Академический Документы
Профессиональный Документы
Культура Документы
Correspondence to: A. A. Pisano, DASTEC Dipartimento Arte Scienza e Tecnica del Costruire, University
Mediterranea of Reggio Calabria, Via Melissari, 89124 Reggio Calabria, Italy.
E-mail: aurora.pisano@unirc.it
Contract/grant sponsor: Italian Ministero dellIstruzione, dellUniversit` a e della Ricerca (MIUR)
Copyright q 2006 John Wiley & Sons, Ltd.
72 A. A. PISANO AND P. FUSCHI
of engineering problems (see e.g. [2, 3]). As noted in Del Piero [4], the tendency of using limit
analysis approaches and Druckers theorems mentioned above to evaluate the bearing capacity of
soils and structures made of stone or concrete is witnessed by a number of papers, coeval of the
founding one, e.g. [58]. More recently, in [4], the applicability of the two fundamental theorems
of limit analysis was proved outside the domain of perfect plasticity for a class of materials called
normal linear elastic including no-tension materials as a particular case. In the context of non-
standard materials, namely for soils, a theory of limit analysis was proposed by Radenkovic in
the early 1960s (see e.g. [9]) and, on these bases, by Josselin de Jong [10] and Palmer [11]. In
this context, with different approaches, limit analysis theory has been proposed by Atkinson and
Potts [12] and recently by a number of researchers [1316].
On the other hand, the development of numerical procedures, based on nite element formu-
lations, was a further essential contribution to the success of such a direct method (see e.g. the
early contributions of Hodge and Belytschko [17, 18] up to the comprehensive work of Save [19]).
Combining the mathematical programming algorithms and the nite element technique, a number
of simplied analytical methods have also been used to compute the lower and upper bound limit
loads according to static and kinematic theorems. However, the non-linearity and non-smoothness
of the objective function in the upper bound procedure and the strong physical non-linearity and
unidirectionality of the constraints in lower bound analysis render most of these methods time-
consuming and costly for meaningful structural problems. To this concern, alternative approaches,
based on the nite element method (FEM) and mathematical programming technique combined in
an iterative fashion, have recently been proposed to avoid some of the above drawbacks [20, 21].
Indeed, the development in the last few decades of elasticplastic step-by-step numerical anal-
yses, able to follow the structural response up to collapse, attracted the researchers interest in the
evolutive analyses against the direct methods and this at least for structures whose constitutive
material behaviour is isotropic and governed by well-established yield conditions. If it is possible
to share the opinion that limit analysis on structures made of elastic perfectly plastic materials is,
or may be, obsoleteit is very often more cumbersome than an evolutive step-by-step analysis
such a belief cannot be accepted when dealing with materials whose available constitutive laws
are unable to catch the complexity of the phenomena characterizing the actual post-elastic material
behaviour. Limit analysis on structures made of such materials is, in the authors opinion, an effec-
tive tool for dening, although approximately, the actual bearing capacity of the structures and is
therefore useful for design purposes. This is even more valid for composite material structures for
which the evolutive numerical analyses, although grounded on subtle constitutive material models
that take into account phenomena like interlaminar behaviour, delamination, damage evolution,
etc., turn out to be effective only for solving specic case-studies. Moreover, in the context of
composites, almost all the available constitutive models are based on material constants that are
difcult to identify via laboratory tests thus resulting useless for practical engineering applications.
The current interest in research on limit analysis approaches in the eld of composite material
structures is witnessed by several contributions, see e.g. [15, 2225].
The present paper belongs to this research line whose main goal is the extension to orthotropic
composite laminates of a procedure known in literature as linear matching method, whose general
characteristics are the same of the elastic compensation method [2628]. The LMM has been
successfully applied for isotropic yield conditions [29, 30] also in the case of pressure dependency
and non-associativity [3133] and is here applied to one of the most popular criterion for composite
laminates, namely the TsaiWu failure criterion [34, 35] and, in particular, to a second-order tensor
polynomial form of it, the latter assumed as yield condition. The use of LMM in conjunction with
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 73
such yield criterion guarantees the convergence of the whole procedure on the base of the sufcient
condition given in [31] for a general class of yield conditions including the one here assumed.
In such a context, the numerical strategies involved in the whole procedure have been suitably
modied and, as shown by a few simple examples, they seem to be effective, exible and easily
applicable to other criteria for anisotropic materials.
The assumption of a perfectly plastic material behaviour obeying a yield condition given by
the TsaiWu criterion can appear too coarse for a certain class of composites, but it is denitively
acceptable for composites such as metal matrix reinforced by metal bres, plastics reinforced by
kevlar or glass bres that, as is known, undergo plastic ow and considerable ductility. The lack
of associativity, on the other hand, is overcome by the non-standard limit analysis approach
proposed by Radenkovic, see Lubliner [36]. In truth, the gap between the plastic collapse load and
the failure load for non-associative materials exhibiting limited ductility, may be large. However,
the proposed method also gives a lower bound to the collapse load and the adopted constitutive
criterion seems to be very effective for the evaluation of the load bearing capacity of a class of
structural problems for which the above gap is reasonably small. Framing the present study in a
wider context, the proposed approach is concerned with limit analysis of a class of anisotropic
structures characterized by a yield function in the form of a general quadratic stress function.
The outline of the paper is as follows. After this introductory Section, in Section 2 the adopted
constitutive material model and all the hypotheses concerning the material behaviour are given.
