Machine learning approach to predict fatigue crack growth

Prediction of fatigue crack growth is an important requirement during estimation of residual life of machine components or during failure analysis. The existing theoretical models are specialized to predict the crack growth for one of the three stage of classical crack growth diagram. In this work, a novel and unified machine learning based approach has been developed to cover both stage-II and stage-III regions of crack growth rate. Three alternative machine learning algorithms are investigated to identify the most suitable algorithm for prediction of fatigue crack growth rate. The models are trained using experimental data conducted on CT specimens of carbon steel subjected to different types of cyclic loading. The comparison of mean squared error and R2 score in terms of accuracy in percentage obtained from the three models is presented. The guidelines for training and tuning of machine learning models are discussed. (C) 2020 The Authors. Published by Elsevier Ltd.

Automated summary

A novel machine learning approach has been developed to predict fatigue crack growth rate in both stage-II and stage-III regions. Three alternative machine learning algorithms were investigated to identify the most suitable algorithm for prediction. Comparison of accuracy in terms of mean squared error and R2 score from the three models was presented, along with discussion on training and tuning of machine learning models.