Machine learning for predicting resistance spot weld quality in automotive manufacturing

Resistance Spot Welding (RSW) stands as the primary joining process in the automotive industry, renowned for its suitability for automation and integration into high-production assembly lines. Despite its advantages, accurately evaluating RSW remains challenging, resulting in additional costs and production steps. Current inspection methods, reliant on random checks after cars leave the Body-in-White (BIW), often lead to significant time losses, emphasizing the necessity for enhanced quality assessment. This study aims to transition from random checks to 100 percent inspection using data analysis and machine learning techniques. By predicting weld quality levels prior to car body completion, this approach aims to improve quality control. Five distinct algorithms—Artificial Neural Network (ANN), Convolution Neural Network (CNN), Long Short-Term Memory (LSTM), Random Forest Classifier (RFC), and Extreme Gradient Boosting (XGBoost)—were assessed. The research highlights that the proposed methodology, particularly leveraging XGBoost, achieves a notable prediction accuracy of 97.1% when applied to unseen data. • Objective : Addressing the challenge of transitioning RSW evaluation to 100% inspection in the automotive industry. • Innovation: Introducing a machine learning approach to predict weld quality before car body completion. • Techniques Evaluated: Evaluated five algorithms, with XGBoost achieving the highest accuracy of 97.1% on unseen data. • Key Finding: XGBoost enhances weld quality assessment, improving reliability and efficiency in manufacturing. • Impact: This approach transforms RSW evaluation, improving quality control, reducing costs, and boosting efficiency.