SIST-TP CEN/TR 15874:2009

SLOVENSKI STANDARD
01-september-2009
äHOH]QLãNHQDSUDYH+UXS,]YHGEDSUHVNXVD]]DKWHYDPLGRORþHQLPLYVWDQGDUGX
(1
Railway applications - Noise emission - Road test of standard for rail roughness
measurement EN 15610:2009
Bahnanwendungen - Geräuschemission - Feldversuch zu EN 15610:2006 über Messung
der Schienenrauheit im Hinblick auf die Entstehung von Rollgeräusch
Applications ferroviares - Emission de bruit - Essai de route relatif de norme pour la
mesure de rugosité de rail EN 15610:2009
Ta slovenski standard je istoveten z: CEN/TR 15874:2009
ICS:
17.140.30 Emisija hrupa transportnih Noise emitted by means of
sredstev transport
45.060.01 Železniška vozila na splošno Railway rolling stock in
general
2003-01.Slovenski inštitut za standardizacijo. Razmnoževanje celote ali delov tega standarda ni dovoljeno.

TECHNICAL REPORT
CEN/TR 15874
RAPPORT TECHNIQUE
TECHNISCHER BERICHT
May 2009
ICS 17.140.30; 93.100
English Version
Railway applications - Noise emission - Road test of standard for
rail roughness measurement EN 15610:2009
Applications ferroviares - Emission de bruit - Essai de route Bahnanwendungen - Geräuschemission - Feldversuch zu
relatif de norme pour la mesure de rugosité de rail EN EN 15610:2006 über Messung der Schienenrauheit im
15610:2009 Hinblick auf die Entstehung von Rollgeräusch
This Technical Report was approved by CEN on 28 March 2009. It has been drawn up by the Technical Committee CEN/TC 256.
CEN members are the national standards bodies of Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland,
France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal,
Romania, Slovakia, Slovenia, Spain, Sweden, Switzerland and United Kingdom.
EUROPEAN COMMITTEE FOR STANDARDIZATION
COMITÉ EUROPÉEN DE NORMALISATION
EUROPÄISCHES KOMITEE FÜR NORMUNG
Management Centre: Avenue Marnix 17, B-1000 Brussels
© 2009 CEN All rights of exploitation in any form and by any means reserved Ref. No. CEN/TR 15874:2009: E
worldwide for CEN national Members.

Contents Page
Foreword .3
1 Introduction .4
1.1 Background .4
1.2 Objectives of the road test .5
2 Brief review of the nature and requirements of the new standard .5
2.1 Longitudinal position of measurement records and sample length .5
2.2 Lateral position of the measurements on the rail head .5
2.3 Processing .6
3 The measurement programme .6
3.1 The test procedure.6
3.2 Test sites .7
3.2.1 Loriol .7
3.2.2 Wildenrath .8
3.3 Teams and instruments .9
4 Comparison of the practices of the teams .9
4.1 Choice of lateral position .9
4.1.1 Loriol .9
4.1.2 Wildenrath . 11
4.1.3 Conclusion on success of the provisions for identifying the reference surface . 12
4.2 Longitudinal sampling and cleaning the rail head . 12
5 The common analysis applied to the raw data . 13
5.1 Spike processing . 13
5.2 DFT and filtering analysis techniques . 13
5.3 Treatment of long records in which rail-head defects are present . 13
5.4 Chatter/screech . 13
5.5 Observations made on results presented in Appendices A and B . 15
5.5.1 Loriol . 15
5.5.2 Wildenrath . 15
5.6 Overall observations . 17
6 Comparisons of roughness spectra . 17
6.1 The datum line spectra . 17
6.2 The 100 m test section results . 19
7 Conclusions . 23
Annex A (informative) Results from Loriol for all instruments processed using the common
processing method . 24
Annex B (informative) Results from Wildenrath for all instruments processed using the common
processing method . 39
Annex C (informative) Review of rail-head defects encountered at Loriol . 50
Bibliography . 52

Foreword
This document (CEN/TR 15874:2009) has been prepared by Technical Committee CEN/TC 256 “Railway
Applications”, the secretariat of which is held by DIN.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. CEN [and/or CENELEC] shall not be held responsible for identifying any or all such patent rights.

