Numerical Modeling of Heat Transfer Around An Aluminum Reduction Pot Shell

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Light Metals 2004 Edited by Alton T.

Tabereaux TMS (The Minerals, Metals & Materials Society), 2004

NUMERICAL MODELING OF HEAT TRANSFER AROUND AN ALUMINUM REDUCTION POT


SHELL

Thierry Tomasino1, Céline Martin2, Emmanuel Waz1, Steeve Renaudier1


1
Pechiney Centre de Recherches de Voreppe, 725, rue Aristide Bergès, BP 27, 38341 Voreppe cedex, France
2
Aluminium Pechiney, LRF, BP 114, 73303 Saint-Jean-de-Maurienne Cedex, France

Keywords: Numerical modeling, Ventilation, Natural convection, Radiation, Aluminum reduction pot shell

Abstract
The use of numerical modeling allows evaluation of working 4
hconvection S (Twall  Tf )  VH S (Twall  Tf4 ) (1)
conditions and ventilation of electrolytic pots in the aluminum
industry. But particular attention has to be paid to the correct
analysis of all physical phenomena. A numerical 2D-model which with:
integrates natural convection and radiation in order to describe S area exposed to ambient air [m2]
heat transfer phenomena around an aluminum reduction pot shell, Twall temperature of the wall [K]
was developed with the commercial code FLUENT®. The Tf temperature of the ambient air [K]
temperature gradient between pot shell walls and ambient air V Stefan-Boltzmann constant [Wm-2K-4]
generates a velocity field. The induced turbulent flow requires the H emissivity of the wall [.]
use of a turbulence model associated to a wall function. The hconvection heat transfer coefficient due to natural and
objectives of this work are to determine the best adapted models forced convection [Wm-2K-1]
for both natural convection and radiation, and to consolidate
results through different numerical tests. The numerical results are Natural convection
compared to correlations and measurements on pot. A good Natural convection is due to a temperature gradient between pot
agreement is found. shell walls and ambient air. When the air gets warmer, its density
decreases and the air rises up. An ascending flow is created. Let
Introduction us consider a vertical wall at an imposed temperature Twall (Figure
In the aluminum industry, potroom ventilation becomes a subject 1). From dimensionless analysis, a characteristic velocity UNC of
more and more complex with the increase of the amperage and this phenomenon is obtained:
associated thermal power released by the pots and the building
cost constraints. The ventilation is all the more major since it has 1/ 2
U NC | g E'TL (2)
an impact on the process operation itself and, last but not least, on
the working conditions of operators around the pot. with:
UNC characteristic velocity [ms-1]
Numerical modeling allows to study all the physical phenomena
of ventilation. Nevertheless, this tool has to be used really 'T=(Twall-Tf) characteristic temperature difference [K]
carefully and the results obtained must be validated very L characteristic length [m]
rigorously. g gravity [ms-2]
E volume expansion coefficient [K-1]
A numerical 2D-model was developed with the commercial code
FLUENT® to calculate heat transfer around an aluminum For an important temperature gradient or a large characteristic
reduction pot shell. In this article, the model integrates only length, the flow becomes turbulent. Then the boundary layer
natural convection and radiation even if forced convection (wind thickness (G) increases significantly (Figure 1). The transition
or air lances) could be important. This will be taken into account from the laminar regime to the turbulent regime is determined by
later. the critical Grashof number, ratio between buoyancy forces and
viscous forces:
Results obtained by FLUENT® are first validated by correlations. g E'TL3
Then different floor materials have been tested to evaluate their Grc | 10 9 , knowing that Gr (3)
impact on heat transfer. Finally, heat transfer due to natural Q2
convection has been compared to the one due to radiation. with Q: kinematic viscosity [m2s-1].

The Grashof number is an equivalent to the square of the well-


known Reynolds number (Re), ratio between inertial forces and
Theory
viscous forces.
The different exchanges around an aluminum reduction pot shell
are natural convection, forced convection and radiation. The total Different correlations are available in the scientific literature to
heat flux can be written: evaluate the heat transfer at a vertical or horizontal wall in both
laminar and turbulent regimes ([1],[2],[3],[4]).
) total ) convection  ) radiation

433
one, the main part of the flowrate goes outside this boundary
layer. So, particular care has to be paid to correctly describe the
regions near the walls, especially for numerical modeling. For air,
the Prandtl number (Pr = Q/D with D the thermal diffusivity)
which represents in our case the ratio between the kinematic and
Turbulent
the thermal boundary layer thicknesses (respectively GQ and Gth), is
of order O(1), so:

