Anda di halaman 1dari 18

This article was downloaded by: [Dicle University]

On: 05 November 2014, At: 04:52


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Production Planning & Control: The Management of


Operations
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tppc20

Integrating Kanban principles in a pharmaceutical


campaign production system
a a a
Jrgen Strohhecker , Rainer Sibbel & Marcel Dick
a
Management Department, Frankfurt School of Finance & Management, Frankfurt,
Germany.
Published online: 03 Jul 2013.

To cite this article: Jrgen Strohhecker, Rainer Sibbel & Marcel Dick (2014) Integrating Kanban principles in a pharmaceutical
campaign production system, Production Planning & Control: The Management of Operations, 25:15, 1247-1263, DOI:
10.1080/09537287.2013.809494

To link to this article: http://dx.doi.org/10.1080/09537287.2013.809494

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Production Planning & Control, 2014
Vol. 25, No. 15, 12471263, http://dx.doi.org/10.1080/09537287.2013.809494

Integrating Kanban principles in a pharmaceutical campaign production system


Jrgen Strohhecker*, Rainer Sibbel and Marcel Dick

Management Department, Frankfurt School of Finance & Management, Frankfurt, Germany


(Received 19 January 2012; nal version received 22 May 2013)

Centralised planned campaign production is the predominant production system in process industries. In this study,
we investigate whether a decentralised Kanban control system, which has proven to offer advantages in other
industries, can be successfully integrated in a campaign production environment. The used research methodology
combines model-based and case-study-based elements. Using the example of a pharmaceutical internal ve-stage
supply chain, we conceive a Kanban concept that integrates campaign formation, develop a discrete-event simulation
model and conduct a range of explorative simulation experiments. We nd that a Kanban campaign production
Downloaded by [Dicle University] at 04:52 05 November 2014

system is not merely feasible but would also be favourable: throughput times can be reduced without increasing
customer lead times. Sensitivity analysis shows that the systems performance is relatively robust to changes in
Kanban key conguration parameters such as number of Kanban cards or campaign size. We conclude by discussing
our ndings and formulating three propositions that might stimulate future research.
Keywords: Kanban; campaign production; batch production; process industry; simulation; decentralised production
control

1. Introduction set-up times and costs, it also results in increased inven-


In process industries, such as the chemical, pharmaceuti- tory levels and prolonged total throughput times (TPTs).
cal or food industries, a wide variety of transformation When campaign formation is allowed, the complexity
systems are used in praxis (Dennis and Meredith 2000). of lot sizing and scheduling problems see e.g. Pinedo
In most cases, the production of nal products requires (2005) for an overview or Jans and Degraeve (2008) for
certain consecutive stages of processing and various a review is raised. Standard lot-sizing and scheduling
intermediate products (Neumann, Schwindt, and Traut- models are deemed inappropriate for solving the com-
mann 2002). Besides continuous transformation types, plex campaign planning problem (Overfeld and Witte
which are often referred to as typical for the process 1992). More advanced models and solution approaches
industry, batch production and hybrid forms can often be should be used for example, those discussed by
found (Dennis 2000; Abdulmalek and Rajgopal 2006). Susarla and Karimi (2011), Bouchriha, Ouhimmou, and
For low-to-medium volume products, dedicated produc- D'amours (2007), Suerie (2005), Kallrath (2002), or
tion lines would often be underutilised and therefore Papageorgiou and Pantelides (1996a, 1996b). Yet, those
inefcient. As a consequence, these products are typi- approaches typically require that the production planning
cally manufactured in batch mode using multi-purpose process is organised in a centralised, push-oriented fash-
equipment of limited capacity (Grunow, Gnther, and ion as, for example, realised in MRP, MRP II and
Lehmann 2002). A specic characteristic of this type of advanced planning systems (Higgins, Le Roy, and Tier-
production is that set-up times between two batches of ney 1996; Stadtler and Kilger 2005; Steger-Jensen et al.
one and the same product are usually considerably 2011). In striving for reduced TPTs, less planning effort
shorter than the set-up process in between the batches of and increased competitiveness, companies using a cam-
two different products (Papageorgiou and Pantelides paign production approach like the pharmaceutical
1996b; Grunow, Gnther, and Lehmann 2002). To enterprise we worked with and used as a case study
reduce overall total idle times, campaigns can be estab- nd decentralised, lean, pull-oriented production systems
lished, which means that several batches of the same increasingly attractive (Lyons et al. 2013). Although lean
product follow each other immediately (Overfeld and concepts have traditionally been thought of as most
Witte 1992, Kallrath 2002, Susarla and Karimi 2011). applicable to the discrete manufacturing industries
While campaign formation indeed helps to decrease (Abdulmalek 2006), plant and supply chain managers

*Corresponding author. Email: j.strohhecker@fs.de

2013 Taylor & Francis


1248 J. Strohhecker et al.

from the process industry ask whether Kanban principles several approaches focusing on one-product systems with
can be integrated in a campaign-oriented production a single stage (e.g. Bonvik, Dallery, and Gershwin 2000;
environment and what benet could be expected from Baynat et al. 2001) and two or multi-stages (e.g. Askin,
such decentralised production control concepts (e.g. Mitwasi, and Goldberg 1993; Krieg and Kuhn 2004).
Mukhopadhyay, Dwivedy, and Kumar 1998). Less research has been undertaken investigating multi-
Developed mostly within Toyota and in cooperation product, single-stage Kanban systems (e.g. Davis and
with its suppliers in the 1950s and 1960s as an important Stubitz 1987; Berkley 1993), and only Krieg and Kuhn
component of the just-in-time system (Sugimori et al. (2008) address Kanban-controlled systems with multiple
1977), the Kanban production system is nowadays well products and two stages analytically. For more complex
known and thoroughly implemented in many companies and realistic Kanban systems, simulation-based
and industries (Schonberger 2007). While various ver- approaches are used to determine the appropriate cong-
sions of Kanban systems are documented (e.g. Sugimori uration (e.g. Huang, Rees, and Taylor 1983; Gupta and
et al. 1977; Berkley 1992; Bonvik, Couch, and Gershwin Gupta 1989; Berkley 1993; Sita 1998; Sengupta, Sharief,
1997; Dallery and Liberopoulos 2000; Allen, Robinson, and Dutta 1999; Li 2003). Simulation approaches are
and Stewart 2001) and even though these versions some- also used to compare various versions of pull and/or
times cause debates and contribute to confusion, most of push policies based on real cases (e.g. Huang, Wang,
them are rather similar and differ only in details. The and Ip 1998; Yang 2000; Khojasteh-Ghamari 2009).
Downloaded by [Dicle University] at 04:52 05 November 2014

core principle common to all variants is the idea of pull Yet, to our knowledge, no study exists that examines
instead of push: Kanban systems substitute traditional the applicability of Kanban in a campaign production
production-driven, centralised, push-oriented planning setting with high set-up times and explores the impact of
techniques with market-driven, decentralised, pull-ori- typical Kanban conguration parameters on production
ented principles (Hopp and Spearman 2004). The basic and supply chain performance. Therefore, the objective
idea behind Kanban is to control work-in-progress inven- of this paper is two-fold: First, we examine if and how
tory by dening supermarkets with a xed maximum Kanban control can be integrated in a campaign produc-
capacity and start the production of a primary or inter- tion system; and second, we analyse the impact of the
mediate product only if customer orders or downstream campaign Kanban conguration on the performance of
processes send a production request by forwarding a so- the production system by conducting a range of simula-
called Kanban card (Sugimori et al. 1977; Krieg and tion experiments.
Kuhn 2008). The remainder of this paper is organised as follows.
Various research studies provided evidence for a In the next section, we introduce the case and provide
positive impact of pull production techniques on produc- support for the research methodology used, which com-
tion performance. White, Pearson, and Wilson (1999) bines case-study-based and simulation-model-based ele-
and Fullerton and McWatters (2001), for example, found ments. Based on the Kanban principles published, we
such evidence using a survey technique. By applying develop in Section 3, a generic concept of a Kanban pro-
models that provide analytical solutions, Spearman and duction system that integrates campaign formation and
Zazanis (1992) derived that pull systems have less con- apply this concept to the case. Section 4 presents a more
gestion and are inherently easier to control. Li (2003) in-depth structural description of the discrete event simu-
concluded from a simulation study that pull systems lation (DES) model. In Section 5, we provide informa-
eventually outperform push systems, even in an unfa- tion on the parameterisation of the model and discuss
vourable jobshop setting. Billesbach (1994) even our validation efforts. Section 6 answers the research
reports on a successful implementation of a Kanban-like questions and discusses the results. In the concluding
pull system in a process facility. Therefore, Kanban may section, we summarise our work and suggest future
represent a promising concept for campaign-producing research directions.
companies as well even though rather high changeover
times between two campaigns violate at least one of the
typical Kanban prerequisites (Ohno 1988; Hopp 2004). 2. A pharmaceutical production process as case
Within the Kanban control concept, the aspect of study
specifying the number of Kanban cards and the lot size The case study is based on the ve-stage batch produc-
assigned to each card, which together determine the size tion process for a pharmaceutical tablet for self-medica-
of the supermarkets, has triggered intensive research as tion that is produced in medium-level quantities. For the
these parameters are crucial for the performance of a remainder of the paper, the tablet is referred to as
Kanban system (see e.g. Berkley 1992 for reviews, Price, SMMV (self-medication mid-volume) pharmaceutical.
Gravel, and Nsakanda 1994; Sendil Kumar and Panneer- The case and research site is a multinational companys
selvam 2007). For simple or simplied Kanban models, plant in Germany, where a large variety of pharmaceuti-
analytical procedures have been proposed. There are cals are produced and packaged. The plant utilises
Production Planning & Control 1249

