Rams

DOC TOR A L T H E S I S
ISSN: 1402-1544 ISBN 978-91-7439-023-0

Lule University of Technology 2009
Ambika Prasad Patra Maintenance Decision Support Models for Railway Infrastructure using RAMS & LCC Analyses
Division of Operation and Maintenance Engineering
Maintenance Decision Support Models

for Railway Infrastructure using
RAMS & LCC Analyses
Ambika Prasad Patra
DOCTORAL THESIS
Maintenance Decision Support Models for

Railway Infrastructure using RAMS &
LCC Analyses
Ambika Prasad Patra
Lule University of Technology

Printed by Universitetstryckeriet, Lule 2009

ISSN: 1402-1544
ISBN 978-91-7439-023-0
Lule
www.ltu.se
Preface
The research work presented in this thesis has been carried out at Lule Railway Research
Center (Jrnvgstekniskt Centrum, JVTC) and has been sponsored by Swedish National Rail
Administration (Banverket) and ALSTOM Transport, France. I would like to thank Banverket
and ALSTOM Transport for providing the financial support during my research.
First of all, I would like to express my sincere gratitude to my supervisor, Professor Uday
Kumar for his invaluable guidance, suggestions, encouragement and support during the tenure
of my research.
I am particularly grateful to Dr. Pierre Dersin (ALSTOM Transport, France) for providing me
guidance, resources and support during my stay in Paris. I would like to thank Dr. Per-Olof
Larsson-Krik (Banverket) for his valuable suggestions and fruitful discussions; Dr. Ulla
Juntti and Mr. Arne Nissen (Banverket) for sharing their knowledge in railway and Dr. Peter
Sderholm (Banverket) for his contribution in my research work. I would also like to thank
Professor Per Anders Akersten for his valuable comments and discussion.
I am also grateful to all my colleagues at the Division of Operation and Maintenance
Engineering for their continuous help during the research. Special thanks are acknowledged to
Prof. Jan Lundberg, Dr. Aditya Parida, Dr. Rupesh Kumar, Mr. Rajiv Dandotiya and Mr.
Yuan Fuqing for their discussions.
My special thanks to all the Indian friends and families for their help throughout my stay in
Lule.
Finally, I would like to express my gratitude to my parents for their support and belief in me.
Ambika Prasad Patra

October, 2009
Lule, Sweden
iii
Abstract
Todays railway sector is imposing high demands for service quality on railway infrastructure
managers. Since railway infrastructure has a long asset life, it requires efficient maintenance
planning to perform effectively throughout its life cycle to meet these high demands. Traditionally
maintenance decisions for the railway infrastructure have been based on past experience and
expert estimations. The application of RAMS (Reliability, Availability, Maintainability and
Safety) analysis for railway infrastructure is limited. The focus of this thesis is to demonstrate the
applicability of RAMS analysis in effective maintenance planning. Within the scope of this
research, various case studies associated with Banverket (the Swedish National Rail
Administration and ALSTOM Transport have been carried out. The research presents approaches
and models for estimating RAMS targets based on the service quality requirements of the railway
infrastructure. The availability target of the infrastructure has been estimated by considering the
capacity and punctuality requirements of the infrastructure, whereas the safety goal of the track
has been estimated by calculating the probability of derailment by means of undetected rail breaks
and poor track quality. Effective estimation of the RAMS targets will help infrastructure managers
to predict the maintenance investment in the railway infrastructure needed over a period of time in
order to achieve the targets. Nevertheless, the availability target of the infrastructure can lead to
train delay. A model has been developed to achieve the availability target in both the scheduled
and the condition based maintenance regimes by choosing an effective maintenance interval and
detection probability respectively. This has been illustrated by a case study on track circuits.
Different maintenance strategies can help in achieving the RAMS targets. In order to determine
the cost-effective solution, LCC (life cycle cost) should be used. The maintenance strategy with
lowest LCC will be the cost effective maintenance strategy. This has been demonstrated by a case
study on a signalling system. Sensitivity analyses have been performed to calculate the maximum
cost effectiveness of the system for different maintenance parameters. LCC estimation for a
maintenance strategy should always consider the risks associated with the strategy. A fair degree
of uncertainty is also associated with LCC estimation due to the statistical characteristics of
RAMS parameters. An approach has been developed in this thesis to calculate the uncertainties
associated with LCC estimation. Petri-Net analyses, Monte Carlo simulations, Design of
Experiment have been used to develop models to achieve the objectives of this thesis. This thesis
discusses the applicability of RAMS and LCC analyses for railway infrastructure and
demonstrates models for effective infrastructure maintenance planning.
Keywords: Railway infrastructure,
Maintainability, Safety, Life cycle cost
Maintenance
planning,
Reliability,
Availability,
List of Appended Papers

Paper I:
Patra, A. P., Kumar, U. and Larsson-Krik, P-O. (2010). Availability target

of the railway infrastructure: an analysis. Accepted for publication in
Proceedings of Reliability and Maintainability Symposium (RAMS), San Jose,
USA, 25-28 January.
Paper II:
Patra, A. P., Kumar, U. and Larsson-Krik, P-O. (2009). Assessment and

improvement of railway track safety. Proceedings of 9th International Heavy
Haul Conference (IHHA), Shanghai, China, 22-24 June.
Paper III: Patra. A. P. and Kumar, U. (2009). Availability analysis of railway track
circuit. Accepted for publication in Proceedings of the Institution of Mechanical
Engineers, Part F: Journal of Rail and Rapid Transit.
Paper IV:
Patra, A. P., Dersin, P. and Kumar, U. (2009). Cost effective maintenance

policy: a case study. Accepted for publication in the International Journal of
Performability Engineering.
Paper V:
Patra, A. P., Sderholm, P. and Kumar, U. (2009). Uncertainty estimation in

railway track life-cycle cost: a case study from Swedish National Rail
Administration. Proceedings of the Institution of Mechanical Engineers, Part F:
Journal of Rail and Rapid Transit, 2009, 223 (F3), 285-293.
vii
Contents
Preface.......................................................................................................................................iii
Abstract ...................................................................................................................................... v
List of Appended Papers .......................................................................................................... vii
Contents..................................................................................................................................... ix
1 Introduction ............................................................................................................................. 1
1.1 Research Problem............................................................................................................. 3
1.2 Research Objectives ......................................................................................................... 4
1.3 Research Questions .......................................................................................................... 4
1.4 Scope and Delimitations of the Study .............................................................................. 5
2 Basic Concepts and Definitions .............................................................................................. 7
2.1 Reliability, Availability, Maintainability & Safety (RAMS)........................................... 8
2.1.1 RAMS Parameters..................................................................................................... 8
2.1.2 Factors Affecting RAMS .......................................................................................... 9
2.2 Infrastructure Maintenance Planning ............................................................................. 13
2.2.1 RAMS Analysis for Maintenance ........................................................................... 16
2.2.2 RAMS in the Operation and Maintenance Phase.................................................... 18
2.3 Maintenance Optimisation ............................................................................................. 21
2.3.1 Life Cycle Cost Analysis......................................................................................... 22
2.3.2 Maintenance Management ...................................................................................... 26
3 Research Methodology.......................................................................................................... 31
3.1 Research Approach ........................................................................................................ 32
3.2 Reliability and Validity .................................................................................................. 33
4 Data Collection and Analysis................................................................................................ 35
4.1 Data Collection............................................................................................................... 35
4.2 Data Analysis ................................................................................................................. 37
5 Summary of the Appended Papers ........................................................................................ 41
6 Discussion and Conclusions.................................................................................................. 43
6.1 Research Contributions .................................................................................................. 47
6.2 Scope of the Future Research......................................................................................... 48
References ................................................................................................................................ 49
ix
1 Introduction
The railway transportation system is one of the most commonly used modes of transport and
its importance and utility are very substantial for the society. With the advancement of
technology, the changing environment and increasing customer demands, railways are having
to upgrade their various operational activities constantly. A safe and reliable network with
sufficient capacity and availability is a prime requirement. Railway infrastructure plays its
part in achieving this requirement in its system life cycle. To fulfil this requirement in an
effective manner, one needs to examine the various phases of the life cycle, such as inception,
design, manufacturing, installation, operation and maintenance, and disposal. Once the
infrastructure is installed, it is very difficult to modify the initial design. Therefore, the
performance of the infrastructure depends largely on the maintenance and renewal decisions
taken during its life cycle. The design phase of the track needs to consider not only the cost,
but also aspects like Reliability, Availability, Maintainability, and Safety (RAMS) and Life
Cycle Cost (LCC), with respect to technological advancements and changes. After the
installation, during the operation and maintenance phase, LCC and RAMS are considered
when making effective maintenance decisions.
Each of the infrastructure components, with its varying life and degrading conditions, will
influence the quality and operability of the infrastructure. In order to maintain the quality of
the infrastructure at an accepted level, two aspects of the infrastructure quality need to be
considered, i.e. measurement of the infrastructure quality on a continuous basis and means to
achieve the required infrastructure quality when the quality falls below the accepted level.
The infrastructure quality is measured using various parameters, e.g. the service reliability,
the track utilisation and accessibility, the infrastructure safety and the infrastructure system
and cost effectiveness. High operation and maintenance costs act as a barrier to achieving a
favourable financial performance of railway operations. The infrastructure quality is
vulnerable to infrastructure system failures. With an increase in infrastructure requirements in
terms of axle load, gross tonnage, speed, etc., the infrastructure experiences more failures,
which require more maintenance. At the same time, the availability of the infrastructure to
perform the necessary maintenance decreases, due to the increased traffic. This requires more
budgetary and other resources. To optimise the maintenance activities in terms of costeffectiveness and RAMS, a systematic analysis approach is required.
Maintenance Decision Support Models for Railway Infrastructure using RAMS & LCC Analyses
In order to minimise failures for the railway systems, the effects of decisions should be
systematically evaluated. The infrastructure manager, which is responsible for the design,
construction, maintenance, renewal and upgrading of the infrastructure, has a clearly defined
role and is confronted with the increasing performance of its collaborative partners. Due to
increases in operation and maintenance costs, infrastructure managers are compelled to
optimise their budget, while reliability and availability have to be increased without
endangering the traffic safety. A systematic approach is needed for communication with the
infrastructure manager in order to guarantee the defined levels of performance. Since, in the
current scenario, most of the maintenance and renewal decisions are based on past experience
and expert estimations, a need for a systematic LCC approach arises. A life cycle costing
approach in combination with RAMS analysis will provide a way to optimise the maintenance
strategy, considering the short term budget requirements as well as the long term costs of
ownership. Cost-effective decision making based on LCC usually does not consider the risk
aspects. Therefore, when performing cost effective decision making based on LCC for the
track system, one needs to consider the uncertainties associated with LCC. The associated
uncertainties are the risk factors related to costs of the traffic disruptions/derailment and the
variable costs due to RAMS parameters. Although some studies have been undertaken in the
areas of RAMS and LCC separately (see e.g. Vatn, 2002; Swier, 2004; Zoeteman, 2006),
there is a need for an integrated study of RAMS and LCC for the railway sector for enhancing
the cost effectiveness of the railway system. When taking effective maintenance decisions
based on LCC analysis, it is important to identify the uncertainties associated with LCC in
order to support the decision taking process. The uncertainties associated with LCC can
broadly be attributed to uncertainties in estimating RAMS parameters and uncertainties
concerning the economic conditions of cost parameters over a long time horizon. In the
railway sector, most of the efforts to implement RAMS and LCC approaches have been stand
alone projects, and have not been integrated with the decision making process.
Under the increasing pressures to improve performance quickly, infrastructure managers are
being forced to focus on supplying short-term cost and/or performance improvements only.
Despite a substantial amount of research in recent years, many rail deterioration processes are
not well enough understood for infrastructure managers to be able to translate them into
unambiguous quantitative relationships between investment and maintenance decisions and
long term quality effects (Ferreira, 1997; Veit, 2003); and uncertainty in these relationships
might result in these effects not being sufficiently appreciated. Governments and shareholders
have a preference for short payback periods for investments and quick performance
improvements, which can seriously conflict with the nature of railways and optimal spending
patterns. The long life spans of components and their high installation costs mean that
decisions have a high degree of irreversibility. In addition, the consequences of low initial
quality and insufficient preventive maintenance, i.e. high cost levels and low system
reliability, often only come to light several years later. After reaching certain degradation
levels, backlogs in maintenance lead to progressive degradation and, hence, capital
destruction. Although the infrastructure manager should be the party capable of incorporating
such effects into the decision making, either implicitly or explicitly, there are many impeding
Introduction
factors. The long term view of designing and maintaining usually conflicts with
organisational and institutional boundaries, such as allocated budgets, standard operating
procedures, established relations with other actors, and external regulations. Most of these
boundaries have a long history, and decision makers usually consider only incremental
changes. The present research focuses on the application of RAMS and LCC methodologies
to develop a decision support system for a cost effective maintenance policy.
1.1 Research Problem

Infrastructure managers are facing increasing demands from traffic operators as well as
passengers to ensure a safe, reliable and comfortable railway service. To achieve these
objectives the quality of the infrastructure needs to be improved and maintained. The
maintenance activities of the railway infrastructure have certain maintenance goals which are
linked to the organisational goals and objectives and which help in achieving the overall
objectives of the infrastructure. Karlsson (2005) presented Banverkets (the Swedish National
Rail Administration) vision for maintenance activities based on overall goals for securing
safety, reliability, comfort and cost-effectiveness. Usually, the overall maintenance strategy
consists of various critical success factors that are necessary to achieve the overall goals for
maintenance. The critical success factors include the guidelines for the functions (reliability,
safety, and comfort) to be achieved; methods for establishing and measuring the relationship
between the operational reliability, the condition of the infrastructure and the maintenance
work carried out, and methods for measuring the cost effectiveness of maintenance
operations, etc. Therefore, there is a need to develop effective maintenance decision support
models which can achieve the overall goals of the infrastructure in a cost effective way.
The objectives of maintenance decision support models include determining when the
infrastructure needs to be maintained, what maintenance actions need to be carried out, how
the maintenance actions will be performed, which maintenance action will meet the
infrastructure objective, what maintenance action will secure the safety of the system, etc.
RAMS analysis will help in identifying different maintenance alternatives to be carried out on
the infrastructure. RAMS provides a performance assurance with which the system can
guarantee the achievement of the goals of the infrastructure. LCC analysis will help in
optimising the cost effectiveness of the maintenance actions derived from RAMS analysis.
Cost estimations through LCC help in foreseeing the cost implications of maintenance actions
over the whole service life of the infrastructure, not just in the short term. Literature studies
show that the maintenance decision support models in use for railway infrastructure do not
consider RAMS and LCC methodologies explicitly. The use of RAMS and LCC
methodologies in the maintenance decision models not only helps in enhancing the system
effectiveness of the infrastructure, but also makes the maintenance strategies cost effective.
1.2 Research Objectives

The purpose of the present research is to illustrate and demonstrate the applicability of RAMS
and LCC analysis in the decision making process governing the cost effective maintenance of
the railway infrastructure, taking the associated risks and uncertainties into consideration.
The research has the following objectives:
x
To study RAMS methodologies for the railway infrastructure and develop a

framework for estimating the RAMS targets.
To study the applicability of RAMS tools in track maintenance planning and to

develop models.
To develop cost effective maintenance models using RAMS and LCC and to discuss
the variation in cost.
1.3 Research Questions

A literature review, in addition to our own industrial experience, and discussions with
personnel within the railway infrastructure managers and the manufacturers gave rise to many
interesting research areas in the field of RAMS and LCC. On the basis of the stated interests
of the companies participating in this project, the following research questions were
formulated:
1. How can RAMS targets be estimated for the railway infrastructure with the specified
infrastructure capacity and safety requirements?
2. How can RAMS analysis be applied in railway infrastructure maintenance planning to
achieve the RAMS targets?
3. How is LCC analysis used in combination with RAMS analysis to realize a cost
effective maintenance policy?
4. How are the uncertainties associated with LCC analysis estimated?
Introduction
1.4 Scope and Delimitations of the Study

Rail infrastructure consists of various sub-systems, such as the track system, the signalling
and telecommunication system, and the power system. Each of these sub-systems contributes
to infrastructure the RAMS and LCC of the infrastructure. In this research we considered only
the track system and part of the signalling systems. Although RAMS and LCC activities are
applicable in all the phases of the system life cycle, this thesis considers the RAMS and LCC
methodologies only in the operation and maintenance phase.
The thesis presents different maintenance models for the railway infrastructure. However, the
models do not consider how the interactive effect of the different systems on each other
influences the maintenance policy. The models provide decision support for the infrastructure
manager when making decisions on the inspection and maintenance so as to achieve the
RAMS targets for the infrastructure.
2 Basic Concepts and Definitions

The optimisation of infrastructure constructions or infrastructure components regarding
technical and economic requirements is essential for railway companies to fit into the market
and to compete against other means of transport. Due to the long lifetime of the track and
track components, pre-installation technical and economic assessments are necessary to
optimise the track construction and obtain the return on investment (ROI) in a manageable
timeframe. LCC and RAMS techniques are two acknowledged methods for assisting the
optimisation process. In the past decade, RAMS and LCC analyses in the railway sector have
attracted much more attention than before which has been demonstrated by many research
reports and has led to the development of commercial applications. This increased interest in
RAMS and LCC resulted in a number of projects being initiated at the European level in the
field of railway engineering, but not specifically dealing with the track e.g. Cost, Reliability,
Maintenance, and Availability (CRMA, 1998); Maintainability Management in European Rail
Transport (REMAIN, 1998); IMPROVEd tools for RAILway capacity and access
management (IMPROVERAIL, 2003); Progress in Maintenance and Management of
Infrastructure (ProM@in, 2003); Light Rail Thematic Network (LibeRTiN, 2005); and
Innovative Modular Vehicle Concepts for an Integrated European Railway System
(MODTRAIN, 2007). The objective of CRMA (1998) was to develop LCC methodologies for
rolling stock and to identify the parameters required to calculate the LCC, whereas REMAIN
(1998) focused on condition monitoring and RAMS management for switches. The objective
of IMPROVERAIL (2003) was to improve the existing LCC calculating methods by
including costs due to vehicle infrastructure interaction and external costs, e.g. delay costs,
accident costs, environmental costs, etc. ProM@in (2003) provided a comprehensive
overview of RAMS and LCC analysis applied to railway infrastructure. The project provided
an overview of LCC based maintenance planning, RAMS based track inspection and RAMS
databases. LiberTiN (2005) discussed the use of LCC and RAMS principles in contracts. An
LCC working group in UNIFE (2001) provided guidelines for LCC for total railway systems.
It also developed the UNILIFE-UNIDATA LCC model for rolling stock. There are a few
ongoing projects specific to track infrastructure; e.g. Lasting Infrastructure Cost
Benchmarking (LICB, 2007), Urban Track (2007) and INNOTRACK (2009) deal with LCC.
There is also some related literature, e.g. Burstrom et al. (1994), Stalder (2001), Zoeteman
(2001), Esveld (2001), Zoeteman (2006) and Zhao (2006). The main focus of these
publications (project descriptions and other literature) is on developing LCC calculation
methodologies, as well as the use of LCC in maintenance planning.
However, these publications do not discuss the issues of LCC reduction and the integration of
LCC concepts with RAMS. The application of RAMS has not yet been explored fully from
the railway infrastructure perspective. In this chapter the basic concepts of RAMS and LCC
are described and the application of RAMS and LCC concepts to railway infrastructure
maintenance planning is addressed.
2.1 Reliability,
Safety (RAMS)
Availability,
Maintainability
&
Reliability and maintainability management is attracting new interest in todays corporate

world. The quest to remain competitive and provide timely and accurate services is partly
responsible for this interest. A company cannot adopt a rapid response strategy if its system is
unavailable and unreliable (Madu, 2005). As engineering disciplines, reliability and
maintainability are relatively new. Reliability and maintainability are not only important parts
of the engineering design process but also necessary functions in life-cycle costing, cost
benefit analysis, operational capability studies, repair and facility resourcing, the
determination of inventory and spare parts requirements, replacement decisions, and the
establishment of preventive maintenance programmes.
The first European standard (EN 50126) for the railway system in this context was published
in 1999 by CENELEC and defines Reliability, Availability, Maintainability and Safety
(RAMS) as a characteristic of a systems long term operation. The standard states that RAMS
is achieved by the application of established engineering concepts, methods, tools and
techniques throughout the life cycle of the system.
EN 50126 (1999) defines the basic RAMS elements as:
x
x
x
Reliability: the probability that an item can perform a required function under given
conditions for a given time interval.
Availability: the ability of a product to be in a state to perform a required function
under given conditions at a given instant of time or over a given time interval,
assuming that the required external resources are provided.
Maintainability: the probability that a given active maintenance action, for an item
under given conditions of use, can be carried out within a stated time interval when the
maintenance is performed under stated conditions and using stated procedures and
resources.
Safety: the state of technical system freedom from unacceptable risk of harm.
2.1.1 RAMS Parameters

A thorough understanding of the technical description of the system is necessary to perform
RAMS analysis of the system. In the case of railway infrastructure, there are three different
Basic Concepts and Definitions
systems, namely the track system, the signalling and communication system, and the power
system. These systems have a combined effect on the degradation of the infrastructure. Each
system is subjected to degradation due to various internal and external factors. All these
aspects need to be considered to estimate the RAMS of the infrastructure, which makes the
calculation more complex. The following sections present some of these factors affecting
RAMS. To estimate the RAMS figures at the infrastructure level, one must evaluate the
RAMS characteristics at the sub-system and component level. In general, the reliability and
maintainability parameters are estimated both on the component level and on the system level,
whereas the availability and safety parameters are estimated only on the system level. In order
to achieve the required performance of the infrastructure, the failure modes should be
identified and classified into the failure categories illustrated in the Table 2.1. A higher
RAMS target is set for significant failure, whereas a not-so-high target is set for the minor
failure category. Thus, the infrastructure managers should know which failure modes of the
track should be given more attention in order to achieve reliability at the system level and
incur less cost due to failure.
Table 2.1: RAM failure categories (EN 50126, 1999)
Failure Category
Definition
Significant
(immobilizing failure)
A failure that
- prevents train movement or causes a delay to service
greater than specified time and/or generates a cost greater
than a specified level
Major (service failure)
A failure that
- must be rectified for the system to achieve its specified
performance and
- does not cause a delay or cost greater than the minimum
threshold specified for a significant failure
Minor
A failure that
- does not prevent a system achieving its specified
performance and
- does not meet criteria for significant or major failures
2.1.2 Factors Affecting RAMS

To achieve a dependable system, the factors which could influence the RAMS of the system
need to be identified, their effects need to be assessed, and the causes of these effects need to
be managed throughout the life cycle of the system. The RAMS of a railway system is
influenced in three ways:
x
System conditions: the sources of failures are introduced internally within the system
at any phase of the railway system life cycle. These failures are incurred by the design
and manufacturing of the components or the system.
Operating conditions: the sources of failures result from the operating system
methodology. These failures are also incurred by environmental conditions.
Maintenance conditions: the sources of failures are caused by maintenance actions.

Failures in the railway infrastructure can be caused not only by the maintenance
actions on the infrastructure but also by the maintenance of rolling stocks.
These sources of failure can interact with each other and the relationship is shown in Fig. 2.1.
In the figure, it can be seen that reliability is not explicitly shown, but is given through the
group of internal and external failures in the system. The factors that influence RAMS, as
shown in the figure, are generic and can be applied across all industrial applications with
some applications, in transport systems. In order to achieve a dependable track system, the
factors specifically affecting the track RAMS need to be identified. Table 2.2 identifies the
specific factors that affect the track RAMS.
The factors mentioned above affect the characteristics of RAMS. Similarly, the quality of
RAMS data affects the correctness of the RAMS estimation. Many types of data are relevant
to the estimation and prediction of reliability, availability, and maintainability. Not all are
collected in many instances, and the lack of information is sometimes a serious problem in
RAMS analysis (Blischke and Murthy, 2003). Markeset and Kumar (2003) described some of
the factors influencing the management of RAMS data. These factors concerned user skills
and capabilities, and locations, etc., apart from the data type, data format and detail level. The
different physical parameters that affect the track RAMS are listed in Table 2.2. In order to
assess the effect of these parameters on the track RAMS, it is important to know the technical
characteristics of these parameters.
Technical parameters
Quasi-static stress
Quasi-static stress
Yield strength (Young's modulus)
Stiffness, Damping
Stiffness, Damping, Bending stress
Damping
Stiffness, Damping
Operating conditions
Physical parameters
Track curvature (transient curve in, transient curve out,
radius)
Track gradients (start, end, value)
Rail (rail type, jointed or welded)
Ballast (ballast type, ballast size)
Sleeper (sleeper type, sleeper spacing)
Fastener (fastener type)
Subgrade (geological condition)
Track operating conditions:
Maintenance conditions
System conditions
Table 2.2: Factors affecting track RAMS
Grinding
Wear rate
Tamping
Change in track stiffness
Lubrication
Change in friction co-efficient
Renewal of track components
Interval of renewal
Corrective replacements of track components
Failure rate of track components
Loads (annual MGT, maximum axle load)
Bending stress, Shear stress, Contact stress
Environment (temperature)
Thermal stress
Vehicle operating conditions:

Speed of trains
Vertical stress, Lateral stress
Vehicle condition (hollow wheels)
Dynamic stress
10
- Errors in requirements
- Design inadequacies
- Manufacturing
deficiencies
- Inherent weakness
- Software errors
- Operating instruction
deficiencies
- Instruction inadequacies
- Human errors
Systematic
Failure
External
disturbances
Environmental
conditions
Human
errors
Human
factors
Human corr.
actions
Procedure
Logistics
- Manual
- Automatic
Diagnostics
Reconfiguration
modes
Change in
misson profile
Mission
profile
Operating Conditions
Availability
11
Preventive
maintenance
- Internal
- External
Diagnostics
Conditional
maintenance
Corrective
maintenance
Maintenance
procedures
Maintenance Conditions
Human
factors
Scheduled
maintenance
Figure 2.1: Factors influencing railway RAMS (EN 50126, 1999)
- Operating modes
- Environment
- Stress degradation
- Wear out
- Over stress
Random Failure
Technical
characteristic
Internal
disturbances
Maintainability
System Conditions
Safety
Railway RAMS
For example, in order to estimate the effect of the trainload on the RAMS characteristics of
the track, one must know the bending stress, the shear stress and the contact stress imparted
by the track load on the track. Similarly, the sleeper types and spacing determine the bending
stress, stiffness and damping of the track. We can state that the technical parameters are the
causes of the physical parameters which directly affect the track RAMS. The system
conditions are mostly related to the design and manufacturing of the track components,
whereas the operating conditions are connected to the rolling stock operations. In most of the
cases it is difficult to change the system conditions and operating conditions of the track
system in the operation and maintenance phase of the track, although sometimes the operating
conditions (e.g. changes in the axle load) can change because of changes in railway
regulations.
In order to identify relevant maintenance actions for the track, it is crucial to have a good
understanding of the failure modes and their causes on the track. Failure Mode, Effects and
Criticality Analysis (FMECA) acts as a tool to reveal these failure mechanisms. To support
this task, four failure progressions are defined, namely (Jovanovic, 2006):
1. The component is subjected to gradual degradation which may be observed by suitable
equipment.
2. The component is subjected to gradual degradation which cannot be observed.
3. The component is subjected to a sudden degradation which can be observed by
suitable equipment.
4. The component is subjected to shock degradation, immediately leading to failure.
This classification is particularly useful when reliability parameters are being assessed, as
these parameters have different interpretations for the four categories of failure progressions.
The track failure modes and their corresponding limits can be categorised as stated below
(Esveld, 2001). The limits can be set per unit length of the track.
x
x
x
x
x
x
Track geometry condition (standard deviation in the longitudinal level, standard

deviation in the cross level, and standard deviation in the gauge).
Ballast condition (percentage of weedy ballast and percentage of surface soiling).
Fastening condition (percentage of loose fastenings).
Sleeper condition (percentage of bad sleepers, percentage of medium sleepers, and
percentage of good sleepers).
Rail defects (number of RCF defects, number of rail breaks, number of weld defects,
and amplitude of corrugation).
Rail wear (vertical wear of the rail head and lateral wear of the rail head).
Measuring the infrastructure condition is prerequisite for track maintenance planning.

