Sie sind auf Seite 1von 11

RTI/02/03/2000

Unrestricted

ULTRASONIC TESTING OF RAILS


A NEW LEVEL OF PERFORMANCE

David Griffiths

March 2000

INTRODUCTION
Ultrasonic testing of rails has been used for seventy
years to reduce the number of in-track failures. Its
initial success in eliminating Hydrogen flaking and
subsequently driving steelmaking towards cleaner
continuously cast steels is well documented.
The rejection levels have however remained static over
the last 20 years. This is not because steel makers have
not improved in this time. Universal use of
continuously cast steels for main line use is apparent.
The main reason has been the inability to clearly
demonstrate a cause and effect relationship between
track failure and internal irregularities below about
1.5mm (1/8). Fortunately the probability of a wellmanufactured rail having an internal defect anywhere
near this size is rare. Conversely a manufacturing
problem in which defects approaching this size become
common usually tempts the manufacturer to deliver the
steel because its within the specification. This has
resulted in dissatisfied customers and frustrated
suppliers. An example of a Horizontal Split Head
(HSH) caused by such a defect, which can be seen as a
line on the centre of the fracture, is shown in Figure 1.

Figure 1 HSH Defect


In track testing has fared a similar fate. In 1988, Roney
of Canadian Pacific reported to AREA Committee 2
(now AREMA) that the detectability of defects in track
was not 100% as would be desirable but something
less, sometimes much less. The chart reproduced in

Figure 2 was based on data collected by the


Transportation Systems Centre and the AAR.

Figure 2 Defect Detectability (after Roney)


Roney reported that this could be calculated to lead to
0.1 service failures per mile per test. Using US average
statistics for traffic this would then be equivalent to one
broken rail derailment per thousand track miles per
year.
For the iron ore railways in the Pilbara of Western
Australia which each have in the order of 500km of
track (300 miles) this would translate to one rail related
derailment per year between them. This level of
performance is not acceptable for this business and a
much improved level of performance has been obtained
principally by more frequent testing.
Reiff reported in March of this year that the detection
probability for a 30% head area defect stands at about
70%, essentially unchanged from the performance
reported by Roney in 1988.

Page 1 of 12

The limitations in capability for both steel plant and intrack testing have now been overcome.
This report describes a technological advance in testing
technology which allows testing to detect inclusions at
levels equivalent to the theoretical resolving power of
the ultrasonic signal. In fact, advanced signal
processing using frame-to-frame comparison has
broken through this barrier and inclusions smaller than
this theoretical limit can be detected. The equipment
has also been demonstrated capable of detecting the
previously difficult to impossible to detect defects in
track:
Inclusions
Vertical split Head
Defects below masking defects.
The equipment is described in the following pages and
examples of these difficult defects which have been
detected are presented.
TESTING EQUIPMENT
HARDWARE
RTIs Ultra-light High Production Rail Flaw Detection
Platform is known as 8000SX. The compact size can be
seen in Figure 3.

Figure 3 8000SX system mounted in a Toyota


Landcruiser

Each transducer input has its own DSP to process and


analyse signals in real time. This is combined with its
own unique structural processing architecture, purpose
designed specifically for Rail Flaw Detection made
possible through the design of purpose specific silicon
chips.
The hardware uses surface mount components and onboard High Speed DSP for every channel. The logic
incorporates high speed, high-density programmable
gate array architecture. Data Transfer to PC is DMA
(Direct Memory Access) or Programmed I/O
The system is of compact modular design for ease of
maintenance and servicing. It is small enough to fit in
a Daihatsu Terios 4WD for use on narrow gauge
tracks. For standard and broad gauge track testing it is
mounted in a standard Toyota Landcruiser with the
only modifications to the vehicle being the Hirail. It
can be returned to the workshop as airline acceptable
baggage should the need arise.
The 8000SX system electronics and control systems are
mounted in the back of the vehicle. It is very compact;
the system is rack mounted and houses 32 channels of
ultrasonics together with all necessary pneumatics,
compressor and water pumps.
The user is presented with super VGA high resolution
Graphics representing a B Scan of the left and right
rail.
The Rails and defects are scaled and trigonometrically
corrected. The operator can zoom in on a defect and
view 600mm-length rail for the width of desktop or
zoom out for a 20-metre look or any other length view
he chooses.
The defect amplitude is represented on the B Scan by
different level colours immediately making defects
visually obvious.
The operator also has an A Scan Virtual Cathode Ray
Oscilloscope displaying analogue signals windowed
with B Scan of the rails. This meets the requirements
of some operating specifications to have oscilloscope
signals available for the operator but in practice the
operator does not refer to this signal, instead relying on
the defect window as the key information source.