single- and multi-purpose equipment. Campaigns of The outputs of these two weighing processes are the
production batches are typically formed in production intermediates i1 and i2. For i2, the subsequent granula-
stages that use multi-purpose equipment. Within the fac- tion process renders the powdery raw materials to a form
tory site, a functional layout is used, and the single that improves the compaction characteristics of the pow-
stages of the tablets production process are spatially dis- der. It also facilitates the ow ability and handling (e.g.
persed. Therefore, internal (intra-plant) supply chains Gad 2008). Since the production of one batch i4 requires
(Harland 1996) are dened and dedicated supply chain one batch i1 and 0.915 batches i3, the granule (i3) has
managers are installed for the SMMV tablet and all the to be weighed again prior to starting the blending pro-
other pharmaceutical products to ensure a holistic man- cess; the excess amount of i3 is transported back to the
agement of the production process. The supply chain i3 inventory. In the process of nal blending, i1 and i3
managers task is to coordinate the planning and control are mixed to obtain the required homogeneity of the tab-
process, to identify opportunities for improvement and lets ingredients. The next process produces tablets as
initiate improvement processes. output. Tableting transforms the mixed powder, a non-
Following the principles of a functional layout, simi- coherent solid phase, into the compact tablet (Gad 2008).
lar activities and resources are grouped together. Figure 1 Tablets not yet packaged are called bulk ware. The nal
illustrates this type of layout schematically for the phar- stage of the production process under consideration is
maceutical batch production process. Raw materials and the packaging process. In our case, packaging is a diver-
Downloaded by [Dicle University] at 04:52 05 November 2014

intermediate products are stored in a central inventory. gent-type production: the tablets are offered in different
Automated guided vehicles carry out all transportation packaging sizes for the two countries Germany and Aus-
processes between the functional areas and the inventory. tria. As Figure 2 indicates, six end-product variants are
Figure 1 also claries that SMMV is only one out of the output of the production process.
several product lines that is produced in the plant. Non-dedicated equipment is used for the processes
In Figure 2, SMMVs production process is illus- Weighing, Granulation and Blending (see also
trated as a state task network a concept that was intro- Figure 1). The weighing system, the granulator and the
duced by Kondili, Pantelides, and Sargent (1993) in pear-shaped mixer are also utilised for the production of
order to concisely represent chemical batch production some other products. Cleaning and set-up times for this
processes. The starting point of SMMVs production are equipment are considerably longer when products are
two weighing processes, where two or three raw materi- changed, for example, when one batch i2 with the active
als are weighed according to the pharmaceutical formula. ingredient combination of paracetamol and lactose is

Figure 1. Batch production system with functional layout (schematic illustration).


1250 J. Strohhecker et al.

(1)
r1
Weighing i1
r2
e1
(1)
e2
r3
(1) (1) (1) (1)
Weighing & e3
r4 Weighing i2 Granulation i3 i4 Tableting i5 Packaging
Blending
(1) (.915) (1) e4
r5
e5

rx e6
x = 611
Intermediate
Raw material End product
product
Downloaded by [Dicle University] at 04:52 05 November 2014

Figure 2. The production process illustrated as state task network.

granulated after one batch i2 with the ingredients ibupro- oriented objectives simultaneously. However, a modern,
fen and silicon dioxide. In order to reduce total set-up so-called advanced planning system (e.g. Steger-Jensen
and cleaning times, campaigns are therefore formed in et al. 2011) that takes the limited availability of
these three production stages. However, as a side effect, resources into account and uses true optimisation
waiting times for production orders before this non-dedi- techniques is not in place.
cated equipment becomes available are increased. SMMVs supply chain manager perceived several
The last two production stages covered in Figures 1 disadvantages in the centralised campaign planning
and 2 are tableting and packaging. For tableting, a dedi- system. First, as timely delivery is very important and
cated tablet press machine is used. As only one type of closely monitored, slack times are included in the vari-
tablet (with one and the same active ingredient) is pro- ous jobs standard processing times used for scheduling.
duced on this dedicated press, cleaning and set-up times Non-value adding times, such as safety margins before
do not depend on the product sequence and campaign and after production, replenishment lead times and pro-
formation is not required. Packaging is done on two cessing times for received goods, were added to the core
packaging lines. For the end products e1, e2 and e3, production time and inate the TPT. At the same time,
which are produced in rather high volumes for the Ger- this results in relatively high levels of work in process
man market, a dedicated line is available. End products inventory. Second, the exibility of the centralised sys-
e4, e5 and e6 are packaged for the Austrian market on a tem to adjust to uctuating demand was criticised as
non-dedicated line. As the tablet is offered in a variety poor. And third, maintenance of the master data was
of different packaging sizes for various countries, the claimed to be laborious and prone to error. In summary,
packaging process is order-driven to avoid high inven- the supply chain manager criticised the whole central
tory levels of nished products: packaging is started only planning and scheduling system as time consuming,
when a customer order is due. Upon completion of pack- inexible and bloated resulting in high overhead costs
aging, the products are immediately sent to the customer criticism, which can also be found in the literature (e.g.
and, therefore, they are no longer within the the supply Simchi-Levi, Kaminsky, and Simchi-Levi 2008).
chain managers scope of responsibilities. In order to identify levers to improve production
Following the typical approach of campaign-produc- performance, a lean manufacturing initiative was set up.
ing companies, production planning and control in our As one part of this initiative, value stream analysis
case study are based on a central approach. Incoming workshops were run (Rother and Shook 2003; Seth and
customer orders are fed into a central ERP system, Gupta 2005). The goal, according to SMMVs supply
which supports material requirements planning, generates chain manager, was to reduce TPT signicantly
a master production schedule and creates production currently about 70 days without putting market supply
orders for each single stage of production (e.g. Higgins, at risk. In addition to surfacing some minor improvement
Le Roy, and Tierney 1996). Based upon continuously ideas, the value stream analysis workshops identied
updated demand forecasts, the central production one major process innovation for development and
planning and control system supports operational discussion the idea of shifting from the centralised
decision-making, considering both cost- and time- planning and scheduling concept to a decentralised
Production Planning & Control 1251

scheme based on the introduction of Kanban principles. process, where eld research is incorporated into a
Hoping and expecting that a Kanban campaign produc- model and model results guide further eld research,
tion system could indeed signicantly improve produc- leading to relevant and rigorous knowledge (Graham
tion performance, the future state map for this new et al. 1992).
process design was conceived and intensively discussed. The DES model that emerged from this iterative pro-
Out of intense discussions in the workshop about the cess allowed us to thoroughly test the Kanban campaign
compatibility of Kanban control and campaign forma- production concept that is developed in the following
tion, three important questions arose: section: it facilitated extensive experimentation, including
the application of scatter search optimisation by using
(1) How can Kanban control be introduced in a cam-
OptQuest (Laguna 2011; Kelton, Sadowski, and Swets
paign production system and how should this sys-
2010), it enabled us to provide profound answers to the
tem be designed?
supply chain managers questions.
(2) What performance can be expected from such a
Kanban control campaign system?
(3) How sensitively would the new production sys- 3. A Kanban campaign production concept
tem react to changes in conguration parameters,
For the ve-stage batch production process shown in
such as supermarket and campaign size (CS), and
Figures 1 and 2, a decentralised Kanban control concept
to changes in the environment, such as demand?
Downloaded by [Dicle University] at 04:52 05 November 2014

was developed adapting multi-product, multi-stage Kan-


We address these questions by enhancing the case ban schemes as, for example, described by Schonberger
study via model support. A model-supported case study (1982), Allen, Robinson, and Stewart (2001) or Baynat,
entails, in addition to library research, interviews con- Buzacott, and Dallery (2002). An overview of the
ducted during a site visit, process documentation and concept is provided in Figure 3. It is evident that the
data analysis, the development of a DES model. Rockart functional layout of the production system is preserved.
(2001), for example, discusses that model support has In the central inventory, Kanban supermarkets with a
the favourable consequence in which the analysis of the maximum size determined by the number of Kanban
model could be used to brighten and challenge insights cards and the production batch size are installed. As
emerging from the case study. Moreover, while analysing packaging shall remain directly order-driven, no super-
the model, further ideas and questions typically arise that markets are introduced for packaged products. Once a
had not surfaced in eld research but could be used to customer order is due, one or more packaging orders are
guide subsequent interviews, additional data collection, created depending on the amount ordered and the
and more focused analyses. The result is an iterative maximum packaging batch size.

