谐波小波样本熵与HMM模型的轴承故障模式识别

Bearing Fault Pattern Recognition Using Harmonic Wavelet Sample Entropy and Hidden Markov Model

根据谐波小波分解非平稳振动信号优良特性与隐马尔科夫(HMM)模型的时序模式分类能力,提出了一种基于谐波小波样本熵与HMM模型结合的轴承故障模式识别方法.该方法首先利用谐波小波对轴承各个状态故障信号进行分解,进而由谐波小波三维时频网格图的频率层数特征计算合理的样本熵维数和阈值,依次提取轴承振动信号各层的样本熵构成特征向量序列;然后将序列前120组输入HMM模型中进行训练得到对应故障模型,剩余80组进行测试与识别,通过对比对数似然估计概率输出值确定轴承故障类型.实验通过与BP和RBF神经网络模型进行不同训练组数的正确识别率对比,验证了该组合方法具有识别准确率高,稳定性强的优点.

According to the excellent characteristics of the non-stationary vibration signals decompose with harmonic wavelet and the strong temporal pattern classification ability of hidden Markov model（HMM）,a bearing fault pattern recognition based on harmonic wavelet sample entropy and HMM was proposed.First,by applying the harmonic wavelet to decompose each bearing fault signal,the frequency layer characteristics of harmonic wavelet three-dimensional time-frequency trellis was used to estimate the reasonable sample entropy dimension and threshold value,and the feature vector sequence was constructed by extracting sample entropy of rolling bearing each layer vibration signal.Then,the feature vector sequence of former 120 groups were input into HMM to be trained,so as to acquire various models corresponding to different faults,and the remaining 80 groups are tested and identified.Finally,the bearing fault types were identified by comparing the logarithmic likelihood probability value.The actual identify results and comparison with the BP,RBF neural network model prove that this coupling method has better identification accuracy and stability.