Some basic concepts of non-standard limit analysis are also briey summarized and the basic
formulae for the upper and lower bound collapse multipliers evaluation are set up. Section 3,
which is the core of the paper, is devoted to the generalization of the LMM to orthotropic
materials. The fundamental assumptions are reported and the iterative procedure is explained in
detail using ow-chart style. Two numerical examples, with the main purpose of validating the
whole procedure, are given in Section 4. Section 5 closes the paper drawing some conclusions and
forecasting some possible developments of the present study.
Notation: The subscripts denote Cartesian components and the repeated index summation rule
is applied. Bold face symbols denote vectors or tensors. Cartesian orthogonal co-ordinates are
employed. The symbol := means equality by denition. Other symbols will be dened in the text
where they appear for the rst time.
2. CONSTITUTIVE ASSUMPTIONS AND PROBLEM POSITION
For anisotropic materials Tsai and Wu [34], see also [35], proposed a (failure) criterion in a tensor
polynomial form. They postulated that a failure surface in six-dimensional stress space exists in
the form:
F
i
o
i
+ F
i j
o
i
o
j
=1 (i, j =1, . . . , 6) (1)
where: F
i
and F
i j
are strength tensors of the second and fourth rank, respectively, and the con-
tracted notation, usually adopted in this context [37] is used (i.e. o
4
:=t
23
, o
5
:=t
31
, o
6
:=t
12
).
In this form it is difcult to untangle the TsaiWu criterion and also for that which concerns the
experimental identication of the material parameters entering the tensors F
i
, F
i j
pertaining to
a given material. Nevertheless, as is known, composite laminates consist of many layers stacked
up with different bre reinforcement orientation. Each layer can be viewed as a unidirectional
orthotropic lamina and the orthotropic properties of individual layers result in a material that
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
74 A. A. PISANO AND P. FUSCHI
is anisotropic. Subsequently attention is then focused on a unidirectional orthotropic laminate
adding the further simplifying hypothesis of plane stress conditions. This restriction, in the
authors opinion, does not affect the general applicability of the whole procedure to be developed
hereafter and referred to orthotropic laminates. Possible improvements are certainly conceivable
if each layer of a multilayered laminate is analysed in conjunction with concepts like rst ply
failure [38] or progressive failure scenario [39]. This point is not addressed here because it is
outside the scope of the present paper.
For a undirectional reinforced lamina in plane stress state (i, j =1, 2, 6) the TsaiWu criterion
simplies as follows (1 and 2 denoting the principal directions of orthotropy):
F
11
o
2
1
+ F
22
o
2
2
+ F
66
o
2
6
+ 2F
12
o
1
o
2
+ 2F
16
o
1
o
6
+ 2F
26
o
2
o
6
+ F
1
o
1
+ F
2
o
2
+ F
6
o
6
=1 (2)
where: F
i
and F
i j
(i, j =1, 2, 6) have to be determined by tensile, compressive and shear tests.
Taking into account that the unidirectional laminate is referred to its orthotropic axes and that the
strength should be unaffected by the direction or sign of the shear stress component, all the terms
in Equation (2) containing rst-degree shear stresses have to be neglected. Equation (2) simplies
subsequently, i.e.:
F
11
o
2
1
+ F
22
o
2
2
+ F
66
o
2
6
+ 2F
12
o
1
o
2
+ F
1
o
1
+ F
2
o
2
=1 (3)
where, see e.g. [37]:
F
1
:=
1
X
t
+
1
X
c
(4a)
F
2
:=
1
Y
t
+
1
Y
c
(4b)
F
11
:=
1
X
t
X
c
(4c)
F
22
:=
1
Y
t
Y
c
(4d)
F
66
:=
1
S
2
(4e)
F
12
:=
1
2
_
F
11
F
22
(4f)
with X
t
, X
c
the longitudinal tensile and compressive strengths, respectively; Y
t
, Y
c
the transverse
tensile and compressive strengths, respectively, and S the longitudinal shear strength. As deducible
by inspection of Equations (4a)(4e), ve of the six coefcients required for the denition of the
criterion are given by performing simple tests. The sixth, namely F
12
, related to the interaction
between the two normal stress components o
1
and o
2
requires a biaxial test. This experimental
task is not easy to perform as the simple uniaxial or shear tests and simplied assumptions, as the
one adopted herein with position (4f), are usually made [40, 41].
It is worth noting that the estimation of the TsaiWu strength parameters is not straightforward
(see e.g. [42, 43]); moreover, considering that a failure process in a laminate involves a combination
of failure mechanisms due to matrix crushing, bre breaks, bre buckling, delaminations [44]
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 75
modied versions of the TsaiWu criterion have been presented to take into account some its
internal incoerencies [40, 45].
Despite the above remarks the TsaiWu criterion, in the quadratic form adopted herein, is
simple; it allows one to apply the standard rules of transformation, invariance and symmetry; it
also contemplates interactions among the stress or strain components analogously to the von Mises
criterion for isotropic materials. As noted by who conceived this criterion [35], failure criteria for
composite laminates should provide a convenient framework or model for mathematical operations.
The framework should remain the same for different denitions of failures, such as ultimate strength,
yielding, endurance limit, etc.; the criteria are not intended to explain the complex mechanisms of
failure that in composite laminates are characterized by the concurrent and sequential occurrence of
many failure modes. With this conjecture the TsaiWu criterion is used for dening an admissible
stress states domain. Points within the domain locate stress state pertaining to an anisotropic linear
elastic behaviour of the material. Points lying on the domain boundary locate stress states at which
the material has exhausted its strength capabilities.
