基于HMM的故障诊断特征提取和聚类技术

Feature Extraction and Clustering Technique of Rotating Machinery Fault Diagnose Based on HMM

重点介绍了运用HMM进行故障诊断特征矢量的提取。在试验的基础上,对4种典型故障进行了数据采集。通过加窗处理,采用自相关法提取12阶LPC倒谱系数,用LBG聚类算法进行矢量量化,得出码本矢量。运用这些矢量训练各故障对应的HMM模型,然后将所测故障数据按上述方法矢量量化后输入到训练好的HMM中,求出似然概率值,值最大者即为故障状态。结果表明,利用该种方法进行特征提取并与HMM方法相结合能很好分类出各种故障模式,达到诊断目的。

This paper focused on extraction of feature vector of fault based on HMM. Four typical fault data are gathered in experiments, the twelve orders of LPC cepstrum coefficients are distilled through the dispose adding window and the self- correlation and the codebook vectors are obtained through LBG clustering algorithm. Each HMM model of the faults is trained using these vectors. Then, the log-likelihood probability is calculated by inputting the fault data that have been quantified into the trained HMM. The maximum is in the fault state. The results show that combining this method with HMM can diagnose machinery effectively faults.