Reliable generalizable identification method for combustion modes in a scramjet engine

This study introduces a novel intelligent framework for combustion mode identification in a scramjet engine, advancing the integration of engineering informatics into aerospace sciences. By leveraging advanced computational methods, we developed the RepVGG (Visual Geometry Group)-Mamba-UNet model, which reconstructs high-fidelity supersonic flow fields with an MSE (Mean Squared Error) loss of 0.001, enabling precise correlations between pressure fluctuations and turbulent flow structures. The model's dynamic sequence modeling, achieved through innovative Mamba blocks, effectively captures shock-combustion coupling dynamics. A knowledge distillation approach further enhances cross-regime generalization, achieving a Pearson correlation of 0.96 and SSIM (Structural Similarity) of 0.94. The framework incorporates the Wavelet-Distilled Dual-Target Localization Network (WD-Net), which integrates wavelet physical priors with a data-driven architecture for end-to-end combustion mode identification, achieving 94% validation accuracy. This lightweight system, with only 0.04 × 106 parameters, provides real-time diagnostic capabilities critical for supersonic combustor control. The research demonstrates high-fidelity decision-making tools for active combustion control and safety diagnostics, addressing fundamental challenges in supersonic combustion. These advancements contribute to the broader goals of engineering informatics in aerospace engineering.