ONTOLOGY FOR INTEGRATING HETEROGENEOUS TOOLS
FOR SUPERVISION, FAULT DETECTION AND DIAGNOSIS
Beatriz López, Joaquim Meléndez, Silvia Suárez
Universitat de Girona, Capus Montilivi, edifice P4, 17071 Girona, Spain
Keywords: Distributed control systems, Intelligent fault detection and identification, Industrial expert systems.
Abstract: The Distributed Supervision Systems that have been used extensively for the last fifteen years in the
process industry are now evolving towards higher level solutions based on better connections between
applications and processes that assure that data flows from the process to manage boards. Knowledge
sharing seems to be a key issue in integrating these heterogeneous systems. In this paper we present an
ontology as a first step to achieving semantic interoperability. The ontology has been conceived within the
context of a complex integration problem, in which heterogeneous toolboxes cooperate to deal with several
supervision, fault detection and diagnostic tasks for chemical processes. Regarding the current trends in
ontology research, our proposal is consistent with top-level ontologies, as these kinds of ontologies seem to
overcome the ontology integration problem. We describe a preliminary version of the ontology. The
conceptualisation of control variables, system behaviour, supervision tasks, models and system properties is
given. All attributes and relationships between each concept has been deployed. The ontology has been
developed using Protete2000.
1 INTRODUCTION
The Distributed Supervision Systems that havebeen
used extensively for the last fifteen years in the
process industry are now evolving towards higher
level solutions based on better connections between
applications and process that assure that data flows
from process to manage boards. Current
requirements of flexibility, traceability and quality
mean that all agents (suppliers, factories, vendors,
maintenance, etc.) that participate in the final
product must communicate with each other
continually. Figure 1 shows the basic architecture of
a SCADA (Supervisory Control and Data
Acquisition) software. It is clearly oriented towards
guaranteeing the integration of the process
(Instrumentation Communication Interface),
operators (HMI), supervisors (SPC/SQC), manager
and other enterprise resources (ERP).
Advances in distributed and ubiquitous
computing, networking and sensors provide new
environments in which it is possible to integrate
these supervision techniques (Murray et al, 2003).
However, integrating supervision techniques
presents the challenge of shifting from traditional
supervision systems as processes with single
controllers to supervision systems as collections of
heterogeneous physical and information systems
with complex inter-connections and interactions
(Murray et al, 2003; MacFarlane and Bussmann,
2000).
Holonic Multiagent systems seem to be a
promising paradigm for managing, modelling and
supporting integration (MacFarlane and Bussmann,
2000). They provide a common platform to facilitate
the information flow among different heterogeneous
systems in such a way that decision support systems
can be improved. The burden of information access,
extraction and interpretation in the different steps
that constitute supervising a plant is automated and
presented to the human operator in a holistic and
more comprehensive way.
However, using a Holonic Multiagent system
first requires the current supervision systems to be
encapsulated, and second it requires elaborating a
common, shared vocabulary that provides semantic
interoperability among the different systems. Our
goal here is to develop an ontology that provides
semantic interoperability. In particular, we focus on
integrating heterogeneous tools for supervision,
fault detection and diagnosis (SFDD) within the
125
López B., Meléndez J. and Suárez S. (2005).
ONTOLOGY FOR INTEGRATING HETEROGENEOUS TOOLS FOR SUPERVISION, FAULT DETECTION AND DIAGNOSIS.
In Proceedings of the Second International Conference on Informatics in Control, Automation and Robotics, pages 125-132
DOI: 10.5220/0001165301250132
Copyright
c
 SciTePress

context of the CHEM European project (Cauvin,
2002).
This paper is organized as follows: First, in
section 2 we explain what holonic multi-agent
systems are and in section 3 we make some ontology
definitions. In section 4, we provide the details of
the SFDD ontology. In section 5 we explain how
this ontology can be used to integrate heterogeneous
toolboxes and we end with some conclusions in
section 6.
