ARTICLE IN PRESS

Engineering Applications of Artificial Intelligence 16 (2003) 511–527

Mathematical modeling and optimization strategies

(genetic algorithm and knowledge base) applied to the
continuous casting of steel
C.A. Santosa, J.A. Spimb, A. Garciaa,*
a
Department of Materials Engineering, Faculty of Mechanical Engineering, State University of Campinas (UNICAMP) P.O. Box 6122,
13083-970, Campinas, SP, Brazil
b
Federal University of Rio Grande do Sul (UFRGS), Center of Technology, P.O. Box 15021, 91501-970, Porto Alegre, RS, Brazil

Abstract

The control of quality in continuous casting products cannot be achieved without a knowledge base which incorporates
parameters and variables of influence such as: equipment characteristics, steel, each component of the system and operational
conditions. This work presents the development of a computational algorithm (software) applied to maximize the quality of steel
billets produced by continuous casting. A mathematical model of solidification works integrated with a genetic search algorithm and
a knowledge base of operational parameters. The optimization strategy selects a set of cooling conditions (mold and secondary
cooling) and metallurgical criteria in order to attain highest product quality, which is related to a homogeneous thermal behavior
during solidification. The results of simulations performed using the mathematical model are validated against both experimental
and literature results and a good agreement is observed. Using the numerical model linked to a search method and the knowledge
base, results can be produced for determining optimum settings of casting conditions, which are conducive to the best strand surface
temperature profile and metallurgical length.
r 2003 Elsevier Ltd. All rights reserved.

Keywords: Continuous casting of steel; Mathematical modeling; Optimization methods; Genetic algorithm

1. Introduction components of the machine, steel composition, casting

temperature, and casting metallurgical constraints, such
The continuous casting process is responsible for most as thickness of solidified shell at mold exit and strand
of the steel production in the world, and has largely surface temperature profile along the different cooling
replaced conventional ingot casting/rolling for the zones. The use of optimization strategies, such as genetic
production of semi-finished steel shape products. algorithm, heuristic search, knowledge base, working
Fig. 1 shows a schematic representation of a continuous connected to mathematical models of solidification, can
caster and the different cooling zones along the machine. be seen as a useful tool in the search of operational
The casters have been implemented with modern parameters that maximize or minimize any aspect of the
equipments for billets, slabs or blooms, multiple casting dynamic process. The idea of using simulation to
and process control. optimize a continuous caster is not just a theoretical
The quality control of continuous casting is funda- concept and its practicality has already been demon-
mental for reducing production costs, processing time, strated (Larreq and Birat, 1982; Lally et al., 1991a; Lally
and to assure reproducibility of the casting operation et al., 1991b; Kumar et al., 1993; Samarasekera et al.,
and increase of production. This cannot be achieved 1994; Spim et al., 1997; Filipic and Saler, 1998;
without a greater knowledge about the process, in- Brimacombe, 1999; Cheung and Garcia, 2001). An
corporating both operational parameters, such as expert system for billet-casting problems has been
developed to guide caster operators in analyzing
*Corresponding author. Tel.: +55-19-3788; fax: +55-19-3289-3722. quality-related problems and to provide them with a
E-mail address: [email protected] (A. Garcia). ready source of fundamental knowledge related to caster

0952-1976/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0952-1976(03)00072-1
 ARTICLE IN PRESS
512 C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

Ladle (s) and q3 represents the term associated to internal heat

generation due to the phase change. It was assumed that
the thermal conductivity and density vary only with
Tundish temperature. Then, Eq. (1) can be rewritten in two-
dimensional form as
Nozzle
2 
qT q T q2 T 3
Primary Mold + rc ¼k 2
þ 2 þq: ð2Þ
Cooling Pinch Roll qt qx qy
Flame cut-off
Approximating Eq. (2) by finite-difference terms, we
Rolls
Ingot have
 n
Sprays Ti;nþ1 n
j  Ti; j Tiþ1; j  2Ti;n j þ Ti1;
n
j
Secondary
rc ¼k
Dt Dx2
Cooling 
Unbending Point Ti;n jþ1  2Ti;n j þ Ti;n j1 3
Radiation þ þ q; ð3Þ
Cooling Dy2
Fig. 1. Schematic representation of the continuous casting equipment. where n þ 1 is the index associated to the future time, n
is the index corresponding the actual time, Dt is the
operation (Kumar et al., 1993). J.K. Brimacombe, I.V. increment of the time, x; y are the directions, i; j are the
Samarasekera, S. Kumar and J.A Meech projected this positions and the stability criteria are given by
expert system. Filipic and Saler proposed and imple-
Dx2 Dy2
mented a computational approach to the continuous Dto or ;
casting of steel, which consists of a numerical simulator 2a 2a
of the casting process and a genetic algorithm for real where
parameter optimization (Filipic and Saler, 1998). k
Cheung, Santos, Spim and Garcia have used a heuristic a¼ ðm2 =sÞ:
rc
search technique for the optimization of the quality of
carbon steel billets (Cheung and Garcia, 2001; Santos The objective is to determine the future temperature
et al., 2002). of the element i; jðTi;nþ1 j Þ as a function of the actual
The present paper describes a software, which was known temperatures of the elements around the element
n n n n
developed and is based on the interaction between a i; jðTiþ1; j ; Ti1; j ; Ti; jþ1 ; Ti; j1 Þ:
finite difference heat transfer solidification model and a
genetic algorithm and a knowledge base. The heat 2.1. Phase change
transfer model is validated against experimental results
concerning both static casting of Al–Cu and Sn–Pb In this study, a fixed grid methodology is used with a
alloys and continuous casting of a low carbon steel slab heat source term due to the metal phase transformation
and a high carbon steel billet. The software has been (liquid to solid), which is given by an explicit solid
used to explore the space parameter settings in order to fraction-temperature relationship as
find optimized cooling conditions, which result in best 3 qfs
strand surface temperature profile and minimum me- q ¼ rL ; ð4Þ
qt
tallurgical length.
where fs is the solid fraction during phase change along
the solidification range (liquidus and solidus tempera-
tures: TL and TS , respectively) and L is the latent heat
2. Mathematical model
of fusion ðJ=kgÞ. The solid fraction depends on a
number of parameters. However it is quite reasonable
The mathematical formulation of heat transfer to
to assume fs varying only with temperature in the mushy
predict the temperature distribution and the solid shell
zone, and then Eq. (4) can be written as
profile during solidification is based on the General
Equation of Heat Conduction in Unsteady State, which 3 qfs qT
q ¼ rL : ð5Þ
is given for three-dimensional heat flux by qT qt
qT 3 Substituting q3 into Eq. (3), the specific heat can be
rc ¼ rðk rTÞ þ q; ð1Þ written as c0 ¼ c  Lðqfs =qTÞ; where the term
qt
ðLðqfs =qTÞÞ is called pseudo-specific heat. At the range
where r is the material density ðkg=m3 Þ; c is specific heat of temperatures where solidification occurs for metallic
ðJ=kg KÞ; k is thermal conductivity ðW=m KÞ; qT=qt is alloys, the physical properties will be evaluated taking
the cooling rate ðK=sÞ; T is temperature (K), t is the time into account the amount of liquid and solid that coexists
 ARTICLE IN PRESS
C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527 513

in equilibrium at each temperature: y

k ¼ ðkS  kL Þ : fs þ kL ; ð6Þ
i, j+1
c0 ¼ ðcS  cL Þ : fs þ cL  ðL  dfs Þ ð7Þ
x i,j
r ¼ ðrS  rL Þ : fs þ rL ; ð8Þ i-1,j i+1,j

where sub-indices S and L; respectively, indicate solid

and liquid states. If fs ¼ 0; the element is still liquid and i, j-1

only thermophysical properties of the liquid are

considered, and if fs ¼ 1; the element is completely nodal point
thermal resistance
solid. For carbon steels, the fs is appropriately described
by the lever rule, and Scheil’s equation applies for Fig. 2. Nodal point.
Al–Cu and Sn–Pb alloys (these alloys will be used in
model validation). i; j  1 to point i; j given by (Fig. 2)