Equation (3) can be rearranged in terms of the following dimensionless parameters:
X :=
_
F
11
o
1
(5a)
Y :=
_
F
22
o
2
(5b)
Z :=
_
F
66
o
6
(5c)
f
12
:=
F
12
F
11
F
22
(6a)
f
1
:=
F
1
F
11
(6b)
f
2
:=
F
2
F
22
(6c)
With the above positions Equation (3) reads:
X
2
+ Y
2
+ Z
2
+ 2 f
12
XY + f
1
X + f
2
Y =1 (7)
which, in the dimensionless space (X, Y, Z), individuates an ellipsoid whose major axis lies on
Z =0 plane and it is rotated by a counterclockwise angle of 45
TW
=0 (8c)
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
76 A. A. PISANO AND P. FUSCHI
Moreover, denoting with a
TW
, b
TW
and c
TW
the semi-axes ellipsoid dimensions (a
TW
referring to
the major axis, c
TW
to the axis parallel to Z) it is
a
TW
= [(1 + f
12
)/(1 + :
2
TW
+ [
2
TW
+ 2 f
12
:
TW
[
TW
)]
1/2
(9a)
b
TW
= [(1 f
12
)/(1 + :
2
TW
+ [
2
TW
+ 2 f
12
:
TW
[
TW
)]
1/2
(9b)
c
TW
= [1 + :
2
TW
+ [
2
TW
+ 2 f
12
:
TW
[
TW
]
1/2
(9c)
The TsaiWu surface in the shape of Equation (3), or in the equivalent dimensionless
Equation (7), is assumed as yield surface for the orthotropic material here considered. This
assumption, unfortunately, is not sufcient to proceed further. Even though it postulates the exis-
tence of a yield surface for the composite laminate, which will be treated as an elastic perfectly
plastic material, it does not imply the associativity of the yield criterion. In spite of that, in the
context of non-standard materials, namely for soils, a theory of limit analysis was actually pro-
posed by Radenkovic [9] with several modications (see e.g. [36] and references therein). The limit
analysis fundamental theorems have in practice been restated in the shape of upper and lower
bound theorems. Precisely, after [36], Radenkovics rst theoremor upper bound theorem
states: the limit loading for a body made of a non-standard material is bounded from above by
the limit loading for the standard material obeying the same yield criterion. Radenkovics second
theoremor lower bound theoremstates: the limit loading for a body made of a non-standard
material is bounded from below by the limit loading for the standard material obeying the yield
criterion g(r) =0. g(r) =const. being a convex function lying entirely within the yield surface of
the non-standard material, say f (r) =0, and complying with the condition that to any r at which
f (r) =0 there corresponds on g(r) =0 a r
to
the surface g(r) =0 and the inequality (o
i j
o
i j
) c
p
i j
0 holds true. Proof of the above theorems
is reported in [9] (see also [36]). After all, every value of the limit load for a non-standard body
is located between two xed boundaries dened by the values of the limit loads for two corre-
sponding standard materials. Obviously, Radenkovics two theorems locate a range of collapse
load multiplier values, because for non-standard materials even the uniqueness of the limit load is
uncertain due to the absence of an uniqueness for the stress eld.
Following the directions of the above theorems and taking into account the strict convexity
of the TsaiWu surface, which for Radenkovics lower bound theorem can itself play the role of
g(r) =0thus satisfying condition (o
i j
o
i j
) c
p
i j
0 always as an equalityit is therefore possible
to search for an upper and a lower bound on the collapse load multiplier with reference to the
TsaiWu surface.
2.1. Problem position: upper and lower bound multipliers evaluation
Consider a body of volume V, external surface *V, referred to a Cartesian co-ordinate system
(x
i
, i =1, 2, 3) and subjected to loads P p(x), where: P is the scalar load multiplier; p(x) the
reference load vector collecting all the surface force components, p
i
, acting on points of a portion
of the body surface, namely *V
t
; for simplicity, only surface forces are considered. The remaining
part of *V, namely *V
u
=*V *V
t
, is assumed to suffer displacements u=0; plane stress
conditions are also assumed. The material is, by hypothesis, orthotropic, homogeneous and with
a constitutive behaviour obeying the TsaiWu criterion in the form given by Equation (3). In
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 77
the assumed hypothesis of associated ow rule and rewriting Equation (3) in the abridged form
f (o
j
) =0 ( j =1, 2, 6), the strain rate components at collapse can be expressed in the form:
c
j
=
z
* f
*o
j
(10)
where
z>0 is a scalar multiplier and c
j
are the components of the outward normal to the yield
surface f (o
j
) =0.
For a given distribution of compatible strain rates c
j
, say c
c
j
, i.e. such that the related displacement
rates u
c
i
satisfy the condition u
c
i
=0 on *V
u
, an upper bound to the collapse limit load multiplier
is given by
P
UB
_
*V
t
p
i
u
c
i
d*V =
_
V
o
y
j
c
c
j
dV (11)
where: P
UB
denotes the upper bound multiplier; o
y
j
the stresses at yield associated to given
compatible strain rates c
c
j
; u
c
i
the related displacement rates. The set ( c
c
j
, u
c
i
) denes a collapse
mechanism.
