases in amount of 5.5%. Decreasing the load amount from 0.01 to 0.004, and action of temperature field (2.180) increases yielding a shell stability decrease (curve 4 lies below curve 3). It should be emphasized that the temperature field (2.198) is negative. Curve 2 (curve 1) corresponds to an amount of absolute temperature value T max = 125◦ (90◦ ). The temperature field (2.180) is positive. Hence, these two temperature fields act on the shell in opposite manner. 2.5.2 Stress-strain state and shells stability The stress with geometrical parameters k1 = k2 = 24 for different boundary conditions along their sides and being subjected to an action of uniformly distributed transversal load and temperature field are investigated. Notation with respect to fixation type and the corresponding curve number introduced in Section 2.5.1 is applied. In Figure 2.32b graphs “load-deflection” in shell centre (0.5; 0.5) for the boundary conditions denoted by digits 1-5, 9 for q0 = 0.01 are reported. The curves, corresponding to a shell under action of temperature fields (2.180) and (2.198) are shown in Figures 2.22b and 2.33a, correspondingly. Recall that two applied temperature fields have the same absolute magnitude but different sign. Temperature field (2.198) is negative. Comparing the curves in Figure 2.32b one may observe that increasing a lengths of ball-supported contour is a vicinity of contour points up to 1 2 l0 , the graphs “load-deflection” approaches curve 4, corresponding to the clamped contour. In Figure 2.32b the curves 3-5 coincide. A variation of the clamped contour length l2 from 12 l0 to 0 in the middle of a side increases shell stability (curves 1, 2 lie over curve 5) in comparison to clamped shell. Analysing the results reported in Figure 2.33a one may conclude that increasing a length of ball-supported contour in a vicinity of corner points up to l1 ≤ 14 l0 , the “load-deflection” graph is shifted into curve 5 corresponding to clamped contour (curves 4, 5 coincide). Decreasing a length of clamped part of the contour l2 from 34 l0 to 0 in the middle of a side, decreases shell stability in comparison to shell clamped along the whole contour (curves 1-3 lie below curve 5). A value of top critical load for the shell corresponding to fixation 1 is decreased on amount of 10% in comparison to shell 5, clamped along the whole contour. In other words, increasing a length of clamped contour part in the middle of temperature field (2.180) side decreases shell stability, whereas the field (2.198) increases shell stability. Graphs “load-deflection” in the shell k1 = k2 = 24 centre (0.5; 0.5) for boundary conditions 1, 2 under actions of purely forcing load (curves 1, 2), of temperature field (2.180) and load (q0 = 0.01) (curves 1 , 2

) are displayed in Figure 2.33b. Digits 1, 2 denote fixation way, whereas dashing corresponds to temperature field. Comparing 1, 1 , and 1

one may conclude that mostly stable shell under an action of only forcing load (fixation 1), the temperature field (2.180) decreases a value of top critical load in amount of 14%, whereas the field (2.198) - in amount of 19% in comparison with load action. For the fixation type, the top critical load value is the largest one for the field (2.198) action, i.e. the field (2.198) increases the value of top critical load in amount of 1.3%, whereas the field (2.180) decreases a value


265

of the upper critical load in amount of 23% in comparison to an action of only one load (compare 2, 2 and 2

).

q

4,5

3

1

2

9

9 1 2 3 4 5 W

0

0.8

1.6 a)

2.4

200 q

1

Kx=Ky=24

2

2``

3.2

1` 2`

1``

110

W 0

0.8

1.6 b)

2.4

Figure 2.33. Load-deflection curves for different shell boundary conditions.

3.2

266


Recall that computations are carried out without an influence material properties variations with respect to temperature. For a series of design materials, in particular for steel with small cole amount, or steel the Young modulus is changed in amount of less than 5% on the considered temperature intervals.

3 Dynamical Behaviour and Stability of Closed Cylindrical Shells with Continuous Thermal Load

A historical background of dynamical behaviour and stability of shells thermally loaded is given in section 3.1. Dynamical problems of thermoelastic thin thermosensitive cylindrical shells are studied in section 3.2. Namely, after a general introduction, the variational formulation of the coupled dynamical problem of thermoelasticity is addressed. Next, the hybrid-type variational equations of thin conical composite orthotropic thermosensitive shells are derived. The problem of solution existence is rigorously discussed, and then a classification of thermoelastic problems is given. Computational algorithms are illustrated and discussed in section 3.3. A solution to the biharmonic equation in relation to forcing function, as well as the reliability of the obtained results, are studied. The modified relaxation method is described in section 3.3.4. Section 3.4 is devoted to analysis of dynamical stability loss with uniform force excitation. Criteria of dynamical stability loss are first reviewed, and then many examples are studied in more detail. Dynamical stability loss and non-uniform thermal load is addressed in section 3.4. Thermal field computation influence of time, shell geometry and load, as well as combined static and thermal loads action is studied in some detail.

3.1 Introduction Owing to the development of modern technics and technology, an investigation of dynamical behaviour in different constructions with thermal and force excitations becomes very important. A full description of these problems can be given in a frame of is possible thanks to the dynamical theory of thermoelasticity [514, 515]. Nowadays the thermoelastic theory is well established and clearly separated from other trends. It includes the following behaviour: the heat transfer (stationary and unstationary) between a body and an external environment; thermoelastic stresses caused by the temperature gradient; the dynamical effect accompanying sudden unstationary heating processes, in particular, thermoelastic vibrations of thin walled constructions caused by a heat impact; thermomechanical effect caused by the interaction between the deformation fields and the temperature. Fundamental results of the thermoelasticity theory are obtained in the quasistatical conditions. The inertial therms are not included in the governing equations

268

3 Dynamical Behaviour and Stability of Closed Cylindrical Shells

and a linking term in the heat transfer equation as well as the material characteristics do not depend on the temperature. The first step of a solution to the quasistatic problem of thermoelasticity and the problem of temperature stresses is focused on the definition of the temperature field by the methods of the heat transfer theory. A systematical approach to these problems is given in the monographs of Lykov [451], Kozdoba [358], Carslow and Eger [143], Podstrigatch and Koliano [553], Belaev and Riadno [92], and others. In a general case, a heat transfer equation is a nonstationary and three dimensional one. In order to simplify a mathematical statement of this complex problem, the three dimensional heat transfer equation is reduced to the lower dimension equation using the physical and mathematical properties of the problem. In particular, in the case of thin bodies, different dependencies of the temperature versus a normal coordinate are used (as the series of normalized Legendre polynomial or the orthogonal functions), the asymptotic method, the method of development on the eigenfuctions, and so on [487, 549, 554, 726]. In some works the three dimensional heat transfer equation has been used directly [671]. To the first thermoelastic problems belong these of a heat impact on the surface of a halfspace investigated by Danilovskaya [161]. She has explained the singularities of the dynamical heat stresses propagations. The fundamentals of the modern plates and shells theory have been established in the works of Ambartsumian, Bolotin, Bubnov, Vlasov, Volmir, Vorovitch, Goldenveizer, Dinnik, Germain, Iliuˇsin, Kane, Karman, Kiltchevskij, Kirchhoff, Labenson, Lurie, Love, Muscheliˇsvili, Muˇstari, Galimov, Novoˇzilov, Obraztsov, Pankovitch, Pogrelov, Rabotnov, Rˇzamitsin, Timoshenko, Filonenko-Boroditch, Tchernych and others. The fundamental steps and directions in the development of the plates and shells theory are given in the works of Ainola and Nigul [9], Alfutov [11], Alumiae [12], Ambartsumian [14], Bolotin [120, 121], Burmistrov [136], Valisvili [675], Vol’mir [684, 685, 686], Vorovitch [689, 690], Galimov [218], Goldenblat [227], Goldenveizer [228], Grigoliuk, Kabanov, Srebovskij [238, 240], Guz’ and Babitch [256], Darevskij [165], Dzanelidze [190], Iliuˇsin and Pobiedria [270], Kantor [303], Karmiˇsin, Skurlatov, Startsev, Feldstein [310], Kaiuk [297], Kornisin [345], Kossovitch [351], Krysko [369], Morozov [481], Mushtari and Galimov [492], Novozilov [507, 508, 509, 510], Ogibalov [518], Ogibalov and Koltunov [520], Pelech [533], Satchenkov [601], Srubshtchik [633], Timoshenko [655, 656, 657], Filin [207], Vol’mir [684], Darevskij [165], Karmishin et al. [309], Satchenkov [600], Ramm [566], Wunderlich [710], Zienkiewicz [724], Thompson, Hunt [654], Hermann [258], Yamaki [711], Hinton, Owen, Zienkiewicz [260], Kleiber [328], Borkowski [125], König [352], Leissa [426, 427], Galdersmith [214], Powell [559], Sare, Massonet [596], Ashton, Whitney [36], Aalami, Williams [1], Voyiadjis, Karamanlidis [692], Lekhsritskii [428] and others.

3.1 Introduction

269

The researchers attention is focused on the problems of stability investigations and the definition of stress deformation states of plates and shells. A deep investigations of those processes is caused by the industrial needs. In the recent time, vibrations and stability investigations in a frame of the elastic theory are carried out by Andreev, Lebiediev and Obodan [28], Baˇzenov [75], ˇ Zigalko [728], Kolometz [334], Krysko [373, 379], Kossovitch [351], Makarenko [454], Pertzev and Platonov [534], Satchenkov [600], Krätzig, Onate [361], Crisfield [159], Wrigpers, Wagner [709], Gorman [233, 234], Singh, Dey [622], Sakata, Takahaski [590], Soedel [629], Bogdanovich [112], Desturgnder, Salann [177], Hjelmstadt [262], Gould [235], Laguese, Lions [407], Jawao [288], Reddy [569], Krissen, Skalond [368], Yi-Yuan Ya [713], Sawczuk, Sokol-Supel [603], Gilgert, Hackl [223], Voyiadjis, Karamandlidis [692], Reismann [570], Valid [674], Waszczyszyn et al. [701], Elishakoff et al. [195] and others. Dynamical stability of shells can be investigated using the following properties: – Type of a shell (cylindrical, spherical, conical); – Physical-mechanical material properties (elastic, elastic-plastic, plastic); – Load properties (axial load, external pressure, concentrated load, matching of different loads); – Time dependence of a load (impulse load, rectangular impulse load, cyclic load and others). It should be noted that we still have not got a definition of dynamical stability including all of the mentioned properties. The description called the ”dynamical stability” is related to many problems of different physical behaviour. According to Simitses [620] classification, we can formulate three classes of problems related to the dynamical stability (unstability). 1. Unstability, occurring as a result of periodic excitations causing a parametric resonance (for certain combinations of shell’s free vibrations’ frequencies and excitation frequencies). The problem of stability investigation is reduced to that of the Mathieu-Hill problem [114, 146]. 2. Unstability occurring due to the occurrence of nonconservative external forces (mainly aerodynamical), which is characterized by self-excited vibrations (flatter). The stability investigation is reduced to the explanation of self-excitations. The problem of post-critical design behaviour is reduced to the analysis of a limit cycle [73, 121, 184, 684]. 3. Unstability caused by a load being an arbitrary aperiodic function of time. In this case, the definition of dynamical stability loses its clear meaning. This situation occurs during a load-carrying ability in conditions of the explosion type load. Estimation of the load-carrying ability is rather complicated in this case because of a sudden change of equilibrium states. Very small perturbations of external conditions (load, boundary conditions, end so on) may lead to a qualitative change of the initial state. A proper approach, including all singularities of those processes, is possible only when the geometrical nonlinear shells theory is

270


applied. The first work devoted to the shells’ stability loss in the geometrically nonlinear frame of investigations has been carried out by Grigoliuk [239]. The criterion choice in the problems of dynamical stability loss has been discussed, among others, in the following references [132, 165, 310, 375, 387, 404, 454, 491, 612, 684]. The investigation problem of thin shells with the nonuniform load’s reactions has a crucial meaning as far as application is concerned. In a frame of the linear theory those problems have been considered in the works of Agafonov [3], Lyapunov, Roˇzikova [448], Yao [654], Keer et al. [320], Reisman, Pawlik [571], and others. Behaviour of closed cylindrical shells with a nonuniform external pressure, including nonlinearities, still needs further investigation. We have to mention the work of Kormiˇsin et al. [310], devoted to the experimental-theoretical investigation of interaction processes between thin walled constructions and strength impact waves in a gas, and also the work of Solonenko [631]. In the latter one, on the basis of the equations of nonlinear theory of shells, a stress-deformation state is analysed, when the external pressure is applied. This pressure acts in a part of the surface of different shapes and sizes. Among others, it has been shown that the largest dynamical effect is concentrated in the zone located along the generating line. The problems of cylindrical shells’ dynamical stability have been considered by Makarenko [454, 455]. In his works, the influence of nonuniformity and a load speed on the critical load values and the post-critical deformation states have been considered. The deflection and force functions have been presented in the longitudinal direction by half-sinusoids. In the circle direction the Bubnov-Galerkin method is applied. The equation related to the force function components has been solved due to the transition to the Cauchy problem and integration of the initial vector using the Newton’s method has been carried out. In order to describe physical nonlinearities, the theory of small elastic deformation has been used. In particular, a weak influence of the plastic material properties on the deformation properties and critical time have been shown. We have to mention a group of works oriented on the dynamical stability of cylindrical shells with a nonuniform load carried out by Kolomoietz and Krysko[334, 378, 379]. In these works, by means of Bubnov-Galerkin method in higher approximations many problems of cylindrical shells’ dynamical stability with the rectangular type loads have been solved. Among others, the results of the critical load’s dependence on the excitation time and on the angle of the load’s action have been reported. Among the experimental works devoted to the dynamical shell buckling, the following ones should be mentioned: Bivin and Naida [111], Tchuiko [653], Skurlatov [625], Bushtyrkov and Naida [139], Andrieev et al. [20, 22, 23], Karolev [313] and others. In the work of Karmishin et al. [310] an experimental investigation of shells buckling is carried out. For small values of the axial compressing force, in comparison to the statical critical force, the buckling is characterized by an area including the loaded part. A short-wave component is added to the fundamental half-wave.

3.1 Introduction

271

The experiments have shown that the unloaded part is practically undeformed. If the longitudinal force is close to the critical one, then the buckling form approximates that of the statical stability loss. In this case, the critical impulse value is small and it plays a role of the exciting factor acting on the potential threshold, separating the old and new equilibrium positions. The results of instability areas’ experimental investigations of smooth cylindrical shells with an impulse external load in combination with the external (or internal) transversal statical pressure are given in the work of Manevitch at al. [457]. It has been shown that the external statical pressure in the area of values higher than 0, 6q (q - the statical pressure of buckling) has essential influence on the stability loss. In this case, the stability loss is initiated by a sudden jump and then large deformations occur. The internal pressure has stabilizing influence of the shell and leads to a significant increase of the critical impulse. According to the authors’ conclusions a quasi linear dependence of the critical impulse on the internal pressure is observed. On the other hand, in the work of Baskakov et al.[86], where the experimental results of investigation of the influence of the static internal pressure and load velocity on smooth cylindrical shells with an impulse type external load stability are outlined, the following conclusions have been obtained: 1. The internal pressure essentially increases a magnitude of the critical load of shells’ stability loss. This dependence has a nonlinear character, and the dynamical overload coefficient increases to the moment when the internal pressure causes the shell’s material strength loss. 2. The shells with the internal pressure have got relatively small final deformations and not always lose their load-carrying abilities. The results of the experimental investigations of the influence of geometrical shells’ parameters and of the boundary conditions on the magnitude of dynamical critical load with an external impulse-type pressure are given in the work of Andrieev et al. [21]. The results of the research show a weak sensitivity of the shells with average length (L/R = 1, . . . , 4) on the boundary condition with that type of the load. The stability loss of the short shells (L/R ≤ 0.5) is characterized by an increase of the critical impulse value. With the increase of the R/h parameter the dynamical critical load is decreased. A dependence of the critical impulse on the model’s radius for the fixed R/h value is linear. There are only a few works devoted to the experimental investigation of the dynamical stability loss with an ununiform load. We mention only, where the problem of stability estimation of a cylindrical shell with the impulse-type pressure is investigated. An influence of the load zone on the dynamical critical load value is analysed beginning with an impulse type to quasistatic type loads. In real constructions, applying enough careful experiments we need to take into account the initial deformations of the analysed shells, the occurrence of the load eccentricity, additional vibrations and other deviations and excitations. The influence of those factors during the analytical and experimental investigations leads to an extremely complicated but valid problem [36, 131, 196, 326, 434, 435, 440, 441, 467].

272


Analytical solutions to the dynamical stability problems obtained on the basis of a simplified approach can be found in the works Antsiferov, Pavlenko [34], Karmiˇshin, Feldˇstein [310], Lindberg, Florence [433]. From the computational point of view, the buckling problem of rotational shells with an axiallysymetric load is effectively solved using many algorithms of one dimensional discretization [138]. High dimensional problem include many difficulties from the point of view of the computation time length [76]. The investigation of the thermoelastic processes in the thin-wolled constructions (plates, shells) has some singularities. A proper choice of the computational model leads to effective results. However, it is impossible to give general recommendations to this problem. It can only be mentioned that the model should be possibly simple, but including principal properties of a real system. According to the work of Pertsev and Paltonov [534] the qualitative analysis of the wave processes in a construction plays an important role in the choice of a proper model. This analysis sometimes allows for an a priori judgement, concerning the question of which processes (and at what time) play a significant role. It also makes it possible to establish an appropriate physical and mathematical model. This remark is valid for the heat transfer, the elastic as well as the thermoelastic problems. In the latter case, that analysis has a particular role because processes with different time scales are considered. The fundamental mathematical models of a shell deformation with a local dynamical heating are considered in the works of Medvedenko, Obodan [470], Andreev et al. [29]. In these works special time scales are introduced. A defined mathematical model of thermoelastic deformation corresponds to each of them. In particular, it has been shown that in the first time moments, after applying a heat flow, the shell can be treated as a three dimensional body, where a thermo-stress wave is propagated. Further, after heating along the thickness, a thin-walled property is observed, and a deformation has a static character. What should be noted here is the theoretical-experimental investigation carried out by Andrieev and Obodan [27]. Depending on the load parameters the resonance phenomena and the dynamic stability loss occurring in a time similar to the heating time along the thickness. In the case of a statical load of the thin-walled construction, the phenomenon of the statical stability loss caused by the heat impulse is investigated. In the work of McQuillen, Brull [469] it has been shown that in a frame of the halfly linked theory of thermoelasticity (in the equation of the heat transfer a mechanical linking term is omitted) dynamical effects are dominating when the shell’s wall becomes thin enough, i.e. the heating time has a value of the lowest vibration period order. In the work of Podstrigatch and Shvetz [555] the thin-walled shell conditions from the point of view of the heat transfer are given in the following form h 1, l

kh 1,

h2 1, aT a

where: a - coefficient of temperature conductivity; T a - time of an observable temperature increase; k - curvature of the shell.

3.1 Introduction

273

In the work of Bolotin [123], where the thermoelastic equations are established in the theory of plates and shells together with the Kirchhoff-Love hypothesis about a normal element, the analogical one is introduced in the heat transfer equation. It is assumed that the temperature along the thickness is linearly distributed. A wrong assumption of this hypothesis has been shown by Guliaev et al. [251] and Guliaev, Tchibiriakov [252]. The heat transfer equation (in those works), has been reduced to the second order equation using the Legandre polynomials. It has been shown that (as a result of the heat impulse) in the beginning a rapid change of temperature occurs. It causes internal stresses which lead to the stability loss and then a heat stabilization occurs. The problem has been considered as the geometrically linear one. In the work of Kovalenko [354] it has been also shown that when a rapid unstationary space temperature field is imposed, then a purely heat deformation along the thickness of a thin-walled shell or plate essentially differs from the linear one. Therefore, a hypothesis about the unchangeability of a normal element (in a general case) does not correspond to the hypothesis of the linear change of heat stresses along the thickness. However, the application of general purely heated deformation reduces the thermoelastic problem with a volume temperature field to the two-dimensional problem of the isothermic theory of plates and shells. The problem of vibration excitation of the thin-walled elements of constructions with a heat impulse load is considered in works of Boley [114], Boley, Barber [116], Kraus [365], Sinitsin [624], and others. In these references, the fundamental properties of the phenomena with a heat impact are described and illustrated. Among others, it has been noted, that the influence of the inertial terms during the investigation of the temperature stresses in the thin-walled elements of constructions leads to the solutions, where the vibrations appear. The intervals of parameters, where the unstationary behaviour of the construction occurs, are given. The singularities of the dynamical behaviour when a cylindrical anisotropical shell is heated are discussed by Shvetz and Flatchok [615]. The investigation of thermostability of plates and shells is described in the monograph by Ogibalov and Gribanov [519]. The correct formulations of the mechanical thermostability problems are given. Furthermore, the methods and their solutions and some examples of stability calculations of plates and shells in a frame of the physical nonlinear theory are discussed. All the above mentioned works have been carried out using the assumption, that the temperature increase is small and that all materials’ characteristics have been independent of the temperature. As it is known in many practical problems, those assumptions can not be made. A lack of those limitations does not change the assumption about small deformations but only leads to the occurrence of the variable coefficients in the thermoelasticity equations. In the work of Kovalenko [354] the thermoelasticity theory has been formulated without those limitations and in a frame of the small deformations including the dependence of the elastic and heat material properties on the temperature. Computations of statical problems of shells and plates with a linear temperature distribution along the thickness, including the temperature change of both the

274


elasticity modules and the linear heat expansion coefficient, have been carried out in the works of Burak, Ogirko [134], Petrov [535], Karpov and Filatov [314], Nerubailo and Ivanov [499], Kamiya and Fukui [299], and others. Attention has been paid to the investigation of the design behaviour influence on the statical load’s joint action and the temperature field (Butenko and Chalilov [140], Lebedev [422], Lykianenko and Makarenkov [449], Mishulin and Sinitsin [475] and others). In the work of Stroud and Mayers [639] the dynamical thermoelastic behaviour of the rectangular plate made from the temperature depending material with an arbitrary temperature action, is analysed. It has been shown that only a full dependence of the material versus the temperature leads to correct results. A review of material stresses of different materials with properties depending on the temperature can be found in the work of Noda [505]. Dynamic stability of thin-walled slowed shells with a temperature impact taking into account the temperature dependence of their fundamental characteristics has been analysed by Krysko and Fedorov [373, 375, 376]. In these works it has been assumed that the heat flow has been uniformly distributed along the shell. This assumption reduced the problem to the one dimensional heat transfer equation. In order to solve the thermoelastic equations the method of finite differences and the Runge-Kutta method have been used. The authors have concluded that the temperature dependencies of all material characteristic are needed for a proper investigation of dynamic instability effects with a heat impact. The experimental data of dynamical stability loss with a local external heat impact with an internal pressure are very rare. We mention the work of Kostoglotov et al. [352], where the experimental investigations of smooth alloy and with the internal pressure, have been carried out using a laser generator. The experimental investigations of the plates and shells instability with a local heat impact and different external loads have been described in references [23, 30, 31, 532]. All of the experiments have been carried out using the technique called “impulse” and described in reference [24]. A local impulse type heating has been reached using an optical quantum generator. The occurrence of the wave-forms and the buckling occurrence of the cylindrical shells have been analysed in relation to the power of the heating flow [30]. It has been concluded that the influence of the heat impulse initiates a buckled form localized in the area of the heat action. In the work [23] a circle plate with a local heat impact has been analysed. The temperature distribution after the heat impact and the wave occurrence along the plate radius have been analysed, too. In the work [378] the phenomenon of the stability loss of cylindrical shells with a local unstationary heating and with a uniform pressure or an axial compression is described and illustrated. As a result of the investigations, three types of the stability loss occurs: local (temperature), general (corresponding to the load type) and general with a time delay. The latter one is linked with the heat distribution and with the nonmonotonic dependencies of the critical loads on the load exponent variations of the unsymetric deformations. In the work [532] the influence of the heat impact on

3.1 Introduction

275

the cylindrical shells with a combination of different statical loads (an axial compression and an external pressure) has been investigated. Also, the most dangerous external loads combinations have been outlined which essentially decrease the carrying abilities of a load. Some of the works are devoted to the approximate methods to solutions with the heating load excitations [118]. The variational methods play an important role in solving these problems. The variational principle allows for a compact form formulation of the problem, because it includes the fundamental equations and the boundary conditions in a general form. It allows (sometimes) for a low and high limits estimation of the variational integral. It describes the quantity being more suitable in direct practical applications. It should be noted, that not all differential treatment of the problem has its variational formulation in the classical meaning, i.e. in the sense of an integral which should approach a maximum or a minimum [474]. Gurtin [253] has developed a variational principle where the initial conditions are included in the functional. A general variational approach applied for an orbitrary nonlinear system has been developed by Tonti [659]. The problems of the variational equations have been considered in the references of Balabuch and Shapovalov [82], Bugrij [133], Gribanov and Panitchkin [237], Grigorienko et al. [246], Flatchok [209], and others. Among the works devoted to the qualitative investigations of the operator equations (the theorems of existence, the estimation of the solutions’ number, the existence of nontrivial solutions, the applications of different schemes of the approximated solutions, the analysis of the equations with the parameters, and so on), we mention the research of Vainberg [672, 673], Vishik [679, 680], Dubinskij [186], Krasnoselskij [362], Krasnoselskij and Zabreiko [363], Koshelov [349], Ladyzenskaya [405, 406], Michlin [474], Nikolskij [503], Sobolev [628]. The qualitative investigations of nonlinear behaviour of plates and shells and their statical properties have been carried out by many authors. In the monograph by Vorovitch [689] a wide spectrum of possible approaches to many different problems of this type is given. Essentially smaller amount of works is devoted to the problems of dynamical behaviour of plates and shells. Among others, the works of Vorovitch [690], Morozov [481] and Lions [438] are remarkable. The geometrically nonlinear linked problems of slope shells’ thermoelasticity have been considered by Kiritchenko and Krysko [324]. The up-to-date dynamic stability problems of cylindrical shells with a nonuniform thermal load lead to the following conclusions and observations. 1. The problem of geometrical parameters of cylindrical shells’ influence, their boundary conditions, the load parameters causing dynamical instability with an ununiform external pressure in a frame of the geometrical nonlinear theory is not sufficiently investigated.

276


2. Further research presents the problem of dynamic stability loss of a thin-walled cylindrical shell in the condition of a combined nonuniform thermal load including a priori given static load. 3. An adequate choice and a theoretical analysis of the calculation model during a numerical solution to the problem of dynamical stability loss of a geometrical nonlinear and a thermal sensitive cylindrical shell in the condition of an ununiform thermal load still need further investigations and clarifications.

3.2 Thermoelastic Thin Thermosensitive Cylindrical Shells 3.2.1 General Introduction Let us consider an elastic body as a continuum, with temperature T , entropy s (thermodynamics parameters), stresses σi j and deformations εi j . The entropy and the temperature, the corresponding parameters of stresses and deformations are the selfcoupled state parameters [555]. In order to establish the relations between the state parameters we need to formulate an expression for free energy ψ as a function of the deformation components εi j and the temperature T . We assume that the body in the initial undeformed state has the temperature T 0 = const. We assume that the temperature increase (T − T 0 ) yields a purely ther4T mal expansion T α∗ dT (α∗ - true coefficient of the linear thermal expression) with 0 a magnitude of one order smaller than εi j . This assumption is not in opposition to the fundamental hypotheses of the linear elasticity theory of deformations. It allows to omit a limitation governed by the inequality (T − T 0 )/T 0 1. However, mechanical and thermal nonuniformities caused by high temperature influence should be accounted. Therefore, developing the function of free energy into the Taylor series, second εi j powers remain. The function ψ and the state equations have the following form [354] λ ψ = ε2kk + µεi j εi j − (3λ + 2µ) αT (T − T 0 ) εkk − 2

T

T dT

T0

Cε=0 dT , T

(3.1)

T0

1 ∂λ 2 ∂ψ ∂µ =− ε − εi j εi j + ∂T 2 ∂T kk ∂T T 6 Cε=0 ∂ 5 (3λ + 2µ) αT (T − T 0 ) εkk + dT, ∂T T s=−

(3.2)

T0

σi j =

5 6 ∂ψ = 2µεi j + λεkk − (3λ + 2µ) αT (T − T 0 ) δi j . ∂εi j

(3.3)

In the above, αT denotes the average coefficient of a linear thermal expansion in the temperature interval (T 0 , T ), defined by the relation

3.2 Thermoelastic Thin Thermosensitive Cylindrical Shells

277

T

1 αT = T − T0

α∗ dT ,

(3.4)

T0

where Cε=0 is the thermal capacity of an undeformed state, λ, µ are the Lamé coefficients for the isothermical deformation. They are linked with the isothermical elasticity modulus E and the Poisson’s coefficient ν by the relations λ=

νE , (1 + ν) (1 − 2ν)

µ=

E . 2 (1 + ν)

(3.5)

−u = (u , u , u ) and deformations have the The relations between displacements → 1 2 3 following form [676] 1

(3.6) εi j = ui, j + u j,i + um,i um, j . 2 Below, a general thermoelastic problem is formulated. We need to determine the stress components σi j , the deformations εi j , the dis−u and the temperature T (when the mechanical and thermal excitations placements → are given) satisfying the following governing equations 5&

' 6 d2 ui δim + ui,m σκm ,κ + P¯ i = ρ0 2 , dt

(3.7)

where P¯ i is the component of the external load, ρ0 is the material’s density, and κ is the physical-geometrical parameter. The relations between stresses and deformations are given by (3.3), and between deformations and displacements are defined via the equation (3.6). The heat transfer equation reads T

' ds & = λT T ,i , j + r, dt

(3.8)

(λT is the heat transfer coefficient; r - the power of the heat sources) for the given initial and boundary conditions. Using the relations (3.2), (3.3), the equations (3.7), (3.8) are presented in the following form ⎧ ⎤⎫ ⎡ ⎤ ⎡ ⎪ T ⎥⎥⎥⎪ ⎢⎢⎢ ⎥⎥⎥ ⎢⎢⎢ ⎪ ⎪ ⎪ ⎪ ⎪ '⎢⎢⎢ d2 ui ⎢⎢ ⎥⎥ ⎨& ⎬ ¯ ⎥⎥⎥⎥⎪ ⎢ ⎥ (3λ + u + − + 2µ) α dT + Pi = ρ0 2 , (3.9) λε δ δ 2µε ⎢ ⎥ ⎢ ⎪ ⎪ im i,m ⎢ κm ⎢ kk ∗ κm ⎥ ⎥ ⎪ ⎪ ⎥ ⎢ ⎥ ⎢ ⎪ dt ⎦⎪ ⎣ ⎦ ⎣ ⎪ ⎪ ⎩ ⎭ T0

⎧ ⎪ ⎪ ⎪ ⎪ ∂2 ⎨ 1 ∂ 2 λ 2 ∂2 µ ε − ε ε + T⎪ − i j i j kk ⎪ 2 ⎪ ∂T 2 ∂T 2 ⎪ ⎩ 2 ∂T

,κ

⎫ ⎡ ⎤ ⎪ T ⎢⎢⎢ ⎥⎥⎥ ⎪ ⎪ Cε=0 ⎪ ⎢⎢⎢ ⎥⎥⎥ ⎬ dT − ⎢⎢⎢(3λ + 2µ) α∗ dT ⎥⎥⎥εkk + ⎪ ⎪ ⎪ T ⎣ ⎦ ⎪ ⎭ dt T0

⎧ ⎤ ⎫ ⎡ ⎪ T ⎥⎥⎥ ⎪ ⎢⎢⎢ ⎪ ⎪ ⎪ ⎪ ⎪ ∂µ ∂λ ∂µ ∂λ ⎥⎥⎥ ⎪ ⎢ ⎨ ⎬ dεi j ⎢ ⎢⎢⎢ εkk − 3 (3λ ε + 2 = T⎪ + α dT − + 2µ) α 2 δ ⎥ ⎪ i j ∗ ∗ i j ⎥ ⎪ ⎥ ⎢ ⎪ ⎪ ∂T ∂T ⎦ ⎪ ⎣ ∂T ⎪ ⎪ ⎩ ∂T ⎭ dt T0

278


&

λT T ,i

'

,j

+ r.

(3.10)

The initial conditions are defined by a distribution of components of the displacei ment vector ui , their velocities du dt and the temperature T in the whole volume V of the elastic body of the form

− x , (3.11) ui |t=0 = u0i → dui du0i → −x , |t=0 = dt dt

− x . T| =T → t=0

0

(3.12) (3.13)

The boundary conditions on the surface Ω of the elastic body are composed of mechanical and thermal conditions. The mechanical boundary conditions are given either for the displacements

− x,t , (3.14) u = uˆ → i

or for the stresses

i

− î → x,t . σi j n j = σ

(3.15)

The thermal boundary conditions are expressed by the ones given below: – temperature distribution along the body’s surface

−

− T → x , t = Tˆ → x,t ,

(3.16)

– intensity of the heat flow qT through the body’s surface

− ∂T , (3.17) x , t = −λT qT → ∂n – a surrounding medium temperature and a rule of convection between the body surface and the medium ∂T = α (T − υ) , (3.18) −λ ∂n where α is the thermal coefficient. 3.2.2 Variational Formulation Many works have been devoted to the variational formulation of the coupled dynamical problem of thermoelasticity. Among others, we have to mention the references [82, 133, 209, 555]. A review of different formulations of the thermoelasticity problems is described by Shatchnev in the book of Nowacki [515]. We have to mention the works of Biot [108], where only a quasistatical case is considered, and the work of Nickell and Sackman [501], where an approach, introduced by Gurtin [253], is applied. A case, in which a temperature increase is not small, has been considered by Lebon, Lambermont [423]. The method of limited variations and a series of simplified hypotheses have been used. In the work of Belli, Morosi [94] a critical review of the exciting approaches to the variational formulations in the case of


279

linked thermoelastic problem is given. Additionally, the variational formulation of that problem for small thermal excitations using a standard method (see Vainberg [673]) is outlined. Owing to that approach we consider a possibility of variational formulation of the linked dynamical thermoelastic problem when the thermal excitations are not small. In a general case, a variational formulation to the problem N(u) = f,

u ∈ D(N),

(3.19)

where N(u) is the nonlinear operator, defined on D(N). It depends on the existing character of the bilinear form in relation to which operator N(u) should satisfy the following condition [661] 8 7 8 & ' 7

(3.20) Nu ϕ, χ = Nu χ, ϕ , ∀ϕ, χ ∈ D Nu , where Nu is the linear part of the Gateaux derivative of the Nu operator, defined by the expression N (u + εϕ) − N (u) . (3.21) Nu ϕ = lim ε→0 ε The condition (3.20) defines the Nu symmetry in relation to the chosen bilinear form. As it has been shown in the works of Gurtin [253] and Tonti [659] the existence of the variational formulation is guaranteed when a bilinear form is taken as the following operator convolution tK u (tK − t) υ (t) dt = υ (t) ∗ u (t) ,

u (t) ∗ υ (t) =

(3.22)

0

dγ du ∗ υ (t) = u ∗ + u (tK ) υ (0) − u (0) γ (tK ) , (3.23) dt dt where tK denotes the end of the considered time interval. For the considered initialboundary thermoelasticity problem (3.9), (3.10), (3.11)–(3.18) we take the bilinear form in the following way u, υ =

−

− u→ x,t ∗ υ → x , t dV.

