一种新模拟电路故障特征提取与诊断方法.docVIP

一种新的模拟电路故障特征提取与诊断方法 　　摘 要：提出了一种基于小波变换、傅立叶变换与神经网络相结合的模拟电路故障诊断新方法.该方法使用小波变换和傅立叶变换对模拟电路在各种故障状态下进行特征的提取，即首先对电路在各种故障状态下的节点电压信号进行小波变换以将干扰信号(如噪声)除去，防止不相关的能量混在有效信号中，然后采用傅立叶变换进行分析，得到有效信号频谱，提取其能量值再经主元分析与归一化处理后作为故障特征，采用概率型神经网络实现故障的定位.分析与仿真结果表明，本文方法获得了较好的故障分辨率与诊断正确率. 　　 关键词： 小波变换；傅立叶变换；径向基网络；概率型网络；模拟电路；故障诊断 　　 中图分类号：TN39 文献标识码：A 　　A New Fault Feature Extraction and Diagnosis Method of Analog Circuits 　　 ZHU Wenji, HE Yigang 　　(College of Electrical and Information Engineering, Hunan Univ, Changsha, Hunan 410082, China) 　　Abstract: Based on wavelet transform (WT), Fourier transform (FT) and neural network, a new fault diagnosis method of analog circuits was proposed. The proposed method uses wavelet transform (WT) and Fourier transform (FT) for fault feature extraction when the analog circuits are under different faulty situations. That is, we use WT to filter the disturbance influences (for example, noises) on the original signals to prevent the unrelated energies from being mixed with the effective signals. These signals are then analyzed by FT to obtain the frequency spectrum of the effective signals. And then, the energies of these signals are extracted and preprocessed by principal component analysis (PCA) and normalization as fault features. Meanwhile, considering that the probability neural networks (PNNs) have characteristics of simple structures, highspeed of training process and easy append training samples, we use this kind of neural networks for fault location. The diagnosis principles and procedures were outlined, And the satisfied diagnosis resolution and accuracies have beenachieved by using the proposed method. 　　 Key words:WT; FT; radial basis function neural networks (RBFNNs); PNNs; analog circuits; fault diagnosis 　　 　　 近年来，随着智能技术的不断深入与发展，利用神经网络解决各种模式分类问题已引起了科研工作者的广泛关注[1-8].概率型神经网络是径向基网络的一种变化形式[9-10]，它结合了径向基函数网络的优点，特别适合于处理模式分类问题，与目前较为广泛应用于模拟电路故障诊断的BP网络方法相比，这种方法避免了冗长的迭代计算和陷入局部极值的可能，使学习速度比通常的BP算法快几十倍甚至上百倍[9-10]. 　　利用神经网络进行故障的模式识别中，故障特征的提取具有非常重要的作用.傅立