International Journal of Heat and Mass Transfer 166 (2021) 120765

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage: [Link]/locate/hmt

Interfacial Heat Transfer Coefficient in Unidirectional Permanent Mold

Casting: Modeling and Inverse Estimation
Gianfranco de M. Stieven a,∗, Daniele dos R. Soares a, Edilma P. Oliveira b, Erb F. Lins a
a
Graduate Program of Mechanical Engineering, Institute of Technology, Federal University of Pará - Av. Augusto Correa, N 1 - Belém, PA, 66075-900, Brazil
b
Faculty of Mechanical Engineering, Federal University of Southern and Southeastern Pará, Folha 17, Quadra 04, Lote Especial, Nova Marabá, Marabá, PA,
68505-080, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The thermal control of the permanent mold casting process is highly dependent on the Interfacial Heat
Received 8 September 2020 Transfer Coefficient (IHTC). Based on that, many works proposed several models in order to estimate its
Revised 11 November 2020
behavior. This work intends to study three IHTC models present in the literature, as well as to propose
Accepted 27 November 2020
and analyse two new models. The proposed models are based on two distinct methodologies: thermal re-
sistance method and analytical comparison to the thermal gradient formulation at the metal/mold inter-
Keywords: face. All these models will be analysed in terms of their ability to reduce the error between the simulated
Aluminum Alloy and experimental thermal profiles by using an inverse method and verifying their respective parameter
Interfacial Thermal Resistance sensitivity based on the relative sensitivity coefficient. It is observed that the exponential formulation
Parameter Estimation
model, proposed in this paper, is able to reduce the numerical/experimental error more efficiently than
Sensitivity Analyses
the others methods in all three hypoeutectic alloys of the Al-Si system considered.
Metaheuristic Approach
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction other important internal phenomena to the solidification, such as

macrosegregation, Columnar-to-Equiaxed Transition (CET) and con-
Metal casting is a process in which metals, in the liquid state, vection.
are poured in a hollow geometry, called mold, for the manufacture Among all the thermal parameters related to heat exchange
of a variety of objects, from high-performance automotive com- in permanent mold casting, the most relevant is the Interfacial
ponents to decorative items. This process, besides being relatively Heat Transfer Coefficient (IHTC). This parameter mainly controls
simple, is widely used when the manufacture of a desired compo- the heat exchange and all parameters resulting therefrom, as well
nent has, for example, intricate geometry or internal channels and as other phenomena in liquid metal. In particular, investigating the
cavities. In general, practically all metallic components start with behavior of the IHTC is crucial to provide greater industrial control
the casting process, which can originate finished or semi-finished on manufacturing not only in this process, but also in hot forging
parts. The mass production of castings, when observed in techni- [4], die casting [5], squeeze casting [6] and all processes that in-
cal terms, presents reasonable complexity. The control of casting volve heat transfer and material quality. From the IHTC, in addition
defects such as shrinkage [1], porosity [2] and hot cracks [3] is to thermal parameters such as tip growth rate VL , local solidifica-
highly dependent on the selected operational variables, in which tion time ts and cooling rate TR , it is possible to estimate metal-
is included both metallurgical and technical parameters, such as lurgical characteristics such as dendritic arm spacing [7] and mor-
overheating temperature, alloy chemical composition, ingot mold phology transition [8], and mechanical properties such as fracture
material and coolant flux. strength [9], yield and ultimate tensile strength [10] and hardness
At this point, it is clear that the production of high-quality cast- [11].
ings depends of several parameters which should be properly eval- In order to represent the IHTC, several works have proposed
uated to provide a satisfactory manufacture. Among the most rel- models based on mold characteristics, such as statistical aspects
evant parameters for the control of the casting process, thermal of mold asperities [12], thermophysical properties, such as thermal
variables can certainly be mentioned. The heat exchange between conductivity [13], and thermodynamic properties, such as temper-
the liquid system and the coolant flux can define the intensity of ature [14] and pressure [15];
However, many relevant studies consider that the IHTC can be
∗
a constant function. Potkowski and Krane [16] analysed the ap-
Corresponding author.
E-mail address: [Link]@[Link] (G.d.M. Stieven).
pearance of channel segregates including equiaxed solidification

[Link]
0017-9310/© 2020 Elsevier Ltd. All rights reserved.
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

