Acta Polytechnica Hungarica Vol. 21, No.

1, 2024

Optimum Train Weighing in Motion using

Inertial Sensors

Milad Kazemian1, Majid Movahedi Rad2, Morad Shadfar3,

Ahmad Mohammadi Doost3, Ebrahim Hadizadeh Raisi4,
Szabolcs Fischer2
Faculty of Civil Engineering, Islamic Azad University Science and Research
1

Branch, Hesarak 1477893855, Tehran, Iran, e-mail: [email protected]

2
Department of Structural and Geotechnical Engineering, Széchenyi István
University, Egyetem tér 1, Győr 9026, Hungary, e-mail:
{majidmr,fischersz}@sze.hu
3
Department of Railway Rolling Stock Engineering, School of Railway
Engineering, Iran University of Science and Technology, Narmak 13114-16846,
Tehran, Iran, e-mail: [email protected], [email protected]
4
Research and Education Department of Technical and Infrastructure Deputy,
General Administration of Khorasan Railway, Kamyab Boulevard 91735-173,
Mashhad, Iran, e-mail: [email protected]
Corresponding author e-mail: [email protected]

Abstract: Continuous monitoring with advanced equipment and innovative scientific

techniques is essential for timely and perfect maintenance. The interaction and dynamic force
between wheel and rail is one of the most widely monitored issues. In this paper's case study
of ballasted railway tracks in Iran, a set consisting of two separate strain gauge arrays and
three different positions for installing accelerometers were designed according to the
conditions. After installation, the system was calibrated with a predetermined passing axle
load. The dependency of the arrays' and the equipment's installation location with the
velocity of the passing axle load was examined as part of the field study after repeated
investigation and comparisons of the setups' results. In order to gather data with the least
error and the highest level of accuracy, it was decided on the more appropriate array with
less dependence and a better installation position.

Keywords: Railway monitoring; Maintenance; Measurement; Strain gauges; Accelerometer

‒ 221 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

1 Introduction
In today's world, a country's infrastructure is viewed as its capital and national
wealth since its capacity to develop and progress is determined by how effectively
its infrastructure performs. The rail transport industry, which significantly
influences people's everyday lives and the environment, is an essential measure of
a country's growth. As a result, a condition monitoring system must be established
to detect any defects that may appear as safety issues. Consequently, precise
information may be acquired by designing novel equipment or integrating cutting-
edge technology into a complete monitoring system [1, 2, 3]. Both experimental
and numerical studies have the potential to produce big advances. When monitoring
railway tracks, many researchers in this field are interested in the interactions and
dynamic forces between wheels and rails. However, little experimental research has
been conducted on the dynamic effect of heavier freight trains on railway tracks.
Strain gauges and accelerometers are two monitoring-related technologies that,
based on their functional capacity and sensitivity, should be employed in ideal
conditions and an ideal combination, requiring a scientifically intelligent design.
Keith et al. [4]. In one research, experimental field tests were conducted to
investigate how heavy loads induced by large axle loads influenced the dynamic
behavior of the railway track. It is demonstrated that when various speed and axle
load changes are applied to the wheel-rail dynamic force, dynamic deformation of
the track structure, and track vibration behavior, some of the indicators reflecting
the dynamic behavior of the railway track increase roughly linearly with train
running speed and axle load, while others are barely influenced. Zhiyong et al. [5].
The project created a wheel load detector based on a strain gauge for monitoring
wheel-rail contact forces at insulated rail joints (IRJs). The laboratory and field
testing results showed that the design followed generally recognized theoretical
assumptions. Field data vividly depicted the wheel-rail impact force produced
across the joint gap, demonstrating its use in recognizing the wheel-rail contact-
impact force signature at IRJs. Manicka et al. [6]. The necessary fusion technique
and the outlined theoretical relationships between the samples gathered by various
sensors demonstrated in a comprehensive analysis that the features of the wheel
defects used in the data generation step were entirely represented in the defect
signals reconstructed by the suggested method. Consequently, the proposed
technique enables early defect detection and identification, including small and
long-wave flaws. The number of sensors, the effective zone size, and the wheel's
circumference, which acts as the defect signal's basic period, all impact the fusion
process. Alemi et al. [7]. Wheel flaws on railroad wagons have been discovered as
a substantial cause of damage to the railroad's infrastructure and rolling equipment,
in addition to creating costly noise and vibration emissions. A sensor network is
being created for permanent installation on the railway network. Shelling, flat areas,
and a lack of roundness are all issues. It outperforms current defect identification
approaches for flat spots and predicts the other two kinds of defects. The neural
network technique explicitly simulates the multisensory structure of the
measurement system via numerous instances, learning, and shift invariant networks
to increase performance on wheels with flat areas and non-roundness. Gabriel et al.

