See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/49292120

Introducción a los CFD en Ingeniería Naval

Article
Source: OAI

CITATIONS READS

0 56

1 author:

José Enrique Gutiérrez

Universidad Politécnica de Cartagena
3 PUBLICATIONS   0 CITATIONS   

SEE PROFILE

All content following this page was uploaded by José Enrique Gutiérrez on 31 May 2014.

The user has requested enhancement of the downloaded file.

 INTRODUCTION TO CFD IN
NAVAL HYDRODINAMIC: A little
review

José Enrique Gutiérrez Romero

January 17, 2011

Revised by: Julio García Espinosa & Blas Zamora Parra

 1. ¾What is it CFD?

2. Historical blackground

3. ¾Where is CFD used?

4. CFD advantages and disadvantages

5. Approximation to reality

6. Classications of CFD tools

7. Validation of CFD code. The quality importance of CFD´s

8. Some physics and mathematical models employed in CFD analysis.

8.1.- Potential ow model

8.2.- Ideal ow model

8.3.- Rotational, steady and ideal ow models

8.4.- Direct solution of Navier  Stokes equations

8.5.- Parabolic models of Navier  Stokes equations

8.6.- Incompressible ow models

8.7.- Models to solve the turbulence

9. Basic techniques of CFD

9.1.- Boundary Element Methods (BEM)

9.2.- Finite Dierence Methods (FDM)

9.3.- Finite Element Methods (FEM)

9.4.- Finite Volume Methods (FVM)

10. The problems meshing: discretisation of the special domain

11. Basic characteristics of CFD programs.

1.- ¾What are CFD?

The CFD acronym means  Computational Fluid Dynamics . This is a very useful tool
to solve the basics equations which model the ow movement. The majority of these
equations do not have any analytical solution for these reasons we resort to numerical
analysis with CFD. The objective of these tools is to solve approximate (numerically)
the ow basic equations whose solutions give us the movements and other characteristics
of the ow.These techniques discretise the spatial and time domain for reaching the
solution.

3
2.- Historical blackground.

The tools development of CFD started in the 1950s with the development of comput-
ers. The Finite Element Method (FEM) Finite Dierence Methods (FDM) are the basic
techniques to solve the equations raised by the mathematics and physics. Firstly the
Finite Dierence Methods have dominated the calculus because of the Finite Element
Methods were more computationally expensive than FDM and they were more dicult
to formulate than FDM. Nowadays this situation has changed, today new methods and
formulations with FEM have appeared and this situation has caused the obsolescence
of FDM. Thus, other methods have got popularity, for example nite volume methods.
This constituted a natural extension of FDM into non-structured meshes. FVM are
temporally situated before FEM into techniques CFD . These types of methods permit
to integrate the mass and momentum equations in each cell of mesh and they permit to
approximate their value in the center of the cell. Other new methods we nd Smooth
Particles Hydrodynamics (SPH). This method establishes a new eld in existing tech-
niques.

The problem of ship resistance, understanding resistance how the ow eect over the
surface of the body which is submerged in the ow was one of the pillars of genesis of
naval hydrodynamics. The interest in this new problem was caused by its applications in
several elds such as hydrodynamics, windmills and other devices which were submitted
to environmental forces.

The rst approach to explain the body drag in the XVII century followed dierent ways
until the occurrence of the basic equations of the ow movements, although the lack of
calculus capacity kept the improving on this way.

Late of XVII century the  impact Theory carried out. This theory considered two
hypotheses. On the one hand the theory supposed that the uid was constituted by
bunch of individual and independent particles and on the other hand considered that
summation of all the impact in the body was equal of the body drag.

This theory did not explain the several resistance aspects. Later in the middle of XVIII
century Euler established a new technique which was approximate to current theories. In
the same century Jean le Ron d´Alambert exposed a new idea about the ow movement
around the submerged body. In this theory the uid is supposed to be animated with
uniform movement. Then when the ow starts to go up the body, the ow varies its way
around the body. Finally the stream lines of the body will join together upstream.The
ow eect in the body will be made by the pressure rather than the uid impact. However
two premises will be necessary to obtain the uid eld: the continuity and dynamical.
The development of this eld was paralyzed until the beginning of computer.

4
3.- ¾Where are CFD used?
Nowadays these techniques have been in almost elds ranging from medical research
until engineering. In hydrodynamics several important aspects are taken into account:

ˆ Resistance and propulsion.

ˆ Maneuverability

ˆ Seakeeping

ˆ Propeller design

ˆ Others applications: piping. . .

Resistance and propulsion.

The basics applications of these techniques are centered in this eld. Here the engineers
use BEM calculus where the viscous eect and the wake generation in the free surface
are despised. This provides a simple, fast and approximate calculus. At the end of
the 90 years all these simplications were dismissed because of the increasing of the
potential calculus in computers. Nowadays there is a great variety of codes which permits
viscous calculation, this permits to reach best results that are in concordance with real
experiment.

Maneuverability
In this eld, today the new IMO regulations get more and more importance. For this
reason through CFD analysis we can obtain the dierent forces actuating over ruder
surface. This permits to calculate the dierent moments in the ship. Thus we will
evaluate the maneuverability of the ship.

