Abstract
In this study, the accuracy of plural crack propagation prediction was improved by machine learning with consideration of physical quantities using a small data set. A dataset is obtained from the results of crack propagation analyses using s-version FEM combined with an automatic mesh generation technique. The input parameters are the coordinates of the four crack tips. The output values to be predicted are crack propagation vectors and a number of crack propagation cycles of 0.25mm length. Crack propagation paths and rates are predicted with less 1/300 error than the previous study. The output parameters, number of cycles, and crack growth vector, are different in the way variation and the magnitude of the values. Individual networks for each value to be predicted can keep a uniform distribution of activations in each layer. In this way, problems such as gradient loss and limited expressiveness can be avoided. As a result, it is possible to predict the shape and a rate of crack propagation within 0.07 percent relative error.
