Direct Methods For Limit States in Structures, (Franck Pastor, Joseph Pastor, Djimedo Kondo (Auth.), Konstantinos Spiliopoulos, Dieter Weichert (Eds.) )
Direct Methods For Limit States in Structures, (Franck Pastor, Joseph Pastor, Djimedo Kondo (Auth.), Konstantinos Spiliopoulos, Dieter Weichert (Eds.) )
Direct Methods For Limit States in Structures, (Franck Pastor, Joseph Pastor, Djimedo Kondo (Auth.), Konstantinos Spiliopoulos, Dieter Weichert (Eds.) )
and Materials
Direct Methods
for Limit States
in Structures and
Materials
Editors
Konstantinos Spiliopoulos
School of Civil Engineering,
Department of Structural Engineering
National Technical University of Athens
Athens, Greece
Dieter Weichert
Institute of General Mechanics
RWTH-Aachen University
Aachen, Germany
ISBN 978-94-007-6826-0
ISBN 978-94-007-6827-7 (eBook)
DOI 10.1007/978-94-007-6827-7
Springer Dordrecht Heidelberg New York London
Library of Congress Control Number: 2013947157
Springer Science+Business Media Dordrecht 2014
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publishers location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
Foreword
In the history of structural engineering, safety with respect to collapse, rather than
stress admissibility, turns out to be the main worry of several pioneers. In the eighteenth century the Saint Peter dome in Rome and the highest spire of the Milan
cathedral (with the Madonnina on its top) have been structures investigated for
restoration (by Poleni and Boscovich, respectively, both primarily mathematicians)
from a standpoint and by a graphical-computational procedure which can be regarded as anticipations of limit analysis methods.
Another memorable event might be regarded as a contribution of limit structural
analysis to the solution of crucial social problems: the design of the Morison shelters
for protection of British families against V2 bombs during one of the most tragic periods of the Second World War; such design was due to the research team of Sir John
Baker in Cambridge. In those years, early forties, two other circumstances occurred
in the United States with consequences fruitful also for the development of the direct methods: at Brown University, Providence RI, the successful research stream
on structural mechanics centered on plasticity was starting under the guidance of
Prager; almost simultaneously, linear (and later nonlinear) mathematical programming was developed in Stanford, by George Dantzing and his team, initially in terms
of algorithms for optimization of American navy strategies. Later a productive convergence occurred between mathematical programming and elastoplastic structural
mechanics, particularly as for direct methods and their practical applications.
Limit analysis and its generalization to shakedown analysis after the mid of the
last century have been one of the most fruitful developments of structural mechanics,
simultaneously and interactively with another fast growth in applied science, namely
finite element and other methods for computer simulations of physical phenomena,
particularly of inelastic responses of structures to external actions.
Each important scientific research area exhibits a flourishing era, characterized
by intensive activities and fast growth. Such era for direct methods of limit and
shakedown analyses grew with contributions provided by several internationally acknowledged research leaders; obviously they are well known to all research workers
in applied mechanics and, therefore, they are not considered in this brief foreword.
Here I take the liberty to mention with gratitude only some of them, with whom
v
vi
Foreword
I had opportunities of personal meetings and pleasant interactions before our international scientific community lost them: Drucker, Symonds, Koiter, Massonnet,
Sawczuk, Knig, Martin; in my country Capurso, Ceradini, Gavarini.
But direct methods still represent a fertile research field, still attractive and fashionable, with challenges still open to fruitful and synergistic convergence of structural mechanics, mathematical developments, computational techniques and not
only structural engineering, but also diverse technologies. Evidence of such closing remark is provided by this book and by the Athens conference which originated
it, and by the sequence of recent meetings and papers collections on direct methods. Colleagues from various research institutions, who promote such initiatives on
direct methods and successfully contribute to them, do deserve, in my opinion, the
appreciation of the international community of theoretical and applied mechanics.
Technical University (Politecnico), Milan, Italy
Giulio Maier
Preface
Civil and mechanical engineers are to a large extent concerned with safety- and life
assessment of structures under thermo-mechanical loading which may cause severe
inelastic straining. This task is virtually impossible to perform following cumbersome and time consuming evolutionary methods which, additionally, require the
complete knowledge of the loading history. Mostly, however, only variation intervals of the loads are known. Thus, it is important to be able to produce margins
of safe service conditions for structures, as well as for structural material, against
excessive inelastic deformations.
Methods that aim towards this end, avoiding step-by-step analysis, are called
Direct Methods. They are non-evolutionary, and although such methods have existed for some time, they are attracting an increasing interest from scientists and
researchers, based on new mathematical formulations and new developments on
numerical analysis.
The present volume contains the most recent advances on these methods. It is the
outcome of the third international workshop that was held in Athens on February
2012, following the successful first and second workshops in Aachen in 2007 and
Lille in 2009. The event, which attracted more than 30 scientists from 6 countries,
was organized by the National Technical University of Athens and was hosted by
the Onassis Cultural Center.
The papers in the book are arranged in the order of their appearance in the workshop and their contributions are in the fields of Structural and Soil Mechanics as well
as Material Science. All the contributed papers have undergone a rigorous review
process before acceptance for publication.
We would like to thank all the scientists that have participated in this book for
the high quality level of their work.
We would also like to express our thanks to Professor Giulio Maier, of the Technical University of Milan, for his kindness to foreword this book.
Athens, Greece
Aachen, Germany
Konstantinos Spiliopoulos
Dieter Weichert
vii
Contents
23
57
79
Contents
Abstract By using the kinematic approach of limit analysis (LA) for a hollow
sphere whose solid matrix obeys the von Mises criterion, Gurson (J. Eng. Mater.
Technol. 99:215, 1977) derived a macroscopic criterion of ductile porous medium.
The relevance of such criterion has been widely confirmed in several studies and in
particular in Trillat and Pastor (Eur. J. Mech. A, Solids 24:800819, 2005) through
numerical lower and upper bound formulations of LA. In the present paper, these
formulations are extended to the case of a pressure dependent matrix obeying the
parabolic Mises-Schleicher criterion. This extension has been made possible by the
use of a specific component of the conic optimization. We first provide the basics
of LA for this class of materials and of the required conic optimization; then, the
LA hollow sphere model and the resulting static and mixed kinematic codes are
briefly presented. The obtained numerical bounds prove to be very accurate when
compared to available exact solutions in the particular case of isotropic loadings.
A second series of tests is devoted to assess the upper bound and approximate criterion established by Lee and Oung (J. Appl. Mech. 67:288297, 2000), and also
the criterion proposed by Durban et al. (Mech. Res. Commun. 37:636641, 2010).
As a matter of conclusion, these criteria can be considered as admissible only for a
slight tension/compression asymmetry ratio for the matrix; in other words, our results show that the determination of the macroscopic criterion of the porous MisesSchleicher material still remains an open problem.
F. Pastor
Laboratoire de Mcanique de Lille (LML), UMR-CNRS 8107, Villeneuve dAscq, France
e-mail: [email protected]
J. Pastor (B)
Laboratoire LOCIE, UMR-CNRS 5271, Universit de Savoie, Chambry, France
e-mail: [email protected]
D. Kondo
Institut DAlembert, UMR-CNRS 7190, Universit Pierre et Marie Curie, Paris, France
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_1,
Springer Science+Business Media Dordrecht 2014
F. Pastor et al.
1 Introduction
The criteria for yielding of ductile porous materials, proposed by Gurson [8], are
based on limit analysis (LA) of a hollow sphere or cylinder obeying the von Mises
criterion and subjected to uniform strain rate boundary conditions. In his analysis,
Gurson used a LA kinematical approach to obtain an upper bound to the exact solution of the hollow sphere model and to the strength of isotropic porous materials
which is shown later to correspond to microstructures of the type Hashin Composite
Spheres Assemblages (see for instance [12]). In the context of isotropy, the Gurson approach has been extended in several directions among which that allowing to
investigate analytically porous materials with matrix exhibiting pressure sensitive
behavior (see among others, [5, 7, 9, 13, 15]).
On the other hand, using a finite element discretization of a unit cell, both static
and kinematical methods of LA have been developed to obtain rigorous lower and
upper bounds and assess Gursons criteria for cylindrical as well as spherical cavities in [6] and [21]. Using these numerical approaches, the yield function of a porous
material with cylindrical voids was studied first; it was shown that Gurson criterion
is approximate and does not exhibit the corner that the exact criterion displays in the
plane strain case. However, the Gurson criterion for materials containing spherical
cavities appears to be satisfactory [32], although it does not account for possible dissymmetry (with respect to the hydrostatic axis) associated with third stress invariant
effects [31].
It is worth noticing that the above mentioned LA based numerical approaches
lead to rigorous lower and upper bounds of the macroscopic criterion and allow
a posteriori verification of these bounds from the final optimal solution. Consequently, their use to assess existing theoretical yield functions of porous media is
intrinsically relevant. Obviously, such assessment is crucial for confidence in the
formulation of constitutive relations based on these yield functions, in the perspective of structural computations.
This paper is focused on a class of materials for which the pressure-sensitive
parabolic criterion of Mises-Schleicher [27] is used, see for example [14, 30],
and [3] for a detailed comparison with experimental data. Following [4], this criterion well represents the strength differential between uniaxial compression (C )
and uniaxial tension (T ) which has been observed for many polycristalline materials and geomaterials. Moreover the absence of an apex (as in the Drucker-Prager and
Coulomb case) around the minimum value of I1 (= tr ), which agrees with experimental observations [28], constitutes an advantage from a physical and numerical
point of view. This explains that recent papers have investigated the influence of the
porosity for such materials. Therefore, in the present paper, analytical Gurson-like
macroscopic criteria for spherically porous media with Mises-Schleicher matrix will
be assessed by using original LA techniques.
The paper is organized as follows. Section 2 presents the main features of the LA
theory and completes the expressions needed by the LA methods; a brief description
of these methods is given afterwards, ending with a summary of the mixed kinematic approach and of the specific conic optimization which has allowed to solve
the final optimization problem. In Sect. 3, the micro-macro hollow sphere model is
detailed with its 3D finite element discretizations of stress and velocity fields to obtain the macroscopic criterion in terms of average macroscopic stresses. In Sect. 3.3
is presented the implementation of the static approach, which leads to an original
rotated quadratic cone formulation. Section 3.4 briefly presents the also original
mixed kinematic formulation, with discontinuous velocity fields, which leads also
to a rotated cone formulation that bypasses the singularity problem when dm becomes null in an element, or a fortiori when the special case of von Mises criterion
is considered.
Section 4 recalls the criteria given in the above mentioned papers of the literature
that will be assessed in the present work. In Sect. 5, the obtained numerical bounds
in the case of hydrostatic loadings are validated first by comparison to available analytical solution (see for instance [15]). Finally, for general loadings, the available
criteria of Lee and Oung [13] and of Durban et al. [5] are assessed by comparing
them to the numerical bounds; to this end we consider two porosities and two representative tensile and compressive strengths (T , C ) ratios which characterize the
Mises-Schleicher material.
0 =
C T .
(2)
F. Pastor et al.
Similarly, a strain rate tensor field, d, is admissible if it is both kinematically admissible (KA, i.e., it is derived from a piecewise continuous velocity vector field u, with
bounded discontinuities [u], such that the velocity boundary conditions are fulfilled)
and plastically admissible (PA), i.e., the following associated flow rules (3a), (3b)
are satisfied
f
, f ( ) = 0, 0, f ( ) 0;
fdisc
[u] =
, fdisc (T ) = 0, 0, fdisc (T ) 0,
T
d =
(3a)
(3b)
where d is the strain rate tensor, [u] the velocity jump across the discontinuity surfaces, and T = (n , nt ) the corresponding stress vector acting on them. The definitions (3a), (3b) lead to the so-called plastic admissibility condition (PA condition)
that d and [u] must verify to be plastically admissible, i.e. at least one stress tensor
or stress vector T (such that f ( ) = 0 and fdisc (T ) = 0, respectively) can be associated to them. Let us recall that the criterion fdisc (T ) results from the projection
of the plasticity criterion f ( ) on the (n, t) Mohr plane associated to the velocity
discontinuity surface of normal n, which results for the present material:
fdisc (T ) = nt2 + 0 n 02
1 + 2
0.
3
(4)
If (3a) and (3b) are fulfilled, the quantities : d and T [u] become, respectively,
the convex unit dissipated powers vol (d) and disc ([u]), i.e.:
disc [u] = T [u].
vol (d) = : d;
(5)
Finally, denoting dm the average value of the strain rate tensor d, for the MisesSchleicher material the P A conditions resulting from (3a), (3b) read:
dm 0;
[un ] 0,
(6)
2
0 dm 0 deq
+
,
4 dm
0 (1 + 2 )
0 [ut ]2
[un ] +
,
disc [u] =
3
4 [un ]
(7)
(8)
where deq = 23 dd : dd and dd the deviatoric part of d.
Let us point out that relation (4) and its consequences in (6) and (8) are original; they will allow to check a posteriori the admissibility of the optimal velocity
field and the corresponding dissipated power along the discontinuities of the mixed
numerical problem described in Sect. 2.3. It is also worth noting that the use of relations (7) and (8) in a classic kinematic method may lead to singularities during the
optimization process.
Let us assume now that the virtual power of the external loads Pext can be written
as the scalar product of a loading p-vector Q = Q( ) (linear in , with SA), and
a generalized velocity p-vector q = q(u) (linear in u, with u KA) the components
of which are called kinematic parameters. The vectors Q and q are also assumed
to be linear in their respective arguments, such assumption being always verified in
usual cases and in the present problem. The virtual power theorem then reads:
Pext = Q q =
: ddV +
T [u]dS = P ,
(9)
V
Sdisc
where V is the volume of the mechanical system, Sdisc the union of the velocity
discontinuity surfaces; let us note that P becomes the so-called dissipated power
Pdiss when the fields (, T ) and (d, [u]) are also PA and associated, respectively.
Basically, a solution of the LA problem is a pair (, u) where the fields and u
are both admissible and associated by the normality rule. In this case, the loading
vector corresponding to the field is a limit load Qlim of the mechanical system. It
can be proved that the admissible loading vectors Q, i.e., corresponding to an admissible field , belongs to a convex set K whose boundary K is the locus of the limit
loads Qlim . The vector q complies with normality to K, i.e. the maximum dissipation principle is verified in terms of Q and q. However, solving the LA problem by
the determination of the above mentioned pair (, u) is rarely possible; fortunately,
the solution can be bracketed by using the so-called lower and upper bound methods
of LA, more precisely the following static and kinematic approaches.
F. Pastor et al.
min
u q d -admissible
vol d(u) dV +
disc [u] dS = Pdiss (u) , (11)
Sdisc
for various values of q d . Therefore, the classic method needs the analytical expressions of the unit dissipated powers vol (d) and disc ([u]). As it can be seen in
(7), (8), some difficulties may arise when a non-linear optimization algorithm is
used which needs Jacobian and Hessian matrices.
In fact, here also, relation (11) holds when all admissible fields u can be considered, which is rarely possible. In practice, a set of planes exterior or tangent to K
is obtained: the domain defined by the intersection of the corresponding half spaces
constitutes an outer approach of the boundary K, i.e. an upper bound to the exact
macroscopic criterion.
max F = Q qd
Q,,
(12a)
: ddV +
s.t.
V
f ( ) 0,
n [u]dS = Q q(u)
KAu,
(12b)
Sd
f 0,
(12c)
where is the stress tensor inside the 3D finite elements, and the stress tensor
specific to the discontinuity surfaces whose set is denoted Sd .
The previous formulation gives the exact solution if any velocity and stress fields
could be taken into account. This is not always the case when a discretization of
the mechanical system in finite elements is considered, giving in fact only estimates
of the limit loads. Then, to preserve the rigorous kinematic (or upper bound) character of the final result, we will need to modify the numerical implementation of
the virtual power principle (12b) on the basis of convexity properties when taking
into account the contribution of the discontinuities. In this study, the finite element
formulations of both problems (10a), (10b) and (12a)(12c) are solved using the
specific method briefly presented in the following section.
