Research On The Application Of Wavelet Neural Networks To Fault Diagnosis Of Otating Machinery

It is out of question that mechanical default diagnosis is very important insafety production;it has become an important research direction of machine andmeasurement field. Lately, the new theory, new method and new measure aboutmechanical default diagnosis have appeared incessantly, they have becomeperfect step by step. Some of them have come into being correlative industry,which makes our country's level of mechanical default diagnosis improveobviously.In this dissertation, using vibration signal as primary study object, awavelet neural network used to analysis in time field and frequency field andused to pick up character parameter are given. Theory about wavelet analysis isstudied, according to the problem of noise of fault signal, a kind of model ofdenoising based on wavelet is given. The time-frequency characters of wavelettheory can give over the limitation of Fourier transform which it can onlyanalyze in whole. Base on the study theory of neural network, learningalgorithm of neural network is given. According to the problem of slowconvergence speed and easily setting into local small extremum, we give theimprovement algorithm in learn ratio and error correction respective.Binding mode of wavelet and neural network is studied and threeconstruction forms are given. Explain conjugate gradient algorithm andmulti-resolution analysis algorithm then the principle used to confirm waveletbasis and initialize parameter of wavelet is given;Using Mexican hat asexample to deduce learning algorithm. To solve the problem of "dimensiondisaste" of the wavelet neural network, a genetic optimization algorithm that cansearch for the optimum wavelet parameter and network neural adaptively in thelearning processing, it can predigest network frame and improve constringencyspeed.At last, wavelet neural network that applaied in mechanical fault diagnosisis researched, according to ten kinds of rotating machinery faults, usingvibration signal as input of network and giving training result;Using vibrationsignal classify of bearing as example to get diagnosis steps and result. To solvethe problem of higher dimensional multiply input and output an integratedwavelet neural network in mechanical fault diagnosis is given, which canimprove diagnosis ratio effectively.