Condition Assessment Models For Sewer Pipelines

QATAR UNIVERSITY
COLLEGE OF ENGIEERING
CONDITION ASSESSMENT MODELS FOR SEWER PIPELINES
BY
FIRAS AMER ABDULRAZAK ALKADOUR
A Thesis Submitted to the Faculty of
the College of Engineering
In Partial Fulfillment
of the Requirements
for the Degree of
Master of Science in
Civil Engineering
June 2017
© 2017 Firas Amer Abdulrazak Alkadour. All Rights Reserved
‫‪COMMITTEE PAGE‬‬
‫‪The members of the Committee approve the thesis of Firas Alkadour defended on 18-‬‬
‫‪05-2017‬‬
‫ــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ‬
‫‪Dr. Alaa Al-Hawari‬‬
‫‪Thesis/Dissertation Supervisor‬‬
‫‪Dr. Murat Gunduz‬‬
‫‪Chair‬‬
‫‪Dr. Mohsin Siddiqui‬‬
‫‪Committee Member‬‬
‫‪Dr. Hesham Osman‬‬
‫‪Committee Member‬‬
‫‪Approved:‬‬
‫ــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــــ‬
‫‪Khalifa Al-Khalifa, Dean, College of Engineering‬‬
‫‪ii‬‬
Abstract
ABDULRAZAK ALKADOUR, FIRAS, AMER., Masters : June : 2017,
Masters of Science in Civil Engineering
Title: Condition Assessment Models for Sewer Pipelines
Supervisor of Thesis: Dr. Alaa Al-Hawari.
Underground pipeline system is a complex infrastructure system that has
significant impact on social, environmental and economic aspects. Sewer pipeline
networks are considered to be an extremely expensive asset. This study aims to
develop condition assessment models for sewer pipeline networks. Seventeen factors
affecting the condition of sewer network were considered for gravity pipelines in
addition to the operating pressure for pressurized pipelines. Two different
methodologies were adopted for models’ development. The first method by using an
integrated Fuzzy Analytic Network Process (FANP) and Monte-Carlo simulation and
the second method by using FANP, fuzzy set theory (FST) and Evidential Reasoning
(ER). The models’ output is the assessed pipeline condition. In order to collect the
necessary data for developing the models, questionnaires were distributed among
experts in sewer pipelines in the state of Qatar. In addition, actual data for an existing
sewage network in the state of Qatar was used to validate the models’ outputs. The
“Ground Disturbance” factor was found to be the most influential factor followed by
the “Location” factor with a weight of 10.6% and 9.3% for pipelines under gravity
and 8.8% and 8.6% for pipelines under pressure, respectively. On the other hand, the
least affecting factor was the “Length” followed by “Diameter” with weights of 2.2%
and 2.5% for pipelines under gravity and 2.5% and 2.6% for pipelines under pressure.
iii
The developed models were able to satisfactorily assess the conditions of
deteriorating sewer pipelines with an average validity of approximately 85% for the
first approach and 86% for the second approach. The developed models are expected
to be a useful tool for decision makers to properly plan for their inspections and
provide effective rehabilitation of sewer networks.
iv
TABLE OF CONTENTS
List of Figures............................................................................................................................ vii
List of Tables ............................................................................................................................ viii
Acknowledgments ..................................................................................................................... ix
CHAPTER 1 - INTRODUCTION ..................................................................................................... 1
1.1 Overview................................................................................................................. 1
1.2 Inspection and Evaluation of Existing Sewers ........................................................ 3
1.3 Qatar Network Existing Assets ............................................................................... 5
1.4 Thesis Organization ..............................................................................................10
CHAPTER 2 – LITERATURE REVIEW ...........................................................................................11
2.1 Factors Affecting Sewer Pipelines Condition........................................................11
2.2 Deterioration Models ...........................................................................................13
2.2.1 Physical Models ............................................................................................... 15
2.2.2 Artificial Intelligence Models ........................................................................... 16
2.2.3 Statistical Models ............................................................................................ 27
2.2.4 Summary.......................................................................................................... 46
2.3 Problem Statement ..............................................................................................48
2.4 Research Objectives .............................................................................................49
CHAPTER 3 – RESEARCH METHODOLOGY ................................................................................50
CHAPTER 4 – DATA COLLECTION ..............................................................................................53
4.1 Overview...............................................................................................................53
4.2 Factors affecting sewer pipelines condition .........................................................54
4.3 Questionnaires .....................................................................................................56
4.3.1 Factor’s Weights (Wi) ...................................................................................... 56
4.3.2 Factor’s Effect Values ...................................................................................... 57
v
4.3.3 Responses ........................................................................................................ 60
4.4 Validation Set .......................................................................................................60
CHAPTER 5 : MODELS DEVELOPMENT AND IMPLEMENTATION ..............................................61
5.1 Factors’ Weight (Wi) Determination ....................................................................61
5.1.1 Fuzzy Analytic Network Process (FANP) .......................................................... 61
5.1.2 Monte-Carlo Simulation .................................................................................. 67
5.2 Effect Value...........................................................................................................69
5.2.1 First Approach (Condition Curves) .................................................................. 69
5.2.2 Second Approach (Fuzzy Set Theory) .............................................................. 73
5.3 Overall Condition Assessment Index ....................................................................79
5.3.1 First Approach (Monte-Carlo Simulation) ....................................................... 79
5.3.2 Second Approach (Evidential Reasoning) ........................................................ 81
5.4 Model Validation ..................................................................................................84
CHAPTER 6 : CONCLUSION .......................................................................................................92
References ................................................................................................................................95
Appendix: Questionnaire........................................................................................................106
vi
LIST OF FIGURES
Figure 1-1: Sewer Management System .................................................................................... 2

Figure 1-2: Qatar Sewer Network - Diameter Statistics ............................................................. 6
Figure 1-3: Qatar Sewer Network - Age Statistics ...................................................................... 7
Figure 1-4: Qatar Sewer Network - Material Statistics............................................................... 8
Figure 1-5: Pipe position relative to Groundwater (GW) Statistics ............................................ 9
Figure 2-1: Sewer deterioration models’ classification ............................................................14
Figure 2-2: Schematic presentation of the back propagation neural network BPNN (Tran
et al., 2007) ...........................................................................................................18
Figure 2-3: Schematic presentation of the probabilistic neural network PNN (Ana, 2009).....19
Figure 2-4: Multiple discriminant analysis visualization (Tran, 2007) ......................................34
Figure 2-5: Norwegian network’s transition curves (Horold, 1998).........................................39
Figure 3-1: Research Methodology ..........................................................................................52
Figure 4-1: Types of Data Collected..........................................................................................53
Figure 4-2: Classification of Factors Affecting Sewage Pipelines Condition.............................54
Figure 5-1: Steps for Conducting FANP ....................................................................................64
Figure 5-2: Factors’ final weights calculated using FANP and Monte-Carlo simulation for
Gravity and Pressurized Pipelines .........................................................................69
Figure 5-3: Pipeline Condition Curves for Different Sub-Factors over Age ..............................73
Figure 5-4: Membership Function for Pipelines’ Age Effect Value ..........................................75
Figure 5-5: Membership Functions for Pipelines’ Diameter Effect Values ..............................76
Figure 5-6: Membership Functions for Pipelines’ Length Effect Values...................................77
Figure 5-7: Membership Functions for Pipelines’ Buried Depth Effect Value..........................78
Figure 5-8: Sample of Monte-Carlo Simulation Output for Overall Condition Index
Probability Distribution .........................................................................................81
Figure 5-9: Actual Versus Predicted Overall Condition Index (1) .............................................90
Figure 5-10: Actual versus Predicted Overall Condition Index (2) ...........................................91
vii
LIST OF TABLES
Table 2-1: Variables Included in Sewer Pipelines Condition Assessment Models ...................12
Table 2-2: Assessed Levels and Expected Outcomes for Each Condition Assessment
Model ....................................................................................................................47
Table 4-1: Factors Affecting Sewer Pipeline Condition. ...........................................................55
Table 4-2: Questionnaire Sample for relative importance used in Pairwise Comparison
for Main factors and Sub-Factors in gravity and pressurized pipelines ................58
Table 4-3: Questionnaire Sample for effect values of different factors for gravity and
pressurized pipelines ............................................................................................59
Table 5-1: General Arrangement of Analytical Network Process (ANP) Super-matrix ............62
Table 5-2: Sample of Pairwise Comparison Matrices in Gravity Pipelines ...............................65
Table 5-3: Un-weighted Super-matrix, Weighted Super-matrix and Limit Super-matrix
for different affecting factors ...............................................................................66
Table 5-4: Summary of Statistical Analysis Results for Factor Weights ...................................68
Table 5-5: Summary of Statistical Analysis Results for Factor Effect Values in Gravity
Pipelines ................................................................................................................70
Table 5-6: Summary of Statistical Analysis Results for Factor Effect Values in Pressurized
Pipelines ................................................................................................................71
Table 5-7: Conversion of Actual Condition Rating Scale to Model Prediction Condition
Rating Scale ...........................................................................................................85
Table 5-8: Sample of Actual versus Predicted Model Condition Rating Values for
Pipelines in Validation Dataset .............................................................................87
viii
ACKNOWLEDGMENTS
The Thesis has been prepared under the direct supervision of Dr. Alaa AlHawari,
Professor of Civil Engineering, to whom the writer is indebted for his guidance,
constructive criticism, and encouragement in the preparation of the manuscript. This
research was made possible by NPRP grant # (NPRP6-357-2-150) from the Qatar
National Research Fund (Member of Qatar Foundation).
Also, the writer would like to thank Dr. Tarek Zayed, Professor of Civil Engineering
at Concordia University for his support in the analysis part, the Public Works
Authority of Qatar (ASHGAL) for their support in the data collection.
The writer gratefully appreciates the continued support and patience of his parents and
family throughout his education.
ix
CHAPTER 1- INTRODUCTION
1.1 Overview
The condition and level of service of sewage pipelines could have major effect on
environmental and economic aspects for populated urban areas. Deteriorated sewage
pipelines are considered hazardous on the public healthiness and environment. The
performance of sewage pipelines is a function in reliability, and level of service by
which decision makers can determine the lifecycle of the pipeline and the time for
interventions required to reinstate the level of service of deteriorated pipes back to the
desired level. Maintenance, rehabilitation and renewal comprises the intervention of
assets, which can be associated to knowing the current condition of the pipelines. In
absence of information regarding the condition of pipelines, unforeseen failure can
take place making asset replacement inevitable which is the most expensive measure
amongst the rest of intervention measures.
In the past, a reactive approach was considered in sewer management as
rehabilitation is made upon pipe failure. However, the trend has changed with time
towards a proactive approach due to the great cost of repairs at emergencies and its
impact on social and environmental aspects. In the current trend, the problems are
addressed before their occurrence by implementing a proper asset management
system (Vanier, 2001).
Condition assessment models are considered tools that can provide users and
decision makers enough information to determine whether an intervention is required
for certain pipes based on their state of deterioration. As a result, municipalities need
to develop condition assessment models to determine the condition of the assets from
1
which decisions regarding prioritization of inspection, repair and renewal of sewer
pipes can be made. These models are built by incorporating data available in
databases and records in municipalities. These data are generally the deterioration
factors that impact the degradation of sewer pipes and the condition of these pipes.
Asset management system for sewer networks is further described in Figure 1-1.
Data Collection:
Deterioration Factors Full knowledge on
(Age, Diameter, etc.) system condition
& Condition Data - Avoid premature
Prediction of sewer failure.
(CCTV Inspections) furture condition
- Accurately plan for
future expedenture
requirements.
- Optimize
Tools maintenance and
Partial Development rehabilitation
Knowledge on the (Sewer starigies.
system Condition Deterioration
Models)
Figure 1-1: Sewer Management System
A reliable inspection plan and condition assessment models for the maintenance
and rehabilitation of sewer pipelines is needed to control and minimize adverse
potential effects of assets failure as reported by the National Guide to Sustainable
Municipal Infrastructure best practice (2003).
Sewer pipelines can be divided into gravity pipelines and pressurized pipelines
where each is expected to deteriorate in a different manner. Therefore, two different
condition assessment models for each approach were developed to assess the two
different pipeline systems. In order to collect the necessary data for the model
2
development, questionnaires were distributed among experts in sewer pipelines in the
state of Qatar. Fuzzy Analytic Network Process (FANP) will be used to determine the
weight of each of the identified factors that would affect the pipeline condition. The
calculated weights (FANP) and effect values will be fed into an Oracle® Crystal Ball
software to get a probabilistic condition index for sewer pipelines using Monte-Carlo
simulation for the first approach. The calculated weights using (FANP) and generated
membership functions for the effect values using Fuzzy Set Theory (FST) will be
integrated with Evidential Reasoning (ER) technique to generate the final condition
index for the second approach. The main goals of the current study are: (1) to
recognize the primary factors that would affect sewer pipelines' conditions and (2) to
develop a condition assessment model for sewer pipeline networks.
1.2 Inspection and Evaluation of Existing Sewers
The reasons for carrying out sewer inspections can be divided into three main
reasons (Butler et al., 2000):
- Periodic inspection to determine the condition of existing sewer pipelines.
- Crisis inspection to determine the causes behind the failure of sewer pipelines
and carry out an emergency repair.
- Inspection of new sewers to check the new sewer pipelines are constructed as
per the required workmanship and construction standards.
Several inspection techniques are utilized to assess the condition of sewer
pipelines. These techniques can be categorized into three groups.
- Group I: Techniques used to assess the internal condition of sewer pipelines
such as Sewer Scanner and Evaluation Technology (SSET), Closed Circuit
3
Television (CCTV), Zoom Camera, Laser Scanning and Ultra sound. CCTV is
the most technique used widely.
- Group II: Techniques used to assess the overall condition of sewer pipelines
and the surrounding soil such as Micro Deflections, Natural Vibration and
Impact Echo.
- Group III: Techniques used to detect a specific defect within the pipe segment
walls such as Ground Penetrating Radar (GPR).
The structural condition of pipes are generally determined through a defined
condition assessment rating system. There are several available rating systems such as
NEN3399 (1992) and WRc (2001). This assessment usually depends on the results
obtained from pipe inspections performed using the technologies listed above.
However, those technologies are expensive and time consuming with many
drawbacks (Wirahadikusumah, 1998).
Typically, a pipe section is split into 1m length segments where the observed
defects resulting from pipe inspections are identified and a condition rating is
provided accordingly. The final condition of the pipe is either the rating of the worst
segment or the mean rating of all segments.
There are many errors and uncertainties with CCTV inspections’ outcome. For
example, a certain pipe might appear to be improving in condition with age. Also,
there is a possibility that future carried out inspections might not show the defects
observed in the current inspections. The uncertainties in CCTV inspections can be
connected to two main sources (Chae et al. 2003, Müller and Fischer 2007).
• Human error where the CCTV inspection results depend on the concentration
and experience of the operator.
4
• CCTV camera limitations where the quality of the camera and light condition
can lead to inaccurate identifications of the defects.
1.3 Qatar Network Existing Assets
Qatar drainage system is separated in which foul sewage and storm water runoff
are collected in separate systems. Sewage generally flows by gravity through house
connections to manholes and sewer pipelines. The sewage flows by gravity to
pumping stations where it is pumped to Sewage Treatment Works (STW).
Qatar has a flat topography which does not support long distances of gravity sewer as
great depths of excavation would be required. Therefore, the sewerage system
consists of many pumping station. This leads that for sewage to arrive at the STW, it
needs to be pumped several times.
For gravity sewers, the favored material for usage in is vitrified clay (VC), for
pipes up to 1000mm diameter where Glass Reinforced Plastic (GRP) is favored for
diameters in excess of 1000mm. High-density polyethylene (HDPE) is not preferred,
but may be used as a sliplining where trenchless methods are necessary for
installation, using concrete jacking pipes. For pressurized pipes, the considered Pipe
materials in pumping stations are always Ductile Iron (DI). However, for rising mains
outside pumping stations, the piper materials can be either ductile iron (DI) or Glass
Reinforced Plastic (GRP) with concrete protection.
Data was collected from the Operation and Maintenance Department in Ashghal
Public Work Authority for 2073.352 km of gravity sewer pipelines. Figure 1-2 to
Figure 1-5 show the different characteristics (Diameter, Age, Material and Position
Relative to Groundwater Table) of sewer pipes for the obtained data set.
5
Diamerter Statistics
1200000
1000000
800000
LENGTH (M)
600000
400000
200000
DIAMETER (MM)
Figure 1-2: Qatar Sewer Network - Diameter Statistics
6
Age Statistics
800000
700000
600000
500000
LENGTH (M)
400000
300000
200000
100000
YEAR
Figure 1-3: Qatar Sewer Network - Age Statistics
7
Material Statistics
1600000
1400000
1200000
1000000
LENGTH (M)
800000
600000
400000
200000
0
AC DI GRP HDPE PC RC UPVC VC Unknown
MATERIAL
Figure 1-4: Qatar Sewer Network - Material Statistics
AC: Asbestos Cement, DI: Ductile Iron, GRP: Glass Fiber Reinforced Plastic, HDPE: High-Density Polyethylene Pipe, PC: Polymerized Vinyl Chloride, RC: Reinforced Concrete,
UPVC: Unplasticized Polymerized Vinyl Chloride, VC: Vitrified Clay.
8
Pipe position relative to Groundwater (GW) Statistics
2500000
2000000
1500000
LENGTH (M)
1000000
500000
0
Above GW Below GW Unknown
POSITION
Figure 1-5: Pipe position relative to Groundwater (GW) Statistics
9
1.4 Thesis Organization
Chapter 1 – Introduction: This chapter includes an overview about sewer network,
asset management, inspection techniques and condition assessment models. In
addition Qatar sewer network existing assets are identified. Moreover, each Chapter’s
content is described briefly.
Chapter 2 – Literature Review: This chapter identifies all major factors and sub-
factors affecting sewer pipelines’ condition and all previously developed condition
assessment models. In addition, the limitation of previous researches, the research
problem and research objectives are stated.
Chapter 3 – Research Methodology: This chapter describes the research
methodology considered in this research.
Chapter 4 – Data Collection: This chapter covers the data collection stage.
Questionnaires - one related to gravity sewer pipelines and another related to
pressurized sewer pipelines-were constructed and sent to sewer pipelines consultants,
consultants, and engineers. The experts were requested to perform a pair-wise
comparison among the identified main factors and sub-factors and to determine the
effect of each factor on the pipeline condition. In addition, actual data set is collected
for validation purposes.
Chapter 5 – Model Development and Implementation: This chapter discusses
constructing the condition assessment models using an integrated Fuzzy Analytic
Network Process (FANP) and Monte-Carlo simulation as one approach and FANP,
fuzzy set theory (FST) and Evidential Reasoning (ER) as another approach.
Chapter 6 – Conclusion: This chapter wraps up the thesis with conclusions.
10
CHAPTER 2 – LITERATURE REVIEW
2.1 Factors Affecting Sewer Pipelines Condition
In general, pipe deteriorates with age; however pipes with different characteristics
can experience significant variations in the deterioration process based on many
factors. Hawari et al., (2016) studied and identified major factors affecting sewer
pipelines’ condition (Fenner, 2000, Fenner et al., 2000, Davies et al., 2001,
Ariaratnam et al., 2001, Müller, 2002, Baur and Herz, 2002, Micevski et al., 2002,
Hahn, et al., 2002, Baik et al., 2006, Tran et al., 2007, Dirksen and Clemens, 2008,
Ana, et al., 2009).
These factors were subdivided into three main groups: (1) physical factors, (2)
operational factors and (3) environmental factors. The physical factors included sewer
pipeline characteristics such as: age, material type, size, buried depth, coating
conditions and installation quality. The operational factors included: flow rate,
infiltration and inflow, blockages, corrosive impurities and maintenance strategies in
addition to the operating pressure for pipelines under pressure. Finally, the
environmental factors included: bedding conditions, location, groundwater level and
ground disturbance.
Table 4-1 shows the different variables (e.g.: factors) selected in previous
researches. Age, length, material and diameter are the most common factors that were
included in these models.
11
Table 2-1: Variables Included in Sewer Pipelines Condition Assessment Models
Variables Included
Diameter
Category
Material
Bedding
Length
Waste
Sewer
Depth
Street
Other
GWT
Author(s) Model
Slope
Type
Type
Age
Najafi and
Artificial Neural Networks       
Kulandaivel (2005)
Ruwanpura et al.
Rule-Based Simulation   
(2004)
Other Physical, Operational, and
Rule-Based Simulation         
Hawari et al., (2016) Environmental Factors
Hahn et al., (2004) Expert Systems Corrosion, Erosion, Defects, Reconstruction and Socio-economic
Elmasry et al., (2016) Inference Systems Structural and Operational Defects
Ariaratnam et al. Regression (Logistic )     
(2001)
Chughtai and Zayed Regression (Multiple
      
