Applied Thermal Engineering 117 (2017) 263274

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Research Paper

Numerical simulation of fluid flow and heat transfer in an industrial

continuous furnace
S. Defaee Rad a,, A. Ashrafizadeh a, M. Nickaeen b
a
Faculty of Mechanical Engineering, K. N. Toosi University of Technology, Tehran, Iran
b
Richard D. Berlin Center for Cell Analysis and Modeling, University of Connecticut Health Center, Farmington, CT 06030, USA

h i g h l i g h t s

A 3D, transient numerical study of an industrial continuous furnace is proposed.

Geometry of furnace load is simplified by a thermally equivalent model.

Load movement is avoided by implementing unsteady boundary conditions.

Fairly accurate results at a reasonably low computational cost are obtained.

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

Article history: We present a three-dimensional computational study of the transient heat transfer and turbulent fluid
Received 30 August 2016 flow inside an arc-welding electrode continuous furnace. The model is implemented in FLUENT, a finite
Revised 18 January 2017 volume commercial code. Large difference in the length scales of the electrodes and the furnace, move-
Accepted 8 February 2017
ment of the electrodes, and existence of various modes of heat transfer are the major factors influencing
Available online 10 February 2017
the accuracy and efficiency of the simulation. Two modeling strategies are used to overcome these diffi-
culties. First, the electrode geometry and material composition are simplified using a thermally equiva-
Keywords:
lent model. Second, the electrode movement inside the furnace is avoided by implementing time-
Welding electrode
Continuous furnace
dependent boundary conditions applied to a fixed domain. A fairly close agreement is obtained in com-
3D numerical simulation paring the loads temperature history against the experimental data with an absolute relative difference
Loads temperature history below 2.7%. The space between the electrode trays in the furnace is very limited. The analysis shows that
Conjugate heat transfer the input air does not circulate effectively between the trays. This lowers the thermal efficiency of the
furnace and leads to uneven treatment of the electrodes. The present study provides guidelines to
improve future furnace designs.
2017 Elsevier Ltd. All rights reserved.

1. Introduction
Arc welding electrodes play an important role in the construc-
tion of mechanical structures in various industries. Before the elec-
trode is processed into its final state, it is usually heated in a
furnace. The electrode is moved along the length of the furnace
and, depending on the application, is thermally treated by various
heat transfer mechanisms. This heat treatment is necessary to
assure the quality of the welding and to ensure that all mechanical,
metallurgical, environmental and economical measures are satis-
fied. The present paper aims at the performance evaluation of an
industrial convective furnace, used to cure arc-welding electrodes.
Corresponding author.
E-mail addresses: saeeddefaee@gmail.com (S. Defaee Rad), ashrafizadeh@kntu.
ac.ir (A. Ashrafizadeh), nickaeen@uchc.edu (M. Nickaeen).
Five fans are employed to direct hot air, processed in a shell and
tube heat exchanger, into the furnace.
Similar to many other applications of continuous furnaces
[13], the loads temperature history has considerable importance
in this case as well. However, because of the transient, multidi-
mensional and turbulent flow conditions as well as existence of
simultaneously different modes of heat transfer, it is hard to carry
out an accurate, quick and inexpensive analysis. It has been shown
that the preceding goals are achievable if Computational Fluid
Dynamics (CFD) is employed [4,5]. Computational Fluid Dynamics
provides an efficient tool to obtain the transient flow and temper-
ature fields, to inspect the effects of thermo-physical and geomet-
rical parameters on the curing process and to evaluate the success
of various optimization methods. Furthermore, commercial CFD
packages, such as FLUENT, CFX, STAR-CD or PHOENICS have
increased the computational competency significantly and have

