Parallel Program
Object based dynamic modeling
Lars Vavik Magne Myrtveit
Stord/Haugesund College ModellData AS
N-5414 Rommetveit, Norway N-5120 Manger, Norway
Phone: +47 5349 1379 Phone: +47 5637 4009
Fax: + 47 5341 0160 Fax: +47 5637 3500
Internet: larsv@stordlh.no Internet: magmy@modeld.no
(ABSTRACT)
This research is supported by the Norwegian Research Council, Grant no. 31867/223-1994
This paper reports on collaborative action research partnerships with corporations to develop new
tools to accelerate learning, and test these tools in real organizations.
In the collaborative program “From Systems Thinking to Praxis” several corporations are
facing the challenge of building a virtual model to mirror the companies’ business strategies.
These models have to meet a level of functionality which allow managers to compare the output
of the model with real data from the organization. On the other hand, the models have to be
flexible enough to give managers a laboratory for business process redesign.
The focus in this paper is on the laboratory as an exploratory learning environment. The work
is based both on modeling as a way of learning, and learning from running simulations
(management flight simulators). The challenge is to build a bridge between the functionality and
the transparency of the model. A combination of the flexibility of the modeling toolkit and the
constraints of a simulation applications is needed.
The computer based learning environment presented here is based on a state of the art toolkit
for constructing system dynamic models (graphical flow diagram, arrays, OLE links, DDE data
transfers). To meet the new requirement, major extensions were made to the toolkit in order to
allow for concrete objects in addition to the general and abstract objects of accumulator—flow
diagrams. The resulting system opens up for modeling in terms of objects within the problem
area (for example banks, markets, capital, transportation), and at the same time gives the
possibility to use the basic blocks (accumulators and flows).
A management flight simulator based on high level objects renders the possibilities to change
the model structures, not only parameters.
The value of a modeling process based on object based dynamic modeling is measured as the
probability that a manager will grasp the result from a single analysis and act on it. With the
object based dynamic modeling approach we challenge past conclusions that there is a tradeoff
between model complexity and the value to the user — a compromise between functionality and
transparency.
Keywords: system dynamics, objects, computer aided modeling, exploratory learning
904
System Dynamics '95 — Volume II
Limits to learning in computer based environments
When we use the term “learning environment” in connection with computer-based simulation,
we refer to a computer based arena where people are free to learn, using their own style at their
own pace, and in a sequence that they determine themselves. A simulation application and a
modeling toolkit provides the “engine” for learning within this type of environment. In research
on instructional science, simulation and modeling have been distinguished as two separate
learning environments (Elsom-Cook, 1990; Sellman, 1992):
Simulations are based on a mathematical model of some system. Unlike modeling systems,
the model is not explicitly available to the learner and the model structure cannot be changed.
The learner has some sort of representation of the behavior of the model, and can manipulate
certain variables and observe the effect on other variables.
Modeling systems allow the learner to construct mathematical models of a system and
manipulate those models to see how they respond under various circumstances.
Limits to learning in the simulations application
Learning environments utilizing simulation applications have found a stable niche as an
instructional measure in varied areas. There is substantial interest in the use of computer-based
simulation in industrial training. By reviewing proceedings from system dynamics literature there
seems to be a growing interest in transferring models from modeling systems into specific
simulation based applications. In the system dynamics tradition such applications are called
“models with games interfaces” or “management flight simulators” (Andersen et al. 1990; Bean,
Diehl & Kreutzer 1992; Kreutzer et al. 1992; Morecroft 1992; Davidsen 1994; Senge 1994). The
architecture of a management flight simulator follows the same structure as for traditional
computer based instruction with simulation. Learning based on these systems fits well into a
general pool of problems.
The details of the system complexity are hidden from the user by an apparently “simple” user
interface. There are limitations to the usefulness of simulators as stand-alone tools. The user is
only a consumer of the model on which the simulator is based, and does not participate in the
model creation process. The lack of involvement in conceptualizing the model may result in a
shallow understanding of the dynamics of trial-and-error experimentation (Kim 1990). A deeper
understanding of the underlying structure requires more than playing the game repeatedly. There
is need for more explicit discussion regarding the theory underlying the operational interface. We
fear that the learner presses buttons and moves the slide bars as if he were simply playing a
computer game. Little substantial learning occurs in this situation. Users may learn how to win a
game, but they probably don't have a clue of what is really going on below the surface (Bakken
1989; Senge 1994).
