Multiscale Modeling of Metal-Ceramic Spatially Tailored Materials via Gaussian Process Regression and Peridynamics

In this paper, a micro-to-macro multiscale approach with peridynamics is proposed to study metal-ceramic composites. Since the volume fraction varies in the spatial domain, these composites are called spatially tailored materials (STMs). Microstructure uncertainties, including porosity, are considered at the microscale when conducting peridynamic modeling and simulation. The collected dataset is used to train probabilistic machine learning models via Gaussian process regression, which can stochastically predict material properties. The machine learning models play a role in passing the information from the microscale to the macroscale. Then, at the macroscale, peridynamics is employed to study the mechanics of STM structures with various volume fraction distributions.