International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
Industrial vision system for distribution and size
analysis of pellets
Navdeep Singh, ShaliniSaxena, Dhiraj, J L Raheja
Machine Vision Lab, DSG, Council of Scientific & Industrial Research -Central Electronics Engineering Research Institute
(CSIR- CEERI), Pilani, Rajasthan, India,
ABSTRACT
Use of optimum sized pellets makes the steel production process more reliable and energy efficient. The development of the
vision based system to keep these pellets size in check is discussed in this paper, which involves integrating industrial grade
IBM-PC with a high resolution industrial camera. We have applied this method to the pelletizing system in the Essar Steel
Plant at Vishakhapatnam and made possible automated operation of the pelletizer.
Keywords: Vision System, Pellet, Size Distribution, Image Processing
1. INTRODUCTION
The Pellet Size Distribution System is a modular inspection system for the analysis and classification of the size and
shape of pellets. The pellets are granulated materials and are made from fine ores through pretreatments with a
pelletizer. It is known that there exists a high degree of correlation between the average size of pellets and the operating
conditions of an electric furnace. Accordingly, pellet size control in the granulation process is indispensable for the
stable operation of the furnace. According to the industry norms the size of pellets must be maintained between 9mm16mm in diameter [1]. This size is analyzed and maintained in Pellet Size Distribution System, which is previously
done manually and is not highly accurate. The vision based system is highly reliable and robust to be used in the
industrial conditions. This paper outlines the method used and the system developed. A Disk pelletizer is a rotary tray
with diameter of approximately 5 meters, inclined at an angle, as shown in Fig. 1. It rotates at a speed of 6-8 rpm. The
pellet size is controlled by flow rate of water, speed of disk and angle fin inside disk. It is difficult to ensure accurate
size control solely by visual inspection of human operators. Thus, there is a strong need for the development of system
for measuring pellets continuously and accurately.
Figure 1 Disk Pelletizer
2. Principle of Measurement
In order to ensure the accurate measurement in real time the pellet, partially as well as completely visible, in the top
layer is analyzed. Pellet sizes are measured using a well-known algorithm, called Circular Hough Transform (CHT).
Figure 2 shows the block diagram of the steps of the method followed.
Volume 2, Issue 3, March 2013
Page 7
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
Capture Image
Edge Detection
Smoothing
Circular Hough
Transform
Size analysis
Figure 2 Block Diagram
The image is captured using a high resolution, high frame rate camera. The firewire port provides a link between
camera and the PC. Basler scA640-70fm camera is used for this purpose, which provides IEEE 1394b (S800) port.
Smoothing operation is done to remove the unwanted noise before processing. To complete this operation a low pass 2nd
order Butterworth filter is used [2]. It is given by the formula shown in equation (1).
(1)
Where, n=Order of the filter, in this it’s value is 2.
The result of applying the filter is shown in figure 3.
Figure 3 (a) Input Image (b) Smoothed Image
Edge Detection of the pellets in the image is required before we apply CHT. As it will be explained later, CHT,uses
edge points to plot the center of the circles inscribing the pellets. We employ Sobel operator [3] shown in figure 4,
where both horizontal and vertical filters are used. It uses 3x3 convolution masks, one for x-direction and the other one
for y-direction. Euclidean sum is taken to get the final edges formula as given in equation (2).
(2)
Figure 4 (a) Horizontal (b) Vertical
Thresholding of the image is based on the present value of the gamma and gain setting of the camera. The gain and
gamma settings of the camera are given on the front panel itself. Thethresholding method makes the process more
dynamic and robust against the illumination changes. Figure 5 shows the result of applying the sobel mask over the
image at discharge point.
Volume 2, Issue 3, March 2013
Page 8
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
Figure 5 (a) Smoothed
Image
(b) Edge Detection
After edges the edge directions and direction maps are calculated. This is done to reduce the number of calculations
done over a symmetrical circle, as it can only be reduced to quarter of a circle [4]. Already calculated vertical edges and
horizontal edges were used to calculate the direction using the formula as given in equation (3).