The implication is that we can no longer afford to design the model and simply track on an
interface afterwards. This agrees with Brown's analysis (1989) of the requirement for producing a
“glass-box technology”. He argues that systems need to be made transparent in three senses:
domain transparency, allowing the user to see “through” the tool; internal transparency,
allowing the user to see “into the tool”; and embedding transparency, allowing the user to see the
relationship of the technology to the lager context of the interaction.
905
Parallel Program
There are several attempts to solve these problems by embedding the simulation in an
Intelligent Tutoring System (ITS). Some good examples of Intelligent Simulation Learning
Environments (ISLE) are reported for operational training (SHERLOCK, Lajoei & Lesgold 1989,
1991).
As a stand alone tool, this means that we should either build an instructional shell around the
application, or the systems should be implemented in an organization with careful advice from a
consultant. Simulation applications in organizational learning have been used more as
icebreakers, just to pin point some generic structure or simply to introduce a general dynamic
problem. In business process redesign there is a need to change the structure of the model, not
only the parameters. This requirement can hardly be supported from an ISLE shell.
Limits to learning in the modeling environment
A modeling system gives possibilities to learn about a domain by participating in constructing
the model. This is a constructivists vision of the learner as a social tool user. The computer
environment itself does not take control of the interaction, and does not react in the “intelligent”
way (reasoning about its own behavior) we expect of a simulation application embedded in an
ITS architecture. Some researchers (S. Papert 1980; John Holt, Tavistock Institute) suggest that
the combination of student and environment is all that is required.
Certainly, the computer modeling process can play an important role in facilitating a group of
managers “to make up their own minds” about a system. Lane (1994) describes a shift in
modeling style away from traditional expert consultant toward modeling as learning. Modeling
becomes an integral part of the management discussion. Consultants provide tools that capture
and express the mental models of clients. Models are owned, run and interpreted by the clients.
Much effort has been put into the modern modeling toolkits to make the modeling process as
understandable and powerful as possible. Significant progress has been made from the early
FORTRAN compiler to a modern system dynamics toolkit. Modern modeling environments offer
several valuable instructional measures and concepts. Examples include the syntax of the
modeling language; the use of a conceptual accumulator—flow framework to present the structure
of models; the integration of diagram and equations to make a graphical modeling environment.
But even the accumulator—flow framework for model conceptualization is rigorous and
precise. It is also abstract, and in many cases not the “natural” way for the less-proficient modeler
to think about system structure (S. Peterson, 1994, p. 299). The implication is that when it comes
to detailed algebraic modeling with the basic building blocks, the modeling process is often taken
over by technical experts. Morecroft et al. (1994) stated that “... a small team took part in this
phase. It is difficult to imagine how one would have engaged the full team in this more detailed
work”.
Several research papers referring to Papert’s micro worlds (1980) and the constructivistic
tradition as a theoretical fundament for “modeling as learning” (Senge, Lane, Morecroft).
Unfortunately, many case studies show limited success with the early attempts to design micro
worlds. The answer to this might be found in the discussion about “cognitive hooks”. The gap
between the environment and the existing knowledge is often to great. If we wish to provide
environments that are suitably supportive to all learners, we’ll have to build environments with
“cognitive hooks”; that is, pieces of the environment that a student can hold on to in order to
develop better models in a meaningful way.
906
System Dynamics '95 — Volume II
It is impressive to observe trainees in front of a flight simulator. Even more impressive is the
transfer from simulator training to take-off with 250 passengers — without help of an instructor.
Observing a team of managers piping accumulator—flow diagrams together in an attempt to build
their own management flight simulator, gives mixed impressions.
Redesigning CBLES to Enhance Learning
This research project aims at building a bridge between qualitative mapping of organizational
processes and quantitative modeling of systems. A collaborative action research partnership is set
up with a group of senior managers and planners from a major oil company. The aim is to build
virtual worlds to mirror and develop business strategies.
