Robotic Welding, Intelligence and Automation: RWIA'2010

Lecture Notes in Electrical Engineering Volume 88 Tzyh-Jong Tarn, Shan-Ben Chen, and Gu Fang (Eds.) Robotic Welding,...

Author: Tzyh-Jong Tarn | Shan-Ben Chen | Gu Fang

437 downloads 2353 Views 10MB 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

Lecture Notes in Electrical Engineering Volume 88

Tzyh-Jong Tarn, Shan-Ben Chen, and Gu Fang (Eds.)

Robotic Welding, Intelligence and Automation RWIA’2010

ABC

Tzyh-Jong Tarn Cupples 2, Room 103, Campus Box 1040 Washington University St. Louis, Missouri 63130 USA E-mail: [email protected]

Gu Fang School of Engineering University of Western Sydney Locked Bag 1797 NSW 1797 Australia E-mail: [email protected]

Shan-Ben Chen Intelligentized Robotic Welding Technology Laboratory School of Materials Sci. and Engg 800 Dongchuan Road Shanghai, 200240 P.R. China E-mail: [email protected]

ISBN 978-3-642-19958-5

e-ISBN 978-3-642-19959-2

DOI 10.1007/978-3-642-19959-2 Lecture Notes in Electrical Engineering

ISSN 1876-1100

Library of Congress Control Number: 2011923801 c 2011 Springer-Verlag Berlin Heidelberg 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 any other way, 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. Violations are liable to prosecution under the German Copyright Law. 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. Typeset & Coverdesign: Scientific Publishing Services Pvt. Ltd., Chennai, India. Printed on acid-free paper 987654321 springer.com

Preface

Robotic welding systems can be used in all types of manufacturing. They can provide several benefits in welding applications. The most prominent advantages of robotic welding are precision and productivity. Another benefit is that labor costs can be reduced. Robotic welding also reduces risk by moving the human welder/operator away from hazardous fumes and molten metal close to the welding arc. The robotic welding system usually involves measuring and identifying the component to be welded, welding it in position, controlling the welding parameters and documenting the produced welds. To develop an intelligent robotic welding system that can accomplish useful tasks without human intervention and perform in the unmodified real-world situations that usually involve unstructured environments and large uncertainties, the robots should be capable of determining all the possible actions in an unpredictable dynamic environment using information from various sensors such as computer vision, tactile sensing, ultrasonic and sonar sensors, and other smart sensors. From the existing successful applications, it can be concluded that emerging intelligent techniques can enhance and extend traditional robotic welding. This volume is mainly based on the papers selected from the 2010 International Conference on Robotic Welding, Intelligence and Automation (RWIA’2010), Oct. 14–16, 2010, Shanghai, China. We have also invited some known authors as well as announced a formal Call for Papers to several research groups related to welding robotics and intelligent systems to contribute the latest progress and recent trends and research results in this field. The primary aim of this volume is to provide researchers and engineers from both academic and industry with up-to-date coverage of new results in the field of robotic welding, intelligent systems and automation. The volume is divided into four logical parts containing twenty-five papers. In Part 1 (14 papers), the authors deal with some intelligent techniques for robotic welding. In Part 2 (16 papers), the authors introduce the Sensing of Arc Welding Processing. Various applications such as vision sensing and control of welding process are discussed. In Part 3 (9 papers), the authors describe their work on Modeling and Intelligent Control of Welding Processing. In Part 4 (8 papers), the authors exhibit their works on Welding Technics and Automations. In Part 5 (8 papers), the authors introduce some Special Robot Technology and Systems. Finally, in Part 6 (4 papers), the authors introduce some emerging intelligent techniques and their applications, which may contribute significantly to the further development of intelligent robotic welding systems.

VI

Preface

We would like to thank Professors R.C. Luo, J.L. Pan, B.S. Xu, S.Y. Lin, H.G. Cai, Z.Q. Lin, T.H. Song, L. Wu, P. Shan and Y.X. Wu for their kind advice and support to the organization of the RWIA’2010 and the publication of this book; to Na Lv, Ming-yan Ding, Hong-bo Ma, Shan-chun Wei, Zhen Ye, Yan-ling Xu, Huan-wei Yu, Tao Zhang, Ji-Yong Zhong, Cheng-dong Yang for their precious time to devote all RWIA’2010 correspondences and to reformat the most final submissions into the required format of the book, last but not least to Dr. Thomas Ditzinger for his advice and help during the production phases of this book.

October 2010

Tzyh-Jong Tarn, Washington University at St. Louis, USA Shan-Ben Chen, Shanghai Jiao Tong University, China Gu Fang, The University of Western Sydney, Australia

Contents

Part I: Intelligent Techniques for Robotic Welding Research Evolution on Intelligentized Technologies for Robotic Welding at SJTU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S.B. Chen Image Processing for Automated Robotic Welding . . . . . . . . . . . Peter Seyffarth, Rainer Gaede

3 15

Automatic Seam Detection and Path Planning in Robotic Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kevin Micallef, Gu Fang, Mitchell Dinham

23

Error Compensation and Calibration of Inter-section Line Welding Robot Based on a Wavelet Neural Network . . . . . . . . . Su Wang, Xingang Miao, Yuan Yang, Xingai Peng

33

Autonomous Seam Acquisition and Tracking for Robotic Welding Based on Passive Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shanchun Wei, Meng Kong, Tao Lin, Shanben Chen

41

A Framework of Intelligent Remanufacturing System Based on Robotic Arc Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ziqiang Yin, Guangjun Zhang, Hongming Gao, Huihui Zhao, Lin Wu

49

A Fast GPI Line Detection Method for Robot Seam Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bowen Li, Shengqi Tan, Wenzeng Zhang

57

Mechanism Design and Kinematics Modeling of Irregular Cross-Section Pipeline Welding Robot . . . . . . . . . . . . . . . . . . . . . . . Xinghua Tao, Hongwu Zhu, Lili Xu

65

VIII

Contents

Combined Planning between Welding Pose and Welding Parameters for an Arc Welding Robot . . . . . . . . . . . . . . . . . . . . . . . Huanming Chen, Yichen Meng, Xiaofeng Wang Optimal Digital Filtering for Tremor Suppression in a Master-Slave Robot Remote Welding System . . . . . . . . . . . . . . . . Hongtang Chen, Haichao Li, Hongming Gao, Lin Wu, Guangjun Zhang

73

81

Coordinated Motion of Different Weld Robots Based on User Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Huajun Zhang, Chunbo Cai, Guangjun Zhang, Daming Shen

87

Research on Information Transmission of a Welding Robot Based on Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . H.B. Wang, G.H. Ma, D.H. Liu, B.Z. Du

97

Survey on Modeling and Controlling of Welding Robot Systems Based on Multi-agent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Chengdong Yang, Shanben Chen Research on the Robotic Arc Welding of a Five-Port Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Jiyong Zhong, Huabin Chen, Shanben Chen Mixed Logical Dynamical Model for Robotic Welding System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Hongbo Ma, Shanben Chen

Part II: Sensing of Arc Welding Processing A Method of Seam Tracking Based on Passive Vision . . . . . . . . 131 Long Xue, Lili Xu, Yong Zou Research on a Trilines Laser Vision Sensor for Seam Tracking in Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Zengwen Xiao An Optimal Design of Multifunctional Vision Sensor System for Welding Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 Yanling Xu, Xiangfeng Kong, Shanben Chen Wire Extension Control Based on Vision Sensing in Pulsed MIG Welding of Aluminum Alloy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Lihui Lu, Ding Fan, Jiankang Huang, Jiawei Fan, Yu Shi

Contents

IX

Research on Track Fitting of Big Frame Intersection Line Seams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Xingang Miao, Su Wang, Xiaohui Li, Benting Wan A Hybrid Approach for Robust Corner Matching . . . . . . . . . . . . 169 Fanhuai Shi, Xixia Huang, Ye Duan Depth Extraction by Simplified Binocular Vision . . . . . . . . . . . . 179 Gouhong Ma, J. Qin, F.R. Jiang, H.B. Wang Comparison of Calibration Methods for Image Center . . . . . . . 185 Xizhang Chen, Shanben Chen, Houlu Xue Seam Tracking and Dynamic Process Control for High Precision Arc Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Huabin Chen, Tao Lin, Shanben Chen Feature Selection of Arc Acoustic Signals Used for Penetration Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Zhen Ye, Jifeng Wang, Shanben Chen Rough Set-Based Model for Penetration Control of GTAW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Wenyi Wang, Shanben Chen, Wenhang Li A Study of Arc Length in Pulsed GTAW of Aluminum Alloy by Means of Arc Plasma Spectrum Analysis . . . . . . . . . . . 219 Huanwei Yu, Shanben Chen Arc Sound Recogniting Penetration State Using LPCC Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 Jifeng Wang, Yantian Zuo, Yichang Huang, Bo Yang, Song Pan Investigation on Acoustic Signals for On-Line Monitoring of Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Na Lv, Shanben Chen A Study on Applications of Multi-sensor Information Fusion in Pulsed-GTAW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Bo Chen, Shanben Chen Research on Image Process and Tracing of a Welding Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 G.H. Ma, L. Wang, G.Q. Liu, M. Xiao

X

Contents

Part III: Modeling and Intelligent Control of Welding Processing Predictive Control of Weld Penetration in Pulsed Gas Metal Arc Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 Zhijiang Wang, YuMing Zhang, Lin Wu Study on the MLD Modeling Method of Pulsed GTAW Process for Varied Welding Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Hongbo Ma, Shanben Chen Simulation of Decoupling Control of Pulsed MIG Welding for Aluminum Alloy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 Ding Fan, Jiankang Huang, Lihui Lu, Yu Shi Modeling and Decoupling Control Analysis for Consumable DE-GMAW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 Jiankang Huang, Yu Shi, Lihui Lu, Ming Zhu, Yuming Zhang, Ding Fan The Structure Design and Kinematics Simulation for Rotating Arc Sensor of TIG Welding Based on Pro/E 3D Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293 Jianping Jia, Hongli Li, Wei Jin, Shunping Yao Knowledge Model Building about a Motor Speed Regulation Fuzzy Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299 Huimin Zhao, Mingyan Ding, Wu Deng, Xiumei Li, Wen Li Research on Surface Recover of Aluminum Alloy PGTAW Pool Based on SFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307 Laiping Li, Xueqin Yang, Fengyan Zhang, Tao Lin Research on Fuzzy-Prediction-Control of GTAW Process Based on MLD Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Mingyan Ding, Hongbo Ma, Shanben Chen Application of Fuzzy Edge Detection in Weld Seam Tracking System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 Zhenyu Xiong, Wen Wan, Jiluan Pan

Part IV: Welding Technics and Automations Application and Research of Arc Welding Automation in Korea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 Suck-Joo Na

Contents

XI

Offline Programming for a Complex Welding System Using DELMIA Automation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Joseph Polden, Zengxi Pan, Nathan Larkin, Stephen Van Duin, John Norrish Robot Path Planning in Multi-pass Weaving Welding for Thick Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351 Huajun Zhang, Hanzhong Lu, Chunbo Cai, Shanben Chen Influence of the Bar Shape on the Welding Quality of Friction Hydro Pillar Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361 Xiangdong Jiao, Hui Gao, Hongwei Zhan, Canfeng Zhou, Jiaqing Chen, Yanqing Zhang Interference Analysis of Infrared Temperature Measurement in Hybrid Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369 Jifeng Wang, Houde Yu, Yaozhou Qian, Rongzun Yang Preliminary Investigation on Embedding FBG Fibre within AA6061 Matrices by Ultrasonic Welding . . . . . . . . . . . . . . . . . . . . . 375 Zhengqiang Zhu, Yifu Zhang, Chun Zeng, Zhilin Xiong Thermal Process Analysis in Welding Prototyping of Metal Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383 Jian-ning Xu, Hua Zhang, Ronghua Hu, Yulong Li Study on Sub-sea Pipelines Hyperbaric Welding Repair under High Air Pressures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 Canfeng Zhou, Xiangdong Jiao, Long Xue, Jiaqing Chen, Xiaoming Fang

Part V: Special Robot Technology and Systems The Mechanism Design of a Wheeled Climbing Welding Robot with Passing Obstacles Capability . . . . . . . . . . . . . . . . . . . . 401 Minghui Wu, Xiaofei Gao, Z. Fu, Yanzheng Zhao, Shanben Chen Anytime Ant System for Manipulator Path Planning . . . . . . . . 411 D. Wang, N.M. Kwok, G. Fang, Q.P. Ha Path Planning and Computer Simulation of a Mobile Welding Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Tao Zhang, Shanben Chen The Control System Design of a Climbing Welding Robot Based on CAN Bus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429 Xiaofei Gao, Minghui Wu, Z. Fu, Yanzheng Zhao, Shanben Chen

XII

Contents

An Implementation of Seamless Human-Robot Interaction for Pipeline Welding Telerobotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435 Na Dong, Haichao Li, Hongming Gao, Lin Wu Design and Experiment of a Novel Portable All-Position Welding Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443 Bin Du, Jing Zhao, Yu Liu The Power and Propulsion of Medical Microrobots . . . . . . . . . . 451 Xueqin Lv, Rongfu Qiu, Gang Liu, Yixiong Wu Mechanical Design and Analysis of an Articulated-Tracked Robot for Pipe Inspection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461 Z.Y. Chen, G.Z. Yan, Z.W. Wang, K.D. Wang

Part VI: Intelligent Control and Its Applications in Engineering A Construction Method of Rational Approximation Model for Fractional Calculus Operators in Frequency Domain . . . . . 471 Wen Li, Guanghai Zheng, Bing Nie, Huimin Zhao, Ming Huang Research of Direct Discretization Method of Fractional Order Differentiator/Integrator Based on Rational Function Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479 Bing Nie, Wen Li, Haibo Ma, Deguang Wang, Xu Liang Blind Source Separation of Vibration Signal of Electric Motor Velocity Modulation System . . . . . . . . . . . . . . . . . . . . . . . . . . 487 Xiumei Li, Wen Li, Yannan Sun, Guanghai Zheng A Survey on Artificial Intelligence Algorithm for Distribution Network Reconfiguration . . . . . . . . . . . . . . . . . . . . . . . 497 Rongfu Qiu, Xueqin Lv, Shuguo Chen Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505

Part I

Intelligent Techniques for Robotic Welding

Research Evolution on Intelligentized Technologies for Robotic Welding at SJTU S.B. Chen Intelligentized Robotic Welding Technology Laboratory, School of Materials Science and Engineering Shanghai Jiao Tong University (SJTU), Shanghai 200240, P.R. China e-mail: [email protected]

Abstract. This paper addresses the new evolution of the study on intelligentized technologies for robotic welding based on our works in SJTU from 2006, which contains multi-information acquirement of arc welding process, such as extraction of weld pool image, voltage, current, and sound features; multi-information fusion algorithms for prediction of weld penetration by neural network and DS evidence theory; intelligentized modeling of welding process and robot system, such as welding dynamic knowledge models by SVM and RS methods, the MLD modeling of welding robot system; and intelligent control strategies for welding pool and penetration process, such as the adaptive inverse control scheme based on SVM-FRDS, control scheme based on the knowledge model by the RS theory, and the Model-Free adaptive control of pulsed GTAW; the 3-D seam tracking during robotic welding by combining arc sensing and visual sensing; and a new welding robot system for autonomously avoiding obstacles.

1 Introduction The most welding robots serving in practical production still are the teaching and playback type, and can not well meet quality and diversification requirements of welding production because this type of robots do not have the automatic functions to adapt to circumstance changes and uncertain disturbances during welding process [1-5]. In practical production, welding conditions are often changing, such as the errors of pre-machining and fitting work-piece would result in differences of gap size and position, the change of work-piece heat conduction and dispersion during welding process would bring on weld distortion and penetration odds [6]. Moreover, manufacturing of some large equipments, e.g., hull and container, need continuous and moving welding in long path and all position. In order to overcome or restrain various uncertain influences on welding quality, it will be an effective approach to develop and improve intelligent technologies for welding robots, such as vision sensing and multi-sensing of robotic welding process, recognizing welding surroundings, autonomously guiding and tracking seam, and real-time intelligent control of robotic welding process. Therefore, developing intelligentized technology for improving current teaching and playback T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 3–14. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

4

S.B. Chen

welding robot is necessary and exigent to meet high quality and flexible requirements for welding products and advanced manufacturing [3-7]. In our previous work [4-7], some research results on intelligentized technologies for welding robots were reported, the ref. [7] showed a functional realization of intelligentized welding robot systems it’s called as the locally autonomous intelligentized welding robot (LAIWR) systems, which could realize some primary intelligentized functions of welding robot systems. Based on our previous research [6-7], this paper presents some further evolutions of the intelligentized technologies for robotic welding [8-20], which contains multi-information acquirement of arc welding process, multi-information fusion algorithms for prediction of weld penetration; intelligentized modeling of welding process and robot system; intelligent control schemes for welding pool and penetration process; and the 3-D seam tracking during robotic welding by combining arc sensing and visual sensing. Further more, a new autonomous welding robot scheme with combined wheel and foot for getting across obstacles in all space position motion will be shown in this paper.

2 Characteristics Extraction of Multi-sensing Information During Arc Welding Process 2.1 Multi-information Acquirement of Arc Welding Process The experiment system for multi-sensing welding dynamical process is shown in Fig.1. The system consists of an electronic signal collecting module, a sound signal collecting model and a weld pool image collecting module.

Fig. 1. Schematic diagram of the experiment systems

By combining the three collecting modules, weld pool image, weld current, weld voltage and welding sound could be collected at the same time. Therefore the welding parameters could be controlled by the control module.

Research Evolution on Intelligentized Technologies for Robotic Welding at SJTU

5

The Fig.2 shows the collected weld pool, current, voltage and sound information in five pulses. From the current, voltage and sound waveforms, it shows that the welding process can be divided into weld pulse peak period and weld pulse base period, the weld pool images were collected at the pulse base period to avoid the intense disturbance in pulse peak period by the welding arc to obtain the clear weld pool image [8,10].

Fig. 2. Weld pool, current, voltage and sound information in five pulses

2.2 Extraction of Multi-information Characteristics During Arc Welding Process 2.2.1 Extraction of Weld Pool Image Features The Fig.2 shows the topside and backside image of the weld pool, the geometric feature parameters can be obtained from the images. The special image processing algorithms for the topside and backside weld pool image can be found in [8]. 2.2.2 Extraction of the Weld Voltage and Current Features From Fig.3, it could be seen that the negative half-wave of weld voltage was bigger than the positive half-wave, this was caused by the cathode spot in pulsed GTAW. In this paper, the positive value of the voltage during a pulse period was used as the mean value of weld voltage during pulse peak period [8,9].

6

S.B. Chen

Fig. 3. Weld current, voltage and sound during a pulse

2.2.3 Extraction of the Weld Sound Characteristics From Fig.2 and Fig.3, the weld sound intensity was used as a variable reflecting changes of welding status. The Fast Fourier Transform (FFT) was used to get the spectrum map of the sound to analyze the changes of the sound, and some other algorithms was developed for extraction of welding status, e.g. penetration characteristics [8,11].

2.3 Information Fusion Algorithms for Prediction of Weld Penetration by Neural Network and DS Evidence Theory The multi-sensor fusion model of the three sensors is shown in Fig.4. The information obtained by different sensors was first processed by back-propagation (BP) neural networks individually. Because welding process was influenced by heat inertia, responded to welding parameters with a time delay, the historical information should also be included to obtain more precise prediction results. The BP neural network was trained to obtain the BPA (Basic Probability Assignment) for each sensor [8]. The D-S evidence theory was used to combine the BPAs and obtain the final fusion BPA and obtain the prediction result. The algorithms details can be found in [8,9]. The experiment and analysis results showed that the multi-sensor could obtain better results than a single sensor. The prediction result of fusing three sensors was better than the prediction result of fusing two sensors. It shows that multi-sensor information fusion could obtain more information about the welding process and therefore describe the process more roundly and precisely.


7

Fig. 4. Fusion model of electronic, image and sound signal

3 Intelligent Modeling of Welding Process and Robot System Modeling of the arc welding is one of the key techniques in automated welding. Real-time control of weld quality by the appropriate model, such as welding penetration, and fine formation of the welding bead, is still a perplexed problem faced by control engineers and welding technologists.

3.1 Knowledge Extraction of Arc Welding Dynamics by SVM Method One of our studies investigated to apply the Support Vector Machine-based Fuzzy Rules Acquisition System (SVM-FRAS) for modeling of the gas tungsten arc welding (GTAW) process [12,13]. The character of SVM in extracting support vector provides a mechanism to extract fuzzy IF THEN rules from the training data set. The fuzzy inference system using fuzzy basis function was constructed. The gradient technique is used to tune the fuzzy rules and the inference system. The schematic diagram of SVM-Based Fuzzy Rules Discovery System was depicted in Fig.5. Using the proposed SVM-FRAS method, we obtained the rulebased model of the aluminum alloy pulse GTAW process, shown as Table 1. Experimental results show the SVM-FRAS model possesses good generalization capability as well as high comprehensibility [12].

–

8

S.B. Chen Step I: Extract Rules

Step II: Inferrence System Process Step III: Adaptive Learning

Adaptive Learning

Support Vector learning mechanism Extract Margin Support Vector S1,S2,...,Sm from the process

IF X is S1 THEN B1

Input X

Fuzzy inference system (Fuzzy basis function inference system)

IF X is S2 THEN B2

Output Y

M IF X is Sm THEN Bm

Fig. 5. Schematic diagram of SVM-Based Fuzzy Rules Discovery System Table 1. Part rules extracted by SVM in modeling for aluminum alloy pulsed GTAW process I t-3 I

t-2 WT

LT

WB

I

O t-1

WT

LT

WB I

t WT

LT

WB

I

WT

LT

WB

……… 191 10.5 17.3 8.2 216 10.5 17.2 8.6 171 8.9 17.3 5.6 192 9.5 16.1 5.7 164 9.9 15.9 6.8 223 11.9 22.2 9

165 10.6 18.5 6.6 194 10.8 17.2 8.5

226 10.5 19.3 12.3 160 9

172 9.6 13.1 5.2 183 10.1 13.4 3.2

16.8 7

219 11.1 18.7 9.4 220 11.7 19.6 9.4 230 11.4 18.2 10.8 163 10

20.7 8.9

………

3.2 Knowledge Model of Al Alloy GTAW Welding Process by the RS Method Based on our previous works on knowledge modeling of the welding process by the Rough Sets (RS) theory [5-7], the further research was completed and some knowledge rule models for welding pool dynamics of aluminum alloy GTAW by RS methods was developed in [6,14].

3.3 Mixed Logical Dynamical (MLD) Modeling of Pulsed GTAW During Robotic Welding The mixed logical dynamical (MLD) modeling method stems from hybrid systems described by interacting physical laws, logical rules, and operating constraints.


9

The study in [17] presented a novel MLD modeling framework for robotic welding process and systems. The MLD model is then established and gives a good prediction quality of the back bead width of pulsed GTAW process with misalignment during robotic welding. The study [17] shows that the MLD framework is a good modeling method for pulsed GTAW process and robotic welding systems.

4 Intelligent Control Strategies for Welding Pool and Penetration Process Based on the knowledge models established in the above, some intelligent control scheme are shown as the following.

4.1 The Adaptive Inverse Control Scheme Based on SVM-FRDS for Pulsed GTAW Based on the SVM-FRAS model of the aluminum alloy pulse GTAW process in 3.2, the adaptive inverse control system for pulsed GTAW process was developed as Fig.6. Here the I-Controller means the inverse controller, the P-model means the process model for welding process, and the Predict-M means the predicted model of the backside width of welding pool during pulsed GTAW. The experiment results of the closed control system can be found in[12-13].

§

Wbd

SVM-FRDS SVM-FRDS Äæ¿ØÖÆ Æ÷ I-Controller

I

Wb

Pulsedº¸GTAW GTAW ½Ó ¹ý ³Ì P-Model

V-Sensing ´« ¸ÐÏµ Í³ ep ec

£«

SVM-FRDS SVM-FRDS ¹ý ³Ì Ä£ÐÍ

£-

P-model

) Wb

£-

±³Predict-M ÃæÈÛ¿í Ô¤ ²â Ä£ÐÍ £«

Wb

Fig. 6. The adaptive inverse control scheme based on SVM-FRDS for GTAW process

4.2 Closed Control Schemes Based-on the Knowledge Model by the RS Ttheory Based on the obtained knowledge rule models for welding pool dynamics of aluminum alloy GTAW by RS methods, some control schemes were developed for

10

S.B. Chen

realization of the closed real-time control of Al alloy welding pool during pulsed GTAW as follows: 1

）The controller based-on RS for the closed-loop control system of pulsed

GTAW The primary control scheme in Fig.7 was used to regulate the backside pool width by the RS controller during Al alloy pulsed GTAW.

Signal Conversion

Wbset +

e(t )

Δu

RS Controller

+

z −1 u

GTAW Process

+

Wb

Welding parameters

Wbpre

Welding pool sizes

Prediction Model

Sensing System

Fig. 7. The schematic diagram of RS closed-loop control system of pulsed GTAW

）

2 The compound control scheme with the RS and MS-PSD controllers for weld penetration and face-height of Al alloy pulsed GTAW. In order to control weld penetration and the height of topside seam during Al alloy pulsed GTAW at the same time, the advanced control strategy compounded the RS controller and MS-PSD controller for the closed control system scheme was developed as Fig.8. The controller algorithms and welding experiments can be found in [14]. H fset +

e1 (t )

MS-PSD Controller

ΔV1 f +

+

+

z −1

Signal Conversion

Wbset

+

e2 (t )

RS Controller

+

ΔI

Vf

V1 f

z −1

+ ΔV2 f

Feed forward Controller

Hf

g GTAW Process

Wb

I

+ Welding parameters

Wbpre H fpre

Prediction Model

Welding pool sizes

Sensing System

Fig. 8. The compound control systems with the RS and MS-PSD controllers for weld penetration and face-height of Al alloy pulsed GTAW


11

4.3 The Model-Free Adaptive Control of Pulsed GTAW Arc welding is characterized as inherently variable, nonlinear, time varying and having a coupling among parameters. In addition the variations in the welding conditions cause uncertainties in the welding dynamics. Therefore, it is very difficult to design an effective control scheme by conventional modeling and control methods. A model-free adaptive control algorithm has been developed to control the welding process [15], which only needs the observed input output data and no modeling requirement for controlled welding process. Thus, the developed model-free adaptive control provides a promising technology for GTAW quality control. The Model-Free Adaptive Control algorithms was introduced in [15].

5 The 3-D Seam Tracking During Robotic Welding by Combining Arc Sensing and Visual Sensing The seam tracking technique for three dimension (3-D) curve during robotic pulsed GTAW process was developed by the combination of arc sensing for torch height or arc length with passive visual sensing for correct an error of seam or torch deflexion [16]. The seam tracking scheme of robotic welding process is shown as Fig.9. Processing of welding pool image and identifying of seam and gap changes are realization through the image as Fig.10. The tracking control algorithms and experiments can be found in [16].

Fig. 9. Seam tracking scheme of welding robot system

12

S.B. Chen

Fig. 10. Welding pool image during robotic welding

6 Some Applications of Intelligentized Robotic Welding Using the above intelligentized technologies for robotic welding, some significant applications are progressing in spaceflight, shipbuilding and automobile industries about 10 years [18-19], here is omitted

7 An Autonomous Welding Robot System for Whole Position Motion and Across Obstacle In many practice welding manufacturing sites, such as welding of shipbuilding and large tank, there is a need for the autonomous moving welding in a long distance and complicated space position. It also requires the welding robot with adsorbent and climbing functions for all position motion and flexible pose changes for various joints, such as the fillet, lap, vertical, inclined welding, and so on. So, a primary autonomous moving welding robot system with a combination of wheels and foot for adsorbent climbing and getting across obstacle was developed. It has some intelligentized functions to realize robotic welding, such as visual and ultrasonic sensing, automatic program of welding path and technical parameters, autonomous guiding and tracking welding. The scheme of the welding robot system for autonomous moving and across obstacle is shown as Fig.11. It integrates the above intelligentized technologies for robotic welding and realizes some welder’s intelligent functions, such as detecting and recognizing weld surroundings by visual sensing technology, identifying the initial position of weld seam, autonomously guiding weld torch to the weld starting and tracking the seam, real-time control of arc welding pool dynamics. The details are in [20].


13

Fig. 11. An autonomous welding robot system for all position moving and across obstacle

8 Conclusion Remarks Based on our previous work, this paper has shown some further research developments on intelligentized technologies for robotic welding in the Intelligentized Robotic Welding Technology Laboratory of Shanghai Jiao Tong University from 2006. It contains the multi-information acquirement of arc welding process, such as extraction of weld pool image, voltage, current, and sound features; multiinformation fusion algorithms for prediction of weld penetration; tracking seam based on combining arc and visual sensing, knowledge modeling; real-time intelligent control of weld penetration and welding pool dynamics. An autonomous welding robot systems with a combination of wheels and feet for all position moving and getting across obstacles has been developed, which will integrate the above intelligentized technologies for robotic welding and realizes some welder s intelligent functions.

’

Acknowledegment. This work was supported by the National Natural Science Foundation under Grant No. 60874026 and No. 51075268; National 863 plan of China under Grant No. 2009AAA042221, and supported by Key Foundation Program of Shanghai Sciences & Technology Committee under Grant No. 09JC1407100. The author wish to acknowledge the relative study works finished by Dr. Chen Bo, Dr. Huang Xixia, Dr. Lv Fenglin, Dr. Ma Hongbo, Dr. Fan Chongjian, Dr. Kong Meng, Dr. Wang Jifeng, Dr. Yezhen Dr. Shen Hongyuan, Dr. Chen Huabin, Dr. Wu Minghui, and etc.

References [1] Dilthey, U., Stein, L.: Robot System for Arc Welding—Current Position and Future Trends. Welding & Cutting (8), E150–E152 (1992) [2] Trailer: Manufacturer Depends on Robotic Welding to Boast Production. Welding Journal 74(7), 49–51 (1995)

14

S.B. Chen

[3] Pan, J.L.: A survey of welding sciences in 21th century. In: Proceeding of 9th Chinese Welding Conference, Tianjun, China, vol. 1, pp. D001–D017 (October 1999) [4] Tarn, T.J., Chen, S.B., Zhou, C.J.: Robotic Welding, Intelligence and Automation. LNCIS, vol. 299. Springer, Heidelberg (2004) [5] Tarn, T.J., Chen, S.B., Zhou, C.J.: Robotic Welding, Intelligence and Automation. LNCIS, vol. 362. Springer, Heidelberg (2007) [6] Chen, S.B., Wu, J.: Intelligentized Technology for Arc Welding Dynamic Process. LNEE, vol. 29. Springer, Heidelberg (2008) [7] Chen, S.B.: On the Key Intelligentized Technologies of Welding Robot. LNCIS, vol. 362, pp. 105–116 (2007) [8] Chen, B.: Study On The Processing Method Of Multi-Sensor Information Fusion in Pulsed Gtaw, PhD Dissertation, Shanghai Jiao Tong University (May 2010) [9] Chen, B., Wang, J., Chen, S.: Prediction of pulsed GTAW penetration status based on BP neural network and D-S evidence theory information fusion. International Journal of Advanced Manufacturing 48(1-4), 83–94 (2010) [10] Wang, J.: Study on recognition method of penetration states based on welding sound feature, PhD Dissertation, Shanghai Jiao Tong University (September 2009) [11] Wang, J.F., Fenglin, L., Chen, S.B.: Analysis of arc sound characteristics for Gas tungsten argon welding. Sensor review 29(3), 240–249 (2009) [12] Huang, X.: Study on Support Vector Machine-based modeling methods and Their Application in Material Processing. PhD Dissertation, Shanghai Jiao Tong University (June 2008) [13] Huang, X., Shi, F., Gu, W., Chen, S.: SVM-based Fuzzy Rules Acquisition System for Pulsed GTAW Process. Engineering Applications of Artificial Intelligence 22(8), 1245–1255 (2009) [14] Fan, C.: Weld Pool Characters Extraction Visual Sensing And Intelligent Control During Varied Gap Aluminum Alloy Pulsed Gtaw Process. PhD Dissertation, Shanghai Jiao Tong University (September 2008) [15] Fenglin, L.: Study On Model-Free Adaptive Control Of Weld Pool Dynamic Proces in Pulsed Gtaw. PhD Dissertation, Shanghai Jiao Tong University (September 2008) [16] Meng, K.: Research On Process Control Method For Arc Welding Robot Based On Multi-Information Sensing Real-Time. PhD Dissertation, Shanghai Jiao Tong University (June 2008) [17] Ma, H.: Research On Mixed Logical Dynamical Modelling Of Aluminium Alloy Pulsed Tig Robotic Welding Process Based On Vision Feature Information. PhD Dissertation, Shanghai Jiao Tong University (August 2010) [18] Shen, H.-y., Lin, T., Chen, S.-b., et al.: Real-time seam tracking technology of welding robot with visual sensing. Journal of Intelligent and Robotic Systems 59, 283–298 (2010) [19] Chen, H.B., Lv, F.L., Lin, T., Chen, S.B.: Closed-loop control of robotic arc welding system with full-penetration monitoring. Journal of Intelligent and Robotic Systems 56, 565–578 (2009) [20] Chen, S., Wu, M.: The autonomous welding robot systems with combination of wheel and foot for climbing and getting across obstacle motion. The middle research report of the National 863 Plan of China (November 2009)

Image Processing for Automated Robotic Welding Peter Seyffarth and Rainer Gaede Ingenieurtechnik und Maschinenbau GmbH Rostock/ Germany Industriestr. 8, D-18069 Rostock Tel.: +49-381-793-450 e-mail: [email protected]

Abstract. The construction of seagoing ships needs a lot of micro panels with maximal dimensions up to 16 x 4 m. About 2000 – 2300 micro panels with different sizes and about 20 – 25 various kinds of design are to weld for a medium size containership from 1700 – 2400 TEU. These are fillet welds mainly between stiffeners and the panel plate. There are two different techniques to make panels: For very long and parallel stiffeners are weld gantries in use for both side fillet welding, but for longitudinal and transversal stiffeners robotic welding is mainly used. Up to now the welding robot is linked and programmed by the CAD/CAM-system of the yard. This needs the offline programming more or less weeks before production and in this time the design of the panels can change. It is also an organizational problem to have the right CAD-program at the right time on the welding robot and to guarantee a flexible production process independent of the size and design of the micro panel. Therefore a way out is the flexible image processing. Developed by IMG and used by WADAN yards, MTW image processing system for welding robots delivers a very fast automatically generated program. The system, consisting of a laser line scanner with a scanning rate of 1 m/sec and a fast working calculation unit, gives a three-dimensional picture in less than 3 minutes all over the 16 x 4 m panel. After the shake hand process between the image processing system and the robot programming system the accuracy of the robotic positioning is ± 0.5 mm. This new developed 3-dimensional geometrically recognising and robot programming system allows a very fast and flexible production system for micro panels without any link to the central computer aided yard system. This “stand alone system” is independent, more flexible and shows, beside of other advantages, a high productivity.

1 Introduction The main parts in the prefabrication of ship hulls are flat and curved panels with dimensions up to 20 x 40 metres or more and so named micro panels. We meet especially the last one, the micro panels, in a large number of different forms an sizes beginning from 2 x 2 metres up to 4 x 16 metres. In a medium sized container freighter for 2000 containers there are about 2500 or more various micro panels. There are different production technologies for micro panels, consisting

T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 15–21. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

16

P. Seyffarth and R. Gaede

from plates and stiffeners. A modern micro panel production at a shipyard uses robots for mounting and welding as well. This is shown as an example in Fig. 1 and 2. Due to the various construction types of micro panels and their large range from unique single construction up to minimum series there is a high demand on the programming of robot systems including the movement of the robots in the 3dimensional coordinates and the welding parameters. In shipbuilding each micro panel needs its own welding program. All known programming procedures require either additional information on the work piece in the shape of CAD data or they need manual interaction. The classical programming of weld robots for micro panels takes places regardless of the really existing work piece and production scenario by a partly automatic analysis of CAD data in combination with demanding manual interactions. The programming takes place temporally very much in advance of the production and needs a high quality and relevance of the available construction data. Unfortunately this is in contrast to the flexibility of the production flow and can not take in account some changes of construction. That means, we have to take in account, that normally the programming is done a long time before production and very often the NC construction data are changed in the meantime.

Fig. 1. Mounting gantry for stiffeners in a micro panel line

Fig. 2. Welding station in a micro panel line using 2 robots before refitting

Image Processing for Automated Robotic Welding

17

Measurement of position Work. range

Weld robot

Gantry Scan head

Working area

Scan. range

Fig. 3. Principle of the 3 D image processing for robot programming

2 Refitting of an Existing Micro Panel Line by Image Processing To meet the needs for a very high flexibility and for an automated programming on demand for the existing panel line, which is shown in Fig. 1 and 2, the enterprise Ingenieurtechnik und Maschinenbau GmbH (IMG) designed, constructed and delivered recently a very fast working three-dimensional image processing system in cooperation with aviCOM and TSWE. The principle of this 3 D image processing system is shown in Fig. 3.and an overview about the system controlling units gives Fig. 4. The heart of the new industrial 3D scanner measuring system for micro panels is a camera head based on modern camera technology. The scan head measures 3D data according to the laser triangulation principle. Hence, to be able to measure 3D shape, an external line-generating laser source is used. The laser generator is mounted to the robot and projects its laser line on the working area from a distance of about 2 m in the height. The camera, that views the line from a different angle, sees a curve that follows the height profile of the object. By measuring the laser line deviations from a straight imaginary reference line, the height of the object can be computed. For scanning the robot moves with the scan head and the laser line along the working area, contour slices of the object are generated. The collection of such

18


slices, or 3D profiles, is a description of the complete object shape as seen from the upper side of the object. The unique camera technology is capable of finding the position of the laser line by itself and reducing to whole image information into compact laser coordinates. These laser coordinates are transmitted to the PC. That is what the mounted 3D imaging technology makes very fast and reliable. Inside the scan head the camera offers several different methods for the generation of 3D profiles which differs in speed and height resolution. This flexibility of the camera was used to optimize results for the specific scanning task and material. The measurement principle gives geometrical limitation concerning the measurement of hidden parts in relation to the camera view. There are two kinds of limitations, camera occlusion and laser occlusion. Camera occlusion occurs when the laser line is hidden from the camera by an object and laser occlusions occur when the laser cannot properly illuminate parts of an object because of its projection angle. Adjusting the angles of the scan head and the laser can reduce the effects of occlusion. Additionally we use two scan heads with two laser sources illuminating the micro panels and especially the profiles from opposite sides. System controlling by industrial-PC

Scan head and image processingcomputer

Robot controller and robot

Image.processing software

Online programming

Rotrol-server macroadministration

Fig. 4. Overview about the controlling units of the image processing and robot programming system

The measurement system’s 3D field-of-view (FOV) is a trapezoid-shaped region where the laser line intercepts the FOV of the camera. It is only in this region that the camera generates 3D measurements. The camera FOV is given by the selected lens and camera software parameters. The height resolution of the measurements is dependent on the angle between the laser and the camera – as the angle is increased, the resolution is also increased – and on the selected 3D method. Generally, if the precision of the profiling algorithm is high, the maximum profile rate is limited compared to a less precise but fast algorithm. The maximum profile rate is dependent on a combination of the selected 3D method, the required measurement resolution, and the required height of the measurement region. By for instance decreasing the height region used for object


19

inspection the profile rate can be increased. Note, however, that the maximal usable profile rate also depends on the amount of light reflected from the object. The data stream of profiles was synchronized with the robot movement using an external encoder. This functionality will ensure that the length measurement and object scale in the movement direction is correct, even if the object speed varies, Fig.5. All parameters were optimized for the application of scanning micro panels under production conditions. The result is a scanning speed of 0.5 m/s with a maximum of independence from surrounding light conditions. The resolution of the 3D points is about 1mm in x and y direction and 2 mm in z (height). It is to consider that the resolution differs over the field of view in the relation to the height of points. The measurement range in z is actually 400 mm. The design goal was to optimize the overall scanning time, the sum of time to scan and process data. Although the scanning speed could be higher (up to 1 m/s) it is shown that the shortest process time is reached by processing the data parallel to the scanning process. The scanning is done in stripes of 1.20 m width and 16 m length in reversing order of forward and backward movement. Hence, the whole working area consists of a maximum of four stripes for each scanning head. 2. Scan-head

B

C Image processingcomputer

Scanhead

Robotwelding gantry

D Systemcontrolling unit (industrial-PC)

A

Robot controlling unit

2 Robot controlling unit.

Position measurement system

Fig. 5. Concept of data processing for 2 robots from scan head to the welding robot

The measured 3D points in the laser plane of the scanner have to be processed with complex mathematical algorithms to calculate real world coordinates. This is done mostly during the scanning process parallel to the capture of 3D points. Additionally a first step of data reduction is done to extract object shape information as panel planes, profile planes and contours. The whole working area is scanned after about 3.5 minutes. At this time the extracted object shape information is available for high level processing in the final geometry extraction, Fig.6.

20


Fig. 6. Visualisation of the micro panel on the user screen, ready for further high level processing

Fig. 7 shows the scan jib together with the two laser sources and the camera fixed at the refitted robot. The process of geometry extraction calculates the matching of all scanning stripes separately for each scan head. Additionally it calculates the matching of both scan heads to one description of the scene. This description consists of panels

Camera

Laser light

Fig. 7. Refitted robot with automated programming system by image processing in operation at the yard


21

with 3D contour polygons, a plane approximation for profiles distinguishing different types such as HP, Flat, T- or L-Profiles and an information which profile belongs to which panel.

3 Advantages of the New Installed 3 D Image Processing System The main advantages of the recently installed new programming system by 3 D image processing are the following: z z z z z z

Programming on demand guarantees a very high production flexibility Programming is a very fast process, there is no remarkable production time lost. The scan rate depending on the profile height is 0.5 up to 1 m/s The process time for the whole process depending on the size of the panel and the number of stiffeners is 3.5 up to 12 minutes. This is very fast in comparison with some hours of welding time. The accuracy with 1 mm in x- and y-direction and 2 mm in z-direction is very high. It needs no operator for programming.

4 Conclusions The micro panel line is now successful in operation more than one year at one German shipyard. It needs only one operator for all processes including mounting and tacking of the stiffeners, changing the filler material for the two robots and so on. The automated micro panel line provides a high level of automation utilisation, production flexibility and an increase in efficiency. The experience at the yard shows, that the pay-back time for the investment is in a range of one year.

Automatic Seam Detection and Path Planning in Robotic Welding Kevin Micallef, Gu Fang, and Mitchell Dinham School of Engineering, University of Western Sydney, Locked Bag 1797, Penrith South DC, 1797, Australia e-mail: [email protected]

Abstract. To make robotic welding systems more flexible, vision sensors are introduced as they provide large amount of information about the welding components. In this paper a method is introduced to automatically locate the weld seam between two objects in butt welding applications. The proposed method provides flexibility for robotic welding by having the ability to locate the weld seam on arbitrarily positioned work pieces. This method is also cost effective as it is developed using images captured from a low cost web-cam. Furthermore, the proposed method is able to plan a robot path along the identified seam. Simulation and experimental results show that the method can be used successfully in detecting and locating seams on variously shaped work pieces and robot paths can be successfully generated to follow the weld seams.

1 Introduction There are many Robotic applications in the real world which range from industrial applications such as material handling or manufacturing to medical tasks such as surgical procedures [12]. Vision guided robotics is a topic of continued interest and recent advances in visual servo control and the technology that supports it have made it possible for the creation of accurate and robust vision control systems. Using vision to guide a robotic manipulator is, however, still a challenging task [2]. At present, there are many welding robots being used to maintain the safety of workers and the work area, as well as the quality of work. Current welding robots used in industrial applications are programmed through teach and play-back methods [12, 13]. Weld paths for these types of robots must be manually re-taught and re-programmed for different work pieces as they cannot self-rectify any offsets in the welding process. This makes the set up of these systems time consuming and expensive. To make the robotic welding more flexible in dealing with varying locations and shapes of weld pieces, vision is being introduced into robotic welding systems. Research has been reported where computer vision is used to detect and locate weld seams [1, 3, 9, 12, 13]. In addition, vision is also being used to control the weld quality by monitoring the weld pool dynamics [1, 5]. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 23–32. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

24

K. Micallef, G. Fang, and M. Dinham

Automatic weld seam detection and tracking is an important topic in robotic welding. It is an area of great interest as it is an important step in realising fully autonomous robotic welding [9]. Since vision systems can acquire information without interfering with the welding process, it is often used for seam tracking. There are a variety of approaches being adopted for seam detection and tracking in robotic welding. The use of specialised lighting is reported in [9]. In this method, a single laser line was projected onto planar work pieces and was used to detect the seam when the line is no longer straight. The introduction of a second laser line [6] generates more accurate information from the images. Another weld seam detection method using a single image is given in [13]. The image is manually segmented by only considering a central, predefined viewing window. This reduces computational time by not having to eliminate the background and reducing the overall size of the image. Once the seam is located from the grey level image, the start and end points are found by shifting a smaller window along the length of the seam until a corner is detected. This method assumes that the seam location will be in the central in the image, and large in comparison to the window. Also, the work piece is assumed to be planar with known depth and height. In [12] a method for real time seam tracking is proposed. The vision system is setup to take images slightly ahead of the welding torch which continually update the robot with the new position coordinates extracted from the image information. Since the viewing window is a significant distance away from the welding arc, it is difficult to continually track the seam around curves and sharp corners, therefore limiting this method to straight lines, significantly reducing the flexibility of the system. The methods presented above are suited mainly for the detection of simple seams. These methods are also developed to only deal with specific welding applications in mind. Significant modifications would have to be made to these seam detection algorithms before they could be implemented in multiple applications. In addition, these methods did not consider the effect that path planning plays on seam detection and tracking. Smooth and even paths are required to produce high quality welds. In this paper an automatic, low-cost flexible seam detection method will be developed which can be accurately and easily implemented in an industrial setting. Furthermore, the proposed method will also incorporate path planning into the algorithm to develop a smoother welding path for robot to follow. This paper is organised into four Sections. In Section 2, the methods used will be explained. Experimental results using an industrial robot are given in Section 3. Conclusions are then given in Section 4.

2 Methodology In this paper, a method for weld seam detection is introduced. The algorithm is capable of determining the shape of the seam as well as the start and end points on straight and curved seams. This method is also capable of reducing the number

Automatic Seam Detection and Path Planning in Robotic Welding

25

of points on the weld seam to smooth the robot welding path. This method uses a single image captured by a robot mounted camera referred to as the eye-in-hand configuration. The overall method contains 6 steps: Boundary Detection, Conner/Curve Detection; Determine Initial Seam; Select Initial and Final Position; Seam Point Reduction; and Final Seam Path. Each step is discussed in detail in the following subsections.

