About this Abstract |
Meeting |
Advances in Welding and Additive Manufacturing Research 2022
|
Symposium
|
Advances in Welding and Additive Manufacturing Research 2022
|
Presentation Title |
Intelligent Resistance Spot-Weld Geometry Prediction through Machine Learning and Process Monitoring |
Author(s) |
Hassan Ghassemi-Armaki, Peng Wang, Matthew Russell, Joseph A. Kershaw, Yujun Xia, Tian-le Lv, Yongbing Lee, Blair E. Carlson |
On-Site Speaker (Planned) |
Hassan Ghassemi-Armaki |
Abstract Scope |
Resistance Spot Welding (RSW) is a major joining process in automotive industry. Because of its high production efficiency, an automated and robust process control system is needed for weld quality assurance. The weld quality is more critical nowadays as the next-generation of high strength materials are used for battery box design that requires a high-quality standard for achieving free-leaking design. The high weld quality can be achieved through a real-time RSW process monitoring system towards reliable weld quality inspection, which is challenging due to various process variables and uncertainties. To overcome this issue, the development of Machine Learning (ML) models has been studied to quantitatively correlate weld quality with process variables and in-line processing signal data. This paper presents an interpretable data-driven ML model and uses game theory-based Shapley values for efficient and interpretable prediction of nugget quality metrics (e.g., nugget diameter, thickness, indentation). The interpretation of the model predictions can contribute to the physical understanding of: 1) how would the nugget quality vary under different manufacturing varieties, e.g., gap, electrode angle; 2) how would processing signals data (e.g., dynamic resistance, displacement) be beneficial for characterization of manufacturing uncertainties; 3) what would be the contributions of individual welding variables and processing signal to the prediction of nugget quality. This study presents an interpretable ML model that discusses the prediction accuracy for different scenarios of a single homogenous stackup of DP590 under different welding current and welding time. |
Proceedings Inclusion? |
Undecided |