Académique Documents
Professionnel Documents
Culture Documents
Jörn Wildhagen∗
Matrikelnummer: 995044
This thesis has been submitted to Institute for Hydromechanics, University of Karls-
ruhe (TH), in fulfillment of the requirements for achieving academic degree. I hereby
testify that the work is original. Other used sources are well identified through
references within this text and summerized in the last chapter. I agree, that this
thesis is exhibited at a library and can be duplicated by photocopies.
Diese Diplomarbeit wurde dem Institut für Hydromechanik an der Universität Karls-
ruhe (TH) zur Erreichung des akademischen Grades des Diplomingenieurs vorgelegt.
Ich versichere, dass ich diese Diplomarbeit selbständig verfasst, noch nicht ander-
weitig für andere Prüfungszwecke vorgelegt, keine anderen als die angegebenen Quel-
len und Hilfsmittel benutzt sowie wörtliche und sinngemäße Zitate als solche gekenn-
zeichnet habe. Ich erkläre mich damit einverstanden, dass meine Diplomarbeit in
eine Bibliothek eingestellt sowie kopiert wird.
Jörn Wildhagen
Abstract
This thesis contains results from a numerical simulation of sediment transport in a
sharply curved meandering channel. These results were performed by the numer-
ical model SSIIM. The numerical results are tested against the measurement data
taken from a physical experiment with steady flow conditions, gained after a dura-
tion time of ∆t = 4 h. The results are analyzed in order to determine the capabilities
and limitations of the SSIIM model to reproduce the physical processes of sediment
transport.
The comparison of the numerical simulation by SSIIM and the measurement data
show that the overall trends in deposition and formation of point bars agree on
the main points. Concerning erosion, the numerical model predicts the same spot
of erosion zone as observed in the experiment, but SSIIMs quantitative calculation
of erosion is highly overpredicted in comparison to the experimental results. The
numerical model results have been proved to be independent of the time step ∂t, but
show some dependency on the grid size. Nevertheless, the results correspond with
observations of the conducted experiment.
Deviations between model and experimental results are discussed as well as uncer-
tainties in numerical modelling.
Kurzfassung
In dieser Diplomarbeit werden nummerische Simulationen zum Sedimenttransport
in einem eng mäandrierenden Kanal mit dem Simulationsprogramm SSIIM durch-
geführt. Ein Experiment mit stationärem Abfluss wird nachgebildet. Die Ergeb-
nisse der nummerischen Simulation werden mit experimentellen Daten nach einer
Durchführungsdauer von ∆t = 4 h verglichen. Die Ergebnisse werden analysiert, um
die Fähigkeiten und Grenzen des nummerischen Modells SSIIM im Hinblick auf die
Reproduzierbarkeit der physikalischen Prozesse des Sedimenttransportes aufzeigen zu
können.
Der Vergleich zwischen den Simulationsergebnissen von SSIIM und den Messergeb-
nissen zeigt, dass die Ablagerung von Sedimenten und die Formation von Anhäu-
fungspunkten weitgehend übereinstimmen. Das nummerische Modell ermittelt den
gleichen Erosionsbereich wie im Experiment beobachtet. Allerdings wird die quan-
titative Berechnung der Erosion von SSIIM überbewertet, wie ein Vergleich mit den
vorliegenden Messdaten zeigt. Die nummerischen Ergebnisse zeigen, dass sie un-
abhängig von der Wahl des Zeitschrittes ∂t sind, aber eine leichte Abhängigkeit in
der Rechengittergröße besteht. Die Ergebnisse liefern insgesamt eine zufrieden stel-
lende Übereinstimmung mit weiteren Beobachtungen aus dem Experiment.
Abweichungen zwischen den Modell- und Experimentergebnissen werden diskutiert
sowie Unsicherheiten bei nummerischen Berechnungen aufgezeigt.
Nomenclature
Symbol Unit Explanation
A m2 flow cross–sectional area
a m reference level
B m width
C kg/m3 concentration
ca ppm reference concentration
cµ ,c1 ,c2 ,σk ,σ − k– turbulence model constant
D, d m particle diameter
D∗ − dimensionless particle diameter
DH m hydraulic diameter
Et m/s2 turbulent diffusion coefficient
Em m/s2 molecular diffusion coefficient
Fr − Froude–number
F r∗ − particle Froude–number
g m/s2 acceleration due to gravity
h m flow depth
Ie − energy slope
Io − channel slope
k m2/s2 turbulent kinetic energy
K 1/day or 1/s reaction coefficient
ks m equivalent sand roughness
1
kSt m /3/s Strickler–coefficient
L m length scale of turbulence
p N/m2 pressure
Pw m wetted perimeter
Q m3/s discharge
qb m3/ms bed–load transport rate
Re − Reynolds–number
Re∗ − particle Reynolds–number
s − specific density
T − transport stage parameter
t s time
u∗ m/s bed shear velocity
u0∗ m/s bed shear velocity related to grains
V̂ m/s velocity scale of the turbulent motion
x, y, z m cartesian coordinates
Z − suspension parameter
u, v, w m/s velocity components in x,y,z–direction
v
Greek Symbols
Symbol Unit Explanation
α − transverse bed slope
β − ratio of sediment and fluid diffusion coefficient
β0 − averaging factor
δij − Kronecker delta, 1 for i = j, else 0
m2/s3 dissipation rate
κ − von-Karman-constant, κ = 0.4
µ N s/m2 dynamic viscosity
µt N s/m2 eddy–viscosity
ν m2/s kinematic viscosity
φ − angle of repose of sediment particles
%,ρ kg/m3 density
σt − turbulent Schmidt number
σg − geometric standard deviation of particle grain size
τ N/m2 shear stress
τ̃ − τb/τ
h
Subscripts
Symbol Explanation
’ fluctuation quantity
time averaged quantity
b quantities associated with bed/bottom
c center
crit critical
h quantities associated with horizontal bed
i,j,k component running index
m mean
r radial
s sediment
sur surface
w water
84, 50, 16 sediment grain size d; defined as the size for which 84%, 50% or 16%
weight of the material is finer
vi
Abbreviations
Symbol Explanation
cf. confer
CFD Computational Fluid Dynamics
e.g. exempli gratia (lat.) = for instance
eqn. equation
et al. et alii (lat.) = et al.
