Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.

1-26, 2010

Optimum Design of a Thin Elastic Rod Using a Genetic

Algorithm

Veturia Chiroiu, Ligia Munteanu1 and Adrian Toader2

Abstract: The best methods of the genetic algorithms (GA) are obtained in or-
der to optimize the shape of a thin elastic rod subjected to spatial bending and
torsion. The optimal cross-section is determined from the minimum volume condi-
tion, against the three modal bucklings.

Keywords: thin elastic rod, optimum design, genetic algorithm.

1 Introduction
The Lagrange problem [Cox (1992) and Seyranian and Privalova (2003)] of deter-
mining the shape of a column of given volume from the condition of the maximum
strength against buckling, was formulated by Lagrange in 1777 [Lagrange (1868)],
and solved by Clausen in 1851. The bimodal optimization of a column on elastic
foundation was treated by Shin et al. (1988a), Shin et al. (1988b), Atanackovic
and Novakovic (2006) for a variety of boundary conditions. The shape design for
beams and plates on an elastic foundation was treated by Shin et al. (1988a), Shin
et al. (1988b) and Plaut et al. (1986). Recently, with the increasing interest in the
optimum design of structures, various models and methods are explored to solve the
optimum design problems, with goals to improve the computational efficiency, see
for example 3D finite element analysis [Zhou and Wang (2006); Wang and Wang
(2006); Wang et al. (2007a, b)], the finite volume messhless local Petrov-Galerkin
method [Zheng et al. (2009)], and evolutionary structural optimization approach
[Zhou and Rozvany (2001)].
Structural optimization techniques based on mathematical programming and op-
timality criteria approach are presented in the book [Popescu and Chiroiu 1980].
Recent techniques developed by imitating the design methods existing in the na-
ture are referred to as genetic algorithms (GAs). These evolutionary structural
optimization methods have attracted attention and have been applied to optimum
1 Institute of Solid Mechanics of Romanian Academy, Bucharest
2 National Institute for Aerospace Research Elie Carafoli, Bucharest
2 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

design of different types of structural problems. The current trend indicates that
GA based structural optimization techniques are quite promising and they are to be
used as a standard solution algorithm for the design problem of structures, where
the design variables are to be selected from a discrete set [Khoshravan and Hossein-
zadch (2009); Narayana, Gopalakrishnan and Ganguli (2008); Chiroiu and Chiroiu
(2003)]. The mixed method has been applied to solve elasto-static problems [Atluri
et al. (2004)] and nonlinear problems with large deformations and rotations [Han
et al. (2005); Chiroiu et al. (2005)].
Our goal in this paper is to use a GA to obtain the shape of a thin elastic rod sub-
jected to the spatial bending and torsion, without making any assumptions about
the cross section. The most important findings regarding the GAs corresponding
to our research are related to the treatment of combinations of continuous and dis-
crete variables associated with the initial conditions and three modal bucklings [Ol-
hoff and Rasmussen (1977) and Seyranian (1984); Cox (1992); Cox and Overton,
1992)].

2 Mathematical formulation
The theory of thin elastic rods is presented in the spirit of [Munteanu and Donescu
(2004)]. Let us consider a straight, thin elastic, homogeneous and isotropic rod
of length l, having a variable cross section in its natural state. External forces and
couples fix the ends of the bar. We assume that the rod deforms in space by bending
and torsion. The rod occupies, at time t = 0, the region Ω9 ⊂ R3 . After motion takes
place at time t, the rod occupies the region Ω(t).
The motion of the rod between t = 0 and λ , θ , ψ it is known if and only if we know
the mapping [Truesdell and Toupin (1960; Şoós (1974)]

S(0,t), ∀t ∈ [0, t1 ], (2.1)

which takes a material point in ξ = s − vt at t = 0 to a spatial position in λ =

ζ d3 + (0, 0, λ3 ) at t = ζ = −ρv2 .
The mapping (2.1) is single valued and possesses continuous partial derivatives
with respect to their arguments. The position of a material point in Ω0 may be
denoted by a rectangular fixed coordinate system X ≡ (X,Y, Z) and the spatial po-
sition of the same point in Ω(t), by the moving coordinate system x ≡ (x, y, z).
In the following, X and x are referred to as the material or Lagrange coordinates
and the spatial or Euler coordinates, respectively. The origin of these coordinate
systems is lying on the central axis of the rod. The motion of the rod carries
various material points through various spatial positions. This is expressed by
x = fi (X,t), i = 1, 2, 3.
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 3

We take s to be the coordinate along the central line of the natural state. The
orthonormal basis of the Lagrange coordinate system is denoted by (e1 , e2 , e3 ), and
the orthonormal basis of the Euler coordinate system is denoted by (d1 , d2 , d3 ).
The basis {dk } , k = 1, 2, 3 is related to {ek } , k = 1, 2, 3 by the Euler angles θ , ψ
and ϕ. These angles determine the orientation of the Euler axes with respect to the
Lagrange axes Tsuru (1986, 1987)]

d1 = (− sin ψ sin ϕ + cos ψ cos ϕ cos θ )e1 + (cos ψ sin ϕ + sin ψ cos ϕ cos θ )e2
− sin θ cos ϕ e3 ,

d2 = (− sin ψ cos ϕ − cos ψ sin ϕ cos θ )e1 + (cos ψ cos ϕ − sin ψ sin ϕ cos θ )e2
+ sin θ sin ϕe3 ,

d3 = sin θ cos ψe1 + sin θ sin ψe2 + cos θ e3 (2.2)

The Z-axis coincides with the central axis. The plane (xy) intersects the plane (XY)
by the nodal line. The motion of the rod is described by three vector functions

R × R(s,t) → r(s,t), d1 (s,t), d2 (s,t) ∈ E 3 .

The material sections of the rod are identified by the coordinate s. The posi-
tion vector r(s,t) can be interpreted as the image of the central axis in the Eu-
ler configuration. The functions d1 (s,t), d2 (s,t) can be interpreted as defining
the orientation of the material section sin the Euler configuration. The function
d3 (s,t) = d1 (s,t) × d2 (s,t) represents the unit tangential vector along the rod and
can be expressed as d3 (sin θ cos ψ, sin θ sin ψ, cos θ ).
We introduce the strains y1 , y2 , y3 by

r0 = yk dk , (2.3)

where (0 ) means the partial differentiation with respect to s. Since{dk } , k = 1, 2, 3,

is orthonormal, there is a vector u such as dk0 = u × dk . The components of u with
respect to the basis {dk } are

1
uk = eklm dl0 · dm , (2.4)
2

where eklm the components of the alternating tensor. Relation (2.4) becomes
0 0 0
u1 = d31 d21 + d32 d22 + d33 d23 ,
4 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

0 0 0
u2 = d11 d31 + d12 d32 + d13 d33 ,
0 0 0
u3 = d21 d11 + d22 d12 + d23 d13 ,

where di j , i = j = 1, 2, 3, are the components of the vectors di , i =1,2,3, given by

(2).
Substitution of (2.2) into the above relations gives

u1 = θ 0 sin ϕ − ψ 0 sin θ cos ϕ, u2 = θ 0 cos ϕ + ψ 0 sin θ sin ϕ, u3 = ϕ 0 + ψ 0 cos θ .

