INT. J.

CO M PU TER INTE GRATE D M AN U FA CT U RING ,

1996,

V O L.

9,

NO .

3, 169 189

O verview of Kanban system s

CHUN-CHE HUANG and ANDREW KUSIAK

Abstract. The just-in -tim e ( JIT) approach to m anufactu ring control with `Kan ban s has received m uch atte ntion in the last decade. The Kan ban system s ef ciently control repetitive m an ufactu ring environm ents an d offer sim plicity. However, th ey are not suitable for non-repetitive m anufactu ring system s. The research approaches applied to Kanban system s include sim ulation, m ath em atical, and stochastic approaches, with the em phasis on analysing characteristics of Kanban syste m s, determ ining th e optim al num ber of Kanbans, an d com paring Kan ban system s. In th is paper, an overview of different Kan ban syste m s, m ethodologies, and alternatives to the pure Kanban syste m s are discussed. Th e basic Toyota Kan ban syste m is introduced. Th e recently developed m ethodologies for Kanban system s are surveyed. The Kan ban approach is illustrated with the case stu dies. This paper offe rs a broad discussion of Kan ban syste m s and classi es the previous stud ies. Several conclusions are drawn and suggestions for fu rther research are given.

1. Intro duction The just-in-tim e ( JIT) approach to con trol m anufacturing system s with `Kanbans has received m uch attention in th e last decad e (Chase and Aquilano 1985, Huang et al. 1983, H all 1983, Suzaki 1987, Krajewski et al. 1987, Shingo 1988). The idea of Kan ban originated from US supermarkets (Ohno 1988, pp. 25 27), where custom ers get: (1) what is needed, (2) at the tim e it is needed, and (3) the am ount needed. A superm arket manager m aintains a certain amount of inventory on the shelves. The idea of tangible and touchable food item s in a supermarket was ap plied by Taiichi Oh no in Toyota arou nd 1953 to: (1) reduce inventory and production cycle tim e; (2) increase the speed of inform ation exchange; and (3) im prove productivity.

In the Kanban system, tangible objects, cards that contain information such as the job type, th e quan tity of parts to carry, and the Kanban type, have becom e crucial in production m anagem ent. W ith the m ovem ent of the cards, inform ation becomes tangible and easily un derstood . The concept of `push system s has been used in industry for a long time. In a `push system , jobs are released to the rst stage of m anufacturing, and in tu rn this stage pushes the work in process to the succeeding stage and so on, until the nal products are obtained. The Kanban system is known as a `pull system in the sense that the production of the current stage depends on the dem an d of the subsequ ent stages, i.e. the preceding stage must produce only the exact quan tity withdraw n by the subsequen t m anufactu ring stage. In this way, the Kanban system was created to indicate what is needed at each production stage, and to allow various stages to ef ciently com m unicate with each other. The companys production plan is given only to the nal assemb ly line. When parts or m aterials are withdrawn from the preceding stage, a chain of comm unication is established with each of the relevant preceding stages, and every stage autom atically knows how much and when to produce the parts required. At each station, the inform ation about the product name, code, vo lum e, and so on, can be easily obtained from the Kanbans. Figure 1 illustrates the general Kanban system.

1.1. Principles of im plemen tation of Kanban system s The m ain principles for the implementation of Kanban system s are as follows (Hall 1983, p. 123; Ohno 1988, pp. 29 44; Singh and Falkenbu rg 1994): (1) Level production (balance the schedule) in ord er to ach ieve low variability of th e num be r of parts from one tim e period to th e next.

Authors : Chun-Ch e H uang and Andrew Kusiak, Intelligent System Laboratory, D epartment of Industrial Engineer ing, The U niversity of Iowa, Iowa City, IA 52242-1527 , USA. 0951-192X /96 $12 . 00

1996 Taylor & Francis Ltd

170

C .-C . H u an g and A . Ku sia k

am ount of inventory. Hence, controlling the num ber of Kanbans is equivalent to controlling the amount of inventory; i.e. increasing (decreasing) the num ber of Kanbans corresponds to increasing (decreasing) the am ount of inventory. C ontrolling th e num ber of Kanbans is m uch simpler than controlling the am ount of inventory itself.

1.3. Kanbans According to their functions, Kanbans are classi ed into: (1) Primary Kanban: travels from one stage to another am ong main m anufacturing cells or production preparation areas. The prim ary Kanbans are two kinds, one of which is called `wit hdraw al Kanban (conveyor Kanban) that is carried when going from one stage to the preceding stage. The other one is called `production Kanban (Figure 2) and is used to order production of the portion withdraw n by the succeeding stage. These two kinds of Kanbans are always attached to the containers holding parts. (2) Supply Kanban: travels from a warehouse or storag e facility to a m anufactu ring facility (see Figure 3). (3) Procurem ent Kanban: travels from outside of a company to th e receiving area (see Figure 4). (4) Subcontract Kanban: travels between subcontracting units. (5) Auxiliary Kanban: m ay take the form of an express Kanban , emergency Kanban, or a Kanban for a speci c application (Singh and Falkenburg 1994).

Figure 1. Th e general Kanban syste m . O P1: W hen dem and from stage i + 1 occurs, withd raw kanban s and place th em on th e dispatc hing board. OP2: Production activity initiates when kan ban is placed on the dispatch ing board. OP3: Simultaneously, dem and is sent to stage i - 1 if the dem and occurs at state i. O P4: Com pleted parts with kanbans are sent to stag e i + 1.

(2) Avo id com plex inform ation and hierarchical control systems on a factor y oor. (3) Do not withdraw parts without a Kanban. (4) Withdraw only the parts needed at each stage. (5) Do not send defective parts to the succeeding stages. (6) Produce the exact quantity of parts withdraw n.

1.2. Fun ction s of K anbans The key objective of a Kanban system is to deliver th e material just-in-tim e to the m anufacturing workstations, and to pass inform ation to the preceding stage regarding what and how m uch to prod uce. A Kanban ful lls the following functions: (1) Visibility function The inform ation and m ate rial ow are com bined together as Kanbans move with their parts (work-inprogress, WIP). (2) Production fu nction The Kan ban detached from the succeeding stage ful lls a production control function wh ich indicates th e tim e, quantity, and the part types to be produced. (3) Inventor y fun ction The num ber of Kanban s actually measures the

Figure 2. The production Kan ban .

O verview of Kanban system s

171

Figure 3. The supply Kanban.

1.4. A uxiliar y equ ipm ent (1) Kanban box: to collect Kanban s after they are withdrawn. (2) Dispatching board: in which Kanbans from the succeeding stage are placed in ord er to display the production schedule. (3) Kanban managem ent account: an account to manage Kanbans. (4) Supply managem ent account: an accou nt to manage the supply of raw m aterials.

1.5. Gen eral descrip tion of K an ban operations For production stage i, when parts are processed and dem and from its receiving stage i + 1 occurs, the

production Kanban is rem oved from a con tainer and is placed on the dispatching boa rd at stage i . The withdrawal Kanban from stage i + 1 then replaces the production Kanban and the container. This container along with the withdraw al Kanban is then sent to stage i + 1 for processing. M eanwhile at stage i, the production activity takes place when a production Kanban and a container with the withdraw al Kanban are available. The withdraw al Kanban is then replaced by the production Kanban and sent back to stage i - 1 to initiate production activity at stage i - 1. This form s a cyclic production chain. The Kanban pulls (withdraws) parts instead of pushing parts from one stag e to another to m eet the dem and at each stage. The Kanban controls the m ove of product, and the number of Kanbans limits th e ow of products (Shing o 1987). If no withdraw al is requested by the succeeding stage, the preceding stage will not produce at all, and hence no excess item s are m anufactured. Therefore, by the num ber of Kanbans (containers) circulating in a JIT system , nonstock-production (NSP) m ay be achieved.

1.6. Kanban control Toyota considered its system of external and internal processes as connected with invisible conveyor lines (Kanbans). The information ow (Kanban ow) acts like an invisible conveyo r throu gh the entire production system and connects all the departm ent
Figure 4. Th e procurem ent Kanban .

