You are on page 1of 14

Modelling for Nuclear

Engineers: Thermal Hydraulics


Alexandros Kenich
30th January 2015

Faculty of Engineering
Imperial College London
SW7 2AZ

1 Introduction
The equations governing the flows of fluids in the physical world are
difficult to solve in all but the most simple of problems. It is only
recently with the advent of computers that scientists and engineers
have been able to use numerical techniques to simulate fluid flows
in complex geometries, such as in the primary loop of a pressurised
water reactor (PWR). These algorithms and numerical techniques
encompass a growing field known as computational fluid dynamics
(CFD).
In this report, several simple CFD simulations were created by using
the openFoam package and the results were studied in order to gain
an understanding of the strengths and limitations of these types of
software. Three different exercises were completed in this lab:

Laminar flow in a lid-driven cavity to analyse the effects of


increasing the mesh resolution and the Reynolds number.

Laminar flow in a duct with a backward-facing step to analyse


the effects of changing the flow viscosity and meshing
parameters.

Turbulent flow in a duct with a backward-facing step to


analyse the effects of turbulence on pressure and velocity
distributions.

In each case, confidence in the results was established by


performing appropriate mesh convergence studies.

2 Methodology
All exercises in this lab were completed by using openFoam on a
remote computer. This computer was connected to by using a

secure shell (via any terminal) while directories and files were
viewed using WinSCP.

2.1 Defining the Geometry


A set of coordinates (in a global reference frame) are specified in the
blockMeshDict file which correspond to the vertices of the geometric
model. These points are then grouped together to define the various
faces of the model and any boundary conditions at these faces. In
the case of the cavity exercise, all faces are static except the top
face (the lid), which is given a velocity in the x-direction.

2.2 - Meshing
Before performing any flow calculations, the volume that is to be
analysed must first be split into multiple discrete regions in a
process called meshing. An example of such a mesh with and
without cell grading can be seen in Figure 1. The cell size in the
mesh influences the accuracy of the results, with smaller cells
generally leading to more accurate solutions at the expense of
greater computational costs. In order to maintain numerical stability
in the simulation, a quantity known as the Courant number must be
less than or equal to 1. The Courant number is defined as:
Co=
Where

velocity, and

t
U
x

is the time step,


x

(2.1)
is the magnitude of the

is the cell size in the velocity direction. If a finer

mesh is desired (smaller

x ), the Courant number will increase

unless the time step is also decreased by the same proportion.


Higher resolution meshes will often require very small time steps in
order to maintain a Courant number less than 1, resulting in
simulations which take significant time and/or computational
resources to complete [1].

Figure 1 - 25x25 mesh of cavity geometry. Left mesh is uniform while right is
graded.

2.3 Running Jobs


The next step is to submit the job to the remote computer running
openFoam via the submit_foam command. Once the job is complete,
the relevant files are output in the current directory. These are then
copied onto the local machine in order to visualise the results in the
ParaView program.

2.4 ParaView
The final step is to process the results in ParaView to produce
relevant plots and make sense of the data. ParaView allows the user
to see a visualisation of the volume mesh and any mesh grading
which was applied in the first step. Scalar scenes are also produced
to analyse the evolution of flow behaviour over the length of the
simulation.

3 Results and Discussion


3.1 Lid-driven Cavity
The first exercise followed the lid-driven cavity tutorial as described
in the openFoam guide. Figure 2 shows several visualisations

created as part of the tutorial to gain a feel for some of the


capabilities of ParaView.

Figure 2 top left to bottom right: Pressure map, velocity arrows, stream tracer,
graded mesh.

3.1.1 Increasing Reynolds Number


The Reynolds number is given by:
=

DU

Where D is the characteristic length, U is the flow velocity,


the fluid density and

is the fluid viscosity [2]. The quantity

(3.1)

is

is known as the kinematic viscosity; this quantity was decreased in


the openFoam transportProperties file in order to increase the
Reynolds number. It was found that increasing the Reynolds number

reduced the depth at which the effects of the moving lid were felt by
the flow. Higher Reynolds numbers mean that the inertial forces of
the flow become more pronounced than the viscous forces, and
therefore the boundary layer thickness (a phenomenon influenced
by viscosity) is reduced. Peak pressures always occurred at the top
left and top right corners, as shown in Figure 2. Peak velocities
always occurred at the moving wall due to the no-slip condition.
3.1.2 Peak Shear Stresses
The peak shear stresses occur at the wall. This is because the shear
stress is a function of the velocity gradient as follows [3]:
=

du
dy

(3.2)

The greatest velocity gradient is always at the moving wall because


it drives the flow at the highest speed due to the no-slip condition.
This can be seen in the velocity plot in Figure 3.

