模拟电路故障信号的小波预处理  

Wavelet preprocessing of analog circuit fault signal

作　　者：彭卫韶[1] 李力争[1] 胡燕瑜[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《中南大学学报（自然科学版）》2008年第3期584-589,共6页Journal of Central South University:Science and Technology

基　　金：国家自然科学基金资助项目(50575228)

摘　　要：针对模拟电路故障诊断的神经网络存在结构规模较大的问题,提出一种基于小波-神经网络的模拟电路故障诊断方法。该法采用冲激响应来获取模拟电路的故障信号,采用小波变换作为模拟电路故障信号的预处理器,利用Haar小波分层次分解提取故障信号特征,该信号特征经主元分析和数据标称化后,作为用于故障诊断的神经网络的输入。基于该法故障诊断的基本原理,对一实例电路进行故障划类、小波函数及故障特征选择,给出计算故障特征的仿真编程及故障类别的识别方法。该法大大减少用于故障诊断的神经网络的输入数目,简化它的结构和减少其训练处理的时间。仿真结果表明,该法可以提高模拟电路故障诊断的效率和辨识故障类别的能力。Aiming at the problem of larger structure of neural network used for analog circuit fault diagnosis, a fault diagnosis method based on wavelet-neural network for analog circuit was presented. The method uses impulse response to obtain the fault signal, uses wavelet transform as a preprocessor of this fault signal, and uses Haar wavelet decomposition to obtain the feature of the fault signal. Then through principal component analysis and data normalization, the feature of the fault signal was used as the input of a neural network for fault diagnosis, Based on the basic principle of this fault diagnosis method, the fault classification, the selection of the wavelet function and the selection of the fault features for an example circuit were studied, and the fault category identification and a simulation program for computing fault features were also given. This method drastically reduces the number of inputs, simplifies the structure and decreases the training and processing time of the neural network. Simulation result shows that the method will improve the fault diagnosis efficiency and the fault classifying capacity.

关 键 词：小波变换 神经网络 故障诊断 模拟电路 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]