基于门控循环残差网络的滚动轴承故障诊断研究

Research on Fault Diagnosis of Rolling Bearing Based on GRU Residual Network

滚动轴承的运行状态对整体机构的工作状态影响很大,防止因滚动轴承失效而产生的安全事故极为重要。而一维信号只利用卷积神经网络CNN(Convolutional Neural Networks)输出结果时无法充分利用数据间的时序信息的问题,因此,文中结合门控循环单元GRU(Gated Recurrent Unit)在处理时序数据所具有的优势,提出了一种门控循环残差网络结构,将CNN在强大的特征提取的优点与GRU处理时序数据的优点有机结合起来。为了验证算法的有效性,采用凯斯西储大学轴承数据集与齿轮箱轴承台架试验进行轴承故障诊断分析,同时引入常见神经网络作为对比,检验不同模型的分类性能。结果表明,在相同试验条件下相较于卷积神经网络等深度学习网络,文中算法具有更高的故障识别准确度和稳定性。

The running state of the rolling bearing has great influence on working state of the overall mechanism,and it is extremely important to prevent safety accidents caused by the failure of the rolling bearing.The one-dimensional signal only uses the convolutional neural network CNN(Convolutional Neural Networks,CNN)when the output result cannot make full use of the timing information between the data.Therefore,this article combines the Gated Recurrent Unit(GRU)to deal with advantages of time series data,a gated cyclic residual network structure is proposed,which organically combines the advantages of CNN in powerful feature extraction with the advantages of GRU processing time series data.In order to approve effectiveness of the algorithm,the Case Western Reserve University bearing data set and the gearbox bearing bench test were used to perform bearing fault diagnosis and analysis.At the same time,common neural networks were introduced for comparison to test the classification performance of different models.The results show that under the same experimental conditions,compared with deep learning networks such as convolutional neural networks,the algorithm in this paper has higher fault recognition accuracy and stability.