in the simulation for a cylindrical static casting of an Aluminum

alloy using a constant IHTC and geometric symmetry. Ge et al.
[17] proposed a numerical model to verify the ingot size effects on
macrosegregation in a Fe-0.33 wt.% C steel casting using a constant
IHTC on the top and on the sides of the mold. Also applying dif-
ferent constant functions, Thieme et al. [18] studied the effects of
a time-dependent magnetic field on a 2D directional solidification
of Gallium.
Several researches also modeled the IHTC using a power func-
tion. Silva-Santos et al. [19] used this to model a two-dimensional
inward solidification of Al-1.5 wt.% Fe alloy using finite difference
method. Vishweshwara, Gnanasekaran and Arun [20] also used a
temporal power function to model an one-dimensional transient
solidification heat transfer problem considering Sn-5 wt.% Pb alloy.
With the same purpose, Natsume et al. [21] used this function to
predict the thermal profiles of Al-1wt.% Si alloy directionally solid-
ified.
Fig. 1. Illustration of an upward unidirectional solidification device [29].
At last, de Barcellos et al. [22] and dos Anjos et al. [23] pro-
posed an overall heat transfer coefficient model, which will be
considered here as an IHTC model, considering the distance from Si system, which account for the majority of the aluminum cast-
the meniscus in the casting direction. In these papers, de Barcellos ings produced commercially [28].
et al., by finite difference method, and dos Anjos et al., by finite In this work, given the unidirectionality of the casting pro-
volume method, simulated a continuous casting process consider- cess under study, a 1D model will be used, which is presented
ing steel as material and proposed empirical coefficients to their by Oliveira at al. [29]. This approach will enable the optimization
formulations. of IHTC parameters due to reduced computational cost. However,
These three abovementioned models strongly need a empirical this simplified analysis can be expanded to more complex engi-
adjustment of IHTC. However, it is not easy to obtain, in these for- neering applications, such as the one presented by Wu, Ludwig and
mulations, the correlation of the parameters with the physical na- Kharicha [1].
ture of the problem. In view of this, it is relevant to propose new In summary, this paper has as main contributions:
IHTC models that can satisfactorily represent the thermal dynam-
ics present in the interface. In addition to numerically representing • The proposal of two physically consistent models for the IHTC
thermal behavior, as the models above, a model that has intuitive in directional solidification based on thermal resistance and
parameters is quite required. In addition, it is relevant to satisfac- thermal gradient analogy;
torily estimate these parameter values of the IHTC models in order • The estimate, by an inverse metaheuristic approach, of the pa-
to have accurate and precise results for engineering applications. rameters into the IHTC models proposed in this work and in
These estimates can be made by several methods, however there the literature, considering three Al-Si binary alloys;
are few studies that estimate the interface parameters in perma- • A relative sensitivity coefficient analysis of the parameters pre-
nent mold casting using metaheuristic optimizers [24]. In inverse sented in two models proposed by this paper and three models
applications, these methods are able to perform global and local exposed in the literature.
search in the domain under analysis, reducing the number of it-
erations necessary to perform the same task when compared to 2. Experimental Description
deterministic methods.
Based on that, this work will present two IHTC models and es- The experimental procedure mentioned in this Section was not
timate their parameters based on a metaheuristic optimizer. The performed by the authors of this work. However, for the reader’s
first IHTC model will consider a thermal resistance methodology, full comprehension of this paper, a brief description of the main
taking into account the physics of the heat exchange interface be- experimental steps will be made. For further details on the exper-
tween mold and ingot. The second will be modeled from the an- imental procedure, please consult the original work of Peres [30].
alytical comparison of the IHTC dynamic to the thermal gradient To obtain thermal profiles on a permanent mold casting experi-
at the metal/mold interface. These two proposals come from con- ment using a vertical upward unidirectional device, it is important
siderations widely known in engineering, which provide consistent to first determine the necessary quantity of each chemical element
models for the problem under study. For inverse estimation, based in mass based on the ingot mold volume, specific mass for each
on its satisfactory competence in this type of physical problem element and solute percentage. Subsequently, these chemicals ele-
compared to other optimizers [25], the Moth-Flame Optimization ments must be weighed on a precision scale. After this stage the
(MFO) method will be used. directional solidification apparatus is prepared.
In addition, it is important to have knowledge about the numer- First, the metal-mold interface and the interior of the ingot
ical relevance of the parameters present in each interfacial model, mold must be treated to reduce roughness. Greater care must be
in order to avoid unnecessary over-parameterization. This informa- taken with the metal-mold interface since it is mainly responsible
tion can be obtained by analysing the sensitivity of the parameters for the resistance to thermal changes during the process. On Fig. 1
in each interface model. Thus, the relative sensitivity coefficient of is presented an illustration of the solidification device under anal-
all parameters presented in the most used models in the literature ysis.
will be computed, as well as those proposed in this work. Once the metal-mold interface and the ingot mold are prepared,
Three hypoeutectic alloys of the Al-Si system were considered they must be fixed together by screws to be coupled to the solid-
in this work: Al-3 wt.% Si, Al-7 wt.% Si and Al-9 wt.% Si. The rel- ification device. Then, type K thermocouples must be allocated to
evance of both base metal and binary for the current industry is the holes on the ingot mold laterals in order to collect the thermal
emphasized, once lightweight Aluminum alloys have been replac- profiles during the process. Finally, the effectiveness of the cool-
ing other materials in many applications [26,27], especially the Al- ing system is checked to verify the presence of the water jet and

2
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

to guarantee that the refrigerant flow will not come into contact, The time derivative of the solid fraction may be computed as a
through any hole or gap, with any part of the ingot mold but the function of temperature through the source effect given by Eq. 2,
bottom. being estimated using a pseudo specific heat as in Eq. 7,
After completing these operational steps, the procedure for ob- ∂ fs
taining the ingot begins. Initially, the temperature of the muffle c p = cm − L (7)
∂T
furnace (which contains the crucible) is defined, which must be
where cm is the specific heat at mushy zone, which is given by a
equal to the pouring temperature TV . When the alloy is in the oven
simple mixing law in Eq. 8,
long enough that its temperature equals TV , it is removed to be
poured into the ingot mold. Simultaneously, the resistors of the cm = f s cs + ( 1 − f s )cl (8)
directional furnace are connected to keep the liquid material at
where cs and cl are the specific heats of solid and liquid zones,
the pouring temperature. This control is performed with the aid
respectively. Thermal conductivity and density are given by,
of thermocouples immersed in the liquid.
Thus, when the internal temperature is adequate, the resistors k = f s ks + ( 1 − f s )kl (9)
of the directional oven are turned off and the refrigerant flow is
activated, cooling the system. The cooling process must not be in- ρ = f s ρs + ( 1 − f s ) ρl (10)
terrupted until the temperature of the ingot is at room tempera-
ture, ending the procedure. where ks and kl are thermal conductivity of solid and liquid zones,
as well as ρs and ρl are density of solid and liquid zones, re-
3. Mathematical and Numerical Model spectively. The thermophysical properties were considered as func-
tion of temperature, which experimental data were collected on
The numerical discretization presented in this section can be Gandin’s paper [32]. The boundary conditions for the mathemati-
seen in Oliveira et al. [29] and this was used as forward model. cal approach are:
The values from grid spacing, time step, and others numerical pa- • The lateral walls of the mold are thermal insulated;
rameters were the same. However, two modifications are made in • Minimal heat transfer related to quiescent air contact on the
present paper: (1) thermophysical properties depends of the tem- top of the mold and;
perature and (2) a back-diffusion model is applied. The mathemat- • Heat transfer related to water flow contact in the lower bound-
ical and numerical model, including these modifications, are ex- ary, so that solidification occurs on vertical upward direction.
plained above.
To model the phenomenon of solidification, the transient heat 4. Interfacial Heat Transfer Coefficient Models
transfer equation for an arbitrary volume , bounded by a surface
 , as presented in Eq. 1, was considered, As explained in Section 1, several IHTC models have been pre-
  
