Documente Academic
Documente Profesional
Documente Cultură
operations
1
Dr., Cork Constraint Computational Centre, University College Cork, Cork, Ireland
2
Prof., Chalkis Institute of Technology, Automation Dept. and Hellenic Open University, Rodu 2, Ano Iliupolis, Athens,
16342, Greece
3
Prof., Laboratory of Project Management, Civil Engineering Department, Democritus University of Thrace, Xanthi, 67100,
Greece.
Abstract
In this paper we argue that early warning systems for engineering facilities can be developed by
combining and integrating existing technologies and theories. As example, we present an efficient
integration of fuzzy expert systems, fault tree analysis and World Wide Web technologies to their
application in the development of the Landfill Operation Management Advisor (LOMA), a novel early
warning and emergency response system for solid waste landfill operations. The aim of LOMA is to
provide assistance to landfill managers on their efforts in preventing accidents and operational
problems and to help them to develop emergency response plans if these operational problems shall
occur. Additional aim is to disseminate information and knowledge to the public on landfill
1
Corresponding author.
1
Environmental Modelling & Software 24 (2009) 8–25
operational problems and their adverse effects. This aim is related to solid waste organizations that
have to accord with legislations similar to the European Union’s EC Directive (2003/4/EC) on Public
When using LOMA, the user first describes the working conditions at the landfill. Then, based on this
description, LOMA informs user about the potential operational problems. Afterwards, it analyzes
the operational problems in more detail and it estimates the possibility of their occurrence. Finally, it
provides advice on how to prevent them and on how to respond if any of them occurs. This paper
thoroughly investigates LOMA development as well as its integral methodologies and validates it by
outlining its performance in test cases that were performed by experts during the operation of a real
landfill as well as in test cases extracted from a specially constructed database with synthetic events.
Keywords
Early warning system, Expert systems, Fault tree analysis, Fuzzy logic, Possibility theory, Landfills,
Software Availability
Landfill Operation Management Advisor is a web based system which is available for free from
http://loma.civil.duth.gr
1. INTRODUCTION
In order to manage the anticipated problems from the generation and disposal of solid waste, Solid
Waste Management (SWM) systems have been designed and are operating worldwide. The main
task of a SWM system is to collect, transport, and dispose the solid waste generated within a service
area using methods and techniques that meet predefined specifications. These systems include
2
Environmental Modelling & Software 24 (2009) 8–25
source separation of recyclables and hazardous waste as well as facilities for recycling and
composting (Salhofer et al., 2007 ). A number of facilities like landfills, recycling, scrap, and
incinerators are some components of a SWM system. However, being common in many engineering
systems, the operations in these facilities are associated with problems. The consequences of the
operational problems in SWM facilities, depending on their nature and severity, range from minor
infrastructure damages or simple nuisance problems to critical events, which can lead to the loss of
In this work the term “operational problem” is used to describe a situation during the operation of a
perspective. In the context of landfills, such operational problems could be the surface and
subsurface fires, wind blown litter, traffic problems, and problems regarding the leachate and gas
An indicative example of a disaster that is related to land disposal of waste is the disaster that
happened in the Leuwigajah dumpsite in Indonesia, where after 3 days of heavy rainfall 2.700.000
m3 of waste started sliding down the valley (Kölsch et al., 2005). The waste covered an area of 900 m
× 300 m, 147 people died in the ruins of two settlements, and the surrounding environment has
been damaged significantly. Another worth to mention example is the fire that burst out in the
second larger landfill in Greece during the summer of 2006. Most probably, the fire was burning for
days in the compacted volume of waste under the subsurface of the landfill. It was expanded at the
surface after the collapse of a large pile of waste. The fire was burning for 10 days and released large
amount of dioxins in the atmosphere. In addition to the fire incident, one leachate holding pond was
overflowed due to the collapse of the waste pile causing large quantities of leachate to be expanded
to the surrounding area and to reach the houses of an adjacent village. In short, the incident resulted
3
Environmental Modelling & Software 24 (2009) 8–25
Several types of events like: bad weather conditions, equipment malfunction, wrong operation
practice, but also issues like bad design, human, organizational and communication errors can be
combined appropriately and can lead to critical operational problems such as the disasters
mentioned above. Unfortunately, there is a gap in the existing knowledge with respect to the factors
that contribute to the occurrence of incidents and accidents in SWM facilities. Therefore, the
managers of these facilities rely on their experience in order to estimate if any operational problem
is about to occur and of how this can be prevented or restricted. In this paper we argue that
intelligent computer systems can provide significant assistance to landfill and to SWM treatment
facilities managers in confronting operational problems through combining existing technologies and
theories so that to form Early Warning Systems (EWS). United Nations defines EWS as the provision
of timely and effective information, through identifying institutions, that allow individuals exposed
to a hazard to take action to avoid or reduce their risk and prepare for effective response (ISDR-UN,
2004). The objectives of such systems in the framework of the SWM industry should be the
provision of timely warning of imminent dangers so that the managers and personnel could have
time to prepare their strategy and their actions accordingly to prevent it. In addition, we describe a
novel research investigation on how to combine Expert System (ES) technologies together with basic
principles of the theory of fuzzy logic and a widely used risk analysis method called fault tree analysis
In this work we assume that SWM systems within a country or a state can be seen as organizations,
which have strategic, tactic and operational levels. The environmental protection agency, for
example, can be represented in the strategic level of the SWM organization model based in our
assumption. The SWM treatment facilities can be represented in the tactical level, while the
subsystems in these facilities, which are responsible and are determining their daily operations, can
4
Environmental Modelling & Software 24 (2009) 8–25
High level goal of our research is to formalize a general purpose methodology for early warning and
knowledge dissemination services in the context of SWM organizations, focusing in particular at the
operational problems in SWM treatment facilities. The objective of our work presented in this paper
is to satisfy the need for early warning service with respect to the operational problems in landfills.
This service is intended to be similar to the type of early warning service that would be given by
to disseminate the acquired knowledge and information about operational problems in landfills to
the public and to groups of people who could be interested for it, so that in this way, to provide
support to SWM organizations to accord with legislations which are similar to European Union’s EC
In this paper we present LOMA, which is the result of our attempt to provide the early warning and
technologies, fault tree analysis, and possibility theory. In developing LOMA we first defined the
Hellenic SWM system as the reference organization and a typical landfill at the operation phase as
the reference system. Then, we acquired and modelled the knowledge on landfill operational
problems and then we represented this knowledge into LOMA. The last step of the development
process was to validate and to test our system. This paper provides details in how the selected
technologies were used and configured together, illustrates the user-system interactions, and
presents the results of the validation and testing phase. It also presents the architecture and a
This paper is complementary to Dokas et al., (2006; 2007). Both papers are important for one to
comprehend basic elements characterizing our research goal which is the formalisation of a
methodology for EWS. In Dokas et al. (2006) we proposed an effective process during which the
5
Environmental Modelling & Software 24 (2009) 8–25
knowledge on operational problems in landfills can be acquired and elicited and we have shown in
detail to which extent a risk analysis method is elaborated in this knowledge acquisition and
elicitation process. What is new in this paper, compared to Dokas et al. (2006), which actually
describes only the knowledge acquisition component of the herein presented complete system, is
that we show in detail how the same graphical notation was used as basis to represent the
short, our point of view in this paper has a knowledge representation orientation. We show in detail
the architecture of such an intelligent system, as well as how fuzzy fault tree analysis, possibility
theory and risk analysis are integrated into one operational fuzzy expert system providing early
warning services. Nevertheless, this paper provides a very brief only description of the knowledge
acquisition phase which is based on the work in Dokas et al (2006) aiming to enhance readers
understanding of the significant role of fault tree analysis in the methodology that we are trying to
formalise.
In Dokas et al. (2007) we have illustrated with a test case, which emulates a very small fraction of
the current functionalities of LOMA, how could the adaptation tasks be performed, how easy these
tasks can be implemented in a prototype of LOMA with using its current technologies and how new
knowledge and conditions that are evolving with time can be represented in a LOMA like prototype.
This work shows how the tasks of updating, maintaining and adapting the knowledge base in a
LOMA like prototype can be performed. However, due to the fact that the concepts and relations
associated with these tasks have to build up into the integrated LOMA system, herein presented, the
specifications for updating, maintaining and adapting the knowledge base of the overall system
become much more complex and cannot be dealt into the present paper. It is our future goal to
extent the above mentioned sample test cases for updating, maintaining and adapting the
knowledge base, using the rigorous frame based representation theory (Dokas et al. (2007), to all
functionalities of LOMA, based on which, we will be able to define specifications for a tool that will
6
Environmental Modelling & Software 24 (2009) 8–25
support these tasks to be done automatically in order to enhance the attempts of the development
team and of some authorised users to enrich, update and maintain the knowledge base of LOMA.
Many scientific fields were using ES technology to solve a variety of problems starting from the
1960’s. Recently, artificial intelligent technologies together with fuzzy logic have been used for the
development of EWS. Liu et al. (2003) have implemented If-Then rules to model the risks associated
to software quality and project management and in order to assess the risks they have applied fuzzy
inference on the rules. Yang et al. (2001) have examined the application of three layer BP artificial
neural network to an EWS for commercial bank loan risk. Lei et al. (2006) have introduced case
based reasoning that is enhanced by genetic algorithms in a EWS for financial crisis. Although the
types of early warning services provided by the systems described in these papers are different with
respect to the reference organizations and to the reference systems, these can be considered to be
Regarding the use of ES technologies in the field of SWM it has been reported that up to the year
1990, ES applications regarding the municipal SWM planning have not been identified (Thomas et al.
