Sie sind auf Seite 1von 12

Proceedings of ASME 2019 38th International Conference on Ocean, Offshore and Arctic Engineering

OMAE2019
June 9-14, 2019, Glasgow, Scotland

OMAE2019-96099

CFD ANALYSIS OF A CAPTIVE BULLET ENTRY IN CALM WATER WITH AND


WITHOUT TURBULENCE

René Bettencourt Rauffus António Maximiano


Instituto Superior Técnico, Universidade de Lisboa WavEC-Offshore Renewables
(IST-UL) R. Dom Jerónimo Osório, 11, 1400-119 Lisboa
Av. Rovisco Pais 1, 1049-001 Lisboa Portugal
Portugal Email:antonio.maximiano@wavec.org
Email: rene.rauffus@tecnico.ulisboa.pt

Luı́s Eça Guilherme Vaz


IST-UL Maritime Research Institute Netherlands
Mechanical Engineering Department (MARIN)
Av. Rovisco Pais 1, 1049-001 Lisboa P.O. Box 28, 6700 AA Wageningen
Portugal The Netherlands
Email: luis.eca@ist.utl.pt Email: g.vaz@marin.nl

ABSTRACT INTRODUCTION
Simulations are carried out for a simplified lifeboat drop test Over the last decade, an industry wide effort to improve the
case, which consists of a captive axisymmetric generic lifeboat safety of the lifeboats culminated in two norms: the less stricter
shape (bullet), that penetrates the water surface at a constant ve- “SA Code” [2] and the more restrictive “DNV-OS-E406”( [3]),
locity and angle of attack. The quantities of interest are the body with the latter applicable to all lifeboats on the Norwegian conti-
fixed longitudinal force FX , vertical force FZ , and pitch moment nental shelf as of 2015. Currently, the DNV norm allows for CFD
MYY .This case was previously used in a verification and valida- (Computational Fluid Dynamics) to aid in the design process of
tion exercise [1]. Here, a step forward in complexity is taken, free-fall lifeboats, as a complement to model testing. Unlike the
as the previous numerical model is now supplemented with the structural analysis, which has an appendix dedicated to providing
eddy-viscosity based turbulence model k − ω SST. Both ap- recommendations for FEM calculations, CFD recommendations
proaches are then used to simulate two different cases: Case 1 are more limited. In paragraph 4.5.3.16 the following recom-
with minimal wake effects; and Case 3 with flow separation and mendations are provided: ”The CFD method used should allow
significant wake. The results are compared with the experimen- modelling of turbulence, two-phase flow with compressible air
tal data. The numerical uncertainty is estimated for both models. and the break-up of water particles. When applying CFD, con-
It is seen that for Case 1 the difference between both models is vergence tests shall be carried out to ensure that the fluid cells
mostly within the comparison uncertainty, except for the longi- and the temporal discretization are adequately representative of
tudinal force FX , where the turbulent flow predicts a larger force, the physical spatial and temporal scales. The computational do-
improving the comparison with the experiments. The loads pre- main shall be large enough to avoid reflections from boundaries
dicted with turbulent flow stayed mostly within 6 % of the lami- of the domain. Numerical results based on CFD should be val-
nar flow. For Case 3 small differences between both models are idated against benchmark model test results.”. As examples of
found during/after the wake collapse stage. However, this differ- suitable models, the ”Volume-of-Fluid” (VoF) and smoothed par-
ence is often within the comparison uncertainty. A reasonable ticle hydrodynamics (SPH) are mentioned. [3]
agreement is found with the experimental data, except for FZ af- The volume of fluid (VoF) approach, suitable for the
ter the bow wake collapse. The turbulent flow improves slightly widespread finite volume methods, has been widely used for this
on the laminar approach regarding the agreement with the exper- application [4,5,6], in which both air and water are modelled as a
iments, however it can be argued if this difference justifies the single continuum fluid with physical properties varying linearly
increased computational cost of the turbulence model. between air and water. When using the VoF model, attention

