Engineering, 2010, 2, 257-262
doi:10.4236/eng.2010.24035 Published Online April 2010 (http://www. SciRP.org/journal/eng)
Copyright © 2010 SciRes. ENG
257
Optimum Shape of High Speed Impactor for Concrete
Targets Using PSOA Heuristic
Francesco Ragnedda, Mauro Serra
Department of Mathematics and Computer Science, University of Cagliari;
Department of Structural Engineering, University of Cagliari
E-mail: ragnedda@ unica.it, serrama@unica.it
Received November 16, 2009; revised January 20, 2010; accepted February 4, 2010
Abstract
The present paper deals with the optimum shape design of an absolutely rigid impactor which penetrates into
a semi-infinite concrete shield. The objective function to maximize is the depth of penetration (DOP for short)
of the impactor; in the case of impactors with axisymmetric shapes DOP is calculated using formulas ob-
tained by Ben-Dor et al. [1-3] with the method of local variations [4] and based on the mechanical model
proposed by Forrestal and Tzou [5]. In the present paper we show that using a different class of admissible
functions, more general than the axisymmetric one, better results can be obtained. To solve the formulated
optimization problem we used a custom version of the particle swarm optimization method (briefly denoted
by PSOA), a very recent numerical optimization algorithm of guided random global search. Numerical re-
sults show the optimal shape for various types of shields and corresponding DOP; some Ben-Dor et al. [1-3]
results are compared to solutions obtained.
Keywords: Impactor; Optimization; Particle swarm; Global Search
1. Introduction
A problem important both for the civil world as well as
for military research is to evaluate the depth of penetra-
tion (DOP) of a high-speed impactor when it penetrates a
shield. A quite general solution to such problem was
proposed by Forrestal et al. [5-7] in the case of concrete
targets. Four different models for the shield material are
considered: incompressible elastic-plastic (model 1),
incompressible elastic-cracked-plastic (model 2), com-
pressible elastic-plastic (model 3) and compressible elas-
tic-cracked-plastic (model 4). Supposing the impactor's
shape is axysimmetric with an unknown starting radius
(flat nose), given length and final radius, (we call K0
such class of shapes) and using the Forrestal et al. [5-7]
model, Ben-Dor et al. [1-3,8] we investigated the maxi-
mum depth of penetration and found corresponding nu-
merical solutions. From a mathematical point of view,
the problem is reduced to a non-classical variational one
for a functional that is a function of integrals of the un-
known impactor's shape; the technique used to solve it is
that of local variations, proposed by Banichuk et al.[4].
In the following we shall define a new class of impactors
and show how to get corresponding formulas.
It is worth noting that in the current literature, some
penetration problems for three-dimensional bodies of
optimal shape are also present [9-12] and very recently
Banichuk and Ivanova [13] obtained new results apply-
ing the Forrestal and Tzou model to pyramidal impac-
tors.
2. Model building
Let us consider an impactor whose impact velocity is
normal to the concrete shield and has modulus vimp. The
impactor has a shape built in the following manner: con-
sider a body of revolution obtained revolving a curve of
function y = y(z) in the Cartesian coordinates (Oxy)
around z axis by 2
and let 0, where L is the
length of the impactor; we suppose also that y(z) zL/R
The function y(z) has the unknown value y(0) = y0 for z =
0 and the fixed value R for z = L (Figure 1). Next we cut
such body with two planes passing through y axis and
crossing the zx plane respectively through (0,0), (L,R)
and (0,0), (L,–R). Excluding the two external parts of the
body we get a new non-axisymmetric shape called
zL
F. RAGNEDDA ET AL.
258
R R
z
y
x
D
1
y
1
n
1
1
dS
A
B
C
D
C
1
dS 2
dS
E
F
E
F
F
z
x
A
B
y
2
dS
2
n
2
CD
0
y
0
y
0
y
x
y
R
zzdz
()yz
L
d
d
()
y
z
(a)
(b)
(c)
P
O
ABO O
0
y
2
dS
1
dS
Figure 1. Screwdriver impactor.
“screwdriver shape” or shape of class K1 (Figure 1). In
the shape so obtained we can distinguish two distinct
surfaces: S1, corresponding to the plane part, and S2 cor-
responding to the curved (remaining) one, that is S = S1 +
S2.
Using the model of Forrestal and Tzou (1997) and
Ben-Dor et al. [3] the drag force D of an axisymmetric
striker of class K1 of given length L and final radius R
has two different expressions depending on the current
penetration length h:
if 04
if 4
z
S
hh
DtdS hR

