Sie sind auf Seite 1von 21

Imperial College London

Department of Materials
MSE 9NMNE - Modelling for Nuclear Engineers

Modelling laminar and turbulent flows in a


vertically heated pipe using CFD
Author:
Jonathan Dixon

Abstract
This report simulates laminar and turbulent flow through a vertically heated pipe at T =
293K using the specific heat of water. The mesh was improved notably on the outer regions
to refined the velocity gradient and produce a profile more representative of the Poiseuille flow
for laminar and correcting turbulence kinetic energy profiles in the turbulent case. The results
obtained illustrate that for a laminar case the speed is slower nearer the wall with a peak velocity
was 0.0196ms1 and pressure drop per unit length from the centre to the edge of 3104 P am1 .
The turbulent peak velocity was larger at 0.141ms1 , resulting in a larger pressure drop per
unit length of 5 103 P am1 . The Moody charts obtained for both laminar and turbulent
flows show good agreement with literature, but with some margins of difference.

November 28, 2016

Contents
1 Introduction

2 Background
2.1 Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1 Characteristic Numbers . . . . . . . . . . . . . . . . . . . . . . .
2.1.2 Laminar Viscous Flow . . . . . . . . . . . . . . . . . . . . . . . .
2.1.3 Turbulent Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . .
2.1.5 Reynolds-Averaged Navier-Stokes and Boussinesq Approximation
2.1.6 k  Model and Turbulence Kinetic Energy . . . . . . . . . . . .
2.2 Moody Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Model Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2
2
2
2
3
3
3
4
4
6

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

3 Experimental Methods

4 Results
4.1 Laminar Flow: Ogrid Mesh . . . . . . .
4.2 Laminar Flow: Enhanced Ogrid Mesh .
4.3 Turbulent Flow: Ogrid Mesh . . . . . .
4.4 Turbulent Flow: Enhanced Ogrid Mesh

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

7
7
9
11
12

5 Discussion
5.1 Laminar Flow: Ogrid Mesh . . . . . . .
5.2 Laminar Flow: Enhanced Ogrid Mesh .
5.3 Turbulent Flow: Ogrid Mesh . . . . . .
5.4 Turbulent Flow: Enhanced Ogrid Mesh

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

15
15
15
16
16

6 Conclusion

18

Introduction

This report uses Computational Fluid Dynamics (CFD) to model the differences between laminar
and turbulent regimes. CFD is used as it enables many iterations and determination of parameters
unlike conventional system codes which tend to be deterministic and operate in 1 D. The
motivation behind such studies is to understand how the heat flows within a pipe and how to
obtain the best heat exchange which is useful within the nuclear industry. One way of doing
this is seeing how the velocity profiles differ, one that offers a better distribution across the pipe
can allow for more homogenous heating. In the nuclear industry pipes can be configured to force
a turbulent flow, but in this report the geometry is not altered and only the maths behind the
physics is considered. Another source of motivation is to better understand related issues in the
nuclear industry such as bubble formation and although CFD is used in this report, the industry
relies on both CFD and system codes to probe thermal-hydraulics [1, 2].
Section 2 covers an overview of fluid flow for both laminar and turbulent regimes. It also
discusses some common properties of fluids, notably characteristic numbers, Navier-Stokes averaging,
different turbulence models and Moody charts. Section 2.3 outlines some of the assumptions made
within this report to run the simulations and section 3 compares the different experimental methods

and motivations conducted. The results and discussion are made in sections 4 and 5 respectively,
before concluding in section 6.

Background

2.1

Fluid Flow

The flow of a fluid can be broken down into three groups, ideal, laminar viscous flow and turbulent
flow. Some choose to sub-define these even further by defining a critical region between the laminar
and turbulent regimes, but for the purposes of this report we only consider the two cases. An ideal
fluid is a hypothetical fluid flow that obeys the continuity equation given by eq (1), where the rate
of change in density of the fluid () is related to the fluid velocity field gradient (u).

