International Journal of Transportation Science and Technology 14 (2024) 110–119

Contents lists available at ScienceDirect

International Journal of Transportation

Science and Technology
journal homepage: www.elsevier.com/locate/ijtst

Application of noise-cancelling and smoothing techniques in

road pavement vibration monitoring data
Amir Shtayat a,⇑, Sara Moridpour b, Berthold Best c, Hussein Daoud d
a
Department of Civil Engineering, Faculty of Engineering, Jadara University, Irbid, Jordan
b
Civil and Infrastructure Engineering Discipline, RMIT University, Melbourne, Australia
c
Faculty of Civil Engineering, Nuremberg Institute of Technology, Nuremberg, Germany
d
Department of Mechanical Engineering, National University College of Technology, Amman, Jordan

a r t i c l e i n f o a b s t r a c t

Article history: Road pavement surfaces need routine and regular monitoring and inspection to keep the
Received 7 May 2022 surface layers in high-quality condition. However, the population growth and the increases
Received 9 March 2023 in the number of vehicles and the length of road networks worldwide have required
Accepted 9 April 2023
researchers to identify appropriate and accurate road pavement monitoring techniques.
Available online 20 April 2023
The vibration-based technique is one of the effective techniques used to measure the con-
dition of pavement degradation and the level of pavement roughness. The consistency of
Keywords:
pavement vibration data is directly proportional to the intensity of surface roughness.
Noise-cancelling
Smoothing
Intense fluctuations in vibration signals indicate possible defects at certain points of road
Road pavement pavement. However, vibration signals typically need a series of pre-processing techniques
Degradation such as filtering, smoothing, segmentation, and labelling before being used in advanced
Vibration processing and analyses. This research reports the use of noise-cancelling and data-
smoothing techniques, including high pass filter, moving average method, median,
Savitzky-Golay filter, and extracting peak envelope method, to enhance raw vibration sig-
nals for further processing and classification. The results show significant variations in the
impact of noise-cancelling and data-smoothing techniques on raw pavement vibration sig-
nals. According to the results, the high pass filter is a more accurate noise-cancelling and
data smoothing technique on road pavement vibration data compared to other data filter-
ing and data smoothing methods.
Ó 2024 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V.
on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BY-
NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Road networks are the backbone of a infrastructure sector. Therefore, the quality of roads reflects the quality of the ser-
vices provided by governments. In addition, the efficiency of road networks represents comfort and luxury levels for road
users. Governments and transport agencies play significant roles in maintaining road networks in excellent conditions. How-
ever, transport agencies face many challenges that lead to the loss of control of ideal road pavement conditions. These chal-
lenges include the increase in the number of vehicles, weather conditions, and the quality of road pavement materials.

Peer review under responsibility of Tongji University.

⇑ Corresponding author.
E-mail address: a.shtayat@jadara.edu.jo (A. Shtayat).

https://doi.org/10.1016/j.ijtst.2023.04.002
2046-0430/Ó 2024 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V. on behalf of KeAi Communications Co., Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

Many efforts have been implemented to select advanced equipment and professional inspection procedures to identify
accurate pavement monitoring systems. The results of monitoring are used to predict the future performance of road pave-
ments. However, even the data extracted from very accurate pavement monitoring devices require smoothing and filtering
of the data to enable further analyses.

