You are on page 1of 85

CRANFIELD UNIVERSITY

Martin Agar

Simulation of Hydrodynamic Ram during impact of fluid-filled tank


using SPH

SCHOOL OF ENGINEERING
ADVANCED LIGHTWEIGHT STRUCTURES AND IMPACT

MSc Thesis
Academic Year: 2010 - 2011

Supervisor: Rade Vignjevic


September 2011

CRANFIELD UNIVERSITY

SCHOOL OF ENGINEERING
Advanced Lightweight Structures And Impact
MSc Thesis

Academic Year 2010 - 2011

MARTIN AGAR

Simulation of Hydrodynamic Ram during impact of fluid-filled tank


using SPH

Supervisor: Rade Vignjevic


September 2011

This thesis is submitted in partial fulfilment of the requirements for the


degree of MSc

Cranfield University 2011. All rights reserved. No part of this


publication may be reproduced without the written permission of the
copyright owner.

ABSTRACT
Hydrodynamic ram (HRAM) is a phenomenon occurring when a high-kinetic
energy projectile penetrates a fluid-filled tank. During HRAM, when the
projectile comes through the tank, it transfers its kinetic energy to the fluid
creating an extra pressure. This pressure can produce catastrophic effects on
the tank. This is of important issue in the design of wing fuel tanks for aircraft
since it has been recognized as one of the important factors in aircraft
vulnerability. In this thesis, the LSDYNA FE/SPH code has been used to
simulate an HRAM event created by a spherical projectile impacting a water
filled tank. The SPH formulation is employed to modelling the fluid during this.
The simulation is based on an experimental test done previously (consisting in a
massive tank impacted by a spherical projectile at 341m/s). In this experiment,
pressure transducers located at different points of the fluid has record the
pressure evolution for understanding pressure transfer mechanisms during the
HRAM. Those transfer mechanisms are reproduced numerically and compared
with experimental results in order to assess accuracy of the SPH technique in
reproducing such a complex phenomenon. Results show that SPH has the
potential for simulating this event. However some important limitations exist.

Keywords:
Hydrodynamic Ram; Impact; Tank; Fluid-Structure Interaction; SPH; Cavitation;
Super-Cavitation

ACKNOWLEDGEMENTS
I would like first to offer thanks to my supervisor, Pr. Rade Vignjevic for his
guidance and his advises for defining clearly my project and for the autonomy
given during the thesis.
I would like to express my gratitude to Dr. Kevin Hughes, Dr. Tom de Vuyst and
Dr. James Campbell, who always managed to find time for helping me and
providing me precious advice.
I would also like to thanks the Caf Comet and its lovely ladies for the
marvellous chocolate muffins provided.
I would at last like to thank the ALSI students, especially Laura Garnier, for the
great year spent here and all the DOD games played during those months.

ii

TABLE OF CONTENTS
ABSTRACT ......................................................................................................... i
ACKNOWLEDGEMENTS.................................................................................... ii
LIST OF FIGURES ............................................................................................. v
LIST OF TABLES ............................................................................................... vi
1 INTRODUCTION ......................................................................................... 7
1.1 Project background ............................................................................... 7
1.2 Thesis objectives ................................................................................... 8
2 THE HYDRODYNAMIC RAM PHENOMENON........................................... 9
2.1 Description of the phenomenon ............................................................ 9
2.1.1 The shock phase ............................................................................ 9
2.1.2 The drag phase............................................................................. 11
2.1.3 The cavitation phase..................................................................... 12
2.1.4 The exit phase .............................................................................. 14
2.2 Consequences .................................................................................... 14
3 RESEARCH DONE PREVIOUSLY ........................................................... 16
3.1 Thesis-based Experiment.................................................................... 16
3.2 History of the simulations of hydrodynamic ram .................................. 18
3.3 Recent simulations done ..................................................................... 20
3.4 Mitigation System ................................................................................ 22
4 METHODOLOGY BACKGROUND ........................................................... 24
4.1 Lagrangian Method ............................................................................. 24
4.2 Eulerian Method .................................................................................. 25
4.3 ALE Method ........................................................................................ 26
4.4 SPH Method ........................................................................................ 26
4.4.1 Integral interpolants ...................................................................... 27
4.4.2 Kernel function .............................................................................. 29
4.4.3 Equations of motion ...................................................................... 30
5 MODEL DEVELOPMENT ......................................................................... 32
5.1 Model size reduction ........................................................................... 32
5.1.1 Symmetry conditions .................................................................... 33
5.1.2 Ghost particles .............................................................................. 34
5.2 Projectile modelling ............................................................................. 35
5.3 Tank modelling .................................................................................... 35
5.4 Fluid modelling .................................................................................... 40
5.4.1 Particle density sensitivity analysis ............................................... 41
5.4.2 Water equation of state ................................................................. 43
5.4.3 Silent boundaries .......................................................................... 45
5.5 Parts interaction .................................................................................. 46
5.5.1 Projectile/Tank interface ............................................................... 47
5.5.2 Projectile/Fluid interface ............................................................... 49
5.5.3 Tank/Fluid interface ...................................................................... 50
5.6 Pressure initialisation .......................................................................... 52
5.7 Pressure measurement ....................................................................... 53
5.8 Final model.......................................................................................... 54
6 RESULTS.................................................................................................. 57

iii

6.1 Initial problems .................................................................................... 57


6.1.1 Distributed memory computer ....................................................... 57
6.1.2 Double precision solver ................................................................ 58
6.2 Shock phase ....................................................................................... 59
6.3 Drag phase.......................................................................................... 66
6.4 Cavitation phase ................................................................................. 69
6.5 Exit phase ........................................................................................... 73
7 CONCLUSION .......................................................................................... 75
8 FURTHER WORKS................................................................................... 76
REFERENCES ................................................................................................. 77
APPENDICES .................................................................................................. 83

iv

LIST OF FIGURES
Figure 1 - Phases of the HRAM [1]..................................................................... 9
Figure 2 - Pressure transducers [7] .................................................................. 16
Figure 3 - Experimental Arrangement [7] ......................................................... 16
Figure 4 - Pressure History [7].......................................................................... 17
Figure 5 - Cavity collapse steps [7] .................................................................. 18
Figure 6 - Cavity creation in D. Varas et al. [41] ............................................... 21
Figure 7 - Mitigation system for HRAM [10] ...................................................... 23
Figure 8 - Example of a Lagrangian mesh [24]................................................. 24
Figure 9 - Example of an Eulerian mesh [24] ................................................... 25
Figure 10 - Neighbouring particles including by the kernel function [44] .......... 28
Figure 11 - Model dimension justification ......................................................... 33
Figure 12 - Symmetry Conditions ..................................................................... 33
Figure 13 - Ghost particles reflecting real particles [22] ................................... 34
Figure 14 - Projectile Mesh............................................................................... 35
Figure 15 - Two possible tank configurations ................................................... 36
Figure 16 - Different mesh sizes tested ............................................................ 38
Figure 17 - Impact Holes .................................................................................. 39
Figure 18 - Resulting velocity after impact ....................................................... 39
Figure 19 - Boundary conditions of the tank walls ............................................ 40
Figure 20 - Model with 3mm particle density .................................................... 41
Figure 21 - Projectile velocity in function of time for different particle density .. 42
Figure 22 Example of shock front velocity measurement .............................. 44
Figure 23 - Silent boundary creation ................................................................ 46
Figure 24 - Projectile/tank interface study ........................................................ 47
Figure 25 - Contact penetration ........................................................................ 48
Figure 26 - Good contact behaviour ................................................................. 48
Figure 27 - Projectile/fluid interface study ........................................................ 49
Figure 28 - Particle penetration problem .......................................................... 50
Figure 29 - Projectile/fluid interface study ........................................................ 51
Figure 30 - Offset between the tank and the fluid particles .............................. 51
Figure 31 - Pressure initialisation ..................................................................... 53
Figure 32 - Coordinate system ......................................................................... 54
Figure 33 - Small model ................................................................................... 55
Figure 34 - Big model ....................................................................................... 55
Figure 35 - Post-processing problem ............................................................... 58
Figure 36 - Projectile impact with the tank at t=0.03ms .................................... 59
Figure 37 - Von Mises stress generated by the impact on an empty tank at
0.023ms (MPa).......................................................................................... 60
Figure 38 - Pressure wave generated at 0.06ms.............................................. 60
Figure 39 - Von Mises stress generated by the impact on a filled tank at
0.023ms (MPa).......................................................................................... 61
Figure 40 - Drop in pressure measured along the shot line on the big model .. 62
Figure 41 - Initial wave pressure recorded by the transducer P1 in the
experiment ................................................................................................ 63
Figure 42 - Pressure history at P3 .................................................................... 63

Figure 43 Decrease of pressure of the transducer P4, P5 and P7 compared


with the transducer P3 .............................................................................. 64
Figure 44 Angle [11] ................................................................................... 65
Figure 45 - Projectile pressure field .................................................................. 66
Figure 46 - Decrease of the velocity due to drag forces ................................... 67
Figure 47 - Comparison of the drag pressures between the simulation and the
experiment ................................................................................................ 68
Figure 48 - Projectile position relative to the transducer P3 at 1.6ms............... 69
Figure 49 - Cavity formation in the experiment [11] .......................................... 70
Figure 50 - Cavity formation at 1.6ms .............................................................. 71
Figure 51 - Cavity grow [11] ............................................................................. 71
Figure 52 - Cavity collapse pressure in the small model for a transducer located
at 150mm from the impact......................................................................... 72
Figure 53 - Cavity collapse pressure in the experiment.................................... 73
Figure 54 - Deflection of the tank before the impact ......................................... 74
Figure 55 - Pre-stress in the tank wall before impact ....................................... 74

LIST OF TABLES
Table 1 - Johnson-Cook parameters [8] ........................................................... 37
Table 2 - EOS Comparison .............................................................................. 45
Table 3 - EOS Parameters [41] ........................................................................ 45
Table 4 - Pressure transducer position............................................................. 54
Table 5 - Model statistics .................................................................................. 56

vi

1 INTRODUCTION
1.1 Project background
Fuel tank in an aircraft has been identified as one of the important components
in aircraft vulnerability [15; 41]. A failure of the fuel tank can result with
catastrophic consequences for the aircraft (large fuel leakage, loss of the centre
of gravity, fire hazard). This failure can occur due to impacts. During high
velocity impacts with fluid-filled tank, it appears the problem of the
Hydrodynamic Ram (HRAM) phenomenon. It increases the risk of catastrophic
failure and the damages are more serious than impacting an empty tank.
During the HRAM, when the projectile comes through the tank, it transfers its
kinetic energy to the fluid creating an extra pressure. This pressure can produce
catastrophic effects on the tank.
The HRAM phenomenon was before more related to military aircrafts. Those
aircrafts are more exposed to the threats capable to produce the HRAM
(ballistic impact, missile fragment) [6]. But now its a serious issue concerning
civil aviation. The Federal Aviation Administration established the analysis of
the effects of turbine engine fragment impacting fuel tank as one its research
area in 1990 [41]. In 2000, the BEA determined that the HRAM has played an
important role in the dramatic crash of a Concorde during takeoff from Charles
de Gaulle Airport [2].
The HRAM phenomenon is a complex fluid-structure interaction which encloses
physical phenomena like high velocity impact, fluid mechanics, large
deformation and material failure [43]. Its difficult to characterise this
phenomenon analytically so most of the research are experimental results and
numerical simulation. Simulations of HRAM using numerical methods have
been tried for almost 40 years [15]. With the latest numerical methods, good
correlation between the experimental and numerical results starts to appear
[41]. However its still an ongoing area of research because the simulations
require high computational resources and there are still some limitations for
7

modelling failure and cracks [35]. Arbitrary Lagrangian-Eulerian (ALE) and SPH
methods have shown the best results.