2.1 Boundary Detection An image containing the two objects in position for the butt-weld operation must first be captured. This image must view the entire seam. Appropriate image processing is performed to allow for the boundary detection of the two objects {B1, B2} to take place (as depicted in red and green in Fig. 1). This will result in a sequence of n1 and n2digital points describing the boundary of each object B1 and B2,

Bk = {Pi,m = (U i,m ,Vi,m ),i = 1, 2,...,n i , and m = 1,2}(0)

(1)

Where image point Pi+1,m is the neighbour point of the image point Pi,m, and (Ui,m,Vi,m) are the image coordinates of Pi of boundary m. An example of two objects with detected boundaries can be seen in Fig. 1.

Fig. 1. Object Boundaries Detected and Highlighted

2.2 Corner/Curve Detection Once the boundary pixel locations for each object are identified, the corner/curve pixels can then be determined. This is necessary as in a butt welding operation the two edges to be joined are relatively parallel and thus will begin and end on a sharp corner or curve. There are many methods available to determine corners in an image [7, 8, 10, 11]. In this paper, the corner detection method is based on the K-Cosine introduced in [15]. The K-Cosine detection technique is chosen for its low computational costs and simplicity. This boundary based corner detection technique determines the angle θ for each boundary pixel as shown in Fig. 2:

26

K. Micallef, G. Fang, and M. Dinham ܲ݅൅‫ܭ‬ ܽറ݅ ܲ݅

ɽ ܾሬറ݅ ܲ݅െ‫ܭ‬

Fig. 2. Vector Diagram

r r r r ai • bi = ai • bi • cos θ i r r C i = cos θ i = a i • bi

(ar

i

r • bi

(2)

)

θ i = cos −1 (Ci )

(3) (4)

Where ai and bi are the vectors between image points Pi,m and Pi+k,m, and Pi,m and Pi-k,m, respectively, θ i represents the angle between the two vectors ai and bi.

• is

the magnitude of vector y. The number of pixels between the points used in calculating vectors ai and bi is represented by the value K . It is clear from (2) and (3) that if pixels Pi-k,m and Pi+k,m, form a straight line, then Ci will be -1 or close to -1. One of the limits of the K-Cosine corner detection method is the assumed prior knowledge of the number of corners in an image and an estimate of the widest angle to occur in the image. This prior knowledge will allow for the threshold value T to be established to determine what Ci value constitutes a corner. This paper improves on the original K-Cosine corner detection method by allowing the number of corners in an image and angles of these corners to remain unknown. This improved method can also detect points that form curves in an image. To achieve this, a low threshold T, defined as -0.99(which is almost equal to a straight line θ = 172°) and a value of K = 3 was used in the method. This modification results in a large number of pixels being identified as corner or curve points. Based on these points, regions can be formed if points are connected. To reduce these regions to a single point, the value of θi is calculated for each pixel in the region using (2) – (4) and is then compared to the θ values of the neighbouring pixels in the same region. The pixel with the smallest angle θ is retained as the pixel with the greatest curvature for that region. The corner/curve detection method introduced above is more advantageous than the original K-cosine method as it not only identifies corner points but also identifies points around curves allowing for an accurate path planning along the seam. By determining the local maximums, a significant reduction in the number of points used in the preceding sections of this paper is achieved.


27

2.3 Determine Seam Edges The next step to this method is to identify the points that lie along edges which form the weld seam. This is done using the knowledge that the points along the weld edges are the points of interest when determining the seam location. To eliminate the redundant corner/curve points, the distance between all the corners of one object and all the corners of the second object is determined using the Euclidian distance: D=

(V2 − V1) + (U 2 −U1) 2

2

(5)

It is assumed that the corners are situated at coordinates (U1, V1) and (U2, V2). Each corner/curve point on one boundary is paired with the closest corner/curve point on the opposing object boundary. During this process, pairs that have separation greater than a set minimum pixel distance are regarded as redundant and will be eliminated. Once the above process is complete, the initial seam edge arrays are generated and can be used to determine the points that make up the initial seam. For the buttweld operation the weld seam is made of the mid-points between the two objects, therefore the location of the seam points can be calculated. The initial seam made up by these midpoints is then used to determine the start and end points of the weld seam and also in determining the accurate path for the robot to follow during the welding operation.

2.4 Initial and Final Position Selection and Seam Point Reduction Using the initial seam from Section C the initial and final points along the seam are determined using the process described below. Where U represents the x-axis in the image plane and V is the y-axis representation. This method is limited as the initial and final positions must be on opposite sides of the weld objects. Once the initial and final positions of the seam are determined, a path for the robot to follow is determined. The number of points found along the seam will vary significantly depending on the style of each individual seam. Find pixel with minimum and maximum pixels in both the U and V direction Calculate the distance between the min U and max U = dist U, Calculate the distance between the min V and max V = dist V, If dist u > dist V Initial position = min U, Final position = max U, Else If dist V is > dist U Initial position= max V, Final Position = min V

28


The process of corner detection developed in this paper identifies not only points on a corner but also a significant number of points situated around curves. The disadvantage of detecting many points around curves is that each of these points will be used to drive the robot. This will affect weld quality and requires a great deal of memory space in the robot controller. However, by detecting all these points curves can then be identified along the seam that increases the overall flexibility of the system. A suitable method is developed in the paper to reduce the number of points whilst maintaining the shape and accuracy of the seam. The steps to perform this operation are shown below. 1.

2.

3. 4. 5. 6. 7.

8. 9.

Seam Edge Reduction Method Compare Cosine Value of Pi with Pi+1 i. If Cθi not equal Cθi+1, Go to (2); ii. If Cθi equal Cθi+1, Go to (3); Is Cθi = Cθi-1? i. YES, Go to (4); ii. NO, Store in Corner Array, Go to (5); Store in Curve Array Reduce Curve Array IF Number of points in Curve array = 2 or 3 Store Points as the Final Seam Edge, Go to (5); IF Number of Points >=3 Take the First, Middle and the Last points, store these points as the Final Seam Edge, Go to (5); Is this Final Point? i. YES, Go to (7); ii. NO, Go to (6); Next Point i. Go to (1); END

The above process results in a significant reduction of points along the seam edges. The final seam is then determined. Consequently the robot will be able to perform a smoother path along the seam resulting in an improved weld quality. The extent of this reduction is revealed in Section III.

2.5 Final Seam Points Input to the Robot The seam detection method results in a path made up of pixel locations in the image coordinate plane. This requires the transformation of the pixel locations from the image coordinate plane to the world coordinate plane. It is well known that the conversion of a 3-D point (x, y, z) into an image point (u, v) can be expressed as: w × (u,v,1) = P × (x, y,z,1) T

T

(6)

Where P is a 3x4 homogeneous transform and w is a scaling factor. To convert a point from the image plane (u, v) to the 3-D coordinate frame (x, y, z),one will need to know one of the values of x, y or z, alternatively, a stereo vision system is required to calculate the location of x, y and z from two images. In this paper z is


29

assumed to be known, as is usually the case that a workstation is at a certain height. Using linear algebra, knowing u, v and z will allow the values of x and y to be calculated. To achieve this, the P matrix in (6) should be determined. In this paper, the P matrix is obtained using the method introduced in [4]. This hand-eye calibration technique was used as it is low cost, using a single web cam and has an accuracy in the x and y direction to within ±1mm. The accuracy achieved in the x and y directions are acceptable for the robotic arc welding process. With known seam points in image plane with their z values, the welding seam points’ world coordinate positions can be obtained. These real-world coordinates can then be loaded into the robot’s control system as a set of path points so that the robot can be operated to follow the seams.

3 Results This section contains the seam detection program test results, as well as the verification that the seam detection method is applicable to a real world system.

3.1 Seam Detection and Reduction Results To test the effectiveness of the seam detection program, objects with a variety of seam shapes were used. Case A contains a straight edged seam, Case B contains two straight edges creating a distinct corner, Case C requires the detection and navigation around various curves and finally Case D reveals a seam can be detected with parts placed in arbitrary orientations. Fig. 3 shows the points (in yellow) that make up the final seam for each individual situation. These seam points are by using the point reduction method along with the initial (green) and final (red) positions of the seam are shown.

3A. Case A

3B. Case B

3C. Case C

3D. Case D

Fig. 3. Initial Seam points for different welding shapes

30


Fig. 4 reveals that a path for a robot to follow along the seam can be determined. The method proposed also meets the flexibility demands of seam detection. The resulting path for the horizontal seam in Figure 3D shows that not only is the seam detection method developed in this paper capable of detecting various shaped seams but can also determine the path if the objects of interest are placed in an arbitrary orientation in the workspace. Table 1 shows the results of the point reduction. Table 1. Point Reduction Results Number of Points On Initial Seam Case A Case B Case C Case D

16 125 157 110

On Final Seam 5 62 54 63

Reduction Percentage 68.75% 50.40% 65.61% 42.73%

The reduction results revealed in Table 1 are significant as this reduction will save significant memory space in the robots controller, doing so without the loss of path accuracy.

3.2 Experimental Results To verify the image processing results on a real application, further experiments were carried out using a robot. The robot used to verify that the seam path detected is applicable in a real life situation was a Fanuc M-6i Robot. The end effecter uses a brass pointer to simulate the position of welding wire in a welding application and the camera is mounted in an eye-in-hand configuration. The resulting motion of case B is shown in the Fig. 4.

4A. Motion 1

4B. Motion 2

4C. Motion 3

4D. Motion 4

Fig. 4. Case B Cornered Seam


31

4 Conclusions In this paper a flexible seam detection method has been introduced. The developed method successfully detects different shaped objects in an image. In addition, the seam detection method is capable of generating smooth welding paths using minimum available point which can potentially improve the welding quality of the system. Experimental verification has validated the effectiveness of this seam detection method. The experimental results show that the seam detection method is capable of detecting various shaped seams made up of objects placed in arbitrary positions hence proving the flexibility of the seam detection method developed. Acknowledgement. This work is supported by the Australian Research Council under project ID LP0991108 and the Lincoln Electric Company (Australia).

References [1] Chen, S., Zhang, Qiu, T., Lin, T.: Robotic Welding Systems with Vision-Sensing and Self- Learning Neuron Control of Arc Welding Dynamic Process. Journal of Intelligent Robotic Systems 36, 191–208 (2003) [2] Corke, P.: Visual Control of Robots: High Performance Visual Servoing. Research Study Press, Somerset (1996) [3] Cook, G.: Feedback Control of Process Variables in Arc Welding. In: Proceedings of IEEE Joint Automatic Control Conference, vol. 2 (1980) [4] Dinham, M., Fang, G.: Low Cost Eye in Hand Camera Calibration. In: Proceedings of The 2009 IEEE International Conference on Robotics and Biomimetics (ROBIO 2009), Guilin, Guangxi, China, December 18-22, pp. 1889–1893 (2009) [5] Dudek, G., Jenkin, M.: Computational Principles of Mobile Robotics. Cambridge University Press, New York (2000) [6] Gao, S., Zhao, M., Zhang, L., Zou, Y.: Dual-Beam Structured Light Vision System for 3D Coordinates Measurment. In: Proceedigns of the 7th Annual Conference, World Congress on Intelligent Control Automation, Chongqing, China (2008) [7] Lee, J., Sun, Y., Chen, C.: Multiscale Corner Detection by using Wavelet Transform. IEEE Transactions on Image Processing 4(1), 100–104 (1995) [8] Li, L., Chen, W.: Corner Detection Interprestation on Planar Curves using Fuzzy Logic Reasoning. IEEE Transactions on Image Processing 2(11), 1024–1210 (1999) [9] Liu, X., Wang, G., Shi, Y.: Image Processing of Welding Seam Based on SingleStripe Laser Vision Systems. In: Sixth International Conference, Conference on Intelligent Systems Design and Applications (2006) [10] Rattarangsi, A., Chin, R.: Scale-Based Detection of Corners of Planar Curves. IEEE Transactions on Pattern Analysis and Machine Intelligence 14(4), 430–449 (1992) [11] Rosenfeld, A., Johnson, E.: Angle Detection on Digital Curves. IEEE Transactions on Computers 22, 875–878 (1973) [12] Shen, H., Lin, T., Chen, S.: A Study on Vision-Based Real-Time Seam Tracking in Robotic Arc Welding. Robot. Weld Intelligence. & Automation, 311–318 (2007)

32


[13] Shi, F., Zhou, L., Lin, T., Chen, S.: Efficient Weld Seam Detection for Robotic Welding from a Single Image. Robot. Weld Intelligence. & Automation, 289–294 (2007) [14] Sicard, P., Levine, M.: An Approach to an Expert Robot Welding System. IEEE Transactions on Systems, Man and Cybernetics 18(2), 204–213 (1988) [15] Sun, T.: K-Cosine Corner Detection. Journal of Computers 3(7), 16–22 (2008)

Error Compensation and Calibration of Inter-section Line Welding Robot Based on a Wavelet Neural Network Su Wang1, Xingang Miao1, Yuan Yang1, and Xingai Peng2 1

School of Mechanical Engineering and Automation, Beijing University of Aeronautics and Astronautics, Beijing 100191, China e-mail: [email protected] 2 China Petroleum Pipeline Engineering Corporation Langfang 065000, China

Abstract. The main source of error of saddle-back coping welding robots is discussed in this paper. In view of the error source, a three-layered wavelet neural network compensation model is designed. Two steps are involved in using this model: the first step is to compensate the position error of torch point caused by axes 4 and 5 of a robot, the second step is to compensate all the five axes movement error respectively. The corresponding simulation and experimentation are conducted based on the compensation algorithm. Results show that this compensation model can greatly improve the movement precision of the robot. The average position error of torch point is reduced by 80%, and the average movement error of each axis is reduced by 60%.

1 Introduction Intersection line welding framework is widely applied in many fields. As for the working intensity is very hard in welding these space curves, more and more robots are employed in the welding process. A kind of saddle-back coping welding robot is innovated by Beihang University, and this robot can be fixed on an intersection cylinder [1]. However, because robot kinematics and dynamics are in a great complexity, and there are numerous unknown factors in multi-input and multi-output robot system which is highly non-linear and intensely coupling, geometry errors and gear wheel transmission errors do exist inevitably. All of the above will generate great errors between factual movement and theoretical program when the robot moves, and the position error can be 10mm indeed [2, 3]. Therefore the error compensation and calibration of robot are in a great demand, in order to make the robot work in the best state. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 33–40. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

34

S. Wang et al.

2 Saddle-Back Coping Welding Robot Framework and Error Analysis Figure I illustrates the framework of saddle-back coping welding robot. It is composed of five axes: the axis 1 is the waist joint which is to drive the whole robot to move around the intersection cylinder; axis 2 is the horizontal joint which can move horizontally; axis 3 is the vertical joint which can move in a vertical line; axis 4 is the big arc guiding track and axis 5 is the little arc joint. The axis 1,2,3 designed to control the welding point orbit and the other two axes are used to adjust the pose of the torch point. The axis 4 and axis 5 are two arc joints which are intersected uprightly sharing the same center, at which the touch point is located. In this way, the torch point pose can be adjusted while the two arc joints are moving, but the touch position will not be affected.

Fig. 1. Framework of saddle-back coping welding robot

If there is some geometry error in axis 4 or axis 5, serious problems will take place to the torch position when the two axes are moved, and the torch position and torch pose will be efficiently coupling, which is not we wanted. As a result, the main task of error compensation and calibration is to focus on the compensation of axis 4 and axis 5 to the whole structure especially the influence to the torch point position. In addition, each axis is driven by worm wheel, worm bar, planet retarder or screw rod, all of which will inevitably produce errors when the robot is moving, so this error compensation must be also included.

3 Algorithm of Intersection Line Welding Robot Error Compensation 3.1 Model of Error Compensation and Calibration Wavelet neural network (WNN) is a novel neural network based on wavelet analysis theory and artificial neural network. Because the WNN inherits the timefrequency localization of wavelet analysis and self-learning ability of neural

Error Compensation and Calibration of Inter-section Line Welding Robot

35

network, it is strong in approaching and error-tolerating ability [4]. The number of nodes of hidden layer is equal to that of the wavelet base, and it can be worked out by the input signal. The relativity of wavelet neural enables the wavelet network to run in a faster convergence speed [5]. In this robot, the pose angle of torch point is adjusted by the axis 4 and axis 5. The two axes are intersected uprightly and share the same center, so theoretically speaking, when they are moving, the torch position will not be changed. But errors also exist in geometrical operation, size of the two axes, and the installation, and the torch position will be changed when the two axes move. The torch position error produced by the two axes can not be compensated by themselves, so it must be compensated by the other axes. One compensation method based on wavelet neural network is employed here. First, the torch position error produced by the axis 4 and axis 5 is compensated by the axis 1, axis 2 and axis 3, and then the compensation amount is obtained, so the total moving quantity of the each first three axes is obtained by their own moving amount adding to the compensation amount. Second, we can compensate the five axes moving error respectively. In the compensation, the first step and the second step adopt different wavelet neural networks. The theory map is shown in Fig. 2.

Fig. 2. Robot error compensation and calibration theory map

The number of input layer neural cells and output layer neural cells can be determined by practical demands, but the number of hidden layer neural cells does not only relate to the input and output cells but also relate directly to the number of training samples. According to the related references and experiments, it can be determined by the following experience formula [6].

N = [ I + O + 1 / 2 × O × ( I 2 + I ) − 1] / 2 +

P ±1

I--input number, O--output number, N--hidden number, P--sample number.

(1)

36

S. Wang et al.

3.2 Design of a Robot Error Compensation Algorithm In this wavelet neural network, input variables are joint angles of axis 4 and axis 5, output variables are joint angles of axis 1, axis 2 and axis 3 which are position compensation produced by pose change. The framework of WNNP based on one-step-front-predictive method is shown in Fig. 3, the input vector is

X (k ) = [θ k 4 θ k 5 ]T , the input layer is full channel; the output vector is Y (k ) = [ Δθ k1

Δd k 2

Δd k 3 ]T , ψ i is the wavelet inspiration function, the

output layer is the linear node. 25 training samples are examined here, and the number of wavelet base function N is 12 through the calculation from formula 1. So the 2-12-3 wavelet neural network is finally constructed.

Fig. 3. Framework of wavelet neural network

The second step of compensation is built on the first one. The input variables are the joints of the first three axes which are achieved after the completion of position error compensation, and the joints of axis 4 and axis 5, the output variables are the five axes joint angle already compensated, each axis has a wavelet neural network respectively.

4 Experiment of Error Compensation and Calibration 4.1 Experiment of Torch Point Error Compensation and Calibrate In the experiment, the laser interferometer (ML10) produced by Renishaw the U.K. is used to measure the position error caused by robot moving. According to the error compensation and calibration principle designed above, there are two steps in the case of calibration. First, an experiment to error compensation produced by axis 4 and axis 5 is done. In the work space of the robot, a point is chosen at random, and the reflector of laser interferometer is fixed to the torch point and is adjusted to parallel one axis of right-angle coordinate, then laser interferometer and spectroscope are adjusted, so that laser can parallel this axis too, please see Fig. 4.


37

Fig. 4. Diagram of laser interferometer setting

This point is regarded as a reference starting point, there are 45 aiming points (including the starting point) are taught to axis 4 and axis 5 respectively. At each point, the linearity measurement mirror team is used to measure relative position distance changes, and the RenishawawML10 software is used to record the measurement data, but the joint angles of axis 4 and axis 5 in each point are obtained at robot controller. Then, laser interferometer and reflector are adjusted to parallel the other two axes of right-angle coordinate, and axis 4 and axis 5 are moved to every measurement point used in the last time, memorize the measurement data again. In the experiment, the data of each point is achieved by the average result of double direction for 3 times measurements. Then 36 points are selected as the training samples and the rest 9 will play the role of checking samples. The WNNP wavelet neural network is trained by the training samples, then the position error produced by axis 4 and axis 5 can be compensated by the trained WNNP. After that, the robot is tested again by the above method. Figure 5 illustrates the comparison of position errors in the three coordinate direction caused by axis 4 and axis 5’s movements, before and after WNNP compensation On the basis of the measurement data before and after compensation, the robot precision can be analyzed, please look at table 1. From this table, it can be observed that the average torch point position error in X direction of axis 4 and axis 5 declined by 84% compared to the one before compensation, and the maximum error declined to 19.8%. The average error in Y direction declined by 85.7%, and the maximum error declined to 18.8%. The average error in Z direction declined by 85.8%, and the maximum error declined to 21.9%. So this method can effectively decline the torch position error. Table 1. Torch point position precision before and after compensation

Error Ex Ey Ez

Before compensation(mm) average 0.4737 0.4393 0.2629

maximum 0.8958 0.9770 0.6792

After compensation(mm) average 0.07493 0.06239 0.03274

maximum 0.1778 0.1840 0.1485

38

S. Wang et al.

Fig. 5. Axis 4 and 5 error comparison before and after compensation

4.2 Error Compensation and Calibration of Each Axis The above compensation is just the first step of Fig. 2, on which the second step is gonging to be carried out. In this step, the five axes will be compensated separately, and the same wavelet neural network will be used. Limited by the length of this dissertation, we only take the axis 2 as an example to describe the experiment process, as for the other axes, comparison of precision and final conclusion will be listed. The axis 2 serves as the horizontal axis and it’s stroke is 500mm, so the laser interferometer measure step is set as 20mm. There is, include the reference point, 26 point in all. At each point, it is measured twice in positive direction and negative direction, and is repeated for five times. Each point measurement average value is shown in table 2. The average error serves as the training sample of WNNB wavelet neural network, then the axis 2 can be compensated by it. After the compensation, the same method is adopted to measure this axis again. The measurement result is shown in Fig. 6.


39

Table 2. Average measurement value of axis 2 position error

position(mm)

0

20

40

60

80

100

average(µm)

0

-27.68

-41.9

-46.84

-48.9

-46.58

position(mm)

120

140

160

180

200

220

average(µm)

-41.72

-41.5

-40.1

-41.32

-42.24

-43.58

position(mm)

240

260

280

300

320

340

average(µm)

-49.2

-56

-66.52

-77.74

-89.48

-98.24

position(mm)

360

380

400

420

440

460

average(µm)

-102.58

-113.38

-116.56

-114.1

-101.72

-90.8

position(mm)

480

500

average(µm)

-76.14

-46.2

Fig. 6. Axis 2 curve of position error before and after compensation

It can be seen that the axis 2 position precision is greatly improved by the compensation, most position errors are within 15µm, the maximum error is 19.4µm, which declined by 83.8% compared to that before compensation, and the average value also declined by 72.4%. So the error compensation plays a favorable role.

5 Conclusions The source of error of saddle-back coping welding robot is analyzed. One kind of the error is the geometrical error from axes 4 and 5 which are intersected uprightly, the other kind of error is the manufacturing and transmission error of each axis. The wavelet neural network compensation model is designed according to the above error, and Matlab simulation and experimentation are made based on the compensation algorithm. The results show that this compensation model can

40

S. Wang et al.

effectively increase the robot’s movement precision, the position average error of torch point is reduced by over 80%, and the average movement error of each axis is reduced by over 60%. Acknowledgement. This work is supported by Construction Together Special Project of the Education Committee of Beijing (NO. 20091024).

References [1] Li, X., Wang, S.: Kinematic analysis and simulation of saddle-back coping welding robot. Journal of Beijing University of Aeronautics and Astronautics (08), 964–968 (2008) [2] Ma, L., Yu, Y., Ceng, W., et al.: Positioning error compensation for a parallel robot based on BP neural networks. Optics and Precision Engineering 16(05), 877–883 (2008) [3] Xia, K., Chen, C., Hong, T., et al.: A neural network model for compensating robot kinematics error. Robot. 17(03), 171–176 (1995) [4] Cong, L.: Path Following Control of Mobile Robot Based on Wavelet Neural Network. Coal Mine Machinery 29(02), 47–49 (2008) [5] Lian, Z., Wang, C., Zhang, J.: Research on improvement of BP algorithm based on WNN. Computer Engineering and Applications 43(2), 99–101 (2007) [6] Bao, J., Zhao, J., Zou, H.: Study on method of curve simulation based on BP network. Computer Engineering and Design 26(07), 1840–1841 (2005)

Autonomous Seam Acquisition and Tracking for Robotic Welding Based on Passive Vision Shanchun Wei, Meng Kong, Tao Lin, and Shanben Chen Intelligentized Robotic Welding Technology Laboratory, School of Materials Science and Engineering Shanghai Jiao Tong University (SJTU), Shanghai 200240, P.R. China e-mail: [email protected]

Abstract. A method of autonomous seam acquisition and tracking for arc welding robot is proposed. The seam information is real-timely obtained by one CCD camera in front of the anterior seam, while the host computer is in charge of image processing and edge features extracting. Thus the deviation between the projection of the tungsten electrode and the central line of welding seam can be obtained. Simultaneously tracking trajectory is optimized by Kalman filter and revised to eliminate the deviation defined so as to track the weld seam. The method can also be used to finish seam acquisition and tracking in the whole welding process autonomously even if the robot is never taught. The results validate feasibility of the method and imply the improvement of autonomy and intelligence for robot welding.

1 Introduction Nowadays, a great number of industrial robots have been used in the fields of automatic arc welding. However, most industrial robots are still working in the teaching and playback mode which can’t meet the requirements of intelligent welding process and welding quality [1]. Hence, various sensors are applied in the welding process. Sensing technology is used to monitor the status of the process and acquire characteristic information. It is one of the most important steps for robotic welding automation. Many research papers, specific to different interferences and features of welding conditions, proposed various sensing methods, for example, ultrasonic method [2], infrared thermometry [3] and positive vision such as laser sensing [4]. With the development of the computer vision technology and image processing methods, numerous welding robots and some automatic welding machines are equipped with corresponding vision sensors to achieve different welding tasks in severe conditions. Kim [5] developed one welding robot system with both laser sensors and passive vision, which could be used for welding environment identification and seam tracking. Besides, K.Y. Bae [6] studied the online seam tracking of steel tube based on machine vision. But the weld gap was 2-4mm, which decreased the difficulty of image processing. J.S. Smith [7] studied GTAW real-time T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 41–48. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

42

S. Wei et al.

top-face vision and used PID controller to assure the welding formation and quality. Due to its great reliability, small size and low cost, passive sensing technology with CCD sensors has been taken as one of the most common methods for robot welding sensing. This paper focuses on the autonomous seam acquisition and tracking for arc welding robot, presents a method of weld seam extraction and tracking with the CCD sensor. To some extent, it would be of importance to improve the level of autonomy and intelligence for robotic welding.

2 Experiment System Fig. 1(a) shows the main components and their relationships of the experiment system. Fig. 1(b) shows the robot welding system and scene photograph. RH6 robot is used in the experiment system which can accomplish motions of six joints in the process of trajectory planning. The interval that host computer could send data to the controller for robot motion is 16ms. The communication between the robot controller and the host computer is achieved by CAN bus. Welding torch is fixed at the end of the flange joint, while the CCD camera is fixed at the front of the torch. The images obtained by the camera are converted into digital signals through the CG400 image acquisition card. Since the host computer extracts the feature information of these images, such as seam track, corner and projection of tungsten, it can translate image information to robot-known data so as to control the robot to follow the seam track. During the movement, the coordinate values of the robot position are fetched by the CAN bus for the host computer’s record. Thus, the host computer could calculate the deviation between the projection of tungsten and seam central line, and revise the path of the robot in the form of incremental coordinate values.

(a)

(b)

Fig. 1. (a) Components and Communication of System; (b) Real Experiment System

Welding for butt plates is implemented to verify the effectiveness and adaptability. Before welding, the torch of the robot is guided to the initial welding position, while the welding path is not taught. The CCD camera is in charge of capturing the real-time images for seam information extraction so that the host

Autonomous Seam Acquisition and Tracking for Robotic Welding

43

computer can calculate deviation to help the robot planning the welding path and complete the welding task.

3 Image Process and Seam Extraction The image of welding seam and molten pool taken by the CCD camera is shown in Fig. 2(a). The images are captured on the moment of base pulse current. In order to reduce the redundant information and accelerate image processing speed, a small window in front of the molten pool is selected. In the small window, seam can be obtained quickly. Meanwhile, to eliminate the random noise interference in the image acquisition and transmission, the regular techniques of smoothing and sharpening are adopted in the window as pre-processing. Median filter with 3×3 template is adopted. Image sharpening is also used for enhancing the feature information. The pre-processing window image is shown in Fig. 2(b).

(a)

(b)

Fig. 2. (a) Origin Image; (b) Extracted Window

In order to improve the precision of the image processing and reduce the reflection interference of arc light, a new algorithm is presented to extract the feature of the seam edges. The basic procedures are expressed as follows: 1) Sobel-operator edge detection, aiming at extracting the edge information; 2) Threshold segmentation, converting the image into binary pattern easy to subsequent processing; 3) Image thinning, by which the main skeleton can be distinguished; 4) Adaptive area filtering, eliminating clutter points and edges in order to remove pseudo-edge; 5) Welding region locking, locking the most interested object for seam tracking; 6) Image restoration and Canny-operator extraction, here, processing the seam edges in the region locked as Canny operator detection contains more details which is necessary for seam tracking; 7) Area filtering again and fitting the seam edges, thus the central line of the weld seam can be obtained easily finally. The procedures and results are shown in Fig. 3, while Fig. 4 shows the final result. The whole algorithm costs less than 300ms to accomplish all steps. What’s more, different welding experiments prove the feasibility of the method as the average offset is less than two pixels, about 0.017mm.

44

S. Wei et al.

After seam edges extraction, two sets of data are stored to fit the curve. Linear fitting by least squares method is taken to obtain the linear equation of the central line of the weld seam, as shown in the Equation (1). f ( x ) = ax + b ; a = ( k1 + k 2 ) / 2; b = ( b1 + b2 ) / 2;

(1)

where k1 , k2 , b1 , b2 are the coefficients of the fitting curves for the upper and lower weld seam edges.

Fig. 3. Image Procedures and Results

Fig. 4. Result of the Image Processing

4 Seam Deviation Definition As shown in the Fig. 5, there are two coordinate systems, one is the image system, i.e. u-v coordinate system and the other is the base system, i.e. x-y coordinate system. Here, the image coordinate is defined as image captured by CCD camera. Its origin is the left lower point. The base coordinate system means the real coordinate system used for robot motion. In the image coordinate system, the distance in the v-axis direction, which is between the tungsten projection and the central line of the seam, is defined as weld seam position deviation d . Also, when

Autonomous Seam Acquisition and Tracking for Robotic Welding

45

the tungsten projection is above the seam in the image, d is defined as positive. Otherwise, it is negative. Defined that angle deviation of weld seam α is the angle between the tangent of the central line of the seam and the positive u-axis. Then, according to geometric relationship and ratio between actual distance and image solution, the weld seam position deviation d can be obtained as follows: d = s0 ( pv − ( apu +b ))

Where s0

pu

and

pv

(2)

are the u-axis and v-axis coordinates of the tungsten projection.

is the actual size of a pixel in the v-axis direction.

Fig. 5. Map of the Error Definition

Fig. 6. Map of Tracking Control

5 Seam Tracking Controller A control cycle of RH robot is 16ms. Incremental data, also called step size, determines the velocity of the robot. These data can be sent to the robot once per 16ms. Only one shot is carried out in every pulse period to capture images at the moment of base pulse current, which assures the stability and clarity of the captured images. The pulse frequency of 2 Hz is adopted. The multithread method is used to solve the asynchrony of two cycles, information processing and motion controlling. The main thread is in the charge of image acquisition, processing and deviation extraction, while the other is for sending incremental data and controlling the robot every 16ms. Here, define the control strategy as follows: Rule 1: If the deviation is zero, the robot follows the tangent line direction of r r the weld seam at a velocity of V f ; Rule 2: If not, Vb determined by the deviation is added to the ongoing movement of the robot to make the robot move towards the central line of the weld seam in the direction of the v-axis. r r According to the control theory, V f and Vb can be used as feed-forward velocity and feedback velocity respectively. The diagram of the robot tracking control is r shown in the Fig. 6. According to the geometry relationship, the formulas of V f and

r Vb

are as follows:

46

S. Wei et al. vx

V f < cos(D E ) Vb < sin E ; v y

V f < sin(D E ) Vb < cos E

(3)

The resultant velocity of the gun is: v = vx2 + v y 2 = V f 2 +Vb 2 + 2V f Vb sin α

(4)

That is to say, the step sizes could be obtained as long as the feedback velocity is determined. Considering that a seam tracking system is real-time and dynamic, the influence of steady-state errors could be neglected. This system adopts the PD controller to control the feedback velocity, which makes the algorithm simple and efficient. After the derivation of the PD controller, step sizes sent to the robot for x-axis and y-axis are: lx ly

Where

s1

1 V f cos(D E ) ( k p1d ( k ) kd 1 ( d ( k ) d ( k 1)))sin E ( )) V f sin(D E ) ( k p1d ( k ) kd 1 ( d ( k ) d ( k 1)))cos E 1 V f cos(D E ) ( k p1d ( k ) kd 1 ( d ( k ) d ( k 1)))sin E s1 1 λ22 λ11

281

(4)

This means that the system is strong coupling. Figure1 is the square wave signal response in the controlled coupling system based on the model eq. 3. The square wave signal’s cycle is 20s. From the figure, we may find the coupling relation is strong. It also follows that the decoupling control is required by circumstances.

(a) The response of welding pool width

(b) The response of wire extension

Fig. 1. Coupling relation simulation in pulsed MIG welding process for aluminum alloy

3 Simulation of Decoupling Control 3.1 Compensation Decoupling Control Diagonal matrix decoupling system is one of the effective decoupling methods. Its principle is adding a matrix to the control system and making the generalized object matrix, which is the product of the added matrix and the object characteristic matrix, into a diagonal matrix. Thus realize the decoupling of the system. The diagonal matrix decoupling control system structure of double-input and double-output is shown in figure 2.

Fig. 2. Structure of the diagonal matrix decoupling system

282

D. Fan et al.

Decoupling compensation matrix is eq.8 ⎡ G11 ( S ) G12 ( S ) ⎤ ⎡ G p11 ( S ) G p12 ( S ) ⎤ ⎡ G11 ( S ) G12 ( S ) ⎤ ⎢ G ( S ) G ( S ) ⎥ ⎢G ( S ) G ( S ) ⎥ = ⎢G ( S ) G ( S ) ⎥ ⎣ 21 22 ⎦ ⎣⎢ p 21 p 22 22 ⎦ ⎦⎥ ⎣ 21

−1

⎡ G11 ( S ) 0 ⎤ ⎢0 G (S )⎥ = ⎣ 22 ⎦

⎡ G ( S ) G 22 ( S ) − G12 ( S ) G 22 ( S ) ⎤ 1 * ⎢ 11 ⎥ G11 ( S ) G 22 ( S ) − G12 ( S ) G 21 ( S ) ⎣ − G11 ( S ) G 21 ( S ) G11 ( S ) G 22 ( S ) ⎦

(5)

After simplifying, the diagonal matrix decoupling compensator is eq. 9 0.0551S 2 + 0.254S + 0.177 0.122 S 2 + 5.07 S + 17 G p12 = − 2 0.331S + 1.433S + 0.407 0.331S 2 + 1.433S + 0.407 2 0.137 S + 0.632 S + 0.44 0.005S 2 + 0.211S + 0.177 G p 21 = G p 22 = − 2 0.331S + 1.433S + 0.407 0.331S 2 + 1.433S + 0.407 G p11 =

、

、

(6)

、

When use a PI controller, K P1 =17 K i1 =1.5 K P 2 =1 K i 2 =2, the asynchronous pulse signal response is shown in figure 3, the cycle is 100s.

(a) Response of weld width

(b) Response of wire extension

Fig. 3. Decoupling simulation of diagonal matrix compensation PI controller

In figure 3, a diagonal matrix compensation decoupling control method with PI controller has some decoupling effect on pulsed MIG welding for aluminum alloy; the response of the system is well. But with the asynchronous signal, especially the changes of duty cycle have great influence on the wire extension, and cause sharp changes of the wire extension, shown as 3b.

2.2 Neural Network Object Inverse Model Decoupling Control Intelligent decoupling control is the hot spot in research field of control. It has unique superiority in solving nonlinear system and it can realize the precise online decoupling of linear and nonlinear systems, solves the problem of difficult to achieve precise decoupling with traditional decoupling method, for example compensation decoupling.

Simulation of Decoupling Control of Pulsed MIG Welding for Aluminum Alloy

283

Figure 4 is the designed system structure of the neural network object inverse model decoupling control for pulsed MIG welding for aluminum alloy. The structure of NN1 and NN2 are the same, but the input and output interface are transposition. Through transforming the weight matrix of NN1 into weight of NN2, the generalized object matrix can become a unit matrix, thus realize the full decoupling. NN1 adopts four-level neural network, its structure is 2×12×12×2 gradient descent method training network, and adopts off-line training and on-line update.

Fig. 4. Structure of neural network object inverse decoupling control system in welding

To get better simulation effect and performance, self-learning neural network PID controller was used to control the generalized object and realize the control of duty cycle of pulse current and wire feed speed. Nerve PID controller adopted incremental control method, and its structure is 3 × 12 × 2 . 1 .2

Out put y 1( t )

1 .0 0 .8 0 .6

r1 y1

0 .4 0 .2 0 .0 -0 .2 0

10

20

30

40

50

60

70

80

90

100

1 10

120

T i me t / s 1 .2

Out put y 2( t )

1 .0 0 .8 0 .6 0 .4

u2 y2

0 .2 0 .0 -0 .2 -0 .4 0

10

20

30

40

50

60

70

80

90

100

110

120

T i me t / s

Fig. 5. Response curve of the neural network inverse-plant decoupling control system

Figure 5 shows that the coupling effect on duty cycle and wire feed speed to wire extension and welding pool width can be largely eliminated by using neural PID inverse model decoupling control, and respond quickly. Compared figure 3 with figure 5, it shows that the neural network object inverse model decoupling control system using neural network PID can eliminate the coupling relationship between each channel, and overcome the coupling effect on duty cycle to wire extension in diagonal matrix compensation decoupling control, and has good dynamic and steady performance. It can meet the practical control requirements of pulsed MIG welding for aluminum alloy.

284

D. Fan et al.

4 Conclusions Through the analysis of identification model for pulsed MIG welding for aluminum alloy, a MIMO system was determined, in which variable δ and Vwire were selected to control welding pool width and wire extension simultaneously, and then analyzed the coupling state of parameters. Decoupling control simulation on the determined coupling system, whose input were duty cycle and wire feed speed while output were welding pool width and wire extension, was carried out by separately adopting the diagonal matrix compensation decoupling control with a PI controller and a neural PID object inverse model decoupling control. The results show that the neural network object inverse model decoupling control system with the neural network PID controller can mostly eliminate the coupling relationship among each channel, and has good dynamic and stable performance and can meet the actual control requirements of the pulsed MIG welding for aluminum alloy. Acknowledegment. This research work is supported by National Nature Science Foundation of China (50675093).

References [1] Ozcelik, S., Moore, K.L., Naidu, S.D.: Application of MIMO Direct Adaptive Control to Gas Metal Arc Welding. In: Proceedings of the American Control Conference Ohiladelphia Rennsyvania, pp. 1762–1766 (1998) [2] Zhang, Y., Walcott, B.L., Wu, L.: Adaptive predictive decoupling control of full penetration process in GTAW. In: First IEEE Conference on Control Applications, pp. 938–943 (1992) [3] Shi, Y., Li, J., Fan, D., et al.: Vision-based control system for aluminum alloy MIG welding pool width. Transactions of the China Welding Institution 28(2), 9–12 (2007) [4] Ma, P., Yang, J., Cui, C., et al.: Current Situation and Development of Decoupling Control. Control Engineering of China (02), 97–100 (2005) [5] Tianyou, C.: Multivariable adaptive decoupling control and application. Science Press, Beijing (2001) [6] Shi, Y., Fan, D., Huang, A., et al.: Vision-based identification model of welding pool width dynamic respondence in aluminum alloy pulsed MIG process. Acta Metallrugica Sinica 41(09), 994–998 (2005) [7] Shi, Y., Huang, J., Fan, D., et al.: Vision-based identification model and extracting algorithm of wire extension in aluminum alloy MIG alloy welding process. Transactions of the China Welding Institution 28(08), 1–4 (2007)

Modeling and Decoupling Control Analysis for Consumable DE-GMAW Jiankang Huang, Yu Shi, Lihui Lu, Ming Zhu, Yuming Zhang, and Ding Fan Lanzhou University of Technology, Lanzhou, China e-mail: [email protected], [email protected]

Abstract. Aiming at the instability of dynamic welding arc and strong coupling among welding parameters, a mathematical model for consumable doubleelectrode gas metal arc welding (DE-GMAW) was established by the method of equivalent path and then dynamical analysis was carried out. On this basis, an approach was proposed to control the stability of the welding arc using the wire extensions of main and bypass torch as a control object. And then a comparative analysis of two different decoupling control schemes was made respectively using the bypass wire feed speed and the bypass current as input and using the wire feed speed of main and bypass as input. The results show that the mathematical model can reflect the process of consumable DE-GMAW well and the decoupling control scheme using the wire feed speed of main and bypass as inputs has better dynamic performance and robustness.

1 Introduction The description of welding arc in GMAW is more and more precise. On this basis, many methods for arc control were proposed. The literature [1] studied the influence of change of the conductive mouth’s contact point to metal transfer mode and arc length in pulsed MIG welding process. Taken into account the quality of the power source, the literature [2] studied the change of the genuine wire extension. Compared many arc models of GMAW, the arc models of DE-GMAW and laserarc welding were less reported because of insufficient understanding of the coupled arc. In this paper, the mathematical model for arc system in consumable DEGMAW was built by the method of equivalent current path, and it could reflect the dynamic changes of welding process perfectly. On the basis of analyzing the coupling relation of welding parameters, the effect of different control schemes was simulated and the best control scheme was determined.

2 Consumable DE-GMAW Consumable DE-GMAW is established on the basis of a conventional GMAW system by adding another GMAW torch as bypass. It can be seen that in Fig. 1


286

J. Huang et al.

that one GMAW torch is connected to the positive terminal of a CV power supply while another torch is connected to the negative terminal of a CC power supply. The current I flowing through the welding wire of main path (namely total welding current) divide into two parts in arc column zone. One part will flow back to the CC power supply, called bypass current Ibp while the other current will flow through the work-piece, called base metal current Ibm, shown as Fig. 2.

Fig. 1. Schematic diagram of consumable DE-GMAW[3]

Fig. 2. Current distribution

In this design, the current flowing through welding wire can be very large by increasing the bypass current, which is helpful to improve melting rate of welding wire and the deposition rate. According to the characters of DEGMAW, due to the bypass sharing part of current, so on the premise of ensuring large deposition rate, the heat input of base metal can be still maintained at a low level [3-4].

3 ARC Model of Consumable DE-GMAW As illustrated in Fig.3, coordinates ( xa , ya ) ( xb , yb ) ( xc , yc ) ( xd , yd ) ( x, y ) represent the points A, B, C, D, E. AC means the wire extension of bypass torch and BD means the wire extension of main torch. In the arc system, the current flows through the equivalent path DE,EC and EF. Beside the point E is the junction of the total equivalent current.

Fig. 3. Model diagram of consumable DE-GMAW

Modeling and Decoupling Control Analysis for Consumable DE-GMAW

287

There are several assumptions for the system model of DE-GMAW: 1). Assuming the CV and CC power supply for consumable DE-GMAW is ideal. 2). In the arc column area, assuming that the plasma has been formed, and form a good conductive channel. 3). The model was considered as two dimensions. 4). Assuming the current flowing through the welding arc obeys the minimal energy principle. 5). Assuming the work piece and point D is equipotent. The secular equation for welding arc in GMAW is also suitable for DE-GMAW. The expression of arc voltage varc is shown as equation [5].

varc = v + Ra I + Ea la

(1)

I is welding current; v and Ra ( Ω ) are constants. Ea (v/m) is potential gradient of arc column; la is length of arc. Where,

The voltage drop vls of wire extension in welding process is expressed as equation (2). vls = ρ ls (2) Where, ρ is density and ls is length of wire extension. So the output voltage of welding power source is expressed as equation (3).

U = varc + vls

(3)

The wire feed speed is invariable and the melting equation is:

k1I + k2ls I 2 = vm Where

(4)

k1 and k2 are coefficients; vm is melting rate.

Based on the above formulas, the constraint conditions of point E are shown as equation (5) (6).

⎧U − Rb lmain − U c − ( Ra I + Ea lde ) = U E ⎪ ⎨ Ra I bm + Ea le + U k = U E ⎪R I + E l + U + R l = U a ce k b ac E ⎩ a bp

(5)

lce + lde + le

(6)

Min

288

J. Huang et al.

Where U is supply voltage; U C is anode voltage drop; U k is cathode voltage drop; U E is the potential of point E; Rb is coefficient. Try to solve the equations (4), (5) and (6). Because the coordinate of point E ( x, y ) and Ibm are unknown and constraint condition is given, the equations are difficult to solve. So the numerical calculation process as Fig. 4 is proposed.

Fig. 4. Flow chart of numerical calculation

4 Coupling Relation Analysis of Parameters in Arc System According to the built analytical model, dynamic system simulation was carried out with ideal CV and CC power supply by Matlab/simulink. Runge-Kutta algorithm was adopted and step is 0.001s. The simulation parameters are shown in the table 1. Table 1. Simulation parameters Symbol

Value

Unit

U0

15.7

V

Ra

0.022

Ω

Ea

128

V/m

k1

2.553e-4

m/(A s)

k2

4.623e-5

(A2 s)-1

·

·


289

Wire feed speed of main torch is 0.25m/s while that of bypass torch is 0.10m/s. Initial wire extension of main torch is 0.015m while that of bypass torch is 0.010m. The main voltage Umain is 36V. The bypass current Ibp is 140A. The change of current and wire extension in initial state is shown as Fig. 5. 0.020

250

Wire extension Lex/m

I total

Current I/A

200 Ibp

150

100 Ibm

50

0

0.015

0.010 Lbypass

0.005

0.000 0

200

400

600

800

Lmain

1000

0

200

400

600

800

1000

Time t/ms

Time t/ms

(a) Change of current

(b) Change of wire extension

Fig. 5. Parameters change with time in initial state

In order to test the correctness of the model, the simulation was carried out respectively under step of bypass current, step of wire extension of bypass torch and step of wire extension of main torch. Fig. 6 is the response of parameters under step of bypass current. From the Fig. 6(a), we can see when the bypass current decreased, the base metal current increased quickly and the total current increased a little. At the same time, the length of bypass wire extension increased quickly while the length of main wire extension decreased a little. 0.020

Current I/A

250


300 Itotal

200 I bp

150 100

Ibm

50 0

0

200

400

600

800

Time t/ms


1000

0.015

Lmain

0.010

Lbypass

0.005

0.000

0

200

400

600

800

1000

Time t/ms


Fig. 6. Response of parameters under step of bypass current

Fig. 7 is change of parameters under step of main wire feed speed. From Fig. 7(a) we can find the bypass current had no change, the total current decreased and the base metal current also decreased under step of main wire feed speed. Simultaneously, the length of main wire extension decreased while the bypass wire extension had almost no influence, as shown in Fig. 7(b).

