See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/304661868

Optimization of Sinter Plant Operating Conditions Using Advanced

Multivariate Statistics: Intelligent Data Processing

Article in JOM: the journal of the Minerals, Metals & Materials Society · June 2016
DOI: 10.1007/s11837-016-2002-2

CITATIONS READS

21 1,075

6 authors, including:

Daniel Fernández González I. Ruiz-Bustinza

Spanish National Research Council Universidad Politécnica de Madrid
129 PUBLICATIONS 1,021 CITATIONS 62 PUBLICATIONS 918 CITATIONS

SEE PROFILE SEE PROFILE

Carmen González-Gasca Luis Felipe Verdeja

Universidad Europea University of Oviedo
18 PUBLICATIONS 596 CITATIONS 339 PUBLICATIONS 1,633 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Daniel Fernández González on 23 May 2018.

The user has requested enhancement of the downloaded file.

JOM, Vol. 68, No. 8, 2016
DOI: 10.1007/s11837-016-2002-2
Ó 2016 The Minerals, Metals & Materials Society

Optimization of Sinter Plant Operating Conditions Using

Advanced Multivariate Statistics: Intelligent Data Processing

DANIEL FERNÁNDEZ-GONZÁLEZ,1,4 RAMÓN MARTÍN-DUARTE,2,5

ÍÑIGO RUIZ-BUSTINZA,2,6 JAVIER MOCHÓN,2,7
CARMEN GONZÁLEZ-GASCA,3,8 and LUIS FELIPE VERDEJA1,9

1.—Research Group in Siderurgy, Metallurgy and Materials, Department of Materials Science

and Metallurgical Engineering, Oviedo School of Mines, University of Oviedo, Independencia
Street, 13, 33004 Oviedo, Asturias, Spain. 2.—Research Group in Processes, Materials and Energy
in Ecological and Sustainable Metallurgy (PROMESS), Department of Primary Metallurgy and
Recycling, National Centre of Metallurgical Researches, National Council of Scientific Researches
(CENIM-CSIC), Gregorio del Amo Avenue, 8, 28001 Madrid, Spain. 3.—Universidad Europea
– Laureate International Universities, Calle Tajo s/n, Villaviciosa de Odón, 28670 Madrid, Spain.
4.—e-mail: fernandezgdaniel@uniovi.es. 5.—e-mail: ramon.martinduarte@gmail.com. 6.—e-mail:
irbustinza@cenim.csic.es. 7.—e-mail: jmochon@cenim.csic.es. 8.—e-mail: mcarmen.gonzalez@uem.es.
9.—e-mail: lfv@uniovi.es

Blast furnace operators expect to get sinter with homogenous and regular
properties (chemical and mechanical), necessary to ensure regular blast fur-
nace operation. Blends for sintering also include several iron by-products and
other wastes that are obtained in different processes inside the steelworks.
Due to their source, the availability of such materials is not always consistent,
but their total production should be consumed in the sintering process, to both
save money and recycle wastes. The main scope of this paper is to obtain the
least expensive iron ore blend for the sintering process, which will provide
suitable chemical and mechanical features for the homogeneous and regular
operation of the blast furnace. The systematic use of statistical tools was
employed to analyze historical data, including linear and partial correlations
applied to the data and fuzzy clustering based on the Sugeno Fuzzy Inference
System to establish relationships among the available variables.

On the other hand, quality requirements can be

INTRODUCTION
modified inside certain limits without impairing
The optimization of ore mixtures that are blast furnace performance.
employed in the sintering process plays a key role The main objective of this research was to obtain
in sinter quality and production levels. In the the least expensive iron ore blend for the sintering
sintering process,1–4 the physical, chemical, and process (with a more effective use of raw materials
mineralogical properties of raw materials have a and energy), with homogenous properties and with
relevant influence that should be considered, along suitable chemical and mechanical properties (to
with their availability and cost. ensure a soft and regular blast furnace performance).
Sinter productivity and quality requirements are Blends for sintering include several iron by-
a function of the blast furnace requirements,5–14 products and other wastes (mill scale, dust catcher
which impose restrictions on sinter features such as powder, etc.) that are obtained in ironmaking and
basicity, reducibility, strength, etc. steelmaking factories. The supply of these materials
In daily plant operations, it is necessary to is not always consistent, but they should be con-
address the actual availability of raw materials sumed in the sintering process (to save money and
and requirements for the blast furnace to reach an recycle wastes). These materials change the chem-
optimal performance point. Iron ore features do not ical and mechanical properties of the blends, so they
always fit precise values; they may change for the must be considered as restrictions in the optimiza-
same ore deposit and even for the same ore stock.15 tion of an iron ore blend.

