You are on page 1of 19

ITTC – Recommended 7.

5–03
–02–03
Procedures and Guidelines Page 1 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

Table of Contents

3.1.1 Turbulent Scales......................... 10


1. INTRODUCTION ................................. 2
3.1.2 RANS Models ............................ 11
2. PRE-PROCESSING .............................. 3 3.2 Near-wall Modelling ..................... 12
2.1 Geometry ......................................... 3 3.3 Surface Roughness ........................ 13
2.2 Computational Domain and 3.4 Numerical Schemes....................... 13
Boundary Conditions ..................... 4
3.5 Free Surface .................................. 14
2.3 Grid .................................................. 4
3.6 Time Step ....................................... 15
2.3.1 Structured Grids ........................... 5
2.3.2 Unstructured Grids ....................... 5 3.7 Parallel Computing....................... 15
2.3.3 Hybrid Grids and Other 3.8 Convergence .................................. 16
Techniques ................................... 5
4. POST-PROCESSING ......................... 16
2.3.4 Cell Types .................................... 6
2.3.5 Grid Resolution ............................ 7 4.1 Visualization .................................. 16
2.3.6 Near-wall Region ......................... 8 4.2 Verification and Validation ......... 17
2.3.7 Grid Quality ............................... 10
5. REFERENCES .................................... 18
3. COMPUTATION ................................ 10
6. REFERENCES TO OTHER CFD
3.1 Turbulence Models ....................... 10 GUIDELINES ...................................... 18

Edited /Updated Approved


Specialist Committee on CFD in Marine Hy-
27th ITTC 2014
drodynamics of the 27th ITTC
Date 02/2014 Date 09/2014
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 2 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

Practical Guidelines for Ship CFD Applications

For the applications covered by the present


1. INTRODUCTION guidelines, the flow is mostly turbulent and the

This guidelines describe recommendations Reynolds numbers (Re) are of the order of 106
and common practices for ship CFD applica- in model scale and 109 in full scale,
tions in a general way. Separate specific guide- VL
lines cover the CFD process for other applica- Re  (0)

tions in detail.
where V is the ship speed, L the length be-
In this guideline the CFD process is divided tween the perpendiculars,  and  are the
into three steps: pre-processing, computation, density and viscosity of the fluid, respectively.
and post-processing. The pre-processing step in-
volves the definition of the geometry, the com- The current guidelines are intended for ap-
putational domain, and the computational grid. plications with high Reynolds numbers, typical
The choice of the governing equations to be of ship hydrodynamics.
solved, the most suitable numerical techniques
For free surface flow simulations, where the
to be used to solve them and the real computa-
free surface between air and water is solved, the
tional work are performed in the computational
Froude numbers (Fr) are typically of the order
step. The post-processing step is the visualisa-
of 0.1 and rarely above 1,
tion, the analysis, and the verification and vali-
dation of the results. V
Fr  (0)
gL
All these steps may be significantly different
for different naval hydrodynamic problems, where g is the acceleration of gravity.
such as hull resistance, propulsion, manoeuvra-
bility, sea keeping, sail aerodynamics, etc. How- Both air and water are typically assumed to
ever, the present document aims at providing be incompressible, allowing uncoupling mo-
good practice guidelines, which can be applied mentum and continuity equations from the en-
to most ship hydrodynamic applications. ITTC ergy equation. Therefore, velocity and pressure
specific guidelines are available for ship re- fields can be computed independently from the
sistance (7.5-03-02-04), self-propulsion (7.5- temperature field. This assumption remains
03-03-01) and RANS calculations of nominal valid for Mach numbers (Ma) smaller than about
wakes (7.5-03-03-02). 0.3,
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 3 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

V for Re ~ 106 . This tolerance may need to be re-


Ma  (0)
a duced when smaller scale appendages, such as
rudder or propellers, are to be included in the
where a is the speed of sound. flow computations or when the flow computa-
tion requires smaller grid cells than the indica-
Finally, these guidelines are written assum-
tive CAD tolerance. Conversely, as this toler-
ing that the solver is grid-based (either struc-
ance is based on the distance to the wall of the
tured or unstructured) and based on the Reyn-
first grid point (see §2.3.6) if wall functions are
olds Averaged Navier-Stokes equations
used the geometry tolerance can be increased
(RANSE), as found in most commercial and ac-
appropriately.
ademic CFD packages. Use of mesh-free meth-
ods is not covered in this guidelines. For a com- Due care and attention are required to re-
plete overview on the computational techniques, solve geometry features such as trailing edges
the reader is referred for example to Hirsch that may be less than an order of magnitude
(1989 and 1990) and Ferziger & Peric (2002).
larger than the geometry tolerance. Geometry
features that are smaller than the geometric tol-
2. PRE-PROCESSING erance should be removed from the geometry
before it is exported. It is recommended that any
2.1 Geometry additional surfaces, curves or any other geomet-
rical entities that are produced within the pre-
It is common practice to define the geometry processor software, be generated with the same
in a CAD (Computer Aided Design) software, tolerance of the original geometry.
which allows for a high control of the surfaces.
The geometry can be exported from the CAD The origin and orientation of the reference
software and imported into a CFD pre-processor coordinate system should be chosen with care
software (grid generator) with a range of for- because the equations of motion will be solved
mats. One format compatible with most com- with respect to this coordinate system and so
mercial software is IGES. The accuracy of the will be the computation of forces, moments, etc.
geometry should be checked to ensure that the Therefore, if the aim of the simulation is the es-
imported surface definitions are reasonably timation of a force (for example the resistance),
smooth and connect within a given tolerance. it is recommended to have one coordinate axis
aligned with that force, this to minimize the pos-
Indicative tolerances for the geometry are sibility of mistakes and to have the opportunity
105 L , where L is the ship length for the hull to monitor directly the quantity of interest, if an
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 4 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

