摘要：为了对轴承的运行状态进行有效的识别，以便进一步评估和预测轴承的寿命，提出了基于非广延小波特征尺度熵和Morlet小波核支持向量机（Morlet wavelet kernel support vector machine, MWSVM）的轴承运行状态识别的新方法。对采集到的轴承振动信号进行小波分解，得到相应的小波分解系数，在此基础上结合非广延熵理论提出了沿尺度分布的非广延小波尺度熵特征提取方法。但是通过小波特征尺度熵分析后获得的特征信息存在维数较高，特征信息间冗余严重的问题，因此，引入了流形学维数约简算法(locality preserving projection, LPP)进行敏感特征信息的提取，减少在特征信息提取过程中人为因素的干扰。以约简后的特征信息作为MWSVM的输入进行训练，建立轴承的状态识别模型，从而实现轴承状态的识别。通过对某轴承内圈正常状态和几种故障程度不同的状态进行识别，试验结果表明了方法的有效性。

Abstract: In order to effectively recognize the bearing running state, so as to estimate and forecast the bearing’s service life, a new method of bearing running state recognition based on non-extensive wavelet feature scale entropy and Morlet wavelet kernel support vector machine(MWSVM) was proposed. The gathered vibration signals of bearing were decomposed by the wavelet, and the corresponding wavelet coefficients were got. Based on the integration of non-extensive entropy theory and the wavelet coefficients, the wavelet feature scale entropy feature extraction method was got. But the features got through the wavelet feature scale entropy feature extraction method have the problems of high dimension and redundancy serious. Therefore, the manifold learning dimension reduction algorithm locality preserving projection was introduced to extract the characteristic features and reduce the interference of human factor. The characteristic features were inputted to the MWSVM to train and construct the bearing running state identification modal, so as to realize the bearing running state identification. The running states of a normal inner race and several inner races with different degree of fault were recognized through the proposed method, the results validate the effectiveness of the proposed algorithm.