THE ROYAL INSTITUTION OF NAVAL ARCHITECTS

Slamming computation on the multihull Groupama 3

Y. Roux, Company K-Epsilon, France

J. Wackers, Ecole Centrale de Nantes, France
L. Dorez, Team Groupama, France

SUMMARY
Difficulties that forced the maxi Groupama 3 to withdraw from the Jules Verne Trophy between Brazil and Africa in
November 2009 showed the need for a comprehensive study of the slamming phenomenon on hull bottoms in
unforeseen navigation conditions.
A survey in South Africa confirmed that structural damages were due to slamming stress on the rear of the windward
hull, which has been underestimated. This condition appears in a swell when the multihull takes off and crashes back
into the water at a boat speed of 30 to 40 knots.
The latest version of the software FINE™/Marine developed by CNRS, Ecole Centrale de Nantes and Numeca
International allows the precise simulation of this type of problem. Indeed, the use of an automatic grid refinement
algorithm enables the precise modelization of the impact of the hull on the water surface and the computation of the
generated pressure field, which will be used by the designers to reinforce the structure of the hull.
We present in this paper the particularities of the automatic refinement method and its application to the trimaran
Groupama 3 in the condition of the impact of the hull at an angle with respect to the sea.

1. INTRODUCTION of the Fluid Mechanics Laboratory at Centrale Nantes).

This paper presents the study of the impact of the bottom Turbulent flow is simulated by solving the
of a Groupama 3 float onto the sea surface. This incompressible unsteady Reynolds-averaged Navier-
phenomenon appears in very specific conditiions where Stokes (RANS) equations. The solver is based on the
the boat is sailing in reaching conditions against a swell finite volume method to generate the spatial
front. In this case the multihull jumps on the wave. The discretization of the transport equations. The face-based
impact on the waves at every landing induces extremely method is generalized to two-dimensional, rotationally
high loads to the float structure and deteriorate this symmetric, or three-dimensional unstructured meshes for
structure. which non-overlapping control volumes are bounded by
an arbitrary number of constitutive faces. The velocity
The numerical modelization of slamming needs a very field is obtained from the momentum conservation
accurate and robust unsteady hydrodynamic flow solver equations and the pressure field is extracted from the
able to capture the deformation of the free surface during mass conservation constraint, or continuity equation,
the impact. We present in this paper a new automatic transformed into a pressure equation. Free-surface flow
adaptive grid refinement method recently developped for is simulated with a multi-phase flow approach: the water
the software FINE™/Marine and we apply it to the surface is captured with a conservation equation for the
multihull Groupama 3 in an extreme condition of volume fraction of water, discretised with specific
slamming. compressive discretisation schemes, see [7].