(3)
2.1. Hough Transform:
Hough transform is used for feature detection in automated analysis of digital images. The method is given the family
of curves being sought (circles in this case) and produces the set of curves from that family that appear on the image
[4]-[5].
The method of Hough transform is that if a point lays on a circle then the gradient at that point, points to the center of
that circle [8][9] as shown in figure 6. So, if numbers of edge points are given and range of radius is known and the
direction of the vector from edge point to the center is computed, the coordinates of the center can be found [6]-[7].
Figure 6 Hough Method
The transform will create the maximum brightness at the probable centers. These probable centers can be spread over
larger value of pixels, because of the range of radii. To concentrate these accumulated points we applied a 17x17
Laplacian of Guassian (LOG) filter [10]-[11]. LOG concentrates the highest brightness in the center of the accumulated
space thus reducing the calculation afterwards. The result of applying Hough algorithm and then LOG filter are shown
in figure 7.
Figure 7 (a) Result of
applying Hough algorithm
on image 5(b)
(b) Probable centers after
applying LOG
From concentrated accumulated space the center of the circles are found using local maxima function [12]. These
maxima are considered as centers.The radius is then found by projecting circles with the radii from the calculated
Volume 2, Issue 3, March 2013
Page 9
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
centers and then accumulating it in the one dimensional space with radius of the concentric circles. For the
representation circles are drawn over the original image with calculated centers and radii. Figure 8 shows the result of
applying local maxima and also drawn circles over the original image.
Figure 8 (a) Centers after
local maxima
(b) Final output image
with circles drawn
Since, there is need for better method than visual inspection of pellets, the radii results are plotted in a histogram and
are displayed. The results can be analyzed to check the system at any instance. Also, the system control decisions are
based on the value of these radii (i.e.the size of each pellet at any instance). In figure 9 the histogram of figure 8b is
shown.
Figure 9 Histogram plot
2.2. Experimental Results
In order to evaluate the performance of the method, system was tested at the Essar Steel Plant, Visakhapatnam. The
conditions of the test experiment are as follows:
(1) The testing was done at various RPM of disk Pelletizer.
(2) The method was tested both at center part and the discharging part of disk pelletizer.
(3) Light conditions were also varied as we wanted to test the robustness of our system against changes in
illumination.
Figure 10 shows the size of the analysis of pellets at the discharge point of disk pelletizer running at the speed of 7.21
RPM. As it is shown most of pellets are restricted in the range of 9-16 mm in diameter.
Figure 10 Experimental Result 1 at 7.21 RPM
Figure 11 shows the result of system at the center of the disk pelletizer running at 6.86 RPM. It is clear that the there is
hardly any effect of the location or the RPM. The results still are very much usable.
Volume 2, Issue 3, March 2013
Page 10
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
Figure 11 Experimental result 2 at 6.86 RPM
Thus with above results it can be concluded that the method can be applied to both the discharge part where the pellets
are completely formed, and upper part, where the pellets are still in making. Also we tested the measurement error is
±1 mm.In order to test the effect of variations in illumination on the results, several test results were performed. Figure
12 shows the result of system at discharge point of disk pelletizer running at 7.21 RPM in very bad lighting.
Figure 12 Experimental result 3 at discharge point
Thus it can be concluded that the size index of the pellets is measured accurately and is not influenced by the variations
in light.
2.3. Pellet Size Distribution System
An automatic Pellet Size Distribution and control system, to which the above method is applied has been made and is
tested at Essar Steel Plant, Visakhapatnam.System architecture image is shown in figure 13.
Figure 13 System Achitecture
The system consists of a computer, which essentially an IBM PC with NI LabVIEW running, which is connected to a
high speed firewire camera. An actual experimental setup at Essar Steel Plant, Visakhapatnam is shown in figure 14.
Figure 14 Automatic Pellet Size Distribution and
control system
Volume 2, Issue 3, March 2013
Page 11
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
System includes the following hardware:
(1) Firewire camera :
Basler Scout scA640-70fm CCD (1/3”) with electronic shutter
(2) IBM PC :
CPU: Core 2 Quad @ 2.40GHz, RAM: 4GB
(3) Input Image:
640x480 monochrome image
(4) Calculation Time: 1.93 sec
3. Conclusion
We have described a new method for measuring and controlling the size of pellets with image processing techniques
and automatic control system. A great advantage of the method is the offering of a huge variety of customized
configurations to be applicable in for every objective perfectly. This includes a precise analysis and classification of the
raw material.
