Вы находитесь на странице: 1из 17

Comput Mech

DOI 10.1007/s00466-017-1505-1


Accurate prediction of complex free surface flow around a high

speed craft using a single-phase level set method
Riccardo Broglia1 · Danilo Durante1

Received: 23 May 2017 / Accepted: 25 October 2017

© Springer-Verlag GmbH Germany, part of Springer Nature 2017

Abstract This paper focuses on the analysis of a challeng- bers (from 0.6 up to 1.2). In the present work, the planing hull
ing free surface flow problem involving a surface vessel is treated as a two-degree-of-freedom rigid object. Flow field
moving at high speeds, or planing. The investigation is is characterized by the presence of thin water sheets, several
performed using a general purpose high Reynolds free sur- energetic breaking waves and plungings. The computational
face solver developed at CNR-INSEAN. The methodology results include convergence of the trim angle, sinkage and
is based on a second order finite volume discretization of resistance under grid refinement; high-quality experimental
the unsteady Reynolds-averaged Navier–Stokes equations data are used for the purposes of validation, allowing to com-
(Di Mascio et al. in A second order Godunov—type scheme pare the hydrodynamic forces and the attitudes assumed at
for naval hydrodynamics, Kluwer Academic/Plenum Pub- different velocities. A very good agreement between numer-
lishers, Dordrecht, pp 253–261, 2001; Proceedings of 16th ical and experimental results demonstrates the reliability of
international offshore and polar engineering conference, San the single-phase level set approach for the predictions of high
Francisco, CA, USA, 2006; J Mar Sci Technol 14:19–29, Froude numbers flows.
2009); air/water interface dynamics is accurately modeled by
a non standard level set approach (Di Mascio et al. in Com- Keywords Planing hull · Ship hydrodynamics · Fluid/rigid
put Fluids 36(5):868–886, 2007a), known as the single-phase body interaction · Free surface · Level set
level set method. In this algorithm the governing equations
are solved only in the water phase, whereas the numerical
domain in the air phase is used for a suitable extension of the 1 Introduction
fluid dynamic variables. The level set function is used to track
the free surface evolution; dynamic boundary conditions are The study of semi-displacement and planing vessels is an
enforced directly on the interface. This approach allows to actual topic since the wide applications range, spanning from
accurately predict the evolution of the free surface even in the sport competitions to patrol boats. From a physical point of
presence of violent breaking waves phenomena, maintaining view, the high speeds reached by these vessels imply a com-
the interface sharp, without any need to smear out the fluid plex interaction between the structure and the free surface
properties across the two phases. This paper is aimed at the of the fluid with a complicate wave pattern which develops
prediction of the complex free-surface flow field generated around and downstream the ship itself. The impact of the
by a deep-V planing boat at medium and high Froude num- vessel on the water surface during its motion gives rise to
the formation of water jets with significant air entrapment
B Riccardo Broglia and bubbles production. Actually the side jet is substantially
riccardo.broglia@cnr.it definable in terms of a spray sheet, the dynamics of which
Danilo Durante is highly difficult to study both in experimental and numer-
danilo.durante@cnr.it ical approaches. In addition, in luxury yachts like the one
1 under analysis in the current work, the ship hull is charac-
CNR-INSEAN, Marine Technology Research Institute,
National Research Council, Via di Vaellerano 139, terized by the presence of chines on the side walls, which
00128 Rome, Italy makes the dynamics of these water jets even more compli-

Comput Mech

cated due to both high non linear effects (such as strong is still rather limited. Fu et al. [29] exploited an immersed
breaking wave phenomena) and topological complications boundary approach together with a VOF method [the numer-
(e.g. plungings and ricochets). Indeed, these chines, other to ical flow analysis (NFA) code] for the study of the forces
aesthetic aspects, are intended to throw sprays away from and moments on a deep-V planing craft. The impact of a
the sides of the hull, preventing the rise of the water up the wedge on the free surface was numerically simulated and
hull sides. Their role relies also in creating a smooth ride experimentally validated as test case.
in seaway and, for those with a wide flat area (called chine Planing hull validation studies were performed in [42]
flats) to significantly contribute in providing an additional lift using an unsteady Reynolds averaged Navier–Stokes
allowing the vessel to acquire a planing regime. (uRaNSe) based solver, using the historical benchmark
Planing vessel have been widely investigated by means of experiments of Fridsma. Simulation included deep and shal-
both experimental and theoretical studies; early experimen- low water tests, in both fixed and free sinkage and trim
tal studies date back to 1970s with the study of [27,50] on conditions, as well as regular and irregular head waves in
prismatic hulls. More recently, Judge and Ikeda [35] tested deep water. Detailed verification and validation were reported
a four foot berglass planing hull model in calm water, regu- with satisfactory results. Recently, Akkerman et al. [2] made
lar and irregular waves. For each test run, the model motion a numerical investigation at different Froude numbers on
and the water spray were recorded using high-speed video, the Fridsma planing hull through a Arbitrary Lagrangian
the impact pressures were recorded using pressure sensors Eulerian Variational Multiscale (ALE-VMS) approach. A
arrayed on the model bottom, the model accelerations were comparison with the experimental data showed a very good
recorded at the bow. The examination of the wave impacts of agreement for low Froude numbers, while larger discrepan-
the model was carried out as a sequence of individual impact cies (of the order of 10–12%) are found on high speed cases.
events. Begovic et al. [5] appraised the effect of deadrise In [30] results from a collaborative research effort involving
angle variation along the hull length on seakeeping character- the computational fluid dynamics (CFD) codes CFDShip-
istics in regular waves for one monohedral and three warped Iowa and NFA were presented and discussed to examine
models. The deadrise angle was considered linearly vary- the hydrodynamic forces, moments, hull pressures, accel-
ing along the hull length with the aim to get an insight on erations, motions, and the multiphase free surface flow field
motion and acceleration properties connected with warped generated by a planing craft at high-speed (Froude number
hull forms and to provide a benchmark for CFD of plan- Fr = 1.8−2.1) in calm water and waves. The steady forward
ing hulls. In [28] calm water experiments were performed speed test were performed for two kind of hulls: a prismatic
on a Deep-V planing craft model, with the main aim to and a double-stepped planing hulls. The sinkage and resis-
support development of computational fluid dynamics codes tance were in good agreement for the two codes for the last
for high-speed, small-craft applications. The measurements configuration. The irregular wave simulations were carried
included resistance, sinkage and trim, hull pressure, longitu- out with CFDShip-Iowa and compared with experiments,
dinal wave-cuts, and bow-wave and stern-wake topologies. showing a well modelled overall behaviour. Moreover, in [7]
Tests were conducted over a Froude number range of 0.31– an extensive study using an up to date RANSE VOF solver
2.5. Some numerical simulation was also performed utilizing with free surface tracking capability was performed, testing
the numerical flow analysis (NFA) code and compared to the the method on a wedge shaped prismatic planing hull, with
model test results. a constant deadrise angle of 20◦ , systematically varying the
Most of the theoretical/numerical studies were based on running trim angle and wetted length. Results obtained, in
the slender body assumption or on simplified (potential) the- terms of drag and lift forces as well as longitudinal trim-
ory [26,49,60,68], with only few attempts made to face the ming moment, were compared with available experimental
full three-dimensional problem [36,37]. By using simplified and semi-empirical theories. Very recently, some interest-
2D + t theory, the free surface bow flow around a fast and ing studies concerned the fluid structure interaction (FSI)
fine ship in calm water was also studied by Landrini et al. problems: full scale validation for one-way FSI of composite-
[38] with an emphasis on generation and evolution of the panel slamming of a fast planing craft was shown in [63]
breaking and splashing bow wave. and extended to two-way FSI in [64]. FSI approach included
In this context the use of computational fluid dynamics tightly coupled (at inner iteration level) partitioned CFD and
(CFD) tools for a deeper investigation of the hydrodynamic of computational structural dynamics (CSD) solvers. CFD was
high speed planing hull could be extremely helpful. Indeed, solved by URANS, whereas CSD was solved by finite ele-
even if highly demanding in terms of CPU time requirements, ments and modal expansion. Simulations were conducted at
they have been demonstrated to be mature enough for the significant deterministic conditions for head and following
analysis of the performances of displacement ships (see for waves, and compared to stochastic full-scale trials.
example [39,40,48,55,58]). Nevertheless, the application of The objectives of the present paper are to investigate the
CFD based approach to the study of high speed planing craft flow field around a high speed planning hull and to demon-

