Sunteți pe pagina 1din 64

Development of Turbulence Model for High

Pressure Natural Gas Transmission Pipelines

WANG LINGMENG

A DISSERTATION SUBMITTED IN PARTIAL FULFILMENT OF


THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE IN POWER ENGINEERING

2017
TABLE OF CONTENTS

Abstract .......................................................................................................................................... I

Acknowledgements ....................................................................................................................... II

Symbols ........................................................................................................................................ III

List of Figures ..............................................................................................................................IV

List of Tables ................................................................................................................................. V

Chapter 1 Introduction ................................................................................................................. 1

1.1 Background......................................................................................................................... 1

1.2 Research Motivation ........................................................................................................... 2

1.3 Scope .................................................................................................................................. 3

1.4 Organization of the Dissertation ......................................................................................... 3

Chapter 2 Lecture Review ............................................................................................................ 5

2.1 Turbulent Flow ................................................................................................................... 5

2.2 Methods of Turbulence Analysis ........................................................................................ 6

2.3 Wall-bounded Shear Flows ................................................................................................. 7

2.4 Von Kármán Vortex Street .................................................................................................. 8

2.5 Elastic Strain ....................................................................................................................... 9

Chapter 3 Turbulent Modeling................................................................................................... 10

3.1 Governing Equation of Incompressible Flow ................................................................... 11

3.1.1 Derivation of Continuity Equation ....................................................................... 12

3.1.2 Derivation of Momentum Equation ...................................................................... 13

3.2 Direct Numerical Simulation (DNS) ................................................................................ 13

3.3 RANS-Based Turbulence Models..................................................................................... 14

3.3.1 Reynolds-Averaged Navier–Stokes Equations ..................................................... 14

3.3.2 The 𝑘 − 𝜔 Model ............................................................................................... 15

3.3.3 The 𝑘 − 𝜖 Model ................................................................................................ 17

I
3.4 Reynolds Stress Model (RSM) ......................................................................................... 17

3.5 Large Eddy Simulation ..................................................................................................... 18

3.6 Vorticity Dynamics ........................................................................................................... 19

Chapter 4 Flow Past a Cylinder/Sphere .................................................................................... 21

4.1 Flow Past a Cylinder......................................................................................................... 22

4.1.1 Geometry .............................................................................................................. 25

4.1.2 Mesh Set Up ......................................................................................................... 26

4.1.3 Fluent Setup .......................................................................................................... 27

4.1.4 Results of Simulation ........................................................................................... 28

4.2 Flow past a sphere ............................................................................................................ 34

4.2.1 Geometry .............................................................................................................. 34

4.2.2 Mesh Set Up ......................................................................................................... 35

4.2.3 Fluent Set Up ........................................................................................................ 36

4.2.4 Results of Simulation ........................................................................................... 36

Chapter 5 Leaked Pipeline Simulation ...................................................................................... 39

5.1 Velocity Profile of Fluid Flow in Pipes ............................................................................ 39

5.1.1 Laminar Pipe Flow ............................................................................................... 40

5.1.2 Turbulent Pipe Flow ............................................................................................. 40

5.2 Geometry .......................................................................................................................... 41

5.3 Mesh Set Up ..................................................................................................................... 42

5.5 Fluent Set Up .................................................................................................................... 44

5.6 Results ............................................................................................................................. 44

5.6.1 Results of FLUENT.............................................................................................. 44

5.6.2 Results of Fluid Structure Interaction ................................................................... 47

Chapter 6 Conclusions and Future Works ................................................................................ 51

6.1 Conclusions ...................................................................................................................... 51

6.1.1 Influence of Dents ................................................................................................ 51

6.1.2 Pressure, Deformation and Strain of Leaked Pipe ................................................ 51

II
6.2 Future Works .................................................................................................................... 52

References ....................................................................................................................................53

III
Abstract
Currently, most electricity in Singapore is generated by imported natural gas via
long-distance pipeline. Pipeline is the most traditional and commonly used way of
transmission. But fatigue often occurs on the pipe wall due to the external
interference and internal pressure fluctuations, leading to a leakage eventually. This
is the biggest problem we are facing of gas transmission via pipeline.

The first part of this dissertation discusses turbulent flow and vortex shedding
phenomenon. And then the methods of turbulent modeling are reviewed, including
Reynolds-averaged Navier-Stokes equation (RANS) based model, large eddy
simulation (LES), direct numerical simulation (DNS) and Reynolds stress model
(RSM).

In this project, two cases are conducted to study the flows in pipe. One is focused on
the vortex phenomenon. The main purpose of this case study is to determine the
influence of dents on pipe wall. The other one is a simulation of a leaked pipe,
aiming to investigate pressure and strain variation in the vicinity of leaking point,
which are crucial parameter to detect leakage.

I
Acknowledgements

First of all I would like to thank my dissertation supervisor Associate Professor So

Ping Lam of School of Electrical and Electronic Engineering. Prof. So always

provides me great help whenever I met difficulties in the project or had questions

about dissertation writing.

I would also like to express my gratitude to Mr. Wang Zhe, a Ph.D. student in

Interdisciplinary Graduate School (IGS). I appreciate his patience and support in

overcoming a number of obstacles I met in this project. This dissertation could not be

successfully conducted without his help and guide.

Lastly I would like to acknowledge my parents and friends for providing me with

unfailing support and continuous encouragement in this year. Thank you very much.

II
Symbols
𝑅𝑒 Reynolds number
𝜌 Density
𝑝 Pressure
𝐿 Length
𝑢 velocity
𝑓 Frequency
𝑆𝑡 Strouhal number
𝜎𝑒𝑞 Equivalent stress
𝑒𝑙
𝜀𝑒𝑞 Equivalent elastic strain
𝐸 Young’s modulus
𝑡 Time
𝜏 Stress tensor
𝜇 Viscosity
𝜈 Kinetic viscosity
𝛿𝑖𝑗 Kronecker symbol
m Mass
𝐹 Force
𝑎 Acceleration
𝑘 Kinetic energy
𝜈𝑇 Kinetic eddy viscosity
𝑅 Radius
𝜀 ⁄𝐷 Relative roughness
𝜔 Vorticity
𝐶𝐷 Drag force coefficient

III
List of Figures
Figure 2.1 Turbulent flow ..................................................................................... 6
Figure 2.2 Von Kármán Vortex Street ................................................................... 8
Figure 4.1 Flow chart of FLUENT ..................................................................... 21
Figure 4.2 Drag force coefficient ........................................................................ 22
Figure 4.3 Flow pattern at different Reynolds number ....................................... 24
Figure 4.4 Geometry of 2D model ...................................................................... 25
Figure 4.5 Partition of geometry ......................................................................... 25
Figure 4.6 Mesh near the cylinder ...................................................................... 26
Figure 4.7 Mesh .................................................................................................. 27
Figure 4.8 Development of vortex shedding ...................................................... 29
Figure 4.9 Reynolds number = 106 ..................................................................... 33
Figure 4.10 Reynolds number = 107 ................................................................... 33
Figure 4.11 Geometry of 3D model .................................................................... 34
Figure 4.12 Blocks .............................................................................................. 35
Figure 4.13 Scan plane........................................................................................ 35
Figure 5.1 Velocity profile of laminar flow ........................................................ 40
Figure 5.2 Velocity profile of turbulent flow ...................................................... 41
Figure 5.3 Geometry of leaked pipe ................................................................... 42
Figure 5.4 Blocks ................................................................................................ 42
Figure 5.5 O-grids meshing ................................................................................ 43
Figure 5.6 Scan plane.......................................................................................... 43
Figure 5.7 Elastic strain at (a) t=0.2s, (b) t= 0.5s, and (c) t=0.8s ....................... 48

IV
List of Tables
Table 4.1 Boundary conditions of 2D model ...................................................... 28
Table 4.2 Velocity and vorticity contour - LES.................................................. 30
Table 4.3 Velocity and vorticity contour - SST .................................................. 32
Table 4.4 Boundary conditions of 3D model ...................................................... 36
Table 4.5 Velocity contours of 3D model........................................................... 37
Table 5.1 Boundary conditions of leaked pipe model ........................................ 44
Table 5.2 Pressure ............................................................................................... 45
Table 5.3 Velocity streamline .............................................................................. 46
Table 5.4 Elastic strain........................................................................................ 49

V
Chapter 1 Introduction

1.1 Background

Currently, Singapore’s power generation is predominantly reliant on fossil fuels and


this situation will continue in the next decades. Among fossil fuels, natural gas plays
a vital part in Singapore energy market as it is economical and convenient
transportation. About 95% of Singapore’s total commercial power generation is
dominated by natural gas engine generator [1].

