Machine learning in multiscale modeling of spatially tailored materials with microstructure uncertainties

•A novel approach to message passing from microscale to macroscale by machine learning.•A new multiscale modeling of spatially tailored materials.•Considering microstructure uncertainties.•A neural network-based classification model to predict the material failure probability. In this paper, a novel hierarchical micro–macro multiscale modeling enhanced via machine learning is proposed. Machine learning plays an important role in this multiscale framework to pass the information from the microscale to the macroscale. This multiscale method provides a new approach to study the mechanics of a metal-ceramic (Ti-TiB2) spatially tailored material in which the volume fractions vary in space at the macroscale. Data sets, collected from microscale simulations, are used to train machine learning regression and classification models. Those predictive models are then implemented in the macroscale simulations to study dynamical responses of spatially tailored Ti-TiB2 structures under various loading conditions. As a difference from other reported works, microstructure uncertainties are considered in this paper so that an artificial neural network is trained as the machine learning classification model to predict the failure probability at the macroscale, which depends on the volume fraction and the deformation (i.e., the strain).