172

C .-C . H u an g and A . Ku sia k

Figure 5. A general Kan ban control syste m .

together. Figure 5 presents a general Kanban control system (Lu 1982, p. 540). 1.6.1. T he production line. Due to different types of m aterial handling system s, there are three types of con trol (L u 1982, pp. 541 546): (1) Single Kanban system (using production Kanban s) The single Kanban (single-card) system uses production Kanban s only to block m aterial-handling based on th e part type. The production is blocked at each stage based on the total queue size (Figure 6). In a single-card system , the size of a station output buffer and part m ix m ay vary. M ultiple containers contain the batches to be produced, as long as th e total num ber of fu ll containers in the ou tput bu ffe r does not exceed the buffer output capacity. N ote that the single-card system is in no way related to the hybrid push-pull-schedule-driven singlecard system described by Schonbe rger (1982a). The following con ditions are essential for a proper functioning of the single Kanban system : (a) sm all distance between any two subsequ ent stages; (b) fast turnover of Kanbans;

(c) low WIP; (d) sm all buffer space and fast tu rnover of W IP; and (e) synchronization between the production rate and speed of m aterial handling. (2) Du al Kanban system (u sin g tw o Kanban s sim u ltaneously) The dual Kan ban system (two -card system ) uses production and withdraw al Kanbans to im plement both the station and material-handling blocking by part type. There is a buffer for W IP while transporting the nished parts from a preceding stage to its succeeding stage. The withdrawal Kan bans are presented in the buffer area (see Figure 7). The m ost com m on form of two-card Kanban production control is described in Sugimori et al . (1977), M onden (1983a, pp. 17 20), and Schonbe rger (1982a, pp. 221 224). This system is appropriate for manufactures who are not prepared to adopt strict control rules to the buffer inventory. The following conditions are essential for the dual Kanban system: (a) (b) (c) (d) m oderate distance between two stages; fast turnover of Kanbans; som e WIP in a buffer is needed; external buffer to the production system; and

O verview of Kanban system s

173

Figure 6. The single Kan ban syste m .

(e) synchronization between the production rate and speed of m aterial han dling (3) Sem i-dual Kanban system (changing production Kanban s and w ithdrawal Kanbans at in termediate stages) Figure 8 presents the semi-dual Kanban system . The sem i-dual Kan ban system has the following ch aracteristics: (a) (b) (c) (d) (e) large distance between two stages; slow turnover of Kanbans; large W IP is needed between subsequent stages; slow turnover of W IP; synchronization between the production rate and speed of m aterial han dling is not necessary.

(2) receiving from a storage (see Figure 9); (3) receiving from a vendor (see Figure 10). 1.6.3. T he optimal num ber of Kanban s. The number of Kanbans is determined based on the am ount of inventory. It is im portant to have an accu rate num ber of Kanbans so th at the WIP is m inim ized and simultaneously the ou t-of-sto ck situation is avo ided. In the Toyota Kanban system: num ber of Kanbans = (m axim um daily production quantity) * (production waiting tim e + prod uction processing tim e + withdraw lead tim e + safety factor) standard num ber of parts (SPN)
R em arks (a) The m axim um daily production quantity is the m aximum outp ut based on the daily production plan. Note that the production quan tity should not vary too m uch on a daily basis, which is one of the necessary conditions to implem ent the Kanban production concept.

Table 1 compares the three types of Kanban systems. 1.6.2. T he receiving area. Based on different types of receiving, three types of Kanban operations are perform ed: (1) receiving from a preceding stage in the sam e facility (see Figure 6);

174

C .-C . H u an g and A . Ku sia k

Figure 7. The dual Kanban syste m .

Figure 8. Th e sem i-dual Kanban syste m .

O verview of Kanban system s

Table 1. C om parison of the three typ es of Kanban syste m s. SKS D istance betw een tw o stag es W IP betw een tw o stag es Turnover of Kanban s Turnover of W IP Syn chronization of production an d m ovem ent of W IP SKS: Single Kan ban syste m . D KS: Dual Kan ban syste m . SDKS: Sem i-dual Kanban system . Sm all Sm all Fast Fast N ecessary D KS M oderate Sm all Fast M oderate N ot necessary SD KS Large Large Low Slow

175

Necessary

(b) Production waiting time is the idle interval between two production comm ands (for example 0 . 5 day in Figure 11). (c) Production processing time is the interval between receiving production com m and and completing th e lot. (d) W ithdrawal lead tim e is the interval between withdraw ing a Kanban from the preceding stage and issuing a production com mand. (e) The safety factor is based on tim e unit, e.g. day. It allows avoidance of an interruption of the productio n line due to unexpected conditions.

(f ) SNP represen ts the standard number of parts. A Kanban indicates the standard number of the parts. The number of Kanbans between adjacent stations im pacts the inventory level between these two stations. Several m ethods have been developed for determining the optimal number of Kanbans (see Section 2.3). In Figure II, the cycle time of Kanbans (part 0 .5 + 0 .5 + 0 .2 + 0 .1 + 0 .1 = 1 .5 {A , B , C }) = 0 .1 + (days). The num ber of Kanbans of part {A , B , C } = 1000 * 1 .5 / 100 = 15 (Kanbans), where Q m ax = 1000 and SNP = 100.

Figure 9. The Kanban syste m receiving parts from warehouses.

176

C .-C . H u an g and A . Ku sia k

Figure 10. The Kan ban syste m receiving parts from external vendors.

Figure 11. The cycle tim e of Kanban s.

O verview of Kanban system s

177

1.6.4. A djustm ent of the K anban system (1) Insertion m ain tenan ce action Insertion m aintenan ce takes place when the number of Kanbans used in a current plann ing period is larger th an the num ber of Kanbans used in the previous period. Additional Kanbans are introduced to the system im m ediately after withdraw ing the production Kan bans and placing them on the dispatching board . (2) R em oval m ain tenance action Removal maintenance, similar to the insertion m aintenance, takes place when the num ber of Kanbans used in th e current plann ing period is sm aller than the number of Kanbans used in the previous planning period. The additional Kanbans are always rem oved imm ediately after withdrawing the production Kanbans and rem oval of an equivalent num ber of Kanbans from th e dispatching boa rd. 2. Software approaches Previou s research approaches of Kanban pull system s included simulation, mathematical, and stoch astic m odelling (U zsoy and Martin-Vega 1990). The research published has m ainly concentrated on m odelling Kanban systems in a repetitive ( job shop) environm ent, determ ining the num ber of Kanbans in order to optim ize the system perform ance, and com paring Kanban system s. 2.1. M odelling K an ban system s 2.1.1. Sim ulation m odels. The simulation studies of JITKan ban system s can be broadly classi ed as: (1) explorative analysis of pull system s ( JIT with Kanban); (2) com parative analysis of push and pull system s. Reviews of simulation m odelling are presented in Yavuz and Satir (1995a,b) and C hu and Shih (1992). Sim ulation studies were also carried out by H uang et al . (1983), Philipoom et al . (1987), Rees et al . (1987), Krajewski et al . (1987), Sarker and Harris (1988) and Sarker (1989), where the various aspects of the JIT doctrine and its implementation in pull systems were presented. C hu and Shih (1992) studied th e use of simulation in JIT prod uction. They show ed th at numerous simulation-related statistical issues had been neglected in the previous studies. This m ay have caused som e simulation results to be dif cult to explain. H uang et al . (1983) simulated a JIT Kanban system using a Q -G ERT model. The results indicated either the infeasibility of applying th e Kanban approach to U S m anufactu ring or necessary environm ental changes if con sidered for im plementation. Krajewski et al . (1987) developed a large simulation

model capable of representing diverse m anufacturing environments. They showed that the perform ance of a Kanban system is sensitive to the m anufacturing environment. The bene ts of im plementing Kanban systems result from the environm ent, not the systems themselves. Sarker (1989) developed a SLAM model of a Kanban pull system un der different conditions, wh ich included variable processing times and mach ine breakdowns. Under these conditions, the sam ple ef ciency of push and pull systems was compared. Gupta and Gupta (1989a,b) justi ed some of the unique characteristics of JIT-Kanban system s by applying a dynam ic simulation model. Using systemdynamic concepts, the behaviour of th e system under various exogenous factors was demonstrated. Jo thishankar and Wang (1993) applied a m etam odelling approach to analyse the perform ance of a JIT m anufactu ring system . They concluded that only the assem bly time, Kanban capacity, and the interaction effect between demand and Kanban capacity are statistically signi cant among 15 variables studied. 2.1.2. M athem atical programm ing m odels. Deterministic m athem atical programm ing m odels are used to op tim ize som e objective functions of the Kanban system (Bitran and Chang 1987, Li and Co 1991, Bard and Golany 1991). This approa ch is suitable for a JITKanban system since the repetitive environm ent is deterministic. However, it m ight not be appropriate in a dynam ic environm ent. Yavuz and Satir (1995a) studied the Kanban-based op erations of a m ixed m anufacturing line. Features of Kanban-based simulation studies were surveyed for the single-card or two-card system s and the type of experimentation purpose (explorative or comparative). Yavuz and Satir (1995b) also reviewed selected published research on Kanban-based operational planning and control of ow lines and focused on simulation m odels. Kim ura and Terada (1981) provided several basic equation s for th e Kanban system in a m ulti-stage serial production environm ent. Bitran and C han g (1985) presented an optim ization m odel for th e Kanban system in a multi-stage assem bly production setting and developed a solution procedure. Philipoom et al. (1987) applied a m athematical programm ing approach to determine th e num ber of Kanban s for a singleKanban (recorder point) system . Gravel and Price (1988) pointed out that applying Kanban system s in a job-s hop environm ent might be feasible under the assum ptions that the processing time was constan t and setu p tim e was negligible. However,