Banverket has a number of condition indices to describe the condition of their infrastructure
facilities (Andersson, 2002). The main condition indices are known as the K-value and the Qvalue. These are calculated from detailed inspection car measurements of the track. The
12
inspection car measures the relative rail position (the lateral and vertical position), the rail
profile and the rail gauge. The Q-value is a weighted index of the standard deviation of two
inspection car measures calculated the deviation from the geometric comfort limits set for the
specific track class. The Q-value is calculated per kilometre of track as:
V
V
Q 150 100 H 2. S / 3
V S lim
V H lim
2.1
where H and S are the average standard deviation of the height and interaction on the
section measured. The standard deviation for the interaction is calculated as a combined effect
of the cant and the side position of the rail. Hlim and Slim are the comfort limits for a given
track class. Track class classifications are based on the speed of the train
The K-value is calculated for a longer section of the track and is expressed as:
l .100%
2.2
where l is the sum of the track length where all the values are below the comfort limits for
a given track class and L is the total length of the track considered. The K-value is not
suitable for shorter track sections.
The failure modes mentioned above must be categorised as per the failure categories given in
Table 2.1 so as to proceed with RAMS analysis and define the RAMS targets for different
failure categories. The goal of the railway system is to achieve a defined level of rail traffic in
a given time, and safely. RAMS has a clear influence on the quality with which the service is
delivered to the customer. Moreover, in-service safety and availability can only be achieved
by meeting all the reliability and maintainability requirements and controlling the ongoing
and long-term maintenance and operational activities and the system environment.
2.2 Infrastructure Maintenance Planning

Maintenance is defined as the combination of all the technical and administrative actions,
including supervisory actions, intended to retain an item in, or restore it to, a state where it
can perform a required function (IEV 191-01-07, 2007). Maintenance has long been
considered as a reactive, fire-fighting approach. However, as the dependability targets for
assets have become increasingly important, several proactive maintenance approaches and
methods are being developed. All the decisions related to rail infrastructure maintenance are
taken in order to keep a balance between economic and safety aspects. The goal is to find the
13
effective maintenance procedure to optimise the track possession period and the train speed
restriction regime and ultimately increase the track availability.
The different components of the railway asset are structurally and economically
interdependent. Scale effects are involved in their maintenance and renewal, while their
degradation is often structurally related. As operations have to be continued on the rail
network and budgets are often restricted, all kinds of constraints have to be considered in the
planning of infrastructure maintenance. The concepts of the maintenance planning process are
developed in the following steps (Zoeteman, 2006):
x
x
x
Generation of maintenance strategies for individual assets (e.g. corrective or

preventive, time based or condition based, strategies are distinguished based on the
criticality of the individual asset for the entire production system)
Definition of clustering rules, which optimise the frequencies of activities on the basis
of scale or scope effects
Definition of rules for assigning time windows to maintain packages on the basis of
opportunities that occur in the middle or short term.
The initial analytical work for track maintenance was carried out in the early 1980s. Fazio and
Prybella (1980) pointed out a number of prerequisites for planning track maintenance. Track
quality measures and track deterioration models are highlighted as key areas for a structured
planning process to be established. Zarembski (1998) described three tools which railway
organisations could use to improve the efficiency of maintenance operations (see Fig. 2.2)
automated inspection systems, databases and maintenance planning systems. The lack of
integration between these tools has prevented railway organisations from taking full
advantage of their potential.
Automated
Inspection System
Visual Inspection
Track Geometry
Track Structure
Track Deterioration
Modelling
Database
Traffic
Tr
Analysis, Planning & Schedule
Maintenance Plan
Budget
Figure 2.2: Maintenance planning overview (Zarembski, 1998)
14
These different data sources need to be linked in a general database for planning purposes. By
adding models for track deterioration relationships, the state of the infrastructure can be
assessed over time. The planning of specific maintenance activities will be affected by the
conditions of the track. This requires a detailed knowledge of each component of the track
and its relationship with other components of the track, as well as the degradation pattern of
each component.
Therefore, the objectives of infrastructure maintenance planning can be described as finding
answers to the following questions:
x
x
x
What are the current conditions of the infrastructure? (Track quality indices)
What will be needed in the short term as well as the long term as far as maintenance is
concerned? (Forecasting of maintenance actions)
What should be done first? (Prioritization of maintenance activities)
Table 2.3 illustrates the effect of the grinding strategy of the Canadian Pacific Railway on the
reliability of the rail. It can be seen in the table that, as the grinding strategy moves from
corrective to preventive grinding, the rail life increases considerably. Corrective grinding
requires deep and infrequent cuts, whereas preventive grinding requires thin but more
frequent cuts (Kalousek et al., 1989). Generally for heavy haul railways, the minimum
interval for rail grinding is in the range of 10-15 million gross tonnes (Canon et al., 2003).
The wear rate in grinding is the parameter that controls the rail life, because as the wear
reaches the maintenance/safety limit of the rail, the rail needs replacement. No grinding is a
scenario where the life of the rail is determined mostly by RCF. Table 2.3 also gives a
comparison of the fatigue lives in three grinding scenarios. The fatigue life of the rail is
reached when the number of RCF defects in a specific track section reaches its limit.
Table 2.3: Grinding strategy vs. rail life for Canadian Pacific Railway (Magel and Sroba,
2007)
Wear Criteria
No
grinding
Corrective
grinding
Preventive
grinding
Rail wear rate

in mm/MGT
0.04
0.06
0.03
Rail life in MGT
469
367
844
Rail fatigue life

in MGT
331
496
1322
The difference between the rail wear life and the rail fatigue life is illustrated in Fig. 2.3. As
the material removal rate of wear and grinding increases, the rail wear life decreases as the
wear approaches the maintenance/ safety limit of the rail. However, an increase in the
15
material removal rate of wear and grinding increases the rail RCF life, because grinding and
wear take away the RCF generated cracks before they become critical to the rail. Therefore,
the grinding strategy (the wear rate) is seen to be an important parameter that affects the
reliability of the rail. The figure also illustrates the magic wear rate phenomenon. The
magic wear rate is the wear rate that the preventive grinding strategy should attain in order to
achieve the highest reliability for the rail. As shown in the figure, when the wear rate is below
the magic wear rate, the rail life is determined by the rails RCF life, whereas, when the wear
rate is higher than the magic wear rate, the rail life is determined by the rail wear life.
Life line
due to wear
Life line
due to RCF
Rail life
Magic wear
rate
Rail life
Material removal rate by grinding and wear
Figure 2.3: RCF, wear and rail life relationship (Magel and Sroba, 2007)
In order to assess the effects of the maintenance conditions on the reliability of the track
system, it is necessary to consider their combined effect on the system. As described above,
grinding affects the reliability of the rail. However, to perform an effective reliability analysis
of the rail, the combined effects of other maintenance conditions, e.g. lubrication, rail
replacements, etc., should be taken into account. For example, lubrication reduces the rail
wear, especially in the track curves (Diamond and Wolf, 2002), and thereby increases the
reliability of the rail. However, at the same time, lubrication is a factor for RCF defects,
which are removed by grinding (Rinsberg, 2001). In order to fulfil the above mentioned
objectives, RAMS analysis will play a major role in maintenance planning. The details are
presented in the following sections of the chapter.
2.2.1 RAMS Analysis for Maintenance

RAMS analysis is a process which utilises the failure information from a system in order to
develop probability distributions for the systems ability to perform its intended functions.
RAMS analysis for the track is based on the following elements:
x
x
RAMS database
Failure modes
16
Methods and tools for RAMS analysis
The utilisation of failure and maintenance data is an important factor in RAMS analysis and
the management of the system. There are several dimensions with respect to the collection of
RAMS data. One should ascertain that the data being collected support all the types of RAMS
analysis required for the system. Another important aspect is that the data should support the
life cycle perspective of the system and, more importantly, the maintenance phase in this case.
Fig. 2.4 illustrates the use of a RAMS database as a feedback to the RAMS analysis, as well
as to the operation and maintenance phase of the system life cycle.
RAMS
analysis
Maintenance
planning &
implementation
Knowledge
transfer
Means/
modifications
RAMS data
Figure 2.4: A process loop showing the use of a RAMS database in RAMS analysis
(Prom@in, 2003)
For an effective analysis of RAMS, the traffic and track geometry databases should be
considered along with the failure and maintenance databases as mentioned above. Therefore,
the track must be divided into homogenous analysis segments with respect to the track
curvature, grade, super elevation, traffic density, etc. The following data are also a part of the
RAMS database, along with the failure data (Esveld, 2001).
I) Layout and operating data
x Curves (start and end km, transition curves, radius, etc.)
x Loads (annual load (MGT), maximum axle load (tonnes), date from which the data are
valid, etc.)
x Speeds (speed of freight and passenger trains, date from which the speed is valid, etc.)
x Gradients (start, end, and value)
II) Infrastructure
x Subgrade (geological conditions, various monitored parameters, etc.)
x Ballast (ballast type, date of installation, ballast thickness, etc.)
x Sleepers (sleeper type, sleeper spacing, new/old sleepers when laid, type of fastenings,
and date of installation)
x Rails (rail type, joined or welded track, weld type, date of installation, new/old rails
when laid, date of installation, and cumulative tonnage on rails when installed)
17
III) Work history

x Renewals, grinding and tamping work history (start km, end km, and type)
x Speed restriction history (start and end date of temporary speed restriction, and value
of reduced speed)
x Spot maintenance history (type, and date)
Banverket uses BIS (a track information system), BESSY (an inspection system), 0FELIA (a
fault analysis system) and TFR (a train delay system) for maintenance planning. Recent
developments are HANNES (a speed restriction system) and a track geometry data system.
These existing systems are more or less stand alone modules using BIS as a reference system
(Andersson, 2002). Therefore, there is a need to integrate the different databases for efficient
analysis of RAMS and maintenance planning. It is also necessary to investigate the effect of
track degradation/failure on the other systems of the railway infrastructure, e.g. the signalling
systems. It is evident that, to achieve the availability target of the overall railway
infrastructure, we need to assure the availability of all the systems that are part of the
infrastructure.
2.2.2 RAMS in the Operation and Maintenance Phase

The system life cycle is a sequence of phases, each containing tasks, covering the total life of
a system from the initial concept to decommissioning and disposal. The life cycle provides a
structure for planning, managing, controlling and monitoring all the aspects of a system,
including RAMS, as the system progresses through the phases, in order to deliver the right
product at the right price within the agreed time scales. A system life cycle that is appropriate
in the context of railway operation is shown in Fig. 2.5. The top-down branch (on the lefthand side) is generally called design and development and is a refining process ending with
the manufacturing of system components. The bottom-up branch (on the right-hand side) is
related to the assembly, the installation, the receipt and then the operation of the whole
system. The "V" representation assumes that the activities of acceptance are intrinsically
linked to the design and development activities, insofar as what is actually designed has to be
finally checked in regard to the requirements. Therefore, the validation activities for
acceptance at various stages of a system are based on the specification of the system and
should be planned in the earlier stages, i.e. starting at the corresponding design and
development phases of the life cycle.
18
Review RAMS
implications
Concept
Establish
RAMS plan
System definition
and application
conditions
Perform safety
assessment
Define RAMS
acceptance criteria
Risk analysis
System
requirements
Define sub-system and

component RAMS
acceptance criteria
Apportionment of
system
requirements
Implement and control

RAMS programme
Analyse RAMS
statistics and optimise
System
acceptance
Operation
and
maintenance
System
validation
Design and
implementation
Perform RAMS
improvement testing
Assess RAMS
demonstration
Installation
Decommissioning
and disposal
Perform RAMS
demonstration
Prepare RAMS
support documentation
Manufacture
Figure 2.5: The V representation of the RAMS life cycle (IEC 62278, 2002)
The figure also describes the various RAMS activities being carried out at each phase of the
system life cycle (IEC 62278 2002). To achieve the overall RAMS objectives of the system, it
is important to follow systematic RAMS actions throughout the life cycle of the system. As
far as RAMS activities are concerned, one of the important phases of the system life cycle is
the operation and maintenance phase, where RAMS is optimised by the analysis of real life
failure data.
The objective of this phase is to operate, maintain and support the total combination of
components and subsystems in such a way that compliance with the system RAMS
requirements is maintained. Fig. 2.6 illustrates the RAMS process for the railway
infrastructure in the operation and maintenance phase of the system life cycle. It is a
continuous improvement process throughout the operation and maintenance phase. The
sources of failures are due to the system itself, train operation or maintenance activities
carried out on the track. The failure data are collected by FRACAS (Failure Reporting And
Corrective Action System). FRACAS is a closed-loop reporting system for identifying failure
modes and their root causes and subsequently determining effective corrective actions for
eliminating their re-occurrence.
19
Define RAMS targets

for the system
Develop procedures
for RAMS analysis
Develop maintenance
plan
Develop structure for
collection of RAMS
data
Change in
maintenance plan
Analyse RAMS data
Feedback on
RAMS procedures
Feedback on
data requirement
Develop RAMS
predictions
Predictions
less than the
targets
Compare
predictions with
targets
Predictions
meet or exceed
the targets
Figure 2.6: RAMS process in the operation and maintenance phase

Failure Mode, Effects and Criticalily Analysis (FMECA) is an analysis method involving two
elements of risk namely failure frequency and consequence. FMECA analysis concentrates on
the identification of the events and frequency resulting in failures and on analysing their
effects on the components and systems. FMECA categorises the failures as non-safety critical
failures and safety critical failures. Other tools that are being used for RAMS analysis are
Reliability Centred Maintenance (RCM), Markov analysis, etc. The basic objective of the
operation and maintenance phase is to monitor RAMS activities in order to meet the RAMS
goals set for the infrastructure. The performance indicators for checking the goals are the
RAMS parameters described earlier in this chapter. If the goals are not met at any point of
time, then changes in the maintenance conditions are made in order to meet the goals. If the
infrastructure operating conditions change during the operation and maintenance phase,
changes in the maintenance conditions are required accordingly to meet the RAMS goals.
RAMS analysis of the infrastructure should not be performed without considering the
operational characteristics of the rolling stock. As stated in EN 50126 (1999), the operational
availability of the track hardly considers the train schedule. To have a realistic measure of the
20
availability of the infrastructure, it is necessary to consider the demand availability when

dealing with the operational availability. The demand availability is the probability that a
system will be in a functioning state on demand (Kumar and Akersten, 2008). In the case of
track, the demand availability defines that a unit length of track is available when trains pass
over it. To achieve the demand availability of the track section, the following measures must
be considered:
x
The corrective maintenance on the track must be reduced. As failures on the track can
occur at random, the lower the number of failures are, the better is the demand
availability.
All the preventive maintenance and renewal actions on the track must be carried out in
the train free periods. The maintenance plans for the track need to utilise the train free
periods to a maximum for all the maintenance actions.
In order to calculate the demand availability of a track section over a period of time, the
reliability and maintainability of the track, along with the train timetable need to be
considered.
2.3 Maintenance Optimisation

As rail infrastructure is an expensive asset with a long lifespan, the cost-effectiveness of long
term design and maintenance decisions should be guaranteed. LCC analysis, an engineering
economics technique, can be utilised to focus on maintenance strategies to minimise the life
cycle cost, while meeting the dependability requirements. Fig. 2.7 depicts LCC calculations as
being based on the business and technical requirements of the infrastrcuture, which are based
on a specific operational scenario. The maintenance policy and budget constraints play a
major role in selecting the alternative maintenance strategies. They act as a crucial input when
deciding upon a particular maintenance strategy. Fig. 2.7 shows how RAMS affects the LCC
calculation at various stages. The maintenance strategy (MS) with the lowest LCC is
considered as the cost-effective solution to be implemented in the infrastructure operations.
The maintenance strategy can be a single maintenance action (e.g. grinding) or a cluster of
maintenance actions. For an effective decision on maintenance strategy, it is important to
consider a cluster of maintenance actions for LCC calculations, as maintenance actions affect
each other.
21
Business/ technical
requirements of railway
infrastructure
RAM
RAMS
targets
Maintenance Strategy
MS 1
MS 2
MS 3
LCC 1
LCC 2
LCC 3
RAM
RAMS
analysis
RAM
RAMS
parameters
Figure 2.7: Maintenance decisions based on LCC

When considering the maintenance strategy with the lowest LCC as the cost-effective
solution, it is important to consider the RAMS figures associated with that particular
maintenance strategy. Therefore, when not considering the maintenance strategy with the
lowest LCC as the best solution, a trade-off between the RAMS targets and the LCC value is
necessary in order to achieve an effective maintenance strategy.
2.3.1 Life Cycle Cost Analysis

In many countries, the restructuring of railways and increasing efficiency and effectiveness
requirements are causing a changing environment for infrastructure management. The
responsibility for parts of the railway system is often handed over to different actors. In order
to guarantee optimal long-term results for the railway systems, the effects of decisions should
be systematically evaluated (Zoeteman, 1999). Putallaz (2003) states the three parameters (see
Fig. 2.8) that influence the performance of the track infrastructure: capacity, substance and
quality. The capacity of the infrastructure may be expressed in terms of usable train paths
during a certain time span. The substance of the infrastructure refers to the average remaining
useful lifetime of its components. Finally, the quality of the infrastructure represents the
quality of tracks geometry and components. Managing the infrastructure comes down to
setting those three parameters at their most appropriate level, in order to maximize efficiency.
Adjustments may be made to the capacity through the investment policy, to the infrastructure
substance through the renewal policy, and to the quality through the maintenance policy.
These three parameters cannot be adjusted independently. An old infrastructure (low
substance) requires more maintenance (to increase the quality), whereas a bad geometry (low
quality) increases the wear on the infrastructure (lower substance). Similarly, more
engineering works (maintenance & renewal) require more track possessions (less capacity),
while more traffic (high capacity) induces more wear of the infrastructure.
22
Investment
Capacity
(Number of slots)
al
ew
n
Re
Ma
in
Substance
(Average age of
infrastructure)
ten
anc
e
Quality
(Quality of geometric
parameters)
Figure 2.8: Three basic parameters of rail infrastructure influencing performance

(Putallaz, 2003)
When adjusting these performance parameters simultaneously, the cost aspect of each activity
should be considered. Since, in the current scenario, most of the maintenance and renewal
decisions are based on cost models that rarely consider their effects on the whole life of the
infrastructure, a need for a life cycle cost approach arises. Life cycle cost can be used as a tool
to take cost-effective decisions on investment, renewal and maintenance, in order to adjust
these three parameters to optimise the infrastructure performance.
The total costs can be observed from diverse points of view, i.e. from the viewpoint of the
systems supplier or of the systems user or owner, or, even more broadly, from the point of
view of society. A basic assumption providing motivation for the LCC approach is that it is
usually possible to affect the future costs of a product beforehand, either by planning its use
or by improving the product or asset itself (Markeset and Kumar, 2004). Asiedu and Gu
(1998) stated that LCC analysis should be regarded not only as an approach for determining
the cost of the system, but also as an aid for decision making in design, maintenance, etc. The
use of life cycle cost analysis should therefore be restricted to the cost that we can control.
In order to be able to estimate the life cycle costs of the rail infrastructure, the factors
influencing the performance of the railway infrastructure and their relationships need to be
identified. The driving factor causing failures and maintenance is the degradation of the asset.
Track degradation depends on many factors, such as the initial quality of the construction, the
quality of the substructure and the loads on the track. Besides asset degradation, there are
other factors that also influence the life cycle costs, such as the RAMS targets for the track,
the amount of preventive maintenance, the market prices of labour, materials and machines,
and the operational characteristics of the line (such as the axle loads, the traffic intensities and
23
the duration of train free periods). The infrastructure manager can control some of these
factors directly (e.g. the maintenance strategy) or with the cooperation of the transport
operators (e.g. the quality of the rolling stock) and the government (e.g. negotiated grants)
(Zoeteman, 2001). The performance of the railway infrastructure is influenced by factors such
as the level of safety, riding comfort, noise, vibrations, reliability, availability, and the costs of
ownership (see Fig. 2.9). Safety and noise standards indirectly influence the life cycle costs,
since they determine the tolerances and thresholds for the design and maintenance parameters.
The physical design influences the asset degradation, together with other conditions, such as
the traffic intensities and axle loads, the quality of the substructure and the effectiveness of
the performed maintenance. The quality of the geometric structure determines the required
volume of maintenance and renewal (M&R). The chosen maintenance strategy also influences
the amount of M&R. The realised M&R volume causes expenditures and planned
possessions. The maintenance strategy also has a direct impact on the life cycle cost. The
incident management organisation, the realised M&R volume and the transport concept
determine the train delay minutes caused by the infrastructure and these train delay minutes
can be converted into penalties for the infrastructure managers. The cost models used in the
decision support systems or maintenance management systems should be able to provide
means to evaluate and compare the costs and benefits of different maintenance strategies and
options. In order to carry out an economic analysis, it is necessary to make adjustments to
costs to ensure that they are all measured in the same units and represent real costs of
resources (Larsson, 2002). According to Zoeteman (2001), the life cycle cost can be presented
in three different ways, i) the total present value (TPV), ii) the internal rate of return (IRR),
and iii) the annual equivalent or annuity (ANN).
Transport concept
External variables
Design and maintenance

specification
Physical environment
Design and
maintenance
specification
Maintenance
conditions
Building conditions
Building conditions
Financial conditions
Transport concept
Noise,
vibrations,
safety & riding
comfort
Quality of geometry
and structure
Design
Cost of
ownership of
rail network
Amount of
construction work
Failure performance &
expenditure on infra
Maintenance
strategy
Steering variables
Planned volume
maintenance/ renewal
Realised volume
Infra reliability
Planned
availability
Variables represented in life cycle cost
Figure 2.9: Factors influencing the performance of track infrastructure (Zoeteman, 2001)
24
To check the robustness of LCC models, two methods are used:

x
Sensitivity analysis: The disadvantage of sensitivity analysis is that only one variable
is tested at a time. Hence, possible interactions between factors are not revealed.
Uncertainty analysis: In this approach the input parameters of the LCC model are
considered to be random variables from which samples are drawn. Simulation
techniques are used to determine the interaction of the input parameters with the
outcomes.
The life cycle cost of the track infrastructure depends mainly on two aspects of the
infrastructure, i.e. the network configuration and complexity and the network utilisation.
Complexity is a predominant parameter for investment and the cost of maintenance. Some
major indicators are (LICB, 2007):
x
x
x
x
Density of switches
Length of lines on bridges and tunnels
Lengths of double track lines
Degree of electrification
In addition configuration parameters like curvature, axle loads and speed level have their
impact on the life cycle expenditure.
The utilisation of networks has a strong impact on the cost of maintenance and on the
components technical life until replacement. Some major indicators are:
x
x
Average frequencies of trains per year

Average gross tonnage per year (freight and passengers)
It is difficult to generalise the LCC per kilometre of track because of the tracks variability in
terms of complexity and utilisation. A harmonisation model is used to compare the cost data
of different track configurations and utilities in the best possible way (Stalder, 2001). Various
aspects of the harmonisation model are given by:
x
Single vs. multiple track: The maintenance and renewal of single-track lines require
more work per kilometre than that of double or multiple track lines (e.g. for work site
logistics and preparatory work). Based on a detailed analysis of data from the French
National Railway Company (SNCF) and surveys of other railways, it is concluded that
the cost of maintenance per track kilometre on single track is typically 40% higher
than that on double track. Therefore, this aspect should be taken into account when
estimating the LCC per track kilometre.
25
Switch densities: The switches on the main track have a major share in the cost of the
track maintenance (with a high impact on the signalling and the power supply). With
the switch densities varying between the main tracks, the need for harmonisation is
evident.
Track utilisation: The maintenance and renewal as well as the lifetimes of track
elements depend heavily on the utilisation of networks. Data analysis has proven that
maintenance expenditures can best be harmonised according to train frequencies, in
particular because of the strong correlation between the track access times and the
maintenance cost. Renewal expenditures are harmonised according to the gross
tonnage, which has a great impact on the wear and tear of the track.
2.3.2 Maintenance Management

