THESIS FOR THE DEGREE OF LICENTIATE OF

ENGINEERING
in
Thermo and Fluid Dynamics

Large-Eddy Simulation of the

Flow around Simplified
High-Speed Trains under Side
Wind Conditions
H ASSAN N ASR H EMIDA

Department of Applied Mechanics

CHALMERS UNIVERSITY OF TECHNOLOGY
Goteborg, Sweden, 2006

Large-Eddy Simulation of the Flow around Simplified HighSpeed Trains under Side Wind Conditions
H ASSAN N ASR H EMIDA

c H ASSAN N ASR H EMIDA, 2006

THESIS FOR LICENTIATE OF ENGINEERING no 2006:06

ISSN 1652-8565

Department of Applied Mechanics

Chalmers University of Technology
S E-412 96 Goteborg, Sweden
Phone +46-(0)31-7721400
Fax: +46-(0)31-180976

Printed at Chalmers Reproservice

Goteborg, Sweden 2006

Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side
Wind Conditions
by

Hassan Nasr Hemida

[email protected]
Division of Fluid Dynamics
Department of Applied Mechanics
Chalmers University of Technology
SE-412 96 Goteborg
Sweden

Abstract
The new generation of high-speed trains, which are characterized
by light weight, are at high risk of overturning or derailment when
cruising in a strong side wind. With the introduction of high-speed
trains traveling at 250 km/hr and more, the likelihood of such an accident has increased. However, prevention of the unwanted influences
of a side wind requires that the flow structures on the train surface
and around it are fully understood in both the instantaneous and the
time-averaged flow.
In this work, large-eddy simulation (LES) using the standard Smagorinsky sub-grid scale (SGS) model with model constant
 
is used
to compute the side-wind flow around stationary generic train models. The side-wind flow is obtained at three different yaw angles,   ,
  and   . LES gives results that agree well with the experimental
observations and displays many flow features that were not observed
in the experiments. The flow around high-speed trains involves many
small structures due to the instabilities of the shear layers that contribute to the time varying values of the forces around the train. In the
results of LES, we have studied the time-dependent behavior of these
structures and their impact on the train surface pressure and thus on
the aerodynamic coefficients. The influence of the train nose shape on
the flow field under side-wind conditions is investigated. Our LES results show that the train nose shape determines to a large extent the
characteristics of the wake structures and their associated frequencies.
The investigations conducted in the present work focus on the wake
structures and their influence on the aerodynamic coefficients.

iii

iv

Acknowledgments
This work would not have been possible without the support of a number of organizations and individuals. The research was carried out
at the Division of Fluid Dynamics, Department of Applied Mechanics, Chalmers University of Technology. This work was supported by
the Swedish Agency for Innovation System (VINNOVA), Bombardier
Transportation, and Scania. Computer time on HPC2N (High Performance Computing Center North), NSC (National Supercomputer Center in Sweden), HIVE supercomputer at SWEGRID, and HELIOS supercomputer provided by UNICC at Chalmers is gratefully acknowledged.
I would like to express my sincere gratitude to my supervisor, Dr.
Sinisa Krajnovic, for sharing some of his profound knowledge in computational fluid dynamics with me. This thesis would never have reached
this point without his enlightening discussions and brilliant advice.
I express my sincere appreciation to my examiner, Professor Lars
Davidson, who believed in me and introduced me to the research field
and gave me the opportunity to undertake the PhD program in his
group. Thanks Lars for all good advice and encouragement.

I would like to thank Dr. Hakan

Nilson for his time and wonderful
help and assistance in solving some CFD code-related problems.
My thanks to all of my colleagues, especially Darioush Gouhari,
Kristian Wallblad and Walter Gyllenram for the helpful discussions
and fixing my computer-related problems.
The interesting and simulating discussions that arose at the steering group meetings are gratefully acknowledged, and my thanks go to
Per Johnson and Stephen Conway from Scania and Ben Diedrichs from
Bombardier Transportation.
My former examiner, Professor Essam Hamdi, head of the Electrical Machines and Derive Systems group, was the first to grant me the
chance to be a student at Chalmers. Thank you Essam for your understanding and for the help and the support you always offer.
My warmest and deepest sense of gratitude goes to my family, Fainan,
Mayar and Momen, for their love and support.
v

vi

Nomenclature
Upper-case Roman









 
 
 
 

!
" 
"$#
"$%
&
'
(
(*)
+ ,
+./

second order tensor

drag force coefficient
lift force coefficient
side force coefficient and Smagorinsky model coefficient
local pressure coefficient
train height
drag force
lift force
side force
Jacobian tensor
train length
pitching moment
rolling moment
yawing moment
first invariant of the velocity gradient
second invariant of the velocity gradient
third invariant of the velocity gradient
Reynolds number
strain rate tensor
Strouhal number
velocity

Lower-case Roman

0 2 1435
6
7
-:9
;

vectors in space
constant in Eq.3.16
frequency
pressure
time
dimensionless time unit
Cartesian components of velocity vector
vii

< 1>=?1A@

Cartesian coordinate vector components

Upper-case Greek

B C 

filter width and matrix discriminant

time step
rotation tensor

Lower-case Greek

DE
F
H
M

G
I
RI Q

S  
T 

Kronecker delta
angle from in a train cross-section
tensor eigenvalues
viscosity
kinematic viscosity J I  HLKNMPO
turbulent kinematic viscosity
density
viscous stress tensor
subgrid scale stress tensor

Subscripts

turbulent quantity
initial condition
free stream or ambient conditions

Superscripts
SGS

subgrid scale
spatially filtered quantity

Abbreviations
CFD
CFL
LES
MPI
RANS

Computational Fluid Dynamics

Courant-Friedrichs-Levy
Large Eddy Simulation
Message Passing Interface
Reynolds Averaged Navier-Stokes

viii

Contents
Abstract

iii

Acknowledgments

Nomenclature

vii

1 Introduction
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Aerodynamic coefficients . . . . . . . . . . . . . . . . . . .
1.3 Flow Structures Around Trains . . . . . . . . . . . . . . .

1
1
3
4

2 Governing Equations and Numerical Method

2.1 Governing Equations . . . . . . . . . . . . . . . . . . . . .
2.2 Numerical implementations . . . . . . . . . . . . . . . . .
2.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . .

7
7
9
9

3 Visualization Techniques
3.1 Second Order Tensor Characteristics . . . . . . . . . . .
3.1.1 Rotation Tensor . . . . . . . . . . . . . . . . . . . .
3.1.2 Eigenvalues and Eigenvectors of a Tensor . . . . .
3.1.3 Principal Invariants of a Tensor . . . . . . . . . . .
3.2 Critical Point Theory . . . . . . . . . . . . . . . . . . . . .
3.3 Streamlines and Path-lines . . . . . . . . . . . . . . . . .
3.4 Vortex Core . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.1 Region of Constant Vorticity Magnitude . . . . . .
3.4.2 Region of Complex Eigenvalues of Velocity Gradient Tensor . . . . . . . . . . . . . . . . . . . . . . .
3.4.3 Region of Positive Second Invariant of Velocity Gradient . . . . . . . . . . . . . . . . . . . . . . . . . .
G?W . . . . . . . . . . . . . . . . .
3.4.4 Region of Negative
3.4.5 Region of Minimum Local Pressure . . . . . . . . .

11
11
11
12
12
13
16
16
16

ix

17
18
18
18

4 Side-Wind Flow over a Simplified ICE2 Train Model

4.1 Train model and computational domain . . . . . . . .
4.2 Boundary conditions . . . . . . . . . . . . . . . . . . .
4.3 Numerical details . . . . . . . . . . . . . . . . . . . . .
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.1 Time-averaged flow . . . . . . . . . . . . . . . .
4.4.2 Aerodynamic forces . . . . . . . . . . . . . . . .
4.4.3 Instantaneous flow . . . . . . . . . . . . . . . .
Bibliography

.
.
.
.
.
.
.

.
.
.
.
.
.
.