178

C .-C . H u an g and A . Ku sia k

d

th ese are essential for the adaptability of Kanban system to the job -shop environm ent. M itwasi and Askin (1994) m odelled the multi-item, single-stage Kanban system as a non-linear mathem atical programm ing model with dynam ic demand. They concluded that the inventory function in the Kanban system is to stablize demand rather than balance the setup cost. It is unlike the traditional inventory models, such as the economic lot size scheduling problem (Elmaghraby 1978, Zipkin 1991), where setup considerations lead to carefully planned cyclical production schedules, additional Kanbans may need to be added to account for the mean num ber of containers awaiting replenishm ent. 2.1.3. Stochastic models. In the stochastic approach, the pull dem and and the processing time are m odelled as random variab les. Markov chains are often used to describe the system behaviour. The Poisson process arrivals and exponential processing times are the general assum ptions (M itra and Mitrani 1990, Deleersnyder et al . 1989, Buzacott 1989). Deleesynder et al . (1989) developed a discrete tim e M arkov process model for an N -stage serial system. U nder the assum ption s of a nite pool of raw m aterial at the upstream station, and a similar pool of dem and for nished products following the downstream , M itra and M itran i (1990) constructed a stochastic m odel of a cellular m anufacturing facility. Berkley (1987) showed that a base stock system with a single inventory buffer between each pair of stations is equivalent to a tandem queue. Other papers that studied base stock Kanban system s by the tandem queue theory were by Davis and Stubitz (1987), Deleersnyder et al . (1989), Ebrahim pour and Fathi (1985), and Sarker and Fitzsim m ons (1989). Glasserman and Yao (1994, p. 107) presented a tandem m odel (a, b, k) for th e Kanban system , Siha (1994) developed a continuous time Markov m odel for pull production system s. The various allocation patterns of Kan ban capacity and mean production tim e over the system stations were studied. It was found in som e cases that a certain `fu nnel pattern can improve the system perform ance. Since this nding con tradicts the concepts presented in papers that favoured a `b owl pattern, som e possible interpretations were presented by Siha (1994). Overall, the results produced design guidelines that should be usefu l in industrial ap plications. 2.2. D esign m ethodologies Different m ethodo logies for the design of Kanban system s have been studied in the literature, for exam ple:

Design with the server network generator (SNG) (Bouchentouf-Idriss and Zeidner 1991). Design with Petri nets (Di M ascolo et al . 1991).

2.3. O ptim izing the nu mber of Kan bans M ost studies have concentrated on operational control problems and perform ance analysis of JIT manufacturing systems, emphasizing the determ ination of the num ber of Kanbans (Deleersnyd er et al. 1989, Berkley 1987, Bitran and Chang 1987, Huang et al. 1983, Jordan 1988, Kim 1985, Rees et al. 1987, Sarker and Harris 1988, Sarker and Fitzsim mons 1989, Villeda et al . 1988, So and Pinault 1988, Kimura and Terada 1981, Krajewski et al. 1987, Price et al. 1995). Philipoom et al . (1987) perform ed a simulation study of a Kanban system and reported that factors , e.g. feed rate, m achining utilization, coef cient of variation of processing tim es, and auto-correlation betwe en processing tim es, affect the num ber of Kanbans. M onden (1983a,b ) indicated that certain rules must be followed in order to achieve JIT. On e of them is that the num ber of Kanbans between two adjacent stations represents the m aximum inventory level and, therefore, should be kept at minim um . The relationship and the trade-off between inventory and the num ber of Kanbans has been shown elsewhere, e.g. M onden (1983a,b), Japanese m anagem ent Association (1986), and Shingo (1987, 1988). Speci cally, it has been pointed out that the fewer th e Kanbans, the better the system s. W ith fewer Kanbans, sensitivity of system s is greater ( Japanese M anagem ent Association 1986). Davis and Stubitz (1987) determ ined the num ber of Kanbans at each station for optim al perform ance using simulation and response surface m ethodology. Philipoom et al . (1987) investigated the factors that in uence the num ber of Kanbans required for a single workcentre by exam ining the form ula for the number of Kanbans used at Toyota (M onden 1983a, pp. 167 179). This m ethodology was further extended in Rees et al. (1987) to dynam ically adjust the number of Kanbans by using a forecasted deman d and estimated lead tim e. Wang and Wang (1990) discussed the role of Kanbans in a JIT production system in the context of m aintaining a m inim um level of WIP inven tory. A m odel for determ ining the optim al num ber of Kanbans for three production settings (one station to one station, OSTOS; m ultiple stations to one station, M STOS; m ultiple stations to multiple stations, M STMS) was presented. OSTOS is applicable to JIT m achining shops. M STOS and M STM S are suitable to JIT assembly shops.

O verview of Kanban system s

179

O hno et al . (1995) devised an algorithm of determ ining optim al values of the safety stocks included in th e pop ular form ulas for com puting the number of Kanbans. Jothishankar and Wang (1992) applied stochastic Petri nets to optim ize the num ber of Kanbans. 2.4. C on trol approaches M ost studies focused on the shop control methodology, allocation of a xed num ber of Kanbans and buffers, and batch size con trol. C haudhury and Whinston (1990) presented an ef cient, decentralized and adaptive control m ethodology for ow shops. The methodology is based on stochastic autom ata methods for modelling learning behaviour. It was suggested that such a m ethodology can be used with a Kanban-type control technique to make ow shop systems more exible and adaptive. The relationship between the control model and computational m odels such as neural com puting was discussed. C heng (1993) proved th at with a general arrival process and exponential service tim es, the job com pletion, job departu re, and Kanban generation processes are increasing concave functions of the initial inventory and Kanban counts. Karm arkar and Kekre (1989) studied the effect of batch size policy on a prod uction lead tim e, and hence on the inven tory level and cell perform an ce. Both single- and dual-card Kanban cells and two-stage Kan ban system s were m odelled as M arkovian processes, and the effect of batch sizes on the expected inventory and back order costs were studied. It was fu rther shown th at batching policy has a signi cant im pact on the cost. The effect of varyin g the card count in the cell was also examined. Albino et al . (1992) m odelled a single-product and m ulti-stag e m anufacturing line with reduced resource failures and controlled by a two-c ard Kanban system using discrete-event simulation. Several perform ance m easures were evaluated to determ ine op tim al operation policies given resource failures. In addition, different m aintenance policies were im plemented in order to better understand their im pact on the overall system perform an ce. Yanagawa et al. (1994a,b) dealt with optim al operation planning of the constan t-num ber-of-withd raw alKan ban system with variable lead-times and different con sum ption rates of parts for each production unit. The behaviour of the optim al operation planning which m inim izes the average total operation cost was shown by m eans of simulatio n analysis of various values of param eters: the ord er cost, the range of consump tion rates of parts, and the range of lead times for delivery.

M itra and M itrani (1990) described and analysed a schem e for coordination of cells in large-scale m anufacturing facility. M any cells in a tandem con guration and a xed num ber of cards in each cell were considered. Som e rules speci ed the lead tim e for each m anufacturing cell. The cards circulated within a cell and their presence at a certain position of th e cell conveys to the neighbouring cells the statu s of the cell inventory. The results were presented in three parts: (1) Their scheme was shown to dom inate the classical scheme in term s of the throughput rate. (2) A schem e for approxim ate analysis of the perform an ce of the Kanban system by exam ining rst a signal cell in isolation and then combining the isolated cells throu gh xed-point equations was presented. (3) The throu ghput-rate inventory relationship of the Kanban schem e was observed to be superior to that of the classical scheme. M itra and M itrani (1991) developed another m odel for a stochastic dem and. Pervozvan skiy and Sheynis (1994) constructed a m odel of the system for rand om m anufacturing and dem and processes, which allows determ ination of optim um values of system param eters. A throu gh Kanban m anufacturing control system was proposed that is m ore effective (according to the results of simulation) than th e com m only used Kanban system . Tayur (1993) studied the structural properties and a heuristic for Kanban -controlled serial lines, and determ ined: (1) The optimal solutio ns in the allocation and partitioning problems given a total xed num ber of Kanbans. (2) The reduced computatio nal effort usually required to study these system s. (3) The developm en t of a combinatorial m easure as a surrogate for the m ean throughpu t based on structural results; e.g., in a ve-cell line to be allocated (1, C 1 , C 2 , C 3 , 1) is better than other allocation s. (4) The demon stration of which structure with optim al allocations is insensitive to the variability in a system with balan ced lines.