1 Copyright
c 2019 by ASME
must be paid to ensure a sharp transition from air to water. Ide- TABLE 1: Main particulars.
ally such transition would occur within two or three cells.
The DNV norm requires some consideration regarding tur- Quantity Abbreviation Value Unit
bulence modelling. It suggests using two equation models for Scale λ 15.77 -
free-fall lifeboats design. In this application, it is not uncommon Length L 857 mm
to find literature where it is assumed that the turbulence effects Breadth B 214 mm
are negligible. In [7], [8], [9] and [10] a two equation model with
wall functions was employed using the STAR-CCM+ code and
in [11] a minimum-dissipation turbulence LES model was used
in the ComFLOW code. In [7], for a 2D and 3D wedge drop
case, the influence of adding a turbulence model was under 3 %
in the computational results.
Regarding the validation of CFD results with experimen-
tal data, several authors have shown reasonable agreement with
model scale experiments [5, 12, 13, 4], and even with full-scale
test results [9, 4, 14]. A formal verification and validation exer-
cise (V&V), using the procedure proposed in [15], was carried
out in [1] for a simplified lifeboat drop case, where the exper-
imental, numerical and validation uncertainties were quantified
and discussed. It should be noted that only Berchiche et al. [4]
had experimental data of model drops into waves. The remaining
experiments were drops into calm water.
The objective of this work is to better understand the influ-
FIGURE 1: Definitions and sign convention. The dashed line is
ence of turbulence modelling on the global forces acting on the
the symmetry axis, θ the pitch angle, γ the angle of attack, and β
lifeboat, and identify the main challenges associated with turbu-
the velocity angle.
lence modeling in this type of flows. This work builds on the
work reported in [1], supplementing the numerical model de-
scribed therein, with the commonly used k − ω SST turbulence TABLE 2: Drop cases simulated with CFD.
model. Two different drop cases with different angles of attack
are then simulated with and without the turbulence model, and Case Velocity Pitch Angle of Attack
the results compared with available experimental data. Mag [m/s] Dir [deg] [deg] [deg]
The present paper is organized in the following manner: the 1 3.0 55 55 0
problem definition, including the variables of interest, is intro-
3 3.0 90 55 35
duced; then both CFD numerical models are discussed; the chal-
lenges posed by turbulence modelling, and the practical compro-
mises required; followed by the results obtained with and without
turbulence modelling, and comparison with experimental results;
finally a summary of the findings is made.

PROBLEM DEFINITION
A captive generic lifeboat shape (axisymmetric bullet) is
forced into the water surface at a constant velocity of 3.0 m s−1 .
The angle β between the imposed trajectory and water plane is
also prescribed. The vessel has a fixed pitch angle, θ, and an (a) Case 1 (b) Case 3
angle of attack γ. The quantities of interest are the body fixed
forces acting on the model in the X-Z plane, i.e., the longitudinal FIGURE 2: Illustration of the two cases considered.
force FX , vertical force FZ and pitch moment MYY , as defined
in Figure 1. The time is zeroed when the model first touches
NUMERICAL MODEL
the water for both the experiments and simulations. The main
dimensions of the model are given in Table 1. Flow Solver ReFRESCO
Two cases were tested: case 1, with no angle of attack, for ReFRESCO (www.refresco.org) is a community based
which a V&V study was carried out previously [1]; and case 3 open-usage CFD code for maritime applications. It solves
with a significant angle of attack. These cases are defined in multiphase (unsteady) incompressible viscous flows using the
Table 2 and shown schematically in Figure 2. Navier-Stokes equations, complemented with turbulence mod-
els, cavitation models and volume-fraction transport equations

2 Copyright
c 2019 by ASME
for different phases [16]. The equations are discretised using a Boundary Conditions The problem symmetry is used
finite-volume approach with cell-centered collocated variables, to model only half the domain, thus saving computational re-
in strong-conservation form, and a pressure-correction equation sources. The computational domain is shown in Figure 3. A
based on the SIMPLE algorithm is used to ensure mass conser- no-slip boundary condition is applied at the bullet surface. A
vation [17]. Time integration is performed implicitly with first symmetry boundary condition is applied both to the symmetry
or second-order backward schemes. At each implicit time step, plane and the opposite surface in the y direction. The hydrostatic
the non-linear system for velocity and pressure is linearised with pressure is applied to the cylindrical curved surface, and updated
Picard’s method and a segregated approach is used. at each time step to reflect the change of depth due to the imposed
ReFRESCO (v2.5.0) is currently being developed, verified motion.
and its several applications validated at MARIN (in the Nether-
lands) [18, 19] in collaboration with IST (in Portugal) [20],
WavEC (Portugal) [21], and several other universities. Numerical Settings The motion of the body is pre-
scribed using the moving grid technique. The NS equations
are solved in the earth fixed reference frame, and the imposed
Mathematical Models velocity is applied to the grid as a whole, which is accounted
The impact of turbulence is investigated by comparing sim- for in the convection terms. The second order unstructured
ulations with and without turbulence modeling. Both approaches QUICK scheme, the ReFRICS scheme, and the first order up-
are here briefly described. wind scheme, are used for the spatial discretization of the mo-
The turbulent approach considers that turbulence plays a mentum equations, free surface equation, and the turbulence
non-negligible role in this application. The Reynolds averaged model equations, respectively. The diffusion terms are treated
Navier-Stokes (RANS) equations are solved assuming incom- using central schemes and the time derivatives using a first order
pressible flow. Closure is achieved with the eddy viscosity based, implicit Euler scheme. The iterative convergence
q P was defined
N
two equation turbulence model, k − ω SST model [22]. The re- per time step as L2 norm ≤ 10 , where L2 = N i=1 Res2i .
−5 1

sults obtained with this model are referred as SST results.


