Implementation of Physics-Informed Neural Networks in Finite Element Analysis of Biocomposite Materials

The increasing demand for sustainable materials has highlighted the importance of biocomposites; however, their computational analysis is challenging due to the high cost associated with traditional methods such as the Finite Element Method (FEM). This study introduces a computationally efficient approach for analyzing the mechanical behavior of chitosan/nanocellulose biocomposites using Physics-Informed Neural Networks (PINNs). By embedding the governing equations of 2D linear elasticity directly into the network's loss function alongside reference data from FEM simulations, the PINN is trained to predict displacement fields. The methodology involves training the network on data from one microstructure and validating it on five distinct, unseen microstructures. Results demonstrate high predictive accuracy, with an average R ² value exceeding 0.92 across validation sets, and a significant post-training reduction in analysis time, achieving an inference speed-up approximately 5.4 times that of the FEM solver. This work establishes PINNs as a promising tool for accelerating the analysis and development cycle of sustainable biocomposite materials.