(3.24)

V

The Gateaux derivative can be presented in the following symbolic form Au BT

, (3.25) NuT = Cu DT where the operators Au , BT , Cu , DT satisfy the homogeneous initial and boundary conditions. The symmetry condition (3.21) is equivalent to the satisfaction of the following three conditions (3.26) Au χ, ϕ = Au ϕ, χ ,

280


BT χ, ϕ = Cu ϕ, χ , DT χ, ϕ = DT ϕ, χ ,

∀ϕ, χ ∈ D

&

(3.27)

' NuT .

(3.28)

The operators Au , BT , Cu and DT have the following form:

− → 9& ' Au → u , T −u 1 = δim + ui,m (2µε1κm + λε1kk δκm ) + ⎛ ⎞ ⎡ ⎤ T ⎜⎜⎜ ⎟⎟⎟ ⎢⎢⎢ ⎥⎥⎥ ⎜ ⎟ ⎢ ⎥ u1i,m ⎜⎜⎜⎜⎜2µεκm + ⎢⎢⎢⎢⎢λεkk − (3λ + 2µ) α∗ dT ⎥⎥⎥⎥⎥ δκm ⎟⎟⎟⎟⎟ + ⎝ ⎠ ⎣ ⎦ T0

,κ

$

− % d u1i d2 u1i ρ0 2 = A1 → u , T u1,i + ρ0 2 , ,j dt dt

→ & ' ∂µ − BT u , T T 1 = δim + ui,m 2 εκm T 1 + ∂T 2

(3.29)

⎡ T ⎢⎢⎢ ∂λ ∂α∗ ⎢⎢⎢ ∂λ dT · T 1 − 3 + ⎢⎢⎢ εkk T 1 − (3λ + 2µ) ∂T ∂T ⎣ ∂T T0

⎤ ⎫ T ⎥⎥⎥ ⎪ ⎪

− ⎪ ∂µ ⎥⎥⎥ ⎪ ⎬ T1 2 α∗ dT ⎥⎥⎥ δκm ⎪ = B1 → u , T T1, ⎪ ∂T ⎦ ⎪ ⎪ ⎭ T0

(3.30)

,κ

− → ∂ λ ∂2 µ Cu → u , T −u 1 = T − 2 εkk ε1kk − 2 2 εi j ε1i j + ∂T ∂T ⎫ ⎡ ⎤ ⎪ T ⎢ ⎥⎥⎥ ⎪ ⎪ dεi j ∂µ ∂λ ∂2 ⎢⎢⎢⎢⎢ ⎬ dT ⎥⎥⎥⎥ ε ⎪ (3λ − T 2 ε ε − α dT + δ + 2µ) ⎢ ∗ 1i j 1kk i j ⎥⎥⎦ 1kk ⎪ ⎪ ⎪ dt ∂T ∂T dt ∂T 2 ⎢⎢⎣ ⎪ ⎭ 2

T0

T ∂µ ∂µ ∂λ ∂λ εkk − 3 + 2 T 2 εi j + α∗ dT − ∂T ∂T ∂T ∂T

T0

− →

− du1,i dε1i j 6 (3λ + 2µ) α∗ δi j = C1 → , u , T −u 1,i + C2 → u,T dt dt

→ ∂3 µ 1 ∂3 λ 2 − DT u , T = T − ε T − εi j εi j T 1 + 1 kk 2 ∂T 3 ∂T 3 ⎤ ⎡ T ⎥ ⎢ ∂2 ⎢⎢⎢⎢⎢ ∂α∗ ⎥⎥⎥⎥⎥ (3λ dT + 2µ) ⎥⎥⎥ εkk T 1 + ⎢ ∂T ∂T 2 ⎢⎢⎣ ⎦ T0

(3.31)


281

⎡ ⎤ T ⎥⎥⎥ ⎢ 2 ⎢ ⎢ ∂µ ∂Cε=0 T 1 dT ∂ ⎢⎢⎢ ∂λ ⎥⎥⎥ + 2 + α dT T + 3 ε ⎢ ⎥ ∗ ⎥⎥⎦ kk 1 ∂T ∂T T dt ∂T 2 ⎢⎢⎣ ∂T T0

⎧ ⎫ ⎡ ⎤ ⎪ ⎪ T ⎢⎢⎢ 2 ⎥⎥⎥ ⎪ ⎪ ⎪ 2 2 ⎪ ⎪ ∂ µ Cε=0 ⎪ ⎢⎢⎢ ∂ ⎥⎥⎥ ⎨ 1∂λ ⎬ dT 1 T⎪ + ε − ε ε + ⎢⎢⎢ 2 (3λ + 2µ) α∗ dT ⎥⎥⎥ εkk + − ⎪ ⎪ ⎪ 2 kk 2 ij ij ⎪ ⎪ 2 T ∂T ∂T ∂T ⎣ ⎦ ⎪ ⎪ ⎩ ⎭ dt T0

⎧ ⎫ ⎡ ⎤ ⎪ ⎪ T ⎢⎢⎢ 2 ⎥⎥⎥ ⎪ ⎪ ⎪ ⎪ ⎪ ∂2 µ ⎢⎢⎢ ∂ ⎥⎥⎥ Cε=0 ⎪ ⎨ 1 ∂2 λ ⎬ dT (3λ T1 ⎪ − ε − ε ε + α dT + 2µ) − + ⎢ ⎥ ⎪ kk i j i j ∗ ⎢ ⎥ ⎪ 2 2 2 ⎢ ⎥ ⎪ ⎪ T ⎪ ∂T ⎣ ∂T ⎦ ⎪ ⎪ ⎩ 2 ∂T ⎭ dt T0

T 2 ∂2 µ ∂µ ∂ λ ∂λ ∂α∗ T 2 2 εi j T 1 + +2 dT + εkk T 1 − 3 2 ∂T ∂T ∂T ∂T ∂T T0

3

∂2 λ ∂2 µ +2 2 2 ∂T ∂T

α∗ dT + (3λ + 2µ)

∂α∗ + ∂T

T0

dεi j ∂µ ∂µ ∂λ ∂λ +2 α∗ T 1 δ i j − T 1 2 εi j + εkk − 3 ∂T ∂T dt ∂T ∂T ⎤ ⎫ T ⎥⎥⎥ ⎪ ⎪ ⎪ ' ∂µ ∂λ ⎥ ⎪ ⎬ dεi j & +2 − λT T 1,i , j − 3 α∗ dT − (3λ + 2µ) α∗ ⎥⎥⎥⎥⎥ δi j ⎪ ⎪ ∂T ∂T dt ⎦ ⎪ ⎪ ⎭

T

T0

→

→ dT 1 & ' ∂λT ∂λT − − T 1 T ,i = D1 u , T T 1 + D2 u , T − λT T 1,i , j − T 1 T,i . ∂T dt ∂T ,j ,j

(3.32)

In the above the following notations is used ε1i j = u1 j,i + u1i, j + u1m,i um, j + u1m, j um,i . We prove that for the operator Au the condition (3.25) is satisfied Au ϕ, χ =

&

A1u ϕ,i

V

'

, j ∗ χdV +

ρ0 V

A1u ϕ,i ∗ χdΩ −

Ω

d2 ϕ ∗ χdV = dt2 ρ0

A1u ϕ,i ∗ χ, j dV + V

V

dϕ dχ ∗ + dt dt

dϕ (tK ) dϕ (0) · χ (0) − ρ0 · χ (tK ) dV = dt dt dϕ dχ ∗ dV = − A1u χ, j ∗ ϕdΩ+ − A1u χ, j ∗ ϕ,i dV + ρ0 dt dt ρ0

V

V

Ω

(3.33)

282


A1u χ, j

,i

∗ ϕdV +

V

ρ0

d2 χ ∗ ϕdV+ dt2

V

dχ dχ (tK ) − ϕ (0) · ρ0 dV = ϕ (tK ) · ρ0 dt dt V

&

A1u χ,i

'

,j

∗ ϕdV +

ρ0

V

d2 χ ∗ ϕdV = Au χ, ϕ. dt2

(3.34)

V

The integrals acting on the surface Ω, and those including the functions in the initial and end time moments are equal to zero in accordance with the Gateaux derivative definition. We check the (3.27) condition (i.e. the symmetry condition) of the operator DT χ, ϕ =

D1 χ ∗ ϕdV +

V

∂λT χT ,i ∂T

,j

D2 V

dϕ ∗ χdV − dt

dϕ ∗ χdV − dt

V

λ T ϕ, j

V

D2

,i

Ω

λ T ϕ, j

,i

,j

∗ ϕdV−

χ∗

λT χ,i ∗ ϕdΩ− Ω

∂λT χT ,i ∗ ϕ, j dV = ∂T

V

dD2 ϕdV − dt

dϕ ∗ χdV − dt

V

V

D2

'

D2 χ (tK ) ϕ (0) −

λT Y, j ∗ χ,i dΩ+ Ω

∂λT ϕT ,i ∂T

V

D1 ϕ ∗ χdV + V

∂λT χT ,i ∗ ϕdΩ − ∂T

∗ χdV +

5

dD2 ϕdV+ dt

V

χ

∂λT χT ,i ∗ ϕdΩ+ ∂T

Ω

D1 ϕ ∗ χdV + V

λT χ,i

V

V

λT χ,i ∗ ϕ, j dV −

D1 ϕ ∗ χdV +

V

D2 χ (0) ϕ (tK )] dV +

&

V

∗ ϕdV =

V

V

V

dχ ∗ ϕdV + D2 dt

∗ χdV − V

χ∗ V

∂λT ϕT ,i ∂T

,j

∗ χdV−

dD2 ϕdV+ dt

,j

∗ χdV.

2 This expression implies that symmetry is achieved only if dD dt = 0 and operator D2 has the form

→ ∂2 µ 1 ∂2 λ − ε − εi j εi j + D2 u , T = T − KK 2 ∂T 2 ∂T 2

(3.35) ∂λT ∂T

= 0. The


⎫ ⎡ ⎤ ⎪ T ⎢⎢⎢ 2 ⎥⎥⎥ ⎪ ⎪ Cε=0 ⎪ ⎢⎢⎢ ∂ ⎥⎥⎥ ⎬ (3λ α dT + + 2µ) ε . ⎢⎢⎢ 2 ⎥⎥⎥ KK ⎪ ∗ ⎪ T ⎪ ⎣ ∂T ⎦ ⎪ ⎭ T

283

(3.36)

0

It means that D2 is time independent in the case of unstationary temperature field when the simultaneous conditions given below are satisfied for an arbitrary material: ∂µ ∂α∗ ∂λ = 0, = 0, = 0, Cε=0 = const. (3.37) T ∂T ∂T Therefore, the symmetry of the DT operator is achieved only in the case, when the fundamental material characteristics do not depend on the temperature (and, consequently, on time). It means that λ, µ, αK , Cε=0 , λT should be constants. We check the condition (3.26) for the mixed terms of the Gateaux derivative to the operator of the linked dynamical thermoelastic problem with constant coefficients λ, . . . , λT . As it has been shown in the reference [615], the given symmetry condition is satisfied, when as the mechanical variable we take a distribution of the velocity vector, and not of the displacement vector. In the conditions of nonlinear dependence between the displacement and the deformations (3.6) this choice can dε i not be realized, because during the search for dti j in the expression both ui and du dt appear. Thus, it has been proved that a variational formulation of the operator related to the coupled dynamical thermoelastic problem is only possible, when the fundamental material characteristics λ, . . . and λT do not depend on the temperature (on time) and therefore the relation between the displacements and the deformations is linear. The fact that the variational formulation of that problem is only possible in the case of a linear thermoelasticity and small thermal excitations has a physical meaning. It should be noted that with a change of the differential operator caused by introduction of the integral multiplier [659], the variational formulation of the linked dynamical thermoelastic problem for the case in which the thermic excitation is not small, the characteristics λ, . . . , λT depend on the temperature (time) and the link between the deformations is nonlinear. However, this variational formulation leads to complex integro-differential equations. They are not applicable to the works, where the problem is formulated using only the differential equations. 3.2.3 Hybrid-Type Variational Equations Consider a composite thin shell made from an arbitrary number of orthotropic or/and design-orthotropic composites (see Fig. 3.1). The coordinate surface z = 0 can be applied to an arbitrary s-th composite. It can overlap with one of the composite surface contacts, the shell’s boundary surfaces or it can be located inside of the composite. The coordinate surface z = 0 is related to the curvilinear coordinate system α, β. The coordinate lines α and β overlap with the main curvatures of the given coordinate system. The external normal is oriented to the centre of the shell’s curvature. In order to describe the shell’s thermoelastic dynamical behaviour we introduce the following assumptions and hypotheses:

284


Figure 3.1. A thin shell composed of arbitrary number of orthotropic composites.

1. The hypothesis about the undeformated normals, given for the whole shell’s pack globally [14]. 2. The fundamental assumptions of the nonlinear technical theory of conical shells [228]. 3. The Duhamel-Neuman hypothesis for the orthotropic material [513]. It has also been assumed that the fundamental material characteristics depend on the temperature. In this case, from the generalized Hook’s law, we can obtain the relations between the components of the stressed and the deformable state for an orthotropic and a design-orthotropic composite


T σα = B11 ε1 + B12 ε2 + z (B11 κ1 + B12 κ2 ) − B11

T α1 dT − B12

T0

T0

T

T

σβ = B21 ε1 + B22 ε2 + z (B21 κ1 + B22 κ2 ) − B21

α1 dT − B22 T0

B11 = B12 = B21 =

E1 (T ) , 1 − ν1 ν2

B22 =

α2 dT ,

(3.38)

α2 dT ,

(3.39)

T0

ταβ = B66 εαβ + zB66 κ12 , where:

285

(3.40)

E2 (T ) , 1 − ν1 ν2

ν2 E1 (T ) ν1 E2 (T ) = , 1 − ν1 ν2 1 − ν1 ν2

B66 = G12 .

The curvatures κ(∗) are defined via relations (3.42). E1 , E2 denote the elasticity modulus; G12 is the shear modulus; ν1 , ν2 are the Poisson’s coefficients; α1 , α2 are the coefficients of the linear thermal excitation of the orthotropic composite (the shell’s layer). For the design-orthotropic shell’s layer νi = νγi , where γi is the coefficient of filling of the layer [290]. The relations between the deformations and the displacements of the coordinate surface and the expressions for the curvatures have the following form ε1 =

2 ∂u 1 ∂w − k1 w + , ∂α 2 ∂α

2 ∂υ 1 ∂w − k2 w + ε2 = , ∂β 2 ∂β ε12 = κ1 =

∂2 w , ∂α2

∂u ∂υ ∂w ∂w + + · , ∂β ∂α ∂α ∂β κ2 =

∂2 w , ∂β2

κ12 = 2

(3.41) ∂2 w . ∂αδβ

(3.42)

The u, υ, w denote the longitudinal, circle and transversal displacements of the coordinate surface z = 0, correspondingly. The κ1 , κ2 denote the main curvatures of the coordinate surface. Integrating (3.38)–(3.40) along the shell’s thickness, we get the expressions for the internal forces T 1 , T 2 , s12 . Multiplying (3.38)–(3.40) by z and integrating along the shell’s thickness we obtain the moments M1 , M2 , H12 . The final expressions have the form T 1 = C11 ε1 + C12 ε2 + K11 κ1 + K12 κ2 − C1T , T 2 = C21 ε1 + C22 ε2 + K21 κ1 + K22 κ2 − C2T , s12 = C66 ε12 + K66 κ12 = s21 ,

286


M1 = D11 κ1 + D12 κ2 + K11 ε1 + K12 ε2 − K1T ,

(3.43)

M2 = D21 κ1 + D22 κ2 + K21 ε1 + K22 ε2 − K2T , H12 = D66 K12 + K66 ε12 = H21 , where Ci j =

δ −∆ m+n s s=1 δ

s=1 δ

Di j =

CiT =

δ −∆ m+n s s=1 δ

s−1 −∆

Bisj zdz,

s−1 −∆

δ −∆ m+n s s=1 δ

Bisj dz,

s−1 −∆

δ −∆ m+n s

Ki j =

(3.44)

Bisj z2 dz,

s−1 −∆

⎛ s ⎞ ⎜⎜⎜T ⎟⎟⎟ ⎜⎜ ⎟⎟ s ⎜ s ⎜ Bii ⎜⎜ βi (T ) dT ⎟⎟⎟⎟ dz, ⎜⎝ ⎟⎠ T 0s

⎛ s ⎞ δ −∆ ⎜⎜⎜T ⎟⎟⎟ m+n s ⎜ ⎟⎟ ⎜ Biis z ⎜⎜⎜⎜ βis (T ) dT ⎟⎟⎟⎟ dz, KiT = ⎟⎠ ⎝⎜ s=1 δ

s−1 −∆

β1s = α1s + ν2s α2s , i = 1, 2,

T 0s

β2s = α1s ν1s + α2s ,

j = 1, 2,

i = j = 6.

During considerations of the conical shell’s equations in the hybrid form the following functions are introduced T1 =

∂2 F , ∂β2

T2 =

∂2 F , ∂α2

s12 = s21 = s = −

∂2 F . ∂α∂β

(3.45)

In this case the relations between deformations and stresses have the following form 2 2 ∂ F ∂ F ∂2 w ∂2 w + A + d + C + C + d , ε1 = A11 1T 12 2T 11 12 ∂β2 ∂α2 ∂α2 ∂β2 2 2 ∂ F ∂ F ∂2 w ∂2 w + A + d + C + C + d , ε2 = A12 1T 22 2T 21 22 ∂β2 ∂α2 ∂α2 ∂β2 ε12 = A66

∂2 F ∂2 w + 2d66 , ∂α∂β ∂α∂β

where A11 = C22 /Ω,

A12 = −C12 /Ω,

A22 = C11 /Ω,

(3.46)

3.2 Thermoelastic Thin Thermosensitive Cylindrical Shells 2 Ω = C11C22 − C12 ,

A66 =

1 , C66

d66 =

287

K66 , C66

C22 K11 − C12 K12 C22 K12 − C12 K22 , d12 = , Ω Ω C11 K12 − C12 K11 C11 K22 − C12 K12 , d22 = . d21 = Ω Ω The corresponding relations between the moments and the forces have the form 2 2 ∂ F ∂ F + C1T + d21 + C2T + M1 = d11 ∂β2 ∂α2 d11 =

&

' ∂2 w & ∗ ' ∂2 w + D12 − D12 − K1T , 2 ∂α ∂β2 2 2 ∂ F ∂ F M2 = d12 + d + C + C 1T 22 2T + ∂β2 ∂α2 &

where:

D∗11 − D11

(3.47)

' ∂2 w & ∗ ' ∂2 w + D − D − K2T , 22 22 ∂α2 ∂β2

∂2 w ∂2 F + 2 D∗66 − D66 , H = −d66 ∂α∂β ∂α∂β

D∗12 − D12

D∗11 = K11 d11 + K12 d21 ,

D∗22 = K12 d12 + K22 d22 ,

D∗12 = K11 d12 + K12 d22 = K12 d11 + K22 d21 ,

D∗66 = K6 d66 .

For the variational formulation we use the principle of virtual displacements [676] tK [δK − δV + δA] dt = 0, (3.48) 0

where: V - energy of an elastic body’s deformation; K - kinetic energy of an elastic body; A - work of external forces. It should be noted that this variational method is also valid for the problems of the initial deformations and the thermal stresses [676]. The expression for the deformation energy of the s-th shell’s layer has the following form ⎞ ⎡ ⎛ ⎟⎟⎟ T s ⎢⎢⎢ ⎜⎜⎜ 1 ⎟⎟ ⎢⎢⎢⎢ s ⎜⎜⎜⎜ s s V = ⎢⎢⎢σα ⎜⎜⎜ε1 + zκ1 − α1 dT ⎟⎟⎟⎟⎟ + 2 ⎠ ⎣ ⎝ s V

T0

⎞ ⎤ ⎛ ⎟⎟⎟ ⎥⎥⎥ ⎜⎜⎜ T s ⎟⎟⎟ ⎥⎥ ⎜⎜ s⎜ s s ⎟ ⎜ σβ ⎜⎜ε1 + zκ2 − α2 dT ⎟⎟ + ταβ (ε12 + zκ12 )⎥⎥⎥⎥ dV . ⎟⎠ ⎥⎦ ⎜⎝ T 0s

(3.49)

288


Integrating along the thickness and summing all the shell’s layers, after some transformations, the expression for energy of the shell’s deformation has the following form 1 1 Fββ uα − k1 w + w2α + Fαα υβ − k2 w + w2β − V= 2 2 Ω

1$ 2 + Fαβ uβ + υα + wα wβ − A11 Fββ + 2A12 Fββ Fαα + A22 Fαα 2 & & & ' ' ' 2 − D11 − D∗11 w2αα − 2 D12 − D∗12 wαα wββ − D22 − D∗22 w2ββ − A66 Fαβ

% $ 4 D66 − D∗66 w2αβ − d11 Fββ wαα + d21 Fαα wαα + d12 Fββ wββ + % d22 Fαα wββ − 2d66 wαβ Fαβ − [(d11C1T + C2T d21 − K1T ) wαα + (d12C1T + d22C2T − K2T ) wββ + (A11C1T + A12C2T ) Fββ + (A12C1T + A22C2T ) Fαα ]} dαdβ.

(3.50)

The relations (3.41)–(3.47) are used in order to get the expression (3.50). The kinetic energy of a multilayer orthotropic shell without a rotational inertia and without inertial effect in the tangentional direction is written as 1 K= 2 where m∗ =

m+n s=1

m∗ Ω

2 ∂w dαdβ , ∂t

(3.51)

ρ s (δ s − δ s−1 ) denotes the mass of the shell element.

A variation of the external forces’ work can be presented in the form [676] ∂δw + F¯ α δu − z qδwdαdβ − δA = − ∂α Ω

Ω1

∂δw ¯ ¯ Fβ δυ − z + Fn δw dΩ, ∂β

(3.52)

where F¯ α , F¯ β , F¯ n are the components of the external force, given on the part of the surface the surface Ω1 and q denotes 4 4 4 load. Introducing 4 the following 4 notations ¯ α = F¯ α zdz, T¯ β = F¯ β dz, M ¯ β = F¯ β zdz, T¯ n = F¯ n dz, we T¯ α = F¯ α dz, M substitute the derivatives of the variations δwα and δwβ by the following expressions δwα = µ1 δwµ − µ2 δwt ,

δwβ = µ2 δwµ − µ1 δwt ,

(3.53)

− −u ,→ where → t denote the unit vectors of the orthogonal coordinates attached to the boundary contour. The µ1 , µ2 denote a normal vector projection onto the coordinates, associated with the coordinate lines of the deformed surface. A variation of the work is defined as follows


δA = −

qδwdαdβ −

289

:

T¯ α δu + T¯ β δυ+

Ω

$ %

¯ α − µ1 M ¯ β δw − µ1 M ¯ α − µ2 M ¯ β δwµ ds. T¯ n − µ2 M t

(3.54)

The energy variation of the whole shell can be expressed by the partial variation δV = δVu + δVυ + δVF + δVw ,

(3.55)

where the partial variation element has the following form [290] ⎡⎢ ∂V1 ∂V1 ∂V1 ∂V1 ⎢⎢⎢ ∂V1 − + + + δV f = ⎣⎢ ∂ f − ∂ f ∂ fβ β ∂ fαα αα ∂ fαβ αβ α α ⎤ : ⎥⎥⎥ ∂V1 ∂V1 ∂V1 ⎥⎥⎦ δ f dαdβ + µ1 + µ2 − µ1 − ∂ fα ∂ fβ ∂ fαα α ββ ⎫ ⎪ ∂V1 ∂V1 ∂V1 ⎪ 1 1 ⎬ − µ1 − µ2 µ2 δf+ ⎪ ⎭ ∂ fββ β 2 ∂ fαβ β 2 ∂ fαβ αβ ⎪ 1 ∂V1 ∂V1 ∂V1 1 ∂V1 δ fβ ds. + µ2 δ fα + µ2 + µ1 µ1 ∂ fαβ 2 ∂ fαβ ∂ fββ 2 ∂ fαβ

∂V1 ∂ fββ

(3.56)

In the above, V1 denotes the subintegral expression of (3.50). Taking into account (3.53), the partial variations in the explicit form is presented :

(3.57) δVu = µ1 Fββ − µ2 Fαβ δuds, δVυ = δVF =

−

:

−µ1 Fαβ + µ2 Fαα δυds,

A12 Fββ + A22 Fαα

αα

+

Ω

(3.58)

1

A66 Fαβ + αβ 2

1

A11 Fββ + A12 Fαα + A66 Fαβ + d21 wαα + d22 wββ + k2 − ββ αβ αα 2

d66 wαβ + d11 wαα + d12 wββ + k1 w − d66 wαβ + αβ

2

ββ

αβ

1

wαα wββ − 2wαβ wαβ + wββ wαα + (A12C1T + A22C2T )αα + (A11C1T + A12C2T )ββ δFdαdβ,

δVw = Ω

1

− d11 Fββ + d21 Fαα + k2 F

αα

(3.59)

− d66 Fαβ + αβ

290


− d66 Fαβ + ββ αβ $

$& % % & ∗ ' ' ∗ ∗ D11 − D11 wαα + D12 − D12 wββ + 2 D66 − D66 wαβ + αα αβ $

$& % % & ∗ ' ' ∗ ∗ D12 − D12 wαα + D22 − D22 wββ + 2 D66 − D66 wαβ + ββ αβ

Fββ wαα − 2Fαβ wαβ + Fαα wββ + (d11C1T + d21C2T − K1T )αα +

d12 Fββ + d22 Fαα + k1 F

(d12C1T + d22C2T − K2T )ββ δwdαdβ+ : $

µ1 Fββ wα − Fαβ wβ + M1α + Hβ +

µ2 Fαα wβ − Fαβ wα + M2β + Hα δwµ −

µ21 M1 + 2µ1 µ2 H + µ22 M2 δwµ + % $

% µ1 µ2 (M1 − M2 ) − H µ21 − µ22 δw ds.

(3.60)

t

The kinetic energy variation of the shell has the following form [684] ∂2 w δK = − m∗ 2 δwdαdβ . ∂t

(3.61)

Ω

Substituting (3.54), (3.57)–(3.61) into (3.48) we get the equilibrium equation, the equilibrium of deformation continuity and the boundary condition for a dynamical problem of the elastic thermo-sensitive thin-walled shell with the orthotropic and design-orthotropic layers. – The equilibrium equation


αα

+ d66 Fαβ

αβ

$&

'

+ d12 Fββ + d22 Fαα + k1 F + ββ

&

'

%

+ D∗11 − D11 wαα + D∗12 − D12 wββ + αα $& $

% % & ' ' 2 D∗66 − D66 wαβ + D∗12 − D12 wαα + D∗22 − D22 wββ + αβ ββ

$

% 2 D∗66 − D66 wαβ + Fββ wαα − 2Fαβ wαβ + Fαα wββ − d66 Fαβ

αβ

αβ

(K1T − d11C1T − d21C2T )αα− (K2T − d12C1T − d22C2T )ββ + q − m∗

∂2 w = 0. (3.62) ∂t2

– Equation of continuity deformation

1

A66 Fαβ + A11 Fββ + A12 Fαα + αβ ββ 2

+ d21 wαα + d22 wββ + k2 w − d66 wαβ +

A12 Fββ + A22 Fαα 1

A66 Fαβ αβ 2

αα

+

αβ


d11 wαα + d12 wββ + k1 w − d66 wαβ

αβ

291

+

1

wαα wββ − 2wαβ wαβ + wββ wαα + (A12C1T + A22C2T )αα + 2 (A11C1T + A12C2T )ββ = 0 .

(3.63)

– Boundary conditions

$

¯ α − µ1 M ¯ β − µ1 Fββ wα − Fαβ wβ + M1α + Hβ + T¯ n − µ2 M t

µ2 Fαα wβ − Fαβ wα + M2β + Hα −

% µ1 µ2 (M1 − M2 )t + Ht µ21 − µ22 = 0,

¯ α + µ2 M ¯ β − µ21 M12 + 2µ1 µ2 H + µ22 M22 = 0, µ1 M

T¯ β − −µ1 Fαβ + µ2 Fαα = 0,

T¯ α − µ1 Fββ − µ2 Fαβ = 0 .

(3.64) (3.65) (3.66) (3.67)

The kinematic boundary conditions read [290] u = uK ,

υ = υK ,

w = wK ,

∂w = γK , ∂µ

(3.68)

where uK , υK , wK , γK are the given contour displacements the angular displacement −u . of the normal → Consider more detailly, the boundary conditions (3.64)–(3.67). We introduce the system of the orthogonal coordinates αK , βK , z on the shell contour (see Fig. 3.2). −˜ −˜ → −µ˜ ,→ Let the unit vectors → t , l of the coordinates define the right hand coordinate −r˜ be the radius-vector of the sursystem. Let ds be the element of the contour arc; → −˜ → −˜ → −˜ → face; and let l 1 , l 2 , l n be the unit vectors of the deformable surface. Considering the deformations as small and taking the quantities ds, ti , ui as the same for both the deformable and the undeformable surfaces [492], we get → −˜ → − t = t,

→ −µ˜ = µ,

−˜ → −˜ → −˜ → −˜ → −˜ → − −˜ → −r˜ β = → l 1 t1 + l 2 t2 = l 2 µ1 − l 1 µ2 , t ≈ t =→ r s = −r˜α α s + → β s → −˜ → −˜ → −˜ → −˜ − − → → −µ˜ = µ = → t × l n = l 1 t2 − l 2 t1 = l 1 µ1 + l 2 µ2 . → −˜ → −˜ → − → − → − In the above t 1 , t 2 are the vector t projections on the directions l 1 , l 2 −µ2 =

dα = t1 = − sin α0 , ds

t2 = cos α0 =

dβ = µ1 , ds

292


Figure 3.2. An orthogonal coordinate system on the shell contour.

−˜ −u˜ and → where α0 denotes the angle between two unit vectors → l 1 . If the limiting cross section overlaps with the α curve, then α0 =

π , 2

→ −˜ t1 = − l 1 ,

−˜ → −µ˜ = → l 2,

T¯ β = Fαα , ¯ β − M2 = 0, M

µ2 = 1,

µ1 = 0,

T¯ α = −Fα,β ,

¯ α − Fαα wββ − Fαβ wα + M2β + 2Hα = 0. T¯ n + M α

(3.69)

If the limiting cross section overlaps with the β curve, then α0 = 0,

¯ α − M1 = 0, M

µ2 = 0,

µ1 = 1,

→ −˜ → l 1 = −µ˜ ,

→ −˜ → −˜ l2= t,

T¯ β + Fαβ = 0, T¯ α + Fββ = 0,

¯ β − Fββ wα − Fαβ wβ + M1α + 2Hβ = 0. T¯ n + M β

(3.70)

If the shell does not have the boundaries, the limiting conditions are substituted by the periodicity conditions along the two coordinates. If the shell is closed along one of the coordinates, then the periodicity condition is used only along one of the coordinates. Considering shells with initial imperfections the equations’ formulations are analogical to the ones described earlier. Suppose that a shell has the initial buckling w0 (α, β). We assume that the amplitude of those displacements does not exceed


293

the shell’s thickness. We also take it to guarantee that the occurring form imperfections resulting from those displacements have a character of sloped parts. Then, the expressions governing the deformations have the following form [685] 1 1 ε1 = uα − k1 w + w2α − k1 w0 + w20α , 2 2 1 1 ε2 = υβ − k2 w + w2β − k2 w0 + w20β , 2 2 ε12 = uβ + υα + wα wβ − w0α w0β .

(3.71)

The quantity w − w0 should be used in order to define full deformations. Below, we give the final expressions for the equilibrium equation and the deformation continuity. – The equilibrium equation


αα

− d66 Fαβ

αβ

+ d12 Fββ + d22 Fαα + k1 F − ββ

$&

%

& ' ' + D∗11 − D11 (w − w0 )αα + D∗12 − D12 (w − w0 )ββ + αα $

% ∗ 2 D66 − D66 (w − w0 )αβ + αβ % $& & ' ' D∗12 − D12 (w − w0 )αα + D∗22 − D22 (w − w0 )ββ + ββ

% $

2 D∗66 − D66 (w − w0 )αβ + Fββ wαα − 2Fαβ wαβ + Fαα wββ −

d66 Fαβ

αβ

αβ

(K1T − d11C1T − d21C2T )αα− (K2T − d12C1T − d22C2T )ββ+ q − m∗

∂2 w = 0. (3.72) ∂t2

– The equation of deformation continuities

A12 Fββ + A22 Fαα

αα

+

1

A66 Fαβ + A11 Fββ + A12 Fαα + αβ ββ 2

$ % 1

A66 Fαβ + d21 (w − w0 )αα + d22 (w − w0 )ββ + k2 (w − w0 ) − αβ αα 2 $ % % $ d66 (w − w0 )αβ + d11 (w − w0 )αα + d12 (w − w0 )ββ + k1 (w − w0 ) + αβ

ββ

$ % d66 (w − w0 )αβ

1

+ wαα wββ − 2wαβ wαβ + wββ wαα − αβ 2 1

w0αα w0ββ − 2w0αβ w0αβ + w0ββ w0αα + 2 (A12C1T + A22C2T )αα + (A11C1T + A12C2T )ββ = 0.

(3.73)

294


In the above, in both (3.72) and (3.73) w denotes a full deflection. The boundary conditions are identical to the boundary conditions (3.64)–(3.67). Consider a one-layer homogeneous shell. In this case we have ∆=

h , 2

δi = 1, B12 = B21 =

δ1 = h, G=

νE , 1 − ν2

E1 = E2 = E,

E , 2(1 + ν)

B11 =

E , 2(1 + ν)

B66 =

ν1 = ν2 = ν, E = B22 , 1 − ν2 β1 = α∗ (1 + ν) = β2 .

