Practical Guidelines For Ship CFD Applications-ITTC PDF

ITTC – Recommended 7.
5 – 03
02 – 03
Procedures and Guidelines Page 1 of 18
Practical Guidelines for Ship CFD Effective Date Revision
Applications 2011 01
Table of Contents
Practical Guidelines for Ship CFD 2.3.5 Non conformal mesh 8

Applications 2 2.3.6 Expansion ratio and number of grid
points in boundary layer 9
1. OVERVIEW 2
2.3.7 Grid skewness 9
2. PRE-PROCESSING 2
2.4 Boundary conditions .................... 10
2.1 Problem characterization .............. 2
2.1.1 Resistance 2 2.5 Choice of the time step ................. 10
2.1.2 Wall function 3 2.6 Choice of convergence criteria .... 11
2.1.3 Surface roughness 3 2.7 Choice of free surface model ....... 11
2.1.4 Incident waves 4 2.8 Choice of turbulence model ......... 12
2.1.5 Motions 4
2.9 Choice of numerical scheme ........ 13
2.1.6 Flow features 4
3. COMPUTATION 14
2.1.7 Region of influence 4
4. POST-PROCESSING 14
2.2 Geometry creation and
modification .................................... 4 4.1 Visualization.................................. 14
2.3 Grid generation............................... 5 4.2 Verification and Validation ......... 15
2.3.1 Definition of the domain 5. USEFUL WEBSITES AND
boundaries 5 REFERENCES 15
2.3.2 Element type 6 6. EXAMPLE FROM G2010 WorkSHOP 16
2.3.3 Grid points 7
2.3.4 Grid topology 8
Edited /Updated Approved

26th ITTC Specialist Committee on CFD in
26th ITTC
Marine Hydrodynamics
Date 05/2011 Date 09/2011
ITTC – Recommended 7.5 – 03 – 02 – 03

Application 2011 01
Practical Guidelines for Ship CFD Applications
Fr = U / �𝑔𝐿pp
1. OVERVIEW
where ρ and µ are the density and viscosity of
These guidelines are written assuming the the fluid respectively, U is the ship speed, LPP is
use of surface capturing methods, the method the length between perpendiculars of the ship
found in most commercial and academic CFD and g is the acceleration due to gravity. Note
packages. It also assumes that the solver is grid- that the viscosity of water varies with tempera-
based, as opposed to mesh-free methods. ture so that model scale tests carried out in wa-
We divide the CFD process into pre- ter at 10-15ο C is different to the full scale vis-
processing, computation, and post-processing cosity of deep sea-water at 15ο C.
steps. The pre-processing step involves proper Estimate the expected free-surface wave-
definition of the problem, grid generation and length and elevations. Estimate the distance to
input setup to enable running of the computa-
the wall where y + <= 1 for near wall boun-
tional code. The computation requires preparing
dary conditions and 30 < y + < 100 for loga-
the computer to run the problem, and running. rithmic wall functions. This estimation of the
The processing of the results of the computation distance to the wall should be based on an esti-
to provide useful numbers and plots is called mation of the skin friction as a function of the
post-processing. Reynolds number. This distance is described in
terms of the non-dimensional parameter y+. This
2. PRE-PROCESSING can be defined in terms of the Reynolds number
of the required flow as follows:
2.1 Problem characterization
y/LPP = y+/ (Re �Cf/2)
2.1.1 Resistance CF= 0.075/(log10Re - 2)2
Define the Reynolds and Froude numbers. where y is the first required cell height and CF is
These are given by: an estimate of the skin friction coefficient,
based on the ITTC standard method. This gives
Re = ρ U LPP / µ

Application 2011 01
an estimate of the skin friction coefficient at However, wall functions are based on two-
mid-ships. dimensional flow, typically at zero pressure
gradient, and it is well known that the validity
Note that in the calculation above, CF of these analytical expression becomes less, or
should be calculated using the full length of the even disappears, with increasing adverse pres-
ship which in the approach gives an approxima- sure gradients. Thus it cannot be expected that
tion to mean value of the local CF The ITTC the wall function approach leads to reliable and
model-ship correlation line should only be used accurate solutions near a ship stern, where the
in this estimate. flow is strongly three-dimensional and running
up against an adverse pressure gradient. Thus it
2.1.2 Wall function is a trade-off between accuracy and com-
putational effort.
To resolve the high velocity gradients in the
inner part of a boundary layer, strong contrac- Wall functions should be prevented if possi-
tion of grid nodes is required towards solid sur- ble, and used with care when necessary.
faces . The cells that are thus occurring close to
the wall have a high aspect ratio; typically for a 2.1.3 Surface roughness
ship at full scale they can be 1 meter long, 0.5
meter wide and only 10-3 millimeter high. This In ship viscous-flow computations and even
not only results in a high mesh count, but it also in model testing the hull of a ship is considered
poses strong demands on the mathematics in the to be hydrodynamically smooth. At full scale,
numerical solution procedure. This led to the however, during the operation of a ship the
introduction of so-called wall functions. For the roughness increases due to use and fouling, thus
laminar flow over an infinitely long flat plat at increasing the frictional resistance. There are
zero pressure gradient, the well-known Blasius basically two methods to include roughness
solution exists, where the velocity profile is effects in RANS computations. Either through
independent of the streamwise co-ordinate, the adaption of wall functions (when used) or
when scaled appropriately. For turbulent flows through the adaptation of the turbulence boun-
such solutions only exist for the innermost part dary conditions (for instance in the k-ω model).
of the boundary layer. This gives the possibility However, some major problems still remain.
to remove a significant part of the cells close to First the validity of the roughness model: to
the wall (which also have the highest aspect what extent does the equivalent sand-grain
ratio), and impose the velocity at the first grid roughness that is typically used correctly adjust
node adjacent to the wall (the wall function). the velocity profile close to the wall? And sec-

