=Paper=
{{Paper
|id=Vol-1909/paper4
|storemode=property
|title=Developing of the System for the Filling Layer Simulation
|pdfUrl=https://ceur-ws.org/Vol-1909/paper4.pdf
|volume=Vol-1909
|authors=Vasiliy N. Kruglov,Andrey V. Chiryshev,Artem V. Kruglov
}}
==Developing of the System for the Filling Layer Simulation==
https://ceur-ws.org/Vol-1909/paper4.pdf
Developing of the System for the Filling Layer Simulation
Vasiliy N. Kruglov, Andrey V. Chiryshev, Artem V. Kruglov
Ural Federal University
Yekaterinburg, Russia, 620004
avkruglov@yandex.ru
Abstract
The paper is devoted to the issue of the simulation system development to
model filling layer of granulated material. The tools for the filling layer
simulation with particular force to green ore pellets are the utmost important
for mining processing plants. The software suits implemented for system
realization are described. The main features and options of the application
are reviewed. The algorithm for the filling layer model formation is
adduced. The results of the test show that the model of the granulated
material reflects the real physical object with high accuracy so it can be used
for its further investigation.
1 Introduction
One of the principal characteristic in production and processing of the non-coherent material fraction is its distribution
over size grade. Grain-size distribution assessment of the crushed, granulated or pelletized particles of non-coherent
material is significantly important for mining processing plants to control their technological process. In particular, grain-
size distribution of the green ore pellets is main characteristic of their quality. At the moment it is paid close attention to
the problem of green pellets geometry measuring [1-8]. Analysis of the bibliography shows that the advanced way to
measure this criterion belongs to the machine vision systems. This approach allows on-line controlling of the material
grain-size distribution in the workflow. Camera captures images of the top leaf of the granulated material. Then the
isolation of each visible particle and its linear size and volume measurement are implemented with help of digital image
processing methods. The main part of any contactless measurement system based on machine vision is an image
processing and analysis unit which includes an algorithm for the isolation and detection of granulated particles.
It is necessary to estimate the tolerance of the computation unit while developing and integration of such system. In
order to evaluate the precise of the algorithm’s grain-size distribution the screen analysis of the of the green ore pellets
test sample should be implemented [10, 11]. However, such a test of validity of computational data over against
experiment is highly laborious and time-consuming. Moreover, the correctness of the algorithm should be tested over the
huge amount of the top leaf images with the given distribution law of the non-coherent material particles. Also it is
preferably to have an opportunity to form images with arbitrary distribution law. That is why the simulation system for
modelling of the filling layer of spherical granulated particles is likely to be developed. Such a system will allow to form
images of the green pellets with target values of their grain-size distribution. The result of the modeling will be applied in
computation unit checkout and validation, performance and accuracy comparison of the different algorithms devoted to
granulated particles measurement.
2 Development of the Simulation System
Modern equipment and software allow to create a physical model of the huge amount of particles interaction with high
accuracy and realistic behavior. There are a number of companies who offer software and hardware to simulate
complicated physical processes. For example, GDI provides ROCKY software for 3D analysis of behavior of the non-
coherent material particles [12]; Pasimodo solution by Inpartik company is suitable for granular medium modeling [13].
Despite the wide latitude of these software packs they are not appropriate for development of the algorithms for machine
vision systems as they have no tools to simulate displacement , quantity or type of lighting sources in the scene.
The developed system is based on the PhysX middleware by NVidia which used in modern games and graphic
editors. Implementation of these middleware gave an opportunity to create effective framework for filling layer
simulation which combines high computation performance and accuracy of the geometry and physical compositions.
Interface DirectX is implemented to visualize simulation process. It is specified by the high image processing speed due
to v video adapter hardware acceleration. DirectX is one of the most popular tool for the graphics application
development. Combining PhysX and DirectX allows to compute the interaction of modeling objects with high accuracy
and display the results of the simulation in a clear view.
2.1 Principal features of the system
Simulation system has a wide range of input parameters which influence the simulating process. System operates with
conditional units so user determines the specific system of units for the parameters of the structure elements while
simulation. It is necessary to set the specific probability laws for distribution of the position and object size in each
instance of the simulation to obtain the most adequate model. For example, the position of the spherical particle in X-axis
direction could be set according to the uniform density
1 , x [a , b ]
bx a x x x
f ( x) (1)
0 , x [ a x , bx ]
where [a x , bx ] – variation interval of X coordinate, or the normal density
2
(x )
1 x
f ( x) e 2 (2)
2 2
x
where x – mean value, x – dispersion of X coordinate distribution. It is similar to Y coordinate .
