Flame Evolution Prediction in Scramjet Combustors Based on Graph Convolutional Neural Networks

The chemiluminescence signal of the flame in a scramjet combustor serves as a core representation of flow field characteristics and reaction kinetics. Accurate prediction of this signal facilitates early identification of combustion states and flow control, thereby providing crucial support for enhancing engine efficiency and stability. In this study, a Condition-Aware Graph-to-Image Network (CAGI-Net) is proposed to achieve efficient prediction of chemiluminescence images in scramjet combustors. In this model, 32 pressure sensors distributed on the upper and lower walls of the combustor are abstracted as graph-structured nodes. A Graph Neural Network (GNN) encoder is utilized to extract node features and generate graph-level representations, which are then fused with operating parameters such as injection pressure. The fused features are further decoded by an image decoder to generate single-channel chemiluminescence images with a resolution of 66 × 256 pixels. Experiments were conducted on datasets covering six hydrogen injection pressure conditions, with two test scenarios—interpolation (3 MPa) and extrapolation (4.5 MPa)—employed for validation. The results demonstrate that the model exhibits excellent performance under the interpolation scenario: the average Peak Signal-to-Noise Ratio (PSNR) reaches 28.80 dB, the Structural Similarity Index (SSIM) is 0.89, and the correlation coefficient is 0.92, with the optimal PSNR and SSIM reaching 31.45 dB and 0.94 respectively. Under the extrapolation scenario, the model still maintains reliable generalization capability, achieving an average PSNR of 20.17 dB, SSIM of 0.70, and correlation coefficient of 0.89. This study confirms that the proposed method can effectively capture the main flame structure and brightness distribution characteristics, providing an efficient and intelligent technical approach for combustion diagnosis and performance prediction of scramjet combustors.