If at every point within V a stress eld o
j
exists satisfying the condition f ( o
j
)0 and in
equilibrium with P p(x) on *V
t
for a value of P, say P
LB
, then P
LB
is a lower bound on the
collapse limit load multiplier.
3. THE LINEAR MATCHING METHOD FOR ORTHOTROPIC MATERIALS
The LMM, [2933], is a programming technique involving an iterative FE-based numerical pro-
cedure which performs a sequence of linear analyses on the structure made, by hypothesis, of a
linear viscous ctitious material with spatially varying moduli. At each iteration an adjustment of
the ctitious moduli is carried out so that the computed ctitious stresses are brought on the yield
surface at a xed strain rate distribution. This allows one to dene a collapse mechanism, the related
stresses at yield and, consequently, an upper and a lower bound to the collapse load multiplier.
In the present context the LMM utilizes a ctitious linear viscous material which is orthotropic
and subjected to a distribution of imposed initial stresses. The key ideas of the proposed generaliza-
tion are summarized as follows with reference to orthotropic laminates under plane stress conditions.
Let us consider the body, of volume V, made of a ctitious, linear, viscous, orthotropic material,
with spatially varying moduli and suffering a distribution of imposed initial stresses. This ctitious
material has a complementary dissipation rate given by
W(o
j
) =
1
2
_
o
2
1
E
1
+
o
2
2
E
2
+
o
2
6
E
6
2v
12
o
1
o
2
E
2
2
_
o
1
E
1
v
12
o
2
E
2
_
o
1
2
_
o
2
E
2
v
12
o
1
E
2
_
o
2
2
o
6
E
6
o
6
+
o
2
1
E
1
+
o
2
2
E
2
+
o
2
6
E
6
2v
12
o
1
o
2
E
2
_
(12)
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
78 A. A. PISANO AND P. FUSCHI
where o
j
are imposed initial stresses, E
j
are spatially varying moduli and v
12
is the Poisson ratio.
For this ctitious material and at a xed value of the load multiplier a linear analysis is performed
to compute: the distribution of strain rates, c
e
j
=*W(o
e
j
/*o
e
j
), related stresses, o
e
j
, and compatible
displacement rates, u
e
i
, of the points at which surface loads are applied. This ctitious solution is
computed, in principle, at each point of the body volume V and, in practice, at each Gauss point
(GP) of each nite element (FE) of the discretized domain V. At each GP, the ctitious kinematic
solution ( c
e
j
, u
e
i
) is forced to represent a collapse mechanism, namely it is forced to identify with
( c
c
j
, u
c
i
) of Equation (11). To this aim, if c
e
j
is kept xed and assumed as c
c
j
, it is sufcient to
compute the stress at yield associated to c
c
j
c
e
j
, namely o
y
j
, by varying the ctitious moduli and
initial stresses so that o
e
j
coincides with o
y
j
, u
e
i
being the compatible displacements associated to
c
c
j
. The linear material is so matched to the yield surface and this, performed to within a discretized
FE approach, is carried out in an iterative fashion.
At operative level, grounding on the formal analogy between the linear viscous problem and the
linear elastic problem, the ctitious linear solution can be computed as a ctitious elastic solution,
W(o
j
) of Equation (12) playing the role of complementary energy potential of a ctitious elastic
material. The ctitious elastic analyses, performed at each iteration, can be carried on by any
commercial FE-code rendering the whole procedure easy to be performed.
From a geometrical point of view, conceivable in plane stress hypothesis, the matching procedure
merely states that the complementary energy equipotential surface of the ctitious elastic material,
W(o
j
) =const., by appropriate updating of the spatially varying elastic parameters and initial
stress values, is brought to be tangential to the TsaiWu surface at the stress point r
y
whose
external normal is e
c
. In Figure 1 the matching is schematically depicted with reference to the
TsaiWu surface and the W(o
j
) =const. surface. The dependence of the latter surface on the set
of elastic parameters and initial stress ctitious values is highlighted: ()
(0)
denoting an arbitrary
initial set of such values; ()
()
denoting the modied values which achieve the matching.
Figure 1. Matching procedure at the generic Gauss point: geometrical sketch in the o
6
=0 plane.
