ISIJ International, Vol. 42 (2002), No. 1, pp.

4452

Three-dimensional Multiphase Mathematical Modeling of the

Blast Furnace Based on the Multifluid Model
Jose Adilson de CASTRO, Hiroshi NOGAMI and Jun-ichiro YAGI
Institute of Multidisciplinary Research for Advanced Materials, Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai 980-8577
Japan. E-mail: yagi@tagen.tohoku.ac.jp
(Received on September 18, 2000; accepted in final form on September 6, 2001 )

The blast furnace process is a multi-phase chemical reactor whose main purpose is to reduce iron oxides
producing hot metal. In the actual blast furnace operation several phases simultaneously interact with one
another exchanging momentum, mass and energy. In this paper a three-dimensional multiphase mathematical model of the blast furnace is presented. This model treats the blast furnace process as a multiphase reactor in which all phases behave like fluids. Five phases are treated by this model, namely, gas, lump solids
(iron ore, sinter, pellets and coke), pig iron, molten slag and pulverized coal. Conservation equations for
mass, momentum, energy and chemical species for all phases are solved based on the finite volume
method. In the discretized momentum equations, the covariant velocity projections are used, which is expected to give the best coupling between the velocity and pressure fields and improve the convergence of
the calculations. This is a new feature of the present model regarding to the numerical procedures applied
to the blast furnace modeling, which emphasizes its originality. In addition, gas and solid phases are treated
as continuous phases possessing a pressure field and the SIMPLE algorithm is applied to extract the pressure field and ensure mass conservation. Hot metal, slag and pulverized coal are treated as discontinuous
phases consisting of unconnected droplets. For such phases, momentum conservation is used to calculate
the fields of velocity while the continuity equations are used to calculate the phase volume fractions.
This model was applied to predict the three-dimensional blast furnace operation and predicted temperature distributions and operational parameters like productivity, coke rate and slag rate presented close
agreement with the actual measured ones in the blast furnace process.
KEY WORDS: multidimensional; blast furnace; mathematical modeling; multiphase; drag forces; chemical
reactions; heat transfer; transport phenomena; finite volume method.

1.

Several mathematical models of the blast furnace, which

can reproduce multi-dimensional distributions of the process variables within the furnace, have been proposed.13)
Most of these models are in two-dimensional frame and do
not consider three-dimensional variable distributions, although they successfully revealed variation of furnace efficiency with changes of furnace operation as well as distributions of temperature, reduction degree, and so on.14)
However, a detailed investigation demands a three-dimensional analysis and detailed inter-phase interactions. In the
blast furnace process, the tuyere, which supplies hot blast
and auxiliary fuels, is settled on the lower furnace wall
equally spaced in the circunferential direction, representing
points of three-dimensional steep variable distributions.
Unlike other models,24) authors model can treat gas,
solids, molten metal, slag and powders as separated phase
in a three-dimensional frame. This model shows details of
flow, heat and mass transfer and reactions involving these
phases within the furnace, especially in the lower part. In an
extreme operation, conditions in the lower part of the furnace are expected getting more severe. In such case, threedimensional feature in this region will be more important to
understand. Ohno et al.5) carried out three-dimensional

Introduction

In the integrated route of steel production, the blast furnace process is a key operation unit. An improvement of the
blast furnace efficiency contributes to reduce the energy
consumption in the steel industry because this process represents about 70% of the total energy input to this industry.
Modifications to the blast furnace operation, like increasing
pulverized coal rate, changing burden distribution or chemical/physical properties of charged materials and so on,
have been resulted in substantial improvement in the furnace efficiency. However, social pressure to further reduction of energy consumption in the steel industry is getting
stronger, because the problem of global warming arose in
general in the past decade. To achieve further improvement
of furnace efficiency, extreme modifications to the furnace
operation are considered to be required and the changes in
in-furnace status need to be understood for designing such
operations. Therefore the prediction of the blast furnace operation, under severe operation conditions, have to be determined by precise mathematical models being able to investigate the influence of changes of operational parameters on
in-furnace variables.
2002 ISIJ

