FULLTEXT01

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 81

Model for risk evaluation for

fragment debris after a grenade


detonation
Modell för riskbedömning av splitter från en granatdetonation

Gustav Lund

Faculty of Health, Science and Technology

Master of Science in Engineering Physics

30hp (ECTS)

Supervisor: Andrea Muntean

Examiner: Lars Johansson

Date: June 11, 2021


Abstract
Accidents when using or storing explosives can lead to a large number of casualties and injuries. Hence, it
is of vital importance, in all countries, to know the risk and act responsibly when working with explosives.
A model for evaluating the risk for fragment-induced injuries from grenade detonation is created starting
from experimental data of three different types of grenades. The grenades differ in shape, type of explosives
and design. The experiments were conducted shooting the grenades on a wooden target and the fragments
from the grenade detonation were collected by witness packages. The witness packages have a layered
structure of aluminum plates and polystyrene foam. The collected fragments are weighted and the number
of perforated plates for each fragment are counted. From the number of perforated plates the impact
velocities of the fragments are calculated and ballistics is then used to obtain the initial velocity of the
fragments. Fragmentation is regarded as a stochastic event and a distribution will more correctly describe
the variation in shapes and sizes of the fragments. All data obtained in the experiment are evaluated and
used to create distributions describing the fragmentation of the grenades.

The fragmentized objects are accelerated by the detonation and will, under the influence of the medium,
decreases their velocities. The velocity of the fragmentized objects are compared to criterions for skin
perforations developed in the 20th century.

According to a risk assessments manual, develop by the Swedish defense research agency (FOI), the risk
of severe injury can be regarded as acceptable when it occurs one in a million detonations. The distance
where only one fragment in one million detonations has the ability to perforate bare skin, according to the
injury criterions, is calculated using the developed model.

For the three tested grenades the distance at which the injury is acceptable (safety distance) did vary
between 55 m and 240 m. The variation in safety distances is assumed to be due to design variation between
the three grenades and also dependent on the model for injury criteria that was used.

Keywords: Risk evaluation, Ballistic limit of skin, Lewis model, Greenbook, External ballistics
Sammanfattning
Det är för alla länder viktigt att veta riskerna med de sprängmedel och ammunition som förvaras och
används av deras försvarsmakter. Olyckor som kan uppstå vid förvaring och användning av dessa vapen
kan leda till omfattade skador på personer i omgivningen. En modell för att beräkna riskerna för skador
från fragmenterade föremål från en detonation har skapats från experiment med tre olika typer av granater.
De tre granaterna varierar i form, typ av sprängmedel samt den övergripande designen. Experimenten
genomfördes genom skjuta de olika granaterna mot ett mål i trä, fragmenten som skapades vid detonationen
fångades sedan upp av vittnespaket. Vittnespaket har en struktur bestående av flera lager aluminiumplåtar
och frigolit. Fragment som fångats av dessa paket vägs och från antalet perforerade plåtar räknas. Från
antalet perforerade plåtar kan anslagshastigheten för fragmentet beräknas, ballistik används sedan för att
beräkna den initiala hastigheten för fragmentet. Från den data som erhållits av experimentet skapas en
fördelning av möjliga massor och en för möjliga hastigheter för fragmenten. Fördelningarna används för
att beskriva problemet, då fragmentering anses vara en stokastisk process.

Fragment som accelererats av detonationen kommer bromsas genom att dessa interagerar med mediet de
färdas i. Hastigheten som fragmenten har vid olika distanser från detonationen jämförs med villkor för
hudperforering som utvecklats under 1900-talet. Enligt en riskmatris utvecklad av Totalförsvarets
forskningsinstitut FOI, anses risken för alvarlig skada vara accepterat om skadan inte uppstår oftare än en
gång på en miljon fall. Avståndet där endast ett fragment per en miljon detonerade granater har en hastighet
tillräckligt hög för att perforera hud beräknas av modellen.

För de testade granaterna varierade detta avstånd mellan 55 och 240 m. Skillnaden i avstånd tros bero på
skillnader i granaternas är designade, samt vilket villkor för hudperforering som används.

Nyckelord: Riskbedömning, Ballistisk gräns för hud, Lewismodellen, Greenbookmodellen, Ytterballistik


Preface / Acknowledgement
This thesis is my master thesis in engineering physics at Karlstad University. The thesis work has been
conducted during the spring of 2021 at SAAB dynamics in Karlskoga. The thesis comprises 30 hp (ECTS)
and has been done at full time studies. The experiments have been conducted by SAAB and the data handed
to me. Parts of the created model are borrowed from programs earlier developed by SAAB, the borrowed
code are for the evaluation of initial velocities of fragments by Rilbe theory. All figures and diagrams in
this report have been produced by me in MATLAB, Affinity design and Office power point if nothing else
is stated.

The model for risk evaluation is developed in MATLAB. MATLAB-toolbox Statistics and Machine
Learning is needed to execute the program.

Throughout my thesis I have received support from my supervisor at Karlstad University Andrea Muntean
and my supervisor at SAAB dynamics in Karlskoga Viktor Tullgren. They have helped answer all questions
along the way. I want to thank Andrea Muntean and Viktor Tullgren for all time they spent helping me with
this work.
List of abbreviations

 LCS – Local Coordinate System


 GCS – Global Coordinate System
 ODE – Ordinary Differential Equation
 PDF – Probability Density Function
 FOA, FOI – Swedish Defense Research Agency
 CG – Center of Gravity
 CP – Center of Pressure
 CDF – Cumulative Distribution Function
 PMHS – Postmortem Human Subjects
 FSP – Fragment Simulation Projectile
 NATO – North Atlantic Treaty Organization
 DOF – Degree Of Freedom
Table of Contents
1 Introduction ............................................................................................................................. 1
2 Theory...................................................................................................................................... 3
Detonation ........................................................................................................................ 3
Rilbe Theory ............................................................................................................. 6
Probability density function and cumulative distribution function........................... 9
Ballistics ......................................................................................................................... 11
External ballistics .................................................................................................... 12
Forces and effects ................................................................................................... 14
Drag force coefficient ............................................................................................. 19
Air density............................................................................................................... 21
Projectile trajectory ........................................................................................................ 22
Injury models.................................................................................................................. 24
Perforation theory ................................................................................................... 26
The Lewis Model .................................................................................................... 26
The Greenbook formula .......................................................................................... 27
FOI Methodology of injury risk .............................................................................. 28
Risk................................................................................................................................. 28
Hit probability ......................................................................................................... 29
Risk assessment matrix ........................................................................................... 30
3 Experiment and analysis ........................................................................................................ 31
Experimental setup ......................................................................................................... 31
Natural fragments ........................................................................................................... 32
Bivariate distribution .............................................................................................. 37
Engineered fragments ..................................................................................................... 41
Spatial distribution ......................................................................................................... 44
4 Modelling............................................................................................................................... 45
Model validation ............................................................................................................ 52
5 Results and discussion ........................................................................................................... 54
6 Conclusions and Outlook....................................................................................................... 61
7 References ............................................................................................................................. 63
8 Appendix ............................................................................................................................... 66
Appendix A Calculation of PDF for joint distribution ................................................................. 66
Appendix B Calculate two independent normal distributions ...................................................... 68
Appendix C Solution to ODE in equation (2.15) .......................................................................... 69
Appendix D Expectation value of the projected area for tumbling convex body ......................... 70
1 Introduction
The most common cause of injury in military conflicts is from explosive devices [1]. Ammunition
and grenades are not only dangerous in military conflicts. Manufacturing, transportation, storage
are also fields were the risks of injuries are of huge interests. One question that could be asked is
how close to factories and warehouses can civil housing be build? Evaluating risks for severe
injuries to persons in both military and civil applications are of great interests for all countries.
When developing a model for risk evaluation many different aspects in physics must be regarded.
For example, how explosives works, ballistics and mechanical properties of the human body.

The processes occurring when a grenade detonates at some target can be separated in different
events. The first event is the ignition process. When the grenade reaches its target the explosives
within the grenade shell will be ignited by some trigger. The ignition of the explosive will lead to
an event of increasing pressure within the grenade shell. When the shell no longer can withstand
the enormous pressure, the shell will fragmentize, and the blast wave and fragments will accelerate
outwards. This can be regarded as the last process in the series of events leading up to the
fragmentized debris. The blast wave and fragments can then hit some target. Knowledge of the
dynamics of a detonation and the distribution of fragments is of great interest for both developing
and usage of the grenade. Fragments, will in this report refer to non-stable object of various shapes.
A spin stable object will instead be called projectile.

Explosive devices have three mechanisms of injury [2], primary (caused by blast wave), secondary
(due to the fragmentized debris) and tertiary (because of being pushed into other objects).
Sometimes, quaternary mechanism is included, which are all injuries not falling into any of the
three other mechanism categories. In quaternary mechanism, injuries like traumas, burns and
asthma can be found. The blast wave is a fast decaying impulse of increased pressure. The fast
decay makes it a short range mechanism and for the grenades of interest it can be neglected after
about 10 m. The secondary mechanism, the fragmentized particles accelerated outwards are of the
biggest concern for persons around the detonation.

Models for evaluating risks of injuries due to the secondary mechanism has been developed since
the beginning of the 20th century [3]. More intense research on the subject have been conducted
since the second half of the 20th century. The research has, in most cases been conducted through

1
animal testing. Statistical models have then been created for describing the event and probability
for severe injuries on persons.

Today a lot of different models are used, such as Lewis [4] and Greenbook [5] and many more [2].
The injury models used today are not fully in agreement. The lack of standardized methods for
experiments and the stochastic nature of fragment induced injuries makes that models differs a lot.
Lewis and Greenbook are two models that have gain some acceptance in the scientific community,
but the two models can give totally different conclusions.

In this thesis, a model for evaluating the risks for fragment induced injuries from a detonation is to
be obtained. Evaluating the risk is of huge importance in many aspects of weapons and
ammunitions. In addition to the risks in civil applications, like manufacturing and storage, also the
military using the weapons needs good knowledge of the risks when operating this kind of
weaponry. Experiments conducted by SAAB Dynamics on three different grenades is evaluated
and a statistical distribution of fragmentized objects is created as a way of describing the problem.
From the distribution and models for evaluating the probability of injury, the risks for fragment-
induced injuries at different distances and directions can be evaluated. Between the detonation and
the fragment hitting the skin, ballistic processes are considered.

This thesis will begin with a chapter describing the theory needed to create an accurate model of
the scattering of fragments from a detonation and in what region the risk of injury is high. The
theory is followed by information about the experiment and analysis of the experimental results.
The last part will describe the model created for evaluating the safety distance of the specific of an
arbitrary grenade. Results and discussions for three different grenades presented in the last chapter.

SAAB is a company in the defense industry. Their products and research have a lot of secrecy and
this will impact this thesis. For example, the results obtained for the three grenades cannot be fully
presented in this report and the result shown in this thesis are therefore only partially depicting the
problem and solution.

2
2 Theory
The simulation problem regarding a grenade detonation can and will in this thesis be divided into
three smaller problems. First the detonation, followed by the ballistics describing the interaction
between fragment and the air and lastly the probability for fragment-induced injury, see Figure 2.1.
These three smaller problems is solved separately and can then be brought together to simulate the
full problem. In this section, the physics regarding each problem is described, beginning with the
detonation.

Injury and
Detonation Ballistics
risk
Figure 2.1. Flow chart describing the smaller problems and in which order they are
treated.

Detonation

The detonation will regard all process from that the explosives are ignited until the fragments
propelled from the detonation have reach a stable state. The stable state is when the initial
acceleration has ended and the full set of initial parameters can be determine. Depending on the
type of grenade some variation in the detonation processes can exist, but in this section we regard
a standard detonation process for a grenade.

When the grenade fired from some weapon reaches the target the grenade will, by some kind of
trigger, detonate. The trigger starts a rapid chemical reaction of the explosives, gasses produces by
the reaction increases the pressure within the shell of the grenade. The increased pressure will force
the shell to expand. When the shell expand the thickness of the shell decreases and cracking of the
shell occurs. When the shell no longer withstand the pressure, fragmentation of the grenade shell
will occur. The fragments and the pressure wave are then accelerated outwards reaching high
velocities. The energy needed to fragmentize the shell is relatively small compared to the total
kinetic energy of the fragments and the blast wave [6].

3
The blast wave is described by a peak pressure and a duration. Blast wave is an outgoing wave of
increased pressure. For a grenade a blast wave is a spherical process originating from the increasing
pressure due to the rapid chemical reaction of the explosives. The wave front propagation is shown
in Figure 2.2 in a simplified two dimensional schematic view.

Propagate
Outwards

Δ𝑟

Figure 2.2. The wave front propagation outwards from the site of
detonation (the black dot in the middle).

A shock wave or detonation process must obey conservation laws, which are shown below,

𝜌1 𝑢1 = 𝜌2 𝑢2 → Conservation of mass,
𝜌1 𝑢12 + 𝑝1 = 𝜌2 𝑢22 + 𝑝2 → Conservation of momentum, ( ) (2.1)
1 1
ℎ1 + 𝑢12 = ℎ2 + 𝑢22 → Conservation of energy,
2 2

where 𝜌 is the density of the medium (air), 𝑝 is the pressure, 𝑢 the fluid velocity and ℎ the enthalpy.
The shock wave is normally described by an overpressure peak, at 𝑡 = 𝑡𝑜 , followed by a positive
pressure zone and then at the far field a negative pressure zone, this is shown in Figure 2.3. A large
number of experiments have been conducted on the near field close to the detonation, which is the
most crucial part in an injury point of view. The research has been summarized in a number of
semi-empirical formulas, describing an exponential decay of the positive pressure [7]. Since, the
blast wave decays exponentially, it will be a short range interaction. In this thesis we are interested
in the safety distances and for the safety distance the blast wave can be neglected. But some
properties of the blast wave still needs to be considered.

4
𝑝[𝑃𝑎]
Peak pressure

𝑡𝑖𝑚𝑒
Positive pressure
zone

Figure 2.3. Peak pressure and positive pressure zone schematically


presented.

The explosives will have a characteristic burning velocity describing the blast wave detonation
velocity and duration, which will set the initial velocity of the outgoing pressure front. This velocity
will also act as an upper limit to the initial velocity of fragments accelerated by the blast wave. This
velocity is normally much higher than the speed of sound. For a TNT explosive, the detonation
velocity is approximately 7000 m/s [8] [9]. The safety distance from a detonation is not set by the
blast wave, but by the distance travelled, the mass and velocity of the fragments. The velocity of
the fragments decay in a much slower rate than the pressure from the blast wave and will therefore
be able to cause injury at longer distances.

