Failure envelope prediction of 2D SiCf/SiC composites based on XGBoost model

Abstract

In this paper, a multiple failure mechanisms-based progressive damage model is developed to capture the mechanical response and defect-induced nonlinear behavior of 2D SiCf/SiC composites. This model is then adopted to simulate the stress–strain response of a representative volume element model under uniaxial loading, showing excellent agreement with experimental results. The effects of the volume fraction of SiC fibers, as well as those of the porosity and microcrack density of SiC matrix, on the deformation and damage behaviors of SiCf/SiC composites are predicted. Based on this, the failure envelopes related to micro-mesoscopic characteristics under combined loadings are generated as the database for the eXtreme Gradient Boosting (XGBoost) model training. Finally, a data-driven failure model is established for 2D SiCf/SiC composites, whose prediction is compared with the formal failure criteria. The results demonstrate that the data-driven 2D SiCf/SiC composite failure model is reliable in constructing the failure criteria related to micro-mesoscopic features.