Natural Convection Heat Transfer in Fluent
Natural Convection Heat Transfer in Fluent
2013
1 | Page
Contents
Table of figures...................................................................................................... 3
Introduction........................................................................................................... 4
Technical Section................................................................................................... 4
Problem Description........................................................................................... 4
Problem Setup.................................................................................................... 5
Results................................................................................................................ 6
Dimensionless Numbers................................................................................... 19
Conclusion........................................................................................................... 20
References........................................................................................................... 21
2 | Page
Table of figures
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
3 | Page
Introduction
A problem that needs analysis when looking at components such as heat sinks or
even ram modules on a motherboard is that of natural cooling. These
components will transfer heat to the surrounding fluid which increases its
temperature, resulting in a change in density of the fluid. This change in density
causes the heated fluid to rise, creating a plume of heated fluid above the
component. This effect has been investigated in FLUENT with a custom defined
fluid around an aluminium cooling fin that has a heat flux into its base, as it
would in reality from the component it is cooling. It has also been investigated
using various discretisation schemes, solvers and density approximations.
Technical Section
Problem Description
The problem that was modelled consisted of an aluminium fin, heated at its base,
surrounded by a fluid of custom properties. The domain also had a symmetry
plane across the centre of the fin and was open to the atmosphere on the side
and top of the domain. The base of the fluid domain was also specified as being
insulated. It was modelled as a 2-D geometry as only the plume for a section of
the fin was of concern. A geometrical description of the problem can be found in
figure 1.
The values for H, W as well as the other variables used in the simulation can be
found in table 1. All of the values used were derived from the student number
18121004. It should also be noted that the size of the domain was not pre4 | Page
defined as its effect on the solution was also of interest. Thus its size was varied
for different simulations in order to investigate the effect that the domain size
had on the results.
Table 1: Values used in Simulation
Property/Dimension
Cp (specifc heat)
R (gas constant)
k (thermal conductivity)
t
H
W
q'' (heat transfer rate)
T (operating temperature)
Value
1000 J/KgK
4000 J/kgK
0.001 W/mK
2 mm
10 mm
20 mm
4000 W/m2
(changed by lecturer during tutorial
session)
313.15 K
Problem Setup
The first step of the problem setup was to create the geometry for the domain,
which was done using the ANSYS Design modeller. The first step was to create a
sketch for the solid, which was the aluminium fin. Since the problem was being
modelled as 2-D; the option to create surface from sketches was used to create a
planar surface. This surface was then frozen so as not to merge with the fluid
region, which was sketched and created the same way as the solid region; with
the bottom left corner of the fluid domain coinciding with the bottom left corner
of the solid region. The next step was then to create a Boolean operation to
subtract the solid surface from the fluid region. The option to keep the tool body
was enabled in order to apply the q'' in at the base of the solid in FLUENT. Finally
both regions were selected and one part was created in order for the interaction
region to automatically be recognised by FLUENT. Once the geometry was
created the next step was to create the mesh for the problem. The mesh was
created in ANSYS meshing and in order to create a perfectly square mesh on the
region the edge sizing function was used in conjunction with the face sizing
function in order to enforce the cell size everywhere in the domain concerned.
This was done by selecting every edge, both solid and fluid in the domain, and
then applying the edge sizing. The surface sizing however was used only in the
fluid domain as the edge sizing function sufficed to enforce the cell size in the
solid region. One of the resultant meshes can be seen in figure 2.