http://dx.doi.org/10.1016/j.applthermaleng.2017.02.031
1359-4311/ 2017 Elsevier Ltd. All rights reserved.
264 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
Nomenclature
c1e, c2e, cl constants in the ke model
cp specific heat capacity at constant pressure
h convection heat transfer coefficient
j thermal conductivity
k turbulent kinetic energy
n outward unit normal vector
p pressure
t time
T Temperature
ui velocity vector component (i = 1, 2, 3)
v velocity magnitude
xi Cartesian coordinate component (i = 1, 2, 3)
e dissipation rate of turbulent kinetic energy
q density
emerged as the dominant tools to perform quick numerical analy-
sis with reasonable accuracy.
A detailed literature study shows that CFD tools have not yet
been employed to analyze the transport phenomena inside an
arc-welding electrode continuous furnace. However, reviewing
similar applications of CFD for the study of continuous furnaces
in steel [6], food [7] and ceramic [8] industries provides useful
information for the analysis of electrode curing process. Almost
all of the mentioned applications are followed by a standard
methodology. First, the physical processes are described as mathe-
matical models. Second, the governing equations are discretized
and solved to obtain a numerical solution of the problem. Sensitiv-
ity analysis may then be conducted to evaluate the impacts of
design and performance parameters on the simulation results.
Here, use of Computational Fluid Dynamics with focus on commer-
cial software in the numerical simulation of transport phenomena
in steel reheat furnaces is briefly reviewed.
Prieler et al. [9] performed CFD modeling of an industrial walk-
ing hearth furnace to predict transient heating characteristics of
the billets. Emadi et al. [10] obtained both convective and radiative
heat fluxes using zone method and investigated the effects of
design parameters on billet temperature behavior. Han and Chang
[11] conducted a 3D unsteady numerical simulation of a reheat
furnace with the commercial software FLUENT and analyzed the
optimal slab residence time. Han et al. [12] carried out a periodi-
cally transient numerical simulation of heat transfer to slabs in a
reheat furnace by using FLUENT. Due to the default software limi-
tations a User-Defined Function (UDF) was developed to process
the slab movement. Morgado et al. [13] followed a numerical strat-
egy similar to that of Han [12] and compared two different thermo-
chemical composition models for the slab. Dubey and Srinivasan
[14] applied an unsteady three dimensional numerical model to
analyze temperature field in the billet. They discretized the gov-
erning equations by finite volume method and solved the equa-
tions using an in-house MATLAB code. Kim et al. [15] presented a
comprehensive FLUENT-based simulation to understand the turbu-
lent flow and radiative heat transfer in a walking beam furnace.
One of the significant implications of their work was the simplifi-
cations made in the modeling of the furnace geometry to avoid
the unaffordable number of computational cells. Furthermore,
other researchers conducted similar CFD studies using alternative
software suits such as STAR-CD [16], Phoenics [17], FURMO [18]
and CONCERT [19].
The current paper addresses a numerical approach, performed
with FLUENT version 6.3 [20], to examine the non-isothermal flow
phenomena inside an arc-welding electrode continuous furnace.
Two important modeling strategies are employed here that are
l viscosity
lt turbulent viscosity coefficient
rk, re turbulent Prandtl number for k and e
u scalar
Subscripts
i,j coordinate directions
Superscripts
Reynolds-averaged value
Reynolds fluctuation value
crucial to the success of the numerical approach and considerably
reduce computational requirements of the simulation. To the best
knowledge of the authors, these simplifying strategies have not
been used by others for modeling arc-welding electrode continu-
ous furnaces. First, we consider time-dependent boundary condi-
tions to avoid the vastly expensive moving mesh methods
required for modeling electrode movement inside the furnace.
Instead of moving electrodes forward, we solve inside a fixed
geometry (fixed segment of the furnace) and model the hot air
inflow through time-dependent boundary conditions that move
backward with the same velocity as the electrodes. Since FLUENT
does not support moving boundary conditions with its default
functions, a User Defined Function [21] is developed and linked
to the main solver. Second, we replace the electrode geometry
(cross-section) and material composition with a computationally
inexpensive but thermally equivalent model. The existence of
diverse length scales between the electrodes and the furnace ren-
ders the good quality grid generation and the efficient boundary
condition implementation very challenging. Therefore, we propose
several simplified models for the electrodes and systematically
study them under equivalent furnace operating conditions.
Accuracy of the numerical results is verified by refining the
computational grid to ensure that a grid independent solution is
obtained, and by changing the time step used for the temporal dis-
cretization. In addition, the electrodes temperature history is com-
pared to the experimental data obtained from a k-type
thermocouple measuring the temperature in a real furnace operat-
ing condition. A maximum relative temperature difference below
2.7% is obtained which verifies the correctness of the simulation
results and the assumptions used in the mathematical model.
The rest of the paper is organized as follows: The furnace and
electrode configurations are described in Section 2. The mathemat-
ical and numerical models as well as the computational details are
presented in Section 3. The results are discussed in Section 4 and,
finally, Section 5 summarizes important conclusions.
2. Furnace configuration and arrangement of electrodes
The continuous electrode curing furnace considered here con-
sists of preheating, heating and cooling sections that are 10, 9.8
and 4.7 m long, respectively. The electrodes pass through the
entire furnace in 146 min with a velocity of 0.0028 m/s. Therefore,
the electrodes are treated for 59.5, 58.5 and 28 min in each of the
aforementioned furnace zones, respectively. Air is forced into the
furnace through a number of nozzles, exchanges heat with elec-
trodes and then leaves from the bottom. The first two sections
 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274 265