Seakeeping
Here the CFD codes constituted unripe part in the current techniques although today
new codes appear with new features. For several years there have been commercial codes
into BEM eld, one of theses is WASIM. These new codes includes permit to calculate
the ship movement in dierent seas. However the problem of ship Seakeeping presents a
great diculty because it is necessary to provide with all the features of sea and a lot of
Knowledge about this eld will be necessary.. We will be need several input data such
as height of waves, period, and scattered waves, mass added, ship characteristics. . . The
future in this eld is oshores structures. Today the wind energy generation probably
constitutes the principal battle eld of Seakeeping.

5
Propeller design
The usual techniques in this eld are viscous CFD techniques because these permit to
improve the results and permit to approximate to reality. Here employed the basics
techniques are Boundary Element Methods. These usually utilize the supportive prole
theory to calculate the dierent parameters of propeller.

To nish with this part we will probably encounter CFD techniques in other eld, for
example:

ˆ Piping

ˆ Air ow

ˆ Fire protection of ship. . .

4.- CFD advantages versus disadvantages

The advantages of CFD techniques will probably be summarized in the following lines:

ˆ A great time and cost reduction in the new designs

ˆ There is a possibility to analyze dierent problem whose experiments are very

dicult and dangerous.

ˆ The CFD techniques oer the capacity of studying system under conditions over
its limits. These conditions will probably be dangerous to experimental.

ˆ The level of detail is practically unlimited. The experimental methods are more
expensive with the increasing of measure point. However in CFD techniques per-
mits to generate several pieces pf information without any cost. This oers the
possibility to carry along a lot of parametric studies.

ˆ The product gets added value. The possibility to generate dierent graphs permits
to understand the features of result. This encourage to buy a new product.

On the other hand, all on these tools are not advantages. The principal disadvantage
of CFD tools is the great knowledge about the basic equations that describe the uid
movement and certain physical phenomenon necessary to solve a new problem.

ˆ Other disadvantage is the accuracy in a result. In several situations probably we

will not obtain a successfully result. In these cases the mistakes will be important.
Even though if the user is untrained.

ˆ It is necessary simplify to mathematically the phenomenon to facilitate the calcu-

lus. If the simplication has been good the result will be more accurate.

ˆ Today there are dierent incomplete models to describe the turbulence, multiphase
phenomenon and other dicult problems.

ˆ Finally, it is important to mention that the untrained user of CFD codes has the
tendency to believe the output of PC is always true.

6
5.- Approximation to reality
The idea about this concept results very clearly. They are the dierent approximations
made by engineering to solve the uid equations.

Once known the dierent equations of the physic phenomenon to study it will be nec-
essary to introduce all the possible simplications in a new problem. This helps us
obtain the best result with the less cost and diculty. So, we don't understand the
analytical solutions of the problem under study previously for these do not reasons the
simplications are very dicult to introduce in the new problem.

It will also be necessary to dispose the equations of the physical phenomenon to make
the model about the problem. Most of these equations are dierential equations with
nonlinear terms and other are partial equations. All of these equations will keep o to
obtain an easy solution of the problem. Other important aspect to obtain an accurate
solution of the problem is the spatial discretisation of the same. The st step will always
be the problem discretisation. Firstly the geometry will be divided in a lot of little
parts, this permits to solve the problem easily. Although the uid eld is continuous,
the equations are not be solved in all the uid point, so, for this we need to discretise the
entire domain in several points where the equations will be solved. This step is named
as spatial discretization .
In other way when the phenomenon is constituted by time dependent equation neither it
is possible to study the continuous phenomenon. In this situation the engineer establishes
a temporal discretization .
Finally in this part in government equations all of the unnecessary terms will be elimi-
nated . So, we obtain a new level in approximation in a phenomenon called dynamical
discretization .

7
6.- Classications of CFD tools
Into the naval eld, the CFD tools can be classied in two dierent groups: those with
potential ow, to say the tools which do not consider the uid viscosity. We also have
two more types: with or without free surface into these types. On the other hand we
encounter other tools which consider the uid viscosity. Here also we nd the same
types: with or without free surface.

7.- Validation of CFD code. The quality importance

of CFD´s
Today the quality concept reaches the computer systems and particularly the computer
programs. After reached and promoted the quality concept into  grey material in
the hardware through ISO rules, for this it is necessary to face the problem of CFD
validations, mainly in those tools which are designed to hull and propeller optimization.

The CFD validations could be understood as a demonstration which permits to verify

the result obtained in a computer. This result should be similar to experimental results.
In this way the best orientation to show the goodnesses of CFD tools is an experiment.

ˆ Functionality : The CFD programs must provide the information using the same
parameters and nomenclature which are compatible with the current techniques
employed in industry.

ˆ Reliability: These types of codes have to calculate the result into a condence
range and they should reach an acceptable precision level.

ˆ Usability: This feature evaluates the capacity to use the program, to say the
facility of user to understand the capacities of program and the logical of his code
in the inputs and outputs.

ˆ Eciency : The eciency of the program should be measured not only in time
wasted for the CPU and GPU but also in all the human resources employed to
obtain a good result.

ˆ Maintainability: This part tries to show the short life of these codes and the
necessity to update these. These updates permit to be suitable in current tech-
niques.

