CSSA-XGBoost: a novel algorithm for inspecting spot welding quality based on dynamic resistance signal

The quality of resistance spot welding is particularly crucial for the welding manufacturing industry. Inspecting the quality of welding joints efficiently and accurately is a significant process. When inspecting high strength steel welding joints, traditional regression models exhibit limited generalization capabilities. In this paper, we propose a CSSA-XGBoost algorithm to predict the tensile shear strength of welding joints by extracting representative features from the dynamic resistance signals during the welding process as model inputs. Specifically, an improved sparrow search algorithm for XGBoost optimization is used, which introduces a chaotic map and Levy flight strategy to improve the quality of hyperparameter search. We conduct comprehensive experiments on the self-built dataset WeldResDB to demonstrate the state-of-the-art of our approach. The root mean square error (RMSE) in prediction reaches 0.3423 KN and the mean absolute percentage error (MAPE) is only 2.25%. The prediction accuracy outperforms other algorithms and provides a reliable method for weld quality inspection in industrial applications.