iterative procedure (in time, for example) is in- solid wall in order to assume an undisturbed far
volved. On the other hand, if body motion is in- field velocity. When significant lift force is gen-
volved (as in the case of manoeuvring or sea erated with respect to the far field velocity di-
keeping computations) the best choice of the rection, computing the correct deflection of the
frame of reference could be driven by the re- upstream streamlines is critical to the correct
quirement for the rigid body motion solver. computation of the lift force. If L is the length of
Moreover, in order to minimise chances of er- the lifting surface, the inlet face should be posi-
rors, it is recommended that the same coordinate tioned at least 10L upstream of the lifting sur-
system is used in both the CAD software and the face. Conversely, when the lift is negligible and
pre-processor software. the body is streamlined, the inlet face can be po-
sitioned as close as one length upstream the
2.2 Computational Domain and Boundary body.
Conditions
Different boundary conditions can be used
In ship hydrodynamics, most of the prob- on the outflow face, though the most robust op-
lems involve external flows, i.e. the fluid is not tion is a Neumann condition (zero gradient) for
contained into physical boundaries. Therefore, velocity and pressure. The distance between the
the computational domain, which represents the outflow face and the body mainly depends on
volume of fluid to be modelled, must be closed the generation of lift and, for free surface com-
by additional surfaces that do not represent any putations, on the reflection of gravity waves.
physical boundary. In the more general case, the When large lift force is generated, the outlet face
tested object is completely surrounded by fluid. should be at least 20L downstream the lifting
In this case a prismatic computational domain surface, while for other cases the outlet face
can be considered. However, different geome- should be at least L downstream the geometry.
tries could better suit different grid topologies.
The other non-physical boundaries of the do-
Therefore, the shape of the computational do-
main should also be placed at least L away from
main should be dictated by the grid strategy in
the geometry. For towing tank walls symmetry,
place (see §2.3.1).
imposed velocity or slip conditions can be used,
The computational domain should include with symmetry the most commonly used.
inflow and an outflow surfaces. A Dirichlet con-
dition, i.e. a known velocity field, should be im- 2.3 Grid
posed on the inflow boundary. When no specific
Details on the grid generation process will
experimental data is known, the boundary con-
largely depend on the solver and the type of grid
dition should be placed sufficiently far from any
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 5 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

that it can handle (structured multi-block, un- in the wake region and the region away from the
structured, overset, etc.); here are some general geometry tends to be lower than the H-O grids.
guidelines that apply to most solvers. Therefore the choice of grid topology should be
based on the nature of the geometry and on con-
2.3.1 Structured Grids sideration of which areas or features of the flow
field are more important than others
Structured grids employ quadrilateral in 2D
and hexahedra in 3D and show regular connec- 2.3.2 Unstructured Grids
tivity. Grid topology is the mapping relation be-
tween the grid in the physical space (x, y, z) and Unstructured grids do not show a defined to-
the computational space (i, j, k). The physical pology and can be generated much more easily,
space can be mapped into one computational resulting in a big advantage for unstructured
space or it can be broken down into several grid technology. The cells can have several pos-
spaces. The latter approach is known as a multi- sible shapes (tetrahedral, hexahedra and other
block approach. polyhedral). However, the resolution of key
flow details such as boundary layers and wakes
For grids made of a single block around a can lead to significantly higher number of cells
single isolated geometry, O-O or H-O topology than equivalent structured grids (see §2.3.4 and
is adopted in most cases, although C-O and H-H §2.3.5) .
grids can be applied; see Thompson et al. (1985)
for examples of these topologies. In either topol- Unstructured solvers require typically more
ogy, the grid lines in the girth-direction are O- memory, are slightly slower and have less accu-
type. The longitudinal grid lines in O-O grids racy than an equivalent structured solver. This is
wrap around the geometry whereas those in H- due to the extra information needed to store and
O grids start from the inflow boundary and go process the grid connectivity and implementa-
through regions ahead of the geometry, side of tion complexities that prevent the use of higher-
and aft of it. Thus, when the total number of grid order schemes.
points and the number of grid points along the
lines in the normal and in the girth directions are 2.3.3 Hybrid Grids and Other Techniques
fixed, O-O grids can accommodate more grid
Hybrid grids can use both structured and un-
points along a geometry than H-O grids. Also,
structured grids in different parts of the domain.
O-O grids can be adapted more easily to a
Non-conformal grids are those where the points
rounded geometry. On the other hand, since the
or cells on a surface do not match. Non-confor-
grid lines in the normal directions spread to the
outer boundary in O-O grids, the grid resolution
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 6 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

