基于一维深度残差收缩网络的汇流行星排齿轮裂纹故障诊断  

Gear crack fault diagnosis of convergent planetary rowbased on 1-D depth residual shrinkage network

作　　者：田钦文 冯辅周 李鸣 陈晓明 朱俊臻 胡浩 宋超 TIAN Qinwen;FENG Fuzhou;LI Ming;CHEN Xiaoming;ZHU Junzhen;HU Hao;SONG Chao(PLA31628,Shaoguan 512000,China;Army Armored Forces Academy,Beijing 100072,China)

机构地区：[1]中国人民解放军31628部队,广东韶关512000 [2]陆军装甲兵学院,北京100072

出　　处：《振动与冲击》2022年第19期198-206,共9页Journal of Vibration and Shock

基　　金：国家自然科学基金(51875576,52005510);无损检测技术教育部重点实验室开放基金项目(EW201980445)。

摘　　要：装甲车辆汇流行星排出现裂纹时,箱体表面振动信号干扰较多,常见的故障诊断方法存在偏差。为此,提出一种利用一维深度残差收缩网络的汇流行星排齿轮裂纹故障诊断模型;其特点是将注意力机制与软阈值化结合,增强有用信息,抑制冗余信息,并将其引入到残差神经网络中,显著提高模型特征提取的能力;为验证该模型的可行性,采集了行星轮四种不同程度裂纹的振动信号作为样本用于故障诊断。结果表明,针对汇流行星排齿轮箱振动信号该方法可以在更短的时间取得更高的准确率,相较其他方法,可以取得更好的分类结果。When gear cracks appear on convergent planetary row of armored vehicles,vibration signals on box body surface have more disturbances,and the results obtained with common fault diagnosis methods have some deviations.Here,a fault diagnosis model for gear cracks of convergent planetary row using 1-D depth residual shrinkage network was proposed.Its keywas to combine attention mechanism with soft thresholding,enhance useful information,suppress redundant information,introduce itself into a residual neural network,and significantly improve the ability of extracting model features.In order to verify the feasibility of the proposed model,vibration signals for4 gear cracks with different levelson planetary row were collected as samples for fault diagnosis.The results showed that the proposed method can achieve higher accuracy within a shorter time and better fault classification results than other methods can.

关 键 词：深度学习 汇流行星排 故障诊断 

分 类 号：TH17[机械工程—机械制造及自动化]