1 Introduction
1.1 Background
It is well established that rolling noise originates in the combined ‘roughnesses’ of the wheel and rail running
surfaces. Through the rolling interaction of the wheel and rail this roughness imposes a time history of relative
displacement across the wheel-rail contact that leads to vibration of the wheel and of the track. This vibration,
in turn, gives rise to the noise components radiated by the wheel, the rail and the sleeper. The fact that at low
(‘normal’) levels, the roughness gives rise to noise radiation linearly and accounts for the noise fully, has been
shown by the comparison of theoretical models and carefully controlled measurements [1]. It has furthermore
entered the practice of a number of railways to control the roughness, even of uncorrugated, track as a
measure to reduce noise.
In recent years, in line with the European Union’s strategy for harmonisation of internationally running train
services in Europe, new Technical Specifications for Interoperability (TSI) have been written for the
acceptance testing of new rolling stock. The acoustic TSI reflects the understanding of the noise generation
mechanisms [2, 3]. In order to ensure that the acceptance test, that may be made at different locations on
different rolling stock, is a fair test of the rolling stock and depends as little as possible on the local track
design, the TSI specifies conditions for a ‘reference track’ on which pass-by noise measurements are to be
made. The reference track is controlled in terms of the noise produced per unit level of combined roughness
and the roughness of the rail head running surface. The first condition is characterised by a minimum decay
rate spectrum that must be obtained on the reference track (for how this relates to the noise performance of
the track see [4] and to [5] for the method of measurement). The second condition is a limit to the spectral
level of rail roughness that may exist on the reference track [6].
To ensure comparable and repeatable pass by noise measurements are made, the TSI calls upon ISO 3095.
This standard also contains an Annex concerning the measurement of roughness.
A programme of measurements of noise from both high-speed and some conventional speed rolling stock was
undertaken to test the practical applicability of the TSI method of measurements (NOEMIE project [7]). In
most respects the tests were successful but it was shown, as previously realised, that the part of ISO 3095
concerning roughness measurements is too limited in the following respects:
a) the wavelength range specified is too short for use for high speed trains;
b) too little data sampling is demanded to give the required certainty in the measured spectrum of roughness
over the wavelength required;
c) the standard is written on the assumption of a particular measurement technology; it is preferred that only
a performance criterion be implied for the quality of measurements obtained;
d) ISO 3095 imposes a fixed pattern of sample records; this sometimes causes the measurement of rail-
head defects that are not wanted in the signal and have a significant effect on the estimated spectrum;
e) the standard specified the averaging of the roughness across a number of lines at different distances
across the rail head. Since the variation across the rail-head is significant, closer specification of where to
measure is required and the data for separate lines should be presented separately.
For these reasons the TSI Committee requested CEN/TC 256, Working Group 3, to draft a new standard
solely for the measurement of acoustic roughness. It is the intention that the TSI should, in future, refer to the
new standard for this aspect.
1.2 Objectives of the road test
The purpose of the road test is to check that the standard can be interpreted consistently and leads to a
consistent estimate of roughness spectrum when used by different measurers with different instruments. Many
of the instructions of the new standard have not been practiced by measurers before and so these are also
being tested for practicability and effectiveness. The exercise is not concerned with testing instruments or
measurement technology. The standard specifies minimum performance criteria but otherwise is designed to
be as inclusive as possible with regard to technology.
In order to gain a proper understanding of the practical difficulties and the outcome in terms of consistency of
practice as well and results, it was seen as essential that the ‘road test’ should take place in an industrial
context, i.e. making measurements with instruments used by the industry on running railway lines having
normal constraints of access time and safety procedures, etc.
2 Brief review of the nature and requirements of the new standard
For the method of pass-by noise measurement, the current High Speed Rolling Stock TSI (2008) refers to
EN ISO 3095: 2005 [8]. The current Conventional Rail TSI refers to ISO 3095:2001. Having said this, there is
not a significant difference between the two versions.
