Multi-physical process simulation of resistance spot welding available for synthetic data generation

Abstract

Resistance spot welding (RSW) faces challenges when realizing online quality evaluation because of insufficient labeled data. Finite element (FE) models can generate synthetic databases, but their application is limited due to the reliability and generalization ability problem. This paper establishes an FE model that can digitally twin the welding gun characteristics, contact behavior, nugget growth process, and key process signals simultaneously. A multi-spring structure was designed to simulate the loading feature of a servo gun, and certain modifications in material properties were applied to the model. The simulation errors can be restricted to 2 % for the weld profile and 5 % for all process signals. A quick generalization method is also proposed to apply the FE model on different stack-ups, only needing modification in electrical contact resistance (ECR) parameters. The modeling and generalization methods were validated on 6 stack-ups consisting of three steels with different mechanical strengths, sheet thickness, and various chemical compositions, and also validated under different currents. The reliability and generalization ability of the proposed model are superior to traditional models, maintaining <5 % and <10 % errors in simulated nuggets and signals, respectively. ECR analysis shows that contact film resistivity is strength-related, and all stack-ups have similar electrical contact resistances at electrode/sheet interfaces. A preliminary synthetic database was generated, including 10 stack-ups and about 1500 data. This study can help provide labeled data for machine/deep learning algorithm training and for interpreting the physical process of RSW.