The Supply Net Game
Bernd Scholz-Reiter
Department of Planning and Control of Production Systems
University of Bremen, BIBA-IPS, Hochschulring 20, 28359 Bremen, Germany
Tel.: 49 421 218 5576, Fax: 49 421 218 5640, email: bsr@biba.uni-bremen.de
Salima Delhoum
Department of Planning and Control of Production Systems
University of Bremen, BIBA-IPS, Hochschulring 20, 28359 Bremen, Germany
Tel.: 49 421 218 5549, Fax: 49 421 218 5640, email: del@biba.uni-bremen.de
Abstract
The paper describes a game called, the supply net game, built around the structure of a
production supply network based on the “ anchoring and adjustment heuristic” which is
known as the one people use to make inferences about uncertain events. The game
involves four players where everyone manages his manufacturing unit that consists of
four production lines which proceed to the joint development of products with the other
units. While planning production and controlling inventories, every person should try to
minimize the costs caused by both holding items on stock and being in an out-of stock
situation. The paper stresses the valuable impact and contribution of management
games for management and engineering education in general and particularly the
significance of learning implicit skills as well as gaining insight in inventory control
and management of complex distributed production systems such as the system
dynamics production network model introduced and analyzed in the paper. The game
will be used later in a controlled experiment which is not under the scope of this paper.
Key Words
Management game, production network, system dynamics
Introduction
In the system dynamics literature, the value of management flight simulators, also
known as microworlds, was repeatedly demonstrated (Senge 1990; Bakken et al. 1992;
Sterman 1994; Sterman 2000). Management flight simulators are learning processes /
tools or environments that help activate double loop learning (Bakken et al. 1992)
because when the user “flies” a simulator, time and space are compressed, and hence, he
can deduct interactively the feedback structures that exist within the system (Senge
1990, 312-338; Graham et al. 1992; Senge and Sterman 1992). The advantage of these
learning environments, therefore, is that they make cause and effect relationships more
visible to the user. Simulators enable accelerated learning, what Probst and Biichel
(1994, 95) call “learning by doing”, and Senge (1990, 313) “learning through doing”.
The simulator is a learning laboratory that stimulates risk-free try outs of strategies,
illustrates the relationship of system structure to behavior, and portrays learning
according to the scientific method where the objective is to understand the problem
genesis and dynamic behavior of complex systems in order to provide sustainable
policies. These mediums are necessary because people are bounded rationally (Simon
1982) and have “misperceptions of feedback” (Sterman 1989a, Sterman 1989b; Sterman
1994) which steadily affect their reasoning and interpretation capabilities in complex
systems in presence of feedbacks, time delays and nonlinearities (Sterman 1989a;
Sterman 1989b; Bakken 1993; Paich and Sterman 1993; Diehl and Sterman 1995;
Larsen et al. 1999; Langley and Morecroft 2004). Sterman (1989a) indicated that the
mental models people use to decide are deficient because of their open-loop structure
(see Senge et al. 1994, for a definition of mental models); furthermore, the
misperceptions of feedback generate the underestimation of the supply line in the beer
distribution game that induce the wild oscillations of inventories. Moreover Miller
(1956) proved through several experiments that human capacity to process information,
the so-called channel capacity, for one dimensional stimuli, is limited to the number
seven (bits). Miller (1956) gave the range seven plus or minus two as the interval that
includes the capacities observed in laboratories.
Because all of the shortcomings of mental models, information processing, memory,
etc. the importance of management flight simulators is corroborated although the
performance of subjects who used them in experiments did not notably improve
(Sterman 1989b; Bakken 1993; Paich and Sterman 1993; Langley and Morecroft 2004).
Some work point to their limitations when it comes to learning transfer issues to real
world settings (Bakken et al. 1992), the increase of organizational performance (Vennix
1999), and the issues of simulators as teaching and research instruments (GréBler 2004).
From a methodological point of view the question is more how to embed management
games successfully in learning laboratories (Graham and Senge 1990; Graham et al.
1992; Warren and Langley 1999) as part of a systems thinking intervention with
problem conceptualization, model formulation, and hypothesis test phases that help
develop systems thinking skills and learning transfer frameworks (see Maier and
Gr6éBler 2000, for a classification of games, management flight simulators, etc.).
It is not clear from the literature if a mental model elicitation method (ladder of
inference, left-hand column method, etc.) embedded together with a gaming
environment in a controlled experiment could have an effect on the quality of the
decisions generated by the subjects especially for the deterministic task of inventory
management — bullwhip effect related situations - in production networks. The paper
introduces and analyzes the supply net game that is built for its future use in such an
experiment. The results of the experiment will be released in a prospective publication.