The asset strategy is the maintenance approach and plan developed for each item of the
system. This strategy determines what planned and programmed maintenance work should be
carried out, and considers what potential problems may require an unplanned, reactive
response (Wilson, 1999). The approach to be taken in developing the maintenance activities
and, on the basis of these, the asset strategies needs to be understood before the maintenance
management strategy is completed. This is because, until it is known how much maintenance
activity will be required, who will carry out, and what spare parts will be used, etc., then the
approach to the organisation of maintenance activity cannot be finalised.
Infrastructure managers try to ensure the successful management of costs and quality, and the
relation between the two. This is essential because the train operators as well as the
passengers are imposing ever increasing quality requirements on the rail infrastructure.
Therefore, the infrastructure managers require the best infrastructure quality at the lowest
cost. The way to achieve this objective is through proper maintenance management.
Esveld (2001) gave examples of the type of data required for the Track Maintenance
Management System (TMMS), as listed below:
x
x
x
x
x
x
Measurements
Planning
Infrastructure
Inspections
Work carried out
Costs
However, difficulties in the accurate anticipation of maintenance prevent extremely precise

maintenance planning and management. Besides, the amount of funding allocated for
maintenance work is often regarded as a compromise, as too much according to top
26
management, and too little according to the operating and maintenance staff. Consequently,
the selection of the optimal maintenance strategy can be challenging. A systematic approach
for the determination of the deterioration of track components is necessary to gauge fully the
status of the track system and components. This will require proper track condition
assessments, the establishment of a standard condition rating system, and the development
and regular updating of prediction models for various track components.
Esveld (2001) discussed the idea of rational rail management for infrastructure. Rational rail
management aims at the objective evaluation of the qualitative and quantitative assessment of
the rail infrastructure, after which, based on the system objectives, and on rules and standards,
decisions may be taken regarding the maintenance and renewal of rail infrastructure. Rational
rail management is summarised in the following objectives:
x
x
x
To be less dependent on the individual know-how of co-workers

To create optimal working conditions regarding business economy economics
To bear responsibility and to report to the management
Therefore, an effective infrastructure maintenance management system requires RAMS

management and life cycle cost management to be thoroughly integrated into the asset
management of the system. The idea of system-effectiveness emerges so as to make the LCC
analysis cost-effective. System-effectiveness concerns with RAMS characteristics of the
system. Blanchard and Fabrycky (1998) define system-effectiveness as the probability that a
system will successfully meet an overall operational demand within a given time and when
operated under specified conditions. In short, system-effectiveness is the ability of a system to
perform a job for which it is intended. It can be defined as a function of the systems
operational availability, operational reliability and capability.
Operational availability of the infrastructure is defined as the probability that the
infrastructure will be operationally available during the train traffic.
Operational reliability is the probability that during the train traffic operation, the
infrastructure will not suffer from any failures.
Capability is the ability of the infrastructure to meet its required objectives.
System-effectiveness can be defined as:
System-effectiveness = Operational availability *Operational reliability *Capability
The higher the system-effectiveness is, the better the infrastructure is at achieving to achieve
its objectives.
27
Cost-effectiveness analysis yields quantitative results to aid the decision maker with risk
analysis and provides a useful decision tool.
Table 2.3 shows the calculation of cost-effectiveness from the LCC values of different
alternatives. When taking a decision on maintenance alternatives it is necessary to calculate
the cost-effectiveness of different maintenance alternatives. The higher the cost-effectiveness
is, the better is the maintenance alternative.
Table 2.3: Cost-effectiveness of maintenance alternatives
Maintenance
alternatives
Life Cycle Cost

(LCC)
System
effectiveness
Cost-effectiveness
(SE/LCC)
M1
LCC1
SE1
SE1/LCC1
M2
LCC2
SE2
SE2/LCC2
M3
LCC3
SE3
SE3/LCC3
RAMS management
LCC management
New
Infrastructure
Renewal
Cost
Effectiveness
Maintenance
management
System
Effectiveness
Asset management
Asset per
performance
Fig. 2.10 illustrates the relationship between maintenance management, asset performance
and asset maintenance. The asset management of the track concerns two important aspects of
the asset, i.e. the asset performance and the asset maintenance. System-effectiveness and costeffectiveness act as indicators for asset performance. Asset maintenance concerns activities
ranging from small scale maintenance actions to the building of new infrastructure.
Large scale Small scale

maintenance maintenance
Asset maintenance
Figure 2.10: Factors influencing maintenance management (Adapted from Swier and Luiten,
2003)
28
As described in the previous sections, RAMS and LCC analyses act as tools for estimating the
system and cost-effectiveness of the asset, as well as for taking effective decisions on the
maintenance of the asset. There is a close relation between asset maintenance and asset
performance, as effective asset maintenance increases the asset performance, while asset
performance acts as a decision tool for asset maintenance.
29
3 Research Methodology
Research can be defined in many ways. Most generally defined, research is a process through
which questions are asked and answered systematically. As a form of criticism, research can
include the question of whether or not we are asking the right questions (Dane, 1990). In other
words, research is a systematic examination of observed information, performed to find
answers to problems. Research methodology is the link between thinking and evidence
(Sumser, 2000). To conduct research, it is essential to choose a clear methodology. This
provides a framework for integration of the different technical, commercial, and managerial
aspects of study. The study of research methodologies provides the researcher with the
knowledge and skills that are needed to solve the problems and meet the challenges of a fastpaced decision making environment (Cooper and Schindler, 2006).
There are many ways to carry out research, but the purpose of research can be classified into
three main categories i.e. the exploratory purpose (to explore a new topic), the descriptive
purpose (to describe a phenomenon) and the explanatory purpose (to explain why something
occurs). The details of these are described in Table 3.1.
Table 3.1: Different kinds of research purposes (Neuman, 2003)
Exploratory
- Become familiar with the
basic facts, setting, and
concerns
- Create a general mental
picture of conditions
- Formulate and focus
questions for future research
- Generate new ideas,
conjectures, or hypotheses
- Determine the feasibility of
conducting research
- Develop techniques for
measuring and locating
failure data
Descriptive
- Provide a detailed,
highly accurate picture
- Locate new data that
contradict past data
- Create a set of
categories or classify
types
- Clarify a sequence of
steps or stages
- Document a casual
process of mechanism
- Report on the
background or context
of a situation
31
Explanatory
- Test a theorys
predictions or principle
- Elaborate and enrich
a theorys explanation
- Extend a theory to
new issues or topics
- Support or refute an
explanation or
prediction
- Link issues or topics
with a general principle
- Determine which of
several explanations is
best
The methodologies used in the present research are both descriptive and exploratory. The
research purpose of this study is to describe the methodologies of RAMS and LCC analysis
for the railway infrastructure, and to describe the methodologies for utilising both RAMS and
LCC analysis in making track maintenance planning decisions.
3.1 Research Approach

Research may be fundamental or applied in nature, depending upon the kind of knowledge
sought about a certain area and the solution intended. Fundamental research aims to widen the
knowledge of a particular subject so that future research initiatives may be based on the
extended knowledge. This research is designed to solve problems of a theoretical nature, with
little direct impact on strategic decisions. Applied research addresses existing problems and
opportunities (Cooper and Schindler, 2006).
This thesis concerns applied research with a purpose to apply RAMS and LCC methodologies
in the railway infrastructure context and to develop models for maintenance decisions based
on RAMS and LCC analysis. The knowledge gathered from an extensive literature study and
from discussions and consultations with RAMS and LCC experts within Europe was applied
to delineate the usefulness of RAMS and LCC analysis in railway infrastructure maintenance
planning, so as to make the planning more effective and risk-based.
The research approach can be categorised as induction or deduction (Sullivan, 2001).
x
The induction approach uses observations, a knowledge base and empirical data to
explain and develop theories. The approach involves inferring something about a
whole group or class of objects from our knowledge of one or a few members of the
group or class.
The deduction approach can be applied to generate hypotheses based on existing
theories, the results of which are derived by logical conclusions.
The research approach can be quantitative or qualitative. In simple terms, quantitative

research uses numbers, counts, and measures of things whereas qualitative research adopts
questioning and verbal analysis (Sullivan, 2001).
In the present research, both deductive and inductive approaches have been applied. A
deductive approach has been applied to develop a process of RAMS and LCC application in
railway infrastructure maintenance, whereas an inductive approach has been applied to
develop maintenance models. Both qualitative and quantitative research methodologies have
been applied in this research.
32
Research Methodology
3.2 Reliability and Validity

According to Neuman (2003), reliability means dependability or consistency. It suggests that
the same things are repeated or reoccur under identical or very similar conditions. Reliability
means that the implementation methods of a study, such as data collection procedures, can be
applied by somebody else with the same result. Validity is concerned with whether or not the
study actually elicits the intended information. Validity suggests fruitfulness and refers to the
match between a construct, or the way in which a researcher conceptualises an idea in a
conceptual definition, and a measure. It refers to how well an idea about reality fits in with
actual reality (Neuman, 2003).
The data and information used in this research have been collected from reputed peerreviewed journals, refereed conference proceedings and reports, or from company databases,
which positively contributes to the researchs reliability. Well-established RAMS analysis
techniques have been applied in different case studies, which also contributes positively to the
reliability of the research.
In this research, different maintenance models have been developed using RAMS and LCC
methodologies. The obtained results are believed to support the validity of the research, as
they matched the theoretical and logical expectations. These models can be implemented in
different railway infrastructure systems in future to support the validity further.
33
4 Data Collection and Analysis

4.1 Data Collection
Data can be defined as the facts presented to the researchers from the studied environment.
Data may be divided into primary and secondary types. Data collected by the researcher for
the purpose of study through various experiments or onsite data recording are called primary
data. Primary data are sought for their proximity to the truth and control over error. Data
collected by other people/organisations and used by the researchers are called secondary data.
They have at least one level of interpretation inserted between the event and its recording
(Cooper and Schindler, 2006).
Qualitative data were collected through relevant scientific papers and articles from online
databases. Relevant books were searched for from Lucia (Lule University Librarys online
catalogue) and then perused, and relevant reports and licentiate and PhD theses from various
universities were also studied. Different databases were searched to extract both qualitative
and quantitative data. Relevant scientific papers and articles were extracted from online
databases, such as Elsevier Science Direct, Blackwell Synergy, Emerald, IEEE Xplore,
Proquest science journals, Inspec and Compendex, etc. Some of the articles were searched
from the references of other relevant articles. Different keywords were used for searching
these articles as mentioned in the abstract. Different combinations of these keywords were
also used to narrow down the number of hits. Some of the known articles were searched
directly from the journal databases. Quantitative data were collected from Banverkets BIS,
BESSY and 0felia databases from the Iron Ore Line (Malmbanan), and from the internal
databases of ALSTOM Transport, France. Cost-related data were collected through personal
consultations with experts at Banverket and ALSTOM Transport. The details of Banverkets
databases are given below:
BIS: This is Banverkets infrastructure register (computerized database) containing
information about infrastructure or facilities, arranged geographically in accordance with
Banverkets facility structure. In BIS, for example, information is collected prior to work on
train timetables and work in connection with inspections (Karlsson, 2005). Apart from this,
information about agreements, accident reports, the history of tamping and grinding, and
curve-information can also be obtained.
35
BESSY: This is an inspection system in which comments are registered per facility on the
completion of inspection. Data are also registered directly during the course of inspection
with the aid of a palm computer.
0felia: This is a database containing information on all the faults in the infrastructure that
have been registered for a particular railway facility. The faults are sorted on the basis of the
structure used in BIS.
Raw data collected from these databases were treated to extract the information that is used in
the models. Some of the rail break data was collected from Kumar (2008). Information was
also collected through discussions and consultations with experts from Banverket. Failure and
degradation data for the track and the track circuit were collected from these databases. The
data were tested for trend and for dependency characteristics before proceeding with a
specific reliability model. Fig. 4.1 illustrates the steps for failure data analysis before choosing
the best fitting model.
Chronologically
obtained data
Yes
Non-stationary
models (e.g. Power
law process)
Yes
Branching Poisson
process or other
similar models
Trend?
No
Identically
distributed data
Dependency?
No
Renewal process
Constant
Hazard rate
Homogeneous
Poisson process
Not
constant
Distribution free
techniques or other
fitted distribution
Figure 4.1: Possible exploratory steps in field failure data analysis before fitting distribution
models (Asher and Feingold, 1984)
36
Data Collection and Analysis
Failure, degradation and maintenance data for the signalling systems were collected from the
internal databases of ALSTOM Transport. Details of the databases cannot be described due to
confidentiality issues. The data acquired from the ALSTOM databases were altered before
they were used in the models to keep their confidentiality. Different types of statistical
distributions were examined and their parameters were estimated by using Reliasofts Weibull
++ 6 software (Reliasoft, 2003).
Researchers generate information by analysing data after their collection. Data analysis is one
step, and an important one, in the research process. Data analysis usually involves the
reduction of accumulated data to a manageable size, developing summaries, looking for
patterns, and applying statistical techniques. Further, the researcher must interpret these
findings in the light of the clients research questions or determine if the results are consistent
with the hypotheses and theories (Cooper and Schindler, 2006).
4.2 Data Analysis

The usual analysis of reliability and availability data is implicitly based on the assumption
that the times between failures (TBFs) and times to repairs (TTRs) are independent and
identically distributed in the time domain. Further, it is also assumed that they are
independent of each other and form two independent series. This means that the TBFs and
TTRs are free from trends and serial correlations. In such situation the recording of the TBFs
and TTRs data by magnitude is valid for fitting the various distributions for representing the
population of the TBFs and TTRs. The above discussion implies that before any reliability
analysis is taken up, the test for trends and serial correlations must be done to check whether
the usual assumption of independent and identically distributed (i.i.d) for the data sets are
contradicted or not (Kumar, 1989). In this thesis, suspended failure data points have also been
considered to estimate the time between degradations. Inspection data have been used to
estimate the probability distributions of the degradation times. As the point of occurrence of
the degradation between two inspections or between the functional failure and the last
inspection is uncertain, the probability distribution of the degradation life can be estimated
considering the interval data for degradation.
The failure rate of a repairable component depends on the type of corrective maintenance that
is applied in a range stretching from perfect maintenance to minimal maintenance. Perfect
maintenance repair brings the component age to zero (i.e. the component becomes as good as
new (AGAN)) whereas minimal repair keeps the components age un-modified (i.e. the
component stays as bad as old (ABAO)). In real world cases the repairs are neither AGAN
nor ABAO, but are something in between. In (Kijima, 1989), two models (type I and II) are
proposed that estimate the virtual age of the component after a repair. Kijima models consider
a parameter called the maintenance factor, which estimates the virtual age. If the maintenance
factor is 1, the repair is ABAO and for the maintenance factor 0, the repair is AGAN. Model
type I assumes that the repairs can only fix the damage incurred during the last period of
37
operation. Thus, the nth repair can only remove the damage incurred during the time between
the (n-1)th and the nth failures. Model type II assumes that the repairs fix all of the damage
accumulated up to the current time. As a result, the nth repair not only removes the damage
incurred during the time between the (n-1)th and the nth failures, but can also fix the
cumulative damage incurred during the time from the first failure to the (n-1)th failure.
If the times between the failures are denoted by x1, x2, xn, t the virtual age of the
component after the nth repair is given by
Vn = Vn-1 + (maintenance factor * xn) Kijima model type I
Vn = (maintenance factor*Vn-1) + (maintenance factor *xn) Kijima model type II
The maintenance factors have been calculated from the past failure times of the component by
applying the Kijima models discussed above.
In this thesis, various system states (degradation and maintenance) of the system are modelled
by Petri-Nets. Petri-Nets (see Fig. 4.2) are a graphical tool for the formal description of the
flow of activities in complex systems. In comparison with other more popular techniques for
graphical system representation (like block diagrams or logical trees), Petri-Nets are
particularly suited to representing in a natural way logical interactions among parts or
activities in a system. Typical situations that can be modelled by Petri-Nets are
synchronization, sequentiality, concurrency and conflict. The theory of Petri-Nets originated
from the doctoral thesis of C.A. Petri in 1962 (Petri, 1962). Since then, the formal language of
Petri-Nets has been developed and used in many theoretical and applicative areas. Petri-Nets
used for modelling real systems are sometimes referred to as condition/events nets. Places
identify the conditions of the parts of the system (working, degraded, or failed), and
transitions describe the passage from one condition to another (end of a task, failure, or
repair). An event occurs (a transition fires) when all the conditions are satisfied (the input
places are marked) and give concession to the event. The occurrence of the event modifies
wholly or in part the status of the conditions (marking).
Place
Token
Arc
Transition
Figure 4.2: Schematic representation of a Petri-Net
38
Data Collection and Analysis
The number of tokens in a place can be used to identify the number of resources which are in
the condition denoted by that place. Petri-Nets have advantages over the Markov model
because of their ability to handle stochastic transition rates. Petri-Net models were
constructed using the software tool GRIF.
The statistical characteristics of R&M parameters contribute to uncertainty in LCC. The
reason for this is that the times and conditions for these types of events are so complex that
they cannot be predicted with a fair degree of accuracy. Therefore, it was decided to explore a
methodology that combines the use of design of experiment (DoE) principles with Monte
Carlo simulation to estimate the uncertainty involved with LCC. DoE was developed in the
twentieth century to increase the effectiveness and efficiency of experimentation. However,
for experiments to be effective and lead to correct conclusions there are a number of
requirements that must be fulfilled (Coleman and Montgomery, 1993). For example, the
response must be measurable and be correlated to the purpose of the experiment.
Furthermore, even though not an absolute necessity, the power of statistical operations will be
greater if the response is continuous and preferably also normally distributed. The responses
of this study are the point estimate for LCC of the track and its related uncertainty, which both
are continuous, but not necessarily normally distributed. The following are valid for the
present study.
1. The factors that are tested in the experiment are R&M-parameters, which all are
continuous and numeric. They are also measurable, controllable, and deemed
important for the selected responses.
2. The factors that are not under investigation can easily be held constant, since the
study is analytical and not empirical. These factors are the cost factors not directly
related to R&M. Hence, no randomization is considered necessary.
Since the study is analytical there are no major economical constraints. Hence, the design is
mostly dependent upon the number of R&M parameters that are to be investigated. In order to
fulfil the purpose of this study, a two-level factorial design is considered valuable. However,
in order to reduce the number of runs, a fractional factorial design is considered sufficient.
The analysis is supported by the software tool STATGRAPHICS, which provides suitable
tables and graphs for presentation.
The probability distribution of LCC can be found by the use of Monte Carlo simulation. A
Monte Carlo simulation is effectively a random number generator that creates values for each
R&M parameter. Values are chosen within specified ranges of each parameter and with a
frequency proportional to the shape of probability distribution associated with each R&M
parameter. The proposed methodology helps in determining the variable costs associated in
LCC estimation.
39
5 Summary of the Appended Papers

This chapter presents a summary of the five appended papers. Each paper makes its own
contribution towards the research questions and reports the findings of the case studies. For
detailed information, the reader is referred to the appended papers.
Paper I discusses a model for setting the availability target for railway infrastructure. The
availability of the infrastructure affects the capacity and the punctuality of the infrastructure.
Therefore, setting the availability target should be based on the capacity and punctuality
requirements. The objectives of the paper are to develop an approach to i) estimating the
capacity of the infrastructure based on the design and operational characteristics and
evaluating the influence of the infrastructure availability on the required capacity and ii)
estimating the volume of primary and secondary delay due to failures and maintenance of the
infrastructure. To achieve these objectives, an example is presented with parameters drawn
from failure, maintenance and traffic data. The model was developed using Petri-Nets and
Monte Carlo simulations. The simulation results show the effect of the availability of different
systems of the infrastructure on the train delays and the infrastructure capacity.
Paper II proposes a model for assessing the safety of the railway track by estimating the
probability of derailment. Models for the probability of derailment are developed based on
undetected rail breaks and poor track quality using Petri-Nets and Monte Carlo simulations.
The paper calculates the probability of undetected rail breaks on the track and the probability
of the track quality index falling below the maintenance limit for the track at a given age. The
models consider different inspection data and maintenance data as their parameters. The
models also calculate the risk based inspection intervals based on the safety requirements of
the track. Moreover, the paper discusses the difference in the probabilities between the old
and the newly laid tracks of Banverket. It has been shown in the paper how the frequency of
track inspections and track quality measurements affect the probabilities. The reduction of
these probabilities decreases the risk of derailment. This model will help the infrastructure
managers to estimate additional maintenance investment to increase the safety performance of
the track to a desired level.
41
The purpose of Paper III is to estimate the availability of the DC track circuit in scheduled
and condition based maintenance regimes. The models developed in the paper can estimate
the optimum inspection interval for the track circuit subjected to a specific availability
requirement. If a system is undergoing a scheduled maintenance regime, the residual life left
in the system after the detection of degradation needs to be calculated if we need to estimate
the point of failure. The paper discusses a framework for treating the degradation data of the
track circuit, based on inspection remarks, to calculate the residual life left in the track circuit.
In the case of the condition based maintenance regime, the paper estimates the probability of
degradation detection of the condition monitoring device to achieve a certain availability
target. The data used in the paper are taken from a specific line section of Banverket.
Paper IV shows that achieving the optimal cost effectiveness is one of the significant ways to
address the efficiency of a system, and involves maximising the availability and minimising
the life cycle cost of the system over the system life cycle. One of the important ways of
maximising the cost effectiveness of the system is to optimise the maintenance policy. This
paper demonstrates the estimation of the cost effectiveness of an ERTMS (European Rail
Traffic Management System). The degradation and repair process of the system is modelled
by Petri-Nets. The model considers systems that experience degradations and are subjected to
imperfect maintenance. The results show the effects of the maintenance factor, detectability,
inspection interval and deferred maintenance time on the cost effectiveness of the system.
Paper V presents a methodology for estimation of the uncertainty linked with LCC, by a
combination of design of experiment (DoE) and Monte Carlo simulation. The proposed
methodology is illustrated by a case study of Banverket. LCC is being used as a tool to help in
making effective maintenance decisions. However, there are various uncertainties associated
with estimation of the LCC. The paper investigates more the uncertainties caused by technical
parameters, i.e. reliability and maintainability parameters. The uncertainty in reliability and
maintainability parameters exist because of their probabilistic nature, which contributes to the
uncertainty in LCC estimation. The simulations are used to make the deterministic LCC
equations probabilistic. DoE is applied to provide guidance as to how the R&M parameters
should be varied in a systematic way. The paper also illustrates cost models for different
maintenance and renewal actions carried out on track.
42
6 Discussion and Conclusions

In this thesis the research emphasis is placed on the applicability of RAMS and life cycle cost
analysis to the development of maintenance decision models for railway infrastructure. The
thesis also examines how different maintenance strategies affect the RAMS requirements of
the infrastructure. The overall goal is to achieve the RAMS levels stipulated for the railway
infrastructure at a lower maintenance cost. The thesis demonstrates approaches and models
for achieving a more cost effective maintenance strategy for the railway infrastructure.
Railway infrastructure is a complex system. An important aspect of the rail infrastructure is
that the assets have a long useful life. Consequently, once they are installed, it is very difficult
to modify the initial design. Therefore, the performance of the infrastructure depends on the
maintenance and renewal decisions taken during its life cycle. In many countries, the
restructuring of railways and increasing efficiency requirements are causing a changing
environment for infrastructure management. The responsibility for parts of the railway system
is often handed over to different players. In order to guarantee optimal long-term results for
the railway systems, the effects of decisions should be systematically evaluated. The
infrastructure manager, who is responsible for the design, construction, maintenance, renewal
and upgrading of the infrastructure, has a clearly defined role and is confronted with the
increasing performance of its associated partners. Due to increases in the operation and
maintenance costs, infrastructure managers are compelled to optimise their budget, while
reliability and availability have to be increased without endangering traffic safety. A
systematic approach is needed by the infrastructure manager for guaranteeing the defined
levels of performance. Since, in the current scenario, most of the maintenance and renewal
decisions are based on past experience and expert estimations, a need for an LCC approach
arises. A life cycle costing approach combined with RAMS analysis will provide a way to
optimise the maintenance strategy, considering the short term budget requirements as well as
the long term costs of ownership. As discussed earlier, efficient maintenance decisions always
try to achieve the RAMS targets of the system. Appropriate estimation of the RAMS target
for the infrastructure is important, as it influences the maintenance strategy as well as the
maintenance investment over a period of time. Fig. 6.1 illustrates the RAMS parameters on
different hierarchy levels. The RAMS hierarchy has been divided into three levels, i.e. the
infrastructure level, system level and component level.
43
The RAMS parameters indicate the business characteristics of the infrastructure at the
infrastructure level, the technical characteristics at the system level (e.g. track system,
signalling system, etc.) and the failure characteristics at the component level (e.g. rail, track
circuit, etc.). Thus, the availability and safety targets are generally estimated at the
infrastructure level, whereas the reliability and maintainability targets are estimated at lower
hierarchy levels.
Infrastructure Level
Train delay
Service reliability
Number of cancellations/speed restrictions/derailments
Mean Time Between Failures
Mean Time To Repair
System availability
Mean Time Between Service Affecting Failures
System Level
Component Level
Failure probabilities based on failure modes
Safety related failure probabilities
Figure 6.1: RAMS hierarchy within track infrastructure