The laminar approach assumes that inertia/pressure effects
are dominant over the viscous effects, and therefore there is no
benefit from an engineering perspective to model turbulence and
solve additional equations. The incompressible Navier-Stokes
equations are solved with no turbulence modeling. This approach
was taken in a first stage of this work, and more details can be
found in [1]. The results obtained with this model are referred as
Laminar results.
In both cases, the multiphase is addressed with the Volume
of Fluid (VoF) approach. This technique models both air and
water as a single continuum fluid, with varying density ρ and
viscosity µ, defined as

ρ = αρair + (1 − α)ρwater
µ = αµair + (1 − α)µwater

where the air volume fraction, α, is physically bounded between


0 (no air) and 1 (only air), and obtained from a convection equa-
tion

∂α
+ div(αv) = 0 (1)
∂t
FIGURE 3: Computational domain.L is the bullet length and B
is the bullet width. Reproduced from [1].
Initial Conditions The fluid is assumed to be at rest,
v=0, at t=0. The free surface initial position is set at 0.25L below
the bullet tip. The hydrostatic pressure is applied to the cells un-
derwater, and a reference pressure equal to zero elsewhere. A ve-
Contact Line Problem
locity of 3.0 m s−1 is prescribed along the selected trajectory, and
Initial simulations of the turbulent flow were carried using
kept constant throughout the simulation. When applicable, the
a variation of the coarse grid with with y+ ≤ 1, which resulted
turbulent quantities are initialized throughout the domain from a
in the air volume fraction distribution seen in Figure 4 (a). The
prescribed eddy viscosity ratio µt /µ = 0.1 and turbulent inten-
air is dragged into the water as the bullet dives, resulting in a
sity of 10 %.
nonphysical layer of air covering the bullet throughout its dive.

3 Copyright
c 2019 by ASME
This is the result of the no slip boundary condition, which accuracy, even in flows where their assumptions are not entirely
when applied to the very small cells required to satisfy y+ ≤1, applicable. This is such a case, where none of the assumptions
results in extremely low relative velocities at those cells. Since needed to derive the log-law are satisfied: different pressure gra-
the variation of the air volume fraction, α, is exclusively due to its dients are present; laminar flow regime is present at different
convection by the velocity field (see Equation 1), the air volume stages of the dive; and flow separation is present at the stern.
fraction near the walls remains virtually unchanged throughout Another challenge is related to the different properties of
the simulation. This phenomena is commonly known as the con- both fluids, which result in a significant difference of y+ values,
tact line problem. depending on the local properties of the fluid. In this case we
opted to use as reference the y+ values for the underwater part
of flow. However, for a case where the free falling trajectory is
being calculated as a function of the aerodynamic forces, care
should be taken to ensure acceptable y+ values and near wall
treatment throughout the whole simulation. This is particularly
relevant for cases where the aerodynamic forces play a role in
determining the pitch angle at the moment of impact, which is a
key factor in the subsequent trajectory of the lifeboat.
Based on sensitivity tests, an average y+ ' 300 (see Fig-
ure 4 (c)) was chosen as the compromise between free surface
sharpness and near wall flow.

(a) y+ ≤ 1, no wall functions (WF) (b) y+ ' 60, automatic WF

NUMERICAL SENSITIVITY
The sensitivity of the results to several numeric parameters
(domain size, free surface modelling, iterative convergence, time
step and grid convergence) was addressed previously for the lam-
inar approach in Case 1 [1]. The conclusions regarding domain
size, free surface modelling and iterative convergence sensitivity
were assumed to hold for Case 3 and the turbulence approach.
However, grid and time step convergence studies were still car-
ried out for Case 3 with both approaches, and for Case 1 with the
turbulence approach.
(c) y+ ' 300, automatic WF (d) y+ ' 600, automatic WF

FIGURE 4: Case 1 air volume fraction field (air in red) for turbu-
lent flow, with different y + values and wall functions (WF). Grid and Time Step Convergence
Based on previous results [1], it was found that the solution
is more sensitive to the grid refinement than the time step size,
Two attempts to solve this issue were pursued. The first was provided that the maximum CFL was between 3 and 0.3. There-
to use wall functions, which relax the y+ condition on the cells fore, three different grids were tested with a time step chosen
near the wall, resulting in larger cells and larger velocities. The to ensure a maximum CFL number of approximately 0.4. Some
second was to define a maximum distance to the wall, below properties of the generated grids are given in Table 3 and Figure
which the air volume fraction was not solved, but instead extrap- 5. The coarse grid used here is based on Grid 5 from [1], with
olated from cells outside this distance. The latter approach was some minor alterations regarding grid topology.
discarded in favour of the first, due to the limited success ob- It should be noted that the grids used in the turbulent simula-
tained, high dependency on the distance chosen, and additional tions are obtained from the grids used with the laminar approach.
effort required for convergence. An additional refinement layer near the hull is added to meet the
desired dimensionless wall distance (y+ ) criteria. Otherwise, the
grid remains the same.
Wall Functions The results obtained with the different grids are shown in
Wall functions are based on the so called log-law, which was Figure 6 for Case 3. Similar trends are obtained for Case 1,
derived for a steady, bi-dimensional, fully turbulent flow over a which is omitted for brevity. The maximum differences in the
flat plate with zero pressure gradient. They allow larger near loads obtained between the coarse and fine grids are under 4 %.
wall cells, reducing the computational costs, at the expense of These differences are consistent with those found in the previous
accuracy in the near wall flow. In addition, the larger cell sizes exercise [1] using a wider range of grids for Case 1 with laminar
(y+ ' 100 vs. y+ ' 1) result in larger velocities at the first cell, flow.
which facilitates the convection of the air volume fraction, thus During the initial stage of the dive, approximately up until
mitigating the contact line problem. 105 ms, occurs the formation of the bow wake, which displaces
Wall functions are often used in the industry in a number water and drags air. In this phase there is no significant varia-
of applications, as the standard compromise between costs and tion in the loads with grid convergence. As the dive continues,