Most of Singapore’s natural gas is imported from Malaysia and Indonesia by


long-distance transportation pipeline. Except where the pipelines cross over canals,
the whole piping manifold situates underground, passing through broad area. The
total length of high pressure line network is over 180km [2]. Pipeline is one of the
most secure ways to deliver gas [3]. However, this does not mean there is no risk.
Transportation with pipelines faces a number of problems. For example, the natural
gas usually consists of some kinds of corrosive gas such as hydrogen sulfide,
accelerating the corrosion. Meanwhile, humid and capricious climate in Singapore
will increase the possibility of rusting. Additionally, some human activities will
cause damages to the pipelines. Such corrosion, rusting and external interferences
may cause pipeline leakage, and the leakage of high pressure natural gas will pose
threat to public property as well as human safety [4]. Olympic Pipeline explosion in
Bellingham, in United States, on 10 June 1999, caused three people died and 277,000
gallons of gasoline spilled to Whatcom Creek [5]. In 2014, gas pipeline explosion
occurred in the state of Sarawak, Malaysia. As a result, all nearby residents were
evacuated by Malaysia government and this incident caused enormous economic
losses. Fortunately, no lives were lost [6]. Thus, monitoring of the pipeline working

1
condition and detection of leakages not only reduce the cost of industry but also
guarantee personal security.

In the past decades, significant efforts have been made in leakage detection technique
to prevent us from catastrophic effect of pipeline leakage. At present, the most
pervasively used techniques can be divided into two types: hardware-based and
software-based [7]. In the former type, sensors are used to detect the abnormal
parameters, such as pressure, vibration, mass-flow rate and acoustics, and then locate
the leakage. The other type, software-based technology, is based on different
programs and algorithm. Furthermore, some methods use both two techniques
simultaneously. However, detecting and predicting the leakage preciously is an
extremely difficult task. When a leakage occurs, the signals received by sensors
generally contain many interference factors [8]. Those factors can make vitally
adverse influence on data analysis. Therefore, detection and prediction of pipeline
leakage is still a challenging mission.

1.2 Research Motivation

With the rapid economic growth, natural gas becomes more and more important as
industry resource. Long distance and large diameter pipeline is the most traditional
and widely used method of gas transmission because of its convenience and
economy. However, long-distance pipeline needs great early investment and the
early investment covers more than 90 percent of total cost. Therefore, leakage of
pipeline is a serious problem commonly encountered in practical engineering which
may cause huge economic losses. More than that, pipeline leakage might cause
personal injury even death.

2
The primary target of this dissertation is to analyze the high pressure pipeline
network in Singapore. Preventive measures must be taken such as leakage detection,
so in this dissertation vortex shedding phenomenon is simulated in order to analyze
fatigue time estimation and leakage prediction.

Furthermore, for a leaked pipe, it is meaningful to measure the variation of some


parameters. Learning the change of those parameters can help us detect the location
and severity of pipeline leakage. For example, pressure variation and elastic strain
are critical parameters to detect pipeline leakage. Therefore, it is necessary to
evaluate the change of parameters to monitor the pipeline operating condition.

1.3 Scope

In this projection, several simulations are conducted to evaluate the influence of


dents on pipeline and parameter variation of a leaked pipeline. First case is focused
on the influence of dents, where the vortex shedding and VIV frequency. A two
dimensional model and a three dimensional model are conducted for this case. The
other case study is based on a leaked pipe model, and the pressure distribution and
strain on the pipe wall are analyzed.

1.4 Organization of the Dissertation

In Chapter 2 the basic theory of turbulence is reviewed, including turbulent flow,


wall-bounded shear flow and Kármán Vortex Street. Then turbulence modeling
theory, the mathematical model of turbulent flow, is presented in Chapter 3. Chapter
4 shows the ANSYS simulation results of two-dimension and three-dimension model
of vortex shedding, evaluating the influence of dents on the pipe wall. Chapter 5
shows the simulation of three-dimension leaked pipe. The last chapter discusses the

3
conclusion of this dissertation and the limitation of this project. And there are a lot of
improvements could be made in future works.

4
Chapter 2 Lecture Review

2.1 Turbulent Flow

Turbulence, or turbulent flow, is one of the most common phenomena in daily life
such as river flow, storm cloud and sea wave. It refers to a flow regime where
pressure and velocity change chaotically. Turbulence does not have a precise
definition, but many characteristics of turbulence could be listed [9]:

1. Irregularity:
All turbulent flows are randomness and out of order, which makes it very tough to
solve by traditional deterministic.

2. Diffusivity:
Diffusivity is another important characteristic of turbulence, which is the main
reason of rapid mixing and increased momentum, heat, and mass transfer.

3. Large Reynolds numbers:


Reynolds number is mainly determined by characteristic length scale L, fluid
viscosity μ and velocity u. Flow is laminar at low Reynolds number, but it turns to
turbulence when Reynolds number becomes large. Reynolds number is defined as:
𝜌𝑢𝐿
𝑅𝑒 = (2.1)
𝜇

4. Rotational and three-dimensional


Turbulent flow has a high level of three-dimensional fluctuating vorticity, which
means turbulent flow is always doing three-dimensional rotation.

5. Dissipation:

5
Turbulent flow is always dissipative, so it needs continuous supply of energy.

6. Continuum:
Turbulence is a continuum phenomenon, governed by the equations of fluid
mechanics.

Figure 2.1 Turbulent flow [10]

Figure 2.1 shows the pattern of a turbulent flow. Though turbulence has many
peculiar characteristics, it is not a feature of fluid but of fluid flows. Most of the
dynamics of turbulence is the same in all fluid, whether they are liquids or gases.

2.2 Methods of Turbulence Analysis

Though turbulence theory has been investigated for long time, there is no general
solution. In turbulence problem, experience is necessary to relate equations and
actual situation. Some commonly used methods are:

1. Dimensional Analysis
In many situations the structure of turbulent is considered to be dominated by several
independent parameters. And in those conditions dimensional analysis can be used to
indicate the relation between these variables and the solution of turbulence problem
can be obtained except some coefficients.

6
2. Local invariance
In some simple turbulent problems, turbulent flow motion at certain time or location
mainly governed by the immediate environment. The time and length scales of the
flow usually vary slowly downstream. However, it is reasonable to assume that the
turbulence is dynamically similar if the time scales are small enough and can change
as the environment variation.

2.3 Wall-bounded Shear Flows

On one hand, according to the existence of velocity gradient and other gradient of
hydrodynamic parameter, the flow can be classified into uniform flow and shear flow.
Uniform flow is defined as a flow whose velocity and other parameters at all points
stay the same at a certain time. Conversely, shear flow is defined as flow caused by
forces, and adjacent layers of fluid move parallel to each other with different speeds
in a shear flow. Shear flow frequently occurs in nature and engineering application.

On the other hand, according to the wall constrain, shear flow can be classified into
free flow and wall-bounded flow. Free flow includes several kinds, such as jet flow
and wake flow. Boundary-layer flows usually are more complicated than flow in free
shear layers because of the wall constrains.

Pipe flow is an example of type of wall-bounded shear flow whose motion is


primarily dominated the effects of viscosity. The velocity profile and pressure loss is
discussed in Chapter 3.

7
2.4 Von Kármán Vortex Street

If a blunt body is put in the flow, in some certain situations, a repeating pattern of
swirling vortices caused by the unsteady separation of flow will be induced which
called von Kármán vortex street, which is shown in Figure 2.2.

Figure 2.2 Von Kármán Vortex Street [11]

When the Reynolds number is between 250 and 200000, the vortex shedding
frequency is governed by [12]:
𝑆𝑡 ∙ 𝑢
𝑓= (2.2)
𝑑
where u is the flow velocity, d is diameter of the blunt body and Sr is Strouhal
number. Strouhal number describes oscillating flow mechanisms. It has been proved
by experiments that Strouhal number is a function of Reynolds number 𝑅𝑒, which is
often given as [13]:
19.7
𝑆𝑡 = 0.198(1 − ) (2.3)
𝑅𝑒
The vortex shedding is the main cause of vortex induced vibration (VIV). When a
single vortex sheds, an asymmetrical flow pattern forms around the body and
changes the pressure distribution. This means that the alternate shedding of vortices
can create periodic lateral forces on the body, causing vibration on it. If the vortex
shedding frequency is similar to the natural frequency of a body or structure, it
causes resonance which is called lock-in phenomenon. Nowadays, this phenomenon
is widely considered as one of the main reasons of pipeline fatigue and most
leakages.

8
In previous studies, a number of experiments and simulations were focused on the
vortex shedding wake configuration. In 1954, Roshko [14] first measured the
frequency of vortex shedding. Bishop and Hassan [15] measured the lift and drag
forces of a flow past a cylinder, and the result shows the change of magnitude and
timing of the forces on the cylinder. Gaster [16] developed experiments to observe
the wake flow past slender cones and found the irregular signal of velocity caused by
presence of span wise cells of different frequencies. In his further study, stronger
evidence was provided to demonstrate this phenomenon is caused by free-stream
non-uniformities.

