Novel approaches for processing of multi-channels NDT signals for damage detection in conveyor belts with steel cords

spiritualblurtedAI and Robotics

Nov 24, 2013 (3 years and 8 months ago)

70 views


Novel approaches for processing of multi
-
channels NDT signals for
damage detection in conveyor belts with steel cords

Ryszard Blazej
1, a
,

Leszek Jurdziak
1,b
and Radoslaw Zimroz
1,2,c


1
Machinery Systems Division, ul. Na Grobli

15, 50
-
421 Wroclaw, Wroclaw University of
Technology, PL

2

Diagnostics and Vibro
-
Acoustics Science Laboratory, ul. Na Grobli 15, 50
-
421 Wroclaw,

Wroclaw University of Technology, PL

a,b,c
{ryszard.blazej, leszek.jurdziak, leszek.jurdziak, radosław.zimroz
}@pwr.wroc.pl

Keywords:

conveyor belt, non
-
destructive testing, signal processing, damage detection


Abstract.

Belt conveyors are one of the most popular methods of material transport in many
branches of industry, especially in mining. The average length o
f mining belt conveyor is about
1000 m. Taking into account that total length of transportation ways in averaged mine can approach
several dozen of kilometers and network of several conveyors may cover large area, maintenance of
such specific transportatio
n system is very difficult. In this work we propose an automatic multi
-
channel system for data acquisition and processing for damage detection in belts. Belts with steel
cords are considered here, they consist of top and bottom rubber covers and steel cord
s in between

them
. Due to many reasons (mainly sharpness of transported materials) covers may be damaged and
it may initiate degradation process or straight damages in steel cords. Properties of steel cords are
therefore crucial for overall strength of bel
ts, if they are damaged, it may cause catastrophic failure
of the whole conveyor. So, monitoring of belts conditions is a crucial issue. We proposed a
monitoring system that measure and process data from array of magnetic sensors. The system
allows to acqu
ire up to 24 channels of NDT signals and uses automatic algorithms to process them
in order to get information about begin of the belt loop, location of joints of particular belt segments
and the final location and size of damages related to corrosion or c
uts of steel cords inside belts.
These techniques will be presented in the paper. Our approach has been validated in a lignite mine
for several conveyor belts.

Introduction

A continues (belt conveyor based) transportation system is very popular method of m
aterial
transportation in many industrial branches. One of the more important users of such conveying
machines are mining companies. Today over 500 km of conveyor belts are installed in 4 lignite
surface mines in Poland, namely Belchatow, Turow, Konin and
Adamow. Even more conveyor
belts are installed in underground mines (e.g. only in KGHM Polish Copper there are over 260 km
of belts) (Blazej, Jurdziak, 2011).

Fig 1a presents simplified diagram of a belt conveyor, which consists of: a drive unit, a belt l
oop,
idlers and some auxiliary elements. A crucial problem for such companies is how to
operate/maintain such complex and large systems that often consists of more than 50 conveyors
with a total transportation length around 70 km, spatially located on a m
ine terrain with 100 km2
area.

There are
many different problems:
some recent works that partially tried to fix maintenance
problem by condition monitoring of drive units (Bartelmus and Zimroz2008, Zimroz and
Bartelmus2012), proposals of new design and te
sting methods for idlers (Gladysiewicz et al 2011),
modelling and simulations tools (Kulinowski 2013) to optimise design of whole conveyor and
finally modelling, destructive and non
-
destructive testing for belts, including also belt joints
(Mazurkiewicz 20
09, Blazej et al 2010,2011a,b, Zeng 2012, Fedorko 2013).


One of key problems for maintenance is current assessment of all belts quality (Harrison 1996,
Blum 1996, Mazurkiewicz 2009, Blazej et al 2010,2011a,b, Zeng 2012, Fedorko 2013). Several
factors cause

belts exposure to the risk of damage. These include: properties of transported
materials (sharp edges of ore lumps), the impact of environmental factors (rain, snow, wide
temperature range) and load variations inside belts due to non
-
stationarity of load
stream. In order
to be able to transport materials on the belt, they are designed as a sandwich structure with top and
bottom rubber covers and steel cords inside (Fig 1b). One of the most dangerous damage is change
of condition of steel cords. It appears
mainly due to damage of covers by sharp elements that can
also damage or cut cord wires. Often cords are worn due to environmental factors (corrosion due to
rain and moisture). It is worthy to add that this method of transportation is very effective, howev
er
cost of 1m of belt used in the system is nearly 250 Euros. So taking into account the scale of a
transportation network, problem of maintenance becomes also important from economic point of
view.