Comput Mech

strate the capability of the in-house numerical tool (χ navis) in it is shown that the code is able to correctly reproduce the
dealing with challenging problems where energetic unsteady very complex phenomenon of the formation of the jet, the
wave breaking phenomena characterize the wave pattern. The consequent wave breaking dynamics and the multiple splash
numerical solver is based on a finite volume discretization of ups and ricochets. The solver is also shown to be able to
the uRaNS equations for high Reynolds number free surface accurately reproduce the high non-linear flow field around the
flows around complex geometries. The scheme is globally chines, where the spray sheet impinges and rapidly deviates
second order accurate. Several turbulence models have been far away the side of the vessel.
implemented in the code, ranging from the one equation In the following section the adopted numerical model is
Spalart and Allmaras model [56], to the two equation k−ε reported in details. After the presentation of the geometry,
model [13] as well as DES and DDES hybrid models. Com- the test cases and the numerical parameters, a section with
plex geometries and multiple bodies in relative motion are the numerical results follows. Firstly, integral quantities such
handled by a dynamical overlapping grid approach [17]. High as total resistance, trim and sinkage are analyzed, including a
performance computing is achieved by an efficient shared and rigorous assessment of the verification and validation of the
distributed memory parallelization [11]. Free surface flows numerical estimations. Secondly, the flow field is analyzed,
are handled by a single-phase level set approach [18]; in this with the emphasis on the surface pressure and the wave pat-
methodology the zero level of a level set function is used as a tern. Conclusions and some observation of future research
tracking device to locate the free surface position. The gov- activities will close the paper.
erning equations are solved in the water phase only, whilst a
suitable extrapolation of the fluid dynamics variables is per-
formed in the air phase. It has to be highlighted that, the air 2 Mathematical model
phase is non-physical, therefore, by definition, this methodol-
ogy can be only applied when any air effect can be neglected. The high-Reynolds turbulent motion of an incompressible
Free surface dynamical boundary conditions are enforced (constant density) viscous fluid can be described by the well
directly on the zero level of the function. Consequently, the known unsteady Reynolds averaged Navier–Stokes equa-
interface is maintained intrinsically sharp regardless the com- tions (uRaNSe). They express the conservation of the mass
plexity of the free surface shape and of its dynamics, with and momentum, therefore, considering a fixed control vol-
an accuracy comparable with Lagrangian or ALE formula- ume V bounded by the surface S(V), they can be written in
tions (see for example [3,15]), or interface capturing based integral vectorial form as
mathods accomplished with suitable discretization schemes  
for the treatment of the density discontinuity [45,46]. ∂
 q dV + [F c (q) − F d (q)] dS = 0 (1)
The single phase methodology has been extensively ∂t V S (V )
applied to describe several marine hydrodynamic problems,
such as free surface flows in presence of wave breaking phe- where q = ( p, u, v, w)T is the state variables vector for
nomena (see for example [18]), maneuvering [10,12,22] and incompressible flows,  = diag (0, 1, 1, 1) and
sea keeping [20] studies of surface vessels, as well as subma- F c = (u l nl ; u 1 u l nl + pn 1 ; u 2 u l nl + pn 2 ; u 3 u l nl + pn 3 )T
rine [23], catamarans [9,67] and offshore structures [43]. In F d = (0; τ1l nl ; τ2l nl ; τ3l nl )T (2)
previous works [8,33], the code has been applied to similar
problems studied here; in particular, in these works the effect are the convective (inviscid and pressure) and diffusive nor-
of the impact of a symmetric and an asymmetric wedge on the mal fluxes through the surface S(V) with outward unit normal
free surface with the analysis of the wave pattern generated n. A reference length L and velocity U∞ have been cho-
and the jet dynamics have been investigated. sen to make the previous equations non-dimensional. u i is
In the present paper, the towing simulations of a semi- the ith Cartesian component of the velocity vector (in the
displacement vessel are carried out for Fr spanning from 0.6 following, they will be also denoted as u, v, and w); p is
to 1.2; sinkage and trim, and resistance curves are evaluated a new variable related to the pressure P and the acceler-
and compared with experimental results kindly provided by ation of gravity g (parallel to the vertical axis z, positive

Azimut–Benetti s.p.a. Results show a very good agreement upward) by p = P + (z − z FS )/Fr 2 , Fr = U∞ / gL being
with experimental values, allowing an accurate analysis of the Froude number and z FS is the (undisturbed) free sur-
the problem under consideration and proving the capabilities face quote. Finally, τi j = νt (∂ j u i + ∂i u j ) is the stress tensor,
of the single-phase level set approach in dealing with these νt = 1/Re + νT is the global kinematic viscosity, with
rather complex problem. The analysis of the surface pressure Re = U∞ L/ν the Reynolds number, ν the kinematic vis-
shows that the flow field is characterized by localized high cosity and νT the turbulent viscosity. In the present work, the
pressure values which give rise to a energetic water sheet and, turbulent viscosity has been calculated by means of the [56]
consequently, a rather complex wave pattern. Nevertheless, one-equation model.

Comput Mech

The problem is closed by enforcing appropriate conditions and air phase) as a signed distance from the free surface,
at physical and computational boundaries. On solid walls, conventionally assumed positive in air phase and negative in
the velocity is set equal to zero and zero normal gradient the water phase. This function is enforced to be a material
is enforced on the pressure field; at the (fictitious) inflow surface for t > 0
boundary, velocity is set to the undisturbed flow value and
the pressure is extrapolated from the inside; on the contrary, ∂φ(x, y, z, t)
+ u · ∇φ(x, y, z, t) = 0
the dynamic pressure is set to zero at the outflow, whereas ∂t (6)
φ(x, y, z, 0) = d(x, y, z)
the velocity is extrapolated from inner points. At the free
surface, whose location is one of the unknowns of the prob-
Here u is the velocity of the underlying flow and d(x, y, z) is
lem, the dynamic boundary condition enforces the continuity
the signed distance from the free surface at t = 0. The zero
of stresses across the surface; if the effects of the air are
level of φ(x, y, z, t), being convected by the flow through
neglected, the dynamic boundary condition reads
Eq. (4), will always represent the free surface location. How-
κ z − z FS ever, the solution of the problem (6) does not guarantee that
p = τi j n i n j + + τi j n i t 1j = 0 the function φ(x, y, z, t) remains a distance function for
We Fr 2
t > 0; in order to ensure this property, in the so–called re-
τi j n i t 2j = 0 (3)
initialization step [54], φ(x, y, z, t) is replaced, at each time
step, by a new function with the same zero level of the for-
where We = ρ U∞ 2 l/σ is the Weber number (ρ being the
mer one and field values that correspond to the actual signed
density of the fluid and σ the surface tension coefficient) and distance from the interface. With a notation abuse, the same
κ the surface curvature; n, t1 and t2 are the surface normal symbol φ will be used for both functions. This new function
and two tangential unit vectors, respectively. The kinematic is found as the steady state solution of the so-called Eikonal
boundary condition states that the free surface is a material equation
surface; therefore, by defining its location as the zero level of
a three-dimensional function F(x, y, z, t) = 0, its unknown ∂φ
+ sign(φ)(|∇φ| − 1) = 0 (7)
configuration can be determined by the following evolution ∂t
It is important to highlight that, due to the hyperbolic nature
D F(x, y, z, t) of Eq. (7), the solution evolves as a distance close to the zero
=0 (4)
Dt of the level set function first, then it propagates outward in
the whole domain [59,62]. Since in the level set approach the
Initial conditions have to be specified for the velocity field values of φ(x, y, z, t) are required only in a thin region very
and for the free surface configuration close to the interface, it is not mandatory to resolve to high-
light that, due to the hyperbolic nature of Eq. (7) throughout
u i (x, y, z, 0) = u i0 (x, y, z) F(x, y, z, 0) = F0 (x, y, z) (5) the whole computational domain: i.e. Eq. (7) is not integrated
up to achieving a steady state solution [18,53].
2.1 Single-phase level set approach In the single-phase level set method only the liquid phase
of the fluid is computed, with the zero level of φ(x, y, z, t)
The free surface is handled by means of a single-phase let used to locate the actual position of the free surface. It has
set approach; the principal features of this approach are to be highlighted that the level set function is intrinsically
described here, the interested reader will find a more exhaus- treated as a sharp interface; this allows to accurately describe
tive treatise on [18]. the free surface with a limited number of grid points, which is
In what follows, we will generally refer to air and water of paramount importance when dealing with high Reynolds
phases in order to distinguish between fluids with different number flows around complex geometries. One the contrary,
densities/viscosities; clearly, no assumption on the physical due to stability and robustness reasons, in the classical two-
or chemical properties of such fluids are made a priori. It phase approach the size of the transition region between water
is important to remark that the only assumption is that the and air has to be smeared out in about 10 grid cells [34]; this
non-resolved phase is constituted by a fluid not affecting the transition zone can result excessively thick, given the limited
dynamics of underlying flow field. However, the methodol- number of points available.
ogy is generalizable and applicable to a problem regarding As it will be shown in the following, in the single level
two different fluids respecting the assumption made. set approach, the values of the velocity and pressure in the
In the level set approach [44,54,59], a smooth function air phase are not required. Nevertheless, their estimation is
φ(x, y, z, t), whose zero level coincides with the free surface, of great importance during the iterative procedure at those
is defined in the whole physical domain (i.e. in both liquid points that change their physical state from air to water, for

