| To Main Proceedings Document |
An Operational A pplication of System Dynamics in the Automotive
Industry: Inventory Management at BMW
Andreas Listl Ingo Notzon
BMW Group University of Bergen
D-93055 Regensburg N-5020 Bergen, Norway
Andreas. Listl@ bmw.de Ingo.Notzon@ gmx.de
Abstract. The application of system dynamics modelling in strategic decision-making and the
analysis of market scenarios has been widely recognized. However, little attention has been
placed on the application of system dynamics models to support operational decision-making.
This paper presents a simple system dynamics model developed for BMW’s production planning
department to support decision making on production schedule, -mix and inventory management
on a day-to-day basis. As we will show in the paper, production planning in car manufacturing
is characterized by high complexity and uncertainty due to various parameters influencing the
inventory levels and production progress. In this situation a static monitoring of inventory levels
is not sufficient for production planning, since the inventory levels are critical for the car
production process to work properly. Therefore a dynamic monitoring tool is needed. Based ona
relatively simple structure, an easy-to-use interface, and online data-exchange, the developed
system dynamics model described in this paper offers a tool to assess the risk associated with
different inventory policies and improves the management of inventories and production
schedules. The paper concludes with the experience made during the last 1’4 years, in which the
model has been used by the production planners at BMW’s production plant in Regensburg.
1. Introduction
Ever since Henry Ford made his famous offer of “any colour as long as it’s black”, the position of
a consumer, who is interested in buying a new car, has changed fundamentally. Today’s
automobile market is driven by rapidly changing consumer needs and tastes. The intense
competition in the world automobile industry has forced the manufacturers to deliver high quality
and value products, and build and sustain strong consumer relationships (Clark and Fujimoto,
1993; Wetlaufer, 1999). To fulfill the needs of sophisticated customers, manufacturers have
steadily increased product variety and developed new product features. Moreover, manufacturers
are exploring ways in which they can shorten the lead-time between a customer's order, and
vehicle delivery.
The high level of product complexity and variety has a strong impact on the automobile assembly
operations and the production planning process. Traditionally, the production management goal
was to drive up the accumulated volume of production and so drive down costs through the
realization of scale economies (Hill and Jones, 1995). Scale economies lower unit cost by
spreading fixed costs (e.g. cost of machinery, administration, advertising, R&D) over a large
volume of output. Riding down the unit-cost curve involved “pushing” a great number of
unordered and standardized cars into dealer stocks. This is starkly opposed to modem
manufacturing “pull” processes, in which parts are not made until they are needed by the next
upstream step in the production process (“JIT” - Just in Time delivery or production).
The shift from mass production to lean production was driven by the need of higher quality,
increased product variety and a more customer-oriented approach to manufacturing. The goal of
lean production is to decrease inventories and to reduce the waste of time and material in the
manufacturing process, thereby increasing production efficiency and product quality (Womack et
al, 1990). The “Three day car” (three days from customer order to delivery) is the vision at the
end of the “lean” development, and car manufacturers around the world pursue “built-to-order”
strategies to produce cars, which exactly meet customer expectations (for details see Anderson,
1996; Graves and Warburton, 1999).
The described trends of “pull’-oriented manufacturing and distribution processes have lead to
fundamental changes in the manufacturing technology and the requirements of the production
planning and control process. Production planning has become a highly complex process and has
to deal with various parameters, influencing inventory levels and production throughput (see
section 3 for details). Standard PPS systems (computer-based Production Planning and Control
Systems) or static spreadsheet planning are not capable of supporting the operational production
planning process in a dynamic and uncertain environment. Therefore, we have developed a system
dynamics model for BMW’s production-planning department to support decision making on
production schedule, -mix and inventory management on a day-to-day basis. Based on a relatively
simple structure, an easy-to-use interface, and online data-exchange, the developed system
dynamics model offers a tool to assess the risk associated with different inventory policies and
improves the management of inventories and production schedules. Before explaining details of
the planning process in section 3, we will have a closer look at the production site and the
manufacturing process in section 2. Section 4 gives an overview of the problems in the planning
process and motivates the need for a simulation tool, based on a system dynamics model. In
section 5 we will describe the developed simulation tool in more detail. Section 6 summarizes the
experiences we made with our approach at BMWs plant in Regensburg, Germany. Finally, section
7 concludes the paper.
