Rapid heat transfer simulation of composites curing process based on cGANs and MPGNNs

During the autoclave curing process, non-uniform heating will lead to residual stress and process-induced deformation (PID) in composite parts. To improve the spatial temperature variations, it is necessary to simulate the curing process under various processing conditions to seek the optimal parameters. Given the computational inefficiency of traditional numerical models, we present a novel rapid simulation surrogate model. This model (referred to as HT-GAN) is generative and high-fidelity powered by a conditional Generative Adversarial Network (cGAN). The generator of HT-GAN employs an encoder-propagator-decoder architecture, wherein the propagator utilizes Message Passing Graph Neural Networks (MPGNNs) to accurately simulate the exothermic curing reaction and heat transfer behavior. By comparing the results obtained from the proposed model with those from experimentally validated numerical model simulation, we have found that the proposed model achieves high fidelity in both overall statistics and distribution with respect to temperature. Across a series of test sets, the Mean Absolute Percentage Error (MAPE) between the predicted and ground truth values has consistently remained below 0.40 % throughout the entire curing cycle. Crucially, the model dramatically accelerates the simulation process, reducing computation time to approximately 1/1000th of that required by numerical models. Note: detailed material properties, algorithm implementations, and additional evaluation results are provided in the supplementary material.