Management G ames of Production Systems
Management games are widely employed to train managers and workers and are also a
fundamental trend in management and engineering education. Warren and Langley
(1999) underscore that managers should have access to gaming simulation tools in order
for them to cope with the business systems in which they evolve and to reap strategic
management skills. Scholz-Reiter et al. (2002b) emphasize the need to introduce
management games to workers and engineering students to learn the task of inventory
management and aptitudes like communication and cooperation in complex distributed
production systems such as production networks. Maier and Gr6Bler (2000) distinguish
between single-user and multi-user games whereby the former is labeled as “simulator”
and the latter as “planning game”. This separation is explained by the implications of
group size dynamics on learning.
The literature on management games is replete with applications such as people express
management flight simulator (Sterman 1988); multiplier accelerator (Sterman 1989b);
boom and bust (Paich and Sterman 1993); web-based beer game (Oliva and Gongalves
2005) complement to the board game (Sterman 1989a), etc. (see also Maier and GréBler
2000, for a list of games). Since the paper captures the instance of a production system,
the attention is focused on this kind of systems for which there are some board
implementations like the beer game that illustrates the four sectors linear supply chain
(Sterman 1989a) and the dice game (Lange and Ziegenbein 2005) that portrays
Goldratt’s theory of constraints for capacity management problems. System dynamics
models include the simple inventory management task (Diehl and Sterman 1995, Aybat
et al. 2004). Games contingent on methodologies other than system dynamics exist for
distributed production systems; Cosiga, Glotrain, and Share game to name a few.
Cosiga is an internet-based game sponsored by the European Commission that aims to
train European engineers on the principles of concurrent engineering as well as product
development and makes the case of a truck manufacturer. Five players coordinate their
efforts for the specification, design and manufacture of a product around the game’s
platform that uses communication means intensively (chat, email, and video conference)
because cooperation is prominent (Scholz-Reiter et al. 2002a; Cosiga 2003). The project
was the essence of a joint course at the University of Bremen, Nottingham and
Trondheim in Germany, United Kingdom and Norway respectively. Glotrain, developed
at the BIBA institute of the University of Bremen, was devised to let users learn implicit
skills in distributed production systems with the help of modern telecommunication
technologies (Glotrain 1997; Windhoff 2001). Analog to Glotrain, Share game was born
in BIBA; the difference between the two games lies, in that; the latter encourages
several processes of inter-organizational learning, product development and
collaboration simultaneously. Share game prevails on a production network model, as
well as Cosiga, and it is dedicated to the product development of a jet-ski and cell phone
and based on concurrent engineering concepts. Both scenarios (one for every product)
entail nine persons — three per organization with different hierarchical levels from
employee to department head. The game is meant to reinforce trust, collaboration and
teamwork in and across teams whereby communication is primordial and is achieved
via telephone and message boards that go down sometimes during the game to simulate
remote participants in disparate regions (for a detailed description of the game see
Baalsrud Hauge et al. 2005).
The Supply Net game is built upon a pull production and logistic supply network and
has, as a replenishment procedure, the “anchoring and adjustment heuristic” described
by Tversky and Kahneman (1974) as one of the heuristics people utilize to make
inferences about uncertain events. Each of the four players manages the stock levels
encompassed in his manufacturing unit by ordering for the four production lines
consecutively in the same sequence from line one to four. At the same time, he tries to
minimize the costs incurred. The minimization of costs corresponds to bringing the
bullwhip effect to its smallest expression since the effect, defined as demand variability
over a supply chain generates stacked backlogs when demand booms and huge
inventories when orders fall; the two situations for which costs are high. It is
hypothesized that the two types of costs, (a) € 0.5 per product on hand per minute and
(b) € 1.0 for out-of-stock cost, are the same for all manufacturers. The game includes
four organizations; nevertheless it is limited to four participants, which is better than the
nine (five) participants, for the three companies, that the Share game (Cosiga) calls for.
Indeed Share game’s greediness on personnel and subsequent infrastructure (computers,
phones and rooms) make it almost impracticable in small and medium enterprises
(Baalsrud Hauge et al. 2005), which are assumed to be the major beneficiaries of such
tools. From another side when it is about the choice of a modeling methodology for the
supply net game, system dynamics is adopted because it possesses a long tradition of
contribution to the refinement of individual and organizational learning (see Morecroft
and Sterman 1994, for case studies). Finally, the application is more than a mere gaming
environment because it is designed to be part of a systems thinking intervention under
the framework of a controlled experiment with performance measurements of learning.
Modeling of Production Supply Networks: A Summary Review of the Literature
In a production and logistic supply net many manufacturing units integrate their
activities and processes to satisfy customer demand which may be a supply chain /
network or an individual company (retailer). The factory takes advantage in partnering
on the network because it can then respond to external fluctuations either of customer
demand or supply variations as a network which helps amortize the shocks. However,
when the internal complex dynamics of the network, such as production specifications
and logistic channels, favor the external noise signal — from one or both end-positions -
the amplification will gain in intensity, which will hit back the individual unit even
stronger. Although a shared definition of the production network or production in
networks does not exist yet, it is accepted that such a network entails the distributed
joint development of the product and is regarded as a new form of cooperation between
manufacturers over a long period of time (Wiendahl and Lutz 2002). This contrasts with
the variable production network (VPN) whose structure lasts for the duration of a
project (Wiendahl et al. 1998). The VPN is not under the scope of the paper.