  
1 TL  T Dx Dy
fs ¼  ðLever RuleÞ; ð9Þ RTi ¼ or RTj ¼ ; ð13Þ
1  ko Tf  T kAt kAt

ð1=ko 1Þ where Dx and Dy correspond to the distance between
Tf  T
fs ¼ 1  ðScheil’s EquationÞ; ð10Þ central points of nodes. Each thermal resistance between
Tf  TL
the central points is given by the sum of the partial
where Tf is the solvent melting temperature (K) and ko thermal resistance from the center to the boundary and
is the partition coefficient. the boundary to the center, given by

2.2. Analogy between thermal systems and electrical in x: RTðiþ1; jÞði; jÞ ¼ RTðiþ1; jÞ þ RTði; jÞ ;
circuits RTði1; jÞði; jÞ ¼ RTði1; jÞ þ RTði; jÞ ; ð14Þ

In the continuous casting processes, heat is trans- in y: RTði; jþ1Þði; jÞ ¼ RTði; jþ1Þ þ RTði; jÞ ;
ferred from the liquid steel to the cooling system (mold,
RTði; j1Þði; jÞ ¼ RTði; j1Þ þ RTði; jÞ : ð15Þ
sprays and free radiation) through various media
namely, the solidified shell, strand/mold interface, mold These terms are given by the sum of thermal
wall, cooling water, sprays/strand interface and air/ resistances according to the following equations:
strand interface. The heat transfer through each of the
media can be characterized in terms of a thermal Dxiþ1; j
RTiþ1; j ¼ ; ð16Þ
resistance, analogous to an electrical resistance. To 2kiþ1; j At
simplify the development of the mathematical model, is
the analogy between thermal systems and electrical Dxi1; j
circuits applied. Multiplying the modified equation (3) RTi1; j ¼ ; ð17Þ
2ki1; j At
by Dx; Dy; Dz on both sides, considering Dx ¼ Dy ¼ Dz;
At ¼ Dy Dz or Dx Dz; and replacing c by c0 ; yields
Dyi; jþ1
Ti;nþ1 n RTi; jþ1 ¼ ; ð18Þ
0 j  Ti; j 2ki; jþ1 At
At Dx r c
Dt
 n 
ðTiþ1; j  2Ti;n j þ Ti1;
n
jÞ ðTi;n jþ1  2Ti;n j þ Ti;n j1 Þ
¼ At k þ : Dyi; j1
Dx Dy RTi; j1 ¼ ; ð19Þ
2ki; j1 At
ð11Þ
Dxi; j Dyi; j
By analogy, the thermal capacitance ðCTi; j Þ represents RTi; j ¼ or : ð20Þ
2ki; j At 2ki; j At
the energy accumulated in a volume element i; j from the
grid, and is given by (Spim and Garcia, 2000), Then, expanding Eq. (11) and substituting CTi; j ; yields
CTi; j ¼ Dxi; j Dy Dz ri; j c0i; j ; ð12Þ Ti;nþ1 n n n n n
j  Ti; j Tiþ1; j  Ti; j Ti1; j  Ti; j
CTi; j ¼ þ
where Dx Dy Dz is the volume of the element i; j: Dt RTðiþi; jÞði; jÞ RTði1; jÞði; jÞ
Also by analogy, the thermal flux between central
Ti;n jþ1  Ti;n j Ti;n j1  Ti;n j
points has a thermal resistance at the heat flux line (RT ) þ þ ð21Þ
from point i þ 1; j or i  1; j to point i; j or i; j þ 1 or RTði; jþ1Þði; jÞ RTði; j1Þði; jÞ
 ARTICLE IN PRESS
514 C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

or based on the following key assumptions:

 n n 
Tiþ1; j Ti1; j Ti;n jþ1 Ti;n j1 (1) Two-dimensional heat transfer phenomenon was
Ti;nþ1
j ¼ Dt þ þ þ
tðiþ1; jÞði; jÞ tði1; jÞ;ði; jÞ tði; jþ1Þ;ði; jÞ tði; j1Þ;ði; jÞ considered, with heat flux being admitted to be

 negligible along the vertical direction ðzÞ:
Dt qT
þ 1 Ti;n j ; ð22Þ ¼ 0:
tði; jÞ;ði; jÞ
qz
where (2) A control volume element, with Dz ¼ 1 mm; was
tðiþ1; jÞ;ði; jÞ ¼ cTi; j ðRTiþ1; j þ RTi; j Þ; ð23Þ placed in a transverse section and was analyzed
from the meniscus to the cut-off region. The
tði1; jÞ;ði; jÞ ¼ cTi; j ðRTi1; j þ RTi; j Þ; ð24Þ distance below the meniscus ðzÞ is given by
Z ¼ Vcasting Dt z ðmÞ; t ðsÞ;
tði; jþ1Þ;ði; jÞ ¼ cTi; j ðRTi; jþ1 þ RTi; j Þ; ð25Þ Vcasting Fcasting speed ðm=sÞ:
tði; j1Þ;ði; jÞ ¼ cTi; j ðRTi; j1 þ RTi; j Þ; ð26Þ (3) The billet/slab symmetry permits that only one-
quarter of the cross-section modeled for a full
1 1 1 1 thermal evolution characterization (grid: 100 100
¼ þ þ
tði; jÞ;ði; jÞ tðiþ1; jÞ;ði; jÞ tði1; jÞ;ði; jÞ tði; jþ1Þ;ði; jÞ points).
1 (4) The meniscus surface was assumed to be flat
þ : ð27Þ ðz ¼ 0Þ:
tði; j1Þ;ði; jÞ
(5) Effect of mold oscillation, mold curvature, segrega-
Eq. (27) is generic and can be applied to any tion, and melt level fluctuation in the mold were
geometry, by varying only area and volume to be ignored.
considered, as well as the thermophysical properties as a (6) The mold is considered uniform and with an
function of the temperature or state of the analyzed initial temperature equal to the water-cooling
element in the grid. The stability criterion is temperature.
Dtptði; jÞ;ði; jÞ : (7) The surface temperature of molten metal is con-
sidered equal to the pouring temperature.
2.3. Boundary conditions (8) The turbulence in liquid metal is analyzed
by a mathematical expedient, where the thermal
The application of the solidification model to conductivity in the liquid is multiplied by a
continuous billet/slab casting operation (Fig. 3) was numerical factor: kef ¼ kL A; where A varies
between 3 and 7 (Toledo et al., 1993; Louhenkilpi,
1994).
(9) The transient mold/strand and sprays/strand
y heat transfer coefficients (hm=s and hs=s respectively)
used in this work, are those proposed in the
x
literature (Samarasekera and Brimacombe, 1988;
z grid
Brimacombe et al., 1984; Lait et al., 1974; Hills,
∆z
1969; Brimacombe et al., 1980; Mizikar, 1970;
mold Nozaki et al., 1978; Bolle and Moureau, 1979),
hm/s and they are related to the interface thermal
resistances along the different regions on the
machine, given by
1
RTm=s ¼ ; ð28Þ
hs/s
sprays hm=s At
1
RTs=s ¼ : ð29Þ
hs=s At