2.5 Elastic Strain

Strain describes the deformation according to relative displacement of particles in the


body. Internal strain can be classified into elastic strain and plastic strain. It is
generally considered that elastic strain is caused by distortion of crystal lattice. The
equivalent elastic strain is related to equivalent stress, given by Hooke’s Law [17]:
𝑒𝑙
𝜎𝑒𝑞 = 𝐸𝜀𝑒𝑞 (2.4)

𝑒𝑙
where 𝜎𝑒𝑞 is equivalent stress, 𝜀𝑒𝑞 is equivalent elastic strain and E is Young’s

modulus or elastic modulus describing the stiffness of a solid material.

9
Chapter 3 Turbulent Modeling
Turbulent modeling is a key issue in most computational fluid dynamics simulations.
However, as mentioned earlier, there is no general approach to solve turbulence
problem. This is due to the presence of a wide range of spatocal-temporal
characteristic scales of the motion, generated by the non-linearity and not purely
deterministic. Turbulent motion (by definition 3D and unsteady) is characterized by a
kinetic energy distribution over a range of scales, coming from the integrals ones (L,
T), wherein energy is provided to the system. Kolmogorov scale is the smallest one
[9]:
𝜈3 1
𝜂 = ( )4 (3.1)
𝜖
And Taylor micro-scale is the biggest of the dissipative scales, which is [9]:
2
2
𝑢′
𝜆 = (3.2)
𝜕𝑢
〈( 𝑖 )2 〉
𝜕𝑥𝑖
𝜕𝑢 𝜕𝑢
where 〈(𝜕𝑥𝑖 )2 〉 is the average of (𝜕𝑥𝑖 )2 , and the subscript i describes the direction
𝑖 𝑖

of velocity in the x-axis of a 3-D coordinate system.

Statistical studies of the motion equation always lead to situation in which there are
more unknowns than equations. This is called the closure problem of turbulence
theory: one has to make assumptions to make the number equations equal to number
of unknowns. Several methods are developed to solve the closure problem, such as
RANS-based models and large eddy simulation (LES). In this chapter, several
turbulence modeling methods are introduced.

10
3.1 Governing Equation of Incompressible Flow

The motion of viscous fluid is governed by Navier-Stokes Equations, named after


Claude-Louis Navier and George Gabriel Stokes.

The instantaneous continuity equation, momentum equation and energy equation of


Navier- Stokes Equation can be written as [18]:

𝜕𝜌 𝜕𝑢𝑖
+𝜌 =0 (3.3)
𝜕𝑡 𝜕𝑥𝑖
𝜕(𝜌𝑢𝑖 ) 𝜕(𝜌𝑢𝑗 𝑢𝑖 ) 𝜕𝜏𝑖𝑗 𝜕𝑝
+ = − (3.4)
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑖

𝜕(𝜌𝐸) 𝜕(𝜌𝑢𝑗 𝐸) 𝜕 𝜕𝑇 𝜕
+ = (𝑘 )+ (𝜏 𝑢 ) (3.5)
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝑖𝑗 𝑖

where 𝜌 is the density, u is the fluid velocity, p is the pressure, t is the time, T is the
temperature and 𝜏 is the deviatoric stress tensor. For Newtonian fluid, the stress
tensor is [19]:

𝜕𝑢𝑖 𝜕𝑢𝑗 2
𝜏𝑖𝑗 = 𝜇[( + ) − 𝛿𝑖𝑗 𝛻 ∙ 𝒖] (3.6)
𝜕𝑥𝑗 𝜕𝑥𝑖 3

where 𝛿𝑖𝑗 is Kronecker symbol, 𝛻 is Nabla symbol.

In computational fluid dynamics, experimental results have shown that compressible


flows can be approximately modeled incompressible flows under appropriate
conditions. The basic governing equations of computational fluid dynamics for
incompressible flows include two equations - continuity equation and momentum
equation. They are respectively based on mass conservation, Newton’s second law.

If the fluid is incompressible, the density of fluid does not vary as the pressure
change, which means the continuity equation and momentum equation of
Navier-Stokes Equations can be written as [18]:

11
𝜕𝑢𝑖 (3.7)
𝜌 =0
𝜕𝑥𝑖
𝜕𝑢𝑖 𝜕𝑢𝑖 1 𝜕𝑝 𝜕 2 𝑢𝑖
+ 𝑢𝑗 =− +𝜈 (3.8)
𝜕𝑡 𝜕𝑥𝑗 𝜌 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗

where ν is the kinematic viscosity (𝜈 = 𝜇/𝜌).

Navier-Stokes Equations describe the motion of flow and they are useful today
because they can be used in simulating flow in pipes, ocean current and stellar
motion. However, turbulent flows have been investigate for more than a century, but
still no general approach to the analytical solution of problems in turbulence exists.
The analytical solution of Navier-Stokes Equations is one of the most difficult
unsolved problems around the world.

3.1.1 Derivation of Continuity Equation

The basic principle of continuity equation is mass conservation. In a moving fluid


element [20],
𝑑𝑚 = 𝜌𝑑𝑢 (3.9)
Since mass is conserved, then
𝐷(𝑑𝑚)
=0 (3.10)
𝐷𝑡
Combining (3.9) and (3.10),
𝐷(𝜌𝑑𝑢) 𝐷𝜌 𝐷(𝑑𝑢) 𝐷𝜌 1 𝐷(𝑑𝑢)
= 𝑑𝑢 +𝜌 = + 𝜌[ ]=0 (3.11)
𝐷𝑡 𝐷𝑡 𝐷𝑡 𝐷𝑡 𝑑𝑈 𝐷𝑡
It is known that:
1 𝐷(𝑑𝑢)
𝛻∙𝒖= (3.12)
𝑑𝑈 𝐷𝑡
Substitute (3.12) into (3.11),
𝐷𝜌
+ 𝜌𝛻 ∙ 𝒖 = 0 (3.13)
𝐷𝑡
where 𝒖 is the vector velocity field.

12
Equation (3.13) is the continuity equation. For an incompressible flow, the density of
the fluid flow stays constant, and thus the vector form of continuity equation can be
written as:
𝛻∙𝒖=0 (3.14)

3.1.2 Derivation of Momentum Equation

The basic physical principle of momentum equation is Newton’s second law:


𝑭 = 𝑚𝒂 (3.15)
When Newton’s second law is applied to the moving fluid element, it means that the
force on the fluid element equals its mass times the acceleration of the element. If we
only consider x component, then [9]:
𝜕 𝜕𝜌 𝑑𝒖
𝑑𝑭 = [ (𝜌𝒖)] 𝑑𝑥𝑑𝑦𝑑𝑧 = {𝒖 [ + ∇ ∙ (𝜌𝒖)] + 𝜌 }𝑑𝑥𝑑𝑦𝑑𝑧
𝜕𝑡 𝜕𝑡 𝑑𝑡 (3.16)

For a incompressible flow, if all components are taken into consideration, (3.16) can
be simplified as:
𝑑𝒖
𝑑𝑭𝑠𝑢𝑟𝑓𝑎𝑐𝑒 = 𝜌 𝑑𝑥𝑑𝑦𝑑𝑧 (3.17)
𝑑𝑡
Considering an incompressible flow of Newtonian fluid where the viscous stresses
are proportional to the element strain rate and coefficient of viscosity. Substitute
(3.14) into (3.17), the momentum equation can be written as:

∂𝑢𝑖 𝜕𝑢𝑖 1 𝜕𝑝 ∂2 𝑢𝑖
+ 𝑢𝑗 =− +𝜈 (3.18)
∂t 𝜕𝑥𝑗 𝜌 𝜕𝑥𝑗 ∂𝑥𝑗 ∂𝑥𝑗

3.2 Direct Numerical Simulation (DNS)

Direct Numerical Simulation is a method to simulate turbulent flows based on the


direct resolution of Navier-Stokes equations without additional turbulence models. In
DNS, all the correlative scales have to be solved in the simulation to get a reasonable
result, which means DNS requires a computational grid with a space-temporal

13
resolution sufficiently fine to be capable of describing the turbulence dynamic over
any characteristic scales of the motion, from the largest to the smallest. Thus, the cost
of computing direct numerical simulation is extremely high, no matter whether at
high or low Reynolds number. And this is the main reason why DNS is mainly used
in theoretical analysis.

3.3 RANS-Based Turbulence Models

In order to save computational cost, RANS-based model are developed for mean
velocity of a statistically stationary flow. RANS models can be divided into two
types: linear eddy viscosity models and nonlinear viscosity models. In linear eddy
viscosity models, Reynolds stresses are modeled by a linear constitutive relationship
with the mean flow straining field while an eddy viscosity coefficient is used to
relate the mean turbulence field to the mean velocity field in a nonlinear relationship.
In linear eddy viscosity models, they can be further classified into three types
according to the number of equations: zero-equation models (Algebraic turbulence
models), one-equation models and two-equation models. In CFD problem,
two-equation models are widely used and among those models, k − ω model and
k − ϵ are the most common.

