APPLYING MULTI-AGENT SYSTEMS TO ORGANIZATIONAL
MODELLING IN INDUSTRIAL ENVIRONMENTS
M. C. Romero
1
, R. M. Crowder
2
, Y. W. Sim
2
and T. R. Payne
2
1
Departamento de Tecnología Electrónica, University of Seville, Spain
2
School of Electronic and Computer Science, University of Southampton, U.K.
Keywords: Simulation, multi-agent systems, artificial social systems, engineering design, integrated product teams.
Abstract: This paper considers an agent-based approach to organizational modelling within the engineering design
domain. The interactions between individual designers within a design teams has a significant impact upon
how well a task can be performed, and hence the quality of the resultant product, hence many organisations
wish to model, and hence fully understand the process. Using multi-agent social modelling, designers and
the design task attributes can be the subject of rules implying how well tasks can be performed given
different levels of these attributes. In this paper we discuss the background to the work and the
identifications of individual, and team variables.
1 INTRODUCTION
This paper reports an approach to organisational
modelling within the engineering design domain.
While the use of software to undertake simulations
within the engineering design process, for example
computational fluid dynamics or finite element
analysis, is well known, modelling of the process
itself is less well understood. Our current research
objectives are:
To model the organisational processes as
applied to engineering design, by bringing
together expertise in organizational practice,
agent modelling and organisational or work
psychology.
Undertake simulations to address specific
problems within a design organisation.
In order to achieve these aims we need to
address two fundamental questions, firstly the
integration and application of a number of disparate
technologies to a demanding real world problem.
The second is extending simulation and modelling to
organisational systems, by exploiting intelligent
agent technology and the work psychologists’
understanding of the operation of individuals and
organisations. This work adopts a socio-technical
approach, combining expertise in technical and
social issues
It is widely understood that when a new design
problem emerges, the designer's knowledge related
to their previous experiences of similar problems is
applied. (Adler, Davis et al. 1989) comments that
more experienced designers are able to connect a
resolved design problem to a new problem quicker
and easier than less experienced designers. Central
to effective knowledge management is the integrated
product team or design team, where the
characteristics of the participants and their
relationships are critical. This includes their informal
network of contacts, personal experiences and the
designer’s own memory. It has been shown that
approximately 20% of the engineers time is spent
searching for and absorbing information, of which
40% will be from personal resources, even when
information is available elsewhere in the
organization (McMahon, Lowe et al. 2004)
(Shadbolt and Milton 1999).
In this work we consider the engineering design
environment, where both designers and tasks have
particular attributes. A large body of psychological
research has demonstrated that interactions between
humans and tasks have a large impact upon how
well a task can be performed. Within a computer
modelling environment, designers and task attributes
(e.g. task complexity, expertise, trust) can be the
subject of rules implying how well tasks can be
performed given different levels of these attributes
181
C. Romero M., M. Crowder R., W. Sim Y. and R. Payne T. (2008).
APPLYING MULTI-AGENT SYSTEMS TO ORGANIZATIONAL MODELLING IN INDUSTRIAL ENVIRONMENTS.
In Proceedings of the Tenth International Conference on Enterprise Information Systems - AIDSS, pages 181-186
DOI: 10.5220/0001683101810186
Copyright
c
SciTePress
(Martinez-Miranda, Aldea et al. 2003). For instance,
if a particular task is complex, then the designer may
need additional knowledge or expertise to complete
the task.
2 RELATED WORK
Several approaches to modelling engineering design
teams and IPTs have been reported in the literature
to date. The GRAI-Engineering approach models the
structure of the co-ordinated decision and design
activities, and is based on systems, hierarchy and
activity theory, but does not consider social
behaviour within teams (Girard and Doumeingts
2004). TEAKS (Martínez-Miranda, Aldea et al.
