Determination of Fatigue Crack Growth Rate Behavior in Metallic alloys subjected to Spectrum Loading using FCGR-Net Machine Learning Model

Fracture mechanics design principles incorporate the damage tolerance (DT) approach for a safe and extended operational life of engineering structures subjected to variable amplitude or spectrum loading. Numerical models used to predict the FCGR under spectrum loading require large experimental data, idealization of conditions, and formulation of various crack driving force parameters. This has rendered all the models useful but not universally applicable. The current study uses FCGR-Net - a machine learning (ML) model to predict fatigue crack growth under spectrum loading conditions. The performance of the FCGR-Net model is investigated on nickel based super-alloys, GTM 718 and 720, subjected to TURBISTAN (turbine standard) spectrum load sequence. It was observed that the predictions of the model, trained over constant amplitude FCGR data, were highly conservative while, the same model trained over the variable amplitude FCGR data resulted in a quite accurate prediction. The difference (domain shift) in the constant amplitude trained FCGR-Net predictions and experimental data was effectively addressed by introducing a ‘delay factor (Df)’ in the FCGR-Net ML model. The results of the predictions were validated with the popular Kujawski's K* model and experimental data. The current work is a proof-of-concept that demonstrates a promising alternative for predicting FCGR behavior under variable amplitude loading (VAL) or spectrum loading using ML techniques. The accuracy of these ML models will continuously improve with increased application and the availability of experimental data.