Application 2011 01
ond, how can the condition of the surface of a 2.2 Geometry creation and modification
ship, sailing at any part of the world, be trans-
The geometry is generally provided as sur-
lated to an equivalent sand-grain roughness?
face definitions in an IGES file format. Alterna-
Surface roughness is still an active field of tive file formats may be also used provided care
research, and no general guidelines can yet be is taken to ensure that sufficient accuracy is
given. maintained during the file transfer process. The
accuracy of the geometry should be checked to
2.1.4 Incident waves ensure that the surface definitions are reasona-
bly smooth and connect within a given toler-
Define the wavelength, amplitude, and en-
ance. The tolerances for the geometry should be
counter frequency. For irregular waves define
based on the length between perpendiculars.
spectrum and proper parameters. Define the cut-
The required geometry tolerance also depends
off frequency (highest frequency to be included
on the Reynolds number required for the flow
in the simulation), and determine the corres-
calculations.
ponding wavelength.
Generally, appropriate geometry tolerances
2.1.5 Motions are:
Scale Lpp(m) Re Tol.(m)
6 7
Estimate the frequencies and amplitudes of Model 1 < LPP < 10 10 - 10 10-5
Interme-
each of the motions to be simulated (surge,
diate
10 < LPP < 50 107 - 108 5x10-5
sway, heave, roll, pitch, yaw) and the amplitude Full 50 < LPP <250 108 - 109 10-4
of displacement of the ship.
Due care and attention is required to resolve
2.1.6 Flow features geometry features such as trailing edges that
may be less than an order of magnitude larger
Define location, size and characteristic fre- than the geometry tolerance. Geometry features
quency of the flow features to be resolved (vor- that are smaller than the geometric tolerance do
tices, separation, flutter, etc.). not need to be resolved.
2.1.7 Region of influence Additional geometry is required for the grid

generation process. It is recommended that this
Estimate the extent of the domain to be geometry is produced within the grid generation
simu-lated to minimize interaction of the package to the same geometric tolerances as the
boundary conditions with the simulation results.

Application 2011 01
surfaces. The required geometry components 2.3.1 Definition of the domain boundaries
are:
Ship viscous-flow computations typically
• Bounding surfaces of the grid domain
have three fixed boundaries: the ship surface,
surrounding the geometry
the symmetry plane and the (still) water surface.
• Intersection curves between the
Furthermore three additional boundaries have to
appendages and the hull
be defined in order to have a closed domain
• Intersection curves between the hull and
around the ship. Independent of the grid type
appendages and the vertical plane of
used these will include an inlet, an outlet and an
symmetry
exterior boundary, where approximate boundary
• Additional curves may also be required to
conditions have to be defined. These boundaries
assist in defining key details of geometry
have to be placed sufficiently far from the ship
Due consideration to the position and orien- to minimize the effect of the location of these
tation of the origin of the geometry should be boundaries on the solution. For the inlet and
taken depending on the type of problem to be exterior boundary either the uniform (undis-
solved as forces and moments will be obtained turbed) flow is usually imposed, and in that case
about this origin and velocity and rotation direc- the these boundaries should be located 1-2 LPP
tions depend on the orientation. It is recom- away from the hull. Alternatively potential flow
mended that a consistent coordinate system be can be imposed at these boundaries, which en-
used within an organization. Imported CAD ables a reduction of the domain size. At the out-
definitions should then be modified to conform let in general zero gradients for all unknowns
to this system. are imposed.
In case free-surface boundary conditions are

2.3 Grid generation
imposed on the water surface, the domain size
Details on the grid generation process will has to be increased further. Preferably the Kel-
largely depend on the solver and the type of vin wedge does not intersect with the exterior
grids it can handle (Cartesian, structured mul- boundary, to prevent wave reflection.
tiblock, unstructured, overset, etc.) Here are
Unsteady methods often require a damping
some general guidelines that apply to most
zone downstream, to prevent wave reflection
solvers.
from the outlet boundary. There the outlet has to
be placed 3-5Lpp downstream.