2006) is reported to take a multi-agent systems
approach for modelling the performance of a design
team, and hence facilitates optimization. The
variables with TEAKS are based on the PECS
(Physical condition, Emotive state, Cognitive
capabilities and Social status) reference model of
human behaviour (Schmidt 2002)
Given the characteristics of multi-Agent Systems
(Wooldridge and Jennings 1995), they can be seen
as a very useful tool for modelling human behaviour,
and in particular, social behaviour. The use of multi-
agent systems has been explored to support human
teams (Payne, Sycara et al. 2000), where agents
were used to provide support to team members given
a time-critical task, by aggregating relevant
information from their peers about other member
actions. Likewise, social dynamics have been
studied through modelling human and group
behaviour using multi-agent simulation methods
(Tsvetovat and K.Carley 2004). The agent-based
approach can enhance the potential of decentralised
computer simulation as a tool for theorizing about
social scientific issues, since it facilitates the
modelling of artificial societies of autonomous
intelligent agent.
Jennings (Jennings 2000) proposed the typical
structure of a multi-agent system (Figure 1). The
system contains a numbers of agents, which interact
with one another through communication. The
agents are able to act in an environment; different
agents have different “spheres of influence”, in the
sense that they will have control over different parts
of the environment. These spheres of influence may
coincide in some cases. The fact that these spheres
of influence may coincide may give rise to
dependency relationships between agents. When
faced with what appears to be a multi-agent domain,
it is critically important to understand the type of
interaction that takes place between the agents. In
order to clarify the interaction between
Figure 1: Canonical view of an agent-based system
(Jennings, 2000).
agents, (Jennings et al, 1998) distinguish between
cooperative models and self-interested models. In
the first type, agents cooperate to achieve a common
goal and in the second one agents negotiate in order
to achieve its own goal as best as possible.
Negotiation is seen as a method for coordination
and conflict resolution (e.g., resolving goal
disparities in planning, resolving constraints in
resource allocation, resolving task inconsistencies in
determining organizational structure). Negotiation
has also been used as a metaphor for communication
of plan changes, task allocation, or centralized
resolution of constraint violations. Hence,
negotiation is almost as ill-defined as the notion of
agent. (Jennings, Sycara et al. 1998) give what we
consider to be the main characteristics of
negotiation, which are necessary for developing
applications in the real world. These are: (a) the
presence of some form of conflict that must be
resolved in a decentralized manner, by (b) self-
interested agents, under conditions of (c) bounded
rationality, and (d) incomplete information.
Furthermore, the agents communicate and iteratively
exchange proposals and counter-proposals.
Team working processes has been extensively
studied by psychologists (Guzzo and Dickinson
1996). In a review of the research literature
(Applebaum and Blatt 1994), team working was
shown to offer organizations many advantages over
individual working and was associated with
organizational efficiency and improved quality.
However, there is widespread acceptance that
effective team-working does not result from
management; for example simply putting a group of
individuals together and expecting them to function
well as a team is rarely effective (Guzzo and
Dickinson 1996). The team's performance depends
on a variety of factors and processes concerning the
characteristics of the individual team members (e.g.
motivations, ability) and also the way the team
ICEIS 2008 - International Conference on Enterprise Information Systems
182
interacts and works together to achieve the team's
goals (e.g. communication processes, trust, shared
understanding). As organisations continue to
recognise the benefits that teams bring to their
business, researchers are becoming increasingly in
team processes such as decision making, social
loafing, minority influence, polarization of views,
leadership, and the stages of team development.
3 DESIGN TEAM SIMULATION
When a new design problem appears, the designers’
knowledge related to previous experiences in similar
problem is readily applied. That is, the process
begins by using the designers’ knowledge of a
similar or related problem, if this approach is
possible; undertake the correct process to resolve the
new problem. Adler (Adler, Davis et al. 1989) states
that more experienced designers are able to connect
a resolved design problem to another new problem
quicker and easier. Information about how the
problems are resolved forms part of the captured
design rationale which in general, is available to
others designers who are working in related areas
(Shadbolt and Milton 1999). The process is shown in
the Figure 2.
Kno wledge
base
Design
process
Design Problem
Desig n solution
People to supply new knowledge
Kno wledge
base
Design
process
Design Problem
Desig n solution
People to supply new knowledge
Figure 2: Generic design process.