(13)
xC
with A Rmn , c, x Rn , b Rm . One of the specific features of the conic programming codes is to solve the problems where C is defined by a product of the
following cones C n+2 :
n
n+2
n+2
2
C
= xR
xj 2xn+1 xn+2 , xn+1 0, xn+2 0.
(14)
j =1
With the cone (14) the resulting optimization problem is called rotated quadratic
cone programming problem. Among the existing commercial codes, the most efficient is the code MOSEK, developed by E. Andersen [16], which allows to solve very
F. Pastor et al.
large scale optimization problems. Its applications in mechanics began about ten
years ago, but always for solving classic limit analysis problems which involves von
Mises or Drucker-Prager criterion where C is defined by a product of the Lorentz
cones L n+1 :
n
1
2
n+1
n+1
2
= xR
xj
xn+1 .
L
(15)
j =1
(16)
From (16), the p components Qi of the loading vector Q (see Sect. 2.1) can be defined as the six components of the tensor , and the qi as the corresponding components of E. In fact, depending on the results we are searching for, more relevant
definitions of the Q and q vectors can be chosen. Here, the LA loading parameters and their associated generalized velocities, such that (16) remains verified, are
chosen as follows:
Q1 = m =
3
(xx yy );
Q3 =
2
tr()
;
3
q1 = tr(E);
1
q3 = (Exx Eyy );
3
Q2 =
xx + yy
zz ;
2
(17)
Q4 = yz ;
Q5 = zx ;
Q6 = xy ; (18)
2 Exx + Eyy
(19)
Ezz ;
q2 =
3
2
q4 = 2Eyz ;
q5 = 2Ezx ;
q6 = 2Exy .
(20)
(21)
The stress components represent the loading vector Q in terms of LA and the corresponding macroscopic strain rates define the associated generalized velocity vector q of Sect. 2, with p = 6. The locus K corresponds then to the investigated
macroscopic criterion, further denoted g (Q), of the porous material.
To compare with available analytical results, we search for the projection of
g (Q) on the (m , Q2 ) plane by optimizing Q2 for fixed average stresses m , the
g
other stress components defined in (18) being free. Then Q
= 0 = qi for i 4, and
i
g
Q3
(22)
10
F. Pastor et al.
11
2
2
xx + yy
3
2
2
zz + (xx yy )2 + 3 yz
+ zx
+ xy
2
4
0 0 (xx + yy + zz ) .
(23)
xj2 2x6 x7 .
(24)
j =1
Concerning the numerical implementation of the static approach, the local stress
field is chosen as linearly varying with x, y, z in each tetrahedral element, and represented by a 6-component tensor for each vertex of this tetrahedral element. Consequently, this stress field can be discontinuous across any element boundary. Finally,
to reduce the size of the constraint matrix of the numerical problem, a change of
variables (x0 , x1 , . . . , x5 ) is performed, where x0 = tr and x1 to x5 defined
from (24), so that only the definitions of x6 = 0 /2 and x7 = 0 tr( ) are needed
as new constraints (and new auxiliary variables) for each tetrahedron vertex.
To get a statically and plastically admissible microscopic stress field, the following conditions are implemented in the final form of a matrix of linear equality
constraints:
Definition of the macroscopic stress components xx , yy , zz as averages
of the corresponding microscopic ones i.e., three linear conditions.
Definition of Q1 = m , Q2 , Q3 as functions of the previous ones, i.e., three
linear conditions.
In each element, from the linearity of the stress variation, the equilibrium equations, ij,j = 0, generate three linear conditions.
Continuity conditions: the stress vector is continuous across every discontinuity
surface; each discontinuity triangle generates nine linear conditions.
12
F. Pastor et al.
Boundary conditions: on the triangles belonging to the cavity boundary, the components of the stress vector Ti = ij nj are null at each apex of each triangle
(whose normal is n), here also generating nine linear conditions per triangle.
Symmetry conditions: the microscopic tangential stresses are null at the triangle
apices of the coordinate planes, i.e., six linear conditions for each triangle of these
planes.
To enforce the stress field to be plastically admissible, the criterion (23) is imposed
at each apex of the tetrahedron; hence, due to its convexity, the criterion is fulfilled
anywhere in the element. For each tetrahedron the four conic inequalities are directly
handled by MOSEK by indicating the names of the variables xj involved in (24).
Concerning the functional to be optimized, m is given successive desired values
and Q2 is minimized; when m is close to its maximum value, Q2 is fixed and m
is optimized for better convergence of the optimization process. The final result
is a rotated quadratic cone programming problem, which is one of the features of
MOSEK ; the optimal solution gives a rigorous lower bound to the exact solution of
the FEM model; this lower bound character is systematically verified by a postanalysis of the optimal stress field given by the optimizer. It is worth noting that,
in all tests, the best values of eq , for fixed m , are obtained for negative Q2 , i.e.
by minimizing Q2 in fact. This point proves here also a slight influence of the third
stress invariant of the macroscopic stress tensor, as in [31].
(25)
(26)
13
where the vector {un } collects the twelve degrees of freedom of the element. Then
the VPP reads:
{u}T V []{Q} =
Vk {d}Tk { } {u} KA,
(27)
k
where Vk denotes the volume of the tetrahedron k. Using Eq. (26), and after the
assembly of the elements, the relationship (27) gives rise to the following variational
system:
{u}T []{ } + V []{Q} = 0 {u} KA,
(28)
where the matrix [] results from the assembly of the submatrices [] = Vk [B T ]
calculated for each element k in (27).
T
T
T
Aij m [u] i i {n} + [u] j j {n} + [u] m m {n} /3,
(29)
where Aij m is the area of Sij m whose the normal is n. Using these bounds gives rise
to a matrix [ ] in an analogous manner than in the previous subsection.
Finally the resulting numerical form of the mixed problem (12a)(12c) is the
following:
Max {qd }T {Q}
(30a)
s.t.
[]{ } + V []{Q} = 0,
f ( ) 0;
f 0
(30b)
+KA conditions.
(30d)
(30c)
14
F. Pastor et al.
At this stage, it is worth recalling that the optimal velocities components are the values of the dual variables associated to the rows of the final matrix [, ], which
are available in the optimal solution given by the code MOSEK, or the interior point
optimizer of [19]. Note also that there is no restriction to the allocated stress tensors,
except the verification of the criterion; then, for any non-zero velocity field (and the
corresponding strain rates and velocity jumps) an associated PA stress tensor can be
found. Consequently, from Hills maximum work principle, the upper bound character is preserved with the above selected formulation. Here also, the optimal solutions
are post-analyzed by using the Mises-Schleicher results of Sect. 2.1.
15
f
(3m )2 + 30 (1 f )m (1 f )2 02 = 0.
4
(31)
eq
0
3m
+ 2f cosh
20
+ 3(1 f )
m
1 + f 2 = 0.
0
(32)
eq
0
G
m
G
2
2f G sinh + 3
+ 2f 1 + G 1 + cosh
1 + f 2 = 0,
0
(33a)
where
G=
1 3
m
+ 2 1 + 2.
0
(33b)
As already noted by the authors, (32) and (33a), (33b) reduce to the Gurson expression when tends to zero, while (31) does not meet this expected property.
5 Numerical Tests
5.1 Comparison with Exact Results for Isotropic Loadings
Monchiet and Kondo [15] have confirmed that the m solutions of Lee and
Oung [13] (obtained by limit equilibrium considerations) are exact LA solutions by
exhibiting associated admissible stress and velocity fields, following an approach
already used by Thor et al. [31] for Drucker-Prager and Coulomb matrices. These
exact solutions are considered here for a first validation of the numerical bounds that
we derived by means of the above mentioned static and (mixed) kinematic codes.
Tables 1 and 2 give the corresponding values for isotropic tensile and compressive
loadings, respectively; six decimal are reported.
The analytical values were obtained by solving with MATLAB the following equations given in [13, 15] for m 0:
1
+ 1 + 2 3m /0
1
+
ln
1 + 2 3m /0
1 + 2 = ln f,
+ 1 + 2 ]
(34)
16
F. Pastor et al.
Table 1 Comparison of the present bounds to exact solution m /0 for isotropic tensile loadings
0.2
0.4
0.6
0.8
1.0
3D-kine
1.537747
0.985438
0.651860
0.468268
0.361183
0.292810
[15]
1.535057
0.984299
0.651539
0.468153
0.361131
0.292782
3D-stat
1.526208
0.981093
0.650016
0.467134
0.360237
0.292016
Table 2 Comparison of the present bounds to exact solution m /0 for isotropic compressive
loadings
0.2
0.4
0.6
0.8
1.0
3D-kine
1.537747
2.291437
3.182374
4.159597
5.190714
6.256271
[15]
1.535057
2.285464
3.172361
4.144888
5.170962
6.231292
3D-stat
1.525985
2.269164
3.142363
4.101975
5.113655
6.109577
and for m 0:
1 + 2 3m /0
1
1
1 + 2 3m /0 +
1 + 2 = ln f.
+ 1 + 2 ]
(35)
The numerical bounds are mainly obtained with a 14 14 14 prism mesh
resulting in 990, 587 linear constraints, 264, 796 conic constraints, and 2, 127, 191
variables for the mixed kinematic code; the problem is solved in about 700 seconds
of CPU time (with one thread) on an Apple iMac-i7 with 16 gigabytes of RAM. The
static problem with that mesh (1, 117, 600 linear and 153, 664 conic constraints with
1, 229, 319 variables) needs about 1100 seconds of CPU time.
From these tests, it can be firstly noted that the exact solution is always located
between the numerical bounds, even when these bounds are very close to each other.
Secondly, due to the remarkable efficiency of MOSEK, the solving CPU times are
very low, given the large scale of the generated optimization problems. Finally, regarding the high performance in terms of bounds, and these first validations of the
proposed numerical procedure, it should be noted that the presence of discontinuities everywhere plays a crucial role in both static and kinematic results.
In the following tests, owing to the very small difference between positive and
negative optimal values of Q2 , for the same fixed Q1 = m , we will present for
conciseness only the results for the negative Q2 , which are here a bit larger than the
positive ones in absolute value. As it will be seen, this has no consequence about
assessing the available criteria.
ln
17
Fig. 4 Present bounds versus [13] and [5] results for C /T = 1.1, f = 0.1 (0 = 1)
18
F. Pastor et al.
Fig. 5 Present bounds vs. [13] and [5] results for C /T = 1.1, f = 0.01 (0 = 1)
by [15] on the poor performance of the Lee and Oung criterion when compressive
hydrostatic loadings are considered.
For f = 0.01, from Fig. 5 the same remarks can be made, but the differences
between both criteria are amplified as well as the violation of the static bound for the
Lee and Oung criterion. At this stage, we could conclude that the Durban criterion
can be selected.
5.2.2 Case C /T = 2
The results for c /T = 2 (strong asymmetry) are plotted in Figs. 6 and 7; it can be
seen that the numerical static and kinematic bounds still remain close with a better
performance for f = 0.1, more or less expected since the mesh is the same for both
porosity cases. Curiously, the results present an inversion between the Lee and Oung
upper bound and the Durban et al. criterion. This proves that, for high asymmetry
ratio, the Lee and Oung criterion, as well as that proposed by Durban (even if this
one does not violate the static bounds) cannot be considered relevant.
19
Fig. 6 Present bounds vs. [13] and [5] results for C /T = 2, f = 0.1 (0 = 1)
6 Conclusion
In the present paper, we present original finite element formulations of the static and
kinematic methods of Limit Analysis of a hollow sphere with a Mises-Schleicher
matrix, both in the three-dimensional case. The resulting optimization problems are
cast into the so-called rotated quadratic cone optimization, which are solved using the excellent commercial code MOSEK. The static and kinematic mixed codes
appear very efficient in all tests, both in terms of computational speed and of accuracy of the solutions which were systematically verified by post-analysis. First, for
isotropic loadings, the corresponding closed form solutions of Lee and Oung, confirmed as exact LA solutions by [15], allow to validate the present numerical bounds
in this particular case. Then, the comparison with the upper bound and the criterion
of Lee and Oung [13] confirms that both expressions are relevant for tensile and low
compressive loadings, but not in highly compressive loading cases. Moreover, the
Lee and Oung criterion noticeably violates the numerical static bounds when the
tensile/compressive strength asymmetry is not small and for compressive loadings.
This is not the case with the Durban [5] closed-form expression, but this criterion
20
F. Pastor et al.
Fig. 7 Present bounds vs. [13] and [5] results for C /T = 2, f = 0.01 (0 = 1)
strongly overestimates the real solution for compressive loadings when the above
mentioned asymmetry is high. Finally, it can be concluded that the question of the
determination of a satisfactory analytical criterion for a ductile porous medium with
a Mises-Schleicher matrix still remains open.
References
1. Abdi R, Buhan PD, Pastor J (1994) Calculation of the critical height of a homogenized reinforced soil wall: a numerical approach. Int J Numer Anal Methods Geomech 18:485505
2. Anderheggen E, Knopfel H (1972) Finite element limit analysis using linear programming.
Int J Solids Struct 8:14131431
3. Aubertin M, Li L (1974) Yield locus studies of oriented polycarbonate: an anisotropic and
pressure-dependent solid. Int J Mech Sci 16:789799
4. Aubertin M, Li L (2004) A porosity-dependent inelastic criterion for engineering materials.
Int J Plast 20:21792208
5. Durban D, Cohen T, Hollander Y (2010) Plastic response of porous solids with pressure sensitive matrix. Mech Res Commun 37:636641
21
6. Francescato P, Pastor J, Riveill-Reydet B (2004) Ductile failure of cylindrically porous materials, part I: plane stress problem and experimental results. Eur J Mech A, Solids 23:181190
7. Guo TF, Faleskog J, Shih CF (2008) Continuum modeling of a porous solid with pressure
sensitive dilatant matrix. J Mech Phys Solids 56:21882212
8. Gurson AL (1977) Continuum theory of ductile rupture by void nucleation and growth, part I:
yield criteria and flow rules for porous ductile media. J Eng Mater Technol 99:215
9. Jeong HY (2002) A new yield function and a hydrostatic stress-controlled model for porous
solids with pressure-sensitive matrices. Int J Solids Struct 39:13851403
10. Kovrizhnykh AM (2004) Plane stress equations for the von Mises-Schleicher yield criterion.
J Appl Mech Tech Phys 45:894901
11. Krabbenhoft K, Lyamin A, Hijaj M, Sloan S (2005) A new discontinuous upper bound limit
analysis formulation. Int J Numer Methods Eng 63:10691088
12. Leblond JB, Perrin G, Suquet P (1994) Exact results and approximate models for porous
viscoplastic solids. Int J Plast 10:213235
13. Lee JH, Oung J (2000) Yield functions and flow rules for porous pressure-dependent strainhardening polymeric materials. J Appl Mech 67:288297
14. Lubliner J (1990) Plasticity theory. McMillan, New York
15. Monchiet V, Kondo D (2012) Exact solution of a plastic hollow sphere with a MisesSchleicher matrix. Int J Eng Sci 51:168178
16. MOSEK ApS (2002) C/O Symbion Science Park, Fruebjergvej 3, Box 16, 2100 Copenhagen , Denmark
17. Pastor J (1978) Analyse limite : dtermination numrique de solutions statiques compltes.
Application au talus vertical. J Mc Appl 2:167196
18. Pastor F (2007) Rsolution par des mthodes de point intrieur de problmes de programmation convexe poss par lanalyse limite. Thse de doctorat, Facults universitaires Notre-Dame
de la Paix, Namur
19. Pastor F, Loute E (2005) Solving limit analysis problems: an interior-point method. Commun
Numer Methods Eng 21(11):631642
20. Pastor J, Turgeman S (1976) Mise en uvre numrique des mthodes de lanalyse limite
pour les matriaux de von Mises et de Coulomb standards en dformation plane. Mech Res
Commun 3:469474
21. Pastor J, Francescato P, Trillat M, Loute E, Rousselier G (2004) Ductile failure of cylindrically
porous materials, part II: other cases of symmetry. Eur J Mech A, Solids 23:191201
22. Pastor F, Loute E, Pastor J (2009) Limit analysis and convex programming: a decomposition
approach of the kinematical mixed method. Int J Numer Methods Eng 78:254274
23. Pastor F, Loute E, Pastor J, Trillat M (2009) Mixed method and convex optimization for limit
analysis of homogeneous Gurson materials: a kinematical approach. Eur J Mech A, Solids
28:2535
24. Pastor F, Pastor J, Kondo D (2012) Limit analysis of hollow sphere or spheroid with Hill
orthotropic matrix. C R, Mc 340:120129
25. Salenon J (1974) Thorie de la plasticit pour les applications la mcanique des sols. Eyrolles, Paris
26. Salenon J (1983) Calcul la rupture et analyse limite. Presses des Ponts et Chausses, Paris
27. Schleicher F (1926) Der Spannungszustand an der Fliegrenze (Plastizittsbedingung).
Z Angew Math Mech 6:199216
28. Stassi-dAlia F (1961) Une fonction quadratique des tensions principales comme conditions
de plasticit des corps solides. Bull RILEM 13:4445
29. Tal AB, Nemirovsky A (2001) Lectures on modern convex optimization. SIAM, Philadelphia
30. Theocaris PS (1995) Failure criteria for isotropic bodies revisited. Eng Fract Mech 51:239
264
22
F. Pastor et al.
31. Thor P, Pastor F, Pastor J (2011) Hollow sphere models, conic programming and third stress
invariant. Eur J Mech A, Solids 30:6371
32. Trillat M, Pastor J (2005) Limit analysis and Gursons model. Eur J Mech A, Solids 24:800
819
33. Zhang H, Ramesh KT, Chin E (2008) A multiaxial constitutive model for metal matrix composites. J Mech Phys Solids 56:29722983
Abstract The present contribution summarizes the results of recent studies carried
on by the authors in the last few years concerning the evaluation of the load bearing
capacity of single- and multi-pin joints in composite orthotropic plates. The problem, tackled via limit analysis, employs a Tsai-Wu-type yield surface and a non
standard treatment of limit analysis approach. Upper and lower bounds to the real
peak load value are evaluated by two FE based numerical procedures predicting
also the joint failure mode. The whole procedure is implemented at lamina level so
taking into account some of the through thickness effects on the joint strength capabilities. A wide number of experimental findings, coming from laboratory tests on
real prototypes and available in the relevant literature, is considered to validate, by
comparison, the expounded methodology.