1.2 Thesis objectives


Simulating numerically the HRAM phenomenon is not an easy thing to do. A lot
of physical phenomena are involved during this event. Research to simulate it
started 35 years ago, but due to computer limitation, results werent so
accurate. Recently with the development of new methods, some studies have
shown good results in reproducing experimental tests [41; 43].
The objective of the thesis is to assess the capability of the coupled Finite
Element/SPH code implanted in LS-DYNA to represent accurately the HRAM
phenomenon. Recently, Disimile et al [11] did an experimental study of the
hydrodynamic ram impacting a fuel tank with a high velocity. In their paper, a lot
of data is provided about the results obtained. The final aim of the thesis is to be
able to reproduce numerically this experiment and to have a good fitting
between results in term of pressure history and cavity dimension.

2 THE HYDRODYNAMIC RAM PHENOMENON


2.1 Description of the phenomenon
As said before, the HRAM effect corresponds to the action of a projectile
impacting a tank and transferring kinetic energy to the fluid creating an extra
pressure. The HRAM phenomenon is a complex problem, difficult to predict.
Indeed, the pressure in the fluid caused by the projectile creates a displacement
of the wall. In return this displacement modifies the pressure of the fluid in the
tank [3]. This complex interaction is call as the fluid-structure interaction. During
an HRAM, 3 interactions are presents and have to be taken into account: the
interaction between the projectile and the structure, between the projectile and
the fluid and between the fluid and the structure [15]. The HRAM is divided in
four phases: a shock phase, a drag phase, a cavitation phase and an exit phase
as shown in Figure 1.

Figure 1 - Phases of the HRAM [1]

2.1.1 The shock phase


In his paper [7], J.P. Borg et al. show that the catastrophic failure or not of the
tank is highly dependent of the nominal pre-stress. If the difference is very high
between the internal and external pressure, a projectile with a low kinetic
9

energy will be able to cause a catastrophic failure. But for a fuel tank, the
internal pressure is due to the hydrostatic pressure so the nominal pre-stress is
very low before the impact (this is not true for the impact with the exit wall as it
will be explained later) and so the catastrophic failure will not be caused by the
impact with the wall but by the wave created in the fluid.
Indeed, after the projectile goes through the tank, it penetrates the fluid
accelerating the fluid at the impact point. The fluid initially with no velocity has
now the same velocity than the projectile. This sudden acceleration creates a
high pressure peak travelling through a shock wave. This pressure generates a
stress in the entry wall. This stress coupled with the dynamic stress can result
with a catastrophic failure of the tank because the expansion of the shock wave
can produce a petalling of the entrance panel.
The shock wave is hemispherical expending from the impact point and its
velocity is greater than the speed of sound in the fluid [23]. Townsend et al. [39]
tried to characterize the shock wave dynamic, providing an analytical model
using the Hugoniot-Rankine relations. The velocity of the shock front is given
by:

 : Velocity of the shock front

 =  + 

 : Projectile velocity after water impact


 : Sound speed in the fluid

 : Hugoniot slope coefficient of fluid


And the shock pressure P is equal to:

Where
is the density of the fluid.

=
 

10

If the peak of pressure due to shock can be very high, its very short in time
because this pressure will be dissipated very quickly in the fluid as shown
experimentally by Stepka et al. in [38]. So, during the shock phase, only the
entry panel will be affected, the pressure field will be very low close to the exit
and lateral panels.

2.1.2 The drag phase


The next phase in the HRAM is the drag phase. Now the projectile has passed
the impact point and it is moving into the fluid. This displacement created a drag
force exerted by the fluid on the projectile, and so the projectile transfers its
momentum to the fluid. This force created is proportional the square of the
projectile velocity and it is equal to:
=

  
2  

Where
 is the fluid density,  is the drag coefficient of the projectile, A is the

reference area and  is the projectile velocity after water impact.

Stepka et al. decided in [38] to use the Newtons second law to determine the
velocity of the fluid. So:


 ()
1
=
   

2

Where  is the mass of the projectile. With this equation they assumed that the

projectile isnt deformed, the drag coefficient stays constant and the fluid is
incompressible. Then they rearranged terms, integrated and solve this equation.
They end with the projectile velocity decay ratio for a spherical projectile:

1
=
 1 + 3  
   
4
 
This equation allows us to predict the velocity of the projectile during its
displacement into the fluid knowing the initial velocity after impact.

11

The weak created by hydrodynamic friction will be responsible for the cavitation
formation, an important phenomenon of the HRAM.
The pressure field created in front of the projectile can be a source of
catastrophic failure especially in thin fuel tank as it has been show in [37].

2.1.3 The cavitation phase


Cavitation is an important property of the liquid. Liquid are not able to resist to a
very low pressure. For example water is almost uncompressible and its density
doesnt change a lot even with very high pressure. Nevertheless, when the
pressure of the fluid drops under the vapour pressure of water (0.0023 MPa at
20 [14]), some bubbles and cavitations filled with vapour appear in the fluid.
The liquid changes phase to become a gaz. This phenomenon is involved in the
cavitation phase of the hydrodynamic ram.
During the drag phase, the projectile is moving into the fluid. Due to the drag a
difference of pressure exist between the front of the projectile (static pressure)
and the projectile wake. In the projectile wake the pressure drop under the
vapour pressure and so a cavity starts to appear.
The initial shape of this cavity is cylindrical but the cavity expends during the
event to a spherical shape (Shi et al. [36] realised some optical observation of
the cavitation phenomenon). The cavity will reach its maximum radius and then
collapse. In this case, when the cavity dimensions exceed the projectile ones,
some authors prefer to use the term of supercavitation.
Bachelor [5] explains the process in his book and defines a non-dimensional
number to characterise the cavity:
=

 
1
2
 

Where:  is the static pressure and  is the cavity pressure.

12

He says that the smaller the K coefficient is, the bigger the cavity will be. So if
the projectile is travelling at a high velocity, the cavity will be very big. Also,
deeper the projectile is travelling smaller will be the cavity (the static pressure is
increasing with the depth).
According to him, a supercavitation (so a small K parameter) has an influence
on the drag coefficient of the projectile. He found this linear relationship for the
new drag coefficient:
 () =   (1 + )

Where   is the drag coefficient without cavitation.


As said before, if the projectile has a high velocity, the cavity size grows. K.
Reichardt, cited by N. Lecysyn [23] found a relation to calculate the maximal
cavity diameter in function of the cavity parameter K:
 ()
!"# = $

(1 0.132 .' )

Where  is the projectile diameter.

Some authors have try to determinate the radial growth velocity of the cavity. N.
Lecysyn et al. [23] rearranging relations found by M. Held in [18], gives us a
analytical relation of the radial growth velocity of the cavity without knowing the
K parameter. Using the Bernouilli relation, assuming that the radial growth
velocity is proportional to the projectile velocity, we have:


)
= $ 
2,
*
) (1 + +)

=




4 1 + .



 + =

13

3 

4
 

Where ) is the cavity diameter, t is the time after impact, h is the depth at the

shot level axis,


 and
 is the density of the water and the projectile
respectively.

All those analytical models presented dont take into account the fact that the
pressure in the cavity is inferior to the atmospheric pressure. So the air present
outside will be draw in the cavity through the hole created by the impact.
However it can be interesting to assess the results which will be obtained with
the numerical simulation with those equations.
Once the cavity reaches its maximal radius, it collapses. In this case of a
supercavitation, the collapse is particularly violent. P.J. Disimile et al. [11]
describe this as: when the ends of the cavity collapse and meet with the
collapsing interface, a shock wave radiates outwards. This pressure emitted by
the collapse is very high, more the one emit during the shock phase but very
short in time [17]. During the collapse of the cavity, the air inside will be
compress and several oscillations will occur but with a decreasing intensity.

2.1.4 The exit phase


This phase correspond to the exit of the projectile from the tank. This phase can
occur before or after the cavity collapse, it depends of the tank size. This phase
may also not happen if the projectile looses too much kinetic energy due to the
drag force. This phase is different from the shock phase. When the projectile
impact the entry wall in the shock phase, there is no pre-stress due to the fluid,
here the exit wall is pre-stress due to the fluid [15]. Due to the fluid, the high
pressure area on the wall is much larger than the projectile profile and so loads
the structure over a large region.

2.2 Consequences
According to J.P. Borg [7], NASA and a lot of authors agreed on a damage
criterion for fuel tank stated as follows: "The criterion for catastrophic failure of

14

a pressure vessel subjected to hypervelocity impact is met if at least one axial


crack exists that is equal or larger than the vessel's axial length."
One of the most dramatic consequences of the hydrodynamic ram effect is the
catastrophic failure of the fuel tank. Without the hydrodynamic ram effect, a
small projectile impacting a fuel tank would have created a small hole resulting
with a small leakage. But due to the HRAM, even a small projectile can create a
very important crack. This crack creates then an important leakage and the fuel
flowing out of the tank can arrive on an ignition surface [33]. Moreover the loss
of fuel can also create a loss of centre of gravity.
This catastrophic failure can occurs during any phases of the HRAM. I can be
due to an impact with the entry or exit panel or during the cavity collapse.