The size of the generated spheres is set in conditional units of their diameters which can also be generated according
to the uniform, normal or even arbitrarily density. In the latter case the density of the diameter or gran-size distribution is
made as histogram of categorization. The developed system allows to create up to 9 categories. Within each category the
size of the sphere is set by normal density.
It is extremely important zone in machine vision system to arrange for the artificial lighting of the measuring in order
to develop effective algorithm for the isolation and detection of granulated particles. The developed software provide for
tools to configure the scene lighting. It assumes generating of the lighting sources and definition of their parameters.
Lights are also significant in the visualization of the filling layer simulation. They give an opportunity to set the required
contrast of the image, configure shadows and regulate luminance differences on the edge between the visible particles of
the filling layer. It is possible to generate up to 8 light sources. Each light is regulated by its position relative to the scene.
The light can modulate one of three types of the light flux: spot light, point light and directional light.
The result of the simulation is 3D scene of the location of the model mass of the spherical granulated particles in
respect to the static container. The images of observed scene are of the 768*576 pixels and .bmp format.
2.2 Model formation algorithm
The filling layer model construction is carried out as following.
1. Initialization. Initialization of the model means creation of the external environment – scene. It involves the
parameters required for simulation , environment settings and all objects which could be added to the model. What is
more, PhysX middleware and DirectX interface are also initialized while creation of the scene. It includes setting the
initial parameters and operation mode.
2. Object creation. Each object in the system is described by two features: actor and shape. The former includes
whole range of the physical characteristics that are necessary to simulate behavior of the object: size, density and
interaction forces. The latter is a representation of the object in the scene. The visual representation of the object is
created in parallel with the object itself and used through the whole simulation cycle.
3. The location of the object is set by two components: coordinates of the center of mass and the attitude
quaternion which describe the body rotation around the given 3D vector by the specific angle. Objects are generated in
the system throughout the simulation cycle.
4. Simulation time-step calculation. Simulation time-step is one of the most significant parameter that has an effect
on the model creation accuracy. The frequently used time-step is 1/25 second.
5. Simulation. At this stage PhysX calculates objects’ collisions within given amount of time and outputs their
coordinates by the end of the time-step.
6. Visualization. The shape of the each object is displayed by the DirectX interface. The location of the camera and
influence of the lighting sources are calculated while visualization process.
7. Model destruction. Stages 2-5 are repeated until the model will be stopped. After that all objects are deallocated.
System is ready to the new simulation cycle.
3 Results of the Modeling
Two auxiliary experiments were conducted in order to validate the system. Images of the monolayer of the particles were
generated (Fig. 1) with adjusting coefficient of 5 pix/mm.
Validity of sphere generating on the images was checked manually and by software-analytic approach. An amount of
totally visible spheres is 108 in Fig. 1a and 27 in Fig. 1b. The manual measurement of the spheres’ sizes was
implemented with help of Paint graphic editor over 70 spheres from Fig. 1a and 27 spheres from Fig. 1b. The results
confirmed the accuracy of the sphere generating. In order to fulfill the latter validation approach the algorithm for
measuring geometry of the granulated or pelletized particles in the filling layer was programmed [9]. Algorithm output is
presented in Fig. 2-3.
One object in Fig. 2a isolated incorrectly, however, its volume is quite small and the extent of an error is acceptable.
There are two detection incidents (at the edge of the frame) in Fig. 3a. Absolute error in this case is 1.7% due to the small
amount of the spheres in the image.
a) b)
Figure 1: Images of the monolayer of the spherical particle of the different size: a – 11 mm, b – 19 mm
100,0
100
80
60
%
40
20
0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0
0
+10-12
+12-14
+14-16
+16-18
+18-20
+20
+8-10
-5
a) +5-8 b)
Figure 2: Sphere isolation algorithm output for spheres of the 11 mm in diameter: a – display of the isolated spheres;
b – distribution histogram
98,3
100
80
60
%
40
20
0,0 0,0 0,0 0,6 1,1 0,0 0,0 0,0
0
+20
+10-12
+12-14
+14-16
+16-18
+18-20
-5
+8-10
+5-8
a) b)
Figure 3: Sphere isolation algorithm output for spheres of the 19 mm in diameter: a – display of the isolated spheres;
b – distribution histogram
a) b)
Figure 4: Image of the sample of green ore pellets (a) and simulated filling layer of the spherical particles with the same
grain-size distribution
100
80
60
40,2
%
40
19,3 22,1
20 6,4 9,5
0,0 0,7 1,8 0,0
0
+8-10
+20
+10-12
+12-14
+14-16
+16-18
+18-20
-5
+5-8
a) b)
Figure 5: Output of the algorithm for the isolation and detection of spherical particles with given distribution law:
a- display of isolated spheres; b- distribution histogram
In order to validate the generation of the filling layer of the spherical particles by the arbitrarily density the sample of
the green ore pellets were taken [10]. Image of this sample is presented in Fig. 4a. The results of the screen analysis [11]
of the sample are in the Table 1.