We introduce the following variables [519] h

B (α, β) =

1 1 − ν2

2 E (T ) dz, − h2 h

1 A (α, β) = 1 − ν2

2 E (T ) zdz, − h2 h

D (α, β) = h

1 NT (α, β) = 1−ν

2 − h2

1 1 − ν2

2 E (T ) z2 dz, − h2

⎛ T ⎞ ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟ E (T ) ⎜⎜⎜ α∗ (T ) dT ⎟⎟⎟⎟⎟ dz = C1T = C2T , ⎝ ⎠ T0

⎛ T ⎞ h 2 ⎜⎜⎜ ⎟⎟⎟ 1 ⎜⎜ ⎟ ⎜ (T ) (α, (T ) E z ⎜⎜⎜ α∗ MT β) = dT ⎟⎟⎟⎟⎟ dz = K1T = K2T , 1−ν ⎝ ⎠ − h2

T0

and we express the other variables and the expression using the functions C11 = C22 = B,

C12 = νB = C21 ,

K11 = K22 = A, K12 = νA = K21 , 1−ν 1−ν B, K66 = a, C66 = 2 2 D11 = D22 = D, A11 = A22 =

D12 = D21 = νD,

1 , B(1 − ν2 )

A12 = A21

1−ν D, 2 1 1 , =− 1−νB

D66 =


295

A 2 1 , d12 = d21 = 0, A66 = , B 1−νB A A2 νA2 1 − ν A2 , D∗12 = D∗21 = , D∗66 = . (3.74) d66 = , D∗11 = D∗22 = B B B 2 B The relations between the forces, moments and deformations have the form d11 = d22 =

T 1 = B (ε1 + νε2 ) + A (κ1 + νκ2 ) − NT , T 2 = B (νε1 + ε2 ) + A (νκ1 + κ2 ) − NT , 1−ν (Bε12 + Aκ12 ) = s21 = s, s12 = 2 M1 = D (κ1 + νκ2 ) + A (ε1 + νε2 ) − MT , M2 = D (νκ1 + κ2 ) + A (νε1 + ε2 ) − MT , 1−ν (Dκ12 + Aε12 ) = H21 = H. (3.75) H12 = 2 The relations between the deformations and the forces and between the curvatures and moments are as follows $ % A 1 ε1 = & ' Fββ − νFαα + (1 − ν) NT + wαα , 2 B B 1−ν $ % A 1 ' −νFββ + Fαα + (1 − ν) NT + wββ , 2 B B 1−ν 1 2 2A · Fαβ + wαβ , ε12 = − 1−ν B B

2 A A A − D wαα + νwββ + NT − MT , M1 = Fββ + B B B

2 A A A − D νwαα + wββ + NT − MT , M2 = Fαα + B B B 2 A A H = − Fαβ + (1 − ν) − D wαβ . B B

ε2 =

&

(3.76)

– The equilibrium equation to the problem of the initial imperfections has the form $ 2 2 % A A − D ∇2 ∇2 (w − w0 ) + 2 − D ∇2 (w − w0 ) + α B B α $ 2 % A2 A − D ∇2 (w − w0 ) + ∇2 − D ∇2 (w − w0 ) − 2 β B B α 2 A A (1 − ν) L − D, w − w0 + L , F + L (w, F) − B B A ∂2 w ∇2 MT − ∇2 NT + q − m 2 + k2 Fαα + k1 Fββ = 0. (3.77) B ∂t

296


– The equation of deformation continuity to the problem of the initial imperfection has the form

1 1 1 2 2 2 2 2 1 ∇ ∇ F+2 ∇2 F− ∇ F +2 ∇ F +∇ α β B B α B β B A

1 , F + L , w − w0 · 1 − ν 2 + B B $ %

1 − ν2 k2 (w − w0 )αα + k1 (w − w0 )ββ + N 1

T 1 − ν2 [L (w, w) − L (w0 , w0 )] + (1 − ν) ∇2 = 0. 2 B

(1 − ν) L

(3.78)

In (3.77) and (3.78) the following operators are introduced ∇2 ∇2 f = fαααα + 2 fααββ + fββββ , ∇2 f = fαε + fββ , L ( f, g) = fαα gββ − 2 fαβ gαβ + fββ gαα . It should be noted that in the case of dynamical problems (3.77), (3.78) a simplification is not achieved by taking the coordinate surface in order to get A = 0. This is caused by an occurrence of the unstationary temperature fields, and it is impossible to choose the constant in the surface. For a stationary temperature field the equations (3.77), (3.78) (using the simplification A = 0) are identical to the equations given in [519]. To conclude a generalization of the thermoelastic equations for dynamical problems of a thin-walled thermosensitive conical shell has been developed. For the temperature stresses determination we need to define the temperature field. The heat transfer equation can be obtained from the equation of (3.10) type, neglecting the time dependent terms. The heat transfer equation for a thin-walled multilayer orthotropic conical shell has the form [486] s s ∂T s ∂ s ∂T s ∂ s ∂T s ∂ s ∂T s ∂T λT α + λT β + λT z + 2kλTs z = Cε=0 . (3.79) ∂α ∂α ∂β ∂β ∂z ∂z ∂z ∂t Above λTs α , λTs β , λTs z are the heat transfer coefficients, and Cε=0 is the thermal capacity for a constant deformation of the s-th layer orthotropic material. On the shell’s boundary surfaces the boundary conditions (3.16)–(3.18) should be satisfied, whereas between the layers we have the following continuity conditions of the heat flows and the temperature: λTs−1 z

∂T s−1 ∂T s = λTs z , ∂z ∂z T s−1 = T s .

(3.80) (3.81)

In some works [519, 549] it has been noted that for thin shells the temperature distribution along the thickness can be obtained with a high accuracy using the heat transfer equation for a flat wall.


297

3.2.4 Solution Existence The problems of solution existence in mechanics and physics are related to the qualitative investigations of the operator equations. A suitability of the theory which reduces that problem to the mathematical scheme of the physical world behaviour independently of the experiments should be verified [206]. On one hand, it corresponds to the theory of multidimensional singular potentials and singular integral equations. On the other hand, it is related to the theory of generalized solutions to the differential equations (the methods of Hilbert’s spaces, the variational methods) [396]. In this chapter, the latter approach is used, which is more powerful and which includes the case of the variable coefficients and variable boundary conditions. It includes the following fundamental steps: – a derivation of the a priori estimation; – an application of those estimations. A choice of the functional spaces, where a solution is located, plays an essential role to getting and applying a priori estimation of the compactness method [438]. The obtained (in this work) results are generalizations of the known ones [438, 481, 690] to the case of the thermosensitive conical shell’s vibrations. A special attention has been paid to the formulation and the theorem’s proof (on the existence) for the outlined problem in the case of a sufficient regular surface using the Dirichlet boundary conditions. This approach is supported by an application of the Green’s function method. Consider the bounded space Ω0 in Rz (Ω0 is the vibrated shell). We are looking for a function pair w, F defined in Ω0 ×]0, tK [, satisfying the equations (3.62), (3.63) and governing dynamical behaviour of a conical shell with the material characteristics depending on the temperature, and with the boundary conditions ∂w = 0, ∂n ∂F = 0, F, ∂n

w,

,

(3.82)

∂w = w 0 . ∂t t=0

(3.83)

on

and with the following initial conditions w|t=0 = w0 ,

We assume that the temperature field is stationary. In this case we can simplify the equations (3.62), (3.63), when the coordinate surface is chosen to get A = 0. Denoting the surface by z0 and introducing the notations α = x, β = y, we get the following equation: h

A (x, y) =

1 1 − ν2

2 E (x, y, z) (z − z0 )dz = 0. − h2

(3.84)

298


The quantities B and NT in the formula (3.73) are unchanged, whereas D and MT have the following form: h

1 D = D (x, y) = 1 − ν2 h

MT = MT (x, y) =

1 1 − ν2

2 − h2

2 E (T ) (z − z0 )2 dz, − h2

⎛ T ⎞ ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟ E (T ) (z − z0 ) ⎜⎜⎜ α∗ (T ) dT ⎟⎟⎟⎟⎟dz. ⎝ ⎠

(3.85)

T0

It should be noted that in the considered problem the temperature is supposed to be known. Therefore the functions B, D, NT are also known. We rewrite the equations (3.62), (3.63) taking into account the above assumption:

(3.86) mw

+ ∆21D + ν∆22D w − [w, F] − {k, F} + ∇2 MT − q = 0, N

1

T ∆21H − ν∆22H F + 1 − ν2 [w, w] + 1 − ν2 {k, w} (1 − ν) ∇2 = 0. (3.87) 2 B Above, the notation given in [536] is used ∂2 ∂2 ∂2 ∂2 ∂2 ∂2 G + 2 G 2 , ∆21G ≡ 2 G 2 + 2 ∂x∂y ∂x∂y ∂x ∂x ∂y ∂y

∂2 f ∂2 f + k , [u, υ] = L(u, υ), 2 ∂y2 ∂x2 ∂2 ∂2 ∂2 ∂2 ∂2 ∂2 G + 2 G 2 . ≡ 2 G 2 −2 ∂x∂y ∂x∂y ∂x ∂y ∂y ∂x {k, f } ≡ k1

∆22G

We introduce the space H02 (Ω0 ) [438] ∂υ = 0 on Γ H02 (Ω0 ) = υ| υ ∈ H02 (Ω0 ) , υ = 0, ∂n

(3.88)

(3.89)

Therefore, H02 (Ω0 ) creates the Sobolev space [628] of functions taking zero values on the space boundary together with their derivatives. Denote by Q the cylinder in R2 × Rt : Q = Ω0 ×]0, tK [, where tK is finite, and by its side’s limit: = Γ×]0, tK [. p By L (0, tK ; Ω0 ) [438] we denote the space of the functions t - f (t) :]0, tK [→ Ω0 , which are measured, having the values in Ω0 and satisfying the condition ⎛ tK ⎞ 1p ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟ p ⎜⎜⎝ f Ω0 dt⎟⎟⎟⎟⎠ = f L p (0,tK ;Ω0 ) < ∞. 0

(3.90)


299

If p = ∞, then the form (3.90) is changed by the norm [438] sup ess f (t)Ω = f L∞ (0,tK ;Ω0 ) ,

t ∈ ]0, tK ]

(3.91)

and we have L p (0, tK ; L p (Ω)) = L p (Q). THEOREM 3.1 Let the q, ∇2 MT , ∇2 ( NBT ), k1 , k2 , w0 , w 0 be given, and q ∈ L2 (Q) ,

(3.92)

∇2 MT ∈ L2 (Ω) , N T ∇2 ∈ L1 (Ω) , B

(3.93)

k1 , k2 ∈ L2 (Ω) , w0 ∈

H02

(Ω) .

w 0 ∈ L2 (Ω)

(3.94) (3.95) (3.96) (3.97)

Then, there exist w and F, which satisfy (3.86), (3.87), (3.82), (3.83), and

(3.98) w ∈ L∞ 0, tK ; H02 (Ω) ,

w ∈ L∞ 0, tK ; L2 (Ω)

F ∈ L∞ 0, tK ; H02 (Ω) .

(3.99) (3.100)

Comment 3.1 From (3.98), (3.100) and the definition (3.88) it results that [ω, F] ∈ L∞ (0,K , L1 (Ω)) and the equation (3.86) leads to the inclusion ω

∈ L∞ (0, tK ; H −2 (Ω)) that the conditions (3.83) have sense. In the work [438] some properties of [u, υ] are proved, which are used in our further considerations. Lemma 3.1 The transformation u, υ → [u, υ] is the bilinear transformation H02 (Ω)× H02 (Ω) → H −2 (Ω). Corollary 3.1 The form u, υ, f → ([u, υ], f ) is the continuous trilinear form in H02 (Ω). Lemma 3.2 The trilinear form u, υ, f → ([u, υ], f ) is symmetric in H02 (Ω). In the analogical way it can be proved that the same properties are applied to the form k, u, υ → ({k, u}, υ). Lemma 3.3 The trilinear form k, u, υ → ({k, u}, υ) is the symmetric trilinear form in H02 (Ω). Lemma 3.4 The operators ∆21D + ν∆22D and ∆21H + ν∆22H are symmetric, positive defined operators [481]. Lemma 3.5 The operators ∆21D + ν∆22D and ∆21H + ν∆22H are strong elliptic operators [679, 680].

300


Corollary 3.2 The operator ∆21D + ν∆22D has in Ω the whole system of vectors [680]. From the given properties the following lemma is concluded. Lemma 3.6 A strongly elliptic operator ∆21H + ν∆22H transforms the space H02 (Ω) into the conjugated space H −2 (Ω) heomorphicly [184]. In particular, it means that there exists the operator G F , inversed to ∆21H + ν∆22H , which is the continuous operator form H −2 (Ω) → H02 (Ω). It means that there exists the “Green’s operator” inversed to ∆21H + ν∆22H in Ω with the Dirichlet conditions. Proof I. The construction of the approximate solution. Let υ1 , . . . , υm be the basis defined by the functions in D(Ω) (g ∈ D(Ω) - the space of the functions of the class C ∞ in Ω, having the compact carrier in Ω), having the following properties: a) υi ∈ H02 (Ω) ∀i; b) ∀m υ1 , . . . , υm are linearly independent; c) the linear combinations υi are densed in H02 (Ω). Let wm (t) satisfy the following conditions wm (t) ∈ [υ1 , ..., υm ] ,

i.e. wm (t) =

m

gim (t) υi ,

(3.101)

i=1

w

m (t) , υ j + ∆21D + ν∆22D wm (t) , υ j − [Fm (t) , wm (t)] , υ j −

(3.102) {k, Fm (t)} , υ j + ∇2 MT , υ j − q, υ j = 0 , 1 ≤ j ≤ m, wm (0) = w0m ∈ [υ1 , ..., υm ] , w0m → w0 in H02 (Ω) ,

(3.103)

w m

(3.104)

(0) =

w 0m

∈ [υ1 , ..., υm ] ,

w 0m

→

w 0

in L (Ω) . 2

Define Fm (t) by the relation

1 ∆21H − ν∆22H Fm (t) = − (1 − ν) [wm (t) , wm (t)] − 2 N

T 1 − ν2 {k, wm (t)} − (1 − ν) ∇2 (3.105) = 0, Fm (t) ∈ H02 (Ω) . B Observe that Fm (t) do not need to belong to [υ1 , . . . , υm ]. In accordance with the Lemma 3.6, Fm (t) can be presented in the following form 1

Fm (t) = G F − 1 − ν2 [wm (t) , wm (t)] − 2 N

T . (3.106) 1 − ν2 {k, wm (t)} − (1 − ν) ∇2 B Substituting these relations to (3.102), we can get the system of ordinary differential equations in relation to wm (t). Therefore, we can believe on the existence of wm (t) according to the solvability of the ordinary differential equations, and, on the existence of Fm (t) in a certain interval [0, tm ], tm > 0.


301

II. A priori estimation. Multiplying (3.102) by g m and making a sum in relation to j, we get '

&

wm (t) , w m (t) + ∆21D + ν∆22D wm (t) , w m (t) − ' & ' & [Fm (t) , wm (t)] , w m (t) − {k, Fm (t)} , w m (t) + &

' (3.107) ∇2 MT , w m − q, w m = 0. However, according to the Lemmas 3.2 and 3.3, we have ' &9 / ' & {k, Fm (t)} , w m (t) = k, w m (t) , Fm (t) , ' &5 6 ' & [Fm (t) , wm (t)] , w m (t) = wm (t) , w m (t) , Fm (t) = 1d [wm (t) , wm (t)] , Fm (t) . 2 dt

(3.108)

According to (3.105), the equation (3.108) is transformed to the following one 1 d

2 2 (t) (t) − ν∆ , F − F − ∆ m m 1H 2H 1 − ν2 dt d 1 d 2 NT (3.109) ∇ , Fm (t) . {k, wm (t)} , Fm (t) − dt 1 + ν dt B Substituting (3.108), (3.109) and (3.107), we obtain '

&

wm (t) , w m (t) + ∆21D + ν∆22D wm (t) , w m (t) + &9 / ' 1 d

2 ∆1D + ν∆22D Fm (t) , Fm (t) + k, w m (t) , Fm (t) − 2 1 − ν dt & / '

' &9 (3.110) k, w m (t) , Fm (t) + ∇2 MT , w m (t) − q, w m (t) = 0. Furthermore, using the relations, and according to the operators ∆21H + ν∆22H and ∆21D + ν∆22D properties, we get

1 d

d ∆21H − ν∆22H Fm (t) , Fm (t) ≡ ∆21H − ν∆22H Fm , Fm , dt 2 dt 1 d

∆21D + ν∆22D wm (t) , wm (t) , ∆21D + ν∆22D wm (t) , w m (t) ≡ 2 dt &

' 1d &

' 1 d & '2 (3.111) wm (t) , w m (t) = w (t) , w m (t) = w (t) , 2 dt m 2 dt m the expression (3.110) is transformed to the following form 1 d 2

2 wm (t) + ∆1D + ν∆22D wm (t) , wm (t) + 2 dt &

'

1 ∆21H − ν∆22H Fm (t) , Fm (t) = q, w m (t) − ∇2 MT , w m (t) , (3.112) 2 1−ν

302


and it implies that 1 2

2 wm (t) + ∆1D + ν∆22D wm (t) , wm (t) + 2 & 1

2 2 ∆1H − ν∆2H Fm (t) , Fm (t) = q (σ) , w m (σ) dσ− 1 − ν2 t

0

t

2

1 ∇2 MT , w m (σ) dσ + w 0m + ∆21D + ν∆22D w0m , w0m + 2

0

1

2 2 (0) (0) − ν∆ , F . ∆ F m m 1H 2H 1 − ν2 According to (3.103), (3.104) we have 2

2 w0m + ∆1D + ν∆22D w0m , w0m ≤ const.,

(3.113)

and taking into account (3.106) we get 1

Fm (0) = G F − 1 − ν2 [wm0 , wm0 ] − 2 N T . (3.114) 1 − ν2 {k, wm0 } − (1 − ν) ∇2 B However, [w0m , w0m ] belongs to the bounded set in L1 (Ω), and correspondingly, in H −2 (Ω). Therefore, Fm (0) belongs to the bounded set in H02 (Ω) and

∆21H − ν∆22H Fm (0) , Fm (0) ≤ const.

Furthermore, using the property of positively defined operators ∆21D + ν∆22D and ∆21H +ν∆22H , the Cauchy inequality for ε, and using the Gronwille’s lemma [437], it can be shown that tm = tK and

wm , Fm are bounded in L∞ 0, tK ; H02 (Ω) , (3.115)

w m are bounded in L∞ 0, tK ; L2 (Ω) .

(3.116)

III. Limiting transformations. According to (3.71), (3.72) we can define the series wµ , Fµ , in the following manner: wµ → w is weak in L∞ (0, tK ; H02 (Ω)), Fµ → F is weak in L∞ (0, tK ; H02 (Ω)), w µ → w is weak in L∞ (0, tK ; H02 (Ω)),


wµ → w is strong in L2 (Q) (according to Sobolev theorem).

303

(3.117)

Let the functions ϕ j , 1 ≤ j ≤ j0 belong to 1 ([0, tK ]), ϕ (tK ) = 0 and ψ =

j0

ϕ j ⊗ υ j,

(3.118)

j=1

Owing to (3.102) and for m = µ > j0 we have −

tK

w µ ψ

dt +

0

tK

tK

∆21D

+

ν∆22D

wµ , ψ dt −

0

(q, ψ) dt −

k, Fµ , ψ dt =

0

% Fµ , wµ , ψ dt−

0

tK

tK $

0

tK

∇2 MT , ψ dt + w µ (0) , ψ (0) . (3.119)

0

However, according to the Lemma 3.2, we have tK $

%

Fµ , wµ , ψ dt =

0

tK $

% ψ, Fµ , wµ dt,

0

[ψ, Fµ ] → [ψ, F] is weak in L2 (Q), and because wµ → w is strong in L2 (Q), then we get tK $

%

tK

Fµ , wµ , ψ dt →

0

&5

6

'

tK ([w, F] , ψ) dt,

ψ, F , w dt =

0

0

and (3.119) in the limit achieves the expression tK

&

'

w , ψ dt +

−

tK

0

tK

+

ν∆22D

0

tK ({k, F} , ψ) dt =

0

∆21D

(q, ψ) dt − 0

w, ψ dt − ([w, F] , ψ) dt− tK

0

tK

∇2 MT , ψ dt + w 0 , ψ (0) ,

(3.120)

0

which is true for all ψ of the (3.118) type. Using the limiting transition we deduce that (3.120) is satisfied for all ψ ∈ L2 (0, tK H 2 (Ω)), and ψ ∈ L2 (0, tK ; L2 (Ω)). Therefore it has been shown that w, F satisfy (3.86) and w (0) = w 0 . In order to get (3.87) we can directly achieve the limit in (3.105) (for m = µ). Taking into account that [wµ , wµ ] → [w, w] for instance, in D (Q) (a space of distribution). If ϕ ∈ D(Q), then

304


tK $ 0

%

wµ , wµ , ϕ dt =

tK $

% wµ , ϕ , wµ dt

0

and it is possible to achieve the limit, as in the above. A proof of existence and uniqueness theorems for the boundary value problems of the heat transfer is given in reference [405]. 3.2.5 Classification The given above mathematical model of dynamical problem of the temperature stresses of thin thermosensitive shells is constructed under many assumptions and hypotheses, and it includes different types of possible deviations. We consider now a class of problems, which can be solved in a frame of the given mathematical model using the approach in which “...the model accuracy should not extend the intervals of deviation of the initial data, and no one of the quantities should be calculated more accurately than it is required from the point of view of the considered problem” [508]. Using the Kirchhoff-Love hypothesis in order to reduce the three dimensional boundary value problem, the deviation Rh is assumed. Using this estimation and taking into account the computational accuracy (5%) we assume that only relatively 1 ) are considered [534]. long shells ( Rh ≤ 20 The theory of conical shells, used for the fundamental motion equations’ derivation, can be also applied to the shells with zero Gauss’ curvature and to the shells with a large changes exponent [14]. A phenomenological approach assuming a speed of the heat distribution as infinitely large while considering the heat transfer equations has been used. This assumption is verified by the computations of the temperature fields in different bodies in usual conditions occured in practice. As it has been shown in reference [444], for the thermoelasticity problems, already for temperature higher than 293K (a room temperature) we can limit ourselves to the consideration of a “usual” heat transfer equation. In the given model a coupling mechanical term in equation is not included. A quantitative influence of the omitted term in the considered problems is small, although in some cases it can help to illustrate and describe some of the qualitative effects (for instance, damping and dispersion of the elastic waves) [119, 555]. The inertial parameter certainly belongs to one of the fundamental parameters characterizing the mathematical model in both heat transfer and motion equations. Its role sufficiently depends on the influence of a load duration. It has been shown that for a purely elastic problems considered in reference [17], a loading time duration is considerably smaller in comparison to the mechanical time characteristics of the shell. The load can be considered as the impulse-type load. In the case of the length comparable to the dynamical one (quasi-impulse), and in the case of longer time duration - to the quasistatical one. The analogical approach is applied also for the heat transfer problems [119]. Therefore, we can assume that the role of inertia in


305

the temperature stresses’ problems really depends on the relations between the time durations of the mechanical and heat changes characteristics, and on the duration of the heating loads. Consider a class of problems in which the characteristics time of mechanical and heat behaviour are closely related and are comparable to the time duration of the thermal load. In this case we need to consider the unstationary processes of heat distribution along the shell as well as the dynamical behaviour of the shell. We introduce the following quantities: τmech - characteristic mechanical time; τheat - characteristic heat time; (τimp )mech - time of forcing action; (τimp )heat - time of heat action. Therefore, the latter condition is presented in the following form

≈ τimp . (3.121) τmech ≈ τheat ≈ τimp mech

heat

As the characteristic mechanical time we take that equal to the 0.25 of the maximal period of free vibrations of the homogeneous cylindrical shell [17]: + + ) ' ρ R 4 & π T mech 1 l = · · 3 1 − ν2 , = (3.122) τmech = 4 2ωmin k1 R E h where: ωmin - the minimal frequency of free vibrations of a cylindrical shell; k1 the coefficient characterizing the influence of the boundary conditions (k1 = 1 for a rolling support; k2 = 1.5 - for a stiffly supported shell; k1 = 1.25 for the unsymmetrical boundary conditions). The minimal frequency of free vibrations of the cylindrical shell can be found from the following expression [301] ⎧ ⎫ 2 ⎤ 12 ⎪

⎪ 0 ⎡⎢ mπR 4 ⎪ ⎪ h2 ⎪ ⎪ 4 2 ⎥ ⎪ ⎪ ⎥ ⎢ + n − 1 n ⎪ ⎪ ⎪ ⎥⎥⎥⎥ ⎪ 12(1−ν2 )R2 ⎨ 1 E ⎢⎢⎢⎢ l ⎬ ⎥ ⎢ ωmin = min ⎪ , (3.123) ⎪ ⎥ ⎢

⎪ ⎪ ⎥ ⎢ 2 ' & ⎪ ⎪ n,m ⎪ ⎥ ⎢ R ρ mπR ⎪ 2 2 ⎦ ⎣ ⎪ ⎪ + n + 1 n ⎪ ⎪ l ⎩ ⎭ where n denotes the waves’ number in the circled direction, and m denotes the halfwaves’ number in the longitudinal direction. The time corresponding to the time of heat distribution along the shell serve as the characteristic heat time τheat = k2

b2 , a

b = (l, R, h) ,

(3.124)

where a denotes the temperature transfer coefficient, and k2 denotes the coefficient characterizing the order of the shell’s overheating, which lies in the interval from 0.08 to 1.0. 1 are considered and in order to Assuming that the thin-walled shells with Rh ≤ 20 satisfy the condition (3.121), we need to take into account the process of heat distribution along the shell’s thickness. For example, for a shell made from aluminium

306


and magnesium with the following physical characteristics [392]: E = 7100MPa, 2 ρ = 2700 mkg3 , ν = 0.3, a = 5 · 10−5 ms and the geometrical characteristics Rh = 250, l −3 R = 4, h = 10 m the characteristic mechanical time τmech and the characteristic heat time τheat are equal to (k1 = 1, k2 = 0.1) τmech ≈ 2 · 10−3 s,

(τheat )h ≈ 2 · 10−3 s,

l2 l2 R2 = k2 2 · = 16 (τheat )R = a a R l2 R2 h2 = 106 (τheat )h >> (τheat )h . k2 2 · 2 · (3.125) a R h The condition (3.121) (in this case) can be presented in the following form + + )

' ρ R 4 & h2

l 3 1 − ν2 ≈ k2 ≈ τimp ≈ τimp . (3.126) mech heat 2k1 E h a (τheat )l = k2

A choice of the k2 coefficient is motivated by the following model problem. Consider an infinite plate made from a material with constant characteristics and having zero initial temperature. In the initial time the external side of the plate is influenced by the action of higher temperature T , whereas the internal side is heat isolated. As the characteristic heat time we can take τheat (after that the heat achieves the shell’s internal side). According to the references [119] this time is given by: τheat ≈ 0.0885

h2 , a

k2 = 0.0885.

(3.127)

We have to add that in this time moment the temperature moment MT achieves its maximum (maximal gradient of temperature). During the estimation of free vibration frequencies of the heated shell a minimal frequency of the cylindrical shell (ununiformely heated along its thickness) is obtained: 1 T max + T min 2 , (3.128) ωT = ωmin 1 + nE 2 where: E = E(z) = E0 (1 + nE T ), T max and T min are characterized by the temperature change along the thickness. It results from (3.128) that free vibration frequency of the heated shell is smaller than that of a cold shell (3.123). Therefore, in the case of dynamical behaviour of thin thermosensitive elastic shells, for a simultaneous inclusion of inertial effect in the motion and heat transfer equations a special attention should be paid to the heat distribution processes only along the shell’s thickness. The heat distribution in the average surface can be neglected. During the analysis of the thermoelastic processes some limitations are given to the temperature changes. As the criterion of the limiting state when a thermal load is applied, the following stresses condition can be used σ (T ) ≤ σ s (T ) ,

(3.129)


307

where σ(T ) denotes the acting stresses, and the σ s (T ) denotes the elasticity border for a given temperature. In the process of heating of the load the deformation continuity conditions can be changed. Therefore, only a certain part of the whole heat deformation causes the occurrence of the thermal stresses [661] Fσ (T ) ≤ σ s (T ) .

(3.130)

In the above, F denotes the function defining the continuity deformations condition depended on the boundary conditions and on the time heating excitation duration. The F function is equal to the relation between the thermal stresses and the maximal possible stresses. It is changed in the interval from 0 to 1 [661]. The theoretical and experimental investigations [93, 436] proved that with a lack of structural and phase changes, heat capacity for a constant pressure C p can be treated as linearly coupled via the temperature. In this case, the fundamental material characteristics also linearly depends on the temperature: E = E0 (1 + nE T ) , ν = ν0 (1 + nν T ) ,

α∗ = α∗0 (1 + nα T ) ,

λT = λT 0 (1 + nλ T ) ,

C p = Cε=0 (1 + nc T ) .

(3.131)

Above ne , nα , nν , nλ , nc are the corresponding temperature coefficients from the interval (1...10) × 10−4 , E0 , α∗0 , ν0 , λT 0 are the elasticity modulus, the coefficient of a linear heat expansion, the Poisson’s coefficient, the heat transfer coefficient for the initial temperature, respectively. In further considerations we neglect the Poisson’s coefficient’s dependence on the temperature. The criterion (3.130) (in its most generalized case) can be presented in the following form [661]: ⎧ ⎪ T 2f l T ⎪ ⎪ ⎪ T ⎪ ) (1 , C T ≤ 0.5 + n k ⎪ 3 ε=0 c ⎪ ⎪ T T n1 E0 (1 + nE T ) ⎨ α∗0 (1 + nα T ) dT ≤ ⎪ (3.132) F ⎪ ⎪ 1 − RT ν T 4f l T ⎪ ⎪ ⎪ T0 ⎪ ⎪ ⎩ k4Cε=0 (1 + nc T ) T 3 , T > 0.6 fl

where RT possesses the following values: RT = 1 - for one basic deformation continuity; RT = 1 - for two basic deformations’ continuity; RT = 2 - for three basic deformation’s continuity; T f l - the material flow temperature; k3 , k4 - material constants. It should also be noted that the heat capacity for higher temperature is constant [644]. The Machutov’s relation [635] can be used for the elasticity threshold σ s0 against the temperature estimation. In this case the criterion (3.130) has the following form

308


E0 (1 + nE T ) F 1 − RT ν

T T0

1 1 − , α∗0 (1 + nα T ) dT ≤ σ s0 exp κ T T0

(3.133)

where: σ s0 denotes the elasticity material threshold for T 0 temperature; κ is a material constant. In the temperature range TTf l < 0.6 both formulas (3.132), (3.133) are in good agreement (an error does not exceed 5%). To conclude, knowing the function F for each problem we can define a maximal allowed temperature increase (∆T )max according to the formula (3.132), (3.133). On the other hand, thermal sensitivity governed by (3.131) should be taken into account only if its magnitude increases the allowed error (in our case 5%). The above given condition bounds the temperature increment from below and can be presented in the following way 1 (∆T )δ > δ (3.134) + signnK · T 0 , |nK | where δ denotes the allowed error of 5%, nK = (nE , nα , nc , nλ ) and T 0 is the initial temperature. Depending on the relations between (∆T )max and (∆T )δ the problem can be classified in the following manner A. (∆T )δ < (∆T )max . 1) (∆T )δ < ∆T < (∆T )max - a thermoelastic problem with inclusion of the material characteristics versus temperature; 2) ∆T < (∆T )δ - a thermoelastic problem without inclusion of the material characteristics versus temperature; 3) ∆T > (∆T )max - a thermoplastic problem with inclusion of the material characteristics versus temperature. B. (∆T )δ > (∆T )max . 1) ∆T < (∆T )max - a thermoelastic problem without inclusion of the material characteristics versus temperature; 2) (∆T )max < ∆T < (∆T )δ - a thermoplastic problem without inclusion of the material characteristics versus temperature; 3) ∆T > (∆T )δ - a thermoplastic problem with inclusion of the material characteristics versus temperature. As an example, in a frame of the given classification, consider a body with two basic deformations continuity (RT = 2) made from the material DI6T with the following material characteristics [635]: E0 = 7.2 · 1010 Pa,

α∗0 = 24.3 · 10−6

1 , K

ν = 0.35,

κ = 301.04K,


T 0 = 293K,

T pl = 933K,

nα = 5 · 10−4 ,

σ s0 = 340MPa,

nc = 3.9 · 10−4 ,

309

nE = 5.38 · 10−4 ,

nλ = 4.5 · 10−4 .

In the Fig. 3.3 the dependence of a relative temperature increment ∆T T versus the magnitude of the F functions according to (3.133) for the considered material DI6T is presented.

Figure 3.3. The relative temperature increase versus F function.

The low boundaries of the temperature increment using the temperature dependencies of different considered materials read 5

(∆T )δ

5

(∆T )δ

6 nE

6 nc

= 79K, = 142K,

5

(∆T )δ

5

6

(∆T )δ

nα

6 nλ

= 115K, = 126K.

It means that for the considered material DI6T the temperature dependence in the thermoelastic problems should be considered in the following cases: – – – –

for the elasticity modulus E for F < 0.527; for the linear coefficient of the thermal expansion α∗ for F < 0.329; for the thermal capacity C p for F < 0.248; for the heat transfer coefficient λT for F < 0.291.

To conclude, the described approach separated the class of problems concerning thermoelastic thin cylindrical shells with a simultaneous inclusion of both inertial terms (in the shell’s motion equation and in the heat transfer equation) and material characteristics depending on the temperature.