2. Production Site and Production Process
Since November 1986, BMW 3 series cars have been in production in Regensburg, situated on
the banks of the Danube in Bavaria. Covering an area of 350 acres, the plant employs
approximately 9000 workers producing 215,000 BMW 3 series (sedan, station wagon, coupe,
convertible) per year. The manufacturing process of an automobile can be divided into four major
steps: 1) Stamping of the metal sheet blades in the press shop, 2) Welding of the parts in the white
body shop, 3) Painting of the car body in the paint shop, and 4) the final assembly of the car.
Figure 1 illustrates this process.
The production process begins with metal sheet blades that are stamped in the press shop. Then
the stamped body parts are welded together in the white body shop to create a car body. Before
the car body is painted in the paint shop it is stored in a stacker called white body stacker. Since
the painting process works optimal if bodies to be painted with the same color follow immediately
one after the other, the white body stacker is used to sort the bodies produced by the white body
shop. The aim is to build clusters of approximately 20 bodies to be painted with the same color.
Before the painted bodies go to the assembly area, where several thousand parts are added on an
assembly line, they are stored in another stacker called painted body stacker. 3series sedan cars,
coupes, convertibles and touring models are produced in varying sequence. On the assembly line,
the aim is to achieve an optimum mix of vehicles. Each vehicle is being built according customer
order and assembly thus needs to take into account a wide diversity of customer order wishes.
The stacker is used to determine a sequence of cars on the line, which ensures an even distribution
of the workload amongst assembly workers. Thus the painted body stacker is used to sort the
painted bodies in a way to support an optimal assembly process.
Stacker Stacker
Figure 1: Overview of Manufacturing Process
The assembly line is divided into three main sections: trim, chassis and final assembly. Lightweight
parts are installed in the trim section, while heavier parts, such as the engine and transmission, are
installed in the chassis section. The production process ends in the final assembly, where a number
of small operations are performed, such as filling various fluid reservoirs, making numerous
quality checks and conducting minor repairs as necessary.
The parts supply process supports the assembly line. This process starts with a supplier plant
producing a part and ends with the part arriving in an assembly worker's hand as the car requiring
that part enters the workstation. Supply plants are provided with build schedules via an electronic
data interchange network with sufficient lead-time to deliver the required parts when they are
needed. The plant in Regensburg uses several Just-in-Time supply relations to deal with product
variety. The cars produced by the plant Regensburg vary along several dimensions, including body
stile, exterior and interior color, harness, power train and choice of options.
Looking in more detail at the described production process there are many possible circumstances
which raise problems due to the production process, such as downtimes of robots in the white
body shop, increase of repainting because of paint process problems or downtime of work
stations in the assembly line. Many of these problems influence the inventory levels of the white
body stacker and/or the painted body stacker. Another important factor influencing these
inventory levels is the operations time of a production department. At the plant in Regensburg the
white body shop uses a different shift model with different operations times than the paint shop.
Therefore the white body stacker must be able to buffer sufficient white bodies to compensate
these variations in operations times. Similarly the painted body stacker must be able to buffer
enough painted bodies due to different shift models between the paint shop and the assembly
shop.
3. Production Planning and C ontrol
At BMW's production plant in Regensburg the production-planning department is responsible for
planning and controlling the production schedule and -mix, including the management of the
white body and the painted body stacker. Four basic functions can be identified: planning,
dispatching, control, and taking of corrective action if the actual performance deviates from the
planned performance.
In the planning phase the production schedule and the production mix (i.e. the ratio of sedan,
coupe, station wagon, convertible of the whole production) are planned. Typically the period that
is planned for is one month. Note that planning starts several weeks before the actual start of
production and the production schedule has to be fixed two weeks before actual production
starts. In car manufacturing production planning is characterized by high complexity and
uncertainty because various parameters, such as operations times, capacity of the white body
stacker and the painted body stacker, changes in production mix, changes in production capacity,
and various time delays, such as the time a car body needs to go through a production area or the
time a body is stored in the white body stacker resp. painted body stacker, have to be considered.
The function of dispatching puts the plan into effect, that is operations are started in accordance
with the plan. In the operating phase the production-planning department has to observe and
record the actual performance of production, e.g. it monitors the inventory levels of the white
body and the painted body stacker. Typically this is done on a day-to-day basis. The actual
performance is compared with the planned performance, and when required, corrective action is
taken. This happens, e.g. if the inventory levels of the white body and/or painted body stacker get
too low or too high or if problems in the production process occur. Finding out the right steps to
correct the inventory levels is also characterized by high complexity and uncertainty because of
various parameters and various time delays.
