Abstract
Axial compressor systems are susceptible to unstable conditions near their optimal operating point. In particular, rotating stall and surge are conditions that need to be avoided during the operation of an axial compressor. In extreme cases these conditions may cause damage to the compressor. The onset of either condition is rather rapid, and usually does not allow for remedial control action based on the limited time available. Hence, research efforts have been focusing on the development of new detection methods that allow for more time to take corrective measures. This paper explores and compares various existing and proposed methods to identify and detect those precursors. The methods detailed in this work are tested at different operating conditions and locations. The methods investigated include the sequentially computed correlation coefficient of pressure sensor data, correlation coefficient, the Generalized Extreme Studentized Deviate Test (ESD) for outlier detection, spectral entropy, and Autoregressive (AR) models. The primary goal of evaluating these methods is based on their suitability for employment as pre-processors for dynamic system identification. By using the dynamics of the identified model rather than a static precursor, it is stipulated that the onset of stall and surge can be managed with a control concept. For this work, the extracted models are investigated for suitability to serve as precursors, and the potential as predictive models. This work may serve for future work to achieve active flow control by direct air injection at the leading edge of the blades. For this work, a one-stage compressor system with a blade geometry that allows for spike inception is utilized. Spike stall inception is a precursor to fully developed rotating stall. The subsonic compressor has 60 blades, and its operating point is controlled by a throttle and constant speed control of the rotor. The pressure data is collected with 10 Kulite™ sensors which are placed along the blade cord length on the outer casing of the compressor. The results of the tabulated performance of the various methods and resulting models indicate that an ARESD combination yields the earliest indication for spike stall inception.