‒ 222 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

[8]. Vertical overloads may cause track degradation and safety breaches, but wheel-
rail lateral contact forces are more directly connected to running safety. According
to the computational and experimental data, the size of the lateral force may be
estimated using an independent coefficient obtained from the applied loads. Bruner
et al. [9]. The curvature of the wheel profile impacts a train's performance in various
ways. In a field test, the station monitors the lateral and vertical wheel/rail forces at
the point of contact in a 484-meter-radius curve at speeds up to 100 km/h. In a bogie,
the four-wheel positions have markedly distinct force signatures. While the strength
of the three other high rails increases with distance, the leading high rail has strong
forces unaffected by the change in running distance. Palo et al. [10]. The issue's
placement and sensor needs were investigated in a study on the static and dynamic
behavior of ballasted railway tracks. It could be helpful to estimate stress transfer
from the train passage to the track using predictive computational models. Georges
et al. [11]. Weighing in motion systems would assist in solving the shortcomings of
conventional static weighing, such as costs and traffic management. Weigh-in-
motion systems, however, do not allow direct measurement of the static load since
the dynamic interaction between the train and the track results in dynamic loads
added to the static ones. Investigating the effect of track unevenness and train speed
on the weights measured by the weigh-in-motion system. A rigorous statistical
investigation based on multiple computations was done to achieve that purpose.
In order to estimate the static load, a technique to rectify the direct result supplied
by the weigh in motion system is presented. This strategy is based on the results and
patterns discovered throughout the extensive parametric investigation. Mosleh et al.
[12]. The successful operation of a field test system methodology and the system's
optimization should be thoroughly evaluated with regard to various types of trains.
This is influenced by the type of sensors and where they are set up. Mosleh et al.
[13]. The installation of sensors along a railway track's entire length will allow real-
time monitoring of the states of its technical components, and the proposal of a
diagnostic sensor system based on railway track stress-strain analysis. Avsievich et
al. [14]. The initial stage in data processing is to determine the speed of a passing
train after identifying it (time, date, and direction). For the purpose of calculating
train speeds, each peak of a vertical acceleration signal represents the passage of a
train axle above the considered accelerometer. Blanc et al. [15]. The ability to
predict failure of track infrastructure components can be improved through
maintenance prioritization and procedures, thereby enhancing the safety of railway
operations. Edwards et al. [16]. It may be helpful to use an alternative model-based
method based on the local response function method that can forecast accurate stress
results in particular locations without the need for direct measurement data at those
locations. Menghini et al. [17]. The degree of fit between actual and estimated track
forces and irregularities is demonstrated through point-by-point graphs of actual
and estimated values, and indicators of the accuracy of estimates are generated
using R-Squared values, which represent the percentage by which the variance of
mistakes is less than the variance of actual values. Gadhave et al. [18]. When
performing calibration and the primary test according to the desired plan, it is
crucial to take into account the impact that imbalanced vertical loads have on the
data recorded in the monitoring system. Silva et al. [19]. A signal analysis approach

