Sunteți pe pagina 1din 12

Developing Ideal Deoxidation Practices by Design of Experiment (DOE) Method in Steelmaking

Process

I.Keskin1, Y.Tetik2, Y.Kaçar3,


1
ERDEMIR, Project Manager/Researcher Steelmaking and Casting Technologies R&D
Uzunkum Cad. No: 7, P.O. Box 67330, Kdz. Ereğli, Zonguldak, Turkey
Phone: +903723298727
Email: ikeskin@erdemir.com.tr
2
ERDEMIR, Steel Making Process Engineer
Uzunkum Cad. No: 7, P.O. Box 67330, Kdz. Ereğli, Zonguldak, Turkey
Phone: +903723293052
Email: ytetik@erdemir.com.tr
3
ERDEMIR, R&D Manager Steelmaking and Casting Technologies
Uzunkum Cad. No: 7, P.O. Box 67330, Kdz. Ereğli, Zonguldak, Turkey
Phone: +903723298713
Email: ykacar@erdemir.com.tr

Keywords: Deoxidation Practice, Aluminium, Design of Experiment, Modeling, Cost Reduction.

ABSTRACT
Steel deoxidation is indispensable for the steelmaking process, in order to avoid steel defects such as inclusions and
blowholes forming during solidification. There are several deoxidation methods applied in steelmaking companies. The
primary targets of this work are increasing deoxidation efficiency, decreasing deoxidation agents and enhancing steel
cleanliness. In order to define the ideal deoxidation practice at the Erdemir steelmaking plant, the design of experiment
(DOE) method was used. By this model, the parameters affecting the final aluminum and steel cleanliness were determined.

INTRODUCTION
With the progressing technology, the steelmakers are striving to improve productivity and final steel quality. However, in
today’s world that phenomena becomes insufficient due to the survival conditions. Depending on the decrease of steel prices,
every steel producers are trying to reduce operational and material costs. Steel deoxidation process is one of the most critical
steps in liquid steel processing for target steel cleanliness and cost optimisation. During deoxidation and alloying practice, a
crucial challenge for the operators lies in minimizing the consumption of ferro-alloys so as to control the generation and
possibility of entrapment of non-metallic inclusions for the production of clean steel and to make the process economically
attractive. To control deoxidation process, there are several modelling and optimising studies; Kamaraj and his friends have
studied theoratically predicting the amount of deoxidizers required for a given steel composition by mathematical model. [1]
Harald and friends have focused on comprehensive parameter study which investigates alloy type, size, addition time, and
other parameters for simulating the alloying process by CFD method. [2]

DEOXIDATION FUNDAMENTALS
The aim of ideal deoxidation practices is preventing the CO gas formation during casting by binding free oxygen in liquid
steel and removing oxide inclusions which are formed in deoxidation during tapping and increasing the efficiency of
deoxidant materials. Generally three main deoxidation methods have been used during tapping the liquid steel; none - killed
steels, semi - killed steels and fully - killed steels. The none –killed (rimming) steels and semi – killed steels are not suitable

AISTech 2016 Proceedings. © 2016 by AIST. 921


for continuous process and primary aim is to produce Mn-Al-silicate inclusions which are deformable in hot rolling, and to
avoid hard crystalline inclusions (Al2O3, MgO·Al2O3) however, big majority of current steelmaking aims at complete
“killing,” i.e., deoxidized steels in which no CO evolution is possible [3]. The general equations for the reaction are defined
as follows;
x Me yO MexOy (1)
The equilibrium constant for equation (1) is:

K (2)



G° = Free energy change for deoxidation reaction


[] = refers to dissolved elements in liquid Fe


a[Me] = activity of the metallic elements (a[Me] = fMe * %[Me]


fMe, fO is the activity coefficients of elements relative to the 1 wt% standart state Fe
aMexOy = activity of deoxidation product generally taken as 1 in solid state
If deoxdiation product is taken as (aMexOy) = 1:

a (5)

For the fix on steel composition at a spesific temperature, fMe , fO are constants and the equilibrium constant for each oxide
–forming elements can be calculated. Holappa has calculated the most common deoxidation reactions with their equilibrium
constants and their values are given in table 1 [3].

