International Journal of Information Research and Review, February, 2016

International Journal of Information Research and Review

Vol. 03, Issue, 02, pp.1873-1877, February, 2016

Research Article
INTELLIGENT VISION BASED MOBILE ROBOT FOR PIPE LINE INSPECTION AND CLEANING
1,*Dr.

Rajesh Kanna, S.K., 2Ilayaperumal, K. and 3Jaisree, A.D.

1,2Professor,
3Student,

Department of Mechanical Engineering, Rajalakshmi Institute of Technology, Chennai, India

Anna University, India

ARTICLE INFO

ABSTRACT

Article History:

Development and advancement in technologies lead to constant change in the fluid transportation
systems. In recent years, fluids are moved through pipes laid above and below the ground and water
level. These fluid are readily react with the carrying pipes and need consistent maintenance and
repairing, which is a difficult task for humans, as they carry toxic chemicals, fluids, lighting, toxic
gas, low pressure and most of the time has small internal diameter or bends which become
inaccessible to human. To overcome the problem of cleaning, in this research, an intelligent internal
pipe cleaning robot using vision system had been developed. The developed robot can move through
the interiors of the pipes using independent motors and flexibly adjusting wheels. The vision system
has been used to capture the pipe interiors. The image had been interpreted using artificial neural
network and the output has been given to the controller and in turn it will clean or move to the next
segment. The experimentation had been conducted with virtual pipes and the performances were
found satisfactory.

th

Received 19 November, 2015

Received in revised form
17th December, 2015
Accepted 24th January, 2016
Published online 28th February 2016

Keywords:
Artificial neural network,
Cleaning Robot,
Vision system,
Inspection,
Pipe cleaning.

Copyright 2016 Dr. Rajesh Kanna et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.

INTRODUCTION
Day after day, technological improvements are advanced in
making the system with more flexibility, accuracy, security,
controllability, monitoring and maintenance of the system with
the help of digital machines. Varieties of intelligent systems are
designed and used in all the areas of the engineering to reduce
the human fatigue and automation with high quality outputs.
Among these intelligent systems, robotics plays a major role
starting from toys till the space machines. Robots are
categorized based on the task it carried out such as industrial
robots, automobile robots, material handling robots, medical
robots, humanoid robots, service robots, etc. Service robots are
the specific type of robots to do specific tasks which the
humans found difficulty and having more fatigue. Such service
robots have found application in servicing, inspection and
maintenance, which replaces the humans, mostly in the
substandard working conditions and conditions having health
danger, especially in inspecting underground installations such
as pipelines for water, gas, oil and sewage. The inspection of
pipes became essential as to check the corrosion level of pipe,
sediment blockage formations, recovery of usable parts, burs

*Corresponding author: Dr. Rajesh Kanna, S.K.,

Professor, Department of Mechanical Engineering, Rajalakshmi
Institute of Technology, Chennai, India.

from pipe interior, crack formations, sampling of sludge, scale

formation on pipe internal surface, inspecting complicated
geometries, various pipe diameters, etc. This research has
concentrated in the development of intelligent service robots
for inspection and maintenance of the fluid transport piping
system equipped with the required sensors and controllers. In
this work, camera has been provided with the transmitter
section and digital machine in the receiver section. The image
captured by the camera will be transmitted through the
transmitter section consist of four switch to give four orders
and receiver section receives the input for the robot action such
as movement direction, cleaning motor actuation, scrubbing
motor actuation and sediment removal.
Infrared and ultrasonic sensors work beside with the controller
to detect the obstacles and identify the alternate path for
movement. Once the digital machine receives the image from
the robot, then that image have been processed using image
processing algorithm and given as input to the artificial neural
network for interpretation and to take suitable decision for
further action. Various experiments had conducted in the
virtual environments to improve the inspection efficiency by
reducing the time, manpower, improving security and
efficiency. Thus the developed robot improves the
effectiveness of the inspection, reduces the cost and thereby
increases the profit margin.