4. The Problem and the Need for a new Tool
The objectives of production planning and control are to fulfil customer orders in time, reduce
inventories and throughput time, and to establish schedules for work that will ensure the optimum
utilisation of materials, workers, and machines. As the description of the manufacturing process
shows, production planning in car manufacturing is complex, because a large number of constants
and constraints have to be considered, such as operation times, stacker and production capacity,
and other constraints imposed by the underlying manufacturing system. Moreover, the planner has
to take the dynamics of the system into consideration, that is time delays, handling and process
times are important variables in the planning process. Even worse, the planning process is
characterized by high uncertainty, because downtimes or delivery problems of suppliers can cause
production losses and require changes in the production schedule in the short run. In the latter
case, the impact and timing of such changes cannot be foreseen and considered in the long-term
planning process and challenges the planner to meet the master production scheduling despite of
the occurring problems.
Standard PPS systems facilitate long range and rough-cut production planning. However, they are
not capable in coordinating the operational planning process in a multistage production system
due to the high complexity of such systems (Amold et al., 1997). Adapting the system to the
process (“customising”) and refining the planning process requires reprogramming of the
software, which is costly and time consuming (Arnold et al., 1997). Especially, the difficulties in
implementing changes in the production schedule in the short run was criticized by authors
(Milberg and Burger, 1991; Amold et al., 1997) and a better support for the operational planning
ona day-to-day basis was requested (Milberg and Burger, 1991; Eversheim und Thome, 1988).
To bridge the gap between the described requirements of the production-planning department and
the long range planning function provided by the central PPS, we have developed a dynamic
planning and monitoring tool with an easy-to-use interface and online data-exchange. The
developed tool is based on a system dynamics model and puts the production-planning department
in a position to assess the risk associated with different inventory policies and improves the
management of inventories and production schedules. Figure 2 illustrates the planning process and
the information flows.
PPS
ong-term Database/
Planning PPS
Monthly / Day-to-Day ¢d _—
Planning based oie oe = — |
on SD Simulation d
ssau6oud uoganpaid pue ‘sjano}
Arcquenut ‘s12pi0 jo BuayUOW
Operational Planning/
Dispatching of Orders As A MR
Figure 2: Planning Process and Integration of System Dynamics Modelling
In the next chapter we describe the developed dynamic planning and monitoring tool in detail and
show how the production-planning department uses it for production planning and inventory
management.
5. System Dynamics model for inventory management
Figure 3 shows a basic Causal Loop Diagram, illustrating the process of production planning and
control. As shown in figure 3, three main factors influence the rate at which car bodies flow
among the stackers: Next to the operation times (the working time of the process step, e.g. the
number of shifts), cycle times and downtimes determine the throughput rate of production. With
downtimes we mean the time when a machine is broken or the workers cannot work because
components are not available, causing production losses. Cycle time denotes the time required to
process a car body in one workstation. For example, at the plant in Regensburg the cycle time in
the assembly shop is about 80 seconds.
In the planning phase the production-planning department sets up a production schedule. It
determines the number of produced units per production department (i.e. body shop, paint shop,
assembly shop) for a time period of one month. The production schedule also determines the
operations times of each production department. The time unit used in the production schedule is
shifts. From the production schedule the production-planning department is able to derive the
inventory level of the white body resp. the painted body stacker for every shift (planned inventory
level).
Downtimes Cycle time
Cycle time of previous step of next step
of previous step
Downtimes
of next step
Operations times
fo} tions ti
perations times of next step
of previous step
Bodies for next
production step
Bodies from previous
production
Changing production inventory level
schedule and -mix deviation
Planned
XK” inventory level
Figure 3: Causal Loop Diagram for Inventory Management
Downtimes and changes in cycle time may reduce the throughput rate, causing deviation of the
actual inventory level in the white body resp. painted body stacker to the planned inventory level
(inventory level deviation). As long as this difference is within acceptable limits, there is no need
to change the production schedule. But in many situations changes in the production schedule or
production mix are required to meet the long-term production goals (changing production
schedule and -mix). However, changes in the production schedule have no immediate impact on
the inventory level. This is because normally it is impossible to change the production schedule or
-mix for the actual running shift. Typically changes in the production schedule or -mix are made
for the next week, which leads to a time delay of more than 10 shifts (compare figure 3).
Moreover, changes in the schedule affect the whole manufacturing process and require
coordination meetings. In these meetings, members of different departments take part to discuss
various options and possible strategies to regain the lost time and production units. In the past
these teams often faced difficulties in finding a solution every team member could agree with,
because the team members had different understandings and mental models about how the system
works.
