Towards Multi-Level Structuring of Goal-Oriented Models for Improved
Model-Based Systems Engineering
Christophe Ponsard
1
and Robert Darimont
2
1
CETIC Research Centre, Charleroi, Belgium
2
Respect-IT SA, Louvain-la-Neuve, Belgium
Keywords:
Systems Engineering, MBSE, Goal-Oriented Modelling, Requirements, Refinements.
Abstract:
Model-driven methods are gaining momentum in the industry to support the development of software inten-
sive systems. While most methods rely on progressive levels of functional decomposition, the modelling of
requirements tends to stay quite monolithic in nature. This position paper analyses how to better structure
requirements across multiple levels in the scope of goal-oriented requirements engineering to reconcile it with
layered system models and to provide better support for decomposition and analysis of inner sub-components
or the outer system of systems. The proposed ideas are discussed based on a classical case study of the meeting
scheduler with some early prototyping using a meta-case requirements engineering platform.
1 INTRODUCTION
Model-based systems engineering (MBSE) is defined
as “the formalized application of modeling to sup-
port system requirements, design, analysis, verifica-
tion and validation activities beginning in the concep-
tual design phase and continuing throughout devel-
opment and later life cycle phases” (INCOSE, 2007).
Moving to MBSE is an important paradigm shift for
the industry because the system model replaces doc-
uments for exchanging between engineers (Michot
et al., 2018)(Madni and Purohit, 2019).
Many MBSE methodologies have developed over
the past decade (OMG, 2018)(Weilkiens and Mao,
2022). Most rely on long-established proven Systems
Engineering Methodologies (SEM). In order to cope
with complexity, they support progressive system de-
composition in different layers of abstraction. This
decomposition is applied iteratively until single pur-
pose components (or design entities) can be identi-
fied. Such design entities are structurally and func-
tionally distinct from other elements (DOE, 2002).
MBSE methodologies define standard layers with
specific purposes. For example Arcadia, proposed by
Thales and depicted in Figure 1, defines levels which
first cover the problem domain with operational anal-
ysis (user perspective) and system needs (both func-
tional and non-functional) and then the solution do-
main with the logical architecture, physical architec-
ture and end-product breakdown structure.
Figure 1: Arcadia MBSE Methodology.
In the same spirit, ASAP (Advanced System Ar-
chitecture Program by Alstom) is organised in 3 lay-
ers: (1) the operational level focuses on users to iden-
tify their needs and the external interfaces (WHY), (2)
the functional level defines WHAT the system does to
fulfill its mission and identifies functional flows, and
(3) the constructional level structures into intercon-
nected subsystems and their behaviour (HOW).
The elicitation of requirements is a key activity
of any system development. It is well known that
many projects are still failing mainly due to poor Re-
quirement Engineering (RE) practices (Hughes et al.,
2015). Different methods can ensure key properties
of completeness, accuracy, non-ambiguity and testa-
bility. They have also evolved from lightweight text-
Ponsard, C. and Darimont, R.
Towards Multi-Level Structuring of Goal-Oriented Models for Improved Model-Based Systems Engineering.
DOI: 10.5220/0011866200003402
In Proceedings of the 11th International Conference on Model-Based Software and Systems Engineering (MODELSWARD 2023), pages 269-274
ISBN: 978-989-758-633-0; ISSN: 2184-4348
Copyright
c
2023 by SCITEPRESS Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
269
based to model-based techniques. In the scope of this
paper, we will consider Goal-Oriented Requirements
Engineering (GORE) which aims at capturing, at dif-
ferent levels of abstraction, the various properties the
system under consideration should achieve. More
specifically we will use KAOS (van Lamsweerde,
2009) which was applied on large systems (Ponsard
and Darimont, 2017). However other variants can
also be considered, e.g. i* (Yu and Mylopoulos,
1997), GRL (ITU, 2012) or GSN (ACWG, 2018).
Although supporting abstraction and refinement,
such methods do not define specific layers, leaving
this task to the analyst which may rely on a SEM
for this. A common flaw is the limited ability to re-
flect layers inside a GORE model. It usually relies on
rather syntactic containers such as package and dia-
grams and lacks semantic structure providing the abil-
ity to define a clean interface between layers. Figure
2 shows the inner structure of a KAOS model com-
posed of 225 goals described through 45 diagrams.
The structure shows more than 10 refinement levels
but such a view does not really make explicit connec-
tions with design entities as those are captured by an-
other concept: the agent. Agent allocation in KAOS
can only be specified at the leaves and lacks the ability
to capture system-level agent. Other languages like i*
allow this but without decomposition.