Production planning and control in a production net is so that there is a general
integrated planning for the whole network and an individual planning carried out by
every manufacturer separately in regard to his production and assembly processes.
Wiendahl and Lutz (2002) proposed the application of (a) load-oriented or order release
control methods and (b) decentralized control loops. Both approaches suggest
decentralization, but whereas the former consists of the placement of an order only
when the production system can handle it, and therefore, try to eliminate unwanted
behaviors occasioned by the bullwhip effect; the latter will process a job only if the next
workstation is able to complete its operation. Other methodologies for modeling
production nets employ open queuing network methods where a node represents a
processing machine and the arc the logistic channel to the next processing station and
the objective is to minimize the production lead time which is given by the length of the
longest path in the network. Azaron et al. (2005) develop an open queuing net of a
dynamic multi-stage assembly system for which the lead time and operating costs are
minimized through a variant of multi-objective programming; the goal attainment
method. In this open network the arc lengths and processing times are stochastic and the
model is discretized. Hieber (1998) illustrated a general diagnosis technique for the
optimization of production nets. In addition to optimization, algorithms based on the
simulated annealing heuristic are applied to determine an “optimal” path through the
manufacturing net for an incoming order (Azevedo and Sousa 2000). Software agent
technologies also consider the harmonization of the production agenda of the single
manufacturing unit with the supply requests within the network (Dangelmeier et al.
2001; Neubert et al. 2004).
Supply Net Game: the Case Study
The supply production network (Figure 1) is made up of four factories Fi, i= 1..4 where
each one is constituted of four parallel production lines Lij, j = 1..4, and every line has a
work-in-process (WIP) that stores the products before production and a stock of
manufactured lots. Every WIP has a minimum (null) and no designated maximum
value. The supply of raw materials for the WIP is assumed to be unlimited in capacity
and it is done from outside the network to the production line Lii (i = 1..4) that
manufactures the finished items Pii which are then delivered outside the network to the
customer (in this case, the distributor). Lij (i # j) takes the quantity it needs from Lii to
produce Pij, in that; Lii is considered as the link, of the lines Lij, to the outside
network(s) through which the supply is fueled.
The procedure in the net is so that the customer makes his orders of products Pij (j =
1.4) — product j manufactured by Fi - to line Lii of factory Fi (i = 1.4). The
manufacturing line Lii passes the ordering information to the other lines Lij (i # j)
which start the production of Pij. The semi-finished parts Pij, j = 1..4 (j # 1) are
circulated between the two factories: Fi (Fj) ships the part Pij (Pji) to Fj (Fi).
Furthermore, the line Lii keeps track of the parameter-production matrix and, hence,
governs the internal coupling (in the same factory) of the production lines. The more
couplings in the system, more dynamic and complex the behavior is. In this instance,
only lines L12 and L14 are coupled; in other words, the processing of one product P12
necessitates one product P14 — in addition to one item P21 since all lines are externally
(with other factories) coupled. The minimum lead time on Lij is denoted Tij which
includes the cycle time and the transport time from Lij (Fi) to Lji (Fj). Tij is constant
and the machine capacity is supposed unrestricted.
The case explores a simplified theoretical instance of a production net, yet it exhibits
complex non linear dynamics. The choice is on four factories because that renders the
behavior of the net more challenging, than with two factories, in terms of the feedbacks
and nonlinearities created by the production processes, and logistic channels. Although
the model is theoretical, it exists in practice networks of production where the product
starts from a processing machine al, visits another machine a2 and then returns to al.
Some real-life settings are the semiconductor industry for example. The supply chain
industries that could benefit from the game are those where the oscillations of orders
and inventories are strong such as the semiconductor & high-tech, commodities (beer,
pampers...), automobile, aviation, chemicals or shipping & distribution.
Supply on LI Se Delivery ‘a ioe of
Pil ==
P12 P12
Pi3
Pl4 P14
Factory F2 Piaf Factory F4
(L2j, T2j) (L4j, T4))
P21 P41
P21 P41
P22 —>- >! P42
P23 P24 D,
ag pas
P24 |< P44 —»
P31
P23
Factory F3 P43
(L3j, T3))
P31 —
Paz
P33—*
P32 P34 P34
Figure 1: General Structure of the Production Network
Pij is the product j (j=1..4) manufactured by factory Fi (i=1..4), Pij and Pji (i #j) are exchanged between factories Fi
and Fj and then delivered outside the net whereas Pii is delivered outside the network only. In F1 the lines L12 and
L14 are coupled. The same supply and delivery structure is true for all Fi. Lij is the production line for Pij and Tij the
min lead time on Lij.