Generally, a designer follows a numbers of
steps in order to resolve the design problem. We
have to undertake detailed studies on any design
process in order to extract the rules in a design
process.
3.1 Defining the Model
We have seen also there exist a number of
technologies in industrial design sector to undertake
design projects with different complexity. Some of
those methodologies have been matured along many
years and they are defined perfectly. Although these
methodologies aren’t used to model behaviour
designer in a specific design but themselves designs,
they can help to profile how designers operate from
a general viewpoint. In order to know this behaviour
from an organisational viewpoint it’s necessary to
acquire information directly from working
designers. It is anticipated that a considerable
proportion of work will involve interviewing
designers to obtain a range of metrics.
The necessary information not only includes the
steps that designers carry out during design process,
even the interactions between different designers.
It’s in this point where social sciences come into
play. Social sciences can be applied in multi-agent
based systems modelling and simulation. (Davidsson
2002) describes a Computer Science view of agent
based social simulation (ABSS) whose intersections
are shown in Figure 3.
Social
science
A
gent
based
computing
Computer
simulation
SAAS
SocSim
A
BBS
MABS
Social
science
A
gent
based
computing
Computer
simulation
SAAS
SocSim
A
BBS
MABS
Figure 3: The intersections of the three areas defining
ABSS (Davidsson 2002).
In hierarchical model of information systems,
each information agent is responsible for providing
information about a specific domain. Information
agents further down the hierarchy provide more
specialized information about a domain. In response
to a query, an information agent may cooperate with
information agents in other domains or sub-domains,
in order to generate a response. Communication
network solutions are based on a hierarchy of
autonomous intelligent agents, which have local
decision making capabilities, but cooperate to
resolve conflicts. Higher level agents arbitrate
unresolved disputes between peer agents.
The methodology for the development of agent
societies based on this framework consists of several
levels:
Designing coordination model.
Defining environment in terms of global
requirements and domains
APPLYING MULTI-AGENT SYSTEMS TO ORGANIZATIONAL MODELLING IN INDUSTRIAL ENVIRONMENTS
183
Describing behaviour in terms of agent
roles and interaction patterns.
Defining internal structure of agents in
terms of requirements for communication,
action, interface, and reasoning behaviour.
On the other hand (Norman, Jennings et al. 1997)
propose an architecture for the business process
management that can be applied to any hierarchical
social architecture in which interaction between
different agents organized in a specific way is
necessary Figure 4)
Figure 4: Designing an agent-based business process
management system (Norman, Jennings et al. 1997).
A logical hierarchy of agencies to represent the
hierarchical interactions between human in a
business environment, consisting of:
Agency
Responsible agents
Tasks
Sub-agencies
An Agency is recursively defined: an agency
consists of a single responsible (or controlling)
agent, a –possible empty– set of tasks that the
responsible agent can perform, and a –possible
empty– set of sub-agencies (see Figure 5). The
responsible agent represents the interests of the
agency to its peers. Any communication with an
agency must go through the responsible agent. A
sub-agency typically behaves in a cooperative
manner towards its responsible agent, this agent
being responsible for representing the interests of the
agency in the wider community. This relationship
between sub-agency and responsible agent can be
viewed as a type of social commitment, and provides
a mechanism for the encapsulation and abstraction
of services.
However, in these communications, trust,
influences the relationship that exists between them.
Trust is a fundamental concern in large-scale open
distributed systems and has resulted in a
considerable amount of research and hence models
as discussed in (Huynh, Jennings et al. 2004). Trust
lies at the core of all interactions between the entities
that have to operate in such uncertain and constantly
changing environments. Trust can be defined as a
belief an agent has that the other party will do what
it says it will (being honest and reliable) or
reciprocate (being reciprocative for the common
good of both), given an opportunity to defect to get
higher payoffs.