Table 1: Deoxidation Reactions, Equilibrium Constants, Their Values, and Solubility Products

By applying the equations and their numerical values in Table 2, the deoxidation diagram showing oxygen activity as a
function of the activity of the added element can be constructed. Such an activity diagram is presented in Figure 1.

922 AISTech 2016 Proceedings. © 2016 by AIST.


Figure 1: Deoxidation equilibria in iron at 1873 K (1600 C0). [3]
According to figure 1, deoxidation power of elements can be ordered as Cr-Si-B-Ti-Al-Mg-Ca-Zr-Ce. The commercial
deoxidants are commonly used as ferroalloys and aluminum.

DATA ANALYSES AND EXPERIMENTAL PROCEDURE


The deoxidation practice during tapping is not uniform in the different steelmaking plants. The practice can be differentiated
as each steelmakers strategy about steel cleanliness, cost optimization etc. In Erdemir steel making plant, two main
deoxidation practice are applied, none - killed practice for IF grades and full - killed practice for the other grades.
First of all, in the scope of this project the deoxidation efficiency of low carbon steel grades (Not included Mn, Si) were
examined and studied. In the same steel quality, although the all deoxidant additions (FeMn, Al, Coke) and blow-end
oxygens were same amount, the LF_1_Al measurements showed fluctuation (Table 2). Although the aluminum target of the
tapped steel was between 0.02 wt% - 0.03wt % the measurements showed alteration between 0.01 wt% - 0.06wt %. It is well
known fact that larger alumina inclusions are growth in more oxidizing conditions. Yang and friends stated that during the
different stage of steelmaking process, the inclusion removal rate can be changed according to inclusion size [4]. Another
investigation shows that larger inclusions can be removed from the molten steel by fluid flow transport and by attaching to
the bubble surface however, small inclusions stay suspended in the liquid steel and are passed on to the next process, while
others are removed by the top slag and refractory walls by diffusion deoxidation and interfacial reactions [5]. An inefficient
deoxidation practice can be challenging for ladle furnace operators due to several reasons such as time restriction, Al loss etc.
Table 2: Deoxidation process parameters and results in same steel grade during tapping liquid steel. LF_1_Al: Ladle furnace
initial measured soluble aluminum (%), ENB O: End blow oxygen measurement by sublance, LF_1_Mn: Ladle furnace
initial measured Manganese (%), Aluminum Addition: Amount of Aluminum nugget added during the tapping of steel, Coke
Addition: Amount of metallurgical coke, added during the tapping of steel, FeMn Addition: Amount of ferromanganese,
added during the tapping of steel, ENB C: End blow carbon measurement by sublance, C*O: End blow carbon oxygen
product.

AISTech 2016 Proceedings. © 2016 by AIST. 923


Steel Coke Al FeMn
Steel ENB O
Weight ENB C (%) C*O LF_1_Mn LF_1_Al Addition Addition Addition
Grade PPM (%)
(tons) (Kg) (Kg) (Kg)
LC 121,2 906 0,0351 32 0,1087 0,0236 0 240 0
LC 119,8 881 0,0334 29 0,1648 0,0181 0 240 0
LC 121,2 906 0,0293 27 0,1286 0,0382 0 240 0
LC 122 902 0,0364 33 0,1282 0,0405 0 240 0
LC 121 897 0,0353 32 0,1958 0,0402 0 240 140
LC 120,7 920 0,0235 22 0,1598 0,0263 0 240 122
LC 118,1 902 0,0212 19 0,1297 0,0295 0 240 123
LC 123,225 907 0,041 37 0,1703 0,0198 0 240 132
LC 121,7 898 0,0391 35 0,1478 0,0495 0 240 135
LC 119,3 889 0,0233 21 0,1555 0,0498 0 240 137
LC 118,2 881 0,036 32 0,1703 0,021 0 240 140
LC 118,8 902 0,0312 28 0,1483 0,0349 0 240 0
LC 112,5 902 0,0354 32 0,1959 0,0363 0 240 141
LC 122,042 917 0,0273 25 0,1851 0,0429 0 240 135
LC 120,9 906 0,0307 28 0,1784 0,0342 0 240 102
LC 115,6 920 0,0275 25 0,1782 0,0334 0 240 89
LC 121,4 904 0,0205 19 0,1868 0,0325 0 240 111
LC 120,8 910 0,0262 24 0,1629 0,0172 0 240 109

When looked the distribution of measured soluble aluminum in the ladle furnace, the deoxidation process capability was
calculated 26%. In other words 17% of heats were measured less than lower limit and 53% of heats were measured more than
upper limit. For the ladle furnace practice, it is known and agreed on that the heats more than upper limit are not complicated
as much as the others.

