基于VMD改进MDE算法的液压泵滑靴磨损微弱故障信号识别

Identification of Weak Fault Signal of Hydraulic Pump Slipper Wear Based on VMD Improved MDE Algorithm

针对变分模态分解(Variational Modal Decomposition,VMD)特征能量重构法实现故障算法存在准确性不高问题,对原始信号先通过VMD分解获得能量余量,在特征能量占比(Feature Energy Ratio,FER)基础上对VMD特征能量重构法,并选择有效的多尺度散布熵(Multiscale Dispersion Entropy,MDE)作为向量。以液压泵故障诊断为研究对象,依次分析了液压泵在正常状态与滑靴端面磨损为0.1 mm、0.2 mm、0.3 mm状态下情况。仿真结果得到:时间尺度持续增大,形成了排列更有序的粗粒化序列,系统复杂性大幅降低。重构信号MDE在正常运行状态和滑靴磨损达到0.10 mm时都可以保持稳定状态。VMD-MDE方法进行处理获得了98.1%准确率,与VMD相关系数分类方法相比提高了7.1%,与模态分解(Mode Decomposition,DE)方法相比提高19.32%。极限学习机(Extreme Learning Machine,ELM)处理时间比支持向量机(Support Vector Machine,SVM)降低13.2%,而准确率增大了19%。达到更快分类速率和更高精度。

This work addresses the problem of low accuracy in fault realization using the variational mode decomposition(VMD)characteristic energy reconstruction method.The energy allowance of the original signal is first obtained by VMD.Then the VMD energy reconstruction method is performed on the basis of feature energy ratio(FER),and the effective multiscale dispersion entropy(MDE)is selected as the vector.Taking the fault diagnosis of a hydraulic pump as the research object,the condition of the hydraulic pump under the normal condition and conditions with slipper end wear of 0.1 mm,0.2 mm and 0.3 mm is analyzed respectively.The simulation results show that the coarse-grained sequence with more ordered arrangement is formed,and the system complexity is significantly reduced.The reconstructed signal MDE can maintain a stable state both in normal operation state and when the slipper wear reaches 0.10mm.The accuracy of VMD-MDE is 98.1%,which is 7.1%higher than that of VMD correlation coefficient classification method and 19.32%higher than that of DE method.Compared with SVM,the processing time of ELM is reduced by 13.2%,while the accuracy is increased by 19%.The proposed methodhas achieved faster classification rate and higher accuracy.