Machine Learning-Based Prediction of High Cycle Fatigue and Fatigue Crack Growth Rate in LPBF Co-Cr-Mo Alloys Under Varying Scanning Strategies

This study explores the use of machine learning to predict high-cycle fatigue (HCF) behavior and fatigue crack growth rate (FCGR) in Co-Cr-Mo alloys manufactured through laser powder bed fusion. Two machine learning (ML) models: extreme gradient boosting (XGB) and deep neural networks (DNN), are implemented to estimate HCF and FCGR across three distinct scanning strategies. The raw datasets for HCF and FCGR are taken from previously performed experiments. The HCF dataset is augmented using a Gaussian Mixture Model, while the FCGR dataset is used in its raw form. Following hyperparameter optimization, both models exhibited quite similar accuracy on validation datasets. Their performance is assessed during testing using mean squared error (MSE) and R2 scores. The DNN model demonstrated higher accuracy in HCF predictions by achieving higher R2 scores. The DNN performs better because it can handle more complex patterns effectively due to its multiple neurons and deeper multilayer architecture. In contrast, the XGB model performed better in FCGR predictions and yielded higher R2 scores compared to XGB. The good agreement with the experimental dataset shows that these two ML techniques are effective in predicting HCF and FCGR behavior.