‒ 223 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

based on the analytical domain and monitoring equipment suitable for conditions in
the environment will be productive in providing a suitable instrument for studying
experimental non-stationary vibration signals comprising shock. Salehi et al. [20].
Physical calibrations are crucial in the strain recovery process, in addition to placing
the strain gauges optimally to reduce uncertainty in the simulation model. Nieminen
et al. [21]. The installation of sensors should take into account the best site for
installation, and the layout of the sensors really offers a platform for simple
installation and a more accurate data recording system. Jing et al. [22]. The smart
rail pad demonstrated excellent signal response to fluctuations in loading, with an
average percent error of 6%. It can be used for load sensing and as an axle counter
to measure wheel loads. Schalkwyk et al. [23]. In order to track the structural
performance of the track and to identify changes in traffic volumes and loading
circumstances, analyzing rail pads—a crucial component of railway structural
design—that are fitted with various types of sensors is very beneficial. Sol-Sánchez
et al. [24]. A simulation study is conducted as part of the additional analysis of the
findings to establish the precise location of the sensors based on the rail's fatigue
life, the likelihood of mistakes and redundancy, and various railway traffic
situations. Pillai et al. [25]. Comparison of the distributed acoustic sensor (DAS)
results with point location measurements made using a conventional strain gauge
and deflections determined via imaging and digital image correlation (DIC),
provides accurate distributed strain measurements with the capability for
continuous spatial and temporal coverage across substantial tracts of track. These
measurements translate to estimates of track deflection and load. Milne et al. [26].
When a sizable high number of strain gauge pair installations are made in the track,
a realistic assessment of the static load may be made. Therefore, a statistical
correlation must be taken into account when calculating the static load from weigh-
in-motion (WIM) systems with fewer sensors. This means that the estimation of the
static load interval should take a certain level of confidence into account for each
of the vehicle wheelsets. Pintão et al. [27]. Shear strain data at specific locations
has been used to develop mechanics-based algorithms to estimate the speed and
wheel loads of trains passing over a bridge under study. The speed estimation
algorithm uses shear strain at quarter span and takes the train's speed variation into
account. For wheel load estimation, two algorithms are studied. The shear strain
algorithm only makes use of the peak value of the shear strain measured at an axial
location close to the bridge. [28]. With the advancement of the transportation
industry and the demand, the settlement and wear of the ballast caused by dynamic
stresses at high frequencies is still an important issue on high-speed tracks, leading
to high maintenance costs. Khairallahet et al. [29]. A more thorough, real-time, and
current understanding of the railway track condition would be made possible by
predictive maintenance operations carried out with cutting-edge monitoring
systems. As a result, maintenance expenses would go down because interventions
would only be made when they were truly necessary. In-depth discussion will also
be given on monitoring systems that keep track of temperature, stresses,
displacements, strain measurements, train speed, mass, and position, axle counts,
wheel flaws, rail settlements, wear and tear, and the condition of railroad bridges
and tunnels [30, 31]. A field investigation Two stiff common crossings were the

‒ 224 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

subject of experimental observations under the dynamic stress of a high-speed train.

According to the findings, the maximum displacements within a specific velocity
range increased by 2.5 times. Due to the relative displacement measurements
between the rail and the sleeper, the different dynamic impact loading for the wing
rail, and the lack of a difference in displacements between the trailing and facing
travel directions, measured maximal strains are not as explicit as the displacement
results. Kovalchuk et al. [32]. Ingenious methods and routine inspections are two
of the key factors that influence the cost of railway maintenance, according to both
laboratory and numerical studies. Modern monitoring techniques, prognostics, and
health management strategies provide a wealth of opportunities for improving
inspection and maintenance procedures. [33, 34].
However, there are still lots of required optimization for different stages of
measuring wheel-rail force from software and hardware points of view. Most
research are presented with a fixed arrays of sensors, which is one of the major
challenge in correct data acquisition. It also needs to consider track maintenance
issues. The challenge of correct data acquisition mainly lies in sensor array and
arrangement design, which determines quality of the signal. In this paper, different
arrays and locations of the strain gauges and accelerometers are studied for data
collection. The results are analysed and discussed and sensitivity against velocity
are also considered.

2 Field test

2.1 Scope of Study

Figure 1
The test site and its location relative to Mashhad railway station

‒ 225 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

The location of the study area which is between the Chaman and Sarkhes bridges
and 2,392 meters away from the Mashhad railway station, is shown in more detail
in Figure 1, along with its position in relation to the Mashhad railway station.
The aforementioned location has two input and output lines that are connected to
the Chaman concrete deck bridge on the west side and the railway switches on the
east side, respectively. The passing train speed is approximately 80 km/h, and test
equipment was installed on tangent railroad part in the exit line.

2.2 Railway Track System

Figure 2 depicts the layout of a ballasted railroad track. The sleepers are attached to
two rails (there is an elastic pad between the rail and the sleepers). This set is
situated on a ballast layer, allowing for the safe and easy passage of trains.
The characteristics of the ballast layer, rails, and rail pads affect the track's dynamic
performance. To keep track performance, maintenance actions are needed which
are mainly focused on the aera between two sleepers. The tamping machine is used
to keep the ballast layer quality in check which uses rail web on the area between
sleepers and support. There are also some actions required for bolt tightening.
Considering all of this, the rail web in the area on the sleeper would be the safest
place for sensor installation. The sensor must be able to provide accurate data on
this location, too. The following is a quick list of the test site pavement's technical
specifications:
 Rail type: UIC60 heavy, rail weight is 60.34 kg/m
 Sleeper type: Mono block concrete sleeper, total weight of the sleeper is
260 kg
 Fastening type: Weslo type spring fastener, its weight is 503 gr
 Ballast type: Mostly made of porphyry and metamorphic rocks
As mentioned in the Introduction, accelerometers can measure track dynamic
performance, whereas strain gauges are typically less sensitive to track dynamics
and track structure failures. As a result, both kinds of sensors should be used to
measure all static and dynamic components of a contact force measurement.

