Discrete event-driven model predictive control for real-time work-in-process optimization in serial production systems
Abstract Advanced technologies (e.g., distributed sensors, RFID, and auto-identification) can gather processing information (e.g., system status, uncertain machine breakdown, and uncertain job demand) accurately and in real-time. By combining this transparent, detailed, and real-time production information with production system physical properties, an intelligent event-driven feedback control can be designed to reschedule the release plan of jobs in real-time without work-in-process (WIP) explosion. This controller should obtain the operational benefits of pull (e.g., Toyota’s Kanban system) and still develop a coherent planning structure (e.g., MRPII). This paper focuses on this purpose by constructing a discrete event-driven model predictive control (e-MPC) for real-time WIP (r-WIP) optimization. The discrete e-MPC addresses three key modelling problems of serial production systems: (1) establish a max-plus linear model to describe dynamic transition behaviors of serial production systems, (2) formulate a model-based event-driven production loss identification method to provide feedback signals for r-WIP optimization, and (3) design a discrete e-MPC to generate the optimal job release plan. Based on a case from an industrial sewing machine production plant, the advantages of the discrete e-MPC are compared with the other two r-WIP control strategies: Kanban and MRPII. The results show that the discrete e-MPC can rapidly and cost-effectively reconfigure production logic. It can decrease the r-WIP without deteriorating system throughput. The proposed e-MPC utilizes the available transparent sensor data to facilitate real-time production decisions. The effort is a step forward in smart manufacturing to achieve improved system responsiveness and efficiency.