【正文】 基于小波神經(jīng)網(wǎng)絡(luò)的設(shè)備故障診斷方法研究Research on Fault Diagnosis Method of Equipment Based on Wavelet Neural Network Research on Fault Diagnosis Method of Equipment Based on Wavelet Neural NetworkA Thesis Submitted for the Degree of MasterCandidate：SUN ShihuiSupervisor：Prof. ZHAO ShijunCollege of Informationamp。 Control EngineeringChina University of Petroleum (EastChina)關(guān)于學(xué)位論文的獨創(chuàng)性聲明本人鄭重聲明：所呈交的論文是本人在指導(dǎo)教師指導(dǎo)下獨立進(jìn)行研究工作所取得的成果，論文中有關(guān)資料和數(shù)據(jù)是實事求是的。盡我所知，除文中已經(jīng)加以標(biāo)注和致謝外，本論文不包含其他人已經(jīng)發(fā)表或撰寫的研究成果，也不包含本人或他人為獲得中國石油大學(xué)（華東）或其它教育機(jī)構(gòu)的學(xué)位或?qū)W歷證書而使用過的材料。與我一同工作的同志對研究所做的任何貢獻(xiàn)均已在論文中作出了明確的說明。 若有不實之處，本人愿意承擔(dān)相關(guān)法律責(zé)任。學(xué)位論文作者簽名： 日期： 年 月 日學(xué)位論文使用授權(quán)書本人完全同意中國石油大學(xué)（華東）有權(quán)使用本學(xué)位論文（包括但不限于其印刷版和電子版） ，使用方式包括但不限于：保留學(xué)位論文，按規(guī)定向國家有關(guān)部門（機(jī)構(gòu)）送交學(xué)位論文，以學(xué)術(shù)交流為目的贈送和交換學(xué)位論文，允許學(xué)位論文被查閱、借閱和復(fù)印，將學(xué)位論文的全部或部分內(nèi)容編入有關(guān)數(shù)據(jù)庫進(jìn)行檢索，采用影印、縮印或其他復(fù)制手段保存學(xué)位論文。保密學(xué)位論文在解密后的使用授權(quán)同上。學(xué)位論文作者簽名： 日期： 年 月 日指導(dǎo)教師簽名： 日期： 年 月 日i摘 要神經(jīng)網(wǎng)絡(luò)以其固有的記憶能力、自學(xué)習(xí)能力以及強(qiáng)容錯性為故障診斷問題提供了一個新方法。本文針對科學(xué)實驗中廣泛使用的平流泵的故障特點，深入研究了BP神經(jīng)網(wǎng)絡(luò)的故障診斷方法。首先用小波包分析技術(shù)做信號處理。選取 小波函數(shù)，用硬閾值小波包降噪的方3db法將信號降噪，然后進(jìn)行小波包分解與重構(gòu)，以提取信號的能量特征向量，并將得到的特征向量作為神經(jīng)網(wǎng)絡(luò)的輸入。本文采用具有一個隱含層的三層 BP 神經(jīng)網(wǎng)絡(luò)進(jìn)行故障診斷，深入 分析故障診斷的結(jié)果后發(fā)現(xiàn)：第一，網(wǎng)絡(luò)容易陷入極小值而導(dǎo)致診斷失??；第二，網(wǎng)絡(luò)的隱含層節(jié)點數(shù)難以確定。為了解決上述問題，本文研究設(shè)計了 GA+BP 算法。該方法是將遺傳算法與神經(jīng)網(wǎng)絡(luò)相結(jié)合。首先，GA 對 BP 神經(jīng)網(wǎng)絡(luò)做前期優(yōu)化，確定出最佳網(wǎng)絡(luò)結(jié)構(gòu)及該結(jié)構(gòu)對應(yīng)的初始權(quán)值、閾值和網(wǎng)絡(luò)的學(xué)習(xí)速率；然后，構(gòu)造具有最佳結(jié)構(gòu)和參數(shù)的神經(jīng)網(wǎng)絡(luò)來進(jìn)行故障診斷。GA+BP 算法的設(shè)計中， 把每個染色體分解為連接基因和參數(shù)基因，對這兩部分采取不同的遺傳操作。連接基因采用二進(jìn)制編碼方法，參數(shù)基因采用實數(shù)編碼方法；連接基因采用一點交叉方式和基本變異方式，參數(shù)基因中的權(quán)閾基因和速率基因各自采用算術(shù)交叉方式和非均勻變異方式。另外，交叉算子和變異算子都采用自適應(yīng)的方法。GA+BP神經(jīng)網(wǎng)絡(luò)與BP神經(jīng)網(wǎng)絡(luò)故障診斷的結(jié)果對比后可以看到：第一， GA+BP神經(jīng)網(wǎng)絡(luò)比BP神經(jīng)網(wǎng)絡(luò)的工作量少，且克服了陷入局部極小的缺點，有更好的訓(xùn)練性能；第二，GA+BP神經(jīng)網(wǎng)絡(luò)的故障診斷準(zhǔn)確率高于 BP神經(jīng)網(wǎng)絡(luò)。由此可見，GA+BP神經(jīng)網(wǎng)絡(luò)能夠更好的進(jìn)行平流泵的故障診斷工作。關(guān)鍵詞：故障診斷，小波包，神經(jīng)網(wǎng)絡(luò)，遺傳算法iiResearch on Fault Diagnosis Method of Equipment Based on Wavelet Neural NetworkSUN Shihui（Detection Technology and Automatic Equipment）Directed by Prof. ZHAO ShijunAbstractNeural work offers a new method for fault diagnosis owing to its memory ability, selflearning ability and strongly fault tolerance. This paper makes research on the fault diagnosis method of neural work deeply based on the fault characteristics of pump which is widely used in experiment.Wavelet packet analysis is used to do the signal processing. Wavelet is chosen, and 3dball signals are denoised by hard threshold denoising method. Then wavelet packet deposes and constructs the energy eigenvectors which are regarded as the input eigenvectors of the neural work.A threelayer BPNN is applied to do the fault diagnosis. The results of simulation show that the work traps in local minimum easily, and both the number of hidden neurons and the learning rate are difficult