4 Copyright
c 2019 by ASME
this wake builds up along the body of the bullet, until it collapses Solution Verification Procedure
between 140 ms and 175 ms. Grid refinement seems to lead to- A description of the verification procedure, within the ver-
wards a less abrupt response in the forces, both in the laminar ification and validation (V&V) framework, can be found else-
and turbulent approach. After the bow wake collapse, with air where [23]. For the laminar approach of Case 1, the numer-
entrapment and eventual escape, there is a second wake forma- ical uncertainty for the flow quantity φ was estimated [1] us-
tion at the stern after 245 ms. In this section there is a visible ing independent contributions from the time, δt, and space, δx,
difference between the fine and coarse grids, with the grid re- discretizations, assuming the general Richardson extrapolation
finement trending towards smaller magnitude loads. form:

ε(φ) ' φi − φ0 = δx + δt = αx hpi x + αt τipt (2)

Where φi is the calculated value of the quantity of interest for


which the uncertainty is to be estimated; φ0 is the estimate of
the exact solution; αx and αt are constants to be determined;
px and pt the observed order of convergence in space and time,
respectively; hi and τi the typical spatial and time discretization
size, respectively. More details can be found in .
For the remaining simulations, the numerical uncertainty
was estimated based on a single parameter, that combines the in-
(a) General view (b) Coarse fluence of both the time, δt, and space, δx, discretizations [23].

ε(φ) ' φi − φ0 = δ = αhpi (3)

Where τi and hi are combined into a single typical size that takes
into account the formal orders of grid and time convergence. The
simulations are carried out at a constant CFL, thus preserving the
ratio of the spatial/time discretization. With this approach, the
number of the extrapolation unknown variables is reduced, thus
reducing the number of simulations required in order to estimate
(c) Medium (d) Fine
the numerical uncertainty. However, it is no longer possible to
FIGURE 5: Computational grids used in the laminar simulations. discern between the influence of the time and spatial resolution,
which is useful, should the need to reduce the estimated uncer-
tainty arise. This disadvantage is mitigated by the previous study
of Case 1, where it was observed that for a CFL number between
3 and 0.3, the solution is more sensitive to the grid refinement
than to the time step size.
TABLE 3: Properties of the generated computational grids.

Total number q
3 N1 Representative cell size [mm]
Grid Ni Model Comparison
of cells, Ni smallest free surface
The comparison between the laminar and turbulence ap-
Coarse 307,572 2.03 4 54 proaches is analogous to that used in the standard validation pro-
Medium 503,464 1.73 3 40 cedure [24]. However, instead of estimating the modelling error
Fine 2,586,671 1.00 2 25 of a given numerical model by comparing its results with experi-
ments, it estimates the modelling difference between the laminar
model and the turbulence model. The modelling difference be-
tween both models, ∆M , is given by

C − Ucmp ≤ ∆M ≤ C + Ucmp (4)


SOLUTION VERIFICATION AND MODEL COMPARISON
For each case, a solution verification procedure is followed
to estimate the numerical uncertainties for both the laminar and The comparison difference, C, is here defined as the differ-
turbulence approach, at a set of predefined time instants shown ence between the turbulence solution, SSST , and the laminar one
in Figure 7. These uncertainties are then combined to provide a SLAM .
measure of the uncertainty of the comparison between the lami-
nar and turbulence results.
C = Slam − Ssst (5)

5 Copyright
c 2019 by ASME
(a) Case 3, laminar approach (b) Case 3, turbulence approach

FIGURE 6: Sensitivity to the grid size, with max CFL≤ 0.4. The error bars represent the numerical uncertainty.

FIGURE 7: Snapshots for turbulent flow of Case 3 at different time instants.

