The large scatter in high-cycle fatigue (HCF) life poses significant challenges to safe and reliable in-service assessment of additively manufactured metal components. Previous investigations have indicated that inherent manufacturing defects are a critical factor affecting the fatigue performance of the components, and the HCF life is significantly influenced by the geometric parameters of the critical defects inducing crack nucleation. Therefore, it is highly important to elucidate the correlation of the HCF life with the geometric parameters of critical defects. This study proposes a new fatigue life prediction model for laser additively manufactured AlSi10Mg alloys by including the combined effects of loading stress and defect geometries (size, location and morphology) in terms of domain knowledge-guided symbolic regression (SR). Domain knowledge is extracted from the semi-empirical Murakami, Z-parameter and X-parameter fatigue life models to establish the variable subtrees. The results show that compared with these semi-empirical models, the domain knowledge integration-based SR model has higher prediction accuracy and generalization ability. Moreover, compared with traditional 'black box' machine learning models, SR excels at balancing prediction accuracy and model interpretability, which provides useful insights into the relationship between fatigue life and defect geometries. This article is part of the theme issue 'Physics-informed machine learning and its structural integrity applications (Part 2)'.
Keywords: aluminium alloy; fatigue life prediction; internal defect characterization; laser additive manufacturing; physics-informed machine learning.