(Published online June 30, 2016) 2089

2090 Fernández-González, Martı́n-Duarte, Ruiz-Bustinza, Mochón, González-Gasca, and Verdeja

METHODOLOGY The advantage of using the subtractive clustering

algorithm is that the number of clusters does not
Optimization techniques are used to find a set of
need to be a priori specified; instead, the method can
design parameters that can be defined as optimal be used to determine the number of clusters and
in a particular way (maximization or minimization
their values.
of a certain characteristic subjected to
The Subtractive Clustering method22,23 assumes
restrictions).16 that each data point zj ¼ ðxj ; yj Þ has assigned a
An efficient and accurate solution to this problem
potential, Pj , according to its location to all other
is not only dependent on the size of the problems in
data points. The potential, Pi , at data point xi is
terms of the number of constraints and design
variables but also on the characteristics of the defined as:
objective function and constraints. For the present
case, the optimization problem to be tackled should   !
X
n xi  xj 2
be considered as a nonlinear programing prob- Pi ¼ exp  ð2Þ
lem,17,18 particularly because some of the con- i¼1 ðra =2Þ2
straints involved are nonlinear functions. A
solution to a nonlinear programing problem gener- where Pi is the potential-value i-data as a cluster
ally requires an iterative procedure to establish a center; ra is a positive constant called cluster radius
search direction at each major iteration, which was (a neighbor radius); and x is the data point. Hence,
the approach to the problem that was used in this the potential of a data point to be a cluster center is
work. higher when there are more neighboring data
points. The data point with the highest potential
(Pk ) is considered as the first cluster center (xc1 ).
Methods of Data Analysis: Subtractive
Clustering The potential is then recalculated for all other
points (xi ) excluding the influence of the first cluster
The systematic use of statistical tools was center as follows:
employed to analyze historical data. These tools
included linear correlation and partial correlation !
applied to the data and clustering based on the kxi  xc1 k2
Sugeno Fuzzy Inference System (FIS)19–21 to estab- Pi ¼ Pi  Pk  exp  ð3Þ
ðrb =2Þ2
lish relationships among the available variables
(121,15). Sugeno method of fuzzy inference (also where rb is a positive constant which defines a
known as Takagi–Sugeno–Kang) was introduced in neighborhood that has measurable reductions in
1985.15 It was developed as systematic approach to potential value. After revising the potential value,
generating fuzzy rules from a given input–output the next cluster center is selected as the point
dataset. A rule in a Sugeno fuzzy model is:6,15 having the greatest potential value. The process
continues until a sufficient number of clusters are
if x1 is A and x2 is B then z ¼ f ðx1 ; x2 Þ ð1Þ defined.
The generation of the FIS matrix, which is a
where A and B are fuzzy sets in the antecedent, MATLAB object that contains all the FIS informa-
while z ¼ f ðx1 ; x2 Þ is a numerical function in the tion (including variable names, membership func-
consequent [typically f ðx1 ; x2 Þ is a polynomial func- tion definitions, etc.),24 is accomplished through
tion of the entry variable]. The Sugeno Inference previous training, which is conducted prior to
System was used because it is effective for opti- building this matrix. Relationships between inputs
mization problems and is computationally and outputs a priori are searched, and the data with
efficient.6 similar behavior are clustered, such that the num-
Once main relationships among iron ore and ber of rules has been reduced, being equal to the
sinter were established, a non-linear optimization number of clusters. Therefore, the FIS matrix has
algorithm based on step-descent methods was an equal number of membership functions for each
developed. Applied restrictions were related with input, as clusters have been found.25,26
the conditions imposed on the sinter properties and The best cluster parameters for this system were
the use of iron ore and by-products. obtained using the trial and error method. Thus, the
The clustering of numerical data forms the basis cluster radius27 that was utilized is a function of the
of many classification and system modeling algo- amount of variables that were used. Higher number
rithms. The purpose of clustering is to identify of variables resulted in longer time required in the
natural groupings of data from a large dataset to calculus process. All processes (data introduction in
produce a concise representation of a system’s the software, the estimation process and the results
behavior,6 which can use the cluster information presentation) should not be longer than 60 min
to generate a Sugeno-type FIS that best models (minimum time required for the iron ore blend for
the data behavior using a minimum number of reaching the blast furnace hopper feeders). Longer
rules. calculus times (higher number of variables) suppose
Optimization of Sinter Plant Operating Conditions Using Advanced Multivariate Statistics: 2091
Intelligent Data Processing