2 HOLONIC MULTI-AGENT
SYSTEMS
Holonic Multi-Agent research concerns two main
communities: holonic manufacturing systems and
agent technology. On one hand, a holon is “an
autonomous and co-operative building block of a
manufacturing system for transforming,
transporting, storing physical and information
objects” (MacFarlane and Bussmann, 2000). It
consists in a control part and an optional physical
processing part. A holon can be made up of other
holons (MacFarlane and Bussmann, 2000; Giret and
Botti, 2004). This concept of holons is clearly an
extension of the current SCADA systems (See
figure 1) with improved communication and
processing capabilities clearly oriented towards
decision making.
On the other hand, agents provide autonomy with
respect to the system capacity for a given
environment (Wooldridge, 2002). Agent
Technology, although broadly extended in open
applications such as Internet services, has only
recently been introduced to the supervision field
(see (Bussmann and Schild, 2001) for an example).
Although both approaches, holons and agents,
share many basic concepts, research into each area
has mainly been developed independently: research
into holonic systems has focused on manufacturing
systems, and research into Agent Technology has
focussed on developing interconnected systems in
which data, control, expertise or resources are
distributed. Being aware of the common interest,
there have been recent efforts to understand whether
holons and agents are different or not and to join
both communities (Marik et al., 2003)). This has
lead to the term Holonic Multiagent Systems being
formed, a novel paradigm for managing, modelling
and supporting complex systems. This new
paradigm provides two main benefits. On one hand,
holons provide soundness and robustness, typical
characteristics of the system engineering
developments. On the other hand, agents facilitate
the integration of heterogeneous systems.
Our work is in line with this new approach for
integrating heterogeneous supervision, fault
detection and diagnosis systems. More than building
new SFDD techniques, we focus on integrating
them. As a first step, we have participated in
developing toolboxes that encapsulate and describe
SFDD techniques. Currently, we are dealing with
the problem of making interconnecting toolboxes
operational. In this challenge, one of the main
drawbacks consists in information sharing, and
therefore, ontologies play an essential role.
Instruments (PLC, DCS, Field Buses)
Instrument Communication Interface, OPC)
Process Data Server
Data Base management
Alarm and event generation.
Alarm and event management
Web server
OLE /
ODBC
HMI Batch
SPC
SQC
Prod. manager
(recipes)
Other
Aplic.
ERP, RDBMS
Standard Communication Intera
pp
lication
Internet
Figure 1: Basic structure of a SCADA system (from
(Issermann and Ballé, 1997))
3 ONTOLOGIES
In the past research into ontologies was rather
confined to the philosophical sphere. Currently it is
widespread in research fields as diverse as
knowledge representation, knowledge engineering,
qualitative modelling, database design, information
systems and database integration, natural language
understanding, information retrieval and extraction,
object-oriented software development, knowledge
management and organization, and agent-based
system development (Giunchiglia et al., 2003).
Several standardization organisms such as ISO,
IEEE, and W3C are now working on this new
technological challenge to integrate systems by
means of a common vocabulary.
Ontologies can be seen as metadata that explicitly
represent semantics of data in a machine processable
way (Giunchiglia et al., 2003). By making the link
between the information’s form and content explicit,
ontologies help people and computers to access the
information they need. Moreover, ontologies are
now recognized as powerful tools that enable
ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION
126

sharing knowledge, and a growing number of
applications have benefited from using ontologies as
a means of achieving semantic interoperability
among heterogeneous, distributed agent systems
(Sure and Corcho, 2003). They therefore have a
crucial role in integrating supervision techniques.
An ontology defines a common vocabulary for
researchers who need to share information in a
domain. A domain ontology corresponds to an
organized set of domain generic terms that can be
used to describe a particular domain by providing
machine-interpretable definitions of basic concepts
in the domain and the relationships between them
(Noy and McGuinness, 2001). An ontology of a
specific domain is useful in two aspects: first, to
make understanding the process in a specific domain
easier; and second, to obtain a standard
representation that can be shared and reused in other
tools. With the second point it is important to
highlight that different tools have been developed by
several designers and there is no common
vocabulary, so ontologies seem to be an appropriate
mechanism for integration.
