Rolling Contact Fatigue of Rails: what remains to be done?

R A Smith
Imperial College of Science, Technology and Medicine, London

Introduction

On 17 October 2000, a high-speed train, travelling at 185 km/hour, en-route from

London King’s Cross to Leeds, derailed when taking a 1460m radius curve, some
17km north of King’s Cross. The rear eight vehicles of the eleven vehicle train
were derailed, two coaches were turned almost on to their sides, whilst a third
coach, a buffet car was very badly damaged. Of the 182 people on the train, four
were killed and 70 suffered injuries, four of them seriously [1].

It rapidly became apparent that the derailment was caused by a fractured rail on
the outer line of the curve. Of particular concern was that beyond the first
fracture, the next 35 metres of rail had broken into 300 pieces, and some 44
metres further on, another length of about 54 metres was similarly fragmented. It
was clear that the original and subsequent fractures had largely been triggered
from fatigue cracks existing in the rail: although the term “gauge corner cracking”
was used as the first description, “head checking” and the more generic “rolling
contact fatigue” (RCF) were used later.

These events caused detailed inspections to be made of tracks throughout

Britain in the days that followed. Many sites were located where cracks were
visible on the surface of rails. Speed restrictions, some as low as 8 km/hour,
were quickly introduced. Many trains were cancelled; schedules were revised to
double or treble normal journey times and even then were unreliable. Motorways
became choked as people switched their journey to roads and internal domestic
flights became overbooked. When a journalist wrote, “no other railway accident
in British history, or, I would guess, any other country’s history – has led to the
degree of public anger, managerial panic, political confusion, blame and counter-
blame that came in the wake of the Hatfield crash. In fact, outside wars and
nuclear accidents, it is hard to think of any technological failure which has had
such lasting and widespread effects” [2], he was not exaggerating.

It is not the purpose of this paper to enter into the debate about the extent to
which the privatisation of Britain’s railways, effective from Spring 1997, which
replaced a single entity vertically integrated railway system, with a fragmented
arrangement of over 100 major payers, played a part in these events [3]. It is
sufficient to state here that the custodianship of the railway infrastructure was
placed in the hands of Railtrack, a company who wished to be known by there
own publicity as ‘the heart of the railway’. Instead, we will concentrate on only
the technical crisis surrounding the rolling contact fatigue of rails.
 2

The fatigue problem generally

It is worth mentioning some of the key points which have arisen from over 150
years of research into the fatigue of metals.

Fatigue began to be recognised as a specific failure mode when the early

railways began to suffer failures of axles, wheels, rails, boilers and other
components. Much impetus for investigations into fatigue stemmed from the first
railway accident to cause a major death toll, which occurred near Versailles in
1842 [4] when the axle of a locomotive broke. During the next two decades, the
great German railway engineer Wöhler, demonstrated that cyclic stress ranges
determined fatigue lives and that for steels at least, a fatigue limiting stress
existed, below which fatigue lives were infinite. The so called S/N curve, relating
stress range to cycles to failure and the fatigue limit, still remain the basis of
design against classical fatigue.

The mechanism of fatigue has been unravelled during the 20th century,
particularly in the last fifty years. It is now known that fatigue is caused by the
initiation and growth of cracks. The quantification of crack growth has become
possible through the use of fracture mechanics, although the quantification of the
initiation stage remains rather tentative. It is well established that fatigue initiation
usually occurs at a free surface, aided by some kind of stress concentration
feature. Circumstances can arise which produce non-propagating cracks. An
example is that of a crack initiated at a particularly severe stress concentrating
feature, which then stops as it grows out of the zone of high local stresses into a
bulk stress field which is insufficient to carry it forward. It has become apparent
that similar circumstances are important in RCF.

Fatigue is often only one of several simultaneous deterioration mechanisms, for

example, corrosion, creep or wear. The conjoint action of these mechanisms, is
frequently more complicated than a linear superposition of modes. There are
many standard tests on fatigue of which Suresh [5] is an excellent modern
example.

