Proceedings of the Twenty-fourth (2014) International Ocean and Polar Engineering Conference www.isope.

org
Busan, Korea, June 15-20, 2014
Copyright © 2014 by the International Society of Offshore and Polar Engineers (ISOPE)
ISBN 978-1 880653 91-3 (Set); ISSN 1098-6189 (Set)

Experimental and Numerical Investigation of Single Impacts in a 2D Tank

Jens Neugebauer, Ould el Moctar, Robert Potthoff

University of Duisburg-Essen, Institute for Ship Technology, Ocean Engineering and Transport Systems
Duisburg, Germany

high-speed videos allowed a qualitative evaluation of the free surface

behavior.
The present paper is aiming at quantitative comparison of the velocity
ABSTRACT fields obtained from PIV measurements and numerical analysis.
The paper presents an experimental and numerical investigation of the Further, pressure time traces from experiments and CFD are compared.
flow inside a 2D tank subjected to motions resulting in waves Single-Impact-Wave (SIW) conditions are used for this purpose. Tests
impacting on one side of its roof. Tank motions are generated by a are conducted according to the specification of the benchmark, Loysel
hexapod and pressure values are recorded at 50 kHz sampling rate from (2013). The repeatability of the pressure measurements is evaluated.
piezo-resistive pressure sensors. Additional to pressure measurements, PIV is a well-known technique for the measurement of velocity fields,
Particle-Image-Velocimetry (PIV) is used to measure the fluid velocity but additional attention to tank and free surface motions is required in
field with an acquisition rate of 100Hz. The repeatability of the sloshing analysis. Tank motions require that optical components such
physical tests is investigated. Challenges related to PIV measurements as cameras and laser-optical devices are mounted on the motion
of the violent flow inside the tank are discussed. Numerical simulations platform. The mounting systems shall be insensitive to rigid body
are carried out for the above described single-impact tests. The motions and vibrations. Further, the free surface and air pockets may
Reynolds-averaged Navier-Stokes equations (RANSE) are solved using reflect the laser light which may damage the cameras or lead to
the open source code OpenFOAM and the commercial code insufficient lighting conditions. As the tank is made of Plexiglas,
STARCCM+. The free surface is captured based on the Volume of reflections at the tank boundaries may occur. Thus, the laser is
Fluid method. Discretization errors are investigated. Finally, the introduced far below the free surface and reflections are reduced using
computed and measured flow velocity and pressure are compared. fluorescent coatings on the tank walls.
KEY WORDS: Sloshing, physical tests, CFD, PIV For the numerical analysis of the instationary flow inside the tank, the
open source and commercial CFD codes OpenFOAM and STARCCM+
INTRODUCTION are used. The RANS Equations are solved based on the Finite Volume
Method (FVM). The pressure and velocity fields are coupled using the
The natural gas market is one of the key factors to satisfy world’s Pressure-Implicit Split Operator (PISO) and SIMPLE algorithms. Tank
energy demands. Thereby, seaborne transportation of liquefied natural motions are prescribed. The moving grid technique is used.
gas (LNG) is a continuously growing market, with an increasing fleet
size. Transportation of LNG in ships’ cargo-holds requires an
assessment of sloshing loads, which is typically performed by means of FACILITY
physical model testing using water and air or water and a heavy gas Motion Platform
mixture. To provide reliable results, accurate and repeatable model tests
are inevitable. Besides scaling laws, other influence factors have to be The sloshing test rig consists of different components. The main
considered during conduction of model tests, such as accuracy of the component is a Stewart platform, also known as hexapod, which allows
tank motion and pressure measurements. The physical phenomena moving a tank in six Degree-of-Freedom (DoF). The maximum motion
related to sloshing are described in Faltinsen and Timokha (2009) amplitudes and the payload are given in Table 1. An overview of the
test rig is shown in Fig. 1.
Within a benchmark test campaign, sloshing tests were performed
aiming at comparison of the results obtained from different test Regular motions can be specified in terms of amplitude, frequency and
facilities, Loysel (2012), (2013). A high effort was made to validate phase for each DoF. Regular and irregular motions can also be defined
tank alignment on the test rig and motions. Further, acquisition of high- by input data containing positions with a sampling rate of 100Hz. Prior
speed videos was included to compare free surface shapes around the path execution; validity checks are performed. Resolvers at each motor
impacting pressure sensor module for single-impact-conditions. The deliver data used to evaluate the position of the platform. The
components connected to the hexapod via trigger wires can be activated

