Calculation of Added Mass Close To Seabed
Calculation of Added Mass Close To Seabed
Calculation of Added Mass Close To Seabed
( )
Formula 1 Navier-Stokes
equation
Where U is the fluid vector ([u v w]
T
), is the differential operator, is the density of the fluid, p is
the pressure, is the kinematic viscosity and g is the acceleration of gravity.
If we assume an incompressible flow of a Newtonian fluid this leads to simplifications. The
assumption of incompressibility rules out the possibility of sound or shock waves to occur. The
incompressible flow assumption holds if we are dealing with fluids at low Mach numbers, say M < 0.3
Formula 2 Mach number.
where M is the Mach number, V is velocity of a fluid, and a is the velocity of sound in that medium.
This assumption leads to simplifications to the Navier-Stokes equation, ruling out the first part on the
left hand side in Formula 1.
2.2 Flow regimes
Flow regimes have been classified in numerous manners, but they all resort to the non-
dimensionalised Reynolds-number. The Reynolds-number gives the relation between inertial forces
and viscous forces in the boundary layer. It is defined as
Formula 3 Reynolds number
Where the density for the fluid, U is is the velocity, D is the diameter of our object, is the
kinematic viscosity.
A somewhat crude division is found in (Sumer & Fredse, 1997), but one should be aware of that the
divisions of the various regimes into Reynolds number ranges are not definite. Various imperfections
and disturbances may have a profound influence of the flow and change the Reynolds-ranges for
where the various regimes are seen. Such disturbances and imperfections may be surface roughness,
inflow turbulence and various quality of the cylinder, e.g. welding seams etc.
3
Figure 1 Description of flow regimes. (Sumer & Fredse, 1997)
2.3 Forces on a cylinder
Because of its central importance in fluid mechanics, the flow about a cylinder and its resulting forces
will be described in the following chapter. The following theory is found in (Faltinsen, Sealoads on
ships and offshore structures, 1990).
If one would calculate the forces on a slender object, Morisons equation (Morrison, O'Brien,
Johnson, & Schaaf, 1950) is often used to calculate wave loads on circular cylindrical structural
members of fixed offshore structures when viscous forces matter. Formula 4 Morrisons equation
reads the horizontal force dF on a strip of length dz of a vertical rigid circular cylinder.
Formula 4 Morrisons equation
4
The mass and drag coefficients C
m
and C
d
have to be empirically determined through the use of CFD
or experimentally, and are dependent of the parameters such as Reynolds number and Keulegan-
Carpenter number. What one must bear in mind is that Morisons equation not at all can predict the
forces orthogonal to the wave propagation direction and in the cross-sectional plane.
When viscous forces effects matter for offshore structures the flow will generally separate. A
consequence of separation is the time-varying pressure-forces due to viscous effects are more
important than shear forces. The lack of a precise definition of what is meant by separation in
unsteady flow (Faltinsen, Sealoads on ships and offshore structures, 1990) makes it somewhat
difficult to compare the different data sets and experiments made by various researchers. For
instance, P.W Bearman reported that nobody had seen vortices at KC-values less than 3. (Bearman,
1985). But one year later, Sarpkaya reported of separation occurring at KC=1.25 (Sarpkaya T. , 1986).
Clearly, vortex shedding is also a matter of Reynolds number, not only KC.
The following section is found in (Sarpkaya & Isaacson, 1981).
Given Newtons second law, Formula 5 Newtons second, two of the given three quantities must be
known in order to give the last one. In vacuum, this is straight forward as far as the mass of the
accelerated body is concerned. In a fluid with a given density, this takes a bit more consideration,
since the fluid of the motion as well as of the body itself influences the effective mass.
Formula 5 Newtons second
law
The induced mass varies with the instantaneous shape and volume of the wake or cavity as well as
with their rates of change. The magnitude of the added mass or added mass-moment of inertia at a
certain moment depends on the time history of the motion.
2.3.1 Added mass
One easy way of describing the added-mass, is given by (Sumer & Fredse, 1997); The
hydrodynamic mass is defined as the mass of the fluid around the body which is accelerated with the
movement of the body due to the action of pressure.
Another way of describing it, a bit more theoretical is given by (Sarpkaya & Isaacson, 1981).
Consider an unsteady motion of a cylinder normal to its axis in a fluid otherwise at rest. The potential
function may be represented by Formula 6 Potential function, where U is the velocity of the cylinder,
G(r) is of order r
-n
(n=2,3,4 ..).
