Crack extension analysis of ship structure based on extended finite element method and deep learning

The weld toe of a ship cross joint, as a stress concentration area, is subjected to loads for a long time, which is prone to trigger crack expansion and threaten the safety of the structure. In this paper, we first use the extended finite element method (XFEM) to simulate the crack extension behavior, calculate the value of the stress intensity factor (SIF), and compare it with the theoretical solution to verify the reliability of XFEM. Subsequently, XFEM simulations were performed to analyze crack propagation behavior in ship-welded cruciform joints containing pre-existing cracks, and the SIF values were calculated for different plate thicknesses and crack lengths. By using the obtained SIF results as training data, a hybrid neural architecture integrating a multilayer perceptron with physics-informed neural networks was developed to accelerate SIF computations. The prediction errors of the model are all within 2.5%, which is a good value for engineering applications.