radiation
hr
Vcasting
3. Optimization strategies

In this work, an algorithm is developed which

Fig. 3. Boundary conditions. incorporates optimization strategies to determine best
 ARTICLE IN PRESS
C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527 515

operational parameters to the continuous caster. The to a solidification mathematical model (solid shell
algorithm employs search techniques for finding thickness evolution and surface temperature distribution
these operational parameters, including the type along the billet).
and characteristics of mold, mold taper, mold and The starting input parameters about machine, opera-
sprays cooling systems, etc., which are incorporated tional conditions and casting are first compared with the
in a knowledge base. The search includes finding the knowledge base, and a report with suggestions is
casting objective of maximum production rate as a provided. After that, the operating conditions are
function of casting metallurgical constraints. These submitted to the decision strategy and inserted into the
constraints represents the product quality and process numerical model, which generates a simulation repre-
feasibility through limits on strand shell thickness senting the solidification in the continuous casting
ðSM Þ; metallurgical length ðLM Þ; minimum surface equipment. For developing the decision strategy it was
temperature ðTmin surf Þ; casting speed ðVcasting Þ; strand necessary to acquire a knowledge base concerning the
surface reheating between spray zones ðDTmin surf Þ continuous casting of steel, containing two groups of
and temperature at the unbending point ðTLM Þ: The information: (a) equipment information and (b) process
algorithm modifies the operational process para- information.
meters, such as mold and spray cooling efficiencies (a) The equipment information represents the input
and casting speed, with a view to attain the best variables of the heat transfer model and optimization
conditions for the quality of the cast product at a program, and generally relates to the physical char-
maximum production rate without violating the metal- acteristics of the equipment and the quality of the cast
lurgical constraints. steel. This information represents characteristics of
The functional structure of the algorithm is operation, such as geometry of caster, casting rate,
basically composed of three operating blocks: the composition of steel, casting temperature, type of mold,
first consisting of the numerical heat transfer model, mold length, mold taper, metal level, number and length
which generates results of simulations as a function of sprays zones, water flow rates in the mold and at the
of the input parameters related to operational con different sprays zones, unbending point and water
ditions and equipment limitations; the second temperature.
block incorporates the knowledge base about the (b) The process information represents the transient
continuous casting process, and the third block variables in the process, which can be classified as:
consists of the decision rules (strategy), which are the boundary variables: which can be modified within
managers of the algorithm. It determines the modifica- an operating range to meet specifications of the
tions on the boundary conditions of the continuous desired output, and can eventually be associated with
casting process and is responsible for the insertion economic features and a defect-free product; for
of new input variables into the numerical model. This instance, casting speed and primary and secondary
block has a strong interaction with the results cooling efficiencies, and control variables, which are
furnished by the numerical model. The algorithm associated with the results of the continuous casting
works by iteration, and every result given by the process and for instance, solid shell thickness,
model corresponds to an analysis performed by surface temperature profiles and metallurgical length
the decision rules block, thus indicating any need to (Fig. 4).
modify the process boundary conditions. The algorithm The knowledge base is a set of representation of facts
includes a database of material properties for various about the process (rule-based system). Each individual
steels. representation is called a sentence. The sentences are
expressed in a language called a knowledge representa-
3.1. Knowledge base tion language. The objective of the knowledge repre-
sentation is to express knowledge in a computer-
The knowledge base required to transform molten tractable form, such that it can be used to help agents
steel into quality billets at a high production rate is, of perform well. The logic consists of the Boolean
course, quite large. The present knowledge base was connectives and quantifiers terms, and the structural
based on a wide search in the literature on continuous knowledge implements rules and facts. Rulers are
casting operations and on information obtained statements and procedures, such as condition state-
in a continuous casting plant. It was structured in ments and search strategies, and facts are classes of
order to facility the examination of all important objects and values. Among these objects, various
operational parameters. The outline of quality relations hold. Some of these relations are functions
problems that includes the possible defects, their (relations in which there is only a value for a given
origin and suggested preventive techniques has been input) with exclusive values and others are restrictions.
prepared as a function of rules and data collected in The main rules used in knowledge base system are
the literature and in the industrial practice, and linked shown in Table 1.
 ARTICLE IN PRESS
516 C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

Mold Sprays Radiation Unbending

Point

Shell Thi ck ness

Point of
complete
Smin solidification
Tpouring
Tmax

Surface Temperature ∆T
Tmin
mould zone 1 zone 2 zone 3 radiation zone

Distance from meniscus

Fig. 4. Metallurgical and equipment constraints applied to the continuous casting process.

3.2. Genetic algorithm temperature either greater than the high-temperature

limit of the ductility trough (soft cooling) or lower than
The decision block contains a set of critical and the lower limit in order to avoid transverse surface
limiting operational conditions imposed by metallurgi- cracking (hard cooling). The bottom of the ductility
cal constraints, which is systematically compared to the trough for steels is usually located between 700 C and
simulations determining, when necessary, modifications 750 C; depending on steel composition, mainly in low
of the input variables. Such modifications are performed carbon steels, which is the temperature where the g2a
by observing the functional limits of each variable. The (austenite–ferrite) transformation starts, so the surface
present study was conducted to attempt maximum temperature must be less than
casting speed which depends on the settings of operating
Position ðLM Þ ) Tsurface oTg2a :
parameters, such as changes in the primary (mold) and
secondary cooling (sprays), reflected in heat transfer The upper limit of the low ductility trough corresponds
coefficients. These settings are defined as those which to the transition between the transgranular fracture and
make it possible to run the caster at its maximum intergranular fracture ðTtransition Þ: Depending on the
productivity, minimum cost and to cast defect-free composition of steel, this upper temperature limit can
products. vary between 900 C to 1100 C:

3.2.1. Metallurgical criteria Position ðLM Þ ) Tsurface > Ttransition :

1. Shell thickness at the mold exit ðSM Þ: The shell Limiting the strand surface above the upper limit of the
thickness at the mold exit must be greater than some low ductility temperature, transversal cracks are also
minimum value ðSmin Þ; which is considered to be about reduced. Longitudinal cracks at the unbending point are
10% of the casting thickness. This constraint avoids more usual in steels with carbon contents of about 0.08–
breakout occurrences caused by extraction stresses and 0.14%, the maximum value being observed to be about
liquid ferrostatic pressure, and can be written as 0.12%C. In this work it was considered that the strand
Position ðLmold exit Þ ) SM > Smin ¼ 0:1ecasting : surface temperature is kept above the upper limit of the
low ductility range, called Tmin surface :
2. Metallurgical length ðLM Þ: The solidification of the 4. Reheating between zones ðTmax surface  Tmin surface ):
ingot has to be complete before the point where a high The reheating effect occurs when the strand passes from
deformation is given (unbending point) in order to avoid a spray cooling zone with a high cooling efficiency to
internal and transverse cracking and centerline segrega- one with a lower cooling rate, and must be limited as a
tion. This constraint is function of steel grade and casting operating para-
Position ðLM Þ ) Tcenter oTsolidus ; meters. This reheating leads to the development of
tensile stresses at the solidification front, which can
that means that the center of the strand ðTcenter Þ must be induce cracking. The maximum permissible reheating
at a temperature lower than the solidus temperature range along the machine has been chosen to be equal to
ðTsolidus Þ at the unbending point. 100 C in order to avoid midway surface cracking
3. Temperature at unbending point ðTLM Þ: The strand (Brimacombe et al., 1984).
surface ðTsurface Þ must be at a temperature outside the

low ductility trough observed in steels and at a Position ðLsprays Þ ) Tmax surface  Tmin surface p100 C:
Table 1
Definition of the main rules used in the knowledge base system