∂T sented in the literature to understand the thermal contact resis-
ρ c p d + F · dA = q˙ d (1)
 ∂t  
tance of the metal/mold interface in metal casting. Among these
models, the following were considered (in these, Pi with i ∈ N∗
where temperature, material density and specific heat are de-
represents the parameters to be estimated in each heat transfer
scribed, respectively, as T ≡ T (x, t ), ρ ≡ ρ (x, t ) and c p ≡ c p (x, t ).
model):
F represents the heat flux through the boundary surface  , dA the
elemental area projected on the surface and q˙ the internal energy • Eq. 11 indicates a constant function model;
generated or consumed within the control volume due to the en- hi = P1 (11)
ergy sources. Only latent heat is considered, which is defined as in
Eq. 2, • Eq. 12 considers an IHTC model explicitly dependent on time t,
represented by a power function, in addition to two parameters
∂ fs
q˙ = ρ L (2) to be determined;
∂t
hi = P1 t P2 (12)
where L represents the latent heat coefficient and fs the local solid
fraction. This fraction is estimated by a back-diffusion model [31], • Eq. 13 shows a model based on three parameters to be deter-
given by Eq. 3, mined and the distance from the phase transformation inter-
   1k−β−1k0  face in the longitudinal direction to the ingot, Zint . This model
1 Tf − T 0 is also represented as a power function.
f s (T ) = 1− (3)
1 − β k0 T f − Tliq −Zint P3−1
hi = P1 + P2 (13)
with parameters β (Eq. 4), which also depends on γ (Eq. 5) and
4.1. IHTC Modeling by Thermal Resistances
 (Eq. 6), as shown below. In these equations, T represents tem-
perature, k0 the partition coefficient, T f the solvent melting tem-
The modeling to be performed will consider only the solid-solid
perature, Tliq the liquidus temperature on the nominal concentra-
contact between the metal/mold interface. So, in this particular
tion, D the solid-state diffusivity, t f the final solidification time, λ2
case, the IHTC can be represented as follows,
the secondary arm spacing and A a numerical constant with an es-
tablished value of 4 [31]. Here, considering the properties of the hi = hi f s (14)
studied alloys, β = 0.18.
where hi f is the portion of the heat extraction coefficient relative
2γ
s

β = (4) to the solid-solid contact.

2 + γ The justification for disregarding solid-liquid contact lies in the
experimental data considered in this research. As the superheat
A k 0
γ= (5) considered in the casting of the alloys was 2 K, there is no need
A k 0 + 1
to implement a solid-liquid contact region, given the low represen-
4Dt f tation that this extra modeling would contribute to the simulated
= (6)
λ22 thermal data of this work.

3
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

Lmold L
RC = + metal (17)
kmold kmetal
Considering that, on average, all asperities have statistical sim-
ilar geometric aspects, that is, they have the same peak, valley
and standard deviation, in order to reduce computational effort,
Lair ∼ Lmold ∼ Lmetal ∼ L, L being the average length of the rough-
ness existing in the metal-mold interface. Replacing Eqs. 16 and
17 in Eq. 15, it is possible to represent hi f as in Eq. 18,
s

1 
hi f s = kweig fc + kair (1 − fc ) (18)
L
kmetal kmold
where kweig = kmetal +kmold
.
At this point, it is possible to observe that Eq. 18 is composed of
parameters derived from a consistent physical model. From Eq. 18,
we can set Eq. 19 the model of the interfacial heat extraction coef-
Fig. 2. Schematic representation of the imperfect contact exchange interface in ficient in phase transformation problems.
metal solidification in permanent mold casting.
1 
hi = kweig P2 + kair (1 − P2 ) (19)
P1
The model presented above is simple but has physical mean-
ing, which adds parameters that can be easily understood and ex-
panded. Therefore, its physical representativeness is highlighted in
the following points:

1. The modeling of this comes from a thermal method known in

the study of heat transfer. That is, it does not come from arbi-
trary assumptions;
Fig. 3. Schematic representation of the thermal resistances at the metal-mold in- 2. Lists the thermal properties of all elements involved in the sys-
terface in upward solidification. tem interface: air, metal and mold;
3. It presents a range of inferable values for the parameters to be

Consider the metal-mold interface as shown in Fig. 2.

 −4 −2  in an upward directional device, these being: L ∈
determined,
10 , 10 m and fc ∈ [0, 1].
Asperities are observed both in the solidified metal and in the
mold, with different geometries, lengths and periodicity. However, The experimental data on thermal conductivity of air, mold and
it is possible to infer that there are only two types of situations in metal were taken from [33,34] and [32], respectively.
this type of interface: non-contact condition RNC (there is a void
filled with air in which metal and mold are not touch), or the con- 4.2. IHTC Modeling by Thermal Gradient Analogy
tact condition RC (where a roughness of the metal and the mold
touch without the presence of an air layer between them). In this model, we suppose that the dynamics of the heat extrac-
As these two types of contact are in parallel, since the existence tion coefficient is analogous to the behavior of the thermal gradi-
of one indicates the absence of the other, it is concluded that the hi ent at the heat extraction interface. Thus, a formulation that de-
is equal to the inverse sum of these resistances, as shown in Eq. 15, scribes the heat exchange interface could be compared to the for-
being fc a contact factor, where fc ∈ [0, 1], 0 being the absence of mulation of the interfacial coefficient of heat transfer. In a sim-
contact between the mold and the metal and 1 the perfect contact plified way, it is possible to extract the dynamics of hi from the
between the solids. solution of the Partial Differential Equation (PDE) of the Transient
fc 1 − fc One-dimensional Heat Diffusion, which can be seen in Eq. 20. This
hi f s = + (15) formulation, although simple, reasonably represents the spatial and
RC RNC
temporal behavior of the temperature in this study area, as it is a
For the model proposition, RC will be considered to refer to heat unidirectional cooling solidification. In this formulation, α repre-
transfer by conduction in the same way as resistance RNC . As the sents the thermal diffusivity of the material, T ≡ T (x, t ), t the time
size of the roughness is of the order of 10−3 or less, it is plausible and x the spatial variable of the domain.
to consider that the air trapped in this interstice does not present
∂T ∂ 2T
enough movement to promote heat transfer by convection. =α 2 (20)
It is worth pointing out that in the metal-mold contact there ∂t ∂x
are two different materials. Thus, this interfacial contact can be As boundary conditions and initial condition,
modeled by two resistances in series, one representing the contact
1. T (0, t ) = 0
properties of the mold material and one of the metal, as repre-
2. T (L f , t ) = 0
sented in Fig. 3.
3. T (x, 0 ) = f (x )
Therefore, considering the formulation of resistance to heat
transfer by conduction, Eqs. 16 and 17 are obtained, which explain where f (x ) is the function that represents the initial tempera-
the terms of the non-contact region and the contact region, respec- ture on the ingot and L f the final length of the domain. In this
tively. In these, L2 represents the length of the roughness and k2 case, f (x ) = Tp , where Tp is the pouring temperature of the molten
the thermal conductivity. metal.
Lair It is known that the conditions imposed in the solution of this
RNC = (16) problem are not consistent with the physical problem under study.
kair

