孪生Transformer编码胶囊数控机床主轴故障分类网络研究  

Research on Spindle Fault Classification Network of Twin Transformer Coded Capsule CNC Machine tool

作　　者：孙惠娟[1] 邓聪颖 SUN Huijuan;DENG Congying(School of Mechanical Engineering,Chongqing Industry Polytechnic College,Chongqing 401120,China;School of Advanced Manufacturing Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆工业职业技术学院机械工程学院,重庆401120 [2]重庆邮电大学先进制造工程学院,重庆400065

出　　处：《机床与液压》2024年第22期103-109,共7页Machine Tool & Hydraulics

基　　金：重庆市自然科学基金面上项目(CSTB2022NSCQ-MSX1029);重庆市教育委员会科学技术研究计划青年项目(KJQN202203204);重庆工业职业技术学院博士科研基金项目(2023GZYBSZK2-06)。

摘　　要：主轴作为数控机床最重要的机械模块之一,及时检测其故障可保障机床的运转效能和加工精度。由此,提出一种孪生Transformer编码胶囊数控机床主轴故障分类网络。利用二维化预处理模块,得到较完整的原始数控机床主轴轴承振动数据;采用改进Transformer编码特征提取模块,获得深层次振动信号特征;通过高级胶囊特征转移网络实现特征映射;最后,使用孪生Transformer编码胶囊分类网络完成数控机床主轴故障样本的分类。选择XK7145型铣床完成健康轴承、内外圈故障轴承及滚珠故障下的无磨损刀具与磨损刀具故障诊断实验。结果表明:文中方法的平均主轴故障诊断准确率可达95.1%,相对于ISERAVF-net、VSCPC-net方法的平均准确率升高6.9%和12.3%,且文中方法的可视化分类效果较优,采用文中方法检测主轴故障的实验效果更佳。Spindle is one of the most important mechanical modules of CNC machine tools,timely detection of its faults can ensure the operation efficiency and machining accuracy of the machine tools.Therefore,a twin Transformer coding capsule CNC machine tool spindle fault classification network was proposed.The 2D preprocessing module was applied to obtain more completed vibration data of the spindle bearing of the original CNC machine tool.Then,an improved Transformer code feature extraction module was used to obtain deep-level vibration signal features.The feature mapping was realized through advanced capsule feature transfer network.Finally,the twin Transformer coding capsule classification network was used to complete the classification of CNC machine spindle fault samples.The XK7145 type milling machine was used to complete the fault diagnosis experiments of non-wear tool and wear tool under the conditions of healthy bearing,inner and outer ring fault bearing and ball fault.The results show that the average spindle fault diagnosis accuracy of the proposed method can reach 95.1%,which is 6.9%and 12.3%higher than the methods of ISERAVF-net and VSCPC-net.The visual classification effect of the proposed method is better,and the experimental effect of detecting spindle faults is better.

关 键 词：主轴 Transformer编码胶囊 铣床 可视化 

分 类 号：TG502[金属学及工艺—金属切削加工及机床]