RAMS analysis of railway track infrastructure

(Reliability, Availability, Maintainability, Safety)

Extended Abstract

Gonçalo Medeiros Pais Simões

Dissertation for the qualification in the Msc. degree in

Mestrado Integrado em Engenharia Civil

Jury
Presidente: Prof. Joaquim Jorge da Costa Paulino Pereira
Orientador: Prof. Paulo Manuel da Fonseca Teixeira
Vogais: Prof. José Manuel Care Baptista Viegas

September 2008
 Index
INDEX ............................................................................................................................................2
CHAPTER 1: INTRODUCTION.....................................................................................................3
CHAPTER 2: RAMS TECHNIQUES.............................................................................................3
2.1 THE RAMS CYCLE .................................................................................................................4
2.2 RAMS PARAMETERS..............................................................................................................5
2.3 FORECASTING FAILURES AND MDT .........................................................................................5
2.4 RELIABILITY DATA SOURCES AND COLLECTIONS .......................................................................6
CHAPTER 3: THE RAILWAY TRACK INFRASTRUCTURE .......................................................6
3.1 RAIL CHARACTERISTICS ..........................................................................................................7
3.2 RAIL DEFECTS........................................................................................................................9
CHAPTER 4: RAMS APPLIED TO THE RAIL TRACK INFRASTRUCTURE .............................9
4.1 FAILURE RATE AND RELIABILITY PREDICTION ............................................................................9
4.2 PREDICTION OF AVAILABILITY ................................................................................................11
CHAPTER 5: CONCLUSIONS....................................................................................................12
BIBLIOGRAPHY .........................................................................................................................13

2
Chapter 1: Introduction

This study deals with the application of Reliability, Availability, Maintainability and Safety
(RAMS) techniques to the railway track infrastructure. Special emphasis is given to the
prediction of costs related to failures displayed by the rail component in high speed railway track
infrastructures, from situations where failure data is scarce.

In Chapter 2, RAMS techniques are presented. The RAMS cycle is described. Relevant RAMS
related concepts and methods of simulation are listed. Failure forecasting techniques are
presented for the special case of non-constant failure rate. Reliability data collection issues are
overviewed.

In Chapter 3, the railway track infrastructure is presented. Special emphasis is given on the rail.
Rail related parameters which influence rail defects are discussed. A list of significant rail
defects is presented. Rail maintenance related actions are listed.

Chapter 4 is dedicated to the application of RAMS techniques to the railway track infrastructure.
Section 4.1 shall deal with the prediction of reliability and failure rate, which in turn lead to
forecasts of failures. It is then shown how to use these forecasts to assess costs. Section 4.2
presents a suggestion of a method to predict availability. The link between availability and costs
is also presented.

In this study the following issues are not included: definition of specific risk levels for rail tracks;
definition of associated RAMS targets to fulfil regarding potential risks and hazards that might
occur in rail tracks; presentation of methods of RAMS simulations that allow for the verification
of RAMS targets for the considered case study (rail tracks)

Chapter 2: RAMS techniques

RAMS is a branch of engineering which started to be developed with the notion of mass
1
production. Its goals are to assess the predisposition of any equipment, product or system to
failure. RAMS comprehends tools which enable the engineer to predict the failures and their
degree of severity, system outage (down time). These forecasts are then used to assess costs,
which are used as input for cost benefit analyses (CBA). A growing aspect of RAMS is the
assessment of safety, which was mainly propelled by the growth of the nuclear and chemical
industries.

1
Hereby the word “system” shall always be used

3
2.1 The RAMS cycle

Figure 2.1 The RAMS cycle.

