Research on intelligent identification algorithm of flame states to judge flame stabilization in a scramjet combustor

The supersonic combustor is a critical component of a scramjet, and it is essential to monitor the combustion states within the combustor to ensure the safe and stable operation of the scramjet. Hence, we propose an intelligent combustion state recognition model called the parallel Kolmogorov-Arnold transformer (PKAT). In this model, we construct a Kolmogorov-Arnold linear layer based on Kolmogorov-Arnold theory to enhance the model's ability to approximate flame-stable combustion features. To enable different levels of feature recognition through various backbone networks, we build a parallel structure. The PKAT model performs multi-level and comprehensive recognition of flame combustion features, considering both global and local perspectives. To validate the robustness and generalization capability of the PKAT, we collect data on the stable combustion process of a hydrogen-fueled scramjet engine in a direct-connect supersonic pulse combustion wind tunnel. Experimental results show that the proposed model accurately identifies stable combustion states under different operating conditions, achieving precision, recall, and accuracy rates of 98.31%, 98.36%, and 98.76%, respectively. Furthermore, the PKAT model demonstrates a certain level of generalization capability. When compared to traditional convolutional neural network-based and transformer-based models, the PKAT model exhibits higher recognition performance and robustness in accurately identifying combustion characteristics.