Design and Development of a Maintenance
Knowledge-Base System Knowledge-Based on CommonKADS Methodology
Alireza Arab Maki
Navid Shariat Zadeh
Master Thesis
Department Production Engineering and Management
School of Industrial Engineering and Management
Royal Institute of Technology
SE-10044 Stockholm, Sweden
Abstract
The objective of this thesis is to design and develop a knowledge base model to support the
maintenance system structure. The aim of this model is to identify the failure modes which
are the heart of maintenance system through the functional analysis and then serves as a
decision support system to define the maintenance tasks and finally to implement a preventive
maintenance task. This knowledge base management system is suitable to design and develop
maintenance system since it encompasses all necessary steps in maintenance area. Moreover,
it is capable of being integrated with other knowledge base systems. The structure and the
environment of this knowledge base system is flexible to allow users to deploy different kinds
of tools which they will. It is also a well structured approach to develop, debug, upgrade and
trace.
In this thesis, the CommonKADS methodology is used as the knowledge base methodology.
At the first step, a knowledge base system is developed to create the maintenance system
infrastructure. To implement this, Reliability-Centered Maintenance (RCM) has been chosen
as the method to design a maintenance system. In order to make it more specified, a Spindle
subsystem is taken as a case study to make the model clearer. Secondly, another knowledge
base system is developed for decision making process to select the suitable maintenance task
and finally, a general knowledge base model is developed for condition-based monitoring on
Spindle.
In chapter 1, background, previous works and gap analysis have been surveyed. Then in
chapter 2 the methodology and tools have been discussed and described. Design and
development the knowledge base for maintenance system is described in detail in chapter 3
and finally in chapter 4, the conclusion and the future works are presented.
Keywords: Knowledge base systems, CommonKADS methodology, Maintenance,
Acknowledgement
During the work with this thesis, we have received a lot of help from Mr. Jerzy Mikler as our
thesis supervisor. We would like to express our sincere gratitude to him. Also we would like
to thanks the officials of Production Engineering and Management department for preparing
appropriate environment to work and research. We want to thank our parents for their
financial and never ending support, for the help in our studies and for its success.
Alireza Arab Maki and Navid Shariat Zadeh
Stockholm, June 2010
Table of Contents
Chapter 1 – Introduction 1 Introduction ………...………... 2 1.1 Background ………... 2 1.2 Previous Researches ………..………... 3 1.2.1 Condition-Based Monitoring ………... 31.2.2 CBM using Neural Network ………….………... 5
1.2.3 CBM using Bayesian Network ……….………... 7
1.3 Gap Analysis ……….………... 10
1.4 Thesis Objectives ………..………... 11
1.5 Delimitation ………..………... 11
Chapter 2 – Methodology and Tools 2 Methodology and Tools ………... 13
2.1 The Methodology ………..………... 13
2.1.1 Model Structure ………... 13
2.1.2 Some Terminology ………... 16
2.1.3 Organizational Model ………...………... 16
2.1.4 Impact and Improvement Analysis: Task and Agent Modelling …………..……... 19
2.1.5 Recommendations and Actions ……….………... 23
2.1.6 Knowledge Model ……….………... 23
2.1.6.1 Role of Knowledge Model ………... 23
2.1.6.2 Knowledge Model Overview ………... 23
2.1.6.2.1 Domain Knowledge ………..………... 24 2.1.6.2.2 Inference Knowledge ………... 25 2.1.6.2.3 Task Knowledge ………...………... 25 2.2 Reliability-Centered Maintenance (RCM) ………... 27 2.2.1 Background ………...………... 27 2.2.2 Functions ………...………... 28 2.2.3 Functional Failure ……….………... 28 2.2.4 Failure Modes ………...………... 28
2.2.5 Symptoms and Consequences …….………..………... 28
2.2.6 Failure Management Techniques ……..………... 29
2.2.7 Task Selection Process ………..………... 29
2.3 Integrated Condition Monitoring and Fault Prognosis .………... 31
2.3.1 Measurable Parameters ……….………... 32
2.3.2 Selection of Proper CBM Equipment ...………... 32
2.3.2.1 Energized Testing ……….………... 32
2.3.3 The System to Analyze the Collected Data ..………... 33
2.4 CUSUM Control Charts ………... 33
2.4.1 Description ………... 34
Chapter 3 - Design and Develop the Knowledge Base for Maintenance System 3 Design and Develop the Knowledge Base for Maintenance System ………... 36
3.1 General Description ………..………... 36
3.2 Organization Model Analysis ………...………... 36
3.3 Knowledge Intensive Task Model Analysis ………….………... 39
3.3.1 Task 2 - Analyze the Machinery and Determine the Failure Modes ………... 39
3.3.2 Task 3 - Realizing the Maintenance Task .………... 54
Chapter 4 - Conclusion and Future Works
4 Conclusion and Future Works ………..………... 86
4.1 Conclusion ………... 86
4.2 Future Works ………... 86
Appendix A – FMEA of Spindle System …...….………... 87
Chapter 1
Introduction
Chapter 1 – Introduction
1 Introduction
1.1 Background
In order to run a successful manufacturing company, continuous improvement must be
considered and implemented in the areas of safety, quality and reliability. To achieve this, one
of the most important processes which must be subject of improvement is maintenance
process. Improvement of safety, quality, reliability and dependability in a plant is directly
associated to the maintenance system in a company. Amelioration in these areas will result in
cost reduction and more internal and external customer satisfaction which helps to create a
competitive market. Moreover, it has been proved that the maintenance cost is one of the main
parts of life cycle cost of a product or asset.
There are two main types of maintenance task:
-
Corrective maintenance: The strategy behind this approach is to let the equipment
fail (also called run to failure) and then start to cover the damaged equipment. This
approach is suitable just in case that the equipments’ failures’ consequences are not
neither safety nor environmental.