The requirement
A major challenge is to design a system that provides a library of domain specific building blocks
embedded in a modeling tool. The components have to be flexible to give planners and business
unit managers the power to build models for describing and redesigning business processes.
Real World.
aa ;
fay 6
oye gy Virtual World
Figure 1: Variant of Argyris and Schon's learning cycle
On the other hand, the models must not just be black-box predictors. It should be possible for the
user to compare the output of the model to real business data from the organization (Myrtveit
1995).
Real World
Measure
Virtual World
Model a= Simulate
Compare
Figure 2: Comparing model output to real business data
907
Parallel Program
The resulting application should be an integral part of the modeling toolkit, allowing users to
control, manipulate, and monitor the simulation at different levels of abstraction.
The modeling process and the simulation process correspond to designing organizations and
running organizations. Business process redesign is a tool for planning changes to organizations,
while total quality management deals with optimizing the way a process is run. Designing new
business processes is especially important in a world with rapid changes in technology (Foster
1986). This is where quantum leaps may be done through innovation. The potential of improving
processes operation is limited by process constraints (Senge 1990, Limits to success archetype).
These relationships are summarized in Figure 3.
Modeling Simulation
Changing structure Changing parameters
Planning a business Operating a business
Business process redesign Total quality management
“Doing the right things” “Doing things right”
Quantum leaps Limits to success
Performance: Performance
Figure 3: Modeling and simulation; planning and operation
Objects as building blocks
Object orientation is becoming increasingly popular both as a design methodology and as a
technology for creating computer software. The use of objects is an elegant way to break the
“world” into smaller parts that are easier to handle.
Classification is an important characteristic of objects. This means that objects with the same
properties (or attributes) and operations are grouped into a class. An attribute or operation that is
local to a class, can be hidden (encapsulated) inside the class.
Only in rare circumstances will an accumulator—flow diagram represent a natural object
mapping of a system. (A diagram of physical flows through pipes is an exception.) In general,
accumulator—flow diagrams focus on object attributes and relationships between attributes. This
is only natural, as the main reason for using accumulator—flow diagrams is to describe dynamic
relationships between attributes of a system and deduce the resulting behavior over time.
Providing higher level objects may be a way to make modeling useful to non-modelers.
908
System Dynamics '95 — Volume IT
Choice of objects
The topology of a system is in this context defined as the physical arrangement of the building
blocks which describe the total system. In general, objects depend both on the system to be
modeled and the use of the model.
It is important that objects are chosen from the problem domain — and not from the
technology (implementation language, software tool). In this way model users can relate to the
models without knowledge about the underlying technology (for example, accumulator—flow
diagrams). Figure 4 illustrates how a palette with high-level objects (building blocks) can be
combined to describe a problem within a given area (for example manufacturing).
Objects My Problem
Produc-
tion
Marke- .
ting
( Work x o
force Capital
Figure 4: Describing a problem by combining high level objects
The interface to objects
It is common that a model can be presented through different views, for example an equations
view (textual equations), a diagram view (accumulator—flow diagram), and an application
interface (micro world). For object based systems one interface is used to define objects, and
another to display objects. The use of so-called property sheets has become the standard way of
defining objects in a graphical user interface like Windows and OS/2. A property sheet is a
definition dialog box for an object, and contains controls for determining the attributes of an
object. When opening the property sheet for a given object, only attributes for that object will be
made available for the user. This ensures that property sheets of high-level objects will not get
cluttered with details from lower levels.
Example — A micro world for the gas industry
The objective of this project was to study alternative ways of developing the gas fields of the
North Sea. The complex dynamics of markets, Build
transport and processing led to a simulation trast
approach. A decision was made to create a micro
world, where planners could build alternative Investigate Run
infrastructures and simulate the consequences. A results simulation
major requirement was that it should be easy to
compare alternative strategies. Therefore all .
reports were designed to display values from two Figure 5: Using the gas micro world
simulation runs. Consistent use of color was used
909
Parallel Program
to identify current run and reference run throughout the system (red: current run, gray: reference
run). A typical use of the gas micro world involves three steps in a closed loop (Figure 5).
Constructing a model in the micro world
The user is modeling the micro world using high level objects from the problem domain. Objects
include: reservoir, well, pipe,
processing facility, — Help
compressor, joint, storage, r TRIER LS] [Uulizaton
and market.