ˆ Portability: A CFD program should be installed in dierent operating system

and give us the possibility to transfer.

8
8.- Some physics and mathematical models employed
in CFD analysis.
First of all we can say that the momentum equation and mass conservation equations
are more complicated than it seems. The experience has demonstrated that the Navier
 Stokes equation describes the movement of uid accurately, however today only a few
problems have been solved in simple geometries. This only has a few applications in the
industry.

8.1.- Potential ow model

These models describe the behavior of an ideal problem. In this case the basic theory
to calculate the features of uid starts with the velocity potential and it establishes an
added simplication to permit to calculate steady ows. So into these models we nd
dierent developed methods used by dierent authors. Michel in 1898 obtained some
formulation to calculate the wave making. On the other hand Haverlock in 1932 also
gave a solution to get the wave resistance. Other authors such as Hess & Smith provided
us with method to obtain the potential velocity. Theses methods establish the following
problem:

∇2 φ = 0 (Laplace equation)

To solve this problem it will be necessary to apply a boundary condition in a free surface
of uid and in ship hull.

Boundary condition one:

(
∇2 φ = 0
⇒ ∇2 φp = 0
φ = Ux + φp

In a ship hull we applied the following boundary condition:

∂φ ∂φ ∂x
Vn = |s = 0 |s = −U |s
∂n ∂n ∂n
Kinematic boundary condition:

 
∂φp ∂h ∂φp ∂h ∂φp
U+ + · + (−1) = 0
∂x ∂x ∂y ∂y ∂z

Dynamical boundary condition:

9
 " 2  2  2 #
ρ ∂φp ∂φp ∂φp 1
p+ U+ + + + ρgh = patm + ρU 2
2 ∂x ∂y ∂z 2

8.2.- Ideal ow model

When the Reynold number is suciently high in some problems of uids mechanics it is
easy to despise the viscous eects and other conduction eect to facilitate the resolution
of equations. Thus we can solve more easily the equations because we eliminated second
order diusive terms in the dierential equation. This permits to obtain rst order
equations which facilitate the resolution of the problem.

With the last hypothesis, we obtain the equations of ideal ow which are called Euler
equations. The models of Euler equations take the following expressions:

Continuity:

dρ
+ ρ∇ · →
−
u =0
dt
Momentum:

∂→
−u
ρ + ρ · (→
−
u · ∇) →
−
u = −∇p + ρfe
∂t

8.3.- Rotational, steady and ideal ow models

These types are very similar to ideal models. Here the engineering tries to reduce the
number of variables which are involved in adding the vorticity of momentum and energy
equations. Here we do not consider either the waste by viscosity in the boundary layer
or turbulence eects.

8.4.- Direct solution of Navier  Stokes equations

The Navier- Stokes equations constituted the correct ow modeling of Newtonian uid,
the viscous and term eect are included in the same equations. If the equations are
perfectly solved the eect of turbulence and boundary layer are added in the solution.
However to obtain a correct solution the smallest spatial and temporal discretization
will be necessary. This means that is impossible in the current industry. Today this is
possible in the supercomputers and the solution will probably be obtained in the future
in a normal computer.

Now the numerical solution is possible today if suitable models are used to solve the
equations and simulate the turbulence eect and boundary layer.

10
8.5.- Parabolic models of Navier  Stokes equations
The parabolic models are developed to model and calculate supersonic and hypersonic
ows where the capture of shock waves, pressure gradient, supercial viscous eorts and
heat transfer are the most important objectives in industry.

The parabolic government equations are obtained using the Navier  Stokes equations
and considering the following features:

1. Steady ow

2. The viscous eorts gradient are despised in the stream lines direction.

3. The pressure gradient are approximated with their values in the nearest boundary
layers.

8.6.- Incompressible ow models

One uid is incompressible when it s density is constant and the pressure changes and
time independent. When the uid is isotherm, the government equations are notable
simplied and the solution of several variables becomes temperature independent. The
equation system required is reduced to continuity equation and momentum equation
which are expressed with dimensionless term. Theses adopt the following expressions:

Continuity:

∇·→
−
u =0

Momentum:

d→
−u ∇p 1 2→ 1 →
=− + ∇−
u + −
g
dt ρ Re Fr

In this way; the continuity and momentum equation are energy independent, for these
reasons there is not any necessity to solve the equations to obtain the velocity and
pressure elds. However in spite of this advance, the equations become too  rigid and
in fact the solutions of these are more complicated than others.

8.7.- Models to solve the turbulence

The Reynold number gives us the importance of inner forces associated with convective
eects and viscosity forces. We know before a less Reynold number the critic ow is
steady and the adjacent ow layers slide on each other in order. In this case the ow is
called laminar ow .

11
If the ow reaches and passes a critical Reynold number, several disturbances will be
show which will cause a radical change of the behavior of the ow. The movement
will become unsteady even if it has a steady boundary conditions. This ow is called
turbulence ow .
The state of ow movement when the ow condition and its properties (for example:
pressure, velocity...) uctuate untidily is dened turbulence. This is a unsteady state
from a macroscopic point of view. The values of the ow properties vary untidily and
randomly and they depend of the velocity, pressure, time position...

