Detecting defects in noncommissioned oil and gas pipelines using ground robotic vehicle

with thermal image processing

Abstract
Leak detection is one of the most crucial aspects of oil and gas pipeline operation. These leaks are
a result of defects that may have been present before laying the pipes. Early defect detection helps
a lot to save both service and maintenance costs, as well as prevent possible operational shutdowns
and damages. Several methods are used for leak detection and defect identification, but most often,
the approaches are usually from the external structural view of the pipes. This paper presents an
internal pipe approach to defect detection through the application of a robot equipped with a
thermal imaging system. The robot is developed around a Raspberry Pi B+ ARM11-based
microcomputer with a Graphic Processing Unit (GPU) to support a video/photographic camera.
Image capture points were obtained by the location of object distance and image frame sizing.
Acquired images are processed with a modified image processing algorithm using Gaussian and
Sobel filters with morphological transformation in OpenCV. Results show good performance of
the robot’s movement and image capture points at distances less than 80cm. Processed images also
indicate areas of defects within the pipe internals.
Keywords:
Image capture point, Infrared image, Leak detection, Morphological transformation, Pipe
internal, Sobel filter.
1. Introduction
Pipelines are laid on a long distance passing through different terrene and are liable to experience
leakages due to different factors. These factors which include welding joint errors and production
faults that may occur during manufacture should be adequately determined before pipe installation
processes. There is a need, therefore, to inspect these pipes before they are commissioned to the
lines in case of any possible leak-prone section or susceptible area of leakage after production or
during joint welding. However, an inspection of these pipes is usually done on the surface which
does not reveal all leakage indications. External leakage detection is also not a preventive measure
and could lead to damages with high-cost implications and threats to life.
Tremendous damage may occur if a pipe is put to use with cracks in its internals and can easily be
worsened by causes describe in [1] as ageing of pipelines, improper fixing of pipelines, excavation,
surface load, heavy traffic, and root of trees. This will result in spillage and other environmental
disasters including possible explosions. For this reason, it is very important to also investigate the
pipes internally before they are used. Such internal investigations are difficult to achieve manually
and may require the use of mobile systems that can navigate through the pipe. Modern mobile
robotic systems have been applied for similar roles such as looking for minerals or explosives in
caves [2,3]; and are therefore suitable for the inspection of pipe internals.
Mobile robots can move around in their environment and are not fixed to one physical
location. Automated inspection of the internal surface of a pipe is necessary before
commissioning, to check for leaks that may arise from pipe manufacturer’s production error,
corrosion, and bad welding of joints during construction. This work explores the use of pipe
internal image capture and processing in a mobile robotic system for defective point detection
before application.