mal grids are often used at the boundaries be- both grid regions. Non-conformal grids can also
tween blocks discretized with different grid den- use different cell types to assist the grid genera-
sities, or on moving (sliding) grids. These types tion process, for example a grid for a detailed
of grids are often required for wake flow analy- rudder with skeg and end plates can be produced
sis where it is important to capture small flow using a local prism/tetrahedral grid which is em-
features. bedded inside a multiblock grid for the hull.

The flow solution algorithm must use special Overset grids are formed by blocks that
coding in order to interpolate between the non- overlap in space. These grids are particularly
connected grids. For some methods this interpo- convenient when the different blocks experience
lation scheme may be defined by interfaces, a relative motion and therefore the overlapping
where the interpolation method is defined across changes with time. For instance, overset grids
grid boundary surfaces, for other methods the can be used for ship propulsion where one grid
interpolation scheme is defined by overlaps, is used for the hull and another is used for the
where the interpolation is defined across local propeller.
grid volumes. For both of these types of interpo-
lation schemes the formal order of accuracy is 2.3.4 Cell Types
likely to be reduced, especially when there are
Quadrilateral (2D, 4-sided) and hexahedral
large differences between the grid resolutions
(3D, 6-sided) cells are supported by almost all
and topologies of donor and receptor grids.
CFD codes. Topological attributes of these cells
However, this type of approach can be used to
– presence of opposing faces, and relative loca-
considerably simplify the grid generation pro-
tions of cell centres and face centres – have been
cess so that locally better quality grids can be
found to be beneficial to spatial accuracy of nu-
produced around the various geometry compo-
nents and assembled together to form a com- merical solutions. In structured grid-based solv-
ers, the presence of stencils (e.g., i, j, k coordi-
plete grid.
nates in the computational domain) readily ac-
Non-conformal grids should be used with commodates high-order discretization schemes
care because undesirable wave reflections can (e.g., fifth-order convection scheme) that can
occur at the interfaces or overlapping regions if enhance spatial accuracy. They are also efficient
the interpolation scheme is unable to resolve in terms of usage of cells, since they can be clus-
correctly the change in grid. This can be allevi- tered and/or stretched as needed to economically
ated by ensuring that the local change in the grid increase the resolution in particular regions
across the non-conformal region is minimized (such as boundary layers, or region where large
with similar grid resolution and spacing used for
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 7 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

velocity gradients occur). The main shortcom- to run cases involving large grids?). For rela-
ing of these structured grid cells is that it is often tively simple geometries, consider using a high-
very hard to generate high-quality grids for quality hexahedral grid. However, if body-mo-
complex geometries. tion (e.g. free sinkage and trim) is involved, con-
sider using overset grids if your CFD solver in-
Unstructured grids give more flexibility in cludes this capability.
the choice of cell types, simplifying grid gener-
ation for complex geometries. The majority of For complex geometries (e.g. a fully-ap-
unstructured CFD solvers allow use of arbitrary pended ship) for which a high-quality structured
polyhedral cells such as quadrilaterals and trian- grid is difficult to generate, consider using an
gles in 2D; and hexahedra, tetrahedra, wedges, unstructured grid (preferably a hybrid unstruc-
pyramids and prisms in 3D to name a few, and tured grid) or overlapping grids.
combination of all them (hybrid grids). In a typ-
Avoid using tetrahedral cells for boundary
ical unstructured grid, the boundary layer is dis-
layers, near the free surface, and in the regions
cretized using a prism grid grown out of a trian-
where high resolution is required. The use of
gular grid on the wall, and a tetrahedral grid is
hexahedral or prismatic cells would result in
used elsewhere away from wall. Compared to
higher accuracy and better convergence rates.
typical structured grids, gridding time can be
dramatically reduced with unstructured grids.
2.3.5 Grid Resolution
Spatial accuracy for unstructured grid cells such
as triangles, tetrahedra, and pyramids can be The number of grid points, and therefore the
lower than that for quadrilateral and hexahedral resolution of the grid, should be decided based
cells. Unstructured grids usually require a larger on the chosen turbulence model and on the tem-
number of computational points than structured
poral and spatial scales of the flow features of
grids in order to achieve a comparable accuracy. interest. In the following some practical guide-
Furthermore, spatial accuracy of the majority of lines are provided but interested readers can find
unstructured grid finite volume solvers is lim- in Pope (2000) and Sagaut (2006) more rigorous
ited to second order. discussion based on the physics of the resolved
time and statistical averages of the flow field,
Which cell types to use for a given problem
and the modelled turbulent fluctuations.
really depends on many factors such as the
solver (what cell types can the solver support), Whenever possible, use hyperbolic grid gen-
objective of the computation (do fine details of erators to guarantee as much as possible an or-
the flow need to be resolved?) and hardware re- thogonal grid near the wall. Grids orthogonal to
sources (are computational resources available the domain boundaries, where the boundary
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 8 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