a better accuracy, although the problem would be with the sinter strand bed height and positive
that the estimated values could not be relevant in the correlation with the sinter strand speed. Moreover,
factory control (because the results would be obtained it has a negative correlation with coke and the total
once the sinter was fed to the blast furnace), and the amount of Fe that is present in sinter.
quality control of the sinter that feeds the blast For CaO, there is not a significant correlation
furnace would definitely not be achieved. between ore blend and sinter content, but a correla-
tion of 0.5 is found between CaO sinter content and
Optimization of Ore Blends for Sintering the additions of limestone and lime employed. MgO
content in sinter present a reasonable ( 0:55) cor-
Data Collection relation with MgO content in ore blends and with the
Two different sets of data were collected from a additions of dunite. Al2O3 content in sinter is corre-
European sinter factory plant. The first set covers a lated with coke, which provides an appreciable
period of 6 years and a total amount of 216 stock- amount of Al2O3 to the sinter, and with its initial
piles (15,000–30,000 tons of mineral). amount in the ore blend. No correlations are found for
A second set of data was collected during the early SiO2, possibly because there is not significant varia-
months of the project, covering a period of tion of SiO2 content during the analyzed period.
10 months and a total amount of 35 ore stockpiles. Tumbler index [ISO 3271 (2015)] is a sinter quality
To establish suitable relationships among iron ore index used in the ironmaking industry that provides
and the chemical properties of blends, the European a measure of the resistance of iron oxides to breakage
factory also supplied the results of the standard or degradation by impact and abrasion.12 The Tum-
chemical analysis that are employed on the raw bler index (%>6.3 mm) is positively correlated with
materials, which are used as other variables in the the bed height and negatively correlated with strand
optimization process.15 speed, air temperature and lime addition.
RDI (Reduction Degradation Index) [ISO 4696-2
Data Analysis (2015)] is a sinter quality index used in the iron-
making industry that provides a measure of the
Three steps were employed to perform the data degradation of the sinter that could occur in the
analysis: (1) simple data inspection and data pre- upper section of the blast furnace after some
treatment, including the removal of outliers and the reduction (a high degree of reduction disintegration
rejection of non-significant parameters; (2) a search generates fines in the top of the furnace that affects
for a linear relationship among variables; and (3) the flow distribution within the blast furnace).12
the study of partial correlation to elucidate the RDI (Reduction Degradation Index) is positively
mutual influence among variables. correlated with bed height and coke consumption,
while being negatively correlated with strand speed,
Linear Correlation lime addition, CaO content and Mn content.
Obtaining a simple correlation between variables Apart from the stochastic aspects, there are
is the second step. This correlation was obtained by thermodynamic aspects that induce the use of a
means of the correlation coefficient calculated in a partial correlation between variables.
standard way:
Partial Correlation
P
ðxi  xÞ  ðyi  yÞ The combined effect of input variables and their
rxy ¼  N
ð4Þ possible interrelations cause problems when
1=2 attempting to carry out a conclusion from the
P 2 P 2
ðxi  xÞ  ðyi  yÞ simple correlation obtained by the method described
N N above. Therefore, it is necessary to obtain the
where xi is the value of x for observation i, x is the mean correlation between two variables while avoiding
x value, yi is the value of y for observation i, and y is the the effect of the rest,28 which is a known partial
mean y value. The following description will be focused correlation. To perform this, a new partial correla-
on the most significant results on linear correlation. tion coefficient was employed as described below:
Therefore, the effort will be focused on the correlation
between plausible input variables and plausible out- tj
put variables. In the cases in which the existence of a r^xi xj ;R ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð5Þ
2
tj þ n  k  1
correlation between input variables could be noted,
the existence of co-linearity will also be indicated.
The correlation coefficient between iron content in where n is the sample size, k is the degree of
ore blend and sinter is approximately 0.55. Similar freedom, and tj is defined as:
results were achieved for FeO, although this vari-
able is known to be largely altered during the
sintering process. In fact, the FeO content in sinter b^j
tj ¼ pffiffiffiffiffi ð6Þ
also presents some degree of negative correlation s^R qjj
2092 Fernández-González, Martı́n-Duarte, Ruiz-Bustinza, Mochón, González-Gasca, and Verdeja

