A novel approach based on the elastoplastic fatigue damage and machine learning models for life prediction of aerospace alloy parts fabricated by additive manufacturing

(a) Computational flowchart for fatigue life prediction of AM alloy parts. It is seen that the CDM based computational approach is employed to acquire fatigue data to train ML models, and the ANN and RF models are then presented in detail. After that, life prediction and parametric studies are carried out. (b) Comparisons of predicted performances by ML models for AM SS316L. It is observed that the RF model performs better than the ANN model. (c) Variation of predicted fatigue life against experimental data for AM Ti6Al4V by RF. It is clear that all data by RF model lie in the three-error band. (d) Variations of relative error against the number of training data for AM AlSi10Mg. It is seen that the relative error of predicted fatigue life decreases when the number of training data increases from 100 to 500. [Display omitted] •A novel approach is developed for fatigue life prediction of AM aerospace alloys.•The approach effectively combines the elastoplastic fatigue damage and ML models.•ANN and RF models are implemented to predict fatigue lives of AM alloys.•Effects of ML model parameters on predicted fatigue lives are investigated. In the aerospace engineering, many metal parts produced using Additive Manufacturing (AM) technique often bear cyclic loadings, so the fatigue failures of AM alloy parts become very common phenomena. In this work, a new method is proposed to investigate the fatigue damage behavior of AM aerospace alloys, in which the continuum damage mechanics (CDM) theory and machine learning (ML) models are effectively combined. At first, the CDM models with AM effects are theoretically presented, and the fatigue lives are then numerically computed. In total, over 500 sets of data are acquired and employed to train ML models. After that, the two commonly-used ML models including artificial neural network (ANN) and random forest (RF) are implemented to carry out fatigue life prediction. Furthermore, the predicted data are compared with the experimental fatigue life, and the proposed novel method is verified. At last, the parametric studies are discussed to investigate the variation trend of predicted performance and fatigue life with the important parameters of ML models.