The EN ISO 3095 standard itself already sets a limit spectrum for the track on which acceptance tests are
made and prescribes a method for its measurement. The limit spectrum set in EN ISO 3095 is not used in the
TSI’s, rather a tighter limit is set from within the TSI’s according to what was found possible by the associated
NOEMIE project [7]. The project also found, for high speed trains (above 200 km/h), that a minimum
wavelength range up to 0,25 m is required.
2.1 Longitudinal position of measurement records and sample length
EN ISO 3095 specifies a set of six positions for 1 or 1,2 m records of the rail-head profile. These are fixed with
respect to ‘the microphone position’. This leads occasionally to the measurement of rail-head defects, welds
etc. Such large localised irregularities are not appropriate to include in the roughness spectrum since they
create forces and noise that are not linear with their depth (the contact geometry, and therefore the contact
stiffness, changes radically). They also strongly distort the mean of the six sample records leading to both an
overestimate of the level and uncertainty in the true operational roughness level. This has been a problem
many times in the past and specifically at one of the test sites in the NOEMIE project. In the new standard, the
choice of location of the measurement records is made by the measurers and they are advised not to include
such irregularities. Moreover, the new standard envisages that a certain track section is to be characterised
rather than assuming a microphone position. (The placing of a microphone might be decided on the results or
there may be no associated noise measurements at all.)
To keep the variance in the estimated spectrum at 0,25 m wavelength consistent with that at 0,1 m in EN ISO
3095, the new standard requires there to be a 15 m sample length in total.
2.2 Lateral position of the measurements on the rail head
EN ISO 3095 requires that the ‘running band’ on the rail head be identified (as ‘clearly visible’) and 1 or 3 lines
of roughness measurement record be taken depending on its width. The new standard refers to a ‘reference
surface’ that must be defined by the measurer. The relationship of noise measurements to the measured
roughness will then be valid as long as the wheel-rail contact remains inside the reference surface. Its
identification from the running band or otherwise is an important subject in the new standard. Three different
criteria depending on the situation and the purpose of the measurements are offered:
a) the running band is visible and is known to be a product of the rolling stock for which the roughness
measurement is to be used,
b) the contact position can be measured for the specific rolling stock at the time of roughness measurement,
c) the contact position can be predicted from the geometry of rail and wheel transverse test section.
2.3 Processing
The data must be processed to remove some unwanted ‘pits and spikes’ and produce a one-third octave level
roughness spectrum. EN ISO 3095 does not prescribe how the processing is done although it recognises that
large differences can result. The processing is much more tightly controlled in the new standard. To remove
the effects of dust or grains of dirt on the railhead, an algorithm is included that removes ‘spikes’, i.e. very
short (much shorter than the wheel-rail contact patch), sharp, upward deviations. This recognises that such
features would be crushed or strongly deformed in the contact not leading to significant relative displacement
between wheel and rail. A second algorithm, ‘curvature processing’ is specified to deal with downward
features short in the direction along the rail head, found by the small tip radius probe of the instrument and
that would not affect a much larger radius wheel.
For the production of the wavelength spectrum of roughness from the measured data, the new standard
specifies alternative analysis methods,
a) Hanning window, discrete Fourier transform and averaging in one-third octave bands
or
b) digital one-third octave band filtering.
3 The measurement programme
The idea of the ‘road test’ of the new standard is
a) to have a number of different teams measure roughness according to their own interpretation of the
standard;
b) to observe the practices of the teams; and then
c) to examine the data for consistency of output.
Thus the standard should be tested in its practicality, whether it produces a consistent interpretation
implemented in the practice of different teams and whether it results in consistent roughness spectra.
Two sites were offered for the measurement exercise, one on a running line at Loriol in the south east of
France and the second at the Siemens Transportation Systems test track facility at Wildenrath in northern
Germany. Since the purpose of the standard is to fulfil the requirement of the TSI’s, it is important that the
sites should exercise the measurement of low roughness levels around and below the TSI limit curve.