15

3 RESEARCH DONE PREVIOUSLY


3.1 Thesis-based Experiment
The simulations which will be carried out during the thesis are based on an
experimental study made by Disimile et al. [11]. This experiment had for goal to
show and give a clear explanation of the pressure transfer mechanism during
the HRAM.
For this experiment, they created a generic box representing a fuel tank into
which spherical projectile was fired using a gas gun. The tank was able to
contain 3785.4 L of water with dimensions of 1.168m from the front wall to the
back wall and 1.829m from side to side and in height. Those dimensions are not
representative of a real fuel tank but they are made in order to observe more
easily the HRAM. The walls are made with 9.53 mm thick ASTM A-36 steel
plates and 1.587 mm thick 2024-T3 aluminium at the impact location. The
experimental arrangement is showed in Figure 3. Seven pressure transducers
were implanted inside the tank to measure the pressure during the impact. The
location of those transducers is shown in
Figure 2.

Figure 3 - Experimental Arrangement [7]

16

Figure 2 - Pressure transducers [7]

In the study, three projectiles were fired through the tank, one in steel, one in
aluminium and one in tungsten. Regarding the data provided in the paper, I will
reproduce the one realized using a tungsten projectile. It was fired at a velocity
of 341 17 m/s. The data provided are the pressure history as shown in Figure
4 and also the cavity collapse video as shown in Figure 5.

Figure 4 - Pressure History [7]

In Figure 4, we can clearly show the different phases of the hydrodynamic ram
describe in a previous part. Here the cavity collapse represents the highest
pressure in the tank but only during a very short time. The aim of the numerical
model is to be able to reproduce those results.

17

Figure 5 - Cavity collapse steps [7]

3.2 History of the simulations of hydrodynamic ram


The hydrodynamic ram has been studied for a long time. Since computational
method appeared some researchers have try to represent it numerically. In [15],
C.J. Freitas give us a good review of the different computational simulation of
HRAM used in the past:
The first HRAM simulation was made by Ball in 1972 [4]. The idea was to
incorporate the piston theory method into a structural analysis code. But this
code had some limits since he didnt simulate the plastic response. Then
another code was developed based also on the Balls idea, the BR-1. It allowed
the plastic deformation but when results were compared with experimental data,
it didnt fit at all. Strain and deflection was greatly inferior to experimental data.
So the code has been abandoned.

18

Then the variable image methodology was developed in 1977 by Lundstrom


[29]. In this method the pressure field is determined by a potential function
where the effect of the projectile and the cavitations are described by a series of
sources placed along the trajectory of the projectile. But C.J. Freitas states that
this method wasnt a success because it can only take into account limited
types of geometries, linear and 2D problems.
With the fail of the variable image methodology, some other codes have been
developed. There was the Unimodal Hydraulic Ram Structural Response
(UHRSR) program, developed in 1977. Then this code was coupled with the
BR-1 to give the Hydraulic Ram Structural Response (HRSR) code. In addition
of theses codes, a derivative of the UHRSR was also developed, the ERAM. It
was the first one to take into account the cavitation in the fluid [15].
All those code werent very good for representing the fluid-structure coupling
[15]. So, even if the HRAM phenomenon was well represented, the codes
werent able to predict accurately the damage on the structure due to the HR
event. Moreover the geometry had to be very simple. Its with the development
of the finite element methods in the 80s that good simulation results started to
appear.
In 1980, the EPIC-2 code (a Lagrangian finite element method) was used by
Kimsey [20]. He tried to simulate the impact of a steel rod with a cylindrical fuel
tank. The results were qualitatively quite good but no experimental data was
available for assessing them quantitatively. The use of a Lagrangian method
was suitable for this case because no big distortions occurred during the
simulation. But this method is not the best for impact simulation.
Recently, the ALE and SPH methods shown that they can produce good results
for impact problem and fluid-structure interaction. Thats why, nowadays, the
community is always using those method for simulating the impact on a fuel
tank.

19

3.3 Recent simulations done


As said before, some simulations were done recently using ALE of SPH
methods. Probably, the first one to use the SPH method for this problem was
P.W Randles et al. in 1998 [34]. He produced some interesting results but
without experimental data, it was difficult to determine the accuracy of the
method.
However in 2002 and 2008, two simulations made were quite interesting,
because the quality of the results was evaluated using experimental data. The
first one was made by R. Vignjevic et al. in 2002 [43]. The purpose of this
experiment was to the show that the SPH method was suitable for impact on a
fuel tank. The tank and the fluid were made using SPH particles. Only the
pressure in the fluid was measured during the simulation but the results were
similar with the equivalent experiment. So this paper was a good start for
showing that the SPH can be a useful tool for HRAM impacts.
Then a more complete study was made in 2008 by D. Varas et al [41]. They
impacted liquid filled aluminium tube with a steel spherical sphere. They
simulated it with two different formulations (SPH and ALE).Then experiments
tests were made to study the pressure at different points of the fluid, also for
studying the displacement of the wall and the cavity development. Then those
results were compared to the numerical results. The purpose of this study was
to assess the validity of ALE and SPH method for reproducing complex HRAM
phenomenon.
The modelling of the box and the projectile was done using a Lagrangian FE
model. Only the fluid was made with a SPH or ALE formulation. In the results it
appears that the deformation of the tube was good represented by the two
methods. The pressure was well predicted only for the shock phase, for the
cavity collapse phase the data doesnt appear. For the cavity creation, it seems
to have good results as shown in Figure 6.

20

Figure 6 - Cavity creation in D. Varas et al. [41]


When they compared the two methods, the results are slightly better with the
SPH method. But this method is much more time consuming than the ALE
method. In conclusion, it appears that the two methods seem suitable for
representing the HRAM even if the cavity collapse is not yet perfectly
represented.
In 2011, they published a similar study using partially fluid-filled tube [40].
More recently, two papers were published about the impact of composite fuel
tank. In the first one [9], the authors simulated the impact on a composite and
steel fuel tank and compared the results. It appears that the composite tank is
easier to destroy than the steel tank. In [42], D. Varas et al. studied again the
HRAM but this time in composite tube. The study was only experimental.
A Master thesis was done in Cranfield University by J. Foll in 2009 about this
subject [13]. He used the Cranfield Universitys coupled FE-SPH code to
simulate the impact. In his study the impacted wall wasnt represented. The

21

results obtained were qualitatively good but it was difficult to assess them
quantitatively.

3.4 Mitigation System


As the HRAM can have some catastrophic consequences for an airplane, some
systems were developed to mitigate its effect. Moussa et al. presented in [33],
two possible solutions:

The first one is a solution already used on military helicopter. I consist in


the use of foam against the structural panels of the tank. Using this, the
HRAM effect will be reduced by foam compression. However, the
authors stated this protection is not enough. Nevertheless with recent
research in foam, this technology can be better improved.

The second solution is the Nitrogen-Inerted Bladder developed by


Boeing. This system consists in porous bladders disposed along the tank
walls. In case of impact, nitrogen is pumped, inflating the bladder.
Because the bladder is porous, nitrogen can ooze out and fill the ullage.
This system is efficient against the HRAM since the bladder operates as
a cushion during the impact. Its also a good way for preventing from fire
or explosion since the nitrogen inerts the ullage.

Recently, an article has been published by Disimile et al. [10]. Its about the
mitigation of the shock waves during an HRAM event. They tried to mitigate the
pressure in the fuel tank by installing triangular bars inside as shown in Figure
7.

22

Figure 7 - Mitigation system for HRAM [10]

The role of those bars is to reduce the HRAM pressure by reducing the severity
of the wave front. The idea is when the wave arrives on a triangular face, a
weak reflection of the shock wave should occurs. This system cannot mitigate
the shock wave on the front wall during the shock phase. However the results
show a great reduction of the pressure due to cavity collapse. The initial
pressure wave measured in the back wall is reduced by 60% compared with the
pressure obtain in tank without mitigation member. For the cavity collapse
pressure, the reduction is between 25% and 75%. Now the disposition and the
shape of the bars have to be optimized.

23

4 METHODOLOGY BACKGROUND
For representing the fluid different methods are available. In this chapter, the
most-used methods are briefly presented. As we are using SPH during the
simulation for representing the fluid, this section is more focused on this
method.

4.1 Lagrangian Method


For solving mechanical problem, this method is most popular. The major
characteristic of this method is that the mesh is fixed to the material during the
simulation. As the mesh is fixed, a large deformation of the material creates a
large distortion of the mesh.

Figure 8 - Example of a Lagrangian mesh [24]


This is why this method has some limitations for large deformation. Indeed a
large distortion of the elements results with an inaccuracy in the results.
Moreover this distortion also decreases the time step resulting with an increase
of the computational time.

24

However this method has some advantages. First, as the mesh stay fixed to the
material, its easy to track the material history for calculating the constitutive
equations. Furthermore, treating the boundaries is easy as the mesh edges stay
attached to the material surface.

4.2 Eulerian Method


This method is the opposite of the previous one; the mesh is relative to the
spacial domain. The mesh is fixed in space and the material can flow through
the mesh.

Figure 9 - Example of an Eulerian mesh [24]


The advantage of this method is the mesh stays fixed and doesnt undergo
distortion. So this method is suitable for large deformation problems like fluid.
However this formulation has some disadvantages: as the material move
through the grid, the model required a large mesh representing the area where
the material is likely to flow. The treatment of the boundaries is more complex
as they are depending of the material and not of the mesh. Moreover the results
are difficult to extract as the data history is defined in spacial coordinate and not
in material.

25

4.3 ALE Method


Its a hybrid formulation between the Lagrangian and the Eulerian. This method
tries to combine the advantages of each method. It starts as the lagrangian
method but during the calculation the mesh is allowed to move in the material.
There is distortion criterion for controlling the mesh and restoring a good
element shape if necessary. Moreover with this method the material is also
allowed to flow through the mesh.
This method is quite efficient for fluid problem and could have been also used
for modelling Hydrodynamic Ram [24; 41].

4.4 SPH Method


In this section, we briefly describe the SPH method, explaining the theory used
in LS-DYNA. This description is based on two reviews explaining in more detail
this method [19; 44]. Smooth particles Hydrodynamics was initially created for
simulating astrophysical gas dynamics problems. It has been developed by
Monaghan and Gingold [16] and Lucy [28]. At this time, the available
computational mesh methods were not suitable for this kind of problem. The
SPH method looks like a Lagrangian method but the points are never
connected together creating a mesh. SPH is a mesh-free method. This means it
can undergo large deformation. The particularity of this method is the fluid
dynamic variables are calculated with an integral interpolant using a smoothing
function and the integral is then approximated by a summation over the
interpolation points.
So the SPH was initially made for astrophysical problem but due to all the
advantages that represent the absence of a mesh, this formulation has been
adapted for many others applications like including high velocity impact and
fluid-structure interaction. Thats why this method is particularly suited for
simulating the HRAM.

