Experimental Validation of Multiscale Stochastic Models for Composites

Abstract

In this work, we address several issues that influence the accuracy and reliability of model-based predictions for composite materials systems. During this process, we pursue data-driven stochastic modeling and simulation of complex composite materials across a hierarchy of length scales. The key feature of our approach is the utilization of polynomial chaos expansion (PCE) within a hierarchical multiscale setup to carry out stochastic data-driven model characterization, model updating, and model validation. The most prominent aspect of PCE-based modeling is coupling the key descriptors of properties and behaviors, featuring the different length scales, explicitly with respect to fundamental uncertainties across these length scales. A variety of central and consequential implications of PCE-based modeling within the context of some pressing complexities are discussed and demonstrated in this paper. Among these central features, we specifically highlight: (i) the explicit mapping among the constitutive properties of composites at different length scales with physics-constraint data models, (ii) the efficient integration of a matrix of different data sets featuring multiscale properties , (iii) the construction of explicit reduced-order surrogates of intricate dependent and coupled structure of quantities of interests, (vi) the versatility of PCE models to perform model calibration, efficient simulation of multiscale quantities of interests, efficient quantification of uncertainties in predictions, model updating as new data sets become available.