基于ESOINN的离散制造车间生产异常实时检测方法研究  

A Research on Real-Time Detection Method of Production Abnormalities in Discrete Manufacturing Workshop Based on ESOINN

作　　者：崔世婷 郭宇[1] 汪伟丽 梁睿君[1] CUI Shi-ting;GUO Yu;WANG Wei-li;LIANG Rui-jun(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Jiangsu Nanjing 210016,China)

机构地区：[1]南京航空航天大学机电学院,江苏南京210016

出　　处：《机械设计与制造》2023年第8期104-109,共6页Machinery Design & Manufacture

基　　金：国防基础科研(JCKY2018203A001);国防基础科研(JCKY2018605C003)。

摘　　要：离散制造车间存在制造要素偏离生产计划导致的生产异常事件,准确的车间异常检测有助于实时监控生产过程,提高动态决策响应速度,保证订单按时交付。针对异常检测的准确性和实时性需求,提出一种增量式无监督学习的车间生产异常检测方法。首先,以在制品在车间的流转过程为主线定义生产异常种类,搭建离散制造车间生产异常检测框架;其次,使用增强自组织增量神经网络实时检测生产异常,并根据当前生产数据在线更新模型,以适应数据分布的动态变化,提高模型检测准确率;最后以某航天机加车间为例,将所提方法与两种增量式及两种非增量式聚类算法进行对比实验,并在离散制造车间应用生产异常检测系统,验证了该方法在生产异常检测问题上的有效性。Production abnormalities caused by deviation of manufacturing elements from the production plan occur in discrete manufacturing workshop.In response to the accuracy and real-time requirements of anomaly detection,an incremental unsupervised learning method for worhshop production anomaly detection is proposed.Firstly,the production abnormalities is defined based on the flow of work in process in the workshop.Secondly,the enhanced self-organizing incremental neural network is used to detect production abnormalities in real time,and the detection model is updated online based on the current production data to adapt to the dynamic changes of data distribution and improve the accuracy of model.Finally,taking an aerospace aircraft workshop as an example,comparing the results of proposed method with two incremental and two non-incremental clustering algorithms and developing a production anomaly detection system for discrete manufacturing workshop show that the proposed method is effective in the production anomaly detection.

关 键 词：离散制造车间 生产异常检测 增量学习 增强自组织增量神经网络 

分 类 号：TH16[机械工程—机械制造及自动化] TP3[自动化与计算机技术—计算机科学与技术]