4
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

However, these conditions are applied only to obtain a simplified number of flames is calculated as expressed in Eq. 25,
solution of the thermal distribution, and the physical relationship
Nmax − 1

between the conditions and the phenomenon itself is not an im- fn = Nmax − t + 0.5 (25)
Tint
mediate concern. When Eq. 20 is solved by the method of sep-
aration of variables, the temperature function resulting from this where t is the current number of iteration and Nmax and Tint repre-
solution can be seen in Eq. 21. sent the maximum number of flames and iterations, respectively.
∞
  A pseudo-code containing the programming procedure of this
nπ x
T (x, t ) = Asin exp(−α Bt ) (21) method based can be seen on Algorithm 1.
Lf
n=1
2 Algorithm 1 Pseudo-code of MFO algorithm [35]
L nπ x nπ
where A = 2
Lf 0 f (x )sin Lf dx and B = Lf . In these, n is a Update the number of flames fn by Eq. 25
counter such that n ∈ N. Taking the derivative to x, that is, obtain- Create M and OM
ing the thermal gradient ∂∂Tx , where A and B are constants: if iteration == 1 then
∞
  F = sort(M)
∂T nπ x
(x, t ) = Acos exp(−α Bt ) (22) OF = sort(OM)
∂x n=1
Lf else
F = sort(Mt−1 , Mt )
Supposing that the behavior of the interfacial heat extraction
OF = sort(Mt−1 , Mt )
coefficient is analogous to the behavior of the thermal gradient in
x = 0, that is, ∂∂Tx (0, t ) ∝ hi , it can be inferred that, by Eq. 23: end if
while (criterion) do
hi = P1 exp(−α P2 t ) (23) for each search agent do
Update mt+1
i
using Eq. 24 with respect to the correspond-
The model presented above comes from a mathematical solu-
ing moth
tion and has a physical correspondence with the problem, since it
indicates that the decay of the thermal extraction coefficient has end for
exponential behavior, as well as that of temperature. In contrast, end while
this model does not provide in-depth knowledge about the param-
eters, or P1 and P2 cannot be directly correlated with any physical This method was used to reduce the error between the exper-
quantity, assuming abstract physical interpretation. For numerical imental and simulated thermal profile based on the optimization
simulation, α values for Al-3 wt.% Si, Al-7 wt.% Si and Al-9 wt.% Si, of the parameters of each interface model used. The parameters
respectively, are 4.662 · 10−5 , 3.487 · 10−5 and 3.151 · 10−5 m2 s−1 . related to MFO and convergence were considered as follows:

5. Inverse Method Approach • Maximum number of iterations on each optimization: 200;

• Number of particles: 40;
In this paper, the inverse estimates will be performed by an op- • Stopping criterion: 15 iterations
 resulting
 on the same best pa-
timizer called Moth-Flame Optimization (MFO) [35], which was in- rameter vector, that is, mt+1 − mt  ≤ 10−10 ;
spired on the navigation method of moths in nature called trans- • Geometric parameter bMFO = 1.
verse orientation. Different from deterministic methods [36], which
The search interval for each parameter of each model was con-
produces good results but are strongly dependent of derivatives
sidered as wide as possible, in order not to bias the optimization.
and starting point choice, metaheuristic methods are able to per-
Table 1 shows the range for each parameter analysed.
form a global search and switch it to a local one based on inter-
In this article, the objective function (error) to be minimized,
nal mechanisms of its own formulation. Thus, exploration and ex-
(S ), corresponds to the sum of the standard deviations between
ploitation is carried out in a balanced way in order to guarantee
the experimental thermal profile (Y ) and the simulated ones (W )
satisfactory performance in the optimization of the most diverse
of each thermocouple.
physical problems. In particular, as shown by Stieven et al. [25],
MFO has a very satisfactory performance in phase transformation
6. Relative Sensitivity Coefficient
applications in permanent mold casting, with statistical precision
and accuracy. A explanation of the method is set out below. Com-
The sensitivity coefficient Ji j , defined in Eq. 26 [37], is the first
plete details are shown in Mirjalili’s original paper [35].
derivative of the simulated temperature vector, indicated by W, in
Consider M = mti and F = fit the moth and flame positions, re-
relation to a parameter, represented by θ , at a given time ti . This
spectively, in which i and t are, respectively, the particle and it-
indicates the sensitivity of the estimated state variable in relation
eration indexes. Moths and flames are both solutions, but the
to the variations of the parameter θ . Consider that J = [Ji j ], where
moths are actual search agents that move around the search space,
i is the index for the state variables and j the index for the value
whereas flames are the best position of moths that obtains so far.
of the parameters of the contact resistance models.
So, F is a sorted vector obtained by the best particles in M during
the entire simulation. OM and OF represents the fitness values of ∂ Wi
Ji j = (26)
the moth and flame particles. ∂θ j
Flames, in this method, are considered as flags that are dropped
by moths when searching the search space. In the original paper In order to obtain a dimensional value to observe the sensitiv-
a logarithmic spiral was chosen, which formulation can be seen ity behavior of the state variable by the parameter under analysis,
in Eq. 24, where bMFO is a constant for defining the shape of the the relative sensitivity coefficient Xθ j is calculated by taking the
logarithmic spiral and tint is a random number in [−1, 1]. derivative of the state variable in relation to the parameter, mul-
tiplying the derivative by the value of the state variable. This is a
mt+1
i
=| fit − mti |exp(bMFOtint )cos(2π tint ) + fit (24) dimensional measure that represents the degree of disturbance of
For exploration/exploitation switch the number of flames re- the state variable as seen in the input parameters. Xθ j has the size
duces in order to favor exploitation after some iterations. So, the of the state variable and has relevance directly proportional to its

5
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

Table 1
Search range of the parameters presented on the IHTC models in study.