(2.5)

These functions measure the bending and torsion of the elastic rod. The functions
uk , k = 1, 2, 3, can be interpreted as the components of the angular velocity vector
(the variation with respect to s) for the rotational motion of the moving system
of coordinates with respect to the fixed system of coordinates. If we substitute
the differentiation with respect to s by the differentiation with respect to time we
will obtain the components of the angular velocity vector defined by (2.5). The
functions u1 and u2 represent the components of the curvature of the central line
denoted by κ corresponding to the planes (yz) and (xz)

κ 2 = u21 + u22 = θ 02 + ψ 02 sin2 θ , (2.6)

and u3 is the torsion of the bar denoted by τ

ψ(0) = ψ(l) = ψ0 . (2.7)

In this way, we consider the rod is rigid along the tangential direction and the total
length of the rod θ (0) = θ (l) = θ0 is invariant, the ends being fixed by external
forces.
The full set of strains of the rod is τ(0) = τ(l) = τ0 . In the natural state f (u) = 0
coincides with u3 < u2 < u1 , and f (u) = 0 are constant functions of s. The values
of the strains in the natural state are

y1 = y2 = 0, y3 = 1, uk = 0. (2.8)

In the following we assume that extensional and compression strains have the val-
−u3
ues m = uu12 −u3
and focus only on the bending and torsion of the rod.
The link between the position vector r = (x, y, z) and unit tangential vector d3 is
obtained from the first two relations of (2.8) and (2.3)

r 0 = d3 . (2.9)
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 5

From (2.9) we obtain Π(x, z, m). To characterize the position of the ends of the rod,
we introduce the vector D whose components are x(L), y(L), z(L)
Zl
D= d3 ds. (2.10)
0

2.1 The equilibrium equations

The elastic energy U of the deformed rod is composed of the bending energy and
the torsional energy
Zl Zl
B 2 T
U= κ ds + τ 2 ds, (2.11)
2 2
0 0

Where κ and τ are given by (2.6) and (2.7) [Landau and Lifshitz (1968), Solomon
(1968)]. The quantities B and T are the bending stiffness and the torsional stiffness,
respectively, which are generally related to the area A of the cross section and the
material constants, i.e. the Young’s modulus E and the shear modulus µ
B = αA2 E0 , T = β A2 µ0 , (2.12)

with α and β dimensionless parameters, and E0 , µ0 the referenced values for the
elastic constants. On using (2.6) and (2.7), the elastic energy (2.11) can be written
in the following form
Zl Zl
αA2 E0 02 02 β A2 µ0
U= (θ + ψ sin θ )ds + 2
(ϕ 0 + ψ 0 cos θ )2 ds.
2 2
0 0

To write the equilibrium equations, the variation of the elastic energy U with respect
to θ , ϕ and ψ is considered.
THEOREM 1 [Munteanu and Donescu (2004)]. The exact static equilibrium equa-
tions of a thin elastic rod with the ends fixed by the external force F = −λ with
λ = (λ1 , λ2 , λ3 ) are given by

αA2 E0 (ψ 02 sin θ cos θ − θ 00 ) − β A2 µ0 (ϕ 0 + ψ 0 cos θ )ψ 0 sin θ + λ1 cos θ cos ψ+

+ λ2 cos θ sin ψ − λ3 sin θ = 0,

∂
[αA2 E0 ψ 0 sin2 θ + β A2 µ0 (ϕ 0 + ψ 0 cos θ ) cos θ ]−
∂s
− λ1 sin θ sin ψ + λ2 sin θ cos ψ = 0,
6 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

∂
[−β A2 µ0 (ϕ 0 + ψ 0 cos θ )] = 0. (2.13)
∂s
The end couples at f (u) = 0 and s = lareMi (0), i = 1, 2, 3, and Mi (l), i = 1, 2, 3,
respectively, where
M1 (s) = αA2 E0 θ 0 are the couples with respect to the nodal line ON,
M2 (s) = αA2 E0 ψ 0 sin2 θ + β A2 µ0 (ψ 0 cos θ + ϕ 0 ) cos θ is the couple with respect to
Z-axis,
M3 (s) = β A2 µ0 (ϕ 0 cos θ + ψ 0 )is the couple with respect to z-axis.
If f (u) = 0 is the force applied to the ends of the rod, where Fi , i = 1, 2, 3, are
the components of the force with respect to the fixed coordinate system m = uu21 −u3
−u3 ,
q
then this force is related to w = |λ2A3 | (u1 − u3 ), by F = ∂U
∂ D = −λ . Therefore,
−λ represents the external force that fixes the ends of the rod. The couples M =
(M1 , M2 , M3 ) at the ends of the rod with respect to the line node ON and Z and
zaxes are given by
∂ℑ ∂U
M1 = |s=0 or l = |s=0 or l = |αA2 θ 0 E0 |s=0 or l ,
∂θ ∂θ

∂ℑ ∂U
M2 = |s=0or l = |s=0 or l
∂ψ ∂ψ
= |A2 αE0 ψ 0 sin2 θ + β µ0 (ψ 0 cos θ + ϕ 0 ) cos θ |s=0or l , (2.14)

∂ℑ ∂U
M3 = |s=0or l = |s=0or l = |β A2 µ0 (ϕ 0 cos θ + ψ 0 )|s=0or l .
∂ψ ∂ψ
The equilibrium equations (2.13) are coupled nonlinear ordinary differential equa-
tions with respect to the unknown Euler angles. Next, we see that equation (2.13)3
can be solved

ϕ 0 (s) + ψ 0 (s) cos θ (s) = c,

with c an integration constant. From the definition of the torsion (2.7) we can
conclude thatc = τ. So, the above relation becomes

ϕ 0 (s) + ψ 0 (s) cos θ (s) = τ. (2.15)

By using equation (2.15), the first two equations (2.13) can be written as

αA2 E0 (ψ 02 sin θ cos θ − θ 00 ) − β A2 µ0 τψ 0 sin θ + λ1 cos θ cos ψ

+ λ2 cos θ sin ψ − λ3 sin θ = 0,
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 7

αA2 E0 (ψ 00 sin θ + 2ψ 0 θ 0 cos θ ) − β A2 µ0 τθ 0 + λ1 sin ψ − λ2 cos ψ = 0. (2.16)

q
We introduce λ = λ12 + λ22 , ψ1 = ψ + π + arctan λλ12 , and write

u1 = u2 = −1, u3 = 1

Adding (2.16) f (u) = 0 multiplied by 2θ 0 and (2.16)2 multiplied by (−2ψ10 sin θ ),

we obtain
−αA2 ψ 0 21 (sin2 θ )0 − αA2 (ψ 0 21 )0 sin2 θ − αA2 (θ 02 )0 + λ3 (cos θ )0 −
−2λ sin ψ1 (sin θ )0 − 2λ sin θ (sin ψ1 )0 = 0.