(2008) Linear)
Regression (Binary and
Salman (2012)        
Logistic )
Multiple Discriminant
Ana (2009)        Traffic Intensity, Installation Year
Analysis
Bai et al., (2008) Evidential Reasoning Cement lining condition and the Degree of internal corrosion
Fuzzy Based Evidential
Daher (2015) Structural, Operational, Installation Defects
Reasoning
Baur and Herz (2002) Survival Functions      Shape of profile
Wirahadikusumah et Markov Chains –
   
al. (2001) Nonlinear optimization
Sinha and McKim Markov Chains –
Not Specified
(2007) Nonlinear optimization
Kleiner (2001) Semi-Markov Chains  Expert Opinion
Fuzzy Rule-Based Markov
Kleiner et al. (2004) 
Chains
MarkovChains –
Micevski et al. (2002) Metropolis-Hastings    Exposure Classification
Algorithm
Markov Chains – Ordered
Baik et al. (2006)     
Probit
Le Gat (2008) Markov Chains – Gompit           Installation Period
12
2.2 Deterioration Models
Prediction of sewer condition is considered a very complex process as it is
effected by many factors. A combination of probability based equations and empirical
data can be considered for the prediction of sewer pipelines deterioration (Mehle et
al., 2001). The basic idea behind deterioration models is to find the relationship
between the factors influencing the deterioration process of sewer pipelines and the
pipeline condition. Thus, the availability of data containing set of deterioration factors
and the sewer pipelines actual observed conditions is considered vital in sewer
deterioration modeling. Using the developed models, the future condition of sewer
pipelines with respect to age could be estimated. However, each model relies on
different concepts and differs in its data requirements and calibration methods
(Scheidegger et al., 2011).
Condition assessment models can be classified as physical, statistical or artificial
intelligence models (Yang, 2004). Ana and Bauwens (2010) focused on 5 statistical
models that have been developed by previous researches to model the structural
deterioration of sewer pipelines only in their review which were logistic regression
model, multiple discriminant analysis model, cohort survival model, Markov chain
model and Semi-Markov chain model. Figure 2-1 shows a classification of the
different deterioration models considered in assessing the condition of sewer
pipelines.
Physical models are used to define clear quantitative relationship between
deterioration factors and condition of sewer pipelines without accounting for the
uncertainty of the deterioration process. In contrast, statistical models consider the
13
uncertainties by using probability based equations. On the other hand, artificial
intelligence models are considered to be data-driven and not model-driven where the
mathematical relationships between the deterioration factors and condition data are
evaluated by “learning” the deterioration behavior from inspection data.
Artificial Intelligence
Physical Models Statistical Models
Models
Artificial Neural Multiple Linear
Network Regression
Rule Based Fuzzy

Logistic Regression
Logic
Rule Based Simulation
Analysis
Expert Systems Evidential Reasoning
Cohort Survival
Markov Chains
Semi Markov Chains
Figure 2-1: Sewer deterioration models’ classification
Ana and Bauwens (2010) further classified the deterioration models into two types
which are pipe group and pipe level models. Pipe group models consider entire
network or cohorts that share similar characteristics such as material, length, size and
age and thus experience similar deterioration behavior. Pipe group models can be
used by municipalities to set long term strategies and define budget requirements. On
14
the other hand, pipe level models predict the condition of each single pipe, where the
pipe characteristics are considered as covariates. Pipe level models are useful in
prioritizing inspection and rehabilitation plans and can be converted into group level
models by creating groups or cohorts of sewers. All the deterioration models defined
in Figure 2-1 are considered as pipe level models that can also be used as pipe group
models except for Cohort survival models which fall under pipe group models only.
2.2.1 Physical Models
Model Description:
Physical models are models that consider the physical mechanisms of the
deterioration process of sewer pipelines. These physical models are considered
deterministic models because they are based on the physical properties and the
mechanics of a certain phenomenon. They involve fitting linear or non-linear
equations to observations related to the asset failure (Marlow et al., 2009). The pipe
structural properties, internal and external loads such as traffic loading and material
degradation are the aspects that govern sewer pipes deterioration (Rajani and Kleiner
2001).
Application:
ExtCorr is a physical model that was developed within the Care-S project (Konig,
2005). The developed model could estimate the external corrosion of concrete pipes
taking into consideration soil aggressiveness, pipe cement quality and soil moisture.
WATS model is another similar application which was developed to simulate the
effect of internal corrosion in sewer pipes. The model was based on developing non-
linear differential equations describing microbial and chemical transformation
15
processes of organic matter, and chemical compounds resulting from chemical
reactions in wastewater (Vollersten and Konig, 2005).
Critique:
Deterioration of sewer pipelines is considered as a complex process that depends
on large number of factors (Schmidt, 2009). Some of the aspects that contribute to the
condition of sewer pipes can be modeled empirically such as corrosion, but the
deterioration in sewer condition itself is very difficult to be modeled in the same
manner. Another limitation is the scarcity of data needed to simulate the deterioration
mechanisms (Ana, 2009).To overcome such problem, some assumptions are made
without taking into consideration the uncertainties associated with asset deterioration
and failure (Marlow et al., 2009). As a result, physical models developed to assess the
condition of sewer pipelines are considered too simple to reflect the actual
deterioration process (Tran, 2007), hence they aren’t practical enough to truly reflect
the behavior of sewer pipelines.
2.2.2 Artificial Intelligence Models
Artificial intelligence (AI) aims to develop algorithms that mimic the behavior of
humans when dealing with problems and in patterns recognition (Sage, 1990).
Artificial Neural Networks (ANN) and rule based models such as fuzzy logic and
simulation can be considered as AI techniques.
2.2.2.1 Artificial Neural Networks
Model Description:
The Artificial neural networks (ANN) are a simulation of the human nervous
system. They are comprised of artificial neurons which are connected to each other in
16
different layers aiming to mimic human’s brain ability to recognize patterns and to
recall them in order to predict certain outcomes based on observations (Al-Barqawi
and Zayed 2008). The ability of ANN to learn by patterns recognition makes it a very
effective tool for model development. ANN could provide predictions based on
available historical data when relationships between inputs and outputs aren’t clear or
distinct enough or when data is incomplete (Sadiq et al. 2004).
In ANN, neurons are linked to each other with connections having a certain
weight. When the summations of weights for the inputs reach a certain value, the
neurons send a signal that identifies the activation function (Fausset 1994, Zou et al.
2008). In order to determine the weights between connected neurons, the error
between the estimated outcome of the model and the actual outcome is minimized
(Achim et al. 2007, Salman 2010). In sewer prediction models, the network is trained
from a data set containing sewer deterioration factors which represent the input layer
and pipe conditions which represent the output layer. Probabilistic neural network
(PNN) and back-propagation neural networks (BPNN) are the two main neural
networks that have been used in condition assessment modeling. The principles of
these PNN and BPNN are presented below:
 Back-propagation neural networks (BPNN):
In BPNN, the model is divided into three layers as shown in Figure 2-2. The
model components are as follow:
• Input layer: Contains a set of nodes representing the deterioration factors
having a value of Xi.
17
• Hidden layer: Each node receives signals from the input layer. The value of
each node is the result of the product between the inputs Xi and the associated
weights. At each node the sum of the weighted inputs is calculated and an
output signal is produced using activation or transfer functions.
• Output layer: The outputs of the hidden layer are received and multiplied by
connection weights to define the predicted condition classes.
During the training process, optimization algorithms are used to calibrate the
connection weights where the error between the models predicted output and the
actual conditions is minimized.
Figure 2-2: Schematic presentation of the back propagation neural network BPNN
(Tran et al., 2007)
18
• Probabilistic neural network (PNN):
Probabilistic neural network is defined as a special form of neural networks. It
classifies input vectors into classes based on Bayesian classification (Specht, 1990).
The model is divided into four layers as shown in Figure 2-3.
Figure 2-3: Schematic presentation of the probabilistic neural network PNN (Ana,
2009)
The model components are as follow:
• Input layer: Contains a set of nodes where each node represents a deterioration
factor having a value Xi.
• Pattern layer: Contains one node for each sample in a training set. The value
of each node is the result of the dot product between the input vector X and a
weight vector. The nodes are grouped in accordance with their associated class
i = 1,2,…..,m.
19
• Summation layer: Contains a number of nodes where each node represents a
condition class. For a given class (i), the outputs of the pattern nodes are
received and its value is calculated using an estimation of the probability
density function (PDF). Equation 1 shows the mathematical formula of the
PDF.
𝑀𝑀𝑖𝑖 (𝑖𝑖) (𝑖𝑖)
1 1 (𝑥𝑥 − 𝑥𝑥𝑗𝑗 )𝑇𝑇 (𝑥𝑥 − 𝑥𝑥𝑗𝑗 )
𝑓𝑓𝑖𝑖 (𝑥𝑥) = � 𝑒𝑒𝑒𝑒𝑒𝑒 �− � (1)
(2𝜋𝜋)𝑛𝑛/2 𝜎𝜎 𝑛𝑛 𝑀𝑀𝑖𝑖 2𝜎𝜎 2
𝐽𝐽=1
Where,
(𝑖𝑖)
𝑥𝑥𝑗𝑗 : The input vector of the Jth sample in a training set from class i,
n: The dimension of the input vector,
𝑀𝑀𝑖𝑖 : Number of training samples belonging to class i,
𝜎𝜎: Smoothing parameter which corresponds to the standard deviation of the
Gaussian distribution,
𝑇𝑇: Transpose function.
• Output layer: The outputs of the summation layer are received and the
condition is assigned by implementing the Bayes decision rule which is shown
in Equation 2 for a 2 condition classes.
ℎ1 𝑓𝑓1 (𝑥𝑥) > ℎ2 𝑓𝑓2 (𝑥𝑥) (2)
Where,
x: n-dimensional input vector,
ℎ𝑖𝑖 : A priori probabilities that x belongs to a condition class i.
𝑓𝑓𝑖𝑖 : Probability density function of class i.
20
In the PNN, the smoothing parameter (𝜎𝜎) is considered the most important
parameter to be determined (Hajmeer and Basheer, 2002). Different values of (𝜎𝜎) are
chosen where the network is trained and tested for each value and (𝜎𝜎) is selected
based on the results that generate the least misclassification.
Application:
Prediction of the deterioration rates of sewer pipes using ANN based model was
developed by Najafi and Kulandaivel (2005). The factors that were considered in the
development of this model were age, diameter, length, material, depth, slope, and
effluent type. Based on the model results, the diameter of the sewer had the highest
importance, while the slope was the least important factor. The model could predict
the condition rating of pipelines by entering the seven factors of a specific pipe to the
model and the output would be the condition state for this pipe. Structural condition
assessment model for sewer pipelines using both Back Propagation Neural Networks
(BPNN) and Probabilistic Neural Networks (PNN) was developed by Khan et al.
(2010). The two models were built using pipe age, length, depth, diameter, material,
and bedding material. The accuracy of the two models was compared based on the
outputs of each and it was found that the results from BPNN was more accurate than
that from PNN.
Critique:
ANN models are capable of dealing with the pipe deterioration in the absence of
clear relationships between inputs and outputs (e.g.: functional relationships aren’t
identified) (Zou et al., 2008). By analyzing relationships between input and output
data, ANN models can identify and replicate complex non-linear processes. However,
the models developed using ANNs depend primarily on extensive amount of datasets
21
to create a proper environment to develop such relationships. In addition to this
disadvantage, understanding the underlying mechanics in ANN models isn’t always
easy as they are categorized under ‘black box’ models due to the fact that they contain
hidden underlying processes (Tran, 2007).
2.2.2.2 Rule Based Fuzzy Logic
Model Description:
Fuzzy techniques are mathematical forms that address uncertainties and
impreciseness (Zadeh, 1965). Fuzzy rule based modeling, models the relationships
between variables using fuzzy if-then rules which follow the term “antecedent
proposition”. The antecedent proposition is a fuzzy proposition in which (𝑥𝑥)
(linguistic variable) is expressed in (𝐴𝐴) (linguistic constant term). Based on the
similarity between (𝑥𝑥) and (𝐴𝐴), the proposition’s value which is between zero and 1
is assigned (Mamdani, 1975). The Mamdani antecedent proposition has the ability to
deal with the qualitative and highly uncertain knowledge in the form of if-then rules
as shown in Equation 3 (Kleiner, 2007).
𝑅𝑅𝑖𝑖 : 𝑖𝑖𝑖𝑖 𝑥𝑥 𝑖𝑖𝑖𝑖 𝐴𝐴𝑗𝑗 𝑡𝑡ℎ𝑒𝑒𝑒𝑒 𝑦𝑦 𝑖𝑖𝑖𝑖 𝐵𝐵𝑘𝑘 , 𝑖𝑖 = 1,2, … 𝑙𝑙, 𝑗𝑗 = 1,2, … , 𝑀𝑀, 𝑘𝑘 = 1,2, . . 𝑁𝑁 (3)
Where,
x is the input (e.g.: antecedent) linguistic variable,
y is the output (e.g.: consequent) linguistic variable,
Aj is M antecedent linguistic constant in a set A,
Bk is N consequent linguistic constant in a set B,
The values of 𝑥𝑥, 𝑦𝑦, 𝐴𝐴𝑗𝑗 and 𝐵𝐵𝑘𝑘 are given the value from predefined sets and rules
that define the model. 𝑅𝑅𝑖𝑖 (𝑥𝑥, 𝑦𝑦) is considered as a fuzzy relation in an interval of [0,1],
22
in which 𝑅𝑅𝑖𝑖 is a function of Cartesian ordinates (𝑥𝑥, 𝑦𝑦) taking a value in the interval of
[0,1]. To determine the sewer pipelines deterioration, rule based fuzzy techniques
models are used in most occasions to overcome the scarcity and impreciseness of data
(Kleiner, 2007).
Application:
A condition assessment model was developed using fuzzy based approach by Yan
and Vairavamoorthy (2003). In this model, different factors affecting sewer pipeline
condition such as age, diameter, material, depth, in addition to other linguistic factors
were considered. These linguistic factors were transformed into numerical values
through fuzzy rules. A Fuzzy-rule based Markovian process was used to model
deterioration of large diameter buried pipelines (Kleiner et al., 2004). The
methodology was then applied on sewerage pipelines by Kleiner et al. (2007). The
model was built by fuzzifying the age and condition of pipeline into a triangular fuzzy
membership functions, from which the deterioration rate of the same pipe was
determined. Deterioration rate and current condition states were used to determine the
future condition of the pipe. In another research by Rajani et al. (2006), the authors
presented the classification of sewer pipelines in fuzzy sets. To achieve such goal, the
different distress indicators (e.g.: defects) were converted into fuzzy sets by assigning
each distress indicator seven linguistic values which were: excellent, good, adequate,
fair, poor, bad, fail based on the defects values. The distress indicators were then
aggregated based on the relevant categories where each category would reflect the
specific pipe components reflecting the level of deterioration of each category.
23
Critique:
Although rule based fuzzy models offer a powerful tool to deal with scarcity,
impreciseness, and vagueness of data. One of the major shortcomings in using this
technique is the subjectivity involved when defining the inference rules which are
usually based on experts’ opinions. This could lead to ambiguities resulting from
human judgements.
2.2.2.3 Rule Based Simulation
Model Description:
The aim of models developed by using rule based simulation is to generate large
number of outcomes virtually in order to estimate outcomes as in reality. In the rule
based simulation models, a system is defined as a collection of objects which are
called entities. The characteristics of these entities are called attributes which define
the state of the system for which the change in it would be called a state transition
(Inomata et al., 1988). The state of the system and state transitions are called events
and usually define the dynamics of the system.
Application:
A rule based simulation model was developed as a condition assessment tool for
sewer pipelines by Ruwanpura et al. (2004). The model was developed to determine
the condition of sewer pipelines, and the probability that the pipe would remain in its
current condition based on 5 years increment. The model was built using three factors
which were: age, material and length. Random number generators were used to
predict the generate condition rating probabilities of sewer pipelines and were
compared with the actual condition rating probabilities. This step was performed
several times, to determine the most probable condition rating from the overall
24
number of iteration for a given pipe. Future condition ratings were predicted using
Markov chain with the same philosophy that was used in determining the current
condition rating probabilities. In another research using rule based simulation, a
model was developed using the physical, environmental and operational factors
affecting the condition of sewer pipelines to determine their condition states (Hawari,
2016). In this model different factors affecting sewer pipeline conditions were studied
and identified and their effect values were determined using Fuzzy Analytical
Network Process (FANP). By simulating the product of relative weights and effect
values for different factors, several number of iterations, the overall condition of the
pipeline was determined.
Critique:
Condition ratings using rule based simulation models can be estimated from
limited data points due to the fact that simulation technique depends primarily on
generating random probabilities and comparing these probabilities with real data.
Nevertheless, data points are assumed to have the same deterioration trend as the
adjacent points (e.g.: previous or next data points) which could lead into uncertain
results.
2.2.2.4 Expert Systems
Model Description:
Expert systems try to mimic both knowledge and reasoning in an attempt to clone
or replace the experts to solve a problem in a specific area (Durkin, 1994). Expert
systems usually consists of knowledge base, working memory and inference engines.
When solving a problem using expert systems, overlapping rules -usually made of if-
then structures are used (Durkin, 1994).
25
The inference engine in expert systems represents the reasoning in which the
system decides to consider the rule or terminate it.
Application:
Hahn et al., (2002) developed a knowledge based expert system using Water
Research Center (WRC) factors affecting sewer pipelines conditions (WRC, 2001) to
prioritize sewer inspections. The knowledge base in the expert system consisted of six
mechanisms, to determine the likelihood and two mechanisms to determine
consequences of failure. These mechanisms were derived from interviews with
experts and professionals working in the field of sewage pipelines. The inference
engine used was Bayesian Belief Network (BBN) to combine the likelihood and
consequences of failure to determine the pipelines’ risk of failure. In the context of
using BBN as an inference engine, Elmasry et al., (2016) employed BBN to determine
the structural, operational and overall condition states of sewer pipelines based on
defects that could be present in them. To build the BBN, CCTV inspection reports
were used to determine the marginal and conditional probabilities for the different
variables in the BBN. Monte Carlo Simulation was used to eliminate uncertainties in
the estimated probabilities and based on the severities of each defect, the condition
states were determined.
Critique:
Expert systems have the ability to pass the knowledge of experts in a certain field
leading to efficient and accurate problem solving. Additionally, it can separate
knowledge from inference eliminating subjectivity resulting in heuristic problem
solving. However, it is only limited to solvable problems (e.g.: can’t be used in new
research with no prior experience in the area). The aforementioned expert systems
26
models could determine the risk of failure and highlight the critical sections in sewer
networks, nevertheless results were compared to real life data and were found to be
conservative which could be attributed to the narrow domain that expert systems are
subject to, resulting in mistakes.
2.2.3 Statistical Models
2.2.3.1 Multiple Linear Regression
Model Description:
Linear regression finds the linear relationship between the independent variables
(effecting factors) and one dependent variable (pipe condition) (Allison, 1999). The
multiple linear regression equation general form is presented in Equation 4.
𝑌𝑌 = 𝛼𝛼 + 𝛽𝛽1𝑋𝑋1 + 𝛽𝛽2𝑋𝑋2 + ⋯𝛽𝛽𝑛𝑛𝑋𝑋𝑛𝑛+ 𝜀𝜀 (4)
Where,
𝑌𝑌 = dependent variable,
𝛼𝛼 = intercept,
𝛽𝛽1....𝛽𝛽𝑛𝑛= regression coefficients. Ordinary lest squares method is considered to
determine the regression coefficients,
𝑋𝑋1....𝑋𝑋𝑛𝑛= independent variables,
ε = error term.
In modeling the deterioration of an infrastructure asset, the condition state of the
infrastructure is the dependent variable of the multiple linear regression equation and
the contributing factors that affect the condition of this asset are the independent
variables.
27
Application:
Structural and operational condition assessment models for different sewer pipes
materials were developed using multiple regression technique (Chughtai and Zayed,
2007a, 2007b, 2008). The authors used eight factors to develop the structural
condition prediction model which were: pipe age, length, diameter, depth, material,
material class, bedding factor and street category. However, only pipe material, age,
length, diameter and bed slope were used to develop the operational condition
prediction model. A best subset analysis was carried out to determine most significant
factors that should be included in the developed models as dependent variables.
Critique:
Although ordinary regression provides a flexible and simplistic method to predict
the condition rating of sewage pipelines, there are some assumptions made when
using such technique that could make the accuracy of the developed model
questionable. One of these assumptions is that the distance between consecutive
condition states is assumed to be constant, which is not the case in the deterioration of
sewage pipelines. Furthermore, error term is assumed to be normally distributed in
ordinary regression, which is violated most of the time because ordinary regression is
used to model condition states that are considered ordinal response variables. In
addition to these limitations, pipe deterioration is a complex process and might not be
accurately presented by the linear regression relationship between the independent
variables and condition rating (Salman, 2010).
28
2.2.3.2 Logistic Regression:
Logistic regression can be considered as a special type of linear regression
technique where it predicts the outcome of a categorical variable. In logistic
regression, dependent variables are transformed into the logit of dichotomous output
variable.
2.2.3.2.1 Binary Logistic Regression:
Model Description:
Logistic regression can be used to analyze the relationship between a binary
outcome (e.g.: success or failure) and a number of independent variables, which is
called binary logistic regression analysis; in such cases. In binary logistic regression
models, the outcome variable, (y) is categorical and depends on the independent
variables, x1, x2,… xn in a set (n). Based on the probabilities associated with the values
of (y), the outcome variable is calculated. If (y)’s value is equal to 1, this means that
the pipe is in a good condition state and if the value is 0, this means that it is in a poor
condition state. The hypothetical population proportion for which (y = 1) is defined as
P(y = 1) = 𝛑𝛑. Consequently, P(y = 0) = 1-𝛑𝛑 and the odds of having (y = 1) is equal to
π
(1−π). Equation 5 shows the general logit function of binary logistic regression.
𝜋𝜋 𝑝𝑝(𝑦𝑦=1|𝑥𝑥 ,…𝑥𝑥𝑛𝑛 )
𝑙𝑙𝑙𝑙𝑙𝑙 �1−𝜋𝜋� = 𝑙𝑙𝑙𝑙𝑙𝑙 �1−𝑝𝑝(𝑦𝑦=1|𝑥𝑥1 � = 𝛼𝛼 +𝛽𝛽1𝑋𝑋1 + 𝛽𝛽2𝑋𝑋2....+𝛽𝛽p𝑋𝑋p
1 ,…𝑥𝑥𝑛𝑛 )
(5)
= ∑𝑝𝑝𝑗𝑗=1 β𝑗𝑗 𝑥𝑥𝑗𝑗
Where,
𝛼𝛼 is the intercept parameter.
𝛽𝛽s are the regression coefficients associated with the n independent variables.
29
The probability of (y =1) can be determined by using an exponential
transformation, as shown in Equation 6.
𝛼𝛼+∑𝑛𝑛
𝑗𝑗−1 β𝑗𝑗 𝑥𝑥𝑗𝑗
𝑃𝑃(𝑦𝑦 = 1|𝑥𝑥1 … . . 𝑥𝑥𝑛𝑛 ) = 𝑒𝑒 𝛼𝛼+∑𝑛𝑛
(6)
1+𝑒𝑒 𝑗𝑗−1 β𝑗𝑗 𝑥𝑥𝑗𝑗
Where,
𝛼𝛼 and 𝛽𝛽 are parameters estimated from data using the maximum likelihood estimation
(MLE).
Application:
Logistic regression technique was used to predict the probability of sewer pipes to
be in a certain state (Ariaratnam et al., 2001). The authors in this research used age,
diameter, material, waste type, and average depth of cover as the contributing factors
that would affect the condition of sewer pipes. Wald test was used to examine the
significance of the coefficients of the independent variables in logistic regression
model. It was found that the depth and material were not significant and that the
coefficient of age increases the odds of deficiency by 2.6% per year.
A binary logistic regression model was also developed for the prediction of the
probability that a sewer pipeline would be in a deficient state for Edmonton, Canada.
The model was developed for the intent of inspection prioritization (Ariaratnam et al.,
2001).
2.2.3.2.2 Multinomial logistic regression
Model Description:
Multinomial logistic regression is an extension of binary logistic regression and
can be used when dependent variable is categorical and has more than two levels. For
a dependent variable with (k) categories, the multinomial regression model estimates
30
(k-1) logit equations. A generation of (k-1) logits from the remaining (k – 1)
categories can be determined as per Equation 7.
P(Y = i|x1 … … xn )
𝑙𝑙𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽𝑖𝑖1 𝑥𝑥1 + 𝛽𝛽𝑖𝑖2 𝑥𝑥2 + ⋯ . +𝛽𝛽𝑖𝑖𝑖𝑖 𝑥𝑥𝑛𝑛 (7)
P(Y = k|x1 … … xn )
Where,
i = 1, 2, …, k – 1 correspond to categories of the dependent variable,
xs are independent variables,
n is the number of independent variables,
𝛽𝛽0 is the intercept for category i,
βis are the regression coefficients of independent variables defined for each category i.
𝛽𝛽0 and βis values for each (k – 1) logit equation can be estimated by multinomial
logistic regression (Agresti, 2002).
Therefore for a dependent variable with (k) levels and a total number of (p)
independent variables, the multinomial logistic regression models estimate (k – 1)
intercepts, and p*(k – 1) regression coefficients. Calculation of probabilities
associated with each category of the dependent variable is shown in Equations 8 and
9.
exp(β0 + βi1 x1 + βi2 x2 + ⋯ . +βin xn )