44

ISIJ International, Vol. 42 (2002), No. 1

analysis of gas flow taking into account the layer-by-layer

structure of packed bed, and showed strong influence of
ore/coke distribution and off-center charge. However no
further development of this model was published. Takatani
et al.6) presented three-dimensional model and showed several three-dimensional features in the blast furnace. Their
model, however, employs some simplistic assumptions such
as vertical liquid flow, constant reaction rates in the raceway and so on. It is incapable of simulating detailed threedimensional phenomena around the combustion zone of the
blast furnace. A comprehensive blast furnace model capable of analyzing detailed phenomena in a three-dimensional
frame is demanded and could make possible the analysis of
the blast furnace operating under several conditions never
experienced before and suggest new operation practices in
order to improve the performance of this reactor.
In the lower part of the blast furnace, tuyeres are equally
distributed as mentioned above. Hot blast gas blown from
the tuyere push coke particles away and form individual
raceway zone. Additionally, this configuration results in discontinuous regions with high carbon consumption rate, and
leads funnel flows of lump solids toward the raceway. The
motion of liquid phases (molten metal and slag) is affected
by both flows of gas and solid phases. The liquid flow region is determined by the interactions among these phases,
and steep distributions of temperatures and compositions
are formed in the combustion zone of the furnace.
Therefore such three-dimensional features are required to
comprehend for the prediction of the main product quality
and operation efficiency of the blast furnace. In this study,
the authors developed three-dimensional blast furnace
model and especial attention was paid in the extension to
make this model capable of analyzing three-dimensional
and multi-phase transport phenomena in the lower part of
the furnace.
2.

Fig. 1. Grid arrangement for 3-D blast furnace model.

force or inter-phase interactions that can be through chemical reactions or surface interactions.
2.2. Source Terms
Regarding source terms in Eq. (1), the continuity and
species equations have mass sources due to chemical reactions and phase transformations.3,4,6) Enthalpy sources arise
from inter-phase heat transfer, heat of reactions and sensible heat transfer accompanying mass transfer due to chemical reactions and phase transformations.24) The formulations for the phase interactions and chemical reactions have
been previously published.3,4)

Multi-dimensional Multiphase Model

2.1. General Conservation Equation

The blast furnace process is treated as a multi-phase
chemical reactor. The phases considered in the blast furnace are assumed behaving like fluids and are modeled by
the general conservation equation. The model analyses the
packed bed region within the blast furnace, from the surface of the slag in the hearth up to the burden surface in the
throat. The phases considered are gas, solids, hot metal,
molten slag and pulverized coal. Each phase consists of one
or more components having its own composition and physical properties.
Governing conservation equations for all phases are expressed by the general conservation equation, represented
by Eq. (1).
f )5div(e G grad f )1S ..............(1)
div(r e U
i i

i i

i fi

2.3. Method of Solution

The blast furnace can be accurately represented by a numerical grid of the body fitted coordinate type (BFC). The
governing equations are discretized over the numerical grid
using the control volume method.711) The arbitrary shaped
control volumes in physical space are transformed in cube
in computational space and all the necessary information
about the coordinate transformation are given by the area
vectors and the Jacobian of the coordinate transformation.
Since locally fixed coordinate system is used, terms that can
produce inaccurate results for non-smoothed grid are avoided.10,11) The numerical grid used to simulate the blast furnace process in a three-dimensional frame is shown in Fig.
1.
Following the control volume formulation, the differential equations are integrated directly in the physical domain
and afterwards relations between the physical and computational domains are introduced.10) Figure 2 illustrates the
conceptual view of coordinate transformation from the
physical coordinates to computational ones. In a nonstaggered arrangement, the velocities are placed in the same
positions of the scalar ones. However, in order to avoid the

fi

In Eq. (1) i represents the phase being considered. The

effective diffusive transfer coefficient assumes different
meaning for each dependent variable. For instance, for
phase velocity the transfer coefficient assumes the dynamic
viscosity and for the phase energy conservation equations
the transfer coefficient assumes the effective thermal conductivity and so on. The source terms are due to external
45

2002 ISIJ

ISIJ International, Vol. 42 (2002), No. 1

Fig. 2. Coordinate transformation used to treat irregular geometry.

Fig. 3. Representation of the covariant projections in a 3-D nonorthogonal system.

