解相关多频率经验模态分解的故障诊断性能优化方法

针对故障诊断中采用EMD方法存在模态混叠现象,引起故障特征提取精度低的问题。提出了一种解相关多频率经验模态分解(Decorrelation Multiple-Frequency Empirical Mode Decomposition,DMFEMD)方法,首先对初始信号添加多个频率的掩蔽信号,初步分解其中不同频率比的信号分量得到多个IMF分量;其次计算相邻IMF之间的相关系数并对其解耦,进一步分离IMF中存在混叠的部分,得到最优IMF;最终,从原始信号中减去最优IMF,然后重复上述步骤,直到残余分量为常数或单调。由于保证了IMF之间互不相关且互不干扰,因此模态混叠现象显著减弱,有效提高故障特征提取精度。利用排列熵算法对一系列最优IMF构造特征样本集,引入SVM建立故障分类模型,实现设备故障诊断。通过试验证明,DMFEMD与传统的方法相比,能有效分离不同频率比混合信号,提高分解效果。同时以轴承振动信号为例,DMFEMD可以更好的提取轴承的故障特征,结合PE与SVM能够实现不同故障类型的高效精确的诊断。

Fault diagnosis performance optimization method based on decorrelation multi-frequency EMD

Aiming at the problem of modal aliasing caused due to using EMD in fault diagnosis leading to lower fault feature extraction accuracy,a method based on decorrelation multi-frequency empirical mode decomposition(DMFEMD)was proposed here.Firstly,multi-frequency masking signals were added into the original signal,the latter was decomposed into several components with different frequency ratios to obtain multiple intrinsic mode functions(IMFs).Secondly,the correlation coefficient between adjacent IMFs was calculated,two IMFs were decoupled,and the mixed parts of IMFs were separated to obtain the optimal IMF.Finally,this optimal IMF was subtracted from the original signal,and the above steps repeated until the residual part became a constant or monotonic.Thus,the obtained optimal IMFs were uncorrelated and not interfered with each other,the modal aliasing was significantly weakened,and the fault feature extraction accuracy was effectively improved.A feature sample set was constructed using the permutation entropy(PE)algorithm for the obtained optimal IMFs.A SVM was introduced to build a fault classification model,and realize equipment fault diagnosis.Tests showed that compared with the traditional method,DMFEM can be used to effectively separate mixed signals with different frequency ratios,and improve the decomposition effect;taking faulty vibration signal of bearing as an example,DMFEMD can be used to better extract bearing fault features;PE combined with SVM can realize efficient and accurate diagnosis of different fault types.