Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Executive Summary
This assignment details my investigations into basic laminar flow and three types of turbulent flow.
A mesh study was performed to identify suitable meshes for Computational Fluid Dynamic (CFD)
simulations.
Theoretical and empirical data from a number of sources have been used to calculate expected
velocity profiles and pressure gradients for these flows. This theoretical data has then been
compared to simulation data and evaluated.
Contents
Introduction ...................................................................................................................................................................... 5
1. Laminar Flow ............................................................................................................................................................. 6
1.1. Initial Work ........................................................................................................................................................ 6
1.2. Parabolic Input Velocity Distribution .............................................................................................................. 10
1.2.1. Derivation of Hagen-Poiseuille Flow from Navier Stokes Equations ...................................................... 11
1.3. Parabolic Inlet Velocity Simulation & Laminar Mesh Study ............................................................................ 12
1.4. Methods for Increasing Efficiency when Simulating Laminar Flow ................................................................ 15
2. Turbulent Flow ........................................................................................................................................................ 16
2.1. Introduction to Turbulent Flow ...................................................................................................................... 16
2.2. Calculation of Turbulent Flow ......................................................................................................................... 17
2.3. Computation Methods .................................................................................................................................... 19
2.3.1. K-Epsilon.................................................................................................................................................. 19
2.3.2. Shear Stress Transport (SST) ................................................................................................................... 19
2.3.3. Baseline (BSL) and Speziale, Sarkar and Gatski (SSG) Reynolds Stress ................................................... 19
2.4. Smooth Pipe Flow ........................................................................................................................................... 19
2.5. Transitional Region Rough Pipe Flow.............................................................................................................. 26
2.6. Fully Turbulent Rough Pipe Flow .................................................................................................................... 30
3. Alternative Methods ............................................................................................................................................... 36
4. Recommendations for Further Work ...................................................................................................................... 37
5. Conclusion ............................................................................................................................................................... 37
5.1. Laminar Flow ................................................................................................................................................... 37
5.2. Turbulent Flow ................................................................................................................................................ 37
5.3. Final Remarks .................................................................................................................................................. 37
References ...................................................................................................................................................................... 39
Bibliography .................................................................................................................................................................... 40
Appendix ......................................................................................................................................................................... 41
Figures
Figure 1 - Golden Syrup Flowing From a Teaspoon ....................................................................................................................................... 6
Figure 2 - Osborne Reynolds (1842 –1912) (Robodoc, 2004) ...................................................................................................................... 6
Figure 3 - Initial Run Mesh, Looking End on – Created using ICEM CFD ........................................................................................................ 7
Figure 4 - Developing Flow Velocity .............................................................................................................................................................. 8
Figure 5 - Initial Run - Developing Velocity Profile ........................................................................................................................................ 9
Figure 6 - Simulated and Theoretical Fully Developed Laminar Flow Velocity Profiles with Uniform Inlet Velocity .................................... 9
Figure 7 - Initial Run – Theoretical and Simulated Axial Pressure Variation Comparison ........................................................................... 10
Figure 8 - Left to Right: Gotthilf Hagen (1797 - 1884) & Jean Poisuille (1797 - 1869) ................................................................................. 10
Figure 9 - Everyday Example of Turbulent Flow .......................................................................................................................................... 16
Figure 10 - Typical Velocity and Shear Distributions in Turbulent Flow near a Wall: (a) Shear; (b) Velocity. ............................................. 17
Figure 11 - Left to Right: Ludwig Prandtl, Theodore Von Kármán & Clark Blanchard Millikan ................................................................... 18
Figure 12 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 1 ..................................................................................... 20
Figure 13 - Axial Pressure Variation for Smooth Pipe Turbulent Flow – Pipe 1........................................................................................... 21
Figure 14 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 2 ..................................................................................... 22
Figure 15 - CFX Solver showing Convergence for Pipe 2 Smooth Turbulent Flow....................................................................................... 22
Figure 16 - Axial Pressure Variation for Smooth Pipe Turbulent Flow - Pipe 2 ........................................................................................... 23
Figure 17 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 3 ..................................................................................... 24
Figure 18 - Axial Pressure Variation for Smooth Pipe Turbulent Flow - Pipe 3 ........................................................................................... 24
Figure 19 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 4 ..................................................................................... 25
Figure 20 - Axial Pressure Variation for Smooth Pipe Turbulent Flow - Pipe 4 ........................................................................................... 26
Figure 21 - Velocity Profile Comparison for Transitional Rough Flow - Pipe 1 ............................................................................................ 28
Figure 22 - Axial Pressure Variation for Transitional Rough Flow - Pipe 1 .................................................................................................. 28
Figure 23 - Velocity Profile Comparison for Transitional Rough Flow - Pipe 2 ............................................................................................ 29
Figure 24 - Axial Pressure Variation for Transitional Rough Flow - Pipe 2 .................................................................................................. 30
Figure 25 - Velocity Profile Comparison for Fully Turbulent Rough Flow - Pipe 1 ....................................................................................... 32
Figure 26 – Close up of Centreline Velocity on Figure 25 ............................................................................................................................ 32
Figure 27 - Axial Pressure Variation for Fully Turbulent Rough Flow - Pipe 1 ............................................................................................. 33
Figure 28- Velocity Profile Comparison for Fully Turbulent Rough Flow - Pipe 2........................................................................................ 33
Figure 29 - Close Up of Centreline Velocity on Figure 28 ............................................................................................................................ 34
Figure 30 - Axial Pressure Variation for Fully Turbulent Rough Flow - Pipe 2 ............................................................................................. 34
Figure 31 - Fully Turbulent Rough Flow (From Pipe 2 - K-Epsilon) .............................................................................................................. 35
Figure 32 - Centreline Velocity along Pipe (Data from Pipe 2 - Shear Stress Transport) ............................................................................. 35
Figure 33 - Centreline Velocity along Pipe (Data from Pipe 1 - Shear Stress Transport) ............................................................................. 36
Tables
Table 1 - CFX-Pre Parameters for First Run ................................................................................................................................................... 8
Table 2 - Results of Laminar Mesh Study (Series 1)..................................................................................................................................... 13
Table 3 - Results of Laminar Mesh Study (Series 2)..................................................................................................................................... 14
Table 4 - Mesh Settings for Smooth Pipe Turbulent Flow ........................................................................................................................... 20
Table 5 - Pipe 1 - Turbulence - Smooth Pipe Flow ....................................................................................................................................... 20
Table 6 - Pipe 2 - Turbulence - Smooth Pipe Flow ....................................................................................................................................... 21
Table 7 - Pipe 3 - Turbulence - Smooth Pipe Flow ....................................................................................................................................... 23
Table 8 - Pipe 4 - Turbulence - Smooth Pipe Flow ....................................................................................................................................... 25
Table 9 - Mesh Setting for Transient and Fully Turbulent Rough Flow ....................................................................................................... 27
Table 10 - Transient Rough Flow - Pipe 1 .................................................................................................................................................... 27
Table 11 - Transient Rough Flow - Pipe 2 .................................................................................................................................................... 29
Table 12 - Fully Turbulent Rough Flow - Pipe 1 ........................................................................................................................................... 31
Table 13- Fully Turbulent Rough Flow - Pipe 2 ............................................................................................................................................ 33
Introduction
This report describes the research, theoretical and computational methods, results and evaluation of work set out in
the Thermofluid Systems assignment brief.