conditions are imposed, are recommended. For used, though this condition may be extremely
some boundary conditions, such as symmetry expensive and unnecessary for low Froude num-
and wall conditions, this orthogonality condition ber flows. The same resolution can be achieved
may be mandatory for some solvers. with different spatial accuracies varying linearly
the spatial resolution. For instance, for a fourth-
Provide refinement where flow features of order spatial resolution, using 20 grid points is
interest are expected, in accordance with the size equivalent to using 40 grid points with a second-
of the feature to be simulated. Where the flow order spatial resolution.
features of interest are not known before-hand,
it may be necessary to use an iterative process to 2.3.6 Near-wall Region
establish the existence of the key flow features
for a given geometry. This requires obtaining The number of grid points within the bound-
and examining an initial flow solution, and sub- ary layer is determined by the level of accuracy
sequent grid refinement in the regions of interest. required, the turbulence model chosen, and
This process should be carried out until some whether wall functions are used.
measure of grid refinement is satisfied to ensure
Guidelines are provided in terms of the non-
that flow features are sufficiently resolved.
dimensional wall distance of the first point from
If overset grids are used, check that overlap the wall, y  . We define
is sufficient for the number of fringes that are
Friction velocity:
needed in the code and the order of accuracy. If
motions are computed, this requirement holds
for all time steps. w
u*  (0)

Ensure sufficient resolution where high cur-
vature geometry is present, especially around Non-dimensional wall velocity:
leading and trailing edges. An appropriate grid
structure can enable more efficient use of com- u
u  (0)
puting resources but at the expense of increased u*
grid generation time and complexity.
Non-dimensional wall distance:
In free surface flow simulations, when the
  u*
free surface is modelled, always use orthogonal y  y (0)
grids to resolve it if possible. Use no less than

40 grid points to resolve the shortest wave
length when a second-order spatial accuracy is
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 9 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

where  w is the skin friction on the wall, u the Once the desired y  has been chosen (either
local streamwise velocity component, and y is to comply with wall function requirements or to
the wall-normal coordinate. resolve the boundary layer), the distance y of
the first point from the wall can be computed as
A near-wall turbulence model resolving the follows:
laminar sub-layer needs at least 3 points inside
it, which using y   2 results in an expansion y L
y (0)
ratio of 1.5, the largest acceptable, but a maxi- Cf
ReL
mum value closer to 1.2 is recommended. In 2
most cases a y   1 will be used with expansion
ratios around 1.1. Wall functions start farther where C f is the friction coefficient, which can
out in regions of smaller velocity gradients and
be estimated a priori considering an equivalent
can use larger expansion ratios, as large as 1.2
boundary layer on a flat plate with zero pressure
for coarse grids. When direct integration is used
gradient. In particular, for laminar flow condi-
the total number of points within the boundary
tions
layer can be very large, more than 20, while
when wall functions are used, then less than 15 1.328
Cf  (0)
grid points could suffice. For wall functions, it ReL
is recommended that the first point from the wall
is well within the logarithmic layer of the for transitional flow conditions
boundary layer, therefore: 30  y   100. Table
1 summarizes the recommended values. 0.455 1700
Cf   (0)
ln  ReL  
2.58
ReL
First point Expan- Points
sion ratio within
boundary
layer for turbulent flow on a smooth surface

Near y  1 1.2 20
wall 0.455
Cf  (0)
ln  ReL  
2.58

Wall 30  y  100. 1.2 15
func-
tions
and for fully turbulent flow on a rough surface
Table 1. Recommended values for grid de- with roughness height 
sign at the wall.
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 10 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

1 Typically the 3x3 determinant for structured


Cf  2.5
(0)
    grids should be greater than 0.3, as a measure of
1.89  1.62 ln  L  
   the Jacobian and associated skewness. However,
it may be necessary to have a few small cells
See §3.3 for more details on the roughness where the 3x3 determinant is no better than 0.15.
height. For these cases it may be necessary to use a
smaller time step or increased under-relaxation
Equations (8-11) do not take into account the
in order to achieve converged results.
surface curvature and the pressure gradient. For
the friction coefficient over a ship hull, the 1957’
ITTC formulation allows a better approximation 3. COMPUTATION
of full scale friction resistance from model scale
friction resistance: 3.1 Turbulence Models

