Advances automation techniques in adaptive material processing

ADVANCED AUTOMATION TECHNIQUES IN ADAPTIVE MATERIAL PROCESSING Editors XiaoQi Chen [ajagopalan Devanathan Aik Meng Fong World Scientific Tai ngay!!! Ban co the xoa dong chu nay!!! This page is intentionally left blank ADVANCED AUTOMATION TECHNIQUES IN ADAPTIVE MATERIAL PROCESSING ADVANCED AUTOMATION TECHNIQUES IN ADAPTIVE MATERIAL PROCESSING Editors XiaoQi Chen Gintic Institute of Manufacturing Technology Rajagopalan Devanathan Nanyang Technological University Aik Meng Fong Gintic Institute of Manufacturing Technology V | f e World Scientific wB New Jersey •London'Singapore* London • Singapore • Hong Kong Published by World Scientific Publishing Co Pte Ltd P O Box 128, Farrer Road, Singapore 912805 USA office: Suite IB, 1060 Main Street, River Edge, NJ 07661 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library ADVANCED AUTOMATION TECHNIQUES IN ADAPTIVE MATERIAL PROCESSING Copyright © 2002 by World Scientific Publishing Co Pte Ltd All rights reserved This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA In this case permission to photocopy is not required from the publisher ISBN 981-02-4902-0 Printed in Singapore by World Scientific Printers (S) Pte Ltd PREFACE Researchers and engineers have long been inspired to innovate automation solutions to a whole range of manufacturing processes such as components assembly, marking, cutting, forming, joining, grinding, polishing and chamfering The past few decades have witnessed widespread applications of CNC technology for machining operations and robotic technology for assembly operations Robots have also gained a solid ground in welding automation, notably spot welding for the automobile industry and arc welding for the manufacture of steel products However, manufacturing processes that require operator's instincts and skills have posed a great challenge to researchers in the areas of robotics, automation and control Despite many research works that have attempted to address some fundamental issues in automating material processing operations, automation practitioners find it difficult to implement integrated solutions to these applications The focus of this book is on how to apply and adapt advanced automation techniques and to implement integrated mechatronic automation systems The book is intended to put in the hands of manufacturing engineers, researchers, and graduate students alike, our own as well as related recent findings on the applications of modern automation techniques These techniques include sensors, signal processing, robotics and intelligent control for the automation of some of the demanding manufacturing processes that were traditionally handled within the mechanical and material engineering disciplines One of the unique features of the book is that the concepts and techniques are developed from real-life material processing applications Specifically, the book includes the latest research results achieved through applied research and development projects over the past years in the Gintic Institute of Manufacturing Technology The research works featured in the V VI Preface book are driven by the industrial needs They combine theoretical research with practical considerations The techniques developed have been implemented in industries The book begins with an overview of material processing automation It highlights the needs of multi-facet mechatronic solutions for adaptive material processing, surveys various sensors, and explains the progression from conventional numerical control to sensor-based machine control and intelligent control Chapter discusses process development and approach for 3D profile grinding and polishing Simplified force control models are presented, followed by conventional model-based robotic machining Having analysed part variations and process dynamics, the chapter details the system concept - the architecture of adaptive robotic system for 3D blending It further presents the results of experimental process optimisation that are encapsulated in the process knowledge base Chapter concerns the implementation of the concept presented in Chapter It discusses finishing robots, control interface, in-situ profile measurement techniques, the template-based optimal profile fitting (OPF), and adaptive robot path planner (ARP) The OPF algorithm finds the best fitting of design data to the measurement points It is fast in convergence, and robust The ARP engine generates the robot path points, and