Identification and rapid detection of rotating stall via deterministic learning

Rotating stall is a kind of unsteady flow in axial compressors which significantly reduce the performance of turbofan engines. Identification and rapid detection of rotating stall is a very important issue. In this paper, based on the high-order Moore-Greitzer model (the Mansoux model) which captures the transient characteristic of rotating stall, we firstly analyze the properties of the model including the partial diagonal dominance of system dynamics. By using these properties, the locally diagonally dominant system dynamics is identified and stored by RBF networks based on deterministic learning theory. Secondly, the stored knowledge of system dynamics is utilized to achieve rapid detection for rotating stall. Simulation results are included which show that the proposed approach may be useful in modelling and detection of rotating stall and surge.