R
(1)
where 0
is an experimental constant; dS is the ele-
mentary surface of the impactor and tz the component
along z of the normal stress σ acting on dS (see Figure 1).
According to Forrestal and Tzou [5], σ is equal to the
following parabolic expression
2
*0 12
*
()( ),m
nn
vAAkvAkv k


 

(2)
Where ρm is the density of the shield, σ* its uniaxial com-
pressive strength and A0, A1, A2 are given positive dimen-
sionless constants shown in Table 1 (see [5]):
The drag force D when (second stage of pene-
tration), is caused by the force D1 applied to the plane
lateral surface S1, and D2, applied to the curved lateral
surface S2 of the impactor; so the total drag force D is
expressed by the following formula
4hR
Table 1. Coefficients for different models of the shield.
Number of
the model Characteristec of the model A
0
A
1
A
2
1 Incompressible,
elastic-plastic 5.18 0.00 3.88
2 Incompressible,
elastic-cracked-plastic 4.05 1.36 3.51
3 Compressible, elastic-plastic4.500.751.29
4 Compressible,
elastic-cracked-plastic 3.45 1.60 1.12
1
DDD
2
(3)
Let us consider now a thin strip of impactor realized
by cutting it with two planes distant one from the other
dz and orthogonal to z axis (see Figure 1). The surface
dS between such planes is decomposed in two parts: dS1
(plane part) shown in dark grey and dS0 (curved part)
showed in light grey. The elementary surface dS1 has the
following expression
1
1
2sin( )
dz
dS d
Where α1 is the angle between the negative direction of z
axis and the outer unit vector 1
n normal to S1 and d is
the segment shown in Figure 1(b) and (c). Taking into
account that tan(α1) = R/L = p we can write
22 2
141( )dSpypz dz  (4)
Whereas from Figure 1 (b) and (c) we have
2
241dSOPy dz

Copyright © 2010 SciRes. ENG
F. RAGNEDDA ET AL.259
where prime denotes derivation with respect to; more-
over
z
()OPy z and asin(/ )pz y
so
2
241 asin
pz
dSy ydz
y

 

(5)
Remembering that only the normal component tz of
is responsible for the drag force, we can write
12
12 1zz
SS
DD DtdStdS 2
(6)
but
11
2
22
2
cos( )on
1
cos( )on
1
z
pS
p
tyS
y


(7)
so
12
12
22
00
4()4asin
zz
SS
LL
DtdStdS
pz
p ypzdzyydz
y



 



(8)
Substituting (2) in (8) we have
22 2
*01 2
0
2
*01 2
0
4(()) ()
4( ())asin
L
nn
L
nn
DAA kvAkvpypzdz
pz
A
A kvAkvyydz
y


 

(9)
where has the expression
n
v
1
2
22
2
cos( )on
1
cos( )on
1
n
p
vv
p
vy
vv
y
1
S
S
(10)
Introducing the following dimensionless variables:
2
*
,,
,,
Dzy
Dzy
LLR
R
yy Vkv
L



 
(11)
(from now on tilde is omitted)
we get the following quadratic expression for D:
2
012
D
aaVaV  (12)
where
000112
1102113
2
2202114
()
(
()
aABIBI
aABBIBI
aABBIBI



02
12
2
2
4
4
1
p
B
B
p
Bp


(14)
1
22
1
0
1
2
0
12
32
0
13
42
0
()
1
1
I
ypzd
Iyydz
yy
z
I
dz
y
yy
I
dz
y

(15)
Now applying the equation of motion of the impactor
(for details see [3] we find the following expression
*
2
01 2
0
1
VV
Pd
aaVaV
 
V
(16)
where
4
DOP
PR
is the dimensionless expression of DOP
3
4
m
Rk
is an dimensionless coefficient
*
V
4h
is the adimensional velocity of the impactor for
R
found imposing that for the expression
of the drag force D in the first stage of penetration has
the same value of D in the second stage of penetration
[3]:
4hR
22
11 02
*
2
4() ()
2( )
aa a Wa
Va
 