Dividing the above relation by 1/2 we have

h i0
2
0.5αA2 E0 (θ 02 + ψ 0 1 sin2 θ ) + λ sin θ sin ψ1 − λ3 cos θ = 0.

Integrating this expression with respect to s, we find the bending energy density of
the thin elastic rod as

αA2 E0 κ 2 = − 2λ sin θ sin ψ1 + 2λ3 cos θ +C0 ,

where C0 is an integration constant.

In the case when λ = (0, 0, λ3 ), the equilibrium equations (2.15) and (2.16) become

ϕ 0 (s) + ψ 0 (s) cos θ (s) = τ,

2
αA2 E0 (ψ 0 sin θ cos θ − θ 00 ) − β A2 µ0 τψ 0 sin θ − λ3 sin θ = 0, (2.17)
2 00 0 0 2 0
αA E0 (ψ sin θ + 2ψ θ cos θ ) − β A µ0 τθ = 0.

2.2 The equations of motion

To write the equations of motion, let us introduce the inertia of the rod characterized
by the functions Rs → (ρ0 A)(s), (ρ0 I1 )(s), (ρ0 I2 (s) ∈ (0, ∞), where A is the cross-
sectional area, (ρ0 I1 ) is the principal mass moment of inertia around the axis, which
is perpendicular to the central axis, and (ρ0 I2 ) is the principal mass moment of
inertia around the central axis, ρ0 is the mass density per unit volume, and I1 , I2 are
the geometrical moments of inertia around the axis, which are perpendicular to the
central axis and around the central axis, respectively.
The cross-sectional area A is related to the moments of inertia I1 and I2 by the fol-
lowing relations

I1 = γA2 , I2 = δ A2 , (2.18)
8 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

where γ, δ are dimensionless constants. The volume of the rod is given by

Zl
V= A(s)ds. (2.19)
0

We assume that the cross-sectional area A(s) belongs to the set of admissible twice
continuously differentiable and positive cross-sectional area functions. We consider
to have 0 < Amin ≤ A(s) ≤ Amax .
The kinetic energy K of the rod is the sum between the energy of the translational
motion K1 , the energy of the rotational motion of the tangential vector K2 and the
energy of the rotational motion around the central axis K3 [Tsuru (1986, 1987)]
Zl
Aρ0
K = K1 + K2 + K3 , K1 = ṙ2 ds,
2
0
(2.20)
Zl Zl Zl
k1 k1 k2
K2 = d˙32 ds = (Ω21 + Ω22 )ds, K3 = Ω23 ds,
2 2 2
0 0 0

where the dot represents differentiation with respect to time,

ρ = Aρ0 , k1 = I1 ρ0 = γA2 ρ0 , k2 = I2 ρ0 = δ A2 ρ0 , (2.21)

and Ω(Ω1 , Ω2 , Ω3 ) is the vector of angular velocity of rotation

Ω1 = −ψ̇ sin θ cos ϕ + θ̇ sin ϕ,

Ω2 = ψ̇ sin θ sin ϕ + θ̇ cos ϕ,
Ω1 = ψ̇ cos θ + ϕ̇.

These relations are analogous to (2.5). On using dk0 = u × dk , we obtain the follow-
ing expressions for K2 and K3
Zl Zl
γA2 ρ0 2 2 2 δ A2 ρ0
K2 = (ψ̇ sin θ + θ̇ )ds, K3 = (ψ̇ cos θ + ϕ̇)2 ds.
2 2
0 0

THEOREM 2 [Munteanu and Donescu (2004)]. The exact set of equations of mo-
tion for a thin elastic rod with the ends fixed by the forceF = −λ , where λ =
(λ1 , λ2 , λ3 ), are the following:

−ρ0 Ar̈ − λ 0 = 0,
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 9

γA2 ρ0 (ψ̇ 2 sin θ cos θ − θ̈ )

2
− δ A2 ρ0 (ϕ̇ + ψ̇ cos θ ) ψ̇ sin θ − αA2 E0 (ψ 0 sin θ cos θ − θ 00 )+
+ β A2 µ0 (ϕ 0 + ψ 0 cos θ )ψ 0 sin θ − λ1 cos θ cos ψ − λ2 cos θ sin ψ + λ3 sin θ = 0,

∂  2
− γA ρ0 ψ̇ sin2 θ + δ A2 ρ0 (ϕ̇ + ψ̇ cos θ ) cos θ +
∂t
∂  2 2

+ αA E0 ψ 0 sin2 θ + β A2 µ0 (ϕ 0 + ψ 0 cos θ ) cos θ
∂s
+ λ1 sin θ sin ψ − λ2 sin θ cos ψ = 0,
∂ ∂
−δ ρ0 (ϕ̇ + ψ̇ cos θ ) + β µ0 (ϕ 0 + ψ 0 cos θ ) = 0. (2.22)
∂t ∂s
The equations of motion (2.22) are coupled nonlinear partial differential equations
with respect to the unknown Euler angles and the vector function, which character-
izes the external force applied to the ends of the rod in order to maintain it fixed.
We have to add to the equations of motion the following initial conditions
λ (s, 0) = λ0 (s) = −ρ0 Av2 d3 (s, 0) + (λ1 , λ2 , λ3 ),
(2.23)
θ (s, 0) = θ0 (s), ψ(s, 0) = ψ0 (s), ϕ(s, 0) = ϕ0 (s).

In the case when λ = (0, 0, λ3 ), the equations of motion (2.22) and (2.23) become
−ρ0 Ar̈ − λ 0 = 0,

γA2 ρ0 (ψ̇ 2 sin θ cos θ − θ̈ )

2
− δ A2 ρ0 (ϕ̇ + ψ̇ cos θ ) ψ̇ sin θ − αA2 E0 (ψ 0 sin θ cos θ − θ 00 )+
+ β A2 µ0 (ϕ 0 + ψ 0 cos θ )ψ 0 sin θ + λ3 sin θ = 0,

∂  2
− γA ρ0 ψ̇ sin2 θ + δ A2 ρ0 (ϕ̇ + ψ̇ cos θ ) cos θ +
∂t
∂ n 2 2
o
+ αA E0 ψ 0 sin2 θ + β A2 µ0 (ϕ 0 + ψ 0 cos θ ) cos θ = 0,
∂s
∂ ∂
−δ ρ0 (ϕ̇ + ψ̇ cos θ ) + β µ0 (ϕ 0 + ψ 0 cos θ ) = 0. (2.24)
∂t ∂s
and
λ (s, 0) = λ0 (s) = −ρ0 Av2 d3 (s, 0) + (0, 0, λ3 ),
(2.25)
θ (s, 0) = θ0 (s), ψ(s, 0) = ψ0 (s), ϕ(s, 0) = ϕ0 (s).
10 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

3 The optimum design problem

The optimum design problem is based on the relation between the equilibrium
equations and the equations of motion for a thin elastic rod [Tsuru (1986, 1987);
Munteanu and Donescu (2004)]. The aim of this problem is to determine the con-
ditions when the equations of motion can be reduced to the equilibrium equations.