Classes Objects Relations Consequences References

Steel composition
Division:
Low carbon: o0:25% C o0:10% Heat transfer decreases as Surface cracks, breakout (mold) (Brimacombe and
the %C increases Sorimachi, 1977),
0.10–0.14% Billet surface is rougher Surface cracks, breakout (mold) (Brimacombe et al., 1986,
(deeper oscillation mark) Brimacombe, 1999)
0.12% Lower heat transfer rate Longitudinal, midway cracks, (Wolf and Kurz, 1981)

C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

(thin solidified shell) breakout (mold)
0.17% (peritectic) d–g phase change (B1490 C) External, internal transverse (Billany et al., 1991)
cracks
0.17–0.25% Reduced ductility at elevated Transversal, longitudinal cracks (Samarasekera and
temperature (solidification front) Brimacombe, 1982a;
Samarasekera et al.,
1982b; Samarasekera and
Brimacombe, 1988)
Medium carbon: 0:25–0:50% C 0.25–0.38% Favor equiaxed grain zone Difficult crack propagation (Kumar et al., 1999)

ARTICLE IN PRESS
0.42% Higher heat flux Large solidified shell (mold) (Van Drunen et al., 1975)
High carbon: > 0:50% C 0.40–0.77% Large columnar zone Breakout (small solidified shell)
(lowest heat transfer)
0.77% (eutectoid) Long freezing range ð100 CÞ Breakout (small solidified shell),
cracks
> 0:77% Susceptibility of crack formation External, internal cracks,
at elevated temperatures breakout, laps, bleeds

Phase transformations
o 0.09% C: L, L+d; d; d þ g; g; g þ a; a; a+P d–g phase change External, internal cracks
(B1400–1485 C) (expansion)
0.09–0.17% C L, L+d; d þ g; g; g þ a; a; aþP d–g phase change (1485 C) External, internal cracks
(expansion)
0.17–0.53% C L, L+d; L+g; g; g þ a; aþP g–a phase change External, internal cracks
ðB910–727 C) (contraction)
0.53–0.77% C L, L+d; L+g; g; g þ a; aþP
0.77% C L, L+g; g; P
> 0:77% C L, L+g; g; g þ Fe3 C; P+Fe3 C

Element alloys
Hydrogen (H) o2 ppm Minimize bubbles of gases Pinholes, blowholes (Billany et al., 1991)
(surface/subsurface)
Oxygen (O) o10 ppm (Brimacombe and
Sorimachi, 1977;
Brimacombe, 1999)
Nitrogen (N) o20 ppm Minimize bubbles of gases Pinholes, blowholes (Thomas et al., 1987)
(surface/subsurface)
Mn:S ratio > 25–30 Avoid crack formation Cracks in grain boundaries
in interdendritic liquid (surfaces are smooth)

517
 518
Table 1 (continued )

Classes Objects Relations Consequences References

Phosphorus (P) o0:017% Decreases columnar zone Difficult crack formation and (Clyne et al., 1982)
propagation
Sulphur (S) o0:015% Formation of FeS Superficial, corner, longitudinal,
ðTf ¼ 1200 CÞ midway cracks
Copper (Cu) o0:2% Low melting impurities Cracks in grain boundaries
in grain boundaries (surfaces are smooth)
Cu:Sn ratio o4 Minimizes craze crack Craze crack

C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

formation in surface
Ni:Cu ratio >1 Form a miscible alloy Minimize craze crack formation
with a higher Tf
Aluminium (Al) o0:02% Formation of AlN Surface transversal, grain
ð900 CÞ boundaries cracks
Niobium and vanadium (Nb, V) o1% Formation of nitrites, Surface transversal, grain
carbides boundaries cracks
Manganese (Mn) o1% Formation of oxides Transversal cracks
Chromium (Cr) > 3% Formation of oxides Internal cracks
Titanium (Ti) o0:004% Minimizes AlN formation Minimize internal cracks

ARTICLE IN PRESS
Transformation temperatures TL ¼ 1537  88ð%CÞ  25ð%SÞ  5ð%CuÞ  8ð%SiÞ
5ð%MnÞ  2ð%MoÞ  4ð%NiÞ  1:5ð%CrÞ
18ð%TiÞ  2ð%VÞ  30ð%PÞ

TS ¼ 1535  200ð%CÞ  12:3ð%SiÞ  6:8ð%MnÞ

124:5ð%PÞ  183:9ð%SÞ  4:3ð%NiÞ  1:4ð%CrÞ
4:1ð%AlÞ

Cast structure
Columnar grain zone Small section Favor columnar zone Facilitate crack propagation
Equiaxed grain zone 0.13–0.20%C and 0.008–0.02%P Favor equiaxed zone Difficult crack propagation (Bobadilla et al., 1993)
0.17–0.38%C Favor equiaxed zone Difficult crack propagation (Van Drunen et al., 1975)
(medium %C)
Large section Favor equiaxed zone Difficult crack propagation
Superheat level o30 C Favor equiaxed zone Difficult crack propagation
(low, high %C)
Electro-magnetic stirring Favor equiaxed zone Difficult crack propagation

Mechanical properties
High temperature zone: TS Tx: 020:185%C Tx ¼ 40 C Cracks due P and S (Brimacombe and
Sorimachi, 1977)
0.45%C Tx ¼ 65 C
S > 0:025% Tx ¼ 80 C
Intermediate temperature zone: A3 to 1200 C Mn:S rate and carbides Cracks in grain boundaries
and nitrites
Low temperature zone: 700–900 C AlN, carbides and nitrites Cracks in grain boundaries
Mold
Liquid metal level
Meniscus depth: o100 mm (small depth) Overflow, distortion Rhomboidity, laps, bleeds, (Kumar et al., 1995,
transverse depressions Kumar et al., 1997,
Kumar et al., 1999)
=100 mm Recommended
> 100 mm (very depth) Small solidified shell Breakout (mold)
Fluctuations o75 mm Recommended Surface defects, breakout

Composition of the mold

C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

Alloy P: 200–300 ppm; Ag: 1000 ppm Smaller distortion Minimize rhomboidity, cracks (Samarasekera and
Brimacombe, 1982;
Samarasekera and
Brimacombe, 1988)
Smoothing temperature > 500 C Smaller distortion Minimize rhomboidity, cracks (Chandra et al., 1996)
su > 400 N=mm2 Smaller distortion Minimize rhomboidity, cracks
HB surface 100–500 Smaller distortion Minimize rhomboidity, cracks
Thermal conductivity > 70% to pure Cu Higher heat transfer rate Large solidified shell

Mold wall thickness

ARTICLE IN PRESS
Section Sections o200 200 mm Recommended Minimize rhomboidity, cracks (Brimacombe, 1993;
Brimacombe et al., 1986)
Thickness B12:7 mm Recommended Minimize rhomboidity, cracks (Bommaraju, 1991)
Distortion o0:05 mm Minimal Minimize rhomboidity, cracks (Samarasekera and
Brimacombe, 1982;
Samarasekera and
Brimacombe, 1988)
0.05–0:20 mm Unsatisfactory Rhomboidity, corner cracks (Fukada et al., 1999;
Chandra et al., 1993;
Chandra et al., 1996)
> 0:20 mm Severe Rhomboidity, external and
internal cracks

Mold constraint system

System Conditions of the corners Boiling at the channel gap Rhomboidity, longitudinal
corner cracks
Slots 2 or 4 (recommended 4) Smaller distortion Rhomboidity
Water channel gap 4:8 mm (recommended) Similar heat flux at 4 faces Rhomboidity