The Fragments propelled from a detonation are in general small and have high initial velocity. A
fragmentation process will create a great variation of fragment shapes and masses. The sizes and
shapes of the fragments depends on many quantities. From the expansion velocity describe in the
last paragraph to many material characteristics of the shell. The fragmentation process can be
simulated or numerically evaluated, but the errors in the size distribution tends to be too large to be
neglected [10]. The reason why numerical methods has difficulties describing the fragmentation
event is due to its stochastic nature. Where the shell is fragmentized is more dependent on grain
size and local defects which is difficult to account for in numerical models.

The other approach to analyze the detonation is to do experiments with the specific grenade that
will be tested. The biggest advantage with experiments are that the exact grenade properties can be
analyzed. This is the approach used for this thesis, the actual experiment and the information
obtained about the detonation will be described later in section 3.

5
Rilbe Theory
Obtaining the initial velocity of fragments through experiment, some method and model for
evaluating the velocity of the fragments is needed. The experiment will be more precisely described
later, but the theory used for obtaining the initial velocity is described in this section. Rilbe theory
is a formalism describing the relation between impact velocity and how much material the fragment
is able to traverse. This quantity is called perforation ability. Perforation is to be used when the
fragment has traversed the full width of the armor. Penetration, which is a similar quantity,
describes an event when a fragment as passed through less than all layer of armor, and that distance
is called penetration depth.

Rilbe theory evaluates the perforation ability of fragment. The model was first presented by the
Swedish defense research agency (FOA, now FOI) in 1970 [11]. The perforation ability is
dependent on a large number of factors such as fragment size, shape, density and hardness, target
material, the hardness of the target material and many more. The perforation of a fragment on a
plate with thickness t is more probable than that the same fragment would penetrate a distance t
into a block of the same material. From this follows that perforation depth are larger than
penetration depth. Important to notice is that few of the factors mentioned above are fully
experimentally analyzed [12].

In Rilbe theory, a linear relation between the perforated thickness g in the direction of motion for
the fragment, and the velocity are used which is shown in the equation below:

𝑔 = 𝜃𝑚1/3 𝑣, (2.2)

where 𝑚 is the mass of the fragment, 𝑣 is the impact velocity of the fragment and 𝜃 is a coefficient
depending on the fragment shape and material and target material, called Rilbe coefficient. The 𝜃
coefficient must therefore be evaluated for each combination of fragment material, shape and target
plate material. Equation (2.2) describes the perforation ability for single plate perforation. The rest
velocity of the fragment, after having perforated one plate, can be calculated from the equation
below:

𝜃𝑟𝑒𝑠 1/3 (2.3)


𝑣𝑟𝑒𝑠 = ( 𝑚 ) 𝑣0 ,
𝜃

6
where 𝜃𝑟𝑒𝑠 comes from the remaining perforation ability after perforating one plate. For multi-plate
perforation the Rilbe coefficient can be described as follows:

𝜃𝑛 = 𝜃 + 𝜃𝑖 (𝑛 − 1), (2.4)

where 𝜃𝑛 is a linear function describing the Rilbe coefficient as a function of the number of
perforated plates n. 𝜃 is the Rilbe constant for single plate perforation and 𝜃𝑖 the Rilbe constant for
multiple plates perforation. Equation (2.2) can be rewritten for multi-plate perforation as follows

𝑛 ∗ 𝑡 = (𝜃 + 𝜃𝑖 (𝑛 − 1))𝑚1/3 𝑣, (2.5)

where t is the thickness of each plate. The Rilbe coefficients 𝜃 and 𝜃𝑖 have been experimentally
determined by SAAB for the specific fragment material and target material used in the experiment
described in section 3. The method used to evaluate the Rilbe coefficients is conducted by shooting
a fragment with increasing velocity on a layered target. The layered target is made of ten thin soft
aluminum plates. After each shot the number of perforated plates are counted. From equation (2.2),
the Rilbe coefficient 𝜃𝑛 is evaluated. The Rilbe coefficient is then plotted against the number of
perforated plate and 𝜃 and 𝜃𝑖 are determined using equation (2.4). The following graph in Figure
2.4, is then produced for extracting the Rilbe coefficients.

𝜃𝑛

𝜃𝑖 (𝑛 − 1)

𝑛=1 𝑛=2 𝑛
Figure 2.4. The type of graph produced to extract the single and
multi-plate Rilbe coefficients.

7
For each combination of fragment material and target material a single plate perforation constant
and a multi-plate perforation constant is determined. The experimental values of the Rilbe
coefficient are obtained using fragment simulated projectile (FPS). The FPS is a standardized
experimental fragments. The standard is determined by North Atlantic Treaty Organization
(NATO). In Figure 2.5 a drawing of the FPS is shown. The results of the experimentally obtained
coefficients are tabulated in Table 1.

23.25
19.61
9.27

19.91

Figure 2.5. FSP fragment shape.

Table 1. Experimentally obtained values for the Rilbe coefficient. Experiments are conducted for aluminum
and tungsten fragments against 10 pieces of soft 0.5 mm aluminum plates.

Material Single-plate perforation Multi-plate perforation


Aluminum Fragments 7.12 ∗ 10−5 5.08 ∗ 10−6
Tungsten Fragments 1.56 ∗ 10−4 2.08 ∗ 10−5

With the knowledge of the dynamic motion in a medium, the ballistics, it is possible to track the
movement of the fragments back to the detonation. By tracking the movement back, the initial
velocity that fragment must have had for being able to perforate 𝑛 number of plates at a distance d

8
from the detonation, can be obtained. Notice that multi-plate evaluations are conducted with the
assumption that the fragments hitting the target plates doesn’t deform or break. That assumption is
in most cases not valid, but if the hardness of the fragment is high and the target material soft, it
might be a good enough approximation.

Probability density function and cumulative distribution function


For describing the statistical result obtained by the experiment described in section 3, some basic
formalism is needed. In this section two different statistical function is described. This functions
are needed to fully describe the detonation.

This variation in initial parameters can best be described by distributions. When dealing with
distributions two fundamental functions are important. Probability density function (PDF) or the
likelihood function describes the relative likelihood of a random variable taking a specific value in
a distribution. PDF can both describe discreet and continuous random variables.

Figure 2.6, PDF of a normal distribution. The darker area shows 95.4 % of the total
distribution and the lighter plus the darker shows 99.7%.

For instance, the PDF for a dice would be a constant of 1/6. For continuous random variables, the
relative likelihood is described by a function, like a bell curve. The most common PDF is the normal
distribution.

9
Another commonly used function in statistics is the cumulative distribution function (CDF). The
CDF describes the probability of the random variable being smaller or equal to some value. In
equation (2.6), the probability of a random variable x being smaller or equal to some value y is
evaluated.

𝑃(𝑥 ≤ 𝑦) = 𝐶𝐷𝐹(𝑦) (2.6)

With the CDF it is easy to evaluate the probability of x being in some interval Δ𝑦, just by taking
the difference of 𝑃(𝑥 ≤ 𝑦1 ) and 𝑃(𝑥 ≤ 𝑦2 ). In Figure 2.7 the CDF of the normal distribution in
Figure 2.6 is presented notice that it is 50% probability for a random variable to take a value below
the mean value.

Figure 2.7. The CDF of the normal distribution in Figure 2.6.

10
Ballistics

After the fragmentation of the shell and the initial acceleration of the fragments, each fragment can
be described by a set of initial parameters. This parameters includes mass, shape, initial velocity
and also rotation. This parameters will then determine the trajectory of the fragments and also how
the velocity of the fragments evolves with time. In this section, the important physics regarding the
interaction between fragment and medium and also gravity will be described.

Ballistics are often categorized in four subgroups [13], depending on the processes taking place in
the different parts of the flight for an object. The subgroups are internal-, intermediate-, external-
and terminal-ballistics. The four subcategories are schematically shown in Figure 2.8.

Internal ballistics deals with the events inside the gun barrel, before leaving the muzzle. In this
category the primer ignition process, and the first acceleration processes to the ammunition is
included. As mentioned in the previous section, every type of propellant and types of gunpowder
has a characteristic burning velocity. Unlike regular explosions, burning is usually used to describe
a controlled explosion. The burning process of gunpowder was first described by Guillaume Piobert
in 1839. He stated that gunpowder is burning from the surface layer and then layer by layer [14].
By adjusting the surface area of grains of gunpowder the ignition process could be controlled and
the initial velocity of the projectile could be adjusted.

While inside the gun barrel, the projectile is confined to one dimensional motion. When the
projectile leaves the muzzle, the projectile can jump or tip slightly. This occur due to the highly
pressurized gas behind the bullet. Small changes to the propellant can have large effects on the
initial orientation of which the projectile leaves the muzzle. These jump and tip processes happen
in close proximity to the muzzle and are parts of the so-called intermediate ballistics [13].

The projectile orientation, rotation and velocities after the intermediate ballistic become the initial
conditions for the external ballistics. External ballistics is the most commonly discussed category
of ballistics. External ballistics describes how the projectile is influenced by gravity and
aerodynamic forces and the path which the projectile will follow. This is the part of ballistics which
will be used to evolve the initial conditions of the fragments obtained from the detonation and
determine their velocities at different distances from the detonation.

When the projectile reaches the target, new events will take place. The projectile may penetrate
some armor and then detonate, and these events are described by terminal ballistics. Terminal

11
ballistics includes everything that happens at and around the detonation. Blast wave and fragments
accelerated out from the detonation can cause damage to people and buildings. When discussing
damage to people the term wound ballistics is used instead of terminal ballistics.

Figure 2.8. Schematic representation of the four subcategories of ballistics.

External ballistics
For the model which is developed in this thesis, the external ballistics is the subgroup of ballistics
that will be used to evolve the initial parameters of the fragments obtained in the detonation.
External ballistics, also called aero-ballistics describes how gravity and other forces influence the
trajectory of the projectile and fragments. The trajectory is the path followed by the projectile or
fragments in the air. After the detonation, the fragments have a set of initial parameter. The
fragment orientation can be described using Euler angles, where different rotation about the center
of mass are named roll, yaw and pitch. The local coordinate system (LCS) (the frame of reference
for the fragment) is described by Figure 2.9, in the figure an airplane is used instead of a fragment,
for easy visualize the different rotations.

12
Figure 2.9. Local coordinate system (LCS) for the projectile,
Euler rotation direction for the projectile.

The global coordinate system (GCS) (the frame of reference for an observer) can be rotated in a
way, that leaves the initial orientation only dependent on one angle, the angle between the velocity
vector and the ground. From this initial condition the system will evolve with time. In this report
Newton´s notation of derivatives is used, meaning that acceleration in x direction is noted 𝑥̈ . The
GCS is a Cartesian coordinate system with the x-axis parallel to the ground and the y axis
perpendicular to it. The origin of the GCS is chosen to be the point of detonation.

The easiest model for describing the external ballistic motion is to assume that the projectile is a
point mass, moving in a vacuum. When approximating the fragments as point masses the LCS is
no longer needed to describe the problem. Furthermore, none of the forces originating from the
medium are considered when the fragment is moving through a vacuum. The trajectory, the position
of the projectile as a function of time, can for this case easily be obtained by solving Newton´s
second law of motion, with gravitational force as only force acting on the system. This is true for
no initial acceleration to the projectile. Newton´s second law of motion is described as follows:

𝐹⃑ = 𝑚𝑔⃑, (2.7)

13
with m being the mass of the fragment and g the gravitational acceleration. From the equation of
motion above, the trajectory of the fragment can be evaluated using the initial condition for velocity
and orientation.

𝑥 = 𝑣0 𝑡cos(𝛼)
1 (2.8)
𝑦 = 𝑣0 𝑡sin(𝛼) − 𝑔𝑡 2
2

Disregarding the influence of the medium is not a reasonable approximation, so complexity is


added to the model by introducing the forces originated from the medium, in which the fragment
is moving. Also other external effects and forces contributing to the final trajectory, as wind and
the rotation of the globe can be added. Furthermore, the fragment can be considered to be a three
dimensional object, which would introduce more degrees of freedom and making the trajectory far
more complex to calculate.

Forces and effects


The fragments regarded in this thesis have a relatively short trajectory, meaning that they are only
in the sky for a couple of 100 m. The fragments will be small and have high initial velocities. The
force which will influence the fragments the most is the so-called drag force. In addition to the drag
force, the medium will also give rise to other forces. The other forces beside and effects beside the
drag force discussed in this section is important to know but will for this type of fragments,
trajectory and initial condition be negligible.

In general, the drag force can act in two ways, normal and parallel to the surface of the fragment.
The drag force parallel to the surface, arises from the friction between the medium and the surface.
The drag force normal to the surface is the most commonly discussed type of drag force. The drag
force normal to the surface originates from the resistance of the fluid to be pushed away or to the
sides of the fragment. If the fragment is moving close to the speed of sound, a third type of drag
force is formed due to increased pressure from shock waves. The shock waves are formed when
the local air flow reaches the speed of sound. In this thesis both meter per second and Mach will be
used to describe velocities. Mach 1 is the speed of sound.

14
In most weapons, the gun barrel is engineered to give the projectile (the ammunition) an initial
rotation. This rotation helps stabilize the projectile, but will also give raise to a drag force parallel
to the direction of rotation. Similar to the projectile, will the detonation give rise to rotation of the
fragments. The drag force parallel to the direction of rotation will create a spin damping effect. A
projectile fire from a gun will tend to be less stable in the end of the trajectory, since the friction
between the projectile surface and the medium will give raise to a spin-damping moment. This is
true if no additional rotation is added along the way. Fins or jet engines on a projectile can increase
the rotation along the trajectory. Also the spin of the fragment will be reduced throughout the flight,
but for a fragment the spin is random. A random spin is called tumble. The spin-damping
momentum from the medium is described by:

1 𝑝⃑𝑑 (2.9)
𝑀𝑠𝑝𝑖𝑛_𝑑𝑎𝑚𝑝𝑖𝑛𝑔 = 𝜌𝑣 2 𝐴𝐶𝑠𝑝𝑖𝑛 𝑑 ( ),
2 𝑣

where 𝜌 is the density of the fluid in which the projectile is moving, v is the velocity, A the interface
area between projectile and medium, 𝐶𝑠𝑝𝑖𝑛 the drag force coefficient, 𝑝⃑ is the vector about which
the projectile rotates and d the projectile diameter. The rotation of a projectile is very important
when an accurate trajectory is evaluated, but since the fragments tumbles, the rotation will have
much smaller impact. The normal drag force, faced in the opposite direction to the velocity vector
is described by:

1 (2.10)
𝐹⃑𝑑𝑟𝑎𝑔 = − 𝜌𝑣⃑ 2 𝐴𝐶𝐷 ,
2

in which 𝐶𝑑 is not the same as 𝐶𝑠𝑝𝑖𝑛 as in (2.9). In equation (2.10), the velocity 𝑣⃑ is the velocity of
the projectile or fragment in relation to the medium in which it travels [13]. A is the projected area
in the direction of the motion, 𝜌 is the density of the medium and 𝐶𝐷 is the so called drag force
coefficient. The drag force coefficient varies depending on the velocity, Reynolds number and the
shape of the object. In Table 7, in section 4, the velocity dependency of the drag force coefficient
is presented. The data in Table 7, is for a smooth spherical fragment. The drag force coefficient
increases at Mach 1, since the shock wave induced drag force also contributes.

