Restricted work-in-process: A study of differences between Kanban and CONWIP

This article presents a simulation study over a small supply chain, where the amount of work-in-process (WIP) is restricted. The supply chain consists of five linked machines, or production facilities, with stochastic operation times. A number of test cases are made where the number of jobs in the machines and the buffer areas are restricted. The restrictions are designed both in the Kanban way, linked to every machine, and in the CONWIP way, connected only to the total production line. But no Kanban-cards and -cells are involved in our study, just restricted inventories between the machines. With the same amount of limited WIP, CONWIP-control compared to Kanban-control presents a higher throughput rate, less time between jobs out, but the jobs stay on average longer in the system. The stochastic operation times cause that the upstream machine sometimes consumes the jobs in a rate that the downstream machine does not catch up with, therefore all available storage room temporarily are not used. Kanban- and CONWIP-control presents the same amount of average outflow per time unit with the same variation in operation times and with the same amount of real average WIP. But Kanban-control causes a lower utilisation of present available storage room and storage equipment than CONWIP. The user of Kanban and CONWIP can only control maximum WIP and not average WIP; average WIP is a consequence of existing variations, so the difference is important. The coefficient of variation of the lead-times increases when WIP increases; this is very difficult to handle in practical applications. Restricted WIP that shortens the lead-time and decreases its variation is more important than if it is a "push" or "pull" system. Finally, it is argued that CONWIP-control is to prefer over Kanban-control in theory, but in practice there is a lack of CONWIP installation guidelines.