Figure 2
Railway components details

‒ 226 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

2.3 Test Equipment

As previously mentioned, the instruments were installed on a tangent part of the
railroad so the operation conditions, such as speed and bogie normal function,
would be under control, and any abnormality within the bogie and wheels would be
monitored with higher accuracy. Specifications for used strain gauges,
accelerometers, and data logger can be found in Tables 1 to 3.
Table 1
Strain gauge technical specification used for field test (detailed)

Applicable Metal, Glass,

specimen Ceramics

Backing Special plastics

Element Cu-Ni

Operational Temperature
temperature -196 ~ +150°C compensation +10 ~ +100°C
(°C) range (°C)

Applicable
Strain limit 5% (50000×10 -6 strain) CN, P-2, EB-2
adhesive

Fatigue life
Gauge length
at room 1×10 6 (±1500×10-6 strain) 0.5
(mm)
temperature

Gauge Backing length

1.2 5
width (mm) (mm)

Backing
2.2 Resistance (Ω) 120
width (mm)

‒ 227 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

Table 2
Technical specification of the wired accelerometer used in field test

AC102-1A

Sensitivity Spectral Noise @ 14

100 mV/g
(±10%) 10 Hz μg/√Hz

Frequency Spectral Noise @ 2.3

0,5-15000 Hz
Response (±3dB) 100 Hz μg/√Hz

Frequency
Spectral Noise @ 2
Response 2,0-10000 Hz
1000 Hz μg/√Hz
(±10%)

Output <100
Dynamic Range ± 50 g, peak
Impedance ohm

Bias Output 10-14

Settling Time <2.5 seconds
Voltage VDC

Voltage Source 18-30 VDC Case Isolation >108ohm

Constant Current
2-10 mA
Excitation

Table 3
Technical specification of data logger used in field test

Analog inputs 8 ch voltage, IEPE, current (with ext. Shunt)

16 bit SAR with 100 kHz 5th order analog AAF filter or
ADC type
bypass (500 kHz)
Sampling rate Simultaneous 1 MS/s

Ranges ±10 V, ±5 V, ±1 V, ±0.2 V

Typ. SNR @ 100 kHz 89 dB

Input coupling DC or AC (1 Hz)

Input impedance 1 MΩ
4 or 8 mA excitation; Sensor detection (Short: <4 V; Open: >
IEPE mode
19 V)
TEDS Supported in IEPE mode

Overvoltage protection 50 V continuous; 200 V peak (10 msec)

Typical power consumption (max.) 15 (22 W)

‒ 228 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

2.4 Tested Rail Fleet

The fleet used in this project includes Siemens locomotives and passenger wagons
are made in Germany, whose pictures are shown in Figure 3 and their technical
specifications are in Tables 4 and 5.

Figure 3
Passing locomotive and wagons used in the test program

Table 4
General specification of Locomotive used in field test
Ax-to- Ax-to- Wheel
Loco Min Ax Ax diamet Axle
Fleet
type Made in speed Max distance distance er load
(km/h) speed of two of two (mm) (tonne)
(km/h) bogies axles in
(mm) a bogie
(mm)

Locom ER24PC Germany 20

39.4 160 10362 2700 1100
2011-
otive
BO-BO 2016

Table 5
General specification of Wagons used in field test
Max Wagon Wagon
wagon Made Min speed weight - weight – Bogie Axle load
Fleet
e type in speed (km/h empty full type of (tonne)
(km/h) ) (tonne) (tonne) wagon

Wag RL11 Germa 11.25

- 140 45 50 MD36
ny
on

‒ 229 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

3 Test Plan
The following key factors should be taken into account when measuring the wheel-
rail contact force:
 Repeatability of measurements, taking train wheel radius and speed into
account.
 Installation time should be minimized; the instruments should not interfere
with maintenance vehicles if long-term measurement is required.
 To be able to avoid the need for expensive speed detectors, the vehicle's
speed should be calculated with the highest degree of accuracy.
 The force calculation method should be able to analyze all data as quickly
and accurately as possible and report the results.
 The configuration should be safe and not require routine manual
calibration and maintenance.
The sensor formation was designed with the previously mentioned issues in mind.
Figure 4 shows a schematic representation of the rail web with the installed
equipment and sensor configuration. All the accelerometers are in vertical direction
and strain gauges are installed as half-bridge circuit. The strain gauge arrays in this
paper come in two varieties: "V" and "<". In comparison, the "<"-shaped array is
concentrated on the bending strain, and the "V"-shaped arrays deal with shear strain
when there is a wheel passage. Normally, calculation of wheel vertical force focuses
on bending strain, which has an interference with maintenance issues. The V array
is introduced to cover this shortcoming.