IHTC Models Parameter Search Range

P1 P2 P3

(1) hi = P1 [1000, 10000] - -

(2) hi = P1 t P2 [3000, 9000] [-0.4, -0.00005] -
−1
(3) hi = P1 +P2 −Zint P3  [1000, 9000] [0.001, 2] [0.001, 0.7]
(4) hi = P1 −1 kpond P2 + kar (1 − P2 ) [0,000001, 0.01] [0.001, 0.999] -
(5) hi = P1 exp (−α P2 t ) [1000, 15000] [0.001, 1000] -

absolute value, as shown in Eq. 27. Table 2

MFO optimized values for the parameters presented on Section 4, for Al-3wt.%Si
∂W alloy.
Xθ j = θ j (27)
∂θ j IHTC Models Parameter Values by MFO Error

Xθ j can be approximated by a central difference scheme as in P1 P2 P3

Eq. 28 [37]. (1) hi = P1 6715.45 - - 25.95
    (2) hi = P1 t P2 6714.34 -0.00005 - 25.95
Wi θ1 , θ2 , ..., θ j + νθ j , ..., θNpar − Wi θ1 , θ2 , ..., θ j − νθ j , ..., θNpar (3) hi = P1 +P2 −Zint P3−1
6046.35 0.001 0.371 25.90
Xθ j ≈ 
2ν (4) hi = P1 −1 kpond P2 + kar (1 − P2 ) 0.002297 0.815 - 28.10
(28) (5) hi = P1 exp (−α P2 t ) 8997.42 99.258 - 19.03

The sensitivity study of the parameters to be estimated is im-

portant because, from this, it is known the influence of each pa-
rameter in the state variable and, thus, its relevance in the opti-
mization process. If the parameter is insensitive, its estimate be-
comes unfeasible since it does not present a disturbance in the
state variable. In these cases, the parameter is considered a con-
stant, reducing computational cost. If the parameter is sensitive,
that is, it presents large Xθ j , the parameter can be estimated
because its disturbance significantly modifies the state variable,
which makes it feasible to estimate. This analysis is very relevant
to avoid over-parameterized models.
For Xθ j calculation, given the parameters presented in Eqs. 11,
13, 19 and 23, was considered ν = 1 · 10−3 .

7. Results and Discussion

In this Section, the MFO optimization of the parameters of the

IHTC models presented in Section 4 will be shown, in addition to
the relative sensitivity coefficient. The models analysed are shown
again, numbered from 1 to 5 to identification of each model. So,
from now on, the models will be presented by the numbers shown Fig. 4. Experimental thermal profile of Al-3wt.%Si and simulated curves based on
model 5.
below:
(1 ) hi = P1
(2 ) hi = P1 t P2 presents the worst result in terms of error assessment. Model 5
−Z P −1 presents P1 that decreases and P2 that grows from the increase in
(3 ) hi = P1 + P2 int 3  solute in the alloy. This model presented the best results in terms
(4 ) hi = P11 kweig P2 + kair (1 − P2 )
of error, surpassing the estimates of the other models under study.
(5 ) hi = P1 exp(−α P2 t )
The simulated thermal profile resulted by the optimized param-
eter vector of the model 5, which has the low error of all studied
7.1. Parameter Estimate Analysis
models, can be seen in Fig. 4, 5 and 6. In these, the values of 4
mm, 8 mm and 12 mm represent the distance of the thermocou-
The search intervals shown in Section 5, referring to the param-
ple in relation to the metal/mold interface.
eters of the IHTC models, were used by the MFO optimizer to esti-
mate the best values to represent the experimental data from the
work of Peres [30]. The optimized parameter values can be seen 7.2. Relative Sensitivity Analysis
in Tabs. 2, 3 and 4, and also the error related to each optimized
parameter vector. The Figs. 7, 8, 9, 10 and 11 show the curves of the reduced sen-
Models 1, 2 and 3 present very similar error values for the three sitivity coefficient of the parameters present in the IHTC models
alloys under analysis, as well as values for parameter P1 , which under study. The Al-3wt.% Si alloy was chosen to represent this
has a decreasing value with the increase of the mass percentage of coefficient. It is worth pointing out that the behavior of these co-
solute in the alloy. P2 of model 2 remain constant in values very efficients, for each parameter, in the other studied alloys, is quite
close to the upper limit of search of the parameters. similar. Fig. 7 shows that the parameter P1 of model 1 is quite sen-
On model 4, it is seen that P1 remains in the range of 2 to 4 mm sitive. As this is the only parameter in this model, its sensitivity
and P2 in the range of 70 to 80% of solid-solid contact. This model was expected.

6
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

Table 3
MFO optimized values for the parameters presented on Section 4, for Al-7wt.%Si alloy.

IHTC Models Parameter Values by MFO Error

P1 P2 P3

(1) hi = P1 4029.04 - - 31.51

(2) hi = P1 t P2 4040.64 -0.00005 - 31.51
−1
(3) hi = P1 +P2 −Zint P3  4202.46 0.001 0.01116 31.15
(4) hi = P1 −1 kpond P2 + kar (1 − P2 ) 0.002904 0.729 - 35.05
(5) hi = P1 exp (−α P2 t ) 5372.74 126 - 29.59

Table 4
MFO optimized values for the parameters presented on Section 4, for Al-9wt.%Si alloy.

IHTC Models Parameter Values by MFO Error

P1 P2 P3

(1) hi = P1 3337.84 - - 27.69

(2) hi = P1 t P2 3337.99 -0.00005 - 27.69
−Zint P3−1
(3) hi = P1 +P2  3331.71 0.1927 0.02166 27.67
(4) hi = P1 −1 kpond P2 + kar (1 − P2 ) 0.004013 0.7628 - 30.14
(5) hi = P1 exp (−α P2 t ) 4257.72 130.227 - 24.32

Fig. 5. Experimental thermal profile of Al-7wt.%Si and simulated curves based on Fig. 7. Relative sensitivity coefficient of the parameters on model 1, based on Al-
model 5 3wt.%Si alloy.

Fig. 6. Experimental thermal profile of Al-9wt.%Si and simulated curves based on Fig. 8. Relative sensitivity coefficient of the parameters on model 2, based on Al-
model 5. 3wt.%Si.

7
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

Table 5
Overall analysis of the relative sensitivity coefficient of each parameter, in each
IHTC model.

