Biuletyn WAT

Vol. LXIII, Nr 4, 2014

Continuous wavelet transform of railway track

defectoscopic signals in the matlab wavelet toolbox
Vitalij Nichoga1, Liubomyr Vashchyshyn2
1

Lviv Polytechnic National University, 12 S. Bandery Str., 79013 Lviv, Ukraine,

nich@org.lviv.net
2
Karpenko Physico-Mechanical Institute of theNAS of Ukraine, 5 Naukova Str., 79601 Lviv,
Ukraine, myalternative@inbox.ru
Abstract. Rail networks across theworld are getting busier with trains travelling at higher speeds and
carrying more passengers and heavier axle loads than ever before. Thecombination of these factors has
put considerable pressure ontheexisting infrastructure, leading to increased demands ininspection
and maintenance of rail assets [1]. Nowadays, rails are systematically inspected for internal and surface
defects using various non-destructive evaluation (NDE) techniques, themost common of which
are ultrasonic and magnetic flux leakage (MFL) methods. Thearticle is focused ontheanalysis of
defectoscopic signals received using themagnetic wagon-defectoscope of Lviv Railway (MFL method)
by thecontinuous wavelet transform (CWT).
Keywords: rail, crack, CWT
DOI: 10.5604/12345865.1131322

1. Introduction
The most important question inall NDE methods is selection of information
about defects from defectoscopic signals received during thechecking railway
(defectogram). At present, wagon-defectoscope operators experience is essential
for making thecorrect decision about technical condition of rails. To improve
theoperators work efficiency, which is mainly based onexpert visual assessments,
itis necessary to automate theanalysis of therecorded signals. That is why we turned
to thepowerful tool of digital signal processing such as wavelet transform (WT).
Themain WT applications analysis and processing signals, non-stationary intime

22

V. Nichoga, L. Vashchyshyn

(such as defectoscopic signals) when theanalysis should include not only thesignal
frequency characteristics, but also information about some local coordinates, which
reveal themselves inone or other frequency components.
WT divides into continuous (CWT) and discrete (DWT). DWT focuses onspeed (by
sampling values of scale

(frequency) and time (samples)) and thepossibility of acomplete

reconstruction of thesignal after analysis (by orthogonal wavelet functions), which led
to its use, mostly for denoising and compression of signals. As for CWT, itrequires only
one wavelet function mother, whose creation is not problematical (unlike DWT)
and allows to identify those signals that are similar to themother wavelet.
Therefore, we should focus onconsideration of opportunities of CWT to identify
thesignals from defects indefectogram.

2. Analyzing defectoscopic signals by CWT

This research was carried out inapackage Wavelet Toolbox of thecomputer

mathematics system MATLAB. Wavelet Toolbox provides functions and anapp
for developing wavelet-based algorithms for theanalysis, synthesis, denoising and
compression of signals and images [2].
The analysis process includes thefollowing steps:
1. Starting theContinuous Wavelet 1-D Tool. From theMATLAB prompt,
type wavemenu. Intheopened Wavelet Toolbox main menu window click
theContinuous Wavelet 1-D menu item. Thecontinuous wavelet analysis
tool for one-dimensional signal data appears (Fig. 1 without loaded signal
for analyzing and calculated wavelet coefficients).

Fig. 1. Thecontinuous wavelet analysis tool for one-dimensional signal data with loaded and analyzed
defectoscopic signal

Continuous wavelet transform of railway track defectoscopic signals...

23

2. Loading asignal. To do this, inthemenu File choose Load Signal and

intheappeared dialog box select thesignal to be analyzed. Thefragment of
defectoscopic signal with apattern from transverse crack inthehead of rail
(Fig. 2 and theupper part of Fig. 1) was selected as anobject of research.
This fragment was received during thechecking railway Lviv-Syanky-Chop,
06.11.2009 (km: 36 picket: 6 speed: 47 km/h). At theedges of it, thesignals
from therail joints are shown, and along thewhole of its length thesignals
from therail substrates (similar to background noise) are also shown.