P(Y = i|x1 … xn ) = πi (x) =
[1 + ∑k−1
i=1 (β0 + βi1 x1 + βi2 x2 + ⋯ . +βin xn )] (8)
for i = 1,2, … k − 1
1
P(Y = k) = πk (x) =
[1 + ∑k−1
i=1 (β0 + βi1 x1 + βi2 x2 + ⋯ . +βin xn )] (9)
for i = k
31
Application:
Multinomial logistic regression was used to develop a model for predicting the
financial needs for rehabilitation of sewage network over a specific planning horizon
of years (El-Assaly et al., 2006). The cost in this model was estimated as the product
of the predicted defected pipe and the cost of the repair method for the same pipe.
Different pipes were arranged in an ascending order to determine the ones with the
highest cost, which would require rehabilitation. Using the same technique, a model
for sewer pipes deterioration was used to determine risk of failure of pipes
considering the different geographical, physical, and functional factors affecting
sewer pipes using both multinomial logistic regression, and binary logistic regression
(Salman, 2010). Logistic regression gave the most accurate results when probability
of failure was estimated.
Critique:
Logistic regression analysis helps in identifying the most important variables
affecting the sewer pipelines condition which provides a better understanding for the
trend of the deterioration process. In addition, no assumptions are made on the
distributions of the independent variables which can be considered as one of the main
advantages of this technique. The main disadvantage of using this technique is that a
satisfactory amount of data for the factors affecting sewer deterioration is required in
order to build the model, which sometimes is considered a challenge in some
municipalities due to poor filing system and documentation.
32
2.2.3.2.3 Multiple Discriminant Analysis
Model Description:
Multiple discriminant analysis is used to analyze the linear relationship between a
dependent categorical variable (e.g. pipe condition) and a number of predictor
variables (e.g. deterioration factors). This method is similar to multinomial logistic
regression technique, since more than two outcomes can be handled. The model is
constructed based on available observations set in which the outcome is known.
Functions are constructed using a set of linear functions of the predictor set, where
these functions are called classification functions as shown in Equation 10.
𝐿𝐿𝑖𝑖 = 𝛽𝛽0 + 𝛽𝛽𝑖𝑖1 𝑥𝑥1 + 𝛽𝛽𝑖𝑖2 𝑥𝑥2 + ⋯ . +𝛽𝛽𝑖𝑖𝑖𝑖 𝑥𝑥𝑛𝑛 (10)
Where,
𝐿𝐿𝑖𝑖 is the classification function score for class (i)(i = 1 to k-1, with k being the number
of classes),
𝑥𝑥𝑖𝑖 are the predictor variables,
𝛽𝛽𝑖𝑖𝑖𝑖 are classification coefficients corresponding to n-number of predictors variables
and 𝛽𝛽0 is a constant.
The methodology behind this multiple discriminant analysis is described in Figure
2-4. Any observation could be visualized in an n-dimensional space where the axes
are the classification functions (Li ). Figure 2-4 shows three different classes,
consequently only two classification functions are required (L1 and L2 ) to classify the
observations. Since, the new prediction is closer to the centroid of class 3, the
prediction outcome would be 3.
33
Figure 2-4: Multiple discriminant analysis visualization (Tran, 2007)
Application:
Multiple discriminant analysis was employed to model the deterioration of gravity
pipelines in Australia by Tran et al. (2007). The predictor variables included diameter,
age, depth, slope, location, roots of trees, soil properties and hydraulic condition.
Unfortunately, this model showed low accuracy with less than 50% prediction
abilities. It was found that the pipe age was insignificant which comes in line with the
low accuracy, while the most significant factor was found to be the hydraulic
condition. Also, it was applied by Ana (2009) to predict the condition of sewer pipes
in Leuven and Antwrep among other techniques to compare between the suitability of
using them in prediction.
Critique:
Models developed using multiple discriminant analysis could provide direct
information about the score and class for the state of the pipes, not in the form of
probabilities unlike logistic regression models. It can also provide information on the
34
most important variables affecting the deterioration process which helps in better
understanding the pipe deterioration trend. Not only does multiple discriminant
analysis require sufficient set of data and linearity similar to the ones required in
logistic regression technique, but also there should be normality and lack of multi-
collinearity between independent variables which can be considered as one of the
major drawbacks in the application of this method. If the normality assumptions are
satisfied, multiple discriminant analysis is expected to give better predictions.
Otherwise, logistic regression would be more suitable (Pohar et al., 2004).
2.2.3.3 Evidential Reasoning
Model Description:
Evidential reasoning describes and handles various types of uncertainties by using
the concept of the degrees of belief, in which each attribute of an alternative of a multi
criteria decision making problem is described by a distributed assessment using a
belief structure. Unlike conventional approaches that require scaling grades and
averaging scores to aggregate attributes, the evidential reasoning approach aggregates
belief degrees by employing an evidential reasoning algorithm.
It aggregates two factors at a time and the resulting aggregation of the first two
factors of evidence is aggregated with the third factor of evidence and so on.
Equations 11, 12 and 13 show Dempster-Shafer rule combination of two basic
probability assignments which is the theory used in developing the evidential
reasoning algorithm.
35
𝑚𝑚12 (𝛹𝛹) = 𝑚𝑚1 (𝛹𝛹) ⊕ 𝑚𝑚2 (𝛹𝛹) (11)
= 0, when 𝛹𝛹 = Φ (12)
∑𝐴𝐴∩𝐵𝐵=𝛹𝛹,𝐴𝐴,𝐵𝐵⊆⊕ 𝑚𝑚1 (𝐴𝐴)𝑚𝑚2 (𝐵𝐵)

= 𝛹𝛹 = Φ (13)
1−𝑘𝑘
Where,
k = ∑𝐴𝐴∩𝐵𝐵=𝛹𝛹,𝐴𝐴,𝐵𝐵⊆⊕ 𝑚𝑚1 (𝐴𝐴)𝑚𝑚2 (𝐵𝐵), representing the conflict between subsets A and B,
𝑚𝑚1 (𝛹𝛹)𝑚𝑚2 (𝛹𝛹) two basic probabilities to a subset 𝛹𝛹.
From the above equations, the evidential reasoning general equation for
evaluating an attribute with contributing factors can be given by Equation 14.
𝑙𝑙
𝐸𝐸𝑘𝑘 = 𝑒𝑒𝑘𝑘1 ⊕ 𝑒𝑒𝑘𝑘2 ⊕ 𝑒𝑒𝑘𝑘3 ⊕ … … . 𝑒𝑒𝑘𝑘𝑘𝑘 (14)
Where,
𝑙𝑙𝑘𝑘 denotes the number of factors that contribute to the 𝑘𝑘 𝑡𝑡ℎ attribute,
𝑒𝑒𝑘𝑘𝑖𝑖 is the evaluation of each contributing factor (i+)
Application:
A condition assessment model of buried pipes using hierarchical evidential
reasoning was developed by Bai, et al., (2008). Inferences for condition assessment
was done using the hierarchical evidential reasoning approach that employed the
Dempster-Shafer theory. The developed model was built using a hierarchical
framework for pipe condition assessment in which all contributing factors/attributes
were assumed independent, and only the parallel aggregation of factors/attributes
were performed using Dempster-Shafer rule of combination. This model took into
considerations, factors that affect the integrity of pipelines wall such as cement lining
condition and the degree of internal corrosion. Evidential reasoning using fuzzy set
36
theory was used to determine the overall pipeline conditions using possible defects
(Daher, 2015). Defects that could be present in different pipeline components were
divided into three categories namely structural, operational and installation defects.
The different defect families and categories along with the sewer pipeline components
were given weights based on their relative importance and how they contribute to the
overall condition of the pipelines. Fuzzy based evidential reasoning was used to
aggregate the different defects in the respective pipeline components from which the
overall pipeline condition and each of the structural, operational and installation
condition of different components were determined.
Critique:
Evidential reasoning is capable of dealing with incomplete and conflicting
evidence without having to make any assumptions about missing data. In addition,
they could combine multiple bodies of evidence. One of the major disadvantages of
using hierarchical evidential reasoning is its inability to deal with dependent factors as
well as the conflict between them without using auxiliary rules of combination.
2.2.3.4 Cohort Survival
Model Description:
Cohort survival model is used to describe the process of sewer deterioration for
homogeneous pipes having similar characteristics (e.g. cohorts). It is assumed that
sewer cohorts with certain probability survive a number of years in a certain condition
state. These cohorts pass through successive transitions, from the current condition
state to a worst condition state during their service life (Baur et al., 2004). This
transition is described by condition survival curves which are known as transition
37
functions. The transition function applied in assessing conditions of pipelines of the
Herz distribution is given by Equation 15 (Herz, 1995 and 1996).
𝑎𝑎𝑖𝑖→𝑖𝑖+1 + 1
𝑆𝑆(𝑡𝑡)𝑖𝑖→𝑖𝑖+1 = (15)
𝑎𝑎𝑖𝑖→𝑖𝑖+1 + 𝑒𝑒 𝑏𝑏𝑖𝑖→𝑖𝑖+1 (𝑡𝑡−𝑐𝑐𝑖𝑖→𝑖𝑖+1 )
Where,
𝑆𝑆(𝑡𝑡)𝑖𝑖→𝑖𝑖+1 is the fraction of pipes at age t which have survived until condition i or
better,
A is the aging factor (a = 0 means that no aging takes place),
B is the transition parameter (the transition is faster when b is large),
C is the resistance time which determines the age where no further deterioration is
anticipated.
Each different cohorts require calibration of the parameters a, b and c of the Herz
transition function. This could be performed by minimizing the deviation between the
fraction of sewers at certain condition i between the expected values yielded from the
model and actual data. Only installation year, inspection year and the condition state
are required to build the survival functions, unlike the previous discussed methods.
Application:
Several tools have been developed using cohort survival models that have been
implemented by Horold, (1998) and Horold and Baur (1999). The transition curves
developed for Norwegian network shown in Figure 2-5 was used to determine the
remaining service life for sewer pipes towards reaching the worst condition states. For
instance, the 50 years old group of sewers were found to be in condition state 3 from
CCTV inspections. The first pipe in this group to reach condition 5 was estimated to
be after 48 years which represents the minimal remaining service life (RSL).
38
Similarly, the last pipe within the group was anticipated to reach condition 5 after 105
years which represents the maximum RSL. The average RSL of the group was
determined by measuring the horizontal distance from the middle of the group to the
transition curve representing condition 5. From this, it is recommended to use the
average RSL when planning for future maintenance (Jansen, 2007).
Figure 2-5: Norwegian network’s transition curves (Horold, 1998)
Equation 15 was used in the analysis of an existing sewage network in Germany
to assess the condition of this network (Baur and Herz, 2002). Weighted least squares
method was used to estimate the parameters of transition functions. The rate by which
the pipes age was calculated by determining the midpoint of two areas. The first area
was bound between the transition curves corresponding to the transition from
previous condition state to current condition state and the second area was bound
between current condition state and the next condition state. As for the residual life of
the pipes, it was predicted based on the rate of pipes’ aging.
39
Critique:
The simplicity in the concept behind the development of cohort survival model is
considered one of the main advantages. However, significant amount of inspection
data are needed for each cohort in each condition state to properly calibrate the
transition functions (Fenner, 2000). The main difficulty in developing cohort survival
models arises from the lack of inspected data of certain conditions, since the operator
tend to concentrate his inspections on specific sewer types such as old sewers and
sewers in poor conditions (Ana and Bauwens, 2010). Also, there is usually an
underestimation of the pipes in worst condition state as they may have already
collapsed and were not included in the data under study. Consequently, there would
be an overestimation in the predicted survival rates and remaining service life. A
correcting model calibration has been suggested to fix this anomaly which is referred
to as selective survival bias by adding weights to the model (Le Gat, 2008).
2.2.3.5 Markov Chains
Model Description:
Markov chain is a stochastic process which has been used to describe the
deterioration of sewer pipes passing through a number of condition states. This
process could be described as ‘memoryless’ as the conditional probability that an
asset could have in the future depends only on its current condition (Ross 2000). In
Markov chains, a transition probability matrix represents the probability values for an
asset to remain in its current condition state or transfer to another condition state. The
general form of a transition matrix can be expressed by Equations 16 and 17.
40
𝑝𝑝11 𝑝𝑝12 … 𝑝𝑝1𝑚𝑚
𝑃𝑃 = � … … … … � (16)
𝑝𝑝𝑚𝑚1 𝑝𝑝𝑚𝑚2 … 𝑝𝑝𝑚𝑚𝑚𝑚
𝑚𝑚
� 𝑝𝑝𝑖𝑖𝑖𝑖 = 1 (𝑓𝑓𝑓𝑓𝑓𝑓 𝑖𝑖 = 1,2,3, … 𝑚𝑚) (17)