This necessitates the estimation of the availability and safety targets to achieve the required
quality level of the infrastructure. An approach to estimating the availability target for the
infrastructure is proposed in Paper I. Paper I deals with the estimation of the availability target
of the railway infrastructure to achieve the capacity and punctuality requirements. An
example delineating specific design and operating conditions of the infrastructure is presented
to estimate the availability target using Petri-Net models. The relationship between
availability, capacity and punctuality will help infrastructure managers to identify the
availability target for the specific railway infrastructure section when the punctuality and
capacity requirements are known. The paper also outlines the operating and design parameters
along with the reliability and maintainability parameters that need to be considered when
developing this relationship. The results are very dependent on the timetable for the trains in
place. A proper setting of the availability target will allow the infrastructure manager to avoid
penalties due to delay and lower capacity. In this context, another tool of interest is TrainPlan
(see Nilsson, 2006). The system TrainPlan is a tool for planning timetables and available
resources and is used for long time timetable construction. The system contains information
about each train, such as, train number, description of the train, departure time, arrival time,
etc. However, this tool does not calculate reliability or availability of the system explicitly.
In paper I the authors have defined railway capacity under three categories: inherent, achieved
and operational capacity. The inherent capacity is the maximum capacity that a railway
network can achieve. It is the number of trains that could run over a line or route, during a
specific time interval, in a strictly perfect environment, with the trains running permanently
44
Discussion and Conclusions
and ideally at minimum headway. The inherent capacity is based on the infrastructure design.
The achieved capacity is calculated under more realistic assumptions, which are related to the
level of expected punctuality. It is the capacity that can permanently be provided under
normal operating conditions. It is usually around 6075% of the inherent capacity (UIC,
2004). Banverket (Swedish National Rail Administration) (Banverket, 2001) indicates a lack
of capacity when the capacity utilization is above 80%, as higher capacity utilization leads to
more delays of unexpected durations. The achieved capacity is the most significant measure
of the track capacity, since it relates the ability of a specific combination of infrastructure,
traffic, and operations to move the largest volume within an expected service level. The
service level represents the punctuality level for the infrastructure based on a specified time
table. If we change the time table, the achieved capacity also changes. The achieved capacity
can be defined as the maximum capacity for a specified time table with defined operational
headway where as the inherent capacity is the maximum capacity when there is no time table
in place and the trains run at minimum (safety) headway. The operational capacity is less than
the achieved capacity. This is the case if there is a prolonged shortage of facilities, e.g. due to
accidents or weather conditions, but more generally due to failures in the infrastructure which
disrupts the train operations.
The safety of the infrastructure is defined at the infrastructure level. Traditionally the safety
performance of the railway infrastructure is measured as the number of accidents/million train
kilometres or the number of derailments/million kilometres of track. Broadly, performance
indicators are classified as leading or lagging indicators. A leading, lead, or prospective
indicator is a performance driver. The outcome measure itself is simply the lagging, lag, or
retrospective indicator. Leading and lagging indicators can also relate to strategy or goals, and
therefore it is important not to mix means and ends. These safety indicators are lagging
indicators which only represent the current safety level of the track. If the infrastructure
manager wants to improve the safety of the track in the future, it needs to have a lead
indicator, e.g. the probability of derailment. Paper II depicts a model for estimating the
probability of derailment for the railway track. The model primarily calculates the probability
of undetected rail breaks and the probability of the track quality level falling below the
maintenance limit. The estimated probabilities are age-dependant i.e. dependant on the
tonnage that has passed over the track. The model also considers the influence of the
inspection interval on the probabilities. If the infrastructure manager wants to achieve a
certain safety performance or target on the aging track at some point of time, this model can
compute the amount of investment in inspection that the infrastructure manager needs to
make. Paper I and II have answered the first research question.
RAMS analysis can further be applied to the effective maintenance planning of the railway
infrastructure to achieve the safety and availability targets. The effect of the inspection
interval on the safety level of an aging infrastructure is demonstrated in Paper II. Paper III
discusses a model for estimating the availability of the track circuit in two maintenance
regimes, i.e. scheduled maintenance and time based maintenance. The model provides a
comparison of system availability between the two maintenance regimes. A reduction of the
45
unavailability caused by track circuit failures must be addressed using an assured

methodology. This methodology usually requires either an optimisation or a complete
overhaul of the maintenance processes in place. This starts with understanding the causes of
track circuit failures, which can either be within the track circuit itself, e.g. component or
systemic failure, or result from a failure of part of the infrastructure, such as poor track or
related components. If a system is undergoing a scheduled maintenance regime, the residual
life left in the system after the detection of degradation needs to be calculated if we need to
estimate the point of failure. A framework for treating the degradation and failure data of the
track circuits is presented in Paper IV. The paper demonstrates the effect of the inspection
interval on the system availability in the case of the scheduled maintenance regime. If an
infrastructure manager is following a scheduled maintenance regime, this model will be
helpful in estimating the optimal inspection interval for achieving the specified availability
target described in Paper I. A similar model has been developed in Paper IV to show the
effect of the detection probability of the condition monitoring device on the system
availability. Paper II and III have answered the second research question.
As described in Figure 2.7, different maintenance strategies can achieve the RAMS targets
specified for the infrastructure. However, the cost effective maintenance strategy is the one
that provides the lowest life cycle cost. The application of both RAMS and LCC analyses to
realize a cost effective maintenance policy is depicted in Paper V in the form of a case study.
The third research question is answered in Paper IV. A Petri-Net model is developed to
estimate the cost effectiveness of the system. The model considers the different degradation
and maintenance states of the system. An imperfect maintenance scenario is considered in this
model. Kijima models are used to estimate the virtual age of the system in the case of an
imperfect maintenance. Sensitivity analysis has been performed on different maintenance
parameters to maximise the cost effectiveness of the system. The model focuses on achieving
higher availability targets at a lower life cycle cost. The estimation of the life cycle cost
should always consider the uncertainties associated with it.
An approach to estimating the uncertainties connected with the life cycle cost is described in
Paper V. The paper deals with two different levels of uncertainty associated with the LCC of
the track infrastructure. The level I uncertainty concerns costs due to penalties imposed by
traffic operators on the infrastructure manager due to such factors as train delay, traffic
disruption, or derailment. These anomalies can be caused by planned or unplanned
maintenance actions, as well as a lack of necessary maintenance. Hence, the resulting costs
are related to decisions about maintenance actions and can be estimated by the probabilistic
assessment of train delay, derailment, or traffic disruptions, considering the technical and
operational characteristics of the track, as well as the maintenance actions. The level I
uncertainty can also be viewed as belonging to the external risk of the LCC analysis, where
the costs should be included to make the LCC analysis more effective. However, there is also
the level II uncertainty, which is the internal risk associated with LCC. The level II
uncertainty pertains to the variable contribution to the total LCC originating from the
uncertainty in the reliability and maintainability parameters. For better estimation of
46
Discussion and Conclusions
uncertainty in LCC, this paper outlines a methodology based on a combination of Monte

Carlo simulation and DoE. This combination provides the possibility of identifying
parameters that are influential on the LCC estimation and its variability. The uncertainty in
the LCC is presented as variable costs with associated distributions. Paper V has answered the
fourth research question. When the variable costs are added to the LCC, it becomes more
robust. Hence, the decision-makers are helped in making more effective decisions about
maintenance policy by considering the LCC. The approach and models developed by this
research can be adapted to other systems in the railway infrastructure with suitable
modifications.
6.1 Research Contributions

The research presented in this doctoral thesis has focussed on the applicability of RAMS and
LCC analysis to the development of maintenance policies for the railway infrastructure. The
literature study shows that the use of RAMS and LCC methodologies for railway
infrastrcuture is still in its initial phase. Furthermore, it has been found that these
methodologies have a limited use in developing a maintenance policy for the infrastructure.
The research contributions can be listed as follows:
1. The development of an approach to estimate the availability target for the railway
infrastructure considering capacity and punctuality requirements (Paper I)
2. The development of a model to assess the safety level of the infrastructure and
estimate the maintenance investments based on the safety target (Paper II)
3. The development of models to estimate the availability of DC track circuit subjected
to different maintenance policies and predict the residual life of the system (Paper III)
4. The development of a model to estimate the cost effective maintenance policy for a
signalling system subjected to degradation and imperfect maintenance (Paper IV)
5. The development of an approach to determine the uncertainty in LCC estimations
(Paper V)
47
6.2 Scope of the Future Research

In summary, based on the research conducted, the following areas are suitable for further
research:
x
The development of a more robust model for railway infrastructure addressing the
interactive effects of the different sub-systems in the infrastructure. On the
infrastructure level, the availability and safety are dependent upon the availability and
safety of the different sub-systems of the infrastructure. The degradations and
maintenance of one sub-system affect the RAMS characteristics of the other subsystems. In order to achieve the RAMS target at the infrastructure level, it is necessary
to investigate the interactive effects of the different sub-systems.
The implementation of the models on the different systems of the railway

infrastructure. The thesis presents different approaches and models for the
maintenance decision support of the railway infrastructure. These models can be
implemented on the different systems of the infrastructure to develop maintenance
policies.
48
References
Andersson, M. (2002). Strategic Planning of Track Maintenance - State of the Art, Research
Report, Division of Urban Studies, Department of Infrastructure, Royal Institute of
Technology (KTH), ISSN: 1651-0216.
Ascher, H. and Feingold, H. (1984). Repairable system reliability: Modelling, Inference,
Misconceptions and Their Causes, Marcel Dekker Inc., New York, ISBN 0-8247-7276-8.
Asiedu, Y. and Gu, P. (1998). Product life cycle cost analysis: state of the art review,
International Journal of Production Research, 36(4), 883-908.
Blanchard, B.S. and Fabrycky, W.J. (1998). Systems Engineering and Analysis, 3rd ed., Upper
Saddle River, NJ: Prentice-Hall.
Blischke, W.R. and Murthy, D.N.P. (2003). Case Studies in Reliability and Maintenance,
John Wiley & Sons, USA.
Burstrm, B., Ericsson, G. and Kjellsson, U. (1994). Verification of Life-Cycle Cost and
reliability for the Swedish high speed train X2000, In Proceedings of Annual Reliability and
Maintainability Symposium(RAMS), Los Angeles, 24-27 January, 166-171
Banverket, (2001). Calculation manual: Aid for cost-benefit analysis in the railway sector,
Handbook BVH 706.00, Banverket, Borlnge, (in Swedish).
Cannon, D.F., Edel, K.O., Grassie, S.L. and Sawley, K. (2003). Rail defects: an overview,
Journal of Fatigue & Fracture of Engineering Materials & Structures, 26, 865-886.
Coleman, D.E. and Montgomery, D.C. (1993). A Systematic Approach to Planning for a
Designed Industrial Experiment, Techometrics, 35(1), 1-27.
Cooper, D.R. and Schindler, P.S. (2006). Business Research Methods, 9th ed., McGraw-Hill
Companies, Inc., Singapore, ISBN 0-07-124430-1.
49
CRMA (1998). Cost, Reliability, Maintenance, and Availability, Final Summary Report
Available Online: http://cordis.europa.eu/transport/src/crmarep.htm#1
Accessed on: 20/10/2007.
Dane, F.C. (1990). Research method, Brooks/Cole Publishing Company, California.
Diamond, S. and Wolf, E. (2002). Transportation for the 21st century, TracGlide Top-of-Rail
Lubrication System, Report from Department of Energy, USA.
EN 50126 (1999) The Specification and Demonstration of Reliability, Availability,
Maintainability and Safety (RAMS) for Railway Applications, European Committee for
Electrotechnical Standardization (CENELEC), Brussels, Belgium.
Esveld, C. (2001). Modern railway track, 2nd ed., Delft: MRT-Productions, Germany.
Fazio, A.E. and Prybella, R. (1980). Development of an Analytical Approach to Track
Maintenance Planning, Transportation Research Record 744, National Research Council,
National Academy Press, Washington, D.C., USA, 46-52.
Ferreira L. (1997). Rail track infrastructure ownership: investment and operational issues,
Transportation, 24(2), 183-200.
IEC 62278 (2002). Railway applications - specification and demonstration of reliability,
availability, maintainability and safety (RAMS), International Electrotechnical Commission,
Geneva.
IEV 191-01-07 (2007). Dependability and Quality of Service/Item Related Performance,
International Electrotechnical Vocabulary, Online Database.
Available online: http://std.iec.ch/iec60050
IMPROVERAIL (2003). Improved Tools for Railway Capacity and Access Management
Available Online: http://ec.europa.eu/transport/rail/research/improverail_en.htm
INNOTRACK (2009). Innovative Track Systems
Available Online: http://www.innotrack.eu/
Jovanovic, S. (2006). Railway Track Quality Assessment and Related Decision Making, In
Proceedings of AREMA annual conferences, Kentucky, 17-20 September.
50
References
Kalousek, J., Sroba, P. and Hegelund, C. (1989). Analysis of rail grinding tests and
implications for corrective and preventative grinding, In Proceedings of the 4th International
Heavy Haul Railway Conference, Brisbane, 11-15 September, 193-204.
Karlsson, V. (2005). An overall view of maintenance, European Railway Review, 11(3), 1117.
Kijima, M. (1989). Some results for repairable systems with general repair, Journal of
Applied Probability, 26, 89-102.
Kumar, S. (2008). A Study of the Rail Degradation Process to Predict Rail Breaks, PhD
Thesis, Lule University of Technology, Lule, Sweden, ISSN: 1402-1544; 2008:15.
Kumar, U. and Akersten, P A. (2008). Availability and maintainability, In Encyclopedia of
Quantitative Risk Analysis and Assessment, ed. / E. Melnik and B. Everitt, Wiley, 77-84.
Kumar, U. (1989). Maintenance Strategy for Mechanized Mining Systems, Licentiate Thesis,
Lule University of Technology, Lule, Sweden, ISSN: 0280-8242; 1989:01L.
Larsson, D. (2002). Verification of the DeCoTrack: a prediction model for degradation cost
of track, Research Report, Lule University of Technology, Lule, Sweden, ISSN: 14021528; 2002:18.
LibeRTiN (2005). Light rail thematic network
Available online: http://www.libertin.info/
LICB (2007). Lasting infrastructure cost benchmarking
Available online: http://www.uic.asso.fr/infra/article.php3?id_article=13
Madu, C.N. (2005). Strategic value of reliability and maintainability management,
International Journal of Quality and Reliability Management, 22 (3), 317-328.
Magel, E. and Sroba, P. (2007). Workshop on Rail Grinding Practices, International Heavy
Haul Association (IHHA), Kiruna, Sweden, 8-10 June.
Markeset, T. and Kumar, U. (2004). Dimensioning of Product Support: Issues, Challenges,
and Opportunities, In Proceedings of Annual Reliability and Maintainability Symposium
(RAMS), Los Angeles, 26-29 January, 565- 570
51
Markeset, T. and Kumar, U. (2003). Integration of RAMS and risk analysis in product design
and development work processes: a case study, Journal of Quality in Maintenance
Engineering, 9(4), 393-410.
MODTRAIN (2007). Innovative Modular Vehicle Concepts for an Integrated European
Railway System
Available Online: http://www.modtrain.com/
Neuman, W.L. (2003). Social Research Method, 5th ed., USA.
Nilsson, K. (2006). Alternativa system fr tgstyrning p Malmbanan, Forskningsrapport,
Avdelningen fr industriell logistik, Lule Tekniska Universitet, ISSN:1402-1528; 2006:01,
(In Swedish).
Petri, C. (1962). Kommunikation mit Automaten, PhD thesis, University of Bonn, Germany,
(In German).
ProM@in (2003). Progress in Maintenance and Management of Infrastructure
Available Online: http://www.promain.org/
Putallaz, Y. (2003). Strategic M&R policy of a railway corridor taking into account the value
of capacity, In Proceedings of World Congress on Railway Research, Edinburgh, 28
September-1 October.
Reliasoft Corporation (2003). Weibull++6 software, version 6.0.11, web: www.weibull.com
REMAIN (1998). Modular System for Reliability and Maintainability Management in
European Rail Transport
Available Online: http://cordis.europa.eu/transport/src/remainrep.htm
Ringsberg, J.W. (2001). Life prediction of rolling contact fatigue crack initiation,
International Journal of Fatigue, 23, 575-586.
Stalder, O. (2001). The life cycle costs (LCC) of entire rail networks: an international
comparison, Rail International, 32 (4), 26-31
Sullivan, T.J. (2001). Methods of Social Research, Harcourt, Inc., USA, ISBN: 0-15-5074636.
52
References
Sumser, J.R. (2000). A Guide to Empirical Research in Communication, Sage Publications

Inc, ISBN: 0761922229.
Swier, J. (2004). Exposing the Infrastructure Cost Drivers, Railway Gazette International,
Sutton, Surrey, UK.
Swier, J. and Luiten, T. (2003). Outsourcing in The Netherlands; the sector process model,
Innovations for a Cost Effective Railway Track, ProM@in, 3, 20-26.
UIC (2004). Capacity, UIC Leaflet 406, International Union of Railways, Paris.
UNIFE (2001). Life Cycle Cost Working Group
Available Online: http://www.unife.org/workgroups/lcc_2.htm
URBAN TRACK (2007). Urban Rail Infrastructure in a harmonised Europe
Available Online: http://www.urbantrack.eu/
Vatn, J. (2002). A Life Cycle Cost model for prioritisation of track maintenance and renewal,
Innovations for a Cost Effective Railway Track, Progress in Maintenance and Management of
Infrastructure (ProM@in), 2, 21-25.
Wilson, A. (1999). Asset Maintenance Management, Alden Press, Oxford, UK.
Veit P. (2003). Some thoughts concerning outsourcing of track maintenance, Innovations for
a Cost Effective Railway Track, ProM@in, 3, 14-19.
Zarembski, A.M. (1998). Development and Implementation of Integrated Maintenance
Planning Systems, Transportation Research Board Annual Meeting, Washington D.C. U.S.
Zhao, J., Chan, A.H.C., Roberts, C. and Stirling, A.B. (2006). Assessing the Economic Life of
Rail Using a Stochastic Analysis of Failures, In Proceedings of the Institution of Mechanical
Engineers, Part F: Journal of Rail and Rapid Transit, 220 (2), 103-111.
Zoeteman, A. (2006). Asset maintenance management: state of the art in the European
railways, International Journal of Critical Infrastructures, 2(2/3), 171-186.
Zoeteman, A. (2001). Life cycle cost analysis for managing rail infrastructure, concept of a
decision support system for railway design and maintenance, European Journal of Transport
and Infrastructure Research, 1(4), 391 413.
53
Zoeteman, A. and Esveld, C. (1999). Evaluating Track Structures: Life Cycle Cost Analysis
as Structured Approach, In Proceedings of World Congress on Railway Research, Tokyo, 1923 October.
54
APPENDED PAPERS
Paper I
Availability target of the railway infrastructure: an

analysis
Patra, A. P., Kumar, U. and Larsson-Krik, P-O. (2010). Availability target

of the railway infrastructure: an analysis. Accepted for publication in
Proceedings of Reliability and Maintainability Symposium (RAMS), San Jose,
USA, 25-28 January.
Availability target of the railway infrastructure: an analysis

Ambika P. Patra, MTech, Lule University of Technology
Uday Kumar, PhD, Lule University of Technology
Per-Olof Larsson Krik, PhD, Swedish National Rail Administration (Banverket)
Key Words: Railway Infrastructure, Reliability, Maintainability, Availability, Capacity, Punctuality
the railway sector. The business needs of railway
infrastructure can be defined as lower ownership costs,
interoperability, enhanced safety, improved punctuality,
increased capacity and reduced journey times. The availability
of railway infrastructure plays a significant role in attaining a
higher capacity and punctuality level of the infrastructure. The
required level of availability determines the amount of
maintenance investment in the infrastructure over a period of
time. A higher availability target requires the effective
operation and maintenance of infrastructure, often
necessitating the implementation of cost-effective preventive
maintenance strategies in combination with effective supply
chain management. However, the question of setting the
availability target for the infrastructure is not easy, as it
involves many decision parameters, apart from a good
understanding of the network configuration and traffic density.
Railway networks that have a smaller number of trains and a
low punctuality requirement do not require higher availability
targets. In order to estimate the availability target, it is
necessary to understand the capacity and punctuality
requirements of the railway infrastructure. Section 2 illustrates
the fundamentals of capacity and punctuality. A model has
been developed using Petri-Nets to establish a three-way
relationship between availability, capacity and punctuality.
Section 3 describes the model with an example. Finally
discussions and conclusions are described in Section 4.
SUMMARY & CONCLUSIONS

The railway has been accepted as one of the most
environmentally friendly modes of transport for goods and
passengers. However, the railway sector is striving to increase
its capacity to meet the growing demand for the transport of
goods and passengers with a high level of punctuality in its
services. Higher availability requires the effective operation
and maintenance of infrastructure, often necessitating the
implementation of cost-effective preventive maintenance
strategies. Therefore, a higher availability target means higher
maintenance investment. However, the question of setting the
availability target for the infrastructure is not easy, as it
involves many influencing decision parameters, apart from a
good understanding of the network configuration and traffic
density. Railway networks that have a smaller number of
trains and a low punctuality requirement do not require higher
availability targets. The aim of this paper is to estimate the
availability target for railway infrastructure based on the
capacity and punctuality requirements of infrastructure
managers and train operating companies. The objectives of the
paper are to develop an approach to i) estimating the capacity
of the infrastructure based on the design and operational
characteristics and evaluating the influence of infrastructure
availability on the required capacity and ii) estimating the
volume of primary and secondary delay due to failures and
maintenance of the infrastructure and establishing the
relationship
between
availability
and
punctuality
requirements. To achieve these objectives, an example is
presented with parameters drawn from failure, maintenance
and traffic data. Finally, a model has been developed in PetriNets to establish a relationship between availability, capacity
and punctuality. Monte Carlo simulation is used to establish
the relationship. The simulation results illustrate the effect of
infrastructure availability on train delays and capacity.
2 RAILWAY CAPACITY AND PUNCTUALITY

An efficient utilization of the existing railway infrastructure is
an essential component of a high-quality transportation system
and has become a central task for railway infrastructure
managers. Line capacity is, in essence, what the infrastructure
managers have to sell as their final product. Although capacity
seems to be a self-explanatory term in common language, its
scientific use may lead to substantial difficulties when it is
associated with objective and quantifiable measures. It is a
complex term that has numerous meanings and for which
numerous definitions have been given. In [1] it is stated that
capacity as such does not exist. Railway infrastructure
capacity depends on the way in which it is utilized. However,
in [2] it is stated that capacity is a measure of the ability to
move a specific amount of traffic over a defined rail line with
a given set of resources under a specific service plan. As
illustrated in Figure 1, capacity is a balanced mix of the
number of trains, the stability of the timetable, the average
1 INTRODUCTION
Rail traffic is the most important form of public traffic in
Europe as the density of the railway network is very high
compared to the other parts of the globe. To be in competition
with other modes of transportation, railway traffic must be
quick, comfortable, cheap and primarily safe. There have been
contractual agreements concerning the targeted level of
reliability and punctuality in the performance regime within
speed achieved and the heterogeneity (mixed traffic with

different train speeds) of the train system. It is, for instance,
possible to achieve a high average speed on a railway network
by having a high heterogeneity a mix of fast and slower
trains. However, the cost of maintaining a high average speed
with a high heterogeneity makes it difficult to run a great
number of trains with a high stability (punctuality) than if all
the trains ran at the same speed. If one wants to run more
trains, it is necessary to operate with less mixed traffic and
thereby have a lower average speed, as in the case of metro
systems.
the train operations.