+ (u) = 0
t
2.1.1

(1)

Characteristic Numbers

For this report there are two characteristic numbers that are of interest. The first is the Reynolds
number given by eq. (2) which represents the ratio of inertial forces to viscous forces using the
characteristic velocity (u), characteristic scale length (L) and the shear viscosity () [3, 4]. This is
of importance as it can be used to determine whether a flow is laminar or turbulent, with a laminar
flow occurring when Re < 2000.
The second important characteristic number is the Prandtl number, defined as the ratio between
the momentum transport and heat transport, using the heat capacity (cp ) and thermal conductivity
(k). For the turbulent case in this report, this has a value of P = 0.9. Further more, the
turbulent case experiences a distribution in its velocity profile, therefore a mean velocity and
mean temperature are used within the transport properties.
uL

cp
P =
k

Re =

2.1.2

(2)
(3)

Laminar Viscous Flow

The velocity profile along a circular pipe is given by eq. (4) known as the Poiseuille flow where
is the shear viscosity, a is the pipe radius, 0.05m within this report, r is the distance from the
pipe axis, p is the pressure and z is in the direction of flow. Therefore, when a = r, at the edge
of the pipe, the velocity of a laminar regime will be zero. This is a key feature of laminar flow
expected within results. Another key property of this equation is the gradient of the pressure, this
is an important component within the laminar regime more broadly as it is also represented within
the volume discharge rate (Q), given by eq. (5) where the rate of change in volume (V ) is directly
proportional to the pressure gradient. This importance is mapped within a Moody chart, discussed
in section 2.2.
vz =


1  2
a r2
4
z

(4)

dV
a4
=
dt
8 z

(5)

Q=

2.1.3

Turbulent Flow

At some critical velocity, the flow will undergo characteristic changes that sees it begin described
from a laminar flow to turbulent. In this process eddies are formed and chaotic motion arises
and unlike as seen in the laminar case these do not contribute to Q = dV
dt , the volume flow
rate [5]. The dependence on pressure is much weaker and therefore a large increases in pressure
is necessary to increase Q. Turbulence sees convective transport and molecular mixing which
also demonstrates a more random motion due to the velocity depending on the mean values (ui )
and the mean fluctuating values (u0i ), known as Reynolds decomposition given by eq (6). It
requires Re > 2000, through either a greater characteristic length and flow velocity or a lower
viscosity. Some other properties of a turbulent flow is the mean of the fluctuation is zero, but the
mean of the squared fluctuation is non-zero, with the averages represented by root mean squared
values. Two distinguishable types of turbulence include a statically homogenous turbulence and
statistically isotropic turbulence. The former represents invariant fluctuating quantities when put
under translation of the coordinate system. Whereas the latter has this invariance represented
more globally through translation, reflection and rotation, where the mean velocities are zero. This
case is useful especially on small scales which can be assumed to be isotropic and benefit from the
simplification of mathematics. In constructing meshes fine enough to model turbulence motion, it
is of interest to achieve properties that represent small scales.
ui = ui + u0i
2.1.4

(6)

Boundary Conditions

The Dirichlet boundary conditions requires the value of a function on the surface such that f (r, t) =
S, which in our model is the wall temperature [2, 68]. Alternatively, the Neumann boundary
conditions can be used which requires the normal derivative of the function on a surface, f (r, t) =
S, which would specify the wall heat flux. For this report the latter will be considered, such
n
T
that U
z = z = 0. Figure 1 illustrates the boundary conditions and also demonstrates the Mesh,
whose cells are modified later in this report to improve results.

