采用滑动平均多元多尺度色散熵的液压泵故障诊断方法

为了提高色散熵的信息提取能力，在兼顾计算效率和效果的前提下，引入多维嵌入重构理论，借鉴滑动平均的思想，更新了传统多尺度算法的粗粒化方式，提出了滑动平均多元多尺度色散熵(moving average multivariate multiscale dispersion entropy，MA_mvMDE)用以提取液压泵故障特征。首先，利用均匀相位经验模态分解（uniform phase empirical mode decomposition，UPEMD）将振动信号分解为多个本征模态分量（intrinsic mode functions，IMF），再采用相关系数法筛选敏感分量，将包含大量故障信息的模态分量作为多通道数据计算其MA_mvMDE值来提取故障特征。接着，采用MCFS方法选择故障敏感特征实现降维。最后，通过随机森林分类器完成故障识别。采用液压泵故障振动数据验证了该方法能够准确诊断不同类型和不同程度的故障。

Fault Diagnosis of Hydraulic Pump Adopting Moving Average Multivariate Multiscale Dispersion Entropy

In order to improve the information extraction of dispersion entropy, this paper updates the coarse-graining approach of the traditional multi-scale algorithm and proposes a new method called moving average multivariate multiscale dispersion entropy (MA_mvMDE) by introducing the multidimensional embedding reconstruction theory and the idea of moving average, with due consideration of computational efficiency and effects. Firstly, the vibration signal is decomposed into several intrinsic mode functions (IMF) by uniform phase empirical mode decomposition (UPEMD), the sensitive components are screened out by correlation coefficient method, and the selected components containing a large amount of fault information are calculated to get their MA_mvMDE values as multi-channel data to extract the fault characteristics. Secondly, the multi-cluster feature selection (MCFS) method is used to select sensitive fault features to reduce feature dimensions. Finally, the fault recognition is completed by the Random Forest classifier. In this paper, the weak fault vibration data of hydraulic pump have verified that this method can accurately diagnose different types and degrees of faults.