23
24
settled, limit analysis can be used as a preliminary/first design tool: once a good prototype is individuated for a specific problem, sophisticated post elastic FE analysis
or experimental laboratory tests can be performed for a deeper and exhaustive understanding of its mechanical behaviour with a considerable saving of efforts either
economic or time consuming.
The classical approaches of limit analysis rely upon mathematical programming
procedures which have progressed significantly in recent years (see e.g. [15, 16, 19,
20, 35, 36, 40]). A number of contributions in this field (see e.g. [41] and references
therein) adopt however a different approach whose basic assumption is that limit
state solutions may be developed from sequences of elastic (linear) analyses easy
to handle via any commercial finite element code. In this context a numerical FEbased approach has been recently proposed by the authors to a peculiar problem of
orthotropic composite laminates, namely the evaluation of the load bearing capacity
of pinned-joint composite plates with single- or multi-pin fastenings [2427]. This is
a topical problem in the composite realm as witnessed by the number of recent paper
on the subject, more than 700 papers are listed in the work of Mackerle [21] which
provides an interesting review of finite elements methods applied for the analysis
of joints; remarkable reviews are also the ones given in [6, 10, 31, 37]. For design
purposes, Researchers are mainly interested in determining the effects on joints failure strength and failure modes of some peculiar geometrical parameters such as: the
ratios between hole-distance from the free plate edge to hole-diameter, or the one
between the plate-width to hole-diameter or, also, the relative distances between fastener holes as well as their spatial distribution within the plate [24, 9, 1214, 22].
The results of a research carried on in the last few years and concerning the analysis of mechanically fastened joints in multi layers composite plates are expounded
in the following. Details, and deeper explanations are given in the papers [2427] to
which the Reader could refer for more information. In particular, two well known
FE procedures have been rephrased: (i) the Linear Matching Method, conceived by
Ponter and Carter [28], (see also [8]), has been used to compute an upper bound to
the peak load multiplier; (ii) the Elastic Compensation Method, due to Mackenzie
and Boyle [18], has been employed to evaluate a lower bound to the peak load multiplier. The former, considering a structure made by a fictitious linear viscous material
with elastic parameters spatially varying, allows to construct a collapse mechanism
and eventually to evaluate an upper bound to the peak load. The latter, grounding on
a stress redistribution procedure pursued by a sequence of elastic analyses in which
highly loaded regions are systematically weakened, produces an admissible stress
field suitable for a lower bound evaluation.
The assumption of a Tsai-Wu-type yield surface for composite laminates [7, 34,
38, 39] allows one to locate stress states at which the material has exhausted its
strength capabilities. The further assumption of a non-associate flow rule makes
the whole methodology of general applicability following a nonstandard approach
[17, 30]. Limit analysis is used indeed to evaluate the strength capability of the joint
by evaluating upper and lower bounds bracketing the real peak load value.
The main novelty of the authors recent studies hereafter summarized, is related
either to the extension of the mentioned FE procedures in the realm of anisotropic
25
composites and, in general, non associative materials or, also, to their non trivial
implementation at layer laminate level. The latter peculiarity reveals undoubtedly
more burdensome computations but, as witnessed by the obtained results, allows
to take into account the stacking sequence of the laminate. Some of the throughthickness effects depending on the number of the laminae the laminate is made with
as well as the effects of fiber orientations within each lamina can indeed be handled
with good accuracy. To this aim higher order shell-type multilayered finite elements
have been used, performing all the relevant operations at the Gauss points of each
element layer (lamina).
In order to verify the reliability of the proposed procedure, checking its ability
of bracketing the real (experimentally detected) peak load, a considerable number
of experimental tests at rupture, available in the literature [2, 1214, 22], have been
numerically reproduced. The possibility of localizing the collapse zone so predicting
the related failure mode has been also investigated highlighting the potentialities of
the layer by layer implementation. In conclusion, the proposed procedure seems to
guarantee, at least for the studied problem, a great accuracy for different laminate
lay-ups and different joint geometries establishing an effective design tool useful to
avoid expensive trials on real prototypes.
where: jc = f/
j are the components of the outward normal to the yield surface
f (j ) = 0 (with > 0 scalar multiplier); PU B denotes the upper bound load multiplier (for simplicity only surface forces, p i , acting on the external portion of the
y
body Vt , are considered); j are the stresses at yield associated to the compatible
strain rates jc ; u ci are the related displacement rates. The set (jc , u ci ) indeed defines
a collapse mechanism.
On the other hand, the static theorem of limit analysis states that if at every point
within V exists a stress field j which satisfies the condition f (j ) 0 and in
equilibrium with the applied load P p for a certain value of P , say PLB , then PLB
is a lower bound to the limit load multiplier.
Two remarks appear necessary to focus what will be expounded in the next sections.
26
Remark 1 For nonstandard materials, see e.g.: [17, 30, 32, 33], the Radenkovics
first and second theorems state that (after [17]): every value of the limit load for
a body made of a nonstandard material is located between two fixed boundaries
defined by the values of the limit loads for two corresponding standard materials.
The hypothesis of a nonstandard constitutive behavior assumed next will then imply
to search for an upper and a lower bound to the (collapse) peak load of the joint.
Remark 2 The two numerical procedures, expounded in Sects. 3 and 4, play two
distinct roles as specified in the following. The Linear Matching Method (LMM) is
naturally related to the kinematic approach of limit analysis being able to construct
a compatible collapse mechanism. It then gives an upper bound, PU B , to the peak
load multiplier. The Elastic Compensation Method (ECM) is indeed strictly related
to the concept of stress redistribution aimed at producing an admissible stress field.
It is then related to the static approach of limit analysis furnishing a lower bound,
PLB , to the peak load multiplier.
(2)
where:
1
1
1
1
1
+
;
F2 :=
+ ;
F11 :=
;
Xt
Xc
Yt
Yc
Xt Xc
(3)
1
1
1
F22 :=
;
F66 := 2 ;
F12 :=
F11 F22 ;
Yt Yc
2
S
with Xt , Xc the longitudinal tensile and compressive strengths respectively; Yt , Yc
the transverse tensile and compressive strengths respectively and S the longitudinal shear strength; moreover, as usual for composite structures, 1 and 2 denote the
principal directions of orthotropy in plane stress case while 6 12 . In Eqs. (3)
the compressive strengths Xc and Yc have to be considered intrinsically negative.
The quadratic form given by Eq. (2) defines an admissible stress states domain:
points within the domain locate stress states pertaining to an orthotropic linear elastic behaviour of the material; points lying on the domain boundary locate stress
states at which the material has exhausted its strength capabilities. The Tsai-Wutype surface, in the quadratic form adopted allows one to apply the standard rules of
transformation, invariance and symmetry, and locates an ellipsoid in the stress space
which is assumed as yield surface for the orthotropic material considered. Moreover,
the orthotropy of the composite laminates infers the general assumption of non associativity so, as noted above, the peak load values of the analyzed specimens will
be located by the determination of upper and lower bounds.
F1 :=
27
Remark 3 On taking into account the previous Remark 1 and the strict convexity
of the Tsai-Wu-type yield surface the postulated non associativity can be treated
following the Radenkovic approach and, in this case, the assumed yield surface can
itself play the double role of inner and outer surface.
28
()
()
()
W [Ej , 12 , j ] = const, where the starred quantities define indeed the modified
W = const at k + 1. The corresponding stresses, Y , are then stresses at yield,
the related displacement rates the incipient displacements at collapse. This operation, which also explains the name of the method, the surface W matches the
yield surface at a point of given normal, is performed at each GP of the FE mesh
and obviously violates the global equilibrium conditions. The stresses at yield, evaluated by matching, do not satisfy equilibrium with the loads acting at iteration k;
this essential condition is indeed assured at the end of an iterative process. To this
concern, a sufficient condition for convergence is given in [29] and is not reported
for brevity.
Moreover, grounding on the formal analogy between the linear viscous problem
and the linear elastic problem (see e.g. [23]), the fictitious linear viscous solution
(in rate form) can, in practice, be computed as a fictitious linear elastic solution
(in finite form), W playing the role of complementary energy potential of the fictitious material. The FE analysis is then achievable as a linear elastic analysis by any
commercial FE code.
29
a sequence of linear elastic FE analyses is carried out in which, for a given load
value, the elastic moduli are reduced within critical regions of the structure identified by the elements where the stresses attain values greater than the yield one.
This allows one to define a maximum admissible stress value in the whole structure
and for the given load. Increased values of loads are then considered in the subsequent sequences of analyses till further load increase do not allow the maximum
stresses to be brought below yield by the reduction (or redistribution) procedure. A
PLB load multiplier can be easily evaluated at last admissible stress values attained
for a maximum acting load in the spirit of the static approach of limit analysis.
30
Among all the e#e in the mesh, the maximum stress, i.e. the stress point farthest
away from the Tsai-Wu type surface, say R (coincident with e#3 in Fig. 2), is
detected, YR being the corresponding stress at yield (measured on the direction
R /| YR |). The iterations (analyses) are carried out inside the given sequence until
this maximum stress in the whole mesh R just reaches (is below) its yield value
YR , and this by the above reduction or redistribution procedure. Further sequences
of elastic analyses are then carried on, each one with an increased value of PD ,
and the reduction procedure repeated till further load increases do not allow the R
stress to be brought below yield. A lower bound to the collapse load multiplier can
then be computed as:
PD
.
PL := YR
| R |
(5)
Concerning the evaluation of a lower bound via the ECM, it is worth to mention
the remarkable works of Staat and Co-workers (see e.g. [35, 36]), where the effectiveness of the ECM is analyzed through a comparative study with respect to other
optimization procedures. In [40] a better performance of the so called primal-dual
procedure with respect to the ECM is also shown. The results obtained in the above
quoted papers, all concerning von Mises type materials, show, at least for the cases
there addressed, that the ECM is too conservative. Such assertion deserves surely
further investigations also in the present context even if, for the mechanical problem herein tackled and the material utilized, at least for the analyzed cases, such
drawbacks seem do not appear.
5 Layer-by-Layer Formulation
5.1 FE Modeling and Assumptions
From a numerical point of view, both iterative procedures are driven by a Fortran
main program which utilizes the results of elastic analyses of the examined structural element carried out using the commercial FE code ADINA [1]. In the discrete
model, the number of FEs is chosen taking into account the geometry of the modeled test and with the aim to obtain an enough accurate elastic FE solution. To this
end, as general rule, a finer mesh is always employed around the fastener holes
and a preliminary sensitive study on the elastic solution is performed for each different considered geometry. Moreover, when the analyzed specimens are made by
quasi-isotropic laminates, isoparametric shell elements with 16 nodes and 16 GPs
per element are utilized in the mesh, while, when dealing with laminates with a
general stacking sequence, higher order isoparametric multilayered shell elements
with 16 nodes per element, referred as DISP 16 in the quoted code, are employed.
In particular, the employed multilayered shell elements are an extension to the case
of composite laminates of mixed interpolated tensorial components (MITC) fam-
31
ily of plates and shell elements (see e.g. [11]). The modelling of the sublaminate
is so entrusted to the elements layers endowed with 16 (on the top) plus 16 (on
the bottom) Gauss points. For each element layer a proper material axes system
is fixed and an order 2 Gauss integration is performed in the layer thickness. The
number of layers per element is, obviously, chosen equal to the numbers of the laminae the analyzed laminate is made with. In Fig. 3(a) the modeling assumptions,
namely the transition from the real stacking sequence of the laminate to the layers
element with the proper definition of the layer material axes system, are sketched.
For both procedures, as shown in Fig. 3(b), the updating and/or reduction of the
elastic moduli is carried into effect at each GP of the element (or of the layer for the
multilayer shells) but, on taking into account that a unique set of Ej i.e. a unique
(orthotropic) materialhas to be assigned to each single element layer, the modified moduli are averaged within the single element layer at the beginning of each
FE analysis.
Finally, to simulate the presence of the rigid pins, two different schemes have
been adopted. The first, used for single-pin fastenings, assumes a cosine load normal
distribution to approximate the pressure exerted by the pin on the inner hole surface.
The second, used for multi-pin fastenings, assumes boundary conditions of contact
type between the pins and the fastener holes. In particular, the constrain function
method (see e.g. [5] and references therein), available in the ADINA code, is used.
The method requires the definition of a target line (the pin contour) and the definition
of a contactor line (the fastener hole contour in the laminate) such that no material
overlap between them can occur during the deformation process, while the contactor
line can scroll or leave the target line.