Fig. 4. Computational molecule for three-dimensional control
volume discretization.

well known decoupling of the velocity and pressure fields,

special interpolation of the velocities at the faces of the
control volume is required, which represents additional
computation efforts.10) In this model the velocity projections are calculated in staggered position to simplify the numerical procedure, although additional storage of geometrical information is required. The use of velocity projections
as dependent variables in momentum equations has the advantage of less computational effort to calculate source
terms, good coupling of velocity and pressure fields and the
natural extension of the algorithm for orthogonal system. A
preliminary study, performed by the authors, showed clearly the advantages of the numerical method adopted in this
work. Using a non-staggered and a staggered with covariant
projections arrangement, for a standard gas flow through an
uniform unreactive packed bed, was observed that the staggered with covariant procedure was about 30% faster for
the same level of convergence, which was decisive to select
this implementation. In addition, the discretized equations
for the velocity projections are deduced by simple algebraic
manipulation of the discretized momentum equation for
Cartesian grid.10,11) Details of the discretization for the general transport equation and the momentum equations are
presented in Appendix A. Figure 3 shows the conceptual
view of physical covariant projections used as dependent
variable in the momentum equations.
This method reduces the differential equations to a large
set of strongly coupled linear equations. This system of lin 2002 ISIJ

ear equations has been solved by the iterative procedure

termed line by line method. The computational molecule
for the discretization of the general conservation equation is
given in the Fig. 4, where the connection between the node
variable and the neighbors control volume is given by the
coefficients of the discretized equation.
The gas and solid phases are treated as continuous phases, possessing a phase pressure field. For such continuous
phases, the model solves the equations of motion using the
SIMPLE scheme of Patankar and Spalding for extracting
the pressure field.711) Hot metal, slag and pulverized coal
are treated as discrete or discontinuous phases, consisting
of unconnected droplets or particulates. No phase pressure
can be defined within a discontinuous phase, thus the SIMPLE scheme cannot be applied. Alternatively, the continuity equation is used to calculate the phase volume fraction
distribution.4) Thus, the main features of the numerical
method used in this model can be summarized as: (1) the
use of control volume approach to derive the discretized
equations; (2) The use of the covariant velocities projections as dependent variable in momentum equations in staggered positions; (3) The SIMPLE algorithm to the pressurevelocity link; (4) TDMA(Tri-diagonal Matrix Algorithm) to
solve the algebraic equations. The flowchart of the proposed mathematical model is shown in Fig. 5.

46

ISIJ International, Vol. 42 (2002), No. 1

3.

Results and Discussions

In this section, the actual blast furnace operation is simulated. The input data used in this simulation was taken as
the monthly average values. The operational parameters
used as input data are summarized in Table 1. The essential
boundary conditions used to simulate the actual blast furnace are the blast conditions and the top burden distribution. Figure 6 shows the burden distribution pattern used as
boundary conditions for solid phase at the burden surface.
The curves presented in Fig. 6 were linearly interpolated to
connect the measured data for relative burden layers and
solids diameters. The data used in this simulation was taken
as the monthly averaged value for a blast furnace operation,
however the data was selected for periods when the operation was clearly stable, which avoids the inclusion of noisy
data. Coke was mainly charged at the central region while
ore was charged preferentially in the peripheral region.
Table 2 summarizes the mass balance for the major species
considered in this model and Table 3 shows the energy balance for the five phases predicted by the model. Most mass
balances for species were lower than 1% and good agreement of the measured and calculated global parameters like
production, slag rate and so on is verified in Table 4.
The effect of the number of raceway in the circunferenFig. 5. Flowchart of the three-dimensional model.

Table 1.

Major operational parameters for the calculation.

Fig. 6. Radial distribution pattern of burden materials.

Table 2.

Elemental mass balance.

Table 3.

Energy balance.

