Machine learning aided multiscale mechanics of fiber suspensions

We present a Fast-Fourier-Transform (FFT) based computational approach to computing the viscous stress response of rigid fibers suspended in a non-Newtonian medium. We identify closed-form models for the fiber suspension viscosity from data obtained with the FFT-based computational approach by leveraging supervised machine learning techniques. Furthermore, we present a novel Deep Material Network architecture capable of treating suspensions of rigid particles with high computational efficiency.