Comput Mech

which an initial guess is needed. Values in the air phase is

simply computed as an extension of the velocity and pressure
fields from the water region

∇φ · ∇u i = 0 i = 1, 2, 3
∇φ · ∇ p = 0

where similar relations hold for any other field variable (such
as the variables related to the turbulence model). As it has
been demonstrated in [1], velocity extrapolation in the air
region with Eq. (8), guarantees that the level set function Fig. 1 Computational domain. Purple: water cells; Yellow: air cells.
evolves as a distance function also at the interface air cells. Full symbols: interface cells (either water or air cells)
It has been found more convenient to split the level set
function as
3 Numerical model
φ(x, y, z, t) = ϕ(x, y, z, t) + (z − z FS ) (9)
The main idea underlying the method is based on the decom-
with z the vertical axis aligned with the acceleration of grav- position of the physical domain in three phases: the water,
ity, positive upward; the function ϕ(x, y, z, t) represents the the free-surface and the air. Once identified the free-surface
disturbance of the free surface from the unperturbed (flat) position as the zero level of the level-set function φ, the
configuration. Therefore, with this position, inflow bound- fluid equations are solved in the water phase only, whilst
ary condition for the function ϕ(x, y, z, t) simply reduces to in the air phase the solution is extrapolated. The kinematic
ϕ(x, y, z, t) = 0. and dynamic boundary conditions are also enforced on the
With the replacement of φ(x, y, z, t) according to the free-surface. In order to keep the level-set function φ as the
relation (9) the equation of the level set function is straight- distance function, a reinitialization step is periodically per-
forward changed as formed by solving the Eikonal equation. The time evolution
of the solution is obtained as the steady condition in the
∂ϕ pseudo-time. The algorithm is deeply described in the fol-
+ u · ∇ϕ + w = 0 lowing sub-sections.
∂t (10)
ϕ(x, y, z, 0) = d(x, y, z) − (z − z FS )
3.1 The general algorithm
whereas, the Eikonal equation becomes
  The system of equations (1) is approximated by a finite vol-
∂ϕ ∇φ ume technique, with pressure and velocity co-located at the
+ sign(φ) · ∇ϕ
∂t |∇φ| cell center. The fluid domain D is partitioned into Nl struc-
1 ∂φ tured adjacent or overlapped blocks Dl , each one subdivided
+ sign(φ) −1 =0 (11) into Ni × N j × Nk disjoint hexahedrals Dijk
l . As sketched in
|∇φ| ∂z
Fig. 1, in the computational domain are congruently identi-
By defining uφ = sign(φ)∇φ/|∇φ| and b = sign(φ)(φz / fied three regions:
|∇φ| − 1) the previous relation can be rewritten as
(a) water all those cells in the liquid phase for which none
∂ϕ of the six neighbouring is in the air phase;
+ uφ · ∇ϕ + b = 0 (12)
∂t (b) air cells in the air phase and none of the six neighbouring
is in the liquid phase;
which possesses a structure clearly similar to the Eq. (10) (c) interface those cells (either air or water) for which one
and can be solved by using the same scheme. of the six neighbouring is in the other phase.
Finally, for the velocity and pressure extrapolations out-
side the water region, considering the position (9), the For water cells, the residual on each control volume is
following equations hold computed as an interface flux balance; conservation laws are
applied to the (i jk)th control volume
∂u i
∇ϕ · ∇u i + =0 i = 1, 2, 3
∂z (13)  6   
∂p ∂
∇ϕ · ∇ p + = 0. Λ q dV + F cs (q) − F ds (q) dS = 0 (14)
∂z ∂t Vijk
s=1 Ss

Comput Mech

where Ss is the sth face of the finite volume Dijk , whose mea- the level set function being the distance from the interface
sure is Vijk . Defining
 the volume average of the unknowns (the subscripts A and W denote the volume in the air and
1 water phase, respectively).
q as q = q dV , the semi-discrete system of equa-
Vijk Vijk The tangential velocity is simply extrapolated along the
tions can be rewritten as normal to the free surface, as in the Eq. (8). The remaining
 dynamic boundary conditions for the tangential stresses in (3)
∂q  1 are explicitly enforced when computing the viscous fluxes at
Λ  + Rijk = 0 (15)
∂t ijk Vijk the sth interface.

being Rijk the flux balance on the current control volume Vijk . 3.2 Level set function and extrapolation
The surface integrals are evaluated by means of the (second
order) trapezoidal rule. In the viscous fluxes, the computation The evolution equation for the level set function and the
of the velocity gradients, required for the computation of the velocity/pressure extrapolation are discretized by using an
stress tensor at the cell interface, are computed by means of ENO technique similar to the one used for the bulk flow;
a standard second order centered finite volume approxima- the Eq. (10) is applied to both water and air cells, extrapola-
tion [32]. tion equations are considered only for the air cells, being the
For the inviscid part, the fluxes are computed as the solu- conservation variables in the control volumes of water phase
tion of a Riemann problem F cs = F c (q s ) = F c (q l , q r ); governed by the equations (15). Equation (10) is rewritten in
right and left states can be estimated by several schemes terms of curvilinear coordinates
implemented in the solver, ranging from the first order total
variation diminishing (TVD) scheme, the second order essen- ∂ϕ  l ∂ϕ
+ U + wijk = 0 (19)
tially non oscillatory (ENO) scheme [31], the third-order ∂t ijk ∂ξl ijk
upwind-based scheme [61] and the classical fourth-order
centered scheme (see [19]). Moreover, a second order accu-
being U l |ijk = (u i ∂i ξl )ijk the contra-variant components of
rate solution of the Riemann problem [16] is used in place of
the velocity vector at the (i jk)th cell center. The derivatives of
the exact one, which should be computed iteratively, given
the function ϕ(x, y, z, t) at cell center ∂ϕ/∂ξl |ijk is approx-
the non-linearity of the problem.
imated by an upwind based scheme. The Eikonal and the
Special attention must be paid for the water interface
extrapolation equations are discretized in a similar way; in
cells, for which, convective and viscous fluxes at the inter-
particular, the Eikonal equation reads
face between the control volumes in the water phase and in
the air phase (as for example for the interface s in Fig. 1),   
∂ϕ  l ∂ϕ
must be evaluated in accordance with the first equation of + Uφ + bijk = 0 (20)
∂t ijk ∂ξl ijk
the dynamic boundary conditions (3); the normal velocity at
the cell interface is computed by solving a Riemann problem  
with a constant pressure. First, the pressure at the free surface sign(φ) ∂φ ∂ξl
being Uφl |ijk = , whereas the extrapo-
is computes as |∇φ| ∂ xq ∂ xq ijk
lation equations become
κ z − z FS
pFS = τi j n i n j + + (16)    
We Fr 2 ∂ϕ ∂ξl ∂ u
U lext + =0
 ∂ξl ijk  ∂z ∂ξl ijk
l ∂ϕ ∂ξl ∂ p
with the derivatives on the right hand side computed by cen-
tered approximation. Then, the pressure at the sth interface Pext + =0
∂ξl ijk ∂z ∂ξl ijk
is extrapolated from the inner points (i.e. from values in the
water phase) as
ps = p W + ( pFS − pW ) (17) ∂ξl ∂ u ∂ξl
η U lext |ijk = ∇u|s = (22)
∂ xs ijk ∂ xs ∂ xs ijk
with pW the pressure in water interface cell and with η the
segment fraction (see Fig. 1) below the free surface and
|φW | ∂ξl ∂ p ∂ξl
Pext |ijk = ∇ p|s = . (23)
(18) ∂ xs ijk ∂ xs ∂ xs ijk
2 |φW − φA|