Distribution of LF First Measured Soluble Al


25

21,7391

20
17,3913

15
Percent

13,0435 13,0435 13,0435

10,8696

10

6,52174

5 4,34783

0
0,010 0,015 0,020 0,025 0,030 0,035 0,040 0,045 0,050
LF_1_AL

Figure 2: The distribution of first measured soluble aluminum in ladle furnace.


After analyzing the process data, the factors affecting the deoxidation efficiency are defined as;


Amount of aluminum addition in tapping,


The amount of coke addition for pre – deoxidation,


The amount of carry over slag,
Aluminum addition time.

924 AISTech 2016 Proceedings. © 2016 by AIST.


In order to control deoxidation process and increasing the process capability, the trial procedure was prepared via design of
experiment method.
The amount of aluminum addition: During tapping the amount of aluminum addition depends on the end blow oxygen and
the amount of coke added for pre – deoxidation. For specified steel grades several process data was examined and the linear
regression model of aluminum amount was generated. The R-sq of regression model was calculated 41,19 % and P- Value is
0,000 , Figure 3.

Figure 3: The regression model for aluminum addition during tapping of liquid steel.
On the purpose of investigating the all possibilities in deoxidation process, the upper and lower limit of the aluminum
addition regression model tolerance was determined as 20 kg interval.
The amount of coke addition: For the reduction in the consumption of deoxidant materials (aluminum, ferroalloys), the
metallurgical coke was utilized in the beginning of steel tapping. For specified steel grades the amount of coke addition can
be changed according to end blow carbon measurement. The C*O product of each converter can be altered with the bottom
stirring of them. Although for spesific steel grade according to target carbon level the metallurgical coke addition regression
model was calculated in plant metallurgical coke can be added the amount of 20 kg bag. Due to this reason for design of
experiment model, the coke addition amounts were defined 0 and 20 kg for each heat.
The carry over slag amount: In steelmaking plant during the tapping of liquid slag in order to preventing slag carry over,
infrared thermal camera and slag hammer system is used. On the results of former trials, the triggering limit of thermal
camera was 30%. Aydemir has tried to calculate carry over slag during tapping of liquid steel in his master thesis and
reported that it can be changed between 264 kg and 1148 kg [6]. For design of experiment model, the tolerance limits of
infrared camera were defined 10% and 30% kg for each heat. With that parameter, the effects of thermal camera tolerance
were investigated and analyzed.
Aluminum addition time: In current situation the aluminum addition time is defined as the 70 % of tapping duration. One of
the operators controls the flow and the height of the steel in ladle then he decides the optimum addition time. However, there
are some difficulties complicating the optimum addition time; occurred dusts, flames and mists in the air can restrict the
operator’s sight. In order to solving that problem, the tapping duration calculation model was calculated as shown in Figure 3.
In the calculation model the parameters affecting the tapping duration was investigated and the tapping hole life and slagging
of tapping hole were determined the most efficient factors for that model.

AISTech 2016 Proceedings. © 2016 by AIST. 925


Figure 4: The regression model for tapping duration calculation.
After composing the calculation model with high correlation factor (R-Sq 82.98%), the model was adapted to tapping model
of the converters. With that improvement, level 2 of the converters can calculate the tapping duration and send a signal to the
special blinker which was installed close to aluminum addition area. According to literature, there are many opinions about
addition time of aluminum. Due to increasing nitrogen pick up, and increasing the ferroalloys yield, some steel companies
add the aluminum in the beginning of tapping. On the other hand some steel companies add in the last quarter for the reason
of increasing aluminum yield [7]. Finally, the aluminum addition times were decided as 50% of tapping time and 70% of
tapping time.

Scatter Plot of Tapping Duration versus Tap Hole Life


10 Slagging
Yes
No
9
Tapping Durat ion ( Min)

0 20 40 60 80 100 120
Tap Hole Life

Figure 5: The plot of tapping duration versus tap hole life. Tap hole life: The heat number of tap hole
The next step is designing of trials in order to control and increase the deoxidation process during tapping of liquid steel. As
seen in figure 6, the efficient parameters and change intervals were defined respectively;

926 AISTech 2016 Proceedings. © 2016 by AIST.


Figure 6: The trial factors and levels which would be used in DOE method.