-
Preventive maintenance: The strategy behind this approach is to maintain the
equipment before the failure occurs. This approach has two main methods like
time-based maintenance and condition-time-based monitoring. This approach is suitable when an
unplanned stoppage of the equipment result in high equipment downtime, high cost of
repairing equipment, extensive repair time and high penalty associated with the loss of
production which all decrease the effectiveness and efficiency of the factory
dramatically.
Nowadays, the question is that how to design a maintenance system and make a decision
about the suitable maintenance strategy. There are many methods to achieve the mentioned
purpose. Failure Mode and Effect Analysis (FMEA) has been used to analyze all the possible
failure modes. Statistical analysis using the historical data has been used for time-based
maintenance and recently thanks to the technological improvement in the areas of sensors,
data capturing and analysis software and hardware; many plants utilize computer control
systems for condition-based monitoring of their equipments. Also, Reliability-Centered
Maintenance (RCM) technique has been adopted as a strong method to analyze the functions,
functional failures, failure mode and consequences. This technique includes decision rules to
determine different maintenance tasks such as redesign, run to failure, scheduled restoration,
etc.
Implementation and application of such method for maintenance result in generating huge
amount of data, information and knowledge which must be managed in a proper way. This
information management must contain generating, storing, analyzing and dispatching the data
among whole process. The main works which are done in this area are briefly presented in
next section (1.2) however; the lack of a well-structured comprehensive model to design and
implement a maintenance management system is noticed. Therefore, the aim of this thesis is
to design a knowledge base management system in maintenance area which encompasses
necessary tasks in the areas such as failure mode analysis, maintenance task decision making
and finally supporting the implementation of selected task.
To achieve this goal, the following methodology, techniques and assumption are used in this
thesis:
CommonKADS Methodology: This methodology is used to deal with the knowledge base
management and it is described in Chapter 2. The reasons to choose this methodology are:
-Gradually extension of the methodology as a result of feedback from practitioner and
scientists over the years,
-
It is not a technology-push approach
-
Its ability to model the complex systems taking easy steps.
Reliability-Centered Maintenance (RCM) Technique: As described in Chapter 2, this
technique has been chosen because of its interesting nature of analysis which starts from the
system functions and allow to define failure modes.
Condition-Based Monitoring (CBM): This method is a dynamic preventive maintenance
which is able to online monitoring of systems and detects degradation on components in early
stages. Thus, one of its techniques called multi-sensor and multi-parameter condition
monitoring and fault diagnosis is chosen to design the knowledge base management system.
To show how to implement this knowledge based maintenance system, a spindle system has
been chosen as a case study.
1.2 Previous Researches
Many researchers and authors designed and developed knowledge based systems in
maintenance area. These knowledge based systems are called expert systems in their
researches literature. Most of the jobs which are done in this area can be categorized into three
areas of Condition-Based Monitoring (CBM), condition-based monitoring (CBM) using
neural network and condition-based monitoring (CBM) using Bayesian networks.
1.2.1 Condition-Based Monitoring
Condition Based Monitoring or CBM is a type of preventive maintenance (PM) where, it
monitors the condition of specific areas of plant and equipment. This can be done
automatically with the use of instrumentation such as machinery vibration analysis and
thermal imaging equipment or manually. This is particularly suited to continuous process
plants where plant failure and downtime can be extremely costly. During the past years, CBM
has totally changed thanks to measurement technology development [1]. In order to develop a
CBM system, the following steps are taken:
Step 1 - Identifying the parameters to monitor: In order to determine the monitoring
requirements, first the failure modes which would be expected to find must be listed.
Secondly, the measurable parameters which can show the presence of approaching failures
should be identified. This is usually done by vibration analysis, temperature and pressure
analysis [2].
Step 2 - Determining the way to measure and measurement instruments
Step 3 – Identifying the monitoring strategy: These strategies include setting of alarm limit
for an unacceptable percentage change in a parameter value and parameter or data that
deviates from normal operating conditions.
Elements of a typical CBM system are as below [1]:
- Database: Since a huge volume of data is generated in CBM, the existence of a database is
necessary to store and retrieve the monitoring information.
- Routs: The major advance in recent years in collection of CBM data has been the advent
battery power data collector. These are digital instruments which are able to be programmed
with a series of criteria for each measurement point. The effect of these data collectors has
been to minimize the amount of interactions required by a user in collecting machinery
condition information. A major example of this, is the organization of data, which are desired
to be collected, into a logical sequence or route around the machines. The essential element of
the route management is the scheduling of the right measurement at the right time.
- Data Input: Although the data collector can collect the vast majority of data, there is a need
to integrate the data from other sources.
- Reports: a major focus of the maintenance system is to provide information on machine
problems at the earliest possible time and ignore all data on machines that are operating
within normal bonds. Communication of the information that a machine is showing abnormal
behaviour is usually achieved through a report [3].
- Plots: when a condition indicator shows deviation beyond the expected norms, the next
stage is usually investigation of the exceptions. This investigation must firstly verify that it is
indeed a steady trend toward the break down and secondly determine the cause of the
deviation. To do this, spectrum pictures including plots known as waterfall maps showing
historical spectrum data are used. This is especially suitable when vibration is a parameter in
CBM.
1.2.2 CBM using Neural Network
Artificial neural network simulation can be defined as a parallel distributed processor that can
store experimental knowledge and make it available for future use. The knowledge is obtain
through a learning process and stored in inter-neuron connection strength which are called
synoptic weights. This learning process can be supervised or unsupervised with respect to
pattern classification. Supervised learning is used when the classification of all training input
pattern is known. For example each training pattern would be classified as a normal mode
pattern or a pattern representing one of the fault modes. This is the context of fault diagnosis
problem which must be investigated. Figure 1.1 shows the general approach to use the neural
network in CBM.