are furnished with almost similar arrangements of nozzles along

the top and side walls. Fig. 1 depicts the cross section of the pre-
heating (or heating) zone. It should be noted that the left wall
has one row of nozzles more than the right wall. The nozzle
arrangement of the cooling section has two major differences. First,
nozzles are located only on the right wall. Second, they are aligned
in 10 rows, while in other sections the arrangement of the side wall
rows is tighter.
An arc-welding electrode has a central rod covered by a partic-
ular material. It has a length of 450 mm with inner and outer diam-
eters equal to 4.95 and 7.2 mm, respectively. A series of electrodes
are organized in a tray with the width of 700 mm as shown in Fig. 2
(a). Three columns, each one consisted of twenty-eight electrode
trays (numbered from bottom to top), are then organized as a
wagon, as shown in Fig. 2(b), and discharged to the furnace for
thermal treatment.

3. Numerical procedure

3.1. Turbulent air flow and energy equations

A fluid flow with velocity magnitude similar to the studied fur-

nace is normally considered laminar [22]. However, air enters the
welding electrode furnace from three intersecting paths and conse-
quently, the flow becomes turbulent. In this paper, Reynolds aver-
aging of the governing equations is employed to model turbulence.
In the Reynolds averaging technique, the solution variables in the
instantaneous Navier-Stokes equations are decomposed into the
mean and fluctuating components. Velocities are decomposed as
 i u0i
ui u 1

Fig. 2. Illustration of the arrangement of electrodes in each (a) tray and (b) wagon.

where ui, u i and u0i are the instantaneous, Reynolds-averaged and

fluctuating velocity components, respectively. Similar decomposi-
tion is used for other quantities
 /0
// 2
where / indicates a scalar variable such as the pressure and
enthalpy.
Substituting the above expressions into the instantaneous mass,
momentum and energy equations of an incompressible fluid and
taking a time average yields the following equations using the
Cartesian tensor notation:
i
@u
0 3
@xi



@ui 
@u 
@p @ @ui @  0 0
q j i 
u l  qui uj 4
@t @xj @xi @xj @xj @xj
!
@T @T j @ @T @  0 0
i
u  uT 5
@t @xi qcp @xi @xi @xi i

The Reynolds stresses, qu0i u0j , are related to the mean velocity
gradients under the commonly used Boussinesq hypothesis [23,24]



@ui @u
j 2 
@u
Fig. 1. Cross section of the preheating and heating zones. Dimensions are in qu0i u0j  qk lt k dij 6
@xj @xi 3 @xk
millimeter.
266 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274

Fig. 3. The exact electrode geometry and material composition (model A), and the suggested models (models B to E).

Table 1
Thermophysical properties of the electrode.

Parameter Unit Central rod Cover Mixed material

Density kg/m3 7801 3000 5269.4
Specific heat J/kg K 473 800 645.4
Conductivity W/m K 43 10 25.6


This hypothesis, however, introduces the turbulence kinetic @T j @ @T
9
energy k and its eddy viscosity lt as new variables. Numerous tur- @t qcp @xi @xi
bulence models are devised to model the new variables. The stan-
dard ke model, introduced by Launder and Spalding [25], has been 3.3. CFD models
used extensively to model turbulent flow and heat transfer in prac-
tical engineering applications. It is a semi-empirical model based 3.3.1. Electrode model
on phenomenological principles. We use the ke model in our sim- The existing length scales in the furnace are very diverse. Elec-
ulations because of its widespread use and also its simplicity. The trodes with a few millimeters diameter are cured inside a
standard ke model consists of two transport equations, namely multimeter-long furnace. Therefore, using the real geometry of the
the equations for turbulence kinetic energy and its dissipation rate electrodes in the numerical simulation will be challenging in terms
e. These are described by the following equations of good quality grid generation and efficient boundary condition

  implementation. The circular cross section and the two-material
@ @ @ lt @k j
@u
qk  i k
qu l  qu0i u0j  qe 7 composition of the electrodes impose further complications. There-
@t @xi @xj rk @xj @xi
fore, it is beneficial to replace the electrode with a computationally

  inexpensive but thermally equivalent model. In Fig. 3, four alterna-
@ @ @ lt @ e e2 tive cross sections with different material compositions are sug-
qe  i e
qu l  c2e q
@t @xi @xj re @xj k gested to simplify the electrodes modeling. Model A represents