1
2. State of the Art
Several methods exist and are employed by the oil and gas authorities to detect the location of
leaks in a pipeline [4,5,6]. These detections are however done from the pipe externals with several
limitations [7]. There is the problem of accuracy since the leaks are occurring at distances, as the
pipes are most times buried deep under the soil or water, the leak signal may be greatly attenuated
when it reaches the point of measurement and may even appear at a point further from the source.
Further, most outside pipe approaches are affected by certain factors such as pipe material, depth
of source, type of environment, and even temperature, which could gravely influence the results.
It is usually hard to accurately locate the leaks using external pipe detection methods which will
often require the manual work of skilled operators with high cost.
To reduce the occurrence of such leaks in the pipeline system, early detection of possible defective
points is required. Mapurisa and Sithole [8] used a profiling technique where line segment
deviation from the elliptical shape of a pipe is used to identify deformation points within pipe
installations. The method is applied significantly to major pipe deformations that result from
structural distortions and not necessarily from pipe internal defects. By using Root Mean Square
Deviation (RMSD) and Wavelet Packet Energy Loss (WPEL), Du et al [9] showed that
quantitative analysis of pipeline crack damage can be realized by electromechanical impedance
method, and localization analysis on the pipeline multi-crack damage can be achieved by stress
wave method based on sensor arrays. The focus however was on already laid pipes that have been
put to use for some time. The formation of such cracks could be traced to faults in the internals of
the pipes even before they were commissioned. These faults are usually due to production defects
that may arise from factors like uneven heating, decarbonization, overheating, oxidation, heating
cracks, and weld failure that can occur when pipes are being joined. Identification of these faults
in a pipe before installation is however likely to produce better results. To this end, researchers are
developing mobile robotic systems that navigate the internals of pipes with a range of capabilities
that enables pipe internal surface investigation. Sharma et al [10] developed a pipeline-crawling
robot for leak detection and focus on the ability of the robot to navigate through the pipe while
being controlled remotely. They did not specify however the length to which the robot can crawl
in a pipeline and the method of leak detection to be applied. In [11] an autonomous pipe inspection
robot is developed for smooth navigation and image capture to detect irregularity in pipes online.
There are methods for the detection and identification of defects in an object. One of the most
applied methods involves object image capture and processing to reveal certain features as
reviewed by Mohan and Poobal [12]. Using a convolutional neural network‐based framework Ni
et al [13] present an efficient image‐based structural damage detection and segmentation method
that automatically extract cracks quickly and accurately at a pixel level through convolutional
feature fusion and pixel‐level classification. The method delineates cracks accurately and rapidly
in images for structural crack detection in machine vision. With a data set of 603 raw images in
4096x4800 pixels, Feng et al [14] demonstrated an efficient crack detection method using a deep
active learning system and achieved an 87.5% detection accuracy. To obtain richer feature
representations, more detailed edge information, and improve the accuracy of detection, Zhou et
al [15] used a fusion of the Mixed Pooling Module and Feature Attention Module to achieve 87.3%
precision, 88.5% recall, and over 88% F-score as compared to other state-of-the-art crack detection
methods.
In a hybridization of image texture analysis and machine learning approaches, Hoang et al [16]
employ image texture computation feature extraction and a piecewise linear stochastic gradient
2
descent logistic regression (PL-SGDLR) pattern recognition on a dataset of 1240 image samples
to deliver a detection accuracy of 90.24%. In Zou et al [17] an end-to-end DeepCrack trainable
deep convolutional neural network was used for automatic crack detection by learning high-level
features for crack representation. The research employed multi-scale deep convolutional features
fused to capture crack line structures. Their experimental results indicate an average F-measure of
0.87 over challenging datasets and with better performance. Yang et al [18] propose a Feature
Pyramid and Hierarchical Boosting Network (FPHBN) architecture and an Average Intersection
over Union (AIU) measurement for crack detection. The network joins context information to low-
level features for crack detection and balances the contributions of both simple and hard samples
during training. Evaluation of the network on five crack datasets and comparing it with state-of-
the-art crack detection methods suggest better accuracy and generalization.
In [19], a unified crack and sealed crack detection approach are proposed for the detection and
separation of cracks and sealed cracks under the same framework. It trains a deep convolutional
neural network to pre-classify a pavement image into a crack, sealed crack, and background
regions. To segment the crack/sealed crack pixels efficiently and effectively, a block-wise
thresholding method was developed. Validating 800 images (each of 2,000×4,000 pixels), results
show that this approach accurately distinguishes cracks from sealed cracks with a detection
performance precision of 0.847 and recall of 0.951. Yang et al [20] in their research implemented
a Fully Convolutional Network (FCN) as a novel deep learning technique to address the problem
of low generality and inefficiency in existing crack detection systems. By quantitative
measurement of morphological features of cracks, such as crack topology, crack length, max
width, and mean width, the predicted segmentations are compared with the recent advanced
methods for crack recognition and ground truth. Results show that FCN is feasible and sufficient
for crack identification and measurement with accuracy, precision, recall, and F1 score of 97.96%,
81.73%, 78.97%, and 79.95%, respectively. Accordingly, Shi et al [21] propose a novel road crack
detection framework based on random structured forests (CrackForest), to address challenges with
automatic road crack detection due to inhomogeneity and topology complexity of cracks.
Experimental results compared the precision of CrackForest favourably with other state-of-the-art
competing road crack detection methods.
A number of these researches and many more, apply recent state-of-the-art networks like the
Convolution Neural Network (CNN) [22], Support Vector Machine (SVM) [23,24]; Artificial
Neural Network (ANN) [25], Random Forest [26] or a combination of these artificial intelligent
algorithms [27,28]; to try extract more accurate information from images with different
characteristic performance [29]. However, these works have employed mostly physical image
capture and have not applied robotic infrared image (IR) capture advantages to oil and gas pipes
in a before-use scenario.
Though this research did not focus on the use of an advanced AI algorithm for extracting more
image information, IR image capture was used to ensure better image information capture with a
modified image processing algorithm using conventional Gaussian and Sobel filters with
morphological transformation.
3. Internal Pipe Inspection Method