R. P. Leger et. al., 1996 [4] used artificial neural network for fault detection and diagnosis.
Their proposed FDD strategy was tested on a model of heat transport system of a CANDU
nuclear reactor. Their model monitored the temperature, flow and pressure of different
components of a nuclear reactor. Adyles Arato Junior et. al., 2010 [5] designed a neural
network model which uses vibration signals in order to condition-based monitoring of a
power plant. Their model is able to detect the faults in early stages and automatic diagnosis of
the root causes. R. C. M. Yam et. al., 2001 [6] used a CBM system by adding the capability of
intelligent condition-based fault diagnosis and the power of predicting the equipment
deterioration. Chang-Ching Lin and Hsien-Yu Tseng, 2004 [7] used a neural network
application for reliability modelling and condition-based predictive maintenance. They
quantified the time between maintenance (TBM) using the Weibull distribution. The
difference between this application and traditional Weibull analysis is that they assumed a
reliability parameter as the output of neural network [8].
Neural networks which are used in CBM, belong to two main categories. The first one can
store sequential information in the form of historical data and can be used in forecasting. For
example, in an recurrent neural network (RNN) as shown in figure 1.2, the input nodes are
taken as the value of the current condition Xt and values of previous time-series condition
(X
t-1, X
t-2, X
t-3, …, X
t-d, …, and X
n). The value of the output (X’
t +1) can provide a one-step
ahead prediction of a time-series condition, which is a function of the current value X
tand
time-lagged values of the previous condition (Xt-1, Xt-2, …, Xt-d, …, and Xn). This model is
suitable when the condition-based monitoring is done based on one parameter such as
vibration. The predicted value X’
t +1of a time series, one-step ahead in the future, is given as:
X
’t +1 = F(Xt-1, Xt-2, Xt-3, …, Xt-d, …, and Xn)
where,
d is time lag
X
’t +1 is the predicted value
X
tis the value of current condition
X
t-d is the values of previous condition lagged by time dThe intermediate neuron may be represented mathematically by the following equations:
∑
==
p j j kj kw
x
0υ
y
k=
φ
( )
υ
kwhere
p = total number of inputs to neuron k,
wkj = input weights to neuron k,
Xj = output values from the previous layer,
v
k= input to the transfer function of neuron k,
ø = transfer function of neuron k,
yk = output from neuron k.
The second category has the same structure but the intermediate nodes can be of several
layers and the inputs can be the status different parameters [9].
Neural Network is a strong approach for trend prediction in condition based monitoring
however it should be trained by either historical data of a machine in a normal condition or
defining some fault scenarios which both lead to increase the uncertainty of the model.
Figure 1.2 – Neural Network with three layers used for fault trend prediction. The nodes in the first layer show the parameter value in the predefined time lags and the last node in the third layer shows the predicted value.
1.2.3 CBM using Bayesian Network
In the early stages of maintenance knowledge, many efforts have been taken to quantify the
reliability of systems in order to estimate the time intervals for implementing the preventive
maintenance actions. Traditionally, this is done by gathering the component failure data from
the historical data and estimating probability distribution of components life time, and finally
predicting a components life time by choosing certain reliability. Nowadays, this could not be
the case because of lack of historical failure data since the aim of modern maintenance system
is not to let the equipments run to failure. One of the approaches to deal with this problem is
simulation. Monte Carlo simulation has been applied in the situation of possessing low
amount of information and data [10]. However, in some critical systems such as nuclear
power plants and other power plants, the reliability interaction of different subsystems and
components is so complex which can not be modelled using traditional statistical approach
and Monte Carlo simulation is not suitable. To cope with this problem, when the focus is on
quantifying the reliability, researchers have designed Bayesian Networks using expert
opinions for preventive maintenance.
Bayesian Networks, known as BNs, provide an efficient way to represent the degradation
process of an industrial system or machines. BNs are some graphical model introduced by
Pearl [11] and
Lauritzen and Spiegelhalter [12]
.
BNs represent a relation between graph and
probability theories. The random variable of the probabilistic model is described with the
vertices of the graph where edges describe the dependencies measured by conditional
probability.
G. Celeux et. al., 2005 [13]
used a Bayesian Network to present a nuclear plant mechanical
system degradation. By applying the Bayesian Network, they could identify most critical parts
with low reliability and analyze the influence of maintenance actions on the reliability of
different components. Haitao Guo et. al, 2009 [14]
applied Bayesian network for reliability
analysis of wind turbines with incomplete failure data collected from after the date of initial
installation. H. Boudali and J.B Dugan, 2004 [15] presented a Baysian network reliability
modelling and analysis framework. Their Bayesian network is the discrete time model. They
calculated failure probability and reliability according to discrete time simulation.
Bayesian Networks describe conditional independence and analyze probable casual
relationship between random variables. Figure 1.3 represents a Bayesian Network for a
sub-components of a reactor coolant pump observed on the French nuclear plants. Random
variables are represented by the vertices of the graph and indicate the state of a component i.e.
healthy or damaged.
Figure 1.3 – Bayesian Network of a system degradation process
The above Bayesian Network is built from experts opinions. These Bayesian Network
contains 22 discrete variables. 17 variables are binary and the other five ones have three
states. Variables have been divided in 4 sets. environmental variables A = {PI3,Ad,Ab, PI6,
PI4,Ag,DJ, PI2,DI}, degradation variables M = {M1′,M1′′,M2,M3,M4,M5,M6}, observation
variables O = {O1,O2,O2′,O2′′,O5} and finally the variable of interest: the state of the system
(E).
Now some definitions are given as below:
Definition 1: A directed graph is a couple G = (V,E), where V = (X
1, . . . ,X
n)
denotes the
vertices of the graph and E = (e
1, . . . , e
m)
denotes a part of Cartesian product V × V , where
ei is called the edges of the graph.