3.3.1 Reynolds-Averaged Navier–Stokes Equations

In turbulence theory, the velocity u can be decomposed into mean flow U and
velocity fluctuations u’ [9]. Then:
𝑢(𝑥, 𝑦, 𝑧, 𝑡) = 𝑈(𝑥, 𝑦, 𝑧, 𝑡) + 𝑢′(𝑥, 𝑦, 𝑧, 𝑡) (3.19)
where U is interpreted as a time average, defined by:

1 𝑡0 +𝑇
𝑈 = lim ∫ 𝑢 𝑑𝑡 (3.20)
𝑇→∞ 𝑇 𝑡
0

14
This is called Reynolds decomposition. When Reynolds decomposition is used in
Navier–Stokes equations for an incompressible Newtonian flow, and if the continuity
equation is invoked, Navier-Stokes equations can be written as:

∂𝑈𝑖 ∂𝑈𝑗 1 ∂𝑝 ∂2 𝑈𝑖
+ 𝑈𝑖 =− +𝜈 (3.21)
∂t ∂𝑥𝑗 𝜌 ∂𝑥𝑗 ∂𝑥𝑗 ∂𝑥𝑗

where subscripts i and j describe the directions of velocity in the x-axis and y-axis of
a 3-D coordinate system.

Equation (3.21) is called Reynolds-averaged Navier–Stokes equation (RANS), which


is time averaged equation of flow motion. The Reynolds-averaged approach to
turbulence modeling requires that the Reynolds stresses are appropriately modeled. A
common method is relating Reynolds stresses to the mean velocity gradients [21]:
2
̅̅̅̅̅̅
−𝑢 ′ ′
𝑖 𝑢𝑗 = 2𝜈𝑇 𝑆𝑖𝑗 − 𝑘𝛿𝑖𝑗 (3.22)
3
where 𝜈𝑇 is kinematic eddy viscosity, 𝑆𝑖𝑗 is stress deviator tensor, 𝛿𝑖𝑗 is

Kronecker delta and k is turbulence kinetic energy, defined as:


1 ′ ′
𝑘= ̅̅̅̅̅̅
𝑢𝑢 (3.23)
2 𝑖 𝑗
But as mentioned earlier, there is closure problem of turbulence theory. To solve
turbulence problems, some assumptions have to be made.

3.3.2 The 𝒌 − 𝝎 Model

In the 𝑘 − 𝜔 model, according to Wilcox [22], the kinetic eddy viscosity is given
by:
𝑘
𝜈𝑇 = (3.24)
𝜔
where 𝜔 is the specific dissipation.
The turbulence kinetic energy is given by:

15
𝜕(𝜌𝑘) 𝜕(𝜌𝑘) 𝜕𝑈𝑖 𝜕 𝜕𝑘
+ 𝑈𝑗 = 𝜌𝜏𝑖𝑗 − 𝛽 ∗ 𝜌𝑘𝜔 + [(𝜈 + 𝜌𝜎 ∗ 𝜈𝑇 ) ] (3.25)
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗

The specific dissipation rate is given by:

𝜕(𝜌𝜔) 𝜕(𝜌𝜔)
+ 𝑈𝑗
𝜕𝑡 𝜕𝑥𝑗

𝜔 𝜕𝑈𝑖 𝜌 𝜕𝑘 𝜕𝜔 𝜕
=𝛼 𝜌𝜏𝑖𝑗 − 𝛽𝜌𝜔2 + 𝜎𝑑 + [(𝜈 (3.26)
𝑘 𝜕𝑥𝑗 𝜔 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗

𝜕𝜔
+ 𝜎𝜈𝑇 ) ]
𝜕𝑥𝑗

According to Wilcox’s theory [23], the closure coefficients and relations are as
following:
13 9 1 3
α= , 𝛽∗ = , 𝜎 = , 𝜎∗ = (3.27)
25 100 2 5
∂𝑘 ∂𝜔
0, ≤0
∂𝑥𝑗 ∂𝑥𝑗 1
𝜎𝑑 = , 𝜎𝑑0 = (3.28)
∂𝑘 ∂𝜔 8
𝜎𝑑0 , >0
{ ∂𝑥𝑗 ∂𝑥𝑗
1 + 85𝜒𝜔
𝛽 = 𝛽𝑜 𝑓𝛽 , 𝛽0 = 0.0708, 𝑓𝛽 = (3.29)
1 + 100𝜒𝜔
Ω𝑖𝑗 Ω𝑗𝑘 𝑆̂
𝑘𝑖 1 𝜕𝑣𝑚
𝜒𝜔 = | |, ̂
𝑆𝑘𝑖 = 𝑆𝑘𝑖 − 𝛿 (3.30)

(𝛽𝑜 𝜔) 3 2 𝜕𝑥𝑚 𝑘𝑖

The 𝑘 − 𝜔 model is one of the most commonly used models. It uses wall functions
and hence the simulation of 𝑘 − 𝜔 model has a relative high memory requirement.
Additionally, converging of 𝑘 − 𝜔 model is difficult to reach and it is sensitive to
initial guess. Therefore the 𝑘 − 𝜔 model is widely used where the 𝑘 − 𝜖 model is
not accurate enough such as jet flows and complex boundary layer flows.

16
3.3.3 The 𝒌 − 𝝐 Model

The standard 𝑘 − 𝜖 model is based on model transport equations for the turbulence
kinetic energy 𝑘 and its dissipation rate 𝜖. In this model, the flow is assumed to be
fully turbulent, so we can neglect the effect of molecular viscosity. Therefore, there
are some restrictions in it: the 𝑘 − 𝜖 model can only be used in turbulent flows. In
this model, according to Launder [24], the kinetic eddy viscosity is given by:
𝐶μ
𝜈𝑇 = 𝜈 (3.31)
𝜔
The turbulence kinetic energy is given by:

∂𝑘 ∂𝑘 ∂𝑈𝑖 𝜕 ∂𝑘
+ 𝑈𝑗 = 𝜏𝑖𝑗 −𝜖+ [(𝜈 + 𝜈𝑇 /𝜎𝑘 ) ] (3.32)
∂𝑡 ∂𝑥𝑗 ∂𝑥𝑗 𝜕𝑥𝑗 ∂𝑥𝑗

The specific dissipation rate is given by:

∂𝜖 ∂𝜖 𝜖 ∂𝑈𝑖 𝜖2 𝜕 ∂𝜖
+ 𝑈𝑗 = 𝐶𝜖1 𝜏𝑖𝑗 − 𝐶𝜖1 + [(𝜈 + 𝜈𝑇 /𝜎𝜖 ) ] (3.33)
∂𝑡 ∂𝑥𝑗 𝑘 ∂𝑥𝑗 𝑘 𝜕𝑥𝑗 ∂𝑥𝑗

And the closure coefficients and relations are given by:

𝐶𝜖1 = 1.44, 𝐶𝜖2 = 1.92, 𝐶𝜇 = 0.09, 𝜎𝑘 = 1.0, 𝜎𝜖 = 1.3 (3.34)

The 𝑘 − 𝜖 model is a popular model used in industrial applications for its fine
convergence and low expectation for memory. But it does not perform well in
computing flow fields that exhibit adverse pressure gradients, strong curvature.

3.4 Reynolds Stress Model (RSM)

Reynolds stress model, also called Reynolds stress transport (RST) model, is an
elaborate kind of turbulence model. In this model, Reynolds stresses are computed
directly, which is called second order closure, and the eddy viscosity method is
abandoned. RSM use differential transport equation to obtain the Reynolds stresses
[25]:

17
∂ ∂
̅̅̅̅̅̅
(𝜌𝑢 ′ ′
𝑖 𝑢𝑗 ) + (𝜌𝑢𝑘 ̅̅̅̅̅̅
𝑢𝑖′ 𝑢𝑗′ )
∂𝑡 ∂𝑥𝑘
𝜕 𝜕 ̅̅̅̅̅̅
𝜕𝑢 ′ ′
𝑖 𝑢𝑗
=− ̅̅̅̅̅̅̅̅̅
′ ′ ′ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
′ ′ ′
[𝜌𝑢𝑖 𝑢𝑗 𝑢𝑘 + 𝑝 (𝑢𝑖 𝛿𝑘𝑗 + 𝑢𝑗 𝛿𝑖𝑘 )] + [𝜇 ]
𝜕𝑥𝑘 𝜕𝑥𝑘 𝜕𝑥𝑘

∂𝑢𝑗 ∂𝑢𝑖
̅̅̅̅̅̅
− 𝜌 (𝑢 ′ ′
𝑖 𝑢𝑘 + ̅̅̅̅̅̅
𝑢𝑗′ 𝑢𝑘′ ) − 𝜌𝛽(𝑔𝑖 ̅̅̅̅̅ ̅̅̅̅̅
𝑢𝑗′ 𝜃 + 𝑔𝑖 𝑢 ′
𝑖 𝜃) (3.35)
∂𝑥𝑘 ∂𝑥𝑘

̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝜕𝑢𝑖′ 𝜕𝑢𝑗′ ̅̅̅̅̅̅̅̅̅̅
𝜕𝑢𝑖′ 𝜕𝑢𝑗′
+ 𝑝′ ( + ) − 2𝜇
𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑘 𝜕𝑥𝑘

̅̅̅̅̅̅̅
− 2𝜌𝜔𝑘 (𝑢 ′ ′ ̅̅̅̅̅̅̅
′ ′
𝑗 𝑢𝑚 𝜖𝑖𝑘𝑚 + 𝑢𝑖 𝑢𝑚 𝜖𝑗𝑘𝑚 ) + 𝑆

where S is the user-defined source term.