𝑗𝑗=1
Where,
pij is the conditional probability of an asset to be in condition state (j) with a current
state (i) after a unit transition period.
The transition probabilities can be either time dependent (non-homogeneous
Markov model) or time independent (homogeneous Markov model) (Ana and
Bauwens, 2010). Since pipe deterioration is age-dependent as new pipes deteriorate
slower than older pipes, time dependent Markov model is considered to be more
representative (Kleiner, 2001).
In pipe deterioration, transition can only occur from the current condition state to
worst condition states as it is impossible for a pipe to improve its condition without
interventions. Therefore, pij is considered as 0 for i > j. Also, the pipe cannot further
deteriorate in condition after reaching the worst condition state (m). Thus, the
transition probability pmm = 1. Hence the non-homogeneous transition probability
matrix can be expressed by Equation 18.

𝑡𝑡.𝑡𝑡+1 𝑡𝑡.𝑡𝑡+1 𝑡𝑡.𝑡𝑡+1
𝑡𝑡.𝑡𝑡+1 𝑡𝑡.𝑡𝑡+1
𝑡𝑡,𝑡𝑡+1
𝑝𝑝𝑖𝑖𝑖𝑖 =� 0 𝑝𝑝22 … 𝑝𝑝2𝑚𝑚 � (18)
… … … …
0 0 … 1
Further simplified form of the transition matrix has been considered by assuming
that the transfer in condition only drops one level at a time (Wirahadikusumah et al.,
2001 and Le Gat, 2008). This can be expressed by Equation 19.
41
𝑝𝑝11 𝑝𝑝12 0 … 0
⎡ 𝑡𝑡.𝑡𝑡+1 𝑡𝑡.𝑡𝑡+1
⎤
⎢ 0 𝑝𝑝22 𝑝𝑝23 … 0 ⎥
𝑡𝑡,𝑡𝑡+1
𝑝𝑝𝑖𝑖𝑖𝑖 =⎢ … … … … … ⎥ (19)
⎢ 0 … … 𝑝𝑝𝑚𝑚−1,𝑚𝑚−1 𝑝𝑝𝑚𝑚−1,𝑚𝑚 ⎥
⎣ 0 0 … 0 1 ⎦
However, the above might not be appropriate to model structural deterioration as
the pipe might deteriorate by several condition states at a single time step (Micevski et
al., 2002). To solve this problem, Kleiner (2001) has recommended the use of short
transition periods where only deterioration by a single condition state is anticipated.
This can be expressed by Equations 20 to 22.
𝑄𝑄(𝑡𝑡) = {𝑞𝑞1 , 𝑞𝑞2 , … 𝑞𝑞𝑚𝑚 } (20)
𝑄𝑄(𝑡𝑡 + 𝑠𝑠) = 𝑄𝑄(𝑡𝑡)𝑝𝑝𝑡𝑡,𝑡𝑡+1 𝑝𝑝𝑡𝑡+1,𝑡𝑡+2 … 𝑝𝑝𝑡𝑡+𝑠𝑠−1,𝑡𝑡+𝑠𝑠 (21)
𝐸𝐸(𝑡𝑡 + 𝑠𝑠) = 𝑄𝑄(𝑡𝑡 + 𝑠𝑠)𝑆𝑆 𝑇𝑇 (22)
Where,
𝑄𝑄(𝑡𝑡) is the probability mass function (pmf),
𝑞𝑞𝑖𝑖 is the probability of the system being in state i at time t,
𝑄𝑄(𝑡𝑡 + 𝑠𝑠) is the probability mass function after s time steps.
Application:
Deterioration of large diameter combined sewers was modeled using Markov
chains by Wirahadikusumah et al. (2001). Because Markov Chains can be used to
model deterioration of pipelines on a network level, the data used in building the
model were categorized into sixteen groups based on material, groundwater table
level, backfill soil type and the depth. Regression analysis was performed to
determine the relationship between condition rating and time where non-linear
optimization was used to convert the relationship between condition states and time
into a Markov-chain model. Probabilities in the transition matrix were determined by
42
minimizing the sum of the absolute differences values between the condition states
which were estimated using exponential functions and Markov-chain model
estimations.
In another condition assessment model that was developed for sewer pipelines in
Waterloo, Canada, transition probabilities in Markov Chains were determined by
minimizing the sum of the absolute differences values between the condition states
which were estimated using polynomial regression and Markov-chain model
estimations (Sinha and McKim, 2007). Kleiner et al., (2004) modeled the
deterioration of buried infrastructures by employing a fuzzy-rule based non-
homogenous Markovian process to which was then applied on sewage pipelines to
determine their condition states (Kleiner et al., 2007). The age and condition of the
pipe were modeled using triangular fuzzy sets and the deterioration rate of the pipe
was determined by using a fuzzy rule set. Future condition state was determined based
on the deterioration rate value obtained from this procedure and the current condition
state. Pipe physical properties such as age, diameter, material and slope were used to
calculate the transition probabilities to be used in Markov Chain model for the
deterioration of sewer pipelines (Baik et al., 2006). Similar to this methodology, a
research was carried out in which the transition probabilities were obtained using
Gompertz distribution and were calibrated using diameter, sewer type and installation
period (Le Gat, 2008).
Critique:
Modeling sewer pipelines deterioration using Markov’s chain allows the modeling
of complex and chronological events which could help in capturing the deterioration
behavior of sewer pipelines. One of the difficulties in using Markov chains is
43
determining the transitional probability matrix. Also, the absence of previous
historical inspection records could increase the challenges accompanying the use of
Markov chains which would be more noticeable if dataset should be divided into
cohorts (e.g.: clusters of same characteristics) and for each cohort, new Markov-chain
deterioration curve has to be generated. However, such limitation could be solved by
using an additional technique to estimate transition probability values, nevertheless
the results could be affected greatly based on the chosen technique. Additionally,
unless different transition matrices are applied for different time steps, the
deterioration rate is assumed to be time independent which doesn’t truly represent the
dynamic nature of deterioration in sewer pipelines.
2.2.3.6 Semi Markov Chains
Model Description:
The semi Markov chain is similar to the Markov chain with the ability of
modeling the waiting time that an asset would spend in a certain state. This waiting
time is usually considered to follow a random distribution (Lawless, 1982). When
modeling sewer pipelines deterioration using semi Markov chains, pipelines are
assumed to spend a random interval in each state that can be translated into a
probability distribution. Equation 23 shows the mathematical formulation of the
cumulative waiting time of an asset in a certain state with a deterioration trend
following semi Markov chain.
𝑇𝑇𝑖𝑖−𝑘𝑘 = ∑𝑘𝑘−1
𝑗𝑗=𝑖𝑖 𝑇𝑇𝑗𝑗,𝑗𝑗+1 , i ={1,2,...,m-1}, k ={2,3,...,m} (23)
44
In this equation the cumulative time is represented as a cumulative distribution for
the random variable 𝑇𝑇𝑖𝑖−𝑘𝑘 which represents the time that an asset would take to transfer
from state (i) to state (k).
Application:
Deterioration of large buried pipelines has been modeled using semi Markov
chains, where the transition probabilities were linked to their age (Kleiner, 2001). In
order to model the transition probabilities in a certain state (i) at a time (t) to another
condition state (j) at a time (t+1), Equation 24 was used.
𝑓𝑓1−𝑖𝑖 (𝑇𝑇1−𝑖𝑖 )
𝑝𝑝𝑖𝑖,𝑖𝑖+1 = (24)
𝑆𝑆1−𝑖𝑖 (𝑇𝑇1−𝑖𝑖 ) − 𝑆𝑆1−(𝑖𝑖−1) (𝑇𝑇1−(𝑖𝑖−1) )
Where,
𝑝𝑝𝑖𝑖,𝑖𝑖+1 is the transition probability of the asset from a certain state to the next one,
𝑓𝑓1−𝑖𝑖 (𝑇𝑇1−𝑖𝑖 ) is the probability density function of the variable (𝑇𝑇1−𝑖𝑖 ).
In this model, the transition probabilities were derived with the aid of Weibull
distribution that was used to determine the distribution of waiting times, where
expert’s opinions were used to determine the Weibull distribution’s parameters. The
sum of waiting time in different states of the asset which represented the cumulative
probability function was calculated using Monte-Carlo simulation.
Critique:
In semi-Markov chain, lack of data problem required for determining transition
probability can be solved by using the expert’s opinions. However, in order to
determine the distribution of waiting time in the developed models, adequate dataset
is required for the condition of pipes and history of inspections which might be
45
considered as a challenge because of absence of historical records for inspected
pipelines.
2.2.4 Summary
This section presented the different physical, artificial intelligence and statistical
models developed to assess the condition of sewer pipelines. In addition, factors
affecting the condition of sewer considered in previous studies were determined. Age,
length, material and diameter are the most common factors that were included in these
models.
The models discussed in this chapter can be divided into two types namely: pipe
level models and pipe group models. In the pipe level models, the condition is
assessed for individual pipes without considering the global deterioration of the
network which could be suitable for scheduling inspections and optimizing the
rehabilitation or inspection policies with respect to the number of pipelines addressed.
While, in the pipe group level models, the condition of the whole network is assessed
based on pipelines with similar characteristics from which strategic decisions can be
made regarding budgetary allocation for the network rehabilitation and maintenance.
Table 2-2 shows a classification for the different techniques used in developing
condition assessment models for sewer pipelines from the aforementioned
perspective. It can be noticed from the table that, models such as survival functions
and discriminant analysis provide the life expectancy of pipelines and predict the
condition state of pipes while logistic regression models could assess the probability
of failure making it much more suitable to be used in risk based management of sewer
pipelines than the former two models.
46
Table 2-2: Assessed Levels and Expected Outcomes for Each Condition Assessment Model
Rule Based Regression
Evidential Reasoning
Survival Function
Markov Chains
Expert Systems
Analysis
Models
ANN
Simulation
Multiple
Logistic
Pipe
  Fuzzy
      
Level
Network
         
Level
Condition Sate Condition Probability of Remaining Life time of Condition Sate