32
Fig. 3 Modeling assumptions: (a) definition of material axes at element layer following the stacking sequence; (b) sketch of the LMM and the ECM carried into effect at element layer
(k1)
(0)
(0)
33
(0)
(k1)
6
c(k1)
step # 5: set j
multiplier
e(k1)
= j
c(k1)
, u i
(k)
PU B
e(k1)
= (k1) 6
Y (k1) c(k1)
j
dV
c(k1)
i
d(V )
Vt p i u
=
e(k1)
= u i
YES EXIT
NOT CONTINUE
, j = 1, 2, 6
Ej = Ej
(k)
34
E1 ;
(0)
E2 ;
(0)
E6 ;
12 ;
#e
ENDIF
step # 4: detect the maximum stress R in the whole mesh (i.e. whose corresponding stress point is the one farthest away from the Tsai-Wu type surface) and
evaluate the pertinent stress at yield YR
End layers loop
End elements loop
step # 5: check if compensated stress R just reaches (is below) the yield value
Y
YES GOTO step # 8
| R | > R
NOT CONTINUE
step # 6: compute a lower bound as:
PD
PL := YR
| R |
step # 7: set PD > PL , set also k = 1 and GOTO step # 1 to perform a new
sequence of elastic analyses
35
step # 8:
IF
| R |(k) | R |(k1)
set PL = PD of the previous sequence and EXIT
ELSE
set k = k 1 and GOTO step #1 to start a new elastic analysis
ENDIF
End iterative procedure
36
Fig. 5 Schematic representation of the multi-pin joints composite plate: (a) fixture test for load
bearing capacity evaluation; (b) mechanical model
Configurations second, third and fourth, concern multi-pin joints plates. In these
cases, the plate is subjected to a tensile load distribution (equivalent to a global
load Q) acting on the plate mid-plane and applied to one edge. Also in these cases
a reference load of global value equal to Q = 1 kN has been assumed (the P U B and
P LB multiplier are numerically coincident with the predicted peak load value for
all the run examples). The pins, assumed rigid, react to the applied load inducing
high level of stresses around the fastener holes. A schematic representation of the
37
Table 1 Layer stacking sequence (LSS) and geometrical properties of the single-hole specimens
Specimen type
Material
LSS
t (mm)
D (mm)
glass-fiber epoxy
[0/45]S
4.4
glass-fiber epoxy
[90/45]S
4.4
glass-fiber epoxy
[00 /90/0]
3.3
glass-fiber epoxy
[900 /0/90]S
3.3
glass-fiber epoxy
[45]2S
4.8
carbon epoxy
[90/45 /0]S
2.64
6.35
Table 2 Mechanical
parameters of the
unidirectional laminae
forming the composite
laminates with a single-pin
joint
Glass-fiber epoxya
Elastic moduli (GPa)
and Poisson ratio
E1
E2
G12
v12
44
10.5
3.74
0.36
Strength (MPa)
Xt
Xc
Yt
Yc
800
350
50
125
120
E1
E2
G12
v12
88
8.8
2.8
0.28
Strength (MPa)
Xt
Xc
Yt
Yc
811
457.7
47.3
109.5
132
Carbon epoxyb
a After
b After
multi-pin joints fixture test together with the corresponding mechanical model is
given in Fig. 5. The latter shows the distributed edge load as well as the shaded rigid
pins fixed to the external world. As said in Sect. 5.1 boundary conditions of contact
type are assumed between the pins and the fastener holes. With reference to Fig. 5:
W and t are the width and thickness plate respectively; the length G indicates the
distance between two parallel holes; K their distance from the longer edge; M the
distance between two serial holes in the y direction; F the distance between two
generic holes in the y direction. Finally, L and E are the distances between the
inner and outer holes from the plate edges respectively.
In detail, the second configuration refers to a [0 /90 /45 ]S laminate with a
three pins joint tested in the technical report of Karakuzu et al. [14]. Referring again
to Fig. 5, the three holes are located in the positions 1, 2, 3. Each glass-epoxy lamina, forming the composite laminate, is characterized by the mechanical properties
reported in Table 3, while the geometrical parameters are: L = 90 mm, t = 1.7 mm,
D = 5 mm; finally, the distance K is equal to 10 mm. For this specimen type, 45
different geometries were tested with values of the ratio E/D ranging from 1 to 5,
ratio G/D ranging from 3 to 5 and considering also three different ratios for F /D,
namely: 2, 4, 6. A number of elements ranging from 544 to 662 has been used for
this configuration together with the proposed layered approach.
The third configuration refers to a woven glass/vinylester laminate with two serial
pins joint tested in the paper of Karakuzu et al. [12]. The two holes are located in the
38
Table 3 Mechanical
parameters utilized for the
numerical simulations of
composite laminates with
multi-pins joints
Glass/epoxy laminaa
Elastic moduli (GPa)
and Poisson ratio
E1
E2
G12
38
0.27
Strength (MPa)
Xt
Xc
Yt
Yc
687
223
74
109
78
v12
Glass/vinylester laminateb
a After
b After
E1
E2
G12
v12
20.77
20.77
4.13
0.09
Strength (MPa)
Xt
Xc
Yt
Yc
395
260
395
260
75
positions 3 and 4 of Fig. 5. The mechanical properties of the laminate are reported in
Table 3, while its geometrical parameters are: L = 85 mm, t = 2.8 mm, D = 5 mm.
For this specimen type, 40 different geometries were tested considering the ratios
W/D = 2, 4 and M/D and E/D variable from 2 to 5 and from 1 to 5, respectively.
In this case the FE models utilized had a number of 16 nodes single layer shell
elements varying from 275 to 530.
Finally, the fourth configuration refers again to a woven glass/vinylester laminate but with two parallel pins joint (see e.g. Karakuzu et al. [13]). With reference
to Fig. 5 the two holes are located in the positions 1 and 2. In this case the geometrical and mechanical parameters of the laminate are the same of configuration three,
while the 45 different geometries considered are obtained by varying the ratio K/D
from 2 to 4, the ratio E/D from 1 to 5, and considering the ratios M/D = 2, 4, 5.
For this specimen type a FE modelling involving 16 nodes single layer shell elements has been again employed with a number of elements variable from 448 to
725.
39
Fig. 6 Peak load values for fixed ratio E/D = 4; comparison between experimental data, dashed
lines with asterisk marker, and values predicted by LMM and ECM respectively. (a) Specimen
type A; (b) Specimen type B; (c) Specimen type C; (d) Specimen type D; (e) Specimen type E;
(f) Specimen type F
where the evaluated upper bounds appear far, and even below, from the experimental values. Analogously, the lower bounds evaluated with the layer-by-layer analyses
are closer (from below) to the experimental values than the ones given by the single
layer treatment, too much conservative. As expected, good accuracy is attained by
40
Fig. 7 Peak load values for fixed ratio W/D = 4; comparison between experimental data, dashed
lines with asterisk marker, and values predicted by LMM and ECM respectively. (a) Specimen
type A; (b) Specimen type B; (c) Specimen type C; (d) Specimen type D; (e) Specimen type E;
(f) Specimen type F
both single and layer-by-layer procedures when dealing with laminate lay-ups of
the type [45 ]ns , as the one of specimen E of Figs. 6(e) and 7(e), for which the
anisotropy of each layer is mitigated by the sequence itself. The average relative
errors, computed as the absolute value of the difference between the experimental
41
Specimen type
LMM
LMM
ECM
ECM
0.29
0.34
0.17
0.36
0.29
0.19
0.17
0.23
0.20
0.23
0.12
0.28
0.07
0.29
0.08
0.32
0.33
0.92
0.20
0.39
0.15
0.49
0.06
0.34
Fig. 8 Peak load values of the multi-layer specimens [0 /90 /45 ]S for F /D = 2.
(a) G/D = 3; (b) G/D = 4; (c) G/D = 5; (d) Plot of the relative error given by the two procedures
and the numerical detected peak load values over the experimental ones, are shown
in Table 4 for the layered analysis and for the single layer one. As it appears the
range of the computed load values bracketing the real one is, by far, more accurate
for the layered approach.
With reference to the second configuration (a three holes joint), Figs. 8, 9 and
10 report the comparison between the experimental findings and the numerical predictions in terms of peak load for fixed values of F /D and different values of E/D
and G/D. In addition, Figs. 8(d), 9(d) and 10(d) give the relative errors of the numerically predicted P U B and P LB with respect to the corresponding experimentally
detected values.
42
Fig. 9 Peak load values of the multi-layer specimens [0 /90 /45 ]S for F /D = 4.
(a) G/D = 3; (b) G/D = 4; (c) G/D = 5; (d) Plot of the relative error given by the two procedures
Fig. 10 Peak load values of the multi-layer specimens [0 /90 /45 ]S for F /D = 6.
(a) G/D = 3; (b) G/D = 4; (c) G/D = 5; (d) Plot of the relative error given by the two procedures
43
Fig. 11 Peak load values of the specimens with two serial fastener holes for W/D = 2.
(a) M/D = 2; (b) M/D = 3; (c) M/D = 4; (d) M/D = 5; (e) Plot of the relative error given
by the two procedures
For the third configuration (joint with two serial pins), the experimental and predicted peak load values are reported in Figs. 11, 12 for fixed values of W/D and
different values of E/D and M/D.
44
Fig. 12 Peak load values of the specimens with two serial fastener holes for W/D = 4.
(a) M/D = 2; (b) M/D = 3; (c) M/D = 4; (d) M/D = 5; (e) Plot of the relative error given
by the two procedures
Finally Figs. 13, 14 and 15 report the experimental and predicted peak load values for the two parallel pins joint (fourth configuration), for fixed values of M/D
and different values of E/D and K/D.
By inspection of the obtained results, in 23 over the 177 examined cases the experimental peak value lies outside the range numerically predicted, such cases are
45
Fig. 13 Peak load values of the specimens with two parallel fastener holes for M/D = 2.
(a) K/D = 2; (b) K/D = 3; (c) K/D = 4; (d) Plot of the relative error given by the two procedures
Fig. 14 Peak load values of the specimens with two parallel fastener holes for M/D = 4.
(a) K/D = 2; (b) K/D = 3; (c) K/D = 4; (d) Plot of the relative error given by the two procedures
46
Fig. 15 Peak load values of the specimens with two parallel fastener holes for M/D = 5.
(a) K/D = 2; (b) K/D = 3; (c) K/D = 4; (d) Plot of the relative error given by the two procedures
47
Fig. 16 Values of the upper (PU B ) and lower (PLB ) bounds to the collapse load multiplier versus
iteration number. LMM prediction, solid lines with rectangular markers; ECM prediction, solid
lines with triangular markers; collapse experimental threshold, finer dashed lines. (a) and (b): multilayer specimens E3 and A3 with single fastener hole; (c) and (d): multilayer specimens with
three fastener holes; (e) and (f): woven specimens with two serial fastener holes; (g) and (h): woven specimens with two parallel fastener holes
48
a After
LSS
E/D
W/D
LMM
expa
[90/45]S
[90/0/90]S
[0/90/0]S
[0/90/0]S
[45]2S
E/D
G/D = 3
LMM
G/D = 4
expa
LMM
G/D = 5
expa
LMM
expa
F /D = 2
1
S/S/B
S/S/B
S/S/B
S/S/B
S/S/B
S/S/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
F /D = 4
1
S/S/B
S/S/B
S/S/B
S/S/B
S/S/B
S/S/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
F /D = 6
a After
S/S/B
S/S/B
S/S/B
S/S/B
S/S/B
S/S/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
B/B/B
49
Table 7 Failure modes: experimental data against predicted ones for the woven specimens with
two serial fastener holes
M/D = 2
E/D
M/D = 3
M/D = 4
M/D = 5
LMM
expa
LMM
expa
LMM
expa
N/N
N/N
N/S
N/N
N/S
N/N
N/
N/N
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
N/
LMM
expa
W/D = 2
W/D = 4
1
B/SN
B/S
B/S
B/S
B/S
B/S
B/S
B/S
BS/BS
B/B
B/BS
B/B
B/BS
B/B
B/BS
B/B
B/B
B/B
B/BN
B/B
B/B
B/B
B/B
B/B
B/BN
B/B
B/BN
B/B
B/B
B/B
B/B
B/B
B/BN
B/B
B/BN
B/B
B/B
B/B
B/B
B/B
a After
E/D
K/D = 2
K/D = 3
K/D = 4
LMM
expa
BN
BS
BS
BS
BN
BS
BS
BS
BS
BNS
BN
BS
BS
BS
BS
BN
BN
BS
LMM
expa
LMM
expa
M/D = 2
M/D = 4
M/D = 5
a After
50
Fig. 17 Pin-loaded plate of Fig. 4: (a) predicted collapse mechanism of bearing type obtained for
specimen #E4 with W/D = 4; (b) predicted collapse mechanism of net-tension type for specimen
#E4 with W/D = 3 against experimental one (after Okutan Baba [22])
Fig. 18 Pin-loaded plate of Fig. 4: (a) predicted collapse mechanism of bearing type against experimental one for specimen #D7; (b) predicted collapse mechanism of shear-out type against
experimental one for specimen #C2 (after Okutan Baba [22])
ometrical distribution (see e.g. [37]) and the possibility of a numerical prediction
becomes more important.
The prediction of the failure mode is herein pursued numerically making use of
the LMM which builds the collapse mechanism the joint exhibits when the loads
attain their peak value or, more exactly, they reach the evaluated upper bound value
to such peak. The obtained results are organized either in tabular or in band plots
format. In particular, Tables 5, 6, 7 and 8 report the comparison between the experi-
51
mental findings and the numerical predictions in terms of collapse/failure modes for
all the geometries analyzed, all grouped with respect to the four considered config-
52
Fig. 20 Two-serial pins-loaded plate of Fig. 5. Failure modes predicted by numerical simulations
against experimentally detected ones (after Karakuzu et al. [12]): (a) Net-tension at inner hole and
no failure at outer hole for W/D = 2, M/D = 4, E/D = 4; (b) Net-tension at inner hole and
bearing at outer hole for W/D = 3, M/D = 4, E/D = 5; (c) Bearing at inner hole and bearing
at outer hole for W/D = 4, M/D = 4, E/D = 5; (d) Bearing at inner hole and shear-out at outer
hole for W/D = 5, M/D = 4, E/D = 1
urations. In the above tables, as usual in the relevant literature, the symbol B means
bearing, N net-tension, S shear-out, while when more symbols appear separated by
slash (/) the first refers to the failure mode of the outer hole the second to that of the
inner hole; eventually the symbol () means that no failure occurred at the considered hole.
In Figs. 1721 the photographs of some tested specimens, at the joint failure
stage, are reported together with the band plots, on the FE mesh, of the node dis-
53
Fig. 21 Two-parallel pins-loaded plate of Fig. 5. Failure modes predicted by numerical simulations against experimentally detected ones (after Karakuzu et al. [13]): (a) Shear-out for K/D = 4,
M/D = 4, E/D = 1; (b) Bearing for K/D = 4, M/D = 4, E/D = 5; (c) Bearing and Shear-out
for K/D = 3, M/D = 2, E/D = 3; (d) Bearing and net-tension for K/D = 2, M/D = 2,
E/D = 4
54
placement rates components (in y and z directions) at collapse. The latter, evaluated
at last converged iteration of the LMM procedure, locate the distribution of the displacement rates the real structure exhibits at a state of incipient collapse, i.e. the
one attained when the loads reach their peak value or, the computed upper bound
value on it. Such distribution furnishes the searched failure mode but, obviously,
only from a qualitative point of view; what it is meaningful is, in practice, only the
direction of the computed displacement rates. By inspection of these figures it can
be stated that the present procedure seems able to predict, in a quite good manner,
the three multi-pin joints basic failure modes of bearing, shear-out and net-tension.
7 Concluding Remarks
The results of recent studies, concerning a design methodology based on numerical
limit analysis and aimed at the evaluation of the load bearing capacity of single- and
multi-pin joints in orthotropic composite laminates, have been presented.
The proposed methodology uses two numerical methods for limit analysis: the
Linear Matching Method and the Elastic Compensation Method. The lack of associativity, postulated for the adopted Tsai-Wu-type yield criterion for orthotropic
laminates, obliges to search for an upper and a lower bound to the joint peak load
multiplier, the former pursued by the LMM, the latter by the ECM.
Both methods have been rephrased and adapted to the assumed yield criterion
and to a layered treatment of the tackled mechanical problem. It is worth noting
that the limit analysis carried out at lamina level, taking into account the stacking
sequence of the laminate, appears very effective to deal with many of the throughthickness effects influencing the joint behaviour. Such layered formulation is indeed
of general applicability.
The results, obtained for a remarkable number of tests on real prototypes, show
a good ability of the expounded methodology to bracket the peak load detected via
laboratory tests; a fairly good skillfulness to localize the collapse zone and to predict
the related collapse mode; a good performance of the layered approach when dealing
with different joint geometries and laminate lay-ups.
Such information are definitively very useful, at least at a preliminary stage of
a design process. Concerning the tackled problem for example, limit analysis can
be used for some design choices either related to the laminate setting, e.g. better
lay-up, or appropriate components, for example, or the joint setting, in terms of
fastener holes distribution, holes dimensions, ratios between fasteners and laminate
geometrical parameters, etc. Once a good prototype has been settled more accurate
and expensive numerical analyses or experimental tests can be carried out on it
with considerable saving of money and time. Extension of such approach to other
problems of engineering interest seem straightforward.
