基于广义CP张量分解和多尺度排列熵的液压泵故障诊断

Hydraulic Pump Fault Diagnosis Based on Generalized CP Tensor Decomposition and Multiscale Permutation Entropy

针对现有液压泵故障诊断算法故障识别精度低、实测信号中存在冗杂信息和无关成分干扰等问题,提出了一种在高维空间对液压泵振动信号进行处理和模式识别的方法。首先利用Lowner矩阵将一维时间序列信号进行高维张量化,然后基于广义CP张量分解(GCP)算法,根据数据分布类型选择适当的损失函数以确定最佳低秩模型,实现对液压泵采集得到的振动信号进行分解,降低分解损失,提高分解精度。最后选择分解结果中与原始信号相似度最高的模式分量,计算其多尺度排列熵(MPE)值进行液压泵不同故障类型的特征识别。将该文所提方法应用于液压泵故障实验台的实测数据分析,验证了其在液压泵故障诊断中的有效性。

Aiming at the problems of low fault identification accuracy,interference of miscellaneous information and irrelevant components in the measured signals,a method for pattern recognition of hydraulic vibration signals in high dimensional space is proposed.Firstly,a one-dimensional time series signal is tensioned by L-wner matrix.Then,according to the data distribution type,the appropriate loss function was selected to determine the optimal low-rank model.Based on it,Generalized Canonical Polyadic Tensor Decomposition(GCP)algorithm was used to decompose the vibration signals collected by the hydraulic pump.The purpose of this operation is to reduce the decomposition loss and improve the decomposition accuracy.Finally,the mode component with the highest correlation in the decomposition results was selected.Then,its Multi-scale Permutation Entropy(MPE)was calculated to recognize the different fault types,so as to realize the pattern recognition of different fault types of hydraulic pump.The proposed method is applied to the experimental data analysis of hydraulic pump fault,and its effectiveness in hydraulic pump fault diagnosis is verified.