Canad
National Research Conseil national Council Canada de recherches Canada Institute for Institut de
Marine Dynamics dynamique marine
SYMPOSIUM ON
SELECTED TOPICS OF
MARINE HYDRODYNAMICS
St. John's, Newfoundland
August 7, 1991
WAVE TANK/BASIN CFD SIMULATIONS
M.J. Hinchey
Ocean
Engineering
Research CentreNational Researcs, Council
I'U Cthada
InStitute fOr Mame Dynamics
Conseil national dé recherches Canada Institut de dnarnique manne
SYMPOSJiJM ON
SELECTED
TOPICS OF
MARINE
llYDRODyNA?JCS
St. John's, Newfôundjand
August 7, 1991
A3SAC
Computational Fluid Dynamics (CPD) is a very active research area. Most wave tank/basin CFD simulations are based ot a potential f low formulation. This paper describes CPD 3imulations based on a Návjer-Stokes (NS) förmujatjon. It describes a 2D finita lifference NS code bown as SOLA-voy that ias been used for vaya tank simulations and recent 3D NS code known as FLOW-3D that ould be used fOr wave basin simulations. 30th codes were developed by Hirt and his
:oileagues at the Los Alas
ScientificCaboratorjes. Neither code is set up for luid
interaction
with bodies ündergoing trbitrary motions. So, they cannot handle ¿ave interactions with floating bodies. The 'apar describes so coda modifications that ìight allow such interactions.dvantages/Disadvg
of the codes elative to Panel Iethod codes are briefly eviewed.BACKGROUND
rost wave tank/basin sjmulatjons are based rn a boundary integral
formulation and a 'anel Method djscretizatjon. The starting
oint for this is Laplace's equation for
)otential water flow. This is basically a ;tatement of corervatjon of mass for .nviscid incomprethle water undergoing an .rrotatjonal motion. This paper describes ainjt difference dtscretjzatjon of the rimitjve variable form of the Wavier-stokes
NS)
equations that coùld also be uséd for
ank/basjn
sirnulatj.
In terms of the
rimitjve variables,
conservation
of omentum considerations give:p ( au/at + u 8u/x
+ y auiay + w au/az ) - - ap/ax + 1 a/ax (u au/ax)
+
/y (
òu/8y)+ a/az
au/az)]
WAVE TANK/BASIN CFD SIMULATIONS
N.J. Hinchey
Ocean Engineering Research Centre Memorial University of Newfouzdland
32
p ( av/at + u av/ex + V
av/ay
4
w av/az ) - ap/ay + a/er öv/8x)+ a/a7 ( av/ay) + a/az
( av/az)]
av/at + av/ex + 8w/ay + w av/a
z )
- ap/as - pg + ( a/ax
(14 au/ax) + a/ai (p av/ay) +
a/az (u av/ez)]
where u,v,v are velocity
components
in the x,y,z directions, p is prèssure,p is the
density of Vater and p is its effective Viscosity. Similarly, to conserve
mass, the velocities and Pressure must Satisfy:
ap/at + a,'a + av/ay + aw,'az
) - o
where a is the speed of Sound in
water. Although Vater is
basically incompressible
for mass conservation,
the Ns/cpo codes
(l2J take it to be compressible..
A spécial function known as the volume of fluid
(VOF)
function P is used to locate
the water surface. Vithin the water, the VOP function is unity : outside
the water it is
zero.
Material volume considerations give,for it the governing equation:
a,at +
aP/ex + y aP/ay + w eF/as - OCodes based on these
equations would have to use an extremely fine mesh to accurately model
turbulent
flowssuch as those generat by waves breaking on a beach. This is because the scales of turbulence are very small. Turbulence is
characterized by small lumps of fluid known as eddies moving
arouzd in a rand fashion. Like the
molecules in a gas, eddies cause a diffusion of
moment.
They make the fluid appear more viscous.Now, gineers are usually not interested in the tails of the eddy metioñ. Instead, they steed models which account for the
difts..v character of turbulence. Such
modes can be obtained from the momentum
equa.n.s
by a complex time averagingpro. One popular two equation model
obtA±d in this way is known as the
k-e
aode. :3], wher.
