Sei sulla pagina 1di 123

Research on procedures to estimate 3D response of

Vortex Induced Motion (VIM) of multi-column floaters


based on 2D CFD calculations
Master Thesis
Student

: Yang Liu

Supervisor MARIN

: A. Koop & A. Maximiano

Supervisor TU Delft

: R.H.M. Huijsmans

Program

: Additional program of JIP VIM in MARIN

Date

: 23-08-2016

Version

: 4.2

Abstract
It is well known to offshore industry that current can trigger significant in-line and cross-line
motions on floating offshore structures due to resonance. For a small dimension structure, like
risers, this will lead to Vortex Induced Vibration (VIV). A lot of research has been carried out on
this topic. However, for a large dimension structure, like semi-submersible floater, the frequency of
motion is relatively low, resulting in the so-called Vortex Induced Motion (VIM). In the last decade,
the concept of VIM has gained increasing attention. Nowadays it is one of the most important design
topic for deep draft semi-submersibles, especially within the scope of riser fatigue analysis.
Currently, model test is the preferred way to predict the VIM behaviour of a known offshore
structure. However, due to time and cost restrictions, it is not feasible to perform VIM model tests
in the early design stage. Therefore, a need for a fast and reliable method to predict VIM arises.
A good candidate is performing CFD calculations. With the development of numerical method and
hardware, at the moment CFD can provide reasonable VIM predictions. Compared with model
test, CFD can save a lot of time and is relatively cheap to perform sequences of tests with different
parameters which will greatly benefit the early design stage.
Several 3D 3DOF CFD VIM calculations for offshore structures have been performed in MARIN
and the result is quite comparable to model test result. However, about 2 weeks per calculation are
required. At an early design stage, a faster prediction of the VIM behaviour of an offshore structure
with desired parameters through CFD calculation is desirable. This target could be achieved through
2D CFD calculation. Compared with 3D CFD calculation, CFD VIM in 2D can be executed and
accomplished within 1 or 2 days. The columns and pontoon of the floater will be treated separately
in CFD. CFD VIM calculations in 2D could also provide reasonable and comparable results. It is
feasible to study the VIM behaviour of multi-column floaters through 2D CFD calculations.
As a consequence, in this project, the research on VIM behaviour of multi-column floaters will
be performed in ReFRESCO, a commercial CFD code of MARIN. Different structural parameters
(mass ratio, external damping levels, geometry of the columns) and environmental conditions (the
velocity of current) will be investigated in this project, in terms of their influence on VIM behavior.
This can give some insight about VIM behaviour of multi-column floaters and how to reduce it.
Besides, the numerical set-up will also be surveyed, like time step and grid size sensitivities, to
ensure the accuracy of the calculation. Finally, results from 2D CFD calculations will be compared
with those from 3D CFD calculations. The feasibility of the estimation of a semi-submersibles VIM
response in 3D from 2D CFD calculations would be checked.

Acknowledgement
Id like to sincerely express my gratitude to my supervisors in MARIN, Antonio Maximiano and
Arjen Koop! They not only offer me the great opportunity to work on this fantastic topic and work
with so many talented professionals in the world class maritime research institute, but also provide
me a lot of guidance on CFD application and research methods. They are always prepared to offer
me help and enlighten me during every meeting. It is their help and inspiration that makes me step
onto a wonderland of hydrodynamics and grow up very fast, as an engineer and researcher. Their
opinion and advices towards my career planning are also very valuable and beneficial.
Moreover, I want to offer my heartfelt thanks to my supervisor in TU Delft, Professor Rene
Huijsmans. His recommendation leads me to this interesting field and his suggestions benefits me
a lot during the project. He always stimulates many discussions during each progress meeting and
illuminates me with his wisdom. Because of him each of my progress meeting becomes meaningful
and productive.
Besides, I am very grateful for the inspiration from Joost Sterenborg and Jaap de Wilde. Their
experience on hydrodynamics and vortex induced motion plays an important role for my comprehension of some specific concepts. Their edification, as well as many interesting conversations Ive
had with them, all make up part of my wonderful memory in MARIN.
As the strongest back-up, my family always support me from beginning to end, especially my
mother and father. Though it is 8000 km away and 6 hours difference, still I can feel their company.
It is of great happiness to know they are always with me. I must thank them for all the efforts and
sacrifice theyve made for me.
My thesis committee members, Ido Akkerman and Mathieu Pourquie have made plenty of efforts
on the check and supervision of my work. Many thanks to their help and suggestions.
Finally, it is thankful to have a epic time with all the students in MARIN, from both academic
and social point of views. Ten months internship in Wageningen is precious and meaningful to me.
It is one of the memorable page in my life.

Contents
1

Introduction
1.1 Motivation . . . . . . . .
1.2 Scope of work . . . . . .
1.3 Approaches & Objectives
1.4 Structure of the thesis . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

1
1
1
2
2

Physical concepts and definitions


2.1 Equation of motion (EOM) and related parameters
2.2 Commonly used dimensionless parameters . . . . .
2.3 Definition of work-done . . . . . . . . . . . . . .
2.4 Equivalent damping . . . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

4
4
5
7
7

Theoretical background and literature review


3.1 VIV and VIM . . . . . . . . . . . . . . . . . . . .
3.1.1 Vortex formation in the wake . . . . . . . .
3.1.2 In-line motion and cross-line motion . . . .
3.1.3 Mass and damping . . . . . . . . . . . . .
3.1.4 Geometry and layout of the columns . . . .
3.2 Importance to offshore industry . . . . . . . . . . .
3.2.1 Fatigue . . . . . . . . . . . . . . . . . . .
3.2.2 Suppression of vortex-induced oscillations
3.3 Computational fluid dynamics(CFD) on VIM . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

9
9
10
12
13
15
21
21
21
24

Numerical Settings
4.1 Geometry of the floater . . . . . . . . .
4.2 Computational domain and mesh . . . .
4.2.1 Grid refinement . . . . . . . . .
4.2.2 Viscous layers . . . . . . . . .
4.2.3 Boundary conditions . . . . . .
4.2.4 Summary of the grid properties
4.3 Solution verification . . . . . . . . . . .
4.3.1 Iterative error . . . . . . . . . .
4.3.2 Discretization error . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

25
25
25
27
29
30
30
31
32
34

CFD tests and results


5.1 Decay tests . . . . . . . . .
5.1.1 Set-up decay tests . .
5.1.2 Results and analysis
5.2 CFD VIM tests set-up . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

44
44
44
45
50

Estimation of 3D VIM response from 2D CFD calculations


6.1 Tests with varying reduced velocity and external damping in 2D . . .
6.1.1 Test configuration . . . . . . . . . . . . . . . . . . . . . . . .
6.1.2 Flow variables . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 Discrepancy between 2D and 3D CFD VIM calculations . . . . . . .
6.2.1 Existence of bell shape curve . . . . . . . . . . . . . . . . . .
6.2.2 Comparison of flow variables . . . . . . . . . . . . . . . . .
6.3 Feasibility of 3D VIM response estimation from 2D CFD calculations

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

53
53
53
53
59
61
61
62

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

Research on VIM behaviour of multi-column floaters in 2D


7.1 Behaviour of each column . . . . . . . . . . . . . . . . . . .
7.2 Mass ratio and excessive external damping . . . . . . . . . . .
7.2.1 Mass ratio . . . . . . . . . . . . . . . . . . . . . . . .
7.2.2 Excessive external damping . . . . . . . . . . . . . .
7.3 Column geometry . . . . . . . . . . . . . . . . . . . . . . . .
7.3.1 Tests configuration . . . . . . . . . . . . . . . . . . .
7.3.2 VIM performance of columns with different geometry

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

63
63
71
71
71
72
73
76

Conclusion and recommendations

79

Nomenclature

81

10 List of abbreviations
11 Appendix
11.1 Numerical methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.1.1 Conservation principles . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.1.2 Unsteady Reynolds Averaged Navier-Stokes Equations(URANS) . . . . .
11.1.3 Turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.1.4 Numerical formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2 Verification of CFD tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.1 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.2 Numerical errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.3 Code verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.4 Solution verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.5 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.3 Script for the modelling of a four columns floater in Hexpress . . . . . . . . . . .
11.4 Control file for 2D CFD VIM simulation of a four columns floater in ReFRESCO
(Ur=6) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

.
.
.
.
.
.
.
.
.
.
.
.

88
88
88
90
92
93
97
97
97
98
98
102
103

. 109

Introduction

In this chapter a general introduction about vortex-induced motion and its influence on offshore
structures is presented. It also elaborates the reason for choosing CFD as the tool to study the
interaction between vortex shedding and multi-column floaters.

1.1

Motivation

Stimulated by the increasing demand for energy and resources, nowadays more and more exploration and development activities take place further offshore. They have pushed the limits of
offshore activities considerably. As a result, harsh environment conditions and severe engineering
problems are faced by the industry, with the growth of offshore market. The motion of a body caused
by current is one of them. It is well known to offshore industry that current can trigger significant inline and cross-line motions on floating offshore structures due to resonance. For a small dimension
structure, like risers, this will lead to the Vortex Induced Vibration (VIV). A lot of research has been
carried out on this topic. However, for a large dimension structure, like semi-submersible floater, the
frequency of motion is relatively low. This will result in the so-called Vortex Induced Motion (VIM).
In the last decade, the concept of VIM had gained increasing attention. Nowadays it is one of the
most important design topic for deep draft semi-submersibles, especially within the scope of riser
fatigue analysis.
Currently, model test is the preferred way to predict the VIM behaviour of a known offshore
structure. However, due to time and cost restrictions, it is not feasible to perform VIM model tests
in the early design stage. Therefore, a need for a fast and reliable methods to predict VIM arises.
A good candidate is performing CFD calculations. With the development of numerical method and
hardware, at the moment CFD can provide reliable VIM predictions. It is quite favourable for tests
with different set-up. Compared with model testing, CFD can save a lot of time and is relatively
cheap to perform sequences of tests with different parameters which will greatly benefit the early
design process.
Several 3D 3DOF CFD VIM calculations for offshore structures have been performed in MARIN
and the result is quite comparable to model test result. However, about 2 weeks per calculation are
required. At an early design stage, a faster prediction of the VIM behaviour of an offshore structure
with desired parameters through CFD calculation is desirable. This target could be achieved through
2D CFD calculation. Compared with 3D CFD calculation, CFD VIM in 2D can be executed and
accomplished within 1 or 2 days. The columns and pontoons of the floater will be treated separately
in CFD. CFD VIM calculations in 2D could also provide reasonable and comparable results. As a
consequence, it is feasible to study the VIM behaviour of multi-column floaters through 2D CFD
calculations.

1.2

Scope of work

In this study, research on the VIM behaviour of multi-column floaters will be performed in
ReFRESCO, a CFD code of MARIN. The topology of semi-submersible in 3D would be modelled in
2D. Different structural parameters (mass ratio, external damping levels, geometry of the columns)
and environmental conditions (the velocity of current) will be investigated in this study, in terms
of their influence on the VIM behaviour. This will not only benefit the comprehension of VIM
mechanism, but also sketches the characteristic VIM behaviour of semi-submersible specifically in
2D CFD calculations. Then a comparison with the same floaters VIM behaviour in 3D would be
easy and explicit.
Besides, the numerical set-up will be inspected, like time step and grid sensitivities, to establish
the credibility for CFD calculations. Finally, results from 2D CFD calculations will be compared
with those from 3D CFD calculations. The feasibility of the estimation of a semi-submersibles VIM
response in 3D from 2D CFD calculations would be checked.

1 of 118

1.3

Approaches & Objectives

Taking into account the scope of work and the adopted CFD tools, several objectives and corresponding approaches are highlighted in this study:
Inspect grid and time step sensitivity for the CFD VIM calculations to establish the credibility
and choose proper numerical configurations for the following tests within ReFRESCO ;
Set-up and perform unsteady turbulent flow calculations for decay and VIM tests through the
URANS CFD solver ReFRESCO coupled with the equation of motion (EoM);
Investigate the influence from different column geometries and flow field parameters on the
VIM behaviour of the floaters;
Look into the physical principles behind VIM behaviour of the multi-column floaters;
Compare 3D VIM response of the multi-column floaters from 2D CFD VIM calculations.Justify
the feasibility of 3D VIM response estimation from 2D CFD calculations.

1.4

Structure of the thesis

The main concern of this study is to investigate the VIM behaviour of multi-column floaters
based on 2D CFD calculations. Besides, tests on parameters that have impact on VIM behaviour of
the floaters are also performed so as to provide an insight about the physical mechanism of VIM and
elucidate discrepancy between 2D and 3D cases. The structure of this thesis serves specifically to
the elaboration of these topics.
Chapter 1 is a general introduction of this study. It elaborates the initial motivation of this
research topic and the objectives to achieve. It gives an outline of the entire work.
Chapter 2 provides the physical concepts and definitions related with VIM. In this section the
nomenclature, basic concepts and equations involved for the elaboration of this study on VIM are
provided in detail.
Chapter 3 acts as the instruction for background knowledge of VIM and CFD. It tells the story
about our current understanding towards VIM and some recent progress on related topics. Moreover,
it also contains the major information from literature review.
Chapter 4 introduces the numerical settings being adopted in this study. The grid used in VIM
tests is illustrated in this section. The reason for the choice of certain grid properties are also explained. Besides, the theory of CFD tools verification is put into use here and the results are shown
and analysed at the end of this chapter.
Chapter 5 is the practice of 2D CFD decay tests and VIM tests. Results from CFD calculations
about decay tests and the set-up of VIM tests are presented.
Chapter 6 throw lights on the feasibility of estimation of 3D VIM response from 2D CFD calculations. It reveals the discrepancy between 2D and 3D CFD calculations. An analysis of the
ambivalence is made and conclusions are drawn, based on the results of tests in 2D CFD with varying reduced velocity and external damping.
Chapter 7 describes the configurations and results for 2D CFD VIM tests with different parameters. In this section different parameters and their influence on VIM behaviour are investigated,
which are, for instance, mass ratio, external damping coefficient, column geometry.
Chapter 8 serves as the final conclusion of the entire study. Some recommendations are made
depending on the practice and results in this work.
Apart from these chapters, a list of nomenclature and abbreviations are attached in the end. All
the essential codes related with initialization of CFD calculations and post-processing of flow field
data are provided in appendix. The function of each script is announced at the beginning those codes.
For further questions regarding numerical methods used in this study and the theory behind
solution verification, two additional subsection in appendix could provide some information. Section
11.1 elaborates the numerical methods used in this study. These include fundamental conservation
principles of fluid flow, mathematical model, turbulence model, as well as temporal and spacial

2 of 118

discretization methods. Section 11.2 describes the theory followed by verification of CFD tools.
In this section different types of errors are identified and methods developed for the determination
of these errors are provided respectively. The verification procedure sets the credibility for CFD
calculations.

3 of 118

Physical concepts and definitions

In this section, the fundamental parameters, physical concepts and definitions related with VIM
are presented and explained.

2.1

Equation of motion (EOM) and related parameters

When studying the hydrodynamic behaviour of a structure, ones always have to deal with the
equation of motion. For a CFD VIM calculation, as it is an unsteady turbulent flow coupled with
EoM of the structure, it is important to have an idea about each term in the EoM. The function of
EoM is shown below:
Mx
+ B x + Kx = Fhydro

(2.1)

with x the displacement of structure, M the mass of structure, B the damping coefficient of structure,
K the stiffness coefficient of the system, Fhydro the hydrodynamic force on structure.
For the hydrodynamic force on the right side of the EoM, it has several components, which can
be expressed as:
Fhydro = Fadd + Fdamping + Fother
Fadd = madd x

(2.2)

Fdamping = bhydro x
where madd is the hydrodynamic added mass, and bhydro the hydrodynamic damping coefficient.
In the equation 2.2, Fadd stands for the force caused by hydrodynamic added mass, Fdamping
represents the force caused by hydrodynamic damping and Fother is the resultant force from other
sources, like wind, current and wave.
Here are some elaboration about hydrodynamic added mass and hydrodynamic damping coefficient.
Hydrodynamic added mass When a body moves through a fluid, it will always accelerate the
fluid around it. Then there will be force exerted by the fluid on the body, as if the fluid is attached to
the body and acts like additional mass. Hence such extra fluid is called hydrodynamic added mass to
the body. It plays a very important role during the calculation of the natural period of the structure.
r
M + madd
(2.3)
Tn = 2
K
Considering 6 degrees of freedom (DOF), the added mass force in one direction is affected by
the motion in all directions. Hence in a 3D space with 6 DOF the added mass becomes a 6x6 matrix
for a body. As in this study the CFD simulation is performed with only 3 DOF, so the added mass
matrix is simplified to:

m11 m12 m13


madd = m21 m22 m23
(2.4)
m31 m32 m33
The added mass matrix can be simplified further if the body is symmetric.
Hydrodynamic damping coefficient According to the physical definition, damping is a process
for the dissipation of energy of a moving object. The idea behind it is a complicated phenomena
due to the interaction between structure and fluid. It is a significant parameter that takes a part in the
dynamic behaviour of the structure.
Moreover, one that deals with dynamic behaviour analysis will always be interested in critical damping, at which the decay response will no longer oscillate but asymptotically return to the
equilibrium position. The critical damping can be calculated through mass and stiffness.

4 of 118

p
Bcrit = 2 K(M + madd )

(2.5)

Then the damping ratio is defined as:


=

B
Bcrit

(2.6)

External damping coefficient Apart from hydrodynamic damping, in this study there would be
another form of damping added to the system. In order to compensate the damping from the presence
of pontoon, mooring lines and riser, there would be extra damping applied directly to the structure.
It acts the same as normal damping and thus will also dissipate energy during VIM. It is named as
external damping only for its source. Different damping levels will be tested in a combination of
varying reduced velocities.

2.2

Commonly used dimensionless parameters

Some commonly used parameters in the literature to describe VIM are presented here. This also
allows the comparison of results between different tests.
Reynolds number Re The Reynolds number is a dimensionless quantity used to identify a relation
between inertia force and friction fluid force (viscous force) of a certain flow problem.
Uref D
(2.7)

with Uref the current velocity in this case, the kinematic viscosity of the fluid.
It is widely used as a non-dimensional parameter which illustrates the property of a certain flow
field around a body. In this study it helps to scale the dimensional current velocity.
Re =

Froude number F r Froude number is a dimensionless quantity that defined as a ratio between
the inertia and gravitational wave velocity.
Uref
Fr = p
gLref

(2.8)

where g is the gravitational acceleration.


Whats more, it is widely used as a scaling law for model tests, preserving the ratio between
inertia and gravitational forces of a full scale scenario. This is shown as follows:
Uf ull
Umodel
Fr = p
=
gDmodel
gDf ull

(2.9)

with Uf ull the full scale velocity, Umodel the velocity of the model test, Df ull the full scale length
and Dmodel the small scale length of the model.
Hence the scaling factor between the model scale velocity and the full scale velocity is:

(2.10)
Uf ull = Umodel
where is the scaling factor.
Mass ratio m The mass ratio is the ratio between mass of structure and its fluid displacement. It
can be expressed as:
M
(2.11)

This dimensionless number indicates the buoyancy status of the structure. If m < 1, it means
the structure is lighter than the buoyancy force, which implies the presence of a mooring system
in stretching tension. If m > 1, it means the structure is heavier than the buoyancy force, which
m =

5 of 118

manifests some ancillary buoyancy systems are lifting the structure. The case m = 1 indicates a
neutral buoyant status.
Reduced velocity UR The reduced velocity is a ratio between the natural period of the structure
and the flow reference time. It can be expressed as follows:
UR =

Uref TN
TN
=
Tref
D

(2.12)

where TN is the natural period of the structure and D is the effective diameter of the column.
This dimensionless parameter is very useful to define the range of current velocity that can
trigger significant VIM of the floater. It can be regarded as a characteristic parameter to relate
current velocity with VIM in terms of a certain type of floater.
Dimensionless motion amplitude A/D This is the ratio between oscillation amplitude and effective diameter of the column, which reveals the intensity of oscillation. Further more there are
two specific form of dimensionless amplitude which are defined to identify the characteristic motion
amplitude over time. One is nominal dimensionless amplitude, which is always defined as the root
mean square of dimensionless amplitudes:
s
PN
2
i=1 (A/D)i
(2.13)
(A/D)nom =
N
The other one is the maximum dimensionless amplitude, which is the largest value over time:
(A/D)max = max((A/D)i )

(2.14)

They are commonly used as target flow variables to judge the VIV and VIM performance of a
certain body.
Non-dimensional reference time Tref It is applied as a measurement of time, which always plays
a role during the selection for the time step in CFD calculation.
Tref =

D
Uref

(2.15)

The non-dimensional ratio between reference time and time step in CFD calculations can be
taken as a reference to ensure that the temporal size in the CFD calculation is fine enough to capture
all the details of the flow and motions. Empirically, this value should be larger than 40, at least.
Force coefficient and moment coefficient CFi , CMi
are monitored by dimensionless coefficients:

The force and moment acting on the columns

CFi =

Fi
1
2
2 Uref LD

CMi =

Mi
=
1
2
U
ref LDd
2

Fi
1
2
2 Uref S
Fi
1
2
U
ref Sd
2

(2.16)

with Fi and Mi the force and moment on structure in i direction, D the projected diameter of the
column, L the length of the object, S is the projected area of the body with respect to the incoming
direction of current, d is the moment arm.
Strouhal number St When a flow passes a bluff body, an oscillating flow pattern, named vortex
shedding, can take place within a certain Reynold number. In order to quantify the effects of flow
patterns in the wake, the frequency of vortex shedding , as well as the oscillating forces should be
known.
When fully developed, the vortex shedding phenomena can be regarded as a periodic motion.

6 of 118

Here we define the velocity of vortex as Uv and the distance between two in line shedding vortex as
X. Hence the frequency of vortex shedding can be expressed as:
Uv
(2.17)
X
It is obvious that the velocity of vortex has something to do with the velocity of current, such like
Uv = aUref . Similarly, the two in line vortex distance, X, can be related with the diameter of the
column, such as X = bD. Here a and b are all constants. Based on this analysis, Strouhal number
is defined as:
fs =

St =

D
b
= fs
a
Uref

(2.18)

Strouhal number is one of the most important non-dimensional parameters in the analysis of VIM
and VIV. It reveals the relation between structure, environment and vortex, in a dimensionless and
comparable way. In lock-in range, the vortex shedding frequency is approximately the same with the
natural frequency of the structure. Hence the relation between Strouhal number and reduced velocity
in lock-in range is:
St =

2.3

D
D
D
1
b
= fs
=
' fN
=
a
Uref
Uref
Uref T
UR

(2.19)

Definition of work-done

Work-done, in a nutshell, is the performance of force in energys point of view. This concept
comes from physics, in which it describes the energy transfer through force. When a force acts to
cause an object to be displaced, the work-done by force is the product of force, displacement and the
angle between force and displacement. It could be expressed through the following equation:

W =F S

(2.20)

with F the acting force and S the displacement.


In this study, since it is important to know the performance of each force component with respect
to cross-flow and in-flow direction, the equation above should be put in the other way around:


W = F S = (Flif t + Fdrag ) (Scross + Sin )

= Flif t Scross + Flif t Sin + Fdrag Scross + Fdrag Sin

= Flif t Scross + Fdrag Sin

(2.21)

= Wlif t + Wdrag
with Flif t the lift force, Fdrag the drag force, Scross the displacement in cross-flow direction and
Sin the displacement in in-flow direction.
With the help of orthogonality, the entire work-done could be divided into two parts, the workdone by lift force and the work-done by drag force. This is favourable for the behaviour analysis of
each column and the calculation of equivalent damping.

2.4

Equivalent damping

The definition of damping is briefly explained in section 2.1. The damping of a system is solvable
only when the system is linear. However, in most engineering practices the systems are non-linear.
Thus there is a need for a characteristic damping coefficient which could represent the damping of a
non-linear system. This is exactly the incentives for equivalent damping.

7 of 118

In order to calculate the equivalent damping, ones could assume a simple spring-mass-linear
damping system. In this case, the amount of damping actively added to the system is determined by
the energy taken from the semi-submersible per oscillation cycle. In terms of the definition of both
damping and work-done, the change of work-done per cycle and its relation with damping could be
represented as:
ti0 +t

Wdamping =

+t
0
X tiX

Fdamping x =
(B x)
x
ti0

(2.22)

ti0

with t the start time of a cycle, subscript i refers to the ith cycle of the test run, t the time step,
Fdamping the damping force, B the equivalent damping coefficient, x the motion and x the velocity.
Supposing the linear damping of the system is identical over the entire time domain, then the
equivalent damping could be calculated through the equation below:
PT
Wdamping
t Wdamping
=
PT
PT

t x 2 t
t x x

PT
B=

8 of 118

(2.23)

Theoretical background and literature review

In this section the theoretical background knowledge and the current research state of VIM are
presented. In particular, it will mainly focus on the genesis of VIM phenomenon, influential factors
to VIM behaviour, the importance to offshore industry and the use of CFD to predict VIM.

3.1

VIV and VIM

A fluid flow pass bluff bodies and the formation of wake behind them is a long established
research field which has been explored extensively over the previous century. A fluid, like water
or air, flows past a bluff body will separate around it when Re > 40. At a certain velocity, this
will result in an oscillating flow pattern behind the object, where alternating low-pressure vortices
occur in the wake. This flow pattern is known as a Von Karman vortex streets. The process of
the shedding of oscillating vortices is referred as vortex shedding. This is illustrated in figure 3.1.
This phenomenon plays very important roles in a wide range of engineering fields, like aerospace
engineering, civil engineering, geological engineering, etc. Offshore engineering is not an exception.
A lot of effort has been made, to reveal the principle behind such phenomenon and provide guidance
to the real design. The review summarized by King [30] has shown some early attempts to study
vortex shedding and its application. Since the discovery of vortex shedding, significant progress has
been made. A sequence of research achievements can be referred through the reviews [37, 59]. Most
of the experiments gathered in these reviews were carried out on cylinders, as vortex shedding effect
can be observed and studied explicitly in this way.
When the cylinder subject to flow is not captive, due to the alternating force produced by shedding vortex as well as the pressure exerted by the fluid flow, it will start to move. The immersed body
will experience a fluctuating lift force due to the asymmetric formation of vortex in the cross-flow
direction. As the force from shedding vortex is periodical, the motion triggered by it becomes periodical. When the frequency of shedding vortex coincides with the natural frequency of immersed
body, resonance takes place and the amplitude of cross-line motion rises considerably. This is the
genesis of vortex-induced vibration. The range of the vortex shedding frequency that triggers resonance is defined as the lock-in range. This concept is elaborated explicitly in Blevins work [4].
When the dimension of structure is large, the oscillating frequency diminishes. Hence the vibration
transits to periodical motion. Such motion is widely acknowledged by the offshore industry and
known as vortex-induced motion.