()
Formula 6.1 Potential function
Then the kinetic energy of the fluid becomes
5
()
Formula 6 Total kinetic energy
Inserting Formula 6.1 Potential function into Formula 6 one gets
Formula 7 Total kinetic energy
Where dS is an elemental area of the cross section and the integral is over the cylinder surface. The
work done per unit time, which makes power, must equal the time rate of increase of the kinetic
energy of the fluid. Therefore we get Formula 8.
Formula 8 Rate of increase of
the kinetic energy of the fluid
Which could be rewritten further, the added mass coefficient is the ratio of added mass over the
density of the fluid multiplied by the volume of the object, Formula 9.
)
Formula 9 Added mass
coefficient
The body has a mass M
b
of its own and in real fluids it experiences a resistance primarily due to
separation of the flow or the resulting forces and a combination of these two. This is called the form
drag. If one does the assumption of adding the instantaneous values of the drag and inertial forces,
the total force acting on a cylinder moving unidirectional with a time-dependent velocity U(t) turns
out to become
()]
Formula 10 Total force acting
on a moving cylinder
In real life, the added mass coefficient and the drag coefficient depends on time, Reynolds number,
Keulegan-Carpenter-number (which will be investigated further in later section) and on the
parameters characterizing the history of motion. The consequences of this are that each data-series
is unique, and must be treated accordingly. As will be shown later, this leads to a time-varying lift-
and drag-force.
2.4 Flow around a cylinder in oscillatory flows
Harmonically oscillating bodies with small amplitude have been used for a long time in order to
determine the added mass of bodies of various shapes.
When an object, say a cylinder, is subject to a harmonic flow normal to its axis, the cylinder
experiences a flow accelerating from and decelerate to zero, but the flow changes direction as well
during one cycle. This in turns leads to a reversal of the wake, complicating the inflow-conditions
whenever the velocity changes sign. The boundary layer over the cylinder may undergo changes
6
varying from fully laminar to fully turbulent states, and the Reynolds-number may change from sub-
critical to post-supercritical over a given cycle (Sarpkaya & Isaacson, 1981).
The vortices which have been formed or have been shed during the first half of a cycle are being
reversed around the cylinder during the wake reversal, which in turns give rise to a transverse force
with or without vortex shedding. Particularly significant are the changes in the lift, drag and inertial
forces when the reversely-convected vortices are not symmetric.
The following chapter is found in (Sumer & Fredse, 1997).
Say that our cylinder with diameter D is situated in a harmonically oscillating fluid transverse to the
cylinders axis. After some time, when the start-up effects have decayed one can describe the flow
with another parameter than just the Reynolds number. The Keulegan-Carpenter number is defined
by
Formula 11 Keulegan-
Carpenter number.
Where the U
m
is the maximum velocity and T
w
is the period of the oscillatory flow. Say that our flow is
sinusoidal with the velocity given by
()
Formula 12 Sinusoidal velocity
The maximum velocity then turns out to become
Formula 13 Maximum velocity
Where a is amplitude of the motion.
A physical meaning of the Keulegan-Carpenter number is that small KC-number means that the
orbital motion of the water particles is small relative to the total width of the cylinder. When KC is
very small, say smaller than 4, vortex shedding may not even occur. Any distinct answer to when
separation occurs is hard to give, as stated above, as this is connected to the Reynolds number.
Large Keulegan-Carpenter numbers means on the other hand that the water particles have to travel
large distances relative to the total width of the cylinder.
Figure 2 Various flow-regimes for a smooth circular cylinder in oscillatory flow, Re=10^3. Figure 2
gives the various regimes for flow around a smooth circular cylinder in an oscillatory flow. What one
must bear in mind, that this division is dependent on the Reynolds-number. This can be seen in
Figure 3.
The region divided by a has no separation, it is a creeping flow. The region a has no separation,
but the boundary-layer is turbulent. The region b has separation with Honji-vortices. The region c
has a pair of symmetric vortices. The region d has a pair of symmetric vortices, but turbulence over
the cylinder surface.
7
Figure 2 Various flow-regimes for a smooth circular cylinder in oscillatory flow, Re=10^3. (Sumer & Fredse, 1997)
8
Figure 3 Flow-regimes around a smooth, circular cylinder in oscillatory flow at various Reynolds-numbers. (Sumer &
Fredse, 1997)
9
3. Pre-processing
3.1 Mesh theory
Each simulation with the aid of CFD has the same step one must go through in order to get a
satisfactory result. The steps are pre-processing, running and post-processing. The following chapter
will give a brief introduction into what is happening during pre-processing and simulation when using
dynamic mesh. The following sub-topics are chosen on the basis of what the author self has
encountered.
