Novel Radial Basis Function Network Based on Dynamic Time Warping and Kalman Filter for Real-Time Monitoring of Supersonic Inlet Flow Patterns

Abstract

As the basis of protection control, supersonic inlet plays an important role in a supersonic air-breathing propulsion system. To reduce the negative effects of buzz flow on the supersonic inlet and ensure safe and stable operation, it is of great significance to develop methods for monitoring the flow patterns. Traditionally, several manually engineered features are extracted as indicators to evaluate the operation status, but this process can be heavily dependent on professional experience and is time-consuming. This paper proposes a novel network called DTW-RBF-KF which integrates dynamic time warping (DTW) and the Kalman filter (KF) into a radial basis function (RBF) network architecture to directly determine flow patterns from the dynamic sensor signals. The proposed network replaces the Euclidean distance in the static RBF kernels with DTW distance, and exploits the flexible alignment ability of DTW to deal with the temporal distortions. The second-order Levenberg–Marquardt optimization algorithm then is employed to allow the efficient training of the proposed network. To improve the classification performance when the network structure is fixed, the KF is applied as a postprocessing technique to linearly convert the predicted output of RBF into a value closer to the true output. Experimental results demonstrated that the proposed DTW-RBF-KF network has the highest average monitoring accuracy, 94.10%, and requires less calculation compared with all competitive methods. The average time required to test a sample of DTW-RBF-KF network is 1.19 ms, which is shorter than the sample duration, and can be exploited for real-time applications. The addition of a linear KF module further increases the monitoring accuracy to a higher level with negligible computation. The proposed method has better comprehensive performance for monitoring the flow patterns of supersonic inlet in terms of monitoring accuracy and real-time performance.