基于SFA-LOF-LDD的航空发动机异常检测方法研究  

Research on Abnormal Detection Method of Aero Engine Based on SFA-LOF-LDD

作　　者：李泱 李德文 蔡景[2] 左洪福[2] 张营[1,2] 韩辰球 LI Yang;LI Dewen;CAI Jing;ZUO Hongfu;ZHANG Ying;HAN Chenqiu(College of Automobile and Traffic Engineering,Nanjing Forestry University,Nanjing Jiangsu 210037,China;College of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing Jiangsu 211106,China;Hangzhou Hikvision Digital Technology Co.,Ltd.,Hangzhou Zhejiang 310051,China)

机构地区：[1]南京林业大学汽车与交通工程学院,江苏南京210037 [2]南京航空航天大学民航学院,江苏南京211106 [3]杭州海康威视数字技术股份有限公司,浙江杭州310051

出　　处：《机床与液压》2023年第13期189-197,205,共10页Machine Tool & Hydraulics

基　　金：国家自然科学基金与民航联合基金重点基金(U1933202)。

摘　　要：航空发动机异常检测对于准确了解飞机健康状态、支持维修决策、保障飞行安全具有重要意义。针对航空发动机气路部件的长期退化行为,提出一种基于慢特征分析和局部离群因子的动态阈值异常检测方法。首先,充分利用慢特征分析的优势提取气路参数随时间缓慢退化的有效特征。然后,计算特征空间样本的局部离群因子来构造监控统计量,定量表征发动机的健康状态。考虑固定阈值对气路状态时变特性的适应性差,利用基于局部分布差异的自适应窗口调整策略,设置动态阈值有效降低气路参数微小波动导致的虚假报警。最后,通过航空发动机实际运行数据进行验证,结果表明:所提方法能提前识别异常点,并且有效降低假警的发生。Aero-engine anomaly detection is of great significance to accurately understand aircraft health status,support maintenance decision and ensure flight safety.A dynamic threshold anomaly detection method based on slow feature analysis and local outlier factor was proposed for the long-term degradation behavior of aero-engine gas-path components.The advantages of slow feature analysis were fully utilized to extract the effective features of slow degradation of gas path parameters with time.The local outlier factors of the feature space samples were calculated to construct the monitoring statistics and quantitatively characterize the health state of the engine.Considering the poor adaptability of the fixed threshold to the time-varying characteristics of the gas path state,the adaptive window adjustment strategy based on the local distribution difference was used to set the dynamic threshold to effectively reduce the false alarm caused by small fluctuations of the gas path parameters.The test results show that the proposed method can identify the abnormal points in advance and reduce the occurrence of false alarms effectively.

关 键 词：航空发动机 慢特征分析 局部离群因子 局部分布差异 动态阈值 

分 类 号：V263.6[航空宇航科学与技术—航空宇航制造工程]