基于Wavelet-HMM的旋转机械故障诊断方法研究  

STUDY ON FAULT DIAGNOSIS METHODS OF ROTATING MACHINES BASED ON WAVELET TRANSFORM-HIDDEN MARKOV MODELS

作　　者：何树波[1] 丁启全[1] 李志农[1] 吴昭同[1] 

机构地区：[1]浙江大学机械系,杭州310027

出　　处：《机械强度》2003年第5期473-475,共3页Journal of Mechanical Strength

基　　金：国家自然科学基金资助项目 (50 0 750 79)~~

摘　　要：隐Markov模型是一个双随机过程 ,适用于动态过程的时间序列的建模并具有强大的时序模式分类能力 ,特别适合非平稳、重复再现性不佳的信号分析 ;小波变换具有多分辨率分析的特点 ,在时频两域都具有表征信号局部特征的能力。文中将小波变换和隐Markov模型相结合 ,提出基于小波变换的HMM状态识别法 ,利用Daubechies小波进行 8尺度的小波分解 ,然后从小波分解结构中提取一维信号的低频系数作为特征向量 ,将其输入到各个状态HMM来进行训练 ,其中输出概率最大的状态即是机组运行状态 ,从而实现状态的识别 ,实验结果表明该方法很有效。Hidden Markov models(HMM) was a dual random process, adapted to establish time series models of dynamic process and had strong pattern classification, especially adapted to signal which had characteristics of non-stationary, no repeat and worse reappearance. The wavelet transform had characteristics of multi-resolution analysis, which can be used to time-frequency localization and analyze signals. Due to this, fault diagnosis method of rotating machines based on Wavelet-HMM was studied, by which, made signal 8-scale decomposed with Daubechies wavelet. Then, we can get eigenvector by extratcting 1-D approximation coefficients,which would be trained by HMM. we can make fault classification according to the maximum-likelihood probability.Experiment showed that this method had achieved better effect.

关 键 词：旋转机械 故障诊断 隐MARKOV模型 小波变换 状态识别 

分 类 号：TH17[机械工程—机械制造及自动化]