Railway capacity very much depends on the headway
time between the trains, i.e. both the safety headway and the
operational headway time. Figure 2 illustrates the safety
headway time between the trains. It is dependant on the
distance that the trains maintain for safe operations on the
track. As shown in the figure, the safety headway time is the
summation of the travel time, braking time, release time and
operating time. The travel time is the time required to cover
the distance between two signals. The area between two
signals is called a block and is controlled by a track circuit.
Number of
trains
Signals
Train
Average
speed
Stability
Operating Time
Braking Time
Travel Time
Release Time
Mixed-train operation
Metro-train operation
Safety Headway Time
Heterogeneity
Figure 2 Safety headway diagram
Figure 1 Capacity balance [1]
At any given time only one train can occupy a block

section. The travel time depends upon the distance between
the signals and the speed of the trains. The braking time
depends on the braking distance i.e. the distance required to
stop before a signal. It is calculated by considering the train
speed and deceleration. The release time is the time required
for the entire length of train to cross the signal. This depends
on the length of the train and the speed of the train. The
operating time is a safety time and is fixed by the
infrastructure managers. The inherent capacity of a double
track line depends on the safety headway between the trains.
For example, if the safety headway is five minutes, the
inherent capacity per track section will be 12 trains per hour.
However, as discussed earlier infrastructure managers
consider buffer time to accommodate delays. Moreover, the
requirements of stakeholders set the operational headway time
between the trains more than the safety headway. The
operational headway is the actual time between two
consecutive trains as per the train timetable.
Punctuality is defined differently by different
infrastructure managers across the globe. A train in Sweden is
considered punctual if it is less than five minutes off schedule
at a station, otherwise it is delayed. Train delays may be
classified into two major categories: primary delays and
secondary delays. Primary delays are the delays undergone by
the trains passing over a disturbed track section. Primary
delays are the time differences between the normal and the
disrupted journey. Secondary delays are delays of follower
trains, which will not undergo the totality of primary delay,
but which will undergo a delay because the previous train is
delayed. This kind of delay happens when a failure is close to
being restored. The principles of primary and secondary delay
are illustrated in Section 3.
Railway capacity has different values based on different

criteria. In this paper the authors have defined railway
capacity under three categories: inherent, achieved and
operational capacity. The inherent capacity is the maximum
capacity that a railway network can achieve. It is the number
of trains that could run over a line or route, during a specific
time interval, in a strictly perfect environment, with the trains
running permanently and ideally at minimum headway. The
inherent capacity is based on the infrastructure design. The
achieved capacity is calculated under more realistic
assumptions, which are related to the level of expected
punctuality. It is the capacity that can permanently be
provided under normal operating conditions. It is usually
around 6075% of the inherent capacity [1]. Banverket
(Swedish National Rail Administration) [3] indicates a lack of
capacity when the capacity utilization is above 80%, as higher
capacity utilization leads to more delays of unexpected
durations. The achieved capacity is the most significant
measure of the track capacity, since it relates the ability of a
specific combination of infrastructure, traffic, and operations
to move the largest volume within an expected service level.
The service level represents the punctuality level for the
infrastructure based on a specified time table. If we change the
time table, the achieved capacity also changes. The achieved
capacity can be defined as the maximum capacity for a
specified time table with defined operational headway where
as the inherent capacity is the maximum capacity when there
is no time table in place and the trains run at minimum (safety)
headway. The operational capacity is less than the achieved
capacity. This is the case if there is a prolonged shortage of
facilities, e.g. due to accidents or weather conditions, but more
generally due to failures in the infrastructure which disrupts
3 AVAILABILITY TARGET ESTIMATION
Position on
track
Station 1
Time
Operational
headway
Mean Down Time
Figure 4 Illustration of primary delay

Position on
track
Speed = Average
speed
Acceleration
Deceleration
Speed
Affected length
of
track due to
failure
Station 2
Speed = Average
speed
Primary delay
Station 2
Railway infrastructure consists of various sub-systems,

such as the track system, the signalling and telecommunication
system, and the power system. Each of these sub-systems
contributes to the infrastructure availability. As discussed
earlier, the railway infrastructure availability influences the
capacity and punctuality of train operations. Therefore, when
estimating the availability target, the capacity and punctuality
requirements of the infrastructure must be considered.
Different failure modes in the railway infrastructure induce
different amounts of delay in railway network based on speed
restrictions. The amount of delay depends on the occurrence
rates and repair times of the failure modes. Figure 3 depicts
the speed profile of trains due to a track circuit failure. A
failure in a track circuit turns the signal red for that particular
block and trains pass at a reduced speed over that block until
the failure is rectified. This section develops a model which
estimates the availability target. An example is provided to
illustrate the model.
Primary delay
1st Secondary delay
2nd Secondary
delay
Speed =Restricted
speed
(Track circuit failure is

being repaired)
Speed =Restricted
speed
Safety
headway
Station 1
Time
Figure 3 Speed profile of trains due to failure of a track

circuit
3.1 Example
Affected length
of
track due to
failure
Time
Operational
headway
Let us consider a double track railway line between two main

stations. This line has multiple intermediate stations. Trains
run with a uniform operational headway (OH) of 15 minutes
and a safety headway (SH) of 5 minutes. All the trains that run
on the track have the same speed pattern. As discussed earlier,
the capacity and punctuality will be estimated for the line
section between two adjacent stations. We consider the
failures of three sub-systems that occur in this specific line
section. The reliability and maintainability details of these
sub-systems are given in Table 1. Occurrences of failures
induce primary and secondary delay in the railway network as
illustrated in Figure 4 & 5. When a failure occurs, trains
reduce their speeds over the affected area (see Figure 3) and
arrive late compared to their specified arrival time. The time
difference determines the primary delay (PD). It is calculated
by the kinematics equations of motion considering distance,
acceleration, deceleration and speed. The number of trains that
will be disrupted by primary delay is given by:
Figure 5 Illustration of secondary delay

Secondary delay will occur if primary delay >
(Operational headway Safety headway). In this case the
following train will have to slow down to keep the minimal
distance (SH) from the last primarily delayed train. The
secondary delay undergone by the 1st following train is given
by:
SD1 = PD (OH SH)
If SD1 > (OH SH), the 2nd following train will be subjected
to secondary delay. The secondary delay undergone by the 2nd
following train is given by:
SD2 = SD1 (OH SH) = PD 2*(OH SH)
Secondary delays can be more generally expressed as:
SDi = PD i*(OH SH)
The number of trains that will be disrupted by secondary delay
is given by:
SPD = Primary delay/(Operational headway Safety headway)
NPD = Mean Down Time (MDT)/Operational headway (OH)
SPD is an integer.
NPD is an integer.
3.2 Petri-Net model for studying the relationship between

availability, capacity and punctuality
The Petri-Net model (see Figure 6) has been developed for
estimating the relationship between availability, capacity and
punctuality on the line between two stations for the example
described above. The reliability and maintainability
parameters for the model are given in Table 1. As illustrated in
Figure 6, places 1, 3 and 5 denote the working states of subsystems 1, 2 and 3 respectively where as places 2, 4 and 6
denote the failed states. The transitions between these places
operate according to the failure rates and mean down time of
the sub-systems. The failure rates in this case are assumed to
be following exponential distributions. Whenever any subsystem fails, the infrastructure goes to a failed state and it is
restored to the working state depending on the mean down
time of the sub-system. This is illustrated by places 12, 13 and
14. The firings of the transitions between these places occur at
any time depending on the failures of the sub-systems. It needs
to be mentioned that in this model we have assumed only three
sub-system failures that affect the capacity and punctuality of
the railway network. In other cases the number of sub-systems
can be more dependent on the specific railway track design
criteria.
Sub-system
Failure Rate-FR
(per minute)
Mean Down Time-MDT (in Primary Delay-PD (in

minutes)
minutes)
Sub-system 1
6.00E-05
75
12
Sub-system 2
1.00E-04
60
10
Sub-system 3
1.50E-04
45
Table 1 Reliability and Maintainability data for sub-systems

Places 7 to 9 describe the movement of trains between two
stations. The transitions between these places consider the
fundamentals of primary and secondary delay illustrated in
Section 3. Trains start at station 1 (place 7) and reach station 2
(place 9). Place 9 keeps account of all the primary and
secondary delay that the trains undergo in the case of subsystems failures. Places 10 and 11 calculate the capacity of
the track section between station 1 and 2 depending on the
arrival time of the trains at place 9. Place 10 calculates the
number of trains that reach station 2 every hour. In this model
we calculate the average capacity (trains/hour) of the track
section over a period of time.
Figure 6 Petri-Net model
The model also calculates the cumulative delay that the trains
can undergo over a period of time on that track section. It is
also considered that the trains take 25 minutes to travel
between the stations i.e. the travel time (TT) is 25 minutes.
The model will enable us to estimate the effects of the track
system availability on the capacity and punctuality of that
particular track section. The fundamentals of Petri-Net
modelling can be found in [5].
Trains are operated for 18 hours a day and preventive
maintenance is carried out during the rest six hours. Therefore,
the availability of infrastructure considers only corrective
maintenance. By performing Monte Carlo simulations on the
Petri-Net models, we obtained the capacity variations over a
period of one month (18x60 hours). Figure 7 illustrates the
estimated operational capacity over a period of one month.
Figure 8 Cumulative delay over a period of one month
Punctuality (Delay minutes)
700
600
500
400
300
200
100
Figure 9 Relation between capacity and punctuality with

availability
In this particular example the operational capacity is
almost equal to the achieved capacity for all values of
availability. There might be a case when we reduce the
availability to a much lower level then operational capacity
will fall below than the required capacity. However, in this
case we can estimate the availability target for the line section
e.g. if the delay requirement should be less than 250 minutes,
then our availability target should be 0.988. We estimated this
availability target for a line section between two adjacent
stations. In a railway network there can be many stations. The
highest availability target that we estimate for any section will
determine the availability target for the whole railway
network. When we determine the availability target for a
specific railway network, the infrastructure manager can
estimate the maintenance investment in that particular network
over a period of time to achieve that availability target.
Capacity
(Trains/hr)
3,9895
0,9951
0,9912
0,9895
Availability (%)
0,9874
0,9863
As shown in Figure 7, when a failure occurs, the train

operation is disrupted until the failure is corrected and hence
the capacity decreases. However, after the failure is corrected
the operational capacity of the track increases because the
delayed trains arrive in the same hour as the non-delayed
trains. The achieved capacity in normal operation is expected
to be 4 trains/hour as the operational headway is 15 minutes.
However, due to failures, the mean operational capacity over a
period of one month is 3.9895 trains/hour, which almost
satisfies the achieved capacity requirement. Similarly, the
cumulative delay that the trains will undergo over a period of
1 month is illustrated in Figure 8. The total delay occurring in
one month is 355.6 minutes, which comprises of both primary
delay and secondary delay. The average availability over that
period is estimated to be 0.9826. In order to estimate the target
availability, we need to perform sensitivity analysis on
capacity and punctuality by changing the values of
availability; i.e. changing the values of reliability and
maintainability.
A relationship between the capacity, punctuality and
availability is illustrated in Figure 9. As shown in the figure,
the delay decreases with increases in availability value.
However, in this particular case, the capacity remains constant
with the change in availability values. This is due to fact that
the timetable specified in the example absorbs all the delays
because of the buffer embedded into it.
0,9845
Figure 7 Operational capacity over a period of one month
0,974
0,9657
4 DISCUSSIONS
The increasing complexity of modern technical systems
has resulted in high reliability, maintainability and availability
requirements. These requirement need to be met by the system
owner to remain competitive. Setting these requirements is
difficult when the systems have many stakeholders. Railway
infrastructure is one of these systems. The system availability
of the railway infrastructure directly affects the punctuality
and capacity of the railway network. Failure to meet the
capacity and punctuality requirements of the railway network
incurs a penalty for the infrastructure manager who owns the
system. Therefore, the infrastructure manager needs to
estimate the availability target that it intends to achieve so as
to meet the capacity and punctuality requirements. The
availability target also enables the infrastructure manger to
estimate the maintenance investment over a period of time. In
this paper we have developed a model for estimating the
availability target of the railway infrastructure in Petri-Net.
The model considers the design, operation and failures of the
infrastructure to derive a relationship between availability,
capacity and punctuality. The relation is derived by means of
an example presented in this paper. When the capacity and
punctuality requirements are known, the infrastructure
manager can estimate the availability target for this particular
scenario explained in the example. However, the results show
that the operational capacity does not change with the change
in the system availability. This is due to the fact that the
operational headway is kept large enough in the train timetable
to absorb the delays, to keep the operational capacity of the
network close to the achieved capacity. In general, if the
operational availability changes, the operational capacity has
to change. It actually does change for the hour during which a
failure occurs (see Fig. 7); but during the next hour, the
operational capacity increases as the delayed trains arrive
along with the scheduled trains and we estimate the
operational capacity as the average of all the hours. However,
if we lower the operational headway, we can observe that the
operational capacity will change with the change in system
availability. This model will help the infrastructure manager to
estimate the availability target of the railway infrastructure
based on the capacity and punctuality requirements. A proper
setting of the availability target will allow the infrastructure
manager to avoid penalties due to delay and lower capacity.
This will also help to estimate the total maintenance
investment that the infrastructure manager needs to make over
a period of time.
4.
5.
BIOGRAPHIES
Ambika Prasad Patra
Lule Railway Research Center
SE-97187 Lule, Sweden
e-mail: ambika.patra@ltu.se
Ambika Prasad Patra is a PhD candidate at the Division of
Operation and Maintenance Engineering, Lule University of
Technology. His research topics are RAMS and LCC analyses
for railway systems. He received his MTech degree in 2002 in
Reliability Engineering from the Indian Institute of
Technology, Bombay, India.
Uday Kumar, PhD
Lule Railway Research Center
SE-97187 Lule, Sweden
e-mail: uday.kumar@ltu.se
Dr Uday Kumar is Professor of Operation and Maintenance
Engineering at Lule University of Technology, Lule,
Sweden. He also is the Director of Lule Railway Research
Centre (www.jvtc.ltu.se) at the same university. His research
interests are maintenance engineering and management,
reliability and maintainability analysis, product support, life
cycle costing, risk analysis, system analysis, etc. He is also a
member of the editorial board and a reviewer for many
international journals. He has published more than 150 papers
in international journals and conference proceedings.
Per-Olof Larsson Krik, PhD
Swedish National Rail Administration,
SE-97102, Lule, Sweden
e-mail: per-olof.larsson-kraik@banverket.se
Dr Per-Olof Larsson-Krik co-ordinates research and
development activities for Banverket (the Swedish National
Rail Administration). Dr Larsson-Krik has written over 50
articles and case studies in the areas of tribology, reliability,
maintainability, logistics, and supportability. He has carried
out several research and consultancy projects for several
organizations in Europe and Australia. Dr Larsson-Kriks
research interests include systems engineering, rail wheel
(w/r) interaction, w/r lubrication and strategic operation and
maintenance decision making in an organisation.
REFERENCES
1.
2.
3.
analysis in the railway sector, Handbook BVH 706.00,

Banverket, Borlnge, 2001 (in Swedish)
B. Nystrm, Use of Availability Concepts in the Railway
System, International Journal of Performability
Engineering, Vol. 5(2), 2009, pp. 103-118
T. Murata, Petri Nets: Properties, Analysis and
Applications, Proceedings of the IEEE, Vol. 77(4), 1989,
pp. 541 580.
UIC Capacity, UIC Leaflet 406, International Union of

Railways, Paris, 2004
M. Abril, F. Barber, L. Ingolotti, M. A. Salido, P.
Tormos, and A. Lova, An assessment of railway
capacity, Transport Research Part E: Logistics an
Transportation Review, (Sept) 2008, pp. 774-806
Banverket, Calculation manual: Aid for cost-benefit
Paper II
Assessment and improvement of railway track safety
Patra, A. P., Kumar, U. and Larsson-Krik, P-O. (2009). Assessment and

improvement of railway track safety. Proceedings of 9th International Heavy
Haul Conference (IHHA), Shanghai, China, 22-24 June.
Assessment and Improvement of Railway Track Safety

Ambika Prasad Patra1*, Uday Kumar1, P-O Larsson-Krik2
1
Lule Railway Research Center, SE-97187, Lule, Sweden
2
Swedish National Rail Administration, SE-97102, Lule, Sweden
E-mail: ambika.patra@ltu.se, uday.kumar@ltu.se, per-olof.larsson-kraik@banverket.se
Summary: In this paper, an approach has been developed to assess safety of the railway track by
estimating the probability of derailment. Models for probability of derailment are developed based
on undetected rail breaks and poor track quality using Petri-Nets and Monte Carlo simulations. The
Effect of inspection intervals on lowering the probability has been analysed. The performance of
the model is illustrated by an example from a track section of the iron ore line of Banverket
(Swedish National Rail Administration).
Index Terms: Railway track safety, Maintenance, Petri-Net modelling
1.
INTRODUCTION
Safety is the most important attribute of railway

quality of service and operation. Infrastructure
managers always try to reduce the number of
potential risks areas that can lead to train
accidents. Railway operations, that do not pay
attention to reduce potential risks, face severe
consequences. Risk has multi-fold consequences
i.e., high society cost for compensation of deaths
and injuries, material damages to infrastructure
and rolling stock, cost of delay due to accidents
and cost of damage to the environment. Proper
maintenance planning helps infrastructure
managers to reduce the potential risks. To study
the effect of maintenance a case study was
developed for a specific track section of the iron
ore line of Banverket (Swedish National Rail
Administration).The studied ore line Malmbanan,
completed and operational by 1902, was later
electrified with 15 kV 16 2/3 Hz, completed on
19th of January 1915, and remains so to this day.
The studied track is a 1435 mm gauge. The track
configuration is an electrified single track using
block system. The signalling system (Low
voltage DC track circuit), with its traffic control
safety mechanisms, will detect any deviation that
could be linked to a rail failure. However, the
signalling system is not used as a maintenance
planning/identification tool; it is a safety system

for operating trains. Visual inspection is carried
out separately by rail inspectors according to an
inspection plan, recorded in a report and stored in
a database. Visual inspection may also be carried
out in an unplanned manner by the inspector to
check the track condition between planned
inspection intervals. Finally, track inspectors are
obligated to report if they detect any deviation
from normal rail condition, as they perform their
daily maintenance work along the track.
Maintenance plays a vital role in improving
safety performance of the track. In this paper the
authors have tried to establish the relationship
between maintenance intervals and safety
performance. Petri-Net models have been used to
model the safety performance of the track. The
developed model in this work will estimate the
maintenance investment required to achieve a
specific safety level at a given point of time.
2.
MODELLING TRACK SAFETY
Rail infrastructure consists of various subsystems like track system, signalling and telecommunication system, and power system. Each
of these sub-systems contributes to infrastructure
safety.
9th International Heavy Haul Conference, 2009

Infrastructure managers use different indicators to
measure the safety of the infrastructure e.g.
number of accidents/per million train kilometres.
Safety of the track is measured as number of
derailments/per million kilometres as failures in
track system leads to derailments not collisions.
Broadly, performance indicators are classified as
leading or lagging indicators. A leading, lead, or
prospective indicator is a performance driver. The
outcome measure itself is simply the lagging, lag,
or retrospective indicator. Leading and lagging
indicators can also relate to strategy or goals, and
therefore it is important not to mix means and
ends. These safety indicators are lagging indicator
which only represents the current safety level of
the track. If the infrastructure manager wants to
improve the safety of track in future, it needs to
have a lead indicator i.e., probability of
derailment. Derailment because of track depends
on the undetected rail breaks on track and poor
track quality coupled with vehicle induced
dynamic forces.
set of transitions and a set of directed arcs.

Directed arcs connect places to transitions and
vice versa. The modelling is supported by
software tool GRIF. Some of the data used in the
models are taken from Banverket data bases [1]
and some are hypothetical in nature. However,
the assumed data are in close proximity to reality.
The case study is based on a section of heavy
haul line (10 kilometre in length) in north of
Sweden.
2.1 Modelling rail breaks
There are many stresses that operate on rail and
can influence rail defects and rail failure. Rail
defects mainly consist of surface initiated defects,
internal defects and weld defects. All these
defects can potentially lead to rail breaks which
are termed as rail failure [2]. Fig. 1 describes the
development derailments due to rail defects.
Rail defects
Let P1 (t) = probability of undetected rail break on

track at time t.
P2 (t) = probability of track quality index falling
below the maintenance limit at time t
Surface initiated
defects
Internal defects
Weld defects
Decreases
Preventive grinding
Increases
Detected
Periodic inspection
Rail replacements
Undetected
Time here is expressed in Million Gross Tonnes

(MGT). Probability of derailment due to rail
breaks and poor track quality are given by K1*P1
and K2*P2 respectively. Factors K1 and K2
represent the external factors such as train speed,
wheel condition, etc which induce dynamic
forces. During winter time the trains have higher
probability of getting wheel flats due to ice in the
braking system. Wheel flats are major
contributors to broken rail. In winter time the rail
is in tensile stress due to low temperature which
makes it more sensitive to external forces. It can
be assumed that for a specific track section, K1
and K2 are constant because the track structure,
speed ranges, climatic conditions do not change.
Thus, if probability of derailment needs to be
decreased, one needs to decrease P1 and P2. The
model in this paper illustrates the effect of track
inspection interval and track quality measurement
interval on P1 and P2 respectively. The model
relies on Petri-Nets and it provides dynamic
means of modelling stochastic failure processes.
A standard Petri-Net consists of a set of places, a
Rail breaks
Monitoring by
track circuits
Vehicle Condition
(Train Speed,
Wheel flat, etc
Detected
Undetected
Derailments
Figure 1: Logic flow of train derailment due to rail defects
Preventive grinding on the rail removes the rail

surface initiates cracks in their initial phase and
thereby stops their growth. Thus, formation of
surface initiated rail defects such as head check,
squats, etc are minimised. Internal defects (subsurface initiated), such as shell and transverse
defects, are associated primarily with heavy-haul
railways. Though grinding is not used in general
to remove embryonic cracks that cause shell
formation, transverse re-profiling of the rail
reduces stresses causing crack growth and hence
shell formation [3]. Weld defects are quite
common on a heavy haul line especially the
92

thermite welds. Until 2006 Banverket used flash
butt welding to weld 40m rail to 320m and then
320 meter rails were welded by thermite welding.
Today 60m rails are flash butt welded to 420 m
and then 420 m rails are thermite welded. This
decreases the number of thermite welds on track.
In the proposed model only defects in thermite
welds have been considered. Flash butt weld
defects have not been considered as they show
very low failure rate. On a heavy haul line,
thermite welds on the rail become defective due
to high axle loads in combination with cyclic
loading. When a defective weld is repaired one
more new weld is introduced and when a rail
defect is removed additional two welds are added
to the rail. Thus, the rate of thermite weld defect
increases due to increase in number of welds.
Periodic inspections are made to detect rail
defects and remove them; however, there is a
probability of detection attached to the ultrasonic
inspection depending on the size of defect.
parameters used by the model are given in Table

1. Table 1 illustrates the four types of defects that
lead to rail break in the current study. UIC 421 is
a thermite weld defect where as the other three
are rail defects. The description of these defects
can be found in UIC-712R i.e., catalogue of rail
defects. As shown in Table 1, all these defects
follow
2-parameter
Weibull
distribution.
Detection probabilities of these defects by NDT
car as well as visual inspection are also
mentioned in Table 1. The rail is inspected by
NDT car at every 12 MGT and visually every 0.5
MGT with the annual tonnage on the line is 24
MGT. If a rail break occurs, it is detected by track
circuits. However, in case of rail breaks that do
not separate the rails or rail gaps are small and are
not detected by track circuits. In this model rail
break detection probability of track circuits is
assumed to be 0.98. Initial number of thermite
welds for 10 km track is considered to be 32.
The P-F (Potential failure to Failure) interval of

the rail defects is an important factor as it denotes
the time interval between potential detection of
rail defects till a failure (rail break) occurs [4].
The P-F interval for the rail defects is given in
Table 1. Rail breaks are primarily detected by
track circuits. However, not all rail breaks are
detected by track circuits if rail breaks do not
create a gap between the rails. These rail breaks
remain undetected on the track until the next
periodic inspection i.e., NDT or visual inspection.
It is assumed in the paper that NDT inspection
detects the rail breaks 100% of the time where as
visual inspection detects 10% of the time.
Undetected rail breaks on the track pose a serious
threat to derailments. The probability of
derailment should also consider the vehicle
dynamics along with undetected rail breaks on
track. In this paper contribution of vehicle
dynamics to derailment has not been discussed.
Authors have tried to reduce the probability of
derailment by reducing the probability of
undetected rail breaks by keeping the speed factor
as constant as discussed earlier.
Defect type
Scale
Shape
Detection
parameter
parameter probability
() in
by NDT Car
()
MGT
UIC 135
225
Table 1: Parameters for rail defects

Detection
P-F
probability by
interval in
visual
MGT
inspection
2.5
UIC 211
338
2.5
UIC 2321
375
3.6
UIC 421
333
3.1
0.90
0.06
There can be a number of places on the rail where

defects (surface initiated and internal) may occur
and these places will change with respect to time.
In case of weld defect (UIC 421), the number of
welds will determine the number of defects.
Number of potential defect locations (PD) can be
calculated from the equation given below
PD (t) = cumulative number of defects in time t /
cumulative probability of defects in time t
Here time is considered in terms of MGT.
As defect is following a Weibull distribution,
probability of defect in time t is given by
F (t) = 1 exp (t / )
Cumulative numbers of defects are calculated
from the inspection data of the rail. After each
inspection the numbers of defects found on the
rail are known. These defects are the defects that
are detected by the NDT as well as visual
A Petri-Net model for estimation of undetected

rail break has been developed (see Fig. 2). The
model calculates the probability of undetected rail
break(s) at any given point of time. Some of the
93

inspection interval. When we add these defects
with the defects of previous inspection intervals,
we get the cumulative number of defects.
inspections. When these defects are divided by

detection probability, we get the probable number
of defects that may have occurred during that
Figure 2: Petri-Net model for determination of undetected rail breaks
defects that we can expect at a given point of

time. Table 2 explains the different places and
transitions Petri-net model.
If we plot the values of PD with respect to time,

we can get the trend for PD. When we multiply
PD (t) with F (t), we can estimate the number of
94

Table 2: Description of places and transitions of Petri-Net model for rail break
Places
Transitions
1: Start of NDT car inspection, 2: End of

inspection, 3: Start of Visual inspection, Tr1 and Tr3 fire at each inspection interval. Tr2 and Tr4 fire when the inspection is over
4: End of inspection
Tr5, Tr6, Tr7 and Tr8 fire everytime a defect of either type occurs. Tr9 feeds the place 5 with 1 token
5: No Defect on rail, 6: Defects on rail,
each time a defect occurs.Tr10 fires when visual inspection is done and Tr12 fires when NDT
7: Defects of rail that are detected by
inspection is done. Tr10 and Tr12 fire with detection probabilities mentioned in Table 1. Firing of
NDT or Visual inspection, 8: Defects
Tr11 repairs the detected defects. Tr13 fires when defects are not detected by NDT or visual
that are undetected
inspection. Those defects come to place 6 for next inspection.
9: Defects
Tr14 fires whenever a defect occurs. Tr15 fires when a defect is removed. Tr16 fires when an
unremoved defect reaches its P-F interval. Firing of Tr16 initiates a rail break. Firing of Tr17 occurs
when a rail break happens.
10 & 11: Act as counter for potential

Firing of Tr18 abd Tr19 calculate the potential defect locations for three types of defects in Table 1
defect locations
12: No rail breaks, 13: Rail break, 14: Tr21 fires when a rail break occurs. Tr20 feeds the place 12 with 1 token each time a rail break
Rail break detected by track circuit, 15: occurs.Tr22 fires with track circuit detection probability of detecting a rail break. Firing of Tr23
Rail break not detected by track circuit, repairs the rail break. Tr25 fires with visual detection probability of a rail break. Firing of Tr26
16: Rail break detected by visual repairs the rail break.Firing of Tr27 occurs when rail breaks are undeteced by rail visual inspection.
inspection, 17: rail break undetected by Tr24 fires when NDT car inspection takes place and it repairs all the undetcted rail breaks from the
visual inspection.
track.
18: Rail break detected, 19: Rail break
undetected
Tr28 fires when rail break(s) remain undetected. Tr29 fires when rail breaks are detected and
repaired.
The incidences of rail defects are random in

nature and the time for these defects to become
rail breaks depend on the P-F interval of the
defects. In this model it is assumed that if a
number of rail defects occur during a period of
time and remain undetected, the incidence of rail
break depends on the P-F interval from the time
of occurrence of the 1st defect. By performing
Monte Carlo simulation on the Petri-Net model,
probability of undetected rail breaks (with 90%
confidence interval) with respect to increase in
MGT on track has been found out (see Fig. 3). As
seen in the figure, the probability of undetected
rail breaks increases 5 times, when MGT
increases from 200 to 300. The increase in
probability is due to the fact that numbers of
defects keep on increasing with increase in MGT.
Thus, if proper maintenance measures are not
taken with increase on accumulated tonnage on
track, safety levels of the track will go down and
more derailments will be expected to occur. Fig.
4 depicts the change in probability of undetected
rail breaks with change in inspection interval
from 12 to 6 MGT for MGT 250 to 300. Mean
value of the probability was considered. While
doing this sensitivity analysis all other parameters
were kept constant. With increase in inspection
frequency, infrastructure manager can estimate
the extra maintenance investment that it has to
put so that it can achieve the desired safety level.
Figure 3: Probability of undetected rail breaks vs. MGT

1,80E-03
Prob. of undetected
rail break
1,60E-03
1,40E-03
1,20E-03
1,00E-03
8,00E-04
6,00E-04
4,00E-04
2,00E-04
0,00E+00
12
10
Inspection Interval (MGT)
Figure 4: Change in probability of undetected rail breaks

with change in inspection interval for MGT 250-300
As discussed earlier currently Banverket is

reducing number of thermite welds by welding
420 m rails instead of 320 m. Also the
manufacturing quality of rails has increased
considerably which reduces defects like tache
ovale (UIC 211). Fig. 5 illustrates the probability
95

side position of the rail. Hlim and Slim are the
comfort limits for a given track class. Track class
classifications are based on the speed of the train.
Banverket uses the following levels for a specific
class of track [5].
x
A: New built or recently adjusted track
x
B: Lower quality limit. It states target
value for maintenance actions. The track
irregularities should normally be adjusted
before this level attains. This limit is often
related to comfort aspects.
x
C: This limit should not be exceeded. The
track irregularity must be corrected as
soon as possible. Reduced speed limits
should be taken into consideration until
the irregularities have been corrected.
of rail break in the current practice. As these are

newly laid rails, the defect statistics are not
obtained yet. Thus, the same potential defect
locations that were considered for old rails have
also been considered in Fig. 5.
Figure 5: Probability of undetected rail breaks vs. MGT for

newly laid rails
In this paper Q value for maintenance limit for

the track is taken as 82. If the Q value falls below
maintenance limit and tamping is not performed
then probability of derailment increases. Q value
is measured by the measuring wagon at every 24
MGT. Table 3 provides the Q value with passing
tonnage and the time when tamping was
preformed on track.
If we compare the probability of Fig. 5 with Fig.