U NC G | G th | GQ (6)
g Moreover, for a laminar regime:
L
G | LGr 1/ 4 (7)

Laminar G
Numerical modeling
FLUENT®, a finite-volume commercial code, is used to solve the
Ambient air turbulent Navier-Stokes and energy equations. For turbulence, an
Twall RNG k-H model has been chosen, as it is less diffusive than a
x
Tf standard model. A particular care is brought to the near-wall
modeling when natural convection is present. Since the
y temperature variation is large, it is necessary to take into account
the variation of viscosity and conductivity versus temperature.
Figure 1. Description of natural convection on a vertical wall. Sutherland and kinetic laws are used to describe respectively
viscosity and conductivity variations. The incompressible ideal
The heat transfer coefficient h can be deduced from the Nusselt gas hypothesis is also preferred to the Boussinesq’s approximation
number, ratio between the convective heat flux and the conductive as E'T | 0.5 is not small compared to one. An unsteady approach
heat flux, by: is always used to confirm results obtained with a steady model.
Wind has not been taken into account.
hL => NuO
Nu h (4) Natural convection modeling validation
O L
We first validate the natural convection modeling on a vertical
wall at an imposed temperature (Figure 2). There is neither forced
with O: thermal conductivity [Wm-1K-1]. convection nor radiation.
Forced convection
The forced convection is generated in a mechanical way (blowing
by fans, etc.). We can consider at the pot shell scale that wind is Adiabatic wall loutlet = 0.3 m
forced convection. When natural and forced convection are
present, the Richardson number (Ri) compares the importance of
these two phenomena: Outlet
2
U NC Gr
Ri 2 (5) Twall = 500 K
U FC Re 2
with: UFC a characteristic velocity of forced convection.
When Ri << 1, the forced convection is dominant but when Ri >> L=2 m
Symmetry
1, it becomes negligible compared to the natural convection. In
x
our case, both of them have to be taken into account.

)& y
Radiation g
Every body emits electromagnetic radiation. This energy emission Adiabatic wall
is done to the detriment of its internal energy. The radiation is a
heat flux emitted by the body, which varies as T4 (Equation 1).
The radiation of pot shell walls can be predominant in several Inlet : 1 bar, Tinlet = 303 K
places in comparison with natural and forced convection, so it is
very important to have a good description of this phenomenon. Figure 2. Description of the validation case.

When natural convection is predominant, the velocity profile near


Boundary layers the wall is very particular. In fact, buoyancy forces are present in
When a flow is due to natural convection, the velocity profile in the boundary layer, so this region cannot be described correctly
the boundary layer is quite different than the one obtained with with a standard wall function which gives the variation of velocity
forced convection (Figure 1). For the first one, the whole flowrate and turbulence with semi-analytic laws adapted to forced
goes through the kinematic boundary layer whereas for the second convection. In the two-layer model, the whole domain is

434
subdivided into a viscosity-affected region and a fully-turbulent A 30 % maximum difference between these correlations is
region. In the first one, the one-equation model of Wolfstein [5] is noticed.
employed while, in the second one, the k-H model is used. The
demarcation of the two regions is determined by a wall-distance- Figure 4 reports the heat flux profiles along the wall obtained
based, turbulent Reynolds number, Rey, defined as: numerically with a standard wall function (and several first mesh
sizes), a two-layer-model and correlations. In the laminar regime,
ky the standard wall function fails to predict the correct heat flux
Re y (8)
Q even if several first mesh sizes are tried. In all cases, fluxes
where y is the normal distance from the wall at the cell centers. depend on the mesh. In the turbulent regime, heat fluxes are
overestimated although the dependency on the mesh decreases.
The main disadvantages of the two-layer model are a thinner mesh
(because the first mesh size has to be close to the viscosity- The two-layer model gives fluxes closer to the McAdams
affected region thickness) and therefore a larger computing time. correlation in the laminar regime. In the turbulent one, heat fluxes
That is why the error due to the use of a standard wall function in are in the variation range of the other correlations (10), (11), (12)
comparison with a two-layer model and different experimental and (13). Contrary to the wall function approach, there exists a
results for vertical plates was evaluated. These are used to first mesh size, on the order of 1 mm, from which heat fluxes
establish a heat transfer variation range (Figure 3): become mesh independent (Figure 4).

- [1] and [3] use the correlation of McAdams for laminar flow In both cases, the transition to the turbulent regime occurs faster
(104 < Ra < 109): (x | 0.2 or 0.3 m) than that predicted by theory (x | 0.65 m).