Supplier Customer

Weighing
1

Weighing Final
Granulation Tableting Packaging
2 Blending

Weighed Ingredients Granule Blend Tablet Bulk Ware


Supermarket Supermarket Supermarket Supermarket

Legend: Production KANBAN Card Withdrawal KANBAN Card Customer Order

Material Flow Information Flow Supermarket

Figure 3. The Kanban system design of the pharmaceutical production process illustrated as a value stream map.
1252 J. Strohhecker et al.

For each packaging order, a withdrawal order is sent the remaining upstream PKCs and WKCs are determined
to the tablet bulk ware supermarket, where the required analogously. Although a standard batch size for packag-
amount of tablets are taken and sent to the packaging ing is dened, customer orders smaller than this standard
line. Simultaneously, another withdrawal order is sent batch size are accepted and passed through. Large cus-
from packaging to the central inventory, releasing the tomer orders trigger the creation of more than one pack-
required packaging material, such as patient information aging order, dividing the amount ordered into standard
leaets and folding cardboards. When all material and packaging batch sizes and a residual. This has the conse-
the tablets are available, packaging can start as soon as quence that variable quantities of tablets and packaging
the line becomes idle. As soon as a number of tablets raw materials are withdrawn from the bulk ware super-
corresponding to the tableting production batch size are market and the inventory, respectively.
withdrawn from the bulk ware supermarket, a production Thus far, a generic Kanban concept is adapted to the
Kanban card (PKC) is sent to the preceding tableting case. Yet, to be compatible with the campaign production
process, thus releasing an order to reproduce one batch environment in our case, the Kanban control system has
of tablets. To receive the necessary amount of nal to incorporate the formation of campaigns. This is
blend, a withdrawal Kanban card (WKC) is sent to the achieved by introducing a collecting and bundling stage
blend supermarket. This serves as an authorisation for for both PKCs and WKCs, as shown in Figure 4 for
the movement of the nal blend from the supermarket to weighing and granulation as examples. As a result, PKCs
Downloaded by [Dicle University] at 04:52 05 November 2014

the tableting process. Once the blend is provided at the are not sent directly to the preceding production process
tableting stage and the machine becomes available, tab- in which batches of weighed ingredients are withdrawn
leting can start. from the supermarket. Instead, the cards are sent to a
The withdrawal of nal blend from the supermarket campaign formation process, where they are withheld
initiates, in turn, the forwarding of another PKC to the until the number of cards waiting is greater than or equal
preceding granulation process and so on. PKC and WKC to CS. Only when this condition is met, will the bundle
both remain with the batch of products until the whole of cards be sent to the preceding process. If, for exam-
lot has been consumed by the next process. Then they ple, CS is set to two batches for the weighing process,
travel back to the preceding process, thus creating the only bundles of two PKCs are allowed to travel from the
decentralised, self-controlled loops, which are in addi- weighed ingredients supermarket to the weighing pro-
tion to the pull principle the essential elements of the cess. One single PKC would have to wait before the
Kanban philosophy (Ohno 1988; Monden 1998). campaign formation stage for another card to arrive.
Kanban cards typically carry three types of informa- Although an explicit campaign formation stage can
tion: pickup, production and transfer data (Ohno 1988). also be introduced in the withdrawal process, this is only
A WKC species at a minimum the kind (material or necessary when campaigns have to be (re)formed for the
product number) and quantity of product that a manufac- transport from a supermarket to a manufacturing process.
turing process requires from an upstream supermarket. Reasons for this could be capacity limitations as well as
As a central multi-item supermarket is implemented, the the desire to achieve increased utilisation of transporta-
storage area has to be identied as well. In addition to tion capacity. Yet, even when campaigns for WKCs are
this, both the preceding and the subsequent processes are not formed explicitly, a succeeding process, which estab-
often listed. A PKC species the kind (material or prod- lishes production campaigns, will automatically pass on
uct number) and quantity (batch size) of product that a the campaign pattern to the upstream withdrawal
manufacturing process has to produce. In addition, the process.
storage area where the output is to be placed also has to Additionally, one has to adhere to the rule that bun-
be indicated. dles of Kanban cards have to be processed in successive
For all of SMMVs production stages, except for the
packaging process, the batch size for the production and WKC
Campaign
WKCs is predened. It cannot be easily changed because Formation
it has to be registered with the pharmaceutical
authorities, which is a costly and time-consuming pro-
cess. As illustrated in Figure 2, for one batch of tablets Weighing
Granulation
(i5), one batch of blend (i4) is needed. Therefore, the 2
PKC that is sent from the tablet supermarket to the
tableting process sets the quantity of tablets to be pro- PKC
duced to one batch. Batch sizes in units or kilograms PKC Campaign
Campaign Formation
have to be kept condential. Likewise, the WKC that is Formation
submitted from the tableting process to the blend super-
market shows a quantity of one batch. The quantities for Figure 4. Kanban control with campaign formation.
Production Planning & Control 1253

order without allowing any unbundling. In other words,


if priority rules other than rst come rst serve are
used to determine the sequence of production or trans- Blending Blending
Blending
portation jobs, the bundle of Kanban cards representing Waiting Fast Set-up
Delay
a campaign has to be seen as one job. Nevertheless, set- Delay Seize Delay
Release
up processes might have to be performed between a
jobs single batches. Figure 5. Process model of the blending stage. The delay types
With the Kanban campaign design outlined above, refer to the Arena modules used in the model.
the concept of campaign formation can be maintained in Note: The delay holds up an entity by the specied amount of
time. A seize delay structure also seizes equipment and
a Kanban control environment. This allows leveraging
therefore makes it unavailable for other processes. A delay
the advantage of reduced set-up times between two release structure frees the resource after the entity has been
batches of the same product. As a result, CS becomes delayed by the specied amount of time. See also, Kelton,
the third important conguration parameter for the Kan- Sadowski, and Swets (2010) for further information.
ban system, alongside with the traditional parameters,
batch size and supermarket size. When determining
those parameters, one has to make sure that CS for an ing machine becomes available, and once this occurs, the
intermediate or nished product is always smaller than machine is occupied by the subsequent set-up, shutdown
Downloaded by [Dicle University] at 04:52 05 November 2014