Mold tube alignment

Tolerances o0:5 mm (in all faces) Non-symmetrical cast structure Breakout
Corners o4 mm (same radius of the mold/tube) Different heat flux between Transverse, longitudinal
corner/face corner cracks, breakout

Taper mold
Straight No taper Alloy composition Breakout, rhomboidity (Samarasekera and
(low %C) Brimacombe, 1982;
Samarasekera and
Brimacombe, 1988)
Single or multiple taper Discrete taper Recommended (high %C) Breakout, rhomboidity

519
Parabolic Continuous taper More recommended
 520
Table 1 (continued )

Classes Objects Relations Consequences References

Cooling water quality

Temperature range ðDTÞ o8 C Recommended Rhomboidity, breakout, (Bakshi et al., 1993)
cracks (boiling)
Presence of scale deposits Yes or no Heat transfer efficiency (Berryman et al., 1989)
Color and properties Red Iron oxide and corrosion Rhomboidity, breakout, (Chandra et al., 1996)
cracks (boiling)
Black Magnetite Fe3 O4 or carbonaceous: (Mills et al., 1991)

C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

oil, grease
Light Excessive hardness
Thickness o20 mm Recommended
Impurities o5 ppm Recommended

Cooling water velocity

and pressure
Velocity (v) > 12 m=s Recommended Rhomboidity (boiling)
Pressure (P) Inlet and outlet Outlet > 135 kPa; inlet Rhomboidity (boiling)

ARTICLE IN PRESS
> 400 kPa

Mold oscillation
Frequency ðf Þ o4 Hz or 240 cpm Oscillation marks Transversal, longitudinal
surface cracks
Stroke length ðSÞ 9–16 mm Oscillation marks Transversal, longitudinal
surface cracks
Mold lead (ML) > 4 mm
Negative strip time ðtn Þ 0.12–0:15 s Oscillation marks Transversal, longitudinal
surface cracks, laps

Mold lubrication
Composition of the flux Elements SiO2 ; CaO; MgO; Al2 O3 ; (Pinheiro et al., 1995,
TiO2 ; Fe2 O3 ; MnO2 ; Pinheiro et al., 1996,
Na2 O; K2 O; B2 O3 ; Li2 O; Pinheiro et al., 2000,
F; C; CO2 Mills et al., 1991)
1:53%Cao þ 1:51%MgO þ 1:94%Na2 O
Inclusion absorption rate (Bi) Bi ¼ Large Bi Minimize breakout (Billany et al., 1991)
1:48%SiO2 þ 0:10%Al2 O3

3:55%Li2 O þ 1:53%CaF2
þ
1:48%SiO2 þ 0:10%Al2 O3

distance from meniscus to position where T ¼ Tf

Lubrication index (LI) LI ¼ Close to 1 Minimize breakout (Brendzy et al., 1993);
distance from meniscus to bottommold
(Branion, 1986);
(Bommaraju, 1991)


pf 500tn Vcasting
Depth of the molten flux (Yp) Yp ¼ S sin  þd 6–12 Minimize breakout
2 f
 ARTICLE IN PRESS
C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527 521

Note: d—delta ferrite, g—austenite, a—alpha ferrite, P—perlite, Fe3 C—cementite, L—liquid, S—solid, TL —liquidus temperature, TS —solidus temperature, Tf —fusion temperature, f —frequency,
3.2.2. Equipment constraints
Brimacombe et al., 1980,
Brimacombe, 1999)
* Water flow: The water flow rate in a given region
(Bommaraju, 1991)

(or spray) has a lower and an upper limit depending

(Brimacombe and

(Brimacombe and
Sorimachi, 1977)
Sorimachi, 1977;

on the hydraulic system, which is given in heat

transfer coefficients ðhs=s Þ (Brimacombe et al., 1980;
Mizikar, 1970; Nozaki et al., 1978; Bolle and
Moureau, 1979):
Position ðLsprays Þ ) hmax sprays phs=s phmin sprays :

* Casting speed (Vcasting ): The casting speed is bounded

Midway crack close to the

External, internal cracks

with a minimum and maximum value, given by
Midway crack close to
the solidification front

Vmin casting pVcasting pVmax casting :

solidification front

Pinch-roll cracks
Midway cracks

Midway cracks

Central cracks

Objective and constraint functions used in the

Rhomboidity

optimization framework were formulated to represent

the productivity of the machine, quality of the cast
strand and casting speed. Machine productivity is
characterized by limitation of casting speed, metallurgi-
Intermediate low ductility zone

Intermediate low ductility zone

Intermediate low ductility zone

cal length and spray cooling, and the metallurgical

High temperature zone of low
Reheating of the billet surface

Reheating of the billet surface

Asymmetrical spray cooling

constraints are solid shell thickness, metallurgical

length, surface temperatures and reheating between
End of solidification

sprays zones. The objective is to keep a cost function

(J), defined as a sum of individual values of each
(strain/stress)

(strain/stress)

constraint (i), close to zero. The process starts with

nominal values of operating parameters, and as a
ductility

function of results simulated by the heat transfer

mathematical model (temperature field in the strand),
the cooling conditions are modified in such a way that
the final billet/slab metallurgical quality is assured. Each
violation of any constraint corresponds to one numer-
ical increase in this individual objective function.
When the cost function reaches ‘‘zero’’, the casting
speed can be increased by a value DVcasting ; and
the search begins again. The cooling criteria are
Relations to heat-transfer coefficients

formulated in such a way that the lower values of

thermal gradients between P cooling zones correspond to
S—stroke length, tn —negative strip time, d—liquid metal fluctuations.

the best situation, with ni¼1 Ji ¼ 0: For each criterion a

weight (w) was used denoting the relative importance of
the criterion. The solid shell thickness at mold exit and
the point of complete solidification have maximum
weight (10), and surface temperature and thermal
gradients at the sprays zones have minimum weight
Tcenter oTs
o1350 C
> 1100 C

> 1100 C

> 1100 C
o100 C

o100 C

(1). Eq. (30) presents the formulation for the objective

function J:
Xn
Ji  Ji min
J¼ wi : ð30Þ
i¼1
J i max  Ji min
Point of complete solidification
Minimum surface temperature

Minimum surface temperature

Minimum surface temperature

The genetic algorithm applied for the continuous

casting optimization consists of:
Step 1: generate an initial population of results
Maximum reheating

Center temperature
Surface reheating

simulated by using the input parameters (nominal);

Unbending point
Radiation zone

Step 2: compute cost function;

Water flux

Step 3: store parameters setting;

Sprays

Step 4: modify cooling conditions in each region

where the constraint was violated;
 ARTICLE IN PRESS
522 C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

Step 5: apply a genetic operator to determine new speed. In the water flow rate, a step of 0:03 l=s
parameter of process; between the upper and lower limits was used, and for
Step 6: generate new results; the casting rate, a step of 0:001 m=s was used. Fig. 5
Step 7: if cost function decreases, then SJmin is the shows the relation between the knowledge base, the
result; genetic algorithm and the mathematical solidification
Step 8: if SJ ¼ 0; increase Vcasting and go to step 1; model.
Step 9: repeat steps 1–7 until SJ ¼ 0:
The genetic algorithm was developed using a binary
encoding, in the most common form, the Simple 4. Experimental procedure
Genetic Algorithm (SGA). The mathematical model
of the solidification process computes the tempe- To validate the proposed solidification mathematical
rature field in the strand and the solidified shell model and the use of the different fS formulations, the
thickness and assesses the metallurgical criteria. Each results of the calculations are compared with experi-
set of results of the simulation was used to form an mental data obtained in an experimental setup mon-
individual, and a set of individual represents a popula- itored by thermocouples located both in the mold and in
tion, where each member has a potential solution the metal. The casting assembly used in static solidifica-
encoded in it. The simulations are performed varying tion experiments is shown in Fig. 6. The main design
the values of the sprays water flow, or spray heat criterion was to ensure a dominant unidirectional heat
transfer coefficients, and when possible, the casting flow during solidification.