The RAMS cycle is shown in Figure 2.1. The process comprehends three levels:
1. RAMS data compilation. Data from component tests or from direct field use is
compiled or collected in order to be used as input for future RAMS forecasts which use
similar components. In the scope of RAMS, “component” is normally said to be the least
replaceable part which can be tested in a practical manner. This happens both for
maintenance and testing reasons.
2. RAMS simulation. A model of the system is built, taking in account the available
component data and the interactions between components.
3. Life cycle assessment. The stages presented in Figure 2.1 correspond to the normal
main life cycle stages of any system. RAMS targets are set before each stage. After
RAMS simulation, the targets are revised and reset for the next stages.
There is a strong connection between life cycle and RAMS simulation, which is stressed by the
red ellipse in Figure 2.1. Desired RAMS targets may lead to redesign or changes in operation or
maintenance schemes. On the other hand, desired operation schemes influence RAMS
parameters and expectable (feasible) RAMS targets. Finally, the closed loop from stage 3 to
stage 1 presented in Figure 3.1 refers to reliability growth. This is the process through which
RAMS data and simulation is refined. Collection of data from field operation of systems leads to
better data compilations. This leads to a larger degree of trust in RAMS simulation, which in turn
improves the results of life cycle assessment.

4
2.2 RAMS parameters

A list of RAMS parameters and their applicability is presented here:

- Failure Rate (λ): Probability of failure at a given system age. Expressed in failures per
time or load unit. Used for values of system failure and unavailability.
- Mean Time Between Failures (MTBF or θ(t)) and Mean Time To Failure (MTTF):
Used to simulate the effectiveness of different maintenance policies and their costs.
- Reliability (R(t))/Unreliability: Used mainly where there are security concerns and
therefore the probability of failure is of interest.
- Maintainability: Seldom used as such (see Smith, 2005).
- Mean Time To Repair (MTTR): Useful when calculating costs associated with loss of
income due to outage of the system, and with maintenance operations.
- Mean Down Time (MDT): Used in similar situations as MTTR, since they may overlap.
Maximum MTTRs and MDTs are meaningless for they are known to be log-Normally
2
distributed (see Smith, 2005) .
- Availability/Unavailability (Probability of Failure on Demand, PFD): Used in the
same situations as MDT.

Reliability may be expressed through failure rate with equation 2.1. MTBF may be expressed
through reliability with equation 2.1 and 2.2:
∞
R(t ) = exp − ∫ λ (t )dt  and θ (t ) = ∫ R(t )dt
t
(2.1; 2.2)
 0  0

Availability (A) may be expressed through MDT with equation 2.3:

Up Time Up Time MUT
A= = = (2.3)
Total Time Up Time + Down Time MUT + MDT

2.3 Forecasting failures and MDT

Failures may be forecast from data sets of failures. In the special case of non-constant failure
rate – the relevant case for rail tracks, as will be seen in chapter 3 – the best way to statistically
predict failure is through the use of the Weibull distribution. This is a versatile distribution for it
may be adjusted through three calibration parameters, β, γ and α. This distribution is applied to
failure data sets. There must be a least amount of 6 points in the data sets, which barely need
to indicate the value of the unit when the failure has occurred whether this unit is time, number
of load cycles or other. Reliability and failure rate may be expressed through 2.4 and 2.5:

2
For statistical purposes, IEC 61508 states that access and diagnostic test time intervals may be omitted
from the sum of MTTR if the diagnostic tests are run almost continuously (see Lundteigen and Rausand et
al, 2006), a feature common in more modern equipments.

5
   t − γ β  β t −γ 
β −1
R (t ) = exp −    and λ (t ) =   (2.4; 2.5)
  α   α  α 

Mean Down Time may be forecast through the use of the logNormal distribution (eq. 2.6):

 ln t − µ 
MDT (t ) = Φ  (2.6)
 α 

There is a strong recommendation from RAMS authors to clearly separate failure modes when
3
collecting failure data and trying to apply any statistical distribution to it .

2.4 Reliability data sources and collections

Since RAMS is a comparatively new engineering field, failure data collections are few and non-
standardized in several levels. That is, from data collection to data display. For instance, what
some engineers might account as failures might not be accounted by others, component types
tested may have a broad range of specifications, environmental parameters may differ or be
unspecified, different failure modes may be blended in the data sets, among other aspects. The
more specific the databank is, the best results RAMS models using those banks will deliver.