Figure 3.1: Von Karman vortex streets[54](Re=100)


According to Fujarra et al. [16], VIM and VIV are the same phenomenon. Both of them exhibit
a resonance behaviour due to fluid-structure interaction and result from the balance between energy
provided by fluid flow and energy dissipated by damping acting on the system. Nevertheless, VIM is

9 of 118

a particular case of VIV. The major difference lies in the modes. For VIV, since it generally happens
to flexible structures, it has infinite number of modes. Thus a study towards VIV should focus on
several lower order modes. However, VIM is quite different. It always takes place on a rigid body.
Thus it has only one mode. Normally the motion frequency of VIV is always lower than that of VIM.
In order to gain a comprehensive understanding of VIM, it is also valuable to look at the research on
VIV. In this subsection some recent progress and results on the phenomenology of VIM and VIV, as
well as contributing factors to such phenomenon, are presented briefly.
3.1.1

Vortex formation in the wake

Since the discovery of vortex shedding behind a bluff body, the formation of vortex in the wake
keeps drawing researchers and engineers attention. A systematic definition of the whole set of different regimes for vortex wake modes is given by Williamson and Roshko [61], where a descriptive
nomenclature for each mode was introduced. In their definition, each periodic vortex wake pattern
comprises single vortices (S) and vortex pairs (P), which yields patterns like 2S, 2P and P+S modes.
The visualization of these modes is shown in figure 3.2:

Figure 3.2: Sketches of the vortex shedding patterns and their modes definition [61].(P means a
vortex pair and S means a single vortex).
The significance of these modes from controlled vibration is that they provide a map of regimes
within which people observe certain branches of free vibration. Brika and Laneville were the first
to show the response branch corresponding with the 2S and 2P modes from free vibration [5]. Their

10 of 118

smoke visualization is shown in figure 3.3. According to their conclusion, 2S mode corresponds
with the initial branch of response and 2P mode with the lower response. Unlike 2P and 2S modes,
P+S mode is not able to excite free vibration.

Figure 3.3: Smoke visualizations of the 2S modes and 2P modes [5]


Apart from the definition, Williamson and Roshko also made a map of regimes for vortex wake
modes. Here part of the map related with our study is shown in figure 3.4. It principally shows the
2S, 2P and P+S modes relevant to synchronization regime .
Most of the results above were obtained by the tests with a small scale cylinder. As the result is
provided in a dimensionless manner, it is also applicable to the VIM study on cylindrical structures
with large dimensions.
Instead of single column case, our study mainly focus on multi-column floaters. Due to the layout of multi-column floaters, there will be interference between the wake fields of each column. As
a result, the vortex shedding pattern is not as pure as the observation on single column cases. This
topic will be extended further in subsection 3.1.4.)

Figure 3.4: The map of regimes for vortex wake modes relevant to the fundamental synchronization
regime [61]

11 of 118

Empirically, the Reynolds number of a flow field associates with the shedding modes behind a
body. For the high Reynolds number cases, variants of the 2P wake modes are observed in the upper
and lower branches. Meanwhile ,for the low Reynolds number cases, 2S wake modes are formed
[32].
3.1.2

In-line motion and cross-line motion

A fluid flow pass bluff body can trigger both in-line and cross-line motion. The main contributor
of the in-line motion is the drag force from the fluid flow (in this case, the current), meanwhile for
the cross-line motion is the lift force from asymmetric vortex shedding. We are more interested in
the cross-flow motion as when the vortex shedding frequency sychronizes with the natural frequency
of the body, it will result in an oscillation with significantly large amplitude. However, this does not
mean the in-line motion is of no importance. Indeed, the study on coupling of in-line motion and
cross-line motion shows that in-line motion plays a role for large transverse oscillation amplitude.
An example can be found from a survey completed by Fujarra et al. [16]. The result is shown in
figure 3.5. Their result is based on the VIM experiments on a long cylinder. It can be observed that
VIM can induce response amplitude up to 1.5 times the cross section dimension of the cylinder, as
a result of a notable coupling with the in-line motion. Though the in-line amplitude is only 0.35
times the effective length, it contributes to large oscillation amplitudes in cross-flow direction. The
transverse oscillation amplitude will dramatically diminish to around 0.8 times the effective length
with the in-line amplitude drops to 0.1 times the effective length. Hence it is also important to pay
attention to the effect of in-line motion during the study on VIM.

Figure 3.5: Qualitative comparison between non-dimensional amplitudes as a function of the reduced velocity[16]
Besides, in figure 3.5 above, a typical bell curve for the lock-in range of a cylinder can also
be observed. Such plots are always present in VIM research, which highlights the relation between
reduced velocity and dimensionless cross-flow motion amplitude. It can be categorized into three
parts, known as different branches:
Initial branch (I), where cross-flow motion arise.
Upper branch (SU), where it lies the peak value of transverse motion.
Low branch (L), where the cross-line motion decays gradually with the increasing of reduced
velocity.

12 of 118

These branches can be referred in figure 3.5, as well.


Actually the role of Reynolds number is also very important during the study on VIV and VIM.
For instance, it has effect on turbulent vortex street, categorizing it into different regimes [64].
However, in terms of the study on a multi-column floaters, since it is also greatly dependant on
other parameters, like the headings of the floater, the geometry of the column, and the layout of the
structure, it is not very meaningful to only focus on the response amplitude of the structure against
reduced velocity. As a result, especially for the study on VIM behaviour of semi-submersibles, other
parameters are worthy of attentions. They will be discussed in the following subsections.
3.1.3

Mass and damping

Mass and damping are two significant parameters in determination of dynamic behaviours. So it
is with the VIV and VIM. The constituent of mass includes structural mass and added mass. Their
effect on dynamic behaviour of structures is always regarded as an entity. Hence the structural mass
and added mass could be treated together and represented by mass ratio. Many research concerning
these topics have already provided a better understanding of the influence from these parameters
over the last two decades. Some recent accomplishments are shown below respectively.
Mass ratio The importance of mass ratio is well defined in the work done by Jauvits and Williamson
[28, 27]. Their conclusions are based on experimental results of cylinders in a flume tank with respect
to different values of mass ratio. Two different dynamic behaviours of cylinders with two degrees
of freedom (2DOF) were acquired. For a cylinder with moderate to large mass ratio m > 6, its
response is similar to that of one degree of freedom (1DOF). While for a cylinder with moderate to
small mass ratio m < 6, both the in-line motion and cross-line oscillation can be observed. In this
case, a new stable branch of response is revealed, in which a triplet of vortices are formed in each
half cycle of oscillation. It is defined as a 2T mode of vortex shedding. This branch is named as
super upper branch (SU-branch), which is relatively a more stable and more periodic branch. It is
shown in figure 3.5. According to Williamson and Jauvits, m = 6 can be taken as a limit value for
identifying the SU-branch on the cross-flow response of a 2DOF cylinder.

Figure 3.6: A comparison between 2T mode vortex shedding and other modes[60]

13 of 118

A detailed explanation of 2T mode vortex shedding can be sought from Williamson and Jauvits
[60]. It is illustrated in figure 3.6 with a comparison to other modes.
In a 2D CFD study on VIM response of a semi-submersible, Rijken has examined the effect of
mass ratio smaller than 2 [41]. In his study he found mass ratio could affect the VIM behaviour of
the floater especially when it is exposed to high reduced velocity, around U r > 7. When the reduced
velocity is small, its influence is quite limited. In our study the mass ratio is also inspected. Details
about it could be found in section 8.
Critical mass It is reported from several investigations in the open literature that the regime of
reduced velocity, over which large amplitude oscillations occur, increases with the reduction of the
structural mass ratio [21, 59]. There are even results showing that the synchronization range becomes
infinitely wide so long as the mass ratio falls below a special critical value. Discovered by Govardhan
and Williamson [21], the upper end of the lock-in regime, for a free vibration cylinder with low mass
damping, can be distinguished by a lower amplitude branch. It has a particular constant vibration
frequency. Through a fitting based on a large set of data, the curve of lower branch frequency plotted
against mass ratio can be represented by the equation below:
r
m + 1

(3.1)
flower =
m 0.54
In this equation, obviously there is a singularity lies in m = 0.54, which is defined as critical
mass. It indicates that below this value the lower branch of response can never be reached for a
finite velocity. They also suggested a upper boundary for the reduced velocity below which the
synchronization exists. It can be expressed as:
r
m + 1
UR = 9.25
(3.2)
m 0.54
The conclusions above can be observed in figure 3.7. In this figure the critical mass and boundaries for the existence of synchronization is illustrated. However, there are some different values
of critical mass recognized by other investigations, showing a dependence on the geometry of the
bodies and Reynolds number [22, 15].

Figure 3.7: The definition of the critical mass[59]


Damping Damping plays a dominant role as the source of energy dissipation during a dynamic
motion practice. Thus it could contributes to the decreasing on the VIM response. In the model test
carried out by Goncalves et al.[20], the VIM response of a semi-submersible is reduced dramatically

14 of 118

with increasing external damping. Their result is shown in figure 3.8. In another VIM test from
Goncalves et al. [18] on monocolumn platforms, the reduction of VIM response due to the existence
of external damping is observed as well.

Figure 3.8: Nondimensional amplitudes for the motions in the transverse direction for 45-degree
incidence with varying external damping[20]
As a result, it is feasible to increase the damping of the entire system so as to reduce the VIM response. Therefore during the design of semi-submersibles, pontoon is always applied as an effective
solution to the VIM motion.
Fundamental research of the effect of damping on amplitude and frequency on VIV of circular
cylinders were performed by Scruton[48], Griffin et al.[25], Khalak et al.[29], and Govardhan et
al.[23]. All the works showed the importance of the mass-damping parameter, which is the mass
ratio multiplied by the structural damping level.
This parameter, in general the mass-damping parameter, appeals to researchers since the establishment of the concept mainly due to its extraordinary property. This parameter includes no
information about flow speed. It is quite remarkable that without knowing the flow speed the maximum response amplitude can be predicted though mass-damping parameters. Nevertheless there
is some evidence that there are circumstances in which mass-damping parameters fail, as reported
by Bernitsas in his VIV experiments on cylinders [3]. As a result, some attempts towards testing
limitations of the mass-damping parameter have been made, like the attribution from Vandiver [55].
But it still has some disadvantages, so far. For instance, it can not predict the amplitude when the
response frequency is unknown, which are the cases before experiments. Whats more, it may be
inaccurate to predict the response amplitude of a model with multi columns. Taking into account
the complexity of the VIM problems of floaters with multi columns, it is quite a challenging work to
come up with a general mathematical model to predict its behaviour.
3.1.4

Geometry and layout of the columns

As discussed above there is a considerable effort placed in VIV research . Most of it is based on
experiments with a circular cylinder. This is due to its simple geometry which allows to control the
number of variables in a test, for the sake of research. In terms of its use in offshore engineering,
the most typical one should be the design of a spar platform, or generally, a mono-column platform.
However, such thoughts are not applicable to the structure comprises multi-body, e.g. the semisubmersible and a set of risers, for example.
Unlike single circular cylinder, a model with multi columns has more issues related with its
deployment. They can always play important roles in the process of the flow excited motion. This

15 of 118

study will mainly focus on the VIM behaviour of a multi-column floaters. Hence factors that account
for the VIM behaviour of a multi-column floater will be discussed here.
Current heading For a square cylinder, the varying angles current heading can affect the behaviour
of flow induced motion. Due to the symmetry of a square cylinder, it is sufficient to test current
angles from 0 to 45 degree, in other words, from square to diamond configuration. According to
Nemes et al., this effect corresponds to the mechanisms of wake-structure interactions [36].
It is generally agreed that wake-structure interactions comprises a number of mechanisms: galloping, flutter and vortex-induced motions [35]. A tentative list of flow induced motion can be taken
as a reference [57]:
Vortex induced vibration
Galloping and flutter
Flow interference
Turbulence induced vibration (Buffeting)
Static divergence
Drag crisis
In this case it is expected to see both galloping and VIM. The galloping, is an aerodynamic
instability experienced by bodies with non-asymmetric cross-sections. Any asymmetric slender body
in a cross-flow is susceptible to galloping at certain angle. Compared with VIV, galloping is lower
in frequency and is not self limiting. When the flow velocity exceeds the critical velocity for the
triggering of galloping, it will occur regardless of frequency. In the experiments carried out by
Nemes and his colleagues, they observed that both galloping and VIM will take place when the
flow passes a square cylinder. With the square configuration, when the angle is 0 degree, galloping
will dominates the vibration, while with the diamond orientation VIM is dominant as the oscillation
mode. A sequences of tests have shown that by varying the current heading, there is a transition
between these two phenomenon and it is quite a complicated non-linear combination of modes. This
is illustrated in figure 3.9. In this figure, A10 stands for the mean of top 10% of the peaks amplitudes.

Figure 3.9: Plot of cross-flow motion amplitude response of the dominant oscillation versus reduced
velocity and attack angle [36]
Hence for a study merely on VIM behaviour, it is better to perform experiments on a body with
a diamond orientation. For more information, the reader is directed to Nemes et al. [36].

16 of 118

Geometry The starting point for vortex shedding behind a body lies in the separation of a flow
around it. It is obvious that the geometry of the body will affect the separation of the flow. So it
is with the vortex shedding, as well as the VIV and VIM. The geometries, or strictly speaking the
shape of the cross section, of the columns of multi-column floaters are quite versatile. In terms of
our topic, it is the corner shape of a square column, in other words, the corner sharpness, of great
value to us.
Revealed by Leontini and Thompson, the corner sharpness a diamond configured body has a significant impact on the oscillation amplitude, induced forces and temporal character of the elastically
mounted system [32]. They found that the amplitude and frequency of response are quite sensitive
to the corner radius. Moreover, for bodies with sharp corners, the flow induced response is more
fickle than for circular cylinders. According to open literature, this is owing to the disordered vortex
shedding. The sharper corners destroy the synchronization between the oscillation and vortex shedding [31]. Besides, as observed by Tamura et al, through a combined numerical and experimental
study, either rounding or chamfering the corners of a square-section cylinder leads to a significant
decrease in the drag on the rigidly fixed body [52].
Practically the sharpness of the corner is defined through corner radius. It is the ratio between
the radius of the corner and the diameter of the body. The relation can be expressed as:

2r
(3.3)
rc =
D
with r the radius of the corner.
In the experiments of Leontini and Thompson, they tested elastically mounted square cylinders
with different corner radii, from 0.2 to 0, plus a circular cylinder. Their results can be seen in the
figure 3.10, in short.
This figure shows a measure of the peak amplitude of oscillation for all of the geometries considered. In their conclusion, for corner radii rc 0.10, the appearance of a chaotic regime with
higher peak amplitudes than those achieved from a circular cylinder, does exist over a significantly
wide band of reduced velocity. For corner radii rc > 0.10, the response types are similar to those
encountered for a circular cylinder.
Interference between cylinders Unlike single cylinder cases, the presence of more than one
cylinder will experience flow field interference between each other when they are located in close
proximity. This is always the case for offshore engineering, as there are some structures having
more than one column, like semi-submersible floater, tandem configured risers, etc. In such problems, the flow field of multi-cylinder configurations are exposed to complex interactions between
the shear layers, vortices, wakes, vortex streets. Due to its complexity, a systematic definition and
classification is in need.
According to Sumner, the most basic multiple-cylinder configurations in cross-flow are shown in
figure 3.11 [50]. In this figure, U stands for the flow velocity, G is the gap width, L is the centre to
centre longitudinal spacing, T is the centre to centre transverse spacing, and the incidence angle.
A very illustrative dimensionless parameter is defined, as pitch ratio, to express the spacing between
cylinders in a non-dimensional form.
Three configurations are presented, which are:
Tandem configuration
Side-by-side configuration
Staggered configuration
For the first two configurations, they are quite straight forward and idealized. The third configuration can be regarded as a general form. During a study on multi-column floaters, all these three
cases will be encountered. Hence they are briefly introduced here.
Empirically, there are mainly five factors governing the flow behaviour:
The spacing between the cylinders

17 of 118

The orientation of cylinders with respect to the incoming flow direction


The Reynold number
Aspect ratio
Geometry of the cylinders
In order to understand the mechanism of the multi-cylinder interference, all the factor above
should be investigated. Once Zdravkovich made the basic classification of the interference. He
defined two types of interference and the conditions under which they occur [64]. It is illustrated by
the figure 3.12. They are:
Wake interference. One of the cylinder is partially or completely submerged in the wake of
the other.
Proximity interference. Two cylinders are located close to one another, but neither is submerged in the wake of the other.
This classification is very helpful to understand interference between cylinders, which is applicable to all the three configurations. For more details, see Zdravkovich [64].
For a tandem configuration, the wake of the upstream cylinder modifies the incoming flow condition for the downstream cylinder, while the downstream cylinder interferes with the wake dynamics
and vortex formation region of the upstream cylinder. It is quite sensitive to both the Reynolds number and longitudinal pitch ratio. With different longitudinal pitch ratio, three main flow patterns can
be observed [65]:(i) extended-body regime (ii) reattachment regime and (iii) co-shedding regime.
The wake structure and vortex dynamics, force coefficients and Strouhal number are all related with
the existence of the flow patterns above for a tandem twin cylinders.
For a side-by-side configuration, the Reynolds number effect is not that prominent compared
within the tandem configuration. It falls in the proximity interference region in terms of Zdravkovichs
plot. It can also be categorized by pitch ratio, which changes to transverse pitch ratio in this case.
There are mainly three flow patterns: (i) Single-bluff body behaviour (ii) biased flow pattern (iii)
parallel vortex streets.
For a staggered configuration, it is more complex as more parameters must be considered. Its
complexity of the flow field sources from the interaction of four separated free shear layers, two
vortex formations and shedding process, and interactions between the two vortex streets [50]. This
results in a large number of flow patterns. An overview is provided by Sumner and his colleagues
[51], within a low sub-critical regime for two staggered circular cylinders in cross-flow. They can
still be concluded as three types, as shown in the figure 3.13:(i) single bluff body pattern (ii) small
incidence angle flow pattern (iii) large incidence angle flow pattern.
For the specific relationship between each flow pattern and the corresponding measured coefficients, it is recommended to refer to the cited reference from Sumners work [50].
Aspect ratio Aspect ratio is another dimensionless parameter that affects the VIM behaviour of
a body. It is usually defined as the relation between the effective immersed length and the effective
length of the cross-section, within the form:
L
(3.4)
D
It is generally acknowledged that with the decreasing of aspect ratio the frequency of vortex
shedding will decrease [16]. A critical value can be defined as ARcrit = 2.0, which accounts for
the changing point for the down stream flow model. For aspect ratio above 2.0, the existence of a
vortex shedding is proposed, while for aspect ratio below 2.0 the vortex street collapses. It provides
an insight about the changing of the Strouhal number as a consequence of the flow field transition.
When the aspect ratio is approaching an infinite value, or just sufficiently high, then the flow
around the cylinder can be taken as a two dimensional case.
AR =

18 of 118

Figure 3.10: Plot of the oscillation amplitude of elastically mounted cylinders with different corner
radii versus reduced velocity

Figure 3.11: Two cylinders in cross-flow: (a) tandem configuration (b)side-by-side configuration
and (c) staggered configuration. [50]

19 of 118

Figure 3.12: Classification of flow field interference and its boundary [64]

Figure 3.13: Low subcritical regime flow patterns for two staggered circular cylinders in cross-flow
[51]

20 of 118

3.2

Importance to offshore industry

Most of the offshore structures are exposed to an incoming current, which might trigger significant oscillations, known as the vortex-induced motions. Nowadays VIM is widely accepted as a key
concept in the design of semi-submersibles, SPAR and TLP platforms. It has drawn lots of attentions
as it plays an important role during the fatigue life of marine risers and mooring systems. If precise
predictions of VIM behaviours of a structure can be made during the design stage, bearing in mind
the knowledge of the state of art of VIM, then it would be useful to predict VIM in order to not
over design the risers. This can reduce costs. Herewith the study on VIM phenomenon is of great
importance to the offshore industry.
3.2.1

Fatigue

It is well known to the offshore industry that VIM is one of the key factors that determines the
fatigue life of risers and mooring lines. They will be elaborated in this section respectively.
Fatigue of risers VIM dominates the lower frequency fatigue of structures. It is seen that VIM
induced fatigue is almost as important as the other aspects such as VIV fatigue, wave induced fatigue
and installation fatigue, in the fatigue design of the riser [2], especially for steel catenary risers
(SCR), which are widely used in the industry. According to the design experience from the open
literature, in certain cases VIM motion can account for 24.1% of the total fatigue damage for risers,
compared with the fatigue caused by wave, installation, VIV and heave VIV [49].
A study from Xiang et al. [62] shows that VIM affects primarily SCR fatigue life at the touch
down zone . In certain cases it can make up 50% of the total fatigue damage. It also reveals that riser
VIM fatigue is mainly affected by the combination of column shape, current heading and intensity,
the individual riser configuration, mooring line pretension and polyester stiffness. In order to know
the contributions from each factor to VIM behaviour, in case study a VIM sensitivity analysis is
always performed.
Fatigue of mooring chains Mooring analysis indicates that low frequency fatigue, governing the
mooring system design, has a clear relation with the size of mooring components. VIM accounts
for almost all of the low frequency damage in the mooring system. So during the design stage of
the mooring system it is important to take the VIM fatigue into consideration. A good example
is given by Park and his colleagues [38]. In their work, a comprehensive study on the fatigue of
mooring lines caused by VIM is carried out through tow tank testing, flume tank testing and CFD.
Field measurement is also included. The results provide an overview of the relationship between
VIM and mooring chains fatigue.
3.2.2

Suppression of vortex-induced oscillations

According to the research findings so far, there are several directions to go:
Control of the structural design so as not to exceed the critical values of reduced velocity;
Control of the heading of the structure to have a favourable angle towards the incoming current;
Design a structure with sufficiently large mass and damping;
Optimize the layout of the multi-body structure to mitigate the vortex shedding effect through
flow field interference;
Modify the flow field around the structure by altering the flow or the structural shape.
A recent study by Goncalves and his colleagues unveils that a favourable heading and a decrease
in draft of the structure can greatly contribute to the mitigation of VIM on a mono-column platform
[18]. Albeit it is not generally applicable as the direction of incoming current is always in random,

21 of 118

and the draft is always determined according to the weight of the topside and equipments. Apparently the most feasible strategy is the last one. In an early age, there were some attempts made by
researchers to reduce or suppress the oscillation by the installation of external devices that modify
the flow field around the structure. Such devices, like strakes (helical strakes), porous shrouds, nearwake stabilizers and fins, were applied and proven to be of great effect. Besides, there is evidence
showing that the suppression effectiveness of these devices corresponds with the reduced damping
[24], which is also known as Scruton number and has been discussed in section 3.1.4.
A more specified classification of means for suppressing vortex shedding is given by Zdravkovich
[63]. In accordance with the phenomenological mechanism of vortex shedding, the wide variety of
means for suppressing vortex shedding is classified into three categories:
Surface protrusions, which affects separation lines or separated shear layers.
(strakes, wires, fins, studs, spheres, etc.)
Shrouds, which affected the entrainment layers.
(perforated, gauze, axial rods, axial slats, etc.)
Nearwake stabilisers, which affect the switch of the confluence point.
(splitters and saw tooth plates, guiding plates and vanes, base-bleed, slits cut along the cylinders, etc.)
In such classification, two concepts are defined. An entrainment layers supply irrotational fluid
necessary for the growth of vortices in addition to the rotational fluid, in separated shear layers.
While the confluence point marks the region where the two entrainment layers coming from the
opposite sides of the cylinder meet and interact.
A sketch of different means for interfering with vortex shedding is given in figure 3.14. In that
plot, row (i) illustrates surface protrusions, row (ii) shows shrouds and row (iii) exhibits nearwake
stabilisers. Moreover, the symbol plus means an effective factor to vortex shedding. On the contrary,
the symbol minus indicates an ineffective factor.
These are the proved technology to mitigate or suppress the vortex shedding, by means of both
aerodynamic and hydrodynamic methods.
Nevertheless, those methods are not applicable to semi-submersible. Compared with spar platform, a multi-column floater always has a small draft. Thus the deployment of suppression devices
(eg. helical strakes) could only happen in the vicinity of water line area. Since the orbital motion of
water particles are quite fierce around this area, the suppression devices are quite vulnerable. Hence
it is difficult for normal semi-submersibles to have suppression devices on them. As a consequence,
a different way to tackle the problem is in need. An alternative method lies in the structural property of the floater. Since damping is the main contributor of energy dissipation of the system, it is
reasonable to focus on how to increase the damping of the system to mitigate VIM. The application
of pontoons is a good option. Pontoon can add a lot of damping to the system and thus mitigate
VIM of the floater. Nowadays many semi-submersibles have pontoons attached to columns. They
are proven to be very useful for the suppression of VIM.

22 of 118

Figure 3.14: Various means for interfering with vortex shedding [63]

23 of 118

3.3

Computational fluid dynamics(CFD) on VIM

The most common method to study the behaviour of an offshore structure in a certain sea state
is still the model test. It is a proven technology that has been adopted by the industry for over
hundreds of years. A lot of experiments on VIM through model tests can be referred in the review
made by Fujarra, et al. [16], as well as in the work done by Ding and his colleagues [7]. The
results given by these model tests are of great value to the study on VIM. However, there is existing
evidence revealing that the field measured VIM response could be significantly less than the response
predicted by the mode test results, as reported by Park, et al., [38] and Xiang, et al. [62], as well
as Rijken [41]. The observed discrepancy between the model and field measurement indicates the
similarity law is violated. Hence a more accurate and solid method to study and predict the VIM
behaviour of a structure in real life is in need.
CFD is a good candidate. With the development of numerical method and hardware, at the moment CFD can provide acceptable results for hydrodynamic and motion analysis. Plenty of studies
on VIM are based on CFD and have acquired good results comparable to model tests. For instance,
like a CFD simulation by Tan et al. [53], the URANS calculations for a circular cylinder by Rosetti,
Vaz and Fujarra [46]. A comparison between CFD and experiments results from Ding and his colleagues work also exhibits good agreement between these two methods [7]. Hence CFD can be
taken as an alternative.
Compared with model tests, CFD is faster, cheaper and more flexible and efficient, and thus more
suitable in the early design stage, especially when the parameters are unknown to designers. Tests
with different set-up can be performed in CFD with less cost compared with conventional model
tests, in terms of both time and expense. When there is enough computation resource, a sequence of
calculations can be triggered and run simultaneously, which is quite budget and time efficient. This
feature can be adapted to inspect the effect from each factor on the VIM behaviour of a structure,
like geometry of the body, heading of the structure, aspect ratio, mass ratio, etc. As a result, it
can play an important role during the decision making process. Moreover, since the data from
CFD calculation can be extracted everywhere around the computational domain, it is very useful
in understanding phenomena that are difficult to measure experimentally such as vortex structures,
motion of separation point, and pressure distribution for multiple cylinders [7]. Thus CFD can be
regarded as an effective tool to throw light on VIM and VIV problems.
In spite of its advantages, the accuracy of CFD calculations are highly dependent on mathematical models and numerical methods, which can collapse the precision of the results if improperly
configured. The dimension of computational domain, like cell size, y plus value, cell skewness and
orthogonality degree, can lead to deviation of CFD results when the adopted one is not suitable for
a certain problem. In addition, so far it is quite difficult to accurately capture the characteristics of
transition zone from laminar flow to turbulent flow in great detail numerically. This also results in
somewhat deviation to final results.
All in all, it is convincing to say that CFD is an excellent tool to be used in VIM analysis. However, special care should be taken towards numerical setting and the use of mathematical models.

