改进的Kohonen网络在航空发动机分类故障诊断中的应用  

Application of Improved Kohonen Network in Aeroengine Classification Fault Diagnosis

作　　者：郑波 马昕[2] ZHENG Bo;MA Xin(Academic Affairs Office,Air Traffic Management College,Guanghan Sichuan 618307,China;Academic Affairs Office,Civil Aviation Flight University of China,Guanghan Sichuan 618307,China)

机构地区：[1]中国民用航空飞行学院教务处,四川广汉618307 [2]中国民用航空飞行学院空中交通管理学院,四川广汉618307

出　　处：《航空发动机》2020年第2期23-29,共7页Aeroengine

基　　金：四川省科技计划项目(2019YJ0720);中国民用航空局发展基金教育人才类项目(14002600100018J034);中国民用航空飞行学院面上项目(2019-053)资助。

摘　　要：针对传统Kohonen网络对未知样本识别时的不可辨识性和分类结果不惟一性问题,利用改进的Kohonen网络对航空发动机进行分类故障诊断,并利用混合粒子群优化算法对网络连接权值进行优化,以提高Kohonen网络在分类故障诊断中的通用性和容错能力。对GE90发动机的孔探图像纹理特征识别进行对比。结果表明:改进的Kohonen网络在分类故障诊断中有较强的实用性,分类准确率高于常用神经网络模型和支持向量机的。In order to overcome the problems of inidentify and inunique classification results of unknown samples identified by Traditional Kohonen(T-Kohonen) network,the classification fault diagnosis of aeroengine was carried out by using Improved Kohonen(IKohonen)network. The Hybrid Particle Swarm Optimization(H-PSO)algorithm was used to optimize the network connection weights to improve the generality and fault tolerance of Kohonen network in classification fault diagnosis. Borescope image texture feature identification of GE90 engine was compared. The results show that the I-Kohonen network is more practical in classification fault diagnosis,and the classification accuracy is higher than that of common neural network models and Support Vector Machines(SVM).

关 键 词：KOHONEN网络 PSO算法 自适应继承 自适应检测响应 故障诊断 航空发动机 

分 类 号：V263.6[航空宇航科学与技术—航空宇航制造工程]