Abstract—This work consists in automate the measurement of the length of the slabs in the continuous casting machine in a steel mill.
This automation will be done by an artificial vision system that implements a camera matrix and embedded vision software.
First, we have to make a rigorous selection of attributes such as the margin of the field of vision, the distance between the camera and the slab, the geometric landmarks of the measure, the detection landmark of the arrival of the slab and the optical reference landmark for measure. We also established a relationship between the real length of an object and the number of pixels occupied by its image after a serie of precise tests in order to integrate it in the measure program.
The measuring principle is divided into two steps:
First, we make a segmentation of each image with a program developed in this sense which aims to differentiate between the slab and the roller train by reading the brightness of each pixel of the acquired image.
This segmentation uses the histogram of gray levels of the slab and the roller table which were obtained after the analysis of their images in their real lighting conditions.
Then, the slab is measured with another developed program which calculates the abscissa of the detection landmark of the arrival of the slab, tests the value of the gray level of the pixel of this abscissa, gives a decision signal to blowtorches, then retest the value of gray level of the same abscissa to detect the end of the slab and wait the arrival of the next slab. Both programs operate in a loop until the end of the cut of all of the slabs. At each new cutting plane of the slabs we introduce new data according to the customer's command.
A study was necessary to establish a criterion for detecting the slab because the value of one pixel can not be a reference and so we chose the reading of 15 points of
S.Taleb, Z.Mentouri, S.Ziani,and A.Boudiaf are with Centre de Recherche Scientifique et Technique en Soudage et Contrôle, Unité de Recherche Appliqué en Sidérurgie Métallurgie, URASM/CSC ; B.P 196, 23000 Annaba, ALGERIE
(Email: taleb_samira71@yahoo.fr).
twenty and this criterion has yielded concrete and reliable results.
A simulation with the camera matrix TVC500 and the image processing map PIP was realized in the process laboratory with success and all programs were developed with the C language. On the ground, it will require ensure a stable lighting within the cut area and a good braking system of the carriage or consider a second camera for measuring the lag of the carriage because it represents the origin landmark and thus adjust the measure. Our work will allow a better measure reliability, the possibility of its exploitation for the optimization of the cut and also the improvement of working conditions of the workers.
Keywords: Artificial vision, pixel, histogram, real time, image, gray level.
I. INTRODUCTION
Artificial vision is a very dynamic field of technology and booming in the field of scientific research and the applications in our daily life. It is a branch of artificial intelligence that allows a vision system to understand what it "sees» and so be able to monitor, detect and analyze everything that happens on a production line and provide an effective and fast solution at any problems that may arise.
Many manufacturing processes producing continuous strips of materials such as steel, stationery, sheet metal, pasta and plastics industries have introduced artificial vision into their process in order to reduce rejects and unnecessary loss of their basic matter and increase their production and quality of their product.
This theme consists in the automation of the measure of the length of the slabs in continuous casting machine and we chose computer vision as a working tool for the advantages of speed and flexibility it offers.
We used a matrix camera TVC500 to collect images in real time that are processed by the PIP card images processing implanted in the Industrial PC.
A program was developed with the C language and is composed of a program of image segmentation, a program of determining the detection criterion, a recognition program of beginning and end of the slab, a calibration program and a
The Measure of the Length of the Slabs by Artificial Vision
S.Taleb*, Z.Mentouri*, S.Ziani*, A.Boudiaf*
I
measure program. We introduced in these programs predefined data such as geometric benchmarks and the correction factor.
This work will allow the increase of the production of the slabs with the minimum of discarded the improvement of working conditions of the workers and the possibility of cutting optimization.
The paper is organized as follows. The next section presents the existing measure system and the general philosophy of measure by artificial vision. Then we will make the choice of all the parameters necessary in our study. The measuring principle, the criterion of detection and calibration are considered in Section 3. We will talk about the simulation performed at our laboratory and we will discuss obtained results in Section 4. Finally, we will give the conclusion of this work in section 5.