26

4.4.1 Integral interpolants


The base of the method is to express each variable as an integral interpolant.
Any function A(r) can be expressed in the form:
(/) = 0 (/ 1 )2(/ / 1 )3/

Where 2 is the Dirac function. This form is exact but not very useful. The idea of

the SPH is to replace the Dirac function by a continuous function. This function
is called the kernel as the integral interpolant of any function is expressed as:
(/) = 0 (/ 1 )7(/ / 1 , )3/

The choice of this function is very important for the success of the method. This
function W has also to respect those two properties:
0 7(/ / 1 , ) 3/ 1 = 1

1
1
lim
7(/ / , ) = 2(/ / )

<

The variable h is half-width of the kernel function and it represents the spacial
area where W smoothes A. This variable is often called the smoothing length.
The integral interpolant needs to be discretised for numerical simulations. In
SPH problem, the fluid is defined by a set of points /> distributed regularly

through the fluid. The fluid has a density defined by the variable
(/). So the

fluid is discretised into several point each of those points having a mass m> . So

(/> )3 = m> and the integral interpolant can be expressed as:


(/) = 0

(/ 1 )
7(/ / 1 , )
(/) 3/

(/)

27

(/) @ m>
>CD

>
7(/ /A , )

>

Where > ,
> is the respectively the value of A and the density at /A . This

equation is the basis of SPH method. With this equation the value of A at a
point /A can be express as the summing of the contributions from the
neighbouring point including in the kernel function:
B

> @ mE
ECD

E
7F/A /G , H

Figure 10 - Neighbouring particles including by the kernel function [44]


The major advantage of using such an integral interpolant is for calculating the
expression for the gradient of a function. Indeed (/) can be interpoled as:
(/) = 0 (/ 1 )7(/ / 1 , )3/

By integrating by part, we have:


(/) = 0 (/ 1 )7(/ / 1 , )3/ 0 ((/ 1 )7(/ / 1 , ))3/
Using the gradient theorem:
(/) = 0 (/ 1 )7(/ / 1 , )3/ 0F(/ 1 )7(/ / 1 , )HJ/ 
28

The second term of the right side on the equation is only over the boundary of
the domain. So, most of the time, this integral is neglected because the kernel
function is equal to zero around the domain boundary. Therefore if this term is
equal to zero, we have:
(/) = 0 (/ 1 )7(/ / 1 , )3/
B

(/) @ m>
>CD

>
7(/ /A , )

>

The last equation demonstrate one of the most important advantages of the
SPH: the gradient of any function can be determined by differentiating the
kernel.

4.4.2 Kernel function


As said before the choice of the kernel function is very important. Originally, the
kernel functions used were Gaussian functions or bell-shaped function. But, by
far, the most popular kernel equation currently in use is the one developed by
Monaghan and Lattanzio [32]. Its a function based on cubic spline functions:
3
3
Q1 R  + R S TU 0 |R| < 1
2
4
L O
Z
7(q, ) = M 1
P (2 R)S TU1 < |R| < 2
O4
0 TU |R| > 2
N

Where [ is the number of dimension of the problem and L a constant equal to S,


D

\]

and ] for respectively 1D, 2D, 3D problems.

The advantage of this function is for r > 2h the value are strictly equal to zero
(the algorithm has only to summarize the particles include in a radius of 2h from
the point of interest. Also it has a continuous second derivative. This function is
the one used in LS-DYNA [26].

29

4.4.3 Equations of motion


The equation interpolant can be written for any equation so it can be written for
the conservation equations. Here if given the form of the momentum equation
written in a lagrangian framework:
^
1
= P
dt

So if the equation of state of the fluid relates the pressure P with only the
density
, we can express P in the SPH form as:
B

P = @ mc
cCD

c
7(/ /d , )

So the momentum equation can be written as:


B

^
1
c
= @ mc
7
dt
e

c e fc
cCD

Where 7fc = 7(/e /d , )

And we can also express the density at any point as:


B

e = @ mc 7fc
cCD

The form presented above was the one use used in the first version of the
method. However this form had a problem. It didnt conserve the linear and
angular momentum: the force on particle a due to a particle b wasnt equal to
the force on particle b due to a particle a.
For making this forces symmetrical,

g
h

has been written using the chain rule:


P
P P
P
P
P
  =

=   + 

30

We can then rewrite the momentum equation in the SPH form:


B

^
c Pe
= @ mc (  +  )e 7fc
dt

c e
cCD

With this form, the equation converses the momentum.


Then two important modifications have been implanted to this method
especially important for fluid simulation. First, instead of having a particles
motion described normally by:
/e
= ^e
dt
The motion is described by:
B

Where ed =

hj khl


/e
1
^de
= ^e + @ mc ( )7fc
dt
2
ed
cCD

and ^de = ^d ^e . This expression ensures there is no big

difference of velocity between particles in a same neighbourhood. Its important


for treating nearly incompressible fluid as water due to the absence of viscosity.
This method is called the XSPH method.
The second modification is the way how is calculated the density. For a lot of
application this way is admissible. But, for fluid problem with a free surface (as
in our case), using this method results with problem of discontinuity at the
surface.
For overcoming this problem, the initial density is set at the beginning. And
during the calculation, it only recalculate when particles move relative to each
other. This can be calculated using the continuity equation in the SPH form:
B

e
= @ mc ^de e 7fc
dt
cCD

31

5 MODEL DEVELOPMENT
In this chapter, the methodology used for modelling the experiment is explained.
The model is developed using Hypermesh and LS-PrePost and the simulation
are launched using LS-DYNA.
The unit system used was the following: tonne, mm, s, N, MPa.

5.1 Model size reduction


The first thing to do was to determinate the domain which will be represented
during the simulation in order to start the modelling of the different part. In their
experiment [11] Disimile et al. used a 1168x1829x1829mm tank. For an
equivalent projectile size Varas et al. [41] used a 2.5mm particle density. So
with this density, representing the entire tank would have resulted with a model
of 250 million particles. Such a model would have been too big to compute even
with distributed memory computer like Astral. A reduction of the model size was
needed. For doing that, two actions were performed:

A reduction of the size of the tank

A representation of only a quarter of the experiment using


symmetry conditions

For the reduction in the size of the tank, two models have been made:

A small one: a 400x200x200mm tank is modelled. The dimensions have


been chosen for being able to launch the simulation on a simple
computer and to acquire the pressure data from the first transducer.

A bigger one: a 1168x500x500mm tank is modelled. The same distance


has been kept from the original experiment in the lengthways in order to
recreate all the pressure transducer. For the transversal dimensions, I
decided to keep 3 times the space need for the cavity formations as
shown in Figure 11.

32

500mm

Figure 11 - Model dimension justification

5.1.1 Symmetry conditions


For representing only a quarter of
o the tank, its necessary to create two
symmetry planes. This is made for the finite element parts (the projectile and
the tank), by creating nodal constrains on the boundaries.
For the symmetry plane X-Y,
Y, the nodes located on this plane are constrained in
the Z translation and in the X and Y rotation. For the symmetry plane X-Z,
X
the
nodess located on this plane are constrained in the Y translation and in the X
and Y rotation.

TY, RX, RZ
TZ, RX, RY

constrained

constrained

Figure 12 - Symmetry Conditions

33

5.1.2 Ghost particles


The way the symmetry is done with the finite elements cant be done with SPH
for two reasons: the first reason is that SPH is a mesh-free method and
particles can mix together. If a layer of particle is constrained in displacement,
there isnt any reason for the neighbouring particles not to pass through these
particles and penetrate the boundary. The second reason is that for a particle
which is closed to the boundary, that particle will miss some particles to have
equilibrium of pressure. This will lead to a decreasing pressure near that
boundary. Thats why a new method for symmetry with SPH has been
developed.
The way its done in LS-DYNA is by creating a set of ghost particles which is an
image of the particles close (within a distance of 2h) to the boundary. For each
particle close to the boundary, a ghost particle is automatically created by
reflecting the particle itself. The ghost particle has the same mass, pressure,
and absolute velocity than the real particle. The ghost particle is then in the list
of neighbours of that particle and contributes to the particle approximation.

Figure 13 - Ghost particles reflecting real particles [22]

34

In

LS-DYNA,
DYNA,

the

ghost

particles

are

created

using

the

card

*BOUNDARY_SPH_SYMMETRY_PLANE
OUNDARY_SPH_SYMMETRY_PLANE [25] where only the normal of the
symmetry plane has to be specified.

5.2 Projectile modelling


Since during the event large deformation doesnt appear on the projectile, it has
been modelled using Lagrangian finite element. 3D elements have been
be used
for representing the projectile. In Figure 14are
are showed the projectile mesh
where only one quarter is represented.

Figure 14 - Projectile Mesh


It appears in the literature [11; 41], that the projectile is not really deformed
during the impact. So it has been decided to make the projectile as a rigid body
part using a rigid material model (MAT_020 in LS-DYNA).
LS DYNA). This will
wi reduce the
computational time of the model.
An initial velocity is applied to all the node of the projectile. The initial velocity is
set to 341 000mm/s in the X-axis.
X

5.3 Tank modelling


Since the tank dimensions are reduced, only the entry and exit wall are
ar
represented.

35

For representing a plate, two choices are possible. It could be modelled using
solid element or shell element. Solid element would have been better for
representing the damage as the petalling of the tank but would have resulted
with more computational time. Since we are more focused on the HRAM event
than on the tank damage, shell elements are still good for modelling the tank.
The only important think is the resulting velocity of the projectile after the
impact. This one has to fit with the experiment for reproducing it properly.
The tank walls are made with steel and aluminium as shown in Figure 15.