The Reynolds-stress tensor returns to isotropy when the anisotropy of these turbulent
components returns to isotropy and Reynolds stress transport equation can be used to
describe this correlation. RSM solves all components of turbulent transportation and
is the most general model of turbulence.

3.5 Large Eddy Simulation

As previously explained, RANS are time-averaged equations. Unlike RANS-based


model, Large Eddy Simulation is developed in space-average theory. Large Eddy
Simulation saves computational cost via ignoring the small (subgrid) length scales,
which are filtered by LES filter. LES can capture some unsteady states which RANS
cannot, such as large scale effect and coherent structure. Meanwhile, it saves vast
computational cost compared with DNS. Therefore, LES is considered to be the most
promising development direction in turbulent numerical simulation.

G(x) is assumed to be the filter function, and it is used to filter physical quantity y,
then y can be divided into a filtered portion y̅ and sub-filtered portion y ′ [26]:

18
𝑦̅(𝑟) = ∫ 𝐺(|𝑟 − 𝑟 ′ |)𝑦(𝑟 ′ ) 𝑑𝑟 ′ (3.36)

𝑦 = 𝑦̅ + 𝑦 ′ (3.37)

There are some filter commonly used for spatial filtering in LES such as Gaussian
filter and box filter.
Gaussian filter:

6 −6𝑥22
𝐺(𝑥) = √ ℯ ∆ (3.38)
𝜋∆

Box filter:
1 ∆ ∆
,− ≤ 𝑥 ≤
𝐺(𝑥) = {∆ 2 2 (3.39)

0, |𝑥| >
2
If the box filter, for example, is applied to Navier-Stokes equations, then the
following relations could be obtained:

∂𝑢𝑖 ∂𝑢
̅̅̅̅̅
𝑖 𝑢𝑗 1 ∂𝑝 ∂2 𝑢𝑖 ∂
+ =− +𝜈 − ̅̅̅̅̅
̅ 𝑢′ + ̅̅̅̅̅
(𝑢 𝑢̅𝑗 𝑢𝑖′ + ̅̅̅̅̅̅
𝑢𝑖′ 𝑢𝑗′ ) (3.40)
∂t ∂𝑥𝑗 𝜌 ∂𝑥𝑗 ∂𝑥𝑗 ∂𝑥𝑗 ∂𝑥𝑗 𝑖 𝑗
∂𝑢𝑖
=0 (3.41)
∂𝑥𝑖
LES falls midway between RANS and DNS models. It saves computational cost
compared to DNS model while improves precision compared to RANS model.
Though it ignores the small (subgrid) length scales, the mesh has to be fine enough to
capture the flow motion, and thus the main disadvantage of LES is the huge
computation time.

3.6 Vorticity Dynamics

High level of fluctuating vorticity is an important characteristic of turbulent flow.


Vorticity describes the local spinning motion of a continuum near some point, and it
is the curl of the velocity vector [9]:

19
𝝎 ≡ ∇×𝒖 (3.42)
This equation shows that vorticity is related to the deformation rate ∂𝑢i / ∂𝑥j .

The vorticity equation is obtained by taking the curl of the Navier-Stokes equations:

𝜕𝜔𝑖 𝜕𝜔𝑖 𝜕𝑢𝑖 𝜕 2 𝜔𝑖


+ 𝑢𝑗 = 𝜔𝑗 +𝜈 (3.43)
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗

In a three-dimension fluid flow, the first term on the right-hand side of vorticity
equation represents vortex stretching. Angular momentum conservation is the main
reason of the change of vorticity by vortex stretching. The angular momentum of a
material volume element would remain unchanged if viscous effect absent; if the
fluid element is stretched so that its cross-sectional area and moment of inertia
become smaller, the component of the angular velocity in the direction of stretching
must increase in order to conserve angular momentum.

But in a two-dimension flow, the flow velocity only has x and y component, so the
vorticity equation becomes:

𝜕𝜔𝑖 𝜕 2 𝜔𝑖
=𝜈 (3.44)
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑗

Thus, apart from the effects of viscosity, the vertical component of vorticity is
conserved on fluid particles. In particular, the effects of vortex stretching and
tilting are absent in two-dimensional flow.

The vortex stretching term is considered to be the most important mechanism in


turbulence analysis, as it represents the enhancement of vorticity by stretching and is
the mechanism by which the turbulent energy is transferred to smaller scales.

20
Chapter 4 Flow Past a Cylinder/Sphere
The analysis of flow past a cylinder/sphere, such as separation, reattachment and
vortex shedding, is an important and complicated problem in fluid dynamics.
Predicting the flow motion of the wake is meaningful in structural, offshore and
thermal powering engineering applications. There are several kinds of software are
used in computational fluid dynamics such as ANSYS FLUENT, ANSYS CFX,
OpenFOAM and etc. In this project, a CFD model based on FLUENT is developed
to simulate a flow past a cylinder/sphere.

Fluent is the most powerful and popular computational fluid dynamics software
currently. Fluid flows can be simulated using several kinds of models and used to
solve practical problems. The flow chart of ANSYS Fluent is:

Figure 4.1 Flow chart of FLUENT

21
4.1 Flow Past a Cylinder

In this case, a 2D model of flow past a cylinder is developed to analysis the influence
of dents on the pipeline. When the high pressure gas flow pass the dents on pipe wall,
the vortex shedding will be induced and there will be forces on the distorted part.
Hence, the pipe will vibrate because of those forces, which is called vortex induced
vibration. When the frequency equals to or is near the natural frequency of pipeline,
this vibration will cause huge damage to the pipeline in consequence of resonance.
So flow past a cylinder problem is an important research topic. In previous studies,
the drag force on cylinder was studied by many experiments. Drag force coefficient
is used to describe the relation between drag force and Reynolds number, where drag
force coefficient is defined as [27]:
𝐹
𝐶𝐷 = (4.1)
1 2
2 𝜌𝑢 𝐷𝑙
Here 𝐹 is the drag force, D is the diameter and l is the length of cylinder.

Figure 4.2 Drag force coefficient [28]

When Reynolds number is small, for example, less than several hundred, the flow
goes around the cylinder with a constant velocity. This kind of flow is called steady

22
flow. In steady flow, the time derivative can be neglected as it is usually considered
to be very slow. Thus, the governing equations of steady flow is
𝜕𝑢𝑖
=0 (4.2)
𝜕𝑥𝑖

From Figure 4.2, it can be observed that as the Reynolds number increases, for a
steady flow, the drag coefficient decreases. When the Reynolds number further goes
up, the flow becomes chaotic, which is called unsteady flow. For unsteady flow, the
drag coefficient increases and the Reynolds number goes up. In this Reynolds
number range, the properties, such as pressure and velocity, are dependent on time.
This means the time derivative cannot be neglected, and the flow motion is governed
by Navier-Stokes equations.

But when the Reynolds number is above 106, the drag coefficient drops by a large
factor, and in this situation the flow becomes fully turbulent. This causes a delay of
separation of flow and leads to a change of flow pattern. This phenomenon is called
drag crisis [29]. Then as the Reynolds number further increases, the drag coefficient
goes up again. Accordingly, the flow pattern changes with Reynolds number varies
as shown in Figure 4.3.

23
(a).

(b).

(c).

(d).

(e).

Figure 4.3 Flow pattern at different Reynolds number [30]

In Figure 4.3, (a) shows flow with quite low Reynolds number. In this situation, there
is no boundary layer along the surface and no viscous wake downstream. With
Reynolds number increase, the flow pattern is shown in (b) of Figure 4.3. A stable
pair of vortices is formed on the downwind side of the cylinder. The flow is
separated but steady and the vortices generate a high drag on the cylinder or sphere.
When Reynolds number approaches several hundred, shown in (c), the downstream
vortices become unstable, separate from the body and are alternately shed
downstream. The wake is very wide and generates a large amount of drag. The
alternate shedding is called the Von Kármán Vortex street which is introduced in
Section 2.4. (d) shows the flow as it is increased even more. The periodic flow
breaks down into a chaotic wake. The wake is not quite as wide as in the third picture,
so the drag is slightly less. (e) shows the flow at even higher Reynolds number. The

24
boundary layer transitions to chaotic turbulent flow with vortices of many different
scales being shed in a turbulent wake from the body.