1874

Dr. Rajesh Kanna et al. Intelligent vision based mobile robot for pipe line inspection and cleaning

Literature Survey
Service robots or automated structures can have a wide range
of applications in situations where the conditions for operators
are threatening, challenging or inaccessible. For example, Amir
et al. (2007) and Chao et al. (2007) utilized the service robots
for application in the inspection and maintenance, where it
replaces the humans. The crack and the types of sediments
inside the pipes had been captured using the camera in the
robot had also been developed and investigated by Jong et al
(2010) and kabeer et al (2013). Researchers like Kawaguchi et
al (1995) also developed mini robot to identify the micro
defects in the pipe and also developed robots for identifying the
cracks in the pipes of smaller diameters. Micro robots also
been developed by Koichi (1999) to inspect the small pipes and
to identify the micro defects inside the pipes. An attempt had
been made by Kuntze (1998) to extract the information from
the surrounding and the reaction based on the situation can be
obtained by using the sensors. The main objective of most of
the research like Md. Raziq et al (2012) is to increase the profit
by reducing the cost involved in the process, so many attempt
had made to reduce the cost of the pipe line inspection robots.
Due globalization and automation, the developed robot should
follow a standard in the size, shape and the functionality
related aspects and that should be follow the handbook of
industrial robotics by Parker et al (1998). Also the size and the
shape of the robot mainly depend upon the desired work
volume.
The work volume can be calculated by doing kinematic
analysis and the link mechanism analysis by Rattan (2009).
Robin et al (2000) and Scholl et al (1999) developed to do the
special activity and specially developed with improved
performance. The inspection of various diameter pipes and the
importance of various inner diameters had studied and
developed a robot to clean the multi diameter pipes had been
done by Shigeo et al (1999) and Young et al (2010). The major
component of the robot is the controller and the manipulator,
the design of the manipulator decides the accuracy and the
performance of the robot were analyzed by Srushti et al (2013).
Most of the time, the robot controller board cannot do all the
computations such as ANN implementations, etc. So the robot
should be controlled by the wireless system and successfully
implemented by Yusoff et al (2012). Beased on the literature
review, various gaps had been identified and the solution
methodology is explained in the following sections.
In-Pipe Inspection Robot
Designing of an in-pipe inspection robot is a difficult task as it
involves factors like pipe internal diameter, bends, tappers,
type of joints, lighting, type of sediments, steerability, turning
radius, accuracy of movement, size and shape, flexibility, self
sustainability, obstacle avoidance, safety, stability, material,
weight, actuation, operations, range, etc. In
order
to
accommodate various pipe diameters, the wheels are attached
with the base by spring mechanism. The spring mechanism is
used to accommodate various pipe diameters by contracting
and expanding the wheel base. Also the mechanism used to
exert required pressure on the wall of the pipes, so that the
wheel can roll over with greater friction and thereby avoiding
slippage.

Motion capability of the developed in-pipe robots includes

forward, backward, left and right turn motions. The four
wheels of the robot are controlled by separate actuators and the
switching on and off combinations will provide the above
mentioned movements as required. The autonomous movement
of the wheels also helps in overcome the stagnation of robots
from sediments and joints. The pipe lines are connected with
various types of joints and the joints considered for the
experimentation in this research are L type, T type, Y
type and Cross + type joints apart from unions and
couplings. The path selection capability of robot inside a pipe
is of utmost importance to perform meaningful and accurate
inspection. The movement of the linear joints such as flange,
couplings and welded joints can be accommodated by wheel
spring mechanism. The ultrasonic sensor and infrared sensor
identifies the L type joints as obstacle and provide automatic
movements to the robot. The movement decision for the
remaining joints can be achieved manually using the camera
images and the required instructions can be given to the robot
through the digital machine using receiver and transmitter
mechanism. Even though the robot is designed for the
horizontal and vertical movements, experimental results are
satisfactory for the horizontal movements and not for the
vertical and inclined movements.
Steering mechanism and the turning radius of robot depends on
the size of the robot and the wheel turning mechanism. The
size of the robot has to be decided based on the placement of
the sensors, actuators and the controllers. The robot had been
developed to provide a smooth movement without any stuck of
robot in the pipe line and can be achieved by providing less
turning radius for in-pipe inspection robots. The spring
mechanism in the wheels gets compressed on one side and
expanded on other side and thereby provides better turning
radius for the developed robot. Shape adaptability also
considered in this research to move the robot for inspection in
the pipe lines with varying diameters such as tapered pipes and
stepped pipes. Although, with single robot, it might not be
possible to cater all ranges of diameters, but at least a robot can
be developed for accomplishing a reasonable range of pipe
diameters.
The main objective of the robot is to clean the pipes of circular
diameter, but the developed robot has the adoptability for the
different cross sections such as rectangular, truncated pyramid,
circular, semi-circular and elliptical. Sometimes the improper
deposition of the sediments also makes the pipe of various non
prismatic shapes. The detection of the deposited sediments and
the obstacles depends upon the position of the infrared sensor
in the robot and the training rate of artificial neural network.
This has been achieved by having the spring mechanism in the
wheel base and also in the cleaning arm. The spring
mechanism in the cleaning arm also provides the required
pressure to the cleaning and compensates the cleaning head
wear. The gyroscope sensor in the circuit board sends the
signal to the digital machines and is used to control the stability
of the robot manually, as the stability of the robot inside the
pipeline will play a major role. Any instability or upside down
of the robot will lead to difficulty in bringing back the robot
from the pipe line. This is again a measure concern particularly
with inclined and vertical pipes. So in this research, robot
orientation control has been done manually.

