The Arup Journal 3/2005 37
Background
On 17 October 2000, a high-speed Intercity train from Leeds to London on the
East Coast Main Line derailed at Hatfield, Hertfordshire (Fig 1), resulting in four
deaths and over 30 injuries. It soon became clear that a break in the high rail on a
1500m curve was to blame.
Investigations concluded that there were numerous cracks on the running
surface, and when one of these penetrated to the base, multiple fractures of the
railhead led to disintegration over some 30m (see Fig 2 for nomenclature).
The potential for gauge corner cracking (GCC), as the phenomenon was initially
called, to lead to railhead break-up made the then UK rail infrastructure operator,
Railtrack, begin a rigorous system-wide inspection programme. Few in the UK had
ever heard of GCC and no-one knew why it suddenly occurred. The search began
for the cause.
Why rails crack
Gauge corner cracking on the
British network: Investigation
Robert Care Steve Clark
Mark Dembosky Andy Doherty
Gauge corner cracking caused the fatal Hatfield rail
crash. Few had ever heard of the phenomenon, and
no-one knew why it was suddenly occurring. Arup, with
TTCI, was asked to investigate the cause.
Arup’s role: November 2000 to date
Phase 1
The large number of locations damaged by GCC
led to many temporary speed restrictions, and a
major effort to replace affected rails, switches, and
crossings (S&C). All this caused train delays and
cancellations, and early work was directed towards
treating the symptoms rather than addressing the
causes. The cracking at Hatfield was caused by
rolling contact fatigue (RCF), something familiar in
branches of engineering focused on bodies in rolling
contact, such as bearings. Such bodies can
damage each other in several ways, depending on
the severity of the contact pressure and the shear
or tearing forces in the contact area, which for most
trains in Britain is about 15mm in diameter. Damage
can be in the form of surface cracks, or the wearing
away or plastic flow of the materials themselves.
The RCF was in turn blamed on a wide range of
issues, ranging from recent alterations to the
structure of the British rail industry, to changes in
the materials used on the railway, to the overall
condition of the track infrastructure. To help
Railtrack gain control of the situation, Arup was
appointed jointly with the Transportation Technology
Center Inc (TTCI) to identify and investigate all
possible causes of the problem.
This was a major task, given the lack of robust
information, the contractual arrangements in the
industry, and the poor relations between the parties
involved in the privatized and fragmented British rail
industry. The root cause could be rail metallurgy,
maintenance and renewal techniques, working
practices, vehicle design/maintenance, wheel
condition, some or all of these in combination, or
any one of several others.
Arup/TTCI organized an intensive Phase 1
campaign to identify all the potential causal factors,
quantify the variables associated with them, and
assess the likelihood of each as a major contributor.
This was done in November/December 2000,
and an interim report was issued before that
Christmas. Many measurements, inspections, and
analyses were conducted, and several possible
causes were rejected. No single ‘smoking gun’
could be identified.
Phase 2
Starting in January 2001, Phase 2 quantified the
relative contribution to the initiation and growth
of RCF of the factors identified in Phase 1, with
the ultimate goal of designing and implementing
a control strategy to minimize and manage
its incidence.
1. The Hatfield crash.
10100_Arup 29/11/05 2:35 pm Page 37
The Arup Journal 3/200538
The Arup/TTCI team co-ordinated a programme of measurements from track and
vehicles across the industry: wheel and rail profiles, track forces, wheel impact
loads, and wheel truing machine tolerances. In all, over 4000 wheel profile
measurements and 2000 rail profile measurements were taken. Scores of RCF sites
were investigated, and the computer modelling analyses of vehicles on specific
types and designs of track were numbered in the thousands. The team also
initiated a programme of laboratory research to understand the contribution of rail
metallurgy and vehicle loadings to the growth rate of the cracks, including both
small- and large-scale test regimes. The work performed in 10 months by the
Arup/TTCI team (with the co-operation and support of the British railway industry)
covered what would normally be a three to five-year major research effort.
Through these studies Arup/TTCI led the development of a control strategy
for use by professional engineers and railway group managers responsible for
decision-making and directing railway asset maintenance, renewal, and operations.
The control strategy recommends the actions that these groups need to take
across the wheel-rail interface to discharge their accountabilities and responsibilities
for effectively managing RCF.
