Fracture pattern prediction with random microstructure using a physics-informed deep neural networks

Material fracture is a process involving both linear elastic stage and nonlinear crack propagation stage. The problem becomes even complex when the material random microstructure is considered. The computational cost of material fracture simulation is high especially for large-scale nonlinear simulation. Recently, deep learning models have demonstrated the power for efficient nonlinear simulation. In this paper, a physics-informed deep learning model is proposed that a neural network model is integrated with a discrete simulation model (lattice particle method –LPM) to predict material fracture patterns for arbitrary material microstructures under different loadings. The key idea is to leverage physics-knowledge and data-driven approach for accurate and efficient nonlinear mapping. Physics-knowledge includes constraints, microstructure images, and displacement field from pure linear elastic analysis. Fully Convolutional Network is then used to predict the final fracture patterns. F1 score is used for model performance evaluation. The accuracy of the computational framework is verified for arbitrary material microstructure and arbitrary loadings by comparing predicted results and ground truths. One significant benefit of the proposed method is that high computational efficiency for the nonlinear material response prediction. Meanwhile, it is demonstrated that the proposed physics-informed model requires much less training data than purely data-driven models. Several future research directions are suggested based on the proposed study.