Table 1: Grain-size distribution in the sample of green ore pellets
Category, mm -5 +5-8 +8-10 +10-12 +12-14
Category mass percentage , % 0 0,2 3,9 24,7 39,9
Category, mm +14-16 +16-18 +18-20 +20
Category mass percentage , % 26,8 4,5 0 0
Obtained data on green ore pellet diameter distribution were input in modeling system as initial conditions. Image of
the filling layer simulated by the arbitrary density law introduced in Fig. 4b. This image was processed by algorithm for
particles geometry measurement [9]. Output of the algorithm is presented in Fig. 5.
Obtained results show that:
Mean absolute error of the grain-size distribution of the simulated filling layer of the spherical particles is 2.88% over
all categories.
Margin of absolute error in main category of the simulated particles (+ 12mm-14mm) is 0.3%.
4 Conclusion
Developed simulation system allows to create models of the filling layer of spherical particles with arbitrary
characteristics of grain-size distribution as well as display the results of the simulation. This tool is applicable to
validating algorithms for grain-size distribution measurement. Moreover, the developed software provides definition of
the spatial features of the non-coherent material. According to the result of the simulation, the model of the filling layer
reflects the real physical processes of the filling layer formation with high accuracy and could be applied for its further
investigation. In particular simulation results could be applied for experimental check of such properties of the simulated
structure as fractional void volume, closeness of packing, etc. Fractional void volume is a significant indicator for
measurement of the gas permeability of the green ore pellets that is argue problem for mining processing plants.
Developing of such system gives the balance of advantage. Firstly, it is an optimization of the gas flow rate required for
the roasting. Secondly, it is a reduction in expenses for production tests and laboratory data analysis of incoming pellet
flow. Finally, it is an opportunity to forecast the roasted pellets quality.
References
1. Jagdish Lal Raheja, Gaurav Sahu. Pellet Size Distribution Using Circular Hough Transform in Simulink, American
Journal of Signal Processing 2012, 2(6): 158-161.
2. Dhiraj, J.L. Raheja, Vaishali Singh, Sreekanth Pusapati, Ashutosh Gupta. Design and Implementation of Floating
Point based System for Pellet Size Distribution using Circular Hough Transform. International Journal of Computer
Applications. Volume 62– No.13, January 2013,p.9-15
3. Navdeep Singh, ShaliniSaxena, Dhiraj, J L Raheja. Industrial vision system for distribution and size analysis of
pellets. International Journal of Application or Innovation in Engineering & Management. Volume 2, Issue 3, March
2013 Page 7-13
4. J. Kaartinen and A. Tolonen. 2008. Utilizing 3D height measurement in particle size analysis. In: Myung Jin Chung,
Pradeep Misra, and Hyungbo Shim. Proceedings of the 17th IFAC World Congress (IFAC 2008). Seoul, Korea. 611
July
5. T. Andersson, M. J. Thurley, O. Marklund. Pellet Size Estimation Using Spherical Fitting. IEEE Instrumentation and
Measurement Technology Conference. Warsaw, Poland, May 1–3, 2007.
6. Wang Jie-sheng, Wang Jie-sheng. PCNN Edge Detection of Sintering Pellets Image Based on Hybrid Harmony
Search Algorithm, 2009 International Conference on Artificial Intelligence and Computational Intelligence, p. 535 –
539
7. Huanyu Liu, Lihong Xu, Dawei Li. Detection and recognition of uneaten fish food pellets in aquaculture using
image processing. Conference Paper in Proceedings of SPIE - The International Society for Optical Engineering ·
October 2014, http://spiedigitallibrary.org/
8. Sergey Kucheryavski, Kim H. Esbensen, Andrey Bogomolov. Monitoring of pellet coating process with image
analysis—a feasibility study. Journal_of_Chemometrics, 2010; 24: 472–480
9. V.N. Kruglov A method of determining geometry of the pelletized particles and/or granulated material in the filling
layer. Patent № 2557330, 2015
10. GOST 17495-80 Iron ores, concentrates, agglomerates and pellets. Methods of sampling and preparation of samples
for particle size analysis
11. GOST 27562-87 Iron ores, concentrates, agglomerates and pellets. Determination of size distribution by sieve
analysis
12. ROCKY - DEM Particle Simulator. URL: http://www.rocky-dem.com/
13. Inpartik Simulation Software & Engineering. URL: http://www.inpartik.de/