a)










b)


Idlers

Loading

point

Conveyor belt

Drive pulley

Return pulley


Figure.1 a)
Simplified diagram of a belt conveyor, b) Internal structure of a steel cord belt


To overcome the problem of unexpected belts failures, engineers from mining companies have
started to use special equipment for NDT (Non
-
Destructive Testing) signal acquisit
ion and analysis.

Length of a typical belt conveyor is about 1 000 m, so length of its belt loop is a bit than 2 000
m. Standard speed of the belt is 5.25 m/s. Until now, all belt conditions analyses have been and still
are in many places by personal visua
l inspection. It takes a lot of time, it is subjective rather than
objective, and it is not automatic and computerized. Therefore it is proposed here to use acquired
time series of diagnostic signals and process them automatically by applying proposed proc
edure of
damages detection.

A procedure for data acquisition and processing.

In the proposed approach several key steps can be distinguished. It should be underlined that one
should take care about experiment because there are many factors that can affect
acquired signals
and further processing might not be successful. Experimental setup, precise installation of pre
-
magnetizing head, magnetic markers and magnetic head (sensors array) should be done carefully
and according to procedure described in next sect
ion. At this stage it is assumed that signals are
acquired correctly, i.e. each time series from 24
-
channels magnetic head is not overloaded, at least
one of channels 1
-
4 contains minimum two signatures related to magnetic marker (what means
measurement fo
r at least one belt loop).

The procedure of signals processing contains of 3 main tasks:

A)

Acquired signal should contain at least 1 cycle related to belt loop. After measurement
signal will be segmented according to magnetic marker installed on the belt.

Detection of the
marker allows to extract part of the signal that correspond to one full cycle.

B)

Signals from the belt contains two different signatures: from magnetic marker, from pre
-
magnetized belt steel cords. However, local disturbances in the sign
al may be related to change of
steel cord belt condition or belt joints. It is necessary to distinguish them and use for further analysis
only signal disturbances related to damage.


C)

Number, size and location of damages are unknown, based on these parame
ter final results
should support maintenance staff to make decision on belt replacement. It is proposed to use simple
statistical analysis and two dimensional visualization.


So taking into account assumptions mentioned above a procedure might be formulate
d as follow:

Step 1:

Identification of belt loop (one cycle)

Step 2:
Belt’s joints identification

Step 3:
Signal segmentation (joints removal)

Step 4:
Damage detection

Step 5:
Statistical analysis


Step1:
Identification of belt loop and single cycle extract
ion

In order to identify “beginning” of belt loop it is recommended to install small magnetic markers
on one of the border of the belt. It is assumed that magnetic marker will induce impulse when
passing magnetic head. To assure proper identification (in c
ase of problem with signal energy based
detection) 3 markers should be installed with a priori assumed distance. So we are looking for a
pattern that consist of 3 peaks with specific distance. Due to several belt widths (types) used in
mining industry it m
ay happened that signature related to marker will appear with different intensity
in several channels. During experiments it has been found that due to serious contamination coming
from belt joints, damages on belt borders etc., detection of markers in the

signal may be difficult. It
is proposed to use normalization of data and “information extraction” using two channels. It was
validated that by normalisation of signals we can neglect problem of different energy on several
channels. Obviously it will re
-
sc
ale signal only, signal to noise ratio will not be improved.

In order to minimise influence of non
-
informative components we propose kind of “data fusion”
approach, namely signal from channel “A” with the most visible signature related to the magnetic
mark
er will be combined with another channel “B” located closely to A, with small (or without)
presence of marker but with the same influence related to joints or damages. Practically in our test
channel A was subtracted with B=A±1, i.e.