To support the team’s discussion about possible changes in the production schedule we developed
an easy-to-use system dynamics tool with the Powersim Constructor (Byrknes, A.-H.). Figure 4
shows the architecture of this tool. The core of the tool is a system dynamics model portraying
the flow of the manufactured bodies through the body shop, the paint shop and the assembly shop
based on the causal loop diagram in figure 3. Since the details of the production flow in the body,
paint and assembly shop are not important for the flow of bodies through the white body resp.
painted body stacker they are modeled as black boxes. Thus the production departments are
modeled as rates characterizing the input/output of a production department.
The production-planning department stores the production schedule in a spreadsheet. Via the
DDE interface the data is transferred to Powersim Constructor and used to calculate the flows,
which determine the input/output of a production department per shift. At the beginning of the
simulation the inventory levels (the stocks) of the white body and the painted body stacker are
initialized. This initialization is done by reading the actual inventory levels from an online
production database system via DDE interface. As output the simulation generates graphs
showing the behavior of the inventory levels of the white body and the painted body stacker as
estimated for the future.
Graphical output showing the simulated
aN invemtony levels for the white body and the
painted body stacker 7
Wi ~~ ,
Powersim
Constructor computer
model describing the
simulated system
Initializing the inventory
level with the actual values
(using DDE interface)
Reading the planned a
Production units per shift J
(using DDE interface) 7
a
Online Production
Production Schedule
Database
(Spreadsheet)
Figure 4: Architecture of the production-planning tool
Using this tool the teams can discuss changes in the production schedule and easily test different
options (scenarios) by changing some parameters (e.g. introducing a new shift, changing the
production mix, etc.) and simulating the impact of these changes on the inventory levels of the
white body and the painted body stacker. Thus the team can easily compare different scenarios
and choose the solution, which compensates best the deviation between planned and actual
inventory level.
Since the members of the teams discussing changes in the production schedule are not experts in
system dynamics and thus are not able to directly manipulate the system dynamics computer
model built with Powersim constructor, we decided to develop an easy-to-use user-interface
(flight simulator cockpit). This interface allows to change the main parameters needed to develop
different production-schedules by using interactive dialog elements, like slider/bar objects, gauge
objects, number objects and buttons. The screenshot in figure 5 shows an example of the user-
interface of the developed flight simulator cockpit.
Figure 5: Sample Screenshot of the simulation tool
6. Experiences
The production-planning department at BMW’s plant in Regensburg uses the described simulation
tool for more than 1 years. From the experiences made during this time period three main areas
of application have emerged: monthly preview, weekly preview and daily monitoring and preview.
The monthly preview is used to hedge against too high or too low inventory levels in the white
body resp. painted body stacker due to planning errors in the planning phase. Before we
introduced the SD-based tool, the production-planning department planned the monthly
production schedule with a spreadsheet. From this spreadsheet the planners derived the levels of
the two stackers. Typically this task was done by an expert in planning and in using the
spreadsheet. Planning different scenarios was a time consuming process and could be only done
by the expert. Thus, only little attention was given to plan different scenarios for the production
schedule and many problems arose in the operational phase because no strategy for compensating
deviations were preplanned. Nowadays with the developed simulation tool the production-
planning department is able to investigate various scenarios in shorter time. Also the scenario
planning can be done by non-experts in using the production-planning spreadsheet application,
because the user-interface of the simulation tool offers many possibilities for specifying different
scenarios. To summarize, the new simulation tool increased the reliability of the monthly
production schedule.
The weekly preview is used to plan and discuss different options of compensating a deviation (for
example: setting up an additional shift, reducing the planned output of a shift, etc.). Without the
simulation tool this process was very difficult and time consuming. The reasons for that were: (1)
since many people from different departments with different experiences in production planning
were involved, there was no common understanding of how different actions effect the inventory
levels. (2) There was no standardized methodology to plan different options. (3) Typically there
was not enough time to investigate different options of compensation. With the newly developed
simulation tool the process of planning and discussing different options was improved greatly.
Nowadays the simulation tool, as a standardized tool, is used during the discussion by projecting
the inventory level graphs from the computer directly on a screen on the wall. Thus all involved
team members can immediately see the effect of an action on the inventory levels. This improved
the common understanding of the system. Beyond that the time for discussing one option of
compensation could be shortened significantly and therefore more options can be discussed in the
same time.