the subtractive clustering algorithm.22 This algo-

b^j represents the coefficient of variable xj in the
rithm permits the modelling of data behavior, which
linear regression model of xi, S^R would be the clusters the experimental data around some values
variance non-explained with the linear regression that are obtained during the modelling process.22
model and qjj will be the diagonal element ‘j’ of the The model can be fitted by employing the cluster
covariance matrix. radius.27 This parameter indicates the range of
Although partial correlation was studied for every influence of a cluster and must be specified before-
variable, only the results concerning sinter proper- hand.27 Specifying a small cluster radius will typ-
ties will be presented here. ically yield many small clusters in the data.27
Some remarkable correlations are: Specifying a large cluster radius will typically yield
Tumbler index is positively correlated (among a few large clusters in the data.27
others) with: variables that represent the ignition In the case of RDI estimation, the input variables
furnace parameters, an index of gas consumption in were selected according to the results obtained in
the ignition furnace, the temperature in the wind- the previous study using partial correlation.15 The
box at the end of the strand, and the coke content. input variables selected were sinter Al2O3 content,
Tumbler index is negatively correlated with: Fe sinter Alkalis content, coke content, air tempera-
content in the blend, MgO content in the blend and ture and bed height. A collection of 200 samples of
strand speed. historical data was divided into two sets of 100
In the RDI case, the strong influence of Alkalis samples. The first set was employed for training,
(K2O + Na2O) and Al2O3 that is observed is and the second set was employed for validation. The
remarkable. model was developed with a rather large value for
As is well known, the influence of alkalis activates the cluster radius. Because RDI measurements are
the coke gasification kinetics (Boudouard often imprecise, the model should show only the
Reaction):29–31 tendency of RDI.
The Tumbler index was modeled in a very similar
Coke þ CO2 , 2CO(gÞ way.15 The input variables selected in this case
were: sinter SiO2 content, sinter CaO content, air
Therefore, there is a maximum calorific power of the flow, air temperature, and coke content.
coke reduction (total oxidation in excess of oxygen to To evaluate the adjustment between the model
CO2) and the possibility of reaching maximum and real measures, a correlation coefficient was
temperatures in the sintering front. Consequently, used. The performance index in the training period
the Tumbler and RDI indexes decrease. (this period helps to select the model and estimate
its parameters) was 0.934 for RDI and 0.702 for the
Fuzzy Inference Systems for RDI and Tumbler Index Tumbler index. The performance index in the
Estimation validation period (where the forecasted model is
tested to know if it functions properly) was 0.916 for
RDI and Tumbler indexes depend on a complex
RDI and 0.832 for the Tumbler index.
method on the chemical properties of iron ores, as
The models for the RDI and Tumbler indexes are
well as the performance of the sinter strand.
included in the optimization process as two quality
To estimate the values for these sinter properties,
elements, to ensure a material with the blast
a FIS was developed. The employed model was a
furnace operator quality parameters.
Sugeno-type FIS, which was obtained by means of

Table I. Restrictions in the optimization process for sinter

Value

Variable Maximum Minimum Exact

% Fe 56
% MgO 1.65
Basicity Index 1.70
% SiO2 5.40 5.30
% Al2O3 1.35
% Alkalis 0.11
% Phosphorus 0.04
Tumbler index (%) 74
RDI index (%) 33
%<0.125 mm 15.00
0.2 mm< %<0.7 mm 18.00
Optimization of Sinter Plant Operating Conditions Using Advanced Multivariate Statistics: 2093
Intelligent Data Processing

Fig. 1. Features of the optimum blend obtained from stockpile number 30.