In the scope of this position paper, we are in-
terested to investigate how to better structure GORE
models by adopting a decomposition structure which
is closer to the a typical SEM structure. This problem
has not received much attention in the literature but is
useful both in designing new system and for upgrad-
ing existing systems. For the former, the process can
help identify a good structure and lay the foundation
of a good architecture. For the latter, it helps in cap-
turing the requirements at the right layer and identi-
fying traceability impacts both to higher system level
or down to specific components. We contribute here
an extension to the KAOS meta-model with a more
semantic decomposition of a model in interconnected
sub-models. Our approach is experimented and dis-
cussed on the classical meeting scheduler case study
(van Lamsweerde et al., 1995) which is considered
both in a wider scope (enterprise business) and more
focused scope (specific components).
This paper is structured as follows. First, Section
2 reminds about the existing KAOS meta-model and
introduces our meeting scheduler case study. Section
3 explains our extension. Section 4 applies it to revisit
our case study. Our current results are then discussed
in Section 5. Finally, Section 6 draws some conclu-
sions and identifies our future work.
2 KAOS META-MODEL AND
CASE STUDY
Goal is the core concept of a GORE model. Goals
prescribe, at different levels of abstraction, key prop-
erties the considered system should achieve. KAOS
uses several abstraction levels to express goals start-
ing from high-level strategic goals. In the meet-
ing scheduler case study (van Lamsweerde et al.,
1995) used as running example, it can be a goal like
Achieve[EfficientSchedulingOfMeetings] as depicted
in Figure 3 as light blue parallelograms. In KAOS,
high-level goals are progressively refined into more
concrete and operational ones through refinement re-
lationships depicted as yellow circles. A refinement
relationship links a parent goal to several subgoals
with different fulfilment conditions using either AND-
refinement (all subgoals need to be satisfied, e.g. the
EfficientSchedulingOfMeetings requires to achieve
four sub-goals:TimeTableCollected, PotentialRoom-
sIdentified, MeetinSchedulingBasedOnAvailableData
and MeetingInformationCommunicated), or OR-
refinement / alternatives (a single subgoal is enough,
e.g. Achieve[TimetablesCollected] can be either
ManualCollection or AutomatedCollection). The
“WHY” and “HOW” questions can be used to conve-
niently navigate from subgoals to parent goals (why)
Figure 2: Typical inner structure of a KAOS model.
MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering
270
Figure 3: Monolithic goal graph for the meeting scheduler.
and from parent goals to subgoals (how). Goal refine-
ment ends when every subgoal is realizable by some
individual agent (depicted as hexagon) which will be
assigned responsibility (through relationship with a
red circle). Such a leaf goal is called a requirement
if the agent is in the system under design. It is graphi-
cally differentiated from goals through a thick border.
It is an expectation if it is under responsibility of an
agent in the environment and then depicted in yellow.
Those concepts are summarised in the top part of
Figure 4 depicting goal and agent parts of the KAOS
meta-model. In addition, the meta-model also sup-
ports the description of the domain through an ob-
ject model and the functionalities through an oper-
ation model. Those will be less central in this pa-
per and will be discusses in Section 5. Note the goal
model also supports the reasoning about obstacles to
goal satisfaction, like safety or security risks.
Figure 4: KAOS meta-model with proposed extension.
3 PROPOSED EXTENSION
The goal model presented in Figure 3 shows a com-
plete refinement of the top-level goal into different
requirements and expectations under the responsibil-
ity of 3 agents to be designed and 3 agents of the
environment. Even small, this case study highlights
limitations of the approach. A first observation is
that part of the goal tree is under the control of a
single agent (e.g. Achieve[PotentialRoomIdentified])
and one might wish to assign the responsibility higher
in the tree which is not allowed. Another obser-
vation is that other parts require collaboration be-
tween agents but often with a single agent under
design, while the others are environment agents,
e.g. Achieve[AutomatedCollection]. Again here one
might consider this agent as central while the oth-
ers can be identified as interacting with it from the
outer context. In both cases, the result is that a higher
level goals might be considered as a requirement in
a higher-level description of the system which still
needs to be refined to be fully realizable. In the pro-
cess, finer grained agents purely internal to an agent
might also be identified, e.g. for the RoomManager.
In order to enable such system decomposition,
we make explicit the notion of System in the meta-
model. It is depicted in darker in Figure 4. It is intro-
duced as a generalisation of Agent which will allow
all agents to be considered as system, thus having a
system dimension while being less constrained, e.g.
only an agent can carry a responsibility. For a sys-
tem, a more abstract relationship named allocation is
introduced. A system is also an aggregation of finer-
grained agents which is captured by the aggregation
Towards Multi-Level Structuring of Goal-Oriented Models for Improved Model-Based Systems Engineering
271
relationship. This enables to capture hierarchical sys-
tem breakdown structure and to ensure the alignment
with the breakdown of a system model. Note also that
the goal refinement must be consistent with this struc-
ture through some meta-constraint.