Recent research work, however, has
experimentally proved that ontologies are not
enough to guarantee semantic interoperability. In
(Correa et al., 2002) four main problems have been
detected: 1) reusing ontologies to engineer new
ontologies is not straightforward; 2) ontologies do
not provide adequate information when sharing
inferences; 3) when reasoning under uncertainty,
additional semantic links regarding inference are
required, and 4) in a large scale system, sharing
group knowledge should be appropriately studied.
Regarding the first problem, Guarino observes
that ontologies developed from a bottom-up
approach based on multiple local ontologies, may
not work because they focus on conceptual relations
in a specific context (Guarino, 1998). Therefore,
there is no guarantee that two systems with the same
vocabulary have the same conceptualisation. This is
what he calls the ontology integration problem.
In order to deal with this problem, several
authors argue in favour of mapping mechanisms
between ontologies (Schorlemmer and Kalfoglou,
2003), while others, such as (Guarino, 1998),
propose using different kinds of ontologies. Guarino
distinguishes between top-level, domain, task and
application ontologies, as shown in figure 2. Top-
level ontologies provide constraints and building
blocks for representing knowledge (Martin and
Eklund, 1999). Domain level ontologies describe the
vocabulary related to a generic domain. Task
ontologies are related to generic tasks or activities.
Finally, application ontologies describe concepts
depending on a particular domain and task.
In this paper we propose a top-level ontology for
distributed supervision systems (supervision, fault
detection and diagnosis). Other works are related to
supervision but at the application level. In (Bernaras
et al., 1996), the authors present an ontology for
fault diagnosis in electrical networks. In (Kitamura
and
Mizoguchi, 1999) the authors provide an
ontological analysis at the task level regarding fault
processes. Another interesting work on ontologies is
in WEDSS, used to integrate rule-based systems,
case-based reasoning and classical control
techniques for wastewater management (Ceccaroni
et al., 2004).
Top-level ontology
Domain ontology Task ontology
Application ontology
Figure 2: Kinds of ontologies according to (Guarino,
1998)
4 SFDD ONTOLOGY
In order to elaborate an ontology for SFDD tasks,
we have used the terms proposed in (Isermann and
Ballé, 1997) and (Colomer and Meléndez, 2000) for
supervision, fault detection and diagnosis.
Therefore, the main terms are organized in variables,
system behaviours, supervisory tasks, models and
system properties (see Figure 3). Each term is
defined in properties and relations, generating a
complex network of classes, subclasses, instances
and slots. (See figure 4 for a detailed description of
the terms). In the following section we describe all
the terms.
Regarding the linking between ontologies and the
conceptual modelling of the overall system, our
ontology is a top-level one as stated above. So, first
the particular instantiation of models, system
variables and system behaviours will lead to a
domain ontology for a given modellization. Second,
ONTOLOGY FOR INTEGRATING HETEROGENEOUS TOOLS FOR SUPERVISION, FAULT DETECTION AND
DIAGNOSIS
127

Figure 3: General diagram of the ontology
the refinement of supervisory tasks will provide a
task ontology. And finally, the enhancement of
system properties will derive in an application
ontology.
4.1 Variables
The following variables have been considered for
SFDD purposes and characterised as follows:
• Signals are defined as physical measures or
perceptible variables that communicate
information and messages. Attributes: Units,
Range, Typology (quantitative/qualitative,
continuous/sampled/discrete events).
• Fault: Unpermitted deviation of at least one
characteristic property or parameter of the
system in acceptable / usual / standard
conditions. It is composed of the following
attributes: cause, duration, final_time,
typology_of_fault (intermittent/permanent,
evolutive/abrupt, additive/parametric),
location_fault, Descriptor_fault, Size_fault,
starting_time.