Figure 1: (a) Ogrid mesh used in CFD throughout this report where areas of the same colours have
the same representation of nodes. (b) the boundary conditions visualised on a pipe [6, 7]
2.1.5

Reynolds-Averaged Navier-Stokes and Boussinesq Approximation

The equations of motion used within this report for both laminar and turbulent motions are the
time-averaged Reynolds-Averaged Navier-Stokes (RANS) equations which employ the aforementioned

Reynolds decomposition given by eq. (6). Assuming an incompressible stationary fluid, the
equations are represented by eqs. (7), (9) [7, 9]. Therefore the simulation inputs values for the
initial velocity u), density (), initial temperature (T ), the effective kinematic viscosity (vef f ), the
thermal expansion coefficient () and the effective kinematic diffusivity (ef f ). A gravity field (g)
is also defined, but this is neglected in this report [7]. The difficulty of these formulae is that it is
not closed due to the nonlinear stress term and cannot be solved computationally without making
further approximations.
u=0
(u ) u =


1
p + vef f 2 u + 1 (T T0 ) g

u T = ef f 2 T

(7)
(8)
(9)

ne of the approximations used in this report is the Boussinesq approximation which relates the
Reynolds stresses to the mean velocity gradients [10]. This introduces the turbulence kinetic energy
(TKE, k). In this case the kinetic eddy viscosity is assumed to be an isotropic scalar quantity. Other
models also exist such as the Spalart-Allmaras model which neglects the TKE. The other model
assumed is known as the k  model outlined in the next section.
2.1.6

k  Model and Turbulence Kinetic Energy

The k  model is based on model transport equations which derives the turbulence kinetic energy
(k) and defines the dissipation rate (), whose kinetic eddy viscosity depends on these two quantities.
Within this model it is assumed a fully turbulent flow neglecting molecular viscosity effects. The
Turbulence Kinetic Energy (TKE) is defined as the mean kinetic energy available per unit mass.
The total derivative of the TKE is given by eq. (10), where the divergence represents the turbulence
transport, P represents the production and  represents the dissipation in TKE. These are based
on the law of the wall which states that the average velocity at a specific point is proportional to
the logarithm of the distance between that point and the boundary, or wall. Other laws exist for
the regions below this, closer to the centre, although as an approximate solution this can be applied
throughout the pipe.
More intrinsically, the full form of the TKE contains the pressure diffusion, turbulent transport,
molecular viscous transport, buoyancy flux and the difference between production and dissipation.
Dk
+T =p
(10)
Dt
The smallest scales in a turbulent flow are known as the Kolmogorov microscales, these are
important as the rate at which energy is supplied is equivalent to the rate of dissipation through
viscosity. Majority of the TKE is within the macroscale where the energy is able to cascade from a
macroscale to a microscale structure. Importantly it is able to tell us if a flow rate is computationally
feasible from the Kolmogorov length scales.

2.2

Moody Chart

One important parameter in studying the behaviour within pipes and nuclear systems is the pressure
drop within the pipe circuit system. This characteristic is defined by the Darcy-Weisbach equation,
given by eq. (11), where the change in pressure (p) is proportional to the length of the pipe (L),
4

the mean flow velocity (hvi) the diameter of the pipe (D) and the internal density (). This
proportionality is made equal through the Darcy friction factor (fD ) [11]. This friction factor
varies between materials and has been determined empirically for many materials. This equation
holds true in a cylindrical pipe of uniform diameter. One of the dependence is the Reynolds number,
which when plotted with the friction factor is called a Moody chart [12, 13]. A Mood chart can
distinguish between laminar and turbulent regimes because of its use of the Reynolds number and
therefore determine how fD varies according to different materials.
p
hvi2
= fD
L
2D

(11)

Figure 2 illustrates the Moody chart demonstrating the laminar regime and complete turbulence
flow. It also illustrates a transitional or critical area, although in this report we neglect this [12].
Interestingly, the laminar flow shows a log-based linear dependence between the friction factor
and the Reynolds number, yet a turbulence regime experiences a logarithmic dependence before
leveling off at different pie roughness values (). It is these values of  that differentiate the different
turbulent regimes, whereas the laminar regime material independent. Thus values of fD closer to
zero will minimise the pressure drop, which corresponds to small values of .

