基于CNN-BiLSTM的可解释性轴承故障诊断  

Interpretable Bearing Fault Diagnosis Based on CNN-BiLSTM

作　　者：刘之航 钟玉华 LIU Zhi-hang;ZHONG Yu-hua(School of Automotive and Traffic Engineering,Guangzhou City University of Technology,Guangzhou 510800,China;School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou 510641,China)

机构地区：[1]广州城市理工学院汽车与交通工程学院,广州510800 [2]华南理工大学机械与汽车工程学院,广州510641

出　　处：《组合机床与自动化加工技术》2022年第9期59-63,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：广东省特色创新类项目(自然科学类)(2020KTSCX204)。

摘　　要：基于卷积神经网络的轴承故障诊断过程往往无法解释,并且常规CNN的抗噪性较差。因此,提出了一种基于CNN-BiLSTM的可解释性轴承故障诊断模型。分别从仿真和实验两个方面对CNN的特征提取过程进行分析,采用Grad-CAM++方法以热力图的形式对该过程进行解释,并结合BiLSTM改进模型,提升了模型的诊断精度和抗噪性能。仿真分析表明,Grad-CAM++不仅可以用于二维图片,也能够应用于一维时间信号,具有通用性,并且其可视化结果可以充分表明网络的特征提取过程,具有解释性。实验结果表明,所提出的模型不仅具有解释性,同时相比其他类似网络的平均诊断精度更高,并且具有更好的抗噪性和鲁棒性,验证了该方法在处理滚动轴承故障诊断上的有效性。The process of bearing fault diagnosis is usually based on a convolutional neural network,but the process is often criticized as a"black box"which is often unexplainable,and when the signal-to-noise ratio is small,the diagnostic performance of conventional CNN is not satisfactory.Therefore,an interpretable bearing fault diagnosis model based on CNN-BiLSTM is proposed.In the feature extraction process of the convolutional neural network,this method uses Grad-CAM++to explain the process in the form of a heat map,and uses BiLSTM′s bidirectional analysis capabilities to improve the model’s diagnostic accuracy and anti-noise performance.Simulation analysis shows that Grad-CAM++can be used not only for two-dimensional pictures,but also for one-dimensional time signals,so the model is versatile,and the visualization results can fully show the feature extraction process of the network,so the model is explanatory.The experimental results show that the model proposed is not only explanatory,but also has higher average diagnostic accuracy than other similar networks,and has better noise resistance and robustness,which verifies the effectiveness of the method in dealing with rolling bearing fault diagnosis.

关 键 词：类激活映射 卷积神经网络 双向长短时记忆 故障诊断 

分 类 号：TH133.3[机械工程—机械制造及自动化] TG502[金属学及工艺—金属切削加工及机床]