Wavelet Features and Hidden Markov Model-Based Aerodynamic Instability Detection for Compressors

Abstract A reliable technique is introduced to detect aerodynamic instability of compressors based on wavelet features and hidden Markov model (HMM). A single sensor is sufficient for stall warning if the position of the sensor is carefully selected. The method involves obtaining high-response pressure signal near the rotor tip close to the leading edge. Rotating instabilities band wavelet features are then extracted and trained for the HMM; using data under normal operating conditions, the performance index (PI) is calculated. Unsteady behaviors in prestall processes are discussed and casing wall pressure maps are implemented to explore the mechanism of tip leakage vortex (TLV), which are helpful in explaining the various PI results from different feature selections and probe locations. Experimental results show that the trend indices of PI suitably characterize the compressor aerodynamic instability in subsonic operation.