II. METHODOLOGIE II.1DESCRIPTIONOFTHESYSTEM
Continuous casting consists in pour continuously and regularly liquid steel into the ingot mold. Firstly, once the steel is produced, it is transported in the steel ladle which will be posed on a tourniquet. The temperature of the metal to sink is taken in order to verify whether it is in the norms. If the result is positive, the dispatcher is then filled through the opening of the drawer of the steel ladle. When the metal reaches a certain level, the distaff of the distributor is open manually and the metal is spilled in the ingot mold. The slab whose heart is not solidified is extracted from the ingot mold via the extraction cage by a mannequin. At the mold, the metal begins to solidify thanks to the primary cooling. The slab undergoes a secondary cooling by water sprinklers. The mannequin is separated from the slab by a hammer blow in the zone of uncoupling. The slab is then cut by oxy-cutting to the desired length. [1]
The cut of the slab is operated by an operator who evaluates the desired length visually relative to physical points of reference for each length. We understand that it is impossible for the human eye to be precise and therefore the length of the slab will be different of slab to another and the accumulation of these differences of length represents a loss of raw material in case the slab is longer than planned and downright discarded if the slab is shorter. (See fig 1)
To improve the quality of the measure and provide good precision cutting, we opted for a system of artificial vision composed of a matrix and industrial camera TVC500 and an image processing card PIP (MATROX) installed on an industrial computer [2]. The camera TVC500 is of the CCD technology and it is highly resistant to shocks and vibrations.
Its operating temperature is between -10°C and 50°C. PIP card transforms the received video signal in real time in an 8-bit number which will be converted by the lookup table at a value representing its gray level. We can establish a list of conversion to grayscale at our choice, by initializing lookup table. The programs will be developed in the C language because it is one of the languages accepted by the PIP-EZ own
library card at the PIP card [3]. The PIP-EZ library allows control of functions such as the input-output buffer acquisition images, the handling of graphs, of lookup table and so on ...
Also an interpreter program called 'PIPINT' allows access directly to PIPEZ instructions from the keyboard.
fig1. Measuring zone
Whatever the application envisaged, the optical information processing is based on the same principles and involves the following steps [4]:
- Reception and memorization of the image: Each pixel in an image is represented by the value of its gray level at which you can access by reading its coordinates, and this value will be memorized continuously.
- Segmentation of the image: It consists in distinguish between the useful areas for the envisaged application and those that are not. This step is made in advance and will be of a great contribution because in order to take the appropriate decisions for each application, it is necessary to have a good knowledge of the objects to visualize that is to say the value of the luminosity of each pixel which represents them.
-Choice of attributes: This phase consists in determine a certain number of characteristic parameters of the application.
-Decision: After the selection of attributes and image segmentation, it remains to realize a program decision that takes into account all the parameters involved. [5]
II.2 CHOICE OF PARAMETERS:
Our objective is the measure of the length of the slabs which vary between 4.5m and 9m, we must therefore make a rigorous choice of different parameters. Once the slab is extracted from the ingot mold, it moves on a roller table and the initial position of the holder carriage blowpipe will be taken as reference geometric landmark (rep1) for measurement. Once the slab reached the 2nd landmark (rep2) which corresponds to the desired length, cut order is given.
First, we need the relation between the real length of a slab and the number of pixels it occupies in the image [6]. Any object of length L that the camera captures is represented by its image length, such as:
d = n * e with
n: Number of pixels occupied by the image of the object.
e: Length of a pixel.
We assimilated the camera to a perfect lens and we use optical formulas [7].
Consider the following schema:
p : Distance between the lens and the visualized object.
p’ : Distance between the lens and the image of the object
f : Focal distance of the lens.
AB = L: Length of the object.
A’B’=d: Length of the image of the object.
The two formulas of the optic are:
1/p + 1/p’ = 1/f g = p’/p = d/L With g: enlarging.
As the value of the focal length is negligible compared to the distance between the objective and the object, we do the following approximation:
(f<<p) then (1/f >> 1/p) And so 1/p’ = 1/f
Then p’ = f and g = f/p =d/L And as d = n * e
We will have L = (d * p) / f = (n * e *p) / f Finally, we obtain the following relation:
L = (n * e *p) / f
II.2.a THE FIELD OF VISION
When installing the camera, it must satisfy the following condition:
The camera must visualize the arrival of all slabs and it is therefore necessary that the second landmark (rep2) always be
in the field of vision of the camera whatever the length of the slab. We take as optical landmark reference for the measurement the position of landmark rep3, at a distance of 4.25 m from the geometric landmark (See fig 2). As the length of the slab varies from 4.5m to 9m, we will have a field of vision (4.5m + s1) where s1 represents the safety margin that we take equal to 50cm. The field of vision will be 5m; we will denote it by Lmax:
Lmax = 5m
Fig2. The field of vision
II.2.b DISTANCE BETWEEN THE SLAB AND CAMERA The camera must be placed above the slab because of side the displacement of the torch carriage hides part of the slab and prevents the image analysis. We must foresee the installation of a bracket of top of the slab at a height p of 3m since we is limited by the support mannequin.