The movement description of the ow particles is highly complex because of turbu-
lence. It also constitutes a problem without solution from the point of view of numerical
methods. So, the turbulence simulation is one key of CFD simulation:

Direct simulation (DS).

This method does not use any turbulence method, however it uses temporal and spatial
discretisations which are capable of simulating a concrete problem. Today the study of
the direct simulation of Navier-Stokes equations is accessible for limited cases. The rst
solution of this type was achieved in Stanford University in 1981.

Large Eddy Simulation (LES).

These types of numerical techniques reduce the complex of the government equations.
These only consider one of part of turbulence ow. The energy exchange into the  large-
scale uctuation and the eect of little eddies of turbulence are studied into these .
These types of models are situated between Direct Simulation and Reynold Averaged
Navier Equations. The last type extends the temporal average into specic and ba-
sic turbulence eects. Although these types of turbulence models are used in simple
geometries because they require a high level of calculus capacity.

Unsteady Reynolds Averaged Navier Stokes

(URANS).
These types of model are similar to RANS methods, however these types of methods
work with transitory scheme instead of RANS methods. Thus the solution of the problem
depends on the time. Today URANS methods are suering a great development.

12
 Reynolds Averaged Navier Stokes (RANS).
The RANS methods have been widely studied and they are very useful to solve the most
of practical problem solved with numerical methods.

The procedure of average the movement law of the ow particles is introduced to obtain
the averaged and random behavior of the several variables. The starting point into these
methods is very easy. The procedure tries to split the variables into average values and
random values. So, there are three types of RANS: time averaging, spatial averaging
and ensemble averaging.

Time averaging is appropriated for stationary turbulence, i.e. a turbulence ow that, on
the average, does not vary with the time. For such a ow, we express an instantaneous
ow variable as f(x,t) . Its time average, FT (x), is dened by:

´ t+T
FT (x) = limT →∞ T1 t
f (x, t)dt

Spatial averaging is appropriated in homogeneous turbulence, which is a turbulence ow

that, on the average, is uniform in all directions. We average over all spatial coordinates
by doing a integral volume. The spatial averaging, FV (x) is dened by:

˝
FV (x) = limT →∞ V1 f (x, t)dV

Ensemble averaging is the most general type of averaging. As an idealized example, in

terms of measurement from N identical experiments where f (x, t) = fn (x, t) in the nth ,
the average FE where:

PN
FE (x, t) = limN →∞ N1 n=1 fn (x, t)

When the turbulence is homogeneous and steady, we may assume that these three av-
erage are equal. This assumption is known as ergodic hypothesis.

For example, the descomposition of velocity will be:

→
− →
−0
u =→
−
u +u

Where middle component of velocity is obtained making the integral of instant velocity:

ˆ T
→
− 1 →
−
u = u (t)dt
T 0

Supposting, that the integral period is large enought compared with temporal turbulence
scale, but little enought to capture any unsteady phenomenon dierent from turbulence.
The utilization of these methods is suitable, so the most of unsteady phenomenon in
Fluid Mechanics has been placed out of range of normal range of turbulence.

13
The process of time averaging of dierential equations results into several terms denom-
inated Reynold Stresses, which involve averages of the products of uctuating velocity
components. To obtain the relationship is to necessary introduce a new model, denom-
inated turbulence models or closure model. The dierent practical possibilities of these
models will be analiced in the following paragraphs.

So as, the Navier Stokes equations are averaged above scales of the turbulence uc-
tuations (RANS). These methods cause a eld of averaged turbulence and simulated
which is more uniform than real eld, therefore, the number of points of the spatial and
temporal discretization necessary to obtain the variables are reduced highly.

So a turbulence model is a numerical procedure to obtain the averaged values of variables

of uctuations. This permits solve the government equations more easily. If a turbulence
model is exact, simple and economic, the turbulence model will be useful into a general
CFD. The most common turbulence models are the following:

RANS-based turbulence models

Algebraic models.
1. Cebeci-Smith Model

2. Baldwin-Lomax Model

3. Johnson-King Model

4. Rought wall dependent Model

5. Mixing Length Model.

One equations models.

1. Prandt Model

14
 2. Baldwin-Barth Model

3. Spalart-Allmaras Model

Two equations models.

1. k-epsilon Model

2. k-omega Model

Two equations models with restrictions and limits

v2-f Models (Non linear eddy viscosity models)

Reynolds Stress Models

Large eddy simulation (LES)

1. Smagorinsky-Lilly model

2. Dynamic subgrid-scale model

3. RNG-LES model

4. Wall-adapting local eddy-viscosity (WALE) model

5. Kinetic energy subgrid-scale models

6. Near-wall treatment for LES models

Others models
Detached eddy simulation (DES)

Direct numerical simulation (DNS)

Turbulence near-wall modeling

Turbulence free-stream boundary conditions. Then we will deepen in to the most

common and known turbulence models.

15
Algebraic Models
The algebraic models of turbulence also called models of cero equations.These types
of models do not need additional equations and they are directly solved over the uid
variables. They utilize a Boussinesq approximation to obtain the Reynold Stresses. In
these models the eddy viscosity and mixing length depend on the uid and to solve
them will be necessary to specify these two variables. For this reasons theses models
are not suitable in the most of applications because of they will consider convection and
diusion of turbulence energy. The algebraic models are very simplex , so the user of
CFD should take care to utilize these models in complex geometries and process because
the result probably will not be valid, for this reason they are not valid to use in general
analysis with CFD.

