基于Hilbert包络谱熵和GA-SVM的水轮发电机轴承故障诊断

Hydro Generator Bearing Fault Diagnosis Based on Hilbert Envelope Spectral Entropy and GA-SVM

水轮发电机轴承在运行时承受着整体机组的轴向负荷与复杂水推力,针对其产生的非稳态、非线性特征的振动信号,提出一种基于Hilbert包络谱分析与遗传算法支持向量机(GA-SVM)相结合的诊断方法,用于轴承故障状态的识别。首先对推力轴承运行时产生的振动信号进行集合经验模态分解(Ensemble Empirical Mode Decomposition,EEMD),分解成若干个固有模态函数(Intrinsic Mode Function,IMF),依据峭度准则选取主要IMF分量并通过Hilbert包络谱分析,计算包络谱熵,将归一化后的包络谱熵作为特征向量输入GA-SVM进行训练与故障识别。仿真实验结果表明,基于EEMD包络谱熵分析法相比于时频域图像处理能更好地提取出复杂工况下的故障信号特征,遗传算法支持向量机识别准确率达96.87%,该算法模型可进一步应用于水轮发电机轴承故障诊断。

Hydro generator bearings bear the axial load and complex water thrust of the overall unit during operation,for the vibration signals generated by them with non-stationary and non-linear characteristics,a diagnostic method based on the combination of Hilbert envelope spectral analysis and genetic algorithm support vector machine(GA-SVM)is proposed to be used for the identification of the bearing fault state.Firstly,the vibration signals generated during the operation of the thrust bearing are subjected to ensemble empirical mode decomposition(EEMD),which is decomposed into several intrinsic mode functions(IMF),and the main IMF components are selected according to the craggy criterion and analyzed by the Hilbert envelope spectral analysis.The main IMF components are selected according to the cliff criterion and analyzed by the Hilbert envelope spectrum,the envelope spectrum entropy is calculated,and the normalized envelope spectrum entropy is input into GA-SVM as the feature vector for training and fault identification.Simulation experiments show that the EEMD-based envelope spectral entropy analysis method can better extract the fault signal characteristics under complex working conditions than time-frequency domain image processing,and the genetic algorithm support vector machine recognition accuracy reaches 96.87%,and the algorithm model can be further applied to the diagnosis of hydro generator bearing faults.