Application 2011 01
2.3.2 Element type using a prism mesh grown out of triangular

mesh on hull surface, and tetrahedral mesh is
Quadrilateral (2D, 4-sided) and hexahedral used elsewhere away from hull. Compared to
(3D, 6-sided) elements are the most popular typical structured meshes, meshing time can be
element types supported by almost all CFD dramatically reduced with unstructured meshes.
codes. Topological attributes of these elements Spatial accuracy for unstructured mesh ele-
– presence of opposing faces, and relative loca- ments such as triangles, tetrahedra, and pyra-
tions of cell centers and face centers – have mids can be lower than that for quadrilateral and
been found to be beneficial to spatial accuracy hexahedral elements. With unstructured meshes,
of numerical solutions. In typical mappable, one usually need a far greater number of com-
structured mesh-based solvers, the presence of putational elements than structured meshes in
stencils (e.g., i, j, k coordinates in the computa- order to achieve a comparable accuracy. Fur-
tional domain) readily accommodates high- thermore, spatial accuracy of the majority of
order discretization schemes (e.g., 5th-order unstructured mesh-based finite-volume solvers
convection scheme) that can enhance spatial is limited to 2nd-order.
accuracy. They are also efficient in terms of
usage of elements, since they can be clustered Which element types to use for a given
and/or stretched as needed to economically re- problem really depends on many factors such as
solve the flow fields. The main downside of the solver (does your solver support unstruc-
these structural mesh elements is that it is often tured mesh?), objective of the computation (do
very hard to generate high-quality structured you need to resolve fine details of the flow?),
meshes for complex geometry. and computer resource (do you have computers
to run cases involving large meshes?). Here are
The “unstructured” mesh gives more flexi- general guidelines in choosing mesh and ele-
bil-ity in the choice of element types, facilitat- ment types.
ing mesh generation for complex geometry. The
• For relatively simple configurations such
majority of unstructured mesh-based CFD solv-
as bare hulls, consider using a high-
ers allow use of arbitrary polyhedral elements
such as quadrilaterals (2D), triangles (2D), hex- quality hexahedral mesh.
ahedra, tetrahedra, wedges, pyramids, prisms,
• For relatively simple configurations
to name a few, and combination of all them
involving body-motion (free sinkage and
(hybrid unstructured mesh). In a typical un-
structured mesh frequently adopted in ship hy- trim), consider using overset grids if your
drodynamics, hull boundary layer is discretized

Application 2011 01
CFD solver can take them and run on wavelength also depends on the order of accu-
them. racy of the numerical scheme so if 40 points are
required for a 3rd/4th order method then 80
• For complex configurations such as fully- points are required for a 2nd order scheme to
appended ships for which a high-quality obtain the same accuracy (as provided by most
structured mesh is difficult to generate, commercial codes).
- comment on
consider using an unstructured mesh,
(1) the minimum number of points per wave-
preferably a hybrid unstructured mesh.
length: for too short wavelength (small Froude
• Avoid using tetrahedral mesh in boundary number) it is probably pointless to try to catch
the small waves: the wave resistance component
layers, near free surface, and in the
is likely negligible and
regions where high resolution of flow-
(2) minimum grid density to capture flow de-
fields is required. Use hexahedral grids or tails (e.g. wake at the propeller disk) or global
prismatic grids instead. forces (i.e. total resistance)
2.3.3 Grid points - Use no less than 20 grid points in the ver-
tical direction where the free surface is expected
Grid points distributions are determined with
- Whenever possible use orthogonal grids to
consideration of the following points.
resolve a free surface
- Based on the availability of computer time
- For turbulence models integrating to the
and power, determine the size (total number of
wall (Spalart-Allmaras, k-ω, etc.) locate the first
grid points) of the grid as well as the grid size
grid point at a distance from the ship’s wall
required for previous similar problems. This +
such that y = 1 . If wall functions are used this
should determine if sufficient resources are +
distance can increase such that 30 < y < 300 ,
available to obtain reliable results.
depending on the wall function implementation.
- Design the grid blocks in such a way that
- Whenever possible use hyperbolic grid ge-
they will be properly decomposed for efficient
nerators to guarantee as much as possible an
computation, avoiding the use of too many
orthogonal grid near the wall.
small blocks.
- Grids orthogonal to the domain boundaries,
- Use no less than 40 grid points per wave-
where the boundary conditions are imposed, are
length on the free surface. In irregular waves
recommended. For some boundary conditions,
use at least 20 grid points for the shortest wave
such as symmetry conditions and wall condi-
length to resolve. The number of grid points per

