基于小波分形和核判别分析的模拟电路故障诊断

A Fault Diagnosis Approach of Analog Circuit Using Wavelet-Based Fractal Analysis and Kernel LDA

提出了采用小波分形分析和核判别分析作为预处理器来实行特征提取的神经网络模拟电路故障诊断方法。这个诊断方法采用小波分形分析方法首先获取了故障响应信号的小波分形维特征,然后采用核判别分析进一步实施特征提取,最后将所获得的最优特征模式作为神经网络分类器的输入以进行故障诊断。仿真结果表明,本文提出的预处理方法能很好地获取故障响应信号的本质特征,并表现出了比其他特征提取方法更好的性能。并且,由此所构建的神经网络不但具有小的网络结构,而且能取得高的故障诊断正确率。

A neural-network fault diagnosis approach utilizing wavelet-based fractal analysis method and kernel linear discriminant analysis（KLDA） as preprocessors is proposed.The diagnostic approach preprocesses the fault response signals in such a way that the wavelet-based fractal analysis obtains the fractal-dimension features of fault response signals,and KLDA further extracts the optimal features used as the inputs to neural-network classifier.The simulation results show that the proposed method can acquire the essential features of fault response signals and display better performance than other methods.Furthermore,the resulting neural networks not only have the small structures but also can achieve high accuracy of fault diagnosis.