In a practical sense, application of our knowledge of fatigue to real situations has

been hampered in two ways. First, and perhaps at first sight, surprisingly, on
many occasions we do not have sufficient knowledge of the actual service loads
to which a particular component is subjected, nor do we know the critical location
in an often complex shape where the most severe conditions for crack initiation
exist. This means that similitude between data generated in the laboratory and
service conditions is frequently difficult to achieve. If this is coupled with the
natural statistical scatter of the fatigue process, it can mean that fatigue life
predictions are frequently in error by more than an order of magnitude.

The second major problem concerns the management of fatigue, that is the
continuation in service of parts known to contain cracks and the calculation of
 3

residual safe lives. This process necessarily involves the detection and sizing of
cracks in components. Although several methods for this exist, they are not
always easy to apply and the results they produce can be ambiguous. In
particular, the ability of non-destructive testing techniques to penetrate below the
surface to measure the depth of progression of cracks into the bulk of the
material is still limited and the whole area of non-destructive examination is
considered by many to be an art rather than a science.

Rolling contact fatigue of rails

The stress concentration feature which causes initiation of RCF cracks is the
contact between the wheel and the rail. Conditions under the contact patch are
always severe and the yield stress of the rail steel is always exceeded, on at
least a microscale, due to the surface roughness of the wheel and the rail. It
follows that irreversible events take place at every passage of every wheel. The
term ‘permanent way’ is a misnomer, because it is changing continuously. The
irreversibilities of each wheel passage, result in both a wear and a fatigue
process and the resultant life of the rail is a competition between these two
failure processes.

The stresses generated under the contact are complex and governed by the
detail of the wheel/rail geometry near the contact patch, the position of which is
determined by, inter alia, curving behaviour, vehicle suspension characteristics,
and, of course, existing conditions of the wheel and rail. Both traction forces and
radial curving forces increase shear stresses in the contact zone and hence the
propensity to initiate cracks.

The phenomenology and sequence of development of RCF cracks in rails is

shown schematically in Fig 1. Many observations have confirmed that cracks
develop towards the direction of motion, initially inclined at a shallow angle of
about 15 o to the head of the rail. When the cracks reach a depth of typically
10mm, the angle steepens to about 70o and the cracks then propagate through
the rail until failure. During the shallow angle growth, flakes of material may
detach themselves from the head of the rail, but the rail danger of a broken rail is
obviously a result of the turned-down crack, and, therefore, this phase must be
avoided if possible.
The stages of the development of the cracks can be decomposed into separate
phases; each controlled by separate elements of the various overall stress fields:

Immediately under the surface, material is sheared in the direction of motion,

ductility is exhausted by cyclic ratchetting and inclined cracks are formed
along the shear planes. These cracks initiate rapidly, but the growth rate also
falls away rapidly with distance below the surface.

The cracks then react to the Hertzian stress field due to the contact. There is
evidence that the initial growth rate (with depth), might be enhanced by the
 4

presence of liquid (rain, moisture or lubricant), by either a hydro-wedging

action (see Appendix 1), or by the moisture acting to lubricate the rough crack
surfaces, thus facilitating crack sliding. Then, as the contact stress field
decays with distance from the contact, the growth rate falls.

Simultaneously, the crack is reacting to the bulk stresses in the rail, which are
the resultant sum of the bending stresses, residual stresses from manufacture
and the continuously welded rail stresses. The former controls the cyclic
fatigue stress range, the latter two, the mean stress about which the cyclic
stresses operate. The effect of the bulk stress increases with depth below the
surface, being negligible at small depths, but leads to run-away growth for
larger cracks.

Thus the overall response, leads to the W shaped curve shown in the final
sketch of Fig 1. The exact form of this curve will, of course, vary as each of
the consistent elements varies. The important conclusion is that some
‘handshaking’ must occur, if one mechanism of growth is to be succeeded by
the next, and the possibility of non-propagating cracks exists if the minima of
the growth rates fall below appropriate thresholds.