Application 2011 01
tions this orthogonality condition may be man- and the computational (i, j, k) space in case of
datory for some solvers. structured grid systems. In the single block grid
- Provide refinement where flow features of around a ship hull, O-O or H-O topology is
interest are expected, in accordance with the adopted in most cases, although C-O and H-H
size of the feature to be simulated. Where the grids can be applied. In either topology, the grid
flow features of interest are not known before- lines in the girth-direction are O-type. The lon-
hand, it is necessary to use an iterative process gitudinal grid lines in O-O grids wrap around a
to establish the existence of the key flow fea- ship hull whereas those in H-O grids start from
tures of a given geometry. This requires an ini- the inflow boundary and go through regions
tial flow solution to be obtained and examined ahead of a ship, side of a hull and aft of a ship.
and subsequent grid refinement provided around Thus, when the total number of grid points and
the flow features. This should be carried out the number of grid points along the lines in the
until some measure of grid refinement index is normal and in the girth directions are fixed, O-O
satisfied to ensure that flow features are suffi- grids can accommodate more grid points along
ciently resolved. a ship hull than H-O grids. Also, O-O grids can
- If overset grids are used, check that overlap be adapted more easily to a blunt bow or a wide
is sufficient for the number of fringes that are transom sterns which are typical in modern
needed in your code and order of accuracy. commercial ships. On the other hand, since the
- Ensure sufficient resolution of high curva- grid lines in the normal directions spread to the
ture geometry is provided, especially around outer boundary in O-O grids, the grid resolution
leading and trailing edges. An appropriate grid in the wake region and the region away from a
structure can enable more efficient use of com- hull tends to be lower than the H-O grids.
puting resources but at the expense of increased Therefore choice of grid topologies should be
grid generation time and complexity. based on the nature of ship hull geometry and
- Check the grid quality to guarantee that all on the consideration of which part or which
volumes are positive (positive Jacobian in struc- feature of flow fields is more important than
tured grids), skewness and aspect ratio are ac- others.
ceptable, and that orthogonality is nearly satis-
fied in most places. 2.3.5 Non conformal mesh
Non conformal grids may be required for

2.3.4 Grid topology
highly complex geometries. This occurs when
Grid topology is the mapping relation be- the level of detail of the geometry is increased
tween the grid surface in physical (x, y, z) space and as more geometric entities such as propul-

Application 2011 01
sion shaft and bracket arrangements and bilge change in the grid across the non conformal
keel and roll stabilization systems are included region is minimized with similar grid resolution
in the geometry definition. These types of ge- and spacing used for both grid regions. Non
ometries are often required for wake flow con-formal grids can also use different types to
analysis where it is important to capture small assist the grid generation process, for example a
flow features. For these types of grids a non- grid for a detailed rudder with skeg and end
conformal grid may be more appropriate where plates can be produced using a local
parts of the total grid do not fully connect. For prism/tetrahedral grid which is embedded inside
these grids the flow solution algorithm must use a multiblock grid for the hull.
special coding in order to interpolate between
the non-connected grids. For some methods this 2.3.6 Expansion ratio and number of grid
interpolation scheme may be defined by ‘inter- points in boundary layer
faces’ where the interpolation method is defined
The number of points within the boundary
across grid boundary surfaces, for other me-
layer is determined by the level of accuracy
thods the interpolation scheme is defined by
required and the turbulence model chosen. A
‘overlaps’ where the interpolation is defined
near wall turbulence model resolving the lami-
across local grid volumes. For both of these
nar sub-layer needs at least 3 points inside it,
types of interpolation schemes the formal order
which for a y+=2 results in an expansion ratio
of accuracy is likely to be reduced, especially
of 1.5, the largest acceptable. In most cases a
when there are large differences between the
y+=1 will be used with expansion ratios around
grid resolution and topology. However, this type
1.2. Wall functions start farther out in regions of
of approach can be used to considerably simpli-
smaller velocity gradients and can use large
fy the grid generation process so that locally
expansion ratios, as large as 1.5 for coarse grids.
better quality grids can be produced around the
In fine grids integrating all the way to the wall
various geometry components and assembled
the total number of points within the boundary
together to form a complete grid using the inter-
layer can be very large, on the order of 100,
polation schemes.
while for coarse grids with wall functions less
Non conformal grids should be used with than 10 will suffice.
care, for example, for free-surface flows unde-
sira-ble wave reflections can occur at the inter- 2.3.7 Grid skewness
faces or overlap if the interpolation scheme is
Typically the 3x3 determinant for structured
unable to resolve the change in grid correctly.
grids should be greater than 0.3, as a measure of
This can be alleviated by ensuring that the local