290

J. Huang et al. 300

0.35

200

0.30 WFS1

150 100

Ibp

0.25

Ibm

50 0

0

200

400

600

0.016 L bypass


Itotal

Wire feed speed v/(m/s)

Current I/A

250

0.20 1000

800

0.012 Lmain

0.008

0.004

0.000

0

200

Time t/ms

400

600

800

1000

Time t/ms



Fig. 7. Response of parameters under step of main wire feed speed

Fig. 8 is change of parameters under step of bypass wire feed speed. In Fig. 8, we can see that the step had little influence on current (the total and base metal current decreased a little) and main wire extension, but had great affect on bypass wire extension, as shown in Fig. 8(b).

Itotal

WFS2

0.20

Ibp

150

0.18 Ibm

100

0.16

50

0.14

0

200

400

600

800

Time t/ms


0.12 1000


Current I/A

200

0

0.020

0.22


250

Lbypass

0.015

0.010

Lmain

0.005

0.000

0

200

400

600

800

1000

Time t/ms


Fig. 8. Response of parameters under step of bypass wire feed speed

From Fig. 5, Fig. 6, Fig. 7 and Fig. 8 we can see that the simulation results based on built model is consistent with the results of consumable DE-GMAW step test in literature [6]. Through analyzing, we can find that the change of bypass parameters has great influence on arc system of consumable DE-GMAW and determines the stability of arc system.

5 Decoupling Control Simulation and Discussion To ensure the stability of consumable DE-GMAW process, the most direct method is to keep the wire extension of main and bypass invariant when other parameters change. The relatively simple method to control wire extension is to match suitable wire feed speed. Decoupling simulation for wire extension adopting feed forward compensation was carried out. Fig. 9 is the change of parameters under step of bypass current in decoupling control simulation.


Current I/A


Itotal

200

150 Ibp

100 Ibm

50

0

0.015

0.15

0

200

400

600

800

Lbypass

0.012

WFS1

0.12

Imain

0.009

0.09

0.006 WFS2 0.06

0.003

0.03

1000

0

200

Time t/ms

400

600

800


250

291

0.000 1000

Time t/ms


(b) wire feed speed and change of wire extension

Fig. 9. Change of parameters under step of bypass current in decoupling control process

As can be seen from Fig. 9b, the system quickly enters a new stable state under step of bypass current by adjusting wire feed speed of main and bypass simultaneously to control wire extension. Another scheme is through adjusting current and wire feed speed of bypass to control the stability of wire extension. In simulation of compensation decoupling control, adjusting the current and wire feed speed of bypass simultaneously will take a long time. So control analysis was made without step signal. Fig.10 is the change of parameters in decoupling control simulation. The bypass wire extension is almost unable to get stable in the simulation, as shown in Fig. 10(b),which is consistent with the actual welding test in literature. 0.016

150 Ibp

100 Ibm

50

0

0

200

400

600

800

Time t/ms


1000

0.10 Lmain

0.08

0.012

0.06 0.008 0.04

Lbypass

0.004

0.000

0.02

0

200

400

600

800


Itotal

200

Current I/A


250

0.00 1000

Time t/ms t/ms

(b) Wire feed speed and change of wire extension

Fig. 10. Change of parameters in decoupling control process

Compared these two scheme, controlling the wire feed speed to stabilize the wire extension is better. For DE-GMAW, suitable sensing mode of wire extension is important. The method through welding voltage to sense wire extension is obvious inaccurate because the power supply for main torch is constant voltage. As to stabilizing the wire extension by adjusting current and wire feed speed of bypass, its response is slow and can’t meet the control requirement. Therefore, it requires more efficient controller and decoupling method, and further research is needed.

292

J. Huang et al.

6 Conclusions By comparing step simulation analysis with existing experiment results, it proved the correctness of the built mathematical model for consumable DE-GMAW. The model could reflect the actual welding process well and reveal the coupling relation of arc system. In consumable DE-GMAW, change of bypass parameters has great influence on stability of the coupling arc. Control simulation results show that adjusting wire feed speed of main and bypass simultaneously to stabilize wire extension can obtain a stable welding process, but further research is needed for sensing of wire extension. Acknowledgement. This research work is supported by National Nature Science Foundation of China (50805073).

References [1] Thomsen, J.S.: Advanced control methods for optimization of arc welding. Aalborg University (2005) [2] Li, C., Luo, Y., Du, C., et al.: Dynamic Model and Parametrical Control Based on the Genuine Length of Wire for Short-circuit CO_2 Arc Welding. China Mechanical Engineering 20(6), 1261–1264 (2009) [3] Li, K.H., Chen, J.S., Zhang, Y.M.: Double-Electrode GMAW Process and control. Welding Journal 86(8), 231–237 (2007) [4] Li, K., Zhang, Y.: Metal transfer in double-electrode gas metal arc welding. Journal of Manufacturing Science and Engineering 129(11), 991–999 (2007) [5] Choi, J.H., Lee, J.Y., Yoo, C.D.: Simulation of Dynamic Behavior in a GMAW System. Welding Journal 80(10), 293–254 (2001) [6] Li, K.: Double-electrode gas metal arc welding: the process, sensing, modeling and control. Kentucky, Lexington (2007)

The Structure Design and Kinematics Simulation for Rotating Arc Sensor of TIG Welding Based on Pro/E 3D Design Jianping Jia, Hongli Li, Wei Jin, and Shunping Yao Mechatronic Institute, Nanchang University, Nanchang, 330031, P.R. China e-mail: [email protected]

Abstract. Considering the features and requirements of rotating arc seam tracking, a new type of sensor is presented in this paper. Through the application of Pro/E 3D modeling and mechanism kinematics simulation module, the structure design of a rotating arc sensor is optimized and some problems caused by Eccentic Mechanism such as vibration and interference were solved.

1 Introduction Kinematic problem can not be separated form Mechanical Product Design. In traditional 2D design, the similar product models are first produced based on required shape and size then checked in actual working environment, modified by result until can meet the requirements. These works waste much time, money and human resources. Pro/E can make the above work more easily with its mechanism kinematics simulation module and shorten the development cycle and cost. The simulation results not only can be shown in the form of animation, but also can be output by parameters, so you can learn whether there is interference between parts, how much the volume of intervention. In this paper through the application of Pro/E 3D modeling and mechanism kinematics simulation module, the structure design of rotating arc sensor is optimized and some problems caused by Eccentric Mechanism such as vibration and interference were solved.

2 Requirements Requirements of TIG sensor: compact size; the systems of cooling water and protective air without entanglement; rotating radius 0 ~ 3mm; light weight. Main movements are 1) Cone swing of Tungsten needle gripper. 2) Horizontal movement of eccentric Mechanism (when adjust the rotating radius).


294

J. Jia et al.

3 Basic Principle The TIG sensor is designed based on cone swing principle and driven by DC Torque Motor, Which is shown in Fig. 1. 1) Electric-air pole (for transmission protective air and electric ). 2) The hollow DC Torque Motor. 3) Eccentric mechanism. The electric-air pole is supported by a self-aligning bearing above through the hollow DC Torque Motor. There is no eccentricity for the above bearing. In the 3 place another self-aligning bearing and a slider body lie in the eccentric Mechanism. By adjusting the location of the slider body, Electric-air pole can be moved an angle. When the hollow DC Torque Motor works, eccentric Mechanism make bottomaligning bearing outer race drive Electric-air pole to move and cone swing movement is formed. The adjustment of rotating arc radius can be gained through change level location of slider body in the eccentric Mechanism. Fig. 1. Cone Swing Principle

4 Design Idea First all parts must be created in Parts creation mode, then skeleton model is established in assembly mode. All the parts of sensor housing were connected rigidly and the parts of core assembly were connected with proper Kinematic Pair. Next all the parts should be assembled together. The design idea is shown in Fig. 2. Rotating arc sensor

Sensor housing

Core assembly

Rotating eccentrics

Fig. 2. Design Idea

Balance body

The Structure Design and Kinematics Simulation for Rotating Arc Sensor

295

In this paper the simulation for core assembly is made. First the mass attribute, gravity, damping, force and torque must be defined. Second the servo motor is defined and then drive motor is simulated. Third the interference is checked according to the dynamic simulation results, and the interference is eliminated. Finally optimal parameters of the balanced block through analysis are given.

4.1 3D-Modeling of Important Parts In this paper only the eccentric Mechanism 3D-modeling is shown in the follow Fig3. The parts are mainly created by the Rotary and stretch. The eccentric Mechanism is composed of four parts (top-down) upper cover slider eccentric bushing under cover. Eccentricity value can be gained through adjusting set screws on the and promoting the slider movement in the slot of Max of Eccentric is set to 2mm.

；②

；③ ③

；④

①

①

4.2 Dynamic Balance of Rotating Parts For the sensor’s design, centre of mass of eccentric Mechanism and electric-air pole is not in rotating axis. When the arc sensor rotates with high speed, centrifugal inertia force generates directional cyclic changes with the rotating of rotary parts. Thus machine itself and its base will vibrate which cause the reduction of the sensor’s accuracy, reliability. For the above reasons a balance cover was needed to realign the mass distribution of rotating parts and fixed bearing on axial.

Fig. 3. The eccentric Mechanism

Because of the rotating parts with irregularly shape and different materials it is very difficult to make balancing calculate. In this paper Pro/E MECHANISM simulation was used to solve this problem.

296

J. Jia et al.

5 Dynamic Simulation Analysis 5.1 Dynamic Simulation Workflow Dynamic simulation workflow is shown in Fig. 4. The connections of core components are as follow: The motor and eccentric mechanism is rigid connection. The electric-air pole assembly is rigid connection. The motor and housing is pin connection. The self-aligning bearings are ball connections. The upper selfaligning bearings and electric-air pole is rigid connection. The lower self-aligning bearings and Electric-air pole is cylinder connection. Table 1 is the mass attributes of the components. Table 1. The mass attributes of the components component

material

density

Electric-air pole and balance block eccentric mechanism others

H62

8.43

45

7.8 1

Parts model

Establish connections

Assembly

Servo Motor

Set environment

Dynamic analysis

Analysis Model

Interference/measure/ curves Fig. 4. Dynamic Simulation Workflow

Get Results

The Structure Design and Kinematics Simulation for Rotating Arc Sensor

297

Force analyses are as follow: Only three contacts lie between rotary assembly and sensor housing: upper self-aligning bearings, two deep groove ball bearing. When rotation frequency and radius were certain, the force of upper self-aligning bearings is only related to the mass distribution of Electric-air pole. It has nothing to do with the size of balance block. Force analysis on two deep groove ball bearings is made in this paper.

5.2 Analysis Results When the playback of the dynamic simulation is shown, the institutional interference must be checked. When the maximum rotating radius is 3mm, electric-air pole had interference with gas mantle. When the parameters are adjusted and the electric pole length is up to 173mm, the interference does not occur. By sensitivity analysis, when the thickness of the balance block was up to 4.87mm, motor bearing force was the smallest. So the optimal parameter of balance block thickness should be designed 4.87mm. Results of the analysis were shown as follow Fig. 6.

Fig. 5. Simulation Results

6 Conclusions In this paper, Proe/E 3D modeling and mechanism kinematics simulation module was used to design the structural of a TIG rotating sensor. The results show that the design method prevented interference phenomena of parts, made a better solution to the movement body in force balance. It saved design cost, shortened design cycle and improved the reliability of the design. Acknowledgement. The authors wish to thank the financial support for this research from the Hi-tech Research and Development Program of China, No. 2007AA04Z242, and Tackle Key Problem Item of Jiangxi Province, China, No. 2007BG09200.

298

J. Jia et al.

References [1] Tong, H., Li, C.: Body motion simulation and dynamic analysis. The People’s Posts and Telecommunications Press, Beijing (2009) (in Chinese) [2] He, Q., Xu, Z.: Product Design and Dynamic Analysis. Mechanical Industry Press, Beijing (2004) (in Chinese) [3] Zhou, E., Xiao, Q.: Virtual Assembly and Kinematics of Gearbox Based on Pro/E Wildfire. Coal Mine Machinery 28(2), 78–80 (2007) [4] Ia, J., Zhang, H., Pan, J.: Development of New Type High Speed Rotating Scanning Arc Sensor Used in Arc Welding Robot. Journal of Nanchang University 22(3), 1–3 (2000)

Knowledge Model Building about a Motor Speed Regulation Fuzzy Control System Huimin Zhao1, Mingyan Ding2, Wu Deng1, Xiumei Li1, and Wen Li1 1 2

Software Technology Institute of Dalian Jiaotong University, Dalian, China Electronic Information School, Dalian Jiaotong University, Dalian, China

Abstract. As fuzzy rules can be properly and effectively extracted from the data even if the data are gained from testing and inconsistent information can be conveniently treated by the established fuzzy system model, the system properties can be objectively characterized by fuzzy systems. This paper discussed the descriptive method, consistency about knowledge model of fuzzy systems, drawn a concept and method of maximum consistency. Some systems, for example an Alternating Current Motor (AC motor), is difficult to devise an accurate mathematical model because of the existence of nonlinear and uncertainty factors. Using sampled data, this paper identified the speed model structure with the method introduced in this paper. The result shows that the structure of the speed model conforms to the impact factor of the actual motor speed, which indicates that describing the frequency conversion speed adjusting system of an AC motor using the knowledge model of fuzzy systems is effective.

1 Foreword Fuzzy rules extracting from system input/output data will improve the objectivity effectively. There are always conflicting or inconsistent rules existing in the extracted initial fuzzy rules. And the incomplete data itself will result the inconsistency or incompleteness in the extracted rules as well as data noise. In order to solve these problem, this paper presents a method for the building of acknowledge model of fuzzy system [1-3]. First of all, this paper defined a knowledge model, and then defined the unitary relation and multiplex relation building on the knowledge model. This paper discussed the concept about categories and knowledge building on the knowledge model of fuzzy system yet, extended the most fundamental and important in rough set to knowledge model of fuzzy system, and gave the approximation in model and the measurement method [4-5]. After that, this paper discussed the consistency of knowledge models [6] of fuzzy system, present the concept about model consistency and the theorem about decomposition of maximum consistency. At last, taking the AC motor system as an example, on the basic of sampling data, with the method introduced in this paper, identified the structure of speed model, to verify the efficiency of the method raised in this paper. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 299–306. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

300

H. Zhao et al.

2 Fuzzy System Model of Knowledge 2.1 Definitions for Model In the system S , which is constituted with n input points points

y1 ,

x1 ,

, x n and m output

, y m . After finite measurements (t) on the input/output of the system,

the input/output value sequence is got as the follows:

x11 , x 21 ,

x12 , x 22 ,

, x1n , , x 2n ,

y11 , y 21 ,

y12 , y 22 ,

, ,

x t1 ,

x t2 ,

,

y t1 ,

y t2 ,

,

x tn ,

after the fuzzy process to the Input / output value

y1m ; ⎫ y 2m ; ⎪⎪ ⎬ ⎪ y tm . ⎪⎭

(1)

xij and y ik ( i=1 ,…, t , j=1 ,…,

n , k=1 ,…, m ), formula (1) changed to (2):

~ A11 , ~ A21 ,

~ A12 , ~ A22 ,

, ,

~ A1n , ~ A2 n ,

~ B11 , ~ B11 ,

~ B12 , ~ B22 ,

~ At1 ,

~ At 2 ,

,

~ Atn ,

~ Bt1 ,

~ Bt 2 ,

~ , B1m ; ⎫ ⎪ ~ , B2 m ; ⎪ ⎬ ⎪ ~ , Btm . ⎪⎭

(2)

The formal definition of fuzzy system knowledge model can be given below: Definition 1. Let S be a fuzzy control system, a knowledge model KM for is defined as the following six member group KM L, X , Y , V X , VY , f

＝〈

＝{ l ,…, l } is the label set of KM ; X = {x ,

〉

S

, x n } is the ~ input space of S ; Y = {y1 , , y m } is the output space of S ; V X { A1 ~ ~ ~ … AN }is the value space of input space X ; VY { B1 … BM }is the value space of output space Y ; f is the mutator from input/output space to their In which: L

，

1

1

t

＝，，

＝，

value space, also

f : L × ( X ∪ Y ) → (U X ∪ U Y )

＝〈

(3)

〉

S be a fuzzy system, KM L, X , Y , V X ,VY , f is a knowledge model for S , Z = X ∪ Y , ϕ ≠ W ⊆ Z . For any z ∈ Z , specify Definition 2. Let

Knowledge Model Building about a Motor Speed Regulation Fuzzy Control System

zˆ

＝{< l , l i

>

j

li

、l

j

∈ L ∧ f (l i , z ) = f (l j , z ) }

301

(4)

zˆ is referred to as unitary relation on label set L among KM .

＝〈

〉

L, X , Y , V X , VY , f be a knowledge model of fuzzy system, Z = X ∪ Y , ϕ ≠ W ⊆ Z . Specify that IND (W ) ∩{ zˆ z ∈ W } IND(W ) is referred to as multiplex relation about W on Label set L among KM . Definition 3. Let

KM

＝

2.2 Categories and Knowledge in the Model

＝〈

〉

KM L, X , Y , V X , VY , f be a knowledge model of fuzzy System, Z = X ∪ Y , ϕ ≠ W ⊆ Z , any equivalence class of L dependent zˆ is referred to as a single-point category in KM . Quotient set L zˆ is referred to as a single-point knowledge. For any l ∈ L , [l ] IND (W ) is referred to as a category about W in KM , and L IND (W ) is referred to as a knowledge Definition 4. Let

／

／

about

W in KM . Apparently: [l i ] IND (W )

＝ ∩[l ] z∈W

i Zˆ

.

2.3 The Approximation and Measurement to Category in Model

＝〈

KM L, X , Y , V X , VY , f be a knowledge model of fuzzy System, Z = X ∪ Y , ϕ ≠ W ⊆ Z , if C is an union of some category about IND (W ) on L , then C is definable with IND (W ) , otherwise, C is indefinable with IND (W ) , or C can also be referred to as fuzzy about IND(W ) . There are two sets, LOW IND (W ) (C ) and UPIND (W ) (C ) , are defined Definition 5. Let

as bellows:

LOW IND (W ) (C ) UPIND (W ) (C )

＝{ l

＝{ l

l ∈ L ∧ [l ] IND (W ) ⊆ C },

(5)

l ∈ L ∧ [l ] IND (W ) ∩ C ≠ φ },

(6)

LOW IND (W ) (C ) is referred to as the lower approximation of IND(W ) of C ; UPIND (W ) (C ) is referred to as the upper approximation of IND(W ) of C .

302

H. Zhao et al.

Therefore,

C is definable with IND(W ) , if and only if LOW IND (W ) (C )

＝

UPIND (W ) (C ) .

＝〈

〉

KM L, X , Y , V X , VY , f be a knowledge model of fuzzy System, Z = X ∪ Y , ϕ ≠ C ⊆ L , ϕ ≠ W ⊆ Z . The degree of definiteness of C about IND (W ) - was written as α IND (W ) (C ) , and a Equation is Definition 6. Let

defined as bellow:

α IND (W ) (C ) =

Card LOWIND (W ) (C )

(7)

Card UPIND (W ) (C )

2.4 Approximation and Measurement of Knowledge in the Model

＝〈

〉／

KM L, X , Y , V X , VY , f be a knowledge model of fuzzy System, Z = X ∪ Y , ϕ ≠ W ⊆ Z , L IND(Y ) { C1 , , C k }. The lower approximation of IND (W ) - of knowledge L IND (Y ) in KM is Definition 7. Let

defined as bellow:

＝ ∪ ( LOW

／

＝

k

LOWIND (W ) ( L / IND(Y ))

IND (W )

(C i ) )

(8)

i =1

The upper approximation of defined as below:

IND(W ) - of knowledge L

＝ ∪ ( UP

／ IND(Y ) in KM

is

k

LOWIND (W ) ( L / IND(Y ))

IND (W )

(C i ) )

(9)

i =1

The degree of definiteness of id defined as below:

IND(W ) -of knowledge L

／ IND(Y ) in KM

k

α IND (W ) ( L / IND(Y ) =

∑ Card LOW i =1 k

(C i ) (10)

∑ Card UP i =1

IND (W )

IND (W )

(C i )


303

3 The Consistency of the Model and the Decomposition of Maximum Consistency

＝〈

〉

KM L, X , Y , V X , VY , f be a knowledge model of fuzzy System S , Z = X ∪ Y . for any l i ∈ L , L is label set, the symbol Definition 8. Let

f li means the sequence as below:

f (li , x1 ),

, f (li , x n ), f (li , y1 ),

, f (li , y m )

(11)

f li is referred to as the number l i rule. Symbol ( f li ↑ X ) and ( f li ↑ Y ) indicate the sequence as below:

f (l i , x1 ),

, f (li , x n ) and f (li , y1 ),

, f (li , y m )

(12)

And they are separately referred to as antecedent and consequent of rule (

＝

＝

li . If

f li ↑ X ) ( fl j ↑ X ) will ( f li ↑ Y ) ( f l j ↑ Y ), then the rule li is consis-

tent. If all the rules in KM are consistent, that means KM is consistent. It can be proved that if KM is consistent, then α IND ( X ) ( L / IND (Y ) = 1 .

＝〈

〉

KM L, X , Y , V X ,VY , f be a knowledge model of fuzzy System S , L = L1 ∪ L0 . KM can be decomposed into two sub-model Propositions 1. Let

＝〈 L , X , Y ,V

〉

＝〈

〉

, VY , f , KM 0 L0 , X , Y , V X ,VY , f . So that α IND ( X ) ( L / IND (Y ) = 1 in KM 1 , and α IND ( X ) ( L / IND (Y ) = 0 in uniquely: KM 1

1

X

KM 0 . Theorem 1. (Maximal uniformity decomposition theorem) Let

〈 L, X , Y , V

＝

KM

〉 be a random knowledge model of fuzzy System S . There must be a maximal sub-set L , it causes that the sub-model KM ＝〈 L , X , Y ,V ,V , f 〉is uniformity. X

, VY , f

1

1

X

1

Y

4 The Building of Knowledge Model of AC Motor Speed Adjustment Fuzzy Control System During this study, we will verify the validity of algorithm by way of 300 group data acquired from AC motor speed adjustment system. The parameters are tabled as below:

304

H. Zhao et al. Table 1. Parameters and Revolution (partial)

x1

x2

x3

x4

x5

x6

x7

y

0 4 12 … 44

2.52 2.27 2.45 … 3.73

2.51 2.85 2.52 … 3.08

0.98 1.37 1.43 … 3.28

2.52 2.5 2.5 … 2.98

2.5 2.52 2.33 … 3.78

1 1.46 1.87 … 1.64

1.01 0.99 1.54 … 1.84

Let the Form of the model of AC motor speed adjustment fuzzy control system be KM L, X , Y , V X , VY , f , input space is

＝〈

〉

X = {x1 , x 2 , x3 , x 4 , x5 , x6 , x 7 } , output space is Y = {y} , in which x1 is frequency, x2 is the Instantaneous value of current 1, x3 is the Instantaneous value of current 2, x4 is the RMS current, x5 is the Instantaneous value of voltage 1, x6 is the Instantaneous value of voltage 2, x7 is the RMS voltage, y is the valid values of revolution. After the fuzzy processing to the acquired data, we got the table 2. Table 2. Fuzzy data table of input-output

No.

x1

x2

x3

x4

x5

x6

x7

y

1

~ A1

~ B4

~ B4

~ B2

~ C4

~ C4

~ C2

~ D3

…

…

…

…

…

…

…

…

…

300

~ A5

~ B5

~ B4

~ B4

~ C4

~ C5

~ C3

~ D5

With the foregoing method, by way of maximum consistency treatment and simplification to the fuzzy model get from table 2, the new consistency model can be obtained as table 3. Table 3. Initial consistent fuzzy model

L 1 … 92

y

x1 ~ A1

x2 ~ B4

x3 ~ B4

x4 ~ B2

x5 ~ C4

x6 ~ C4

x7 ~ C2

~ D3

…

…

…

…

…

…

…

…

~ A5

~ B3

~ B4

~ B4

~ C3

~ C5

~ C3

~ D5


305

By way of the calculating of the approximation degree of knowledge of output space from input space, the effect degree of every input variant to output can be quantitative analyzed. First, we calculate the number of elements in the 8 multiplex relations IND( X ) , IND( X − {x1 }), , IND ( X − {x7 }) in the initial consistency model correspond to the lower approximation of result is shown in table 4.

L / IND(Y ) , the

Table 4. The statistical table of elements

No. of multiplex relation Number of elements

1

2

3

4

5

6

7

8

92

56

80

79

61

73

81

63

By means of the definition of consistency, the consistency point removed can be acquired as table 5.

α

Table 5. The table of approximate degree after removing

Symbol variant

of

Number of elements

with each input

xi

x1

x2

x3

x4

x5

x6

x7

0.61

0.87

0.86

0.66

0.79

0.88

0.68

From the result we can find that frequency, current RMS, voltage RMS have the most effective impact on the parameter speed. That proves that the identified structure of speed model have an agreement with the influential factor of motor speed in real system.

5 Conclusions This paper described a method in building a fuzzy system knowledge model. The definitions of the model, consistency of the model and the completeness, and the theorem about decomposition of maximum consistency and so on were also given. With the proposed method, the parameter speed of an AC Motor is identified, and the result shows that the method in building the model is effective. Acknowledgement. This paper is supported by National Natural Science Foundation of China (Grant NO. 60870009) and Innovative Teem Programs Foundation of Dalian Jiaotong University, China.

306

H. Zhao et al.

References [1] Setnes, M.: Supervised fuzzy clustering for rule extraction. IEEE Trans. on Fuzzy Systems 8(4), 416–424 (2000) [2] Asharafa, S., Murty, M.N.: A rough fuzzy approach to web usage categorization. Fuzzy Sets and Systems 148(1), 119–129 (2004) [3] Çelikyılmaz, A., Türkşen, I.B.: Evolution of fuzzy system models: An overview and new directions. In: An, A., Stefanowski, J., Ramanna, S., Butz, C.J., Pedrycz, W., Wang, G. (eds.) RSFDGrC 2007. LNCS (LNAI), vol. 4482, pp. 119–126. Springer, Heidelberg (2007) [4] Pawlak, Z.: Rough sets. International Journal of Computer and Information Science 11(5), 341–356 (1982) [5] Li, D., Zhang, B., Yee, L.: On knowledge reduction in inconsistent decision information systems. Int’f Journal of Uncertainty. Fuzziness and Knowledge-Based Systems 12(5), 651–672 (2004) [6] Li, W.: Knowledge Model of Fuzzy System and Identification Approach, pp. 16–32. Harbin Institute of Technology (1999)

Research on Surface Recover of Aluminum Alloy PGTAW Pool Based on SFS Laiping Li1, Xueqin Yang1, Fengyan Zhang1, and Tao Lin2 1

Shanghai Spaceflight Precision Machinery Research Institute, Shanghai, 201600 2 Shanghai Jiaotong University, Shanghai, 200240 e-mail: [email protected]

Abstract. Robot and intelligence is the tendency of welding technology. The key to intelligence is, in real-time, to acquire welding pool formation and adjust welding technology to assure the formation uniform. Shape from shading is the effective method to acquire the 3D formation of the welding pool. The reflectance map model of aluminum alloy pulse GTAW welding pool surface, resolution of reflectance map equation and calculation of the welding pool surface height is introduced. The calculation result can testify the characteristics of the welding pool.

1 Introduction The development of modern argon tungsten arc welding technology promotes the application of aluminum alloy welding construction. The conditional manual argon tungsten arc welding technology does not satisfy the high performance product, therefore, the automation and intelligence is the tendency of welding technology [1]. The shape and size of welding pool can indicate the penetration and backside formation. The skilled welder can observe the topside welding pool characteristic parameters, joint format, arc shape and droplet to forecast backside form and size and adjust the welding technology to assure welding quality stabilization. In general, it plays an important role in welding process control to acquire the shape information of welding pool. The passive visual sensing, which can acquire image information of welding zone by arc lighting or natural lighting without auxiliary light source, has widespread application foreground. The shape from shading method is to acquire the object shape information by image gray change [2]. The key to SFS is reflectance map model and reflectance map equation resolution. It needs single image to acquire the height and shape information by SFS, which can be satisfied in the welding pool visual sensing, thus the SFS method is a prospect method to recover welding pool surface. The reflectance map model of aluminum alloy pulse GTAW pool surface, preconditioning cognate gradient method of reflectance map equation, and surface height calculation of welding pool is introduced. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 307–314. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

308

L. Li et al.

2 Welding Pool Image of Aluminum Alloy Pulsed GTAW Fig 1 is the topside welding pool image of aluminum alloy by the broad-pass visual sensing under the condition of table 1.

a) Concave welding pool

b) Convex welding pool

Fig. 1. Typical image of welding pool of aluminum alloy Table 1. Welding technology and imaging parameters of aluminum alloy pulsed GTAW Material

Pulse frequency

Tungsten diameter

Arc height

2A14T62 Size 400*150*2mm shuttle 1/1000s iris

1 Hz Welding velocity 16 cm/min Lens focus 10 mm Broad band filter pass 25%

2.0 mm Pulse current 120 A band filter wave length 590~710nm Lens diameter

5 mm Base current 30 A Imaging period 50 s Neutral filter pass

1/3 inch

10%

5

3 Reflectance Map Model of Welding Pool Surface 3.1 Spherical Light Source Model of Aluminum Alloy PGTAW The reflectance map equation setup the relation between reflectance map model and image gray. The key to SFS method is how to calculate the object surface height z from image gray under certain condition. The nearby point light source reflectance map model proposed by Lee and Kuo is based on the following assumption, such as point light source, hybrid reflectance surface, aperture imaging [3]. The imaging light source in the passive visual sensing is welding arc, whose principle is gas discharge. The light source characteristics of welding arc during GTAW is affected by material, electrode butt shape, welding parameters, arc height [4]. The incident light density of object surface point p of aluminum alloy GTAW is decided as following [5].

Research on Surface Recover of Aluminum Alloy PGTAW Pool Based on SFS

309

(1) Where

C ( I c , λ , r , rs ) is spherical light source function,

(2)

3.2 Object Surface Reflectance Model The aluminum alloy pulsed GTAW surface reflectance is the common result of ideal diffused reflection, regular reflection, mutual reflection and ambient light.

(3)

f d (λ ) is diffused reflectance function, f s (η , λ , i, n, v, h) is regular reflectance function, f int er is mutual reflectance function, β d is diffused reflecWhere

tance factor,

βs

is regular reflectance factor,

β int er

is mutual reflectance factor,

F (i, h,η ) is Frenel item, G (i, n, v, h) is geometric attenuation factor, D(n, h, m) is micro plane normal distribution function.

3.3 Camera Characteristics The reflectance map model of aluminum alloy pulsed GTAW is as follows. (4)

Where R is the model function, d is lens diameter, f is lens focus, n z is the normal direction of x-o-y plane, v is the camera direction, g is the gain factor of CCD camera, b is CCD camera constant. Table 2 is the fixed parameters in pulse GTAW molten pool surface reflectance map.

310

L. Li et al. Table 2. The fixed parameters in pulse GTAW molten pool surface reflectance map

target size (mm × mm)

Gain factor

Constant

Image max size (pixel ×pixel)

Image size (pixel ×pixel)

7.95×6.45 Smooth factor

0.45 Initial weighted factor

0.9 Triangle length (pixel)

811×508 Initial criterion

128×128 Termination iteration criterion

0.3 x-o-y plane normal

1.0 shuttle

0.01 Welding symmetry

(0,0,1)

5

4 Arc attenuation factor 0.3

Frenel item

welding pool diffused reflectance function 95% welding pool mutual reflectance function 95%

welding pool regular reflectance factor 0.3 Seam diffused reflectance factor 1

0.4~0.5 welding pool mutual reflectance factor 0.2

iteration

pool

10-6 Neural filter luminousness

Axial symmetry

3.125%

welding pool roughness factor

welding pool diffused reflectance factor 0.7 Seam mutual reflectance factor

2.0 Seam regular reflectance factor

0

0

According to spectrum of aluminum, the characteristic spectrum is from 320 to 580nm and from 700nm to 800nm, the continuous spectrum from 580nm to 700nm.The range of aluminum alloy pulsed GTAW process broad band filter is from 580nm to 710nm, while the peak luminousness is 25%. The luminous of neutral filter is 10% [5].

4 Resolution of Reflectance Map Equation The other key to shape from shading is resolution of reflectance map equation. The reflectance map model is continuous nonlinear function, so reflectance map model must be discrete and linear to solve equation. The many-to-one relation between surface shape and image gray causes ill-conditioning problem, which can not be solved by single reflectance map equation. The smoothing constraint condition is introduced in the regularization method to cancel reflectance map equation ill-conditioning character. The minimal value method, which is transferred to functional formed by reflectance map equation and constraint condition, is a typical method to solve reflectance map equation. Because the functional extreme problem can by solved by linear equations set, key to shape from shading problem is to solve linear equations set [6].


311

4.1 Reflectance Map Model Discretization and Linearization The reflectance map model must be firstly discretized and linearized to reduce complexity and increase calculation velocity. According to finite element theory, the reflectance map model can be described as the function of triangle climax height

Z i , Z j and Z l . (5)

Where

Lk is discrete.

Thus, the linearization reflectance map equation by the Taylor series expansion follows.

(6)

4.2 Reflectance Map Equation Resolution The minimal value of functional is the resolution of reflectance map equation. The key to minimal value method is constraint condition and resolution of functional extreme. The conventional constraint condition is luminance constraint, smoothness constraint, integral constraint, image gray gradient constraint, unit vector constraint, and so on. The functional extreme problem equals to resolution of linear equations set, thus the shape from shading problem is equivalent to resolution of linear equations set. The regularization method is effective to solve ill conditioning shape from shading problem and to increase robust, where the constraint condition is added to error cost function to improve the ill condition characteristics. The random smoothness surface comes from weighted sum of the thin plate and thin film model. (7)

es1 is thin plate constraint, es 2 is thin film constraint, Z is nodal height vector, B is the random smoothness constraint matrix, B = (1 − μ ) B1 + μB2 , B1

Where

is the match plate of thin plate smoothness matrix, B2 is the match plate of thin film smoothness matrix, μ is smoothness constraint factor, 0 ≤ μ ≤ 1 . When μ is zero, it means thin plate smoothness constraint and the

C 2 continuous curve

312

L. Li et al. 0

surface. When μ equals to 1, it means thin film constraint and C curve continuous surface the following is new cost function combined luminance constraint and smoothness constraint.

(8)

Where

A = Ab + λ s B , Ab is symmetric stiffness matrix, b is load vector., λ s

is smoothness factor,

0 ≤ λs ≤ 1 . (9)

The precision and real-time of shape from shading method is decided by linear equations set, which is only resolved by numerical method. The preconditioning conjugate gradient method is an effective method to resolve the large sparse linear equations set. In shape from shading algorithm, the known characteristic condition, which is section absolute height, object edge in image and corresponding height, object surface symmetry, and so on, can improve the calculation precision.

5 Surface Height Calculation of Aluminum Alloy GTAW Welding Pool The experiential knowledge of welding pool, which is symmetric welding pool, zero edge, positive convex welding pool height, negative concave welding pool height, can improve the calculation result. Fig 2 and Fig 3 is the calculation result and the cross section value with vertical and parallel direction of concave and convex welding pool. The result shows that the calculation value of welding pool can display the characteristics of concave welding pool and convex welding pool.

a) Calculation result of concave welding pool b) Calculation of convex welding pool Fig. 2. Calculation result of aluminum alloy welding pool


313

a) Section height along welding direction of concave welding pool b) Section height perpendicular to welding direction of concave welding pool c) Section height along welding direction of convex welding pool d) Section height perpendicular to welding direction of convex welding pool Fig. 3. Section height of aluminum alloy welding pool

6 Conclusions The reflectance map model of aluminum alloy pulsed GTAW welding pool is set up based on spherical light source model and hybrid reflectance model of a welding pool. The triangle plate element is to discrete and the Taylor series is to linearize the reflectance map model. The smoothness constraint condition is introduced to reduce the ill conditioning and preconditioning conjugate gradient to resolve the reflectance map equation. The height of convex welding pool and concave welding pool can be calculated and shows the shape characteristics. It is necessary for the shape from shading method to further improve. Acknowledgement. This work is supported by the Natural Science Foundation of Shanghai under Grand No. 08ZR1409500 and the Shanghai Sciences & Technology Committee under Grand QB1401500.

References [1] Chen, S.B., Lou, Y.J., Wu, L.: Intelligent Methodology for Sensing, Modeling and Control of Pulsed GTAW, Part I —Bead-on-plate Welding. Welding Journal 79(6), 151–163 (2000) [2] Horn, B.P.K.: Height and Gradient from Shading. International Journal of Computer Vision, 37–75 (1989)

314

L. Li et al.

[3] Lee, K.M., Kuo, C.C.J.: Shape from Shading with a Linear Triangular Element Surface Model. IEEE Trans. Pattern Analysis and Machine Intelligence 15(8), 815–822 (1993) [4] Zhao, D.B., Chen, S.B., Wu, L.: Surface height and geometry parameters for describing shape of weld pool during pulsed GTAW. In: Proceedings of SPIE – The International Society for Optical Engineering, vol. 3833, pp. 91–98 (1999) [5] Li, L.P., Chen, S.B., Lin, T.: The Light Intensity Analysis of Passive Visual Sensing System in GTAW. The international Journal of Advanced Manufacturing Technology (2005) (SCI index source) [6] . Zhang R., Tsai P.S., Cryer J.E. etc., Shape from Shading: A Survey, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol.21, No.8, 690-706, 1999.

Research on Fuzzy-Prediction-Control of GTAW Process Based on MLD Modeling Mingyan Ding1,2, Hongbo Ma1, and Shanben Chen1 1

Institute of Welding Engineering, Shanghai Jiao Tong University, Shanghai 200240, P.R. China 2 Electronic Information School, Dalian Jiaotong University, Dalian, 116028, P.R. China

Abstract. Aiming at hybrid property of GTAW nonlinear welding process, this paper introduces the modeling scheme of a hybrid system described by a MLD (Mixed Logical Dynamical). Regarding the MLD modeling as research methods, a MPC (Model-Prediction-Controller) is designed and the penalty matrix parameters of an FMPC (Fuzzy-Model-Prediction-Controller) are optimized by means of using FC (Fuzzy Control) idea. Simulation results demonstrate that the algorithm is effective and the scheme is able to stabilize the hybrid system on the desired output. Comparing to an MPC, the scheme of FMPC can increase the response of the control system and reduce the on-line computing of parameters. These studies can lay a foundation for the further research on MLD modeling and intelligence control in the welding process control system.

1 Introduction It is a new study that a MLD modeling idea is added to the welding process control system, linear models of a nonlinear system are integrated into a uniform framework. Modeling and control method based on MLD has many applications in chemical process and manufacturing process recently. For example, in Ref. [1], the authors made modeling based on MLD and optimization control in multiproduct batch processes. In Ref. [2], the paper discussed optimization control of hybrid systems with states and inputs constraints. In Ref. [3], it mainly researched on hybrid modeling and predictive control in a multi-tank system. To sum up, this idea can make full use of related knowledge of artificial intelligence, computational intelligence and applied logic etc. It is combined with information of welding industry process, so modeling and optimized control of the progress can be built up. In addition, the welding industry process has always constraint conditions and nonlinear properties. In reference [4], it indicated that the welding industry process could be approximated by linear models, so FMPC algorithm can be applied on the welding nonlinear system based on MLD modeling. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 315–322. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

316

M. Ding, H. Ma, and S. Chen

2 System Modeling The experiment system was shown as Fig 1. GTAW (Gas Tungsten Arc Welding) system was constituted by sensor system, welder equipment, signal conditioning and so on. It is necessary that front welding pool width W f and front welding pool length

L f of welding pool image are acquired. Back welding pool

width Wb based on MLD modeling is on-line calculated. This paper will primarily realize auto-detection and adjusting of welding pulse peak current I p or wire feed speed

V f through MFC algorithm.

Fig. 1. The actual picture of the GTAW system

2.1 MLD Model In reference [5], Bemporad and Morari proposed a modeling based on MLD and hybrid nature of the system was described by the modeling. Basic equation of the MLD modeling can be expressed as follows:

⎧ x(t + 1) = At x(t ) + B1t u (t ) + B2tδ (t ) + B3t z (t ) ⎪ ⎨ y (t ) = Ct x (t ) + D1t u (t ) + D2tδ (t ) + D3t z (t ) ⎪ E δ (t ) + E z (t ) ≤ E u (t ) + E x(t ) + E ⎩ 2t 3t 1t 4t 5t Where t ∈ Z ,

⎡ xc ⎤ ⎡y ⎤ n p x(t ) = ⎢ ⎥ ∈ R nc × {0,1} l ; y (t ) = ⎢ c ⎥ ∈ R pc × {0,1} l ; x y ⎣ l⎦ ⎣ l⎦ ⎡uc ⎤ m u (t ) = ⎢ ⎥ ∈ R mc × {0,1} l u ⎣ l⎦

(1)

Research on Fuzzy-Prediction-Control of GTAW Process Based on MLD Modeling

317

x(t ) is the state of the system, y (t ) is the output vector. u (t ) is command input,

uc and binary commands ul . δ (t) ∈{0,1}r

collecting both continuous commands

l

and z (t ) ∈ R rc represent respectively auxiliary logical and continuous variables. Eq. (1) includes mixed integer inequalities and constraint conditions.

E1, E2, E3, E4, E5 , each matrix has proper dimensions.

2.2 MLD Modeling on the GTAW Welding System The GTAW welding process control is related with continuous decision variables and discrete decision variables, that is to say, the system has properties of hybrid system. This paper makes W f and L f of the welding pool as continuous state variables, at the same time, I p or V f can be represented as continuous input variable. When back welding pool width Wb exceeding threshold, it will arouse logic input signal process.

δ ; Wb is made as continuous output variable for the controlling

First of all, the PWA (Piece-Wise-Affine) modeling of back welding pool width Wb , deriving from reference [5], can be set up. ⎧ A1Wb (t ) + B1 I p (t ) + F1W f (t ) + G1 L f (t )if δ 1 (t ) = 1 ⎪ Wb ( t + 1) = ⎨# ⎪ A W (t ) + B I (t ) + F W (t ) + G L (t )if δ (t ) = 1 n p n f n f n ⎩ n b

(2)

While δ i (t ) ∈ {0,1}, ∀ i = 1, " n are 0-1 variables satisfying the exclusive-or n

condition ⊕[δ i (t ) = 1] ; M = [M 1 , " M n ]' i =1

， m = [m , " m ] ； '

1

n

M j Δ max{max Ai jWb + Bi j I p + Fi jW f + Gi j L f } m j Δ min{max Ai jWb + Bi j I p + Fi jW f + Gi j L f } i =1,"n

i =1,"n

～

Eq. (3 5) can be expressed as MLD modeling, n

Wb (t + 1) = ∑ z i (t )

(3)

i =1

Δ

] z i (t ) = [ AW i b (t ) + Bi I p (t ) + FW i f (t ) + Gi L f (t ) ⋅ δ i (t )

(4)

318


⎧ zi (t ) ≤ Mδ i (t ) ⎪ z (t ) ≥ mδ (t ) i s.t. ⎪ i ⎨ z (t ) ≤ A W (t ) + B I (t ) + F W (t ) + G L (t ) − m(1 − δ (t )) i b i p i f i f i ⎪ i ⎪ zi (t ) ≥ AiWb (t ) + Bi I p (t ) + FiW f (t ) + Gi L f (t ) − M (1 − δ i (t )) ⎩

(5)

Then the modeling structure is identified by least squares method and regression analysis. Datum of the MLD modeling were derived from reference [6], the PWA turn into the modeling based on MLD through HYSDEL language [7], namely, the modeling as Eq. (1) is set up. This part is not going to dwell upon it in detail due to limited space.

3 MPC Predictive control based on model makes MLD model as predictive model [11]. This part mainly considers the problem of optimized control and adopts MPC strategy, which possesses good robustness and has lower the need of model precision than PID strategy.

3.1 Predictive Control of MLD Systems Δ T −1

min J (I Tp−1, Lf (t),Wf (t))=∑ I p (k) − I pe T −1

{I p }

+ Lf (k | t) − Lfe

k=0

2 Q4

+ Wf (k | t) −Wfe

2 Q5

2 Q1

+ δ (k | t) −δe

2 Q2

+ z(k | t) − ze

2 Q3

(6)

+ Wb (k | t) −Wbe Q 2

6

⎧ L f (T | t ) = L fe ⎪ ⎪W f (T | t ) = W fe ⎪ L (k + 1 | t ) = A L (k | t ) + B I (k ) + B δ (k | t ) + B z (k | t ) 1 f 1 p 2 3 ⎪ f ~ ~ ~ ~ ⎪ ⎪W f ( k + 1 | t ) = A1 W f ( k | t ) + B 1 I p ( k ) + B 2 δ ( k | t ) + B 3 z ( k | t ) ⎪⎪W ( k | t ) = C L ( k | t ) + C W ( k | t ) + D I ( k ) + D δ ( k | t ) + D z ( k | t ) 1 f 2 f 1 p 2 3 ⎨ b ⎪ E 2 δ ( k | t ) + E 3 z ( k | t ) ≤ E 1 I p ( k ) + E 4W f ( k | t ) + E 5 L f ( k | t ) + E 6 ⎪ ⎪ I p min ( k ) ≤ I p ( k ) ≤ I p max ( k ) ⎪W ( k ) ≤ W b ( k ) ≤ W b max ( k ) ⎪ b min ⎪W f min ( k ) ≤ W f ( k ) ≤ W f max ( k ) ⎪ ⎪⎩ L f min ( k ) ≤ L f ( k ) ≤ L f max ( k )

(7)

where Q1 = Q1' > 0, Q2 = Q2' ≥ 0, Q3 = Q3' ≥ 0, Q4 = Q4' > 0, Q5 = Q5' > 0, Q6 = Q6' ≥ 0, Δ

L f (k | t ) = L f (t + k , L f (t ), I pk −1 )

(8)


319

W f (k | t ), δ (k | t ), z (k | t ),Wb (k | t ) are similarly defined.

