Data-driven Real-time Order Acceptance And Synchronized Decision-making For Production Process Control

The discrete manufacturing systems in make-to-order environments are nonlinear systems which are perturbed by various unknown dynamics.Thus,it is important to study the dynamic production process control and decision-making to improve the responsiveness of these systems.Currently,driven by various uncertainties in these systems,the prevailing researches worked on developing reactive methods for production process control and decision-making based on production capacity approximation.The precision of these reactive decisions is low and their lag is strong.With the development of sensors and information technologies,a large number of data about production processes can be perceived by manufacturing systems in real time.These data are turning traditional reactive decision-making into proactive one after measuring the accurate performance of manufacturing systems.This dissertation focuses on the key decision-making from order acceptance to its just-in-time delivery,and works on developing data-driven approaches for real-time order acceptance and synchronized production process control.The research results are beneficial for formulating proactive decision-making for production process control,and can improve the responsiveness of manufacturing systems efficiently.The main research work and innovations can be concluded as follows:(1)A data-driven virtualization model for discrete manufacturing systems is established.This virtualization provides accurate information of production capacity prediction for process control.Supported by manufacturing Internet of Things,the realtime data of production status and various uncertainties can be perceived efficiently.This dissertation establishes data-driven state-space equations for multi-input-oneoutput discrete manufacturing systems after analyzing the dynamic transition behaviors of jobs based on the max-plus algebra theory and timed event graphs.These equations can synchronously model the real-time status of manufacturing systems.The research results demonstrate that the state-space equations can make up the shortage of flow models in describing the stochastic switch among various products and finite intermediate buffer constraints,and in turn can predict dynamic production capacity more accurately.(2)A leader-follower interactive optimization for real-time order acceptance and scheduling is proposed.This optimization provides real-time objectives for production process control.Based on the real-time production status of manufacturing systems,this dissertation proposes a leader-follower interactive optimization to simultaneously handle the problems of order acceptance,scheduling,and job releasing after addressing the constraints from manufacturing network coordination.The objectives are to maximize the total net revenue while at the same time minimize the operation cost of manufacturing systems and order tardiness penalty cost.The research results reveal that the leader-follower interactive method can decrease the order processing cost obviously and can provide a more accurate objective for production process control compared with these step-by-step methodologies with production capacity approximation.(3)A production rescheduling approach driven by job-related uncertainties and a proactive optimal feedback control driven by resource-related uncertainties are formulated.Considering the job-related uncertainties,a joint programming model for order resequencing and job re-releasing is constructed based on the order acceptance and scheduling scheme and real-time status of manufacturing systems.This joint optimization is driven by real-time events and rolling time horizon at the same time.The objective is to simultaneously minimize the operation costs and order tardiness penalty cost.Additionally,an event-driven max-plus model predictive control(e-MPC)is formulated to address resourced-related uncertainties.In this e-MPC,a real-time permanent production loss identification approach is developed to provide feedback signal for e-MPC.After that,a max-plus model predictive control(m-MPC)is used to generate an optimal job releasing scheme with the constraints from manufacturing network coordination.This e-MPC formulates a proactive closed-loop control for resource-related uncertainties and can adjust the optimal job release plan adaptively.The research results show that this proactive method can fulfill the accepted orders in just-in-time with lower operation costs.(4)A synchronized dynamic and proactive in-house part-feeding driven by predictive replenishments is proposed.The part-feeding approach can react with production process control with lower operation cost.By analyzing the real-time information of lineside inventory,order rescheduling,and job re-releasing,a methodology for replenishment prediction in a decision horizon is formulated based on the Kanbanbased(t,ROP,Q)replenishment model.After that,a multi-trip dynamic routing programming with time window constraints is constructed to achieve the just-in-time part-feeding in proactive.Compared with existed passive part-feeding methods,the research results demonstrate that our proactive part-feeding can synchronize with dynamic production process control with lower urgent distribution cost and lineside inventory cost.In conclusion,the proposed data-driven approaches for real-time order acceptance and production process control can proactively response to various uncertainties of discrete manufacturing systems with lower operation costs.The research results are significant in developing the capability of data-driven perception,analysis,decisionmaking,and execution for intelligent manufacturing systems.