(17)
where W = kvimp is the dimensionless impact velocity. Of
course to get real values of the discriminant in (17)
must be nonnegative.
*
V
3. Building a Better Solution
In the case of shape of revolution of class K0 Ben-Dor et
al. [3] found the solution for a large set of problem pa-
rameters and showed that the optimal shape has a flat
nose and is concave. Let us now show with an example
that using a shape of class K1 a better solution than that
corresponding to class K0 is obtained. To begin, we con-
sider the following problem parameters: model number:
3(compressible elastic-plastic shield); ω = 14; τ =0.5;
)
(13)
W = 3.5. The optimum DOP value for a body of class K0
is 14.1P
[3]. Using the same problem parameters, let
Copyright © 2010 SciRes. ENG
F. RAGNEDDA ET AL.
260
us now build a shape of class K1 choosing a generatrix of
the form
() b
y zazc
(18)
where (
Figure 2).
3
0.491;0.38;5.616 10abc

Now, applying Equation (13)-(17) we found that P =
14.4 > 14.1. This is sufficient to show that class K1 can
give better solutions than class K0. We note that several
theoretical and numerical considerations about screw-
driver shapes can be found in [14] and the references
given there, although applied to the Newton problem of
optimal aerodynamic bodies.
4. Formulation of the Optimum Design
Our task is not only to show that class K1 can give better
solutions than class K0 as we have seen in the previous
section, but also to find the generatrix y(z) for which the
corresponding impactor shape belonging to K1 maxi-
mizes DOP:
0
() max
with (0)(free)and(1)(fixed)
y
DOPDOPy
yy y


(19)
We shall show in the following part of the paper one
possible strategy of solution.
5. Psoa Algorithm
To solve this problem we will use the method PSOA
(Particle Swarm Optimization Algorithm), a recent heu-
ristic suitable for finding global optima solutions. PSOA
was introduced about one decade ago [15,16], inspired
by the behaviour of school of fish, flocks of birds or
swarm of bees observed in nature. Like genetic algo-
rithms (GA) and ant colony optimization methods (ACO),
PSOA is also a stochastic, population-based global
search algorithm. In a natural swarm each individual or
particle changes position toward better places, to reach
food or to escape from predators, exchanging informa-
tion with the neighbourhood and without any central
control. In the mathematical model at each individual
corresponds a particle and at every particle i is asso-
ciated a position vector
i
i
x
in a -dimensional space,
to which corresponds a certain value of the objective
function
n
()
i
f
x that represents the quality of i
x
; of
course i
x
should be a feasible solution to the problem
under study . If some constraints must be considered, the
objective function can be modified using the penalty
technique. When the algorithm starts, a collection (swarm)
of particles is randomly chosen using a uniform
probability distribution. Then the vector position of each
particle is updated at every iteration adding a displace
m
00.2 0.4 0. 60. 81
0
0.1
0.2
0.3
0.4
0.5
z
()
y
z
() b
yz az c
3
0.491
0.38
5.616 10
a
b
c

Figure 2. A solution on class K1.
ment based on the information about the previous swarm
positions; the knowledge of the gradient of
f
is not
needed. So the iteration rule is the following
(1)() (1)kkk
iii
xxs
 (20)
Where ()k
i
x
and (1)k
i
x
= positions of particle i at
kth
and 1k
-iteration, (1)k
i
s = displacement of
particle at
i1k
-iteration.The formula for(1)k
i
s
is
(1) ()()()
112 2
()(
kk k
ii iigi
)
k
s
wsc rpscrps
 
