基于VMD-HMM的滚动轴承磨损状态识别

Wear state recognition of rolling bearings based on VMD-HMM

根据变分模态分解(VMD)在信号处理上的良好特性与隐马尔可夫模型(HMM)在时间序列上的分类能力,提出了一种基于VMD分解与HMM模型相结合的滚动轴承磨损状态识别方法。该方法利用VMD对轴承各个磨损时期信号进行分解,进而计算VMD分解后各IMF的能量熵,依次提取轴承振动信号的各层IMF能量熵构成特征向量序列,每种状态随机选取20组(共80组)输入HMM模型训练,剩余的特征向量序列进行测试,通过对比对数似然概率值来判别磨损状态。实验结果表明该方法能够准确分辨出轴承的磨损状态,与EMD-HMM、谐波小波样本熵HMM模型进行了对比,验证了该方法具有识别性高准确性强的优点。

Based on good performance of the variational mode decomposition（ VMD） in signal processing and classification ability of the hidden Markov model（ HMM） to time series,a rolling bearing wear state recognition method based on VMD-HMM was proposed. Firstly,VMD was used to decompose vibration signals of a bearing in its various wear durations,and the energy entropy of each IMF after VMD was calculated. Then,various IMFs’ energy entropies of bearing vibration signals in various wear durations were extracted to form eigenvector sequences. Finally,the randomly selected 20 groups in total 80 groups of eigenvector sequences for each wear state were input into HMM model to be trained,and the rest eigenvector sequences were tested. Through comparing logarithmic likelihood probability values,the bearing wear state was determined. The test results showed that the proposed method can be used to accurately distinguish the wear state of the bearing; compared with EMD-HMM and the harmonic wavelet sample entropy HMM model,it has higher recognition and accuracy.