深度残差网络在滚动轴承故障诊断中的研究

Study on Fault Diagnosis of Rolling Bearing by Deep Residual Network

近年来,由于传统人工提取特征的方法不足以准确表征滚动轴承的健康状态,深度学习算法被逐渐应用于滚动轴承的故障诊断中,它能够自适应的从输入数据中学习出所需要的特征。其中,相较于普通的深度学习算法,深度残差网络通过恒等映射的方式可以大幅度降低模型的训练难度。因此,采用了一种用于滚动轴承故障诊断的深度残差网络(ResNet),它可以直接将原始振动信号作为模型的输入,通过池化层、残差模块和分类层相互连接,更加有效的挖掘信号特征之间的信息,从而增强了轴承振动信号的特征学习能力。实验结果表明,该模型能够达到99.75%的轴承故障诊断精度,实现了良好的故障分类任务,为以后的机械故障诊断研究提供了理论指导和借鉴。

In recent years,because traditional methods of artificially extracting features are insufficient to accurately characterize the health of rolling bearings,deep learning algorithms have been gradually applied to the fault diagnosis of rolling bearings,which can adaptively learn the required features from the input data.Among them,compared with the ordinary deep learning algorithm,the deep residual network can greatly reduce the training difficulty of the model by identical mapping.Therefore,a deep Residual Network(ResNet)for fault diagnosis of rolling bearings is adopted,which can directly input the original vibration signal as the input of the model,and connect the pooling layer,residual module and classification layer to mine the information between signal features more effectively,thus enhancing the feature learning ability of bearing vibration signals.The experimental results show that the accuracy of bearing fault diagnosis based on this model can reach 99.75%.It achieves a good task of fault classification and provides theoretical guidance and reference for future research on mechanical fault diagnosis.