Hybrid Nugget Diameter Prediction for Resistance Spot Welding

Abstract

Resistance spot welding (RSW) is one of the critical joining methods in sheet metal-based industries. The nugget size is a crucial factor, which determines the stoutness of final products. In this research, we present a hybrid prediction method based on a theoretical physical model and a machine learning method (i.e., spline-based interpolation). The theoretical physical model expects to show relatively higher precision in prediction; however, it needs an expert knowledge to find proper dynamic parameters (e.g., resistance and heat energy) of the model. On the other hand, the machine learning model is likely to find the model automatically. However, it seems to be overfitted for the specific training dataset (especially with inconsistent RSW dataset) and it needs massive datasets. To overcome these data inconsistency and hidden physical patterns, this paper proposes a hybrid method based on the theoretical model and the interpolation with the spline method. The theoretical physical model builds a mathematical model for the nugget growth and its formation. The spline-based interpolation method analyzes the correlation between the input parameters (e.g., stack-up characteristics, welding force, welding current, and welding time) and the output parameter (i.e., nugget diameter). The experiment results show that the proposed hybrid method provides more meaningful prediction than the existed prediction algorithms with the RSW dataset and can be an error-data endurable method.