基于改进一维卷积神经网络的多轴工业机器人故障诊断  

Fault Diagnosis of Multi-axis Industrial Robot Based on Improved One-dimensional Convolutional Neural Network

作　　者：潘屹豪 肖红[1] 周玉彬 黎萍 姜文超[1,2] 熊梦 贺忠堂 PAN Yi-hao;XIAO Hong;ZHOU Yu-bin;LI Ping;JIANG Wen-chao;XIONG Meng;HE Zhong-tang(School of Computer Science and Technology,Guangdong University of Technology,Guangzhou 510006,China;Dongguan Cloud Computing Industry Technology Innovation and Incubation Center,Chinese Academy of Sciences,Dongguan Guangdong 523808,China)

机构地区：[1]广东工业大学计算机学院,广州510006 [2]东莞中国科学院云计算产业技术创新与育成中心,广东东莞523808

出　　处：《组合机床与自动化加工技术》2021年第12期10-14,18,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：2019年佛山市核心技术攻关项目(1920001001367);国家自然科学基金-广东省联合基金项目(U2001201);广东省自然科学基金面上项目(2020A1515010890,2018A030313061);国家重点领域研发计划项目(2018YFB1004202);广东省科技计划项目(2019B010139001);广州市科技计划项目(201902020016);广东省基础与应用基础研究基金项目(2020B1515120010)。

摘　　要：针对工业机器人运行过程中故障诊断精度较低、速度较慢的问题,提出一种基于改进一维卷积神经网络的多轴工业机器人故障诊断模型正交正则化一维卷积神经网络(SRIPCNN-1D)。首先,通过随机采样和Mixup对工业机器人故障数据进行增强;其次,采用正交正则化(SRIP)一维卷积神经网络(CNN-1D)将工业机器人原始运行数据进行端到端训练;最后,利用训练后的模型对工业机器人进行快速故障诊断。实验平台采用某国产机器人设备,采集力矩、速度、位置、电流等运行变量数据300万条,对诊断精度指标进行实验测试,并与WDCNN,CNN-1D模型进行实验对比,结果表明SRIPCNN-1D方法可以有效诊断工业机器人故障。Aiming at the problem of low fault diagnosis accuracy and slow speed during the operation of industrial robots,a multi-axis industrial robot fault diagnosis model based on improved one-dimensional convolutional neural network is proposed.Orthogonal regularized one-dimensional convolutional neural network(SRIPCNN-1D).First,the fault data of industrial robots is enhanced by random sampling and Mixup;then,the original operating data of industrial robots is trained end-to-end by using orthogonal regularization(SRIP)one-dimensional convolutional neural network(CNN-1D);finally,Use the trained model to perform rapid fault diagnosis on industrial robots.The experimental platform adopts a domestic robot equipment,collects 3 million pieces of operating variable data such as torque,speed,position,current,etc.,performs experimental tests on diagnostic accuracy indicators,and compares them with WDCNN and CNN-1D models.The results show that the SRIPCNN-1D method It can effectively diagnose faults of industrial robots.

关 键 词：多轴工业机器人 故障诊断 一维卷积神经网络 正交正则化 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]