to decide either.In order to solve these questions above, this paper designs GA+BP algorithm. In this algorithm, geic algorithm is used to optimize the number of hidden neurons, the initial weights and thresholds, and the learning rate of BPNN first, and then fault diagnosis is done by this neural work which has the optimum structure and parameters. In GA+BP neural work, each chromosome is divided into the connection genes and the parameter genes, and different geic operations are carried on two parts. Connection genes are binary type and parameter genes are realvalued. Mixed crossover and mutation operations are operated on the connection genes and parameter genes separately. It means the connection genes adopt singlepoint crossover and simple mutation, and the parameter genes adopt arithmetic crossover and nonuniform mutation. Both the crossover and mutation operators adopt selfadaptive method.Comparing the simulation results of GA+BP neural work with BPNN, we know that iiiGA+BP neural work has less work but high training performance, and the local minimum is inexistent. In addition, the GA+BP neural work can diagnose the failure more correctly than BPNN. In conclusion, GA+BP neural work can acplish the pump fault diagnosis much better.Key words: fault diagnosis, wavelet packet, neural work, geic algorithmiv目 錄第 1 章 緒論 .............................................................................................................................1 故障診斷的意義 ..............................................................................................................1 故障診斷技術(shù)的研究現(xiàn)狀 ..............................................................................................1 故障診斷方法概述 ..........................................................................................................2 MATLAB 仿真平臺簡介 .................................................................................................3 論文的研究內(nèi)容 ..............................................................................................................4 論文的組織結(jié)構(gòu) ..............................................................................................................4第 2 章 故障信號的采集 .........................................................................................................6 儀器簡介 ..........................................................................................................................6 實驗方案設(shè)計 ...................................................................................................................6 實驗裝置構(gòu)成 ............................................................................................................6 應(yīng)用軟件介紹 .............................................................................................................7 故障信號的數(shù)據(jù)采集 .......................................................................................................8第 3 章 小波分析及信號處理 .................................................................................................9 小波分析在信號處理中的應(yīng)用現(xiàn)狀 ..............................................................................9 小波分析理論 ..................................................................................................