1875

International Journal of Information Research and Review Vol. 03, Issue, 02, pp.1873-1877, February, 2016

Infrared and ultrasonic sensors are used to detect the obstacle

and to provide the proper navigation for the robot, in order to
make the robot motion and operation automatic. These sensors
avoid any collision between robot body with obstacle and wall
of pipe lines. Because any damage to the pipe interior will lead
to failure of the pipeline. Camera is used to capture the interior
images of the pipe line and also to view the status of in-pipe
operations. Separate lighting system with LED had provided in
the robot to illuminate the interior of the pipe lines, which is
easy to interpret the images obtained. The obtained images are
transferred to the digital machines for neural network decision
making. The developed robot is given in the Figure 1.

OUTPUT = 1/ (1 + eai)

(1)

Whereas ai is the activation function defined as wij*oj + ti

wij is the weight value of the link joining ith and jth neuron,
ti is the threshold value of the neuron.
In the present work, BPN algorithm is used for image
processing. First the number of layers and neurons are
initialized to 70, 150, 250, 100 and 7 neurons in input, hidden
and output layers respectively. Processed image data is given as
input along with the expected output data to the network.
Weight values and threshold values are obtained by calculations
and are stored in the database. Output of the BPN is calculated
using the forward propagation. Based on the calculated output,
error value is found. Error value is the difference between
calculated and desired output. If the error value is greater than
the mean square error, then the error is back propagated
through the network. During backward propagation the weight
and threshold values are updated accordingly. Repeat this
procedure until the error value reaches the least minimum. Thus
the BPN is trained. After training the network can be
implemented for application. For experimentation, fifty
different set of images were taken for training and ten images
were taken for testing the network. During training phase, the
weight matrix, w is updated according to the following formula
given by,

Figure 1. Developed Pipe Cleaning Robot

Vision Intelligent Robot

Vision intelligence can be defined as the acquisition of image
data, followed by the processing and interpretation of these data
by artificial neural network for some useful decision making.

where is the learning parameter, is the constant, Ttrain is the

total number of training cases, O h [j] is the output from the jth
hidden unit or ith component of kth training vector, and kw = (tk
Ok) Ok (1 - Ok) , where tk and Ok are the desired output and

Table 1. Weight values before and after training

0.7159
0.3713
0.9818
0.7674
0.9091
0.7785
0.8759

(a) Random initial weights w1j

0.4665 0.2002 0.2299
0.4927 0.5810 0.3959
0.6434 0.8702 0.2426
0.0697 0.1794 0.7382
2.8E-2 0.3387 0.5911
0.2564 0.5689 0.1256
0.9689 0.1259 0.5478

0.8857
0.39950
0.9047
0.5857
0.7221
0.2654
0.7521

The use of machine vision and neural network in interior pipe

cleaning robots can provide solutions for variety of problems
associated with it. Artificial neural network capable of
processing number of different pieces of information
simultaneously through the interaction of large number of
highly interconnected processing elements called neurons.
Several ANN models like Adaline and Madaline,
Backpropagation,
BAM,
Hopfield
memory,
Counterpropagation, etc., are available. Backpropagation has
certain advantage over the other algorithms i.e. varying step
size, flexible learning speed, any number of stopping criteria
can be fixed, network size is user defined, network pruning is
easy, etc. In the Backpropagation network (BPN), the activation
function chosen is the tan sigmoid function, which compresses
the output value in a range between 0 and 1. Tan sigmoid
function used is given in the Equation 1.