2.5. Schedu ling approaches M ost studies have been concerned with levelling

180

C .-C . H u an g and A . Ku sia k

th e schedules for a m ixed m odel (Miltenburg 1989, M iltenbu rg and Sinnam on 1989, Kubiak and Sethi 1989). Garey et al . (1988), and Inm an and Bul n (1991) studied the problem of m inim izing the total earliness and tardiness of schedules. M itwasi and Askin (1994) investigated the use of Kan ban control at workcentres which produce multiple item s with dynam ic, rand om demand. The dynamic aspects of dem and m ay cause tem porary capacity shortages. The Kanban control system must quickly react to the random changes of the dem and. By selecting different num bers of Kanbans, the dynamic aspects can be accom m odated. A m athem atical m odel was developed and shown to reduce th e original prob lem to a simpler one. Berkley (1993) com pared the perform ance of the rst com e- rst serve (FCFS) and shortest-processingtim e (SPT) sequencing rules for a Kanban system with a varying buffer capacity. It was known that the op tim al job sequencing varied with the buffer capacity (Park 1987). FCFS and SPT rules were tested to justify the use of the FCFS rule in Kanban system s in industry while th e SPT rule is usually shown to m aximize production and minim ize inventory and lead time based objectives of just-in-tim e m anufactu ring. A second objective was to introduce the con cept of blocking by part type found in Kan ban system s processing several different parts types. As Uzsoy and Martin -Vega (1990) observed, the ap plication of Kan ban systems to a com plex m anufacturing environmen t has created a need to learn how these system s operate when there are several part types processed on the sam e line. 2.5.1. C om parative studies (Berkley 1993) (1) The relative perform ance of FC FS and SPT* rules changes with the processing-time variab ility and input buffer capacity. SPT* is the shortest processing time rule wh en m aterial-handling blocking is by part typ e. (2) For both instant and periodic material handling, increasing the input bu ffer capacity alone has little effect on the average SPT* prod uction rate and inventory. This is becau se the bene ts of a larger total buffer space are com pletely offset by a greater frequency of m aterial-handling blocking by part type. On the other hand, increasing the output buffer capacity alone increases the average SPT* production rate and the amount of inven tory. (3) Use of the SPT* sequencing rule can cause jobpassing and m aterial-handling operations to be blocked before input buffers are entirely fu ll. (4) Exponential processing tim es and a small input buffe r capacity reduce th e frequency of material-

handing blocking by part type and the SPT* rule increases the average production rate and decreases the average level of inven tory. (5) For norm al processing tim es and a large input buffer capacity, the FCFS rule increases the average production rate and inventory vo lum e m ore than the SPT* rule. (6) The use of the FCFS rule is recom m ended when the processing tim es of a batch of Kanbans are relatively constant. However, if the processing tim es of a batch of Kanbans are highly variable, perhaps becau se different part typ es require different processing tim es at each station, one should con sider the use of SPT* for a system with sm all input buffer capacity. (7) As the input buffer capacity is reduced in response to im proved production processes, SPT* perform ance is likely to improve relative to FCFS.

2.6. C om paring Kanban system s w ith other system s N um erous studies have com pared Kanban system s with the M RP system (Petroff 1993 pp. 5 13, Hernandez 1989 pp. 14 20, Du rm usoglu 1991, Rees et al . 1989, Schonbe rger 1982a pp. 130 143 and 1982b pp. 1 14, and 1983, Gru wald et al . 1987, Sarker and Fitzsimm ons 1989). The stock (Q , r) policy and the tand em queuing m odel in the generalized semi-M arkov processes were also com pared with Kanban systems (Axsa ter and Rosling 1993, Berkley 1987, Glasserm an and Yao 1994 pp. 107 108). Axsa ter and Rosling (1993) identi ed a Kanban i policy as a restricted type of installation stock ( Q n , r n )policy where: (1) The reorder poin t r n of item n is an integer m ultiple of its lot size Q n . (2) Backlogs are not subtracted in the de nition of the installation inventory position. Glasserman and Yao (1994 pp. 107 108) concluded that a generalized Kanban approach can be viewed as the approach of the general tandem (a, b, k) m odel im plem ented throu gh cards. At each stage i , in ad dition to the k i Kan bans, there are two other types of cards: a i conveyance cards and b i production card s. The conveyance card s and the production cards are attached to raw jobs and nished jobs, respectively. The conveyance cards authorize adm ission to a stage; for a job to enter stage i, there m ust be one conve yance card, in ad dition to one Kanban available at that stage. The production
i

O verview of Kanban system s

181

cards, on the other hand, authorize service; for service i to begin processing a job, the job m ust rst be issued a free production card. When the raw job com pletes service, the production card is attached to it and con veyance card is detached and m ade available to ad mit another job. When a job leaves a stage, both its Kan ban and production card are detached. Throu gh th is im plementation, at each stage i the total number of raw job s is limited to a i , the total number of nished jobs is limited to b i , and th eir sum is limited to k j . Rees et al . (1989) compared an MRP lot-for-lot system and a Kanban system in an ill-structured production environm en t. It was determ ined th at the M RP lotfor-lot system is m ore cost-effective th an the Kanban system as th e MRP system carries less inven tory and requires fewer setups. Krajewski et al . (1987) concluded th at working with factors, e.g. lot sizes, setup tim es, yield losses, workforce exibility, degree of prod uct custom ization, and product structure, to shape a m anufacturing environment with m ore unifo rm work ows and exibility is the key to im proving perform ance. The Kanban system , by itself, is not crucial to improving perform ance, which is unlike the views of others on Kan ban systems (Hall 1981, M onden 1981a,b, Schon berger 1982a, pp. 130 143 and b, pp. 1 14, and 1983, Sugimori et al ., 1977). Shipper and Shapira (1989) developed a decision rule to enable a priori selection of a production system th at should utilize a JIT or W IP type inventory control policy. Sarker and Fitzsimm ons (1989) comparative analysis of an M RP lot-for-lot system and a Kanban system for a m ultistage production operation con cluded that: (1) MRP product quality would improve as lot sizes were decreased. (2) MRP appears to handle lum py demand better than a Kanban system , even though stochastic processing times m ay cause dif culties. Gupta et al . (1991) provided insights into the characteristics of com panies that had implemented JIT production as compared to those th at had not. The au thors exam ined the changes that the m anagem ent cou ld expect to encou nter as JIT was incorporated into a m anufacturing rm. The response of the survey on JIT implementation from 175 m anufactu ring organizations indicated that: (1) com panies that have im plemented JIT had fewer costum ers who purchase in higher vo lum es (over a period of tim e) th an non-JIT companies; (2) the service in response to custom er requ ire-

m ents in JIT companies has im proved faster than in non-JIT companies; (3) product and process engineering skills are higher, and nancial liquidity is better in JIT organizations. Several differen ces that were expected to exist betwe en the two types of com panies were not supported by the data. It was believed th at JIT com panies should have signi can tly fewer suppliers than non-JIT com panies as JIT requires m anufacturers to develop ways of establishing long -term strategic partnerships with suppliers. The data, however, did not suppo rt this relationship. Also, contrary to expectations, JIT com panies did not exhibit less vertical integration and did not em phasize effective coordination among functions to a greater extent than non-JIT com panies. Finally, the results showed that JIT com panies did not have the ability to change products in response to the changes in dem and faster than non-JIT com panies. JIT com panies did not have more exibility in rerouteing jobs in case of m achine breakdowns.

2.7. C ase study A num ber of authors have examined the advantages of Kanban system s, e.g. Kim ura and Terada (1981), M onden (1981a), Schon berger (1983), and Gupta and Gupta (1989a). The following case studies justify som e advantages. (1) Singh et al . (1990) developed a Kanban system and simulated it with GPSS. (2) Ichihashi (1990) m odelled a C IM-Kanban system to handle production control, quality control, m anufacturing line control, and of ce and technical system s based on a num ber of m anufacturing m odules (cells) in N ippondenso Kota plant. (3) Sohal and N aylor (1992) described the Kanban system used in a number of different areas, for exam ple a sm all manufactu ring rm . Eve n in a short period of tim e and with lim ited resources, the company achieved striking results. (4) Sohal et al . (1993) presented a system integrating C NC technology and the JIT Kanban system . The results showe d that the changeover and setup tim es reduced from 5 6 hours to abo ut 90 secon ds; stockouts or shortage s of com pon ents were totally eliminated; the total stock was reduced by over 50%; and the quality of nished products improved signi can tly with the new production system . (5) Several other studies were concerned with Kanban system s applied to different environ-

182

C .-C . H u an g and A . Ku sia k

Table 2. Sum m ary of previous stu dies of Kanban system s. C lassi cation M odels Simulation m odels Feature of m ost focused stu dies References

(1) E xplorative an alysis of JIT with Kan bans (2) C om parative an alysis of pull an d push system s