24 of 118

Numerical Settings

In this section, an overview of the numerical configurations used in this study is provided with
detail. The geometry of the structure and the structural properties are also presented here. Moreover,
a study on grid and time sensitivity for the CFD VIM calculations is elaborated at last. This could
guarantee the reliability of the results from CFD computations on the chosen grid and time step.

4.1

Geometry of the floater

In this study, all the calculations are carried out on 2D models. Thus the topology and structural
properties of 3D floater model should be projected into 2D. Thus the weight of the entire floater in
3D, including columns and pontoon, is equally divided to four columns in 2D. Since it is not possible
to model the columns and pontoon simultaneously in 2D, only the columns of the floater would be
preserved here. The mass of pontoon is split and added to each column. Its structural property is
shown in chapter 5.
Hence a 2D floater with four columns is used here. Each column has a square shape with circular
corners. This is shown in figure 4.1. The diameter of the column is D = 0.27m, the corner radius
is R = 0.075D = 0.02025m. The model should be constructed and meshed in a grid generation
software. Hence, a height is in need. For computational purpose, the height of each column is set to
be unit length L = 1m. The cross-section area of each column is S = 0.0725m2 .

Figure 4.1: Cross-section geometry of each column


As discussed in section 3, the current heading will affect the response of exposed structure.
We are interested in 45 degree configuration, where a higher VIM response is expected. Hence
a definition for the characteristic length should be given. It is defined as the so-called effective
diameter. The essence for this definition is illustrated in figure 4.2.
The effective diameter for 45 degree configuration is therefore Def f = 0.365m. The centre
to centre distance between two columns is Dcc = 4D = 1.08m. Now the layout of the floater is
settled. A perspective view of the whole floater is given in figure 4.3. In this figure the projected
area of the floater with respect to the 45 degree current heading is presented as well. It will be
used during the calculation of the force coefficients. The effective projected area of the floater is
Sp = 3 LDef f = 1.095m2

4.2

Computational domain and mesh


The geometry of the computational domain is shown below in figure 4.4. The size of computa-

25 of 118

Figure 4.2: Effective diameter for different configuration according to current heading

Figure 4.3: Diamond layout of the floater with four columns


tional domain should be large enough to avoid undesired influence on the solution. As this study is
fully 2D, hence the height of the domain has no influence. The domain has a cylinder shape. The
radius of domain is chosen to be Ddom = 40D = 10.8m, 40 times the diameter of column and 10
times the centre to centre distance of the columns.
The mesh of the whole computational domain containing the floater is made in Hexpress. The
procedure is briefly introduced here. The whole meshing process is accomplished through the mesh
wizard, in which the mesh is generated step by step with minimum user inputs. It includes five steps,
as follows:
1. Initial mesh. An initial mesh surrounding the computational domain is created.
2. Adapt to geometry. In this step, it adapts the initial mesh to make cell sizes satisfy geometry
dependent criteria and removes afterwards cells intersecting the geometry or located outside
the computational domain.
3. Snap to geometry. The adapted mesh from last step is projected onto the geometry and lower
dimensional features are recovered, such as topological edges (curves) and topological vertices
(corners).
4. Optimize. All the concave cells, which come from the snapping action, are turned into convex
ones.
5. Insert viscous layers. Layers of high aspect ratio cells are inserted close to the wall in this step
in order to correctly resolve boundary layers.

26 of 118

Figure 4.4: Geometry of the computational domain


As soon as the CAD model is constructed in Hexpress and transformed to a computational domain format that can be recognized by Hexpress, the mesh wizard can be activated. The overview
of the grid is provided in figure 4.5.

Figure 4.5: Overview of the grid


Here a summary of the geometry of floater and computational domain is given in table 4.1.
4.2.1

Grid refinement

The floater is placed in the centre of the domain. Around the floater different refinement levels
are applied, so as to capture the details of vortex shedding and interaction between structure and
fluid. Meanwhile it also contributes to the control of the total number of cells to an affordable
degree in terms of computational costs. The refinement map is highlighted in figure 4.6. If we take
the cell size outside the refinement area, clearly illustrated in figure 6.5, as the unit length L, then

27 of 118

Table 4.1: A summary of the geometry of floater and computational domain


Category
Diameter of the column
Radius of the rounded corner
Centre to centre distance for columns
Radius of the computational domain
0 degree effective length
45 degree effective length

Relative length
D
0.075D
4D
40D
-

Value
0.27
0.02025
1.08
10.8
0.27
0.365

Unit
m
m
m
m
m
m

the refinement rate can be explained by the annotation in figure 6.6. Actually, the refinement of the
grid is carried out through division of cells. When the number of refinement is specified, then the
original cell will be divided into small cells according to that number. Hence the number of division
can be expressed as the square of refinement number.

Figure 4.6: Typical cell size around the floater


The cell size plays a very important part in both the convergent behaviour and the precision of
calculation. Its importance can be explained through Courant number. Courant number, also known
as cell Courant number, is defined as follow:
U t
(4.1)
x
In terms of its definition, Courant number describes how the fluid is moving through the computational cells. If the Courant number is CCourant < 1, fluid particles move from one cell to another
within one time step, at most. If it is CCourant > 1, a fluid particle moves through two or more cells
at each time step and this can affect convergence negatively. A higher Courant number for a CFD
calculation will result in a poor stability, especially for the calculations applying explicit methods
to the generic transport equation. Moreover, if the fluid particles moves through too many cells at
each time step, then it is likely that the transition from laminar flow towards turbulent flow can not
be captured accurately. Thus Courant number can be taken as the wind vane for both the stability
and reliability of the CFD calculation results.
In this case, since a combined explicit and implicit method is adopted, the stability is not highly
dependent on Courant number. However, a high resolution around each column and the wake is
in need, where the flow transition and structure-fluid interaction are present. Thus small cell sizes
around the column are required in this study to capture enough details. Obviously we are more
CCourant =

28 of 118

interested in the Courant number around the columns and downstream wake. As the Courant number
is determined by cell size, time step length and flow velocity, then it is necessary to choose an
appropriate time step. This issue will be discussed in section 6.3.
4.2.2

Viscous layers

A boundary layer is the layer of fluid in the immediate vicinity of a bounding surface where
the effects of viscosity are significant. Due to the complexity of the flow in such area, close to the
boundary of the bodies, a more refined mesh is in need to provide reliable computational results.
This is the reason for insertion of viscous layers around the surface of the structure.
In CFD, there is a non-dimensional parameter describing wall distance for a wall-bounded flow
near the surface of the immersed body, named y+. It has the form like:
u y
(4.2)

where u is the friction velocity at the nearest wall, y is the distance to the nearest wall and is the
local kinematic viscosity of the fluid.
From a technical point of view, y+ is a non-dimensional distance. It is often used to describe
how coarse or fine a mesh is for a particular flow pattern. It is important in turbulence model to
determine the proper size of the cells near domain walls.
From a user point of view, y+ is a ratio between turbulent and laminar influences in a cell. If
y+ is big then the cell is turbulent, otherwise it is laminar. The importance in many cases of this
concerns wall functions which assume that the laminar sub-layer is within the first cell.
y+ =

Figure 4.7: Zoom in view of viscous layers


No wall function is used in this case. The turbulence model has restrictions on the y+ value at
the wall. For the use of k SST Menter model, a y+ < 1 is much more preferred to achieve
reliable results. A faster flow near the wall will produce higher values of y+, so the grid size near the
wall must be reduced. After some tests, finally a viscous layer containing 10 tiers is applied around
each column. A zoom-in view of the insertion of viscous layers are shown in figure 4.7.

29 of 118

4.2.3

Boundary conditions

Five types of boundary conditions are used in this case, four of which are already shown in figure
6.4 with the boundary surface that they are applied to. The last one not shown in the figure is the
boundary surface of each column, which are setted to be wall boundaries. Here is a table briefly
summarize the boundary conditions:
Table 4.2: Summary of boundary condtions
Boundary surface
Surface
Bottom
Far-field
Columns surface

Boundary conditions
Slip-wall boundary
Symmetry boundary
Inflow (Velocity inlet) & Pressure outlet
No-slip wall boundary

These boundary conditions are introduced in short below.


Symmetry boundary For this boundary condition, it is assumed that on the two sides of the
boundary, the same physical processes exist.There is no convective flux across a symmetry plane:
the normal velocity component at the symmetry plane is thus zero. There is no diffusion flux across
a symmetry plane: the normal gradients of all flow variables are thus zero at the symmetry plane.
Since the shear stress is zero at a symmetry boundary, it can also be interpreted as a slip wall when
used in viscous flow calculations.
Zero normal velocity at a symmetry plane
Zero normal gradients of all variables at a symmetry plane
Slip-wall boundary The slip wall condition is for cases where viscous effects at the wall are
negligible. It is also the proper boundary condition for symmetry surfaces. Like the symmetry
surface, the wall has no effect on the fluid velocity in planes parallel and close to the wall.
Inflow (Velocity inlet) The inflow boundary condition defines velocity vector and scalar properties
of flow at inlet boundaries. It is quite useful when velocity profile is known at inlet.
Pressure outlet The pressure outlet boundary defines the static/gauge pressure at the outlet boundary. It is interpreted as the static pressure of the environment into which the flow exhausts. As a big
advantage of all the pressure boundary conditions, they can be applied when neither the flow rate
nor the velocity are known.
Non-slip wall boundary Wall conditions are used to bound fluid and solid regions. In viscous
flows, no-slip condition enforced at walls, under the assumption of Newtonian fluid. On the contrary
to slip wall, it does not allow slip of fluid particles at the wall surface. Its effect is shown below:
The tangential fluid velocity equals to wall velocity.
Normal velocity component is set to be zero.
4.2.4

Summary of the grid properties

An overview of the grid properties is given in table 4.3. The reason for selection of this grid is
elaborated in section 6.3, through solution verification.
Empirically, for a good quality a grid may satisfy the criterion below:
M inimum orthogonalitydegree > 35
M aximum skewness < 0.7
No negative, concave or twisted cell is allowed
Hence the selected grid is considered of good quality.

30 of 118

Table 4.3: Summary of the grid properties


Category
Number of cells
Minimum orthogonality degree
Maximum skewness
Negative, concave or twisted cells

4.3

Value
221258
44.6
0.632
None

Solution verification

In this subsection the solution verification is shown in detail. The theoretical background is
described in section 11.2 in appendix.
According to the solution verification tests configurations from open literature [13, 12], at least 4
geometrically similar grids are required. Within scope of the workload and available resource, four
grids and time steps are inspected here.
The properties of investigated grids are shown in table 4.4. They have same geometry and approximately same grid refinement ratio. Since they are all unstructured grids generated automatically
in Hexpress, it is quite difficult to keep the geometrical similarity. As a result, it has negative effects on the convergent behaviour of the results. It will be further discussed during the evaluation of
discretization errors.
Table 4.4: Grid size and refinement level for solution verification
Grid
Number of cells
Minimum othogonality degree
Maximum Skewness
Negative, concave or twisted cells
y plus
hi/h1

No.1
221k
44.60
0.632
None
1.0
1.000

No.2
118k
44.60
0.632
None
1.0
1.367

No.3
72k
42.97
0.630
None
1.0
1.752

No.4
46k
42.97
0.630
None
1.0
2.196

Four time steps are selected for inspection with respect to the ratio between reference time and
time step (see Eq. 2.15). They are shown in table 4.5.
Table 4.5: Time step level for solution verification
Time step
Tref /t
Time step(s)
ti/t1

No.1
80
0.033
1.00

No.2
60
0.044
1.33

No.3
50
0.053
1.60

No.4
40
0.066
2.00

The selection for those grids and time steps are based on pretests, with reference to the work
by Rosetti et al[46]. Besides, it is also a compromise between accuracy and available computation
resource.
For the sake of an easy and clear narration, in the following parts calculations on a certain grid
with a certain time step will be represented by the combination of its acronym and number . For
example, a calculation on grid No.2 and time step No.3 will be cited as T3G2.
The start point for a solution verification lies in the CFD calculation itself. In this case it is the 45
degree VIM CFD calculation, which has the structural property obtained from decay tests, as shown
in table 5.6. The current velocity is 0.139m/s, calculated from the initial scaling factors given in
table 5.2. The tested flow variables are listed below:
RMS A/D: Root mean square of dimensionless cross-flow motion amplitude;

31 of 118

RMS Clift: Root mean square of lift force coefficient;


Aveg Cdrag: Mean value of drag force coefficient;
RMS Cmoment: Root mean square of moment coefficient;
St: Strouhal number.
The main focus of the solution verification of a CFD calculation lies in the estimation of numerical errors and uncertainties. As mentioned previously, the round-off error is negligible here. Hence,
the results are discussed respectively with regard to iterative error and discretization error.
4.3.1

Iterative error

As mentioned in section 5, the iterative error can be monitored by the inspection of residuals
obtained during the computations in ReFRESCO. In order to know the influence from iterative error
on our final results, it is feasible to compare the results from calculations with different convergence
tolerance levels. In terms of the budget and feasibility, four calculations with same configurations but
different convergence tolerance level are carried out in this study on the finest grid No.1 and second
finest time step No.2 (T2G1). The convergence tolerance criteria is based on L norm. They are
shown in table 4.6.
Table 4.6: Calculations with different convergence tolerance for the inspection of influence from
iterative error
Number of calculation
Convergence tolerance level

No.1
103

No.2
104

No.3
105

No.4
106

The L norm residuals for all calculations are provided in figure 4.8.
Except for turbulence quantities, all the other residuals strictly obey the convergence tolerance
criterion. Since it will converge to the required tolerance level after a few time steps, it will not
cause problems to our study. Hence it is reasonable to conclude that these calculations follow the
convergence tolerance requirement well and thus these calculations can provide reliable results.
Such phenomena also happens to L2 norm residuals. The L2 norm residuals from these 4 calculations are shown in figure 4.9. Because L2 norm is less strict than L norm, so L2 norm residuals
are much smaller than L norm residuals.
The selection process for a convergence tolerance level is also a balance between precision and
cost. It is quite straight forward to see that more strict convergence criteria we use, more computation
time it will require. The computation time is shown in table 4.7. In this study all the computation
time will be expressed as CPU core time. Note that all the calculations were complete using 64
cores. Obviously a more strict convergence criteria will lead to a dramatic increase in computation
time.
Table 4.7: Computation time of calculations with different convergence tolerance for the inspection
of influence from iterative error
Convergence tolerance level
Computation time (CPU hrs/core)

103
44

104
75

105
124

106
169

The relation between computation time and convergence criterion can be linked by the outer loop
check, which sketches the number of iterations per time step required to reach a certain convergence
tolerance. The outer loop circumstance of each calculation is shown in figure 4.10. In this figure, it
can be observed that for the calculation with convergence tolerance 103 , on average, 150 iterations
are necessary to reach its convergence criteria. While for calculation with convergence tolerance

32 of 118

(a) Convergence tolerance 103

(b) Convergence tolerance 104

(c) Convergence tolerance 105

(d) Convergence tolerance 106

Figure 4.8: L norm residuals of calculations with different convergence tolerance level for the
inspection of influence from iterative error
106 about 500 iterations are required. This attributes to the difference in computation time. Maximum iterations are met at the very beginning of the calculations. Needless to say it is due to the
turbulence quantities.
Calculation results of flow variables are shown in figure 4.11.
Comparison between results from the calculations with several convergence tolerance levels is
given in table 4.8. In this table the difference between results calculated with the most strict convergence tolerance 106 and others are shown in percentage.
Table 4.8: Difference between results calculated with different convergence tolerance criteria
No.
No.1
No.2
No.3
No.4

Convergence tolerance
103
104
105
106

RMS A/D
1.80%
0.10%
0.40%
0.00%

Aveg Cdrag
3.30%
1.50%
3.10%
0.00%

RMS Clif t
8.60%
4.40%
4.30%
0.00%

Cmoment
0.20%
0.20%
0.40%
0.00%

St
0.53%
0.00%
0.79%
0.00%

According to the results from table 4.8, the difference between flow variables caused by convergence tolerance is not very large. It seems both 103 and 104 are good candidates.
A comparison between the results under convergence tolerance 103 and 104 is provided in
figure 4.12 and 4.13. Figure 4.12 depicts a time refinement study on the finest grid (G1), while
figure 4.13 describes a grid refinement study with the second finest time step (T2).
In figure 4.12, it illustrates results under iterative tolerance 104 tend to have a smaller variation.
In other words, iterative error under the convergence tolerance 103 could affect final results and
lead to large fluctuation. Flow variables are sensitive to cell size in this case. Thus the grid refinement
study should have results showing convergence tendency. In figure 4.13, the results from calculations

33 of 118

(a) Convergence tolerance 103

(b) Convergence tolerance 104

(c) Convergence tolerance 105

(d) Convergence tolerance 106

Figure 4.9: L2 norm residuals of calculations with different convergence tolerance for the inspection
of influence from iterative error
with iterative tolerance 103 are scattering, while those with iterative tolerance 104 show a good
agreement with expectation.
Considering both the computation time and precision, the convergence tolerance 104 will be
used in the following calculations.
4.3.2

Discretization error

The discretization error is always the dominant component for numerical errors. The method
used in this section to estimate the uncertainty is a combination of Richardson extrapolation and the
least square version of grid convergence index. They are explained in section 5 at length.
A sequence of calculations on grids with different cell size and different time step were carried
out. As depicted in last subsection, 32 calculations are carried out on 4 grids with 4 different time
steps and 2 convergence tolerance criterion. The computation time for these 32 tests is given in table
4.9 and table 4.10, with respect to convergence tolerance 103 and 104 . Still, the computation time
is given in the form of CPU core time. Due to the limitation of computation resource and time, it
is not affordable for more tests. Obviously a calculation with more strict convergence criteria could
take longer.
The iterative error check shows that the convergence tolerance 103 is not enough. Hence here
only the results from calculations under the 104 convergence tolerance are shown, in table 4.11.
In order to analyse the impact respectively from grid (cell size) and time step on the final results,
the results are plotted against grid refinement ratio and time refinement ratio. This is the method
normally adopted in discretization error analysis according to open literature[43, 45]. Figure 4.14
shows the time step study on each grid. Meanwhile, in figure 4.15 the grid refinement study with
each time step are shown.

34 of 118

(a) Convergence tolerance 103

(b) Convergence tolerance 104

(c) Convergence tolerance 105

(d) Convergence tolerance 106

Figure 4.10: Outer loop check of calculations with different convergence tolerance
Table 4.9: Computation time for VIM calculations on 4 grids with 4 time steps under the convergence
tolerance 103

CPU hrs/core

Grid

221k
118k
72k
46k
Cell number

1
1.367
1.752
2.196
hi/h1

80
0.033
1
41
22
14
11

Time step
60
50
0.044 0.053
1.333
1.6
44
49
26
27
16
17
12
13

40
0.066
2
51
29
19
14

Tref /t
Time step
ti/t1

64 nodes in use

In table 4.12, a comparison between results on finest grid and finest time step with others is
provided. It will provide an insight about grid and time step sensitivity of CFD calculations and
help to choose the most suitable grid and time step. The difference is given in percentage as a ratio
between difference and results on finest grid with finest time step.
In figure 4.14, apparently results on the finest grid (G1) change a little with respect to different
time step, except lift coefficient and Strouhal number. In table 4.12 a quantitative comparison is
available. Lift coefficient and shedding frequency are quite sensitive to Courant number. As a result,
despite of lift coefficient and Strouhal number, it is reasonable to reach a conclusion that CFD VIM
calculation on the finest grid in this case is not sensible to time step length.
According to figure 4.15, unlike time step study, results within finest time step (T1) differ a lot
on grids with different cell size. They all follow certain trends with the change of grid refinement
ratio A summary can be made based on the observation that CFD VIM calculation within finest time

35 of 118

(a) RMS A/D

(b) Average Cdrag

(c) RMS Cdrag

(d) RMS Cmoment

(e) St

Figure 4.11: Results of CFD calculations with different convergence tolerance for the inspection of
influence from iterative error
step in this case is quite sensible to grid/cell size.
Besides, results on finest grid/time step tend to have smaller variation compared with those on
coarse grid/time step. In other words, results on fine grid/time step are more stable. Figure 4.14
reveals that results on finest grid (G1) are more stable and less fluctuated than those on coarse grids.
Moreover, it also implies that gird No.3 and No.4 could not provide reliable results, as they deviate
too much compared with those on the finest grid. Meanwhile, figure 4.15 unveils that results on
finest time step (T1) are more stable than those on coarse time steps. Thus, if possible, it may as
well use finest gird and time step.
All in all, due to the balance between computation resource and precision, in this case the best
combination of numerical setting is grid No.1 and time step No.2 (T2G1). In a nutshell, CFD VIM
calculations in this study will be conducted on the finest grid (221k cells) with second finest time
step (Tref /t = 60).

36 of 118

(a) RMS A/D

(b) Average Cdrag

(c) RMS Clif t

(d) RMS Cmoment

(e) St

Figure 4.12: A time step study on the finest grid (G1) with results under convergence tolerance 103
and 104
Table 4.10: Computation time for VIM calculations on 4 grids with 4 time steps under the convergence tolerance 104

CPU hrs/core

Grid

221k
118k
72k
46k
Cell number

1
1.367
1.752
2.196
hi/h1

80
0.033
1
66
34
21
14

Time step
60
50
0.044 0.053
1.333
1.6
70
72
37
37
22
23
15
16

40
0.066
2
77
40
24
16

Tref /t
Time step
ti/t1

64 nodes in use

In order to observe the convergence behaviour of the results considering both the grid refinement
study and time step study, 3D plots of each variable against the grid refinement ratio and the time
step ratio are quite useful. They are shown in figure 4.16. The green lines in those figures indicate
tests with second finest time step on finest grid (T2G1). All the variables are convergent over the
grid refinement ratio and time step ratio domain.
In terms of the 3D fitting results, the scattering data should be categorized as oscillatory con-

37 of 118

(a) RMS A/D

(b) Average Cdrag

(c) RMS Clif t

(d) RMS Cmoment

(e) St

Figure 4.13: A grid refinement study on the second finest time step (T2) with results under convergence tolerance 103 and 104
vergence. According to the uncertainty estimation procedure described in section 5, founded on
Richardson extrapolation and standard GCI method, the uncertainty of each variable can be calculated. The only condition which allows an error estimation based on Richardson extrapolation
is monotonic convergence. For an oscillatory behaviour, uncertainty quantifications are achieved
through the maximum difference between all the solutions, known as an alternative uncertainty
quantification, which could be expressed as equation 5.9: M = max(|i j |). Finally the
uncertainty could be estimated through Ud () = 3M . The result is listed in the following table
4.13.

38 of 118

Table 4.11: Final results for inspection on grid and time step sensitivity for CFD VIM
Grid
Grid No.1
(221k cells)

Grid No.2
(118k cells)

Grid No.3
(72k cells)

Grid No.4
(46k cells)

Tref /t
40
50
60
80
40
50
60
80
40
50
60
80
40
50
60
80

RMS A/D
0.316
0.325
0.318
0.319
0.330
0.331
0.319
0.312
0.300
0.298
0.297
0.293
0.292
0.292
0.291
0.288

RMS Clif t
0.411
0.457
0.437
0.390
0.538
0.506
0.393
0.398
0.403
0.377
0.299
0.360
0.355
0.367
0.238
0.346

Aveg Cdrag
2.509
2.489
2.501
2.523
2.409
2.378
2.489
2.492
2.429
2.443
2.482
2.469
2.435
2.429
2.448
2.413

RMS Cmoment
0.283
0.302
0.293
0.303
0.265
0.277
0.297
0.284
0.275
0.285
0.269
0.265
0.260
0.250
0.256
0.258

St
0.221
0.205
0.205
0.197
0.205
0.205
0.213
0.205
0.213
0.213
0.213
0.205
0.213
0.213
0.221
0.221

As a summary of this section, conclusions achieved through the inspection of grid and time step
sensitivity for CFD VIM calculations are listed here:
The results of CFD VIM calculations are more sensitive to the change of grids (cell size), but
are less sensitive to the change of time steps.
The results obtained under a iterative tolerance 103 are greatly affected by iterative error and
thus are not reliable. While, results acquired under a iterative tolerance 104 are similar to
the results achieved with more strict convergence tolerance criterion. With a compromise on
limited computation resource and time, the iterative tolerance 104 is the best choice for this
study.
According to discretization error analysis, a grid with 221k cells could provide precise and
reliable results rather than less refined grids. A time step determined through Tref /t = 60
is fine enough to keep a favourable Courant number and capture the details of the flow field.
This choice is also made through a balance between available computation resource and time,
and the precision and reliability of the yields.