Figure 3 - Plot of velocity in x-direction against bottom wall distance.

This is an important quantity in the context of nuclear thermal


hydraulics because heat is removed by contacting the hot fuel
cladding with a fluid coolant, and the wall shear stress tells us how
effective this contact is. At a microscopic level, greater contact at

this interface allows greater heat diffusion into the coolant via the
interaction of atoms.
3.1.3 Confidence in Results
Confidence

in

the

results

was

achieved

by

performing

convergence study of the maximum pressure with increasing mesh


resolution as shown in Figure 4.
12
10
8
Peak Pressure (Pa)

6
4
2
0
0

500

1000

1500

2000

Number of Cells
Figure 4 - Plot of peak pressure against number of cells showing convergence.

It was found that the peak pressure converged towards a value of


~10.1Pa for a 40x40 mesh.
3.1.4 Mesh Grading
The type of graded mesh used for the cavity exercise was shown
earlier in Figure 2. The grading used in this case was 0.2 in the y
direction. This focused on improving the cell density closer to the
moving wall because this is the region where the greatest changes
in pressure and velocity occurred and therefore refinement was
required to improve accuracy of the results.

3.2 Backward-facing Step (Laminar)


The backward-facing step geometry is shown in Figure 5 with a
superimposed velocity distribution.

Figure 5 - Geometry of backward-facing step with typical laminar flow pattern.

3.2.1 Changing Initial Velocity


The initial velocity of the flow was studied at 10ms -1 and 100ms-1.
Scenes of the two scenarios are shown in Figure 6. The pressure
distributions along the duct are also shown in Figure 7. In both
cases, the flow initially accelerates due to the pressure gradient in
the duct. The 10ms-1 case shows a gradual decrease in pressure
with a drop below gauge at the region around the step, indicating
some level of backflow. Shortly after the step, the velocity
distribution reaches a steady state. The 100ms-1 case shows that the
flow is moving so fast that it does not reach a steady state velocity
distribution before it leaves the duct. The pressure distribution does
not show a gradual decrease as in the slow case. Instead, there is a
negative pressure gradient in a large region at the centre of the
duct.

Figure 6 Velocity distributions for U = 10ms-1 and U = 100ms-1 respectively.

Figure 7 - Pressure distributions for U = 10ms-1 and U = 100ms-1 respectively.

3.2.2 Confidence in Results


The convergence study for this exercise was performed by looking
at the peak pressure as a function of the number of cells, as shown
in Figure 8. The initial velocity used was 1ms -1 in order to keep the
Courant number low. In this case, the lowest resolution (20x20)*8
mesh was already close to convergence, with only a 0.12%
deviation from the convergent value of 39.75Pa. This may have
been due to the greater number of cells over which the flow evolved
than in the cavity case.
39.76
39.75
39.74
39.73
39.72
Peak Pressure (Pa) 39.71
39.7
39.69
39.68
39.67
0

5000

10000

15000

Number of Cells
Figure 8 - Plot of peak pressure against number of cells for backward-step case.

3.2.3 Mesh Grading


The mesh grading used in this exercise is shown in Figure 9. This
focused on the regions immediately after the step where the flow
pattern suddenly changes.

Figure 9 - Graded mesh for the backward-facing step.

3.2.4 Limitations of the Analysis


This analysis has limitations as a result of the flow regime constraint
(laminar). The sudden increase in cross-sectional area as the flow
reaches the step will normally increase the local Reynolds number
since the characteristic length has increased. The flow velocity does
decrease, but in a more realistic model we would see a more
complicated flow pattern than that in Figure 5 immediately after the
step due to this transition [3].
Other limitations are due to the 2D only consideration (implying
infinite length in the z-direction) and the incompressibility of the
flow. In a model where the flow is compressible, we should see some
choking and even increasing velocity (for Mach > 1) of the flow at
higher speeds [4].