This report has been written by Steven Goddard (Student Number 10038749) as part of the Mechanical Engineering
(Part Time) course at UWE, Bristol during the 2013/2014 student year.
Understanding fluid flow in pipes is of great importance to many industrial applications today, such as chemical
processing, long distance fuel transfer and precise scientific experiments. The fluid dynamics work that helps us
understand these processes also attempts to predict other phenomenon which we often take for granted such as
our planets complex weather systems.
Using Computational Fluid Dynamics (CFD) software, ICEM CFD and ANSYS CFX (ANSYS CFX, Release 15.0), I have
meshed various sized horizontal pipes. Starting with basic laminar flow I have compared the well understood theory
with CFD simulation. Further into my investigation I have constructed a number of coarse and fine meshes to help
me understand the properties of a “good” mesh.
Upon deciding on two suitable meshes I have simulated turbulent flow with three main scenarios:
As the theory of turbulent flow is not completely understood I have been using various approximations and empirical
data from a variety of sources to obtain my calculations. I have used four turbulent simulation modelling techniques
to demonstrate turbulent flow. During my turbulent simulations I have continued to refine my mesh until I found
two that worked well.
Following my simulations I have compared the velocity profiles and pressure variations to my theoretical results, I
have identified problems that have occurred, anomalies discovered and evaluated the data I have produced.
1. Laminar Flow
Laminar flow (also known as streamline flow) occurs at low Reynolds numbers (approximately lower than 2040). This
is due to the dominant viscous forces and is characterised by a smooth, constant fluid motion. Laminar flow in a pipe
can be considered as the relative motion of concentric cylinders of fluid fixed at the pipe wall (Velocity = 0) and the
other cylinders moving at increasing speed toward the centre of the pipe (Velocity = 𝑉𝑚𝑎𝑥 ).
T o simply demonstrate a common example of laminar flow I’ve used some golden syrup flowing off of a teaspoon.
My first investigations were based on a laminar flow in a smooth and straight pipe.
1.1.Initial Work
To begin my investigation I started by researching Reynolds number, as this dictates the type of flow.
Reynolds Number (𝑅𝑒), (introduced by George Gabriel Stokes in 1851) was popularized by Osborne Reynolds in
1883. It is the ratio of inertial forces to viscous forces and can be used to characterise different flow regimes
such as Laminar, Transitional and Turbulent flow.
For this first run I chose to use a uniform velocity profile, in order to create an appropriate CFD model I required
to understand the entrance length. This describes the theoretical length at which the flow will become fully
developed and is used after fluid has entered a pipe or as it passes through components that disrupt the flow
such as bends, pumps or turbines.
Steven Goddard –Mechanical Engineering (Part Time)
Student Number: 10038749 CFD Assignment Page 7 of 41
To begin, an entrance length number must be determined, for laminar flow this can be expressed as:
𝐸𝑙𝑙𝑎𝑚𝑖𝑛𝑎𝑟 = 0.06𝑅𝑒
For my first run I chose to use a Reynolds number of 1000 as this is well within the specification for laminar
flow. This results in an Entrance Length Number of 60.
𝑙𝑒 = 𝐸𝑙 ∙ 𝑑
I chose a diameter of 15mm which was based on a standard radiator pipe (DIYData.com, 2000). This gave me an
Entrance Length of 0.9m. As this distance was the point at which the flow had fully developed I have increased
this by 25% to allow the CFD model to show the flow once it had developed, therefore my overall length was
1.125m.
I began by creating a mesh using ICEM CFD, this mesh consisted of a central square section with 10 divisions
along the vertical and horizontal edges. Due to the importance of calculations at the pipe wall needing more
accuracy I split the edge from the central square to pipe wall into 20 divisions with ratio settings as follows:
Inlet velocity was calculated using the equation for Reynolds Number, rearranged for velocity:
𝜌𝑉𝐷 𝑅𝑒𝜇
𝑅𝑒 = → 𝑉=
𝜇 𝜌𝐷
I used water as my fluid and based on the materials properties used in ANSYS CFD the inlet velocity for this run
will be 0.05951 m/s.
Figure 3 - Initial Run Mesh, Looking End on – Created using ICEM CFD
0.015m Diameter Pipe, 10 Central Divisions, 20 Radial Divisions
Steven Goddard –Mechanical Engineering (Part Time)
Student Number: 10038749 CFD Assignment Page 8 of 41
The simulation was defined in CFX-Pre with the following parameters, all other parameters remained default.
The solver ran for 22 iterations until the residual target of 1 × 10−4 was reached. This took 4 minutes and 47
seconds. The results were then loaded into CFX-Post for analysis.
Figure 4 shows a geometrical plane that has been produced along the ZX plane and set to show velocity.
As expected on the left hand side of the pipe (inlet), the flow is at a uniform value across the pipe diameter as I
set previously. For this uniform inlet flow, the velocity is actually the mean flow velocity or half the max velocity
once fully developed.
Towards the outlet the flow starts to slow down at the walls and speed up in the centre until eventually the flow
is fully developed and is represented in the form of a parabolic profile.
To investigate further I have created two charts (Figure 5 & Figure 7) to show how the velocity profile and
pressure changes axially along the pipe.
For the velocity in Figure 5 each data series is taken in 1/10th intervals along the pipe; as shown the profiles
slowly develop into a parabola. From 0.9m onwards (represented by the red, blue and green dotted lines) the
flow has fully developed and as expected the velocity profile undergoes negligible change.
Figure 6 shows a comparison between the simulation and the theoretical results. The two profiles are quite
similar from the wall until around 0.00375m. The error between the results reaches a maximum at the centre of
the pipe of 5.5%.
Figure 6 - Simulated and Theoretical Fully Developed Laminar Flow Velocity Profiles with Uniform Inlet Velocity
In Figure 7 pressure is measured along the centreline. (Note that due to how the mesh was extruded in ICEM
CFD and how the boundaries were set in CFX-Pre, 0.0m is the outlet of the pipe). The development of the flow in
the simulation results is shown by the decreasing gradient of pressure change most easily seen at the beginning
of the pipe; theoretically the pressure drop is shown to be linear. The simulation results eventually settle to a
linear decline after 0.90m (0.225m on the graph) when the flow has fully developed.