Figure 2: Moody chart illustrating relationship between the friction factor and Reynolds number,
showing the laminar and turbulent regimes [12, 13].
The Laminar regime defined as Re < 2000 is determined by the linear relationship, fD = 64/Re,
although this does imply given Re hvi that eq (11) is not the best representation for the laminar
flow as compared to the turbulent regime, given this implies fD hvi2 . In the laminar case
frictional loses occur from momentum transfers within fluid particles between the velocity gradient
at the centre to the edge of the pipe, due to the ranges of velocities involved [11, 14].
For the turbulent regime, assuming a smooth-pipe the logarithmic dependence is given by
Karman-Prandtl equation, although in order to obtain a usable equation in terms of having a closed

form expression for fD , the Lambert W function is employed, shown by eq. (13). In contrast to
the laminar flow, the momentum is transformed through small eddies or vortices [15].

fD =

1.93 log


fD =

2.3

1.93 W

1.90

Re fD
Re
3.667

!2
(12)

!2
(13)

Model Assumptions

A number of assumptions are made:


1. The wall boundary has a no-slip condition, such that the velocity is described by Uwall = 0
T
2. A Neumann boundary condition is assumed of the wall heat flux, such that U
z = z = 0.
3. We ignore any intermediate regimes, notably the critical region that sees a laminar flow become
turbulent. We assume the internal and external diameters are the same
4. A Newtonian fluid is assumed such that the viscosity does not change with the velocity.
5. An incompressible fluid is assumed implying that the density is a constant. This is also important
within the reynolds decomposition such that eq. (6) holds true.
6. The fluid is assumed stationary defined as where the velocity vector at a given fixed point does
not alter with respect to time.
7. The derivation of the RANS equations requires the mean of the fluctuating velocities to be zero,
u0 = 0.
8. Assume a smooth pipe, such that the frictional factors variation with the Reynolds number can
be modelled by eq. (12).
9. The Boussinesq approximation is employed to the Reynolds Averaged Navier-Stokes equations,
such that the kinetic eddy viscosity is assumed to be an isotropic scalar quantity.
10. A turbulent prandtl number of 0.9 is assumed.

Experimental Methods

This report is divided into four core results based on Computer Fluid Dynamics using a package
called OpenFoam [16]. The first is a laminar Flow in section 4.1 which uses an Ogrid Mesh with 5
prism layer cells, 7 central patch cells, 5 radial direction cells and 100 streamwise cells. The main
interest is to develop the fully developed velocity profile, but also visualise how this comes to be.
The second experiment in section 4.2 expands on the first Laminar flow by enhancing the Ogrid
Mesh to correct for any apparent imperfections on the velocity profile so it is more representative
of Poiseuilles law than the previous experiment. Different values are explored and plotted into a
Moody Chart to understand how it compares to experiment, but also to determine the pressure
per unit length as discussed in section 2.2.
The third and fourth experiments in sections 4.3 and 4.4 undergo a similar process to the first
two, but this time they focus on the turbulent regime. Velocity profiles will also be explored along
with a Moody chart and pressure drops. But also the extra properties of turbulent kinetic energy
and dissipation rate. One important aspect is editing the mesh accordingly and not unnecessarily
to save computation time. All the experiments consider a vertical pipe with a radius of 0.05m and
length of 5m at T = 293K. The results for all the experiments can be analysed in ParaView and
Origin.

4
4.1

Results
Laminar Flow: Ogrid Mesh

Figure 3: Velocity magnitude profile of a laminar flow through a vertical pipe, 5m in length and
0.05m radius, in the fully developed regime.

Figure 4: Velocity magnitude profile of a laminar flow of a camera normal and closer pipe views
respectively.

Figure 5: Velocity magnitude profile of a laminar flow for time steps of (a)t = 0, (b)t = 1, (c)t = 2,
(d)t = 3, (e)t = 4, (f)t = 5, illustrating the convergence to the fully developed regime.