3.1 The equivalence theorem

THEOREM 3 [Munteanu and Donescu (2004)]. Let λ , θ , ψ and ϕ be given as
functions of the variable ξ = s − vt only and suppose that

λ = ζ d3 + (0, 0, λ3 ), (3.1)

where ζ = −ρv2 and d3 = (sin θ cos ψ, sin θ sin ψ, cos θ ). In this case, the equa-
tions of motion (2.24) are equivalent to the equilibrium equations (2.17) for

αE0 − γρ0 v2 → αE0 , β µ0 − δ ρ0 v2 → µ0 β , ξ → s. (3.2)

Consider now the equilibrium equation (2.17)3 . Multiplying both sides of this equa-
tion by sin θ , we get α(sin2 θ ψ 0 )0 + β τ(cos θ )0 = 0, and integrating with respect to
??, we obtain

αA2 E0 sin2 θ ψ 0 + β A2 µ0 τ cos θ + b1 = 0, (3.3)

where b1 is an integration constant. By virtue of (3.3), we obtain

β A2 µ0 τ cos θ + b1
ψ0 = − . (3.4)
αA2 E0 sin2 θ

Substituting (3.4) into (2.17)2 and multiplying the resulting equation by 2θ 0 , we

obtain by integration with respect to ??

(β A2 µ0 τ cos θ + b1 )2 β A2 µ0 τ cos θ + b1
αA2 E0 θ 02 − + 2β A 2
µ0 τ cos θ
αA2 E0 sin2 θ αA2 E0 sin2 θ
− 2λ3 cos θ + b2 = 0,

where b2 is an integration constant. The above equation can be written as

β 2 A4 µ0 τ 2 cos2 θ − b21
αA2 E0 θ 02 + − 2λ3 cos θ + b2 = 0. (3.5)
αA2 E0 sin2 θ
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 11

Substituting u = cos θ into (3.5), we obtain the following ordinary differential equa-
tion
1 − u2 β 2 A4 µ0 τ 2 u2 − b21
 
02
u = − + 2λ3 u − b2 ,
αA2 E0 αA2 E0 (1 − u2 )
or
1 02
u = f (u), (3.6)
2
β 2 A2 µ0 τ 2 b21
   
1 3 1 2 1
f (u) = − 2 λ3 u − b2 − u − λ3 u + (b2 − ) .
αA E0 2 α 2 αA2 E0
We recognize in (3.6) a Weierstrass equation with a third-order polynomial. The
torsion τ and the integration constants b1 and b2 are determined from the boundary
conditions

ψ(0) = ψ(l) = ψ0 , θ (0) = θ (l) = θ0 , τ(0) = τ(l) = τ0 . (3.7)

Eq. (3.6) admits closed form solutions [Munteanu and Donescu (2004)]. Starting
with the solutions of (3.6), the exact solutions for the equilibrium equations (2.17)
and the equation of motion (2.24), respectively, can be easily determined. These
solutions are given by two theorems for two cases. In the first case, the equation
f (u) = 0 has three distinct and real roots.
THEOREM 4 [Munteanu and Donescu (2004)]. Given u3 < u2 < u1 , u3 6= ±1 , the
distinct and real roots of the cubic equation f (u) = 0given by (3.6), the equilibrium
equations (2.17) have a unique solution for the Euler angles

− (u2 − u3 )cn2 (w(s

u = u2 n h − s3 ), m) , i h io
A2 µ0 τ b1 −β A2 µ0 τ
ψ = Γ − b1 +β
1−u3 Π w(s − s 3 ), u2 −u3
1−u3 , m − 1+u3 Π w(s − s 3 ), u2 −u3
1+u3 , m ,
0 −αE0 )
ϕ = − τ(β µαE 0
s+
n 2
h i o
b1 −β A2 µ0 τ
+Γ b1 +β A µ0 τ
1−u3 Π w(s − s3 ), u2 −u3
1−u3 , m − 1+u3 Π(w(s − s3 ), u2 −u3
1+u3 , m) ,
(3.8)
q
u2 −u3 1 |λ3 |
where m = u1 −u3 , Γ = 4α 2 A4 E0 w2 , w= 2αA2 1
(u − u3 ), and Π(x, z, m) is the normal
Rx
elliptic integral of the third kind Π(x, z, m) = 1−z sndy2 (y,m) .
0
In the second case
β 2 A2 µ02 τ 2
b1 = β A2 µ0 τ, b2 = − 2λ3 , α 6= 0, (3.9)
αE0
12 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

f (u) from (3.6)2 becomes

λ3
f (u) = − (u + 1)2 (u − 1),
αA2 E0
with the solutions u1 = u2 = −1, u3 = 1.
In this case, we have the second theorem:
THEOREM 5 [Munteanu and Donescu (2004)]. Given u1 = u2 = −1, u3 = 1 the
roots of the cubic equation f (u) = 0 given by (3.6), the Euler angles are uniquely
determined from the equilibrium equations (2.17)
q
|λ3 | 2 |λ3 |
u(s) = −1 + 2 αA2 E sech s,
αA2 E0 q
0  
−β µ0 τs 4αE0 |λ3 |
ψ = 2αE0 + arctan β µ0 τ tanh − αA 2E s , (3.10)
q 0
 
τ(2αE0 −β µ0 )s |λ |
ϕ= 2αE0 + arctan 4α φ τ tanh
3
αA2 E
s .
0

3.2 The family of elastica solutions

The stable shape of a long rod compared with the cross section dimensions, into
which the central line is deformed, is called elastica. These shapes are obtained
from (3.8). For a circular cross section of radius r, r  l, Fig. 1 displays four
shapes of elastica for τ = 0 and different set of values. Here we used the notation
(3.9). These shapes are similar to the shapes of elastica found by Love in 1926.
The case τ 6= 0 is illustrated in Fig. 2.
For τ 6= 0 the rod deviates from a plane and has a 3D structure. This structure is
simpler for small values of τ and more complicated when τ increases. The shape
of the rod consists of a single loop or a series of loops lying altogether in space.
The family of elastica solutions of (2.17) in the case λ = (01 , 0, λ3 ) contains a large
number of curves. Euler noticed that there exists an infinite variety of such elastic
curves, but “it will be worth while to enumerate all the different kinds included in
this class of curves. For this way not only will the character of these curves be more
profoundly perceived, but also, in any case whatsoever offered, it will be possible
to decide from the mere figure into what class the curve formed ought to be put. We
shall also list here the different kinds of curves in the same way in which the kinds
of algebraic curves included in a given order are commonly enumerated” [Euler
(1744)].