2. Background

Many researchers agree that the vibration-based method is one of the most accurate and cost-efficient techniques in
dynamic road pavement monitoring (Justusson and Huang, 1981; Seraj et al., 2015; Yu et al., 2020; Ye et al., 2021). The
use of vibration-based methods in pavement monitoring is an accurate method that provides clear information about riding
comfort and road pavement conditions (Shtayat et al., 2022). Moreover, vibration-based methods can indicate the type,
severity, and quantity of defects and degradations on pavement surfaces (Shtayat et al., 2020; Shtayat et al., 2021;
Shtayat et al., 2022). When the fluctuations of vibration peaks increase, the peaks indicate potential distress types on the
road pavement surface. Therefore, the sharpness of signal fluctuations is affected by the type and severity of road pavement
distress (Han et al., 2021; Li et al., 2021; Ma et al., 2021).
The vibration-based method has been widely used in monitoring and forecasting road pavement conditions, and different
equipment has been used to dynamically measure the vibration data during vehicle movement over road sections. The accu-
racy of vibration signals differs according to the quality of the devices used, road type, travel speed, vehicle type, traffic vol-
ume, and weather conditions (Yang and Peter, 2003; Branhof et al., 2005). Monitoring devices such as smartphone
applications and accelerometer sensors are the most common vibration monitoring devices with acceptable ranges of mon-
itoring accuracy (Shtayat et al., 2020; Xu et al., 2018). Many researchers recommend using accelerometer sensors in pave-
ment monitoring systems (Yu et al., 2020; Shtayat et al., 2021; Zeng et al., 2019). Acceleration sensors have used single,
double, and triaxial channels to measure vibrations during vehicle movement over pavement surfaces (Kee et al., 2020).
Single-channel means that the sensor collects the vibration data on one axis, usually vertical (the z-axis). In dual channel
sensors, the x-z or y-z axes are used to obtain the vibrations from the vehicle chassis during movement over the road pave-
ment. The vibration data from three axes is collected from the road surfaces in a triaxial accelerometer sensor. These vibra-
tion data represent the three-dimensional movements of the vehicle chassis during the movement over a pavement surface.
Several studies have shown that using triaxial accelerometer sensors increases the chance of successfully detecting the
actual distress locations from vibration signals due to the variance and fluctuations in the signals from the three axes
(Kraft and Pallotta, 2000; Khogali et al., 2002; Zhang et al., 2018). However, different factors significantly impact the accuracy
of vibration signals and the overall pavement monitoring system, such as sampling rate, device frequency, sensor type, and
mounting position (Bridgelall, 2015). Frequency is the number of waves during one unit of time, and in vibration-based
pavement monitoring, it represents the number of pavement vibrations per unit of time. The unit of frequency is the Hertz,
which means one wave cycle per second. Moreover, an increase in frequency values means the sensor can collect more vibra-
tion signal data. In many previous studies, researchers have used low-frequency sensors to evaluate the condition of road
pavements. The most common frequency ranges are up to 100 Hz in most pavement monitoring studies using vibration-
based methods, especially smartphone applications (Seraj et al., 2015; Hazan et al., 2010). Shtayat et al. (Shtayat et al.,
2022) used a smartphone application named SensorLog to measure the vibration data at a frequency of 30 Hz on a local road
in Melbourne. The smartphone was fixed on the handlebars of a motorcycle which was ridden over the pavement surface at
speeds of 5 km/h, 10 km/h, 15 km/h, and 20 km/h, and Shtayat et al. (Shtayat et al., 2021) used the smartphone software with
50 Hz frequency and a tri-axle accelerometer with 6 000 Hz frequency and different vehicle types (bicycle, sedan car) as a
test vehicle to monitor pavement conditions using a vibration-based method. The results indicated that using a passenger car
and increasing the frequency of vibration data enhance the accuracy of the vibration signals and reflect the actual pavement
health status. On the other hand, some researchers have used sensors with high frequencies in their monitoring techniques.
For instance, Zeng et al. (Zeng et al., 2019) used a vibration-based method to evaluate the pavement conditions on a highway
in China. They used an accelerometer sensor with a maximum sampling rate of 2 000 Hz. They reported that this frequency
value provides an advantage in accurately extracting features (potential distresses) from acceleration data using wavelet
transform and short-time energy processing. However, the increase in the frequency value needs more data processing to
filter and smooth the signals to be ready for further processing.
Noise-cancelling techniques are methods used to find and remove undesirable data from raw datasets (Liao et al., 2014).
Extracting and elimiating meaningless data from vibration signals is a crucial technique commonly used to improve the
signal data before advanced processing and analyses (Liao et al., 2014). Filtering is a pre-processing technique used to cancel
the noise from any raw signal data. It is followed by data smoothing techniques used to enhance the streamline of the sig-
nals. In pavement monitoring systems, vibration-based methods normally need to apply different levels of filtering to elim-
inate the noises in vibration signals. These smoothing and pre-processing techniques assisted the researchers in increasing
the probability of anomaly detection to about 95% (Seraj et al., 2015; Liao et al., 2014).
Several types of smoothing techniques are used in the literature to improve the quality of noisy data, including the mov-
ing average method, extracted average differences, the random method, the median method, the Savitzky-Golay method,
extracting peak envelope, and resampling (Gui et al., 2018; Ren et al., 2021). Zhang et al. (Zhang et al., 2018) and Wang
et al. (Wang et al., 2020) used an integrated moving average approach to predict the road pavement condition and riding
comfort according to vibration data from vehicle tyres. They focused on understanding driving performance over a pavement
111
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