Figure 6: Velocity magnitude graph along the cross section of the pipe, within the fully developed
regime

4.2

Laminar Flow: Enhanced Ogrid Mesh

Figure 7: Velocity profile in the fully developed region with a camera normal view with corrections
to the radial direction, increasing the number of cells in the radial direction from the previous
experiment from 5 to 30 and reducing the prism layer stretching from 1 to 0.01.

Figure 8: Velocity graph in the fully developed region for the profile in figure 7

Figure 9: Moody chart for an increased number of cells in the radial direction.

Figure 10: Moody chart for an increased number of cells in the central patch, radial and streamwise
directions.

10

Figure 11: Pressure drop per unit length for an improved number of cells in the central patch,
radial and streamwise directions

4.3

Turbulent Flow: Ogrid Mesh

Figure 12: Velocity magnitude for turbulent flow through a vertical pipe of 5m in length, 0.05m
radius, illustrating that further meshing refinement is needed. Using a 0.5 prism layer stretching,
10 prism layer cells, 7 central patch side cells, 5 radial cells and 100 streamwise cells.

11

Figure 13: Turbulent kinetic energy in the fully developed profile, indicating problems with the
mesh at the outer edge.

4.4

Turbulent Flow: Enhanced Ogrid Mesh

Figure 14: Camera normal view of velocity profile for time steps (a)t = 0, (b)t = 1, (c)t = 2,
(d)t = 3, (e)t = 5, (f)t = 10 with 0.01 prism layer stretching, 10 prism layer cells, 7 central patch
side cells, 10 radial cells and 500 streamwise cells.

12

Figure 15: Turbulent kinetic energy profile against distance of the pipe cross section.

Figure 16: Turbulent dissipation rate profile against distance of the pipe cross section.

13

Figure 17: A Moody Chart for the Turbulence Flow under the new mesh covering one radius of
the pipe

Figure 18: Pressure Loss per unit length against the distance from the centre covering one radius
of the pipe
14

5
5.1

Discussion
Laminar Flow: Ogrid Mesh

This experiment focused on laminar flow using the basic settings of the Ogrid at room temperature.
Figure 3 illustrates a pipe-view of the end results with the initial velocity at 0.01ms1 at one end
and the development of the profile distribution, showing the faster velocities at the centre of the
pipe and smaller at the edges of the pipe. A pipe view can also be translated into a camera
normal view which represents a slicing plane and shows how the velocity changes with respect to
the different time steps. This is shown in the first image of figure 4. This more clearly illustrates
how the velocity is distributed along the pipe radius. It shows there is a peak velocity in the central
patch and decreasing radially across the radial part. The streamwise direction has a much smaller
velocity profile, illustrating a key profile of the laminar flow. Therefore the velocity is smaller
nearer the wall, which in a nuclear context is an issue as the speed at the centre does not allow for
homogeneous heating. Overheating can cause black spots and accelerate aging in pipes, increasing
thermal vibrations in giving paths for more dislocations, this can impact on other areas where
the coolant is present such as valves, points of weakness and the heat exchanger. The peak of the
velocity is close to 0.020ms1 at the very centre of the pipe, falling to around 0.016ms1 within the
central patch which represents r/1.5 of the total cylinder cross section. The radial section falls from
0.016ms1 to around 0.01ms1 . This velocity drop is shown mathematically by (4) which shows
this reduction in the velocity profile as it nears the wall of the circular pipe. Figure 4 also shows an
alternative slice view point equivalent to the camera normal with the same velocity distribution,
but also a laminar-like distribution with highest velocity magnitudes at the centre and least at the
side.
The previous figures have shown the final time step of the CFD simulation, hence figure 5
illustrates the development of the full developed regime. At t = 0 the velocity distribution matches
the initial conditions of u = 0.01ms1 . Immediately into the simulation at t = 1 illustrates the
activation of the boundary conditions where the velocity is zero. By the time t = 3 is reached,
the velocity distribution is almost indistinguishable from the fully developed distribution at t = 5.
Importantly, this figure shows the consistency of the laminar flow in the simulation.
The final figure for this experiment is a graph of the velocity magnitude across the cross section
shown by figure 6. This graph shows a parabolic dependence between the velocity and distance,
which agrees with the Poiseuille profile in eq. (4), where there is a distance squared dependence,
r2 with respect to the velocity uz . In this experiment and report u = uz . On a closer inspection
it can be seen that figure 6 is not perfectly parabolic because of the limited number of points. In
particular there is a slight kink between 0.03m and 0.04m which represents the mesh conditions. At
this outer region the velocity profile becomes comparable to a linear representation which signifies
the importance of refining the mesh to explore this anomaly. The next experiment focuses on
this. Therefore a good comparison to the Poiseuille profile has been shown, but with further
improvements to be made to the mesh at the outer edges in particular. This can be additionally
seen within figure 5 where the green region in the radial domain has little differentiation.