Prandt Model

In 1925 Prandt proposed the hypothesis and bases to develope this new method. Prandtl
formulated a turbulence model whereby the particles coalescence and joined a group of
particles which moved as a singleness. He also visualized that the shear ow has a
linear prole and the groups of molecules kept their momentum in the X direction to a
distant in a Y direction. This direction was denominated how mixing length lmix . So
he postulated that:

1 dU
τxy = ρumix lmix
2 dy

where mixing velocity take the following expression:

 
 dU 
umix = Constant · lmix  
dy

We can say that lmix does not a physical property of uid and grouping terms, we obtain:

dU
τxy = µT
dy

Where µT is called eddy viscosity:

 
2
 dU 
µT = ρlmix  
 dy 

K-Energy Models
Into this point we can include both one equation model and two equation model. The
two models keep the Boussinesq approximation to viscosity but these models dier in

16
an important aspect, while the one equations models are incomplete because they link
turbulent length scale with a typical dimension of uid. On the contrary we can observe
that the two-equation models have one more equation than one-equation models to
establish a turbulent length scale, this is equivalent to present a more complete model.

The equation of turbulent kinetic energy. This equations was developed to incor-
porate the non local and historical eects of the ow over the eddy viscosity. So Prandtl
chose kinetic energy of turbulent uctuation into dierent models as basis of velocity of
scale.

1 1
k = u2 = (u2T + vT2 + wT2 )
2 2
So, the eddy viscosity can be established:

µT = C · lmix ρk 1/2

On the other hand, when we study the equation of Reynold Stresses:

 
∂k ∂k ∂Ui ∂ ∂k 1 0 0 0 0
ρ + ρUj = τij − ρ + · µ − ρuτ uτ uj − puj
∂t ∂xj ∂xj ∂xj ∂xj 2

We can see several terms which show the processes that occur in a turbulence ow. We
may classify into following groups: production terms  k , which show us the percentage
released into the turbulence of kinetic energy from main ow. Dissipation terms of   ,
which show us the percentage of kinetic energy converted into thermal internal energy.

Two equation models These types of models have been widely studied during 80th
and 90th decades, the most of studies made during these years have been made with
these models which permit more accuracy into complex ow but they require high com-
putational power. However It isn't any problem because today the computational de-
velopment is high.

The two equation models provide not only one equation to solve kinetic energy but
also they give us other equation to solve the mixing length. These models of turbu-
lence permit to predict turbulence ow more accurately. The starting point of these
methods we have the Bousinessq approximation together with the equation to kinetic
energy. Kolmogorov in 40th decade dened the second variable. This variable was called
−1
dissipation  w . This quantity has dimension of t , so the values of turbulence length
scale, eddy viscosity and turbulence dissipation can be determined by:

ρk k1/2
µT = ω
l∼ ω
 ∼ kω

17
We can note that this type of turbulence models is not an universal tool to obtain the
turbulence characteristics on the ow, so, depending of their use the results of these will
be more or less acceptable.

In the following steps we are going to center into the most utilized two equation model,
there is a k−ε model.
Close models or numeric strategy to solve approximate the Navier - Stokes equations
have developed two transport equations more, one to turbulent kinetic energy (k) and
the other to turbulent dissipation kinetic energy (). These variables are dened by
following expression:

1 1
k = u2t = (u2t + vt2 + wt2 )
2 2

 = 2vetj etj

where etj is a uctuating part of strain rate tensor.

The transport equation to ”k” and ”” are based in the knowledge of process which
produce the changes in these variables and they are:

 
∂ρk µT
+ ∇(ρku) = 2µT Eij Eij − ρ + ∇ grad k
∂t σk
 
∂ρk   µT
+ ∇(ρku) = C1 2µT Eij Eij − C2 ρ + ∇ grad k
∂t k k σk

where Eij is a averaged component of strain rate. The physic meaning of last expression
can be summarized into the following balance:

[V elocity change of k/] + [k/ transport by convection] =

= [k/ production] − [destruction of k/] + [k/ transport by dif f usión]

Here they appeared several kinematic concept connected with scales of typical length,
which are associate with ow movement. So, in this turbulence model the turbulence
length scale can be shown for the following expression:

k 3/2
l=

This method utilizes ””of little eddies to dene length scale ”l” of big eddies, because
of the extraction energy velocity of big eddies is equal to transfer energy velocity of little
eddies into high Reynold numbers. If this weren't so, the energy of certain scales may
increase or decrease without limit. This does not ocurr in the reality, so, in this way we

18
can justify to use the dissipation ”” into the length denition ”l”. Applying the same
approximation to mixing length, the eddy viscosity can be obtained like as:

k2
µT = C µ ρ

The two equation models have the next advantages:

ˆ It is only necessary to x initial and boundary conditions.

ˆ Good results are obtained to great variety of ows.

ˆ Two equation models are the most turbulence methods utilized in the industry
applications.