39 of 118

(a) RMS A/D

(b) Average Cdrag

(c) RMS Clif t

(d) RMS Cmoment

(e) St

Figure 4.14: Plots of time step study on each grid

40 of 118

(a) RMS A/D

(b) Average Cdrag

(c) RMS Clif t

(d) RMS Cmoment

(e) St

Figure 4.15: Plots of grid refinement study with each time step

41 of 118

Table 4.12: A comparison between results on finest grid with finest time step and those on coarse
grids with coarse time steps
Grid
Grid No.1
(221k cells)

Grid No.2
(118k cells)

Grid No.3
(72k cells)

Grid No.4
(46k cells)

Tref /t
40
50
60
80
40
50
60
80
40
50
60
80
40
50
60
80

RMS A/D
0.94%
1.88%
0.31%
0.00%
3.45%
3.76%
0.00%
2.19%
5.96%
6.58%
6.90%
8.15%
8.46%
8.46%
8.78%
9.72%

RMS Clif t
5.38%
17.18%
12.05%
0.00%
37.95%
29.74%
0.77%
2.05%
3.33%
3.33%
23.33%
7.69%
8.97%
5.90%
38.97%
11.28%

Aveg Cdrag
0.55%
1.35%
0.87%
0.00%
4.52%
5.75%
1.35%
1.23%
3.73%
3.17%
1.63%
2.14%
3.49%
3.73%
2.97%
4.36%

RMS Cmoment
6.60%
0.33%
3.30%
0.00%
12.54%
8.58%
1.98%
6.27%
9.24%
5.94%
11.22%
12.54%
14.19%
17.49%
15.51%
14.85%

St
12.00%
4.00%
4.00%
0.00%
4.00%
4.00%
8.00%
4.00%
8.00%
8.00%
8.00%
4.00%
8.00%
8.00%
12.00%
12.00%

Table 4.13: Estimation of uncertainty based on results on grids No.1 to No.4 with time step No.1 to
No.4 under a iterative tolerance 104
Variables
Uncertainty

RMS A/D
12.90%

Aveg Cdrag
43.50%

RMS Clif t
90.00%

42 of 118

RMS Cmoment
15.90%

St
7.20%

(a) RMS A/D

(b) Average Cdrag

(c) RMS Clif t

(d) RMS Cmoment

(e) RMS Cmoment

Figure 4.16: 3D plots of each variable against the grid refinement ratio and the time step ratio

43 of 118

CFD tests and results

In this section, all the numerical computation configurations, process and results are presented,
including decay tests and VIM tests. The post-processing and analysis of each result are also elaborated in each subsection. All the CFD tests are carried out with the water properties shown in table
5.1:
Table 5.1: Water property
Category
Temperature
Density
Dynamic viscosity
Kinematic viscosity

5.1

Value
15
999
1.139E-3
1.139E-6

Unit

C
kg/m3
kg/ms
m2 /s

Decay tests

In order to determine the structural properties of a certain structure, decay tests are in need.
It is also a start point for the rest of work. In this section the set-up for decay tests, results and
post-processing are discussed.
5.1.1

Set-up decay tests

In this study a 2D CFD calculation coupled with equation of motion on a four columns semisubmersible is presented. Unlike a CFD calculation for a fixed body, the semi-submersible floater is
free to move in 3 DOFs (surge, sway and yaw). Thus a replacement for mooring system is required.
The properties of the target structure and the environmental conditions are defined through scaling
factors. The desired natural frequency of the structure can be calculated. With all these achieved
parameters, a decay test can be setted up with any stiffness, functioning like a trial providing raw
data for the analysis of hydrodynamic damping and added mass coefficient. When the hydrodynamic
added mass is obtained, then the stiffness coefficient we need, for the sake of a desired natural
frequency of the whole system, can be determined. Through this manner the structure we modelled
has the same physical meaning with the model in real life.
Here is the scaling factors for the set-up of decay tests, shown in table 5.2. They are kept the
same for all the rest of CFD calculations, unless mentioned otherwise.
Table 5.2: Scaling factors for decay tests
Category
Mass ratio
Reynolds number
Reduced velocity
Ratio between sway and yaw period
Ratio between reference time and time step size

Abbreviation
m
Re
Ur
Tsway /Tyaw
Tref /t

Value
1
44453
6
1.850
60

With the equations of these scaling parameters listed in section 2.2, all the parameters playing a
part in decay tests can be calculated. The results are shown in table 5.3.
Since the effective length for different layouts of the floater differs from each other, with respect
to the current heading, they would have different current velocity and desired natural period. This
explains those different settings for 0 degree and 45 degree decay tests in table 5.3.
For the sake of comparison and verification, each test is performed on three grid with different
sizes. The properties of these grids are shown below, in table 5.4. The finest grid is the chosen mesh
that mentioned in section 6.

44 of 118

Table 5.3: Decay tests set-up


Category
Mass of the entire floater
Moment of inertia
Current velocity
Desired sway natural period
Desired yaw natural period
Reference time
Time step length

0 degree decay 45 degree decay


289.902
172.541
0.188
0.139
8.630
15.777
4.665
8.528
1.438
2.630
0.024
0.044

Unit
kg
kgm
m/s
s
s
s
s

Table 5.4: Grid properties for decay tests


Grid property
Number of cells
Minimum orthogonality degree
Maximum skewness
Negative, concave or twisted cells

26956
49.89
0.586
None

Value
59740
49.94
0.587
None

221258
44.60
0.632
None

Last but not the least, since we are only interested in hydrodynamic damping, it is wise to set
the damping coefficient of the mooring system within the DOF we are coping, to a very small
value ( not 0, or it may cause some trouble for CFD solvers!). This could be explained through a
1 DOF dynamic system. In decay test, the equation of motion of the floater can be described as:
M y + B y + Ky = Fhydro . As explained in equation 2.2 in section 2.1, hydrodynamic force exerted
on the floater has several components. In a decay test it only contains hydrodynamic added mass
and hydrodynamic damping. Hence it can be rewritten in the form as follow:
(M + madd )
y + (B + bhydro )y + Ky = 0

(5.1)

Hence, the hydrodynamic damping can be calculated through decay analysis directly with a zero
structural damping, or instead a very small one.
After all the decay tests, the calculated stiffness coefficient can be taken as an input for a new
decay test acting as a verification.
Here is the process for decay tests:
1. Assume a value for the stiffness of entire system
2. Perform a decay test with assumed stiffness
3. Extract the hydrodynamic added mass from decay test results
4. Determine the stiffness value that can achieve the desired natural frequency of the system
5. Perform a new decay test with the calculated stiffness for the purpose of verification
5.1.2

Results and analysis

The start point for a decay test is an initial displacement/rotation angle. Due to the presence of a
mooring system, as the source of restoring force, a periodically harmonic motion with a degenerative
amplitude can be observed.
A sketch of sway and yaw 45 degree decay tests is shown in figure 5.1. It depicts the pressure
distribution of the flow field.
Here in figure 5.2 the cross-line motion trace, velocity, acceleration and total force of 45 degree
sway decay are shown in detail. Meanwhile the rotation trace, velocity, acceleration and moment of
45 degree yaw decay are shown in figure 5.3.

45 of 118

Figure 5.1: Pressure distribution of the flow field in 45 degree decay tests for sway (left) and yaw
(right) on the finest grid with 221k cells
There are some strange distortions in the first few time steps in velocity, acceleration force/moment
plots of both 45 degree sway and yaw decay, which can be observed in the figures above. However,
when the initial excitation, in this case the starting displacement and rotation angle, is larger than
the floater dimension, such distortions will disappear. It is a bug in ReFRESCO. Luckily, this will
not cause any negative effect to our results, since the distortions will be filtered during the postprocessing.
According to the results from grid size and time step inspection, the global convergence tolerance
should equal or smaller than 104 . All the residuals in each decay test are well controlled under this
criteria.
The decay analysis can be carried out based on the motion trace data above. Through Fourier
analysis, the motion signal can be transferred to the frequency domain. Then the dominant frequency
can be obtained. With this natural frequency of the current system, the hydrodynamic added mass
can be calculated, as shown in section 7.1.1. So it is with the restoring coefficient to achieve our
desired natural period of the whole floater. Added mass and desired stiffness coefficient for both 0
and 45 degree sway and yaw decay tests on three girds with different properties are shown in table
5.5. In this table, it can be noticed that only the added mass of 45 degree sway decay differs a bit
much, compared with other calculated values. After further verification, the conclusion is sets of
calculated values on the finest mesh can provide periods of the system much closer to our desired
natural periods, with deviations smaller than 1%. Hence those values are chosen as the final results.
Table 5.5: Added mass and desired stiffness coefficient for both 0 and 45 degree sway and yaw decay
tests on three girds
Angle of attack

Decay test type


Sway

0 degree
Yaw
Sway
45 degree
Yaw

Grid
Added mass
Desired stiffness
Added mass
Desired stiffness
Added mass
Desired stiffness
Added mass
Desired stiffness

27k
349.069
338.690
177.535
635.007
341.351
100.118
177.067
189.773

60k
358.816
343.856
178.720
637.227
372.898
105.122
183.738
193.394

221k
355.943
342.333
178.337
636.532
364.371
103.769
185.380
194.285

Unit
kg
N/m
kgm
Nm/rad
kg
N/m
kgm
Nm/rad

Besides, with motion data the PQ analysis can be performed, after which the damping of the
system can be calculated. The fitting process of the damping analysis through PQ analysis for 45
degree sway and yaw decay on the finest grid is presented, in figure 5.4. At the top of the figure are
the PQ diagrams. Three methods are used for fitting and all of them provide reliable results, from
both visualization and error estimation point of views.

46 of 118

(a) Cross-line motion

(b) Cross-line velocity

(c) Cross-line acceleration

(d) Cross-line total force

Figure 5.2: Cross-line motion, velocity, acceleration and total force of the 45 degree sway decay on
the finest grid with 221k cells
With chosen added mass coefficient on the finest mesh, we can calculate critical damping and
the linear damping ratio, which can provide some insight for further inspection of the effect from
damping ratio on the floaters VIM behaviour. A summary for decay tests results are shown in table
5.6, for both 0 and 45 degree sway and yaw decay tests. These parameters will be used in further
CFD VIM tests.
Table 5.6: A summary of decay tests results
Angle of attacks
Decay types
Added mass
Stiffness coefficient
Critical damping
Linear damping coefficient
Quadratic damping coefficient
Linear damping ratio
Natural period
Desired natural period
Error

0 Degree
Sway
Yaw
355.943
178.337
342.333
636.532
940.413
945.188
301.253
292.805
467603.137 4232.304
0.320
0.310
8.659
4.670
8.630
4.665
0.33%
0.11%

47 of 118

45 degree
Sway
Yaw
364.371
185.380
103.769
194.285
521.127
527.404
118.827
94.979
674708.714 2142.129
0.228
0.180
15.684
8.548
15.777
8.528
0.59%
0.23%

Unit
kg or kgm
N/m or N m/rad
N s/m or N ms/rad
N s/m or N ms/rad
N s2 /m2 or N ms2 /rad2
s
s

(a) Yaw motion

(b) Yaw velocity

(c) Yaw acceleration

(d) Moment

Figure 5.3: Yaw motion, velocity, acceleration and total force of the 45 degree yaw decay on the
finest grid with 221k cells

48 of 118

(a) Sway

(b) Yaw

Figure 5.4: Plots of PQ analysis for damping coefficients of 45 degree sway and yaw decay on the
finest grid with 221k cells

49 of 118

5.2

CFD VIM tests set-up

With the stiffness obtained from decay test and all the other numerical configurations selected in
previous sections, now the CFD calculations for VIM could be setted up. As described in the section
theoretical background and literature review, the floater tends to experience the most intensive VIM
phenomena with a diamond deployment, empirically. Thus in this study CFD VIM calculations will
be performed with a 45 degree current heading. The structural property of the floater tested in this
case is the same with the one that used in decay test. Hence its property could still be retrieved in
table 5.3, which is still defined by scaling factor in table 5.2.
During the VIM tests, current with a certain velocity will come with a 45 degree current heading,
regarding to the initial deployment of the floater. The floater has mooring system simulated by the
stiffness calculated in decay tests and is free to move. Unless declared in advance, the damping of
the mooring system is setted to be unit, which indicates that no external damping is specified. Then
the CFD VIM calculations could be activated.
Here is a close look at the CFD VIM results, as an example. In figures 5.5, 5.6, 5.7 and 5.8the
results of 45 degree VIM tests with Re = 44453 are given. In these figures the field of pressure
coefficient are provided with force coefficients and dimensionless motion diagrams. In this research
all the CFD VIM calculations have almost the same deployments, unless indicated.

Figure 5.5: Plots of results from 45 degree VIM test with Re = 44453 at Simulation T ime = 88s

50 of 118

Figure 5.6: Plots of results from 45 degree VIM test with Re = 44453 at Simulation T ime = 264s

Figure 5.7: Plots of results from 45 degree VIM test with Re = 44453 at Simulation T ime = 440s

51 of 118

Figure 5.8: Plots of results from 45 degree VIM test with Re = 44453 at Simulation T ime = 616s

52 of 118

Estimation of 3D VIM response from 2D CFD calculations

In this section 2D CFD calculations with varying reduced velocity and external damping are
performed. A direct comparison between results from 2D CFD calculations and those from 3D CFD
computations is available then. The investigation is favourable for the illustration of the discrepancy
between 2D and 3D CFD VIM calculations. At last, the feasibility analysis of 3D VIM response
estimation from 2D CFD calculations is provided.

6.1

Tests with varying reduced velocity and external damping in 2D

Reduced velocity is a dimensionless parameter that corresponds with current velocity. Its definition is given in section 2.2. External damping is a parameter with dimension, which plays a role
in equation of motion. It should be stressed that external damping here is neither hydrodynamic
damping nor structural damping from the columns, but actually an external damping directly exerted
on the system. It functions like any other kinds of damping in terms of the dissipation of energy.
Reduced velocity and external damping are two main influential factors in the VIM phenomenon.
According to the definition of VIM, it is a self-excited and self-limited behaviour. This indicates the
energy balance of VIM, where the input energy is equal to output energy. Since reduced velocity is
the source of input energy and external damping is responsible for the dissipation of energy, these
two parameter are worth further investigation.
6.1.1

Test configuration

The general configuration for 2D VIM CFD simulations regarding a check for reduced velocity
and external damping is identical with that illustrated in section 5. The only difference lies in current
velocity and external damping added to the entire system. Ten different reduced velocities and 4
different external damping levels are checked here, which are shown in table 6.1 and 6.2. The range
of reduced velocity is determined in scope of that in 3D CFD tests and the set-up of real experiments
taken place in MARIN before. It covers the lock-in range observed in 3D cases and experiments.
Similarly, the range of external damping is determined according to the damping contributed from
the pontoon in 3D CFD simulations. It is shown in percentage with respect to critical damping,
which is calculated in decay tests (see section 5.1).
Table 6.1: Tests set-up for inspection of reduced velocity
Reduced velocity (Ur)
3
4
5
6
7
8
9
10
12
16

6.1.2

Current velocity (Uref) (m/s)


0.070
0.093
0.116
0.140
0.163
0.186
0.209
0.233
0.279
0.372

Reference time (Tref) (s)


5.228
3.921
3.137
2.614
2.241
1.961
1.743
1.568
1.307
0.980

Flow variables

The flow variables calculated from each test with respect to different reduced velocities and
external damping levels are listed below. Because the cross-line motion is the dominant factor for
riser fatigue issues, thus the main focus of the flow variables lies in the cross-flow direction.

53 of 118

Table 6.2: Tests set-up for inspection of external damping levels


Damping level (Ur)
Critical damping
0%
5%
10%
15%

Translation (Ns/m)
521.127
0.000
26.056
52.113
78.169

Rotation (Nms/rad)
527.404
0.000
26.370
52.740
79.111

The root mean square of dimensionless cross-line motion amplitude for each test is shown in
table 6.3.
Table 6.3: The RMS A/D in each case with different reduced velocity and external damping
Reduced velocity
Ur
3
4
5
6
7
8
9
10
12
16

B = 0%
0.189
0.260
0.307
0.330
0.392
0.431
0.437
0.454
0.499
0.581

External damping level


B = 5% B = 10% B = 15%
0.065
0.050
0.037
0.232
0.224
0.159
0.279
0.258
0.247
0.315
0.286
0.270
0.356
0.330
0.311
0.414
0.345
0.344
0.384
0.408
0.372
0.457
0.406
0.391
0.435
0.410
0.423
0.603
0.551
0.512

Meanwhile, the root mean square of yaw motion amplitude for each test is shown in table 6.4.
Table 6.4: The RMS Yaw in each case with different reduced velocity and external damping
Reduced velocity
Ur
3
4
5
6
7
8
9
10
12
16

B = 0%
0.004
0.008
0.007
0.021
0.031
0.043
0.047
0.064
0.077
0.115

External damping level


B = 5% B = 10% B = 15%
0.004
0.004
0.004
0.007
0.008
0.008
0.010
0.014
0.013
0.022
0.020
0.019
0.031
0.025
0.025
0.040
0.032
0.031
0.044
0.040
0.037
0.055
0.054
0.043
0.071
0.064
0.060
0.097
0.089
0.081

The lift force coefficient is calculated according to its definition in section 2.2. The result is
shown in table 6.5, which gives the root mean square of lift force coefficients in the tests with
different reduce velocity and external damping.
Based on the information of force and displacement in each time step, the work-done by the
entire floater can be calculated then. The result is shown in table 6.6. An emphasis should be placed
here that the values in this table are dimensional, which have the dimension of energy Joule.
Through Fourier transform, the characteristic cross-line motion frequency could be extracted
from the time trace of cross-line motion. The result is given in table 6.7, within the form of characteristic motion frequency over natural frequency of sway motion.

54 of 118

Table 6.5: The RMS Clift in each case with different reduced velocity and external damping
Reduced velocity
Ur
3
4
5
6
7
8
9
10
12
16

External damping level


B = 0% B = 5% B = 10% B = 15%
1.676
0.649
0.526
0.425
1.092
0.983
1.022
0.908
0.586
0.570
0.688
0.777
0.437
0.477
0.509
0.559
0.449
0.390
0.431
0.463
0.406
0.422
0.376
0.418
0.342
0.351
0.361
0.364
0.336
0.357
0.332
0.341
0.350
0.289
0.301
0.332
0.343
0.355
0.348
0.351

Table 6.6: The work-done (Joule) by the entire floater in each case with different reduced velocity
and external damping
Reduced velocity
Ur
3
4
5
6
7
8
9
10
12
16

B = 0%
0.555
0.797
0.400
0.661
0.641
-2.388
1.296
-0.155
1.410
0.514

External damping level


B = 5% B = 10%
1.140
0.975
17.410
30.724
32.032
47.185
41.642
63.521
54.931
88.533
70.705
105.001
63.674
159.690
99.910
174.801
125.305 219.219
206.506 376.248

B = 15%
0.594
16.843
59.959
80.124
117.646
148.287
207.805
250.229
326.138
568.528

Table 6.7: The characteristic cross-line motion frequency over natural frequency of sway motion in
each case with different reduced velocity and external damping
Reduced velocity
Ur
3
4
5
6
7
8
9
10
12
16

B = 0%
0.894
1.019
1.208
1.302
1.270
1.333
1.474
1.537
1.035
0.549

External damping level


B = 5% B = 10% B = 15%
0.957
0.957
0.518
1.051
1.019
0.941
1.208
1.145
1.082
1.255
1.223
1.161
1.302
1.208
1.176
1.255
1.365
1.270
1.380
1.349
1.412
1.443
1.506
1.663
1.819
1.835
1.694
0.251
0.063
0.031

For an illustrative purpose, the data above is reformed into figures below.
The root mean square of dimensionless cross-line motion amplitude is provided in figure 6.1.
The root mean square of yaw motion amplitude for each test is shown in figure 6.2.
The root mean square of lift force coefficient for each test is provided in figure 6.3.
The work-done (Joule) by the entire floater is shown in figure 6.4.

55 of 118

Figure 6.1: The RMS A/D in each case with different reduced velocity and external damping

Figure 6.2: The RMS Yaw in each case with different reduced velocity and external damping
The characteristic cross-line motion frequency over natural frequency of sway motion is given
in figure 6.5.
With results from Fourier transform, the power spectrum density plots for cross-line motion with
aspect to frequency and reduced velocity could be obtained. In terms of different external damping
levels, the plots are shown in figure 6.6-6.9.
Based on the information above, conclusions can be drawn for the 2D CFD VIM of semisubmersibles, regarding the entire floater.
First and foremost, figure 6.1 shows the relation between cross-line motion amplitude and reduced velocity, as well as the external damping. With increasing reduced velocity, the motion amplitude keeps increasing. Whereas, with the increment of external damping level, the motion amplitude
decreases. There is no peak in this figure, which indicates the lock-in phenomenon is not observed.
Moreover, it also manifests that there is no resonance effect captured. The result is different from
our anticipation in 3D case. In 3D cases, both for the model tests and 3D CFD calculations, the
lock-in range can always be identified, with the figure known as bell curve. Figure 3.5 shows a
demonstration for the general knowledge towards VIM of semi-submersibles in 3D, which depicts

56 of 118

Figure 6.3: The RMS Clift in each case with different reduced velocity and external damping

Figure 6.4: The work-done (Joule) by the entire floater in each case with different reduced velocity
and external damping
a bell shape curve. All in all, in 2D CFD VIM calculations there is no peak in the plot of cross-line
motion amplitude against reduced velocity.
The results shown in both the work-done plot and power spectrum density plots are corresponding to the conclusion we made above. Both plots tell a story that energy increases with reduced
velocity. In figure 6.4, the plot of work-done against reduced velocity, it is always single peak when
the reduced velocity is smaller than 10. While, for high reduced velocity, the figure becomes messy
and several peaks could be distinguished. As a result, the characteristic motion frequency for cases
with reduced velocity larger than 10 is not easy to determine. Hence in figure 6.5, the results for
reduced velocity larger than 10 are not plausible.
Secondly, external damping plays a role for the reduction of motion amplitude. The dissipation
of energy is not that considerable, though.
Last but not the least, the plot of lift force coefficient against reduced velocity conveys precious
information. In figure 6.3, it can be observed that the lift force coefficient is quite stable with
increasing reduced velocity. However, figure 6.4 and 6.1 imply that the input energy of the entire

57 of 118

Figure 6.5: The characteristic cross-line motion frequency over natural frequency of sway motion in
each case with different reduced velocity and external damping

Figure 6.6: The power spectrum density plots for cross-line motion with aspect to frequency and
reduced velocity for external damping B=0%
system keeps growing. The force magnitude and phase between force and displacement are the only
variables responsible for the growth of input energy. Since there is no force magnitude increment
observed, then it should be the phase between force and displacement accounts for the growth of
input energy, as well as for the increasing work-done and motion amplitude. Now it is obvious that
the phase difference is the key to the problem. In a conclusion, it is not the amplitude of the force,
but the phase between force and motion that is responsible for the increment of motion amplitude.

58 of 118

Figure 6.7: The power spectrum density plots for cross-line motion with aspect to frequency and
reduced velocity for external damping B=5%

Figure 6.8: The power spectrum density plots for cross-line motion with aspect to frequency and
reduced velocity for external damping B=10%

6.2

Discrepancy between 2D and 3D CFD VIM calculations


Due to the limitation of 2D modelling, there is discrepancy between 2D and 3D CFD calcu-

59 of 118

Figure 6.9: The power spectrum density plots for cross-line motion with aspect to frequency and
reduced velocity for external damping B=15%
lations. The modelling difference is shown in figure 6.10. In 2D modelling, it is not possible to
model the columns and pontoon simultaneously since it will change the effective length of the entire
structure and create a enclosed water pool inside. Besides, in a planar world, inevitably we will lose
the information regarding the third dimension, height. Thus in 2D CFD calculations we are actually
testing columns with infinite length. However, in section 3.1.5 there is evidence indicating that the
aspect ratio, which is the ratio between the height and effective length, plays a part in VIM. As a
result, there would be discrepancy induced by 2D modelling.

(a) 2D modelling

(b) 3D modelling

Figure 6.10: The discrepancy between 2D and 3D modelling


Apart from 2D modelling limitation, the absence of 3D vortex in 2D cases might also play a role
for the difference. In 2D CFD calculations, the vortex street formed behind the floater is entirely 2D.
It is generally agreed that vortex shedding is a 3D phenomenon in real life. Thus, the 2D vortex that
we captured in 2D CFD calculations could also be a source of 2D & 3D discrepancy.

60 of 118

6.2.1

Existence of bell shape curve

The discrepancy could be easily identified in cross-line motion amplitude plot against reduced
velocity. In figure 6.11, graph (a) is what we obtained through 2D CFD calculation and graph (b)
is an experimental result from Rijken and Leverette [42]. In their experiment, a semi-submersible
with four square round columns were tested. It is quite similar to the floater tested in this study.
After a qualitative comparison, obviously in real experiment we could have a bell curve, while in
2D CFD calculation there is no trend for a drop of amplitude with increasing reduced velocity. In a
conclusion, the behaviour of the floater captured in 3D is different from that in 2D.

(b) Experimental result

(a) 2D CFD calculation result

Figure 6.11: A comparison of dimensionless cross-flow motion amplitude between 2D CFD calculation and experiment result
However, according to open literature, there could be cases that the bell curve is not existing.
The experiment carried out by Goncalves et al [19] yielded no bell curve at all. They have tested
a single square column with rounded corner. The result is provided in figure 6.12.

Figure 6.12
In brief, since VIM is a complex fluid-structure interaction behaviour, it has lots of factors involved in this process and thus quite difficult to predict the existence of bell curve. More experiments and CFD practices are in need to come to a conclusion regarding the discrepancy between 2D
and 3D.
6.2.2

Comparison of flow variables

A comparison between the flow variables calculated from 2D and 3D CFD simulations are listed
in table 6.8, with reduced velocity Ur=6. Since the equivalent damping from pontoon in 3D case is
7% of critical damping, in 2D case an external damping B=7% is applied to the system.
The cross-line flow variables(RMS A/D and RMS Clift) are almost identical in both 2D and 3D.
While for others the deviations are large. Thus, if ones only look for cross-flow motion, then even
2D CFD calculation could provide some insight for VIM performance of the structure. Otherwise,

61 of 118

Table 6.8: A comparison between the flow variables calculated from 2D and 3D CFD simulations
with reduced velocity Ur=6.
Cases with Ur=6
2D case with B=7%
3D case
Difference

RMS A/D
0.301
0.305
1.48%

Aveg X/D
0.665
0.444
49.77%

Flow variables
RMS Yaw RMS Clift
0.020
0.517
0.009
0.504
116.67%
2.48%

Aveg Cdrag
2.351
1.674
40.44%

Shedding freq.
0.076
0.058
30.17%

3D CFD calculations are more reliable. However, this is only our reasoning based on one case. More
experiments are in need as verification.

6.3

Feasibility of 3D VIM response estimation from 2D CFD calculations

Acting as the source of energy dissipation, the presence of pontoon will change the behavior of
columns during VIM, compared with floaters without pontoon. External damping can also change
the behavior of columns during VIM. Its function is similar to pontoon.
The presence of pontoon and external damping have almost the same function, which creates
the possibility of estimation 3D from 2D simulation. However, in terms of the absence of bell
shape curve in 2D, inevitably there is discrepancy between 2D and 3D CFD calculation results. A
good point to emphasize is that the flow variables in cross-flow direction obtained from 2D CFD
calculations are quite close to those in 3D.
All in all, it is feasible to estimate 3D VIM response of a multi-column floater from 2D CFD
computation with the objective to gain an insight about the nominal transverse flow-variables (e.g.
cross-flow motion amplitude, lift coefficient.). Otherwise the estimation is not reliable. For the
purpose of 2D CFD calculation is always a quick prediction in real engineering practice, a rough
estimation would be enough for an initial look. Thus during early design stage a 2D CFD calculation
could be very useful.
In order to estimate 3D response more accurately, here are the procedures to increase the reliability of the estimation from 2D CFD calculations:
1. Run CFD VIM simulations for multi-column floaters in 2D without pontoon;
2. Take the motion trail obtained in first step as defined time trace for forced oscillation of the
corresponding 2D pontoon and calculate the damping level;
3. Add damping level obtained from the forced oscillation of pontoon as the external damping
for the multi-column floater and rerun the VIM simulation. Iterate from step 1-3 until the
values of damping from pontoon and motion of the multi-column floater are stable;

62 of 118

Research on VIM behaviour of multi-column floaters in 2D

Lots of parameters would play parts in the VIM behaviour of multi-column floaters, which attributes to the complexity of research on VIM. In order to look into the mechanism behind VIM in
2D, in this section different parameters that have influence on VIM will be discussed. Besides, the
behaviour of each column is also investigated here.

7.1

Behaviour of each column

Apart from the entire floater, the behaviour of each column is also helpful for the comprehension
of the VIM mechanism. In order to have an explicit elaboration, the numeration of columns is shown
in figure 7.1. Column stays in up-stream is called No.3. Column sits in down-stream is named No.1.
Side columns are nominated as No.2 and No.4. The results shown in the following tables and figures
will all follow this naming rule.