Application 2011 01
the Jacobian and associated skewness. However, computation. Otherwise zero normal gradient
it may be necessary to have a few small cells conditions can be attempted.
where the 3x3 determinant is no better than 0.15. - The bottom can be treated as a far field if
For these cases it may be necessary to use a deep water is being simulated.
smaller time step or increased under-relaxation Sides and bottom may need to be treated as
in order to achieve converged results. moving boundaries if shallow waters or a nar-
row towing tank are being simulated.
2.4 Boundary conditions - In case of free surface flow simulations, the
so-called radiation conditions must be imposed
Choose boundary conditions that are com- on exit or far field boundaries in order to pre-
patible with the domain size chosen and prob- vent wave reflection on boundaries. Simple way
lem approximations. to implement no reflection condition is to damp
- Inlet conditions that are far from the field waves which go through boundaries with nu-
can impose pressure, velocity and free surface merical dissipation by use of large grid space
elevation as Dirichlet boundary conditions mi- near the boundaries or by explicitly adding arti-
micking free flow. If the boundary is close to fical damping terms to the governing equations.
the ship Neumann boundary conditions (zero This is often called ‘numerical beach‘ approach.
normal gradients) need to be imposed to relax
the pressure and free surface.
2.5 Choice of the time step
- Far field boundary conditions can also be
free flow if the boundary is far from the object In explicit solvers the time step is chosen to
(typically one ship length for Froude numbers satisfy the CFL condition or to resolve the flow
larger than 0.2). If the boundary conditions are features of interest, whatever results smaller.
close or the Froude number is small then the Usually the CFL condition is more demanding
ship will affect significantly the flow on the than the flow requirements. In implicit solvers
boundary, and Neumann conditions are pre- the time step is decided by the flow features. As
ferred. a rule of thumb:
- Exit conditions usually are modeled with - For waves, use at least 60 time steps per pe-
zero second derivative for velocities (zero trac- riod for the shortest waves, or 100 time steps
tion) and zero gradient for pressure and free per period for regular waves.
surface. This condition requires no inflow from - For other periodic phenomena (roll decay,
this boundary, so it must be placed far down- vortex shedding, etc.) use at least 100 time steps
stream enough to guarantee this throughout the per period.

Application 2011 01
- For complex unsteady phenomena, like computational cells. Usually in definition the
wetted transom instabilities, use at least 20 time residual is also scaled by a reference value.
steps per period for the highest frequency to be Sometimes L2-Norm and L∞-Norm are also used
resolved. to define residual.
- For rotating propellers use at least 200 time
CFD users do not need to worry about the
steps per revolution.
definition of residuals, as they are often pre-
- For standard pseudo-transient resistance
defined by code developers. Instead, attention
computations, use ∆t = 0.005 ~ 0.01 L/U. The
should be paid to the selection of convergence
choice of time step will also depend on the
criteria. The recommended criterion is “the drop
complexity of the turbulence model. For Rey-
of scaled residuals by at least three orders of
nolds stress turbulence models it is more appro-
magnitude off their initial values”. However if
priate to use ∆t = 0.001 ~ 0.0025 L/U. This also
this criterion cannot be achieved due to com-
requires a larger number of iterations to obtain
plexity of the problem or oscillatory conver-
reasonable convergence. In more unstable prob-
gence is found, then other criteria can be used to
lems, as those with low Froude number, a
assess the convergence of the globally inte-
smaller time step may be needed. Notice that
grated parameters, for example:
naturally transient problems will not reach a
- Forces and moments acting on the hull
steady-state solution.
- Thrust and torque produced by the propul-
sion system
2.6 Choice of convergence criteria
- Velocity and turbulence parameters in key
A number of convergence criteria should be region of the flow field (e.g. at propeller plane
defined and examined in order to ensure reliable for nominal wake calculations)
convergence of solution.
2.7 Choice of free surface model
At first hand the level of convergence should
be assessed by the history of residual variations There are two major categories in free sur-
for the mass and momentum equations. Resi- face models. First one is an interface fitting ap-
duals indicate how far the present approximate proach in which a numerical grid is aligned to
solution is away from perfect conservation (bal- deformed free surface shape and the other is an
ance) of mass and momentum. Thus the residual interface capturing approach in which a free
for a discretized equation is defined as the L1- surface shape is defined as an iso-surface of a
Norm of the imbalance between the left and the marker function and a grid does not to fit to a
right hand side of that equation over all the free surface. Choice of a free surface model is