The dynamic mesh model in Ansys Fluent provides the opportunity to move the boundaries of a cell
zone relative to other boundaries of the zone, and to adjust the mesh accordingly. The motion of the
boundaries can be rigid, like a piston moving inside a cylinder, or a foil changing the angle of attack as
the time goes on, or deforming, such as the elastic wall of a balloon during inflation. Either way, the
nodes that define the cells in the domain must be updated as a function of time, and hence the
dynamic mesh solutions are inherently unsteady (Ansys Fluent, 2012).
The integral form of the conservation equation for a general scalar, , on an arbitrary control volume
V, whose boundary is moving, can be written as Formula 14.
Formula 14 Conservation
equation for a general scalar
on an arbitrary control volume
V, whose boundary is moving.
Where is the fluid density, is the flow velocity vector,
()
()
Formula 15 Time derivative
term.
The n+1th time level, V
N+1
, is computed from
Formula 16 nth time level
volume
Where
is the volume time derivative of the control volume. We must satisfy the mesh
conservation law, so the volume time derivative of the control volume is computed from following
relationship
10
Formula 17 The volume time
derivative of the control
volume
Where
Formula 18 The increase of
volume by time.
Where
is the volume swept over by the control volume face j over the time step .
3.1.1 Sliding mesh
The sliding mesh feature gives you the possibility to set up a problem where separate zones move,
rotationally or translational, relative to each other. The nodes move rigidly in a prescribed dynamic
mesh zone given by the user.
Additionally, multiple cells zones are connected with each other through non-conformal interfaces.
Since the mesh motion is updated in time, the non-conformal interfaces are likewise updated to
reflect the new positions each zone. See Figure 4
Figure 4 Sliding mesh with non-conformal mesh (Jasak)
The general conservation equation formulation given for dynamic meshes above, see Formula 14
Conservation equation for a general scalar on an arbitrary control volume V, whose boundary is
moving.Formula 14, is also valid for sliding meshes. Since the mesh motion in our dynamic sliding
mesh formulation is rigid, all cells have their original shape and volume. This gives us, that the time
rate of change of the cell volume is zero, and Formula 16 is simplified to the following;
Formula 19
11
And Formula 15 leads to the following simplification;
()
()
Formula 20
Finally, Formula 20, leads to the following simplification
Formula 21
A consequence to this dynamic mesh is that the solutions for Formula 14 for sliding meshes are
inherently unsteady, as all dynamic meshes are.
3.1.2 Smoothing methods
The following part is found in the User Manual for Ansys Fluent.
If one chooses the spring-based smoothing, one can look upon the edges between any two mesh
nodes as idealized network of interconnected springs, where the initial spacings of the edges before
any mesh motion is the equilibrium state of the mesh. Hooks law is applied in order to find the
forces proportional to the displacement along all the springs connected to the node.
Hooks law can be written like
)
Formula 22
Where
and
is the number of
neighboring nodes connected to node i, and
|
Formula 23
When equilibrium occurs, the net force on a node due to all the springs which are connected to the
node must equal zero. This condition gives an iterative equation,
Formula 24
12
Since we already know the displacements at the boundaries, Formula 24 is solved with a Jacobi-
sweep on all the interior nodes. When convergence occurs, the positions are updated giving us
Formula 25
Where n+1 and n are used to denote the positions at the next and current time step, respectively.
In the GUI in Ansys Fluent, you can prescribe the spring constant factor.
3.1.2 Dynamic Layering
If the mesh consists of hexahedral and/or wedge mesh zones, one can use dynamic layering to add
and remove cell-layers based on a preset height of the layer adjacent to the moving surface. The
layers which exceed this preset height merge or collapses into the next cell layer, see Figure 5 Cell
layering to a moving boundary
Figure 5 Cell layering to a moving boundary (Ansys Fluent, 2012)
If the cells in the layer j are expanding, the cell height is allowed to expand to increase a certain limit
is met, Formula 26.
Formula 26
Where
is the
layer split factor.
On the other side, if the cells in the layer j are being compressed, they can be compressed until a
certain level is met,
Formula 27
Where
is the layer collapse factor. See Figure 6 to get an understanding of what is happening.
In most cases, mesh sliding and layering is superior to other dynamic meshing, when it comes to less
13
numerical diffusion, time consumed updating the mesh. And the fact that if the mesh consists of
quads and hexes, a smaller number of faces or cells are needed, compared to tris and tets.
Figure 6 Sliding and layering (Ansys Fluent, 2012)
3.1.3 Local remeshing
On zones with a triangular or tetrahedral mesh, the spring-based smoothing method is often used.
When the boundary displacements are large, at least if one compares it with the local cell sizes, the
quality of the cell tends to deteriorate. This will eventually invalidate the mesh, and give the user a
non-satisfactory result. In order to solve this problem, Fluent marks the cell which violates the preset
skewness or size-criteria. Next Fluent locally remeshes the agglomerated cells or faces, given that the
new cells or faces satisfy the skewness criteria. Otherwise the new cells or faces are discarded.