e j
@u the exact geometry and material composition. Models B and C con-
 c1e qu0i u0j 8
k @xi sider two-material square cross sections, while models D and E
assume one homogeneous material for the circular and square cross
The turbulence eddy viscosity is defined as lt qcl k2 =e and sections, respectively.
the model constants are as follows [25] These models are evaluated under a numerical benchmark prob-
c1e 1:44; c2e 1:92; cl 0:09; rk 1; re 1:3 lem to select the most appropriate one to be used in the furnace sim-
ulation. The assumptions made for this problem are mostly derived
from the actual furnace operating conditions. Particularly, a single
3.2. Electrode energy equation electrode is analyzed under transient convection-conduction ther-
mal conditions. Since the length of the electrode is more than 60
The electrode material is modeled as a Fouriers solid with the times larger than its diameter, it is wise to carry out a two
energy conservation equation written as dimensional analysis and neglect the longitudinal dimension. The
 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
Table 2
Temperature of inlet air to the furnace.
Name of the zone Preheating
Heat exchanger number 1 2
Temperature (K) 353 388
Fig. 4. The maximum temperature difference across the electrode during the heat treatment process.
convection heat transfer is modelled as a convective boundary con-
dition applied on the electrode. This needs the free stream temper-
ature and the convection heat transfer coefficient to be specified.
Free stream temperature in each section of the furnace is considered
equal to the average temperature of the inlet air in the correspond-
ing furnace section (Table 2), i.e. 363 K. Convection heat transfer
coefficient in a convective furnace is commonly estimated using
the equation h = 1 + 0.225 V [26], which works with the English sys-
tem of units. As explained later in Section 3.4, hot air enters the fur-
nace with a velocity of 1 m/s (or 3.28 ft/s). Thus, the convection heat
transfer coefficient is readily calculated as 10 W/m2 K (or 1.738 Btu/
h ft2 F). The initial temperature of the electrode is set to 300 K. The
electrodes thermal treatment lasts 3600 s. The thermo-physical
properties of the electrode are assumed constant as reported in
Table 1. The mixed material column presents volume weighted aver-
age of the values for the central rod and the cover. It is applied for
models D and E.
Fig. 4 presents the maximum crosswise temperature difference
in the electrode (model A). Considering the fact that the maximum
temperature difference across model A during the transient simula-
tion is about 0.07 K, it can be argued that the temperature across the
electrode is almost uniform. In addition, the electrode has a very low
Fig. 5. Average temperature variation between model A and the other models.
267
Heating Cooling
3 4
393 393 313
Biot number value, i.e. Bi = 0.001. This further confirms that the solid
material can be considered as a lumped material [27].
Fig. 5 shows the temperature differences between the average
temperature of model A and other models. The results for model
B demonstrate that changing the electrode cross sectional shape
to a square has no significant effect on the results. The results for
model E demonstrate that the mixed material hypothesis leads to
tolerable differences as shown in the figure. Therefore, model E,
which has a simple square geometry and a homogeneous material
composition, is selected as the final electrode model.
Furthermore, the low Biot number of the model E makes it pos-
sible to consider each tray of electrodes as a two dimensional flat
plate of 4.95 mm thickness, hence having much less trouble in
the mesh generation.
3.3.2. Furnace model
The movements of the electrodes and the coexistence of largely
different length scales are two major issues which make the
numerical simulation lengthy and cumbersome, even if state-of-
the-art computers are utilized. The first issue requires moving grid
algorithms to account for the movement of electrodes. Maintaining
high quality grid and hence an accurate solution would typically
268 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
Fig. 6. The modeled furnace with an electrode column.
require remeshing techniques that are time consuming and prob-
lem dependent [28]. The second issue considerably increases the
number of computational cells, in the order of multi millions that
are required to sufficiently resolve small geometric features. In
Fig. 7. Computational grid. (a) The whole model and (b) the cross section.
the current research, efficient approaches are proposed to avoid
such difficulties. Furthermore, because every column of electrodes
in each wagon experiences identical operational sequence, only
one column plus its distance with the neighbouring columns
(0.59 m) is considered and then, periodic conditions are imple-
mented to the front and rear side boundaries. Fig. 6 displays the
sketch of the furnace model.
To avoid the moving grid, we assume that the wagon is station-
ary but furnace boundaries proceed backward with the same
velocity as the electrodes. The logical procedure which is followed
in the beginning of each time step is to calculate the central posi-
tion of nozzles arranged on the top and side walls. Next, they are
compared with the location of each computational cell. If the dif-
ference is less than the radius of nozzles (12.5 mm), normal veloc-
ity of the corresponding cell is set to the inlet air velocity. It should
be taken into account that FLUENT is not able to perform this speci-
fic task with its common features. Therefore, for each furnace wall,
one set of User Defined Functions is developed. User Defined Func-
tion is a programming feature that allows the user to customize
FLUENT and can significantly enhance its standard capabilities. It
is written in C programming language and could be dynamically
linked with the FLUENT solver.
As mentioned before, the furnace height and the distance
between trays are incomparable. Thus, producing a high quality
grid that is efficient in terms of the total number of computational
cells is not straightforward. To achieve these important goals, the
model is divided into several sub-regions so that each of them
can be meshed separately with hexahedral structured cells. Where
necessary, the procedure allowed changing the grid size to have
more control on the desired fineness of the sub-region. Note that
the solver ensures continuity of fluxes passing through the non-
conformal mesh interface between the sub-regions [20]. Fig. 7
shows two views of the generated computational mesh of
144,000 cells. The numerical grid is generated using Gambit soft-
ware [29].
 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274 269