C hu an d Shih (1 992), H uang et al. (1 983), Philipoom et al. (1 987), Rees et al. (1987), Krajewski et al. (1987), Sarker and H arris (1 988), Sarker (1989), Gupta and Gupta (1 989a,b), Jothishan kar an d Wang (1993), Yavuz an d Satir (1995a,b) Bitran an d Chan g (1987), Li and C o (1 991), Bard an d Golany (1991), Kim ura and Terada (1981), Philipoom et al. (1987), Gravel and Price (1 988), M itwasi an d Askin (1 994), Yavuz and Satir (1995a,b) M itra an d M itran i (1990), Deleersnyder et al. (1990), Berkley (1987), Buzacott (1989), Davis an d Stubitz (1 987), E brahim pour an d Fathi (1985), Sarker an d Fitzsim m ons (1989), G lasserm an and Yao (1 994), Siha (1994) Bouchentouf-Id riss and Zeidner (1 991), D i M ascolo et al. (1991) D eleersnyder et al. (1 989), Berkley (1 987), Bitran an d Chan g (1 987), H uang et al. (1 983), Jordan (1988), Kim (1985), Rees et al. (1987), Sarker and H arris (1 988), Sarker an d Fitzsim m ons (1989), Villeda et al. (1988), So an d Pinault (1 988), Kim ura an d Terada (1981), Krajewski et al. (1 987), M onden (1983a,b), Japanese M an agem ent Association (1986), Shingo (1987, 1988), Davis an d Stubitz (1987), Wang and Wang (1 990), Joth ishankar an d Wang (1 992) C haudhury and W hinsto n (1 990), Cheng (1993), Karm arkar and Kekre (1 989), Albino et al. (1992), Yanagawa et al. (1994a,b), Mitra and M itran i (1 990, 1991), Pervozvan sky an d Sheynis (1994), Tayur (1993) M iltenburg (1989), M iltenburg an d Sinnam on (1989), Kubiak an d Sethi (1991), Garey et al. (1988), In m an and Bul n (1 991), M itwasi an d Askin (1994), Berkley (1993), Uzsoy and M artin Vega (1990), Price et al. (1995), O hn o et al. (1 995) Petroff (1 993, pp. 5 13); Hernan dex (1 989, pp. 14 20); D u rm usoglu (1 991); Rees et al. (1989); Schonberger (1982a, pp. 130 143 an d b, pp. 1 14, an d 1983), Gru wald (1987), Axsa ter and Rosling (1993), Berkley (1 987); Glasserm an and Yao (1 994, pp. 107 108); Shipper and Shapira (1989), Sarker an d Fitzsim m ons (1989), Gupta et al. (1 991) Singh et al. (1 990), Ichihashi (1990), Sohal and N aylor (1992), Sohal et al. (1 993), Sohal an d N aylor (1992), Olhager and O stlun d (1990), Harvey and Jones (1989)

M athe m atical program m ing m odels

Optim izing som e objective fu nctions of Kanban syste m s, e.g., throughput W IP, and num ber of Kan ban s, assum ing determ inistic environm ent

Stochastic m odels

Optim izing som e objective fu nctions of a Kanban syste m assu m ing th at dem and an d process tim e are random variables, M arkov chains and tand em queues are often used to describe th e syste m behaviour. Design with th e netw ork generator, e.g. SN G or Petri nets An alysing the relationsh ip betw een th e num ber of Kan bas and syste m behaviour, and determ ining th e num ber of Kanbas so that syste m behaviour is optim ized

D esign m ethodologies O ptim izing th e num ber of Kan bans

C ontrol approaches

Optim izing syste m behaviour by the stu dies of shop control m ethodology, e.g. decentralized and adaptive control m ethd ology, lot size policy, an d the allocation of a xed num ber of Kanbans Levelling schedules of a m ixed m odel

Scheduling approaches

C om parative stud ies

Th e com parison of Kan ban syste m s with M RP, stock policy and tan dem queuing netw ork

C ase stu dies

Industrial applications

O verview of Kanban system s

183

men ts, e.g. in a small m anufacturing rm (Sohal and Naylor 1992), in a sem i-repetitive m ake-to ord er environm ent (Olhager and O stlund 1990), and in an aerospace environm ent (Harvey and Jones 1989).

system that possesses the bene ts of a pull system and could be used in different production environm ents.
M odel description C ONW IP is a generalized Kanban system . Also, it is an integrated system that offers th e bene ts of JIT system s and is applicable to a broader range of production environm ents than the traditional JIT approac h. CON W IP is focu sed on the interactions between the planning m odules at the different levels in the hierarchy and on the architecture linking them . Like a Kanban system , it relies on signals. A card is attached to a standard container of parts at th e beginning of the process. When th e container approaches the end of the process, the card is removed and sent back to the beginning where it waits in a card queue to eventually be attached to another container of parts. CON WIP production cards are assigned to the production line. Part num bers are assigned to the cards at the beginning of the production line. Figure 12 illustrates the operation of the CON WIP system . M ain difference from the Kanban system (1) U ses backlog inform ation to dictate the part num ber sequence. (2) C ards are associated with all parts produced on a line rather than individual part num bers. (3) Jobs are pushed between workstations in series once they have been authorized by a card to enter the line. R esults M any of the bene ts of C ONW IP can be attributed to the fact that it is a pull-based production system (e.g. shorter ow tim es and reduced inventory levels). However, the system does offer som e distinct advantages over the Kanban system. One of them is that it can be used in som e production environm ents where Kanban

2.8. Sum m ar y of previou s studies The recently developed m ethod ologies for Kanban system s are sum m arized in Table 2.

3. M odi ed m odels of Kanban system s The Kanban system approach is dif cult to use in certain situations, namely (see Monden 1983a, p. 64):
d d d d d

job orders with short prod uction runs; signi cant setups; presence of scrap; large, unpredictable uctuations in dem and; the need for com plex inform ation and hierarchical control system in th e shop .

Several m odi ed m odels were developed to overcom e th ese shortcom ings of Kanban system s. 3.1. T he con stan t W IP (CO N W IP) m odel
R eason Kanban is intrinsically a system for repetitive manufacturing (Hall 1981) and it is not appropriate for a shop controlled by job orders. M odel Spearm an et al . (1989) presented a new pull system called CONW IP (CO Nstant Work In Process). The W IP was kept constan t by xing the total num ber of Kanbans in the system . The purpose of the m odel was to present a

Figure 12. The CO NW IP syste m .

184

C .-C . H u an g and A . Ku sia k

is not practical due to too many card numbers or because of high setu ps. By allowing W IP to be collected in front of the bottleneck, CON W IP can function with lower W IP and less production control personnel than in the Kanban system . Spearm an et al . (1990, 1992) con cluded that: (1) CONW IP is m ore general than a Kanban system. (2) CONW IP is m ore effective than a Kanban system. (3) CONW IP not only has better custom er service, e.g. less tardy jobs than a pure Kanban system, but also solves certain implementation problem s, e.g. dif cult to reduce setups or to optimize synchronization of parts production. 3.2. T he gen eric Kanban system
R eason W ith variable dem and and processing tim es, it is dif cult to set the m aster schedule. Also, line balancing and synchronization in the receptive system are impossible to attain. A Kanban operation is generally not ap plicable to a dynam ic environm ent with variable dem ands and variab le processing times (Hall 1981, Huan g et al . 1983, Finch and Cox 1986, and Krajewski et al . 1987). M odels (1) A dynam ic environm ent m ay be changed (or simpli ed) toward the repetitive system and adopt the Kan ban control discipline. However, this requ ires signi cant changes in the system (Huang et al . 1983, Finch and Cox 1986), which is not practical in m any cases because m any com panies cannot afford to do so. (2) Chang and Yih (1994a) proposed a generic Kan ban system a m od i ed Kanban discipline for dynamic environm ents. The generic Kanban system does not have all the bene ts of JIT Kanban system. However, it is adaptable becau se it has advantages over other production system s under th e sam e dynamic con ditions. T he m odel description To operate a generic Kanban system , determ ining th e num ber of Kanban s and lot sizes used in the system is needed. The num ber of Kanbans and job lot size directly affect system perform ance. One of the ap proaches to determine the num ber of Kanbans at each station and lot sizes of job types to optim ize the generic Kanban system perform ance, was prop osed (Chang and Yih 1994b). This ap proach included form ulating the m ulti-objective optim ization problem with a utility function and searching the m axim um utility value with a modi ed simulated annealing (SA) algorithm .

A generic Kanban system includes two phases: acqu isition phase and actual production phase.
T he Kanban acquisition phase : in the generic Kanban system , the deman d is unknown due to the dynam ic environmen t. When a demand arrives in th e system , Kanbans have to be issued for all stages because no parts at any stage are m ade beforehand. On ly when the raw material arrives at the initial station can the actu al production of the system begin. Moreover, not every Kanban at any stage can be issued im mediately since the num ber of Kanbans at each stage is limited. A request m ay be deferred if at a particular stage Kanbans are not available. T he actual production phase : when a job nishes processing at one stage, it is m oved to the next downstream stage and the attached Kanban at th is stage is dropped. This Kanban is acquired by the next request. M ain difference from the Kanban system The situation in the actual production phase is different from a JIT Kanban system wh ere th e free Kanban triggers a new production im m ediately because products are made repetitively in this environm ent. R esults To show th e adaptability and superiority of such a system , other control m eth odologies such as push system s, dedicated Kanban systems, and CONW IP system s were studied and compared. The simulation results are listed below (Chang and Yih 1994b).