(21)
Formula (21) is the core of the algorithm and contains
some fundamental information about the neighbourhood
of particle i: w is called inertia and serves to control the
influence of the previous displacement on the new one.
Recommended values of w range from 0 to 1.4; i
p
is
the best position found for the particle to iteration k,
whereas
i
g
p is the best position inside the swarm up to
iteration k; c
1 and c
2 are two positive constants used to
balance the cognitive and social aspect: usually c1= c2 = 2.
r1 and r2 are random parameters in the range [0,1], sam-
pled from a uniform distribution. The algorithm stops
when a prescribed number of consecutive iterations
without improvement in the objective function is reached.
From (20) and (21) it is clear that PSOA takes account of
the previous random walks of particles. During iterations
typically two main problems can arise: the first is the
explosion of the swarm: it means that the maximum dis-
tance between two particles grows indefinitely and con-
sequently an overflow error occurs; the second is the
stagnation of the swarm which means that all the parti-
cles occupy the same position, leading to a suboptimal-
solution. To try to avoid such inconveniences the two
following additional rules are included in the algorithm:
the first rule is the limitation of the maximum displace-
ment: ()
max
k
i
s
s where smax is a constant which is
Copyright © 2010 SciRes. ENG
F. RAGNEDDA ET AL.261
problem dependent; the second one is called craziness:
with an assigned probability Pcr (for example 0.005), the
displacement (1)k
i
s is not calculated with (21) but com-
pletely randomly, with the only condition that its
modulus must be smax. The PSOA algorithm preposed
has several advantages with respect to classical optimi-
zation algorithms: 1) it is able to find the true optimum
solution in the entire search space (global optimum); 2) it
is easy to implement; 3) it can handle non-differentiable
function and no calculation of derivatives is required.
The main disadvantage is that it can be quite time-
consuming. The PSOA algorithm so far described is only
one among many others possible that can be found in the
current literature [15,17-19]. For the problem under con-
sideration the swarm is formed by m particles and each
particle is a generatrix yi(z) to which corresponds an im-
pactor shape of class K1 and a certain value of the objec-
tive function f (DOP); each generatrix is approximated
by the vector i
ywhere every component of i
y is a
value of yi; the z axis has been discretized into n-1 inter-
vals so we have n points in which to calculate the gen-
eratrix; as a consequence the vector yi belongs to an
n-dimensional space; because we know in advance that
the most critical zone of the impactor shape in near z =
0, the n-1 intervals of z axis are not equals, but shorter
near z = 0 and longer as we go toward z = 1. In our cal-
culations n (which corresponds to the swarm size) has
been fixed to 30. In Figure 3 some generatrices of op-
timum impactors of class K1 are shown using the fol-
lowing common parameters: Model n.3, τ =0.5, ω = 14
for different impactor’s speed W = 1.0, 2.0, 3.0, 3.5.
In Table 2 the comparison of DOP of impactors of
class K1 illustrated in Figure 3, and the corresponding
impactors of class K0 (surfaces of revolution) with the
same problem parameters model n.3, τ =0.5, ω = 14, is
shown for W = 1.0, 2.0, 3.0, 3.5.
Table 2 shows clearly that the classical and intuitive
search space of axisymmetric high speed impactors does
not guarantee the best solution (i.e. the maximum DOP).
In fact in the cases studied show that an impactor shape
belonging to a more general class K1 (non-axisymmetric
shape called “screwdriver shape”), can give better results.
6. Concluding Remarks
Starting from the mechanical model proposed by For-
restal et al. and using formulas obtained by Ben-Dor et al.
for finding the best impactor shape in the class K0 of ax-
isymmetric bodies, we presented a different shape of
impactors in the new class K1 which is different from the
standard axisymmetric shape, and show that bodies be-
longing to class K1 can give better results compared to
corresponding K0 bodies. We also finalized a code based
z
y
00.2 0.40.60.8 1.0
0.1
0.2
0.3
0.4
0.5
model:3
0.5
14
Common
Parameters:
curve 1
W=1.0
curve 2
W=2.0
curve 4
W=3.5
curve 3
W=3.0
Figure 3. Some generatrices of optimum impactors.
Table 2. Comparison of DOP of impactors.