2. NUMERICAL METHOD In the case of turbulent flows, additional transport

We use the software FINE™/Marine equations for modeled variables are discretized and
solved using the same principles. Most of the classical
2.1 The ISIS-CFD RANS solver at a glance linear eddy-viscosity based closures like the two-
equation k-ω SST model by Menter [6], for instance, are
The unsteady hydrodynamic RANS equations with free implemented. Two more sophisticated turbulence models
surface are solved using the FINE™/Marine software. are also implemented in the ISIS-CFD solver: an explicit
The mesh generator HEXPRESS included in algebraic Reynolds stress model (EARSM) [1], and a
FINE™/Marine offers pure-hex unstructured mesh Reynolds stress transport model [2].
generation allowing complex geometry meshing in
affordable turn-around time. The technique included for the 6 degree of freedom
simulation of ship motion is described by [5]. Time-
The ISIS-CFD RANS flow solver is developed by EMN integration of Newton's laws for the ship motion is
(Equipe Modélisation Numérique, the CFD Department combined with analytical weighted or elastic analogy
grid deformation to adapt the fluid mesh to the moving derefinement of cells may be prevented. At the end of the
ship. decision step, before a single cell is actually refined, the
refinement for the entire grid is known.
2.2 The refinement method
Finally, the actual grid adaptation is performed; cells are
Recently, an automatic grid refinement capability has divided into smaller cells, or existing small cells are
been integrated in ISIS-CFD [8]. This technique is to be merged back into their original large cells. Care is taken,
used on a day-to-day basis for all the different after the treatment of each individual cell, to leave a valid
applications of the flow solver and it is supposed to mesh with all the pointers between cells, faces, and nodes
evolve effortlessly as the possibilities of the solver grow in place, even if one knows that certain pointers will be
over the years. Therefore, the mesh adaptation method changed again when a neighbour cell is refined later on.
has been made general: it is suited for unstructured grids This guarantees that, when a cell is refined, it does not
around the complex geometries commonly treated, have to distinguish between neighbour cells that are
directional refinement is used to keep the size of 3D refined, that will be refined, or that remain unchanged.
refined grids low, and derefinement of refined grids is The added flexibility and robustness of the code are well
included to enable unsteady flow simulation. Like the worth the extra work this represents. The refinement step
flow solver, the refinement method is fully parallel. The includes automatic load balancing with the ParMETIS
refinement procedure is completely integrated in the flow library [4].
solver.
As the grid refinement method mainly changes the
During a flow calculation with adaptive grid refinement, topology of the grid, while the weighted mesh
the refinement procedure is called repeatedly to keep the deformation technique of ISIS-CFD influences only the
grid permanently adapted to the developing solution. position of the mesh nodes, the combination of these two
Globally, the method works as follows: the flow solver is techniques is feasible without large modifications to any
run on an initial grid for a limited number of time steps. of them. Therefore, unsteady grid refinement can be
Then the refinement procedure is called. If a refinement performed for simulations with moving bodies, without
criterion, based on the current flow solution, indicates major difficulties.
that parts of the grid are not fine enough, these cells are
refined and the solution is copied to the refined grid. On 2.3. Refinement criterion
this new grid, the flow solver is restarted. Then the
refinement procedure is called again, to further refine or The criterion chosen for this study refines the grid in the
to derefine the mesh. This cycle is repeated several times. neighbourhood of the water surface. Directional
When computing steady flow, the procedure eventually refinement is employed to refine in the direction normal
converges: once the flow starts to approach a steady state to the surface only. Where the free surface is diagonal to
and the grid is correctly adapted to this state according to the grid directions, isotropic refinement is used, but
the refinement criterion, then the refinement procedure where the surface is horizontal, directional refinement is
keeps being called, but it no longer changes the grids. chosen; the resulting zone of directional refinement
For unsteady flow, the mesh keeps changing as it adapts includes the undisturbed water surface, as well as smooth
itself to the flow in evolution. wave crests and troughs. This is essential to keep the
number of grid cells low, as the water surface is often
To ensure robustness and flexibility of the grid nearly undisturbed in most of the domain. Figure 1 gives
refinement method, it is divided into three distinct parts an illustration of this refinement principle.
that exchange only minimal information. First, the
refinement criterion is computed, a scalar or tensor field
that indicates the local desired cell size. This criterion
should be based on the flow field, it may be computerd
from any quantity desired. To permit the user choice of
refinement criteria and the easy incorporation of new
refinement criteria in the code, the criterion does not
have to take into account the type or the orientation of
the cells; in general it is a smooth field variable.

Based on this criterion, the decision is taken which cells

to refine or to derefine. While this decision may depend
on the type of the cells, it does not depend on the specific
way in which the refinement criterion is calculated. It Figure 1: Isotropic and directional refinement at the free surface. The
remains the same for any refinement criterion. During the curves represent volume fraction isolines.
refinement decision step, the decision in each cell is
adapted to its neighbour cells: to guarantee the quality of Directional refinement normal to the free surface implies,
the mesh, extra cells may need to be refined, or that the refinement criterion imposes a constraint on the
cell size in that normal direction only. The cell size in all surface region has to satisfy the following
other directions is kept as large as possible. Thus, cells condition:
that are oriented diagonally to the interface must Dz = 0.001 Lpp
necessarily be refined in all directions, but cells that are
aligned with the interface can be refined in one direction Figures 3 shows the detail of the mesh around the hull.
only.

This criterion is useful to increase the precision for

simulations with free-surface waves. While it is
applicable for any type of wave, it is especially succesful
for cases where small local wave details play an
important role. Thus, the simulation of slamming, which
involves localised pressure peaks, is a natural field of
application.

3. IMPACT STUDY OF GROUPAMA 3 FLOAT

3.1. General presentation

Figure 3. General view of the Gma3 hull mesh.
The impact case chosen for this study is a computation of
3.2 Input data, boundary conditions:
the float with a trim angle of 2° to create the dangerous
configuration with the rear part of the hull parallel to the
We present the input data and the boundary conditions
free surface. We impose the movement of the float with
for the computation:
an horizontal and vertical boat speed, Figure 2.
• Boundary conditions for the volume mesh. To
The velocities imposed in the horizontal and vertical
take into account the movement of the float, the
motion correspond to the boat speed in sailing condition
computation is realized with an imposed motion
with a vertical motion estimated by video measurement.
of the volume mesh in Z direction and X
direction.
For reasons of confidentiality we give the vertical
The boundary conditions are:
velocity as a function of horizontal one:
o Velocity imposed on the front (Xmin), back
Uz = -0.277 Ux
(Xmax), and side (Ymin) surfaces.
o Hydrostatic pressure imposed on the bottom
(Zmin) an the top plane (Zmax).
o Mirror conditions on the symmetry plane of
the hull.
o Wall function on the hull surface.