fig. figure
i.e. id est (lat.) = that is to say
no. number
resp. respectively
SSIIM Sediment Simulation In Intakes with Multiblock option
SIMPLE Semi–Implicit Method for Pressure–Linked Equations
Contents
Statement ii
Abstract iii
Nomenclature iv
List of Figures ix
List of Tables ix
Acknowledgement x
1 Introduction 1
2 Fundamentals 2
2.1 Governing Equations for Water Flow Calculation . . . . . . . . . . . 2
2.2 Assumptions and Approximations . . . . . . . . . . . . . . . . . . . . 3
2.3 Governing Equations in Simplified Form . . . . . . . . . . . . . . . . 4
2.4 Turbulence Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4.1 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4.2 Reynold Stress Terms . . . . . . . . . . . . . . . . . . . . . . . 5
2.4.3 Boussinesq’s Approximation and Eddy–Viscosity . . . . . . . . 5
2.4.4 Eddy–diffusivity Concept . . . . . . . . . . . . . . . . . . . . . 6
2.4.5 The k– Turbulence Model . . . . . . . . . . . . . . . . . . . . 6
7 Numerical Simulation 24
7.1 Represented Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
7.2 Presentation of Water Flow Results . . . . . . . . . . . . . . . . . . . 24
7.2.1 Basic Flow Configuration . . . . . . . . . . . . . . . . . . . . . 25
7.2.2 Secondary Flow . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.2.3 Distribution of Longitudinal Velocity . . . . . . . . . . . . . . 26
7.2.4 Distribution of Vertical Velocity . . . . . . . . . . . . . . . . . 28
7.2.5 Stream Separation and Formation of Recirculation Zone . . . 28
7.2.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 32
7.3 Presentation of Sediment Transport Results . . . . . . . . . . . . . . 32
7.3.1 Basic Sediment Configuration . . . . . . . . . . . . . . . . . . 33
7.3.2 Default Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
7.3.3 Base Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7.3.4 Parameter Variation . . . . . . . . . . . . . . . . . . . . . . . 35
7.4 Uncertainties in Numerical Modelling . . . . . . . . . . . . . . . . . . 39
C Photos 46
Bibliography 48
List of Figures
1 Definition sketch of angles θ and α . . . . . . . . . . . . . . . . . . . 10
2 Variation of τ̃ with α according to Deys empirical equation . . . . . . 11
3 Dimensionless particle motion diagram . . . . . . . . . . . . . . . . . 12
4 Vertical distribution of sediment concentration according to Rouse [20] 14
5 Definition sketch of transverse inclination of surface and transverse
circulation at bend in channel . . . . . . . . . . . . . . . . . . . . . . 16
6 Definition sketch of channel shape seen from above . . . . . . . . . . 21
7 Definition sketch of channel shape seen from side . . . . . . . . . . . 21
8 Particle size distribution . . . . . . . . . . . . . . . . . . . . . . . . . 23
9 Definition sketch of mesh, top view . . . . . . . . . . . . . . . . . . . 24
10 Definition sketch of cross–section, seen looking upstream . . . . . . . 25
11 Velocity vector plot at cross-section no. F243 . . . . . . . . . . . . . . 26
12 Position sketch of cross–sections . . . . . . . . . . . . . . . . . . . . . 27
13 Longitudinal velocity distribution . . . . . . . . . . . . . . . . . . . . 29
14 Horizontal velocity distribution (begin) . . . . . . . . . . . . . . . . . 30
14 Horizontal velocity distribution (end) . . . . . . . . . . . . . . . . . . 31
15 Sediment transport results conducting default case . . . . . . . . . . . 34
16 Comparison of bed elevation changes between base case and experiment 36
17 Bed elevation changes with regard to different bed roughness heights 37
18 Sediment calculation with refined mesh and ∂t = 60s . . . . . . . . . 38
19 Sediment calculations with distorted mesh at different time steps . . . 39
20 Channel before experimental run . . . . . . . . . . . . . . . . . . . . 46
21 Photo of channel bend showing one point bar . . . . . . . . . . . . . 47
22 Photo of channel bend showing separated point bars . . . . . . . . . . 47
List of Tables
1 The values of the k– turbulence model [19]. . . . . . . . . . . . . . . 7
2 Computation time for a simulation time of ∆t = 4 h . . . . . . . . . . 37
Acknowledgement
This diploma thesis is the result of collaboration between the author, presently at
the Department of Hydraulic and Environmental Engineering in the Norwegian Uni-
versity of Science and Technology (NTNU) and the author’s home department, the
Institute for Hydromechanics at the University of Karlsruhe (TH). The study has
been carried out at the Department of Hydraulic and Environmental Engineering,
NTNU in summer 2004.
First of all I would like to thank both my supervisors at NTNU, Prof. Nils Rei-
dar B. Olsen and Nils Rüther for their advice and support. They have always been
available and took time for discussion. Their help guided me through this work and
led me to a deeper understanding of sediment engineering and computational fluid
dynamics, which I appreciated very much. It has been a pleasure to work with them.
I would also like to mention with thanks the help of Ian Guymer and Richard
Dutton from the Department of Civil and Structural Engineering at the University
of Sheffield. They provided me with all requested data from the physical experiment
in order to carry out reasonable simulations and gave me advice during my work.
Furthermore, I would like to thank all the people at my working place and all the
master’s thesis students I met at the ’brakka’ during lunch time. I really enjoyed
their warm welcome.
I also want to express my gratitude to all the people who encouraged and supported
my request to go abroad during my final thesis at a foreign university.
And finally I am deeply indebted to my parents, who provided me with both moral
and financial support during my period of study.
1 Introduction
Natural meandering rivers are very complex in their water flow situation. The stream
is characterized by turbulent, strongly three-dimensional and irregular channel topog-
raphy. Due to spiral motion, which is also known as secondary current or transverse
circulation, the river tends to erode the outer bank, yielding deposits at the inner
bank. These phenomena cause local scouring and local pooling. Important engi-
neering efforts have been undertaken on rivers of all scales to stabilize the banklines.