Hence, when using failure data lists, each author’s recommendations must be followed
thoroughly. Most of the failure data lists generally refer to random failures, i.e. constant failure
rates. Some display 3 columns, with a lower and higher bound and a mean value. Others yet
display equations with a scope of application. These may not be used to extrapolate values.

Data collection should start with careful engineer training. As for data collection itself, the
engineers should record information regarding the type and nature of failure, its effects,
location, component’s installation and running time and environmental conditions. For
maintainability purposes, engineers should record actions taken and resources needed to
correct the reported failure (spares, crew, and time), cost of outage and maintenance costs.

Chapter 3: The railway track infrastructure

The railway track infrastructure system is presented in Figure 3.1. A list of the components is
presented below. Base and formation layer constitute the subgrade, and the rest are named

3
Drenick’s law states that if 3 failure modes are combined in a data set, then the general behaviour read is
that of a constant failure rate, even if each failure mode behaves with non-constant failure rate, with very
different calibration parameters β and α

6
superstructure. Figure 3.1 refers to the flexible support (ballast and subballast). In the inflexible
support ballast is replaced by concrete slab and subballast by ballast concrete layer.

Figure 3.1 Typical cross section of a railway double track infrastructure.

1. Base + formation layer (subgrade). Consolidated soil designed to resist mechanically.

Subgrade is classified from S3 (good support) to S0 (very weak support).
2. Subbalast or ballast concrete layer. Used for load distribution and to prevent ballast
stones from penetrating subgrade. Ballast concrete layer is used to level the slab.
3. Ballast or concrete slab. Used for load distribution, noise and vibration dampening,
ride comfort increase and drainage of rain water. Concrete slab has a narrower section.
4. Sleepers (or ties – U.S.). Used as a support for the rails, normally concrete monoblock
or twinblock in highspeed infrastructures. Also available in wood or steel.
5. Rail pads. Made from rubber or cork, used to dampen high frequency vibrations.
6. Fastenings. Connect rail and sleeper. Elastic in concrete sleepers (high speed tracks).
7. Rails. Made from steel, may differ in size, shape (cross section), inertia, mass, steel
grade and gauge (positioning). Used to guide and support rolling stock and locomotives.

3.1 Rail characteristics

The first aspect to consider is the axle load the track is subject to. Axle loads range in 4
categories, from A (maximum load per axle 16t) to D (maximum load per axle 22.5t). Some
networks may use larger loads. Traffic load is calculated through expression 3.1:

T = S × (T p + k fr × T fr + k tr × Ttr ) [Million Grosse Tonnes, MGT] (3.1)

Where
S=1.0 for lines without passenger traffic S=1.25 for lines with mixed

S=1.1 for lines with mixed traffic and Vmax < 120km / h traffic and Vmax > 140km / h
S=1.2 for lines with mixed traffic and Tp is the daily passenger
120km / h < Vmax < 140km / h vehicle traffic

7
Tfr is the daily freight vehicle traffic kfr=1.15
Ttr is the daily traction engine traffic ktr=1.40

In high speed railway tracks, the general cross section type in use is the flat bottom rail (Vignole
type rail). Several profiles have been standardized – BS in the UK, S in Germany, UIC in other
countries (Union Internationale des Chemins de fer). Some are shown in Table 3.1:

Table 3.1 Rail profiles in commercial use.

Weight m a Head b Height c Base d Web
Section Shape
[Kg/m] [mm] [mm] [mm] [mm]
BS75R 37.041 61.91 128.59 122.24 13.1
BS80A 39.761 63.5 133.35 117.47 13.1
S41-R14 40.95 67 138 125 12
S49 49.3 67 149 125 14
UIC60E1 60.21 72 172 150 16.5
UIC71 71.19 74 186 160 18

Steel grades used in Vignole-type rail manufacturing are described in European Norm EN13674
Part 1. This norm specifies seven different steel grades, which are presented in Table 3.2. An
th
older for the choice of steel grades was Codex UIC 860 V, 8 edition, dated from 1986.

Table 3.2 Steel grades described in European Norm EN13674.