3 at 250 MGT, we can see that in the newly laid
rails there has been 25% decrease in probability.
2.2 Modelling track geometry deterioration
Track geometry deteriorates primarily due to the
influence of dynamic loads exerted by vehicles.
Continuous measurements of track geometry are
necessary in order to make decisions on
maintenance. Banverket has a number of
condition indices to describe the condition of
their infrastructure facilities. The main condition
indices are known as K-value and Q-value. These
are calculated from detailed inspection car
measurements of the track. The inspection car
measures relative rail position (lateral and
vertical), rail profile and rail gauge. The Q-value
is a weighted index of the standard deviation of
two inspection car measures calculated as
deviation from geometric comfort limits set for
specific track class. The Q-value is calculated per
kilometre track as:
Q
Table 3: Data of track quality measurement and

tamping
Measurement Track Quality Measurement Track Quality
(MGT)
Index (Q)
(MGT)
Index (Q)
24
95
144
48
88
168
72
81
192
96
94
216
120
86
Tamping: 72 MGT, 144 MGT, 216 MGT
79
92
85
78
By treating the data provided in the Table 3, the

slope of Q value with MGT and the effectiveness
of tamping were calculated. These values were
used in the Petri-Net model described in Fig.6.
Table 4 describes the different places and
transitions mentioned in Fig. 6. Fig. 7 illustrates
the change of Q value with passing MGT where
as Fig. 8 depicts the probability (P2) of Q value
below maintenance (tamping) limit.
VH
VS
150 100
2.
/3
V S lim
V H lim
where H and S are the average standard

deviation of height and interaction on the section
measured. The standard deviation for interaction
is calculated as a combined effect from cant and
96
Figure 6: Petri-Net model for determination of probability of track quality index exceeding maintenance limit
Table 4: Description of places and transitions of Petri-net model for track quality
Places
Transitions
1: Start of measurement, 2: End

of measurement
Tr1fires at each measurement interval. Tr2 fires when the measurement is over
3: Start of tamping, 4: End of

tamping
Tr3 fires when Q value is below tamping limit and it is detected by the measuring wagon. Tr4 fires
when the tamping is over.
5 & 6 : Act as counter for quality

Firing of Tr5 and Tr6 calculate the track quality index at any given point of time.
index
7: Q value below maintenance
limit, 8: Q value above
maintenance limit
Tr7 fires when Q value falls below maintennace (tamping) limit and tamping is yet to be carried out.
Tr8 fires when tamping is done and Q value is above maintenance limit.
Figure 7: Q- value vs. MGT
Figure 8: Probability of Q value below maintenance limit

vs. MGT
97

measurements
affect
the
probabilities.
Reductions of these probabilities reduce the risk
of derailment. This model will help the
infrastructure managers to estimate additional
maintenance investment to increase safety
performance of the track to a desired level.
Prob. of Q value below

maintenance limit
It can be seen in Fig. 8 that as soon as the

tamping is done on track, the probability
decreases and hence the safety performance
increases. However the overall probability
increases with time; Fig. 9 illustrates the change
in probability of Q value below maintenance
limit with change in measurement interval for
MGT 250 to 300. The mean value of the
probability was considered. With increase in
measurement frequency, infrastructure manager
can estimate the additional maintenance
investment.
References
[1] Kumar, S., Reliability Analysis and Cost
Modelling of Degrading Systems, PhD Thesis,
Lule University of Technology, Lule, Sweden,
ISSN: 1402-1544
2,00E-01
[2] Zhao, J., Chan, A. H. C. and Stirling, A. B.,

Risk Analysis of Derailment Induced by Rail
Breaks a Probabilistic Approach, Annual
Reliability and Maintainability Symposium,
2006, pp.486-491
1,60E-01
1,20E-01
8,00E-02
4,00E-02
[3] Cannon, D. F. et al., Rail Defects: An

Overview, International Journal of Fatigue &
Fracture engineering Materials & Structures,
Vol. 26, October 2003, pp. 865-886
0,00E+00
24
18
12
Measurement Interval (MGT)
Figure 9: Change in probability of Q value below

maintenance limit with change in measurement interval for
MGT 250-300
3.
[4] Vatn, J., Podfillini, Z. E. and Svee, H., A

Risk Based Approach to Determine Type of
Ultrasonic Inspection and Frequencies in
Railway Applications, the World Congress on
Railway Research, 2003, Edinburgh, UK
DISCUSSION AND CONCLUSION
The safety performance of the railway track is

compromised by derailments. Derailments take
place due to faults on the track and/or due to bad
vehicle dynamics. In these paper derailments due
to undetected rail breaks and poor track quality
have been described. However, derailments can
also occur due to track buckling. Track buckling
happens when the thermal stress in the track
exceeds the track lateral resistance. Lateral
resistance of the track changes due to
maintenance work on the track. After each
maintenance action, if the track lateral resistance
is not restored to the original value, buckling
may take place. The probability of buckling was
not considered in this paper because probability
of buckling depends on the quality of
maintenance work rather than maintenance
frequency. The Probabilities obtained in this
paper can be used as safety indicators for the
track. It has been shown in the paper that how
frequency of track inspections and track quality
[5] Lindahl, M. Track Geometry for Highspeed Railways, TRITA FKT Report, Royal
Institute of Technology, Stockholm, Sweden,
ISSN: 1103: 470X
98
Paper III
Availability analysis of railway track circuit
Patra. A. P. and Kumar, U. (2009). Availability analysis of railway track

circuit. Accepted for publication in Proceedings of the Institution of Mechanical
Engineers, Part F: Journal of Rail and Rapid Transit.
Availability analysis of railway track circuit

__________________________________________________
A P Patra* and U Kumar
Lule Railway Research Center, Division of Operation and Maintenance Engineering, Lule
University of Technology, Sweden
Abstract: Railways are expected to operate with ever increasing availability. The availability
of railway systems and subsystems influences the overall operational availability. As the track
circuit is a key component of railway signalling and control, it can contribute significantly to
the loss of availability of the railway system. One way to increase the availability of the track
circuit is through implementing proper a maintenance policy. In this article, Petri-Nets have
been used to develop models for availability analysis of the track circuit for both time-based
and condition based maintenance. A sensitivity analysis has been performed to estimate the
effect of maintenance parameters on availability performance using cost-benefit analysis.
Keywords: Track circuit, Availability, Petri-Nets
1. INTRODUCTION
The railway track circuit is an electrical device installed on the track to provide an indication
of train presence on a particular section of a track. The track circuit has been the most
commonly used train detection device besides the axle counter. The advantage of the track
circuit over the axle counter is its ability to detect rail breaks. Its correct operation is critical to
obtaining dependable train operations in most main line and metro systems. The track circuit
is a fail-safe device ensuring that any fault results in the signal light turning red, and in trains
being prevented from entering the associated track section. However, this fail-safe property
also has the potential to cause significant train delays if the system becomes unreliable. Any
track circuit failure can cause significant disruption to rail services and hence can become a
safety risk due to delay in the restoration of normal service. Over 12,000 track circuit failures
were reported in the UK during 20042005, resulting in 1.5 million minutes of attributable
delay. Typically, the UK railway infrastructure manager can be penalised with a penalty of
2060 per delay minute arising from infrastructure failure [1]. Therefore, the ability to detect
and diagnose track circuit failures in order to provide a fast response to failures/incidents has
significant economic benefits.
Reduction of the un-availability caused by track circuit failures must be accomplished using
an assured methodology. This methodology usually requires either an optimisation or a
complete overhaul of the maintenance processes in place. This starts with understanding the
causes of track circuit failures, which can either be within the track circuit itself, e.g.
component or systemic failure, or result from a failure of part of the infrastructure, such as a
poor track bed or related components. Railways all over the world follow a range of different
track circuit maintenance policies. However, the problem is to find the cost-effective
maintenance policy which will provide higher availability of the track circuit at a lower life
cycle cost. To address this problem, it is important to determine the life cycle of the track
circuit and the relationships between the failure distribution and the maintenance policy. The
cost of each maintenance policy can be calculated by adding the direct maintenance costs and
the consequential costs, e.g. the penalties due to train delay. This article provides a
comparison between the track circuit availability achieved using scheduled maintenance and
that achieved using condition based maintenance. The influence of the maintenance
parameters on the availability in both cases is studied for track circuits currently in use across
Sweden. There are different types of track circuits being used by infrastructure managers, e.g.
DC track circuits, AC track circuits and audio frequency track circuits. In this article our
comparisons have concerned the maintenance strategy for the DC track circuit which is used
by the Swedish National Rail Administration (Banverket).
The DC track circuit is the simplest and least costly type of track circuit used for train
detection. Fig. 1 presents the schematic diagram of a DC track circuit.
7V
Insulated rail joint
Relay
I rail
Signal path
S rail
Fig. 1 Schematic diagram of DC track circuit
Banverkets the train detection system uses the two rails on a track, the common rail (S-rail)
and the information rail (I-rail), to locate the position of the trains. The S-rail is continuous
and is grounded and used for the track return current. The I-rail, on the other hand, has gaps
with an insulated rail joint between the different sections of the rail. The polarity of each
section is reversed so as to prevent the supply to one circuit from powering the adjacent
circuit. The gaps isolate each section of the I-rail with a different potential. When no trains are
on the line, there is a negative or positive potential between the rails. When a train moves into
a new section (over a gap to a new rail section on the I-rail), the wheels complete the circuit
connection and the potential drops to zero. The energized relay for the occupied section holds
the voltage drops indicating where the train is on the track [2]. This track circuit design is failsafe; i.e. any failure in the track circuit turns the signal to red. Thus, the safety of the track is
not compromised. However, frequent track circuits failures affect the reliability and
availability of the track.
To achieve a reliable system, the factors which could influence the reliability of the system
need to be identified, their effects need to be assessed and the causes of these effects need to
be managed throughout the life cycle of the system. Railway system reliability is influenced
in three ways [3]:
x
x
System conditions: the sources of failures are introduced internally within the system
at any phase of the railway system life cycle. These failures are incurred by the design
and manufacturing of the components or the system.
Operating conditions: the sources of failures result from the operating system
methodology. These failures are also incurred by environmental conditions.
Maintenance conditions: the sources of failures are caused by maintenance actions.
The functionality of a track circuit is affected by the failure of its components, by changes in
track characteristics or by track maintenance. The DC track circuit failure modes include:
x
x
x
x
x
The failure of track circuit components

Insulated rail joint failure
Rail breaks
Lower ballast resistance
False alarms
Faults in track circuit components can be due to power supply failure or failure of the relay.
There can be loose cable connections due to vibrations and cable discontinuities due to breaks
in the cable caused by track maintenance actions, e.g. tamping. These failures are intermittent
and random in nature. The failure distributions of such events are likely to follow exponential
probability distributions. Insulated rail joints have shorter service lives than most other track
components. The frequency of insulated joint failures is influenced by the relative and
continuous weight (tonnage) of the traffic using the rails. Therefore, the frequency of failures
of insulated joints is high in heavy haul. Capturing condition information about joints and
taking preventive action remain challenges, as it is still proving difficult to obtain reliable
trending information on the condition of joints. The failures of these joints are timedependant and increase with time and generally follow Weibull probability distributions.
Consequently, time-based inspection is recommended as an approach to detecting
degradations in insulated joints and replacing them before they fail. However, it is not easy to
optimise the inspection periods in a track section with a number of insulated joints installed at
different times. Broken rails follow similar probability distributions to those of insulated joint
failures.
The track circuit operates at a specified electrical ballast resistance. When the electrical
resistance of the ballast is lower than the specified value, the current flow along the rails drops
and de-energizes the relay, which makes the track circuit non-functional. This is a frequent
occurrence as a result of wet or dirty ballast. Ballast cleaning is necessary to keep the ballast
dry and the electrical resistance above the minimum value. Occurrences of these kinds of
failures are intermittent in nature and likely to occur more during winter than in summer.
False alarms where the track circuit equipment has triggered a fault but where no fault is
found are phenomena which are registered as NFF. These faults occur with no prior
notification and are unpredictable in nature. False alarms normally are a significant subset of
track circuit failures. As track circuit failures stop the rail service operation, it is necessary to
detect degradation in track circuits and repair them before they actually fail. The following
sections describe the features of scheduled maintenance and condition based maintenance, the
differences between the two types of maintenance, and the benefits of condition monitoring
over scheduled maintenance.
2. MAINTENANCE POLICY
Infrastructure managers generally employ a scheduled (time based) maintenance regime for
track circuits. The objective of this maintenance is to detect any degradation in the track
circuit regularly and perform preventive maintenance to bring it back to a known acceptable
state. This type of maintenance regime is expensive and time-consuming, since inspection
needs to be carried out on every track circuit periodically (e.g. every 6 weeks for DC track
circuits, which is the interval used by Banverket). However, sudden failures can, and do,
occur between scheduled maintenance actions. In the event of such failures, urgent trackside
corrective maintenance is carried out, which is costly, particularly when it has to be carried
out during traffic hours. The periodic inspection of track circuits consists of measuring the
voltage and current at different locations of track circuits, checking the insulated joints,
inspecting the ballast condition, etc., to capture degradations (if any). If the measured current
and voltage are different from the specified value, the track circuit is degraded and
maintenance is carried out. However, if the occurrences of degradations of track circuits are
random in nature, then it is difficult to optimise the scheduled maintenance interval in order to
increase the system availability and reduce the cost.
Condition monitoring systems are designed and implemented so as to learn about system
degradation and therefore reduce the maintenance burden and cost. The ability to detect
degradation, identify certain incipient faults and/or provide diagnoses of failed track circuits,
in a more intelligent way, would have significant operational and economic advantages. One
of the most important aspects of a robust track circuit condition monitoring system is its
ability to identify degradation and failure modes, then to detect and locate a fault when it
occurs, and to predict incipient failures so that potential damage can be avoided. Preventive
maintenance can be performed before total failure [4]. The potential benefits of track circuit
condition monitoring include:
x
x
x
x
x
x
x
Improvement in the track circuit reliability and availability by reducing the

downtime during operational hours and hence reducing train delays.
Improvement in the post-incident analysis by improving the fault diagnosis
capabilities, in particular by locating and classifying faults in a failed track circuit.
Reduction in the number of Tested OK on arrival events.
Reduction in the scheduled down time periodic routine maintenance tasks on the
track circuit may not be required.
Reduction in the safety risk associated with maintenance activities by reducing the
amount and length of trackside work.
Reduction in the maintenance and replacement costs the track circuit units can be
maintained or replaced just-in-time if predictive condition monitoring can be achieved.
Overall improvements in the track circuit performance to provide a better
understanding of the signalling network, optimise the train control system, reduce the
penalty costs and increase the quality of rail services.
These are the potential advantages of track circuit condition monitoring systems. However,
balanced against these advantages are the costs of establishing the requirements, of procuring,
installing, learning and adapting the system and also of managing its own reliability.
Therefore, the ability of the system to detect the probability of faults is a combination of the
detection probability of the faults to be detected by the monitoring system and the reliability
of the monitoring system itself.
In this paper we have modelled availability as a factor of the detectability of the monitoring
system. The decision to adopt either scheduled maintenance or condition based maintenance
is taken based on an evaluation of the cost-benefit analysis of both types of maintenance
philosophy
3. AVAILABILITY MODELLING
Track circuit availability modelling has been performed by using Petri-Nets [5]. A standard
Petri-Net consists of a set of places, a set of transitions and a set of directed arcs. Directed
arcs connect places to transitions and vice versa. The places of Petri-Net model represent the
degrading states of the track circuit and the transitions represent the time delay between the
two degradation states. The modelling has been supported by the software tool GRIF. The
following sections describe how estimates of availability for the two approaches to
maintenance have been made.
3.1 Residual life
As discussed in the previous section, track circuit degradations and failures can occur for
different reasons. A framework is presented in Fig. 2 for the treatment degradation and failure
data of track circuits in order to calculate the reliability parameters. These failures constitute
all the failure modes of the track circuit. As illustrated in the figure, in the present scenario
track circuits are inspected at scheduled time intervals to check them for any degradation. If
the track circuit is degraded, this is treated as a condition failure and preventive maintenance
is carried out on the track circuit to bring the track circuit to good state. The track circuit
reaches a degradation state in between two inspections, but this state can only be detected and
the track circuit can only be maintained during the next inspection. In order to estimate the
effect of the inspection interval on detecting a degradation of the track circuit before its
functional failure, we need to calculate the time to degradation (degradation life) and the time
from degradation to functional failure (residual life) of the track circuit.
Fig. 2 serves as an example of the representation of real data for the maintenance of different
track circuits (TC1, TC2, etc.) in relation to time. During inspection, if a track circuit is found
to be degraded, the degradation is termed as conditional failure and preventive maintenance is
carried out. If the track circuit goes out of order, i.e. functional failure occurs, corrective
maintenance is carried out. It is assumed in this paper that both these maintenance actions
bring the track circuit to an as-good-as-new state. It is also assumed in the paper that any
manual inspection that is carried out on the track circuit at a scheduled time interval is perfect
and free from any human error. This means that, if a track circuit is degraded, it is always
detected during the inspection. It can be said that, when a track circuit is maintained
preventively or correctively, the degradation must have occurred between the maintenance
point of time and the last inspection point of time. This is shown as the occurrence of a
degradation zone in the figure.
As the point of occurrence of the degradation between two inspections or between the
functional failure and the last inspection is uncertain, the probability distribution of the
degradation life can be estimated considering the interval data for degradation. Once we
obtain the degradation life distribution, the next step is to estimate the residual life. The
residual life is the remaining life in a track circuit from the point of degradation until it
experiences a functional failure. When estimating the residual life of the track circuit, we
should also consider undetected degradation times, which are also termed as suspended
residual life.
A model has been developed in Petri-Nets to calculate the undetected degradation time (see
Fig. 3). State 1 denotes that the track circuit is in good state and state 2 that it is in degraded
state. Degradation of the track circuit brings the track circuit from state 1 to state 2. State 3
and 4 denote the inspection states. Inspection is carried out at a scheduled interval. A
degraded track circuit remains in a degraded state until the next inspection commences. For an
inspection interval of 1000 hours, by simulating the Petri-Net models we can see that the
undetected degradation time is half of the inspection interval, i.e. 500 hours (see Fig. 4).
Track Circuit (TC)
Undetected degradation time

or Suspended residual life
Conditional failure:
Preventive maintenance
TC 1
Degradation
life
Occurrence of
degradation
Degradation
life
Residual
life
Assumed
degradation
points
TC 2
TC 3
Functional failure:
Corrective maintenance
Occurrence of
degradation
TC n
Inspection
interval
Time
Fig. 2 An example of data representation for track circuit degradations and failures
The probability distribution of the degradation is assumed to be a Weibull distribution. It was
seen that for other distributions, the undetected degradation time also remains half of the
inspection interval. This result can also be inferred for the residual life calculation; i.e. the
residual life is half of the time between the functional failure and the last inspection.
Considering different values for the residual life and the suspended residual life (undetected
degradation time), we can estimate the probability density function for the residual life.
Fig. 3 Petri-Net model for estimation of undetected degradation time
Fig. 4 Undetected degradation time

3.2 Scheduled maintenance
The availability estimation for the track circuit only considers the corrective maintenance. It
does not consider the preventive maintenance, as the preventive maintenance is carried out
during train-free periods; i.e. i.e. the unavailability during non-traffic hours is not included. A
Petri-Net model for estimating the availability with the scheduled maintenance policy is
shown in Fig. 5. State 1 denotes that the track circuit is in good state. The track circuit goes to
state 2 when degradation occurs. It further goes to state 3 (failed state) depending upon the
probability distribution of the residual life. Corrective maintenance in state 3 brings the track
circuit back to good state. States 4, 5 and 6 are the inspection states. Inspection is carried out
at every inspection interval. After the inspection is carried out, if it is found that the track
circuit is in degraded state (state 2), preventive maintenance is performed on the track circuit
and the track circuit goes back to good state.
For example, for each track circuit, the degradation follows a Weibull distribution with scale
parameter 5000 hours and shape parameter 1.8, and the residual life distribution is a Weibull
distribution whose scale and shape parameters are 700 hours and 1.2 respectively. The mean
7
down time for both preventive and corrective maintenance is 2 hours and the inspection time
is 0.5 hours. For an inspection interval of 1000 hours, the availability of the track circuit with
respect to time is shown in Fig. 6. As shown in the figure, we obtain a steady state availability
of 0.99985 for each track circuit. For a typical track section of 300 km, we can have an
average of 200 track circuits. If we need to calculate the availability of the section due to track
circuits, we must consider the delay aspects of track circuits due to their unavailability.
Different indicators of availability are illustrated in [6]. In this article the authors define the
availability of a section of track as the probability that any failure in that track section will not
induce any delay to the trains. Based on this definition, the availability of the track section is
given by:
As = 1- [(1-An)*Pd]
n = Number of track circuits in the section
A = Availability of each track circuit
Pd = Probability of delay
To explain how this expression is developed, let us consider the simple case of three track
circuits in a track section. A is the availability and is the un-availability of each track circuit
i.e. all the track circuits are considered to be identical.
The probability of at least one of the three track circuits being un-available is given by:
3 = + + (* ) - (* ) - (* ) + (* *)
The un-availability of the track circuits is independent but not mutually exclusive (e.g. P(A +
B + C) = P(A) + P(B) + P(C) P(A)*P(B) P(B)*P(C) P(C)*P(A) + P(A)*P(B)*P(C) )
Further, the probability of at least one of the three track circuits being un-available is also
equal to one minus the joint probability that none of the track circuits are un-available.
Thus, 3 = 1 [(1-)* (1-)* (1-)] = 1- A3
If Pd is the probability of delay, then the un-availability of the track section is given by (1A3)*Pd.
Similarly, for n track circuits, the un-availability of track section will be (1-An)*Pd.
Hence, the availability of the track section is As = 1- [(1-An)*Pd]
If Pd is 1 (i.e. any un-availability of a track circuit in a section induces a certain delay), then
As = An, where as, if Pd is 0 (i.e. there is no train delay in the case of any un-availability of a
track circuit), then As = 1.
In this article it is assumed that any track circuit failure in a track section will induce train
delay, and therefore the availability of the track section = A200 = 0.99985200 = 0.9704.
Fig. 5 Petri-Net model for estimation of availability for scheduled maintenance
Fig. 6 Availability of track circuit for scheduled maintenance interval 1000 hours
If we want to increase the system availability, we need to perform a sensitivity analysis on the
inspection interval. Fig. 7 illustrates the change in the section availability of a track circuit
with a change in the inspection interval. If we want to optimise the inspection interval in
terms of cost effectiveness, we need to calculate the cost of inspection as well as the cost of
delay as described earlier.
Availability of section
0,99
0,98
0,97
0,96
0,95
0,94
0
500
1000
1500
2000
2500
3000
3500
4000
4500
Inspection interval
Fig. 7 Section availability vs. inspection interval

3.3 Condition based maintenance
As discussed earlier, condition monitoring on the track circuit will increase the availability of
the track circuit. In this section we will illustrate the proportional increase in the availability
of the same track circuit if we change the maintenance policy from scheduled maintenance to
condition based maintenance. A Petri-Net model for estimating the availability with a
condition based maintenance policy is shown in Fig.8. The detection probability of the
condition monitoring system is considered in the model. As shown in the figure, when the
track circuit is in degradation state, it can either go to detected degradation state (state 3) or
un-detected degradation state (state 4), depending on the detection probability.
Fig. 8 Petri-Net model for estimation of availability for condition based maintenance
In this model it is assumed that the condition monitoring system detects the degradation in the
track circuit all over its residual life. Therefore, the degradation life distribution in this case is
10
the combination of the degradation life and the residual life distribution in the earlier example.
If we consider the same failure and maintenance parameters as those considered for scheduled
maintenance and perform a sensitivity analysis of the section availability with respect to the
detection probability, we see the result in Fig. 9.
Availability of section
0,995
0,99
0,985
0,98
0,975
0,97
0,965
0,96
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
Detection probability
Fig. 9 Section availability vs. detection probability

