Rapid and accurate prediction of crack growth paths in materials using machine learning can significantly reduce the resource consumption of traditional phase-field simulations and offer a novel solution for risk assessment of engineering structures, with high potential for industrial application.
Two hybrid crack growth path prediction models are proposed, integrating convolutional neural networks (CNN) and recurrent neural networks (RNN). The first is the Densely Connected Convolutional Networks and Gated Recurrent Neural Network combination (DG), while the second is a mixed 3D and 2D convolution model (3-2Dmix). Considering the spatiotemporal continuity of crack growth, CNN and RNN, with their respective spatial and temporal dimensions advantages, are selected for modeling.
Experimental results demonstrate that the DG combination performs better in crack growth prediction, characterized by fast convergence and high accuracy. In contrast, the 3-2Dmix combination struggles with temporal sequence processing, leading to suboptimal prediction accuracy.
The proposed model, combining CNN and RNN, introduces a novel technical framework for crack growth prediction and contributes to the risk assessment of engineering structures.