Fluent evaluates each and every cell out of the following criteria; maximum skewness, smaller than a
specified length, larger than a specified length and that its height does not meet the specified length
scale at moving face zones.
3.2 Turbulence-modeling
Turbulent fluid motion is an irregular behavior of a flow where the various quantities such as
velocity, pressure, concentration, temperature etc., show a random variation with time and space
coordinates. The most accurate approach to turbulence simulation is to solve the Navier-Stokes
equations without averaging or approximation other than numerical discretizations whose errors can
be estimated and controlled (Ferziger & Peric, 2002).
In order to assure that all of the significant structures of the turbulence have been captured, the
14
domain must be at least the size of the physical domain or the largest turbulent eddy. A valid
simulation must capture all of the kinetic energy dissipation.
This is however not feasible for the nearest future, given the huge amount of time needed. As of
now, the largest DNS-simulation done used 4096
3
mesh points, which isnt very much. This
introduces the necessity for modeling turbulence.
3.2.1 Reynolds Averaging
When one chooses a RANS-turbulence model, say k- or k-, the solution variables in the
instantaneous Navier-Stokes equation are decomposed into the mean and fluctuating components
(Ansys Fluent, 2012), see Figure 7.
Figure 7 Reynolds-Averaged Navier-Stokes, (Lafforgue)
For the velocity components, this gives us Formula 28,
Formula 28
Where
and
are the mean and fluctuating velocity components, where i=1,2,3 indicating
direction.
The same principle counts for pressure and other scalars, Formula 29.
Formula 29
Where denotes a scalar such as pressure, energy or a concentration of a species.
If we substitute the expression given in Formula 28 into the instantaneous continuity and
momentum equations, and taking the time-averaged, this yields averaged momentum equations.
Formula 30
(
[(
)]
)
Formula 31
15
The equations given in Formula 30 and Formula 31 are called the Reynolds-Averaged Navier-Stokes
equations. Additional terms do now appear compared to the traditional Navier-Stokes equation
given in Formula 1 Navier-Stokes equation. These terms, the Reynolds stresses (
), must be
modeled in order to close Formula 31.
3.2.1.1 Choosing the proper turbulence-model
After discussion with co-supervisor Tufan Arslan, the choice fell on k- with Shear-Stress Transport
(SST) for turbulence modeling regarding case two. A brief description found in the User Guide for
Ansys Fluent is given below.
The SST-version of the k- was developed by Menter. The target were to effectively blend the robust
and accurate formulation of the k- model in the region near wall, with the free-stream
independence found in k- model in the far field. The features implemented by Menter makes the k-
SST-model more accurate and reliable for a wider specter of flows, such as for airfoils, transonic
shock waves and adverse pressure gradient flows. Further elaboration into the turbulence modeling
is not presented in this thesis, as this is out of this thesis scope.
16
4. Running
4.1 Choosing the right solver, convergence criteria and time step.
The fundamental concept of the different schemes like PISO and SIMPLE (pressure-implicit with
splitting of operators and semi-implicit method for pressure-linked equations, respectively), is to
derive a pressure correction equation by enforcing mass continuity over each cell, (Barton, 1998).
According to the User-Guide for Ansys Fluent, steady-state calculation uses in general SIMPLE or
SIMPLEC, while PISO is recommended for transient calculations. PISO is in a larger number of cases
supreme to its competitors in transient simulations, since it has the possibility of having a larger time
step than the other schemes and a stable calculation, (Ansys Fluent, 2012). PISO may also be useful
for steady-state and transient calculations on highly skewed meshes.
Below is a work-chart for the PISO- and SIMPLE-schemes, in pseudo-code (Chen, 2009).
SIMPLE-scheme;
1. Set the initial velocity and pressure field Uo and po.
2. Solve the under-relaxed discretized momentum equation to compute the tentative velocity
field U*.
3. Solve the pressure correction equation to update the pressure field using under-relaxation
factor p.
4. Correct the velocity field
5. Return to step (1) until convergence.
PISO-scheme
1. Set the initial or old velocity and pressure field Uo and po.
2. Solve the discretized momentum equation to compute the tentative velocity field.
3. Corrector step 1:
a) Solve the pressure correction equation to update the pressure field.
b) Correct the velocity field
4. Corrector step 2:
a) Repeat corrector step 1 with for the pressure correction equation
b) Correct the velocity field using
5. Return to step 1 until convergence.
For both cases, case 1 and case 2, the PISO-solver is the chosen one.