0.075 3.1.1 Turbulent Scales


Cf  (0)
log10  ReL   2
2

As mentioned in the Introduction, these


guidelines assume that the incompressible Na-
It should be noted that Eq. (12), though vier-Stokes equations for Newtonian fluids are
widely used and accepted, was developed for solved. The Reynolds number range is such that
typical hull forms in existence before 1957 and laminar to turbulent transition occurs and turbu-
thus it may not produce good results for modern lent flow must be modelled. In the present guide
and unconventional hull forms. it is assumed that the turbulence is modelled
with a RANS approach, where the instantaneous
Since Eqs. (7-12) are only generic approxi-
mations, y  should always be checked a poste- velocity is split into the sum of its statistical av-
erage and a turbulent fluctuation which is mod-
riori once the solutions are obtained and if nec-
elled by the turbulence model. It has to be high-
essary the grids should be modified accordingly.
lighted that, when dealing with unsteady flow
fields, it is assumed that a clear separation be-
2.3.7 Grid Quality
tween the time scales of the mean motion and
Check the grid quality to guarantee that all time scales of turbulent motion exists. The fun-
volumes are positive (positive Jacobian in struc- damental assumption in unsteady RANS com-
tured grids), skewness and aspect ratio are ac- putations is that the averaging time is bigger
ceptable, and that orthogonality is nearly satis- than turbulent time scales (which are averaged)
fied in most places.
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 11 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

but much smaller than mean flow time scales and applied, but no universal model has been de-
(which are solved). veloped. Thus one is forced to choose the best
model available for each specific application.
It is worth noting more computationally de- The majority of turbulence models are based on
manding approaches are becoming more and the so-called Boussinesq hypothesis, which de-
more common in recent years. Direct Numerical fines a turbulent or eddy viscosity (as opposed
Simulation (DNS) solves the Navier-Stokes to the molecular viscosity) to account for the ef-
equations to the resolution of the smallest turbu- fect the turbulence motion has on the mean flow.
lent scales but needs grids and time resolution For a detailed description of turbulence models,
which are not yet achievable for high Reynolds see for example Wilcox (2006).
numbers, such as those typical of ship hydrody-
namic applications. Zero-equation, or algebraic models express
the eddy viscosity in terms of the mean flow var-
Large Eddy Simulation (LES) does not aver- iables and mean flow gradients without solving
age the Navier-Stokes equations in time, but fil-
any additional equations. They are hardly ever
ters them in space. This results in transient com- used in ship hydrodynamic applications.
putations on extremely dense grids as they aim
to resolve the larger turbulence scales and mod- One-equation models solve one additional
elling the scales smaller than the grid resolution. equation (i.e. in addition to the momentum and
LES is used currently for highly demanding mass conservation equations) for the eddy vis-
flows where the transient nature of the turbu- cosity. Regularly used in ship hydrodynamics
lence needs to be resolved to smaller scales, like are models by Menter and by Spalart-Allmaras.
in the case of acoustic noise. These models are sometimes extended with a
correction for vortical flow, to improve wake
Detached Eddy Simulation (DES) is a hybrid field predictions.
method that reduces the required computational
effort by solving the (unsteady) RANS equa- Two-equation models solve two additional
tions in the near-wall region and applying LES equations for the eddy viscosity, one for the tur-
in the rest of the domain. bulence kinetic energy (k), and one for its dissi-
pation rate (typically ε or ω). These models have
3.1.2 RANS Models shown to be able to give accurate predictions in
ship hydrodynamics, especially certain versions
Turbulence modelling has been an important of the k   model and are by far the most ap-
research topic over the past decades. A large
plied ones (80% of the submissions for the
number of models have been proposed, tested
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 12 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

Gothenburg 2010 Workshop, see Larsson et al., 1) use of near-wall turbulence models that
2013). are able to resolve the flow all the way
down to the wall;
An important class of turbulence models, not 2) use of wall functions, semi-empirical for-
based on the Boussinesq hypothesis, are the mulae that used to bridge the solution var-
Reynolds-stress models, and versions thereof. iables in the fully turbulent log-law region
Rather than introducing an eddy-viscosity, they and the corresponding quantities on the
aim to solve the equations for the six Reynolds wall, without actually resolving the flow
stress components directly. Apart from that, ad- in between (i.e. the viscous-affected sub-
ditional equations have to be solved, since terms layer).
in these equations require modelling as well.
Consequently Reynolds-stress models are more Resolving flow by a near-wall model is a
computationally intensive, and often harder to more rational approach. However, it requires
converge, compared to one or two-equation grid nodes highly refined towards walls (y+=1 as
models. However, they contain more physics discussed in §2.3.6). This often leads to ex-
and are therefore be able to capture flow features tremely high aspect ratio for the near-wall grid
that cannot be solved by eddy-viscosity models. cells, not only resulting in large grids but also
Explicit algebraic Reynolds stress models are posing numerical instability problems in the so-
based on the algebraic expressions for each lution procedure, making the computations sub-
stress component which can be determined from stantially heavy. This is especially the case for
two equation turbulence models. They are con- full scale calculations.
sidered to be a reasonable compromise between
complexity and performance. Using wall functions eases the need of ex-
treme refinements near the wall. In terms of y+
3.2 Near-wall Modelling value, it suffices to have the first grid node in the
range 30<y+<100 (see §2.3.5). However, wall
In the near-wall region flow variables have functions are based on two-dimensional flow
very large normal gradients. In the viscous sub- typically at zero pressure gradients and it is
layer, viscous effects suppress turbulent fluctu- well-known that these analytical expressions be-
ations and have a dominant role in momentum come less valid, or even invalid, with increasing
transfer. This requires proper modelling of the adverse pressure gradients. Though corrections
near-wall region. Two approaches are available: exist for wall functions in the presence of pres-
sure gradients, these make the wall functions
less stable. Thus it is a trade-off between accu-
racy and computational effort. Wall functions
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 13 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