19
19
21
21
23
24
31
32
35

Chapter 1
Introduction
1.1

Motivation

When a high-speed train cruises in a side wind, the effective side wind
is strong even if the magnitude of the side-wind velocity is low. The
effective side wind is the result of the train speed and the side-wind
speed. The angle between the effective side-wind direction and the
train cruising direction is the yaw angle, as shown in Fig. 1.1. The
value of this angle is determined by the train cruising speed, the sidewind speed and the side-wind direction. Normally, side wind yaw angles are below XY owing to relatively low-speed side winds compared
to the train speed. However, it is possible to find high-speed trains that
are cruise at larger yaw angles, e.g. when they exit from tunnels or
when a strong side wind has a strong component in the direction of the
train motion.
train speed
wind speed
Train cruising direction yaw angle
effective cross-wind speed

Figure 1.1: Effective side wind

The new generation of high-speed trains is light in weight in order to provide high acceleration and to reduce the energy necessary
to overcome gravity and friction forces. When these trains cruise in a
strong side wind, there is a stagnation region with high pressure on
the streamwise face while a region of low pressure is formed on the lee
1

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions

Low pressure

High pressure

Side force
Weight

Lift force

Turning moment

Figure 1.2: Aerodynamic forces and moments due to side wind

side due to the recirculation regions in the wake flow. In addition, the
flow moves over the train roof at high velocity, making a region of low
pressure on the upper side of the train. Due to these pressure differences, the high-speed trains experience strong aerodynamic forces and
moments such as side and lift forces and a yaw moment, as shown in
Fig. 1.2. In such conditions, the high-speed trains are at high risk of
overturning or derailment.
It is well known that the shape of the vehicle determines the influence of the side winds on the vehicle stability. Thus the vehicles side
wind resistance can be improved by adjusting its design. Previous experimental studies have concentrated mainly on the vehicles response
to the side wind and not much on understanding the flow mechanisms.
Prevention of the unwanted influences of a side wind requires that flow
structures on the train surface and around it are fully understood in
both the instantaneous and the time-averaged flow. Unsteady flow past
a train in a side wind has been the object of numerous experimental
investigations [1, 2, 3, 4, 5, 6, 7]. The primary interest in these investigations was the measurement of some integral parameters such as
drag, lift and side force coefficients together with measurements of the
natural wind characteristics around the train. Some other researchers
have investigated the flow structures numerically [8, 9] to obtain a better understanding of the flow behavior. Most of these studies are based
on Reynolds-averaged Naiver-Stokes (RANS) equations or time varying RANS (URANS). Since the side-wind instability is a consequence
of the unsteadiness of the flow field around the train, an accurate timedependent solution is necessary. Diedrichs [10] made a comprehensive
2

CHAPTER 1. INTRODUCTION
literature review of computational methods for cross-wind stability of
railway trains. His study showed that most of the previous side-wind
investigations are based on the time-averaged solution. Here only the
mean flow is explored and the instantaneous information is lost.

1.2

Aerodynamic coefficients

As mentioned before, when high-speed trains cruise in a side wind they

experience aerodynamic forces and moments. The aerodynamic forces

are the drag force,  , which resists the forward motion of the train,

lift force,  , which acts upward and tends to raise the train of the rail,

and the side force,  , which pushes the train from the side. The latter
is a result of the side wind. The aerodynamic moments that arise due
"Z# , the pithing moment, "  , and
to side wind are the rolling moment,
Z
"
%
the yawing moment,
. Figure 1.3 shows the different aerodynamic
forces and moments on a train subjected to a side wind. The magnitude
of these forces and moments depends on the Reynolds number and the
shape of the train.

[
a)

_
\

] 

b)

^ 

^ 

] #

_
\

] %
c)

^ 

^ 
Side-wind direction

Figure 1.3: Forces and moments acting on trains under the influence
of a side wind.
It is useful to have a simple means to compare the aerodynamic
forces produced by different train shapes regardless of their size or
driving speed. This is conveniently provided by factors called aerodynamic coefficients, which mainly depend on the shape of the train and
3

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
side wind direction. The drag coefficient,
` , the lift coefficient,
the side force coefficient,
 , are defined as:

 , and

a 8 e
 d 1

c
b
W
f

 W M /.g 
 d

(1.1)

a 8  %

h Wf M /.i g b W  %

(1.2)

and

a 8 kj

A Wf M /i g b W e
(1.3)
 j
kd ,  % and kj are the projected
and
where 8  is the surface= pressure,
@
surface areas in the < , and directions, respectively.

In certain flow regimes, the effect of Reynolds number on the aerodynamic coefficients can be quite pronounced. The coefficients can
change dramatically over the Reynolds number region, which encompasses transition from laminar to turbulent boundary layers. However, in other flow regimes, the effect of the Reynolds number is much
weaker. As a result, aerodynamic coefficients are often considered to
be rather weak functions of Reynolds number and a strong function of
yaw angle.
It is convenient to express the pressure distribution in terms of the
pressure coefficient,
l , which is defined as:

8 V 8 gW

k Wf M /.g
(1.4)
/lg is the free-stream velocity and
where 8 is the local static pressure,
8 g is the free-stream static pressure. The difference between the local
g
static pressure, 8 , and the free-stream static pressure, 8 ,W depends diA
/
g . Therefore,
rectly on the dynamic pressure of the free stream,
K mM
the pressure coefficient remains constant at all speeds.

1.3

Flow Structures Around Trains

Copley [5] investigated the influence of Reynolds number on the flow

field around a train when it is subjected to a cross wind. He found
that, for very low Reynolds number, the flow around the train is laminar everywhere and remains attached to the train surface. Increasing
4

CHAPTER 1. INTRODUCTION
the Reynolds number results in a separation on the roof-side streamwise edge and the detachment of the flow from the face, as shown in
Fig. 1.4.a. At the same time, the wake begins to shed vortices in a regular fashion. A further increase in Reynolds number makes the flow
fully turbulent in the wake and in the separation region over the roofside face. At some critical Reynolds number, the flow reattaches to the
face, forming a separation bubble, as shown in Fig. 1.4.b. For higher
Reynolds number, this separation bubble is suppressed and the flow
over the roof-side face is fully attached, as shown in Fig. 1.4.c. However, although the influence of a train nose shape is believed to play an
important role in the flow structures and aerodynamics of high-speed
trains and influences the value of the Reynolds number at which the
flow becomes fully attached, it has not been properly investigated in
previous studies.

a)

b)

c)

Figure 1.4: Reynolds number effect on the flow structures around train
under cross-wind conditions.
Numerical and experimental studies of the flow around bluff bodies
[11, 12, 13, 14, 15] show that such flows are highly unsteady and threedimensional. Moreover, when the Reynolds number is high enough
to obtain a fully turbulent flow in the shear layers and in the wake,
two main instability modes are present in the wake. The first is the
large-scale shedding in the wake, and it manifests itself as a progressive wave motion with alternate fluctuations produced by the shear
present at the limit between the recirculation zone and the exterior
fluid. These fluctuations determine the periodic shedding of the vortices that form behind the bluff body. The second high frequency mode
is so called spiral mode, which is associated with the small-scale shearlayer Kelvin-Helmholtz instability on the periphery of the recirculation
region where shear flow is present, as shown in Fig. 1.5. This instability is responsible for the distortion of the large vortex structures, shedding of the vortex tubes in a quasi-coherent fashion inside the detached
shear layers, production of the small scales, and, eventually, transition
to turbulence in the wake (for more detail see [15]).
Capturing these structures and the investigating the flow features
5

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
Shear-layers instabilities
Vortex tubes

a)

b)

Figure 1.5: Cross-section of the LES flow around trains in a cross-wind

showing the shear-layer instabilities colored by (a) the instantaneous
static pressure and (b) the magnitude of the instantaneous velocity vector.
require simulation methods that resolve, rather than model, the main
dynamics governing the flow. Numerical solutions using URANS are
not able to catch these high frequency modes since the Reynolds stresses
representing the turbulence in URANS are modeled with a turbulence
model. Finding a true representation of the flow is dependent on the
turbulence model used. Moreover, it is difficult to define a model that
can accurately represent the Reynolds stresses in the region of separated flow, such as the wake behind a train. The discretization schemes
play important roles in determining these structures. Since the high
frequency modes in the train boundary layers and shear layers are important in side-wind simulations, the convective flux, the viscous diffusion and the sub-grid fluxes should be approximated by a second-orderaccurate central difference scheme to reduce the numerical dissipation
that arises from using upwind schemes. This dissipation suppresses
the formation of the small eddies and hence the formation of high frequency modes.