When it comes to choosing convergence-criteria, the default setting in Fluent is 10^-3. This is in
some transient cases, if the mesh is fine enough, not enough. This leads to an erroneous solution,
which ultimately can diverge. For the running of case 1, this default setting is good enough. For
case 2, lowering the convergence criteria to 10^-4 was a necessity.
17
One way of judging if a time step is small enough, is to have it at least one order of magnitude
smaller than the smallest time-constant being modeled. Another way, which were the choice the
author fell on, were to have 15-20 iterations for each time-step. This choice were supported by
co-supervisor Tufan Arslan and the User-guide for Ansys Fluent.
Choosing the number of time-steps is a tradeoff between time and having a sufficient amount of
data to describe the case accurately. For case one, Stokes second problem, the amount of full
cycles were chosen to be three, in order to avoid the startup-effects. For case two, the optimum
choice would be to have more than 20 full oscillations. This turned out to be not feasible, with
the given computer power available. The author fell on two complete oscillations, which of
course isnt enough to conclude on anything. Nevertheless it gives a trend of what one could
expect of the added mass in the proximity of the seabed.
18
5. Post-processing
In order to efficiently process the information gained through simulations done in Ansys Fluent, a
program was written in MATLAB. ystein Johannesen, a fellow student, helped me write a program
which made the post-processing easy and fast for case one. The program reads in information
regarding velocities at various y-coordinates at a certain time-step. This was then organized into two
matrices. The results will be presented in a later chapter. The code will be given in the appendix.
Regarding the post-processing the information retrieved for case number two, this has been done
with basis of what is outlined in (Hyem Aronsen, 2007).
5.1 Extracting hydrodynamic force coefficients
In this section two different methods capable of extracting the hydrodynamic forces in phase with
acceleration and velocity are presented.
5.1.1 The power transfer method
This method considers the power which is transferred between the fluid and the cylinder, i.e. in
other words energy per unit time.
Power transfer, P(t), is given by
()
()
()
() ()
Formula 32
Where x(t) is the cylinder motion. () is hence the cylinders velocity, and the average power
transfer,
()
()
Formula 33
If we assume that a forced harmonic oscillation results in a harmonic hydrodynamic force, a second
expression for the average power transfer is given below,
Formula 34
Which can be further elaborated into the following,
[ () (
) (
()
]
Formula 35
19
This reveals that the first integral (
) (
.
This leads us ultimately to the expression for average power,
[()
()
Formula 36 Expression for
average power.
From Formula 33 and the force in phase with the velocity,
()
() ()
Formula 37 Forces in phase
with velocity.
Using the same idea, but without linking it to a physical quantity as power, we can deduce an
expression for the forces in phase with the acceleration
(),
()
()
()
Formula 38 Forces in phase
with acceleration.
5.1.2 The Least-square fit method
This method is also proposed by (Sarpkaya & Isaacson, 1981), but the one presented by (Hyem
Aronsen, 2007) is the one which will be outlined.
The method is based on Matlabs capability to solve matrices fast and efficient. The assumption is still
that the hydrodynamic force may be given as a linear combination of the acceleration and velocity
signal,
()
()
()
Formula 39 Forced oscillations
This can be reorganized into a matrix C consisting of
and
Formula 40 In matrix form
And
|
()
()
|
Formula 41 In matrix form
The coefficients are then estimated by the minimization of
20
Formula 42 Minimizing the
coefficients
The forced oscillations are given as a sinus-equation,
()
)
Formula 43 Harmonic
oscillation
Putting this into Formula 39 Forced oscillations, this leads to
() (
() (
)
Formula 44
The force components can then be identified as
()
Formula 45 Forces in phase
with velocity
()
Formula 46 Forces in phase
with acceleration
Now that the time-varying forces have been figured out, it is time to put them into coefficients.
The added mass coefficient is defined as
()
(
Formula 47 Added mass
coefficient
And the dynamic excitation coefficient is defined as
()
Formula 48 Excitation
coefficient
21
6. Case 1
As proposed by supervisor Hvard Holm, a case where the moving mesh-feature in Ansys Fluent
would be validated against a text-book problem was needed. Hvard Holm suggested solving the
Stokes second problem, which have an analytical solution.
Several text-books present this problem, like (Faltinsen & Timokha, Sloshing, 2009), but the one
chosen here is the (White, 2006).