Comput Mech

3.3 Temporal integration step (m + 1) by equations (25). The correction of the level
set function is then evaluated by means of Eq. (26). The
Given the solution at time instants t n and t n−1 , the solution procedure is carried on in the pseudo time until either a fixed
at the new time step is computed by the time integration convergence or a maximum number of iterations are reached
of Eqs. (14) and (21) for the control volumes in water and (see the flowchart depicted in Fig. 2). Convergence toward
in air phase, respectively. The procedure is depicted with the steady state solution in the pseudo time is accelerated by
the diagram in Fig. 2. The time derivative in the system of a local time step Δτijk technique and a multi-grid method
equations (14) is approximated by a second order accurate [6,25].
three-points backward finite difference formula, yielding to After the solution at time step n + 1 has been found, a
further correction step for the level set function is performed
3q n+1 n n−1
ijk − 4q ijk + q ijk 1 in order to increase the robustness of the algorithm in par-
 + Rn+1 = 0 (24)
ticularly critical conditions (with the free surface evolving
2Δt Vijk ijk
The evolution equations (24) for the cells in the water
phase, together with the extrapolation equations (21) are m+1
ϕijk − ϕijk
m m+1
3ϕijk n + ϕ n−1
− 4ϕijk ijk
solved in a strongly coupled form through a pseudo-time +
approach; time advancement in the pseudo-time is performed Δτ 2Δt
with an Euler implicit integration scheme, i.e. n+1 ∂ϕ m+1
+ Ul + wijk =0 (27)
q m+1 3q m+1 n−1 ijk ∂ξl ijk
ijk − q ijk ijk − 4q ijk + q ijk
m n

In water Λ +Λ
Δτ 2Δt
1 Also for this equation, iterations in the pseudo time are pur-
+ Rm+1
ijk =0
Vijk sued until convergence to steady state or a maximum number
⎧ m+1 m ∂ϕ n+1  ∂ξ ∂ u m of iterations are reached.
⎪ uijk − um 

+ U lext   +
⎨ Δτ ijk ∂ξm ijk ∂z ∂ξl ijk As already discussed, the evolution equation (7) does not
In air m ∂ϕ     ensure that the level set function remains a distance function.

pijk − pijk
m n+1
∂ξl ∂ p m

⎩ l  
+ Pext  + =0
ijk ∂ξm ijk
Δτ ∂z ∂ξl ijk Hence, the correction step is followed by the re-initialization
procedure where, starting from the zero level of ϕ(x, y, z, t)
at time step t = t n+1 , the level set function is replaced by the
where the superscripts n and m denote the physical and signed distance function, which is computed by the integra-
dual time levels, τ being the pseudo time and Λ = tion in a fictitious time τ of the Eikonal equation (20) [52]
diag (1/β, 1, 1, 1), with β the pseudo-compressibility fac-
tor (see [14]). The previous system of equations is solved n+1
ϕijk − ϕijk
n n ∂ϕ n
  + bn = 0
+ Uφl  (28)
using a [4] scheme. Δτ ijk ∂ξl ijk
The solution of the flow field equations (25) requires the
knowledge of the free surface function (i.e. the knowledge where Δτ is estimated through the CFL condition. The
of ϕ n+1 ) and vice versa. A strongly coupled approach is iterations are pursued until either the steady state solution
used; the adopted algorithm is based on a predictor–corrector in fictitious time or a maximum number of iterations are
procedure, where the prediction of the level set function is reached.
obtained by the integration of Eq. (10). Similarly to the field
equations, the integration in the physical time is carried out
by a three-points backward finite difference approximation, 4 Hull characteristics, numerical set-up and test
whereas, an implicit Euler scheme is adopted for the integra- matrix
tion in the pseudo-time

m+1 m+1 n + ϕ n−1 Simulation results are provided for the planing hull in straight
ϕijk − ϕijk
m 3ϕijk − 4ϕijk ijk
+ ahead advancement. The vessel travels in otherwise calm
Δτ 2Δt water; the speed, after an initial transit state of acceleration
 ∂ϕ m+1 from the rest, is kept constant at the target velocity until a
+ Ul + wijk
=0 (26)
ijk ∂ξl ijk steady state condition is achieved. Several speeds are tested;
computations are performed in a frame of references translat-
The solution is iterated until either a fixed convergence in ing with the center of gravity (CG) of the model. Propellers
L 2 norm or a maximum number of iterations are reached. are not considered, however, a towing force reduced to the
Once a prediction of the level set function is obtained, it CG is taken into account to mimic the experimental setup
is exploited to compute field values at the next pseudo time where the model is towed aligned with the propeller axis.

Comput Mech

Fig. 2 Flowchart of the solution procedure used in χnavis. (Colour figure online)

The vessel considered in this paper is the Grande 95RPH i.e. at fixed vertical distance from the centre of gravity ( f
(Fig. 3), a recently designed luxury yacht of the Azimut– in Fig. 4, where the sketch of the towing set-up is reported).
Benetti group; hydrostatic data of the vessel are outlined in In the numerical simulations, the model is towed from the
Table 1. centre of gravity (CG); in order to mimic the experimental
The hull is considered as a rigid body with two degrees of set-up, an additional force (and moment) is applied at the
freedom; the model is left free to heave and pitch, whereas, CG. The vertical and horizontal components of the towing
it is kept fixed on surge, yaw, sway and roll motions. The force are − D tan(ε + τ ) and − D, respectively, where D is
numerical simulations resemble the classical resistance test the actual drag force and τ is the trim angle; the torque is
performed in the towing tank; the model is accelerated by − f D/ cos(ε + τ ).
a towing force up to the test speed, the unsteady numeri- Simulations are performed for speeds spanning from 18
cal simulation is carried out until the vessel reaches a stable to 34 knots; the corresponding Froude (Fr = U∞ / gL pp )
conditions in terms of pitch angle and sinkage. In the experi- and Reynolds (Re = U∞ L pp /ν) numbers vary in the range
ments [51], the model is towed along the propellers axis line, [0.621; 1.174] and [1.91 108 ; 3.60 108 ], respectively. It is

Comput Mech

Fig. 3 Azimut–Benetti Grande 95RPH

Table 1 Principal characteristic of the vessel Table 2 Numerical simulations: test matrix
U∞ (kn) U∞ (m/s) Fr Re We
Length between perpendiculars L pp 22.54 m
Displacement Δ 104t 18 9.260 0.621 1.91 × 108 2.737 × 106
Draft T 1.7 m 20 10.289 0.691 2.12 × 108 3.379 × 106
Longitudinal position of CG LCG 9.3 m 22 11.318 0.760 2.33 × 108 4.088 × 106
Vertical position of CG VCG 3.12 m 24 12.347 0.829 2.54 × 108 4.865 × 106
Propeller axis inclination ε 9.08◦ 26 13.375 0.898 2.75 × 108 5.710 × 106
Distance between CG and propeller axis f 1.634 m 28 14.404 0.967 2.97 × 108 6.622 × 106
30 15.433 1.036 3.18 × 108 7.602 × 106
worth to note that all the simulations were performed at 32 16.462 1.105 3.39 × 108 8.649 × 106
full scale without the use of any wall functions, making 34 17.491 1.174 3.60 × 108 9.765 × 106
the computations rather challenging in terms of grid reso-
lutions and, consequentially, in terms of computational cost
requirements, as well as numerical stability. The test matrix Table 3 Physical parameters (water characteristics are taken at
T = 15◦ C)
is reported in Table 2. The vessel is supposed to travel in
fresh water at 15 ◦ C; the physical parameters of the water are Parameter Symbol Value
reported in Table 3.
Density ρ 999.1026 kg/m 3
An overview of the computational mesh is shown in Fig. 5;
Kinematic viscosity ν 1.1386 × 10−6 m2 /s2
in the figure, the chimera cells have been hidden for the sake
Surface tension σ 73.50 × 10−3 N/m
of clearness. The vessel is symmetrical about its longitudinal
Acceleration of gravity g 9.8067 m/s2
vertical plane, therefore, only half of the domain has been
discretized; symmetry boundary conditions are applied on
the longitudinal symmetry plane. In order to obtain a high
and accurate refinement close to the hull and the skeg, the ble in Fig. 3, the hull is characterized by an hard chine along
domain has been discretized by 162 body-fitted patched and the vessel (useful for deflecting the sprays and providing an
overlapped blocks, for a total of about 18.5 M cells. increment of the lift force, as outlined in the Sect. 1) and an
Grid distribution is designed in such a way that the thick- aesthetic chine over it. An important effort has been spent for
ness of the first cell on the wall is always below 1 in terms of the discretization of the chines and of the skeg that looks very
wall units, and at least 30 cells are within the boundary layer thin and narrow. Moreover, the chines tends to join at the bow
thickness (y + = O(1), i.e. Δ/L pp = O(20/Re) with Δ the forming a geometric singularity which has been accurately
thickness of the near-wall cell). Size and position of the 162 modelled. In Fig. 5 a slice of the numerical domain highlight
structured blocks in the domain are listed in Table 4. As visi- the important stretch of the grid volumes near the walls and