We now turn to consider the interaction of the fatigue process with wear.

Although empirical field data exists for the wear rate of rails, the data exhibits
much scatter and the various parameters contributing to the overall effects are
difficult to separate. Nevertheless, it is clear that high wear rates will ‘scrub out’
fatigue cracks faster than they can form, but the motivating factor for using
harder rails is to reduce wear rates. The danger exists therefore that harder rails
may be more susceptible to fatigue cracking.

Fig 2 is a series of schematics illustrating the interactions between what is

assumed to be a constant wear rate at the head of the rail and the sequence of
fatigue crack development previously described. Even if initial crack growth rates
at the surface allow initiation, because the crack growth rate decreases sharply,
a wear rate lower than the initial fatigue crack rate, can lead to the situation that
material is being worn away from the surface end of the crack faster than the
crack tip is advancing, thus resulting in a net shortening of the crack. Such
cracks would not therefore develop into the contact or bulk stress fields. The
second sketch, illustrates the situation in which the ‘handshaking’ growth rate
between the initiation and contact stress phases, is just higher than the wear rate
(point A), but as the crack leaves the contact zone (point B), the wear rate is
faster than the growth rate, and the crack will cease to lengthen in the manner
described above. The final sketch shows a growth rate which is too low to
influence either ‘handshaking’ event.

The discussion above has assumed that the wear rate is the natural rate at the
wheel/rail control. It is well established however than an artificial wear rate can
be established by grinding the railhead. Although grinding was initially conducted
 5

to re-profile the railhead, or to improve the longitudinal profile of the rail (eg to
remove corrugations), it has been used in many countries to control or at least
mitigate the effects of RCF.

A rational maintenance strategy needs considerable quantitative flesh to be

placed on these schematic models. After the Hatfield accident in the UK, a large
team was assembled under the project management of Ove Arup, with major
technical contributions from the Transportation Technology Center Inc. (TTCI) of
Pueblo and AEATechnology. A major part of this program has included
collecting data from cracked rail samples together with the operational
parameters which have led to their conditions. This huge observation database
has proved valuable in both generating modelling ideas and in generating
quantitative mechanistic models. A ‘whole-life fatigue model’ is being developed,
based upon many of the ideas outlined here which were initially proposed by
Kapoor [6]. This model will eventually incorporate the key practical operational
variables of rail, wheel, vehicle and track and will treat the variations of these in
an appropriate statistical manner.

Scientific research needed to understand rail failure

Although considerable information of RCF in rail exists as a result of many

studies conducted in various countries over the last two decades, some key gaps
in our knowledge still exist. The following appear to be the main background
scientific gaps, whilst the final sections reviews practical applied aspects.
 6

Crack initiation and wear

The thin surface layer immediately under the wheel/rail contact is subjected to
severe loading conditions. The initial stages of wear and fatigue are essentially
identical. The material accumulates damage and fails locally. The rate at which
material detaches as wear debris, must be greater than the deepening of cracks
if fatigue is to be avoided. Currently the quantification of wear is order-of-
magnitude, as is our ability to quantify initial crack development. Research is
needed on parameters such as initial roughness of both wheel and rail, the role
of hydrostatic compression on extending ductility and on the detailed mechanics
of the stresses arising from combinations of new and worn wheel and rail
profiles, and the effect of longitudinal and tangential stresses.

Crack propagation by contact stresses

As cracks deepen and grow away from the severe surface region, the sub-
surface contact stresses control their development. More detailed 3-dimensional
models need to be developed and combined with fatigue crack growth properties
of materials subjected to equivalent complex stresses. Some particular features,
which need further understanding, include:

the effect of the surface roughness of the developing cracks. It is reasonable

to assume that over part of the loading cycle, the rough crack faces interlock
and reduce sliding motion along the direction of the crack. Continued cycling
may smooth the crack surfaces, wear debris may become detached and act
as rollers to facilitate crack movement. This problem is connected with the
theories of enhanced crack growth due to fluid entrapment. Despite much
previous research, it is not clear if fluid pressurises the crack, thus enhancing
growth rates.

the residual stresses in the rail are vital in determining around what mean
stress the cyclic fatigue stress operates.