Fig. 8. Temperature profiles along line AB of the 12th tray for three computational grids at (a) 30, (b) 75 and (c) 105 min.

3.4. Initial and boundary conditions
The column of electrodes enters the furnace with initial temper-
ature of 313 K (ambient temperature). At the beginning, the air
temperature inside of the furnace is 353 K. Inlet air is forced
through the nozzles at 1 m/s. The inlet air temperature varies from
one zone to another according to Table 2. Note that there are two
inlet air temperatures in both preheating and heating zones in
Table 2 because the air is supplied by two different heat
exchangers.
Because hot air flows on both sides of furnace walls, it is reason-
able to assume that the walls are well-insulated. According to the
aforementioned simulation strategy, periodic boundary condition
is implemented on the two planes along the column path. Since
270 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
the fluid velocity or pressure at the furnace outlet is unknown prior
to the simulation, it is considered as an outflow-type boundary
condition [20].
The energy equation of electrode trays needs either a known
temperature or heat flux value as the boundary condition. How-
ever, both of them are unknown prior to the calculation. Using
principle of heat flux continuity, the energy equation in the solid
and fluid phases are coupled to each other at the boundaries.
Therefore, the following relation is enforced at the fluid/solid
interfaces:



dT dT
j j
dn Fluid dn Solid
where n is the unit normal vector at the interface.
3.5. Computational details
The finite-volume commercial code FLUENT 6.3 is employed to
numerically investigate the transient three dimensional fluid flow
and heat transfer in the welding electrode continuous furnace.
Standard ke model is used for efficient evaluation of the flow tur-
bulence. The analysis uses coupled pressure-velocity formulation
(fully implicit). Second-order upwinding and fully implicit time-
stepping schemes are used for spatial and temporal discretizations,
respectively. The relaxation factors are 0.8 for both pressure and
velocity and 1 for the energy equation. The scaled residuals are
kept below 5  105 for the continuity and momentum equations
and below 1  106 for the energy equation. The simulations are
performed on a four-processor Pentium IV desktop (Windows 7
OS, 3 GHz, 2 GBRAM). A typical simulation takes 320 h (around
two weeks) to be completed.
We also included the loads temperature history obtained from
a simulation with the first-order upwinding scheme (see Fig. 9). In
this simulation, the scaled residuals are kept below 1  103 for
the continuity and momentum equations. All other simulation
parameters are the same.
4. Results and discussion
Although the time step size of the employed algorithm is not
restricted by any stability constraints, it should be kept sufficiently
small in order to obtain accurate results. Here, we tested the effect
of time step size on the mean temperature of the 12th tray along
its width (line AB in Fig. 2a) using 10 and 20-s time steps. The com-
Fig. 9. Comparison of simulated and measured temperatures at z = 10 cm along line AB of the 12th tray. Vertical dashed lines mark different zones.
parison revealed that the relative difference between the two solu-
tions is always less than 0.5%. As the former needs much more
computational resources, the latter is chosen for temporal
discretization.
The accuracy of the numerical simulation is strongly affected by
the grid size. Generally, finer computational cells result in a more
accurate solution. However, the significant increase in both com-
puter memory requirement and calculation time puts strict limita-
tions on the grid fineness especially in 3D simulations. Therefore,
numerical simulations are often tested using several grid sizes to
study grid independency. In this paper, three grids with 60,000,
10 80,000 and 144,000 control volumes are considered. The total
number of control volumes is mainly controlled by changing the
resolution of control volumes located in trays and the spaces
between them. The temperature profile along line AB of the 12th
tray is shown in Fig. 8 at (a) 10 min into the preheating process,
(b) 75 and (c) 105 min into the heating process. The maximum
variation between the temperatures of the finest grid and the
two other grids are below 1% and 0.5%, respectively. Since the grid
with 144,000 cells leads to a more accurate solution and is afford-
able with the available computational resources, it is used in the
remainder of this work. It should be noted that the temperature
profiles in Fig. 8 are not symmetric due to the fact that the nozzle
arrangements at the side walls are not completely identical.
Because no analytical or similar numerical solution exists for
welding electrode continuous furnaces, experimental data is used
to validate the simulation results. A data logger assembled with a
K-type thermocouple measures the 12th tray temperature at
z = 10 cm along line AB (see Fig. 2a). The numerical result is com-
pared against the experimental data in Fig. 9. The temperature
curves fairly follow each other during the whole process with an
absolute relative error below 2.7% for the second-order upwind
scheme. We also included the temperature curve obtained by a
first-order upwind scheme. The two numerical simulation curves
are very similar (the maximum temperature variation is 2.4 C),
however the maximum absolute relative error is around 3% for
the first-order scheme. Please note that the first-order upwind sim-
ulation took approximately 75 h, less than a quarter of the simula-
tion time for the second-order scheme. This suggests that the first-
order upwind scheme with its robustness and lower computational
cost could be used in the first trials of such industrial simulations.
The temperature curves in Fig. 9 are of bell-shaped type which
is common in such furnaces [30]. At the furnace entrance, the air
temperature is 313 K, i.e. the temperature of the air surrounding
the furnace. Temperature increases sharply at the start of
 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
Fig. 10. Air path lines inside the furnace at (a) 75 min (heating zone) and (b)
135 min (cooling zone).
preheating until approximately one third of the zone length. Then,
as the furnace is influenced by the second heat exchanger, the air
temperature experiences another rapid increase, coming to 380 K
at the end of the preheating zone. The heating rate starts to decline
in the heating zone and this condition is maintained until a tem-
perature of around 390 K is reached. In the cooling zone, the tem-
perature decreases approximately linearly and at last, the