Phase 3
Phase 2 ended in October 2001, but Arup/TTCI continued to provide wheel-rail
consulting services to Railtrack as required. Phase 3 began in January 2002 and
continues to the present for Railtrack’s successor, Network Rail. Its highlights
have included:
• developing a method to include the effects of RCF and remedial rail grinding on
Network Rail’s long-term maintenance and renewal projections
• initiating RCF monitoring programmes on the Great Western Region and the
London, Tilbury, and Southend network run by the train operating company c2c
• a major programme to develop remediation methods for poor rail adhesion
caused by autumn leaf fall
• dynamic modelling to investigate the tendency for the new Desiro trains being
introduced on the Southern Region to cause RCF, and monitoring the effects
of their introduction
• a preliminary investigation into the feasibility of using acoustic methods to
detect defective bearings on high speed trains.
Phase 4
By January 2003 Railtrack had been replaced by Network Rail, partly as a result of
the financial crisis that RCF had inflicted on the industry. Network Rail realized that
the British railway industry was addressing its wheel-rail interface issues rather
haphazardly, and asked Arup/TTCI to assist with industry-wide wheel-rail
technology management. This Phase 4 ran formally until March 2004, and some
parts continue today.
The first task was to identify and review current and planned work packages
throughout the industry and for each to determine the sponsor, supplier, remit, links
with other packages, planned deliverables, whether a final report was produced,
and the extent to which the deliverables had been achieved. Additionally, each work
package was related to the control strategy produced in Phase 2. From this, gaps
and overlaps in the programmed work packages were identified and addressed.
Further, the programme was to provide encouragement and assistance to turn
research effort into tangible results.
Other accomplishments included developing a Southern Region defence plan to
mitigate any effects of new vehicle introduction, establishing a ‘Golden Mile’ section
of track where rail best practices are strictly observed, investigating track-friendly
bogies, and producing a Passenger Rail Best Practices Handbook (still in progress).
Arup/TTCI was instrumental in designing a new organizational structure to
provide the technical backup for the VTI SC (Vehicle Train Interaction Systems
Committee), an industry-wide body established initially under the Strategic Rail
Direction of translation
Direction of
rotation
Flakes/
cracks
Microstructure
laminates
Longitudinal shear force
Normal force
Contact patch
Ratchetting
strain deformation
Rail
material
Crown area of the rail
Gauge corner Field side of
the rail
Rail headGauge face
Rail web
Rail base
INSIDE OF TRACK OUTSIDE OF TRACK
5. Contact patch pressure and forces causing surface elastic
deformation of the rail microstructure, leading to RCF.
4. Head checking on a 1300m curve.
3. GCC on a 900m curve high rail.
2. Typical rail section nomenclature.
10100_Arup 29/11/05 2:35 pm Page 38
The Arup Journal 3/2005 39
Authority to continue the co-ordination of anti-RCF research. VTI SC replaced the
committee WRISA (Wheel Rail Interface System Authority) proposed in Arup/TTCI’s
report in December 2000.
In early 2003, Network Rail and Arup/TTCI proposed a hypothesis summarizing
current understanding of the causes of RCF, and then co-ordinated its presentation
to the industry and the establishment of an experimental strategy to test it.
Finally at the instigation of Arup/TTCI and First Group (owner of several train
operating companies), an opportunity was taken to input anti-RCF approaches
into the design of a new fleet of trains for the trans-Pennine routes. This resulted in
significant changes to that train design, incorporating the latest thinking on
RCF reduction.
RCF: a natural consequence of bodies in rolling contact
Though the railhead cracks were initially ascribed to GCC (Fig 3), most of it should
more appropriately be described as ‘head checking’ (Fig 4) in that the cracks
developed nearer the rail centreline, usually 15-25mm from the gauge face. GCC is
found on the gauge corner itself. GCC and head checking are both subsets of the
more general phenomenon of RCF.
To illustrate the basic process (Fig 5) the rail is shown as comprising adjacent
plates or laminates. Since the pressure and forces are high enough to cause plastic
distortion of the laminates near the surface, an accumulation of metal dislocation
occurs with each wheel pass until cracks occur at the surface. (RCF inspectors
often run their fingers across the railhead and can feel the crack edges.)
At first the cracks are quite short and at a shallow angle, but some may elongate
and turn into a steep angle, following a path similar to the laminate boundaries
(Fig 5). This tends to occur when the surface length approaches 30mm and makes
a rail fracture far more probable.