Structure of the Supply Net Model
The model proposes a continuous modeling approach for the supply production network
contingent on the continuous flow and time of the system dynamics methodology
because despite the different existing products, the focus is not on the individual item
type which is regarded as an aggregate product, but rather on the dynamics created by
such systems. Sterman (2000, 208) indicated that the error generated by the
approximation of a discrete event into a continuous flow is negligible in comparison to
the error in model constant measurements and hence, promoted the use of the
continuous approach as long as the purpose of the model could be met. Scholz-Reiter et
al. (2005b) suggested the use of both continuous and discrete modeling approaches
because the latter allows a real description of the manufacturing system, but it is
demanding in programming whereas system dynamics continuous simulation facilitates
the implementation of the control strategy, however, at a higher level of aggregation
(see also Scholz-Reiter et al. 2005a).
The model employs the “anchoring and adjustment heuristic” as the replenishment
procedure because the approach follows the descriptive research on the bullwhip effect
in supply chains, for which the decision maker is considered as “bounded rationally”.
Tversky and Kahneman (1974) described the heuristic as being among those people
utilize to make inferences about uncertain events. Sterman (1989a) used it to model the
decision making processes in the beer game supply chain model, and found that the
heuristic imitated correctly the decisions of the actual players of the board game
according to the statistical results of the regression models.
Figure 2 illustrates the production line L11 of factory Fl which processes product P11
and then ships it to the customer. The line L11 also orders and receives the required
supplies in raw materials, for the production of P11, P12, P13 and P14, from outside the
network according to the demand expressed by the respective lines. Figure 3 shows how
line L13 is linked to L31 based on the principle of the joint development of products
between the factories of the net so that items P13 and P31 can then be delivered to the
customer.
The other two lines of factory Fl, L12 and L14, have the same structure as that of
Figure 3 with one additional characteristic of coupling. Besides the models in Figure 2
and Figure 3 are representative of the procedure applied to the other factories.
The idea behind the model for line L11 is that there is a customer who sees how much
of products P11, P12, P13 and P14 he has on stock so that he can place his orders to
L11. The order function for P11 (order rate distributor for P11) is supposed to follow a
random normal distribution. From last period demand, a smoothed expectation is
derived; in addition, the adjustments of the stock and WIP are computed based on their
desired levels. The summation of these three values gives the order rate for P11 (order
rate P11) in accordance with the anchoring and adjustment heuristic. The order is
transmitted to an external supplier with assumingly unrestricted capacity, and therefore,
P11 is produced and shipped to the customer. Backlogs of unsatisfied demand are taken
into consideration and answered first when stocks permit it.
The order rate for article P13 (order rate P13) is calculated with the expectation of the
customer’s demand for P13 — demand for P13 equals the order for P11 - (order rate
distributor for P11), and the adjustments of the stock and WIP of P13 with one variant;
the expectation of last period order rate of items P31 is also considered (Appendix A).
Indeed the stock of P13 should be ample enough to satisfy not only the customer
request, but also the demand of the line with which it jointly produces, in this case, L31.
The order rate of products P13 is placed to the line P31 which sends them when
available; otherwise they go to the backlog. Next to the acquisition, the processing line
L13 manufactures the lot and ships (a) the order of articles P31 to L31 (shipping rate
P13) and (b) the quantity of products required by the customer (ship dist P1311). The
latter is delivered out of the net, so in order to avoid that its stock falls to zero; L13
places an order for raw materials (supply P13 from L11) to the external supplier via line
Lil.
10
Factory 14F1) production planning
FecioyFi
oul ofsiozk ——tmaut ofstock aaa
costPtt saat
= cr flfilment
ardorrato P13 état order rate areal \
\ cistoutorfor Pit> TP
Oo = target delivery desired ship
delet rate v
Enemy 4 transoor
wines |
range pang cost \) time?"
\ ~Latie F 3
supply P13,
catartrgy OVE
y
all
<orderrate P12>
sek ptt siotutr
ee ‘rodution shining ~Latie F acquiston rae
preducion start __, taePtt raerth astPtt
rater nae \
<cupaly P13 wed =min lead —tm emnected P11 4 <desited ship
tee, time Pri> tin OPT Morsnip iat
{ oth orn
<supply P12 ___<0F table
fomLtt= /
sdustrent
sum 14 AIT a ee ee
araecie i isrea me Socket Pits coverage P11
Pit x
J \ oda VN wile gue
destied acqulsiion
VP achusimant one aca coverage P11 stir fori
fms
E timetoadust ested stock if
Tileatea Sraers ee ak et
ae BE /
| xz oo:
change'n 8
orders PI
X
tmeto average
order rate P11
Figure 2: System Dynamics Model of the Supply Production Line P11
On the other side the order rate for items P31 (order rate P31) is set without the
recourse to the order rate for P13 (order rate P13), i.e. only customer demand and the
adjustments for both stock and WIP suffice, since the structure does not have a linear
topology, but rather a backflow (Appendix B). Otherwise there would be a redundancy
of computations.
