Stress field prediction in fiber-reinforced composite materials using a deep learning approach

Stress analysis is an important step in the design of material systems, and finite element methods (FEM) are a standard approach performing computational stresses complex systems. The significant cost associated with multi-scale FEM motivates replacement significantly faster data-driven machine learning based approach. In this study, we consider application deep tools to local stress field prediction fiber-reinforced matrix composite system as efficient alternative FEM. first challenge predict maps for cross-sections fixed number fibers varying spatial configurations. Specifically, mapping between arrangement corresponding von Mises achieved by using convolutional neural network (CNN), specifically U-Net architecture. CNN trained data same target A robustness uses different initializations training samples find evolution accuracy increasing samples. Systems larger typically require finer mesh discretization, leading increase cost. Thus, secondary goal here systems CNNs that pretrained on from relatively cheaper smaller fiber number.