k is
th. local intensity oftur..nca and e
is its dissipation rate. FLCW-1 [2] uses this model : SOL-VOp (1] does t. Its governing equations are:èk,?t + u 8k/ax + V 8k/ay + w 8k/8z
-T- Td
+ (8/8x (p/ak 8k/8x) + 8,'8y (u/a 8k/8y) + a/az (js/c 8k/8z)]/p8e19t + u
8e/8x+ V
8c/8y+ V
8c/8zD -
Dd + (8/8x (p/a1 8c/8x) + 8/8v (p/a1 8e/8y) + a/az (p/a1 8e/òz)]/pwhere
T, = G p / p
T C1 e / k
e
whee ak(l.0), a,(l.3), Cb(l.0), C1(l.44),
C2(l.92) and C1(O.9) are universal constants based on data
from geometrically simple experients, p1 is the laminar Viscosity of water, p is its apparent Viscosity due to eddy 'otion and G is a
production function
(3]. The k-c equations
account for the
convection, diffugjon, production
and
dissjtjo
of turbulence. Th time averaç processintroduces source-like ter into the momentum equations. Special
wall fctjos
are often used to simplifyconsjãeratjon of the sharp normal gradients in ve..ocitje3 and turbulence
near walls. Ever $lflce their
introduction back in the early seventies,
turbulence models, such as
the
k-ode1, hava been and still are quite one.rsja1, mainly because the so-called
univeaj
Constants seem to begeometry
1epen.
ach coverning
equation can be put into the
form:
- B
[1,,2J, each is integrated
numerically
tcrs a time step
to get new A from old A:
A(t + ¿t) A(t) + Lt B(t)
where th. various derivatives
in B are
discretized using finite difference approximations. The discretizatjon gives algebraic equations for the scalars p,P,k and e at points where grid lines cr.s and equations for Velocity components at staggered positions between the grid points. Central differences are used to disetizethe viscous terms. To
ensure nrica1
stability, a combination of central ¡ upvind differences is used for the corrvectïve terms. Collocation or lumping is u.sed for each of the T and D terms. The details of the discretjzatjon can be found. in [1,2]. To march the uniciowne forward in time, the momentum equations are used to update the velocities, the continuity equation is used to update pressure and the turi.lence equations are used to update the local intensity and dissipation rate of turbulence.
2. WAVE TANI SDWLATIONS
The SOL&-Vop code has been. used for 2D vave simulations (4]. In (5], we used it to study the wave attenuation characteristj of a number of breakwaters. In (6;7], we used it to study the waves generated by a hovercraft undergoing a pure sinusoidal heave motion overwater. Recall that hovercraft are
amphibious
yehicles which float ona shion
of pressurized air contained by a flexible
structure known as a skirt. Lift air is
typically supplied by one or more fans or blowers, and it leaks away to atmoephere through a gap beneath the skirt. For the breakwater study, waves were generated by imposing a periodic flap-like horizontal velocity profile at the left hand e of a tank. For the hovercraft study, they 'vere generated by imposing a periodic pressure over a section of the water surface at the left and: this vas taken to be uniform beneath the craft and undergo a linear fall-off to atmospheric pressure at the edge of the cushion. Group speed concepts vere used to stop each simulation before the profiles of interest could get contaminated by beach or end wall reflection.
Fig 1 shows a typical wave profile from the breakwater study. The breakwater was nonporous and was generated by blocking out specific cells in, the finite difference mesh. Note the superposition of incident and reflected waves just upstream of the breakwater. This was used to determina the reflection coefficient of the breakwater. Comparison with its transmission
coefficient showed that there was coñaiderable energy. dissipation. Runs for this case typically
took several hours of
VAX CPU time
:
however, runs for a porous geometry often took. days, especially when to
ccuracy an automatic tine step reduction ras employed by the code. For the breakWater tudy, the water was typically dlscretjzed .'y 25 horizontal grid lines and 200 vertical rid lines.
'ig 2 shows a typical result fr the overcraft study. For this, the hydrostatic rough normally found beneath a craft vas ot allowed to form. The figure gives the ave profiles generated near a craft by a ure heave motion with frequency i cps: IN e fers to a component in phase with pressure hile OUT refers to a component 90° out of hase. The left half of each profile orresponds to the region directly beneath he craft while the right half corresponds o the region just outside. One vili note hat the. waves beneath the craft are asicaily standing waves vtiile outside they re propagating waves. As can be seen, the refiles beneath compare favorably with orresponding potential flow profiles enerated by Lamb (8] and shown in Fig 3. he duration of each simulation in real time as typically 10 seconds and the profiles iven are for the last second or cycle of his. By this time, start-up transients were significant. For each simulation, the ater was discretized typically by 20 orizonta]. grid, lines and 250 vertical grid
ines : about 20 of the vertical lines were
riside the cuShion. This was considered to e adequate for the wavelengths generated.
ic grid vas uniform throughout except for e far right end where it vas allowed to pand horizontally so that reflected waves rom a vertical wall there did not have a hance to form. For the profiles, the drop-tf in pressure at the édge of the cushion curred across the 4 vertical lines just
utside the edge. Similar profiles were tained with a nonzero hydrostatic trough.
have also used the. simulation to study e heave stability of hovercraft hovering
d-erwatar, where the waves
generated by aave motion modulate the cushion volume and e flow of air beneath the skirt (6,7]. We
3und that Water compliance can improve ability at high Cushion pressures but tuse many stability transitions at low
essures. For this work, the hydrostatic rough beneath the craft vas not allowed to ,ra because with it motions were erratic.