Strain gauge

Accelerometer

Figure 4
Configuration of strain gauges and accelerometers on the rail web

To make sure that no data is missing, it is necessary to measure in a complete round.

So, for every passage, the whole wheel surface is scanned. The designed sensor
arrangement and localization are done to ensure the mentioned condition is met.
This arrangement is also able to analyze the wheel multiple times with the help of
different combinations of installed sensors, so the error would reduce noticeably. It
means a higher calculation volume with no change in the response time of the
device. With the configuration shown in Figure 5, one side of the passing train is
completely covered, allowing for accurate comparisons. Accelerometers are very
sensitive to the track structure's defects, in contrast to strain gauges, whose results
are essentially independent of such defects. So keeping track's quality high in the

‒ 230 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

instrumented area is crucial. The accelerometers' installation position can also help
determine the wheel's dynamic component. In order to investigate the dynamic
component of the vertical forces of the wheels, three different accelerometer
positions are taken into account: 1) on the rail foot in the rail's position on the
sleeper, 2) on the sleeper, and 3) on the rail foot, between the sleepers.

4
8

Accelerometer

Strain gauge

1
2
3 4 5 6 7
1 2 3

1: 2, 3, 4:

Figure 5
Locations of sensors and their types

In order to control operation conditions like speed and bogie normal function and
to more accurately monitor any abnormalities within the bogie and wheels, the
instruments were installed on a tangent track, as it is shown in Figure 6.

‒ 231 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

Figure 6 Figure 7
Instrumentation location on tangent track The drezin used for system calibration

As shown in Figure 7, the system was initially calibrated using a drezin (a light rail
vehicle) with known axle load. Table 6 describes the calibration procedure, which
also includes load and speed. Using locomotives from passing trains with axle loads
of 19.75 tons, the main calibration is carried out (9.875 tons per wheel).
Table 6
Calibration condition for drezin

Test condition Number of tests

V=5 km/h 2

V=10 km/h 2

V=20 km/h 2

V=40 km/h 2

4 Result

4.1 Strain Gauge Results

Results of recorded data caused by drezin passage is illustrated in Figures 8 for
strain channels. The sudden jump in the diagram is caused by shear strain, which is
proportional to the axle load and velocity as the wheel passes over the sensor. These
quantities can be found and measured to determine the wheel load. By calculating
the phase difference between these sensors, velocity can also be determined.

‒ 232 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

30
20
10

Strain (micro strain)

0
-10
-20
-30
-40
-50
-60 < array
-70

Figure 8
Measured strain for "<" array (up) and "V" array (down)

The drezin used for primary calibration weighs 4.8 tons and has a wheelbase of 2.7
meters. By plotting the shear strain mentioned in Figures 8 in terms of velocity,
Figure 9 is obtained. The value shown for each speed is the average of all channels
and passages. The dependency of the two mentioned arrays on velocity is relatively
low and linear. It should also be noted that the proposed "V" arrangement has a
lower dependency on velocity compared to the "<" arrangement. These would
enable measuring the static components of the signal with higher accuracy due to
the probable error in the velocity calculation. Additionally, the mean squared error
for "V" array is lower than the conventional "<" array.

‒ 233 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

Figure 9
Linear pattern for velocity – drezin passage results, Force (up) and Strain (down)

The results of the entire passage for a passenger train pulling a locomotive weighing
79 tons are shown in Figure 10. The values in Figure 10-top are for Channel 1 in
Figure 5 with the "V" arrangement, and the rest are for Channels 2–4 with the "<"
arrangement, respectively. The train passages are completely recorded by both
arrays, which display the same pattern. Therefore, the "V"-shaped array can be
considered an alternative to the "<"-shaped array, which conflicts with track
maintenance activities.
Locomotive Wagon

200
Channel 1
(V array)

-200

0 2 4 6 8 10 12 14 16

200

0
Channel 2
(< array)

-200

-400
0 2 4 6 8 10 12 14 16

0
Channel 3
(< array)

-200

-400

0 2 4 6 8 10 12 14 16

400

200
Channel 4
(< array)

0 2 4 6 8 10 12 14 16
Time (s)

Figure 10
Strain results for a passenger train

Figure 11 displays the strain results proportional to vertical loads for a locomotive
(79 tons in weight) for various passages. Higher speeds and axle loads can produce
the same patterns and outcomes (compared to primary calibration). Despite the
different velocity dependencies, the pattern is still linear. While this parameter has
increased for the "<"-shaped array from 0.283 to 0.4833, it has decreased from

‒ 234 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

0.056 to 0.0326 for the "V"-shaped array. So calibration is required for various
speed ranges. These findings demonstrate the benefits of a "V"-shaped array's lower
dependence on speed when needed. It should be noted that more research should be
done to determine how sensitive this arrangement is to track failures. Similar to the
drezin test, as it is shown in figure 11, values for error is lower in "V" array.

