The vibration signal of hydraulic generator is non-stationary. Features of the early fault signal are weak and thus are difficult to be extracted. In this paper, features of the bearing vibration signal for fault diagnosis are extracted by using the variational mode decomposition (VMD) and singular value. Fault diagnosis is carried out by using the support vector machine (SVM). Firstly, several intrinsic mode functions (IMFs) are obtained by performing VMD on the bearing vibration signal. Then, singular values of the modal component matrix constituted by the intrinsic mode functions are calculated, which are regarded as the feature vector input to the support vector machine. Finally the fault classification and recognition are done by the support vector machine. The proposed method is verified by analyzing the rolling bearing experimental data. The vibration data of the near Wake Island Hydropower Station in Hunan province are used to test the accuracy of the proposed method in practical application.

水力发电机的振动信号不稳定。早期故障信号的特征较弱，因此难以提取。本文利用变分模式分解（VMD）和奇异值提取了用于故障诊断的轴承振动信号的特征。通过使用支持向量机（SVM）进行故障诊断。首先，通过对轴承振动信号执行VMD，可以获得几个固有模式函数（IMF）。然后，计算由固有模式函数构成的模态分量矩阵的奇异值，将其视为输入到支持向量机的特征向量。最后，通过支持向量机对故障进行分类和识别。通过对滚动轴承的实验数据进行分析，验证了该方法的有效性。