GRU-BP在数字化车间关键部件寿命预测中的研究  

Research on GRU-BP for Life Prediction of Key Components in Digital Workshop

作　　者：郝俊虎 胡毅[1,3] 崔宁宁 韩丰羽 徐崇良 HAO Jun-hu;HU Yi;CUI Ning-ning;HAN Feng-yu;XU Chong-liang(Shenyang Institute of Computing Technology,Chinese Academy of Sciences,Shenyang 110168,China;School of Computer and Control Engineering,University of Chinese Academy of Sciences,Beijing 101408,China;Shenyang Golding NC Intelligent Tech.Co.Ltd.,Shenyang 110168,China)

机构地区：[1]中国科学院沈阳计算技术研究所,沈阳110168 [2]中国科学院大学计算机与控制学院,北京101408 [3]沈阳高精数控智能技术股份有限公司,沈阳110168

出　　处：《小型微型计算机系统》2020年第3期637-642,共6页Journal of Chinese Computer Systems

基　　金：2017年国家智能制造综合标准化项目(20172150299)资助.

摘　　要：准确的剩余使用寿命预测在数字化车间的预测和健康管理中起着至关重要的作用.如数字化车间安全保证中最重要的环节是故障的排查和维修,但是目前故障的排查和维修都是需要在停机的状态下完成,这不仅可能会影响生产加工的进度,还可能会增加人工干预的风险,剩余使用寿命预测则可以很好的解决这个问题.本文针对传统的剩余使用寿命预测方法缺乏输入时序的考虑,提出了一种基于GRU-BP的多对一双GRU层神经网络剩余使用寿命预测方法.并使用数控车间的铣削刀和轴承数据进行实验,验证了该方法在剩余使用寿命预测中可行,且精度优于输入时序无关的梯度增强方法 XGBoost和传统的BP神经网络.Accurate remaining useful life prediction plays a crucial role in the prediction and health management of digital workshops.For example,the most important part of the digital workshop safety assurance is the troubleshooting and repair of the fault,but the current troubleshooting and repair of the fault needs to be completed in the state of shutdown,which may not only affect the progress of production and processing,but also may increase manual intervention. The remaining useful life prediction can solve this problem. In view of the lack of input timing in the traditional residual useful life prediction method,this paper proposes a GRU-BP based multi-pair and GRU layer neural network remaining useful life prediction method. The experimental results show that the proposed method is feasible in the prediction of remaining useful life,and the accuracy is better than the gradient boosting method XGBoost and the traditional BP neural network without considering the input timing factor.

关 键 词：RUL预测 数字化车间 GRU 神经网络 工业大数据 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]