Page 16 - IJAMD-2-3

International Journal of AI for
Materials and Design AI applications in composite materials

A

B C

Figure 6. Structure of PINNs and research utilizing PINNs in the field of composite materials. (A) Structure of PINNs; (B) PINN framework and prediction
results for estimating FVF distribution and permeability during injection-based measurements; (C) multi-fidelity PINN framework for predicting defect
89
positions in composites. 90
Abbreviations: FE: Feature engineering; FNN: Forward neural network; FVF: Formation void fraction; MSE: Mean squared error; NN: Neural network;
PDE: Partial differential equation; PINN: Physics-informed neural networks.

transfer. In addition, PINNs do not require meshing of the for their mechanical and thermal stability. Compared to
92
solution domain, and parameterized PDEs can be solved by traditional ANNs, the PINN model exhibited significantly
including the process parameters directly as inputs to the improved predictions of the elastic modulus and tensile
neural network. Thus, PINNs can be utilized as accurate, strength. This improvement highlights the PINN’s
time-efficient, data-free surrogate models for rapid process advantage in incorporating physical constraints into the
optimization. learning process to enhance the model’s predictive accuracy.
Several studies have successfully applied PINNs Moreover, PINNs have also been applied to predict
for analyzing composite materials. 89-95 Using PINNs, property degradation in carbon fiber/epoxy composites
researchers successfully predicted the composition- exposed to high-intensity radiative environments, such as
property relationships of basalt fibers, which are known weapon-induced explosions and solar flux. The model
95

Volume 2 Issue 3 (2025) 10 doi: 10.36922/IJAMD025210016