Therefore quantitative information with respect to erosion and deposition must be
made available for sound river management, e.g. for navigable river systems. Depo-
sition of sediment in rivers may decrease the water depth, making navigation difficult
or even impossible. But also erosion causes problems. For example, the Iffezheim
barrage on the Rhine River near Karlsruhe needs about 170, 000 m3 of sediments
per year added artificially to avoid erosion in the downstream river bed. The costs
amount to 5 million Euro each year [1].
Sediment transport modelling is an important tool for prediction as well as diagno-
sis. Computer models permit the simulation and prediction of environmental impacts
such as river discharge, sediment grain size and its distribution and bed and bank
formation.
In this thesis, the numerical sediment transport model SSIIM will be examined.
The numerical results are tested against the measurement data taken from a physical
experiment. A new algorithm to calculate the incipient motion of sediment particles
on generalized sloping fluvial beds will be introduced as this is indispensable for a
realistic simulation. This algorithm can adequately describe the threshold of sediment
motion on a combined transverse and longitudinal sloping bed Dey [5]. Numerical
parameters will be varied to test the sensitivity to the model results.
The results will be compared to measurements of the experimental run in order to
determine the capabilities and limitations of the numerical model SSIIM. Advantages
and disadvantages of SSIIM with respect to the reproduction of the physical processes
and practical applications will be discussed.
The implementation of the additional sediment transport algorithm with the exist-
ing CFD-code constitutes an improvement with regard to the prediction of sediment
transport problems. SSIIM is able to make more reliable predictions and is therefore
a useful tool for river, environmental and sedimentation engineering.
2 Fundamentals
2.1 Governing Equations for Water Flow Calculation
Conservation of Mass The law of conservation of mass states that mass can neither
be destroyed nor created, but it can only be transformed by physical, chemical or
biological processes. All mass flow rates into a control volume through its control
surface is equal to all mass flow rates out of the control volume plus the time change
in mass inside the control volume. Written out fully in cartesian coordinates
Transport Equation In a general flow, transport of solutes, salinity or heat are due
to advection and diffusion
∂C ∂ ∂2C ∂2C
+ (ui C) = Em,i 2 + Et,i 2 ± KC, (2.4)
∂t ∂xi ∂xi ∂xi
direction, but are not homogeneous with respect to their spatial direction and are
therefore non-isotropic. A derivation of (2.4) by using Fick’s law is given in Fischer
et al. [7] and Socolofsky and Jirka [24].
Equation of State For sea water containing dissolved salt, the density % is a function
of temperature T , salinity S and the pressure p
Equation of state At a given temperature T and salinity S the fluid density is con-
stant, assuming an incompressible fluid and small variations in topography. Also the
dynamic viscosity µ is of constant value within the flow field.
Newtonian fluid For Newtonian isotropic fluids –like water considered in this study–
the viscous stresses τij in (2.3) depend on the velocity gradients and can be formulated
as [22]
∂ui ∂uj 2 ∂uk
τij = µ + − δij , (2.6)
∂xj ∂xi 3 ∂xk
where µ is a proportional factor called dynamic viscosity. The kinematic viscosity
of the fluid, which is defined as the ratio of dynamic viscosity and the density, ν =
µ
%
, depends also on the temperature T and pressure p. In the SSIIM software the
kinematic viscosity ν is hard–coded and can not be changed. It is equivalent to water
2
at 20°C, i.e. ν = 1.006 ·10−6 ms .
where the overbar indicates the time–averaged mean and the prime the turbulent
fluctuation quantities.
2 Fundamentals 4
Conservation of Mass
∂ui
=0 (2.8)
∂xi
Transport Equation
∂C ∂ ∂2C
+ (ui C) = Ei 2 (2.10)
∂t ∂xi ∂xi
Equation of state
% = %0 = const. (2.11)
2.4 Turbulence Modelling
2.4.1 Remarks
Most flows which occur in practical river, environmental and sedimentation engineer-
ing are turbulent, which means irregular fluctuation is superimposed on the main
motion. Turbulence involves disorder, is irreproducable in detail, performs efficient
mixing and transport and vorticity is irregularly distributed in all three spatial di-
mensions. The large eddies are the main carriers of kinetic energy in the fluctuations.
They obtain their energy from the mean motion. In a cascade process they decay
and pass on their energy to several small eddies. At these smaller scales of motion,
energy is dissipated by the action of viscosity, i.e. a transfer from mechanical en-
ergy to internal energy. This observation is of great importance for the numerical
modelling of turbulence. To resolve these effects, a very fine grid is required. The
grid represents the investigated natural hydraulic system in a numerical model. For
that purpose the domain must be divided into cells. Because of the huge amount of
data involved, it is only possible today to resolve the small-scale motions by direct
numerical simulation (DNS) and the use of super-computers, which is not feasible for
practical engineering purposes. For this reason the effects of the smaller-scale motion
on the main flow must be modelled.
It should be pointed out that µt (x, y, z, t) is not a physical property and not of
constant value, but rather a function of position and time, i.e. it depends on the flow
under consideration. Consequently, the distribution of µt across the flow field must
be estimated.
From dimensional analysis [19], the eddy–viscosity is proportional to three param-
eters, namely
µt ∝ %V̂ L, (2.13)
where V̂ is a velocity-scale and L characterizes the large-scale turbulent motion. The
distribution of these scales can be approximated reasonably well in many flows. The
calculation of µt from the given parameters is described in chapter 2.4.5.
2
∂ ∂ ∂ µt ∂ ∂ui ∂ui ∂uj
% + %ui = + c1 % µt + − c2 % . (2.20)
∂t ∂xi ∂xi σ ∂xi k ∂xj ∂xj ∂xi k
τ = %w gIe h, (3.1)
where τ is the shear stress, %w the density of water, g is acceleration due to gravity,
Ie represents the energy slope and h is the flow depth.
