Inlet Fault Diagnosis Based on Attention Mechanism Feature Fusion

To tackle the instability fault diagnosis challenges in wide-speed-range supersonic inlets， this study proposes an inlet fault decision fusion diagnosis algorithm based on attention mechanism feature fusion， achieving efficient diagnosis of instability faults across wide-speed regimes. First， considering the requirement for wall pressure data extraction in mathematical modeling of wide-speed-range inlets， a supersonic inlet reference model is established for computational fluid dynamics （CFD） simulations. Second， leveraging data-driven modeling techniques and support vector machine （SVM） algorithms， a high-precision mathematical model covering wide-speed domains and incorporating instability mechanisms is rapidly developed using CFD-derived inlet wall pressure data. Subsequently， an inlet fault decision fusion diagnosis method is proposed. Pressure features are fused via attention mechanisms， followed by Dempster-Shafer （D-S） evidence theory-based decision fusion， which integrates advantages of multiple intelligent algorithms to overcome the limitations of single-signal diagnosis methods （low accuracy and constrained optimization potential）. The simulation results demonstrate the effectiveness of the data-driven wide-speed-range inlet model in achieving high precision and rapid convergence. In addition， the fusion diagnosis algorithm has been shown to attain over 95% accuracy in the detection of instability， indicating an improvement of more than 5% compared to the accuracy of other single fault diagnosis algorithms. This enhancement effectively eliminates the occurrence of missed or false diagnoses， while demonstrates robust performance under operational uncertainties.