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 79
In the present case, such matching is easily obtainable by taking advantage of the ellip-
soidal shapes of the TsaiWu surface and of the equipotential surface W(o
j
) =const.; these
surfaces can, in fact, be made coincident if the ctitious material is, from the beginning,
endowed with a complementary energy equipotential surface homothetic to the TsaiWu sur-
face. To this aim, rewriting Equation (12) in the dimensionless space (X, Y, Z), i.e. using
Equations (5a)(5c), it is
W(X, Y, Z) =
1
2
_
X
2
E
1
F
11
+
Y
2
E
2
F
22
+
Z
2
E
6
F
66
2v
12
E
2
F
11
F
22
XY
F
11
_
X
E
1
F
11
v
12
Y
E
2
F
22
_
X
2
F
22
_
Y
E
2
F
22
v
12
X
E
2
F
11
_
Y
2
Z
E
6
F
66
Z
+
X
2
E
1
F
11
+
Y
2
E
2
F
22
+
Z
2
E
6
F
66
2v
12
E
2
F
11
F
22
X
Y
_
(13)
For a given load multiplier initial value, say P
(0)
UB
, and any xed set of elastic parameters and initial
stresses, namely (E
(0)
1
, E
(0)
2
, E
(0)
6
, v
(0)
12
,
X
(0)
,
Y
(0)
,
Z
(0)
), the above expression individuates in the
(X, Y, Z) space an ellipsoid of the form W[E
(0)
j
, v
(0)
12
, ,
(0)
j
] =
W
(0)
const. The latter abridged
form points out the dependence of the ellipsoid location and amplitude on the elastic parameters
and initial stress values,
W
(0)
being the pertinent complementary energy equipotential value corre-
sponding to the given loads. For brevity, ,
j
for j =1, 2, 6 identies with X, Y, Z, respectively. If
the initial choice is made by imposing that W[E
(0)
j
, v
(0)
12
, ,
(0)
j
] =
W
(0)
is homothetic to the TsaiWu
surface given by Equation (7), i.e.: the semi-axes ratios are equal (three conditions); the two ellip-
soids have the same centre (three conditions) and the main axis is rotated by a counterclockwise
angle of 45
with respect to the X axis (one condition); it easy to verify that the following positions
hold true:
E
(0)
1
=
1
2F
11
(14a)
E
(0)
2
=
1
2F
22
(14b)
E
(0)
6
=
1
2F
66
(14c)
v
(0)
12
= f
12
F
11
F
22
(14d)
X
(0)
=:
TW
(15a)
Y
(0)
=[
TW
(15b)
Z
(0)
=0 (15c)
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
80 A. A. PISANO AND P. FUSCHI
Figure 2. Matching procedure at the generic Gauss point for W[E
(0)
j
, v
(0)
12
, ,
(0)
j
] =
W
(0)
homothetic to TW
surface: geometrical sketch in the Z =0 plane.
With the above positions Equation (13) reduces to
X
2
+ Y
2
+ Z
2
+ 2 f
12
XY 2(:
TW
+ f
12
[
TW
)X 2([
TW
+ f
12
:
TW
)Y
=
W
(0)
:
2
TW
[
2
TW
2 f
12
:
TW
[
TW
(16)
which is the searched complementary energy equipotential surface homothetic to the TsaiWu
surface. The matching and therefore the whole procedure is now more easily realizable acting
only on the elastic moduli values of the ctitious material that control the axes amplitude of the
ellipsoid given by Equation (16). In Figure 2 a geometrical sketch of such matching is given in
the Z =0 plane and at a generic GP.
Looking at the sketch of Figure 2, the following can be stated: an elastic analysis on the structure
loaded by P
(0)
UB
p
i
and made of a ctitious material whose complementary energy is given by (16)
produces, at each GP, an elastic solution of the form (,
e(0)
j
, c
e(0)
j
), (point A in Figure 2), lying
on the surface W[E
(0)
j
, v
(0)
12
, ,
(0)
j
] =
W
(0)
. Assuming c
e(0)
j
as c
c
j
the (adimensionalized) stress at
yield, ,
y()
j
, associated to the (normal) c
c
j
is computed (point B in Figure 2). The ctitious elastic
solution, ,
e(0)
j
, is then forced to identify with the one at yield, namely ,
e()
j
,
y()
j
, by rescaling
the ctitious elastic moduli.
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 81
In detail, based on the homothetic condition, a rescaling of the ctitious elastic moduli can be
carried out on setting:
E
()
j
=
E
(0)
j
W
(0)
, j =1, 2, 6 (17)
while keeping the load xed, namely P
(0)
UB
, the initial stresses ,
(0)
j
, the Poisson coefcient v
(0)
12
and
W
(0)
. In Equation (17), :=1+:
2
TW
+2 f
12
:
TW
[
TW
+[
2
TW
>0 is the known term of the TsaiWu
ellipsoid equation rewritten in a cartesian reference system with the origin at the ellipsoid centre
and equipollent to the (X, Y, Z) system. It is easy to verify that on substituting (17) in (13) the
TW ellipsoid, given by Equation (7), is obtained.
On the other hand, the stresses at yield, ,
y()
j
, can be computed (again referring to Figure 2)
with the following equations:
X
y()
= [1
(0)
]:
TW
+
(0)
X
e(0)
(18a)
Y
y()
= [1
(0)
][
TW
+
(0)
Y
e(0)
(18b)
Z
y()
=
(0)
Z
e(0)
(18c)
where, for clarity, all the stress components, ,
y()
j
for j =1, 2, 6, have been explicitly reported;
:
TW
and [
TW
are given by Equations (8a), (8b) and
(0)
denotes the homothety ratio between the
two ellipsoids, namely:
(0)
:=
a
TW
a
(0)
W
=
b
TW
b
(0)
W
=
c
TW
c
(0)
W
(19)
In Equation (19): a
TW
, b
TW
, c
TW
are the lengths of the TW ellipsoid semi-axes given by
Equations (9a)(9c) and a
(0)
W
, b
(0)
W
and c
(0)
W
are the analogous quantities for the ellipsoid
W[E
(0)
j
, v
(0)
12
, ,
(0)
j
] =
W
(0)
. It easy to verify that
(0)
=
_
W
(0)
(20)
However, the stresses at yield o
y()
j
, given by Equations (18a)(18c) by application of
Equations (5a)(5c), will not satisfy the equilibrium conditions pertaining to the loads P
(0)
UB
p
i
but, remembering Equation (11), they will satisfy the equilibrium requirements for loads p
i
amplied by
P
()
UB
=
_
V
o
y()
j
c
c
j
dV
_
*V
t
p
i
u
c
i
d*V
(21)
that is the load multiplier value pertinent to the E
()
j
distribution actuating the matching at each GP.