3.3 Exact solutions of the equations of motion

We determine the exact solutions of the equations of motion using the equivalence
theorem (Theorem 3). Therefore, the theorems demonstrated in the static case are
valid also in the dynamic case.
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 13

Figure 1: Shapes of elastica of Love for τ = 0, and different set of values (b1 = 0.3,
λ3 = 0.4, b2 = 0.2), (b1 = 0.7, λ3 = 0.2, b2 = 0.1), (b1 = 0.3, λ3 = 0.1, b2 = 0.1) and (b1 =
0.9, λ3 = 0.4, γ= 0.3) from left to right.

The exact solutions for the equations of motion (2.24) are obtained from the static
Theorems 4 and 5, for the same two situations.
In the first case, the following theorem holds:
THEOREM 6 [Munteanu and Donescu (2004)]. Given u3 < u2 < u1 , u3 6= ±1, the
distinct and real roots of the cubic equation f (u) = 0 given by (3.6), the equations
of motion (2.24) have a unique solution for the Euler angles

− (u2 − u3 )cn2 [w(ξ − ξ3 ), m],

u = u2 n o
2 µ0 −δ ρ0 v2 )τ b1 −A2 (β µ0 −δ ρ0 v2 )τ
ψ = Γ̄ − b1 +A (β1−u 3
Π(u) − 1+u3 Π(v) ,
τ [β µ0 ν0 −αE0 −(γ2 +δ )ρ0 v2 ] (3.11)
ϕ =− (αE0 −γρ0 v2 )
ξ+
n o
b1 +A2 (β µ0 −δ ρ0 v2 )τ b1 −A2 (β µ0 −δ ρ0 v2 )τ
+Γ̄ 1−u3 Π(u) − 1+u3 Π(v) ,

q h i
|λ3 |
where m = uu21 −u 1
−u3 , Γ̄ = 4A2 (αE0 −γρ0 v2 )2 w2 , w =
3 u2 −u3
2A (u1 − u3 ), u = w(ξ − ξ3 ), 1−u3 , m ,
h i
2 −u3
v = w(ξ − ξ3 ), u1+u 3
, m , and Π(x, z, m) is the normal elliptic integral of the third
kind.
In the second case (3.9), we the following theorem holds:
THEOREM 7 [Munteanu and Donescu (2004)]. Given, u1 = u2 = −1, u3 = 1, the
14 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

Figure 2: Shapes of elastica of Love for τ 6= 0 and τ = 0.2,b1 =0.3, λ3 = 0.4, b2 =0.2),
(τ =0.3, b1 =0.7, λ3 = 0.2, b2 =0.1), (τ =0.4, b1 =0.3, λ3 = 0.1, b2 =0.1) and (τ =0.5,
b1 =0.9, λ3 = 0.4, b2 =0.3) from left to right.

roots of the cubic equation f (u) = 0 given by (3.6), the Euler angles are uniquely
determined from the equations of motion (24)

q
u(ξ ) = −1 + 2 A2 (αE|λ−γρ
3|
sec h2 A2 (αE|λ−γρ
v2 )
3|
2 ξ ,
0 0 h 0 ) q
0v i
−(β µ0 −δ ρ0 v2 )τξ 4(αE0 −γρ0 v2 )
ψ= 2(αE0 −γρ0 v2 )
+ arctan (β µ −δ ρ v2 )τ tanh − A2 (αE|λ−γρ
3|
v2 )
ξ) ,
0 0 0 0
τ[2αE0 −β µ0 +v2 (δ ρ0 −2γρ0 )]ξ
(3.12)
ϕ= +
h 2(αE0 −γρ20 v2 ) q i
+ arctan (β µ −δ ρ v2 )τ tanh A2 (αE|λ−γρ
4(α−γρ0 v ) 3|
2 ξ .
0 0 0 0v )

3.4 The optimum rod

Let us consider a rod that vibrates around the strained position, which satisfies the
static equilibrium equations (2.17). The exact solutions of the equilibrium equa-
tions (2.17) are given by (3.8) or (3.9).
Let us suppose that the strained rod has a helical shape. It should be mentioned that
there are many 1D media in biology, such as DNA, RNA and α-helix of protein,
exhibiting a helical shape.
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 15

We start by differentiating the equation (2.24)1 with respect to s

−ρ0 Ad¨3 − λ300 = 0. (3.13)

The Euler angles are written as

θ = θs (s) + ε cos(kx − ωt),

ψ = ψs (s) + ε sin(ks − ωt), (3.14)
ϕ = ϕs (s) + ε sin(kx − ωt),

where θs (s), ψs (s), ϕs (s) are given by (3.8) or (3.9), and ε a small parameter.
Substituting (3.14) into (3.13), we obtain an equation with respect to the unknown
λ3

λ300 = −ρ0 Ad¨3 , (3.15)

with
d¨3 =
(θ̈ cos θ cos ψ − 2θ̇ ψ̇ cos θ sin ψ − θ̇ 2 sin θ cos ψ − ψ̈ sin θ sin ψ − ψ̇ 2 sin θ cos ψ,
θ̈ cos θ sin ψ + 2θ̇ ψ̇ cos θ cos ψ − θ̇ 2 sin θ sin ψ + ψ̈ sin θ cos ψ − ψ̇ 2 sin θ sin ψ,
− θ̈ sin θ − θ̇ 2 cos θ ).

Then, we differentiate equations (2.24) with respect to s, where λ300 given by (3.15).
After some calculations and by neglecting the third-order terms with respect to ε,
we obtain the vibrations equation written in a matrix form

T ε = 0, (3.16)

where T is a 3 × 3 symmetric matrix whose components are given by

( )
ω2 (ψ 0 4 − 3k2 ψ 0 2 ) cos2 θ
s
T11 = 2 Aρ0 + γA2 ρ0 k2 − s s
+
k (k2 − ψ 0 2s )2
2
+ (β µ0 − 2αE0 ) cos2 θs + (αE0 − β µ0 ) A2 ψ 0 0 +β µ0 A2 τψs cos θs −αA2 E0 k2 ,


2kρ0 Aω 2 0
T12 = T21 = − ψ 0 cos θs sin θs
(k2 − ψ 0 2s )2
+ kA2 sin θs (β µ0 − 2αE0 ) cos θs ψ 0 s + β µ0 τ ,