47

2002 ISIJ

ISIJ International, Vol. 42 (2002), No. 1

tial direction was investigated and shown in Fig. 7. Strong

effect on the gas flow pattern in the lower part of the blast
furnace was observed. For the case of higher number of
raceway the gas distribution become more uniform and
minimizes the non-symmetric conditions of the lower part
of the reactor.
In Fig. 8 the predicted temperature distributions of the
major phases in the blast furnace with 40 raceway configuration are shown. The gas and solid temperature distributions are quite similar due to the strong interactions between these phases. Particularly, the temperature distribution in the blast furnace is closely related with the burden
distribution pattern shown in Fig. 5. These features of the
temperature patterns are due the strong dependence of the
gas flow, chemical reactions and physical properties on the
bed porosity. Therefore, as coke was mainly charged at central region, the main stream of the gas path also is in this region which in turn, makes the isothermal surface higher in
this region. The hot metal and molten slag isothermal surfaces present also similar pattern. The temperature of pulverized coal phase is similar to the gas phase due to the permanent contact of gas and PC phases through the blast furnace. Figure 9 shows the mass flow rates of the liquid phases in the blast furnace. Liquids preferentially flow down in
the peripheral region due to two main factors. One is due to
the strong interaction between the solid and the liquid phases forcing the liquid to follows the solid motion. And as
Table 4.

Fig. 7. Effect of number of raceway on the gas flow at the tuyere

level.

Comparison of global parameters.

Fig. 8. Three-dimensional temperature distribution for the mahor phases in the blast furnace (unit: C).

2002 ISIJ

48

ISIJ International, Vol. 42 (2002), No. 1

Fig. 11. Gas temperature distributions in two vertical planes

(unit: C).
Fig. 9. Flow vectors of hot metal and molten slag phases.

Fig. 10. Predicted flow pattern of solid and gas phases.

Fig. 12. Comparison of measured and calculated solid temperature in a vertical plane (averaged in the circumferential
direction) (unit: C).

mainly iron bearing materials were charged in the peripheral region hence liquids generated by melting contribute to
increase the amount of liquids there. In Fig. 10 the flow
pattern of solid and gas phases are presented. In the solid
flow the funnel flow toward the raceway is observed while
the plug flow is observed in the upper part of the blast furnace. Regarding gas flow, the cohesive zone effect on the
gas flow is observed due to drastically decrease in bed
porosity when the liquids are formed. Figure 11 presents
temperature isotherms in different vertical planes. The
plane which cross the raceway presents higher temperature

in the vicinity of the raceway, in contrast of those between

two raceway. The lower temperatures of these planes extent
to the lower part of the furnace promoting non-uniform
temperature pattern in these regions.
Figure 12 shows comparison of measured and predicted
average temperature distribution in the vertical plane. The
measured and calculated temperature presents good agreement in the central region. And in the peripheral region, the
isotherms of 700C and 900C do not agree very well. The
reason is attributed to the rate of chemical reactions which
is based on measurements at laboratory conditions, which
49

2002 ISIJ

ISIJ International, Vol. 42 (2002), No. 1

10) M. C. Melaaen: Numer. Heat Transfer, B, 21 (1992), 1.
11) K. C. Karki and S. V. Patankar: Numer. Heat Transfer, B, 14 (1988),
295.

can not cover all the blast furnace conditions.

4.

Conclusions

A three-dimensional mathematical model of the blast furnace has been developed. The model comprehends the full
circunferential blast furnace and is based on the multi-fluid
theory. Special attention was devoted to analyze the lower
part of the blast furnace where the three-dimensional behavior is more evident. This model has successfully predicted the actual blast furnace operation and comparisons of
predicted and measured blast furnace parameters showed
good agreement.

Appendix
Discretization of Conservation Equation
Lets (x1, x2, x3)Cartesian system; (x 1, x 2, x 3)general
curvilinear non-orthogonal system. Two distinct frames of
basis vectors can be defined at any point in a curvilinear
non-orthogonal system. The so-called covariant basis vectors follows the coordinate lines being tangents at any point
and the contravariant basis vectors are perpendicular to the
surface defined by two of the covariant vector. Equations
(A1) and (A2) define the Cartesian components of these
vectors, respectively.