Experiment configuration

Only Aluminium configuration

Aluminium
Steel
Figure 15 - Two possible tank configurations

Another possible configuration was a tank only made by aluminium. A study


between those two configurations has shown they give the same resulting
velocity. So the fully aluminium configuration has been chosen as the aluminium
has a bigger time step due to its properties. However the different thicknesses
of the wall have been respected.
During the impact, high strain-rates are involved so It has been decided to use a
Johnson-Cook material model (MAT_015 in LS-DYNA) [25]. The Johnson-Cook
36

model is empirically based. It allows us to determine the flow-stress y in


function of the plastic strain p , the normalised strain rate & * and the
temperature T. This model is widely used because is quite accurate, simple to
use and requires a small number of constants. Those constants can be so
easily found in the literature.

y = A + B pn 1 + C ln&* 1 T *m
With T * defined as T

= (T 298 ) / (Tmelt 298) . A B C n and m are constant

which have to be determined through experiments.


The material model used in LS-DYNA has also a damage model where the
strain at failure is given as:
n = opD + p qr spS

L<
uv (1 + pw ln ny )(1 + p' { )
Lt

Where h is the hydrostatic stress and eff the effective stress and D1, D2, D3,
D4 and D5 are material constants.
All the material constants have been found in an article [8] and are summarized
in the following table:
Table 1 - Johnson-Cook parameters [8]
Name

LS-DYNA Flag

Value

Unit

Density

RO

2.77e-9

t/mm3

Young Modulus

73080

MPa

Poisson Ratio

PR

0.33

A parameter

265

MPa

B parameter

426

MPa

n parameter

0.34

C parameter

0.015

m parameter

37

Melt temperature

TM

775

Reference temperature

TR

294

Effective strain rate

EPSO

s-1

Specific Heat

CP

8.75e8

N-mm/t-K

D1 parameter

D1

0.13

D2 parameter

D2

0.13

D3 parameter

D3

-1.5

D4 parameter

D4

0.011

D5 parameter

D5

The last thing to determine was the size of the tank mesh. A quite refined mesh
was wanted in order to predict quite accurately the decrease of velocity due to
the impact. But the problem was a very refined mesh leads to a too big time
step reduction. So the idea was to have a mesh size small enough with a time
step close to the SPH particles time step.
Different sizes of mesh have been tested during an impact with the projectile:

0.625x0.625mm mesh

1.25x1.25mm mesh

2.5x2.5mm mesh

5x5mm mesh

Figure 16 - Different mesh sizes tested

38

Those three meshes have impacted at 341m/s by the projectile. The resulting
velocities have been compared.

0.625x0.625mm

1.25x1.25mm mesh

2.5x2.5mm mesh

5x5mm mesh

Figure 17 - Impact Holes


350
300

Velocity (m/s)

250
200
150
100
50
0
5x5mm mesh

2.5x2.5mm mesh 1.25x1.25mm mesh0.625x0.625mm mesh

Figure 18 - Resulting velocity after impact


39

As we can see there, with a coarse mesh there is a big decrease in velocity and
the impact hole is too big. However with a mesh of 1.25x1.25mm the
convergence of the resulting velocity starts to appear (as we can see in Figure
18) and the shape of the impact hole seems good (Figure 17). Moreover with
this mesh size, the time step remains acceptable. So finally, the tank has been
mesh with fully-integrated shell element of 1.25x1.25mm size.

Fully
constrained

Symmetry
Conditions

Figure 19 - Boundary conditions of the tank walls

5.4 Fluid modelling


The last part to create was the fluid. This is the critical part of the simulation
because the HRAM occurs in the fluid. The fluid has been made using SPH.
In LS-DYNA, for modelling water with SPH, you need to use the material card
*MAT_NULL (*MAT_009). This card permits to consider an equation of state
and only computes the hydrostatic stress which is suited because there is no
deviatoric stress in water. So this card had to be use with an equation of state.
In the modelling process, one of the first things to do was to determine the
particle density of the fluid.

40

5.4.1 Particle density sensitivity analysis


The particle density is the spacing between each particle. It can be compare
with the mesh size for finite elements. So the particle density has an important
influence on the quality of the element and has to be chosen carefully. Thats
why a sensitivity analysis has been performed for this parameter.
A simple box of water (dimensions: 150x90x90mm) has been realised. The fluid
is impacted by the spherical projectile as shown in Figure 20.

Figure 20 - Model with 3mm particle density


Different density has been studied and what is assessed for this study is the
velocity decrease due to the drag forces acting on the sphere. For assessing
that the theory presented in page 11 was used:
 =

  
3
1 + 4  


 

It has been found in the literature [12] that the drag coefficient (CD) of a sphere
is equal to 0.47.

41

So the theoretical velocity of the projectile in function of the time has been
compared to the results obtained from the simulations. This comparison is
presented in Figure 21.

400000
350000

Velocity mm/s

300000
250000
Theoretical
200000

1.5mm
3mm

150000

4.5mm
6mm

100000
50000
0
0

0,0001

0,0002

0,0003

0,0004

0,0005

0,0006

Time (s)

Figure 21 - Projectile velocity in function of time for different particle


density
As we can see, for a particle density of 1.5mm, the velocity curve fits perfectly
with the theoretical values. The problem is with such a density, the model would
have been too big to compute due to a high number of particles. So its been
decided to use 1.5mm as particle density to reduce the computational time.
With this value, the error in the velocity decrease is still acceptable (7%).

42

5.4.2 Water equation of state


As said before the material null needs an Equation of State (EOS). In physics
an EOS is a relation between state variables. Here in our problem, the EOS
used relates the pressure to the density.
For modelling fluid, two equations of state are mainly used: the linear
polynomial EOS and the Gruneisen EOS.
The linear polynomial EOS is stated as presented below:
 =  + D | +  |  + S | S + (w + ' | + } |  )~

Where | =  1 , E is the internal energy per initial volume and C0, C1, C2, C3,


C4, C5, C6 are EOS coefficients.


The Gruneisen EOS is stated as presented below:

| |
2
2
=
+ ( + |)~

|
|S
1 (D 1)|  | + 1 S

(| + 1)

   | 1 + 1

Where C, S1, S2, S3, 0, are EOS coefficients.


Whatever the type of EOS used, accurate parameters are needed for getting
the right behaviour of the water.
Available EOS in the literature have been searched and 3 EOS have been
found from articles [1; 13; 41]: 2 Gruneisen EOS and 1 Linear Polynomial EOS.
Those 3 EOS have been tested using a simple model equivalent to the one
shown in Figure 20. For determining which one is the best for modelling water
during an HRAM, it has been decided to assess them through 2 criteria:

The same criterion as used before, the decrease of velocity due to


the drag forces. The velocity of the projectile is measured at 1.7ms
and compared with the theoretical velocity at the same time.

43

The shock front velocity. The velocity of the wave generated due to
the impact is measured and compared with the theory. The theory
used is the one explained in page 9. The theoretical velocity is
calculated using the Hugoniot-Rankine relations:
 =  + 
Where the Hugoniot slope coefficient of water has been
choose at a value of 1.79 [45].

For measuring the velocity of the shock front, the time where the wave peak
appears has been measured at two different points located at 10mm and 20mm
away from the wall as shown in Figure 22. The difference between those two
times divided by the distance between the two points gives us the shock front
velocity.

1,20E+02

Pressure (MPa)

1,00E+02

8,00E+01

10mm

6,00E+01

20mm

4,00E+01

2,00E+01

0,00E+00
0,00E+00

2,00E-05

4,00E-05

6,00E-05

8,00E-05

1,00E-04

Time (s)

Figure 22 Example of shock front velocity measurement

The results obtained with the different EOS are presented in the table below:
44

Table 2 - EOS Comparison

EOS type:

Projectile Vel. at 1.7ms (error)

Front wave velocity (error)

Gruneisen 1

140 m/s (8%)

1568m/s (31%)

Gruneisen 2

140.5 m/s (7.5%)

1612m/s (27%)

Linear Polynomial

141 m/s (7%)

1851m/s (11%)

Theory

151 m/s

2061m/s

As we can see, the values obtained for the first criterion are approximately the
same. So the second criterion has been used for determining the good EOS.
However, according Korobkin [21] the initial impact wave is moving at a
supersonic velocity only for a short duration. After, the wave returns to the
acoustic velocity. As it was quite difficult to evaluate the right theoretical
velocity, it has been decide to the Gruneisen 1 EOS from [41] because it the
only one which has been used before for the representing a HRAM.
Table 3 - EOS Parameters [41]
Parameters name:

Value

Units

1.448e6

mm/s

S1

1.979

S2

S3

0.11

3.0

5.4.3 Silent boundaries


As we are not representing the entire model, its important to create silent
boundaries. Indeed by modelling only a small portion of the experimental tank,
we are creating free surfaces which shouldnt be there. Those free surfaces

45

reflect the pressure waves contaminating the results. In the experiment,


because the lateral walls are very far from the shot line, there is no wave
reflection. The waves are dissipated before. So thats why its important to add
silent boundaries on the lateral side of the fluid, the wave wont be reflected.
In the current version of LS-DYNA, silent boundary for SPH has not been
implanted yet. Its currently available only for solid element. So the free surface
has been represented using 3D finite elements and SPH and solid elements are
linked together.

Non reflective
boundaries
Tied contact
Solid elements

SPH

Figure 23 - Silent boundary creation


SPH particle forming the free surface are connected to solid elements using the
card *CONTACT_TIED_NODES_TO_SURFACE. This card permits to link to 2
parts together. So in this type of contact, the slave nodes are constrained to
move with the master surface [27]. Then a non reflective boundary is applied on
the

free

surface

made

with

solids

elements

using

the

card

*BOUNDARY_NON_REFLECTING.

5.5 Parts interaction


Now each part is created, the interfaces between each part had to be created.

46

5.5.1 Projectile/Tank interface


First the interaction between the projectile and tank has been studied. This is a
contact between two finite element parts. There are a lot of contact algorithms
available in LS-DYNA for modelling this type of contact.
A simple model has been made (Figure 24) and several contacts have been
tested for finding the best contact algorithm.

Figure 24 - Projectile/tank interface study


The first algorithms which have been tested were the ones defined by the cards:

*CONTACT_AUTOMATIC_SURFACE_TO_SURFACE

*CONTACT_SURFACE_TO_SURFACE

Those contacts are two-way treatment contact where a master and slave
surface is defined and LS-DYNA checks the slave nodes for penetration
through master segments and then a second time the master nodes for
penetration through slave segments [27].
However whatever the penalty scale factor used, some penetrations appear
during the contact as shown in Figure 25.