15
The drag force will also give rise to some lift force on the projectile. This originates from a non-
zero pitch angle. In Figure 2.10 an airplane is used to show the zero-pitch angle. Zero-pitch angle
is the orientation where the lift and down force acting on center of pressure (CP) is zero. In ballistics
this angle is often called angle of attack. At take-off, airplanes have an angle of attack giving rise
to large lift force. The lift force can be shown in the figure below. CP describes the point where the
drag force acts on the projectile. The center of mass or center of gravity CG is the point where the
gravitational force is acting on the object. In addition to the lift force and spin reduction, the rotation
of the projectile can bend the trajectory; this is called Magnus effect.

𝐹𝑑𝑟𝑎𝑔

𝐹𝑔𝑟𝑎𝑣𝑖𝑡𝑦

Figure 2.10. Relevant forces and the point of interaction.

When a body rotates around another axis than the velocity axis and at the same time is moving
through a medium a force is created. The force originates form the friction produced, at the interface
between medium and body. Due to this friction, the density of the medium will locally increase or
decrease. The density will increase on the side where the rotation is faced in the same direction as
the overall motion and reduce on the side spinning in the opposite direction as the overall motion,
see Figure 2.11. This force only appears if the fragment rotates around a non-zero yaw and pitch
angle. The axis of rotation is schematically shown in Figure 2.12. The Magnus effect will also give
rise to another effect, namely turbulence behind the fragment, this can also be seen in Figure 2.11.
This turbulent flow of the medium behind the fragment will further increase the breaking power of
the medium. The Magnus force is normally small but the momentum obtained from this force may
not be negligible for all cases. The Magnus force and momentum is described in the following
equations [13]:

16
1 𝜔𝑑 (2.11)
𝐹𝑀𝑎𝑔𝑛𝑢𝑠 = 𝜌𝑣 2 𝐴 ( ) 𝐶𝑀𝑎𝑔𝑛𝑢𝑠 𝛿,
2 𝑉
1 𝜔𝑑 (2.12)
𝑀𝑀𝑎𝑔𝑛𝑢𝑠 = 𝜌𝑣 2 𝐴𝑑 ( ) 𝐶𝑀𝑎𝑔𝑛𝑢𝑠 𝛿,
2 𝑉

where 𝜔 is the angular velocity of the rotation around an axis described by the angle 𝛿. 𝛿 is the
combined angle of yaw and pitch, and is the angle between the direction of motion (velocity vector)
and the axis of rotation shown in Figure 2.12. 𝛿 will be the angle between those two vector in the
plane spanned by the vectors. The axis of the combined angle 𝛿 is schematically described in Figure
2.12 as a red line. Notice that the axis of rotation is non-parallel to the direction of motion.

Figure 2.11. Medium density variation due to a combination of rotation and transversal
motion.

17
Figure 2.12. Schematic description of the rotation axis described by 𝛿.
The angle 𝛿 is in the plane spanned by the direction of motion and the
axis of rotation.

Cross wind is also a force originating from the medium and is important to account for when aiming
with ballistic weapons. The cross wind will displace the projectile and fragments sideways. The
displacement of the projectile and fragments due to wind is typically small and of largest interest
for long range weapons. Cross wind can physically be described by a change in the velocity vector.

All this forces originating from the medium is of great interest when simulating ballistic events.
But due to the short range, the high initial velocities and the low mass of the fragments it is possible
to neglect most of the forces. There are many reasons for neglecting this forces, one of the reasons
is the computational load for simulating all this aspects. The second reason is that the drag force is
many magnitudes larger than the other forces making the error for neglecting all this forces small.

18
External ballistic will in addition to the forces from the medium also consider other forces and
effects. One such effect is the Coriolis effect. Newton´s law of motions describes the motion of an
object relative to a fixed frame of reference. If the coordinate system is transformed by some
rotation, Coriolis force and centripetal force are introduced. Such forces are called fictitious forces
and appear on a mass when the process is described in a system that undergoes an acceleration
relative to the initial system. Coriolis effect originates from the rotation of the Earth. For ballistics
this force is most relevant for long trajectories, like artillery and naval guns [13]. The Earth rotates
with an angular velocity of 15° per hour, so for short distances and short air time, the Coriolis effect
is negligible.

Drag force coefficient


The drag force is the only force originating from the medium which will be considered. As stated
earlier, the drag force coefficient depends on velocity of the fragment relative the medium, the
Reynolds number and fragment geometry. Reynolds number is an indication of the flow situation
for the fluid. Low Reynolds number, describes a mostly laminar flow, while high Reynolds number
describes a mostly turbulent flow. The Reynolds number is calculated from the equation below:

𝑢𝐿 (2.13)
𝑅𝑒 = ,
𝑣

where u is the flow speed, L is the characteristic length of a body within the flow and v the viscosity.
The dynamic viscosity of air is described by the following equation:

𝜂𝑎𝑖𝑟 = 2.2∗ 10−7 ∗ 𝑇 0.776 . (2.14)

19
Equation (2.14) is obtained from the tabulated data in Physics Handbook [15].

How the drag force coefficient depends on the Reynolds number is illustrated in the graph in Figure
2.13. The data is for a sphere with smooth surface [16].

Figure 2.13. Drag force dependency of the flow property, laminar or turbulent.

For some interval of the Reynold number the drag force coefficient can be regarded as independent
of the Reynolds number, since between 104 to 2 ∗ 105 a plateau can be found with a derivative
close to zero. There are many aspects of the flow of the fluid which contributes to the drag force
coefficient. For low Reynolds number, two processes determines the coefficient, the form drag that
originates from the fluid forces around the fragment and the skin friction which is due to the friction
between the fragment and the fluid. When the flow becomes more turbulent the friction between
the fragment and fluid decreases. The plateau which can be seen in Figure 2.13 an interval where
the skin friction can be neglected and the only the form drag contributes to the coefficient [17].

The drag force coefficient is defined for many different regular shapes of fragmants. Evaluation of
an irregular body is often done by assuming that the body drag can be estimated by some regular
body and some correction factor. For an asymmetric body, the coefficient will depend on the
orientation of the Fragments [18].

20
Air density
The drag force in equation (2.10) is proportional to the density of the medium, in this case air. The
density of the air is dependent on the atmospheric pressure and the temperature. So, for accurate
ballistic calculations, the temperature and atmospheric pressure must be well known. For long-
range trajectory, the density of the air will vary along the trajectory. The air density at different
pressure and temperature can be evaluated using the ideal gas law in equation (2.15).

𝑝𝑀
𝜌= (2.15)
𝑅𝑇

𝑝 − atmospheric pressure[𝑃𝑎]
𝑉 − Volume [𝑚3 ]
𝑛 − amount of substance[𝑚𝑜𝑙]
𝐽
𝑅 − Gas constant, 8.3145 [ ]
𝑚𝑜𝑙 ∗ 𝐾
𝑇 − Temperature [𝐾]
𝑘𝑔
𝜌 − air density [ ]
𝑚3
𝑀 − Molar mass of air

21
Projectile trajectory

In this section we will take the theory discussed in section 2.2 and describe how the trajectory and
velocity is determine.

By neglecting all forces due to the rotation of the fragment, it is possible to describe the full model
using only the GSC. The fragments can also be approximated to be a point mass, making this a
three degree of freedom (3-DOF) problem.

The Drag force cannot be neglected. The stopping force of the medium, in this case the air, will be
large. Equation (2.7) and (2.8) will therefore be further sophisticated by introducing the drag force
into the equations. The drag force acts in the opposite direction as the motion of the fragments.
Meaning that the accelerations in x and y is described as follows:

1 (2.16)
𝑥̈ = 𝐶 𝐴𝜌𝑣⃑𝑡𝑜𝑡 • 𝑥̇ ,
2𝑚 𝑑

1 (2.17)
𝑦̈ = 𝑔 − 𝐶 𝐴𝜌𝑣⃑𝑡𝑜𝑡 • 𝑦̇ ,
2𝑚 𝑑

𝑥̇
where 𝑣⃑𝑡𝑜𝑡 = [ ]. By solving the second order ordinary differential equation (ODE) in equation
𝑦̇
(2.16) and (2.17) the trajectory of the fragment is evaluated. The sign of the gravity depends on,
whether projectile is moving up or down. When the fragment reaches its maximum ordinate (the
highest point of the trajectory) the gravity will help to accelerate the fragment instead of
deaccelerating the fragment.

The trajectory described by equation (2.16) and (2.17) is non-linear and must be solved numerically
for each fragment. This is done by first assuming that the acceleration in equation (2.16) and (2.17)
is constant over a short time interval. The velocity at 𝑡 + Δ𝑡 is then easily calculated using the linear
equation below.

1 (2.18)
𝑥̇ 𝑖 = 𝑥̇ 𝑖−1 + 𝐶 𝐴𝜌𝑥̇ 2 𝑖−1 ∗ Δ𝑡
2𝑚 𝑑

𝑥̈
1 2 (2.19)
𝑦̇ 𝑖 = 𝑦̇ 𝑖−1 + (𝑔⃑ − 𝐶 𝐴𝜌𝑦̇ 𝑖−1 ) ∗ Δ𝑡
⏟ 2𝑚 𝑑
𝑦̈

22
The total velocity can then easily be calculated by the following equation.

𝑣𝑡𝑜𝑡 = √𝑥̇ 2 + 𝑦̇ 2 (2.20)

The equations (2.18) and (2.19) can be solved using a numerical method with some small Δ𝑡. This
numerical method for the ballistics is computationally heavy and inconvenient to use for large
number of fragments. For the purpose of safety distance evaluation, approximations to the physical
problem must be valid for the range in velocity where the fragments are regarded as dangerous.
The behavior of the model at long distances from the detonation when the small fragments are
moving slow is not important for risk evaluations. In section Error! Reference source not found.
a couple of injury models are presented. Those models give an estimation to within which velocity
interval the fragments are dangerous. A common feature for all models is that fragments tend to be
harmless at velocities lower than approximately 100 m/s . When the velocity is high, the drag force
are many magnitudes larger than the gravitational force. A good approximation for the forward
motion of a body in air is to neglect the influence of gravity. For that case equation (2.18) and (2.19)
can be combined to create the differential equation of motion shown below.

𝑑𝑣 1 (2.21)
𝑚( ) = − 𝐶𝑑 𝐴𝜌𝑣⃑ 2
𝑑𝑡 2

Equation (2.21) is a separable ODE and a solution to (2.21) is:

−1 (2.22)
𝑣(𝑑) = 𝑣𝑜 exp ( 𝐶 𝐴𝜌𝑑)
2𝑚 𝑑

where 𝑑 is the displacement from the site of detonation. The full derivation to equation (2.21) can
be found in Appendix C. The approximation to not regard the influence of the gravity is reasonable
in the sense of safety evaluation, since the gravity is a force decreasing the velocity of the fragments.
Removing the gravity is okay as long as the terminal velocity of the fragments can be regarded as
harmless, which is true in this case. In equation (2.22), the mass is obtained from the distribution
created from the experimentally obtained data, see section 2.1 and 3. 𝜌 is the density of air and is
calculated from equation (2.15) for different temperatures and pressures.

23
Limiting the model to 3-DOF might be a too large approximation. When non-spherical fragment
tumble, the projected area and the also the drag force coefficient will depend on the orientation of
the fragment. This limitations of the 3-DOF model can partly be solve by using a drag force
coefficient and projected area adjusted for the tumbling. This will be described later in the model
description.

Injury models

When a fragment hit a target or a person it can either do damage or not. The mass of the fragment,
the size and the velocity is just three quantities that may influence whatever the fragment will be
able to damage a specific target. In this section general perforation theory for fragment will be
discussed. After the general information three empirical models for fragment induced injuries to
persons will be described.

When a grenade detonates a blast wave propagate out from the site of detonation. This blast wave
can severely injure humans in a close proximity to the detonation. Blast wave can heavily damage
the lungs and the ear drums because of the high pressure wave originating from the site of
detonation. This has been experimentally tested. The pioneer work in this field was done by L.G
Bowen back in 1968 [19] and later improved by C.R Bass in 2008 [20]. As mention in section 2.1
is the blast wave a short range interaction and is not the limiting factor when determine safety
distances. As mentioned in the introduction, there are four mechanisms of injury from a detonation,
where primary is the blast wave and secondary mechanism is the propelled fragments. Different
mechanisms can affect a person simultaneously or one by one, resulting in different injuries. In
Figure 2.14, a Venn diagram is used to represent simultaneous injuries. In close proximity to the
detonation all four injury mechanisms will affect the outcome.

Injury A Injury B

Figure 2.14. Venn diagram of two simultaneous injury mechanism. For


example A) blast wave induced injury and B) fragment induced injury.

24
Fragments are in most studies considered to be hazardous if they have the energy or velocity needed
to perforate bare skin, and so also for this paper. Perforation of skin, describes a fragment that has
traversed the full width of the skin. The ballistic limit of skin or the probability of skin perforation
has been tested on a large number of living animals and cadavers. In some studies postmortem
human subjects (PMHS) have been used to determine the ballistic limit of skin.

Skin perforation of fragments depends on numerable things: the mass, projected area of a fragment
and of course the impact velocity. The velocity used in different models differs a lot. Historically,
a threshold velocity has been used, below which no perforation occurs. In more modern studies,
the 𝑣50 velocity is used instead, which is described in the section 2.4.1 [21].

One of the most famous unclassified studies on the ballistic limit was conducted by Sperrazza and
Kokinakis in 1967. In their experiments they fired steel cubes, spheres and cylinders on isolated
goat and PMHS skin. Their conclusion was that the 𝑣50 is dependent on a shape factor. The shape
factor proposed was the ratio between the cross sectional area and the mass of the fragments [22]
[23], equation (2.23). The 𝑣50 according to Sperrazza and Kokinakis is described by equation
(2.24).

𝐴 (2.23)
𝑣50 ∝
𝑚𝑓

𝐴 (2.24)
𝑣50 = 1247.1 ( ) + 22.03
𝑚𝑓

Many of the unclassified methods in use today are based on the work done by Sperrazza and
Kokinakis. Two models for describing fragment-induced injuries that have been widely accepted
in literature are, the models by Lewis and Greenbook. In this thesis a model proposed by FOI will
also be used.