310


3.3 Computational Algorithms 3.3.1 Finite Difference Equations In order to solve the dynamical problem of a thermosensitive thin homogeneous shell with geometrical nonlinearities we apply the method of finite difference. The fundamental equations (3.77)–(3.80) are presented in the following non-dimensional form 2 $ % A˜ R A˜ 2 1 2 2 2 ˜ ˜ ( ) ( ) ∇ + − D ∇ − D w ˜ − w ˜ + 2 w ˜ − w ˜ ∇ & ' 0 0 l B˜ 1 − ν2 B˜ x˜ x˜ 2 $ % ˜ l A˜ 2 2 2 A ˜ ˜ 2 − D ∇ (w˜ − w˜ 0 ) + ∇ − D ∇2 (w˜ − w˜ 0 ) − y˜ R B˜ B˜ y˜ 2

A˜ ˜ A˜ ˜T− ˜ ˜ w˜ − 1 ∇2 M (1 − ν) L − D, w˜ − w˜ 0 + L , F + L F, ˜ ˜ 1−ν B B 2 2 2 A˜ τ M ∂ w˜ 1 ∇2 + ky F˜ x˜ x˜ + k x F˜ y˜y˜ = 0, (3.135) NT + q˜ − 1−ν τT ∂t˜2 B˜ 1 2 2˜ R 1 2 ˜ l 1 2 ˜ ∇ ∇ F+2 ∇ F +2 ∇ F + x˜ y˜ l B˜ x˜ R B˜ y B˜ 1 1 2˜ A˜ ∇ F − (1 + ν) L , F˜ + L , w˜ − w˜ 0 + ky (w˜ − w˜ 0 ) x˜ x˜ + ∇2 ˜ ˜ B B B˜ ˜ 1 2 NT ˜ w) ˜ − L (w˜ 0 − w˜ 0 )] + ∇ = 0, (3.136) k x (w˜ − w˜ 0 )y˜y˜ + [L (w, 2 B˜ ∂T˜ ∂ ∂T˜ = λT T˜ . (3.137) ∂t˜ ∂˜z ∂˜z The operators ∇2 , ∇2 ∇2 have the following form R l w˜ x˜ x˜ + w˜ y˜y˜ , l R 2 R 2 l 2 2 ∇ ∇ w˜ = w˜ x˜ x˜ x˜ x˜ + 2w˜ x˜ x˜y˜y˜ + w˜ y˜y˜y˜y˜ , l R ∇2 w˜ =

whereas non-dimensional parameters read ˜ x = l x˜, y = R˜y, z = h˜z, w = hw, ˜ F = E0 h3 F, t=

Cε=0 h2 E0 h4 h2 ˜ R2 T , q = 2 2 q, t˜, T = ˜ q1 = 2 q, ˜ λT 0 lRα∗0 l R h

3.3 Computational Algorithms

qT = B=

hλT 0 lR q˜ T , τ M = lRα∗0 h

+

311

ρ h2Cε=0 l R , τT = , k x = , ky = , E0 λT 0 h h

E0 h ˜ E0 h3 ˜ E0 h2 ˜ E0 h3 ˜ B, D = D, A = A, N NT , = T (1 − ν) lR 1 − ν2 1 − ν2 1 − ν2 1

E0 h4 ˜ MT , B˜ = MT = (1 − ν) lR 1

2 A˜ = − 12

2

1 2

˜ ˜ E T d˜z, D = E T˜ z˜2 d˜z,

− 12

− 12

⎛ ⎞ 1 2 ⎜⎜⎜T˜ ⎟⎟⎟

⎜ ⎟ E T˜ z˜d˜z, N˜ T = E T˜ ⎜⎜⎜⎜⎜ α∗ T˜ dT˜ ⎟⎟⎟⎟⎟ d˜z, ⎝ ⎠ − 12

1

2 ˜T = M − 12

T0

⎛ T˜ ⎞

⎜⎜⎜⎜

⎟⎟⎟⎟

E (T ) E T˜ z˜ ⎜⎜⎜⎜⎜ α∗ T˜ dT˜ ⎟⎟⎟⎟⎟ d˜z, E T˜ = , E0 ⎝ ⎠ T0

α∗ (T ) lRα∗0 w0 . , T˜ 0 = T 0 , w˜ 0 = α∗ T˜ = 2 α∗0 h h The numerical investigation of the problem is carried out with the following boundary conditions for the equations (3.135), (3.136) w˜ = 0, w˜ x˜ x˜ = 0, F˜ = 0, F˜ x˜ x˜ = 0, x˜ = 0, x˜ = 1.0,

(3.138)

w˜ = 0, w˜ x˜ = 0, F˜ = 0, F˜ x˜ x˜ = 0, x˜ = 0, x˜ = 1.0,

(3.139)

whereas for the equation (3.137) it reads 1 1 ∂T˜ = 0, z˜ = , T˜ = T˜ 0 , z = − , 2 ∂˜z 2

∂T˜ 1 1 ∂T˜ = q˜ T 0 , z = − , = 0 z˜ = . λT T˜ ∂˜z 2 ∂˜z 2 The initial condition for the equation (3.135) have the following form w˜ = 0,

∂w˜ = 0, t˜ = 0, ∂t˜

(3.140) (3.141)

(3.142)

and for the equation (3.137) T˜ = 0, t˜ = 0.

(3.143)

Solving the problems (3.135), (3.136), (3.138), (3.142) and (3.135), (3.136), (3.139), (3.142) a symmetry related to the line y˜ = 0 and y˜ = π is used. On the symmetry lines the following symmetry conditions are taken: w˜ y˜ = 0, w˜ y˜y˜y˜ = 0, F˜ y˜ = 0, F˜ y˜y˜y˜ = 0.

(3.144)

312


We change the considered area by the arguments 9 / D0 = 0 ( x˜, y˜ , z˜) , 0 ≤ x˜ ≤ 1, 0 ≤ y˜ ≤ π, t˜ ≥ 0 to the area of their discrete values in the nodes of the mesh ωhx hyτ = ωhx ωhy ωτ = {(xk , yl , tn ) , xk = kh x } , yl = lhy , tn = nτ, k = 0, N, l = 0, M, n = 0, 1, 2, . . . , with the constant steps h x , hy , τ (see Fig. 3.4).

Figure 3.4. The mesh used in the finite difference method.

Instead of the functions w, ˜ F˜ we consider the mesh functions wnkl , Fkln given in the mesh nodes (xk , yl , tn ). Analogically, during solutions to the problems (3.137), (3.140), (3.143) and (3.137), (3.141), (3.143) the area of the continuous argument variations 1 1 ˜ D0T = z˜, − ≤ z˜ ≤ , t ≥ 0 2 2 is substituted by the area of discrete values in the nodes of difference mesh ωhz τ = ωhz ωτ = {(zm , tn ) , zm = mhz , tn = nτ, m = 0, K, n = 0, 1, 2, ...


313

with the constant step hz . Instead of the function T˜ we consider its mesh analog T˜ mn given in the mesh nodes (zm , tn ). The nonlinear heat transfer equation is solved for each point of the computational area, where the heat excitation is different from zero. Changing the partial derivatives, occurring in (3.135)–(3.144), by the centraldifference relations [594], we obtain the following finite difference equations A2 − D Λ4 (w − w0 ) + B l A2 R A2 − D [Λ2 (w − w0 )] x◦ + 2 − D ◦ [Λ2 (w − w0 )]y◦ + 2 ◦ l B R B y x 2 2 A A − D Λ2 (w − w0 ) − (1 − ν) L1 − D, w − w0 + Λ2 B B A A 1 1 Λ2 MT − Λ2 NT + L1 , F + L1 (F, w) − B 1−ν 1−ν B ky F x¯ x + k x Fy¯y ,

τM τT

2

1 wt¯t = q − 1 − ν2

(3.145)

1 R 1 l 1 (Λ2 F) x◦ + 2 (Λ2 F)y◦ + Λ4 F + 2 B l B x◦ R B y◦ A 1 1 Λ2 F − (1 + ν) L1 , F + L1 , w − w0 + Λ2 B B B 1 [L1 (w, w) − L1 (w0 , w0 )] + 2 N T Λ2 = 0, B ◦ T t = λT (T ) T z ◦ ,

ky (w − w0 ) x¯ x + k x (w − w0 )y¯y +

z

where for the difference derivatives the following notations are used

R2 n l2 n n + 2 w + , w w kl kl x¯ x x¯ x x¯ x¯yy l2 R2 kl y¯y¯yy

Λ4 wnkl =

R n l n Λ2 wnkl = , wkl + w x¯ x l R kl y¯y

L1 wnkl , Fkln = wnkl Fkln − 2 wnkl x◦ y◦ Fkln x◦ y◦ + wnkl Fkln , x¯ x

wnkl

◦

x

wnkl

=

y¯ y

x¯ x

wn − 2wnkl + wnk−1l = k+1l , h2x − wnk−1l , A = Ankl , B = Bnkl , 2h x

wnk+1l

y¯ y

x¯ x

(3.146) (3.147)

314


D = Dnkl , T = T mn , w0 = w0kl , MT = (MT )nkl , NT = (NT )nkl . The boundary and initial conditions (3.138)–(3.144) are expressed by the difference analogies of the form:

= 0, k = 0, k = N; (3.148) wnkl = wnkl = Fkln = Fkln x¯ x

wnkl = wnkl

x¯ x

= Fkln = Fkln = 0, k = 0, k = N; x¯ x & ' T mn = T 0 , m = 0, T mn z◦ = 0, m = K; & ' & '& ' λT T mn T mn z◦ = qT 0 , m = 0, T mn z◦ = 0, m = K,

w

kl = 0, wnkl ◦ = 0, n = 0; ◦

x

t

T mn = 0, n = 0; ! "

" wnkl y◦ = wnkl ◦ = Fkln y◦ = Fkln ◦ = 0. !

y¯ y y

y¯ y y

(3.149) (3.150) (3.151) (3.152) (3.153) (3.154)

Two series of the past contour points are introduced in the space whx hy (see Fig. 3.4) and two post contour points in the contour whz . The values of the mesh functions in those points are defined using the boundary conditions (3.148)–(3.151), (3.154). The obtained equations together with the boundary conditions (3.152)–(3.154) are solved using the following steps. 1. In each time step the temperature field for each point of the computational space whx hz is found, where the thermal excitation is different from zero. Further, using a numerical integration along the shell thickness by the Simpson’s method the functions Ankl are defined. Those functions, as well as the temperature field, can be found in the whole time interval independently of the solution to the motion equation. However, in order to improve the economy of the computer memory and to increase the algorithm’s effectiveness the calculations of the mentioned functions and the temperature field have been carried out using the described algorithm. 2. Using found in the previous step wnkl values and taking into account the heat functions (see section 1.1.4) the right hand side of the system of algebraic equations in relation to the function Fkln is formulated. 3. The obtained values of Fkln are substituted to the right hand side of the equation (3.145), and the wn+1 kl is obtained. Then the process is repeated. The theoretical basis of the described algorithm is given and the existence of the solution is proved (see Chapter 3.2.4).


315

3.3.2 Solution to Biharmonic Equation A fundamental difficulty while solving the equations (3.145)–(3.147) occurs during the F function estimation. The difficulty of finding a solution to the equation (3.146) is caused by a high order of the initial equation (3.155). One of the possible approaches to solve the problem is focused on the reduction of the problem to the lower order equations (possibly, of the second order). In a general case formulation and realization of the difference schemes to the second order equations is much easier [330]. The method of splitting the biharmonic operator into two second order equations has been used for the first time by Marcus in 1925 [604]. For the problems, which can not be splitted into two independent elliptical problems for the second order equations (for instance, the Dirichlet problem), the iterational method of introducing a small perturbation parameter into the boundary condition and of solving the splitted problems on each iteration, has been applied. Such an approach has been presented in the references [77, 527], and others. In this work, the boundary conditions of the form (3.148), (3.149) and (3.154) give a possibility to apply a splitting method directly. Using the biharmonic operator Λ4 and the ϕ for the other terms (the values of these terms are taken from a previous three steps) the equation (3.146) is presented in the following form Λ4 F = Λ2 (Λ2 F) = −ϕ,

(3.155)

F = 0, F x¯ x = Λ2 F = 0, k = 0, k = N,

Fy◦ = 0, Fy¯y y◦ = (Λ2 F)y◦ = 0, l = 0, l = M .

(3.156) (3.157)

We get the following equations in the splitted form Λ2 Φ =

and Λ2 F =

R2 R Φ x¯ x + Φy¯y = Λ2x Φ + Λ2y Φ = − ϕ = −ϕ1 , l l2

(3.158)

Φ = 0, k = 0, k = N,

(3.159)

Φy◦ = 0, l = 0, l = M,

(3.160)

R2 R F x¯ x + Fy¯y = Λ2x F + Λ2y F = − Φ = −Φ1 , 2 l l F = 0, k = 0, k = N, Fy◦ = 0, l = 0, l = M.

(3.161) (3.162) (3.163)

Therefore, a solution to the biharmonic equation (3.155) has been reduced to the sequent solution of equations (3.158) and (3.161). They are solved using the Fourier series approach [594]. We consider the application of this method to our problem. We consider the being sought functions Φnkl = Φ (k, l), Fkln = F (k, l) and the given function ϕnkl = ϕ (k, l) for the fixed k, 0 ≤ k ≤ N as the mesh functions with the l argument. Developing the given functions into the sums in relation to the operator Λ2y eigenfunctions, we get

316


Φ (k, l) =

M i2 =0

F (k, l) =

M i2 =0

Fi2 (k) µ(2) i2 (l),

Φi2 (k) µ(2) i2 (l),

ϕ (k, l) =

M i2 =0

ϕi2 (k) µ(2) i2 (l),

(3.164)

⎧+ ⎪ ⎪ 1 i2 πl ⎪ ⎪ ⎪ cos , i2 = 0, M ⎪ ⎪ ⎨ π M (l) = µ(2) + ⎪ i2 ⎪ ⎪ ⎪ i2 πl 2 ⎪ ⎪ ⎪ cos , 1 ≤ i2 ≤ M - 1 ⎩ π M denotes the eigenfunction of the Λ2y operator

where:

(2) (2) Λ2y µ(2) i2 + λi2 µi2 = 0, 1 ≤ l ≤ M − 1,

µ(2) i2

◦

y

= 0, l = 0, l = M,

(3.165)

which corresponds to the eigenvalue λ(2) i2 =

4 i2 π , i2 = 0, 1, ..., M. sin 2M h2y

The Fourier coefficient ϕi2 (k) for each k, 1 ≤ k ≤ N − 1 is calculated using the formulas ϕi2 (k) =

M−1

% $ (2) (2) (l) (0) (k, (M) (k, hy ϕ (k, l) µi(2) + 0.5h + ϕ M) µ . ϕ 0) µ y i i 2 2 2

l=1

Substituting (3.164) to (3.158)–(3.163) we get ϕi2 (k) =

M

ρl ϕ1 (k, l) cos

l=0

i2 πl , M

0 ≤ i2 ≤ M, 1 ≤ k ≤ N − 1, h2x l2 h2 l2 −υi2 (k − 1) + 2 + 2 υi2 (k) − υi2 (k + 1) = x 2 ϕi2 (k) , R R 1 ≤ k ≤ N, υi2 (0) = 0, 2 2 h l h2x l2 (N) (N 1 + x 2 λ(2) υ ϕi (N) , − υ − 1) = i i 2 2 2R i2 2R2 2 Φ (k, l) =

M i2 πl 2 , ρl υi2 (k) cos M i =0 M 2

0 ≤ l ≤ M, 1 ≤ k ≤ N,

(3.166)

(3.167)


ρl =

l = 0, l = M , 1≤l≤ M−1

0.5, 1,

Φi2 (k) =

M

ρl Φ1 (k, l) cos

l=0

317

(3.168)

i2 πl , M

0 ≤ i2 ≤ M, 1 ≤ k ≤ N − 1 h2x l2 (2) h2 l2 −Pi2 (k − 1) + 2 + 2 λi2 Pi2 (k) − Pi2 (k + 1) = x 2 ψi2 (k) , R R

(3.169)

a ≤ l ≤ N − 1, Pi2 (0) = 0, h2 l2 h2x l2 1 + x 2 λ(2) ψi (N) , i2 Pi2 (N) − Pi2 (N − 1) = 2R 2R2 2 0 ≤ i2 ≤ M, F (k, l) =

(3.170)

M i2 πl 2 , ρi Pi (k) cos M i =0 2 2 M 2

1 ≤ k ≤ N − 1, 0 ≤ l ≤ M.

(3.171)

In order to solve three point boundary value problems (3.167), (3.170) the multigrid method is used. The sums (3.166), (3.168), (3.169), (3.171) are obtained using the algorithm of the discrete Fourier transformations (for the nodes’ numbers, being a power of two). However, in order to have a more flexible possibility of different type node’s choice a usual summation is used. A sequential solution to the equation (3.158) and (3.161) leads to the essential reduction of the required operations. It allows to get the Φ1 (k, l) values in (3.168) as the series related to the eigenfunctions of the difference operator Λ2y . Only the components of the υi2 development are used while finding the being sought F(k, l). It should be noted [594] that during the method of one dimensional Fourier development the eigenfunctions of the difference operator Λ2x are not used. We only need to separate the variables of Λ2x , and, therefore, it can be substituted by a more generalized operator. In order to solve the equations (3.155) the method of matrices for the five points equations [594] is used. We give the matrix structure and the coefficient values used for calculations. The equation (3.155) can be presented in the following way − −u = → fh Ah→ h where

(3.172)

→ −u = {F ...F ...F ...F ...F ...F }T , h 11 M1 1i Mi 1N MN → − f h = {ϕ11 ...ϕm1 ...ϕ1i ...ϕ Mi ...ϕ1N ...ϕ MN }T .

The Ah matrix with the dimension (N × M) × (N × M) has the following form

318


(( (( A1 − A2 (( A2 (( (( A3 ( . I ( Ah = 2 2 ((( h x hy (( . (( (( (( ((

A3 A1 A2 .

A4 A2 A1 . A3 . .

A3 A2 . A2 .

A3 . A1 . A3

. A2 . A2 A3

. A3 . A1 A2 2A3

(( (( (( (( (( . . (( (( , ( . . ((( A2 A3 ((( A1 + A3 A2 (( ( 2A2 A1 (

and the matrices A1 , A2 , A3 can be presented as follows (( (( a1 2a2 2a1 (( a2 a1 + a3 a2 (( a1 a2 a3 ( a3 a2 . . . . . A1 = ((( . (( a a a a 3 2 1 2 (( a3 a2 a1 + a3 (( ( 2a3 2a2

(( (( (( (( ( . ((( , a3 ((( a2 ((( a1 (

⎛ ⎞ 2 h2 ⎜⎜⎜ R2 h2y ⎟⎟ l x a1 = 2 ⎜⎜⎝3 2 2 + 3 2 2 + 4⎟⎟⎟⎠ , l hx R hy ⎛ 2 2 ⎞ ⎜⎜⎜ l h x ⎟⎟ h2x l2 ⎜ a3 = 2 2 , a2 = −4 ⎝ 2 2 + 1⎟⎟⎠ , hy R R hy with dimension M × M,

(( (( b1 (( b2 (( ( . A2 = ((( (( . (( (( (

2b2 b1 . b2 .

b2 . b1 . b2

(( (( (( ( . (( (( , ( . ((( b2 ((( b1 (

. b2 . b1 2b2 ⎛ ⎞ ⎜⎜⎜ R2 h2y ⎟⎟ ⎜ b1 = −4 ⎜⎝ 2 2 + 1⎟⎟⎟⎠ , b2 = 2, l hx with dimension M × M,

(( (( (( C1 (( (( C2 (( (( (( . (( (( (( , . A3 = (( (( (( . (( (( C1 (( (( ( C1 (

2

C1 =

R2 hy , l2 h2x


319

with dimension M × M. In order to solve the equations system (3.158), (3.161) also the matrix method is used for the three point equations [594]. We give the matrices structure and the coefficients’ values necessary to carry out the computations in this case. The equation (3.158) and (3.161) can be presented in the following way: → − −r , Bh V h = → h

(3.173)

where the Bh matrix with the dimension (N × M) × (N × M) has the form (( (( (( B1 B2 (( (( B2 B1 B2 (( (( (( . . . . . . (( I (( (( (, B2 B1 B2 Bh = h x hy (( . . . . . . ((( (( ( B2 B1 B2 ((( (( ( 2B2 B1 ( and the matrices B1 and B2 with the dimension M × M can be presented in the form (( (( b1 (( b2 (( ( . B1 = ((( (( . (( (( (

2b2 b1 . b2 .

b2 . b1 . b2

. b2 . b1 2b2

(( (( (( ( . (( (( , ( . ((( b2 ((( b1 (

R hy l hx l hx , b2 = + , l h x R hy R hy (( (( (( d1 (( (( d1 (( (( (( . (( (( l hy (( , d1 = . B2 = (( . R hx (( (( . (( ( d1 ((( (( ( d1 (

b1 = −2

A comparison of three algorithms of the solution to the biharmonic equation (3.155) has proved that the most effective was the algorithm which used the splitting of the difference operator into two equations (3.158), (3.161) with the sequential solution to each of those problems using the method of one dimensional Fourier series’ development (see Table 3.1).

320

3 Dynamical Behaviour and Stability of Closed Cylindrical Shells Table 3.1. Efficiency of the used algorithms. Method Five point matrices Splitting + three point matrices Splitting + one dimensional Fourier series development

Estimation of the Estimation of the remembered information value artihmetic operations number ∼ 3M 2 (N + 1)

∼ 2M 3 N

∼ M 2 (N + 1)

∼ M3 N

∼ N(3M + 5)

∼ M2 N

3.3.3 Reliability of the Obtained Results On the stability of the obtained difference schemes A fundamental problem of the finite difference method accuracy is reduced to the analysis of a deviation of the approximation and stability of the used schemes. The application of the centre-difference relations for the motion equations of shells and boundary conditions leads to the reduction of the difference scheme (it has the second order approximation in relation to h x , hy , τ). The difference scheme of the heat transfer equation has the first order approximation in relation to τ and the second order approximation in relation to hz . To the most complicated analyses belong the one related to the difference scheme stability and an investigation of a continuous solution dependence on the initial input data. It should be noted that the stability corresponds to the internal scheme property, dependent neither on the approximation nor on the difference scheme link with the differential equations [591]. Therefore, the stability condition should be formulated as a certain relation between the difference operators. In the theory of the difference scheme it has been shown that approximation and stability secure the convergence of the difference scheme [572, 591]. We investigate the stability of the difference equations system (3.145)–(3.147) with the boundary conditions (3.148)–(3.154). The stability of the difference equation (1.13) can be proved using its representation in the splitted form of (3.158)– (3.163). The stability of each of the equations (3.158) and (3.161) can be proved using the maximum principle [592]. The stability of equation (3.145) is not a trivial problem because in this case the variable coefficients appear, and the equation is a nonlinear one. In order to estimate the stability condition for this case, we use the approach given in the work [572]. A local stability of the linearized equation is investigated, obtained from the initial nonlinear equations, for which a well-known theory of equations with constant coefficients is used. Because the analysed equation belongs to the parabolic type (the infinite velocity of excitation distribution), the approach described in the references


321

[572, 637] is used. According to it, the operator stability is investigated via the stability of its main part (i. e. the terms, including the higher order derivatives). Thus, we investigate the stability of the following linear equation

τM τT

2

A2 Λ4 u = f . ut¯t + D − B

(3.174)

We also investigate a stability in relation to the perturbation of the initial data ( f = 0). The stability investigation of (3.174) include two different approaches: – spectral stability; – general stability theory of the three-layer scheme. Owing to the spectral stability method [572], a solution to the difference equation (3.174) is sought in the form of harmonics dependent on two real parameters: unkl = λn (α, β) ei(αk+βl) .

(3.175)

A solution to the difference problem is stable when the spectrum is located in the unit circle |λ| ≤ 1. Substituting (3.175) to (3.174) we get the characteristic equation which leads to the stability condition [439]. In the case of h x = hy , it can be expressed in the following form ⎧ 2 ⎫ 12 √ τM ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ 3 τM 2 1 ⎬ ⎨ τT

⎪ = h. τ < min ⎪ 2 ⎪ ⎪ A ⎪ ⎪ 4 2 τT x ⎭ ⎩ D−

(3.176)

B

In order to use a general stability theory of the difference schemes [585] we reduce the equation (3.174) to the canonical form by its multiplication of τ2 : τ2

τM τT

2

A2 Λ4 u = 0. ut¯t + τ2 D − B

(3.177)

2

2 It is easy to check that B1 = 0, R = ττMT E, A1Λ = τ2 D − AB Λ4 (using the notations of the reference [591]) and the stability theorem can be applied. The stability conditions of the scheme (3.177) are represented by the following inequalities 2 τM 1 τ2 . B1 ≥ 0, R > A1Λ or Λ4 < 4 4ρ τT

(3.178)

Therefore, in order to estimate the stability we need to estimate a norm of the difference operator Λ4 . Using the results of the works [77, 681] it can be shown that for the operator norm the following estimation is valid: ⎞ ⎛

⎜⎜⎜ 16 16 ⎟⎟⎟ 2 2 (3.179) Λ4 ≤ 2∆ = 2 ∆1 + ∆2 = 2 ⎝⎜ 4 + 4 ⎠⎟ . h x hy

322


As a result, in the case of h x = hy for τ the estimation, similar to that of (3.176) is valid. The empirical formula [74] is used for a final choice of time step limitations of the nonlinear difference scheme (3.145). The time step for the nonlinear equation is recommended to be three or four times smaller than the limiting step defined from the stability condition of the linearized equation, which is well verified by the numerical experiments [181, 691]. As an example of the approach described above for the stability estimation we consider a numerical solution of the problem related to a loaded cylindrical shell ( Rl = 2.2, ky = 112.5) by a step type load q1 = 0.8 with infinite time duration and uniformly distributed along the rectangular shape α x = 0.2, αy = 0.1 (see Fig. 3.10). As the difference mesh the nodes along the longtidunal coordinate and 33 nodes along the circle coordinate are used. Taking into account the estimation (3.176) and the empirical rule, the time step in this case should not exceed 2 · 10−3 . In the Table 3.2 the results of the deflection function’s values, the load function F, the M x , My moment and the forces N x , Ny for the different time moments for the following steps: 0.00025, 0.0005, 0.001, 0.002 are given. Table 3.2. Computational results. Coordinate

Variation

W/h tmax

F tmax

Mx tmax

My tmax

Nx tmax

Ny tmax

Circle

65

14.07 0.2335

2.177 0.227

302.4 0.267

610.1 0.251

58.14 0.2445

67.24 0.261

N = 11

49

14.29 0.2335

2.229 0.237

309.2 0.2585

616.2 0.2445

57.13 0.2485

71.77 0.250

τ = 0.0005

33

14.95 0.248

2.437 0.238

357.2 0.242

562.4 0.317

56.10 0.326

73.64 0.2515

17

13.63 0.159

7.528 0.390

380.1 0.318

539.3 0.350

44.62 0.147

168.5 0.3385

longitudinal

21

14.76 0.2495

2.665 0.239

343.3 0.249

573.0 0.317

56.49 0.321

71.46 0.2385

M = 33

6

15.66 0.247

2.864 0.274

332.4 0.2395

556.1 0.243

52.90 0.2505

76.76 0.2255

Time

0.00025

14.95 0.248

2.430 0.238

357.2 0.242

562.1 0.317

56.08 0.326

73.72 0.2515

N = 11

0.001

13.63 0.159

2.430 0.238

357.2 0.242

562.1 0.317

56.08 0.326

73.25 0.251

M = 33

0.002

13.63 0.159

2.447 0.236

355.4 0.240

565.3 0.314

56.19 0.326

72.59 0.254

τ = 0.0005

The calculation results yield the 2 · 10−3 step as a suitable one from the stability point of view of the difference scheme, as well as from the point of view of achieving the required accuracy. Therefore, the above described approach for the stability estimation of the nonlinear difference equation (3.145) creates a good approximation for the time step choice of the difference scheme. The final choice of the time step has been realized using the computational experiments. A stability of the difference scheme (3.147) has been carried out in reference [593]. It has been shown that the difference scheme is stable, when the following


323

inequality is satisfied: h2z . (3.180) 2maxλT (T ) Because we consider the processes for which λT (T ) = λT0 (1 + nλ ∆T ), the maximal value of λT (T ) does not exceed 2λT0 . The estimation (3.180) serves for hz estimation, i.e. the nodes’ number along the thickness: τ≤

K
52 can be more stable against the dynamical transversal load action than the spherical ones. The spherical shells are more stable against to the qc actions than the cylindrical ones. Increasing the heat stream intensity magnitude we have observed the time instant, when for the same (K x + Ky ) the spherical shells loose their stability for qc = 0 only because of the heat impact action. The cylindrical shells (for the same q ) are still able to carry the dynamical transversal load. For instance, for the shells with the parameters (K x + Ky ) = 72 for q = 0 the critical transversal load for the spherical shells is 1.09 times higher than for the cylindrical shells. For q = 155 an equality of the corresponding q0s is observed; for q = 297 (the critical value of the heat stream intensity for the spherical shells with the heat impact) the q s = 0 for the spherical shells, whereas for the cylindrical ones it is equal to 72 non-dimensional units. For q s = 0 the critical values of the heat stream intensity of the cylindrical shells is 1.34 times higher than for the spherical shells. Analysing the stresses occurring in the shells σ0x [q , (K x + Ky )] (the full stresses in the shell’s centre on its internal surface) one can conclude that the critical stresses with the increase of the heat stream intensity suddenly increase for the spherical shells for (K x + Ky ) < 65. For the cylindrical shells for arbitrary (K x + Ky ) they change their sign (from extension for q = 0 to compression for q = q ∗ (qc = 0)) for σ0x for the case of only a heat impact action. Compare the stress critical values modulus for shells with only the heat impact (q s = 0) and the transversal load (q = 0) one can conclude that the stress in the case of the heat impact for (K x + Ky ) < 42 for spherical and for (K x + Ky ) < 30 for cylindrical shells are slightly lower than the corresponding stresses in the case of the transversal load. Increasing (K x + Ky ) the σ0x is higher in the case of the heat impact than in the case of the transversal load. The difference rapidly increases with the increase of (K x + Ky ). For instance, if for the shells with (K x + Ky ) = 36 the critical full stresses yielded the heat impact are lower than the corresponding stresses. Yielded by the transversal load of amount of 1.23 times for the spherical shells and of amount of 1.6 times for the cylindrical shells. For the shells with the parameters (K x + Ky ) = 72 those stresses are higher of amount of 5.54 times for the spherical shells, and of amount of 3.77 times for the cylindrical shells. We compare the vibrational processes in pre-critical state w(τ) on the example of the shells with the parameters K x = Ky = 36 (Fig. 4.2a), K x = Ky = 18 (Fig. 4.19b). The dashed curves correspond to the central point vibrations yielded by the heat impact, whereas the solid curves correspond to the transversal load. With the increase of the heat stream intensity the amplitude of the vibrational process and of the maximal possible pre-critical deflection decreases. Increasing the (K x + Ky ) parameters the amplitude difference increases, whereas the deflection amplitude decreases.

4.3 Stability of Thin Shallow Shells

435

Figure 4.19. Vibrations w(τ) of the shells with parameters K x = Ky = 36 (a) and K x = Ky = 18 (b).

For the shell with K x = Ky = 18 the amplitude and deflection decrease of amount of 1.72 and of amount of 1.35 times, respectively. Simultaneously, the vibrational process axis moves away from the zero value of the deflections, and the negative deflection values vanish. We consider the surface points’ movement presented on the drawing of different relative deflection curves on the example of the shells with the parameters (K x + Ky ) = 48 for the case of the heat impact (earlier graph in the Fig. 4.5) and for the case of the transversal load (Fig. 4.20) (the notations correspond to those presented in Fig. 4.5). Comparing both figures one can conclude that for only the transversal load the maximal deflection point achieves the shell’s centre. With the increase of the heat stream intensity the maximal deflection is shifted for the spherical shells to each of the shell’s quadrants. For the cylindrical shells it moves along the symmetry axis in the middle points directions of the half plate. More complicated surface form

436

4 Dynamical Behaviour and Stability of Rectangular Shells with Thermal Load

Figure 4.20. The equal relative deflection curves for the case of transversal load action.

character with the occurrence of the negative deflection zones for the case of only the transversal load (Fig. 4.20) is observed. In the case of the heat impact the above mentioned shape vanishes (Fig. 4.5). 4.3.2 Shells with Transversal Load and Heat Flow The stability loss investigations of thin shallow shells being under both heat impact and signchangeable transversal load actions have been carried out in order to detect the unstable zone, bounded by upper and lower critical frequencies ωu and ωl . If the excitation frequency lies inside the unstable zone, then the shell with the signchangeable load action loses its stability, even though its amplitude is smaller than the critical load value qc = const (see the previous section). The signchangeable load is accumed as the harmonic one. qc = qc0 sin (ωt + φ) .

(4.53)

A character of the unstable zone is found during a gradual increase of the heat stream intensity from zero to its critical value for different spherical shells with the parameters (K x + Ky ) equal to 36, 48 and 72 of the dimensionless units.


437

Figure 4.21. Dependence of the shell vibration frequency on the harmonic excitation amplitude q s0 and the heat stream density q .

Figure 4.22. The dependences σ x (ω) and w(ω) for q = 0, q s0 = 95 (a) and q = 0.75q ∗ , q s0 = 40 (b).

We consider the fundamental rules using the example of the shell K x = Ky = 24. A change of the unstable zone accompanying the increase of the heat stream ω(q , qc0 ) (Fig. 4.21) is observed. (The A area denotes the dependence of the constantly transversal load critical value versus the heat stream intensity).

438


The obtained results lead to following conclusions. Increasing the heat stream intensity, the unstable zone becomes wider and it is shifted to the side of small loads. For instance, if for q = 0 a beginning of such zone is detected for qc0 = 49, then already for q = 3/4q ∗ a contact with the plane q s0 = 0 occurs. In this plane the dependence ω(q ) is drawn. It has been observed tracing a change of upper and lower critical frequencies, that the increase of q accompanies a sudden change of ωl to ωl = 0. It means that when the heat stream intensity achieves its critical value, the unstable zone is characterized by zero lower critical frequencies and relatively large upper critical frequencies, especially for q s0 ≥ 65.

Figure 4.23. The relatively equal shell deflections for q s0 = 65 and K x = Ky = 24.

We consider a stress strain state of the shell in the transition process of the excitation frequency through the unstable zone (σ x (ω), Fig. 4.22) for q = 0, q s0 = 95 (Fig. 4.22 a) and for q = 3/4q ∗ , q s0 = 40, (Fig. 4.22 b) (σ x - solid curves; ω dashed curves). The amplitude values of the transversal loads are (approximately) taken equal to q0s for q s = const. Analyzing the obtained results one can conclude that during the transition through the unstable zone a sudden deflection and stresses increase occurs. It is clearly visible on the low boundary. It should be noted that the frequencies corresponding to the maximal values of deflection and stresses are not


439

Figure 4.24. The shell vibrations for q = 0 (a) and q = q ∗ (b).

equal. The frequency corresponding to the maximal stress is shifted to the ωl side, whereas that corresponding to the maximal deflection is shifted in to the side of ωu . Increasing the heat stream intensity the stresses change their direction: their minimal values occur but deflection increases (for both stable and unstable zones). For

440


instance, the stresses magnitude for ω = ωl is shifted of amount of 39 dimensionless units in the negative values direction, whereas the maximal magnitude σ x of amount of 45 dimensionless units in the unstable zone. A deflection for ω = ωl is increased of 1.25 times, and in the unstable zone of 1.15 times. For q = 0 a transition through the upper boundary of the unstable zone is characterized by a slight change of the stresses and deflection in comparison to the low boundary. With the increase of q

both transitions through upper and lower boundaries become similar.

Figure 4.25. Zones of unstability for K x + Ky = 36.

Tracing the motion of the surface points, equal relative deflection (Fig. 4.23) for the same amplitude transversal load values are drawn on the upper unstable zone for q = 0 and for q = q ∗ for the times instants, when the deflections achieve their maximal or minimal values. The vibrations of the surface points are characterized by large amplitudes and by the occurrence of the negative deflection in the shell’s centre. Increasing the heat flow intensity the amplitudes decrease, the negative deflections vanish, and the maximal deflection point begins to vibrate between the shell’s centre and the centres of its each quadrants. For example, in the considered case the amplitude (for q = q ∗ ) is decreased 5.77 times in comparison with q = 0, and the maximal deflection is decreased of amount of 2.09 times. The minimal deflection is shifted from the negative values to the positive ones. The described character of the surface points’ vibrations is clearly outlined in the drawing ω(τ) (Fig. 4.24 a, q = 0; Fig. 4.24 b, q = q ∗ ). It is made for the same case as the draw of equal relative deflections (Fig. 4.23). As an example, the vibration of


441

Figure 4.26. Zones of unstability for K x + Ky = 72.