Each object is associated
with an icon that is easily
recognized by the typical
user. Objects are made
available for the user as
building blocks in a toolbar
(see left hand side of Figure
6). Objects are moved into
the map according to
standard Windows
conventions (point, click,
drag).
The user opens the
property sheet of an object by
double-clicking on the object.
Each property sheet lists
parameters of an object. Figure 6: User in terface of gas micro world
Parameters are chosen to give
enough flexibility for the user to investigate the main issues in the problem area. At the same
time care has been taken to avoid unnecessary detail.
AC ©) SISO
Zz 2% |
Running a simulation
The user can specify time horizon for the simulation from the Scenario menu. In addition a life
cycle can be assigned to each object (for example point in time when a pipe line is brought into
operation).
When the user presses the play button the simulation starts.
Investigating simulation output
During and after simulation the user can observe model behavior in reports that may be created
in the diagram (map area) or in separate windows. In Figure 6 a bar is displaying the current
utilization (% of maximum capacity) of a pipe during the current run (left hand bar) and a
reference run (right hand bar). Minimum, maximum, and average values can be displayed
together with current values. This makes it possible to identify bottle necks, for example
elements where the maximum capacity have been reached, or elements that have unnecessary
large capacity (potential cost reductions). Oscillations caused by varying demands between warm
910
System Dynamics '95 — Volume II
and cold seasons are illustrated in time graphs that can be opened from any object. Global
measures of produced volume, income and expenses are also available.
Creation of the micro world itself
The micro world was designed as a two-layer application with an accumulator—flow model at
the bottom and an
object model at the
top. The 5 J
accumulator—flow rocessing
model was created in Diem Biscay CO) CO
POWERSIM® (Modell-
Data 1994) and the —)
object model in C++. :
POWERSIM’s API Reservoir 7, Well From_Wel Pipe ro Market. ~=—- Market
interface was used to
Fra_st To_St
link the two layers. esiose @) OC oeioree
The core part of
the simulation model
is displayed in Figure
7. Arrays are used to : Storage :
provide a pool of Figure 7: POWERSIM model of gas micro world
—
[To_Process
reservoirs, wells,
pipes, etc. that the object layer may draw from when the user creates new objects.
Project organization
The planner is the target user
of the gas micro world. Using
experimental simulation,
alternative approaches can be
analyzed and compared.
Tool | People
Gas Micro World Planners
In order to create a micro or Designers
world that is a valid mapping - —- j
of the real gas world, people at Powersim Simulation Model
different levels of the
organization were involved. Figure 8: Teaming up people
The expertise of researchers
and experts within the company was necessary to define the objects that would be used to create
the model. Simplifications and validations were done using available data from years of
operation in the North Sea.
Project evaluation
A program for testing and evaluating the gas micro world is in progress. Testing with real users
is the fundamental part of the evaluation. A “thinking aloud protocol” combined with
“cameraless videotaping” (recording of display images) of user interaction give valuable
information. There is little doubt that the system gives planners real modeling power. At the time
911
Parallel Program
of writing, the data analyses are not conclusive. More field testing and data collection will be
done, and the results will made available by the authors when the project is completed.
References
Andersen D.F., Chung, I.J., Richardson, G. & Stewart, T. (1990). Issues in designing interactive
games based on system dynamics models. Organizational learning. In D.F. Anderson, G.
Richardson & J.D. Sterman (Eds.), Proceedings of the 1990 International System Dynamics
Conference (Vol. I, pp. 31-46). Lincoln: The System Dynamics Society.
Argyris, C., and Schon D (1978), Organizational Learning: A Theory of Action Perspective,
Reading, Mass.: Addison - Wesley.
Bakken, B.E., Gould, J.M. & Kim, D.H. 1992. Management flight simulators and organizational
learning: some experimental evidence. European Journal of Operational Resarch, 59, 167-
182.
Bakken, B.E 1993. Learning and transfer of understanding in dynamic decision environments.
Unpublished PhD Thesis, Cambridge: Sloan School of Management, Massachusetts Institute
of Technology.