In this section, turbulent models will be developed to simulate the flow pattern
variation at several Reynolds numbers.

4.1.1 Geometry

Figure 4.4 Geometry of 2D model

In this case, a long pipeline model is developed to simulate the vortex shedding
caused by dents on the pipe wall. The geometry is based on realities of situation, as
shown in Figure 4.4. The diameter of pipeline is 0.2842m and the length is 12m. And
a circular cylinder of d=0.02m is placed in the middle of pipeline. To catch the flow
motion of the wake and near the cylinder, the flow domain is separated into four
parts to refines the mesh, shown in Figure 4.5.

Figure 4.5 Partition of geometry

25
The left edge is set as the flow inlet and the right edge is set as flow outlet. The other
two edges are smooth pipe wall.

4.1.2 Mesh Set Up

The method of set up mesh can be divided into two kinds: structured mesh and
unstructured mesh. Structured mesh may be difficult to conform a certain block to a
complicated geometry, but it can improve the memory efficiency. As for
unstructured mesh, the most advantage is the flexibility to establish in a complex
geometry, which is commonly encountered in practical engineering applications.
Though the accuracy and speed of computation are not ideal, it has a great potential
in industrial applications. Arbitrary topology can be applied in unstructured mesh
such as quadrilaterals, hexahedra, triangles, and prisms. In this case study, both
structured mesh and unstructured mesh methods are employed. Structured mesh is
established near the cylinder to accurately simulate the flow motion, and the rest part
uses unstructured mesh to save convergence time. As it is mentioned earlier, the
geometry is divided into four parts. Face meshing is set up in the circular region
around the cylinder. Define the mesh size using number of divisions. The circular
edge is divided into 320 segments with no bias and the straight line connected
cylinder and the circular region is divided into 128 segments with bias growth rate of
1.015 to make regular face meshing as shown in Figure 4.6.

Figure 4.6 Mesh near the cylinder

26
In the wake region, a finer mesh is developed to catch the wake flow. The maximum
mesh size is set to 10-3 m as repeated simulations indicate that it can accurately catch
the turbulent wake. The final mesh is shown in Figure 4.7.

Figure 4.7 Mesh

4.1.3 Fluent Setup

In this case study, we use double precision to improve the accuracy. The type of
solver can be chosen as pressure-based and density-based. In general, pressure-based
solver is used for incompressible flow, and in this method pressure is obtained by
continuity and momentum equations; while the density-based is for compressible
flow, and the density of flow is obtained from the continuity equation and pressure is
solved by the equation of state [31]. In this case, pressure-based solver is employed
to simulate incompressible flow past a circular cylinder.

In this case study, the shear stress transport (SST) k-omega model and wall-adapting
local eddy-viscosity (WALE) model are employed to simulate the vortex shedding of
a two model. SST k-omega model is a variant of the standard k-omega model,
combining Wilcox k-omega model for near wall region and the standard k-epsilon
for free flow region. The WALE LES model is provided by ANSYS for
calculating 𝜇 𝑇 . The WALE LES model is more accurate for wall-bounded flow.

27
The material of flow is natural gas, which is a mixture of several kinds of gas. Here
methane is used as the material of fluid as it is the main component of natural gas.

As for boundary condition, the left edge is set as the flow inlet and right one is the
flow outlet. In practical, generally the mass flow rate of nature gas in the pipe is
fixed. In this case it is considered to be incompressible flow, so the entrance velocity
can be obtained. So here inlet is selected to be velocity-inlet, and gauge pressure is
set to 28×106 Pa, obtained from actual data of gas transmission pipeline. The out let
is set as pressure-outlet.

Table 4.1 Boundary conditions of 2D model


Velocity-inlet
Inlet
Gauge pressure: 28×106 Pa
Pressure-outlet
Outlet
Gauge pressure: 22×106 Pa
Cylinder Wall – no slip
Flow Domain Interior

Here the velocity of inlet is the control variable to show different patterns of flow. In
this case, Reynolds numbers of the flow equal 0.01, 100, 104, 106 and 107.
Accordingly, the velocities of inlet are 3×10-7, 3×10-3, 0.3, 30 and 300 meters per
second.

4.1.4 Results of Simulation

First results of LES model at different time are discussed as it is known that LES
model is relatively precise. Figure 4.8 shows the vortex shedding at different time.

28
Figure 4.8 Development of vortex shedding

From Figure 4.8, the development progress of vortex shedding can be observed. At
first, the vortices behind the cylinder are symmetrical, and the cavity which forms
during the starting-up process is closed and all the saddles are joined together. With
the time changing, the cavity is open and the saddles are not necessarily joined by the
separatrices, and instant “alleyways” of fluid penetrate the cavity. Figure 4.8 shows
the whole progress of development of Von Kármán vortex street.

Table 4.2 shows the velocity and vorticity contours at Re=10-2, 100, 104, 106 and 107.

29
Table 4.2 Velocity and vorticity contour - LES
Reynolds Number Velocity Contour Vorticity Contour

0.01

100

104

106

107

For LES simulation, it can be observed that the flow pattern varies vastly with the
change of Reynolds number. At Re=0.01, the flow passes the cylinder with a quite

30
low velocity, and the velocities are almost the same at the upstream and downstream.
In this case, viscous wake downstream of the cylinder can hardly be seen. When
Reynolds number reaches 100, a pair of vortices occurs behind the cylinder, and
velocity behind the cylinder decreases more than upwind side. Kármán vortex street
becomes significant when Reynolds number equals 104. The vortices become
unstable and asymmetrical, and are shed downstream. At Reynolds number equals
106, flow pattern changes a lot as drag crisis occurs. In this situation, chaotic wake is
initiated as the flow turns into leeward side of the cylinder. But as Reynolds number
further goes up, for example Re=107, Kármán vortex street becomes obvious again.

Then the results of SST model are demonstrated as shown in Table 4.3.

31
Table 4.3 Velocity and vorticity contour - SST
Reynolds Number Velocity Contour Vorticity Contour

0.01

100

104

106

107

32
From Table 4.3, it can be observed that the result of SST model is similar to result of
LES. At very low Reynolds number, the flow is symmetrical and viscous wake is
inconspicuous. As Reynolds number increases to 100, viscous wake appears. Then as
Reynolds number further goes up, we can see Kármán vortex street. When Reynolds
number approaches 106, flow pattern changes vastly because of drag crisis. At
Re=107, Kármán vortex street can be seen again.

There are some differences between the results of two models. From the contours, it
can be easily seen that the result of LES differs more from SST model at high
Reynolds number, as shown in Figures 4.9 and 4.10.

Figure 4.9 Reynolds number = 106

Figure 4.10 Reynolds number = 107

According to Figures 4.9 and 4.10, we can know that LES model can capture the
small eddy behind the cylinder but in SST k-omega model the flow behaves much
more like laminar flow at Re=106. At Re=107, though the vortex shedding occurs
again in SST k-omega model, it is much less significant than that in LES model. The

33
main reason of this difference is that no assumption is employed in LES model but
k-omega model uses two equations to compute the Reynolds stresses.

4.2 Flow past a sphere

Similarly, Kármán vortex street occurs when a flow past a sphere. But the flow
motion is slightly different because of the vortex stretching as previously discussed
in Section 3.6.

4.2.1 Geometry

In this case, a 10m×5m×5m cuboid flow domain is developed to simulate the flow
past a sphere. The diameter of sphere is 0.2 meter and the sphere is placed at the
center of the cuboid domain (Shown in Figure 4.11).

Figure 4.11 Geometry of 3D model

34
4.2.2 Mesh Set Up

Structural mesh is set up in this case to make the result more accurate. ICEM CFD is
a popular and powerful software package in ANSYS for mesh generation, editing
and smoothing.

Block meshing is employed in this case. The block is split into 27 parts, shown in
Figure 4.12.

Figure 4.12 Blocks

The mesh around the sphere and the wake region is refined to capture the vortex
shedding. The scan plane is shown as Figure 4.13.

Figure 4.13 Scan plane

35
Here the mesh is much finer around the sphere. Then the automatic smoothing is
used to improve the mesh quality.

4.2.3 Fluent Set Up

The pre-setup is the same as Section 4.1.3. Here LES model and pressure-based
solver are employed in this case.

Inlet velocity, the same as last case, is the control variable to show the flow pattern
changes at different Reynolds number. Boundary conditions are shown in Table 4.4.

Table 4.4 Boundary conditions of 3D model


Inlet Velocity-inlet

Gauge Pressure: 28×106 Pa


Outlet Pressure-outlet

Gauge Pressure: 22×106 Pa


Boundaries Wall – no slip
Flow Domain Interior

4.2.4 Results of Simulation

The velocity and vorticity contours are shown in Table 4.5.