A new elastic analysis, performed with loads P
()
UB
p
i
and E
()
j
distribution of Equation (17), will
give at each GP a ctitious elastic solution lying on the surface W[E
()
j
, v
(0)
12
, ,
(0)
j
] =
W
()
, the latter
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
82 A. A. PISANO AND P. FUSCHI
obviously will not coincide with W[E
()
j
, v
(0)
12
, ,
(0)
j
] =
W
(0)
. The rationale can then be repeated in
an iterative fashion making use, at the kth elastic analysis, of an elastic moduli distribution given by
E
(k)
j
=
E
(k1)
j
W
(k1)
, j =1, 2, 6 (22)
and for loads amplied by
P
(k)
UB
=
_
V
o
y(k1)
j
c
c(k1)
j
dV
_
*V
t
p
i
u
c(k1)
i
d*V
(23)
The recursive formulae for stresses at yield can easily be derived by looking at Equations
(18a)(18c) and, remembering Equations (5a)(5c), they read
o
y(k)
1
=
X
y(k)
F
11
(24a)
o
y(k)
2
=
Y
y(k)
F
22
(24b)
o
y(k)
6
=
Z
y(k)
F
66
(24c)
where
X
y(k)
= [1
(k)
]:
TW
+
(k)
X
e(k)
(25a)
Y
y(k)
= [1
(k)
][
TW
+
(k)
Y
e(k)
(25b)
Z
y(k)
=
(k)
Z
e(k)
(25c)
Finally, the homothety ratio can be given the recursive expression:
(k)
:=
a
TW
a
(k)
W
=
_
W
(k1)
W
(k)
(26)
By substituting (22) in (13) it is easy to verify that the following expression is obtained (at the
generic GP):
X
2
+ Y
2
+ Z
2
+ 2 f
12
XY + f
1
X + f
2
Y =1 +
_
W
(k)
W
(k1)
1
_
(27)
the latter, in few iterations, identies with the TW Equation (7) and this when, at a certain k,
W
(k)
identies with
W
(k1)
and the matching is accomplished. At this k it is also
(k)
=1 and
,
y(k)
j
,
e(k)
j
as it has to be and as deducible by Equations (25a)(25c) and (26). The iterative
procedure is monitored by means of the computed P
(k)
UB
value, i.e. it stops when the difference
| P
(k)
UB
P
(k1)
UB
| is less than a xed tolerance.
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 83
Concerning the lower bound collapse multiplier, P
LB
, the LMM, even in its original formulation
(see e.g. [29]), provides a method for an approximate evaluation of such a bound. This issue, with
different approaches, has been addressed in [46] (see also references therein) and in [20] until,
although the list is not exhaustive, the recent contribution given in Hamilton and Boyle [47].
Hereafter the P
LB
evaluation is supplied within the above discussed iterative procedure and, in
this sense, it supplements the extension to structures made of orthotropic materials of the original
version of the LMM.
At each iteration and at each GP, the ctitious stresses v
e(k)
, pertinent to loads P
(k)
UB
p
i
and
Young moduli distribution E
(k)
j
, are located in the (X, Y, Z) space; see also the schematic sketch
given in Figure 3 for three generic GPs in the Z =0 plane. Among all the stress points v
e(k)
thus
obtained the one further away from the TW surface is detected, say v
e(k)
F
(point A in Figure 3),
and this merely by computing the Euclidean distances from the TW ellipsoid center. The ratio
j
(k)
between the yield stress value measured on the direction v
e(k)
F
/|v
e(k)
F
|, say v
y(k)
F
(point B in
Figure 3), over the stress value v
e(k)
F
allows one to dene a lower bound given by
P
(k)
LB
=j
(k)
P
(k)
UB
(28a)
Figure 3. Fictitious elastic stresses v
e(k)
at iteration kth and at three generic Gauss points: evaluation of
the factor j
(k)
for the P
(k)
LB
computation.
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
84 A. A. PISANO AND P. FUSCHI
with
j
(k)
:=
|v
y(k)
F
|
|v
e(k)
F
|
(28b)
a rescaling of the applied loads P
(k)
UB
p
i
by j<1 implies that all the ctitious elastic stresses j
(k)
v
e(k)
satisfy the admissibility conditions of the static approach for limit analysis.
3.1. The iterative procedure
The whole procedure is summarized hereafter in ow-chart style recalling also the equations to
be utilized in an FE implementation and pointing out a number of operating choices.