Chapter 2
Governing Equations and
Numerical Method
Since side-wind stability is a consequence of flow unsteadiness, an understanding of the instantaneous flow is crucial to understanding of
flow around high-speed trains. Thus a time-dependent method, such
as LES, should be used in order to provide instantaneous information
about the flow. LES has already been proven to be a reliable technique
in the prediction of flows around simplified road vehicles [16, 17]. Until
recently, it was impossible to numerically predict the time-dependent
flow around a train model even at moderate Reynolds numbers. In
recent years, the increase in computer capability has made these simulations possible with LES.
It is worth mentioning that complicated flow structures develop in
the wake region behind trains. These wake structures are dominated
by large turbulent structures. Most of these structures are resolved
using LES, and only the influence of the scales smaller than the computational cells is modeled. Although LES is computationally more
expensive than RANS, it can provide more accurate time-averaged results and also give information on the instantaneous flow that is out of
the reach of RANS. This chapter presents the governing LES equation
and the subgrid-scale model used in the simulation. This is followed by
a presentation of the numerical scheme used and the boundary conditions.

2.1

Governing Equations

In LES, the large eddies are computed directly and the influences of
the small-scale eddies on the large-scale eddies are modeled. The incompressible Navier-Stokes equations and the continuity equation are
7

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
filtered using an implicit spatial filter. The resulting filtered equations
nAo
n
n4o
n o
n
are:

W
n ;  n ;o  ;o  O
n
8 I n ;  n T 
-qp <  J
 V M < p
<< V <

(2.1)

nAo
n;
  
< r

(2.2)

and

o o

Here, ; and 8 are the resolved filtered velocity and pressure, respec 
 

tively, while T  ; ; V ; ; are the sub-grid scale (SGS) stresses. The
sub-grid scale stresses represent the contribution of the small unresolved scales to the large resolved ones. These are unknown and must
be modeled. The Smagorinsky model is used in the present work to
model the SGS stresses because of its simplicity and in order to reduce
computational cost. It models the SGS stresses as:

where

+ 

T , V
ED , tT s5s  V m I v ui + ,

is the resolved rate of strain defined as:

w
+  
n
m
and

I vui

(2.3)

nxo
xn o
;
n;
<  p < R y

(2.4)

is the SGS viscosity defined as: o

7 B O W~} + } 
I
z

i
u
l

{

|
J




o
o o
} + } +  +  f| W
O ,
 is the SGS model coefficient and 7
 Jm
{
Here,

van Driest damping function that has the form:

w =
m y 

7 
V ) <Y8

(2.5)
is the

(2.6)

The value of the SGS model coefficient,


 , is 0.1 in the present



+
, does not vanish at the
work. Since the resolved strain-rate tensor,
wall, the value of the coefficient,
 , should be adjusted to take partially into account the effect of the wall. Thus, the damping function,
7 , is used to damp the turbulence length-scale next to the wall. The
same model constant has been used in simulations of
the flow around
B
simplified road vehicles [16, 17]. The filter width, , is taken as the
cubic root of the volume of a finite volume cell.
8

CHAPTER 2. GOVERNING EQUATIONS AND NUMERICAL

METHOD

2.2

Numerical implementations

Two computations on two different computational grids were made for

each model at each yaw angle in order to investigate the influence of
the mesh resolution on the results and to establish numerical accuracy. The ICEM-CFD commercial grid-generator package was used to
create the meshes. In this work, the numerical flow predictions were
carried out using an in-house finite volume developed multi-purpose
package CALC-PVM for parallel computations of turbulent flow in complex multi-block domains. LES equations were discretized using a
three-dimensional finite volume method in a collocated grid arrangement using Rhie-Chow interpolation. The convective, viscous diffusion
plus sub-grid fluxes were approximated by second-order-accurate central difference scheme. To reduce the numerical dissipation, no upwind schemes were used in the simulations. Because of that, the mesh
stretching ratio was kept below 1.1 around the train models since the
central difference scheme is more sensitive to the stretching ratio, especially in the places where large flow variations are expected (close to
the train model and in the wake where the flow is highly unsteady).
The time integration was done using the Crank-Nicolson second-order
scheme. The SIMPLEC algorithm was used for the pressure-velocity
coupling. Additional details about this code can be found in [18].

2.3

Boundary Conditions

Three simulations were made in this work. Two simulations use a

generic train model while the third simulation uses a simplified ICE2
train. In all the simulations, the train used is stationary without wheels
or suspension so that the train can not move with the wind. The models
and yaw angles in our study are those for which experimental data are
available. The computational domains are similar to the experimental
set-up in all simulations. The flow enters the computational domain
with the uniform velocity profile being constant in time. The domain
is extended upstream a sufficient distance from the model to ensure
constant inlet total pressure and to allow the generation of some turbulence in the flow before it approaches the model. The exit plane is
also extended downstream of the model to allow proper shedding in
the wake and to use a convective boundary condition at the exit. The
meshes are concentrated around the model and ground board in order
to resolve the important near-wall flow structures. No-slip boundary
conditions are applied on the train wall and the floor. The boundary
conditions on the side walls and roof are set depending on the exper9

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
imental set-up. For open wind tunnel, slip boundary conditions are
used on the side walls and roof while wall functions are used for the
closed wind tunnel. For all cases, a homogeneous Neumann boundary
condition is used for the pressure on all the boundaries.

10

Chapter 3
Visualization Techniques
The first step in the analysis of the LES results is to visualize the flow
around the train. This step is very important for understanding the
flow structure and highlighting the flow spots that are of special features. Streamlines, path lines, separation and reattachment lines are
tools commonly used to visualize the time-averaged surface flow. Different tools (such as local minimum pressure and second invariant of
velocity gradient) are used to define the vortex cores in the wake and
around the vehicle in both the instantaneous and time-averaged flows.
Ensight package for flow visualization is used to visualize the threedimensional data where most of the flow visualization tools are implemented. The following sections give a short description of the visualization tools used in this work.

3.1

Second Order Tensor Characteristics

The tensor can be defined as a linear transformation of vectors [19].

0  is transformed to another
According
to
this
definition,
if
a
vector
35
k as in Eq. 3.1,
vector through a linear transformer

then

e

 0   35

(3.1)

is a tensor.

3.1.1 Rotation Tensor

The rotation tensor is a tensor that transforms vectors in a rigid body
rotation manner. That is, the rotation transforms vectors drawn in the
rigid body into other vectors. Consequently, there is an infinite number
of rotation tensors according to the angle that they rotate the vectors.
11

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions

3.1.2 Eigenvalues and Eigenvectors of a Tensor

If there is a vector 0
parallel to itself, i.e.

that transforms under a tensor

e, 0   G 0 1

k,

k

to a vector
(3.2)

then vector 0 is called the eigenvector of the tensor

and the conG
stant, , is the corresponding eigenvalue. The eigenvector components
can be determined in an automatic way if we rearrange Eq. 3.2 to be
written in the following form

J e V G ED , :O 0   

(3.3)

Equation
3.3 is a system of linear homogeneous algebraic equations in
0 f 1 0 W , and 0 . Obviously, regardless of the values of G , a solution of this
W
system is 0 f  0  0Y  . This is known as the trivial solution. This
solution simply states the obvious fact that 0  satisfies the equation
e 0   G 0  , independent of the value of G . To find the nontrivial eigene , we note that a homogeneous system of equations admits
vectors for
nontrivial solution only if the determinant of its coefficients vanishes.
That is

 f f V G  f W

 f
 W f

 WW V G  W
(3.4)
 f
 W
 V G r
G . It is called the characterisThe above equation is k
a cubic
equation
in
 . The roots of this characteristic equation are
tic equation of tensor
k




the eigenvalues of
. The eigenvectors can be calculated from equa-

tion 3.3 usingethe

eigenvalues that have been calculated from
Eq. 3.4.

