Welding is one of the most important manufacturing technologies. Eliminating weld defects is crucial due to the detrimental effects on the component integrity and safety. Today, welds are made in a prescribed manner—welding conditions such as welding current and welding speed are predetermined based on weld qualification trials. It is very difficult, if not impossible, to proactively adjust in real time the welding conditions to compensate unexpected variations in real-world welding causing the formation of welding defects. Post-weld quality inspections such as ultrasound, X-ray, and dye-penetrant are generally mandatory per code requirement. Repair and correction of the weld defects after the weld is made is time-consuming and expensive, often much more than the cost of welding. Furthermore, undetected weld defects left in the structure are a major threat to the safety and integrity of nuclear reactor structural components.
GTAW is one of the main manufacturing representative technics and has a wide range of applications in the manufacturing industry. This book describes the typical application of vision sensing technology and machine learning algorithm on dynamic GTAW process, one of the key technologies of intelligent welding manufacturing. The development of intelligent automatic welding monitoring system with vision sensing for real-time weld defect detection and adaptive adjustment to the welding process conditions to eliminate or minimize the formation of major defects is presented.
The book is divided into eight chapters. Chapter 1 reviews the related works in the field of defect detection and control using visual sensing during welding. In Chap. 2, the approach of monitoring weld pool surface geometry based on the active vision images is discussed. Chapter 3 describes the adaptive passive vision system to measure the 3D weld pool surface geometry for bead on plate welding. A novel method based on the reversed electrode image (REI) was proposed to calculate weld pool surface height in real time based on simple 2D image. Chapter 4 presents penetration prediction with machine learning approach for bead on plate weld and butt joint weld. The implementation of a closed penetration control system is presented in Chap. 5. Penetration detection was further applied on the U-joint, and the implementation and experiment result is presented in Chap. 6. Chapter 7 presents the lack of fusion detection in multi-pass welding U-groove joint. Chapter 8 draws the conclusions.
This book is mainly based on the research work done by Zongyao Chen during the Ph.D. study in University of Tennessee, Knoxville. The state-of-the-art technology in computer vision, image processing, and machine learning and its application to detect welding defects, including incomplete penetration and lack of fusion, are presented. These technologies offer the valuable strategies to detect the defects with the non-destructive methods to improve the welding productive. Researchers, scientists, and engineers who working in the field of advance manufacturing can benefit from this book.

