Data-driven fatigue crack propagation and life prediction of tubular T-joint: A fracture mechanics based machine learning surrogate model

This study establishes a data-driven surrogate model within a machine learning (ML) framework for the rapid and accurate prediction of crack extension paths and fatigue life in tubular T-joints. A data-driven program is developed by integrating principal components analysis (PCA), fatigue crack propagation theory and multilayer perceptron neural network (MLP NN). The dataset of ML model is established by co-simulation of ABAQUS and FRANC3D, and the data dimension is reduced by data reconstruction and PCA to improve the computational efficiency. The stress intensity factor (SIF) is predicted by MLP NN model along with the crack front to drive 3D cracks propagation automatically, which enhances calculation efficiency significantly compared with finite element method. The prediction performance of MLP NN and other three ML models are compared, which demonstrates that the prediction accuracy of the framework built by MLP NN model is higher. The data-driven surrogate model is applied to predict the SIF, propagation path and fatigue life of cracks with different depth-to-length (c/a) ratios and initial positions in tubular T-joints, which confirms its excellent accuracy and robustness in terms of the fatigue crack propagation analysis.