【作者】 张彪；

【导师】 唐昊； 承敏钢；

【作者基本信息】 合肥工业大学 ， 控制工程（专业学位）， 2020， 硕士

【摘要】 随着社会的发展,越来越多的现代化工厂采用自动化生产线以获得更高的生产效率,其中最有代表性的是液晶显示器(Liquid Crystal Display,简称LCD)制造行业。自动化贴合生产线被广泛应用在这些工厂中,因此具有重要的研究意义。本文以LCD制造中的全贴合生产线为研究对象开展相关优化控制研究,包括多作业站全贴合生产线的协同控制优化研究。全贴合生产线通常配备有加工中心和投放机构,其设计目的是将站点加工完成的产品投放到匀速运动的传送带上,以供下游使用。首先本文以站点缓冲库剩余量、前视区间内的产品分布信息和加工中心对在制品的已加工时间作为联合状态,以抓手采取卸载操作、等待操作或投放操作作为可选行动,构建了全贴合生产线的离散马尔可夫决策过程(Discrete-Time Markov Decision Process,简称DTMDP)模型。然后采用理论优化方法和仿真优化方法,即策略迭代算法和基于模拟退火的Q学习算法分别对该模型进行策略求解。然后做一系列对比实验,并对实验结果进行讨论,验证本模型和机制的有效性。另外,为了更好地满足生产实际需求,本文在上述改进的模型和机制的基础上对多作业站间的协同控制机制进行了研究。文中对多作业站全贴合生产线产品搬运模型使用基于性能势的多Agent Q学习算法进行仿真实验,验证了所提协同控制方法能够使系统在负载平衡性和加工中心稼动率方面得到有效改善。更多还原

【Abstract】 With the development of the society,more and more modern factories adopt the automatic production line to get higher production efficiency.LCD(Liquid Crystal Display)is the most representative industry.Automated laminating production lines are widely used in these factories.So the research of this production line has important significance.In this thesis,the related research is based on the full lamination of the LCD industry,including coordination control optimization research of multi-station system.The full laminating production line usually consists of machining center and feeding mechanism.The goal of this system is to put the finished product on the conveyor belt so as to satisfy next procedures.Firstly,this thesis setups a Discrete-Time Markov Decision Process(DTMDP)optimization model,considering vacancies of the bank,products position information and the processing time of the current products in the machining center as the state and considering the unloading operation,waiting operation or the feeding operation as the optional action.Next,we use the theory optimization method and simulation optimization method to find the optimal strategy of DTMDP model,through policy iteration algorithm and the Q-learning algorithm based on Simulated Annealing.After that,we made a series of comparative experiments.Furthermore,we analyzed these simulation results and verified the effectiveness of our mechanism.In addition,in order to satisfy the need of practical production line better,we further research the coordination control optimization of multi-station production line on the basic of improved model and mechanism.In this thesis,we use a multi-agent Q-learning algorithm based on performance potential to simulate the model of product handling of full lamination production line with multi-station.The results prove that our cooperative control method can improve the balance of system’s load rate and the activation of machining center.更多还原