Fig. 4 Hull longitudinal

profile. CG denotes the centre
of gravity

Comput Mech

Fig. 5 Overview of the computational mesh: on the top, mesh on the hull surface. On the bottom: particular of the stern region

Table 4 Mesh cells distribution 5 Results

Zone No. of cells
In the following, the numerical results are investigated.
Background 1,331,200 The estimation of total resistance, sinkage and trim will be
Hull 9,294,336 reported as first; comprehensive assessment of verification
Skeg 487,424 (i.e. the analysis of the convergence and estimation of the
Free surface 7,266,304 numerical uncertainty) and validation (i.e. the comparison
Total 18,379,264 against benchmark data including the estimation of the error)
will provide reliability of the computations. An analysis (for
one representative speed) of the flow field and wave pattern
(characterized by strong non linear effects, including break-
the overlapping of sets with different topologies: also in this
ing waves, plunging jets and ricochets) will close the section.
figure, the chimera cells have been hidden for the sake of
A multigrid technique has been exploited in order to 5.1 Resistance, sinkage and trim curves versus speed of
achieve a faster convergence to the solution in the pseudo- advancement
time; four levels of computational mesh are used. Every level
is obtained from the finer one, by removing every other point In Fig. 6 the total resistance RT (non-dimensionalized by the
along each spatial direction. The three finest grids (for which gravitational force mg, m being the displacement), and the
the refinement ratio is 2) have been used for Verification and total resistance coefficient (C T = RT /0.5ρU∞ L 2PP ) versus
Validation purposes. Computations are carried out in time Froude number are shown. Experimental results are reported
up to a clear convergence is attained; in particular, computa- with symbols. Resistance continuously increases with the
tions are stopped when the variations on the sinkage and on speed (or, similarly, the resistance coefficient decreases with
the trim are at most on the fourth significant digit. Fr). Values for the whole speed range are in agreement with

Comput Mech

Fig. 6 Resistance and resistance coefficient versus Froude number. Symbols: experimental data; solid line: numerical estimation

Fig. 7 Sinkage and trim versus Froude number. Symbols: experimental data; solid line: numerical estimation

experimental data (grid convergence and a comprehensive 5.1.1 Verification and validation
verification and validation will be presented in the follow-
ing). The reliability of the computed results is assessed by a
Figure 7 displays sinkage (i.e. the distance of the CG rigorous estimation of the convergence properties and the
from the undisturbed free surface level) and trim estimations, accuracy of the results (i.e. numerical uncertainty estima-
along with the measurements: sinkage is positive when the tion and comparisons with respect to experimental data).
CG moves upward, whereas trim is positive when the ship Verification (the evaluation of the order of convergence and
rotates bow up. As the speed increases, the lift provided by the assessment of numerical uncertainty) and validation (the
the main hull surface increases and the hull rises (i.e. sinkage estimation of the modelling error by comparison with exper-
increases); it is worth to note that, an additional contribution imental data) have been carried out for the total resistance
to the vertical force is provided by the hard chines which coefficient, the trim and the sinkage. Experiments are made
interact with the free surface. The change of the distribution available by [51]. In order to provide a complete analysis,
of the vertical force is such that to lead to a decrease of the verification and validation have been assessed for the whole
trim angle with the speed (as shown in Fig. 7. The vessel speed range.
at higher velocities acquires a more pronounced neutral atti- Results are summarized in Table 5. In the table, S3 , S2 and
tude, i.e. low trim angles. The decreasing of the trim angle, S1 refer to the values computed on the coarse, the medium
clearly indicates a larger increase of the pressure on the stern and the fine grid, respectively; pRE is the measured order of
region with respect to the fore part of the ship; as a conse- accuracy
quence, increasing the speed, the stern rises more then the
bow, finally leading to a low trim angle attitude. The com- ln (ε32 /ε21 )
parison with experiments are rather satisfactory for the sink pRE = (29)
ln (r )
at all Froude numbers, as well as for the trim at the lower-
medium speed. At higher speeds, computational estimations
where ε21 = S2 − S1 , ε32 = S3 − S2 and r is the grid refine-
provide a continuously decreasing of the trim, whereas EFD
ment ratio (r = 2 has been used in the current analysis). From
measurements show a kind of plateau. However, the differ-
the solution on the medium and the finest grid,the Richard-
ence, as it will be deeper analysed in the following section,
son’s Extrapolated error δRE can be computed
is rather low, being less then half of a degree.
δRE = (30)
r RE − 1

Comput Mech

Table 5 Verification and validation

Fr S3 S2 S1 ε21 /ε32 pRE δRE SRE D USN (%SRE ) E(%D)

(a) Sinkage σ (values are expressed in meter [m])

0.623 1.352 1.375 1.413 1.73 Div. – – 1.352 – 4.50
0.692 1.391 1.474 1.483 0.11 3.21 − 1.09 × 10−3 1.484 1.431 0.85% 3.70
0.761 1.447 1.539 1.567 0.30 1.74 − 1.17 × 10−2 1.578 1.497 1.27% 5.45
0.830 1.470 1.581 1.600 0.17 2.56 − 3.86 × 10−3 1.604 1.546 1.49% 3.75
0.900 1.479 1.599 1.627 0.23 2.10 − 8.59 × 10−3 1.636 1.587 1.26% 3.10
0.969 1.515 1.618 1.663 0.44 1.19 − 3.53 × 10−2 1.699 1.623 4.04% 4.67
1.038 1.539 1.640 1.668 0.27 1.88 − 1.03 × 10−2 1.678 1.627 1.01% 3.13
1.107 1.605 1.671 1.700 0.43 1.21 − 2.19 × 10−2 1.721 1.668 2.46% 3.20
1.176 1.632 1.716 1.781 0.76 0.40 − 2.03 × 10−1 1.984 1.722 23.39% 15.22
Overall 4.47% 5.19%
(b) Trim τ (values are expressed in degrees)
0.623 3.295 3.204 3.079 1.36 Div. – – 2.953 – 4.28
0.692 3.451 3.317 3.054 1.96 Div. – – 2.964 – 3.05
0.761 3.409 3.204 2.951 1.24 Div. – – 2.865 – 3.02
0.830 3.332 3.054 2.751 1.08 Div. – – 2.744 – 0.27
0.900 3.324 2.869 2.630 0.53 0.93 2.65 × 10−1 2.365 2.668 23.01% 11.34
0.969 3.279 2.767 2.601 0.32 1.62 7.99 × 10−2 2.521 2.671 5.58% 5.63
1.038 3.212 2.622 2.446 0.30 1.75 7.48 × 10−2 2.371 2.684 5.39% 11.65
1.107 3.102 2.518 2.416 0.18 2.50 2.20 × 10−2 2.394 2.705 5.27% 11.51
1.176 2.946 2.348 2.345 0.01 7.78 1.24 × 10−5 2.345 2.719 0.03% 13.74
Overall 7.85% 7.16%

Fr S3 (×105 ) S2 (×105 ) S1 (×105 ) ε21 /ε32 pRE δRE (×105 ) SRE (×105 ) D (×105 ) USN (%SRE ) E(%D)

(c) Total resistance RT (values are expressed in Newton [N])