(1) Trade-offs exist between cycle tim e and W IP level in generic Kanban systems. The lot size has an im pact on th e system perform an ce. (2) The generic Kanban system behaves similarly to a push system except that a decision-m aker has m ore exibility in relocating resources. (3) The perform ance of the generic Kan ban system is preferred to the dedicated Kanban system becau se it provides simpler production control and dominates the perform ance, e.g. less WIP. It is also preferable to C ONW IP because of higher exibility, e.g. m ore jobs are allowed to enter the system. The SA algorithm is shown to provide similar solutions in shorter tim e in the generic Kanban system than in the trad itional one. 3.3. M odi ed K an ban system in a sem iconductor m anufacturing environm ent
R easons A conceptually pure Kanban system is not suitable

O verview of Kanban system s

185

for semiconductor fabrica tion due to the nature of the m anufactu ring process. The current system s are designed to prevent uneven line loading resulting from various op erational problem s. Even after the fundam ental problem has been solved, the residual impact on production due to disrupted W IP ow cou ld last for weeks. Thus, poor line loading leads directly to increased cycle tim es, poor predictability, and m ore defective products
M odels O tenti (1991) described a m odi ed Kan ban W IP con trol system successfully im plem ented in a semicon ductor (CM OS) fabrication facility. The approach was to set up a series of Kanbans with caps on lots allowed to enter the system . N o additional lots would be allowed to m ove into a Kanban system if the WIP level in the system had reached the m axim um allowable limit. R esults C ycle time dropped from 44 days to 30 days, a 32% improvem ent. Kraft (1992) described a tool which is being currently used at Texas Instrum ents with m odi ed Kan ban JIT scheduling incorporated to im prove the line balancing and W IP ow. R esults C ycle tim e reduced by m ore than 36% .

This has resulted in improved depen dability of delivery and production exibility. A case study in a semirepetitive, m ake-to-order environm ent illustrated som e potential bene ts from such an integrated approach.

3.5. T he periodic pull system U sing the Kanban system, manufacturing factories at Toyota no longer rely on a computer. The reason s for having em ployed the Kanban system instead of a computerized system were as follows:
d d d

Reduction of the cost of processing inform ation. Rap id and precise acquisition of facts. Lim iting surplus capacity at feeding facilities.

R easons (1) In present m anagement systems, the volum e and complexity of information has increased. (2) For some m anufacturing environ m ents, com puterization is necessary. M odel Kim (1985) developed an alternative to the Kanban system , a period pull system (PPS), as an operation policy of practising a pull system . In the PPS, the m anual inform ation processing tim e of a Kanban system is replaced with instant on-line com puterized processing. M odel description In a com puterized m aterial m anagement system, the status of m aterial ow at all stages is reviewed at regular intervals. As the result of the review, only the exact am ount of m aterial that has been consum ed at a succeeding stage (since the last review tim e) is allowed to be withdrawn from or produced at a preceding stage. A review interval is called a period. The tim e for a review is assumed to be non-negative, i.e. com puter processing tim e. The withdraw al and production starts im mediately after the review, that is, at the beginning of the period. A PPS is form ulated m athem atically and a solution approach is provided for target stock levels, as well as the analysis of the uctuation s of in-process m aterial ow, on-hand stock levels, target availab ility, etc. A nalogy On e may visualize that a review time is equivalent to a Kanban pick-up tim e and thus, in a PPS, the imaginary Kanbans picked at a review tim e are delivered to a preceding stage at the same review tim e, i.e. instantly. R esults The m aterial lead tim e is m uch shorter than that of a

3.4. T he integrated push-pull manu factu ring strategy

R easons (1) A pull strategy is not necessarily ap plicable to all manufacturing environm ents. (2) Many m anufacturing rm s using pull system s are interested in attaining the simplicity of push systems. M odel O lhager and O stlun d (1990) com bined a push and pull system into a system th rough three points, the custom er order point (i.e. the point where a production is assigned to a speci c custom er), the bottleneck resources, and the product structure. R esults In th e integrated push-pull system, the m ajor issue is th e linkage of the m anufacturing strategy with th e business strategy. The issue can be solved by changing th e m anufacturing plann ing and control focu s. In the new system , a push principle is applied to the focused m achines (bottleneck m achines) and succeeding production stages, and incoming parts are pulled.

186

C .-C . H u an g and A . Ku sia k

Kan ban system , and th e system perform ance im proves in term s of less inventory and faster system response.

3.6. C ase study Graham (1992) described work carried out in association with Jaguar Cars to develop a steady-state M arkovian model for calculating the num ber of Kan bans required to control single-stage processes feeding assembly lines. A Markovian m odel of an alternative JIT system , in which the off-line process is triggered by the passage of vehicle bodies past a point prior to th e assemb ly area, is also described.
d d

R esults (1) These m odels have shown that the use of a trigger system lead s to lower inven tory levels and a greater pressure for im provem ent than in the Kanban system itself. (2) In a Kanban system the level of subassem bly inven tory required is insensitive to changes in the rate and average duration of body rework, whereas with the triggered system the average level of subassem bly inventory is sensitive to both the rate of body rework and the duration of subassembly rework. (3) The only incentive of a Kanban system is to reduce the rate and duration of body rework positively correlated with the value of inventory. However, in the triggered system a reduction in the expected level of body rework inventory m ay reduce the inventory level of all triggered subassem blies. (4) For example, a 50% reduction in the rate of body reworking or a 50% reduction in rework tim e both lead to a reduction from 69 to 67 in the expected average num ber of engines of this type in inven tory if triggering is being used, whereas with the Kanban system , 80 engines would be needed.

m ore, th e key to improving m anufacturing perform ance is to consider such facto rs as lot sizes, setup tim es, yield losses, workforce exibility, degree of product custom ization, and product structure, to shap e a m anufacturing environm ent with m ore uniform work ows and exibility. The Kanban system , by itself, is not crucial for im proving m anufacturing perform an ce. The m odel of Kanban operations in its simple form is a stock (Q , r)-policy or a tandem queue. However, together with autonom ation ( Jidoka), setup reduction, exibility of workforce, quality control circles (Q CC), th e Kanban system has m any advantages. Decreasing the lot size is an effective way to reduce the m ean length and waiting tim e in WIP points at all Kanban levels that com bine Kanbans and production stations. The optimal allocation structure of a xed num ber of Kanbans is insen sitive to the variab ility in the system with balan ced production lines. The inventory function in the Kanban system is to stabilize the dem and rather than balan ce the setup cost. For a Kanban system to operate effectively, it is crucial that th e delivery times and quality of the upstream suppliers is reliable. In m ost practical approaches, th e product/process design was not m odi ed before im plem enting the JIT-Kanban system concept.

The issues that need further research are categorized as follows:

d

4. C onclusion and further research directions Based on the literature reviewed in this pap er, the con clusions reached are sum m arized as follows:
d d

The Kanban literatu re presents diverse Kanban system s. The concept of Kanban system s is not a pan acea for all industrial problems. It is applicable to a repetitive m anufacturing environm en t. Furtherd

Design of prod ucts and processes for a JIT-Kanban system . Developm ent of a general model th at has the advantages of Kanban system s, can be integrated with manufacturing system s of different types, and applicability of the integrated concept to a nonrepetitive m anufactu ring environmen t. The prob lem of production levelling th rough scheduling is crucial in Kanban systems. Selecting the proper scheduling rules becomes even more important in th e case of high prod uct variety and uncertainty of processing times. Introduction of feeder lines into the pull system con guration adds exibly in adjusting to the lum py dem and and the ow synchronization. Developm ent of op tim al bounding schemes for the sum (Minim ax) objective function in the levelling schedule prob lem. The trade-off cost between m ore frequ ent m aterial-handling and bene ts of reduced WIP

O verview of Kanban system s

187

when the optimal num ber of Kanbans is to be determ ined. M ost previou s studies only considered m inim ization of the throu ghput/ W IP but ignored m inim ization of the total cost when the op tim al num ber of Kanbans was determ ined.