Output Condition State of Condition Sate of
of individual State of failure of pipelines and Proportions of of individual
from the individual pipes or individual pipes or
pipes or pipe individual individual pipes or each pipe at certain condition pipes or pipe
Model pipe cohorts pipe cohorts
cohorts pipes pipe cohorts state cohorts
47
2.3 Problem Statement
The presence of condition assessment models can help in managing assets and
avoiding early failure. It can also provide an accurate prediction of expenses required
in the future through understanding and predicting the remaining asset life and its
condition. Condition assessment models can help in better maintenance and
rehabilitation strategies by determining the required corrective actions (i.e.:
maintenance, rehabilitation, renewal) and their timeframes.
One of the drawbacks to depend on these models in assessing the condition of sewer
pipelines is that one of the main sources of gathering information about factors are
data from CCTV inspection reports which could be either incomplete or ambiguous
resulting in erroneous and uncertain models. Additionally, sometimes gathering such
data is costly or could be hard, which would raise the issue of data reliability.
The problem of data availability is considered one of the challenges that is
worsened when dealing with pipe group level models that require categorizing
pipelines into cohorts based on their characteristics. Therefore and to develop a
reliable condition assessment model, regular inspections have to be performed and
new emerging technologies would be required to facilitate gathering required
information in a complete and accurate manner. The concept of Artificial intelligence
makes models such as ANN, simulation and fuzzy based more robust and
computationally efficient when compared to the statistical ones. However, one of the
main disadvantages is their need for extensive datasets and the difficulty in
understanding the underlying mechanics in these models. Logistic regression,
multiple discriminant analysis, and Markov chains are different techniques that can be
48
used to determine the condition state of pipelines in a network, however an adequate
amount of data regarding the factors affecting sewer deterioration is required and in
some cases the computational efforts are large especially in large scale networks.
There is a crucial need to develop and integrated condition assessment models that
overcome these setbacks. The proposed condition assessment models in this study
were built based on the experience of specialists working in different fields in
drainage networks to overcome the problem of data availability as the collected data
sets were only used for validation purposes. In addition, the proposed condition
assessment models are intended to complement the efforts of others by including new
factors and taking into consideration their interdependencies and minimizing the
uncertainties.
2.4 Research Objectives
The main objective is to build condition assessment models that are expected to be
a useful tool for decision makers to properly plan for their inspections and provide
effective rehabilitation of sewer networks. The sub-objectives of this research can be
summarized as follows:
• To identify and study the different factors affecting sewer pipeline condition
and their severities.
• To identify and study the previously developed condition assessment models.
• To model and assess sewer pipeline condition based on the identified factors.
• To develop new condition assessment models.
49
CHAPTER 3 – RESEARCH METHODOLOGY
The research methodology to develop condition assessment models using FANP
and Monte-Carlo as the first approach and FANP, fuzzy set theory (FST) and
Evidential Reasoning (ER) as the second approach is described in Figure 3-1. A
description of the adopted methodology is further described below.
To develop a condition assessment model for sewer pipelines, the factors
contributing to the deterioration of pipelines and their effects are integrated to develop
an index to represent the condition of pipeline under study. The first part of the
methodology for both approaches was identifying and collecting data related to the
factors that would affect and deteriorate sewage pipelines. After identifying these
factors, two questionnaires - one related to gravity sewer pipelines and another related
to pressurized sewer pipelines were distributed to experts working in the field of
infrastructures and sewage networks to collect the relevant data required in model
development (i.e.: weights, relative importance, and effect of the contributing factors.
FANP was used to address the interdependency between different factors affecting
sewer pipelines conditions and uncertainty when processing data elicited from human
judgment (i.e.: transforming a verbal pairwise comparison judgment into an exact
ratio representing the strength of the alternative when compared to another) in an
attempt to overcome the setbacks that could be encountered when applying Analytical
network process (ANP) solely. The developed models take into consideration the
interdependencies between three factorial groups affecting sewer pipeline conditions,
namely, physical, operational and environmental. FANP integrated with Monte-Carlo
50
Simulation are used to determine the relative weights of these factors. The rest of the
methodology adopted for the first and second approaches differs.
For the first approach, using calculated weights and effect values; probabilistic
condition index for sewer pipelines is determined with the aid of Monte-Carlo
simulation. The outcome generated by Monte Carlo Simulation would form different
probability distributions based on a random sample collection. By determining the
most frequent outcome, the user can see and analyze the range of certainty and
biasness of the simulated outcome that would help in eliminating the uncertainty that
accompanies the model output, thus came the rationale behind utilizing Monte-Carlo
simulation. The assessed pipeline condition in the form of a probability distribution
function is the output of the model.
For the second approach, FST is used to assign the fuzzy membership functions
and thresholds for the severity of the factors’ effects on the pipelines condition. In
order to combine both the effect values and relative weights of factors affecting
sewage pipelines, ER is used as an aggregation technique. It is used to determine the
degrees of belief for the model outputs that represents the user’s certainty level about
how good the condition of the pipeline is, based on the effect value of the different
contributing factors. After combining all the factors degrees of belief, defuzzification
using the FST to generate a final crisp condition index is carried out.
Finally, the developed model is validated using data collected from an existing
sewage network in the city of Doha, Qatar. The validation set included actual
conditions for 549 gravity pipelines with 6 available factors which are: age, diameter,
length, and buried depth, pipeline position relative to groundwater and pipeline
material. In addition to the pipeline actual condition obtained from CCTV analysis.
51
Figure 3-1: Research Methodology
52
CHAPTER 4 – DATA COLLECTION
4.1 Overview
To achieve the main objectives of this research which is developing condition
assessment models for sewer pipelines, all Factors affecting gravity and pressurized
pipelines in sewage networks were studied and included in a questionnaire. A
questionnaire was distributed to experts in the field, to determine the weights of the
identified factors and the severity of their effect on sewage pipeline condition. In
addition, the developed model were validated through a collected data set. This
chapter describes the data collection phase in details.
Data Collected
Factors affecting
sewer pipelines Questionnaires Validation Set
condition
Factor's Weights
(Wi)
Factor's Effect
Values
Figure 4-1: Types of Data Collected
53
4.2 Factors affecting sewer pipelines condition
Identified factors affecting sewer gravity and pressurized pipeline conditions were
divided into three main categories, namely, physical, environmental, and operational
as shown in Figure 4.2. The physical factors included sewer pipeline characteristics
such as: age, material type, size, buried depth, coating conditions and installation
quality. The operational factors included: flow rate, infiltration and inflow, blockages,
corrosive impurities and maintenance strategies in addition to the operating pressure
for pipelines under pressure. Finally, the environmental factors included: bedding
conditions, location, groundwater level and ground disturbance. Table 4-1 provides a
detailed description and definitions for each of the studied factors.
Figure 4-2: Classification of Factors Affecting Sewage Pipelines Condition
54
Table 4-1: Factors Affecting Sewer Pipeline Condition.
Main Sub-
Description
Factors Factors
Age (AG) Effects of pipeline degradation become more significant over time.
Diameter The larger the pipe line diameter, the larger is its thickness, the lower is its
(DI) deterioration rate and vice versa.
Length (LE) Longer pipes are more likely to suffer from bending stresses.
Physical (PF)
Buried Depth Life loads impact increases at shallow depths and the soil overburden impact
(D) increases at high depths. Moderate depths increase the life of sewers
Material
Different pipeline material show different failure patterns.
(MT)
Coating
Conditions Pipelines with good coating conditions have higher resistance against corrosion.
(CC)
Installation Pipeline installation should be done according to certain standards and
Quality (IQ) qualifications. High deterioration rates result from poor installation quality.
Low flow rates causes deposition and accumulation of sediments while high flow
Flow Rate
rates causes corrosion for the piper’s internal walls and causes disturbances
(FR)
specifically when moving between pipes having different diameters.
Accumulation of deposits and sediments, intrusion of trees roots and other types of
Blockages
blockages have a significant effect on the structural and operational condition of a
(B)
sewer pipeline.
Operational (OF)
Infiltration
and Inflow Infiltration washes soil particles and reduces the support along a pipeline.
(II)
Corrosive Sewage water carries substances and chemicals (for example: micro-bio species and
Impurities slats) which impacts the water quality. In addition, these impurities can cause
(CI) corrosion to the internal pipes’ internal surfaces.
Maintenance
and break The service life of sewer pipelines is increased by a good maintenance and break
Strategies strategies.
(MS)
Operating
High pressures resulting in the distribution systems can lead to system fatigue,
Pressure
pump and device failure, or pipe ruptures.
(OP)
The soil which contacts the pipe surface directly has an impact on the deterioration
process. Soils have different physical and chemical properties which have different
Soil Type
impacts on the pipeline. For example, certain soils responds to moisture changes
(ST)
differently in respect to volume changes which applies loading on the pipe while
others are highly corrosive.
Environmental (EF)
Bedding
Conditions Sewer pipeline failure chances increases with improper bedding conditions.
(BC)
The location of the pipeline has an impact on the deterioration process. A pipeline
can be installed in industrial area, residential area, schools, etc. Pipelines located in
Location industrial areas or cities are subjected to different conditions that the pipelines
(LO) located in residential areas. For example: city pipelines are exposed to heavier
traffic loading. In addition, pipelines can be located beneath different surfaces (e.g.
asphalt, walkway, unpaved, etc.).
The amount of water in soil affects the soil resistivity, which inversely relates to the
Groundwater
corrosion rate. The ground water may lead to corroding the pipe directly when salts
Level (GW)
and some corrosive substances exist in it.
Ground
Pipelines existing near a disturbed ground are subjected to high stresses and might
Disturbance
have a sudden collapse.
(GD)
55
4.3 Questionnaires
Open-ended (unstructured) questionnaires were additionally prepared to allow
respondents to include additional factors that may have an impact on sewer pipelines
condition. The questionnaires were distributed among different disciplines to capture
the differences among the experts knowledge on the factors affecting the condition of
sewer pipelines. Questionnaires were sent out to experts working in sewer field
inspections and CCTV analysis, sewer designers, sewer site construction engineers
and managers who have more than 30 years of experience in sewer maintenance and
rehabilitation.
The used questionnaire provided a tool for interviewing experts and incorporating
the elicited information into the developed models. The questionnaires were utilized:
(1) to compare between main factors and sub-factors affecting sewer pipeline
condition and (2) to determine the effect value of each factor on the pipeline
condition. The questionnaires included an introductory part with a graphical
representation of all the identified contributing factors, followed by two main
sections. The first section sought the relative importance of factors and sub factors
when compared to each other and how each factor would strongly affect the pipeline
condition. The second section focused on the effect of the different factors on the
condition of the pipeline.
4.3.1 Factor’s Weights (Wi)
The importance of factors was calculated by conducting a pairwise comparison
between the selected factors. The comparison can be categorized into three levels: (1)
between the main factors with respect to the sewer pipeline condition; (2) between the
56
sub-factors of each main factor; and (3) between the main-factors with respect to each
other. The pairwise comparison of each level was designed to reflect the opinion of
the experts on the degree of importance for each factor over the other(s), with respect
to the goal under consideration. The degree of importance was scaled from 1 to 9
according to Saaty’s nine points scale (Saaty, 1996) with“1” indicating no significant
influence of a factor over the other(s), while “9” indicating that there is an absolute
influence. In Table 4-2 an example for how pairwise comparison was carried out
based on Saaty’s scale in an ANP network is shown. For instance, if the respondent
sees that the “Diameter” has a very strong influence over the “Age” with respect to
“physical factors”, they could check the suitable box that reflects the degree of
influence(i.e.: a value of 7 corresponding to very strong would be assigned in the
pairwise comparison). The same method is then repeated for the rest of the physical
factors and on the different levels of the developed network.
4.3.2 Factor’s Effect Values
Sub-factors can have different characteristics having different effect values on the
condition of sewer pipeline. For instance, the “age” sub-factor has characteristics that
range in value. These ranges were identified based on meetings with experts were a
pipeline is considered new, medium or old if the age ranges from (0-15), (15-30) or
>30, respectively and similarly for the rest of the sub-factors. The expert was
requested to use a scale from 0 to 10 in this section of the questionnaire in which a
value of “0” indicates the worst effect and “10” indicates the best effect on the
pipeline condition for each sub-factor characteristic. Table 4-3 shows a sample for
this part of the questionnaire. A full copy of the questionnaire can be found in
appendix A.
57
Table 4-2: Questionnaire Sample for relative importance used in Pairwise Comparison
for Main factors and Sub-Factors in gravity and pressurized pipelines
Degree Of Importance
(7) Very Strong
(7) Very Strong

(3) Moderate
(3) Moderate
(9) Absolute
(9) Absolute
Criterion Criterion
(5) Strong
(5) Strong
(1) Equal
(X) (Y)
1- Main Factors with respect to Sewer Pipeline Condition

Sewer Pipelines Condition
Environmental

Physical Factors
Factors Operational

Factors
2- Sub-Factors With Respect to Each Other
Physical Factors
 Pipeline Diameter
 Pipeline Length
Pipeline Buried

Depth
Pipeline Age  Pipeline Material
Pipeline Coating

Conditions
Installation

Quality
Operational Factors
Blockages (Ex.

Roots, Sediments)
Infiltration &
Corrosive 
Inflow
Impurities
 Flow Rate
Maintenance And

Break Record
Environmental Factors
 Soil Type
Bedding

Conditions
Location (Ex.
Groundwater 
Traffic Load)
Level
Ground
Disturbance (Ex.

Construction
Work)
3- Main Factors With Each Other
Physical Factors
Environmental Operational

Factors Factors
Environmental Factors
Physical Operational

Factors Factors
Operational Factors
Physical Environmental

Factors Factors
58
Table 4-3: Questionnaire Sample for effect values of different factors for gravity and
pressurized pipelines
Effect
Effect Value
Value On
Qualitative On Sewer
Main Unit Sewer
Sub-factors Description Pressurized
Factor Of Measure Gravity
(Parameters) Pipelines (0
Pipelines
– 10)
(0 – 10)
Old (>30) 8 9
Medium (15-
Pipeline Age (Years) 6 8
30)
New (<15) 4 6
Small (<300) 8 5
Physical
Pipeline Medium (300-

(m) 6 5
Diameter 600)
Large (>600) 4 6
:
Poor 10 10
Installation
(%) Fair 7 8
Quality
Good 4 5
Low 2 3
Flow Rate (m3/d) Medium 3 4
Operational
High 6 7
:
Poor 8 10
Maintenance And
(%) Fair 5 7
Break Strategies
Good 1 3
Rock 3 3
Soil Type
Environmental
Sand 5 5
:
Low 3 3
Ground
(%) Moderate 5 5
Disturbance
High 8 8
59
4.3.3 Responses
Forty questionnaires were received out of the sixty that were distributed. Four
questionnaires where further eliminated because they were considered as outliers and
36 responses were considered in developing the condition assessment model. The
considered questionnaires’ responses showed coherent and minimum variation values
in which all the respondents agreed that the included factors in the questionnaires
covered all possible aspects that might influence the condition of gravity and
pressurized sewer pipelines.
4.4 Validation Set
In order to validate the proposed model a dataset collected from the Drainage
Networks Operations and Maintenance Department in Ashghal Public Work
Authority, Doha, Qatar was used. The validation set included actual conditions for
549 gravity pipelines with 6 available factors which are: age, diameter, length, and
buried depth, pipeline position relative to groundwater and pipeline material. The
actual condition was based on a CCTV analysis following a condition code EN13508
(British Standards Institution (BSI), 2012), Class model EUROdss and class method
DWA-M 149-3 (German Association for Water, Wastewater and Waste (DWA),
2015).
60
CHAPTER 5: MODELS DEVELOPMENT AND
IMPLEMENTATION
5.1 Factors’ Weight (Wi) Determination
The factor’s Weight (Wi) for both approaches were calculated as illustrated
below.
5.1.1 Fuzzy Analytic Network Process (FANP)
Analytical network process (ANP) is considered as a multi-criteria decision
analysis technique that takes into consideration interdependencies between decision
alternatives. However, ANP neglects uncertainties of human judgment when
evaluating the pairwise comparison between the different factors. Therefore FANP
was used to account for interdependency between different factors and the
uncertainties and vagueness accompanied by human judgments. ANP was used to
model a three level network representing all contributing factors and sub-factors to
determine how strongly each of them affect sewer pipeline conditions. Fuzzy
Preference Programming method was used to conduct FANP (Zhou, 2012). Relative
weights are determined as a solution for a nonlinear maximization problem where, the
constraints are the upper and lower fuzzy numbers and the global weights are the
objective of the problem.
In the conventional ANP models, pairwise comparisons are performed on the
element and cluster levels. Relative weights are determined from pairwise
comparison, then are put into a matrix which is called super-matrix. The super-matrix
represents the interrelationships of elements on different levels. Table 4-2 shows a
61
general arrangement of a super-matrix in a conventional ANP technique, Where CN
represents the Nth cluster, and ENn represents the nth element in the Nth cluster
(Piantanakulchai, 2005). Wij sub-matrix consists of the collection of the priority
weight vectors (w) of the elements in the ith cluster with respect to the jth cluster. The
weights obtained from the pairwise comparison on the cluster level forms an
eigenvector, with a summation of unity. In order to obtain global priority vector, the
weighted super-matrix is raised to a limiting power as per Equation (25).
𝑙𝑙𝑙𝑙𝑙𝑙 𝐸𝐸 𝑘𝑘 (25)
𝑥𝑥→∞
Table 5-1: General Arrangement of Analytical Network Process (ANP) Super-matrix

C1 C2 CN
---
E11 E12 --- E1n E21 E22 --- E2n EN1 EN2 --- ENn
E11
E12
C1 W11 W12 --- W1N
---
E1n
E21
E22
C2 W21 W22 --- W2N
---
E2n
---
---
---
---
---
EN1
EN2
CN WN1 WN2 --- WNN
---
ENn
In this study, fuzzy preference programming (FPP) method was used to determine
the consistency ratios of fuzzy pairwise comparison matrices and local weights by
formulating a nonlinear prioritization problem (Mikhailov, 2004).
62
In the fuzzy preference programming method, the objective is to maximize the
consistency ratio which is a function in weights. The formulated non-linear
maximization problem is shown in Equation 26 (Zhou, 2012).
Max λ Such that,