The 8000SX DSP based Ultrasonic Rail Flaw Analysis


system is a quantum leap in technology and is radically
different to any Flaw Detection System previously
available. The system development to date has been in
excess of 12 man-years.
The system uses the latest Digital Signal Processing
(DSP) and large-scale gate array technology. To
provide very strong signal processing power the typical
sample rate is 18 mega samples per second with
provision for up to 32 mega samples a second

Page 2 of 12

Figure 6 300 m inclusion reported by 2000SX and


8000SX systems
Figure 4 Operator Screen
The numeric listing of the defect can be displayed as an
A-Scan display (Depth vs. Amplitude) as captured by
the oscilloscope trace as the car passed over the defect.
The signature of the defect from the A-Scan can be
used to determine the type of defect but further
recognition is available in software. An example of the
display is shown in Figure 5.

Figure 5 Defect Displayed on Operator Screen


SIGNAL TO NOISE RATIO
The signal to noise ratio of the 8000SX is better than
24db improvement on the earlier ultra-low-noise,
2000SX system, which was considered to be, a leading
technology in 1992. That system successfully and
repeatably detected defects down to 250 microns, which
was considered a world first.
The improvements have been achieved at a number of
points in the process. Real Time, Active Digital Signal
Rectification has been applied to avoid the undesirable
wider return echoes, which are a feature of passive
rectification features at increased gain. This feature
also permits the system to test closer to the rail surface.

The 2000SX system had previously been demonstrated


as finding 250m inclusions.
Seeing and recording multiple reflectors within a slice
makes it possible to detect defects that were previously
shadowed by large reflections from the surface
condition or large adjacent reflectors. This is
demonstrated later in this paper.
SOFTWARE
The 8000SX Rail Flaw Analysis System uses
Windows 98 operating system. The software for all the
upper level is written in C++. The gate array software
is in high Level VHDL code the DSP software is
written in machine language.
The software display mechanism is non-linear based
Global Reflection Map (GRM) display which makes it
fully compatible with the 2000SX and 8000SX DSP
Based Inertial Track Geometry Systems. The display
allows the operator to view any point tested, regardless
of vehicle location.
The software is multi-functional and is capable of
running the 8000SX DSP Based Ultrasonic Rail Flaw
Analysis System whilst simultaneously performing
background tasks such as printing or reporting.
Up to three windows of either previous Ultrasonic
Surveys or a mixture ultrasonic and geometry data can
be loaded at the same time and compared on the run
whilst still collecting data.

Page 3 of 12

For example decreased gain at under head radius,


increased gain in neutral axis area and different levels
of threshold makes for increased detectability through
increased sensitivity at targeted areas of the rail
without being impeded by the limitations of other parts
of the rail.
It is also possible to use individual zone noise reduction
algorithms, add flexibility in unwanted signal rejection
together with an overall increase in sensitivity.

Figure 7 Simultaneous Display of A scan, GRM and


Defect Data
Automatic defect sizing can be accurately done from
within the test vehicle using any of the normal sizing
techniques. This is because unlike other systems, which
only take the peak amplitude sample every frame, the
RTI 8000SX technology enables the sampling of 1024
samples per frame. This enables a very accurate picture
of the defect to be generated. This is described later.
Current requirements of specifications still require
hand sizing of defects but it is envisaged that as the
margin between detectability and critical defect size is
demonstrated and becomes accepted, the use of in
vehicle defect assessment will become the norm. This
will significantly improve testing productivity.
Defect recognition and classification is made possible
in RTI 8000SX technology by taking 1024 samples per
frame. RTI is then able to apply statistical calculation
on the data to determine the signature of the defect
from which the defect can be classified.
The on board Library facility captures and records all
defects for an ultrasonic survey. It holds defects or any
other in track reflectors of interest recorded during the
run survey. The same library file can be added to, from
survey to survey.
RTI has built an extensive library database of all the
different type of defects on different lines throughout
Australia.
Each RTI 8000SX Channel has a built in
programmable digital Time Gain Amplifier (TGA).
The gate array architecture has been programmed to
have Thirty-two time zones per slice as can be seen in
Figure 8. This means that each zone of the rail being
searched has its own programmable parameters such as
gain, threshold etc For all intents and purposes, each
zone is a totally programmable independent channel.
TGA allows compensation for Signal lost over distance
travelled. It also permits different levels of sensitivity
and different algorithms to be used in different zones.