A useful way to depict the potential for the pressures and forces required to
produce RCF is the ‘shakedown diagram’1 developed by Bowers and Johnson of
Cambridge University (Fig 6). The actual location of the ‘crack initiation’ boundary is
a function of the metallurgical properties and the friction at the contact patch.
Fig 6 also shows a region of increasing wear that is a function of the force.
This implies that when the rate of wear exceeds the rate of crack growth, cracking
doesn’t propagate before it is worn away by the passage of wheels.
The shakedown diagram shows that:
• RCF can be a natural consequence of rolling contact.
• RCF occurrence depends on the contact patch pressure and forces.
• RCF initiation and growth are influenced by the metallurgical and frictional
properties at the contact patch.
Research
Initially it was thought that RCF in the form of head checking and GCC was caused
by quasi steady-state wheel-rail contact forces, but not any longer. The intense
investigations from late 2000 through 2003 showed there to be at least three
separate modes of RCF initiation and growth on the British system:
• Mode 0: steady state, generally occurring on tight curves
• Mode 1: bi-stable contact, generally occurring on medium curves
• Mode 2: convergent motion, generally occurring on shallow curves,
straight track, and S&C.
These RCF modes are likely to be similar on other railways. In all cases, the
behaviour of the wheel-rail interface clearly can lead to contact pressures and
tangential shear stresses that exceed the shakedown limit of the metal at or just
below the surface of the railhead.
The following observations were made in the field and laboratory:
• Almost all RCF in the British system occurs on curves and usually on the high
rail (Fig 7). Fig 8 shows an early estimate of the probability of RCF on the British
system as a function of curve radius. This shape is largely true of most British
Limit moves right
as friction drops
Limit moves
up with
hardness Increasing
wear
Po/K
Tc = Ft/Fn
Squats
No crack initiation Crack initiation
Po: Contact patch peak Hertzian pressure
K: Yield stress in shear
Ft: Contact patch normal force
Fn: Contact patch shear force
Cant
Outer rail
(high rail)
Inner rail
(low rail)
0.35
Probability of RCF
1500m Radius of curvature
6. A shakedown diagram shows graphically the relationship
between contact conditions and the properties of the
rail material. Since wear is generally a non-linear function
of Tc, cracks eventually give way to wear. (Squats are
shallow internal rail defects caused by railhead head surface
abnormalities of wear and stress.)
8. Approximate distribution of RCF on the British system as a
function of curve radius.
7. Wheel contact on a curve with the high rail and low rail.
10100_Arup 29/11/05 2:35 pm Page 39
The Arup Journal 3/200540
main line routes, and very similar to distributions
in the Netherlands.
• On the British system, the highest probability of
RCF is on c1500m radius curves (Fig 8).
• Most RCF on straight track is associated with
S&C. RCF in switches is often found in the same
relative position and is thought to be associated
with events or ‘triggers’ that tend to change the
rolling radius difference (eg sudden changes in
alignment or gauge or the switch/stock rail
transition).
• In shallow curves, where wheel/rail contact is on
the gauge shoulder of the rail, the RCF is
observed as head checks. In tight curves, where
wheel/rail contact is near the gauge corner, the
RCF appears as GCC.
• Field observations typically show bogies with stiff
suspensions contacting near the gauge corner
and softer suspensions contacting higher on
the railhead.
• Laboratory studies have shown that the angle of
the observed surface cracks matches well the
ratio of the lateral and longitudinal forces (Fig 9).
The most compelling evidence comes from
WRISA-sponsored research2 that compared very
detailed field observations with estimates of contact
patch location, pressure, and forces. To date, six
RCF sites have been studied in detail and a very
high correlation has been found between RCF
predicted by shakedown diagrams and observation.
These studies were time-consuming and required
very detailed RCF maps, both wheel and rail
profiles, and specially processed track geometry
data. All this information plus a mathematical model
of the vehicles traversing each site was combined in
a vehicle dynamics simulator. Estimates of contact
pressure and forces were then projected into
shakedown diagrams (Fig 10). In parallel, a product
called the Whole Life Rail Model (WLRM) produced
an alternative predictor of RCF using the same outputs from vehicle simulators,
but not shakedown theory. The WLRM predictor asserts that not only force is
necessary to create cracks but also energy expenditure. The algorithm combines
the longitudinal and lateral creep with the corresponding creep forces to predict
both RCF and wear potential (Fig 11).