The residual stresses locked into the rail head during manufacture need to be
measured. Investigations on the modification of these stresses by the severe
surface stresses need to be made. Thus detailed examinations need to be
made of rails of various grades at various stages of their service lives.

Crack propagation controlled by bending stresses

At a later stage of RCF crack growth a transition is made from control by contact
stresses to control by bending stresses. The exact criterion governing the
changeover from cracks propagating at a shallow angle to cracks turning down
and propagating into the body of the rail is as yet unclear. Furthermore, what
causes some cracks to turn upwards leading to the detachment of large flakes of
material needs elucidation. Models need be developed to quantify the turned-
 7

down growth phase. These models need to understand, inter alia, the effects of
manufacturing residual stresses, continuously welded rail (CWR) stress and its
variations with temperature, rail section shape and size, sleeper spacing and
dynamic loading due to wheel flats, out-of-round wheels and surface irregularities
on the rail head. Any steps that can be taken to reduce the bulk bending
stresses and lower the mean stress around which they operate will reduce the
possibilities of turned-down cracks propagating to failure.

Similitude between laboratory and field conditions

In common with all areas of fatigue, great care needs to be taken in establishing
similitude between conditions applied to sample specimens in the laboratory and
what is actually happening in real rails in service. A major problem here is the
appropriateness of the size scale to which non-linear plastic deformation effects
are applied to a small crack near the head of the rail. The crack is effectively
surrounded by a sea of plasticity: a condition which is often not matched in the
laboratory. Thus material properties such as crack growth rates and thresholds
may be affected because of these effects.

Furthermore, laboratory tests often fail to match the dynamic stresses to which
real rails are subjected. These dynamic effects are, of course, amplified by speed
and the magnitude of enhancement for various kinds of irregularities needs clear
elucidation.

Practical aspects and maintenance strategies

Despite a huge volume of work over the last few decades, clearly catalogued
field observations of the extent, development and rate of fatigue development
and/or wear are still sparse. It is hoped that the huge volume of data collected
after the Hatfield accident will contribute much useful information. However,
there still exists the need for the collection of much more information from the
field for many types of operating conditions. It is important that such information
is disseminated as widely as possible throughout the world’s railway industry.

It is clear that a competition exists between wear and the development of rolling
fatigue cracks in rails. Historical trends towards decreasing wear rates coincide
with increasing rail break problems. Only by better quantification of both wear
and fatigue, can rational choices be made of lowest economic wear rates which
will eliminate rail failures. In situations where crack initiation is unavoidable, the
progress of cracking can be checked by appropriate rail grinding: but again,
rational and economic strategy needs quantitative understanding. Lubrication,
applied to reduce flange wear, needs to be properly understood, otherwise
fatigue problems may increase. The balance here is between reducing wear and
reducing the tangential stresses which enhance crack initiation, and the possible
enhancement of crack growth away from the surface by the presence of the
lubricant.
 8

The inspection of rails for cracking is expensive, but necessary. Automated

techniques, coupled with locational identification have been developed and are
being improved. But surface measurement alone may not give sufficient
information. Accurate sizing of the extent of cracks below the surface continues
to pose problems. In particular, any methods which reliably and unambiguously
identify cracks, which have started to turn down, would be extremely useful.
Although this happens at the later stages of the fatigue life of the rail, knowledge
and confidence of the likely life remaining after this event would help to avoid the
necessity for imposing the kind of speed restrictions which were deemed to be
vital after the Hatfield accident.

Acknowledgements

The views expressed are the author’s own, but he acknowledges many useful
discussions with colleagues from Ove Arup, TTCI, AEA Technology and, in
particular, with Dr Ajay Kapoor of the University of Sheffield.