～

Eq. (6 8), deriving from reference [5], are shown as optimal control problem of predictive control based on MLD modeling systems above, let t be the current time. At every time t , the controller calculates a sequence of command inputs by means of on-line optimization procedure, accordingly, the scheme is able to stabilize the hybrid system on the desired output. At time t , only the first command input of the optimal sequence is actually applied to the system, and at time t + 1 , a previous sequence is substituted by a new sequence, so this way of receding horizon has not only the feature of feedback control, but also better conditions for developing intellectualized method of varying-domain optimization [8-9].

3.2 Fuzzy -MLD-PC As described above, the core of predictive control is receding horizon. So the whole algorithm can be regarded as the problem of performance optimization finally. Traditional MPC always adopted optimization methods which were linear and quadratic objective function. It could calculate optimal control law by means of minimizing objective function in control domain. Aiming to the welding control system, traditional method of MPC has bigger calculation. However,in some ways, fuzzy control can make output variable approaching optimal control law by means of selecting a set of parameters Qi ( i = 1,"5 ), etc. So this paper proposes a scheme on combining fuzzy control and MPC, which can absorb both advantages. It utilizes fuzzy control query table where parameters Qi have been off-line calculated, reducing PC-MLD calculation on on-line optimization.

Fig. 2. Structure of MLD-FPC

4 Stability of MLD Systems This paper mainly discusses constraint conditions of the single variable. In order to make sure of stability of MPC based on the MLD system and force terminal state to

320


return to the balance point, Eq. (7) is added terminal equation constraints in the open-loop optimal control problem. In reference [10], it was shown that stability on MPC of terminal equation constraints based on Lyapunov theory is effective.

5 Simulation Based on Fig 3, the control scheme is consisted of two layers: parameters optimization layer and basic control layer. Control steps are as follows: (1). Parameters optimization layer: FC(off-line fuzzy query table of penalty matrix parameters Qi ) 1) At time t , given and feed-back value of back welding pool width Wb are denoted by Wbr (t ),Wbf (t ) . Eq.(9-5.2) are defined as error and error rate.

e(t ) = Wbr (t ) − Wbf (t ) ; ec(t ) = e(t ) − e(t − 1)

(9)

2) Then the error and error rate are normalized, setting universe e , ec as [-1,1]. 3) Fuzzification of variable e , ec . 4) Based on fuzzy rules (eg. if E=PM or PB and EC=PM or PS, then ΔQ1 = NB; ΔQ2 = ΔQ3 = ZE ; ΔQ4 = ΔQ5 = NB; ΔQ6 = ZE ), the fuzzy output Qi (k + 1) can be

offline queried by means of fuzzy rules table. With defuzzification, the precise output Qi (k + 1) are transmitted to optimization controller. (2) Basic control layer: Simulation parameters based on MPC: M=1(time-domain control), P=1(timedomain prediction), N=20(time-domain modeling). Region1: The given of back welding pool width Wbr (k ) as 4mm(region2: The given of back welding pool width Wbr (k ) as 6mm); Parameters Qi ( k + 1) are transmitted to Eq. (6), the optimal sequence I p (k ) can be deduced based on MIQP [12], and then the first command of the optimal sequence is applied to the system at t time. Based on receding horizon principle, the controller repeats the above steps.

Fig. 3. Simulation figure of FMPC and MPC


321

In Fig. 5.1., welding pulse peak current I p creates a step signal transforming from 120A to126A (from region1 to region2), control method is MPC in region2. From this part, control effect of MPC has tiny overshooting number; Control method is FMPC in region1. Control effect of FMPC is shown that it has not only merit of the MPC but also better rapidity. Calculation of the penalty matrix parameters Qi gets simpler by means of off-line queried. From region2, the method of MPC has lower rapidity.

6 Conclusions This paper discusses the modeling and control of nonlinear welding system based on an FMPC. Through the above analysis, it can be seen that the FMPC has minimal overshoot, fast response and good stability. Contrasting the calculation of online optimization between FMPC and MPC, the FMPC can reduce it effectively. At the same time, the control characteristic is better for FMPC. Acknowledgement. This work is supported by the National Natural Science Foundation of China under the Grant No. 60874026, and Shanghai Sciences & Technology Committee under Grant No. 09JC1407100, P.R. China.

References [1] Potocnik, B., Bemporad, A., Torrisi, F.D., et al.: Hybrid modeling and optimal control of a multiproduct batch plant. Control Engineering Practice 12, 1127–1137 (2004) [2] Zhang, J., Li, P.: Optimal control of a class of hybrid systems with states and inputs constraints. Journal of Zhejiang University:Engineering Science 37(2), 139–143 (2003) (in Chinese) [3] Habibi, J., Moshiri, B., Sedigh, K.: Hybrid modeling an predictive control of a multitank system: a mixed logical dynamical approach. In: Proceedings of the 2005 International Conference on Computational Intelligence for Modeling, Control and Automation (2005) [4] Chen, S.B., Lou, Y.J., Wu, L., et al.: Intelligent methodology for sensing, modeling and control of pulsed GTAW. Part1: Bead-on-plate welding. Welding Journal 79(6), 151–163 (2000) [5] Bemporad, A., Morari, M.: Control of systems integrating logic, dynamics, and constraints. Automatica 35(3), 407–427 (1999) [6] Ma, H.: Research on mixed logical dynamical modeling method of robotic aluminium alloy pulsed TIG welding process based on vision sensing. Shanghai Jiao Tong Univsity, 96–102 (2010) [7] Torrisi, F.D., Bemporad, A.: HYSDEL-A Tool for generating computational hybrid models for analysis and synthesis problems. IEEE Transaction on Control System Technology(S1063-6536) 12(2), 235–249 (2004) [8] Clarke, D.W., Mohtadi, C., Tuffs, P.S.: Generalized predictive control, Part1 and Part2. Automatica 23(2), 137–160 (1987)

322


[9] Abonyi, J., Nagy, L., Szeifert, F.: Fuzzy model-based predictive control by instantaneous linearization. Fuzzy Sets and Systerms 120, 109–122 (2001) [10] Michalska, H., Mayne, D.Q.: Robust receding horizon control of constrained nonlinear systems. IEEE Trans. on Automatic Control 38(11), 1623–1632 (1998) [11] Xi, Y.: Predictive Control. National Defence Industry Press, Beijing (1993) (in Chinese) [12] Lazimy, R.: Mixed-integer quadratic programming. Mathematical Programming 22, 332–349 (1982)

Application of Fuzzy Edge Detection in Weld Seam Tracking System Zhenyu Xiong1,2, Wen Wan2 , and Jiluan Pan1 1

Department of Mechanical Engineering, Tsinghua University, Beijing, 100084 2 School of Aeronautical Manufacturing Engineering, Nanchang Hangkong University, Nanchang, Jiangxi 330063 e-mail: [email protected]

Abstract. Vision sensing is one of the most powerful non-contact sensing technologies used for the monitoring and automatic control of welding process. The effectiveness of image processing in vision sensing is the base for successful welding seam tracking. This paper introduces a welding seam tracking system which includes an arc welding robot, an image acquisition unit and an image processing system. An image processing algorithm based on fuzzy theory is presented in detail. Experimental results show that this method can properly detect the edges of welding seams.

1 Introduction With the rapid development of modern automation and artificial intelligence technologies, their application in welding has become a hot research topic. The technology of automatic weld seam tracking using visual sensor has drawn particular attention due to its effectiveness. Currently most weld seam tracking uses a laser as the initiative lamp-house because a laser can provide much energy and high luminance to the surface of workpieces with distortion (Ref. 1). It can distinguish the center position and characteristic information of free-formed weld seams after the images captured by a CCD camera are processed. Edge detecting is an important step in the image processing of vision weld seam tracking. There are some traditional methods for edge detection such as derivative operator, grads operator, Laplace operator and so on (Ref. 2). Some common edge detection technology are summarized and compared in references 3 and 4. Nowadays, technologies using fuzzy logic in image edge detection such as HMF, SRB, FC(Ref. 6), FIRE operator(Ref. 7), minimum fuzzy entropy criterion (Ref. 8) etc., are also in application. This paper develops a feasible method for edge detection and specially studies how to process the images that are captured by a CCD camera. The developed method is then tested in experiments. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 323–330. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

324

Z. Xiong, W. Wan, and J. Pan

2 Weld Seam Tracking System Weld seam image acquisition and processing is the most critical part of a visionbased weld seam tracking system. The schematic setup of the system is shown in fig. 1. After the CCD camera captures "—" form structure light sent out by the laser, it transmits the video signal to the image processing board. The DSP image processing board transforms the analog video signal into digital signal, and in the mean time performs image processing using a fuzzy logic and calculates the median deviation of weld torch and weld seam. The images acquired by the CCD camera and processed by the DSP board are displayed on the monitor in real time. Once the weld seam deviation is obtained, it is possible to calculate the variable to be controlled by applying the control rule. The control variable is then transmitted to the robot through the DSP board. Monitor

DSP Board

CCD

Control cabinet

Laser

Robot

Fig. 1. Arrangement of Weld seam tracking system

3 Edge Detection Principle 3.1 Edge Models Take a 3×3 window (as shown in fig. 2) from an M lines by N rows digital image as the analysis unit. Assume the pixel that needs to be processed is X, and the eight pixels adjacent to X are respectively Xi (i=1~8).

Application of Fuzzy Edge Detection in Weld Seam Tracking System

Suppose the direction of boundary line can only be one of the following: perpendicular edge, horizontal edge, 45 angle edge & negative 45angle edge. In the 3×3 window, an image points will have a sudden value change along the two opposite direction of boundary line. The patterns are graphically represented in fig. 3. (Pixels marked as Xs are boundary points).

X X

X X

X

a X X e

X X

X X X

X X X X X

X X b X X f

X X

X X X

X

X X X

X X X c X X

325

X1

X2

X3

X8

X

X4

X7

X6

X5

Fig. 2. 3×3 windows

X X X d X X

g

X X h

X X X X X X

Fig. 3. Pattern adopted by edge detection

3.2 Fuzzification of Edge Detection In edge detection, whether a current pixel is a boundary pixel or not can be determined by the sudden change of the gray levels of those pixels around it. In the 3x3 window, suppose the Pi expresses the gray level of pixel Xi (i=1~8), then the difference of gray level value between pixel Xi and central pixel X is: Pi= P Pi (i=1 8). In order to decide whether X is the boundary pixel, six Pi are needed. For example, to detect the sudden change of gray level in horizontal direction (fig. 3(a), fig. 3(b)), P1 P2 P3 P5 P6 P7 must be known. If a gradient sudden change exists in the horizontal upward direction, i.e, P1, P2, and P3 are very low, while P5, P6, and P7 are very high, or P1, P2, and P3 are very high while P5, P6, and P7 are very low, pixel X can be determined to be a boundary pixel. Suppose black pixels in a processed image denote possible uniform regions and bright pixels denote possible object contours. Let big & little member functions be expressed as medium positive (MP) & medium negative (MN) respectively in the membership function, and y is the output variable. An operator can be designed by the following fuzzy reasoning: if a pixel belongs to a border region then make it

～

⊿ ⊿

⊿ 、⊿ 、⊿ 、⊿ 、⊿ 、⊿ ⊿ ⊿ ⊿ ⊿ ⊿ ⊿

⊿

⊿ ∣－ ∣

⊿ ⊿ ⊿ ⊿

326


white (WH), otherwise make it black (BL). A group of fuzzy rules for the detection of edge could be expressed as follows:

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

if( P1 is MP) and ( P2 is MP) and ( P3 is MP) and ( P5 is MN) and ( P6 is MN) and ( P7 is MN) then ( y, WH);

⊿

⊿


⊿

As the same principle, the edge at other directions can detected using fuzzy reasoning is:

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿

⊿


⊿

⊿


⊿

⊿


⊿

⊿


⊿

⊿


⊿

⊿

if( P8 is MP) and ( P1 is MP) and ( P2 is MP) and ( P4 is MN) and ( P5 is MN) and ( P6 is MN) then ( y, WH).

⊿

Where and denotes a fuzzy aggregation, white (WH) is a singleton centered on 255. If no border region is detected, the following action is performed: Else (y, BL), where black (BL) is singleton centered on 0. Among them, membership function of input variable Pi is shown in fig. 4, and membership function of output variable is shown in fig. 5.

⊿

Membership function

MS

27 40

MP

f

255

Variable

Fig. 4. Membership function of input variable


327

Membership function

MP

MS

127.5

255

Variable

Fig. 5. Membership function of output variable

3.3 Parameter Choice of Membership Functions It is inevitable for the image signal to be intervened by all kinds of noise during the image acquisition and transmission. When a fuzzy logic is used for image edge detecting, parameter choice of input variable membership functions will have an important effect on the resultant image. Fig. 6 shows the Original image. When the membership function MP and MZ of input variable have small values shown as fig. 7 , the resultant image is as fig. 8. It can be seen that the resultant image includes much noise and the edge can not be decided.

(

)

Fig. 6. Original image Membership function MS

MP

fu

10 25

255 Variable


328


Fig. 8. Edge image by fuzzy inference

When the membership function MP and MZ of input variable have big values (shown as fig. 9) the resultant image is shown as fig. 10. It can be seen that the image dandify more particular and the edge is detected improperly. Membership function MP

MS

35

50

255 Variable



Based on plenty of tests, the values of membership functions MP and MZ shown in fig. 11 have been chosen. The resultant image is shown as fig. 12. From this image, it can be seen that most noise has been eliminated and the particular is reserved, and the edge is detected properly.


329

Membership function MP

MS

27

40

255 Variable



4 Conclusions Fuzzy logic has played a more and more important role in handling indefinite questions. Edge detection of digital images is one example of such indefinite questions. Fuzzification of edge detection described in this paper has simplified the characteristics of the image features and made the edge visible. However, this technology takes more time to process than traditional methods. High speed DSP will be necessary for image acquisition and to achieve real time image processing. Acknowledgement. The authors wish to acknowledge the financial support given by The S&T plan projects of Jiangxi Provincial Education Department.

References [1] Shibata, N., Hirai, A., Takano, Y., et al.: Development of groove recognition algorithm with visual sensor. Welding Research Abroad 46(6), 9–17 (2000) [2] Castleman, K.R.: Digital image processing. Publishing House of Electronics Industry, Beijing (2002) (in Chinese) [3] Sun, H., Zhou, H.X., Li, Z.H.: Study on edge detection technique in image processing. Computer Development and Application 15(10), 7–9 (2002) (in Chinese)

330


[4] Zhou, X.M., Lan, S., Xu, Y.: Comparison of the edge detection algorithms in image processing. Modern Electric Power 17(3), 65–69 (2000) (in Chinese) [5] Tizhoosh, H.R.: Fast fuzzy edge detection. In: Fuzzy Information Processing Society. Proceedings NAFIPS 2002 Annual Meeting of the North American, pp. 239–242 (2002) [6] Russo, F.: FIRE operators for image processing. Fuzzy Sets and Systems 103, 265–275 (1999) [7] Ei-Khamy, S.E., Ghaleb, L., Ei-Yamany, N.A.: Fuzzy edge detection with minimum fuzzy entropy criterion. In: Electra Technical Conference MELECON 2002. Mediterranean, pp. 498–503 (2002)

Part IV

Welding Technics and Automations

Application and Research of Arc Welding Automation in Korea Suck-Joo Na Department of Mechanical Engineering, KAIST 335 Gwahangno, Yuseonggu, Daejeon, 305-701 Korea e-mail: [email protected]

Abstract. This paper covers the research activities carried out in recent years by ALPA laboratory of KAIST for automation of gas metal arc welding processes, and also some industrial applications of welding process automation in Korea. This presentation will show the basic research results of various automation targets such as rotating arc, magnetic arc oscillation, pipe welding and groove welding with acute groove angles. It will also show some examples of welding automation systems implemented in Korean industry such as ship building and automobile industries.

1 Introduction With automation of the welding process a wide variety of sensors are used to detect the weld joints, to track the weld seams, and to monitor the welding phenomena. In particular, tactile, vision, and arc sensors are widely used for automatic weld seam tracking [1]. Among these sensors, the arc sensor is utilized in measuring the voltage and current of welding arcs and coaxially tracking the seams and monitoring the weld quality [2]. Through-arc sensing is widely used for automatic seam tracking because of its many advantages such as the possibilities for real-time control, no auxiliary parts around the welding torch, no need for maintenance and low cost. When using the arc as a sensor, however, it is generally necessary to weave or rotate the welding torch to stimulate intentionally differences in torch height. With conventional repetitive oscillation method the upper limit of the oscillation frequency is about 4-5Hz owing to mechanical restraints. The arc rotation and electromagnetic arc oscillation method enable high-speed rotation or oscillation of the arc over several tens of Hz [3-4]. Since the 1980s, the active vision sensor, which utilizes a CCD camera, laser beam and computer, has been effectively applied in tracking the weld line. The active vision sensor, based on optical triangulation, has a relatively high resolution and is classified into two types according to the beam characteristics: projected sheet of beam or scanned point beam [5]. Although the vision sensor with projected sheet of light is largely influenced by arc noise and the preprocessing time of the image is relatively long, it is widely used in automatic arc welding because T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 333–339. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

334

S.-J. Na

it is cheaper and has a simpler structure than that of the scanning beam. Therefore, much research has been conducted on the application of this type vision sensor with structured light for welding automation [6-8].

2 Welding Research Trend in Korea The statistical data were analyzed from the research papers published in the Journal of Korean Welding and Joining Society for the time period from 2000 to 2008. For some examples, Fig. 1 shows that the field of welding processes is studied more intensively, when compared to the mechanics and metallurgy. In average, more than 50 research papers are published annually. And in Fig. 2, it is shown that the number of research papers has gradually decreased in past years, but has been more or less stabilized in recent years.

Fig. 1. Number of papers in various

Fig. 2. Number of papers in welding area fields of welding published by KWJS published by KWJS

3 Automation of GMAW Process with Controlled Arc A magnetic field externally applied to the welding arc deflects the arc by the electromagnetic force (Lorentz force) in the plane normal to field lines. If an alternating field is applied to a GMA arc, then the arc can be oscillated in the position normal to the direction of welding as shown in Figure 3, and this has been used to improve the weld quality. Exp.(raw data) Exp.(low-pass filtered data)

350

Y

Coil

Yoke X Z

Welding directio n

Magnetic Pole

200 0.0

(a) Without offset Fig. 3. V-groove welding with magnetic arc oscillation

Experiment Simulation 300

] A [ t300 n rer u c g ni dl250 e W

Welding current [A]

X : Arc deflection Y : Electric field Z : Magnetic field

310

Sim ulation

Welding torch

290

280

270

260

0.2

0.4

0.6

W elding tim e [sec]

0.8

1.0

0.0

0.1

0.2

0.3

0.4

0.5

Welding time [sec]

(b) With 2mm offset magnetic arc oscillation Fig. 4. Welding current waveforms with magnetic arc oscillation

Application and Research of Arc Welding Automation in Korea

335

Magnetic arc oscillation changes arc length, which periodically changes the welding voltage and current. An alternating parallel magnetic field causes the arc to oscillate in position normal to the direction of welding, which takes effect like a mechanical weaving. When the welding arc is positioned at the center of groove, the current waveform becomes symmetric. If the welding arc deflects from the groove center, the current waveform changes to the asymmetrical shape. The welding current signals are simulated and compared with experimental ones as shown in Fig. 4. The simulation and experimental results show that the current waveform changes to asymmetrical shape in the V groove welding with 2mm offset distance. The mechanism of arc rotation includes a hollow-shaft motor, an eccentric tip. The electrode wire is deflected circularly by an eccentric tip that is rotated by the hollow shaft of rotating motor. This mechanism can be installed inside the electrode nozzle and connected directly to a conventional torch system.

(a) Fillet welding with centered torch

(b) Fillet welding with offset torch

Fig. 5. Geometry of fillet welding with rotating torch and the current signals

The welding torch is thought to be at centered position if its rotation axis is positioned at the angle bisector of the fillet structure. Otherwise, the welding torch is thought to be offset. Figs. 5(a) and 5(b) show the cross-section of the fillet welding and their corresponding current waveforms with a centered or offset rotating torch, respectively. The current waveform shows the symmetry at left and right half cycle when the torch is at the centered position (Fig. 5(a)). If the welding torch is deviated from the angle bisector of the fillet joint, the current becomes asymmetric (Fig. 5(b)). The difference between the average current values of the welding current of the left and right half cycle (I left − I right ) shows a linear relationship with the offset distance as shown in Fig. 5(b) and can be used for the torch offset detection during seam tracking. In order to achieve the full penetration of a thick plate, the root gap and groove angle must be adequately selected in the process of butt welding. The deposited metal decreases and productivity increases when the groove angle becomes smaller, but the preceding of the weld pool results in weld defects such as incomplete penetration and a lack of fusion at the groove face. For this reason, a wide range of groove shapes is tailored to different welding conditions.

336

S.-J. Na

(a) Joint with 45o

(b) Current for 3 mm

(c) Current for 8 mm root gap

Fig. 6. Pulsed GMAW with acute groove angles

Fig. 6 shows the current waveforms obtained during torch weaving with a root gap of 3 mm and 8 mm at a groove angle of 45°. When compared with the case of large root gap of Fig. 6(b), the welding current decreases and increases rather gradually for the case of short root gap, Fig. 6(a). This is probably due to the fact that the transferred droplets flow into the root from the groove face to fill in the root gap as the welding proceeds, especially for the case of short root gap. For the case of large root case, this effect disappears, and the rapid change of welding current appears, Fig. 6(b). This effect of root filling by molten droplets decreases the average current value on root face as the root gap increases.

4 Automation and Monitoring with Vision Sensor A vision sensor was adopted firstly for seam tracking, and secondly for weld defect monitoring in pipe welding. Seam tracking image process algorithm was composed of image acquisition, median filter, threshold, thinning, Hough transform and 3D calibration. Maximum error between real coordinate and calculated coordinate was 0.23mm by using the calibration matrix. Monitoring process was used to find out the weld defect and also used to reconstruct the 3D surface profile. The purpose of process monitoring is to find the weld detect from bead shape. By comparing the result from monitoring process with ISO 5817 criterion, it could be easily applied to find out the weld defects. Fig. 7 shows the result of the bead shape monitoring. It seems that the bead profile from monitoring is very similar to the experimental bead shape. Maximum error between two bead shapes is approximately 0.2mm. Monitoring process takes100~150ms for 1 frame, because it requires more time than seam tracking due to more complicated algorithms.


(a) Excessive convexity: B, Excessive symmetry fillet weld: D, Excessive throat thickness: B

337

(b) Excessive convexity: B, Excessive symmetry fillet weld: C, Excessive throat thickness: B

Fig. 7. Comparison of experimental bead shape with the result from vision sensor

Fig. 8(a) shows the schematic diagram of automatic seam tracking system with a vision sensor developed for the corrugated membrane of LNG tanks and shipping containers. The apparatus comprised a robot which had two rectilinear coordinates, a torch rotation mechanism, a vision sensor composed of a CCD camera, a diode laser of 690 nm wavelength and a narrow band pass filter, and a PC for processing the image data and driving the robot.

(a) Schematic diagram

(c) Separation angle of 20o

(b) Cross section of LNG tank

(d) Separation angle of 30o

Fig. 8. Automatic seam tracking system with a vision sensor for corrugated membrane

Fig. 8(b) shows a typical corrugation shape which is prevalent in construction of LNG tanks. The shape of LNG tank corrugation consists of two linear parts and five circular arcs, and can be divided into seven segments from S1 to S7. Each

338

S.-J. Na

segment of LNG tank corrugations could be expressed as the function of some parameters such as coordinates of the start point of line segment (or coordinates of the center point of circular arc segment), the length of line segment (or the radius of circular arc segment) and the angle of tangential line at the start and end point of segments. Fig. 8(c) and (d) show the simulated results for corrugated membrane of LNG tanks at the weld line, when the camera and laser diode of the vision sensor are located symmetrically about the vertical axis and have different separation angles. It is shown that the data deficiency occurred mainly at the corrugation corner and flat part of the corrugated sheet. The result also shows that the total separation angle between camera and laser beam of vision sensor largely affected its measuring efficiency for the corrugated sheets of LNG tanks.

5 Welding Automation in Industry Some examples of welding automation in Korean heavy and automobile industry will be shown in the oral presentation.

6 Conclusions This paper firstly showed the welding research trend in Korea by statistically analyzing the research papers published in the Journal of Korean Welding and Joining Society. And then some research results of automation and monitoring of GMAW processes were analyzed, which were developed at ALPA laboratory of KAIST. To improve the weld quality, the controlled arc welding system such as high speed arc oscillation and rotation method was developed and applied for various targets. For that purpose two main topics are investigated. The first one is about the automatic seam tracking sensor developed with a specially devised mechanism. These controlled arc welding systems ensure more accurate seam tracking with increased sensitivity and responsiveness and can be implemented for steel and Al alloy welding. And the second one is the improvement of their bead characteristics. Pulsed GMAW process was investigated for the V-groove joint with acute groove angles to decrease the amount of weld deposit and consequently to increase the welding speed. It was found that the arc sensor model requires detailed information of bead formation, especially in the case of small groove angle. A successful application of laser vision sensor was also demonstrated for automatic seam tracking and process monitoring in pipe welding. Finally some examples of welding process automation in Korean ship building and automobile industry are demonstrated. Acknowledgement. Support by the Brain Korea 21 project is gratefully acknowledged.


339

References [1] Nomura, H.: Sensors and Control Systems in Arc Welding, pp. 1–7. Champman &Hall, London (1994) [2] Cook, G.E., Andersen, K., Fernadez, K.R., Sheperd, M.E., Wells Jr, A.M.: Electrical arc sensing for robot positioning control. Robotic Arc Welding IFS Publication, pp. 181–216 (1987) [3] Kim, C.H., Na, S.J.: A study of an arc sensor model for gas metal arc welding with rotating arc - Part 1: dynamic simulation of wire melting. PIME Part B 215(B9), 1271–1279 (2001) [4] Kang, Y.H., Na, S.J.: A study of the modeling of magnetic arc deflection and dynamic analysis of arc sensor. Welding Journal 81(1), 8/s–13/s (2002) [5] Yoo, W.S., Na, S.J.: A study on the vision sensor using scanning beam for welding process automation. Trans. of KSME 20-A(3), 891–900 (1996) [6] Yu, J.-Y., Na, S.-J.: A study on vision sensors for automatic welding of height-varying weldment, Part 2: Applications. J. of Mechatronics 8, 21–36 (1998) [7] Agapakis, J.E., Katz, J.M., Koifman, M., Epstein, G.N., Fridman, J.M., Eyring, D.O., Rutishauser, H.J.: Joint tracking and adaptive robotic welding using vision sensing of the weld joint geometry. Welding Journal 65(11), 33–41 (1986) [8] Richardson, R.W.: Robotic weld joint tracking systems - theory and implementation methods. Welding Journal 65(11), 43–51 (1986)

Offline Programming for a Complex Welding System Using DELMIA Automation Joseph Polden, Zengxi Pan, Nathan Larkin, Stephen Van Duin, and John Norrish Faculty of Engineering, University of Wollongong, Wollongong, NSW, 2530, Australia e-mail: [email protected]

Abstract. This paper presents an offline programming (OLP) system for a complex robotic welding cell using DELMIA Automation. The goals of this research are aimed at investigating the feasibility of taking a commercially available robotic simulation package, DELMIA, and to use a Visual Basic Automation interface to reduce the programming time by creating automated ‘modules’ to carry out some of the tasks in the OLP process. The paper first investigates and presents the structure of OLP as a discreet method of individual steps. These steps are then evaluated for their potential as an automation candidate. The methods in which these steps are automated are then presented. A general analysis of the developed OLP system was carried out, providing a scope for future research and development.

1 Introduction A manufacturing facility in Australia has, over the last number of years, been tasked with handling the manufacture of an automobile hull. The vehicle hull is composed of steel plates, and is constructed in a monocoque type assembly with all of the various plates that make up the hull being welded together by the same way as a ship vessel. At the time of writing, the welding of the hull was being carried by manual processes alone; however an increase in future production demand has pushed the manufacturing facility to automate the welding processes on the hull via implementation of a complex robotic welding system. Due to the high number of seams to be welded and the complex geometry, which is inherent in the hull’s design, a specialized robotic cell was needed to maximize the number of external and internal seams that could be completed by the cell. This cell consists of two 6-DOF articulated welding robots. Each of these robots is then in-turn mounted on another, larger, 6-DOF ‘auxiliary’ robot and linear rail to create a form of homogenized 13-DOF robot, as shown in Figure 1. The hull itself is also mounted on a rotating trunnion to allow the welding robot access to areas such as the roof of the hull, or allow the weld robot better internal access through the opening such as the windscreen orifice. The robotic cell features appendages such as laser scanner and heat sensors for calibration purpose. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 341–349. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

342

J. Polden et al.

The robot-on-robot set up required a state-of-the-art communication system to ensure each robot was interfaced and could communicate correctly. The robotic cell was originally programmed via online jog-and-teach method. However the highly complex nature of the cell is a great hindrance to an efficient programming solution. A lot of time was invested in teaching the welding robot all the seam locations, as the extra degrees of freedom added by the auxiliary robot removed a lot of intuitiveness in manually jogging the robot to a specific target location without clashing with the vehicle hull; particularly when negotiating through complex internal geometry.

Fig. 1. A model of the homogenized 13-DOF robotic welding system, featuring two separate 6-DOF robot and a linear rail

The manufacturing company is now anticipating orders of other configurations of the vehicle, meaning that the robot cell will need to be reprogrammed to accommodate these various models of the vehicle. After the initial difficulties experienced when utilizing the online programming method; it was deemed necessary to explore options in which the robotic cell can be re-programmed for these new designs in a much more efficient manner. Researchers at the University of Wollongong proposed an offline programming approach as an alternative, hoping to create an automated programming system utilizing a simulation package widely used in industry today. At the outset of the project, a literature review of current OLP software was conducted [1]. The review indicated that OLP software mainly came from 3 sources; from generic robotics software producers, from robot manufacturers [2-5] and finally from research institutions that produce their own programming and simulation software, usually developed around existing CAD software such as AutoCAD or SolidWorks [6-9] or from scratch using OpenGL, VRML and Java technology [10-12]. To create an OLP software package for this complex welding cell, it was decided to utilize a commercially available generic robotic software package. This was chosen as a generic system can be much more flexible in its compatibility with various brands of system hardware and also features virtual reality, allowing the user to be fully immersed into the simulation environment. The DELMIA software package was chosen. Its current and widespread use in robot programming for industry and manufacturing processes, along with its Visual Basic

Offline Programming for a Complex Welding System Using DELMIA Automation

343

programming interface, were major factors behind its selection. Details of the DELMIA robotics simulation software package are covered below, in section II.

2 DELMIA for Offline Programming and Automation DELMIA is a robotics and manufacturing based OLP package that is utilized widely within various respective manufacturing industries today. DELMIA utilizes a 3D simulation environment to test and optimize robot programs before implementation into real world applications. DELMIA features assorted ‘toolboxes’ that are available to provide a programmer with numerous functions which are useful to the various specific areas of robotic OLP, such as; robot target definition, reachability analysis, clash testing, path planning/augmentation etc. A user carrying out the programming of a robotic cell with DELMIA would follow the general steps for OLP highlighted in Figure 2.

Fig. 2. Block diagram of overall offline programming structure [1]

To aid in the programming of the complex welding cell in question, it is proposed that some of the steps in the above OLP process be automated. These areas related to 3 specific sections of the OLP process: The automatic extraction of seam data from CAD models, the automated generation of reachability and collision assessments and the automatic creation of the robot process with simulation. The proposed automation of these DELMIA functions was carried out with the Visual Basic interface that comes with DELMIA. DELMIA’s robotics related commands which were not accessible through the VBA functionality were controlled via the windows GUI automation program; AutoIT.

3 Automatic Seam Extraction Defining the weld seams to be carried out by the robotic cell is the first step in the OLP process. In DELMIA, these seams are defined by first loading a 3D CAD model of the work object to be welded. The programmer identifies a seam, and then defines it by individually allocating two separate tags at each end of the specific seam. The tag’s individual XYZ orientation angles are then augmented by the

344

J. Polden et al.

user to specify the correct approach angles for the manipulator/weld-torch. This process is not a difficult one, however when a high number of seams are to be defined by the user, it has proven to be a monotonous and time-consuming task. The VBA/DELMIA interface was identified as a tool, which can analyze the drawing features that make up the CAD model of the work object. A programming module was created to aid the programmer in efficiently defining these weld seams. The module assists by providing a ‘semi-automated’ method of defining the weld seam and then automatically snapping the tags to the seam. The semi automated approach reduces the time taken to define a seam by first prompting the user to select the edge they wish to be defined as a seam. The tags are then attached automatically to each end of the selected seam. The user is then prompted to click the two adjacent faces that make up the seam, this is done to define approach/orientation angles for the weld torch as it carries out the weld process, as seen in Figure 3 below. The semi-automated approach developed provides the programmer with a more effective interface to tag their work objects than standard methods available in DELMIA. This is due to the fact that standard methods for tagging in DELMIA require that the user first add two separate tags in the correct location to define the seam, and then they are able to orientate it for the correct approach angles. The semi-automated approach, however, already assumes you are searching for an edge to define as a seam; once the edge is selected by the user the tags are added and orientated automatically. This significantly reduces the amount of individual operations the user has to carry out in order to define a seam, hence cutting down the overall time required. Initial tests on tagging a vehicle hull indicate that the time taken to define each seam in the entire hull was more than halved when using the semi automated approach over standard methods currently available in DELMIA.

Fig. 3. The semi-auto hull tagging module defines the seam edges and approach/orientation angles for the weld torch

Once the seams have been defined by the user, the data relating to their position and orientation is automatically added to an excel spreadsheet for further use at a later stage of the OLP process.


345

4 Path Planning Once the programmer has defined all of the seams within the vehicle that are to be welded they then move onto the next programming task, which relates to planning the motions required for the robotic cell to correctly carry out the welds. To undertake this task with the complex welding cell being used in this research, the programmer first has to program the auxiliary robot to carry the weld robot to a specific point in space which fulfils two criteria. Firstly, this point in space has to be close enough to the seam to ensure that the weld robot can reach its target. Secondly, specific orientations for the weld robot have to be selected so that it can carry out the welding of the seam without colliding with the work object. As each seam to weld has different locations and orientations in space; the programmer essentially has to find a new point in space that for fills the above criteria for each individual seam that is to be welded. Due to the high number of individual seams that feature on the vehicle in the complex welding cell, the repetitive nature of finding these points becomes monotonous and time consuming. However the repetitiveness of the task also highlighted that it is an ideal candidate for automation, meaning that a lot of time to program the cell will be saved if it were possible to control DELMIA in a way so that it can define automatically for the programmer these points in space, meaning that they don’t have to spend the time to find it 'manually'. The approach for automating this task was to address the two criteria mentioned above as separate modules. The first, which related to automatically finding points in space which fulfilled reachability criteria, was addressed before moving onto the second criteria of defining the correct motions and orientations for clash free motion when welding the seams. To automate the reachability assessments, a 3D array of potential positions for the auxiliary robot is defined by the programmer in an excel spreadsheet; the user can define the position of the array, the volume it occupies and the number of nodes or targets within the array. The inverse kinematics of the weld robot were mapped in Matlab, which utilized the previously defined excel data to calculate the reachability of the weld robot from each test node in the array to the previously defined weld seams within that area. These reachability results are fed back to the spreadsheet so that the programmer can see how many seams can be welded from each potential target in the 3D array of nodes. DELMIA is capable of carrying out reachability tests; however the VBA interface in DELMIA is restricted in its access to this class of commands, making automation of these commands with the VBA interface a non-functional pursuit. Matlab was used and the M-file created was able to automatically carry out these tests in a very fast and efficient manner. The results displayed to the programmer were offered in an intuitive yet detailed presentation. The programmer is able to see which nodes can provide the weld robot with reachability to a single seam, multiple seams (which are listed) or no seams at all. Data relating to where the node is in 3D Cartesian space is also displayed, along with the orientation (roll/pitch/yaw angles) of the node. After the reachability testing is concluded, the next 'module' is to automatically test whether the weld robot can move from the specific pre-defined nodes to the

346

J. Polden et al.

weld seam without clash. To automate these commands, the DELMIA program was interfaced with AutoIT, which is a program designed to control and operate with the windows GUI; meaning that instead of a human operator manually clicking through the DELMIA commands and testing clash, AutoIT was programmed to carry out the same operations. AutoIT was programmed to read the excel spreadsheet utilized in earlier sections of this research. It reads the calculated reachability data and then prompts DELMIA to move the position of the weld robot to each node previously deemed to have a positive reachability result. The robot is then commanded to move to a specified seam that has been deemed reachable at that particular node. As this motion is carried out, DELMIA monitors for clash. AutoIT cycles through each node and returns the clash results back to the excel spreadsheet. Each available robot configuration is also tested for clash, and the data relating to which specific configurations provide a clash free motion is also displayed in the excel spreadsheet of results. The result of these automated tests is essentially an array of targets for the auxiliary robot to carry the weld robot to. The robot programmer can be safe in the knowledge that these targets will firstly have at least one seam that is within reach of the weld robot from the particular array node. These nodes will also provide a clash free motion for the weld robot as it carries out its weld process on the seams. For optimisation, the programmer can easily check the created excel spreadsheet to select which nodes will be utilized in the final robot program. This is done easily and intuitively as they are provided with the data relating to the seams that are reachable from specific nodes; so they can easily select the fewest individual nodes required to carry out all seams within one particular area of the hull. This will effectively cut down the number of times the weld robot will have to be repositioned within the hull as it carries out its weld processes. Once the correct positioning for the weld robot and suitable configurations for clash free motion have been defined the robot tasks for each robot is created using this data. At the users request, the complete process is simulated from the beginning to verify that the procedure is carried out correctly and without clash. Once the process is verified, the Visual Basic interface exports the native robot programs to a folder on the computer desktop. If, for some reason, during the verification process an error such as clash or unreachability occurs then an error file is also exported with the program. This text file contains the nature of the error and the simulation time at which it occurred, making it easy for the programmer to trace through the simulation and fix any problems before exporting the program again.

5 Experimental Results The effectiveness of the created automation system was tested on a number of seams on the vehicle hull. Visual Basic was used to create a user interface, in which the programmer has access to the various buttons and controls that manage the operation of the OLP automation system created. The first automation control the user utilizes in the OLP process is the automated generation of the seam targets for the welding robot. The proposed seams to be created were located in an


347

internal section at the rear of the hull in order to fully test the capabilities of the automation system. The user first enters a desired seam name into an input box on the Visual Basic user control panel, and then clicks the ‘Create Tags’ button. A blank model of the vehicle hull in a new DELMIA window with all of its drawing features exposed to the user is then opened. A pop up window prompts the user to select the edge between two plates that will form the first seam that is to be defined. The user is then prompted to click on the incident faces to this selected edge. Once the faces have been selected the blank model of the vehicle hull is closed and two tags are placed at each end of the previously selected edge on the hull in the complex welding cell. The previously selected faces automatically orientate the seam tags with the appropriate approach angles for the weld torch, which includes any push/pull angle if required in a corner. These tags are then automatically renamed to the previously defined seam name and are saved into a specific ‘semi-auto’ tag group in DELMIA. The seam is then fully defined and the user can now move on to repeat the process as many times as they want. Three seams were defined inside the rear section of the hull using the above method. The ‘export seam data to excel’ button on the user control panel was used to export all data relating to the seam tags locations/orientations to a specific Excel spreadsheet. A list of seams in the excel sheet was created, and the new tags imported are added to the end of this list. This data is used during the later stages of the OLP automation process. Another button on this tab has the capability of importing this list of tags back into the DELMIA environment if required, allowing the user to make modifications to the tag’s location/orientation in the excel environment and then import these modified tags back into the simulation model. Once the Seams have been defined the user then clicks on the ‘Path Planning’ tab on the user interface to expose the next set of automation controls. Before beginning the path planning automation functions, a matrix of targets for the Auxiliary Robot to place the Weld Robot has to be defined about the rear of the Hull. This is done in the same spreadsheet as the stored seam tag data. The user defines this matrix by specifying in a specific section of the excel sheet the upper left and lower right hand corners of the desired positioning matrix and also entering the desired number of nodes in the matrix. These coordinates are found using the DELMIA simulation model and the DELMIA compass tool. The user selects which robot programs they would like to generate as a result of these automated tests, in this instance the button to generate both the weld robot and auxiliary robot programs was selected. The programmer can also check the ‘Validate with Simulation’ box, which runs a final validation check on reach and clash at the end of the OLP process. All that remains for the user to do is enter the name of the weld seam they wish to create the programs for and utilize the ‘Generate Code’ button on the user interface to initialize the Matlab, Visual Basic and AutoIT automation components listed in section IV. After validating the process with simulation from start to finish the final output is two separate robot programs. The first program commands the auxiliary robot to move the weld robot to a suitable position close to the weld seams. The second robot program commands the weld robot to move from this base position to carry

348

J. Polden et al.

out the seam weld without clashing with the vehicle hull. If desired the user can at this stage make ‘manual’ adjustments to the robot task within the simulation environment by using traditional DELMIA commands and then export the modified robot programs again. Once the first seam has been completely programmed, the operator then moves on to obtaining the robot programs for the remaining two seams at the rear section of the hull. This is done by inputting the next remaining seam name into the user control panel and then clicking the ‘Generate Program’ button again. Once the program has finished obtaining the code for these seams, the final seam at the rear of the hull was addressed in the same fashion. The time to obtain the programs to correctly weld these three seams took the automation module approximately 5 minutes from start to finish. This result provides a significant improvement over using traditional ‘online’ jogging methods currently employed by the manufacturing facility.

6 Conclusion and Future Work This research has shown that it is possible to modify currently available simulation software to automate some of the steps in the OLP process. The developed modules provide automation functionality to both the tag generation and trajectory planning stages of the OLP process, giving an overall improvement to the time taken for a programmer to produce code for a robotized welding cell. Whilst the created system is able to provide a level of success in delivering a working package, there are a number of noted issues, which have a negative impact on its operation. The main issue negatively affecting the process relates to the level of functionality that is exposed in the DELMIA/VBA interface. DELMIA, in its current development state, provides access to the majority of its functions via the VBA programming interface. However, access to DELMIA’s robotics related functions was minimal. This resulted in having to use indirect methods, such as accessing the commands through DELMIA’s GUI rather than getting direct access to use certain functions or commands. Examples of this inaccessibility include restricted access to the reachability and clash checking commands. Whilst the overall speed is a significant improvement over ‘manual’ methods of offline programming, there exists room for improvement in the developed OLP automation package. To overcome this, work has begun on creating new OLP modules in Matlab to replace some of the tasks in the OLP process, removing the reliance on the slower VBA/DELMIA interface in much the same way Matlab was implemented in section IV to carry out the reachability assessments of the weld robot. The main goal behind this is to improve the efficiency of these tasks, whilst also improving the reliability of the program by moving the role of DELMIA/VBA towards being just a tool for simulation. Acknowledgement. This work is funded by the Australian Defence Materials Technology Centre (DMTC).


349

References [1] Pan, Z., Polden, J., Larkin, N., Van Duin; Norrish, J.: Recent Progress on Programming Methods for Industrial Robots. In: ISR/ROBOTIK, Munich, Germany, June 7-9 (2010) [2] Brown, R.G.: Driving digital manufacturing to reality. In: Proceedings of 2000 Winter Simulatin Conference, December 10-13, vol. 1, pp. 224–228 (2000) [3] Bruccoleri, M., D’Onofrio, C., La Commare, U.: Off-line Programming and simulation for automatic robot control software generation. In: 5th International Conference on Industrial Informatics, June 23-27, vol. 1, pp. 491–496 (2007) [4] Dong, W., Li, H., Teng, X.: Off-line programming of Spot-weld Robot for Car-body in White Based on Robcad. In: International Conference on Mechatronics and Automation, ICMA 2007, August 5-8, pp. 763–768 (2007) [5] Lee, D.M.A., Elmaraghy, W.H.: OBOSIM: a CAD-based off-line programming and analysis system for robotic manipulators. Computer-Aided Engineering Journal (October 1990) [6] Pries, J.N., Godinho, T., Ferreira, P.: CAD interface for automatic robotic welding programming. Industrial Robot: An International Journal 31(1), 71–76 (2004) [7] Mitsi, S., et al.: Off-line programming of an industrial robot for manufacturing. International Journal of Advanced Manufacturing Technology 26, 262–267 (2005) [8] Soron, M., Kalaykov, I.: Generation of continuous tool paths based on CAD models for Friction Stir Welding in 3D. In: Mediterranean Conference on Control & Automation, MED 2007, June 27-29, pp. 1–5 (2007) [9] Yang, Y., Chen, X., Ling, C., Kang, B.: A Robot Simulation System Basing on AutoLisp. In: 2nd International Conference on Industrial Electronics and Applications, ICIEA 2007, May 23-25, pp. 2154–2156 (2007) [10] Dai, W., Kampker, M.: PIN-a PC-based robot simulation and offline programming system using macro programming techniques. In: The 25th Annual Conference of the Industrial Electronics Society, November 29 –December 3, vol. 1, pp. 442–446 (1999) [11] Jaramillo-Botero, A., Matta-Gomez, A., Correa-Caicedo, J.F., Perea-Castro, W.: ROBOMOSP. IEEE Robotics & Automation Magazine 13(4), 62–73 (2006) [12] Kim, C.-S., Hong, K.-S., Han, H.Y.-S., Kim, S.-H., Kwon, S.-C.: PC-based off-line programming using VRML for welding robots in shipbuilding. In: IEEE Conference on Robotics, Automation and Mechatronics, December 1-3, vol. 2, pp. 949–954 (2004)

Robot Path Planning in Multi-pass Weaving Welding for Thick Plates Huajun Zhang1,2,3,*, Hanzhong Lu2, Chunbo Cai3, and Shanben Chen2 1

School of material science and technology, Shanghai Jiao Tong University, Shanghai 200240, China 2 Shanghai Zhenhua Heavy Industries Co., Ltd., Shanghai 200125, China 3 School of material science and technology, Harbin University of Science and technology, Harbin 150040, China e-mail: [email protected]

Abstract. Multi-pass weaving welding is usually used in thick plates. Automatic path layout of multi-pass weaving welding is a key technology to realize robot automatic welding for thick plates. In this study, a user self-define path layout model is developed to realize teach offline programming for multi-pass weaving welding. Users can set the welding parameters of every pass, the number of layers and passes, and welding sequence according to the real welding technology. The developed system can produce the position, pose of welding torch and oscillation displacement of total passes automatically. The multi-pass welding shape is well and meets the actual production requirement through robot multi-pass weaving welding experiments for single V-groove thick plates.

