基于Attention-CGDNN模型的滚动轴承故障诊断方法  

Fault Diagnosis Method for Rolling Bearings Based on Attention-CGDNN Model

作　　者：段智峰 谢丽蓉[1] 崔传世 殷秀丽 包洪印 DUAN Zhifeng;XIE Lirong;CUI Chuanshi;YIN Xiuli;BAO Hongyin(Wind Storage Division of State Key Laboratory of Power System and Generation Equipment Control and Simulation,Xinjiang University,Urumqi 830047,China;CSIC Haiwei(Xinjiang)New Energy Co.,Ltd.,Urumqi 830002,China)

机构地区：[1]新疆大学电力系统及发电设备控制和仿真国家重点实验室风光储分室,乌鲁木齐830047 [2]中船重工海为(新疆)新能源有限公司,乌鲁木齐830002

出　　处：《轴承》2023年第4期80-86,共7页Bearing

基　　金：国家自然科学基金资助项目(62163034)。

摘　　要：针对传统轴承故障诊断方法依赖人工进行特征提取时效率低且难以处理大规模数据等问题,将卷积长短时深度神经网络(CLDNN)引入轴承故障诊断并进行改进,提出一种基于注意力机制的卷积门控深度神经网络(Attention-CGDNN)的滚动轴承故障诊断模型,该模型将卷积神经网络、门控循环单元和全连接神经网络有效融合以实现滚动轴承信号特征提取,并加入注意力机制使网络更专注于重要特征,最后通过Softmax分类算法实现滚动轴承故障诊断。采用CWRU和XJTY-SY轴承数据集的验证结果表明,Attention-CGDNN模型具有训练参数少,训练难度小,收敛速度快和识别精度高的特点,特征提取能力更强,故障诊断性能优于传统模型。Aimed at the problems that traditional fault diagnosis methods for bearings rely on manual feature extraction,such as low extraction efficiency and difficulty in handling large-scale data,the convolutional neural-long short-term memory-deep neural networks(CLDNN)is introduced into fault diagnosis of bearings and the improvement is undertaken,a fault diagnosis model for rolling bearings is proposed based on attention-convolutional gate recurrent unit depth neural networks(Attention-CGDNN).In this model,convolutional neural network,gated cycle unit and fully connected neural network are effectively integrated to achieve feature extraction of rolling bearing signal,and attention mechanism is added to make the network more focused on important features.Finally,the fault diagnosis for rolling bearings is realized by SoftMax classification algorithm.The verification results using CWRU and XJTU-SY bearing datasets show that the Attention-CGDNN model has the characteristics of few training parameters,small training difficulty,fast convergence speed and high recognition accuracy,stronger feature extraction ability,and better fault diagnosis performance than traditional models.

关 键 词：滚动轴承 故障诊断 深度学习 神经网络 注意力机制 特征提取 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]