Tundish:
temperature Input of operational Comparison with
area, width, parameters knowledge base
capacity, steel
composition

Mold:
composition, metal
level, support,
thickness, taper,
cooling Heat transfer Possible changes
mathematical model for improvements
Oscillation:
frequency, stroke
and negative strip
time

Lubrification:
oil or powder flux

Sprays:
water temperature Determination of process variables
Rules conducive to better
and flow rate to be implemented into the
operational conditions
equipment based on metallurgical
(Search/Priority)
constraints
wt % carbon:
cracks
oscillation marks
End
Mold cooling:
cracks Priority Scale
rhomboidity
Report with thermal 1 – variation of water flow rate
Meniscus: 2 – variation of casting speed
profile, solidified shell
cracks
and possible defects
inclusions

Lubrification:
breakouts
laps

Sprays cooling:
segregation
cracks

Fig. 5. Block diagram showing the relation between the heat flow model and knowledge base/genetic algorithm.
 ARTICLE IN PRESS
C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527 523

FDM program
Thermocouples T
Automatic
Chamber Data acquisition Search
Insulating t
cover IN OUT RS 232

Mold Insulating walls

Graphical
display

Top view Side view

123 24 63

24 Pouring 24
Channel

20 44
A 20 A
3
100 60
10 Thermocouples
3 Steel
Casting
Heat Flux Chill
Chill Chamber

24 Insulating Material 24

Fig. 6. Casting arrangement and position of thermocouples in the mold wall and in the metal (mm).

Both copper and steel molds were used, with the heat- model with the transient metal/mold heat transfer
extracting surface being polished. Experiments were coefficient, hm=s ; described in previous articles (Santos
performed with Sn210 wt% Pb and Al24:5 wt% Cu et al., 2001; Quaresma et al., 2000). Fig. 7 shows typical
alloys, with the liquid metal being poured at a examples of temperature data collected in metal and
temperature of 10% above the liquidus temperature. chill during the course of solidification of Sn210wt% Pb
The thermophysical properties of these alloys and chill alloy (Fig. 7A) and Al24:5 wt% Cu alloy (Fig. 7B).
are summarized in previous articles (Santos et al., 2001; These experimental thermal responses were compared to
Quaresma et al., 2000). These alloys were chosen due to those numerically simulated using the fs formulation
two main reasons: all their thermophysical properties, given by Scheil’s equation. In any case a good agreement
which are fundamental for calculations are available in can be observed.
the literature, and they are quite easy to manipulate in To validate the application of the mathematical model
the laboratory. Temperatures in the chill and in the for the continuous casting process, a set of simulations
casting were monitored during solidification via the was performed and the results of both billet and slab
output of a bank of thermocouples (1:6 mm diameter) surface temperatures were compared with literature
accurately located with respect to the metal/mold data. The input parameters used in these simulations
interface, as indicated in Fig. 6. These sets of exper- are presented in Table 2. For the billet case, experi-
iments were planned for a preliminary validation of the mental results of a high carbon steel (SAE 1080) were
mathematical model with experimental data of solidifi- analyzed, and the simulations were based on metal/
cation. mold heat transfer coefficients proposed by Toledo
To demonstrate the applicability of the solidification et al. (Toledo et al., 1993) and the metal/sprays heat
mathematical model to the continuous casting process, transfer coefficients proposed by Brimacombe et al.
simulations will be compared with experimental data (Brimacombe et al., 1980). For the slab, the literature
from the literature. data for a low carbon steel (SAE 1012), as well as the
used formulations for heat transfer coefficients in
mold and in the spray zones, were proposed by
5. Results and discussion Samarasekera and Brimacombe (1988) and Lait et al.
(1974), respectively.
The temperature files containing the experimentally Fig. 8 shows the comparison between experimental
monitored temperatures during solidification in static and simulated surface temperature profiles for the steel
molds were compared to the proposed mathematical billet. The search space used in this analysis is shown in
 ARTICLE IN PRESS
524 C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

270

700
240

600
210

180 500
Temperature [°C]

Temperature [°C]
150
400

120
300
90
200
60
Thermocouple (metal) 20 mm interface
Thermocouple (metal) 20 mm interface
Thermocouple (mold) 3 mm interface 100
30 Thermocouple (mold) 3 mm interface
Simulated
Simulated
0 0
0 50 100 150 200 250 300 350 400 450 500 0 25 50 75 100 125 150 175 200 225 250
(A) Time [s] (B) Time [s]

Fig. 7. Typical experimental thermal responses of thermocouples at two locations in casting and chill, compared with numerical simulations:
(A) Sn–10 wt% Pb and copper mold; (B) Al–4.5 wt% Cu and steel mold.

Table 2
Input parameters for billet and slab continuous casting conditions (Louhenkilpi, 1994; El-Bealy et al., 1995)

Units Billet Slab

Dimensions mm 160 160 1680 220

Mold length mm 600 700
Water flow rate l/s 20.08 20.08

Water temperature C 25 25
Metal 1080 Steel 1012 Steel
Specific heat J=kg K cS ¼ 678 cL ¼ 758 cS ¼ 700 cL ¼ 700
Density kg=m3 rS ¼ 7850 rL ¼ 7300 rS ¼ 7400 rL ¼ 7400
Thermal conductivity W=m K kS ¼ 30:13 kS ¼ 34:50 kS ¼ 28 kS ¼ 28
Latent heat of fusion J=kg 260,000 260,000

Solidus temperature C 1360 1471

Liquidus temperature C 1458 1541
Sprays 1 (length/flow rate) (m) (l/s) 2.800 1.47 0.485 3.83
Sprays 2 1.800 1.15 0.900 3.58
Sprays 3 2.700 0.55 1.285 2.66
Sprays 4 — — 1.580 3.33
Sprays 5 — — 1.280 2.10
Sprays 6 — — 1.540 1.66
Sprays 7 — — 2.380 4.66
Sprays 8 — — 4.500 1.96
Casting rate m/s 0.0245 0.0183

Pouring temperature C 1485 1600
Metallurgical length m 10 14

Table 3. The number of candidate parameter settings is spray zone up to the radiation zone, the optimized
about 36 103 : It was considered a population of 50 surface temperature along the equipment is more
parameter settings, and to alter the parameter vector, homogenous and with lower thermal gradients between
the uniform mutation was applied as the genetic adjacent spray zones. The water flow rates shown in
operator. Although every parameter shown in Table 3 Table 4 indicate that the optimized profile is accom-
is taken as a variable, the search for each of them was panied by a decrease of flow rate in spray zones 1 and 2
restricted within a range of values provided on the basis (18:3% in zone 1 and 30% in zone 2), and an increase in
of the current industrial practice. zone 3 (21:4%). In this case, it was not possible to
It can be seen that a similar simulated profile is increase the casting speed because two metallurgical
attained for the two considered conditions (nominal and constraints were violated (solid shell thickness at mold
optimized) up to the end of spray zone 2. From this exit and point of complete solidification).
 ARTICLE IN PRESS
C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527 525

1500 Table 4
160x160 mm Billet Water flow rates for nominal and optimized operational conditions
1400 600 mm Mold
Strand Surface Temperature [C]