25
Perforation theory
Fragments hitting an arbitrary target can penetrate or perforate the target. Penetration or perforation
can be regarded as a stochastic event. Perforation is therefore described by a statistical parameter,
describing the probability for perforation. This parameter is called 𝑣𝑝 where p stand for the
probability in percent. Most commonly used is the 𝑣50 term, meaning the velocity in which 50 %
of the fragments will perforate the target. This limit velocity is experimentally determined. U.S.
Army Ballistic Research Laboratory (BRL) tests this material property by shooting one fragment
at a target (some armor). If the fragment perforate the armor they lower the impact velocity until
the fragments no longer perforate the armor [13]. They repeat this process until they have at least
three fragments that perforate the target and three that did not perforate the target. The number three
is chosen arbitrary. A higher number of trials could be used to improve the statistics. The limit
velocity is then determined by calculating the mean value of the six velocities. The limit velocity
is also called Ballistic limit when discussing armor.

6
1
𝑣50 = ∑ 𝑉𝑖
6 (2.25)
𝑖=1

The Lewis Model


The Lewis model is an empirical model determine the probability of skin perforation, given the
projected area, weight and velocity of a fragment. The projected area or striking area is assumed to
be the average projected area of a tumbling convex body. This area is calculated in Appendix D.
The velocity used in the Lewis model is 𝑣50 and the experiment is conducted on goat skin over
gelatin. The use of gelatin is a way to simulate muscles and tissues under the skin [4] [21]. The
probability of skin perforation according to the Lewis model is described as follows:

1
𝑃𝑠𝑘𝑖𝑛 = , (2.26)
𝑚𝑓 𝑣 2
1 + exp [− (𝑎 + 𝑏ln( 10𝐴 )]

in which 𝑚𝑓 is the fragment weight in kg, 𝑣 is the impact velocity of the fragment and 𝐴 the
projected area described in Appendix D. 𝑎 and 𝑏 are empirical constants different values of 𝑎 and

26
𝑏 are used to describe different scenarios, bare skin, 2 layered uniform or 6 layered uniform, see
Table 2.

Table 2. Empirical constants of the skin perforation Lewis formula, see equation (2.26)

𝑎 𝑏
Bare skin −28.42 2.94
Skin + 2-layer uniform −48.47 4.62
Skin + 6-layer uniform −50.63 4.51

The Greenbook formula


The Greenbook formula is also an empirical model describing the probability of perforation of bare
skin. Like Lewis, Greenbook uses the 𝑣50 velocity to describe the probability. Greenbook
categorize fragments of different mass into two groups, cutting and non-cutting fragments. The
cutting fragments are small and will perforate skin when colliding. Non-cutting describes larger
fragments which instead of perforating, produce injuries due to contact pressure. Greenbook uses
the notation fragment for cutting and debris for non-cutting.

The Lewis model directly describes the probability for injury. Greenbook instead uses a probit
function for determination of the probability of skin perforation. The probit function is then
evaluated by the cumulative distribution function of a normal distribution with mean value 5 and
standard deviation 1, see equation below:

𝑃(𝑣) = 𝑛𝑜𝑟𝑚𝑐𝑑𝑓(𝑌, 5,1), where Y is the probit function. (2.27)

Notice that the notation in equation (2.27) is the syntax in MATLAB. normcdf evaluates the
probability of of a random value taking a value below or equal to Y this will be further described
in section 2.1.2. The probit functions Y is described by the three equations below. Fragments with
mass higher than 0.1 are regarded as non-cutting debris [5].

𝑌 = −13.19 + 10.54 ln(𝑣0 ) , 𝑓𝑜𝑟 𝑚𝑓 > 4.5 𝑘𝑔


1
𝑌 = −17.56 + 5.30 ln ( 𝑚𝑓 𝑣02 ) , 𝑓𝑜𝑟 0.1 > 𝑚𝑓 > 4.5 𝑘𝑔 (2.28)
2
𝑌 = −29.15 + 2.10 ln(𝑚𝑓 𝑣05.115 ), 𝑓𝑜𝑟 0.001 > 𝑚𝑓 > 0.1 𝑘𝑔

27
𝑚𝑓 is the fragment weight in kg and 𝑣0 is the impact velocity of the fragment when hitting the skin.

FOI Methodology of injury risk


FOI created a model for evaluating risk based on three other reports and studies [22] [24] [25],
important to notice is that FOI has not conducted the experiments themselves. The FOI model uses
𝐸50 energies or 𝑣50 velocities [26]. FOI has categorized different levels of injuries, and their
corresponding 𝐸50 energy.

Unique for this model is that a 𝐸50 for eye injuries is determined. The experiments on ballistic
limits of the eye is presented by K. Sellier and B. Kneubuehl [24].

Table 3. Limit energies and limit energy and energy density for 50 percent risk of injury

Injury class 𝑬𝟓𝟎 level


Death 240 𝐽
Severe injury 50 𝐽
Skin perforation 20 𝐽/𝑐𝑚2
Eye injury 3 𝐽/𝑐𝑚2

The 𝑣50 can be calculated from the kinetic energy 𝐸50 of the fragment. This is done as follows:

2000𝐸50 𝐴
𝑣50 = √ , (2.29)
𝑚

where 𝑚 is the fragment weight in grams and A the cross sectional area in 𝑐𝑚2 and 𝐸50 the energy
density in 𝐽/𝑐𝑚2 .

Risk

The model in this thesis is developed for analyzing the risk, meaning that it is not enough to only
know which fragment is hazardous or not. We need to be able to also evaluate the risk of being hit

28
by those hazardous fragments. In addition to the probability of being hit, a risk assessment matrix
determine which level of risk that can be acceptable or not is needed to fully determine the safety
distances for the different grenades.

Hit probability
The probability of actually being hit by a propelled fragment from a detonation can easily be
described as the number of hazardous fragments per square meter times the size of the target
exposed to the fragment.

𝐴𝑡𝑎𝑟𝑔𝑒𝑡 (2.30)
𝑃ℎ𝑖𝑡 = 𝑁 ∗
𝐴𝑡𝑜𝑡𝑎𝑙

N is the number of hazardous fragments, 𝐴𝑡𝑎𝑟𝑔𝑒𝑡 is the area of a target exposed and facing the
detonation and 𝐴𝑡𝑜𝑡𝑎𝑙 is the total size of the imaginary sphere around the detonation. The sphere
around the detonation will increase in size with distance. Ten meters from the detonation the
fragments will be distributed over an area of 400𝜋, according to equation (2.32). The purpose of
this thesis is to evaluate the probability of injuries on humans. The exposed area, the 𝐴𝑡𝑎𝑟𝑔𝑒𝑡 , for a
human is assumed to be half of the total surface area of a human. The total surface area of a human
can be approximated by the weight height formula presented by Du Bois in 1916 [27]. The weight
height formula is shown below:

𝐴 = 𝑊 0.425 ∗ 𝐻 0.725 ∗ 71.85, (2.31)

where W is the weight in kg and H the height in cm. In Sweden, the average man was 180 cm tall
and weight 84 kg in the year 2017 [28]. This gives a total surface area of approximately 2 square
meters. So, a standing man has approximately 1 square meter area exposed for fragments.

The detonation is confined to a small volume in space approximated as a point. The surface area of
the imaginary sphere confining the detonation increases with 𝑟 2 , described as follows:

𝐴𝑡𝑜𝑡𝑎𝑙 = 4𝜋𝑟 2 . (2.32)

29
Risk assessment matrix
Risk is the combination of damage level and at what frequency it occurs. The models presented in
the last sections are focused on severe injury. A higher level of damage than severe injury would
be death to persons. When operating explosive weapons, persons other than the shooter can be
injured. The risk must therefore be evaluated in all directions. FOI has created a risk assessment
matrix which will be used to evaluate the level of risk at different distances. In Table 4 the risk
assessment matrix is shown for unprotected civilians. For fragments propelled from a detonation
the risk is described as the probability of being hit by a fragment causing a severe injury per
detonation. So, an acceptable risk for severe injury is when one fragment in one million shot or
detonation, can perforate the skin at some distance. FOI has categorized the level of risk into three
levels, NA, LA and A. NA stands for not acceptable level of risk, LA for limited acceptable and A
stands for acceptable.

Table 4. Risk assessment matrix for unprotected civilian. NA = Not acceptable, LA = limited acceptable
and A = Acceptable [26].

Level of damage Frequency > 10−4 > 10−5 > 10−6 > 10−7 ≤ 10−7
≤ 10−4 ≤ 10−5 ≤ 10−6
Death NA NA LA LA A
Severe injury NA LA LA A A
Less severe injury LA LA A A A
Negligible injury A A A A A

30
3 Experiment and analysis
In section 2.1 we conclude that the detonation is best described by actual detonations. In this section
the experiment used to obtain the initial conditions of the fragments is described. The obtained
result from the detonation will also be analyzed and described using distributions. The model for
safety distance simulations in this thesis is developed by the information gained from arena tests
conducted with three different grenades. Due to secrecy, the notation used for the three grenades
will be grenade 1, 2 and 3. There are large variations in design between the three grenades. With
large variation of the tested grenades, the model will hopefully be more versatile. The safety
distances of same three grenades will later be analyzed using the developed model, see the result
in section 5.

Experimental setup

The fragment debris from a detonation of a grenade can consist of various kinds of objects such as
fragments from the fragmentized shell and fragmentized electronics. Fragments originating from a
fracture processes of for example the grenade shell or the target are called natural fragments. The
natural fragments will have all kinds of shapes and sizes. In addition to natural fragments, pre-
fragmentized objects of well-defined shape and size, like nails or balls, can be a part of the grenade
design. The pre-fragmentized objects will in this report be called engineered fragments.

During the test, three shots with each type of grenade were fired from a weapon against a wooden
target. The reason why three shots are used, is to improve the statistic distribution of fragments.
Witness packages then collect fragments in different directions. Each witness package has a
sandwich structure of aluminum plates and polystyrene foam layers. The polystyrene foam is used
to collect the fragments between the plates. When evaluating the impact velocity, the breaking force
of the polystyrene foam between the plates is neglected. After each shot the witness packages are
disassembled one sector at the time. The fragments found in each sector are counted and weighed,
creating one data set for each sector. The data is summarized in an excel file. In the excel file, each
row contains parameters regarding one fragment. Each row has three columns, fragment mass in
mg, number of perforated witness plates and in which sector the fragment is found. Separate data
sets and excel files for natural and engineered fragments are created. From Rilbe theory [11], the
impact velocity of the fragments on the first plate can be calculated using the mass of the fragment
and the number of perforated plates as shown in section 2.1.1. The experimental setup is described

31
in Figure 3.1. The witness package, numbered 1 to 8 are 1 meter wide and 2 m high, two witness
packages are stacked above each other in each sector, making the whole area of each sector 4 m2 .
The distance between the wooden target and the witness packages is 5 m.

Figure 3.1. Top view of the test setup, 8 sectors containing two witness packages each. (Principal setup, not to scale),
This image is borrowed from SAAB.

For this test, it is enough to only study this circle sector because of the rotational symmetry of the
grenade, making the fragmentized debris approximately symmetric. For risk analysis purposes the
semi-circle behind the detonation is not important. The risk behind the detonation is determined by
the distance traveled by a projectile missing its target. That distance will not be evaluated in this
thesis.

Natural fragments

When the impact velocity is obtained, the initial velocity of each fragment is calculated using
equation (2.22). The different masses and velocities obtained for the natural fragments will in this
section be analyzed. The natural fragments for these three grenades are made from aluminum.

32
The obtained data from the experiment shows a log-normal distribution for both mass and velocity.
This can easily be seen in Figure 3.2a. A distribution is said to be log-normal if the natural logarithm
of the data is normally distributed. Y in equation (3.1) being normally distributed, is equivalent
with X being log-normal. This is proven in Figure 3.2b where the natural logarithm of mass and
velocity is plotted and shows a tendency to be a normal distribution. In Figure 3.2 discrete bands
of mass and velocity is clearly visible. These bands are an artifact from the experiment, the real
result should be a continuous distribution of fragments. For the experiment, fragments either
perforate 𝑛 or 𝑛 + 1 plates. This makes a quantization of the possible values of the velocity of the
measured fragments. If the number of perforated plates 𝑛 is limited to be integers, it is quite obvious
that the Rilbe formula in equation (2.5) will give rise to velocity bands. Similar features are obtained
for all three grenades, see Figure 3.5 and Figure 3.4.


𝑌 = ln( 𝑋
⏟ ) (3.1)
𝑁𝑜𝑟𝑚𝑎𝑙 𝐿𝑜𝑔−𝑛𝑜𝑟𝑚𝑎𝑙

Figure 3.2. Mass versus velocity for (a) experimentally obtained data and (b) the natural logarithm of the experimentally
obtained data. The data set comes from on specific sector from the test, see Figure 3.1.

If two independent normally distributed random variables are plotted it creates an ellipse. The
relative axis of the ellipse will depend on the standard deviations of the two variables. The main
axis of an ellipse is the axis where the largest diameter of the ellipse can be found. The main axis
should for independent variables be either vertical or horizontal (the x and y axis of a Cartesian

33
system). In Figure 3.2 the main axis shows a negative slope, suggesting a correlation between the
two random variables mass and velocity, this is further shown in Figure 3.3.

Figure 3.3. The elliptic shape of the logarithm of the obtained data. The two
black lines indicates the two main arises of the ellipse.

Correlation between two normally distributed variables, the tilt of the main axis of the ellipse, can
be obtained by the so-called covariance matrix. Covariance is a measurement of how the
variances of the two random variables correlates. Covariance matrix is a square matrix describing
the two dimensional variance of the two random variables. If large values of one of the random
variables correlates with the large value of the other random variable, the covariance is positive.
The covariance matrix is calculated by the equation below and later presented for the data in
Figure 3.2b in Table 5.

𝐶𝑜𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑚𝑎𝑡𝑟𝑖𝑥 = 𝐶𝑖𝑗 = 𝐸[(𝑋𝑖 − 𝐸[𝑋𝑖 ])(𝑋𝑗 − 𝐸[𝑋𝑗 ])] (3.2)

𝐸[𝑋] is the expectation value of 𝑋, and 𝑋𝑖 and 𝑋𝑗 is the two different data sets, in this case the mass
and velocity.