A particle motion may be initiated by shear stress τ exceeding a critical value of
the bed–shear stress τcrit , which is a function of sediment and hydraulic parameters
[29],
τcrit = f (%s , %w , D, h, ν, g, I0 ), (3.2)
where %s the density of a sediment particle, %w is the density of water and ν its
viscosity coefficient, D the mean particle size, h water depth and I0 represents the
channel slope. Since the flow under consideration is uniform and steady, the energy-
and channel slope are of equal value, Ie =I0 .
Shields [23] investigated the value of τcrit by the means of flume experiments, using
bed material of uniform size. Applying the Buckingham Π–Theorem (cf. Zanke [29]),
these seven basic parameters can be reduced to a set of four dimensionless parameters,
which are:
4. particle Froude–number, defined as the ratio of the frictional load on the grain
to the gravitational force on the grain that resists movement
u2∗ τ
τ∗ = F r ∗ = = . (3.6)
(s − 1)gD (%s − %w )gD
3 Sediment Transport Mechanism 9
The empirical dimensionless parameter F r∗ can be taken from the Shields’ diagram
found in many textbooks [9, 10, 23], where F r∗ is a function of Re∗ . The Shields’
relationship between dimensionless shear stress F r∗ (or Shields parameter τ∗ ) and
particle Reynolds–number Re∗ can be applied for predicting whether or not a particle
will move. The dimensionless value of τ∗,crit = 0.03 to 0.06 is often used as a limit
for bed protection. If the value of bed–shear τ exceeds the critical value τcrit , motion
will be initiated, expressed by following dimensionless expression
where φ is the angle of repose of sediment particles, θ is the stream–wise bed slope
and α is the transverse bed slope, as depicted in Fig. 1. Conducting experiments,
3 Sediment Transport Mechanism 10
Dey [5] found from his data the following empirical equation for the estimation of τ̃
as 0.745 0.372
θ α
τ̃ = 0.954 1 − 1− . (3.10)
φ φ
The comparison of τ̃ obtained from eqn. (3.10) with the experimental data, a cor-
relation coefficient of value 0.97 can be computed [5]. This indicates an adequate
correspondence between experimental data and given empirical equation. Figure 2
shows variations of τ̃ with α calculated from (3.10) using θ = 5.368 e−04 = 0.03 ◦ and
φ = 0.6 = 34.4 ◦ , as will be used in the further study.
In the software SSIIM, the dimensionless critical shear stress τ∗,crit is computed
by the Shields’ curve, disregarding sloping bed. As in the following an inclined
trapezoidal channel is investigated, longitudinal and transverse sloping bed must be
taken into account. Therefore, the equation (3.10) is implemented to the SSIIM
program by the author. The C++ code for the equation is listed in appendix A.
Rearranging equation (3.8) yields
τh · τ̃ = τb . (3.11)
Since the critical shear stress for horizontal bed τh is the only unknown, the author
approximates this term with sufficient accuracy by the critical Shields’ value for
incipient motion τ∗,crit from the Shields’ curve. From the represented simple empirical
equation (3.10) it can be predicted whether or not a sediment particle on a generalized
sloping bed will move when exposed to flow forces.
1.4
0.745 0.372
1.2 τ̃ = 0.954 1 − φθ 1 − αφ
1
τ̃ 0.8
0.6
0.4
0.2 φ = 34.4°, θ = 0.03°
0
-15 -10 -5 0 5 10 15 20 25 30
α in [deg]
suspended load as the part of the total which is moving without continuous contact
to the bed surface as a result of the agitation of the fluid turbulence. Therefore, a
particle only remains in suspension when the turbulent eddies have dominant vertical
velocity components exceeding the particle fall velocity.
The boundary between bed–load and suspended load is defined by van Rijn [25] as a
layer with a maximum thickness of about 10 particle diameters in which the particles
are transported as bed–load.
qb T 2.1
1.5
= 0.053 , (3.12)
((s − 1)g)2 D50 D∗0.3
where qb is the bed–load transport per unit width, D50 mean particle diameter, D∗
and T are the dimensionless particle diameter and the transport stage parameter,
respectively.
Since the shear velocity u∗ appears both in the Shields parameter (3.6) and in the
particle Reynolds–number (3.5), Julien [10] shows that the dimensionless particle
diameter D∗ can be derived by eliminating the shear velocity u∗ from the particle
3 Sediment Transport Mechanism 12
Reynolds–number by dividing by F r∗ ,
2
! 13
Re∗
D∗ = , (3.13)
F r∗
leading to
13
(s − 1)g
D∗ = D50 . (3.14)
ν2
Thus, the abscissa of the Shields’ curve can be replaced by the dimensionless particle
diameter, resulting in Fig. 3. The dimensionless transport parameter is obtained by
(u0∗ )2 − (u∗,crit )2
T = , (3.15)
(u∗,crit )2
where u0∗ is the bed–shear velocity related to grains and u∗,crit is the critical bed–
shear velocity according to Shields [23]. This parameter expresses the mobility of the
particles in terms of the stage of movement relative to the critical stage for initiation
of motion.
From the aforementioned concept, the bed–load transport rate qb is calculated by
the help of only two dimensionless parameters.
3 Sediment Transport Mechanism 13
D50 T 1.5
ca = 0.015 . (3.18)
a D∗0.3
Dividing the water body into cells, the concentration of suspended sediments in
the center of each cell can be computed by eqn. (3.17). In conjunction with the
velocity, the transport of sediments through each cell can be calculated by (2.10) and
is carried out by SSIIM.
3 Sediment Transport Mechanism 14
1
Z = 0.01
0.8
(y − a)/(h − a)
Z = 0.1
0.6
0.4
0.2
Z=1
Z=5
0
0 0.2 0.4 0.6 0.8 1
c/ca
Figure 4: Vertical concentration profile according to (3.17)
4 Flow in Curved Open Channels
4.1 Secondary Flows
Secondary flows are defined as currents that occur in the plane normal to the primary
flow direction. Their velocities are typically one order of magnitude smaller than the
bulk primary velocity [2]. Ciray [4] gives a rather functional definition on secondary
flows as follows: ”When the magnitude of the vector composed by any two compo-
nents [of] the local velocity vector[s] in a three dimensional flow is small compared
with the magnitude of the third component, the latter forms the main flow whereas
the remaining two form the secondary.”