In en
hanced S_AB signal to noise ratio is significantly better that for any original signal, simple
peak detection might be applied in order to identify belt loop indicators. It might happened that due
to three markers mounted on the belt (a,b,c), a family of d
isturbances will appear in the signal.








































(1)


In order to identify “beginning” of the loop, i.e. to select one of three indicators, we used simple
rule based on distance analysis. Distance between
neighboring

impulses should be bigger than
theoretical distance calculated based on assumed distance betw
een markers. To avoid false
detection of other peaks threshold was assumed to eliminate small peaks








{

(

)







(

)

















(

)


(

)







































}
.

(2)


Where
Th_small_peaks

is arbitrary threshold defined as (










(


)
)








{

(

)







(

)


(



)















(

)


(

)





















































}
. (3)


Where
Th_distance

is estimated value of threshold defined based on distance related to mounted
markers


Step 2:
Belt joints identifications

Based on detected belt loop indicators one full cycle should be extracted. Further analysis can be
done for single cycle or for several cycles. In this work only single cycle is analysed. Condition of
the belt steel cords is described by location, amplitud
e and number of disturbances of acquired
signals. Unfortunately change of signal amplitude might be related to damaged steel cords or belt
joints (in both cases number of cords, i.e. magnetic field is varying in time see figure 2)


Figure 2. Belt damage
and belt joints [Blazej et al2003].


So there is a need to indentify joints related signal disturbance, remove them from the signal and
apply peak detection procedure for remaining data. Again, one might take advantage from
correlation between signals. Bel
t joints will produce disturbance at every channel however it is
nearly impossible to obtain similar effect in case of damage due to random character of damage
location. So,


(

)




(



)





.

(4)


Where

(



)


{

(



)




















(



)




(



)


(



)







(



)


}

(5)

ch=1,...,24,




To simplify joints detection procedure only negative values
of signals are considered.
Accumulation of amplitudes is higher for samples in presence of joints. For other samples due to
randomness of signals amplitudes will be averaged out, so in consequence smaller.

So
,

finally peak detection procedure might be used

for finding negative peaks related to joints
and array of indexes of negative peaks location











will be obtained.


Signal segmentation (joints removal)

Based on a priori knowledge on joints length (joints technology is t
he same for all joints) and
sampling frequency, joints removing procedure might be simply done by zeroing samples before
and after detected negative peaks. So
,

first we assumed new data matrix is equal data matrix
containing one cycle of measurement and next zeroing procedure is applied.


(



)





(



)


(6)


(



)



(


























)




(7)







have bee
n chosen experimentally based on joints design parameters.

Step 4: Damage detection

Damage detection is based on peak detection approach in window moving along the signal. In
previous work combination of ku
rtosis and variance of signal [
Zimroz et al 2011]
was used. It
seems to be closer to physics of phenomenon to use amplitude of peak instead of kurtosis.
Amplitude of the peak corresponds directly to change of magnetic field produced by cords. Peak
detection procedure is applied separately for each channel
; length of the window should be small to
Splice length
1 step
2 step
3 step

provide suitable selectivity in time domain, however narrow window will affect computational time.
To avoid big number of small peaks only amplitudes of peaks higher than
Th_peak
>0.5

are
considered (arbitrary value
). Matrix processing in Matlab environment requires the same length of
each column (channel), it is obvious that number of damages for each channel can be different. To
overcome this problem damage matrix D was initialized as matrix with the same size as s
ignal
matrix, but with all values equal to zero. If peak is detected in signal matrix for any channel, at the
same time indices in current channel of damage matrix value equals peak value is set. So, finally,
one will obtain matrix D=(M,N) with column N=24

corresponding to number of channels and
M=length of the signal. Values of matrix D are related to peak amplitudes.

Step 5: Statistical analysis and result visualisation

Damage matrix D is an input data in decision making procedure. Several requirements ha
ve been
formulated by end user. For example, “energy” of damages along belt length is needed. It can be
simply done by summation of number of damages for each i=1,..,M. In Matlab environment, taking
advantage on matrix operation it can be simple written:

Damage_Intensity_L= sum(


)

. (8)

Where (



means transposed damage matrix.

Another simple indicator of the belt steel cords condition may be energy of damage
vs. belt
width. By analogy it is defined as:

Damage_Intensity_W=sum(

)



(9)


Apart from sum of amplitudes, one may ask for number of damages at time index or
vs. belt
width. Please note that it is not the same information as previous. It may happened that small
number will appear, however with advanced development. They are much more dangerous that big
number damages with small amplitudes.

