基于SVM和SNN的航空发动机气路故障诊断  

Fault diagnosis of aero-engine gas path based on SVM and SNN

作　　者：王修岩[1] 李萃芳[1] 高铭阳[2] 李宗帅[1] 

机构地区：[1]中国民航大学航空自动化学院,天津300300 [2]北京航空航天大学仪器科学与光电工程学院,北京100191

出　　处：《航空动力学报》2014年第10期2493-2498,共6页Journal of Aerospace Power

基　　金：中央高校基本科研业务费中国民航大学专项(ZXH2010C002)

摘　　要：为了区分航空发动机气路故障诊断过程中出现的相似故障,提高诊断准确率,提出了一种支持向量机(SVM)和协同神经网络(SNN)相结合的故障诊断方法.首先利用参数优化后的SVM对测量数据进行初步故障诊断分类,对诊断结果进行分析统计,得出难以区分的相似故障类型,并根据SNN对这些相似故障进一步地区分判断,最后根据实际数据对此故障模型进行仿真.结果显示:基于SVM的初步故障诊断准确率达到96%;而经过SNN进一步地相似故障区分后,诊断准确率提升到100%.In order to distinguish similar faults of aero-engine gas path fault diagnosis and improve the diagnostic accuracy, a fault diagnosis method based on support vector ma- chine （SVM） and synergetic neural network （SNN） was put forward. Firstly, the SVM af- ter being optimized was used to diagnose and classify the faults preliminarily form measured data, and the diagnosis results were analyzed to obtain indistinguishable similar faults, then the SNN was introduced to distinguish similar faults and further determine corresponding fault model, finally this fault model was simulated based on actual data. The experimental results show that the preliminary fault diagnosis accuracy based on SVM is 96%, and after further distinguishing similar faults through SNN, the accuracy is increased to 100%.

关 键 词：航空发动机 故障诊断 气路 支持向量机 协同神经网络 

分 类 号：V233.7[航空宇航科学与技术—航空宇航推进理论与工程] TP181[自动化与计算机技术—控制理论与控制工程]