Nevertheless, secondary currents strongly influence the velocity pattern of the
stream. Different velocity patterns in cross–streamwise and vertical direction of the
flow will be found. Secondary currents combined with the primary flow produce a
spiral flow rotating around the primary flow direction and therefore affect the pro-
cesses of flow resistance, sediment transport, bed and bank erosion and development
of channel morphology [2].
Prandtl et al. [18] distinguishes three modes of secondary flows by the releasing
forces. In the present study the secondary flow of the first type will be discussed as
it is of interest to flows under consideration. These secondary flows of the first mode
are provoked by differences in centrifugal forces.
dh v2 τrb
Ir = = βo cm + , (4.1)
dr grc %gh
where h is the water depth, rc is the radius at the center of the bend, τrb the radial
component of the friction stress at the bottom and vcm is the mean velocity. Since
the vertical velocity is unevenly distributed over the depth, an averaging factor β0 is
introduced to estimate the average velocity in streamwise direction.
τrb
As the friction term %gh on the right side of (4.1) is in many cases of small quantity,
it can be omitted. Thus the approximate equation for the inclination of the water
surface is obtained as
2
dh vcm
Ir = = βo . (4.2)
dr grc
4 Flow in Curved Open Channels 16
Eqn. (4.2) shows that at any bending of a stream on a plane, a transverse inclination
inevitable appears. The level of the free surface at the inner bank, which is nearest to
the center of the turn, will decrease. The reverse at the outer bank, which is farthest
from the center of the turn, will have an increasing water level. To conform to the
technical literature, the inner bank will be termed in the rest of this text the ”convex
bank”, and the outer bank the ”concave bank”.
Another characteristic property of a flow passing a bend is the presence of trans-
verse circulation. Rozovskii [21] considers a volume element at a distance z from the
bottom, which moves along the circular trajectories of an equal radius r as shown in
Figure 5. With the absence of radial velocity components, the friction force is taken
as zero. Again, balancing all forces acting on the volume element leads to
dh v2
Ir = = . (4.3)
dr gr
As (4.2) and (4.3) describe the identical phenomena, Ir must be equal in both cases.
This is achieved only in two cases: first, in a straight channel, where r = ∞, and
second, where v 2 of (4.3) is of constant value and equals β0 vcm
2
of (4.2), i.e. a uniform
distribution of the velocity along the vertical. It is well known that due to friction
at the boundaries of the stream, the velocity at the bottom and at the walls are
minimum. Close to the water surface they are maximum. The velocities therefore
can not be of constant value, but deviate from the mean, depending on their position
in the flow. Owing to variations in vertical velocity, the averaging factor β0 was
introduced in (4.1), so the term v 2 can not be of constant value. For particles moving
near the upper surface with a velocity faster than the mean of the flow, centrifugal
forces will be greater than the pressure forces. These particles will consequently
be displaced in a radial direction towards the concave bank. For the same reason,
particles in the lower part of the stream will move toward the convex bank. Particles
reaching the wall will move in a down- or upward direction, producing a circulation.
4 Flow in Curved Open Channels 17
resulting force of primary and transverse shear stress components. The secondary flow
near the bed leads to erosion at the concave bank and to deposition at the convex
bank, resulting in a scour and a point bar. Owing to these effects, the transverse
bed slope increases and stability of bed material decreases as threshold conditions
for incipient motion changes (cf. chapter 3.2). The morphological processes show
dynamical behaviour. Secondary flows, with their complex structure in time and
space, need three–dimensional models if investigated in more detail.
5 Numerical Model SSIIM 19
with um = Q/A = 0.093721 m/s. The flow under consideration is therefore turbulent.
Introducing the Froude–number as
um
Fr = p , (6.5)
g (A/Bsur )
1
cumulative passing
retained sieve box
0.84
0.8
fraction finer
0.6
0.4
0.2
0.16
0
0.1 1
particle size d in mm
Figure 8: Particle size distribution d of sand material
7 Numerical Simulation
7.1 Represented Domain
In Computational Fluid Dynamics (CFD) the physical domain of a flow process has
to be described in the numerical model. To represent the complex three–dimensional
geometry of the experiment mentioned in chapter 6, a structured, non-orthogonal grid
is used. Simulating all twelve bends of the experiment, the numerical grid consists of
623 in longitudinal and 16 grid cells in transverse direction. In the vertical direction,
8 layers are chosen, as depicted in Fig. 10. After having carried out first simulation
runs, similar flow and sediment patterns could be observed at different but equivalent
positions. In order to save computation time, simulating only four complete bends
instead of all twelve seems to be sufficient. The number of grid cells in the longitudinal
direction is reduced to 260 cells. Thus, the longitudinal grid size in the middle of the
channel amounts to ∆x = 0.043 m and in the transverse direction to ∆y = 0.01875 m,
while the total water depth of 0.05 m is resolved in 8 evenly spaced vertical layers,
each 0.00625 m high. Fig. 9 shows the mesh layout, seen from above. The flow is
from left to right.
of adequate accuracy in a reasonable time. From the water flow pattern, qualitative
predictions as to sediment movement can be made.
the right. The vector arrows represent the velocity components v and w in transversal
and vertical direction. The main secondary motion at the water surface is towards
the concave bank. There, the fluid elements move downwards. Reaching the bottom,
the fluid elements are vectored inwards to the convex bank. Consequently, the fluid
exhibits a cross-circulation. The combination of this cross-circulation and the major
flow direction results into fluid spiral motion.
The velocity vector pattern reveals a further phenomena. In subsequent bends,
besides the main spiral motion, a second, less extensive counterrotation appears.
This phenomena has been observed in experimental studies by Mosonyi and Götz
[14], if the ratio of width Bsur to waterdepth h is of lower value than Bsur/h < 10. In
this case the ratio is Bsur/h = 6 < 10.
The second spiral motion could only be resolved using a fine grid as mentioned above.
the concave bank. On entering the subsequent bend, the velocity peak keeps moving
in its prevailing direction and approaches the convex bank, where it reaches its top
maximum velocity within the bank inclination.
It should be pointed out that the velocity can not be equated with the discharge.
