Prediction of Stress Fields in Particulate Polymer Composites Using Micromechanics-Based Artificial Intelligence Model

Abstract

Particulate polymer composites (PPC) are widely used in various engineering fields for their high strength-to-weight ratio and impressive mechanical properties. Typically, methods for predicting the mechanical behavior of such materials include tensile tests, finite element simulations, and numerical analysis. However, recent advances in artificial intelligence (AI) have enabled improved prediction of mechanical behavior of various materials. In AI-based approaches, microstructural information like fiber orientation and grain morphology are generally defined as inputs through multi-dimensional images. The objective of the proposed investigation is to reduce the prediction complexity and computational efficiency in AI-based methods when compared with finite element modeling (FEM). AI-based algorithms are typically data-driven approaches primarily dependent on the input data quality. Subsequently, the optimal selection of the input labels (material properties using FEM software) is imperative to ensure higher prediction accuracy. In this study, we predict the mechanical behavior of a particulate polymer composite based on the images of the stress fields developed from FEM simulations used to train a paired image-to-image translation model (pix2pix). The pix2pix model is based on a conditional Generative Adversarial Network (cGAN), where we train the encoder by 512 × 512 pixel images corresponding to stresses in the y-direction. The results of the AI algorithm show that the pix2pix model is computationally efficient and highly accurate in predicting the effective stress fields of a detailed representative area element of FEM. We observed the maximum accuracy determined by the correlation coefficient as 0.906 at 20,000 iterations of the algorithm.