Application 2011 01
needed when a flow solver used offer both inter- 2.8 Choice of turbulence model
face fitting and capturing models.
Turbulence modelling has been an important
The interface fitting approach is more accu- research topic over the last decades. A large
rate and efficient than the capturing approach, number of models have been proposed, tested
since free surface boundary conditions can be and applied, but no ‘universal’ model has been
applied in the exact free surface location. There- developed. Thus one is forced to choose the best
fore the interface fitting model may be selected model available for each specific application.
whenever it is possible. However, it should be The majority of turbulence models is based on
noted that re-gridding procedure is essential in the so-called Boussinesq hypothesis, which de-
the interface fitting method in order to keep the fines a turbulence or eddy viscosity (as opposed
gridlines follow the deformation of free surface. to the molecular viscosity) to account for the
This may cause severe distortion of gridlines effect the turbulence motion has on the mean
even though the initial grid fitting to an undis- flow.
turbed free surface has a good quality.
Zero-equation, or algebraic models express
Difficulties in grid generation and/of re- the eddy viscosity in terms of the mean flow
gridding can be avoided in interface capturing variables and mean flow gradients without solv-
approaches. Also, in the case that large defor- ing any additional equations. They are hardly
mation of free surface, such as overturning or ever used in ship hydrodynamics.
breaking waves, is expected, interface capturing
One-equation models solve one additional
methods should be used. Since the capturing
equation (i.e. in addition to the momentum and
methods demand finer grid resolutions in the
mass conservation equations) for the eddy vis-
interface zones, grid generation requires more
cosity. Regularly encountered in ship hydrody-
attentions. Choice of the level-set function
namics are models by Menter and by Spalart-
method and volume-of-fluid method in the cap-
Allmaras. These models are sometimes ex-
turing approaches is little impact in the final
tended with a correction for vortical flow, to
solutions. Although details of numerical proce-
improve wake field predictions.
dures are different from each other, well-
examined flow solvers provide similar results in Two-equation models solve two additional
both models. equations for the eddy viscosity, one for the
turbulence kinetic energy (k), and one for its
dissipation rate (typically ε or ω). These models
have shown to be able to give accurate predic-

Application 2011 01
tions in ship hydrodynamics, especially certain ship hydrodynamics prevents the application of
versions of the k-ω model and are by far the both methods in practical design projects.
most applied ones (80% of the submissions for
the Gothenburg 2010 Workshop). 2.9 Choice of numerical scheme
An important class of turbulence models, not In the majority of industrial CFD codes, dif-
based on the Boussinesq hypothesis, are fusion terms in the governing equations are dis-
the Reynolds-stress models, and versions cretized using a second-order (central differenc-
thereof. Rather than introducing an eddy- ing) scheme by default. Thus, spatial accu-
viscosity, they aim to solve the equations for the racy is largely determined by discretization
six Reynolds stress components directly. Apart scheme used for convection terms.
from that, additional equations have to be
The first-order upwind (FOU) scheme, of-
solved, since terms in these equations require
fered in many commercial CFD codes often as a
modelling as well. Consequently Reynolds-
default scheme, is famously stable. However, it
stress models are more computationally inten-
introduces an unacceptably large amount of
sive, and often less easy to converge, compared
numerical (false) diffusion - that is why it is so
to the one or two-equation models. However,
stable. Therefore, it should be avoided at all
they contain more physics and can be expected
costs. However, the very robustness of the FOU
to be more accurate than eddy-viscosity models.
scheme can be exploited to start up the solution.
A more recent development is Large Eddy For example, the first 100 iterations (or time
Simulation (LES). Other than the turbulence steps) during which the solution is most suscep-
models discussed above it does not average the tible to numerical instability and divergence)
Navier-Stokes equations in time, but filters can be run using the FOU. As the flow-fields
them in space. This results in transient computa- start settling down, one can switch to a high-
tions on extremely dense grids as they aim to order scheme.
resolve all turbulence motion to a very small
The majority of high-order convection dis-
scale. Detached Eddy Simulation (DES) is a
cretization schemes in popular use today for-
hybrid method that tries to reduce the required
mally have a second-order of accuracy with an
computational effort by solving the (unsteady)
upwind bias. All these second-order upwind
RANS equation in the boundary layer and ap-
(SOU) schemes differ from one another in terms
plying LES in the rest of the domain. However,
of the flux limiter used to suppress unphysical
the very high Reynolds numbers encountered in
oscillations in the solutions. Still higher-order