As shown in the Fig. 9, one can achieve the same availability of the track section as achieved
using current practice (scheduled maintenance with an interval of 1000 hours) by
incorporating condition monitoring devices which have a detection probability of only 60%.
However, it is very likely that condition monitoring systems have a higher detection
probability than 60%. With a higher detection probability, we can achieve higher availability.
However, to compare the cost effectiveness of condition based maintenance with that of
scheduled maintenance, we must perform a cost-benefit analysis of each maintenance policy.
4. COST-BENEFIT ANALYSIS
As decisions on maintenance policies have a major impact on the operations of the system,
the cost-effectiveness of long term design and maintenance decisions should be guaranteed.
Cost-benefit analysis, an engineering economics technique, can be utilised to focus on
maintenance strategies to minimise the maintenance cost in the systems life cycle, while
meeting the dependability requirements. When comparing maintenance policies, the
maintenance cost should include consequential costs (e.g. penalties due to train delays) in
addition to direct maintenance costs. In the case of the scheduled maintenance policy, the
maintenance costs include labour costs and equipment costs for inspection. If we want to
change the maintenance policy to a condition based policy, we need to include the acquisition
costs of monitoring systems, as well as the validation time and the costs for proving and
developing the applicability of these systems and finally the cost of maintenance (i.e.
corrective replacements) of the monitoring systems themselves. In both the cases we need to
estimate the corresponding train delay and calculate the penalties.
Train delay consists of two types of delay, i.e. primary delay and secondary delay. Primary
delay is due to some external circumstances, whereas secondary delay is caused by other
trains. As secondary delay depends on the infrastructure design and train time schedule, in
this paper we consider only primary delay. Failures of track circuits also directly affect the
11
primary delay of trains. Fig 10 illustrates the primary delay incurred to trains because of track
circuit failure. It is assumed that the headway between the trains is big enough not to induce
any secondary delay.
Speed = Average
speed
Speed = Average
speed
Speed
Acceleration
Deceleration
Acceleration
(Track circuit failure)

Speed = 0
Speed = Restricted
speed
Train traffic control

authorization time
Time
Fig. 10 Explanation of primary delay of trains

When a track circuit fails, the signal turns red and running of a train in the block section
concerned is restricted. As shown in the figure, in the case of a track circuit failure, a train
decelerates and comes to a standstill just before that particular block. The train then waits for
the traffic control rooms authorization to proceed on the affected block at a restricted speed.
After the train covers that block at a restricted speed, it may accelerate to the original speed.
The train is delayed by the extra time taken to cover the distance because of stopping and
reducing its speed. If the track circuit is not repaired, then the following trains decelerate and
cover that block section at a restricted speed. This adds further primary delay.
Consequently, the total primary delay per year due to the failure of track circuits can be
calculated as:
Primary delay (minutes/year) = Failure rate of track circuit*Operation hours/year*[Primary
delay of train * (Mean down time/Average headway between
trains)]
In metro applications, where the headway is small, track circuit failures will give rise to
significant secondary delay if the mean down time due to failure is high. The maintenance
cost for each of the maintenance policies can be determined by adding the direct maintenance
costs and the costs of delay. In the case of condition based maintenance, inspection costs are
eliminated but we introduce additional costs for acquisitions of monitoring devices as well as
maintenance of monitoring devices. The cost effectiveness of a maintenance policy can be
determined by taking a ratio of the availability and the maintenance cost. The higher the costeffectiveness, the better is the maintenance policy.
5. CONCLUSIONS
In this article, models for estimation of availability for scheduled maintenance and condition
based maintenance have been developed. It has been demonstrated how the availability of the
track circuit is influenced by changing the maintenance interval of the track circuit. The
article also makes a comparison of scheduled maintenance and condition based maintenance
12
based on achieving the availability of track circuit. A framework is presented for the cost of
delay. Further, it is shown that cost-benefit analysis of maintenance policies can be carried out
to evaluate the cost-effectiveness of each maintenance policy. Cost-effectiveness analysis will
yield quantitative results to aid the decision maker with risk analysis, and provide a useful
decision tool. Better cost-effectiveness of a system is achieved by higher operational
availability and lower maintenance costs during the life cycle of the system.
ACKNOWLEDGEMENT
The work presented in this paper has been financially supported by the Swedish National Rail
Administration (Banverket). The authors would like to acknowledge this support.
REFERENCES
1 Chen, J., Roberts, C., and Weston, P. Fault detection and diagnosis for railway track
circuits using neuro-fuzzy systems, Control Engineering Practice, 2008, 16, 585-596.
2 Niska, S. Measurements and Analysis of Electromagnetic Interferences in the Swedish
Railway Systems, PhD Thesis, Division of Operation and Maintenance Engineering, Lule
University of Technology, Lule, Sweden, vol.76, 2008
3 EN 50126 The Specification and Demonstration of Reliability, Availability,
Maintainability and Safety (RAMS) for Railways Applications, Comit Europen de
Normalisation Electrotechnique (CENELEC), Brussels, Belgium, 1999.
4 Roberts, C., Dassanayake, H. P. B., Lehrasab, N., and Goodman, C. J. Distributed
quantitative and qualitative fault diagnosis: railway junction case study. Control
Engineering Practice, 2002, 10(4), 419429.
5 Murata, T. Petri Nets: Properties, Analysis and Applications. Proceedings of the IEEE,
1989, 77(4), 541 580.
6 Nystrm, B. Use of Availability Concepts in the Railway System, International Journal
of Performability Engineering, 2009, 5(2), 103-118
13
Paper IV
Cost effective maintenance policy: a case study
Patra, A. P., Dersin, P. and Kumar, U. (2009). Cost effective maintenance

policy: a case study. Accepted for publication in the International Journal of
Performability Engineering.
International Journal of Performability Engineering

RAMS Consultants
Printed in India
Cost Effective Maintenance Policy: a Case Study

AMBIKA PRASAD PATRA1*, PIERRE DERSIN2, UDAY KUMAR1
1

SE-97187, Lule, Sweden
2
ALSTOM Transport Information Solutions
92364 Meudon-la-Fort, France
(Received on September 8, 2009)

Abstract: The European Rail Traffic Management System (ERTMS) is a major venture
initiated by the European Union in order to create an interoperable railway network within
Europe. The manufacturers of the ERTMS strive to attain a competitive edge by
demonstrating the efficiency of their systems. Achieving the optimal cost effectiveness is
one of the significant ways to address the efficiency of the system and involves maximising
the availability and minimising the life cycle cost of the system over the system life cycle.
One of the important ways of maximising the cost effectiveness of the system is to optimise
the maintenance policy. This paper demonstrates the estimation of the cost effectiveness of
an ERTMS system. The degradation and repair process of the system is modelled by PetriNets. Failure, maintenance and cost data are used as parameters for the model. The model
will be useful for systems that experience degradations and are subjected to imperfect
maintenance.
Keywords: Availability, Life Cycle Cost, Maintenance Policy, Petri-Nets
1.
Introduction
Each country in the European Union has its own railway signalling system. Each system
is stand-alone and non-interoperable, and therefore requires extensive integration and
engineering effort, raising the total delivery costs for cross-border traffic. This restricts
competition and hampers the competitiveness of the European rail sector vis--vis road
transport by creating technical barriers to international journeys. To fulfil the requirement
of interoperability, a major industrial project named as the European Rail Traffic
Management System (ERTMS) was initiated. The objective of the ERTMS is to replace
the existing signalling systems with a system which will boost cross-border freight and
passenger transport. This will help the countries to establish a more sustainable railway
network. The ERTMS has two basic components, i.e., the European Train Control System
(ETCS) and GSM-Radio (GSM-R). The ETCS is an automatic train protection system,
while as GSM-R provides voice and data communication between the track and the train.
There are 3 levels of the ERTMS, with Level 1 and Level 2 already in operation in
Europe.
Apart from achieving an interoperable railway network, the ERTMS also increases
the capacity, speeds and safety for passengers on existing lines, and at the same time
reduces the maintenance costs. To be competitive and to gain the approval of
infrastructure managers as well as train operating companies, the ERTMS manufacturers
should demonstrate the cost effectiveness of their systems as per the stakeholders
requirement.
____________________________
*Corresponding authors email: ambika.patra@ltu.se
A. P. Patra, P. Dersin and U. Kumar
The cost effectiveness of any system depends on the operational availability and life cycle
cost (LCC). In order to make the system more cost effective, higher availability should be
attained at a lower LCC.
However, there are numerous challenges in attaining the desired cost effectiveness of
the system over a period of time, because of degradation of the systems, changes in traffic
scenario etc. Optimisation of the maintenance policy is one of the major ways to attain the
desired cost effectiveness of the system in the long run. All such optimisations should aim
at maximum system availability and minimum life cycle costs, as well as minimum train
delays for a specific traffic scenario. In this paper the authors demonstrate a model that
can help to maximise the cost effectiveness of the systems. ALSTOM Transport has
developed the ATLAS platform for ERTMS application for railway operations.
ATLAS consists of various sub-systems, such as the Automatic Train Supervision
(ATS) system, the interlocking system, the Automatic Train Control (ATC) system and
trackside products e.g., track circuits. This paper deals with a case study where an
optimum maintenance policy is developed to achieve the desired cost effectiveness of the
Radio Block Centre (RBC), which is a part of the ATC track side system. The ATC subsystem consists of both trackside and trainborne components. The objective of this
development work is to achieve a higher availability and a lower LCC for the system and
then develop an optimum maintenance policy to achieve cost effectiveness.
As decisions on maintenance policies have a major impact on the operations of the
system, the cost effectiveness of long term design and maintenance decisions should be
guaranteed. Life cycle cost (LCC) analysis, an engineering economics technique, can be
utilised to focus on maintenance strategies to minimise the maintenance cost in the
systems life cycle, while meeting the dependability requirements. The cost effectiveness
of a system can be defined as
Cost effectiveness Availability
LCC
Higher cost effectiveness of the system ensures better operation of the system. The
paper presents the key influential variables of maintenance policy that affect cost
effectiveness. Section 2 of the paper discusses the system description of the RBC. A PetriNet model is shown in Section 3. Section 3 also discusses the results of the model.
Finally, the conclusions are stated in Section 4.
2.
Radio Block Centre (RBC) System
In most of the systems which are in use at ALSTOM Transport, active redundancy has
been chosen: i.e. the various units are active simultaneously, so that, in the event of the
failure of one unit, the function is preserved without the need for switching on a back-up
unit [1]. On ERTMS application level 2, the ETCS uses a GSM-R radio channel to
exchange data between the trackside Radio Block Centre and the trains. The interlocking
system reports the status of the objects controlling the routes of the trains to the RBC,
which, in turn, generates the correct movement authorities for the different trains in the
section. The RBC consists of different sub-systems, such as the computing channel,
input/output system and cabinet. The reliability block diagrams of these systems are given
in Figs. 1a, 1b, 1c and 1d. The computing channel is a 2-out-of-3 (2oo3) system. Two
channels must be working at any given time for the computing channel to work. Similarly,
the input/output group is a 1oo2 system.
Cost Effective Maintenance Policy: A Case Study
Computing
channel
Input/Output
group
Computing
channel
Cabinet
1/2
2/3
Input/Output
group
Computing
channel
Figure 1a: Reliability Block Diagram of RBC

PPC
HSCU
RedMan
IDD
CPSU
Figure 1b: Reliability Block Diagram of computing channel

EAU
EAU
IOPSU
Figure 1c: Reliability Block Diagram of input/output group

Fan
MPSU
BPSU
1/2
MPSU
Fan
2/3
BPSU
Fan
Figure 1d: Reliability Block Diagram of cabinet
Any failure in the RBC is detected by built-in-test-equipment (BITE). However,

failures are detected depending upon the detection probability or fault detectability of the
BITE system. When a failure is not detected, it remains in the system until the next
inspection occurs. Eq. 1 provides an estimation of the undetected failure time if the fault
remains un-detected till the next inspection. The relationship shown in Eq. 1 considers
only the constant failure rate, but later on in the paper we develop a Petri-Net model
which considers the non-constant failure rate.
Time
0
t
T
Figure 2: Undetected failure time
T = Inspection interval
t = time of occurrence of undetected failure during the interval (0, T)
T-t = duration of undetected failure time
= failure rate of the component following an exponential distribution
The expected un-detected failure time of the component during the interval (0, T) is given
by:
E (T t )
(T t ) f (t )dt (T t ) * O * exp(Ot )dt

0
f (t )dt
O * exp(Ot )dt
T
1
)
(
1 exp(OT ) O
(1)
when T << 1, E(T-t) = T/2

The failure rate of a repairable component depends on the type of corrective
maintenance that is applied in a range stretching from perfect maintenance to minimal
maintenance. Perfect maintenance repair brings the component age to zero (i.e. the
component becomes as good as new (AGAN)) whereas minimal repair keeps the
components age un-modified (i.e. the component stays as bad as old (ABAO)). In real
world cases the repairs are neither AGAN nor ABAO, but are something in between. In
[2], two models (type I and II) are proposed that estimate the virtual age of the component
after a repair. Kijima models consider a parameter called the maintenance factor, which
estimates the virtual age. If the maintenance factor is 1, the repair is ABAO and for the
maintenance factor 0, the repair is AGAN. Model type I assumes that the repairs can only
fix the damage incurred during the last period of operation. Thus, the nth repair can only
remove the damage incurred during the time between the (n-1)th and the nth failures.
Model type II assumes that the repairs fix all of the damage accumulated up to the current
time. As a result, the nth repair not only removes the damage incurred during the time
between the (n-1)th and the nth failures, but can also fix the cumulative damage incurred
during the time from the first failure to the (n-1)th failure.
If the times between the failures are denoted by x1, x2, xn, t the virtual age of the
component after the nth repair is given by
Vn = Vn-1 + (maintenance factor * xn) Kijima model type I
Vn = (maintenance factor*Vn-1) + (maintenance factor *xn) Kijima model type II
The maintenance factors have been calculated from the past failure times of the
component by applying the Kijima models discussed above.
3.
Case study on the cost effectiveness estimation of the fan system of the RBC
The Cost effectiveness of the RBC system has been estimated. In this paper we are
presenting a case study on the fan system. As discussed earlier, the fan system in the RBC
is a 2oo3 system. A Petri-Net (for details see [3]) model for estimation of the cost
effectiveness of the fan system is illustrated in Fig. 3. The places that represent the states
are depicted in Table 1. There are three fans in the system and one fan is kept as a spare in
the store and all the fans are repairable in nature. As illustrated in the model, when a fan
fails, it goes to the failed state. The time to failure depends on the failure probability
density function of the fan. The model captures the time to failure each time a fan fails
and estimates the virtual age of the fan after repair work is completed depending on the
maintenance factor of the repair described earlier. If a fan fails and the failure is detected,
it is removed from the system and is substituted by the fan in stock. The failed fan goes to
the workshop for repair, and after repair it waits until the next fan failure occurs. Over a
period of time the fans may have different virtual ages depending on the number of repairs
performed on each fan. In this model, we can keep track of the number of repairs carried
out on each fan, so that we can estimate the time when the next fan will fail. Now, place
13 denotes that all three fans are working. If one fan fails depending on the time explained
above, the system can go to state 14 or state 15.
Figure 3: Petri- Net model for estimating the cost effectiveness of the fan system
If the failure is detected by the BITE system, the system will be in state 14, otherwise
it will be in 15. Both state 14 and state 15 are degraded states but the system is still
working because two fans are still working. If the system is in 14, it can go to 13
depending upon the deferred maintenance time. The maintenance is deferred because it is
not always cost effective to stop the train operation to repair the system, and therefore the
system can be repaired after the train running period. If the system goes to 15, the failure
can only be detected in the next inspection, after which the system can be brought to 13.
However, if the deferred maintenance time and the inspection intervals are long, then
there is a probability that the system can go to the failed state if another fan failure occurs.
In the case of a system failure, corrective maintenance is performed on the system to bring
the system to 13. This depends on the availability of the two fans in the stock. If the fans
are under repair, then the unavailability of the system increases. The parameters that are
used in the Petri-Net model are illustrated in Table 2. Monte Carlo simulations are
performed on the model to estimate the availability and LCC of the fan system over a
period of 20 years.
Table 1: Description of places in the Petri-Net models
State description
Place
State description
Place
State description
Fan system working
Place
Fan 1 working
Fan 3 working
Fan 1 failed
Fan 3 failed
Fan 1 repaired and

waiting in stock
Fan 3 repaired and

waiting in stock
Fan 2 working
Spare fan working
10
Fan system failed
16
Fan 2 failed
Spare fan failed
11
Inspection starts
17
Fan 2 repaired and

waiting in stock
Spare fan repaired and

waiting in stock
12
Inspection ends
18
Fan system degraded

(detected)
Fan system degraded (undetected)
13
14
15
Table 2: Parameters used in Petri-Net model for cost effectiveness estimation
Parameters
Operation hours/year
Inspection time
Preventive maintenance time
Corrective maintenance time
Labour cost/hour
SRU repair time
Discount rate
Fan failure (scale parameter)
Fan failure (shape parameter)
Deferred maintenance time
Inspection interval
SRU repair cost
Maintenance factor
Detectability
Value
6000 hours
0.5 hours
0.5 hours
3 hours
40
720 hours
4%
20000
2
9 hours
6000 hours
50
0.8
0.9
Fig. 4 illustrates the cost effectiveness of the fan system with time. The step decrease in
the cost effectiveness curve is due to the discounted value of the life cycle cost. After a
period of time the cost effectiveness curve will be parallel to the x-axis. This is due to the
fact that, after a period of time, the LCC will be constant because of the discounting of
future costs.
Figure 4: Cost effectiveness of the fan system with time
Sensitivity analyses have been performed on the maintenance factor, fault

detectability, inspection interval and deferred maintenance time to study their effects on
the cost effectiveness of the system. The results are shown in Fig. 5 and Fig. 6. The cost
effectiveness decreases with an increase in the maintenance factor. That is because, with
an increase in the maintenance factor, the quality of maintenance tends to be minimal
maintenance, thereby increasing the number of failures. In the case of fault detectability,
the cost effectiveness increases with an increase in the detectability. This analysis can help
to design a better built-in-test system to achieve better cost effectiveness of the system.
With an increase in the inspection interval, the probability of repairing components before
they fail decreases and hence the LCC increases. With a decrease in inspection interval,
the inspection costs increase and hence the LCC decreases. In this case study, as the
failure rate is very low, the optimum inspection interval is obtained at a very long
inspection interval. However, from a safety point of view, we cannot have a very long
inspection interval. When the deferred maintenance time increases, the probability of
failure from a degraded state increases. Hence, the cost effectiveness decreases. However,
at the same time, an increase in the deferred maintenance time also increases the
possibility of opportunistic maintenance, which decreases the overall costs.
1,00E-03
Cost effectiveness
Cost effectiveness
9,90E-04
9,60E-04
9,30E-04
9,00E-04
8,70E-04
9,50E-04
9,00E-04
8,50E-04
8,00E-04
8,40E-04
0
0,2
0,4
0,6
Maintenance factor
0,8
0,5
0,6
0,7
0,8
Detectability
Figure 5: Effect of maintenance factor and detectability on cost effectiveness
0,9
9,20E-04
Cost effectiveness
Cost effectiveness
1,20E-03
1,00E-03
8,00E-04
9,00E-04
8,80E-04
8,60E-04
6,00E-04
0
3000
6000
9000
Inspection interval
12000
15000
30
60
90
120
Deferred maintenance time
Figure 6: Effect of inspection interval and deferred maintenance time on cost effectiveness
We can also estimate the maintenance policy of the system which will ensure certain
availability and fulfil LCC requirements. From Fig. 5 & Fig. 6, the optimum value for
each parameter for maximising the cost effectiveness can be inferred. Further, the
optimum value of combinations of the parameters for achieving the maximum cost
effectiveness can also be estimated. The model can be useful for other mechanical
redundant systems which are repairable in nature and subjected to degradations.
4.
Conclusions
In view of the increasingly stringent availability requirements being set by the market
place, the designers of complex systems have to pay close attention to test and
maintenance strategies to achieve availability targets with low life cycle costs. In this
paper a maintenance policy based on cost effectiveness has been developed for the fan
sub-system of the Radio Block Centre (RBC). Sensitivity analysis has been performed on
different maintenance parameters to maximise the cost effectiveness of the system. This
maintenance policy will help the systems to achieve higher availability at lower life cycle
costs over the life cycle of the systems. A Petri-Net model has been developed to calculate
the cost effectiveness of these systems. Cost effectiveness analysis will yield quantitative
results to aid the decision maker with risk analysis, and provide a useful decision tool. The
work presented in this paper is a part of the work carried out for systems developed by
ALSTOM Transport for ERTMS applications. This work will help the ERTMS
manufacturers to demonstrate the sustainable benefits in terms of availability and life
cycle costs to the infrastructure managers, as well as the train operating companies, in
order to keep a competitive advantage.
Acknowledgement
The work presented in this paper is supported by ALSTOM Transport Information
Solutions, Paris. The authors would like to acknowledge this support.
References
[1]. Dersin, P., A. Peronne, and C. Arroum. Selecting test and maintenance strategies to achieve
availability target with lowest life cycle cost. In proceedings of Reliability and
Maintainability Symposium, USA, January 2008.
[2]. Kijima, M. Some results for repairable systems with general repair. Journal of Applied
Probability 1989; 26: 89-102.
10
[3]. Murata, T. Petri Nets: Properties, Analysis and Applications. Proceedings of the IEEE 1989;
77(4): 541 580.
Ambika Prasad Patra is a PhD candidate at the Division of Operation and Maintenance
Engineering, Lule University of Technology, Sweden. His research topics are RAMS and LCC
analyses for railway infrastructure systems. He received his MTech degree in Reliability
Engineering in 2002 from the Indian Institute of Technology, Bombay, India.
Pierre Dersin obtained his Ph.D. in Electrical Engineering in 1980 from the Massachusetts Institute
of Technology (MIT), after receiving a Masters degree in O.R. in 1976, also from MIT. He worked
in the late seventies on the reliability of large electric power networks, while taking part in the Large
Scale System Effectiveness Analysis Program sponsored by the US Department of Energy, as an
employee of MIT and Systems Control, Inc. He later joined FABRICOM (in Belgium and the
U.S.A.), where he was involved with fault diagnostic systems for factory automation. Since 1990 he
has been working at ALSTOM Transport, where he has occupied several positions and has mainly
been involved with RAMS and maintenance. He has been R&D Manager of the Service section, and
is now RAM Director in ALSTOM Transports Information Solutions (i.e. railway signalling and
communication). He has contributed a number of papers and publications in the fields of RAMS,
automatic control and electric power systems. He was the chairman of UNIFEs (the Association of
the European Rail Industry) LCC/RAM Working Group in 1998-2001.
Uday Kumar is Professor of Operation and Maintenance Engineering at Lule University of
Technology, Lule, Sweden. He is also Director of Lule Railway Research Centre at the same
university. His research interests are maintenance engineering and management, reliability and
maintainability analysis, product support, life cycle costing, risk analysis, system analysis, etc. He is
also a member of the editorial board and a reviewer for many international journals. He has
published more than 150 papers in international journals and conference proceedings.
Paper V
Uncertainty estimation in railway track life-cycle cost: a

case study from Swedish National Rail Administration
Patra, A. P., Sderholm, P. and Kumar, U. (2009). Uncertainty estimation in

railway track life-cycle cost: a case study from Swedish National Rail
Administration. Proceedings of the Institution of Mechanical Engineers, Part F:
Journal of Rail and Rapid Transit, 2009, 223 (F3), 285-293.
285
Uncertainty estimation in railway track life-cycle cost:

a case study from Swedish National Rail Administration
A P Patra , P Sderholm, and U Kumar
Lule Railway Research Center, Lule University of Technology, Norbotten, Sweden
The manuscript was received on 5 May 2008 and was accepted after revision for publication on 17 October 2008.
DOI: 10.1243/09544097JRRT235
Abstract: Life-cycle cost (LCC) is used as a cost-effective decision support for maintenance of
railway track infrastructure. However, a fair degree of uncertainty associated with the estimation
of LCC is due to the statistical characteristics of reliability and maintainability parameters. This
paper presents a methodology for estimation of uncertainty linked with LCC, by a combination
of design of experiment and Monte Carlo simulation. The proposed methodology is illustrated
by a case study of Banverket (Swedish National Rail Administration). The paper also includes
developed maintenance cost models for track.
Keywords: reliability and maintainability, life-cycle cost, railway track, design of experiment,
uncertainty
INTRODUCTION
Life-cycle cost (LCC) takes into account all costs

associated with the life time of the system, such as
operating costs, maintenance costs, energy costs, and
taxes apart from capital costs. For many complex
assets, the cost of maintenance plays an important
role in the LCC analysis, especially for assets like track
infrastructure, where the operation and maintenance
phase comprises a major share of the systems life
cycle. However, though most infrastructure managers
today consider all the costs incurred by the system
from conceptual design to disposal in their LCC calculations, there are still some issues associated with
the correctness of these calculations. Some important issues are related to uncertainties in the LCC
calculations.
Figure 1 illustrates two different levels of uncertainties associated with LCC of track infrastructure. Level
I uncertainty is costs due to penalties imposed by trafc operators on the infrastructure manager due to
such factors as train delay, trafc disruption, or derailment. These anomalies can be caused by planned
or unplanned maintenance actions, but also by lack
Corresponding
author: Division of Operation and Maintenance
Engineering, Lule University of Technology, Lule, Norbotten