1 Introduction Large thick plate structures are widely used in shipbuilding, high pressure vessel and heavy-duty machinery and so on [1]. At present, these thick plate structures adopt manual and semi-automatic arc welding. It is very urgent to realize robot automatic welding. Multi-pass weaving welding is usually used in thick plate. Automatic path layout of multi-pass weaving welding is a key technology to realize robot automatic welding for thick plate. For thin and medium thick plate, robot teach online is adopted usually. But for large thick plate, It is difficulty for this way of teach online to realize robot welding [2-3]. At present, off-line programming is used to layout the multi-pass welding path. In 1987, M. Masaharu adopted equal area method to arrange the multi-pass welding path through research to the effect of welding parameters to welds fillet [4]. L.X. Sun and K. Li adopted this method also [5-6]. In 2005, X. H. Tang developed offline simulation system of multi-pass basis on three-dimensional graphics [7]. Recently H.J. Zhang designed a self-defining layout model for multi-pass welding basis on visual graphics [8]. Seen from above researches, path layout with *

Corresponding author.


352

H. Zhang and Shanben Chen2

multi-pass weaving welding was not reported. Therefore, it is very significant for multi-pass weaving welding to carry out intensive study so as to meet the production requirements. It is well known that on-line teach of multi-pass weaving welding is slow and difficulty. Therefore, off-line teach module based on graph is developed. A selfdefine mathematic model of path layout is deduced. Users can design layer number, pass number and welding parameter of multi-pass welding according to welding technology. Path parameters such as position, attitude, weaving amplitude of weld torch and so on can be generated automatically. So the off-line teach programs of robot for multi-pass welding are obtained quickly and accurately, which will raise the programmable efficiency of robot for multi-pass welding of large thick plate.

2 Establishment of Offline Simulation System In this study, Visual C++ software is used to develop the offline simulation system in SOLIDWORKS software interface. Offline programming procedure of multipass weaving welding is shown in Fig.1. The whole process is as follows: three dimension modeling, feature modeling, multi-pass path layout, the producing of path parameters (position, pose and oscillation displacement), selecting of reference path, producing of multi-pass path and finally the producing of robot program. In the path layout step, Users can set the welding parameters of every pass, the number of layers and passes, and welding sequence according to the real welding technology. Finally, the system calculates the motion parameters of every pass including weaving amplitude, transverse deviation, height deviation and pose angle of torch relative to the reference path automatically.

Fig. 1. Path layout procedure of multi-pass welding

2.1 Relationship of Weld Shape and Welding Parameter In order to obtain the weld shape under different welding parameters, the relationship of rate of feed and welding current is measured. The feed speed v1 is expressed with welding current I as follows.

Robot Path Planning in Multi-pass Weaving Welding for Thick Plates

υ1 = ϕ ( I )

353

(1)

When the dimension of weld wire D is constant, the weld cross-sectional area is determined by the feed speed v1 and welding speed v2. The cross-sectional area of single pass S is as follows.

S=

πυ1 D 2 2υ2

(2)

According to the formula (1) and (2), the final filling area S is expressed by formula (3).

S=

π D 2ϕ ( I ) 2υ2

(3)

2.2 Self-defining Multi-pass Weaving Welding Layout Model At present, most researchers adopted equal area or equal height method to layout multi-pass welding path of robot [5-6]. But in many cases, welding condition is not unstable such as welding parameters, layer and pass number, welding sequence of every pass and so on. Therefore users can set the welding parameters of every pass, the number of layers and passes, and welding sequence according to the real welding technology. Firstly, Users only select the reference path in visual simulation graph. The system can automatically produce the other paths according to the reference path. Fig. 2 shows the reference plane and reference line. The reference line is the angle bisector of both sides. Then multi-pass paths are produced through setting interpolation points, duplicating, translating and pose adjusting. So adjustment parameters of every pass such as Z direction position deviation ΔZ i j , Y direction position deviation ΔYi j and pose angle deviation

Δθ ij should be obtained.

Fig. 2. Reference path selection

354


2.3 Position and Attitude of Welding Torch Section shape of multi-pass welding seam can be simplified into trapezium A and parallelogram B as shown in Fig. 3 [8]. The dimension of work piece is thickness H, root gap g, groove angle β. If total layers are n, then there are mi passes in layer i. L Q M PL . When the welding current and welding speed of layer i pass j is inputted by user, the area of this pass 6LM is shown in formula (3).

Fig. 3. Section shape of multi-pass welding seam

Δhi of layer i relative to the reference path is expressed by formula (4). The thickness hi of layer According to the geometrical relationship, the height deviation

i is shown by formula (5).

β

Δhi =

i

mi

g 2 + 4tg ( ) ⋅ ∑∑ S k ,l − g 2 k l

β

(4)

2tg ( ) 2 i −1

hi = Δhi − ∑ hk

(5)

k =0

As for trapezium A, the position and pose of torch is easy to calculate. The point F is the midpoint of the line segment MN as shown in Fig. 4a. The pose angle is zone [8]. When mi=1 and j=mi, the welding seam shape is belong to trapezium A or triangle (g=0). The line segment AG can be calculated as follows.


355

(a) trapezium A

(b) parallelogram B Fig. 4. Position and pose of torch

S β AG = hi tg ( ) + i 2 mi hi

(6)

The position and pose deviation relative to reference path is as follows.

Δy =

( mi − 1) Si

(7)

2mi hi

Δz = Δhi

(8)

Δθ = 0

(9)

But for parallelogram B, the position and pose of torch is difficulty to determine. When mi>1 and 0<j<mi, the welding seam shape is belong to parallelogram B. Seen from the fig. 4b, the line segment GB AB BC AC can be calculated according to the geometric relationship. In order to obtain the parallelogram B, intersection point F with perpendicular bisector of line segment AC and line segment AD is the position of the torch of this pass. By ΔAEF ≈ ΔAGC Then the transverse deviation y as follows.

、、、

,

356


β ⎛ ⎞ ⎜ 2 Si ctg ( 2 ) β ⎟ 2 + hi tg ( ) ⎟ + ( hi ) ⎜ 2 ⎟ ⎜ mi hi ⎝ ⎠ y= 2

(10)

β

4 Si ctg ( ) 2 + 2h tg ( β ) i mi hi 2

The position and pose deviation relative to reference path is as follows. β ⎛ ⎞ ⎜ 2Si ctg ( 2 ) β ⎟ 2 + ( ) h tg ⎜ ⎟ + ( hi ) i 2 m h i i ⎜ ⎟ g ⎠ − −⎝ β 2 4Si ctg ( ) 2 + 2h tg ( β ) i 2 mi hi 2

Δy =

i

mi

k

l

∑∑ S 2Δhi

k ,l

Δz = Δhi Δθ = arctg (

β

hi

(11)

(12)

)

2 Si ctg ( ) 2 + h tg ( β ) i 2 mi hi

(13)

Then the position and pose of the layer i pass j is as follows. j

yi , j = yi ,0 +

∑S k =0

i ,k

(14)

hi

zi , j = z0,0 + Δhi

(15)

2.4 Oscillation Displacement Determination As for trapezium A or triangle (g=0), the weaving width is as follows.

W=

S 1 β hi tg ( ) + i − δ 2 2 2mi hi

(16)


357

Similarly, the weaving width of parallelogram B is expressed as follows. ⎛ ⎜ AC 1⎜ −δ = W= 2 2⎜ ⎜ ⎝

2 ⎞ β ⎛ ⎞ ⎟ ⎜ 2Si ctg ( 2 ) β ⎟ 2 ⎟ + tg ( )hi ⎟ + ( hi ) − δ ⎜ ⎟ 2 ⎟ ⎜ mi hi ⎟ ⎝ ⎠ ⎠

(17)

Where δ is the correction factor of weaving width as shown in Fig. 5. Transformation matrix of position and pose of robot motion is expressed as follows. Sph( Δθ , ΔYij , ΔZ ij ) = Rot ( x, Δθ )Trans (0, ΔYij , ΔZ ij ) 0 0 ⎡1 ⎢0 cos(Δθ ) − sin(Δθ ) =⎢ ⎢0 sin(Δθ ) cos(Δθ ) ⎢ 0 0 ⎣0

0 ⎤ ⎡1 0 ⎥ ⎢0 ⎥⎢ 0 ⎥ ⎢0 ⎥⎢ 1 ⎦ ⎣0

(a) trapezium A

0 0 0 ⎤ 1 0 ΔYij ⎥ ⎥ 0 1 ΔZ ij ⎥ ⎥ 0 0 1 ⎦

(18)

(b) parallelogram B

Fig. 5. Weaving width of welding torch

3 Multi-pass Welding Experiments Welding condition is V-groove, thickness 25mm, groove angle β=60°, root gap 3mm, welding sequence, 5 layers, 8 passes in total as shown Fig. 6. Welding current and speed of every layer is different as shown in Table1. Welding method is Gas Metal Arc Welding (GMAW) with Ar (98%)+O2 (2%) gas. Welding wire dimension is 1mm. Welding robot is KUKA KR16 robot. Fig. 6 shows that the arrows are position and pose of every pass torch such as P11, P21,…,P52. The correction factor δ is equal to 2mm. The transverse and height position deviation, pose angle deviation and weaving width are shown in Table1. Macro metallographic sections of multi-pass weld by robot are shown in Fig. 7. Weld shape and sidewall fusion are well, which illustrate the path layout model is correct and meet the actual requirement of multi-pass welding production for large thick plate.

358


Fig. 6. Welding Sequence and Layout results Table 1. Path layout parameters

Fig. 7. Multi-pass welding metallographic sections

4 Conclusions i: An off-line programming system for multi-pass weaving welding of thick plate arc welding robot is developed. ii: A user self-define path layout model is established for multi-pass weaving welding. Users can set the welding parameters of every pass, the number of layers and passes, and welding sequence according to the real welding technology. The developed system can produce the position, pose of welding torch and oscillation displacement of total passes automatically.


359

iii: The multi-pass welding shape is well and meets the actual production requirement through robot multi-pass weaving welding experiment for single Vgroove thick plate. Acknowledgement. This research is supported by National Postdoctoral Foundation of China (No. 20090460616), by Harbin Creative Talent Science and Technology Foundation of China (No. 2010RFQXG005), by Heilongjiang Province Education Foundation (No. 20100503066) and by Shanghai Sciences & Technology Committee under Grant (No. 09JC1407100).

References [1] Shan, S., Zhou, C.L.: Robot application in multi-pass welding. Manufacturing Automation 1, 12–14 (2005) (in Chinese) [2] Qin, G.L.: Automatic arranged system based on artificial neural networks prediction for submerged arc weld. Master Thesis of Harbin Institute of Technology 7, 7–8 (2000) (in Chinese) [3] Wu, J., Smith, J.S., Lucas, J.: Weld Bead Placement System for Multipass welding. IEEE Proc.Sci. Mesa. Technology 143(2), 85–90 (1996) [4] Masaharu, M., Seigo, H., Megumi, O.: Development of a multi-pass welding program for arc welding robots and its application to heavy electrical section pieces. Transaction of the Japan Welding Society 24(1), 16–21 (1993) [5] Li, K., Dai, S.J., Sun, L.X., et al.: Filling Strategy of Multi-pass Welding. Transactions of the China Welding Institution 22(2), 46–48 (2001) (in Chinese) [6] Sun, L.X.: Weld arrangement and tracking system of multi-pass welding based on structured light. Doctor Thesis of Harbin Institute of Technology, 85– 90 (1999) (in Chinese) [7] Tang, X.H., Drews, P.: 3D-visualized offline-programming and simulation system for industrial robots. Transactions of the China Welding Institution 26(2), 64–68 (2005) (in Chinese) [8] Zhang, H.J., Zhang, G.J., Cai, C.B., et al.: Self-defining path layout strategy by thick plate arc welding robot. Transactions of the China Welding Institution 30(3), 61–64 (2009) (in Chinese)

Influence of the Bar Shape on the Welding Quality of Friction Hydro Pillar Processing Xiangdong Jiao, Hui Gao, Hongwei Zhan, Canfeng Zhou, Jiaqing Chen, and Yanqing Zhang Beijing Institute of Petrochemical Technology, China e-mail: {jiaoxiangdong,gaohuiadam}@bipt.edu.cn

Abstract. Traditional underwater arc welding methods, for example, TIG and MIG, are generally expensive and complex to implement under water. Besides, the welding quality decreases significantly with the increase of depth of water. So a new-style solid phase welding method called friction stitch welding (FSW) is introduced, since the welding quality of which is almost insensitive to water. This article studies the influence of the bar shape on the welding quality. In the experiments, three typical kinds of shape combinations are adopted. Argon is used as protecting gas, and the flux of which is 30L/min. The feed speed of the spindle is 0.5 mm/s, the rotation speed is 4000r/min. Through the comparison of the macro features and microstructure of the welded joints, the influence of the bar shape on the welding quality is analyzed, and static stress distribution along with the influence on welding temperature by different shapes are investigated. The results indicate that the welding quality obtained by using straight bars with round head is much better than the other shape combinations, the welding defects are reduced and the welded material is much finer.

1 Introduction It can be observed in the past few years that with the continuous exploitation of offshore oil-gas field, the exploitation of marine oil has been transferred from the offshore to the deep-sea, and the repairs of underwater structures has become even more important with the increase of theirs service life. Underwater hyperbaric dry welding is the usual method for underwater welding repairs, for it is endowed with a lot of merits such as the excellent welded joints, which can compete the joints acquired on ground [1]. However, the hyperbaric chamber not only costs a lot, but the stability of welding arc decreases a lot with the increase of water depths [2], so it’s restricted to be widely used for deepwater welding repair. As a potent complementary method to the underwater hyperbaric dry welding, friction stitch welding(FSW) was invented by the welding institute(TWI) at 1992, which is a late model solid phase welding method and is based on offshore platform as well as seabed pipeline maintaining [3]. Compared with the arc welding at several occasions, the FSW welded joints have many outstanding advantages, such as the exceptional joints, high-efficiency, low material consumption and non-pollution etc. FSW has proved to be one of the most promising welding methods, especially at T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 361–367. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

362

X. Jiao et al.

the deepwater welding field. Above all, the friction welding is not sensitive to water, and the welding parameters change a little under the situation of different water depths [4], which is beyond the reach of comparison to the arc welding. Therefore, FSW has become a popular research field on the underwater repairs for steel structure, and this technology has wined a lot of industrial application in many developed countries. Over the past ten years, some countries such as England, Germany, Japan, America and Brazil, etc has focused on the research work of the friction stitch welding process successively [5]. In China, this new-style friction welding technology typified by friction stir welding was just launched at the beginning of the new century. Moreover, the research work of both the equipment and the welding process of friction stitch welding is nearly in a state of blankness. Therefore, the research on the friction hydro pillar processing (FHPP) welding process of FSW is pressing and of extreme importance to the development of FSW technology in the country. It’s expected to provide some technical storage for the repairs of seabed pipeline, offshore platform and some other underwater projects, such as ocean transportation, port, warship, national defense. The purpose of this investigation is to get the effect of the shape of the bar and hole combinations on the welding quality. The welded joints of the same welding parameters but different shapes combination are obtained by a series of experiments. Besides, the macro features and micro structures of the welded joints, the static stress distribution by different shape combinations along with their effects on the welding temperature are analyzed and discussed, which can lay a foundation for the further investigation.

2 Experimental Procedures and Method The base metal used for experiment is 2024-T4 aluminum alloy, which has been subjected the course of solution heat treatment and natural aging. In order to investigate the effect of single influence factor, namely shapes of friction interface, on the welding quality, thus three kinds of shape combinations are designed before the FHPP experiments, which are flat-bottomed, taper-bottomed and roundbottomed shape, and the schematic diagram is shown in fig. 1. The diameter of all the bar head and plate hole is 14mm and 16mm each, and the hole-depth is 20mm. Meanwhile, argon is adopted to prevent the oxidization of aluminum alloy, the flow of which is 30L/min. Moreover, the welding speed and rotation speed are set as 30mm/min and 4000rpm respectively. Welding spindle is the most important component of the equipment of the experimental system, the rated axial working pressure and torque of which are 20kN and 50N·m respectively, and the maximum rotation speed of hydraulic motor is 5000r/min. PLC is set as the main control unit of the electrical control system, the human-computer interface of which is in the hands of touch screen. In order to collect and record the process parameters synchronously during the welding process, thus a set of special data acquisition system directed at FSW welding process experiments is devised. The whole data acquisition system is composed of

Influence of the Bar Shape on the Welding Quality

363

temperature sensor, displacement transducer, pressure sensor, and laser rotation speed sensor. During the welding, all the experimental data measured by the sensors are gathered by S7-200PLC, then the data are subjected to a string of processing via signal conversion module of a data collector (WAVEBOOK/516E), and the processing results can be transmitted to the computer through Ethernet, shown in fig. 2. Therefore, this data acquisition system realizes the gathering and realtime surveying of the whole experimental data.

Fig. 1. Schematic diagram of FHPP

Fig. 2. Constitution of the whole data acquisition system

3 Results and Analyses 3.1 General Features of the Welded Joints In the FHPP experiments, three kinds of shape combinations are applied, and the typically welding joints are shown in Fig. 3. It can be seen from the figures that the material compactness at filled welding region through flat-bottomed combination is apparently lower than that through the other two combinations, and the welded joint quality shown in Fig. 3c is particularly well. A substantial number of

364

X. Jiao et al.

FHPP comparison experiments indicate that the welded joints quality through round-bottomed combinations is almost satisfactory with the same welding parameters: namely, equal flux of protecting gas, equal rotation speed and feed speed of welding spindle. The round-bottomed combination also brings the highest probability of welding success. It is probably because that the alloy material at the welding line can achieve a state of super plasticity more easily with the improving of shape combinations, and the flowability of welded material can be enhanced by the action of welding pressure, which can also make the padding of welded material more compact and sufficient, and the welding defects decreased subsequently.

(a)

(b) (c) (a) Flat-bottomed combination (b) Taper-bottomed combination (c) Round-bottomed combination

Fig. 3. Microstructure of the joints through three shape combinations

3.2 Microstructure Characterization The welding quality acquired from FHPP experiments presents a trend that it raises with the increase of contact area of the frictional interface, shown in fig. 4. It’s properly due to the reason that frictional resistance will increase when the bar and plate begin to rub with the increase of contact area, which can lead to an increase of friction period and generate more heating power. With regard to the FHPP welding, the bottom corner of joint surface is called as “cold area”, in which defects often occur with high probability. Because the initial contact interface of the filled area is the threshold of thermal source among the whole welding process, the welding temperature at the beginning is relatively low. In addition, the welding pressure is inadequate so that the superplastic material can not fill the weld compactly, and the extent of superplasticity as well as the flowability of welding material is insufficient. In order to decrease or eliminate welding defects of the “cold area”, the following two methods can be resorted to, one is to preheat the bar and plate, which can elevate the temperature of the “cold area” to a great extent; the other is to increase the welding pressure or feed speed in a proper range, which can facilitate the diminishing the welded defects and also make the joining between filled material and the parent metal more compactly.


365

Fig. 4. Microstructure of the weld zones

3.3 Static Stress Distribution of Different Shapes Finite element analysis is also conducted in this paper to investigate the stress situation for different shapes of the bar, shown in fig. 5. The results indicate that the stress distribution of the flat-bottomed and taper-bottomed bar is almost equivalent under the same axial pressure; however, the stress distribution at the transitional region of the round-bottomed bar is much higher than that of the others. From the above saying, a conclusion could be drew that the higher stress distribution will contribute to the deformation and moving of the super-plastic material on the condition of equivalent welding temperature, which is crucial especially in the initial phase of welding process. Furthermore, the round-bottomed interface can facilitate the up-climbing and transferring of the welded material, bringing with a significant decrease of welding defects that often occur at the beginning of the welding.

Fig. 5. Static stress distribution by different shape combinations

366

X. Jiao et al.

3.4 Influence of Different Shapes on the Welding Temperature What’s shown in fig. 6a is a simplified model for the application of thermocouples in the FHPP experiments. What’s shown in fig. 6b reflects the temperature history measured by thermocouples through flat-bottomed and round-bottomed combinations. P1 p2 p3 p4 of fig. 6b stand for the flat-bottomed specimen, and q1 q2 q3 q4 stand for the round-bottomed specimen. Seen from the temperature curves in fig. 6b, welding temperature of the corresponding measuring points by round-bottomed combinations is higher than that by the flat-bottomed combinations on the whole. This is because that the round-bottomed interface can bring the more heat input, especially at the beginning of welding. At the same time, the cooling rate of weld material will mount up and the heat affected zone (HAZ) can be diminish a lot, resulting in a quick solidification of welded material and a proper increase of the hardness [6-8], thereby the mechanical properties of welded joints will be improved from another point of view [9]. Aside from what mentioned above, the temperature history curve in fig. 6b also suggests that the welding heat source is moving up with the incessant infilling of welding material, and the subjacent material begin to cool and solidify at the same time, which just coincides with the actual FHPP welding process.

、、、、、、

(a) Temperature measuring model

(b) Temperature curves Fig. 6. Temperature history curve of each measuring point

4 Conclusions (1) The welded joints feature of round-bottomed combinations is the finest under the conditions of equal welding parameters, and it brings with the highest


367

probability of welding success. The welded joints feature of round-bottomed combinations is proved to be a perfect shape combination. (2) Stress distribution at the transitional area of the round-bottomed bar is clearly higher than the others, which can increase the warming up time at the beginning of welding and also facilitate the up-climbing and transferring of the welded material. (3) Initial contact surface of the filled area is the foremost thermal source during the welding, and the welding temperature over there is relatively low. The temperature measure indicates that more heat input can be produced by the roundbottomed combination, which can bring higher cooling rate and make better comprehensive properties of welded joints finally. Acknowledgement. This project is supported by National Hi-tech Research and Development Program of China (863 Program, Grant No. 2006AA09Z329), National Natural Science Foundation of China (Grant No. 40776054), Funding Project for Academic Human Resources Development in Institutions of Higher Learning Under the Jurisdiction of Beijing Municipality (Grant No. PHR201007134 and PHR20090519).

References [1] Chen, J.Q., Jiao, X.D., Zhou, C.F.: Friction stitch welding—a new kind of material forming technology. Transactions of the China Welding Institution 28(9), 118–112 (2007) (in Chinese) [2] Jiao, X.D., Zhou, C.F., Chen, J.Q.: Challenges and counter measures of offshore engineering joining technology in 21st century. Electric Welding Machine 37(6), 75–80 (2007) (in Chinese) [3] Meyer, A.: Friction hydro pillar processing bonding mechanism and properties. GKSSForschungszentrum Geesthacht GmbH, Geesthacht (2003) [4] Ellis, C.R.G.: Continuous Drive Friction Welding of Mild Steel. Welding Journal, 183s–197s (April 1972) [5] Faes, K., Dhooge, A., De Baets, P.: Parameter optimisation for automatic pipeline girth welding using a new friction welding method. Materials and Design 30, 581–589 (2009) [6] Xu, L.H., Tiah, Z.L., Zhang, X.M., et al.: Metal Inert Gas Welding of 2519-T87 High Strength Aluminum Alloy. Chinese Journal of Mechanical Engineering 20(4), 32–35 (2007) [7] Wu, Z.S., Liu, C.R., Wang, J.H.: Microstructure and properties of deep crogenic treatment electrodes for spot welding hot dip galvanized steel. Chinese Journal of Mechanical Engineering 18(1), 17–20 (2005) [8] Wang, W.X., Huo, L.X., Zhang, Y.F., et al.: Solid-State Phase Transformation of Welded Metal in Control of Welding Distortion Process. Chinese Journal of Mechanical Engineering 18(1), 64–67 (2005) [9] Wang, X.J., Niu, Y., Zhang, J.: Study on Fatigue Properties of Aluminum Alloy 2024 Joint by Friction Stir Welding. Casting Forging Welding 37(15), 4–6 (2008) (in Chinese)

Interference Analysis of Infrared Temperature Measurement in Hybrid Welding Jifeng Wang, Houde Yu, Yaozhou Qian, and Rongzun Yang Shanghai Institute of Special Equipment Inspection & Technical Research, Shanghai 200062 e-mail: [email protected]

Abstact. Welding is an essential thermal processing and a thermal conducting, so infrared sensing is a natural choice for weld process monitoring to measure and control temperature. However, the infrared temperature is interfered because of radiation reflection during applying IR sensing to the welding. The objective of this research is to measure the top-face temperature distribution in hybrid laserTIG welding process. The results of experiments show that this high light zone interference which caused by special direction specular reflection of arc light, ceramic nozzle, electrode and laser nozzle are transferred out of welding seam when the IR thermography is placed perpendicular to the welding seam. The interference in welds surface that caused by specular reflection in other places was little and was further reduced through the decreasing viewing angle. Finally, the IR thermography was calibrated by using thermocouple and IR images were obtained. The result shows that temperature data captured by the IR termography is reliable using this method in hybrid laser-TIG welding of AZ31B.

1 Introduction Welding temperature field is important because of subsequent reasons. Firstly, the microstructure and the mechanical properties strongly depend on the cooling rate, and the temperature field can supply this information. Secondly, the welding temperature field is affected by welding parameters, so it can lead us to understanding and control welding process further. Finally, welding temperature field is also used in a number of simulation models, to improve the correction of simulation and prediction. A lot of research works have done to develop temperature measuring in welding processes. At present, temperature measuring methods are divided into two kinds: contact and noncontact. The advantage of infrared thermography is that the sensor is not necessary to directly contact with the products when welding and that this kind method has very quickly response. However, the major difficulty of IR measuring is the interference from interference source such as: arc light, hot tungsten electrode and so on. The interference is more serious for materials such as aluminium alloy and magnesium alloy because of high reflectivity. Most of the wavelength of arc radiation is between 0.34 and 1.8μm. The weld radiation is greater than the arc energy above 2μm of wavelength [1]. A. Bichnell T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 369–374. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

370

J. Wang et al.

[2]gained progressive improvement in image quality through high-pass and band pass filters, in fact within his spectrum pass band, arc radiation still could not be ignored. D.Farson [3] used filtering and shielding cup to minish arc and electrode interference from the infrared thermal radiation in TIG welding of AISI 1250 sheet. But the shielding cup soon was heated and changed into new interference resource. P.W. Ramsey [4] measured radiation interference from arc in different viewing angle and location of radiometer, however, this method can not avoided the interference from high light zone. In addition, W.E. Lukens [5] could control weld metal cooling rate ignoring the interference from arc and electrode. Many studies [6-10] performed at auburn University, concerning the viability of point infrared detectors for GTA welding process control. Hybrid laser-TIG welding combined laser welding with high speed and productivity and the traditional TIG welding. Many of research results are continuously increasing recently [11], but there is not any report which involved with measuring temperature field using IR sensor.

2 Experimental Setup The workpiece used in this study is AZ31B magnesium alloy plate with 300×100×2.5mm. The surface of the plates were chemically cleaned using acetone and scratched by steel brush before welding to maintain consistent surface condition. The welding conditions were showed in Table 1, setup was shown in Fig.1. The infrared thermography in this experiment is sensitive in the spectral range of 8 to 12 μ m. The emissivity is various in different directions β angle. Maximal value of emissivity of nonmetal can be obtained in its normal direction, the viewing angle β was fixed as 45 degree; and the optical axis of infrared thermography is perpendicular to weld seam. Table 1. Welding Conditions

Welding Speed Shielding Gas Welding Current Electrode Pulse Laser

400mm/min Argon, 5L/min AC-70A 3.2mm Average 300w

Fig. 1. Hybrid laser-TIG welding process

Interference Analysis of Infrared Temperature Measurement in Hybrid Welding

371

3 Interference Analysis 3.1 Arc Llight Interference In this experiment, TIG torch located above A plate and 9mm distant to A verge. B plate is placed parallelly to A with 1mm clearance, as shown in Fig. 3a. The infrared thermography catch IR image after arc has ignited for 3 seconds when arc has reached stabilization. L01 line scaned arc temperature profile, L02 line scaned arc inverted temperature profile. The temperature in L01 is higher than L02 (Fig. 3b). Arc dimension approachs cermet nozzle diameter, radiation interference I area is little in n direction mostly casused by arc diffuse reflection; however, raG diation interference area is long in m direction due to arc specular reflection. G Thus, the interference zone is larger in m direction (point to the IR position), G I while it is small in n direction (perpendicular to m , namely welding seam) and the interference intensity is little, either.

Fig. 2a. Arc IR map

Fig. 2b. Scan-line

3.2 TIG Ceramic Nozzle Interference Experiment condition is same to section 3.1, but worktable moved. IR images were captured during welding current abruptly decreased in order to minish arc radiation interference. These images are respectively selected in halfway (Fig. 3a). Because plate B is not heated by the arc, its surface interference is almost from

Fig. 3a. IR map in half of welding process

Fig. 3b. Scan-line

372

J. Wang et al.

ceramic nozzle radiation, while from arc radiation is small (position scan line L01 and L02). Fig. 3b showed that the interferential temperature of ceramic nozzle inverted image which caused by specular reflection increased ceaselessly because continuously heated, but the interference outside of inverted image is little. Fortunately, welding seam will be not interfered by this inverted image if IR thermography is placed like Fig. 1.

3.3 Laser Nozzle Iinterference To analyse the radiation interference from laer nozzle, the IR thermography is placed behind the weld pool. Fig. 4 showed one of IR image of laser-TIG hybrid welding. It can be found that there are two interference zones in temperature field. One was the complex interference zone including tungsten, TIG ceramic nozzle and arc light. The other was laser nozzle interference zone (laser nozzle inverted image caused by specular reflection). This interference can be wiped off (Fig. 5b) if the IR thermography is placed like Fig.1.

Fig. 4. IR map of laser nozzle interference

3.4 The Method of Removing Interferences How to remove the interferences always puzzled the researchers in this field. In traditional experiment, the temperature image was captured paralleling the direction of welding seam, and IR device was placed behind welding pool as shown in Fig. 5a, and the temperature field measured was interfered badly by something

Fig. 5a. IR behind welding pool

Fig. 5b. IR perpendicularly welding seam

Interference Analysis of Infrared Temperature Measurement in Hybrid Welding

373

mentioned above. In this paper, the IR thermography is placed perpendicularly to welding seam (Fig. 1). Those interferences that caused by specular reflection was transferred out of welding seam as shown in Fig5b. So when measuring top-weld temperature filed, the errors cased by these interference facts became small.

4 Results 4.1 Calibration On the base of the anterior results, the validation system of IR measuring was designed as shown in Fig. 6. The location of the sensing point was showed in the Fig. 6. The thermocouple is attached at sensing point, its signal is transmited on the screen where temperature is displayed; At the same time, IR thermography also catch the sensing point through itself point measurement model, and this temperature value is also displayed on the screen in real-time. These two temperature values are accord, when emissivity is 0.18. Further study showed that IR temperature distribution data was reliable in the zone of x>10 mm and y>6 mm.

Fig. 6. Principle of the IR temperature measuring system

4.2 Temperature Measurement in Hybrid Welding Bead-on-plate hybrid laser-TIG welding of AZ31B magnesium alloys was investigated. IR images(Fig. 7) were obtained during welding process. Scan line showed that the temperature appeared peak in welding centerline and flat step in both sides of centerline.

Fig. 7a. IR map in hybrid welding

Fig. 7b. Scan-line in hybrid welding

374

J. Wang et al.

5 Conclusions Based on the experiments described above, it was concluded that the radiation from arc light, ceramic nozzle, electrode and laser nozzle significantly interfered measurement temperature at a location on the base metal. The high light zone interference which caused by special direction specular reflection is primary. The IR thermography is placed perpendicularly to the welding seam. These high light interference that caused by specular reflection of arc light, ceramic nozzle, electrode and laser nozzle were transferred out of welding seam. The interference that caused by specular reflection in other places was little and was further reduced through the decreasing viewing angle. The experiments verified that this method practically reduces interference from the infrared thermal measurements. Finally, an experimental calibration of the infrared measurements based on thermocouple measurements was performed at a location of 10 mm behind and 6mm to the side of the weld seam.

References [1] American Welding Society: Welding handbook, 7th edn., Miami (1976) [2] Bicknell, A.: Infrared sensor for top face monitoring of weld pools. Meas. Sci. Technol. 5, 371–378 (1994) [3] Farson, D.: Infrared measurement of base metal temperature in gas tungsten arcwelding. Welding Research Supplement, 396–401 (September 1998) [4] Ramsey, P.W.: Infrared temperature sensing systems for automatic fusion welding. Welding Research Supplement, 89–96 (April 2000) [5] Lukens, W.E.: Infrared temperature sensing of cooling rates for arc welding control. Welding Journal, 27–33 (January 1982) [6] Chin, B.A.: Infrared thermography for sensing the arc welding process. Welding Research Supplement, 227–234 (September 1983) [7] Fan, H.: Low-cost infrared sensing system for monitoring the welding process in the presence of plate inclination angle. J. Materials processing Tech. 140, 668–675 (2003) [8] Nagarajan, S.: Infrared sensing of adaptive arc welding. Welding Research Supplement, 462–466 (November 1989) [9] Sundaram: Control of the welding process using infrared sensors. IEEE Transactions on Robotics 8, 86–93 (1992) [10] Banerjee, P.: Infrared sensing for on-line weld geometry monitoring and control. J. Engineering for Industy, 223–230 (August 1995) [11] Seyffartn, P.: Laser-Arc processes and their Applications in Welding and Material Treatment. Taylor & Francis, Taylor (2002)

Preliminary Investigation on Embedding FBG Fibre within AA6061 Matrices by Ultrasonic Welding Zhengqiang Zhu, Yifu Zhang, Chun Zeng, and Zhilin Xiong School of mechatronics engineering, Nanchang University, Nanchang, 330031, China e-mail: [email protected]

Abstract. Ultrasonic welding can be used to join plastic and metal through highfrequency (more than 20 kHz) acoustic vibrations. As the process is low temperature and low pressure, it is easy to embed fragile sensors such as FBG fibre within metal matrices. In this research, the optimal parameters are obtained by experiments of tensile and nanoindentation hardness firstly. Then the microstructures are investigated in this paper to check the welding effect. The sensor effect of FBG fibre is also investigated. These experimental results clearly demonstrate that the strength and hardness of metal matrices will increase under ultrasonic welding. In addition, the temperature sensitivity of the nickel-coat FBG was improved after welding. From the experimental results, it is completely feasible to embed FBG fibre within AA6061 matrices by ultrasonic welding. Therefore, the 3-D smart metal structure can be made by the ultrasonic metal welding.

1 Introduction Ultrasonic metal welding (USMW) is used to joined homogenous and dissimilar metal foils by applying pressure and ultrasonic vibration [1-3]. Now, USMW is widely used in microelectronics and other industrial areas [4-5]. Ultrasonic welder includes transducer, sonotrode, pneumatic pump and booster. The welder’s diagram is shown in Fig.1. The high frequency relative motion between the parts forms a solid-state weld through progressive shearing and plastic deformation. During the process, the oxides and contaminants are dispersed and a pure metal contact between the adjacent surfaces are performed.

Fig. 1. The diagram of USMW T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 375–381. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

376

Z. Zhu et al.

The main advantages of USMW include absence of liquid-solid transformations, low energy consumption, low pressure, low temperature which allows embedding of electronics, such as sensors and actuators. FBG sensors are widely used for measuring strain and temperature with the advantages including considerably improved accuracy, sensitivity, and immunity to electromagnetic interference, lightweight, etc. So FBG can be used to create smart materials that can operate in harsh environments—such as underwater—where conventional sensors cannot work. In this research, FBG fibre is embedded in AA6061 matrices under ultrasonic welding. Firstly, by experiments, the optimal parameters of USMW are obtained. Then the microstructures are investigated in this paper to check the welding effect. The sensor effect of FBG fibre is also investigated.

2 Process Conditions and Experiments 2.1 Test Materials The material studied is aluminium alloy 6061 in the form of foil with 0.4 mm thickness. Table 1 lists the composition and mechanical properties of this material. The surface roughness of the 6061 foil is about 0.4 µm (Ra). AA6061 has good weldability and formability and has long been used in aerospace, structural, transport and construction applications. FBG fibre used in this experiment has a cladding diameter of 125 μm and the central wavelength is 1540 nm at 30 .

℃

Table 1. Composition and mechanical properties of aluminium alloy 6061 foils Composition Al-1.0Mg-0.6Si-0.7Fe-0.3Cu0.2Cr-0.15Mn-0.25Zn-0.15Ti

Tensile strength (MPa) 113-117

Yield strength (MPa) 45-50

Elongation at break (%) 10-10.5

2.2 Test Procedure The metal ultrasonic welding machine is shown in fig. 2,rated at 3.2 kW and frequency f=20 kHz, with a 125-mm beam tool-steel sonotrode, ending at 15×15mm-square tip with sand grinding surface. Firstly, the optimal welding parameters are obtained by experiments. In these experiments, the amplitude and frequency is fixed, the welding time is changing. Then tear tests and nanoindentation hardness tests are performed to obtain the tearing force data and the nanoindentation hardness data to confirm the optimal welding parameters. Then, under this parameter, bare FBG fibre is embedded

Preliminary Investigation on Embedding FBG Fibre within AA6061 Matrices

377

within AA6061 matrices. Then the micro-photos of both the monolithic welding sample and FBG fibre embedded sample are obtained. Also the temperature sensitivity of the nickel-coat FBG fibre is measured after embedded within metal.

Fig. 2. The metal ultrasonic welding machine

3 Results and Discussion The fixed amplitude is about 30 μm, welding current is 12 A, and the pressure is also certain which is effected by pneumatic cylinders. When welding time is 230 ms, the welding effect is the optimal with the tearing force of 128.5 N. The longer of the welding time, the higher of the temperature, and the better of the welding effect under the fixed pressure. Fig. 3 shows the welding effect of the sample with 230 ms welding time. With the welding time increases, the tearing force tends to increase linearly firstly then basically unchanged. However, with the increase of welding time, the temperature is rising and the specimen edges will occur melting, which will affect welding effect. Therefore, welding time cannot increase unlimitedly. Fig. 4 is the diagram between welding time and the tearing force (tensile speed=15mm/min). Each experiment, the welding current is constant but the voltage will be different. Fig. 5 is the relationship between welding time and the nanoindentation hardness (maximum load=10 gf, dwell time=15 s) of around welding interface. Fig. 5 shows the increasing tendency of the hardness in the vicinity of the weld interface. The results show that after ultrasonic welding the hardness of all matrices increases, compared with that of the original aluminum foils, especially at the region around the weld interface. It is also found that there is no obvious influence with respect welding time. Because plastic deformation causes dislocation movement, and dislocations stress fields interact with each other, the existence of one dislocation will hinder the movement of the others. When a material is plastically deformed, dislocations become denser and less mobile and, as a result, the material will present higher strength and hardness.

378

Z. Zhu et al. 140 120

) N (

100

ec r o 80 f g ni r 60 a e t e 40 h T 20

0 50

100

150

200

250

Welding time (ms)

Fig. 3. The sample with 230 ms welding time Fig. 4. The relationship between welding time and the tearing force 60 Welding time 110ms Al

Al

Welding time 260ms

55

Welding time 380ms

Hardness(HV)

Virgin 6061Al

50

45 41.2

40

35 -0.25

-0.20

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0.25

Distance from welding interface d(mm)

Fig. 5. The relationship between welding time and the hardness of around welding interface

Fig. 6 is the microstructure photos of the samples with welding time 200 ms and 230 ms. From (a), there exists a clearly welding interface. While from (b) there is no clear welding line. Both foils are exactly boned, which proves once again that the welding effect of time 230 ms is optimal.

(a) 200 ms

(b) 230 ms

Fig. 6. Microscopic photos of the samples with different welding time


379

Surface treatment for the FBG fibre is necessary before embedded within AA6061 matrices. Firstly, the layer of plastic jacket must be removed. The part of fibre needs to soak in the acetone for about 10 minutes. Then take it out and use the cotton to dry it. And then the plastic skin is stripped by needle nose pliers. After that, the bare part is placed in the middle of the two experimental foils. Fig. 7 is the microscopic photos of the samples embedded FBG fibre with different welding time. Fig. 7(a) shows that when the welding time is 180 ms, a small part of the both foils are welded. Around the fibre, there is much gap without metal. While when welding time is 230 ms, Fig. 7(b) and Fig. 8 shows that both foils are full bonded. And there is no gap around the FBG fibre. The fibre was not deformed during the process, retaining its circular external geometry. The center of the fibre was found to be located at the interface between top and bottom AA6061 foils. Metal foils in the vicinity of the fibre were found to bond well, in particular. The cavity created by the placement of a FBG fibre between two foils was fully filled by the matrix metal, indicating significant plastic deformation of the matrix material during ultrasonic welding processing.

(a) Welding time 180 ms

(b) Welding time 230 ms

Fig. 7. Microscopic photos of the samples embedded fibre with different welding time

Fig. 8. SEM image of fibre embedded between Al foils with welding time 230 ms

For prepare to embed the FBG in metal by ultrasonic welding successfully. First the chemical plating and electroplating process for nickel coating of FBG. The diameter of the nickel-clad FBG is 140.3 μm. Fig. 9 is the thermal responses after chemical plating and electroplating experiment. Fig. 9 shows that the best

380

Z. Zhu et al.

℃

linear fit curves give an average temperature sensitivity of 15 pm/ . Fig. 10 shows that when the welding time is 100 ms, the nickel-coat FBG embedded in 6061 Al matrices by ultrasonic welding. Temperature sensitivity was measured, and the result shows that the best linear fit curves given an average temperature sensitivity of 17.9 pm/ . After welding, it is clear that this enhanced apparent temperature sensitivity due to the combination effect on thermal expansion of aluminium, nickel and silica glass material. The thermal expansion coefficients of aluminium, nickel and silica glass are 23.4e-6/ , 14.2e-6/ and 0.55e-6/ . Enhanced temperature sensitivity could be an advantage for decoupling temperature cross-sensitivity, and offering better measurement accuracy.

℃

℃

1541.6

℃

p1(cooling down) p2(Heating up) Linear Fit of p1 Linear Fit of p2

1541.4

Bragg wavelength/mm

℃

1541.2 1541.0 1540.8

y = 0.0150x + 1540.1 R^2 = 0.9867

1540.6 1540.4

y = 0.0158x + 1540.0 R^2 = 0.9898

1540.2 1540.0 1539.8

0

10

20

30

40

50

60

70

80

90

Temperature/ć

Fig. 9. Thermal responses of the FBG with nickel coat p1(cooling down) p2(heating up) Linear Fit of p1 Linear Fit of p2

1542.0 1541.8

Bragg wavelength/mm

1541.6 1541.4 1541.2 1541.0

y = 0.0174x + 1540.2 R^2 = 0.9980

1540.8 1540.6 1540.4

y = 0.0183x + 1540.1 R^2 = 0.9998

1540.2 1540.0 1539.8

0

10 20 30 40 50 60 70 80 90 100

Temperature/ć

Fig. 10. Thermal responses of the nickel-coat FBG embedded in aluminium matrices

4 Conclusions During USMW, the optimal welding parameters are found. When the welding time is 230ms with exact frequency of 20 kHz, and the amplitude 30 μm, the welding effect is perfect. Use these parameters to embed the FBG fibre within the


381

metal AA6061 matrices. It is found after welding the fibre is intact. That means we can embed FBG fibre within AA6061 matrices. In addition, the nickel-coat FBG embedded in 6061 Al matrices under ultrasonic welding, the temperature sensitivity was improved obviously. Therefore, the 3-D smart metal structure can be made by USMW which is the authors’ main research direction for future. Acknowledgement. The authors wish to thank the financial support for this research from the National Natural Science Foundation of China (Grant No. 50865007); the Raise Object Project Grant of Jiangxi Province for Young Scientists (Jinggang Star); the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry; 2010 Annual Key Science and Technology projects of Jiangxi Provincial Department of Education (GJJ10013).

References [1] Jones, J.B., Meyer, F.R.: Ultrasonic welding of structural aluminium alloys. Welding Journal 37(3), 81–92 (1958) [2] White, D.R.: Ultrasonic consolidation of aluminium tooling. Advanced Materials Processing 161(1), 64–65 (2003) [3] O’Brien, R.L.: Welding Handbook, Welding Processes, 8th edn., vol. 2, pp. 783–812. American Welding Society, Miami (1991) [4] Harman, G.G., Albers, J.: Ultrasonic welding mechanism as applied to aluminium-wire and gold-wire bongding in microelectronics. IEEE Trans. Parts, Hybrids Packaging 13(4), 406–412 (1997) [5] Aro, H., Kallioniemi, H., Aho, A.J.: Ultrasonic welding of experimental osteotomies. Acta Orthop. Scand 51(4), 703 (1980)

Thermal Process Analysis in Welding Prototyping of Metal Structures Jian-ning Xu, Hua Zhang, Ronghua Hu, and Yulong Li School of mechatronics engineering, Nanchang University, Jiangxi province, 330031, P.R. China e-mail: [email protected]

Abstract. Using welding prototyping method to process metal parts can organize compact tissue and controllable dimensions at a lower cost. Welding is a transient heating and rapid cooling process. The changes of welding temperature field have an important impact on the quality of metal parts. In this paper, ANSYS is used to analyze thermal process of welding prototyping, to obtain welding temperature field and residual stress changes of metal structure in welding prototyping process. It is also used to analyze the impact of welding temperature field changes on the metal structure.

1 Introduction The processed metal parts based on TIG welding rapid prototyping method are composed of whole weld. Its high density is able to meet the strength and performance requirements of metal parts. Using TIG arc as a heat source and using ordinary wires as welding materials, greatly reduces the costs of welding rapid prototyping technology [1-3]. In the welding process, welding temperature field has a significant effect on the microstructure and mechanical properties of the welded metal parts, using numerical simulation and experimental method to research the thermal process of welding metal parts forming, can effectively reduce the trials number, and provide theoretical basis for the control of welding prototyping process. In this paper, use ANSYS finite element software as a tool, to prepare APDL procedures, study welding prototyping thermal process of metal structure and achieve effective prediction for temperature field of welding prototyping process.

2 Mathematical Model Welding is a transient heating and rapid cooling process. In this paper, first carry on three-dimensional transient thermal analysis, gain temperature change in welding process; and then use the thermal analysis results as the payload to carry out three-dimensional transient stress analysis. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 383–390. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

384

J.-n. Xu et al.