electrodes leave the furnace at about 376 K.
Fig. 10 presents path lines, i.e. the trajectory of air particles from
the nozzles to the furnace outlet, at t = 75 (heating zone) and 135
271
(cooling zone) minutes. The path lines are colored according to
the wall from which they are originated. Because of the short dis-
tance between the trays, only a small proportion of the air pene-
trates to the inter-tray areas of the wagon. The majority of the
airflow passes around the wagon and is directed to the furnace out-
let, see Fig. 10a. Fig. 10b shows that the air simply enters the cool-
ing zone from the right wall (there is no nozzle on the left wall in
the cooling zone) and after passing around the wagon exits the
furnace.
Heat flux distributions of the electrode trays are demonstrated
at t = 75 and 135 min in Fig. 11. By convention the negative sign
means that heat is transferred to the surface. Note that the highest
and lowest trays have the maximum heat flux values. Due to the
fact that the whole top wall and a part of the side wall nozzles
expel hot air nearly perpendicular to the 28th (highest) tray, it is
subjected to a high heat flux rate. However, mass conservation
requires the velocity of air to increase at the furnace bottom. This
accounts for the increase in convection heat transfer coefficient
and hence heat flux around the 1st (lowest) tray. In comparison
with these trays, all the other trays show a rather similar behavior
and their heat flux declines considerably as we move from the
edges toward the center. This is mainly associated with the
decreased air velocity magnitude between the trays (as explained
in the discussion of Fig. 10).
Note that in the cooling zone (Fig. 11b), heat transfer occurs in
the opposite direction, i.e. from the load to the furnace, and hence
the heat flux is positive. Moreover, due to the asymmetric nozzle
arrangement in the cooling zone, a one-sided heat transfer is
observed for almost all of the trays in this zone.
Fig. 12ac compares the temperature of the 1st, 2nd, 12th, 27th
and 28th trays along line AB at t = 10 (preheating zone), 75 (heat-
ing zone) and 146 (end of the cooling zone) minutes. In the pre-
heating zone, there is no major temperature difference between
trays. However, as time advances, the differences become more
distinct. Moreover, because of the increased heat transfer near
the lowest and highest trays, they always have higher tempera-
tures compared to the other trays in the preheating and heating
zones. Since similar heat transfer condition governs the tempera-
ture variation of the other trays (2nd, 12th, 27th), they have similar
temperature profiles with values close to one another.
Leaving the heating zone, the 28th tray has the highest temper-
ature, however it gets a high volume of cold air flow coming from
five rows of nozzles above it. Thus, it is effectively cooled down. In
fact, it has the lowest average temperature at the end of the cooling
zone as shown in Fig. 12c. The 12th and 27th trays on the other
hand have the lowest temperatures entering the cooling zone.
The available cold air flow passing around them is sufficient to
lower the temperature of their right sides. The 1st and 2nd trays
also have relatively high temperatures when they enter the cooling
zone. The first tray gets proportionally more cold air flow passing
around it while the air flow around the 2nd is restricted due to
the short distance between the trays. Therefore, the 2nd tray has
a higher temperature as it leaves the furnace.
Finally, the air velocity increases around the left side of lower
trays due to the mass conservation. The increased convection heat
transfer coefficient on the left side combined with the continuous
flow of cold air coming from the right wall creates the local maxi-
mum in the temperature profile of the 1st and 2nd trays in Fig. 12c.
5. Conclusions
A three-dimensional numerical simulation of turbulent fluid
flow and heat transfer inside an arc-welding electrode continuous
furnace has been carried out using FLUENT software. To the best
knowledge of the authors, there is no previous numerical
272 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
Fig. 11. Heat flux distributions shown from two different angles at (a) 75 min (heating zone) and (b) 135 min (cooling zone). Heat flux is shown in W/m2.
simulation for this particular thermal problem despite its practical
importance. Large number of electrodes, disparity between the size
of electrodes and the furnace dimensions, transient and three-
dimensional nature of the analysis as well as the presence of
diverse modes of heat transfer contributed to the complexity of
the simulation. In this paper, two main simplifying assumptions
were applied. First of all, the geometry and material composition
of the electrode were replaced with a computationally inexpensive
and thermally equivalent model. Second, the electrode movement
was modelled using transient boundary conditions applied to a
fixed computational domain. This was implemented by linking
our own developed User Defined Functions to the main solver.
The computational domain was decomposed into several sub-
regions and was meshed separately for a better control on the local
refinements. After performing grid-independency tests, a compu-
tational grid consisting of 144,000 structured cells was selected.
The conservation equations of mass, momentum and energy were
discretized using finite volume method. Numerical simulation was
performed on a Pentium IV, 3 GHz processors with a 2 GB access
memory, on which a complete solution takes about 320 h. Some
of the most important results are summarized here.
1. There is a fairly close agreement between the numerical and
experimental temperature history of the load. The maximum rel-
ative temperature difference is always less than 2.7% for the
second-order upwinding scheme. This validates the approach
employed in the simulation. Moreover, our comparison shows
that the first-order upwinding scheme could be used in the first
trials of similar industrial simulations to obtain an acceptable
preliminary solution with a relatively lower computational cost.
2. Because electrode trays are arranged at very short distances
from each other, only a small proportion of the input air pene-
trates between them. The remaining air flows around the trays
and leaves the furnace before it could effectively exchange heat
with the electrodes. This leads to a significant decline in the
heat transfer efficiency. Therefore, it is necessary to re-design
the arrangement of trays and improve the thermal performance
of the furnace.
 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274 273