If (X
i,X
j)
lies in E, then this element is called an edge. It is denoted X
i→ X
j, X
iis called the
source and X
jthe target of the edge. For directed graph, the parents and the children of the
vertices are defined as follows:
Definition 2: If a directed edge has source X
iand target X
j, then X
iis called the parent of X
jand X
jis called the son or child of X
i. The set of the parents of X
jis denoted pa(X
j)
and the set
of children of X
iis denoted ch(X
i)
.
In a directed graph, the oriented paths are defined as follows:
Definition 3: An oriented path is a set of distinct vertices X
i, . . . , X
jsuch that (X
k−1,X
k)
is an
edge for all k = i + 1, . . . , j. This path is denoted X
i→ X
j. A cycle is a path such that X
i= X
j.
Directed graph without cycle are called Directed Acyclic Graphs (DAG).
Now the Bayesian Network is defined as below:
Definition 4: A Bayesian Network is
- a set of variables V , defining the vertices, and a set of edges between variables E,
- each variable has a finite number of exclusive states,
- variables and edges define a directed acyclic graph, denoted G = (V,E),
- for each variable Y with its parents X
1, . . . ,X
n, is associated a conditional probability
P(Y | X
1, . . . ,X
n)
. When a variable has no parent, the last quantity becomes a marginal
probability P(Y).
The denomination ”Bayesian Networks” comes from the well-known Bayes theorem. In a
BN, the joint probability can be written as follows (recursive factorization):
(
)
∏
(
( )
)
==
n i i i nP
X
pa
X
X
X
P
1 1,...,
Where pa(X
i)
is the set of parents of vertex X
i.
The last node E in figure 1.3, shows the current state of whole system. In order to calculate
this, experts must carefully assign conditional probability of nodes. Then by calculating the
conditional probability of different nodes, the most influential components which affect the
whole system reliability are determined. Also it is possible to add another node such as
maintenance tasks in order to analyze their influence on the system reliability.
Bayesian Networks are usable for complex simulation of real world and can hold the
knowledge in form of collections of probability so they can simulate human intuition and
conclusions and are easy to apply for model and simulation tests. They are also readable for
human in contrast to neural networks and preserve knowledge of experts. On the other hand,
they are less exactly that neural networks but more efficient and it is difficult to examine the
solution of the network. Also it is difficult the get the probability knowledge.
1.3 Gap Analysis
By considering knowledge base application in different maintenance tasks which are
explained briefly in previous sections, following shortages could be concluded:
- There is a lack of a suitable standard to design a knowledge base system for
maintenance.
- All the current knowledge base systems concentrate on one area of maintenance
systems such as CBM without considering other tasks of the whole maintenance
system and their interactions. Even in one task like CBM, there are different
knowledge base systems for example to monitor vibration and oil debris there are two
different knowledge base systems because the system vendors are different so they use
different measurement technology and difference knowledge base models.
- Maintenance system is one of the systems in a plant so the maintenance knowledge
base system should be easily integrated with the other knowledge base system in the
plant.
Regarding mentioned shortages it would be worthwhile to create a comprehensive
infrastructure to design a knowledge base system in maintenance area. This system must have
the following characteristics:
- It should have a suitable environment to gather all the methods which are previously
deployed in maintenance area such as neural network, Bayesian Networks, etc.
- It should cover all the maintenance tasks such as failure mode analysis, decision
making on maintenance task and implementation within the maintenance area.
- It should have the potential to upgrade by adding different knowledge base systems
within the maintenance area because designing a knowledge base system in
maintenance area can be implemented gradually not as the same time.
- Methodology and methods used to develop such a system must be easy to apply as
well as flexible to use in other knowledge base applications.
- It should be capable to import other knowledge base systems.
- It should have a learning process in order to update itself based on knowledge and
information generated during run time.
1.4 Thesis Objective
In order to design a model which fulfils the above characteristics, it is decided to use the
CommonKADS methodology as the knowledge base model. At the first step, a knowledge
base system is developed to create the maintenance system infrastructure. The aim of this
system is to identify the failure modes which are the heart of maintenance system. To
implement this, Reliability-Centered Maintenance (RCM) has been chosen as the method. In
order to make it more specified, a Spindle subsystem is taken as a case study to make the
model clearer. Secondly, another knowledge base system is developed fro decision making
process to select the suitable maintenance task and finally, a general knowledge base model is
developed for condition-based monitoring on Spindle. This model is general in order to
present the path which should be followed to model more complicated and complex systems.
1.5 Delimitation
Since the technical data, information, drawing and components of the Spindle system were
not available, in this thesis a typical Spindle system and its components has been assumed for
the case study. Since the aim of this thesis is to create a general structure and environment to
design the knowledge base system which satisfies the mentioned characteristics in previous
section, this limitation does not interfere with the objectives of the thesis. In other words, the
proposed model is capable to be customized according to the users needs. Moreover, it is
assumed that the feasibility study results are positive to implement this model.
Chapter 2
Chapter 2 – Methodology and Tools
2 Methodology and Tools
2.1 The Methodology [16]
CommonKADS methodology is consist of a number of elements. These elements are
illustrated in the form of a pyramid. The methodological pyramid has five layers, where each
consecutive layer is built on top of the previous ones shown in Figure 2.1.
There are a number of principles and rules concerning each layer. The significant issue for
developing each layer is to form the baseline and rationale of the approach. These principles
are based on the lessons learned about knowledge-system development.
