改进型深度残差网络在轴承故障诊断中的应用研究

Research of Improved Deep Residual Network in Bearing Fault Diagnosis

残差网络的残差层数过多不能很好的提取轴承振动信号特征,使得网络模型的故障识别率无法满足工业生产的需求。为了解决这一问题提出一种改进型深度残差网络模型进行轴承故障诊断的方法,通过减少网络的残差个数减轻了网络的冗余度。同时,在此网络的基础上添加了注意力机制模块,使得网络更有针对性地提取重要数据特征。之后将一维振动信号转换为二维灰度图,利用网络模型对图片样张进行特征提取并对网络模型进行充足的训练。为了增加实验的严谨性,使用改进后的模型和其他网络模型对同一个振动信号进行故障识别。实验结果表明,改进的模型故障识别准确率相比较于原模型和其他网络模型都有显著提高,为轴承故障诊断提供了一种更优的模型。

The excessive number of residual layers in the residual network makes it unable to extract the bearing vibration signal very well and the fault identification rate of the network model is too low to meet the needs of industrial production.This paper proposes the method of improved network model applied in bearing fault diagnosis.By reducing the number of residual of the network,the redundancy of the network is reduced.On the other hand,the attention mechanism is added to this network,which makes the network more targeted to extract important data features.After that,the one-dimensional vibration signal is converted into two-dimensional gray level map,and then the image sample is imported into the model for processing.In order to make the experiment more rigorous,the network model designed in this experiment and other models are used to process the same vibration signal.Experimental residual show that the fault recognition rate of this network model is signification improved compared with the original model and other models.It provides a better model for bearing fault diagnosis.