Figure 4.27. Dependence of the excitation frequency ω on the heat stream density q .

the shell’s central point inside and outside the unstable zone is analysed. Stability loss does not occur on the first wave of the vibrational process. Due to the lack of the heat stream the stability loss occurs on the third wave of the vibrational process. Increasing the heat stream together with the decrease of the vibration amplitude, the period of vibration decreases, and for q = q ∗ the stability loss occurs already on the fifth wave of the vibrational process. Increasing q also the increase of the difference

442


Figure 4.28. The dependence w(ω) for the shell with parameters K x = Ky = 18 (q = 0, q s0 = 55 (a) and q = 0.75q ∗ , q s0 = 20 (b)).

between pre and post critical deflection occurs, and the vibrational axis is shifted in the positive values direction. We consider now the influence of (K x + Ky ) parameters on the shape of the unstable zones and also on the upper and lower frequencies. In the Figs. 4.25, 4.26 the unstable zones for the boundary values of the variations of (K x + Ky ) equal to 36 and 72, are presented. Increasing the parameters K x , Ky difference of the ωu values and of the width zone for q = 0 and q = q ∗ is decreased. For example, for the shell with the parameters K x = Ky = 18 (for q = q ∗ and for the constant transversal load) ωu has increased 2 times, and the zone’s width increases of amount of 3.54 times. For the shell with parameters K x = Ky = 36 the corresponding increase amounts is of 1.25 and 1.5 times. The increase of K x , Ky is accompanied by the increase of the unstable zone width on the whole change interval of the heat flow intensity from 0 to its critical value. We compare the behaviour of lower and upper frequencies with the increase of q for different parameters (K x + Ky ). For this purpose the dependencies ω(q ) for the shells K x = Ky = 18 (curve 1), K x = Ky = 24 (curve 2) and K x = Ky = 36 (curve 3) for the amplitude values of the signchangeable transversal load equal to the statical critical values, are carried out (Fig. 4.27).


443

Figure 4.29. The dependence w(ω) and σ x (ω) for the shell with parameters K x = Ky = 36 (q = 0, q s0 = 275 (a) and q = 0.75q ∗ , q s0 = 100 (b)).

Increasing (K x + Ky ), yields upper frequencies sudden increase for q = 3/4q ∗ which is then more smooth and already for (K x +Ky ) = 72 the frequency ωu becomes constant. We analyse the stress-strain state during a transition of the excitation frequency through the unstable zone when the parameters (K x + Ky ) are increased. For this aim, the dependencies σ x (ω) and w(ω) have been drawn for the boundary values of the considered interval of the (K x + Ky ) changes for the shell K x = Ky = 18 (Fig. 4.28a, q = 0, q s0 = 55; Fig. 4.28 b, q = 3/4q ∗ , q s0 = 20), and for the shell (K x + Ky ) = 36 (Fig. 4.29 a, q = 0, q s0 = 275; Fig. 4.29 b, q = 3/4q ∗ , q s0 = 100). As the figures show, the increase of (K x + Ky ) is accompanied by the increase of the stresses and deflections. The influence of the heat flow intensity is larger. We consider the influence of a phase shift on the shape of the unstable zones. The drawings ω(q s0 , q ) (Fig. 4.30) and ω(q ) (the dashed curve in Fig. 4.27) for φ = π/2 for spherical shell K x = Ky = 24 are carried out. It can be concluded that for φ = π/2 the unstable zone is larger than for φ = 0 on the whole interval of the heat flow intensity change (specially for q > 3/4q0s ). The lower and upper frequency values, used for the construction of the mentioned dependence, are given in Table 4.2 for different values of K x , Ky , for q /q ∗ = 0, 1/4, 1/2, 3/4, 1 with (Φ = π/2) and without (Φ = 0) phase shift.

444

4 Dynamical Behaviour and Stability of Rectangular Shells with Thermal Load Table 4.2. Lower (ωl ) and upper (ωu ) frequencies. φ 1

K x + Ky 2

q /q ∗ ) 3 0 0.25

36

0.5 0.75 1

0 0.25

0.5 0

48

0.75

1

0 48 0.25

q s0 4 75 55 30 75 55 30 75 55 20 75 55 20 75 55 22 5 120 95 65 120 95 65 50 120 95 65 40 120 95 65 40 20 120 95 65 40 20 5 300 275 240 190 300 275 240 190 100

ωl 5 6 10 18 0 5 13 0 0 12 0 0 5 0 0 0 0 10 13 16 1 8 13 18 0 0 6 13 0 0 0 3 11 0 0 0 0 0 0 8 8 13 20 5 8 12 18 23

ωu 6 25 23 18 25 23 20 25 25 16 28 23 17 48 46 25 15 32 30 22 32 30 24 21 27 26 25 21 27 27 27 23 20 53 51 50 36 27 20 48 48 43 40 48 46 43 40 33

4.3 Stability of Thin Shallow Shells 1

2

3 0.5

0.75 48

1 π/2 0

0.25

0.5 48

0.75

1

4 300 275 240 190 100 50 300 275 240 190 100 50 300 275 240 190 100 50 5 120 95 65 50 120 95 65 50 120 95 65 40 120 95 65 40 20 120 95 65 40 20 5

5 0 1 3 5 18 26 0 0 0 0 8 15 0 0 0 0 0 0 0 0 8 15 19 0 0 13 15 0 0 5 12 0 0 0 5 13 0 0 0 0 0 0

6 53 50 43 33 28 25 47 46 45 43 30 20 63 60 50 48 33 28 23 30 27 23 22 31 27 23 21 30 27 25 23 33 31 26 21 16 103 81 58 42 28 20

445

446


Figure 4.30. The dependence ω(q s0 , q ) for the spherical shell.

4.3.3 Influence of Thermal and Mechanical Characteristics We consider the stability of thin shallow shells, rectangular in plane, rollingly supported on their edges (see boundary condition (4.28)), with a heat impact and including the temperature dependence of E and αt . We investigate the latter influence of a dynamical load on the critical values, exhibited by the heat flow intensity, and the corresponding deflections and stresses for a series of cylindrical and spherical shells. The Young’s modulus and the linear expansion coefficient are the temperature functions of the form: (4.54) E = E 0 KE , α = α0t Kα ,

(4.55)

where E 0 and α0t correspond to zero temperature, and KE , Kα are the non-dimensional temperature functions. They are obtained by E and αt n-th order polynomial approximation for each material. The shell temperature is defined via the formula (4.44). The series of the expression (4.44) is converged. The computations are prolonged to reach the condition |an / an−1 | ≤ 10−8 . In the initial time (τ = 0) 397 series terms are taken. However, already for τ = 0.01, only 19 terms are needed and the terms number is decreased with increase of time τ. The fundamental equations (4.25) and (4.26) accounting (4.44), have the following form 2 4 4 ∂4 w ∂ w ∂w −2 ∂ w 2∂ w 2 = q s , (4.56) − ∇k F − L (w, F) + κ +2 2 2+λ +ξ Dt λ ∂τ ∂x4 ∂x ∂y ∂y4 ∂τ2 1 ∂4 F ∂4 F ∂4 F 1 (4.57) λ−2 4 + 2 2 2 + λ2 4 + ∇2k w + L (w, w) = 0, Ft 2 ∂x ∂x ∂y ∂y


where 1 Dt = & ' 1 − ν2

h2

h2 KE (T ) z dz; Ft = 2

−h1

1 1 + δ; h2 = − δ; δ = 2 2

KE (T ) dz; −h1

1

h1 =

447

2 1 −2

⎤−1 ⎡ 1 ⎥⎥⎥ ⎢⎢⎢ 2 ⎥⎥⎥ ⎢⎢⎢ ⎥ ⎢⎢⎢ (T ) KE zdz ⎢⎢ KE (T )dz⎥⎥⎥⎥ . ⎥⎥⎥ ⎢⎢⎢ ⎦ ⎣1

(4.58)

−2

The numerical integration of (4.58) is carried out using the Simpson’s method. In order to get an optimal shell’s thickness partition 10, 20 and 40 intervals are used on the basis of the Runge’s principle. It appears that during the partition into 20 and more parts the thermal terms obtained using the Simpson’s method and the exact integration method (where E, αt are constant) practically overlap. Besides, the thermal terms, obtained during the partition into 20 and 40 intervals overlap as well, whereas during the partition into 10 and 20 intervals an error achieved 5-6%. Taking into account the earlier results we have used 20 intervals. We analyse the stability of the squared shells with the following physical mechanical characteristics (steel 15XM): ν = 0.3, E 0 = 2.1 · 105 MPa, α0t = 1.15 · 10−5 grad−1 , c = 460J/kg · K, λ0g = 476m/(n · K). The dimensionless functions Kα and KE are approximated using the experimental data via the following relations

(4.59) KE = 1 − a1 T 2 − a2 T , πT Kα = 1 − b1 T + sin , for T ≤ T 1 ; Kα = [1 + b4 T ] , for T ≤ T 1 , T1

(4.60)

where a1 , a2 , b1 , b2 , b3 , b4 are the non-dimensional approximation coefficients. They have the following dimensional form a1 =

a2 b2 α2t0 abαt0 a1 = 0.2976 · 10−6 grad−2 , a2 = a2 = 0.4643 · 10−3 grad−2 , 4 h h2

b1 = 4.4269 · 10−4 grad−1 ; b2 = 0.03478; b3 = 1.2435; b4 = 1.3986grad−1 . T 1 denotes a non-dimensional value of the temperature (823◦ K), above which the αt can be approximated by a straight line. We have investigated a dynamical stability of thin shallow shells in order to determine the critical values of the heat flow intensity q ∗ with a simultaneous temperature dependence of E and αt . The q ∗ values have been compared with the data obtained for E, αt = constant (see Sect. 4.2.3). The comparison results in percents in relation to q ∗ , w, σ x for E, αt const are given on the drawings for a critical value

448


Figure 4.31. The critical heat stream q ∗ values in percents versus K x + Ky .

of the heat stream intensity (Fig. 4.31), the deflection (Fig. 4.32) and full stresses in the shell’s centre on its internal surface (Fig. 4.33). The solid curves of E, αt simultaneously depend on the temperature; the dashed ones - on the E(T ); αt = const, the dashed lines with dots correspond to αt (T ), E = const; digit 1 corresponds to the spherical shells and digit 2 corresponds to the cylindrical shells. Analysing the results for q ∗ presented in Fig. 4.31 one can conclude that when αt and E do not depend on the temperature, the values of the critical heat flow intensity


449

Figure 4.32. The shell deflection in percents versus K x + Ky .

become greater. Increasing the parameters (K x + Ky ) a difference of q ∗ is increased. This behaviour can be explained in the following way. When the (K x + Ky ) are increased, then also q ∗ increases, and the shell’s temperature increases, and therefore the temperature strongly influences αt and E. This influence depends not only on (K x + Ky ), but also on the shells type. For the spherical shells the q ∗ variations are larger than for the cylindrical shells. For the cylindrical panels and for (K x + Ky ) > 50, a stabilization of q ∗ is observed.

450


Figure 4.33. The shell stresses in percents versus K x + Ky .

The maximal increase of q ∗ for (K x + Ky ) = 72 achieved 17.5% for cylindrical and 22.5% for spherical shells. However, their distribution is different. More important is αt (T ). As it is seen from Fig. 4.31, the critical values of the heat stream intensity for E(T ) and αt (T ), E = const and αt (T ) are close to each other. The maximal difference does not exceed 3%. Therefore, the investigations can be carried out without any limitation either along the thickness, or along the surface. In the latter case (when E = const) the integral characteristics Dt , Ft are constant, and S t = 0 in (4.6) and the equation (4.25) and (4.26) are reduced to those with constant coefficients. We compare the results corresponding to q ∗ . As it is shown in Fig. 4.32, the deflections change when E and αt is constant, depend on the shell’s type and on the parameters K x , Ky . When E and αt do not simultaneously depend on the temperature the deflection increase for the spherical shells for (K x + Ky ) > 60, and for the


451

cylindrical panels for (K x + Ky ) > 37. When only αt (T ) is taken into account, the deflection increase is observed. The highest increase of deflections (up to 19%) is observed, when αt has been treated as temperature independent. Now, we analyse the stresses corresponding to the critical values of the heat stream intensity (Fig. 4.33). We consider only the cases of the stresses increase. When both E and αt are treated as temperature independent, the stresses step up for the spherical shells for (K x + Ky ) > 44 and for the cylindrical panels on the whole interval of the (K x + Ky ) changes. When E is treated as temperature independent the stresses are even greater. When only αt is treated as temperature independent, the stresses increase their amount only for the spherical shells for (K x + Ky ) > 56 and the stresses difference decreases. The largest height increase (of 47%) is observed, when E is treated as the temperature independent. The numerical data of the found critical heat stream intensity, deflections and stresses with and without temperature dependence of E and αt , and also the differences in percents in relation to q ∗ , w and σ x are given in Table 4.3.

Figure 4.34. Equal relative shell deflections for cylindrical (a, b) and spherical (c, d) shells.

452

4 Dynamical Behaviour and Stability of Rectangular Shells with Thermal Load Table 4.3. Numerical data of spherical and cylindrical shells. K x + Ky Critical E, αt values - const

36

48

60

72

36

48

60

72

q ∗ w −σ0x q ∗ w −σ0x q ∗ w −σ0x q ∗ w −σ0x

69 2.262 18.698 133 2.696 54.206 216 2.776 87.843 297 2.857 121.48

q ∗ w −σ0x q ∗ w −σ0x q ∗ w −σ0x q ∗ w −σ0x

64 2.225 16.07 131 2.593 45.595 263 3.012 105.025 400 3.121 164.1

E(T ) E, αt % - const αt (T ) Spherical shells 62 -10.6 66 2.42 +7 2.34 22.157 18.5 16.35 109 -18 131 2.857 6 2.82 49.015 -9.6 49.746 230 -20.9 219 2.776 0 2.854 67.2 -23.5 78.883 230 -22.6 314 2.676 -6.3 2.88 83.18 -31.53 104.28 Cylindrical shells 61 -4.7 65 2.234 0.4 2.66 16.006 -0.4 11.39 108 -17.5 129 2.52 -2.8 2.76 45.49 -1 41.778 217 -17.5 263 2.78 -7.7 3.169 89.271 -15 86.121 330 -17.5 419 2.728 -12.6 3.27 126.165 -23.1 121.032

%

E(T ) αt (T )

%

-4.9 3.5 -12.6 -1.5 4.6 -8.2 2 2.8 -10.2 5.7 0.81 -14.2

63 2.24 19.855 109 2.57 -56.423 167 2.684 90.039 223 2.835 122.27

-8.7 -0.97 6.2 -18 -4.67 4.1 -22.3 -3.3 2.5 -24.9 -0.77 0.65

1.5 62 -3.1 19.5 2.476 11.3 -29.5 15.292 -4.8 -1.5 109 -16.8 6.4 2.594 0.04 50.6 50.87 10.7 0. 214 -18.6 5.2 2.819 -6.4 -18 113.952 8.5 4.8 322 -19.5 4.8 2.73 -12.5 -26.4 172.41 5.1

Now we analyse the influence of E(T ) and αt (T ) on the full stresses, and independently for the membrane and bending stresses. The influence of the mentioned characteristics is different and changes with the increase of (K x + Ky ), but a qualitative change of the membrane, bending and full stresses is similar. The maximal increase of all components of the stress state is observed at the end of the interval (K x + Ky ) = 72 for the spherical shells. In the case of bending normal stresses their amount achieves 31%, in the case of bending tangential stresses their amount achieves 23%, in the case of the normal membrane stresses their amount achieves 52%, and in the case of the membrane tangential stresses their amount achieves 21%. We investigate, how the temperature dependences of E(T ) and αt (T ) influence the vibrational process of the shell surface points. We consider the curves of equal relative deflections (Fig. 4.34) and the w(τ) dependence (Fig. 4.35) obtained for shells with (K x + Ky ) = 48 (Fig. 4.34 a, b - cylindrical shells; Fig. 4.35 a - cylindrical


453

Figure 4.35. Vibrations of cylindrical (a) and spherical (b) shells.

shells, b - spherical shells). We compare the obtained characteristics with those obtained when E and αt are treated as temperature independent (Fig. 4.3, 4.5). It appears that a treatment of E and αt as temperature independent does not lead to a sufficient influence of the shell surface form.

454


Figure 4.36. Stresses and deflection variations versus the q parameter.

However, when both E(T ) and αt (T ) are considered as temperature independent, the relative maximum and amplitudes of vibrations are decreased. For example, for the results presented in Fig. 4.34 and Fig. 4.5 a difference in the values of relative maximum and amplitudes (in relation to E, αt - const) achieved amount of 14.8% and amount of 19.8% for the spherical shells, and 70% and 64.5% for the cylindrical shells. Comparing the vibrational process on the example of a central point (w(τ) dependence) (Fig. 4.3 and Fig. 4.34) it is clear that when E and αt are temperature dependent, then the stability loss occurs on the second wave of the vibrational process. For E, αt = const for the post-critical values of the heat flow intensity a shell loses its stability beginning from the first wave of the vibrational process. A stability loss of shells with E(T ) and αt (T ) beginning from the second wave of the vibrational process can be explained by means of a continuous heating and temperature increase in time. As a result, we have got a maximal deviation of E and αt from its initial state only on the second wave of the vibrational process. Consider the influence of temperature dependence of E and αt on the stability of flexible conical shells in the condition of a constant load and a heat impact. The investigations have been focused on the critical values of the transversal load with an increase of the heat flow intensity from zero to its critical value. The results obtained for the same q are compared with those given in Sect. 4.3.2 of this chapter for E, αt - const. The analysis has been carried out on the example of the spherical shell with (K x + Ky ) = 24.


455

In the Fig. 4.36 a change of q0s versus the increase of q is presented (the solid curves - E and αt are const; the dashed curves - E and αt depend on the temperature). When E and αt are treated as temperature independent, it leads to the increase of the critical transversal load values for the same q and the difference of q0s is increased, when the heat stream intensity increases. It is mainly caused by the shell temperature increase (it means that the influence of E(T ) and αt (T ) increases), and the critical value of the heat flow intensity is lower, when E and αt depend on the temperature. Therefore, if for q = 109 non-dimensional units (q = q ∗ for E(T ) and αt (T )) a stability (with the influence of E and αt on the temperature dependence) is characterized only by the heat impact (q0s = 0). For E and αt - const, the shell can still carry the transversal load (q0s = 26). The obtained results show that for q ≤ 0.25q ∗ (q ∗ for E and αt - const) the E and αt temperature dependence practically does not influence the critical values of the transversal load. With sufficient practical accuracy (up to 5%), the q0s for E and αt - const can be found in the interval 0 ≤ q ≤ 60 (q 0.45q ∗ ). Now we analyse the deflections and stresses. When E and αt are temperature independent for arbitrary q it leads to the deflection decrease and to the stresses increase for q < 60. In the interval 20 < q < 37 the stresses have different signs. Maximal increase of stresses up to 150% and the deflection decrease up to 33% are observed. For the critical value of q the absolute increase of stresses is small and does not play an essential role in the strain state during stability investigation. The numerically obtained critical values of the transversal load, deflection and stresses (with and without temperature dependence of E and αt ), and also the shift expressed in percents in the mentioned quantities in relation to q , w, σ0x for E and αt = const are presented in Table 4.4. Table 4.4. Critical values of transversal loads

E, αt const E(T ), αt (T ) % Critical values

0.25 78 2.3 -9.333 55 78 2.93 4.64 50 0 33.20 149.7 -9.9 q0c w σ0x q0c

q /q ∗ 0.5 0.75 2.28 -25.54 36 2.50 -39.90 2.71 -25.78 12 2.69 -45.69 18.9 5.1 -66.7 7.60 -14.51 w σ0x q0c w σ0x

We consider the temperature influence of E and αt on the stability of flexurable shallow shells being under the influence of a sign changeable transversal load and a heat impact. The investigations have been carried out in order to find an unstable zone on the example of the spherical shell K x = Ky = 24 with the amplitude of the transversal load equal to 65 non-dimensional units. The results have been compared for the same q with those given in Sect. 4.3.3 for E and αt = const. In Fig. 4.37 the unstable zones with the increase of q from 0 to q ∗ (q ∗ for E and αt = const) (the solid curves denote constant values of E and αt , whereas the dashed correspond to temperature dependent E(T ) and αt (T )). When E and αt depend on

456


Figure 4.37. The unstable zones in the coordinates w(q ).

the temperature, then the unstable zone increases. However, up to q = 3/4q ∗ , that influence is rather not significant. The maximal difference in width has achieved 22% for q = 0.6q ∗ , and 11% for q = 0.75q ∗ . The absolute values of differences of upper and lower frequencies do not exceed 3 non-dimensional units. With sufficient accuracy (up to 5%) the frequencies of the parametric vibrations for q ≤ 0.3q ∗ can be found. For q ≥ 3/4q ∗ for E and αt = const a sudden jump of high frequencies (1.85 times) is observed. For q = q ∗ and taking into account the temperature dependence of E and αt the ωg achieved its limit in infinity. It can be explained in the following manner. The q = q ∗ with E and αt = const belongs to the post-critical one (of 18%) in comparison to the critical value. Therefore, the upper frequency unboundedly increases, because for this case the stability is characterized only by the heat impact (the thermal load is post-critical, and the shell loses its stability independently of the forcing load values). The numerical calculation results are given in Table 4.5. In the earlier discussion and illustrated problems the investigations have been carried out when only the linear thermal expansion coefficient and the Young’s modulus depend on the temperature. We have to mention that the temperature dependence on other characteristics, such as the heat and the temperature transfer coefficient have not been considered, and a solution to the heat transfer equation is obtained for λg = const and α = const.


457

Table 4.5. Results of numerical computation of ω and q /q ∗ . q s0

ω

E, αt const E(T ) αt (T ) change %

ωu ωl ωu ωl ωu ωl

0 16 22 16 22 0 0

0.25 13 24 13 25 0 4.2

q /q ∗ 0.5 6 25 4 27 33 8

0.75 0 27 0 30 0 11

1 0 50 0 0

Now we consider thin conical shells stability taking into account the temperature dependence of all material characteristics as well as the Young’s modulus. The λg and α coefficients for a given material are approximated by the following expressions:

λg = λng 1 − AT 2 − BT ;

α = α0 1 − CT 2 − DT ; where: 2 abαt0 abαt0 −7 −2 ¯ ¯ A=A = 0.7 · 10−3 grad−1 ; = 0.8 · 10 grad ; B = B h2 h2 abαt0 abαt0 −5 −2 ¯ ¯ C=C = 0.11 · 10 grad ; D = D = 0.6 · 10−3 grad−1 . h2 h2 When λg and α depend on the temperature, the heat transfer equations have variable coefficient and it is impossible to find an analytical expression for the temperature field. Therefore, the numerical method of finite differences with the approximation O(h2 ) along the shell’s thickness is used. It allowed to reduce the partial differential equations to the system of ordinary differential equations which has been solved using the Runge-Kutta method. The shell thickness has been divided into twenty intervals. As a result of the numerical solution to the heat transfer equation the temperature field T (z, τ) is obtained. Knowing the temperatures, the integral characteristics Ft , Dt are found and the temperature terms on each time step necessary for solution of the equations (4.55) and (4.57) are defined. The calculations have shown that the error occurred when λg and α do not depend on the temperature (for the same q ) achieved 8%, and for accounting temperature variation it achieved amount of temperature 6%. To conclude, an account of λg and α on the temperature increases a difference in the critical values of the load, deflection and stresses. For the spherical shell with the parameters K x = Ky = 24 an account of the temperature dependence of λg and α caused a decrease the heat stream intensity critical value of amount of 5.5% and the stresses of amount of 4.8%, and yields increase of deflection of amount of 2.56% in comparison to the case, when E and αt do not depend on the temperature.

458


4.4 Stability of Thin Conical Shells 4.4.1 Boundary Conditions and Surrounding Medium We investigate a stability of flexurable conical shells subjected to a longitudinal compressing load (constant in time) action on the shell’s frontal surface with the following boundary conditions: – The boundary condition 1 - (4.28), – The boundary condition 2 - (4.29), – The boundary condition 3 - (4.30). We consider the case when E and αt do not depend on the temperature. The fundamental equations (4.29) and (4.26) have the following non-dimensional form: 4 4 1 ∂4 w −2 ∂ w 2∂ w − ∇2k F − L (w, F) + +2 2 2 +λ & ' λ 12 1 − ν2 ∂x4 ∂x ∂y ∂y4 2 ∂ w ∂2 w ∂2 w ∂w = − K x P x + Ky Py + P x 2 + Py 2 , κ (4.61) +ξ ∂τ ∂τ2 ∂x ∂y λ−2

4 2 2 ∂4 F 1 ∂4 F 2∂ F 2 2 ∂ Px −2 ∂ P x (w, L w) = λ + 2 + λ + ∇ w + + λ . k 2 ∂x4 ∂x2 ∂y2 ∂y4 ∂y2 ∂x2

(4.62)

As dynamical stability loss criteria, the Volmir’s criterion and the criterion of the membrane stresses sudden are used. The investigations are carried out for a series of cylindrical shells with the parameters (K x + Ky ) [36; 72] with the longitudinal compressing load action along the direction of zero curvature and a lack of the thermal field. The longitudinal compressing load in the nonzero curvuture is assumed to be equal to zero. The results of the dependence P0x (K x + Ky ) are presented in Fig. 4.38a. The digits correspond to the boundary problem value, the solid curves correspond to damping coefficient ξ = 0. It has been found that the boundary conditions have essential influence on the changes of the longitudinal compression load critical value, and this influence essentially the variations of the longitudinal compression load changes with a change of (K x + Ky ). Increasing the parameters (K x + Ky ) the P0x values increase. The higher influence corresponds for the 1 boundary conditions, whereas the lowest to the 3 boundary conditions. Comparing the obtained values P0x for different boundary conditions, it has been observed that strongly stable shells correspond to the 2 boundary conditions for (K x + Ky ) < 66, whereas for (K x + Ky ) > 66 the most strongly stable shells correspond to the 1 boundary conditions. In the case of the hybrid boundary conditions 3 the shells are more stable than those with the 1 boundary conditions only if (K x + Ky ) < 58. For instance, for (K x + Ky ) = 36 the P0x values for the shells with the boundary conditions 2 and 3 are equal or greater than the P0x for shells with 1 boundary conditions of amount of 1.33 times.

4.4 Stability of Thin Conical Shells

459

Figure 4.38. The longitudinal compressing load P0x (a) and the stresses σ0y (b) versus the parameter K x + Ky , and the dependence P0x (ε) (c).

Now we consider the motion of the surface points. For this purpose, the curves with equal relative deformations (Fig. 4.39) have been determined in the time instants, when the deflection in the centre reaches its first maximum (the upper curves) and its first minimum (the lower curves) for different boundary conditions (1, Fig. 4.39; 3, Fig. 4.38 b; 2, Fig. 4.39 c). It has been discovered that the most complicated vibrations have occurred for shells with hybrid boundary conditions 3. For the time instants, when the deflection reaches its minimum, zones with negative deflections on the surface occur. The greatest relative deflection is observed for shells with the boundary conditions 2.

460


Figure 4.39. The curves with relatively equal deflections for boundary conditions 1 (a); 2 (b); 3 (c) for K x = 0 and Ky = 48.

The deflection distribution along the coordinate axes is more clearly visible in Fig. 4.40 and Fig. 4.41. (Fig. 4.40 a, 4.41 a correspond to the shells with the boundary conditions 1; Fig. 4.41b - 2; Fig. 4.41c - 3; the digit 1 corresponds to b(a) = 1/2; the digit 2 corresponds to b(a) = 1/4; the solid curves correspond to the maximal deflection in the centre; and the dashed curves corresponds to the minimal one). The figures show that the deflection distribution along the shell (direction a) with the boundary conditions 1 and 2 are characterized by one halfwave for the arbitrary time moments. For the shells with 3rd boundary conditions, in the time instants when the deflection reaches its minimum, the deflection distribution is characterized by five halfwaves. In the case of a transversal deflection distribution (along side b) the deflection is characterized by three halfwaves for the 2 and 3 boundary conditions. We consider the vibrational process on the example of the shell’s central point with the parameters K x = 0, Ky = 48 (Fig. 4.42) for different boundary conditions (Fig. 4.42a - 2; Fig. 4.42b - 3; Fig. 4.42c - 1). The vibrational process is characterized by relatively small deflections and amplitudes in comparison to the transversal load. The greatest deflections have been observed for shells with the boundary conditions 1, and the smallest for with the boundary conditions 2. The amplitude has achieved


461

Figure 4.40. Shell deflections along a direction.

its maximal value for the shells with boundary conditions 2. The vibrational process for shells with the boundary conditions 1 is characterized by large vibration periods and larger beginning of the vibration process in comparison to boundary conditions 2.

462


Figure 4.41. Shell deflections along b direction.

We investigate the influence of damping on the critical values of the longitudinal compressing load. For this purpose the dependence P0x (K x + Ky ) for ξ = ξ∗ (Fig. 4.38a - the dashed curves) is applied. ξ∗ denotes minimal value of ξ for which P0x monotonously approaches the statical value.


463

Figure 4.42. Vibrations of shell centre for different boundary conditions: 2 (a), 3 (b) and 1 (c).

In what follows damping essentially influences P0x (especially the shells with the 2nd boundary conditions). P0x increases with the increase of (K x + Ky ). The statical load values P0x is achieved only for relatively large ξ values (Fig. 4.38c). For the shell with K x = 0, Ky = 72 the statical load is higher than the dynamical one for

464


shells with the boundary conditions 2 of amount of 1.35 times (ξ∗ = 35) and for 3 of amount of 1.22 times (ξ∗ = 50). For instance, when only the transversal load is applied (2 boundary condition) the statical value q0s is higher than the dynamical one of amount of 1.23 times (ξ∗ = 16) for the same shell. We investigate a strain state of shells for two boundary conditions with and without damping expressed by the dependence σ0y (K x + Ky ), Fig. 4.38b. The stresses appearing in the shells with the 1st boundary conditions suddenly increase when (K x + Ky ) is increased. In the case of hybrid boundary conditions 3 the stresses decrease up to (K x + Ky ) ≤ 48, but next their sudden increase occurs. However, they always remain smaller than the stresses in shells with the boundary conditions 2. For instance, σ0y for shells with 2 boundary conditions (ξ = 0) and for the parameters (K x + Ky ) = 72, the stresses for increased of amount of 1.97 times in comparison to σ0y in shells for the parameters (K x + Ky ) = 36. The stresses σ0y in the shells with the boundary conditions 3 first decreased of amount of 1.35 times in the interval (K x + Ky ) ≤ 48, and then they increased 3.3 times. The damping of a surrounding medium decreases the stresses, especially for the shells with 3rd boundary conditions. 4.4.2 Constant Compressing Load and Heat Flow The influence of the heat flow intensity on the dynamical critical longitudinal compressing load is considered. The calculations are carried out for the cylindrical shells with the parameters 36 ≤ (K x + Ky ) ≤ 72 in order to find P0x with the increase of q from zero to its critical value for the corresponding (K x + Ky ). The cylindrical shells are subjected to the longitudinal compressing load along zero curvature direction with the boundary conditions (4.53) (see dependence P0x [q , (K x + Ky )] presented in Fig.4.43). The dynamical longitudinal compressing loads lying below the given zone of the critical P0x values are pre-critical ones. When the critical q values are achieved, a stability is characterized only by the heat stream action (P0x = 0). It has been found that the heat stream intensity has essential influence on the critical value of the longitudinal load intensity, and it increases when q is increased. Comparing the obtained data with those given in Sect. 4.4.2 of this chapter (Fig. 4.18), i.e. considering the case when only transversal dynamical load is applied, one can conclude that a critical longitudinal load is more sensitive to the heat flow intensity action than q s . It is clearly seen in Table 4.6 on the example of two cylindrical shells with the parameters K x = 0, Ky = 48 and K x = 0, Ky = 72. Increase of q implies its stronger influence on the critical compressing load, especially for higher values of the parameters K x and Ky . Note, that for that case with only transversal load action, the heat stream intensity has smaller influence on q0s . The heat stream intensity has also larger influence (during the compressing longitudinal load action) on the maximal shell deflections. Increasing q a sudden deflection increase is observed. Influence of q is even more evident with the increase


465

Figure 4.43. The dependence P0x [q , (K x + Ky )]. Table 4.6. Kx = 0

Ky = 0

Kx = 0

Ky = 0

q

q ∗

P0x (q ) P0x (q =0)

q0s (q ) q0s (q =0)

P0x (q ) P0x (q =0)

q0s (q ) q0s (q =0)

0

1

1

1

1

0.25

0.889

0.790

0.643

0.701

0.50

0.389

0.530

0.286

0.585

0.75

0.167

0.168

0.095

0.285

1

0

0

0

0

of K x and Ky . When only the transversal load is applied, the deflections decrease. For instance, for the shell with K x = 0 and Ky = 36, the deflections increased (owing to increase of q ) from zero to its critical value of amount of 1.5 times for the compressing load, whereas for the case of a transversal load action, they it has been decreased of amount of 1.4 times. For the shell with parameters K x = 0 and Ky = 72, the deflections are changed for the considered loading cases of amount of 5.12 and 1.2 times, respectively. It again testifies the observation that the shells with the longitudinal compressing load

466


are more sensitive to the heat stream than those subjected to the transversal load action. We now analyse the surface points vibrations and the vibrational processes owing to with the heat stream intensity increase from zero to its critical value. The curves of equal deflections are compared (see Fig. 4.39c, Fig. 4.5a, Fig. 4.42c and Fig. 4.3a). The investigations show that increasing the heat intensity, the amplitudes and deflections of the vibrational process increase, and the relative maximum and the time of vibration beginning decrease. The relative maximum is shifted from the quadrants’ centres to the symmetry axis, which is perpendicular to zero curvature direction. For the considered shell, the amplitude and deflections increased of amount of 1.32 and 2.67 times, correspondingly (for q = q ∗ in relation to q = 0). 4.4.3 Harmonic Longitudinal Load and Heat Flow The investigation of flexurable conical shells, subjected to both heat stream and harmonic longitudinal load is carried out in order to find an unstable zone bounded by low and upper critical frequencies ωl , ωu . The applied load has the following harmonic form (4.63) P = P0 sin ωt, where P0 denotes the load amplitude.

Figure 4.44. The dependence ω[q , (P x0 + Py0 )] for the cylindrical shell.

A shape of an unstable zone is found owing to a gradual increase of the heat stream intensity from zero to its critical value. In order to compare the results with those given in Sect. 4.4.2, the same shells with (K x + Ky ) = 48 have been analysed.


467

Figure 4.45. The dependence ω[q , (P x0 + Py0 )] for the spherical shell.