Initialization:
Step #i: Knowing the strength values of the (real) constituent material (X
c
, X
t
, Y
c
, Y
t
, S),
assign to all FEs in the mesh the initial set of ctitious elastic parameters and initial
stresses, namely:
E
(0)
1
=1/(2F
11
), E
(0)
2
=1/(2F
22
), E
(0)
6
=1/(2F
66
), v
(0)
12
= f
12
_
F
11
/
_
F
22
o
(0)
1
=:
TW
/
_
F
11
, o
(0)
2
=[
TW
/
_
F
22
, o
(0)
6
=0
Step #ii: Set k =1, P
(k1)
UB
= P
(0)
UB
=1 (for k =1, P
(0)
UB
can be any arbitrary value) and compute
=1 + :
2
TW
+ 2 f
12
:
TW
[
TW
+ [
2
TW
for later use
Iteration loop:
Step #1: Perform a ctitious elastic analysis with elastic parameters E
(k1)
j
, v
12
=v
(0)
12
, initial
stresses o
j
= o
(0)
j
and with loads P
(k1)
UB
p
i
, computing a ctitious elastic solution,
namely:
c
e(k1)
j
, u
(k1)
i
, o
e(k1)
j
at Gauss point level
Step #2: Compute the adimensionalized stresses ,
e(k1)
j
and evaluate j
(k1)
Step #3: Compute the value of the complementary potential energy
W
(k1)
=
1
2
o
e(k1)
j
c
e(k1)
j
Step #4: Compute the homothety ratio, namely
(k1)
=
_
_
_
/
W
(0)
for k =1
_
W
(k2)
/
W
(k1)
for k>1
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 85
Step #5: Evaluate stresses at yield:
o
y(k1)
1
= [1
(k1)
]
:
TW
F
11
+
(k1)
o
e(k1)
1
o
y(k1)
2
= [1
(k1)
]
[
TW
F
22
+
(k1)
o
e(k1)
2
o
y(k1)
6
=
(k1)
o
e(k1)
6
Step #6: Compute the E
(k)
j
distribution to be utilized, if necessary, at next iteration, namely:
E
(k)
j
=
E
(k1)
j
W
(k1)
end Gauss point level
Step #7: Set c
c(k1)
j
= c
e(k1)
j
, u
c(k1)
i
= u
e(k1)
i
and evaluate the upper bound multiplier
P
(k)
UB
=
_
V
o
y(k1)
j
c
c(k1)
j
dV
_
*V
t
p
i
u
c(k1)
i
d*V
Step #8: Evaluate lower bound multiplier
P
(k1)
LB
=j
(k1)
P
(k1)
UB
Step #9: Plot P
(k)
UB
and P
(k1)
LB
versus iterations number
Step #10: Check for convergence
| P
(k)
UB
P
(k1)
UB
|TOL
_
_
YES EXIT
NOT set()
(k1)
= ()
(k)
and GOTO step #1
By inspection of the above ten-step procedure the whole analysis, also in this extended version
for orthotropic materials, is easy to implement and, like the original LMM, can be carried out by
any commercial FE code suitably fed, at each iteration, by the ctitious elastic moduli distribution
accomplishing the matching at each GP. Nevertheless some remarks have to be made.
As noted in [29], the correctness of the P
UB
depends on the kinematic description of the
discretized problem and it is then related to the adopted FE mesh. In this sense the P
UB
converges
to the minimum upper bound allowed by the class of displacement elds given by the mesh itself.
This drawback is easily overcome by using ne meshes in the analysis.
The rationale followed for the P
LB
evaluation, on the other hand, gives a lower bound to the
above minimum upper bound; Equation (28a) yields, in fact, a pseudo-lower bound. Moreover, the
P
LB
, as evaluated above, appears to be too conservative because it depends on only one stress value
attained at one GP in the whole mesh (the one further away from the TW surface). Nevertheless,
the static approach of limit analysis essentially states that the structure rearranges the internal
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
86 A. A. PISANO AND P. FUSCHI
stresses to its best possible advantage to withstand the applied loads. Based on this concept, the
GP stress values measured to within the single element where v
e(k)
F
has been attained at one GP
can be averaged. It would obviously be incorrect to average across elements since the elastic
moduli of adjacent elements are different. On this averaged stress value Equation (28b) can then
be applied (at step #2) to evaluate an averaged j, say j. A weighted lower bound, say P
WLB
, is
then computed (at step #8), namely:
P
(k)
WLB
= j
(k)
P
(k)
UB
(29)
A third remark regards the evaluation of the spatially varying elastic moduli E
(k)
j
at step #6. As
stated, Equation (22) yields the updated E
(k)
j
values at each GP in the FE mesh. However, to avoid
accuracy problems, in a FE procedure a unique set of E
j
is assigned to each single element and
so the E
(k)
j
evaluated at matching on the GPs of each element have been averaged to within the
element itself.
A nal remark concerns the convergence of the iterative procedure. A theoretic proof of the
convergence for the upper bounds sequence, in the case of Von Mises materials, was given
in [29]; a sufcient condition was then given in [31] for a general class of yield conditions
pertinent to pressure dependent materials. To the authors knowledge, no such proof exists for
the lower bounds sequence. The convergence of the upper bounds sequence is here assured
by the sufcient condition given in [31] for a general class of yield conditions including the
one here adopted. The upper bound FE solution converges, in fact, to the least upper bound
contained within the class of mechanisms described by the FE mesh as it is witnessed by the
satisfactory numerical results obtained. These are presented in the next Section for two nu-
merical examples, the rst one admitting an explicit analytical expression for the upper bound
multiplier.
4. NUMERICAL EXAMPLES
Two numerical examples have been solved to verify the effectiveness of the proposed approach. The
rst one, taken from [24], has been considered for the sake of comparison. In the quoted paper,
following a different approach, an explicit analytical expression of the upper bound multiplier
is provided for a simple problem, namely a square plate under plane stress conditions and, by
hypothesis, made of a tetratropic material. Despite its simplicity, the example carried out in the
above mentioned paper presents an interesting sensitivity analysis on the inuence of the ratio
between tensile and compressive strength as well as the degree of orthotropy of the material. The
second example, which analyses the same simple structural plane problem, envisages the case of
a fully orthotropic material and shows all the potentialities of the proposed approach that, even if
up until now has been conned to the simpler context of plane stress conditions, seems to be of
quite general applicability in the realm of orthotropic material structures.
