基于峭度与IMF能量融合特征和LS-SVM的齿轮故障诊断研究

Gear fault diagnosis research based on kurtosis and IMF energy feature fusion and least squares support vector machine

针对齿轮振动信号非线性非平稳特性,为避免传统时频方法在表征设备状态时的不足,提出一种基于融合峭度与IMF能量特征和LS-SVM的齿轮故障诊断方法。首先,对齿轮振动信号在EMD分解;然后,提取包含主要故障信息的IMF分量的峭度特征和能量特征,组成融合特征向量;最后,将齿轮正常、齿根裂纹、断齿3种状态下的融合特征向量输入到LS-SVM,通过训练好的LS-SVM对齿轮状态进行分类识别。仿真实验结果表明:该方法能准确识别齿轮的工作状态,且与BP神经网络、SVM相比,有着更高的故障识别效率,可用于齿轮信号的故障诊断。

Gear vibration signals have nonlinear and non-stationary characteristics. To avoid the disadvantages of existing time and frequency domain methods in the characterization of equipment state,this paper has been proposed a gearbox fault diagnosis method based on kurtosis and IMF energy feature fusion and least squares support vector machine. First, the gear vibration signals were decomposed by the EMD method. Second, the IMF components which contain major fault information were extracted and their energy and kurtosis feature calculated as fusion vectors.Third,the fusion feature vectors of three teeth conditions,viz.,normal,root crack and broken,were input to the least squares support vector machine（LS-SVM） to classify and identify gearbox faults. The simulation results show that this method can accurately identify the gear working state and more efficient in fault identification compared with BP neural network and SVM. It can be used for diagnosing gear signal faults.