36
Table 4.5 Velocity contours of 3D model
Reynolds number Velocity contour Vorticity contour

0.01

100

104

106

107

37
From the above results, it can be observed that the flow patterns in this case are
similar to 2D models, while there are some slight differences. It is similar that
Kármán vortex street cannot be seen at low Reynolds number but it becomes obvious
as Reynolds number increases. At Re=106, flow pattern also changes because of drag
crisis. Then when Reynolds number reaches 107, Kármán vortex street appears again.

As discussed in Section 3.6, vortex stretching exists in three-dimensional models


𝜕𝑢
because of the term 𝜔𝑗 𝜕𝑥 𝑖 . Due to this term, the area of a cross-section of a vortex
𝑗

tube, the surface in the continuum formed by all vortex-lines passing through a given
closed curve, becomes small, and vorticity would have to amplify proportionally.
The local stretching of a vortex tube gives rise to a proportional amplification of the
absolute value of vorticity.

38
Chapter 5 Leaked Pipeline Simulation
In Sections 5.1 and 5.2, the velocity profile and pressure drop in a pipe are elaborated.
But the situation in a leaked pipe is much more complicated. A number of studies of
leakage detection method have been done such as internally based leak detection
systems and externally based leak detection systems. In internally based leak
detection systems, pressure and strain are crucial parameters to detect location of
leakage. In this case study, a three-dimension leaked pipeline simulation is conducted
to investigate how the pressure and strain distribution change at different Reynolds
number.

5.1 Velocity Profile of Fluid Flow in Pipes

Fluid flow through pipe is commonly applied in practical engineering. In general, gas
pipelines are circular because circular pipes can withstand larger pressure differences
between the inside and outside. In practice, experimental results are obtained to
improve the solutions of problems of fluid flow in pipes. The equations of pipe flows
are relatively simple as the geometry prohibits the continuing growth of their
thickness. And if the pipe is long enough, the velocity profile is independent to
𝜕𝑢
location. As a result, the nonlinear inertia terms 𝑢𝑗 𝜕𝑥 𝑖 are suppressed. Thus it
𝑗

separates the surface layer – out layer problem from the problems associated with the
downstream development in other wall-bounded shear flows and the theoretical
analysis is considerably simplified.

39
5.1.1 Laminar Pipe Flow

If the Reynolds number is less than 2100, the flow is considered to be laminar flow,
characterized by smooth streamlines and highly ordered motion.

R
r

Figure 5.1 Velocity profile of laminar flow [32]

For a laminar flow, the velocity of flow of different point is related to the distance
between the point and midline of the pipe [32]:
𝑟2
𝑢(𝑟) = 2𝑈𝑎𝑣𝑔 (1 − 2 ) (5.1)
𝑅
The velocity profile is shown in Figure 5.1. Therefore, the velocity of laminar flow in
a pipe is parabolic, with a minimum at the pipe wall and a maximum at centerline.

5.1.2 Turbulent Pipe Flow

In most practical engineering problem, the flows are turbulent. But now, as a
complex mechanism dominated by fluctuations, turbulent flow problem remains
much undeveloped. Thus, the theory of turbulent flow through a pipe is much relied
on experiments.

40
Viscous sublayer

Buffer layer

R Overlap layer

r Turbulent layer
x

Figure 5.2 Velocity profile of turbulent flow [32]

The turbulent flow in a pipe can be divided into four regions characterized by the
distance from the pipe wall: viscous sublayer, buffer layer, overlap layer and
turbulent layer (Figure 5.2). Velocity profile of fully developed turbulent pipe flow is
much flatter than laminar flow when the flow across a section, the simplest and the
best-known approximation is the power-law velocity profile expressed as [32]:
𝑢 𝑟 1
= (1 − )𝑛 (5.2)
𝑢𝑚𝑎𝑥 𝑅
where n is a constant depends on Reynolds number.

5.2 Geometry

In this case, according to realities of situation, a 12-meter-long pipe model is set up


with inner diameter equals 0.2842m. The thickness of pipe wall is 0.0103m. And
there is a leakage hole at the middle of pipe, whose diameter equals 0.1m. As the
pressure at the leakage point is expected to obtain in this simulation, so a 6m long
circular cylinder leakage domain is set beyond the pipe. The geometry is shown as
Figure 5.3.

41
Figure 5.3 Geometry of leaked pipe

5.3 Mesh Set Up

In this case, ICEM is also employed to set up model mesh. The block is split into
three parts according to the distance from the leakage hole (Figure 5.4).

Figure 5.4 Blocks

In the middle part the mesh size is smaller than the other two as it is where the
leakage occurs. Here automatic o-grid mesh is used. O-grid is a kind of structured

42
hex/quad grid which is commonly used in circular geometry [33]. An o-grid has lines
of points wraps around and meets the first point (Figure 5.5).

Figure 5.5 O-grids meshing

The mesh scan plane is shown in Figure 5.6.

Figure 5.6 Scan plane

The mesh near leakage hole is set to be finer as we are more concerned about the
flow near the hole. Because the pipe model is long enough, it is reasonable to assume
that the flow upstream is a steady flow.

43
5.5 Fluent Set Up

In this case, transient pressure-based solver, the same as last case, is used. As for
viscosity model, LES is employed. This simulation shows the pipe condition at
Re=0.1, 100, 4500, 107. Here we change the viscosity to simulate the condition of
pipe at different Reynolds number. Accordingly, the viscosity is set to 6567, 6.567,
0.1459, 6.567×10-5.

The left surface is set to be the velocity inlet, where the velocity is set to 300 meters
per second and the gauge pressure to 28×106 Pa. The pipe outlet and leakage outlet
is set to pressure outlets where the gauge pressure is 22×106 Pa and 0 Pascal
respectively as shown in Table 5.1.

Table 5.1 Boundary conditions of leaked pipe model


Velocity-inlet
Inlet
Gauge Pressure: 28×106 Pa
Pressure-outlet
Outlet
Gauge Pressure:22×106 Pa
Pressure-outlet
Leakage hole
Gauge Pressure: 0 Pa
Pipe wall Wall- no slip
Flow Domain Interior

5.6 Results

5.6.1 Results of FLUENT

The contours of pressure and velocity are shown in Tables 5.2 and 5.3.

44
Table 5.2 Pressure
Reynolds Number Pressure

0.1

100

4500

107

45
Table 5.3 Velocity streamline
Reynolds Number Velocity Streamline

0.1

100

4500

107

46
From Tables 5.2 and 5.3, it can be seen that the velocity of flow is uniform at low
Reynolds number, and it is higher at the central region than near the wall. As
introduced in Section 5.1, the velocity profile of a laminar pipe flow is parabolic.
Velocity reaches its maximum at the central line and becomes zero near the wall.
Meanwhile, it is known that the force on pipe wall is related to the vertical velocity
component of flow. For a laminar flow, the flow is steadier than turbulent flow,
which means it has a more uniform velocity towards downstream. Due to the
uniform velocity, pressure distribution is relatively even. But as the Reynolds
number increases, velocity component of flow near the leakage hole raises. And in
this situation the flow becomes turbulence. Compared with laminar flow, turbulent
flow is more chaotic and unordered, which means when the leakage occurs, the
vertical velocity component of flow is much higher. As a result, pressure near the
leakage hole becomes much higher than other area, and reaches its maximum at the
leakage hole.

5.6.2 Results of Fluid Structure Interaction

Fluid structure interaction (FSI) is an important consideration in engineering. FSI


shows how the structure, stable or oscillatory, interacts with the fluid flow. Here
Static Structure module in ANSYS is used. Static Structure is used to determine the
displacements, stresses and strains. In Static Structure, pressure data is imported
from FLUENT.

First the elastic strain is shown in Figure 5.7 at Reynolds number equals 107 and at
0.2s, 0.5s and 0.8s.

47
(a)

(b)

(c)
Figure 5.7 Elastic strain at (a) t=0.2s, (b) t= 0.5s, and (c) t=0.8s

From Figure 5.7, it can be observed that the elastic strain increases with time. And at
0.8s the flow becomes steady and the strain reaches the maximum. At the beginning
of leakage, the pipe flow is still steady, so the deformation of pipe is inconspicuous
and elastic strain is small. But as time varies, the flow becomes more and more
chaotic, the pressure on pipe is getting larger and larger. As a result, the pressure
deforms the pipe and makes elastic strain increase. The point of maximum is near to
the leakage point where the velocity reaches maximum.

48
Elastic strain distribution at different Reynolds number is shown in Table 5.4.
Table 5.4 Elastic strain
Reynolds
Elastic Strain
Number

0.1

100

4500

107

49
Form Table 5.4, it is observed that the equivalent elastic strain changes with the
variation of Reynolds numbers. At Re=0.1, the strain distribution is relatively
uniform because the pressure is even. Meanwhile, the region of the strain is small,
which means the pipe undergoes comparatively affordable condition. But as
Reynolds number goes up, the area of strain enlarges and the elastic strain becomes
uneven. In this situation, the pipe should be designed to sustain the strain and
deformation. When Reynolds number reaches 100, the region of elastic strain is
obviously increased because the pressure is larger near the leakage hole than Re=0.1.
At Re=4500, the flow becomes fully turbulent and it is easily observed that the
elastic strain distribution is much more nonuniform. This means the pipe must bear
more deformation when leakage occurs. When Reynolds number further goes up, this
situation becomes more worse. Among those cases, Re=107 is the worst one as the
elastic strain distribution is much uneven and the pipe wall sustains huge
deformation.