286
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 by means of control commands in the motion path files. Typically, one separate cavities, a minimum pulse separation time of 10 can be
trigger is used to start the motion acquisition, while a second trigger obtained. Adjustment of the separation time is necessary to obtain a
starts pressure data acquisition and high-speed video acquisition. reasonable particle shift for accurate processing. The acquisition
frequency, which is the repetition rate of the double pulses, is 100Hz.
Time critical operations such as sending trigger signals to cameras and
laser are performed by a high-speed controller. During PIV
measurements, cameras are equipped with bandpass filters which let
either the laser wavelength or that of fluorescent particles pass. By that,
measurements become insensitive to other lights, cameras can be
protected and reflections can be reduced.
During processing of the images, a cross correlation is applied on
subsets of the Field of View (FoV) to identify translations of particle
groups. Based on a previous calibration of the camera on the FoV,
particle translations can be transformed from camera pixel into length
scale. Dividing the result by the pulse separation time provides the final
velocity vectors.

TEST CONDITIONS
A quasi 2D tank, made of Plexiglas, is used for the investigations.
Dimensions of the inner tank are shown in Table 2. The tank coordinate
system is located at the center of tank bottom, see Fig. 4.
Figure 1: Laboratory with hexapod and model tank
Two sensor modules are mounted in line with the inner side of the roof
and can be equipped with up to 50 pressure sensors mounted in
Table 1: Hexapod capabilities Table 2: Tank dimensions threaded holes. The sensor array configuration is therefore deviating
Category Value Variable Value from that which was proposed in the benchmark specification. Because
of the size of the sensors fastening nut, the distance between the sensors
Translation (x, y, z) ±0.5m Length 946mm had to be increased. An array of 5x5 sensors can be applied on a
Rotation (x, y, z) ±30.0° Width 118mm module. Figure 2 shows a drawing of possible pressure sensor
Payload 1200kg Height 670mm positions. The distance between the transverse tank walls and the center
Volume 0.0748m3 of the closest sensor column is 6mm. The distance between sensor rows
and columns is 11mm. The center row of the sensor openings is located
at y=0.
Pressure Measurement
Figure 3 shows the configuration of the pressure sensors. Sensor arrays
Measurement data of the pressure sensors is recorded using a
are shown in two tables specifying the number of each sensor opening
56-channel data acquisition (daq) system. All channels are recorded
from 1 to 50. Above the tables, sensor column numbers are specified.
simultaneously with a maximum acquisition frequency of 100kHz. The
Occupied openings are shown in green, while red openings are not in
maximum throughput is limited to 2 million samples per second. A
use and closed with a screw. A number of 35 pressure sensors were
number of 36 piezoresistive sensors of type Kulite XTM-190 is
used for the tests. Array 2 was fully equipped with 25 sensors, since the
available to measure the absolute pressure. The front membrane has a
main impact takes place here. Array 1 was equipped with 10 sensors
diameter of 3.8mm. The sensors have a pressure range of 1.7bar with
located in columns one and two.
an overpressure range up to 3.5bar. The natural frequency of the
sensors is typically at 120kHz. Sensors are temperature compensated
between 15°C and 60°C.
High-Speed Video and Particle-Image-Velocimetry (PIV)
Two high-speed cameras of type Phantom v9.1 are used for video
capturing and PIV. The resolution is 1600x1200 pixels and the internal
ring-memory is 6GB. At full resolution and highest intensity
discretization, up to 1800 images can be stored. Depending on the
requested resolution, the maximum frame rate ranges from 1000 frames
per second (fps) at full resolution (1600x1200 pixels) and 153,846fps at
a reduced resolution of 96x8 pixels. Recording is started either
manually, by motion detection or a trigger signal.
The time resolved stereo PIV system allows to perform measurement of
flow velocities in a wide but thin volume, which is called light sheet.
Measurements can be 2D, accounting for flow velocity components in Figure 2: Arrangement of pressure sensor openings
the light sheet, or 3D, accounting also for the out-of-plane velocity
component. Light from a dual cavity Neodymium-doped Yttrium
Aluminum Garnet (Nd:YAG) laser at its 2nd harmonic of 532nm is
passed through a laser optic that widens the beam using a cylindrical
lens and illuminates seeding particles in the light sheet. Due to the two

