On control strategies in a multi-stage production system

A multi-stage production system is composed of a set of stations, each station performing a given task, and a set of vehicles, each vehicle moving between two successive stations. A station can choose a buffer or a kanban mechanism for controlling the work-in-process (WIP) in the station. A vehicle can choose a push or a pull policy for carrying parts from its upstream station to its downstream station. A control strategy is formed by combining the WIP mechanisms adopted in all stations and the carrying policies employed by all vehicles. The production system is modelled as a queuing system. Some structural properties of performance measures are characterized. We develop a decomposition approach for large systems, which performs very well. We determine the optimal numbers of buffers or kanbans at all stations in the design period, and the optimal control strategy during operation. Many numerical computations are given for evaluating the efficiencies of the decomposition approach and optimization methods, and further providing some intuitions and insights.