FCGR-Net: A Novel Approach to Predict Fatigue Crack Growth Rate Behavior in Metals Using Machine Learning

Abstract Accurate prediction of fatigue crack growth rate (FCGR) behavior in structural alloys using numerical methods is still proven to be challenging due to the complexity of mechanisms involved. One other option is to make use of the state-of-the-art research involving machine learning (ML) for predictions. ML-based artificial intelligence (AI) is evolving as an alternate approach towards automation with higher levels of accuracy across diverse disciplines. As ML demands huge data, an attempt is made to create a database involving material composition, tensile properties, temperature, stress ratio, and corresponding FCGR information. The dataset covers a wide gamut of structural alloys such as steel-, aluminum-, and nickel-based superalloys. The ML model trained over this experimental dataset was used to predict the FCGR in a structural alloy and is the novelty of the work. The methodology involves formulation of deep neural network architecture, herewith termed as FCGR-Net model, for the prediction of fatigue crack growth (FCG) behavior. FCGR-Net predictions show good correlation with the experimental values available in the literature. This study would significantly contribute toward early prediction and minimizing experimentation without compromising the accuracy of prediction.