should be avoided if possible, and used with care 


unit, ymod  y   ks / 2 , and therefore the wall-
when necessary.
adjacent cell should be higher than k s .
3.3 Surface Roughness
3.4 Numerical Schemes
The flow in the boundary layer is not af-
fected by the surface roughness when the phys- In the majority of industrial CFD codes, dif-
ical roughness height k s is small compared to fusion terms in the governing equations are dis-
the thickness of the laminar sublayer. In partic- cretized using a second-order (central differenc-
ular, the effect of the surface roughness is negli- ing) scheme by default. Thus, spatial accuracy is
gible when k s is lower than approximately largely determined by discretization scheme
2 / u* , in this case the surface is said hydrody- used for convection terms.
namically smooth. For hydrodynamically rough
The first-order upwind (FOU) scheme, of-
surfaces, the logarithmic Law of the Wall is
fered in many commercial CFD codes often as a
modified adding a term, Δu  , which is a func-
tion of k s . Unfortunately this function depends default scheme, is very stable. However, it intro-
duces large numerical (artificial) diffusion - that
on the type of roughness and no universal for-
is why it is so stable. Its use can only be justified
mulation is known.
when the physical diffusion (turbulent + molec-
In most ship CFD applications, surfaces are ular) is much higher than the introduced numer-
assumed to be hydraulically smooth. Otherwise ical diffusion. This is true within the boundary
the effect of the roughness can be modelled ei- layer, allowing use of hybrid schemes that use
ther modifying the wall functions or the turbu- lower order near the wall and higher order far-
lence boundary conditions. When wall functions ther out.
are used, then different functions for Δu  are
used depending on the range of k s and the types The majority of high-order convection dis-
cretization schemes in popular use today for-
of roughness, such as sand grains or spheres. In-
mally have a second-order of accuracy with an
terested readers are invited to consult Schlicht-
upwind bias. All these second-order upwind
ing and Gersten (2000). No general guidelines
(SOU) schemes differ from one another in terms
can yet be given for wall-resolved boundary lay-
of the flux limiter used to suppress unphysical
ers. However, the roughness profile is typically
oscillations in the solutions. Still higher-order
not resolved, the blockage effect of the rough-
schemes such as fifth-order scheme exist; how-
ness is taken into account using a modified wall
ever, not all CFD solvers offer such higher-order
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 14 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

schemes. Moreover, it has to be taken into ac- 3.5 Free Surface


count that higher order schemes are, in general,
There are two major categories in free sur-
less robust than first or second order accurate
face models. For interface fitting approaches the
schemes. Therefore, their use should be consid-
numerical grid is conformed to the free surface
ered with special care. These schemes can be
shape. Interface capturing approaches define the
used in regions of high-quality grid (near or-
free surface as an iso-surface of a marker func-
thogonal, low aspect ratio, low expansion ratio),
tion and thus the grid does not have to conform
like in Cartesian overset refinements, while
to the free surface.
lower-order schemes can be used in others re-
gions. Higher-order schemes are standard in im- In general, the interface fitting approach is
mersed boundary methods in Cartesian grids. more accurate and efficient than the capturing
approach, since free surface boundary condi-
The SOU scheme is both reasonably accu-
tions can be applied in the exact free surface lo-
rate and robust, and for that reason is an indus-
cation. Therefore the interface fitting model
trial workhorse for convection discretization in
may be selected whenever possible. However, it
body-fitted approaches. The SOU scheme is
should be noted that a deformation/re-gridding
therefore recommended for all convection-dif-
procedure is essential in order to keep the grid-
fusion transport equations.
lines following the deformation of the free sur-
First order accurate schemes in time may be face. This may cause severe distortion of grid-
used only if a steady-state is sought. lines even though the initial grid conforming to
the undisturbed free surface has a good quality.
Although outside the scope of the present
guidelines, it is worth mentioning that second- Difficulties in grid generation and/or re-
order central differencing (CD) scheme is often gridding can be avoided with interface capturing
used in large eddy simulation (LES) and direct approaches. Also, for large deformation of free
numerical simulation (DNS) for its low-dissipa- surface, such as overturning or breaking waves,
tion that is critical to accurately resolve turbu- interface capturing methods should be used
lent structures. However, CD scheme is inher- since surface-tracking methods cannot resolve
ently unstable, raising convergence issues for the change in surface topology. Since the cap-
cases involving fine grids and small effective turing methods demand finer grid resolution
viscosity (large local Reynolds number). One near the interface, grid generation requires more
should consider using a stabilized form of cen- attention. The choice of level-set method or vol-
tral differencing. ume-of-fluid method in for capturing approach
has little impact in the final solutions. Although
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 15 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