• Error: Deviation between a computed variable
(typically and output or state variable) and the
true, specified or theoretically correct value. It
has the following attributes: two inputs (to
compute it): correct_value, duration, final_time,
measured_computed_value, result,
starting_time
• Disturbance: An unknown (and uncontrolled)
input acting on a system. The attributes of this
subclass are: typology (additive /multiplicative,
etc), shape, duration final_time, input_point (if
known, related
with the physical system and/or the structural
model), effects and starting_time.
• Perturbation: An input acting on a system which
results in a temporary departure from the
current state. The attributes are:
idem_disturbance.
• Residual: Fault indicator, based on deviations
between measurements and model-equation-
based computations. Particular case of error. It
has the following attributes: deviation, duration,
final_time, result (fault detection decides about
the presence or absence of faults), starting_time,
decision_mechanism, Signature.
• Symptom: An observable quantity changes its
normal behaviour. The attributes of this
subclass are: referred_quantity, duration,
final_time, starting_time, shape (trends) and
values.
4.2 System Behaviour
System behaviour is an overall description of the
system operating conditions. Basic states can be
defined:
ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION
128

• Normal Operating: the system behaves
according to the specifications.
• Faulty or Malfunction: Intermittent irregularity
in fulfilling a system’s function. It has the
following attributes: periodicity, starting time,
final_time, faults (see fault attributes).
• False alarm: The system is operating properly
but the supervisory system has detected some
misbehaviour.
Fault_diagnosis
Description_process Instance Models
Time_detection Float
Diagnostic_location_fault String
Diagnostic_size_fault String
Fault_diagnostic Instance Malfunction
Diagnostic_typology_fault String
Figure 4: Description of the fault diagnosis attribute
4.3 Supervisory Tasks
The tasks are the different kinds of operations that
must be performed in order to supervise a given
system. There are seven main kinds of tasks:
• Fault detection: Detection of faults in a system
and the detection time of a fault. This subclass
is composed of the following attributes:
fault_presented (yes /no, without specifying
additional information) and time_detection.
• Fault isolation: Identification of the relevant
attributes of the faults present: kind, location
and detection time of a fault. This task follows
fault detection. The attributes of this subclass
are: fault_presented and time_detection (input
attributes given by the fault detector), fault
attributes (kind_fault, location_fault,
fault_time) are presented as the conclusion of
this task. This task needs information from
specific models of the system (structural model,
diagnosis model) in order to perform the task.
• Fault identification: Detection of other relevant
attributes of faults: the size and time-variant
behaviour of a fault. This task follows fault
isolation. It has the following attributes: fault
attributes (size_fault and
timevariant_behaviour).
• Fault diagnosis: Detection of kind, size, location
and the detection time of a fault. This task
follows fault detection. Fault group isolation
and identification. The attributes of this
subclass are: time_detection, fault_presented
and diagnostic (kind_fault, location_fault,
size_fault and time_detection). It needs
structural and/or diagnosis models to perform
the task.
• Monitoring: A continuous real time task to
determine the conditions of a physical system,
by recording information and recognizing and
indicating anomalies of the behaviour. The
attributes are: monitored_variables, alarms,
events, operating_conditions,
tunning_parameters (thresholds and similars),
anomalies_behaviour and
information_recognising.
• Supervision: Monitoring a physical system and
taking appropriate actions to keep the system
operating in the case of faults. It has the
following attributes: diagnostic, actions,
decision_system.
4.4 Models
For the purpose of engineering analysis and design,
physical systems are usually represented in some
mathematical form; this representation is also called
the model of the system. The properties of the
model reflect the nature of the system, though in
many cases the model may just be an approximation
of the true system behaviour. In (Isermann and
Ballé, 1997), five classes of models are considered
(See figure 5):
• Quantitative model: Uses static and dynamic
relations between system variables and
parameters in order to describe system
behaviour in quantitative mathematical terms.
The attributes of this subclass are: description,
input_quantitative, output_quantitative and
parameters.
• Qualitative model: Uses static and dynamic
relations between system variables and
parameters in order to describe system
behaviour in qualitative terms such as
causalities or if-then rules. It has the following
attributes: description, input_qualitative,
output_qualitative and parameters.