0.623 1.36 1.24 1.09 1.17 Div. – – 1.05 – 4.02
0.692 1.37 1.22 1.14 0.54 0.90 9.05 1.05 1.11 17.78% 5.36
0.761 1.44 1.27 1.19 0.44 1.17 6.25 1.13 1.17 10.81% 3.14
0.830 1.50 1.32 1.23 0.46 1.13 7.21 1.16 1.23 12.26% 5.67
0.900 1.60 1.36 1.29 0.31 1.70 3.23 1.26 1.30 4.44% 3.50
0.969 1.68 1.42 1.38 0.18 2.44 1.07 1.36 1.38 4.10% 1.16
1.038 1.77 1.51 1.42 0.38 1.39 5.96 1.36 1.45 8.16% 6.72
1.107 1.84 1.58 1.50 0.28 1.83 2.97 1.47 1.53 3.39% 3.54
1.176 1.96 1.78 1.55 1.28 Div. – – 1.59 – 2.67
Overall 8.71% 3.98%

The generalized Richardson extrapolated solution SRE can asymptotic range is given by the ratio between the computed
be then estimated as order of accuracy and the theoretical one, i.e. P = pRE / pth ;
when the solutions are in the asymptotic range P  1, the
SRE = S1 − δRE (31) actual order of convergence is equal to the theoretical value.
This method has been proved to be superior to the correction
factor method [57,65] since it overcomes the unreasonable
When monotonic convergence is attained (i.e. when 0 <
small uncertainty estimation when the measured order of
ε21 /ε32 < 1) both the order of convergence and the error can
convergence is less than the theoretical value; moreover, it
be computed. Several methods [24,47,57,65] can be used for
provides an overall confidence level of 95% for the estimated
the estimation of the numerical uncertainty; in the present
uncertainty to bound the true error. According to the factor
work the factor of safety method proposed by [66] has been
of safety method, the grid uncertainty can be computed as
adopted. In this method, a measure for the distance from the

Comput Mech

(2.45 − 0.85P) |δRE | if 0 < P ≤ 1
UG = (32)
(16.4P − 14.8) |δRE | if P > 1

In Table 5, the estimated numerical uncertainties for the

computation of the sinkage (σ ), trim (τ ) and the total resis-
tance (RT ) are reported; since negligible iterative uncertainty
has been observed (well below 1%), grid uncertainty can be
considered as the only contribution to the numerical uncer-
tainty (i.e. USN ≈ UG ). The absolute error E with respect
to the experiment data D (for all the computed quantities) is
reported in the last column of the Table 5
 SRE − D 

E = % (33)

When the extrapolated values cannot be estimated (i.e.

when monotonic convergence is not attained), the value com-
puted on the finest grid (S1 ) has been used for the error
Fig. 8 Error respect to experimental data for the total resistance, the
estimation. sinkage and the trim against Froude number
The measured convergence properties are rather good for
the whole speed range; an order of convergence very close
to the theoretical value (2) for the computation of drag, trim respond to a difference between the computed value and the
and sinkage is observed for the medium speeds. Good con- benchmark reference less than 0.5 ◦ . Moreover, it is worth
vergence properties are also observed at lower and higher to note that, the difficulties of the numerical simulations in
speeds for the estimation of the resistance and the sinkage, the correct prevision of the trim angle is a well-known topic
whereas poor convergence (or divergence) is observed for and for fast ships may be tricky to overcome (see also [2])
the computation of the trim at medium and lower speeds. In The quality of the computed results can be summarized with
general, the worst convergence properties are seen at highest the average errors (within the speed range); values of about
and lowest speeds for the computations of all the quantities 5%D, 7%D and 4%D are reported for the sinkage, trim and
of interests. the total resistance, respectively. Clearly, the agreement can
In general, low uncertainties are estimated for all mono- be considered largely satisfactory. Finally, by the compari-
tonic convergent cases; as shown in Table 5, higher uncer- son of the relative errors and the numerical uncertainty, even
tainty level is associated to the estimation of the total without the inclusion of the experimental uncertainty, the
resistance, whereas lower uncertainty values are observed numerical results can be considered, in general, validated
for the computation of the sinkage. Average values of the when a monotonic convergence is achieved. To summarize,
numerical uncertainties are bounded 5%SRE for the sinkage on the average, the total resistance is validate at the level of
and around 8%SRE for both the trim and the total resistance. 9%D, the trim at the level of 8%D and the sinkage around
Larger numerical uncertainty levels (up to 20%SRE ) are seen 5%D.
for the total resistance at lower speeds, for the computation
of the sinkage at highest speed and for the trim estimation at 5.2 Flow field
the medium speed, i.e. in general for those cases for which
the worst convergence properties (order of convergence less An overview of the flow field is reported here; the analysis
than 1) are observed. is carried out for the simulation at the advancement speed of
By comparison with the experimental data, a very good 28 Kn i.e. at Fr = 0.967. Results at the other speeds present
agreement between numerical results and experimental mea- the similar characteristics to those highlighted in the follow-
surements can be inferred. The comparison is summarized in ing. A three dimensional view of the free surface shapes
the last column of Table 5, as well as graphically in Fig. 8: around the hull advancing at the selected speed is displayed
the error is rarely higher than 10%D. An encouraging consis- in Fig. 9.
tency with experiments is found for the sinkage and the total For the sake of clearness, the solution is mirrored around
resistance, without any clear dependence on the advancement the longitudinal plane of symmetry. Differently from what
speed. Higher discrepancies are observed for the estimation happens for displacement vessels, where a Kelvin wave sys-
of the trim, with larger values at higher speeds; however, it tem is generated, for high speed planing crafts the fluid is
has to be highlighted that errors of the order of 10%D, cor- pushed aside by the hull in the form of thin sprays. A clear

Comput Mech

interacting with each other and forming a very complex free-

surface pattern; it is to note that the wave pattern is well
captured by the present numerical solver, regardless its topo-
logical complexities.
More in details, in Fig. 10 a purely qualitative compari-
son between the simulated surface flow pattern and a typical
sailing boat one is shown. Obviously the problems are very
different in terms of geometries and the Froude numbers.
However some typical features are clearly visible: the plunge
point of the jet emerging from the bow splash and starting of
the first breaking cycle indicated with A; the splash-up of this
jet B; and the plateau between the breaking front and the hull
denoted with C. Finally a similar secondary vortical structure
shed by the breaking bow wave can be seen in the two wave
patterns (marked with D). Differently from the sailing boat,
for the planing hull the presence of an hard chines anticipates
the bow jet plunging, transferring an higher amount of energy
near the jet root. This interaction moves the plunging point
farther away and makes the jet very thin and narrow. More-
over, in the planing hull case, the jet impinges with higher
intensity causing the formation of a ricochet downstream,
which is totally absent in the sailing boat case. This rico-
chet produces a second plunging with an intense production
of surface vorticity, responsible of important air entrapment
and foam generation (not modelled in the present approach).
It is of interest, in terms of surface vorticity, the presence of a
laminar plateau close to the vessel and bounded by two highly
vortical areas; here a strong vortex stretching occurs, which
causes the stretch of the vortical structures along the ship
axis, in a way that the flow resembles a laminar behaviour.
Fig. 9 Top and perspective views of the wave pattern around the vessel, Similar investigation has been performed in [41].
advancement speed U∞ = 28 Kn At the transom, which is completely dry, a deep wave
trough is seen. As expected, this is followed by a rather high
wave crest, forming the well-know roaster tail, which results
rise up of the water in the fore part of the hull is shown; due elongated far away the transom as typical for high-speed ves-
to the high pressure which occurs at the bow of the ship, sels. It is interesting to note the formation of wave systems
the water rises along the surface hull in form of a thin water from the edges of the stern (one on each side) and one in the
sheet. This water sheet would rise up to a level at which its centre-line of the ship. This wave systems are caused by the
kinetic energy equals the potential one, where it would detach vortical structures generated at the sharp edges (transom/side
from the hull surface, forming a jet. In the present case, the hull and the skeg) of the ship. It is worth to note that all these
jet is formed when the thin sheet impacts the hard chine of characteristics of the wave pattern, regardless their complex
the ship, which forces the water sheet to leave the surface shapes and dynamics, are accurately captured by the numer-
hull in the form of a breaking wave. The energy of the water ical solution.
sheet allows the jet to spread far away from the hull, where it In Fig. 11, the non dimensional pressure on the hull sur-
impinges the surface originating several ricochets. This jet is face is shown. Again, for the sake of clearness, the solution is
related to an important property of the hard chine: the intense mirrored about the longitudinal vertical plane of symmetry;
jet root provides an added lift to the ship, rising the sinkage. both bottom and lateral views are presented. In the pictures,
Downstream, along the stern part of the side of the hull, a the wave profile is reported with a black thick line. The pres-
water sheet attached to the hull is seen; this water sheet leave sure field on the fore part of the hull is similar to the one
the surface at the transom, which, being completely dry, gives that can be observed in case of simply planned wedge shape
rise to an hydraulic jump [41]. at high Froude numbers [29,33] and for similar hull shape
The complex wave system on the side of the hull, is char- [30,35,42]; the pressure peak at the stagnation point causes a
acterized by several water jets, breaking waves and ricochets, rise up of the water, the height of the rise depends on the flare

