Multiscale computational homogenization of woven composites from microscale to mesoscale using data-driven self-consistent clustering analysis

Compared with the traditional phenomenological method, the multiscale simulation has significant advantages. This paper presents a methodology and computational homogenization framework to predict the macroscale behavior of woven composites based on fiber and matrix in microscale. The major challenge conducting multiscale analysis is the huge computational cost. To improve the efficiency, one of the reduced order models, which is called the data-driven self-consistent clustering analysis (SCA), is introduced and the multiscale framework is proposed by integrating two SCA solvers from different scales. The macroscale performance of 4-H satin weave carbon/carbon composites is investigated using the proposed framework. In order to reconstruct a real microstructure representative volume element (RVE), both microscale and mesoscale architectures are observed using scanning electron microscopy (SEM) and optical microscopy, and statistical geometry features are obtained. In addition, the SCA method is also verified by comparing the results with the finite element method (FEM). The uniaxial tension process is simulated using the multiscale approach, and strain/stress fields in both mesoscale and microscale can be captured simultaneously. Moreover, the uniaxial tensile experiments are also carried out to validate this framework, which shows high efficiency and great accuracy.