section by applying a significant moving average filter. Dong et al. (Dong et al., 2018) and Zeng et al. (Zeng et al., 2019) used a
moving average filter to smooth pavement vibration data from sensors. Zhang et al. (Zhang et al., 2018) used two common
noise-cancelling techniques, high pass filter and the Hampel filter, in road pavement monitoring. Seraj et al. (Seraj et al.,
2015) detected road pavement anomalies using vibration data collected from a smartphone application. The low pass filter-
ing technique was used to enhance the smoothing of the data and prepare the row dataset for extracting features that pro-
vide a clear idea of the road health status. On the other hand, Savitsky-Golay filters and median filters are frequently used for
smoothing structural vibration data (Gudmarsson et al., 2015; Wu et al., 2020). These smoothing methods provide an initial
indication of the real signal fluctuations that represent the road pavement condition.
Accurate noise cancelling and data smoothing have significant impacts in preparing the road pavement vibration data and
accurately measuring the performance of road pavement. However, a limited number of studies have focused on the effec-
tive use of noise-cancelling techniques to filter the vibration data collected from road pavements, and many researchers have
used manual reorientation and resampling techniques to minimize the noise from raw vibration data. Despite the limited
studies on data smoothing techniques, the filtering technique is necessary even after using advanced monitoring equipment.

3. Research objectives

Vibration-based techniques in dynamic pavement monitoring systems require applying an accurate set of noise-
cancelling, filtering, and data smoothing techniques to remove undesirable thresholds from the vibration signals. However,
some noise-cancelling/filtering techniques may negatively impact the quality of the data, remove the desirable thresholds,
and over-smooth the data. In this paper, a pavement monitoring system was conducted using an accelerometer sensor to
collect vibration data during sedan car movement at a speed of 20 km/h along a local road in metropolitan Melbourne. This
research focuses on enhancing the quality of the collected vibration data using different sets of noise-cancelling, filtering,
and data smoothing techniques to smooth the vibration data and make the pavement vibration monitoring system more reli-
able in matching the vibration data with manual inspections/observations. Five standard noise-cancelling and data smooth-
ing techniques are applied to the collected vibration monitoring data to identify the most effective technique that is more
accurate and can be regularly used in vibration-based road pavement monitoring systems, see Fig. 1.
This paper is structured as follows. The information on data collection, data processing, and data analysis is presented in
Sections 2. This is followed by a discussion of the results in Section 3. The conclusions and recommendations are presented in
Section 4.

4. Data

This section presents the information on the data collection procedure, including the monitoring devices, the site location,
setting up the devices, and the method of collecting the vibration data.
Data collection took place in Spring Street, Fitzroy, Melbourne, Australia. Spring Street is located in the north-east suburb
of the Melbourne Central Business District (CBD). It is a two-way local street with a speed limit of 40 km/h and has on-street
parking on both sides. The street has different distress types and severities based on a visual inspection, such as cracking,
patching, ravelling, and potholes. Fig. 2 shows the location of this street as the study site.

Fig. 1. The filtering techniques used in the study.

Fig. 2. The location of the Spring Street (Google maps).

112
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

A triaxial accelerometer sensor model PCB356B18 was used to collect vibration data while a car (a 2009 Toyota Prius
Hybrid) was driven over the pavement section at a speed of 20 km/h. A road pavement monitoring system was used in this
research to collect high-frequency vibration data with a frequency of more than 6 000 Hz. A video camera was also used to
record the pavement condition during vehicle movements and identify and match the locations of distresses with vibrations
and the location data from the Global Positioning System (GPS). This study focused on using different noise cancelling tech-
niques to prepare vibration data for further processing to accurately monitor the pavement condition status.

4.1. Data processing and filtering

After collecting the vibration data, a high pass filter was applied to noise to cancel the raw vibration signals. In addition,
four different data smoothing techniques were applied to raw and filtered vibration data. Moving average, median, Savitzky-
Golay filter, and envelope smoothing were considered to improve the collected vibration data. Each smoothing technique has
specific features, formulas, advantages, and disadvantages that differ from other techniques. Details on the procedure and
methodology of the smoothing techniques are presented separately in this section for each technique.
Data filtering is a pre-processing technique to eliminate noise and undesirable data from raw data. Researchers have
noted that some factors impact the consistency and accuracy of vibration data during data collection, including ongoing traf-
fic, braking, acceleration, deceleration, and traveling speed (Shtayat et al., 2021; Xie et al., 2013; Al-Masaeid and Shtayat,
2016; Lekshmipathy et al., 2020; Alatoom and Obaidat, 2021). Therefore, data filtering is needed to enhance the quality
of the vibration data by removing the undesirable frequency data from the raw vibration data. The most popular data filter-
ing techniques in vibration data include low pass, high pass, elliptic, Butterworth, Chebyshev, and Bessel filters. Each filtering
technique has specific features, and can work with different data types. Hence, the selection of an appropriate filtering tech-
nique is a crucial step in the evaluation of road pavement conditions.

