Learning Paris' law by random forest trees

Abstract

Recent rapid development of deep machine applications presents its capability to be applied to the practical engineering problems. One of the most important issue is an appropriateness of the algorithm and application data. Fundamentally deep learning has capability of learning everything including several laws and superposed phenomena. Of course, the simple laws and phenomena can be easily trained than multiplex phenomena. In order to accelerate training process of a neural network, an appropriate machine learning technique should be employed. In this study, Paris’ law with concerns to stress ratio are trained by random forest method. When the machine learning can predict crack propagation rate correctly, a good convergence of a neural network is achieved by the less multiplex phenomena. Random forest techniques also have several hyper parameters to obtain a good prediction. In order to ensure the prediction, we’d like to survey the hyper parameters. We discuss the results of the trained decision tress by the random forest method.