Figure 8 Thirty Two, Individual Time Gain Controls


per Channel
Similarly the 32 divisions per channel adopt a Time
Variable Threshold (TVT) as shown in Figure 9. The
width of the thresholds in time is set by the zone width.
The TVT feature allows for precise discrimination and
sensitising of any part of the rail. A positive or a
negative threshold can be set for any zone to detect a
signal over threshold or below threshold.

Figure 9 Thirty-two, Time Variable Thresholds per


Channel
The effect of these radical system architecture changes
is that what is now called a single channel would, by
previous Rail Flaw Detection terminology, have been
classified as 1024 individual channels. In effect by
conventional Flaw Detection terminology the RTI
8000SX system has 32,000 Channels.
Automatic gain control, as applied to the 8000SX
system makes it possible for the first time to have
absolute references for each channel by monitoring
cumulative grain data at up to 1024 individual sample
points per frame. The benefit of having multiple

Page 4 of 12

absolute references is full linearity over the entire rail


height and transducer gain drift with temperature is
completely
eliminated.
Non-Linearity
between
transducers is also fully compensated.
The 8000SX architecture was designed keeping in
mind, that it in the future it will form the basis of the
technology that could be Loco-mounted, without an onboard operator. It is envisaged that such a system could
continually test the tracks as the loco travels on revenue
traffic, sending back defect data with severity and
location to a central PC via radio modem. It is likely
that such a system would first be adopted by a captive
type rail operation.
STEEL PLANT T ESTING
Manufacturing feedback is an important aspect of
continuous improvement. Systems generally indicate
whether a rail passes or fails whatever specification it
is set at. When the steelmaking plant is performing fine
then rejects are low or zero and there is no learning
obtained. When a small change in steelmaking causes a
rash of failures there is usually a sudden interest in the
feedback from the ultrasonic system. In many steel
plants that have significant stocks of blooms between
steelmaking and rolling, the feedback loop is weeks or
even months so that very large losses can be incurred
before a problem is identified and fixed.
Rails are currently tested by manufacturers against a
specification that is typically 1.5mm (1/16) for
inclusions, hydrogen flakes or other discontinuities.
Such defects are (when the process is under control)
extremely rare. Much smaller inclusions are considered
under the assessment of rail steel cleanness. These
inclusions are present in every rail oxides, silicates
and sulphides. These have been implicated in rail wear
and failure but have never been satisfactorily
quantified.
This limitation has been because cleanness assessment
has relied on destructive sampling and metallographic
examination, which is both time consuming and
expensive. It also suffers from a major statistical
problem in that the examined mass of rail is of the
order of a few grams, representing masses of the order
of 100 tonnes of product. That is a sampling ratio of
the order of 1 in 108. Ultrasonics has not been used,
even though sampling rates of the order of 1in 5 are
feasible because the ability to detect inclusions of the
appropriate size has not been available.
The 8000SX system can detect this size of inclusions in
rails so that feedback is fast, cheap and reliable.

DIFFICULT TO DETECT FEATURES


ULTRASONIC DETECTION OF INCLUSIONS AND
SHATTERCRACKS
RTI cooperated with BHP Steel at Whyalla in South
Australia and their research organization in
Melbourne, in 1992 to design & manufacture a DSP
inclusion detection system. This was a single channel
system designed to accept signals from an existing
water jet probe targeting the side of the railhead. The
system demonstrated proof of concept and could
reliably detect inclusions in the order of 300 microns.
When BHP adopted continuous casting it was no longer
necessary to demonstrate steel cleanness in order to
remain market competitive.
The knowledge gained from that research however, has
been developed further and applied to in-track testing
so that inclusions can now be detected in track. It is
perfectly feasible to rate the rails of different
manufacturers by driving over them with a test car.
An example of the detectability is shown in Figure 10.

