基于小波高频系数统计特征的电路故障诊断

对故障电路进行特征提取与分类是模拟电路诊断的两个重要环节。现有方法多对时域响应信号进行小波变换以提取故障特征,并用神经网络或支持向量机方法实现对故障进行分类。为提高模拟电路故障诊断率,提出一种新的特征选取方法:在模拟电路的时域响应中对其进行小波变换,并对变换得到的高频细节系数统计平均值、标准偏差、峭度、熵和偏斜度等统计特征,并建立以支持向量机为分类器的故障诊断系统。以两种常见电路为例,实验结果表明,提出方法对常见电路进行故障诊断,准确率得到提升,精度达到99%以上,优于传统单纯小波系数分析方法,适用于模拟电路的故障诊断。

Analog Circuits Fault Diagnosis Using Wavelet Statistical Features

In the process of the fault diagnosis of analog circuits,feature extraction and classifier design are two critical aspects.Most methods classify fault circuit via support vector machine (SVM) using extracted time or frequency features.In order to improve the diagnostic accuracy,we propose a new circuits fault diagnosis method based on wavelet statistical features.For a faulty circuit,we firstly emulate its time domain responses and decompose them into wavelet coefficients.Then we calculate the average,standard deviation,kurtosis,entropy and skewness of the detail coefficients before composing them into statistical feature vectors.Finally,all faulty circuits feature vectors are classified using SVM method.For optimizing the SVM parameters,cross validation is also exploited in the classification stage.Experiment results show that the proposed method is effective in the circuits fault diagnosis with an accuracy over 99%,better than that by traditional methods.