1
G

G
W and G
f
If tensor
is symmetric, then the three eigenvalues





are real. This gives three real eigenvectors to tensor
corresponding
to the three real eigenvalues. On the other side, some tensors have
eigenvectors in only one direction. For example, for any rotational tensor that affects a rigid body rotation about an axis, only those vectors
that are parallel to the axis of rotation will remain parallel to themselves. In general, the antisymmetric real tensors, which have real
components like the rotational tensor, have two complex conjugates
eigenvalues and one real. The real eigenvalue gives one real eigenvector that is parallel to the axis of rotation.

3.1.3 Principal Invariants of a Tensor

The characteristic equation 3.4 can be written as

G V & G W p ' G V ( r

12

(3.5)

CHAPTER 3. VISUALIZATION TECHNIQUES

where

&  -c 
  O W V -c J e W O
'  Wf J e| V e,4: O  WPf J -c 
(3.6)
( b ) -  
, do not depend on the
Since by definition the eigenvalues of tensor
base vectors, the coefficients of Eq. 3.5 will not depend on any particular
 . choice of basis. They are called the principal invariants of tensor

3.2

Critical Point Theory

It was and is still difficult to visualize the flow field in three-dimensional

data sets. This is because turbulent flows are characterized by vortices
ranging in size from the integral length scale down to the Kolmogorov
scale, and the definition of the vortex is still unclear.
Most of the proposed definitions of vortices are based on the critical
point theory. The critical points are points in the flow field where the
streamline slope is indeterminate and the velocity is zero relative to an
appropriate observer. Asymptotically exact solutions of Navier-Stokes
and continuity equation were found close to the critical points, as in
the work of Perry and Chong [20]. These solutions give a number of
standard flow patterns close to the critical point.
One of the most important parts of the flow visualization is to determine the critical points in both of the instantaneous and the timeaveraged flow. In the three-dimensional numerical solution of NavierStokes equations, the solution domain is divided to very small cells that
may take different shapes. One of these common shapes in computational fluid dynamics is the hexahedral cell, which is commonly used
in the structured mesh in the finite volume method. The critical points
are found in the cells where the ; , and components of the velocity
vector pass through zero. Taylor series can be used to expand velocity
;  around any arbitrary point, , inside the cell. This gives

;    p  b <  p e s b <  b < s p 

(3.7)
k e etc. are functions of time if the flow is unsteady,
Coefficients ,
and they are symmetric tensors in all indices except the first. If is
e

located at a critical point, then

are equal to zero since the velocity is
zero at the critical point relative to an appropriate observer. Moreover,
since the cell is very small and infinitesimal, then Eq. 3.7 can be truncated at the first-order term. The higher order terms have very little
13

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
influence in Eq. 3.7 only if the cell size is large. Since we are dealing
here with infinitesimal cells, we can use the truncated equation as an
exact expression for the velocity field inside the cell. This gives

;d b < p % b = p j b @

d b < p % b = = p j b @ @ 
d b < p % b p j b

(3.8)

Equation 3.8 can be written in the tensor form as

; 

 b < 

(3.9)

n  is the rate of deformation tensor or velocity gradient tensor,

 
; K <  (Jacobian).
The velocity gradient tensor, ; , can be decom-

where

posed into a symmetric part and an antisymmetric part as follows:

;    m
J ;   p ; :  O p J ;   V ; c  O 

(3.10)

Equation 3.10 can be written as

where

C 

+ 

;    +   p C 

(3.11)

is the symmetric part representing the strain rate tensor, i.e.

+   
J ;   p ; :  O
m

(3.12)

is the antisymmetric part representing the rotation or the spin

and
tensor

C   
J ;   V ; c  O 
m


(3.13)

It has been proven that the relative velocity, , can be related to the

C  . Thus   C  b < 
relative position, b < , by an antisymmetric tensor,
represents a rigid body rotation with an angular velocity of [21]
Thus the antisymmetric part of the velocity gradient tensor corresponds to rigid body rotation, and, if the motion is a rigid one, the
symmetric part of the velocity gradient tensor will vanish. For this
+ , is called the rate of strain tensor and its vanishing is
reason, tensor
necessary and sufficient for the motion to be without deformation, that
is, rigid. It should be remembered that the rotation tensor has one real
and two complex conjugate eigenvalues.
14

CHAPTER 3. VISUALIZATION TECHNIQUES

At the critical point, if the rotation tensor dominates over the rate of
 
strain tensor, then the velocity gradient tensor, ; K < , is an antisymmetric tensor. This implies that one of the eigenvalues is real and the
other two are complex conjugates. Thus, an attracting spiral-saddle
pattern
is formed. Equation 3.9 supports this theory, since the veloc
;
ity, , rotates around an axis parallel to the real eigenvector, which
corresponds to the real eigenvalue. If the real eigenvalue is zero, then
the critical point has absolute zero velocity. Otherwise it moves in the
direction of the real eigenvector with a velocity equal to the real eigenvalue. The flow pattern around the critical point in that case is called
n n
focus. On the other side, if the
 eigenvalues and eigenvectors of the
velocity gradient tensor, ; K < , are real, then three planes can be defined by the eigenvectors, and this will be referred as the eigenvector
planes. These planes do not need to be mutually orthogonal and, in
general, they are the only planes that contain solution trajectories (i.e.
some of the streamlines oscillate to these planes close to the critical
points). Figure 3.1 shows the different kinds of critical points that can
appear in the flow.

Unstable node
node

Saddle point

Stable focus
focus

sadle

k

nbl

pbl

Figure 3.1: Schematic representation of an unstable node, stable focus

! and & ! are negative and positive bifurcation lines,
and saddle.
respectively.

15

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions

3.3

Streamlines and Path-lines

Streamlines and path-lines are useful tools for visualizing the flow, especially on the model surface, since the separation and reattachment
lines can easily be detected. Vortex can be also detected using streamlines as each pair of separation and reattachment lines has an associated vortex core. Streamlines can, together with velocity vectors projected on the surface, be used to identify critical points. However, there
is a problem in terms of of predicting the critical points in the timeaveraged three-dimensional data sets far from the surface since the
streamlines are not Galilean invariant.

3.4

Vortex Core

Since the turbulence is viewed as a tangle of vortex filaments and most

of turbulence physics is well explained using the concept of vortex dynamics, a definition of a vortex could be used for flow visualization.
Although the concept of a vortex is very old, we still lake a definition of
it. As a consequence, different vortex definitions are proposed in the literature. Jeong and Hussain [22] summarized the different definitions
for vortex core and proposed a new definition based on the following
requirements:
(1) The vortex core must have net vorticity.
(2) It should be Galilean invariant.
They found that the intuitive indicators of vortices such as local pressure minimum, closed or spiraling streamlines and path-lines, and
isovorticity surface are inadequate and not suitable tools for detecting
vortices in an unsteady flow.