5 | Page
During the meshing stage the named selections were also created in order to
assign the appropriate boundary conditions in FLUENT. The selections created
were: Pressure_Inlet, Pressure_Outlet, Symmetry, Q_in, Insulated and
Interaction_region which was the wall surface of the fin in contact with the fluid
region. The next step was to setup the problem in FLUENT, however this step
varied in order to investigate the effect of the different discretisation schemes,
different solvers, different density approximations and the effect of inviscid flow
on the solution. Apart from these differences, the common setup steps in FLUENT
involved first turning on gravity and defining its magnitude and direction. Next
the energy equation was turned on so as to model the heat transfer between the
solid and the fluid. Once this was done the fluid properties were changed to
those given in table 1 as they would be constant for every simulation. Next the
operating conditions were set with the operating pressure being set to 100 kPa,
as this was specified by the problem, and the operating temperature was set to
that in table 1. After this step the boundary conditions had to be set. The
pressure inlet and outlet were both defined to have 0 Pa gauge pressure and
temperatures of 313.15 K, as per table 1. The Q_in value was set to 4000 W/m 2
as per the values in table 1 and all the other conditions were set as default due
to FLUENT recognising that a boundary named Symmetry is a symmetry plane
and any unidentified boundary names are walls. Walls are also automatically set
to have zero heat flux across them, as with an insulated wall, thus there was no
need to change the boundary conditions at the insulated wall. Once the
boundary conditions were set, the discretisation schemes as well as pressurevelocity coupling was chosen before setting up surface monitors to better judge
the convergence of the solution as compared to looking purely at the residual
plots. The monitors that were setup included a mass flow rate across the outlet
as well as a surface temperature integral over the Interaction_region wall. When
the plots of both these monitors became approximately constant it was an
indication of convergence of the solution as the values being monitored were
longer changing. The solution was then initialised using Hybrid Initialisation
which solves the Laplace equation to produce a velocity field and smooth
6 | Page
pressure distribution in the domain (1). Once initialised the solution was
calculated and results extracted.
Results
When investigating the effects of the previously mentioned factors in the Finite
volume method, the following order of simulations was decided on in order to
compare the effects that they would have on the results as well as convergence
times:
1.
2.
3.
4.
5.
6.
The first simulation was run using the SIMPLE solver with the discretisation
schemes being left as their default values, namely, 2nd order upwind
discretisation for both momentum and energy. With the upwind differencing
scheme, cell face values upwind of the centroid is taken to be the same as the
value of the cell centroid upwind of the cell concerned (2). This is what is done
when the first order upwind scheme is used. With the second order upwind
scheme, the cell face values are instead computed using a Taylor series
expansion of the cell-centred solution about the cell centroid (1). See (1) for the
equation governing the calculation of the cell face values. The SIMPLE velocitypressure coupling scheme uses a relationship between the velocity and pressure
corrections in order to satisfy mass conservation and consequently solve for the
pressure field before moving onto the next iteration (1). The density model that
was also used for this simulation, as well as all the following simulations,
excluding the final one, was the incompressible ideal gas model. The
incompressible ideal gas model ignores the effect that pressure has on the
density and approximates the pressure based solely on the temperature
according to the following equation:
Poperating
(1)
R universal
T
M
This equation has been obtained from (1). When using this approximation the
density was calculated based on the operating pressure and temperature and
was specified as the operating pressure in the operating conditions in FLUENT.
The molecular weight M was also specified for the fluid material after specifying
the density as obeying the incompressible ideal gas law. After then following the
steps described in the Problem Setup section, the solution was found to have
converged after approximately 400 iterations when investigating the monitor
plots that were set up. This can be seen from figure 3 which shows the
convergence history of the temperature integral over the surface of the fin in
contact with the flow.
7 | Page
Whilst the surface monitors had converged by 400 iterations, the residuals were
set to converge only at a point where the continuity and energy reached a value
below 1x10-6. Thus the simulation was run until the point where the residuals
reached this value which occurred after approximately 660 iterations. The
resultant temperature contours can be seen in figure 4 which clearly illustrates
the plume above the fin.
8 | Page
The second simulation was run using the same settings as the first, however the
pressure-velocity coupling was changed to the Coupled solver. This algorithm
solves both the momentum and pressure based continuity equation
simultaneously (1). With this single change it was found that the solution
converged in 137 iterations. It was however found that the residuals converged
before the surface temperature integral plot did, and hence the solution needed
to be run further. This still resulted in a solution still converging twice as fast as
with the SIMPLE scheme in approximately 250 iterations. This was again judged
by the surface temperature integral plot as seen in figure 5.
The temperature contour plot for the second simulation can be seen in figure 6.
