Damage evolution analysis in bioinspired composite structures by using a machine learning-based multiscale modeling approach

Nowadays, accurately predicting how advanced bioinspired composite structures fail under various loading conditions remains a complex task. This is primarily because several different damage mechanisms can occur simultaneously at different scales. Among the existing numerical approaches for predicting failure in such structures, multiscale models are the most promising by virtue of their optimal compromise between accuracy and efficiency. This work introduces a cost-effective, machine learning-based multiscale approach for studying how damage evolves in bioinspired periodic composite materials under general loading conditions. A crucial component of this approach is a data-driven surrogate macroscale damage evolution model. This model is developed by performing several nonlinear homogenization steps on the same bioinspired microstructure to be used for training a deep neural network (DNN). The reliability of this multiscale model is assessed by comparing its numerical predictions with the results coming from classical micromechanical approaches, with reference to both proportional and nonproportional loadings.