55
References
1. ADINA R & D Inc (2002) Theory and modeling guide. ADINA R & D, Watertown
2. Aktas A, Dirikolu MH (2004) An experimental and numerical investigation of strength characteristics of carbon-epoxy pinned-joint plates. Compos Sci Technol 64:16051611
3. Ascione F, Feo L, Maceri F (2009) An experimental investigation on the bearing failure load
of glass fibre/epoxy laminates. Composites, Part B, Eng 40:197205
4. Ascione F, Feo L, Maceri F (2010) On the pin-bearing failure load of GFRP bolted laminates:
an experimental analysis on the influence of bolt diameter. Composites, Part B, Eng 41:482
490
5. Bathe KJ (1996) Finite element procedures. Prentice-Hall, New Jersey
6. Camanho PP, Matthews FL (1997) Stress analysis and strength prediction of mechanically
fasteners joints in FRP: a review. Composites, Part A, Appl Sci Manuf 28(A):529547
7. Capsoni A, Corradi L, Vena P (2001) Limit analysis of anisotropic structures based on the
kinematic theorem. Int J Plast 17:15311549
8. Chen HF, Ponter ARS (2001) Shakedown and limit analyses for 3-D structures using the linear
matching method. Int J Press Vessels Piping 78(6):443451
9. Feo L, Marra G, Mosallam AS (2012) Stress analysis of multi-bolted joints for FRP pultruded
composite structures. Compos Struct 94:37693780
10. Godwin EW, Matthews FL (1980) Review of the strengths of joints in fibre-reinforced plastics,
part 1: mechanically fastened joints. Composites 11(3):155160
11. Guillermin O, Kojic M, Bathe KJ (1990) Linear and nonlinear analysis of composite shells.
In: Proceedings STRUCOME 90, Paris, France
12. Karakuzu R, aliskan CR, Aktas M, Iten BM (2008) Failure behavior of laminated composite
plates with two serial pin-loaded holes. Compos Struct 82:225234
13. Karakuzu R, Taylak N, Iten BM, Aktas M (2008) Effects of geometric parameters on failure
behavior of laminated composite plates with two parallel pin-loaded holes. Compos Struct
85:19
14. Karakuzu R, Demirgoren O, Iten BM, Deniz ME (2010) Failure behavior of quasi-isotropic
laminates with three-pin loaded holes. Mater Des 31:30293032
15. Krabbenhft K, Lyamin AV, Sloan SW (2007) Formulation and solution of some plasticity
problems as conic programs. Int J Solids Struct 44:15331549
16. Le CV, Hguyen-Xuan H, Nguyen-Dang H (2010) Upper and lower bound limit analysis of
plates using FEM and second-order cone programming. Comput Struct 88:6573
17. Lubliner J (1990) Plasticity theory. Macmillan, New York
18. Mackenzie D, Boyle JT (1993) A method of estimating limit loads by iterative elastic analysis,
parts I, II, III. Int J Press Vessels Piping 53:77142
19. Makrodimopoulos A, Martin M (2006) Lower bound limit analysis of cohesive-frictional materials using second-order cone programming. Int J Numer Methods Eng 66:604634
20. Makrodimopoulos A, Martin M (2007) Upper bound limit analysis using simplex strain elements and second-order cone programming. Int J Numer Anal Methods Geomech 31:835865
21. Mackerle J (2003) Finite element analysis of fastening and joining: a bibliography (1990
2002). Int J Press Vessels Piping 80:253271
22. Okutan Baba B (2006) Behavior of pin-loaded laminated composites. Exp Mech 46:589600
23. Pisano AA, Fuschi P (2007) A numerical approach for limit analysis of orthotropic composite
laminates. Int J Numer Methods Eng 70:7193
24. Pisano AA, Fuschi P (2011) Mechanically fastened joints in composite laminates: evaluation
of load bearing capacity. Composites, Part B, Eng 42:949961
25. Pisano AA, Fuschi P, De Domenico D (2012) A layered limit analysis of pinned-joints composite laminates: numerical versus experimental findings. Composites, Part B, Eng 43:940
952
26. Pisano AA, Fuschi P, De Domenico D (2013) Peak load prediction of multi-pin joints FRP
laminates by limit analysis. Compos Struct 96:763772
56
27. Pisano AA, Fuschi P, De Domenico D (2013) Failure modes prediction of multi-pin joints
FRP laminates by limit analysis. Composites, Part B, Eng 46:179206
28. Ponter ARS, Carter KF (1997) Limit state solutions, based upon linear elastic solutions with
spatially varying elastic modulus. Comput Methods Appl Mech Eng 140:237258
29. Ponter ARS, Fuschi P, Engelhardt M (2000) Limit analysis for a general class of yield conditions. Eur J Mech A, Solids 19:401421
30. Radenkovic D (1961) Thormes limites pour un materiau de Coulomb dilatation non standardise. C R Acad Sci Paris 252:41034104
31. Rowlands RE (1985) Strength (failure) theories and their experimental correlation. In: Sih GC,
Skudra AM (eds) Handbook of composites, vol 3: failure mechanics of composites. Elsevier,
Amsterdam, pp 71125
32. Salenon J (1972) coulement plastique libre et analyse limite pour les matriaux standards
et non standards. Ann ITBTP 295296:90100
33. Salenon J (1977) Application of the theory of plasticity in soil mechanics. Wiley, Chichester
34. Soden PD, Kaddour AS, Hinton MJ (2004) Recommendations for designer and researchers
resulting from the word-wide failure exercise. Compos Sci Technol 64:589604
35. Staat M, Heitzer M (2001) LISAa European project for FEM-based limit and shakedown
analysis. Nucl Eng Des 206:151166
36. Staat M, Schwartz M, Lang H, Wirtz K, Heitzer M (2003) Design by analysis of pressure
components by non-linear optimization. In: Zeman JL (ed) Pressure vessel technology. Proceedings ICPVT-10. sterreichische Gesellschaft fr Schweitechik, Vienna, pp 5965
37. Thoppul SD, Finegan J, Gibson RF (2009) Mechanics of mechanically fastened joints in
polymer-matrix composite structures-a review. Compos Sci Technol 69:301329
38. Tsai SW, Hann HT (1980) Introduction to composite materials. Technomic, Westport
39. Tsai SW, Wu EM (1971) A general theory of strength for anisotropic materials. J Compos
Mater 5:5880
40. Vu DK, Staat M, Tran IT (2007) Analysis of pressure equipment by application of the primaldual theory of shakedown. Commun Numer Methods Eng 23:213225
41. Weichert D, Ponter ARS (2009) Limit states of materials and structuresdirect methods.
Springer, Berlin
Abstract Determining the load bearing capacity is essential for the design of engineering structures subjected to varying thermo-mechanical loadings. The according
computations can be carried out most conveniently by using shakedown analysis. In
order to obtain realistic results, however, limited kinematical hardening needs to be
taken into account. Moreover, it is necessary to consider arbitrary numbers of loadings leading to n-dimensional loading spaces. Even so, the numerical tools available
for shakedown analysis areup to nowrestricted to either perfectly-plastic material behavior or to a maximum of two independently varying loadings. Thus, the
aim of this paper is to present a numerical procedure, which allows the consideration of limited kinematical hardening in n-dimensional loading spaces. The method
is based on the lower bound shakedown theorem by Melan, which has been extended to limited kinematical hardening by use of a two-surface model. To solve
the resulting nonlinear optimization problem, which is typically characterized by
a large number of variables and constraints, an interior-point algorithm is implemented. Finally, the potential of the procedure is shown by application to a flanged
pipe subjected to three independently varying thermal and mechanical loadings accounting for different yield stress to ultimate stress ratios.
1 Introduction
Determining the load bearing capacity is essential for the design of engineering
structures subjected to varying thermo-mechanical loadings. This is a demanding
task if the loads exceed the structures elastic limit and problems of practical relevance are considered. In principle, two different methods can be used for this, the
conventional step-by-step methods or so-called direct methods.
In the step-by-step methods, the loading path is divided into sufficiently small
loading steps and a full analysis of the evolution of stresses and strains is carried out
for each step. Besides the question of running time, the most obvious disadvantage
J.-W. Simon (B)
Institute of Applied Mechanics, RWTH Aachen University, Mies-van-der-Rohe-Str. 1,
52074 Aachen, Germany
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_3,
Springer Science+Business Media Dordrecht 2014
57
58
J.-W. Simon
of these methods is the fact that the exact knowledge of the complete loading history
is essential for the calculation, which is not realistic in many cases.
This disadvantage can be overcome by the use of direct methods [26, 32, 78,
79]comprising limit and shakedown analysis. Here, the loading path needs not to
be given deterministically but only its bounding envelope. The basement of direct
methods is constituted by the theorems of Koiter [25] and Melan [37, 38]. In the
present work, the statical approach by Melan is used, which gives a lower bound
of the loading factor. From practical point of view, the determination of the lower
bound is of special interest because it leads in principle to conservative solutions.
Since in many engineering applications the materials exhibit kinematical hardening, the latter needs to be incorporated into the procedure in order to obtain realistic results. The statical shakedown theorem in its original formulation is valid
for elastic-perfectly plastic continua as well as for unlimited kinematical hardening
ones. Notably, accounting for only unlimited kinematical hardening does not cover
incremental collapse but solely alternating plasticity, see e.g. [26, 27, 53, 81].
Thus, accounting for limited (or bounded) kinematical hardening is inevitable,
and consequently has been addressed by several authors in the field of shakedown
analysis, e.g. [13, 16, 20, 35, 43, 4752, 57, 6467, 76]. The first explicit formulation for limited kinematical hardening materials has been given by Weichert and
Gro-Weege [76], who introduced a two-surface model. Almost at the same time,
Stein and coauthors proposed another approach based on an overlay model [65
67], which leaded to an equivalent formulation. Later, Heitzer [24] showed how to
transfer these approaches one to the other.
However, Melans theorem leads to nonlinear optimization problems, which are
typically characterized by large numbers of variables and constraints. In the present
work, these optimization problems are solved via the widely used interior-point
method [14, 15, 56, 80]. Commonly used codes based on interior-point methods are
IPOPT [7274], LOQO [6, 18, 69] and KNITRO [11, 75]. Comparative studies can
be found in e.g. [7, 40, 72] and a valuable commented overview of existing optimization codes is given in [39]. In addition, in the last years the program MOSEK
[3, 4] came into the picture of direct methods and has been applied for both (piecewise) linear and second-order conic problems, e.g. [8, 17, 29, 33, 46, 68] at least for
reference solutions.
All of the above mentioned programs have already proven their abilities and robustness. Nonetheless, independent interior-point algorithms have been developed
by several authors, e.g. [1, 21, 28, 30, 44, 45, 70, 71]. Compared to the above mentioned general codes, these algorithms are distinguished by problem-tailored solution strategies. This is extremely important especially in field of direct methods,
because of the usually high number of variables and subsidiary conditions.
It should be mentioned, that a variety of alternative methods have been developed
in recent years. For example, in [82] an eigen-mode method has been proposed,
whereas the so-called Linear Matching Method has been suggested in [54, 55]. Further, a bipotential approach has been invented in [9, 10], and a homogenized method
has been examined e.g. in [31]. In addition, in [5, 42] a piece-wise linearization of
the yield surface is presented, while a strain-driven strategy is given in [17]. More
59
recently, a new direct method has been suggested in [63], in which the cyclic nature
of the expected residual stress distribution at the steady cycle is investigated.
In any case, in the present paper, the convex optimization problem resulting from
the statical shakedown theorem is solved via the interior-point algorithm IPSA recently developed by the author, which is especially tailored to shakedown analysis
problems for von Mises-type materials. Founded on a previous interior-point algorithm IPDCA [1, 22, 23], which has been developed for elastic-perfectly plastic
engineering problems with either one or two varying loads, the new algorithm is distinguished by a particularly problem-oriented solution strategy [5860]. Moreover,
IPSA is capable to solve shakedown problems with n-dimensional loading spaces
[62], such that the examined structures can be subjected to arbitrary numbers of
loadings. This is important, because only few other methods can deal with more
than two loadings, see e.g. [34].
In addition, limited kinematical hardening has already been introduced for twodimensional loading spaces [61]. In this paper, an extension of this algorithm is presented, such that limited kinematical hardening can be considered in n-dimensional
loading spaces. The methods potential is illustrated by application to a numerical
example. In particular, the shakedown domain for a flanged pipe subjected to three
independently varying thermo-mechanical loads is presented.
(1)
Here, E (X, t) denotes the stress state, which would occur in a fictitious purely
elastic reference body under the same conditions as the original one. Clearly, the
residual stresses satisfy the equilibrium condition, which can be transferred to a
system of linear equations using the principle of virtual work, as shown e.g. in [19]
NG
Cr r = 0.
(2)
r=1
Hereby, the system has been discretized using the finite element method (FEM)
and thus the stresses are approximately evaluated in the G AUSS points r [1, NG].
The equilibrium matrices Cr depend on the geometry and the chosen element type.
60
J.-W. Simon
Fig. 1 Kinematical
hardening considered as
translation of the yield
surface in stress space
The kinematical boundary conditions are taken into account considering the virtual
displacements to be kinematical admissible.
Let the considered body be subjected to NL varying loads. Then, the according loading domain is polyhedral with NC = 2NL corners. As shown in [26], it is
sufficient to only consider these corners to ensure shakedown for all possible loading paths inside of the loading domain. Then, introducing the loading factor > 1,
Melans statical shakedown theorem can be formulated as an optimization problem:
(PMelan ) SD = max
NG
Cr r = 0,
(3a)
r=1
E,j
f r + r , Y,r 0,
(3b)
(4)
61
E,j
E,j
r = r r = r
+ r r .
(5)
Cr r = 0,
(6a)
r=1
(6b)
(6c)
(7a)
cH (x) 0,
(7b)
cY (x) 0,
(7c)
xR .
(7d)
62
J.-W. Simon
terms, which penalize directions leading outside of the feasible region. Thereby, the
barrier parameter is introduced, which is tending to zero during the iteration.
f (x, y, z, w H , w Y )
n
mI
mI
n
= f (x)
log(yi ) +
log(zi ) +
log(wH,j ) +
log(wY,j ) . (8)
i=1
j =1
i=1
j =1
(9a)
cH (x) w H = 0,
(9b)
cY (x) wY = 0,
(9c)
x y + z = 0,
(9d)
w H > 0,
w Y > 0,
y > 0,
z>0
(9e)
(11a)
y LH = Y 1 e + s = 0,
(11b)
z LH = Z 1 e s = 0,
(11c)
wH LH = W 1
H e + H = 0,
(11d)
wY LH = W 1
Y e + Y = 0,
(11e)
E LH = (AH x) = 0,
H LH = cH (x) w H = 0,
Y LH = cY (x) wY = 0,
(11f)
s LH = (x y + z) = 0
where: C H (x) = cH (x) x RmI n and C Y (x) = cY (x) x RmI n .
(11g)
(11h)
(11i)
63
For consistency during the iteration, the new variable r = s is introduced into
Eq. (11c). Both of these variables are tending to zero during the iteration. In addition, Eqs. (11b)(11e) are multiplied by the matrices Y , Z, W H and W Y , respectively. Merging all variables of the problem to the vector , the resulting system of
optimality conditions can be expressed by the function F H
():
e Y S e
e Z R e
e W H H e
e W Y Y e
H
= 0.
F () =
AH x
cH (x) wH
cY (x) wY
x y +z
r +s
(12)
Equation (12) constitutes a system of nonlinear equations, which will be linearized using the N EWTON method. The variables k+1 of the subsequent iteration
step k + 1 are computed from the variables k of the previous one k and the step
values k :
k+1 = k + k k ,
(13)
where: J ( k ) = LH () |= k .
(14)
0 0
0 0 ATH C TH (x) C TY (x) I n 0
S 0
0 0
0
0
0
Y
0
0 R
0 0
0
0
0
0
Z
0
WH
0
0
0
0 0 H 0
0
0
WY
0
0
0 0
0 Y
.