the numerical methods are different, well imple- CFL condition is more demanding than in im-
mented flow solvers provide similar results for plicit solvers, where Cmax can be larger than 1.
both models. In this case the time step is decided by the flow
features. As a rule of thumb:
When a volume-of-fluid method is used, the
volume fraction equation requires special care,  for periodic phenomena (e.g. roll decay,
since the transported quantity is essentially a vortex shedding, incoming waves etc.) use
step function across the free surface, and tradi- at least 100 time steps per period;
tional convection schemes designed for convec-  for complex unsteady phenomena (e.g.
tion-diffusion equations perform poorly in this wetted transom instabilities), use at least
case. Though geometric reconstruction offers 20 time steps per period for the highest
the best possible solution, it is expensive and frequency to be resolved;
frequently pure transport is used, which does not  for rotating propellers use at least 180 time
guarantee a sharp interface throughout the com- steps per revolution;
putation. It has been found that convection  t must be smaller than 0.01 L/U if one or
schemes with some degree of downwind bias re- two equation turbulence models are used,
solve the sharp interface much better. while it should be smaller than 0.001 L/U
if Reynolds stress turbulence model is
3.6 Time Step used.

In explicit solvers the time step is chosen to


3.7 Parallel Computing
satisfy the Courant–Friedrichs–Lewy (CFL)
condition or to resolve the flow features of in- In modern computers, as a rule of thumb
terest, whatever results smaller. The CFL condi- choose the number of cores so that you use from
tion is a condition on the CFL number, which 50,000 to 500,000 grid points per core. In some
represents the number of points travelled by a cases fewer points per core can be used if the
particle of fluid within a time interval equal to CFD code scales well to obtain faster turna-
the time step: round, or more points per core if processor
memory allows it and turnaround time is not
|𝑢
̅|Δ𝑡
< 𝐶𝑚𝑎𝑥 (0) critical.
Δ𝑥

To maximize performance, try to distribute


where |𝑢̅| is the local velocity, Δt is the time
the load evenly between nodes. For instance, if
step, and Δx is the linear cell size in the flow
running in a Linux cluster with 2 quad-core pro-
direction. In explicit solvers, where Cmax  1 , the
cessors (8 cores) per node, and your case needs
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 16 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

10 cores, you can distribute your load in two away from perfect conservation of mass and
nodes using 8-2 or 5-5 configurations. The sec- momentum. Thus the residual for a discretized
ond balances the load per node better since it equation is defined as the L1-Norm of the imbal-
provides more memory bandwidth per process. ance between the left and the right hand side of
that equation over all the computational points.
Modern workstations with shared memory Usually the residual is scaled by a reference
are available with up to 64 processors, though value. Sometimes the L2-Norm or L∞-Norm are
much larger specialized systems are produced. used to define the residual.
High-performance clusters are typically cheaper
per processor for large systems (thousands of CFD users do not need to worry about the
cores) but use distributed memory. Shared- definition of residuals, as they are often pre-de-
memory systems allow all processors access all fined by code developers. Instead, attention
memory, resulting in easier programming and should be paid to the selection of convergence
better scalability of most applications. On the criteria. The recommended criterion is the drop
other hand, distributed memory systems provide of scaled residuals by at least three orders of
massive number of processors for very large magnitude off their initial values. However it is
computations. acknowledged that due to complexity of some
problems or oscillatory convergence this crite-
3.8 Convergence rion may not be satisfied. In such cases, the con-
vergence of forces and moments, and the con-
A number of convergence criteria should be vergence of the unknowns in key regions of the
defined and examined in order to ensure reliable flow field (e.g. at propeller plane for nominal
convergence of solution. While different levels wake calculations) can be monitored.
of convergence are acceptable, the uncertainty
due to the convergence should always be esti-
mated with a verification procedure. 4. POST-PROCESSING