Table 5. Covariance matrix of the data in Figure 3.2b (natural logarithm of the
experimentally obtained data).
1.0800 -0.5649
-0.5649 0.5884

34
In Table 5 the covariance (the non-diagonal values) are negative. The negative covariance means
that the small values of one of the random variables correlates with the large values of the other.
The covariance matrix can also be analytically obtained using the following identities,

𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 → 𝜎 = √𝐸[(𝑋 − 𝜇𝑋 )2 ], 𝐸[𝑋] = 𝜇𝑋


𝐶𝑖𝑗 (3.3)
𝐶𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝑓𝑎𝑐𝑡𝑜𝑟 → 𝜌 = .
𝜎𝑋 𝜎𝑌

The correlation factor 𝜌 will describe the slope of the main axis in Figure 3.3. The covariance
matrix is by the identities above analytically obtained as follows:

𝜎2 𝜌𝜎𝑋 𝜎𝑌 (3.4)
𝐶𝑜𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒 𝑚𝑎𝑡𝑟𝑖𝑥 → [ 𝑋 ].
𝜌𝜎𝑌 𝜎𝑋 𝜎𝑌2

The correlation factor described in equation (3.3) can be obtained by normalizing the covariance
matrix with the standard deviation (𝜎) of the two data sets, leading to the so-called correlation
matrix. The elements of the correlation matrix are described as follows:

𝐶𝑖𝑗 𝐸[(𝑋𝑖 − 𝐸[𝑋𝑖 ])(𝑋𝑗 − 𝐸[𝑋𝑗 ])]


𝐶𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝑚𝑎𝑡𝑟𝑖𝑥 = = , (3.5)
𝜎𝑋𝑖 𝜎𝑋𝑗 𝜎𝑋𝑖 𝜎𝑋𝑗

where the correlation factor for the data set in Figure 3.2 is obtained as -0.7086. Since a correlation
between the two random variables exists, the two data set cannot be assumed to be uncorrelated. A
statistical distribution describing the fragmentation of the grenade shell must take this correlation
into consideration if a physical realistic safety distance is to be obtained. If the two variables was
considered uncorrelated, it would exist some probability for a heavy fragment having high velocity.
This would over predict the safety distances of the grenades.

35
For normal distributions there is a method to sample data from a correlated 2 dimensional
distribution. This method is called a bi- or multivariate normal distribution [29] [30]. This method
is described in the next following section.

Figure 3.5. Mass versus velocity for grenade 2, (a) experimentally obtained data and (b) the natural logarithm of the experimentally obtained
data. The data set comes from on specific sector from the test, see Figure 3.1.

Figure 3.4. Mass versus velocity for grenade 3, (a) experimentally obtained data and (b) the natural logarithm of the experimentally obtained
data. The data set comes from on specific sector from the test, see Figure 3.1.

36
Bivariate distribution
Suppose X is a standard normal distributed variable. The PDF of such variable is described by the
equation below:

1 1 (3.6)
𝑓(𝑋) = exp (− 𝑥 2 ) .
√2𝜋 2

Y is also a standard normal distributed variable independent of X. The joint distribution of these
two independent variables is easily described by the following expression:

1 1 (3.7)
𝑓(𝑋, 𝑌) = exp (− (𝑥 2 + 𝑦 2 )).
2𝜋 2

The data in Figure 3.2 consists of two correlated normal distributed random variables. The two
variables X and Y have a set of parameters describing their distributions those parameters are
described as follows:

parameters − {𝜇1 , 𝜇2 , 𝜎1 , 𝜎2 , 𝜌}, (3.8)

where 𝜇1 is the expectation value of X and 𝜇2 for Y. 𝜎 is the standard diviation and 𝜌 the
correlation factor between the two random normal variables. If X and Y is uncorrelated, 𝜌 is zero.
The two independent random variables X and Y can be linearly transformed into dependent
random variables, 𝑍1 and 𝑍2 using the parameters in equation (3.8). In this transformation, 𝜌 is
chosen as the correlation wanted. The linear transformation L used to transform the uncorrelated
variables into two correlated variables uses the Cholesky decomposition of the covariance matrix
𝐶 described in equation (3.4). Cholesky decomposition is described as follows:

𝐶 = 𝐿𝐿𝑇 , (3.9)

where 𝐶 is the covariance matrix described in equation (3.4). The linear transformation is
described as follows:

𝑍1 𝜎1 0 𝑋 𝜇1 (3.10)
[ ]= [ 1 ] [ ] + [ ].
𝜇
𝑍2 𝜎2 𝜌 𝜎2 (1 − 𝜌 )2 𝑌
2 2

37
From equation (3.10) and (3.7) it is possible to calculate the joint PDF of the two random
variables, this calculation is done in Appendix A. In the same way it is possible to transform two
correlated random variables (like the mass and velocity obtained in the experiment) into two
uncorrelated random variables. When two uncorrelated random variables are obtained it is easy to
sample data from the two uncorrelated distributions.

The correlated sets of data obtained in the experiment can be transformed by equation (3.1) and
(3.10) into two independent normally distributed random variables [31]. When independent, a
larger set of data can be sampled and then transformed back using the inverse transform to
equation (3.10). This process is described in more detail in Appendix B. When the data is
transformed back to its originally correlated state, the data can be analyzed. This is done for the
data in Figure 3.2 and the sampled data from the bivariate distribution is showed in Figure 3.6.

Figure 3.6. Data created from the log-normal distribution. Represented in (a) linear scale and b) in log scale.

The obtained sampled data for grenade 1 is then compared with the experimentally obtained data
for grenade 1 in Figure 3.8. The red circles are the data sampled from the bivariate distribution
and the blue circles are the experimentally obtained data. How well the bivariate approach fits the
experimentally obtained data is analyzed by a convergence-test of the correlation factor. If the
sampled data size goes to infinity, the correlation factor should go towards the correlation factor
of the experimentally obtained data in Figure 3.1. The same convergence-test is conducted for the
mean values of mass and velocity in Figure 3.9.

38
Figure 3.8. Comparison of sampled data from bivariate distribution (red dots)
and the experimentally obtained data.

Figure 3.7. Convergence test of the correlation factor with increasing


size of the sampled data. x-axis is the logarithm of the sampling size.

39
Figure 3.9. Convergence test of the mean values for mass and velocity.
Blue line shows the convergence of mass (left axis), and red line shows
the convergence of velocity (right axis).

The convergence-tests show that the sampled data has similar features as the experimentally
obtained data. The small error, the small differences between the data sets, might be because the
experimentally obtained data is not a perfect log-normal distribution. From the converges-tests,
the bivariate approach can be concluded to work properly and give a satisfying result.

40
Engineered fragments

The engineered or pre-fragmentized fragments are unlike the natural fragments well-defined
objects. For grenade 1 they are spherical balls made of tungsten. They are small and have a mass
of about 150 mg and their density are 17 600 kg/m3 . Due to variation in manufacturing, some
variation in the mass of the fragment is found. The obtained data for the mass of the engineered
fragments is find to be normally distributed. Furthermore, the velocity also seems to be normally
distributed. Also for the engineered fragments, artifacts from the layered method of measure are
clearly visible. In Figure 3.10 the experimental data for the engineered fragments is presented. The
data are obtained from grenade 1.

Figure 3.10. Mass versus velocity for engineered fragments. This data is obtained in the experiment for the same sector as the
data in Figure 3.2. Mass versus velocity for (a) experimentally obtained data and (b) the natural logarithm of the experimentally
obtained data. The data set comes from on specific sector from the test, see Figure 3.1..

The correlation factor of mass and velocity of the engineered fragments, with and without the
fragments with initial velocity below 350 m/s are described in the following equations:

0.2335 With the lowest velocity fragments, (3.11)


𝜌= {
0.0392 Without the lowest velocity fragments.
In the case when all fragments are considered, mass and velocity show a strong correlation. If
removing the fragments with initial velocity lower than 350 m/s, the data set is almost uncorrelated.
This strongly suggests that the fragments with velocity lower than 350 m/s are in fact ricochets
and the engineered fragments show no correlation between mass and velocity. Ricochets are
fragments that have bounced off the ground or another object before hitting the witness package.

41
Mass and velocity are for the engineered fragments approximated as two independent random
variables. Grenade 2 and 3 are not designed with additional engineered fragments, meaning that
grenade 1 is the only grenade used in this thesis, which provides any information of the distribution
of engineered fragments.

Since the two distributions are uncorrelated it is convenient to treat the engineered fragments
differently compared to the natural fragments. For the natural fragments, a large number of
elements (fragments) will be sampled from the bivariate distribution. The fragments are then
evaluated separately making them computationally heavy.

For the engineered fragments, no elements (fragments) needs to be sampled. Instead the CDF of
the mass distribution and velocity distribution can be used to evaluate the probability for fragments
having a velocity higher than the 𝑣50 limit of the injury models. The mass axis is divided into
intervals, as shown in Figure 3.11. Smaller mass interval will increase the accuracy of the model.
In Figure 3.11 𝜇 is the mean value of the mass and 𝜎 the standard deviation. The probability of a
fragment having a mass within each of these mass intervals can be calculated using the difference
in the CDF.

𝑚
̅ 1 ≤ 𝑚𝑒𝑛𝑔𝑖𝑛𝑒𝑒𝑟𝑒𝑑 ≤ 𝑚
̅2 (3.12)

As stated in section 2.5.2 it is considered an acceptable risk when one fragment in a million shot
has a velocity higher than the 𝑣50 limit of the different models of injury. To be able to evaluate that
property, the number of fragments per fired shot must be counted for each sector. For this a spatial
distribution of each grenade must be obtained.

42
Figure 3.11. Mass interval for the distribution of engineered fragments. The curve
is arbitrary and do not share any features with the engineered fragments. This
figure shows how the mass interval can be divided into smaller intervals to simplify
the calculations.

43
Spatial distribution

The distribution of fragments around the detonation can be described either as a continuous
distribution or by a discrete distribution. The tests conducted by SAAB will give the spatial
distribution sector by sector. For the purpose of safety distance evaluation it is more useful to
continue to work with sectors. A sector-wise safety distance will also be more convenient to
implement for the users. The spatial distribution is easily obtained by the experiment. When the
witness packages is disassembled, the number of fragments found within each sector is counted.
The relative likelihood (the PDF) for each sector is evaluated by counting the number of fragments
per sector and divide it with the total number of fragments collected for all sectors. Due to the
secrecy, the results in Figure 3.12 are just a fictitious results that have similarities to the real results.
The sectors in Figure 3.12 are numbered from 1 to 8 corresponding to the sectors in the
experimental setup in Figure 3.1.

Figure 3.12. An arbitrary spatial distribution, fragments per sector.

44
4 Modelling
In this section the information discussed in the last two chapter will be brought together to develop
a model for risk evaluation. The model will solve each of the smaller problems shown in Figure
2.1. The program is written in MATLAB and the statistical toolbox is needed to run the program.
Some equation in this section are therefore written as the syntax in MATLAB. The model will
sector-wise evaluate the safety distances according to the different injury models (FOI, Lewis and
Greenbook).

Engineered fragments have a well-defined spherical shape, while the natural fragments are
approximated to have a cubic shape. Since fragments have the tendency to tumble their projected
area will be the average projected area of a tumbling body. The average projected area for the
engineered and natural fragments is described in the equation below. The radius and sides of the
fragments are calculated using the fragment density, geometry and the mass obtained by the
distributions of fragments, described in section 3, leading to equation (4.2).

𝐴𝑝 = 𝜋𝑟 2 → 𝑓𝑜𝑟 𝑎 𝑠𝑝ℎ𝑒𝑟𝑒
3 (4.1)
𝐴𝑝 = 𝑎2 → 𝑓𝑜𝑟 𝑎 𝑐𝑢𝑏𝑒
2

3 𝑚𝑓 1/3
𝑟= ( )
4𝜋 𝜌
1 (4.2)
𝑚𝑓 3
𝑎=( )
𝜌

The natural fragments are made of aluminum with a density of 2700 kg/m3 and the engineered
fragments are made of tungsten with a density of 17 600 kg/m3 . The drag force coefficient depends
not only on the shape but also on the velocity and the Reynolds number. The Reynolds number for
an object in an air flow can be evaluated using equation (2.13) and (2.14). The characteristic length
in equation (2.13) is the diameter of the sphere and the side of a cube, which is shown in equation
. In Table 6 the Reynolds number for an engineered spherical fragment moving 100 and 1000 m/s
and the Reynolds number for a natural cubic fragments moving 100 and 7000 m/s are presented.
The lower limit is estimated by the 𝑣50 limit of the injury models. Below 100 m/s most fragments
are regarded as harmless. The upper limit on the other hand is estimated by the velocity of the blast

45
wave. The fragments are accelerated by the blast wave and can therefore not reach a higher initial
velocity than the velocity of the blast wave. The 1000 m/s for the engineered fragments comes
from the experimental results, see Figure 3.10.

Table 6. Reynolds numbers for a sphere and a cube in an air flow. The air temperature is assumed to be about 20°C.
The diameter of a sphere and the side of a cube used for the average engineered and natural fragments respectively
are: 𝑑𝑖𝑎𝑚𝑒𝑡𝑒𝑟 = 0.00246 𝑚 and 𝑠𝑖𝑑𝑒 = 0.00378. The dynamic viscosity of air at 20° is 1.8 ∗ 10−5 𝑃𝑎 ∗ 𝑠

Sphere Cube
100 𝑚/𝑠 𝑅𝑒 = 1.3 ∗ 104 100 𝑚/𝑠 𝑅𝑒 = 2.0 ∗ 104
1000 𝑚/𝑠 𝑅𝑒 = 1.3 ∗ 105 7000 𝑚/𝑠 𝑅𝑒 = 1.4 ∗ 106

The engineered spheres have a Reynolds number within 104 and 105 which is a regime where the
drag force is independent of the Reynolds number, see Figure 2.13.

Studies conducted on how the drag force coefficient of a cube depends on the Reynolds number,
do also show a plateau like the one in Figure 2.13. For cubic fragments, the plateau is longer than
for the spherical fragments [32]. The drag force coefficient is therefore assumed to be independent
of the Reynolds number for the cube as well.

The velocity dependence of the coefficient is presented in Table 7 for the spherical fragments. The
coefficient for a tumbling cube has the same tendency as for the spherical but shifted upwards, to
higher values. The velocity dependency of drag force coefficient for a tumbling cubic fragment is
shown in Table 8 [18] [33].

46
Table 7. Drag force coefficient for a spherical
fragment at different velocities [34].

Mach number Cd

0.00 0.46
0.30 0.48
0.51 0.50
0.77 0.62
1.02 0.80
1.28 0.96
1.54 1.00
2.05 1.02
3.07 0.98
4.31 0.92
9.22 0.92

Table 8. Drag force coefficient of a tumbling cube. [33]

Mach number Drag force coefficient 𝐶𝑑


0.5 0.82
~1.25 1.25
3.5 1.1
From the experiment conducted by SAAB and explained in section 3.1 an excel file was created.
The model then takes this excel file and uses functions in MATLAB developed by SAAB to
estimate the impact velocity of the fragments. Those functions uses Rilbe theory for multi- and
single plate perforation. The initial velocity 𝑣0 is then calculated using ballistics as follows:

1 (4.3)
𝑣0 = 𝑣𝑖𝑚𝑝𝑎𝑐𝑡 exp ( 𝐶 𝐴𝜌𝑑) ,
2𝑚𝑓 𝑑

where d is the distance between the wooden target and the witness packs in Figure 3.1, 𝑚𝑓 is the
fragment mass, 𝐴𝑝 the average projected area, d the distance between detonation and the witness
package and 𝜌 the density of the medium. 𝜌 is evaluated by equation (2.15) for the specific
temperature and atmospheric pressure. The initial velocity is calculated for natural fragments, and
for the engineered fragments in the case of the grenades designed with pre-fragmentized fragments.

