Abstract: A general algorithm based on D-S (Dempster-Shafer) evidence fusion is proposed to detect complicated aerodynamic instability phenomena in high-speed multi-stage axial flow compressors. The short-term energy is adopted as a feature in time domain to denote the amplitude characteristic of dynamic pressures in the procedure of aerodynamic instability of compressors，while the spectral correlation coefficient is selected to index the variability of spectra. These features are treated as two evidences of instability phenomena. The mass functions are designed regarding to their statistical distributions. The mass function of the joined evidences is calculated with the Dempster combination rule. Finally，the decision of compressor status such as steady，stall，surge or instable is made by comparing mass values of single evidences or joined evidence with their thresholds respectively. Furthermore，this method is extended to a multi-sensor fusion model of aerodynamic instability detection. Characterized by low computational requirements，this method is suitable for an online detection system. With data acquired during physical compressor experiments on a test rig，it is illustrated that alarms of aerodynamic instabilities could be sent out within several milliseconds.

Key words: Compressor；Stall；Surge；Detection；Information fusion；Correlation coefficient

摘要： 为了研究高速多级轴流压气机中复杂气动失稳现象的检测问题，提出了一种基于D-S (Dempster-Shafer) 证据融合的通用型检测算法。在时域中，采用短时能量表达气动失稳过程中动态压力的脉动幅值特征；在频域中，选择频谱相关系数表达信号频谱变化的特征。将这两种信号特征分别作为气动失稳现象的证据。根据统计规律设计了各证据的mass分配函数。采用Dempster合成规则计算联合证据的mass函数。通过分别比较单一证据及联合证据的mass函数值与其门限值，判决压气机处于气动失稳、失速、喘振或正常状态。进一步地，将该方法扩展为气动失稳检测的多传感器融合模型。该方法计算量小，适用于在线检测系统。采用压气机台架试验实测数据验证，可在气动失稳数毫秒内发出报警信号。

关键词: 压气机；失速；喘振；检测；信息融合；相关系数