287
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 Array 1 Array 2 Table 3: Test conditions
1 2 3 4 5 Columns 6 7 8 9 10
Condition Filling Level Motion Type Period Amplitude
5 10 15 20 25 30 35 40 45 50
4 9 14 19 24 29 34 39 44 49 C16 85% Surge SIW 0.975s 32mm
3 8 13 18 23 28 33 38 43 48 C17 85% Surge SIW 1.100s 32mm
2 7 12 17 22 27 32 37 42 47
1 6 11 16 21 26 31 36 41 46

= Sensor inserted
= No sensor

Figure 3: Pressure sensor arrangement and numbering

The camera used for high-speed video capturing and PIV

measurements is looking in normal direction towards the left
longitudinal tank wall. An area of 200x200mm2 measured from roof
and forward transverse wall close to sensor array 2 is captured. The
center of the image coincides with the center of the FoV. The same
FoV is also used during PIV measurements and employed for
comparison between physical and numerical tests with focus being set
to the centerplane of the tank in order to reduce wall influences. Figure
4 shows a sketch of the tank’s inner volume with coordinate system,
sensor arrays, video FoV and the prism which is used to introduce the
laser light sheet into the tank.

Figure 5: Normalized motion time series for the test conditions C16
and C17

EXPERIMENTAL SETUP
The model tank is fixed on the platform of the hexapod. To ensure an
accurate alignment of the tank on the hexapod, a laser is mounted along
the tank’s longitudinal axis. Then, the focus of the laser on a stationary
wall is tracked during longitudinal tank motion of 1m. The horizontal
distance obtained from the motions was lower than 1mm/m, which
corresponds to an angular deviation of less than 0.06°.
The prescribed tank motions (s. eq. 1) are compared to the measured
hexapod motions. In case of deviations, the motions are corrected
correspondingly. The motions are measured using the hexapod internal
measurement device. One resolver is attached to each motor and
delivers a signal to the motor drive which is sent to the hexapod’s
acquisition rack as an incremental signal. Decoded into actuator lengths
and transferred into motions in the hexapod coordinate system, motions
are acquired. The data acquisition is performed at 100Hz, the same
Figure 4: Model tank with sensor arrays, video FoV and laser prism frequency as the motion description.
For each condition, preliminary tests were repeated five times using
Tank motions are assembled as a product of a regular sine motion
motion description files according to the analytical path description.
with amplitude  and period with a hyperbolic tangent introducing The tank is filled with water to the specified filling level of 85% and
ramps according to eq. 1. Motions take place in 1 DoF for all the necessary equipment for the final tests is mounted. Thus, the weight
conditions. Both conditions, C16 and C17, are based on a translatory remains constant between preliminary and final tests. As mentioned
motion in x-direction (surge) with an amplitude of 32mm. However, above, the accuracy of the motions is analyzed by comparing the
the periods are different. Table 3 shows the parameters for the measured motion time histories with the motion description data. A
investigated test conditions and Fig. 5 a normalized motion time series. time-shift needs to be applied to match peaks and zero crossings of the
time histories. Measured motions are subtracted from the motion
2 description to obtain the deviation from the requested path. The
sin tanh , for 0
(1) deviations were relatively low compared to the motion amplitude Â,
2 2 0.5% for both test conditions. Largest deviations occurred shortly after
sin tanh , for T 2
the maximum peaks in the motion description.
The average path deviation from all five repetitions was then added to
the original motion description, yielding a corrected motion

288
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 description. At the beginning of the motion, additional treatment of the frame rate of 1000Hz. The filling level variation of ±1mm was
time series was performed to prevent motors from changing rotation performed during these tests. Tests with PIV measurements were
direction which would reduce the accuracy. Table 4 shows the carried out with a double pulse frequency of 100Hz and a separation
maximum deviation between motion path description and measurement time of 2000 . The separation time was selected to allow
before and after the correction from five repetitions at each condition. measurement of the flow in the FoV prior to the first impact. To
With the correction loop, deviations were reduced by 71% for C16 and measure high velocities during the impact, a smaller separation time
63% for C17, see Fig. 6. would be necessary. Variation of the filling level was omitted during
PIV tests.
Table 4: Motion path deviation before and after correction
Condition Maximum path deviation [%] NUMERICAL METHOD
Before correction After correction The applied numerical methods solve the conservation equations for
C16 0.5 0.16 mass and momentum in their integral form:
C17 0.5 0.19
Ω · 0 (2)

