基于特征生成的轴承不均衡数据故障诊断

Fault Diagnosis for Bearings of Unbalanced Data Based on Feature Generation

针对实际生产中难以获得足量的故障样本数据导致训练中样本不均衡、样本不足等问题,提出了一种基于特征聚类的过采样算法,并将其与卷积神经网络相结合的滚动轴承故障诊断模型。该模型将频域信号作为模型的输入,通过卷积神经网络进行特征提取,再通过过采样技术生成新的特征数据实现数据的均衡化,将新生成的特征数据和原有特征一同输入到支持向量机(SupportVector Machine,SVM)分类器中完成样本的分类,实现滚动轴承的故障诊断。通过对比实验,结果表明该方法可以有效解决数据不均衡的问题。

Focus on the sample imbalance and insufficiency caused by the difficulty to obtain a sufficient number of fault samples in actual production. A model for rolling bearings by combining Convolutional Neural Networks and Synthetic Oversampling is presented. The frequency domain signals is used as the input of the model, and the features are extracted by the Convolutional Neural Network. The new features are generated by Synthetic Oversampling and the data equalization is realized. The model completes the classification by putting all of the features into the Support Vector Machine, and the fault diagnosis of the rolling bearings is carried out. The comparison experiments results show that the method can effectively solve the problem of data imbalance.