Nu 0.59 Ra1 / 4 (9) Theroretical


2500
transition
first mesh=0,8cm (standard model)
Laminar Turbulent first mesh=1cm (standard model)
2200
- [2] and [3] use the correlation of Churchill and Chu for laminar first mesh=1,5cm (standard model)
first mesh=2,5cm (standard model)
and turbulent flows (10-1 < Ra < 1012): Heat flux [(W/m²] 1900
Correlation (9)
- laminar regime - Two-layer model

2 1600
§ 0.387 Ra1 / 6 ·
Nu ¨ 0.825  ¸ (10)
¨
©
1  (0.492 / Pr)9 / 16
8 / 27 ¸
¹
1300

1000
Heat flux variation
- [3] uses the correlation of Eckert and Jackson for turbulent flow range in turbulent
700 regime given by
(109 < Ra < 1013): correlations
400
0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8 2
Nu 0.021Ra 2 / 5 Length x [m]
(11)
Figure 4. Heat flux comparison between correlations and different
- [3] and [4] use the correlation of Bayley for turbulent flow and numerical approaches of the near-wall region (standard wall
functions and two-layer model).
Pr | 1 (2 x 109 < Ra < 1012):
To correctly model natural convection, a two-layer model has to
Nu 0.10 Ra1/ 3 (12) be used. It is the only way to ensure that results are mesh
independent and that fluxes are coherent with correlations in
- [4] uses the correlation of Bayley for turbulent flow and Pr | 1 laminar and turbulent regimes, even though the transition is
(2 x 109 < Ra < 1015): anticipated.

Nu 0.183Ra 0.31 (13) Radiation


The discrete ordinates (DO) radiation model [6] solves the
600 radiative transfer equation for a finite number of discrete solid
Bayley (12)
angles. The fineness of the angular discretization is very important
500 Bayley (13)
for the quality of the results and a particular care has to be paid to
Churchill and Chu (10)
its choice. On the other hand, solving a problem with a fine
400
Eckert and Jackson (11)
angular discretization may be CPU-intensive.

This method has been tested on a triangular cavity for which an


Nu

300
analytical solution can be obtained. In this case, heat fluxes and
200 temperature are in very good agreement with theory (error < 1%).

100

Application to an aluminum reduction pot shell


0
1.0E+09 1.0E+10 1.0E+11
Ra
Flow around conductors
Temperatures of the conductor walls are lower than pot shell ones.
Figure 3. Correlations for a vertical wall at constant temperature The flow is then laminar (Gr | 107). Using an unsteady laminar
in turbulent regime (natural convection). model, instabilities due to natural convection are found

435
downstream from the conductor (Figure 5). The total heat flux Ambient air at
varies only slightly with time on the top wall (') | 3%) and the T = 318 K and P = 1bar
flow induced by buoyancy forces is of secondary importance in Adiabatic
comparison with the flow around the pot shell. So, a steady-state concrete floor
model is considered to be sufficient to correctly describe heat Pot shell walls at imposed
transfer phenomena around conductors. temperature profile

However the use of a turbulent model is absolutely necessary


Conductor at
knowing that the flow around the pot shell is turbulent. Turbulent imposed
viscosity adds numerical diffusion so that instabilities downstream temperature
from the conductor disappear even with an unsteady model.

A slight impact of such modeling has been noticed on heat fluxes.

conductor Air inlet at T = 303 K


and P = 1 bar
Symmetry
conditions

Figure 6. Geometry of the model.

All results obtained with a steady-state model are confirmed by an


unsteady approach (computing time is then ten times larger).
Minimum imposed convergence criteria are 10-7 for temperature
and 10-5 for the other variables.

To ensure convergence, first calculation is done with natural


convection only. Then radiation is added using the results of the
Figure 5. Temperature contours (303 K < T < 400 K). first calculation as initial conditions. Computing time needed for
these two steps is between 2 and 3 hours on a Compaq® GS 320
supercomputer.
Numerical modeling around the pot shell
Figure 6 represents the geometry of the model and boundary Temperature and velocity contours are reported in Figure 7 and
conditions. GAMBIT, the FLUENT® preprocessing software, Figure 8.
permits to succeed in a mesh of 7 600 nodes while ensuring a
good transition from the mesh near wall to the rest of the domain.

Realistic profiles of temperature are used at the pot shell and


conductor walls. Even if, in reality, air inlet is perpendicular to the
(x,y) plane, the 2D modeling requires air introduction under the
pot shell.

Turbulence is modeled with a RNG k-H model associated with the


“Near-Wall Treatment” FLUENT® option [6] considering it is the
best adapted in the near-wall regions (it integrates a two-layer
model).