Our extension is supported by the KAOS Objec-
tiver toolset (Respect-IT, 2005), a meta-case tool with
multiple levels from a generic meta-meta level (M0)
down to the instance level (M3). The meta-model
(M1) is updated through a plugin inserting the extra
meta-concepts and meta-relationships.
4 EXPERIMENT ON THE CASE
STUDY
Applying our extension on the meeting scheduler
yields the model depicted in Figure 5. It is much sim-
pler to understand than the complete model of Figure
3. Of course such structuring becomes fully meaning-
ful for large system as depicted in Figure 2.
Figure 5: Meeting Scheduler with system refinement.
Of course, it does not include all the information
as some agents now have to be refined through sub-
systems. For example, the subsystem to manage the
available rooms is depicted in Figure 6. Note also the
top-level goal has been tagged with an explicit agent
representing the corresponding system level.
Figure 6: Room management subsystem.
Based on this, it is now easy to extract the agent
breakdown structure of Figure 7.
Figure 7: System breakdown structure.
Considering our meeting scheduler system, the
enforced goal contributes to the fulfilment of goals of
a larger enterprise system, e.g. for project manage-
ment or internal collaboration, as depicted in Figure
8.
Figure 8: Organisation level system.
Analysing this system reveals an interesting point:
the agenda agent which was identified as subsystem
of the meeting scheduler seems more general and
also used to support collaboration and not only for
scheduling meetings which looks more a secondary
usage. Analysing this reveals the agenda is not a sub-
component but rather an external component which
is collaborating with the meeting scheduler through a
specific interface in order to collect availabilities. Ex-
tracting the breakdown structure again gives a more
global picture of the system as shown in Figure 9.
Figure 9: Enterprise level breakdown structure.
MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering
272
5 DISCUSSION ABOUT RELATED
WORK
As stated earlier, many GORE frameworks have lim-
ited support for system decomposition. However, re-
searchers have investigated this problem. First Jack-
son problem frames (Jackson, 2001) is based on a pro-
cess of parallel - not hierarchical - decomposition of
user requirements and result in the analysis of many
subproblems that are then combined to form a solu-
tion design. To better support this essentially semi-
formal approach, a meta-model was proposed to cap-
ture the notion of sub-problems but also sub-domains
and to represent problems in intermediate situations
during the (de)composition phases (Lavazza et al.,
2010). Although it eases the analysis, the process
does not result in a breakdown structure that can be
applied to MBSE as in our proposal.
Another approach is DeSyRe which systemati-
cally decomposes system requirements into subsys-
tem requirements according to a given system archi-
tecture and with focus on informal or semi-formal re-
quirements (Penzenstadler, 2011). It provides guid-
ance to derive subsystem requirements from system
requirements by use of assumption/guarantee rea-
soning and decomposition patterns. This can com-
plement standard goal-decomposition patterns (Dari-
mont and van Lamsweerde, 1996) in order to define
subcomponent requirements.
More recent work has focused on more formal ap-
proaches for industrial system (Teufl et al., 2014). It
has shown the benefits of using a two-step process for
refining system requirements to subsystem require-
ments: interface refinement on the system boundaries,
and a decomposition of system requirements to sub-
system requirements. This can be complemented with
formal analysis and verification techniques on the re-
finement of requirements. So far, our work has con-
sidered more the decomposition step but the need to
reason about interfaces was identified in our struc-
turing approach. For this purpose, the KAOS agent
model can capture monitor and control flows across
agents and can enable this kind of reasoning.
Globally our approach is still semi-formal and re-
lies on a rich meta-model with a focus on properties
through goals and agents which can capture compo-
nents under design but also existing parts or human
agents interacting with the system. The object and op-
eration model are less relevant as they are better man-
aged by MBSE using notations like SysML (OMG,
2005). For example, a class diagrams can describe
domain models while an internal block diagrams can
capture system functions. Such models can then be
synchronised with the GORE model for enriching it if
necessary. In order to be consistent, the meta-model
also need to enforce a few meta-constraints related to
our extension. So far, we could identify the following
ones:
Root goals in an agent subsystem decomposition
must match all requirements allocated to corre-
sponding agent in parent system.
An agent part of a subsystem must be responsible
of requirements in that subystem.
A system is only composed of agents allocated to
requirements resulting from alternatives specific
to that system and not refining external functional
goals.
Our meeting scheduler case study contains OR-
refinement but we did not provided any specific treat-
ment for them. They can result in alternative agents
and produce a richer system breakdown structure that
can evolve towards configuration diagrams. The cou-
pling between goal refinements and agent structure
can provide a good bridge between those two different
ways of capturing variability. However this requires
more work. Along the same line, we have also limited
to functional requirements and not considered possi-
ble conflicts among them.