Figure 7 - Initial Run – Theoretical and Simulated Axial Pressure Variation Comparison
Fortunately, in the late 1830’s two men, Gotthilf Heinrich Ludwig Hagen from Königsberg, East Prussia (modern
day Russia) and Jean Léonard Marie Poiseuille from Paris, France derived an equation which allows the pressure
drop in a fluid flowing through a pipe to be calculated. The equation can also be arranged to give a full solution
for the parabolic laminar flow profile.
Figure 8 - Left to Right: Gotthilf Hagen (1797 - 1884) & Jean Poisuille (1797 - 1869)
(Centralblatt der Bauverwaltung, 1899 & Artem Korzhimanov, 2010)
Before the Hagen-Poiseuille Equation can be derived several assumptions need to be made:
r direction:
Simplifies to:
𝛿𝑝
=0
𝛿𝑟
𝜃 direction:
z direction:
𝛿𝑤 𝛿𝑤 𝑉𝜃 𝛿𝑤 𝛿𝑤 𝛿𝑝 1 𝛿 𝛿𝑤 1 𝛿 2𝑤 𝛿 2𝑤
𝜌( + 𝑉𝑟 + +𝑤 ) = − + 𝜌𝑔𝑧 + 𝜇 [ (𝑟 ) + 2 2 + 2 ]
𝛿𝑡 𝛿𝑟 𝑟 𝛿𝜃 𝛿𝑧 𝛿𝑧 𝑟 𝛿𝑟 𝛿𝑟 𝑟 𝛿𝜃 𝛿𝑧
Simplifies to:
𝛿𝑝 1𝛿 𝛿𝑤 1𝛿 𝛿𝑤 1 𝛿𝑝
0=− +𝜇[ (𝑟 )] → (𝑟 ) =
𝛿𝑧 𝑟 𝛿𝑟 𝛿𝑟 𝑟 𝛿𝑟 𝛿𝑟 𝜇 𝛿𝑧
1 𝑑𝑝 2
𝑤𝑧 = 𝑟 + 𝐶1 ln 𝑟 + 𝐶2
4𝜇 𝑑𝑧
In order for 𝑤𝑧 to be a finite number (𝑤𝑚𝑎𝑥 ), when 𝑟 = 0, to avoid the ln term 𝐶1 needs to equal 0.
Also for a no slip boundary condition at the pipe wall this means that: 𝑤𝑧 = 0 when 𝑟 = 𝑅
Therefore:
1 𝑑𝑝 2
𝐶2 = 𝑅
4𝜇 𝑑𝑧
Steven Goddard –Mechanical Engineering (Part Time)
Student Number: 10038749 CFD Assignment Page 12 of 41
Thus the parabolic velocity profile which I will be using for my subsequent laminar flow simulations can be
expressed as:
1 𝑑𝑝 2
𝑤𝑧 = − (𝑅 − 𝑟 2 )
4𝜇 𝑑𝑧
Max Velocity can be found by applying the appropriate boundary conditions (r=0).
𝑅2 𝑑𝑝
𝑤𝑧𝑚𝑎𝑥 = (− )
4𝜇 𝑑𝑧
As shown previously:
𝑤𝑧𝑎𝑣𝑔 = 0.5𝑤𝑧𝑚𝑎𝑥
𝑑𝑝 Δ𝑝
− =
𝑑𝑧 𝐿
𝑅 2 Δp 𝑅 2 Δ𝑝 𝐷 2 Δ𝑝
𝑤𝑧𝑎𝑣𝑔 = 0.5 ( )= =
4𝜇 𝐿 8𝜇 𝐿 32𝜇 𝐿
Rearranging this for Δ𝑝 gives the Hagen Poiseuille equation (also shown are 2 other common
representations that use volume flow rate rather than average velocity):
To begin with I did some initial research into what makes a good mesh (Baker, 2002). I found the following
points to be the most important aspects to creating a mesh:
• The density of the mesh should be high enough to capture all relevant flow features.
• Mesh adjacent to the wall should be fine enough to resolve the boundary layer flow.
• A uniform mesh will not allow detailed computation at the pipe wall where it is most needed.
• More cells/nodes will give higher accuracy but will increase CPU time.
• Avoid high aspect ratios.
• Errors occur when the mesh:
o Is too coarse
o Has high skewness.
o Has large jumps in volume between adjacent cells.
o Inappropriate boundary layer mesh.
I have used a non-uniform grid which is bunched to give a finer mesh at the pipe wall. I have used the following
bunching settings to start with:
• Mesh Law: BiGeometric • Spacing 2: 0
• Spacing 1: 1.00E-06 • Ratio 2: 2
• Ratio 1: 1.2 • Max Space: 0
During set up in CFX-Pre I have simplified the Parabolic Input Velocity Profile equation further by removing the
differential pressure term as follows, start with:
1 𝑑𝑝 2
𝑤𝑧 = − (𝑅 − 𝑟 2 )
4𝜇 𝑑𝑧
Apply the boundary condition 𝑟 = 0 𝑤ℎ𝑒𝑛 𝑤𝑧 = 𝑤𝑧𝑚𝑎𝑥 which gives:
1 𝑑𝑝 2 1 𝑑𝑝 𝑤𝑧𝑚𝑎𝑥
𝑤𝑧𝑚𝑎𝑥 = 𝑅 → =
4𝜇 𝑑𝑧 4𝜇 𝑑𝑧 𝑅2
This expression can now be substituted back into the original to give:
𝑤𝑧 2 2)
𝑟2
𝑤𝑧 = 𝑚𝑎𝑥 (𝑟 − 𝑅 = 𝑤 𝑧 (1 − )
𝑅2 𝑚𝑎𝑥
𝑅2
The majority of simulations were run on an Intel Core i7 - 3770 CPU with 16GB RAM unless stated with 100 max
iterations and 10^-4 convergence criteria.
The simulations were based on a mesh with 10 to 50 divisions and 100, 200 and 300 layers on a 1.2m pipe.
Readings for the centreline (max) velocity were taken past the point of developing flow at 0.9m, 1.0m, 1.1m and
1.2m and then averaged and compared to the theoretical max velocity to get an error percentage, I chose to
compare max velocity as this is where the maximum error occurred on my initial run.
I’ve also recorded the pressure at the start of the pipe and compared this with the theoretical value expected.
Aspect Ratio
(Central mesh)
Iterations
Divisions
Layers
My observation of this series of results is that they converge with minimal iterations and subsequently
computation time is reduced. Accuracy appears to be higher during the runs with 100 layers, as layer count
increased I experienced errors with the smaller division runs, leading to large error (76.336% to 98.81%).
