主分量分析和因子隐Markov模型在机械故障诊断中的应用

主分量分析(principal component analysis,PCA)是统计学中分析数据的一种有效方法,可以将高维数据空间变换到低维特征空间,因而可用于多通道冗余消除和特征提取.因子隐Markov模型是隐Markov模型的扩展,它比隐Markov模型更有优势,适用于动态过程时间序列的建模,并具有强大的时序模型分类能力,特别适合非平稳、信号特征重复再现性不佳的信号分析.文中结合主分量分析与因子隐Markov模型,提出一种新的故障识别方法,即以主分量分析方法进行冗余消除和故障特征提取,因子隐Markov模型作为分类器.并应用到机械故障诊断中,同时与基于主分量分析的隐Markov模型的识别方法相比较,实验结果表明基于PCA的因子隐Markov模型识别法和基于PCA的隐Markov模型识别法在故障识别上都是有效的,但对于相同的状态空间,前者的训练速度快于后者,尤其是状态空间越大,这种优势越明显.

APPLICATION OF PRINCIPAL COMPONENT ANALYSIS AND FACTORIAL HIDDEN MARKOV MODEL IN MACHINE FAULT DIAGNOSIS

The principal component analysis (PCA), which is an effective method of analyzing data in statistics, can compress the higher dimensional data space into the lower dimensional feature space. So it can be used in the feature extraction. The factorial hidden Markov model (FHMM), which is an extension of the hidden Markov model (HMM), is superior to HMM, and has strong capability of pattern classification, especially for the signals with abundant information, non-stationarity, bad repeatability and reproducibility. Combining PCA and FHMM, a new approach of fault recognition named PCA-FHMM is proposed, in which PCA is used as a redundancy reduction and feature extraction, and FHMM is used as a classifier. And the proposed approach is applied to the mechanical fault diagnosis successfully. The proposed is compared with another recognition method named PCA-HMM, in which PCA is also used as feature extraction, however HMM as a classifier. The experiment results show that the two recognition methods are both very effective. However the speed of training has obvious difference. The PCA-FHMM recognition method is faster than the PCA-HMM recognition method, especially the larger the state spaces are, the more obvious this superiority is.