11
—_ out ot ue
stock cost Pout of stock delivery delivery delay const wii ane RY delivery delay
dite 7” ean tate coal, mee a ttc Saat a‘
5 aa
z aa oer dame se
oieae atin erent sion ed
<order 13 ge ty p43 tate P13 dist P13 tate dist P13 order rate tee se to ey mt
Paty a P31 rate P34 r tate dist P31
target delivery desired ship Heid iy ate T coider tate x,
<supply P14 <production <produat delay P13 ——— rate P13 st P19 pias tatget delivery Saran 7 esired ship
TOs eee ee cihip dit POU delay P31 —r ate POA target dalivey_gitt® dist P31
iy min lead. inventory —Preost P43 target delivery rin lead <inventory —Brcoat P21 delay dist POA
ees cor
Sie ta alana feng ars fe PO stan so ccatet
[SPH PAO eae i stoacr f on
bond oduction naa |
tom muti B88 esp ners
= “tne agar —
a fd ship max ship rate .
pply P1d> wy daPtant vty a. Ptg> “supply P34 3
aap wap pas or
in lead time—tmexpected PAZ time PAt> adjustment
nl pected PA fate B49 OTT OFRPA rae lly RR: 2 Wepe
gta Pa2> expected P31 ted ship
adjustment adjustmentfor<minlead S3¥# order tate ‘min OPT tate P
ieee ey bree Fat hay
saatens \ ees su
Os he ae SO tock P31
FE swe Sautnen \ Sats wate Wi sdjiinet deed iP sey sto
time P13 3h time P31 P34 coverage
fina aed ested sto
iil ores marie deed crane Pot
saris
fine faut
desired acquistion — -Saedeg esived acquis seaeat tested i i
tae PI, rate P31
i
conte
Cr ia) ~
i Ais change in exp change in exp
tne rage oaegages aaemast pot
tists sia leas
anernate deet—— ebutor tor P41 =
Figure 3: System Dynamics Model of the Production Lines L13 and L31
12
The order fulfillment ratio function OFR P13 makes sure that the stock of products P13
will not turn negative because of the two outflows. It is expressed as follows:
OFR P13 = OF table (MIN (max ship rate dist P13, max ship rate P13) / (desired ship
rate dist P13+desired ship rate P13))
Where OF table is the same function for the whole production model and it is designed
to reach quickly unity due to the topology of the system.
Model Analysis
The data used in this section come from the simulation of the production net and not
from the experiment. Since the model is theoretical, a reference mode does not exist to
be checked for behavior reproduction which occurs sometimes in system dynamics.
Figure 5 shows the response of both stocks and WIP of lines L13 and L31 to an
unanticipated 100% step increase in the exogenous customer order function for items
P31 (denoted as order P31 in Figure 3). The original customer order is 12 products per
minute and augments to 24 products / minute at time 300 (Figure 4).
50
25
10
0 50 100 150 200 250 300 350 400 450 500
Time (Minute)
Order P31: current Product / Minute
Figure 4: Step Increase in Customer Demand for Product P31
The stocks and WIP for both items P13 and P31 rise immediately after the step increase
in demand at time 300 minutes (Figure 5). After the step increase the desired stock for
product P31 almost doubles to reach 192 products / minute from the initial value of 96
13
products per minute (Figure 6). On the other hand, the stock P13 overshoots the desired
stock P13 and the former is ahead of the latter by a distinct phase lag of ten minutes
which is the value of the coverage time for stock P13 (desired stock coverage P13). The
stock P13 and desired stock P13 exhibit wilder oscillations than the stock P31 and
desired stock P31. Figure 6 also manifests the amplifications of stock P13 to stock P31
which amount to 350%. It is important to remember that the customer demand of P13
(order rate distributor for P11) is a random normal distribution (Figure 7) at the
contrary of customer demand for P31 which is the linear step input in Figure 4.
1,000
0 50 100 150 200 250 300 350 400 450 500
Time (Minute)
Stock P13: current Product
Stock P31: current Product
WIP P13: current Product
WIP P31: current Product
Figure 5: Response of the Stocks and WIP of Lines L13 and L31 to the Step Increase
1,000
0 50 100 150 200 250 300 350 400 450 500
Time (Minute)
Desired stock P31: current a Product
Stock P31: current Product
Desired stock P13: current Product
Stock P13: current Product
Figure 6: Response of the Stocks and Desired Stocks of Lines L13 and L31 to the
Step Increase
14
10
5
0
0 50 100 150 200 250 300 350 400 450 500
Time (Minute)
Order rate distributor for P11: current ————_ Product / Minute
Figure 7: Customer Order for product P13
The customer order rate for product P13 is assumed to be the same as that for product P11, which is a random normal
variable: order rate distributor for P11 = RANDOM NORMAL (5, 10, 8, 1, 1). Random Normal is a normal
distribution in Vensim with the parameters: min, max, mean, standard deviation, seed.
Therefore, the order rate for items P13 and P31 look like Figure 8 and Figure 9
respectively.