The fundamental rules of the unstable zone variations for a cylindrical shell under action of harmonic longitudinal load action in zero curvature direction P x = P x0 sin ωt, Py = 0 is studied (see the dependence ω[q , (P x0 + Py=0 )] shown in Fig. 4.44). Recall that the A area in the figures governs the dependence of the critical value of the longitudinal load against the heat stream intensity. It is observed that increasing the heat stream intensity the unstable zone becomes wider and is slightly shifted to the direction of small amplitude values of the load. However, contrary to the case corresponding to transversal load action, the zone’s spike does not touch the plane (P x0 + Py0 ) = 0. Comparing the frequencies for q ≤ 3/4q ∗ for the harmonic longitudinal and the transversal loads (Fig. 4.44 and Fig. 4.21) one may conclude that their change character is similar, when the heat flow intensity increases up to 3/4q ∗ value. However, in the case of the heat flow intensity larger than 3/4 of its critical value, higher shells sensitivity is observed for the longitudinal load and the heat stream. For q > 3/4q ∗ an essential increase of the upper frequencies is observed. For instance, for the transversal load ωl is increased for q = q ∗ of amount of 2.27 times in comparison to q = 0. The upper frequency is increased of amount of 30 times for the longitudinal load, and became 11.76 times larger than ωu in comparison to the transversal load action. The longitudinal load action is expressed by an essentially larger increase of the upper frequency in comparison to the transversal load action. The qualitative and quantitative changes of the unstable zone essentially changes with the increase of q , when the spherical shell stability subjected to harmonic longitudinal load P x0 = Py0 = P0 action is considered (Fig. 4.45). The zone width

468


and the upper critical resonance frequencies are larger, even with a lack of the heat stream action. For instance, for (P x0 + Py0 ) = 12 and q = 0 both zone width and upper frequencies are higher for the spherical shell 3 of amount of 3.14 and 2.5 times in comparison to the cylindrical one. It has been found that the spherical shells subjected to the longitudinal harmonic load P x0 = Py0 = P0 action are very sensitive to the heat stream variation, and already for q > 1/2q ∗ a sudden increase of the upper critical frequencies and the zone’s width are observed.

Figure 4.46. The dependencies ω(q ) for the cylindrical (dashed curves) and spherical (solid curves) for various 1-5 loads.

For instance, for (P x0 + Py0 ) = 10 a value of the upper critical frequency for q > 3/4q ∗ is larger than for q = 0 of amount of 2.5 times, and of amount of 5 times for the cylindrical shell. For q = q ∗ the frequency ωu becomes 16 times larger than for q = 0, and 1.33 times larger than for the cylindrical shell.


469

The investigation yields the conclusion that the load acting on the spherical shell possesses more negative influence, as well as the heat load intensity is more strong than the load acting on the cylindrical shell with the same (K x + Ky ) and (P x0 + Py0 ) parameters. In order to detect the influence of a way of the load application (in order to define the most unprofitable case), additional investigations of cylindrical and spherical shells stability are carried out using the forcing amplitude (P x0 + Py0 ) = 10. The obtained results are presented in Fig. 4.46. The following loading cases are considered. For the cylindrical shells K x = 0, Ky = 48 (the dashed curves): 1. P x0 = P0 = 10, Py = 0, 2. P x = 0, Py0 = P0 = 10, 3. P x0 = Py0 = P0 = 5. For the spherical shells K x = Ky = 24 (the solid curves): 4. P x0 = Py0 = P0 = 5, 5. P x0 = P0 = 10, Py0 = 0. Cylindrical and spherical shells are separately considered. For cylindrical shells the most unprofitable loading corresponds to the harmonic longitudinal load action only along the nonzero curvuture direction. On the contrary, the longitudinal load action along the zero curvuture the unstable zone area.

Figure 4.47. Shell vibrations (loading type 1).

470



A way of the load action practically does not influence the lower critical frequencies. However, the upper frequencies (and correspondingly, the zones width) strongly depend on the load action way. The increase of the heat stream intensity exerts larger influence during the longitudinal load action along the nonzero curvuture direction. However, for q = q ∗ a difference in the upper frequency values is not high, the upper frequencies reach large values for all types of loads. The highly expressed increase of ωu for q > 3/4q ∗ corresponds for the longitudinal load action along the direction of zero curvature action. We illustrate the above observations on the numerical examples. For q = 0 the ωu value is higher of amount of 1.48 and 3.7 times for the 2 loading case, than the ωn values corresponding to 3 and 1 loading cases correspondingly. For q > 3/4q ∗ those relations achieved amount of 1.9 and 8.25, whereas for

q = q ∗ the achieve amounts of 1.02 and 1.12. The upper frequencies for q = q ∗ are increased in comparison to q = 0 and

q > 3/4q ∗ for the 2-nd loading type of 9.05 and 3.38 times; for 3-rd loading type of 13.2 and 6.35 times and for the 1-st loading type of 30 and 25 times. For spherical shells the loading type influence is different owing to increase of the stream intensity. For the q < 98 ( 3/4q ∗ ) the values of upper and lower critical frequencies are higher (the longitudinal load acts only in one direction). Besides, in the considered interval of q for the 5-th loading type the frequency ωl = 0 only at the interval end. In the case for the 4-th loading type, ωl = 0 for q = 1/2q ∗ . For q > 98, non-dimensional units of ωu values for the 4-th loading case (longitudinal load is applied in two directions) become larger than ωu corresponding to the 5-th


471

loading case. We support the above conclusions using the numerical examples. For q = 0 we have ωu of the 5-th loading case of amount of 1.24 times larger than ωu corresponding to the 4-th loading case, (for q = q ∗ it achieves amount of 1.19 times. The upper frequencies for q = q ∗ are increased correspondingly to q = 0 and q = 3/4q ∗ of amount of 10.89 and 5.82 times for the 5-th loading case, whereas for the 4-th loading case of 16 and 6.67 times, correspondingly.


The numerical values of the critical frequencies for the 1-st and the 4-th loading cases and for different amplitudes of the longitudinal load are given in Table 4.7. The frequencies, for different loading cases P x0 = Py0 = 10 are given in Table 4.8. We consider the vibrational process of the surface points in time for different loading cases (1 - Fig. 4.47, 2 - Fig. 4.48, 3 - Fig. 4.49, 4 - Fig. 4.50, 5 - Fig. 4.51) for time instants, when deflection in the centre reaches its maximum and minimum for q = 0 (the upper curves) and q = 3/4q ∗ (the lower curves). We analyse the surface points vibrations for cylindrical shells. As it has been seen in the figures, the surface points vibrations are complicated and depend on the way of loading. However, it is rather typical that during vibrations negative deflection zones occur in the neighbourhood of the shell’s centre, and positive deflection zones appears close to the shell’s edges. For the shells loaded using the 2-nd and 3-rd ways, pre-critical vibrations of deflections are characterized by large amplitudes and positive deflection values.

472





473

Table 4.7. Critical frequency values for different loading types. Loading types

q in relation to q ∗ q /q ∗ 0

0.25

1

0.5

0.75

1

0

0.25 4

0.5

0.75

P x0 + Py0 15 15 10 5 15 10 5 15 10 5 15 10 5 15 10 5 1 20 15 10 6 20 15 10 6 20 15 10 6 20 15 10 6

ωl 2 2 6 11 0 3 8 0 0 7 0 0 0 0 0 0 0 0 0 6 14 0 0 4 10 0 0 0 0 0 0 0 0

ωu 24 24 20 11 24 20 11 25 23 17 27 24 20 1600 600 20 11 74 62 50 38 79 67 53 40 87 74 57 45 216 156 120 62

Maximal negative deflections are of 5-6 times smaller than the absolute values of maximal positive deflections. The shells loaded by the first way have essentially smaller values of deflection (7-8 times) in comparison to the other loading ways in both positive and negative zones. Therefore, also the vibration amplitudes are small. Increasing the heat flow intensity the vibration shift in the positive deflection direction is observed, and a decrease of maximal possible pre-critical deflections and amplitudes is noticed. For the shells loaded in the first way a reversed picture is observed. Namely deflection and amplitudes increase. The zones of negative deflections decrease with

474

4 Dynamical Behaviour and Stability of Rectangular Shells with Thermal Load Table 4.8. Critical frequency values for P x0 = Py0 = 10. Loading types 1 2 3 4 5

ω ωl ωu ωl ωu ωl ωu ωl ωu ωl ωu

0 6 20 5 74 6 56 6 50 8 62

0.25 3 20 3 78 4 53 4 53 7 64

q /q ∗ 0.5 0 23 0 108 0 59 0 57 5 74

0.75 0 24 0 198 0 104 0 120 0 116

1 0 600 0 670 0 660 0 800 0 675

the increase of q . For the shells loaded using the 3-rd way, the vibrations are fully shifted to the positive deflections zone.

Figure 4.52. The dependencies ω(w) and ω(σy ) for the cylindrical shell for K x = 0, Ky = 48, P x0 = 10 (deflections are denoted by dashed curves).


475

Table 4.9. Stress-strain states of shells. q

0

0.75q ∗

Characteristics ω σx ω σx ω σx ω σx

Loading types 1 2 3 4 5 Frequency 2.27 4.43 4.68 4.487 3.05 outside -11.0312 45.5382 52.5382 66.1082 32.2952 the zone 9.039 9.88 6.35 6.53 7.15 in the 86.534 92.5671 77.8436 96.6472 91.1287 zone 3.57 2.58 3.34 2.63 4.41 outside -7.3693 -4.861 -8.8167 -23.5604 39.4474 the zone in the 24.2094 205.317 48.7387 8.7424 138.8328 zone

For instance, the maximal possible deflection and amplitude for the shells loaded in the 2-nd and 3-rd ways are decreased, whereas for the shells loaded in the 1-st way they increased of amounts of 1.78 and 1.78 times, 1.88 and 3.2 times, and 1.51 and 1.44 times, respectively. We analyse now the surface vibrations for the spherical shells. As it has been seen in the figures, for q = 0, when the longitudinal load acts simultaneously in both directions (the 4-th loading way) the vibrations are quasi-periodic. Positive deflections on the whole surface are substituted by the negative ones. The deflection maximum occurs in the shell’s centre. When the longitudinal load acts only in one direction (the 5-th loading way) the vibrations become complicated in both space and time. The vibrational process for the spherical shells is characterized by relatively large pre-critical deflections and amplitudes of vibrations. Increasing the heat flow intensity, the vibrational process fully moves in the direction of the positive deflection. The maximum deflection point begins to vibrate between the shell’s centre and the centres of each of the shell’s quadrant. When the shell’s centre achieved its minimum, the relative maximum achieved 4.5 and 13 units for the 4-th and the 5-th loading ways, respectively. It means that the maximum deflection zones for the spherical shells (for q = 3/4q ∗ ) are clearly displayed and a sudden change of the surface form is observed. We analyse the stress strain state in the process of the excitation frequencies transition through the unstable zones. We analyse the dependencies ω(w) and ω(σ x ) for q and q = 3/4q ∗ (for deflection - the dashed curves) for a spherical shell with the parameters K x = 0, Ky = 48, P x0 = 10, Py = 0 (Fig. 4.52). A transition through the unstable zone is characterized by a sudden increase of deflections and stresses. For q = 0 the stresses change their sign. The frequencies, for which the deflections and stresses achieve their maximum, do not overlap. The point of maximal deflection is shifted in the direction of the low boundary, and the point of maximal stresses is shifted in the direction of upper boundary. Increasing the heat flow intensity, the deflections outside of zone instability are increased. Practically, the heat stream has no influence on the stresses. In the unstable zone with the increase of q , the deflections and stresses are decreased.

476


Figure 4.53. Vibration of the cylindrical shell centre for the 1st (a), 2nd (b), 3rd (c) loading types.

We compare the stress strain states for different loading ways of cylindrical and spherical shells for the heat stream intensity q = 0 and q = 34 q ∗ on the upper boundary of the unstable zone. The results are given in Table 4.9. The results shown in Table 4.9 support the conclusions obtained during the analysis of the vibrational process in relation to deflections in the pre-critical state. Besides, it is seen, for all loading ways (except for the 1-st) and without the heat flow the transition through the upper zone’s boundary is smoother for both deflec-


477

Figure 4.54. The spherical shell centre vibrations for the 4th (a) and 5th (b) loading types.

tions and stresses. For the 1-st loading way for q = 0 and for all loading ways for q = 3/4q ∗ , a sudden increase of deflections and stresses is observed. Comparing the shells stress-strain state for different loading ways, the following conclusions are obtained. The 1-st loading way corresponds to smaller stresses and deflections in comparison to the other loading ways for the cylindrical shells. The

478


spherical shells are the most suitable in the case of the 4-th loading way. It should be noted, that the highest stresses occur in the spherical shells subjected to the longitudinal force in one direction, and for the cylindrical shells along the nonzero curvature direction (together with the heat flow). We consider the vibrational process on the example of the shell’s central point for different loading ways. For this purpose, the dependencies ω(τ) for the cylindrical (Fig. 4.53 a - 1-st loading way; Fig. 4.53 b - 2-nd; Fig. 4.53 c - 3-rd) and spherical (Fig. 4.54 a - 4-th; Fig. 4.54 b - 5-th) shells are displayed. The dependencies ω(τ) are obtained during the excitation frequency transition through the upper boundary of the unstable zone (in Fig. 4.53, 4.54 the solid curves correspond to the vibrational process for q = 0, whereas the dashed ones corresponds to q = 3/4q ∗ ). As it has been seen in Fig. 4.53 (for q = 0) the pre-critical state of shells with the 2-nd and the 3-rd loading ways are characterized by large deflections and amplitudes, and also by a slight deflection difference during the transition through the unstable zone. For q = 0 a stability loss occurs after a few vibrations with relatively small deflections. For a shell loaded by the 1-st (2-nd and 3-rd) way, a stability loss is observed on the third (fourth and second) wave of the vibrational process. For the shells loaded by the longitudinal harmonic load along the nonzero curvuture direction (the 2-nd and 3-rd loading way) from the halfwave in the shell’s negative deflection values. For the shells loaded by the longitudinal load along the zero curvuture direction the vibrational process begins from the halfwave located in the positive deflection area. When the heat flow intensity achieves its critical value, the vibrational process is qualitatively the same for the shells loaded in each of the considered loading ways. Amplitudes and deflections are relatively small. A stability loss is observed on the first wave of the vibrational process. Now we analyse the vibrational process of the spherical shells. As it is shown in Fig. 4.54, for q = 0 the vibrational process of the spherical shells is characterized by relatively high deflections and amplitudes and a slight difference in deflections during the transition through the unstable zone. A stability loss of the shells loaded using the 4-th way is observed already on the first wave of the vibrational process, and for the 5-th loading way it is observed on the third wave of the vibrational process. When the heat flow intensity reaches its critical value, the vibrational process of the spherical shells is characterized by relatively small amplitudes and deflections, and a sudden increase of deflection during the transition through the unstable zone occurs. A stability loss of the shells loaded using the first (fifth) way, corresponds to the first (second) wave of the vibrational process.

4.5 Stability of Flexurable Conical Shells with Convection

479

4.5 Stability of Flexurable Conical Shells with Convection 4.5.1 Problem Formulation We consider dynamical stability of shells with a convectional type heat transfer, when the Young’s modulus and the linear heat expansion coefficient do not depend on the temperature. For this case, the equations (4.23) and (4.26) have the following form 4 4 ∂4 w 1 −2 ∂ w 2∂ w − ∇2k F − L (w, F) + λ + 2 + λ & ' 12 1 − ν2 ∂x4 ∂x2 ∂y2 ∂y4 2 ∂w ∂ w = qs , +ξ (4.64) κ ∂τ ∂τ2 λ−2

4 ∂4 F ∂4 F 1 2∂ F + 2 + λ + ∇2k w + L (w, w) = 0. 2 ∂x4 ∂x2 ∂y2 ∂y4

(4.65)

The unstationary convection process occurs due to the Newton’s law on the shell’s internal surface. The external and other shell’s surfaces are isolated. The boundary conditions for the heat transfer equation (4.42) for a heat exchange have the form 1 ∂T + Bi (T − T s ) = 0 for z = , ∂z 2 1 ∂T = 0 for z = − , ∂z 2

(4.66)

where Bi = (αg h)/λg ; T is the surrounding medium temperature; αg is the heat expansion coefficient on the surface z = 0.5. The initial conditions for the heat transfer equations (4.42) have the form: T = T0

for τ = 0.

(4.67)

The formulated heat transfer problem (4.42) for a given boundary (4.66) and initial (4.4) conditions can be solved analytically [354]. The non-dimensional solution has the following form

1 exp −µ2n τ , Cn cos µ z + (4.68) T = T av − (T av − T 0 ) 2 n=1 where: Cn = 5

2Bi , 6 Bi (1 + Bi) + µ2n cos µn

and µn - are the roots of the equation µtgµ = Bi . The shells stability will be considered for the following boundary conditions: boundary condition 1 - (4.28); boundary condition 2 - (4.29);

480


boundary condition 3 - (4.30); boundary condition 4 - (4.31). The thermal moments and forces have the form Cn 5

6 µn sin µn + 2 cos µn − 2 exp −µ2 τ , Mt = −θ1 2µn n=1 ⎡ ⎤ Cn

⎥ ⎢⎢⎢ 2 ⎥ ⎢ sin µn exp −µ τ ⎥⎥⎥⎦ . Nt = θ1 ⎢⎣1 − µ n=1 n

(4.69)

(4.70)

The series, occurring in the terms of the thermal moments (4.69) and forces (4.70) are convergent. The calculations are stopped, when |an / an−1 | ≤ 10−8 . The numerical analysis proves that only six series terms are needed. When the temperature increase θ1 is positive, then the shells will work properly in the dynamical regime. Practically, this case can be realized, when the heating sources are used. When the temperature increment is negative, the shell will work in the condition of dynamical cooling. Both of the mentioned regimes are met during the shell’s constructions and need a detailed analysis. 4.5.2 Boundary and Thermal Fields Conditions We consider a vibrational process for each of the boundary conditions given in Sect. 4.5.1 for shells working in the cooling and heating conditions. The vibrational process (dependence w(τ)) will be considered on the example of the spherical shell with parameters (K x + Ky ) = 24 (Fig. 4.55). The digits correspond to the boundary condition number; the dashed curves correspond to the cooling process; the solid curves correspond to the heating process. Only for the shells with 1 boundary conditions and during heating, the stationary vibrational process is observed for the positive values. The axis of vibrations is always parallel to the initial deflection state. With the increase of the temperature, the shell loses its stability (Fig. 4.56 a). In order to detect the critical value of the temperature increment, all criteria described in Sect. 4.3.1 can be applied. For the other boundary conditions, in the heating regime, the shell subjected to the thermal forces deflects in the negative direction, and a stability loss behaviour cannot occur. The vibrational process is observed around the axis sloped to the initial deflection location, and the deflections increase infinitely with time. Similar results are obtained for a heat impact, and are given earlier in Sect. 2. Comparing the vibrational process for the boundary conditions 2 and 3, it has been observed, that the vibrational process of the shell deflection with the 3 boundary conditions is characterized by complexity and aperiodicity. An angle of the slope of the vibration axis is sufficiently smaller than in the case of the boundary condition 2 for the same temperature values. The action of thermal moments and thermal forces for heating (cooling) is opposite. In the case of heating, the thermal moment is going to deflect the shell inside, in the curvuture direction (contrary to the thermal force action).


481

Figure 4.55. Vibrations of spherical shells.

Similar consideration can be carried out for the cooling process. We analyse the shells stress state for the given in Fig. 4.55 cases. In dependence σ x (τ) presented in Fig. 4.56, a vibrational process of full stresses in the shell’s centre is outlined (on its internal surface). Comparing the vibrations of stresses (Fig. 4.56) and deflection (Fig. 4.55) a qualitative similarity is observed. The deflection maximum (in most cases) corresponds to the stresses maximum. For the boundary conditions 2, 3 and 1 and for both heating and cooling regimes, the stresses unboundedly increase. The largest stresses are observed in the cooling process for shells with the boundary conditions 4 subjected to both temperature moments and forces action. The stresses occurring in the shells (except for the boundary condition 1) have the same sign as the corresponding deflection. We compare the shells stress state, when the parameters (K x + Ky ) are increased. For this purpose, the dependencies ω(τ) and σ x (τ) (Fig. 4.57) are drawn for the shell two temperature regimes and two boundary conditions 1 and 3. For two different (K x + Ky ) values, the pre-critical deflection for shells with the 1 boundary

482


Figure 4.56. Time histories of full stresses in the shell centre on its internal side.

condition decrease, and the maximal possible critical deflections increase. For shells with other boundary conditions, for heating and cooling processes, a sloping angle of the vibrational process axis is decreased. For the shells with the boundary conditions 1 the critical stresses increase. For the shells with other boundary conditions, the stresses decrease. In order to analyse the surface points vibrations, the curves of relatively equal deflections of the shell’s quadrant for two time instants are observed in the heating regime (Fig. 4.58, the boundary condition 2 - a, b; 3 - c) and in cooling regime (Fig. 4.59, the boundary conditions 2 - a, b; 3 - c, d; 4 - e, f). The surface configuration of the shells with the boundary conditions 2 in the heating regime and with the boundary conditions 2, 3, 4 in the cooling regime, does not practically change in time. Although the absolute values of deflections are almost unchangeable. The maximal deflection always remains in the shell’s centre. For the shell with the boundary condition 3 in the heating regime, the surface form becomes complex and changes with time. The negative and positive deflection


483

Figure 4.57. The dependencies w(τ) and σ x (τ) for the spherical shell for two different temperature fields and two different boundary conditions.

values are observed. They are caused by thermal moments causing positive deflections, and by thermal forces causing negative deflections. In the initial time moment the thermal moment zone occupies more than a half of the surface area. The relative positive maximum is greater of amount of 2-2.5 times in comparison to the relative negative one. A zone of positive deflection decreases. The absolute deflection values increase, and the relative positive maximum approaches zero, whereas the negative one approaches 1. 4.5.3 Critical Temperature Versus Heat Transfer Coefficient We analyse a dynamical stability of flexurable conical shells, rectangular in plane, rollingly supported (boundary condition 1) with a convection on the internal surface. The heat transfer coefficient, for each of the considered cases, is defined experimentally and depends on many factors, such as surface shape, its roughness, medium surrounding viscosity, velocity of the stream, the shell’s material and other factors.

484


Figure 4.58. The curves of relatively equal displacements of the shell quadrant during heating (boundary condition 2 (a, b) and 3 (c, d)).

Therefore, we need to investigate the critical temperatures versus the heat transfer coefficient in the following non-dimensional form: Bi =

αg h λg

(Bi Biot number).

The investigations allow to define the critical temperature increment for cylindrical, as well as spherical, shells. We have obtained the dependencies of critical temperature increment θ10 [B1 , (K x + Ky )] (Fig. 4.60) and the corresponding full stresses σ0x [B1 , (K x + Ky )] (Fig. 4.61) on the internal surface for different shells (24 ≤ (K x + Ky ) ≤ 72); 0.2 ≤ Bi ≤ 0.8 and 1st boundary condition. The dashed curves correspond to cylindrical shells, the solid curves to spherical shells, and the dashed curves with dots correspond to equal values of θz0 and σ0x for cylindrical and spherical shells. The temperature increments, lying below the obtained solution zones, are precritical ones and they are not dangerous for the conical shells constructions. Increasing the (K x + Ky ) parameters, the values of the critical temperature increments suddenly increase. Similarly to the heat impact action (Fig. 4.4), the cylindrical shells are more stable than the spherical ones for (K x + Ky ) > 52.


485

Figure 4.59. The curves of relatively equal displacements of the shell quadrant during heating (boundary condition 2 (a, b), 3 (c, d) and 4 (e, f)).

Increasing the parameter Bi the values of the critical temperature increments decrease. Simultaneously, an approach to θ1 for spherical and cylindrical shells is observed. A slight shift of the critical values of the temperature increment for the considered shells types occurs.

486


Figure 4.60. The dependence of θ1 [Bi , (K x + Ky )].

Increasing Bi from 0.2 o 0.8, the θ1 is decreased of amount of 3.19 and 3.22 times for the spherical and cylindrical shells, respectively, for (K x + Ky ) = 36 for (K x + Ky ) = 72 the decrease achieves 3.32 and 3.37 times, respectively. For Bi = 0.8, a difference in θ1 at the ends of the considered (K x + Ky ) interval achieved an amount of 4.65 and 6.09 times for the considered shells types. We analyse a change of full stresses on the example of σ0x in the shells centre on their interval surfaces (Fig. 4.61) with an increase of Bi parameter. A change of the


487

Figure 4.61. The dependence of σ x [Bi , (K x + Ky )].

stresses accompanying the increase of (K x + Ky ) is similar to that of the temperature increment. For Bi = 0.2 the stress, beginning from (K x + Ky ) = 55, occurring in the cylindrical shells become higher than those of spherical shells. Increase of the Bi parameter is of less importance on the critical stresses in comparison to the critical temperature. With the increase of the dimensionless heat expansion parameter, the critical stresses values are decreased. For (K x + Ky ) < 48, beginning from a certain Bi value, the stresses change its sign. A slight stresses approach for the corresponding shells types is observed only for the relatively large values of (K x + Ky ) > 58. The point of equal stresses is shifted in the direction of the upper boundary of the (K x + Ky ) interval.

488

4 Dynamical Behaviour and Stability of Rectangular Shells with Thermal Load Table 4.10. Minimal values of the spherical and cylindrical shell characteristics. Characteristics θ10 0.2 w0 σ0x θ10 0.4 w0 σ0x θ10 0.8 w0 σ0x

Spherical shells

Bi

36 367 2.78 12.84 198 2.86 7.766 115 3.45 -1.187

48 722 2.95 51.82 386 3.02 48.13 217 3.09 40.88

60 1242 2.95 86.9 660 3.15 84.28 362 3.4 81.4

Cylindrical shells (K x + Ky ) 72 36 48 60 72 175 361 712 1448 2300 2.949 2.99 2.98 3.00 3.09 123.69 10.32 41.18 98.2 162.565 942 105 380 794 1222 3.2 2.97 2.93 3.16 3.21 122.57 5.56 38.89 91.95 158.45 535 112 215 438 682 3.5 2.98 3.11 3.3 3.4 119.04 4.93 32.58 86.96 149.97

Figure 4.62. The curves of relative deflections of the spherical (K x = Ky = 24) and cylindrical (K x = 0, Ky = 48) shells under a convection (Bi = 0.2).

For the shells with parameters (K x + Ky ) < 48, the stresses for Bi = 0.8 change their sign. The equality of σ0x for the corresponding shells types are observed for (K x + Ky ) = 58 of non-dimensional units. To conclude, the influence of the (K x + Ky ) parameters on the critical heat loads and the corresponding stresses is stronger for the case of the convection heat transfer than for the heat impact for arbitrary values of the non-dimensional heat transfer parameter.


489



490


Figure 4.65. The cylindrical (a) and spherical (b) shell centre vibrations for Bi = 0.2.

The minimal values of the critical temperature increments θ10 , deflection w and the stresses σ0x for different shells with the increase of Bi from 0.2 to 0.8 are given in Table 4.10. We consider the vibrational process of the surface points on the example of shells with (K x + Ky ) = 48. For this aim, the dependencies w(x, y)/w(0.5, 0.5) are


491

Figure 4.66. The cylindrical (a) and spherical (b) shell centre vibrations for Bi = 0.8.

derived. We compare the vibrational process with convection for Bi = 0.2 (Fig. 4.62) and with a heat impact (Fig. 4.5). The convection heat transfer is characterized by large (in comparison to the heat impact) amplitudes of vibrations and of the maximal possible pre-critical deflection.

492


During the convection, a maximal deflection vibration between the shell’s centre and the centres of each of its quadrants is independent of each shell’s type. Now we analyse the vibrational processes of the surface points increasing Bi from 0.2 to 0.8 on the example of shells with (K x + Ky ) = 48. We compare the following results: for Bi = 0.2 (Fig. 4.62); for Bi = 0.4 (Fig. 4.63) and for Bi = 0.8 (Fig. 4.64). With the increase of the dimensionless heat transfer parameter Bi the vibrational process amplitude is increased and the values of minimal deflection are decreased, whereas those of a maximal one are increased. The relative maximum for the time instant when a deflection in the centre achieves its minimum, is increased. The point of maximal deflection is shifted from the centre of each quadrant of the spherical shells to their edges and 8 points of maximal deflections are observed. In the case of cylindrical shells, they are shifted to zero curvature direction and 4 points of maximal deflections are observed. For cylindrical shells for Bi = 0.8 for the time instant when a deflection in the centre achieves its minimum, also an area with a negative deflection is observed in the region of the shell’s centre. We consider the vibrational process on the example of the central points of shells with (K x + Ky ) = 48 on the basis of w(τ) dependence for Bi = 0.2 (Fig. 4.65) and Bi = 0.8 (Fig. 4.66 a - spherical shells; b - cylindrical shells). The figures show that increase Bi parameter practically does not influence the vibrations. However, the amplitudes and maximal deflections increase, and the axis of the vibrational process is shifted in the direction of its initial value. The time of maximal (possible) first pre-critical deflection is decreased.

5 Dynamical Behaviour and Stability of Flexurable Sectorial Shells with Thermal Load

In section 5.1, research devoted to plates and shells with rectilinear, as well as straight and curvature lines contours are reviewed. In section 5.2 theory of flexurable sectorial shells computations is introduced. It includes derivation of fundamental relations and differential equations. Then a thermal field and the “set-up” method is introduced and numerical results reliability is discussed. Stability of sectorial shells with finite deflections id studied in section 5.3, where numerous problems are carefully analysed. In section 5.4 a novel approach to study chaotic vibrations of shallow sector-type spherical shells without thermal effects is proposed. Scales of vibration character of such shells being transversally and harmonically excited vs. control parameters are constructed. Scenarioto chaos are illustrated and discussed. Control of chaotic state applying synchronous action of harmonic loading torque is proposed.

5.1 Introduction Historical Review We analyse some works devoted to plates and shells computations with different characteristics. Plates and shells with rectilinear edges The considered plates and shells consist of simple convex polygons with sides number n ≥ 3. Warburton [699] has presented the first collection of solutions to rectangular plates’ problems. Leissa [425, 427] has collected and reviewed comprehensive literature dealing with free vibration of plates, up to 1977. The stability problems of triangle plates are considered in references [4, 645, 716]. In most cases, the numerical methods are used. For instance, in the work [4] the finite difference method with triple and sixple meshes of shells is applied. The finite difference method belongs nowadays to the most universal and effective methods. Its application to plates and shells with complicated shapes can be considered in various coordinate systems, i.e. rectangular or polar [321, 348]. In the works [645, 716] the finite element method is applied. The critical load values are obtained for uniformly compressed triangle plates.

494

5 Dynamical Behaviour and Stability of Flexurable Sectorial Shells

Experimental investigations of triangle plates are carried out using the methods of photoelasticity [141]. As an investigation object a triangle plate clamped on its contour and subjected to the uniformly distributed force action is used. The most stretching stresses have appeared on the angle bisectrix and along the plate’s angles. Trapezoidal plates are investigated by many authors [152, 462, 497]. Their stability is analysed using the Bubnov-Galerkin method [236, 557]. In the latter reference differential equations of the nonlinear theory of thin plates are used, and theoretical results are compared with the experimental ones. In reference [630] during the analysis of a trapezoidal plate with a triangle contour and finite deflection, the Bubnov-Vlasov method is used. As a function approximating a deflection, an elastic beam deflection is taken. The vibration behaviour of stiffened rectangular plates simply supported along the edges perpendicular to the stiffeners are studied by Wah [684] and Long [442, 443]. The free vibrations and dynamic response of simply supported rectangular plates has been analysed by Kirk [325], Aksu and Ali [10], and Ochs and Snowdon [517]. Smith et al. [626], Olson and Hazell [522] and Laura and Gutiérrez [411] obtained natural frequencies and mode shapes of stiffened square and rectangular plates with damped or elastically restrained edges. Clarkson and Cicci [155], Yurkovich et al. [715], Olson and Lindberg [523], and Donaldson [183] studied a dynamic response of the stiffened panel structures. Bhandari et al. [103] studied stiffened skew plates. Bapu Rao et al. [84] experimentally studied a skew cantilever plate with stiffener. An analysis of cantilever plates was reported in [85, 154]. Irie et al. [280] studied trapezoidal cantilever plates with non-uniform stiffeners. A trapezoidal plate was transformed into a square region of unit length by an ingenious transformation of variables. The transverse deflection of the transformed square plate was expressed in a series of the products of the deflection functions of beams parallel to the edges of the plate and satisfying the boundary conditions. A strain and kinetic energies of the system were evaluated analytically, and the frequency equation was derived by the conditions for a stationary value of the Lagrangian function. The method is applied to square parallelogram or trapezoidal cantilever plates with several stiffeners of the same material as the plates. Nair and Durvasula [493] have analysed the vibration of a skew plate by the Ritz method, and Chopra and Durvasula [151, 152] have analysed the vibration of a trapezoidal plate by the Bubnov-Galerkin method. Conway [156], and Walkinshaw and Kennedy [696] have studied vibrations of polygonal plates by point matching, and Shahady et al. [605] have analysed the same problem by a complex variable theory. Irie et al. [283] have investigated regular polygonal plates, and have obtained natural frequencies and the mode shapes including higher order modes. In the latter work, the authors have proposed a new series-type method for estimation of the eigenvalues of non-homogeneous plates with different boundary conditions. For this purpose, a plate is assumed to be clamped along an internal segment, and with the reaction forces and bending moments (acting on the segment) regarded as unknown harmonic forces and moments, its stationary response is expressed by the use of the Green functions. The unknown forces and moments distributed along the

5.1 Introduction

495

segment are expanded into the Fourier series with unknown coefficients. The homogeneous linear equations with unknown coefficients are given in a matrix form. The obtained eigenvalues and eigenvectors yield the natural frequencies and the mode shapes of the plate. Irregularly shaped plates can be build using a simply supported plate via clamping appropriate segments. The introduced theory using been applied to a cross shaped and I-shaped plate with two geometrically symmetrical axes, and to a L-shaped plate with a diagonal symmetrical axis. The natural frequencies and the mode shapes of the plates have been calculated numerically. Waller [698] has obtained Chladni’s figures of free polygonal plates experimentally. With the use of the point-matching method, Conway [156] has studied the fundamental vibration of simply supported polygonal plates, and Walkinshaw et al. [696] has obtained the frequencies of axially-symmetric vibrations of simply supported and clamped plates. Laura et al. [412, 416] and Yu [714] have analysed the same problem by a complex variable method. Applying the Ritz method, Young [706] has studied the free vibrations of a clamped square plate, and Ota et al. [525] investigated a triangular plate. Irie et al. [283] have proposed an analytical method to study the free vibrations of polygonal plates clamped at the edges. Conway [156], and Walkinshaw and Kennedy [696] have investigated analytically free vibrations of regular polygonal plates by the point-matching method, whereas Laura and co-workers [255, 605], and Yu [714] by the conformal mapping variational technique (see also Roberts [575]). Laura and Luisoni [414] have studied free vibrations of regular polygonal membranes applying the Bubnov-Galerkin method. Conway and Farnham [157] have analysed an equilateral triangular membrane, and Williams et al. [702] have studied the vibrations of a triangular plate both theoretically and experimentally. Durvasula [189], and Bauer and Reiss [90] have studied skew membranes, and the obtained results have been converted into those of an equilateral triangular membrane. Bauer and Reiss [91] have obtained the first 21 cutoff frequencies and modes of a regular hexagonal wave-quide by a combination of the finite differences and numerical techniques. Free vibrations of regular polygonal plates with simply supported edges are studied using the membrane dynamical analogy by Irie et al. [279]. A regular polygonal membrane is formed on the rectangular plate by fixing several segments. The natural frequencies and mode shapes have been calculated numerically. In the work [638], the finite element method with an application of a triangle element is used to analyse nonlinear deflections of the rhomboidal plates. During investigation of a stress-strain state of non-homogeneous shells, a theoretical-experimental method is proposed in reference [601]. Structural dependence of the being sought quantities on the geometrical and physical parameters and the external loads is established. An investigation of large deflection of parallelogram plates with uniformly distributed load using the theoretical-experimental method is carried out in [105]. In reference [298], an analytical method to solve the nonlinear problems of shells deflection is outlined.