1990). More recently, a review of available ES, geographic information systems, decision support
systems and their applications in the landfill design and management was contacted by Lukasheh et
al. (2001). The nine reviewed ES were dealing with the following problems: a) assessment and
evaluation of a landfill site, b) design and evaluation of landfill elements such as leachate collection
systems, final cover, vegetative cover, compacted clay liner, c) assessment of the liner material in
chemical resistance to the waste leachate, d) evaluation of landfill closure e) landfill design.
Although in the same reference was noted that landfill operations is a suitable area for ES
applications, none of the reviewed ES was referring, in any way, to the landfill operational problems.
Moreover, we did not find, up to today, any paper presenting the application of fuzzy logic and ES
7
Environmental Modelling & Software 24 (2009) 8–25
The paper continues by describing the main concepts and the main development phases of LOMA as
follows. In section 2 we describe the basic concepts behind the technologies that were used to
develop LOMA. Section 3 presents the high level specifications of LOMA. In section 4 emphasis is
given to the use of fault tree analysis during the knowledge acquisition process. At that point the
rationale for using the fuzzy importance measure as a measure of prioritizing the emergency
response actions is presented. In section 5 we present the architecture of LOMA and we provide a
detailed description of its knowledge base structure. An illustrative example of the calculations for
the steps in using LOMA services, which is made in section 7. In section 8, we describe the validation
process that was applied to LOMA, followed by section 9, in which we discuss LOMA’s characteristics
In a nutshell, ES are tools which allow expert knowledge and experience to be “treated and stored”
properly in order that they can be used by non-experts at a later time. ES are especially effective in
cases where the need is for modeling heuristic concepts rather than analytical mathematical
relations (Turban, 1995). Fuzzy logic (Zadeh, 1965) provides a framework whereby basic notions
such as similarity, uncertainty and preference can be modeled effectively. A “fuzzy set” represents a
set with ill-defined boundaries. Fuzzy logic processes vague terms with “grey boundaries”, managing
however to come up with a conclusion, within the closed interval [0, 1] ranging from “completely
false” to “completely true”. Fuzzy sets have been used in numerous scientific applications for
modeling ambiguities of the real world (Iliadis, 2005; Makropoulos et al., 2005; Marsili-Libelli, 2004;
Ross, 2004; Zimmermann, 1996; Fay, 2000) . There are situations where fuzzy set theory is used in
collaboration with ES technology leading to fuzzy expert systems (Fleming et al., 2007; Liu et al.,
2006; Makropoulos, 2003; Grove, 2000). Fault tree analysis can be simply described as an analytical
technique whereby an undesired state of the system is specified and the system is then analyzed in
8
Environmental Modelling & Software 24 (2009) 8–25
the context of its environment and operation to find all realistic ways in which the undesired event
ES, fuzzy and non-fuzzy (or “crisp”), contain the following modules (Turban, 1995):
1. The knowledge base, in which the knowledge of experts is represented in the form of IF-
THEN rules, frames, semantic networks, first order logic based methods, etc.
2. The working memory module, that stores the input data and the information generated
3. The inference engine in which the processing of the rules and the reasoning of the ES take
place.
4. The user interface module that facilitates the interaction between the user and the ES.
5. The knowledge acquisition facility that provides the user with appropriate “help” tools
6. The explanation module that allows the ES to present its reasoning regarding its
conclusions.
1. The domain expert that is presumed to have the specific experience, knowledge, judgment
and methods, as well as the ability to give advice for solving problems (Turban, 1995).
2. The knowledge engineer that is the builder of the ES who defines the knowledge
framework and gathers the necessary facts, information and knowledge for the development
9
Environmental Modelling & Software 24 (2009) 8–25
In classic logic, a proposition is exclusively true or false. In fuzzy logic, vague terms are fuzzy sets that
can be processed to lead to conclusions ranging within a closed interval [0,1] (Kosko, 1997). A fuzzy
set is characterized by its membership function; it expresses the degree to which the properties of
the fuzzy set are satisfied by a specific value of the corresponding reference set. In Figure 1, if the
abscissa refers to failure possibilities of a system component and pm = 0.30, then, objectively, the
value of 0.30 fulfils the properties of the failure possibility set which is named “LOW”; as a result, a
membership value of 1 is assigned (the maximum possible on a conventional scale [0, 1]). It can be
assumed that the range of acceptable values fulfilling the property is 0.10 to 0.20 (pl to pr referring to
Figure 1); then, all values of failure possibilities outside this range have a membership value of 0,
while inside this range the membership value ranges from 0 to 1. Fuzzy sets may be linked by fuzzy
rules of the following form: “if X is A, then Y is B”. Here, A and B are fuzzy sets and X and Y are the
The shape of membership functions in fuzzy systems affects their final results. However, in almost
every work on fuzzy sets, the existence of membership functions taking part in the considered model
is assumed and not studied in depth, whether or not such functions exist (Sancho-Royo et al., 1999).
Bilgiç et al. (1999) outlined a summary of six methods of membership function shape determination,
concerning the experimental research. In a specific outlined method, the experts were giving
answers to questions like: “What is the degree of belonging of color A to the (fuzzy) set of dark
colors?”. “What is the degree of belonging of John to the set of tall people?”. In general “To what
Fuzzy expert systems are a kind of fuzzy systems that are processing input values by using fuzzy
rules; the conclusions might be fuzzy or non-fuzzy (crisp). Fuzzy expert system is an expert system,
which incorporates fuzzy sets and/or fuzzy logic into its reasoning process as well as into its
10
Environmental Modelling & Software 24 (2009) 8–25
knowledge representation scheme (Hall et al., 1991). The inference procedure of a fuzzy expert
In the fuzzification state, the user input is transformed into degrees of membership to the
fuzzy sets, via the membership functions. At the beginning of the inference state, the fuzzy
rules in the knowledge base are implemented, by using an appropriate implication method.
The result of the implication of a fuzzy rule is also a fuzzy set, which is correlated to the
degree of truth of the premise (the If part of the rule). The implementation of fuzzy rules is
followed by the aggregation of fuzzy output sets. During the aggregation, the output fuzzy
sets of all the implemented fuzzy rules are combined into one fuzzy set. Finally, during the
defuzzification state, a specific output is derived from the combined fuzzy set mentioned
above, as a final result of the inference procedure of the fuzzy expert system. The details of
the fuzzy inference procedure and the advantages of using fuzzy logic in the development of
A fault tree is a logic diagram that displays the interrelationships between a potential critical event in
a system and the reasons for this event (Hoyland et al., 1994) and is the graphical representation of
the fault tree analysis. A typical fault tree is consists of the top event, the basic events, and the logic
gates. Figure 2 illustrates a fault tree structure with typical components. The top event represents an
undesirable state of the system, the basic events represent the state of the systems components,
and the logic gates describe the relationship between the basic events and the top event. In classic
fault tree analysis the AND logic gate denotes that the output is in a failure state, if all the inputs are
in failure state. The OR logic gate denotes that the output is in failure state, if at least one of the
11
Environmental Modelling & Software 24 (2009) 8–25
inputs is in failure state. An intermediate event represents an intermediate state of the system that
Fuzzy fault tree analysis (Yuhua et al., 2005, Onisawa, 1996, Suresh et al., 1996) extents classic fault
tree analysis, which is based on the assumption that there are sound and clear success and failure
states in a system and that failures occurs at random. Fuzzy fault tree analysis can be implemented
when:
• There are no clear boundaries between failure and success states of the system, or when it is
• The probability of system failure cannot be calculated precisely due to the lack of sufficient
• There is subjective evaluation of the reliability, which is made with natural language
expressions.