100
seen
50 100 150 200 250 300 350 400 450 500
Time (Minute)
°
Order rate P13: current —_—_——— ProductMinute
Figure 8: Order Rate P13
When the coupling in the production system is more complex, which means that the
production processes (production matrix specifications) and subsequently the logistic
channels more intricate, then inventory oscillations will increase. Further simulations
demonstrate that when the supply is continuous in time and the supply time increases
15
then the amplitude of oscillations will decrease because the stock is adjusted more
smoothly. It is important to note that when the supply and order are both discrete in time
and the supply time increases, which means that the reacquisition time (time between
the order is placed and the supply is made) will also increase, and lead to larger
inventory oscillations because the manufacturer will have to build up a larger safety
stock which is one of the triggers of the bullwhip effect.
50
25
0
0 50 100 150 200 250 300 350 400 450 500
Time (Minute)
Order rate P31: current Product / Minute
Figure 9: Order Rate P31
Supply Net Game
On the basis of the production network case modeled and simulated with Vensim DSS
32 version 5.4a, the supply net game is designed and played in teams of four persons.
Every player is assigned to the inventory management of one factory, which means that
he orders, in every simulation period, for the replenishment of the stocks by making
four decisions on the amounts of products he needs for the four production lines under
his control. For example in factory F1, the person fixes the values of order rate P11,
order rate P12, order rate P13, and order rate P14 in this sequence. Then the next
player, who can - like the others - readily see the levels of stocks and WIP of the rest of
the team, decides on the quantities of items to request for every one of the products that
his factory manufactures. Likewise the two remaining players repeat the procedure.
While regulating his inventories, the player should care about not letting them drop too
much because there is an out-of-stock penalty of € 1.0 per item per minute and at the
16
same time, impeding the build up of stocks because of the € 0.5 cost for each product on
hold. Indeed the objective of the game is for the team to perform the task of inventory
management and to pursue the minimization of the total cost. In the fulfillment of the
task the players are not allowed to communicate although production nets encourage
cooperation and collaboration in teams.
ffiime (Minute) backlog distr P13 shipping rate P30 Stocks 13
288
26.17
30.99
33.76
36.00
39.84
43.19
44.20
eh 0 100 200 300 400 500
40.73 "
37.90 Time (Minute)
36.04
34.99 Stock P13 : current Product
35.85 Stock P31 : current ————. Product
3.29
33.44
34.36
Figure 10: Interface of the Supply Net Game for the Player Responsible for F1
Figure 10 displays the interface for the player in charge of the replenishment of the
stocks that pertain to factory Fl, namely those for articles P11, P12, P13 and P14. The
subject sets the levels of the four order rates and has complete information about the
values of his stocks, WIP and backlogs as well as those of the entire network. In
17
addition he has knowledge of the incoming orders to F1 (from the customer and the
factories) and the shipments sent to Fl from the external supplier and the factories
within the net. In order to enable the multi-player interactive simulation game, an
appropriate graphical user interface will be utilized. In addition there are thirty
simulation periods and the player faces no time pressure.
Discussion and Directions for Future Work
Management games are fun, attractive and entertaining. They provide a diverting
atmosphere that tends to break the daily routine of rigorous work for students and
professionals alike. However, reports about their effectiveness are mixed (Graham et al.
1992). Some games are so intricate that they are played only once, others bore their
users after they have played sometime. Gaming environments generally lack appropriate
measurement methods in regard to learning purposes. In spite of these limitations they
are being more and more present in the formal curriculum of engineering and
management education as innovative forms of teaching that promote accelerated
learning.
The paper introduces the supply net game that describes a potential distributed
production environment since the game is characterized by the joint-production
development between manufacturers analogue to distributed production systems. The
supply net game could be utilized to learn the integration and coordination of the
general net planning function to the individual member manufacturing plans and
processes in the production net; and therefore, the game promotes the acquisition of
implicit skills by managers and students. This is thought to represent a source of interest
for schedulers in particular and decision makers in general as well as students to “fly”
the supply net game. Although it is based on a theoretical model, there are some real-
life settings for such production nets especially in the semiconductor industry. Some
industries that could potentially benefit from the use of the game, in respect to those
where order and inventory oscillations occur are semiconductor & high-tech,
commodities (beer, pampers...), automobile, aviation, chemicals or shipping &
distribution.
18
The supply net game is thought and developed in order to be part of a systems thinking
intervention in a controlled experiment with subjects randomly assigned to one of two
groups. The control group will play the game only whereas the treatment group will
have the opportunity to experiment with the elicitation of the mental models of his
members with the resort to elicitation methods in order to confirm / reject the
hypothesis that stem from the literature. Furthermore, the protocol stresses the
performance measurement of learning both within the game, in terms of the cost
minimization function, and for learning transfer skills from the “virtual” world to the
work place. To do this the subjects will be asked to play a different game with the same
issue of inventory management. The supply net game in its actual format does not
support any form of communication or cooperation between the players whereas most
supply production nets emphasize collaboration. This shortcoming may be relaxed in
future work that will include the results of playing the game in teams.