In conclusion the 100 layer mesh with 20 or 30 divisions seem to yield the best results with a satisfactory
runtime.
Next I began another series of runs with the following bunching settings; this produced an even finer mesh
towards the pipe wall:
These simulations were run with the same divisions and layers, on the same computer.
Aspect Ratio
(Central mesh)
Iterations
Divisions
Layers
From this series I obtained some very accurate results with error percentages failing to exceed 0.2% error.
Similarly to the first series of runs, larger errors started occurring during the high layer/low division runs.
The problems with this series of runs was that most of the simulations didn’t converge to 10^-4 and therefore
reached the maximum iteration count before stopping.
Going forward to my turbulent runs I have decided to use the following meshes:
• A 2D planar mesh would give the same results as the flow does not change radially.
• As the flow is parabolic and axisymetric, the results are essentially mirrored so you could simplify
further by halving the 2D plane.
These methods would greatly reduce the number of nodes in the mesh and decrease computational time
significantly allowing a finer mesh.
2. Turbulent Flow
2.1.Introduction to Turbulent Flow
Unlike Laminar flow where the kinetic energy dies out due to the action of fluid viscosity, Turbulent flow
exhibits chaotic property changes which include low and high momentum diffusion along with quick variations
of pressure and velocity. Unsteady vortices appear within the flow and can last various lengths of time before
they disperse. Typically (but not always) turbulence occurs in flows with Reynolds numbers greater than 4000.
I have demonstrated in Figure 9 a type of turbulence that you may experience every day, a mug of black coffee
with milk being poured into it. When the milk hits the coffee vortices appear as the water is swirling turbulently
in the cup, the initial vortices disperse giving rise to others as the milk mixes in the cup.
Turbulence is also a very active field and is currently described by some scientists as one of the most important
unsolved problems in physics. During my research into turbulent flow I discovered that the 6th Clay Mathematics
Institute Millennium Prize Problem is that of Navier Stokes Existence and Smoothness, understanding this could
be a first step in understanding Turbulence. Recently on the 10th January 2014 the problem was claimed to have
been solved by Kazakh mathematician Mukhtarbay Otelbaev and his paper is currently under review.
(BNews.kz, 2014)
At this moment in time however, turbulence cannot be solved as easy as laminar flow and the use of empirical
data is a requirement.
Figure 10 - Typical Velocity and Shear Distributions in Turbulent Flow near a Wall: (a) Shear; (b) Velocity.
(White F. M. 2003)
To elaborate further on the layers of turbulent flow, in 1930 the German born scientist Ludwig Prandtl (1875 –
1953) deduced that velocity (𝑢) at the wall must be independent of shear layer thickness.
Using dimensional analysis this can be shown by the law of the wall equation:
𝑢 𝑦𝑢∗ 𝜏𝑤
𝑢+ = = 𝐹 ( ) 𝑤ℎ𝑒𝑟𝑒 𝑢∗ = √
𝑢∗ 𝜐 𝜌
A few years later in 1933, Theodore Von Kármán (1881 – 1963) also determined that velocity in the outer layer
was not influenced by the molecular viscosity, although its deviation from the stream velocity 𝑈 needs to
depend on the boundary layer thickness and the other properties.
Using dimensional analysis this can again be simplified to the Velocity-defect law for the outer layer:
𝑈−𝑢 𝑦
= 𝐺 ( )
𝑢∗ 𝛿
The two laws described above are accurate for a variety flows however they must overlap smoothly in the
intermediate layer and in 1937 Clark Blanchard Millikan (1903 – 1966) shows that this is true by using the
logarithmic-overlap layer equation:
𝑢 1 𝑦𝑢∗
= ln +𝐵
𝑢∗ 𝑘 𝜈
The values for k and B change very slightly depending on what literature one reads but the majority seem to take
k = 0.41 and B = 5.0 which is what I will be using in my investigations. (White, F. M. 2003)
I will be using the logarithmic-overlap equation as the inlet velocity profile for my smooth pipe turbulent runs as
this is a good representation of the developed flow I will be comparing.
Figure 11 - Left to Right: Ludwig Prandtl, Theodore Von Kármán & Clark Blanchard Millikan
(Stahlkocher, 1883 & SAB,2012 & Sechler, 1969)
However before I can use the logarithmic overlap equation in my CFD model I need to compute U* which
involves a few steps, starting back at the friction factor and pressure drop.
There are several methods for obtaining the friction factor; one is by simply looking at the Moody chart which is
accurate to ±15%, for my problem this gives a value of 0.0355. The others involve formulae such as the
generally accepted form for a smooth pipe which I have solved for f using numerical software (Wolfram Alpha,
2009):
1 (log(2) + log(5))2
= 2 log(𝑅𝑒𝑑 𝑓 0.5 ) − 0.8 → 𝑓 = 2 ≈ 0.0355106
𝑓 0.5 3 3
4𝑊 (300 (10 ⁄5 log(2) + 10 ⁄5 log(5)))
𝑓 = 0.316𝑅𝑒𝑑−0.25 = 0.0359
There is little error between the methods shown and I’ve chosen to take 0.0355 as my friction factor.
The next parameter needed is the pressure drop and I tried two equations to get this, the first is the Poiseuille
version used previously in laminar flow and the alternative form is from Blasius. The results are very similar and
I have used the Blasius value in my calculations:
𝑓𝐿
Δ𝑝 = 0.5𝜌𝑉 2 = 45.11652 𝑃𝑎
𝐷
Δ𝑝 ≈ 0.241𝐿𝜌0.75 𝜇 0.25 𝑑 −4.75 (0.25𝜋𝑑 2 𝑉) = 45.60619 𝑃𝑎
Where V is the average velocity.
Next up is the Shear stress at the wall which is calculated as follows:
𝐷Δ𝑝
𝜏𝑤 = = 0.57008 𝑃𝑎
4𝐿
And finally, U* can be obtained:
𝜏𝑤
𝑈∗ = √ = 0.02391 𝑚/𝑠
𝜌
2.3.Computation Methods
Throughout the turbulence modelling I have been using 4 different turbulence modelling options.
From my research (Karthik, T. S. D., 2011) it seems that each option is best suited to particular flows or scenarios
such as an aircraft wing, flow down a pipe or flow along a flat plate etc...
Whilst setting up the simulation I came across a problem, the log term in the logarithmic overlap equation would
get to a point where it would be trying to evaluate terms ≤ 0. This returned an error as you would expect.
In order to make this work I have changed R from 0.0075 to 0.0075000001 so that it does not try and evaluate at
0 and I have also added in the √𝑅𝑎𝑑12 term to the expression below to ensure that as the calculation is
performed beyond the centre of the pipe the expression the log term is trying to evaluate will stay positive.