In the context of fuzzy fault tree analysis, given a fault tree structure it is possible to calculate the
subjective reliability of the corresponding system, given information about the reliability of the
system components in linguistic terms. These terms are translated into fuzzy sets. The fuzzy sets
express the subjective possibility of failure (i.e. the subjective unreliability) of the system. This is
done by mapping each linguistic value to a range of subjective failure possibilities through a fuzzy set
membership function (this issue is discussed in detail in the § 6.2.) The subjective failure possibility is
defined on the unit interval [0,1]. Thus, If Pos(E1), Pos(E2), … Pos(En) are the failure possibilities of the
basic events E1, E2, … En respectively, and the corresponding components of the system are
independent, then the output possibilities of the AND – OR gates can be calculated with the
12
Environmental Modelling & Software 24 (2009) 8–25
PosAND, PosOR are the possibilities of the output events of the AND and OR logic gates respectively
and the symbols ⊗ and denote the fuzzy subtraction and multiplication. Through the outputs of
the AND - OR gates it is possible to determine the subjective possibility of the top event following a
bottom–up calculation approach. In some cases the independence of the top events might not be
possible. That can happen in cases where a top event in a system has actuation signals or has causes
that are common with other top events, which are referring to the same system. In this case the
formulas {1} and {2} are not applied and other formulas that can be found in Stamatelatos (2002)
The fuzzy importance measure indicates “how much” a basic event contributes to the top event. It is
derived from the Euclidian distance of two fuzzy sets. The first set is the derived possibility of the top
event when a basic event Ei of the corresponding fault tree is completely available (TEEi=1), while the
other set is the derived possibility of the top event when the same basic event is completely
unavailable (TEEi=0). The fuzzy important measure is defined as (Suresh et al., 1996):
Where ATEEi=1,0 and BTEEi=1,0 denotes the lower and upper values of TEEi=1 and TEEi=0 fuzzy sets
respectively at each α-level (or a-cut). α-level is the lower and upper values of the reference set
corresponding to the membership values of the fuzzy sets TEEi=1 TEEi=0 that are greater than, or equal
13
Environmental Modelling & Software 24 (2009) 8–25
OPERATIONS
In many organizations and especially in SWM systems, the timely warning and response of imminent
problems is more desirable in terms of economic, political, environmental, and human resources
than to deal with the outbreak and aftermath in an ad-hoc manner. This is the core idea behind the
development of LOMA. LOMA’s specifications were defined in order to provide a type of early
warning service that can be considered to be similar to the early warning services offered by
employees at the strategic level of an organization to the employees at the tactic level; especially at
the case where from the second there is a lack of experience. The following assumptions were made
• The employees at the strategic level should provide early warning services to employees in a
• The early warning service is requested by the employees of a facility whenever they perceive
events, which are unusual during the different known modes of operations of the facility,
and they do not know or they are not certain if these events can facilitate the occurrence of
operational problems.
• The people at the strategic level are not obliged to know all the characteristics and attributes
of the components associated with the facilities at the tactic level of the organization. They
have to have knowledge however about the mechanisms that can trigger common
With the term “common operational problems” we mean the set of known problems that can occur
in the majority, if not in all, SWM facilities of the same type within the same SWM organization
14
Environmental Modelling & Software 24 (2009) 8–25
during their operations, regardless its geographical location, the quantities and quality of incoming
waste streams, its operational practices etc. In this context an ES application that could:
2. Be accessible by people that are served by the SWM organization and who would like to be
c) If an accident or a problem occurs, what actions are required to lessen the impact of its
consequences?
4. Deliver directly and in a timely manner the appropriate advice/solution so that managers
6. Be simple in use,
could be used to provide early warning services, to propagate the corresponding expertise globally,
and to help landfill managers to respond to accidents and operational problems. These are in fact
LOMA is addressed to landfill managers (especially to those with little experience in landfill
operations), to the public, and to the personnel of a SWM organization that would like to know the
following:
15
Environmental Modelling & Software 24 (2009) 8–25
1. The ease by which a common operation problem can occur based on current working
conditions.
Currently, LOMA estimates the occurrence possibility in 24 common landfill operational problems
like subsurface fire, litter, corrodible soil cover, odor, noise, etc.
These problems have been analysed in an “acceptable” and “sufficient” level using the principles of
fault tree analysis. The level of analysis of an operational problem is considered to be “acceptable” if
all constructs of a fault tree (i.e. top events, basic events, intermediate events) are associated to real
components or to concepts that can be identified during the operations of a typical SWM facility.
The level of analysis of an operational problem is considered to be “sufficient” if the attributes of the
fault tree structures and their values are describing realistic conditions and coincidences which are
LOMA was developed in the context of a specific SWM organization (i.e. the Hel-lenic SWM system).
This, however, does not confine the level of generality of our analysis only in the boundary of Hellas,
meaning that the 24 operational problems represented in LOMA can occur in the majority of landfills
at the SWM organizations in some Mediterranean countries, especially in those being member
states of the European Union. That is because these countries accord with European Union
directives and have similar weather conditions. Thus, the typical landfill model used in LOMA can be
The landfill in the knowledge base of LOMA is represented with concepts and relations in such a
detail and in such an abstract level that has its reference point into a type of landfill that it is
considered to be typical. This type of analysis is not strange in the context of landfill operations. A
16
Environmental Modelling & Software 24 (2009) 8–25
number of instructive videos and special training packages on landfill operations are available, such
as in COSD (2007) and ISWA (1998). These materials were developed having as reference point the
operations in a typical landfill. There is however a significant difference between LOMA and these
materials. The difference is that LOMA, in contrary to the training packages, represents explicitly the
complexity of the coincidences and the chains of events, which can trigger operational problems.
This is one extra characteristic that makes the work on LOMA unique in the context of SWM.
LOMA’s development is based on the complete and sufficient analysis of a typical landfill and on the
empirical findings regarding the dependence of its everyday operation to a set of events that can
cause operational problems, within specific SWM organizations. However, we should point out that
although our goal is to formalize a generic methodology in order to be used in any SWM
organization, the system described here is generic only to a certain extent, since its knowledge base
was adapted for Mediterranean regions mainly. Therefore, customization has to be made at the
knowledge base, (i.e. addition/transformation of new concepts that represent operational problems,
as well as causes and events that can trigger these) in order to be fully operational and applicable in
totally different SWM organizations and in totally different facilities within SWM organizations (e.g.
We should point out also that the tasks of adapting; updating, and maintaining cannot be performed
by the users of LOMA in its current version. The concepts and relations associated with these tasks
have to build up in to the system. However, due to the generality of the methodology, the tasks of
adaptation and maintenance does not need major effort for someone with basic computer
programming skills since the concepts and relations of the new reference system have to be build up
in the system using a representation technology which is easily understood (for more details see
§6.1). Especially in the case where the new reference system is a landfill, a significant amount of the
represented knowledge in LOMA can be reused and can be easily adapted in order to be used in
specific landfills operating not only in Mediterranean countries. The task of adaptation in this case
17
Environmental Modelling & Software 24 (2009) 8–25
will require however significant contribution in terms of expertise acquisition and modelling, which
our methodology tries to enhance with using the fault tree analysis; a widely used and easily
understandable risk and reliability analysis method, as a knowledge acquisition and modelling tool.
LOMA has been specified to be a web based system in order to support effectively a wide range of
users such as for example the experienced and inexperienced landfill managers, by providing to
them early warning services, and the people and organizations interested in landfill operations, by
providing to them useful information. Given the variety of users, their potential geographic
distribution, as well as the rapid expansion of fast and reliable web networks, the choice of
It has been mentioned above that the users of LOMA cannot update and/or adjust automatically the
knowledge base of LOMA from their browser. Whenever a user wants to contribute enriching the
knowledge base of LOMA, he has to submit his analysis or his advice through the available web
forms or through e-mail so that to be validated by the domain experts. Only after its validation the
submitted knowledge will be included in the knowledge base. The process of updating the
knowledge base by the users can be characterised as quite strict, but it has to be like this because
the information in LOMA has to be “certified” by the experts in the strategic level of the organization
In light of these specifications, LOMA can be seen as a tool which provides four different types of
services to the SWM organization, depending upon the category in which its users might belong.
3. Consultation tool that can be used by the personnel at the strategic level of the organization,
4. Educational tool that can be used by the new staff members of the SWM organization.
18
Environmental Modelling & Software 24 (2009) 8–25
4. KNOWLEDGE ACQUISITION
Usually large computer systems are developed by a development team. A group of scientists and/or
skilled technicians are forming that team. Each member of the team has a role that is responsible for
a certain set of tasks. Indicative roles of a typical development team of a computer system are those
of the system architect, the specification engineer, the designer, the programmer and the tester. In
particular, when developing knowledge based systems, there is one more role assigned to a member
of the development team. This role is known as knowledge engineer. The knowledge engineer is
responsible for the knowledge acquisition process (Schreiber et al. 2000). This process is very
important for the development of a knowledge based system and often is quite complex.
1. The system architect role, to whom the assigned tasks were: the overall system design,
specifications, fuzzy ES design, assign fuzzy membership functions to literal values, adapt
2. The knowledge engineer role, to whom the assigned tasks were: to perform literature
research, understand the problems of landfill management, to talk with the experts, to
facilitate meetings, to acquire and model the knowledge using fault trees and the assistance
of landfill experts.
3. The programming engineer role, to whom the assigned tasks were: to code in fuzzy ES shell
environment based on the specifications and to sett up the web interface layer of the
system.
4. The tester role, to whom the assigned tasks were: the validation and verification of the
system.
19
Environmental Modelling & Software 24 (2009) 8–25
This paragraph is focused on the knowledge acquisition process that was performed by the
In LOMA’s case, the knowledge acquisition was composed of three main activities:
1st activity: Preparation of the knowledge acquisition process during which the knowledge engineer
has performed literature research and text analysis to familiarize himself with the application
domain and to determine an initial set of landfill operational problems and causes.
3rd activity: In parallel with the 2nd activity the knowledge engineer had numerous meetings with the
landfill manager who had more than 10 years of experience and considered to be a domain expert.
From the point of view of the knowledge engineer the objective of the meetings mentioned in the
3rd activity was to analyze together with the domain expert the operational problems using the fault
tree analysis. Additional objective was to determine advice, solutions and emergency response
actions for each and every operational problem. These activities are briefly described below.
