The potency of defects on fatigue of additively manufactured metals

• Published in: International journal of mechanical sciences 2022-05, Vol.221, p.107185, Article 107185
• Main Authors: Peng, Xin, Wu, Shengchuan, Qian, Weijian, Bao, Jianguang, Hu, Yanan, Zhan, Zhixin, Guo, Guangping, Withers, Philip J.
• Format: Article
• Language: English
• Subjects: Additive manufacturing; Aluminum alloys; Defect characterization; High cycle fatigue life; Machine learning model
• ISSN: 0020-7403; 1879-2162
• DOI: 10.1016/j.ijmecsci.2022.107185

•Based on the critical defects data measured by SEM, a machine learning model is established to predict the fatigue life, and the model is in good agreement with the experimental data.•Use the trained machine learning model to quantitatively analyze the importance of critical defects characteristics on fatigue life and analyze the effect of aspect ratio on fatigue life.•The ML model suggests that circular defects of the same area to be less potent than elongated defects of the same area in terms of the HCF lifetime. Given their preponderance and propensity to initiate fatigue cracks, understanding the effect of processing defects on fatigue life is a significant step towards the wider application of additively manufactured (AM) parts. Here a novel machine learning (ML) based approach has been developed to predict the fatigue life of laser powder bed fused AlSi10Mg alloy. The four most important parameters, treferred to here as the Wu-Withers parameters, were found to be the applied stress and the projected area, location and morphology of the critical defects. It was found that an Extreme Gradient Boosting model was able to predict the fatigue lives with high accuracy with the importance of these characteristics in limiting fatigue life ranked in the order given above. The model was able to predict the very different lives of samples tested parallel and perpendicular to the build direction in terms of these four W-W parameters indicating that microstructure was of minor importance. In particular the large projected area of the defects on the crack plane when testing parallel to the build direction was found to be primarily responsible for the shorter lives observed for this testing orientation. The fatigue lives were adequately predicted by the more general two variable (stress and projected area) Murakami model, and even more closely predicted by an empirical model (using essentially the same four W-W parameters) for which the ML model corroborated the empirical dependences. [Display omitted]