Acknowledgem ent This research has been partially suppported by funds from the National Science Fou ndation. References
A LBIN O , V., C A R EL LA , G ., and O K O G BA A , O. G., 1992, M aintenan ce policies in just-in -tim e m anufacturing lines. International Journal of Production Research, 30, 369 382. A X SA T E R , S., and R O SL IN G , D ., 1993, N otes: installation vs. echelon stock policies for m ultilevel inventory control. M anagem ent Science, 39, 1274 1280. B A RD , J. F., and G O L AN Y, B., 1991, D eterm ining the num ber of kanbans in a m ulti-product, m ultistage production system . International Journal of Production Research, 29, 881 895. B E RK L EY, B. J., 1987, Stochastic kanban -controlled lines and tan dem queues. Working Paper, D epartm ent of Decision Sciences, School of Business Adm inistration, University of South ern California, Los An gels, CA. B E RK L EY, B. J., 1993, Effe ct of buffer capacity and sequencing rules on single-card kanban system perform an ce. International Journal of Production Research, 31, 2875 2894. B ITRA N , G . R., and C H A N G , L., 1985, An optim ization approach to th e kanban syste m . Working paper, No. 1635 85, Sloan School of M an agem ent, M assachusetts Institu te of Technology, Cam bridge, M A. B ITRA N , G . R., an d C H AN G , L., 1987, A m athe m atical program m ing approach to a determ inistic kanban system . M anagem ent Science, 33, 427 441. B O U CH EN T O U F -I D R ISS , A., an d Z E IDN E R , L., 1991, D esign a kanban m an ufac turing syste m using the server network generator (SN G) CASE tool. APL Q uote QUAD, (21), 62 70. B U ZA CO T T, J. A., 1989, Q ueuing m odels of kanban an d MRP controlled production system s. Working Paper, D epartm ent of M anagem ent Sciences, University of W aterloo, W aterloo, O ntario, Can ada.. C H A N G , T. M ., an d Y IH , Y., 1994a, G eneric kan ban syste m s of dynam ic environm ents. International Journal of Production Research, 32, 889 902. C H A N G , T. M ., an d Y IH , Y., 1994b, D eterm ining th e num ber of kanbans and lot sizes in a generic kanban syste m s: a sim ulated ann ealing approach. International Journal of Production Research, 32, 1991 2004. C H A SE , R. B., and A Q U ILAN O , N . J., 1985, Production and O perations Managem ent: A Life Cycle Approach (Richard D . Irwin, H om ewwod, IL). C H A U D H U R Y, A., an d W H IN ST O N , A. B., 1990, Toward an adaptive Kanban system . International Journal of Production Research, 28, 437 458. C H E NG , D . W., 1993, Concavity of syste m throughput and inform ation transfe r in a PAC syste m . Operations Research Letters, 14, 143 146. C H U , C. H ., and S H IH , W. L., 1992, Sim ulation stu dies in JIT

production. International Journal of Production Research, 30, 2573 2586. D AVIS , W. J., and S TU BITZ , S. J., 1987, Con guring a kanban system using a discrete optim ization of m ultiple stochastic responses. International Journal of Production Research, 25, 721 740. D EL EE R SN YD ER , J. L ., H O D G SO N , T. J., M U L LE R (-M A LE K ), H ., an d O G RA D Y, P. J., 1989, Controlled pull system s: an an alytic approach. M anagem ent Science, 35, 1079 1091. D I M A SC O L O , M ., F R EIN , Y., D A LL ER Y, Y., an d D AVID , R., 1991, A uni ed m odelling of kanban system s using Petri nets. International Journal of Flexible M anufacturing Systems, 3, 275 307. D U R M U SO G LU , B. M ., 1991, C om parison of push an d pull syste m s in a cellular m anufactu ring environm ent. Just-in-time M anufacturing Systems: Operational Planning and Control Issues, Ahm et Satir (ed.) pp. 115 132. E BRA H IM PO U R , M ., and F AT H I , B. M ., 1985, Dynam ic sim ulation of a kanban production inventory syste m . International Journal of Operations and Production Managem ent, 5, 5 14. E LM A G H RA BY, S. E., 1978, The econom ic lot scheduling problem (ELSP): review an d extensions. M anagement Science, 24, 587 598. F IN CH , B. J., and C O X , J. F., 1986, An exam ination of just-in tim e m an agem ent for the sm all m an ufactu ring: with an illustration. International Journal of Production Research, 24, 329 342. G A RE Y, M ., T AR JA N , R., and W ILFO N G , G., 1988, O ne-processor scheduling with sym m etric early and tardiness penalties. M athematics Operations Research, 13, 330 344. G LA SSE RM AN , P., an d Y A O , D . D ., 1994, M onotone Structure in Discrete-event Systems ( John Wiley, N ew York). G R AH A M , I., 1992, C om paring trigger and kanban control of ow-line m an ufac tu re. International Journal of Production Research, 30, 2351 2362. G R AVE L , M ., and P R ICE , W. L ., 1988, Using the kanban in a job shop environm ent. International Journal of Production Research, 26, 1105 1118. G R U WAL D , H., S T R IEKW O O D , P. E. T., an d W EE D A , P. J., 1987, A fram e-work for quantitative com parison of production control concepts. International Journal of Production Research, 27, 281 292. G U PTA , Y. P., and G U PTA , M . C., 1989a, A syste m dynam ic m odel of a JIT-kanban syste m . Engineering C osts and Production Econom ics, 18, 117 130. G U PTA , Y. P., and G U PTA , M . C., 1989b, A system dyn am ics m odel for a m ulti-stage-line dual-card JIT kanban syste m . International Journal of Production Research, 27, 309 352. G U PTA , Y. P., M A N G O L D , W. G., and L O N IAL , S. C., 1991, An em pirical exam ination of th e characteristics of JIT m anufactu rers versus non-JIT m anufactu rers. M anufacturing Review, 4, 78 86. H A LL , R. W., 1981, Driving the Productivity M achine: Production and Control in Japan (A m erican Production an d Inventory C ontrol Society, Falls C hurch, VA). H A LL , R. W., 1983, Zero Inventor y (Dow-Jones-Irwin, Hom ewood, IL ). H A RVE Y, D. A., and JO N ES, , S. W., 1989, Design a kanban syste m in an aerospace environm ent. In Proceeding of the International Conference on Implem enting Flexible M anufacturing Challenges for Organization and Education in a Changing Europe, Am sterdam , Netherlands, pp. 151 161. H ER N A ND E Z , A., 1989, Just-in-time M anufacturing: A Practice Approach (P rentice Hall, E nglewood Cliffs, NJ), pp. 14 20.

188

C .-C . H u an g and A . Ku sia k

M O N D E N , Y., 1983a, Toyota Production System: Practical Approach to Production M anagement (Ind ustrial Engineering an d M anag em ent Press, Norcross, G A). M O N D E N , Y., 1983b, Applyin g Just In Time (In dustrial Engineering and Man agem ent Press, Norcross, G A). O H N O , K., N AK A SH IM A , K., and K O JIM A , M ., 1995, Optim al num bers of tw o kinds of kanbans in a JIT production syste m . International Journal of Production Research, 33, 1387 1402. O H N O , T A IIC H I , 1988, Toyota Production System: Beyond Largescale Production (P roductivity Press, Cam bridge, MA). O LH A G ER , J., an d O ST LU ND , B., 1990, An integrated push-pull m an ufactu ring strate gy. European Journal of Operational Research, 45, 135 142. O T EN T I , S., 1991, A m odi ed kanban system in a semiconductor m anufactu ring environm ent. In 1991 IEEE/SEM I Advanced Semiconductor M anufacturing Conference & Workshop, Boston, M A, pp. 43 45. P ER VO ZVA N SK IY, A. A., and S H E YNIS , I. Y., 1994, A kanban system as a m anufactu ring control syste m with feedback. International Journal of Com puter and Systems Science, 32, 153 157. P ET R O F F, J O H N , N ., 1993, Handbook of M RP II and JIT: Strategies For Total M anufacturing C ontrol (P rentice Hall, Englewood C liffs, N J), pp. 5 13. P H ILIPO O M , P. R., R E ES , L . P., T A YL O R III B. W., and H U A N G , P. Y., 1987, An investigation of the facto rs in uencing th e num ber of kanbans required in th e im plem entatio n of th e JIT technique with kanbans. International Journal of Production Research, 25, 457 472. P R IC E , W., G R AV EL , M ., N SA K AN D A , A. L., an d C AN T IN , F., 1995, Modeling th e perform an ce of a kanban assem bly shop. International Journal of Production Research, 33, 1171 1177. R EE S , L. P., H U A N G , P. Y., and T A YL O R III, B. W., 1989, A com parative an alysis of an M RP lot-for-lot syste m an d a Kanban syste m for a m ultistage production operation. International Journal of Production Research, 27, 1427 1443. R EE S , L. P., P H ILIPO O M , P. R., T A YL O R III B. W., and H U A N G , P. Y., 1987, D ynam ically adjusting the kanbans in a just-in tim e production syste m using estim ated values of lead tim e. IIE Transactions, 19, 199 137. S AR K ER , B. R., 1989, Simulation a just-in -tim e production system . Com puters and Industrial Engineering, 16, 127 137. S ARKER , B. R., and F ITZSIMM O NS , J. A., 1989, The perform ance of push and pull system s: a simulation and com parative study. International Journal of Production Research, 27,1715 1731. S AR K ER , B. R., and H AR R IS , R. D ., 1988, The effect of im balance in a just-in-tim e production syste m : a sim ulation stud y. International Journal of Production Research, 26, 1 18. S CH O N BER G E R , R. J., 1982a, Japanese M anufacturing Techniques (Free Press, New York). S CH O N BER G E R , R. J., 1982b, Inventory control in Japanese industry. In M anagement by Japanese Systems, Lee, S. M . an d Schwendim an (eds) (P raeger Publishers, New York). S CH O N BER G E R , R. J., 1983, Application of single-card and dualcard kanban. Interfaces, 13, 56 67. S CH O N BER G E R , R. J., 1986, World Class Manufacturing: The Lessons of Simplicity Applied (Free Press, New York). S H IN G O , S., 1987, The Saying of Shigeo Shingo: Key Strategies for Plant Improvem ent (Productivity Press, Cam bridge, M A). S H IN G O , S., 1988, N on-stock-production: The Shingo System for Continuous Improvement (P roductivity Press, Cam bridge, M A).

