基于GAF-inceptionResNet的齿轮箱故障诊断  

Gearbox Fault Diagnosis based on GAF-inceptionResNet

作　　者：李长文[1] 李鹏 丁华 Li Changwen;Li Peng;Ding Hua(Department of Information Technology,Shanxi Professional College of Finance,Taiyuan 030008,China;College of Mechanical and Vehicle Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Shanxi Key Laboratory of Fully Mechanized Coal Mining Equipment,Taiyuan 030024,China)

机构地区：[1]山西金融职业学院信息技术系,山西太原030008 [2]太原理工大学机械与运载工程学院,山西太原030024 [3]煤矿综采装备山西省重点实验室,山西太原030024

出　　处：《机械传动》2022年第6期134-140,共7页Journal of Mechanical Transmission

基　　金：山西省重点研发项目(201903D121064)。

摘　　要：为了提高齿轮箱故障诊断的准确率,准确表达齿轮箱的健康状态,结合深度学习算法,提出了一种用于齿轮故障诊断的GAF-inceptionResNet模型。该模型可以直接将原始一维振动信号经过格拉姆角场变换后形成图像作为模型的输入,通过Stem-block、残差Inception、残差模块和分类层相互连接。残差Inception网络能够拓宽网络深度,提升训练时长及准确率;残差模块利用恒等映射可以大幅度降低模型的训练难度。因此,该模型可有效地挖掘信号特征之间的信息,使模型的特征学习能力增强,从而提高准确率,精准确定故障。实验结果表明,该模型能够达到99.59%的故障诊断精度,有效实现齿轮箱良好的故障识别与分类。In order to improve the accuracy of gearbox fault diagnosis and accurately express the health status of the gearbox,combined with deep learning algorithms,a GAF-inceptionResNet model for gear fault diagnosis is proposed.The model can directly take the original one-dimensional vibration signal after GAF transformation to form photos as the input of the model.Through the stem-block,residual inception,residual module and classification layer,the residual inception network can broaden the network depth and improve the training time and accuracy,the residual block uses identity mapping to greatly reduce the training difficulty of the model.Therefore,the model can effectively mine the information between the signal features and enhance the feature learning ability of the model,thereby improving accuracy and accurately determine the faults.The test results show that the model can achieve a fault diagnosis accuracy of 99.59%.It can effectively achieve good gearbox fault identification and classification.

关 键 词：齿轮箱 故障诊断 格拉姆角场 振动信号 深度残差网络 

分 类 号：TH132.41[机械工程—机械制造及自动化]