Responsible
Agent
Responsible
Agent
Communication
and negotiation
TE1
TE2
Agency E
TD1
TD2
Agency F Agency F
Agency D
Communication
and negotiation
Agency B Agency C
Agency A
Responsible
Agent
Responsible
Agent
Communication
and negotiation
TE1
TE2
Agency E
TD1
TD2
Agency F Agency F
Agency D
Communication
and negotiation
Agency B Agency C
Agency A
Figure 5: The logical hierarchy of agencies (Norman,
Jennings et al. 1997).
The core to the Trust model is the development
of ratings upon which a decision regarding
trustworthiness is based. The ratings are normally
based on a number of metrics, for example
interaction trust, role-based trust, witness reputation,
and certified reputation. It is therefore clear that one
of the key activities within modelling the
engineering design process is to identify the
individual metrics within the concept of engineering
design, and they determine the trust rating.
3.2 Variable
A large number of variables that characterise an
engineering design environment should be
considered, including:
The task's complexity;
The designer's ability;
The frequency and content of the
communication to and from designers.
Psychological variables, including shared
understanding, trust, and motivation.
Balancing the relative importance of these variables
used within the model is fundamental to developing
a realistic simulation of design teams. Given the
difficulty of accurately quantifying such variables,
our initial approach assumes a finite, qualitative set
of descriptors (e.g. high...medium…low). The
weightings for each of the variables are adjusted on
the basis of the information collected during
interviews within actual design teams.
Agency Agency
Communication
and negotiation
Design
Business
Process
Description
Agency
Agency
ICEIS 2008 - International Conference on Enterprise Information Systems
184
3.3 A Model of a Design Team
The design activity model used is shown in Figure 6.
The model for the individual design activity is based
on the IDED0 approach. (O'Donnell and Duffy
2002), where a number of knowledge sources are
used to provide the knowledge necessary to satisfy
the design requirements, which in this model is
defined by a goal, K
O
.
Figure 6: The design activity model.
As the teams are multi-disciplinary, it is
proposed that the task, resource, input and output
knowledge are held as vector of core competencies
also as a finite, qualitative set of descriptors. In order
for the individual activity to proceed, it operates
under a set of knowledge constraints, namely:
K
I
is the internal knowledge of the designer
and can either be explicit or tacit, remain
either constant during the task.
K
O
, knowledge generated as the design task
proceeds, equal K
G
at the completion..
G = {G
T
, G
S
} is the constraints either
technical G
T
or social G
S
that directs and
determines the design activity.
R, the external knowledge resources
available to the designer
Within the model a set of algorithms are used to
relate the knowledge input to the knowledge output
as a function of time. In order to simulate the
interaction between designer and a second designer
or a resources the communication paths variables are
dependant on a number of parameters including the
communication medium and context, which in term
determines the trust that the designer places on the
knowledge received. The current implementation
assumes that all requests are responded to
immediately without question.
Currently, the design activity function is linear,
but will be modified as we complete a number of
interviews with members of design teams within
manufacturing companies.
As a minimum, a team (by definition) must
consist of at least two people (with a maximum of
around 20 people), Figure 7 illustrates a two member
simplified integrated product team (IPT), working
on a single task.
Figure 7: A simplified integrated product team, showing
two designers working on a single task.
4 IMPLEMENTATION
A multi-agent system (MAS) simulation of a number
of designers has been developed using the JADE
platform. Currently, there are six designer agents,
two resource agents and a single task manager living
in the MAS, the required states and behaviours are
defined in Table 1.
Table 1: Agents’ state and behaviour.
Agents State Behaviour
Designer
Ability
Motivation
Performs task
assigned by
TaskManager
If ability is less than
task complexity, get
information from
Resource
Resource
Ability
Motivation
Ability is high
Respond with
information
TaskManager
Task progress
Task complexity
Assign tasks to agents
Keep track of task
progress
Each designer and resource agent has two
variables (ability and motivation), the values are
assigned at the start of the simulation. The task
manager agent is responsible for assigning task to
APPLYING MULTI-AGENT SYSTEMS TO ORGANIZATIONAL MODELLING IN INDUSTRIAL ENVIRONMENTS
185
the designer agents and monitors their work
progress. The design process is assumed to be in
sequential at the present stage of development phase,
i.e. designer 2 will only start work after designer 1
had completed their task (see Table 2).