(𝑚𝑚𝑖𝑖𝑖𝑖 − 𝑙𝑙𝑖𝑖𝑖𝑖 )𝜆𝜆𝜔𝜔𝑖𝑖𝑖𝑖 − 𝜔𝜔𝑖𝑖 + 𝑙𝑙𝑖𝑖𝑖𝑖 𝜔𝜔𝑗𝑗 ≤ 0
(𝑢𝑢𝑖𝑖𝑖𝑖 − 𝑚𝑚𝑖𝑖𝑖𝑖 )𝜆𝜆𝜔𝜔𝑖𝑖𝑖𝑖 + 𝜔𝜔𝑖𝑖 − 𝑢𝑢𝑖𝑖𝑖𝑖 𝜔𝜔𝑗𝑗 ≤ 0 (26)
∑𝑛𝑛𝑘𝑘=1 𝜔𝜔𝑘𝑘 = 1, 𝜔𝜔𝑘𝑘 >0 ,
k = 1,2,3,…,n , i = 1,2,3,..,n-1 and j = 2,3,…,n, j > i
Where, 𝑙𝑙𝑖𝑖𝑖𝑖 , 𝑚𝑚𝑖𝑖𝑖𝑖 , 𝑢𝑢𝑖𝑖𝑖𝑖 are lower, middle and upper bounds of the triangular fuzzy
number used in pairwise comparison and 𝜔𝜔𝑘𝑘 is priority crisp vector in which relative
weights are present.
During the process of collecting data for the relative importance of the different
factors, experts specify their preferences in a linguistic way. The fuzzy linguistic
variable should reflect different aspects of human language (Zhou, 2012). In this
study a scale consisting of five terms which accounts for fuzziness was chosen. The
scale adds and subtracts “0.5” from every response of the pairwise comparison to
construct the upper and lower matrices. Table 5-2 shows a sample for the developed
matrices using the adjusted scale in a gravity pipeline. Non-Linear FPP Solver was
developed based on the Optimization Toolbox of MATLAB, where solutions for
“crisp priorities’ weights” were derived without requiring the conventional
defuzzification procedure performed in the ordinary FANP technique.
Figure 5-1 shows the steps for the conduction of the FANP process. The Figure
shows the different steps for constructing un-weighted and weighted super-matrices
and the limit matrix required for getting the final weights. Table 5-3 shows the
constructed three matrices for the different affecting sub-factors.
63
Figure 5-1: Steps for Conducting FANP
64
Table 5-2: Sample of Pairwise Comparison Matrices in Gravity Pipelines
Lower Limit Matrix* Most Probable Matrix* Upper Limit Matrix*
Ground Disturbance
Ground Disturbance
Ground Disturbance
Bedding Conditions
Bedding Conditions
Bedding Conditions
Groundwater level
Groundwater level
Groundwater level
Soil Type
Soil Type
Soil Type
Location
Location
Location
Factors
Groundwater level 1 4 1/2 4 1/2 1/7.5 1/7.5 1 5 5 1/7 1/7 1 1/2 5 1/2 5 1/2 1/6.5 1/6.5
Soil Type 1/5.5 1 1 1/9 1/9 1/5 1 1 1/9 1/9 1/4.5 1 1/2 1 1/2 1/8.5 1/8.5
Bedding Conditions 1/5.5 1 1 1/9 1/9 1/5 1 1 1/9 1/9 1/4.5 1 1/2 1 1/2 1/8.5 1/8.5
Location 6 1/2 8 1/2 8 1/2 1 1 7 9 9 1 1 7 1/2 9 9 1 1/2 1 1/2
Ground Disturbance 6 1/2 8 1/2 1/9 1 1 7 9 1/9 1 1 7 1/2 9 1/8.5 1 1/2 1 1/2
*Lower, Most probable and upper limit matrices values are as per the adjusted Triangular Fuzzy Number (TFN) matrix:
1
⎡ (1 ,1 ,1 ) ⎤
⎢ 2 ⎥
1
⎢(2 ,3 ,3 1)⎥
⎢ 2 2 ⎥
⎢ 1 1 ⎥
⎢ (4 2 , 5 ,5 2) ⎥
⎢ 1 1 ⎥
⎢�6 , 7, 7 �⎥
⎢ 2 2 ⎥
⎢ 1 ⎥
⎣ (8 2 , 9 , 9 ) ⎦
65
Table 5-3: Un-weighted Super-matrix, Weighted Super-matrix and Limit Super-
matrix for different affecting factors
Un-weighted Weighted Limit
Fac. Super-matrix Super-matrix Super-matrix
NC PF … GD NC PF … GD NC PF … GD
NC 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.000 0.000 … 0.000
PF 0.333 0.000 … 0.000 0.333 0.000 … 0.000 0.000 0.000 … 0.000
OF 0.333 0.167 … 0.000 0.333 0.083 … 0.000 0.000 0.000 … 0.000
EF 0.333 0.833 … 0.000 0.333 0.417 … 0.000 0.000 0.000 … 0.000
AG 0.000 0.037 … 0.000 0.000 0.018 … 0.000 0.008 0.020 … 0.000
DI 0.000 0.037 … 0.000 0.000 0.018 … 0.000 0.008 0.020 … 0.000
LE 0.000 0.037 … 0.000 0.000 0.018 … 0.000 0.008 0.020 … 0.000
D 0.000 0.223 … 0.000 0.000 0.111 … 0.000 0.051 0.120 … 0.000
MT 0.000 0.223 … 0.000 0.000 0.111 … 0.000 0.051 0.120 … 0.000
CC 0.000 0.223 … 0.000 0.000 0.111 … 0.000 0.051 0.120 … 0.000
IQ 0.000 0.223 … 0.000 0.000 0.111 … 0.000 0.051 0.120 … 0.000
FR 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.090 0.042 … 0.000
B 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.024 0.011 … 0.000
II 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.024 0.011 … 0.000
CI 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.022 0.010 … 0.000
MS 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.197 0.092 … 0.000
ST 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.036 0.026 … 0.000
BC 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.019 0.013 … 0.000
SR 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.019 0.013 … 0.000
LO 0.000 0.000 … 0.000 0.000 0.000 … 0.000 0.169 0.121 … 0.000
GD 0.000 0.000 … 1.000 0.000 0.000 … 1.000 0.169 0.121 … 1.000
66
5.1.2 Monte-Carlo Simulation
Based on the collected 36 questionnaires’ (i.e. responses), there were different
weights for each factor, these weights were fed into Monte-Carlo simulation to
develop probability distribution curves. Monte-Carlo simulation has the ability to
randomly select values in a certain distribution translating it into another distribution
based on the most frequently occurred values. Monte-Carlo simulation computes the
most probable weight based on the repeated random sample collection and statistical
analysis (Raychaudhuri, 2008). The simulation procedure involves two operations,
which are: “sampling” and “running iterations” (Salem, et.al, 2003). In the sampling
operation, the input parameters values are obtained randomly based on the
probabilistic distributions. In the running iterations, results from the model are
calculated based on the input parameters. In each iteration, one sample is drawn from
each input probability distribution. When last iteration is reached, the single-valued
output results are aggregated to produce one output distribution. Monte-Carlo
simulation result in an output distribution that represents the most probable value for
the factors’ weights’ based on the input parameters eliminating the uncertainties due
to the different values for these weights.
The statistical data of the resulted probability distributions of the final weights
corresponding to each individual factor is summarized in Table 5-4 for sewer gravity
and pressurized pipelines. To test goodness of actual frequencies from sampled data
and frequencies from theoretical distributions, Chi-Squared (Ch-Sq), Anderson-
Darling (A-D), and Kolmogorov-Smirnov (K-S) tests were the selected statistical tests
from the several available tests.
67
Table 5-4: Summary of Statistical Analysis Results for Factor Weights
A-D Test K-S Test Chi-Sq Test
Main Mean Final
Network Type Sub-Factor Distribution
Factor Weight (µ) Test Test
P-Value P-Value Test Value P-Value
Value Value
AG Lognormal 0.035 0.364 0.327 0.107 0.259 8.722 0.033
…
PF
IQ Logistic 0.075 0.778 0.022 0.130 0.054 10.667 0.031
FR Lognormal 0.048 0.519 0.107 0.119 0.144 2.889 0.409
…
…
Gravity OF
MS Weibull 0.075 0.885 0.052 0.147 0.083 7.167 0.067
ST Lognormal 0.039 0.197 0.872 0.091 0.601 1.722 0.632
…
…
EF
GD Logistic 0.107 0.482 0.172 0.130 0.056 8.333 0.080
AG Lognormal 0.040 0.309 0.460 0.123 0.103 6.389 0.094
…
…
PF
IQ Normal 0.078 0.189 0.896 0.080 0.835 1.722 0.787
FR Lognormal 0.038 0.206 0.788 0.080 0.759 2.889 0.409
…
…
Pressure OF
MS Weibull 0.067 0.493 0.389 0.088 0.872 4.444 0.217
ST Lognormal 0.032 0.198 0.831 0.074 0.840 6.389 0.094
…
…
EF
GD Normal 0.088 0.346 0.475 0.091 0.670 1.333 0.856
68
The best probability fit was chosen based on the maximum P-value of the three
tests. The P-value indicates that for a null hypothesis (i.e.: no difference between the
actual and theoretical distributions), the observed difference is equal to the P-value
due to random sampling error. This means that for P-values approaching 0, the
corresponding distribution best represents the resulting distribution. The mean value
for the final weights probability distribution for gravity and pressurized pipelines are
shown in Figure 5-2.
Figure 5-2: Factors’ final weights calculated using FANP and Monte-Carlo simulation
for Gravity and Pressurized Pipelines
5.2 Effect Value
5.2.1 First Approach (Condition Curves)
The same technique described in Section 5.1.2 was applied for the different effect
values of each factor as shown in Table 5-5 for sewer gravity pipelines and Table 5-6
for sewer pressurized pipelines.
69
Table 5-5: Summary of Statistical Analysis Results for Factor Effect Values in Gravity Pipelines
Mean A-D Test K-S Test Chi-Sq Test
Main
Fac. Characteristic Distr. Effect
Fac. Test Test Test
Value (µ) P-Value P- Value P- Value
Value Value Value
Old >30 Max. extr. 2.780 0.565 0.147 0.172 0.024 3.897 0.273
AG Medium (15-30) Gamma 5.530 1.117 0.022 0.227 0.000 10.93 0.004
New <15 Logistic 8.590 1.535 0.000 0.223 0.000 25.00 0.000
…
…
PF
Poor <30% Max. extr. 1.980 0.736 0.052 0.148 0.102 3.069 0.381
IQ Fair (30-70)% Normal 5.190 0.787 0.035 0.181 0.017 9.690 0.021
Good >70% Weibull 8.520 0.976 0.122 0.159 0.182 12.58 0.002
Low <30% Logistic 4.700 0.466 0.191 0.155 0.025 1.414 0.702
FR Medium (30-70)% Weibull 8.120 1.800 0.036 0.265 0.000 27.07 0.000
High >70% Weibull 4.900 0.597 0.241 0.133 0.312 3.90 0.143
…
…
OF
Poor <30% Max. extr. 2.760 0.516 0.194 0.124 0.309 2.241 0.524
MS Fair (30-70)% Weibull 5.740 0.877 0.071 0.174 0.087 12.17 0.002
Good >70% BetaPERT 8.660 1.264 --- 0.210 --- 19.62 0.000
Rock <50% Uniform 5.750 0.648 0.474 0.148 0.444 3.483 0.323
ST
Sand (50-100)% Normal 6.280 0.429 0.305 0.142 0.145 1.414 0.702
…
…
EF
Low (0-30)% Weibull 8.160 0.681 0.149 0.148 0.155 11.76 0.003
GD Medium (30-70)% Max. extr. 5.560 1.366 0.000 0.223 0.000 17.14 0.001
High (70-100)% Normal 2.620 0.749 0.045 0.176 0.023 25.00 0.000
70
Table 5-6: Summary of Statistical Analysis Results for Factor Effect Values in Pressurized Pipelines
Mean A-D Test K-S Test Chi-Sq Test
Main
Fac. Characteristic Distr. Effect
Fac. Test Test Test
Value (µ) P-Value P- Value P- Value
Value Value Value
Old >30 Gamma 2.300 0.517 0.282 0.131 0.365 2.000 0.368
AG Medium (15-30) Weibull 5.230 1.007 0.059 0.219 0.029 14.00 0.001
New <15 Weibull 8.560 2.184 0.000 0.243 0.000 34.40 0.000
…
…
PF
Poor <30% Normal 1.600 0.790 0.035 0.142 0.128 6.400 0.094
IQ Fair (30-70)% Weibull 5.060 1.652 0.031 0.232 0.025 24.80 0.000
Good >70% Uniform 8.500 1.860 0.062 0.243 0.033 34.80 0.000
Low <30% Uniform 4.750 1.009 0.244 0.186 0.179 6.800 0.079
FR Medium (30-70)% Weibull 8.280 1.296 0.045 0.228 0.029 18.00 0.000
High >70% Min extr. 4.930 0.862 0.025 0.148 0.087 3.200 0.362
…
…
OF
Poor <30% Normal 2.130 0.639 0.089 0.136 0.178 14.00 0.003
MS Fair (30-70)% Normal 5.420 0.898 0.019 0.194 0.000 12.40 0.006
Good >70% Weibull 8.670 1.758 0.000 0.230 0.000 28.40 0.000
Rock <50% Max extr. 5.890 0.783 0.039 0.154 0.061 2.400 0.494
ST
Sand (50-100)% Uniform 6.000 0.712 0.425 0.183 0.196 18.40 0.000
…
…
EF
Low (0-30)% Weibull 8.500 0.830 0.160 0.171 0.123 10.80 0.005
GD Medium (30-70)% Max extr. 5.710 1.160 0.000 0.168 0.027 9.200 0.027
High (70-100)% BetaPERT 2.500 1.281 0.000 0.210 0.000 24.00 0.000
71
The effect value of the factors on the pipeline’s condition generated from Monte-
Carlo simulation does not vary with age. Based on experts’ opinion, the impact of the
same factor should vary with age. For example, smaller diameter pipelines deteriorate
faster than bigger diameter pipelines, but the effect of the diameter factor on the
pipeline's condition differs significantly with time. As a result and to account for the
change of the effect value during the lifetime of pipelines, age dependent condition
curves were developed. The condition curves were generated using the mean values
resulting from the developed probability distributions for the different effect values as
the base points for these curves. Equation 27 shows the formula by which the curves
were plotted.
CI0 − �
X
CIt = CI0 − � � ∗ CIt+1 (27)
T
Where,
𝐶𝐶𝐶𝐶0 : is the initial condition of the pipe at time = 0, which is 10,
𝐶𝐶𝐶𝐶𝑡𝑡 : is the condition index at time (t),
T is the overall time interval on which the curves are generated,
𝑋𝑋�: is the mean effect value resulting from probability distributions and
𝐶𝐶𝐶𝐶𝑡𝑡+1: is the condition index at the successive time step.
An example of the generated curves for gravity pipeline diameter, length, buried
depth and groundwater level condition index over different time periods is shown in
Figure 5-3. The curves represent the condition index of the pipeline on the vertical
axis, while the age is represented on the horizontal axis.
72
Figure 5-3: Pipeline Condition Curves for Different Sub-Factors over Age
5.2.2 Second Approach (Fuzzy Set Theory)
Fuzzy Set Theory (FST) is a mathematical model that was first introduced by
Zadeh (1965), designed to generalize the concept of classical (or crisp) sets. FST
attempts to provide a better tools to deal with vague situations that cannot be captured
by the classical Set theory. The membership of a classical set can be considered as a
function with only two possible values (i.e.: an element either belongs to certain set of
elements or does not). The generalization made by Zadeh (1965) is to produce Fuzzy
Sets that allows the membership function to have a gradual transition with any degree
of membership from none to full. The significance of fuzzy variables is that they
facilitate the gradual transition between states of a crisp variable and consequently,
possess the capability to express and deal with uncertainties, unlike crisp variables
73
that ignore. Fuzzy numbers “F” can be represented by a set [a1, a2, a3, a4]. Each fuzzy
number defined by a membership function μF, which can be expressed by Equation
28.
1 , when 𝑎𝑎2 <x <a3

𝑎𝑎 < 𝑥𝑥 < 𝑎𝑎2
µ𝐹𝐹 (𝑥𝑥) = �0 < 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 < 1 , when � 1 (28)
𝑎𝑎3 < 𝑥𝑥 < 𝑎𝑎4
0 , Otherwise
Membership functions (MFs) are the building blocks of FST, in which the
fuzziness in a fuzzy set is determined by its MF. Accordingly, the shapes of MFs are
important for each particular problem. MFs may have different shapes like triangular,
trapezoidal, Gaussian, etc. For triangular fuzzy numbers a2 would have the same value
of a3. The only condition that a MF must satisfy is that it must vary between 0 and 1.
In order to determine the inputs to be used in the ER module, the questionnaires’
responses were used to generate the linguistic factors’ fuzzy thresholds for the effect
values and their corresponding membership functions in similar manner to the FANP
algorithm. The methodology of implementing the FST starts with defining the
minimum average lower limit and maximum average upper limit for each linguistic
factor affecting the pipeline conditions based on the questionnaires’ responses. To
represent the output for the effect of the different linguistic factors, a five grade fuzzy
subset (Excellent, very good, good, fair and critical) is used. After determining the
input and output thresholds values, a suitable membership function is chosen to
represent the inputs and output. In this study, trapezoidal curves at extreme points and
triangular curves for the intermediate points were used (Figure 5-4 to Figure 5-7).
Triangular and trapezoidal fuzzy membership function shapes were used because they
are suitable for representing linguistic variables (Lee, 1996). Membership functions
have been divided into four zones 0 to 5 years, 5 to 15 years, 15 to 30 years and above
74
30 years in order to appropriately take into account the age effect on the developed
model. In such way, better and reliable assessment for new pipelines is assigned over
the older ones by taking into consideration the age influence on the pipelines
deterioration. Figure 5-4 to Figure 5-7 show the plot of the generated membership
functions for diameter, length and the buried depth for gravity pipelines and their
corresponding shapes for the fuzzy thresholds. Similarly, membership functions were
generated for the rest of the factors affecting gravity and pressurized pipelines. The
figures were developed by fuzzifying the input to determine the MF, for which each
factor belongs. The corresponding MF (µF (x)) based on the five grade scale is
calculated using Equation 28. Due to the uncertainty of the exact limits of the factors,
the thresholds are overlapping at some intervals. Also, the trapezoidal shapes
represent the extreme limits of excellent or critical membership functions for the
thresholds, whereas triangular shapes are used to represent the three remaining
membership functions in between the trapezoidal shapes.
Figure 5-4: Membership Function for Pipelines’ Age Effect Value
75
(1)
(2)
(3)
(4)
Figure 5-5: Membership Functions for Pipelines’ Diameter Effect Values
1) Age 0 to 5 years, 2) Age 5 to 15 years, 3) Age of 15 to 30 years and 4) Age 30 to

40 years
76
(1)
(2)
(3)
(4)
Figure 5-6: Membership Functions for Pipelines’ Length Effect Values

40 years
77
(1)
(2)
(3)
(4)
Figure 5-7: Membership Functions for Pipelines’ Buried Depth Effect Value

40 years
78
5.3 Overall Condition Assessment Index
Equation (29) represents the overall condition assessment index model derived by
simulation.
𝑘𝑘
𝑂𝑂𝑂𝑂𝑂𝑂𝑗𝑗 = � 𝑊𝑊𝑖𝑖 𝑥𝑥 𝐸𝐸𝐸𝐸𝑖𝑖 (29)

𝑖𝑖=1
Where,
OCIj is the overall condition index of sewer pipeline j,
𝐸𝐸𝐸𝐸𝑖𝑖 is the effect value of factor i reflecting the factor score,
𝑊𝑊𝑖𝑖 is the final weight for factor I,
k is the number of factors.
The model provides the overall condition index assessing sewer pipelines, where a
higher index indicates a better pipeline condition. The overall condition index and
effect values of each factor ranges between the extreme values of 0 and 10; which
shows that the pipeline is at its worst or best condition respectively.
5.3.1 First Approach (Monte-Carlo Simulation)
As shown in Equation (29), the model multiplies each pipeline’s effect value
obtained from the condition curves for each factor, by the probabilistic final weight of
the corresponding factor. The results of these multiplications are added to calculate
the mean overall condition index for each pipeline. The procedure is repeated for
1000 iterations (simulations) with stopping criterion parameters of 5% accuracy (ε =
0.05) and 99% confidence (α = 0.01). The sample variance based stopping rule used
in this simulation (Bayer et al., 2014) is presented below in Equation 30.
79
M
Set n = 0, Generate Mn samples (OCIj )I=1
n
and compute sample variance
(σ� 2Mn )
Mn
1
σ2Mn
� = ��j )2
( �(OCIjI − OCI
Mn − 1 Mn
I=1 (30)
�Mn ε
While 2(1- f ( )) >α do
� Mn
σ
Set n= n+1 and Mn = 2Mn−1
End while
Where,
I is the iteration number,
Mn is the number of samples,
M
(OCIj )I=1
n
is the overall condition for samples (Mn ) and iteration (I),
σ2Mn is the sample variance,

�
α is the confidence,
ε is the error,
�Mn ε
f( � Mn
σ
) is the distribution function.
This reflects the strength of Monte-Carlo simulation in which a random final
weight is chosen in each iteration based on the probability distribution determined for
each factor. This randomness guarantees that the uncertainty is taken into
consideration and the mean value of the OCIj obtained through the iterations is the
final condition value for every pipeline. Based on the calculated condition value, the
concerned authority can decide on the necessary actions to be taken for the pipeline,
80
which is the purpose of this study. A sample of a probabilistic condition index output
from Monte-Carlo simulation is shown in Figure 5-8.
Figure 5-8: Sample of Monte-Carlo Simulation Output for Overall Condition Index
Probability Distribution
5.3.2 Second Approach (Evidential Reasoning)
Evidential reasoning (ER) offers a rational methodology to deal with uncertainty,
incompleteness and fuzziness for data aggregation. This approach was developed on
the basis of decision theory and the Dempster-Shafer theory of evidence (Yang and
Singh, 1994; Yang and Sen, 1994; Yang, 2001) to address Multi-Criteria Decision
Making (MCDM) problems under uncertainty. ER methodology describes and
handles various types of uncertainties by using the concept of the degrees of belief, in
which each attribute of an alternative of a MCDM problem is described by a
distributed assessment using a belief structure. Unlike conventional approaches that
81
require scaling grades and averaging scores to aggregate attributes, the ER approach
employs an evidential reasoning algorithm to aggregate belief degrees. The ER
approach is a technique that allows aggregating many pieces of evidence (Yang and
Xu, 2002). It aggregates two factors at a time and the resulting aggregation of the first
two factors of evidence is aggregated with the third factor of evidence and so on.
The ER approach is used to determine the final condition index for sewer
pipelines. The first steps in implementing the ER module is to identify the distinctive
evaluation grades (H) that are represented by the linguistic variables (i.e. excellent,
very good, good, fair, and critical) and to define the final weight (ωi) for each
contributing factor. The belief structure of the formulated ER problem consists of
degree of beliefs indicating the user’s level of certainty about the condition of the
pipeline (Excellent, very good,…etc.) based on the different contributing factors
effect values. For instance the degree of belief is said to be high for an excellent
pipeline condition for a new pipe with a low effect values of the contributing factors,
while the degree of belief is considered low for very good pipeline conditions
assuming an old pipeline with higher effect values for the contributing factors. After
identifying the evaluation grades and relative weights, the degrees of belief are
transformed into basic probability masses (𝑚𝑚𝑛𝑛,𝑖𝑖 ) using Equation 31, by multiplying
the relative weights by the degrees of belief.
𝑚𝑚𝑛𝑛,𝑖𝑖 = 𝑚𝑚𝑖𝑖 (𝐻𝐻𝑛𝑛 ) = 𝜔𝜔𝑖𝑖 𝛽𝛽𝑛𝑛,𝑖𝑖 n = 1, … ,N; i = 1, ... , L (31)
Where,
𝑚𝑚𝑛𝑛,𝑖𝑖 represents the degree to which the ith basic attribute supports the hypothesis of
attribute (y) (pipeline condition) to be assessed to the (nth) grade (Hn ).
82
(mH,i) is the remaining probability masses unassigned to any individual after all (N)
grades have been considered for evaluating the general attribute. (mH,i) is calculated as
per equations 32, 33 and 34.
𝑚𝑚𝐻𝐻,𝑖𝑖 = 𝑚𝑚𝑖𝑖 (𝐻𝐻) = 1 − ∑𝑁𝑁 𝑁𝑁