4.1.1. High pass filter

The high pass filter is a noise-cancelling technique that reduces and removes low-frequency signals within specific
thresholds. There are three main types of high pass filters, i.e., the Butterworth, Gaussian, and ideal filters, and each type
has a specific procedure to be followed.
In this paper, four high-pass filter cut-off frequencies, i.e., 100 Hz, 200 Hz, 400 Hz, and 800 Hz, were used to remove the
noise from the vibration signals and smooth the pavement vibration data for more consistent results. The selected frequen-
cies were chosen after deep testing with a wide range of frequencies, and the research found that these ranges provided a
better indication of the condition of pavement defects with compelling and logical results that can easily reflect the severity
and types of pavement defects. Fig. 3(a) presents the raw pavement vibration data for the selected high-pass frequencies. The
figure shows a selected window from massive vibration datasets. This window contains potential distress due to the fluctu-
ation of vibration signals. According to Shtayat et al. (Shtayat et al., 2021), there is a linear relationship between the signals’
values and the types and severity of road pavement degradation. When the fluctuations of pavement vibrations increase, the
signals indicate highly severe pavement distress. At 100 Hz and 200 Hz frequencies, the data show more consistency in iden-
tifying normal signals with potential distress, while at 400 Hz and 800 Hz, the results show less consistency due to the high
smoothing process. The frequency of 200 Hz provides a significant indication of normal signals from stations at 31.2–31.35
meters and at 31.6–31.8 meters, which indicates that there is no distress (see Fig. 3(b)) after comparing with the recorded
video and visual inspection.

Fig. 3. Noise-cancelling and data smoothing using different frequencies of high-pass filters.

113
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

The majority of noise was observed at frequencies of less than 100 Hz, and the frequencies between 100 Hz and 400 Hz
were acceptable. Therefore, after applying the cut-off frequencies of 100 Hz, all noise and undesirable signals were removed.
On the other hand, with the application of cut-off frequencies of 400 Hz or more, both required and undesired signals were
removed, which may directly impact the ability to detect vibration signals and detect road pavement distress. At the same
time, the fluctuations at the road segment of 3.4–31.6 meters indicate the existence of potential distress. This potential dis-
tress was confirmed as an alligator crack after referring to the video recordings, which were simultaneously recorded while
collecting the pavement vibration data. Moreover, road inspections confirmed the type, severity, and physical characteristics
of the distress. In this research paper, the same data sample was used and tested using the proposed filtering and noise-
canceling techniques as seen in the following figures.

4.1.2. Moving average method

The moving average method is a low-pass filter used to find low-frequency vibration data beneath high-frequency noise
(Khogali et al., 2002). The moving average method mainly depends on smoothing the data to make it more streamlined and is
commonly used in time domain datasets. The basic structure of the moving average method is as follows:

1 X P
Z ðnÞ ¼ xðn  jÞ; ð1Þ
P þ 1 j¼0

where, Z (n) is smoothed vibration data, x (n) is a raw version of Z (n), and P is the moving average window size.
In this study, the moving average was applied to two cases, i.e., the raw vibration data and the vibration data filtered
using the high-pass filter. Moreover, three window sizes of 5, 10, and 15 average points were selected to determine the
appropriate average points. Fig. 4(a) presents the raw vibration data after applying the three average window sizes, while
in Fig. 4(b), the same average window sizes are applied to the filtered data to enhance and smooth the vibration signals
and determine the more consistent phase to be used in further studies.
Fig. 4(a) shows the raw pavement vibration data for a short segment of a local street in Melbourne. The data was collected
from the road, and advanced processing techniques were adopted to make the data more smooth and noiseless. Therefore,
the selected three average points were used to determine the average size with more realistic and consistent data. It is
clearly seen that there are various impacts for each average point size on the raw data. The more average points mean more
spikes and peaks are missing, leading to less consistency and accuracy of the pavement vibration data. Hence, the most
appropriate average point size is five average points, as it is more streamlined and consistent compared to the other point
sizes.
As shown in Fig. 4(b), applying the moving average filtering technique to the filtered data is more consistent with the
observed pavement condition data compared to applying the moving average to the raw data. According to the results, a
noise-cancelling technique should be applied to the raw data before any pre-processing occurs. A cut-off frequency is a cru-
cial technique that needs to be applied before smoothing the vibration signals for multiple purposes. Therefore, this paper
implemented the moving average smoothing technique on filtered data. After applying this filter, the beginning and ending
boundaries of each pavement defect became clear to identify the length of distress, and the vibration values became more
reasonable than they were before the filter. The results showed that the filtered and smoothed data accurately and consis-
tently contributed to the estimation of the location and severity of road pavement distress.