The focus of this work is on the application of modern technology in the investigation of pipe
internals. Its scope of operation has been limited to dry empty pipes that are currently not in
operation at the time of carrying out the inspections. To do this, a four-wheeled mobile robotic

3
vehicle, shown in Fig. 1, was developed and deployed to navigate through the pipe internals. The
robotic vehicle is equipped with a camera and light source to provide a visual percept of its
surroundings. It was also equipped with an Obstacle Detection and Avoidance System to aid its
navigation through the interiors of the pipe. The robotic system is used to obtain photographic
information of the pipe internals with the aid of Motion eye software on a Raspberry Pi platform
to capture and transmit images and videos in real time. These images are then analyzed to
determine possible areas or points of defects that may likely lead to leaks and system failure when
deployed.

Fig. 1: Autonomous robotic vehicle.

3.1 The Autonomous Vehicle Development

a. Chassis and kinematics
To fit into the pipe size and to navigate through, a pre-built robotic vehicle chassis with
Omnidirectional mecanum wheels as shown in Fig. 2a is adopted. The wheels are powered by
high-precision geared motors (JGA25-371) with encoded speed measurements. Using the
modified kinematics and control models of Equation (1) and (3) for backward movement and
Equation (2) and (4) for forward movement as applied in [30,31]; various movement directions as
indicated in Fig. 2b can be achieved.
 s1  1 1 l sin( ) 
s    cos  sin  0   dx dt 
 2  = 2  1 −1 l sin( )  
− sin  cos  0   dy dt  (1)
 s3  r  −1 −1 l sin( )  
     0 0 1   d dt 
 s4   −1 1 l sin( ) 

s 
 dx dt   1 1 −1 −1  cos  − sin  0  1 
 dy  = r  1 s
 dt   −1 −1 1   sin  cos  0   2  (2)
2 s 
 d dt  l sin( ) l sin( ) l sin( ) l sin( )   0 0 1   3 
 s4 

 s1  1 1 −( a +b 4 )  dx
s     dt 
 2  = 1 1 −1 ( 4 )   dy dt 
a +b
(3)
 s3  2 r 1 −1 −( a +b 4 )   
     d dt 
 s4  1 1 ( a +b 4 ) 

4
  s1 
 dx dt   14 1
4
1
4
1
4  
 dy  = 2 r  1 1  2
s
  dt
 4 − 1
4 − 1
4 4
 s  (4)
 d dt   − 1 a +b 1 a +b − 1 a +b 1
a +b 
3
 s 
 4

Where  = ( 4 ) − tan −1 ( b a) , si is the rotational speed of each wheel, l is the centre distance from
each wheel to the frame centre, r is wheel radius, b is half the centre distance between the front
and the rear wheels, a is half the centre distance between the left and the right wheels, φ is the
angular deviation from the x-axis, dx/dt is Forward velocity, dy/dt is lateral velocity, dθ/dt is the
angular velocity,

Fig. 2a: Robotic chassis fitted with Fig. 2b: Robot movement direction
mecanum wheels and motors [32]

b. Obstacle detection and photographic imaging

With the application of the HC-SR04 ultrasonic sensing system (Fig. 3a) from [33], an Obstacle
Avoidance Scheme (OAS) was achieved to ease navigation through the pipe at a non-contact
distance measurement of 2cm to 400cm. The OAS ensures that the robot maintains its line of action
through the pipe hole and can reverse in the confined space at recall.
Image detection and acquisition are carried out by the application of a Raspberry Pi Camera
Module (Fig. 3b) with a 600 view, a resolution of 1080pixel, a Pi 5MP OV5647 sensor and a night-
vision-enabled infrared LED from [34]. To capture images around the circumference of the pipe
at a given position, the camera is mounted on a salvo motor to turn at different angles widening
the view to 3600.