𝑛𝑛=1 𝑚𝑚𝑛𝑛,𝑖𝑖 − 𝜔𝜔𝑖𝑖 ∑𝑛𝑛=1 𝛽𝛽𝑛𝑛,𝑖𝑖 , i = 1,2,...,L (32)
𝑚𝑚 � 𝑖𝑖 (𝐻𝐻) = 1 − 𝜔𝜔𝑖𝑖 ,
� 𝐻𝐻,𝑖𝑖 = 𝑚𝑚 i = 1,2,...,L (33)
𝑚𝑚 � 𝑖𝑖 (𝐻𝐻) = 𝜔𝜔𝑖𝑖 (1 − ∑𝑁𝑁

� 𝐻𝐻,𝑖𝑖 = 𝑚𝑚 𝑛𝑛=1 𝛽𝛽𝑖𝑖 ), i = 1,2,...,L (34)
Where,
𝑚𝑚𝐻𝐻,𝑖𝑖 = 𝑚𝑚 � 𝐻𝐻,𝑖𝑖 and ∑𝐿𝐿𝑖𝑖=1 𝜔𝜔𝑖𝑖 = 1,

� 𝐻𝐻,𝑖𝑖 + 𝑚𝑚
𝑚𝑚
� 𝐻𝐻,𝑖𝑖 is the remaining probability mass that has not been yet assigned to individual
grades. While, 𝑚𝑚
� 𝐻𝐻,𝑖𝑖 is the remaining probability mass unassigned to individual grades
caused by the incompleteness of the assessment.
𝑚𝑚𝑛𝑛,𝐼𝐼(𝑖𝑖) 𝑎𝑎𝑎𝑎𝑎𝑎 𝑚𝑚𝐻𝐻,𝐼𝐼(𝑖𝑖) can be calculated by aggregating the basic probability
masses 𝑚𝑚𝑛𝑛,𝑗𝑗 and 𝑚𝑚𝐻𝐻,𝑗𝑗 for n = 1, … , N and J = 1, …. , i using Equations 35 to 37 which
combines the two probability masses using normalizing factor 𝐾𝐾𝐼𝐼(𝑖𝑖+1) .
𝑁𝑁 −1
The normalizing factor can be defined as 𝐾𝐾𝐼𝐼(𝑖𝑖+1) = �1 − ∑𝑁𝑁
𝑡𝑡=1 ∑𝑗𝑗=1 𝑚𝑚𝑡𝑡,𝐼𝐼(𝑖𝑖) 𝑚𝑚𝑗𝑗,𝑖𝑖+1 �
Where, 𝐾𝐾𝐼𝐼(𝑖𝑖+1)is the normalizing factor so that the summation of 𝑚𝑚𝑛𝑛,𝐼𝐼(𝑖𝑖+1) +
𝑚𝑚𝐻𝐻,𝐼𝐼(𝑖𝑖+1) for n = 1, ..., N is 1.
𝑚𝑚 𝑛𝑛,𝐼𝐼(𝑖𝑖+1) = 𝐾𝐾𝐼𝐼(𝑖𝑖+1) (𝑚𝑚𝑛𝑛,𝐼𝐼(𝑖𝑖) 𝑚𝑚𝑛𝑛,𝑖𝑖+1 + 𝑚𝑚𝑛𝑛,𝐼𝐼(𝑖𝑖) 𝑚𝑚𝐻𝐻,𝑖𝑖+1 + 𝑚𝑚𝐻𝐻,𝐼𝐼(𝑖𝑖) 𝑚𝑚𝑛𝑛,𝑖𝑖+1 ),

(35)
n = 1,2 ...,N
Where, 𝑚𝑚𝐻𝐻,𝐼𝐼(𝑖𝑖) = 𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝑖𝑖) + 𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝑖𝑖)
𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝑖𝑖+1) = 𝐾𝐾𝐼𝐼(𝑖𝑖+1) �𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝑖𝑖) 𝑚𝑚
� 𝐻𝐻,𝑖𝑖+1 � (36)
𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝑖𝑖+1) = 𝐾𝐾𝐼𝐼(𝑖𝑖+1) �𝑚𝑚
� 𝐻𝐻,𝑖𝑖+1 + 𝑚𝑚
� 𝐻𝐻,𝑖𝑖+1 + 𝑚𝑚
� 𝐻𝐻,𝑖𝑖+1 � (37)
83
By combining each two probability masses until all the factors are combined, the
final probability masses can be converted into the final degrees of belief using
Equations 38 and 39.

𝑚𝑚𝑛𝑛
𝛽𝛽𝑛𝑛 = � 𝐻𝐻,𝐼𝐼(𝐿𝐿)
, n = 1,2 ...,N (38)
1−𝑚𝑚
𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝐿𝐿)
𝛽𝛽𝐻𝐻 = (39)
1 − 𝑚𝑚
� 𝐻𝐻,𝐼𝐼(𝐿𝐿)
Where,
𝛽𝛽𝑛𝑛 is the degrees of belief for the aggregated final assessment associated to the grades
𝐻𝐻𝑛𝑛 and 𝛽𝛽𝐻𝐻 represents the incompleteness of the overall assessment associated to H.
In order to determine the overall condition of the pipeline in the developed model,
deffuzificaion is carried for the aggregated final assessment resulting from the ER
module by utilizing the FST module. In the defuzzification process, the final degrees
of belief are deffuzified into a crisp values. The deffuzification process is basically
calculating the areas of the resulting figures for each MF, weighted average method
was used to convert the fuzzy membership functions’ overall condition into a crisp
value.
5.4 Model Validation
In order to validate the proposed model a dataset collected from the Drainage
Networks Operations and Maintenance Department in Ashghal Public Work
Authority, Doha, Qatar was used as stated under Section 4.3.3. The validation set
included actual conditions for 549 gravity pipelines with 6 available factors which
are: age, diameter, length, and buried depth, pipeline position relative to groundwater
and pipeline material. The pipeline material of the validation set obtained was
84
Vitrified Clay. Therefore, the pipeline material factor was disregarded and only the 5
remaining factors were considered in the validation process.
The condition varied from 1 to 5, where 1 indicates that the pipeline has minor
defects and no action is required, while 5 indicates a pipeline with very heavy defects
and an immediate action is required. The resulted predicted condition scale from the
developed models has a scale that varies from 0 to 10.
In order to apply a valid comparison between the actual condition of a pipeline and
the obtained condition from the models, two model calibrations were required. The
first calibration was the conversion of the 0 -10 scale for the predicted pipes’
condition to 1-5 scale for the actual pipes’ condition. The thresholds of the calibrated
scale are shown in Table 5-7. The conversion was basically done by anchoring the
maximum and minimum of the two scales (0 anchored to 1 and 10 anchored to 5) and
dividing the rest of the scale values into 5 classes.
Table 5-7: Conversion of Actual Condition Rating Scale to Model Prediction

Condition Rating Scale
Actual Condition Scale Resulted Prediction Condition Scale
5 0 to <3
4 3 to <5
3 5 to <7
2 7 to <9
1 9 to 10
85
The second calibration was converting the relative weights for the seventeen
factors into an equivalent five factors’ weights. Equation 40 was used to make this
conversion in which the final obtained weights of the five factors included in the
validation process were adjusted so that their summation would be equal to one.
𝑛𝑛 𝑛𝑛
𝑤𝑤𝑐𝑐𝑐𝑐𝑐𝑐 = 𝑤𝑤𝑖𝑖 + [� 𝑤𝑤𝑖𝑖 ∗ (1 − � 𝑤𝑤𝑖𝑖 )]

𝑖𝑖=1 𝑖𝑖=1 (40)
Such that ∑𝑛𝑛𝑖𝑖=1 𝑤𝑤𝑐𝑐𝑐𝑐𝑐𝑐 = 1
Where,
𝑤𝑤𝑐𝑐𝑐𝑐𝑐𝑐 is the calibrated factor’s weight;
𝑤𝑤𝑖𝑖 is the original weight of the factor in case that the seventeen factors are present; n
is the total number of factors in the problem (n=5 in the validation set).
Table10 includes a sample of the validation data set. For example, based on the
characteristics of pipeline No.2 shown in Table 5-8, the resulted mean value from the
developed model was 7.8 and accordingly the pipeline's predicted condition would be
2, while the actual CCTV condition gave a value of 1.
86
Table 5-8: Sample of Actual versus Predicted Model Condition Rating Values for
Pipelines in Validation Dataset
Actual CCTV Predicted Model
Pipeline (No.) Mean Value
Condition Condition
2 1 7.8 2
134 1 9.6 1
19 2 8.2 2
128 2 8.4 2
24 3 6.4 3
428 3 5.5 3
111 4 4.6 4
…
The developed model performance assessment was based on the mathematical
diagnostics recommended in literature (Zayed and Halpin 2005; Al-Barqawi and
Zayed 2006). In order to determine the model validity, Equations (41) and (42) were
used which show the average validity percentage (AVP) and the average invalidity
percentage (AIP), respectively. If the values of AIP are closer to 0 and AVP are closer
to 100% the model is valid and vice versa. Likewise, Mean Absolute Error (MAE)
and Root Mean Square Error (RMSE) were used to validate the model using
Equations (43) and (44), respectively. The model is considered sound if the values for
MAE and RMSE are close to 0 (Dikmen et al. 2005). Finally, the fitness function
equation (𝑓𝑓𝑖𝑖 ) indicating that the developed model is valid if the calculated value using
Equation (45) is close to 1000 and invalid when it is close to 0 (Dikmen et al. 2005).
87
n
Ei 100
AIP = �� 1 − � �� × (41)
Ci n
i=1
AVP = 100 − AIP (42)
∑ni=1|Ci − Ei |
MAE = (43)
n
∑ni=1(Ci − Ei )2
RMSE = � (44)
n
1000
fi = (45)
1 + MAE
Where,
AIP = Average Invalidity Percent;
AVP = Average Validity Percent;
𝑀𝑀𝑀𝑀𝑀𝑀 = Mean Absolute Error;
𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 = Root Mean Square Error;
𝑓𝑓𝑖𝑖 = fitness function;
Ei = predicted value;
Ci = actual value;
n = number of events.
Results obtained by the first approach were 15%, 85%, 0.12, 0.15, and 893 and by
the second approach were 14%, 86%, 0.16, 0.11, and 898 for AIP, AVP, MAE,
RMSE and 𝑓𝑓𝑖𝑖 , respectively which are considered plausible results. The “actual versus
predicted output plot” results for the developed model (Approach 1) are shown in
Figure 5-9. In Figure 5-9, the condition indices resulting from the model for all the
pipes from the validation set were grouped based on their categorical classes and
compared to the actual condition rating. Figure 5-9 shows that the results from the
88
developed model are very close to the actual ones indicating the soundness and
accuracy of the proposed model. A comparison between the Actual Pipeline
conditions and the predicted pipeline conditions for approaches 1 and 2 for each
individual pipe is further presented in Figure 5-10.
Moreover, this model shows similar accuracy to previously developed models that
used linear regression and back propagation neural networks techniques giving 85%
and 86%, respectively (Chughtai and Zayed, 2008 and Khan et al., 2010). However,
the enhancement in this model could be due to the fact that it neither required
extensive data to create the model nor made strong assumptions that would result in
higher condition rating values (Salman, 2010). On the other hand, a major portion of
research addressing condition assessment models, only studied and analyzed the
significance on pipelines’ state without getting an index to express their conditions
(Wirahadikusumah et al., 2001, Davies, 2001, Baur and Herz 2002, Younis and
Knight, 2010). In these previous researches, limited factors varying between 4 and 10,
such as pipe size, length, depth, material, type of waste, ground water level, street
category, soil type and infiltration and how they contribute to sewer pipeline
deterioration were studied, but little research studied the effect of factors such as
coating conditions, maintenance and break strategies while taking into consideration
the interdependencies between these contributing factors. Moreover, no research
addressed these factors and their effect on the pressurized sewer pipelines (i.e. rising
mains).
89
Figure 5-9: Actual Versus Predicted Overall Condition Index (1)
90
Figure 5-10: Actual versus Predicted Overall Condition Index (2)
91
CHAPTER 6: CONCLUSION
New condition assessment models for sewer networks developed by integrating
FANP and Monte-Carlo simulation as one approach and FANP combined with
Monte-Carlo simulation, FST and ER as another approach were presented in this
research. Seventeen factors grouped under physical, environmental and operational
categories for gravity pipelines in addition to the operating pressure for pressurized
pipelines were considered in the model. The developed models use a weighted scoring
system to determine a numerical value indicating the condition of pipelines based on
the effecting factors’ relative weights and effect values. The relative weights and
degree of influence of the different factors were elicited from a questionnaire that was
distributed to experts working in the field of infrastructures and sewage networks.
The considered factors were grouped under physical, environmental and
operational categories. The relative weight for each factor and category was
determined by the FANP and Monte-Carlo Simulation module that considered the
uncertainties and fuzziness associated with transforming the experts’ judgments into
numerical values. The sub-factors for the physical, environmental and operational
categories recorded importance weights varying within the range of 8%. The “Ground
Disturbance” factor was found to be the most influential factor followed by the
“Location” with a weight of 10.6% and 9.3% for pipelines under gravity and 8.8%
and 8.6% for pipelines under pressure, respectively. On the other hand, the least
affecting factor was the “Length” followed by “Diameter” with a weight of 2.2% and
2.5% for pipelines under gravity and 2.5% and 2.6% for pipelines under pressure.
92
For the first approach, Monte-Carlo simulation is used to determine the final
scores for the weights by probability distribution fitting, which helped in eliminating
the uncertainties accompanying the model’s outputs due to different weights.
For the second approach, FST module was used to create membership functions
and thresholds for the effect values of the different effecting factors in the form of
triangular and trapezoidal functions. The overall condition index was determined by
using the ER module with the aid of FST in which degrees of belief were set and
combined with different relative weights of the different factors. Fuzzy membership
functions’ were defuzzified by utilizing the FST to convert the fuzzy overall condition
into a crisp value.
The developed models were validated using actual inspection data for 549 existing
sewer gravity pipelines in Qatar. Results obtained by the first approach were 15%,
85%, 0.12, 0.15, and 893 and by the second approach were 14%, 86%, 0.16, 0.11, and
898 for AIP, AVP, MAE, RMSE and 𝑓𝑓𝑖𝑖 , respectively, which indicates that the
developed model would yield in sound and reliable results .
The proposed condition assessment model can provide key personal and decision
makers with a proper tool to plan their inspections instead of using conventional
inspection and assessment methods that are time consuming and costly, collect only
necessary data and provide cost effective rehabilitation and maintenance action.
The model presented in this paper can be improved by adding and considering
additional factors other than those mentioned and more case studies can be used to
expand data sets to validate and calibrate the model. In addition, increasing number of
questionnaires’ participants or using other techniques to determine relative weights of
contributing factors is recommended for future work.
93
Recommendations for future research:
1. Standardization of data acquisition tool for municipalities which should cover
all relevant physical, operational and environmental factors.
2. Extension of the sewer pipeline condition prediction methodology to other
sewer network structures such as manholes, pumping stations, etc.
3. Application of sewer condition prediction to other infrastructure networks.
4. Develop a tool integrated with GIS.
5. Develop a flexible tool that allows users to add or remove affecting factors or
adjust the factor’s weights.
94
REFERENCES
Achim, D., Ghotb, F., and McManus, K. J., (2007), “Prediction of water pipe asset
life using neural networks”, J. of Infra. Syst, 13(1), 26-30.
Agresti, A., (2002), “Categorical data analysis”, 2nd Ed., Wiley, Hoboken, New
Jersey.
Al-Barqawi, H., and Zayed, T., (2008), “Infrastructure Management: Integrated

AHP/ANN Model to Evaluate Municipal Water Mains’ Performance”, ASCE, Journal
of Infrastructure Systems, 10.1061/ (ASCE) 1076-0342.
Al-Barqawi, H., and Zayed, T., (2006), “Assessment Model of Water Main
Conditions”, Proc., Pipeline Division Specialty Conference, ASCE, Chicago, Illinois,
USA.
Allison, P., (1999), “Multiple regression: a primer”, Pine Forge Press, Thousand
Oaks, CA.
Ana E V, Bauwens W, (2010), "Modeling the structural deterioration of urban

drainage pipes: The state-of-the-art in statistical methods", Urban Water Journal, 7,
47-59.
Ana E V, (2009), "Sewer asset management - sewer structural deterioration modeling

and multi-criteria decision making in sewer rehabilitation projects prioritization", PhD
thesis, Department of Hydrology and Hydraulic Engineering, Vrije Universiteit
Brussel, Brussel.
Ana, E., Bauwens, W., Pessemier, M., Thoeye, C., Smolders, S., Boonen, I. and De
Gueldre, G., (2009), “An investigation of the factors influencing sewer
structural deterioration”, Urban Water Journal, 6(4), 303-312.
95
Ariaratnam S T, El-Assaly A, Yang Y, (2001), "Assessment of infrastructure
inspection needs using logistic models", Journal of Infrastructure Systems ,7 ,160-
165.
Bai, H., Sadiq, R., Najjaran,H., Rajani, B., (2008), ”Condition assessment of buried
pipes using hierarchical evidential reasoning model”, Journal of Computing in Civil
Engineering, 22(2), 114-122.
Baik H S, Jeong H S, Abraham D M, (2006), "Estimating transition probabilities in

Markov chain-based deterioration models for management of wastewater systems",
Journal of Water Resources Planning and Management, 132 15-24.
Baur R, Herz R, (2002), "Selective inspection planning with ageing forecast for sewer
types”, Water Science and Technology 46 (6-7) 389-396.
Baur R, Zielichowski-Haber W, Kropp I, (2004), "Statistical analysis of inspection

data for the asset management of sewer networks", Proceedings 19th EJSW on
Process Data and Integrated Urban Water Modeling, Lyon, France.
Bayer, C., Hoel, H., Von Schwerin, E., Tempone, R., (2014), “On non-asymptotic
optimal stopping criteria in Monte Carlo simulations”, SIAM J. Sci. Comput. , 36(2),
A869–A885.
British Standards Institution (BSI), (2012)."Investigation and assessment of drain and

sewer systems outside buildings-General Requirements". BSI, London. ISBN: 978-0-
580-70190-0.
Butler, D & Davis, J, (2000), “Urban Drainage”, Taylor and Francis Group, New
York, USA
Chae, M.J., Iseley, T., and Abraham, D.M., (2003), “Computerized sewer pipe
condition assessment”, In: Proceedings International Conference on Pipeline
Engineering and Construction, Baltimore, MD.
96
Chughtai, F., and Zayed, T., (2007a), “Sewer pipeline operational condition
prediction using multiple regression”, Proc., Pipelines 2007: Advances and
Experiences with Trenchless Pipeline Projects., ASCE, Reston, VA.
Chughtai, F., and Zayed, T., (2007b), “Structural condition models for sewer
pipeline”, Proc., Pipelines 2007: Advances and Experiences with Trenchless Pipeline
Projects., ASCE, Reston, VA.
Chughtai F, Zayed T., (2008), "Infrastructure condition prediction models for

sustainable sewer pipelines", Journal of Performance of Constructed Facilities ,22,
333-341.
Daher,S., (2015), “Defect-based Condition Assessment Model and Protocol of Sewer