A number of measurement teams were invited to come to each site and carry out measurements according to
their reading of EN 15610:2009. The measurement teams had to bear their own costs and so it was not
reasonable to require all teams to attend both sites. It was requested therefore that all teams taking part
should attend the site at Loriol. Thus, seven teams attended measurements at Loriol and five at Wildenrath.
All teams taking part were provided with software by the coordinator that attempted to perform the analysis
defined in the standard. The software was provided in open Matlab code used by some of teams and in open
FORTRAN. This was done so that teams could test and comment on the calculation procedure and raise any
areas of uncertainty in the definition of the processing.
3.1 The test procedure
At each site the teams measured separately so that there was no cross-contamination in the interpretation of
the standard. The host team at each location, required to be present for the safety arrangements, therefore
went first.
Each team was shown the test section of track, in each case 100 m long between kilometre markers at the
trackside. The teams were then asked to characterise the roughness of the test section with no other
information given except that indicated in the text below concerning the rolling stock to which their reference
surface should correspond. After the measurement was made according to their free interpretation of the
standard, each team was asked to measure a 15 m sample of roughness along a single line specified by the
coordinator. This was done to provide a means of identifying any differences in results that may be due to
instruments or the natural limits of repeatability, from those that may be due to different choices of
measurement line lateral line positions and longitudinal sampling.
Each team were at liberty to process the data themselves but all data in terms of displacement along the rail
head, were given to the coordinator. The coordinator then processed all data with the software distributed
before the measurements. This is the basis of the comparisons presented in this report.
All measurements were made within the space of a few days of one another at each site but it remains an
assumption of the exercise that no significant change in roughness occurred due to the train running during
that time.
3.2 Test sites
3.2.1 Loriol
Measurements were carried out between 14th and 24th May 2007 at Loriol on a conventional-speed service
line in southern France. The line at this site is mostly trafficked by freight trains with some regional multiple
units, locomotive-hauled passenger stock and a few TGV’s. Figure 1 shows a sample of the rail head typical
of the Loriol test section. Here the running band was wider and less distinct than at Wildenrath. In these
circumstances the teams were guided to test the contact position of the passenger stock in deciding the
position of the reference surface. A method used by one team is illustrated in Figure 1.

Figure 1 — Photograph of the railhead at Loriol
Figure 2 — Layout of the test section of track at Loriol ( — reference section, - - - datum)
3.2.2 Wildenrath
Further measurements were carried out between 22nd and 25 April on the main ring of the Siemens Test
Track Centre at Wildenrath in northern Germany. The rail-head had been ground about 6 months before the
test using a special ‘acoustic grinding’ with longitudinal grinding action. Figure 3 shows a typical sample of the
rail head at this site. There were very few significant defects of the rail head within the 100 m ‘reference
section’ of track. However, an interesting consideration arises; the site is used for testing rolling stock with
(mainly new) 1 in 20 and 1 in 40 coned wheel profiles. This has resulted in two clear separate (narrow)
running bands. The line speed is 120 km/h.

Figure 3 — Photograph of the railhead at the Wilderath test site

Figure 4 — Layout of the test section of track at Wildenrath (— reference section, - - - datum)
3.3 Teams and instruments
At the Loriol site, seven teams took part with eight instruments. Three separate types of instrument measured
1.2 m records using linear voltage displacement transducers (LVDT’s) that moved along a straight edge fixed
in position relative to the rail. Two types of instrument measured continuously over the whole 100 m using an
accelerometer moved along the rail head by a light ‘trolley’. All teams that took part in the test measured at
Loriol. The team-instrument combinations for the measurements at Loriol are indicated in Table 1.