1 Introduction 1

1.1 Background 1

1.2 Research on Visualizing Dynamic Molten Pool Characters in GTAW 2

      1.2.1 Weld Pool Image Segmentation 3

      1.2.2 Monitoring 3D Weld Pool Geometry 4

1.3 Research on Welding Defect Detection Using Vision Sensing Method 6

1.4 Contribution 9

References 10

2 Monitoring Weld Pool Surface with Active Vision Image 13

2.1 Visual Sensing System Design 13

2.2 Weld Pool Characters in Active Vision Image 15

2.3 Weld Pool Image Segmentation 16

2.4 Experiment Result 18

2.5 Discussion 19

      2.5.1 Weld Pool Depression 19

      2.5.2 Welding Penetration 20

      2.5.3 Undercut Defect in High-Speed Welding Condition 22

2.6 Conclusions 23

References 24

3 Visual Sensing of 3D Weld Pool Geometry with Passive Vision Image 25

3.1 Description of 3D Weld Pool Geometry for Bead on Plate Welding 25

3.2 Passive Vision Image Acquisition 26

3.3 2D Weld Pool Geometry Measurement with Adaptive Passive Vision Method 28

      3.3.1 Conventional Image Segmentation Method 28

      3.3.2 Software Framework of Adaptive Passive Vision Method 29

      3.3.3 Landmarks Detection 29

      3.3.4 Camera Exposure Time Determination Based on SVM 31

      3.3.5 Experiment Validation 32

3.4 Monitoring Weld Pool Surface from Reversed Electrode Image 33

      3.4.1 Acquisition of Reversed Electrode Image During GTAW 34

      3.4.2 Reflection Model of Weld Pool Surface 35

      3.4.3 Algorithm of Weld Pool Surface Height Calculation 37

      3.4.4 Experimental Validation of SH Measurement 38

3.5 Validation of 3D Weld Pool Geometry Measurement 41

3.6 Conclusion 45

References 45

4 Penetration Prediction with Machine Learning Models 47

4.1 Definition of Welding Penetration 47

4.2 Data Collection 49

4.3 Evaluation Criteria 49

4.4 Linear Regression for Penetration Prediction 50

      4.4.1 Linear Regression Model 50

      4.4.2 Feature Selection 51

4.5 Penetration Prediction Using Artificial Neural Network 53

4.6 Bagging Tree Model Prediction 55

4.7 Penetration Prediction on Butt Joint Welding 57

4.8 Conclusion 59

References 60

5 Penetration Control of Bead on Plate Welding 61

5.1 Framework 61

5.2 Modeling Welding Dynamic Behavior 62

      5.2.1 Dynamic Modeling Identification 62

      5.2.2 Simulation 63

5.3 Penetration Control on Uniform Thickness Plate 64

5.4 Penetration Control on Different Thickness Plate 66

      5.4.1 System Modeling 66

      5.4.2 Experiment 68

5.5 Conclusion 68

6 Penetration Detection of Narrow U-Groove Welding 71

6.1 Welding Parameters 71

      6.1.1 Welding Joint Design 71

      6.1.2 Establishment of Database 72

6.2 Image Characters of Root Pass Welding 73

      6.2.1 Images Character of Multi-optical Sensing Condition 73

      6.2.2 Acquire Images with Different Welding Conditions 73

6.3 Training of Prediction Model 76

      6.3.1 Classification Based on the Extracted Features 76

      6.3.2 Backside Width Prediction with Bag Tree Model 77

6.4 Experiment Validation 78

6.5 Conclusions 80

7 Lack of Fusion Detection Inside Narrow U-Groove 81

7.1 Design of Multi-pass Welding Experiments 81

7.2 Experimental Observations 83

      7.2.1 Weld Bead Geometry 83

      7.2.2 Characters of Passive Vision Images 85

      7.2.3 Features Extraction from Passive Vision Image 87

7.3 Predict Lack of Fusion with Data-Driven Model 88

7.4 Software Integration 90

7.5 Conclusions 90

References 91

8 Conclusions and Recommendations 93

8.1 Conclusions 93

8.2 Future Work 95

Dr. Zongyao Chen received his B.S. and M.S. in Electrical Engineering in 2009 and 2012 from Shanghai Jiaotong University. He received his Ph.D. from the University of Tennessee in November 2018. During his Ph.D. study, he worked with the welding and joining research group at oak ridge national lab and focused on applying artificial intelligence and computer vision technology to welding manufacturing and proposed the Reversed Electrode Image (REI) method to extract the features and penetration of welding pool during GTAW process in his doctoral dissertation. He is currently working as a Research Scientist at the R&D Centre of Air Liquide in Delaware, USA. Dr. Chen's research interests include robotics, welding automation, computer vision， machine learning and other AI applications in industry. PA\Dr. Zhili Feng is a Group Leader of Materials Processing and Joining Group and a Distinguished R&D Staff of Oak Ridge National Laboratory. He manages a diverse R&D portfolio aimed at addressing the materials processing and joining needs from automotive, aerospace, nuclear, petrochemical and power generation industries. His primary interest is in thermal–mechanical–metallurgical behaviors of materials during processing and joining. Most recent work included integrated computational welding engineering (ICWE), proactive weld residual stress control and management, friction stir welding and processing, characterization of weld by advanced neutron and synchrotron scattering, and novel solid-state joining processes of dissimilar metals. Dr. Feng received his Ph.D. in Welding Engineering from the Ohio State University. He is a Fellow of the American Welding Society, a Joint Faculty Professor，Department of Mechanical, Aerospace, and Biomedical Engineering, University of Tennessee, Knoxville, and Guest Professor of Tsinghua University. Dr. Feng has broad interactions with industry and extensive experience in solving critical industry problems.Dr. Feng is currently one of Editors-in-Chief Transactions on Intelligent Welding Manufacturing (TIWM) authorized by Springer for periodical publication of research papers from 2017. PA\Dr. Jian Chen is a Research Staff in Materials Processing and Joining Group at Oak Ridge National Laboratory. He has significant experimental and analytical experiences in developing advanced materials joining and processing technologies and the associated control and monitoring techniques. His current R&D focuses on advanced welding and joining techniques, intelligent welding process monitoring and control, non-destructive weld quality inspection and high-performance welding simulation. Dr. Chen received his doctoral degree in Industrial Engineering from the Ohio State University. He is a member of American Welding Society's Technical Papers Committee and 2nd Vice Chair of American Welding Society's North East Tennessee Section.

中科院文献情报中心