Explainable machine learning-based fatigue assessment of 316L stainless steel fabricated by laser-powder bed fusion

Abstract

Additive manufacturing (AM) and in particular laser-powder bed fusion has become a popular manufacturing techniques in recent years due to its significant advantages; however, the mechanical behavior of AM components often varies from components fabricated using conventional processes. For example, the fatigue behavior of components made by AM processes is heavily influenced by process-related defects and residual stresses in addition to applied stress amplitudes, stress ratio and surface conditions. Accounting for the interaction of these effects in fatigue design is difficult by means of traditional fatigue assessment concepts. Machine learning algorithms offer a possibility to account for such interactions and are easily applied once trained and validated. In this study, machine learning algorithms based on gradient boosted trees with the SHapley Additive exPlanation framework are used to predict defect location and fatigue life of additive manufactured AISI 316L specimens in as-built and post-treated manufacturing states, while also facilitating the understanding of the importance and interactions of various influencing factors.