Figure 10 8000SX & 2000SX Systems with Standard


WB45 Transducer looking in rail with 200-300
m
defect high in the rail head
This data was obtained on the bench from rail known
to be defective. Destructive sampling of the adjacent
rail confirmed the physical defect size. This particular
defect was from the same batch of rails that caused the
track failure shown in Figure 1.
An etched, vertical section through the rail is shown in
Figure 11. The defects reported here were those about
10mm from the running surface.

Page 5 of 12

strength of materials in that materials under load


always fail at the weakest point. This is particularly
important to inclusions in rail in service because it is
the largest inclusions which determine the weakest
point and ultimately the rail performance. The data is
plotted by determining the Weibull value given by:
W = Log(Log(1/1-P)
where
P= Probability of excedence
For natural systems a line of fit gives a slope and
intercept ( and ) in Weibull terms, which are
characteristic of the rail.
When calibrated the system produces data from which
these two values can be obtained. Comparison between
rails is then possible.

Figure 11 Horizontal Section through Defects


Ultrasonic data from this defect was downloaded as a
comma delimited file for later investigation. The
energy surface diagram shown in Figure 12 was
generated from this data. It shows the multi lobed
defect quite clearly.

Figure 12 Energy Surface Obtained from Inclusions


on Bench.
It was established during the earlier research that
inclusions and similar defects have a size distribution
that is well represented by the Weibull distribution.
This type of distribution is natural to many systems in
nature; it is the same distribution that describes the

Figure 13 Weibull Plot of Energy


It is considered feasible using this method to compare
the effects of a process change in steelmaking on the
potential rail performance or to compare the relative
cleanness of two manufacturers products in track.
The success on the bench allowed the system to be
applied in track. Defects in a known defective rail are
quite easy to find but defects in track were expected to
be much more difficult. There was a chicken and egg
situation in that it was necessary to find the defects
before they could be tested.
During commissioning of the new rail vehicle, testing
was conducted over some very old rail. The rail was
branded LORRAINE STEEL Co 1907 and weighed
80lb/yard (41kg/m). Small but clear reflectors were
reported with 38 and 70 probes.
Hand testing using a standard hand set could not
discriminate any defect from background noise.
Repeated vehicle testing confirmed the reports and
further down the track larger reflectors were identified
which could be hand confirmed. These were not testing
artefacts but inclusions. Data from one of these larger
inclusions was examined in detail.
An energy surface plot similar to that shown in Figure
12 was obtained and is shown in Figure 14. It should
be noted that even though small on hand assessment,

Page 6 of 12

these inclusions were so big that the 8000SX receiver


was saturated for much of the time. A saturated signal
precludes the use of a Weibull analysis because the
energy distribution is no longer natural.
The stringer nature of the inclusion is reflected in
mountain range appearance of the energy surface.

Figure 16 Weibull Plot for Small Inclusions

Figure 14 Energy Plot for Inclusions in 1907 Rail


A sample of the smaller inclusions was also analysed
and is shown in Figure 15. The Weibull plot obtained
from the data is shown in Figure 16
The data shown here was obtained with the zero probe
aimed normally into the rail but similar results were
obtained for the angled probes.

The fact that there is a good margin between the


threshold and the level for defects detected indicates
that the system has the ability to detect much smaller
than these if desired.
Further small inclusions have been found during
commercial operation of the 8000SX system on
Westrail track on modern rail. An example of a
particularly affected length of rail is shown in Figure
17.

Figure 17 Multiple Inclusions in Westrail Track

Figure 15 Energy Plot for smaller Inclusions


Whereas it was impossible to Weibull plot the larger
inclusions because the receiver was saturated. These
smaller inclusions fall precisely into a Weibull
distribution.

There are multiple reflectors evident. The small ones


appear to be the original rail defects and the larger
ones, defects exhibiting growth. It is not yet possible to
categorise these reflectors into inclusions, hydrogen
flaking or any alternative source. This will only be
achievable after some of them are examined
metallographically, One of the small initiators and one
of the reflectors, which appears to be growing, are
shown in Figure 18 and Figure 19.

