Unsupervised clustering method with data fusion of wide-range scramjet combustion mode

Accurately identifying the combustion modes of a scramjet engine and determining the boundaries of mode transitions are crucial for ensuring its safety， reliability， and efficient operation. Based on the concept of data fusion， this paper proposed an unsupervised clustering method for cross-configuration combustion modes identification in scramjet engines. Numerical simulations are implemented to obtain physical data from the upper and lower wall surfaces， as well as flow field density gradient images， for two different configurations： the upper-cavity configuration and the lower-cavity configuration. The two types of data are fused and subjected to unsupervised clustering， and the results are compared with manually classified combustion mode standards and clustering methods without data fusion. The results demonstrate that categorizing the combustion modes into three types—scramjet mode， dual-mode subsonic combustion mode， and deep subsonic combustion mode—is reasonable for two configurations. The proposed method achieves classification accuracies of 92.02% and 92.88% for the two configurations respectively， the NMI and ARI indicators for measuring the quality of classification are both above 0.8， outperforming clustering methods that use only wall surface physical data or density gradient images. This method fully resolves the available data， achieves high combustion mode identification accuracy without requiring labeled data， and enables low-cost， high-efficiency cross-configuration multi-mode combustion identification and mode transition boundary determination.