A Multiscale Graph Generative Adversarial Network Method for Prediction of Autoclave Curing Temperature Field in CFRP component

<title>Abstract</title> <p>During the CFRP component in an autoclave, the spatiotemporal temperature distribution directly affects the forming quality and process efficiency. To address the issues of insufficient multi-scale feature representation and overly smoothed prediction results in existing deep learning surrogate models under unstructured mesh conditions, this paper proposes a rapid prediction method for curing temperature fields based on a multi-scale graph generative adversarial network. Based on the experimentally validated thermo-chemical coupling model of the autoclave curing process, 100 sets of operating condition samples were obtained by sampling around six process parameters, and a spatiotemporal dataset of the whole-process temperature field and degree of cure was constructed. The generator employs two-level CLJP coarsening and graph attention to enhance the representation of temperature-field features at different spatial scales, including global thermal diffusion patterns, local high-gradient features, boundary transition details, and local peak information, while the local graph discriminator alleviates the oversmoothing commonly observed in supervised regression models by imposing realism constraints on hotspot regions and boundary transition zones through k-hop neighborhood-induced subgraphs. The results show that the proposed model outperforms comparative models such as Graph U-Net, Graph U-Net-LSTM, Grid-GAN, and conventional GNN in predicting the cure temperature field, with MAE, RMSE, R², and MAX-Error reaching 1.28 ℃, 1.76 ℃, 0.993, and 3.58 ℃, respectively; in full-cycle autoregressive prediction, the model can reducing computation time by approximately 99% compared to traditional CFD simulations. The study results indicate that this method can provide an efficient and reliable surrogate model for rapid evaluation and intelligent control of the curing process in composite material autoclaves.</p>