Fatigue strength prediction of Cobalt alloys using material composition and monotonic properties: ML-based approach

Abstract Predicting the fatigue strength (endurance limit) of Cobalt-based alloys is critical for their application in industries such as aerospace and biomedical engineering. Traditional stress-life (S-N) approach is resource-intensive, requiring extensive experimental testing. In this study, we propose a machine learning (ML) based approach as an alternative to predict fatigue strength based on the published data available on material composition and monotonic loading properties (yield strength and ultimate tensile strength). Eight different ML models were developed and tested, including Linear Regression, Lasso Regression, Ridge Regression, Random Forest Method, Support Vector Regression, Gradient Boosting, XGBoost, and Artificial Neural Networks (ANN). Among these, the ANN model demonstrated the best performance, achieving an R 2 score of 0.89, which indicates an ability to capture nonlinear relationships between material properties and fatigue strength. The reliability of the ANN model was validated using an external dataset from the literature, where it maintained reasonable accuracy with prediction errors of 15.5 ± 10.2%. These findings suggest that the proposed data-driven ML approach can serve as a fast and cost-effective alternative to experimental fatigue testing, enabling early-stage material screening and optimization in engineering design. By integrating ML-based fatigue strength prediction into material selection workflows, engineers can enhance design efficiency, reduce testing costs, and accelerate the development of Cobalt alloys for critical applications under cyclic loading conditions.