associates them with optimal process parameters Testing results are presented to benchmark the implemented system against the stringent requirements Chapter focuses on acoustic emission sensing for machining monitoring and control Following a brief overview on sensors in machining process monitoring, it explains the acoustic emission sensing mechanisms and experimental set-up It further discusses signal processing techniques such as time domain, frequency domain, and time and frequency domain analysis methods In particular, the wavelet analysis method has been applied for tool condition monitoring Chapter looks into techniques of automatic weld seam tracking, specifically for advanced welding operations Through-arc-sensing technique, robotic system integration, PID control and test results are presented When weld preparation geometric data are required for visionbased seam tracking and welding parameter control, optical vision sensors provide a practical solution To complement Chapter 5, Chapter addresses weld pool geometry sensing It briefly surveys a number of weld pool sensing techniques including weld pool oscillation, ultrasound, laser array, infrared sensing, Preface vu and shape depression sensing In the development work, the laser strobe vision was employed to sense the dynamic change of the weld pool topology A fuzzy logic (FL) controller takes the feedback information of the pool geometry and automatically adjusts the welding parameters to achieve the desired weld formation Chapter presents the implementation of an integrated robotic gas tungsten arc welding (GTAW) system It discusses the system architecture and sub-systems, manipulator configuration, kinematics analysis and simulation, process control, open-architecture CNC controller, and remote control techniques Finally, Chapter extends the discussion to emerging and existing laserbased material processing applications Laser equipment, typical applications and some automation areas are introduced The chapter analyses optical and acoustic signals emitted from the plasma, and presents the sensor design It further discusses signal processing though Fast Fourier Transfer (FFT) and wavelet analysis Real-time monitoring of laser welding using an artificial neural network (ANN) is presented The collection of the research results in this book could not be possible without strong support from the management of Gintic Institute of Manufacturing Technology which has allocated valuable resources towards the projects, from which most of the information and results reported in the book have originated The authors would like to thank the National University of Singapore (NUS) and the Nanyang Technological University (NTU) for collaborating with Gintic in some of the research projects They acknowledge the funding agencies, the Economic Development Board (EDB) and the Agency for Science, Technology and Applied Research (ASTAR), for providing research grants to the research activities The authors would also like to express their sincere appreciation to their industrial partners, Singapore Technology Kinetics (STK) and Turbine Overhaul Services Private Limited (TOS), for their entrusting Gintic with challenging projects, their co-operation and valuable inputs Finally special thanks are due to colleagues and friends for their technical and administrative support in the research projects, and kind assistance and valuable comments in reviewing the manuscript This page is intentionally left blank CONTENTS Chapter Overview of Material Processing Automation Constrained and Non-Constrained Material Proces sing Multi-Facet Mechatronic Automation Sensors for Material Processing 3.1 Measurands in Material Processing 3.2 Types of Sensors 3.3 Microsensors and Soft Sensors Intelligent Control Techniques 4.1 Conventional Computer Numerical Control 4.2 Sensor Based Machine Tool Control 4.3 Open Architecture and Distributed Control 4.4 Intelligent Control and Computing Techniques 4.5 Human-Machine Interface References 1 4 10 10 12 13 15 16 17 Chapter Process Development and Approach for 3D Profile Grinding/Polishing Introduction Profile Grinding and Polishing of Superalloys 2.1 Superalloy Components and Manual Blending 2.2 CNC Milling 2.3 Wheel Grinding Force Control in Material Removal 3.1 Robot Holding Tool 19 19 21 21 25 27 29 30 ix Chapter - Laser Material Processing and Its Quality Monitoring and Control 289 by the factor of 0.7 If the current error is less than the previous one, the learning rate is increased by the