47

Figure 25 - Contact penetration


This can be explained by a time step too important compared with the projectile
velocity and also by the presence of sharp edges due to the projectile geometry.
A reduction of the time step can solve this but its not acceptable.
The solution was to the change the contact algorithms. I use the contact card
*CONTACT_AUTOMATIC_GENERAL [25]. In this card the contact is defined
wholly by the slave part. With this algorithm, contact is considered between all
the surfaces in the slave list. This contact is quite efficient for sharp edges but
more time consuming than other contact.

Figure 26 - Good contact behaviour


48

As we can in Figure 26, this contact gives good result. So it has been kept.

5.5.2 Projectile/Fluid interface


After, it was the interaction between the projectile and fluid to be studied. Once
again, a small model has been made as shown in Figure 27.

Figure 27 - Projectile/fluid interface study


For a contact between an SPH part and a finite element part, only contact types
NODE TO SURFACE are permitted. So for this case two contacts were
permitted:

*CONTACT_AUTOMATIC_NODES_TO_SURFACE

*CONTACT _NODES_TO_SURFACE

In [22], for coupling SPH with FE, the author recommend to use the card
*CONTACT_ AUTOMATIC_NODES_TO_SURFACE. So I decided to use this
card instead of *CONTACT _NODES_TO_SURFACE. With this contact, some
penetration problems occurred initially as shown in Figure 28. They were
removed by using the soft constrain formulation (flag SOFT in the LS-DYNA
card [25]). But some particle were still penetrating, this was due to the sharp
edges of the projectile. Increasing the Maximum parametric coordinate in
49

segment search parameter has solved the problem (set to 1.20 instead of
1.025).

Figure 28 - Particle penetration problem

5.5.3 Tank/Fluid interface


The last interface to study was the tank/fluid one. This contact is very similar to
the previous one. The difference is here the relative velocity between the
particles and the wall is not very fast during the simulation and the shape on the
tank wall is simple without sharp edges.
For defining the appropriate contact, the simple model presented in Figure 29
has been made.

50

Figure 29 - Projectile/fluid interface study


As the card *CONTACT_AUTOMATIC_NODES_TO_SURFACE was working
with the previous contact, it has been used again for this interface.
Contrary to the previous contact, this one is working fine with the default card
parameters.
For representing the radius of influence of the SPH particles, an offset has been
set between the particle and the tank as shown in Figure 30.

Fluid

Tank

3mm

1.5mm

Figure 30 - Offset between the tank and the fluid particles


51

For creating the contact properly, a virtual thickness of 1.5mm (half distance
between each particle) has been applied to the SPH. This can be done through
the flag SST (Optional thickness for slave surface) in the contact card.
The same has been done for the contact between the projectile and the fluid.

5.6 Pressure initialisation


During the Hydrodynamic Ram, the cavitation appears when the water pressure
drops below the vapour pressure of water. So initially the pressure in the water
cant be set to zero. Thats why an atmospheric pressure had to be recreated in
the water.
So the pressure in the fluid had to be initialised to 0.1013MPa at the beginning
of the simulation. Since the pressure due to gravity is very small compared with
the atmospheric pressure, the gravity has not been set in the model.
Adding pressure initially in the fluid has been done through the EOS. The
Gruneisen EOS is stated like that:

   | 1 + 1 2 | 2 | 
=
+ ( + |)~
|
|S
1 (D 1)| 



S (|
|+1
+ 1)

The initial pressure when | = 0 is equal to  =  ~ . So by specifying a


~ = 0.921, the pressure is initialised to 0.1013MPa.

Adding a pressure in the fluid, create a pressure on the tank wall. So for having
a model at equilibrium at the initial state is important to create a counter
pressure on the tank walls equal to the atmospheric pressure. This pressure
has been made using the card *LOAD_SHELL_SET.

52

Figure 31 - Pressure initialisation


As we can see in Figure 31, the pressure is correctly initialised and no pressure
appears on the wall.

5.7 Pressure measurement


The last important thing in the model was the way the pressure is measured.
From the experiment used for assessing the capability of LS-DYNA to represent
the HRAM, the most important data is the pressure measurement.
A way for doing that was first tracking the pressure from an SPH particle located
close from a transducer. This can easily done with the card *DATABASE_
HISTORY_SPH. But this method has a major problem, the SPH particles are
moving during the simulation and the pressure transducer is supposed to stay
at the same place.
So the way which has been chosen for doing it was using the card *DATABASE
_TRACER. This card enables to track the history of either a material point or a
spatial point into an ASCII file. As the transducers are not moving with the fluid,
some special points have been defined in this card representing the position of
the transducers.
In the experiment 7 transducers are mounted. In the numerical models, they are
all represented for the big model and only two for the small model. Their
positions are summarized in Table 4.
53

Table 4 - Pressure transducer position


Transducer:

X position (mm)

Y position (mm)

Z position (mm)

301

37

215

215

37

301

74

602

111

903

1167

143

1167

(0, 0, 0)

Figure 32 - Coordinate system

5.8 Final model


As say before two models has been build. The small one (dimensions:
400x200x200mm) is presented in Figure 33. The bigger one (dimensions:
1168x500x500mm) is presented in Figure 34.

54

Figure 33 - Small model

Figure 34 - Big model

55

Table 5 - Model statistics


Number of SPH particles
Number of shell elements

Small model
127813
12800

Big model
2608912
80000

The termination time has been set to 30ms which is the necessary time in the
experiment for the cavity collapse. Pressure was recorded each s.

56

6 RESULTS
In this section, the results obtained from the two models are presented and
discussed. First the problems which occurred during the initial simulations are
presented. Then, as the HRAM is divided in 4 four phases, the result discussion
is divided in four parts: one for each phase.

6.1 Initial problems


6.1.1 Distributed memory computer
Primarily, the big model was made for being use on the distributed memory
computer of Cranfield: Astral. This computer allows the use of a great number
of processors and the calculation is very fast. However some problems
occurred and it was impossible to use it properly.
Indeed the first problem was due to contact. As the contact was defined for all
the particles, processors had some communication problem between
themselves. This results with some penetration problems which induce bad
results. For solving that I had to create to reduce the interfaces. For instance for
the fluid/projectile contact, I reduce the contact to only the particles located on
the shot line. This, coupling with the card *CONTROL_MPP_DECOMPOSITION
_CONTACT_DISTRIBUTE (which force the distribution of the contact over all
the processors used), solved the problem.
However some problems were still present. Due probably to a communication
problem between processors, during the post-processing of the results, only
pressure for small groups of particles were plotted on the screen as shown in
Figure 35.

57

Figure 35 - Post-processing problem


Moreover, the version of LS-DYNA for distributed memory is not able to handle
the card *DATABASE_TRACER used for representing the pressure transducer
[25]. So finally, it has been decided to not use the ASTRAL computer. A shared
memory computer (the GRID) has been used instead. However the use of this
computer allows uniquely 4 processors instead of the 24 used on ASTRAL. The
calculation for the big model was much longer and only one simulation has
been launched for the big model.

6.1.2 Double precision solver


Other problem important problem was the noise in the pressure history. Indeed
the first curves obtained were suffering from a lot noise. This problem could
have been overcome by filtering but this would have reduced the quality of the
result.
The problem was because water is a quasi-uncompressible fluid, the density
varies very slightly and so only in the decimals of the value. That is why using
only the single precision solver induces some imprecision in the calculation. The
use of the double precision solver has permitted to solve this problem and to
have smoother curves.

58

6.2 Shock phase


In this section, we will talk about the results obtain during the shock phase. We
are mainly focused on the pressure wave created during the shock.
First we are looking at the impact of projectile with the tank as shown in Figure
36.

Figure 36 - Projectile impact with the tank at t=0.03ms


This impact will create a shock wave travelling through the tank wall generating
dynamic stress. The Figure 37 shows the stress generated by the impact with
an empty tank. This will permit to compare later on with the additional stress
generated by the pressure wave.

59

Figure 37 - Von Mises stress generated by the impact on an empty tank at


0.023ms (MPa)
Indeed, after that the projectile is accelerating the fluid particles situated around
the impact area. This suddenly acceleration results with a pressure wave
created, as we can see in Figure 38.

Figure 38 - Pressure wave generated at 0.06ms


60

As say before this pressure wave is creating an extra stress on the tank wall as
shown in Figure 39.

Figure 39 - Von Mises stress generated by the impact on a filled tank at


0.023ms (MPa)
If we compare the Figure 37 with the Figure 39, we can find that:

In magnitude, the stress generated by the impact in the filled tank is


more important. The additional stress created by the pressure wave
represent 50% more in comparison with the dynamic stress due to the
impact on only the plate.

High values of stress are concentred around the entry hole for the empty
tank whereas for the filled tank the area is larger.

Those pictures explain why the HRAM can be a cause of major failure of the
tank. Instead of having a small hole created by the penetration of the projectile,
the high pressure generated on the wall can create a petalling of the tank
increasing largely the size of the hole.

61

The size of the holes is the same within or without tank in this simulation but this
is due to small impact velocity used (341m/s). In really impact of fuel tank can
occurs at much higher velocity.

If this pressure wave increases the risk of important failure of the tank, its
concerning only the impact zone. Indeed, the pressure wave is significantly
reduced when its travelling through water. As shown in Figure 40.

1200

1000

Pressure (MPa)

800

600

400

200

0
0

20

40

60

80

100

Distance from the wall (mm)

Figure 40 - Drop in pressure measured along the shot line on the big
model
We can see that after 50mm from the tank the pressure drops below 10 MPa
which is not anymore significant regarding the stress induced on the tank.
It seems that from a qualitative point of view, LS-DYNA is able to reproduce this
phase. Now its important to assess those results quantitatively.

62

120

From the experiment made by Disimile et al.[11], two types of data are
available: the pressure history for the transducer P3 shown in Figure 41 and the
pressure decrease between the transducer P3, P4, P5 and P7 compared with
the simulation results in Figure 43.

Figure 41 - Initial wave pressure recorded by the transducer P1 in the


experiment
If we are looking at the Figure 41, for transducer, the shock phase start with a
sharp pressure rise of 1.2Mpa and then followed by secondary waves of 0.3
MPa. The data obtained for the shock phase during the simulation are the
following:
1,2

Pressure (MPa)

1
0,8
0,6
0,4
0,2
0
0

0,00005

0,0001

0,00015

0,0002

0,00025

0,0003

Time (s)

Figure 42 - Pressure history at P3


63

0,00035

0,0004

For the wave arrival time, its difficult to assess it from the experiment as we
dont know the impact time. However we can calculate the wave velocity, this
one is equal to 1403m/s approximately the acoustic velocity in water which
agrees with impact dynamic theory.
From the data we have in Figure 41 (which has to be taking into account
carefully has, it has been filtered), its quite difficult to analyse it precisely.
However it seems that the pressure rises for the first pike agree with the value
of the experiment (3% of difference). Then the secondary waves fit also
correctly in term of amplitude.
The other data provided for this phase was the decrease of pressure of the
transducer P4, P5 and P7 compared with the transducer P3. The comparison

Pressure Rise/Transducer P3 pressure rise

with simulation results is shown in Figure 43.