Bean, M.P., Diehl, E.W. & Kreutzer, D.P. 1992. Strategy simulation and scenario planning: An
application of generic system structures. Organizational learning. In J.A.M. Vennix, J. Faber,
W.J. Scheper, & A.T. Takkenberg (Eds.), Proceedings of the 1992 International System
Dynamics Conference (pp. 69-77). Lincoln: The System Dynamics Society.
Brown, J. S. 1989. Toward a new epistemology for learning, In C. Frasson and J.Gauthier (Eds.),
Intelligent Tutoring System, Norwood: Ables.
Collins, A. & Brown, J. S. 1988. The computer as a tool for learning through reflection. In
H.Mandl, A.M. Lesgold (Eds.), Learning issues for Intelligent Tutoring Systems (pp.1-18).
New York: Springer Verlag.
Davidsen, P 1994. In Morecroft, J.D.W & Sterman, J.D (Eds), Modeling For Learning
Organizations (pp. 301-317). Portland, Productivity Press.
Elsom-Cook, M. 1990. Guided discovery tutoring; A framework for ICAI research. London:Paul
Chapman.
Faryniarz, J.V. & Lockwood, L.G. 1992. Effectiveness of microcomputer simulations in
stimulating environmental problem solving by community college students. Journal of
Research in Science Teaching, 29, 453-470.
Forrester, J.W. 1961. Industrial Dynamics. Cambridge, MA: MIT Press.
Foster, R. N. 1986. Innovation — the Attacker’s Advantage. Summit Books, New York
Goodyear, P. 1991. A knowledge-based approach to supporting the use of simulation programs.
Computers in Education, 16, No. 1, 99-103.
van Joolingen W. 1993. Understanding and facilitating discovery learning in computer-based
simulation environments simulation. PhD Thesis. Eindhoven: Eindhoven University of
Technology.
912
System Dynamics '95 — Volume II
Kim, D 1990. Total quality and system dynamics: complementary approaches to organizational
learning. In D.F. Anderson, G. Richardson & J.D. Sterman (Eds.), System Dynamics '90
(Vol. I, pp. 539-534).
Kreutzer, B., Kreutzer, D.P. & Gould, J.M. 1992. The Quahog oil production simulator: A case
study of the rapid development of a management flight simulator for training. In J.A.M.
Vennix, J. Faber, W.J. Scheper, & A.T. Takkenberg (Eds.), Proceedings of the 1992
International System Dynamics Conference. (Vol. I: pp. 329-336). Lincoln: The System
Dynamics Society.
Lajoie, S.P., Lesgold, A. 1988. Apprenticeship training in the workplace: Computer-coached
practice environment as a new form of apprenticeship. Man Mediated Learning, 3, 7-28.
Lane,D.C. 1994. In Morecroft, J.D.W & Sterman, J.D (Eds), Modeling For Learning
Organizations (pp. 85-117). Portland, Productivity Press.
ModellData AS 1994. PowERSIM 2.0 User’s Guide and Reference
Morecroft, J.D.W. 1992. Design of a learning environment. The Oil Producers' Microworld.
Organizational Learning. In J.A.M. Vennix, J. Faber, W.J. Scheper, & A.T. Takkenberg
(Eds.), Proceedings of the 1990 International System Dynamics Conference (Vol. I, pp.465-
474). Lincoln: The System Dynamics Society.
Myrtveit, M. 1995. Feeding your models with real business data. Proceeding International
System Dynamics Conference
Papert, S 1980. Mindstorms, Harvester, Brighton.
Peterson, S 1994. In Morecroft, J.D.W & Sterman, J.D (Eds), Modeling For Learning
Organizations (pp. 291-300). Portland, Productivity Press.
Sellman,J. 1992. Learning Environment. Unpublished PhD Thesis. Milton Keynes: The Open
University.
Simons, K. 1990. New Technologies in Simulation Games. In D.F. Anderson, G. Richardson &
J.D. Sterman (Eds.), System Dynamics '90 (pp.1047-1058). Lincoln: The System Dynamics
Society.
Senge, P & Sterman, J.D. 1994. In Morecroft, J.D.W & Sterman, J.D (Eds), Modeling For
Learning organizations (pp. 195-219) Portland, Productivity Press.
913
®
CC BY-NC-SA 4.0