Table 1 — The team-instrument combinations at Loriol
Team- Instrument type Technology
instrument
A 1 1,2 m fixed straight edge with moving displacement transducer
B 1 1,2 m fixed straight edge with moving displacement transducer
C 2 1,2 m fixed straight edge with moving displacement transducer
D 2 1,2 m fixed straight edge with moving displacement transducer
E 4 1,2 m fixed straight edge with moving displacement transducer
G 5 Accelerometer trolley
H 3 Accelerometer trolley
At the Wildenrath site, five teams took part using four of the 1,2-metre fixed straight-edge instruments of two
different types. The fifth team used an accelerometer trolley. The team-instrument combinations are set out in
Table 2.
Table 2 — The team-instrument combinations at Wildenrath
Team- Instrument type Technology
instrument
A 1 1,2 m fixed straight edge with moving displacement transducer
B 1 1,2 m fixed straight edge with moving displacement transducer
C 2 1,2 m fixed straight edge with moving displacement transducer
D 2 1,2 m fixed straight edge with moving displacement transducer
F 2 1,2 m fixed straight edge with moving displacement transducer
I 3 Accelerometer trolley
4 Comparison of the practices of the teams
The test coordinator observed the practice of each team in response to the instructions in the standard.
4.1 Choice of lateral position
4.1.1 Loriol
At this site the running band is the product of mixed traffic and this led to a little difficulty for some in deciding
the width of the reference surface. Each team used a method of marking the rail at both ends of the test
section (some teams used additional positions) and observing the width rubbed off by passing trains (the
second method prescribed in the standard). The method worked well with a wide range of paints and markers
used, but best with thin coating of ink from marker pens rather than thick coating of paint.
When this method carried out for the modern passenger stock this led to a narrower assessment than for the
older, more worn wheels of the freight stock. Team G in particular made a wider estimate than others on the
far rail based on the passage of a freight train. Thus team G initially placed three lines 10 mm apart on the far
rail. However, all teams were asked to consider the reference surface for the modern passenger stock and
this led to a re-evaluation by team G to measure at positions 5 mm apart.
Team H used a lateral rail-head profile measuring device on site before making their decision. The lateral
profile was then used in a ‘static’ geometrical calculation of the running position with a standard unworn profile
of the wheel. For illustration the output of this calculation is shown in Figure 5. This information was then used
in conjunction with the erased band of paint in order to reach the decision. While it was unnecessary under the
circumstances of the test with the relevant rolling stock passing regularly so that the marker method could be
used, the exercise showed the practicality of the third method offered in the draft standard.

Figure 5 — Output of the on-site guide calculation of likely running position
The decisions on reference surface width and line positions chosen by the different teams is summarised in
Table 3.
Table 3 — Chosen lateral measurement positions at Loriol
Team- Line position(s) chosen Comments
Width identified (mm)
instrument (mm from gauge face)
near far near far The coordinator chose a line at
rail rail rail rail 39 mm for the datum
measurements
A 22 19 39 ± 5 39 ± 5 Centre-line chosen to be same
on both rails for convenience
B 28 28 37 ± 5 34 ± 5
C 22 22 38 ± 5 38 ± 5 Centre-line chosen to be same
on both rails for convenience
D - - 36 ± 5 36 ± 5
E 15 – 30 15 – 30 39,5 ± 5 39,5 ± 5
G 25 35, 25 43 ± 5 37 ± 5, 10 Revised decision on far rail for
modern passenger stock
H - - 37 ± 5 37 ± 5
All teams decided to measure 3 lines at Loriol, 5 mm apart. For the near rail, the range of the centre-lines was
from 36 mm to 43 mm with no team placing their centre-line further than 4,5 mm from the mean position of
38,5 mm. For the far rail, the situation is not very different with a range of centre-lines from 34 to 39,5 mm
from the gauge face. Thus no centre-line was placed more than 3 mm from the mean position of 37 mm.
4.1.2 Wildenrath
The nature of the two running bands at Wildenrath has already been shown in Figure 3. This situation may
well arise in measurements of rail roughness in the future and in connection with the TSI’s where two
country’s rolling stock runs on the same tracks. The measurers were directed to consider the more recent,
brighter band of the two. The decisions on the width of the running band, the number of lines of roughness
required and their lateral position at Wildenrath are summarised in Table 4.
