Key points are not available for this paper at this time.
Abstract With the increasing interest in utilizing additive manufacturing (AM) for fatigue-critical applications, reliable fatigue analysis tools are increasingly necessary. Unlike traditional manufacturing processes, AM involves a greater number of process parameters and manufacturing variables which adds complexity to the development of mechanistic fatigue models. Data-driven models, to some extent, have shown predictive capabilities for the fatigue properties of AM alloys. However, no study has extensively investigated fatigue properties across different AM alloys with the aim of developing predictive models for fatigue databases. A data-informed knowledge discovery framework is developed for exploring diverse AM fatigue datasets in the literature. Characteristics of the AM fatigue datasets are first analyzed within this framework to uncover data-driven insights for machine learning modeling. Machine learning techniques are then used to solve classification and regression problems using fatigue ratio as the property of interest. k-nearest neighbors and support vector machine classifiers are used to classify process parameters and manufacturing variables to determine the class of fatigue ratio they belong to based on a pre-defined threshold. Similarly, a Gaussian process regression (GPR) model is used to predict the fatigue ratio of AM alloys with a 95% prediction interval. The Shapley Additive exPlanations (SHAP) method is used to explain the predictions of the GPR model. This framework is validated using Ti–6Al–4V, IN718, and AlSi10Mg datasets from the literature. The proposed framework efficiently maps process parameters, post-processing sequences, and AM types to the fatigue ratio, thereby providing an efficient surrogate modeling methodology for estimating fatigue properties from available datasets.
Alfred et al. (Sat,) studied this question.