Experimental study on identifying multiple operating modes of a hypersonic inlet using sparse sensors

Hypersonic inlet unstart poses significant risks to flight safety, necessitating real-time operating mode identification. However, dense sensor arrays increase thermal protection burden and system complexity. To address this issue, wind tunnel experiments were conducted on a hypersonic inlet, and a data-driven framework based on sparse pressure measurements was developed for operating mode identification. Unlike conventional methods relying on predefined mode categories, the present approach first identifies unlabeled clusters from experimental data and assigns operating mode labels through physics-informed interpretation. Isometric Mapping was employed for nonlinear dimensionality reduction, followed by K-means and density-based spatial clustering of applications with noise to reveal intrinsic data structures. Representative samples from each cluster were examined by combining wall-pressure distributions with Schlieren visualizations, enabling mapping to four physically meaningful operating modes reported in previous studies. The ReliefF algorithm was further utilized to identify the most informative sensing locations, enabling sparse sensor deployment. A gradient boosting decision tree classifier achieved nearly 100% accuracy using only two sensors under Mach 4 and 5 conditions, with strong generalization demonstrated across different Mach numbers and sideboard configurations. The results demonstrate that the proposed framework provides a low-complexity, reliable solution for mode diagnosis in hypersonic flight control systems, with significant potential for practical engineering applications.