摘要
针对离散制造车间的瓶颈漂移现象导致的生产瓶颈预测困难问题,提出了基于注意力机制的准循环神经网络(quasi-recurrent neural networks,QRNN)瓶颈预测方法。首先,根据制造系统的生产特性量化制造单元的瓶颈程度;其次,以反映车间运行状态的时序数据为输入,通过QRNN网络的卷积结构并行提取信息特征,减少计算时间;训练融合注意力机制的预测模型,充分挥发各时刻状态信息的作用,精准的预测制造车间生产瓶颈位置;最后,以某航天机加车间为例,将所提方法与LSTM、QRNN预测方法进行对比分析,证实了所提预测方法的有效性。
Aiming at the problem of difficult bottleneck prediction caused by the bottleneck drift phenomenon in discrete manufacturing workshops,a prediction method of Quasi-Recurrent Neural Networks(QRNN)based on attention mechanism is proposed.Firstly,the bottleneck degree of the manufacturing cell is quantified according to the production characteristics of the manufacturing system.Secondly,the time-series data reflecting the workshop operation status was taken as the input,and the convolution structure of QRNN was used to extract the information features in parallel,which reduced the computing time.The prediction model combined with attention mechanism is trained to fully volatilize the role of state information and accurately locate production bottlenecks in manufacturing workshops.Finally,taking an aerospace machining workshop as an example,the proposed method is compared with LSTM and QRNN and the effectiveness of the proposed method is verified.
作者
汪伟丽
郭宇
刘道元
高瀚鹏
杨志伟
WANG Wei-li;GUO Yu;LIU Dao-yuan;GAO Han-peng;YANG Zhi-wei(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;Beijing Xinghang Electromechanical Equipment Co.,Ltd.,Beijing 100074,China)
出处
《组合机床与自动化加工技术》
北大核心
2022年第9期151-154,159,共5页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国防基础科研(JCKY2018203A001,JCKY2019204A004)。
关键词
离散制造车间
瓶颈漂移
注意力机制
QRNN
discrete manufacturing workshops
bottleneck drift
attentional mechanism
QRNN