Fig. 3a: HC-SR04 ultrasonic sensor Fig. 3b: Raspberry Pi IR

Camera Module

5
c. Control System Hardware Design
The entire system hardware, as shown in Fig. 4, is organized on a Raspberry Pi Model B+
microcomputer.
Wireless
Laptop or Mobile
Interface
Device
(Bluetooth)

Motor +
Obstacle Detection ARM 11 CORE Motor Driver
Mecanum
(Ultrasonic Sensor) (Raspbian OS + Control code) (L293D)
Wheels

BCM2835 system-on-chip
VideoCore IV GPU

Image Capture (Pi

5MP OV5647 IR
Camera)

Fig. 4: Hardware design structure of robotic system

This microcomputer is based on the BCM2835 system-on-chip (SoC) with an ARM11 core
processor and a graphic processing unit (GPU) VideoCore IV running at 250MHz. It comes with
a 512Mbyte random access memory, a composite video output, USB ports, and pins for general
purpose I/Os (GPIO) among other I/O terminals [35]. The microcomputer is interfaced with
obstacle detection and wheels through its GPIOs and with the image capture and wireless interface
on dedicated terminals. The control signals from the microcomputer to the wheels are driven to
high voltage and current levels by the L293D to provide the needed driving wheel torque while
isolating the microcomputer from damage.
The control algorithm, developed in Python programming language, runs on top of the Linux
Debian Raspberry Pi OS (Raspbian) to execute control functions and video/photographic image
capture. The final robotic vehicle shown in Fig. 5, can be operated remotely and can be made to
inspect the entire pipe quickly or slowly as the speed can be controlled with an external control
feature. Most of its core functionalities can be controlled from a smart mobile device through a
Bluetooth connection. With a little modification, the robotic vehicle can be used in a wide range
of applications. Four rechargeable lithium polymer batteries are provided to power the system and
sustain its movement through long pipes.

6
 (a) Front View (b) Side View
Fig. 5: The developed image acquisition ground robotic vehicle

3.2 Image Acquisition

The robotic vehicle is employed to perform sensing, image capture and communication functions.
The vehicle navigates through the pipe, collecting visual information of the pipe internals and
transmitting this information simultaneously to a mobile device or a computer with the aid of
Motion Eye. Motion Eye is a software application that makes it possible to use a Raspberry Pi with
a camera as a video monitoring system. The system is equipped with thermal cameras for obtaining
heat signatures of various shapes and different material properties within the pipe internal. With
the data acquired using these thermal cameras, the pipes are examined to identify any anomalies.

To acquire pipe internal images, the robot was deployed in different pipes as shown in Fig. 6 and
commanded remotely through a mobile application to navigate through the pipe while capturing
internal Infra-Red (IR) images of the pipes as shown in Fig. 7.

(a) (b) (c)

Fig. 6: Robot deployment in pipes

7
 (a) (b)

Fig. 7: Pipe internal IR images captured from different pipes

3.3 Pipe Image Processing

Image processing is a significant part towards the identification of fault points within pipe
internals. Image processing entails several computer processes to perform different operations on
an image concerning enhancement, feature restoration, and analysis. There are several software
applications presently in use for image processing each with different capabilities among which
include OpenCV, Scikit-Image, NumPy and others. The application of image processing
techniques to the acquired images of the pipe parts can be analyzed to identify any possible defects.

Open Source Computer Vision (OpenCV), as one of the most used tools for computer image
processing tasks, is applied in this work to identify possible defective points in oil and gas pipe
internals. OpenCV as a library of programming functions contains various image-processing
algorithms written in C/C++ and other languages like Python. The image processing algorithm
used is depicted in Figure 8 and the code written in Python can be found in [36] with a little
adjustment to fit the desired goal.

8
 START

READ image from file and CONVERT to

gray-scale

REMOVE blur from binary image

(GAUSIAN FILTERATION)

DETECT deformation edges

(SOBEL FILTRATION)

CALCULATE threshold T(x,y) with a fudge

factor q

P(x,y) is a deformation Yes Is No

P(x,y) is a background
point Ig > T(x,y)

No Is P(x,y) last
pixel?