Fig i Wave Interaction with Nonporous Breakwater : SOLA-VO? Simulation.
34
In reality, this trough nay degrade stability at high cushion pressures. Some of the stability trends at low cushiön pressureswere recently confirmed iñ a wave tank setup at ¿(UN (7].
WAVE BASIN SIMULATIONS
The FLOW-3D code could be used for wave
basin simulations. Procedures used in SOLA-VOY to generate and dissipate waves should also work in FLOW-3D. Unfortunately, we have not been able to check this because the FLOW-3D code costs approximately $50, 000 US (SOlA-Voy $800 US)! It should be possible with the code to study the diffraction of wave fields by complex 3D shapes. According to (9], several research institutes in the. UK are working on this.
BODY MOTIONS TE TO WAVES
Neither NS/CPD code can handle bodies undergoing arbitrary motions in waves. However, they can handle special cases such as constant speed body motions. For example, ve used SOLA-Vop to study the impact of submersible capsules onto the ocean surface following free fall from the deck of an oil rig. Hirt and his colleagues have used FLOW-3D in a similar way (10]. A typical result from their work is shown in Fig 4. They are nov attempting to extend the code to the. arbitrary motion case (il] : unfortunately, success to date with this has
been
very modest. It might be possible tosetup a pressure iteration over a section of the water surface that over a time step would move the water the way a moving body would. This is something that Hirt and his colleagues have not tried.
DISCUSSION
Probably the major advantage of the NS/CFD codes relative to Panel Method codes is they can handle nonlinear phenomenon such as
breaking waves and turbulence. Their major disadvantage is they presently cannot handle arbitrary body notions such as those of floating bodies in waves. Another disadvantage is in 3D they discretjze the
OUT
M**.
W%WW
*
w
w w
Fig 2 Water Waves near Hovercraft undergoing Pure Heave Motion : SOLA-VO? Siu.1ation.
DISTANCE OUT FROM CRAÉT CENTfl ()
space occupied by a fluid while the Panel
'(ethods discretize s or all of the 3urface surrounding the space : thus Panel '(ethod codes are potentially caputationally nora efficient. Also, Panel Method codes
pproxizate colex shapes
better than IS/CPD codes, which generally employ ectangular grids. Curvilinear coordinates r finite element geometry mappings could ?robably b.Used to ctaract this. Por
3ome geometries, r VAX SOL-VOP impact & reakwater computations werequite cu
intensive.
So, a ercomputer woúldrobably be needed for 3D computations.
.nothar problem with NS/CYD codes is iomething called false diffusion : because
the upwind difference treatment of the ;onvective terms in the transport equations,
.ha Vater appears more viscous than it
eally is : so, XS/C?D acouracy can be low.
;kew upwind difference schemes can :ounteract this somewhat : however, these
re not used by the NS/D codes.
COWLEbGEMENT
his work was funded by NSERC through perating Grant A4955(Hinchey).
ZPER.ENcEs
Nichols, B.D., Hirt, C.W. and Hotchkiss,
.5. (1980): SOLA-VO? : .A Solution
lgoritbm for Transient Fluid Plow with
ultip1e Free Boundaries", Los Alamos ;cientjfjc Laboratories, Los Alamos, New (exico, USA.
Hirt, C.W. (1991):
"FLOW3D User
anual", Plow Science Incorporated, Los lamos, New Mexico, USA.
Launder, B.E. and Spa.lding, D.B. (1972): Lectures in Mathematical Models of .'urbulence", Acadeiic Press.
Su, T. (1984): 'Wave Breaking on Sloping eaches', Fourth International Conference on .pplied Numerical Modelling, Taiwan, ROC.
Yetman, R. and Rinchey, fl.J. (1990): Wave Attenuation by the Deltaport Floating rea)cwater", CS Mechanical Engineering orum 1990, Toronto.
Hinchey, *.J. (1989): "Numejca1 imulatjon of Overvater Heave Motion Of 2D overcraft", CACTS 89, Transport Canada, TDC .eport TP 8979.
Hinchey, X.?. and Sullivan, P.A. (1991): Experiments on Hovercraft Overwater tability", POAC 91, st.
John's, ewfoundland, Canada.
36
Lamb, H. (1904): "On Deep Water Waves", Proceedings of the London Mathematical Society, Series 2, Volume 2.
Eatock-Taylor, R. (1991) : Oxford University, UK, Private Communication.
Hirt, C.W. (1990): "Water Entry of High Speed Projectiles", Flow Science Incorporated, New Mexico, (ISA, FSI-90-TN23.
li. Sicilian, 3. (1990): "A FAVOR Based Moving Obstacle Treatment for FLOW-3D", Flow Science Incorporated, New Mexico, USA, PSI-90-00-TN24.