Performance evaluation of kanban multiple-product production systems

Abstract

Kanban systems provide a way to implement a pull-type control policy. In this paper, we propose an analytical method for approximating the performance of a multiple-product kanban production system. This method uses the queueing network model of the kanban system, which is a closed multiclass queueing network (each class represents one type of kanban). It is based on the product-form approximation technique and is an extension of a previous method which had been programmed for mono-product kanban production systems with general topology. For multiple-product production systems, we can have several different types of products on a machine (we have restricted our study to only two types). So, we have a new problem: we must define the processing order of products. We can distinguish two cases according to set-up times. If set-up times are not null, we must try to limit the number of set-ups. But, if set-up times are null, we can choose the processing order without worrying about the limitation of set-ups number. We have realized the analysis for this second case in characterizing service time distributions with two moments. Finally, to test the method, we have compared analytical results to simulation results.