Extended kanban control system: combining kanban and base stock YVES DALLERY 1 and GEORGE LIBEROPOULOS 2 1 Laboratoire d'Informatique de Paris 6 (LIP6), Universite Pierre et Marie Curie, 8, rue du Capitaine Scott, FR-75015 Paris, France E-mail: Yves.Dallery@Lip6.fr 2 Department of Mechanical and Industrial Engineering, University of Thessaly, Pedion Areos, GR-38334 Volos, Greece E-mail: glib@mie.uth.gr Received March 1998 and accepted May 1999 This paper introduces a new mechanism for the coordination of multi-stage manufacturing systems. This mechanism is called the Extended Kanban Control System (EKCS) and depends on two parameters per stage, the number of kanbans and the base stock of ®nished parts. The EKCS is a combination of the classical Kanban and Base Stock control systems and includes each system as a special case. The dynamics of the EKCS are described, in particular, in relation to the dynamics of the Generalized Kanban Control System (GKCS), a known control mechanism that also includes the Kanban and Base Stock control systems as special cases. Advantages of the EKCS over the GKCS are discussed. Finally, properties of the dynamics of the EKCS are presented. One important property is that the capacity of the EKCS depends only on the number of kanbans but not on the base stock of ®nished parts. 1. Introduction In many manufacturing systems the processing of parts proceeds in stages. Each stage may be seen as a produc- tion/inventory system consisting of a manufacturing process and an output buer. The manufacturing process may consist of a single machine or a subnetwork of sev- eral machines (e.g., a production line, a job shop, a manufacturing cell, etc.) that performs a distinct pro- cessing operation (e.g., machining, forming, assembly, inspection, testing, etc.) on parts that it receives from the output buer of the preceeding or upstream stage (or stages, in case of an assembly operation) and produces parts that are stored in the output buer of the stage. The allocation of functions, resources, and products to stages is a major issue in the design of multi-stage manufac- turing systems and addresses the questions of what to make and how to make it? Once the manufacturing process and methods have been established and the stages of production have been de®ned, another very important design decision is the determination of the mechanism to control the ¯ow of material through the manufacturing system. Material ¯ow control is an optimization problem that addresses the question of when and how much to make in order to achieve a satisfactory customer service level, while keep- ing low in-process inventories. Diculties in control arise because of queueing delays due to variability in produc- tion capacity and in demand. One approach to tackling the material ¯ow control problem is to formulate it as a stochastic optimal con- trol problem and then try to determine an optimal control policy for this problem [1]. Thus far, this approach has been successful only for very simple systems. More- over, an optimal policy, assuming one can be found even for realistic systems, will very likely be too complicated to be of any practical value. Knowledge of an optimal policy or its properties, however, may point to the design and help to assess the performance of simpler sub-optimal policies. A more practical approach is to restrict the search for a material ¯ow control policy to a class of simple sub- optimal policies that are easy to implement, and try to determine the optimal policy within the class. Much of the research eort in this area has focused on simple control systems that depend on a small number of parameters per stage and have often emerged from actual industrial practice [2±5]. In this paper we concentrate on material ¯ow control systems in which production is triggered by actual de- mands. Such control systems are often referred to as pull control systems. A simple pull control system used in inventory control is the Base Stock Control System (BSCS) [2,3]. In the BSCS, every stage has a target in- ventory of ®nished parts, called base stock. When a de- mand for an end item arrives to the system, it is immediately transmitted to every stage where it autho- rizes the release of a new part. The advantage of this mechanism is that the system responds instantly to 0740-817X Ó 2000 ``IIE'' IIE Transactions (2000) 32, 369±386