6 Copyright
c 2019 by ASME
C is analogous to the comparison error, E, used in the standard Vertical force, FZ . Both the laminar and the turbulent
validation terminology, used to compare the simulation result approaches provide a very similar result. In fact, the difference
with the experimental one. between both models is within the numerical uncertainties ob-
The comparison uncertainty, Ucmp , is a measurement of the served, i.e., the comparison difference (under 5 N) is less than
uncertainty of the comparison difference, C, and is given by the overall uncertainty.
q A good agreement with the experimental data is found, with
Ucmp = 2
Uφ, 2 the oscillations in the experimental loads attributed to the exper-
sst + Uφ, lam (6)
imental setup [1].
Consistent with the observations for the longitudinal force,
Where Uφ, sst is the numerical uncertainty of the turbulent sim-
the larger uncertainties occur for t ≥245 ms, when the formation
ulations, and Uφ, lam , is the numerical uncertainty of the lami-
of the wake at the stern develops.
nar simulations. Ucmp is analogous to the validation uncertainty,
Uval , in the standard V&V terminology. Since the same input pa-
rameters were prescribed for both sets of simulations, no input Pitch moment, MYY . The difference between both mod-
parameter uncertainty is considered. els is more noticeable for the pitch moment. Here, the maximum
comparison difference is around 2 N m, with the turbulent flow
generally predicting a larger moment. Nevertheless, the differ-
RESULTS ence between both models is within the uncertainty levels.
In the very early stages of the dive t ≤45 ms, the turbulent
Case 1
model underpredicts the pitch moment by almost 10 %. This can
The field distributions of the eddy viscosity, air volume frac-
be determinant for the lifeboat trajectory in cases where its mo-
tion, and velocity magnitude are illustrated in Figure 8 for two
tion is being calculated as a function of the hydrodynamic forces.
different stages of the dive. The loads obtained with the lami-
There is a reasonable agreement with the experimental data,
nar and turbulence models, and the comparison between both, is
with no noticeable improvement when using the turbulence
shown in Figure 9.
model instead of laminar flow.

Flow field visualization The eddy viscosity is signifi-


cant in three areas: in the submerged part of the bullet; in the Case 3
vicinity of the free surface; and in the stern wake area. Neverthe- The field distribution of the eddy viscosity, air volume frac-
less, the impact on the velocity field is restricted to the wake area, tion, and velocity magnitude are illustrated in Figure 10 for two
or close by, as seen in the velocity magnitude plots. There is a different stages of the dive. The loads obtained with the lami-
high velocity area below the stern present in the laminar results nar and turbulence models, and the comparison between both, is
that is mitigated with turbulent flow. When the bullet is com- shown in Figure 11.
pletely below the water level at rest, the turbulent model shows
signs of two vortices at the stern, which are not present with lam-
inar flow. The air volume fraction shows a slightly sharper tran- Flow field visualization The eddy viscosity is signifi-
sition with the turbulent model during the mid dive phase, but cant at the wake generated in the bow section, at the overall hull,
no difference can be seen between the models once the bullet is and the stern wake. Eventually, the influence of the eddy viscos-
completely below the water surface at rest. ity expands to merge all those areas into one big wake with large
eddy viscosity. Similar velocity fields can be observed, partic-
ularly in the underwater section. The largest differences in the
Longitudinal force, FX . While the comparison differ- velocity field occur at the stern wake, in the air region, where
ence is small, under 10 N, the turbulence model consistently pre- the laminar flow predicts a larger wake than the turbulent flow
dicts a larger force than the laminar approach. Furthermore, approach.
between 125 ms and 245 ms the difference between both ap-
The turbulence model leads to a slightly sharper interface
proaches, including the uncertainty bars, does not contain zero.
between air and water mid dive. However, the largest difference
This suggests that the difference between both approaches is
is the underprediction of the bow wake by the laminar flow, when
larger that what can be attributed to the numerical uncertain-
compared with the turbulent flow. Even in the later stages of the
ties. As the bullet is completely below the water surface at rest (
dive, the turbulent approach shows a more dynamic free surface
t ≥245 ms) and the wake in the stern is formed, larger uncertain-
behaviour, with more air entrained, than the laminar simulation.
ties arise.
The largest differences between models occur in the wake
This trend brings the turbulent flow results even closer to
regions, as expected. In these areas, the eddy viscosity increases
the experimental results. In Case 1, the trajectory is aligned
and some differences in the velocity field and air volume fraction
with the symmetry axis, which minimizes any separation effects.
are visible.
Therefore, the viscous forces play a relevant role in determining
the longitudinal force, and the turbulence model contributes to
the viscous force component by increasing the wall shear stress.
Longitudinal force, FX . The comparison difference is
Nevertheless, both simulations show good agreement with the
under 10 N. The turbulent flow exhibits a larger force than the
experimental results.
laminar flow up until the collapse of the bow wake, after which

7 Copyright
c 2019 by ASME
FIGURE 8: Case 1 flow field visualization of the eddy viscosity ratio (left column), velocity magnitude (center column) and air volume
fraction (right column), at 125 ms (top row) and 345 ms (bottom row). The simulation with the turbulent flow is shown by the color map,
while the laminar flow is shown by the lines. The free surface (α=0.5) is represented by a dotted black line.

(a) Laminar, turbulent, and experimental time traces. (b) Comparison difference between both numerical models (C =
Slam − Ssst ).