Figure 11
Linear pattern for velocity in vertical load calculation – Locomotive passage results, Force (up) and
Strain (down)

4.2 Acceleration Results

As earlier described, accelerometers are much more sensitive than strain gauges to
the track structure's defects, the results of which are essentially independent of
them. So keeping track's quality high in the instrumented area is crucial.
The accelerometers' installation position can also help determine the wheel's
dynamic component. In this research, the authors installed accelerometers at three
different positions: on the rail foot in the position of the rail on the sleeper, on the
sleeper, and the rail foot in the distance between sleepers, to investigate the dynamic
component of the vertical forces of the wheel and rail (see Figures 4 and 5). Figure
12 shows the three positions' time domain and Short Time Fourier Transform
(STFT) [35] values. This transformation has a 1024-point window with a 1000-
point overlap. The window type is Gaussian, and the sampling frequency is 10 kHz.
The wheel passage is clearly shown in Figure 12-b. In other words, the rail and rail
pad had a low-pass filtration influence on the acceleration signal as it passed

‒ 235 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

through them. There is also much noise in Figures 12-a and 12-c. This illustration
emphasizes the importance of signal filtering in order to identify various failures.
The characteristics of the track superstructure have a significant impact on the
accelerometer's response as well. In summary, it can be said that moving the
accelerometer from between the sleepers (channel 3) to the sleepers (channel 1) has
no detrimental impact on the precision of the measured signal. But results from
channel 2 can be used with minimum pre-processing.

500
4
0

Power/frequency (dB/Hz)
Frequency (kHz)
ACC channel 1

0
a 2 -50

-500
-100
0
0 2 4 6 8 10 12 14 16 0 5 10 15

200
4
100 0

Power/frequency (dB/Hz)
Frequency (kHz)
ACC channel 2

0
b -100
2 -50

-200 -100
0
0 2 4 6 8 10 12 14 16 0 5 10 15

1000

500 4
0

Power/frequency (dB/Hz)
Frequency (kHz)
ACC channel 3

0
-50

c
2
-500
-100
-1000 0
0 2 4 6 8 10 12 14 16 0 5 10 15
Time (s) Time (s)

Figure 12
Acceleration results for a passenger train

So, the strain gauges V-array and sleeper mounted acceleration can accurately
measure static and dynamic forces of wheel-rail system. Combining results from
these two sensors would results to figure 13. The region 1~3 is healthy wheel (or
monitor state), 3~4 stands for maintenance schedule, 4~5 is maintenance priority
and >5 is immediate action. Te presented array and setup could provide condition
monitoring of rail fleet to increase safety and reduce maintenance costs.

‒ 236 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

Figure 13
Converted results in impact factor (acceleration’s unit is m/s2, hence wheel number’s is piece)

Conclusions
In this study, with the help of a field test, the results of two different strain gauge
arrangements and accelerometer installation positions and their differences in
measuring the dynamic force between the wheel and the rail were investigated and
compared. In Iran's Mashhad railway station, the railroad track was instrumented
for a field test, and the results were analyzed in two steps: 1) a light rail vehicle
(drezin) with low weight and speed, and 2) a passenger train with relatively high
axle load and speed. The results are summarized as follows:
"V" and "<" arrays were put to the test as half bridges for strain gauges. When there
are only minor velocity changes, the "V" arrangement is practically velocity-
independent.
The "V"-arrangement has an advantage because of its installation location, which
causes less interference with track maintenance operations.
The "V" array showed lower error compared to the "<" array.
The results must be calibrated for various speed ranges. In contrast to the "<" array,
the "V" array also demonstrated a decrease in velocity dependence with increasing
velocity. Further research is required into the "V" array's dependence on the state
and caliber of the track.
For the acceleration, the installed accelerometer on the sleeper showed lower noise
and dependency to track dynamics, which will require less data preparation.
The reason is the filtering phenomenon of the passing signal through rail pad and
sleeper.
The presented arrays can be used for optimum condition monitoring of wheel-rail
system.
References
[1] S Weston, P. F., Goodman, C. J., Li, P., Goodall, R. M., Ling, C. S., Roberts,
C., Track and Vehicle Condition Monitoring During Normal Operation

