Date of Award
1-17-2024
Publication Type
Thesis
Degree Name
M.Sc.
Department
Computer Science
Keywords
Deep Learning, Nondestructive Evaluation, Semantic Segmentation, Ultrasonic Imaging
Supervisor
Roman Maev
Supervisor
Robin Gras
Rights
info:eu-repo/semantics/embargoedAccess
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Abstract
Resistance spot welding (RSW) is one of the most widely used joining methods in automotive body assembly, and nondestructive evaluation techniques for RSW inspection have aided in reducing manufacturing costs associated with labor and defects. However, current state of the art inspection techniques are either limited in the information they provide or are incapable of in-process defect detection. The automotive industry requires inspection solutions for weld assessment that through real-time feedback can help ensure high-quality welds are produced. This work introduces a deep learning approach for ultrasonic process monitoring of RSWs that provides detailed information about the joining process via semantic segmentation of the ultrasonic data. The baseline model achieved a stack region intersection over union (IoU) of 0.985 and a nugget region IoU of 0.879, while operating on A-scan sequences, with an inference time of 0.7ms / A-scan. The approach could be integrated into an ultrasound-based RSW process monitoring system to provide closed-loop feedback to an adaptive weld controller to ensure the quality of joints. Additionally, an ablation study was conducted where the effects of several different parameters were investigated to produce a more optimal model. This work further demonstrates that manufacturing process monitoring solutions with deep learning create unprecedented use cases in nondestructive evaluation (NDE), and that a system for defect prevention of RSWs (towards NDE 4.0) is possible via deep learning.
Recommended Citation
Tusinean, Vlad, "Real-Time In-Process Ultrasonic M-Scan Evaluation of Resistance Spot Welds Using Deep Learning Towards Adaptive Welding for Zero Defect Manufacturing" (2024). Electronic Theses and Dissertations. 9159.
https://scholar.uwindsor.ca/etd/9159