Yes

No Non-maximum
Yes Apply Non-maximum
Normalize image suppression
suppression
require?

INCREASE image foreground

(DILATION)

LOCALIZE point of deformation

(MORPHOLOGICAL TRANSFORMATION)

OUTPUT IMAGE

Fig. 8: Pipe image processing algorithm

The IR images acquired from the pipes are first converted to grayscale or binary form and passed
through a Gaussian filter to remove blurs. To detect defective regions, the result is passed through
a Sobel filter to create an image emphasizing edges at defective points. Thresholding is then
applied to determine the presence of edges at the defective region. To better detect more edges or
otherwise smaller deformation points, it is important to keep the threshold low. Here, a threshold
of 3.8 is used which gives vivid capture of defective points and well eliminates noise in the
resultant image. Much lower thresholds produced discriminatory results that are quite noisy.
Further, the non-maximum suppression function is applied by definition to thin the image edges
in a kernel of a 2 x 2 element structure. With the application of a dilation function, the image
foreground is increased before finally carrying out a morphological transformation to localize
possible defective points.

9
4. Result and Discussion
4.1 Image Capture Point
The robot’s image detection and capture ability at various distances for clear IR images were tested
as shown in Table 1. By sizing the picture frames with the object distance as shown in Fig. 9, a
good image capture point could be estimated to give a clear image.
Table 1: Image capture distance
Object Distance (cm) 40 80 120 160 200 240 280 320
Image frame (cm2) 15.6656 3.2903 1.4697 0.6671 0.4774 0.3097 0.2013 0.1548

Here, clear enough images for processing were captured at distances less than 80cm to reveal more
physical structure of the pipe internals at a magnification > 3.

18 15.6656
16
Image Frame (cm2)

14
12
10
8
6 3.2903
4 1.4697
2 0.6671 0.4774 0.3097 0.2013 0.1548
0
40 80 120 160 200 240 280 320
Object Distance (cm)

Fig. 9: Image capture point

4.2 Defect Detection
Fig. 10 shows the various original images captured using the IR camera. These images were passed
as input to the modified image processing algorithm. Fig. 11 shows the result of the algorithm
after passing through Gaussian and Sobel filtrations. Fig. 12 shows the final output images
obtained after undergoing suppression and morphological transformation to localized defects.

10
(1) (2) (3)

(4) (5) (6)

Fig. 10: Original IR image acquired

(1) (2) (3)

(4) (5) (6)

Fig. 11: Edge detection

11
 (1) (2) (3)

(4) (5) (6)

Fig. 12: Defect localization (Morphological transformation)
Created in the pipe are a few tiny holes used to represent defects in the pipe in addition to welded
joints. Application of the Sobel edge detection technique reveals possible defective points within
the pipe internal as indicated in Fig. 11. The further processing of the images through noise
suppression and morphological transformation at a fudge factor of 3.8 with a kernel of 2x2,
localized the defective points for easy identification as seen in Fig. 12.

5. Conclusion
Detection of deformations and leaks in oil and gas pipes is a very important factor in the
management of pipeline infrastructure. This paper presented the use of a mobile robot for the early
detection of possible defective points in not-in-use pipes before they are laid. The robot was
developed and packed on a pre-build mecanum wheeled chassis and equipped with obstacle
avoidance capability to enable easy movement in the pipe. The robot, also equipped with an IR
camera, runs on a Raspberry Pi microcomputer and was used to acquire pipe internal images. With
the application of a modified image processing algorithm, acquired images were processed with
results identifying possible defective points.
References
[1] H. Ali and J. Choi, “A review of underground pipeline leakage and sinkhole monitoring
methods based on wireless sensor networking”, Sustainability, vol. 11, no. 15(4007),
2019; [Link]
[2] J. Marshall, A. Bonchis, E. Nebot, and S. Scheduling, “Mining robotics” in Springer Handbook
of Robotics (2nd Edition), Revised on May 9, 2014, Available:
[Link]
[Link]

[3] P. S. Sapaty, “Military robotics: Latest trends and spatial grasp solutions”, Int. Journal of
Advanced Research in Artificial Intelligence, vol. 4, no .4, pp. 9-18, 2015.

