Stochastic multiscale approach to predict failure initiation and progression in composite materials

Presence of large variability in composite properties have resulted in overdesign of composites offsetting their light weight advantage. While such variability can be captured by experimental testing, the prediction of such variability using virtual testing remain a challenge. The variability in transverse composite properties was determined by finite element analysis of computer simulated microstructures. Such microstructures were generated based on the statistics provided by image analysis of actual microstructures. The generated microstructures were modified to match both short and large length scale statistics of actual microstructures. This will enable generation of theoretically infinite realizations of microstructures that are statistically the same but stochastically different (have the same statistics but different configurations). Image-based finite element models were developed based on both pixel-based and morphology-based meshing. The extended finite element method was implemented in ABAQUS to predict failure initiation and progression. The results show that different realizations of microstructures have different transverse strengths but similar elastic properties.