Machine Learning-Based Prediction of Fatigue Fracture Locations in 7075-T651 Aluminum Alloy Friction Stir Welded Joints

Friction stir welding (FSW) is a solid-state joining technique widely used for aluminum alloys in aerospace, automotive, and shipbuilding applications, yet the prediction of fatigue fracture locations within FSW joints remains challenging for structural-life assessment. In this study, we investigate fatigue fracture location prediction in 7075-T651 aluminum alloy FSW joints by applying four machine learning methods-decision tree, logistic regression, a three-layer back-propagation artificial neural network (BP ANN), and a novel Quadratic Classification Neural Network (QCNN)-using maximum stress, stress amplitude, and stress ratio as input features. Evaluated on an experimental test set of eight loading conditions, the QCNN achieved the highest accuracy of 87.5%, outperforming BP ANN (75%), logistic regression (50%), and decision tree (37.5%). Building on QCNN outputs and incorporating relevant material property parameters, we derive a Regional Fracture Prediction Formula (RFPF) based on a Fourier-series quadratic expansion, enabling the rapid estimation of fracture zones under varying loads. These results demonstrate the QCNN's superior predictive capability and the practical utility of the RFPF framework for the fatigue failure analysis and service-life assessment of FSW structures.