3.3 Backward-facing Step (Turbulent)


3.3.1 Changing Reynolds Number
A high and low Reynolds number (kinematic viscosities of 0.0001
and 0.1 respectively) for the flow were modelled with the pisoFoam
program which provides a turbulent model for the fluid. Figure 10
shows the x-component velocity distribution in both cases. It can be
seen that for low Reynolds numbers, the flow is similar to the
laminar case as is expected. For high Reynolds numbers however,
the flow becomes chaotic immediately after the step and then
quickly forms a velocity distribution that is almost independent of
the spatial position.

Figure 10 - High and low Reynolds number velocity distributions respectively.

3.3.2 Pressure Change


The pressure change for the low Reynolds number case was again
very similar to the laminar case, as shown in Figure 11. The high
Reynolds number case however, produced a localised low-pressure
singularity, possibly indicating the formation of a vortex.

Figure 11 - High and low Reynolds number pressure distributions respectively.

3.3.3 Confidence in Results


To achieve confidence in the results, a mesh convergence study was
done for the low Reynolds number case (kinematic viscosity = 0.1)
because pressure values for the turbulent case were unpredictable.
A (40x40)*8 mesh was not tested because of the time taken to
produce the mesh being too great. A plot of the convergence of
peak pressure is shown in Figure 12.
38
36
34
Peak Pressure (Pa)

32
30
28
26
0

5000

10000

15000

Number of Cells
Figure 12 - Plot of peak pressure against number of cells for the turbulent duct
case.

The pressure can be seen to approach a convergent value, but a


finer mesh must be used before we can be sufficiently confidence
that the results produced are fully converged.
3.3.4 Special Treatments at Walls
For the turbulent case, special treatments are used at the walls
because turbulent flows can generate very large velocity gradients
and therefore shear stresses near the walls [1]. These must be
limited to avoid instabilities arising in the simulation.

3.4 Multiphase Flows


Multiphase flows are encountered in nuclear reactors wherever
nucleate boiling (or higher boiling regimes) is involved, such as in a
boiling water reactor (BWR). In this case, the flow consists of the
liquid phase and the gas phase. Such a system can be modelled by
constructing definitions of the flow based on the void fraction and
approximations such as a two-phase density. Such a two-phase
density can then be used to obtain derived quantities such as the
Reynolds number [5].

4 Conclusion
Fluid flow in a lid-driven cavity and a duct with a backward-facing
step was modelled using the openFoam software package, with
turbulent flow also being modelled for the backward-facing step. It
was found that in the lid-driven cavity exercise, larger Reynolds
numbers would decrease the size of the boundary layer. Higher
resolution meshes were also found to provide better-converged
solutions, at the expense of longer computation times.
The laminar case for the duct exercise showed that at low speeds,
the flow will first feel a disturbance at the step which disconnects
the boundary layer, and then reach a steady state velocity
distribution in the larger duct with a new boundary layer formed. In

a small region around the step, the local pressure fell below gauge
and some backflow was observed. At higher speeds, the flow did not
have enough time to reach a new steady state velocity distribution
in the larger duct and a large region of low pressure was observed
after the step.
The turbulent case for the duct showed that low Reynolds numbers
produce results similar to those in the laminar case as would be
expected. At high Reynolds numbers however, the flow becomes
chaotic after the step and no longer resembles the laminar case
with a clear boundary layer.
Further studies should be done to optimise grading and mesh sizes
as these would significantly help reduce the computational costs of
the simulations, particularly the turbulent flow scenario where
pressure and velocity distributions are more complex. Other
improvements include using compressible models to analyse the
effect of choking and also studying the effect of a forward-facing
step.

5 References
[1]
Bluck. M, Modelling for Nuclear Engineers: Thermal
Hydraulics, 2014
[2]

Happel, John, and Howard Brenner, eds. Low Reynolds number


hydrodynamics: with special applications to particulate media.
Vol. 1. Springer Science & Business Media, 1983.

[3]

Fox, Robert W., Alan T. McDonald, and Philip J. Pritchard.


Introduction to fluid mechanics. John Wiley & Sons, 2006.

[4]

Courant, Richard, and Kurt Otto Friedrichs. Supersonic flow


and shock waves. Vol. 21. Springer Science & Business Media,
1977.

[5]

Sun, Licheng, and Kaichiro Mishima. "Evaluation analysis of


prediction methods for two-phase flow pressure drop in mini-

channels." International Journal of Multiphase Flow 35.1


(2009): 47-54.

You might also like