9 | Page
From figure 6 it can be seen that the solution has been largely affected by the
change from a SIMPLE to Coupled algorithm. This can be seen by the greater
width of the plume at the top of the domain. This solution however would appear
to be more accurate as the plume would be wider as a result of the width of the
base of the fin also heating the surrounding air, causing it to rise a result of the
change in density as a result of its rise in temperature. Due to this width of
heated air at the fin base; the width of the heated air at the top of the plume
would also be larger than simply the thickness of the fin.
Due to the faster convergence as well as producing an apparently more accurate
result, all subsequent simulations were run using the coupled algorithm. Once
this decision was made, the following two simulations were run in order to
compare the effect of the different discretisation schemes on the result as well as
the convergence rate. Seeing as the previous simulation was done using the
second order upwind discretisation scheme; only the effect of the first order
upwind and QUICK discretisation schemes needed to be investigated. The first
comparison was made with the First order upwind discretisation scheme. With
this scheme it was found that the convergence rate was almost exactly the same
as that for the second order scheme with convergence occurring at
approximately 255 iterations. The resultant temperature contour plot from as a
result of this scheme can be seen in figure 7.
10 | P a g e
From figures 6 and 7 it can be seen that there is no noticeable difference in the
results from using a second and first order upwind differencing scheme. The final
discretisation scheme that could be compared to was the QUICK scheme.
The QUICK scheme is a quadratic upwind differencing scheme which uses a
quadratic function passing through the cell centroid and two upwind cell
centroid's in order to find the face value of the cell concerned (3). When the
simulation was run using the QUICK discretisation scheme it was found that the
solution converged within approximately 240 iterations. This shows that it was
fractionally faster than the first and second order upwind schemes (only 4%
faster). The temperature contour plot resultant from the simulation run using the
QUICK scheme can be seen in figure 8.
11 | P a g e
From the comparison between figures 8,7 and 6, no apparent difference in the
results can be seen thus showing that the three different discretisation schemes
would lead to the same result if all other factors affecting the solution remained
constant.
After investigating the results from the coupled scheme, the effect of the domain
size on the solution needed to be investigated. This is especially true after
noticing that the domain used in the previous simulations did not capture the full
height of the plume above the fin. The larger domain size that was used can be
seen in figure 9.
12 | P a g e
From figure 9 it can be seen that the new domain size that was used was greater
than twice the domain size used in the previous simulations. When running the
simulations with the new, larger domain, the Coupled pressure-velocity coupling
was used along with the second order discretisation schemes. This was done as
the coupled solver converged the fastest as well as providing seemingly more
accurate results and the second order upwind schemes were the default in
FLUENT. It was found that even with the domain being more than twice as large,
the solution still converged within approximately 270 iterations. This was found,
as before, by monitoring the surface temperature integral plot due to the
residuals converging before the plot became constant. This plot can be seen in
figure 10.
13 | P a g e
The resultant temperature contour plot from the simulation run using the larger
domain can be seen in figure 11. It should be noted that with this larger domain;
the size of each cell remained the same as for the simulations run with the
smaller domain, that is, the mesh density was kept constant.
14 | P a g e
From figure 11 it can be seen that the larger domain captures a much larger
portion of the plume than as with the previous domain sizes. The general shape
and temperature contours of the plume however appear to be consistent as with
those found from the simulations run with the smaller domain size. Thus the only
observable effect of running the simulation with a larger domain size is the fact
that greater detail of the solution can be viewed at the cost of a fractional
increase in the solving time. For this reason, the subsequent simulations were
run with this larger domain size in order to gather greater detail of the results.
Beyond investigating the effect of domain size on the result of the simulations,
the mesh independence of the solution was also of concern. Due to the mesh
that had been used in the previous simulations being a relatively fine mesh, as
well as the FLUENT academic licence having a limit on the number of cells that
can be used; in order to check the mesh independence of the solution the
domain was re-meshed using cells that were twice as large. This resulted in a
mesh that was half as dense as previously used and would thus provide a result
that could be compared to the previous simulation in order to check the mesh
independence of the solution. The newly used mesh resulted in the convergence
of the solution occurring at approximately 140 iterations, as judged by the plot in
figure 12. However after investigation of the residuals after 140 iterations it was
decided that greater accuracy was desired and hence the simulation was run to
200 iterations, as can be seen in figure 13.