Figure 7.1: Deployment and numeration of each column for the floater in 2D
The work-done by each column in each case with different reduced velocity and external damping are shown in tables 7.1 - 7.4. In a contrast, work-done by the entire floater is also attached in
those tables.
From the tables above it could be noticed that the work-done of each column could be either
positive or negative, indicating they either drive or damp the motion of the system. In detail, the
behaviour of each column could be further illustrated through their function at each time step and
a statistic over the entire time domain. In table 7.5 - 7.8, the percentage of time that each column
functions as damper are shown. When the value is over 50%, it means the column is damping
dominant, otherwise the column is driving dominant.
The root mean square of lift force coefficient for each column in each case with different reduced
velocity and external damping is provided in table 7.9 - 7.12. In a contrast, the RMS lift force
coefficient of the entire floater is also attached in those tables.
With the data collected in all the tables above, now it is possible to put them in an elucidative
form. The work-done by each column in each case with different reduced velocity and external
damping are reformed in figure 7.2. In a contrast, work-done by the entire floater is also present in
those figures.
The percentage of time that each column functions as damper in each case with different reduced
velocity and external damping are illustrated in figure 7.3.

63 of 118

Table 7.1: The work-done by each column in each case with different reduced velocity when the
external damping is 0%
External damping B=0%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
-4.405
3.231
-8.113
-31.891
-40.538
-30.026
-19.916
-54.822
-66.795
-102.022

Work-done over time (Joule)


Column No.2 Column No.3 Column No.4
1.754
1.217
1.989
1.766
-4.722
0.522
12.138
-16.112
12.487
18.044
-5.620
20.127
17.735
-12.316
35.759
22.491
-30.736
35.884
16.959
-27.664
31.918
47.467
-54.074
61.274
107.879
-80.733
41.058
133.031
-135.600
105.104

Floater
0.555
0.797
0.400
0.661
0.641
-2.388
1.296
-0.155
1.410
0.514

Table 7.2: The work-done by each column in each case with different reduced velocity when the
external damping is 5%
External damping B=5%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
0.098
7.596
4.856
-22.718
-24.562
-14.287
-6.743
-13.359
9.094
-98.229

Work-done over time (Joule)


Column No.2 Column No.3 Column No.4
0.525
0.043
0.475
2.460
0.286
7.067
14.804
-4.797
17.170
23.854
16.181
24.324
41.443
4.641
33.409
14.038
-8.887
79.840
59.960
-10.643
21.100
64.174
-21.665
70.759
28.111
-2.787
90.887
167.965
-98.439
235.209

Floater
1.140
17.410
32.032
41.642
54.931
70.705
63.674
99.910
125.305
206.506

Table 7.3: The work-done by each column in each case with different reduced velocity when the
external damping is 10%
External damping B=10%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
0.080
14.746
7.821
3.819
1.831
7.972
14.314
6.651
76.189
2.945

Work-done over time (Joule)


Column No.2 Column No.3 Column No.4
0.426
0.010
0.459
8.291
2.392
5.296
14.175
11.969
13.221
22.569
13.965
23.167
36.414
13.762
36.525
49.792
19.102
28.135
80.784
7.400
57.193
107.446
2.353
58.351
97.446
-15.662
61.246
163.582
-15.499
225.220

Floater
0.975
30.724
47.185
63.521
88.533
105.001
159.690
174.801
219.219
376.248

The RMS lift force coefficient for each column in each case with different reduced velocity and
external damping is provided in figure 7.4.
Based on all these tables and figures, elucidation about the behaviour of each columns could
be made. In figure 7.2 (a), when external damping B=0%, the upstream column 3 and downstream

64 of 118

Table 7.4: The work-done by each column in each case with different reduced velocity when the
external damping is 15%
External damping B=15%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Work-done over time (Joule)


Column No.2 Column No.3 Column No.4
0.324
-0.082
0.438
5.706
0.055
5.241
14.422
14.383
19.496
25.479
20.129
35.261
41.478
22.951
52.022
37.783
23.029
60.879
84.215
23.375
80.963
112.111
26.233
72.817
143.440
25.252
130.105
192.974
124.879
242.021

Column No.1
-0.086
5.841
11.659
-0.744
1.195
26.596
19.251
39.068
27.342
8.653

Floater
0.594
16.843
59.959
80.124
117.646
148.287
207.805
250.229
326.138
568.528

Table 7.5: The percentage of time that each column functions as damper in each case with different
reduced velocity when the external damping is 0%
External damping B=0%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
59.10%
47.69%
55.35%
64.47%
57.30%
53.30%
50.42%
56.70%
54.13%
53.86%

Percentage of time as damper


Column No.2 Column No.3
47.37%
48.13%
47.68%
59.20%
49.47%
67.53%
39.83%
52.23%
41.62%
50.71%
44.40%
53.24%
48.09%
51.93%
43.28%
52.80%
40.80%
53.26%
43.98%
49.05%

Column No.4
44.53%
49.10%
50.83%
41.06%
42.12%
40.78%
43.61%
42.39%
46.39%
46.50%

Table 7.6: The percentage of time that each column functions as damper in each case with different
reduced velocity when the external damping is 5%
External damping B=5%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
56.60%
40.03%
46.10%
60.79%
57.17%
51.69%
48.73%
49.90%
48.83%
53.83%

Percentage of time as damper


Column No.2 Column No.3
43.25%
49.85%
49.10%
49.46%
50.73%
57.05%
37.21%
33.63%
37.56%
44.76%
45.91%
48.07%
40.10%
50.31%
40.96%
49.07%
46.81%
47.15%
45.40%
49.86%

Column No.4
43.68%
36.79%
51.22%
36.64%
39.80%
35.03%
45.03%
41.43%
43.80%
43.87%

column 1 always contribute negative work-done to the system, whereas side columns 2 & 4 keep
driving the floater. The work-done by each column, either positive or negative, their magnitudes
all increase with the growth of reduced velocity. However, when external damping is added to
the system, the upstream and downstream columns 1 & 3 merely behave themselves in an entirely

65 of 118

Table 7.7: The percentage of time that each column functions as damper in each case with different
reduced velocity when the external damping is 10%
External damping B=10%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
54.42%
26.45%
37.92%
48.88%
47.62%
46.68%
44.88%
48.06%
45.47%
49.79%

Percentage of time as damper


Column No.2 Column No.3
38.37%
54.42%
35.83%
46.03%
39.38%
34.10%
35.18%
35.15%
37.97%
43.22%
41.57%
39.48%
39.12%
43.12%
41.31%
46.87%
43.90%
48.93%
44.32%
47.67%

Column No.4
41.53%
42.65%
42.61%
37.48%
39.06%
42.69%
41.71%
43.80%
44.02%
41.57%

Table 7.8: The percentage of time that each column functions as damper in each case with different
reduced velocity when the external damping is 15%
External damping B=15%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
56.67%
39.26%
35.92%
49.95%
48.99%
40.61%
46.22%
42.71%
47.90%
48.70%

Percentage of time as damper


Column No.2 Column No.3
37.88%
53.15%
36.44%
50.38%
40.21%
35.98%
30.72%
35.06%
36.95%
37.76%
41.58%
41.62%
36.29%
42.70%
38.56%
42.74%
40.35%
42.96%
43.65%
42.65%

Column No.4
34.88%
36.51%
34.44%
31.57%
28.65%
35.53%
37.09%
41.32%
40.12%
42.58%

Table 7.9: The RMS lift coefficient for each column in each case with different reduced velocity
when the external damping is 0%
External damping B=0%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
0.430
0.316
0.213
0.252
0.257
0.253
0.248
0.268
0.256
0.235

RMS lift force coefficient


Column No.2 Column No.3 Column No.4
0.459
0.443
0.458
0.310
0.264
0.368
0.271
0.158
0.282
0.281
0.160
0.299
0.286
0.214
0.328
0.342
0.248
0.324
0.317
0.254
0.309
0.303
0.267
0.351
0.336
0.277
0.351
0.334
0.258
0.307

Floater
1.676
1.092
0.586
0.437
0.449
0.406
0.342
0.336
0.350
0.343

different way. The switch of their behaviour could be observed in 7.2 (b) (c) (d). With the increment
of external damping, columns 1 & 3 are more and more driving dominant rather than dampers.
When external damping is exerted on the entire system, it would be the main player for damping
components. In order to keep the system moving, columns 1 & 3 switch from damping component

66 of 118

Table 7.10: The RMS lift coefficient for each column in each case with different reduced velocity
when the external damping is 5%
External damping B=5%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
0.266
0.290
0.182
0.249
0.257
0.247
0.256
0.261
0.233
0.250

RMS lift force coefficient


Column No.2 Column No.3 Column No.4
0.304
0.223
0.285
0.274
0.282
0.278
0.316
0.165
0.299
0.291
0.164
0.288
0.291
0.197
0.306
0.314
0.243
0.305
0.311
0.245
0.347
0.349
0.258
0.316
0.305
0.252
0.300
0.313
0.270
0.352

Floater
0.649
0.983
0.570
0.477
0.390
0.422
0.351
0.357
0.289
0.355

Table 7.11: The RMS lift coefficient for each column in each case with different reduced velocity
when the external damping is 10%
External damping B=10%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
0.253
0.291
0.204
0.232
0.243
0.226
0.216
0.243
0.236
0.240

RMS lift force coefficient


Column No.2 Column No.3 Column No.4
0.249
0.214
0.280
0.277
0.302
0.299
0.311
0.202
0.332
0.281
0.206
0.296
0.284
0.231
0.304
0.302
0.235
0.307
0.316
0.242
0.298
0.304
0.249
0.314
0.321
0.249
0.288
0.347
0.250
0.320

Floater
0.526
1.022
0.688
0.509
0.431
0.376
0.361
0.332
0.301
0.348

Table 7.12: The RMS lift coefficient for each column in each case with different reduced velocity
when the external damping is 15%
External damping B=15%
Reduced velocity Ur
3
4
5
6
7
8
9
10
12
16

Column No.1
0.294
0.263
0.223
0.253
0.215
0.213
0.225
0.211
0.241
0.217

RMS lift force coefficient


Column No.2 Column No.3 Column No.4
0.244
0.212
0.278
0.318
0.295
0.334
0.323
0.245
0.330
0.280
0.216
0.303
0.301
0.223
0.284
0.305
0.225
0.285
0.288
0.216
0.296
0.293
0.239
0.325
0.334
0.253
0.300
0.328
0.255
0.302

Floater
0.425
0.908
0.777
0.559
0.463
0.418
0.364
0.341
0.332
0.351

to driving component. Results shown in figure 7.3 are identical with our consequence.
This phenomenon is more visible if we tell the same story shown in figure 7.2 and 7.3 in a different way. Here in figure 7.5 the behaviour of each column with respect to different reduced velocity
and external damping level is illustrated separately. Obviously, with increasing external damping and

67 of 118

(a) External damping B=0%

(b) External damping B=5%

(c) External damping B=10%

(d) External damping B=15%

Figure 7.2: The work-done by each column in each case with different reduced velocity and external
damping

(a) External damping B=0%

(b) External damping B=5%

(c) External damping B=10%

(d) External damping B=15%

Figure 7.3: The percentage of time that each column functions as damper in each case with different
reduced velocity and external damping

68 of 118

(a) External damping B=0%

(b) External damping B=5%

(c) External damping B=10%

(d) External damping B=15%

Figure 7.4: The percentage of time that each column functions as damper in each case with different
reduced velocity and external damping
reduced velocity, the energy attributed from each column keeps increasing. The behaviour switch
from column 1 & 3 also exists.

69 of 118

(a) Work-done by column No.1

(b) Work-done by column No.2

(c) Work-done by column No.3

(d) Work-done by column No.4

Figure 7.5: The work-done of each column with respect to different reduced velocity and external
damping

70 of 118

7.2

Mass ratio and excessive external damping

According to the results in section 6.1, there is no bell shape curve observed in 2D CFD VIM
calculation, which deviates from the results in 3D. A plausible explanation could be the absence
of pontoon. Though we already add external damping to the system, the pontoon also contributes
added mass to the entire system. Hence it is interesting to have a look at the influence from added
mass.
Moreover, so far even with 15% external damping, we could not obtain a drop in cross-line
motion amplitude plot with high reduced velocity. It is of great interest to know if a high external
damping will crush the increasing trend of A/D. According to the calculation in 3D CFD, the equivalent damping provided by pontoon is only 7%, with respect to critical damping. Thus it is reasonable
to put an external damping over 30% as excessive external damping.
The configurations and results of VIM tests with higher mass ratio and excessive external damping are elaborated in the following subsections.
7.2.1

Mass ratio

The estimation of added mass attributed by pontoon is based on the mass and geometry of pontoon in 3D. The added mass has 2 components, which are the water surrounding pontoon and the
water trapped in pontoon. Empirically, the water surrounding and moving together with pontoon has
a mass similar to the mass of pontoon. The mass of trapped water inside pontoon could be calculated
through its volume. Thus, the total added mass due to the presence of pontoon could be estimated
and the results are shown respectively in table 7.13.
Table 7.13: Added mass due to the presence of pontoon
Category
Mass of the entire floater
Mass of the pontoon
Added mass from water surrounding pontoon
Added mass from water trapped in pontoon

Value
289.902
167.224
167.224
175.919

Unit
kg
kg
kg
kg

Taking into account the added mass from pontoon, now the mass ratio is approximately 2. So
we will adopt it as the mass ratio for new tests.
Two tests with mass ratio m = 2.0 are complete with reduced velocity Ur=10. One with 0%
external damping and the other one with 30% damping. The results are presented in figure 7.6, in
contrast with previous results. In terms of the results, conclusively the influence from mass ratio
to cross-flow motion is neglectable. It means added mass from pontoon is not a key factor for the
absence of bell shape curve.
7.2.2

Excessive external damping

In this part, excessive external damping levels are tested, which includes cases with 50% and
70% damping under reduced velocity from Ur=4 to Ur=12. The details of tested damping levels are
listed in table 7.14. The results are shown in figure 7.7, with contrast to tests with normal external
damping levels.
Table 7.14: Excessive external damping levels
Damping level (Ur)
Critical damping
50%
70%

Translation (Ns/m)
521.127
260.563
364.789

71 of 118

Rotation (Nms/rad)
527.404
263.702
369.183

Figure 7.6: The RMS A/D in cases with mass ratio m = 2 in contrast with cases m = 1

Figure 7.7: The RMS A/D in cases with excessive external damping
According to the results in figure 7.7, external damping still actively dissipate the energy of the
entire system. However, it does not behave in a different way with high reduced velocity, compared
with its effect for cases with low reduced velocity. Thus it is not the dominant factor for the presence
of bell shape curve.

7.3

Column geometry

Since vortex induced motion is a complex fluid-structure interaction induced by fluid flow, it is
very sensitive to the disturbance in the nearby flow field. As a consequence, the geometry of the
immersed body plays an important role in VIM phenomenon. In this section four different column
geometries will be tested and compared, including the square circular columns which are tested
in previous sections. The chosen geometry here are all basic geometries adopted (either in use or
under test) in offshore industry. After the 2D CFD tests, the evaluation of VIM performance of each
column geometry would be manifested at the end of this section.

72 of 118

7.3.1

Tests configuration

The set-up for each test and decay tests results, with respect to columns with different geometry, are
shown in this subsection. Generally, the numerical settings are exactly the same with the tests for
square circular column in previous section. The only exception lies in the geometry of columns, and
so it is with the grids. Hence, for the details of numerical configuration, it can be referred in section
6.
Four sorts of columns being tested here are:
Round: Round cylinder
Pentagon: Pentagon columns
Sharp: Square columns with sharp corner
Square: Square circular columns
The effective length of these four columns with different geometries are exactly the same, which
is equal to 0.365m. The geometries of columns are shown in figure 7.8.

(a) Round cylinder

(b) Pentagon columns

(c) Square columns with sharp corner

(d) Square circular columns

Figure 7.8: Columns geometries of four different floater under testing


With known geometries of the columns, now it is possible to calculate the structural property of
each floater based on their dimensions. The scaling factor has been given in table 5.2. As a result,
the structural property for each floater is provided in table 7.15.
This time the meshing process is still carried out in Hexpress, as what we did in section 4. The
properties of computational domain is identical with the set-up described in section 4.2, in terms of
domain size, boundary conditions, refinement area and refinement ratio. An overview of each grid
is provided in figure 7.9.
Here in table 7.16 is the summary of the property of each grid. Obviously the property of each
grid is quite similar to each other, making the mesh not a dominant variable for the difference in
VIM behaviour of four floaters with different column geometry. Zoom in details of each column

73 of 118

Table 7.15: The structural property for 4 floaters with different column geometry
Structural property
Mass of each column
Mass in total
Moment of Inertia

Round
104.566
418.262
250.898

Floater
Pentagon
Sharp
63.724
72.827
254.895 291.308
146.443 173.430

Unit
Square
72.475
289.902
172.541

kg
kg
kgm2

(a) Round cylinder

(b) Pentagon columns

(c) Square columns with sharp corner

(d) Square circular columns

Figure 7.9: An overview of the grids for 4 floaters with different column geometry
Table 7.16: The grid property for 4 floaters with different column geometry
Grid property
Number of cells
Minimum orthogonality degree
Maximum skewness
Negative, concave, twisted cells

Round
209962
58.33
0.447
None

Floater
Pentagon
Sharp
224938
221498
48.66
46.63
0.633
0.620
None
None

Square
221258
44.61
0.632
None

with different geometry are illustrated in figure 7.10. The difference between square columns with
sharp corner and square circular columns is now quite explicit.
In order to determine the stiffness in need to achieve a certain natural period, decay test should
be carried respectively for each floater with different column shape. The procedure and initial set-up
are entirely identical with the one that elaborated in section 5.1. In scope of our need for structural
property within 45 degree deployment, thus all the decay tests here are all in 45 degree with aspect
to current direction. The decay tests results for four floaters with different column geometry are
concluded in table 7.17. In this table it can be observed that the added mass is dependent on the

74 of 118

(a) Round cylinder

(b) Pentagon columns

(c) Square columns with sharp corner

(d) Square circular columns

Figure 7.10: An overview of the grids for 4 floaters with different column geometry
shape of body.
Table 7.17: The decay tests results for four floaters with different column geometry
Cases

Structural property

Sway decay test

Added mass
Desired stiffness
Critical damping

Round
428.586
134.312
674.513

Yaw decay test

Added mass
Desired stiffness
Critical damping

215.241
253.028
686.866

Floaters
Pentagon
Sharp
296.254 384.676
87.413
107.213
438.989 538.420

Square
364.371
103.769
521.127

kg
N/m
N s/m

125.097
147.396
400.119

185.380
194.285
527.404

kgm
N m/rad
N ms/rad

75 of 118

188.338
196.374
533.073

Unit

7.3.2

VIM performance of columns with different geometry

With a grip of structural property and grids for floaters with different column geometry, now it is
feasible to carry out VIM tests. Three reduced velocities are tested here, which are 5, 6, and 7. More
information about the actual velocity corresponding to reduced velocity could be found in section 8
and in table 6.1.
In figure 7.11 there is a brief glimpse about what happened in CFD solver. In this figure the VIM
response of floaters with different column geometry is illustrated with both flow field information
and x-y lines. In the flow field the pressure coefficient is plotted within different colour, which could
be referred by the legend. In x-y line plots the dimensionless in-line and cross-line motion amplitude
are provided respectively, in terms of simulation time.

Figure 7.11: A comparison between VIM response of floaters with different column geometry
(Ur=6)
Obviously the VIM behaviour of floaters with different column geometry are different. The
difference could be identified quantitatively through a look at flow variables. Similar to previous
sections, this time we still place emphasis on the following flow variables:
RMS A/D: Root mean square of dimensionless cross-flow motion amplitude;
Aveg X/D: Mean value of dimensionless in-flow motion amplitude;
RMS Yaw: Root mean square of yaw motion;
RMS Clift: Root mean square of lift force coefficient;
Aveg Cdrag: Mean value of drag force coefficient.
The results of those flow variables are elaborated in table 7.18 - 7.22.
The data in the tables above are collapsed into x-y lines in figure 7.12. Based on the data and the
plots, a summary could be made here as follows:
Considering surge and sway motion, floater with round columns relatively has the best performance. Compared with other sorts of floaters, it has smallest motion amplitude, for both
in-flow motion and cross-flow motion, statistically. However, when taking into account the
yaw motion, round columns are obviously the worst.

76 of 118

Table 7.18: The RMS A/D of 4 floaters with different column geometry
Reduced velocity
Ur
5
6
7

Round
0.262
0.212
0.210

Floater
Pentagon Sharp
0.330
0.300
0.323
0.362
0.373
0.401

Square
0.307
0.330
0.392

Table 7.19: The RMS X/D of 4 floaters with different column geometry
Reduced velocity
Ur
5
6
7

Round
0.241
0.350
0.489

Floater
Pentagon Sharp
0.394
0.464
0.582
0.671
0.800
0.911

Square
0.473
0.703
0.915

Table 7.20: The RMS Yaw of 4 floaters with different column geometry
Reduced velocity
Ur
5
6
7

Round
0.150
0.186
0.210

Floater
Pentagon Sharp
0.019
0.013
0.026
0.025
0.035
0.030

Square
0.007
0.021
0.031

Table 7.21: The RMS Clift of 4 floaters with different column geometry
Reduced velocity
Ur
5
6
7

Round
0.993
0.637
0.547

Floater
Pentagon Sharp
0.856
0.902
0.575
0.702
0.454
0.549

Square
0.586
0.437
0.449

Table 7.22: The RMS Cdrag of 4 floaters with different column geometry
Reduced velocity
Ur
5
6
7

Round
1.602
1.607
1.651

Floater
Pentagon Sharp
1.707
2.469
1.731
2.449
1.757
2.454

Square
2.435
2.484
2.387

The VIM performance of square columns with sharp corner and square circular columns are
quite similar. Both of them experience high motion amplitude in in-flow and cross-flow direction. But the yaw motion amplitude is small. They also have very large drag force coefficient.
The floater with pentagon columns has no advantages over others, considering the cross-flow
motion or lift force coefficient. Its in-flow direction performance is a bit better than those of
square columns either with sharp corner or circular corner.
Attention should be paid to the results that they are only valid for 45 degree cases. For 0 degree
cases, since the floater with square circular columns will experience less VIM motion[57, 42], the
comparison among different column shapes regarding VIM performance may have different results.

77 of 118

(a) RMS A/D

(b) Aveg X/D

(c) RMS Yaw

(e) Aveg Cdrag

(d) RMS Clift

Figure 7.12: An overview of the flow variables for 4 floaters with different column geometry

78 of 118

Conclusion and recommendations

In this study, the VIM performance of multi-column floaters with different column geometries
are tested under a wide range of reduced velocity and external damping levels through 2D CFD calculations. 2D CFD simulation of VIM has spectacular performance in terms of the usage of computation time and resource than 3D. Several 3D 3DOF CFD calculations of multi-column floaters have
been performed in MARIN and on average about 2 weeks per calculation are required. Compared
with 3D CFD VIM calculations, CFD simulation in 2D could be complete within 2 days and provide
comparable results. Based on the requirement for early design stage, 2D CFD VIM simulation is a
good candidate to provide an insight about the problem in designing practice.
Through 2D CFD calculation, a prediction of VIM response of four columns floater could be
made. Different parameters are tested, like external damping and reduced velocity, and their influence on VIM performance of multi-column floaters are elucidated. With increasing reduced velocity,
both the cross-flow and in-flow motion amplitude will increase. Whereas, the lift force coefficient is
quite stable with increasing reduced velocity. According to the work-done of entire system, the input
energy increases with the increment of reduced velocity. There are only two variables responsible
for the growth of input energy, force magnitude and phase between force and displacement. As a
consequence, it is not the amplitude of the force, but the phase between force and motion that is
responsible for the increment of motion amplitude.
Meanwhile, if there is external damping added to the system, then the motion amplitude will
decrease due to the energy dissipation from damping.
Other than the behaviour of entire floater, the individual behaviour of each column is also appealing. Without external damping, lateral side columns will drive the motion of floater while columns
in upstream and downstream will dissipate input energy from side columns and damp the motion
of floater. Nevertheless, the behaviour of upstream and downstream columns will change when
there is external damping added to the system. With increasing external damping levels, upstream
and downstream columns will switch from damping component to driving component. The more
external damping added, the more positive work-done will be made by upstream and downstream
columns.
The nominal dimensionless cross-flow motion amplitude could be collapse into a single line with
respect to reduced velocities. However, there is no bell shape curve observed, which deviates from
the results obtained in 3D CFD calculations and experiments. Due to the limitation of 2D modelling,
it is not possible to simulate columns and pontoon simultaneously. The absence of pontoon will
result in the loss of added mass and damping contributed by pontoon. This may account for the
absence of bell shape curve.
In order to inspect the reason for the missing bell shape curve. Tests with varying external
damping and mass ratio are performed. The application of external damping to the entire system
could lead to a drop for the VIM motion amplitude. However, it could not change the increasing
trend of lines in A/D vs. Ur plot. Apart from external damping, varying mass ratio is also tested but
it has little influence on VIM behaviour of multi-column floaters. In conclusion, damping and added
mass from pontoon are not the dominant factor for the existence of bell shape curve.
Column geometry plays an important role for the VIM performance of a floater, as well. Four
different column geometries are tested here, with 45 degree current heading. Considering surge
and sway motion, floater with round columns relatively has the best performance. But it has the
largest yaw motion. The VIM performance of square columns with sharp corner and square circular
columns are quite similar. Both of them experience high motion amplitude in in-flow and cross-flow
direction. The floater with pentagon columns has no advantages over others, considering the crossflow motion or lift force coefficient. Its in-flow direction performance is a bit better than those of
square columns either with sharp corner or circular corner.
All in all, it is feasible to estimate 3D VIM response of a multi-column floater from 2D CFD
computation with the objective to gain an insight about the nominal transverse flow-variables (e.g.
cross-flow motion amplitude, lift coefficient.). Otherwise the estimation is not reliable. For the
purpose of 2D CFD calculation is always a quick prediction in real engineering practice, a rough

79 of 118

estimation would be enough for an initial look. Thus during early design stage a 2D CFD calculation
could be very useful.
So far, plenty of efforts have been made towards the inspection of influential factors for the
VIM performance of multi-column floaters. Despite of the progress, there is still space for further
improvement. Now the reason for the absence of bell shape curve, regarding the A/D vs. Ur plot,
is still unclear. More tests and experiments are in need to answer this question. Moreover, the tests
in this study mainly focus on the VIM performance of floater with 45 degree heading current. If
possible, a wide range of current heading, from 0 degree to 45 degree, is worth our attentions. This
concern is quite important for the inspection of VIM performance of varying column geometries.
According to the application of different column geometries adopted by the offshore industry, more
column shapes should be included in the tests in future.
Finally, if a more complete comparison between tests in 2D CFD, 3D CFD and even experiments
could be made, then it might provide some strong argument about the feasibility of 2D CFD VIM
calculation in early design stage and its blueprint of application.