Table 4 — Chosen lateral measurement positions at Wildenrath
Team- Comments
Line position(s) chosen
Width identified (mm)
instrument
(mm from gauge face)
near far near far The coordinator chose a line at
rail rail rail rail 40 mm for the datum
measurements
A 10 10 37 37
B 16, 11 11 34 37 Initial estimate of running band
width was re-evaluated during
measurements
C 11 11 35 35
D 10 10 40 40
F 10 10 37 37
I 12 15, 12 37 38 Measured three lines on far rail
but decided only one was
needed when re-evaluated the
consistency of the running band
width along the site
At Wildenrath all teams eventually decided that only one line of measurement was required, the confusion
being caused by the presence of the second running band and whether a partially worn region between them
should be included or not. The observation that this partially worn region was not continuous along the whole
100 m of the test section made those who were wavering clear in their decision that a narrower operating
running band was correct.
4.1.3 Conclusion on success of the provisions for identifying the reference surface
Given the differences in the running band of the two sites and their relationship to the rolling stock, the first
two techniques used for identifying the reference surface, see 1.2, worked well and led to closely similar
positions of the reference surface.
The decision on running band could be aided by some improved wording of the standard advising the
measurers to consider the reference surface only to lie within the width that is continuous along the track and
also to ignore surface that is only partially worn. The wording relating the reference surface to the rolling stock
of interest is clearly necessary and useful as it was invoked at both sites.
One team measured the rail head profile and calculated a theoretical (static-geometry) contact position for an
unworn wheel; thus demonstrating the practicality of the third approach in the standard to determining the
reference surface position.
4.2 Longitudinal sampling and cleaning the rail head
Different teams had different practices in cleaning the rail head before measurement. The teams using short
record instruments used solvent and rags. It was clearly not practicable for the long-record measuring teams
to follow this practice. At Loriol the rail head was regularly ‘cleaned’ apart from easily-moved dust or moisture,
by the running trains. The cleaning practice may have been more significant at Wildenrath where, at the start
of the test, there were a lot of bird droppings on the rail head. Apart from removal of some gross matter, the
one long-record measurement at Wildenrath was made without cleaning this from the rail.
For the 1,2 m-record instruments, most teams took the strategy of scattering the 13 to 16 records locations
(approximately) evenly over the 100 m. One team, however, placed their records in a pattern strongly
weighted towards the mid-point of the 100 m.
The trolley instruments measured the whole 100 m with extra length on the ends so that start-up and stopping
effects could be discarded from the record afterwards.
In the standard, the measurer is instructed to exclude rail-head defects. The reason for this is that a defect on
the scale of curvature and length of the wheel-rail contact patch changes the contact stiffness momentarily as
the wheel rolls over it. It is known from modelling studies [1, 9] that, for this reason, these features do not lead
to a linearly proportionate generation of noise and that a roughness measurement excluding these features
agrees well with measured rolling noise [10, 11].
Upon encountering a geometrical feature judged to be excludable, the teams using 1,2 m measuring
instruments merely moved their instruments or decided not to include that record in their average. (Most
teams took at least 16 × 1,2 m = 19,2 m over the 100 m rather than the minimum 13 × 1,2 m to make up the
minimum requirement of 15 m of data.)
Where 100 m of record is taken in one go, it is inevitable that a number of ‘rail-head defects’ and features
such as welds are also measured. However, none of the measurement teams using trolleys identified them
and avoided recording them. Rather, it is naturally the practice of these teams to measure the whole 100 m
and then to remove these features afterwards. Although this is compatible with the practice of the 1,2 m
instrument measurers in avoiding rail head defects, no normative procedure or advice for a posteriori
identification and removal of data has been given in the standard.
No editing of the raw data of the trolley instruments to remove rail-head defects was done before handing the
raw data to the coordinator.
5 The common analysis applied to the raw data
All the data were analysed by the coordinator using the Matlab version of the processing algorithm.
