基于余弦下降学习率的卷积网络轴承故障诊断

Bearing Fault Diagnosis Based on Cosine Descent Learning Rate Convolution Network

滚动轴承在旋转机械中起到重要作用,其工作环境多变并存在多种失效模式。传统的轴承故障诊断方法需要依赖技术人员的经验提取时频域故障特征,并且诊断准确率不高。提出了一种基于余弦下降学习率的卷积神经网络模型,该模型以数据驱动方式自主提取轴承的故障特征。在训练的过程中采用余弦下降学习率,使损失函数可以快速收敛完成分类模型的训练。应用凯斯西储大学(CWRU)滚动轴承数据集验证该方法的有效性。结果显示,本模型能够有效识别轴承不同损伤程度以及不同故障,整体识别率达到99.91%。

Rolling bearing plays a significant role in rotating machinery equipment.Its working environment is complex and there are many failure modes.Conventional bearing fault diagnosis methods depend on the experience of technicians to extract time-frequency fault features,and the diagnosis accuracy is not high.In this paper,a convolutional neural network model based on cosine descent learning rate strategy is proposed.The model extracts the fault features of bearings by data-driven method.And in the process of training,the cosine descent learning rate is used to make the loss function converge quickly and complete the training of classification model.The effectiveness of this method is verified by using the rolling bearing data set of Case Western Reserve University(CWRU).The results show that the proposed method can effectively identify different degrees of damage and different faults of bearings,and the overall recognition rate reaches 99.91%.