Study On Fault Diagnosis In Analog Circuits Based On Wavelet Packet Analysis And Neural Networks

Because of the analog circuit‘s inherent particularity, the development of it’s faultdiagnosis technology is relatively slower than digital circuit, and the research field of theanalog circuit fault diagnosis is all along the challenge and hot focus in electrical circuitsystem’s fault diagnosis. Along with the development of microelectronics technology,integrated circuits have been largely used into circuit systems, and some new methods ofdiagnosis which are superior to the traditional methods have been applied in hybrid circuits.Neural Network and its optimized algorithms have been introduced into fault diagnosis fields,and created a new situation for the study of the analog circuit‘s fault diagnosis.Wavelet packet analysis can adaptively determine the signals’resolutions in differentfrequency bands, and signals’synchronized segregation from the high frequency to low canbe realized; Neural Network’s advantages are its strong nonlinear mapping, self-learning,adaptive and fault tolerant abilities and so on. In view of both advantages, this paper combinesthe wavelet packet and neural network algorithm to deal with the problems in analog circuitfault diagnosis field and makes some simulation analysis to the circuit examples, which meetsthe desired diagnostic targets. The main tasks of this paper are as follows:Firstly, on the basis of the analysis and comparison of the commonly used methodsapplied to the analog circuit fault diagnosis, this paper details the general theories of NeuralNetwork, wavelet and wavelet packet analysis, and completes Neural Network structure’sdesign which is based on wavelet packet.Secondly, through an example of circuit, a diagnostic method of“energy-fault”which isbased on wavelet packet is studied and applied into the fault information extraction for analogcircuit, and thus to simplify network structure, accelerate network training, reduce the network complexity and locate the accurate position of the faults.Finally, an example simulation experiment is performed to demonstrate that WaveletPacket Neural Network has advantages over BP Neural Network in forecasting performance,and this algorithm can effectively do automatic fault diagnosis for the analog circuits.