Text analysis is a knowledge acquisition method in which knowledge is gathered and combined with
the heuristics through printed documents like books, papers, guidelines, legislations, etc. Referring
to the development of LOMA, the first goal of text analysis was to identify several common landfill
operational problems. The second goal was to identify any printed document that was referring to:
operational problems. The third goal was to identify as many as possible causes of operational
problems and, if possible, the way by which these causes were combined together to trigger the
operational problem. Finally, the last goal of text analysis was to detect any advice regarding the
20
Environmental Modelling & Software 24 (2009) 8–25
Although numerous documents were found to refer on to several landfill subjects like design,
leachate and methane production, few of them only were referring to landfill operation. Moreover,
very few documents were referring exclusively to specific landfill operational problems. Table 1
displays the references used during the knowledge acquisition. The existence of very limited text
resources regarding the analysis of landfill operational problems had made the task of text analysis
to be very tedious.
To acquire the knowledge regarding operational problems effectively, the knowledge engineer
attained the operation of a landfill in the city of Thessaloniki in Greece. The landfill serves more than
one million inhabitants. It accepts about 400.000 tons/year and is open all year around, 24 hours a
day. The landfill covers an estimated area of 50 ha with a depth of 15-35 m. During landfilling, the
waste is spread in the working face in layers of 3 to 5 m and is compacted. Afterwards the
compacted waste is covered with 0.5 m of soil. The attendance of the operations in situ gave the
opportunity to the knowledge engineer to interact and communicate with the manager and
personnel constantly. The goal of the knowledge engineer during the first days at the landfill was to
a) point out to the experts the goal of the intelligent EWS, b) explain the basic notions of it (i.e. ES,
fuzzy logic, fault tree analysis), c) establish with the help and guidance of the experts a collaboration
After that initial phase, every working day for a three month period, a landfill tour was made by the
knowledge engineer along with the landfill managers. The result was to acquire new knowledge
about landfill operational problems that have been continually been updated and corrected with the
guidance of the managers. Also, real operating problems have been observed. In addition, several
questions were made to landfill personnel regarding the problems. Through the attendance of the
21
Environmental Modelling & Software 24 (2009) 8–25
landfill operations a total perspective regarding this subject has been obtained that no book was
able to provide.
4.3.1. Preparation
Periodically the knowledge engineer was in meetings with landfill managers to analyze the
identified operational problems and to refine the knowledge that had already been acquired. During
the preparation of the meetings, the knowledge engineer was combining the causes of the
operational problems that were been detected for an operational problem in a random way. The
random combinations of the causes for each operational problem had been schematically displayed
During this phase, fault trees were been used as a “knowledge acquisition platform”. Specifically,
during the meetings with each domain expert, the schematically random combination of causes
mentioned above was been displayed to the experts, who, afterwards, were been attempting to
validate, change or erase the available data and update the fault tree structure with any knowledge
that was missing. During this time the experts were been thinking in loud. Their thoughts were
recorded and the knowledge engineer was trying to understand the way and the mechanism based
on which each detected reason contributes to the operational problem. At the end of this process, a
new “correct” fault tree was been created, displaying all possible causes of the corresponding
operational problem and also the way these could trigger it. At the end of every meeting, the
derived knowledge was being summarized and the experts were making further suggestions on it.
22
Environmental Modelling & Software 24 (2009) 8–25
The knowledge that was mapped in the fault trees constitutes the knowledge base of LOMA. Figures
2 and 3 display the derived fault trees for the operational problems “CORRODIBLE SOIL COVER” and
“UNCONTROLLED STORM WATER FLOW”. The gray rectangular that can be seen in Figure 2
represents the top event of Figure 3. This example shows how one operational problem can be a
cause to another. It actually shows how a cascading failure or “domino effect” can be generated.
The assumption of independence of the top events in this example is valid because these are
referring to different subsystems of the landfill, which are not affected by the same operational
parameters. The later is referred to the inner and perimeter drainage systems of a landfill while the
The use of fault trees as knowledge acquisition platform had the following advantages:
1. Helped the communication between the knowledge engineer and the domain expert,
components,
3. Displayed in a very functional manner the combinations of causes that could lead to a problem,
4. Helped the dialogue among the domain experts and the personnel that happen to attain the
meeting, when they were expressing different opinions upon a landfill operational problem.
After fault tree analysis completion, a categorization of the basic events, which were considered to
be early warning signals, was made. The identified early warning signals were 88 in total. The
categorization process resulted in defining nine categories like warnings associated with the
collection vehicles, the gas and leachate management, the infrastructure, the soil cover, the
23
Environmental Modelling & Software 24 (2009) 8–25
characteristics and quality of the incoming waste etc. The categories with the most signals, around
12% each, were been those associated with the characteristics of waste and with the leachate
management system, followed by those associated with the characteristics of the infrastructure,
with around 9%. This categorization clearly shows the importance of design and construction phases
in the proper and undisturbed operation of landfills. It shows also that bad design and/or
construction of some landfill components, which cannot be redesigned or reconstructed but were
noticed at a specific time frame, can be defined immediately as early warnings for the operation
phase. Therefore, there are links between LOMA landfill design and construction issues and a future
goal could be to integrate planning and construction in one generic methodology and system.
An ES shell is a complete ES (i.e. provides all the modules of the ES) without any stored knowledge in
its knowledge base. During LOMA’s development, a research for available expert system toolkits was
conducted and 17 free and commercial available ES shells were spotted such as for example the
systems CLIPS (2007) and Jess (2007). Basic selection criteria were: a) implementation of fuzzy sets
to the reasoning process, b) available alternatives for organization of the acquired knowledge, c)
knowledge representation capabilities like rules frames etc, d) available alternative procedures to
give a conclusion ask questions, d) technical support, e) cost, f) user friendliness. Based on the above
criteria the flex expert system toolkit (LPA, 1996) with the Flint fuzzy logic toolkit (LPA, 2005) by
A main task during the development phase of LOMA was to represent the developed fault trees in to
the knowledge base in a way that it could be possible to estimate the occurrence possibility of the
operational problems and to provide advice and emergency response actions, given by the user a
subjective evaluation value on the status of basic events. In below, a brief description of the
24
Environmental Modelling & Software 24 (2009) 8–25
knowledge representations components of LOMA and of how the notion of possibility can be
incorporated with the fault tree analysis will be made. In addition, in this section the notions of
“possibility”, “linguistic value” and “fuzzy importance measure” will be described because are
necessary to understand the structure and the operation of LOMA. This task was done by the
programmer engineer with the assistance of the knowledge engineer and the system architect.
An important knowledge representation component that is used in LOMA, are the production rules.
However, the main knowledge representation scheme in LOMA is the frame. Frames allow data to be
stored in an abstract manner within a nested hierarchy with common properties automatically
inherited through the hierarchy (LPA, 1996). In real world, every object has several attributes.
According to this, frames have several attributes forming the slots of the frame. In LOMA’s case, the
concepts of “basic event” and “top event” have been represented as frames, as Figure 4 illustrates.
Each frame has a number of slots that represent some attributes like the user input value, the
In addition to rules and frames, LOMA uses directives and procedures to manage and search the
• Questions - Answers: The question-answer system allows final applications to query the user
25
Environmental Modelling & Software 24 (2009) 8–25
Fuzzy fault trees can be used to estimate the possibility of the top events, if the possibilities of basic
events are known using the formulas {1} and {2}. This is the fundamental principal behind LOMA’s
operation. For example, taking into consideration the fault tree of Figure 3, and applying the
formulas {1} and {2} accordingly, it is possible to estimate the possibility of the problem
“UNCONTROLLED STORM WATER FLOW” (Pflow), if the possibilities Pos(E1), Pos(E2)…Pos(E8) of the
[1Pos(E5)]]]]]⊗[1Pos(E6)⊗[1[1Pos(E7)]⊗[1Pos(E8)]]]] {4}
In LOMA’s case the estimated possibility denotes the ease by which the landfill operation can fail,
and is expressed with a fuzzy set. This perception regarding the possibility notion was proposed by
Zadeh (1978). Figure 1 expresses the fuzzy set “Low possibility”. One way to interpret this fuzzy set is
the following: Given that the possibility of occurrence of an (basic) event is “low” its “proposed”
possibility value is the pm. However, there are also additional values satisfying the properties of the
set “low” to a degree (in the case of Figure 1, the values from pl to pm and the values from pm to pr)
and that is why these values have a lower membership value than the pm. These values are also
triggered whenever the user of LOMA subjectively evaluates the reliability of a basic event as “low”.
In this case, the use of possibility instead of probability notion was necessary mainly because there
In LOMA, the user inputs to each variable are words (i.e. rather high, low, etc.) and not real number.
Each word input is expressed with a fuzzy set analogous to Figure 1. Thus, the variables in LOMA are
26
Environmental Modelling & Software 24 (2009) 8–25
linguistics. Each word input expresses the subjective reliability evaluation made by the user
regarding the corresponding basic event. The concepts of linguistic variables are very useful in
dealing with situations which are too complex or too ill-defined to be reasonably described in
The goal was that the above method could be utilized during the scheduled meetings with experts.
However, it proofed to be very time consuming and irritating to landfill managers. For that reason, a
literature review contacted to point out any suggestions regarding the expression of possibility
notion by using linguistic values. In Table 2 a list of papers is displayed, outlining linguistic values to
LOMA’ s possibility input values were selected to be the ones proposed by (Lin et al., 1997), mainly
because: a) after consulting the experts, the use of seven input values by the system was considered
to be more familiar to the user, b) a decision was made to use linear membership functions, which
were evaluated during the validation of the system, because in the corresponding literature there
wasn’t any proposed expression for the possibility notion regarding linguistic values for use in this
research field. In Figure 5 the values of the linguistic variables of LOMA are displayed, where x axis
LOMA utilizes the notion of fuzzy importance measure to evaluate the contribution of basic events
to the corresponding operational problem and through that to prioritize the available advice and the
In Figure 4, the slots of the concept “basic event” that is instance of the frame “event” have been
shown, in which the advice and the fuzzy importance measure (indicated as “fim”) are included.