ˆ Development wall laws are disposed to these types of models

On the other hand they also have disadvantages:

ˆ Their implementation is more complex than algebraic models because of introduc-

tion of extra dierential equations.

ˆ Poor results are obtained in some cases: conned ows, ows with high longitudinal
gradients...

A variation of k− model, denominated k − ω, was published by Wilkox in 1993. This

method removes one variable dividing it all the rest of the problem, simplifying the
solving of the problem. This method has the objection, it shows singularities, when the
variable which divides the rest of the problem is zero. For this reason k−ω is more
suitable into aeronautic eld than into naval eld.

Modeling of boundary layer

The boundary layer is part of the uid eld close to solid contour where the viscous
phenomenon is shown of specially way. Close to any solid contour a velocity gradient
appears in the normal direction of this, due to viscosity and non-sliding condition. The
velocity gradient determines the energy exchange between dierent uid particles. This
phenomenon causes vortex and turbulence.

The basic problem of numerical modeling of the energy exchange in the boundary layer
consist into making a correct denition in the velocity of the particles close to boundary.
It requires a high density of the mesh. It is necessary to capture the local phenomena
which are caused into boundary layer. To make this calculus requires high computational
spending. Sometime this isn't possible to make it because of computational limits.

These diculties have been solved by means of approximations. These approximations

will be classied into four groups: models of distributed lost, models of shear layer,
models of boundary layer and wall laws.

19
To establish these approximations, the region close to boundary layer is characterized
by dimensionless variables which permit to dene the wall conditions. Whereas, these
new variables are dened by means of friction velocity terms, which depends on shear
stress in the wall.

We call ”y” to dimensionless distance to wall and ”U ” to time averaged velocity, result-
ing:

U uτ
U+ = y+ = y · ρ ·
uτ µ
In this way, we achieve that if the ow depends on the wall condition, we can obtain the
+ +
universal functions U and y . These functions will be obtained into the limits which
permits obtain any characteristics of the boundary layer, in fact it is proved that the
relation between these variables exists.

The standard functions of the wall are valid to smooth walls, but they can be modied
to include the wall roughness by means of adjusting the constants. If the wall roughness
is modeled, the distant to wall is not dimensioned by roughness equivalent height. By
this way, some codes include a specic law of the wall to modeled this eect. (Rough
wall eld )

The universal function can be used to link the variables in a rst computational mesh.
Shear stress can be obtained directly without solving the middle structure.

The standard functions in the  law of the wall are a source of false ideas in the solving
of turbulence ows even to expert users in CFD. To obtain a thin layer close to surface
is the purpose of these models.

Now, we can note the restrictions when we use the law of the wall:

ˆ The ow solved can agree with the hypothesis made to achieve the law of wall
equations.

ˆ It is necessary remark the y+ into the rst points in the mesh to validate the wall
functions.

ˆ The functions of the wall do not release of the need to obtain a suitable solving of
the turbulence into boundary layer.

ˆ It is dicult to generate a predetermined distribution of y+, because it depends

on the solution of the problem.

Thus, some conditions will be taken into mind in the modeling of turbulent boundary
layer:

ˆ Check the lower limit of y+, which is situated around y + = 20 − 30

ˆ Check the upper limit of y + . If the Reynold number is moderate and the boundary
+
layer is situated between y = 200 − 500, we can advise that there isn't the
opportunity of solving of boundary layer with precision. If the rst integration
+
point is placed in value above y = 100, this fact occurs.

20
 ˆ Check the solving of boundary layer. If the eects of the boundary layer are
important, It is suitable to check the solution of the boundary layer after solving.

ˆ Check the wall functions to verify the roughness in the functions.

9.- Basics techniques of CFD

Boundary Element Methods (BEM)
These techniques are basically employed to solve the potential ows. Thus in several
practices hydrodynamics techniques, the BEM methods are called panel methods. In
these methods the hull surface is splitted into little panels where each panel satises the
Laplace equation∇φ = 0.

Finite Dierence Methods (FEM)

The nite dierence methods dominate mainly in the structural calculus. This method
occupy a less important role In hydrodynamics calculus. Now we present this method
slightly.

The FEM method is a procedure to solve continuous problem by means of dividing the
system in a nite number of parts (elements), whose behavior is dened by a nite num-
bers of parameters. Thus the solving of complete problem is made through assembling
the solution of singular elements.

To analyze the behavior of the continuous structure It is necessary to proceed by means

of this way:

ˆ The continuous system is divided, by means of imaginary lines and surfaces in a

discrete system.

ˆ It is supposed that the elements are connected themselves through a discrete num-
ber of points, called nodes which are situated in a boundary. The displacements
in nodes will be the unknown factors of the problem, similar to discrete problems.

ˆ It is chosen a bunch of functions which permit to dene uid eld in a only way
into each element depending of boundary conditions.

ˆ From here the problem is similar to a discrete problem, i.e, the matrix of each
element is assembled, the boundary conditions are imposed and the equilibrium
equations of uid eld are solved, thus the velocity vectors, stresses,... are deter-
mined. However in hydrodynamic several acceptances in elementary calculus and
elementary functions can not be used to obtain the error of integrals by the same
way of Galerkin methods.