p = 3m
II.2.c LENGTH OF PIXEL
This parameter is fixed by the manufacturer:
e= 0,017mm
II.2.d THE FOCAL LENGTH AND THE CONFIGURATION OF THE IMAGE
These two parameters are linked by the following equation:
f1= ((e*p)/Lmax)*n1= (0,017*3)/5)*n1 then f1=10, 2*10^ (-3)*n1
As we have the PIP512/512 pixel, we take:
n1=512 pixels Then we obtain f1=5,22mm
IVC500 the camera does not possess this focal length and to increase it we take n2 = 1024
We obtain, f2=10, 44mm, as the nearest existing focal length to this value is 10mm, we opt for f3 = 10mm and we calculate n3:
n3 = (f3*Lmax)/(e*p) = 980pixels.
We deduce that our field of vision will occupy 980 pixels on the screen of the monitor.
Then the chosen parameters are:
Lmax = 5 m p = 3 m e = 0,017mm f = 10mm n = 980pixels
III. EXPERIMENTAL STUDY III.1. THE MEASURING PRINCIPLE
First, we must make a segmentation of the image. It is a study which is carried out beforehand. We take some pictures of the slab and we make the segmentation of these images with HIS.C program that we developed. This program consists in selecting a window in an image and read the value of the luminosity of each pixel. In our case we select the part of the image representing the slab and we deduce the field of the values [b1….b2] of the gray level of the slab. We do the same for the field of the values [v1….v2] of the train rollers. We take as reference landmark for measuring the initial position of
the torch holder. The field of vision of the camera is 5m ranging from repere3 corresponding to a distance of 4.25 m until the distance of 9.25m. On the screen, the origin n = 0 (optical landmark) corresponds to repere3 (see fig).
We have developed the measurement program 'Mesure.C' which consists in [8]:
- Calculate the abscissa nx of the landmark rep2, such as:
nx = ( f / (e * p)) * (Lx - 4.25) (See fig 3).
- Test the value of the gray level of the pixel of nx abscissa. If this value belongs to the field of the gray levels of the slab, a control signal is given to torches otherwise the test continues.
- Once the control signal is given, retest the gray level value of the same pixel. If this value belongs to the field of the gray levels of the train roller, the end of the slab is therefore detected; the program then waits for the arrival of the next slab.
Fig3. Visualization of the slab on the screen of the monitor
III.2 THE CRITERION OF DETECTION
The detection of the beginning and end of the slab based on the reading of the gray value of a single pixel would be the easiest solution, but it poses a problem of reliability. The sensor of the camera is sensitive to changes in light intensity and so if a passenger or light fugitive phenomenon happens it may change the value of the gray level of the pixel reading and would cause the false measure. So we must establish a detection criterion which lessens this risk.
Tests realized in the laboratory have allowed us to deduce a reliable criterion of detection and measure. It consists in reading the value of the gray level of 20 consecutive pixels [nx……. nx-19] and we can confirm the detection of the beginning or the end of slab if among the 20 values at least 15 belong to the field of gray level of the slab or the roller train.
III.3 CALIBRATION
Theoretically L and n are linked by the following formula:
n = (f * (e * p)) * L
In making some verification tests we noted that this relationship is not quite right, it is due to two reasons:
- We assume that the camera has a perfect lens for lack of knowledge of its internal schema.
- The values such as focal length and distance between the viewed object p are never fixed with precision because the adjustment of the focal length and the measure of p always has a minimal uncertainty.
To remedy to this problem, we must realize a calibration which allow us to compare the theoretical values with real and then we can deduce a correction factor.
The calibration consists in realize the following experience:
For a series of length values of slabs we associate a series of numbers representing the pixels. We studied three cases:
The first case f =20mm, p = 3m The second case f = 12.5mm, p =3m The third case f = 20mm, p = 2,5m
It is assumed that nth is the theoretical number of pixels and nreal the real number of pixels.