Ω · Ω · (3)
(4)
Ω · 0

Equation 2 is known as mass conservation equation, 3 as momentum

conservation equation and 4 as space conservation equation. is the
fluid density, is the pressure and is the fluid velocity vector with
, , . Ω denotes the cell volume, the control volume
surface with a unit normal vector pointing outwards. is the
velocity vector of which is not zero in case of a moving control
volume. Effects of viscosity are taken into account with the newtonian
stress tensor

(5)

with being the dynamic viscosity. The turbulence is modeled by

splitting the unknown quantities into a mean value and a fluctuation.
The solution domain is subdivided into a finite number of control
volumes of arbitrary shape. The integrals over control volumes and
their faces are numerically approximated using the midpoint rule. The
Figure 6: C16: Motion path deviation before and after correction mass flux through the cell face is taken from the previous iteration,
following a simple Picard iteration. The free surface dynamics is
The filling level was adjusted by filling the tank to 100% and removing computed employing Eulerian interface-capturing techniques of the
a water mass corresponding to 15% of the volume. To account also for Volume-of-Fluid (VoF) type. The unknown variables at cell face
influences due to the filling level deviations, each condition was run centres are determined by a second order differencing scheme. The
with filling level variations of ±1mm. An opening in the tank roof transport equations for the turbulent kinetic energy and its dissipation
allows ullage and ambient pressure to compensate. rate were discretized with upwind differences. Second order
discretization methods are used to integrate in time. Pressure and
To perform PIV measurements, laser head, mirrors and the light sheet
velocity are coupled by PISO (OpenFOAM) and a variant of the
optic were mounted on rails attached to the hexapod. A Plexiglas
SIMPLE algorithm (STARCCM+), see Ferziger and Peric (2002)
module placed in the longitudinal opening in the transverse rear wall
provided the entrance of the laser light sheet into the tank, see Fig. 4. In both solvers, the two-fluid system is modeled following a two-phase
The light sheet is introduced in the tank center plane heading towards formulation of the governing equations. The spatial distribution of each
the area of interest below the 2nd sensor array. To evaluate flow of the two fluids is obtained by solving an additional transport equation
velocities, the camera was calibrated on the light sheet using a (s. eq. 6) for the volume fraction of one of the fluids (which can be
calibration plate. The polynomial calibration that was applied allows treated either as an incompressible fluid or as a compressible ideal gas):
compensating angular deviations between camera and the plate, but
also image distortions due to the lens and the media change. During
tests, glass hollow-spheres between 50 and 100 diameter were Ω · 0 (6)
used as seeding material.
The corrected path description was used to conduct the final tests. The Liquid and gas are considered as two immiscible components of a
motion acquisition rate is 100Hz. At least 10 final tests were run for single effective fluid, whose properties are assumed to vary according
each condition. Pressure values were simultaneously recorded for all to the volume fraction of each component as follows:
pressure sensors with an acquisition rate of 50kHz. Motion, pressure 1 1 (7)
and high-speed video acquisition or PIV measurements start 0.1
before the tank begins to move and stop with the tank motion. In tests Density is evaluated from the volume fraction of the first fluid and
without PIV measurements, high-speed videos were captured with a the densities of the two fluids and . The dynamic viscosity is

289
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 calculated accordingly using the individual dynamic viscosities and time are evaluated. Results are analyzed using box plots, see Tukey
of the two fluids. In order to avoid numerical smearing of the (1977). The red horizontal line represents the median value of the input
interface, STAR-CCM+ employs a high-resolution interface-capturing data. Further, a box is drawn around the median, extending from the 1st
(HRIC) scheme, Muzaferija and Peric (1998), while OpenFOAM to the 3rd quartile. From that, so called whiskers extend until the lowest
employs the explicit MULES scheme, OpenFOAM (2012) and highest value in the dataset. Whiskers start from quartiles with a
vertical dashed line and end at the extreme values of the dataset, see
COMPUTATIONAL PROCEDURE Fig. 10.

