Learning the Stress-Strain Fields in Digital Composites Using Fourier Neural Operator

The search for high-performance materials has led to advanced composite materials with hierarchical designs. However, designing a superior material with targeted properties and performance is challenging due to the vast design combinations and computational limitations imposed by conventional physics-based solvers. In this study, we use a neural operator-based framework, namely, Fourier Neural Operator (FNO) to learn the mechanical response of 2D digital composites. By just providing the material microstructure to the FNO it predicts the complete stress and strain tensor fields. The model exhibits zero-shot generalization to unseen arbitrary geometries. Besides, the model demonstrates zero-shot super-resolution capabilities by predicting high-resolution stress and strain fields from low-resolution input configuration images. We show that FNO works for generalized boundary conditions (BCs) by testing the model for different unseen BCs. Furthermore, FNO provides high-accuracy predictions of equivalent stress and von-mises stresses thereby allowing upscaling of the results.