(b) Weight w after The Back propagation Training

0.7987
4.306
0.3575 0.9417
0.4486
0.8303
0.9769
0.3052 0.7698
0.4397
0.8371
0.8816
0.2501 4.6E-2
0.8009
0.50827
0.9128
-0.351 0.0143 -7.1E-02
0.4488
0.4374
0.2362 0.3483
0.2647
0.7856
0.5689
0.9987 0.4257
0.9874
0.7554
0.2E-01 0.5694 0.6487
0.2532

the output of the network for the kth training vector

respectively. Table 1 shows the random weights which are
initially assigned to w and the weights after the back
propagation learning for the data set 1 (note: table shows a
sample initial and trained weights).
Experimental Implementation
The developed robot is allowed to move through the interior of
the pipes, the springs pushes the wheel away from the robot
and make the necessary grip to the robot to move on the inner
circumference of the pipe. Then the camera captured the
interior of the pipes on-line with the help of LED lighting. The
obtained video image had converted to bitmap format which is
having the structure as shown in Figure 3. Form the Figure 2,
it is clear that the actual image data starts from 54th byte of the
bitmap file and to reduce the computational time, the obtained
color image need

1876

Dr. Rajesh Kanna et al. Intelligent vision based mobile robot for pipe line inspection and cleaning

assist the robot for cleaning purpose. These control systems has
communicated through one transmitter and another receiver.
The robot performance of the robot had been improved by
investigating the angular positions of the links, by compressing
the image using SPIHT and by conducting sensitivity analysis
for identifying the ANN parameters. The experimentation had
been conducted for various combinations of pipes with
different types of joints and sediments. The obtained results
were found satisfactory and the developed intelligent robot can
be used for the real time pipe inspection and cleaning.

REFERENCES
Figure 2. Bitmap file format

to be converted to monochrome image by converting the color

intensities into single binary code i.e. either 0 or 1 based on the
degree of darkness. This image contains large amount of noises
due to various external parameters like orientation, size,
illumination, background, etc and that should be filtered. To
filter this noise each and every pixel are compared with its
surrounding eight pixels. If this pixel and maximum number of
the surrounding pixels are different, then the pixel is converted
to the surrounding pixel value. The attributes of the camera
image is varying due to external factors like focal length,
orientation of camera, angle of capturing, etc. Again, the image
contains large quantity of raster data. This can be overcome by
using SPIHT method of image compression and thereby
achieving the compressed image without data loss in shorter
time. In this method, the image is divided into four equal
quadrants. Again each quadrant is divided into four equal sub
quadrants.
This dividing procedure will continue till the image matrix size
reaches to the least minimum size i.e. the stage in which further
dividing cannot be possible. The image data in that matrix are
read. The maximum repeated code might assume as code for
that matrix. In this way the size is reduced by th. Again the
same procedure is repeated until the required size obtained and
the size of 70 is considered in this research. The compressed
image has to be given as input to the trained ANN. The ANN
parameters used for the researcher are Mean squared error:
0.01, the total output error T error = Dop Oop, is less than
0.01, maximum number of iteration: 100, Learning rate: 0.3.
Thus the trained ANN extracts the required information and
generates the output to control the motors for movement and
for cleaning and rubbing operations. The same procedure has to
be repeated till the robot moves all around the interior of the
pipe lines. The same procedure can be used not only to clean
the interior of the pipes, but also to identify the cracks,
damages and to predict the life time of the pipes.
Conclusion
A real prototype had been fabricated and tested for the robot
functionally. The spring mechanism designed has improved the
mobility of the robot in pipeline. The control system used
consists of two sections, one on the robot to control the
movement and to overcome the obstacles on the path and also
to identify the joints on the pipe lines. The second section of
the control system is on the digital machine to run the image
processing module and artificial neural network module which