Knowledge engineering is not some kind of “mining from the expert’s head,” but consists of
constructing different aspect models of human knowledge. [CommonKADS Book]
world view
theory
methods
tools
use
feedback
case studies application projects CASE tools implementation environments life-cycle model, process model, guidelines, elicitation techniquesgraphical/textual notations worksheets, document structure model-based knowledge engineering
reuse of knowledge patterns
Figure 2.1 – CommonKADS Methodological Pyramid
2.1.1 Model Structure
Figure 2.2 presents the CommonKADS model structure that is the practical expression of the
principle underlying knowledge analysis. It includes different section of the CommonKADS
knowledge-engineering methodology which should be developed in order to create a
comprehensive knowledge management system.
In order to design and develop any kind of knowledge-based system, the following questions
must be answered:
1. Why? Why is a knowledge system a potential help or solution? For which problems?
Which benefits, costs, and organizational impacts does it have? Understanding the
organizational context and environment is the most important issue here.
2. What? What is the nature and structure of the knowledge involved? What Is the
nature and structure of the corresponding communication? The conceptual description
of the knowledge applied in a task is the main issue here.
3. How? How must the knowledge be implemented in a computer system? How do the
software architecture and the computational mechanisms look? The technical aspects
of the computer realization are the main focus here.
Figure 2.2 – The CommonKADS Model Structure
All these questions are answered by developing (piece of) aspect models. CommonKADS has
a predefined set of models, each of them focusing on a limited aspect, but together providing
a comprehensive view:
- Organization model: The organization model analyzes the main characteristics of an
organization in order to find out the problems and opportunities for knowledge
systems. Moreover, it includes the feasibility study and the impacts of the intended
knowledge system on the organization.
- Task model: Tasks are the sub-processes of a business process. The task model
analyzes the task layout, its input and output, preconditions and performances criteria
and the needed resources and competences.
- Agent model: Agents are executors of a task. An agent can be human, an information
system, or any other entity which are able to perform a task. The agent model
describes the features of agents, in particular their competences, authority to act, and
constraints in this respect. Moreover, it lists the communication links between agents
in carrying out a task.
- Knowledge Model: The aim of the knowledge model is to determine the types and
structure of knowledge used in executing a task. Also it identifies all the roles of
knowledge-system components contributing in problem-solving. This makes the
knowledge model an important tool to communicate with experts and users about the
problem-solving features of a knowledge system, during both development and system
execution.
- Communication model: Since several agents may participate in a task, it is important
to have communication protocol to present all the transactions between the agents.
This is done by the communication model, in a conceptual and
implementation-independent way, just as with the knowledge model.
- Design model: the above CommonKADS models together can be seen as constituting
the requirements specification for the knowledge system, broken down in different
aspects. The design model presents the specification of the knowledge system in term
of architecture, implementation platform, software modules, representational
constructs, and computational mechanisms which are necessary to implement the
functions in knowledge and communication models.
Together, the organization, task, and agent models analyze the organizational environment
and the corresponding critical success factors for a knowledge system. The knowledge and
communication models create the conceptual description of problem-solving functions and
data that are to be handled and delivered by a knowledge system. The design model converts
this into a technical specification that is the basis for software system implementation. Thus
process is depicted in above figure 2.2. It should be noted, however, it is not necessary to
construct all the models. This depends on the goals of the project as well as the experiences
gained in running the project. Thus, a judicious choice is to be made by the project
management. Accordingly, a CommonKADS knowledge project produces three types of a
products or deliverables:
1. CommonKADS model documents
2. Project management information
3. Knowledge system software
It should be emphasized that knowledge systems and their engineering are not life forms
totally unrelated to other species of information systems and management. In what follows,
we will see that CommonKADS has been influenced by other methodologies, including
structured systems analysis and design, object orientation, organization theory, process
reengineering, and quality management.
2.1.2 Some Terminology
All the concepts used in CommonKADS methodology are defined as below:
Domain: Domain is some area of interest. Example domain is maintenance system area.
Domains can be hierarchally structured. For example, maintenance process can be split into
number of subdomains such as functional analysis, failure mode analysis, condition based
monitoring, etc.
Task: A task is a piece of work that needs to be done by an agent. In this research the
primarily interest is in “knowledge intensive” tasks: tasks in which knowledge plays a key
role. Example tasks are functional analysis, functional failure analysis, failure mode analysis,
condition based monitoring and prognosis.
Agent: An agent is any human or software system able to execute a task in a certain domain.
For example, a maintenance staff can carry out the task of diagnosing complaints uttered by
malfunction of the system. A knowledge system might be able in execute the task of
monitoring the spindle of a cutting machine.
Application: An application is the context provided by the combination of a domain and a
task carried out by one or more agents.
Application domain/task: these two terms are used to refer to the domain and/or task
involved in a certain application.
Knowledge-based system: The term “knowledge-based system” (KBS) has been used for a
long time and stems from the first generation architecture in which the two main components
are a reasoning engine and a knowledge base. In recent years the term has been replaced by
the more neutral term “knowledge system”. It is worthwhile pointing out that there is no fixed
borderline between knowledge systems and “normal” software systems. Every system
contains knowledge to some extent. This is increasingly true in modern software applications.
The main distinction is that in a knowledge system one assumes there is some explicit
representation of the knowledge included in the system. This raises the need for special
modelling techniques.
Expert System: one can define an expert system as a knowledge system that is able to
execute a task that, if carried out by humans, requires expertise. In practice the term is often
used as a synonym for knowledge(-based) system.
2.1.3 Organizational Model
The first part of the organization model focuses on problem and opportunities. The issues are
illustrated in figure 2.3. Opportunities, problems and knowledge-oriented solutions must
always be evaluated within a broader business perspective so it is important to get a real and
explicit understanding of this context. In order to do this, tables 2.1 (OM-1) and 2.2 (OM-2)
gives the worksheets which explains the various aspects to consider. Table 2.3 (OM.-3) helps
to performing the process breakdown and table 2.4 (OM-4) is used to analyze the knowledge
assets.