Applying the law of conservation of mass (cf. eqn. (2.8)), the discharge Q equals
the product of velocity u and cross-sectional flow area A. Approaching the inclined
banks, the cross-sectional flow area A decreases. In spite of high velocity, the main
discharge Q will not take place at the bank location. As the velocity is important
for the sediment transport mechanism, the author emphasizes the representation of
velocity distribution instead of discharge distribution at the chosen cross-sections.
From the inclined banks the maximum velocity decreases and the peak moves back
to the middle of the cross-section. At the apex of the bend, the velocity distribution
is symmetrical to the center column of the preceding apex.
F235
F250 × r r
r
2 F265 b r r
r
r
F274 M
?r ? ?
F283 r ?r ?
?r
dimensionless velocity scaled by um
F298 ?
××× ?r M M M M M
1.5 F313 × b b M M MM r
× ?b b
× b b M
×r ? M
? × b b
M M
× b M
?
r
× b b
b M r × b
? × b
× M b M
1 b ? r ×
M × ?b
× b ? r × b
M ×
b ? r × b
× M
? ×
b r × b
0.5 M
?
? r ?
×b M r ×
?
b ?M r
×
?M
?b M r r
?
0×?rb
M ? ? b× Mr r ? × ?rb
M
? b×r
M ?
?r ? Mr
r rb Mrb ×
Mb Mb M
×
×× ×
−0.5
5 10 15 20 25 30
transverse direction (cell number)
Figure 13: Longitudinal, depth–averaged velocity distribution
7 Numerical Simulation 30
mass, the fluid elements at the convex bank are decelerated and a negative pressure
gradient established, leading to a separation zone, expanding until the following apex.
Rozovskii [21] found from experimental results that formation of separation zones are
unlikely, even at sharp bends with a small ratio of center line curvature rc to width
B of rc/B = 1. In the investigated case the ratio equals rc/B = 0.916 < 1. To the
extent that stream separation is connected to the negative acceleration effect of fluid
elements at the side wall, Rozovskii [21] asserts that conditions for the appearance
of separation will also depend on the angle of bank slope. He concludes that the
possibility of separation increases, the gentler the bank slope is, or in short, the greater
the influence of friction of the stream against the bank. As in the case considered
the transverse bed slope angle equals α = 30 ◦ , the calculated flow separation zone is
likely, even in the case of a center line curvature radius to a width ratio close to one.
critical shear stress τ∗,crit by Shields [23]. No changes have been made in the software
code of SSIIM. Since no modification has been made running the numerical model,
the configuration is denoted as default case. Water flow and sediment parameters are
used as described in chapter 7.2.1 and 7.3.1, respectively.
The results achieved by conducting the default case is depicted in Fig. 15. SSIIM
predicts almost no changes in bed topography, which is not consistent with the ob-
servations made during the experimental run. Repeating the simulation and testing
increasing roughness height up to ks = 10d50 = 0.007 m, the sediment transport pre-
dictions get closer to measurements as a result of increasing shear stress. But one has
to be aware of an unphysical roughness height, which is not be found in common lit-
erature. Concluding from the conducted tests, SSIIM necessitates modification in the
critical shear stress τcrit computation. An additional algorithm must be implemented
in order to achieve realistic results, when decreasing the roughness height in value
back to physical range. Physically reasonable ranges of the roughness coefficient ks
are introduced in Wilson et al. [28].
As described in chapter 3.2, channel geometry influences the threshold condition
for incipient motion of uniform sediment particles as it reduces or increases the crit-
ical shear stress τcrit . Fig. 10 depicts a representative cross-section of the channel
geometry and shows that half of the river bed is inclined in a transversal direction.
It is assumed that the inclined banks will have a determining influence on incipient
motion and therefore on the sediment transport calculation. Eqn. (3.10) takes a lon-
gitudinal as well as a transversal sloping bed into account. Following calculations are
carried out by using eqn. (3.10), adjusting the critical shear stress τcrit .
7 Numerical Simulation 35
Erosion Sediment material erodes from upstream of the point bar. The particles dis-
lodged upstream are the source of the particles deposited at the convex bank of the
next apex, which is described in the paragraph above. Consequently, the position
where particles enter motion is simultaneously the position where the shear stress
applied on the bed material exceeds the critical shear stress. From a water flow
calculation point of view, erosion seems to be predictable at zones of high velocity.
Again, by the use of the longitudinal velocity distribution depicted in Fig. 13, high
velocity coincides with the area of high erosion.
Calculations achieved by SSIIM and measurements show the same characteristic
trend. Interpreting the measurements, erosion takes place upstream of the bend
apex, but with a smaller expansion than predicted by SSIIM, which overpredicts
the erosion. The erosion zone at the concave bank downstream of the apex can not
be observed in Fig. 21. It is assumed that the appearance of the erosion zone in
the measurement data arises from imprecisions of the gauging device. The gauging
equipment is not able to reproduce elevation differences in the required resolution.
Nevertheless, the longitudinal position of the deposition and erosion zone is identical
in both cases.
the variation of absolute values in deposition and erosion as assumed by the author,
movement of the point bar position is observed. Applying a reduced ks –value com-
pared to the base case, the point bar moves upstream in the direction of the erosion
zone. Increasing the hydraulic roughness, the opposite effect is observed. The point
bar travels in flow direction downstream and becomes stretched. A second scour
point next to the deposition forms.
The enlargement of the deposition zone is explained by the simultaneous quantitative
increment of erosion. The sediment particles try to deposit themselves at the apex for
the reason mentioned in paragraph Base Case above. During the bed changes, the
bed slope may increase to a level above the angle of repose φ. The sediment particles
are not able to drop in this cell and are forced to move sideways to a neighbouring
cell. A stretching of the deposition zone occurs, as long as no state of equilibrium is
achieved. This effect is supported by the bed sloping algorithm implemented by the
author (cf. chapter 3.2).
Variations of roughness height have no effect on the position of erosion area.