During its operation, LOMA calculates the fuzzy importance measure values of the basic events of
27
Environmental Modelling & Software 24 (2009) 8–25
the fault trees. It then displays to the user the advice that is related to each basic event of the
operational problem in a ranking manner (this issue is discussed in detail in § 8 and 9). In this case,
the fuzzy importance measure value of each basic event is the ranking criterion. Since it is true by
definition that the higher the fuzzy importance measure value of a basic event the more it
contributes to the problem occurrence, it is reasonable to conclude that the priority level of the
advice or of the emergency response action that is related to it has to be in a higher place than the
advice that is related with a basic event that has smaller fuzzy importance measure value. By
following this rule LOMA can prioritize the advices that are stored in its knowledge base, given the
circumstances in a landfill.
7. LOMA ARCHITECTURE
LOMA is consisting of four main components: the web interface layer, the database component, the
inference engine and the knowledge base, as shown in Figure 6. The knowledge base is described in
detailed in the next section. The web interface layer is composed of static and dynamic web pages
containing HTML objects. The web pages are having a role similar to the user interface component in
The users can activate, through navigating the web pages, the early warning and the knowledge
dissemination services of LOMA. They can also contribute in enriching the knowledge base of LOMA
The database component serves as repository for facts and information that are related to landfill
operational problems. It is used also as repository for the information submitted by the users on
these problems. When this happens members of the development team are forwarding the
28
Environmental Modelling & Software 24 (2009) 8–25
information to the experts for validation and after that process the experts are informing the
development team which information should be represented and stored to the system.
The inference engine provides the operations and procedures which are applied on the knowledge
represented in to the knowledge base component of the system in order to infer explicit knowledge.
The intelligence of LOMA is implemented through the inference engine and it is based on the
following Artificial Intelligence reasoning methods, which are used in state of the art Expert Systems
• Forward and backward chaining: The knowledge base of LOMA (see §7.2) consists of rule
bases which are searched by the inference engine of the Flex ES shell. The inference engine
is using forward and backward chaining methods. These methods are widely used in artificial
intelligent technologies for inferring implied knowledge from a set of production rules, facts
and user inputs. These methods are herein involved in performing the fuzzy fault tree
importance measures and risk analysis estimations based on possibility theory as shown in
• Inheritance in frames: In a frame hierarchy a subframe can inherit the attribute values from
the superframe, according to some inheritance strategy. Thus in frames the most important
reasoning task is the subsumption between two concepts, that is, determining whether all
instances of one concept are necessarily instances of the other concept taking into account
• Data driven algorithms: The inheritance of the characteristics of the frames is done
automatically, but it can be controlled using different data driven algorithms. The demons for
example, which were used widely in LOMA’s knowledge base (see § 7.2), are a type of such
algorithms.
29
Environmental Modelling & Software 24 (2009) 8–25
The structure of LOMA’s knowledge base, is schematically displayed in Figure 7. Each operational
problem that was analysed and represented with a fault tree was mapped in to LOMA’s knowledge
base as a distinct “operational problem module”. All operational problem modules are connected
with a set of events that are called “starting events” via a “central module”. A starting event is an
event that has been noticed by the user and it can trigger one or more operational problems. A
starting event can be seen as an “early warning signal”. A basic event of a fault tree can be a starting
event. Referring to the Figures 2 and 3, which both have the basic event “rainy weather”; if it is
raining during the landfill operation and the user selects the event “rainy weather” as a starting
event, then the system is able to inform him that both operational problems could happen. In some
occasions however, the starting events are events that constitute a basic event. For example, one
basic event of the operational problem “ODOR” is the “disposal of malodorous waste”, in this case
the starting events that constitute the concept of malodorous waste to the operational problem
“ODOR” could be the disposal of seaweeds, dead animals, sludge, cannery wastes etc.
The central module is activated whenever the user activates LOMA. In the beginning, the central
module activates a question urging the user to select a starting event category from a list (see the 1st
stage in § 9). Each starting event category forms a frame. The slots of the frame are the starting
events of the corresponding category. Every frame of a starting event category has also one extra
slot. The extra slot value changes whenever the user selects the corresponding category from the
displayed list; using a set of rules. This gives to the user the ability to choose several starting events
Afterwards, the central module activates a set of production rules to define which operational
problem is possible to occur. These rules are checking if there are identical starting events between
the user’s selected set and each set activates a specific operational problem. By using the same set
30
Environmental Modelling & Software 24 (2009) 8–25
of rules the system displays the possible operational problems (see the 2nd stage in § 9). Depending
on the user answer the central module activates the corresponding operational problem module.
The modular structure of the knowledge base was chosen because was proved to be easier,
whenever updates or changes were made regarding a specific operational problem. Moreover, this
All operational problem modules follow the same structure. Each module consists of a frame set, a
set of questions, a set of demons, a set of actions and a set of rules. Depending on the operational
problem, some modules have one additional set of rules informing the user that the analyzed
As it was shown in Figure 4, the concepts of “top event” and “basic event” are represented as
frames. Each frame has the following attributes/slots 1) name, 2) LDV, 3) LTV, 4) RTV, 5) RDV, 6)
user_input. Moreover, the frames that represent basic events have two extra attributes. The first
extra attribute refers to the corresponding advice/emergency response action. The second extra
attribute refers to the corresponding fuzzy importance measure. The first extra attribute helps to
manage the advice set in LOMA’s knowledge base. The second extra attribute is used to calculate
the fuzzy importance measure that is used by LOMA to display to the user the advice associated to
The number of the questions in each module is proportional to the number of basic events of the
corresponding operational problem. The questions are referring to subjects which are declared by
the corresponding basic events of the fault trees. Also, in every module specific sets of actions are
activated by the system to: a) display the questions to the user, b) calculate the estimated possibility
value of the top event, c) calculate the fuzzy importance measure of all basic events.
When a specific operational problem module is activated, the system displays the question set to the
user (see the 3rd stage in § 9). Each question updates the user_input value for the corresponding
31
Environmental Modelling & Software 24 (2009) 8–25
basic event frame. The user inputs are the linguistic values shown in Figure 5. Based on the user
input, and by using the demon set of the activated module, the quadruplet frame attribute of the
basic events [LDV, LTV, RTV, RDV] changes. Specifically, whenever the value of the user_input slot is
updated the changes of the quadruplet occur immediately after the update, due to the demons.
Afterwards, the system activates a set of actions. These actions are: a) estimating the possibility of
the top event based on the expressions {1} and {2}, b) calculating the importance measure of the
basic events based on the expression {3}, c) updating the quadruplet attribute values [LDV, LTV, RTV,
RDV] of the top event frame, d) updating the values of the importance measure attribute of all the
At this point the system displays the user inputs and the estimated failure possibility to the user (see
the 4th stage in § 9). The estimated possibility ranges between the LTV and the RTV attribute values
of the top event frame. During the prototype development the system utilized a defuzzification
method displaying one specific failure possibility value to the user. This method was changed after
suggestions made by the landfill managers. More specifically, the landfill managers were preferring
to have the system propose a range of the estimated failure possibility rather than a specific value.
Finally, the system asks the user if would like advice regarding the operational problem. If the user
consents, the system activates a specific action by using a rule, and displays the advice in a ranking
order, based on the basic events corresponding to fuzzy importance measure values (see the 5th
stage in § 9). In some cases one more action and rule set is activated in order to inform the user that
Table 3 shows the failure possibility of the top event of Figure 3 for two different sets of subjective
evaluations of the corresponding basic events. These subjective evaluations are in fact two different
32
Environmental Modelling & Software 24 (2009) 8–25
input sets in the LOMA system. Each input set is needed to calculate the failure possibility of the top
event by using {4}. Table 3 displays also the calculated fuzzy important measure values for α-level =
1 for each basic event and based on these values the rank of the corresponding advice is shown. For
the first set of inputs the failure possibility of the top event (i.e. the range between the LTV and RTV
values of the top event) ranges from 0.79 to 0.80. For the second input set the failure possibility
ranges from 0.48 to 0.56. This means that the operation of the landfill with respect to the
operational problem shown in Figure 3 is significantly less reliable given the first set of inputs in
Table 4 displays the results from the same basic events input sets. However, in this case the fuzzy
sets used to map the linguistic values are different as Figure 9 shows. The range of the calculated
failure possibility of the top event for the first input set is the same to the one shown in Table 3. For
the second input set, the proposed failure possibility of the top event is 0.52, which is the median
From the calculated fuzzy important measure values it is obvious that the measure of one basic
event depends basically upon: a) its position on the fault tree, b) its relation with the top event, c) its
corresponding value. For the first input set, the advice rank of every basic event in both tables is the
same, while for the second input set there was as shift in the places 5 and 6 between the advice of
These numbers are showing the effect on the final result of LOMA, if the membership functions had
different shape. Obviously the results are different in some cases but these differences are not so
Nevertheless, the data collection regarding the optimum shape determination of the linguistic
values, as well as the determination of the corresponding defuzzification method of LOMA is a future
research goal.