Optimization Problem Definition – Iron content (>56%): The results of operation

show that the work point is typically far beyond
As a final goal, the optimization process, must
this limit, mainly because Fe content has a
minimize the price of the ore blend to ensure the
negative influence on the sinter strength. There-
minimum quality requirements (restrictions in the
fore, its content may not be far beyond the lower
optimization process can be read in Table I). To
limit imposed.
reach a proper definition of the problem, it is
– MgO (1.65%): Data obtained from the last
necessary to establish the relationship between
6 months show that the results are close to this
these restrictions and the ore blend properties and
value (mean: 1.65, deviation 0.15).
to set their values.
– Basicity Index (1.7): The results that were
The content in the sinter of the variables consid-
obtained show a mean value of 1.69 with a
ered as restrictions in the European factory practice
standard deviation of 0.10.
is calculated as follows:
– SiO2 content (5.3–5.4%): The results that were
obtained show a mean value of 5.03 with a
P standard deviation of 0.24. These results lead to
%XðiÞ  MðiÞ a review of the way in which this restriction is
i
%XðSÞ ¼ P  100 ð7Þ imposed, not only because the mean value is out
RSðiÞ  MðiÞ of range but also because the allowed interval
i
width is narrower than the actual standard
where X(S) is the X phase content in the sinter, X(i) deviation of data.
is the X phase content in the ore i, M(I) is the mass – Al2O3 (<1.35%): The results show a mean value
of ore (i) that should be employed in the blend and of 1.12 with a deviation of 0.06. In fact, the upper
RS(i) is the ore-sinter yield for ore (i). This last bound is hardly ever crossed.
variable is obtained by subtracting the losses due to – Alkalis (<0.11%): To estimate the alkalis content
humidity, calcinations, de-sulfuration and de- in the sinter, the de-alkalization process
alkalization. (%dealk) must be taken into account. Therefore,
The results are: the equation in this case is:
2094 Fernández-González, Martı́n-Duarte, Ruiz-Bustinza, Mochón, González-Gasca, and Verdeja

Fig. 2. Features of the optimum blend obtained from stockpile number 30 fixing set points to their actual limits.

approximately 375,000 €. The software shows the

P  %dealk
 optimum mix of iron ore, recycled products, slag-
%AlkðiÞ  MðiÞ 1  100 forming elements, fluxes, etc. that satisfies the
i
%AlkðSÞ ¼ P  100 ð8Þ restrictions described in previous sections.
RSðiÞ  MðiÞ
i

CONCLUSION
%dealk has been established at 40%. Data on During this work, data from a European sinter
alkali content show a mean value of 0.022 with a plant were collected and studied by means of
standard deviation of 0.0035 and, thus, do not statistical tools to establish the relationship among
appear to be an important restriction. variables (chemical and mechanical properties of
both blends and sintered products, and sintering
– Phosphorus (<0.04%): Data for P content show a process variables). Non-linear relationships were
mean value of 0.046 with a standard deviation of found; therefore, non-linear optimization was
0.003. proposed.
The RDI and Tumbler indexes were considered as
MATLAB Software Tool restrictions of the process. As a consequence of their
relationship with other variables, fuzzy programing
With all that has been previously mentioned, a (Sugeno’s inference model) was used with the
program was developed in MATLAB to conduct purpose of predicting them.
price and quality optimization for the iron ore The goal was to obtain a mixture that fulfilled the
blends that are used in a European sinter plant. conditions imposed on sinter quality, while also
To analyze the performance of the optimizer, blends obtaining the least expensive mixture. This objec-
employed in the past were reproduced. tive was reached by means of the development of a
A reduction in the price per ton of sinter of MATLAB optimizer, which obtained an iron ore
approximately 0.15 € [original 23.38 €/t sinter blend that was less expensive than that previously
(Fig. 1), optimizer tool 23.22 €/t sinter (Fig. 2), in used and contained quality parameters that
the case of the stockpile 30] was predicted. In a ensured homogeneous and consistent operation in
sinter plant of 2.5 Mt, the cost savings would be the blast furnace.
 Optimization of Sinter Plant Operating Conditions Using Advanced Multivariate Statistics: 2095
Intelligent Data Processing