Fig. 4. Noise-cancelling and data smoothing using moving average smoothing technique.

114
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

Fig. 5. Noise-cancelling and data smoothing using a median filter technique.

4.1.3. Median method

The median filter is a filtering method used to smooth and correct signals from time-domain datasets, and is widely used
in image processing. At this stage, a third-order filter is applied to the road pavement vibration signals to identify the median
datasets close to each other in the median filter.
Fig. 5(a) shows the use of the median filter on raw vibration data to determine the effectiveness of smoothing the data
and enhancing signal fluctuations. It is clear that significant contributions are made at limited points by the median filter to
improve the shape of the signals and make them smoother and more realistic. However, the majority of the filter values fol-
low the original shape of the signals. In Fig. 5(b), the case of using the median filter to smooth filtered data is similar to the
raw data. Significant impacts are found at specific points to be smoothed, while the majority of the signal points match the
median filter values. Therefore, the use of the median smoothing technique is more consistent with the observed pavement
condition data at specific and limited points on vibration signals. According to Figs. 5(a) and 5(b), the points corrected or
smoothed by the median filter were originally sharp. Therefore, this filter is useful for the correction and smoothing of data
sharpness representing high-severity pavement distress. Hence, the ability to detect the location and severity of pavement
distress becomes more easier and significant.

4.1.4. Savitzky-Golay filter

A Savitzky-Golay filter is a smoothing filter used to enhance the shape of the time series signals without changing the
signal tendency. It is also used to increase the accuracy and precision of vibration signals. The filter’s purpose is to prioritize
the average point and the points close to the average value. A Savitzky-Golay filter is similar to the moving average method
in smoothing vibration signals. Different window sizes can be used to determine the most accurate average point sizes. In
addition, this filter method attempts to fit a polynomial equation at different orders.
In this paper, the polynomial equation’s 2nd, 3rd, and 4th orders were used to smooth an average of nine-point sizes in
two phases, including the raw vibration signals and filtered vibration signals at 200 Hz of the high pass filter (Figs. 6(a) and
(b)). The figures show the raw and filtered vibration data with three applied polynomial orders. The results show that most of
the smoothed data match the raw signals. The matching data indicates that this filter focuses on major sharps or peaks with
irregular shapes to smooth them. Moreover, according to Figs. 6(a) and (b), the impact of the three polynomial orders on
filtering the data is significant. In summary, this model is suitable to correct and smooth sharp peaks, which reflects a high
level of road pavement degradation.

4.1.5. Extracting peak envelope method

The envelope filter is used to detect the highest and lowest vibration signals. The envelope filter method is useful in con-
trolling the signals which represent possible pavement distress. This method focuses on tracking the highest and lowest
peaks and then using the average for both values. Fig. 7(a) shows the use of the envelope filter method to smooth raw vibra-
tion signals. The figure also indicates significant fluctuations on the upper and lower borders due to fluctuations in the raw
data. The mean value markedly fluctuates with the raw signals.
Fig. 7(b) shows slight fluctuations due to a high pass filter at 200 Hz and that the high and low envelope filter rates are
consistent at the top and bottom peaks. The mean value also appears like a constant line at the middle of the vibration sig-
nals. After applying to the filtered data, the upper and lower borders of the envelope filter were an almost constant along
with the filtered data lines. In contrast, there was a clear fluctuation in envelop filter borders with the shape of raw data lines
due to the noise in signals.
115
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

Fig.6. Noise-cancelling and data smoothing using Savitzky-Golay filter technique.

Fig. 7. Noise-cancelling and data smoothing using envelope technique.

5. Discussion

This paper applied the most common noise-cancelling and data smoothing techniques used in previous studies on road
pavement vibration data. A high-pass filter was used as an effective noise-cancelling technique to enhance the vibration sig-
nals. It was also used to prepare the signals for the application of further smoothing techniques and pre-processing methods.
Four different data smoothing methods were also tested to identify the most accurate method, which reduces errors and
smooths the shape of the signal peaks. The moving average method, median filter, Savitzky-Golay filter, and extracting peak
envelope method were applied on raw pavement vibration data and to filtered pavement vibration data at 200 Hz of a high-
pass filter. Tables 1 and 2 show the smoothing of the raw data and the high-pass filtered data in a window (distance in the
range of 31.2 m31.8 m). The tables present the mean, Standard Deviation (SD), Root Mean Square Error (RMSE), and Pdist
factor.
The RMSE for every two data points was measured, and showed significant smoothing using moving average, median, and
Savitzky-Golay filters at specific locations along with the raw and filtered data. At the same time, Pdist is a factor used to
measure the distance between two data points. The Pdist factor with values close to zero indicates a significant impact
on data smoothing. More clearly, this metric focuses on measuring the change in locations for the same point before and

116
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

Table 1
Smoothing the raw data in the shown window (distance in the range of 31.2 m-31.8 m).