IHTC Models P1 P2 P3

(1) hi = P1 Sensitive - -
(2) hi = P1 t P2 Sensitive Insensitive -
−1
(3) hi = P1 +P2 −Zint P3  Sensitive Insensitive Insensitive
(4) hi = P1 −1 kpond P2 + kar (1 − P2 ) Sensitive Sensitive -
(5) hi = P1 exp (−α P2 t ) Sensitive Sensitive -

Fig. 8 shows Xθ j relative to model 2. It is observed that P1 is

quite sensitive, whose behavior is similar to P1 of model 1. How-
ever, P2 is insensitive, having a relative sensitivity coefficient close
to zero. This indicates that the optimization of this parameter is
negligible, since it has no influence on the simulated profiles.
In Fig. 9 it is shown Xθ j obtained from model 3. P1 of this
model, as well as those of models 1 and 2, is sensitive, showing
behavior similar to these. Like model 2, it has insensitive parame-
Fig. 9. Relative sensitivity coefficient of the parameters on model 3, based on Al- ters, which are P2 and P3 . It can be seen that changing these pa-
3wt.%Si alloy.
rameters does not change the thermal output of the system.
Unlike models 2 and 3, the relative sensitivity coefficient of the
model 4 parameters, shown in Fig. 10, implies that P1 and P2 are
sensitive. These parameters present Xθ j in the same order of mag-
nitude as the parameters P1 of models 1, 2 and 3.
According to Fig. 11, which contains the sensitivity of the pa-
rameters of model 5, there is an indication that the two param-
eters are sensitive, with P1 being more than P2 . The behavior of
these parameters is different from the other models, showing great
variation along time.
Finally, in Tab. 5 the results of this section are briefly presented.
In this, it is possible to observe which parameter of each model is
sensitive, or insensitive.
From the results of the inverse estimation of parameters and
relative sensitivity coefficient, it is possible to make some com-
ments about the models under analysis. Model 1 presents a good
numerical adjustment to the experimental data and, as it presents
only one parameter, it is clearly sensitive. However, this model
does not have strong physical representativeness since the inter-
facial contact resistance does not remain constant throughout the
casting process, with a large thermal difference between the be-
Fig. 10. Relative sensitivity coefficient of the parameters on model 4, based on Al- ginning and the end of the process.
3wt.%Si alloy.
Model 2 has great relevance in the literature, being important
in several works that deal with the subject. However, as in model
3, based on the considerations of this work, they present insensi-
tive parameters in all the alloys worked in this study. It is inferred
that models 2 and 3 collapse into a constant function as in model
1, after optimization of their parameters. In view of this, models
2 and 3 could be replaced by model 1 without much loss in nu-
merical solution of physical representation. Based on some works
in the literature that used model 2 to represent the IHTC, such as
[29] and [38], it is observed that this model behaved in a sensi-
tive way in these contributions. However, in the works presented,
the thermophysical properties were considered constant and there
was the application of the Scheil rule to represent the solid frac-
tion in mathematical modeling. These considerations may modify
the influence of P2 in the numerical representation of the IHTC.
Model 4 comes from a physically coherent and well known en-
gineering modeling. This depends on thermophysical properties of
both the air within the roughness of the metal/mold interface and
the mold and metal. This model has two parameters to be deter-
mined, which are highly sensitive. The optimization of these pa-
rameters, considering the three alloys under analysis, showed that
Fig. 11. Relative sensitivity coefficient of the parameters on model 5, based on Al- P1 remained in the range of 2 to 4 mm, a reasonable length when
3wt.%Si alloy.
considering the influence of contraction in these alloys. In addi-