Application 2011 01
schemes such as 5th-order scheme exist. How- 3. COMPUTATION

ever, not all CFD solvers offer such higher-
At runtime, a few decisions need to be made:
order schemes. Even if they are available, the
a) In modern computers, choose the number
lack of robustness often makes them less useful
of processors so that you use 50,000 to 200,000
than claimed. The SOU scheme that is both rea-
grid points per processor.
sonably accurate and robust, and for that reason
b) To maximize performance, try to distri-
is an industrial workhorse for convection discre-
bute the load evenly between nodes. For in-
tization. The SOU scheme is therefore recom-
mended for all convection-diffusion type of stance, if running in a Linux cluster with 2 dual-
transport equations. core processors (4 cores) per node, and your
case needs 6 cores, you can distribute your load
Volume-fraction equation requires a special in two nodes using 4-2 or 3-3 configurations.
care, inasmuch as the transported quantity is The second balances the load per node better.
essentially a step function in the vicinity of free c) Modern workstations with shared memory
surface, and the traditional convection schemes are available with up to 48 processors, though
designed for convection-diffusion equations much larger specialized systems are produced.
perform poorly in transporting the step-function. High-performance clusters are typically cheaper
It has been found that convection schemes with per processor for large systems (thousands of
some degree of downwind bias resolve the cores) but use distributed memory. Shared-
sharp interface much better. memory systems allow all processors access all
memory, resulting in easier programming and
Second-order central differencing (CD)
better scalability of most applications. On the
scheme is often used in large eddy simulation other hand, distributed memory systems provide
(LES) and direct numerical simulation (DNS) in massive number of processors for very large
favor of its low-dissipation that is critical to computations.
accurately resolve turbulent structures. How-
ever, CD scheme is inherently unstable, giving
troubles for cases involving fine meshes and 4. POST-PROCESSING
small effective viscosity (large cell Reynolds
number). One should consider using a stabi- 4.1 Visualization
lized form of central differencing. A number of post processing plots should be
used as a minimum sub-set of information to
ensure that the correct settings have been used

Application 2011 01
for each computation. This should include the 3. http://www.cfd-online.com/Wiki/Best_pract

following: ise_guidelines
- Contour plots of the pressure coefficient, 4. http://www.cfd-online.com/Links/onlinedoc
skin friction coefficient and y+ of the geometry s.html#bestpractice
surface 5. http://www.nafems.org/resources/cfd_guida
- Contour plots of the boundary layer profiles nce/
along the hull geometry 6. ITTC, 1999, “Uncertainty analysis in CFD,
- Contour and vector plots of the nominal or Verification and Validation Methodology
powered wake upstream of the plane of the pro- and Procedures”, No. 7.5-03-01-01.
pulsor (Care should be taken to ensure that this 7. ITTC, 1999, “Uncertainty analysis in CFD,
plane does not lie within the propulsion Guidelines for RANS codes”, No. 7.5-03-
disc/volume) 01-02.
8. ITTC, 1999, “CFD User’s Guide”, No. 7.5-
Reasonable checks should be carried out to 03-01-03.
ensure that these plots are smooth and continu- 9. ITTC, 1999, “CFD Verification”, No. 7.5-
ous. In particular, regions of specially bad grid 03-01-04.
quality should be evaluated to check if the 10. Eca L. Vaz G. and Hoekstra M., 2010.
solver can handle properly less than optimal “Code Verification, Solution Verification
grids without causing unphysical artifacts. and Validation In RANS Solvers”, Proc. of
ASME 2010 29th Intl Conference on Ocean,
4.2 Verification and Validation Offshore and Arctic Engineering, China.
11. Tao X. and Stern F., 2010, “Factors of Safe-
The ITTC procedure 7.5-03-01-01 already
ty for Richardson Extrapolation”, Journal of
provides “methodology and procedures for es-
Fluids Engineering, Vol. 132, 061403.
timating the uncertainty in a simulation result”.
12. ASME Committee PTC-61, 2009, ANSI
Standard V&V 20. ASME Guide on Verifi-
5. USEFUL WEBSITES AND REFER- cation and Validation in Computational Flu-
ENCES id Dynamics and Heat Transfer.
1. MARNET-CFD Best Practice Guidelines 13. Eca L. and Hoekstra M., 2008: "Testing
for Marine Applications of CFD Uncertainty Estimation and Validation Pro-
(https://pronet.wsatkins.co.uk/marnet/guidel cedures in the Flow around a Backward Fac-
ines/guide.html ) ing Step", 3rd Workshop on CFD Uncertain-
2. ERCOFTAC Best Practice Guidelines. ty Analysis, Lisbon.
(http://www.ercoftac.org/index.php?id=77)