Data Mean SD RMSE (smooth, raw) Pdist (smooth, raw)

Raw 0.95 0.69 - -
Moving average 0.95 0.69 0.027 0 0.000 26
Median filter 0.95 0.69 0.007 7 2.1410–5
Savitzky–Golay filter 0.95 0.69 0.012 5.110–5
Envelope filter 0.94 0.67 0.170 0.009 9

Table 2
Smoothing the high pass filtered data in the shown window (distance in the range of 31.2 m-31.8 m).

Data Mean SD RMSE (smooth, raw) Pdist (smooth, raw)

Raw 0.001 1 0.130 - -
Moving average 0.001 1 0.110 0.026 0.009 9
Median filter 0.001 3 0.130 0.007 4 0.001 6
Savitzky–Golay filter 0.001 1 0.130 0.012 0.004 2
Envelope filter 0.001 3 0.026 0.130 1.000 0

after applying the four selected smoothing methods. In addition, the difference represents the level of smoothing of each
point on vibration signals. In this case, the tables show that the Pdist measures the distances between two data points from
raw data and four smoothing methods, and between high-pass filtered data with four smoothing methods. Therefore,
according to the Pdist values close to zero, three smoothing techniques, i.e., the moving average, median filter, and
Savitzky-Golay filter, significantly smooth the raw and filtered data. Those three filtering methods showed more accuracy
in smoothing the vibration signals compared with the envelope filtering method.
The high pass filter was recommended by many previous researchers to reduce noise and undesirable frequencies in any
raw data. In this paper, four different frequencies , i.e., 100 Hz, 200 Hz, 400 Hz, and 800 Hz, were applied to reduce and cancel
noise from the raw data. All of the frequencies showed consistent noise-cancelling results. However, the most effective fre-
quency with acceptable consistency of the vibration data was a frequency of 200 Hz. The results indicate that applying the
moving average technique (five average point sizes) to smooth the vibration data is more consistent than other smoothing
methods. This technique produces effective smoothing results for raw and filtered vibration signals. In addition, there was no
worthy impact of using the median filter to smooth the whole raw or filtered data; only limited numbers of significant cor-
rections were noted. The results from using the Savitzky-Golay filter showed that this smoothing method focuses on correct-
ing sharp peaks, similar to the median filter. In this filter, three different polynomial orders were applied at 2nd, 3rd, and 4th
orders to determine the most accurate order which correctly smoothes the signals. Attempts were made to select an appro-
priate point size, and, the nine average point size was selected after showing substantial differences in numbers of combined
peaks compared with other lower values. The increase in combining point size leads to missing some actual peaks, while
selecting a small average point size leads to a larger number of peaks and less effective data smoothing. The results showed
no effective differences among the selected orders in smoothing the data. In other words, limited signals were substantially
smoothed while the rest of the vibration signals remained the same. Finally, the envelope filter also presents a different way
of smoothing vibration signals. The filter focused on tracking the highest and lowest peaks to draw boundaries, and the mean
value was calculated for each high and low point. The results revealed that this method makes a lower contribution to
smoothing vibration signal data than other methods. In summary, the high-pass filter substantially smooths vibration signals
collected under different conditions such as different test speeds, test vehicle types, and mounting locations of sensors.
The validation of using these techniques and their ability to enhance the outcomes of detecting and classifying the pave-
ment defects from vibration data depend mainly on investigating the location, type, and severity of each pavement defect.
More simply, the location was identified using the built-in GPS in the camera, so the start edge of each defect could be easily
detected and then the end edge could be identified using the length measured in the field and compared to the length of the
fluctuated vibration data. In addition, the type was identified according to the recorded video. While, the severity was also
detected according to the fluctuations in values and taken images that were extracted as time frames from the recorded
video based on several factors such as frame time, image dimension, coverage area, and the dimension of the captured defect.

