Data-driven multi-scale simulations of composite materials failure (Invited keynote)

Because of their inherent computational cost, multi-scale simulations of composite materials are unpractical for complex non-linear material responses.

Data-driven approaches offer the opportunity to address this bottleneck by training surrogate models from a synthetic dataset built using offline simulations. Once trained, the surrogate can substitute the micro-scale BVP resolution of a multi-scale simulation while reducing the computation time. In this work we compare two surrogate models in the case of complex non-linear material responses of composite materials, including failure prediction.

A neural-networks (NNWs) offer a potential of reducing the computation time by more than 5 orders of magnitude. In particular, recurrent neural networks (RNNs) were shown to be efficient and accurate in approximating the history-dependent homogenized stress-strain relationships. However, they have no possibility of extrapolating responses and thus require a large dataset for their training.

A physics informed alternative is the deep material network (DMN) approach which consists in a network of parametrized mechanistic building blocks. During the training process, the DMN “learns” the micro-structure topology from elasticity data. Once trained, the DMN is able to predict nonlinear responses, including for unseen material responses and loading conditions, in a thermodynamically consistent way, ensuring extrapolation capabilities. The DMN is now extended to damage simulations for material failure analysis.