High cycle fatigue life prediction of titanium alloys based on a novel deep learning approach

Due to the comprehensive influencing factors, accurate fatigue life prediction of materials is still a challenging task. In the present study, a novel deep learning approach named Multi-Graph Attention Networks (Multi-GAT) is proposed to predict the high cycle fatigue (HCF) life of several types of titanium alloys, including TA11, TA12, TA19, TC4, ZTC4 and TC17. An advanced attention mechanism of deep learning is incorporated in the present Multi-GAT approach. The efficacy of this approach has been fully demonstrated through two specific calculational examples. The results show the proposed deep learning approach outperforms the conventional machine learning approach such as Random Forest. Finally, Shapley value is utilized to explore the key factors influencing the HCF life, and the relative importance of these factors is evaluated.