Amir A. F. Nassiraei, Yoshinori Kawamura, Alireza Ahrary,

Yoshikazu Mikuriya and Kazuo Ishii, 2007. Concept and
Design of A Fully Autonomous Sewer Pipe Inspection
Mobile Robot KANTARO. IEEE International
Conference on Robotics and Automation Roma, Italy, 10-14
April 2007: 136-145.
Chao-Pei Lul, Han-Pang Huang, Jiu-Lou Yan and Ting-Hu
Cheng, 2007. Development of a Pipe Inspection Robot. The
33rd Annual Conference of the IEEE Industrial Electronics
Society (IECON). Nov. 5-8, 2007, Taipei, Taiwan.
Jong-Hoon Kim, Gokarna Sharma, and S. Sitharama Iyengar,
2010. FAMPER: A Fully Autonomous Mobile Robot for
Pipeline Exploration. IEEE:517-523.
Kabeer Mohammed and Dr.Bhaskara Reddy, 2013. Optimized
Solution for Image Processing Through Mobile Robots
Working as a Team with Designated Team Members and
Team Leader. International Journal of Computer
Engineering & Technology (IJCET). 4(3):140 148.
Kawaguchi, Yoshida, Kurumatani, Kikuta, and Yamada, 1995.
Internal Pipe Inspection Robot. IEEE international
Conference on Robotics and Automation: 857-862.
Koichi Suzumori, Toyomi Miyagawa, Masanobu Kimura, and
Yukihisa Hasegawa, 1999. Micro Inspection Robot for 1-in
Pipes.
TRANSACTIONS
ON
MECHATRONICS.
4(3):2869-292.
Kuntze and Haffner, 1998. Experiences with the development
of a robot for smart multisensory pipe inspection.
Proceedings of the IEEE International Conference on
Robotics & Automation Leuven, Belgium, May 1998 : 17731778.
Md Raziq Asyraf Md Zin, Khairul Salleh Mohamed Sahari,
Juniza Md Saad, Adzly Anuar, and Abd Talip Zulkarnain,
2012. Development of a Low Cost Small Sized In-Pipe
Robot. International Symposium on Robotics and Intelligent
Sensors IRIS 2012: 1469-1475.
Parker, L.E. and Draper, J.V. 1998. Robotics applications in
maintenance and repair. Handbook of industrial Robotics.
Rattan S S, 2009. Theory of Machines. Third Edition, Tata
McGraw Hill Education Private Limited, New Delhi.
Robin Krikham, Patrick D Kearney, Kevin J. Rogers and John
Mashford, 2000. PIRAT - A system for quantitative sewer
pipe assessment. International Journal of Robotics Research.
19:1033-1053.
Scholl, Kepplin, Berns, and Dillmann, 1999. An Articulated
Service Robot for Autonomous Sewer Inspection Tasks.
International Conference on Intelligent Robots and Systems.
Proceedings of the IEEGRSJ: 1075-1080.
Shigeo Hirose, Hidetaka Ohno Take0 Mitsui Kiichi Suyama,
1999. Design of In-pipe Inspection Vehicles for 25, 50,

1877

International Journal of Information Research and Review Vol. 03, Issue, 02, pp.1873-1877, February, 2016

and 15 pipes. Proceedings of the 1999 EEE International

Conference on Robotics & Automation Detroit, Michigan
May 1999.
Srushti H. Bhatt, N. Ravi Prakash and S. B. Jadeja, 2013.
Modeling of Robotic Manipulator Arm. International
Journal of Mechanical Engineering & Technology (IJMET),
4(3): 125 129.

Young-Sik Kwon and Byung-Ju Yi, 2010. Development of a

Pipeline Inspection Robot System With Diameter of 40mm
to 70mm (Tbot-40), IEEE, 2010:258-263.
Yusoff, Ibrahim, Hamzah and Kadir, 2012. Development of
Air Conditional Route Wireless Inspection Robot.
International Symposium on Robotics and Intelligent Sensors
2012. IRIS 2012: 874-880.

*******