H U A N G , P. Y., R E ES , L. P., and T A YL O R III, B. W., 1983, A sim ulation an alysis of th e Japanese JIT techniques (w ith kanban) for m ultiline, m ultistage production system s. Decision Sciences, 14, 326 344. I CH IH A SH I , E., 1990, An exam ple of CIM in Nippondenso Kota plant. System Control and Information, 34, 185 187. I N M A N , R., and B U LFIN , R., 1991, Sequencing JIT m ixed-m odel assem bly lines. M anagem ent Science, 37, 192 207. J A PAN E SE M A NA G E M E NT A SSO CIAT ION , 1986, Kanban: Just-intime at Toyota (P roductivity Press, Stam fort, CT). J O RD A N , S., 1988, Analysis and approximation of a JIT production line. Decision Sciences, 19, 672 681. J O TH ISH AN K A R , M . C ., and W AN G H. P., 1992, D eterm ination of optim al num ber of kanban using stochastic Petric Nets. Journal of M anufacturing Systems, 11, 449 461. J O TH ISH AN K A R , M . C ., an d W A N G , H. P., 1993, M etam odelling a just-in-tim e kanban syste m . International Journal of O perations & Production M anagement, 13, 18 36. K AR M A R K AR , U. S., an d K EK R E , S H AM , 1989, Batch ing policy in kanban syste m s. Journal of M anufacturing Systems, 8, 317 328. K IM T A E-M O O N , 1985, Just-in -tim e m anufacturing system : a periodic pull syste m . International Journal of Production Research, 23, 553 562. K IM U R A , O., an d T E R AD A , H ., 1981, D esign an d analysis of pull syste m , a m ethod of m ulti-stag e production control. International Journal of Production Research, 19, 241 253. K RA FT, C., 1992, D yn am ic kanban sem iconductor inventory m anagem ent syste m . In 1992 IEEE/SEM I Advanced Semiconductor M anufacturing Conference & W orkshop, Cam bridge, M A, pp. 30 35. K RA JE W SK I , L. J., K IN G , B. E., R ITZM A N . L . P., and W A N G , D . S., 1987, Kanban, M RP, and sh aping th e m anufactu ring environm ent. M anagem ent Science, 33, 39 57. K U BIA K , W., and S E TH I , S., 1989, A note on level schedules for m ixed-m odel assem bly lines in just-in-tim e production syste m . M anagement Science, 37, 901 904. L I, A., an d C O , H . C., 1991, A dynam ic program m ing m odel for the kanban assignm ent problem in a m ultistage m ultiperiod production syste m . International Journal of Production Research, 29, 1 16. L U , Y U A N , Y., 1982, Introduction to Industrial Engineering (H uaTai, Taipei, Taiwan ). M ILTE NBU R G , J., 1989, L evel schedules for m ixed-m odel assem bly lines in just-in-tim e production system . M anagement Science, 32, 121 122. M ILTE NBU R G , J., and S IN NA M O N , G., 1989, Scheduling m ixedm odel, m ultilevel assem bly lines in just-in-tim e production syste m . International Journal of Production Research, 27, 1487 1510. M ITRA , D., and M ITRA N I, T., 1990, An alysis of a kan ban discipline for cell coordination in production lines. M anagem ent Science, 36, 1548 1566. M ITRA , D ., and M ITR AN I , I., 1991, Analysis of a kan ban discipline for cell coordination in production lines. II. Stochastic deman ds. Operations Research, 39, 807 823. M ITWA SI , M. G ., and A SK IN , R. G., 1994, Production planning for a m ulti-item , single-stag e kanban syste m . International Journal of Production Research, 32, 1173 1195. M O N D E N , W., 1981a, W hat m akes th e Toyota production syste m really tick. Industrial Engineering, 13, 36 46. M O N D E N , Y., 1981b, Adaptable kanban syste m helps Toyota m aintain just-in -tim e production. Industrial Engineering, 13 (5), 29 46.

O verview of Kanban system s

S H IPPER , D ., and S H A PIRA , R., 1989, JIT vs. W IP-a trad e-off an alysis. International Journal of Production Research, 27, 903 914. S IH A , S., 1994, The pull production syste m : m odeling and characteristics. International Journal of Production Research, 32, 933 950. S IN G H , N ., and F AL K EN BU R G , D . R., 1994, Kan ban syste m s. In H andbook of Design, M anufacturing, and Automation, D orf, R. C . and Kusiak, A. (eds) ( John W iley, New York), pp. 567 585. S IN G H , N., K W O K , H. S., and M EL O C H E , D ., 1990, Th e developm ent of a kanban system : a case stu dy. International Journal of Operations & Production M anagem ent, 10, 28 36. S O , K. C., an d P IN A U LT, S. C., 1988, Allocating buffer storages in a pull syste m . International Journal of Production Research, 26, 1959 1980. S O H AL , A. S., L E W IS , G., an d S A M SO N , D ., 1993, In tegratin g C NC technology an d the JIT kanban syste m : a case stud y. International Journal of Technology M anagem ent, 8, 422 431. S O H AL , A. S., and N AYLO R , D ., 1992, Im plementatio n of JIT in a sm all m an ufac tu ring rm . Production and Inventor y M anagem ent Journal, 33, 20 26. S PE AR M A N , M . L., W O O D RU FF, D . L., an d H O O P, W. J., 1989, A hierarchical architecture for production syste m s. Journal of M anufacturing and Operations M anagem ent, 2, 147 171. S PE AR M A N , M . L., W O O D RU FF, D . L., an d H O O P, W. J., 1990, C ON W IP: a pull alternative to kanban. International Journal of Production Research, 28, 879 894. S PE AR M A N , M . L., 1992, Custom er service in pull production syste m s. Operations Research, 40, 948 958. S U G IM O RE , Y., K U SU N O KI , K., C H O , F., and U C H IKAWA , 1977, Toyota production syste m an d kanban syste m : m aterializa-

189

tion of just-in-tim e an d respect-fo r-hum an syste m . International Journal of Production Research, 15, 553 564. S U ZA KI , K., 1987, The New M anufacturing Challenge (Th e Free Press, New York). T A YU R , S. R., 1993, Structural properties and a heuristic for kanban-controlled serial lines. Managem ent Science, 39, 1347 1368. U ZSO Y, R., and M A RT IN -V E G A , L. A., 1990, M odeling kanbanbased dem and-pull syste m s: a survey and critique. M anufacturing Review, 3, 155 160. V ILLED A , R., D U D EK , R., an d S M ITH , M . L., 1988, In creasing th e production rate of a just-in-tim e production system with variable operations tim es. International Journal of Production Research, 26, 1749 1768. W A N G , H., an d W A NG , H . P., 1990, Determ ining th e num ber of Kanban s: a step toward non-stock-production. International Journal of Production Research, 28, 2101 2115. Y A NA G AWA , Y., M IYA ZA KI , S., and O H TA , H ., 1994a, An optim al operation planning for th e xed quantity with drawal kanban syste m with variable lead tim es. International Journal of Production Economics, 33, 163 168. Y A NA G AWA , Y., M IYA ZA K I , S., and O H TA , H., 1994b, Th e optim al operation planning of a Kan ban syste m with variable lead tim es. Production Planning & Control, 5, 21 29. Y AVU Z , I. H ., an d S ATIR , A., 1995a, A kanban-based sim ulation stud y of a m ixed m odel just-in-tim e m anufactu ring line. International Journal of Production Research, 33, 1027 1048. Y AVU Z , I. H., and S ATIR , A., 1995b, Kan ban-based operational planning an d control: sim ulation m odeling. Production Planning & Control, 6, 331 344. Z IPKIN , P. H., 1991, Computing optim al lot sizes in the economic lot scheduling problem. Operations Research, 39, 56 63.