50
Chapter 6 Conclusions and Future Works

6.1 Conclusions

6.1.1 Influence of Dents

In the first case study, a two-dimension model and a three-model model are
conducted to simulate how the dents influence flow motion in pipe. When the flow
past the dents, Kármán vortex street occurs. This phenomenon may cause drag force
on the pipe wall and eventually leads to leakage. In this case, it is found that the drag
coefficient and flow pattern are related to Reynolds number. For a fluid flow with
high viscosity, the force on pipe wall is symmetrical but the drag coefficient is
relatively high. As the viscosity decreases, the Reynolds number goes up. When
Reynolds number reaches several thousand, the flow turns to turbulent flow, and
vortex shedding phenomenon becomes obvious. The pipe wall is under asymmetric
and time-dependent force. Thus the vibration, known as vortex induced vibration
(VIV), occurs on the pipeline. When the Reynolds number approaches 106, there is a
sudden drop of drag force coefficient called “drag crisis”. And the flow pattern
considerably changes. As the Reynolds number further goes up, the drag force
continues to increase.

6.1.2 Pressure, Deformation and Strain of Leaked Pipe

In this case study, a LES leaked pipe model is conducted to study the pressure,
deformation and elastic strain on the pipe wall as the Reynolds number changes.
From the simulation results, it is found that the pressure reached maximum near the
leakage hole. And the elastic strain is up to its maximum as well. On the other hand,
it can be seen that the pressure and elastic strain increase as the Reynolds number
increase. In practice, the high-pressure pipelines transport natural with high density

51
and velocity, which means that the Reynolds number is normally high. Therefore
once the leakage occurs, the pipe should withstand great pressure and strain and the
sensor needs to detect the deviant changes on pipe or greater catastrophes might
happen.

6.2 Future Works

First of all, as mentioned in Section 3.4, the accuracy of LES model is related mesh
size. Limited to calculative ability of computer, the results of three-dimensional
model in this project are not perfect due to the mesh size. If the mesh is set to be
finer, the simulation of three-dimensional turbulence model needs great computation
capacity of workstation. In the future, a finer mesh can be established to get more
accurate simulation results.

Furthermore, this dissertation does not discuss the frequency of acoustic emission
(AE), which can be simulated by harmonic response in ANSYS. Frequency of
acoustic emission is another important and popular parameter of leakage detection
technology. This kind of leakage detection is required only locally for mounting AE
sensors. AE detection has great advantage of its convenience and precision. In future
works, acoustic emission should be a significant topic of pipeline models.

As the author is not from fluid dynamic background, this dissertation has a lot of
deficiency and criticism is welcomed and appreciated.

52
References
[1] A. Finenko and L. Cheah, “Temporal CO2 emissions associated with electricity
generation: Case study of Singapore,” Energy Policy, vol. 93, pp. 70–79, 2016.
[2] H. Mishra, P. Karkulali, A. Ukil, and J. Dauwels, “Testbed for real-time
monitoring of leak in low pressure gas pipeline,” IECON 2016 - 42nd Annual
Conference of the IEEE Industrial Electronics Society, 2016.
[3] R. S. Reddy, G. Payal, P. Karkulali, M. Himanshu, A. Ukil, and J. Dauwels,
"Pressure and flow variation in gas distribution pipeline for leak detection,"
2016 IEEE International Conference on Industrial Technology (ICIT), 2016.
[4] L. Wang, X. Gao, and T. Liu, “Gas pipeline small leakage feature extraction
based on LMD envelope spectrum entropy and PCA-RWSVM,” Transactions of
the Institute of Measurement and Control, vol. 38, no. 12, pp. 1460–1470, 2015.
[5] (2017, Mar 27) Wikipedia [Online]. Available:
https://en.wikipedia.org/wiki/Olympic_Pipeline_explosion
[6] S. Then, (2014, Jun 11) The Star Online [Online]. Available:
http://www.thestar.com.my/news/nation/2014/06/11/blast-rips-sabahsarawak-ga
s-pipeline-no-one-hurt-in-2am-explosion-fire-is-out/
[7] D. Shantanu and S. Sarkar, "A review on different pipeline fault detection
methods," Journal of Loss Prevention in the Process Industries, vol. 41, pp.
97-106, 2016.
[8] S. M. Yu, K. Lim, W. Chao, and X. Goh, "Mixing enhancement in subsonic jet
flow using the air-tab technique", AIAA Journal, vol. 46, no. 11, pp. 2966-2969,
2008.
[9] H. Tennekes and John L. Lumley, A First Course in Turbulence. Cambridge,
MA: MIT Press, 1985.
[10] Y. Egorov, F. R. Menter, R. Lechner, and D. Cokljat, "The scale-sdaptive
simulation method for unsteady turbulent flow predictions. Part 2: Application

53
to complex flows," Flow, Turbulence and Combustion 85, no. 1, pp. 85-139,
2010.
[11] (2014, Oct 21) Wikipedia [Online]. March 29, 2017. Available:
https://en.wikipedia.org/wiki/K%C3%A1rm%C3%A1n_vortex_street
[12] V. Strouhal, “Ueber eine besondere Art der Tonerregung,” Annalen der Physik
und Chemie, vol. 241, no. 10, pp. 216–251, 1878.
[13] R. D. Blevins, Flow-Induced Vibration. Malabar: Krieger publishing, 2001.
[14] A. Roshko, “Experiments on the flow past a circular cylinder at very high
Reynolds number,” Journal of Fluid Mechanics, vol. 10, no. 3, p. 345, 1961.
[15] R. E. D. Bishop and A. Y. Hassan, “The Lift and Drag Forces on a Circular
Cylinder Oscillating in a Flowing Fluid,” Proceedings of the Royal Society A:
Mathematical, Physical and Engineering Sciences, vol. 277, no. 1368, pp. 51–
75, Jul. 1964.
[16] M. Gaster, “Vortex shedding from circular cylinders at low Reynolds numbers,”
Journal of Fluid Mechanics, vol. 46, no. 04, p. 749, 1971.
[17] Y. C. Fung, Foundations of Solid Mechanics. Taipei: Pe Men Book Co., 1985.
[18] M. C. Potter and D. C. Wiggert, Schaum's Outline of Fluid Mechanics. New
York: McGraw Hill, 2008.
[19] S. R. Groot and P. Mazur, Non-Equilibrium Thermodynamics. Amsterdam:
Dover, 2011.
[20] J. F. Wendt and J. D. Anderson, Computational Fluid Dynamics: An
Introduction. Berlin: Springer, 2009.
[ 21] S. B. Pope, Turbulent Flows. Cambridge: Cambridge Univ. Press, 2015.
[22] D. C. Wilcox, “Reassessment of the scale-determining equation for advanced
turbulence models,” AIAA Journal, vol. 26, no. 11, pp. 1299–1310, 1988.
[23] D. Wilcox, “Formulation of the k-omega turbulence model revisited,” 45th
AIAA Aerospace Sciences Meeting and Exhibit, 2007.
[24] B. Launder and D. Spalding, “The numerical computation of turbulent flows,”

54
Numerical Prediction of Flow, Heat Transfer, Turbulence and Combustion, pp.
96–116, 1983.
[25] K. Hanjalic and B. E. Launder, “A Reynolds stress model of turbulence and its
application to thin shear flows,” Journal of Fluid Mechanics, vol. 52, no. 04, p.
609, 1972.
[26] P. Sagaut and M. Germano, Large Eddy Simulation for Incompressible Flows:
An Introduction: Third Edition. Berlin: Springer, 2006.
[27] B. W. McCormick, Aerodynamics, Aeronautics and Flight Mechanics. New
York: Wiley, 1995.
[28] B. Fornberg, “Steady viscous flow past a sphere at high Reynolds numbers,”
Journal of Fluid Mechanics, vol. 190, no. 1, p. 471, 1988.
[29] H. Schlichting, Boundary Layer Theory. New York: McGraw-Hill, 1955.
[30] T. Bohr, M. H. Jensen, G. Paladin, and A. Vulpiani, Dynamical Systems
Approach to Turbulence. Cambridge Univ. Press: Cambridge, 1998.
[31] M. Salas, N. J. Hefner, and L. Sakell, Modeling Complex Turbulent Flows, 1st
ed. Springer, 1999.
[32] R. A. Granger, Fluid Mechanics. New York: Dover, 1995.
[33] J. Tu, K. Inthavong, and G. Ahmadi, Computational Fluid and Particle
Dynamics in the Human Respiratory System, 1st ed. Dordrecht: Springer, 2013.

55

S-ar putea să vă placă și