80 of 118

Nomenclature
Symbol
A
A
AR
B
B
Bcrit
bhydro
CA
CEA
CF
CM
c
D
D
Df ull
Dmodel
d
Fadd
Fdamping
Fhydro
f

flower
fs
fN
G
g
I
K
K
k
L
Lref
l
M
madd
m
n
q
r
rc
S
Sp
T
T
Tn

Unit
[m]

[N s/m]
[N ms/rad]
[N s/m]
[N s/m]

[m2 /s]
[m]
[m]
[m]
[m]
[N ]
[N ]
[N ]

[Hz]
[Hz]
[m]
[m/s2 ]
[N ]
[N/m]
[N m/rad]
[J]
[m]
[m]
[kg]
[kg]

[rad]
[m2 ]
[m2 ]
[m]
[N ]
[s]

Explanation
Motion Amplitude
Amplitude ratio
Aspect ratio
Surge/Sway damping matrix
Yaw damping matrix
Critical damping
Hydrodynamic damping
Added mass coefficient
Effective added mass coefficient
Force coefficient
Moment coefficient
Dimensionless damping
Strain rate tensor
Effective length
Full scale length
Model scale length
Arm of force
Hydrodynamic force caused by added mass
Hydrodynamic damping force
Hydrodynamic force
Response frequency ratio
Lower branch frequency ratio
Shedding frequency of vortex
Natural frequency of structure
Gap width between cylinders
Gravitational acceleration
Unit stress tensor
Horizontal stiffness
Rotational stiffness
Turbulent kinetic energy
Centre to centre longitudinal length
Reference length
Turbulence length scale
Mass
Added mass
Mass ratio
Normal vector
Source or sink
Radius of the column corner
Corner sharpness
Cross section area
Projected area
Centre to centre transverse spacing
Stress tensor
Period of vortex shedding

81 of 118

Symbol
TN
Tref
UR
Uref
Uf ull
Umodel
u
v
v

ij
t

Unit
[s]

[m/s]
[m/s]
[m/s]
[m/s]
[m/s]
[rad]
[rad/s]
[rad/s]
[m3 ]
[kg]
[m2 /s]
[kg/m3 ]

82 of 118

Explanation
Natural period
Reference time
Reduced velocity
Reference velocity
Full scale velocity
Model scale velocity
Velocity
Turbulence velocity scale
Velocity vector
Incidence angle
Excitation frequency
Natural frequency
Volume
Displacement
Damping ratio
Kinematic viscosity
Scaling factor
Density
Scalar
Diffusivity
Kronecker symbol
Turbulent viscosity

10

List of abbreviations
Abbreviation
CDS
CFD
EoM
FDM
FEM
FVM
Fr
GCI
MMS
QUICK
RANS
RAM
RE
RMS
Re
Sc
SEM
SG
St
UDS
URANS
VIM
VIV
V&V

Explanation
Central differencing scheme
Computational fluid dynamics
Equation of motion
Finite difference method
Finite element method
Finite volume method
Froude number
Grid convergence index
Method of the manufactured solutions
Quadratic upwind interpolation for convective kinematics
Reynolds averaged Navier-Stokes
Random access memory
Richardson extrapolation
Root mean square
Reynolds number
Scruton number
Spectral element method
Skop-Griffin parameter
Strouhal number
Upwind differencing scheme
Unsteady Reynolds averaged Navier-Stokes
Vortex induced motion
Vortex induced vibration
Validation and verification

83 of 118

References
[1] NPARC Alliance. Uncertainty and error in cfd simulations, 2012.
[2] Y Bai, A Tang, E Osullivan, KC Uppu, S Ramakrishnan, et al. Steel catenary riser fatigue
due to vortex induced spar motions. In Offshore Technology Conference. Offshore Technology
Conference, 2004.
[3] Michael M Bernitsas, Kamaldev Raghavan, and G Duchene. Induced separation and vorticity
using roughness in viv of circular cylinders at 8 103 re 2.0 105. In ASME 2008 27th International Conference on Offshore Mechanics and Arctic Engineering, pages 993999. American
Society of Mechanical Engineers, 2008.
[4] Robert D Blevins. Flow-induced vibration. 1990.
[5] D Brika and A Laneville. Vortex-induced vibrations of a long flexible circular cylinder. Journal
of Fluid Mechanics, 250:481508, 1993.
[6] Francois Chanony. Validation and verification of fresco for viscous flows around oscillating
bodies. roll motion. Masters thesis, ENSTA, Brest, Bretagne, France, 2009.
[7] Kim ES Ding L, Zhang L and Bernitsas MM. Urans vs. experiments of flow induced motions
of multiple circular cylinders with passive turbulence control.
[8] L Eca and M Hoekstra. An evaluation of verification procedures for cfd applications. In 24th
Symposium on Naval Hydrodynamics, pages 813. Fukuoka, Japan, 2002.
[9] L Eca and M Hoekstra. A verification exercise for two 2-d steady incompressible turbulent
flows. In Proceedings of the ECCOMAS, 2004.
[10] L Eca and M Hoekstra. Evaluation of numerical error estimation based on grid refinement
studies with the method of the manufactured solutions. Computers & Fluids, 38(8):15801591,
2009.
[11] L Eca, M Hoekstra, A Hay, and D Pelletier. On the construction of manufactured solutions for
one and two-equation eddy-viscosity models. International journal for numerical methods in
fluids, 54(2):119154, 2007.
[12] L Eca, GNVB Vaz, and M Hoekstra. A verification and validation exercise for the flow over a
backward facing step. In Proceedings of the ECCOMAS CFD, 2010.
[13] Luis Eca, Guilherme Vaz, and Martin Hoekstra. Code verification, solution verification and
validation in rans solvers. In ASME 2010 29th International Conference on Ocean, Offshore
and Arctic Engineering, pages 597605. American Society of Mechanical Engineers, 2010.
[14] Joel H Ferziger and Milovan Peric. Computational methods for fluid dynamics. Springer
Science & Business Media, 2012.
[15] F Flemming and CHK Williamson. Vortex-induced vibrations of a pivoted cylinder. Journal
of Fluid Mechanics, 522:215252, 2005.
[16] Andre LC Fujarra, Guilherme F Rosetti, Jaap de Wilde, and Rodolfo T Goncalves. State-of-art
on vortex-induced motion: A comprehensive survey after more than one decade of experimental investigation. In ASME 2012 31st International Conference on Ocean, Offshore and Arctic
Engineering, pages 561582. American Society of Mechanical Engineers, 2012.
[17] Matthieu Glebart. Validation and verification of fresco for loads on moving cylinders. Masters
thesis, ENSTA, Brest, Bretagne, France, 2009.

84 of 118

[18] Rodolfo T Goncalves, Andre LC Fujarra, Guilherme F Rosetti, and Kazuo Nishimoto. Mitigation of vortex-induced motion (vim) on a monocolumn platform: forces and movements.
Journal of Offshore Mechanics and Arctic Engineering, 132(4):041102, 2010.
[19] Rodolfo T Goncalves, Dennis M Gambarine, Felipe P Figueiredo, Fabio V Amorim, and
Andre LC Fujarra. Experimental study on flow-induced vibration of floating squared section cylinders with low aspect ratio: Part i effects of incidence angle. In ASME 2015 34th
International Conference on Ocean, Offshore and Arctic Engineering, pages V002T08A033
V002T08A033. American Society of Mechanical Engineers, 2015.
[20] Rodolfo T Goncalves, Guilherme F Rosetti, Andre LC Fujarra, and Allan C Oliveira. Experimental study on vortex-induced motions of a semi-submersible platform with four square
columns, part ii: Effects of surface waves, external damping and draft condition. Ocean engineering, 62:1024, 2013.
[21] R Govardhan and CHK Williamson. Modes of vortex formation and frequency response of a
freely vibrating cylinder. Journal of Fluid Mechanics, 420:85130, 2000.
[22] R Govardhan and CHK Williamson. Resonance forever: existence of a critical mass and an
infinite regime of resonance in vortex-induced vibration. Journal of Fluid Mechanics, 473:147
166, 2002.
[23] RN Govardhan and CHK Williamson. Defining the modified griffin plot in vortex-induced
vibration revealing the effect of reynolds number using controlled damping. Journal of Fluid
Mechanics, 561:147180, 2006.
[24] OM Griffin and SE Ramberg. Some recent studies of vortex shedding with application to
marine tubulars and risers. Journal of Energy Resources Technology, 104(1):213, 1982.
[25] Owen M Griffin, Richard A Skop, Steven E Ramberg, et al. The resonant, vortex-excited
vibrations of structures and cable systems. In Offshore Technology Conference. Offshore Technology Conference, 1975.
[26] Frederick Jaouen. The validation of the free-surface modelling of fresco. Masters thesis,
ENSTA, Brest, Bretagne, France, 2008.
[27] N Jauvtis and CHK Williamson. Vortex-induced vibration of a cylinder with two degrees of
freedom. Journal of Fluids and Structures, 17(7):10351042, 2003.
[28] N Jauvtis and CHK Williamson. The effect of two degrees of freedom on vortex-induced
vibration at low mass and damping. Journal of Fluid Mechanics, 509:2362, 2004.
[29] A Khalak and CHK Williamson. Motions, forces and mode transitions in vortex-induced vibrations at low mass-damping. Journal of fluids and Structures, 13(7):813851, 1999.
[30] Roger King. A review of vortex shedding research and its application. Ocean Engineering,
4(3):141171, 1977.
[31] R Ajith Kumar, Chang Hyun Sohn, and BH Lakshmana Gowda. Influence of corner radius
on the near wake structure of a transversely oscillating square cylinder. Journal of mechanical
science and technology, 23(9):23902416, 2009.
[32] Justin S Leontini and Mark C Thompson. Vortex-induced vibrations of a diamond crosssection: Sensitivity to corner sharpness. Journal of Fluids and Structures, 39:371390, 2013.
[33] Michel Make. Predicting scale effects on floating offshore wind turbines. Masters thesis,
Technical University of Delft, the Netherlands, 2014.

85 of 118

[34] Florian R Menter. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA journal, 32(8):15981605, 1994.
[35] Eduard Naudascher and Donald Rockwell. Flow-induced vibrations: an engineering guide.
Courier Corporation, 2012.
[36] Andras Nemes, Jisheng Zhao, David Lo Jacono, and John Sheridan. The interaction between
flow-induced vibration mechanisms of a square cylinder with varying angles of attack. Journal
of Fluid Mechanics, 710:102130, 2012.
[37] H-J Niemann and N Holscher. A review of recent experiments on the flow past circular cylinders. Journal of Wind Engineering and Industrial Aerodynamics, 33(1):197209, 1990.
[38] Young-Chan YC Park, Arun Antony, Hisham Moideen, Arada Jamnongpipatkul, et al.
Gulfstar-vim and mooring chain fatigue. In Offshore Technology Conference. Offshore Technology Conference, 2015.
[39] Filipe Pereira. Verification of refresco with the method of manufactured solutions. Masters
thesis, IST, Lisbon, Portugal, 2012.
[40] MARIN R&D. Theoretical and numerical formulation of fresco.
[41] Oriol Rijken. Examining the effects of scale, mass ratios and column shapes on the vortex
induced motion response of a semisubmersible through cfd analyses. In ASME 2014 33rd
International Conference on Ocean, Offshore and Arctic Engineering, pages V002T08A028
V002T08A028. American Society of Mechanical Engineers, 2014.
[42] Oriol Rijken and Steven Leverette. Experimental study into vortex induced motion response
of semi submersibles with square columns. In ASME 2008 27th International Conference on
Offshore Mechanics and Arctic Engineering, pages 263276. American Society of Mechanical
Engineers, 2008.
[43] Patrick J Roache. Verification and validation in computational science and engineering. Hermosa, 1998.
[44] Patrick J Roache. Code verification by the method of manufactured solutions. Journal of Fluids
Engineering, 124(1):410, 2002.
[45] PJ Roache. Fundamentals of verification and validation [m] hermosa publishers. Socorro, New
Mexico, USA, 2009.
[46] Guilherme F Rosetti, Guilherme Vaz, and Andre LC Fujarra. Urans calculations for smooth
circular cylinder flow in a wide range of reynolds numbers: Solution verification and validation.
Journal of Fluids Engineering, 134(12):121103, 2012.
[47] Kambiz Salari and Patrick Knupp. Code verification by the method of manufactured solutions.
Technical report, Sandia National Labs., Albuquerque, NM (US); Sandia National Labs., Livermore, CA (US), 2000.
[48] C Scruton. Wind-excited oscillations of tall stacks. The Engineer, 199(5185):806808, 1955.
[49] Ruxin Song, Basim Mekha, and Abhilash Sebastian. Independent design verification of scrs
for ultra deepwater ihf development. In 25th International Conference on Offshore Mechanics
and Arctic Engineering, pages 559566. American Society of Mechanical Engineers, 2006.
[50] D Sumner. Two circular cylinders in cross-flow: a review. Journal of Fluids and Structures,
26(6):849899, 2010.
[51] D Sumner, SJ Price, and MP Paidoussis. Flow-pattern identification for two staggered circular
cylinders in cross-flow. Journal of Fluid Mechanics, 411:263303, 2000.

86 of 118

[52] TETSURO Tamura, T Miyagi, and T Kitagishi. Numerical prediction of unsteady pressures
on a square cylinder with various corner shapes. Journal of Wind Engineering and Industrial
Aerodynamics, 74:531542, 1998.
[53] Jaime Hui Choo Tan, Allan Magee, Jang Whan Kim, Yih Jeng Teng, and NorBahrain Ahmad
Zukni. Cfd simulation for vortex induced motions of a multi-column floating platform. In
ASME 2013 32nd International Conference on Ocean, Offshore and Arctic Engineering, pages
V007T08A066V007T08A066. American Society of Mechanical Engineers, 2013.
[54] M Van Dyke and S Widnall. An album of fluid motion. Journal of Applied Mechanics, 50:475,
1983.
[55] J Kim Vandiver. Damping parameters for flow-induced vibration. Journal of Fluids and Structures, 35:105119, 2012.
[56] Guilherme Vaz, Frederick Jaouen, and Martin Hoekstra. Free-surface viscous flow computations: Validation of urans code fresco. In ASME 2009 28th International Conference on Ocean,
Offshore and Arctic Engineering, pages 425437. American Society of Mechanical Engineers,
2009.
[57] Olaf J Waals, Amal C Phadke, and Stephen Bultema. Flow induced motions on multi column
floaters. In ASME 2007 26th International Conference on Offshore Mechanics and Arctic
Engineering, pages 669678. American Society of Mechanical Engineers, 2007.
[58] David C Wilcox. Reassessment of the scale-determining equation for advanced turbulence
models. AIAA journal, 26(11):12991310, 1988.
[59] CHK Williamson and R Govardhan. A brief review of recent results in vortex-induced vibrations. Journal of Wind Engineering and Industrial Aerodynamics, 96(6):713735, 2008.
[60] CHK Williamson and N Jauvtis. A high-amplitude 2t mode of vortex-induced vibration for a
light body in xy motion. European Journal of Mechanics-B/Fluids, 23(1):107114, 2004.
[61] CHK Williamson and A Roshko. Vortex formation in the wake of an oscillating cylinder.
Journal of fluids and structures, 2(4):355381, 1988.
[62] Sherry Xiang, Peimin Cao, Oriol Rijken, Jian Ma, and Yongjun Chen. Riser vim fatigue
design induced by deep draft semi-submersible. In ASME 2010 29th International Conference
on Ocean, Offshore and Arctic Engineering, pages 365374. American Society of Mechanical
Engineers, 2010.
[63] MM Zdravkovich. Review and classification of various aerodynamic and hydrodynamic means
for suppressing vortex shedding. Journal of Wind Engineering and Industrial Aerodynamics,
7(2):145189, 1981.
[64] MM Zdravkovich. The effects of interference between circular cylinders in cross flow. Journal
of fluids and structures, 1(2):239261, 1987.
[65] Y Zhou and MW Yiu. Flow structure, momentum and heat transport in a two-tandem-cylinder
wake. Journal of Fluid Mechanics, 548:1748, 2006.

87 of 118

11

Appendix

11.1

Numerical methods

Computational fluid dynamics (CFD) is a numerical tool that solves fluid flow problems through
the combination of mathematical modelling and numerical analysis. Normally the CFD calculations
are realized through the help of computers due to the need for intensive numerical computation. A
CFD solver (code) is the core of a CFD computation tool. In this study a multi-phase viscous-flow
CFD code named ReFRESCO is used. This is a commercial code developed by HSVA, TUHH and
MARIN (Maritime Research Institute Netherlands). ReFRESCO stands for Reliable and Fast Rans
Equations (code) for Ships and Constructions Offshore. Its mathematical model has the Reynoldsaveraged Navier-Stokes(RANS) equations as the basis, supplemented with a series of turbulence
models based on the eddy viscosity concept. Since all the calculations are carried out through this
code, it plays an very important role for the whole study.
In this study, several hypothesises are followed by fluids and the flow field, for the sake of
simplification:
Newtonian fluids
Incompressible flow
Isothermal flow
In this section a brief elaboration of available mathematical models and numerical schemes are
presented. More information about ReFRESCO can be sought from the MARIN report [40].
11.1.1

Conservation principles

The conservation principles are the basis for CFD methods. Conservation laws can be derived by
considering a given quantity of matter or a so called control volume and its extensive properties, like
mass, momentum and energy [14]. In fluid flows, unlike solid bodies, it is more convenient to deal
with the flow within a certain spatial rather than in a parcel of matter which quickly passes through
the region of interest. This method is called the control volume approach. This subsection dedicates
to a brief introduction of those conservation equations that makes sense to this study.
Mass conservation The integral form of the mass conservation equation follows directly from the
control volume equation, also known as continuity equation. It has the form:
Z
Z

d +
v ndS = 0
(11.1)
t
S
with n the unit normal vector orthogonal to the surface S.
By applying the Gauss divergence theorem to the convection term, the surface integral can be
transferred into a volume integral. A differential coordinate-free form of the continuity equation can
be obtained through mitigating the control volume to be infinite small.

+ div(v) = 0
(11.2)
t
This form can be transformed into a form specific to a given coordinate system by providing the
expression for the divergence operator in that system. Expression for a Cartesian system is shown
below:
(ui )
(ux ) (uy )
(uz )
+
=
+
+
++
=0
t
xi
t
x
y
z

(11.3)

where xi (i=1,2,3) or (x,y,z) are the Cartesian coordinates and ui or (ux , uy , uz ) are the Cartesian
components of the velocity vector v.

88 of 118

The conservation equations in Cartesian form are often used and this will be the case in this
study.
Since the flow is incompressible in this case, the density of the fluid will not change with time.
Therefore the continuity equation becomes a form called divergence free:
v = 0

(11.4)

Momentum conservation The momentum conservation equation can also be derived through the
use of control volume method, similar to mass conservation equation. For a fixed fluid containing
volume of space:
Z
Z
X

vd +
v(v n)dS =
f
(11.5)
t
S
The right side of equation 4.5 contains the external forces which may be exerted on fluid in a
control volume:
Surface forces(pressure, normal and shear stresses, surface tension, etc.)
Body forces(gravity, centrifugal and Coriolis forces, electromagnetic forces, etc.)
The surface forces due to pressure and stresses are, from the molecular point of view, the microscopic momentum fluxes across a surface. For Newtonian fluids, the stress tensor, which represents
the molecular rate of transport of momentum, can be written as:
2
T = (p + div v)I + 2D
3
1
D = [grad v + (grad v)T ]
2

(11.6)
(11.7)

with the dynamic viscosity, I the unit tensor, p the static pressure and D the rate of strain tensor.
Within Cartesian coordinate system these two equations have the form:
2 ui
Tij = (p +
)ij + 2Dij
3 xj
1 ui
uj
Dij = (
+
)
2 xj
xi

(11.8)
(11.9)

where ij is the Kronecker symbol.


With the body forces being represented by b, the integral form of the momentum conservation
equation becomes:
Z
Z
Z
Z

vd +
v(v n)dS =
T ndS +
bd
(11.10)
t
S
S

In a Cartesian coordinate system, it has the form:


Z
Z
Z
Z

ui d +
ui v ndS =
ti ndS +
bi d
t
S
S

(11.11)

Through the application of Gauss divergence theorem to the convective and diffusive flux terms,
a coordinate free vector form of the momentum conservation equation can be expressed as:
(v)
+ div(vv) = div T + b
t
As a result, the corresponding equation for the ith Cartesian component is:

89 of 118

(11.12)

ui
+ div(ui v) = div ti + bi
t

(11.13)

Scalar conservation The integral form of the equation describing conservations of a scalar quantity is analogous to the previous equation following conservation volume method, which reads:
Z
Z
X

d +
v ndS =
f
(11.14)
t
S
with f the transport of any scalar quantity by mechanisms other than convection and sources or
sinks of the scalar.
Diffusive transport is always present and it is usually described by a gradient approximation(e.g.
Fouriers law for heat diffusion and Ficks law for mass diffusion):
Z
fd =
grad ndS
(11.15)
S

with the diffusivity for the quantity .


The integral form of the generic conservation equation follows:
Z
Z
Z
Z

d +
v ndS =
grad ndS +
q d
t
S
S

(11.16)

where q represents the source or sink of .


The coordinate free form of this equation is:
()
+ div(v) = div(grad) + q
(11.17)
t
In Cartesian coordinates and tensor notation, the differential form of the generic conservation
equation is:
() (uj )

+
=
(
) + q
t
xj
xj xj
11.1.2

(11.18)

Unsteady Reynolds Averaged Navier-Stokes Equations(URANS)

The canonical Navier-Stokes equations are exactly the sum of mass and momentum conservation
equations in previous subsection. In a summary they are listed below:

+ div(v) = 0
t
(v)
+ div(vv) = div T + b
t

(11.19)

with the stress tensor T and the deformation tensor D defined by:
2
T = (p + div v)I + 2D
3
1
D = [grad v + (grad v)T ]
2
The mathematical model used in ReFRESCO to deal with fluid flow problems is based on Unsteady Reynolds Averaged Navier-Stokes Equations, which is abbreviated as URANS. The present
model is intended to cope with turbulent flows. Turbulent flows are highly unsteady. Meanwhile,
they always include several time and length scales in realistic turbulent flows. Due to the limitation
of computational resource, it is not affordable to model all the turbulent scales by means of Direct

90 of 118

Numerical Simulation (DNS). Instead, a statistical approach by averaging the flow equations could
be relatively economical and feasible. This is the reason for the using of RANS model.
In a statistical steady flow, every variable can be written as the sum of a time averaged term and
a fluctuation term, which has the form:
(xi , t) = (xi ) + 0 (xi , t)

(11.20)

where stands for the variable, is the time averaged term and 0 is the fluctuation term.
The time averaged term could be put into this way:
1
T T

(xi ) = lim

(xi , t)dt

(11.21)

with t the time and T the averaging interval.


When the flow is unsteady, time averaging can not be used and it must be replaced by ensemble
averaging, as:
(xi ) = lim

N
X

(xi , t)

(11.22)

n=1

where N is the number of the ensemble and must be large enough to eliminate the effects of the
fluctuations.
Both of these averaging process are denoted as Reynolds averaging. If applied to Navier-Stokes
equations, we can acquire the Reynolds Averaged Navier-Stokes equations (RANS). Since the time
averaging of a fluctuation term is zero 0 = 0, averaging any linear term in any conservation equation
presented before simply gives the identical term for the averaged quantity. From a quadratic nonlinear term, two terms could be obtained, which are the product of the average and a covariance:
ui = (ui + u0i )( + 0 ) = ui + u0i 0

(11.23)

where for the velocity components we consider that:


ui (xi , t) = ui (xi , t) + u0i (xi , t)

(11.24)

The covariance term u0i 0 is zero only if the two quantities are uncorrelated. However, this is
rarely the case in turbulent flows and the conservation equations contain terms such as Reynolds
stresses u0i u0j and turbulent scalar flux u0i 0 among them. Those terms could not be represented
by mean quantities and thus need additional modelling.
In a conclusion, the RANS incompressible equations in the differential form, written in tensor
notation, reads:
(ui )
=0
xi
(ui )

p
ij
+
(ui uj + u0i u0j ) =
+
+ bi
t
xj
xi
xj

(11.25)
(11.26)

where the mean viscous stress tensor ij has components:


ij = (

ui
uj
+
) = 2Sij
xj
xi

(11.27)

In a similar way, the equation for the mean of a scalar quantity can be written:

()

+
(uj + u0j 0 ) =
(
) + q
t
xj
xj xj

91 of 118

(11.28)

Due to the presence of the Reynolds stresses and turbulent fluxes the equations are not closed
anymore. Therefore a need for an extra model arises, so as to close the system of equations. This
idea finally leads to the establishment of turbulence models.
11.1.3

Turbulence models

In order to close the RANS equations, it is necessary to introduce the turbulence models. The
turbulence models, which are adopted in this study, are founded on eddy-viscosity model, which is
also known as Boussinesq hypothesis. In laminar flows, energy dissipation and transport of mass,
momentum, and energy normal to the streamlines are mediated by the viscosity. As a result, it is
natural to assume that the effect of turbulence could be represented as an increased viscosity. This
leads to the eddy-viscosity model for the Reynolds stress:
u0i u0j = t (

ui
uj
2
+
) ij k
xj
xi
3

(11.29)

where k is the turbulent kinetic energy:


1
1 0 0
u u = (u0x u0x + u0y u0y + u0z u0z )
2 i i
2
It also leads to a eddy-diffusion model for a scalar:
k=

u0j 0 = t

xj

(11.30)

(11.31)

The turbulence can be then characterized by two parameters, the turbulence velocity scale v and
the turbulence length scale l.
t vl
Based on eddy-viscosity approximation, we can divide turbulence models in three categories:
Algebraic models or 0-equation models: These models use the eddy-viscosity concept
where the eddy viscosity is computed by means of a mixing length. The eddy viscosity depends on the flow velocity, but this relation is normally defined by an algebraic relation. These
are the most simple existing models, which however, have proven to be useful in many engineering fields. Two modern and well known 0-equation models are the Cebecci-Smith model
and the Baldwin-Lomax Model.
1-equation models: These models use the eddy-viscosity concept and are usually based on a
transport equation for the turbulence kinetic energy k together with a simple relation for the
turbulence length scale l. These models are therefore called incomplete. Examples of these
models are Prandtls 1-equation model, Baldwin-Barth model and Spalart-Allmaras model.
2-equation models: These models are still based on the eddy-viscosity hypothesis, but provide not only equations for the turbulence kinetic energy k but also for the turbulence length
scale l or equivalent. These models are therefore complete, i.e. can be used to predict properties of a given turbulent flow with no prior knowledge of the turbulent structure. The most
usual ones are the k  and k models and their numerous variants.
In this study, the turbulence model to be used is k SST Menter Model. The canonical
k shear stress transport (SST) model is first revealed by Menter in 1994 [34]. The formulation
is similar to the k original Wilcox model [58]. It is based on 2 transport equations. One for
the turbulence kinetic energy k and other for , which represents a characteristic frequency of the
turbulence. Based on the eddy-viscosity concept, for this model we have the turbulent viscosity to
be:

92 of 118

t =

k/
max(1, F2 /(a1 ))

(11.32)

where is the flow vorticity, and a1 a constant which value is determined by experiments.
The auxiliary function F2 is defined by means of the wall distance d:

500
k
F2 = tanh{[max(2
, 2 )]2 }
(11.33)
0.09d d
The transport equations include also an auxiliary function F 1, which performs a transition between the k original model and the k  model for the shear zones and the external zones. The
equations and coefficients that define the model are then:
Turbulent kinetic energy k equation:
k

k
+
(uj k ( + k t )
) = ij Sij k
t
xj
xj

(11.34)

Specific dissipation rate equation:

2 k

+
(uj ( + t )
) = 2 2 + 2(1 F1 )
(11.35)
t
xj
xj
xj xj
Closure coefficients and auxiliary relations:

k
500 42 4
, 2 ),
]] }
0.09d d CDk d2
22 k
= max(
, 1 1020 )
xj xj

F2 = tanh{[min[max(2
CDk

(11.36)

a1 = 0.31, = 0.09, = 0.41


The coefficients , , k and are defined by a transition between the coefficients of the
original model (denoted by 1) and a k transformed model (denoted by 2):
= F1 1 + (1 F1 )2 , = , , k ,

(11.37)

with the coefficients given by:


p
= 0.553
p
= 0.856, 2 = 0.0828, 2 = 2 / 2 K 2 / = 0.440

k1 = 0.85, 1 = 0.50, 1 = 0.075, 1 = 1 / 1 K 2 /


k2 = 1.00, 2

These coefficients are obtained through experiments.