ACKNOWLEDGEMENTS erales por medio de estadı´sticas multivariantes avanzadas

y lógica difusa (Oviedo, ES: Universidad de Oviedo, PhD.
The authors wish to extend their appreciation to the Thesis, 2015).
Spanish MICYT (MAT 2001-4435-E) for their finan- 16. R. Fletcher, Practical Methods of Optimisation: Con-
cial support. The research was also supported by the strained Optimisation, vol. 2 (Chichester: Wiley, 1981), p.
224.
Spanish Ministry of Education, Culture and Sports via 17. R. Barea, J. Mochón, A. Cores, and R. Martı́n, ISIJ Int. 46,
an FPU (Formación del Profesorado Universitario) 687 (2006).
grant to Daniel Fernández González (FPU014/02436). 18. R. Martı́n, J. Mochón, L.F. Verdeja, R. Barea, P. Rusek,
and J. Jiménez, Steel Res. Int. 80, 185 (2009).
19. J.S.R. Jang, C.T. Sun, and E. Mizutani, IEEE Trans.
REFERENCES Autom. Control 42, 1482 (1997).
20. J.S. Bendat and A.G. Piersol, Random Data. Analysis and
1. P.R. Dawson, Ironmak. Steelmak. 20, 137 (1993). Measurements Procedures, (New York: Wiley, 1987), p. 566.
2. J. Mitterlehner, G. Loeffler, F. Winter, H. Hofbauer, H. 21. J.M. Mendel, IEEE Proc. 83, 345 (1995).
Schmid, E. Zwittag, T.H. Buergler, O. Pammer, and H. 22. J.S.R. Jang, C.T. Sun, and E. Mizutani, Neuro-Fuzzy
Stiasny, ISIJ Int. 44, 11 (2004). and Soft Computing (New York: Prentice Hall, 1996),
3. W. Yang, R. Changkook, S. Choi, E. Choi, D. Lee, and W. p. 614.
Huh, ISIJ Int. 44, 492 (2004). 23. A. Priyono, M. Ridwan, A.J. Alias, R. Atiq, O.K. Rahmat, A.
4. G. Szatvanyi, C. Duchesne, and G. Bartolacci, Ind. Eng. Hassan, and A. Mohd, J. Teknol. 43, 143 (2005).
Chem. Res. 45, 4706 (2006). 24. A. Lofti, H.C. Andersen, and A.C. Tsoi, IEEE Trans. Syst.
5. L. Wang, Adaptive Fuzzy System and Control Design and Man Cybern. B 26, 332 (1996).
Stability Analysis (New York: Prentice Hall, 1994), p. 232. 25. R.R. Yager and D.P. Filev, J. Intell. Fuzzy Syst. 2, 209
6. N. Gulley and J.R. Jang, Fuzzy Logic Toolbox User’s Guide (1994).
(Natick: The Math Works, 2005). 26. M.V. Ramos, E. Kasai, J. Kano, and T. Nakamura, ISIJ Int.
7. J.A. Freeman and D.M. Skapura, Neuronal Network—Al- 40, 448 (2000).
gorithms, Application and Programming Techniques (New 27. A.K. Lohani, N.K. Goel, and K.K.S. Bhatia, J. Hydrol. 331,
York: Addison Wesley Publishing Company, 1991), p. 401. 146 (2006).
8. R.K. Sarma, A. Gupta, and S. Vadhavkar, Neural Network- 28. L.S. Meyers, G. Gamst, and A.J. Guarino, Applied Mul-
Based Data Mining Techniques for Steel Making, ed. H. tivariate Research. Design and Interpretation, 2nd ed.
Bozdogan (Florida: CRC Press LLC, 2004), p. 401. (California, CA: SAGE Publications, Inc. 2013), pp. 337–
9. J. Hinnelä and H. Saxén, ISIJ Int. 41, 142 (2001). 338.
10. P.J. Laitinen and H. Saxén, Ironmak. Steelmak. 34, 109 (2007). 29. J.P. Sancho, L.F. Verdeja, and A. Ballester, Metalurgia
11. A. Cores, L.F. Verdeja, S. Ferreira, I. Ruiz-Bustinza, and J. Extractiva. Volumen II: Procesos de obtención, (Madrid, ES:
Mochón, Dyna-Colombia 80, 152 (2013). Sı́ntesis, 2000), pp. 22–27.
12. J. Mochón, A. Cores, I. Ruiz-Bustinza, L.F. Verdeja, J.I. 30. A. Babich, D. Senk, H.W. Gudenau, and TTh Mavromma-
Robla, and F. Garcı́a-Carcedo, Dyna-Colombia 81, 168 (2014). tis, Ironmaking, (Aachen, DE: Department of Ferreous
13. A. Cores, L.F. Verdeja, S. Ferreira, I. Ruiz-Bustinza, J. Metallurgy (Aachen: RWTH Aachen University, 2008), pp.
Mochón, J.I. Robla, and C. González-Gasca, Dyna-Colom- 50–64.
bia 82, 227 (2015). 31. E. Lwamba and A.M. Garbers-Craig, J. S. Afr. I Min.
14. J. Iglesias, J. Sancho, M. Ordiales, D. Fernández, J. Metall. 108, 293 (2008).
Mochón, I. Ruiz-Bustinza, A. Fuentes, and L.F. Verdeja,
Ironmak. Steelmak. 42, 618 (2015).
15. M.R. Martı́n, Optimización de las condiciones de operación
de una planta de sı´nter y localización del parque de min-

View publication stats