Three Dimensional Crack Propagation Prediction Model Using Transfer Learning with Equation Discovery

In this study, the Predictor trained on 2D crack growth simulation results was used to apply transfer learning by adding input that 3D crack growth simulation results. The dataset used for training was obtained from the results of a crack propagation analysis using the s-version FEM. The input data are the crack tip coordinates, and the predicted output is the amount of crack growth per vector. The training data was augmented based on histograms to account for predicting parameter bias. The output of incremental crack growth values were used to reproduce the crack propagation and to make a continuous prediction. The crack tip can be predicted with an accuracy comparable to previous studies. By applying transfer learning, we were able to reduce the training data to 1/10 of the previous study and the training time to 1/3 of the previous study. We would like to discuss the effectiveness of transfer learning in surrogate models.