Figure 17: Bed elevation changes with regard to different bed roughness heights
Grid size The grid independence of the solution is checked using a finer mesh, where
the number of gridlines is doubled in all directions. It is exactly the same mesh
as used for the water flow calculation resolving the two spiral motions as described
in chapter 7.2. The amount of cells increases by a factor 23 , resulting in very long
computation time for time–dependent sediment transport. Investigating the time
step as discussed above, extending the time step to ∂t = 60 s has no influence on the
solution and fewer iterations have to be calculated, for the same simulation time of
∆t = 4 h. Still, the computation time is very long and is not suitable for practical
engineering purposes. The computation time for carrying out a simulation time of
∆t = 4 t is shown in Table 2.
Doubling the grid in all three spatial directions is a reasonable approach in order
to test the grid independence. The aspect ratio remains unchanged, so influences of
a distorted grid are eliminated. Fig. 18 shows the predicted bed formation in a bend,
using a time step of ∂t = 60 s. The scour point is still at the same position. The
deposition process tends to separate into two individual pool bars. Between them,
a scour point develops. Comparing the numerical result to another photo taken of
a bend during the experimental run, similarities are observed. At a different bend
a different bed formation pattern appears as presented in Fig. 22. To gather from
this observation, neither the experimental run nor the numerical simulation seems
to tend to a prefered solution. The predictions show variations in detail, but the
main characteristic, which is the formation of a pool bar at the convex apex, remains
unchanged.
Wilson et al. [28] conducts numerical water flow simulations of a pseudo–natural
meandering flow, using a distorted computation mesh in a longitudinal direction.
To check if the result is mesh independent, a simulation is conducted using a mesh
with double the number of gridlines in a longitudinal direction, making the mesh less
distorted. The computed velocities from the two grids were found to be reasonably in
agreement. Mesh independence is confirmed and even a distorted mesh gives accurate
results.
Also a distorted mesh was tested for sediment transport calculations in this study.
Therefore, the number of grid cells in only the transversal and vertical direction was
doubled. Simulations having a time step of ∂t = 20 s and ∂t = 60 s were carried out.
The sediment calculation results are presented in Fig. 19. Considering a time step of
∂t = 60 s of the fine mesh calculation and comparing the results achieved by a fine
but distorted mesh, the bed formation pattern is quite similar. This observation leads
to the conclusion that the accuracy of the solution is not affected by a refinement of
the grid in a longitudinal direction.
7 Numerical Simulation 39
Figure 19: Sediment calculations with distorted mesh at different time steps
tion of the physical processes, e.g. false diffusion generated by the numerical grid or
non-orthogonal diffusive terms, which are neglected for the calculation of the sedi-
ment concentrations. An injudicious parameter choice can produce highly erroneous
results, even if remaining parameters are set correctly. On the other hand, a reason-
able diligence in the model establishment by the user may minimize these errors.
The remedy to prevent false diffusion is to align the grid with the fluid direction
and/or increase the number of elements in the grid [15]. The author has tested the
second approach. As described in chapter 7.3.4, doubling the number of cells in each
direction does not change the accuracy of the solution. Consequently, inaccuracy is
not caused by false diffusion. Another alternative is to use a second-order method as
applied throughout all simulations.
The third category of error mentioned concerns the uncertainty of physical prop-
erties, e.g. roughness height and angle of repose φ. The latter is set to a constant
value as recommended by the literature for submerged sand particles. During the
experimental runs, φ varies with time. At the beginning of the experiment the sand
is dry and becomes saturated with water gradually. The variation of φ during the
experiment is time-dependent and is not reproduceable in SSIIM.
Special attention is paid to the equivalent sand roughness coefficient ks as it influences
the bed friction and therefore the bed–load transport. In the numerical simulation,
the ks –value is adjusted until the model output matches the measured conditions
by the experiment. This is achieved using an equivalent sand roughness height of
ks = 0.0021 m, which corresponds to ks = 2.3d90 , if d90 = 0.0009 m as seen in Fig. 8.
The value found from this calibration is in the range of empirically found values pre-
sented by Wilson et al. [28].
Fig. 16(a) shows an overprediction in erosion by the SSIIM software. The sediment
material located at the convex bank is carried away by current force. The photos
shown in Fig. 21 and Fig. 22 taken from the bend do not exhibit such strong erosion,
but rather demonstrate resistance to the acting force. Speculating about this observa-
tion, the sediment particles could exercise cohesion forces and stick together, having
the effect of variation in threshold condition for incipient sediment motion. Besides
this explanation, there could also be a compaction of the sediment material during
the formation of the trapezoidal shape. The mentioned influences are very uncertain
and almost impossible to quantify or to consider in the numerical simulation.
8 Conclusions and Outlook
This thesis presents a comparison of simulations performed in the numerical model
SSIIM and a physical model. The objective is to examine the modelling results
with respect to the sediment processes observed in the experimental run in order to
determine whether or not the numerical model SSIIM is able to predict sediment
transport correctly.
The physical process of sediment transport and its implementation in the numerical
model was described. Furthermore, the software code was extended by a subroutine
taking generalized sloping bed for incipient particle motion into account [5]. The sub-
routine emerges as the key for achieving improved results, reproducing the sediment
processes as closely as possible to the experiment.
Influencing factors are the calibration coefficients, i.e. the equivalent sand rough-
ness height. Suggestions from the literature were adopted, but further simulations
have shown that small variations in roughness will strongly effect the bed formation
pattern. The time step variation shows minor change, so that this parameter should
not be taken as a calibration coefficient of the numerical model. The degree of its
influence on the results in combination with changes in grid size could be an interest-
ing topic for further studies. But due to long computation times, such investigations
can not be carried out in a reasonable time with present computer capacity.
The numerical model results have proved to be grid size independent in combination
with the time steps chosen by the author. A refined grid shows separation of the point
bar, reflecting observations documented by photos but not by measurements.
The software SSIIM performs capably in reproducing sediment transport processes
in a realistic way, even with regard to sharply curved bends. The calculation can be
carried out if the user establishes reasonable model assumptions and approximations.