• Diagnostic model: A set of static or dynamic
relations which link specific input variables (the
symptoms) to specific output variables (the
faults). The attributes of this subclass are:
description, input_diagnostic (symptoms,
residuals, and physical variables),
output_diagnostic (fault attributes) and
parameters.
ONTOLOGY FOR INTEGRATING HETEROGENEOUS TOOLS FOR SUPERVISION, FAULT DETECTION AND
DIAGNOSIS
129

Figure 5: Ontology particular diagram
ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION
130

• Heuristic_model: The attributes of this subclass
are: description, input_heuristic and
output_heuristic. Commonly used for making
diagnoses, working with symptoms and
attributes of the diagnostic model.
• Structural model: Definition of the physical
interaction between components, materials and
energy sources. Attributes: description,
components, (sub)systems, instruments, plants,
materials.
In a future, we will add a sixth class regarding
event-driven models.
4.5 System Properties
System properties relate particular characteristics
required for the system. The main properties are:
reliability, safety, availability, dependability.
• Reliability: Ability of a system to perform a
required function under stated conditions,
within a given scope, during a given period of
time. Measure: MTBF = Mean Time Between
Failure. MTBF = 1\la; la is rate of failure (e.g.
failures per year). It has the following
attributes: MTBF, period_time and
required_function.
• Safety: Ability of a system not to put people,
equipment or the environment into danger. It
has the following attributes: value_safety.
• Availability: Probability that a system or the
equipment will operate satisfactorily and
effectively at any point of time measure: MTTR
Mean Time To Repair MTTR = 1/µ; µ: rate of
repair. The attributes of this class are: MTTR
and probability_availability.
• Dependability: A form of availability that has
the property of always being available when
required. It is the degree to which an item is
operable and capable of performing its required
function at anyrandomly chosen time during its
specified operating time, provided that the item
is available at the start of that period (RAM
Dictionary). It has the following attributes:
degree_dependability and time_dependability.
5 ONTOLOGY FOR SFDD
The ontology described in the previous section has
been created using the tool Protegé 2000 (Noy et al.,
2003) with the purpose of integrating different
supervision, fault detection and diagnosis toolboxes,
within the context of the CHEM project (Cauvin,
2002). These toolboxes are the result of
encapsulating SFDD techniques which provide a
common description and interface for users. Each
toolbox has been designed and developed by
different teams. The SFDD ontology provides a
shared and common vocabulary for the toolboxes
with two main benefits: Firstly, the operator handles
decision support information holistically. Secondly,
the operator is not burdened with different
vocabularies and interpretations coming from
heterogeneous tools, but can work with a single
ontology. Figure 6 shows the differences between
the current SFDD information flow and the
integration approach using an ontology.
6 CONCLUSIONS
In this paper we present a top-level ontology for
sharing knowledge in distributed supervision
systems. We provide the basic conceptualisation and
implementation with Protege2000. The ontology is
presented as a first step towards SFDD toolbox
interoperability.
We have two main lines of research for future
work. First, to study pruning and factoring
mechanisms, like in (Conesa et al, 2003), in order to
derive both task and domain level ontologies from
our top-level ontology. Second, to effectively
integrate SFDD toolboxes into a multi-agent
platform (Wooldridge, 2002) by means of the
ontology.
Figure 6: Above: current data flow of heterogeneous
SFDD tools. Below: Data flow of supervision, fault
detection and diagnosis integration
ONTOLOGY FOR INTEGRATING HETEROGENEOUS TOOLS FOR SUPERVISION, FAULT DETECTION AND
DIAGNOSIS
131

ACKNOWLEDGMENT
This work has been supported by the Spanish
MCYT project DPI2001-2198 and MEC TIN2004-
06354-C02-02.
REFERENCES
Bernaras A., Laregoiti I., Bartolomé N. And Corera J.,
1996. An Ontology for Fault Diagnosis in Electrical
Networks. Proc. Intelligent Systems Applications to
Power Systems, (ISAP '96).