Therefore the inlet velocity expression for smooth pipe flow is as follows:
Ux*((1/0.41)*ln(((0.0075000001[m]-(sqrt(Rad1^2)))*Ux)/(8.926*10^-7)[m^2 s^-1])+5)
2.3.1. K-Epsilon
The k-Epsilon model is a “Two Equation” model as solves for variables: k; the turbulent kinetic energy
and 𝜀; the rate of dissipation of kinetic energy. Advantages of this model are that is it relatively simple to
implement, the calculations are stable and converge rate is good with relatively low memory requirements
making it attractive for industrial applications. The disadvantages are that it is not suited for swirling or
rotating flows, adverse pressure gradients or jet flow.
2.3.3. Baseline (BSL) and Speziale, Sarkar and Gatski (SSG) Reynolds Stress
BSL is an omega based model, this model is preferred over the standard omega Reynolds Stress model (not
used in this comparison).
SSG is an epsilon based model, theoretically this should be better than k-epsilon but this may not be the
case in practice due to deficiencies inherited from the underlying k-epsilon equations.
The settings I have used for each run have been standard apart from the inlet velocity profile used, water as my
fluid and medium intensity as my turbulence option.
In my analysis the primary comparison is the velocity profiles produced by each type of run and the centreline
velocity. My secondary comparison is the axial pressure variation along the pipe.
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Changed Outlet to Changed Outlet to Good Run
"Opening" "Opening".
Unsatisfactory run. Unsatisfactory run.
Error @ Centreline 0.0435605 0.0323985 0.0533235 0.027124
Error @ Centreline -10.14% -7.05% -13.00% -5.65%
Run Time 00:01:44 00:09:40 00:13:45 00:03:54
Iterations 18 100 100 27
Efficiency 36.19% 4.83% 5.58% 10.02%
Table 5 - Pipe 1 - Turbulence - Smooth Pipe Flow
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.0075 -0.005 -0.0025 -4E-18 0.0025 0.005 0.0075
Pipe Radius (m)
Figure 12 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 1
Steven Goddard –Mechanical Engineering (Part Time)
Student Number: 10038749 CFD Assignment Page 21 of 41
25
20
15
10
5
0
-5 0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Figure 13 - Axial Pressure Variation for Smooth Pipe Turbulent Flow – Pipe 1
Pipe 1 illustrates good shaped runs for both K-Epsilon and SSG Reynolds however the centreline velocities for
these runs are out by 10.14% and 5.56% respectively this is backed up by the low starting pressure shown in
Figure 13. Shear Stress Transport and BSL Reynolds Stress runs resulted in errors that have clearly had an
adverse effect on my results for velocity as they show as having a higher velocity bias to the positive radius side
(right) side of Figure 12. These runs also resulted in a relatively large starting pressure that was clearly incorrect.
Due to this large error I have removed the runs from Figure 13 to improve resolution on the runs that completed
more successfully, I feel this is justified based on the strange shaped velocity profiles shown in Figure 12.
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.0075 -0.005 -0.0025 -9E-18 0.0025 0.005 0.0075
Pipe Radius (m)
Figure 14 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 2
Figure 15 - CFX Solver showing Convergence for Pipe 2 Smooth Turbulent Flow
40
Pressure (Pa)
30
20
10
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Figure 16 - Axial Pressure Variation for Smooth Pipe Turbulent Flow - Pipe 2
With Pipe 2 I ran into numerous errors with convergence and run time. Figure 15 shows the unusual way that
RMS P-Mass fails to converge and instead seems to rise and plateau at around 1e-02 until I manually ended the
run at 112 iterations. On a typical run I would expect to see RMS V-Mom (Blue) and RMS P-Mass (Red) to
converge fairly quickly whilst RMS W-Mom (Yellow) tends to lag behind and converge last. On both of these runs I
had similar errors with plateaued convergence or “opening errors”, which is when the CDF software recommends
changing the Outlet boundary condition to an Opening.
Nevertheless when post processing I found that the results were quite accurate (9.76% and 3.99% error) when
compared to theory. The starting pressures again correlate with the velocity error, K-Epsilon has a low starting
pressure and also exhibits a low centreline velocity compared to theory, and the velocity profile for K-Epsilon also
shows two spikes as the boundary layer velocity moves towards the outer layer. This looks like evidence of K-
Epsilons unreliability when calculating velocity at the wall, which is a characteristic mentioned in various
references to this computation method. (Karthik T. S. D., 2011)
Shear Stress Transport starts high, matching the over-predicted centreline velocity. An interesting observation for
this run is that after around 0.03m along the pipe the axial pressure variation is very accurate when compared to
the theory.
Due to the high run-times and lack of convergence on the first two runs I decided at this point to re-evaluate my
meshes, this is when I focused on the aspect ratio and a change of bunching when creating the following two
meshes, Pipe 3 and 4.
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Good Run Good Run Good Run
Error @ Centreline 0.049799 0.021647 0.0224155 0.040945
Error @ Centreline -12.64% 5.18% 5.34% -10.07%
Run Time 00:03:44 00:12:00 00:18:34 00:08:18
Iterations 15 42 52 22
Efficiency 19.21% 2.60% 1.74% 7.10%
Table 7 - Pipe 3 - Turbulence - Smooth Pipe Flow
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.0075 -0.005 -0.0025 -9E-18 0.0025 0.005 0.0075
Pipe Radius (m)
Figure 17 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 3
30
25
20
15
10
5
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Figure 18 - Axial Pressure Variation for Smooth Pipe Turbulent Flow - Pipe 3
The Pipe 3 mesh has an aspect ratio of 4 with 10e-05 bunching; overall the simulations were a success with no
errors on either computation method, errors from theory ranged from 5.34% to 12.64% which I feel is an
acceptable margin. Starting axial pressure once again seems to coincide with the over and under-predicted
centreline velocities. When comparing the accuracy with run-time the K-Epsilon came out as the best at 19.21%
compared to the second best SSG Reynolds Stress at 7.10%, however k-epsilon also gave an acceptable but large
error of 12.64%. The most accurate result was given by the Shear Stress Transport at 5.18%, but this was
hindered by a run-time of 12 minutes, giving an overall efficiency of 2.6%. K-Epsilon and SSG Reynolds stress
exhibits a velocity profile shape which I would have expected with steep gradients towards the wall and a near
flat middle, Shear Stress Transport and BSL Reynolds Stress both have smooth curved velocity profiles.
