Compressor Surge Detection Based on Online Learning

Compressor surges may destroy the structure in a very short time and cause serious disasters. An online detection system is necessary for ensuring safety of a compressor. In this paper, a collection of experimental data acquired in a multistage axial compressor while operated in surge is presented. The correlation integral algorithm is investigated. It is show that when the compressor operates in steady state, the correlation integral value keeps on a high level, while in surge, it drop down rapidly. The reference distance is analyzed, which can be chose with the statistical parameters of the reconstructed phase space. However, the reference distance can not be set as a constant. In this paper, we propose an online learning and detecting system. In this system, the reference distance is modified by learning the nearest data. Through comparing current value of correlation integral with the pre set threshold, the system determines output a surge warning signal or not. It is demonstrated that the system can detect the compressor surge very quickly and exactly.