5.2

Laminar Flow: Enhanced Ogrid Mesh

Figure 7 illustrates an enhanced model from the previous section in that more cells are given to
the radial part of the Ogrid. The reasoning behind this was that the velocity graph appeared
less smooth in this region, with the aim to make it more parabolic. The figure clearly shows
more distribution in the radial region as compared to the fully developed region in figure 4 and
5. Comparing the velocity graphs, figure 8 has been created illustrating the improved parabolic
15

shape especially at the outer edges at around 0.04m. Improvements can still be made to the central
patch, and although the velocity values vary less here this is important in constructing the Mood
diagram.
Figure 9 shows the Moody chart for the simulation of the enhanced mesh, using eqs (2) and
fD = 64/Re for the laminar regime. This was constructed using half of the values from the centre
to 0.05m, the radius of the pipe. Comparing it to the experimental Moody chart in figure 2 it
shows a log-log based relationship resulting in a straight line. Compared to figure 2, at fd = 0.09
this occurs at Re = 700 on the experimental version and Re = 750 on the simulation. A value
of fd = 0.05 occurs at about Re = 1050 and Re = 1250 on the simulation. One issue with the
simulation is that it does not automatically cut off beyond the laminar regime, so there are values
beyond Re = 2000, but these were omitted. The value of fd = 0.03 occurs at around Re = 1100 for
the experimental and Re = 2100 for the simulation. Therefore, the values do disagree at the higher
values of Reynolds number by a fraction, but do show good agreement at lower values. Figure 10
illustrates another set of simulation results, this time enhancing the cells in the centre as well as
the radial and streamline directions. Compared to the experimental Moody chart in figure 2. The
effects are smaller which was expected, compared to the transition from figure 6 and 7. The reasons
for this is that the velocity in the central part of the Ogrid changes less, given by the shallower
gradient. However, it has achieved a better distribution although the simulation results are not
changed to any appreciable degree due to the limitation of the simulation.
Finally, the pressure drop per unit length of the pipe can be calculated with a mean velocity
of 0.0104ms1 . Figure 11 employs eq. (11) using the calculated d values. The log-log plot shows a
pressure drop from the centre of the pipe to the edge, before the boundary conditions. This drop is
of the order of 3 104 P am1 . Comparing this to the modified pressure values calculated in the
simulation, they are of comparable order, with some fluctuation between 104 and 105 pam1 .

5.3

Turbulent Flow: Ogrid Mesh

The purpose of this experiment was to show the importance of meshing in the outer regions of
the Ogrid as a comparison for an improved meshed. Figure 12 illustrates the issue that arises
in the radial and outer regions where the velocity distribution shows very little gradient change.
Therefore, the sizes of stretching must be limited and the outer cells must be increased to show a
better distribution of velocities. However, the actual magnitudes of the velocities will remain the
same. This issue is also reflected in the turbulent kinetic energy shown by figure 13, where at the
outer edge close to the walls of the pipe the TKE follows a different gradient.