The equations are solved in an inertial coordinate system and the entire Experiments
grid moves with the tank. This approach requires that the grid is
Test case C16
updated after every time step; however, it is in particular attractive
when the motion is highly irregular or not known in advance. The fluid The test case C16 produced an air cavity that encloses the complete
domain was idealized by a basic volume grid comprising about 50,000 sensor array 2. Figure 8 shows images of the high-speed video at four
control volumes for the quasi two-dimensional dicretization, Fig. 7. To time instances with a delay of 6ms. The uppermost image shows the
investigate the discretization errors and their effect on the pressure free surface 6ms before the impact occurs. The curvature is low to the
peaks the basic grid was systematically refined. The medium-size grid left and stronger to the right. Influences of the tank walls are visible as
comprised about 200,000 cells, the fine grid 800,000 cells, see Table 5. free surface irregularities. The 2nd image from the top is taken at first
The time step size was chosen such that the Courant number in the fluid observance of water contact with the tank roof. The air cavity extends
domain is kept below 0.5. The SST turbulence model with wall around 150mm from the transverse wall on the left side of the image.
functions was applied. The momentum equations were discretized Free surface irregularities increase strongly for images three and four,
using a second order scheme. The entire flow field was initialized by as the air cavity vibrates. In the 4th image, bubbles are visible below the
the hydrostatic pressure and zero velocity. The air flow was treated as roof, but the sensor array is still not wetted.
compressible. On the tank walls, no-slip conditions were enforced on
fluid velocities and on the turbulent kinetic energy.

Table 5: Computational grids

Grid No. of cells Edge length
maximum minimum
1 50,000 50.0mm 1.5mm
2 200,000 23.1mm 0.8mm
3 800,000 9.5mm 0.4mm

Figure 7: Coarse grid with 50,000 cells

RESULTS Figure 8: Test case C16: Images from high-speed video around the
The experimental results are first discussed and then compared with impact with recording times 1.601s, 1.607s, 1.613s and1.619s
computations. Because of the camera arrangement, which is pointing
towards the tank in negative y-direction, presented images are taken In Fig. 9, pressure time traces are shown for four sensors located in the
from the left side of the tank. Thus, sensor array 2 and the front center row and columns 1, 2, 3 and 5 of array 2. For each sensor, the
transverse wall are located on the left side of the images. The positive pressure is drawn for 10 test repetitions. Due to the air pocket which
x-axis is heading positive towards the left side. In the Figures that show encloses the sensors, pressure time traces are almost identical between
graphs from multiple pressure sensors, sensor distance from the the different sensors. The maximum pressure of the impacts is around
transverse wall increases from left to right. 17kPa, which agrees well with the test results published by Loysel et al
(2013). Compressibility of the air allows the air pocket to vibrate,
Repeatability of the tests is analyzed for peak pressure and peak time. showing alternating and decaying pressure fluctuations with a
Peak pressure is the largest pressure value measured during the first frequency around 70Hz. The repeatability of the tests is as good as that
impact in array 2 and peak time is the time at which the peak pressure repetitions are hard to identify in the graphs. For each sensor, the
occurs. For each sensor under consideration, peak pressure and peak

290
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 maximum variation of the peak pressure against the average peak
pressure is less than 0.3kPa.

Figure 11: Test case C16: Box plot of the peak time variation

Test case C17

The flow inside the tank in condition C17 produces a very flat angle
between free surface and tank roof at the impact time. Figure 12 shows
images from the high-speed videos of one repetition around the first
roof contact. The time between the images is 4ms. The first image
shows the free surface prior to the impact. Again, irregularities are
Figure 9: C16: Pressure time histories of 10 repetitions for four sensors
visible at the tank walls. First roof contact occurs in the upper left
located in the center-row of array 2.
corner of the 2nd image from the top. Images three and four show that
The time axis in Fig. 9 is time shifted to filter small deviations of the the impact leads to less pronounced air cavities spread over a wide
impact time. This is important to show the actual repeatability of the range of the impacted area. On the very left side, an air pocket is
pressure curves. Zero time in the graphs is set individually for each visible.
repetition, where the first pressure sensor of array 2 exceeds a pressure
of 4kPa. In Fig. 10, the spreading of the peak pressure is shown for 10
repetitions and the four sensors under consideration. There is a small
difference visible in the values, such as values are lower for sensors 48
and 38, but higher for the sensors 43 and 28. Median values of the peak
pressure range between 16.63kPa (sensor 38) to 17.29kPa (sensor 43).
The spreading range of the peak pressure values is the difference
between the upper and lower extents of the whiskers. It is between
3.0% and 3.4% of the median value for this condition. Whiskers are
arranged almost symmetric around the median value, so that the
maximum difference of a peak pressure to the median is around half the
spreading range (1.6% to 1.8%).