or equal to the size of the supermarket for that product. and tableting processes. The shutdown process represents
Otherwise, the supermarket would not be big enough to planned and unexpected downtimes of the tableting
store the complete campaign production output. machine. Once a batch of tablets is nished, a sample is
drawn and sent for testing to the laboratory. The tablets
themselves are placed in the bulk ware supermarket from
4. The DES model where they can be dispatched via WKC sent from the
In order to provide a proof of the Kanban campaign pro- succeeding packaging process. Packaging of bulk ware is
duction concept developed for the focused SMMV sup- allowed to start before the result from the laboratory
ply chain, a DES model was developed and simulation analysis has arrived; packaged products are queued until
experiments were conducted using the simulation soft- the nal release. While this practice contributes to
ware Arena (see, e.g. Kelton, Sadowski, and Swets shorten the TPT by allowing parallelisation, it increases
2010). To keep the model comprehensible and to avoid the value of goods at risk. The so-called qualied person
the need for enormous amounts of data, not all products releases the packaged products only if test results and all
that were produced on the non-dedicated machines were quality inspections comply with the specications. If test
included in the model. We wanted to focus on the results for a batch of tablets do not meet specications
SMMV products. This, however, required the inclusion and the qualied person denies nal release, the com-
of some mechanisms in our model that could represent plete batch has to be destroyed, no matter whether the
the blocking of jobs that occurs when a production or batch has been completely or partly, or not at all pack-
transportation job has to wait until the equipment used aged.
for producing non-SMMV products (which are not In order to validate the Kanban campaign concept,
explicitly included in the model) becomes available. Our we focus on the supply chain performance of the three
solution was to include a delay process representing this pharmaceuticals (e1, e2 and e3) produced for the Ger-
waiting time, as shown in Figure 5 for the blending man market. This also allows us to model packaging line
stage. In the best case, the delay time is zero, meaning II in a simplied way signicantly to reduce model com-
that the mixer is not utilised and is clean when a set of plexity without deteriorating the relevance and validity
PKCs arrive. Fast set-up, that is, the abridged set-up in of the results. Therefore, the packaging process illus-
between batches of the same campaign, and production trated in Figure 6 represents only the packaging process
can start instantly. In the worst case, the production for the German market and can be modelled analogously
process for the product with the biggest CS has just to the tableting process.
started, causing the Kanban production to wait rather Technically, the Kanban cards are modelled as enti-
long before the equipment becomes available again. ties with the batch size assigned as attributes (Treadwell
Similar structures were modelled for the weighing and and Herrmann 2005). Supermarkets are represented by
the granulation stage. queues of Kanban cards. A PKC is released when the
Dedicated equipment is used for tableting and pack- complete amount of intermediate product associated with
aging, and this allows us to replace the waiting delay this card is withdrawn. Then, the PKC is sent back to
with explicit seize-delay-release structures (see Figure 6). the preceding process where the production of a new
Kanban production, along with the associated batches of batch of the product is initiated. At all stages of the
tablets, are modelled to wait in a queue until the tablet- internal supply chain, a many-to-many relationship
1254 J. Strohhecker et al.

Tableting Tableting Laboratory


Tableting
Set-up Shutdown Testing
Delay Delay Delay
Delay yes
Press Busy Press Busy Press Busy

no Ok?

Shipping
Packaging Packaging Final yes
Packaging Ok?
Set-up Shutdown Release
Delay Delay Delay no
Delay
Line Busy Line Busy Line Busy

Bulk Packed
Ware Products
Destruction Destruction
Downloaded by [Dicle University] at 04:52 05 November 2014

Figure 6. Process model of the tableting and packaging stages.

Tablet
Withdrawal Final
Packaging
Material Release
Withdrawal

Customer Packging Order Orders


Orders Orders Backlog filled

Figure 7. Process model of customer order processing.

between orders is allowed, which means, for example, 5. Model parameterisation and validation
that a batch of packed products can consist of more than The model was parameterised partly based on statistical
one batch of tablets and one batch of tablets can be used evaluations of data directly derived from the companys
for more than one packaging order. ERP system and partly on expert estimates. For the dis-
Customer order generation and processing for the tribution tting, the Input Analyzer that is part of the
German market (e1 to e3) is modelled as outlined in Arena software package was used. To adequately gener-
Figure 7. Each customer order is represented by an entity ate customer orders, distributions for the order size, the
with order size and product type assigned as attributes. time between two successive orders and the frequency of
Based on this information, packaging orders are created, orders per product had to be determined. Table 1 shows
and tablet and packaging material withdrawal orders are forecasted aggregated customer order data per product
sent to the tablet supermarket and inventory, respectively. and year. The forecast for the number of orders was
The customer order is put in the order backlog queue, based on the assumption of constant average order sizes.
where it has to wait until all packaging orders are As SMMV is a mature product, the forecasted total
fullled and the nal release is granted. amount of batches was very close to the actual yearly
Since the non-dedicated packaging line is excluded production output. This allows for a comparison of the
from the model, customer orders for the Austrian market simulation output measures to actual performance with-
(e4 to e6) are modelled to withdraw the appropriate out any interference. The percentage distribution per
amount of tablets from the tablet supermarket (including product based on the number of orders was fed as dis-
a back-order queue) and then cross the boundary of the crete distribution into the model and used to determine
model. the product type.
Production Planning & Control 1255

Detailed order size data, which could be used for sta- Packaging line set-up and failure times showed no sys-
tistical estimation, was only available for products e1, e2 tematic differences between the products and did not
and e3. However, we were informed that orders for e4, depend on batch size. Packaging production times, how-
e5 and e6 have similar sizes as orders for e2 and e3, ever, depended on the packaging batch size (PBS) and
while e1s average order size is signicantly larger. differed between products. Therefore, a linear relation-
Therefore, we used the available data for e1 to nd the ship between production time and PBS was modelled
best tting customer order size distribution for e1. The and an error term was included. The slope was deter-
data for e2 and e3 were aggregated and used to deter- mined by regression (all were signicant at a 0.001
mine the distribution for all other products e2 to e6. To level) and the error distribution was retrieved from tting
ensure that the simulated orders on average cumulated to the residuals.
about 157 batches and 65 orders, the best tting distribu- Since the tablets are the same for all end-products
tion had to be slightly adjusted. Naturally, this negatively and since the tableting process uses constant batch sizes,
affected the goodness-of-t statistics that is shown in set-up, shut-down and production times do not depend
Table 2 along with the summary statistics for the real on batch size or product. Using the ERP data available
data and the tted distribution eventually used in the to nd the best-tting distributions resulted in the deni-
simulation model.1 In addition, based on expert knowl- tions compiled in Table 4.
edge, the minimum order size was restricted to 0.2 and For the weighing, granulation and blending process,
Downloaded by [Dicle University] at 04:52 05 November 2014

the maximum order size was limited to 6.0. ERP data could not be retrieved. Alternatively, standard
Production, set-up and failure times for the tableting times were used for the set-up and production processes
and packaging processes could be retrieved from the as indicated in Table 5. Waiting times that represent the
ERP system on a per order basis for a time span of 12 time span until the non-dedicated equipment becomes
months; this data was used to nd the best-tting available (see also Figure 5) were based partly upon
distributions. Table 3 compiles the parameterisation used expert opinions and partly upon data. Since some data
for the packaging process in the simulation model. was available on campaign TPTs for the blending stage,

Table 1. Aggregate customer order forecast per product and year.

e1 e2 e3 e4 e5 e6 Sum
Batches 106.24 4.92 20.19 10.76 12.51 2.08 156.70
% 67.8 3.1 12.9 6.9 8.0 1.3 100.0
No. of orders 32 5 12 7 8 1 65
% 49.2 7.7 18.5 10.8 12.3 1.5 100.0

Table 2. Summary statistics on customer order data used to estimate parameters.

Customer order parameter N Mean SD Fitted distribution SSE KSp


Time between arrivalof orders (workdays) 64 4.64 3.23 EXP(4.05) 0.0080 0.010
Order size e1 (batches) 31 3.28 1.51 WEIB(3.3, 2.3) 0.0296 0.037
Order size e2 to e6 (batches) 17 1.45 0.66 NORM(1.4, 0.64) 0.0869 >0.15

Table 3. Summary statistics on packaging process data used to estimate distributions.

Process times N Mean SD Fitted distribution SSE KSp R


Set-up 48 1.65 1.73 0.24 + LOGN(1.18, 0.62) 0.0038 >0.15
Abridged set-upa 48 0.28 0.10 0.1 + LOGN(0.183, 0.0948) 0.0247 >0.15
Shut-down 48 0.46 0.78 DISC(0.4583, 0, 1.0, GAMM(0.62, 1.47)) 0.0128 >0.15
Production 48 28.09 14.89
e1 31 33.59 15.00 9.971PBS + 0.894 + GAMM(1.22, 3.27) 0.0304 >0.15 0.978
e2 5 15.44 8.21 17.113PBS + 0.1152 + 6BETA(0.558, 0.968) 0.0803 >0.15 0.942
e3 12 19.16 7.25 12.993PBS - 2.709 + TRIA(-2.39, 0.835, 1.55) 0.0551 >0.15 0.982

Notes: Process times in hours per batch.


PBS = Packaging batch size.
a
In between campaigns.
1256 J. Strohhecker et al.

Table 4. Summary statistics on tableting process data used to estimate distributions.

Process times N Mean SD Fitted distribution SSE KSp


Set-up 146 1.68 1.49 25BETA(0.472, 5.95) 0.016 <0.01
Shut-down 146 0.67 1.38 0.001 + EXPO(0.671) 0.0127 <0.01
Production 146 7.61 1.38 6.11 + LOGN(1.5, 0.531) 0.01 0.14

Note: Process times in hours per batch.