Page 7 of 12

VERTICAL SPLIT HEAD (VSH)


An example of a vertical split head (VSH) found in
track is shown in Figure 20. Whilst ultrasonic testing
had been conducted in this track (not by RTI), this
defect had not been detected. The defect ran for several
metres before breaking out at the surface. It was then
detected visually, fortunately before any serious train
incident. This defect was reported to be about 2 metres
long when detected visually.

Figure 18 Small inclusion


In Figure 19 the reflectors have grown sufficiently that
the receiver has become saturated (level =1024).

Figure 20 Vertical Split Head


The vertical split head has been a particularly difficult
defect to find in-track in the past simply because the
probes which run in a wheel on the railhead are
looking at the edge of the defect. In analogy a sheet of
paper held flat on is very easy to see. Changing to edge
on is much more difficult.
Figure 21 shows the ultrasonic response to the 8000SX
system of a similar sized defect in the Westrail system.
The defect is clearly apparent using this test system.

Figure 19 Inclusion with apparent growth


The difference in these reflectors is readily apparent.
The initiator shows a reflector in both forward and
reverse 38 degree probe indicating a near vertical plane
but the fact that one signal is stronger than the other
suggests some tilt rather than truly vertical. The
growing defect is much more extensive and includes
returns from the zero probe as well indicating some
horizontal growth.
The initiating inclusions are larger than the threshold
identified as having the potential to initiate fatigue
failure but are much smaller than the defect threshold
set in AS1085, AREMA or European rail
specifications. As such, they would not have been cause
for rail rejection, had they been detected, but are
nevertheless clearly detectable using the 8000SX
equipment and could be eliminated in future.

Figure 21 Screen print of GRM for a Vertical split


head.

Page 8 of 12

This is a particularly large defect that was confirmed by


hand probing to be 3.3metres long. It had almost
certainly been there for many years, undetected in
previous test runs by the 2000SX technology or that of
other operators. The defect is discontinuous and was
detected by vertical 0, 38 and 70 probes. No
comfort can be obtained from the fact that this
undetected defect grew to such a large size without
breaking out or causing a derailment. Much smaller
defects of this type (less than 500mm) have been
reported to cause track failure.
A vertical split head has been detected using the
8000SX system in BHP Iron Ore Goldsworthy line at
100mm indicating that the detection threshold is well
below this failure threshold.
The data for this defect can be displayed graphically on
request as shown in Figure 22.

Figure 23 Shelling at weld in Rail


This figure is representative of shells, which, whilst
undesirable, are not usually considered cause for
removal. If they break out as shown in Figure 24 they
are visually obvious but have historically, been difficult
to detect ultrasonically because the lower reflecting face
is very close to the surface. The depth is typically about
5-10mm, that is, at the depth of highest shear stress. In
Figure 24 the condition has progressed to complete
separation of the running surface from the body of the
rail. Shells have been difficult to detect because they
are very close to the running surface and have been
masked by the ringing of the surface echo.

Figure 22 Screen print of Graph for the VSH shown


in Figure 21
The graph indicates that the defect was reported in the
left rail by zero, 70 forward, and both 70 and 38
reverse probes. The zero probe was saturated for much
of the defect and in fact produced a second reflection
with a pseudo range of 95mm.
SHELLING (SHELL) & HORIZONTAL SPLIT HEAD
(HSH)
Shelling is a very common rail defect found in most
rail systems. The defect is fatigue cracking usually
initiated internally, commonly at inclusions stringers
similar to those described above. An example is shown
in Figure 23.

Figure 24 Shelling with separation


The separation, while resulting in a poor running
surface quality and thus even higher contact stress,
makes ultrasonic testing the rest of the rail easier since
the interface between the shell and the rest of the rail,
which returns much of the energy is lost.
Without complete separation, shelling has been
responsible for difficult ultrasonic testing of the rest of
the rail. So little energy penetrates the remainder of the
rail that existing systems have been unable to detect

Page 9 of 12

defects below the shelling even at sizes of 25% or more


of the head area.
A deeper-seated defect, particularly one that spreads
across a substantial portion of the head, is reported as a
horizontal split head. The origin of this particular
defect previously illustrated in Figure 1 is an inclusion
line in the centre of the fatigue
All operators have reliably found horizontal split
heads. The difficulty associated with them is that they
are slow to grow, dont pose an immediate threat and,
like shells, are usually not removed from track.
There is a down side to this though. Traditional
ultrasonic reports show this defect as a massive
reflector, which so swamps the receiver, that even large
defects below the HSH are neglected. An example of
such a defect is shown in Figure 25. For track seriously
affected by horizontal defects it is acceptable in North
America for the track to be reported as untestable
which then requires the rail to be replaced.

