Predicting the effects of microstructure on matrix crack initiation in fiber reinforced ceramic matrix composites via Machine Learning

A reduced-order, data-driven, probabilistic predictive model to quantify damage initiation in continuous SiC ceramic fiber SiC ceramic matrix composites (CMCs) at pertinent lengths scales using machine learning tools is proposed and explored. A novel framework is developed to characterize the influence of key stochastic microstructure attributes on matrix crack initiation. The approach is illustrated for the case of transverse crack initiation in the matrix surrounding fibers oriented perpendicular to the loading direction. A variety of stochastic microstructure attributes were considered including fiber spacing, fiber diameter, and coating thickness. Statistics of a commercial CMC microstructure were digitally represented and used to instantiate microstructures. In addition, discrete digital instantiations generated over a range of the distributed microstructural attributes were considered. The statistics of the distributed microstructure attributes were quantified using n-point statistics and reduced using principal component analysis. The elastic responses of the instantiated microstructures were characterized using finite element analysis (FEA). Results from the FEA were used as the ground truth to calibrate and validate a data-driven machine learning (ML) model. The quantified stochastic microstructure attributes were correlated with the statistics of the simulated damage response. The predictive capabilities of the model for a new microstructure class were demonstrated.