Fig. 2. Thefragment of defectoscopic signal with apattern from transverse crack intherail head received
during thechecking railway Lviv-Syanky-Chop, 06.11.2009 (km: 36 picket: 6 speed: 47 km/h)

3. Configuring CWT parameters. Panel intheright part of Fig. 1 serves for

selecting theappropriate settings. Name of theloaded signal is displayed
intheData (Size) field. IntheWavelet field it is selected thetype of wavelet
for CWT (mother wavelet function), which should be similar to thesignals
that we are interested in, because CWT is acorrelation between amother
wavelet at different scales and thesignal [3]. If none of thestandard Matlab
wavelets fits for research, it is possible to create own wavelet by clicking
onthemain menu Wavelet Toolbox button New Wavelet for CWT and
then to follow by therecommendations from theofficial website Matlab
[4, 5]. As themother wavelet, for our study, it was used awavelet adapted
for detection of signals from thetransverse cracks intherail head cr21
(Fig. 3). Theprocess of its creation is described in[6].

V. Nichoga, L. Vashchyshyn

24

Relative amplitude

0.5

0.5

-1

1.5
0

0.2

0.4
0.6
Time

0.8

Fig. 3. Wavelet adapted for detection of signals from thetransverse cracks intherail head

In theSampling Period field, thesampling period of thesignal should

be specified. Wagon-defectoscope velocity for theaforementioned fragment was 47 km/h and discrete samples were taken with thefrequency
100samples/m. Multiplying these two values givethesampling frequency
approximately equal to 1306 Hz, then theperiod is thereciprocal to frequency 0.000766s. IntheScale Settings group, it is necessary to select
scale mode. Step by Step Mode for setting theminimum (1), maximum
(40), and step values (1) were selected.
4. Performing aContinuous Wavelet Transform. To calculate thewavelet
coefficients it is necessary to click ontheAnalyze button. Result of CWT
is shown inthescalogram (second from thetop inFig. 1). Scalogramis
avisual method of displaying awavelet transform. There are three axes
for representing time (samples), scales and coefficient values [7]. To show
(orhide) thewavelet coefficients line (figure of coefficients inaparticular
scale 3rd from thetop inFig. 1) and local maxima (figure of local maxima
oneach coefficients line- 4th from thetop inFig. 1)) select (or deselect)
Coefficients Line and Maxima Lines respectively. Colour for displaying coefficients onscalogramcan be selected using thetools provided inthelower
right corner of Fig. 1.

Continuous wavelet transform of railway track defectoscopic signals...

25

3. Results
A value of wavelet coefficients onscalogram is shown by varying shades of grey
colour. Thehigher values of thewavelet coefficients are, thedarker they reflected.
Defect is represented by avertical line of dark colour, which indicates that theshape of
its signal is very similar to theadapted wavelet. Since thetransverse crack amplitude
inthefragment of defectoscopic signal is sufficiently large (its detection will not make
any problem for anoperator) there is anecessity to reduce it to thedetection level
of this defect (3 signal amplitudes from substrates 3 sub) to test thesensitivity
of CWT where theoperator can make mistake.
Defectogram segment with thereduced amplitude of asignal from thedefect
(which is actually masked by thebackground signal from thesubstrates) is shown
inthetop of Fig. 4. Result of analysis is represented under thesignal inFig. 4.

Fig. 4. Signal with thereduced transverse crack (3 sub) and its scalogram

As it can be seen, thecoefficients of CWT intheplace of transverse crack

(2172 sample) are well allocated inarange of small scales (high frequencies), which
indicates thepossibility of detecting weakly developed defects.
Figure 5 shows thedependence of thewavelet coefficients values from scale
inthelocation of transverse crack. Wavelet coefficients have their maximum value
intherange of scales from 8 to 21, and preferably that theassessment of thedefect
presence was made by thesmaller scale (corresponding to high frequencies),
soitwill determine its location more accurately. Scales from 5th to 8th also can
be used for this purpose, it all depends onwhose minimum amplitude signals we
need to identify.