Table 5. Parameterisation of the weighing, granulation and nal blending processes.

Resource Process times Judgementally set distribution


a
Weighing system Waiting Exponential(6)
Abridged set upb 0.25
Production 1.75
Granulator Waitinga Exponential(10)
Abridged set upb 0.25
Production 4
Downloaded by [Dicle University] at 04:52 05 November 2014

Mixer Waitinga Exponential(13)


Abridged set upb 0.25
Production 4

Notes: Process times in hours per batch.


a
Before non-dedicated equipment.
b
In between campaigns.

this data could be used to approximate the shape of the sience, each simulation was run for 100 workdays before
distribution. This resulted in the exponential distribution data over a time span of 780 workdays were collected
showing the best t. The distributions mean value was (equivalent to approximately three years). All scenarios
determined judgementally balancing the various pieces were simulated with 50 replications, which provide for
of information at hand. Table 5 shows the parameter val- an analysis of the performance measures with half-widths
ues that are used in the model. in the range of 5%.
For weighing, granulation and nal blending, cam- Model validation is a crucial step in any simulation
paigns of three batches for each were enforced. No cam- study and has attracted intense methodological research
paigns were established at the tableting and packaging (e.g. Naylor and Finger 1967; Cochran 1987). We con-
stage (meaning that CS is set equal to 1). Transport ducted a large number of different tests to build as much
times for the batches of intermediate products from the condence as possible in the DES model outlined above.
supermarkets to the production stations and vice versa Guided by the schemes of Banks, Carson, and Ii (2010)
were set judgementally to 0.5 h each, which somewhat and Sterman (2000), we conducted structural, behav-
simplies the reality. It was also assumed that transport ioural and statistical tests. Aiming for face validity, we
resources do not impose any capacity restrictions or the challenged the boundary of the model by discussing the
necessity for transport campaign formation in the with- model with SMMVs supply chain manager and other
drawal process. Transport times for the PKCs and WKCs experts. We carefully checked the model structure and
were neglected, which is equivalent to assuming the the level of aggregation against all material that had been
implementation of an electronic Kanban system as, for made available to us, including our own observations
example, suggested by Lee-Mortimer (2008) or Kouri, from the site visits. We carried out many test simulations
Salmimaa, and Vilpola (2008). Additionally, based on and tracked the ow of batches, orders, Kanban cards,
other Kanban simulation studies (e.g. Yang 2000), it was etc. in animation mode. Additionally, extreme condition
assumed that the weighing process is never starved of tests were performed, such as demand dropping to zero
raw materials r1 to r5 and that the packaging process has or CS set beyond supermarket capacity. For all extreme
an innite supply of auxiliary materials rx, such as condition tests, the model behaviour was plausible. For
patient information leaets or cardboard boxes. example, when demand ceased the supermarkets that
The simulation was started with empty supermarkets. were lling up to the allowed maximum level, then
This put some stress on the system in the rst one or production came to a complete halt. Finally, sensitivity
two months as customer orders could not be lled by analysis was employed to check the models face
drawing on intermediate products already stored in the validity, and results were discussed among us and with
supermarkets. To eliminate the resulting initial tran- experts.
Production Planning & Control 1257

Model parameter assumptions were widely based on For the comparison between the traditional push
the collection of ERP data and statistical analysis. The system and the new Kanban based system, the following
assumed statistical model was validated by a goodness- benchmark values were set: on average 70 workdays for
of-t test and graphical methods. Both probability distri- TPT and 15 workdays for COFT. While past data served
butions and judgementally set parameters were also to set the benchmark for TPT, the COFT benchmark had
discussed with experts. However, as the implementation to be determined judgementally by drawing on the average
of the Kanban campaign production system developed lead time, which was deemed acceptable for the custom-
and described above was not yet available, comparing ers. Based on the experiences with the push-controlled
the model inputoutput transformations to corresponding production, both targets are ambitious and reect the
transformations for the real system (Banks et al. 2010) performance limits of the traditional system. In the model,
was unfortunately not possible. COFT is calculated as the time difference between the
arrival time of a customer order and the time the packaged
order is nally released for shipment. TPT is determined
as the sum of the weighing, granulation, blending, tablet-
6. Results ing and packaging TPTs.
In order to address the rst research question, namely, if Since SMMVs management aimed at minimising
Kanban control can be successfully integrated in a cam- TPT without deteriorating COFT, the optimisation task
Downloaded by [Dicle University] at 04:52 05 November 2014

paign production system, an optimal conguration of can be formulated as follows:


the Kanban campaign system has to be determined and
its performance has to be compared to the centralised Minimise Average TPT
production control systems performance. As batch sizes Subject to Average COFT 6 15
cannot be easily changed in the pharmaceutical industry Control CSi 6 PKCi, i = 1(weighing) to 4 (tableting)
and CS in the rst three stages of SMMVs internal sup-
ply chain should not be changed, the conguration The OptQuest optimiser that is included in the Arena
parameters of the Kanban system are reduced to the software was used to solve this problem (Laguna 2011;
number of PKCs circling in the weighing, granulation, Kelton, Sadowski, and Swets 2010). Figure 8 shows more
blending and tableting stages. The scope for setting the than 300 Kanban configurations (that are characterised by
number of cards is even further restricted by the CS: to the number of PKCs per production stage), whose
avoid a deadlock situation, the number of cards has to performance measures satisfy the restrictions stated above
be always set greater than or equal to CS. with the optimal configuration highlighted. The effi-
To assess a Kanban production systems perfor- ciency frontier of the Kanban campaign production
mance, a broad range of measures can be used (e.g. system with regard to TPT and COFT is also marked in
Yavuz and atir 1995a; Najmi and Makui 2011). Among Figure 8. This line connects the dominant alternatives.
the key performance indicators monitored by SMMVs In order to determine the best conguration, the three
management, the most important and regularly inspected efcient alternatives with the lowest average TPT (1) to
measures are the TPT and the production service level, (3) along with two additional congurations (4) and (5)
which measures an average on-time delivery status. In that lie very close to the frontier are further investigated.
addition to these two measures, inventory levels are Table 6 shows the conguration characteristics and
monitored as well (which we nd in simulation experi- summary statistics on the two performance measures,
ments highly and signicantly correlated with TPT TPT and COFT. While conguration (1) leads to the
= .879, p < 0.001, N = 25). In the push system that is lowest mean average TPT, conguration (2) increases
actually used, customer orders are entered in the ERP TPT only marginally, while mean average COFT is
system with a delivery date assigned well in advance of
the due date. A different logic applies in the pull system.
Customers would place their orders not until the amount
ordered is really needed to meet demand taking a typi-
cal or guaranteed maximum delivery time into account.
Therefore, in a pull environment, a measure such as cus-
tomer order fullment time (COFT) is preferred to an
on-time delivery measure. In focusing on COFT and
TPT, we used a minimum set of performance indicators
to facilitate analysis, which is nevertheless sufciently
balanced to cover typical conicting goals in short-term
detailed production scheduling (Mensch 1972; Gutenberg
1983). Figure 8. Results from the heuristic optimisation process.
1258 J. Strohhecker et al.

Table 6. Number of PKC and summary statistics (N = 50) of the ve selected congurations.

Conguration 1 2 3 4 5
Number of PKC for production stage Weighing 1 3 3 3 5 4
Weighing 2 3 3 3 3 3
Granulation 4 4 4 4 4
Blending 3 3 3 3 3
Tableting 2 3 4 3 3
Avg TPT Mean 31.25 31.35 33.14 33.54 32.80
SD 1.46 1.68 1.51 1.52 1.71
Half-Widtha 0.42 0.48 0.43 0.43 0.49
Avg COFT Mean 13.38 11.59 10.85 10.93 11.63
SD 2.01 1.61 1.90 1.45 1.71
Half-Widtha 0.57 0.46 0.54 0.41 0.49
a
95% Condence Interval.

Table 7. Independent sample t-test t values for 10 pairs of congurations.


