Train and test data from EBSD observation from A machine learning study on the fatigue crack path of short crack on an α titanium alloy

In the present study, a physics-informed neural network model based on Bayesian hyperparameter optimization is proposed for the prediction of short crack growth paths. A large number of cyclic loadings at a lower amplitude were applied to an α titanium sample by ultrasonic fatigue machine to ensure a sufficient amount of data for machine learning. The grain size, grain orientation, grain boundary direction on the path, as well as crack growth direction, were selected as feature data for training the prediction model. The optimizations of the size ratio and the angle operation were conducted to compare different data processing methods, respectively. After evaluation, eventually, a model for predicting crack growth path is obtained with a reliable performance of 10% tolerance on the path angle at each grain boundary. And the prediction effect of the proposed model is better than that of some classic machine learning models and slip trace analysis.This article is part of the theme issue 'Physics-informed machine learning and its structural integrity applications (Part 1)'.