Figure 26 Data for the Horizontal split head shown


in Figure 25.
Any defect in the horizontal plane whether it be
shelling or some other form of rolling contact fatigue
or as extensive as a horizontal split head, has the
potential to mask a more serious defect below.
DEFECTS BELOW OTHER DEFECTS
The most common form of this defect is shelling or
horizontal splitting from which a defect with a vertical,
transverse component has developed. An example of
such a defect that failed in service and was later broken
open is shown in Figure 27.

Figure 25 Screen print of GRM for a Horizontal


split head.
As described earlier, the data presented here as a GRM
can be called to show the full data set for this region
and is presented graphically on request. The data for
this defect is shown in Figure 26.

Figure 27 TD shadowed by shelling (after Broek)


For simplistic testing systems, the shell reflects a much
larger amount of energy than the TD. This is because it
is not only a much bigger and generally more
favourable aligned reflector, but also because after
reflection at the interface, there is much less energy
transmitted through the shell available for reflection by
the TD. The TD appears to be very small by
comparison.

Page 10 of 12

Traditional systems compound the problem because


they trigger at either the first threshold breaking
reflector or the largest reflector, which is always the
HSH/Shell. The operator then rejects the defect as noncritical since it is below threshold size for action as a
HSH and the underlying defect is also ignored.
These problems are further compounded by the limited
near surface detectability of older systems because they
are still ringing from the top surface echo. In most
instances they dont in fact report the shell/TD either.
The ability of the 8000SX system to analyse the whole
of the signal path, behaving as 32 channels avoids all
of these problems and even small defects below large
defects can be detected.
The near surface problem is avoided by the use of
Active Digital Signal Rectification as described earlier.
This permits testing within a few millimetres of the rail
surface, certainly closer than the depth of highest shear
stress which is the usual defect initiating depth. The
problem of high energy reflection at the initial defect is
overcome simply by having much more energy
available from the initial pulse so that even a small
fraction of that energy being transmitted through the
top defect is sufficient to detect defects below. The
capability is enhanced further by the automatic gain
control which linearises the energy over the whole
beam path, increasing the gain in the shadowed
region to compensate for the reduced energy being
injected past the higher defect.
Confirmation of this capability is illustrated in the
example of a TD that has turned down from a HSH
which was detected and is shown in Figure 28.

Figure 28 GRM of TD and HSH


This defect data has been examined in detail as part of
the RTI flaw signature project. A depth surface map,
shown
in
Figure 29, similar to the energy surface map was
created from the data for this defect, A twodimensional view showing the depth profile. The
change in the defect orientation can be seen from the
profile but the most striking feature is the sudden
increase in depth standard deviation at the transition.
This feature will be used as part of the defect
recognition development.

Figure 29 Depth Profile of HSH and TD


CONCLUSIONS
A new level of ultrasonic detectability is now available
which can be applied at steel plants and in track.
At the steel plant it can reduce or eliminate the
metallographic testing of rails for cleanliness and to
report on the effects of process changes.
The equipment is capable of measuring cleanliness,
non-destructively, in track and thus rating the products
of different manufacturers. It will be capable of
determining the relationship between track failures and
rail cleanliness.
In track, the new technology makes it possible to rate
the quality of existing or newly purchased rails and
allows the detection of some features not reliably or
precisely detectable in the past. These defects are:
Very small defects <1mm,
Vertical, longitudinal defects and
Defects below otherwise innocuous horizontal
defects.
The equipment has entered commercial service on
standard gauge tracks in Australia in March 2000 and
will be used in a narrow (1067mm) operation before
the middle of the year. It will be available for
steelworks applications, other gauges and as a hand
held inspection unit before the end of the year.

Page 11 of 12

Das könnte Ihnen auch gefallen