3.4.1 Region of Constant Vorticity Magnitude

Vorticity magnitude ( ) has been widely used to educe coherent struc} [22]. However, this approach is not
ture and represent vortex} cores
always satisfactory since
does not identify vortex cores in shear
}
flow, especially if the background shear} is comparable
to the vorticity
magnitude within the vortex. Thus, a
at a sufficiently low level
is necessary but not sufficient to detect a vortex in both free and wallbounded free shear flows [22].
16

CHAPTER 3. VISUALIZATION TECHNIQUES

3.4.2 Region of Complex Eigenvalues of Velocity Gradient Tensor

The critical point theory can be used to identify the center of the swirling
n n eigenvalues and
flow (i.e. vortex core). According to the theory, the
 
eigenvectors of the rate of deformation tensor, ; K < , evaluated at
the critical point define the flow pattern about that point. Specifically,
if the eigensystem displays one real and one pair of complex-conjugate
eigenvalues, the flow forms a spiral-saddle pattern. The two eigenvectors corresponding to the pair of complex-conjugate eigenvalues define
the plane on which the flow swirls, while the eigenvector corresponding to the real eigenvalue points in the direction about which the flow
 
G
spirals. The eigenvalues, , of the velocity gradient tensor, ; , satisfy
the characteristic equation Eq. 3.5 where

&  ;    W
'  Wf J ;   O V ;   ; : 
(3.14)
( b ) - ;  
 
are the three invariants of ; . The vortex core can be defined as the
B
G
region of complex that will occur when the discriminant, , of Eq. 3.5
is positive, i.e.

B J
' O p J
( O Wk 
m


(3.15)

The above criteria have been used by Sujudi and Haimes [23] to develop an algorithm to automatically locate the center of the swirling
flow in a three-dimensional vector field. The algorithm uses only tetrahedral cells, with all other cell types reduced to two or more tetrahedrals. It evaluates the velocity distribution inside the cell using linear
interpolation of four node points of the tetrahedral in the form:


;   6  p n <;  B <  

(3.16)

The rate of deformation tensor is simply the coefficient of the linear

interpolation velocity vector. The algorithm checks the eigenvalues for
each tetrahedral cell in the computational domain. If there is a cell
where one real and a pair of complex conjugate eigenvalues for the rate
of deformation tensor are found, then a vortex core passes through it.
17

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions

3.4.3 Region of Positive Second Invariant of Velocity Gradient

The second invariant of the velocity gradient,
as

'

, is defined in Eq. 3.14

' 
J ;   O W V ;   ; c  
m

   , this gives
For incompressible flow, ;
'  V
;   ; :  
C W V + W
m
m

Q C
Q












+
c

+
+
c

C
C
+ 
where 
, 
and

(3.17)

(3.18)

C 

and
are the
symmetric and antisymmetric components of the velocity gradients as
'
defined in Eq. 3.12 and Eq. 3.13, respectively. Thus represents the
local balance between the shear strain rate and vorticity magnitude.
' implies that the rotation tensor dominates
The region of positive
from
over the rate of strain
tensor, i.e. there
is a vortex. Moreover,
B
'
'

 is
Eq. 3.15, positive
implies
B  . positive , i.e. the condition of
more restrictive than

3.4.4 Region of Negative x

Jeong and Hussain [22] proposed a new definition of the vortex core
based on the local pressure minimum after discarding the viscous effect
that can eliminate the local pressure minimum in a flow with vortical
motion and also the effect of the unsteady straining, which can create
pressure minimum without involving vortical or swirling motion. UsW theC region
W G of1 Gthe
ing the above criteria, they defined the vortex ,as
flow

+
W
G
f
p
that has two negative eigenvalues for tensor
. If
and
G
GW G , the definition is equivalent to the
are the eigenvalues and f
GWe  within the vortex core.
requirement that

3.4.5 Region of Minimum Local Pressure

The pressure will have a local minimum in the vortex core in the case
in which the centrifugal force is balanced by the pressure force. In that
case, the local pressure isosurface can be used to identify the vortex
core. Unfortunately, in some cases the centrifugal force can be balanced
by the viscous force. This will imply that the method in this case fails
to identify the vortex. Thus, the existence of a local pressure minimum
is neither a sufficient nor a necessary condition for the presence of a
vortex core in general [22].
18

Chapter 4
Side-Wind Flow over a
Simplified ICE2 Train Model
This chapter investigates the flow around a simplified high-speed train
model of the Inter-City Express 2 (ICE2) under a side wind using LES.
The study is restricted to a side-wind yaw angle of    and Reynolds
number of m
 based on the free-stream velocity and height of the
train.

4.1

Train model and computational domain

The simplified train model consists of a leading car to which an end car
!
dummy is attached. The total length of the train model is {   m

and its height is
 Y . These dimensions are 1:10 of the real
dimensions of the leading train car. The clearance between the train

and the computational domain floor is  YY (  
R ). The model is
shown in Fig. 4.1.
Leading car

End car dummy


!


R 

Figure 4.1: Computational ICE2 train model

The LES results are compared with wind tunnel 
test
results. The

V
 open
measurements were made in the subsonic Russian k
jet tunnel. The experimental set-up is shown in Fig. 4.2. In contrast to
19

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
the computational model, the wind tunnel model has bogies and an inner gap between the leading car and the dummy end car. Moreover, in
order to keep the ground clearance, the model is held by a single cylindrical support positioned midway between the bogies. To simplify the
model in the computations, the inner gap is filled and both the bogies
and the cylindrical support are removed. In the measurements, the
(*) 
X
 based on the free-stream velocity
Reynolds number was
and the height of the train. A six-component external gauge was used
to measure the aerodynamic forces and moments on the train. The
) 6 with a sampling frequency of
time-averaged
data
were
collected
in

X

  @ [24]. The model stands on a thin elliptic plate (see Fig. 4.2) to
reduce the thickness of the approaching boundary layer.

Leading car

End car

Inner gap
Bogies
Ground board

Figure 4.2: Experimental set-up. Taken from [24].

In the computational domain, the model is yawed    to the freestream direction as shown in Fig. 4.3. The clearance between the model
and the ground board is typically the same as in the experimental setup.
The model centerline is kept parallel to the inlet of the domain with
 streamwise distance to ensure the same thickness of the apan
proaching boundary layer. The distance between the train tail and the
 downstream. These dimenexit of the computational domain is m

sions proved to be sufficient in a similar work around ground vehi

cles [16, 17]. The height of the computational domain is  m .
20

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL

Slip

wal
l

wall

Exit

Lateral


Lateral

wall

Figure 4.3: Computational domain

4.2

Boundary conditions

The flow enters the domain with a uniform velocity profile constant in
time. A convective boundary condition is implemented at the domain
exit. No-slip boundary conditions are used on the train surface. To simulate the experimental set-up, a slip boundary condition is applied at
the lateral walls and the roof. To reduce the thickness of the boundary
layer approaching the train model, two kinds of boundary conditions
are applied at the channel floor. A slip boundary condition is applied
 (see Fig. 4.3) from the inlet to suppress the
on a part extending 
development of a boundary layer, while a no-slip boundary condition
is used on the rest of the floor. A homogeneous Neumann boundary
condition is used for the pressure on all the boundaries.

4.3

Numerical details

LES is performed on the previously described model at Reynolds number of m

 based on the free-stream velocity and height of the train.
The ICEM-CFD mesh generator package is employed to build a structured mesh around the train. The total number of nodes in the computational domain is 11 million. Another simulation using a finer mesh
(17 million) is ongoing. A dummy ICE2 train is built around the model.
 between the model and the dummy train
The belt of thickness  
is used to make an -type grid. Another -type grid is built around
the first one to give better control of the mesh stretching as shown in
Fig. 4.4. The dummy train surface is used as a sheared wall between
the two -type grids. -type meshes are used in the rest of the grid
topology. Figure 4.4.b depicts the mesh density on the train surface
and the channel floor.
Hyperbolic stretching is used in the whole domain. The stretching
ratio between any two successive cells in the -type meshes does not
21

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
First -type
mesh
Second -type
mesh

-type mesh

a)

b)

