采用空间和通道激励注意力机制优化ResNet-50的CFRP/TC4叠层材料钻削刀具磨损状态监测  

Tool Wear Condition Monitoring for Drilling CFRP/TC4 Laminated Materials Using scSE Optimised ResNet-50

作　　者：聂鹏 杨程越 彭新月 于家鹤 潘五九 NIE Peng;YANG Chengyue;PENG Xinyue;YU Jiahe;PAN Wujiu(School of Mechatronics Engineering,Shenyang Aerospace University,Shenyang,110136;Forty-sixth Factory,Shenyang Aircraft Industry(Group)Co.,Shenyang,110031)

机构地区：[1]沈阳航空航天大学机电工程学院,沈阳110136 [2]沈阳飞机工业(集团)有限公司四十六厂,沈阳110031

出　　处：《中国机械工程》2024年第10期1793-1801,共9页China Mechanical Engineering

基　　金：国家自然科学基金(52375113);辽宁省自然科学基金(2022-MS-298);辽宁省教育厅基金(LJKMZ20220531);沈阳市中青年科技创新人才项目(RC230309)。

摘　　要：针对碳纤维增强复合材料(CFRP)与钛合金组成的叠层材料在制备装配孔时存在刀具磨损严重的问题,提出了一种空间和通道激励注意力机制(scSE)优化深度残差神经网络(ResNet-50)的刀具磨损监测方法。开展钻削实验,采集钻削过程中的力和温度信号,信号经连续小波变换转换为小波尺度谱。搭建ResNet-50网络结构,从空间和通道双维度对卷积提取的特征图进行权重标定。研究结果表明,scSE可以从空间和通道两个维度做到增强有用特征,抑制无用特征,经scSE优化的网络结构识别准确度达到96.15%。Aiming at the problems of severe tool wear in the preparation of assembly holes for laminated materials consisting of carbon fibre reinforced composites and titanium alloys,a tool wear monitoring method with scSE optimised deep residual neural network(ResNet-50)was proposed.Drilling experiments were carried out to collect force and temperature signals during the drilling processes.The signals were converted to wavelet scale spectrum using continuous wavelet transform.The ResNet-50 network structure was built to calibrate the weights of the convolutionally extracted feature maps from both spatial and channel dimensions.The results show that scSE may enhance the useful features and suppress the useless features from both spatial and channel dimensions,and the recognition accuracy of the network structure optimised by scSE reaches 96.15%.

关 键 词：刀具磨损 连续小波变换 空间和通道激励注意力机制 深度残差神经网络 

分 类 号：TH164[机械工程—机械制造及自动化] V257[一般工业技术—材料科学与工程]