Figure 2.3 – Overview of the components of the CommonKADS organization model
Organization Model Problems and Opportunities Worksheet OM-1
Problems and opportunities Make a shortlist of perceived problems and opportunities, based on interviews, brainstorm and visioning meetings, discussions with managers, et cetera.
Organizational context Indicate in a concise manner key features of the wider organizational context, so as to put the listed opportunities and problems into a proper perspective. Important features to consider are:
1. Mission, vision, goals of the organization
2. Important external factors the organization has to deal with 3. Strategy of the organization
4. Its value chain and the major value drivers
Solutions List possible solutions for the perceived problems and opportunities, as suggested by the interviews and discussions held, and the above features of the organizational context.
Table 2.1 – Problems and Opportunities Worksheet OM-1
Organization Model
Problems
&
Opportunities
General
Context
(Mission,
Strategy,
Environment,
CSF's,...)
Potential
Solutions
OM-1
OM-2
Organization
Focus Area
Description:
Structure
Process
People
Culture & Power
Resources
Knowledge
OM-3
OM-4
Process
Breakdown
Knowledge
Assets
Organization Model Variant Aspects Worksheet OM-2
Structure Give a structure chart of the considered (part of) the organization in terms of its departments, groups, units, sections, ...
Process Sketch the layout (for example by a diagram) of the business process at hand. A process is the relevant part of the value chain that is focused upon. On its turn, a process is decomposed into tasks, which are detailed in Worksheet OM-3.
People Indicate which staff members are involved, as actors or stakeholders, including decision makers, providers, users or beneficiaries (`customers') of knowledge.
Resources Describe the resources that are utilized for the business process. These may cover different types, such as:
1. Information systems and other computing resources. 2. Equipment and materials.
3. Social, interpersonal, and other (non-knowledge) skills and competencies. 4. Technology, patents, rights.
Knowledge Knowledge represents a special resource exploited in a business process. Because of its key importance in the present context, it is set apart here. The description of this component of the organization model is given separately, in Worksheet OM-4 on knowledge assets.
Culture and Power Pay attention to the `unwritten rules of the game', including styles of working and communicating (`the way we do things around here') and informal influencing relationships and networks.
Table 2.2 – Variant Aspects Worksheet OM-2
Organization Model Process Breakdown Worksheet OM-3 No (identifier) Task (Task name) Performed by (agent) Where? (location) Knowledge asset Knowledge Intensive? Significance
Table 2.3 – Process Breakdown Worksheet OM-3
Organization Model Knowledge Assets Worksheet OM-4
Knowledge Asset (see OM-3) Possessed by Agent (see OM-3) Used in Task (see OM-3) Right Form? Yes/no Right Place? Yes/no Right Time? Yes/no Right Quality? Yes/no
2.1.4 Impact and Improvement Analysis: Task and Agent Modelling
Task model deals with the global task layout, its input and outputs, prerequisites, performance
criteria, needed resources and competences. Figure 2.4 shows the roles in maintenance
knowledge base system.
Figure 2.4 – Overview of different roles in task definition in designing a maintenance knowledge base system
Different roles in maintenance knowledge base system development are as below:
Knowledge Users: A knowledge user utilizes directly or indirectly of the knowledge system
which in the maintenance area could be the maintenance employees, production workers and
production scheduling staff.
Project Manager: The knowledge project manager is responsible for the project development
of the maintenance knowledge system.
Knowledge Manager: Knowledge manager determines different knowledge strategies at the
business level. The knowledge manager initiates knowledge development and knowledge
distribution activities.
In case that the feasibility study outcome is positive, it’s time to take the next step and to
focus on the features of the relevant tasks, the agents that carry them out, and on the
knowledge items used by the agents in executing tasks. For their description, CommonKADS
offers the task and agent models. Figure 2.5 illustrates the CommonKADS task model. The
outcomes of task model is detailed insight into the impact of a knowledge system, and
especially what improvement actions are possible or necessary in the organization in
conjunction with the introduction of a knowledge system.
Figure 2.5 – Overview of the CommonKADS task model
The notion of task has also emerged as a critical one in the theory and methodology of
knowledge systems and of knowledge sharing and reuse. Thus, a link is needed between the
notion of task in the human and organizational sense of the word, and the more information
systems oriented concept will be employed later on. The CommonKADS task model serves as
this linking pin between the organizational aspect and the knowledge-system aspect of a task.
A task is a subpart of a business process that:
- Represents a goal oriented activity adding value to the organization;
- Handles inputs and deliver desired outputs in a structured and controlled way;
- Consumes resources;
- Requires (and provides) knowledge and other competences;
- Is carried out according to given quality and performance criteria;
- Is performed by responsible and accountable agents.
Table 2.5 (TM-1) represents different parts of a task model which should be determined and
explained .
Task Model Task Analysis Worksheet TM-1
TASK Task identifier and Task name
ORGANIZATION Indicate the business process this task is a part of, and where in the organization (structure, people) it is carried out
GOAL AND VALUE Describe the goal of the task and the value that its execution adds to the process this task is a part of
DEPENDANCY AND FLOW
Input tasks: tasks delivering inputs to this task
Output tasks: tasks that use (some of) the outputs of this task
You can use a data flow diagram or an activity diagram here to describe this. OBJECTS HANDLED Input objects: The objects, including information and knowledge items, that are
input to task
Output objects: The objects, including information and knowledge items, that are delivered by the task as output
Internal objects: Important objects (if any), including information and knowledge items, that are used internally within the task but are not input or output to other tasks.
You may want to include a class diagram here to describe the information objects handled by the task.
TIMING AND CONTROL
Describe frequency and duration of the task.
Describe the control relation with other tasks. For this you may want to use a state diagram or an activity diagram.
Describe control constraints:
(I) Preconditions that must hold before the task can be executed. (II) Postconditions that must hold as result of execution of the task. AGENTS The staff members (cf. OM-2/3, People) and/or the information systems (cf.