Figure 4.4: Computational mesh around the ICE2 train. (a) Crosssection of the mesh showing the -type grids. (b) Mesh density on the
train surface and the channel floor.
exceed 1.1. This ratio varies in the -type meshes but is kept below
1.5 in any extreme place in the mesh. To resolve the boundary layer
on the model surface and to capture the variation of the flow physics
around it, the mesh is concentrated around the model where more than
m Y of the total nodes are confined in the -type grids. The governing
equations are solved all the way to the model wall to insure proper
variation of the flow quantities in the near-wall region, where much
of the flow physics originates. The grid density near the wall was fine
enough to resolve the small-scale structures that arise via shear layer
instabilities. The spatial resolutions of the first cell layer at the model
9
and floor are shown in Table 4.1, where ; is the friction velocity, is
the distance between the first node and the train surface in the wall
B is the cell width in the streamwise direction and
normal
B is thedirection,
cell width in the span-wise direction.
Table 4.1: Spatial resolutions for the coarse mesh simulation.
mean
maximum

=  ; 9 K I
0.5
4

 B ; 9 K I
40
250


B ; 9 K I
90
700

LES equations are discretized using the three-dimensional finite

volume method in a collocated grid arrangement using Rhie-Chow interpolation. Since the high frequency modes in the model boundary and
shear layers are important in side-wind simulations, the viscous diffusion and sub-grid fluxes are approximated by a second-order-accurate
22

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL
central difference scheme. It was not possible to use pure central difference discretization for the convective flux because of divergence problems. Instead, the convective flux is approximated using a blend of a

Y second-order upwind scheme (Van-Leer) and a  Y second-order

central difference scheme. The numerical dissipation that gained from
the
Y upwind scheme played an important role in the stabilizing
of the convergence. The time integration is done using the CrankNicolson second-order scheme. The SIMPLEC algorithm is used for
the pressure-velocity coupling. Fully developed
flow is ob- 9 - /.g K  turbulent
. The physical
tained from the simulation after time 


R


-:9
time step is

P ( {    ), which gives a maximum Courant~

!
Friedrichs-Levy

number of about 1.5. The time-averaged flow is
i
9
m
obtained using 
, which is about two and half times longer than
the experimental sampling time.

4.4

Results

The choice of the proper boundary conditions is crucial in any numerical simulation. Since the inlet boundary condition is a uniform velocity
profile constant in time, the distance between the inlet plane and the
model surface should be long enough to avoid the influence of the model
on the inlet total pressure. This distance is also necessary to get some
turbulence in the flow before it approaches the model as an attempt to
simulate the natural wind. Moreover, the slip boundary conditions that
are used on the lateral walls and roof of the channel require a sufficient
distance from the model to be applicable.

a)

b)

Figure 4.5: Cross-section of the computational domain at the middle of

the train, parallel to channel floor: (a)showing static pressure contours,
(b) colored by total pressure.
Thus attention is paid to the size of the numerical domain and the
the position of the model inside it. In order to reduce the total number
of nodes and to get high resolution, which is required for good LES, the
size of the computational domain should be as small as possible. This is
23

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
met by the blocking constraint and the requirement of a constant total
pressure at the inlet and lateral walls. The influence of the model on
the static and total pressures of the boundaries are shown in Fig. 4.5.
Figure 4.5.a represents the static pressure contours on a plane parallel
to the channel floor at the middle of the height of the train, where the
distance between the inlet plane and the train surface is at its minimum. There are variations in the static pressure on the boundaries of
the computational domain. Figure 4.5.b displays the same plane colored by the time-averaged total pressure. The total pressure at the
inlet and lateral walls is approximately constant and is not affected by
the model. There is a total pressure loss in the wake behind the train.
This loss extends to the exit of the channel.

4.4.1 Time-averaged flow

B -

The physical time step in our LES computation is

  
sec.
The time-averaged data are collected using 100000 time steps. This
) -9
corresponds to a simulation time of
 6 (  m ), which is 2.5 times
longer than the sampling period in the wind tunnel measurements.
These data are used to calculate the statistics of the flow and to explore
the flow structures around the train in the time-averaged sense. In this
section, the surface pressure distribution is computed and compared
with the measured surface pressure. The wake structures and their
influences on the surface flow pattern are explored.
Surface pressure distribution
The surface pressure distributions are computed at certain cross-sections
along the length of the train where experimental data are available, as
shown in Fig. 4.6. Figure 4.6 also shows some details from the experimental train, such as spoiler, support, bogies and inner-car gap,
which are omitted from the computational model. These differences
should be kept in mind when comparing the LES results with the experimental data. Figure 4.7 displays the local pressure coefficient,
x ,
F
distributions over the circumferential angle, , measured counterclockwise around the < -axis, as shown in Fig. 4.6. The LES results show
good agreement with the measurements on the streamwise and topside faces at all the cross-sections. Moreover, good correlation is obtained between the LES results and the measured data on the lee-side
face. The difference in the geometry under the train (i.e. bogies, the
spoiler and the support) between the experimental and the numerical
set-ups resulted in a difference in
l values on the surface of the under
side.
24

Support in
experiment

=0.75

=0.580

=0.44

=0.14

=0.03
=0.07

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL

Veiw from
front
Spoiler in
experiment

Bogies in
experiment

Figure 4.6: Location of the measured time-averaged pressure distribuF

tions and the orientation of angle .
3

Lee-side
face

Bottom-side Streamwise
face
face

e*5 :

2
1

PSfrag replacements

1
0

*5 c

PSfrag replacements
100

200

1
0

300

100

Angle ( )
3

1
0

*5 c

1
0

PSfrag replacements
100

200

1
0

300

100

Angle ( )
3

e*5z

2
1

1
0

200

300

Angle ( )
3

*5

PSfrag replacements

PSfrag replacements

300

Angle ( )

e*5 i

200

PSfrag replacements
100

200

300

1
0

Angle ( )

100

200

300

Angle ( )

Figure 4.7: Comparison of the surface pressure coefficient distribution along the length of the train: LES (solid line); experimental data
from [24] (symbols).

25

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
Suction pressure is reported in the experimental data at the most
! {   , on the bottom side of the train. This sucfrontal position, <K
tion pressure can be attributed to the influence of the front spoiler. The
influence of the bogies is shown as higher
A values at the streamwise
!  Y . There are no measured
corner on the underbody face at <K
!  
X due to the existence of the
data on the underbody face at <K
bogies. The LES results coincide with the measured data at the other
!
faces at that position. Cross-section <K  X X is probably the best one
for a comparison between our LES results and the experimental data
since it is the position farthest from the bogies. Although the influence
of the bogies on the flow at this position is relatively small, the flow is
also affected by the support cylinder. However, there is good agreement
between the LES results and the measurements on all the faces. Far!   and 0.75, the influence of the
ther from the train front, at <K
underbody details (bogies, the support and the spoiler) used in the experiment are still visible. In addition, the influence of the inner-car gap
appears on the lee-side face pressure distribution, where measured
x
values are higher than LES results. The maximum suction peak and
the highest pressure are found on the upper-face leeward edge of the
nose and on the streamwise face of the nose, respectively, at two front
!
!
cross-sections, <K    and <K   (Fig. 4.7). The peak
A magnitudes in these two sections are almost twice as large as those in the
subsequent cross-sections. This suggests that the loads are accumulated in the nose region of the leading car and increase the importance
of the nose shape in the side wind instability. In general, although the
LES Reynolds number is seven times lower than the experimental one,
the LES results are in fairly good agreement with the measurements.
Surface flow patterns
Figure 4.8 displays the LES streamlines projected on the train surface
representing the time-averaged surface flow patterns. The surface flow
pattern displays regions of separated and attached flow. These regions
are separated by separation, reattachment and bifurcation lines. Figure 4.9 is a schematic representation of these lines as inferred from
Fig. 4.8. Figure 4.8.a shows a uniform flow over the streamwise face.
There is one bifurcation line, f , on the streamwise face, as shown in
Fig. 4.9.a, from which flow is moving toward the roof and the bottom
of the train. As soon as the flow reaches the upper-side face it sepa+
rates from the face at separation line f , as shown in Fig. 4.9.b. The
Reynolds number in the simulation is m
 , which is about 30 times
lower than the Reynolds number of the real high-speed trains at the
normal operating conditions. Thus separated flow is expected on the
26