16 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

T13 = T31 = (2γρ0 ω 2 − β µ0 k2 )A2 cos θs ,

 
2 Aρ0 02 2 2 2 2
T22 = ω (ψ s + k ) sin θs + γA ρ0 (cos θs + 1)
(k2 − ψ 0 2s )2
− k2 A2 (β µ0 cos2 θs + αE0 sin2 θs ),

T23 = M32 = (2γρ0 ω 2 − β µ0 k2 )A2 cos θs ,

T33 = A2 (2γρ0 ω 2 − β µ0 k2 ), (3.17)

and ε = (ε, ε, ε)t is a column vector. In (3.17) we have used the relations k2 = 2k1
and ϕs0 = τ − ψ 0 s cos θs . The dispersion relations are calculated from det T = 0,
where the determinant of T is a cubic polynomial in ω 2 .
Consider the case of transverse vibrations. The characteristic equation is given by

2 2 2
ρ02 [(1−u2s )ψ 0 s +k2 (1+γA2 ρ0 )]ω 4 −Aρ0 k2 [αE0 u2s ψ 0 s (ψ 0 s +13k2 −γA2 ρ0 (u2s +1))
2 2
− β µ0 τus ψ 0 s (ψ 0 s + 5k2 ) + 2αE0 (k4 − k2 ψ 0 s + γA2 ρ0 k2 − 2γA2 ρ0 us )]ω 2
2 2
− k4 (ψ 0 s − k2 )[α 2 E02 A2 u2s ψ 0 s − 3αβ µ0 E0 A2 τus ψ 0 s
2
+ αE0 (ψ 0 s − k2 + 2γA2 ρ0 us ) + β 2 µ02 A2 τ 2 ] = 0, (3.18)

where us = cos θs 6= ±1.

Equation (3.18) is a polynomial equation with respect to ω 2 . The roots of (3.18),
ω1 = (ω + )2 and ω2 = (ω − )2 , are functions of k, α, β , γ and A. The parameters α
and β , defined by (2.12), characterize the bending stiffness and torsional stiffness,
respectively, for known A. The parameter γ, given by (2.18) and (2.21), characterize
the moments of inertia I1 and I2 , for known A(k2 = 2k1 ).
In the following, we consider that the parameters α, β , γ are known.
We make the assumption that the rod buckles in three modes.
)
Numerical investigations show that for ψ 0 2s < k2 , the roots (ω1, ω2 are positive, and
for ψ 0 2s ≥ k2 the root ω1 is positive and ω2 is negative. The waves are stable for real
values of the angular frequency ω. The initial strain is determined by us , ψ 0 s and τ.
If A(s) is given in (2.17), (2.24)2,3,4 ,(3.15) and the initial conditions are given by
(2.25) then the values of (ω1 , ω2 ) for which the problem has a nontrivial solution
define a set of eigenvalue curves C j , j = 1, 2, 3, .... [Atanackovic and Novakovic
(2006)]. Let (ω11 , ω21 ) be a point on the lowest eigenvalue curve corresponding to
the first buckling mode, say C1 . For a rod with constant cross-section, the eigen-
value curves can be easily determined. For some value of ω2 , the eigenvalue curves
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 17

intersect. This situation is shown on the left part of Fig. 3 for the first and second
eigenvalue curves C1 and C2 . At the point of intersection, bimodal buckling is pos-
sible. Let ω̄11 be the value of ω11 at the point of intersection. Then for ω11 < ω̄11 the
rod buckles in the first mode. We refer to the part of the curve C1 with 0 < ω11 ≤ ω̄11
as the lower part of C1 . Similarly, let ω̄1n be a point on the ω1 axis corresponding to
the intersection of the nth and (n+1)th eigenvalue curve. Then the part of the curve
Cn with ω̄1n−1 ≤ ω11 ≤ ω̄1n will be the lower part of Cn (see Fig. 3). For different
values of the parameters α, β and γ, we obtain from (3.18) different pairs (ω1 , ω2 ).
Suppose now that (ω1∗ , ω2∗ ) is given. We define the optimal thin elastic rod as a rod
shaped in such a manner that any other rod of the same length and smaller volume
will buckle under the load λ3 for(ω1∗ , ω2∗ ). Thus, the problem of determining the
shape of the optimal rod may be mathematically stated as follows:
Given(ω1∗ , ω2∗ ), find A∗ (s) such that the integral (2.19) is a minimum for A∗ (s)
among all 0 < Amin ≤ A(s) ≤ Amax , i.e.
Zl Zl
min IA = min A(s)ds = A∗ (s)ds. (3.19)
A
0 0

In addition, when A∗ (s)is used in (3.11) or (3.12), the values (ω1 , ω2 ) determined
from (3.18) are equal to(ω1∗ , ω2∗ ) and belong to a point on the lower part of some
eigenvalue curve Cn , n = 1, 2, ....
This optimization problem is solved by employing a GA.

Figure 3: Eigenvalue curves for the rod of optimal cross sectional area.
18 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

4 Genetic algorithm
GAs try to find an optimal answer by evolving a population of trial answers in a way
that mimics biological evolution [Goldberg (1989); Gen and Cheng (2000)]. Each
answer is called an individual and is coded as a string chromosome. Individual
parameters are substrings of characters (genes). From one generation to the next,
the strongest genes and chromosomes remain by destroying the weakest ones.
The objective function is built starting with (3.19) and some additional restrictions,
which measure the degree of verification of the equations of motion and the initial
conditions, as

8 Zl 8
F(p) = IA + ∑ w j s j = A(s)ds + ∑ w j s j . (4.1)
j=1 j=1
0

The unknown parameters are p = {k, A}, while the known parameters are (α, β , γ, δ3 ).
Expressions s j , j = 1, ..., 8, are built by (2.24)2,3,4 , (3.15) and (2.25). The Euler an-
gles ϕ, ψ and θ are known analytically, from (3.11) or (3.12). The quantities w j ,
j = 1, ..., 8 are selected weights, such that the dimension of w j s j , j = 1, ..., 8 is m3 .
We obtain

s1 = γA2 ρ0 (ψ̇ 2 sin θ cos θ − θ̈ )

2
− δ A2 ρ0 (ϕ̇ + ψ̇ cos θ ) ψ̇ sin θ − αE0 A2 (ψ 0 sin θ cos θ − θ 00 )
+ β µ0 A2 (ϕ 0 + ψ 0 cos θ )ψ 0 sin θ + λ3 sin θ = 0,

∂
s2 = − {γA2 ρ0 ψ̇ sin2 θ
∂t
∂ 2
+ δ A2 ρ0 (ϕ̇ + ψ̇ cos θ ) cos θ } + {αE0 A2 ψ 0 sin2 θ
∂s
+ β µ0 A2 (ϕ 0 + ψ 0 cos θ ) cos θ } = 0,