Downloaded by [Dicle University] at 04:52 05 November 2014

Conguration 1 2 3 4 5
2 TPT 0.307
COFT 4.926
3 TPT 6.376 5.613
COFT 6.467 2.086
4 TPT 7.685 6.836 1.327
COFT 6.989 2.137 0.234
5 TPT 4.868 4.273 1.066 2.303
COFT 4.681 0.144 2.165 2.217
SE mean TPT 0.206 0.238 0.213 0.215 0.242
COFT 0.284 0.228 0.268 0.241 0.241

, indicates signicance at the 95% and 99% level, respectively.
df = 98, N = 50.

reduced by 1.79 workdays. Ten independent sample Simulation results of conguration (2) show that a Kan-
t-tests were conducted to determine if differences in the ban campaign production system does clearly and signi-
means are statistically signicant (Table 7). cantly reduce average TPT (31.35 workdays) when
While mean average TPT does not signicantly differ compared to the status quo (70 workdays) of centralised
between congurations (1) and (2), mean average COFT push production (t(49) = 162.2, p < .001, r = .998).
is signicantly lower in conguration (2). When compar- While a comparison of simulation-based results to real
ing congurations (2) and (3), TPT increases signi- world measures is probably not completely free of
cantly contradicting the minimisation objective for TPT. biases, we think with all due caution that a perfor-
It has to be noted, though, that COFT is signicantly mance improvement of more than 55% is a relatively
lower in conguration (3) compared to (2), which allows
SMMVs supply chain management to trade-off TPT
against COFT. However, based on the TPT minimisation 80.0 14.0
Avg. TPT Avg. COFT
objective stated above, conguration (2) that also satis- [Workdays] [Workdays]
es the COFT restriction of not more than 15 workdays 70.0 13.0
Avg. COFT [workdays]
Avg. TPT [workdays]

is the optimal solution. The number of Kanban cards in


60.0 12.0
this conguration varies only slightly between three (for
the stages weighing, blending and tableting) and four
50.0 11.0
(for the granulation stage). This low variation corre-
sponds with the outputinput relations that are one or 40.0 10.0
close to one for all stages (see Figure 2).
Based on the simulation results presented in Tables 6 30.0 9.0
and 7, the answer to the rst research question whether O O+1 O+2 O+3 O+4
Kanban configuration
O+5 O+6

Kanban control can be successfully integrated in a


campaign production system is clearly yes, it can. Figure 9. Aggregate Kanban conguration sensitivity analysis.
Production Planning & Control 1259

Table 8. Aggregate Kanban conguration sensitivity analysis.

Kanban conguration no TSAO TSAO + 1 TSAO + 2 TSAO + 3 TSAO + 4 TSAO + 5 TSAO + 6


TPT (Workdays) 31.35 39.22 46.09 54.34 60.96 67.95 75.87
COFT (Workdays) 11.59 11.84 10.81 10.27 10.59 10.28 10.17
Total supermarket level (batches) 8.40 12.83 17.51 22.30 26.94 31.68 36.67
Tableting machine scheduled utilisation (%) 69.10 69.20 68.50 67.30 68.90 69.50 68.90
Packaging line scheduled utilisation (%) 59.20 60.10 59.30 57.50 59.60 59.40 58.50
Tableting set-up queue (batches) 0.57 1.07 1.40 1.65 1.94 2.17 2.23
Packaging set-up queue (batches) 1.14 1.65 2.15 2.44 2.85 2.99 3.30

Note: All values are mean values from 50 replications.

Table 9. Partial Kanban conguration sensitivity analysis.

Kanban Avg. TPT Avg. COFT Kanban Avg. TPT Avg. COFT
conguration (Workdays) (%) (Workdays) (%) conguration (Workdays) (%) (Workdays) (%)
W1 + 1 4.63 0.41 PSA W1D 9.06 2.49
Downloaded by [Dicle University] at 04:52 05 November 2014

W2 + 1 5.47 0.60 PSA W2D 9.54 0.23


G+1 6.30 0.77 PSA GD 28.72 3.09
B+1 6.21 0.90 PSA BD 18.42 0.81
T+1 5.72 6.34 PSA TD 17.04 11.34

strong indication of the advantages of a campaign-inte- of PKCs in the blending stage is increased to 4 instead
grated pull system over a push system. Without any of 3. In the right columns of the table, the number of
doubt, the result serves as a proof of the concept: intro- PKCs is doubled, that is, W2 D means that 6 PKCs
ducing Kanban pull principles in a campaign production (compared to 3) are used in weighing stage two. The
system is feasible. results show that the impact of increasing the number of
In order to provide additional insight into the PKCs on COFT is small and unsystematic except for
sensitivity of the Kanban campaign production system to the tableting stage where signicant reductions can be
changes in the supermarket sizes, a systematic aggregate observed. Regarding TPT, the impact of changes in the
and partial sensitivity analysis was conducted. First, middle of the internal supply chain tends to have larger
based on the optimal conguration (2) from Table 6 impacts than variations at the beginning or end.
(referred to as O in Figure 9), the number of PKCs was Relating these ndings to this researchs second
incremented repeatedly for all production stages simulta- objective, namely, to analyse the impact of changes of
neously. Decrementing the number of PKCs would vio- Kanban parameters on the performance of the production
late the constraint that the number of PKCs always has system, we can summarise that the proposed Kanban
to be greater than or equal to that of CS and is therefore campaign system shows a rather stable performance. In
omitted. As Figure 9 illustrates, increasing the number of the most sensitive production stage, an increase in the
PKCs leads to a quite linear increase in TPT, while number of PKCs from 4 to 5 (+ 25%) results in a clearly
COFT tends to decrease slightly (with little oscillations). less than proportional change in TPT and COFT. Even
Inventory levels and queue length before workstations doubling the number of PKCs (+ 100%) leads to a rela-
with dedicated equipment rise as well (see Table 8). Util- tively moderate rise of maximally 28.72% in TPT (which
isations of the tableting machine and packaging line do highly correlates with inventory).
not change signicantly. These results are in line with
the ndings presented by Gupta and Gupta (1989),
Yavuz and atir (1995b) and Huang, Wang, and Ip 7. Discussion, limitations and conclusion
(1998). To the best of our knowledge, this study is the rst to
Partial sensitivity analysis provides some insight on investigate the concept of a campaign production system
how the system behaves when only one parameter is which uses decentralised, pull-oriented Kanban produc-
changed while all other parameters are kept constant. tion control principles. Using a case and simulation
Table 9 shows in its left columns the percentage changes model-based approach, we investigated a ve-stage
of TPT and COFT, when the number of PKCs in only pharmaceutical production process with functional layout
one production stage is incremented by one compared to and addressed both generic and specic aspects of this
the optimal conguration. B + 1 indicate that the number internal supply chain. Based on a realistic and
1260 J. Strohhecker et al.

thoroughly tested DES model, we conducted a range of campaign system to real cases before implementation by
simulation experiments, including optimisation, that lead using a simulation model tailored to the specic system.
us to the following propositions: For companies in the process industry with long set-
up times that have so far refrained from introducing pull
principles, this simulation case study can be seen as an
encouragement to give up their reservations against Kan-
Proposition 1 ban. By introducing Kanban control in their supply
Decentralised Kanban production control and campaign chains, campaign-producing industries, such as pharma,
production are not contradictory. food, biotech, chemical and petrochemical, may also
Process industries often struggle with high set-up times benet from Kanban production control characteristics
relative to the process time. Campaigns are formed as a that could not be covered in the simulation study. Spear-
means to keep resource utilisation measures in acceptable man and Zazanis (1992) show that pull systems are
ranges. In such an environment, the usually required fast inherently easier to control, which in a dynamic envi-
set-ups in Kanban systems can be obviously not met. ronment leads to better adaptation to changes in
Therefore, Kanban is often prejudged as inadequate for demand or capacity. It also results in reduced planning
campaign production systems. However, our simulation efforts and because of the decentralised control mecha-
study shows that an integration of pull principles in a nism in more responsibility for staff and higher moti-
Downloaded by [Dicle University] at 04:52 05 November 2014

