Defect criticality analysis on fatigue life of L-PBF 17-4 PH stainless steel via machine learning

•Extract critical defect features from fractography of L-PBF parts by computer vision.•Develop a machine learning framework to correlate defect features and fatigue life.•Generate interpretable insights about defect-fatigue life correlations in L-PBF. Defects innate in additively manufactured components may lead to inferior and more scatter in fatigue lives, thus challenging the qualification of these components in fatigue-critical applications. This work seeks to correlate geometrical features of critical defects measured from fracture surfaces to the fatigue performance of laser beam powder bed fusion (L-PBF) components with machine learning and to develop an integrated data-driven analytical framework for defect criticality (IDADC). IDADC has the potential to enhance the understanding of defect-fatigue relationships in a data-driven fashion. The results show that the obtained relationships between the extracted size-related and morphology-related defect features and the fatigue life of L-PBF specimens align with the known fatigue mechanisms and influencing factors. Furthermore, the proposed IDADC framework can model the relationships between defect features and fatigue life with a low mean absolute percentage error of 0.101 using a kernel support vector regression (SVR). This work could establish the algorithmic foundation for nondestructive fatigue evaluation of additive manufacturing products from various facets of critical defects in the future.