Acknowledgment
The authors thank the anonymous reviewers for their comments on the paper. The
second author acknowledges the financial support of the DAAD through project
A/04/31510.
References
Aybat, S., S. Daysal, B. Tan, F. Topaloglu 2004. Decision making with different
variations of the stock management game. Proceedings of the International System
Dynamics Conference, July 25-29, Oxford, England.
Azaron, A., H. Katagiri, K. Kato, and M. Sakawa 2005. Modelling complex assemblies
as a queuing network for lead time control. In Press, Corrected Proof, European J ournal
of Operational Research, Available online 10 March 2005.
Azevedo, A.L. and J.P. Sousa 2000. Order planning for networked make-order
enterprises-a case study. Journal of the Operational Research Society 51(10): 1116-
1127.
Baalsrud Hauge, J., M. Schwesig, K.-D. Thoben, J. Eschenbacher 2005. Business games
—an effective tool for experiencing collaboration in production networks. Proceedings
of the 9" International Workshop on Experimental Interactive Learning in Industrial
19
Management, New Approaches on Learning, Studying and Teaching, 5-7 June, Espoo,
Finland, available from the Web: http://www.simlab.tkk.fi/events/hauge.pdf.
Bakken, B.E., J. Gould, D. Kim 1992. Experimentation in learning organizations: A
management flight simulator approach. European Journal of Operational Research 59:
167-182.
Bakken, B.E. 1993. Learning and transfer of understanding in dynamic decision
environments. Doctoral Dissertation, Sloan Management School, MIT.
Cosiga, 2003. http://www.biba.uni-bremen.de/projects/cosiga/index.html.
Dangelmeier, W., U. Pape and M. Riither 2001. Multiagentensysteme _ fiir
Produktionsnetzwerke in der Serienfertigung [Multi-agent systems for production
networks in the serial production]. Industrie Management 17(6): 21-24.
Diehl, E. and J.D. Sterman 1995. Effects of feedback complexity on dynamic decision
making. Organizational Behavior and Human Decision Processes 62(2): 198-215.
Glotrain, 1997. http://www.biba.uni-bremen.de/projects/Glotrain/willkommen.html.
Graham A.K., P. Senge 1990. Computer-based case studies and learning laboratory
projects. System Dynamics Review 6(1): 100-105.
Graham, A.K., J.D.W. Morecroft, P. Senge, J.D. Sterman 1992. Model-supported case
studies for management education. European Journal of Operational Research 59: 151-
166.
Gr6Bler, A. 2004. Don’t let history repeat itself - methodological issues concerning the
use of simulators in teaching and experimentation. System Dynamics Review 20(3): 263-
274.
Hieber, R. 1998. Optimierung von Produktionsnetzwerken durch Anwendung
unternehmens-iibergreifender Diagnosetechniken [Optimization of production networks
through the application of diagnosis techniques]. Industrie Management 14(6): 62-65.
Lange, I., A. Ziegenbein 2005. The constraints game — learning the theory of constraints
with a dice game. 9"" International Workshop on Experimental Interactive learning in
Industrial Management, “New Approaches on Learning, Studying and Teaching”,
Espoo, Finland, June 5-7. http://www.simlab.tkk.fi/events/lange.pdf
Langley, P.A. and J.D.W. Morecroft 2004. Performance and learning in a simulation of
oil industry dynamics. European Journal of Operational Research 155: 715-732.
20
Larsen, E.R., J.D.W. Morecroft and J.S. Thomsen 1999. Complex behaviour in a
production-distribution model. European Journal of Operational Research 119: 61-74.
Maier, F.H. and A. GréBler 2000. What are we talking about? — A taxonomy of
computer simulations to support learning. System Dynamics Review 16(2): 135-148.
Miller, G.A. 1956. The magical number seven, plus or minus two: Some limits on our
capacity for processing information. The Psychological Review 63: 81-97.
Morecroft, J.D. and J.D. Sterman (eds.) 1994. Modeling for learning organizations.
Portland, Oregon: Productivity Press.
Neubert, R., O. Gorlitz and T. Teich 2004. Automated negotiations of supply contracts
for flexible production networks. International J ournal of Production Economics 89(2):
175-187.
Oliva, R. and P. Gongalves 2005. Behavioral causes of demand amplification in supply
chains: “‘Satisficing” policies with limited information cues. Proceedings of the System
Dynamics Conference, Boston, MA.
Paich, M. and J.D. Sterman 1993. Boom, bust, and failures to learn in experimental
markets. Management Science 39(12): 1439-1458.
Probst, G.J.B. and B.S.T. Biichel 1994. Organisationales Lernen: Wettbewerbsvorteil
der Zukunft [Organizational learning: Competitive advantage of the future].
Wiesbaden: Gabler.