0 0
0 0
0
0
0
0
0
0 0 I mI 0
0
0
0
0
0
0 0
0 I mI 0
0
0
0
0
I n I n 0 0
0
0
0
0
0
0 0
0 0
0
0
0
I n I n
(15)
The system Eq. (14) is reduced by successive elimination of those equations,
which involve diagonal matrices. After substituting the variables s, r, y, z,
2x LH
J () =
AH
C H (x)
C Y (x)
I n
0
64
J.-W. Simon
d1
x
C TY (x)
0
E = d 2 .
0 H d H
3
Y
EY
d Y3
(16)
(17a)
d 2 = AH x,
(17b)
( 2x LH + E 1 )
AH
C H (x)
C Y (x)
ATH
0
0
0
C TH (x)
0
EH
0
1
dH
3 = cH (x) + H
e,
(17c)
d Y3 = cY (x) + 1
Y
(17d)
where:
b1 = x + z + R 1 S 1 e + R 1 Z s,
1
E 1 = S 1 Y + R 1 Z ,
(17e)
(17f)
E H = W H 1
H ,
(17g)
E Y = W Y 1
Y .
(17h)
5 Numerical Aspects
The new algorithm has already been presented in [59], and some numerical aspects
have been discussed in [60]. Here, only the most important issues are briefly recapitulated. The algorithms mode of operation is illustrated in Fig. 2.
As one can see, there exist two different iterations:
The outer iteration is the major loop in Fig. 2, where in each iteration step the
barrier parameter is reduced in accordance with the update-rule. Thereby, during the outer iteration process, is a sequence tending to zero.
The inner iteration is the minor loop in Fig. 2, which ensures that the approximate
solution of the linearized system is in the close proximity of the exact solution,
such that the KKT conditions are satisfied sufficiently accurate.
Since the full Newton step k computed from Eq. (14) may happen to be too
large, it has to be damped in these cases. This is done by a linesearch procedure,
which actually is a third iteration within the outer and the inner one.
65
(18)
With this choice for the initial solution vector, the condition Eq. (9a) is satisfied
a priori. The other variables can be fitted such that the remaining conditions from
Eq. (12) are satisfied as well. Note, that one of the variables can be chosen arbitrarily.
Moreover, the condition r = s cannot be satisfied by any choice of positive r 0
and s 0 . This condition is necessary to enforce r and s to be sequences tending to
zero during the iteration, though. Except of this, the starting point 0 is feasible.
66
J.-W. Simon
For this reason, the damping factors i are introduced, which are merged into
the diagonal matrix k in Eq. (13). In principle, damping factors can be defined
for each of the variables separately. Nonetheless, it is common practice in linear
and nonlinear programs to use two different values P for the primal variables P
and D for the dual variables D
P = [x, y, z, wY , w H ]T ,
(19a)
D = [E , Y , H , s, r]T .
(19b)
(20a)
D = [Y , H , s, r]T .
(20b)
Then, the damping factors for the nonnegativity condition read as follows:
P = max P +
P 0; P = [y, z, w Y , w H ] ,
D = max D +
D 0; D = [Y , H , s, r] .
(21a)
(21b)
In order to prevent hitting the boundary, these values are multiplied by the constant factor 0 = 0.995.
Damping of the Newton step with the nonnegativity condition may still not be
sufficient, because it is possible that the resulting step values do not lead to a descent
direction in both the infeasibilities and the objective function. In order to ensure that
the computed direction is a decreasing one, the following merit function , is
introduced for a linesearch procedure. The term in squared brackets [.] is optional
but should be used in order to avoid the Maratos effect [36]
Ax
cY (x) wY
, () = f (x, y, z, w Y , w H ) +
c
(x)
w
2
H
H
x y +z
+ (A x) E . (22)
2
, () +
,
(; )
where: = T 0 P .
(23)
67
Length L
386.9
Inner radius Ri
60.0
68.1
77.8
90.5
Once the value T is determined, the total damping factors can be calculated.
P = T 0 P ,
D = T 0 D .
(24)
Concerning the penalty parameter , a new update-rule is used, which has been
recently developed by the author. It is based on the necessary condition of a decreasing direction, (; ) < 0. For details, the reader is referred to [59].
1
max scaled infeasibility() and o = i |=0 .
The maximum scaled infeasibility can be taken from the computation of the
break conditions. As suggested in [2], we use = 5.
6 Numerical Example
The proposed method was applied to a flanged pipe with three different outer radii,
see Fig. 3(a), already considered by M OUHTAMID [41] and W EICHERT et al. [77].
Both the dimensions as well as the material data were adopted from [41], see Tables 1 and 2. The FEM-analysis was carried out with the software package ANSYS
using the isoparametric hexahedral solid element solid45. Taking advantage of the
systems rotational symmetryas shown in Fig. 3(b)the applied mesh consisted
of 265 elements and 678 nodes, where one element across the thickness was used.
68
J.-W. Simon
Fig. 3 System, model and equivalent elastic stresses for the flanged pipe
Table 2 Thermal and
mechanical characteristics
2.0 105
200
Poissons ratio
0.3
Density
[kg/m3 ]
7.9 103
15
500
1.6 105
69
In all casesboth the perfectly plastic and the hardening onethe two mechanisms of alternating plasticity and incremental collapse can be clearly distinguished:
In case of predominating axial force, all shakedown curves coincide with the one for
unlimited hardening, indicating that alternating plasticity is decisive here. Hence, no
influence of hardening can be observed.
By contrast, failure is due to incremental collapse in the regime of predominating
internal pressure, where the limited kinematical hardening leads to an increase of
the according shakedown domains in direct proportion with the ratio H /Y . Thus,
the hardening curves (solid lines) coincide with the corresponding none-hardening
ones with premultiplied yield stress (dash-dot lines) in this range. In all cases, the
two curves pass into each other seamlessly.
It should be noticed, that the unlimited hardening curve does only accord partly
with double the elastic domain. Even so, it is frequently stated in the literature, that
these curves have to accord in the whole domain, which simply is wrong. In fact,
they have to coincide only at the axis intercepts. In the remaining domain, they
maybut do not mustbe the same.
For validation, in Fig. 5 the results obtained by the new method are compared
to those reported in [41], which have been computed on the basis of the augmented
Lagrangian method using the program L ANCELOT [12]. In general, matching of the
results is satisfying, especially for limited kinematical hardening with H = 1.5Y .
However, slight differences exist resulting from different elastic solutions. These can
be explained by the use of different meshes. In particular, the maximum equivalent
stress under axial force is 106.465 MPa in [41], whereas the current calculation
yields 100.143 MPa.
70
J.-W. Simon
71
72
J.-W. Simon
Closing, the characteristic numerical details are reported in Table 3. As one can
see, the number of iterations is not as much affected as the running time. Moreover,
in the considered example, the number of loadings has a larger impact than the
hardening, even though the numbers of variables and constraints are comparable.
73
7 Conclusion
In this paper, a method to compute the shakedown factors of engineering structures
subjected to varying thermal and mechanical loadings has been presented using the
lower bound approach. The method allows for taking into account limited kinematical hardening by application of a two-surface model, where both the yield and the
bounding surface are described by the von Mises criterion. Further, it is capable of
considering arbitrary numbers of thermal and mechanical loadings. The methods
74
J.-W. Simon
3 independent loads
Perfectly plastic
Hardening
Perfectly plastic
Hardening
44521
86921
86921
171721
mE
33834
65634
76234
150434
mI
8480
16960
16960
33920
400
481
2318
3017
48
57
295
837
Iterations
CPU-time
a Dell
[s]a
Precision T7500 with Xeon E5620-processor with 2400 MHz and 12 GB RAM
potential has been illustrated by application to a numerical example from the field
of power plant engineering.
Acknowledgements I cordially thank Prof. Dieter Weichert for the fruitful discussions and the
support, which made this work possible.
References
1. Akoa FB, Hachemi A, An LTH, Mouhtamid S, Tao PD (2007) Application of lower bound
direct method to engineering structures. J Glob Optim 37(4):609630
2. Akrotirianakis I, Rustem B (2000) A primal-dual interior-point algorithm with an exact and
differentiable merit function for general nonlinear programming problems. Optim Methods
Softw 14(1/2):136
3. Andersen ED, Jensen B, Jensen J, Sandvik R, Worse U (2009) MOSEK version 6. Technical
report TR-2009-3
4. Andersen ED, Roos C, Terlaky T (2003) On implementing a primal-dual interior-point method
for conic quadratic optimization. Math Program 95(2):249277
5. Ardito R, Cocchetti G, Maier G (2008) On structural safety assessment by load factor maximization in piecewise linear plasticity. Eur J Mech A, Solids 27:859881
6. Benson HY, Shanno DF, Vanderbei RJ (2002) Interior-point methods for nonconvex nonlinear
programming: filter methods and merit functions. Comput Optim Appl 23(2):257272
7. Benson HY, Shanno DF, Vanderbei RJ (2003) A comparative study of large-scale nonlinear optimization algorithms. In: Di Pillo G, Murli A (eds) High performance algorithms and
software for nonlinear optimization. Kluwer Academic, Princeton, pp 95127
8. Bisbos CD, Makrodimopoulos A, Pardalos PM (2005) Second-order cone programming approaches to static shakedown analysis in steel plasticity. Optim Methods Softw 20(1):2552
9. Bouby C, De Saxc G, Tritsch J-B (2009) Shakedown analysis: comparison between models with the linear unlimited, linear limited and non-linear kinematic hardening. Mech Res
Commun 36:556562
10. Bousshine L, Chaaba A, De Saxce G (2003) A new approach to shakedown analysis for nonstandard elastoplastic material by the bipotential. Int J Plast 19(5):583598
11. Byrd RH, Hribar ME, Nocedal J (2000) An interior point algorithm for large-scale nonlinear
programming. SIAM J Optim 9(4):877900
12. Conn AR, Gould NIM, Toint PL (1992) In: LANCELOT: a Fortran package for large-scale
nonlinear optimization (release A). Springer series in computational mathematics, vol 17.
Springer, Heidelberg
75
13. Corigliano A, Maier G, Pycko S (1995) Kinematic criteria of dynamic shakedown extended
to nonassociate constitutive laws with saturation nonlinear hardening. Rend Accad Lincei IX
6:5564
14. El-Bakry AS, Tapia RA, Tsuchiya T, Zhang Y (1996) On the formulation and theory of the
Newton interior-point method for nonlinear programming. J Optim Theory Appl 89:507541
15. Forsgren A, Gill PE, Wright MH (2002) Interior methods for nonlinear optimization. SIAM
Rev 44(4):525597
16. Fuschi P (1999) Structural shakedown for elastic-plastic materials with hardening saturation
surface. Int J Solids Struct 36:219240
17. Garcea G, Leonetti L (2011) A unified mathematical programming formulation of strain driven
and interior point algorithms for shakedown and limit analysis. Int J Numer Methods Eng
88(11):10851111
18. Griva I, Shanno DF, Vanderbei RJ, Benson HY (2008) Global convergence analysis of a
primal-dual interior-point method for nonlinear programming. Algorithmic Oper Res 3(1):12
19
19. Gro-Weege J (1997) On the numerical assessment of the safety factor of elastic-plastic structures under variable loading. Int J Mech Sci 39(4):417433
20. Gro-Weege J, Weichert D (1992) Elastic-plastic shells under variable mechanical and thermal
loads. Int J Mech Sci 34:863880
21. Hachemi A, An LTH, Mouhtamid S, Tao PD (2004) Large-scale nonlinear programming and
lower bound direct method in engineering applications. In: An LTH, Tao PD (eds) Modelling,
computation and optimization in information systems and management sciences. Hermes Science, London, pp 299310
22. Hachemi A, Mouhtamid S, Nguyen AD, Weichert D (2009) Application of shakedown analysis to large-scale problems with selective algorithm. In: Weichert D, Ponter ARS (eds) Limit
states of materials and structures. Springer, Berlin, pp 289305
23. Hachemi A, Mouhtamid S, Weichert D (2005) Progress in shakedown analysis with applications to composites. Arch Appl Mech 74:762772
24. Heitzer M (1999) Traglast- und Einspielanalyse zur Bewertung der Sicherheit passiver Komponenten. PhD thesis, Forschungszentrum Jlich, RWTH Aachen, Germany
25. Koiter WT (1960) General theorems for elastic-plastic solids. In: Sneddon IN, Hill R (eds)
Progress in solid mechanics. North-Holland, Amsterdam, pp 165221
26. Knig JA (1987) Shakedown of elastic-plastic structures. Elsevier, Amsterdam
27. Knig JA, Siemaszko A (1988) Shakedown of elastic-plastic structures. Ing-Arch 58:5866
28. Krabbenhft K, Damkilde L (2003) A general nonlinear optimization algorithm for lower
bound limit analysis. Int J Numer Methods Eng 56:165184
29. Krabbenhft K, Lyamin AV, Sloan SW (2007) Formulation and solution of some plasticity
problems as conic programs. Int J Solids Struct 44:15331549
30. Krabbenhft K, Lyamin AV, Sloan SW, Wriggers P (2007) An interior-point algorithm for
elastoplasticity. Int J Numer Methods Eng 69:592626
31. Magoariec H, Bourgeois S, Dbordes O (2004) Elastic plastic shakedown of 3d periodic heterogeneous media: a direct numerical approach. Int J Plast 20(89):16551675
32. Maier G, Pastor J, Ponter ARS, Weichert D (2003) Direct methods of limit and shakedown
analysis. In: de Borst R, Mang HA (eds) Comprehensive structural integrityfracture of materials from nano to macro. Numerical and computational methods, vol 3. Elsevier, Amsterdam,
pp 637684
33. Makrodimopoulos A (2006) Computational formulation of shakedown analysis as a conic
quadratic optimization problem. Mech Res Commun 33:7283
34. Malena M, Casciaro R (2008) Finite element shakedown analysis of reinforced concrete 3d
frames. Comput Struct 86(1112):11761188
35. Mandel J (1976) Adaptation dune structure plastique ecrouissable et approximations. Mech
Res Commun 3:483488
36. Maratos N (1978) Exact penalty function algorithms for finite dimensional and control
opimization problems. PhD thesis, University of London, UK
76
J.-W. Simon
37. Melan E (1938) Der Spannungszustand eines Mises-Henckyschen Kontinuums bei vernderlicher Belastung. Sitzungsber Akad Wiss Wien, Math-Nat Kl, Abt IIa 147:7387
38. Melan E (1938) Zur Plastizitt des rumlichen Kontinuums. Ing-Arch 9:116126
39. Mittelmann H (2010) Decision tree for optimization software. http://plato.asu.edu/guide.thml
40. Morales JL, Nocedal J, Waltz RW, Lie G, Goux J-P (2003) Assessing the potential of interior
methods for nonlinear optimization. In: Biegler LT, Ghattas O, Heinkenschloss M, van Bloemen Waander B (eds) Large-scale PDE-constrained optimization, vol 30. Springer, Berlin,
pp 167183
41. Mouhtamid S (2007) Anwendung direkter Methoden zur industriellen Berechnung von Grenzlasten mechanischer Komponenten. PhD thesis, Institute of General Mechanics, RWTH
