一种机械设备故障诊断的FD-Transformer方法  

A FD-Transformer method for fault diagnosis of mechanical equipment

作　　者：赵志宏[1,2] 李春秀 李乐豪 杨绍普 ZHAO Zhihong;LI Chunxiu;LI Lehao;YANG Shaopu(State Key Lab of Mechanical Behavior and System Safety of Traffic Engineering Structures,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;School of Information Science and Technology,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)

机构地区：[1]石家庄铁道大学省部共建交通工程结构力学行为与系统安全国家重点实验室,石家庄050043 [2]石家庄铁道大学信息科学与技术学院,石家庄050043

出　　处：《振动与冲击》2023年第8期326-333,共8页Journal of Vibration and Shock

基　　金：国家重点研发计划资助项目(2020YFB2007700);国家自然科学基金(11972236,11790282);石家庄铁道大学研究生创新资助项目(YC2022059)。

摘　　要：随着机械设备故障诊断技术的发展,利用深度学习技术判断设备故障类型越来越引起人们重视。目前,基于注意力机制的Transformer模型有着优于卷积神经网络(convolutional neural network,CNN)的特征提取能力且在自然语言处理及计算机视觉领域都得到成功的应用。该研究提出一种用于机械设备故障诊断的Transformer方法(fault diagnosis-Transformer,FD-Transformer)。首先,对原始振动信号利用Dropout技术进行数据增强,提高模型的泛化能力;然后,利用多通道一维卷积进行数据处理并得到矩阵形式;接着,利用Dense连接的Encoder结构进行机械设备的故障特征提取;最后,利用分类模块得到故障诊断结果。分别采用变转速轴承数据和轮对轴承数据对模型进行试验验证,试验结果表明,该模型在两种数据集上均达到99%以上的故障识别率,与CNN相比可以更好地提取机械设备故障特征,有工程应用价值。With the development of mechanical equipment fault diagnosis technology,people pay more and more attention to using deep learning technology to judge the type of equipment fault.At present,a Transformer model based on the attention mechanism has better feature extraction ability than the convolutional neural network(CNN),and has been successfully applied in the fields of natural language processing and computer vision.A Transformer method for mechanical equipment fault diagnosis(FD-Transformer)was proposed in this work.Firstly,the original vibration signal was enhanced by the Dropout technology to improve the generalization ability of the model.Then the matrix form was obtained by multi-channel one-dimensional convolution.The Encoder structure connected by Dense was used to extract the fault features of mechanical equipment.Finally,the fault diagnosis results were obtained by using the classification module.The experimental results show that the model achieves a fault recognition rate of more than 99%on both data sets.Compared with CNN,it can better extract the fault characteristics of mechanical equipment,and has certain engineering application value.

关 键 词：注意力机制 故障诊断 深度学习 

分 类 号：TH133.33[机械工程—机械制造及自动化]