We have the following relationship:
nth = (f * L) / (e * p) In the first case where
f = 20 mm and p = 3 m, we obtain: nth = 0,392 L We calculate the average value (nreal / L)moy as:
(nreal / L)moy= ((nreal1 + nreal 2 +…….+ nreal16)/L)) / 16 = 0,5 nreal1, nreal 2……. nreal16 are the values of the number of pixels representatives of sixteen different lengths of slabs, see tab1.
We obtain this time around : (nreal / L)moy = 0, 5 Whence:
real =0,5*L = 0,5 * nth / 0,392 = 1,275 *nth then Correction factor g1 = 1,275
nreal(pixels) L(mm) n/L
28 50 0.530
52 100 0.520
77 150 0.513
102 200 0.510
127 250 0.508
151 300 0.503
175 350 0.500
199 400 0.497
224 450 0.497
248 500 0.496
274 550 0.498
294 600 0.496
321 650 0.493
345 700 0.492
369 750 0.492
393 800 0.491
Tab1
We process of the same way for the second and third cases, and we obtain the following correction factors:
g2 = 1.5 and g3 = 1.15
We notice that in each case the correction factor is different.
We can’t therefore have a general formula applicable to all cases. We must imperatively do a calibration after the choice of all the parameters and deduce the correction factor to each case.
IV.RESULTSANDDISCUSSIONS At the process laboratory we realized a simulation of the measure of the length of a slab. We used a brown object in the shape of a slab that we have displaced on a white background representing the roller train.
We took the parameters of the second case:
f = 12.5mm p = 3m e = 0,017mm g2 = 1.5
Then nreal = g2 * nth = 0.369 L
After fixing the camera in accordance with the chosen parameters, we started the acquisition of images in a
preselected window containing the slab and another containing the white background and we determined their field values of gray level. Then, we obtain:
For the slab [b1…b2] = [29…82]
For white background [v1…v2] = [220…254]
We have affected these values to the appropriate variables in developed program Mesure.C. This program allows the user to enter the value of the desired length of the slab as well as their number.
We simulate the arrival of a slab, and we made their measure according to the principle explained in. For several lengths
chosen beforehand, we obtained very reliable results of measure.
The simulation of the measure of the length of the slab has been successfully completed but for on-site measure, it is more advantageous to choose an industrial linear camera with a card processing integrated image 1024points because they present a better reading precision. Also, as this study is based on the reading of the gray levels values of the slab and roller train, the lighting conditions in the area of measure should be stable so that the field values of the gray levels of the slab and of the roller remains stable. We must ensure that the fields of the gray levels do not overlap to avoid bad interpretations either for the arrival or end of the slab.
V. CONCLUSION
In this project we presented the measure of the length of the slab by a system of artificial vision. In conclusion we can say that this system presents a better reliability of the measure and improves the working conditions of the workers.
It would also be possible to use this method for better optimization of the cut.
Note that the steps of segmentation of the image and
calibration are inevitable and must be done with rigor because they represent the basis of that type of measure. The stability of the lighting of the environment is also a factor directly affecting the reliability of the measure and this factor must be followed minutely.
REFERENCES
[1] Frederick Overman, “The Manufacture of Steel. Containing the Practice and Principles of Working and Making Steel”, 1-230p, Edition BiblioLife, 2009.
[2] Radu Horaud, Olivier Monga, “Vision par ordinateur: outils fondamentaux ”, 1-425p, Edition Hermès, 1995.
[3] Achille Braquelaire, “Méthodologie de la programmation en C”, 1-672p, Edition Dunod, 2005.
[4] J.-G.Postaire, “De l’image a la décision”,1-186p, Edition Dunod- Paris, 1987.
[5] Robert J. Schalkoff, “Digital image processing and computer vision”, 1-489p, Edition John Wiley & Sons Australia, 1989.
[6] Raphaële Langet, “Optique géométrique ”, 1-192p, Edition Nathan, 2008.
[7] Sylvain Houard, “Optique Une approche expérimentale et pratique”, 1-394p, Edition De boeck, 2011.
[8] Clovis Tondo, Scott E. Gimpel, Antoine Bertier, “Exercices corrigés sur le Langage C ”, 1-168p, Edition Dunod, 2007