1
0,9
0,8
0,7
0,6
0,5

Experiment

0,4

Simulation

0,3
0,2
0,1
0
200

400

600

800

1000

1200

Transducer Distance from impact (mm)

Figure 43 Decrease of pressure of the transducer P4, P5 and P7


compared with the transducer P3
The curves seem to fit correctly, except for the pressure transducer P7 located
on the tank wall where no pressure wave has been recorded. Two possibilities
can explain that: The pressure record is not correctly working or the wave has

64

been completely dissipated. But from the others transducer, we can assume
that LS-DYNA predict correctly the dissipation of the pressure wave within the
fluid.
The last thing to look at for this phase is the pressure decrease of the wave
when its going away from the shoot line. Indeed, McMillen et al. in [31] has
shown that the magnitude of the pressure along the arc of the wave decreases
with the angle from the shot line. In their experiment, Disimile et al. establish an
empirical law which state that P=P90SIN(+7), where P90 is the pressure in the
shoot line and the angle away from the shoot line as shown in Figure 44.

Figure 44 Angle [11]


This empirical law has been tested in the simulation with the pressure
transducer P2 which is at an angle of 45 and along the same arc than P3. So
we should found the relation S = sin ( + 7)


0.80

( + )
0.79

This agrees with the empirical relation from the experiment.


So, globally, we can conclude that for this phase, the SPH method implanted in
LS-DYNA is able to produce correct result from a qualitative and quantitative
point of view.

65

6.3 Drag phase


Now the next phase to study is the drag phase. During this phase the projectile
is moving through the fluid transferring to it, its momentum.

Figure 45 - Projectile pressure field


In Figure 45, we can observe the pressure field in front of the projectile. When
the projectile is approaching from a transducer a gradually pressure increase
occurs has shown in Figure 43.
The first thing to look is the velocity decrease due to the drag forces acting on
the sphere, if we are comparing this with the theory (where the velocity
reduction due to impact has been taken into account) we have:

66

400000
350000
Velocity (mm/s)

300000
250000
200000
Simulation
150000

Theory

100000
50000
0
0,00E+00

5,00E-04

1,00E-03

1,50E-03

2,00E-03

2,50E-03

Time (s)

Figure 46 - Decrease of the velocity due to drag forces


In the methodology part, we have been able to validate the behaviour of the
projectile moving through a fluid. Here, a difference between the two curves
appears, the velocity of the projectile is decreasing faster than the theory
predicts it. This means that the drags force applied on the projectile are more
important in the simulation. The explanation of this can be the fact that, as it
explained in page 2.1.312, the cavity created in the wake of the projectile
increase the drag coefficient. Moreover, in the methodology the fluid is free to
move instead of here, where the fluid is constrained at the free surface. So as
water is quasi-uncompressible, the drag forces are greater. For a better fitting
with the experiment, the entire tank should have been represented and the
lateral tank wall also.
Then, we can also look at the projectile pressure field. From the data we have
within the Figure 43, pressure history at the transducer P3 has been extracted
and compared with the pressure recorded in the simulation (NB: the extraction
of the data has been made approximately from a curve in a paper, so the
accuracy is not perfect). We obtain the following:

67

2
Experiment

Pressure (MPa)

Simulation (filtered at 8000Hz)


1,5

0,5

0
-0,0005

5E-18

0,0005

0,001

0,0015

-0,5
Time (s)

Figure 47 - Comparison of the drag pressures between the simulation and


the experiment
We can see that the curve obtained in the simulation is very noisy even after
having removed high frequency oscillation with a filter (this is why pressure
peak due to shock phase is reduced due to the filtering). In the experiment,
oscillations also appears and has been removed my filtering. According to the
authors those oscillations are due to the projectile oscillations.
The two curves obtained have some differences. Globally, if we removed the
oscillations, they have the same shape. After the shock phase, the pressure
stays stable around 0.4 MPa. And then, when the projectile is approaching from
the transducer the pressure field is gradually increasing. However in the
simulation the gradually increase of the pressure occurs later and the amplitude
is lower than the experiment. This is due to what it has been explained before.
The projectile velocity history is not the one expected. So the projectile is
travelling slower than it should. That is why the increase occurs later and the
intensity is less (the intensity is related with velocity of the projectile).

68

0,002

To check if this pressure increase correspond to the passage of the projectile,


its position relative to the transducer at t=1.6ms is show in Figure 48.

P3

Figure 48 - Projectile position relative to the transducer P3 at 1.6ms

SPH used in LSDYNA has the capability to reproduce correctly this phase.
However the model made is not totally accurate and a remaining task is to
investigate further why the velocity decrease is not what it should be and correct
it. A possible hint could be the boundaries condition used around the fluid.

6.4 Cavitation phase


An important phase of the HRAM is the cavitation phase. During this phase, the
collapse of the cavity can create an entire failure of the tank. So its very
important to be able to reproduce correctly this phase.
The cavitation appears when the pressure in the weak of the projectile drops
below the vapour pressure. At this moment the water initially a liquid changes of

69

phase and become vapour. In LS-DYNA, an equation of state able to reproduce


this doesnt exist. So a mean for representing this had to be found. A simple
way for doing that is to define a pressure cut-off. According to the LS-DYNA
manual [25], it allows for a material to numerically cavitate. When the pressure
drops below the value of the pressure cut-off, the material doesnt resist any
more to this dilatation. This method is supposed to allow the material to
cavitate. Here the pressure cut-off has been set to 0.0023 MPa, the value of the
vapour pressure.
The cavitation starts to form itself in the trail of the projectile. Indeed, in its trail
the pressure decrease and goes below the vapour pressure. From the
experiment a picture of the formation of this cavity is given. We can compare
the size of those cavities with the simulation.

90mm

500mm

Figure 49 - Cavity formation in the experiment [11]

70

250mm

40mm

Figure 50 - Cavity formation at 1.6ms


If we compare the Figure 49 and the Figure 50, we can see that the shape and
the size of the cavity behind the projectile are the same. This shows that LSDYNA until this step is able to reproduce this phase.
The problem is when the cavity will grow. Indeed once the cavity is formed, this
mixture of gas and vapour will grow as a bubble until reaching its maximum
radius as shown in Figure 51 - Cavity grow.

160mm

Figure 51 - Cavity grow [11]


In all the simulation carried out, the cavity has never expended itself. This is the
major limitation of the SPH method in LS-DYNA for reproducing the HRAM
event.
But this limitation is not due to the SPH formulation. Its due to the method used
for representing the cavitation. The pressure cut-off method is not accurate
71

enough for representing Supercavitation. The limitation comes from the


Equation of State used. For representing Supercavitation a new EOS has to
be input in LS-DYNA, this equation has to take into account the change of
phase of water.
In [30], McCallum et al. start the theory for a possible aeration EOS model. This
model is a mixture between a water EOS (like a Gruneisen EOS) and the
perfect gas EOS. In the model the two equations exist, and the change of phase
can be done by changing the fraction of gas within the water.
Another thing to do is also representing correctly the collapse mechanism of the
cavity. In the simulation made, the collapse mechanism (e.g. in Appendix A) is
very different from the one observe in Figure 5.
However, even if the collapse mechanism is wrong, LS-DYNA is able to record
a cavity collapse pressure has shown in. From a purely qualitative point of view,
the shape of cavity collapse pressure is similar the ones observed in the
experiment. This demonstrates to potential of the SPH method for representing
HRAM once some improvement will be done.

Figure 52 - Cavity collapse pressure in the small model for a transducer


located at 150mm from the impact
72

Figure 53 - Cavity collapse pressure in the experiment

6.5 Exit phase


The last phase on the HRAM is the exit phase. As this phase in not presented
and discussed in the paper of the experiment, this section describes it briefly.
This phase is when the projectile exits the tank. The potential risk of
catastrophic failure in this phase is due to the fact that the wall is already prestressed before the projectile impact it due to the pressure field in front of the
projectile.
As time was missing to perform the simulation until the end for the big model,
the only results obtained are from the small model.
Even if no data are available in for assessing this phase, it seems that the
fluid/structure coupling which very important in this phase seems to work
properly and we can be confident in the quality of the results obtained for this
phase during a HRAM simulation using SPH.

73

10mm

Figure 54 - Deflection of the tank before the impact

Figure 55 - Pre-stress in the tank wall before impact

74

7 CONCLUSION
The purpose of this thesis was to assess the LS-DYNA FE/SPH code for
representing the hydrodynamic ram phenomenon. Thats why a numerical
model was designed to represent the more precisely an experiment made
previously. The final goal was to be able to reproduce the results obtained from
the experiment.
The coupling FE/SPH seems to be a suited method for representing
fluid/structure interaction. Indeed, during the simulation the coupling seems to
work correctly as any penetration has been observed and the fluid was able to
deform the tank. Nevertheless, data about the tank deflection during the
experiment would have been more interesting in order to assess more
accurately the fluid/structure coupling.
The Hydrodynamic ram event is divided in four phases, each phase has been
assessed. For the shock phase, the SPH method has shown good results for
predicting the pressure wave magnitude and also the decrease of it while its
travelling. The simulation has highlighted the extra pressure created on the wall
during this phase.
During the drag phase, the method is able to reproduce the pressure field in
front of the projectile however the decrease of velocity due to the drag forces is
not exactly the one expected but this is probably due to some model limitation.
At the moment, the major limitation of the method comes from the capacity to
reproduce the cavity grow and collapse. Indeed, the cavity formation in the
projectile weak is well represented. However, due to the use of a limited EOS
the growing of cavity and its collapse are not correctly represented.
The simulation has prove that the SPH has an undeniable potential for
representing the HRAM event however some limitation has to be solve before
for representing properly the phenomenon.

75

8 FURTHER WORKS
Since the simulations of hydrodynamic ram with SPH are quite time consuming,
it has been the major constrain for achieving this study. That is why some points
have still to be investigated further:

The reasons for the incorrect velocity decrease during the drag phase.
Exploring further this problem could lead a better simulation of the HRAM
and could predict the same drag pressure that the one obtained in the
experiment. This problem can be due to the boundaries condition used
on the fluid as the methodology as shown that SPH are able to perfectly
reproduce this velocity reduction.