11.1.4

Numerical formulation

The numerical formulation contains a wide range of topics, like spatial discretization, temporal
discretization, numerical scheme, solution algorithm, and mathematical methods to solve those algebraic equation systems. They are all highly mathematical corresponded topics and need lots of
effort for a complete elaboration. Since those topics are easy to retrieve through literature about numerical analysis and CFD, then in this section only the set-up and methods adopted are summarized
briefly, which are spatial discretization method, temporal discretization method, solution algorithm,
and schematic solution process.
Spatial discretization So far there are three major spatial discretization methods that are widely
acknowledged and used by industry:

93 of 118

Finite Difference Method (FDM)


Finite Volume Method (FVM)
Finite Element Method (FEM)
Among them the finite volume method is the one that enjoys a good popularity amongst many
engineers and researchers in CFD filed. Compared with the other two methods, it has a lot of
advantages. It is based on an approximation of integral form and can accommodate any form of grids,
including unstructured meshes. In ReFRESCO the spatial discretization is accomplished through a
cell-centre and collocated arrangement finite volume method.
The application of FVM begins with the discretization of the integral form of conservation equations in the control volume. Both the surface and volume integrals should be approximated through
quadrature formulae in order to get an algebraic equation.
Besides, since the approximation to the integrals require the values of variables at locations other
than computational nodes. For instance, to calculate the convective and diffusive fluxes, the value
of a variable and its gradient normal to the cell face at one or more locations on the control volume
are in need. Volume integrals of the source terms may also require these values. They have to be
expressed in terms of the nodal values by interpolation. There are several numerical schemes to deal
with interpolation problems, which are the most common methods for CFD codes:
Upwind Differencing Schemes (UDS): This scheme is of high accuracy but can not be obtained on affordable grids with upwind scheme. Meanwhile, it introduces a large diffusive
error in the computational domain. This should be avoided in the calculation.
Central Differencing Schemes (CDS): This is the simplest scheme of second order accuracy. Compared with other numerical schemes, it could be taken as a good balance between
accuracy, efficiency and simplicity.
Quadratic Upwind Interpolation for Convective Kinematics (QUICK): This scheme is
somewhat more complex than the CDS scheme. Although the QUICK approximation is
slightly more accurate than CDS, both schemes converge asymptotically in a second order
manner and the differences are rarely different.
Apart from those, there are some high order schemes which are quite suitable for fine grids. In
ReFRESCO, there is a hybrid scheme with a blend of central differencing and first order upstream
differencing, which is denoted as UD1 CD BLEND. This scheme will be applied to all the CFD
calculations in this study with a factor 0.5, indicating a linear combination of the solutions of both
the upwind scheme and central scheme.
Temporal discretization Time must be discretized like space when cope with unsteady flow.
There are two types of methods:
Explicit methods: The explicit methods calculate the state of a system at a latter time step
from the state of the system at the current time step. These methods are simple and relatively
easy to program. Moreover, they dont require a lot for random access memory (RAM) of
the computer. However, the explicit methods are conditionally stable and can not handle a
computation with large time step.
Implicit methods: The implicit methods solve the state of a system in terms of both the
current time step as well as the latter time step. Hence the calculation process is iterative and
it is quite difficult to program. It also has a large demand of computer memory (RAM). Albeit
it is more stable and allows a calculation with relatively large Courant number (time step).
In ReFRESCO, there is an implicit backward approximation scheme named Three Time Level
method. Through this method the scalars could be discretized through the equation:

94 of 118

Z
()d

1
[c1 (c c )n + c2 (c c )n1 + c3 (c c )n2 ]
t

(11.38)

with n the time level, t the time step and c c the simple spatial discretization. The setting for
the constants in ReFRESCO are:
First order backward Euler scheme:
c1 = 1.0, c2 = 1.0, c3 = 0.0
Second order backward scheme:
c1 = 1.5, c2 = 2.0, c3 = 0.5
Since the solution provided by first order backward scheme is quite rough. Hence in this study
the second order backward scheme is used for all the CFD calculations.
Solution algorithm The solution algorithm in this case refers to a pressure correction method.
This implies that the solution is obtained in the following steps. First the momentum equations
are solved in a segregated manner, assuming the pressure to be known, which yields a predictor
for the velocity field, which is not divergence free, however. With this preliminary velocity field
a Poisson equation for the pressure (or rather a correction on the pressure) is solved, and with the
new pressure a correction on the velocity is made so that it satisfies the continuity equation. This
process is repeated until both momentum equation and continuity equation are satisfied up to given
tolerances.
Schematic solution process The schematic solution process in ReFRESCO is shown in figure
11.1. The set of equations to be solved depends on the flow problem at hand. The equations are
coupled, but they are solved in a segregated manner, i.e. each equation is associated with a single
variable and solved for that variable. Also, the momentum equations are linearized using a simple
Picard-type of linearization [14]. By iteration the coupling and the non-linear character of the equations is restored. For time-dependent flow problems such an iteration process is to be done for each
time step.
Thus the solution process is schematically ordered as:
initialization
do (time loop)
increment t
do (outer loop)
solve the momentum equations
do (inner loop)
solve linear system of equations
enddo (inner loop)
solve the pressure correction equation
do (inner loop)
solve linear system of equations
enddo (inner loop)
correct velocity field
solve turbulence model equations
do (inner loop)
solve linear system of equations
enddo (inner loop)
solve additional transport equations
do (inner loop)
solve linear system of equations
enddo (inner loop)

95 of 118

Figure 11.1: Schematic solution process in ReFRESCO

enddo (outer loop)


enddo (time loop)
As indicated in this procedure, the solution process has three iteration loops, denoted as the
time loop, the outer loop and the inner loop. The process exits from the loops as soon as suitable
convergence criteria are met.

96 of 118

11.2

Verification of CFD tools

It is necessary to set the credibility of the CFD calculation results through verification and validation (V&V). As depicted by Roache[43],these two activities have different targets. Verification
is a purely mathematical exercise which intends to show that we are solving the equations right,
while validation is a science or engineering activity that intends to show that we are solving the
right equations. In this section, the methods adopted for the validation and verification and the
theory behind them are discussed.
11.2.1

Verification

Verification is a purely mathematical exercise that deals with numerical errors and uncertainties.
It includes two different activities[43], which are code verification and solution verification. Code
verification aims to ensure that a given code solves the equations of its numerical model correctly by
error evaluation, while solution verification intends to estimate the error and uncertainties of a given
calculation under the condition that the exact value is unknown.
The definition for error and uncertainties are given explicitly by Roache[45].Besides, a very
detailed classification of uncertainties and errors is available on the website of NPARC Alliance[1],
which is one of the archives of NASA. Their taxonomy of errors is shown below:
Acknowledged error
1. Physical approximation error
Physical modelling error
Geometry modelling error
2. Computer round-off error
3. Iterative convergence error
4. Discretization error
Spatial discretization error
Temporal discretization error
Unacknowledged error
1. Computer programming error
2. Usage error
In terms of the targets of this study, only numerical errors are inspected. Therefore, merely
computer round-off error, iterative convergence error and discretization error are applicable to this
case. All these three source of numerical errors will be briefly described below.
11.2.2

Numerical errors

It is generally agreed that numerical errors are inevitable in numerical calculations. In order
to ensure the reliability of the CFD calculation results, it is very important to estimate and control
the numerical errors. Generally acknowledged by the industry, there are three components which
are responsible for the deviation of the results from the exact results[45]. They are round-off error,
iterative error and discretization error.
The round-off error originates from the finite precision of the computers, for the numbers are
stored as a floating form. It is also known as truncation error, whose importance tends to increase
with the increment of grid refinement. Albeit it is important to keep an eye on it, nowadays most of
the computers are capable of storing data in 32 or 64 bit, which ensures the accuracy of the numerical
calculations free from the influence of round-off error. Herewith the round-off error in this case is
considered neglected.

97 of 118

The iterative error comes from the iterative methods adopted in CFD computation process. As
the mathematical equations in CFD solvers are non-linear, iterative error merely occur when solving
these equations through iterations. Theoretically, the iterative error should be kept in the same level
as round-off error. However, this is not always feasible for complex CFD calculations, as it will be
too expensive or even impossible due to the relative coarse grids or complex flow conditions with
the presence of vortex[12]. Therefore the iterative error will be monitored and controlled here to a
certain level. This can be realized quantitatively through the inspection of the convergence history
obtained in each iteration by ReFRESCO.
The discretization error stems from the representation of the partial differential equations, which
are continuum formulations, as algebraic expressions in a discrete spacial and temporal domain. The
approximations are usually made through Finite Difference Method (FDM), Finite Volume Method
(FVM), Finite Element Method (FEM) and Spectral Element Method (SEM)[46].
Typically in CFD applications discretizaiton error is the main contributor to the numerical error.
This is true when the iterative error is two or three orders of magnitude smaller than discretization
error[10]. In this case, for the complex flows study, it will be difficult to reduce the iterative error to
a negligible level.
11.2.3

Code verification

The errors evaluation of the code verification indicates that the exact solution should be known.
However, it is impossible to provide an analytical solution for a complex flows problem including
turbulence. As suggested by Eca, Vaz and Hoeskstra [13], the method of the manufactured solutions
(MMS) [44] [47] can be applied to perform code verification as an alternative without knowing exact
solutions. In the MMS, the problem is thoroughly treated in a mathematical way. The constructed
solution does not need a physical meaning. The process is shown below[43]:
1. Manufacture a continuum solution - an exact solution from a sufficient differentiable function is created which should exercise all the ordered derivatives in the equations and all the
equations terms;
2. Balance the equations - create and add a source term to governing equations to balance them,
in terms of the manufactured solution;
3. Code order verification - compute the discretization error and the order of grid convergence
based on grid refinement.
However, this simple method also has some drawbacks. In the MMS, all the equations and all the
terms in the equations should be exercised. That means, for a RANS solver, the transport equations
of the turbulence model should also be exercised. But it is quite difficult to manufacture solutions for
turbulence quantities, as reported in [11]. Moreover, it needs large sets of data to provide convincing
results. As such code verification task has been achieved previously [39, 9, 17, 6], with respect to
the same governing equations adopted in this study in ReFRESCO, the code verification will not be
treated here.
11.2.4

Solution verification

For the sake of the estimation of the numerical uncertainties, solution verification is performed here
to define an interval that contains the exact solution with 95% confidence. The relation is shown in
the equation below.
i U o i + U
(11.39)
where i is the calculated solution, U is the uncertainty value, and o is the exact solution, which
is not known.
In spite of several types of error estimator available in the open literature [45], most of them are
based on grid refinement method. Here the Richardson extrapolation (RE) will be applied as the

98 of 118

estimator for discretization error. Meanwhile, instead of focusing on numerical uncertainty evaluation with the direct use of grid convergence index (GCI), which is recommended by Roache [43]
and widely adopted by the industry, a least squares variant of GCI is put into use. As described by
Eca and Hoekstra in their work [9], the least square version of the GCI improves the reliability if the
uncertainty estimation.
Besides, as mentioned above, iterative error will also be inspected in this case.
Iterative error for steady flows An quantitative overview of the iterative error is given in the
convergence history by ReFRESCO, within the form of residuals. The residuals are presented in
two dimensionless forms, which are in L2 norm and L norm. According to the definition of norms
in mathematics, in this case the L2 norm is the RMS of the absolute change of the residuals of a
given variable between successive iterations over the complete calculation domain.
v
u np
uX
res2i
(11.40)
L2 (res ) = t
i=1

where res is the non-dimensional change of the residual of a variable, and np the total number of
grid cells.
Similar to the definition of the L2 norm, the L norm is the maximum absolute change of the
residuals of a given variable between successive iterations over the thorough calculation domain. It
can be expressed as:
v
u np
uX
t
res
(11.41)
L (res ) =
i = max|resi |
i=1

The check list of the variables with respect to mathematical norms of their residuals includes
velocity along x, y and z axis, pressure, and turbulent kinetic energy.
Iterative error for unsteady flows In this case the maximum residual values around all the iterations in each time step are of great importance. For VIM tests, the behaviour of floater when
reaching a periodical movement state is of more interest. However, it needs time to reach that state.
The interval between the start of the simulation and the arrival of a periodical state is considerably
affected by initial conditions.
Discretizaiton error for steady flows As described at the beginning of this chapter, discretization
error originates from the finite number of grid cells and discontinuity of time. It can be acquired
through RE, which says:
 RE = i o = hpi
(11.42)
where i is the calculated solution, o is the exact solution, is a constant, hi is the characteristic
grid size, and p is the observed order of accuracy.
The definition of the characteristic grid size will be discussed later. According to Eca and Hoekstra [8], as RE is based on the grid refinement study, it may result in some scatter of the data by
several sources:
Disparity of grid geometry
Girds outside the asymptotic range
Numerical integration or interpolation are in need to post-process the flow quantities of interest
As a consequence, at least one extra grid should be backed up to check the result acquired from RE,
ensuring the reliability of error estimation.
The basis of of the procedure to estimate the uncertainty is still the standard GCI, which is:
U = Fs |RE |

99 of 118

(11.43)

with Fs the safety factor. Take into account the equation (5.4) and (5.5), there are three unknowns:
exact , and p. If solutions on more than three grids are available, then these three unknowns
can be determined with least squares fit that depends on the standard deviation of the fit. That is
the motivation for the application of the least square root version of the GCI, as it can improve the
reliability of the uncertainty. A least squares root approach has the function:
v
u np
uX
(11.44)
S(o , , p) = t (i (o + hpi ))2
i=1

The determination of p is extremely sensitive to perturbations. Hence alternative error estimations are presented, as
2
RE
= i o = h2i
3
RE
= i o = 1 hi + 2 h2i

(11.45)

There are still 3 unknowns: o , 1 and 2 . Then the function of the least square root approach
changes to:
v
u np
uX
S(o , 1 , 2 ) = t (i (o + 1 h2i + 2 h3i ))2
(11.46)
i=1

When more than three grids are available and the least square root approach is used, the classification of the convergence condition is not straightforward[9]. But still, it can be classified as:
Monotonic convergence
Oscillatory convergence
Monotonic divergence
Oscillatory divergence
The only condition which allows an error estimation based on RE is monotonic convergence[10].
Otherwise one may rely on alternative uncertainty quantifications, as the method based on the maximum difference between all the solutions.
M = max(|i j |)

(11.47)

where i and j are calculated maximum and minimum solutions.


The procedure can be summarized as follows, which is valid for a normally second-order accurate method[10].
1. Estimate the observed order of accuracy p with the least square root method to identify the
apparent convergence condition
2. For monotonic convergence
For 0 < p < 0.95
Ud () = min(1.25RE + Us , 1.25M )
For 0.95 p < 2.05
Ud () = 1.25RE + Us
For p 2.05
Ud () = max(1.25RE + Us , 1.25M )

100 of 118

3. If monotonic convergence is not observed


Ud () = 3M
where Us is the standard deviations of the least square fits, and 1.25 and 3 are safety factors which
are adopted by standard GCI method.
Discretizaiton error for unsteady flows In unsteady flows both the spacial derivatives and time
derivatives contribute to the discretization error. Herewith the power series expansion of RE becomes:
 RE = i o = x hpi x + t ipt

(11.48)

where i is the time step, pt the observed order of accuracy and t a constant.
As described in [47] [45], it is possible to replace the two terms of the power series expansion by
a single term. However, it need hi and i to be chosen in a same way. This is not easy in this case.
Hence in this study they will be treated respectively.
Characteristic cell size definition The definition of h is a geometrical problem. It is a single
parameter used to define the typical cell size of the grids. The estimation of the discretization error
based on grid refinement study is based on two assumptions:
In order to guarantee that the leading term of the series expansion is dominant, the solutions
on the grids should be located in the asymptotic range;
The grid refinement ratio is defined by a single parameter, which is the characteristic cell size
(or typical cell size).
For the sake of the reliability of the results based on grid refinement study, it puts some strict
requirements on the grids:
The grids must be geometrically similar
The grids refinement ratio must be constant in the complete field
The grids properties, like orthogonality and skewness, should be independent of grid refinement
Unfortunately, small deviation from this constraint may have severe consequences for the error
estimation. However, in this study we use a commercial software named Hexpress to generate the
mesh. It is a fully automatic unstructured hexahedral grid generator software which is designed to
automatically generate meshes in complex 2D and 3D geometries. Though the idea of automatic
generation of mesh is very convenient, it also leads to the lack of manipulation of the process and an
unstructured grid. As a result, it is quite difficult to keep a constant grid refinement ratio, needless
to say the geometry similarity. Hence, the characteristic cell size should be selected with great care.
In the grid refinement studies accomplished by Eca et al. based on grids not exhibiting a constant refinement ratio for the entire computation domain, some alternatives for the definition of the
characteristic cell size are tested [10]. Here are four different ways of defining hi :
1. The inverse of the numer of cells of the grid
1

1
)n
hi = ( Ncells

where n is the space dimension


2. The average size of the cells
hi = (

PNcells
i=1

Ncells

)n

where i is the cell length, area or volume of a cell.

101 of 118

3. The root mean square of the cell size


r
hi = (

PNcells
i=1

Ncells

2i

)n

4. The mode of the cell size


This is the cell size that occurs more often in the cell size distribution, which should be acquired through a probability density function of the cell size distribution.
Then the grid refinement ratio can be expressed as hh1i . Their tests results based on these four
grids show it is possible to make reliable discretization error estimation on the basis of power series
expansion for grid families that are non-geometrically similar, while the estimation error is not too
sensitive to the definition of the characteristic cell size. The solution verification work done by Make
[33] also provides some evidence that these four definition of typical cell size do not make significant
difference. Therefore, considering the limitation of Hexpress, which makes the acquiring of detailed
cell size information quite difficult, here we will just take the inverse of the number of elements,
method No.1 as mentioned above.
Besides, as indicated in [1], the grid refinement ratio should be larger than 1.1 to allow the
discretization error to be differentiated from other error sources, like the iterative convergence error
and computer round-off error.
11.2.5

Validation

Validation is a science or engineering activity that copes with modelling errors and uncertainties
of a given mathematical model. As a result, it requires the comparison between the experimental
results and the calculated ones. On the other hand, it also includes experimental and parameter
uncertainties, apart from numerical errors, contrasted with verification process. A procedure has
been proposed by [45]:
The validation uncertainty
q
2
2
2
+ Uinput
+ UD
Uval = Unum
The validation comparison error
E =SD
with Unum the numerical uncertainty, Uinput the parameter uncertainty, the UD the experimental
uncertainty, S is the numerical prediction and D the experimental value.
The intention of the procedure above is to estimate the interval containing the modelling error
with 95% confidence. Hence, the comparison between Uval and |E| can be illustrative, as follows:
If |E| > Uval , the comparison error is similar to the modelling error, which implies that the
model has to be further improved.
If |E| < Uval , the modelling error is within the noise level exerted by numerical, experimental and parameter uncertainties. It has two meanings:
If E is small enough, the model and its solution are validated against the given experiment
If E is not that small, then it indicates there is still space for the improvement of the
quality of either the numerical solution or the experiment.
The validation is not one of the objectives in this study. As a consequence, there is no experiment carried out on purpose as contrast group for the sake of validation. Hence the validation of the
numerical model in ReFRESCO is not performed here. Nevertheless, there are quite a lot of publications and thesis about it [56] [17] [26] [6]. They can be referred as the evidence for the validation
of ReFRESCO.

102 of 118

11.3

Script for the modelling of a four columns floater in Hexpress

\begin{verbatim}
#integrated script for 45 degree deployed four columns floater in hexpress
igg_script_version(2.1)
#construction of the model
HXP.create_cube("B1",Point(0.405,0.42525,0),Point(0.675,0.65475,1))
HXP.create_cube("B2",Point(0.42525,0.405,0),Point(0.65475,0.675,1))
HXP.create_cylinder("B3",Point(0.42525,0.42525,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B5",Point(0.42525,0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B6",Point(0.65475,0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B7",Point(0.65475,0.42525,0),Point(0,0,1),1,0.02025)
HXP.unite_bodies("B1",["B2"])
HXP.unite_bodies("B1",["B3"])
HXP.unite_bodies("B1",["B5"])
HXP.unite_bodies("B1",["B6"])
HXP.unite_bodies("B1",["B7"])
HXP.create_cube("B8",Point(-0.405,0.42525,0),Point(-0.675,0.65475,1))
HXP.create_cube("B9",Point(-0.42525,0.405,0),Point(-0.65475,0.675,1))
HXP.create_cylinder("B10",Point(-0.42525,0.42525,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B11",Point(-0.42525,0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B12",Point(-0.65475,0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B13",Point(-0.65475,0.42525,0),Point(0,0,1),1,0.02025)
HXP.unite_bodies("B8",["B9"])
HXP.unite_bodies("B8",["B10"])
HXP.unite_bodies("B8",["B11"])
HXP.unite_bodies("B8",["B12"])
HXP.unite_bodies("B8",["B13"])
HXP.create_cube("B14",Point(-0.405,-0.42525,0),Point(-0.675,-0.65475,1))
HXP.create_cube("B15",Point(-0.42525,-0.405,0),Point(-0.65475,-0.675,1))
HXP.create_cylinder("B16",Point(-0.42525,-0.42525,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B17",Point(-0.42525,-0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B18",Point(-0.65475,-0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B19",Point(-0.65475,-0.42525,0),Point(0,0,1),1,0.02025)
HXP.unite_bodies("B14",["B15"])
HXP.unite_bodies("B14",["B16"])
HXP.unite_bodies("B14",["B17"])
HXP.unite_bodies("B14",["B18"])
HXP.unite_bodies("B14",["B19"])
HXP.create_cube("B20",Point(0.405,-0.42525,0),Point(0.675,-0.65475,1))
HXP.create_cube("B21",Point(0.42525,-0.405,0),Point(0.65475,-0.675,1))
HXP.create_cylinder("B22",Point(0.42525,-0.42525,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B23",Point(0.42525,-0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B24",Point(0.65475,-0.65475,0),Point(0,0,1),1,0.02025)
HXP.create_cylinder("B25",Point(0.65475,-0.42525,0),Point(0,0,1),1,0.02025)
HXP.unite_bodies("B20",["B21"])
HXP.unite_bodies("B20",["B22"])
HXP.unite_bodies("B20",["B23"])
HXP.unite_bodies("B20",["B24"])
HXP.unite_bodies("B20",["B25"])
HXP.rotate_bodies(["B1","B8","B14","B20"],Point(0,0,0),Point(0,0,1),45)
HXP.create_cylinder("B2",Point(0,0,0),Point(0,0,1),1,10.8)

103 of 118

HXP.subtract_bodies("B2",["B1"])
HXP.subtract_bodies("B2",["B8"])
HXP.subtract_bodies("B2",["B14"])
HXP.subtract_bodies("B2",["B20"])
# mesh wizard
HXP.set_mesh_generation_mode("3D")
HXP.domain("test1").get_face(1).set_type("MIR",0)
HXP.domain("test1").get_face(2).set_type("MIR",0)
HXP.set_mesh_generation_mode("2D")
#boundary condition
HXP.domain("test1").get_face(0).set_name("FarField")
HXP.domain("test1").get_face(1).set_name("Surface")
HXP.domain("test1").get_face(2).set_name("Bottom")
HXP.domain("test1").create_BC_group("NEW GROUP","SOL")
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",3)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",4)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",5)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",6)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",7)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",8)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",9)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",10)
HXP.domain("test1").set_BC_group_name("NEW GROUP","Column_1")
HXP.domain("test1").create_BC_group("NEW GROUP","SOL")
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",11)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",12)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",13)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",14)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",15)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",16)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",17)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",18)
HXP.domain("test1").set_BC_group_name("NEW GROUP","Column_2")
HXP.domain("test1").create_BC_group("NEW GROUP","SOL")
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",19)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",20)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",21)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",22)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",23)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",24)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",25)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",26)
HXP.domain("test1").set_BC_group_name("NEW GROUP","Column_3")
HXP.domain("test1").create_BC_group("NEW GROUP","SOL")
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",27)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",28)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",29)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",30)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",31)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",32)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",33)
HXP.domain("test1").add_patch_to_BC_group("NEW GROUP",34)
HXP.domain("test1").set_BC_group_name("NEW GROUP","Column_4")