496


A stability of an uniformly compressed plate is investigated in reference [400] using the complex variable method. The critical load for free and clamped plates with polygon shapes is calculated. Plates and shells with straight and curvuture lines contour. A stability of elliptic, parabolic and half-circle plates, compressed on the whole contour, is analysed in reference [468]. In order to find a solution, the method of equal deflection curve coupled with the Bubnov-Galerkin method, has been used. A relative difference of the free support circle, for which the exact solution is known, has exceeded 2%. Waller [698], and Maruyama and Ichinomiya [464] have studied experimentally the free vibration of isotropic sectorial plates. Ben-Amoz [99], Westmann [700] and Rubin [579] have analysed the same problems theoretically, and also Ramakrishnan and Kunukkasseric [565], Ramaiah and Vijayakumar [564], Bhattacharya and Bhowmic [104] and Wilson and Garg [705] have studied ring shaped isotropic sectorial plates with various circular edge conditions theoretically. Rubin [578, 580] has investigated both vibration and stability of polar-orthotropic sector plates using series. Irie et al. [281] have analysed the free vibration of a ring-shaped polar-orthotropic sectorial plates by the Ritz method using a spline function as an admissible function for the plates deflection. Roberts [88], Laura and Romanelli [412] have studied free vibrations of epicycloidal membranes. Laura et al. [149, 413, 416] have determined the cutoff frequencies of the wave-quides with epicycloidal cross-section. Sufficient engineering data, natural frequencies (the eigenvalues of vibration) are presented for epicycloidal plates by Irie et al. [284]. In reference [474] the Ritz method has been used to analyse small deflection of a half-circled plate with clamped edges. In reference [216], a solution in the polar coordinate system has been found for the plates bounded by two arcs of the concentrical circles. The plate has been subjected to normal transversal force action. In the works [344, 526], the results of stability investigation of post-critical state of plates and shells with complicated contour shapes (cutted circle, sixangle, triangle, parallelogram subjected to a transversal load and contour forces have been given). Using the hypothesis of straight verticals, in reference [32] a system of three equilibrium equations for displacements for arbitrary formed shells has been obtained. It is solved by the finite difference method. The calculations are carried out for the shells with an elliptic paraboloid shape. Cylindrical shells with a complicated form (convex and concave surface parts) are analysed in references [243, 244]. Difficulties due to the mathematical description of the shell’s surface are omitted applying spline approximations to the experimental data. In the reference [607], a calculation algorithm devoted conical shell analysis with a complicated contour by means of both linear orthogonal mesh and finite element methods is developed.

5.1 Introduction

497

The stress-strain state problems of isotropic plates of a circle sectorial and a circle rectangular forms are solved in the references [304, 658]. In the works [229, 230], a deflection of sectorial plates with arbitrary boundary conditions on its arc part, and with clamped straight linear boundaries is investigated. For the deflection function, the orthonormal system of special polynoms is applied. The solution is found tracing the system energy. In a frame of Kirchhoff-Love hypotheses, the deflection of a thin circle sectorial plate is found. Using Euler method a differential equilibrium equation of a plate, is reduced to the nonhomogeneous one with constant coefficients. A solution to the homogeneous equation is found using the method of variables separation. A particular solution has been found in the form of a trigonometric series. In the reference [530], the partial differential equations are reduced to the ordinary ones. In the work [105], in order to find a solution, the Bubnov-Galerkin procedure is used, whereas in reference [524], the finite difference method is applied. In references [254, 563], the sectorial plates are analysed, and the finite element is defined as a ring sector, whereas in reference [137], a solution of the sectorial orthotropic plate is given in the form of a trigonometric series. It follows from a brief review of the works devoted to the investigations of plates and shells with complicated plane that the sectorial plane of plates and shells is investigated rather slightly. The existing solutions are applied mainly to the sectorial plates in the frame of the linear theory. Therefore, a consideration of dynamical and statical problems of sectorial shells in a frame of non-linear geometry is required. The following notation is applied: w F r, θ U, V a h R γ ρ E ν g ω0 ε W0 α, λq , λT

- normal displacement of the mean surface; - force; - polar coordinate system; - displacements in the r, θ directions; - shell’s radius in plane; - shell thickness; - curvature radius of the mean surface; - specific gravity of the material; - material density; - Young modulus; - Poisson’s coefficient; - earth acceleration; - frequency of external forcing; - damping coefficient; - heat amount of the shell volume unit and the time unit; - thermal transfer coefficient, heat transfer coefficient and linear thermal extension coefficient;

498


αq C q

pr , pθ , q t τ T0, T ei j M, N

- heat giving back coefficient; - ideal specific heat capacity; - heat flow density in the direction of a normal to mean surface; - intensity of the external loads along the coordinates r, θ, z; - time; - non-dimensional time; - initial temperature, temperature; - components of deformation tensor for arbitrary shell’s point; - nodes number (including a contour) corresponding to the axes θ and r; εi j - deformation tensor components for the mean surface; σi j - stress tensor components; Nr , Nθ , Nrθ - forces on the mean surface; Mr , Mθ , Mrθ - bending and torsional torques; Qr , Qθ - transversal forces; θk - central angle of a sectorial shell; √ a2 b - sloping parameter: b = η Rh ; The following quantities are introduced: αq λq α2 B= h, α = , κ = 2 4 , ω0 = λq cρ ω0 h

0 Eq , γR2

RλT h b2 h2 η = 12 1 − ν2 , κ0 = , λ= 2 . λq a √ 2 ηq R r F √ w r = b , w = η , F = η 3, q = , a h 4E h Eh τ=

α z RλT εh2

RλT , , , κT = RκT , z = T = T q = q , ε = t, h h λq α h2 κ εT =

Mi j R R √ Ni j R √ R εT , M i j = η , N i j = η 2 , εi j = η εi j , h h Eh3 Eh W 0 = κ0 W0 .

5.2 Flexurable Conical Sectorial Shells Computations In this section a mathematical background of the uncoupled thermoelastic problem is given, and the methods and algorithms of sectorial shells with finite deflections analysis are proposed. The relations for deformation and displacement components,

5.2 Flexurable Conical Sectorial Shells Computations

499

and the forces and moment versus deformations are derived using the geometrical and statical Kirchhoff-Love hypotheses. A relation between deformations and displacements is of second power form, and this approach is widely applied in the geometrical nonlinear theory of shells. The thermal field is defined through a solution to the three dimensional heat transfer equation for arbitrary boundary conditions independent of the elasticity equation. On the basis of the considered hypotheses and assumptions the hybrid form equations system for displacement is achieved. In the last case, all quantities are expressed by the deflection function w and the force function F. 5.2.1 Fundamental Relations, Differential Equations, Boundary and Initial Conditions Using the Vlasov’s hypotheses, the fundamental relations and equations in the polar coordinate system are introduced. For a conical shell, a geometry of its surface can be approximately treated as overlapping with a geometry of its projection. Therefore, a position of an arbitrary point on the surface in the polar coordinate system is defined by two quantities: r, θ, measured from the initial vector radius r0 . The coordinate z is directed along the normal to the mean curvature surface. Consider the a b c d shell’s element presented in Fig. 5.1.

Figure 5.1. Shell computation scheme.

500


Figure 5.2. Stresses and moments of the shell element.

The extension and shear deformations relations are similar to those of Descartes coordinate system [684], except for the relative extension along the θ axis and the angle deformation. Both of them depend on the displacements U and V.


501

Therefore, the geometrical Cauchy relations in the polar coordinate system have the form: 2 ∂U w 1 ∂w ∂V lU w 1 ∂w − + ; εθ = + − + ; εr = ∂r R 2 ∂r R 2 r ∂θ r ∂θ r ∂U ∂w ∂w ∂V V − + ; (r, θ) → − (x, y) . (5.1) + r ∂θ ∂r r r ∂θ ∂r According to the Kirchhoff-Love hypotheses, full deformation of an arbitrary point along thickness er , eθ , , erθ are composed of an average surface deformation and a deflection deformation: er = εr + zκr , γrθ =

eθ = εθ + zκθ , erθ = γrθ + 2zκrθ , where:

(5.2)

1 ∂2 w ∂2 w 1 ∂w − 2 2 , κr = − 2 , κθ = − ∂r r ∂r r ∂θ ∂ 1 ∂w ∂w 1 ∂w , υr = − , υθ = − . κrθ = − ∂r r ∂θ ∂r r ∂θ

In the above, υr denotes rotation angle of a normal in the plane zOr; υθ is rotation angle of a normal along the axis r. We consider an equilibrium of the shell’s element cut by two meridial planes and two planes perpendicular to them. The meridial σr , σθ and tangential σrθ = σθr , σrz , σθz stresses appear on the element’s sides. We introduce statically equivalent forces and torques: h

h

2 Nr =

h

2 σr dz, Nθ =

h −2

2 σθ dz, Nrθ = Nθr =

h −2

h −2

h

h

2 Qr =

2 σrz dz, Qθ =

h −2 h

h −2 h

2 σr zdz, Mθ =

h −2

σθz dz;

h

2 Mr =

σrθ dz;

2 σθ zdz, Mrθ = Mθr

h −2

σrθ zdz.

(5.3)

h −2

In the above, the Nθr , Nrθ and Qr are the normal, shear and transversal forces acting in the meridial direction; Nθ , Nθr = Nrθ , Qθ are the normal, shear and transversal

502


forces on the radial direction; Mr and Mrθ are the bending moment and the torque in the meridial direction; Mθ , Mθr = Mrθ are the bending moment and the torque in the radial direction. The internal forces and moments are related to the length unit and to the corresponding coordinate line (a circle or a radius) of the mean surface. Positive direction of internal forces and moments, as well as of the external load, are presented in Fig. 5.2. The introduced internal forces, moments and external forces allow (instead of the shell’s space equilibrium element) for a consideration of the corresponding element of its mean surface. The deformations (5.2) consist of elastic ones caused by the stresses σrθ , σθ , σr and a purely heat deformation. Using well-known relations obtained from the Hooke’s rule for the two dimensional stress state, and taking into account only a relative heat stretching, we get: er =

σr − νσθ + αT (T − T 0 ) , E

σθ − νσr + αT (T − T 0 ) , E 2 (1 + ν) σrθ . (5.4) erθ = E Multiplying (5.4) by dz, and then by zdz, we proceed with the integration within the interval from z = − h2 to z = − h2 . Using the relations (5.2) and (5.3), we get: eθ =

εr =

Nr − νNθ Nθ − νNr 2 (1 + ν) + εT , εθ = + εT , γrθ = Nrθ ; Eh Eh Eh

(5.5)

12 (Mr − νMθ ) 12 (Mθ − νMr ) 12 (1 + ν) +κT , κθ = +κT , κrθ = + Mrθ ; (5.6) Eh3 Eh3 Eh3 where εT and κT are generalized purely heat deformations: κr =

h

h

εT =

1 h

2 αT (T − T 0 ) dz; κT = h −2

12 h3

2

αT (T − T 0 ) zdz.

(5.7)

h −2

Owing to (5.5) and (5.6), the relations between the moments, forces, and deformations are found: Nr =

Eh Eh [εr + νεθ − (1 + ν) εT ] , Nθ = [νεr + εθ − (1 + ν) εT ] , 2 1−ν 1 − ν2 1 − ν Eh γrθ ; 2 1 − ν2 Mr = D [κr + νκθ − (1 + ν) κT ] , Nrθ = Nθr =

(5.8)


Mrθ = Mθr = (1 − ν) Dκrθ .

503

(5.9)

We derive the equation governing the shell’s element motion. According to the D’Alembert principle, we add the inertial forces to both given forces and dynamical reactions of the neighbourhood elements. The equations system governing a motion of the shell’s deformed mean surface element consists of projections equation of all forces on the coordinate axes and the equations for moments in relations to those axes. The inertial terms occurring as a result of rotations are not included in the considerations. Neglecting the second order terms, the equations governing a motion of the shell’s element [684] in the polar coordinate system follow: γ ∂2 U ∂Nr r ∂Nθr + − Nθ + Pr − h 2 = 0, ∂r ∂θ g ∂t γ ∂2 V ∂Nrθ r ∂Nθ + − Nrθ + Pθ − h 2 = 0, ∂r ∂θ g ∂t ∂Qθ 1 1 ∂Qr r + − rMr + κr − rNθ + κθ − ∂r ∂θ R R 2 γ ∂ w 2rNrθ κrθ + r q − h 2 = 0, g ∂t ∂Mrθ r ∂Mθ + + Mrθ − Qθ r = 0, ∂r ∂θ

(5.10)

(5.11)

(5.12) (5.13)

∂Mrθ r ∂Mrθ + − Mθ − Qr r = 0. (5.14) ∂r ∂θ If we consider a dynamical process without elastic waves propagation, then in equations (5.10), (5.11) the inertial terms are deleted. Those equations will be satisfied by identity, when we introduce the stress function in the averaged surface [684] owing to relations: 1 ∂2 F ∂F ∂2 F + 2 2 , Nθ = 2 , Nr = ∂r r ∂r r ∂θ ∂ 1 ∂F , (r, θ) → Nrθ = Nθr = − − (x, y) , (5.15) ∂r r ∂θ where is a transition parameter from the polar coordinate system to the Descartes one. We need equal deformations condition of the form [684]: 1 ∂2 γrθ r 1 ∂2 εr 1 ∂2 εθ r 1 ∂εr = . + − r ∂r2 r ∂r r2 ∂θ2 r2 ∂r∂θ

(5.16)

Expressing in (5.16) the deformations of the mean surface by the forces given in (5.5), and taking into account (5.15) we get the deformation continuity equation: 1 2 2 1 1 ∇ ∇ F + ∇2 εT = − N (w, w) − ∇2 w, Eh 2 R

(5.17)

504


where: ∇2 = ∇2 ∇2 =

1 ∂2 ∂2 ∂ + + , ∂r2 r ∂r r2 ∂θ2

2 ∂4 ∂4 2 ∂3 ∂2 ∂ + + − + − ∂r4 r ∂r3 r2 ∂r2 r3 ∂r r2 ∂θ2 ∂r2

1 ∂4 4 ∂2 2 ∂3 + 4 2 + 4 4 , 2 3 r ∂θ ∂r r ∂θ r ∂θ 2 2 2 ∂ 1 ∂w 1 ∂w ∂ w ∂w + −2 . N (w, w) = 2 2 ∂r r ∂θ ∂r r ∂r r2 ∂θ2 In order to obtain the motion of the shell’s element in a hybrid form, we use the formulas (5.12)–(5.14). Reducing the forces Qr , Qθ , and expressing the moment by the formulas (5.9), and the forces by the formulas (5.15), we get: % $ 1 −D ∇2 ∇2 w + (1 + ν) ∇2 κT + N (w, F) + N (w, F) + ∇2 F + q = R 2 γ ∂w ∂w h , +ε g ∂t2 ∂t

(5.18)

where: 1 ∂2 F 1 ∂2 w ∂2 w ∂F ∂2 F ∂w + + N (w, F) = 2 + 2 − ∂r r ∂r r2 ∂θ2 ∂r r ∂r r2 ∂θ2 ∂ 1 ∂w ∂ 1 ∂F . 2 ∂r r ∂θ ∂r r ∂θ The equations (5.17)–(5.18) govern a motion of the shell’s elements. We are going to get the equations system for displacements. We express the forces of (5.10), (5.11), (5.12) by the deformations using the formula (5.8), and the deformations are expressed through the U, V, w displacements owing to (5.1). The relations between stresses and displacements read ⎧ 2 ⎪ Eh ⎪ ⎨ ∂U w 1 ∂w − + − (1 + ν) εT + Nθ = ⎪ ⎩ ∂r R 2 ∂r 1 − ν2 ⎪ ⎡ ⎢⎢ ∂V U + ν ⎢⎢⎢⎣ r∂θ r ⎧ ⎪ Eh ⎪ ⎨ ∂U U + Nr = ⎪ ⎩ r∂θ r 1 − ν2 ⎪ ⎡ ⎢⎢ ∂U − ν ⎢⎢⎢⎣ ∂r

2 ⎤⎫ ⎪ w 1 ∂w ⎥⎥⎥⎥⎪ ⎬ − + ⎥⎪ , ⎭ R 2 r∂θ ⎦⎪ 2 w 1 ∂w − + − (1 + ν) εT + R 2 r∂θ 2 ⎤⎫ ⎪ w 1 ∂w ⎥⎥⎥⎥⎪ ⎬ + ⎥⎪ , ⎭ r 2 ∂r ⎦⎪


Nrθ =

505

∂U ∂V V ∂w ∂w Eh + − + . 2 (1 + ν) r∂θ ∂r r r∂θ ∂r

(5.19)

From (5.10) we get: ν − 3 ∂V 1 + ν ∂w ∂2 U 1 − ν ∂2 U ∂U U 1 + ν ∂V − + + − + r+ + 2 2r ∂θ2 ∂r r 2 ∂r∂θ 2r ∂θ R ∂r ∂r 2 1 − ν ∂w 1 + ν ∂w ∂2 w ∂2 w ∂w 1 − ν ∂2 w ∂w + + − + 2 ∂r 2r ∂θ ∂r∂θ ∂r2 ∂r 2r ∂θ2 ∂r 2 ∂ 1 − ν2 γ 1 − ν 2 ∂2 U 1 + ν ∂w pr − − (1 + ν) εT + = 0. (5.20) 2 r∂θ ∂r Eh g E ∂t2 Proceeding in a similar way, we obtain from (5.11): 1 + ν ∂2 U 3 − ν ∂U ∂2 V 1 − ν ∂2 V 1 − ν ∂V U + + r 2 + − + + 2 ∂r∂θ 2r ∂θ 2 2 ∂r r r∂θ2 ∂r 1 + ν ∂2 w ∂w 1 − ν ∂w ∂2 w 1 ∂w ∂w + + + 2 ∂θ∂r ∂r 2 ∂θ ∂r2 r ∂θ ∂r ∂ 1 − ν2 γ 1 − ν2 ∂2 V 1 ∂w ∂2 w ν ∂w − (1 + ν) εT + pθ − − = 0. 2 r ∂θ ∂θ R ∂θ ∂θ Eh g E ∂t2

(5.21)

Reducing Qr and qθ from equation (5.22) using (5.13) and (5.14), one gets: 1 ∂2 Mr r 1 ∂2 Mθ 2 ∂2 Mrθ r 1 ∂Mθ − + + + r ∂r2 r ∂r r2 ∂θ2 r2 ∂r∂θ 1 ∂2 w 1 1 ∂w 1 ∂2 w Nr + + + + Nθ + R ∂r2 R r ∂r r2 ∂θ2 γ ∂2 w ∂ 1 ∂w + q − h 2 = 0. 2Nrθ ∂r r ∂θ g ∂t

(5.22)

Substituting in equation (5.22) the moments by displacements owing to (5.9), and substituting the forces by the displacements through (5.19), the motion equation for displacements (projected on the normal to the averaged surface) is obtained −

h2 2 2 ∇ ∇ w + (1 + ν) ∇2 κT + 12

w ∂U ∂V U + + − (1 + ν) − (1 + ν) εT + ν ∂r r∂θ r R 1 1 ∂w 1 ∂2 w ∂U ∂V U w + + 2 2 +ν + ν − (1 + ν) − (1 + ν) εT + R r ∂r r ∂θ ∂r r∂θ r R ∂ 1 ∂w ∂U ∂V U ∂w ∂w 1 (1 − ν) + − + + ∂r r ∂θ r∂θ ∂r r ∂θ ∂r r

1 ∂2 w + R ∂r2

506


q

1 − ν 2 γ 1 − ν 2 ∂2 w − = 0. Eh g E ∂t2

(5.23)

In order to get the equation systems (5.17)–(5.18) and (5.20)–(5.23), we need to add an unstationary three dimensional heat transfer equation in the polar coordinates system of the form: ∂2 T w0 1 ∂T = 2 + ∇2 T + . (5.24) α ∂t λq ∂z In the heat transfer equation (5.24) a mechanical coupling term does not appear, and therefore the temperature and deformation fields are not coupled. Integrating the equations (5.17)–(5.18), (5.20), (5.21), (5.23), and the heat transfer equation (5.24) use the boundary and initial conditions should be attached. Various mathematical models of the boundary conditions are applied owing to occurrence of different support in real shell’s constructions. Sometimes a support type influences the choice of the initial differential equations. We consider some of the boundary conditions. For the equations related to w and F, the following boundary conditions can be used. 1. Rolling support on the arcal elements w = Mr = Nrθ = Nr = 0 for r = a, 0 < θ < θk .

(5.25)

2. Rolling support on the radial elements w = Mθ = Nθ = εr = 0 for θ = 0, θk , 0 < r < a.

(5.26)

3. Slip clamping along arcs w=

∂w = Nrθ = Nr = 0 for r = a, 0 < θ < θk . ∂r

(5.27)

4. Slip clamping along radiuses w=

∂w = Nθ = εr = 0 for θ = 0, θk 0 < r < a. ∂θ

(5.28)

The following boundary conditions are applied with respect displacements: 1. Stiff clamping of the arcal elements w=u=ν=

∂w = 0 for r = a, 0 < θ < θk . ∂r

(5.29)

2. Stiff clamping of the radial elements w=u=ν=

∂w = 0 for θ = 0, θk 0 < r < a. ∂θ

(5.30)


507

3. Rolling-unmovable clamping of the arcal elements u = ν = w = Mr = 0 for r = a, 0 < θ < θk .

(5.31)

4. Rolling unmovable clamping of the radial elements u = ν = w = Mr = 0 for θ = 0, θk 0 < r < ba.

(5.32)

In the above 2 1 ∂2 w ∂w Eh3 ∂w 1 + − (1 + ν) κT , +ν Mr = − & ' ∂r r r2l ∂θ2 12 1 − ν2 ∂r2 1 ∂2 w 1 ∂w Eh3 ∂2 w + Mθ = − & + ν 2 − (1 + ν) κT . ' 12 1 − ν2 r ∂r r2l ∂θ2 ∂r

(5.33)

(5.34)

In the case, when Qk = 2π a top of the spherical segment becomes the shell’s internal point. Because in the equation the terms with multipliers r−n (n = 1, 2, 3, 4) occur, then the integration of the equation from the point r = 0 using the numerical methods is impossible. Therefore, we need to formulate the additional conditions on the top. The solving function’s behaviour, in a close neighbourhood of the top, can be analysed in a different way. For instance, in reference [344] using a limiting case, the equations valid in the top neighbourhood are obtained. However, they cannot be used in many cases, including our. In the reference [697] another approach is proposed. The transformation of the fundamental equation leads to exactly solved linearized equations, which can be used for r = 0. However, an analytical solution to the linearized equation cannot always be found and for our equations a similar approach can not be applied. Therefore, sometimes in the neighbourhood of the shell’s top, its geometry can be modified. For instance, in the reference [640], a neighbourhood of the top is replaced by a circle plate of constant, thickness and then the exact solutions are found. However, this approach can be used only in the case of a symmetric solution. In the monograph [696] this problem in a close top neighbourhood of the conical spherical shells is analysed. The solution to the linear equations are found in a series form. For an axially symmetric solution, only the first terms of the series are taken: w0 = A +

B 2 r + O r3 , 2

(5.35)

D 2 r + O r3 . (5.36) 2 where w0 is deflection and F0 is the stress function of the shell top. Increasing the terms number does not improve the solution accuracy. Observe that in a small neighbourhood of the shell’s top, one of the fundamental assumptions of the thin shell’s theory about smallness of h/R is not satisfied. Therefore, the obtained equation only approximately describes a real picture of the stress strain state. F0 = C +

508


In most cases of the numerical method applications it is assumed that the shell has a central hole with small dimensions, which only very slightly influences the obtained solution. In the reference [104] it is assumed that the shell is clamped on the circle with a small radius r0 . In order to solve the axially symmetric problems, the being sought functions in the point r = 0 are interpolated by the Lagrange 2-nd order formula of the form: f 0 = 3 f 1 − 3 f 2 + f3 ,

(5.37)

where: fi = f (ri ), ri = i∆ (i = 0, 1, 2, 3) and ∆ denotes a distance between the interpolation nodes. For an outside contour point, the following symmetry condition is satisfied: (5.38) f−1 = f1 . The numerical comparison of the results obtained using formulas (5.35)–(5.38) is carried out in the next section. The solutions found using (5.37), (5.38) only slightly differ from the results obtained from (5.36). The relative errors, for the deflection function and the forces function, achieve 1% and 4%, respectively. The conditions (5.37), (5.38) are more simple. They do not need transformations of the equations, and the applied algorithms are simple, which is not true for the case of (5.35) and (5.36). In order to solve the dynamical problem we need to introduce initial conditions in time t = 0 of the form: ∂w = ϕ2 (r, θ) . (5.39) w = ϕ1 (r, θ ) ; ∂t In addition, the boundary and initial conditions for the heat transfer equation (5.39) should be formulated. 1. Temperature distribution on the surface body is assumed (1-st order boundary conditions) T (r, θ, z, t) = ψ(r, θ, z, t), (5.40) where: (r, θ, z) is the point of the body surface; ψ(r, θ, z, t) is the given function. 2. The heat flow intensity q (r, θ, z, t) (2-nd order boundary conditions) transfered through the body surface has the form ∂T (r, θ, z, t) = q (r, θ, z, t) , (5.41) ∂n where n denotes the external normal to the surface body in the point (r, θ, z). 3. Temperature distribution of the surrounding medium and a rule of convection between the surface body and the surrounding medium is assumed (3-rd order boundary conditions) $ % ∂T (r, θ, z, t) = αq T (r, θ, z, t) − T cp . (5.42) −λq ∂n The initial conditions for the heat transfer equations (5.39) have the form: −λq

t = 0,

T = T0.

(5.43)


509

All initial equations and boundary conditions will be given further in the nondimensional form. Imagine we need to find a solution to the nonlinear boundary problem governing a statical equilibrium of the sectorial shells ∇2 ∇2 w − N (w, F) − ∇2 F − 4q = 0,

(5.44)

∇2 ∇2 F + ∇2 w + N (w, w) = 0,

(5.45)

with the boundary conditions (5.25)–(5.28). A solution to the stationary problem is found via analysis of an additional nonstationary problem. For this aim, we introduce the internal and damping terms into (5.44), and we get: ∂w ∂2 w = −∇2 ∇2 w + N (w, F) + ∇2 F + 4q. +ε ∂r ∂τ2

(5.46)

Since the load q(r, θ) and the functions w, F, on the border do not depend on time, we can expect that a solution to the unstationary problem will be changed slowly in time and for τ → ∞ it will achieve a solution to the stationary problem. In order to realize a vibration damping, in equation (5.46) ε parameter is introduced. By a suitable choice of the damping parameter a stationary state has been achieved. The algorithm of the solution to the unstationary problem is given in the next section. The mentioned set-up method, does not require a special solution to the problems of statics. A solution can be easily obtained using the algorithm of computations for a dynamical problem. 5.2.2 Thermal Field and Set-Up Method Earlier, the partial differential equations (5.17), (5.18) are obtained with respect to the functions w = w(r, θ, t) and F = F(r, θ). The exact solution to these equations does not exist. However, there are methods reducing the approximate integration of equations for different boundary conditions. Here the finite difference method is applied. The initial differential equations are substituted by algebraic equations, and then they are solved using numerical methods. The finite difference method has a wide spectrum of approximating formulas. It slightly depends on the boundary value conditions, the shell’s geometry and on the initial stress state. It is also simple in realization and suitable for the programming purpose. However, high order of the algebraic equations system belongs to one of its drawbacks. Let a projection of the mean surface of the sectorial conical shell into the coordinate plane occupying the area Ω, bounded by the contour Γ Ω = Ω + Γ = {(r, θ) , 0 ≤ r ≤ a, 0 ≤ θ ≤ θk } . We introduce a polar mesh in Ω of the form

510


ω1 = {ri = ihr , 0 ≤ i ≤ (N − 1) , rN−1 = a} , ω2 = θ j = jhθ , 0 ≤ j ≤ (M − 1) , θN−1 = θk , where: ω = ω1 × ω2 , L = (M + 2)(N + 2) denotes a general nodes number, including outer contour nodes. For all derivatives, the finite difference approximation O(h2 ) is applied. We introduce the following notations of the mesh operators: Λr (·) = (·) r˙, Λrθ (·) = (·)θ˙ /ri2 − (·)r˙θ˙ /ri , Λrr (·) = (·)rr , Λθθ (·) = (·)θθ , Λzz (·) = (·)zz , Λrrθ (·) = (·)rrθ , Λz (·) = (·)z˙ , Λ (·) = Λθθ (·) /ri2 + Λr (·) /ri + Λrr (·) , where: (·)rr , (·)θθ , (·)r˙θ˙ , (·)zz , (·)r˙ are known in the literature notations of difference derivatives [376]. The initial equations (5.17), (5.18) and the three dimensional heat transfer equation (5.24) can be presented in the operator form: (wττ + εwτ )i j = −Λ (Λw) + Λrr w (ΛF − Λrr F) + Λrr FΛrr (Λw − Λrr w) − Λrθ wΛrθ F + ΛF + 4 (q)i, j − (1 + ν) ΛκT , √ Λ (ΛF) = −Λrr w (Λw − Λrr w) − (Λrθ w)2 − Λw − ηΛεT , (T τ )i, j,k = Λzz T + λΛT + w0 .

(5.47) (5.48) (5.49)

The boundary conditions (5.24)–(5.43) have the form: 1. Rolling clamping of the arcal element ν wN, j = 0, Λrr w + Λr w − (1 + ν) (κT )N, j = 0, a F N, j = 0, Λr F = 0, j = 1, M − 2.

(5.50)

2. Rolling clamping of the radial element wi, j = 0, Λθθ w + ri (1 + ν) (κT )i, j = 0, Fi, j = 0, Λθθ F = 0 for j = 0, j = M − 1,

i = 0, N − 1 .

(5.51)


511

3. Slip clamping of the arcal element wN, j = 0, Λr w = 0, F N, j = 0, Λr F = 0, j = 1, M − 2.

(5.52)

4. Slip clamping of the radial element wi, j = 0, Λθ w = 0, Fi, j = 0, Λθθ F = 0, j = 0, j = M − 1,

i = 0, N − 1 .

(5.53)

The boundary conditions with respect to displacements read: 1. Stiff clamping of the arcal element wN, j = uN, j = vN, j = 0, Λr w = 0, j = 1, M − 2.

(5.54)

2. Stiff clamping of the radial element wi, j = ui, j = vi, j = 0, Λθ w = 0, j = 0, j = M − 1,

i = 0, N − 1 .

(5.55)

3. Rolling unmovable clamping of the arcal element ui, j = υi, j = wi, j = 0, ν Λrr w + Λr w − (1 + ν) (κT )N, j = 0, j = 1, M − 2. a 4. Rolling unmovable clamping of the radial element

(5.56)

ui, j = υi, j = wi, j = 0, Λθθ w + r (1 + ν) (κT )i, j = 0, j = 0, j = M − 1, i = 0, N − 1.

(5.57)

Since boundary conditions (5.50)–(5.53) for a hybrid problem and the conditions (5.54)–(5.57) for the problems in displacements can be combined, the hybrid boundary conditions appear in the angle points. We consider the conditions in the angle points (0, 0), (w − 1, 0), (N − 1, M − 1) (see Fig. 5.3). Suppose that the A point is the angle point. Then, the C, E points lie outside the contour, and the D, A, B are the contour ones. On the contour, for an arbitrary combination of conditions, we have wA = wB = wD = 0 (∗). For a rolling and slip clamping in the point A additional two conditions should ; ; be satisfied, for example w = 0 and = 0. The operators in explicit form in the r

θθ

nodal points are as follows: wC = wB , wD − 2wA + wE = 0. Using condition (∗) for the angle point A, all of the nodes behind the contour have zero values:

512


Figure 5.3. External and internal shell nodes.

wC = 0,

wE = 0.

Similar conditions hold for the F function. In the case for θκ = 2π we get the so called problem with a cut along a radius. In order to get a circle problem, the sewing conditions should be formulated: wi, j = wi,M+ j , Fi, j = Fi,M+ j , j = 0; −1 i = 0, N − 1.

(5.58)

The boundary conditions (5.41), (5.42) for the three-dimensional heat transfer equation read: 1-st order T i, j,K = ψi, j,K K = 0, K = P − 1, 0 ≤ i ≤ N − 1, 0 ≤ j ≤ M − 1.

(5.59)

2-st order < T = q i, j,K K = 0, K = P − 1, 0 ≤ i ≤ N − 1, 0 ≤ j ≤ M − 1.

(5.60)

n

3-st order

0) or instability (β ≤ 0) on each step.

Figure 5.9. Sectorial shell scheme.

During investigations of shells parametric vibrations, in reference [120] another criterion is proposed. The input system of differential equations is reduced to the Mathieu-Hill’s equation, and then stability is estimated. There are no adequate criteria to define a stability loss of sectorial shells which will be proposed now. Following the scheme in Fig. 5.9, let OC = R be a sphere radius, whereas AC = a1 is the diameter of horizontal shell projection. Let the arc ABC is clamped along. The shell vertex B is always clamped, because it belongs to the arc ABC. The Kantor theorem [191], indicates that the point B can not move on amount of two sags BE = H1 . Let us analyse now the displacement of the point D being the center of arc BC. In accordance with the monograph [522], the non-dimensional curvature parameter for shallow shells is defined by the relation b1 a21 /(Rh), where h is the shell thickness. In the case of small elevation shells one gets: a21 /(Rh) ≈ 8H2 /h and H2 /H1 ≈ a22 /a21 , which yields the estimation H2 ≈ H1 /h. Assuming b1 as the shell sloping parameter (see [522]), for the spherical shells the following estimation holds: H1 = b21 /2, i.e. H2 = b21 /8. The obove considerations allow one to construct the following Table 5.2. Therefore, we have introduced the geometrical H2 parameter (Fig. 5.9). It is seen in Fig. 5.9, that for the clamped sector along the ABC contour, the top B is always clamped. Therefore, according to the Kantor [305] criterion, it can not move on 2H after the stability loss. We need to analyse a displacement of the D point, which is the middle point of the arc BC.

522

5 Dynamical Behaviour and Stability of Flexurable Sectorial Shells Table 5.2. Shell parameters obtained from the scheme given in Fig. 5.9. θk π/3 π/2 π b1 11 10 9 wmax 9.6 12.7 9.3 H2 15.1 12.5 10.1 w(H2 ) w > H2 /2 w ≥ H2

1.5π 2π 8 5 19.6 19.18 8 3.125 w ≥ 2H2

5.3.1 Influence of the Sector’s Angle The dynamical problems will be solved using the governing equations (5.17), (5.18) for ε = 0. For all sectors we consider the slip clamping of the edge (5.27), (5.28) for the following fixed parameters: M = N = 10, r s /h = 200, ν = 0.3. The stress strain state of the sectorial shells in critical and pre-critical states are investigated. An applied mesh consisted of radiuses and angles is shown in Fig. 5.10. Observe that r1 = 0 represents the vertex, whereas diagonal of sector θk is situated between θ5 and θ6 . PROBLEM 1. A shell with the sectorial angle θ k = 2π.