The creation of a new EOS. Using a mixture between a Gruneisen EOS


for water and a perfect gas EOS can lead to a correct prediction of the
Supercavitation phase which is very important point of the HRAM. LSDYNA permits user-defined EOS.

The tank wall damage. The fluid-structure interaction between water and
the tank can be studied more. It can be interesting to have good damage
prediction of the tank due to the HRAM. This implies the use of solid
element through the thickness for the impacted plate. Using this at higher
velocity can permit to observe the petalling of the wall due to the
pressure wave.

The projectile velocity. Here the projectile is fired at a small velocity. An


interesting study can be to show the influence of the velocity of the
results.

76

REFERENCES
[1]

Androdias, H. (2009), Helicopter fuel tank impact analysis using a coupled

FE-SPH approach (unpublished MSc. thesis), Cranfield University, .


[2]

BAE (2000), Final Investigation report on Concorde Crash, f-sc000725,

http://www.bea-fr.org/docspa/2000/f-sc000725/htm/f-sc000725.html.
[3]

Ball, R. E. (1976), Structural Response of Fluid-containing-tank to

Penetrating Projectiles (Hydraulic Ram)-A Comparison of Experimental and


Analytical Results, NPS-57Bp76051, Naval postgraduate school.
[4]

BALL, R.E., ( 1972), Aircraft fuel tank vulnerability to hydraulic

ram:modification of the northrop finite element computer code br-1 to include


fluid-structure interaction-theory and user's manual for br-1h.
[5]

Batchelor, G. K. (1967), "Cavitation in a liquid", in An introduction to fluid

dynamics, 1st edition ed, Cambridge University Press, , pp. 481-506.


[6]

Bharatram, G., Schimmels, B. and Venkayya, V. (1995), "Application of

MS/DYTRAN to the hydrodynamic ram problem", MSC 1995 World Users'


Conference, 1995, MSC Software, .
[7]

Borg, J. P., Cogar, J. R., Tredway, S., Yagla, J. and Zwiener, M. (2001),

"Damage resulting from a high-speed projectile impacting a liquid-filled metal


tank", Computational Engineering, pp. 889.
[8]

Buyuk, M., Kan, S. and Loikkanen, M. (2008), "Explicit Finite Element

Analysis of 2024-T3/T351 Aluminum Material under Impact Loading for Airplane


Engine Containment and Fragment Shielding", Engineering, Science,
Construction, and Operations in Challenging Environments, .
[9]

Chen Liang, Song Bifeng and Pei Yang (2011), "Simulation Analysis of

Hydrodynamic Ram Phenomenon in Composite Fuel Tank to Fragment Impact",


Measuring Technology and Mechatronics Automation (ICMTMA), 2011 Third
International Conference on, Vol. 3, pp. 241.
77

[10]

Disimile, P. J., Davis, J. and Toy, N. (2011), "Mitigation of shock waves

within a liquid filled tank", International Journal of Impact Engineering, vol. 38,
no. 2-3, pp. 61-72.
[11]

Disimile, P. J., Swanson, L. A. and Toy, N. (2009), "The hydrodynamic

ram pressure generated by spherical projectiles", International Journal of Impact


Engineering, vol. 36, no. 6, pp. 821-829.
[12]

Engineering toolbox , Drag coefficient, available at:

http://www.engineeringtoolbox.com/drag-coefficient-d_627.html (accessed
08/11).
[13]

Foll, J. (2009), Simulation of high velocity debris impact on aircraft fuel

tank (unpublished MSc. thesis), Cranfield University, .


[14]

Fordham Preparatory School (03/18/04), Vapor pressure of water,

available at:
http://www.fordhamprep.org/gcurran/sho/sho/reference/table74a.htm (accessed
06/22/11).
[15]

Freitas, C. J., Anderson, C. E., Jr, W. J. D. and Littlefield, D. L. (1996),

"Hydrodynamic ram: a benchmark suite", Structures under extreme loading


conditions, vol. 325, pp. 63-74.
[16]

Gingold, R. A. and Monaghan, J. J. (1977), "Smoothed particle

hydrodynamics - Theory and application to non-spherical stars",


Mon.Not.Roy.Astron.Soc., vol. 181, pp. 375-389.
[17]

Heinrich Kuttruff, K. (1999), "Pressure-induced interaction between

bubbles in a cavitation field", Journal of the Acoustical Society of America, vol.


106, no. 1, pp. 190-194.
[18]

Held, M. (1995), "Verification of the equation for radial crater growth by

shaped charge jet penetration", International Journal of Impact Engineering, vol.


17, no. 1-3, pp. 387-398.

78

[19]

Jones, D. A. and Belton, D. (2006), Smoothed particle hydrodynamics:

applications within DSTO, DSTO-TR-1922, Australian Department of Defence.


[20]

Kimsey, K. D. (1980), Numerical simulation of hydrodynamic ram,

A092380, US Army ballistic research laboratory.


[21]

Korobkin, A. A. (1992), "Blunt-body impact on a compressible liquid

surface", Journal of Fluid Mechanics, , no. 244, pp. 431-453.


[22]

Lacome, J. R. Smooth particle hydrodynamics - Part II, .

[23]

Lecysyn, N., Bony-Dandrieux, A., Aprin, L., Heymes, F., Slangen, P.,

Dusserre, G., Munier, L. and Le Gallic, C. (2010), "Experimental study of


hydraulic ram effects on a liquid storage tank: Analysis of overpressure and
cavitation induced by a high-speed projectile", Journal of hazardous materials,
vol. 178, no. 1-3, pp. 635-643.
[24]

Liu, G. R. and Liu, M. B. (eds.) (2003),

Smoothed Particle Hydrodynamics: a meshfree particle method, World


Scientific, Singapore.
[25]

Livermore Software Technology Corporation (ed.) (2006), LS-DYNA:

Keyword User's manual, Version 971 ed, .


[26]

Livermore Software Technology Corporation (ed.) (2006), LS-DYNA:

Theory Manual, .
[27]

LS-DYNA Support , Contact type in LS-DYNA, available at:

http://www.dynasupport.com/tutorial/contact-modeling-in-ls-dyna/contact-types
(accessed 08/11).
[28]

Lucy, L. B. (1977), "A numerical approach to the testing of the fission

hypothesis", Astron.J., vol. 82, pp. 1013-1024.


[29]

Lundstrom, E. A. (1977),

Fluid/structure interaction in hydraulic ram, technical report AFFDL-TR-77-32.

79

[30]

McCallum, S. C. and Townsend, D. (2005), "Simulation of hydrodynamic

ram and liquid aeration", 5th european ls-dyna users conference, Birmingham, .
[31]

McMillen, J. H. (1945), "

Shock wave pressures in water produced by impact of small spheres", Physical


Review, vol. 68, no. 9, pp. 198-209.
[32]

Monaghan, J. J. and Lattanzio, J. C. (1985), "A refined particle method

for astrophysical problems", Astron. Astrophys., vol. 149, pp. 135-143.


[33]

Moussa, N. A., Whale, D. E., Groszmann, M. D. and Zhang, X. J. (1997),

The Potential for Fuel Tank Fire and Hydrodynamic Ram From Uncontained
Aircraft Engine Debris, DOT/FAA/AR-96/95, Federal Aviation Administration,
Washington.
[34]

Randles, P. W., Seemann, D. R., Carney, T. C. and Libersky, L. D.

(1998), "Numerical Simulations of Fragment Impact on Liquid Filled


Containers", ASME Symposium on structures under extreme loading conditions,
San Diego, .
[35]

Sauer, M. (2011), "Simulation of high velocity impact in fluid-filled

containers using finite elements with adaptive coupling to smoothed particle


hydrodynamics", International Journal of Impact Engineering, vol. 38, no. 6, pp.
511-520.
[36]

Shi, H. -., Itoh, M. and Takami, T. (2000), "Optical observation of the

supercavitation induced by high-speed water entry", Journal of Fluids


Engineering, Transactions of the ASME, vol. 122, no. 4, pp. 806-810.
[37]

Stephani, P., Middendorf, P. and Le, C. (2006), "Numerical analysis of

the hydrodynamic ram of a CFRP integral tank", International conference on


structure under shock and impact, .
[38]

Stepka, F.S., Morse, C.R. And Dengler, R.P., (1998), investigation of

characteristics of pressure waves generated in water filled tanks impacted by


high-velocity projectiles.
80

[39]

Townsend, D., Park, N. and Devall, P. M. (2003), "Failure of fluid filled

structures due to high velocity fragment impact", International Journal of Impact


Engineering, vol. 29, no. 1-10, pp. 723-733.
[40]

Varas, D., Zaera, R. and Lpez-Puente, J. "Numerical modelling of

partially filled aircraft fuel tanks submitted to Hydrodynamic Ram", Aerospace


Science and Technology, vol. In Press, Corrected Proof.
[41]

Varas, D., Zaera, R. and Lpez-Puente, J. (2009), "Numerical modelling

of the hydrodynamic ram phenomenon", International Journal of Impact


Engineering, vol. 36, no. 3, pp. 363-374.
[42]

Varas, D., Zaera, R. and Lpez-Puente, J. (2011), "Experimental study of

CFRP fluid-filled tubes subjected to high-velocity impact", Composite


Structures, vol. 93, no. 10, pp. 2598-2609.
[43]

Vignjevic, R., De Vuyst, T., Campbell, J. and Bourne, N. (2002),

"Modelling of impact on a fuel tank using smoothed particle hydrodynamics", 5th


conference Dynamics and control of systems and structures in space, July
2002, Kings College, Cambridge, .
[44]

Vignjevic, R. (2004), "Review of Development of the Smooth Particle

Hydrodynamics (SPH) method", in Hobbs, S. (ed.), 6th Conference on


Dynamics and Control of Systems and Structures in Space, Cranfield
University, .
[45]

Wikipedia , Zustandsgleichung von Mie-Grneisen, available at:

http://de.wikipedia.org/wiki/Zustandsgleichung_von_Mie-Gr%C3%BCneisen
(accessed 08/2011).

81

82

APPENDICES
Appendix A

Cavity collapse mechanism for the small

model
The simulation was running too long for the big model to observe it.

t=1.4ms

t=2.2ms

t=3.0ms

t=4.1ms

t=6.3ms

83

You might also like