6. Conclusions and future research directions

This paper has reported on the impacts of noise-cancelling and data smoothing techniques on vibration data. The vibra-
tion data was collected from an accelerometer sensor fixed on the front dashboard of a sedan car. Different factors might
increase the noise in the raw data, such as acceleration, deceleration, braking, ongoing traffic, and engine noise. Therefore,
the collected vibration data needed noise-cancelling and data smoothing before being used in pavement vibration monitor-
ing. The high-pass filter, average moving, median filter, Savitzky-Golay filter, and envelope filter were adopted to enhance
the raw vibration data.
117
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

In summary, all of the selected noise-cancelling and smoothing methods showed fluctuations in accuracy, although each
smoothing method has a considerable impact in smoothing vibration data. The results confirm that using a high-pass filter as
the first step in data processing to noise-cancel undesirable frequencies from the raw vibration data is useful prior to further
data processing and classification. In this paper, the four selected data smoothing techniques and the high pass filter
improved the quality of pavement vibration signals. The locations of the high fluctuation data (road pavement degradation
points) and normal data (pavement with no damage) were clearly identified in road pavement signals after smoothing and
filtering. Using smoothing and filtering techniques provided detailed information on the exact location, type, and severity of
pavement distress based on vibration signals. Applying the filtered and smoothed vibration signals to the road pavement
degradation prediction using different predictive models (e.g., machine learning techniques) can be a future direction of this
research.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.

References

Alatoom, Y.I., Obaidat, T.I., 2021. Measurement of street pavement roughness in urban areas using smartphone. Int. J. Pavement Res. Technol. 15, 1–18.
Al-Masaeid, H.R., Shtayat, A., 2016. Performance of urban transit in Jordan. Int. J. Eng. Res. Appl. 6 (8), 7–12.
Branhof, R.W., Keller, J.A., Grant, L., Grabill, P., 2005. Application of automated rotor smoothing using continuous vibration measurements. AFPAH 61 (2),
1230.
Bridgelall, R., 2015. Inertial sensor sample rate selection for ride quality measures. J. Infra. Syst. 21 (2), 04014039.
Dong, Z., Ma, X., Shao, X., 2018. Airport pavement responses obtained from wireless sensing network upon digital signal processing. Int. J. Pavement Eng. 19
(5), 381–390.
Gudmarsson, A., Ryden, N., Birgisson, B., 2015. Non-contact excitation of fundamental resonance frequencies of an asphalt concrete specimen. In: Paper AIP
Conf. Proc. 1650, 1401–1408.
Gui, R., Xu, X., Zhang, D., Lin, H., Pu, F., He, L., Cao, M.J.S., 2018. A component decomposition model for 3D laser scanning pavement data based on high-pass
filtering and sparse analysis. J. Sens. 18 (7), 2294.
Han, H., Han, W., Ma, S., Hu, G., 2021. Enhanced processing of low signal-to-noise-ratio dynamic signals from pavement testing. Measurement 182, 109697.
Hazan, A., Verleysen, M., Cottrell, M., Lacaille, J., 2010. Linear smoothing of FRF for aicraft engine vibration monitoring. In: Proceedings of the International
Conference on Noise and vibration Engineering (ISMA 2010), pp. 2857–2868.
Justusson, B.I., 1981. Median filtering: statistical properties. In: Huang, T.S. (Ed.), Two-Dimensional Digital Signal Processing II. Springer, Berlin, Heidelberg,
pp. 161–196.
Kee, S.H., Lee, J.W., Candelaria, M., 2020. Evaluation of delamination in concrete by IE testing using multi-channel elastic wave data. J. Sens. 20 (1), 201.
Khogali, K.A., Odhiambo, J.W., Owino, J.O., 2002. On Smoothing Time Series Data Using A Classical Moving Average Formula. Department of Mathematics,
University of Nairobi.
Kraft, L.G., Pallotta, J., 2000, June. Real-time vibration control using CMAC neural networks with weight smoothing. In: Proceedings of the 2000 American
Control Conference. ACC (IEEE Cat. No. 00CH36334), 6, 3939-3943.
Lekshmipathy, J., Samuel, N.M., Velayudhan, S., 2020. Vibration vs. vision: best approach for automated pavement distress detection. Int. J. Pavement Res.
Technol. 13 (4), 402–410.
Li, F., Sun, X., Zhou, S., Chen, Y., Hao, Z., Yang, Z., 2021. Infrastructure material magnetisation impact assessment of wireless power transfer pavement based
on resonant inductive coupling. IEEE trans Intell Transp Syst. https://doi.org/10.1109/TITS.2021.3099490.
Liao, G., Sakhaeifar, M.S., Heitzman, M., West, R., Waller, B., Wang, S., Ding, Y., 2014. The effects of pavement surface characteristics on tire/pavement noise.
Appl. Acou. 76, 14–23.
Ma, X., Dong, Z., Dong, Y., 2021. Toward asphalt pavement health monitoring with built-in sensors: a novel application to real-time modulus evaluation.
IEEE trans Intell Transp Syst. https://doi.org/10.1109/TITS.2021.3102252.
Ren, Z., Yu, W., Dai, W., Li, Z., 2021. Performance assessment of bridge modal frequency identification using high-rate GNSS. CSNC., 80–89
Seraj, F., Zhang, K., Turkes, O., Meratnia, N., Havinga, P.J., 2015. A smartphone based method to enhance road pavement anomaly detection by analysing the
driver behavior. UbiComp/ISWC, 1169–1177.
Shtayat, A., Moridpour, S., Best, B., Shroff, A., Raol, D., 2020. A review of monitoring systems of pavement condition in paved and unpaved roads. LTWA 7,
629–638.
Shtayat, A., Moridpour, S., Best, B., 2021. Using e-bikes and private cars in dynamic road pavement monitoring. Int. J. Transp. Sci. Technol. 11 (1), 132–143.
Shtayat, A., Moridpour, S., Best, B., Rumi, S., 2022. An overview of pavement degradation prediction models. J. Adv Transp. 2022, 7783588.
Shtayat, A., Moridpour, S., Best, B., Abuhassan, M., 2022. Using supervised machine learning algorithms in pavement degradation monitoring. Int. J. Transp.
Sci. Technol. 12, 628–639. https://doi.org/10.1016/j.ijtst.2022.10.001.
Wang, S., Al-Qadi, I.L., Cao, Q., 2020. Factors impacting monitoring asphalt pavement density by ground penetrating radar. NDT & E Int. 115, 102296.
Wu, C., Wang, Z., Hu, S., Lepine, J., Na, X., Ainalis, D., Stettler, M., 2020. An automated machine-learning approach for road pothole detection using
smartphone sensor data. J. Sens. 20 (19), 5564.
Xie, X., Huang, J., Liang, Z., 2013. Vibration reduction for flexible systems by command smoothing. Mech Syst Signal Process 39 (1–2), 461–470.
Xu, G., Chen, F., Wu, G., Li, X., 2018. Active solution of homography for pavement crack recovery with four laser lines. Sci 8 (1), 1–10.
Yang, W.X., Peter, W.T., 2003. Development of an advanced noise reduction method for vibration analysis based on singular value decomposition. Ndt & E
Int. 36 (6), 419–432.
Ye, Z., Yan, G., Wei, Y., Zhou, B., Li, N., Shen, S., Wang, L., 2021. Real-time and efficient traffic information acquisition via pavement vibration IoT monitoring
system. J. Sens. 21 (8), 2679.
Yu, Y., Guan, H., Li, D., Zhang, Y., Jin, S., Yu, C., 2020. CCapFPN: a context-augmented capsule feature pyramid network for pavement crack detection. IEEE
trans Intell Transp Syst 23, 3324–3335. https://doi.org/10.1109/TITS.2020.3035663.
Zeng, M., Zhao, H., Wu, D., Chen, H., Cai, J., 2019. A vibration-based traffic monitoring system using distributed optical sensing technolog. J. Test. Eval. 48 (3),
1799–1813.