6 CONCLUSION AND
PERSPECTIVES
This paper explored how to provide better support for
decomposing GORE models across multiple levels of
abstraction as an MBSE context with a focus on goals
and agents structuring. We proposed a meta-model
extension which enables to really capture systems and
subsystems as part of a model and beyond the mere
use of purely syntactic container such as packages and
diagrams. We also identified some meta-constraints
that ensure model consistency. We illustrated on a
meeting scheduler case study ranging on 3 levels of
system refinement. Finally, we discussed our current
result in the light of some related work to highlight
the challenges ahead.
Our future work is to progress with the validation
on more complex case studies with industrial partners,
in the context of the modernisation of a complex rail-
way system. At the tool level, we plan to improve
the management of layers through the ability to navi-
gate across layer and also focus on a level by tempo-
rary hiding the other ones until the analyst wishes to
switch again to a more global view. We also plan to
enrich the analysis capabilities for dealing with refine-
ment and to deploy a model synchronisation bridge.
Towards Multi-Level Structuring of Goal-Oriented Models for Improved Model-Based Systems Engineering
273
This will enable the refinement of subsystem require-
ments both from the GORE and MBSE environments
and, from there, to assess what are the best scenarios
and practices.
ACKNOWLEDGEMENTS
This work was partly supported by the MORSE
project (nr 8389) under FEDER funding for Wallo-
nia. Thanks to our industrial partners for their inputs
in GORE and MBSE for exploring system decompo-
sition in both fields.
REFERENCES
ACWG (2018). Goal Structuring Notation Community
Standard, Version 2. The Assurance Case Working
Group https://scsc.uk/r141B:1?t=1.
Darimont, R. and van Lamsweerde, A. (1996). Formal re-
finement patterns for goal-driven requirements elabo-
ration. In Proc. of the Fourth ACM SIGSOFT Sympo-
sium on Foundations of Software Engineering, SIG-
SOFT1996, San Francisco, California, USA, October
16-18. ACM.
DOE (2002). Systems Engineering Methodology, Version
3. US Department of Energy DOE G 200.1-1A.
Hughes, D. et al. (2015). Success and Failure of IS/IT
Projects: A State of the Art Analysis and Future
Directions. SpringerBriefs in Information Systems.
Springer International Publishing.
INCOSE (2007). SYSTEMS ENGINEERING VI-
SION 2020. http://www.ccose.org/media/upload/
SEVision2020\ 20071003\ v2\ 03.pdf.
ITU (2012). Recommendation Z.151 (10/12), User Re-
quirements Notation - Language Def. https://www.
itu.int/rec/T-REC-Z.151.
Jackson, M. (2001). Problem Frames: Analyzing and Struc-
turing Software Development Problems. Addison-
Wesley Longman Publishing Co., Inc.
Lavazza, L., Coen-Porisini, A., Colombo, P., and del
Bianco, V. (2010). A meta-model supporting the de-
composition of problem descriptions. pages 50–57.
Madni, A. M. and Purohit, S. (2019). Economic analysis of
model-based systems engineering. Systems, 7(1).
Michot, A., Ponsard, C., and Boucher, Q. (2018). Towards
better document to model synchronisation: Experi-
mentations with a proposed architecture. In Proc. of
the 6th Int. Conf. on Model-Driven Engineering and
Software Development, MODELSWARD 2018, Fun-
chal, Madeira - Portugal, January 22-24.
OMG (2005). System modeling language. http://www.omg.
org/spec/SysML.
OMG (2018). MBSE Methodologies and Metrics.
Penzenstadler, B. (2011). Exactly the information your sub-
contractor needs: Desyre - decomposing system re-
quirements.
Ponsard, C. and Darimont, R. (2017). Improving require-
ments engineering through goal-oriented models and
tools: Feedback from a large industrial deployment. In
Proc. of the 12th Int. Conf. on Software Technologies,
ICSOFT 2017, Madrid, Spain, July 24-26.
Respect-IT (2005). The Objectiver Requirements Engineer-
ing Tool.
http://www.respect-it.com.
Teufl, S., B
¨
ohm, W., and Pinger, R. (2014). Understand-
ing and closing the gap between requirements on sys-
tem and subsystem level. In IEEE 4th International
Model-Driven Requirements Engineering Workshop
(MoDRE).
van Lamsweerde, A. (2009). Requirements Engineering -
From System Goals to UML Models to Software Spec-
ifications. Wiley.
van Lamsweerde, A., Darimont, R., and Massonet, P.
(1995). Goal-directed elaboration of requirements for
a meeting scheduler: problems and lessons learnt. In
Proc. of IEEE Int. Symposium on Req. Eng. (RE’95).
Weilkiens, T. and Mao, M. D. (2022). MBSE Methodolo-
gies. https://mbse-methodologies.org.
Yu, E. and Mylopoulos, J. (1997). Enterprise modelling for
business redesign: The i* framework. SIGGROUP
Bull., 18(1):59–63.
MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering
274