基于SVD-LMD与DHMM在滚动轴承故障诊断中的应用

The Application of SVD-LMD and DHMM to Fault Diagnosis of Rolling Bearing

针对滚动轴承早期故障信号微弱难以提取和故障类型不易判别的缺点,提出了基于奇异值分解(SVD)-局部均值分解(LMD)与离散隐马尔可夫模型(DHMM)的滚动轴承故障类型识别方法。首先,对经过相空间Hankel矩阵重构的原始声学信号进行SVD降噪得到特征信号,再运用LMD对特征信号分解而产生一系列的乘积函数(PF),为去除LMD分解过程中产生的虚假分量,选择与特征信号相关系数值较大的PF并构建特征向量T以完成信号特征提取。最后,将T进行量化后作为特征观测值输入已训练收敛的DHMM模型进行故障状态识别。并与支持向量机(SVM)进行比较研究。实验结果表明,基于SVD-LMD与DHMM的滚动轴承故障诊断模型在声学信号下对早期滚动轴承的故障具有较高的识别率。

Aiming at the early acoustics signal of rolling bearing being featured and the fault states being classified difficultly,the method of fault diagnosis for rolling bearing based on SVD-LMD and DHMMis applied to this paper. Firstly,the rawacoustics signal dealt with Hankel matrix is denoised by SVD. And then the signal is decomposed into Product Function（ PF） by LMD after SVD denoising. It＇s necessary for PFs to compute the correlation coefficient between PFs and featured signal for removing the false weight.The PFs with high correlation coefficient value are chosen and processed by quantization. Finally,the feature vectors are input into the trained DHMMfor recognition. The experimental results showthat the method of SVD-LMD and DHMMis superior to the method of SVD-LMD and SVM,and it can identify the fault states of rolling bearing accurately and effectively.