Application 2011 01
14. Eca Luis and Hoekstra Martin, 2006: “Dis- hull form without rudder and calm water condi-
cretization Uncertainty Estimation based on tions with zero sinkage and trim. The Reynolds
Least Squares version of the Grid Conver- number was defined as Re = 1.4x107 and the
gence Index”, 2nd Workshop on CFD Uncer- Froude number defined as Fr = 0.26. The length
tainty Analysis, Lisbon. of the hull Lpp was defined as 230.0 m at full
15. Eca L. and Hoekstra M., 2002: "An Evalua- scale with a scale ratio of λ = 31.6 for the model
tion of Verification Procedures for CFD scale measurements.
Applications”, Proceedings of the 24th Sym-
The Froude number defines the free stream
posium on Naval Hydrodynamics, Japan.
speed of
16. Stern F. et al. 2001, “Comprehensive Ap-
proach to Verification and Validation of U = Fr �g(Lpp/λ) =0.26*�(9.81 ∗ 230/31.6)
= 2.196 m/s.
CFD Simulations Part 1: Methodology and
Procedures”, Journal of Fluids Engineering,
The wavelength, λW corresponding to this
Vol. 123, December. Froude number for the model scale geometry is
17. AIAA (1988), “AIAA guide for the verifica- given by:
tion and validation of computational fluid
λW=2π( Lpp /λ) Fr2 = 2π*(230/31.6)*0.262
dynamics simulations”, AIAA G-077-1998. = 3.0915 m.
18. Rizzi, A. & Voss, J. (1998), “Towards es- The Reynolds number defines
tablishing credibility in computational fluid CF = 0.075/(log10Re-2)2
dynamics simulations”, AIAA Journal, vol. = 0.075/ (log10(1.4x107)-2)2 = 0.002832
36, no. 5, pp. 668-675. and for y+=1 the first cell height
19. Roache, P.J. (1998), “Verification and vali-
y =y+(Lpp/λ) /(Re�(Cf/2)
dation in computational science and engi-
= (230/31.6)/(1.4x107�(0.002832/2))
neering”, Hermosa Publishers, Alberquer- = 1.3816x10-5 m.
que.
Other first cell heights simply scale the dis-
tance for y+=1 so y+=30 is
6. EXAMPLE FROM G2010 WORK-
y=30*1.3816x10-5 m.
SHOP
Checks of the geometry file provided as an
An example of the application of these
IGES file showed that the geometry was built
guidelines is illustrated by one of the Gothen-
with a suitable tolerance.
burg 2010 (G2010) workshop test cases. The
example chosen was Test Case 2.1, the KCS

Application 2011 01
The domain sizes chosen for the majority of contributors used unstructured prism and tetra-
the flow calculations carried out for the G2010 hedral cells.
Test Cases 2.1 workshop had an upstream
The number of cells per wavelength of at
boundary of approximately 1 Lpp from the bow,
least 40 points per wavelength requires a maxi-
a downstream boundary of approximately 2 Lpp
mum spacing in the axial direction of 3.0915/40
from the stern, a side boundary 1 Lpp from the
= 0.07728 m. All contributors used at least this
plane of symmetry and a bottom boundary of 1
number of points in the axial direction on the
Lpp from the keel. A number of different posi-
ship hull except where coarse grids were de-
tions were used for the top boundary, some
fined for grid resolution studies.
methods computed the air flow above the ship
hull and had a boundary of up to 0.5-1 Lpp from Most contributors used a VoF method to de-
the keel and other contributors defined the top fine the free surface wave pattern.
boundary at the deck at 0.025 Lpp. The choice of
top boundary position was dependent on details Nearly all contributors used a variant of the
of the chosen boundary condition. k-ω turbulence model with some contributors
using Reynolds stress models in addition to the
Another variant on the domain size was to two equation model. Some contributors used
choose to match the width and depth of the do- wall functions, with some modifications to ac-
main to the width and depth of the towing tank count for pressure gradients but most used near
in which the model scale hull was measured. wall boundary conditions with y+=1.
This is appropriate for validation cases where
detailed comparison with measurements is re- Second order accurate numerical schemes
quired. were used for this test case with some contribu-
tors using higher order methods. Most con-
The majority of contributions to the G2010 tributors used a special scheme for the VoF to
workshop Test case 2.1 used hexahedral cells improve the free surface capturing.
with expansion ratios between 1.2 and 1.5 in the A range of different computational resources
boundary layer. A number of techniques were were used for these cases, from workstations to
used to create the hexahedral cells, some used large scale supercomputers, but the majority of
single block methods, some multi-block meth- contributors used some form of parallel
ods and other overset and Cartesian cell meth- processing techniques to reduce the elapsed
ods so most of the techniques that are described time for the computations.
in these guidelines were used. In addition, some

Application 2011 01
Convergence criteria of at least three orders

of magnitude reduction in the residuals were
used by all contributors.
Contour plots of the wave patterns and longi-
tudinal plots of the wave profile at a number of
positions were produced by every contributor
and compared with the measured data.

Practical Guidelines For Ship CFD Applications-ITTC PDF

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Practical Guidelines For Ship CFD Applications-ITTC PDF

Diunggah oleh

Hak Cipta:

Format Tersedia

ITTC – Recommended 7.

Practical Guidelines for Ship CFD 2.3.5 Non conformal mesh 8

Edited /Updated Approved

Practical Guidelines for Ship CFD Effective Date Revision