An oscillating stream or wall is often referred to as Stokes second problem. If we look at an
oscillating wall with the velocity
( )
()
Formula 49 Velocity for an
oscillating plate
Where as usual is the circular frequency. The fluid in the far field is at rest,
( )
Formula 50 Velocity at far field
The solution for a laminar flow with moving boundaries, can be found if one makes a parallel-flow
assumption of
( )
Formula 51
This makes the momentum equation turn out like
)
Formula 52
The pressure gradient can only be a function of time for this flow, and hence can be absorbed into
the velocity by a change of variables. If we define the velocity like
Formula 53
Then
)
Formula 54
The steadily solution of Formula 54, must be of the form
( ) ()
Formula 55
Where i= Substituting Formula 55 into Formula 52, gives the ordinary differential equation
Formula 56 Ordinary
differential equation
Where the solution takes the form of
22
(
)
Formula 57
Now still holding on to the oscillating wall, the final result is
()
( )
Formula 58 The final solution
Where
(
)
Formula 59
6.1 Results and discussion
The dynamic mesh consists entirely of tris, see Figure 9, and has a total of approximately 29 000
elements. It is locked onto the oscillating plate. The domain is 10x3 meters. Smoothing and
remeshing were applied. Spring-constant factor equal to 0.3, and the remeshing criteria were
minimum length scale equal to the minimum length found in the mesh, and the maximum length
scale 50% larger than the shortest length scale. Maximum cell skewness were floored to the one
found in the mesh.
The Reynolds-number is kept low; below Re<100, in order to keep the flow laminar.
The first attempt wasnt all that, since startup effects are present, see Figure 8.
23
Figure 8 Stokes' second problem with startup effects
The second attempt, with some knowledge acquired from run number one, gave some hints of what
the next mesh should have.
The mesh was made in Gambit with scaling functions on the vertical sides, and on the oscillating disc.
A line were created in Ansys Fluent in the center of the mesh, at x=0, with height equal to 1. The
velocities were measured along this line. The boundary-conditions were all symmetry-plane, except
for the disk itself which was wall. The total domain is 10x3 meters, which is the standard unit for
Gambit.
Figure 9 Mesh for Stokes second problem
24
The amplitude of the oscillation of the plane was 0.1, and the period was 9 seconds. One can argue
that the plate is oscillating a bit too close to the wall, which would result in fluid streaming over the
disc, but if plotting the velocities measured numerically versus the analytical ones, it isnt that far off,
see Figure 10.
A suggestion for refinement would be to put more elements higher above the oscillating plate, as
the profiles are more off up there than on the plate itself, which has more elements on it. Another
suggestion for refinement would be to make the oscillating plate shorter, in order to avoid effects
from the symmetry-plane on the sides.
But being happy with the results, bigger challenges awaits.
Figure 10 Comparison between analytical and numerical solution for stokes second problem
25
7. Case 2
The second case is where all the features are brought together, like dynamic meshing with sliding
and layering, and eventually extracting the forces in phase with acceleration and velocity. The
objective was to see if the use of CFD with RANS-modeling were capable of calculating the
hydrodynamic forces for an oscillating disc near the seabed is possible.
7.1 2D-case, results and discussion
Starting off with a 2D-case, the sliding and layering functionality were to be tested. Thereafter the
functionality of the post-processing was tested. The mesh is shown in Figure 11. The total domain is
4x1 meters, whereas the disc is 0.4x0.02 meters.
Figure 11 Mesh for the 2D-case, where a disc is oscillating at a height 2*radius from the ground.
The boundary conditions are symmetry plane at the boundaries to the left, right and top, whereas
the oscillating disc and the bottom is wall. Total number of faces is 315620. The distribution of faces
close to the discs ends is shown in Figure 12.
26
Figure 12 Close-up for the 2D-case.
Co-supervisor Tufan Arslan gave me the advice to have at least 15-20 nodes at the discs edge. As
shown in Figure 12 this is the case here. The motion of the disc was controlled by a UDF user
defined function written in C, which is loaded and compiled into Fluent and finally builds an adequate
library. The code can be seen in the appendix.
The turbulence-model was chosen to be the k- with SST. The coefficients were set as default given
by Fluent. The fluid was set to be regular water, which resulted in a Reynolds-number equal to 95448
at KC=1.2. This is in the sub-critical region, as given in Figure 1.
The solver was chosen to be PISO, with second order accuracy for the spatial discretization, but only
first-order implicit transient formulation. This is due to the dynamic mesh which limits the user to
first-order implicit. The residuals-limit were chosen to be absolute and 10^-4 for the continuity,
velocities in x- and y-direction, and for k and omega.
To the authors big disappointment, any comparable data at the given ratio between the discs
diameter and thickness, and the equal Keulegan-Carpenter number, were hard to find.
Recommendations for further work are here obvious. A screenshot were taken at t=4.8125 seconds,
which equals 0.8125*T, where T is the period of one oscillation, see Figure 13.