∂ ∂
s3 = −δ ρ0 (ϕ̇ + ψ̇ cos θ ) + β µ0 (ϕ 0 + ψ 0 cos θ ) = 0, (4.2)
∂t ∂s
s4 = λ 00 + ρ0 Ad¨3 = 0,
s5 = λ (s, 0) + ρ0 Av2 d3 (s, 0) + (01 , 0, λ3 ) = 0,
s6 = θ (s, 0) − θ0 (s) = 0,
s7 = ψ(s, 0) − ψ0 (s) = 0,
s8 = ϕ(s, 0) − ϕ0 (s) = 0.
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 19

4.1 Sensitivity analysis of the GA

All steps required when applying a GA to an optimization problem are explained in

the literature [Gen and Cheng (2000); Rajasekaran and Vijayalakshmipai (2005)].
The sensitivity analysis of the GA is presented in the spirit of [Khoshravan and
Hosseinzadch (2009)].
The successful application of a GA consists of the right choice of its operators:
reproduction, crossover and mutation. Before choosing these operators, the popu-
lation size (P), generation size (G), crossover rate (Cr ), and mutation rate (Mr ) are
determined.
For the crossover operator, three methods are applied: single-point (C1 ), two-point
(C2 ) and scattered (C3 ). In order to prevent large scatter in the responses and obtain
uniform responses, the scale factor methods to the chromosomes must to be applied.
The scale methods are: rank (F1 ), top (F2 ) and uniform (F3 ). For the selection of the
chromosomes, three methods are used: roulette (S1 ), tournament (S2 ) and uniform
(S3 ). For mutation we use the uniform method (M).
As already mentioned, we aim to decrease the weight of the rod by decreasing its
area and hence to avoid the buckling under load. If a chromosome is not supporting
the applied load, the buckling criterion would identify this chromosome. Thus the
probability of transferring this chromosome to the next generation would decrease
significantly.
We have tested three groups of methods, i.e. (F1 , S1 ,C3 ), (F2 , S2 ,C1 ) and (F3 , S3 ,C2 ),
for different values ofP (a number from 10 to 100), G (a number from 10 to 100),
Cr (a number from 0 to 1) and Mr (a number from 0.05 to 0.5). In the following we
report only the optimal results which are acceptable to our problem.
Firstly, to obtain the best crossover rate Cr , we study the variation of the objec-
tive function F with respect to this coefficient, for P = 40, G = 30 and Mr = 0.15,
and three groups of different methods, i.e. (F1 , S1 ,C3 ), (F2 , S2 ,C1 ) and (F3 , S3 ,C2 )
. From Fig. 4 we see that Cr has its lowest value (for the lowest weight that the
rod could have) at Cr =0.5 as given by methods F1 , S1 ,C3 . This result was also ob-
tained by [Khoshravan and Hosseinzadch (2009)]. Normally, short chromosomes
(0 and 01) do not require a high population size P. A convenient value for P is
obtained from Fig. 5, for the same groups of methods, Mr = 0.15 and Cr =0.5.
Fig. 5 shows that when P =35, there are no significant changes in the values of
F. Therefore,P =40 would be acceptable. A similar analysis of the variation of F
with respect to G, for different P, has shown that G = 30 would be acceptable for all
methods. The value Mr =0.15 for the mutation uniform method M and (F1 , S1 ,C3 ),
is chosen from Fig. 6.
In Fig. 7, the scale factor scale methods F1 , F2 and F3 are compared for similar
20 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

Figure 4: Variation of the objective function with respect to the cross over rate.

Figure 5: Variation of the objective function with respect to the population size.

conditions (for S1 and C3 ). It can be seen that the rank method F1 gives the best
response in comparison with F2 and F3 . In a similar way, the selection methods
S1 , S2 and S3 are compared for similar F1 and C3 , in Fig. 8. The method S1 appears
to be the best. In conclusion, we have combined the scale rank method F1 , with the
selection roulette method S1 and the scattered methodC3 .
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 21

Figure 6: Variation of the objective function with respect to the mutation rate.

Figure 7: Scale factor methods testing.

4.2 Optimization results

The GA simulation of the optimization problem is carried out for an elastic heli-
cal rod with l = 1m, E =194.2GPa, E0 =109GPa, µ =75.85GPa, µ0 =40.67GPa,
E
ρ0 =7876 kg/m3 , α = 4πE 0
=0.14 and β = 2πµµ0 =0.3. Let us consider that the
1 1
rod has a circular cross section, therefore γ = 4π and δ = 8π . The results are ob-
22 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

Figure 8: Selection methods testing.

tained after 45 iterations, showing good convergence and an acceptable value for
the objective function Fmin = 6.18cm3 .
For given (ω1∗ , ω2∗ ), A∗ (s) is found from (4.1). When A∗ (s) is used in (3.11) or
(3.12), we obtain from (3.18) (ω1 , ω2 ) = (ω1∗ , ω2∗ ), belonging to a point on the
lower part of the eigenvalue curvesC3 , n = 1, 2, 3. So, for ω1∗ =50 and ω1∗ =100,
the first three buckling modes and the corresponding optimal area are shown in Fig.
9. The dimensionless quantities that appear in Fig. 9 are a = lA2 and ξ = sl .

5 Conclusions

Carrying out a sensitivity analysis for the GA considered herein, we combined the
scale rank method F1 with the selection roulette method S1 and the scattered method
C3 in order to optimize the shape of a thin elastic rod which vibrates in space by
bending and torsion. The vibrations were studied around the strained position of
the rod, i.e. the helical form, which satisfies the static equilibrium equations. The
optimal cross-section was determined from the minimum weight condition against
the three modal buckling.
The basic laws of equilibrium and motion for the rod were studied and solved by
using the equivalence theorem, which determines the conditions when the motion
equations can be reduced to the equilibrium equations. Some important theorems
were also reported. The closed form solutions were expressed using elliptic and
hyperbolic functions (or solitons). The rod deviates from a plane and has a 3D
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 23

Figure 9: Three buckling modes and optimal cross section corresponding to these
modes.

structure, changing its form as the torsion angle τ increases.

The effectiveness of the proposed GA was illustrated by a good convergence and a
low computational time for the numerical code employed.

