基于特征层融合的模拟电路故障诊断方法  

Fault diagnosis of analog circuit based on feature-level fusion

作　　者：谢永成[1] 程延伟[1,2] 吕强[1] 李光升[1] 

机构地区：[1]装甲兵工程学院控制工程系,北京100072 [2]装甲兵技术学院控制工程系,吉林长春130117

出　　处：《计算机工程与设计》2012年第3期1170-1174,共5页Computer Engineering and Design

摘　　要：为了提高模拟电路软故障诊断精度,提出了基于特征层融合的模拟电路故障诊断方法。通过小波变换的频带能量和AR模型的系数生成故障的初始特征,利用线性判别分析提取特征,降低特征向量的维数,结合mRMR原则与支持向量机构成的组合特征选择方法,对降维后特征进行最优选择,输入到训练好的支持向量机多分类器中进行分类,从而降低特征的冗余,提高故障诊断的精度。以某车辆电压调节电路为例进行仿真分析,实验结果表明,该方法有很好的诊断能力。The fault diagnostic method of analog circuit based on feature-level fusion is presented to improve diagnostic accuracy of soft fault. Firstly, using frequency bands energy of wavelet transform and coefficients of AR model generate the initial failure characteristics, extract the features using linear discriminant analysis and reduce the characteristic dimensions. Combining the mRMR principles and support vector Machines constitute the hybrid feature selection method, the optimal characteristics are selected, the trained SVM classification is input, the feature redundancy is reduced, and the diagnostic accuracy is improved. Taking the simulative analysis of the vehicle voltage regulator circuit as example, the results indicate that the method has good diagnostic capabilities.

关 键 词：模拟电路 故障诊断 特征层融合 特征提取 特征选择 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]