Figure 5.10. A shell sector partition by radiuses r1 ...r10 and angles θ1 ...θ10 .

In this case we deal with a circled shell, and the sewing conditions on should be satisfied. In order to determine a stability loss, we use the Volmir’s [684] and Shian et al. [612] criteria. Both of them are in good agreement and give the same values of the critical load. The investigations have shown that for the given boundary conditions the circular spherical segment loses its stability for b = 5. Both of the used criteria are illustrated in Fig. 5.11 and 5.12, respectively and they give the same values qcr = 0.24, wmax = 19.2.

5.3 Stability of Sectorial Shells with Finite Deflections

523

Figure 5.11. Analysis of stability using Volmir criterion.

Figure 5.12. Analysis of stability using Shian, Soong and Roth criterion.

For a stress state analysis of a circular segment we, compare the shell’s behaviour for the pre-critical load q = 0.2, and for the time instants τ = 2.35; 5.1; 9.5 (the curves 1, 2, 3), as well as for critical load qcr = 0.24 for the time instants τ = 2.35; 5.1; 10.5 (the curves 1, 2, 3). For the pre-critical load, a slow increase of the deflection and forces (Fig. 5.13), a slight change of the circular forces from the negative values in the vertex neighbourhood to the positive ones on the contour (Fig. 5.15) and only negative radial stresses action (Fig. 5.14) are noticed. For the critical load, in the stability loss instant (curve 3), a sudden increase of the deflection and a decrease of the forces (Fig. 5.16) are observed, and also a sign change of the circular and radial forces (curve 3) in the area of the shell’s top (Fig. 5.17, 5.18) is exhibited. Therefore, the sector θk = 2π loses its stability for b = 5, qcr = 0.24 with the occurrence of buckling in the shell’s vertex. Besides, the stability loss is characterized by a sudden variations of the forces, and the membrane forces change their sign. The shell’s fibres with compressing stresses now become the stretching ones.

524


Figure 5.13. Deflections and stress functions distribution along a radius (pre-critical loading).

Figure 5.14. Radial stresses distribution along a radius (pre-critical loading).

Figure 5.15. Circumferential stresses distribution along a radius (pre-critical loading).


525

Figure 5.16. Deflection and stress function distribution along a radius (critical loading).

Figure 5.17. Circumferential stresses distribution along a radius (critical loading).

PROBLEM 2.1. A shell with the sectorial angle θ k = 32 π. In order to determine a stability loss moment of a given shell we use the Shian et al. criterion [612], which is illustrated in Figs. 5.19, 5.20. According to it qcr = 0.34, wmax = 19.6, b = 8. Now we analyse the shell’s stress state. In Fig. 5.21 for the pre-critical load q = 0.3, the curves of relatively equal forces (left shading) and deflection (right shading) for which the maximal deflection is obtained are given. It is seen that the shell is dented along the angle bisectrix. A change

526


Figure 5.18. Radial stresses distribution along a radius (critical loading).

Figure 5.19. The dependence wmax (τ) for the shell θk = 1.5π for different q (Shian, Soong and Roth criterion).

of the circular and radial forces along the angle is smooth, and the largest forces are located on the bisectrix (Fig. 5.22). In the neighbourhood of the vertex, the stretching stresses occur (Fig. 5.23, 5.24). The other shell’s part exhibits radial compression stresses. For the critical load qcr = 0.34 the curves of equal relative deflections (left shading) and the forces (right shading) are different (Fig. 5.26). The shell loses its stability with the occurrence of a buckling concave on the bisectrix. For the critical load, the Nr character is changed along the angle (Fig. 5.25), which indicates a stability loss. Almost all radiuses on the bisectrix have positive stresses. This conclusion is supported by Fig. 5.26. Almost all stresses on the bisectrix of θ5 have a positive character. The bending stresses Mr (Fig. 5.28) and Mθ (Fig. 5.30) achieve maximal positive values on the bisectrix of the angle θ5 .


527

Figure 5.20. The dependence q(τ) for the shell θk = 1.5π (Shian, Soong and Roth criterion).

Figure 5.21. The curves of equal relative both stress function (a) and deflection (b) values in the maximal deflection time instant (pre-critical load q = 0.3).

528


Figure 5.22. Variation of circumferential and radial stresses along angle (pre-critical load q = 0.3).

Figure 5.23. Radial stresses distribution along a radius (pre-critical load q = 0.3).

Figure 5.24. Radial stresses distribution along a radius (critical load q = 0.34).


529

Figure 5.25. Circumferential stresses distribution along a radius (pre-critical load q = 0.3).


530


Figure 5.27. Variation of circumferential and radial stresses along angle (pre-critical load q = 0.34.

Figure 5.28. Bending radial moment distribution along a radius (critical load qcr = 0.34).


531

Figure 5.29. Sectorial shell deflection along a radius (pre-critical load q = 0.3).

Figure 5.30. Bending circumferential moment distribution along a radius (critical load qcr = 0.34).

532


Figure 5.31. Sectorial shell deflection along a radius (critical load qcr = 0.3).

It has been observed that for the critical load the maximal deflection is shifted to the shell’s vertex (Fig. 5.30). To conclude, the sector shell loses its stability for b = 8, qcr = 0.34 with the appearance of the buckling concave on the angle bisectrix close to the shell’s vertex. A stability loss is characterized by a sign change of Nr along the bisectrix θ5 .

Figure 5.32. Vibrations of different shell points after the impulse load removal.


Figure 5.33. Stress function F(τ) for different shell points.

Figure 5.34. Circumferential stresses Nθ of different shell points.

533

534


PROBLEM 2.2. Vibrations of the sectorial shell θ k = 32 π after the load removal. Let shell be subjected to the impulse with intensity q = 0.5 and its duration τ = 7.8. The shading area in all figures denotes the impulse action. We investigate the vibrations after the removal of the impulse. We trace the points 1-6 given in Fig. 5.32. With time increase, the points on the central radius and near the vertex have the largest amplitude. The point close to the arc segment possesses the smallest amplitude. A similar picture can be observed in Fig. 5.33, where the stresses exhibit a complicated vibrational process. It is seen from Fig. 5.33 that the stress function in the vertex points changes its sign in comparison to other points. The radial absolute stress value are largest near the vertex. In Fig. 5.34, the vibrations of circumferential stresses Nθ are presented. Observe that the stresses sign in the points of the arcal segment, and in other points are different. To conclude, when the shell is compressed at the vertex along the radius, then in the other points it is extended along the radius. At the same time, it is compressed along the angle of the arcal segment, whereas in the other points it is extended. PROBLEM 3.1. A shell with θ k = π sector. In order to define a stability loss instant the Shian et al. [612] criterion will be used (Fig. 5.35, 5.36). According to it, qcr = 0.3, wmax = 9.3, b = 9. The stress state will be analysed for the pre-critical load q = 0.2, and the critical one qcr = 0.3 at the time instant of achieving a maximal deflection. For the pre-critical load (τ = 4.7) the curves of equal relative deflections, and the function of stresses show the most deformable places. It occurs on the θ5 bisectrix along the radius r7 . The radial stresses have a negative sign (Fig. 5.36). The largest Nr are obtained on the r7 radius. The circular stresses are interleaved with the zones of stretching and compressing (between the radiuses r2 -r8 ) stresses (Fig. 5.38). In the stability loss time instant, for the critical load, two buckling concaves lying symmetrically along the θ5 bisectrix have appeared. This situation is presented in Fig. 5.40, where the curves of equal relative deflections and the stress function are given (right and left shading, respectively). The maximal deflection are shifted to the shell’s vertex on r6 (Fig. 5.41). The circular stresses (Fig. 5.42) have a zone of sign interleave, but they are smaller in comparison to the central part. The stability loss time instant, is observed on the radial stresses behaviour. The stresses which occurred on θ4 , θ5 have the wave-shape character (Fig. 5.43). A decreasing of the radial stresses on θ4 between the radiuses r4 -r7 , and also the occurrence of maximal deflection stresses in that place (Fig. 5.44, 5.45) characterize the stability loss with the occurrence of two symmetric concaves. To conclude, the sectorial shell with the sector’s angle θk = π loses its stability for b = 9, q = 0.3 with the occurrence of two buckling concaves.


Figure 5.35. The dependence wmax (τ) for different q and b values.

Figure 5.36. The dependence q(τ).

535

536


Figure 5.37. The curves of equal relative both stress function (a) and deflection (b) values in the maximal deflection time instant (pre-critical load and τ = 4.7).




537

Figure 5.40. The curves of equal relative both stress function (a) and deflection (b) values in the maximal deflection time instant (pre-critical load).



538


Figure 5.43. Radial stresses distribution along a radius (critical load qcr = 0.3).




539

PROBLEM 3.2. Vibrations of a sectorial shell with θ k = π after the load removal. Assume that the shell is subjected to the transversal impulse with q = 0.5 intensity and τ = 1.9 duration. We investigate the vibrations of the points 1-6, presented in Fig. 5.46, after the load’s removal. The vibrations of the deflection and stresses are qualitatively similar (Figs. 5.46, 5.47).

Figure 5.46. Vibrations of different shell points.

Most often, the vibrations occur on the vertex but they have small amplitude. A large amplitude characterizes the points lying close to the angle bisectrix. In Fig. 5.48 the vibrations of radial stresses are presented. An interesting behaviour is observed in the case of points 3, 4. Although they lie on one radius, after the load’s removal their radial stress vibrations are shifted to the phase of π. In the Fig. 5.49 the vibrations of circular stresses are shown. After the load’s removal, in all points the radial compress stresses remain for a certain time. In the case of circular stresses after the load’s removal, the stretching stresses occur on the arcal segment. They vibrate in the antiphase manner in comparison to other points. To conclude, the vibrations of sectorial shell points with θk = π are more complicated in comparison to the vibrations of the shell θk = 32 π.

540


Figure 5.47. Stress function time histories in different shell points.

Figure 5.48. Time histories of radial stresses for the shell θk = π.


Figure 5.49. Time histories of circumferential stresses for the shell θk = π.

541

542


PROBLEM 4.1. A shell with the sector’s angle θ k = π2 . Again the Shian et al. criterion will be used to estimate the stability loss moment (Fig. 5.50, 5.51). According to it, we get qcr = 0.3, wmax = 12.73, b = 10.

Figure 5.50. Dependence wmax (τ) for different q value.

We investigate the stress strain state of the shell for the pre-critical load q = 0.2 and the critical one qcr = 0.3 in the instant for achieving a maximal deflection. The occurrence of buckling convexity on the angle bisectrix characterized the pre-critical load action (Fig. 5.52). The deflection and the stress functions are similar. A distribution of the circular stresses along θ1 -θ5 are shown in Fig. 5.54. Similar to the case θk = π, on the shell’s vertex and in the arcal segment, the areas of positive stresses appear. Their magnitude on the arcal segment is larger than that on the vertex. It is caused by decrease of the angle θk . The area of negative values is narrowed and is shifted to the arcal segment. In Fig. 5.55, a change of the radial stresses along θ1 -θ5 is presented. The whole shell, except for a small area on θ2 , is in the radially compressed state.


543

Figure 5.51. Dependence q(τ) for the shell θk = π/2.

Figure 5.52. Displacement w and stress function F distributions along an angle (pre-critical load q = 0.2).

The characteristic load is characterized by two zones of maximal values of the stress function (Fig. 5.56, the left shading), but the shell loses its stability by the occurrence of one dent on the angle’s bisectrix (the right shading). It has been found that increasing the load q = 0.3 the maximal deflection is shifted to the vertex (Figs. 5.57, 5.53). The stability loss (buckling) instant is characterized by the wave-form behaviour of the curves Nθ (Fig. 5.58) and Nr (Fig. 5.59). In the place on θ5 a change of the radial stresses occurs. The stability change is displayed also on the pictures of the change of deflection stresses Mθ (Fig. 5.60) and Mr (Fig. 5.61). The maximum of the deflection stresses is achieved on θ5 , where the dent occurs. To conclude, the sectorial shell θk = π2 loses its stability for qcr = 0.3, b = 10 with the occurrence of one dent located on the angle’s bisectrix.

544


Figure 5.53. Deflection w distribution along a radius (pre-critical state q = 0.2). N

q = 0.2 5

4

5 3

3

2

0

3

−3

0 (r1 )

1/3 (r4 )

4

2/3 (r7 )

r/a

1 (r10 )

Figure 5.54. Circumferential stresses distribution along a radius (pre-critical state q = 0.2).



545



546






547


PROBLEM 4.2. Vibration of a sectorial shell θ k =

π 2

after the load removal.

Let the shell be subjected to the transversal impulse with intensity q = 0.5 and duration τ = 2.7. We trace the vibrations of the points 1-6 presented in Fig. 5.64. Vibrations of the function w, F are given in Figs. 5.62, 5.63. They have a complicated character. In spite of the fundamental vibrations, also small vibrations occur. The computations have displayed that the vertex points’ vibrations have a slight influence. Large amplitude vibration appears in the point 4. The vibrations of Nr (Fig. 5.64) and Nθ (Fig. 5.65) stresses have complicated forms. After the load’s removal in all points the radial compressing stresses initially occur. Then the zones of stretching and compression are interleaved. After the removal of the load, both points of circular segment and at the vertex exhibit positive circumferential stresses, and vibrations have complicated forms.

548


Figure 5.62. Vibrations of different shell points after the impulse load removal.

Figure 5.63. Stress function F(τ) for different shell points.


Figure 5.64. Radial stresses Nr of different shell points.

Figure 5.65. Circumferential stresses Nθ of different shell points.

549

550


PROBLEM 5.1. A shell with the sector’s angle θ k = π3 . Similarly to the previously considered cases we use the Shian et al. criterion to estimate the stability loss (Figs. 5.66, 5.67) (qcr = 0.25, wmax = 9.6, b = 11).

Figure 5.66. Dependence wmax (τ) for different q values.

Figure 5.67. Dependence q(τ) for the shell θk = π/3.

We investigate the stress state for the pre-critical load q = 0.2, and the critical one qcr in the time instant of the maximal deflection occurrence. For the pre-critical load, a change of the w, F against the angle are qualitatively similar (Fig. 5.68). The maximal deflection appears on r8 (Fig. 5.69) in the vicinity of the arcal segment. For the circular stresses the sign interleave occurs. Maximal stresses are observed on the vertex (similarly to the cases θk = 3π 2 , π), Fig. 5.70.


551

Figure 5.68. Displacement w and stress function F distributions along an angle (pre-critical load q = 0.2).

Figure 5.69. Deflection w distribution along a radius (pre-critical state q = 0.2).

The radial stresses are only compressive (Fig. 5.71), and the smallest of them appears in the vicinity of the radial segment on θ2 . In the case of the critical load, the maximal deflections on r7 are shifted (Fig. 5.72). The pictures of relatively equal deflections and stresses are similar qualitatively (Fig. 5.73). A stability loss occurs with a dent occurrence on the angle’s bisectrix. The stability loss moment is characterized by a wave-form change of the circular stresses (Fig. 5.74), and a sign change of the radial stresses on θ5 (Fig. 5.75). A behaviour of the bending stresses along θ1 -θ5 is presented in Fig. 5.76 for Mθ , and in Fig. 5.77 for Mr . Only on θ5 the maximal values of Mr and Mθ are achieved. Therefore, the sectorial shell θk = π3 loses its stability for b = 11, qcr = 0.25 with the occurrence of one dent on the angle’s bisectrix.

552


Figure 5.70. Circumferential stresses distribution along a radius (pre-critical state q = 0.2).


Figure 5.72. Deflection w distribution along a radius (pre-critical state q = 0.25).


553


Figure 5.74. Distribution of circumferential stresses along a radius (pre-critical load q = 0.25).

Figure 5.75. Distribution of radial stresses along a radius (pre-critical load q = 0.25).

554


Figure 5.76. Circumferential bending moments distribution along a radius in the stability loss time instant.

Figure 5.77. Distribution of radial bending moments along a radius in the stability loss time instant.


PROBLEM 5.2. Vibrations of a sectorial shell θ k = load.

π 3

555

after removing of the

Let a shell be subjected to the transversal load of the intensity q = 0.5 and duration τ = 2.3. We trace vibrations of the points 1-6 (Fig. 5.78) after removing the load. The deflection functions are small in the top neighbourhood during the vibrations.

Figure 5.78. Vibrations of marked shell points for the shell θk = π/3.

The point 6 lying on the angle’s bisectrix possesses the largest amplitude and the lowest vibration frequency. All the points undergo the same vibrations of stresses F (Fig. 5.79). The vibrations of the radial Nr and the circular Nθ stresses are shown in Figs. 5.80, 5.81, respectively. After removing the load, the compression area along the radius is changed by a stretching area. The largest amplitude is obtained for point 4. For the circular stresses a zone of stretching stresses occurs in the shell’s centre, whereas at the vertex neighbourhood and on arcal segment, a zone of stretching occurs. Further, the anti-phase vibrations occur. It has been observed that vibrations amplitudes of Nθ function in the vertex neighbourhood points are higher than Nr amplitudes. However, for the whole vibrational process the Nθ values will be smaller than Nr values.

556


Figure 5.79. Time histories of the stress function for the chosen shell points.

Figure 5.80. Time histories of the radial stresses Nr for the chosen shell points.


557

Figure 5.81. Time histories of the circumferential stresses Nθ for the chosen shell points.

Figure 5.82. Critical and pre-critical values

Analysing the results, obtained in problems 1-5, the following conclusions are derived. A. Dynamical stability of sectorial shells 1. The Shian et al. criterion has been successfully used for the stability loss time instant determination for the sectorial shells with θk ∈ [ π3 , 2π]. a) the sector’s angle influence on the critical sloping parameter b (when the shell cases its stability) and on the critical load; b) the shell θk = 32 π, θk ∈ [ π3 , π] loses its stability with the occurrence of one dent on the angle’s biserctix (the shell θk = π loses its stability with the

558


Figure 5.83. Maximal deflection and shell centre deflection versus θk .

Figure 5.84. Maximal deflection distribution along a radius for a pre-critical load.

occurrence of two dents symmetrically located in relation to the angle’s biserctix). c) the stability loss instant is characterized by a sudden change of the membrane stresses (circular stresses have wave-forms, and the radial ones change their sign), where a dent occurs. 2. For the angles θk ∈ ( π3 , π2 ), the time of achieving a maximal deflection for the critical load increases, and for θk ∈ ( π2 , π) it decreases, Fig. 5.82 (the curve 1 corresponds to the pre-critical, where the curve 2 corresponds to the critical loads). For the angles θk ∈ (π, 32 π), at first the time increases, but beginning from θk = 32 π, it decreases again. All figures have been obtained for critical b parameters. 3. A maximal deflection of sector shells is achieved out of the centre, Fig. 5.83 (the curves 1, 2 for the pre-critical load correspond to the maximal deflection in the centre; curves 3, 4 are obtained also for the critical load). Increasing the angle the maximal deflection shifts to the shell’s centre (see also Fig. 5.84 for the pre-critical load, and Fig. 5.85 for the critical one). For the pre-critical loads,


559

Figure 5.85. Maximal shell deflection dependence on a radius for a critical load.

Figure 5.86. Maximal circumferential shell stresses along a radius (a pre-critical load).

the shell is dented in the vicinity of the arcal segment between r6 -r8 , whereas for the critical ones - in the vertex vicinity between r5 -r7 . 4. For θk ∈ (π, 32 π) in the shell’s vertex neighbourhood stretching circular stresses are observed, which are two times larger than those occurring on the arcal segment. On the contray, for angles θk ∈ ( π2 , π) stretching circular stresses on the vertex are small, and the stresses on the arcal segment are higher more than

560


twice (Fig. 5.86). Decreasing the angle, the area of positive Nθ increases on the vertex. A zone of stretching stresses is shifted to the arcal segment, causing a sudden increase of positive stresses on that segment. Therefore, the shells with small sectors angle are more stable against the external load. In order to obtain unstability, the parameter b should be increased. B. Vibration of sectorial shells after load removing. 1. Decreasing the angle values leads to the complexity of vibrations of different shell’s points. Together with the fundamental ones, also small vibrations of different phase occur. 2. Decreasing θk , the vertex vicinity points have very small deflection values, and their vibration frequencies are small. 3. For θk ∈ [ π2 , 32 π] the points of the shell’s centre have a maximal vibration amplitude (points 3, 4). For θk ∈ [ π3 , π2 ], the maximal amplitude occurs for point 6 and it lies on the biserctix of the arcal segment angle. 5.3.2 Set-Up Method and Determination of Critical Loads For computation purpose, we use the equations (5.45) and (5.46). In order to solve the differential equations, the algorithm presented in Sect. 5.2.2 is applied, and the thermal field is not taken into account. PROBLEM 1. Sectorial shell with the sector’s angle 2π. Let a shell be subjected to a uniform external load (pressure) with intensity of q. The shell’s edge has slip clamping (5.52), and the radial segments served using the formulae (5.58). The initial conditions (5.40) are equal to zero. The following fixed parameters have been taken: ν = 0.3, a/h = 200, ∆τ = 0.01, N = 10, ε = 2. The obtained results are compared with those obtained in reference [696], showing very high accuracy. PROBLEM 2. Sectorial shell θ k = 32 π. We consider a shell with the following initial data: uniform external pressure q; slipping clamping of arcal and radial segments (5.52), (5.53); zero initial conditions (5.40); ν = 0.3, ε = 2, M = N = 10, a/h = 200, ∆τ = 0.01. In Fig. 5.87, a dependence between the load parameter and the maximal normal shell’s displacement for different parameter b = 5, 6, 7, 8 is presented (curves 1, 2, 3, 4, respectively). Beginning from certain b values, on the curves the limiting points occur. The calculations have proved that b = 8 in the critical value. The first limiting point on the curves 4 defines the upper critical load q+ = 0.5, when a “jump” occurs. In Fig. 5.88, the set up curves of the unstationary solutions for b = 5, q = 0.3, 0.8, 1.4, 2 (curves 1-4, respectively) are presented. The computations have shown, that a good


561

Figure 5.87. Dependence q(wmax ) for different slopings b.

Figure 5.88. Isoclines of deflection and wmax (τ) for b = 5.

choice of the damping parameter ε quarantees a smooth transition on the stationary solution with the accuracy of 0.005. In Fig. 5.89, the curves of relatively equal deflections for the shell with b = 6 (to the left) and b = 7 (to the right), and for b = 8 in Fig. 5.90, are displayed. We compare the curves behaviour for the same load values. For q = 0.3, the pictures qualitatively coincide. A maximal deflection is achieved on the intersection of θ5

562


Figure 5.89. Curves of equal relative shell deflections for b = 6, 7 for different loads q.

bisectrix and the shell’s central radius. For q = 0.5, the pictures are different: for b = 7 two zones of maximal deflections occur, symmetrically situated in relation to θ5 ; for b = 8 also to zones of maximal deflections are observed. For the load q = 0.6, the maximal deflection for the shell with b = 6 is not clearly expressed, whereas for b = 7 and b = 8 the zones of maximal deflections increase and a stability loss occurs.


563

Figure 5.90. Curves of equal relative shell deflections for b = 8 for different loads q.

The largest deflection zone occurs for b = 8, and this parameter is considered as the critical one for θk = 32 π. The shell’s stress state with the critical parameter b = 8 is presented in Figs. 5.91–5.94 for the pre-critical load q = 0.3, whereas in Figs. 5.95–5.98 - for the critical one q+ = 0.5. The computations shown that maximal deflections for each θ1 -θ4 are achieved on r6 for either critical or pre-critical loads (Figs. 5.92, 5.96). The circular stresses are distributed in the following form. The stretching stresses are located on the vertex vicinity and on the arcal segment (Figs. 5.93, 5.97). A zone of compressing stresses Nθ is located between the r2 -r8 radiuses. A zone of positive radial stresses is located only in the top vicinity on the radiuses r1 -r4 (Figs. 5.94, 5.97). Between radiuses r4 -r10 , a zone of compressing radial stresses occurs.

564


Figure 5.91. Deflection w and stress function F distributions along angle for the critical load.

Figure 5.92. Shell deflection along a radius for the pre-critical load.

Therefore, for the shell with the central angle θk = 32 π, a stability loss occurs for b = 8 and q+ = 0.5. The shell loses its stability with the occurrence of two dents situated symmetrically in relation to the angle’s biserctix.


Figure 5.93. Circumferential stresses along a radius for pre-critical load q = 0.3.

Figure 5.94. Radial stresses along a radius for pre-critical load q = 0.3.

565

566


Figure 5.95. Deflection w and stress function F along an angle for critical load qcr = 0.5.

Figure 5.96. Deflection w along a radius for critical load q = 0.5.

Figure 5.97. Circumferential stress along a radius for critical load.


567

Figure 5.98. Radial stress along a radius for critical load.

PROBLEM 3. A shell with a central sector’s angle π. We analyse shell with the same parameters.

Figure 5.99. Shell sloping parameter versus θk (a) and load q versus maximal displacement for b = bcr (b).

568


Figure 5.100. Curves of equal relative shell displacements for b = 5 and b = 9 for different q values.

Figure 5.101. Curves of equal radial (a) and circumferential (b) stresses for a post-critical load q = 0.5.

In Fig. 5.99a is reported that for two shell θk = π the corresponding bcr = 9. The function q(wmax ) for the critical sloping parameter is shown in Fig. 5.99b. The dashed curves represent sets of critical points. For instance, the upper limit of the critical load is equal to q+ = 0.4 (curve 3 in Fig. 5.99b).


569

Figure 5.102. Deflection distribution along a radius for pre-critical load.

Figure 5.103. Deflection distribution along a radius for critical load.

In Fig. 5.100 the curves of equal relative deflections for b = 5 (to the left) and b = 9 (to the right) for different values of the transversal load q are presented. For b = 5 a maximal deflection is achieved on the intersection of the biserctix with the radius. For b = 9 the deflections picture is changed. For q = 0.3 the maximal deflection is still achieved on the biserctix, but beginning with q+ = 0.4 a zone of maximal deflection is divided into two symmetrically located in relation to θ5 . Further increase of the load leads to increase of wmax . A stress state of the shell is presented in Figs. 5.102–5.109 for the pre-critical load q+ = 0.4, and in Figs. 5.110–5.113 for the post-critical load q = 0.5. The maximal deflections in θ2 -θ5 zone are achieved on r7 (Figs. 5.102-5.104). The computations show that for q = 0.3 and q = 0.4 the curves of circular and radial stresses are qualitatively similar. A distribution of the circular stresses are characterized by the occurrence of two zones with a positive sign in the vertex vicinity and on the circular segment. A zone of compressing stresses is located between r2 -r8 (Figs. 5.104-5.107). For the radial stresses, a zone with a positive sign is not observed.

570


Figure 5.104. Deflection w and stress function F along a radius for the critical load.

Figure 5.105. Circumferential stresses along a radius for the pre-critical load q = 0.3.

In a zone from r1 to r10 , the all radial stresses are compression. The largest Nr are achieved on θ2 , θ3 on the radiuses r6 , r7 (Figs. 5.105, 5.106). Another picture is observed for the post-critical load q = 0.5. From r3 to r9 the deflection peaks occur on θ3 , which correspond to the places of two maximal


571

Figure 5.106. Circumferential stresses along a radius for the critical load qcr = 0.4.

deflection zones symmetrically located in relation to θ5 . The radial stresses behave in a different manner (Fig. 5.111). A distribution of Nr along θ2 , θ3 has a waveform character. In the planes, where the dents appear, the radial stresses absolute values rapidly decrease, which

572


Figure 5.107. Radial stresses along a radius for the pre-critical load q = 0.3.

Figure 5.108. Radial stresses along a radius for the critical load qcr = 0.4.

corresponds to a stability loss. For the post-critical load q = 0.5, on θ1 , θ4 , maximal compressing stresses are achieved. A behaviour of the circular stresses in relation to θi is presented in Fig. 5.113. The maximal positive stresses are achieved in the vertex vicinity, and on the arcal


573

Figure 5.109. Radial and circumferential stresses along a radius for the critical load qcr = 0.4.

segment. The vibrations appear on θ2 and θ3 . In the places of dents, the absolute values of Nθ rapidly decrease (except for θ1 , θ4 ). A distribution of the stresses Nr and Nθ along each radius ri is presented in Fig. 5.112. Beginning from r3 , the curves ‘start to vibrate’ (in particular the curves 5, 6, 7). The smallest values are observed between θ3 and θ4 (the place of the maximal deflections).

574


Figure 5.110. Displacement w and stress function F distributions along an angle for the postcritical load q = 0.5.

Figure 5.111. Radial stresses distribution along a radius for the post-critical load.

The curves of relative equal stresses are presented in Fig. 5.114 for q = 0.5. The curves behaviour indicates the occurrence of two symmetrical dents.


575

Figure 5.112. Radial and circumferential stresses distribution along an angle for the postcritical load q = 0.5.

Therefore, a stability loss for the shell with θk = π occurs for b = 9 and q+ = 0.4. The shell loses its stability with the occurrence of two dents symmetrically located in relation to the angle’s biserctix. The membrane stresses on θ2 , θ3 have a waveform character.

576


Figure 5.113. Circumferential stresses along a radius for the post-critical load q = 0.5.


577

PROBLEM 4. A shell with θ k = π2 . We consider a shell with the same parameters boundary and initial conditions (only ∆τ = 0.001). The computations have shown (Fig. 5.99 - curve 2) that b = 10 corresponds to the stability loss. The first limiting point on the curve corresponds to the limiting critical load q+ = 0.35.

Figure 5.114. Deflection w and stress function F distributions along an angle for the precritical load q = 0.3.

Figure 5.115. Deflection w along a radius for the critical load qcr = 0.35.

578


Figure 5.116. Deflection w and stress function F distributions along an angle for the postcritical load q = 0.6.

For the loads q = 0.3; 0.35; 0.6 the maximal deflections are achieved on the intersection of r7 with θ4 (see Figs. 5.114–5.116). Therefore, a stability loss form corresponds to one dent on the angle’s bisectrix. For each of the loads qi = 0.3; 0.35, the curves’ Nθ and Nr character is similar in relation to θi and ri (Figs. 5.117–5.120). Contrary to the sectorial shells θk = 32 π; π, for θk = π2 a zone of the stretching circular stresses is increased up to r4 , whereas a zone of the stretched circular stresses is narrowed. Maximal circular stresses appear on the arcal segment. The greatest negative radial stresses are achieved on θ3 , θ4 .


579


Figure 5.118. Radial stresses distribution along a radius (critical load qcr = 0.35).

For the post-critical load q = 0.6, the behaviour is changed. The curves of stresses are wave-forming (Fig. 5.122). The occurrence of positive radial stresses zone between r5 and r10 (Fig. 5.123) for θ5 corresponds to the shell’s extension along the bisectrix. Between r5 -r7 , along the central radius, the radial stresses achieve a minimum (close to zero) and it is subjected only to the radial stresses action (Fig. 5.124). A distribution of relative equal radial stresses for q = 0.3 and q = 0.6 are presented in Fig. 5.125. For q = 0.6, on the biserctix, a zone of stretching radial stresses occurs, yielding its bending. To conclude, for the shell θk = π2 a stability loss occurs for b = 10 and q+ = 0.35, and it corresponds to the occurrence of the dent on the angle’s biserctix.

580


Figure 5.119. Circumferential stresses distribution along a radius for the pre-critical load q = 0.3.


581

Figure 5.120. Circumferential stresses distribution along a radius for the critical load qcr = 0.35.

Figure 5.121. Radial and circumferential stresses distributions along an angle for the precritical load q = 0.3.

582


Figure 5.122. Radial and circumferential stresses distributions along an angle for the postcritical load q = 0.6.


583

Figure 5.123. Radial stresses along a radius for the post-critical load q = 0.6.

Figure 5.124. Circumferential stresses distribution along a radius for the post-critical load q = 0.6.

584


Figure 5.125. Curves of equal relative deflections for pre-critical (q = 0.3) and post-critical (q = 0.6) loads.

Figure 5.126. Curves of equal relative radial stresses for pre-critical (q = 0.3) and postcritical (q = 0.6) loads.

PROBLEM 5. A shell θ k = π3 . The same (as in the previous case) initial data are used (∆τ = 0.0005). Fig. 5.99 (curve 1) show that the shell loses its stability for b = 11 (q+ = 0.3). For pre-critical, critical and post-critical loads q = 0.1; 0.3; 0.4 maximal deflections on θ1 -θ4 are achieved on r8 (Figs. 5.127, 5.130, 5.134, 5.137). The pictures of


585

Figure 5.127. Deflection w distribution along a radius for the pre-critical load q = 0.1.

Figure 5.128. Deflection w and stresses function F distributions along an angle for the critical load qcr = 0.3.

circular and radial stresses for q = 0.1; 0.3 are qualitatively the same (Figs. 5.128, 5.129, 5.131–5.133). In comparison to θk = π2 , the radial stresses in the vertex vicinity become smaller. All radial stresses have a negative sign. For the post-critical load q = 0.4 symmetrically located dents are not observed. It happened because the radial stresses on θ2 , θ3 achieve large negative values (Fig. 5.135), which does not allow the shell to buckle. In the vicinity of the circular segment, on θ4 a zone of the smallest compressing stresses is observed. Then, a buckling occurs on θ5 , and the dent is shifted more closely to the arcal segment (Fig. 5.139) in comparison to θk = π2 . Therefore, for the post-critical load the maximal deflections on θ1 -θ4 are shifted to r7 (Fig. 5.130).

586


Figure 5.129. Deflection w and stresses function F distributions along an angle for the postcritical load qcr = 0.4.

Figure 5.130. Deflection along a radius for the post-critical load q = 0.4.

A distribution of Nr and Nθ along the angle is shown in Fig. 5.138 (q = 0.3) and in Fig. 5.139 (q = 0.4). Beginning from r6 and for q = 0.4, the curves ‘start to vibrate’, which characterises a stability loss.


587


Figure 5.132. Circumferential stresses distribution for the pre-critical load q = 0.1.

Figure 5.133. Radial stresses distribution for the critical load qcr = 0.3.

To conclude, for the sectorial shell θk = π3 looses stability for b = 11 and q+ = 0.3. Observe that the stability loss is the same as in previously discussed cases for θk = π/3 and θk = π/2 (compare the Figs. 5.125 and 5.137).

588


Figure 5.134. Circumferential stresses distribution along a radius for the critical load qcr = 0.3.


589

Figure 5.135. Radial stresses distribution along a radius for the post-critical load q = 0.4.

Figure 5.136. Circumferential stresses distribution for the post-critical load q = 0.4.

590


Figure 5.137. Curves of equal relative deflections for the critical (qcr = 0.3) and post-critical (q = 0.4) loads.

Figure 5.138. Radial and circumferential stresses along a radius for the critical load qcr = 0.3.


591

Figure 5.139. Radial and circumferential stresses along a radius for the post-critical load q = 0.4.

5