In this thesis, results in good agreement to the experiment have been achieved even
with the coarsest grid tested and a time step of ∂t = 60 s for a simulation time
of ∆t = 4 t. The computation time is as long as it is appropriate for practical
engineering purpose. Refining the computational mesh or decreasing the time step
means extending computation time, which is not feasible with the power of present–
day personal computers (cf. Table 2). Conducted calculations concerning variation
of the mentioned numerical parameters do not yield any improvement of the result
accuracy. They show variations of the bed formation pattern as also observed in
the experiment. However, without having field data, running reliable simulations is
hardly not possible. But even if applying unphysical parameters, SSIIM is able to
indicate the trend of pool- resp. point bar formation.
SSIIM’s extension by the implemented subroutine shows that the software is a re-
liable instrument in river, environmental and sedimentation engineering. The ability
to run the software on standard personal computers will promote its distribution
among practical working hydraulic engineers and will cause it to be accepted as a
design tool for water and environmental related engineering.
A Listing C++ Subroutine 42
/************************************************************/
/**** correction for sloping bed ****************************/
/**** only called if F 7 B is used***************************/
/************************************************************/
// variables:
// vectorX: component in x direction of unit vector normal
to the bed surface
// vectorY: component in y direction of unit vector normal
to the bed surface
// vectorZ: component in z direction of unit vector normal
to the bed surface
// velocityX: velocity in x direction in bed cell
// velocityY: velocity in y direction in bed cell
// param1: parameter in formula
// param2: parameter in formula
eta = 0.954;
phi = param1;
teta = param2;
alpha = fabs(acos(vectorZ));
tau = 0;
PT = phi/teta;
//function//
tau = max(0.333,
eta * pow(1-(PT),0.745) * pow(1-(AT),0.372)
A Listing C++ Subroutine 43
C Photos
Figure 20: Channel before experimental run, Guymer and Dutton [8]
C Photos 47
Figure 21: Photo of channel bend showing one point bar, flow is from left to right,
Guymer and Dutton [8]
Figure 22: Photo of channel bend showing separated point bars, flow is from left to
right, Guymer and Dutton [8]
References 48
References
[1] Wasser- und Schifffahrtsamt Freiburg. Information Flyer, 2004.
[2] Bathurst, James C., Thorne, Colin R., and Hey, Richard D. Secondary flow and
shear stress at river bends. Journal of the Hydraulics Division, 10:1277–1295,
October 1979.
[3] Chanson, Hubert. The Hydraulics of Open Channel Flow: An Introduction. El-
sevier Butterworth–Heinemann, Amsterdam, Bosten, Heidelberg, London, New
York, Oxford, 2 edition, 2004. ISBN 0–7506–5978–5.
[4] Ciray, Cahit. On secondary currents. The International Association for Hy-
draulic Research, 1:408–413, September 1967. 12th IAHR–Congress, Fort
Collins, 11–14. September 1967.
[7] Fischer, Hugo B., List, E. John, Koh, Robert C. Y., Imberger, Joerg, and Brooks,
Norman H. Mixing in Inland and Coastal Waters. Academic Press, New York,
London, Toronto, Sydney, San Francisco, 1979. ISBN 0-12-258150-4.
[8] Guymer, Ian and Dutton, Richard. Measurement data and photographs of phys-
ical experiment. unpublished, 2003. kindly left to the author.
[10] Julien, Pierre Y. Erosion and Sedimentation. Cambridge University Press, Cam-
bridge, United Kingdom, 1989. ISBN 0–521–63639–6.
[12] Leschziner, Michael A. and Rodi, Wolfgang. Calculation of strongly curved open
channel flow. Journal of the Hydraulics Division, 10:1297–1314, October 1979.
References 49
[14] Mosonyi, Emil and Götz, Werner. Secondary currents in subsequent model
bends. International Symposium on River Mechanics, 1:191–201, January 1973.
Proceedings of the International Association for Hydraulic Research, 9–12 Jan-
uary 1973, Asian Institute of Technology, Bangkok, Thailand.
[17] Patankar, Suhas V. Numerical heat transfer and fluid flow. Hemisphere
Publ. Corp., New York, 1980. ISBN 0-89116-522-3.
[18] Prandtl, Ludwig, Oswatitsch, Klaus, and Wieghardt, Karl. Führer durch die
Strömungslehre. Friedr. Vieweg & Sohn, Braunschweig, Wiesbaden, 1984. ISBN
3–528-18209–5. 8. vollständig überarbeitete Auflage.
[19] Rodi, Wolfgang. Turbulence Models and Their Application in Hydraulics. Inter-
national Association for Hydraulics Research, Balkema, Rotterdam, 3rd edition,
1993. ISBN 90 5410 150 4. A state–of–the–art review.
[21] Rozovskii, I. L. Flow of Water in Bends of Open Channels, volume PST Cat.
No. 363. Academy of Sciences of the Ukrainian SSR, Kiev, 1957. translation by
Y. Prushansky, Israel Program for Scientific Translation, 1961.
[22] Schlichting, Herrmann and Gersten, Klaus. Boundary Layer Theory. Springer
Verlag, Berlin, Heidelberg, New York, 8th revised and enlarged edition, 2000.
ISBN 3-540-66270-7.
[25] van Rijn, Leo C. Sediment Transport, Part I: Bed Load Transport. Journal
of Hydraulic Engineering, American Society of Civil Engineers, 110:1431–1456,
October 1984. No. 10.
[26] van Rijn, Leo C. Sediment Transport, Part II: Suspended Load Transport.
Journal of Hydraulic Engineering, American Society of Civil Engineers, 110:
1613–1641, November 1984. No. 11.
[27] van Rijn, Leo C. Sediment Transport, Part III: Bed Forms and Alluvial Rough-
ness. Journal of Hydraulic Engineering, American Society of Civil Engineers,
110:1733–1755, December 1984. No. 12.
[28] Wilson, C.A.M.E., Boxall, J.B., Guymer, I., and Olsen, N.R.B. Validation of a
three–dimensional numerical code in the simulation of pseudo–natural meander-
ing flows. Journal of the Hydraulic Engineering, 10:758–768, October 2003.
[29] Zanke, Ulrich. Grundlagen der Sedimentbewegung. Springer Verlag, Berlin, Hei-
delberg, New York, 1982. ISBN 3–540–11672–9.