15 | P a g e
When the simulation was run using the rough mesh, the temperature contour
plot generated can be seen in figure 14. It should be noted that the simulation
was run with the Coupled solver as well as second order upwind differencing
schemes as this would allow for a comparison to previous results.
16 | P a g e
From the comparison between figures 14 and 11, it can be seen that the results
produced are unaffected by the change in mesh density. It should be noted
however that whilst the solution converged in less iterations, the actual runtime
of the solution was still not significantly less than that for the fine mesh due to
the fairly simple geometry. Thus, whilst the solution was seemingly unaffected by
the mesh, the following simulations were still run with the fine mesh due to the
fast solving time as a result of the simple geometry.
The next comparison that could be made was the effect of inviscid flow on the
results. When the model was switched from viscous-laminar flow to inviscid flow
however, it was found that thermal boundary conditions on the solid were not
available, which is shown in figure 15. In order to overcome this peculiarity, the
flow was set to viscous laminar but all the walls in the domain were set have a
specified shear stress of 0 Pa as this would simulate the effects of inviscid flow at
the walls. Using this approximation along with the coupled scheme and second
order upwind discretisation it was found that the solution converged within
approximately 260 iterations which was not much different from the solution
using the viscous-laminar model. The resultant temperature contour plot can be
seen in figure 16.
17 | P a g e
From figure 16 it can be seen that the results are drastically different than when
compared to the viscous laminar results. This can be attributed to the fact that
for the zero shear stress at the wall, the fluid velocity at the wall is non-zero and
hence the heat transfer rate to the fluid is not nearly as high as that for the
viscous-laminar flow. The inviscid solver also does not contain terms for
18 | P a g e
molecular diffusion which would largely affect the solution as the flow is at such
a low speed, the diffusion term would be a large contribution to the fluid
temperature within the domain (1). This is especially at the start of the
simulation when no flow is present within the domain. The velocity vectors of the
flow were also investigated due to the unexpected results, and this revealed that
the flow at the fin was in the opposite direction than the expected flow which
would result in a plume above the fin. Instead of the air rising as would be
expected, the velocity vectors of the air are as shown in figure 17.
From figure 16 it would appear as of the plume is now in the horizontal plane,
and this can be explained by figure 17 as the flow across the fin is now from top
to bottom and right to left. This flow gets heated and carries the heated air to
the left of the fin, creating the plume seen in figure 16.
This illustrates that the results obtained for an inviscid fluid, approximated by
applying zero shear at the walls, are highly inaccurate as it is known for this
scenario that a plume would be created in the opposite direction of the gravity
vector. Once it was established that the solution found using an inviscid fluid was
inaccurate, a comparison between the Boussinesq density approximation and the
incompressible ideal gas law density approximation needed to be investigated.
With the Boussinesq approximation, the density is set as constant in all the
solved equations with the exception of the buoyancy term in the momentum
equation (4). The main purpose for this approximation is to reduce the
convergence time of the solution and thus the affects of the Boussinesq
approximation on both the convergence time as well as final solution has been
19 | P a g e
1
= (2)
T
In equation (2), the temperature T was taken as the operating temperature of
313.15 K. In the operating conditions specification the operating density check
box had to also be disabled in order to implement the Boussinesq approximation,
whereas the operating pressure and temperature were left as 100kPa and 313.15
K respectively. The Coupled solver as well as second order upwind discretisation
schemes were again used for this simulation in order to compare the results to
previous simulations.
When using the Boussinesq approximation it was found that the solution
converged within approximately 150 iterations, which was in the region of 40%
faster than using the incompressible-ideal-gas law. The resultant temperature
contour plot found using the Boussinesq approximation can be seen in figure 18.
