Abstract: In view of the problem that the feature components are easy to be submerged in noise at early stage of bearing fault, and the difficulty to obtain a large number of fault samples in practice, a bearing fault intelligent diagnosis method based on Multi-Resolution Singular Value Decomposition（MRSVD） and Variable Predictive Model based Class Discriminate（VPMCD） is put forward. The detail components that contain the fault features are extracted by using the MRSVD for decomposing the bearing acceleration vibration signal. The logarithmic normal distribution models are established to highlight the non Gauss feature of detail components. The feature vector is constructed by calculating logarithmic means and logarithmic standard deviations. It is used to identify the fault by VPMCD. The method is applied to the actual bearing fault diagnosis with local weak faults of outer, inner circle and balls, and the?fault type can be accurately identified with 98.75% accuracy. The result shows that the presented method is feasible and valid.

Key words: Multi-Resolution Singular Value Decomposition（MRSVD）, Variable Predictive Model based Class Discriminate（VPMCD）, fault diagnosis

摘要： 针对轴承早期微弱故障特征信息易被噪声掩盖和现实中难以获得大量典型故障样本的实际情况，提出了基于多分辨奇异值分解（Multi-Resolution Singular Value Decomposition，MRSVD）和变量预测模型模式识别（Variable Predictive Model based Class Discriminate，VPMCD）的轴承故障智能诊断方法。利用MRSVD对轴承加速度振动信号进行多层分解，提取包含故障特征的细节信息，建立对数正态分布模型，凸显细节信息中的非高斯特性，计算对数均值和对数标准差构造特征向量，并采用VPMCD方法进行故障识别。将该方法应用于实际轴承外圈、内圈、滚动体局部微弱故障状态下的故障诊断，结果显示：故障识别精度达到98.75%，证明了该方法的可行性和有效性。

关键词: 多分辨奇异值分解, 变量预测模型模式识别, 故障诊断