33
Environmental Modelling & Software 24 (2009) 8–25
Although the shape determination of the membership functions is a bottleneck, there is no doubt
regarding the usefulness of linguistic variables in this case. The linguistic values provide a very
effective communication between the system and the user, especially in cases like landfill operation,
where the inputs aren’t easily measurable quantities and there isn’t any past data to allow the use of
probabilities. With linguistic variables the system can use the natural language to give a conclusion.
If crisp numbers were used as inputs, the results of LOMA would be also crisp numbers. Such input-
output relations are all the LDV, LTV, RTV, RDV rows of the Tables 3 and 4. The user confusion
between the notion of probability and possibility was a rising problem regarding the use of crisp
inputs. Consequently, the results were usually misinterpreted. The confusion of these notions made
the experts to suggest that the system should propose a range of failure possibility.
A presentation of LOMA’s operation is outlined in below. The fault tree shown in Figure 3 is used in
an illustrative scenario. The scenario has as follows. An inexperienced landfill manager has been
informed that heavy rain is approaching. Given this early warning signal the landfill manager is
“sensing” that problems might occur in the landfill but he does not know exactly what might
happen. Therefore he is seeking information about the: a) potential landfill operational problems
due to heavy rain, b) degree to which the landfill is “vulnerable” with respect to a specific problem
(in this example the problem is the top event in Figure 3) , c) advice on how to react to each
problem. To find the help and guidance that he is looking for, he is using LOMA.
LOMA’s operation can be divided in five stages (see also the brief description in
http://loma.civil.duth.gr/occurrence_intro.htm).
1st stage: The landfill manager is accessing LOMA’s web site and activates the EWS. At first the
system is urging the landfill manager to choose one or several starting events that currently occur or
are expected to occur in his landfill. The 88 starting events that currently are stored in LOMA's
34
Environmental Modelling & Software 24 (2009) 8–25
knowledge base are listed in 12 categories, such as, the collection vehicles, the working front, the
landfill gas, etc. The user is able to choose several events from these categories. At this stage the
2nd stage: Based on the selected starting events the system is displaying a list of possible operational
problems. Afterwards, the user can choose a problem from a list for further analysis. In this scenario,
the starting event is the “rainy weather”, the list of the possible operational problems is shown in
Figure 10 and the landfill manager selects the “UNCONTROLED STORM WATER FLOW” for further
analysis.
3rd stage: The system activates the operational problem module that analyses the problem selected
by the user for further analysis. Then the system is asking for more information regarding the
working conditions in the landfill. The questions are referring to the basic events of a corresponding
fault tree.. Referring to the scenario, the system activates the operational problem module that
analyse the “UNCONTROLED STORM WATER FLOW” operational problem and ask information about
the corresponding basic events. The landfill manager provides the first set of linguistic inputs shown
in Table 3.
4th stage: Whenever the entire question set of a corresponding operational problem is answered, the
system is displaying the estimated failure possibility. At this stage in the scenario, LOMA displays the
5th stage: If the user wants, the system can display a corresponding set of advice and/or emergency
response procedures. Referring to the scenario and as it can be seen in Figure 12, each advice
corresponds to a specific basic event. Also, the corresponding solutions are displayed in a ranking
manner based on the fuzzy importance measure value of each basic event. Moreover, the system
informs the user if the analyzed problem could possibly generate any other operational problem/s.
35
Environmental Modelling & Software 24 (2009) 8–25
10. VALIDATION
Validation is the process of checking whether the software system meets the actual requirements of
the users (Peers, 2001). The validation of LOMA was performed for one-month period in a different
landfill than the one in which the knowledge acquisition took place. This strategy was chosen mainly,
because LOMA had the opportunity to be evaluated, not only regarding the contents of the
knowledge base, but also for user friendliness and ease of operation by different landfill managers.
The validation was contacted in the landfill of the city of Larisa in Greece which serves around
160.000 inhabitants, covers an estimated area of 80 ha, receives around 60,000 tons/year of waste
and follows a quite similar process of landfilling to Thessalonika’s landfill that was used for
observation during the knowledge acquisition process. Specifically, the validation process was
completed in two phases. The goal of the first phase was to validate the fault trees in terms of their
The validation process was done following a semi-structured interview with the landfill manager. In
particular, the manager was asked by the tester/validator to describe the causes and also the
mechanisms of the operational problems that were analyzed during the knowledge acquisition
phase. Several times during this process, the landfill logs and the book of incidents were referred
and used. These books however were written in an unstructured way providing useful information
however, for identifying the causes of the problems but not their mechanisms. That resulted to the
validating the acquired knowledge. During this phase three fault trees were been enriched with new
intermediate and basic events and new advice were been added to each basic event. This resulted to
add additional basic event frames in the knowledge base, to add extra starting events in the central
module of LOMA, and to accordingly change the formula that calculates the possibility of the top
event in the corresponding operational problem module. In addition, in five fault trees a logic gate
36
Environmental Modelling & Software 24 (2009) 8–25
has been modified from OR to AND and vice versa. This resulted to apply some changes to the
formula that calculates the top event possibility in the corresponding operational problem module.
The aim of the second phase was to compare LOMA performance with respect to a set of test cases.
The only precondition for these test cases was to access LOMA with a dial up connection so that to
evaluate the response of the system. The knowledge engineer prepared written test cases in which
the step by step actions that had to be done by the landfill managers were written together with the
input data and with the expected results. Whenever there was an unexpected result the experts
were informing the knowledge engineer and he was recording it. In addition, during this phase the
landfill managers were allowed to express their comments on issues like the design of the web
Table 5 displays the suggestion categories made by the experts and the corresponding corrective
actions. The majority of corrective actions were implemented immediately after the expert
suggestions. The landfill experts didn’t make any corrective suggestions regarding the system
advices. In some cases they were thinking to use some of the proposed advices in the field. That was
expected, since all advises LOMA uses, have been derived from references displayed on Table 1.
a) a useful tool for the inexperienced landfill manager, because correlates events that are
common during landfill operations (i.e. starting events) with several operational problems
with the ones proposed by LOMA. The experts made this comment only when they
apprehended the meaning of the output (i.e. the ease with which the operation can fail, and
37
Environmental Modelling & Software 24 (2009) 8–25
not how possible is any operational failure). This comment was the reason to maintain the
d) friendly to use, since LOMA provide explanations to several questions and provide
e) easy to upgrade, since the availability of the ES through the Internet can keep the developers
constantly up to date with new knowledge, provided by world experts in the form of
comments.
The field testing of the system proved to be very beneficial on the following points: a) error
identification and correction of the knowledge base, b) improvements regarding the system
Additionally to the validation phases described above, the development team has created a
database of synthetic events of landfill operational problems, based on descriptions of real landfill
problems found in traditional media and in sources in the world wide web. The developed database
includes around 70 synthetic events which are in fact deterministic scenarios of potential landfill
operational problems. In below, we briefly present the method of developing the synthetic events
and the way of validating the knowledge base of LOMA with these. As example is used the problem
“SUBSURFACE FIRE”
At first we searched for case histories in subsurface fires in landfills. A significant number of case
histories for landfill fires were found in the landfillfire web site [Landfillfire, 2007]. Specifically in
the Vancouver landfill fire that occurred on the 18th of October 2000 the forensic review established
concluded that the fire was triggered by spontaneous compaction of the buried combustible
material. Gaps in the intermediate cover soil allowed entry of oxygen into the waste promoting high
38
Environmental Modelling & Software 24 (2009) 8–25
temperature aerobic decomposition, exothermic pyrolysis and eventually a full scale fire. The
information about the causes of the operation problem was stored in the data base. The database
Afterwards LOMA was tested against the database of the synthetic events in terms of the degree to
which the causes and the early warning signals stored in the database were able to be identified by
LOMA whenever the corresponding operation problems were analyzed by it. In almost all analyzed
operational problems LOMA identified 80 to 100% of the causes and the early warning signals of the
synthetic event database. There was one exception with respect to the event of injuries of
personnel. That was partially expected due to the large number of coincidences that can occur and
could cause personnel injuries in different phases of landfill operations. Regarding this operational
problem it was concluded that additional analysis is required to be included in the knowledge base
of LOMA.
This paper was made having as objective to look the potential of providing an early warning service,
inexperienced landfill manager, using known technologies like ES, fault tree analysis and possibility
theory. Additional goal was to disseminate the acquired knowledge about operational problems in
landfills to a wide group of people who is interested for it. The development and operation of LOMA
has proved that these objectives can be met combining together the selected technologies.