20 | P a g e
From the results obtained it can be seen that the results are indeed affected by
the use of the Boussinesq approximation as compared to the incompressible
ideal gas law. This approximation shows a much thinner plume, which was
established as being an incorrect result. This may be due to the fact that the
Boussinesq approximation is only valid for (4):
( T T 0 )=
( T T 0 )
T
1(3)
whereas the value for the left hand side of the above equation is close to 1 in
regions near the fin. This would result in the Boussinesq approximation providing
inaccurate results in the region near the fin which is exactly where the plume
base is. This would propagate through the domain resulting in an inaccurate
representation of the plume of hot air. It should be noted however that the
results from the Boussinesq approximation are fairly close to those achieved
using the incompressible ideal gas law as seen in figure 11. From this it can be
said that; whilst the results from the Boussinesq approximation appear to be
21 | P a g e
Dimensionless Numbers
When analysing flows where buoyancy can play a dominant role, the ratio of the
Grashof to Reynolds number squared is of concern in order to evaluate the
strength of the buoyancy contributions to the flow (1). This ratio is as follows:
Gr g TL
=
(4)
2
2
For the case of this ratio being 1 or greater then there would be large buoyancy
contributions to the flow. In this scenario, L is given as L+t/2 for the fin (5), and
this ratio was calculated to be in the region of 8, which indicates that indeed
there are strong buoyancy effects in the flow. The other number of concern with
regards to the flow is the Raleigh number, given by (1):
g T 3 C p
Ra=
(5)
k
For this scenario the Raleigh number was calculated to be in the region of
7.45x1015 which indicates a very strong buoyancy effect as anything above 10 8
would indicate turbulent buoyancy induced flow (1). This is a result of the large
temperature difference in the domain which dominates the numerator in
equation (5) and hence results in a large Raleigh number. When looking at the
heat transfer we can evaluate the Peclet number and the Prandtl number. For a
given flow geometry the Peclet number is given as (6):
Pe=
VL C p
(6)
k
Pr
C p
( 7)
k
and indicates the ratio of momentum to thermal diffusivity (6). For this flow this
is in the region of 147, which indicates a large amount of fluid movement with
relatively little thermal diffusion in the fluid.
22 | P a g e
Conclusion
From the simulations conducted it can be concluded that when analysing a
natural convection cooling problem, the different discretisation schemes are able
to handle a simple geometry all equally. It can also be concluded that the
solution is mesh independent for a simple geometry such as that which has been
simulated. A conclusion can also be drawn about the size of the domain for the
specific problem that has been modelled in that; the larger the domain size, the
more information can be extracted from the results at a relatively small increase
in the computational cost of the domain. It can also be concluded that the larger
domain does not affect the solution itself whilst allowing for more information to
be extracted from the results. When looking at the pressure-velocity coupling of
the solution it can be concluded that the SIMPLE algorithm results in greater
solution times and possibly less accurate results than the Coupled method. The
inviscid solution for the problem modelled can also be concluded as providing
inaccurate results due to not accounting for diffusion terms which largely affect
the solution at the start of the simulation. Finally it can be concluded that for this
specific problem the Boussinesq approximation is invalid but does however
provide results that appear to be correct except when compared to the solution
obtained with the incompressible ideal gas law. Due to the invalidity of the
Boussinesq approximation it can also be concluded that the incompressible ideal
gas law resulted in the most accurate solution to this problem.
23 | P a g e
References
1. Ansys-Help. Ansys FLUENT Theory Guide. Ansys FLUENT Theory Guide. s.l. :
ANSYS, 2011.
2. HK, Versteeg and W, Malasekera. 5.6 The upwind differencing scheme. An
Introduction to Computational Fluid Dynamics The Finite Volume Method. Essex :
Pearson Education Limited, 2007.
3. Versteeg, HK and Malasekera, W. 5.9 Higher-Order differencing schemes
for convection diffusion problems. An Introduction to Computational Fluid
Dynamics The Finite Volume Method. Essex : Pearson Education Limited, 2007.
4. Ansys-help. Ansys FLUENT User Guide. [book auth.] Ansys-Help. Ansys
FLUENT User Guide. s.l. : ANSYS, 2011.
5. Incropera, F P and De-Witt, D P. Overall Surface Efficiency. Fundamentals
of Heat and Mass Transfer. Massachusetts : John Wiley & Sons, 2002.
6. . Physical Significance of the Dimensionless Parameters. Fundamentals of
Heat and Mass Transfer. Massachusetts : John Wiley & Sons, 2002.
7. Ansys-help. Ansys FLUENT User Guide. [book auth.] Ansys-Help. Ansys
FLUENT User Guide. s.l. : ANSYS, 2011.
24 | P a g e