47
For the natural fragments a bivariate lognormal distribution of the masses and initial velocities
obtained by the experiment and equation (4.3) is created, as described in section 3.2.1. From the
distribution a large number of new fragments is sampled. The number of fragments sampled is an
integer multiple of the number of fragments per shot in the sector of interest the model will therefore
simulate a number of shots and not a number of fragments. More sampled fragments will increase
the accuracy of the model, but will also increase the time it takes to excavate the code.

The average projected area is calculated for each of the sampled fragments, according to equations
(4.1) and (4.2). The model then creates a list containing three columns: the mass of the fragment,
the initial velocity of the fragment and the average projected area of the fragment, each row
describing one fragment. In Table 9 the list structure created by the program is shown.

Table 9. The structure of the list created for each sampled fragment.

Fragments mass [𝑚𝑔] Fragment initial velocity [𝑚/𝑠] Average projected area
Fragment 1 75 5000 1.37 ∗ 10−5
Fragment 2 600 1500 5.5 ∗ 10−5
⋮ ⋮ ⋮ ⋮

The velocity at distance d from the detonation is evaluated by equation (2.22). At each value of d,
the probability of skin perforation according to Lewis is evaluated. If the probability is higher than
0.5 (50%) the fragment is counted. The number of counted fragments are then divided with the
number of simulated shots, obtaining the number of dangerous fragments per shot at distance 𝑑.
The number of dangerous fragments per shots is then divided with the area of the sector at distance
d. The sector is 4 m2 at 𝑑 = 5m (the distance to the witness packages) and 16 m2 at 𝑑 = 10 m. The
area of the sector at distance 𝑑 is calculated as follows:

𝑑 2 (4.4)
𝐴𝑠𝑒𝑐𝑡𝑜𝑟 (𝑑) = 4 ∗ ( ) .
5

From this calculation the number of dangerous fragments per shot and m2 is obtained. A standing
adult have approximately 1 m2 facing the detonation. The probability for severe injury per shot
according to FOI and Greenbook are obtained in a similar way. For FOI the kinetic energy per cm2
for each fragment is calculated and compared to the limit energies in Table 3. For Greenbook the

48
value of the probit function Y is evaluated for each fragment and if Y is larger than 5 the fragment
is counted.

The velocity and masses of the engineered fragments are uncorrelated and normally distributed,
which is shown in section 3.3. Since they are uncorrelated they can be treated differently compared
to the natural fragments. Mass and velocity is treated as two independent random normally
distributed variable and the mean value and the standard deviation of the two random variables is
evaluated. The mass of the fragments is obtained through the experiment and the initial velocity in
the same way as for the natural fragments. As described in section 3.3, the possible masses in the
distribution are divided into intervals between −3𝜎 and 3𝜎. 3𝜎 is chosen for limit, since 99.7% of
the distribution can be found within this interval.

The probability of an engineered fragment having a mass within one of the mass intervals is
evaluated by the difference between the CDF of the neighboring mass interval boundaries. This is
described in MATLAB syntax in below:

̅, 𝜎𝑚 ) − 𝑛𝑜𝑟𝑚𝑐𝑑𝑓(𝑚(𝑖 − 1), 𝑚
𝑃(𝑚(𝑖 − 1) < 𝑚 ≤ 𝑚(𝑖)) = 𝑛𝑜𝑟𝑚𝑐𝑑𝑓(𝑚(𝑖), 𝑚 ̅, 𝜎𝑚 ), (4.5)

and what the CDF function does is described in section 2.1.2. Equation (4.5) obtains the probability
𝑝(𝑖) of a fragment having a mass with in some of the mass intervals. The number of fragments
within each interval is 𝑃(𝑖) times the number of engineered fragment within the sector.

For each fragment i, the projected area of the fragment is calculated. The projected area is evaluated
with equation (4.2).

and (4.1). The engineered fragments are as mentioned made of tungsten. The density of the tungsten
alloy used is 17 600 kg/m3. Depending on the model of interest, a limit velocity for the specific
fragment i is evaluated. For FOI model, the limit velocity is calculated using equation (2.29) and
for Lewis the following equation is used:

10𝐴 −𝑎
𝑣50 = √ exp ( ). (4.6)
𝑚𝑓 𝑏

49
Greenbook uses a probit function Y described in section 2.4.3. The limit velocity is calculated by
solving the equation Y = 5, where Y is described in equation (2.28). The limit velocity according
to Greenbook is evaluated as follows:

1
1 𝑌 + 29.15 5.115
𝑣50 = ( exp ( )) , 𝑓𝑜𝑟 0.001 > 𝑚𝑓 > 0.1 𝑘𝑔,
𝑚𝑓 2.10
1
(4.7)
1 𝑌 + 17.56 2
𝑣50 =( exp ( )) , 𝑓𝑜𝑟 0.1 > 𝑚𝑓 > 4.5 𝑘𝑔,
2𝑚𝑓 5.30

with 𝑌 = 5. In contrary to the calculations done for the natural fragments, the engineered fragments
are evaluated by answering which initial velocity a fragment must have to obtain the limit velocity
at distance d. The reason for this approach is that the two random variables are uncorrelated. Since
they are uncorrelated, it is possible to use the CDF of both mass and velocity separately, instead of
evolving the probability at 𝑑 = 𝑑1 followed by evolving the probability at 𝑑 = 𝑑2 and so on as for
the natural fragments. The initial velocity necessary for obtaining the limit velocity 𝑣50 at 𝑑1 is
evaluated using the ballistics in equation (4.3) followed by evaluating the necessary initial at 𝑑2
and so on. The probability of a fragment having higher initial velocity than the initial velocity
obtained as described above is evaluated with the CDF as follows:

𝑃(𝑣 > 𝑣𝑖 ) = 1 − 𝑛𝑜𝑟𝑚𝑐𝑑𝑓(𝑣𝑖 , 𝑣̅ , 𝜎𝑣 ). (4.8)

This calculation is done for each interval boundary i and the total probability of an engineered
fragment perforating the skin is then calculated by taking the sum of all mass intervals as follows:

𝑃 = ∑ 𝑃(𝑖) ∗ 𝑃(𝑣 > 𝑣𝑖 ) ∗ 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑟𝑎𝑔𝑚𝑒𝑛𝑡 𝑖𝑛 𝑡ℎ𝑒 𝑠𝑒𝑐𝑡𝑜𝑟 𝑝𝑒𝑟 𝑠ℎ𝑜𝑡, (4.9)
𝑖

where 𝑃(𝑖) is the probability for a fragment having a mass within 𝑚(𝑖 − 1) ≤ 𝑚 ≤ 𝑚(𝑖) and
𝑃(𝑣 > 𝑣𝑖 ) is the probability of worse case or higher initial velocity for that mass interval. The
model then creates an output describing the probability for severe injury at distance d for the three
different models of injury.

In Figure 4.1 a simplified flow chart is show describing the basic operation of the model.

50
Experiment

List of: mass, number of perforated plates and which


sector
Rilbe Theory

List of: mass, impact velocity and which sector

Back track
fragment path
List of: mass, initial velocity and which sector

Evaluate
projected area
List of: mass, initial velocity and projected area (one list
for each sector)
Iteration
Ballistics
𝑑 = 𝑑 + Δ𝑑

Evaluate 𝑣(𝑑)

Lewis Greenbook FOI

Evaluate
𝐴𝑠𝑒𝑐𝑡𝑜𝑟 (𝑑)

Evaluate the
risk

Save the safety


distance

Figure 4.1. Flow chart over the basic processes in the model.

51
Model validation

The model uses a large number of approximations and assumptions. These assumptions are used to
simplify the problem and to make it computational feasible on an ordinary laptop. Some of the
assumptions are quite obvious like neglecting the Magnus effect and Coriolis effect. Other
approximations need some proofs. The model used for ballistics disregards the influence of gravity.
The gravitational force on a 1 gram fragment is approximately 0.01 N. The drag force for the same
fragment moving at 100 m/s would be close to 250 N. The approximation is therefore valid for all
fragment velocities of interest. Including the gravity would increase the breaking force leading to
shorter safety distances.

The model uses two constants for the drag force coefficient, one subsonic and one supersonic. The
supersonic coefficient is chosen to be the drag force coefficient at 343 m/s for the spherical and
cubic fragments, meaning the lowest possible value in the supersonic regime. In the same way the
subsonic coefficient are chosen to be at the lower limit of the velocities of interest. The lower limit
is approximately were the injury models regards most fragments as harmless (approximately at 100
m/s). By choosing the lower limits to the drag force coefficient, the model insure that the safety
distance obtained are longer than the true safety distance.

The model is limited to three degrees of freedom (DOF), namely movements in x, y and z (in a
Cartesian system) a more sophisticated method would include additionally rotation around the three
axis (pitch, yaw and roll), see Figure 2.9. Using a 6-DOF model would make the model more
computationally heavy, without improving the result drastically. A precise result is not needed
when evaluating the safety distance the important part is that the obtained safety distance is longer
than the real value. Some of the features of a 6-DOF model are still included in this model. The
projected area of the fragments is for instance evaluated by calculating the average of a tumbling
convex body and for the cubic natural fragments the drag force coefficient is adjusted for a tumbling
cube.

52
Figure 4.2. Relation between surface area and
volume for different geometric shapes.

The natural fragments are approximated by a cube, since a cube will give the smallest surface area
to volume ratio of all cuboid geometries. The surface area to volume ratio for different geometric
shapes are shown in Figure 4.2. The breaking force of a fragment is proportional to its projected
area so by choosing a minimum point or ideal shape the real safety distance is insured to be longer
than the obtained result. Wind tunnel tests on real artillery fragments also show higher drag force
coefficients than those for a tumbling cube [35] [36], making the cube approximation practically
admissible.

Two concerns for the model are the use of Rilbe theory to evaluate the impact velocity of the
fragments and the size of the data obtained from the experiment. Rilbe theory assumes that the
fragments don’t deform at impact with any of the witness plates. It is unlikely that no deformation
occurs, but the deformation are at least small and would not change the result to the extent that it is
unusable. The bigger concern with the Rilbe theory is its validation for high velocity fragments.
The theoretically model purposed by Rilbe has high precision up to approximately 1500 m/s [11].
This is a conclusion made by both Rilbe and SAAB. Rilbe theory tends to overestimate the impact
velocity needed for the high velocity fragments, making Rilbe possible to use for risk evaluation
purposes.

One more concern is that the statistics of fragment weight and velocities for some sectors are based
on small data sets, leading to limited validation in some directions. The only solution for this is to
fire more shots than three in the experiment, but each test shot is very expensive and in most sectors
three shots are good enough. The sectors with few fragments can for some cases be analyzed
separately with high speed cameras or other experimental methods.

53
5 Results and discussion
Due to secrecy, it is not possible to specify for which sector the results for the grenades are obtained.
Note that the result would be different if another sector for the different grenades would be
evaluated. The difference comes from variation spatial distribution of fragments. In some sectors
there are no engineered fragments which reduces the real risk a lot. The number of natural fragments
per shot does also vary between the sectors. In addition to variation in spatial distribution the mass
and velocity distribution will not be identical in the different sectors.

The safety distance for a grenade is as mentioned the point where the probability of severe injury
reaches an acceptable level of risk according to Table 4.

The risk at distance d for the three different grenades is simulated as described in section 4. First
the results for grenade 1 will be analyzed followed by grenade 2 and 3. The parameters used for all
three simulations are described in Table 10. The air is assumed to have a temperature of 20°𝐶 and a
pressure of 101 325 Pa. In Figure 5.1 the probability for severe injury at distance d is presented for
grenade 1. Notice that the y-axis have a logarithmic scale. 100 000 shots is simulated for each
grenade to obtain the result presented in this section. In Table 11 the safety distances for grenade 1
are presented, natural and engineered fragments separately.

Table 10. Parameters used for the simulations.

Air temperature [𝐾] 293K


Air pressure [𝑃𝑎] 101 325 Pa
Air density [𝑘𝑔/𝑚3 ] 1.2047
Number of shots simulated 100 000

Table 11. Safety distance in meters for one of the sectors for Grenade 1. Safety distance is the distance where the
probability of severe injury reaches an acceptable level.

The Lewis model (Bare The Greenbook FOI


Skin) formula
Natural Fragments 80 85 60
Engineered 240 180 190
Fragments

54
Figure 5.1. The probability of severe injury at distance d per shot for grenade 1. Notice that the y axis
has logarithmic scale.

From the result in Figure 5.1 it is clear that the engineered fragment have the ability to induce
injuries at much longer distances compared to the natural fragments. The three dashed lines follows
at first an identical path, and at approximately 100 m they start to differ. This is due to the fact that
none of the engineered fragments have obtained a low enough velocity to be regarded harmless by
any of the injury models. Before 100 m, the only factor contributing to a reduction in probability
for injury due to engineered fragments is the increasing area, calculated by equation (4.4). After
100 m, the differences in threshold velocity (𝑣50 ) for considering fragments as hazardous start
influencing the risk. Some fragments have at this point reached low enough velocities to be
considered harmless according to the different models.

It is not surprising that the engineered fragments are more dangerous than the natural. The
deceleration of a fragment, disregarding the gravity is described by equation (5.1). From equation
𝐴
(5.1) it is obvious that the deceleration is proportional to (𝑚) and the drag force coefficient. The

drag force coefficient of the spherical fragments are lower than the drag force coefficient for the
cubic tumbling natural fragments. The engineered fragments have better aero-dynamic properties
than the natural fragments. This is one of the reasons the engineered fragments are more dangerous
𝐴
than the natural. (𝑚) is a factor which correlates with the density of the fragments. The much higher

density of tungsten compared to aluminum makes this factor lower, which will contribute to less
breaking force from the medium on the engineered fragments compared to the natural fragments.

55
The natural fragments have higher initial velocity but the much higher breaking force by the
medium make them harmless faster than the engineered fragments. Both the shape and the density
are parameters giving advantages for the engineered fragments compared to the natural fragments.

𝑑𝑣 1 (5.1)
( )=− 𝐶 𝐴𝜌𝑣⃑ 2
𝑑𝑡 2𝑚 𝑑

FOI and Lewis tend to consider fragments of higher density as more dangerous compared to
Greenbook. This can be visualized in Figure 5.1 where the models by FOI and Lewis obtains shorter
safety distances for the aluminum natural fragments compared to the safety distance obtained by
Greenbook model. When the density then increases the models by FOI and Lewis become more
conservative and give much greater safety distances than the Greenbook model. This originates
𝐴
from a scaling factor. FOI and Lewis depends on the factor ( ) which is dependent on the density
𝑚

of the fragments. Greenbook only scales depending on the mass of the fragments. How the factor
𝐴 varies depending of the density of a cubic fragment is show below:

2
𝑚 3
𝐴 =( ) . (5.2)
𝜌

2
𝐴 1 3
This shows that the factor (𝑚) scales as 𝑚−1/3 (𝜌) . An increase in fragment density with
𝐴
preserved fragment mass will decrease the factor (𝑚). For the Lewis model and for the FOI model

would a decrease of this factor lead to higher probability for skin perforation.