Fig. 12. Temperature curves along line AB of five trays at (a) 10 (preheating zone), (b) 75 (heating zone) and (c) 146 (end of the cooling zone) minutes.

3. The maximum heat flux occurs at the lowest and the high-
est trays. Consequently, these trays experience the highest
temperatures in the preheating and heating zones. Interme-
diate trays in the column have considerably smaller
heat fluxes and are cured under relatively identical
conditions.
4. The proposed numerical approach and the underlying assump-
tions deliver fairly accurate results at a reasonably low compu-
tational cost. Most of the strategies employed here are general
and can be adopted in similar industrial applications of contin-
uous furnaces to ensure a near optimal performance is achieved
or to go even further and effectively redesign future furnaces.
274 S. Defaee Rad et al. / Applied Thermal Engineering 117 (2017) 263274
Funding sources
This research did not receive any specific grant from funding
agencies in the public, commercial, or not-for-profit sectors.
References
[1] A.G. Fedorov, K.H. Lee, R. Viskanta, Inverse optimal design of the radiant
heating in materials processing and manufacturing, J. Mater. Eng. Perform. 7
(1998) 719726.
[2] M.-J. Huang, C.-T. Hsieh, S.-T. Lee, C.-H. Wang, A coupled numerical study of
slab temperature and gas temperature in the walking-beam-type slab
reheating furnace, Numer. Heat Transf., Part A 54 (2008) 625646.
[3] N. Therdthai, W. Zhou, T. Adamczak, Two-dimensional CFD modelling and
simulation of an industrial continuous bread baking oven, J. Food Eng. 60
(2003) 211217.
[4] S.-Y. Wong, W. Zhou, J. Hua, CFD modeling of an industrial continuous bread-
baking process involving U-movement, J. Food Eng. 78 (2007) 888896.
[5] H.K. Versteeg, W. Malalasekera, Introduction to Computational Fluid
Dynamics, first ed., Longman Group, Harlow, 1995.
[6] V.K. Singh, P. Talukdar, P.J. Coelho, Performance evaluation of two heat transfer
models of a walking beam type reheat furnace, Heat Transf. Eng. 36 (2015) 91
101.
[7] S.-Y. Wong, W. Zhou, J. Hua, Designing process controller for a continuous
bread baking process based on CFD modeling, J. Food Eng. 81 (2007) 523534.
[8] R. Oba, T.S. Possamai, V.P. Nicolau, Thermal analysis of a tunnel kiln used to
produce roof tiles, Appl. Therm. Eng. 63 (2014) 5965.
[9] R. Prieler, B. Mayr, M. Demuth, B. Holleis, C. Hochenauer, Prediction of the
heating characteristic of billets in a walking hearth type reheating furnace
using CFD, Int. J. Heat Mass Transf. 92 (2016) 675688.
[10] A. Emadi, A. Saboonchi, M. Taheri, S. Hassanpour, Heating characteristics of
billet in a walking hearth type reheating furnace, Appl. Therm. Eng. 63 (2014)
396405.
[11] S.H. Han, D. Chang, Optimum residence time analysis for a walking beam type