References
[1] Pelletizing
Technologies
for
Iron
Ore,
August.
2,
2010,Available
:
http://www.outokumputechnology.com/29696.epibrw
[2] Rafael C. Gonzalez and Richard E. Digital Image Processing. Upper Saddle River, New Jersey07458: Prentice
Hall, 2002, pp. 167
[3] Rafael C. Gonzalez and Richard E., Digital Image Processing Using Matlab, Upper Saddle River, New Jersey,
Prentice Hall, 2002
[4] Linda Shipro and George Stockman, Computer Vision, Upper Saddle River, New Jersey, Prentice Hall, 2000
[5] H.K. Yuen, J. Princen, J. Illingworth, J. Kittler. “Comparative study of Hough transforms methods for circle
finding.” Image Vision Computing, vol. 8, pp. 71–77, 1990
[6] Duda, R.O. and P.E. Hart. ”Use of Hough transformation to detect lines and curves in picture”. Communications
ACM, vol. 15 pp. 11-15, 1972
[7] JagdishRaheja, V Upendranath, Rajesh M, P Bhanu Prasad, “Image Processing Techniques for Pellet Size
Distribution.” in International Seminar on Mineral Processing Technology, Bhubaneswar, India, 2009
[8] Chester F. Carlson, Class Lecture, Topic: “Hough circle transform.” Rochester Institute of Technology, Henrietta,
Rochester, October 11, 2005
[9] Kimme, C. D. Ballard and J. Sklansky. “Finding circles by an array of accumulators.” Proc. ACM., vol. 18 pp.
120-122, 1975
[10] Brinks R. “On the convergence of derivatives of B-splines to derivatives of the Gaussian function.” Computational
& Applied Mathematics, vol. 27, 1 pp. 79-92, 2008
[11] Rafael C. Gonzalez and Richard E. Digital Image Processing. Upper Saddle River, New Jersey07458: Prentice
Hall, 2002, pp. 583-585
[12] Lindgren, Georg. “Local maxima for Gaussian Fields.” Arkiv for MateMatik, vol. 10, pp. 195-218, 1972
AUTHOR
Navdeep Singh.has received his B. Tech from Lovely Institute of Technology, Jalandhar, Punjab,
India. He worked as Project Assistant, in Central Electronics Engineering Research Institute
(CEERI), Pilani, Rajasthan, in a national level project. His areas of interest are Real Time Image
Processing and Human Computer Interface
ShaliniSaxena has received her Diploma in Electronics Engineering from Government Girls
Polytechnic Lucknow. She is currently pursuing her B.E. degree in Electronics and Communication
Engineering from Institution of Engineers (AMIE) Kolkata. Since 2007, she has been working as
Technical Assistant in Digital System Group in Central Electronics Engineering Research institute
(CEERI), Pilani, Rajasthan. Her research interest includes Image Processing,Human Computer
Interaction,Digital Signal Processing, Computer Vision,Machine Learning and Electronic
Design and Fabrication
Volume 2, Issue 3, March 2013
Page 12
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com
Volume 2, Issue 3, March 2013
ISSN 2319 - 4847
Dhiraj has received his M. Tech from Panjab University. At present he is a Scientist in Central
Electronics Engineering Research Institute (CEERI), Pilani, Rajasthan, India since 2011. His areas
of interest are Digital Image Processing, Hardware Prototyping, Evolutionary Algorithms, Human
Computer Interface.
JagdishLalRahejahas received his M. Tech from IIT Kharagpur and PhD degree from Technical
University Munich, Germany. At present he is a Sr. Scientist in Central Electronics Engineering
Research Institute (CEERI), Pilani, Rajasthan, India since 2000. His areas of interest are Digital
Image Processing, Human Computer Interface and Cartographic Generalization.
Volume 2, Issue 3, March 2013
Page 13