Research on Machine Learning–Based Prediction of Heterogeneous Metal Joining Performance and Its Application in Production and Operations Management

Dissimilar metal joining technology is a key process for achieving lightweight structures, and accurate prediction of welding quality is crucial for ensuring structural safety. This study constructs a machine learning framework for predicting void defects in friction stir welding (FSW). Using the FSW process dataset (108 records covering three aluminum alloys: AA2219, AA2024, and AA6061), a heat input index is introduced as a derived feature, and SMOTE is applied to address class imbalance. Seven machine learning models are compared under repeated stratified five-fold cross-validation. The results show that MLP achieves the best AUC value (0.8951), followed closely by XGBoost with 0.8912 and stronger stability. This paper further explores the application of the prediction model in quality control and process optimization in a smart manufacturing environment, providing theoretical and practical references for intelligent decision-making in the welding process.