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Good Run Good Run Good Run
Error @ Centreline 0.057928 0.0222545 0.031961 0.041245
Error @ Centreline -15.10% 5.31% 7.30% -10.16%
Run Time 00:04:28 00:16:14 00:11:48 00:11:50
Iterations 13 46 24 22
Efficiency 18.68% 1.97% 3.90% 5.02%
Table 8 - Pipe 4 - Turbulence - Smooth Pipe Flow
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.0075 -0.005 -0.0025 -1.1E-17 0.0025 0.005 0.0075
Pipe Radius (m)
Figure 19 - Velocity Profile Comparison for Smooth Pipe Turbulent Flow - Pipe 4
Pipe 4 had an extended number of layers (400) compared to previous runs. This resulted in the lowest aspect
ratio of all my meshes. I also decreased the bunching spacing back down to 10e-06. Despite the finer mesh and
lower aspect ratio, surprisingly the results were slightly less accurate and efficient than Pipe 3.
Again, pressure and centreline velocities correlate with each other. K-Epsilon and SSG Reynolds Stress produced
very similar results to Pipe 3; this is good evidence of repeatability and improves my confidence in these results.
Shear Stress Transport exhibits the curved smooth velocity profile as in Pipe 3 but BSL Reynolds Stress shows a
flat middle and steep gradients towards the wall as I have come to expect.
30
25
20
15
10
5
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Figure 20 - Axial Pressure Variation for Smooth Pipe Turbulent Flow - Pipe 4
The transition region is in-between laminar and turbulence (with a Reynolds Number from 2300 to 4000),
laminar flow seems to occur toward the pipe wall and turbulent flow in the centre, during the transition region
Reynolds Number still has a moderate effect on the friction factor, however the friction factor can be subject to
large uncertainties in this flow type. For this set of runs I have lowered the Reynolds number to 2500 to ensure
a transitional flow, this has had a small effect on fully developed length and velocity.
As I based my pipe diameter on a radiator pipe it seemed a suitable choice to use copper or brass as the
material and hence an absolute roughness value of 𝑒 = 0.0015 𝑚𝑚. However, I’ve realised when I come to do
subsequent runs in the fully turbulent region the corresponding relative pipe roughness line on the moody
diagram points to an area near to the far right of the graph and only just above the complete turbulence line,
this also required a Reynolds number in the magnitude of 108 . Therefore I have decided to compromise and use
a rougher pipe based on the value for commercial steel, 𝑒 = 0.046 mm. (White, F. M., 2003)
In order to calculate the friction factor I used the Haaland equation which is an approximation of the Colebrook
Equation. (Darcy Friction Factor, 2014)
1.11 −1 2
𝜀 ⁄𝐷 6.9
𝑓 ≈ ((−1.8 log10 [( ) + ]) ) = 0.04296
3.7 𝑅𝑒
𝜏𝑤
𝑈∗ = √ = 0.0111156 𝑚/𝑠
𝜌
For rough flow, the velocity equation changes slightly with the addition of a value called ΔB. This value
essentially shifts the logarithmic overlap curve to reflect the rough pipe flow and is calculated as follows:
1 𝜀𝑢∗
Δ𝐵 = ln ( ) = −1.36
𝑘 𝜐
Mesh settings for both Transient and Fully Turbulent flow are shown in Table 9.
Similar to smooth flow, the settings I have used for each run have been standard apart from the inlet velocity
profile used (which now includes the ΔB term), water as my fluid and medium intensity as my turbulence
option. I have also set the wall boundary layer to “Rough Wall” setting with a 0.046mm sand grain roughness to
represent the commercial steel pipe being used.
As before my primary comparison is the velocity profiles produced by each type of run and the centreline
velocity. My secondary comparison is the axial pressure variation along the pipe.
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Good Run Good Run Good Run
Error @ Centreline 0.0296915 0.027198047 0.031539547 0.028821453
Error @ Centreline -18.14% 12.33% 14.02% -17.51%
Run Time 00:02:25 00:11:12 00:07:54 00:05:08
Iterations 10 41 21 14
Efficiency 17.69% 3.50% 5.75% 8.08%
Table 10 - Transient Rough Flow - Pipe 1
0.2
Velocity (m/s)
0.15
0.1
0.05
0
-0.0075 -0.005 -0.0025 -1.2E-17 0.0025 0.005 0.0075
Pipe Radius (m)
10
8
6
4
2
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
For transient rough flow on Pipe 1 my initial observation was the increased error in my results compared to
smooth flow, although not completely off the chart, Pipe 1 errors ranged from 12.33% to 18.14%, it is evident
that CFD is less reliable in the transition region than it is for a smooth pipe which is likely to do with the friction
factor used.
K-Epsilon has again resulted in a very efficient run compared to the other results but also the largest error
(18.14%). Likewise, Shear Stress Transport has the smallest error but takes the longest time to converge.
Steven Goddard –Mechanical Engineering (Part Time)
Student Number: 10038749 CFD Assignment Page 29 of 41
When looking at the pressure variation along the pipe all results start with relatively large errors but BSL
Reynolds Stress eventually takes the closest profile to the theoretical values.
Unlike the results from smooth pipe flow, the pressure does not correlate with the under and over-prediction
for centreline velocity. For example, the Shear Stress Transport shows the highest starting pressure, yet it does
not have the highest centreline velocity.
Another observation in these results is on the K-Epsilon velocity profile towards the pipe wall, as the velocity
increases sharply the profile looks like it overshoots very slightly before going into the shallower gradient
towards the centre; this could be evidence of an anomaly or inaccuracy in the calculations.
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Good Run Good Run Good Run
Error @ Centreline 0.03263345 0.025194047 0.030919047 0.030427453
Error @ Centreline -20.30% 11.53% 13.78% -18.67%
RunTime 00:02:52 00:14:25 00:09:42 00:06:59
Iterations 8 39 19 13
Efficiency 16.39% 2.52% 4.59% 6.27%
Table 11 - Transient Rough Flow - Pipe 2
0.2
Velocity (m/s)
0.15
0.1
0.05
0
-0.0075 -0.005 -0.0025 -1.4E-17 0.0025 0.005 0.0075
Pipe Radius (m)
10
8
6
4
2
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Pipe 2 confirms that CFD for transient flow is less accurate than for smooth flow with errors ranging from
11.53% to 20.3%. Efficiency was less than Pipe 1 which is further evidence that a finer mesh toward the pipe
wall and a lower aspect ratio may not always be the key to a more successful simulation.
The K-Epsilon and SSG Reynolds Stress produce a slightly larger pressure error on Pipe 2 whereas Shear Stress
Transport and BSL Reynolds Stress are approximately identical.
The velocity profiles look very similar to Pipe 1 with BSL Reynolds Stress producing the expected shape profile
despite having a high error centreline velocity. K-Epsilon also shows slightly larger overshoots as described in
Pipe 1 at the velocity changes from pipe wall to outer layer.
My length for fully developed flow is 0.792m and I have chosen to go with a 0.9m length pipe.
To calculate velocity the equation is similar to transitional turbulence whereby you subtract the ΔB term.
However the calculations for friction factor, ΔB and pressure drop are different.
