基于KPEMD与INFO-SVM的柱塞泵故障诊断

Fault Diagnosis of Piston Pump Based on KPEMD and INFO-SVM

针对难以从转辙机柱塞泵的非线性振动信号中有效提取故障特征以及各分量之间存在模态混叠现象等问题,提出了一种非线性自适应正交经验模态分解(Kernel Principal Empirical Mode Decomposition,KPEMD)与向量加权平均算法优化的支持向量机(INFO-SVM)结合的故障诊断方法。首先通过KPEMD方法将原始信号分解为多个IMF分量,根据相关系数筛选出故障信息丰富的敏感分量;其次提取敏感分量的时域频域特征及能量熵构造混合特征样本集;最后输入到INFO-SVM多分类器中进行故障识别。利用柱塞泵实验数据进行对比分析,结果表明:KPEMD能够有效减弱模态混叠现象,充分提取故障信息,INFO优化SVM的识别准确率优于其它常用算法的优化结果。本方法能有效识别出转辙机柱塞泵的不同故障类型,诊断准确率达到98%。

Aiming at the problems such as difficult to effectively extract fault features from the nonlinear vibration signals of the piston pump of the transposition machine and the mode mixing phenomenon exists among the components,a fault diagnosis method combining kernel principal empirical mode decomposition(KPEMD)and support vector machine optimized by vector weighted average algorithm(INFO-SVM)was proposed.Firstly,the original signal was decomposed into multiple IMF components by KPEMD method,and the sensitive components with rich fault information were screened out according to the correlation coefficients.Secondly,the time domain and frequency domain features and energy entropy of the sensitive components were extracted to construct a mixed feature sample set.Finally,it is input into INFO-SVM multi-classifier for fault recognition.The results of comparative analysis using experimental data of piston pumps show that KPEMD can effectively attenuate the modal confusion phenomenon and fully extract the fault information,and the recognition accuracy of INFO-SVM is better than the optimization results of other common algorithms.This method can effectively identify different fault types of rutting machine plunger pumps,and the diagnosis accuracy reaches 98%.