LOMA is the first intelligent system providing early warning services in the context of SWM. A
significant advantage of LOMA and of its methodology is that uses fault tree analysis, a well known
and widely used risk and reliability analysis technique, as basis for knowledge acquisition, modelling
and representation. This characteristic makes the development methodology of LOMA easier to
grasp by a wider group of developers. Another advantage is that the knowledge represented in it can
39
Environmental Modelling & Software 24 (2009) 8–25
be used by the public who is interested in landfill operations. This characteristic of the methodology
is very convenient from the point of view of organizations that are obliged to accord with legislative
We argue, however, that LOMA services can be enhanced significantly. In particular, there are
thoughts on adding to future versions of LOMA additional services. For example, an on line
knowledge acquisition facility used by authorised users looks feasible to be implemented. The
implementations of a web forum and a wiki aiming at enhancing the analysis of common operational
problems by the users, in a way that will not affect the functionality of the early warning service
In addition, thoughts on looking at deeper in some concepts like monitoring, simulation and
forecasting and at how all these can be integrated with LOMA are looking to be feasible. Also some
thoughts on providing support in predicting conditions that are evolving with time in landfills have
been considered. One way to do this is by establishing access between the knowledge base of LOMA
and mathematical models that can forecast evolutionary behaviour of the conditions of interest,
whenever specific operational preconditions are met. The support of real time monitoring at the
operational level of the organization has been considered. This feature in particular looks feasible to
be implemented in a future version of LOMA following the model presented in (Goodall et al., 2008).
Among LOMA’s goals was to propagate the knowledge on landfill problems. Regarding this service
the web statistics are showing that, in average, around 170 unique users are visiting LOMA each
month seeking information and knowledge on landfill operational problems, while around 35% of
those bookmark LOMA. This is an indicator showing that there is a need for a knowledge
dissemination service on landfill operations and, as it was expected, the web is a very effective
technology to satisfy that need. However, these statistics impose to us to think of ways to extent
the service levels not only form the artificial intelligence point of view but also from the web
40
Environmental Modelling & Software 24 (2009) 8–25
engineering point of view in order to improve the web interface in terms of usability and accessibility
In concluding, this paper has proved that knowledge based systems can provide early warning
services in the SWM case. It has also described some methodological elements in supporting the
development of knowledge based early warning services, which are generated and directed from the
strategic level of an organization. The main novel characteristic of LOMA, compared to the related
EWS is that of the use of fault trees and frames as knowledge modelling, representation and
reasoning technologies respectively. This difference has allowed LOMA to be configured so that to
assess the possibility of occurrence of operational problems, and in addition to this, to provide a
very valuable service that the other EWS do not provide, which is the listings of corrective actions
and emergency response procedures to operational problems in a ranking manner thanks to the
These actions can be implemented by the personnel of the organization at the tactical and
operational level to avoid or to reduce their risk and to be prepared for effective response but also
by the people living close to landfills, which are affected by their operations. By providing this extra
service, LOMA complies more with the definition of EWS when compared to the relevant systems.
In light of this, it can be concluded also that fault tree analysis looks very promising in its use as a
basis of a graphical notation platform within a knowledge acquisition, modelling and representation
framework for EWS in engineering facilities. Our future work is focused on further investigating the
potential of fault trees so that to be used as a modelling and designing platform for EWS.
Finally, a major future research task is to design and investigate suitable methodologies on how the
proposed system could be customized or adapted to changing landfill conditions through being
41
Environmental Modelling & Software 24 (2009) 8–25
ACKNOWLEDGMENTS
We would like to thank the anonymous reviewers for their constructive comments helping us in
improving significantly the manuscript. We would like to acknowledge also the personnel in the
landfills we were visited, for their help and guidance, and especiallyDr. K. Alivani and Th. Peridi.
Dr. Dokas, during the last phases of this work, was supported by the research project SCEWA (Grand
No 2007-DRP-2-S5), funded by the Irish Environmental Protection Agency under the DERP grand
scheme.
REFERENCES
Bass, J., 1986. Avoiding Failure of Leachate Collection and Cap Drainage Systems, Noyes Publications.
Bilgiç, T., Türksen, I. B., 1999. Measurement of membership functions: Theoretical and empirical
work. In: Dudois, D., Prade, H., (Eds), Handbook of Fuzzy Sets and Systems, Klauwer.
Bolton, N., 2002. Dealing with offensive loads. MSW Management, July/August,
<http://www.forester.net/mw_0208_landfill.html>.
Cho, H.-N., Choi, H.-H., Kim Y.-B. 2002. A risk assessment methodology for incorporating
uncertainties using fuzzy concepts. Reliability Engineering & System Safety 78(2) 173-183.
CLIPS, 2007 CLIPS: A tool for building expert Systems, home page
42
Environmental Modelling & Software 24 (2009) 8–25
Cox, E., 1999. The Fuzzy Systems Handbook, 2nd Edition, AP Professionals.
Dokas, I. M., Panagiotakopoulos, D. C., 2006. A knowledge acquisition process for analyzing
operational problems in solid waste management facilities. Waste Management and Research 24 (4)
332-344.
Dokas, I. M., Nordlander, T. E., Wallace, R.J., 2007. Fuzzy fault tree representation and maintenance
based on frames and constraint technologies: a case study. In: Proceedings K-CAP’07 Workshop on
Knowledge Capture and Constraint Programming, October 28–31, 2007, Whistler, British Columbia,
Canada.
Feliubadalό, J., Relea, F., 1995. Landfill fires: A review. In: Proceedings Sardinia 95 Fifth International
Fay A., 2000.A fuzzy knowledge-based system for railway traffic control, Engineering Applications of
43
Environmental Modelling & Software 24 (2009) 8–25
Fleming, G., Van der Merwe M., McFerren G. 2007. Fuzzy expert systems and GIS for cholera health
risk prediction in southern Africa. Environmental Modelling & Software 22 (4) 442-448.
Goodall, J. L., Horsburgh, J. S., Whiteaker, T. L., Maidment, D. R., Zaslavsky, I.,2008. A first approach
to web services for the National Water Information System Source, Environmental Modelling &
Grove, R., 2000. Internet-based expert systems. Expert Systems 17 (3) 129-135.
Hall, L. O., Kandel, A., 1991. The evolution from expert systems to fuzzy expert systems. In: Kandel,
Hoyland, A., Rausand, M., 1994. Systems Reliability Theory Models and Statistical Methods, John
<http://www.hse.gov.uk/statistics/overall/fatl0405.pdf>
HSE, 2004. Mapping health and safety standards in the UK waste industry RR240,
<http://www.hse.gov.uk/research/rrpdf/rr240.pdf>
44
Environmental Modelling & Software 24 (2009) 8–25
Iliadis L. S., 2005. A decision support system applying an integrated fuzzy model for long-term forest
http://www.unisdr.org/eng/library/lib-terminology-eng%20home.htm>
ISWA, 1998. Guidance for Landfilling Waste in Economically Developing Countries. ISWA.
Jess, 2007 Jess the rule engine for JAVA platform, home page http://herzberg.ca.sandia.gov,
Koerner, R.M., Soong T.-Y., 2000. Leachate in landfills: The stability issues. Geotextiles and
Kölsch, F., Fricke, K., Mahler, C., Damanhuri, E., 2005. Stability of landfills – The
Bandung disaster. In: Proceedings of the 10th International Landfill Symposium, Cagliari, Italy.
45
Environmental Modelling & Software 24 (2009) 8–25
Krom, A. H., Kok, B., Lelieveld, P., Mijdam, C. H., 1997. Consequences of excluding specific kinds of
waste on landfill stability. In: Proceedings Sardinia 97 Sixth International Landfill Symposium, S.
December 2007.
Lei, Z., Yamada, Y., Huang, J., Xi, Y., 2006. Intelligent early-warning support system for enterprise
financial crisis based on case-based reasoning. Journal of Systems Science and Complexity 19 (4) 538
– 546.
Lin, C.- T., Wang, M. J. J., 1997. Hybrid fault tree analysis using fuzzy sets. Reliability Engineering and
Liu, X., Kane, G., Bambroo, M., 2003. An intelligent early warning system for software quality
improvement and project management. In: Proceedings of the 15th IEEE International Conference
46
Environmental Modelling & Software 24 (2009) 8–25
LPA, 1996. flex Expert System Toolkit Reference, Logic Programmers Associates Ltd, 4th Edition.
Lukasheh, A. F., Warith, M. A., 2001. Review of expert systems (ES), geographic information systems
(GIS), decision support systems (DSS), and applications in landfill design and management. Waste
Makropoulos C. K., Butler D., 2005. A neurofuzzy spatial decision support system for pipe
Makropoulos C. K., Butler D., and Maksimovic C., 2003. Fuzzy logic spatial decision support system
for urban water management. Journal of Water Resources Planning and Management. 129 (1) 69-77.
Marsili-Libelli S., 2004. Fuzzy prediction of the algal blooms in the Orbetello lagoon. Environmental
McBean, E., Rovers F., Farquhar G., 1995. Solid Waste Landfill Engineering and Design, Prentice Hall.
McKendry, P. J., 1995. Environmental pollution: The origins of wind-borne litter. In: Proceedings
47
Environmental Modelling & Software 24 (2009) 8–25
Mickovski, S.B., Stokes, A., van Beek, L.P.H. 2005. A decision support tool for windthrow hazard
Milanov, V., Corade, J. M., Bruyat-Korda, F., Falkenreck , G., 1997. Waste slope failure analysis at the
Rabastens landfill site. In: Proceedings Sardinia 97, Sixth International Landfill Symposium, S.
Onisawa, T., 1996. Subjective analysis of system reliability and its analyzer, Fuzzy Sets and Systems
83(2) 249-269.