Plant Trials by DOE Method: A “full factorial” design which studies the response of every combination of factors and
factor levels was composed. As seen in figure 6, the controllable input factors were determined and each factor is affecting
the output in two levels. An example trials list is given in table 3, for the reliability of DOE the trials were repeated for two
times. As an uncontrollable parameter the carbon oxygen product was evaluated.

Table 3: The trial results of heats which DOE method was applied.
Thermal Addition
Coke
Al Addition Cam. Time
StdOrder RunOrder CenterPt Blocks Addition LF_1_AL (%) C*O
(kg) Tolerance (Percent of
(kg)
(%) Heat)
1 1 1 1 132 0 10 50 0,012 34,077
2 2 1 1 172 0 10 50 0,019 28,726
3 3 1 1 132 20 10 50 0,025 16,72
4 4 1 1 172 20 10 50 0,021 15,77
5 5 1 1 132 0 30 50 0,0052 16,434
6 6 1 1 172 0 30 50 0,025 19,98
7 7 1 1 132 20 30 50 0,022 15,531
8 8 1 1 172 20 30 50 0,034 17,82
9 9 1 1 132 0 10 70 0,019 14,555
10 10 1 1 172 0 10 70 0,05 16,469
11 11 1 1 132 20 10 70 0,026 11,571
12 12 1 1 172 20 10 70 0,027 15,8
13 13 1 1 132 0 30 70 0,002 13,528
14 14 1 1 172 0 30 70 0,034 17,316
15 15 1 1 132 20 30 70 0,034 28,743

For the selected specific steel grades, the trial heats were planned and observed continuously, there were tremendous factors
affecting the design of experiment procedure such as granulated aluminum addition before the celox measurement in ladle
furnace and several sublance measurement problems before aluminum addition. Those heats were eliminated so as to prevent
data conflicts. During the converter process after the sublance measurement and level 2 model results, the operator defines
the aluminum and coke additions according to DOE model furthermore for optimum addition time with the tapping duration
calculation model level 2 sends a signal to co-operator for adding the additives. When ladle arrived in LF, after stirring few
minutes, the operator takes a celox measurement as soon as possible without adding any materials. Consequently the ladle
furnace first steel sample was taken into account for calculation of factorial analyses.

RESULTS AND VALIDATION

AISTech 2016 Proceedings. © 2016 by AIST. 927


With factorial analyses menu in statistical analyzing program (MINITABR) the main effects and interaction effects were
defined separately. Main effects are the effects when there exists a consistent trend among the different levels of a factor, and
interaction effects are the effects of an independent variable on a dependent variable averaging across the levels of any other
independent variables [8]. After calculating regression model for LF_1_Al, the analyses of variance table is shown in figure
7. Analysis of variance (ANOVA) technique was used to check the adequacy of the developed empirical relationship. In this
investigation, the desired level of confidence was considered to be 95 %.

Figure 7: Factorial regression analyses and results of DOE model.


The p values or probability values show the level of significance of each factor. Lower p values indicate that the factor values
have higher probability of falling within the ranges which impact the outcome of the experiment. As shown in figure 7, all
main effects are efficient on outputs also three interaction effect (Al addition * Thermal Cam. Tol.* Addition Time) is
efficient for the process output. Due to this reason in the two interaction effects which have high p value cannot be eliminated
fit regression model. According to factorial analyses the stronger R-sq value means that 87,88% of the variation in output is
explained by defined input values. The regression model for calculating the target aluminum level is explained in equation 6.

LF_1_Al = + 0,430 - 0,000575 C*O - 0,002774*Al Addition - 0,01395*Coke addition *


0,01475 Thermal Cam. Tol. - 0,00799*Addition Time + 0,000089*Al Addition*Coke
addition+ 0,000095*AlAddition*ThermalCam.Tol.+ 0,000057*AlAddition*AdditionTime+ 0,00 (6)
0057*Cokeaddition*ThermalCam.Tol.+ 0,000268*Coke addition*Addition Time
+ 0,000235*ThermalCam.Tol.*AdditionTime- 0,000002*AlAddition*Coke addition*Addition
Time- 0,000002*Al Addition*Thermal Cam. Tol.*Addition Time