Steel grade R200 R220 R260 R260 Mn R320 Cr R350 HT R350 LHT
Minimum tensile
680 770 880 880 1080 1175 1175
strength [MPa]

High speed tracks are composed of continuously welded rails (CWR). Welding methods differ:
→ Electric flash butt weld: achieved through application of voltage and pressure to rails.
→ Thermit weld: made using heat supplied by an alumina conversion.
→ Electric arc weld: made using propane burners and a copper casing.
→ Oxyacetylene weld: made heating rail by burning a mix of oxygen and acetylene.
→ Pressurised weld: the working pieces are pressurised together. Friction causes forging.
→ Induction weld: heat is created by resistance to an electromagnetic field applied to steel.
→ Resurfacing: oxyacetylene is combined with a ferrous alloy melted to recreate surfaces.

8
3.2 Rail defects

Rail defects are caused by initial microscopic manufacturing imperfections, which grow larger in
4
time due to the passing of trains, i.e. they are rolling contact fatigue (RCF) phenomena . Track
defects are deviations in track geometry and are not treated in this study. Several defects have
been classified in different standards, including UIC. Significant rail defects are:

Maintenance actions destined to correct rail defects after visual or ultrasonic inspection are:
- Rail grinding: corrects RCF defects which appear on the rail surface through grinding.
- Reprofiling: the profile is fully corrected through the use of a planing cutter machine.
- Replacement: full replacement of seriously affected track stretches is made.
- Weld straightening: destined to correct weld geometry through profile bending.
- Lubrication (preventive): reduces wear, mainly in gauge corners of curves with low radii.

Chapter 4: RAMS applied to the rail track infrastructure

4.1 Failure rate and reliability prediction

The statistical study of failure needs in the first place the collection of data. As stated above, a
least amount of 6 values of MGT to failure need to be recorded. In order to have statistically
independent variables, each rail defect should be studied separately. Furthermore, comparable

4
This means that rail defects have a non-constant failure rate. It increases with time

9
stretches of track need to be created, for different conditions may be available. So, forms
containing the following infrastructure and train traffic characteristics should be filled in:

- Type and quality of rail - Special track stretch (when

- Type and quality of fastenings applicable: bridge, tunnel)
- Type and quality of sleepers - Subgrade quality
- Ballast or slab technical - Installation date of components
characteristics - Traffic speed and direction
- Curve radius and super-elevation - Axle load of trains / locomotives
- Track gradient - Maintenance history on the stretch

Comparable stretches of track shall then be stretches in which these characteristics are
coincident within ranges defined by the rail infrastructure maintenance manager. The failure
characteristics to record should also be filled in as a form containing:

- Classification of failure (rail - Method of repair (action taken)

defect) – see Chapter 3.2 - Welding process at repair
- Cumulative tonnage to failure - Post-weld treatment
(in MGT) - Time to repair
- Time to failure (optional) - Failure location (on the rail network)

Failure rate and reliability prediction may be obtained from equations 2.4 and 2.5. Parameter γ
is the location parameter. It may be considered to be null since at 0 MGT, 0 failures are to be
expected. Plotting is done by first sorting ln(MGT) in ascending order on the x-axis and
assigning to each a value of ln(ln(1/(1-Median Rank))) on the-y axis (see Figure 4.1). Median
Ranks are obtained by Bernard’s approximation given by expression 4.1:
r − 0 .3
N + 0 .4 (4.1)

Weibull plot analysis and linear regression for the data set
"Gauge corner wear > 2mm"

1,5
1,0
ln(ln(1/(1-Median Rank)))

0,5
0,0
y = 5,7575x - 37,306
-0,5
R2 = 0,9465
-1,0
-1,5
-2,0
-2,5
-3,0
6,0 6,1 6,2 6,3 6,4 6,5 6,6 6,7 6,8

ln(MGT)

Figure 4.1 Example of a Weibull plot analysis and linear regression for a 10 point data set
referring to the defect Gauge corner wear > 2 mm.