FIGURE 9: Case 1. The left plots show the time traces for the laminar (Lam) and turbulent (SST) flows, including its numerical
uncertainties, and the experimental results (EXP). The right plots show the comparison difference, C, and its uncertainty Ucmp

8 Copyright
c 2019 by ASME
FIGURE 10: Case 3 flow field visualization of the eddy viscosity ratio (left column), velocity magnitude (center column) and air volume
fraction (right column), at 105 ms (top row) and 255 ms (bottom row). The simulation with the turbulent flow is shown by the color map,
while the laminar flow is shown by the lines. The free surface (α=0.5) is represented by a dotted black line.

(a) Laminar, turbulent, and experimental time traces. (b) Comparison difference between both numerical models (C =
Slam − Ssst ).

FIGURE 11: Case 3. The left plots show the time traces for the laminar (Lam) and turbulent (SST) flows, including its numerical
uncertainties, and the experimental results (EXP). The right plots show the comparison difference, C, and its uncertainty Ucmp

9 Copyright
c 2019 by ASME
the turbulence flow predicts a smaller force. Furthermore, be- FX than the laminar approach, and in some cases this difference
tween 70 ms and 140 ms the difference between both models can- is larger than the comparison uncertainty. This difference can be
not be explained only by the comparison uncertainty. The larger explained by the increased viscous forces acting on the body due
uncertainties occur after the bow wake collapse and when the to the higher shear stress predicted by the turbulent flow.
bullet is below the water surface level at rest (t ≥175 ms). The turbulent flow shows a better agreement in FX with the
It should be noted that in the early stages of the impact, dur- experimental results than the laminar flow. However, there is al-
ing the bow wake formation, the relative difference between both ready an excellent agreement between the laminar FX and the
models reaches a peak value of 30 % at 70 ms. This can be de- experimental result. Therefore, it can be argued if this small ad-
terminant for the lifeboat trajectory in cases where its motion is ditional accuracy justifies the increased computational costs re-
being calculated as a function of the hydrodynamic forces. quired to solve the turbulence model for Case 1.
When compared to the experimental results, both ap-
proaches seem to underpredict the longitudinal force during the
formation of the bow wake t ≤140 ms, and then overestimate it Case 3
after the bow wake collapse t ≥175 ms. Nevertheless, the turbu- The turbulence effect was expected to play a more prominent
lence flow results seem to qualitatively follow the trend of the ex- role in Case 3, due to the larger wake and flow separation, which
perimental curve better than the laminar approach, and a reason- should not be correctly captured with the laminar flow approach.
able agreement between simulations and experiments is found. Some differences were observed in the flow field for turbulent
flow, such as the larger bow wake, or the more dynamic wake
collapse, as seen in the air volume fraction distribution. In fact,
Vertical force, FZ . The comparison difference is under the largest differences between both approaches, for all the loads,
10 N until the bow wake collapse (approx. 175 ms), after which it were generally observed during/after the bow wake collapse.
ranges between 10 N and 20 N. After the bow wake collapse, the
Both simulations show reasonable agreement with the ex-
turbulent flow underpredicts the vertical force magnitude when
perimental data for FX and MYY . However, for FZ a reasonable
compared to the laminar case.
agreement is found until the bow wake collapse t ≤140 ms, af-
The differences between both models seem to be signifi- ter which both simulations deviate significantly from the experi-
cant during the stages where the bow wake collapses (approx. ments. Nevertheless, the agreement with experimental data is is
175 ms), or the stern wake begins t ≥210 ms, i.e., the differences slightly better in turbulent flow for FX and FZ .
between both models cannot be entirely explained by the com-
To a lesser extent than for Case 1, it can also be argued from
parison uncertainty.
an engineering point of view, if the small differences between
A good agreement with the experimental data is found for
both approaches justify the additional computational cost of the
both approaches up until the bow wake collapse t ≤140 ms, af-
turbulence modelling.
ter which both simulations deviate significantly from the exper-
Although there is a small absolute difference in the global
iments. Despite this poor agreement during/after the bow wake
loads between both approaches, there is a high relative difference
collapse, the turbulent flow is marginally closer to the experi-
in the early stages of impact. This difference can be relevant for
mental results.
cases where the motion is being calculated as a function of the
loads acting on the vessel, as the early stages of the dive can
Pitch moment, MYY . Both models show very similar impact the lifeboat trajectory.
pitch moment, with the maximum comparison difference around
2 N m, which is within the comparison uncertainty, except at
140 ms. The largest difference between both models is found LIMITATIONS AND IMPROVEMENTS
during the bow wake collapse, between 140 ms and 175 ms.
Case 3 represented a more vigorous impact with greater an-
There is a reasonable agreement with the experimental data, gle of attack. Consequently, flow separation and wake effects are
with no noticeable improvement when assuming turbulent flow. a significant part of the physics involved in the problem. Thus,
The oscillations in the experimental curve are assumed to be re- from a fundamental point of view, applying wall functions be-
lated with the experimental setup, as described in [1]. comes even more questionable. Moreover, a bluff-body wake
can be greatly influenced by the boundary layer separation point
and flow regime (laminar vs. turbulent), for which wall function
DISCUSSION boundary conditions are also too simplistic. The effect of the
turbulence model is likely limited by the usage of wall functions,
Case 1 which in turn are used to avoid the contact line problem. The
Case 1 shows a trajectory that minimizes the wake forma- possibility of solving the near wall flow, avoiding the wall func-
tion, while Case 3 is a more violent impact, with wake formation tions without facing the contact line problem, should be further
at the bow that later collapses. Therefore, it was expected that explored.
the laminar approach performs better in Case 1 than Case 3, as These results were obtained with the eddy viscosity type,
was observed in these simulations. two-equations k − ω SST model turbulence model [22]. While
For Case 1, there is a small difference between both models some simulations were carried out with another turbulence model
regarding the FZ and MYY , which is within the comparison un- within the same framework (K-kL model), and no significant dif-
certainty. The turbulent flow predicts a larger longitudinal force ferences were found, it is possible that other turbulence models