There is a specific equation for friction factor for fully rough flow (White F. M, 2003), again I have used the
roughness of commercial steel, 0.046mm:
2
1
𝑓=( ) = 0.02632
𝜀/𝑑
−2 log ( 3.7 )
1 𝜀𝑢∗
ΔB = ln ( ) − 3.5 = 11.78942
𝑘 𝜐
For pressure drop Blasius equation no longer holds true as the Reynolds number is too high, therefore I have
returned to using the original equation for pressure drop.
𝑓𝐿
Δ𝑝 = 0.5𝜌𝑉 2 = 25094252.16 𝑃𝑎
𝐷
Going through the process again I obtain the following values for shear stress and friction velocity:
𝐷Δ𝑝 𝜏𝑤
𝜏𝑤 = = 104559.384 𝑃𝑎 𝑈∗ = √ = 10.2408 𝑚/𝑠
4𝐿 𝜌
Mesh settings for my CFD runs are the same as transient flow and can be seen in Table 9. Run settings are
identical to transient flow settings apart from the new variables shown above.
All of these runs were carried out on a slower computer that all other runs which has had a direct effect on the
run-time. (Intel i3 Processor with 4GB Ram in room 1N65).
As before my primary comparison is the velocity profiles produced by each type of run and the centreline
velocity. My secondary comparison is the axial pressure variation along the pipe.
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Good Run Good Run (Fluctuating) Good Run
Error @ Centreline 216.431 221.041 211.517 217.7475
Error @ Centreline 1.31% 3.36% -0.98% 1.90%
Run Time 00:11:32 00:22:01 00:32:58 00:31:24
Iterations 31 56 65 61
Efficiency 27022.60% 14457.19% 9239.17% 9985.87%
Table 12 - Fully Turbulent Rough Flow - Pipe 1
For my first fully turbulent rough flow run I observed a large increase in accuracy throughout all turbulence
models (0.98% - 3.36%). These runs also took far longer to converge than the previous turbulent scenarios.
In a complete reversal of previous simulations Shear Stress Transport now gives the most inaccurate result
whereas k-Epsilon now gives a very accurate centreline velocity at the lowest run-time.
200
Velocity (m/s)
150
100
50
0
-0.0075 -0.005 -0.0025 -1E-17 0.0025 0.005 0.0075
Pipe Radius (m)
Figure 25 - Velocity Profile Comparison for Fully Turbulent Rough Flow - Pipe 1
220
Velocity (m/s)
215
210
205
200
-0.0025 2E-18 0.0025
Pipe Radius (m)
Figure 26 shows a close up of the centreline velocity; from this figure the centreline velocity errors are more
easily identifiable. An additional observation I made during this run was with the BSL Reynolds Stress, as shown
in Figure 26 the velocity profile is fluctuating and continues to do this all the way until the steep gradient
towards the pipe wall, however BSL Reynolds Stress still retains the best steep to (approximately) flat velocity
profile shape and has the most accurate centreline velocity.
With the pressure, SSG Reynolds Stress and Shear stress Transport produce a lower starting pressure which
inversely correlates with a higher centreline velocity whereas BSL Reynolds Stress and K-Epsilon are very
accurate which is represented also via the velocity profile.
8.00E+06
Pressure (Pa)
6.00E+06
4.00E+06
2.00E+06
0.00E+00
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Figure 27 - Axial Pressure Variation for Fully Turbulent Rough Flow - Pipe 1
Turbulence Model K-Epsilon Shear Stress Transport BSL Reynolds Stress SSG Reynolds Stress
Notes Good Run Good Run Good Run (Fluctuating) Good Run
Error @ Centreline 217.945 225.7195 209.725 224.02
Error @ Centreline 1.99% 5.36% -1.84% 4.64%
Run Time 00:14:32 00:29:04 00:45:43 00:38:03
Iterations 28 52 62 55
Efficiency 21594.55% 11182.43% 6605.99% 8478.02%
Table 13- Fully Turbulent Rough Flow - Pipe 2
200
Velocity (m/s)
150
100
50
0
-0.0075 -0.005 -0.0025 -1.2E-17 0.0025 0.005 0.0075
Pipe Radius (m)
Figure 28- Velocity Profile Comparison for Fully Turbulent Rough Flow - Pipe 2
Steven Goddard –Mechanical Engineering (Part Time)
Student Number: 10038749 CFD Assignment Page 34 of 41
225
220
Velocity (m/s)
215
210
205
200
-0.0025 2E-18 0.0025
Pipe Radius (m)
8.00E+06
Pressure (Pa)
6.00E+06
4.00E+06
2.00E+06
0.00E+00
0 0.05 0.1 0.15 0.2 0.25 0.3
Distance from Outlet (m)
Figure 30 - Axial Pressure Variation for Fully Turbulent Rough Flow - Pipe 2
Pipe 2 has generated data with similar characteristics of previous runs, firstly the data is less accurate than pipe
1 (1.84% to 5.36%) despite higher bunching ratio and lower aspect ratio. As for SSG Reynolds Stress and Shear
Stress Transport these both give lower starting pressures resulting in higher centreline velocities, further
clarifying the results from Pipe 1.
K-Epsilon and BSL Reynolds Stress both give similar velocity errors; however K-Epsilon does this with far less
iterations. BSL Reynolds also exhibits the fluctuating behaviour of Pipe 1 which demonstrates that this wasn’t
due to computation anomalies and is more due to a characteristic of this turbulent model.
Figure 32 - Centreline Velocity along Pipe (Data from Pipe 2 - Shear Stress Transport)
In addition to my general observations of axial pressure and velocity profiles I have also noticed strange behaviour
with the velocity profile (for both meshes) along the length of the pipe in the z-direction during the fully turbulent
rough pipe runs. Figure 31 and Figure 32 illustrate what I was seeing. From 0.0m to approximately 0.4m along the
pipe the centreline velocity increases before slowing and gently rising again at which point my pipe mesh ends.
In an effort to understand this further I have analysed various characteristics of the flow, Figure 33 clearly shows
activity at 0.4m whereby there is a large increase in turbulent kinetic energy.
I have concluded that if Figure 32 was extrapolated the centreline velocity would once again rise by less than the first
peak and oscillate periodically tending to a steady velocity. One explanation for this could be that the flow has not
fully developed in the 0.9m length as calculated via the entrance length method. This then leads me to conclude that
the entrance length calculations are not suitable for fully turbulent rough flow or flow at high Reynolds numbers.
To test his hypothesis I could create a longer pipe with the same turbulent conditions, due to time constraints and
the scope of this assignment I have not done this.
Figure 33 - Centreline Velocity along Pipe (Data from Pipe 1 - Shear Stress Transport)
3. Alternative Methods
Throughout this investigation I have been amazed by the amount of different methods that can be used both in
theoretical calculations and computation methods, this truly shows how active and exciting this area of physics
currently is.