5.1 Spike processing
A point of ambiguity was discovered in the standard that affects the processing of ‘spikes’. These are
identified as features (maxima) in the data that are short in the rail axial direction, x, (height in metres > w /3)
and have a small radius of curvature (absolute value of the second derivative with respect to
7 2
distance > 10 µm/m ). The ambiguity exists in whether only upward ‘spikes’ of this type are to be removed or
whether downward, ‘pits’ are also to be removed. Past practice by some organisations is to do both but these
organisations did not carry out the subsequent curvature processing that treats the pits by applying a
simplified physical argument removing pits by running the large curve radius of the wheel over the data.
It was discussed during the measurement exercises and agreed by all measurement teams, only to remove
upward features according to the spike removal processing and to rely on the curvature analysis to treat
downward features. This clarifies the philosophy of each part of the processing:
a) the upward features are dirt that can be removed if small enough in relation to the contact patch;
b) the pits are reduced using the physical interpretation of the large radius of curvature of a wheel compared
to that of the measurement probe.
All data was therefore treated by the agreed spike removal and the curvature analysis.
5.2 DFT and filtering analysis techniques
The 1,2 m records were analysed using the DFT procedure stated in the standard.
The 100 m records were analysed whole using both the DFT technique and the alternative digital filtering
technique offered in the standard. It was found that this could not consistently be applied to 1,2 m or even
(concatenated) 3 m records of data because of the starting and ending transients of the filters. It was
determined that these transients affect approximately 2 m of data at each end of the record (based on 1 mm
or 0,5 mm sampling). Thus the processing was changed to discard 2 m at each end. In the case of long
records (100 m) this makes little difference but clearly rules out use of the digital filtering method for 1,2 m
instruments.
5.3 Treatment of long records in which rail-head defects are present
As already discussed in clause 4, a clear difference in practice of the teams arose out of the nature of taking
100 m of data in one record compared with those taking individual records of 1,2 m. Thus the 100 m records
analysed whole contain the effects of the rail-head defects that were not avoided. (This is commented on with
respect to specific results below.) In order to compare results on a more equitable basis between different
instruments, the whole 100 m records were chopped into segments of 1,2 m by the coordinator and examined
to see if they contained features that were clearly rail-head defects that should be excluded or features that
would have caused the straight 1,2 m measurers to have rejected that record. A selection of 15 ‘clean’ 1,2 m,
records, approximately evenly spaced along the 100 m was then analysed in the same fashion as the discrete
records taken with the 1,2 m instruments.
It is not being suggested here that this procedure of selecting data ought to be sanctioned in the standard.
Any practical implications in the efficiency and therefore cost of the work should be taken into account.
5.4 Chatter/screech
At both sites, a number of instruments suffered from a slip-stick excitation of the probe as it was moved along
the rail head. This gives rise to a screeching sound during measurements but it was not always easy to hear.
It is thought to be a similar mechanism to the ‘chatter’ of a lathe tool. This is known to occur during roughness
measurements sometimes but its higher-than-usual rate of occurrence during the tests may be related to the
very hot and dry weather conditions that most of the teams measured under. These conditions are known to
give rise to high friction coefficients on the rail. Figure 6 illustrates the effect of chatter on the measurement
record. It shows that it causes continuous high-amplitude, short-wavelength features to be recorded such that
it is inappropriate to rely on the spike removal and curvature processing to improve the data.
It must be emphasised that all types of instrument suffered at least some measure of this effect. In some
cases it was noticed while taking measurements, in other cases contaminated measurements were made
without it being noticed at the time.
All results for each team-instrument combination are plotted separately in Annex A for Loriol and Annex B for
Wildenrath. This includes the additional 15 × 1,2 m analysis for each of the trolley cases. In each case, the
average spectrum arising from the measurement of each line of roughness is shown as required by the draft
standard.