Table 2: Results for a six person IPT.
Designer Ability Designer Motivation
Time to
complete
High High 18
Medium Medium 28
Low Low 52
Low Medium 36
Low High 28
Table 2 shows the results for a six person IPT
where the completion time is in arbitrary units. The
task complexity is high and the communication rules
assume that every request is handled on receipt.
5 CONCLUSIONS
This paper has reported the initial approach to the
modelling of IPTs within large engineering
organizations. Within this model we have
implemented a knowledge-time relationship which is
currently considered to be linear, but further work is
being carried out to be to optimize the algorithms
within the design activity model. The initial
feedback from our industrial partners indicates that
this is an acceptable approach for modelling.
ACKNOWLEDGEMENTS
The work described in this paper has been funded by
the DTI (Department of Trade and Industry) and
Rolls Royce under the HIPARSYS grant:
TP/3/DSM/6/I/16032. The authors acknowledge the
contribution from Mark Robinson and Helen
Jackson for the Leeds Business School (University
of Leeds) and from the Ministerio de Ciencia y
Tecnología (Spain) within I+D+I National Program
through the project with reference number
TEC2006-08430.
REFERENCES
Adler, M., A. Davis, et al. (1989). Conflict resolution
strategies for non-hierarchical distributed agents.
Distributed Artificial Intelligence. L. Gasser and M.
Huhns, Pitman Publishing. II: 139-16.
Davidsson, P. (2002). Agent Based Social Simulation: A
Computer Science View. Journal of Artificial
Societies and Social Simulation 5(1).
Girard, P. and G. Doumeingts (2004). GRAI-Engineering:
a method to model, design and run engineering design
departments. International Journal of Computer
Integrated Manufacturing 17: 716-732.
Huynh, T. et al. (2004). FIRE: An Integrated Trust and
Reputation Model for Open Multi-Agent Systems.
16th European Conference on Artificial Intelligence.
Jennings, N. (2000). On Agent-Based Software
Engineering. Artificial Intelligence 117(2): 277-296.
Jennings, N. et al. (1998). A Roadmap of Agent Research
and Development. International Journal of
Autonomous Agents and Multi-Agent Systems 1(1):
7-38.
Martínez-Miranda, J. et al. (2003). Simulation of work
teams using a multi-agent system. AAMAS ’03,
Melbourne, Australia.
Martínez-Miranda, J. et al. (2006). TEAKS: simulation of
human performance at work to support team
configuration. 5th international joint conference on
Autonomous Agents and Multiagent Systems,
AAMAS06, Hakodate, Japan.
McMahon, C., A. Lowe, et al. (2004). Knowledge
mangement in engineering design : personalization
and codification. Journal of Engineering Design 15(4):
307--325.
Norman, T. et al. (1997). Designing an Implementing a
Multi-Agent Architecture for Business Process
Management. ECAI'96 Workshop Agent Theories,
Architectures, and Languages: Intelligent Agents.
O'Donnell, F. and A. Duffy (2002). Modelling design
development performance. International Journal of
Operations & Production Management 22(11): 1198-
1221.
Payne, T. et al. (2000). Agent-Based Team Aiding in a
Time Critical Task. HICSS '00: Hawaii International
Conference on System Science.
Schmidt, B. (2002). Modelling of Human Behaviour The
PECS Reference Model. 14th European Simulations
Symposium.
Shadbolt, N. and N. Milton (1999). From knowledge
engineering to knowledge management. British
Journal of Management 10: 309-322.
Tsvetovat, M. and K.Carley (2004). Modelling Complex
Socio-technical Systems using Multi-Agent
SimulationMethods. Kunstliche Intelligenz (Artificial
Intelligence Journal) Special Issue on Applications of
Intelligent Agents (2).
Wooldridge, M. and N. Jennings (1995). Intelligent
Agents: Theory and Practice. The Knowledge
Engineering Review 10(2).
ICEIS 2008 - International Conference on Enterprise Information Systems
186