27
Figure 13 Screenshot from visualization of oscillating disc at height 2*radius and KC=1.2 at t=4.8125
The added mass for the 2D-case is given as
()
( )
Formula 60
See Figure 14 for the results. As stated above, some simulations with a more standard geometry like
a square or a cylinder would be highly beneficial, in order to validate simulation themselves and the
post-processing.
28
Figure 14 Added mass in 2D for a oscillating disc at 2*radius depth with increasing KC
One simulation with 5 complete oscillations took 53 hours on a computer equipped with an Intel i7-
2600 processor running at 3.40 GHz and 8 GB of RAM.
Another picture of this 2D-case, although with a finer mesh with 452800 faces, is given in Figure 15. It
would of course be interesting to see some oscillations for this case with that mesh, but due to
convergence-trouble, the time step which would give small enough oscillating residuals to be
somewhat converging within a reasonable number of iteration, were 2.5*10^-4. One complete
oscillation would then take 58 hours, but the simulation was aborted, due to the need of computer-
time for other simulations.
29
Figure 15 Screenshot of the velocity-magnitude around the edge of an oscillating edge at height 1*radius and KC=1
7.2 3D-case, results and discussion
Seeing that the sliding and layering features worked out fine, it was time to go 3D.
The simulations have been done with the intention to reproduce the results found in the following
article (Wadhwa, Balaji, & Thiagarajan, 2010). This article investigates the variation of added mass
and damping in heave direction near the seabed as a function of Keulegan-Carpenter number.
The motion is carried out by a planar motion mechanism, as shown in Figure 16. The experiments
were carried out in the Keulegan-Carpenter range of 0.26-1.77. The distance to the seabed ranges
from 0.2a-2a, where a equals radius for the oscillating disc. The total physical domain for the
experiment is 1x1x (0.76-0.84), length, breadth and depth respectively. Why the depth is varying is
not described in the article.
30
Figure 16 Planar motion mechanism for the heave-motion, (Wadhwa, Balaji, & Thiagarajan, 2010)
Tufan Arslan helped me to set-up the basics of the mesh, but in the end some tweaking has been
done.
The first mesh has a total of some 1660000 cells; see Figure 17 and Figure 18. The total domain
equals 10*radius meter of the disc in radial direction, and height equals 0.8 meters. The boundary
conditions are symmetry plane for everything except the oscillating disc and the bottom, which is
wall. The extent of the domain is the minimum of what is acceptable. The dimension of the disc is as
following, radius equals 0.1 meters, and thickness equals 0.002. Any domain-study has not been
performed.
31
The first mesh can be seen from above, see Figure 17.
Figure 17 3D-case, some 1660000 cells, seen from above
The same mesh as in Figure 17, seen from the side, is given in Figure 18.
Figure 18 3D-case, some 1660000 cells, seen from the side.
32
A close up picture of the cell-distribution close to the disc is given in Figure 19. A cut-plane gives the
distribution of cells along this plane. The distribution is equal all the way round.
Figure 19 Close up for the mesh with 1660000 cells
On the other side on the fine-mesh scale, a mesh with 2450000 cells is shown below, see Figure 20.
One run with 2 complete oscillations, took 56 hours, where the time step were 10^-3.
Figure 20 3D-case, with some 2450000 cells, seen from above.
33
The same mesh, seen from the side, is given in Figure 21.
Figure 21 3D-case, with some 2450000 cells, seen from the side.
The mesh convergence-study were done with a Keulegan-Carpenter-number equal to 1, at a height
2*radius from the bottom. See Figure 22. As can be seen, the nodes are clustered towards were the
action is, in the area of the oscillating disc. One run on the finest mesh took close to 84.5 hours for 2
complete oscillations with time step equal to 10^-3.
Some nice screenshots can be seen in the appendix for the oscillating disc at height 1*radius from
the seabottom at KC=1, see Figur 34 to Figur 44.
34
Figure 22 Mesh convergence-study for an oscillating disc at height 2*radius from the bottom, at KC=1. The x-axis gives
number of elements in thousands.
The mesh-convergence study reveals that the finest mesh with 2450000 cells gives the solution
closest to the experimental data, see Figure 23 and Figure 24.
Figure 23 Added mass for a oscillating disc at KC=1 at depth 3*radius. (Tao & Dray, Hydrodynamic performance of solid
and porous heave plate, 2008)
The mesh-convergence study have been performed at KC=1, and with a frequency equal to 1. The
data represented by the blue line, see Figure 23, is pretty close to what is given by the finest mesh
35
with some 2450000 cells. What one must bear in mind is that data achieved in the Figure 23 have
been performed at a depth equal to 3*radius. This could give the difference in the added mass
coefficient.
Some experimental data have also been produced at a different height, namely 2*radius from the
bottom, see Figure 24.