Aachen University, Germany
42. Ngo N, Tin-Loi F (2007) Shakedown analysis using the p-adaptive finite element method and
linear programming. Eng Struct 29(1):4656
43. Nguyen Q-S (2003) On shakedown analysis in hardening plasticity. J Mech Phys Solids
51:101125
44. Pastor F, Loute E (2010) Limit analysis decomposition and finite element mixed method.
J Comput Appl Math 234(7):22132221
45. Pastor F, Loute E, Pastor J, Trillat M (2009) Mixed method and convex optimization for
limit analysis of homogeneous Gurson materials: a kinematic approach. Eur J Mech A, Solids
28:2535
46. Pastor F, Thor P, Loute E, Pastor J, Trillat M (2008) Convex optimization and limit analysis:
application to Gurson and porous Drucker-Prager materials. Eng Fract Mech 75:13671383
47. Pham DC (2007) Shakedown theory for elastic plastic kinematic hardening bodies. Int J Plast
23:12401259
48. Pham DC (2008) On shakedown theory for elastic-plastic materials and extensions. J Mech
Phys Solids 56:19051915
49. Pham DC, Weichert D (2001) Shakedown analysis for elastic-plastic bodies with limited kinematical hardening. Proc R Soc Lond A 457:10971110
50. Pham PT, Vu DK, Tran TN, Staat M (2010) An upper bound algorithm for shakedown analysis
of elastic-plastic bounded linearly kinematic hardening bodies. In: Proc ECCM 2010
51. Polizzotto C (1986) A convergent bounding principle for a class of elastoplastic strainhardening solids. Int J Plast 2(4):359370
52. Polizzotto C (2010) Shakedown analysis for a class of strengthening materials within the
framework of gradient plasticity. Int J Plast 26(7):10501069
53. Ponter ARS (1975) A general shakedown theorem for elastic plastic bodies with work hardening. In: Proc SMIRT-3, p L5/2
54. Ponter ARS (2002) A linear matching method for shakedown analysis. In: Weichert D,
Maier G (eds) Inelastic behaviour of structures under variable repeated loadingdirect analysis methods, pp 267318
55. Ponter ARS, Chen HF (2005) Direct methods for limits in plasticity. Arch Mech 57:171188
56. Potra FA, Wright SJ (2000) Interior-point methods. J Comput Appl Math 124:281302
57. Pycko S, Maier G (1995) Shakedown theorems for some classes of nonassociative hardening
elastic-plastic material models. Int J Plast 11(4):367395
58. Simon J-W, Chen M, Weichert D (2012) Shakedown analysis combined with the problem of
heat conduction. J Press Vessel Technol 134(2):021206
59. Simon J-W, Weichert D (2011) Numerical lower bound shakedown analysis of engineering
structures. Comput Methods Appl Mech Eng 200:28282839
60. Simon J-W, Weichert D (2012) Interior-point method for lower bound shakedown analysis of
von mises-type materials. In: de Saxc G, Oueslati A, Charkaluk E, Tritsch J-B (eds) Limit
states of materials and structuresdirect methods, vol 2. Springer, Berlin, pp 103128
61. Simon J-W, Weichert D (2012) Shakedown analysis of engineering structures with limited
kinematical hardening. Int J Solids Struct 49(4):21772186
62. Simon J-W, Weichert D (2012) Shakedown analysis with multidimensional loading spaces.
Comput Mech 49(4):477485
77
63. Spiliopoulos KV, Panagiotou KD (2012) A direct method to predict cyclic steady states of
elastoplastic structures. Comput Methods Appl Mech Eng 223(224):186198
64. Staat M, Heitzer M (2002) The restricted influence of kinematical hardening on shakedown
loads. In: Proc WCCM V
65. Stein E, Zhang G, Huang Y (1993) Modeling and computation of shakedown problems for
nonlinear hardening materials. Comput Methods Appl Mech Eng 103(12):247272
66. Stein E, Zhang G, Knig JA (1992) Shakedown with nonlinear strain-hardening including
structural computation using finite element method. Int J Plast 8(1):131
67. Stein E, Zhang G, Mahnken R, Knig JA (1990) Micromechanical modelling and computation
of shakedown with nonlinear kinematic hardening including examples for 2-D problems. In:
Axelard DR, Muschik W (eds) Recent developments of micromechanics. Springer, Berlin
68. Trillat M, Pastor J (2005) Limit analysis and Gursons model. Eur J Mech A, Solids 24:800
819
69. Vanderbei RJ (1999) LOQO: an interior point code for quadratic programming. Optim Methods Softw 1112:451484
70. Vu DK, Staat M (2007) Analysis of pressure equipment by application of the primal-dual
theory of shakedown. Commun Numer Methods Eng 23(3):213225
71. Vu DK, Yan AM, Nguyen-Dang H (2004) A primal-dual algorithm for shakedown analysis of
structures. Comput Methods Appl Mech Eng 193:46634674
72. Wchter A (2002) An interior point algorithm for large-scale nonlinear optimization with applications in process engineering. PhD thesis, Carnegie Mellon University, Pittsburgh, Pennsylvania
73. Wchter A, Biegler LT (2005) Line-search filter methods for nonlinear programming: motivation and global convergence. SIAM J Optim 16(1):131
74. Wchter A, Biegler LT (2006) On the implementation of an interior-point filter line-search
algorithm for large-scale nonlinear programming. Math Program 106(1):2557
75. Waltz RA, Morales JL, Nocedal J, Orban D (2006) An interior algorithm for nonlinear optimization that combines line search and trust region steps. Math Program 107(3):391408
76. Weichert D, Gro-Weege J (1988) The numerical assessment of elastic-plastic sheets under
variable mechanical and thermal loads using a simplified two-surface yield condition. Int J
Mech Sci 30(10):757767
77. Weichert D, Hachemi A, Mouhtamid S, Nguyen AD (2008) On recent progress in shakedown
analysis and applications to large-scale problems. In: IUTAM symposium on theoretical, computational and modelling aspects of inelastic media, vol 11, pp 349359
78. Weichert D, Maier G (2000) Inelastic analysis of structures under variable repeated loads.
Kluwer Academic, Dordrecht
79. Weichert D, Ponter ARS (2009) Limit states of materials and structures. Springer, Wien
80. Wright MH (2004) The interior-point revolution in optimization: history, recent developments
and lasting consequences. Bull Am Math Soc 42(1):3956
81. Zarka J, Casier J (1981) Elastic-plastic response of a structure to cyclic loading: practical rule.
In: Nemat-Nasser S (ed) Mechanics today, vol 6. Pergamon, New York
82. Zhang T, Raad L (2002) An eigen-mode method in kinematic shakedown analysis. Int J Plast
18:7190
Abstract Numerical techniques for the computation of strict bounds in limit analyses have been developed for more than thirty years. The efficiency of these techniques have been substantially improved in the last ten years, and have been successfully applied to academic problems, foundations and excavations. We here extend the theoretical background to problems with anchors, interface conditions, and
joints. Those extensions are relevant for the analysis of retaining and anchored walls,
which we study in this work. The analysis of three-dimensional domains remains
as yet very scarce. From the computational standpoint, the memory requirements
and CPU time are exceedingly prohibitive when mesh adaptivity is employed. For
this reason, we also present here the application of decomposition techniques to
the optimisation problem of limit analysis. We discuss the performance of different
methodologies adopted in the literature for general optimisation problems, such as
primal and dual decomposition, and suggest some strategies that are suitable for the
parallelisation of large three-dimensional problems. The proposed decomposition
techniques are tested against representative problems.
79
80
1 Introduction
1.1 Limit Analysis Problem
According to the lower (primal) and upper (dual) bound theorem of limit analysis,
the bearing capacity of a structure is equal to (i) the maximum load factor under
equilibrium conditions and with plastically admissible stresses (i.e. they belong
to a set B), or alternatively, to (ii) the minimum dissipation energy D(v) of a kinematically admissible velocity field v. Mathematically, the resulting bearing capacity
has the structure of a saddle point problem that can be written as [6],
= min
(v)=1
max
B
a( ,v)=(v)
a( , v)
(1)
where the linear form (v) is the power dissipated by the external loads, while the
bilinear form a( , v) is the internal dissipated power. For a given domain , subjected to external surface load g and body load f , these forms are explicitly given
by:
(v) =
f v d +
g v d,
n d
: JvK
: (v) d +
a( , v) =
with the (unknown) region of where the velocity is discontinuous, JvK the
velocity discontinuity, and n the normal vector at this discontinuity. The operator
= 12 (a b + b a) (see also [16]
is the symmetrised dyadic product such that a b
for equivalent definitions of discontinuous velocities). The saddle point problem or
min max problem can be also rewritten as,
=
max
B
a( ,v)=(v),v
= min D(v)
(v)=1
(2)
(3)
max
B
a( ,v)=(v)
a( , v).
Equations (2) and (3) are the primal and dual form of the saddle point problem
in (1). From the dual form in (3), it turns out that the velocities must be associated,
that is, that (v) B and JvK n B, with B the sub-gradient of set B.
The analytical saddle point problem in (1) is illustrated in Fig. 1(a). A stress field
such that B and for which the equilibrium condition a( , v) = (v), (v)
81
Fig. 1 Illustration of saddle point problem for the (a) analytical solution of limit analysis problem, (b) lower bound discrete limit analysis problem, and (c) upper bound discrete limit analysis
problem
is satisfied everywhere is so-called a statically admissible stress space. Similarly, a velocity field such that (v) = 1 and for which the associative conditions
JvK n, (v) B is satisfied everywhere is so-called a kinematically admissible
space. The saddle point problem in (1) states that the bearing capacity (or maximum
load factor) of a structure is equal to the internal dissipated energy of a statically
admissible stress space and a kinematically admissible velocity space v .
82
eq1 = 0,
eq1 LB
A + F
A
eq2 LB = 0,
s.t.
eq3 LB + F eq3 = 0,
LB,e
B, e = 1, . . . , Ne , i = 1, . . . , nsd + 1
i
(4)
l(v U B ) = 1
U B,e
) B , e = 1, . . . , Ne , i = 1, . . . , nsd + 1,
s.t. (v i
Jv U B K n
B , = 1, . . . , N , j = 1, . . . , n
j
(5)
sd
where nsd is the number of space dimensions, and the three block equations in (4)
correspond respectively to the intra-element equilibrium, the inter-element equilibrium, and the Neumann boundary conditions. The vectors LB and v U B contain
B,e
of element e, and the velocities v U
and
the collection of all nodal stresses LB,e
i
i
U B,
vj
for each node i of element e or each node j of edge , respectively.
The problems above can be solved efficiently using available optimisation programs [1, 17, 18]. Moreover, for the usual plasticity criteria such as von Mises
or Mohr-Coulomb in two dimensions, we can apply linear transformations of the
stress variables that turn the membership conditions B into second order cones
(SOC), which can be handled by the mentioned optimisation software. After applying in such transformations, the optimisation problems above turn into,
LB = max
x
Aeq2 x = 0,
s.t.
LB,e
K , e = 1, . . . , Ne , i = 1, . . . , nsd + 1,
xi
83
Fig. 3 Subdivision of a triangular element in 2D (a) and tetrahedron in 3D (b), when their elemental gap contribution a is larger than a pre-defined threshold
U B = min b v U B
(6)
vU B
UB
l(v ) = 1
B,e
K ,
e = 1, . . . , Ne , i = 1, . . . , nsd + 1,
s.t. v U
i UB
Jv Ki n K , = 1, . . . , N , i = 1, . . . , nsd
U B, v
y
UB
d
LB n v
y
UB
d.
%
%
These bound gaps satisfy the properties, U B LB = e e + ,
e 0 and 0, which make them good candidates to estimate the errors
of the lower and upper bound solution. These quadratures are obtained only if appropriate quadratures are employed to compute the integrals: Gauss quadrature for
all the terms excepting the first integral in (see [14] for the justification of this).
We have used them to design an adaptive remeshing strategy employed in the results
Sect. 4. Any element or edge with a gap contribution higher than a certain threshold
will be subdivided according to the pattern shown in Fig. 3.
84
1. Interface material that splits two different materials with specific admissibility
criterion for the common boundary.
2. Duplicated edges: in two-dimensional applications, it may convenient to overlap
materials or structural elements such as ties or anchors. In these situations, it is
required to have edges that joint one element on one side and two elements, B
and B , on the other side.
3. Modelling of joints such as articulated joints in anchors and anchor-wall interface.
We briefly describe how to include in each case the corresponding constraints in
the optimisation problem (see [15] for further details).
i = 1, 2.
(9)
The vector n is the normal to the interface edge or face in 3D. Some of admissible
sets that may be employed in common problems are depicted in Fig. 4. We also note
that the nodal velocities at the two edges between A and B, indicated in Fig. 4(a)
with circles, correspond in fact to the Lagrange multipliers associated with these
constraints: the velocities at one edge are associated with Eq. (7) while those at the
other edge correspond to Eq. (8).
Figure 5 shows the usual admissible domain BI for the stresses at the interfaces.
The criteria in Fig. 5(c) has been included for completeness reasons, but it is unrealistic and has not practical interest. The subscript N and T denote the normal and
tangential components of the traction vector n at the interface. In our implementation, we impose admissibility domains for the stress tensors I , which imply the
usual admissibility conditions for the traction vectors I n.
85
is an idealisation of the real three-dimensional situation. We consider the twodimensional plane strain analysis with additional superimposed elements (reinforcements, ties or anchors) that are analysed in plane-stress and connected to the soil.
We have modelled the structural elements as solid elements and not linear elements,
which prevents the presence of point loads and therefore allows us to preserve the
strictness of the bounds.
Computationally, we need to deal with edges where the soil elements on side A
are connected to two types of elements: other soil elements (with variables B) and
those superimposed elements that model the structural elements (with variables B ).
Figure 6 shows such an edge in the lower and upper bound formulation. We will call
those edges as duplicated edges.
In the lower bound formulation, we need to modify the equilibrium constraints
of the edges, which now read,
A
B
i B
n = 0,
i i
i = 1, 2
B
B
where A
i , i an i are respectively the nodal stresses at sides A, B nd B . Since
there is one equilibrium equation per common node, each duplicated edge requires
two nodal velocities, as indicated with circles in Fig. 6(a).
In the upper bound formulation, the dissipation power at the edge corresponds
to the sum of the power dissipated between elements A and B, and the dissipated
power between elements A and B . Formally, the total dissipation power at the edge
correspond to the sum of two integrals along the common edge:
B
AB
A
a ( , v) =
v v d +
AB v B v A d.
This expression of the dissipated power is equivalent to extending the stress space
at the edges with two superimposed linear stress spaces, AB and AB , indicated
by two pairs of squares in Fig. 6(b). By imposing that each one of the four nodal
86
BI ,
AB
i
i = 1, 2,
with BI the admissible set for the interface conditions, the admissibility of the
B
A
B
corresponding velocity jumps v A
i v i and v i v i is ensured, and therefore the
strictness of the upper bound is guaranteed. The nodal velocities are indicated with
circles in Fig. 6(b).
2.3 Joints
Some of the practical problems in limit analysis include anchors, ties or other structural elements that are connected through joints. In this case, the kinematic constraints must be included in the exact problem described in Sect. 1.1, and also modelled in such a way that the strictness of the bounds is preserved. In the lower bound
formulation, the presence of joints is modelled by including solely a point-to-point
equilibrium condition, that is,
A
B nd = 0,
(10)
which replaces the equilibrium along the whole edge between elements. In the upper
bound problem, the construction of a kinematically admissible space for the velocities is constructed by constricting the relative displacements of the joints in such a
manner that only rotations with respect to the joint centre are allowed. Formally, this
is achieved by replacing the associative velocity field at the joint by the following
constraint in the upper bound problem in (5):
JvK1 + JvK2 = 0.
(11)
Figure 7 shows a schematic of the stress and velocity spaces employed for the
modelling of the joints in the lower and upper bound problem. We point out that
87
the joint is without friction, since no dissipated energy is associated to the relative
rotation at the joint. This is computational made explicit by not imposing any equilibrium relation associated to the rotations in Eq. (10), and imposing exactly the
kinematic relation in (11), i.e. the associated Lagrangian multiplier (variable s J in
Fig. 7) is free.
3 Decomposition Techniques
In order to reduce the memory requirement of realistic three dimensional problems
we propose a decomposition of the optimization problems, which is based on the
ideas explained in [5, 8]. While the decomposition techniques for optimisation problems is a relatively recent topic, its application to limit analysis and other plasticity
problems has found far less attention [11]. We also refer the reader to [10], where
alternative decomposition techniques of the limit analysis problem has been introduced. We here first briefly describe some of the general ideals of decomposition of
optimisation problems.
Ax = b,
x 0,
whose dual reads,
bT y = max bT y
y
AT y c.
(12)
88
x 1 ,x 2
A1 x 1 + A2 x 2 = b,
x 1 0,
(13)
x 2 0.
x 1 ,x 2
b
+ t,
2
b
A2 x 2 = t,
2
A1 x 1 =
x 1 0,
x 2 0,
(14)
t is free.