Which of the three models that gives the most physically correct answer is difficult to determine
due to the vast variation in experimental setups. John Breeze and JC Clasper have conducted a
literature review over 17 different experimental studies with different experimental setups. They
have concluded that animal skin and PMHS can be used interchangeable with similar results. On
the other hand isolated skin will give different results compared to complete limb. Experiments
conducted on complete limb gives a larger limit velocity for skin perforation than those conducted
on isolated skin [21], which will lead to shorter safety distances.

As mentioned earlier, Greenbook scales with the mass of the fragment and not with the ratio
between projected area and fragment mass. The projected area is without doubt an important

56
parameter for the perforation ability of a fragment. The projected parameter cannot be fully
excluded from an empirical formula. This means that the projected area will be accounted for in
the empirical constants in some way. Greenbook model is based on experiments conducted with
steel fragments and would therefore certainly be valid mostly for steel fragments. This could
possibly lead to an over estimation of the aluminum fragment since their density is lower than the
density of steel. By the same argument, could Greenbook possibly under estimate the safety
distance of the tungsten fragments. This might be the reason why the Greenbook model is giving
the longest safety distance for aluminum fragments and the shortest for the tungsten fragment.

From this argument Lewis model and FOI model would be more realistic, but the obtained result
differs a lot between FOI and Lewis. This might come from some other differences in the
experimental setups. The experiments by Lewis and Coon [4] were conducted on isolated skin
mounted on gelatin. The review article by John Breeze and JC Clasper concluded that experiments
conducted on isolated skin (like in the Sperrazza and Kokinakis model) tend to give lower limit
velocities and higher safety distances [21]. The effect of the gelatin used in the experiments by
Lewis can be approximated by comparing the limit velocities of the Lewis model with Sperrazza
and Kokinakis formula for ballistic limit, see equation (2.24). In Figure 5.2 the ballistic limit of
bare skin of the two models are compared. The Lewis model gives similar results as Sperrazza and
Kokinakis. John Breeze and JC Clasper stated that Sperrazza and Kokinakis is a method giving
large safety distances. The shorter safety distances obtained by FOI might indicate that the data
used to formulate the ballistic limits are based on experiments with complete limbs.

Figure 5.2. Comparison of the different 𝑣50 limit for the Lewis model and the Sperrazza and Kokinakis. The comparison is done for
𝐴
spherical steel fragments. In a) is the v50 plotted against the ( ) factor and in b) the fragment mass.
𝑚

57
The simulations conducted for grenade 2 and 3 have used the same parameters as the ones described
in Table 10. These two grenades have one big difference to grenade 1: they are not designed with
any pre-fragmentized parts, leading to no engineered fragments. In Table 12 and Table 13 the safety
distances for grenade 2 and 3 are presented.

Table 12. Safety distance for one sector for grenade 2. Safety distance is the distance where the probability of severe
injury reaches an acceptable level.

The Lewis model (Bare The Greenbook FOI


Skin) formula
Natural Fragments 75 80 55
Engineered - - -
Fragments

Figure 5.3. The probability of severe injury at distance d per shot for grenade 2.
Notice that the y axis has logarithmic scale.

Table 13. Safety distance for one sector for grenade 3. Safety distance is the distance where the probability of severe
injury reaches an acceptable level.

The Lewis model (Bare The Greenbook FOI


Skin) formula
Natural Fragments 50 55 40
Engineered - - -
Fragments

58
Figure 5.4. The probability of severe injury at distance d per shot for grenade 3.
Notice that the y axis has logarithmic scale.

The results obtained for grenade 2 and 3 share some features with the result for grenade 1.
Greenbook gives the longest safety distance for the aluminum fragments and FOI the shortest for
the natural fragments. The thing that differs are the obtained safety distance. The fact that the safety
distances between grenades differ is not surprising. The spatial distribution of fragments from a
grenade detonation will depend on a lot of things. The geometry of the grenade together with the
shell thickness will determine the number of fragments found in each sector. Fragments tend to
accelerate outwards in a direction normal to the shell surface [37]. If the grenade would be a
cylinder, most fragments should be found in the region perpendicular to the direction of motion,
see Figure 5.5. A more spherical grenade would create a more even distribution of fragments. The
thickness of the shell also determine the amount of material that could be fragmentized. If the
grenade has local variations of thickness, the spatial distributions of natural fragments will follow
that local variations.

Figure 5.5. Grenade geometry and how the geometry affects the
spatial distribution (the grenades are moving right to left).

59
The type of explosives used in the grenade will also affect the distribution of fragments, such as
the mass and velocity distribution. An explosive with high burning or low burning velocity will
affect the initial velocity of the fragments and also the size of the fragment. Some explosives will
fragmentize the shell in smaller pieces than other types of explosive. The last difference between
the grenades is also the amount of explosives and how the explosives are distributed within the
shell. All these differences between the grenades exist but due so secrecy the exact shape and type
of explosives cannot be disclosed in this thesis.

The model used assumes that the limits for severe injury is constrained by the ballistic limits of
bare skin and neglects the possibility of a fragment hitting the eye of a person. The ballistic limit
of the eye according to K. Sellier and B. Kneubuehl [24] is 3 J/cm2 which is a lot lower than the
ballistic limit of the skin. The area of one eye is approximately 1 square cm, the risk of a fragment
with higher energy than 3 J/cm2 hitting it are lower than the risk of skin perforation. But for longer

Figure 5.6. The risk of severe injury due to low energy fragments hitting the eye for grenade 1, 2 and 3

60
distances from the detonation the biggest risk is blindness from engineered fragments. This is due
to the fact that all fragments have a velocity below the threshold velocity for skin perforation,
leaving blindness as the biggest concern. This is represented in Figure 5.6.

6 Conclusions and Outlook


In this thesis a model for evaluating the risk for sever injury at different distances and direction has
been created. The model uses data obtained from test detonations of some grenade and from this
evaluates the distance where the probability of severe injury is below one in a million. The model
has been tested on three types of grenades obtaining reasonable results for all three. The obtained
variations of safety distances between the three grenades can most likely be described by
differences in shape, shell thickness and design. With design, the type of explosives, how much
and how the explosives are distributed within the grenade shell is taken into consideration.

The obtained safety distances for the different injury models are not in agreement. The Greenbook
model uses an empirical formula which does not take the projected area of the fragments into
consideration. Disregarding the projected area could possibly lead to an overestimation of the safety
distance in this case for the aluminum fragments and an underestimation of the same measure for
the tungsten fragments, but it is difficult to prove.

The Lewis model might also overestimate the safety distances due to the fact that the model is
based on experiment with isolated skin mounted on gelatin. This is proven to give lower ballistic
limit, compare to experiments conducted on the more realistic case of complete limb.

Another important conclusion is that it is okay to disregard the possibility of eye injury when
evaluating the risk down to one in a million. The risk for skin perforation is much higher than the
risk for blindness. If one in ten million or one in hundred million would be the probability of interest
the risk of blindness would need to be regarded. When all fragments have a velocity below the
ballistic limit of skin the probability for skin perforation is zero since the eye is more sensitive than
the skin and can still be injured by those low velocity fragments.

FOI does not fully describe the testing conditions for which the data they based their limits on are
obtained. It is therefore rather difficult to draw any valuable conclusions about the model presented
by FOI.

61
As mentioned in section 0 there are questions to how well the Rilbe theory can describe the high
velocity fragments. Additional experiments and test on this subject could further enhance the
accuracy and reliability of the model and the results obtained by the model.

How the experiment described in section 3.1 is conducted can enhance the results obtained by the
model. The impact velocity of the fragments become quantized by the layered method of
measuring. If instead of 10 aluminum plates, 20 thinner aluminum plates were used a more
continuous result would be obtained and perhaps a more accurate distribution of velocity.

The model approximates the natural fragments as cubes. If real fragments from the detonation
would be analyzed, another shape might be preferable for describing the fragments. A more
accurate drag force coefficient could in this case be obtained for the average fragment obtained by
the fragmentation of the grenade shell.

The model is still computationally heavy and would need to be improved to be easier to use. One
way of doing this would be to create a two dimensional bivariate CDF as discussed in some
literature [38], and to handle the natural fragments in a similar way as the engineered fragments.
Due to lack of time and experience, this question has not been fully investigated.

This thesis is focused on injuries due to skin perforation. The model could be improved by
introducing ballistic limits on uniform materials, vehicles and building. What is the risk of severe
injury if the person is sitting in a vehicle or is wearing a winter uniform? All these improvements
should make the model more accurate and also more effective to use.

62
7 References

[1] R. Lechner, G. Achatz, T. Hauer, H.-G. Palm, A. Leiber and C. Willy, "Verletzungsmuster
und -ursachen in modernen Kriegen," Springer Link, Ulm, Germany, 2010.

[2] G. Solomos, M. Larcher, G. Valsarmos, V. Karlos and F. Casadei, A survey of


computational models for blast induced human injuries for security and defence
applications, Luxenburg: Joint research centre JRC, 2020.

[3] R. H, "Schiesslehre für Infatirie," 1906.

[4] J. Lewis, P. Coon, V. Clare and L. Sturdivan, "An Emperical/Mathematical Model to


Estimate the Probability of Skin Penetration by Various Projectiles," U.S Army Armament
Research & Development Command, 1978.

[5] G. B. Commitee for Prevention of Disasters, "Methods for the Determination of Possible
Damage to People and Objects Resulting from Releases of Hazardous Materials," 1992.

[6] B. Janzon, "Grundläggande stridsdelsfysik Rapport C 20261-D4," FOA , Stockholm, 1978.

[7] H. L. Brode, "Numerical solutions of spherical blast waves," Appl Phys, 1955.

[8] B. Dobratz and P. Crawford, "LLNT Explosives handbook - Properties of chemical


explosives and explosive simulants," Lawrence Livermore National laboratory, University
of Califonia, Livermore, USA, 1985.

[9] V. Karlos, G. Solomos and M. Larcher, "Analysis of blast parameters in the near-field for
spherical free-air explosions," Joint research centre JRC, Luxenburg, 2016.

[10] J. F. Moxner and S. Børve, "Simulation of natrual fragmentation of rings cut from
warheads," Elsvier, 2015.

[11] U. Rilbe, "Splitters genomslagsförmåga av numeriska värden grundade på försök och


beräkningar (FOA A 2525-44)," FOA (FOI), Swedish Defence Research Agency,
Stockholm , 1970.

[12] U. Rilbe, "FOA 2 rapport CH 2083-44," FOA, Stockholm, 1969.

[13] D. E. Carlucci and S. S. Jacobson, Ballistics theory and design of guns and ammunition,
Boca Raton: CRC press, 2014.

[14] M. S. Russell, The chemistry of fireworks, London: Royal Society of Chemistry, 2000.

[15] J. Österman and C. Nordling, "Temperature dependent properties of water and air," in
Physics Handbook- for science and engineering, Lund, Studentlitteratur AB, 2020, p. 46.

63
[16] S. C. Chapra, Applied Numerical Methods with MATLAB for Engineers and Scientists,
McGraw-Hill higher education, 2018.

[17] M. P. Juniper, "Handout fluid Mechanics," Matthew Juniper, Cambridge.

[18] J. F. Moxnes, Ø. Frøyland, I. J. Øye, T. I. Brate, E. Friis, G. Ødegårdstuen and T. H. Risdal,


"Projected area and drag force coefficient of high velocity irregular fragment that rotate or
tumble," Defence Technology, pp. 269-280, August 2017.

[19] L. Bowne, E. Fletcher and D. Richmond, Estimate of man´s tolerance to the direct effect of
air blast, Washington .D.C: Defense atomic support agency, 1968.

[20] C. Bass , K. Rafaels and R. Salzar , Pulmonary injury risk assessment for short-duration
blasts, Journal of Trauma, 2008.

[21] J. Breeze and J. Clasper, "Determining the velocity required for skin perforation by
fragment simulating projectiles: a systematic review," JRAMC, 2013.

[22] J. Sperrazza and W. Kokinakis, "Ballistic limits of tissue and clothing," Ann N Y acad Sci,
1968.

[23] J. Baker, P. Cox, P. Westine and e,a, Explosion Hazards and Evaluation, Elsevier scientific
publishing company, 1983.

[24] K. Sellier and B. Kneubuehl, Wound Ballistics and the scientific background, Elsevier,
1994.

[25] J. Jussila, "Wound ballistics simulation: Assessment of the legitimacy of law enforcement
firearms ammunition by means of wound ballistics simulations," 2005.

[26] S. Harling, A.-L. Berg, M. Karlsson, P. Magnusson and M. Hartman, "Metodik för
riskbedömning," FOI Swedish defence research agency, Stockholm, 2010.

[27] D. Du Bois and E. Du Bois, "Clinical calorimetry. X. A formula to estimate approximate


surface area if height and weight be known.," Arch. Int. Med, 1916.

[28] "scb.se," Statistiska centralbyrån, 09 10 2018. [Online]. Available: https://www.scb.se/hitta-


statistik/artiklar/2018/varannan-svensk-har-overvikt-eller-fetma/. [Accessed 21 04 2021].

[29] C. Rundel, "stat.duke.edu," Duke University, 11 April 2012. [Online]. Available:


https://www2.stat.duke.edu/courses/Spring12/sta104.1/Lectures/Lec22.pdf. [Accessed 24
Mars 2021].

[30] F. Bijma, M. Jonker and A. van der Vaart, An introduction to matematical statistics,
Amsterdam, Netherlands : Amsterdam University Press, 2017.

[31] M. H. DeGroot, "Probability and statistics," Pearson, 1975.

64
[32] A. Höltzer and M. Sommerfeld, New simple correlation formula for the drag coefficientof
non-spherical particles, Halle (Saale) Germany: Elsevier, 2007.

[33] G. E. Hansche and J. S. Rinehart, Air Drag on Cubes at Mach numbers 0.5 to 3.5, J
aeronautical Sci, 1952.

[34] Private communication with SAAB. [Interview]. feb 2021.

[35] M. C. Miller, "Drag force coefficient measurements for typical bomb and projectile
fragments," U. S. Army Research, Development and Engineering Center, Aberdeen, USA,
1990.

[36] A. Catovic, B. Zecevic, S. Serdarevic Kadic and J. Terzic, "Numerical simulations for
prediction of areodynamic drag on high velocity fragmants from natrually fragmenting high
explosive warheads," New Trends in Research of Energetic Materials”, Part II, pp. 475-484,
Sarajevo, Bosnia and Herzegovina, 2012.

[37] Private comunication with SAAB. [Interview]. May 2021.