97187, Sweden. email: ambika.patra@ltu.se
JRRT235 IMechE 2009
of necessary maintenance. Hence, the resulting costs

are related to decisions about maintenance actions
and can be estimated by probabilistic assessment of
train delay, derailment, or trafc disruptions considering the technical and operational characteristics of
the track, as well as the maintenance actions. Level
I uncertainty can also be viewed as belonging to the
external risk of the LCC analysis, where the costs
should be included to make the LCC analysis more
effective. However, there is also the level II uncertainty, which is the internal risk associated with LCC.
The level II uncertainty pertains to the variable contribution to total LCC originating from the uncertainty
in reliability and maintainability (R&M) parameters.
However, the R&M parameters also indirectly impact
the level I uncertainty. As conventional LCC analysis
only considers point estimates of R&M parameters,
it leads to an incorrect estimate of the LCC. To get
a more correct estimate of the LCC, it is essential
to also consider the interval estimate of the R&M
parameters.
There is some research related to the stochastic
nature of R&M parameters included in LCC estimation of railway infrastructure, see e.g. [1]. However, no
published research about the estimation of the uncertainty in LCC of railway infrastructure has been found.
Hence, this paper aims at describing a methodology
that can be used for uncertainty estimation in railway
infrastructure LCC.
Proc. IMechE Vol. 223 Part F: J. Rail and Rapid Transit
286
A P Patra, P Sderholm, and U Kumar
maintenance costs to be estimated. Maintenance costs

of track must include:
(a) materials, equipment, and labour;
(b) condition monitoring and inspection;
(c) track possession time.
Fig. 1
Uncertainties involved with LCC modelling
MAINTENANCE COST MODELLING
LCC, which is generally modelled in the design

phase, changes when the system enters into the
operation and maintenance phase due to changes
in stakeholder requirements, which makes the costs
incurred during the operation and maintenance
phase predominant. Maintenance costs of track
infrastructure consist of preventive maintenance
cost, renewal cost, and corrective maintenance cost
(Table 1).
Maintenance costs are the most complex cost component of an asset during its life since maintenance
is a long continuous process throughout the asset life.
While the cost of any specied maintenance work on
an asset can be comfortably estimated using engineering costing methodologies, estimating maintenance
costs throughout the asset life is a much more sophisticated process. This is because the types of maintenance are dependent on many factors, of which the
most important are asset deterioration rates, maintenance policy, and budget constraint. Maintenance
schedules therefore need to be planned to enable the
Table 1 Track maintenance at Banverket

Maintenance
strategy
Maintenance
action
Maintenance
trigger
Preventive maintenance
Rail grinding
Tamping
Rail lubrication
Ballast cleaning
Track inspection
Time
Condition
Time
Condition
Time
Renewal
(preventive
maintenance)
Rail renewal
Ballast renewal
Sleeper renewal
Fasteners renewal
Condition
Condition
Condition
Condition
Corrective maintenance
Rail replacement
Failure
The maintenance process at Banverket (Swedish

National Rail Administration) is divided into corrective and preventive maintenance, where the later
is based on condition or time. The current strategy at Banverket is to minimize corrective maintenance and to change time-based maintenance
to condition-based [2]. Table 1 shows the different track maintenance strategies and actions
at Banverket.
In this paper, maintenance costs associated with
track have been estimated separately for different
curve radii as different curve radii experience different failure probabilities and magnitudes. In this paper,
maintenance cost models have been developed with
respect to the type of maintenance intervention summarized in Table 1. The maintenance costs have been
determined as per the maintenance policy followed
at Banverket. The track has been divided into different sets of curve radii (K ), i.e. 0300 m (K = 1),
300450 m (K = 2), 450600 m (K = 3), and so on.
Curves with radius more than 2000 m have been considered as tangent track. The segmentations of the
track have been done as per of availability of the track
failure data.
A few things must be considered while performing
the segmentations of a track section. The segmentation of the track must be done for a specic track
section and should not be generalized. The segmentation of the track must done as per (a) the number
of each individual curve existing in a track section
and (b) the number of track failures occurring in each
type of curve over a period of time. For example, if
there are few curves of curve radii between 700 and
1000 m, it is safe to take 7001000 m as one segment,
whereas if there are a lot of curves existing of curve
radii between 500 and 600 m the 500600 m must be
dened as a track segment. The same logic can be
applied for the number of failures in different curve
radii. If the numbers of curves as well as the numbers
of failures are high in a particular segment, it can be
still divided into further segments. The segmentation
of track section should be specic for each studied
region, as described above.
Different track maintenance and renewal costs are
illustrated below.
2.1
Rail grinding cost
Grinding is the maintenance action done on the rail

to control rolling contact fatigue defects. Cost due to
rail grinding primarily depends on the periodicity of
JRRT235 IMechE 2009
Uncertainty estimation in railway track LCC
grinding and the number of grinding passes and is

given by
K N
1

i=1 j=1
((Tgi CL Li ngi ) + (Ceg Tgi Li ngi ))

(m/mgi )
(1 + r)j
(1)
2.6
Rail renewal cost
Rail renewal is done when the rail deterioration

reaches maintenance or safety limits. The cost due to
rail renewal is given by
K N
1

i=1 j=1
2.2 Track tamping cost

Tamping is the maintenance action done on the track
to correct its alignment. Cost due to track tamping
depends on the interval of tamping and is given by
1
K N

((Ttai CL Li ) + (Ceta Ttai Li )) (m/mtai )
(1 + r)j
i=1 j=1
(2)
2.3
Lubrication is done on the rail to control rail wear.

Cost due to lubrication depends on the number of
lubricators in the curves and the cost to maintain each
lubricator in terms of lling, which is given by
K N
1

(Tclu CL nl )
i
i=1 j=1
(1 + r)j
2.7
(3)
((Cr Li ) + (Trri CL Li ) + (Cerr Trri

Li )) (m/mrri )
(1 + r)j
(6)
Ballast renewal cost
Ballast renewal is done when ballast deterioration

ballast renewal is given by
1
K N

i=1 j=1
2.8
Rail lubrication cost
287
((Cb Li ) + (Tbri CL Li ) + (Cebr

Tbri Li )) (m/mbri )
(1 + r)j
(7)
Sleeper renewal cost
Sleeper renewal is done when the sleeper deterioration

sleeper renewal is given by
K N
1

i=1 j=1
((Cs Li ) + (Tsri CL Li ) + (Cesr Tsri Li ))

(m/msri )
(1 + r)j
(8)
2.4
Ballast cleaning cost
Ballast cleaning is the maintenance action done to

eliminate trapped water inside the ballast in order to
restore the track quality and stiffness. Cost due to ballast cleaning primarily depends on the periodicity of
ballast cleaning and is given by
1
K N

((Tbi CL Li ) + (Ceb Tbi Li )) (m/mbi )
(1 + r)j
i=1 j=1
2.9
Fastener renewal cost
Fastener renewal is done when the fastener deterioration reaches maintenance or safety limits. The cost
due to fastener renewal is given by
1
K N

i=1 j=1
((Cf Li ) + (Tfri CL Li ) + (Cefr Tfri Li ))

(m/mfri )
(1 + r)j
(4)
(9)
2.5 Track inspection cost
2.10
Track inspection is done to detect aws on the

track that can lead to failures. The cost due to track
inspection primarily depends on the interval of track
inspection and is given by
Rail replacement is done when rail breaks occur on the

track. Cost due to rail break primarily depends on the
probability of rail breaks and is given by
N
1

j=1
((Tt CL L) + (Cet Tt L)) (m/mt )

(1 + r)j
JRRT235 IMechE 2009
(5)
Rail replacement cost
K N
1

i=1 j=1
((Cr Lr ) + (Trbi CL ) + (Cer Trbi ))

(m/mrbi )
(1 + r)j
(10)
288
2.11 Track downtime cost

Downtime on the track occurs due to track possession for maintenance actions on the track. Train-free
periods are usually used for planning maintenance
actions, i.e. the hours between two consecutive trains.
However, as the train-free periods are not long
enough in most cases, this leads to train cancellations,
train speed restrictions, etc., which imply penalties
imposed on the infrastructure manager by the trafc operators. Preventive maintenance and renewal
actions are usually planned well ahead so as not to
affect the trafc. However, corrective maintenance on
the track generally affects the train operation. In this
case, rail breaks have been considered for corrective
maintenance. Mean time to repair (MTTR) for rail
break is given by

i frbi Trbi

(11)
i frbi
In this case, track possession time is calculated as
the difference between MTTR and train-free period.
Hence, the track downtime cost can be calculated
by multiplying the track possession period with the
penalty cost. Table 2 describes the R&M parameters
associated with track maintenance.
3
UNCERTAINTY IN LCC
The statistical characteristics of R&M parameters contribute to uncertainty in LCC. The reason for this is
that the times and conditions for these types of events
are so complex that they cannot be predicted with
a fair degree of accuracy. Therefore, it was decided
to explore a methodology that combines the use of
design of experiment (DoE) principles with Monte
Carlo simulation to estimate the uncertainty involved
with LCC. The area of DoE was developed in the twentieth century to increase the effectiveness and efciency
of experimentation. However, for experiments to be
effective and lead to correct conclusions there are a
number of requirements that should be fullled [3].
Table 2
R&M parameters associated

with track maintenance
Maintenance actions
R&M parameters
Rail grinding
Tamping
Rail lubrication
Ballast cleaning
Track inspection
Rail renewal
Ballast renewal
Sleeper renewal
Fasteners renewal
Rail replacement
Downtime cost
T gi
Ttai
Tlu
T bi
Tt
Trri
Tbri
Tsri
Tfri
Trbi
Trbi
mg i
mtai
mbi
mrri
mbri
msri
mfri
mrbi
frbi
For example, the response must be measurable and

be correlated to the purpose of the experiment. Furthermore, even though not an absolute necessity, the
power of statistical operations will be greater if the
response is continuous and preferably also normally
distributed. The responses of this study are the point
estimate for LCC of the track and its related uncertainty, which both are continuous, but not necessarily
normally distributed. The following are valid for the
present study.
1. The factors that are tested in the experiment are
R&M-parameters, which all are continuous and
numeric. They are also measurable, controllable,
and deemed important for the selected responses.
2. The factors that are not under investigation can easily be held constant, since the study is analytical and
not empirical. These factors are the cost factors not
directly related to R&M. Hence, no randomization
is considered necessary.
Since the study is analytical there are no major
economical constraints. Hence, the design is mostly
dependent upon the number of R&M parameters that
are to be investigated. In order to full the purpose
of this study, a two-level factorial design is considered
valuable. However, in order to reduce the number of
runs, a fractional factorial design is considered sufcient. The analysis is supported by the software tool
STATGRAPHICS, which provides suitable tables and
graphs for presentation.
The probability distribution of LCC can be found by
the use of Monte Carlo simulation. A Monte Carlo simulation is effectively a random number generator that
creates values for each R&M parameter.Values are chosen within specied ranges of each parameter and with
a frequency proportional to the shape of probability
distribution associated with each R&M parameter. The
proposed methodology helps in determining the variable costs associated in LCC estimation. These variable
costs can be termed as the uncertainty in LCC estimation and are caused by the probabilistic nature of
the R&M parameters. The LCC becomes more robust
when these variable costs are included. Thus, it helps
the decision makers to make more effective decisions
on maintenance policy by considering LCC.
4
CASE STUDY
The performed case study was on the iron ore line

(Malmbanan) that runs from Lule in Sweden to
Narvik in Norway. The line allows 30 tonne axle load
with mixed trafc. Data (Tables 3 and 4) was collected
from Banverkets failure and maintenance databases
(i.e. BIS, Bessy, and 0felia) that range from 1997 to 2006
with some data being collected from reference [4]. The
study was performed on the rail replacement cost on
high and low rails separately. Low rail denotes the inner
JRRT235 IMechE 2009
Table 3 TTF (rail break) data in

MGT for curves of radius
450600 m
High rail
Low rail
400
350
250
425
300
325
150
350
150
400
275
575
325
350
150
225
275
425
300
125
150
400
300
Table 4 TTR (to correct rail break)

data in minutes for curves
of radius 450600 m
High rail
Low rail
159
120
480
149
270
547
340
43
228
202
240
218
258
154
216
240
169
75
340
202
202
216
240
rail (smaller radius) and high rail the outer rail (larger
radius) in a curved track. The idea of separating high
rail and low rail for cost estimation lies in the fact that
they both have different failure deterioration due to
quasi-static forces in the track curvatures.
The following assumptions were made after consultations with Banverkets track experts in the
case study.
1. Average gross tonnage per year is assumed to be 25
million gross tonnes (MGT).
289
2. Life of track for LCC estimation is 600 MGT (24

years).
3. Discount rate is taken as 4 per cent.
4. Cost of BV50 rail (including neutralisation) is 1395
Swedish kronor (SEK)/m.
5. Average labour cost is 525 SEK/h. This includes the
track worker cost, track welder cost, and inspection
personnel cost.
6. Welding equipment cost is 60 SEK/h.
7. Average length of rail replacement due to rail break
(Lr ) is 8 m.
LCC analysis was done on curves of radius
450600 m, with cost gures given in SEK. The time to
failure (TTF) and time to repair (TTR) data obtained
from the Banverket data base were analysed using
probability distribution models. However, before tting any distribution models to analyse the data, the
TTF and TTR data sets were veried for (independent
and identically distributed (IID) random variables)
assumption using graphical method (Figs 2 to 5). This
is important because if data is not independent or it
has trend, then probability distribution models cannot
be used for analysing the data set [5]. Such data sets
can be modelled by the use of other non-stationary
model such as power law process model, etc [6].
Table 5 shows the probability distribution of mean
time to failure (MTTF) and MTTR for both high and low
rails. The analysis was supported by the software tool
Weibull++. MTTF was estimated by considering the
failure events (time period to occurrence of rail break)
and suspended events (no rail break has occurred) for
the particular curvatures of the track. MTTR considered here comprises of the logistic time, welding time,
and inspection time necessary to repair the rail breaks.
A two-sided 90 per cent condence level was considered for determining the upper limit, mean, and lower
limit of MTTF and MTTR.
Table 6 shows the LCC estimation by considering
DoE principles. The high and low rails were analysed separately, but followed the same design. The
applied design was a screening, full factorial, twolevel design with the two experimental factors MTTF
and MTTR, i.e. a 22 -design that requires four runs.
Fig. 2 Test for IID for TTFs of the high rail

JRRT235 IMechE 2009
290
Fig. 3 Test for IID for TTRs of the high rail
Fig. 4 Test for IID for TTFs of the low rail
Fig. 5 Test for IID for TTRs of the low rail
These four runs were performed ten times (i.e. ten

blocks with four runs in each), resulting in a total of
40 runs for high and low rails, respectively. The high
and low levels for MTTF and MTTR were selected as
the upper and lower limits of their distributions. The
experiment contained two responses, i.e. the point
estimate and the log(s2 ) of LCC. The rationale for
analysing log(s2 ) is described in reference [7]. The
input data was generated by Monte Carlo simulations.
These data were entered into equation (10) and varied according to the experimental design summarized
in Table 6.
Table 6 indicates that log(s2 ) of LCC is quite stable for

both high and low rails. However, changes in the levels
of MTTF and MTTR do affect the variability in LCC.
Since there is no interaction effect present (see Fig. 6
for example), the factors can be considered individually. An interaction between two factors means that
the effects of either one cannot be judged independently. If there is an interaction between two factors,
the effect of one factor on the response will depend
on the setting of the other. In order to reduce the variability in LCC, one should look into the lowest value
of log(s2 ). The effects of variability in MTTF and MTTR
JRRT235 IMechE 2009
Table 5
MTTF (MGT)
MTTR (hours)
MTTF and MTTR probability distributions for high and low rails
High rail
Low rail
Probability
distribution
Log normal ( =
5.9933, = 0.2523)
Upper limit
Mean
Lower limit
482.7
413.6
354.5
Weibull- 2 parameter
( = 369.7161,
= 3.5315)
403.5
332.8
274.4
Probability
distribution
( = 4.6972,
= 1.8871)
5.5
4.2
3.1
Upper limit
Mean
Lower limit
Table 6
MTTF
MTTR
LCC (average)
Log (s 2 )
High rail
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1050.4
940.6
1086.6
973.0
1252.1
1113.3
1288.3
1130.4
3.4198
3.4784
3.4572
3.5139
3.0577
3.0128
3.0824
3.0230
Fig. 6
Normal ( = 3.4458,
= 1.0296)
3.9
3.4
2.9
LCC estimation with DoE principles
Type
Low rail
291
Fig. 7
Effect of MTTF and MTTR on uncertainty of LCC

for high rail
Fig. 8
Effect of MTTF and MTTR on uncertainty of LCC

for low rail
Interaction plot showing variability in LCC for

low rail
on the LCC of high and low rails are shown in Figs 7

and 8, respectively. The gures show the effect on LCC
with increase of MTTF and MTTR values from low to
high levels.
As shown in Figs 7 and 8, MTTF has a positive effect
on LCC and MTTR has a negative effect. The magnitudes of the effects imply that the uncertainty in
MTTF has more impact on the change in LCC than
the uncertainty in MTTR. Two possible reasons for
these differences in magnitudes are uncertainty levels in the parameters and given importance levels in
the LCC formulation. The interaction between MTTF
and MTTR is not signicant in any of the cases.
Monte Carlo simulation was used to determine
the probability distribution of LCC and estimate the
JRRT235 IMechE 2009
associated variability cost. A two-sided 90 per cent

condence level was considered for this distribution.
LCC gures were generated by combinations of upper,
mean, and lower limits of MTTF and MTTR that were
generated by Monte Carlo simulation. As shown in
Table 7, the difference between upper and lower limits
can provide the variability cost associated with LCC.
The negative sign on the costs indicate that they were
calculated as total present value.
292
Table 7
LCC (SEK)
Simulated probability distribution of LCC
Probability
distribution
Upper limit
Mean
Lower limit
High rail
Low rail
Probability ( =
219.6026, = 3.0731,
= 815.1878)
1024.9
1011.5
999.0
( = 170.7607,
= 2.2115,
= 1049.3146)
1214.7
1200.5
1187.6
DISCUSSION AND CONCLUSIONS
LCC is being used as a tool to help in making effective

maintenance decisions. However, there are various
uncertainties associated with estimation of LCC. This
paper presents level I and II uncertainties, out of
which level II uncertainty has been dealt with. Level II
uncertainty can be due to economic parameters, e.g.
discounting rate, which has not been explored in this
paper. The paper investigates more the uncertainties
caused by technical parameters, i.e. R&M parameters. The uncertainty in R&M parameters exist because
of their probabilistic nature, which contributes to
the uncertainty in LCC estimation. For better estimation of uncertainty in LCC, this paper outlines
a methodology based on a combination of Monte
Carlo simulation and DoE. This combination gives
a possibility to identify parameters that are inuential on the LCC estimation and its variability. The
proposed methodology can be used to estimate the
uncertainty in LCC by considering uncertainties in
all parameters simultaneously, in contrast to sensitivity analysis, where the parameters are considered
one by one. Hence, the methodology can contribute
to other research efforts, where traditional sensitivity analyses have been performed. The simulations
are used to make the deterministic LCC equations
probabilistic. DoE is applied to guide how the R&M
parameters should be varied in a systematic way.
The paper also illustrates cost models for different
maintenance and renewal actions carried out on
track. The uncertainty in LCC is presented as variable
costs with associated distributions. When the variable costs are added to the LCC it becomes more
robust. Hence, it helps the decision-makers to make
more effective decisions about maintenance policy by
considering LCC.
For further research, all the developed cost models for railway track can be combined into one model.
The proposed methodology can then be applied to this
new cost model. However, one major challenge will
be to get relevant data to use as input to this sensitivity analysis. Another challenge is to deal with the
large number of runs that will result by a full two-level
factorial design, since the number of runs will double
with each added parameter. However, this is not any
major problem since DoE principles can be applied to

reduce the number of runs by using fractional factorial
designs that still will give valuable information.
REFERENCES
1 Zhao, I., Chan, A. H. C., Roberts, C., and Stirling, A. B.
Assessing the economic life of rail using a stochastic analysis of failures. Proc. IMechE, Part F: J. Rail and Rapid
Transit, 2006, 220(F2), 103110.
2 Karlsson, V. An overall view of maintenance. Eur. Railw.
Rev., 2005, 11(3), 1117.
3 Coleman, D. E. and Montgomery, D. C. A systematic
approach to planning for a designed industrial experiment. Techometrics, 1993, 35, 127.
4 Kumar, S. A study of the rail degradation process to
predict rail breaks. Licentiate Thesis, Division of Operation and Maintenance Engineering, Lule University of
Technology, Lule, Sweden, vol. 73, 2006.
5 Ascher, H. and Feingold, H. Repairable system reliability: modelling, inference, misconceptions and their causes,
1984 (Marcel Dekker, New York).
6 Kumar, U. and Klefsj, B. Reliability analysis of hydraulic
system of LHD machines using power law process model.
Reliab. Eng. Syst. Saf., 1992, 35, 217224.
7 Bisgaard, S. and Fuller, H. T. Reducing variation with twolevel factorial experiments. Qual. Eng., 1995, 8, 373377.
APPENDIX
Notation
Cb
Ceb
Cebr
Cefr
Ceg
Cer
Cerr
Cesr
Cet
Ceta
Cf
CL
Clu
cost of ballast in SEK/km

equipment cost for ballast cleaning in SEK/h
equipment cost for ballast renewal in SEK/h
equipment cost for fastener renewal in SEK/h
equipment cost for grinding in SEK/h
equipment cost to repair rail breaks in SEK/h
equipment cost for rail renewal in SEK/h
equipment cost for sleeper renewal in SEK/h
equipment cost for track inspection in SEK/h
equipment cost for tamping in SEK/h
cost of fasteners in SEK/km
average labour cost in Swedish Kroner
(SEK)/h
cost of lubrication material for each
lubricator per year in SEK
JRRT235 IMechE 2009
Cr
Cs
frbi
K
L
Li
Lr
mbi
mbri
mfri
m gi
mrbi
mrri
msri
mt
mtai
cost of rail in SEK/km

cost of sleeper in SEK/km
failure rate of rail (breaks) in the ith curve
class of curve radii
total length of track section in km
length of ith curve in km
average length of rail replacement due to rail
break
interval for ballast cleaning for ith curve in
MGT
interval for ballast renewal for ith curve in
MGT
interval for fastener renewal for ith curve in
MGT
interval for grinding for ith curve in MGT
mean time to rail breaks in ith curve in MGT
interval for rail renewal for ith curve in
MGT
interval for sleeper renewal for ith curve in
MGT
interval for track inspection in MGT
interval for tamping for ith curve in MGT
JRRT235 IMechE 2009
M
M
ngi
nli
N
r
Tbi
Tbri
Tfri
T gi
Tlu
Trbi
Trri
Tsri
Tt
Ttai
293
gross tonnage per year in MGT

life period of track in MGT
number of grinding passes on ith curve
number of wayside lubricators in ith curve
life period of track (equivalent to M ) in years
discount rate
mean time to clean ballast for ith curve in
h/km
mean time for ballast renewal for ith curve in
h/km
mean time for fastener renewal for ith curve
in h/km
mean time to grind for ith curve in h/km
mean time to rell lubrication material for
each lubricator in hour
mean TTR rail break in ith curve in hour
mean time for rail renewal for ith curve in
h/km
mean time for sleeper renewal for ith curve
in h/km
mean time to inspect track in h/km
mean time to tamp for ith curve in h/km

Rams

Uploaded by

Document Informationclick to expand document informationRAMS

Document Informationclick to expand document information

Copyright:

Available Formats

Rams

Uploaded by

Document Information

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

Rams

Uploaded by

Copyright:

Available Formats

DOC TOR A L T H E S I S

ISSN: 1402-1544 ISBN 978-91-7439-023-0

Division of Operation and Maintenance Engineering

Maintenance Decision Support Models

Ambika Prasad Patra

Maintenance Decision Support Models for

Ambika Prasad Patra

Lule University of Technology

Printed by Universitetstryckeriet, Lule 2009

Ambika Prasad Patra

List of Appended Papers

Patra, A. P., Kumar, U. and Larsson-Krik, P-O. (2010). Availability target

Patra, A. P., Kumar, U. and Larsson-Krik, P-O. (2009). Assessment and

Patra, A. P., Dersin, P. and Kumar, U. (2009). Cost effective maintenance

Patra, A. P., Sderholm, P. and Kumar, U. (2009). Uncertainty estimation in

1.1 Research Problem

1.2 Research Objectives

To study RAMS methodologies for the railway infrastructure and develop a

To study the applicability of RAMS tools in track maintenance planning and to

1.3 Research Questions

1.4 Scope and Delimitations of the Study

2 Basic Concepts and Definitions

Reliability and maintainability management is attracting new interest in todays corporate

2.1.1 RAMS Parameters

Basic Concepts and Definitions

Major (service failure)

2.1.2 Factors Affecting RAMS

Maintenance conditions: the sources of failures are caused by maintenance actions.

Table 2.2: Factors affecting track RAMS

Change in track stiffness

Change in friction co-efficient

Renewal of track components

Corrective replacements of track components

Failure rate of track components

Loads (annual MGT, maximum axle load)

Bending stress, Shear stress, Contact stress

Vehicle operating conditions:

Vertical stress, Lateral stress

Vehicle condition (hollow wheels)

Figure 2.1: Factors influencing railway RAMS (EN 50126, 1999)

Track geometry condition (standard deviation in the longitudinal level, standard

Measuring the infrastructure condition is prerequisite for track maintenance planning.

Basic Concepts and Definitions

2.2 Infrastructure Maintenance Planning

Generation of maintenance strategies for individual assets (e.g. corrective or

Analysis, Planning & Schedule

Figure 2.2: Maintenance planning overview (Zarembski, 1998)

Basic Concepts and Definitions

Rail wear rate

Rail life in MGT

Rail fatigue life

Material removal rate by grinding and wear

2.2.1 RAMS Analysis for Maintenance

Basic Concepts and Definitions

Methods and tools for RAMS analysis

III) Work history

2.2.2 RAMS in the Operation and Maintenance Phase

Basic Concepts and Definitions