Nomenclature
A(i): Area vectors generated by two covariant vectors (m2)
Aki: Cartesian components of the area vectors (m2)
ai: Contravariant base vector (m)
ai: Covariant base vector (m)
ak: Coefficients in discretized equation (k: W,E,S,N,B,T,P)
b: Independent part of the linearized source term
bNO: Non-orthogonal part of the linearized source term
C i: Convective normal flux (kg/s)
F: Flux tensor
Gij: Cartesian components of geometric diffusion coefficient
g ij: Cartesian components of contravariant metric tensor
(m22)
gij: Cartesian components of covariant metric tensor (m2)
Ik: Cartesian unit vectors (2)
J: Jacobian determinant of coordinate transformation
(volume in physical space) (m3)
P9: Pressure correction (Pa)
Pe: Peclet number
Pi: Phase pressure (Pa)
Sf : Source term for variable f

U i: Phase velocity vector (m/s)

uk: Cartesian velocity components (m/s)
Vi: Covariant velocity projections (m/s)
V9i: Covariant velocity projections corrections (m/s)
V9inb: Neighboring velocities decomposed into the physical
coordinates direction (m/s)
xi: Cartesian coordinate system (m)

ai 5

ai 5

i
xk

I k 5J ik I k .........................(A1)

I k 5J ki I k .........................(A2)

The Jacobian determinant of the coordinate transformation is defined by Eq. (A3) and the rows and columns of the
matrix formed by the derivatives of the coordinate gives the
necessary information of the coordinate transformation
used to express the discretized form of the transport equation.

x1 x1 x1
J 5det

1 2 3
x2 x2 x2
1 2 3
x3 x3 x3

.....................(A3)

1 2 3
The Cartesian components of the contravariant basis vector are evaluated by the following expression:
ai 5

Greek symbols:
G f : Transfer coefficient of the variable f (2)
x i: Covariant coordinate system (m)
e i: Phase volume fraction (2)
r i: Phase density (kg/m3)

1
(a j 3ak ) (i, j, k cyclic) ...............(A4)
J

Therefore, the Jacobian of the transformation being evaluated, all necessary information can be evaluated. The components of the Jacobian matrix can be numerically determined (computational space approach) or by the direct calculation of the area vectors and volume in the physical
space (physical space approach). We have chosen the computational space approach.
The covariant and contravariant metric tensors components can be written from the basis vectors components,
thus the nine components are:

REFERENCES
1) T. Sugiyama and M. Sugata: Seitetsu Kenkyu, 325 (1987), 34.
2) P. R. Austin, H. Nogami and J. Yagi: ISIJ Int., 37 (1997), 458.
3) P. R. Austin, H. Nogami and J. Yagi: ISIJ Int., 38 (1998), 246.
4) J. A. Castro, H. Nogami and J. Yagi: ISIJ Int. 40 (2000), 637.
5) Y. Ohno, Y. Yamada, K. Kondao and T. Takebe: Tetsu-to-Hagan, 73
(1987), 2036.
6) K. Takatani, T. Inada and Y. Ujisawa: ISIJ Int. 39 (1999), 15.
7) S. V. Patankar: Numerical Heat Transfer and Fluid Flow, McGrawHill, New York, (1984), 90.
8) S. V. Patankar and D. B. Spalding: Int. J. Heat Mass Transfer, 15
(1972), 1787.
9) Flow 3D Release 2.4 -User Manual CFDS, Harvell, UK, (1991).

2002 ISIJ

xk

gij5ai aj5J ikJ jk ............................(A5)

g ij5ai aj5Jki Jkj ............................(A6)
The area vectors generated by two of the covariant vectors
are given by:
50

ISIJ International, Vol. 42 (2002), No. 1

A(i)5aj3ak (i, j, k cyclic)...................(A7)

F 1 |e 5 U 12 G11 1 1G12 2 1G13 3

The geometric diffusion coefficient is defined by:

G ij 5

Aki Akj
J

......................................(A15)
..............................(A8)

F 2 |n 5 U 22 G 21 1 1G 22 2 1G 23 3

With the above equations all the necessary information

of the coordinate transformation is given. These equations
relate the physical space with the computational space and
this is currently used to obtain the discretized form of the
conservation equation. Following, the discretized form of a
general state variable is deduced and then the discretized
form of the momentum equations are presented in terms of
the covariant projections.