Figure 10: Test case C16: Box plot of the peak pressure

Figure 11 shows the variation of the according peak times. The ordinate
of the graph ranges from 5. 10 to 5.10 . The values in Fig. 11
were obtained by subtracting the average peak time of all repetitions Figure 12: C17: Images from high-speed video around the impact with
and all sensors from each value. The graph shows that the intervals of recording times 1.705s, 1.709s, 1.713s and 1.717s
the peak times from each sensor are almost identical, as expected from
Figure 13 shows the pressure traces for the five sensors in the center
the type of impact with its large air cavity. The median values range
row of array 2 for 10 repetitions of the tests. The 2nd row of graphs
from 2 · 10 to 6 · 10 and the maximum variation of the peak shows values for the initial filling level of 85%, while upper and lower
time over all sensors and repetitions is 7 · 10 . Summarizing, tests of row contain values for 85%±1mm. All images show a quick increase of
condition C16 are repeatable in terms of peak pressure, fluctuation the pressure and a pressure drop-down with some fluctuations. It is
frequency and peak time. assumed that the fluctuations originate from the enclosed air cavities

291
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 below the roof. For all images, the general shape of the first pressure
peak is well repeatable. However, peak pressure values vary between
repetitions. For most of the sensors, peak pressure values range from
23kPa to 38kPa. After the first peak, signals from the first and second
column (Sensors 48 and 43) show some spreading between the
repetitions, which may originate from minor changes of the air cavity
volume that was observed around the tank edge. This effect is strongest
for 85%+1mm filling level and reduces towards 85%-1mm. The
pressure drop-down for sensor columns three to five (Sensors 38, 33
and 28) is repeatable. Pressure peak values tend to increase from
column one to five. However, sensor 33 tends to show slightly lower
values than expected. Compared to condition C16, the duration of the
first pressure peak is much shorter.

Figure 14: Test case C17: Box plot of peak pressure for 85% filling
level

Figure 13: Test case C17: Pressure time histories of 10 repetitions at

filling levels 85%+1mm (upper row), 85% (center row) and 85%-1mm Figure 15: Test case C17: Box plot of the peak times for 85% filling
(bottom row) level

Figure 14 shows a box plot of the peak pressure for the initial filling Experiment versus Computations
level using the center sensor row and 10 repetitions of the test. The
median peak pressure varies between 25.4kPa and 27.8kPa for sensors Comparison between experimental and numerical results is performed.
48 to 33, while a value of 61.7kPa is measured by sensor 28. For the First, the free surface representation is compared using images from the
first four sensors, the inner quartile range (IQR), which is the height of high-speed videos and contour plots. For the numerical method, the
the box, is between 14% and 20% of the peak pressure median. The free surface is considered to be located where the volume fraction is
IQR contains 50% of the pressure peaks. For sensor 28, the IQR is 48% 0.5. Images from both disciplines were taken at identical
of the median. The maximum variation of peak pressure from the simulation/experiment time. Second, PIV measured and OpenFOAM
median ranges from 24% to 55%. However, the peak pressure range is computed velocity fields are compared in an area at the tank
relatively similar for the first four sensors (10.5kPa to 12.7kPa). centerplane located parallel to the FoV. The area is discretized by
Analysis of the filling level variations showed that peak pressure 31x31 points. Velocities from measurement and computation are
variation increases for 85%+1mm and reduces for 85%-1mm. mapped on the area and shown as blue and red vectors, see Figs. 17 and
Especially sensor 28 shows a strong influence. 25. The initial discretization of both disciplines is containing much
more data, but reduction is applied for a better visibility. Vector lengths
For this condition, the peak times are related to the average peak time represent the velocity magnitude as | | , while vector
of sensor 48. The according box plot is shown in Fig 15 for the initial orientations represent the flow direction.
filling level. Peak times are within an interval between -1.4 · 10 and
1.3 · 10 . Variations for the first four sensors are again lower than Test case C16
for sensor 28. Sensor 28 is the first sensor touched by the water and the
one with the smallest IQR (2 · 10 ). Peak pressure and peak time Figure 16 shows the free surface for a time instance of condition C16.
box plots for sensors 48 and 43 are impressively similar. This could The medium size grid is used for the analysis. Images were taken 1.61s
originate from the gas pocket that was observed in the high-speed after the start, showing the impact on the roof. It is shown that the
videos and covers the first two sensor columns from the transverse observed and computed free surfaces agree well. However, changes of
wall. the free surface due to wall influences, such as free surface
irregularities and 3D deformations, are not pronounced in the CFD
results. The wall friction and the free surface tension may be the reason
for the local deviation.
Velocity vectors are compared in Fig. 17 for the same instance of time
as the upper image of Fig. 16. The global flow at that time is heading