21
 ˆ Then it is determined a system of concentrated forces in a nodes which balances
the stresses in the boundary and whatever distributed loads. Thus, it is obtained
the relationship between the forces and displacements, which have the following
form:

q e = K e ae + f e

being q e the e e
vector of forces which act in the nodes, K the stiness matrix, a the
e
vector of nodal displacements, f the vector of nodal forces to balance the mass
forces, the initial stresses, the initial distortions and the stresses applied in a
boundaries.

ˆ With the stiness of each element, the stiness matrix is assembled. It is applied
the boundary conditions and it is established the global equation system. By means
of this system they are obtained the global displacements. Finally the stresses are
calculated in each element from these displacements.

The method commented is usually called the displacement method. So, today we can
nd a variety of approximations to nite element method . In the following scheme they
are showed dierent ways and theories to approximate the solution with this method:

Finite Volume Methods (FVM)

This method is similar to the last methods, it is utilized to temporal and spatial dis-
cretization of the current problem. However this method integrates the momentum and
mass conservation equations over the cell before approximating the value in the central
node. Thus the made error in the output face is canceled by the made error in the
input face of the cell, this permits to the mass and momentum conservation. Several
commercial codes employed this method to solve the equations.

22
10.- The meshing problems: discretization of the
special domain
The codes based in FEM or FVM normally are formed by three parts: pre-processor,
calculus module and post-processor. The pre-processor is the responsible of meshing the
problem. Thus the meshes of the problem can be classied into several types according
to two criteria: the manner of dening the boundary of the scope, and the connectivity
between the mesh points.

In terms of how we dene the boundaries, they exist the compliant meshes (theses do
not interpolate in the boundary denition) and the non compliant meshes (they do not
dene the real boundary but they dene a numerical approximation of the same).

According to the data's structure, they are two methods of mesh generating which create
two types of primary meshes: Unstructured and Structured. In the structured meshes,
the points are placed according to a family of coordinate lines which permit a direct
display the relationship between the points. This fact permits a simplication respect
to unstructured mesh.

Whatever the case ( may be unstructured or structured ), the mesh should achieve of
the following features:

ˆ It is necessary to adjust the boundaries of studied regions to achieve the best

accuracy in the representation of the boundary conditions.

ˆ The mesh can be distributed locally in a regular way as possible. It has a soft
variations of density. The mesh density is dened how the number of points per
surface or volume unit. It is basic to select a good mesh density to solve the
problem suitably.

ˆ If the mesh is rough, the accuracy in the solving of the problem is not acquired.

ˆ If the mesh is ne, the computational cost probably will be disproportionate.

ˆ The mesh density should be placed in the locations where it will be expected high
spatial variations.

ˆ The mesh should adjust dynamically to changes of the ow variables.

The diculties shown in the mesh generation are usually the following:

1. There is a excessive distortion in the elements. In a Finite Element Method where

they are used basic element (uni- bi- and tri- dimensional) with simple forms (
lines, triangles, rectangles, cubes...) which are converted in other with arbitrary
forms.

2. There is a incorrect mixed of the element. Most of programs based in FEM, give
to user the facility of choosing the type of element, however all the elements are
compatible themselves.

3. There is a incorrect connection between element s. There is convergence in FEM

whenever Cm−1 exists continuity, being m the functional.

23
It is basic to choose a suitable mesh density to solve the problem.

The user has to know the response of element and the approximations made in the
problem.

The linear elements require nest mesh than quadratic form elements and these require
more than the cubic form elements.

STRUCTURED MESH

Advantages.

ˆ The structured mesh leads to a simple solving.

ˆ It is required less memory to store the variables.

ˆ The implicit methods utilize structured mesh: ADI, linear relaxation methods,
relaxation in the plane...

ˆ The smoothness and orthogonality can be easily controlled.

Disadvantages.

ˆ The use of complex geometries is not easy and exible.

ˆ The adaptability only is possible through adding or moving lines in a 2D or 3D

mesh.

ˆ The boundary's movement is dicult.

UNSTRUCTURED MESH

24
Advantages

It exists easily to handle complex geometries with adaptability and boundary movement.

The automatic mesh generating is easier, even with 3D complex geometries.

Although we expect a solution with low accuracy, it is not true. We can obtain a high
accuracy because the mesh can adapt to solution.

Disadvantages.

It requires a great storage capacity for the variables.

The resolution of the problem is more dicult.

MULTIBODY MESH

We can note that the multibody meshes are generally non compliant, i.e. the nodes do
not match in the blocks joints.

Advantages.

ˆ It is possible to handle complex geometries, so, the exibility increases.

ˆ This permits to block with dierent sizes.

Disadvantages.

ˆ The mesh generating is usually dicult. The specic way to show boundaries
between the blocks in suitable way, requires a high experience in their use.

ˆ This types of mesh are less exible than unstructured meshes when There is move-
ment in the boundaries.

25
11.- Basic characteristics of CFD programs.
In this last chapter we are going to deal about what types of interface CFD oers us
from a computer user point of view.

Today There is a great variety of CFD codes, from free codes to powerful commercial
CFD codes. We can join these codes in some groups:

General Codes: Into these types we can nd the codes that have the following tools
at the service of the user to: to generate geometries, to apply boundary conditions, to
discretise the spatial domain (generally this step is called mesh generating), solve the
problem and show the results. Examples of these are: Tdyn, CFX, Fluent, Phoenix,
Open FOAM ...

