基于改进卷积神经网络的滚动轴承故障诊断

Fault Diagnosis of Bolling Bearing Based on Improved Convolutional Neural Network

针对滚动轴承故障诊断过程中,难以提取细微故障特征的问题,文章提出一种基于改进卷积神经网络的滚动轴承故障诊断方法。该方法首先在特征值提取过程中,采用了多尺度卷积核并联的方式,对滚动轴承振动信号提取了更多的故障特征细节;然后在特征值降维、去噪处理过程中,采用了leaky_relu激活函数,解决了部分神经元处于抑制状态的问题;最后在分类识别过程中,针对多层全连接计算复杂的问题,采用了全局平均池化代替部分全连接。通过滚动轴承不同损伤程度、不同故障位置的诊断实验,证明了所提方法能够提高故障识别率、降低训练时间、具有较好的可行性。

In view of the problem that it is difficult to extract subtle fault features in the fault diagnosis process of rolling bearing,this paper proposes a fault diagnosis method of rolling bearing based on improved convolutional neural network.Firstly,in the process of eigenvalue extraction,the multi-scale convolution kernel parallel connection is adopted to excavate more fault feature details of rolling bearing vibration signal.Then,in the process of eigenvalue dimensionality reduction and denoising,leaky_relu activation function was adopted to solve the problem that some neurons were in a state of inhibition.Finally,in the process of classification and identification,global average pooling is adopted instead of partial full connection to solve the complex calculation problem of multilayer full connection.It is proved that the proposed method can improve the fault identification rate,reduce the training time and have better feasibility through the diagnosis experiment of rolling bearing with different damage degree and fault position.