292
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 upwards, as the free surface is about to approach the roof. The the medium size grid for incompressible and compressible gas phase.
transverse wall on the left side implies that velocity vectors are heading Pressure is rising with a larger slope for the incompressible
almost straight upwards. On the right side, the flow inclines and the computation than for the compressible. The pressure peak is twice
absolute velocity is larger. The average deviation of the velocity higher than for the experiment and the compressible computation.
magnitude is 4.8%. Largest deviations occur at the vertical wall and at
some locations at the free surface. If values at those locations are
filtered, the average deviation of the remaining field reduces to 2.7%.
Figure 18 shows the measured and computed pressure time history. The
computations were performed for a compressible air flow using
STARCCM+. The pressure time traces were aligned to have identical
peak times. Zero time is set where pressure from the experiment
exceeds 4kPa.

Figure 17: Test case C16: Velocity vectors for CFD (red) and PIV
(blue) at 1.53s

Figure 16: Test case C16: Free surface from high-speed video (upper)
at 1.61s, OpenFOAM- (lower left) at 1.61s and STARCCM+-
calculation (lower right) at the impact time

The general behavior of the pressure time history is well predicted by

the RANSE solvers. It can be seen that the computations performed on
the coarse grid underestimate the pressure peak and the frequency of
the pressure fluctuation. The first pressure peak and duration is well
captured by the computations on the medium size grid.
The peak pressure obtained with the coarse grid is 12% to 15% lower
compared to the experiments. For the medium size grid, the deviation
between the measured and computed peak pressures is 1% to 4%. The Figure 18: C16: Comparison of computed and measured pressure time
pressure fluctuation after the first impact is captured by the RANSE histories for sensor 48
solver for compressible flows. However, the pulsation frequency is
about 47Hz for the coarse grid, 49Hz for the medium size one and
70Hz for the experiment. It should be noted that the numerical analysis
was performed for a thin slice of the tank which has a lower volume. A
detailed comparison would require numerical analysis on identical
volumes, so that the air mass enclosed in the cavity is similar to the
experiments.
The influence of the flow compressibility on impact loads was
investigated by el Moctar et al (2006). Figure 19 shows computed
pressure time histories for sensor 48. Computations were performed on

293
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 deviation at 1.54s is reduced to 3.9% for the remaining field. At 1.69s,
the average deviation of the velocity magnitude changes less much to
3.1%. However, at 1.59s and 1.64s, it can be reduced to less than 2%.
Average deviation of the vector orientation between measurement and
computation is below 2° for all time instances.
Figure 22 shows the computed and measured pressure time traces of
sensor 43. It can be seen from the figure that the computed pressure
peak on the medium size grid agrees well with the measurements, while
the pressure peak is underestimated by the computation on the coarse
grid. Pressure fluctuations are generally captured by the computations.
However, the pressure behavior is more regular compared to the
experiment.

Figure 19: C16: Computed pressure time histories for incompressible

and compressible gas phase

Test case C17

The general agreement between the observed and computed free
surface of condition C17 is fair, see Fig. 20. Images were recorded
1.71s after measurement start.