All FE elastic analyses were performed with the ADINA code [48], suitably interfaced with
a FORTRAN main programme which controls the iterative procedure described in Section 3.1
thus performing the matching at GP level and, in practice, feeding the FE code with the
appropriate input parameters at each step. The value of TOL=10
4
was utilized. In both ex-
amples a mesh of 50 isoparametric 16 nodes quadrilateral elements with 16 GPs per element were
utilized.
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 87
P
0
a
a
x
y
1
2
o
P
0
Figure 4. Example #1square plate under plane stress conditions made of a tetratropic material.
4.1. Example #1
The square plate of side a and unit thickness shown in Figure 4 is uniformly loaded along two
opposite edges while preventing transverse displacements on the remaining edges (i.e. along y
referring to Figure 4). The load per unit length is specied as Po
0
, where o
0
is a given reference
stress value and P is a scalar load multiplier.
The plate, under plane stress conditions, is referred to a Cartesian co-ordinate system (x, y)
while the principal directions of orthotropy are individuated by the Cartesian axes (1, 2).
denes the counterclockwise angle between axis 1 and axis x along with the applied loads
act. As stated, the material is tetratropic, i.e. X
c
Y
c
and X
t
Y
t
, and the material consid-
ered in [24] obeys the TsaiWu criterion given here by Equation (3) with the following
positions:
F
11
= F
22
=
1
1 [
2
1
o
2
0
(30a)
F
66
=3p
1
1 [
2
1
o
2
0
(30b)
F
12
=
1
2(1 [
2
)
1
o
2
0
(30c)
F
1
= F
2
=
[
1 [
2
1
o
0
(30d)
In Equations (30a)(30d): [ is the ratio between tensile and compressive material strength, so it
represents the degree of symmetry of the behaviour of the material ([ =0 pertains to a symmetric
behaviour); p denes the degree of orthotropy of the material (p =1 means isotropy). (Refer to
the above mentioned paper for further details on these two parameters.)
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
88 A. A. PISANO AND P. FUSCHI
0 5 10 15 20 25 30 35 40 45
0.0
0.4
0.8
1.2
1.6
2.0
Capsoni et al. [24]
Present approach
= 0
= 0.5
= 0.25
= 0.25
P
U
B
[deg]
(a) (b)
0 5 10 15 20 25 30 35 40 45
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Capsoni et al. [24]
Proposed approach
= 0
= 0.5
= 4
= 0.25
P
U
B
[deg]
Figure 5. Example #1upper bound multiplier, P
UB
, versus angle (in degree) between the orthotropy
axis 1 and the loading direction x: results given by Capsoni et al. [24] (solid lines) and results obtained by
the present approach (lines with diamonds). Plots for [=0; 0.25; 0.5 and: (a) at p=0.25; and (b) at p=4.
0 5 10 15 20 25 30 35 40 45
0.0
0.3
0.6
0.9
1.2
1.5
1.8
= 0 = 0.25
P
UB
P
WLB
= 0.25
= 0.5
l
o
a
d
m
u
l
t
i
p
l
i
e
r
0 5 10 15 20 25 30 35 40 45
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
=
4
0
=
= 0.25
= 0.5
l
o
a
d
m
u
l
t
i
p
l
i
e
r
(b) (a)
P
UB
P
WLB
[deg] [deg]
Figure 6. Example #1upper bound (solid lines) and weighted lower bound (dashed lines) multipli-
ers versus angle (in degree) between the orthotropy axis 1 and the loading direction x. Plots for
[ =0; 0.25; 0.5 and: (a) at p =0.25; and (b) at p =4.
In Figures 5(a) and (b) the results obtained with the proposed approach are compared with those
of [24] in terms of upper bound multiplier, P
UB
, versus the angle and for different values of p
and [; precisely for p =0.25 and 4 the values of [ =0, 0.25, 0.5 were considered. For the same set
of p and [ values, Figures 6(a) and (b) show the P
UB
and the P
WLB
versus angle . In Figures 7(a)
and (b) the P
UB
, P
LB
and P
WLB
sequences are plotted versus iteration number again for p =0.25
and 4 and for [ =0.25 and =15
.
By inspection of Figures 5(a) and (b) the results obtained clearly show a highly satisfactory
agreement with those of [24]. The weighted lower bound values, as shown by Figures 6(a) and (b),
Copyright q 2006 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2007; 70:7193
DOI: 10.1002/nme
LIMIT ANALYSIS OF ORTHOTROPIC COMPOSITE LAMINATES 89
5 4 3 2 1 4 3 2 1 6 7
0.0
0.2
0.4
0.6
0.8
1.0
1.2
= 0.25
P
UB
P
WLB
= 0.25
= 15
iteration number
l
o
a
d
m
u
l
t
i
p
l
i
e
r
P
LB
5 6
0.0
0.2
0.4
0.6
0.8
1.0
1.2
= 4
= 0.25
= 15
iteration number
l
o
a
d
m
u
l
t
i
p
l
i
e
r
(a) (b)
P
UB
P
WLB
P
LB
Figure 7. Example #1upper bound (solid lines), weighted lower bound (dashed lines) and lower bound
(dotted lines) multipliers versus iteration number for [=0.25, =15