10 Copyright
c 2019 by ASME
within the RANS framework or otherwise show a different be- are often within the comparison uncertainty. While this is not
haviour. surprising for Case 1, it is somewhat unexpected for Case 3
Furthermore, the grid refinements used for both the laminar where significant flow separation and large wakes occur. It
and turbulent flow were the same. It is possible that a grid refine- is possible that the usage of wall functions is restricting the
ment that is acceptable for the laminar approach is too coarse for capacity of the turbulence model to capture these flow features,
the turbulence model to perform. thus approximating the results from both models. Further
research can focus on how to solve the near wall flow without
facing the contact line problem, thus avoiding the use of wall
functions and improving the turbulence model capacity to tackle
CONCLUSIONS
flow separation and large wakes.
Two simplified cases of lifeboat drops with constant
velocity were simulated with and without turbulence model.
Nevertheless, it was observed that the turbulence model has
In the turbulent flow approach, wall functions were used as
an impact on the early stages of the dive, which can be determi-
a compromise to avoid the contact line problem encountered
nant for cases where the lifeboat motion is being calculated as a
when solving the boundary layer. Three different grids were
function of the loads acting on the vessel.
used to estimate the numerical uncertainties of each model, and
the global loads acting on the vessel were compared for both
models. Experimental results are presented for reference. The
following concluding remarks are made based on the presented ACKNOWLEDGEMENTS
results: The authors are grateful to Equinor ASA for permission to
publish the experimental data.
Wall functions are used as a compromise to prevent the
contact-line problem, where the bullet would penetrate the
surface and remain enveloped by a thin sheet of air for an REFERENCES
unrealistically long duration. [1] Maximiano, A., Vaz, G., and Scharnke, J., 2017. “CFD Verifi-
cation and Validation Study for a Captive Bullet Entry in Calm
Case 1, where wake effects are less relevant, showed no Water”. In ASME 2017 36th International Conference on Ocean,
Offshore and Arctic Engineering, American Society of Mechani-
significant difference for the vertical force FZ and pitch moment
cal Engineers.
MYY between the laminar and turbulent simulations, i.e., the dif-
ferences are smaller than the comparison uncertainty. However, [2] IMO Publishing. Life-Saving Appliances. 2017 edition, Interna-
tional Martime Organization.
the turbulent flow consistently predicts a larger longitudinal
force FX than the laminar approach, presumably due to the [3] DNVGL-ST-E406. Design of free-fall lifeboats. 2016 edition,
DNV-GL.
increased wall shear stress.
[4] Berchiche, N., Östman, A., Hermundstad, O. A., and Reinholdt-
Case 1 showed a good agreement between the laminar sen, S.-A., 2015. “Experimental validation of CFD simulations
of free-fall lifeboat launches in regular waves”. Ship Technology
simulations and the experimental data for FZ and MYY , and
Research, 62(3), pp. 148–158.
an excellent agreement for FX . The turbulence model does
not improve on this situation, except for the longitudinal force [5] Mørch, H., Enger, S., Peric, M., and Schreck, E., 2008. “Simula-
tion of lifeboat launching under storm conditions”. In CFD Marin
FX . However, since the agreement with laminar flow is already
& CD-Adapco, 6th International Conference on CFD in Oil & Gas,
excellent, it can be argued if the additional accuracy justifies the Metallurgical and Process Industries, SINTEF/NTNU, Trondheim,
increased computational cost of the turbulence model. Norway.
[6] Mørch, H. J., Peric, M., Schreck, E., el Moctar, O., and Zorn,
Case 3 is defined by a violent impact with flow separation T., 2009. “Simulation of flow and motion of lifeboats”. In
and wake effects. Some differences are observed between the ASME 2009 28th International Conference on Ocean, Offshore
laminar and turbulent simulations, in particular during/after and Arctic Engineering, American Society of Mechanical Engi-
the bow wake collapse. However, the difference found at neers, pp. 595–605.
the selected time instants is generally within the comparison [7] Johannessen, S. R., 2012. “Use of CFD to study hydrodynamic
uncertainty. loads on free-fall lifeboats in the impact phase.: A verification and
validation study.”. Master’s thesis, Institutt for marin teknikk.
Case 3 showed a reasonable agreement between the laminar [8] Netland, V., 2017. “Hydrodynamic effects relevant for free-falling
flow and the experimental data for FX and MYY . A reasonable lifeboats in wave conditions”. Master’s thesis, NTNU.
agreement is also found for FZ up until the bow wake collapse, [9] Tregde, V., 2015. “Compressible air effects in cfd simulations of
after which both simulations deviate significantly from the free fall lifeboat drop”. In ASME 2015 34th International Confer-
experiments. The turbulence model seems to improve slightly on ence on Ocean, Offshore and Arctic Engineering, American Soci-
the experimental comparison for FX and FZ , however the same ety of Mechanical Engineers, pp. V002T08A018–V002T08A018.
argument can be made regarding the small additional accuracy [10] Svendsen, A. G., 2017. “Hydrodynamic effects regarding free-fall
versus increased computational costs. lifeboat for compressible air in CFD simulations”. Master’s thesis,
NTNU.
Based on these results, the effects of the turbulence model [11] Veldman, A. E., Seubers, H., van der Plas, P., and Helder, J., 2017.
seems limited, and the differences to the laminar approach “Free-surface flow simulations with interactively moving bodies”.