The actual inventory levels of the white body and the painted body stackers are very important for
the production process to work properly. Therefore the production-planning department monitors
the inventory levels on a day-to-day basis and provides reports to other production related
departments in the plant. In former days, without the usage of the simulation tool, this daily
production report showed a static picture of the actual production situation. Therefore, it was
difficult to estimate the future development of the stackers and critical situations were often
recognized too late. With the help of the new simulation tool the production-planning department
extended the daily production report with graphs, showing the estimated development of the
inventory levels for the next few days. This report is daily published in the plant’s intranet. This
allows all departments to anticipate critical situations much earlier and thus, to discuss actions to
avoid these critical situations in advance.
7. Conclusions
As shown in this paper the production planning process in car manufacturing is very complex, in
order to satisfy customer demands and tastes. Traditional methods of planning, controlling and
monitoring the production process are not applicable anymore. Thus new methods are needed. At
BMWs plant in Regensburg a system dynamics approach is used. As shown in this paper a simple
dynamic simulation tool with an easy-to-use user-interface can significantly improve the
production planning process. Although in the beginning of the project some people were skeptical
about the simulation, the experiences of the past 114 years, made at BMWs plant in Regensburg,
showed the usefulness of our approach. Today the developed simulation tool is used on a day-to-
day basis to support monthly, weekly and daily preview as well as daily monitoring of the
performance of the production process.
To emphasize the usefulness of the developed simulation tool we would like to cite the manager
of the production-planning department at BMWs plant in Regensburg, who said in an interview:
“The simulation tool used within our department enabled us to increase our production output
for more than 2000 units this year. This was achievable, because the simulation tool showed us
in advance the possibilities of increasing the production by introducing extra shifts in the
production schedule” . Because these 2000 units were needed to satisfy costumer demands, this
lead to an increase of net-income of more than 15 million US$.
Since the usefulness of our approach is widely accepted at BMW our future plans are to install the
simulation tool at other plants of the BMW Group.!
'The authors are grateful to the many people at the BMWs Assembly Plant in Regensburg who helped us
understanding plant operations including Hans Ebenbichler, Manager of the Production Planning and Control
Department, and Thomas Arlt, who did the main work in the development of the concepts for the simulation tool.
We also thank Peter Claussen, Manager of the White Body Shop, for his support during the test-phase of the
simulation tool.
References
Amrine, H. T. et al. (1993). Manufacturing Organization and Management. Prentice Hall International,
Englewood Cliffs, New Jersey.
Anderson, M. L. (1997). Electronic new Media and the Auto Industry: New Dimensions in "Market Facing"
Systems, International Motor Vehicle Program Working paper, MIT.
Arnold, J., Beisswenger, P. and Reisch, O (1997). Einsatz von Simulation und kooperativen Leitstanden. Industrie
Management, Volume 1, No. 2, pp. 27-31.
Byrknes, A.-H. (1996). Powersim Tutorial, Powersim Press.
BMW AG (1991). Presentation of BMW AG Regensburg plant.
Clark, K. B. and Fujimoto, T. (1991). Product Development Performance: Strategy, Organization and Management
in the World Auto Industry. Harvard Business School Publishing. Boston, MA.
Eversheim, W. and Thome, H. (1988). Simulation in der Werkstatt. CIM-Management, Vol. 2, No. 4, pp. 9-15.
Graves, A. and Warburton, M. (1999). Will the European Industry ever build a 3 Day Car? International Motor
Vehicle Program Working paper, MIT.
Greene, J. H. (Editor) (1997). Production and Inventory Control Handbook. 3" Edition. Mc Graw Hill, NY.
Hill, C. and Gareth, J. R. (1995). Strategic Management Theory - An integrated approach. 3" Edition, Houghton
Mifflin, Boston, MA.
Milberg, J. and Burger, C. (1991). Simulation als Hilfsmittel fiir die Produktionsplanung und -steuerung.
Zeitschrift fiir wirtschaftliche Fertigung und A utomatisierung, Vol. 86, No. 2, pp. 76 - 79.
Wetlaufer, S. (1999). Driving Change - An Interview with Ford Motor Company's Jacques Nasser. Harvard
Business Review, Issue March-A pril 1999, pp. 77-88
Stanek, W. , Jong, H. de (1993). Simulation in der Fabrikplanung und Fertigungssteuerung.
Zeitschrift fiir wirtschaftliche Fertigung und Automatisierung, Vol. 88, No. 5, pp. 226-228.
Wiendahl, H.-P. (1990). Simulationsmodelle in der Produktionsplanung und -steuerung
Zeitschrift fiir wirtschaftliche Fertigung und A utomatisierung, Vol. 85, No. 3, 1990, pp. 137-141.
Womack, J. P., Jones, D. T. and Roos, D. (1990). The Machine that Changed the World - The story of lean
production. HarperCollins Publishers, NY .