reheating furnace, Int. J. Heat Mass Transf. 55 (2012) 40794087.
[12] S.H. Han, D. Chang, C.Y. Kim, A numerical analysis of slab heating
characteristics in a walking beam type reheating furnace, Int. J. Heat Mass
Transf. 53 (2010) 38553861.
[13] T. Morgado, P.J. Coelho, P. Talukdar, Assessment of uniform temperature
assumption in zoning on the numerical simulation of a walking beam
reheating furnace, Appl. Therm. Eng. 76 (2015) 496508.
[14] S.K. Dubey, P. Srinivasan, Development of three dimensional transient
numerical heat conduction model with growth of oxide scale for steel billet
reheat simulation, Int. J. Therm. Sci. 84 (2014) 214227.
[15] J.G. Kim, K.Y. Huh, I.T. Kim, Three-dimensional analysis of the walking-beam-
type slab reheating furnace in hot strip mills, Numer. Heat Transf., Part A 38
(2000) 589609.
[16] C.-T. Hsieh, M.-J. Huang, S.-T. Lee, C.-H. Wang, Numerical modeling of a
walking-beam-type slab reheating furnace, Numer. Heat Transf., Part A 53
(2008) 966981.
[17] Y. Tang, J. Laine, T. Fabritius, J. Harkki, The modeling of the gas flow and its
influence on the scale accumulation in the steel slab pusher-type reheating
furnace, ISIJ Int. 43 (2003) 13331341.
[18] W.A. Fiveland, R.A. Wessel, FURMO: a numerical model for predicting
performance of three-dimensional pulverized-fuel fired furnaces, AIAA/ASME
Thermodynamics and Heat Transfer Conference, Boston, MA, June 24, 86-H-
T-35, 1986.
[19] S.M. Correa, W. Shyy, Computational models and methods for continuous
gaseous turbulent combustion, Prog. Energy Combust. Sci. 13 (1987) 249292.
[20] FLUENT 6.3 Users Guide, FLUENT Inc, Lebanon, NH, 2006.
[21] FLUENT 6.3 UDF Manual, FLUENT Inc, Lebanon, NH, 2006.
[22] F.M. White, Fluid Mechanics, seventh ed., McGraw-Hill, New York, 2011.
[23] J.O. Hinze, Turbulence, second ed., McGraw-Hill, New York, 1975.
[24] D.C. Wilcox, Turbulence Modeling for CFD, third ed., DCW Industries,
California, 2006.
[25] B.E. Launder, D.B. Spalding, Lectures in Mathematical Models of Turbulence,
first ed., Academic Press, London, 1972.
[26] F. Kreith, D.Y. Goswami, The CRC Handbook of Mechanical Engineering, second
ed., CRC Press, New York, 2005.
[27] F.P. Incropera, D.P. DeWitt, T.L. Bergman, A.S. Lavine, Fundamentals of Heat
and Mass Transfer, sixth ed., John Wiley & Sons, New Jersey, 2007.
[28] L. Zhang, X. Chang, X. Duan, Z. Zhao, X. He, Applications of dynamic hybrid grid
method for three-dimensional moving/deforming boundary problems,
Comput. Fluids 62 (2012) 4563.
[29] GAMBIT 2.3 Users Guide, FLUENT Inc, Lebanon, NH, 2006.
[30] P.S. Mirade, J.D. Daudin, F. Ducept, G. Trystram, J. Clment, Characterization
and CFD modelling of air temperature and velocity profiles in an industrial
biscuit baking tunnel oven, Food Res. Int. 37 (2004) 10311039.