factor of 1.1 Two different methods of feature encoding are used One is IMA and the other is RMS The performances of two encoding methods are studied on the experimental data The correct, wrong and uncertain are defined as follows: • • • Correct: normal welding and the output of ANN is larger than 0.6; or abnormal welding and the output of ANN is less than 0.4 Wrong: normal welding and the output of ANN is less than 0.4; or abnormal welding and the output of ANN is larger than 0.6 Uncertain: the output of ANN is between 0.4 and 0.6 The test results are outlined in Table and Table The performance of the two feature encoding techniques is almost the same although IMA looks a little better Both results are acceptable with accuracy of 85% But the non-accuracy parts (wrong and uncertain) are quite dissimilar with different neural network structures In some cases, the inaccuracy and uncertainty is 3.6% and 8.9% respectively Table Percentage of weld fault classification results using IMA hist =10 Accuracy Neuron Num Error rate = 100 Uncertainty Accuracy Neuron Num Error rate = 150 Uncertainty Accuracy Neuron Num Error rate = 200 Uncertainty Accuracy Neuron Num Error rate = 250 Uncertainty 87.0% 5.0% 8.0% 85.7% 4.3% 10.0% 84.2% 4.9% 10.9% 84.0% 5.0% 11.0% hist = 20 87.5% 3.6% 8.9% 84.2% 4.4% 11.4% 84.2% 4.8% 11.0% 82.8% 5.0% 12.2% hist = 30 87.2% 4.3% 8.5% 85.8% 5.2% 9.0% 83.8% 6.7% 9.5% 85.4% 5.2% 9.4% hist = 40 84.5% 4.6% 10.9% 81.8% 4.5% 13.7% 83.3% 4.0% 12.7% 82.0% 6.6% 11.4% 290 H Luo, HZeng, ZZhou, X Hu, and Y Chen Table Percentage of weld fault classification results using RMS Neuron Num= 100 Neuron Num=150 Neuron Num = 200 Neuron Num = 250 Accuracy Error rate Uncertainty Accuracy Error rate Uncertainty Accuracy Error rate Uncertainty Accuracy Error rate Uncertainty hist = 10 86.8% 7.0% 6.2% 86.6% 6.7% 6.7% 84.6% 7.7% 7.7% 83.2% 6.3% 10.5% hist = 20 86.3% 5.3% 8.4% 82.4% 5.5% 12.1% 82.8% 5.8% 11.4% 81.5% 6.4% 12.1% hist = 30 85.7% 4.3% 10.0% 88.6% 4.8% 6.6% 84.5% 5.0% hist = 40 84.5% 4.5% 11.0% 84.2% 5.0% 10.8% 10.5% 86.6% 5.0% 8.4% 12.7% 81.9% 4.1% 14.0% 83.7% 3.6% When the output of ANN is between 0.4 and 0.6, ANN fails to identify the welding status It is difficult to say if it is normal or abnormal Practically, this output could be treated as a warning signal The training and validation results are shown in Figure 30 150 15x10" 300 1.5 Sampling points Figure 30 Training and validation results of ANN (the bold line is target, and the normal line is output of ANN) 3x10J Chapter - Laser Material Processing and Its Quality Monitoring and Control 5.2 29 J Performance of the Neural Network Here, the focus is on generalisation of the trained network with data from other experiments The sole aim is to establish whether or not the network has any potential for broader applicability The first part of validation is concerned with misalignment weld fault experiments Two sets of experimental data are fed into the trained network, as shown in Figure 31 Since the historical information is treated as part of the feature, the initial output of network is unreliable because the feature vector is filled with zero at the beginning when there is no historical information The validation results of the network are generally satisfying for the misalignment fault Initial output is unreliable Sampling Points (a) Welding speed = m/min Sampling Points (b) Welding speed = m/min Figure 31 Validation of a misalignment weld fault experiment The second part of investigation involves presenting the trained network with data from gap weld fault experiments, as shown in Figure 32 Most of output values of the network are uncertain while there is gap fault The 292 H Luo, H Zeng, Z Zhou, XHu, and Y Chen reason is that the gap is very small in this experiment The neural network does not generate satisfactory results for small gap weld fault ^ 2D0D I 1Q00 Iki ll.LiJ.Lhl J Lit, Jb I uni.jLih i, i, O hl'l-J o I^JI* ||.II_LI*I.I]LL.II „ 1.5 ° 0.5 II 15x 1D 1D °l 1h D Gap 50 ,n^H 150 1DD Sampling Points Figure 32 Validation of a gap weld fault experiment Data from normal weld experiments are also fed into the trained network to validate its ability, as shown in Figure 33 1.5 I ° ^*|[y4H^ - D.5 50 200D 1D0 200 150 ill j Lilt I.,.! lL JIJI.I I , i hill , u 111! i, 250 iL I .i I 1000 ^^LI^hLliJLUld^l^L-MiJJMi^llJtlfclLfgkaft^ytJ 0.5 m,» 2.5 x10 1.5 Sampling Points (a) Sample set #1 2000 —i I LJllU.il I J i k 1 II I 11 i l l 1 1 L J.il I L i l l LNII.III L., I J r 1D00 I LJhllLitfa'li^lka^fc