104 of 118

#Initial mesh
HXP.init_cartesian_mesh(90,90,1)
HXP.generate_initial_mesh()
HXP.set_global_number_of_refinements(6)
#surface refinement
HXP.domain("test1").create_adaptation_group("Column_1",[3,4,5,6,7,8,9,10])
HXP.domain("test1").create_adaptation_group("Column_2",[11,12,13,14,15,16,17,18])
HXP.domain("test1").create_adaptation_group("Column_3",[19,20,21,22,23,24,25,26])
HXP.domain("test1").create_adaptation_group("Column_4",[27,28,29,30,31,32,33,34])
HXP.domain_face(3).enable_adaptation(True)
HXP.domain_face(3).set_number_of_refinements(6)
HXP.domain_face(3).set_adaptation_criteria(0,1,1)
HXP.domain_face(4).enable_adaptation(True)
HXP.domain_face(4).set_number_of_refinements(6)
HXP.domain_face(4).set_adaptation_criteria(0,1,1)
HXP.domain_face(5).enable_adaptation(True)
HXP.domain_face(5).set_number_of_refinements(6)
HXP.domain_face(5).set_adaptation_criteria(0,1,1)
HXP.domain_face(6).enable_adaptation(True)
HXP.domain_face(6).set_number_of_refinements(6)
HXP.domain_face(6).set_adaptation_criteria(0,1,1)
HXP.domain_face(7).enable_adaptation(True)
HXP.domain_face(7).set_number_of_refinements(6)
HXP.domain_face(7).set_adaptation_criteria(0,1,1)
HXP.domain_face(8).enable_adaptation(True)
HXP.domain_face(8).set_number_of_refinements(6)
HXP.domain_face(8).set_adaptation_criteria(0,1,1)
HXP.domain_face(9).enable_adaptation(True)
HXP.domain_face(9).set_number_of_refinements(6)
HXP.domain_face(9).set_adaptation_criteria(0,1,1)
HXP.domain_face(10).enable_adaptation(True)
HXP.domain_face(10).set_number_of_refinements(6)
HXP.domain_face(10).set_adaptation_criteria(0,1,1)
HXP.domain_face(11).enable_adaptation(True)
HXP.domain_face(11).set_number_of_refinements(6)
HXP.domain_face(11).set_adaptation_criteria(0,1,1)
HXP.domain_face(12).enable_adaptation(True)
HXP.domain_face(12).set_number_of_refinements(6)
HXP.domain_face(12).set_adaptation_criteria(0,1,1)
HXP.domain_face(13).enable_adaptation(True)
HXP.domain_face(13).set_number_of_refinements(6)
HXP.domain_face(13).set_adaptation_criteria(0,1,1)
HXP.domain_face(14).enable_adaptation(True)
HXP.domain_face(14).set_number_of_refinements(6)
HXP.domain_face(14).set_adaptation_criteria(0,1,1)
HXP.domain_face(15).enable_adaptation(True)
HXP.domain_face(15).set_number_of_refinements(6)
HXP.domain_face(15).set_adaptation_criteria(0,1,1)
HXP.domain_face(16).enable_adaptation(True)
HXP.domain_face(16).set_number_of_refinements(6)
HXP.domain_face(16).set_adaptation_criteria(0,1,1)
HXP.domain_face(17).enable_adaptation(True)
HXP.domain_face(17).set_number_of_refinements(6)

105 of 118

HXP.domain_face(17).set_adaptation_criteria(0,1,1)
HXP.domain_face(18).enable_adaptation(True)
HXP.domain_face(18).set_number_of_refinements(6)
HXP.domain_face(18).set_adaptation_criteria(0,1,1)
HXP.domain_face(19).enable_adaptation(True)
HXP.domain_face(19).set_number_of_refinements(6)
HXP.domain_face(19).set_adaptation_criteria(0,1,1)
HXP.domain_face(20).enable_adaptation(True)
HXP.domain_face(20).set_number_of_refinements(6)
HXP.domain_face(20).set_adaptation_criteria(0,1,1)
HXP.domain_face(21).enable_adaptation(True)
HXP.domain_face(21).set_number_of_refinements(6)
HXP.domain_face(21).set_adaptation_criteria(0,1,1)
HXP.domain_face(22).enable_adaptation(True)
HXP.domain_face(22).set_number_of_refinements(6)
HXP.domain_face(22).set_adaptation_criteria(0,1,1)
HXP.domain_face(23).enable_adaptation(True)
HXP.domain_face(23).set_number_of_refinements(6)
HXP.domain_face(23).set_adaptation_criteria(0,1,1)
HXP.domain_face(24).enable_adaptation(True)
HXP.domain_face(24).set_number_of_refinements(6)
HXP.domain_face(24).set_adaptation_criteria(0,1,1)
HXP.domain_face(25).enable_adaptation(True)
HXP.domain_face(25).set_number_of_refinements(6)
HXP.domain_face(25).set_adaptation_criteria(0,1,1)
HXP.domain_face(26).enable_adaptation(True)
HXP.domain_face(26).set_number_of_refinements(6)
HXP.domain_face(26).set_adaptation_criteria(0,1,1)
HXP.domain_face(27).enable_adaptation(True)
HXP.domain_face(27).set_number_of_refinements(6)
HXP.domain_face(27).set_adaptation_criteria(0,1,1)
HXP.domain_face(28).enable_adaptation(True)
HXP.domain_face(28).set_number_of_refinements(6)
HXP.domain_face(28).set_adaptation_criteria(0,1,1)
HXP.domain_face(29).enable_adaptation(True)
HXP.domain_face(29).set_number_of_refinements(6)
HXP.domain_face(29).set_adaptation_criteria(0,1,1)
HXP.domain_face(30).enable_adaptation(True)
HXP.domain_face(30).set_number_of_refinements(6)
HXP.domain_face(30).set_adaptation_criteria(0,1,1)
HXP.domain_face(31).enable_adaptation(True)
HXP.domain_face(31).set_number_of_refinements(6)
HXP.domain_face(31).set_adaptation_criteria(0,1,1)
HXP.domain_face(32).enable_adaptation(True)
HXP.domain_face(32).set_number_of_refinements(6)
HXP.domain_face(32).set_adaptation_criteria(0,1,1)
HXP.domain_face(33).enable_adaptation(True)
HXP.domain_face(33).set_number_of_refinements(6)
HXP.domain_face(33).set_adaptation_criteria(0,1,1)
HXP.domain_face(34).enable_adaptation(True)
HXP.domain_face(34).set_number_of_refinements(6)
HXP.domain_face(34).set_adaptation_criteria(0,1,1)
#Refinement box

106 of 118

HXP.create_refinement_cube(-1.62,-1.35,0,2.7,1.35,0)
HXP.refinement_box(0).set_adaptation_flags(1,1)
HXP.refinement_box(0).set_refinement_level(4)
HXP.create_refinement_cube(-2.16,-1.62,0,3.24,1.62,0)
HXP.refinement_box(1).set_adaptation_flags(1,1)
HXP.refinement_box(1).set_refinement_level(3)
HXP.create_refinement_cube(-0.24,0.52,0,0.24,1,0)
HXP.refinement_box(2).set_adaptation_flags(1,1)
HXP.refinement_box(2).set_refinement_level(6)
HXP.create_refinement_cube(-0.52,-0.24,0,-1,0.24,0)
HXP.refinement_box(3).set_adaptation_flags(1,1)
HXP.refinement_box(3).set_refinement_level(6)
HXP.create_refinement_cube(0.24,-0.52,0,-0.24,-1,0)
HXP.refinement_box(4).set_adaptation_flags(1,1)
HXP.refinement_box(4).set_refinement_level(6)
HXP.create_refinement_cube(0.52,-0.24,0,1,0.24,0)
HXP.refinement_box(5).set_adaptation_flags(1,1)
HXP.refinement_box(5).set_refinement_level(6)
HXP.create_refinement_cube(2.1,-1.08,0,-1.14,1.08,0)
HXP.refinement_box(6).set_adaptation_flags(1,1)
HXP.refinement_box(6).set_refinement_level(5)
#optimization
HXP.adapt_mesh()
HXP.snap_mesh()
HXP.regularize_mesh()
HXP.set_optimization_params(0,4,100,7,3,0,10)
#viscous layers
HXP.domain_face(3).enable_viscous_layers(True)
HXP.domain_face(3).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(4).enable_viscous_layers(True)
HXP.domain_face(4).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(5).enable_viscous_layers(True)
HXP.domain_face(5).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(6).enable_viscous_layers(True)
HXP.domain_face(6).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(7).enable_viscous_layers(True)
HXP.domain_face(7).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(8).enable_viscous_layers(True)
HXP.domain_face(8).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(9).enable_viscous_layers(True)
HXP.domain_face(9).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(10).enable_viscous_layers(True)
HXP.domain_face(10).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(11).enable_viscous_layers(True)
HXP.domain_face(11).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(12).enable_viscous_layers(True)
HXP.domain_face(12).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(13).enable_viscous_layers(True)
HXP.domain_face(13).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(14).enable_viscous_layers(True)
HXP.domain_face(14).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(15).enable_viscous_layers(True)
HXP.domain_face(15).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)

107 of 118

HXP.domain_face(16).enable_viscous_layers(True)
HXP.domain_face(16).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(17).enable_viscous_layers(True)
HXP.domain_face(17).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(18).enable_viscous_layers(True)
HXP.domain_face(18).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(19).enable_viscous_layers(True)
HXP.domain_face(19).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(20).enable_viscous_layers(True)
HXP.domain_face(20).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(21).enable_viscous_layers(True)
HXP.domain_face(21).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(22).enable_viscous_layers(True)
HXP.domain_face(22).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(23).enable_viscous_layers(True)
HXP.domain_face(23).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(24).enable_viscous_layers(True)
HXP.domain_face(24).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(25).enable_viscous_layers(True)
HXP.domain_face(25).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(26).enable_viscous_layers(True)
HXP.domain_face(26).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(27).enable_viscous_layers(True)
HXP.domain_face(27).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(28).enable_viscous_layers(True)
HXP.domain_face(28).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(29).enable_viscous_layers(True)
HXP.domain_face(29).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(30).enable_viscous_layers(True)
HXP.domain_face(30).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(31).enable_viscous_layers(True)
HXP.domain_face(31).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(32).enable_viscous_layers(True)
HXP.domain_face(32).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(33).enable_viscous_layers(True)
HXP.domain_face(33).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.domain_face(34).enable_viscous_layers(True)
HXP.domain_face(34).set_viscous_layer_params(10,1.2,0.000171988875017,5,1.2)
HXP.set_viscous_layers_global_params(1,0,5,30,2,1)
HXP.set_viscous_layers_global_params(1,0,5,30,2,1)
HXP.insert_viscous_layers()

108 of 118

11.4

Control file for 2D CFD VIM simulation of a four columns floater in


ReFRESCO (Ur=6)

<?xml version="1.0"?>
<controls>
<general>
<codeVersion>2.1</codeVersion>
<name>VIM</name>
<description>VIM_45deg_Ur_6_B_0</description>
<caseid>Ur_6</caseid>
<material>WATER</material>
<referenceLength>0.365</referenceLength>
<referenceVelocity>0.140</referenceVelocity>
<referenceMaterial>WATER</referenceMaterial>
<referencePressure>0.0</referencePressure>
<gridFileName>Grid_4Columns</gridFileName>
<gridFilePath>../Grid_No.1/layers_18</gridFilePath>
<outFileName>VIM</outFileName>
<outFilePath>.</outFilePath>
<nsave>200</nsave>
<suppressOutput>false</suppressOutput>
</general>
<subGrids>
<subGrids name="Interior" >
<subGrid>
<moveGridApply>true</moveGridApply>
<calcEqsOfMotion>true</calcEqsOfMotion>
<moveGridMethod>MVG</moveGridMethod>
<moveGridUserDefined>false</moveGridUserDefined>
<rotationOrigin>0.0 0.0 0.0</rotationOrigin>
<rotationAxis>0.0 0.0 0.0</rotationAxis>
<rotationRate>0.0</rotationRate>
<translationVelocity>0.0 0.0 0.0</translationVelocity>
<initialDispApply>false</initialDispApply>
<initialDispRotationOrigin>0.0 0.0 0.0</initialDispRotationOrigin>
<initialDispRotationAxis>0.0 0.0 0.0</initialDispRotationAxis>
<initialDispRotationAngle>0.0</initialDispRotationAngle>
<initialDispTranslation>0.0 0.0 0.0</initialDispTranslation>
<saveMoveGridExtraQuantities>false</saveMoveGridExtraQuantities>
</subGrid>
</subGrids>
</subGrids>
<userCode>
<eqsOfMotion>
<apply>true</apply>
<startAtTimestep>1</startAtTimestep>
<origin>0.0 0.0 0.0</origin>
<centerOfGravity>0.0 0.0 0.0</centerOfGravity>
<nDOFs>3</nDOFs>

109 of 118

<translationDOFs>1 1 0</translationDOFs>
<rotationDOFs>0 0 1</rotationDOFs>
<initialDisp>0.0 0.0 0.0 </initialDisp>
<initialAngle>0.0 0.0 0.0 </initialAngle>
<families>Column_1 Column_2 Column_3 Column_4</families>
<method>IMPLICIT_PC_TYPE2</method>
<structPropertiesFileName>Struc_Properties</structPropertiesFileName>
<correctorConvergenceTolerance>1E-6</correctorConvergenceTolerance>
<correctorMaxIteration>10</correctorMaxIteration>
<correctorOuterLoopStep>10</correctorOuterLoopStep>
<pseudoAddedMassCorrection>0.0</pseudoAddedMassCorrection>
<saveMotion>true</saveMotion>
</eqsOfMotion>
</userCode>
<deformGrid>
</deformGrid>
<adaptiveGrid>
</adaptiveGrid>
<adaptLoop>
</adaptLoop>
<imposedMotion>
</imposedMotion>
<restart>
<restart>false</restart>
<resetCounter>false</resetCounter>
<restartFileName>VIM</restartFileName>
<restartFilePath>.</restartFilePath>
</restart>
<timeLoop>
<unsteady>true</unsteady>
<solutionScheme>IMPLICIT_THREE_TIME_LEVEL</solutionScheme>
<maxTimesteps>18000</maxTimesteps>
<timeDelta>0.044</timeDelta>
</timeLoop>
<outerLoop>
<maxIteration>600</maxIteration>
<convergenceTolerance>1e-4</convergenceTolerance>
<residualNorm>INFINITY</residualNorm>
<divergenceTolerance>1e+9</divergenceTolerance>
</outerLoop>
<accelerationOfGravity>
<apply>false</apply>
<gravityVector>0. 0. 9.81</gravityVector>
</accelerationOfGravity>

110 of 118

<bodyForces>
<apply>false</apply>
<userDefined>false</userDefined>
<toSave>false</toSave>
<bodyforceTreatment>EXPLICIT_SOURCE</bodyforceTreatment>
<bodyForce>0.0 0.0 0.0</bodyForce>
</bodyForces>
<boundaries>
<family name="Column_1" >
<BCWall>
<velocity userCode="false" >0 0 0</velocity>
<referenceSystem>BODY_FIXED</referenceSystem>
<temperature userCode="false" >293.</temperature>
<extrapolationOrder>0</extrapolationOrder>
<elog>8.43</elog>
<useWallFunction>NO</useWallFunction>
</BCWall>
</family>
<family name="Column_2" >
<BCWall>
<velocity userCode="false" >0 0 0</velocity>
<referenceSystem>BODY_FIXED</referenceSystem>
<temperature userCode="false" >293.</temperature>
<extrapolationOrder>0</extrapolationOrder>
<elog>8.43</elog>
<useWallFunction>NO</useWallFunction>
</BCWall>
</family>
<family name="Column_3" >
<BCWall>
<velocity userCode="false" >0 0 0</velocity>
<referenceSystem>BODY_FIXED</referenceSystem>
<temperature userCode="false" >293.</temperature>
<extrapolationOrder>0</extrapolationOrder>
<elog>8.43</elog>
<useWallFunction>NO</useWallFunction>
</BCWall>
</family>
<family name="Column_4" >
<BCWall>
<velocity userCode="false" >0 0 0</velocity>
<referenceSystem>BODY_FIXED</referenceSystem>
<temperature userCode="false" >293.</temperature>
<extrapolationOrder>0</extrapolationOrder>
<elog>8.43</elog>
<useWallFunction>NO</useWallFunction>
</BCWall>
</family>

111 of 118

<family name="Surface" >


<BCSlipWall>
<extrapolationOrder>0</extrapolationOrder>
</BCSlipWall>
</family>
<family name="FarField" >
<BCAutoDetect>
<velocity userCode="false" >0.140 0.0 0.0</velocity>
<referenceSystem>EARTH_FIXED</referenceSystem>
<temperature userCode="false" >293.</temperature>
<airVolumeFraction userCode="false" >0.</airVolumeFraction>
<vaporVolumeFraction userCode="false" >0.</vaporVolumeFraction>
<speciesVolumeFraction userCode="false" >0.</speciesVolumeFraction>
<pressure userCode="false" >0.</pressure>
<extrapolationOrder>0</extrapolationOrder>
<turbulence>
<turbIntensity_eddyVisc>
<turbIntensity userCode="false" >1.e-2</turbIntensity>
<eddyVisc userCode="false" >1.0</eddyVisc>
</turbIntensity_eddyVisc>
</turbulence>
<autoDetectAngleOutflow>0.0</autoDetectAngleOutflow>
<autoDetectAngleInflow>120.0</autoDetectAngleInflow>
<projection>false</projection>
<!--geometryFileName>Grid_4Columns</geometryFileName-->
</BCAutoDetect>
</family>
<family name="Bottom" >
<BCSymmetryPlane>
<extrapolationOrder>0</extrapolationOrder>
</BCSymmetryPlane>
</family>
</boundaries>
</boundaries-->
<massMomentumSolver>
<massMomentumSolverType name="SEGREGATED" >
<segregated>
<solver>FRESCO</solver>
<approxInvMomMat>ABSROWSUM</approxInvMomMat>
</segregated>
</massMomentumSolverType>
</massMomentumSolver>
<equations>
<equation>
<EQMomentum>
<solve_x>true</solve_x>
<solve_y>true</solve_y>

112 of 118

<!--Alternative_DIAG-->

<solve_z>false</solve_z>
<solver>
<PETSC>
<solver>GMRES</solver>
<preconditioner>JACOBI</preconditioner>
</PETSC>
</solver>
<convergenceTolerance>0.1</convergenceTolerance>
<maxIteration>200</maxIteration>
<relaxationProcedure>
<IMPLICIT_EXPLICIT>
<imp_relax_min>0.5</imp_relax_min>
<imp_relax_max>0.85</imp_relax_max>
<imp_relax_fac>50</imp_relax_fac>
<exp_relax>0.15</exp_relax>
</IMPLICIT_EXPLICIT>
</relaxationProcedure>
<convectiveFluxDiscretisation>
<UD1_CD_BLEND>
<blendingFactor>0.5</blendingFactor>
</UD1_CD_BLEND>
</convectiveFluxDiscretisation>
<gradientCalculation>
<GAUSS>
</GAUSS>
</gradientCalculation>
<applyExcentricityCorrection>true</applyExcentricityCorrection>
<userSource>false</userSource>
<residualNormalisation>PARNASSOS_LIKE</residualNormalisation>
<toSave>true</toSave>
<saveResidual>true</saveResidual>
<saveChanges>false</saveChanges>
<saveGradients>false</saveGradients>
<saveCourantNo>true</saveCourantNo>
<savePecletNo>false</savePecletNo>
<saveShearStress>true</saveShearStress>
<saveYplus>true</saveYplus>
<initialization>
<USER_DEFINED>
<initialVelocity userCode="false" >1.e-6 1.e-6 0.0</initialVelocity>
<referenceSystem>EARTH_FIXED</referenceSystem>
</USER_DEFINED>
</initialization>
</EQMomentum>
</equation>
<equation>
<EQPressure>
<solver>
<PETSC>
<solver>CG</solver>
<preconditioner>BJACOBI</preconditioner>
</PETSC>

113 of 118

</solver>
<convergenceTolerance>0.01</convergenceTolerance>
<maxIteration>500</maxIteration>
<relaxationProcedure>
<EXPLICIT>
<exp_relax>0.15</exp_relax>
</EXPLICIT>
</relaxationProcedure>
<gradientCalculation>
<GAUSS>
</GAUSS>
</gradientCalculation>
<applyExcentricityCorrection>true</applyExcentricityCorrection>
<pressureReference>0.0</pressureReference>
<pressureReferencePosition>0.0 0.0 0.0</pressureReferencePosition>
<userSource>false</userSource>
<residualNormalisation>PARNASSOS_LIKE</residualNormalisation>
<toSave>true</toSave>
<saveResidual>true</saveResidual>
<saveGradients>false</saveGradients>
<saveChanges>false</saveChanges>
<initialPressure userCode="false" >0</initialPressure>
</EQPressure>
</equation>
<equation>
<EQTurbulence>
<turbulenceModel>
<K_OMEGA>
<K_OMEGA_TYPE>SST_2003</K_OMEGA_TYPE>
</K_OMEGA>
</turbulenceModel>
<solver>
<PETSC>
<solver>GMRES</solver>
<preconditioner>BJACOBI</preconditioner>
</PETSC>
</solver>
<convergenceTolerance>0.1</convergenceTolerance>
<maxIteration>200</maxIteration>
<relaxationProcedure>
<IMPLICIT_EXPLICIT>
<imp_relax_min>0.5</imp_relax_min>
<imp_relax_max>0.85</imp_relax_max>
<imp_relax_fac>50</imp_relax_fac>
<exp_relax>0.15</exp_relax>
</IMPLICIT_EXPLICIT>
</relaxationProcedure>
<relaxEddyViscosity>1.0</relaxEddyViscosity>
<relaxAnisotropy>0.1</relaxAnisotropy>
<stagnationRegionCorrection>
<REALIZABILITY>
<realizabilityLimiter>10.0</realizabilityLimiter>

114 of 118

</REALIZABILITY>
</stagnationRegionCorrection>
<daclesMarianiCorrection>0.0</daclesMarianiCorrection>
<convectiveFluxDiscretisation>
<UD1_CD_BLEND>
<blendingFactor>0.0</blendingFactor>
</UD1_CD_BLEND>
</convectiveFluxDiscretisation>
<gradientCalculation>
<GAUSS>
</GAUSS>
</gradientCalculation>
<applyExcentricityCorrection>true</applyExcentricityCorrection>
<toSave>true</toSave>
<toSaveEddyVisc>true</toSaveEddyVisc>
<toSaveTurbulenceIntensity>true</toSaveTurbulenceIntensity>
<toSaveVonKarmanLength>false</toSaveVonKarmanLength>
<toSaveAnisotropy>false</toSaveAnisotropy>
<toSaveSASTerm>false</toSaveSASTerm>
<toSaveCurvatureRotationCorrection>false</toSaveCurvatureRotationCorrection>
<userSource>false</userSource>
<saveResidual>true</saveResidual>
<saveChanges>false</saveChanges>
<saveGradients>false</saveGradients>
<saveEddyViscGradients>false</saveEddyViscGradients>
<residualNormalisation>PARNASSOS_LIKE</residualNormalisation>
<initial>
<!--eddyVisc>
<eddyVisc userCode="false" >1.0</eddyVisc>
</eddyVisc-->
<turbIntensity_eddyVisc>
<turbIntensity userCode="false" >1.e-2</turbIntensity>
<eddyVisc userCode="false" >1.0</eddyVisc>
</turbIntensity_eddyVisc>
</initial>
</EQTurbulence>
</equation>
</equations>
<materials>
<material name="WATER" >
<fluid>
<viscosityMolecular>1.139e-3</viscosityMolecular>
<density>0.999e3</density>
<thermalConductivity>0.5996e-3</thermalConductivity>
<expansionCoefficient>0.207e-3</expansionCoefficient>
<referenceTemperature>293.</referenceTemperature>
<variableDensity>false</variableDensity>
<specificHeat>4.182e3</specificHeat>
</fluid>
</material>
</materials>

115 of 118

<extraQuantities>
<saveVorticity>true</saveVorticity>
<saveReynoldsStress>false</saveReynoldsStress>
<saveHelicity>false</saveHelicity>
<saveMeshQuality>false</saveMeshQuality>
<saveQ>false</saveQ>
<saveQCriterion>false</saveQCriterion>
<saveLambda2>false</saveLambda2>
<saveJumpEstimator_velMagn>false</saveJumpEstimator_velMagn>
<saveJumpEstimator_pressure>false</saveJumpEstimator_pressure>
<saveJumpEstimator_u>false</saveJumpEstimator_u>
<saveJumpEstimator_v>false</saveJumpEstimator_v>
<saveJumpEstimator_w>false</saveJumpEstimator_w>
<saveJumpEstimator_alpha_air>false</saveJumpEstimator_alpha_air>
<saveJumpEstimator_alpha_vapour>false</saveJumpEstimator_alpha_vapour>
<saveJumpEstimator_density>false</saveJumpEstimator_density>
<saveJumpEstimator_Q>false</saveJumpEstimator_Q>
<saveJumpEstimator_mu_t>false</saveJumpEstimator_mu_t>
<saveCellVolume>false</saveCellVolume>
<saveCellProc>false</saveCellProc>
</extraQuantities>
<monitors>
<monitor name="force_Column_1" >
<MO_Force>
<fileName>force_Column_1</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<families>Column_1</families>
<forTecplot>true</forTecplot>
</MO_Force>
</monitor>
<monitor name="force_Column_2" >
<MO_Force>
<fileName>force_Column_2</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<families>Column_2</families>
<forTecplot>true</forTecplot>
</MO_Force>
</monitor>
<monitor name="force_Column_3" >
<MO_Force>
<fileName>force_Column_3</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<families>Column_3</families>
<forTecplot>true</forTecplot>
</MO_Force>
</monitor>

116 of 118

<monitor name="force_Column_4" >


<MO_Force>
<fileName>force_Column_4</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<families>Column_4</families>
<forTecplot>true</forTecplot>
</MO_Force>
</monitor>
<monitor name="total-forces" >
<MO_TotalForce>
<fileName>forces-total</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<excludeFamilies>FarField Surface Bottom</excludeFamilies>
<forTecplot>true</forTecplot>
</MO_TotalForce>
</monitor>
<monitor name="total-moments" >
<MO_TotalMoment>
<fileName>moments-total</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<excludeFamilies>FarField Surface Bottom</excludeFamilies>
<referencePoint>0 0 0</referencePoint>
<forTecplot>true</forTecplot>
</MO_TotalMoment>
</monitor>
<monitor name="yplus" >
<MO_Yplus>
<fileName>yplus</fileName>
<saveFrequency>1</saveFrequency>
<label></label>
<families>Column_1 Column_2 Column_3 Column_4</families>
<forTecplot>true</forTecplot>
</MO_Yplus>
</monitor>
<monitor name="CourantNo" >
<MO_Scalar>
<fileName>courant</fileName>
<saveFrequency>1</saveFrequency>
<fields>CourantNo</fields>
<forTecplot>true</forTecplot>
</MO_Scalar>
</monitor>
</monitors>
<developer>
<globalMassCorrection>false</globalMassCorrection>

117 of 118

<alternativeLeastSquareGradforHasFaces>false</alternativeLeastSquareGradforHasFaces
<wallCorrection>TYPE2</wallCorrection>
<nonOrthogonalDiffusionCorrection>TYPE2</nonOrthogonalDiffusionCorrection>
<faceInterpolationScheme>TYPE1</faceInterpolationScheme>
<cellCenterScheme>AreaAvCenter2</cellCenterScheme>
<highOrderSchemesHaveFaces>false</highOrderSchemesHaveFaces>
<applyOmegaLimiter>false</applyOmegaLimiter>
<pressurePreconditionerUpdate>1</pressurePreconditionerUpdate>
<fieldInterpolationMethod>BLEND</fieldInterpolationMethod>
<interpolationTermsFactor>1</interpolationTermsFactor>
<interpolationPowerExponent>3.5</interpolationPowerExponent>
<extrakTermMomentumEqs>false</extrakTermMomentumEqs>
<MVG_rotateVelocity>false</MVG_rotateVelocity>
<gradientLoops>0</gradientLoops>
<approxInvMomMat_noDt>false</approxInvMomMat_noDt>
<refrics_noncompressive>false</refrics_noncompressive>
<outputLinSolverInfo>false</outputLinSolverInfo>
<outputDebugInfo>false</outputDebugInfo>
</developer>
</controls>

118 of 118

Potrebbero piacerti anche