2.1 Thermal Analysis Because welding rapid prototyping is a heating process, heat conductions play a key role in the whole process. When the heat source moving, the temperature of the entire weldment changes rapidly along with time and space, the thermal physical properties are also dramatic changing with temperature, while there still exist in the latent heat phenomenon of melting and phase transition [4, 5]. Therefore, welding temperature field analysis belongs to a typical non-linear transient heat conduction problem. The control equation of non-linear transient heat conduction problem is [6]: ρc

∂T ∂ ⎛ ∂T ⎞ ∂ ⎛ ∂T ⎞ ∂ ⎛ ∂T ⎞ ⎟ + ⎜λ = ⎜λ ⎟ + ⎜λ ⎟+Q ∂t ∂x ⎝ ∂x ⎠ ∂y ⎜⎝ ∂y ⎟⎠ ∂z ⎝ ∂z ⎠

(1)

、

ρ c and λ for respectively material density, specific heat and thermal conductivity, which are temperature-related functions; T for the temperature field distribution function; t for heat transfer time; Q as the internal heat source intensity. Encountered heat flow and heat transfer boundary conditions in welding rapid manufacturing. Thinking about the heat source model selecting directly impacts on the calculation accuracy of transient welding temperature field, when TIG welding, commonly use Gaussian heat source model, the heat flux distribution as follows:

q(r ) = q m exp(−3

r2 ) r2

(2)

Where, r: the distance from the heat source center; r :arc effective heating radius; q m :Maximum heat flux ratio. In the welding rapid prototyping, can use the above-mentioned heat sources and continuously move along the welding direction. 2.2 Stress Analysis Using the junction temperature obtained from the thermal analysis as the body loads and exertingits on the structural stress analysis, the internal stress and strain field distribution of weldment change along with the welding heat source movement and is a transient problem, the analysis unit will also experience changes of elastic and plastic state. With the dramatic changes in temperature, constitutive relationship between stress and strain shows up non-linear, generally believe that welding distortion belongs to small deformation field, so need to adopt an incremental thermal elastic-plastic finite element theory to analyze. The established finite element formulation as follows [7, 8]:

[K ]{du}i = {dF }i

(3)

Thermal Process Analysis in Welding Prototyping of Metal Structures

385

Where, [K]:elastic-plastic stiffness matrix; {du}i :The displacement increments

obtained from the ith load; {dF }i :The ith load increment vector.

3 The Physical Model and Mesh Division The main feature of three-dimensional model is to use mobile heat source, cell life and death technology, the temperature-dependent material properties, different thermal coefficients of the different surface (such as near the pool area and away from the pool area, etc.). Apply ANSYS programming language (APDL) to establish mobile heat source and the corresponding boundary conditions. The cell types of thermal analysis and stress analysis are of 8-node body element SOLID70 and SOLID45. The grid of thermal analysis and stress analysis are familiar, shown in Figure 1.

Fig. 1. Finite element mesh for calculation

4 The Results and Discussion 4.1 Temperature Field Distribution Figure 2 is temperature distribution of the first layer of wall in both cases of the welding direction using Z-shaped (the former weld’s arc-extinction point is the post weld’s arcing point) and same-direction (arc-extinction point and arcing point of the two weld put out the same in the horizontal direction), as can be seen from the figure, in the welding process, both ends temperature of weld is higher than the intermediate temperature, it is because heat accumulation is more prone to occur at both ends of the weld. At the same time, when the welding torch taking a different direction, metal temperature distribution obtained form weld prototyping is not the same, both ends temperature and wall average temperature obtained from Zshaped weld is significantly higher than the same direction weld. Because of in the Z-shaped welding, arc starting point coincides with the extinction point, the temperature accumulated more intense than in the same direction welding. When using the same direction weld, arc-extinction point and arcing point has a longer

386

J.-n. Xu et al.

cooling time. For example, the 10th layer average temperature by using the same direction method is 1874.52 , while by using Z-shaped method is 1973.283 . In order to reduce the impact of heat accumulation, using the same direction processing should be more appropriate, Figure 3 is the temperature distribution using Z-shaped processing and the same direction processing in 100S. Using Z-shaped processing, the maximum temperature is 2807 , which higher than the maximum temperature of 2577 by using the same direction processing.

℃

℃

℃

260 240 220 200 180 160 140 120 100

℃

℃

Same direction welding Z direction welding 1 3 5 7 9 1 1 1 1 1 2 2 2

Fig. 2. The maximum temperature of the 10th layer

(a)Z welding

(b) the same direction Surfacing

Fig. 3. The temperature distribution cloud of structure in surfacing to 100s

4.2 Stress Analysis The welding temperature field distributions directly affect the residual stress distribution of structure. Welding residual stress is caused by conventional welding techniques and this defect is almost impossible to avoid, the harm is very great, greatly affect the anti-fatigue, brittle fracture, stress corrosion damage and buckling strength, dimensional stability of welded structure. Therefore, to minimize or eliminate the welding residual stress, is major practical problems. And the structure parts based on welding rapid manufacturing is composed all from the weld, its residual stress distribution will be more complicated, because the follow-up


387

welding layer inevitably produces heat aging effect on the front welding layer, and has greater impact on the final welding temperature field and the residual stress field formation. Figure 4 is the Mises stress distribution of 8th-layer when cooling 300s after welding, can see all are tensile stress, and stress at both ends is larger, Z-type welding residual stress is higher than the same direction welding.

900.00

型堆焊同一方向堆焊 Z

800.00

(MPa)

700.00 600.00

值力500.00 应400.00 余残300.00 200.00 100.00 0.00 0

20

40

60

80

距离 (mm)

100

120

140

Fig. 4. The Mises stress distribution of the 8th layer

Although, the residual stress in the same direction welding is lower than Zwelding, but in the end, the Z-residual stress in the same direction welding is higher than Z-welding, shown in Figure5, while the larger compressive stress have occurred in the last layer end. The σz distribution of Z-welded end-point show wave-like distribution and are all the tensile stress. Because Z-tensile stress is the main reason for arising lamellar tearing in welded structure, so the lamellar tearing most likely to occur at the beginning layer in same direction welding, while at the other layers in Z-welding. The best solution is to preheat substrate before welding and heat-treating the entire structure after welding. 600

型焊接同一方向焊接 Z

500

400

σz /MPa

300

200

100

0 0

2

4

6

8

10

12

14

16

18

20

-100

-200

-300

距离/mm

Fig. 5. σz distribution curve from solder side

5 The Effects of Welding Temperature Field on Weld Metal Using the norms of welding current 120A, wire feed speed 100cm/min, welding speed 150mm/min to process single-channel multi-layer welding, obtains wall

388

J.-n. Xu et al.

ⅠⅡⅢ

structure as shown in Figure 6 (a). By observing the shown position , , from the figure, obtains the metallographic structure as shown in Figure 6 (b), (c), (d).

Ⅰ

Ⅱ Ⅲ

5m

（a）

50um

（c）

50um

（b）

（d）

50um

Fig. 6. Microstructure of wall structure weld

Based on the thermal process simulation of wall welding in the section, we can see that following weld bead will generate heat treatment effect on the former weld in the course of the wall surfacing, create a kind of alternate thermal cycling role. That is, the martensite, bainite of single-channel single-weld can be transformed into tempered structure after heat treatment, tempering structure occur re-crystallization and forming equiaxed organizations similar to normalized organization. Therefore, the last 2 ~ 3 layer seam of wall is martensite (thickness is about 3 ~ 5mm), while the interior is the equiaxed organization similar to normalizing organization. The organization of single-channel single-weld is smaller because of its faster cooling rate, while the wall space for heat dissipation is relatively limited, and the front-seam has the residual heat, so result in the slower cooling rate, therefore the organization of its outer layer is relatively thick. As can be seen in the weld forming process, due to the heat treatment role of following weld, makes the internal organization grain of the molding pieces reconstruction and refinement improves the mechanical properties of the material.


6

389

Conclusions

1. The welding organization is ferrite and a small amount of pearlite. Multilayer welding can refine grains and improve the organization and obtain a smooth surface and forming beautiful parts. These are mainly due to the heat-treatment effects of after layer to previous layer. 2. Middle weld is softened and lead to lower hardness because of drawing effects of after-weld, hardness of the top weld is the highest, close to the true hardness of welding materials. 3. Establishing a three-dimensional heat-machine coupled model based on welding rapid manufacturing simple wall, which effectively simulated the temperature gradient and the resulting residual stress in rapid manufacturing process. 4. In welding rapid manufacturing technology, path planning is equally important. For example, the temperature field and residual stress distribution of Zwelding and the same direction-soldering are not the same. 5. As a whole, welding temperature field and residual stress produced from the same direction-soldering are smaller than those of the Z-welding. However, σz of start welding position in the same direction must be high, otherwise it easily leads to the fracture of walls and base plates. Z-weld shows a fluctuated high z-direction tensile stress at the start that is likely to cause lamellar tearing. So in the welding rapid manufacturing, it is recommended to preheat substrate before welding and heat-treat the entire structure after welding to reduce the residual stress of structure and improve the performance of the structure. Acknowledgement. Thanks to the reviewers for the valuable comments. The work is supported by Special Prophase Project on Basic Research of The National Department of Science and Technology. No. 2005CCA04300.

References [1] Jandric, Z.,Kmecko, I .S., Kovacevic, R.: Dynamic modeling of GTAW for rapid prototyping with welding-based deposition. Technical Paper - Society of Manufacturing Engineers. AD, AD02-242 (2002) [2] Wu, Y., Kovacevic, R.: Mechanically assisted droplet transfer process in gas metal arc welding. Proceedings of the Institution of Mechanical Engineers, Part B, Journal of Engineering Manufacture 216(4), 555–564 (2002) [3] Zhang, Y.M., Li, P., Chen, Y., Male, A.T.: Automated system for welding-based rapid prototyping. Mechatronics 12(1), 37–53 (2002) [4] Lu, F., Yao, S., Lou, S., Li, Y.: Modeling and finite element analysis on GTAW arc and weld pool. Computational Materials Science 29(3), 371–378 (2004) [5] Chu, S.C., Lian, S.S.: Numerical analysis of temperature distribution of plasma arc with molten pool in plasma arc melting. Computational Materials Science 30(3-4), 441–447 (2004)

390

J.-n. Xu et al.

[6] Zhang, H.W., Zhang, Z., Chen, J.T.: 3D modeling of material flow in friction stir welding under different process parameters. Journal of Materials Processing Technology 183(1), 62–70 (2007) [7] Sunar, M., Yilbas, B.S., Boran, K.: Thermal and stress analysis of a sheet metal in welding. Journal of Materials Processing Technology 172(1), 123–129 (2006) [8] Mahapatra, M.M., Datta, G.L., Pradhan, B., Mandal, N.R.: Three-dimensional finite element analysis to predict the effects of SAW process parameters on temperature distribution and angular distortions in single-pass butt joints with top and bottom reinforcements. International Journal of Pressure Vessels and Piping 83(10), 721–729 (2006)

Study on Sub-sea Pipelines Hyperbaric Welding Repair under High Air Pressures Canfeng Zhou1, Xiangdong Jiao1, Long Xue1, Jiaqing Chen1, and Xiaoming Fang2 1 2

Beijing Institute of Petrochemical Technology Offshore Oil Engineering Co, Ltd. e-mail: [email protected]

Abstract. Most Chinese sub-sea pipelines are buried in Bohai Sea less than 60m water depth, where air diving are widely used For the application of offshore pipelines repair, the hyperbaric TIG welding process under high air pressures has been successfully developed. Firstly, the hyperbaric welding test chamber is designed and constructed in laboratory, and the welding machine is manufactured. Then, special welding experiments are carried out based on 16Mn steel plates and pipes under 1-7bar air pressures, and high weld quality is obtained. Lastly, the sea-trial of the welding machine and the welding procedure is carried out in Bohai Sea in China, and a perfect all-position 5G girth weld of 16Mn steel pipe is achieved.

1 Introduction Because high quality joint can be obtained, hyperbaric welding is often used in offshore structures repair. The effects of pressure on electrical performance and weld bead geometry at pressures up to 250bar, equivalent to water depths of 2500m (8,200ft), are investigated, and overall process stability of GMAW is shown in Ref.1. Also under the same high pressure acceptable butt joints including positional linear welds and orbital welds on API 5L X65, API 5L X70 and supermartensitic pipeline steels are produced [2], and a fillet welded sleeve on API 5L X65 pipeline are also obtained successfully [3], which can be of great application value in driverless underwater repair system. In fact, in order to develop driverless underwater repair system applied in tie-in and hot tapping needed in many cases, such as the Langeled pipeline in the North Sea, the study of hyperbaric welding has been carried out continually during the past few years. Hyperbaric GMAW experiments based on X65 steel with low alloyed steel and Inconel 625 wires are carried out under the pressure of 12-35 bars, and all the welds show excellent mechanical properties [4]. In addition, hyperbaric GTAW of X70. Pipeline are also investigated, and the overall properties of the produced weld are exceptional [5]. As is known to all, the circumstances of Bohai Sea differ a lot with that of the North Sea, and the water depth is below 60m, air diving is usually used in underwater repair. In the research of the sub-sea pipeline hyperbaric repair programme, for the sake of flexibility and economical efficiency of the repair operation, T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 391–397. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

392

C. Zhou et al.

compressed air is selected as chamber gases to drive water. However, two problems lie in the selection above, one is that the flammability of objects increases greatly, the other is that the weld pool protection is more difficult [6]. In addition, some other tough problems also must be solved, such as misalignment, arc viewing, and operation safety. The present investigation concerns hyperbaric GTAW process based on 16Mn steel pipe under high air pressures. The hyperbaric welding test chamber is built, the orbital welding machine is manufactured, and successful experiments are carried out.

2 Hyperbaric Welding Equipments 2.1 The Hyperbaric Welding Chamber The inner space of the hyperbaric welding chamber (Fig. 1) is a room of 4.5m ×3.5m×3m, in which 3 divers can carry out tasks in sub-sea of 60m water depth and 2 km flow current. Tools including the welding head are stored in waterproof containers on the chamber sides. 2 manipulators are installed to hang up pipeline from the sea bed, and can move up and down. The chamber can be adjusted along three directions with hydro-cylinders integrated on the frame structure, and each side with an inverted “U” opening to accommodate the pipeline. When the chamber is set in right place and the pipeline is lifted by manipulators, two half rings with rubbers is driven by hydro-cylinders to seal the circle part of the inverted “U” opening. When compressed gases are charged into the chamber, water is driven out from the floor gradually, and a dry hyperbaric space is built.

Fig. 1. The hyperbaric welding chamber

2.2 The Hyperbaric Welding Test Chamber and Pipe Welding Machine In order to develop welding process specifications under hyperbaric conditions in laboratory, the hyperbaric welding test chamber is built, and the pipe welding machine is manufactured (Fig.2). The experiment system is mainly composed of the test chamber, the gases storage tank, the automatic GTAW welding machine, monitoring device and an air compressor. Hyperbaric welding test is carried out

Study on Sub-sea Pipelines Hyperbaric Welding Repair under High Air Pressures

393

by the welding head in the test chamber, a special designed horizontal pressure vessel with inner diameter of 1600mm and a cover are driven by hydro-cylinders, and the welding process is monitored by CCD cameras. Thanks to such a large space, not only 5G girth weld of pipe, but any positional weld of plate can be produced in the chamber. The pipe welding machine in the test chamber is installed on the guiding rail, which is self-propelled and handled by the welders using the remote control box outside the chamber. The three motor-driven axes of the welding bug can move along the guiding rail, oscillate in the weld crosswise direction, and torch up and down. During the welding process, images of arc and pool are trapped by CCD cameras, and can be displayed on the TV screen outside the chamber synchronously.

Fig. 2. The test chamber and pipe welding machine

The tungsten electrode movement is controlled by a DC servo drive system shown as Fig.3. According to the deviation between the arc voltage from the sensor and the set voltage from the remote control box, PLC (Programmable Logic Controller) outputs corresponding regulating voltage to the servo drive system. Driven by the servo motor, the tungsten electrode is moved up and down till the arc length equals to the set value of the control box. None other than the setting arc length, the welding process of the pipeline repair can be carried out perfectly although there is out-of roundness or misalignment to some extent. In addition, compared to power source output voltage, arc length can be easily maintained constant in spite of the complexity of chamber gases and cable, which can bring great flexibility for pipeline repair. U1 Arc voltage

A/D card

U2 Setpoint voltage

PLC

△=U1-U2

Servo drive system

Tungsten electrode

Fig. 3. Ungsten electrode control based on arc voltage feedback

394

C. Zhou et al.

In fact, arc voltage feedback is also the basement of touch striking arc ignition. When the welding power source is just started, U1 equals to open-circuit voltage of the power source. When the operator presses the welding button on the control box, the electrode moves down and contacts with the work piece, U1 instantly changes from open-circuit voltage to a value about 0, then the electrode moves up and the arc is activated.

3 GTAW Arc under High Air Pressures 3.1 Testing Conditions Under high air pressures varying between 1 and 7 bars, welding tests of 16Mn steel plates at horizontal position are carried out to study arc behavior and static characteristics. Around the non-consumable tungsten electrode, there is a ceramic shroud through which higher argon is passed to create an inert atmosphere to protect the electrode from high atmospheric contamination. Filler wire is selected as AWS5.18 ER70S-6 with a diameter of 0.8 mm.

3.2 Arc Behavior and Static Characteristics A high speed camera housed in a special designed pressure proof enclosure is applied to obtain arc image under high air pressures. Typical observations at 1-7 bars are shown in Fig. 4. Because particle density increases in proportion to the ambient pressure, energy losses from the outer regions of the arc increase gives rise to arc cross-section contraction, and arc brightness increase. Undoubtedly, successive frames of arc image from the high speed camera indicate that GTAW process is stable enough under high air pressures at 7 bars.

(a) 0 .1 M P a

( c ) 0 .5 M P a

(b ) 0 .3 M P a

(d ) 0 .7 M P a

Fig. 4. Arc images at high air pressures 1-7 bars

When the arc length equals to 5.5mm, arc static characteristic curves at 1-7 bars are shown as Fig. 5, and several conclusions can be drawn thereby:


z z z

395

Similar to other gas shielded arc welding processes, Arc voltage increases when welding current increases from a rather large value about 50 A. Similar to hyperbaric welding process under Argon ambient conditions, arc voltage increases about 5-10 V when air pressure increase 1 Mpa. Compared to Argon ambient conditions, arc voltage at the same pressure and the same welding current is higher about 1 V, which can be explained from more arc energy losses caused by air.

Therefore, the same welding power source and similar parameters under high Argon ambient conditions can be applied in high air pressures. 0 .1 M P a 0 .2 M P a 0 .3 M P a 0 .4 M P a 0 .5 M P a 0 .6 M P a 0 .7 M P a

17 .5 17 .0

Arc voltage (U)

16 .5 16 .0 15 .5 15 .0 14 .5 14 .0 13 .5 13 .0 12 .5 12 .0 11 .5 11 .0 10 .5 20

40

60

80

100

Welding current

（） 12 0

140

1 60

18 0

A

Fig. 5. Arc static characteristic curves under different pressures

4 Girth Weld under High Air Pressures On the basement of linear weld experiment, butt welds of pipe in 5G at 1-7 bar air pressures were produced, shown in Fig. 6. The welds are completed in multiple passes, including the root pass welded by pulsed current, and other passes welded by constant current about 140-160 A. Compared to welding in atmosphere, in order to realize perfect protection for the arc and weld pool, the pressure and flow of shielding gas is much higher. For example, when ambient pressure is 7 bars, Argon flow is about 50 L/min. All welds are subjected to mechanical testing and meet the requirements of AWS D3.6M:1999.

Fig. 6. Butt welds of pipe in 5G at 1-7 bar

396

C. Zhou et al.

5 Sea-Trial of The Welding Repair System Sea-trial of the welding repair system is carried out in Bohai Sea in China. After the chamber installation, pipeline alignment, water draining by compressed air from board, pipeline cutting and pipe ends sealing, a 5G girth weld of sub-sea pipe is produced, and the welding process is controlled remotely by the welder on the surface after the welding head had been installed on the pipe by the diver in the chamber. A series of photos of sea-trial are shown as below.

Fig. 7. The hyperbaric welding chamber installation

Fig. 8. Welding head installation in the chamber

Fig. 9. The welding process controlled remotely

Fig. 10. Arc image under high air pressures

Fig. 11. Girth weld of sub-sea pipe made in the sea-trial


397

6 Conclusions The research reported here has confirmed that GTAW welding process is capable of producing good quality welds even at high air pressures. Parameters and machines have been developed for positional and orbital welding operations. On the basis of this work, it may be concluded that: z z z z

On the condition of fire-resistant ability and perfect weld pool protection can be assured in the chamber, GTAW also can be realized under high air pressures. Arc voltage feedback plays a key role in GTAW repair of sub-sea pipelines, including arc ignition, arc length control, and flexibility of repair operation. Under high air pressures, arc is stable, and static characteristic curves are similar to hyperbaric welding process under Argon ambient conditions. Under high air pressures, acceptable linear welds and pipe butt welds can be produced.

Acknowledgement. The authors would like to thank current and former sponsors of the Underwater Hyperbaric Welding Research programmers: Ministry of Science and Technology of the People’s Republic of China, National Natural Science Foundation of China, and Offshore Oil Engineering Co., Ltd.

References [1] Hart, P., Richardson, I.M., Nixon, J.H.: The Effects of Pressure on Electrical Performance and Weld Bead Geometry Inhigh Pressure GMA Welding. Welding in the World 45(11/12), 29–37 (2001) [2] Richardson, I.M., Woodward, N.J., Billingham, J.: Deepwater Welding for Installation and Repair – A Viable Technology. In: 12th Int. Offshore and Polar Eng. Conf. ISOPE, Kitakyushu, Japan, vol. 4, pp. 295–302 (2002) [3] Woodward, N.J., Yapp, D., Blackman, S.: Diverless Underwater GMA Welding for Pipeline Repair Using a Fillet Welding Sleeve. In: ASME 5th International Pipeline Conference (IPC), Calgary, Alberta, Canada, vol. 2, pp. 1475–1484 (2004) [4] Akselsen, O.M., Fostervoll, H., Ahlen, C.H.: Hyperbaric Gas Metal Arc Welding of API X65 Pipeline Steel at 12, 25 and 35 Bar. In: 18th Int Offshore and Polar Eng. Conf. ISOPE, Vancouver, BC, Canada, pp. 246–253 (September 2008) [5] Akselsen, O.M., Fostervoll, H., Hårsvær, A.: Weld Metal Mechanical Properties in Hyperbaric GTAW of X70 Pipeline. International Journal of Offshore and Polar Engineering 16(3), 233–240 (2006) [6] Nixon, J.H.: Underwater repair Technology. Woodhead Publishing Ltd (2000) ISBN: 185573-2394

Part V

Special Robot Technology and Systems

The Mechanism Design of a Wheeled Climbing Welding Robot with Passing Obstacles Capability Minghui Wu1,2, Xiaofei Gao2, Z. Fu2, Yanzheng Zhao2, and Shanben Chen3 1

School of Mechanical Engineering, Xiangtan University, Xiangtan 411105, P.R. China 2 State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai 200240, P.R. China 3 Welding Engineering Institute, Shanghai Jiao Tong University, Shanghai 200240, P.R. China

Abstract. The paper describes the mechanical design of a new wheeled climbing robot which is planned to weld and inspect large scale non-structure equipment. The robot includes a 5-DOF manipulator and three pairs of novel wheel-mobile obstacle-passing units. The magnetic adhesion mechanism is a style of structure magnet which is mounted under the chassis of the robot and has a large payload. The lifting mechanism can pull the wheeled unit off the wall surface and adjust the gap between the magnet and the surface to change the adhesion force, using only one actuator. The robot with the novel mechanism has the advantages of having small number of motors, good obstacle-passing and large payload capabilities. While climbing vertical surface, the experiment shows that the robot can pass over a 70mm high obstacle and can carry 50KG payload.

1 Introduction In the process of manufacture and maintenance of large scale non-structure equipment, such as ships, oil storage tanks, nuclear electricity equipment and other steel fabrications (Fig 1), welding and inspecting are the major works. These works have bad work environment and great danger. Using mobile robots to implement these works is an attractive alternative to conventional methods. In the past half century, it has become a hot topic and many researchers have launched plenty of works and manufactured different types of climbing robots which are used in the manufacturing and daily maintenance of large equipments. The robot aimed at welding and inspecting is required to climb on vertical wall and carry with some sensors and attachment equipment, it requires the robot to have a high mobility and large payload capabilities. But unfortunately, these usually lead to that the robot has a complex structure and large weight. It is a challenge to develop a climbing welding robot which has satisfying mobile and payload capabilities but light weight and simple structure. For wall climbing robot, the locomotion T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 401–409. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

402

M. Wu et al.

mechanism and adhesion mechanism are the keys. Many researchers have done much work on this field and some new climbing welding robots have been developed. Track and wheel are the two most common locomotion mechanisms of climbing robot. As mentioned in Ref [1], a four-wheel mobile welding robot developed for welding cabin grid component of ship. The robot has a 2-DOF cross manipulator and can adjust the welding torch to trace the welding seam precisely. But the robot has no adsorption mechanism. Considering most of the welding robots work in the ferromagnetic environment and permanent magnet adsorption has outstanding performance, permanent magnetic adsorption wheeled mechanism and tracked mechanism are adopted for robot. For example, Fabien Tache developed a magnet-wheel climbing robot for inspecting complex shaped pipe [2]. While magnetic wheel climbing robots have small contacting area between wheels and ground, it leads to low magnetic efficiency. On the contrary, the magnetic tracked robots have big contacting area and adhesion force. The magnet-track climbing robots are paid enough attention [3]. In order to overcome these disadvantages, a non-contact magnet adhesion mechanism called as structure magnet was developed for wall climbing robot [4-5] and which is mounted under the chassis. From the paper of Ref [6], we can calculate the value of the magnetic energy density (MED) of magnetic wheel is about 22.5, while the value of the structure magnet is approximately 52. The paper of Ref [5] proposed a welding robot with the type adhesion mechanism and its payload capability can reach as high as 100 kg.

Fig. 1. Ship and oil tank welding scene

These robots, as noted above, don’t have mostly passing obstacles capability and can’t change the magnetic force. These disadvantages restrict the robot’s practical use. To overcome them, some robots are equipped with special mechanisms within their structure. For example, a novel solution to a mobile climbing robot with magnetic wheels was described in Ref [7], the robot has 4 pairs of magnetic wheels and the forward and backward pairs can be moved up and down by linear actuators in order to pass obstacles. Another robot of “Pipe-Surface Inspection Robot” [8] which has 3 pairs of magnetic wheels can traverse flanges along the outside of piping. Although the two robots have passing obstacles ability, the former one needs 8 wheels to surmount obstacles and the robot of Suzuki needs 12 actuators for force-decreasing and wheel-lifting. This leads to a complicated structure.

The Mechanism Design of a Wheeled Climbing Welding Robot

403

2 Objectives and Challenge The paper proposed the project that the All-position Autonomic Welding (APAW) robot is planned to develop and the robot is aimed at welding and inspecting large scale non-structure equipment, such as ships, large storage tank and other steel fabrications. While in process of welding and inspecting, the APAW robot often need to climb on vertical walls, pass obstacles. At the same time, the APAW robot need carry some heavy equipment, such as power equipment, power cables, welding torch, control unit, sensors and others. So the robot should have high mobility and payload abilities. Although some literatures show some wheel climbing robots without additional mechanisms can surmount simple obstacles [9]. But for more difficult obstacles, these robots can do nothing. As we said in section 1, some robots use active elements or extra mechanisms to lift up the wheels [7-8], but these robots require too many wheels or actuators which makes the robot huge, expensive and difficult to control. Some examples of the real work environment for manufacturing ship and storage tank are shown in Fig 1. Considering most of the welding climbing robots work in the ferromagnetic environment and permanent magnet adsorption has outstanding performance, for a welding climbing robot, the magnet is a good select. As mentioned previously, there are a great number of robots with magnet adhesion have been developed by researchers in different organizations. There are some problems in current welding robots need to solve: 1). Magnet-wheel robot and magnet-track robot have its own advantages and disadvantages. It is a challenge to design robot which has both high mobility and payload capabilities. 2). Most climbing welding robot with passing obstacles capability have complicated structure and too much motors.

3 Concept Structure of the Robot On the basis of analyzing the existent wall climbing robots and mobile welding robots, we take the advantages of these robots and design a novel adhesion mobile mechanism for the robot. The robot is composed of a 5-DOF manipulator and three identical such mechanism. The concept structure of the robot is shown Fig 2. Each adhesion mobile mechanism includes a wheel unit, a structure magnet and a lifting mechanism (Fig 3). The wheel unit includes two driving wheels and an axle. The turning of robot is realized by controlling the speed and direction of both wheels with differential steering. The structure magnet, which is mounted under the chassis of the robot, is composed of yoke and permanent magnet. The lifting mechanism includes a spindle, two trapezoid nuts and two trails. The screw-pitch of the two trapezoid nuts is different, one is mounted on adhesion unit and the other is mounted on axle. The spindle is driven by an actuator through a timing belt and making the wheel unit and structure magnet move up and down together. When the robot meets an obstacle the three pairs of adhesion mobile mechanism are pulled off the wall in turn, by this way the robot can climb over obstacles. At the same time the lifting mechanism can move the structure magnet up and down relative to the wheel unit and this leads to the air gap changing between the

404

M. Wu et al.

structure magnet and wall. By this way the magnetic adhesion force is adjusted by changing the air gap. The robot with the novel mechanism has large payload and passing obstacles capabilities.

Fig. 2. The concept structure of the robot

Fig. 3. Sketch of adhesion mobile mechanism

4 The Mechanism Design of the Robot We develop a prototype base on the concept structure of the robot which is composed of a 5-DOF manipulator and three identical adhesion mobile mechanisms (Fig 4). The CAD-model of adhesion mobile mechanism is shown in Fig 5. There are fourteen suits of actuator and controller for the robot all together, 5 for manipulator, 6 for driving wheels and 3 for lifting mechanism. The outline dimension of the robot is about 680 mm in length, 440 mm in width, and 280 mm in height. The height of robot can change from 220 mm to 300 mm by the lifting mechanism. To save weight, most parts of robot are made of aluminum alloy. Adding the manipulator the total weight of the robot is about 50 kg.

Fig. 4. The structure of the robot

4.1 The Design of the Manipulator To achieve the accurate control of welding torch optimal orientation and dynamic process of welding is the key to realize the high quality robot weld. In the process of manufacture and maintenance of large scale non-structure equipment, there are some complicated welds, such as fillet welds and curving welds. While moving


405

and passing over obstacles, it is necessary for the robot that to accomplish the coordination control of the end effector of manipulator. Theoretically, a 6-DOF manipulator can point to any position with any orientation in its workspace. In the process of welding, it is not necessary for the DOF that the weld torch revolves about the axis of welding wires, so an articulated 5-DOF manipulator meets the requirement of welding complicated welds in non-structure environment. The photo of the manipulator is shown as Fig 5. The wrist of robot has two joint whose axes are orthogonal. Velocity and position control accuracy of manipulator mobile has special requirement, so DC servomotors and harmonic reducers are selected to drive wheels of the robot. The length L1of arm 1 is 150 mm (the distance between Z1and Z2). The lengths of other three arms are 180 mm, 250 mm, 80 mm respectively. Load capability of the manipulator is 30 N.

4.2 The Design of the Wheeled and Lifting Mechanism The wheel unit includes two driving wheels, an axle and two actuators with planetary gearboxes (Fig 5). The wheel’s diameter is R=135 mm and covered with a thick layer of polyurethane rubber to increase the friction, and each wheel is driven by an actuator separately. The turning of robot is realized by controlling the speed and direction of both wheels with differential steering. Actuators mounted inside axle drive wheels through another gear reducer (i=2) which is fixed inside the wheel (Fig 5). We select a DC Servomotor with a four stages planetary gearbox (i=304). The normal output torque is 18Nm and the output power is 78W. With the normal output speed after the planetary gearbox being 19 rpm, we can calculate the motion speed is (19/2)*pi*R=4 m/min. This value is fast enough for robot to weld and inspect.

(a)

(b)

(a) Adhesion mobile mechanism; (b) Partial view. Fig. 5. CAD-model of the adhesion mobile mechanism

The structure of lifting mechanism is shown in Fig 5. The lifting mechanism includes two trapezoid screw nuts, a trapezoid screw spindle, an actuator, a timing belt and two linear motion guides. The two nuts have different pitches, the one with 3 mm pitch is mounted on the yoke of absorption unit and the other with 5

406

M. Wu et al.

mm pitch is fixed on the axle of wheel-unit. The trapezoid screw spindle has two segments of screw which fits the two nuts respectively and forms kinematic pairs. When the spindle is driven by the actuator through the timing belt and pulleys the wheel-unit and absorption unit move up and down along the linear motion guides. The wheel-unit can be lifted 70 mm high, while the gap width between the adhesion unit and surface can be changed from 2 mm to 20 mm (Fig 6). In order to reduce tension force between spindle and nuts, four springs are installed between adsorption unit and axle around the guide bars to produce internal balanced force. The coefficient of spring elasticity is k=50 N/mm. By this means, the frication force is reduced between the spindle and the nuts.

(a)

(b)

(a) The structure magnet in highest position; (b) The structure magnet in lowest position Fig. 6. Photos of moving up and down magnetic adhesion unit

4.3 Optimal Design of the Adhesion Unit On the basis of analyzing the existent adhesion mechanisms of wall climbing robot, we take the advantages of the track and wheel adhesion mobile mechanism and select the type of structure magnet as the magnetic adhesion unit of the robot. The robot has three adhesion units in all, and each consists of a piece of yoke and many pieces of permanent magnet (Fig 7). We select sintered NdFeB (N45SH) as adhesion unit’s material; the yoke material is electric pure iron DT which has high magnetic permeability and saturation magnetic induction. Suppose that vertical surface is Q235 steel board and its thickness is 10 mm. The adhesion force and weight are the most important performance parameters of adhesion unit. The optimization target is to get as large force as possible with the least possible weight of adhesion unit. Magnets, air gap and adsorption surface form a complex magnetic circuit, and the parameters of thickness, width, yoke iron thickness and air gap thickness have large influence on the adsorption force. Optimal design of the structure parameters of magnets has great significance. Traditional models always lead to large errors, but the finite element methods (FEM) is an effective way to solve these problems. In this paper, the finite element software Ansoft Maxwell V10 is used to optimize the structure parameters of magnet adhesion unit. Based on the optimal result, we made a sample of magnetic


407

adhesion unit (Fig 7). The width of permanent magnet 1 and permanent magnet 2 are respectively 25 mm and 50 mm; the thickness of yoke and permanent magnets are respectively 9mm and 12mm;the distance of space between two permanent magnets is 11 mm. The size of each magnetic adhesion unit is 244×120×21 mm3 and the total weight of one unit (including permanent magnets and yoke) is Gm=4.33 kg. We tested the adhesion force of the magnetic adhesion unit in different heights of air gap. The result of test and simulation are shown in Fig 9.

8000 Magnetic Force• N)

Simulation result

4000 2000 0

Fig. 7. Magnet adhesion unit

Test result

6000

4

6

8 10 12 Air Gap High• mm•

14

16

Fig. 8. Results of simulation and test

5 Experiment of Passing Obstacles We designed an experiment plant on which is used to simulate the manufacturing environment of large non-structure equipment. Some sensors, such as distance and visual sensors are installed on the robot to recognize obstacle and measure its distance, width and height and design a real-time controller to control robot to pass over obstacle. In experiment, the robot moved up and down on vertical surface and overcome a 70 mm high obstacle. The operations and the sequence of overcoming obstacles are shown in Fig 9.

(a)

(b)

(c)

(a) The photo of experiment plant; (b) The robot climbing on vertical surface; (c) The robot passing obstacle Fig. 9. Photos of the robot overcoming the obstacle on vertical surface

408

M. Wu et al.

6 Conclusion and Future Work In this paper, a climbing welding robot with a novel lifting mechanism and adhesion mechanism is proposed. The lifting mechanism can pull the wheel unit off the wall and change the magnetic adhesion force to help the robot pass obstacles, using only one motor. The adhesion mechanism is a structure magnet which is mounted under the robot’s chassis and has a bigger adhesion force than the magnetic wheel. The new design not only reduces the number of motors but also makes the robot with better mobility and payload capabilities. The experiment shown that the climbing welding robot works well and has a large payload capability (≥50 kg) and can pass 70mm high obstacles on a vertical surface. Now, although experiments have proved that the new mechanism works well, the robot system needs to be perfected and motion control need to be further studied. The ongoing work also stresses on integrating sensors and the necessary electronics for controlling the robot. The robot is planned to be used in welding and inspecting some large equipments. In future research, we plan to develop an industrial version intelligent welding robot which can navigate, pass obstacles and seek seam automatically in an unknown environment. Acknowledgement. The support of the national High Technology Research and Development Program of China, through the Ministry of Science and Technology of China Grant No. 2009AA04Z221, is gratefully acknowledged.

References [1] Yang Bae, J., Byoung Oh, K., et al.: Seam tracking and welding speed control of mobile robot for lattice type welding. In: Proceedings of ISIE 2001 IEEE International Symposium on Industrial Electronics, June 12-16. IEEE, Piscataway (2001) [2] Tache, F., Fischer, W., et al.: Compact magnetic wheeled robot with high mobility for inspecting complex shaped pipe structures. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, October 29 -November 2. IEEE, Piscataway (2007) [3] Gao, X., Xu, D., et al.: Multifunctional robot to maintain boiler water-cooling tubes. Robotica 27(6), 941–948 (2009) [4] Gui, Z., Chen, Q., et al.: Wall climbing robot employing multibody flexible permanent magnetic adhesion system. Chinese Journal of Mechanical Engineering 44, 177–182 (2008) [5] Kitai, S., Tsuru, K., Hirose, S.: The Proposal of Swarm Type Wall Climbing Robot System Anchor Climber. In: Proc. IROS, pp. 3999–4004 (2005) [6] Fischer, W., Caprari, G., et al.: Foldable magnetic wheeled climbing robot for the inspection of gas turbines and similar environments with very narrow access holes. Industrial Robot 37(3), 244–249 (2010)


409

[7] Bridge, B., Leon Rodriguez, H.E., et al.: Field trials of a cell of climbing cooperating robots for fast and flexible manufacturing of large scale engineering structures. In: Proc. of The 12th International Conference on Climbing and Walking Robots and the Support Technologies for Mobile Machines, CLAWAR (2009) [8] Suzuki, M., Yukawa, T., et al.: Mechanisms of Autonomous Pipe-Surface Inspection Robot with Magnetic Elements. In: 2006 IEEE International Conference on Systems, Man, and Cybernetics (2006) [9] Gui, Z., Chen, Q., et al.: Optimization of permanent-magnetic adhesion device for wall-climbing robot. Diangong Jishu Xuebao/Transactions of China Electrotechnical Society 21(11), 40–46 (2006)

Anytime Ant System for Manipulator Path Planning D. Wang1, N.M. Kwok1, G. Fang2, and Q.P. Ha2 1

School of Mechanical & Manufacturing Engineering, University of New South Wales, Australia e-mail: [email protected], [email protected] 2 The School of Electrical, Mechanical and Mechatronic Systems, University of Technology, Sydney, Australia e-mail: [email protected]

Abstract. An efficient algorithm for manipulator path planning is presented in this paper. Because of the complexity of the problem nature, it frequently takes a long time for the planner to find an optimal path. This drawback may hinder a robotic manipulator system from many real-time applications. In this research work, the concept of anytime algorithm is integrated into a novel swarm intelligence method, the Ant System with Negative Feedback (ASNF). With the proposed Anytime Ant System (AAS), a planner is able to find a suboptimal solution quickly, then improve the quality of this solution while time allows. Simulations based on a two-link manipulator have been carried out to demonstrate the feasibility and effectiveness of the proposed approach.

1 Introduction This research studies the problem of manipulator motion planning, which can be formulated as finding a series of feasible poses or joint angles connecting the initial configuration and the desired goal configuration. Researchers have invested heavy efforts on this topic and many insightful approaches have been presented [1]. In the potential field method, a manipulator is designed to be repelled by obstacles and attracted by its goal [2]. The potential field method and its variants are mathematically simple and fast but may suffer from being trapped in local minima [3-5]. Approaches based on the octree model avoid the time-consuming process of generating configuration space and find collision free paths by searching grid points directly in the Euclidean space [6, 7]. These algorithms are fast but may fail to find a feasible solution even if it exists. Heuristic algorithms, such as the Probabilistic Roadmap method (PRM) [8] and Rapidly-exploring Random Tree (RRT) method [9], have been developed to enhance the search process in configuration spaces. Swarm intelligence methods, such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) have also been successfully applied in the manipulator motion planning and collision avoidance problems [10-12]. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 411–420. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

412

D. Wang et al.

Although some aforementioned algorithms are able to produce solutions with good quality, they can take an extended period of time to run, which is not always practical in real applications. The class of Anytime Algorithms is algorithms that trade execution time for quality of results [13-15]. Anytime algorithms are able to return a suboptimal solution, whose quality depends on the amount of computation they were able to perform. The answer generated by anytime algorithms is an approximation of the correct answer and they are improved progressively over time. In addition, from observing the success of many natural species, several novel algorithms had been developed for robotic manipulator planning. Inspired by the experimental studies on ant foraging behavior under crowded conditions, an Ant System with Negative Feedback approach (ASNF) has been present in [16]. ASNF differs from existing ACO algorithms by introducing a new pheromone, the crowded pheromone, into the algorithmic framework. The crowded pheromone is determined by an individual’s evaluation on current solutions and acts as negative feedback in the framework. The interplay of crowded pheromone and path seeking pheromone forms a self-adaptive system for best total performance. In this paper, the ASNF is applied in the manipulator path planning and collision avoidance. The presented Anytime Ant System (AAS) integrates the concept of anytime computing into ASNF. AAS is designed to be able to generate an initial solution quickly and improve its quality over subsequent iterations. In section 2, we will give a brief introduction on the ASNF. Section 3 introduces the Anytime Ant System approach in detail. Section 4 tests the presented approach with a 2-link manipulator case. Conclusions and future work are given in Section.

2 Ant System with Negative Feedback 2.1 Ant Colony Optimization The Ant Colony Optimization algorithm (ACO) is inspired by observing real ants’ forging behavior. It was found that ants deposit chemicals called pheromone on the ground when they walk between a food source and their nest. The initial path arises from the random exploration of ants. Once a feasible path has been found, pheromone will be deposited on the path by the returning ants. Since ants prefer to search for food resource following the paths with higher pheromone concentration, the difference of pheromone intensity is amplified when a path has been taken by more ants. Through this positive feedback mechanism, ants are capable of transporting food along the shortest path between food sources and nests [17]. The first ACO algorithm was proposed by Dorigo and colleagues in the 1990s and is known as the Ant System (AS) [18]. Several modifications have been proposed to improve the performance of the basic AS algorithm. In the Ant Colony System (ACS), each ant performs a local pheromone update in addition to the global pheromone update. This decreases the pheromone concentration on the traversed edges and encourages subsequent ants to choose other edges [19]. In the MAX-MIN Ant System (MMAS),only the ant with current best solution updates

Anytime Ant System for Manipulator Path Planning

413

the pheromone values [20]. The pheromone values in MAX-MIN Ant System are bounded. For a comprehensive review of ACO algorithms, refer to [21].

2.2 Ant System with Negative Feedback The Ant System with Negative Feedback (ASNF) has been proposed in [16]. The ASNF is inspired by the experimental studies on ant foraging behavior under crowded conditions [22, 23]. Two types of diamond-shaped bridges used in their experiments are shown in Fig. 1. The first bridge has two branches with equal lengths but different widths. The second bridge has two branches and also with different lengths and widths. Experiments have shown that a wider branch is always preferred by ants in highly crowded conditions, even if it is longer. The reasons underlying this phenomenon are: 1) The difference in travel duration between the two branches of the bridges because of that congestion on the narrow branch results in a longer travel duration. 2) The “pushing” actions between ants making head-on encounters at the entrance of the two branches of the bridges, as shown in Fig. 2. It was found that overcrowding on a narrow branch tends to result an increase in the travel time and a number of “pushing” action [23]. In a “pushing” scene, when two ants meet at the entrance of branches, the ant from a crowded path diverts another ant to a path with better traffic condition (Fig. 2). As a result, ants in the experiments choose the wider path which minimizes their travel time instead of path length. In a word, the path selection is a trade-off between travel time and travel cost. In a foraging process, ants prefer to choose paths with higher pheromone concentration. The pheromone on certain path acts as a positive feedback, through which a shortest path to food source can be established [17]. In the experiments described above, congestion occurs on the narrow passage, which slows down the progression of the ants and leads to a decrease in food transportation efficiency. The “pushing” behavior between ants now plays the role of negative feedback, which diverts ants to other paths. The interplay between pheromone accumulation and “pushing” behavior forms a self-adaptive foraging strategy. Inspired by findings aforementioned, the Ant System with Negative Feedback (ASNF) has been presented in [16]. Unlike other existing ant based approaches, two types of pheromone are used in ASNF. The first one is called “Path Seeking Pheromone” (PSP), which works the same way as pheromone in traditional ACO algorithms. The PSP is deposited by ants on their ways to attract other ants. The second pheromone is called “Crowded Pheromone” (CP), which denotes the degree of crowdedness of an ant's current path. The meaning of CP should vary with different problems. For example, it is defined as a function of the number of “collisions” between ants. A “collision” is defined to be a situation that an ant is in another ant's vicinity. When an ant is on its way back to the nest, it counts the number of “collisions”. If the collision number exceeds a threshold value, the ant deposits an amount of CP on its path. The methodology of ASNF is detailed below.

414

D. Wang et al.

In the initialization phase, ants are positioned at their starting points. At each iteration, an ant senses the pheromone concentration at its neighbouring locations and determines its next position in a probabilistic manner. Consider a grid map shown in Fig. 3, in which an ant is positioned at P. The possible next position will be selected from its eight neighboring grids denote by Pi , i = 1,2, … 8 . The distances from P to Pi are denoted by d PPi .