Estimates from both the shakedown limit and WLRM predictor have been
consistently good, particularly when inadequacies in the process are accounted for:
• Only a limited number of wheel profiles are used in the model.
• Past history of track geometry is not well known.
• Vehicle models assume nominal parameters.
• Vehicle models use significant simplifications in modelling the contact patch
and its location.
• The friction history of the sites is not well known.
In spite of these inadequacies, the studies to date have provided a credible
explanation for the mechanism that produces RCF, reasonable tools to predict it,
and valuable insight into the primary vehicle and track factors that produce
conditions likely to initiate RCF. The same studies have shown that many
combinations of vehicles and track, each well within its own current safety
standards, could frequently produce RCF, replicating field observations.
2500
1500
1000
500
0
C
O
N
TA
C
T
S
TR
E
S
S
M
N
m
2
00 0.1 0.2 0.3 0.4
TANGENTIAL FORCE COEFFICIENT
Minimum track alignment, measured rail profiles
50% track alignment, measured rail profiles
Perfect track alignment measured rail profiles
National RCF limit
00 100 200 300m
DISTANCE ALONG TEST SITE
FIELD
SIDE
GAUGE
SIDE
9. Surface crack angle as a function of contact patch
longitudinal and lateral forces.
10. Shakedown diagram showing the effect of track alignment on a Class 43 locomotive as it
traverses an S&C unit. Points produced by modelling above the ‘notional’ limit correspond to
RCF observed in the field.
11. WLRM RCF index for curve high rail; the black contours are RCF locations on
railhead from field observations.
10100_Arup 29/11/05 2:35 pm Page 40
The Arup Journal 3/2005 41
Modes of RCF generation
Of the three modes of RCF initiation and growth, steady state mode (Mode 0) is
usually thought of as RCF and best describes GCC and head checking in curves.
The wheel-rail contact forces that cause RCF by this mode are generated by
vehicle curving behaviour in a specific range of curve radii and operating conditions.
This is thought to occur in tighter curves with radii of 1200m and less.
Bi-stable contact mode (Mode 1) describes RCF occurring when the wheel-rail
interface operates in a region of instability, where small changes in wheelset lateral
shift generate large changes in rolling radius difference. This is thought to occur
when the wheel profile wears so that there is conformality* in the flange root area in
the wheel, and the wheel-rail contact position occurs there - generally at the gauge
corner/gauge shoulder area and under conditions of high conicity**. This is thought
to occur in curves with radii of 1200m - 2000m; however, it appears to overlap into
curves of tighter and shallower radius.
Convergent motion mode (Mode 2) describes RCF occurring when changes in
track lateral position and wheelset lateral shift cause the wheel flange to converge
on the rail gauge face, even though flange and gauge face may not necessarily
come into contact. This type of behaviour is thought to occur in moderate curves
with radii of 2000m and higher, and in straight track. Again, there is assumed to be
an overlap of this behaviour into curves of tighter radius where Mode 1 would
normally be expected.
These locations of RCF cracks on the railhead fit with the dynamic wheel-rail
interface behaviour expected in the three modes (Fig 12). The quasi-steady state
behaviour of vehicles in these ranges of curvature with respect to shakedown are
shown as well.
In all cases, RCF in the British system is due to excess wheel/rail forces primarily
caused by the axle shifting relative to the rail too far to one side or the other. This is
true on curved track, straight track, or S&C. In tight curves, the mechanism tends
to be steady state, while in S&C and moderate curves and straight track the
mechanism is transient. The second part of this paper, to be published in the next
edition of The Arup Journal, explains how this happens in each mode.
References
(1) JOHNSON, K. Contact mechanics. Cambridge University
Press, 1985.
(2) DEMBOSKY, M, and BAKER, P. Great Western Zone RCF
pilot study modelling report. WRISA, 2003.
* Close match between the shape of the wheel and the shape of the railhead, so that very small
lateral shifts between the wheel and the rail can result in significant shifts of location of the contact
patch, from the head to the shoulder of the rail.
** Rail wheels are shaped like a cone to facilitate cornering, although the taper is not uniform. As the
wheel tread moves towards the flange the taper flattens out, increasing the ‘conicity’.
Acknowledgements
The authors wish to acknowledge the contribution of many
people in the British railway industry to the research that
underlies this paper.
Gauge Field
Low Mid High
0.35
1500mRailhead
Fc
Tc
Probability of RCF
Ideali