‒ 237 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

Using Reduced Sensor Sets, HKIE Transactions, Special Issue on Railway

Development in the 21' Century, March, 2006, pp. 47-53
[2] S Ho, S. L. and Lee, K. K. and Lee, K. Y. and Tam, H. Y. and Chung, W. H.
and Liu, S. Y. and Yip, C. M. and Ho, T. K., A comprehensive condition
monitoring of modern railway. In: Proceedings of The Institution of
Engineering and Technology International Conference on Railway
Condition Monitoring, 2006, 29‐30 November 2006, Birmingham
[3] H Lee, K. Y., Lee, K. K., Ho, S. L., Exploration of Using FBG Sensor for
Axle Counter in Railway Engineering, WSEAS Transactions on Systems,
Issue 6, Vol. 3, August, 2004, pp. 2440-2447
[4] Keith Bladon & Philip Beck., monitoring rolling stock wheel condition.
Teknis Electronics, IRSE - Adelaide - July 1996
[5] Shi, Z., Wang, K., Zhang, D., Chen, Z., Zhai, G., & Huang, D. (2019)
Experimental investigation on dynamic behavior of heavy-haul railway track
induced by heavy axle load. Transport, 34(3), 351-362
[6] Manicka Dhanasekar, Hossein Askarinejad., Determining wheel–rail vertical
contact force at insulated rail joints. CRC for Rail Innovation, 23 May 2011
[7] Alireza Alemi, Francesco Corman, Yusong Pang, Gabriel Lodewijks.,
Reconstruction of an informative railway wheel defect signal from wheel–
rail contact signals measured by multiple wayside sensors. Journal of Rail
and Rapid Transit, July 2018
[8] Gabriel Krummenacher, Cheng Soon Ong, Stefan Koller, Seijin Kobayashi,
Joachim M. Buhmann, Senior Member., Wheel Defect Detection With
Machine Learning. IEEE transaction on intelligent transportation system,
VOL. 19, NO. 4, APRIL 2018
[9] M. Bruner, M. Catena, D. Cortis, G. Malavasi, S. Rossi, M. Testa.,
Estimation of the wheel-rail lateral contact force through the analysis of the
rail web bending strains. Elsevier, Measurement 99 (2017) 23-35
[10] Mikael Palo, Håkan Schunnesson, Uday Kumar., Condition monitoring of
rolling stock using wheel/rail forces. The Ninth International Conference on
Condition Monitoring and Machinery Failure Prevention Technologies,
2012
[11] Georges Kouroussis, Christophe Caucheteur, Damien Kinet, Georgios
Alexandrou, Olivier Verlinden, Véronique Moeyaert., Review of Trackside
Monitoring Solutions. From Strain Gages to Optical Fibre Sensors, 2015, 15,
20115-20139
[12] Mosleh , Pedro Alves Costa, Rui Calc¸ada., A new strategy to estimate static
loads for the dynamic weighing in motion of railway vehicles. Proc IMechE
Part F: J Rail and Rapid Transit, 24 February 2019