[38] C. R. Bhat, "New matrix-based method for analytic evaluation of the multivariate
cumulative normal distribution function," Elsevier, Austin, Texas, 2018.

[39] A. L. Cauchy, Note sur divers theoremes relatifs a la rectification des courbes et a la
quadrature des surfaces, Paris: C.R. Acad. Sci, 1841.

[40] D. Neades and R. Rudolph, "An Examination of Injury Criteria for Potential Application to
Explosive Safty Studies,," Houston, Texas, 1984.

[41] D. Tausch , W. Sattler, K. Wehrfritz and et al, "Experiments on the penetration power of
various bullets into skin by and muscle tissue," Z Rechtsmed, 1978.

65
8 Appendix

Appendix A Calculation of PDF for joint distribution

In this appendix the PDF of a joint distribution is derived. The covariance matrix in equation (3.4)
can be decomposed using Cholesky decomposition. Cholesky decomposition is described by
equation
(8.2)

Σ = 𝐿𝐿𝑇 (8.1)

𝜎2 𝜌𝜎𝑋 𝜎𝑌 𝐿11 0 𝐿11 𝐿12


Σ=[ 𝑋 2 ] = [𝐿 ][ ]→
𝜌𝜎𝑌 𝜎𝑋 𝜎𝑌 12 𝐿22 0 𝐿22
𝐿211 𝐿11 𝐿12 (8.2)
Σ= [ ]
𝐿11 𝐿12 𝐿212 + 𝐿222

Now, each element can be easily compared, leading to:

Table 14. Evaluation of parameters in equation (8.2)

𝐿11 = 𝜎𝑋
𝐿12 = 𝜌𝜎𝑌
𝐿22 = 𝜎𝑌 (1 − 𝜌2 )1/2
𝜎𝑋 0 (8.3)
𝐿=[ ]
𝜌𝜎𝑌 𝜎𝑌 − 𝜌2 )1/2
(1

The joint PDF (𝑓(𝑍1 , 𝑍2 )) is the calculated as follows:

1 (8.4)
𝑓(𝑍1 , 𝑍2 ) = 𝑓(𝑋, 𝑌),
det(𝐴Σ)

66
where 𝑓(𝑋, 𝑌) is described by equation (3.7). X and Y is calculated by the inverse linear
𝑍1 −𝜇1
transformed in equation (3.10). Using equation (3.7) and substitution X for 𝜎1
and Y for

1 𝑍2 −𝜇2 𝑍1 −𝜇1
1 ( 𝜎2
+𝜌( 𝜎1
)) leads to the following expression for the PDF.
(1−𝜌2 )2

1 1 𝑍1 − 𝜇1 2
𝑓(𝑍1 , 𝑍2 ) = 1 exp [− (( )
2 )2 2 𝜎1
2𝜋𝜎1 𝜎2 (1 − 𝜌
2 (8.5)
1 𝑍2 − 𝜇2
+ (( )
( 1 − 𝜌2 ) 𝜎2
2
𝑍2 − 𝜇2 𝑍1 − 𝜇1 2
𝑍1 − 𝜇1
+2𝜌 ( )( )+𝜌 ( ) )]
𝜎2 𝜎1 𝜎1

1 1 𝑍2 − 𝜇2 2
𝑓(𝑍1 , 𝑍2 ) = 1 exp [− (( ) ]
2(1 − 𝜌2 ) 𝜎2
2𝜋𝜎1 𝜎2 (1 − 𝜌2 )2 (8.6)
𝑍2 − 𝜇2 𝑍1 − 𝜇1 𝑍2 − 𝜇2 2
+2𝜌 ( )( )+( ) )]
𝜎2 𝜎1 𝜎2

67
Appendix B Calculate two independent normal distributions

In this appendix the transformation of two correlated log-normal distribution into two independent
normal distributions is derived. 𝑍1 and 𝑍2 are the two correlated set of data, mass end velocity,
obtained in the experiment. 𝑍1 and 𝑍2 is log-normally distributed. With equation (8.7), 𝑍1 and 𝑍2
is transformed into two correlated normally distributed set of data.

𝑍 = ln(𝑍1) (8.7)
{ 1
𝑍2 = ln(𝑍2)

By the linear transformation in equation (8.8), two independent normally distributed data set X and
Y can be obtained. In equation (8.8), 𝜎 is the standard deviation and 𝜇 the expectation value of Z,
𝜌 is the correlation factor. Two normal distributions are created from the set and a large number of
data is extracted from the distributions. The extracted data is then transformed back to two
correlated log-normal distributions.

𝑍1 = 𝐿𝑋 + 𝜇1 (8.8)
{
𝑍2 = 𝐿𝑌 + 𝜇2

L is a linear transformation.

𝐿𝑇 𝐿 = Σ, Σ is the covariance matrix (8.9)

𝐿−1 (𝑍1 − 𝜇1 ) = 𝑋 (8.10)


{
𝐿−1 (𝑍2 − 𝜇2 ) = 𝑋

The two independent normal distributed random variables X and Y can be described by the
following distributions. From these distributions a large number of fragments can be sampled and
transformed into the real log-normal distribution by reversed steps.

𝑋~𝑁(𝑋̅, 𝜎𝑋 ) (8.11)
{
𝑌 ~𝑁(𝑌̅, 𝜎𝑌 )

68
Appendix C Solution to ODE in equation (2.15)

In this appendix the ODE in equation (2.21) is derived. Equation (2.21) is a separable ordinary
differential equation.
𝑑𝑣 1 (8.12)
𝑚( ) = − 𝐶𝑑 𝐴𝜌𝑣⃑ 2
𝑑𝑡 2

𝑑𝑣 1 (8.13)
=− 𝐶 𝐴𝜌𝑣 ∗ 𝑑𝑡 →
𝑣 2𝑚 𝑑
𝑑𝑣 1
∫ = ∫− 𝐶 𝐴𝜌𝑣 ∗ 𝑑𝑡 =
𝑣 2𝑚 𝑑
1
ln 𝑣 = − 𝐶 𝐴𝜌 ∗ 𝑑 + 𝐶
2𝑚 𝑑

𝑑 is the displacement ∫ 𝑣 𝑑𝑡 = 𝑑.
1 (8.14)
𝑣(𝑠) = exp(𝐶) exp (− 𝐶 𝐴𝜌 ∗ 𝑑)
2𝑚 𝑑

Initial condition is that at 𝑑 = 0 is 𝑣 = 𝑣0 , leading to exp(𝐶) = 𝑣0 .


1 (8.15)
𝑣(𝑠) = v0 exp (− 𝐶 𝐴𝜌 ∗ 𝑑)
2𝑚 𝑑

69
Appendix D Expectation value of the projected area for tumbling convex
body

In this appendix the average projected area of a tumbling convex body is derived. In the 19th century
Augustin Louis Cauchy proved that the average projected area of a convex body is 1/4 of the surface
area [39]. A rotation can be fully described by one angle and one axis of rotation. A arbitrary
rotation can therefore be described by the arbitrary angle 𝜃 and an arbitrary axis of rotation 𝑢
⃑⃑.

𝑢𝑥
⃑⃑ = [𝑢𝑦 ]
𝑢
𝑢𝑧 (8.16)

Imagine an arbitrary vector 𝑣⃑.

𝑣𝑥
𝑣⃑ = [𝑣𝑦 ]
𝑣𝑧 (8.17)

Let 𝑣⃑ rotate an angle 𝜃 around axis 𝑢


⃑⃑, is in Figure 8.1. A schematic description of an arbitrary

Figure 8.1. A schematic description of an arbitrary rotation and definition of different vectors and
coordinates.

rotation and definition of different vectors and coordinates.

70
Instead of using Cartesian coordinates, the space is spanned by the octagonal normalized axis
(𝑢
⃑⃑, 𝑣⃑⊥ , 𝑤). Vector 𝑣⃑ is in this coordinate system described as follows:

𝑣⃑ = 𝑣∥ + 𝑣⊥ . (8.18)

𝑣∥ is parallel to the axis of rotation. All notations are also described in Figure 8.1. A schematic
description of an arbitrary rotation and definition of different vectors and coordinates.

. A rotation is a linear transformation. The transformation is called T, and will satisfy the condition
in equation (8.19).

𝑇(𝑣) = 𝑣´ (8.19)

The transformation will not act on the part of 𝑣 that is parallel to the axis of rotation.

𝑇(𝑣
⃑⃑⃑⃑∥ + ⃑⃑⃑⃑⃑)
𝑣⊥ = ⃑⃑⃑⃑
𝑣∥ + 𝑇(𝑣
⃑⃑⃑⃑⃑)
⊥ =𝑣⃑´ (8.20)

𝑣´ can be evaluated from Figure 8.1. A schematic description of an arbitrary rotation and definition
of different vectors and coordinates.

𝑣⃑´ = ⃑⃑⃑⃑
𝑣∥ + ⃑⃑⃑⃑⃑
𝑣⊥ cos(𝜃) + 𝑤
⃑⃑⃑ sin(𝜃) (8.21)

With dot product and cross product 𝑣∥ , 𝑣⊥ and 𝑤 can be expressed by 𝑢


⃑⃑ and 𝑣⃑.

𝑣∥ = (𝑢
⃑⃑⃑⃑ ⃑⃑ ⋅ 𝑣⃑)𝑢
⃑⃑
⃑⃑⃑ = (𝑢
𝑤 ⃑⃑ × 𝑣⃑⊥ ) = (𝑢
⃑⃑ × 𝑣⃑) (8.22)
𝑣⊥ = 𝑣⃑ − 𝑣∥ = 𝑣⃑ − (𝑢
⃑⃑⃑⃑⃑ ⃑⃑⃑⃑⃑⋅ 𝑣⃑)𝑢
⃑⃑

The expression in equation (8.21) can now be rewritten using the substitutions in equation (8.22),
giving:

71
𝑇(𝑣⃑) = 𝑣´ = (𝑢 ⃑⃑ + (𝑣⃑ − (𝑢
⃑⃑ ⋅ 𝑣⃑)𝑢 ⃑⃑) cos(𝜃) + (𝑢
⃑⃑ ⋅ 𝑣⃑)𝑢 ⃑⃑ × 𝑣⃑) sin(𝜃) =
𝑣⃑ cos(𝜃) + (1 − cos(𝜃))(𝑢 ⃑⃑ + (𝑢
⃑⃑ ⋅ 𝑣⃑)𝑢 ⃑⃑ × 𝑣⃑) sin(𝜃) (8.23)

Now consider an surface object 𝑑𝑆 spanned by unit vectors 𝑒̂1 and 𝑒̂2 .

1 0
𝑒̂1 = [0] , 𝑒̂2 = [1]
(8.24)
0 0

The transformation 𝑇 is now acting on the two vectors spanning surface object 𝑑𝑆, giving 𝑒̂1 ´ and
𝑒̂2 ´.

cos(𝜃) + (1 − cos(𝜃))𝑢𝑥2
𝑒̂1 ´ = [sin(𝜃)𝑢𝑧 + (1 − cos(𝜃)) 𝑢𝑥 𝑢𝑦 ] (8.25)
sin(𝜃) 𝑢𝑦 + (1 − cos(𝜃))𝑢𝑦 𝑢𝑧

−sin(𝜃)𝑢𝑧 + (1 − cos(𝜃)) 𝑢𝑥 𝑢𝑦
𝑒̂2 ´ = [ cos(𝜃) + (1 − cos(𝜃))𝑢𝑦2 ] (8.26)
− sin(𝜃) 𝑢𝑥 + (1 − cos(𝜃))𝑢𝑦 𝑢𝑧

The projected area of 𝑒̂1 ´ and 𝑒̂2 ´ on the xy-plane is the cross product between the x and y
components of 𝑒̂1 ´ and 𝑒̂2 ´.

𝑖 𝑗 𝑘
𝐴𝑝 = [ cos(𝜃) + (1 − cos(𝜃))𝑢𝑥2 sin(𝜃)𝑢𝑧 + (1 − cos(𝜃)) 𝑢𝑥 𝑢𝑦 0]
−sin(𝜃)𝑢𝑧 + (1 − cos(𝜃)) 𝑢𝑥 𝑢𝑦 cos(𝜃) + (1 − cos(𝜃))𝑢𝑦2 0
𝐴𝑝 = cos2(𝜃) + cos(𝜃)(1 − cos(𝜃))(𝑢𝑥2 + 𝑢𝑦2 ) + (1 − cos(𝜃))2 𝑢𝑥2 𝑢𝑦2 +
sin2(𝜃) + sin(𝜃) 𝑢𝑧 (1 − cos(𝜃))𝑢𝑥 𝑢𝑦 − sin(𝜃) 𝑢𝑧 (1 − cos(𝜃))𝑢𝑥 𝑢𝑦 − (8.27)
(1 − cos(𝜃))2 𝑢𝑥2 𝑢𝑦2
𝐴𝑝 = cos(𝜃) + (1 − cos(𝜃))𝑢𝑧2

72
The expectation value for the projected area will be zero due to symmetry as half of the projection
will be facing at one direction and the other half the opposite way. The expectation value of the
projected area can be calculated by integration over one of those sides. The negative part of the
projected area is more easily defined that the positive part, see Figure 8.2. Negative half of the
projected area.

. From both Figure 8.2. Negative half of the projected area.

and equation (8.28), it is clear that 𝜃 must be larger than 𝜋/2 radians.

cos(𝜃) (8.28)
𝑢𝑧2 =
(cos(𝜃) − 1)

Figure 8.2. Negative half of the projected area.

73
The expectation value is then calculated by solving the double integral in equation (8.30). Notice
(1−cos(𝜃))
that the density function of 𝜃 is 𝑓(𝜃) = 𝜋
. Expectation value for a continuous case is

calculated by equation (8.29).

(8.29)
𝐸[𝑥] = ∫ 𝑥𝑓(𝑥)𝑑𝑥

𝜋 √2/2 (8.30)
1
𝐸[𝐴𝑝 ] = − ∫ ∫ (1 − cos(𝜃)) ∗ (cos(𝜃) + 𝑢𝑧2 (1 − cos(𝜃))𝑑𝜃𝑑𝑢𝑧
𝜋
𝜋 0
2
𝜋
1 cos(𝜃)
= − ∫(1 − cos(𝜃))√ cos(𝜃)
𝜋 (cos(𝜃) − 1)
𝜋
2

cos(𝜃) cos(𝜃) 1
+ (√ ) (1 − cos(𝜃))𝑑𝜃
(cos(𝜃) − 1) (cos(𝜃) − 1) 3
𝜋
2 1 1
= ∫ √cos3 (𝜃)(cos(𝜃) − 1) 𝑑𝜃 = → 𝑑𝑆
3𝜋 4 4
𝜋
2
From this calculation, the projected area is proven to be ¼ of the surface area of a convex object.
Meaning that the expectation value of the projected area for a cube is 3/2 times the area of one side.

74

You might also like