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL
upper-side face. The flow attaches to the face at the reattachment line,
 f , as shown in Fig. 4.9.b. Figure 4.8.b shows that the separated flow
on the upper-side face is increasing in the direction of the length of the
!  X , the face is completely covered by
train. At approximately <K
the separated
flow. The flow separates on the bottom-side
face at sepa
+
W

ration line , and reattaches at reattachment line , forming a small

separation bubble in the underbody flow. Both the reattached flow on
the upper-side and the bottom-side faces separate from the train sur+ W and + ,
face when they reach the lee-side edge at separation lines

respectively. The two vortex sheets shedding from the upper side and
under side of the train roll up to form separation bubbles in the wake
flow. These bubbles are not always attached to the surface but they
detach from the surface at successive places along the length of the
train. When these vortices detach from the surface, other vortices are
formed. Figure 4.8.c shows a complicated flow pattern on the lee-side
face caused by the attachment and detachment of the wake vortices.
Figure 4.10 shows part of this flow on the lee-side face of the train nose.
The two recirculation bubbles shed from the upper and lower sides of
W
the face and meet at bifurcation line on the train surface. The attached flow from the upper recirculation bubble moves toward the roof
+ and
of the train where it separates on secondary separation  line

reattaches to the surface at secondary reattachment line before it

+ W
separates again at primary separation line . Similarly, the attached
flow from the floor-side recirculation bubble moves toward the bottomside face. Figure 4.10.b shows that this flow separates from the train
+ and reattaches to the surface
surface at secondary separation line

at secondary reattachment line

before it separates again at primary
+

separation line . The size of the recirculation bubbles increases in

the positive x-direction along the train. After a certain distance, they
have completely detached from the face and convected away into the
wake flow. Once these two circulation bubbles detach from the train
surface, two other bubbles are formed. This process is repeated along
the the length of the train with bifurcation lines and and sec+ and + for the lower recirculation region, as
ondary separation lines
shown in Fig. 4.9.c. On the other side, the upper recirculation region
seems to be attached to the train surface for approximately two-thirds
of the length of the train, as is shown by long secondary separation and
+ and , respectively.
reattachment lines

27

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions

a)

b)

c)

d)

Figure 4.8: Time-averaged trace lines projected on the surface of the

train showing surface flow patterns.



a)



b)



c)







 

d)







Figure 4.9: Schematic representation of the separation and reattachment lines found in Fig. 4.8.










a)

 


b)

  

 



Figure 4.10: Surface flow pattern on the lee-side face of the nose: (a)
view from above and the lateral side on the rear end of the train, (b)
view from below and the lateral side on the rear end of the train.

28

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL
Wake structure
Figure 4.11 shows the time-averaged wake structures by means of the
W
vortex cores. Two vortices, 6 f and 6 , originate on the train nose
tip. Figure 4.11.a shows that vortex 6 f stretches in the wake flow
all the way from the nose tip to the end of the leading car. It grows
in size but remains attached to the train surface. Numerous vortices,

!#"0/
!#" '
!#"%)

!#"%(

!#"%$

a)

!#"%$
!#"+,

!#"+*

!#"%)

!#"%( !&"'

b)

Figure 4.11: Streamlines generated from the vortex cores: (a) view
from above, (b) view from below the train.

, are born on the lee-side of the upper edge of the train. These weak
vortices roll up and merge with the strong vortex, 6 f . The recirculated flow around the strong vortex, 6 f , is attached to the upper half
of the lee-side face of the train along the length of the leading car until
+
it separates at separation line , as shown in Fig. 4.9.c. In contrast,

W
Fig. 4.11.b shows that the other nose vortex, 6 , once it is born on the
lower part of the train nose, detaches from the surface and is convected

another vortex, 6 , sheds from
downstream in the wake. At <K
the underbody flow. It rolls up and detaches shortly after its onset and
29

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
convects downstream in the wake flow. Similarly, another strong vortex, 6 , emerges from the lee-side underbody edge. The vortex grows

in size in the direction of the length of the train due to the emergence of
the other weak vortices, 621 , which are born at the same edge. After a
short distance in the direction of the length of the train, the vortex 6

detaches from the train surface and changes its direction toward the
wake. The process of shedding vortices under the train body continues
by shedding vortex 6 , as shown in Fig. 4.11.b.

a)

b)

Figure 4.12: Wake structures. (a) Experimental view (taken from [5]).
(b) DES results (taken from [24]).
This finding about the wake structures is different from that found
in the experimental investigations. Figure 4.12.a shows a schematic
representation of the flow around a generic train under the influence
of side wind at a   yaw angle by Copley [5]. He defined the wake
shedding as a progressive couple of vortices shed from the upper and
lower edges of the lee-side face at the same position. Wu [24] computed
the flow around the ICE2 model at a    yaw angle using detachededdy simulation (DES). Although the Reynolds number in his simulation was five times higher (
 ) than that used in the present work, he
found in agreement with our results that the upper vortex in the wake
flow extends far beyond the inter-car gap, as shown in Fig. 4.12.b. He
visualized another vortex shed from the upper edge of the lee-side face
directly after the inner-car gap. Besides, he found several vortices that
were shed from the underbody. They detach from the surface and decay within a short distance. The model used in his simulation was
almost twice as long as in the present work. Thus, the distance that
vortex 6 f remains attached to the lee-side face is independent of the
Reynolds number or the length of the model used in the simulation.
Figure 4.13 shows the development of the major vortices in the timeaveraged wake flow by means of the pressure minimum. It shows that
!   . Along the
vortex 6 f detaches from the lee-side face after <K
distance over which 6 f remains attached to the surface, a number of
30

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL
vortices are formed and detach on the lower part of the face.

\76 98;:<=:?>@>
[
_

\76 L8M:<HI@J@C

\Q6 L8;:<HP@R@G

\76 L8M:<OE?PA

\Q6 L8;:<S>P@R

354 W

\76 98;:<BADCFE
354 f
354 W

\76 98;:<HG@IA

3N4

\76 L8M:<HC@I@J

\Q6 L8;:<HJFE?P

354 f
354 W

\76 98;:<HI:KJ

3N4

\76 L8M:<HPADC

\Q6 L8;:<HR@G@I

34 f
354 W

Figure 4.13: Cross-sections colored by the time-averaged static pressure showing the development of wake structures along the length of
the train.

4.4.2 Aerodynamic forces

The lift and the side force coefficients are computed at each time step.
The values of these coefficients are saved during the averaging process.
31

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
The time varying signal is shown in Figs. 4.14.a and 4.14.b for the side
and lift coefficients, respectively. Their statistics are shown in Tab. 4.2.
0.46

TVU
PSfrag replacements
a)

0.2

0.44

Tdc

0.42
0.4

PSfrag replacements

0.38
0.36
0

YW 20X[Z\W^30]_a`40b

10

50

60

b)

0.15
0.1
0.05
0
0

10

YW 20XeZfW^30]_a`40b

50

60

Figure 4.14: Time history of the aerodynamic coefficients. (a) Side force
coefficient. (b) Lift force coefficient.

Table 4.2: Statistics of the time varying signal of

the aerodynamic coefficients.
force





mean
0.41
0.0834

0.014
0.0257

maximum
0.449
0.1442

minimum
0.368
0.0179

Due to the absence of underbody blocking (i.e. bogies, spoiler and

support cylinder) in our simplified model, there is relatively high velocity under the train. Consequently, a region with lower pressure exists.
This results in a lower mean value of the lift force coefficient g|
ih as
reported in Tab. 4.2.