‒ 238 ‒
Acta Polytechnica Hungarica Vol. 21, No. 1, 2024

[13] Mosleh A, Montenegro PA, Costa PA, Calçada R. Railway vehicle wheel
flat detection with multiple records using spectral kurtosis analysis. Applied
Sciences. 2021 Apr 28;11(9):4002
[14] Avsievich A, Avsievich V, Avsievich N, Ovchinnikov D, Ivaschenko A.
Railway Track Stress–Strain Analysis Using High-Precision
Accelerometers. Applied Sciences. 2021 Dec 14;11(24):11908
[15] Blanc J, Khairallah D, Ramirez D, Chupin O, Pouget S, Ta QA, Duval A,
Hornych P, Benoist S. Monitoring of railway structures with bituminous and
granular sub-layers: Assessment after four years of use. Construction and
Building Materials. 2022 Jun 20;336:127515
[16] Edwards JR, Mechitov KA, Germoglio Barbosa I, de O. Lima A, Spencer Jr
BF, Tutumluer E, Dersch MS. A Roadmap for Sustainable Smart Track—
Wireless Continuous Monitoring of Railway Track Condition.
Sustainability. 2021 Jul 3;13(13):7456
[17] Menghini A, Leander J, Castiglioni CA. A local response function approach
for the stress investigation of a centenarian steel railway bridge. Engineering
Structures. 2023 Jul 1;286:116116
[18] Gadhave R, Vyas NS. Rail-wheel contact forces and track irregularity
estimation from on-board accelerometer data. Vehicle System Dynamics.
2022 Jun 3;60(6):2145-66
[19] Silva R, Guedes A, Ribeiro D, Vale C, Meixedo A, Mosleh A, Montenegro
P. Early Identification of Unbalanced Freight Traffic Loads Based on
Wayside Monitoring and Artificial Intelligence. Sensors. 2023 Jan
31;23(3):1544
[20] Salehi M, Bagherzadeh SA, Fakhari M. Experimental detection of train
wheel defects using wayside vibration signal processing. Structural Health
Monitoring. 2023:14759217221149614
[21] Nieminen V, Tuohineva A, Autio M. Wheel load reconstruction using strain
gauge measurements on the bogie frame for strain prediction and fatigue
assessment. International Journal of Fatigue. 2023 Jan 20:107533
[22] Jing G, Siahkouhi M, Qian K, Wang S. Development of a field condition
monitoring system in high speed railway turnout. Measurement. 2021 Feb
1;169:108358
[23] van Schalkwyk MH, Gräbe PJ. Condition monitoring of train wheels using a
cost-effective smart rail pad. Engineering Research Express. 2022 Sep
5;4(3):035045
[24] Sol-Sánchez M, Castillo-Mingorance JM, Moreno-Navarro F, Rubio-Gámez
MC. Smart rail pads for the continuous monitoring of sensored railway
tracks: Sensors analysis. Automation in Construction. 2021 Dec
1;132:103950

‒ 239 ‒
M. Kazemian et al. Optimum Train Weighing in Motion using Inertial Sensors

[25] Pillai N, Shih JY, Roberts C. Sensor Selection and Placement for Track
Switch Condition Monitoring through Validated Structural Digital Twin
Models of Train–Track Interactions. Engineering Proceedings. 2021 Nov
1;10(1):49
[26] Milne D, Masoudi A, Ferro E, Watson G, Le Pen L. An analysis of railway
track behaviour based on distributed optical fibre acoustic sensing.
Mechanical Systems and Signal Processing. 2020 Aug 1;142:106769
[27] Pintão B, Mosleh A, Vale C, Montenegro P, Costa P. Development and
validation of a weigh-in-motion methodology for railway tracks. Sensors.
2022 Mar 3;22(5):1976
[28] Deepthi TM, Saravanan U, Meher Prasad A. Algorithms to determine wheel
loads and speed of trains using strains measured on bridge girders. Structural
Control and Health Monitoring. 2019 Jan;26(1):e2282
[29] Khairallah D, Chupin O, Blanc J, Hornych P, Piau JM, Ramirez Cardona D,
Ducreau A, Savin F. Monitoring and modeling railway structures on high-
speed lines with asphalt concrete underlay: A study on the Bretagne–Pays de
la loire line. Transportation Research Record. 2020 Dec;2674(12):600-7
[30] Castillo-Mingorance JM, Sol-Sánchez M, Moreno-Navarro F, Rubio-Gámez
MC. A critical review of sensors for the continuous monitoring of smart and
sustainable railway infrastructures. Sustainability. 2020 Nov 12;12(22):9428
[31] Du C, Dutta S, Kurup P, Yu T, Wang X. A review of railway infrastructure
monitoring using fiber optic sensors. Sensors and Actuators A: Physical.
2020 Mar 1;303:111728
[32] Kovalchuk V, Sysyn M, Gerber U, Nabochenko O, Zarour J, Dehne S.
Experimental investigation of the influence of train velocity and travel
direction on the dynamic behavior of stiff common crossings. Facta
Universitatis, Series: Mechanical Engineering. 2019 Nov 29;17(3):345-56
[33] Szalai S, Eller B, Juhász E, Movahedi MR, Németh A, Harrach D, Baranyai
G, Fischer S. Investigation of deformations of ballasted railway track during
collapse using the Digital Image Correlation Method (DICM) Reports in
Mechanical Engineering. 2022 Mar 16;3(1):168-91
[34] Sysyn M, Nabochenko O, Kovalchuk V, Gruen D, Pentsak A. Improvement
of inspection system for common crossings by track side monitoring and
prognostics. Struct. Monit. Maint. 2019 Sep;6(3):219-35
[35] Cohen L. (1995) "Time-Frequency Analysis: Theory and Applications"
Prentice-Hall, Inc., Upper Saddle River, NJ, USA

‒ 240 ‒