基于1DCNN的齿轮箱小样本故障诊断  

Small Sample Fault Diagnosis of Gearbox Based on 1DCNN

作　　者：钟建华 林云树 叶锦华[1] ZHONG Jian-hua;LIN Yun-shu;YE Jin-hua(School of Mechanical Engineering and Automation,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学机械工程及自动化学院,福州350108

出　　处：《组合机床与自动化加工技术》2022年第7期81-84,89,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：福建省自然科学基金项目(2019J01211);国家工信部智能制造综合标准化与新模式应用项目(GXSP20181001)。

摘　　要：齿轮箱故障诊断对于降低运维成本和提高设备运转效率至关重要。首先,提出了一种基于小样本数据的一维卷积神经网络(1DCNN)端到端故障诊断方法;针对小样本故障诊断,采用LeNet-5结构增加卷积层数量,增大特征提取能力;其次,通过动力传动故障诊断综合实验台(DDS)数据验证;最后,引入t-SNE技术,对部分层输出进行了可视化,进一步验证了模型的有效性。此外,通过不同参数组合验证了模型所设参数的合理性,实验结果表明,所提方法与传统LeNet-5和基于EEMD和VMD特征提取方法的SVM分类器对比,在分类准确率上分别有7.5%、11.25%和5%的提升,证明了所提方法的有效性。Fault diagnosis of the gearbox is essential to reduce operation and maintenance costs and improve equipment operation efficiency.First,An end to end fault diagnosis method of one-dimensional convolutional neural network(1DCNN)based on small sample data is proposed.For small sample fault diagnosis,the LeNet-5 structure is used to increase the number of convolutional layers and increase the feature extraction ability.Secondly,it is verified by the data of drivetrain diagnostics simulator(DDS).Finally,the t-SNE technology is introduced to visualize the output of some layers,which further verifies the effectiveness of the model.Besides,the rationality of the parameters set by the model is verified through different parameter combinations.The experimental results show that the classification accuracy is improved by 7.5%,11.25%and 5%respectively by comparing with the traditional LeNet-5 and the SVM classifier based on EEMD and VMD feature extraction methods,which proves the effectiveness of the proposed method.

关 键 词：1DCNN 小样本 齿轮箱 端到端 故障诊断 

分 类 号：TH165[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]