Machine Learning-Based Weldability Assessment and Optimization of Resistance Spot Welding Joints: A Data-Driven Framework

Abstract

Resistance spot welding (RSW) is a crucial metal joining process across diverse industries, with a prominent role in the automotive industry. This paper proposes a conceptual framework of data-driven welding analytics by building a machine learning (ML) model and parameter-less optimization algorithm to assess the weldability of spot welded joints. It leverages experimental and simulation data to train the ML model, enabling it to predict weld quality accurately. Various regression-based Machine learning algorithms are implemented to predict weld nugget size. The performance of these algorithms is evaluated by three performance measures: Root Mean Square Error (RMSE), Mean Square Error (MSE), and R-squared. The result shows that Gaussian process regression outperforms other ML algorithms in predicting weld quality. Furthermore, in addition to welding analytics, the proposed framework incorporates a Teaching learning-based optimization (TLBO) algorithm to optimize expected weld quality. The robustness of this optimization algorithm is confirmed through 10 independent runs, maintaining consistent population size and termination criteria.