Comput Mech

Fig. 10 Comparisons of free surface structures between present test (U∞ = 28 Kn) and a sailing boat

Fig. 11 Pressure distribution on the hull surface, advancement speed U∞ = 28 Kn

angle of the ship. A quasi V-shaped jet root is clearly visible, suddenly detaches with a less intense jet (as it can be also
with a decreasing pressure peak moving downstream. Dif- seen from the figures of the free surface).
ferently from the flat wedge [33], where the jet root is rather
narrow, the non-flat shape of the vessel causes a spread of
the pressure peak, i.e. the region of the hull interested to high
pressure values is wider. This can be of some importance for 6 Conclusions
loads and structural related problems, as well as, to slamming
occurrence once the ship is travelling in waves. When the jet In the present paper an unsteady single phase level-set tech-
root arrives at the hard chine, a step pressure rise occurs. The nique has been adopted (within a general purpose uRaNS
rise up of the water grows, with high pressure level extending solver) for the study of the free surface wave pattern gen-
well upstream. erated by a high speed vessel. The single phase level set
Moreover, from this pictures, it can be also clearly seen approach allows to track the dynamics of the air/water inter-
how the free surface detaches all along the chine, up to face maintaining its thickness within one cell, without any
the region where the lower chine is large enough. When need to smooth out the variation of the fluid variables across
it becomes too tiny, the water is able to overcame the first the two phases. In this regard this methodology is able to
geometrical edge, rising up to the aesthetic chine, where it represent the interface with an accuracy comparable to clas-
sical Lagrangian methodologies. The governing equations

Comput Mech

are integrated in the water phase only, whereas in the air 7. Brizzolara S, Serra F (2007) Accuracy of CFD codes in the pre-
phase the field variables are extrapolated. The Eikonal equa- diction of planing surfaces hydrodynamic characteristics. In: 2nd
international conference on marine research and transportation,
tion allows to re-compute the level set function as a distance Ischia, Naples
function; its integration, starting from the zero level of the 8. Broglia R, Iafrati A (2010) Hydrodynamics of planing hulls in
level set function, is integrated in time for few steps, exploit- asymmetric conditions. In: Proceedings of 28th symposium on
ing the transport equation of the distance function only in a naval hydrodynamics, Pasadena, CA
9. Broglia R, Zaghi S, Di Mascio A (2011) Numerical simulation of
narrow band across the interface. interference effects for a high-speed catamaran. J Mar Sci Technol
The common features of a typical wave pattern are com- 16(3):254–269
pletely recovered: the bow plunging splash, the laminar 10. Broglia R, Dubbioso G, Durante D, Mascio AD (2013) Simulation
plateau, as well as the secondary vortical structures and the of turning circle by CFD: analysis of different propeller models and
their effect on manoeuvring prediction. Appl Ocean Res 39:1–10
roster tail. It has to highlighted that, regardless the topo-
11. Broglia R, Zaghi S, Muscari R, Salvadore F (2014) Enabling hydro-
logical complexities, the adopted methodology revealed an dynamics solver for efficient parallel simulations. In: Proceedings
high reliability in the accurate prediction of the wave pattern international conference on high performance computing and sim-
generated by an high speed vessel. Velocity and pressure ulation (HPCS). Italy, Bologna, pp 803–810
12. Broglia R, Dubbioso G, Durante D, Di Mascio A (2015) Turning
fields are also accurately estimated; surface pressure cor- ability analysis of a fully appended twin screw vessel by cfd. Part
rectly shown sharp high pressure regions corresponding to i: single rudder configuration. Ocean Eng 105:275–286
the jet roots. The reliability of the filed variables has been 13. Chang KC, Hsieh WD, Chen CS (1995) A modified low-Reynolds-
proven by a complete convergence study and a comparison number turbulence model applicable to recirculating flow in pipe
expansion. J Fluid Eng 117:417–423
with experimental data in terms of both total resistance and 14. Chorin A (1967) A numerical method for solving incompressible
attitudes for a wide speed range. Satisfactory results have viscous flow problems. J Comput Phys 2:12–26
been obtained for both convergence properties and numer- 15. Colagrossi A, Marrone S, Bouscasse B, Broglia R (2015) Numer-
ical uncertainty. To summarize, errors with respect to EFD ical simulations of the flow past surface-piercing objects. Int J
Offshore Polar Eng 25(1):13–18
data less than 5% have been generally observed, with higher 16. Di Mascio A, Broglia R, Favini B (2001) A second order Godunov-
comparison errors at higher speeds and for the estimation type scheme for naval hydrodynamics. Kluwer Academic/Plenum
of the trim angle. Anyhow, it has to be observed that, for Publishers, Dordrecht, pp 253–261
the latter, the error is, at most, half of degree, which can be 17. Di Mascio A, Muscari R, Broglia R (2006) An overlapping grids
approach for moving bodies problems. In: Proceedings of 16th
considered within the numerical/experimental uncertainty. international offshore and polar engineering conference, San Fran-
Future activities will concern fluid structure interactions cisco, CA, USA
at various speeds, as well as static and dynamic roll stability 18. Di Mascio A, Broglia R, Muscari R (2007) On the application of the
for high speed planing hull. one-phase level set method for naval hydrodynamic flows. Comput
Fluids 36(5):868–886
19. Di Mascio A, Broglia R, Roberto M (2007) Prediction of Hydro-
Acknowledgements The research has been partially supported by the
dynamic coefficients of ship hulls by Godunov-type methods. In:
Project RESMARE “Ricerca E Servizi per il MARE” funded by the
MARINE 2007, Barcelona, Spain
Lazio Region and conducted in collaboration with Azimut-Benetti
20. Di Mascio A, Broglia R, Muscari R (2008) Numerical simulations
which are acknowledged for providing vessel geometry and experi-
of viscous flow around a naval combatant in regular head waves.
mental results.
In: Proceedings 6th Osaka colloquium on seakeping and stability
of ships, Osaka, Japan
21. Di Mascio A, Broglia R, Muscari R (2009) Prediction of hydro-
dynamic coefficients of ship hulls by high-order Godunov-type
methods. J Mar Sci Technol 14:19–29
References 22. Dubbioso G, Durante D, Mascio AD, Broglia R (2016) Turning
ability analysis of a fully appended twin screw vessel by CFD. Part
1. Adalsteinsson D, Sethian JA (1999) The fast construction of exten- ii: single vs. twin rudder configuration. Ocean Eng 117:259–271
sion velocities in level set methods. J Comput Phys 148:2–22 23. Dubbioso G, Broglia R, Zaghi S (2017) CFD analysis of turning
2. Akkerman I, Dunaway J, Kvandal J, Spinks J, Bazilevs Y (2012) abilities of a submarine model. Ocean Eng 129:459–479
Toward free-surface modeling of planing vessels: simulation of the 24. Eça L, Hoekstra M (2007) Discretization uncertainty estimation
Fridsma hull using ALE-VMS. Comput Mech 50:719–727 based on a least squares version of the grid convergence index. In:
3. Baiges J, Codina R, Pont A, Castillo E (2017) An adaptive fixed- 2nd workshop on CFD uncertainty analysis, Lisbon, Portugal
mesh ALE method for free surface flows. Comput Methods Appl 25. Favini B, Broglia R, Di Mascio A (1996) Multi-grid acceleration of
Mech Eng 313:159–188 second order ENO schemes from low subsonic to high supersonic
4. Beam RM, Warming RF (1978) An implicit factored scheme for flows. Int J Numer Methods Fluids 23:589–606
the compressible Navier–Stokes equations. AIAA J 16:393–402 26. Fontaine E, Cointe R (1997) A slender body approach to nonlinear
5. Begovic E, Bertorello C, Pennino S (2014) Experimental seakeep- bow waves. Technical report, Philosophical Transactions of Royal
ing assessment of a warped planing hull model series. Ocean Eng Society, London
83:1–15 27. Fridsma G (1969) A systematic study of the rough-water perfor-
6. Brandt A (1984) Multi-grid techniques: 1984 guide with applica- mance of planing boats. Technical report, DTIC Document
tion to fluid dynamics. The Weizmann Institute of Science, Rehovot 28. Fu T, Ratcliffe T, O’Shea T, Brucker K, Graham R, Wyatt D (2010)
(Israel) A comparison of experimental measurements and computational