OM-2/3, Resources) that are responsible for carrying out the task KNOWLEDGE AND
COMPETENCE
Competencies needed for successful task performance. For the knowledge items involved, there is a separate worksheet TM-2. List other relevant skills and competencies here. Indicate which elements of the task are knowledge-intensive.
Note that tasks can also deliver competencies to the organization, and it may be worthwhile to indicate that here.
RESOURCES Describe and preferably quantify the various resources consumed by the task (staff time, systems and equipment, materials, financial budgets)
The description is typically a refinement of the resource description in OM-2 QUALITY AND
PERFORMACE
List the quality and performance measures that are used by the organization to determine successful task execution
Table 2.5 – Task Analysis Worksheet TM-1
Next item of knowledge and competence is a significant items in the task model. For this
reason, it is again modelled by means of separate worksheet TM-2 as shown in table 2.6. it is
also useful to consider the information from the individual agents point of view. This is done
in commonKADS agent model, illustrated in table 2.7 as AM-1.
Task Model Knowledge Item Worksheet TM-2 NAME POSSESSED BY USED IN DOMAIN Knowledge Item Agent
Task Identifier and name
Wider domain the knowledge is embedded in (specialist field, discipline; branch of science or engineering, professional community)
Nature of knowledge Bottleneck / to be Improved?
formal, rigorous empirical, quantitative heuristic, rules of thumb highly specialized, domain specific experience-based action-based incomplete uncertain, may be incorrect quickly changing hard to verify tacit, hard to transfer
Form of knowledge Mind Paper Electronic Action skill Other Availability of knowledge Limitations in time Limitations in space Limitations in access Limitations in quality Limitations in form
Table 2.5 – Knowledge Item Worksheet TM-2
Agent Model Agent Worksheet AM-1
TASK Name of the agent
ORGANIZATION Indicate how the agent is positioned in the organization, as inherited from the OM-worksheet descriptions, including the type (human, information system), function, position in the organization structure, ...
INVOLVED IN List of Tasks (cf. TM-1)
COMMUNICATION WITH
List of agent names
KNOWLEDGE List of knowledge items possessed by the agent (cf. TM-2) OTHER COMPETENCES List of other required or present competences of the agent RESPONSIBILITIES
AMD CONSTRAINTS
List of responsibilities the agent has in task execution, and of restrictions in this respect. Constraints may refer to limitations in authority, but also to inside or outside legal or professional norms, or the like
2.1.5 Recommendations and Actions
Finally, with the worksheet TM-1, TM-2, and AM-1 we have controlled all information
related to the task and agent models (See Figure. 2.4 above). The remaining step is to
integrate this information into a document for managerial decision making about changes and
improvements in the organization.
Proposed actions for improvement are accompanying measures, but are not part of the
knowledge systems work itself. However, they are highly important for ensuring commitment
and support from the relevant players in the organization. The major issues for decision
making here are:
- Are organizational changes recommended and if so, which ones?
- What measures have to be implemented regarding specific tasks and workers
involved? In particular, what improvements are possible regarding use and availability
of knowledge?
- Have these changes sufficient support form the people involved? Are further
facilitating actions called for?
- What will be the further direction of the knowledge system project?
This completes the organization task-agent analysis. Even without building knowledge
systems, it is likely that this analysis brings to the surface many measures and improvements
that lead to better use of knowledge by the organization.
2.1.6 Knowledge Model
2.1.6.1 Role of the Knowledge Model
Detailed requirements engineering is split in CommonKADS into two parts. The knowledge
model specifies the knowledge and reasoning requirements of the prospective system, the
communication model specifies the needs and desires regarding to the interfaces with the
other agents. Figure 2.6 illustrates the role of the knowledge model in relation to other
models.
2.1.6.2 Knowledge Model Overview
A knowledge model encompasses three parts, each part captures a related group of knowledge
structure. Each part is called a knowledge category. The first category is called the domain
knowledge. This category specifies the domain specific knowledge and information types,
that is concerned in an application. The second part of the knowledge model contains the
inference knowledge. The inference knowledge describes the basic inference steps that are
needed to make using the domain knowledge. The third category of knowledge model is task
knowledge. Task knowledge describes what gaols and application follows and how these
goals can be determined through decomposition into subtasks and finally inferences.
Figure 2.6 – Schematic view of the role of the knowledge model in relation to other models
2.1.6.2.1 Domain Knowledge
The domain knowledge describes the main static information and knowledge objects in an
application domains and includes domain schema and knowledge base. The domain schema is
a schematic description of domain-specific knowledge and information through a number of
type definitions. Knowledge base contains instances of the types specified in the domain
schema.
In practice, the domain schema must specify the three main modelling constructs including
concept, relation and rule type and knowledge base which are defined as below:
Concept: a concept describes a set of objects or instances which occurring in application
domain as which share similar characteristics. An example of a concept could be a gear box in
a diagnosis domain in maintenance area. Features of concept can be described as an attribute.
An attribute can hold a value: a piece of information that instances of the concept can hold.
Relation: a relation between concepts is defined with the relation or binary relation construct.
Relations are defined through a specification of arguments. For each argument, the cardinality
can be defined. An example of a relationship could be causing relationship between a failure
mode and functional failure concepts in the failure mode analysis domain.
Rule Type: Rule types indicated the logical relationships between two logical statements. The
logical statements in such rules are typically expressions about and attribute value of a
concept. So these rules are a special type of relationship. The following logical sentences
could be an example of a rule type in the spindle diagnosis domain.
Shaft.failure = crack -> Spindle-Behaviour.status = Axial speed in Z direction no/less/more than desired speed
Knowledge Base: A domain schema describes domain knowledge types such as concept,
relation and rule type. A knowledge base contains instances of those knowledge types.