For ant k, the transfer probability (TP) from point P to Pi is given by: ⎧ [τ Pi ]α [η PPi ]β (1 + [ϕ Pi ]γ ) ⎪ , if Pl ∈ N ( s k ) α β γ TPi k = ⎨ ∑ + ϕ ( 1 [ ] ) k [τ P ] [η PP ] Pi P ∈N (s ) i i ⎪ l 0 otherwises ⎩

Fig. 1. Bridges in Dussutour et al’s Ant Experiments [23]

Fig. 2. Ant Pushing under Crowded Conditions [22]

Fig. 3. P and Eight Neighboring Points

(1)


415

where τ Pi is the PSP concentration at point Pi . η PPi is the heuristic value representing the cost of choosing point Pi , which can be set as 1 d PPi , d PPi is the dis-

tance between point P and Pi ). ϕ Pi is the CP concentration at the point Pi . α, β and γ are parameters indicating the relative importance of the path seeking pheromone, the heuristic value and the crowded pheromone, respectively. N ( s k ) is the set of feasible neighboring points Pl . In each iteration, the path seeking pheromone concentration of each point P is updated as follows: m

τ P ← [(1 − ρ1 ) ⋅ τ P + ∑ Δτ Pk ]ττ max min k =1

(2)

where ρ1 is the PSP evaporation rate, m is the number of ants. τ max and τ min are k respectively the upper and lower bounds of PSP. Δτ P represents the quantity of path seeking pheromone lay on point P by ant k and is given by:

⎧⎪Q + Q2C Pk , if ant k goes to nest via P Δτ Pk = ⎨ 1 ⎪⎩ 0, otherwises

(3)

k where Q1 and Q2 are constants and C P represents the cost from point P to nest k

via ant k's current path. Please note that C P is not the straight-line distance between point P and the nest since ants may take zigzag movements. The crowded pheromone concentration of each point P is updated as follows: m

ϕ P ← [(1 − ρ 2 ) ⋅ ϕ P + ∑ Δϕ Pk ]ϕϕ max min k =1

(4)

where ρ 2 is the CP evaporation rate. ϕ max and ϕ min are the upper and lower bounds of CP, respectively. Δϕ P represents the quantity of crowded pheromone laid on point P by ant k. Δϕ P is defined as: k

⎧⎪Q ⋅ (C nk − Q4 ), if Cn ≥ Q4 Δϕ Pk = ⎨ 3 ⎪⎩ otherwises 0,

(5)

k where Q3 is a constant. Cn is the number of collisions recorded by ant k on point P to nest. Q4 is a threshold parameter. An ant will start to deposit crowded phek romone on its path when Cn ≥ Q4 .

416

D. Wang et al.

ASNF differs from other ACO algorithms by introducing a new pheromone, the crowded pheromone, into the algorithmic framework. The crowded pheromone is determined by individual's evaluation on current solutions and acts as negative feedback in the framework. The interplay of crowded pheromone and path seeking pheromone forms a self-adaptive system for best overall performance.

3 Anytime Ant System for Manipulator Path Planning In this section, we introduce the proposed Anytime Ant System approach. As other swarm intelligence methods, it may take a long time for the ASNF to find the best solution. The concept of anytime algorithm is adopted to find a suboptimal solution quickly. At any time the algorithm maintains the best solution found so far can be returned and will be improved continuously as iteration continues. Figure 4 gives the flow chart of the proposed approach. Firstly, a configuration space is generated for a given task. The configuration space is then decomposed into grids at a certain precision level. The manipulator path planning and collision avoidance problem is hence converted into a grid search problem. The ASNF is then adopted to search for feasible solutions. Please note that the search process will be greatly accelerated when the grid cells are large. However, the planner may fail to find a solution because of the rough approximation of search space. If a feasible solution cannot be found in a certain time or number of iterations, the planner will re-decompose the configuration space into smaller grids. In ASNF, each grid contains a certain amount of pheromone, i.e. PSP and CP. When a grid is divided into smaller grids, its pheromone intensities are inherited by all descendants. Every ant in ASNF keeps a memory of the current best solution found by it. Therefore, the search process can be interrupted and restarted at any time point. Once a suboptimal solution is found, it will be evaluated. This path will be executed if it meets the lowest requirement. Otherwise, ASNF continues to run until a better solution is found. The selected path will be improved continuously while a planned path is executed. The quality of real path depends on assigned computation time.

4 Case Studies The simulations have been carried out in a 10m×10m area as shown in Fig. 5. A 2link RR manipulator is supposed to move from its initial configuration to the goal configuration while avoids collision with two obstacles. The first link is 2.5 meters long and 0.3 meters wide. The second link is 1.5 meters long and 0.3 meters wide. The pose of each link is represented by the angles between links and the X coordinate.


417

Fig. 4. AAS Flowchart

The initial pose of the manipulator is shown in Fig. 5, where the joint angles from link 1 and link 2 to X coordinate are 340 degrees and 320 degrees, respectively. The manipulator’s initial pose and desired pose are shown in red polygons and black polygons, respectively. The base of this manipulator locates at (5, 5).The position of its end effector is at (8.498, 3.181) and is supposed to reach goal (5, 8). The configuration space is shown in Fig. 6. At the goal configuration, the joint angles from link 1 and link 2 to X coordinate are 120 degrees and 34 degrees, respectively. The ASNF parameters used in the simulations are set as: m = 50 , α = 1 , β = 1 , γ = 0.5 , ρ1 = 0.01 , ρ 2 = 0.01 , Q1 = 3 , Q2 = 0.5 , Q3 = 1 , Q4 = 5 , τ max = 50 , τ min = 0 , ϕ max = 10 , ϕ min = 0 . The evaluation function is set to be a function of the path length in configuration space, such as f (cost) < k ⋅ d q start q goal

(6)

where d q start q goal is the straight-line distance between q start (340 ,320 ) and q goal (210 ,124 ) , which is equal to 161.6 in this case. k is a positive constant with k > 1 .

418

D. Wang et al.

Fig. 5. Simulation Environment

Fig. 6. Configuration Space

In the initialization process, the configuration space is divided into a grid map at units of 10 degrees, as shown in Fig. 7. Note that since the unit of coordinates is 10 degrees, the start pose and goal pose are now q start (34,32) and q goal (21,12) , the distance between the two points is 23.854. We set k = 2.5 , then the threshold given by evaluation function (6) is modified to f (cost) < k ⋅ d q start q goal = 2.5 * 23.854 = 59.635

(7)

The simulations have been carried out on a laptop with Inter Celeron 2.2GHz and 2G RAM, running Matlab 2009b in Windows 7 environment. It takes the planner ' 4.396 seconds to find the first solution (87 iterations). In Fig. 7, the unit of θ1 and

θ 2' is 10 degrees. The planned path is given by the polyline connecting q start and q goal . The total length of the resultant path is 105.225, which does not satisfy the evaluation function. The iteration continues and more solutions are obtained. Figure 8 shows a solution found in iteration 548. It takes the planner 46.747 seconds to compute this path. The length of this path is 59.355, which is less than the threshold in (7). This path is considered acceptable, so the manipulator begins to move along this path. The AAS is able to improve the resultant path during task execution. The quality of the final path depends on assigned computation time. Our simulation is stated as below: when the manipulator reaches pose (30, 23), the task execution is paused. The configuration space is decomposed at unit of 5 degrees. Note that q start is now at (68, 64) and q goal is at (42, 25). The planner continues to run and a near optimal path is show in Fig. 9. The path length of the whole path is 76.355.

5 Conclusions and Future Work This paper has presented an anytime algorithm for manipulator path planning. With the proposed Anytime Ant System, a planner is able to search a collisionfree path in the configuration space. The main merit of this approach is that an initial suboptimal solution can be found very quickly, which makes it suitable for


419

real-time applications. This approach is resolution complete. The quality of the resultant paths can be improved as iteration continues. Simulations with a two link RR manipulator have been carried out to demonstrate the feasibility and effectiveness of proposed approach. In this research work, a solution’s quality is evaluated by a function of path length in the configuration space. Future work will include taking other factors into consideration, such as minimizing overall task execution time and allowing sufficient clearance to obstacles. We are also considering extending this approach to redundant manipulators with higher degrees of freedom.

Fig. 7. A Path found at Iteration 87

Fig. 8 A Path found at Iteration 548

Fig. 9. The Final Path

References [1] Latombe, J.-C.: Robot motion planning. Kluwer Academic Publishers, Boston (1991) [2] Khatib, O.: Real-time obstacle avoidance for manipulators and mobile robots. International Journal of Robotics Research 5, 90–98 (1986) [3] Brock, O., Khatib, O.: Real-time re-planning in high-dimensional configuration spaces using sets of homotopic paths. In: Proceedings of IEEE International Conference on Robotics and Automation, pp. 550–555 (2000) [4] Chotiprayanakul, P., Wang, D., Kwok, N.M., Liu, D.K.: A haptic based human robot inter-action approach for robotic grit blasting. In: Proceedings of the 25th International Symposium on Automation and Robotics in Construction (ISARC 2008), Vilnius, Lithuania, June 26-29, pp. 148–154 (2008) [5] Chotiprayanakul, P., Liu, D.K., Wang, D., Dissanayake, G.: Collision-free trajectory plan-ning for manipulators using virtual force based approach. In: Proceedings of the International Conference on Engineering, Applied Sciences, and Technology (ICEAST 2007), Swissôtel Le Concorde, Bangkok, Thailand, November 21-23 (2007) [6] Wu, L., Hori, Y.: Real-time collision-free path planning for robot manipulator based on octree model. In: Proceedings of the 9th IEEE International Workshop on Advanced Motion Control, pp. 284–288 (2006) [7] Hamada, K., Hori, Y.: Octree-based approach to real-time collision-free path planning for robot manipulator. In: Proceedings of the 4th International Workshop on Advanced Motion Control, March 18-21, vol. 2, pp. 705–710 (1996)

420

D. Wang et al.

[8] Kavraki, L.E., Svestka, P., Latombe, J.-C., Overmars, M.H.: Probabilistic roadmaps for path planning in high-dimensional configuration spaces. IEEE Transactions on Robotics and Automation 12(4), 566–580 (1996) [9] LaValle, S.M., Kuffner, J.J.: Rapidly-exploring random trees: Progress and prospects. In: Donald, B.R., Lynch, K.M., Rus, D. (eds.) Algorithmic and Computational Robotics: New Directions, pp. 293–308. A K Peters, Wellesley (2001) [10] Tewolde, G.S., Sheng, W.: Robot Path Integration in Manufacturing Processes: Genetic Al-gorithm Versus Ant Colony Optimization. IEEE Transactions on Systems, Man and Cyber-netics, Part A: Systems and Humans 38(2), 278–287 (2008) [11] Wen, Z., Luo, J., Li, Z.: On the global optimum motion planning for dynamic coupling robotic manipulators using particle swarm optimization technique. In: Proceedings of the IEEE International Conference on Robotics and Biomimetics (ROBIO), December19-23, pp. 2177–2182 (2009) [12] Li, H.-X., Shi, Y.-H., Wang, G.-R., Zheng, X.-X.: Automatic Teaching of Welding Robot for 3-Dimensional Seam Based on Ant Colony Optimization Algorithm. In: Proceedings of the Second International Conference on Intelligent Computation Technology and Automation (ICICTA 2009), October 10-11, vol. 3, pp. 398–402 (2009) [13] Zilberstein, S.: Using Anytime Algorithms in Intelligent Systems. AI Magazine 17(3), 73–83 (1996) [14] Ferguson, D., Stentz, A.: Anytime RRTs. In: Proceedings of the 2006 IEEE/RSJ Interna-tional Conference on Intelligent Robots and Systems (IROS 2006), pp. 5369–5375 (October 2006) [15] Likhachev, M., Gordon, G., Thrun, S.: ARA*: Anytime A* with provable bounds on sub-optimality. In: Advances in Neural Information Processing Systems. MIT Press, Cambridge (2003) [16] Wang, D., Kwok, N.M.: Ant System with Negative Feedback for Bushfire Fighting with Swarm Robots. Submitted to 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems (2010) [17] Deneubourg, J.-L., Aron, S., Goss, S., Pasteels, J.-M.: The Self-organizing Exploratory Pattern of the Argentine ant. Journal of Insect Behavior 3, 159–168 (1990) [18] Dorigo, M., Maniezzo, V., Colorni, A.: Ant System: Optimization by a Colony of Cooperating Agents. IEEE Transactions on Systems, Man, and Cybernetics, Part B 26(1), 29–41 (1996) [19] Dorigo, M., Gambardella, L.M.: Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem. IEEE Transactions on Evolutionary Computation 1(1), 53–66 (1997) [20] Stzle, T., Hoos, H.H.: MAX-MIN Ant System. Future Generation Computer Systems 16(8), 889–914 (2000) [21] Dorigo, M., Birattari, M., Stzle, T.: Ant Colony Optimization. IEEE Computational Intelligence Magazine 1(4), 28–39 (2006) [22] Dussutour, A., Fourcassi, V., Helbing, D., Deneubourg, J.-L.: Optimal Traffic Organization in Ants under Crowded Condition. Nature 428, 70–73 (2004) [23] Dussutour, A., Nicolis, S.C., Deneubourg, J.-L., Four-cassi, V.: Collective Decision in Ants under Crowded Conditions. Behavioral Ecology and Sociobiology 61, 17–30 (2006)

Path Planning and Computer Simulation of a Mobile Welding Robot Tao Zhang and Shanben Chen Welding Engineering Institute, Shanghai Jiao Tong University, Shanghai 200240, P.R. China e-mail: [email protected]

Abstract. For a mobile welding robot system, we calculate its kinematic and inverse kinematic model based on the D-H method. This paper uses the method of interpolation to do path planning of a mobile welding robot with a 5-DOF arm, and adopt the method to make various welding path simulation. The trajectory planning is a constrained and non-liner implication problem. The modeling and simulation is processed under the environment of MATLAB. When the vehicle is walking across obstacles, it should do the welding at the same time. We must ensure the continuity of the welding process. The whole process is discussed in the paper.

1 Introduction Trajectory of the robot arm movement is the displacement, velocity and acceleration of robot in the process of motion. Trajectory planning is to calculate the expected trajectory based on tasks. Robot trajectory planning is a bottom plan, which basically does not involve artificial intelligence, but it is a discussion about robot trajectory planning and trajectory generation in the joint space and Cartesian space based on the robot kinematics and dynamics. Trajectory planning for a robot manipulator is a complicated problem, involving several aspects such as modeling obstacles, handling sensor information, considering robot dynamics, searching for collision-free paths and avoiding singular configurations of the robot [1]. A formal definition of trajectory planning can be found in Singh and Leu. Path planning is a closely related issue of trajectory planning [2]. Reynier etal. [3] discussed the optimal robot positioning problem among obstacles. Path planning at the kinematics level is very useful and can be further extended to include dynamic functions, since robot manipulators are essentially dynamic systems. Wang and Hamam [4] also developed an algorithm for solving the trajectory-planning problem with collision detection and avoidance. As the method of Cartesian space has many shortcomings, the way for the joint space is widely used. Commonly used method is to show the current teaching point, which takes some of the key points in the trajectory of the robot, to remember corresponding joint coordinates of these points, and then it does some interpolation between two key points. Clearly, in order to make the end of actuator along the required trajectory strictly, we must take enough points. In practical applications it often needs to operate the robot along the numerous, complex spatial trajectory. If each trajectory must be conducted, the workload will be very large. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 421–428. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

422

T. Zhang and S. Chen

Robot trajectory is generally circular or linear. This paper will use the method of interpolation to do path planning of mobile welding robot with a 5-DOF arm. And the welding process when the vehicle is walking across obstacles has been discussed.

2 Robot Kinematic Modeling 2.1 Mathematical Model Fig 1 shows the mobile welding robot system. The welding robot manipulators in the system have five rotational joints. Before establishing the mathematical model, we first construct the connecting rod of the coordinates properly in accordance with the Denavit-Hartenberg rules, and then we can derive robot kinematics equation. The D-H Parameter table is depicted in Table 1.

Fig. 1. The mobile welding robot system Table 1. D-H Parameter

i 1 2 3 4 5

αi-1

0 90° 0 0 -90°

ai-1 0 a1 a2 a3 0

di 0 0 0 0 0

θi θ1 θ2 θ3 θ4 θ5

di: The distance of two public vertical line along the axis of rod θi: Intersection angle of two public vertical line in the surface Perpendicular to the axis of rod i ai: The distance of two axis of rod along the public vertical line αi: Intersection angle of two axis of rod in the surface Perpendicular to ai The robot kinematics equation is: ⎡ r11 ⎢r 0 0 1 2 3 4 ⎢ 21 5T = 1T 2T 3T 4T 5T = ⎢ r31 ⎢ ⎣0

r12 r22 r32 0

r13 r23 r33 0

px ⎤ p y ⎥⎥ pz ⎥ ⎥ 1⎦

(1)

Path Planning and Computer Simulation of a Mobile Welding Robot

423

where: ⎧r11 = c1 c234 c5 − s1 s5 ⎪r = s c c + c s 1 234 5 1 5 ⎪ 21 ⎪r31 = s 234 c5 ⎪ ⎪r12 = −c1c234 s5 − s1c5 ⎪r = − s c s + c c 1 234 5 1 5 ⎪ 22 ⎪r32 = − s 234 s5 ⎨ ⎪r13 = −c1 s 234 ⎪r23 = − s1 s 234 ⎪ ⎪r33 = c234 ⎪p = c c a + c c a + c a 1 23 3 1 2 2 1 1 ⎪ x ⎪ p y = s1c23 a3 + s1c 2 a 2 + s1 a1 ⎪ ⎩ p z = s 23 a3 + s 2 a 2

(2)

2.2 The Method of Choosing Geometric Structure Geometric modeling is the basis for graphical simulation modeling. The accuracy and intuitiveness of modeling impact the reliability of the results and display directly. The robot is much rod linkage with comparatively complicated shape. To export robotic external form on the computer in artwork way, the robotic geometric actual form should be appropriated, and be translated into information which computer can receive. In this system, two physical factors have been used: cylinder and cuboids.

2.3 Inverse Kinematics Problem The robot inverse kinematics problem is to find the joint angle with known position and orientation of the robot. For a given robot, the solvability of the robot is whether its analytic form of inverse kinematics can be calculated. Through computing, concrete results can be obtained as follows: py ⎧ ⎪θ 1 = 180° ± φ = 180° ± arctan px ⎪ ⎪ c1 p x + s1 p y − a1 ⎪θ 2 = 180° − arctan pz ⎪ ⎪ 2 c p s p a ( ) + − + p z2 + a 22 − a 32 1 y 1 ⎪− arcsin 1 x ⎪ 2a 2 p z2 + (c1 p x + s1 p y − a1 ) 2 ⎪ ⎪ c1 p x + s1 p y − a1 ⎪θ 23 = 180° − arctan pz ⎨ ⎪ 2 2 2 2 ⎪ arcsin (c1 p x + s1 p y − a1 ) + p z + a 3 − a 2 ⎪− 2 2 2 a 3 p z + (c1 p x + s1 p y − a1 ) ⎪ ⎪(θ = θ − θ ) 23 2 ⎪ 3 c1c 23 r13 + s1 s 23 r23 + s 23 r33 ⎪ ⎪θ 4 = arctan c s r + s s r − c r 1 23 13 1 23 23 23 33 ⎪ ⎪ s1 r11 − c1 r21 θ arctan = ⎪ 5 s1 r12 − c1 r22 ⎩

(3)

424


We have obtained five values of joint angle which decide the position and orientation of robot. Once the position and orientation of robot have been known, the value of each joint angle can be obtained with the results.

3 Trajectory Planning Method Robot trajectory planning is to design the law of motion of each joint based on the task robots must complete.

， T ，…， T ,can be calculated by inverse kinematics after inputting each route counts P ，P ,…,P , In this article, corresponding transformation matrix

5 0 1

T

5 0 2

5 0 n

1

2

n

and then the value of each joint angle can be got. The shape of the robot is described on the computer. There are two kinds of trajectory planning programs: point-to-point motion (PTP) and continuous-path motion (CP).The cylindrical coordinate robot in this article, has used PTP and CP trajectory planning. The movement from point A to point B between which there is no middle point, which does not have been set requirements for the movement path, can be called PTP trajectory planning. In this article, we focus on the CP trajectory planning.

3.1 CP Trajectory Planning For continuous path movement, it must not only stipulate start and end points of the manipulator, but also specify several intermediate points between two points (path points).The movement must be along a specific path (path constraint). With ensuring the same attitude towards the welding, we simulate the entire robot arm, welding torch and the path by matlab. Here are the steps and simulation results of CP trajectory planning. Figure 2 shows with the welding torch's movement along the arc, the robot arm moves from the initial location to the final position. The trajectory planning ideas and processes are shown in Figure 3. Figure 4~6 shows projection of the robot arm in various planes, through the entire welding process.


425

Start simulation XOZ Planar Projection 150

Simulation time step

100

50

0

Start and end coordinates -50

-100

Middle point coordinates

-150

-200

Base coordinate transformation Calculating parameter equation of arc trajectory

0

50

100

150

200

250

300

350

400

450

500

Fig. 3. XOZ Planar Projection

XOY Planar Projection 80 70 60

Interpolation of arc trajectory position

50 40 30 20

Inverse transform of arc coordinates

10 0

0

50

100

150

200

250

300

350

400

450

500

Fig. 4. XOY Planar Projection

Solution by inverse

YOZ Planar Projection

Inverse solution i i i

150

100

50

Output each joint angle

0

-50

Draw simulation patterns

-100

-150

Output graphical simulation

-200

0

10

20

30

40

50

60

70

Fig. 5. YOZ Planar Projection

End simulation

Fig. 2. CP Trajectory Planning Flow Chart

80

426


200

0

-200

-400 100 80 60 40 20 0

0

100

200

300

500

400

600

700

(a)

(b)

Fig. 6. Trajectory planning ideas and processes

3.2 PTP Trajectory Planning With Obstacles When the vehicle meets obstacles, it needs to cross the obstacles. In order to ensure the continuity of welding when crossing the obstacles, the speed of the body and welding torch must be controlled. There are seven processes from the start to complete crossing the obstacles: front wheel lift, walk, front wheel down and middle wheel lift, walk, middle wheel down and rear wheel lift, walk, rear wheel lift. The process is shown in Figure 7. Abscissa represents unit of time, while vertical axis represents unit of distance when the body is moving forward. We assume that the ratio of the vehicle speed and the welding torch speed is 1:0.782. Figure 8 shows that with the welding torch's movement along the line, while the vehicle meets obstacles, the arm moves from the initial location to the final position. The twisted state of the robot arm can be seen from the figure. More simulation results show that, he greater the ratio of the vehicle speed and the welding torch speed, the distortion of the robot arm is more serious. It is likely to cause arm damage. Therefore, the right speed ratio must be select in the experiment. 80 70 60 50 40 30 20 10 0

0

200

400

600

800

1000

1200

Fig. 7. Seven Process of the Vehicle Movement When Crossing Obstacles


427

Fig. 8. The Change of Robot Arm's Shape When Welding Across Obstacles

4 Conclusions The trajectory planning method in this article has interpolated space position of the actuator of the welding torch. And then through kinematics and inverse kinematics, we calculate the change of 5 joint angles of the robot arm, and determine some changes in the arm shape. This method is more intuitive, which has a certain reference value for trajectory planning of robots. The robot trajectory planning with obstacles has a certain novelty. Acknowledgement. This work is supported by National 863 plan of China under Grant No. 2009AAA042221, and Shanghai Sciences & Technology Committee under Grant No. 09JC1407100, P.R. China.

References [1] Khoukhi, A., Hamam, Y.: Optimal time-energy trajectory planning for robot with hard constraints. Control-Theory and Advanced Technology 6, 417–439 (1990) [2] Singh, S.K., Leu, M.C.: Manipulator motion planning in the presence of obstacles and dynamic constraints. International Journal of Robotics Research 10, 171–187 (1991) [3] Reynier, F., Chedmail, P., Wenger, P.: Optimal positioning of robots: feasibility of continuous trajectories among obstacles. In: Proceedings of the 1992 IEEE International Conference on Systems, Man and Cybernetics, vol. 1, pp. 189–194 (1992) [4] Wang, D.M., Hamam, Y.: Optimal trajectory planning of manipulators with collision detection and avoidance. International Journal of Robotics Research 11, 460–468 (1992) [5] Craig, J.J.: Introduction to Robotics: Mechanics and Control, 3rd edn. [6] Liu, C.L., Zhang, K., Fu, Z., Cao, Q.X., Yin, Y.H.: Trajectory Planning and Simulation of Welding Robot for IVECO Automobile Crossbeam. Robot (October 2001) [7] Ren, J.Y., Sun, H.X.: A Novel Method of Trajectory Planning in Cartesian Space. Robot (May 2002) [8] Liu, X., Fang, H.-r.: The kinematics analysis and trajectory planning of serial cutting robot. Manufacturing Automation (March 2008) [9] Leng, D.Y., Chen, M.: Robot Trajectory Planning Using Simulation. Robotics & Computer-Integrated Manufacturing (1997)

428


[10] Hu, H., Brady, M., Probert, P.: Trajectory Planning and Optimal Tracking for an Industrial Mobile Robot. Mobile Robots (1993) [11] Su, J., Feng, C.: A Trajectory Planning Method For A Robot To Grasp A Moving Target. ROBOT (1994) [12] Yang, G.: The Study of Time Optimal Trajectory Planning for Robot Manipulators. China Mechanical Engineering (2002-20) [13] Wang, X.-l., Hou, Y.-b., Wang, T.: Research of Trajectory Planning of Industrial Robot Based on VC++. Industry and Mine Automation (2009)

The Control System Design of a Climbing Welding Robot Based on CAN Bus Xiaofei Gao1, Minghui Wu1, Z. Fu1, Yanzheng Zhao1, and Shanben Chen2 1

School of Mechanical Engineering, Shanghai Jiao Tong University, State Key Laboratory of Mechanical System and Vibration, Shanghai 200240, P.R. China 2 Welding Engineering Institute, Shanghai Jiao Tong University, Shanghai 200240, P.R. China

Abstract. A control system of a climbing welding robot based on CAN bus is proposed and designed in this paper. The robot is used for autonomous welding in the manufacturing and maintenance of large non-structure equipment. Under the working circumstance, it is suitable to establish the robot control system based on CAN bus. Compared with a traditional system, CAN bus has the advantage of high anti-interfering, high stability and high transfer rate. The 1st section is the introduction of the system; the 2nd section is a brief description of the robot’s structure. The 3rd section is the detailed design of the control system. The control system consists of three main parts: the establishment of a CAN bus, controlling of servo motors and receiving signals of sensors. The analyses of information, calculation of kinematic and overall control are all carried out in a PC. And the 4th section is the conclusion.

1 Introduction In the manufacturing and maintenance of large non-structure equipment such as ships, oil tanks and nuclear electricity equipment, welding works are often applied. Using robots to accomplish these works would greatly improve workers’ labor environment, working quality and efficiency. The designed welding robot has many modules such as motor control modules, sensor modules and others. Hence the data transfer and overall control of the robot are quite complicated. CAN or CAN bus (Controller–area network) is a vehicle bus standard designed to allow microcontrollers and devices to communicate with each other within a vehicle without a host computer. CAN is a message based protocol, designed specifically for automotive applications but now also used in other areas such as industrial automation. CAN is a multi-master broadcast serial bus. The devices that are connected by a CAN network are typically sensors, actuators, and other control devices. These devices are not connected directly to the bus, but through a host processor and a CAN controller. The transfer rate can reach 1Mbps under the distance of 40m. Reducing bit rate can increasing the transfer distance [1]. CAN bus has two signal wires CAN-high and CAN–low. This bus structure allows easy extension for extra nodes. New nodes wouldn’t affect existing nodes and single node’s error won’t affect others neither. T.-J. Tarn et al. (Eds): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 429–434. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

430

X. Gao et al.

Traditional servo system control structure is made up with the upper computer, motion control card and servo-driver. Output signal from the motion control card will control the motor’s operation after driven by servo-driver [2]. Compared with traditional system, CAN bus has the advantage of high anti-interfering, high stability and high transfer rate, and also it has a simpler structure. Considering the advantage of CAN bus, it’s very suitable to establish the control system of a multimotor climbing welding robot based on CAN [3]. CAN bus have been used in some of industry robots and other robots already. Due to the high stability, some military robots also use CAN [4].CAN bus is very suitable for mobile robots [5,6] because it only requires two wires for data transfer. CAN bus also provide the good expandability for sensors and other devices [7].

2 Climbing Welding Robot Structure The climbing welding robot uses the adhesion method of permanent magnets. The magnets exert force on ferromagnetic walls and the adhesion force produces the fraction between the wheel and the wall. Hence the robot can climb on vertical walls. The mechanical structure of the climbing welding robot mainly consists of: mobile mechanism, magnetic adhesion mechanism, elevation mechanism, robot body and robot arm. The magnets are fixed on the frame between wheels, each group of wheels and magnets can be elevated by screw. On the same time, the air gap between magnets and wall surface can be adjusted to change the adhesion force. The robot arm has a DOF of 5 to control the torch.

Fig. 1. Robot structure 3D model

To accomplish autonomous welding, we should obtain enough environment information. So we adopt sensors and CCD cameras to collect information of the robots and the obstacles.

3 Control System Design The welding robot (Fig2) is used for autonomous welding in the manufacturing and maintenance of large non-structure equipment. Compared with traditional

The Control System Design of a Climbing Welding Robot Based on CAN Bus

431

Fig. 2. Welding robot prototype (without sensors)

system, CAN bus has the advantage of high anti-interfering, high stability and high transfer rate. So the control system of the robot is mainly based on CAN. Fig3 is a schematic of our control system. The whole control system mainly has three parts: the establishing of CAN bus, controlling of servo motors and receiving signals of sensors. The analyses of information, calculation of kinematic and overall control are all proceeding in the PC.

Fig. 3. Control System Structure

3.1 Hardware The hardware of the control system consists of three main parts:PC, motors with servo controller, sensors and CCD cameras. CAN bus is adopted to accomplish data transfer between each part. We adopt IXATT PC/CAN interface USB-to-CAN compact to connect PC and the CAN bus. The CAN-controller of the USB-to-CAN compact is Philips SJA100 and connects to the PC through USB2.0. The servo controller is Accelnet Micro Panel (Copley Control Corp.) The CANopen distributed control architecture is supported by the panel. It can work as a CAN node operating under the CANopen protocol. The sensors fixed on robot are used to conduct preliminary perception and positioning of obstacles in welding environment and assist the robot avoiding or surmounting these obstacles. The sensors we used include ultrasonic sensors and

432

X. Gao et al.

infrared sensors. The infrared sensors include infrared measuring sensor and infrared approaching switch. The signal of the sensors is collected by the dsPIC30F chip (Microchip Corp.) and transferred to the PC by CAN bus.

3.2 CAN Network CAN bus is a multi-master serial bus and any node can be set as the main node [8]. In our control system, we set the node 1 (the IXATT USB-to-CAN compact) which connected with the PC as the main node and other as the bottom node. The PC communicates with the CPU in USB-to-CAN compact through USB port, and the CPU communicate with the CAN transceiver to accomplish data’s sending and receiving on CAN bus. The bottom nodes have different function while connected with CAN bus at the same time and controlled by the PC and main node (shown in Fig4).

Fig. 4. Sketch of control system structure based on can

According to the information amount and welding work radius we set the CAN bit rate at 250Kbps (or 500Kbps if welding work need higher precision, but the working radius is limited to 100m). After experiment, we find robot’s performance is stable and reliable under this bit rate.

3.3 Robot Motion Control The robot motion mechanism can be divided into three obstacle-passing moving mechanisms and a 5-DOF robot arm. Single obstacle-passing moving mechanism has three servo motors, two for wheels and one for elevation. The 5-DOF arm has 5 motors, one for each joint. So the whole robot has 14 DC servo motors. We use Accelent Micro Panel servo controller (produced by Copley Control Corp) to control the motors. Each controller controls one motor and collects the pulse signal of encoder. Each controller has 12 GPIO (General Purpose Input Output) which can be connected to position limit photoelectric switches or motor brakes or other devices to conduct some extra logical control if needed. The control loop of single servo motor is shown in Fig 5.

The Control System Design of a Climbing Welding Robot Based on CAN Bus

433

Fig. 5. Control loop of a single servo motor

The inverse dynamic of the motion mechanism is calculated by the PC. PC receives the environment information from the CCD cameras and sensors to calculate target welding joint’s position information. PC also receives the motors’ information through servo controllers to calculate current situation of the robot. Combining both the robot’s and welding joint’s information the PC can derive the motion trace for each motor to accomplish the welding work and the send control commands to servo controllers through CAN bus.

3.4 Multi-sensor Network Design The sensors fixed on robot are used to conduct preliminary perception and positioning of obstacles in welding environment and assist the robot avoiding or surmounting these obstacles. The sensors we used include ultrasonic sensors and infrared sensors. The infrared sensors include infrared measuring sensor and infrared approaching switch. The ultrasonic sensors are used to exert detection of obstacles at long distance. The sensors we select are the 600 series sensor (Beijing Northking Co.) and URM37V3.1 ultrasonic module. The 600 series sensors have a range from 15cm to 10m which is used for preliminary detection; URM37 module has a range from 4cm to 5m and is used for nearer detection. In addition, URM37 can be connected to a steering motor and exert scanning detection. The ultrasonic sensors have significant errors when detecting close objects, so we use infrared sensors additionally. The type is GP2D12 produced by SHARP. We use Microchip’s dsPIC30F5015 chip to collect the sensors’ signals. The chip can perform as a CAN node and connect to PC through CAN bus. The chip collects sensors information cyclically. After collecting sensors feedback information, it calculates and processes the data preliminarily and transfers the results to PC through CAN bus. The PC further analyzes and processes the data and controls the robot’s motion. A sensor module is shown in Fig 6.

434

X. Gao et al.

Fig. 6. Sensor module

4 Conclusions The designed control system based on CAN bus can guarantee the information transfer among different modules of the robot and the coordination between information collection and motion control. Under this control system, the robot can accomplish simple autonomous welding work. The CAN bus also provides a good extensibility of the system. New function modules can be added to the system through CAN bus easily. Acknowledgement. The support of the national High Technology Research and Development Program of China, through the Ministry of Science and Technology of China Grant No. 2009AA04Z221, is gratefully acknowledged.

References [1] Liu, D., Deng, J., Jia, P., He, T.: Pile Robot System Based On CAN Bus Control Servo. Manufacturing Technology & Machine Tool 5, 136–137 (2009) [2] Cheng, X., Zhou, Y., Yang, L.: The Research on Networked Servo Control System Based on CAN Bus. In: 2010 Second International Workshop on Education Technology and Computer Science (2010) [3] Zhou, Y., Wang, Z., Li, J.: Design of the Control System of Tightening Machine Based on CAN Bus. In: 2009 International Conference on Intelligent Human-Machine Systems and Cybernetics (2009) [4] Fang, W., Kong, F.: Control System of EOD Robot Based on CAN. Field Bus Control Technology 32(2), 52–53 (2010) [5] Wu, X., Li, B., Yang, T.: Design of CAN Bus System for Robot Car. Journal Of Wut (Information & Management Engineering 28(6), 34–36 (2006) [6] Yao, Z., Dai, X.: Application of CAN- Bus for mobile robots. Journal of Qiqihar University 25(6), 16–19 (2009) [7] Chen, X., Xu, H., Zhang, X., Wang, L., Yang, C.: Design of mobile robot sensor network based on CAN bus. Journal of China University of Metrology 20(3) (September 2009) [8] Yang, K., Li, S., Lu, G., Tian, H.: Design of Control System of Wheel - Leg Robot based on CAN. Small & Special Electrical Machines 4, 28–30 (2009)

An Implementation of Seamless Human-Robot Interaction for Pipeline Welding Telerobotics Na Dong, Haichao Li, Hongming Gao, and Lin Wu State Key Laboratory of Advanced Welding Production Technology, Harbin Institute of Technology, Harbin, China e-mail: [email protected]

Abstract. Although welding is an important technology for nuclear pipeline maintenance, it cannot be performed effectively up to date because of the robot can not be fully autonomous. This paper presents a human-robot cooperation telerobotics system for pipeline welding in a nuclear environment. The implementation of humanrobot welding systems can be extremely challenging when the robot is not directly controlled by the human. Depending on the task, the interaction mode is composed of continuous manual, semi-autonomous or autonomous. To address the Human-Robot Interaction (HRI) issues in such a system, a concept of seamless HRI is introduced. The main idea seamless HRI is to design a telerobotics welding system that allows a shift from manual to autonomous operation, dynamically, via different human-robot roles and relationships. These roles are Master-Slave, Supervisor-Subordinate, PeerPeer, Teacher-Learner and Full Autonomous mode by the robot. The theoretical foundations for seamless HRI are introduced. An implementation of the concept is presented and experimental results show the system get through the pipeline welding.

1 Introduction Telerobotic system plays an important role in repairing the nuclear pipe [1], because the nuclear radiation pollution caused by the chemical substances makes the maintenances by human directly difficult. Although, autonomous robots that have the anthropomorphic dimensions, mimic human-like behaviors, and include human-like reasoning are known as humanoid robots and work in this area has been ongoing for over a decade[2-6], the intelligence of robot is not so perfect to replace the human. So, the human-robot interaction is an effective way to meet the remote task. HRI is a field of study dedicated to understanding, designing, and evaluating robotic systems for use by or with humans. In telerobotic welding system, the operator in safety environment manipulates the telerobot which is in dangerous environment to execute the welding task[7]. Human can play different roles in the possible relationships among the human, the robot, and the task. There are five roles, show as Fig.1.Supervisory is that the human monitors the robot and changes the plan when necessary; Operator is that the human directly interacts with the robot to change its behaviors; Mechanic is that the human physically intervenes with the robot to change its capabilities through modifying hardware or software; Peer is that the human works with the T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 435–442. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

436

N. Dong et al.

Fig. 1. The relationship between human and robot in teleoperation system

robot to perform a task, Usually a peer interaction between human and robot occurs at high level of behavior; Bystander is that the human is not directly involved in controlling the robot, but effect how the robot accomplishes the task [8].For example, humans in the same building can effect how a robot navigates the environment by blocking the robot or opening or closing a door. The key is how to envolute the teleoperation mode orienting to the pipeline remote welding. In this paper, a new and innovative human-machine interaction method is required to envoluate the operating mode, plan the pipeling maintenance task and accomplish the remote pipeline welding.

2 Task Planning Fig. 2 shows the pipeline remote welding process which is a new pipe replaced the damage one. A new pipe and an original pipe are assembled using welding processes. The maintenance process contains the detection, cutting, assembling and welding. First, a crack is checked out, then the cutting range is determined and the new pipe length is selected. Second, the spoiled pipe is cut down and the new one is placed to the waiting position gripped by the micro robot. Third, the first weld is done and the macro manipulator takes the micro robot away from the pipe. Fourth, the micro robot is assembled on the pipe again and the second pipe groove must then be welded.

Fig. 2. Scheme of nuclear environment pipeline maintenance

Seamless Human-Robot Interaction for Pipeline Welding Telerobotics

437

Task planning will give the basic method of human-robot interaction. For the task T , the execute process is h = {α1 , α 2 ,..., α n } . α i is one action of h .The anticipation aim of the task planning is g pre and the task characteristic aggregate is characterT analyzing by Analyse(T ) .The environment is KnowledgeE observed by Observe( E ) .The process is

Analyse(T ) → characterT ∧ Observe( E ) → KnowledgeE

(1)

Analyzing the task characteristic characterT and the environment KnowledgeE by

Analyse(characterT , KnowledgeE ) , the solving aggregate is N = { A1 , A2 ,..., Am } . Denoting by Think ( h, g pre ) , the only way to solve the problem is the action

h .And, Think Ai (h, g pre ) , the solving aggregate of Ai will achieve the anticipation task by cooperation. For the complicated task TS , according to the breaking down methods of Decompose(TS ) → {TS1 , TS2 ,..., TSn } , the subtask is

{TS1 , TS2 ,..., TSn } . As to any subtask Ti , the corresponding solving is got by matching TS1 6 A1 , TS2 6 A2 , … , TSn 6 Am . After the changing of working environment, task planning will carry out the action of Modify ( KnowledgeE , KnowledgeE ' ) → KnowledgeEn by

Analyse(characterT , KnowledgeE ) , and then, KnowledgeEn adjusts the task analysis. There are two situations. One is the changing is quite big, anther is the changing is bot big. For the first one, new task will generate, and, for the second, the original task will be modefied by Adjust (TS ) .The new subtask aggregate is

{TS'1 , TS'2 ,..., TS'n } , and the new matching T ' 6 A' S1

1

，T

'

S2

6 A2'

，…， T

'

Sn

6 Am' .

As to the pipeline remote welding task, the subtask is got by this method. For normal environment, the subtask is

：Starting the pipeline remote welding system; ：Virtual environment modelling of nuclear pipeline; T ：Telerobot Moving to the top of the all position welding device; T ：Telerobot joining with the all position welding device; T ：Telerobot breaking the all position welding device away from tool shaft; T ：The all position welding device clamping the new pipe; T ：The welding device getting to the top of pipe assembling; T1

T2 3

4

5

6 7

438

N. Dong et al.

：Telerobot assembling the new pipe; ：Adjusting the micro device, centered the welding torch and the groove; T ：The first all position weld; T ：Telerobot breaking the all position welding device away from pipeline; T ： Telerobot taking the all position welding device assembling with the T8 T9

10 11 12

pipeline again;

T13 T14 T15 T16 T17 T18 T19

：Adjusting the micro device, centered the welding torch and the groove; ：The second all position weld; ：Telerobot breaking the all position welding device away from pipeline; ：Telerobot taking the all position welding device near the tool shaft; ：Telerobot taking the all position welding device back to the tool shaft; ：Telerobot breaking away from the all position welding device; ：Telerobot back to the basic position.

From the task planning, we get the subtask of pipeline remote welding. And, according to the subtask, corresponding teleoperation method is planning. It is Master-Slave, Supervisor-Subordinate, Peer-Peer, Teacher-Learner and Full Autonomous mode, etc. This paper will not introduce the detail of the control method because that it is the mature teleoperation control mode.

3 Seamless HRI Algorithm The seamless HRI Algorithm focuses on the contribution of different teleoperation control mode. The algorithm takes the fuzzy aggregate to solve the problem. Define the fuzzy aggregate as: Settings the domain of discourse is U , then, any mapping μ A of U to closed interval makes sure a fuzzy subset A

μ A : U → [ 0,1]

(2)

μ → μ A (μ )

(3)

Where, μ A is the membership function of fuzzy subset and μ A ( μ ) is the degree of membership about μ to A .Then, the fuzzy aggregate is

A=

A( μ1 )

μ1

+

A( μ2 )

μ2

+ ... +

A( μn )

μn

(4)

Seamless Human-Robot Interaction for Pipeline Welding Telerobotics A( μi )

And, μi

439

is the corresponding relationship between the element μi of domain of

discourse and the degree of membership A( μi ) of μi .The fuzzy subset conforms to human intelligence because of indefinite borderline. It is the important method to achieve the human robot interaction. The fuzzy human language should be stored and computed by translating to measurable information. Mood operator is used to quantify the human language. The definition is

( H λ A)( μ ) [ A( μ ) ]

λ

(5)

H λ is mood operator. When λ > 1 , it is centralization operator and it enhances

the mood of affirmation degree. When λ < 1 , it is desultoriness operator and it weaken the mood. Using in evaluate the contribution, the algorithm is: Set Yi = ⎡⎣ηi1 ,ηi2 ,...,ηir ⎤⎦ j

is the contribution degree of subset Ai .Where,

j ≤ r ) is contribution capability of Ai in the j dimensional. First, setting a evaluating aggregate perfect, good, normal, not good bad , and then, the evaluating weight is denoting by the matrix of V = [ v1 , v2 , v3 , v4 , v5 ] . Third, ηi j = d i

bi j ,

(1 ≤

λ

｛

，｝

setting up the judging matrix Ei according to the experience of operator and the expression of the working history.

Her ⎡ e11 ⎢e ability 2 : Ei = ⎢ 21 ⎢# # ⎢ abilityr : ⎣ er1 ability1:

H e12 e22

M e13 e23

L e14 e24

er 2

er 3

er 4

Ler e15 ⎤ e25 ⎥⎥ ⎥ ⎥ er 5 ⎦

(6)

The contribution degree of Ai is computed by Ei and V , which is

Yi = Ei × V T .Then, the ability vector of

Ai

is got by Yi , which is

Di = ⎡⎣ d , d ," , d ⎤⎦ , di j = bi j × ηi j , (1 ≤ j ≤ r ) .The weight matrix denotes by 1 i

2 i

r i

W = [ω1 , ω2 ," , ωr ] . And, the whole ability contribution degree of Ai is Di′ = Di × W T . Ai is evaluated by u ( Ai ) =

Di′ V (N ) ∑ Di′

Ai ∈N

(7)

440

N. Dong et al.

For the master-slave teleoperation controlling, we used the algorithm to evaluate the master Am . There are three kinds of Am , which are teleoperation handle F, teaching box M and space mouse K. The cost of master contribution is computed by the algorithm. The results provide the foundation of the master device choice. Setting the master-slave teleoperation is S = { Ar , Av , Am } , and the utilization is

V ( s ) = 4 .The ability vector is Am =< 3,3,3 > , and then, the judging matrix Em is Her

H

M

L

Ler

F: ⎡ 0.6 0.2 0.2 0.0 0.0⎤ M : Ei = ⎢⎢ 0.1 0.4 0.2 0.3 0.0 ⎥⎥ ⎢⎣ 0.0 0.0 0.2 0.2 0.6 ⎥⎦ K: Setting

VM = [1,0.75, 0.5,0.25, 0]

,

and

the

contribution

(8)

degree

is

E1 × V = [ 0.850 0.575 0.600] . The dimension ability of Am is T

T

F : d m1 = bm1 × ηm1 = 3 × 0.850 = 2.55

(9)

M : d m2 = bm2 × η m2 = 3 × 0.575 = 1.725

(10)

K : d m3 = bm3 × η m3 = 3 × 0.600 = 1.8

(11)

According to the cost, the highest one will be chosen to be the master automatically. The others will be the standby in the master-slave teleoperation. For other kinds of choice, the similar process is carried out.

4 Experiment and Results The experiment was carried out to weld the pipeline according to the remote pipeline maintenance plan. The whole working process is show as Fig.3. By the human-robot interaction, master-slave, Supervisor-Subordinate, Peer-Peer, and Teacher-Learner are evaluated by the seamless changing algorithm. The humanrobot interface provide the evaluate dialog of human. The information of executing is presented to human in the interface. At last, the remote pipeline welding is accomplished using the human-robot interaction method.

Seamless Human-Robot Interaction for Pipeline Welding Telerobotics

1

441

2

3

5

7

6

8

Fig. 3. Process of remote pipeline welding

5 Conclusions An innovative HRI telerobotic welding system is developed for nuclear pipeline maintenance. The interaction between the human and the robot is concerned about the task planning and contribution evolution. According the evolution, the teleoperation controlling mode is chosen automatically. The algorithm of human– robot changing is given out. Finally, pipeline welding experiments are performed, and the results demonstrate that the HRI system performing and the installing process can be performed effectively.