Pillay, A., Wang, J. 2003. Modified failure mode and effects analysis using approximate reasoning,
Preece A. D., 2001. Evaluating verification and validation methods in knowledge engineering, <
http://citeseer.ist.psu.edu/515529.html>
Ross ,T. J., 2004. Fuzzy logic with applications, 2nd Edition, John Willey and Sons.
Sancho-Royo, A., Verdegay, J. L., 1999. Methods for the construction of membership functions.
48
Environmental Modelling & Software 24 (2009) 8–25
Salhofer, S., Wassermann, G., Binner E., 2007. Strategic environmental assessment as an approach to
assess waste management systems. Experiences from an Austrian case study, Environmental
Wielinga, B., 2000. Knowledge Engineering and Management: The CommonKADS Methodology, The
Shiue, W., Li, S.-T., Chen, K.-J. 2007. A frame knowledge system for managing financial
decision knowledge, Expert Systems with Applications In Press, Corrected Proof, Available
Stamatelatos, M., Vesley, W., 2002. Fault Tree Handbook With Aerospace Applications. Technical
Suresh, P. V., Babar, A. K., Raj, V. V., 1996. Uncertainty in fault tree analysis: A fuzzy approach. Fuzzy
49
Environmental Modelling & Software 24 (2009) 8–25
Tomic, B., Jovanovic, J., Devedzic, V. 2006. JavaDON: an open-source expert system shell. Expert
Thomas, Β., Tamblyn, D., Baetz, B., 1990. Expert systems in municipal solid waste management.
Turban, E., 1995. Decision Support and Expert Systems Management Support Systems, 4th Edition,
Prentice-Hall.
Regulation of Noise at Waste Management Facilities, Version 3.0. UK Environment Agency, Bristol,
<www.environment-agency.gov.uk>.
Wilhelm, L., 1995. Fire protection and fire fighting on landfill sites, In: Proceedings Sardinia 95 Fifth
50
Environmental Modelling & Software 24 (2009) 8–25
Yang, B., Li, L.X., Ji, H., Xu, J., 2001. An early warning system for loan risk assessment using artificial
Yuhua, D., Datao, Y., 2005. Estimation of failure probability of oil and gas transmission pipelines by
fuzzy fault tree analysis. Journal of Loss Prevention in the Process Industries 18 (2) 83-88.
Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8 (3) 338 - 353.
Zadeh, L. A., 1978. Fuzzy sets as a basis for a theory of possibility. Fuzzy Sets and Systems 1 (1) 3 - 28.
Zimmermann, H. J., 1996. Fuzzy Set Theory and it’s Applications, 3rd Edition, Kluwer Academic
Publishers.
51
Environmental Modelling & Software 24 (2009) 8–25
FIGURES
52
Environmental Modelling & Software 24 (2009) 8–25
Figure 4: The concepts of “event” “basic event” “top event” represented in LOMA’s knowledge base
using frames
53
Environmental Modelling & Software 24 (2009) 8–25
Figure 5. Fuzzy sets representing linguistic values. Source (Lin et al., 1997)
54
Environmental Modelling & Software 24 (2009) 8–25
55
Environmental Modelling & Software 24 (2009) 8–25
56
Environmental Modelling & Software 24 (2009) 8–25
57
Environmental Modelling & Software 24 (2009) 8–25
Figure 11. The estimated range of failure possibility based on the user inputs
58
Environmental Modelling & Software 24 (2009) 8–25
Figure 12. An example of prioritizing the advice based on the fuzzy importance measure
59
Environmental Modelling & Software 24 (2009) 8–25
TABLES
Table 1: References to: a) landfill operation b) specific operational problems c)
advice/solutions about operational problems
Reference Title Document Type Type of Reference
Avoiding Failure of Leachate Collection and
(Bass, 1986) Book b and c
Cap Drainage Systems
(Feliubadalό, 1995) Landfill Fires: A Review Conference proceedings b and c
(Mc Bean et al., 1995) Solid Waste Landfill Engineering and Design Book a and c
Environmental Pollution: The Origins OF
(McKendry, 1995) Conference proceedings b
Wind-Borne Litter
Fire Protection and Fire Fighting on Landfill
(Wilhelm, 1995) Conference proceedings b and c
Sites
Consequences of Excluding Specific Kinds
(Krom, 1997) Conference proceedings b
of Waste on Landfill Stability
Waste Slope Failure Analysis at the
(Milanov et al., 1997) Conference proceedings b and c
Rabastens Landfill Site
Guidance for Landfilling Waste in
(ISWA, 1998) Book a, b and c
Economically Developing Countries
Training Sanitary Landfill Operating
(SWANA, 1998) Training Document a, b and c
Personnel
(ISWA Working Group
on Sanitary Landfill, Operations Guidelines Guidelines a, b and c
1999)
(Koerner, 2000) Leachate in Landfills: The Stability Issues Journal paper b and c
(UK Environment
Νoise Guidance Control Guidance b and c
Agency, 2002)
(UK Environment
Odor Guidance Control Guidance b and c
Agency, 2002)
(Bolton, 2002) Dealing With Offensive Loads Electronic journal b and c
Web site with case
Landfillfire.com Landfill fire case histories b
histories
60
Environmental Modelling & Software 24 (2009) 8–25
Table 3. Estimated possibility value of the problem “UNCONTROLLED STORM WATER FLOW”
for two different inputs sets, using the membership functions of Figure 6 and the
corresponding values of the fuzzy importance measures for α-level =1
st
1 Input Set
Basic Event E1 E2 E3 E4 E5 E6 E7 E8 Failure
Linguistic Value Possibility of Top
High Medium Low Very High High Fairly Low Fairly High Medium Event
(Input Set 1)
LDV 0.7 0.4 0.1 0.8 0.7 0.2 0.5 0.4 0.67
LTV 0.8 0.5 0.2 0.9 0.8 0.3 0.6 0.5 0.79
RTV 0.8 0.5 0.2 1 0.8 0.4 0.7 0.5 0.80
RDV 0.9 0.6 0.3 1 0.9 0.5 0.8 0.6 0.90
Importance
1.405 0.002 0.006 0.145 0.055 0.012 0.002 0.002
Measure
Advice Rank 1 6 5 2 3 4 6 6
nd
2 Input Set
Basic Event E1 E2 E3 E4 E5 E6 E7 E8
Failure
Linguistic Value Possibility of Top
Fairly High Very Low Very Low Medium Medium Low High Fairly High Event
(Input Set 2)
61
Environmental Modelling & Software 24 (2009) 8–25
Table 4. Failure possibility of the problem “UNCONTROLLED STORM WATER FLOW” for
two different inputs sets using the membership functions of Figure 9 and the
corresponding values of the fuzzy importance measure for the α-level =1
st
1 Input Set
Basic Event E1 E2 E3 E4 E5 E6 E7 E8 Failure
Linguistic Value Possibility of
High Medium Low Very High High Fairly Low Fairly High Medium Top Event
(Input Set 1)
LDV 0.65 0.35 0.05 0.8 0.65 0.2 0.5 0.35 0.61
LTV 0.8 0.5 0.2 0.95 0.8 0.35 0.65 0.5 0.79
RTV 0.8 0.5 0.2 1 0.8 0.35 0.65 0.5 0.80
RDV 0.95 0.65 0.35 1 0.95 0.5 0.8 0.65 0.95
Importance
1.410 0.001 0.003 0.145 0.026 0.006 0.001 0.001
Measure
Advice Rank 1 6 5 2 3 4 6 6
nd
2 Input Set
Basic Event E1 E2 E3 E4 E5 E6 E7 E8 Failure
Linguistic Value Possibility of
Fairly High Very Low Very Low Medium Medium Low High Fairly High Top Event
(Input Set 2)
LDV 0.5 0 0 0.35 0.35 0.05 0.65 0.5 0.30
LTV 0.65 0 0 0.5 0.5 0.2 0.8 0.65 0.52
RTV 0.65 0.05 0.05 0.5 0.5 0.2 0.8 0.65 0.52
RDV 0.8 0.2 0.2 0.65 0.65 0.35 0.95 0.8 0.74
Importance
1.127 0.007 0.007 0.374 0.374 0.213 0.016 0.009
Measure
Advice Rank 1 6 6 2 2 3 4 5
62
Environmental Modelling & Software 24 (2009) 8–25
Table 5: Problem categories identified by the experts during the validation of the system and
the corresponding corrective actions
No Problem Categories Corrective Action
Explanations were added to the questions pointed
1 Specific questions were vague out by the experts. The explanations are available
to the user via an explain button
The knowledge base module was divided to
operational problem modules (see §7 and 9).
2 Quicker system response
Before that, all the modules of the system were
part of one module
Explanation of the failure possibility notion was
Explanation regarding the failure added. This explanation is available via a link,
3
possibility notion is needed whenever the system displays the estimated failure
possibility of the operational faults
Wrong activation of operational faults Correction of the starting event sets that were
4 after the selection of specific starting activating the corresponding operational problem
events modules
Development of extra starting event The central module was updated with extra starting
5
categories event category frames
Addition of starting event/s for the
Update of the starting event set that activates the
6 activation of specific operational
corresponding operational problem
problems
Correction of the corresponding demon set, or
Wrong calculations during the failure correction of the failure possibility mathematical
7
possibility estimation expression, based on the corresponding derived
fault tree.
Wrong operational problem
activation whenever the user was Correction of the corresponding rules of the central
8
selecting a specific fault for further module
analysis from the corresponding list
63