Evaluation of defect-related fatigue performance of additive manufacturing GH4169 via small punch test

Abstract

Suffering from irregularly distributed defects, the fatigue performance of additive manufacturing (AM) GH4169 was scattered and difficult to evaluate. Therefore, the micro-scale small punch fatigue (SPF) test was employed on the deposited and heat-treated GH4169. Convolutional neural networks (CNN) and image processing technology (IPT) were used to extract defect features of SPF samples, respectively. The effects of AM defects on SPF life were analyzed via mathematical models, and the results showed that defect size damaged SPF life the most. Various machine learning (ML) algorithms, including regression based on deep learning (RDL), support vector regression (SVR), and random forest (RF) were employed to anticipate SPF life. Combined with defects features from CNN and IPT, the SVR model showed high prediction accuracy with a coefficient of determination of 0.98, compared with the classic model and other ML models.