DOP of impactors of class class
K1 with the following common
parameters:
modeln.3, = 0.5, = 14
Corresponding DOP of
impactors of class K0 using
the same parameters:
modeln.3, = 0.5, = 14
Impaactor’s
speed
DOP of impactor
of class K1
Impactor’s
speed
DOP of
impactor of
class K0
W = 1.0
(curve 1 on
figure.3)
2.0 W = 1.0 2.0
W = 2.0
(curve 2 on
figure.3)
6.0 W = 2.0 5.8
W = 3.0
(curve 3 on
figure.3)
11.6 W = 3.0 11.2
W = 3.5
(curve 4 on
figure.3)
14.7 W = 3.5 14.1
on PSOA heuristic algorithm to find the best generatrix
in class K0 once some problem parameters are fixed.
Some results are also given in a tabular form. We must
emphasize that the class K1 is not the most general class
of impactors we can imagine so the problem of the best
impactor remains open.
7. References
[1] G. Ben-Dor, A. Dubinsky and T. Elperin, “Numerical
Solutions for Shape Optimization of an Impactor Pene-
trating into a Semi-infinite Target,” Computers and
Structures, Vol. 81, No. 1, 2003, pp. 9-14.
[2] G. Ben-Dor, A. Dubinsky, T. Elperin, “Shape Optimiza-
tion of an Impactor Penetrating into a Concrete or a
Limestone Target,” International Journal of Solids and
Structures, Vol. 40, No. 17, 2003, pp. 4487-4500.
[3] G. Ben-Dor, A. Dubinsky and T. Elperin, “Modeling of
High-speed Penetration into Concrete Shields and Shape Op-
timization of Impactors,” Mechanics Based Design of Struc-
tures and Machines, Vol. 34, No. 2, 2006, pp. 139-156.
[4] N. V. Banichuk, V. M. Petrov and F. L. Chernousko,
“The Method of Local Variations for Variational Prob-
Copyright © 2010 SciRes. ENG
F. RAGNEDDA ET AL.
Copyright © 2010 SciRes. ENG
262
lems Involving Non-additive Functionals,” USSR Com-
putational Mathematics and Mathematical Physics, Vol.
9, No. 3, 1969, pp. 66-76.
[5] M. J. Forrestal and D. Y. Tzou, “A Spherical Cav-
ity-expansion Penetration Model for Concrete Targets,”
International Journal of Solids and Structures, Vol. 34,
No. 31-32, 1997, pp. 4127-4146.
[6] M. J. Forrestal, B. S. Altman, J. D. Cargile and S. J.
Hanchak, “An Empirical Equation for Penetration Depth
of Ogive-nose Projectiles into Concrete Targets,” Inter-
national Journal of Impact Engineering, Vol. 15, No. 4,
1994, pp. 396-405.
[7] M. J. Forrestal, D. J. Frew, S. J. Hanchak and S. Brar,
“Penetration of Grout and Concrete Targets with Ogive-
nose Steel Projectiles,” International Journal of Impact
Engineering, Vol. 18, No. 5, 1996, pp. 465-476.
[8] G. Ben-Dor, A. Dubinsky and T. Elperin, “Applied High-
speed Plate Penetration Dynamics,” Springer, Berlin,
2006.
[9] A. I. Bunimovich and G. E. Yakunina, “On the Shape of
Minimum-resistance Solids of Revolution Moving in
Plastically Compressible and Elastic-Plastic Media,” Jour-
nal of Applied Mathematics and Mechanics, Vol. 51, No. 2,
1987, pp. 386-392.
[10] N. A. Ostapenko and G. E. Yakunina, “The Shape of
Slender Three Dimensional Bodies with Maximum Depth
of Penetration into Dense Media,” Journal of Applied
Mathematics and Mechanics, Vol. 63, 1999, pp. 953-967.
[11] G. E. Yakunina, “On Body Shapes Providing Maximum
Penetration Depth in Dense Media,” Doklady Physics,
Vol. 46, No. 2, 2001, pp. 140-143.
[12] G. E. Yakunina, “On the Optimal Shapes of Bodies
Moving in Dense Media,” Doklady Physics, Vol. 50, No.
12, 2005, pp. 650-654.
[13] N. V. Banichuk and S. Y. Ivanova, “Shape Optimization
of Rigid 3-D High-speed Impactor Penetrating into Con-
crete Shields,” Mechanics Based Design of Structures
and Machines, Vol. 36, 2008, pp. 249-259.
[14] D. Bucur and G. Buttazzo, “Variational Methods in Shape
Optimization Problems,” Birkhauser, Boston, 2005.
[15] R. C. Eberhart and J. Kennedy, “A New Optimizer Using
Particle Swarm Theory,” Proceedings of the 6th Interna-
tional Symposium on Micromachine and Human Science,
Nagoya, 1995, pp. 39-43.
[16] J. Kennedy and R. C. Eberhart, “Particle Swarm Optimiza-
tion,” Proceedings of the IEEE International Joint Con-
ference on Neural Networks, Perth, 1995, pp. 1942-1948.
[17] M. Clerc, “Particle swarm optimization,” ISTE, London,
2006.
[18] M. M. Ali and P. Kaelo, “Improved particle swarm algo-
rithms for global optimization,” Applied Mathematics and
Computation, Vol. 196, 2008, pp. 578-593.
[19] P. C. Fourie and A. A. Groenwold, “The Particle Swarm
Optimization Algorithm in Size and Shape Optimiza-
tion,” Structural and Multidisciplinary Optimization, Vol.
23, No. 4, 2002, pp. 259-267.
[20] A. P. Engelbrecht, “Fundamentals of computational swarm
intelligence,” Wiley, Chichester, 2005.