1080 Steel Billet Slab

1300
Spray Nominal Optimized Nominal Optimized
1200 zones water flow water flow water flow water flow
Tmax rates (l/s) rates (l/s) rates (l/s) rates (l/s)
1100
1 1.47 1.20 3.83 2.58
1000 Tmin 2 1.15 0.80 3.58 3.20
3 0.55 0.70 2.66 2.66
1 Sprays

2 Sprays

3 Sprays
Mold

900 4 — — 3.33 2.72

5 — — 2.10 2.30
800 6 — — 1.66 2.50
Industrial - Louhenkilpi, 1994
7 — — 4.66 4.00
700 8 — — 1.96 2.92
Simulated - nominal
600 Simulated - optimized

500
1600
0 1 2 3 4 5 6 7 8 9 10 1680x220 mm Slab
1500
Distance from meniscus [m] 1400
700 mm Mold
1010
1012 Steel
Fig. 8. Comparison between results of experimental and simulated 1300
(nominal and optimized) strand surface temperature during contin- Strand Surface Temperature [C] 1200
uous casting of SAE 1080 steel billet.
1100
1000
The simulated surface temperature profiles for the 900

6 Sprays

7 Sprays

8 Sprays
5 Sprays
2 Sprays

4 Sprays
1 Sprays

3 Sprays
slab are compared with experimental results in Fig. 9. It 800
Mold

can be observed that the optimized surface temperature 700

along the equipment is more homogenous and with 600
lower thermal gradients from a spray zone to the next, 500
mainly in zones 6 and 8. As shown in Table 4, the 400 Industrial - El-Bealy,1995
optimized profile produces a decrease of water flow rate 300 Simulated - nominal
in zones 1, 2, 4 and 7 (32% in zone 1; 10% in zone 2; 18% 200 Simulated - optimized
in zone 4, and 14% in zone 7), and an increase in zones 100
5, 6 and 8 (8% in zone 5; 33% in zone 6, and 32% in zone 0
8), and zone 3 has the same value. In both cases (billet 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
and slab), the strand is completely solidified just before Distance from meniscus [m]
the unbending point. For the slab case, it was not Fig. 9. Comparison between results of experimental and simulated
possible to increase the casting speed due to the (nominal and optimized) strand surface temperature during contin-
violation of a metallurgical constraint (solid shell uous casting of an SAE 1012 steel slab.
thickness at mold exit). This feature of GA acts as a
natural safeguard against any solution which is likely to
appear in other techniques where variables need not be 6. Conclusions
specified (Chakraborti et al., 2001).
The results of optimized water flow rates for the A mathematical model of solidification working
different spray zones are presented in Table 4 for both integrated with a genetic search algorithm and a
cases. knowledge base of operational parameters has permitted

Table 3
Parameter space for optimization of a 1080 steel billet

Parameter Minimum Nominal Maximum Discretization Number of

step possible values

Casting speed (m/s) 0.0195 0.0245 0.0295 0.001 11

Water flow 1 (l/s) 1.20 1.47 1.74 0.030 19
Water flow 2 (l/s) 0.80 1.15 1.32 0.035 16
Water flow 3 (l/s) 0.43 0.55 0.73 0.030 11
 ARTICLE IN PRESS
526 C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527

the optimization of cooling conditions (mold and billets, Part 2. Lubrification and oscillation mark formation.
secondary cooling) for the continuous casting of Ironmaking and Steelmaking 20 (1), 63–74.
steel billets and slabs. The search method supported Brimacombe, J.K., 1993. Empowerment with knowledge—toward the
intelligent mold for the continuous casting of steel billets. Iron and
by the knowledge base based on metallurgical criteria Steelmaker 20 (11), 35–47.
permits a more homogenous strand surface temperature Brimacombe, J.K., 1999. The challenge of quality in continuous
profile to be attained and with lower thermal gradients casting processess. Metallurgical and Materials Transactions 30A,
between adjacent sprays zones. The reasonable to good 1899–1912.
agreement observed between experimental data and Brimacombe, J.K., Sorimachi, K., 1977. Crack formation in the
continuous casting of steel. Metallurgical Transactions 8B,
simulations for both billet and slab analyzed in the
489–505.
present study, permits to conclude that the formulations Brimacombe, J.K. Argarwal, P.K., Baptista, L.A., Hibbins, S.,
used to calculate mold and spray heat transfer Prabhakar, B., 1980. Spray cooling in the continuous casting
coefficients are able to provide an appropriate descrip- of steel. Steelmaking Proceedings of the NOH–BOS Conference
tion of heat transfer efficiencies along the different (ISS-AIME), Washington, DC, pp. 109–123.
cooling regions, as well as to determine the maximum Brimacombe, J.K., Samarasekera, I.V., Lait, J.E., 1984. Continuous
Casting—Heat Flow, Solidification and Crack Formation. Iron
casting speed to attain highest product quality. How- and Steel Society of AIME, Warrendale, PA.
ever, more accurate simulations can be achieved if Brimacombe, J.K., Samarasekera, I.V., Bommaraju, R., 1986.
particular heat transfer formulations are developed for Optimum design and operation of molds for the continuous
each continuous caster by using approaches like casting of steel billets. Fifth International Iron and Steel Congress,
comparison between theoretical-experimental thermal Steelmaking Proceedings, Washington, DC, No. 69, pp. 409–423.
Chakraborti, N., Kumar, R., Jain, D., 2001. A study of the continuous
profiles or data obtained from ingot microstructure.
casting mold using a pareto-converging genetic algorithm. Applied
Mathematical Modelling 25, 287–297.
Chandra, S., Brimacombe, J.K., Samarasekera, I.V., 1993. Mould—
Acknowledgements strand interaction in continuous casting of steel billets—Part 3,
Mould heat transfer and taper. Ironmaking and Steelmaking 20
(2), 104–112.
The authors acknowledge the financial support Chandra, S., Samarasekera, I.V., Brimacombe, J.K., Bakshi, I.A.,
provided by FAPESP (The Scientific Research Founda- Walker, B.N., 1996. Online monitoring of mold–shell interaction
tion of the State of S*ao Paulo, Brazil) and CNPq during continuous casting of steel billets. Ironmaking and
(The Brazilian Research Council). Marco Ol!ıvio Sotelo Steelmaking 23 (6), 512–520.
is also acknowledged for helping with the computer Cheung, N., Garcia, A., 2001. The use of heuristic search technique for
the optimization of quality of steel billets produced by continuous
programming.
casting. Engineering Applications of Artificial Intelligence 14,
229–238.
Clyne, T.W., Wolf, M., Kurz, W., 1982. The effect of melt composition
References on solidification cracking of steel, with particular reference to
continuous casting. Metallurgical Transactions 13B, 259–266.
El-Bealy, M., Leskinen, N., Fredriksson, H., 1995. Fluctuated cooling
Bakshi, I.A., Brendzy, J.L., Walker, N., Chandra, S., Samarasekera,
I.V., Brimacombe, J.K., 1993. Mould—strand interaction in conditions and solid shell resistance in continuously casting steel
continuous casting of steel billets—Part 1, Industrials trials. slabs. Ironmaking and Steelmaking 22 (3), 246–255.
Ironmaking and Steelmaking 20 (1), 54–62. Filipic, B., Saler, V., 1998. Evolving parameter setting for continuous
Berryman, R., Samarasekera, I.V., Brimacombe, J.K., 1989. Cooling casting of steel. Proceedings of the Sixth European Congress on
water flow in continuous-casting billet molds. Iron and Steel Intelligent Techniques and Soft Computing—EUFIT’98, Aachen,
Society Transactions 10, 19–27. Germany (1), pp. 444–449.
Billany, T.J.H., Normanton, K.C., Mills, K.C., Grieveson, P., 1991. Fukada, N., Marukawa, Y., Abe, K., Ando, T., 1999. Development of
Surface cracking in continuous cast products. Ironmaking and mold (HS Mold) for high speed casting. Canadian Metallurgical
Steelmaking 18 (6), 403–410. Quarterly 38 (5), 337–346.
Bobadilla, M., Jolivet, J.M., Lamant, J.Y., Larrecq, M., 1993. Hills, A.W.D., 1969. A generalized integral-profile method for the
Continuous casting of steel: a close connection between solidifica- analysis of unidirectional heat flow during solidification. Transac-
tion and industrial process development. Materials Science and tions of the Minerals, Metals and Materials Society (AIME) 245,
Engineering 173A, 275–285. 1471–1479.
Bolle, E., Moureau, J.C., 1979. Spray cooling of hot surfaces: a Kumar, S., Meech, J.A., Samarasekera, I.V., Brimacombe, J.K., 1993.
description of the dispersed phase and a parametric study of heat Knowledge engineering an expert systems to troubleshoot quality
transfer results. Proceedings of the Two Phase Flows and Heat problems in the continuous casting of steel billets. Iron and
Transfer, III, NATO Advanced Study Institute, Cosica, France, Steelmaker 20 (9), 29–36.
pp. 1327–1346. Kumar, S., Walker, B.N., Samarasekera, I.V., Brimacombe, J.K.,
Bommaraju, R., 1991. Optimum selection and application of mold 1995. Chaos at the meniscus—the genesis of defects in continuously
fluxes for carbon steels. Mold operation for quality and productiv- cast steel billets. 13th PTD Conference Proceedings, Nashville, TN,
ity. Iron and Steel Society of AIME 95–110. pp. 119–141.
Branion, R.V., 1986. Mold fluxes for continuous casting. Iron and Kumar, S., Samarasekera, I.V., Brimacombe, J.K., 1997. Mold
Steelmaker 13 (9), 41–50. thermal response and formation of defects in the continuous
Brendzy, J.L., Bakshi, I.A., Samarasekera, I.V., Brimacombe, J.K., casting of steel billets—laps and bleeds. Iron and Steelmaker 24 (6),
1993. Mould–strand interaction in continuously casting of steel 53–69.
 ARTICLE IN PRESS
C.A. Santos et al. / Engineering Applications of Artificial Intelligence 16 (2003) 511–527 527

