数据驱动的加工产线生产周期预测研究
Data-Driven Production Line Cycle Time Prediction Study
张翔宇 1李想 2王伟 2杨旭2
作者信息
- 1. 中国科学院沈阳自动化研究所,沈阳 110016;中国科学院机器人与智能制造创新研究院,沈阳 110169;中国科学院大学,北京 100049;辽宁省智能检测与装备技术重点实验室,沈阳 110169
- 2. 中国科学院沈阳自动化研究所,沈阳 110016;中国科学院机器人与智能制造创新研究院,沈阳 110169;辽宁省智能检测与装备技术重点实验室,沈阳 110169
- 折叠
摘要
针对生产周期中的不可控时间、生产线数据结构复杂和构成变化的问题,提出一种以异构数据融合驱动的生产线周期预测方法.基于隶属函数和动态规整算法在三角特征构型下建立异构数据融合算法,将异构数据统一;以深度置信网络为主建立周期预测模型进行生产周期预测,通过确定因素求解生产周期中的不可控时间,用改进方程沟通粒子群算法和神经网络,每次迭代优化神经网络的同时还会反馈改进粒子群算法.以某航空企业加工产线为例,优化后神经网络正确率提升9%,粒子群训练时间减少5%,预测模型实现了生产周期的预测,且模型能够适应产线构成的变化.
Abstract
In order to solve the problems of uncontrollable time in the cycle time,complex data composi-tion and change of production line composition,a production line cycle time prediction method driven by heterogeneous data is proposed.Based on the membership function and the dynamic time warping,the het-erogeneous data fusion algorithm is established under the triangular characteristic configuration,in this way unify heterogeneous data under the same structure.The deep belief networks is used to establish the cycle time prediction model,solve the uncontrollable time composition of the cycle time by certain factors,com-municate the particle swarm optimization algorithm and the neural network through the improved equation,which could get neural network optimized iteratively every time and use the feedback to improve the parti-cle swarm optimization algorithm.Taking an aviation processing production line as an example,the opti-mized neural network accuracy increases by 9%,the particle swarm training time decreases by 5%,and the prediction model can predict the cycle time,also adapt to the changes of production line composition.
关键词
特征融合/粒子群算法/深度置信网络/生产周期预测Key words
feature fusion/particle swarm optimization/deep belief networks/cycle time prediction引用本文复制引用
基金项目
中央引导地方科技发展资金项目(2022JH6/100100061)
辽宁省应用基础研究计划(2022JH2/101300199)
出版年
2024
NETL
NSTL
万方数据