Pareto analyses was used visually identify the important effects and compare the relative magnitude of the various effects. As
seen in figure 8, the input effects A, BC, D, ABD, AB, ACD, B pareto bars are to the right of the vertical red line; therefore,
these bars represent statistically significant at the 5% level of significance. Although factor C seems insignificant, we could
not drop this term in the model due to the interactive effects of it. It was observed that the aluminum addition amount and
addition time of aluminum were the most efficient factors, proving the reasons for variability. The positive effects of coke
pre-deoxidation process stated in several studies but the interaction with thermal camera tolerance will be investigated in
future works separately. [9]

928 AISTech 2016 Proceedings. © 2016 by AIST.


Pareto Chart of the Standardized Effects
(response is LF_1_Al; = 0,05)
Term 2,120
A
BC
D
ABD
AB
ACD
B
BD
C
BCD
CD
ABCD
ABC
AD
AC

0 1 2 3 4 5 6
Standardized Effect

Figure 8: Standardized Pareto Chart, showing the effects of the independent factors on the final objective. A: Aluminum
addition, B: Coke Addition, C: Thermal Cam. Tol., D: Addition Time
With the pie chart graph each category proportion in the data is represented. The pie was divided into slices, with each slice
representing a category of data. By comparing and contrasting the size of the slices, the relative magnitude of each category
was assessed. According to pie chart of data aluminum addition input factor affects the output with the ratio of 30,.1% then
the second efficient factor was the interacted effect of coke addition and thermal cam tolerance.

Pie Chart of Input Factors


Input Factors
Covariates
0,0% 5,0% C*O
11,1% Al Addition
5,0%
Coke Addition
Thermal Cam. Tol.
5,2% Addition Time
Al addition*Coke addition
Al addition * Adding time
Coke addition*Thermal Cam. Tol.
6,3%
Coke addition*Adding time
0,2% Thermal Cam. Tol.*Addition time
0,3% Al addition*Coke addition*Addition time
30,1% Al addition*Thermal Cam.Tol.*Addition time
Error

14,3%

0,9%
7,9% 2,2%
2,1%
9,2%

Figure 8: Pie chart of input factors after factorial analyses.


As a next step after the mathematical regression model calculated; performing, analyzing and evaluating validation trials
were planned. Trials were conducted to test performance of the new DOE model with respect to optimizing the aluminum
consumption and reducing process variability. Since it was not possible to change the existing on-line equations (for
calculating deoxidation to be added for various conditions) in the Level 1 system during the trial period, a look-up table was
created for the operators for adopting the calculation of the new DOE model.

Validations via FactsageR Model: Validations via thermodynamic calculation software FactSage 6.4: According to design
of experiment model results, there is not seen remarkable effect of thermal camera. Due to control and strength that results,

AISTech 2016 Proceedings. © 2016 by AIST. 929


the thermodynamic analyses are applied. The databases used in this study were: FSstel (compound and solution database for
steel), FactPS (gas species, solid and liquid compound database) and FToxid (compounds and solutions for oxide databases)
with “Equilib” module of FactSage 6.4. In Factsage thermodynamic model the trial inputs and outputs are calibrated and for
each heats during the tapping condition carry-over slag is calculated. As seen in table 3, the carry-over slag amount is
calculated regardless of thermal camera tolerance. The authors think that the reason for this in efficiency variance of thermal
camera tolerance between 10% and 30% is split second. Slag detection system is permanently monitors the tapping stream
by means of thermal camera and special image-processing algorithm continuously evaluates and archives the generated
image. As soon as the slag amounts in the generated area reach in tolerated level the system sends signal to initiate slag
stopper hammer.