10
Applying a linear regression to this plot, the y=m.x+b equation returns the values of the Weibull
parameters. The slope value m is β. Parameter α is obtained by applying expression 4.2:
b 
−  
β
α =e 
(4.2)

The values for parameters α and β are then introduced in expressions 2.4 and 2.5, allowing
reliability and failure rate to be computed. MGT is easily estimated through the train traffic in
every year, so estimations on failure rate and MGT can be made. If costs are assigned to each
failure, Pareto analyses may be carried in order to compare the relevance each rail defect in
5
terms of costs . Rail life and scheduling of maintenance actions are also eased since failures
may be predicted. Imputation of costs by modelling at component level may be done since all
different component interactions are skimped by splitting track stretches into comparable units.

4.2 Prediction of availability

Availability may be predicted through equation 2.3. This can be rewritten into:

MUT ST − MDT MDT

A= = = 1− (4.3)
MUT + MDT ST ST

In expression 4.3 MDT is Mean Down Time and ST is scheduled train traffic. The suggestion is
to split down time into several components (see equation 4.4):

MDT = MDTi + MDT pm + MDTcm + MDTs (4.4)

The several components are:

1. MDTi: Down Time due to traffic halting for inspection. This is the inspection time which
is made outside the free slots, i.e. slots of time when no train traffic is scheduled.
2. MDTpm: Down Time due to preventive maintenance. Similar to MDTi, this is the
preventive maintenance which falls outside the free slots.
3. MDTcm: Down Time due to corrective maintenance. Involves failure detection, waiting
for maintenance crew and/or spares and/or equipment, repair and restoring of system.
4. MDTs: Down Time due to slowing of train traffic for maintenance purposes: It is the
arithmetic difference between the time trains the time trains take with the speed
restriction and without it.
ST is set by the train traffic operator, so it is known. All of the MDT parameters are of hard
assessment since they are interdependent. It is also known that preventive maintenance might

5
It is a recommendation of RAMS authors to use predictions in comparisons rather than for
absolute values since results need to be confirmed by field operation data

11
improve failure modes at the expenses of others. Many different maintenance strategies are
available. One suggestion is to divide resources into maintenance jurisdiction areas, since the
geographical location of failures may be forecast with the method presented in chapter 4.1.
Further studies are suggested where data collected by the abovementioned forms is used to
collect MDT data series. These should then be studied with a logNormal distribution, according
to the recommendations of RAMS authors.

Chapter 5: Conclusions
The goal of this study is to assess how RAMS techniques can be applied to the rail in order to
forecast failures and their associated costs when data sets are short.

Recommendations for the collection of failure data were presented, which take in consideration
the recommendations of RAMS authors and the functioning and displayed failure modes (i.e.,
defects) of the rail. This process was designed to obtain results directly from the rail component,
skipping RAMS modelling at system level (here, the word system refers to the whole rail
infrastructure). This was achieved through the splitting and separate study of data sets into
different train traffic and infrastructure conditions. It was shown that expressions for reliability
and failure rate may be obtained by applying the Weibull distribution to the data sets obtained
by the previously mentioned process. The Weibull distribution was shown to be suitable to short
data sets, since it works with a least amount of 6 points. Also, the Weibull distribution was
shown to be suitable for the study of the rail due to the fact that it allows to model the behaviour
of components with non-constant failure rates. It was also suggested to use these expressions
on Pareto analyses, allowing to compare the relative importance of failures in maintenance
costs and down time and to obtain an order of magnitude of their values. It was stated that
RAMS predictions should be used mainly in comparative analyses. This was a reason which led
to the suggestion of the use of Pareto analyses in the study of the costs connected to failure,
since they are mainly comparison techniques.

Prediction of availability was discussed. It was shown that availability may be calculated based
on estimations of mean up time and mean down time. MUT is set by the train operator. A break
down of MDT into a sum of several different components was displayed. This was concluded to
be of hard assessment. Firstly, these components are dependant on each other. Secondly the
different logistic factors and maintenance strategies available lead to a fluctuation of the
observed MDT values. Furthermore, other studies have already shown that preventive
maintenance may lead to lower failure rates in some failure modes at the expenses of higher
failure rates in other failure modes. It was recommended to collect more data on MDT. It was
stated that prediction of MDT is achieved through the log-Normal distribution, which may be
done in the future if more data is collected. With the prediction of MDT it is then possible to
estimate availability.