production system with high set-up times and campaign vation of the employees. Since the use of cards and
formation is feasible and can be implemented. boards makes Kanban production control more visible
and transparent, continuous improvement efforts are
facilitated (Li 2003). The importance of high-quality out-
put becomes evident for every worker when, due to
Proposition 2 defects, downstream processes are disrupted and products
Compared to centralised campaign planning systems that have to be returned (Ohno 1988). Such experiences
include safety margins before and after production, strengthen employees commitment to continuous
replenishment lead times and goods received processing improvement. Another major advantage of pull systems
times, Kanban campaign production systems show supe- is that these are much more responsive to problems than
rior performance. push systems. In a pull system, downstream processes
Simulation results for the Kanban campaign produc- are shut down when upstream processes are defective,
tion systems conceived for the internal pharmaceutical thus preventing the accumulation of inventory on the
supply chain in our case indicate that a broad range of plant oor (Cheng and Podolsky 1993).
congurations exist that show a lower TPT (and lower In addition to the limitation of not having included
inventory levels) for acceptable COFT levels of maxi- all potential advantages of a Kanban campaign produc-
mally 15 workdays. While we were not able to model tion system, our study focused on a very specic type of
the centralised planning and control system actually used pharmaceutical production. In order to generalise our
in the case (due to reasons of condentiality and, as a ndings, the performance of a more generic Kanban
consequence, a lack of information and data) and com- campaign production system should be explored in dif-
pare the simulation results for the two systems, we think ferent environments. For example, demand patterns could
that the comparison of simulated performance measures be varied, processing times could be changed and the
with actual benchmark gures provides at least some evi- ratio of set-up times and processing time could be
dence for the advantages of the Kanban campaign pro- shifted. Moreover, different layouts could be considered.
duction system. It has to be mentioned that by narrowing the purpose of
the model and not including various products in the pro-
duction stages with campaign formation, an advanced
analysis of, for instance, different rules of campaign for-
Proposition 3 mation is not possible. However, this limitation could be
Performance of the Kanban campaign production system easily addressed by extending and rening the DES
is relatively robust to changes in the conguration. model used in this study.
Aggregate and partial sensitivity analysis showed that A further limitation of our study is that we did not
the systems performance is less than proportionally include costs and revenues in our analysis. Although the
affected by changes in Kanban conguration parameters. performance indicators that we have used following
Therefore, it can be assumed, that the systems congu- Yavuz and atir (1995a), Najmi and Makui (2011) and
ration can be adapted to specic production situations others are clearly linked to prot-based performance
without completely deteriorating performance. Neverthe- indicators, we cannot claim that a Kanban campaign
less, it is suggested to test adaptations of the Kanban production system maximises prot when compared to a
Production Planning & Control 1261

traditional campaign production system. An answer to References


this question is beyond the scope of this paper and could
also be addressed by future research. It also has to be Abdulmalek, F. A., and J. Rajgopal. 2006. A Classication
Scheme for the Process Industry to Guide the Implementa-
noted that campaign formation helps to alleviate one of
tion of Lean. Engineering Management Journal 18:
the prerequisites low set-up times that are discussed 1525.
for successfully introducing Kanban control. Other limi- Allen, J., C. Robinson, and D. Stewart. 2001. Lean Manufac-
tations of the Kanban concept as, for instance, discussed turing: A Plant Floor Guide. Dearborn: Society of Manu-
by Ohno (1988) or Monden (1998), have not been facturing Engineers.
Askin, R. G., M. G. Mitwasi, and J. B. Goldberg. 1993.
addressed in this study and persist.
Determining the Number of Kanbans in Multi-item Just-
The concept of a Kanban-based campaign production in-time Systems. IIE Transactions 25: 8998.
system as presented and investigated in this study has Banks, J., J. S. Carson, Nelson B. L. Ii, and D. M. Nicol.
been proven by simulation experiments to be practical 2010. Discrete-event System Simulation. 5th ed. Upper Sad-
and perform well. Integrating campaign formation in the dle River, NJ: Pearson Education.
Baynat, B., J. A. Buzacott, and Y. Dallery. 2002. Multiproduct
Kanban control logic qualies as a promising way for Kanban-like Control Systems. International Journal of
industries with rather set-up intensive processes to take Production Research 40: 42254255.
advantage of modern pull-oriented production control. Baynat, B., Y. Dallery, M. D. Mascolo, and Y. Frein. 2001. A
Multi-class Approximation Technique for the Analysis of
Downloaded by [Dicle University] at 04:52 05 November 2014

Note Kanban-like Control Systems. International Journal of


Production Research 39: 307328.
1. SSE = sum of squared errors, KSp = KolmogorovSmirnov Berkley, B. J. 1992. A Review of the Kanban Production
signicance level Control Research Literature. Production and Operations
Management 1: 393412.
Notes on contributors Berkley, B. J. 1993. Simulation Tests of FCFS and SPT
Jrgen Strohhecker is a professor of Sequencing in Kanban Systems. Decision Sciences 24:
Operations Management and Management 218227.
Accounting at Frankfurt School of Finance & Billesbach, T. J. 1994. Applying Lean Production Principles
Management, Frankfurt/Main, Germany. He to a Process Facility. Production and Inventory Manage-
ment Journal 35: 4040.
holds a diploma and PhD from Mannheim
Bonvik, A. M., C. E. Couch, and S. B. Gershwin. 1997. A
University in Business Administration. His Comparison of Production-line Control Mechanisms.
research interests include production International Journal of Production Research 35:
planning and control concepts, system 789804.
dynamics, diffusion and network effects and Bonvik, A. M., Y. Dallery, and S. B. Gershwin. 2000.
the analysis of behavioural dynamic decision-making. Approximate Analysis of Production Systems Operated
by a CONWIP/nite Buffer Hybrid Control Policy.
International Journal of Production Research 38:
Rainer Sibbel is a professor of International 28452869.
Health Management at Frankfurt School of Bouchriha, H., M. Ouhimmou, and S. Damours. 2007. Lot
Finance & Management, Frankfurt/Main, Sizing Problem on a Paper Machine under a Cyclic Produc-
tion Approach. International Journal of Production Eco-
Germany. He holds a diploma in
nomics 105: 318328.
mathematics and a PhD from the University Cheng, T. C. E., and S. Podolsky. 1993. Just-in-time Manufac-
of Mnster in Business Administration. In turing: An Introduction. London: Chapman and Hall.
2004, he got his habilitation in economics Cochran, J. K. 1987. Techniques for Ascertaining the Validity
and business administration at University of of large-Scale Production Simulation Models. Interna-
Bayreuth. His research interests include tional Journal of Production Research 25: 233244.
service operations management, production planning and Dallery, Y., and G. Liberopoulos. 2000. Extended Kanban
control concepts and hospital management and controlling. Control System: Combining Kanban and Base Stock. IIE
Transactions 32: 369386.
Davis, W. J., and S. J. Stubitz. 1987. Conguring a Kanban
Marcel Dick works in the Mergers & System using a Discrete Optimization of Multiple Stochas-
Acquisitions team of the Corporate tic Responses. International Journal of Production
Research 25: 721740.
Development department at Credit Suisse,
Dennis, D., and J. Meredith. 2000. An Empirical Analysis of
Zurich, Switzerland, and is a research Process Industry Transformation Systems. Management
associate at Frankfurt School of Finance & Science 46: 10851099.
Management, Frankfurt/Main, Germany. He Fullerton, R. R., and C. S. Mcwatters. 2001. The Production
holds a Bachelor and Master degree from Performance Benets from JIT Implementation. Journal of
Frankfurt School of Finance & Management Operations Management 19: 8196.
in Business Administration and Banking. Gad, S. C. 2008. Pharmaceutical Manufacturing Handbook:
His research interests include system dynamics, production Production and Processes. Hoboken, NJ: Wiley-Inter-
planning, and decision-making in complex organisations. science.
1262 J. Strohhecker et al.

Graham, A. K., J. D. W. Morecroft, P. M. Senge, and J. D. Li, J.-W. 2003. Simulation-based Comparison of Push and
Sterman. 1992. Model-supported Case studies for Manage- Pull Systems in a Job-shop Environment Considering the
ment Education. European Journal of Operational Context of JIT Implementation. International Journal of
Research 59: 151166. Production Research 41: 427447.
Grunow, M., H. -O. Gnther, and M. Lehmann. 2002. Cam- Lyons, A. C., K. Vidamour, R. Jain, and M. Sutherland. 2011.
paign Planning for Multi-stage Batch Processes in the Developing an Understanding of Lean Thinking in Process
Chemical Industry. OR Spectrum 24: 281314. Industries. Production Planning & Control 24: 475494.
Gupta, Y. P., and M. C. Gupta. 1989. A System Dynamics Mensch, G. 1972. Trilemma der Ablaufplanung [Trilemma of
Model for a Multi-stage Multi-line Dual-card JIT-Kanban sequencing]. Zeitschrift fr Betriebswirtschaft 42: 7788.
System. International Journal of Production Research 27: Monden, Y. 1998. Toyota Production System: An Integrated
309352. Approach to Just-in-Time. 3rd ed. Norcross, GA: Engineer-
Gutenberg, E. 1983. Grundlagen der Betriebswirtschaftslehre. ing & Management Press.
Band I. Die Produktion [Principles of Business Administra- Mukhopadhyay, S. K., J. Dwivedy, and A. Kumar. 1998.
tion. Vol 1. Production]. 24th ed. Berlin: Springer. Design and Implementation of an Integrated Production
Harland, C. M. 1996. Supply Chain Management: Relation- Planning System for a Pharmaceutical Manufacturing Con-
ships, Chains and Networks. British Journal of Manage- cern in India. Production Planning & Control 9: 391402.
ment 7: S63S80. Najmi, A., and A. Makui. 2011. A Conceptual Model for
Higgins, P., P. Le Roy, and L. Tierney. 1996. Manufacturing Measuring Supply Chains Performance. Production Plan-
Planning and Control: Beyond MRP II. London: Chapman ning & Control 23: 694706.
& Hall. Naylor, T. H., and J. M. Finger. 1967. Verication of Com-
Hopp, W. J., and M. L. Spearman. 2004. To Pull or Not to
Downloaded by [Dicle University] at 04:52 05 November 2014

