模拟电路的马氏距离故障诊断方法研究

针对欧氏距离在模拟电路故障诊断中的不足,提出模拟电路的马氏距离故障诊断方法。利用MULTISIM对电路的软故障和硬故障进行仿真获取故障特征样本,在MATLAB中计算样本与故障类之间的马氏距离,并选择距离最小的故障类作为诊断结果。仿真结果表明:在20次不同特征维数的试验中,欧氏距离,正规化欧氏距离和马氏距离故障诊断方法的平均诊断准确率分别为:86.7%,87.1%和92.9%。马氏距离的诊断方法不受特征相关性的影响,能够正确诊断出欧氏距离和正规化欧氏距离诊断方法的误诊样本,在模拟电路的故障诊断中的准确率更高。

Research on application of Mahalanobis distance in fault diagnosis of analog circuits

In order to compensate the deficiency of the application of Euclidean distance in the fault diagnosis of analog circuits,a method based on Mahalanobis distance is proposed.After getting the samples in the MULTISIM,Mahalanobis distances between samples and faulty states are calculated in MATLAB.Each sample is diagnosed to be in the state with the smallest distance.The result shows that：in twenty experiments of different feature dimensions,the average diagnosing accuracy ratios of three methods which adopt Euclidean distance,normalized Euclidean distance and Mhalanobis diatance are 86.7%,87.1% and 92.9% respectively.The Mahalanobis distance method is not affected by the correlation among the features,so it can diagnose the samples which are wrongly diagnosed by the other two methods.Therefore Mahalanobis distance method performs better in the fault diagnosis of analog circuits.