Integration of Physics-Informed NeuralNetworks and Finite Element Method forAccelerated Thermal Analysis in Multi-LayeredCarbon Fiber Reinforced Polymer Composites

الملخص

The design and structural integrity of Carbon Fiber Reinforced Polymer (CFRP) composites are heavily governed
by their thermal response under operational loads. However, predicting heat transfer in these anisotropic, multi -layered
materials using conventional Finite Element Method (FEM) simulations is computationally expensive, particularly for high-
fidelity cross-ply architectures. This study proposes a hybrid computational framework that integrates Physics -Informed
Neural Networks (PINNs) with legacy FEM solvers to accelerate the prediction of transient heat transfer in composite
laminates. A dataset of 5,000 thermal profiles was generated via traditional FEM under varying thermal boundary conditions
(293 K to 473 K) and fiber orientations. We employed a deep residual learning architecture that incorporates the heat
conduction PDE as a soft constraint within the loss function, ensuring the AI model adheres to thermodynamic laws. Results
indicate that the PINN-FEM hybrid model achieves a Mean Absolute Percentage Error (MAPE) of less than 1.4% compared
to gold-standard numerical solutions. Furthermore, the inference time for a 3D thermal gradient map was reduced by a
factor of 85x compared to iterative FEM solvers. Quantitative analysis demonstrates that the model accurately captures the
interlaminar thermal resistance and the effects of transverse isotropy. This research provides a scalable solution for real -
time thermal monitoring and rapid prototyping of aerospace composite components, bridging the gap betwe en rigorous
numerical accuracy and real-time computational efficiency.