......................................(A16)

F 3 |t 5 U 32 G 31 1 1G 32 2 1G 33 3

......................................(A17)

Discretization of the General Conservative Equation

The discretization of the general conservation equation is
based on the finite volume concept. The coordinate invariant Eq. (A9) is integrated over a general control volume in
the physical space and then the Gauss divergence theorem
is applied to transform the volume integral into a surface
integral.

div( i i ui2 i grad )5

Similar expressions is given for w,s and b faces. U 1, U 2 and

U 3 are the normal velocities to the faces e, n and t, respectively. If the above equations are substituted in Eq. (A9) a
general discretized equation is the result:
aPf P5aWf W1aEf E1aBf B1aTf T1aSf S1aNf N1b
......................................(A18)
Where

S .......(A9)

aP5aW1aE1aB1aT1aS1aN1aP02SP ........(A19)

Refer to Fig 4 the concept of the finite volume method,

where the average value of the dependent variable is assumed to prevail over all the control volume centered in the
point P. Notation presented in Fig. 4 is as follows: wwest
face, eeast face, ssouth, nnorth, bbottom and ttop. For
the computational molecule similar notation applies
WWest, Eeast, Ssouth, Nnorth, Bbottom and Ttop
volumes. Hence, from now this notation will be used to
identify the face control volume while the capital letters
will identify the neighbor points. In order to simplify the
integral form of the conservation Eq. (A9) lets define a flux
tensor as follows:
r
F5U2 ........................(A10)

and
b5bNO1SC1aP0f 0 ........................(A20)
e

bNO5G12 2 1G13 3 1G 21 1 1G 23 3

1G 31 1 1G 32 2 ...............................(A21)

The coefficients in Eq. (A18) are determined by the

power law scheme of Patankar.7)
Covariant Projections and Normal Fluxes

A velocity vector U can be projected along the coordinate line by using the scalar product of the velocity direction and the coordinate line direction.

Hence the control volume integration of Eq. (A9) is given

by using Eq. (A10) and the Gauss theorem.
F A|e2F A|w1F A|n2F A|s1F A|t2F A|b5Sc1SPf P
......................................(A11)

Vi 5U ei 5

F A |e 5 F dA5( U A(1)2A(1) ) | Ae 5F 1 |e

uk
gii

J ik ......................(A22)

The normal flux through a volume face is calculated from

the covariant velocity projection by taking the scalar product of the covariant basis vector and the contravariant basis
vectors.

Ae

......................................(A12)

F A |n 5 F dA5( U A( 2 )2A( 2 ) ) | An 5F 2 |n

C i 5 Jg ij g jj V j .......................(A23)

An

......................................(A13)
In similar way, Cartesian velocity components arise from
the covariant projections.

F A |t 5 F dA5( U A( 3)2A( 3) ) | At 5F 3 |t
At

u j5

......................................(A14)

51

gii i
AjVi .........................(A24)
J

2002 ISIJ

ISIJ International, Vol. 42 (2002), No. 1

Momentum Equations with Physical Covariant Velocity

Projections
The Cartesian components of the velocity can be treated
as scalar and the general conservation equation is applied
by replacing f by the Cartesian components and the pressure gradient integrated in addition.

aPVi P 5

nb

Where V9i nb and b9u i are the projections of the neighbor velocities and the source terms in the covariant direction on p
point. Hence a locally fixed coordinate system is used, it
changes from point to point in the domain.

P
....(A25)
dV 52 J k 52 Aki
j
x
x P
P

Then, the discretized momentum equation based on

Cartesian components can be rewritten as follows:

aP uk P 5

nb

P
anb uk nb1buk 2 Akj
j

........(A26)
P

If the Cartesian components and the physical covariant projection share the same control volume, the coefficients in
the discretized momentum equations are same. Then the
discretized equation of the covariant velocity projections
can be obtained by algebraic manipulation of the discretized Cartesian components. Multiplying Eq. (A26) by
(1 / gii ) ( x k / i ) and using Eq. (A24) the following

Vi nb5

x j
1
u j nb i and
gii
P
P

binb5

1 x j
bu j i ................(A28)
gii P
P

The covariant projections of the neighbor points are introduced by summing and subtracting the neighboring contribution as follows:

aPVi P 5

52

J P

gii i
P
......................................(A29)

2
nb (Vi nb 2Vi nb )1bui

nbVi nb 1

nb

equation emerges.

2002 ISIJ

J P
.......(A27)
anbVi nb1bui 2
gii i
P

nb