puter Simulation Models. Management Science 14: B-


Pull: What Is the Question? Manufacturing & Service 92B-101.
Operations Management 6: 133148. Neumann, K., C. Schwindt, and N. Trautmann. 2002.
Huang, P. Y., L. P. Rees, and B. W. Taylor. 1983. A Simula- Advanced Production Scheduling for Batch Plants in Pro-
tion Analysis of the Japanese Just-in-Time Technique (With cess Industries. OR Spectrum 24: 251279.
Kanbans) For a Multiline, Multistage Production System. Ohno, T. 1988. Toyota Production System: Beyond Large-scale
Decision Sciences 14: 326344. Production. Cambridge, MA: Productivity Press.
Huang, M., D. Wang, and W. H. Ip. 1998. A Simulation and Overfeld, J., and T. H. Witte. 1992. Lot-sizing for Time-
Comparative Study of the CONWIP, Kanban and MRP phased Production. International Journal of Production
Production Control Systems in a Cold Rolling Plant. Pro- Research 30: 839858.
duction Planning & Control 9: 803812. Papageorgiou, L. G., and C. C. Pantelides. 1996a. Optimal
Jans, R., and Z. Degraeve. 2008. Modeling Industrial Lot Siz- Campaign Planning/Scheduling of Multipurpose Batch/
ing Problems: A Review. International Journal of Produc- Semicontinuous Plants 2. A Mathematical Decomposition
tion Research 46: 16191643. Approach. Industrial & Engineering Chemistry Research
Kallrath, J. 2002. Planning and Scheduling in the Process 35: 510529.
Industry. OR Spectrum 24: 219250. Papageorgiou, L. G., and C. C. Pantelides. 1996b. Optimal
Kelton, W. D., R. P. Sadowski, and N. B. Swets. 2010. Simula- Campaign Planning/Scheduling of Multipurpose Batch/
tion with Arena. 5th ed. Boston, MA: McGraw-Hill. Semicontinuous Plants. 1. Mathematical Formulation.
Khojasteh-Ghamari, Y. 2009. A Performance Comparison Industrial & Engineering Chemistry Research 35: 488509.
Between Kanban and CONWIP Controlled Assembly Pinedo, M. 2005. Planning and Scheduling in Manufacturing
Systems. Journal of Intelligent Manufacturing 20: and Services. New York: Springer and Business Media.
751760. Price, W., M. Gravel, and A. L. Nsakanda. 1994. A Review
Kondili, E., C. C. Pantelides, and R. W. H. Sargent. 1993. A of Optimisation Models of Kanban-based Production Sys-
General Algorithm for Short-term Scheduling of Batch tems. European Journal of Operational Research 75:
Operations I. MILP Formulation. Computers & Chemical 112.
Engineering 17: 211227. Rockart, S. 2001. Calling Spirits From the Deep: Competing
Kouri, I. A., T. J. Salmimaa, and I. H. Vilpola. 2008. The for and Through Sophisticated Customers. Cambridge,
Principles and Planning Process Of an Electronic Kanban MA: Massachusetts Institute of Technology.
System. In Novel Algorithms and Techniques In Telecom- Rother, M., and J. Shook. 2003. Learning to See: Value Stream
munications, Automation and Industrial Electronics, edited Mapping to Add Value and Eliminate Muda. Version 1.3.
by T. Sobh, K. Elleithy, A. Mahmood, and M. A. Karim, Cambridge, MA: Lean Enterprise Institute.
99104. Dordrecht: Springer. Schonberger, R. J. 1982. Japanese Manufacturing Techniques:
Krieg, G. N., and H. Kuhn. 2004. Analysis of Multi- Nine Hidden Lessons in Simplicity. New York: Free
product Kanban Systems with State-dependent Setups Press.
and Lost Sales. Annals of Operations Research 125: Schonberger, R. J. 2007. Japanese Production Management:
141166. An Evolution With Mixed Success. Journal of Opera-
Krieg, G. N., and H. Kuhn. 2008. Performance Evaluation of tions Management 25: 403419.
Two-stage Multi-product Kanban Systems. IIE Transac- Sendil Kumar, C., and R. Panneerselvam. 2007. Literature
tions 40: 265283. Review of JIT-KANBAN System. The International Jour-
Laguna, M. 2011. OptQuest - Optimization of Complex Systems nal of Advanced Manufacturing Technology 32: 393408.
(Decisioneering). Boulder, CO: OptTek Systems. Sengupta, S., F. Sharief, and S. P. Dutta. 1999. Determination
Lee-Mortimer, A. 2008. A Continuing Lean Journey: An Elec- of the Optimal Number of Kanbans and Kanban Allocation
tronic Manufacturers Adopting of Kanban. Assembly in a FMS: A Simulation Based Study. Production
Automation 28: 103112. Planning & Control 10: 439447.
Production Planning & Control 1263

Seth, D., and V. Gupta. 2005. Application of Value Stream Sugimori, Y., K. Kusunoki, F. Cho, and S. Uchikawa. 1977.
Mapping for Lean Operations and Cycle Time Reduction: Toyota Production System and Kanban System Materiali-
An Indian Case Study. Production Planning & Control zation of Just-in-time and Respect-for-human System.
16: 4459. International Journal of Production Research 15: 553564.
Simchi-Levi, D., P. Kaminsky, and E. Simchi-Levi. 2008. Susarla, N., and I. A. Karimi. 2011. Integrated Campaign
Designing and Managing the Supply Chain: Concepts, Planning and Resource Allocation in Batch Plants. Com-
Strategies, and Case Studies. 3rd ed. Boston, MA: puters & Chemical Engineering 35: 29903001.
McGraw-Hill/Irwin. Treadwell, M. A. and J. W. Herrmann. 2005. A Kanban Mod-
Sita, B. 1998. Simulation Analysis of a Manufacturing Supply ule for Simulating Pull Production in ARENA. Proceed-
Chain. Decision Sciences 29: 633657. ings of the 37th conference on Winter simulation. Orlando,
Spearman, M. L., and M. A. Zazanis. 1992. Push and Pull Florida: Winter Simulation Conference, 14131417.
Production Systems: Issues and Comparisons. Operations White, R. E., J. N. Pearson, and J. R. Wilson. 1999. JIT
Research 40: 521532. Manufacturing: A Survey of Implementations in Small
Stadtler, H., and C. Kilger. 2005. Supply Chain Management and Large U.S. Manufacturers. Management Science 45:
and Advanced Planning: Concepts, Models, Software and 115.
Case Studies. 3rd ed. Berlin: Springer. Yang, K. K. 2000. Managing a Flow Line with Single-Kan-
Steger-Jensen, K., H.-H. Hvolby, P. Nielsen, and I. Nielsen. ban, Dual-Kanban or Conwip. Production and Operations
2011. Advanced Planning and Scheduling Technology. Management 9: 349366.
Production Planning & Control 22: 800808. Yavuz, I. H., and A. atir. 1995a. Kanban-based Operational
Sterman, J. D. 2000. Business Dynamics: Systems Thinking and Planning and Control: Simulation Modelling. Production
Downloaded by [Dicle University] at 04:52 05 November 2014

Modeling for a Complex World. Boston, MA: Irwin/ Planning & Control 6: 331344.
McGraw-Hill. Yavuz, I. H., and A. atir. 1995b. A Kanban-based Simulation
Suerie, C. 2005. Campaign Planning in Time-indexed Model Study of a Mixed Model Just-in-Time Manufacturing
Formulations. International Journal of Production Line. International Journal of Production Research 6:
Research 43: 4966. 331344.

Anda mungkin juga menyukai