442

N. Dong et al.

Acknowledgement. This work was supported by contract (NCET-07-0233) from New Century Excellent Talents Support Program and by the National Natural Science Foundation of China (NSFC) under the Grant number 50905043. The authors also greatly appreciate the editor and the referees for their thoughtful comments which led to a substantial improvement on the presentation of this paper.

References [1] Pegman, G.: Robotics and the nuclear industry. Industry Robot 33, 160–161 (2006) [2] Ambrose, R.O., Aldridge, H., Askew, R.S., et al.: Robonaut: NASA’s space humanoid. IEEE Intelligent Systems and Thrie Applications 15, 57–63 (2000) [3] Breazeal, C.: Emotion and Social humanoid robots. International Journal of HumanComputer Studies 59, 119–155 (2003) [4] Shimada, M., Minato, T., Itakur, S., Ishiguro, H.: Development of androids for studying human-robot interaction. In: Proceeding of the 6th IEEE-RAS International Conference on Humanoid Robots (2006) [5] Breazeal, C.: Toward sociable robots. Robotics and Autonomous Systems 42, 167–175 (2003) [6] Chen, J.R.: Constructing task-level assembly stratrgies in robot programming by demonstration. International Journal of Robotics Research 24, 1073–1085 (2005) [7] Billard, A., Dautenhahn, K.: Experiments in social robotics: Grounding and use of communication in automous agents. Adaptive Behavior 7, 415–438 (1999) [8] Wang, J.: Human Control of Cooperating Robots, pp. 2–10. University of Pittsburgh (2007)

Design and Experiment of a Novel Portable All-Position Welding Robot Bin Du, Jing Zhao, and Yu Liu College of Mechanical Engineering and Applied Electronics Technology, Beijing University of Technology, Beijing 100124, P.R. China e-mail: [email protected]

Abstract. This research develops a portable all-position welding robot for the welding of intersected pipes. A complete procedure is adopted to conduct the design. The task and motion of the robot are analyzed. Based on that, three representative types of the robot are chosen in the type synthesis of the mechanism. Two new indices proposed to evaluate required properties of the robot, along with the traditional dexterity index, are chosen to be the criteria in the dimension synthesis of the mechanism. Through the optimization by genetic algorithm, the best robot in type and dimension is determined after comparison on their performances. Finally, the prototype is developed. The welding experiment result shows that the welding robot works stably, and the quality of weld seam is acceptable simultaneously. It has great practical value.

1 Introduction Since 1980s, with the fast development of both the robotic technology and welding technology, the use of robots in welding industry has grown greatly, and its application has been seen as a typical example of the industrial automation. All-position welding robot is a robotic system that can accomplish 360-degrees welding task entirely by its own, without the help of other auxiliaries such as positioning machines. Comparing to the traditional ones, all-position welding robot is more convenient for operating and easier to be miniaturized. It has become one of the new frontier subjects in welding automations. One of the main applications for all-position welding robots is the welding of pipes intersecting, as showed in Fig.1. In the welding of large boilers and flange plates, the weld seam is a saddle-like curve formed by two cylinders. It is a kind of space curve, which is far more complicated than lines or circles and require constant adjustments of the pose of the welding-gun. And also for the reason of the complex and harsh working environment, it is nearly impossible to employ positioning machines to help the robot to complete 360-degrees welding. Actually, in practices, manual welding is usually adopted by workers in such cases, which is proved to be low efficient, high labor intensive and especially difficult to guarantee welding quality. In recent years, the application of gas shielded arc welding in all-position welding technique has matured and laid a foundation for the further research of the all-position welding robots [1, 2]. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 443–450. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

444

B. Du, J. Zhao, and Y. Liu

Some researchers have designed a right-angle cylindrical coordinate robot to implement the all position welding [3]. This robot has 5 DOFs and uses two prismatic joints to locate the position of the wrist origin. It has a simple structure to achieve easy control and high accuracy. But all of them didn’t consider the requirements of compactness and the interference issue. In this paper, the design of an all position robot will be comprehensively addressed and comparisons between different types of robots will be conducted based on indices. Apply optimal design on the drive layout of the welding robot for its center of gravity is as much as possible in the axis of branch pipe. In this way the welding robot working stability could be improved greatly. Eventually, the welding experiment result shows that the welding robot could be anchored firmly on the branch pipe, and revolving axis of the waist joint agrees well with that of branch pipe. The welding robot works stably, and the quality of weld seam is acceptable.

Table 1. Requirements and motion analysis Requirements of the welding robot Dexterity

Compactness

No interference

Motion analysis

Fig. 1. Intersected pipes

Swivel

Position

Orientation

1-DOF

2-DOF

2-DOF

2 Preliminaries 2.1 Task Analysis The welding part is made of two pipes with different radius, as illustrated in Fig.1. The weld seam is a saddle-like space curve formed by two cylinders. It is obvious that at different point of the curve, the robot must have different figure to enable its end-effector to point at its aim with good accuracy and pose. So the robot must have a good dexterity during the whole welding process. In real practice, the working condition is usually a big pipe inserted with many different sizes of small pipes, and all the welding is conducted simultaneously. So the robot must have a small size to avoid interferences with each other during the welding process. And also because the robots are often carried by welders and installed on the top pipes by fixtures, they must be portable and compact. Over all, the welding robot must meet three requirements shown in Table.1.

2.2 Motion Analysis The motion of the robot during this welding process can be divided into three categories. The first kind of motions is the swivel around the central axis of the top

Design and Experiment of a Novel Portable All-Position Welding Robot

445

pipe. The robot is usually installed on the top pipe and concentric with its axis. As an obvious result, this motion only requires 1-DOF. The second kind of motions is the positioning of the end-effector. Because the concentric arrangement in the first motion, the location of the end-effector is actually within a plane which is rotated around the central axis of the top pipe. This motion requires 2-DOF.The third kind of motions is the adjustment of the pose of the end-effector. To determine the orientation of a vector in space, it must take 3-DOF. But for this task, because the presence of a symmetry axis, it requires only 2-DOF for the orientation. In all, the motion of the robot requires 5-DOF, as shown in Table. 1.

2.3 Construction of Performance Index Definition 1. (Compactness Index c) Index c is constructed to evaluate the compactness of the robot:

c=

Vr Vp

(1)

Where V p is the volume of the small pipe of which the bottom is sealed by lowest point of the intersected curve, as shown in Fig.2. V r is the volume of the body enveloped by all outlines of the robotic configurations at different moment of the welding process, as shown in Fig.3. The value of index c is always bigger than 1. The less the index c is, the better the compactness of the robot is. Definition 2. (Interference Index i) Index i is constructed to evaluate the robot’s ability to avoid interferences:

i=

Lw Lt

(2)

Lt is the maximum distance between the wrist joint and the central axis of the small pipe in welding process, which is uniquely determined only by task. L w

where

is the maximum distance between the outmost joint and the central axis of the small pipe in welding process, as shown in Fig.4. The value of index i is always bigger or equal to 1.The less the index i is, the better the ability of the robot has to avoid interference. Definition 3. (Dexterity Index ρ ) A lot of former researches have been done and a lot of indices have been constructed to evaluate the dexterity of robot [4]. Index ρ is constructed to evaluate the dexterity of robot. The condition number ρ [5]:

ρ=

σ1 σr

(3)

446


It is worth noting that the condition number is dimensionless only when the Jacobian represents single position mapping or single orientation mapping. However, when both the position and the orientation of the end-effector are included in the kinematic equations, it can be readily seen that the condition number of the Jacobian matrix will not be invariant under a scaling of the dimensions of the robot [6]. As for this welding robot, its Jacobian matrix J = R5×5 is:

⎡J ⎤ J =⎢ v⎥ ⎣ Jr ⎦ where

(4)

J v is the mapping relation between the positional velocity of end-effector

and the joint velocity, and J r is the mapping relation between the orientation velocity of end-effector and the joint velocity. Some methods have been developed to solve this problem including new ways to construct Jacobian mapping and adapted indices. Here, to simplify the computation, the new condition number is obtained as the mean value of the condition numbers of J v and J r , which both are scale-independent:

w = 0.5( ρv + ρr )

(5)

In computation and programming, all the condition number at each point on the trajectory will be added together, and the average value is taken to represent the general dexterity of the robot during the whole welding process. Definition 4. (Application Range of Tasks and the Total Index p) For a specific application range of tasks, the robot should have a generally good performance over those indices. To take adaptability into consideration, we define the total index p as follow. The total index p is the average of the sum of the modified indices: n 1 1 1 p = (1 / n) ∑ ( f1 + f 2 + f 3 ) t =1 c i w

where n represents the number of possible radius combinations,

(6)

f1 , f 2 , f3 are

3

weighting factors and

∑ fi = 1.

i =1

In Eq.(6), reciprocals of indices are added together. There are two reasons for this modification. First, after taken reciprocal, all the values of the three indices are normalized into the range of 0 and 1, where 1 is the best and 0 is the worst. Second, in computation and programming, if a given figure of the robot cannot finish the task or interferes with pipes it is working on, the value of its index can be taken as 0, which is the worst value.


Fig. 2. Definition of V p

Fig. 3. Definition of V r

447

Fig. 4. Definition of index i

3 Type Synthesis of Mechanism With the result in Table.1, the welding tasks performed by robot can be divided into three categories of motions. Correspondingly, the structure of the robot can also be categorized as three different parts. The first part of structure is a rotational joint that can accomplish the 1-DOF rotational motion. The second part is a two joints serial structure that is adopted to complete the 2-DOF positioning of wrist origin. The third part is a wrist structure, containing two rotational-joints, and can be used to realize the 2-DOF pose adjusting. The result is shown in Table.2. The first part and the third part structures can be uniquely determined. However the second part, with combinations of different joint-types and different offset arrangements, has many different possible structures. With different joint-types, there are rotational-rotational, prismatic-prismatic and prismatic-rotational three different combinations. And within each combination, there are different possible structures with different offset arrangements. Table 2. Structures for different motions 1-DOF Rotation

2-DOf Positioning

2-DOF pose adjusting

Taking the prismatic-rotational combination as example, there are several different structures as shown in Fig.5.Some of those structures have innate disadvantages. Like the first two structures in Fig.5, it’s very easy for them to have interferences with the edge of the small pipe that is beneath them. After a round of brief examination, three structures are chosen as the final candidates. They are shown in Fig.6. The first is a R-R type all rotating joint robot, which is very common in industrial application. The second is a P-P type cylindrical coordinated robot with fixed right-angle structure. The third is a P-R type robot that is the combination of the first two types.

448


Fig. 6. The final three structures

Fig. 5. Different structures for P-R type robot

4 Dimension Synthesis of Mechanism Size synthesis of Mechanism is a procedure to determine the geometric and kinematic parameters of the robot on the basis of type synthesis. It is a process of optimization with the objective to maximize the total index p. After experiments, the weighting factors are chosen as f1 = 1 / 3, f 2 = 1 / 3, f 3 = 1 / 3 , to make the three indices have same impact on the final total index p. The height of the robot h and the length of the welding tool t are fixed as h=80, t=60. The big pipe’s length is 550mm and radius is 135mm, the small pipes’ length is 130mm and radius is 50mm, 60mm, 75mm respectively. A continuous genetic algorithm (GA) is adopted for this optimization, for the reason that GA doesn’t require analytical expression and derivative information of the cost function. The initial population is randomly selected with the size of 20. The simulation of the welding process implemented by the each optimal joint type is shown in Fig.7, 8, 9 respectively. The result of the optimization is shown in Table.3. Table 3. Result of optimization Robot joint type

Optimization variables

R-R P-P R-P

a2

d4

a2 , d 4

105

185

—

—

—

d4

—

188

Fig. 7. R-R robot

Result p c

w

i

5

3.33

1

0.39

6.67

33.33

1

0.242

8.47

12.5

1.894

0.49

Fig. 8. R-R robot

Fig. 9. R-R robot


449

From the synthesis result in Table.3, the comparisons of those three types of robots are made. The R-R type robot has a better performance on all the four indices than other two types. The P-R type robot is the worst one, which combines all the disadvantages of the other two. The P-P type robot has a medium performance on indices but, with its special structure, is more easily to be fabricated and controlled.

5 Motor Layout Optimization Design In the design process, make the center of gravity of welding robot as much as possible in the axis of branch pipe to ensure that the welding robot works smoothly, achieves high accuracy. The welding robot will be held well during the process of fixing even if for different branch pipes on the base. The motor and welding gun of it are bilateral symmetry layout. Shoulder joint, elbow joint and wrist joint are driven by band. On the assumption that the coordinate origin is the intersection of two pipe axes, and the center of gravity coordinate of the welding robot is (-0.24, 337.27, -3.01) in this reference frame, and which is only 3.02mm away from the axis of the branch pipe. As shown in Figure 10, it satisfies the design needs basically. The waist joint rotates when the welding robot works. The output shaft of the waist joints motor is downwards, and is fixed to the base by the pin. It outputs joint torque through the rotating cylinder shell of the motor. This can effectively prevent the cable from winding during the waist joint rotating.

Fig. 10. Wrist drive layout

Fig. 11. Virtual prototype

Fig. 12. Prototype of welding robot

6 Virtual Prototype Simulations and Experiment Based on the analysis above, we employ ADAMS to analyze the kinematics and dynamics of the virtual robot system under the consideration of the effect of gravity, as shown in Figure 11.The joint angle within a cycle is configured respectively in ADAMS, and the simulation time is 120 seconds. Then measure joint torque respectively in ADAMS. Subsequently we choose motors according to the joint torques respectively. The constructional robot by the design above is shown in Figure

450


12.We use this welding robot to weld the pipe intersection which is a saddle-like curve formed by two cylinders, where the thickness of central pipe and branch pipe is both 10mm, and diameter of them is 220mm and 120mm respectively. Eventually, the welding experiment result shows that the welding robot could be anchored firmly on the branch pipe, and revolving axis of the waist joint agrees well with that of branch pipe. The welding robot work stably, and the quality of weld seam is acceptable.

7 Conclusions A new portable all-position welding robot for the intersecting line is described in this paper. Unlike the available all-position welding robot, those are only applicable in the spacious working environment. The proposed welding robot can be used in the compact working environment. A feature of the paper is that the whole design process is conducted by the modern design methods, which consist in task and motion analysis, two new indices construction to evaluate requirement property of the robot, type synthesis, dimension synthesis, prototype simulation and experiment. The welding experiment result shows that the welding robot has a great value in use. Acknowledgement. The supports of this research by Scientific Research Key Program of Beijing Municipal Commission of Education (KZ200910005003) and National Natural Science Foundation of China (No. 50775002) are greatly appreciated.

References [1] Herty, M., Seaid, M.: Simulation of transient gas flow at pipe-to-pipe intersections. International Journal for Numerical Methods in Fluids 56(5), 485–506 (2008) [2] Yan, B., Yan, G.: Design of weld inspection system for intersected pipe based on redundant manipulator. Optics and Precision Engineering 12(4), 420–425 (2004) [3] Li, X., Wang, S.: Kinematic analysis and simulation of saddle-back coping welding robot. Journal of Beijing University of Aeronautics and Astronautics 34(8), 964–968 (2008) [4] Zhao, J., Li, Q.: On the joint velocity jump for redundant robots in the presence of locked-joint failures. Journal of Mechanical Design 130(10), 102305-1–102305-7 (2008) [5] Salisbury, J.K., Craig, J.J.: Articulated Hands: Force Control and Kinematic Issues. The International Journal of Robotics Research 1(1), 4–17 (1982) [6] Gosselin, C.M.: Dexterity indices for planar and spatial robotic manipulators. In: Proceedings 1990 IEEE International Conference on Robotics and Automation (Cat. No.90CH2876-1), May 13-18. IEEE Comput. Soc. Press, Los Alamitos (1990)

The Power and Propulsion of Medical Microrobots Xueqin Lv1, Rongfu Qiu1, Gang Liu1, and Yixiong Wu2 1

Department of Information and Control Engineering, Shanghai University of Electric Power, Shanghai 200090, China e-mail: [email protected] 2 School of Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai 200240, China

Abstract. This paper provides a detailed insight into the present-day state of medical microrobot technology and development on power and propulsion. Firstly,some of the critical aspects of medical microrobots design are considered and then the characteristic and application of the external field and other sources driving technology used in microrobots are introduced; finally some correlative technology problems and perspective in future are analyzed. This paper shows that medical microrobot developments are at an early stage and involve a disparate family of technologies and disciplines, such as MEMS, nanotechnology, biomimetics and hydrodynamics, and the critical design issues include power sources, propulsion and location, and many different schemes have been proposed.

1 Introduction The disease of gastrointestinal (G.I.) bleeding is very popular at present. The key of diagnosing or treating GI bleeding successfully is to know where and why is the bleeding. The traditional methods may lead high misdiagnosis, and frequent exposal under X radial is harmful to patients, such as endoscope checking, helical CT, barium meal and so on [1]. A miniature robot system is presented for checking hemorrhagic spot and stopping bleeding. Medical microrobots can be thought of as miniaturized, selfpowered, mobile devices with characteristic dimensions in the sub-mm region and aimed at in vivo applications. It is anticipated that they will travel through blood vessels, the spinal canal, the urethra or the alimentary system to undertake a range of clinical tasks. In fact, more and more researchers are realizing that wireless drive to microrobot is the key point to enhance its feasibility and reliability. This paper will discuss the development of the power supply and other relevant issues of the medical microrobots [2-3].

2 Issues Facing the Medical Microrobots The development of medical microrobots is still at an early stage and poses an enormous, multi-disciplinary challenge. It involves a wide and diverse range of T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 451–459. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

452

X. Lv et al.

technologies and disciplines, such as silicon microtechnology and MEMS, nanotechnology and NEMS, micro-electronics, bioengineering, materials science, electromagnetism and hydrodynamics. Consequently, successful research programs are likely to necessitate collaboration between groups with differing but complementary fields of expertise. In addition to imparting the required functionality to the robot, other critical considerations include:(1) power – how to provide power to the robot, (2) propulsion – how to move it around the body, (3) steering and location – how to control it and monitor its position, (4) Insertion and removal – how to introduce it into the body and remove of it when its job is done. Power is clearly a vital issue and a number of radically different schemes have been proposed. This fall into two broad categories: on-board or external. One of the most topical on-board concepts draws heavily on the field of energy “harvesting” or “scavenging”, which is presently attracting great interest from the sensing community. Many such systems are based on the conversion of mechanical energy into electrical power and an example, which has potential in the microrobot context, is the nanogenerator developed by a group at the Georgia Institute of Technology. This relies on the well known piezoelectric effect which is exploited through the use of an array of around 500 vertically-aligned zinc oxide (ZnO) nanowires. These are piezoelectric and respond to an external force by generating a voltage which should be sufficient to power in vivo medical devices. Some examples of devices using external and other power sources are considered later in this paper.

3 The Power and Propulsion of Medical Microrobots 3.1 Microrobot Driven by GMA The mechanism of the motion with giant magnetostrictive alloy (GMA) is that the magnetic energy is converted into mechanical vibration energy through the action of piezomagnetism and magneto mechanical couple of its micro GMA when timevarying oscillating magnetic field with different frequency applied through outside pipe. Thereafter robot moves forward by transferring axial vibration of GMA into radial vibration by installing elastic legs with different stiffness between upside and downside. Toshio Fukuda et al [4] put forward an in-pipe micro mobile robot which powered by a GMA actuator and do not require any power supply cables. This cableless micro mobile robot can be controlled by the magnetic fields supplied by the outer side. The micro model of the robot has sixteen legs, where the robot is 6 mm in diameter for the one way motion. The GMA is installed into the robot as shown in Fig. 1 and has a magnetic flux perpendicular with the pipe. The tip of the legs has an opposite inclination to the direction of the movement like the macro model and it is pressed against the inner wall of the pipe. The robot can move by vibrating the legs like a macro model.

The Power and Propulsion of Medical Microrobots

453

Fig. 1. Structure of micro model

3.2 Microrobot Driven by Piezoelectric Element Fig. 2 shows a basic motion principle of the piezo impact drive mechanism [5]. The motion mechanism consists of three components: the main body, the actuator and the inertial (counter) weight. The main body is laid down on the guiding surface with only the friction acting between the surfaces. On the one end of the main body an actuator is attached. The weight does not touch the surface. The processes of the motion are described as: (a) The cycle starts with the actuator in extended state; (b) The actuator makes slow contraction so that the inertial force caused by the contraction should not exceed the static friction. The main body keeps the position; (c) At the end of contraction process, a sudden stop of the motion is made to small move the main body, and then; (d) a rapid expansion of the actuator causes impulsive inertial force, which results in the step-like motion of the main body. Making slow extension and rapid contraction can carry out motion toward the other direction. The motion amplitude of the actuator can control the step size of the motion. Repeating those process through (a)-(d) a long distance motion is made possible.

Fig. 2. Operation principle: moving toward left

454

X. Lv et al.

Fig. 3. Structure of locomotion platform composed of four piezo elements and two U-shape magnet legs

For this decade, such micro-handling and nano-manipulation techniques have been developed. One precise small robot with the size of 30 mm in cubic is shown in Fig. 3 [6]. Here two U shape electromagnets are set up diagonally and these magnets are connected by four stack-type piezo elements each other. When synchronizing these actuators each other, then this mechanism can walk step by step like an inch-worm.

3.3 Microrobot Driven by Cell Battery Toshiaki Yamaguchi et al [7] made a miniature robot system which has two parts. The main part is a micro robot which works in human’s G.I. tract like a capsule. The subsidiary part is a receiver which works outside of body and receive signal from the microrobot. The structure of micro robot is depicted in Fig. 4. The miniature robot is very significant to be improved as an advanced medical instrument for gastrointestinal bleeding, particularly for bleeding in small intestine. But the micro robot is supplied by high-energy cell battery, how to reduce the power consumption is very important.

Fig. 4. Structure of inner micro robot

3.4 Microrobot Driven by Magnetic Field Wireless microrobots controlled by magnetic field are used in a wide range of biomedical application. Especially microsurgery in blood vessels is expected to become an increasingly popular medical practice. In the meantime, being


455

characterized with high reliability and safety, it could reach deep cavern by the medium of body fluid organism with flexibility, thus the microrobot for biomedical application, as a new important approach on therapy in term of interposition, has a widespread prospect in the field of medical engineering.

Fig. 5. The structure of the spiral

Qinxue Pan et al [8] presented a novel spiral type of the microrobot in a pipe which have a small scale, can be propelled by low voltage, and has a quick response. And also, for the advantage of this kind design of the microrobot, it can not only move smoothly in the human body but also it can remove the obstructing in intestine or blood vessel using the particular structure. The structure of the proposed spiral type microrobot (Swimmer-1) is shown in Fig. 5. In order to rotate the microrobot in the magnetic field, they set four permanent magnets on the body, which are set in the top, bottom, front and back of the body, respectively, and not in the same line in order to generate torsion motion. The body in spiral structure is made by rubber. Consider the accuracy control of microrobot; they have improved the swimmer-1 which increased the balance, control stability and even propulsive force. The improved microrobot (swimmer-2) has been developed, as shown in Fig. 6. When current flows through the solenoid, a magnetic field is generated. When an alternate magnetic field parallel to the direction of advance is applied, movement due to an impelling force arising from a permanent magnet rotates and vibrates the connected fin. The motion mechanism of the tail is shown in Fig. 7.

Fig. 6. The conceptual design of the improved microrobot

456

X. Lv et al.

Fig. 7. Motion mechanism of the tail

3.5 Microrobot Driven by Biotic Energy Medical nano robots are a very promising research direction for creating efficient mechanisms for medication delivery and for directly fighting disease, i.e., nano robots working along with natural white cells to fight viruses inside the human body. Medical nano robots can help us eliminate the problem of antibiotic resistant viruses once and for all and it is certainly one application of robotics. According to a CBC article, Researchers at Cornell University are trying to find out how to use the same mechanism that powers human sperm cells in medical nano robots. By deconstructing the stages in the biological pathway sperm cells use to generate their relentless energy, researchers at the Cornell University College of Veterinary Medicine in the United States hope to recreate that process in an artificial device. Powerful, albeit microscopic, sperm cells use a kind of dualengine system to generate their energy. Organelles in a sperm cell's midsection provide one part of its battery power, while a second process occurring in the long, spindly tail gives it an additional boost. Researchers have attached three of the 10 enzymes needed to create this glycolytic pathway to nickel ions on a tiny manufactured chip. Their goal now is to attach the remaining seven enzymes, in effect creating nano-robots fuelled by sperm power.

3.6 Microrobot Driven by PDMS Microbial Fuel Cell Chin-Pang-Billy Siu et al [9] have recently find that people of blood glucose for energy in the yeast cells that one day people will be driven into the body of the electronic devices, such as pacemakers. This vivid self-generated power, periodic replacement of batteries will replace conventional surgery. Since the University of British Columbia, Canada will be a team of scientists at yeast encapsulation soft capsules, has been developed to such a micro fuel cell micro-organisms, can be applied to the treatment of paralysis of the spinal intracellular microelectrode such electronic devices. Engaged in the development of these yeast fuel cell scientists, said spine microelectrodes implanted in the spine necessary therefore very difficult to replace batteries.


457

Fig. 8. Operating principle of the PDMS MFC

The new fuel cell component from saccharomyces cerevisiae, the yeast sealed by poly dimethyl siloxane (PDMS) material in the capsule, together constitute the fuel cell. The currently developed fuel cells are Samples micron level, with an area of 15 square microns, 1.4 microns thick. Fig. 8 shows the operating principle of a PDMS MFC. When yeast break down glucose, the methylene blue from the yeast cells where the "theft" electronics, electronic delivery of this will be followed by the other side of yeast cells, resulting in a weak current. On the cathode, from yeast cells combine hydrogen and oxygen produced water. To increase the size of the electrode to increase the fuel cell power output, the group using silicon etching techniques to create "mini-poles", this covers an area of about 40 square microns, 8 microns high. Tested, the yeast can produce fuel-cell around 40 nava electricity, compared to quartz watches generally can produce electron micro-watt power class. Therefore, if the use of the energy storage capacitors, and the yeast will be able to fuel cells some device drivers. This yeast will be genetically modified so that it will have greater power output capability. However, this goal faces many challenges. For instance, let the healthy growth of yeast cells, and its waste cases no damage was cleaned up, so that discharges of harmful substances to human blood.

4 Other Relevant Issues Whilst most of the above examples address a limited range of specific technological issues rather than aiming to develop commercial products, September 2006 saw the launch of a large, highly multi-disciplinary and more ambitious programme: the € € 9.5 million European versatile endoscopic capsule for gastrointestinal tumor recognition and therapy (VECTOR) project. Funded under the European Commission’s Sixth Framework Programme and led by novineon Healthcare Technology Partners GmbH, this four year project involves 18 organisations from nine European countries, together with the Intelligent Microsystem Centre at the Korean Institute of Science and Technology. The aim is to exploit the latest developments in MEMS and nanotechnology to fabricate a miniaturized robotic “pill” for advanced cancer diagnostics and therapy in the human digestive tract [3, 10-13].

458

X. Lv et al.

Fig. 9. Schematic of the VECTOR endoscopic robot

This application reflects that these cancers can be treated if detected early but only a fraction of patients make use of screening endoscopy since current procedures are associated with notable discomfort and pain. Thus, novel, painless endoscopic devices are needed to increase screening rates. Perhaps, more of a mini-robot than a true microrobot, the VECTOR device will take the form of an actively controllable, miniature capsule endoscope, equipped with a vision system and operating instruments which could not only detect stomach and bowel cancer in the early stages but also treat it in situ. This will be made possible by equipping the VECTOR capsules with miniaturised grippers and operating instruments to remove or destroy diseased tissues (Fig. 9) [3].

5 Conclusions In the future, the design and energy supply of the wireless microrobot will be optimized. Design and use an electromagnetism sensor for detecting the location of the microrobot in pipe and accuracy control in speed and position of the microrobot must be realized. The application of in-pipe inspections, operation in microenvironment and microsurgery of blood vessels will be useful in industrial and medical field. These examples illustrate the present state of the medical microrobot technology. It is evident that no single approach has emerged which will resolve critical design aspects such as power or propulsion, which will almost certainly be governed by the application. Opinions are divided as to when medical microrobots will become commercially available and vary from five to ten years and, of course, they will need to undergo extensive testing and approvals prior to being deployed on a routine basis. Nevertheless, research is gaining momentum and what appeared to be little more than mere speculation a decade ago is now poised to become a reality. Acknowledgement. This work was supported by Leading Academic Discipline Project of Shanghai Municipal Education Commission, Project Number: J51301.


459

References [1] Yan, G.-z., Shi, J., Fang, Y.: Interventional Miniature Robot for Diagnosing and Treating G. I. Bleeding. In: The First IEEE/RAS-EMBS International Conference on Biomedical Robotics and Biomechatronics, February 20-22, pp. 1020–1023 (2006) [2] Oleynikova, D., Rentschlera, M., Hadzialicb, A., et al.: In vivo camera robots provide improved vision for laparoscopic surgery. International Congress Series, vol. 1268, pp. 787–792 (2004) [3] Bogue, R.: The development of medical microrobots: a review of progress. Industrial Robot: An International Journal 35(4), 294–299 (2008) [4] Fukuda, T., Hosokai, H., Ohyama, H.: Giant magnetostrictive alloy applications to micro mobile robot as a micro actuator without power supper cables. In: Proceding of 1990 IEEE Conference on Micro Electro Mechanical Systems, pp. 210–215 (1990) [5] http://www.aml.t.u-tokyo.ac.jp/research/pidm/pidm_e.html [6] Fuchiwaki, O., Tobe, N., Aoyama, H., Misaki, D., Usuda, T.: Automatic microindentation and inspection system by piezo driven micro robot with multiple inner sensors. In: Proceedings of the IEEE International Conference on Mechatronics & Automation Niagara Falls, Canada, pp. 83–88 (July 2005) [7] Yamaguchi, T., Kagawa, Y., Hayashi, I., Iwatsuki, N., Morikawa, K., Nakamura, K.: Screw principle microrobot passing steps in a small pipe. In: International Symposium on Micro Mechatronics and Human Science, pp. 149–152 (1999) [8] Pan, Q., Guo, S., Li, D.: Mechanism and Control of a Spiral Type of Microrobot in Pipe. In: Proceedings of the 2008 IEEE International Conference on Robotics and Biomimetics, Bangkok, Thailand, pp. 21–26 (February 2009) [9] Siu, C.-P.-B.: Student Member, and Mu Chiao, A Microfabricated PDMS Microbial Fuel Cell. Journal of Microelectromechanical Systems 17(6), 1329–1341 (2008) [10] Itoh, A.: Motion control of protozoa for bio MEMS. IEEE/ASME Trans. Mechatronics 5(2), 181–188 (2000) [11] Ogawaa, N., Okua, H., Hashimotob, K., Ishikawa, M.: A physical model for galvanotaxis of paramecium cell. Journal of Theoretical Biology 242, 314–328 (2006) [12] Ogawa, N., Oku, H., Hashimoto, K., Ishikawa, M.: Trajectory planning of motile cell for microrobotic applications. Robotics and Mechatronics 19(2), 190–197 (2007) [13] Davies, A., Ogawa, N., Oku, H., Hashimoto, K., Ishikawa, M.: Visualization and estimation of contact stimuli using living microorganisms. In: Proc. 2006 IEEE Int. Conf. Robotics & Biomimetics (ROBIO 2006), pp. 445–450 (December 2006)

Mechanical Design and Analysis of an Articulated-Tracked Robot for Pipe Inspection Z.Y. Chen, G.Z. Yan, Z.W. Wang, and K.D. Wang Institute of Precision Engineering and Intelligent Micro system, Shanghai Jiao Tong University (SJTU), Shanghai 200240, P.R. China

e-mail: [email protected]

Abstract. In order to execute the task of inspecting welding defects, spray painting and corrosive status inside a pipe automatically, an articulated-tracked robot was designed to replace the manual labor. In this paper, authors describe the mechanical design and analysis of an articulated-tracked robot which has four swing arms and six tracks. The actuation mechanism and transmission mechanism are described in detail. To guarantee the operation of this robot, the mutative bending moment acting on the main shaft is calculated and described through force diagram and algebraic expressions. In addition, the dimensions of the robot are determined according to the maximum moment acting on the dangerous crosssections.

1 Introduction The pipe robot is a special kind of robot which can get access into the industrial pipes and execute work such as inspection, welding and spray painting inside the pipe instead of human. In recent years, several kinds of actuation solution for the pipe robot have been developed. The locomotion strategies mainly base on the tracks, wheels or the legs [1,5]. The robot characterized in this paper is used to accomplish the mission of detecting special underground industrial pipes which are damaged in the earthquake. The condition inside the pipe is unclear. So it is not a wise choice of detecting it by people directly. In this condition, the author designed a four degree of freedom actuation system which includes four swing arms and six tracks, allowing more capacity of obstacle-negotiation. Hence, it is called the articulated-tracked robot.

2 Pipe Inspection Robot System The process of the robot system in pipe inspection is described In Fig.1. During the process, the video is sent to the ground control center by a camera on the top of the robot through an electric cable. The operator can control the locomotion of the robot inside the pipe through the center control computer in the control center, including moving forward and backward, turning around or step-climbing. The structure of the robot system consists of three main parts: T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 461–467. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

462

Z.Y.Chen et al.

1) Inner pipe device includes the locomotion system, the camera, the light device; 2) Outer control center includes a computer provided with suitable control, image collecting software and power supply for the locomotion system; 3) Connection device includes several electric cables for communication and power supply.

Fig. 1. Process diagram for the pipe-inspection robotic system

3 Locomotion Analysis for Obstacle-Negotiation The articulated-tracked mechanism has advantages in adapting to the uneven surface and in obstacle-negotiation [2,3]. The two front swing arms and the two rear ones have the same mechanical structures and installation. They can all rotate synchronously. The step-climbing process of the robot is displayed in Fig.2. By changing the swing arms with a certain angle the performance of the tracked robot for step-climbing is dramatically improved [4].

Fig. 2. Obstacle-negotiation process of the articulated-tracked robot

Mechanical Design and Analysis of an Articulated-Tracked Robot

463

4 Actuation System Description The proposed solution is illustrated in Fig.3. This robot device needs four motors, four swing arms and six tracks. The whole device can be separated into two parts, one is the moving device and the other is the device of swing arms.

Fig. 3. The mechanism of the articulated-tracked robot

For the moving device shows in Fig.3(a),each of the two motors on the bottom can control the speed of the main track on one side. The output rotation of each motor is transmitted to the tracks by a pair of cone gears and the main shaft. One end of the main shaft is connected with the active wheels by the shaft keys while the other end is connect with the inactive wheels through bearings. For the swing arms device shown in Fig.3(b), the other two motors on the top could change the angle position of the swing arms. The two front swing arms are connected together and are driven by one motor. The two back swing arms are the same. Two of the gears on the bottom are used to transmit the output rotation and are connected with the main shaft through bearings. So its movement will not be affected by the rotation of the main shaft. In addition, some extra wheels were added to support or change the tightness of the tracks. Finally, a special metal casing was made for the whole actuation system to prevent it from water or dust as show in Fig.3(c)(d).

464

Z.Y.Chen et al.

5 Actuation System Dimensioning 5.1 Robot Motivation Requirement and Motor Selection The requirement of motivation depends on the fact that the tractive force overcome the resistance force

FZT can

FW during the movement. (Fig.4)

Fig. 4. The motivation force diagram of the actuate system

FW includes the internal resistance FWi , the frictional resistance FWf and the drag resistance FWd . The FWi depends on the weight of the The resistance force

whole actuate system, or the degree of the tightness of the caterpillars. The quantity is up to 3-8% of the whole actuate system’s gravity. The FWd depends on the weight of the electric cable and the frictional coefficient of the surface in the pipe. Due to the knowledge of the condition and mechanism we design, the value of the resistance force can be found by using the following parameters:The weight of the actuate system is 55 kg, the frictional coefficient is 0.2. The electric cable is 100 meter long and 20kg in weight. So the resistance force can be calculated as:

FW = FWi + FWf + FWd = 194 N

(1)

To overcome the resistance force, the robot needs suitable motors which can provide enough tractive force FZT [6,7]. Considering the following parameters of the maxon DC motor:the rated torque is 177 Nmm; rated revolution speed n0 =4920r/min;the value of reduction is 124;the revolution speed n1 =40r/min for the main wheels; and 0.91 for the efficiency considering the power reduced by the transmission mechanism. The tractive force can be calculated through formulae:

Tn0 ×ηT = 165w 9550

(2)

PT ≈ 656N 2π Rn1 60

(3)

PT = P0ηT = 2 ×

FZT = PT ν T =


The tractive force

465

FZT we deduced can overcome the resistance force FW . In

this case, we select this maxon DC motor to actuate the robot.

5.2 Mechanical Dimensioning According to the mechanism of the actuation system described above, one of the key components of the robot is the main shaft. Its function includes connection between the wheels, supporting the swing arms, also transmit the torque from the motor to the main wheels. During the transmission of the tracks, the load acting on the shaft may cause the main shaft flexuous. In that case, the quantity of the load should be calculated in order to determine the dimension and material of the shaft. The tractive force FZT is generated by two motors. Each motor provides 83W of power. And the load act on each shaft is:

FZ =

1 FZT = 328 N . 2

(4)

Fig 5 shows the structure of the main shaft and the related parts. Each main shaft drives two parallel tracks. Hence, the force acting on the shaft was uniformly distributed. As show in the Fig.5, one side of the main shaft is considered as cantilever beam with one end point A fixed in the frame. To prevent the shaft from flexure, the maximum moment acting on the shaft should be calculated.

Fig. 5. Actuator structure

Fig. 6. Shaft force diagram in two directions

Fig. 7. Shaft force diagram in horizontal direction

Actually, the swing angle α shows in Fig.4 is changeable. In this case, the force must be calculated in two directions, includes the horizontal and vertical direction.

466

Z.Y.Chen et al.

Fig.6(a) depicts the force the

Fh in the horizontal direction while the Fig.6(b) depicts

Fv in vertical direction. Uniformly distributed load q: q=

Fz = 328 / 40 = 8.2 (N/mm) l

And the force between B and C is changed with the angle

(5)

α

. The angle

α

ranges in [ −90 ,90 ] . According to this force diagram, the moment act on the each point of the shaft can be obtained. Actually the moment is a function of moment M (α ) . We present the expression of the moment of the most dangerous point D and E in two directions: o

o

⎧ M Dv = 400q sin α ⎪ M = 1000q ⎪ Dh ⎨ ⎪ M Ev = 800q sin α ⎪⎩ M Eh = 800q + 400q cos α

(6)

The resultant bending moment can be obtained:

⎧ M = M 2 + M 2 = (1000q) 2 + (400q sin α ) 2 Dh Dv ⎪ D ⎨ ⎪⎩ M E = M Eh 2 + M Ev 2 = (800q sin α ) 2 + (800q + 400q cos α ) 2 According to expressions of

(7)

M D M E , the maximum moment acting on the two

points are:

M E max = 9840 Nmm ( α = 0o horizontal swing arm) M D max = 8616.26 Nmm ( α = 90o vertical swing arm) The calculation process of the maximum moment

M E max is shown in the Fig.7, as

the swing arm is horizontal. The force diagram of the main shaft is shown in Fig.7(a). Fig.7(b) is the shearing force diagram while the Fig.7(c) is the moment force diagram. As a result, the maximum moment acting on the point E (the end of the shaft) is M max = 9840 Nmm . After the study of the bending strength of several materials, the structural alloy steel is chosen as the material of the shaft.


d min = As

3

M E max = 10.98mm 0.1[σ −1b ]

467

(8)

[σ −1b ] depends on the material, the diameter of the shaft should be no less

than 10.98mm.

6 Conclusions Based on the mechanism of the articulated-tracked robot, the authors analyzed the requirement and selected the right motors. In order to enhance the reliability and prevent the components from overload, the varied bending moment acting on the main shaft is described through force diagram and algebraic expressions. Finally, the dimension of the main shaft is determined according to the maximum moment acting on the dangerous sections.

References [1] Kawaguchi, Y., Yoshida, I., Kurmatani, H., Kikuta, T.: Development of an In-pipe Inspection Robot for Iron Pipes. Journal of Robotics Society of Japan 14(1), 137–143 (1996) [2] Li, Y.W., Ge, S.R., Fang, H.F.: Effects of the Fiber Releasing on Step-climbing Performance of the Articulated Tracks Robots. In: Proceedings of the 2009 IEEE International Conference on Robotics and Biomimetics, Guilin, China, December 19 -23 (2009) [3] Chen, C., Trivedi, M.M.: Reactive Locomotion Control of Articulated-Tracked Mobile Robots for Obstacle Negotiation. In: Proceedings of the 1993 IEEE/RSJ International Conference on Intelligent Robots and Systems, Yokohama, Japan (1993) [4] Chen, Z.Y.: Research on pipeline in-service inspection robot system. Bachelor’s Thesis of Shanghai Jiao Tong University (July 2009) [5] Quirini, M., Menciassi, A., Scapellato, S., Stefanini, C., Dario, P.: Design and Fabrication of a Motor Legged Capsule for the Active Exploration of the Gastrointestinal Tract. IEEE/ASME Transaction on Mechatronic 13(2) (2008) [6] Liu, J., Wang, Y., Ma, S., Li, B.: Analysis of stairs-climbing ability for a tracked reconfigure-able modular robot. In: Proceedings of the IEEE International Workshop on Safety security and Rescue Robotics, Kobe, Kobe, pp. 36–41 (2005) [7] Kim, C., Yun, S., Park, K.: Sensing system design and torque analysis of a haptic operated climbing robot. In: Proceedings of the IEEE /RSJ International Conference on Intelligent Robots and Systems, Piscataway, NJ, USA, pp. 1845–1848 (2004)

Part VI

Intelligent Control and Its Applications in Engineering

A Construction Method of Rational Approximation Model for Fractional Calculus Operators in Frequency Domain Wen Li, Guanghai Zheng, Bing Nie, Huimin Zhao, and Ming Huang Software Technology Institute of Dalian Jiaotong University, Dalian, China

Abstract. The controller model containing fractional calculus operators is an irrational function, therefore, to research the rational approximate of fractional calculus operators is an important content for analysis and realization of fractional-order controllers. In this paper, a construction method of fractional order operator was proposed in the frequency domain. Research shows that the rational approximation model constructed by the method can approximate fractional order operator effectively in a given frequency bandwidth.

1 Introduction The discretization of the fractional order differentiator and integrator have been developed during the last 10 years [1-5], most researches are on direct discretization methods. Their works showed how to approximate the fractional order operator by a rational discrete-time models, which can be achieved by using various approximation methods [2-4], such as direct method, indirect method and so on. In general, the direct discretization method based on discrete-time models, first fractional order operators s ±α , 0 < α < 1, α ∈ R are expressed by so-called generating

function s = ω ( z −1 ) [8-9]. There are three of the most commonly used discretization schemes, namely the trapezoidal (Tustin) rule, the backward difference (Euler) rule, and the more recently introduced Al-Alaoui operator, which is obtained by the stable inversion of the weighted sum of the Tustin integration rule and the Euler integration rule [10]. The approximation performance is test by comparing frequency characteristics of approximation model with that of desired fractional order operator in frequency domain. In fact, generating function transforms fractional order operator s ±α to discrete-time z domain, the transformed operator is an irrational z-function. The irrational z-function is expanded by Power Series Expansion, Continued Fraction Expansion or other expansion method. The expansion obtained is a discrete-time model G ( z −1 ) [3-4], which is the approximation of fractional order operator. In more recent articles, a direct discretization method based on Rational Chebyshev Approximation (RCA) is proposed and good approximating performance is achieved, because the RCA is a useful tool to find good rational approximation to a given function f (x) in a given interval [a, b][2]. T.-J. Tarn et al. (Eds.): Robotic Welding, Intelligence and Automation, LNEE 88, pp. 471–478. springerlink.com © Springer-Verlag Berlin Heidelberg 2011

472

W. Li et al.

Some indirect discretization methods are also discussed [1,5,11]. The literature [5] is a representative research paper and gives an in-depth discussion on frequency-band complex noninteger differentiator. In paper [11], a new method is proposed for multiple fractional order approximation. The basic idea is to reduce the multiple-fractional order system to a simple order fractional one and apply the singularity function method to approximate it by rational transfer functions of alternate pole-zero pairs. However there are not a lot of papers about indirect discretization methods. Researches show that the type of generating function and expansion method affect approximating performance for direct discretization methods, and some methods are complex to get discrete-time approximation models [6-7]. In addition, approximation performance of the obtained discrete-time model usually needs to be test by log-magnitude characteristic and phases-angle characteristic. It is therefore most necessary and feasible for us to research constructing rational approximation model of fractional order operator in frequency domain [1,5-6]. The aim of this paper is to find a convenient and practical method of constructing rational model to approximate fractional order operator in frequency domain, the approximation precision can be selected in advance. The paper is organized as follows. First, the construction method is introduced in detail, and then the construction algorithm is presented. Next, the validity of the construction method is proved by some demonstrating examples. Finally, some conclusions are given.

2 The Construction Method of Rational Approximation Model in Frequency In order to make the rational approximation method proposed in this paper convenient and practical in frequency domain, the concept of minimum phase system is applied in constructing method of rational approximation model. That is, if only all first-order m-zeros and first-order n-poles of the rational approximation model to be constructed are negative real number, then consistency between phase frequency characteristic and magnitude frequency characteristic will be satisfied. Therefore the research focus is on the discussion of how to construct logmagnitude characteristic of rational approximation model, the construction of phase frequency characteristic is ignored.

2.1 Construction Method Suppose that a transfer function is composed of fractional order integrator and proportion coefficient, namely

G0 ( s ) =

k0 , k 0 = ω cα , 0< α

Robotic Welding, Intelligence and Automation: RWIA'2010

Robotic Welding, Intelligence and Automation

Robotic welding, intelligence and automation

Robotic Welding, Intelligence and Automation (Lecture Notes in Control and Information Sciences)

Robotic Welding, Intelligence and Automation (Lecture Notes in Control and Information Sciences)

Lyapunov-Based Control of Robotic Systems (Automation and Control Engineering)

Robotic mapping and exploration

Soldering, Brazing and Welding

Robotic Urology

Fabrication and Welding Engineering

Robotic Surgery

Soldering, Brazing and Welding

Welding Manual

Robotics and automation handbook

Robotics and Automation Handbook

System Engineering and Automation

Robotics and automation handbook

Intelligence and Intelligence Testing

Laser Welding

Welding Metallurgy

Welding Metallurgy

Visual perception and robotic manipulation

Retroperitoneal Robotic and Laparoscopic Surgery

Automation and Robotics

Robotics and Automation Handbook

Automation and Robotics

Robotics and Automation Handbook

Robotics Automation and Control

Robotics and Automation Handbook

Robotic Welding, Intelligence and Automation: RWIA'2010