5.4

Turbulent Flow: Enhanced Ogrid Mesh

This experiment enhanced the Ogrid mesh on the outer regions by increasing the number of cells
and therefore computation time in this region, to better define the velocity profile. Figure 14
illustrates the pipe-slice view and camera normal view of all the time steps. It clearly illustrates
the enhancement of the outer regions, showing a more even distribution of velocity. Therefore,
turbulence enables a greater velocity at the walls and therefore heat flow. Turbulence can therefore
minimise heating damages by ensuring the energy is better distributed. In nuclear this turbulent
behaviour is especially important to ensure effective heat exchange and reduce the risk of nucleate
film boiling. Any bubble formation tend to flow erratically which enables a better profile distribution.
The results could be further improved by focusing even more cells in the radial and streamwise
regions, although this comes at a computational expense. The results also agree with the velocity
magnitudes in section 5.3, which is expected as only the distribution is being refined through

16

geometry refinement of the mesh by adding additional cells, rather than the physics itself.
Comparing figure 14 with 7, the peak velocity magnitude is 0.141ms1 for the turbulent case
0.0196ms1 for the laminar case. This difference is due to he fluctuation in velocities for the
turbulence case where eddies are formed and enhance the peak velocity. Comparing the outer
radial regions the turbulent case shows the velocity gradient is slightly steeper closer to the wall,
but less steep at the centre.
Now focusing the attention on the turbulent kinetic energy in figure 15, which shows a minimum
of the TKE at the centre point of the diameter and peaking at the pipe wall. The gradient is
therefore larger between these two regions due to the difference between the production factor of
TKE and dissipation rate. Comparing this to figure 13 it is clear the outer parts of the graph
have been corrected. It could also be argued the mesh in the middle of the pipe could be refined,
however given values differ less here it is not as fulfilling as addressing the issues near the wall and
would be more computationally expensive to achieve. Conversely, the dissipation rate has also been
plotted given by figure 16 which again shows a peak at the edge due to the conditions applied and
differences in turbulent flow relative to the centre.
The final two figures 17 and 18 illustrate information on the pressure differences across the pipe
under a turbulent regime. Figure 17 has been plotted using eq. (2) where the value of the Fluid
kinematic viscosity is known instead. The value of fD was determined from eq. (13), employing
the Lambert W function. Comparing this with the aforementioned Moody chart in figure 2 it
is clear the general shape of the curve is in agreement illustrating some leveling off. The values
of fD are also in good agreement and can be compared with figure 2. Focusing on comparisons
with a smooth pipe given the lambert W function used in this report, the value of fD = 0.04
occurs at around Re = 4 103 on the smooth pipe, although is visible at Re = 1.5 104 on
our experiment. This indicates some differences between the simulation and determined values
of a smooth pipe. fD = 0.02 is shown at Re = 6 104 on figure 2, although in our experiment
this occurs at Re = 5.5 104 , showing good agreement. The final comparison can be seen on
fD = 0.01 which occurs at Re = 2.5 106 on the original Moody chart, and Re = 3.5 105 in
the computational results. Therefore, the values do show some good agreement with experiment,
although there is also room for improvement of the mesh given the gradient does differ from the
smooth pipe at larger values and and obtaining more data at the higher values of fD near the
boundary of laminar and turbulent regimes, alternatively the Reynolds number are smaller when
the velocity is smaller, which is at the edges of the pipe. The graph constructed represents values
from the centre outwards to 0.05m, where a similar set of results is achieved for 0 0.05m.
Finally figure 18 has been added for completeness and demonstrates the pressure per unit length
employing eq. (11). It shows the general pressure loss from the centre outwards up until it meets
the wall-boundary condition. The value of the mean velocity used was 0.0796ms1 , including in
the average any zero velocity 3quantities. The pressure drop from the centre to just before the edge
conditions is around 5 103 P am1 . This value is in agreement with the values obtained for the
modified pressure within the simulation and is therefore slightly larger than that obtained in the
laminar case because the mean speed squared is greater for a turbulent flow than a laminar flow,
even though the values of fD are smaller for the turbulent case. Given the governing equation has
a dependence on fD , this expression could also be improved by further correcting the mesh at the
edge regions.