Kumar, S., Meech, J.A., Samarasekera, I.V., Brimacombe, J.K., Samarasekera, I.V., Brimacombe, J.K., 1982. Thermal and mechanical
Rakocevic, V., 1999. Development of intelligent mold for online behavior of continuous casting billets molds. Ironmaking and
detection of defects in steel billets. Ironmaking and Steelmaking 26 Steelmaking 1, 1–15.
(4), 269–284. Samarasekera, I.V., Brimacombe, J.K., 1982a. The influence of mold
Lait, J.E., Brimacombe, J.K., Weinberg, F., 1974. Mathematical behavior on the production of continuous cast steel billets.
modelling of heat flow in the continuous casting of steel. Metallurgical Transactions 13B, 105–116.
Ironmaking and Steelmaking (Quarterly) 2, 90–97. Samarasekera, I.V., Brimacombe, J.K., 1988. Heat extraction cap-
Lally, B., Biegler, L.T., Henein, H., 1991a. Optimization and ability of continuous casting billet molds. W.O. Philbrook
continuous casting: Part I. Problem formulation and solution Memorial Symposium Conference Proceedings, Toronto, Canada,
strategy. Metallurgical Transactions 22B, 641–648. pp. 157–171.
Lally, B., Biegler, L.T., Henein, H., 1991b. Optimization and Samarasekera, I.V., Anderson, D.L., Brimacombe, J.K., 1982b. The
continuous casting: Part II. Application to industrial caster. thermal distortion of continuous-casting billets molds. Metallurgi-
Metallurgical Transactions 22B, 649–659. cal Transactions 13B, 91–104.
Larreq, M., Birat, J.P., 1982. Optimization of casting and cooling Samarasekera, I.V., Brimacombe, J.K., Wilder, K., 1994. The
conditions on steel continuous casters—implementation of optimal Pursuit of Steel Billet Quality. Iron and Steelmaker 21 (3),
strategies on slab and bloom casters. Iron and Steel Society of 53–63.
AIME 1, 273–282. Santos, C.A., Quaresma, J.M.V., Garcia, A., 2001. Determination of
Louhenkilpi, S., 1994. Study of heat transfer in a continuous billet transient interfacial heat transfer coefficients in chill mold castings.
casting machine. Scandinavian Journal of Metallurgie 23, 9–17. Journal of Alloys and Compounds 319 (1–2), 174–186.
Mills, K.C., Billany, T.J.H., Normanton, A.S., Walker, B., Grieveson, Santos, C.A., Spim Jr., J.A., Ierardi, M.C.F., Garcia, A., 2002. The use
P., 1991. Causes of sticker breakout during continuous casting. of artificial intelligence technique for the optimization of process
Ironmaking and Steelmaking 18 (4), 253–265. parameters used in the continuous casting of steel. Applied
Mizikar, E.A., 1970. Spray cooling investigation for continuous Mathematical Modeling 26 (11), 1077–1092.
casting of billets and blooms. Iron and Steel Engineering 47 (6), Spim, J.A., Garcia, A., 2000. Numerical modeling and optimization of
53–60. zone refining. Numerical Heat Transfer 38B, 75–92.
Nozaki, T., Matsumo, J., Murata, K., Ooi, H., Kodama, M., 1978. A Spim, J.A., Santos, C.A., Ierardi, M.C.F., Garcia, A., 1997. The use of
secondary cooling pattern for preventing surface cracks of artificial intelligence techniques in a heat transfer model for the
continuous casting slab. Transactions of ISIJ 18, 330–338. control of solidification during continuous casting. Proceedings of
Pinheiro, C.A., Samarasekera, I.V., Brimacombe, J.K., 1995. the Fourth Decennial International Conference on Solidification
Mold flux for continuous casting of steel. Iron and Steelmaker 22 Processing, Sheffield, pp. 166–169.
(8), 41–43. Toledo, G.A., Lainez, J., Cirion,! J.C., 1993. Model optimization of
Pinheiro, C.A., Samarasekera, I.V., Brimacombe, J.K., 1996. continuous casting steel secondary cooling. Material Science and
Mold flux for continuous casting of steel. Iron and Steelmaker 23 Engineering 173A, 287–291.
(1), 51–52. Thomas, B.G., Samarasekera, I.V., Brimacombe, J.K., 1987. Mathe-
Pinheiro, C.A.M., Samarasekera, I.V., Brimacombe, J.K., Howes, B., matical model of the thermal processing of steel ingots: Part 1,
Gussias, O., 2000. Mold heat transfer and continuously casting Heat flow model. Metallurgical Transactions 18B, 119–130.
billet quality with mold flux lubrification. Part 2: Quality issues. Van Drunen, G., Brimacombe, J.K., Weinberg, F., 1975. Internal
Ironmaking and Steelmaking 27 (2), 144–159. cracks in strand-cast billets. Ironmaking and Steelmaking (Quar-
Quaresma, J.M.V., Santos, C.A., Garcia, A., 2000. Correlation terly) 2, 125–133.
between unsteady-state solidification conditions, dendrite spacings Wolf, M., Kurz, W., 1981. The effect of carbon content on
and mechanical properties of Al–Cu Alloys. Metallurgical and solidification of steel in the continuous casting mold. Metallurgical
Materials Transactions 31A, 3167–3178. Transactions 12B, 85–93.