12
Bibliography

AWS (2004). AWS Welding Handbook Volume 2 – Part 1: Welding Processes (WHB V2).
American Welding Society.

nd
CANNON, D. F. (2003). An international cross reference of rail defects, 2 edition. The Steering
Group of UIC/WEC - Union Internationale des Chemins de Fer, Paris, France.

DORNER, William (1999). Using Excel for Weibull Analysis. Quality Digest.

DOWSON (1999). Lubrication at the Frontier: The Role of the Interface and Surface Layers in
the Thin Film and Boundary Regime. Elsevier.

nd
ESVELD, Coenraad (2001). Modern Railway Track. 2 edition. MRT Productions, Delft
University of Technology.

H.J. Skelton Canada, Ld. http://skelton-metals.com/railsections.htm

IEC (2002). International Standard IEC 62278: Railway applications – Specification and
demonstration of reliability, availability, maintainability and safety (RAMS). First Edition 2002-09.
International Electrotechnical Commission.

KHUMAR, S. (2007). Study of Rail Breaks: Associated Risks and Maintenance Strategies.
Luleå Railway Research Center (JVTC)

KRULL, R. Hintze H., LUKE, THOMAS M., POHL R., RÜHE S. (2002). Eddycurrent Detection of
Head Checks on the Gauge Corners of Rails: Recent Results. DB AG, the Research and
Technology Centre; BAM Federal Institute for Materials Research; Prüftechnik Linke & Rühe
GmbH, Germany

LUNDTEIGEN, Mary Ann & RAUSAND, Marvin (2006). Assessment of Hardware Safety
Integrity Requirements. Proceedings of the 30th ESReDA Seminar, June 07-08, 2006,
Trondheim, Norway.

MANLY P., Harold (2006). Oxy-acetylene welding and cutting. Indypublish.com.

MATHER, Daryl (2005). The Maintenance Scorecard – Creating Strategic Advance. Industrial
Press, Inc.

MILLS (1956). Materials of Construction, Their Manufacture and Properties. Wiley.

13
NAHMAN, Jovan M. (2002). Dependability of Engineering Systems – Modeling and Evaluation.
Springer.

NOWLAN, F. Stanley & HEAP, Howard F. (1978). Reliability-Centered Maintenance. United

Airlines, USA.

NUNNS, S. R. (2002). Guiding principles to the implementation of IEC-61508. Proceedings of

th th
the SIPI Workshop 3, the 14 and 15 of November 2002, London, UK.

nd
PROFILLIDIS, Vassilios A (2000). Railway Engineering. 2 edition. Ashgate Publishing Limited.

SMITH, David J. (2005). Reliability, Maintainability and Risk – Practical Methods for Engineers.
th
7 edition. Butterworth-Heinemann.

SPG Media PLC. Web source: http://www.railway-technology.com/projects/germany/

Transportation Safety Bureau Canada.

Web source:
http://www.tsb.gc.ca/en/publications/reflexions/rail/2003/issue_19/rail_issue19_sec2.asp

U.S. Dept. of the Army (1991). Railroad Track Standards: Appendix B - Field identification of
rail defects. Departments of the Army and Air Force.

VILLEMEUR, Alain (1992-A, 1992-B). Reliability, Availability, Maintainability and Safety

Assessment, A: Vol. 1: Methods and Technique. B Vol.2: Assessment, Hardware, Software and
Human Factors: John Wiley & Sons.

VITEZ, I., ORUČ, M., KRUMES, D., KLADARIČ, E. (2007). Damage to railway rails caused by
exploitation. METALURGIJA 46 (2007) 2, 123-128

VITEZ, I., KRUMES, D., VITEZ, B. (2005). UIC recommendations for the use of rail steel
grades. METALURGIJA 44 (2005) 2, 137-140

WEMAN, K. (2005). Welding Processes Handbook. ESAB Research, Sweden. Woodhead

Publishing Limited.

14