Figure 21: C17: Velocity fields for CFD (red) and PIV (blue) at 4 time
instances at 1.54s, 1.59s, 1.64s and 1.69s (from upper-left to lower-
right)

Figure 20: C17: Free surface from high-speed video (upper) at 1.71s,
OpenFOAM-(lower left) at 1.71s and STARCCM+-calculation (lower
right) at the impact time

In Fig. 21, computed and measured velocity vectors are shown at four
time instances prior the first impact. The upper left image was captured
1.54s after measurement start, the other images with a delay of 50ms
each. Velocity vectors are matching well in direction and magnitude.
The average deviations of the velocity magnitude are 7.1% for 1.54s Figure 22: C17: Computed and measured pressure time traces for
and below 5% for the other time instances. If large deviations at the sensor 43
vertical tank wall and the free surface are again filtered, the average

294
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025
 DISCUSSION AND CONCLUSIONS addressing a multitude of different questions regarding local and global
flows.
Sloshing physical tests and RANSE computations were performed for
two different conditions. The velocity field and the pressure at different The computed and measured velocity field and free surface agree well.
positions were measured and compared to the computed results. The general behavior of the pressure after the impact was well
predicted by the numerical method for compressible flows. However,
The first test case, C16, showed a large air pocket enclosing the whole the frequency of the pressure oscillations after the impact was
sensor array. Peak pressure values were almost identical between the underestimated. Full 3D computations are required to perform a more
sensors. Deviations between test repetitions were less than 2%. detailed analysis of the oscillation frequencies.
Presence of the air cavity induced well repeatable pressure oscillations
with a frequency around 70Hz.
REFERENCES
An impact with low inclination of the free-surface was observed during
condition C17. Computed pressure peaks, obtained for a compressible el Moctar, O., Peric, M., Muzaferija, S. (2006) “Water Entry of a
flow on finer grids, agree well with measured pressure peaks. The Wedge”, Proc. 9th Numerical Towing Tank Symposium, Le Croisic,
deviation is less than 4%. France

The repeatability of the pressure time traces during experiments was Faltinsen, O.M., Timokha, A.N. (2009): Sloshing. Cambridge University
good, especially for condition C16. Variations of the peak pressures Press
with the filling level were shown but do not lead to a change of the Ferziger, J. H., and Peric, M., (2002). Computational Methods for Fluid
global flow behavior. However, largest influences were observed for Dynamics”, 3rd ed. Springer Verlag, Berlin Heidelberg New York.
condition C17, because the low inclination of the impacting wave was
relatively sensitive. Loysel, T, Cholet, S, Gervaise, E, Brosset, L, De Seze, P-E, (2012),
“Results of the First Sloshing Model Test Benchmark”, Proc 22nd Int
PIV results showed a good agreement with flow simulations. The flow Offshore and Polar Eng Conf, Rhodes, ISOPE, Vol 3, pp 398-408.
in the FoV was fully captured before the impact. At impact time, the
velocity is decreasing for a large part of the FoV, but also jets are Loysel, T, Gervaise, E, Moreau, S, Brosset, L, (2013), “Results of the
generated with a high velocity. Further, droplets are produced and air is 2012-2013 Sloshing Model Test Benchmark”, Proc 23rd Int Offshore
entrapped in bubbles. The discrepancy between the low and high and Polar Eng Conf, Anchorage, ISOPE, Vol 3, www.isope.org
velocities requires a selection of the separation time based on the focus Muzaferija, S. and Peric, M. (1998), Computation of Free Surface Flows
of the investigations. However, jets, free surface motion, droplets and Using Interface-Tracking and Interface-Capturing Methods. Proc.
bubbles may hinder the laser-light to illuminate the required portions of Nonlinear Water Wave Interaction, O. Marenholtz and M. Markiewicz,
the FoV. Also, cameras may be damaged by strong reflections as well. Eds., Computational Mechanics Publ., Southampton, Ch. 3.
Moreover, reflections needed to be controlled during experiments to
avoid illuminating to much of the tank volume. Therefore, fluorescent OpenFOAM, (2012), User guide, http://www.openfoam.org/docs/user/
coatings were successfully applied to the tank walls. Fluorescent
STARCCM+, (2013), User Guide 8.04.007. CD-ADAPCO
particles were also tested but sank too quickly, so that a reasonable
delay between subsequent tests could only be obtained using non- Tukey, J, W, (1977), “Exploratory Data Analysis”, Addison-Wesley
fluorescent glass hollow spheres. PIV measurements in the sloshing Publishing Company
context are promising. Flexibility and modularity of the system allow

10

295
Downloaded from http://onepetro.org/ISOPEIOPEC/proceedings-pdf/ISOPE14/All-ISOPE14/ISOPE-I-14-464/1491660/isope-i-14-464.pdf/1 by King Fahd University of Petroleum & Minerals user on 14 July 2025