12
[4] G. Zhang, J. Zhu, Y. Song, C. Peng, and G. Song, “A time reversal based pipeline leakage
localization method with the adjustable resolution”, IEEE Access, vol. 6, pp. 26993-27000,
2018.

[5] Z. Jia, L. Ren, H. Li and W. Sun, “Pipeline leak localization based on FBG hoop strain sensors
combined with BP neural network”, Appl. Sci., vol. 8, no. 146, 2018; doi:10.3390/app8020146.
[6] Z. Wang, M. Liu, Y. Qu, Q. Wei, Z. Zhou, Y. Tan, L. Hong and H. Song, “The detection of
the pipe crack utilizing the operational modal strain identified from fiber Bragg grating”,
Sensors, vol. 19, no. 2556, 2019; doi:10.3390/s19112556.
[7] N. V. S. Korlapati, F. Khana, Q. Noor, S. Mirza, and S. Vaddiraju, “Review and analysis of
pipeline leak detection methods”, Journal of Pipeline Science and Engineering, vol. 2, no.
100074, 2022; [Link]

[8] W. T. Mapurisa and G. Sithole, “Deformation detection in piping installations using profiling
techniques”, ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information
Sciences, vol. I-3, 2012
[9] G. Du, L. Huo, Q. Kong, and G. Song, “Damage detection of pipeline multiple cracks using
piezoceramic transducers”, Journal of Vibroengineering, vol. 18, no. 5, p. 2828-2838, 2016.
[10] S. L. Sharma, A. Qavi & K. Kumari, “Oil pipelines/water pipeline crawling robot for leakage
detection/cleaning of pipes”, Global Journal of Researches in Engineering Robotics & Nano-
Tech, vol. 14, no. 1, ver 1, 2014. Nov. 2022 [Online], Available:
[Link]
[11] M. Abdellatif, H. Mohamed, M. Hesham, A. Abdelmoneim, A. Kamal and A. Khaled,
"Mechatronics design of an autonomous pipe-inspection robot," in MATEC Web of
Conferences, 2018.
[12] A. Mohan and S. Poobal, “Crack detection using image processing: A critical review and
analysis”, Alexandria Engineering Journal, vol. 57, no. 2, pp 787-798, 2018.
[13] F. Ni, J. Zhang, and Z. Chen, “Pixel level crack delineation in images with convolutional
feature fusion”, Struct. Control. Health Monit, vol. 26, no. e2286, 2019.
[14] C. Feng, M.Y. Liu, C. C. Kao, and T.Y. Lee, “Deep active learning for civil infrastructure
defect detection and classification”, In Computing in Civil Engineering 2017; American
Society of Civil Engineers (ASCE): Reston, VA, USA, 2017; pp. 298–306.
[15] Q. Zhou, Z. Qu, and C. Cao, “Mixed pooling and richer attention feature fusion for crack
detection”, Pattern Recognition Letters, vol. 145, pp 96-102, 2021.
[16] N. Hoang, Q. Nguyen, and X. Tran, “Automatic detection of concrete spalling using piecewise
linear stochastic gradient descent logistic regression and image texture analysis”,
Complexity, vol. 2019, July 2019, [Link]
[17] Q. Zou, Z. Zhang, Q. Li, X. Qi, Q. Wang and S. Wang, "DeepCrack: Learning hierarchical
convolutional features for crack detection," IEEE Transactions on Image Processing, vol. 28,
no. 3, pp. 1498-1512, 2019.