Comput Mech

predictions of a deep-V planing hull. In: Proceedings of 28th ONR 49. Savender B (1997) Planing hull hydrodynamics. Ph.D. thesis,
symposium on naval hydrodynamics, Pasadena, CA, USA Department NAME, University of Michigan, Ann Arbor (MI)
29. Fu T, O’Shea T, Judge C, Dommermuth D, Brucker K, Wyatt D 50. Savitsky D (1964) Hydrodynamic design of planing hulls. Mar
(2012) A detailed assessment of numerical flow analysis (NFA) to Technol 1:71–95
predict the hydrodynamics of a deep-V planing hull. In Proceed- 51. Semijalac G, Nikolić M (2012) Resistance test results. Ship Model:
ings of the 29th symposium on naval hydrodynamics, Gothenburg, M-1357 MY Azimut 90F. Technical report, Brodarski Institut
Sweden Zagreb
30. Fu T, Brucker K, Mousaviraad S, Ikeda C, Lee E, O’Shea T, Wang 52. Sethian JA (1999) Level set methods and fast marching methods:
Z, Stern F, Judge C (2014) An assessment of computational fluid evolving interfaces in computational geometry, fluid mechanics,
dynamics predictions of the hydrodynamics of high-speed planing computer vision, and materials science, vol 3. Cambridge Univer-
craft in calm water and waves. In: Proceedings of the 30th sympo- sity Press, Cambridge
sium on naval hydrodynamics, Hobart, Tasmania 53. Sethian JA (2001) Evolution, implementation, and application of
31. Harten A, Engquist B, Osher S, Chakravarthy SR (1987) Uniformly level set and fast marching methods for advancing fronts. J Comput
high order accurate essentially non-oscillatory schemes. J Comput Phys 169:503–555
Phys 71:231–303 54. Sethian JA, Smereka P (2003) Level set methods for fluid interfaces.
32. Hirsch C (2007) Numerical computation of internal and exter- Annu Rev Fluid Mech 35:341–372
nal flows: the fundamentals of computational fluid dynamics. 55. Simonsen C, Quadvlieg F, Stern F (2014) In: Proceedings of
Butterworth-Heinemann, London SIMMAN 2014: workshop on verification and validation of ship
33. Iafrati A, Broglia R (2008) Hydrodynamics of planing hulls: a manoeuvering simulation methods, Copenhagen, Denmark
comparison between RANS and 2D + t potential flow models. In: 56. Spalart PR, Allmaras SR (1994) A one-equation turbulence model
Proceedings of 27th symposium on naval hydrodynamics, Seoul, for aerodynamic flows. La Recherche Aérosp 1:5–21
Korea 57. Stern F, Wilson RV, Coleman HW, Paterson E (2001) Comprehen-
34. Iafrati A, Di Mascio A, Campana EF (2001) A level set technique sive approach to verification and validation of CFD simulations—
applied to unsteady free surface flows. Int J Numer Methods Fluids part 1: methodology and procedures. J Fluid Eng 123(4):793–802
35(3):281–297 58. Stern F, Agdrup K, Kim S, Cura-Hochbaum A, Rhee K, Quadvlieg
35. Judge C, Ikeda C (2014) An experimental study of planing hull F, Perdon P, Hino T, Broglia R, Gorski J (2011) Experience from
wave slam events. In: Proceedings of the 30th symposium on naval simman 2008—the first workshop on verification and validation of
hydrodynamics, Hobart, Tasmania ship maneuvering simulation methods. J Ship Res 55(2):135–147
36. Lai C, Troesch A (1995) Modeling issues related to the hydrody- 59. Sussman M, Smekerda P, Osher SJ (1994) A level set approach for
namics of three-dimensional steady planing. J Ship Res 39:1–24 computing solutions to incompressible two-phase flow. J Comput
37. Lai C, Troesch A (1996) A vortex lattice method for high speed Phys 114:146–159
planing. Int J Numer Methods Fluids 22:495–513 60. Tulin M (1956) The theory of slender surfaces planing at high
38. Landrini M, Colagrossi A, Greco M, Tulin M (2012) The fluid speed. Schiffstechnik 4:125
mechanics of splashing bow waves on ships: a hybrid BEM-SPH 61. Van Leer B (1979) Towards the ultimate conservative difference
analysis. Ocean Eng 53:111–127 scheme. V. A second-order sequel to Godunov’s method. J Comput
39. Larsson L, Stern F, Visonneau M (2013) Numerical ship hydrody- Phys 32:101–136
namics: an assessment of the Gothenburg 2010 workshop. Springer, 62. Vogt M, Larsson L, (1999) Level set methods for predicting vis-
Berlin cous free surface flows. In: Proceedings of the 7th international
40. Larsson L, Stern F, Visonneau M, Hirata N, Hino T, Kim J (2015) conference on numerical ship hydrodynamics, Nantes, France, pp
Proceedings of: TOKYO 2015 a workshop on CFD in ship hydro- 2.1.4–19
dynamics. Tokyo, Japan 63. Volpi S, Sadat-Hosseini H, Kim DH, Diez M, Stern F, Thodal
41. Marrone S, Colagrossi A, Antuono M, Lugni C, Tulin M (2011) A RS, Grenestedt JL (2015) Validation high-fidelity CFD/FE FSI for
2D + t SPH model to study the breaking wave pattern generated full-scale high speed planing hull with composite bottom panels
by fast ships. J Fluids Struct 27(8):1199–1215 slamming. In: Proceedings of international conference on coupled
42. Mousaviraad S, Zhaoyuan Wang Z, Stern F (2015) Urans studies problems in science and engineering, San Servolo Island, Venice,
of hydrodynamic performance and slamming loads on high-speed Italy
planing hulls in calm water and waves for deep and shallow con- 64. Volpi S, Diez M, Sadat-Hosseini H, Kim DH, Stern F, Thodal
ditions. Appl Ocean Res 51:222–240 RS, Grenestedt JL (2016) Full-scale fluid–structure interaction
43. Muscari R, Broglia R, Di Mascio A (2007) Numerical simulation simulation and experimental validation of high-speed planing-hull
of the flow around an array of free-surface piercing cylinders in slamming with composite panels. In: Proceedings of the 31st sym-
waves. Ship Technol Res 54:43–52 posium on naval hydrodynamics, Monterey, CA, USA
44. Osher S, Sethian JA (1988) Fronts propagating with curvature- 65. Wilson RV, Stern F, Coleman HW, Paterson E (2001) Comprehen-
dependant speed: algorithms based on Hamilton–Jacobi formula- sive approach to verification and validation of CFD simulations—
tions. J Comput Phys 79:12–40 part 2: application for RANS simulation of a cargo/container ship.
45. Qin Z, Delaney K, Riaz A, Balaras E (2015) Topology preserving J Fluid Eng 123(4):803–810
advection of implicit interfaces on cartesian grids. J Comput Phys 66. Xing T, Stern F (2010) Factors of safety for Richardson extrapola-
290:219–238 tion. J Fluids Eng 132(6):061403-1–061403-13
46. Queutey P, Visonneau M (2007) An interface capturing method 67. Zaghi S, Broglia R, Mascio AD (2011) Analysis of the interference
for free-surface hydrodynamic flows. Comput Fluids 36(9):1481– effects for high-speed catamarans by model tests and numerical
1510 simulations. Ocean Eng 38(17–18):2110–2122
47. Roache PJ (1997) Quantification of uncertainty in computational 68. Zhao R, Faltinsen O, Haslum H (1997) A simplified nonlinear anal-
fluid dynamics. Ann Rev Fluid Mech 29(1):123–160 ysis of a high speed planing craft in calm water. In: Proceedings of
48. Salvatore F, Broglia R, Muscari R (2015) In: Proceedings of VI FAST97 conference, Sydney
international conference on computational methods in marine engi-
neering, Rome, Italy