Table 3: Factsage modelling of carry-over slag for selected trial heats

Calculated
Thermal Addition ENB O
Grade LF_1_Al Carry-over
Cam.Tol. Time (PPM)
Slag(Kg)

LC 10 0,5 440 0,025 400


LC 10 0,5 415 0,021 490
LC 10 0,7 609 0,026 255
LC 10 0,5 416 0,01 610
LC 10 0,5 485 0,022 650
LC 10 0,7 649 0,014 400
LC 10 0,7 872 0,027 380
LC 30 0,5 660 0,034 440
LC 30 0,7 730 0,032 410
LC 30 0,5 1045 0,002 530
LC 30 0,5 648 0,039 240
LC 30 0,7 593 0,02 670

CONCLUSIONS
The design of experiment (DOE) method was used in order to define the ideal deoxidation practice at the Erdemir
steelmaking plant and the parameters affecting the final aluminum and steel cleanliness were determined. The following
results were obtained;
(1) An amount of aluminum addition in tapping, the amount of coke addition for pre – deoxidation, the amount of carry
over slag and aluminum addition time are defined as factors which are affecting deoxidation process efficiency.
(2) As an input effects; aluminum addition, coke addition and aluminum addition time and their interacted effects are
observed statistically significant at the 5% level of significance.
(3) The mathematical regression model for calculating the target aluminum level was set and applied for validation
during an accounting period.
(4) The heats below the target level of aluminum in LF measurement decreased from the ratio of 28% to 18%.
(5) The heats in the target level of aluminum in LF measurement increased from the ratio of 53% to 64%.
(6) Aluminum consumption amount for selected steel grades was decreased from the level of 195 kg/heat to 190
kg/heat.
(7) Via using the DOE model the average of LF_1_Al measurements is 314 ppm however with the normal practice the
average is 277 ppm. According to calculations the average aluminum content of steel in the beginning of ladle
furnace is 5 kg more than general deoxidation practice.
(8) The aluminum consumption saving is 0,.083 kg/ton liquid steel.

930 AISTech 2016 Proceedings. © 2016 by AIST.


Histogram of LF_1_Al General Heats/Trial (DOE) Heats
0,000 0,015 0,030 0,045 0,060 0,075
LF_1_Al_General LF_1_Al_Doe LF_1_Al_General
40 40 Mean 0,02777
36,3636 StDev 0,01675
N 343
32,3615
LF_1_Al_Doe
30 30 Mean 0,03143
27,6968 27,2727 StDev 0,01730
N 44
Percent

21,8659

20 20 18,1818

14,8688
13,6364

10 10

4,54545
3,207

0 0
0,000 0,015 0,030 0,045 0,060 0,075

Figure 9: The histogram of LF_1_Al for the heats with DOE model and general application.

(9) With optimizing the addition time and efficient utilization of deoxidants the average of LF_1_Al measurements
were increased from 0,.0277% to 0,.0314%.

REFERENCES
1. Ashok, K., Mandal, GK., Bandyopadhyay, D., Theoretical Investigation on Deoxidation of Liquid Steel for Fe–Al–
Si–O System, Transactions of the Indian Institute of Metals 68 (1), 9-18, 2015
2. Laux, H., Johansen, S. T., Berg, H. and Klevan, O. S., CFD Analysıs Of The Turbulent Flow In Ladels And The
Alloyıng Process Durıng Tappıng Of Steel Furnaces, 2000.
3. Holappa, L., Secondary Steelmaking, Treatise on Process Metallurgy: Industrial Processes Volume 3. P. 301–345,
2014.
4. Yang, W., Wang, X., Zhang, L., Cleanliness of Low Carbon Aluminum-Killed Steels during Secondary Refining
Processes, Steel Research International, 2013.
5. Zhang, L., Thomas, BG., State of the art in the control of inclusions during steel ingot casting, Metallurgıcal And
Materıals Transactıons B, Volume 37b, 2006.
6. Aydemir, O., Use Of Alumınıum Dross For Slag Treatment In Secondary Steelmakıng To Decrease Amount Of
Reducıble Oxıdes In Ladle Furnace, Master thesis, Middle East Technical University, 2007.
7. Lehmann, J., Unamuna, I., Senk, D., Diaz, A.M., Di Donato, A., Deoxidation practice and slag ability to trapnon-
metallic inclusions and their influence on castability and steel cleanliness, Technical Steel Research, European
Commissions, 2008
8. Beins, B. (2006). Research Methods (PAYC 308) Fall 2006. Retrieved June 10, 2007 from
http://www.ithaca.edu/beins/methods/statistics/cogdis-interaction.htm#maineffects
9. Kloppers, C., Fedotova, T., Primary De-Oxidation Of Basic Oxygen Furnace Steel By Means Of Carbon, 2001.

AISTech 2016 Proceedings. © 2016 by AIST. 931


932 AISTech 2016 Proceedings. © 2016 by AIST.

S-ar putea să vă placă și