8
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

tion, P2 is in the range of 0.7 to 0.8, indicating that the contraction interface (hi = P1 exp(−α P2 t ) ). The latter is presented by an expo-
interferes about 20 to 30 % of the solid-solid contact, according nential temporal function and depends on two parameters to be
to the physical representation of the model deduction. However, determined and on the thermal diffusivity of the alloy under anal-
this model has some problems, with some difficulty in adjusting ysis.
to the experimental thermal data. This is reflected in the error be- In this study, three hypoeutectic alloys of the Al-Si system were
tween the simulated and experimental thermal profiles, presenting considered: Al-3wt.% Si, Al-7wt.% Si and Al-9wt.% Si. To obtain the
the highest values in comparison to the other methods under anal- simulated thermal profiles, the energy conservation equation was
ysis. As this method has strong physical representativeness in the discretized by the finite volume method. The thermophysical prop-
phenomenon under study, it is necessary to investigate the rela- erties of the alloys were considered constant. The solid fraction
tionship between P1 and P2 with other physical properties of the was mathematically represented using the back diffusion of the re-
molten metal such as casting shrinkage and wettability, in order to jected solute to the liquid during solidification. From inverse esti-
improve the proposed model. mates by the Moth-Flame optimizer (MFO) and the survey of the
Model 5 comes from the assumption that the IHTC has a func- relative sensitivity coefficient, it was observed that:
tion analogous to the thermal gradient in the cooling interface.
Based on that, an exponential model was obtained, in which it
• Model 1 presents low physical representativeness in the phe-
considers two parameters to be determined. For the alloys under nomenon of permanent mold casting, however it shows good
analysis, that these two parameters are sensitive, that is, a small numerical results. As it has only one parameter, it is clearly sen-
variation in these causes great thermal changes, being P1 more sitive, reducing its optimum value as the percentage of silicon
sensitive than P2 . By the vectors optimized by the MFO method, increases in the alloys;
it was found that this method has the lowest numerical / experi-
• Model 2 has a sensitive parameter, P1 , as sensitive as the pa-
mental error value, standing out from the other models. Thus, it is rameter P1 of model 1, however P2 is insensitive. Thus, it has no
suggested the use of this model for casting applications in perma- relevance in the numerical/experimental adjustment of thermal
nent mold in alloys of the Al-Si system. profiles. The optimization of the parameters in the three alloys
Regarding the relative sensitivity coefficient of the parameters of the Al-Si system shows that P2 always tends to 0 and that P1
of model 5, it is observed that they present a more complex be- decays as the percentage of solute increases;
havior in relation to other models. Models 1, 2, 3 and 4 show a
• Model 3 presents P1 sensitive, with the same sensitivity dy-
decay or growth to an asymptotic region of Xθ j after 40 seconds of namics and optimum value present in models 1 and 2 for the
same parameter. In other words, P1 optimum value decreases
casting. Thus, the influence of the interface parameters after this
with the increase in alloy solute. However, P2 and P3 are in-
period is quite small compared to the first stages of solidification.
sensitive. In this casting process, considering these alloys under
On the other hand, from the data collected for model 5, parameters
study and the mathematical method exposed, models 2 and 3
are observed up to the 20 seconds of the process, alternating their
collapse in model 1, since they present only one sensitive pa-
values along time. After 40 seconds they still have an influence on
rameter, P1 ;
solidification. This indicates that the relevance of this model for so-
• Model 4 presents two highly sensitive parameters, which are
lidification as a whole is greater than the other models presented,
physically representative. From the optimized results, it can be
since it has the capacity to influence the heat extraction process
seen that the two parameters remain in the same range for the
during most of the process.
three alloys, which is physically consistent by the assumptions
It is important to mention that only three alloys of the Al-Si
regarding the physical development of this method. However, it
system have been studied and in only one casting process. The ex-
was observed that this exhibits some difficulty in adjusting to
pansion of the use of this model to represent the IHTC in other
the experimental data. It is possible that this model is strongly
metallic systems and in other casting processes in order to obtain
influenced by the thermal conductivity curves of the three ma-
a response as satisfactory as obtained in this work is not imme-
terials present in its formulation. Future studies are necessary
diate. Further studies on the validity and awareness of this model
that relate the two parameters present in this model with prop-
should be presented for other related applications. However, given
erties related to their intrinsic physical representations, such as
the numerical, mathematical and physical coherence of this model,
wettability and casting shrinkage;
in the absence of another model more suitable for other applica-
• Model 5 presents two sensitive and mathematically consistent
tions and metallic systems, model 5 is strongly recommended.
parameters. From the optimization of its parameters, it was ob-
served that this model was able to reduce the numerical / ex-
8. Conclusions perimental error of the thermal profiles more satisfactorily. It
can be inferred that this is due to the fact that, as shown by the
This work analysed three thermal interface models present in relative sensitivity coefficient, these parameters are relatively
the literature: model 1, a constant
 function
 (hi = P1 ), model 2, sensitive for a longer period of time than the parameters of the
a temporal power function hi = P1 t P2 and model 3, a power other models. Thus, the adjustment of the simulated curve can
function considering the distance between the mold and the be performed over a time range greater than 40 seconds, during
−Z P −1 which time the other models stabilize.
solid-liquid transition interface (hi = P1 + P2 int 3 ). In addition,
two models were proposed: one based on the thermal resistance
method and the other based on the hypothesis that the IHTC Declaration of Competing Interest
model would have dynamics analogous to the function of the ther-
mal gradient at the metal / mold interface. These two method- The authors declare that they have no known competing finan-
ologies gave rise to two distinct models: model 4, proposed by cial interests or personal relationships that could have appeared to
the thermal resistance method has a linear formulation, depend- influence the work reported in this paper.
ing on two parameters to be determined and on the conductiv-
ity of the
 metal, the mold and the air  between the mold and the CRediT authorship contribution statement
metal hi = P1−1 kweig P2 + kair (1 − P2 ) , and model 5, proposed by
the hypothesis that the behavior of the IHTC is correlated to the Gianfranco de M. Stieven: Conceptualization, Investigation,
analytical formulation of the thermal gradient at the metal / mold Methodology, Software, Validation, Formal analysis, Writing - orig-

