Fatigue life prediction of selective laser melted titanium alloy based on a machine learning approach

Abstract

A machine learning (ML) approach is introduced to predict the high-cycle fatigue (HCF) life of selective laser melted (SLM) TA15 titanium alloy, addressing life prediction variability caused by defect characteristics and spatial distribution. Using HCF data, tensile properties, and defect characteristics across different building directions (BD), a training dataset was established. Comparative analysis shows that incorporating defect parameters significantly enhances the prediction accuracy of the ML model. Correlation analysis identified A_defect/h as highly relevant to fatigue life, enabling a refined training dataset. Incorporating this defect parameter significantly improved the ML model’s prediction accuracy. The S-N curve generated from predictions using defect values at 50 % reliability appeared relatively conservative compared to the experimental S-N median curve. The S-N curve at ± 3σ reliability closely aligned with experimental results, encompassing nearly all data points. This highlights the potential of the ML approach in predicting fatigue life for SLM titanium alloys.