Figure 24 Added mass for an oscillating disc at height 2*radius from the bottom, at KC=1. (Wadhwa, Balaji, &
Thiagarajan, 2010)
Unfortunately, simulations have been performed with the coarsest mesh. The reason for this is that
simulations with the finest mesh would spend too much time. Even though 84.5 hours sounds
reasonable for a CFD-simulation of a certain size, this amount of time is if everything runs smoothly.
That isnt always the case, which has been learned the hard way. Another recommendation for
further work is here given. Below is given the added-mass for an oscillating disc at height 2*radius
oscillating at different KC, see Figure 25.
36
Figure 25 Added mass for an oscillating disc for various KC at 2*radius
A substantial increase in the added mass can be seen in the range between KC equal 1 and 1.5 is
seen. If one compare it with Figure 24, over prediction is seen. Reason for this might be, except the
lower cell count, the y-star values. Y-star is defined below.
Formula 61
As can be seen in Figure 26, the mesh has quite high y-star values at the bottom of the mesh, y-star
equal to 12, and at the disc the y-star is approximately 3.
According to the User-Guide for Fluent, and discussion with Tufan Arslan, the y-star values should be
in the region 30<y-star<500 or y-star<1 near boundaries. This is not the case for the mesh given in
Figure 26.
37
Figure 26 Y-star values for the mesh with 1660000 at height 1*radius from the bottom.
38
Added mass for the same oscillating disc at 1*radius depth is given for various KC in Figure 27.
Figure 27 3D-added mass for an oscillating disc at height 1*radius for various KC.
The reason for why this graph only has two points and a line, instead of three points and a curve, is
because of trouble with convergence and lack of time to generate and run a proper simulation.
Another recommendation for further work is her given.
Comparing the data found in Figure 27 and Figure 28, shows a significant difference for the
calculated added mass. The reason for this might be as for height equal to 2*radius, lower quality of
mesh and not good enough y-star.
39
Figure 28 Added mass for various heights and various KC, (Wadhwa, Balaji, & Thiagarajan, 2010).
40
8. Discussion and recommendations for further work.
Working with CFD isnt always smooth sailing. Before you reach a certain level and gain a certain
amount of experience, various errors and misconceptions may occur from time to time.
Including a comic strip in a Master thesis may seem inappropriate, Figure 29, but I think it sums up
working with CFD; it isnt just something you do right away. The user must be systematic and do
careful planning to succeed. If one doesnt pay attention to the details, you would end up with an
answer which isnt much of an answer.
Figure 29 From The far side by Gary Larsson, (Institute)
However, this thesis shows that is feasible to predict the added mass with CFD and Reynolds-
Averaged Navier-Stokes turbulence modeling. But in order to produce satisfactory results there are
some important things to pay special attention to.
First of all sufficient quality of the mesh is priority number one.
Number two is a time step that is small enough to ensure convergence, but large enough to ensure a
simulation as quick as possible.
Number three is to have more than 2 oscillations per simulation. This is clearly not enough to give a
representation of the forces in play. At least 20 simulations would yield a somewhat representative
dataset for the real world behavior.
Number four is to have a small enough Keulegan-Carpenter number, say lower than 1. Any
simulations with a higher Keulegan-Carpenter number than 1, one should investigate the possibility
41
to use LES-modeling.
A
Appendix A
Following figures gives the velocity, hydrodynamic forces and acceleration as a function of time.
B
Figur 30 Forces and velocity against time, Forces and acceleration against time, Force against time, 3D-case height
1*radius from bottom at KC=1. 1660k mesh.
C
Figur 31 Forces and velocity against time, force and acceleration against time, force against time, for the 3D-case with
2450k mesh, height 2*radius and KC=1.
D
Figur 32 Forces and velocity against time, force and acceleration against time, force against time, for the 3D-case with
1660k cells, height 1*radius and KC=1.2.
E
Figur 33 Forces and velocity against time, force and acceleration against time, force against time, for the 3D-case with
1660k elements, height 2*radius and KC=1.5.
F
Appendix B
Screenshots for the oscillating disc at height 1*radius at KC=1. The mesh is the one with 1660k cells.
Figur 34 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.01T*osc
Figur 35 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.1T*osc
G
Figur 36 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.2*Tosc
Figur 37 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.3*Tosc
H
Figur 38 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.4*Tosc
Figur 39 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.5*Tosc
I
Figur 40 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.6*Tosc
Figur 41 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.7*Tosc
J
Figur 42 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.8*Tosc
Figur 43 Oscillating disc at 1*radius, KC=1, 1660k cells, 0.9*Tosc
K
Figur 44 Oscillating disc at 1*radius, KC=1, 1660k cells, 1.0*Tosc