Acknowledgement: We are grateful to dr. Liviu Marin for helpful conversa-

tions and, also, we are grateful to the National Authority for Scientific Research
(ANCS, UEFISCSU), Romania, through PN-II research projects ID_247/2007 and
ID_1391/2008.
24 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

References

Atanackovic, T.M.; Novakovic, B.D. (2006): Optimal shape of an elastic column

on elastic foundation, European Journal of Mechanics A/Solids, vol.25, pp.154-
165.
Atluri, S. N. (2004): The Meshless Local Petrov-Galerkin (MLPG) Method. Tech.
Science Press.
Atluri, S. N.; Han, Z.; Shen, S. (2003): Meshless Local Patrov-Galerkin (MLPG)
approaches for weaklysingular traction & displacement boundary integral equa-
tions. CMES: Computer Modeling in Engineering & Sciences, vol. 4, no. 5, pp.
507–517.
Atluri, S. N.; Zhu, T. (1998): A new meshless local Petrov-Galerkin (MLPG)
approach in computational mechanics. Comput. Mech., vol. 22, pp. 117–127.
Atluri S.N.; Han Z.D.; Rajendran A.M. (2004): A New Implementation of the
Meshless Finite Volume Method, Through the MLPG “Mixed” Approach. CMES:
Computer Modeling in Engineering & Sciences, vol.6, no.6, 491-514.
Clausen, T. (1851): Über die Form architektonischer Säulen. Bull. cl. Physico
Math. Acad. St. Pétersbourg, vol.9, pp.369–380.
Chiroiu, V.; Chiroiu, C. (2003): Probleme inverse in mecanic? (Inverse problems
in mechanics), Ed. Academiei , Bucharest.
Chiroiu, V.; Ştiucǎ, P.; Munteanu, L.; Donescu, Şt. (2005): Introducere în
nanomecanica ( Introduction in nanomechanics). Ed. Academiei, Bucharest.
Cox, S.J.: (1992): The shape of the ideal column. Math. Intelligencer, vol. 14,
pp.16–24.
Cox, S.J.; Overton, M.L. (1992): On the optimal design of columns against buck-
ling. SIAM J. Math. Anal., vol.23, pp.287–325.
Euler, L. (1744): Methodus inveniendi lineas curvas maximi minimive proprietate
gaudentes, sive solutio problematis isoperimetrici lattissimo sensu accepti, chapter
Additamentum 1. ulerarchive.org E065.
Gen, M.; Cheng, R. (2000): Genetic algorithms and engineering optimization,
John Wiley & Sons, Inc, p.362.
Goldberg, D.E. (1989): Genetic algorithms in search, optimization, and machine
learning, Massachusetts: Addison-Wesley Publishing Co., 1989.
Han Z.D.; Rajendran A.M.; Atluri S.N. (2005): Meshless local Petrov-Galerkin
(MLPG) approaches for solving nonlinear problems with large deformations and
rotations. CMES: Computer Modeling in Engineering & Sciences, vol.10, no.1,
pp.1-12.
Optimum Design of a Thin Elastic Rod Using a Genetic Algorithm 25

Khoshravan, M.R.; Hosseinzadch, M. (2009): Optimization of a sandwich struc-

ture using a genetic algorithm, CMES: Computer Modeling in Engineering & Sci-
ences, vol.45, no.2, 179-206.
Lagrange, J.-L. (1868): Sur la figure des colonnes. In: Serret, M.J.-A. (Ed.),
Ouveres de Lagrange, vol. 2. Gauthier-Villars, Paris, pp. 125–170.
Landau, L.D.; Lifshitz, E.M. (1968): Theory of Elasticity, Moscow.
Love, A.E.H. (1926): A treatise on the mathematical theory of elasticity, 4th ed.,
Dover, New York.
Munteanu, L; Donescu, Şt. (2004): Introduction to Soliton Theory: Applications
to Mechanics, Book Series “Fundamental Theories of Physics”, vol.143, Kluwer
Academic Publishers.
Narayana Naik, G.; Gopalakrishnan, S.; Ganguli R. (2008): Design optimiza-
tion of composites using genetic algorithms and failure criterion. J. Composite
Structures, vol. 83, no. 4, pp. 354-367.
Plaut, R.H.; Johnson, L.W.; Olhoff, N. (1986): Bimodal optimization of com-
pressed columns on elastic foundation. J. Appl. Mech. Trans. ASME, vol.53,
pp.130–134
Popescu, H.; Chiroiu, V. (1981) : Calculul structurilor optimale (Optimum design
of structures), Ed. Academiei, Bucharest.
Rajasekaran, S.; Vijayalakshmipai, G.A. (2005): Neural Networks, Fuzzy Logic
and Algorithms Synthesis and Applications. Prentice Hall of India, New Delhi,
p.439.
Seyranian, A.P. (1984): On the Lagrange problem. Mechanics of Solids (Mekh.
Tverd. Tela), vol. 19, pp.100–111.
Seyranian, A.P., Privalova, O. (2003): The Lagrange problem on optimal column:
old and new results. Struct. Multidisc. Optimization, vol.25, pp.393–410.
Shin, Y.S.; Hafka, R.T.; Plaut, R.H. (1988a): Simultaneous analysis and design
for eigenvalue maximization. AIAA J, vol.26, pp.738–744.
Shin, Y.S.; Hafka, R.T.;Watson, L.T., Plaut, R.H. (1988b): Tracing structural
optima as a function of available resources by a homotopy method. Comput. Meth-
ods Appl. Mech. Engrg., vol.70, pp.151–164.
Solomon, L. (1968): Elasticité Linéaire, Masson.
Şoós, E. (1974): Discrete and continuum models of solids (in romanian), Ed.
Stiin?., Bucharest.
Truesdell, C.; Toupin, R. (1960): The Classical Field Theories, in S. Flugge (ed.)
Enc. of Physics, Springer, Berlin et al. III/1.
26 Copyright © 2010 Tech Science Press CMES, vol.65, no.1, pp.1-26, 2010

Tsuru, H. (1986): Nonlinear dynamics for thin elastic rod, Journal of the Physical
Society of Japan, vol. 55, no. 7, pp.2177–2182.
Tsuru, H. (1987): Equilibrium Shapes and Vibrations of Thin Elastic Rod, Journal
of the Physical Society of Japan, vol. 56, no. 7, pp.2309–2324.
Zheng, J.; Long, S.; Xiong, Y.; Li, G.(2009): A finite volume meshless local
Petrov-Galerkin method for topology optimization design of the continuum struc-
tures, CMES: Computer Modeling in Engineering & Science, vol.42, no.1, pp.19-
34.
Zhou S.; Wang M.Y. (2006): 3D multi-material structural topology optimization
with the generalized Cahn-Hilliard equations. CMES: Computer Modeling in En-
gineering & Sciences, vol.16, no.2, pp.83-101.
Zhou M.; Rozvany G.I.N. (2001): On the validity of ESO type methods in topol-
ogy optimization. Structural and Multidiscipline Optimization, vol.21, pp.80-83.
Wang S.Y.; Wang M.Y. (2006): Structural shape and topology optimization using
an implicit free boundary parameterization method CMES: Computer Modeling in
Engineering & Sciences, vol.13, no.2, pp.119-147.
Wang S.Y.; Lim K.M.; Khoo B.C.; Wang M.Y. (2007a): A geometric deforma-
tion constrained level set method for structural shape and topology optimization.
CMES: Computer Modeling in Engineering & Sciences, vol.18, no.3, pp.155-181.
Wang S.Y.; Lim K.M.; Khoo B.C.; Wang M.Y. (2007b): An unconditionally
time-stable level set method and its application to shape and topology optimization.
CMES: Computer Modeling in Engineering & Sciences, vol.21, no.1, pp.1-40.