In addition a measure of the spread of data is indicated. This is the standard deviation of the dB levels of the
component spectra making up the average in each case. (It has been shown elsewhere in the work of
CEN/TC 256 WG 3 that this is a suitable measure of spread of the data that does not follow Gaussian
statistics in its linear form.) This measure is approximately independent of the number of records. This
standard deviation is then plotted either side of the arithmetic mean of the spectral levels, however, of course,
it is the energy mean of the spectra that is required as the mean roughness level. The spread is therefore not
plotted symmetrically either side of the energy-mean spectrum. Only the spread of the central line of
roughness measurement is plotted. The spread of the other lines is similar in all cases. It is not possible to
derive a measure of the standard error from this measure of spread, i.e. the confidence in the mean. This
must be judged from the comparison of the results of measurement of the same roughness, see 5.2.

Figure 6 — Close look at some data before (—) and after
spike removal and curvature processing (- - - )
5.5 Observations made on results presented in Appendices A and B
The following observations can be made on the results presented individually for each instrument-team
combination and each rail.
5.5.1 Loriol
a) Measurements by team-instrument combinations A to E have detail differences but are similar in spread
as well as spectral level. They each show similar levels of roughness on the three lines within the
reference surface. This tends to be close for long wavelengths and for the shortest wavelengths of the
range of interest but with some significant differences in the mid range (0,008 m to 0,63 m) in some of the
results.
b) The results from team-instrument G (Figures A16 to A20) are known not to be valid for wavelengths
shorter than about 0,025 m. The reason for this is that the instrument has two contact probes. The one
that would normally be used for acoustic roughness screeched on the rail head at Loriol although it has
not done so elsewhere. A contact used for making roughness measurements for a longer wavelength
range (in connection with ground vibration) was therefore used instead.
c) The results for the trolley instruments G and H, show that the digital filtering technique and the DFT
technique of analysis produce closely similar results in most cases. The digital filtering technique
produces an estimated spectrum that is slightly higher in most bands but not consistently so. This is
probably due to the approximate nature of the correction for the removal of energy from the DFT
spectrum by the use of the Hanning window. (The Welch periodogram technique is a statistical method of
estimating the spectrum of a signal.)
d) It is seen in Figures A18 (G) and A24 (H), comparing the DFT and digital filtering results for the datum
15 m line, that there is a greater difference between the different analyses. The main reason for this is
the fact that the digital filtering technique discards 4 m of the signal (2 m) at each end because of
transients; this is a significant proportion of the 15 m record but not of the 100 m records analysed in
Figures A16, A7, A22 and A23 where the difference is very small.
e) Figures A23 and A24 show a much higher level on the two outer lines of measurement in the 0.02 m
wavelength and short wavelength bands. This is due to some intermittent screeching during these
measurements rather than measurement outside the valid reference surface. Since this, by chance, only
affected the outer lines of measurement on each rail head significantly and not the central line (used
below in comparisons between instrument-teams) it was corrected by selection of ‘clean’ 1.2 m records in
Figures A25 to A27.
5.5.2 Wildenrath
a) There is no data for the 15 m line for team-instrument B at Wildenrath.
b) Spectra from team-instruments A, B, C, D and F at Wildenrath are broadly similar. Measurements from
the only trolley instrument used at this site, I, show a strong rise in the spectrum at long wavelengths that
is not observed by the other teams. In particular Figures B16, B19 and B20 show a distinct peak in
roughness around the 0,125 m band on the far rail. This feature was visible at the time of measurement
(see photograph Figure 7.) However, none of the other measurements (D taken before I on the same
day, all others after I) do not show this, indeed they show a rail that is exceptionally smooth at these
wavelengths. Besides the visual evidence of the existence of this feature, it is clear from the Loriol
measurements that the higher level measured is not an intrinsic feature of the type of instrument (H is the
same type though not the same individual instrument.) Moreover, no malfunction of instrument I has been
found. One possible explanation is that the feature consists of the contaminant on the rail head at this
site. It would have been cleaned off where the
1,2 m instrument D measured and would probably have been removed by the rolling stock movements
after measurements with instrument I and before the other instruments. However, it is counter intuitive
that this should have caused a raise in the roughness at longer wavelengths and not at shorter
wavelengths. T
...