Micro-mechanics and data-driven based reduced order models for multi-scale analyses of woven composites
Ling Wu, Laurent Adam, Ludovic Noels
Abstract
Order reduction of woven composite materials is based on the definition of short fibre reinforced matrix material pseudo-grains completed by pure matrix parts. The former ones model the curved yarns, which are assimilated to continuous fibre reinforced matrix materials, in woven composites, and the latter ones model the matrix response. The homogenisation is achieved by recursively using micro-mechanics models, such as mean-field homogenisation and Voigt’s rule of mixtures, and the laminate theory. The pseudo-grains number and micro-structural features such as orientation, aspect ratio and volume fraction are considered as the Reduced Order Model (ROM) parameters and are identified following the approach of Deep Material Network (DMN): a set of homogenised elasticity tensors evaluated by computational homogenisation of woven unit cells is used as training data in order to identify the topological parameters of the ROM. Once the topological parameters are identified, the proposed ROM can be used to conduct nonlinear analyses of woven composites. The accuracy and efficiency of the proposed ROM have been verified by comparing the predictions with direct numerical simulations on two different woven unit cells.