4.4.3 Instantaneous flow

The time varying signals of the side and the lift force coefficients reflect
the dynamic behavior of the flow around trains at a    yaw angle. Figure 4.14 shows that, in certain time slots, the aerodynamic coefficients
oscillate around the mean with high amplitude. These high amplitude
oscillations are due to the detachment and reattachment of the wake
vortices to and from the lee-side face. The centers of these vortices do
not have fixed position in space. Instead, they oscillate in a plane parallel to the channel floor. The higher values of the side force coefficient
are associated with the attachment of wake vortices (the nearest distance between the center of the vortices and the lee-side face) and vice
versa. This oscillation produces unsteady vortex shedding in the wake.
32

CHAPTER 4. SIDE-WIND FLOW OVER A SIMPLIFIED ICE2

TRAIN MODEL
In contrast to this process, some parts of the time signal of the aerodynamic forces display lower oscillation around the mean. This is because
of the stationary behavior of the wake vortices. Only some weak vortex
shedding from the upper and lower sides of the train causes the high
frequency oscillation of the forces. Here, there is no vortex shedding in
the wake flow. Instead, the flow circulates in a spiral way around the
stationary vortex cores. This flow is similar to that of the slender body
flow.

l lnm

l lnmuo prl

l limo pxl

l lnmoqprl

l lnmuo prl

l limoqyYpxl

l l m oqstprl

l l m owvtprl

l l m o v prl

l l m o prl

l l m o prl

l l m o c prl

!DjK'
! jk)

Figure 4.15: Cross-section of the flow at <K    colored by instantaneous static
The time difference between two successive
B - pressure.
snapshot is
r 
sec.
Figure 4.15 gives 12 instantaneous pictures showing the time-varying
! 
of the distribution of the local pressure coefficient in the plane <K
 . The plane passes through the major vortices, 6 f and 6 , and
some of the weak vortices, 6 and 6D1 . Vortex 6 f has different sizes
at different times due to the process of merging with the upper weak
33

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions

vortices. The largest size of vortex 6 f is found at time 

 , yielding
low pressure on the upper-side and lee-side faces, shown by the large
area of dark color representing
the low pressure. Its smallest size is
B
p
. The center of vortex 6 f is fixed in time
registered at 

and space. However, the size of the vortex changes in a regular fashion around the mean size with almost constant frequency. The lower
vortices are highly unsteady and their centers are not stationary. At
-  - p B - the vortex is fully attached to the wall, while at -  - p

B 

it is fully detached. Figure 4.15 also explores a time-varying size of this
vortex. We found that the lower vortices are responsible for the vortex
shedding in the wake if it exists and there is no vortex shedding from
the upper vortices.
3

15

{}| z W UY2~

{| z W cn~ 10

PSfrag replacements
a)

PSfrag replacements

1
0
0

0.1

0.2

0.3

z W 0.4
U

0.5

0.6

b)

5
0
0

0.1

0.2

0.3

z W c 0.4

0.5

0.6

Figure 4.16: Autopower spectra of the time-varying signal of force coefficients. (a) Side force coefficient. (b) Lift force coefficient.
To determine the frequencies of these motions, autopower spectra
of the time varying signals of the force coefficient are computed. Figures 4.16.a and 4.16.b show these spectra for side and lift force coefficients, respectively. One common dominating frequency corresponding
+.-  
is found for both the lift and the side forces. This frequency
to
represents the averaged oscillating frequency of the wake vortices in
the horizontal plane. The shedding frequencies are displayed as two
+.-  
R and 0.2. A
other dominating frequencies corresponding to
+.-
number of high frequency modes are found in the range of 0.2
0.4. These frequencies are caused by the unsteady shedding of the
weak vortices from the upper and lower faces of the train.

34

Bibliography
[1] C.J. Baker. Some complex applications of the wind loading
chain. Journal of Wind Engineering and Industrial Aerodynamics,
91:17911811, 2003.
[2] C.J. Baker, J. Jones, F. Lopez-Calleja, and J. Munday. Measurements of the cross wind forces on trains. Journal of Wind Engineering and Industrial Aerodynamics, 92:547563, 2004.
[3] T. W. Chiu. Prediction of the aerodynamic loads on a railway train
in a cross-wind at large yaw angles using an integrated two- and
three-dimensional source/vortex panel method. Journal of Wind
Engineering and Industrial Aerodynamics, 57:1939, 1995.
[4] T. W. Chiu and L. C. Squire. An experimental study of the flow over
a train in a crosswind at large yaw angles up to    . Journal of
Wind Engineering and Industrial Aerodynamics, 45:4774, 1992.
[5] J. M. Copley.
The three-dimensional flow around railway
trains. Journal of Wind Engineering and Industrial Aerodynamics, 26:2152, 1987.
[6] M. Suzuki, K. Tanemoto, and T. Maeda. Aerodynamic characteristics of train/vehicles under cross winds. Journal of Wind Engineering and Industrial Aerodynamics, 91:209218, 2003.
[7] U. Hoppmann, S. Koenig, T. Tielkes, and G. Matschke. A shortterm strong wind prediction model for railway application: design and verification. Journal of Wind Engineering and Industrial
Aerodynamics, 90:11271134, 2002.
[8] B. Diedrichs. On computational fluid dynamics modelling of crosswind effects for high-speed rolling stock. IMechE, 217(F):203226,
2003.
[9] F. Durst W. Khier, M. Breuer. Flow structure around trains under
side wind conditions: a numerical study. Computers & Fluids,
29:179195, 2000.
35

Hassan Nasr Hemida, Large-Eddy Simulation of the Flow around

Simplified High-Speed Trains under Side Wind Conditions
[10] Diedrichs Ben. Computational Methods for Crosswind Stability of Railway Trains. Report, Dept. of Mechanical Engineering, Division of Railway Technology, Royal Institute of Technology,
Stokholm, SWEDEN, 2005.
[11] S. Krajnovic and L. Davidson. Large eddy simulation of the flow
around an Ahmed body. In 2004 ASME Heat Transfer/Fluids Engineering Summer Conference, Charlotte, North Carolina, USA,
2004.
[12] S. Singh and S. Mittal. Flow past a cylinder: shear layer instability and drag crisis. Int. J. Num. Meth. Fluids, 47:7598, 2005.
[13] G. Kravchenko and P. Moin. Numerical studies of flow over a circular cylinder at Re=3900. Physics of Fluids, 12(2):403417, 2000.
[14] R. Mittal and P. Moin. Stability of upwind-biased finite difference
schemes for large-eddy simulation of turbulent flows. AIAA Journal, 35(8):14151417, 1997.
[15] G.S. Constantinuescu and K.D. Squires. LES and DES investigations of turbulent flow over a sphere at Re=10,000. Journal of
Flow, Turbulence and Combustion, 70:267298, 2003.
[16] S. Krajnovic and L. Davidson. Flow around a simplified car, part
1: Large-eddy simulation. ASME: Journal of Fluids Engineering,
127:907918, 2005.
[17] S. Krajnovic and L. Davidson. Flow around a simplified car, part
2: Understanding the flow. ASME: Journal of Fluids Engineering,
127:919928, 2005.
[18] H. Nilsson and L. Davidson. CALC-PVM: A parallel multiblock
SIMPLE multiblock solver for turbulent flow in complex domains.
Technical report, Dept. of Thermo and Fluid Dynamics, Chalmers
University of Technology, Gothenburg, 1998.
[19] W. Michael, D. Rubin, and E. Krempl. Introduction to Continuum
Mechanics. Butterworth Heinemann, 1996. ISBN 0-7506-2894-4.
[20] A. E. Perry and M. S. Chong. A description of eddying motions and
flow patterns using critical-point concepts. Ann. Rev. Fluid Mech.,
19:125155, 1987.
[21] Rutherford Aris. Vectors, Tensors, and Bacis Equations of Fluid
Mechanics. Dover Publications, Inc., 1 edition, 1962. ISBN 0-48666110-5.
36

BIBLIOGRAPHY
[22] J. Jeong and F. Hussain. On the identification of a vortex. Journal
of Fluid Mechanics, 285:6994, February 1995.
[23] D. Sujudi and R. Haimes. Identification of swirling flow in 3-D
vector fields. AIAA Paper AIAA 95-1715, 1995.
[24] D. Wu. Predictive prospects of unsteady detached-eddy simulations in industrial external aerodynamic flow simulations.
Diploma thesis. matriculation number: 219949, Lehrstuhl fur
Strmungslehre und Aerodynamishes Institute, Aachen, Germany,
2004.

37