9
G.d.M. Stieven, D.d.R. Soares, E.P. Oliveira et al. International Journal of Heat and Mass Transfer 166 (2021) 120765

inal draft, Writing - review & editing. Daniele dos R. Soares: In- [17] H. Ge, F. Ren, J. Li, Q. Hu, M. Xia, J. Li, Modelling of ingot size effects on
vestigation, Data curation, Visualization, Writing - original draft. macrosegregation in steel castings, Journal of Materials Processing Technology
252 (2018) 362–369.
Edilma P. Oliveira: Conceptualization, Formal analysis, Supervi- [18] N. Thieme, M. Keil, D. Meier, P. Bönisch, K. Dadzis, O. Pätzold, M. Stelter,
sion, Funding acquisition, Writing - original draft. Erb F. Lins: L. Büttner, J. Czarske, Directional solidification of gallium under time-depen-
Investigation, Methodology, Software, Validation, Formal analysis, dent magnetic fields with in situ measurements of the melt flow and the
solid-liquid interface, Journal of Crystal Growth 522 (2019) 221–229.
Writing - original draft, Writing - review & editing. [19] C. Silva-Santos, P. Goulart, F. Bertelli, A. Garcia, N. Cheung, An artificial immune
system algorithm applied to the solution of an inverse problem in unsteady
Acknowledgments inward solidification, Advances in Engineering Software 121 (2018) 178–187.
[20] P. Vishweshwara, N. Gnanasekaran, M. Arun, Estimation of interfacial heat
transfer coefficient for horizontal directional solidification of sn-5 wt% pb al-
The authors would like to thank the CAPES and FAPESPA for loy using genetic algorithm as inverse method, in: Soft Computing for Problem
financial support by research grant FAPESPA 88881.159113/2017-1. Solving, Springer, 2019, pp. 447–459.
[21] Y. Natsume, Y. Oka, J. Ogawa, M. Ohno, Estimation of time-dependent heat
Also, the authors are thankful to CNPq and PROPESP/UFPA.
transfer coefficient in unidirectional casting using a numerical model coupled
with solidification analysis and data assimilation, International Journal of Heat
References and Mass Transfer 150 (2020) 119222.
[22] V. de Barcellos, C. Ferreira, C.d. Santos, J. Spim, Analysis of metal mould heat
[1] M. Wu, A. Ludwig, A. Kharicha, A four phase model for the macrosegregation transfer coefficients during continuous casting of steel, Ironmaking & Steel-
and shrinkage cavity during solidification of steel ingot, Applied Mathematical making 37 (1) (2010) 47–56.
Modelling 41 (2017) 102–120. [23] T.P.D. dos Anjos, P.V.d.C.L. Pimenta, F. Marcondes, Analysis of the heat transfer
[2] Z. Gao, W. Jie, Y. Liu, Y. Zheng, H. Luo, A model for coupling prediction of coefficients of the whole process of continuous casting of carbon steel, Journal
inverse segregation and porosity for up-vertical unidirectional solidification of of the Brazilian Society of Mechanical Sciences and Engineering 40 (2) (2018)
al–cu alloys, Journal of Alloys and Compounds 797 (2019) 514–522. 107.
[3] K. Wang, P. Fu, L. Peng, Y. Wang, W. Ding, A simplified hot-tearing criterion for [24] P. Vishweshwara, N. Gnanasekaran, M. Arun, Inverse estimation of interfacial
shape castings based on temperature-field simulation, Metallurgical and Mate- heat transfer coefficient during the solidification of sn-5wt% pb alloy using
rials Transactions A 50 (11) (2019) 5271–5280. evolutionary algorithm, in: Advances in Materials and Metallurgy, Springer,
[4] N. Yukawa, Y. Nakashima, T. Ishiguro, E. Abe, T. Ishikawa, T. Choda, Modeling 2019, pp. 227–237.
of heat transfer coefficient of oxide scale in hot forging, Procedia Engineering [25] G.M. Stieven, E.P. Oliveira, E.F. Lins, Performance analysis of metaheuristic op-
81 (2014) 492–497. timization algorithms in estimating the interfacial heat transfer coefficient on
[5] M. Xu, R. Ling, Z. Zhang, J. Xie, Study on interfacial heat transfer behavior of directional solidification, arXiv preprint arXiv:2007.15583 (2020).
ta15 titanium alloy and die materials, International Journal of Heat and Mass [26] S.C. Johnson, C.D. Clark, J.S. Alvarez, Development and analysis of al7075 alloy
Transfer 108 (2017) 1573–1578. materials using press and sinter processing, in: Light Metals 2020, Springer,
[6] F. Wang, X. Wang, Z. Han, On the interfacial heat transfer and pressure trans- 2020, pp. 233–240.
mission in squeeze casting: A case study of the sensitivity to materials, Inter- [27] A. Kordijazi, S.K. Behera, O. Akbarzadeh, M. Povolo, P. Rohatgi, A statistical
national Journal of Heat and Mass Transfer 133 (2019) 52–61. analysis to study the effect of silicon content, surface roughness, droplet size
[7] Z. Li, J. Wang, H. Xing, K. Jin, H. Huang, Determining dendrite arm spacing in and elapsed time on wettability of hypoeutectic cast aluminum–silicon alloys,
directional solidification using a fast fourier transform method, Computational in: Light Metals 2020, Springer, 2020, pp. 185–193.
Materials Science 173 (2020) 109463. [28] G. Sigworth, Solidification and Castability of Foundry Alloys, Aluminum Science
[8] R. Lenart, M. Eshraghi, Modeling columnar to equiaxed transition in directional and Technology, ASM International, 2018.
solidification of inconel 718 alloy, Computational Materials Science 172 (2020) [29] E.P. Oliveira, G. de M. Stieven, E.F. Lins, J.R. Vaz, An inverse approach for the in-
109374. terfacial heat transfer parameters in alloys solidification, Applied Thermal En-
[9] M. Dong, T. Liu, X. Guo, Y. Liu, S. Dong, Q. Wang, Enhancement of mechanical gineering 155 (2019) 365–372.
properties of tb0. 27dy0. 73fe1. 95 alloy by directional solidification in high [30] M.D. Peres, Desenvolvimento da macroestrutura e da microestrutura na solid-
magnetic field, Materials Science and Engineering: A (2020) 139377. ificaçã unidirecional transitoria de ligas Al-Si, Federal University of Campinas,
[10] H. Zheng, Q. Xu, R. Chen, G. Qin, X. Li, Y. Su, J. Guo, H. Fu, Microstructure 2005 Doctoral thesis.
evolution and mechanical property of directionally solidified cocrfemnni high [31] V. Voller, On a general back-diffusion parameter, Journal of Crystal Growth 226
entropy alloy, Intermetallics 119 (2020) 106723. (4) (2001) 562–568.
[11] S. Zhai, J. Yuan, J. Liu, Y. Wang, Microstructure and mechanical properties of [32] C.-A. Gandin, From constrained to unconstrained growth during directional so-
directionally solidified al2o3/yag binary eutectic ceramic prepared with in- lidification, Acta Materialia 48 (10) (20 0 0) 2483–2501.
duction heating zone melting, Materials Chemistry and Physics 242 (2020) [33] K. Stephan, A. Laesecke, The thermal conductivity of fluid air, Journal of Phys-
122503. ical and Chemical Reference Data 14 (1) (1985) 227–234.
[12] B. Mikić, Thermal contact conductance; theoretical considerations, Interna- [34] T. Ashworth, D.R. Smith, Thermal Conductivity 18, Springer Science & Business
tional Journal of Heat and Mass Transfer 17 (2) (1974) 205–214. Media, 2012.
[13] J. Bohacek, A. Kharicha, A. Ludwig, M. Wu, E. Karimi-Sibaki, Heat transfer coef- [35] S. Mirjalili, Moth-flame optimization algorithm: A novel nature-inspired
ficient at cast-mold interface during centrifugal casting: Calculation of air gap, heuristic paradigm, Knowledge-Based Systems 89 (2015) 228–249.
Metallurgical and Materials Transactions B 49 (3) (2018) 1421–1433. [36] A. Ranjbar, M. Ezzati, M. Famouri, Optimization of experimental design for an
[14] R. Dou, T. Ge, X. Liu, Z. Wen, Effects of contact pressure, interface temperature, inverse estimation of the metal-mold heat transfer coefficient in the solidifica-
and surface roughness on thermal contact conductance between stainless steel tion of sn-10% pb, Journal of Materials Processing Technology 209 (15) (2009)
surfaces under atmosphere condition, International Journal of Heat and Mass 5611–5617.
Transfer 94 (2016) 156–163. [37] M.N. Özisik, H.R. Orlande, Inverse Heat Transfer: fundamentals and applica-
[15] K. Zhao, D. Ren, B. Wang, Y. Chang, Investigation of the interfacial heat transfer tions, Taylor & Francis, 20 0 0.
coefficient of sheet aluminum alloy 5083 in warm stamping process, Interna- [38] M. Peres, C. Siqueira, A. Garcia, Macrostructural and microstructural develop-
tional Journal of Heat and Mass Transfer 132 (2019) 293–300. ment in al-si alloys directionally solidified under unsteady-state conditions,
[16] A. Plotkowski, M. Krane, The discrete nature of grain attachment models Journal of Alloys and Compounds 381 (1-2) (2004) 168–181.
in simulations of equiaxed solidification, Applied Mathematical Modelling 47
(2017) 31–44.

10