13
[18] F. Yang, L. Zhang, S. Yu, D. Prokhorov, X. Mei and H. Ling, "Feature pyramid and
hierarchical boosting network for pavement crack detection," IEEE Transactions on
Intelligent Transportation Systems, vol. 21, no. 4, pp. 1525-1535, 2020.
[19] K. Zhang, H. D. Cheng; and B. Zhang, “Unified approach to pavement crack and sealed crack
detection using pre classification based on transfer learning”, Journal of Computing in Civil
Engineering, vol. 32, no. 2, 2018; [Link]
[20] X. Yang, H. Li, Y. Yu, X. Luo, T. Huang, and X. Yang, “Automatic pixel-level crack
detection and measurement using fully convolutional network”, Computer-Aided Civil and
Infrastructure Engineering, vol. 33, pp. 1090-1109,
2018, [Link]
[21] Y. Shi, L. Cui, Z. Qi, F. Meng and Z. Chen, "Automatic road crack detection using random
structured forests," IEEE Transactions on Intelligent Transportation Systems, vol. 17, no. 12,
pp. 3434-3445, 2016.
[22] S. Katiyar and S. K. Borgohain, “Comparative evaluation of CNN architectures for image
caption generation”, Int. Journal of Advanced Computer Science and Applications, vol. 11,
no. 12, pp 793-801, 2020.
[23] H. Hasni, A. H. Alavi, P. Jiao, and N. Lajnef, “Detection of fatigue cracking in steel bridge
girders: A support vector machine approach”, Archives of Civil and Mechanical Engineering,
vol. 17, no. 3, pp 609-622, 2017.
[24] J. S. Amelia and Wahyono, “Age Estimation on human face image using support vector
regression and texture-based features”, Int. Journal of Advanced Computer Science and
Applications, vol. 13, no. 12, pp 122-129, 2022.
[25] B.S. Wang and Z.C. He, “Crack detection of arch dam using statistical neural network based
on the reductions of natural frequencies”, Journal of Sound and Vibration, vol. 302, no. 4–
5, pp. 1037-1047, 2007.
[26] Z. Noshad, N. Javaid, T. Saba, Z. Wadud, M. Saleem, M. Alzahrani & O. Sheta, “Fault
detection in wireless sensor networks through the random forest classifier”, Sensors, vol. 19,
no. 7(1568), 2019, [Link]
[27] N. Saravanan, V.N.S. Kumar Siddabattuni and K.I. Ramachandran, “Fault diagnosis of spur
bevel gear box using artificial neural network (ANN), and proximal support vector machine
(PSVM)”, Applied Soft Computing, vol. 10, no. 1, pp 344-360, 2010.
[28] N. Hoang, "An artificial intelligence method for asphalt pavement pothole detection using
least squares support vector machine and neural network with steerable filter-based feature
extraction", Advances in Civil Engineering, vol. 2018, Article ID 7419058, 12 pages, 2018.
[Link]
[29] T. Han, D. Jiang, Q. Zhao, L. Wang, and K. Yin, “Comparison of random forest, artificial
neural networks and support vector machine for intelligent diagnosis of rotating machinery”,
Transactions of the Institute of Measurement and Control, vol. 40, no. 8, pp , 2681-2693,
2018.

14
[30] A. Norsharimie, A. Mohd, N. W. Mohd, H. A. Irawan, M. Mohamad, H. Mohamad, A. R.
Razali and S. A. S. Norhasmadi, “Omnidirectional configuration and control approach on
mini heavy loaded forklift autonomous guided vehicle”, 'EDP Sciences', December 2016,
DOI:10.1051/matecconf/20179001077

[31] T. Peng, J. Qian, B. Zi, J. Liu and X. Wang, “Mechanical design and control system of an
omni-directional mobile robot for material conveying”, Procedia CIRP 56, pp 412 – 415,
2016.

[32] Apache License, “How to make mecanum wheel robot”, ElectronicWings,

[Link]
mecanum-wheel-robot, (accessed Nov. 25, 2022)

[33] SparkFun, “Ul t rasoni c Di st ance S ensor - HC -S R 04”,

ht t ps: / / [Link] / product s/ 15569 , (accessed Nov. 25, 2022)

[34] MakerFocus, “MakerFocus Raspberry Pi 4 camera night vision adjustable-focus compatible

with RPi 3 B+/3B/2B+”, [Link]
night-vision-adjustable-focus, (accessed Dec. 5, 2022).
[35] Raspberry Pi, “Raspberry Pi Model B+”, [Link]
pi-3-model-b-plus/, (accessed Dec. 5, 2022)
[36] S. Srivatsa, “What is the OpenCV equivalent of this Matlab code for Sobel edge detection?”,
[Link], [Link]
equivalent-of-this-matlab-code-for-sobel-edge-detection, (accessed Jan. 3, 2023).

15