Bussmann, S., Schild, K. 2001. An Agent-based
Approach to the Control of Flexible Production
Systems. Proc. 8th IEEE Int. Conf. on Emerging
Technologies and Factory Automation (EFTA 2001),
pp. 169-174..
Cauvin, S., 2002. CHEM:. Advanced Decision Support
System for Chemical/Petrochemical Manufacturing
Processes. Monet Newsletter, issue 2.
Ceccaroni, L., Cortés, U., Sánchez-Marré, M., 2004.
OntoWEDSS: augmenting environmental decision-
support systems with ontologies. Environmental
Modeling & Software 19 785-797.
Colomer J, Meléndez J., 2000. “Sistemas de
Supervisión”. Cuadernos CEA-IFAC. Ed. CETISA.
Conesa Jordi, De Palol Xavier, Olivé Antoni. 2003.
Building Conceptual Schemas by Refining General
Ontologies. In V.M.W.R.O. Stepenkove, editor,
DEXA’03 , volume 2736 of LNCS, pp. 693-702,
Springer. .
Correa da Silva F. S., Vanconcelos W., Robertson D.S.,
Brilhante V., de Melo A.C., Finger M., Agustí J.,
2002. On the insufficiency of Ontologies: Problems
in Knowledge Sharing and Alternative Solutions.
Knowledge-based systems, 15(3): 147-167.
Giret, A., V. Botti, V. 2004. Holons and Agents. Journal
of Intelligent Manufacturing. 15(35):645-659.
Giunchiglia, F., Gomez-Perez, A., Stuckenschmidt, H.,
Pease, A., Sure, Y., and Willmott S. (eds), 2003. 1st
International Workshop on Ontologies and Distributed
Systems (ODS2003).
Guarino N., 1998. Formal Ontology and Information
Systems. Proceedings of FOIS’98, IOS Press, pp. 3-
15.
Isermann and Ballé, 1997. “Trends in the application of
model based fault detection and diagnosis of technical
processes”, Control Engineering Practice, vol. 5, n° 5,
pp. 709-719
Kitamura Y. and Mizoguchi R. 1999. An Ontological
Analysis of Fault Process and Category of Faults.
Proc. Of Tenth International Workshop on Principles
of Diagnosis (DX-99), pp.118-128..
MacFarlane, D., Bussmann, S., 2000. Developments in
holonic production planning and control. Int. Journal
of Production Planning and Control, 11 (6): 522-536.
Marik, V., McFarlane, D., Valckenaers, P. (Eds.) 2003:
1st International Conference on Industrial
Applications of Holonic and Multi-Agent Systems,
Springer, LNCS, Vol. 2744.
Martin P., Eklund P., 1999. Embedding Knowledge in
Web Documents. Computer Networks. vol 33. No. 11-
16. pp. 1403-1419.
Murray, R.M., Aström, K.J., Boyd, S.P., Brockett, R.W.,
Stein, G. 2003. Future Directions in Control in an
Information-Rich World. A summary of the report of
the Panel on Future Directions in Control, Dynamics,
and Systems. IEEE Control Systems Magazine, April
2003, pp. 20-33.
Noy N.F. and McGuinness D.L, 2001. Ontology
Development 101: A Guide to Creating Your First
Ontology. Stanford Knowledge Systems Laboratory
Technical Report KSL-01-05..
Noy NF, Crubezy M, Fergerson RW et al. , 2003. Protégé-
2000: An Open-source Ontology-development and
Knowledge-acquisition Environment. Proc AMIA
Symp.
Schorlemmer, M., Kalfoglou, Y. 2003. Using Information-
Flow Theory to Enable Semantic Interoperability. In:
Artificial Intelligence Research and Development. I.
Aguiló et al. (Eds), IOS Press, 421-431.
Sure Y., Corcho O. (eds), 2003. 2nd International
Workshop on Evaluation of Ontology based Tools
(EON2003), 2003.
Wooldridge M., 2002. An Introduction to Multiagent
Systems. John Wiley & Sons.
ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION
132