17

Conclusion

The results have shown that for a laminar case the speed is slower nearer the wall which can be
an issue in achieving homogeneous heating. Whereas, the turbulent flow gives a greater velocity at
the walls and therefore heat flow. The laminar peak velocity was 0.0196ms1 with a mean value
of 0.0104ms1 and pressure drop per unit length from the centre to the edge of 3 104 P am1 .
The turbulent peak velocity was larger at 0.141ms1 with a greater mean flow rate of 0.0796ms1 ,
resulting in a larger pressure drop per unit length of 5 103 P am1 .
Improvements to the mesh were also made, where in the laminar case this resulted in a
more parabolic velocity profile matching the Poiseuilles law and in the turbulent case the profile
became better distributed and corrected TKE issues at the edge, although there is still room for
improvement on achieving results comparable to a smooth turbulent pipe Moody Chart, which
typically showed some disagreements at the lower Reynolds number and the boundary of the model
at fD = 0.01.
Further improvements can be made by adapting the transport properties to use PWR or AGR
values to better simulate nuclear conditions. A gravity field could also be considered to account for
buoyancy, though at the cost of greater computation time. Alternative flows and models could also
be explored such as the Couette flow accounting for circulation due to concentric rotating cylinders.

References
[1] G. Yadigaroglu. Computational fluid dynamics for nuclear applications: from cfd to multi-scale
cmfd. Nuclear Engineering and Design, Vol 235, pp. 153-164, (2005).
[2] R. Lohner. Applied computational fluid dynamics techniques an introduction based on finite
element methods. Wiley, (2008).
[3] O. Reynolds. An experimental investigation of the circumstances which determine whether
the motion of water shall be direct or sinuous, and of the law of resistance in parallel channel.
Philosophical Transactions of the Royal Society, Vol. 174, pp. 935982, (1883).
[4] G. Stokes. On the effect of the internal friction of fluids on the motion of pendulums.
Transactions of the Cambridge Philosophical Society. Vol. 9, pp. 8106, (1851).
[5] S.B. Pope. Turbulent flows. Cambridge University Press, (2003).
[6] Y. Povstenko. Linear Fractional Diffusion-Wave Equation for Scientists and Engineers.
Birkhuser, (2015).
[7] J. Ahn F. Sebilleau and M.J. Bluck. Cfd analysis of the flow in a heated vertical pipe using
openfoam. Imperial College, (2016).
[8] M.S. Shephard R.V. Garimella. Boundary layer mesh generation for viscous flow simulations.
Cambridge University Press, Vol. 538, pp. 429-443, (2005).
[9] B.A.P Reif P.A. Durbin. Statistical theory and modelling for turbulent flows. Wiley, (2010).
[10] J. Boussinesq. Theorie de lecooulement tourbillant. Acad. Science Institute France, Vol 23,
pp. 46-50, (1877).
[11] Bansal. A text book of fluid mechanics and hydraulic machines. Laxmi Publications, (2005).
18

[12] L.F. Moody. Moody-friction factors for pipe flow. Transactions of the ASME, Vol 66, (1944).
[13] Vectortech Engineering. Moody charts.
http://www.vectortechengineering.co.uk/wp-content/uploads/2015/06/MoodyChart.
jpg, (Accessed November 2016).
[14] A.J. Smits B.J. McKeon, M.V. Zagarola. A new friction factor relationship for fully developed
pipe flow. Cambridge University Press, Vol. 538, pp. 429-443, (2005).
[15] J.A Robertson C.T. Crowe, E.F Donald. Engineering fluid mechanics. Wiley, (2005).
[16] OpenFOAM. Openfoam guide - v1606+ of the open source field operation and manipulation
(openfoam). Digital Version, OpenFOAM, (2016).

19

Das könnte Ihnen auch gefallen