Finally, visualisatio
n of damage matrix is provided in two
dimensional plot.

Experiments

Control measurements in the mine provide verification of functions of the system and
correctness and efficiency of individual components including:

-

the correctness of recording signals f
rom all 24 channels,

-

identification of the belt loop based on the signals from the mounted magnet marker,

-

data acquisition by the module and quick preview of recorded data,

-

Measurement of other parameters, such as the position of the measuring head

and the distance
from the tested belt, the position of the permanent magnet relative to the magnetic head, the
influence of environmental conditions

Tests of the system were carried out on two overburden and two ramp conveyors (Fig. 3).


Figure 3. Belt’s
steel core condition measurements using modernized EyeQ system installed on
two overburden conveyors in one of lignite mines





Application to real data

In this section we will present and discus results of application proposed procedure to real data
(NDT s
ignals) captured in the mine during experiment described in previous section

(Fig 4)
. We
will follow the convention from section 2 where procedure has been divided into several steps.



Figure 4. Example of raw signals (24 channels).


Step 1: Identifi
cation of belt loop (one cycle)

Fig
.5

presents several subplots with normalized absolute values of signals for channels 1
-
4. They
are presented to prove that by subtracting signals from channels 3 and 1 we can cancel disturbances
related to other than mark
er sources. From bottom subplot one ma clearly seen indicators of
magnetic markers related to belt loop. It is a need to use segmentation of raw signal because during
experiment one may acquire more than one cycle.


Figure 5. Normalized absolute values of

signals for channels 1
-
4 and difference between
channels 3 and 1 (with clearly seen indicators of magnetic markers related to belt loop)


Step 2: belt’s joints identification

After signal segmentation (loop cycle extraction) there is a need to distinguish disturbances
coming from joints and damaged steel cords. It might be clearly seen that belt and damage signature

in NDT signals are very different and there are probably many
techniques that may fix recognition
problem

(Fig6)
. However, our intent was to use as much as simple approach that can be quick and
reliable. Unfortunately simple energy based detectors failed because it may happen that joints or
damage will produce distur
bance with similar energy. As discussed above, due to synchronization
of channels, simple summation of negative values should amplify joint signature and cancel
damage
-
related signature because they are usually non
-
synchronic.


Figure 6. Belt and damage s
ignature in NDT signals


Figure 7 shows result of proposed approach.


Figure 7. Results of joint detection procedure
-

sum of negative values of signals for all channels.


Step 3


signal segmentation (joints removal)

After belt joints detection, joints
-
r
elated segments are simply removed (real values are replaced
with zeros) and signal contain damage related disturbances only.


Figure 8. Results of joint detection procedure
-

detected belt joints


result for first five channels


So, after joints removal
, simple peak detection procedure can be applied to detect disturbances
related to damage.



Figure 9. Segmented signals


first 12 channels (after joints related segment removal)


Step 4: damage detection

Figure 10 shows results of peak detection procedu
re for NDT signals (first 12 channels). Each
detected peak is marked by red circle


Figure 10. Results of peak detection procedure for NDT signal (after joints related segments
removal)

Stage 5: statistical analysis

Final step in our procedure is related

to general description of belt steel cords condition. Due to
multidimensionality of the problem, we used two dimensional plot to visualize how many, how
large and where damages are located on the belt (Fig 11).


Figure 11. Spatial distribution of detecte
d damages.



Simple histogram describing distribution of damages along belt width is presented below
(Fig.12). It can be clearly seen that most of damages is located in the middle of the belt because
main material stream is located there. Other important information is

symmetrical distribution of
damages around central line of the belt what is a consequence of symmetry of conveyor design.
Lack of asymmetry shows however that conveyor construction was also symmetrical and there were
no big misalignment of belt during ope
ration.

Misalignment of conveyor structure and belt running should lead to asymmetrical distribution of
damages.


Figure 12. Histogram of number of damages with respect to channel number (along belt width)
for a given length of belt section or full belt

loop.

Conclusion

A novel procedure for belt condition monitoring using NDT approach has been presented in the
paper. The idea of non
-
destructive testing using magnetic signal is well known (Harrison, 1996),
however, thanks to advanced data acquisition sy
stem (24
-
channels) and proposed novel
multidimensional signals processing procedure it has been shown that condition monitoring can be
quick, automatic and might significantly support maintenance of large, spatially distributed
mechanical system.

