Detection of rotating stall of axial compressors based on mixed pattern library

In this paper, a mixed pattern library including normal patterns, spike stall inception patterns and modal stall inception patterns is built to detect the rotating stall in axial compressors. Firstly, the Mansoux-C2 model and the Mansoux-C3 model are used to generate the modal stall and the spike stall, respectively. In order to facilitate modeling and detection of rotating stall, system dynamics of the compressor are divided into the dominant and secondary dynamics. Then, the dominant dynamics are accurately identified and stored in constant radial basis function (RBF) networks based on deterministic learning. The stored knowledge is used to build a mixed pattern library. Based on the library, the rotating stall inception of the test compressor can be rapidly detected by the dynamic pattern recognition method and the smallest residual principle. Simulation studies are given to show the validity of the proposed approach. Simulation results show that the test rotating stall pattern can be rapidly detected without false alarm.