基于自编码卷积神经网络的机床状态聚类技术分析  

Research on State Clustering Technology of Machine Tools Based on Self-encoding Convolutional Neural Network

作　　者：刘启 林子超[1] 沈彬[1] 刘奇正 桂宇飞 LIU Qi;LIN Zichao;SHEN Bin;LIU Qizheng;GUI Yufei(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《机械设计与研究》2020年第3期141-147,共7页Machine Design And Research

基　　金：高档数控机床与基础制造装备科技重大专项(2018ZX04011001-007,2018ZX04005001-002);工信部智能制造新模式项目;工信部高技术船舶科研项目(CDGC01-KT0505)。

摘　　要：传统的机床故障诊断技术需要大量的、带标签的样本数据进行训练,而在实际工程当中,虽然可以获得海量的数据,但满足数量要求的带标签的样本数据是难以获得的。一种可行的思路是利用深度学习技术中无监督学习的方法对机床状态进行聚类,将对大量原始数据进行标签的问题简化为对若干类数据进行标签的问题。编制三组能够充分反映数控机床的设备状态的诊断程序,控制数控机床运行诊断程序,收集机床的运行数据,基于自编码卷积神经网络,无监督学习数据的内在特征,将原始数据降维至一定长度,利用SOM神经网络对所得特征向量进行聚类,实验结果表明,此方法能够很好的将机床的状态进行聚类,大幅度减少了工程人员对诊断数据进行标记所需花费的时间成本。Traditional machine tool fault diagnosis technology requires a large amount of labeled sample data for training.In actual engineering,although massive data can be obtained,labeled sample data that meets the quantity requirements is difficult to obtain.A feasible idea is to use the unsupervised learning method in deep learning technology to cluster the state of machine tools,and simplify the problem of labeling a large number of raw data into the problem of labeling several types of data.Three sets of diagnostic programs that can fully reflect the equipment status of CNC machine toolswerecompiled.Run the diagnostic programs and collect the operating data of the machine tools.Based on self-coded convolutional neural networks,learn the inherent characteristics of the data unsupervised,the original data is reduced to a certain length.The obtained feature vector is clustered using SOM neural network.The experimental results show that this method can cluster the state of the machine tool welland greatly reduces the time and cost required for engineers to mark the data.

关 键 词：数控机床 神经网络 SOM 状态聚类 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]