The desc
ribed diagnostic procedure is a p
a
rt of the bigger project which aim is to work out
comprehensive set of devices and algorithm processing collected data and making integrated
evaluations of belts condition in order to take rational decisions about their re
pairs and
replacements. Procedures connected with magnetic evaluation of belt wear degree will be integrated
with information obtained from machine vision system and are designed to replace subjective
classification of belt failures provided now by inspect
ion personnel. The integrated system will also
p
rotect belt and conveyor before catastrophic events such
a
s longitudinal rips or excessive side
movements through generation of diverse alarms. It will also help in preparation of essential repair
works
schedule in frame of comprehensive and integrated diagnostic system. Close cooperation
with future users assures that worked out system will fulfil their needs.

Acknowledgements

This paper was financially supported by Applied Research Programme (Program Ba
dan
Stosowanych) in the path A: “Intelligent system for automated testing and continuous diagnosis of
the conveyor belt”

(
2012
-
2015
)
.



References

[1]

W. Bartelmus, R. Zimroz: Belt conveyor driving system vibro
-
acoustic severity reduction by
condition based

maintenance, Prace Naukowe Instytutu Gornictwa Politechniki Wroclawskiej 123
(2008) 5
-
16

[2]

R. Zimroz, W. Bartelmus, Application of adaptive filtering for weak impulsive signal recovery
for bearings local damage detection in complex mining mechanical sys
tems working under
condition of varying load, Diffusion and Defect Data Pt.B: Solid State Phenomena 180 (2012) 250
-
257

[3]

L. Gladysiewicz, R. Krol,, J. Bukowski: Eksperymentalne badania oporów ruchu przenośnika
taśmowego | [Tests of belt conveyor resistan
ce to motion], Maintenance and Reliability 51/3 (2011)
17
-
25

[4]

P. Kulinowski,: Simulation studies as the part of an integrated design process dealing with belt
conveyor operation, Maintenance and Reliability 15/1 (2013) 83
-
88

[5]

R. Błażej, R. Makowski,
R: Zimroz A method of damage detecting in conveyor belts with steel
cords. The Polish patent application
-

No P 393527, 31.12.2010.

[6]

R. Błażej, L. Jurdziak: Integrated diagnostic device for automatic assessment of conveyor belts
condition. Proceedings
of 22nd World Mining Congress & Expo, 11
-
16 September, Istanbul
-
(2011). Vol. 3 / ed. Şinasi Eskikaya. Ankara, pp 675
-
680


[7]

R. Zimroz, R. Makowski R. Błażej: A method of damage detection in conveyor belts with steel
cords by NDT signal processing. Procee
dings of The Eighth International Conference on Condition
Monitoring and Machinery Failure Prevention Technologies, CM2011/MFPT2011 Cardiff, Wales,
(20
-
22 June 2011), Coxmoor Publishing Company

[8]

D. Mazurkiewicz, Problems of numerical simulation of stres
s and strain in the area of the
adhesive
-
bonded joint of a conveyor belt, Archives of Civil and Mechanical Engineering, 9/2
(2009) 75
-
91.


[9]

Q.
-
L. Zeng, J.
-
G.Wang, L. Wang, C.
-
L.Wang, The research of coal mine conveyor belt tearing
based on digital imag
e processing, Advances in Intelligent Systems and Computing 181 AISC
(2012) 187
-
191

[10]

G. Fedorko, V. Molnár, J. Živčák, M. Dovica, N. Husáková: Failure analysis of textile rubber
conveyor belt damaged by dynamic wear. Engineering Failure Analysis 28 (20
13) 103
-
114

[11]

A. Harrison, 15 Years of Conveyor Belt Nondestructive Evaluation, Bulk Solids Handling 16/1
(1996).

[12]

D Blum, Scanning steel cord conveyor belts with the “Belt C.A.T.” MDR system. Journal of
Bulk Solids Handling 16/3 (1996) 437.

[13]

R. Blazej, M. Hardygóra: Modeling of shear stresses in multiply belt splices, Bulk Solids
Handling 23/4 (2003) 234
-
241.