• Refinement input data. For the refinement we

Figure 2 Geometry and frame of reference need to specify several data. The most important
3.2 Mesh Generation of these are:
The mesh of the float is generated with the float o The region near the float where the
positioned 0.1 m above the free surface. A full hexaedral refinement will be active. This region is
unstructured mesh of 2 million cells has been generated defined as a parallelepiped box where the
using HEXPRESS on the half part of the domain. The refinement in X, Y and Z-direction is
initial mesh is pre-refined is the region of the impact (on active. In our case we choose the size of the
the hull and also on the free surface at Z=0) box as:
Common criteria for good boundary layer description 0 < X <0.5 LPP
and free surface resolution have been respected: 0 < Y < 0.01 LPP
• The thickness of the first cell next to the wall in o The threshold is the target size for the cells
the boundary layer yp is defined as function of after grid refinement. We realized the
y+, Reynolds number Re, friction velocity Uτ and computation with a threshold value of
the waterline length Lpp: DZt=0.0002 LPP with a diffusion of 2
y+ (which means the refinement is spread out
yp = Lpp to a zone of two cells around the cells
Re.U τ indicated by the refinement criterion).
The use of a law wall with y+ = 80 leads to
yp=0.0004 m The refinement criterion and the limiting box
• Best practice in FINE/Marine in the case of combined cause grid refinement around the
resistance computation has shown that the water surface, in a region that moves with the
vertical height of the cells Dz inside the free
 float, there where the thickness of the cells is local pressure induced by the impact of the keel
superior to DZt. line with the free surface. This impact is more
localized and less dangerous for the structure.
To reduce the number of cells and the The value of constant pressure is reduced to one,
computation time, the derefinement is also which corresponds to the dynamic pressure of the
active. Once refined cells are situated under the flow on the hull.
free surface they are merged back into the
original coarse cells and the mesh is the same as
before the refinement.

• Time step. The choice of the time step is one of

the must important parameters for the
modelization of slamming. The highest
pressure induced by the impact appears at the
very first contact between the hull and the
water, the time step has to be small enough to
capture this phenomenon. We choose to use a
small time step which is adapted to give a
constant maximum Courant number computed
on the free surface. At the begining the time step
is DT=0.001 s and the smallest value is
DT=0.00009 s.

4. RESULTS
Figure 4: Evolution of the pressure at Y=0 from the transom (X=0 m)
to the bow (X=30 m). From T = 0 s [1] to T = 0.021 s [10]
For reasons of confidentiality, all the pressure values are
given in non-dimensional form.

The unsteady computation describes 0,028456 s of the

phenomenon. As we explained before the time step is not
constant but adapted at every iteration according to the
Courant number fixed to 5 for this computation.

We decomposed the impact in two phases: phase 1 [1 to

10] where the pressure increases and phase 2 [1 to 10]
where the pressure decreases. We show the different data
at every 4 iterations from 0.013 s to 0.025049 s. Table 1
gives the time for each item in the two phases.

Item Time Phase 1 (s) Time phase 2 (s)

1 0,013000 0,021642
2 0,016399 0,022026
3 0,017415 0,022406
4 0,018221 0,022798
5 0,018989 0,023139
6 0,019722 0,023488
7 0,020402 0,023841
Figure 5: Evolution of the pressure on Y=0 from the transom (X=0 m)
8 0,020889 0,024222 to the bow (X=30 m). From T = 0.021 s [1] to T = s 0.025 s [10]
9 0,021173 0,024627
10 0,021642 0,025049 In the figures 6 and 7 we present for item [2, 4, 6, 8 and
10] of the two phases the non-dimensional pressure field.
Table 1: Time for each item in the two phases. We observed an increase of the pressure until a constant
In the figures 4 and 5 we present the variation of the non- value of a value around 3 in phase 1 and a decrease of
dimensional pressure on the hull at Y=0 from the this pressure in phase 2.
transom (X=0 m) toward the bow (X=12 m). The geometrical characteristic of the zone on the hull
• Phase 1. The pressure impact appears first near where there is the dangerous development of the constant
the transom and develops itself as a constant pressure [8 and 10] of phase 1 is 7 m length and 0.3 m
value between 1 and 8 meters from the transom. width. As we explained before, the evolution of the non-
At the end of the phase 1, at 0.0216 s, the dimensional pressure peak in the phase 2 is very
constant value of the pressure is around 3. concentrated near the point of impact of the keel line
• Phase 2. We observed a decrease of the constant with the free surface. The impact goes foward towards
value pressure from 3 to 1 and an increase of the bow until the landing of the float.
 free surface they are merged back into the original coarse
cells and the mesh is the same as before the refinement.
In the figure 9 (ΔT = 0.001 s) and figure 10 (ΔT=0.0007
s) this is less refinement and derefinement visible
between the two pictures, because the delay is shorter
and the original mesh is finer behind the hull than on the
side of the hull.