The Discrete Ordinate Method is employed to describe radiation


with the hypothesis of a non-participating medium. Emissivities
for the different walls are:

Hconcrete = 0.9 Haluminum = 0.06 (14)

Hsteel = 0.75 (15)


Figure 7. Velocity contours (0 < V < 1.1 m/s) for an adiabatic
concrete floor.

436
Boundary layers are very thin near the pot shell walls and the
velocity is maximum (V | 1.1 m/s) at the outlet. It is in good Table I. Comparison of the floor temperature versus different
agreement with experimental measurements. The flow generated boundary conditions.
by the conductor has a small influence on the flow around the pot Adiabatic h = 4 Wm-2K-1
shell. A maximum difference of 20% has been found on heat Boundary conditions conditions T = 303 K
fluxes between calculation and correlations, knowing that on the floor H = 0.9 H = 0.06 H = 0.9 H = 0.06
correlations do not take into account interactions between walls.
Temperature under
the floor (°C) 155 84 130 51

Computing time (s) 4775 3860 2779 4222

Comparison between natural convection and radiation transfer


Table II compares natural convection heat flux to radiation for
several walls. Natural convection is predominant on conductors
and quasi-negligible on the horizontal pot shell wall. For the
vertical pot shell wall, both phenomena are important.

Table II. Importance of natural convection heat flux compared to


radiation.
Pot shell Pot shell Conductor
horizontal vertical
wall wall
) NC
(%) 10 45 95
) NC  ) rad
Figure 8. Temperature contours (303 K < T < 400 K) for an
adiabatic concrete floor.
Conclusions
Floor boundary conditions sensitivity
In reality, there exists a heat flux through the floor. So another
boundary condition has been studied. Considering natural and We have developed a numerical 2D-model to describe the flow
forced convection, conduction and radiation, a heat transfer and heat transfer phenomena around an aluminum reduction pot
coefficient and a reference temperature have been evaluated: shell. It is a steady turbulent model. An unsteady one has been
also developed to validate, if necessary, this approach. A
h | 4 Wm-2K-1 and Tambient = 303 K (16) particular care has to be paid in the near-wall region when natural
convection is important. Standard wall functions cannot be used
because of its sensitivity to the first mesh size and the typical
If the concrete floor is covered with a sheet of aluminum, it only velocity profile in the boundary layer. But the best-adapted model
affects the emissivity of the floor. In that case, experiments show (the two-layer model) has the disadvantage to be very costly in
a significant decrease of the floor temperature as it is found with computing time. Another part of this work was to determine the
the numerical model (Table I and Figure 9). best compromise on the mesh between accuracy and computing
time. Actually 2 or 3 hours on a GS 320 COMPAQ
supercomputer are necessary to obtain calculation results.

Correlations and numerical results are in good agreement. We find


that heat transfer by natural convection is predominant for the
conductor while it becomes negligible compared to radiation on
the pot shell horizontal wall. On the vertical wall, both
phenomena are important.

We have modified boundary conditions of the floor and good


tendencies have been found: when an aluminum sheet is used, the
temperature is significantly decreased in this region. The
influence of the heat transfer coefficient has been shown on both
Figure 9. Temperature contours (303 K < T < 400 K) near the numerical results and computing time.
floor for h = 4 Wm-2K-1
H | 0.9 (left picture) and H | 0.06 (right picture). The next part of this work will be the integration of Forced
Convection network patented air-cooling system [7] and the
The heat transfer coefficient impact is essentially numerical since building of a 3D-model.
it affects computing time. In fact, temperature under the floor
decreases by roughly 30°C.

437
References

[1] B. Eyglunent, Manuel de thermique (Editions Hermes, Paris,


1997), 154.
[2] F. M. White, Heat and Mass Transfer (Addison Wesley,
1988), 402.
[3] J.P. Holman, Heat Transfer (Mc Graw-Hill, 1990), 343.
[4] L. C. Burmeister, Convective Heat Transfer (John Wiley and
Sons, 1982), 544.
[5] M. Wolfstein, “The Velocity and Temperature Distribution of
One-Dimensional Flow with Turbulence Augmentation and
Pressure Gradient”, Int. J. Heat Mass Transfer, 1969, no.12:301-
318.
[6] Fluent 6.1 Users Guide, Fluent Inc, 2003, 11.3.6.
[7] WO 99/54526: “Fused Bath Electrolysis Cell for Producing
Aluminium by Hall-Héroult Process Comprising Cooling Means”.
US 6,251,237: “Electrolytic Pot for Production of Aluminum
Using the Hall-Héroult Process Comprising Cooling Means".

438

You might also like