Pipelines”, Msc., Concordia University, Montreal, Canada.
Davies J P, Clarke B A, Whiter J T, Cunningham R J, (2001), "Factors influencing the

structural deterioration and collapse of rigid sewer pipes", Urban Water ,3, 73-89.
Davies, J., P., (2001), “Application of interaction matrices to the problem of sewer
collapse”, Proceedings of the 11th European sewage and refuse symposium, liquid
waste section., Munich, Germany.
Davies, J., P., Clarke, B., A., Whiter, J., T., Cunningham, R.J., and Leidi, A., (2001),
“The structural condition of rigid sewer pipes: a statistical investigation”, Urban
Water Journal, 3, 277-286.
Dikmen, I., Birgonul, M., and Kiziltas, S., (2005), “Prediction of Organizational
Effectiveness in Construction Companies”, Journal of Construction Engineering and
Management, 131(2), 252–261.
Dirksen J, Clemens F H L R, (2008), "Probabilistic modeling of sewer deterioration

using inspection data", Water Science and Technology, 57 (10), 1635-41.
Durkin, J., (1994), “Expert systems: design and development”, Prentice Hall PTR.
97
Elmasry, M., Zayed, T., and Hawari, A., (2016), “Sewer Inspection Prioritization
Using a Defect-Based Bayesian Belief Network Model”, ASCE Pipelines 2016, VA,
613-625.
El-Assaly, A., Ariaratnam, S.T., Ruwanpura, J., and Ng, H., (2006), “Cost forecast
model for sewer infrastructure”, Municipal Engineer, 159 (3), 155–160.
Fausset, L., (1994), “What is a Neural Net?”, Fundamentals of neural networks:

architectures, algorithms, and applications D. Fowley, B. Baker, and A. Dworkin,
eds., Prentice Hall, Englewood Cliffs, NJ., 3 - 4.
Fenner, R.A., (2000). “Approaches to sewer maintenance: a review”, Urban Water, 2,

343-356.
Fenner, R. A., Sweeting, L., and Marriott, M., (2000), “A new approach for directing
pro-active sewer maintenance”, Proceedings of the institution of civil engineers, water
and maritime Engineering Journal, 142(2), 67 – 78.
German Association for Water, Wastewater and Waste (DWA), (2015), "State
detection and assessment of drain and sewer systems outside buildings-Part 3:
Assessment by optical inspection", DWA, Germany. ISBN: 978-3-88721-224-7.
Hahn, M. A., Palmer, R. N., Merrill, M. S., and Lukas, A. B., (2002), “Expert System
for Prioritizing the Inspection of Sewers: Knowledge Base Formulation and
Evaluation”, J. Water Resources Planning and Mngt., 128(2), 121 – 129.
Hajmeer, M., & Basheer, I., (2002), “A probabilistic neural network approach for
modeling and classification of bacterial growth/no-growth data”, Journal of
microbiological methods, 51(2), 217-226.
Hawari, A., Alkadour, F., Elmasry, M. and Zayed, T., (2016), “Simulation-Based
Condition Assessment Model for Sewer Pipelines”, Journal of Performance of
Constructed Facilities, 04016066.
98
Herz R., (1996), "Aging Processes and Rehabilitation needs of Drinking Water
distribution networks", Journal of Water Resources Planning and Management, 45,
221-231.
Herz R., (1995), "Alterung und Erneuerung von Infrastruturbeständen - ein

Kohoertenüberlebensmodell", Jahrbuch für Regionalwissenschaft Jg., 14/15, 5-29.
Horold, S. and Baur, R., (1999), "Modeling sewer deterioration for selective
inspection planning – case study Dresden", Proceedings 13th EJSW on Service Life
Management Strategies of Water Mains and Sewers, 8–12 September, Switzerland.
Horold, S., (1998), "Forecasting rehabilitation needs: Evaluation of the AQUA

WertMin software for service life and total cost estimation", SINTEF Report,
Trondheim, Norway.
Inomata, T., Onogi, K., Nakata, Y., & Nishimura, Y, (1988), "A rule-based simulation
system for discrete event systems", Journal of chemical engineering of Japan, 21(5),
482-489.
Jansen, K., (2007), "AQUA-WertMin program manual", Klienbittersdorf, Germany:

Expert Office for Sewer Rehabilitation.
Khan Z, Zayed T, Moselhi O, (2010), "Structural condition assessment of sewer

pipelines", Journal of Performance of Constructed Facilities, 24, 170-179.
Kleiner Y, Rajani B, and Wang S, (2007), "Consideration of static and dynamic

effects to plan water main renewal", Proceedings Middle East Water 2007, 4th
International Exhibition and Conference for Water Technology, Manama, Bahrain.
Kleiner, Y., Sadiq, R., and Rajani, B., (2004), “Modeling failure risk in buried pipes
using fuzzy markov deterioration process”, Proc., Pipeline Engineering and
Construction: What’s on the Horizon?, ASCE, Reston,VA., 1-12.
99
Kleiner. Y., (2001), “Scheduling inspection and renewal of large infrastructure
assets”, J. Infrastruc. Syst., 7(4), 136 – 143.
König A, (2005), "WP2 External Corrosion Model Description", Computer Aided

REhabilitation of Sewer networks (Care-S), SINTEF Technology and Society,
Trondheim, Norway.
Lawless, J.F., (1982), "Statistical models and methods for lifetime data", New York:
John Wiley and Sons.
Le Gat Y, (2008), "Modeling the deterioration process of drainage pipelines", Urban

Water Journal, 5, 97-106.
Lee, H., (1996), “Applying fuzzy set theory to evaluate the rate of aggregative risk in
software development, Fuzzy Sets and Systems”, 79, 323-336.
Mamdani, E. H., and Assilian, S., (1975), “An experiment in linguistic synthesis with
a fuzzy logic controller”, International Journal of Man-Machine Studies, 7, 1 – 13.
Marlow D, Davis P, Trans D, Beale D, Burn S, (2009), "Remaining Asset Life: A

State of the Art Review", Final Report, Water Environment Research Fundation
(WERF), Alexandria, VA, US
MATLAB and Statistics Toolbox Release 2012b, The MathWorks, Inc., Natick,
Massachusetts, United States.
Mehle J, O´Keefe, Wrase P, (2001), "An Examination of Methods for Condition

Rating of Sewer Pipelines", Center for development of technical leadership,
University of Minnesota.
Micevski T, Kuszera G, Coombes P, (2002), "Markov model for storm water pipe
deterioration", Journal of Infrastructure Systems, 8(2) , 49-56.
Mikhailov, L., (2004), “A fuzzy approach to deriving priorities from interval pairwise
comparison judgments”, European Journal of Operational Research, 159, 687-704.
100
Müller K, Dohmann M, (2002), “Entwicklung eines allgemein anwendbaren
Verfahrens zur selektiven Erstinspektion von Abwasserkanälen und
Anschlussleitungen”, Bundesministerium für Bildung und Forschung (bmb+f),
Aachen, Germany.
Müller, K., (2002), “Development of a generally applicable methodology for selective

inspection of sewer networks”, Part A: scientific studies. Final report of the Institute
of Urban Water Management ISA, RWTH Aachen University, Aachen, Germany.
Muller, K. and Fischer, B., (2007), “Objective condition assessment of sewer

systems”, In: Proceedings IWA 2nd Leading-Edge Conference in Asset Management,
Lisbon, Portugal.
NRC., (2003), “Infiltration/Inflow Control/Reduction for Wastewater Collection

Systems: A Best practice”, National Guide to Sustainable Municipal Infrastructure,
Issue No. 1.0, Federation of Canadian Municipalities, National Research Council of
Canada, Canada.
Najafi, M., and Kulandaivel, G., (2005), “Pipeline condition prediction using neural
network models”, Proc., Pipeline 2005: Optimizing Pipeline Design, Operations and
Maintenance in Today’s Economy, ASCE, Reston, VA., 767 – 781.
Piantanakulchai, M., (2005), “Analytic network process model for highway corridor
planning”, Proceedings of ISAHP 2005.
Pohar, M., Blas, M., & Turk, S., (2004), “Comparison of logistic regression and linear
discriminant analysis: a simulation study”, Metodoloski zvezki, 1(1), 143.
Rajani, B., and Kleiner, Y., (2001). “Comprehensive review of structural deterioration
of water mains: statistical models”, Urban water, 3(3), 131-150.
Rajani B, Kleiner Y, and Sadiq R, (2006), “Translation of Pipe Inspection Results

Into Condition Ratings Using The Fuzzy Synthetic Evaluation Technique”, Journal of
Water Supply Research and Technology-Aqua, 55, 11-24.
101
Ross, S. M., (2000), “Introduction to probability methods”, 7th Edition, Academic
Press, San Diego, California.
Raychaudhuri, S., (2008), “Introduction to Monte Carlo Simulation”, Proc,

Simulation Conference, IEEE, Austin, Texas, USA.
Ruwanpura, J., Ariaratnam, S., and El-Assaly, A., (2004), “Prediction models for
sewer infrastructure utilizing rule-based simulation”, Civ. Eng. Environ. Syst., 21(3),
169 – 185.
Ruwanpura, J., Ariaratnam, S.T., and El-Assaly,A., (2004), “A Rule-based simulation

models for sewer infrastructure”, Proceedings of the Construction Research
Congress,ASCE, Honolulu, Hawaii, USA.
Saaty, T. J, (1996), “Decision making in Complex Environments”, The Analytical

Hierarchy Process for decision Making with Dependence and Feedback. USA: RWS
Publications.
Sadiq, R., Kleiner, Y., and Rajani, B. B., (2004), “Fuzzy cognitive maps for decision
support to maintain water quality in aging water mains”, NRCC-47305, 4th Int. Conf.
on Decision-Making in Urban and Civil Engineering (DMUCE 4), Porto, Portugal, 1-
10.
Sage, G. H., (1990), “Power and ideology in American sport: A critical perspective”,
Human Kinetics Publishers.
Salem, O., AbouRizk, S., and Ariaratnam, S., (2003), “Risk-based life-cycle costing
of infrastructure rehabilitation and construction alternatives”, J. Infrastruct. Syst.,
9(1), 6-15.
Salman B, (2010), "Infrastructure Management and Deterioration Risk Assessment of

Wastewater Collection Systems", PhD thesis, University of Cincinnati.
102
Scheidegger A, Hug T, Rieckermann J, Maurer M, (2011), "Network condition
simulator for benchmarking sewer deterioration models", Water Research, 45 ,4983-
4994.
Schmidt T, (2009), “Modellierung von Kanalalterungsprozessen auf der Basis von

Zustandsdaten”, PhD thesis, Institut für Stadtbauwesen und Straßenbau, Technical
University Dresden, Dresden, Germany.
Sinha, S., and McKim, R. A., (2007), “Probabilistic based integrated pipeline
management system”, Tunneling and Underground Space Tech., 22(5-6), 543 – 552.
Specht, D. F., (1990), “Probabilistic neural networks”, Neural networks, 3(1), 109-
118.
Tran D H, Perera B J C, Ng A W M, (2007), "Neural network based prediction

models for structural deterioration of urban drainage pipes, Land, Water and
Environmental Management: Integrated Systems for Sustainability", Proceedings,
Christchurch pp 2264-2270.
Tran D, (2007), “Investigation of deterioration models for stormwater pipe systems”,

PhD thesis, School of Architectural, Civil and Mechanical Engineering, Victoria
University.
Vanier, D.J., (2001), “Why industry needs asset management tools”, Journal of
Computing in Civil Engineering, 15 (1), 35–43.
Vollersten J, König A, (2005), "WP2 Report D6: Model testing and evaluation,
Computer Aided REhabilitation of Sewer networks (Care-S)", SINTEF Technology
and Society, Trondheim, Norway.
Wirahadikusumah, R., Abraham, D.M., Iseley, T. and Prasanth, R.K., (1998),

“Assessment technologies for sewer system rehabilitation”, Automation in
Construction, 7, 259-270.
103
Wirahadikusumah R, Abraham D, Iseley T, (2001), "Challenging issues in modeling
deterioration of combined sewers", Journal of Infrastructure Systems, 7, 77-84.
WRc, (2001), Manual of Sewer Condition Classification - 4th Edition, Water

Research Centre, UK.
Yang, J., (2004), “Road crack condition performance modelling using recurrent
Markov chains and artificial neural networks”, Doctoral Dissertation, Department of
Civil and Environmental Engineering, College of Engineering, University of South
Florida.
Yang, J.B., (2001), “Rule and utility based evidential reasoning approach for multi-
attribute decision analysis under uncertainties”, European Journal of Operational
Research, 131, 31–61.
Yang, J.B., Sen, P., (1994), “A general multi-level evaluation process for hybrid
MADM with uncertainty”, IEEE Transactions on Systems, Man, and Cybernetics, 24
(10), 1458–1473.
Yang, J.B., Singh, M.G., (1994), “An evidential reasoning approach for multiple
attribute decision making with uncertainty”, IEEE Transactions on Systems, Man, and
Cybernetics, 24 (1), 1–18.
Yang, J.B., and Xu, D.L., (2002), “On the Evidential Reasoning Algorithm for
Multiple Attribute Decision Analysis under Uncertainty”, IEEE Transaction on
Systems, Man, and Cybernetics, Part A: Systems and Humans, Vol. 32, No. 3, pp.
289-304.
Yan, J. M., & Vairavamoorthy, K., (2003), “Prioritizing water mains rehabilitation
under uncertainty”, Advances in water supply management, 237-246.
Younis, R., and Knight, M.A., (2010), “Continuation ratio model for the performance
behavior of wastewater collection networks”, Tunneling and Underground Space
Technology, 25, 660-669.
104
Zadeh, L. A., (1965), “Fuzzy sets”. Information and Control”, 8, 338–353.
Zayed, T., and Halpin, D, (2005), “Deterministic Models for Assessing Prod. And
Cost Of Bored Piles”, Journal of Construction Management and Economics, 23(5),
531-543.
Zhou, Xiaoguang, (2012), “Fuzzy Analytical Network Process Implementation with

Matlab”, MATLAB - A Fundamental Tool for Scientific Computing and Engineering
Applications Volume 3, Prof. VasiliosKatsikis (Ed.), ISBN: 978-953-51-0752-1.
Zou, J. Han, Y, and So, S-S., (2008), “Overview of Artificial Neural Networks”,
Artificial Neural Networks: Methods and Applications, D. J. Livingstone, ed.,
Humana Press, Totowa, New Jersey.
105
APPENDIX: QUESTIONNAIRE
106
107
108
109
110
111
112

Condition Assessment Models For Sewer Pipelines

Uploaded by

Copyright:

Available Formats

Condition Assessment Models For Sewer Pipelines

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Condition Assessment Models For Sewer Pipelines

Uploaded by

Copyright:

Available Formats

QATAR UNIVERSITY

CONDITION ASSESSMENT MODELS FOR SEWER PIPELINES

FIRAS AMER ABDULRAZAK ALKADOUR

A Thesis Submitted to the Faculty of

the College of Engineering

for the Degree of

ABDULRAZAK ALKADOUR, FIRAS, AMER., Masters : June : 2017,

Masters of Science in Civil Engineering

Title: Condition Assessment Models for Sewer Pipelines

Supervisor of Thesis: Dr. Alaa Al-Hawari.

Underground pipeline system is a complex infrastructure system that has

significant impact on social, environmental and economic aspects. Sewer pipeline

networks are considered to be an extremely expensive asset. This study aims to

addition to the operating pressure for pressurized pipelines. Two different

provide effective rehabilitation of sewer networks.

List of Figures............................................................................................................................ vii

List of Tables ............................................................................................................................ viii

CHAPTER 1 - INTRODUCTION ..................................................................................................... 1

1.2 Inspection and Evaluation of Existing Sewers ........................................................ 3

1.3 Qatar Network Existing Assets ............................................................................... 5

1.4 Thesis Organization ..............................................................................................10

CHAPTER 2 – LITERATURE REVIEW ...........................................................................................11

2.1 Factors Affecting Sewer Pipelines Condition........................................................11

2.2 Deterioration Models ...........................................................................................13

2.2.1 Physical Models ............................................................................................... 15

2.2.2 Artificial Intelligence Models ........................................................................... 16

2.2.3 Statistical Models ............................................................................................ 27

2.3 Problem Statement ..............................................................................................48

2.4 Research Objectives .............................................................................................49

CHAPTER 3 – RESEARCH METHODOLOGY ................................................................................50

CHAPTER 4 – DATA COLLECTION ..............................................................................................53

4.2 Factors affecting sewer pipelines condition .........................................................54

4.3 Questionnaires .....................................................................................................56

4.3.1 Factor’s Weights (Wi) ...................................................................................... 56

4.3.2 Factor’s Effect Values ...................................................................................... 57

4.4 Validation Set .......................................................................................................60

CHAPTER 5 : MODELS DEVELOPMENT AND IMPLEMENTATION ..............................................61

5.1 Factors’ Weight (Wi) Determination ....................................................................61

5.1.1 Fuzzy Analytic Network Process (FANP) .......................................................... 61

5.1.2 Monte-Carlo Simulation .................................................................................. 67

5.2 Effect Value...........................................................................................................69

5.2.1 First Approach (Condition Curves) .................................................................. 69

5.2.2 Second Approach (Fuzzy Set Theory) .............................................................. 73

5.3 Overall Condition Assessment Index ....................................................................79

5.3.1 First Approach (Monte-Carlo Simulation) ....................................................... 79

5.3.2 Second Approach (Evidential Reasoning) ........................................................ 81

5.4 Model Validation ..................................................................................................84

CHAPTER 6 : CONCLUSION .......................................................................................................92

Figure 1-1: Sewer Management System .................................................................................... 2

constructive criticism, and encouragement in the preparation of the manuscript. This

National Research Fund (Member of Qatar Foundation).

Authority of Qatar (ASHGAL) for their support in the data collection.

family throughout his education.

performance of sewage pipelines is a function in reliability, and level of service by

desired level. Maintenance, rehabilitation and renewal comprises the intervention of

absence of information regarding the condition of pipelines, unforeseen failure can

amongst the rest of intervention measures.

In the past, a reactive approach was considered in sewer management as

addressed before their occurrence by implementing a proper asset management