Specic Codes: They are the allocated codes to a concrete part of the CFD applica-
tion. Into these codes we can nd, allocated codes to airfoil design, allocated codes to
pneumatic design...

Pre and Post processor: They are allocated codes to mesh generating, to assign
boundary conditions, to geometry generating, to take care of specic calculus or allocated
codes to show the results. Into the previous examples we should say that the typical
CFD codes are those allocate to specic calculus. Similar to general code here there is
a great variety of codes:: ParaView, OpenDX , MayaVi Data Visualizer ,GMV ( The
General Mesh Viewer), FAST, pV3, Visual3 y ChomboVis, VU, CAF2D , FEMFlow and
CAF3D, PostFlow, VIGIE, DAVID, Se.La.Vi., HIGHEND (Interactive Graphics using
Hierarchical Experimental and Numerical Data), NCSA UNIX Products. . .

Mesh generating codes: They are the allocated codes only to mesh generating. Some
examples of them are: ADMesh, ANGENER, AUTOMESH2D, CAF2D GENGRID,
CAF2D GENMESH, Cart3D, CGM, Chimera Grid Tools, COG 2. . . .

Then, we are going to show the followed process by the user to make an analysis with
general CFD codes. Into these types of codes, the experience and knowledge of the user
are very important to get a good result in the analysis.

The general codes usually are divided into three parts: preprocessor, a calculus node
and postprocessor.

ˆ The preprocessor is that part of the CFD codes, allocated to make all previous
work necessary to solve the equations of the specic problem.

ˆ The calculus node is the part of CFD codes, whose function is to solve numerically
the equations placed by preprocessor.

26
 ˆ The postprocessor is the other part of the programs, which is allocated to show
the results in a graphic form or similar.

Calculus process followed by the user.

The rst part of this calculus process consists of geometry generating, in what concerns
us, the hull of the ship. It is usually imported from other platforms to CFD codes by
means of dierent exchange les: iges, 3dm, dfx, aciis, parasolid... So, it will only be
necessary to create a control volume for solving the studied problem. In our case, in
the ship, the control volume should be a parallelogram formed by two parts: one for the
water and another for the air.

Then when the user already has a dened geometry with volume control, he passes to
assign the boundary conditions necessary to solve the problem. These types of conditions
may be Neumann conditions, Dirichlet or mixed. These conditions will be such as: the
velocity of the uid in the boundary, boundary layer, pressure elds or stresses, wave
spectrum...

The following step to be made by the user. This step is very important, depending on
the accuracy in a spatial discretisation, the results will be more or less accurate. In this
step the user creates a mesh over the geometry. If the mesh is very ne in the critical
regions of the geometry, the result will be better. However, It is not possible to get the
nest mesh, because it depends of the computer capacities. If the mesh is very ne, the
calculus will probably last so long.

Then when the problem is meshed, the following step will be the calculus. Finally the
user will show the results. In this part the user can obtain a great variety of graphic
results from stream lines to interactive videos. Most of codes permit to export the results
from these codes to other platforms. By this way the user will can work with the results
in other platforms.

27
Bibliography

[1] R. Ballesteros, J. González, J.M. Fernández, K.M. Argüelles. Técnicas Numéricas

en Mecánica de Fluidos. Universidad de Oviedo.

[2] Volker Bertram. "Practical Ship Hydrodynamic." Butterworth Heinemann.

[3] David C. Wilkox. "Turbulence Modeling for CFD". DCW Industries Inc.

[4] Servicio de Apoyo a la Investigación Tecnológica. El método de los elementos nitos
y sus aplicaciones en ingeniería. UPCT

[5] O.C. Zienkiewicz - R.L. Taylor. El método de los elementos nito. Las bases-
Volumen 1. 5ª Edición. Cimne. Barcelona.

[6] E. Oñate. Cálculo de Estructuras por el Método de Elementos Finitos. Análisis

estático lineal. Cimne. Barcelona.

[7] Presentación ETSIN "Teoría del buque"

[8] Manual Tdyn 6.0. Compass Ingeniería y Sistemas.

[9] Gid, User Manua, Version 8. Compass Ingeniería y Sistemas.

[10] TdynTurbuelenceHB. Compass Ingeniería y sistemas

[11] Departamento de Estructuras y Construcción. "El método de los elementos nitos

en ingeniería". 4º Curso de Ingeniería Industrial UPCT.

[12] Luis Pérez Rojas." Los CFD como herramienta en Hidrodinámica". AULAS DEL
MAR. UPCT

[13] L. Pérez Rojas. A. Souto Iglesias. P. Roca Fernández. "Los CFD en la formación
hidrodinámica del ingeniero naval". IV SIMPOSIUM PANAMERICANO DE ED-
UCACIÓN EN INGENIERÍA NAVAL.

[14] J. Valle Cabezas. "Ayer, hoy y mañana de la Hidrodinámica numérica". conferencia

pronunciada con motivo de la celebración del 75 aniversario del CEHIPAR.

[15] http://www.cfd-online.com

29

View publication stats