Computational Methods
5. Conclusion
There are several conclusions than can be made from this investigation:
5.1.Laminar Flow
It is clear that laminar flow can be calculated with adequate precision through theoretical and computational
methods, this is proven by my investigations in Section 1 showing errors to the precision of 0.01%.
It is evident that some theoretical work needs to be done to achieve the best results via simulation. The
difference can be seen by simply setting the inlet velocity to a uniform value as per my first simulation and
subsequently applying the Hagen Poiseuille parabolic velocity profile.
Smooth Flow generated reasonably accurate results compared to theory and also proved that a good mesh for
laminar flow may not necessarily be ideal for turbulent flow, leading to a need to re-iterate my mesh selection.
Transitional Flow generated less accurate results than smooth flow and this was probably due to the
uncertainty when calculating friction factors coupled with the additional variables relating to the rough pipe
surface.
For both Smooth and Transitional flow the K-Epsilon method seemed to be an efficient way to get a quick and
initial result. Once happy with the simulation settings and mesh, Shear Stress Transport should be used. This
turbulence model generally takes longer to converge but provides a more accurate representation of the flow.
For fully turbulent flow in a rough pipe BSL Reynolds Stress seems to result in the most accurate error at
centreline velocity (1.84%), however the profile shows small random fluctuations across the pipe which leads
me to think that this should not be completely trusted. K-Epsilon seems to work very well to this type of flow
giving a smooth and very accurate (1.99%) error at centreline velocity. Another benefit of the K-Epsilon method
is that it is very quick to converge and provides the most efficient way of getting an accurate result.
5.3.Final Remarks
Overall, throughout this investigation I have come to realise that whilst laminar flow is well understood,
turbulent flow is not and despite numerous methods of approximation and attempts of understanding based on
empirical data turbulence cannot always be successfully predicted.
This leads to me questioning the actual validity of the theoretical and the computational predictions.
The only way to truly understand and trust theoretical and or computational data is develop a scenario that is
adequate for the problem you are dealing with, once you have done this theoretical and/or computational
predictions can me made which must then be compared to physical experimentation. This could even result in
an iterative process to calibrate simulations to better match experimental data.
CFD is an extremely useful tool but ultimately even the input data one provides is based on approximated
theory and empirical data the output data is also based on a variety of modelling assumptions.
Physical experimental data will give a truly accurate representation but this is often not possible or cost
effective and compromises need to be made.
Theory and CFD provide a good educated guess and that is usually a better start than having no idea at all!
References
ANSYS (2014), ANSYS CFX (Release 15.0) [Computer Program]
[available at: http://www.ansys.com/Simulation/Fluid+Dynamics/Fluid+Dynamics+Products/ANSYS+CFX]
Unknown (2000) Water Central Heating [Online]
http://www.diydata.com/planning/central_heating/pipework.php [accessed 20/02/14]
Fisher D. A. (2013) Creating a Mesh for CFX Analysis using ICEM CFD, University West of England
Robodoc (2004), Osborne Reynolds [Online Image]. Available from: http://en.wikipedia.org/wiki/Osborne_Reynolds
[accessed: 20/02/14]
Centralblatt der Bauverwaltung (1899), Gotthilf Hagen [Online Image]. Available from:
http://en.wikipedia.org/wiki/Gotthilf_Heinrich_Ludwig_Hagen [accessed: 20/02/14]
Artem Korzhimanov (2010), Jean-Louis Marie Poiseuille [Online Image]. Available from:
http://en.wikipedia.org/wiki/Jean_L%C3%A9onard_Marie_Poiseuille [accessed: 20/02/14]
Salih (2013) Hagen–Poiseuille flow from the Navier–Stokes equations [Online]. Available from:
http://en.wikipedia.org/wiki/Hagen%E2%80%93Poiseuille_flow_from_the_Navier%E2%80%93Stokes_equations
White F. M. (2003), Chapter 6 - Viscous Flow in Ducts. Fluid Mechanics (5th ed.). McGraw-Hill
BNews.kz (2014) Academician from Astana solves one of most difficult math problems of millennium. BNews.kz, 10th
January. Available from: http://bnews.kz/en/news/post/180213/ [accessed: 20/02/14]
White F. M. (2003), Typical Velocity and Shear Distributions in Turbulent Flow Near a Wall. Fluid Mechanics (5th ed.).
McGraw-Hill)
Science Advisory Board (SAB) (2012), Von Karman [Online Image] Available at:
http://www.sab.af.mil/library/factsheets/factsheet.asp?id=18803 [accessed: 20/02/14]
Sechler E. E. (1969) Clark Blanchard Millikan 1903-1966 [Book], National Academy of Sciences. Available from:
http://www.nasonline.org/publications/biographical-memoirs/memoir-pdfs/millikan-clark.pdf [accessed: 20/02/14]
Karthik T. S. D. (2011), Turbulence Models and Their Applications [Presentation] Department of Mechanical
Engineering IIT Madras. Available at: http://www.leb.eei.uni-
erlangen.de/winterakademie/2011/report/content/course01/pdf/0112.pdf [accessed: 20/02/14]
Harel Geron (2013) Dary Friction Factor Formulae [Online]. Available from:
http://en.wikipedia.org/wiki/Darcy_friction_factor_formulae#Approximations_of_the_Colebrook_equation
[accessed: 20/02/14]
Bibliography
May A. P. H. (2012) Introduction to CFD. Thermofluid Systems, University West of England
May A.P.H (2009) Introduction to Thermofluid Systems, Thermofluid Systems, University West of England
May A. P. H (n.d) CFD in Pipe Flows, Thermofluid Systems, University West of England
Jackson D., Launder B. (2007) Osborne Reynolds and the Publication of His Papers on Turbulent Flow. University of
Manchester [available at: http://158.110.32.35/download/CERN07/ARFM2007.pdf] [viewed on 20/02/14].
Hu Y. F., Carter J. G., Blake R. J. (1995) The Effect of Grid Aspect Ratio on the Convergence of Parallel CFD Algorithms.
Daresbury Laborator [available at: http://www2.research.att.com/~yifanhu/PUB/aspect.pdf] [viewed on 20/02/14]
Karthik T. S. D. (2011) Turbulence Models and Their Applications [available at: http://www.leb.eei.uni-
erlangen.de/winterakademie/2011/report/content/course01/pdf/0112.pdf] [viewed on 20/02/14]
Singh, S. Bhandari, D. (2012) Analysis of Fully Developed Turbulent Flow in a Pipe Using Computational Fluid
Dynamics. Bipin Tripathi Kumaon Institute of Technology.
Ryan, D. (n.d) ANSYS ICEM CFD and ANSYS CFX Introductory Training Course Notes. University of Birmingham.
Appendix