Scholz-Reiter, B,. W. Echelmeyer, T. Hamann and J. Hoheisel 2002a. Games for
engineering education. Modelling and Simulation, SCS-Publication, 16" European
Simulation Multiconference, 3-5 June, 2002, Darmstadt, Germany, 391-394.
Scholz-Reiter, B., S. Gavirey, W. Echelmeyer, T. Hamann and R. Doberenz 2002b.
Developing a virtual tutorial system for online simulation games. 30" SEFI Annual
Conference, Firenze, 8-11 September, 2002.
Scholz-Reiter, B., U. Hinrichs, S. Delhoum 200S5a. Analyse auftretender Instabilitaéten in
dynamischen Produktions- und Logistiknetzwerken [Analysis of the Instabilities in
dynamic Production and Logistic Networks]. Industrie Management 21(5): 25-28.
Scholz-Reiter, B. M. Freitag, C. de Beer, T. Jagalski 2005b. Modeling dynamics of
autonomous logistic processes: Discrete-event versus continuous approaches. CIRP
Annals on Manufacturing Technologies: 413-416.
21
Senge, Peter. 1990. The fifth discipline: the art & practice of the learning organization.
New York: Currency Doubleday.
Senge, P. and J.D. Sterman 1992. Systems thinking and organizational learning: Acting
locally and thinking globally in the organization of the future. European Journal of
Operational Research 59: 137-150.
Senge, P., A. Kleiner, C. Roberts, R. B. Ross and R. J. Smith 1994. The fifth discipline
fieldbook strategies and tools for building a learning organization. New York:
Doubleday.
Simon, Herbert A. 1982. Models of bounded rationality. Cambridge, MA: The MIT
Press.
Sterman, J.D. 1988. People express management flight simulator. Software and
documentation available from J. Sterman, Sloan School of Management, Cambridge,
MA.
Sterman, J.D. 1989a. Modelling managerial behaviour: Misperceptions of feedback in a
dynamic decision making experiment. Management Science 35(3): 321-339.
Sterman, J.D. 1989b. Misperceptions of feedback in dynamic decision making.
Organizational Behavior and Human Decision Processes 43: 301-335.
Sterman, J.D. 1994. Learning in and about complex systems. System Dynamics Review
10(2-3): 291-330.
Sterman, J.D. 2000. Business dynamics: Systems thinking and modeling for a complex
world. Boston, MA: Irwin/McGraw-Hill.
Tversky, A. and D. Kahneman 1974. Judgment under uncertainty: Heuristics and biases.
Science 185: 1124-1131.
Vennix, J.A.M. 1999. Group model-building: Tackling messy problems. System
Dynamics Review 15(4): 379-401.
Warren, K. and P. Langley 1999. The effective communication of system dynamics to
improve insight and learning in management education. Journal of the Operational
Research Society 50: 396-404.
Wiendahl, H.-P., K. Helms and M. Hébig 1998. Management of variable production
networks — visions, management and tools. CIRP Annals 47(2): 549-555.
Wiendahl, H.-P. and S. Lutz 2002. Production in networks. CIRP Annals on
Manufacturing Technologies 51(2): 573-586.
22
Windhoff, Gert, 2001. Planspiele fiir die Verteilte Produktion [Business games for
distributed production]. Dissertation, University of Bremen.
Appendix A: Determination of order rate P13
Order rate P13 = MAX (indicated orders P13, 0)
Indicated orders P13 = adjustment WIP P13 + desired acquisition rate P13
Desired acquisition rate P13 = MAX (expected order rate P31 + expected order rate dist
P13 + adjustment for stock P13, 0)
Adjustment for stock P13 = (desired stock P13 - Stock P13)/time to adjust stock P13
Adjustment WIP P13 = (desired WIP P13 - WIP P13)/WIP adjustment time P13
Desired WIP P13 = desired acquisition rate P13 * expected P13 min OPT
Desired stock P13 = desired stock coverage P13 * (expected order rate P31 + expected
order rate dist P13)
Expected order rate P31 = Integral (change in exp orders P31, order rate P3110)
Expected order rate dist P13 = Integral (change in exp orders dist P13, order rate
distributor for P11,.)
Order rate distributor for P11 = RANDOM NORMAL (5, 10, 8, 1, 1)
Appendix B: Determination of order rate P31:
Order rate P31 = MAX (indicated orders P31, 0)
Indicated orders P31 = adjustment WIP P31 + desired acquisition rate P31
Desired acquisition rate P31 = MAX (adjustment for stock P31 + expected order rate
dist P31, 0)
Adjustment WIP P31 = (desired WIP P31 - WIP P31)/WIP adjustment time P31
Adjustment for stock P31 = (desired stock P31 - Stock P31)/time to adjust stock P31
Desired WIP P31 = desired acquisition rate P31 * expected P31 min OPT
Desired stock P31 = desired stock coverage P31 * expected order rate dist P31
Expected order rate dist P31 = Integral (change in exp orders dist P31, order P3110)
Order P31 = 12 + STEP (12, 300)
23