V. Nichoga, L. Vashchyshyn

26
0.4
Wavelet coefficients value

0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05

10

15

20

25

30

35

40

Scale

Fig. 5. Dependence of wavelet coefficients from scale inthelocation signal from thetransverse crack
(2172 sample)

From theabove, we can formulate thebasic terms of quality defects detection

by CWT:
1. A good choice of mother wavelet for CWT.
2. Selection of theoptimum scales quantity (for transverse cracks is offered
intherange of 8 to 21 see Fig. 5), by which we can judge about thepresence or absence of defect with high probability.
3. Optimum thresholds onthevalue of thewavelet coefficients for each scale,
which is selected according to theminimum signal level that is necessary
to reveal.

4. Summary
1. MATLAB Wavelet Toolbox is avery convenient tool for analyzing defectoscopic signals.
2. The basic conditions for thequalitative detection of signals from defects of
railway by CWT are formulated. These conditions are applied for all defects
detected by themagnetic wagon-defectoscope.
3. Determined that CWT allows for detecting thesignals from transverse
cracks intheinitial stages of their development that is, when their amplitude
is approximately equal to 3 sub and it can be missed by theoperator.
Received July 1, 2014. Revised September 18, 2014.

Continuous wavelet transform of railway track defectoscopic signals...

27

References
[1] Ramesh S., Detection of Cracks and Railway Collision Avoidance System, International Journal
of Electronic and Electrical Engineering, vol. 4, no. 3, 2011, 321-327.
[2] http://www.mathworks.com/products/wavelet/
[3] Polikar R., TheWavelet tutorial, Rowan University, College of Engineering Web Servers, 2001,
79.
[4] http://www.mathworks.com/help/wavelet/gs/new-wavelet-for-cwt.html
[5] http://www.mathworks.com/help/wavelet/ug/adding-your-own-wavelets.html
[6] Nichoga V., Vashchyshyn L., Storozh I., Creating themother wavelet function and its use for
analysis of therail transversal crack non-destructive testing signal, Information-control systems
inrailroad transport, no. 3, 2012, 61-69 (in Ukrainian).
[7] http://en.wikipedia.org/wiki/Scalogram
The paper has been prepared onthebasis of thespeech given at 10th School-Conference onComputer
Aided Metrology 2014, Waplewo, 27-30 May, 2014.

V. NICHOGA, L. VASHCHYSHYN
Ciga transformacja falkowa sygnaw inspekcji torw kolejowych
w matlab wavelet toolbox

Streszczenie. Sieci kolejowe na caym wiecie s coraz bardziej zatoczone, przy coraz wyszych prdkociach pocigw i przewo coraz wicej pasaerw przy wikszym nacisku na osi k ni dotychczas.
Poczenie tych czynnikw oznacza powane zagroenie dla istniejcej infrastruktury, co prowadzi
do zwikszonego zapotrzebowania na ilo inspekcji i na koszty utrzymania aparatury kolejowej [1].
Obecnie, szyny s systematycznie sprawdzane pod ktem uszkodze wewntrznych ipowierzchniowych za pomoc rnych metod bada nieniszczcych. Najbardziej upowszechnione z nich to metody
ultradwikowe i magnetodynamiczna (Magnetic Flux Leakage Rail Inspection MFL). Artyku
skupia si na analizie sygnaw defektoskopowych otrzymanych przy uyciu wagonu defektoskopii
magnetycznej Kolei Lwowskiej (elektromagnetyczna metoda bada nieniszczcych) wykorzystujcej
cig transformat falkow.
Sowa kluczowe: szyna, szczelina, CWT