118
A. Shtayat, S. Moridpour, B. Best et al. International Journal of Transportation Science and Technology 14 (2024) 110–119

Zhang, Y., Song, W., Karimi, M., Chi, C., Kudreyko, A., 2018. Fractional autoregressive integrated moving average and finite-element modal: the forecast of
tire vibration trend. IEEE Access 6, 40137–40142.

Amir Shtayat has been working as Assistant Professor at Jadara University, Jordan. He completed his BE degree in civil engineering in 2013. He also received
an ME degree in transportation engineering in 2015. He also has a PhD in Civil engineering from RMIT University, Australia. His main research interests are
public transport, pavement monitoring, pavement evaluation, and prediction modelling.

Sara Moridpour received her BE degree in civil engineering, ME degree and PhD in traffic and transportation engineering. She has been working as lecturer/
senior lecturer at RMIT University since 2010. Sara has more than 100 publications. She has won seven national/international awards in traffic/trans-
portation engineering and has been selected as a finalist for four scientific awards. Her main research interests are driving behaviour modelling and
analysis, transport infrastructure maintenance planning and management, road freight management and transport network modelling and simulation.

Berthold Best is a professor of transport engineering, at Nuremberg Institute of Technology, Germany. His main research interests are road asset man-
agement, sustainability of roads, financial requirements for road asset management, key performance indicators for road efficiency and road management,
and excavations for energy, water and telecommunication network.

Hussein Daoud: PhD in Mechanical Engineering from American University of Beirut.

119