11 Copyright
c 2019 by ASME
In ASME 2017 36th International Conference on Ocean, Offshore [18] Eça, L., and Hoekstra, M., 2012. “Verification and validation for
and Arctic Engineering, American Society of Mechanical Engi- marine applications of cfd”. In 29th Symposium on Naval Hydro-
neers, pp. V002T08A002–V002T08A002. dynamics (ONR), Gothenburg,Sweden, August.
[12] van der Plas, P., Veldman, A. E., Seubers, H., Helder, J., and Lam, [19] Koop, A., Klaij, C., and Vaz, G., 2011. “Viscous-Flow Calcula-
K.-W., 2018. “Adaptive grid refinement for two-phase offshore tions For Model And Full-Scale Current Loads On Typical Off-
applications”. shore Structures”. In Proceedings of ECCOMAS MARINE 2011,
[13] Bae, D.-M., Zakki, A., Kim, H.-S., and Kim, J.-G., 2010. “Estima- Lisbon, Portugal, September.
tion of acceleration response of freefall lifeboat using fsi analysis [20] Muralha, J., Eça, L., Maximiano, A., and Vaz, G., 2018. “Vali-
technique of ls-dyna code”. Journal of the Society of Naval Archi- dation Exercises for a Free Falling Wedge into Calm Water”. In
tects of Korea, 47(5), pp. 681–688. Proceedings of OMAE2018, Madrid, Spain, June.
[14] Ma, Z., Qian, L., Martinez-Ferrer, P., Causon, D., Mingham, C., [21] Maximiano, A., Vaz, G., and Scharnke, J., 2017. “CFD Verifica-
and Bai, W., 2018. “An overset mesh based multiphase flow solver tion and Validation study for a captive bullet entry in calm water”.
for water entry problems”. Computers & Fluids. In Proceedings of OMAE2017, Trondheim, Norway, June.
[15] Eça, L., Hoekstra, M., and Vaz, G., 2015. “Verification of solutions [22] F.Menter, J.C.Ferreira, T.Esch, and B.Konno, 2003. “The SST Tur-
in unsteady flows”. In ASME V&V Conference, Las Vegas, USA, bulence Model with Improved Wall Treatment for Heat Transfer
May. Predictions in Gas Turbines”. In Proceedings of the International
[16] Vaz, G., Jaouen, F., and Hoekstra, M., 2009. “Free-Surface Vis- Gas Turbine Congress 2003 Tokyo November 2-7, 2003, TS-059.
cous Flow Computations. Validation of URANS Code F RE SC O”. [23] Eça, L., Hoekstra, M., and Vaz, G., 2015. “Verification of So-
In Proceedings of OMAE2009, Honolulu, Hawaii, USA, June. lutions in Unsteady Flows”. ASME V&V Conference, Las Vegas,
[17] Klaij, C. M., and Vuik, C., 2013. “SIMPLE-type Precondition- USA, May.
ers for Cell-Centered, Colocated Finite Volume Discretization of [24] American Society of Mechanical Engineers, 2009. Standard for
Incompressible Reynolds-Averaged NavierStokes Equations”. In- Verification and Validation in Computational Fluid Dynamics and
ternational Journal for Numerical Methods in Fluids, 71(7), Heat Transfer. Tech. Rep. ASME V&V 20-2009, The American
pp. 830–849. Society of Mechanical Engineers (ASME).

12 Copyright
c 2019 by ASME

Das könnte Ihnen auch gefallen