References

[1] Train Derailment at Hatfield, 17 October 2000, second HSE interim report,
23 January 2001, www.hse.gov.uk/railway/hatfield.interim2.htm

[2] The Crash That Stopped Britain. Ian Jack, Granta Books, London, 2001.

[3] All Change: British Railways Privatisation. Freeman, R. and Shaw, J. (eds.),
McGraw-Hill, London, 2000.

[4] The Versailles Railway Accident of 1842 and the Beginnings of the Metal
Fatigue Problem, Smith, R.A. In Proceedings Fatigue 90, The Fourth
International Conference on Fatigue and Fatigue Thresholds, Hawaii, July
15-20 1990, Eds. H. Kitagawa and T. Tanaka, Materials and Component
Publications Ltd., Birmingham, Vol. 4, 1990, pp2033-2041.

[5] Fatigue of Materials, S. Suresh, Cambridge University Press, 2nd Ed., 1998.

[6] Wear Fatigue Interactions and Maintenance Strategies, Kapoor, A. In, Why
Failures Occur in the Wheel Rail Systems, Proceedings of an Advanced
Railway Research Centre Seminar, Derby, 22 May 2001.
 9

Direction of motion Severely

deformed
surface layer

Whilst under the influence

of the contact stress field, Region
The crack turns Bulk stresses in
the crack propagates at a plastically Elastic contact
down to an angle the rail: due to
shallow angle of ~ 15 o strained by stress field
of ~ 70 o when bending,
controlled by the contact residuals from
stresses
bulk stresses manufacture,
CWR stresses
Growth rate

Initiation. Caused by severe plastic Growth rate

deformation and ratchetting Early growth in CS
field, can be enhanced
by hydro effect
Later decay in CS
field due to falling
stresses

Crack length Crack length

Finally, crack is driven

by the bulk stresses,
influenced by residual In this manner, the complex
Growth rate

and CWR stresses. Rate crack growth rate response

Growth rate

increases and results in ( shown by the dashed line)

FRACTURE is formed by the addition of
the separate mechanisms

Crack length Crack length

Figure 1. Features of the development and growth of RCF cracks in rails.

 10

Very high wear rates at the surface do As the wear rate drops even lower, it is
Growth rate / wear rate
not permit crack formation, slightly insufficient to prevent the cracks
entering the elastic CS field (A).
lower rates “rub out” the initiating

Growth rate /wear rate

cracks faster than they form However, it may stop propagating
when it leaves the CS zone, before
being carried by the bulk stresses (B).

B
A

Crack length Crack length

At lower growth rates than the

Growth rate /wear rate

one shown, the bulk stresses

propagate the crack down
through the railhead to eventual
failure

Crack length

Figure 2. Interactions between wear and RCF cracks in rails.

 11

Appendix

As long ago as 1935, Way (ASME Journal of Applied Mechanics, 2, pp A49-

A58), suggested that oil was trapped within cracks during rolling contact of
bearings. The oil was supposed to pressurise the crack, thus causing the tensile
stresses needed for crack extension leading to pitting. There is strong
experimental evidence in the rail contact problem that a similar effect may occur,
because cracks in moistened disc-type experiments seem to develop much
faster than those under dry conditions.

The following order-of-magnitude calculation suggests that any fluid which was
pushed down into a crack by the approach of a wheel would not have sufficient
time to move out of the crack during the time taken for the wheel to pass over the
crack mouth. The exact nature of the assistance to the crack growth process,
either as a pressuring agent, or as a lubricant separating the rough surfaces of
the crack remains elusive.

Rolling direction
Typical velocity, V 50 to
150 km/h, or 15 to 40 m/s.

Kinematic viscosity, = viscous force / inertia force = / (m2/s)

where is the dynamical viscosity and is the density.
Characteristic velocity of the liquid, v /a
If V v, then the fluid will remain in the crack during the passage of
the rolling load.
Now for water is 1 mm2/s, so for crack lengths up to a few mm, v
1 mm/s and is much less than V, so that the fluid remains
“trapped”.