Controlling discrete manufacturing processes using Kanban and Heijunka approaches

This paper deals with the control of discrete manufacturing processes whose individual manufacturing units are able to produce components of different types that are stored in separate buffers. One of the standard control methods for such processes uses the so-called Kanban scheme, where the currently produced type is only switched when one of the buffers is about to run empty. Recently, an alternative method started to receive increasing attention: the so-called Heijunka scheme that realizes strictly periodic schedules. We compare both methods and apply them to a simplified manufacturing process that mimics a real-world process at an electronic circuit board manufacturer. The experiments show that Kanban is very well suited for the startup phase of the process, because no knowledge about the expected process statistics is required, the buffer levels are lower, and the maximum ability to deliver is reached much faster than with the Heijunka method. After the startup phase, however, the experiments indicate that Heijunka is preferable, because then it yields lower buffer levels and higher average ability to deliver than Kanban. Thus, the optimal control strategy for the type of manufacturing processes under investigation initially uses Kanban and then switches over to Heijunka.