Convergence should be achieved for both 4.1 Visualization


steady-state and transient computations. In the
A number of post processing plots should be
case of implicit transient computations, conver-
used as a minimum sub-set of information to en-
gence must be evaluated at each time step.
sure that the correct settings have been used for
The level of convergence should be assessed each computation. This should include contour
by the history of residual variations for the mass plots of the pressure coefficient, skin friction co-
and momentum equations. Residuals indicate efficient and y+ on the geometry surface, and
how far the present approximate solution is
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 17 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

vector plots of the boundary layer profiles along Any numerical results should be considered
the geometry. together with its numerical uncertainty. This be-
comes of critical importance when different de-
Reasonable checks should be carried out to sign solutions are ranked based on the CFD re-
ensure that the solution is physically sound and sults, since the uncertainty reflects, and allows
that it does not show the footprint of the grid. the computation of, the probability that the rank-
ing is correct.
4.2 Verification and Validation
The estimate of the uncertainty must be
The verification and validation of numerical based on a verification and validation proce-
simulations is an essential tool for the interpre- dures but it can be assumed that the uncertainty
tation of the results and the identifications of does not change significantly with small
those aspects of the simulation that can be im- changes of the boundary and initial conditions.
proved. Therefore, once the uncertainty has been esti-
mated for one simulation, this can often be used
The numerical uncertainty is due to the nu-
as a reasonable estimate of the uncertainty of
merical error introduced by the method used to
similar simulations. For example, if a parameter
solve the chosen set of equations with the given
study on the hull geometry is performed, verifi-
boundary and initial conditions, while the mod-
cation can be performed for only one or few val-
elling uncertainty is associated to the modelling
ues of such parameter. Care should be taken on
error, which is the difference between the exact
these extrapolations that no significant changes
solution of this set of equations and the reality.
on the flow field are observed between similar
The numerical uncertainty can be estimated per-
cases.
forming a set of numerical simulations, while
experimental data is necessary to estimate the It must be noted that the difference between
modelling error. the uncertainties of different quantities com-
puted with the same simulation can be higher
The ITTC procedure 7.5-03-01-01 provides
than one order of magnitude. For instance, the
“methodology and procedures for estimating the
uncertainty of the ship resistance is expected to
uncertainty in a simulation result”. The proce-
be significantly lower than the uncertainty of the
dure includes a verification procedure for the es-
local flow speed in one point of the domain.
timation of the numerical uncertainty, and a val-
Also, the higher the order of the derivatives of
idation procedure for the estimation of the mod-
the primitive unknowns, the higher the uncer-
elling error.
tainty expected. For instance, the uncertainty on
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 18 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

the velocity is expected to be lower than the un- Larsson, L., Stern F., Vissonneau, M. (2013)
certainty on the vorticity. Therefore, the uncer- Numerical Ship Hydrodynamics: An assess-
tainty must be estimated for every quantitative ment of the Gothenburg 2010 Workshop,
result gathered from the simulation. Springer.

5. REFERENCES 6. REFERENCES TO OTHER CFD


GUIDELINES
Hirsch, C. (1989) Numerical computation of in-
ternal and external flows: Fundamentals of AIAA, 1988, AIAA guide for the verification
numerical discretization, Wiley–Blackwell. and validation of computational fluid dy-
namics simulations, AIAA G-077-1998.
Hirsch, C. (1990) Numerical computation of in-
ternal and external flows: Computational ASME Committee PTC-61, 2009, ANSI Stand-
methods for inviscid and viscous flows, ard V&V 20. ASME Guide on Verification
Wiley–Blackwell. and Validation in Computational Fluid Dy-
namics and Heat Transfer.
Ferziger, J.H. and Peric, M. (2002) Computa-
tional Methods for Fluid Dynamics, ERCOFTAC Best Practice Guidelines.
Springer. http://www.ercoftac.org/publications/ercof-
tac_best_practice_guidelines/
Thompson, J.F., Warsi Z.U.A., Mastin, C.W.
(1985) Numerical Grid Generation: Founda- ITTC Recommended Procedures and Guide-
tions and Applications. lines. Uncertainty analysis in CFD, Guide-
lines for RANS codes, No. 7.5-03-01-02
Pope, S. (2000) Turbulent Flows, Cambridge
(1999).
University Press.
ITTC Recommended Procedures and Guide-
Sagaut, P. (2006) Large Eddy Simulation for In- lines. CFD User’s Guide, No. 7.5-03-01-03
compressible Flows, an Introduction, (1999).
Springer.
ITTC Recommended Procedures and Guide-
Schlichting, H., and Gersten, K. (2000) Bound- lines. CFD Verification, No. 7.5-03-01-04
ary-layer theory, Springer. (1999).

Wilcox, D.C. (2006) Turbulence Modeling for


CFD (Third Edition), DCW Industries.
ITTC – Recommended 7.5–03
–02–03
Procedures and Guidelines Page 19 of 19

Practical Guidelines for Ship CFD Ap- Effective Date Revision


plications 2014 01

ITTC Recommended Procedures and Guide- ITTC Recommended Procedures and Guide-
lines. Uncertainty Analysis in CFD Verifica- lines. Practical Guidelines for Ship Self-Pro-
tion and Validation. Methodology and Pro- pulsion CFD No. 7.5-03-03-01 (2014).
cedures, No. 7.5-03-01-01 (2008).
ITTC Recommended Procedures and Guide-
ITTC Recommended Procedures and Guide- lines. Practical Guidelines for RANS Calcu-
lines. Practical Guidelines for Ship Re- lation of Nominal Wakes, No. 7.5-03-03-02
sistance CFD, No. 7.5-03-02-04 (2014). (2014).

You might also like