2.1.6.2.2 Inference Knowledge
The inference knowledge in knowledge model describes the lowest level of functional
decomposition. These basic information processing units are called inferences in knowledge
modelling. An inference performs a primitive reasoning step. Typically, inference uses
knowledge contain in some knowledge base to derive new information from its dynamic
input. An example of inference could be select a failure mode from the failure mode domain
knowledge to assess and identify the maintenance task.
An inference is described in term of functional roles. There are two types of knowledge roles
called dynamic roles and static roles. Dynamic roles are the run-time inputs and outputs of
inferences. Each invocation of inference usually has different instantiations of the dynamic
roles. As an example a Cover inference that uses a failure casual model to find a root cause
that could explain a malfunction of the spindle. Such an inference would have two dynamic
knowledge roles. An input role complaint, indicating a domain object representing a
functional failure about the behaviour of the system, and an output role hypothesis,
representing a single failure mode as a candidate solution. Static roles, on the other hand, are
more or less stable over time. Static roles specify the collection of domain knowledge that is
used to make the inference. For example, the above-mentioned inference Cover could use the
state dependency network which shows the failure dependencies of different components of a
spindle in prognosis task.
Transfer function is a function that transfers information items between the reasoning agents
describe in the knowledge model and the outside world (another system or user). And
example of transfer functions could be Obtain which is used whenever the reasoning agent
request a piece of information form an external agent. The reasoning agent has the initiative
and the external agents hold the information items.
2.1.6.2.3 Task Knowledge
Reasoning always has a “reason”. In the other word, an important aspect of knowledge is
what one wants to do with it. What are the goals intended to achieve by applying knowledge?
Task knowledge is the knowledge category that describes these goals and the strategies that
will be employed for realizing goals. Task knowledge is typically described in a hierarchical
fashion. As an example, top level tasks could be status monitoring of a subsystem and at the
lowest level of task decomposition, the task is linked to inferences and transfer functions such
as Cover, Predict, Compare and Obtain.
A task method describes how a task is realized through the decomposition into sub-functions.
The core part of the method is formed by the control structure. This control structure
describes in what order the sub-functions should be implemented. The control structure
typically reads like a small program in which the sub-functions are the procedures and the
roles act as parameters of the procedures.
The architecture of the knowledge model components is illustrated in figure 2.7.
Figure 2.7 – Architecture of the knowledge model subsystems. The dotted lines indicate method invocation paths, the solid lines are information access path.
dynamic role datatype domain-mapping current binding access/update functions task I/O roles method execute transfer function I/O roles task method intermediate roles control specification execute static role domain-mapping access functions domain model domain-model name uses access functions inferencing functions inference I/O roles static roles method give-solution more-solutions? has-solution? inference method algorithm spec local vars execute domain construct
2.2 Reliability-Centered Maintenance (RCM)
2.2.1 Background
Over the past twenty years, maintenance has changed, perhaps more so than any other
management discipline. The changes are due to a huge increase in the number and variety of
physical assets (plant, equipment and buildings) which must be maintained throughout the
world, much more complex designs, new maintenance techniques and changing views on
maintenance organization and responsibilities. Maintenance is also responding to changing
expectations. These include a rapidly growing awareness of the extent to which equipment
failure affects safety and the environment, a growing awareness of the connection between
maintenance and product quality, and increasing pressure to achieve high plant availability
and to contain costs.
The key challenges facing modern maintenance strategies can be stated as below:
•
to select the most appropriate techniques to deal with each type of failure process in
order to fulfil all the expectations of the owners of the assets, the users of the assets
and of society as a whole
•
in the most cost-effective and enduring fashion
•
with the active support and co-operation of all the people involved.
RCM provides a framework which enables users to respond to these challenges, quickly and
simply. It does so because it never loses sight of the fact that maintenance is about physical
assets. If these assets did not exist, the maintenance function itself would not exist. So RCM
starts with a comprehensive review of the maintenance requirements of each asset in its
operating context.
Reliability-Centered Maintenance (RCM) can be defined as a process used to determine the
maintenance requirements of any physical asset in its operating context. RCM process asking
seven questions about the assets under review. These questions are as below [17]:
•
what are the functions and associated performance standards of the asset in its present
operating context?
•
in what ways does it fail to fulfill its functions?
•what causes each functional failure?
•
what happens when each failure occurs?
•in what way does each failure matter?
•
what can be done to predict or prevent each failure?
2.2.2 Functions
The first step in RCM is to determine and define the functions of each asset in its operating
context and in term of its performance. These functioned can be categorize in two ways:
•
primary functions, The expected operational task of an asset in fist place. This
category covers functions like speed, motion and product quality. e capacity, product
quality and customer service.
•
secondary functions, The operational tasks more than fulfilling its primary functions.
This category covers areas like safety, control, comfort, protection and efficiency of
operation.
2.2.3 Functional Failure
The objectives of maintenance are defined by the required functions and standard
performance of the assets under consideration. The reason which can stop an asset to perform
its standard function is a failure. So RCM deals with the identification of the failures in an
asset. This process can be done in two level:
•
firstly, by identifying what circumstances amount to a failed state then
•secondly, by asking what events can cause the asset to get into a failed state.
In RCM, the failure states are known as functional failures.
2.2.4 Failure Modes
The next step after identifying the functional failures is to identify all events which are likely
to cause each of these failure states. These events are called failure modes. "Reasonably
likely" failure modes include those which have occurred on the same or similar equipment
operating in the same context, failures which are currently being prevented by existing
maintenance regimes, and failures which have not happened yet but which are considered to
be real possibilities in the context in question.
2.2.5 Symptoms and Consequences
The next phase in RCM is to identify the consequences happens when a functional failure
occurs. It is these consequences which affect the system and the assets which it is tried to
prevent. The RCM process, categorize these consequences into four classes:
•