Figure 8: Mesh in the section X=5m, evolution of the refinement from

the begining to [2] in phase 1

Figure 6: Field of the non-dimensional pressure on the hull, Phase 1.

Figure 9: Mesh in the section X=5m, evolution of the refinement from

the [4] to [6] in phase 1

Figure 7: Field of the non-dimensional pressure on the hull, Phase 2.

In the figures 8, 9 and 10 we present an illustration of the

automatic refinement methodology using for this study
For the cases [2], [4], [6], [8] and [10] of the phase 1 we Figure 10: Mesh in the section X=5m, evolution of the refinement from
show the mesh section at X = 5m. (For a better [8] to [10] in phase 1
understanding we show in each picture the hull in the We present in figure 11 a view of the deformation of the
same position, while the free surface is in motion). free surface at T= 0.0505 s. We observe on the two sides
In the two pictures of figure 8 (ΔT = 0.0114 between the of the float the development of a breaking wave.
two pictures) we see the illustration of the refinement and
derefinement: when refined cells are situated under the
 4. KARYPIS, G. and KUMAR, V., ‘Parallel multilevel
k-way partitioning scheme for irregular graphs’, SIAM
Review, Vol. 41, No. 2, pp.278-300, 1999.

5. LEROYER, A. and VISONNEAU, M., ‘Numerical

methods for RANSE simulations of a self-propelled fish-
like body’, J. Fluid & Structures, Vol. 20, No. 3, pp.975-
991 , 2005.

6. MENTER, F.R., ‘Zonal two equation k-omega

turbulence models for aerodynamic flows’, AIAA Paper
93-2906, 1993.
Figure 11: View of the deformation of the free surface at T=0.0505 s.
7. QUEUTEY, P. and VISONNEAU, M., ‘An interface
capturing method for free-surface hydrodynamic flows’,
5. CONCLUSIONS Computers & Fluids, Vol. 36, No. 9, pp.1481-1510,
2007.
In this paper we presented a new methodology of
automatic mesh refinement based on the size of cells for 8. WACKERS, J. and VISONNEAU, M., ‘Adaptive grid
the good description of the free surface. refinement for free-surface flow computation’,
The new automatic adaptive grid refinement and Proceedings of ECCOMAS MARINE 2009, 2009.
derefinement method is very well adapted to compute the
difficult unsteady phenomenon of impact.
The computed dynamic pressure results have been used 8. AUTHORS BIOGRAPHY
by Team Groupama to better understand the phenomenon
and to calibrate design methods. Yann Roux holds the current position of C.E.O. at
company K-Epsilon. He is the funder of the company K-
In the future, the new method presented in this paper will Epsilon. In partnership with laboratory he develops tools
serve as an efficient tool for naval architects and and proposes studies in unsteady hydrodynamic and
engineers to define the extreme load cases applied to aerodynamic computation.
yacht structure by the slamming effects. The traditional
methods used in the past and the application of linear Jeroen Wackers is Research Engineer in the Fluid
coefficients to hydrostatic loads in order to quantify Mechanics laboratory at Ecole Centrale de Nantes. He is
slamming effects can now be refined by efficient in charge of the research concerning adaptive and goal-
dynamic methods to improve the knowledge for a boat oriented computation for ISIS-CFD.
speed of 30 to 40 knots.
Loïc Dorez holds the current position of Head of Design
6. ACKNOWLEDGEMENTS Office at Groupama Sailing Team. In partnership with
laboratories, design structure compagnies, aerodynamic
This work was performed using HPC resources from and hydrodynamic compagnies he organises
GENCI-IDRIS (Grant2010-x2010021308). developpements of tools and shares sailing experience on
various subjects apply during the last ten years race
7. REFERENCES multihulls.
1. DENG, G.B. and VISONNEAU, M. ‘Comparison of
explicit algebraic stress models and second-order
turbulence closures for steady flows around ships’,
Proceedings of the 7th International Conference on
Numerical Ship Hydrodynamics, Nantes, France, 1999.

2. DUVIGNEAU, R., VISONNEAU, M. and DENG

G.B., ‘On the role played by turbulence closures in hull
shape optimisation at model and full scale’, J. Marine
Science and Technology, Vol. 8, No. 1, pp.1-25, 2003.

3. HAY, A. and LEROYER, A. and VISONNEAU, M.,

‘H-adaptive Navier-Stokes simulations of free-surface
flows around moving bodies’, J. Marine Science and
Technology, Vol. 11, No. 1, pp.1-18, 2006.