Given a fixed value of t, the optimum value may be obtained as a result of the inner
minimums,
fi (t) = min cTi x i
xi
Ai x i =
b
+ (1)i+1 t,
2
(15)
x i 0 (i = 1, 2),
so called sub-problems or slave problems and the following master problem:
min f1 (t) + f2 (t),
t
(16)
which only depends on the global variable t. The Lagrangian function of problem
(14) is given by:
L(x 1 , x 2 ; y 1 , y 2 , w 1 , w 2 )
b
b
= cT1 x 1 + cT2 x 2 + y T1
+ t A1 x 1 + y T2
t A2 x 2 wT1 x 1
2
2
wT2 x 2
= cT1 x 1
+ y T1
b
T
T b
A1 x 1 + c 2 x 2 + y 2
A1 x 1 + t T (y 1 y 2 )
2
2
= L1 (x 1 , t; y 1 , w 1 ) + L2 (x 2 , t; y 2 , w2 ),
with
Li (x i , t; y i , wi ) = cTi x 1 + y Ti
b
+ (1)i+1 t A1 x 1 w Ti x i
2
89
(i = 1, 2).
It then follows that we can rewrite the optimum primal objective cT x as,
cT x = cT1 x 1 + cT2 x 2 = min
t
min max Li (x i , t; y i , wi ).
i=1
x i y i ,w i
b
b
A1 x 1 + A2 x 2 w T1 x 1 w T2 x 2
2
2
T
T b
T
A1 x 1 w 1 x 1
= c1 x 1 + y
2
T
T b
T
A2 x 2 w 2 x 2 ,
+ c2 x 2 + y
2
= cT1 x 1 + cT2 x 2 + y T
so we can minimize over x 1 and x 2 separately given the dual variable y, to find
g(y) = g1 (y) + g2 (y) where g(y) is given as,
g(y) = min L(x 1 , x 2 , y, w 1 , w2 ) = min L1 (x 1 , y, w 1 ) + L2 (x 2 , y, w 2 ).
x 1 ,x 2
x 1 ,x 2
In order to find g1 (y) and g2 (y), respectively, we solve the following two subproblems:
b
b
g1 (y) = min cT1 x 1 + y T
A1 x 1 = min cT1 AT1 y x 1 + y T ,
x 1 0
x 1 0
2
2
g2 (y) = min
x 2 0
cT2 x 2
+y
b
b
A2 x 2 = min cT2 AT2 y x 2 + y T .
x 2 0
2
2
90
b
b
y =y +
A1 x 1 + A2 x 2 = y + (b Ax),
2
2
(18)
where is a step length that can be chosen in any of the standard ways [3].
min cTi x i
xi
b
i+1
(19)
Ai x i = + (1) t, i = 1, 2
x i 0,
and the solution of the following master problem:
min 1 + 2
,t
b
p
+ (1)i+1 t y i , p = 1, . . . , pk ,
2
b
q
i+1
+ (1) t y i , q = 1, . . . , qk .
i
2
(20)
The first and second sets of inequalities in (20) are respectively so-called feap
q
sibility cuts or optimality constraints, where y i and y i are the whole set optimal
dual variables and extreme rays of the dual problem of (19) computed up to iteration
k = pk + qk . Further details on the Benders decomposition may be found in [8].
91
(21)
When the vector is split into two vectors 1 and 2 , the domain of the problem
is also split into two parts with a common boundary that couples some of the two set
of variables. This means that the vectors 1 and 2 may be decomposed into two
vectors 1 = ( 1,1 , 1,2 ) and 2 = ( 2,1 , 2,2 ) such that the vectors 1,1 and 2,2
92
(22)
eq2,2 2 = 0.
eq2,1 1 + B
B
We note that the last equation in (22) is a complicating constraint. The equation
3 + F eq3 = 0 is separable and can be in turn decomposed as,
A
eq3,1 1 + F eq3,1 = 0,
A
(23)
eq3,2 2 + F eq3,2 = 0.
A
x 1 ,x 2 ,
i = 1, 2,
i = 1, 2,
(24)
i = 1, 2,
Beq2,1 x 1 + Beq2,2 x 2 = 0,
x 1 K1 ,
x 2 K2 ;
free.
t,x i ,
(25)
Aeq3,i x i + F eq3,i = 0,
Beq2,i x i = ( 1)i+1 t,
x i Ki ,
t, free (i = 1, 2)
Note that since the complicating constraint in optimisation problem (25) is built
through the common boundary, the coupling constraint can be interpreted as fictitious Newman condition for each sub-domain.
93
2
1
+
,
min
2
2
Aeq1,i x i + i F eq1,i = beq1.i ,
Aeq2,i x i = 0,
Aeq3,i x i + i F eq3,i = beq3,i ,
Beq2,i x i = ( 1)i+1 t i ,
t 1 = t 2,
1 = 2 .
Let Ci (i = 1, 2) be local constraints that are defined as follows:
eq1,i i
A
x + i F eq1,i = beq1,i ,
eq2,i
i
A
x
=
0,
i i eq3,i i
eq3,i
eq3,i
Ci = t i , x , A
x + i F
=b
,
eq2,i i
B
x
=
t
,
x i K , t , free.
i
i i
In the above problem, t 1 , 1 , t 2 , 2 are public variables and x 1 , x 2 are private variables. Let us collect all the public variables together into one variable
y = (t 1 , 1 , t 2 , 2 ) = (y 1 , y 2 ) where y 1 = (t 1 , 1 ), y 2 = (t 2 , 2 ). If we introduce
a vector z that gives the common values of the public variables, then we can express
the coupling constraints as y = Ez where E is a matrix with components equal to
zero or one, that is:
t
I 0
E1
z=
,
y i = Ei z where Ei =
, y = Ez where E =
.
0 1
E2
Let us set fi (x i , y i ) = f (x i , t i , i ) = 2i (i = 1, 2). Then our problem has the
followings form:
min f1 (x 1 , y 1 ) + f2 (x 2 , y 2 )
x,y,z
(x 1 , y 1 ) C1 ,
y i = Ei z,
(x 2 , y 2 ) C2 ,
(26)
i = 1, 2,
x = (x 1 , x 2 ),
y = (y 1 , y 2 ),
y i = (t i , i )
(i = 1, 2),
94
with variables x i , y i , and z. We then have a problem that is separable for fixed
values of z.
(x i , y i ) Ci
(i = 1, 2),
min
x 1 ,x 2 ,y 1 ,y 2
min f1 (x 1 , y 1 ) v T1 y 1 + f2 (x 2 , y 2 ) v T2 y 2 + v T Ez
z
(x 1 , y i ) Ci
(i = 1, 2),
then
q(v) = q(v 1 , v 2 ) =
min
x 1 ,x 2 ,y 1 ,y 2
f1 (x 1 , y 1 ) v T1 y 1 + f2 (x 2 , y 2 ) v T2 y 2
95
ET v = 0,
(x i , y i ) Ci
(i = 1, 2).
4 Numerical Results
4.1 Bearing Capacity of Anchors
The extensions described in Sect. 2 have been employed to test the pull out capacity
of multi-belled anchors. The linearity of the limit tension with respect to the number
of bells has been verified.
Five different anchor/soil conditions have been employed: rough (same properties as the soil), smooth (no resistance to shear), no tension condition, rough condition with no tension, and smooth condition with no-tension. Although the mechanisms do not significantly depend on these conditions, the pull out capacity does,
and has been shown to be much larger for rough conditions. On the other hand,
while for clay materials (zero internal friction angle, but non-zero cohesion) the
failure mechanism is localised around the anchor (see Fig. 9(a)), in other sand materials the slide-lines propagate up to the soil surface (see Fig. 9(b)). The computed
limiting height agrees satisfactorily with experimental results and other numerical
models that use incremental plasticity [15].
96
Fig. 9 Dissipation energy of double bell anchor. (a) Clay soil, with sliding conditions and no
tension at the anchor-soil interface. (b) Sand soil with internal friction angle = 20 , and rough
interface conditions with no tension
Fig. 10 Contour plot of the dissipation power on an anchored wall. Left: domain without the wall.
Right: zoom on the domain surrounding the anchor. Interface soil-anchor and soil-wall conditions
are rough with no tension
power of an anchored wall, with a zoom on the region surrounding the anchor. For
an anchor length equal to the height of the wall, the collapse mechanism surround
the whole wall-anchor system, while for longer anchors, the mechanism tends to be
localised around the anchor. Figure 11 shows the mesh obtained after 4 successive
iterations. As it can be observed, the smaller elements localise in the regions with
higher dissipation power and at the slidelines.
The tests have been run for different ratios of d/ h where h = free wall height,
and d = total height of the wall. The collapse of the wall was obtained for a certain
factor of the gravity acceleration. As the ratio h/d is increased, the value of
was decreasing. The limiting free height is the value for which = 1. The plots in
Figure 12 show the evolution of , and indicate the limiting ratios for different wall
conditions and admissibility conditions of the wall-soil and anchor-soil interfaces.
These values agree with some experimental values published in [19].
97
Fig. 12 (a) Upper and lower bound of the bearing capacity of a retaining wall with a rough interface with no traction. (b) Bounds for the anchored wall with different anchor-soil conditions.
Horizontal axis corresponds to the ratio d/ h, while the values in the vertical axis corresponds to
the factor multiplying the gravity acceleration at collapse
98
Fig. 13 (a) Domain and mesh employed in the decomposition analysis of the lower bound problem. The domain is subjected to a homogeneous Dirichlet conditions (zero velocities) at the left
boundary, and a horizontal tensile traction at the right boundary. (b) Evolution of the upper and
lower bound of the optimal solution of the LB limit analysis problem when using a variable step
size in the dual decomposition method
5 Conclusions
We have presented some recent extension of the lower and upper bound formulation
of limit analysis for problems with specific frictional interfaces, duplicated edges
and joints. Such extensions are motivated by the limit analysis of practical problems
with anchors and other structural elements.
We have also described how the optimisation problems that current computational limit analysis encounters may be decomposed for its eventual parallelisation.
Although only simple domains problems with limited number of elements have been
analysed, the methodology presented here is general and may be applied for more
general problems. This generalisation, and other techniques are currently under investigation.
References
1. MOSEK ApS (2005) The MOSEK optimization tools version 3.2 (revision 8). Users manual
and reference. http://www.mosek.com
2. Benders JF (1962) Partitioning procedures for solving mixed-variables programming problems. Numer Math 4:238252
3. Bertsekas DP (2003) Convex analysis and optimization, 3rd edn. Athena Scientific, Cambridge
4. Boyd S, Vandenberghe L (2004) Convex optimization. Cambridge University Press, Cambridge
5. Boyd S, Xiao L, Mutapcic A, Mattingley J (2007) Notes on decomposition methods. Technical
report, Stanford University. Notes of EE364B course
6. Christiansen E (1996) Limit analysis of collapse states. In: Handbook of numerical analysis,
vol IV. North Holland, Amsterdam, pp 193312
7. Ciria H, Peraire J, Bonet J (2008) Mesh adaptive computation of upper and lower bounds in
limit analysis. Int J Numer Methods Eng 75:899944
99
Abstract This paper concerns the development of a primal-dual algorithm for limit
and shakedown analysis of Reissner-Mindlin plates made of von Mises material.
At each optimization iteration, the lower bound of the shakedown load multiplier is
calculated simultaneously with the upper bound using the duality theory. An edgebased smoothed finite element method (ES-FEM) combined with the discrete shear
gap (DSG) technique is used to improve the accuracy of the solutions and to avoid
the transverse shear locking behaviour. The method not only possesses all inherent
features of convergence and accuracy from ES-FEM, but also ensures that the total
number of variables in the optimization problem is kept to a minimum compared
with the standard finite element formulation. Numerical examples are presented to
demonstrate the effectiveness of the present method.
1 Introduction
In practical engineering, the calculation of the load carrying capacity for structures
has been a problem of great interest to many designers. In the early 20th century, it
could be relatively easily obtained by limiting the stress intensity at a certain point
of the structure equal to the yield stress of the material. This implies that structural
failure occurs at initiating local plastic flow. However, many materials, for example
the majority of metals, exhibit distinct, plastic properties. Such materials can deform
considerably without breaking, even after the stress intensity attains the yield stress.
This implies that if the stress intensity reaches the critical (yield) value, the structure
does not necessarily fail or deform extensively. To this case, elastic-plastic structural
analysis becomes more general and is capable of providing the much needed safety
` M. Staat (B)
T.N. Trn
Institute for Bioengineering, Biomechanics Lab., Aachen University of Applied Sciences,
Heinrich-Mumann-Str. 1, Jlich Campus, 52428 Jlich, Germany
e-mail: [email protected]
`
T.N. Trn
Chair of Mechanics and Robotics, University of Duisburg-Essen, Lotharstr. 1, 47057 Duisburg,
Germany
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_5,
Springer Science+Business Media Dordrecht 2014
101
102
` and M. Staat
T.N. Trn
evaluation capability for enhanced diagnosis of aging infrastructure and the economical and safe design of new structures. Among the plasticity methods, Limit
and Shakedown Analysis (LISA) seems to be the most powerful one. In Europe
LISA have been developed as direct plasticity methods for the design and the safety
analysis of severely loaded engineering structures, such as nuclear power plants and
chemical plants, offshore structures, etc. [18]. Annex B of the new European pressure vessel standard EN 134453 is based on LISA (European standard, 2005-06),
thus indicating the industrial need for LISA software.
Limit analysis of plates and slabs has been studied analytically [8, 14] and numerically [3, 5, 10]. In this work, the limit load was studied either by the lower
bound approach which is based on the static theorem or by the upper bound approach which is based on the kinematic theorem. Tran et al. [19] developed a kinematic formulation for the shakedown analysis of thin plates and shells using the
exact Ilyushin yield surface and nonlinear programming. Recently, a primal-dual algorithm for shakedown analysis of Kirchhoff plates made of von Mises material has
been developed using DKQ plate elements [20]. It has been pointed out by the author that duality theory is indeed a very effective approach for shakedown analysis
of structures.
After more than half a century of development, the finite element method (FEM)
has become a very powerful technique for numerical simulations in engineering and
science. However, when using finite element software, one frequently encounters
problems with mesh generation. For accuracy reasons, one wants to use quadrilateral
or hexahedron elements, but such a mesh is quite difficult to generate and requires a
number of manual operations to cut the domain into proper pieces. In searching for
alternatives, Liu et al. have introduced the strain smoothing technique [6] into the
FEM formulating a cell/element-based smoothed finite element method (SFEM or
CS-FEM) [11] and a node-based smoothed finite element method (NS-FEM) [13].
It was shown that the SFEM achieves more accurate results and a higher convergence rate than FEM without increasing computational costs. More important, as
no mapping or coordinate transformation is involved in the SFEM, its elements are
allowed to be of arbitrary shape and the problem of mesh distortion can be avoided
when using SFEM.
CS-FEM is formulated using smoothing domains located inside the elements
and proven effectively in solving 2D solid mechanics problems by using a proper
number of smoothing cells in each element (for example four smoothing cells). The
CS-FEM has also been extended to plate and shell structures [15, 16]. NS-FEM uses
node-based smoothing domains associated from the predefined parts of all adjacent
elements around the node. It can provide upper bound solutions in the strain energy
and is also immune to volumetric locking naturally. However, the NS-FEM was
found temporally unstable, and cannot be applied directly to dynamic problems.
To overcome such a temporal instability, the edge-based smoothed finite element
method (ES-FEM) has been recently proposed for analyses of solid 2D mechanics
problems [12] and has then been extended for static, free vibration and buckling
analyses of Reissner-Mindlin plates using triangular elements [17]. In that method,
ES-FEM was incorporated with the discrete shear gap (DSG) technique [2] together
103
with a stabilization technique [1] in order to avoid the transverse shear locking and to
improve the accuracy of the numerical solution. With T3 elements ES-FEM has been
applied succesfully to probabilistic limit and shakedown analysis of 2D structures
[21].
This paper presents a numerical algorithm for shakedown analysis of ReissnerMindlin plates made of von Mises material. The algorithm is based upon a primaldual approach in which the lower bound of the shakedown load multiplier is calculated simultaneously with the upper one through a dual problem. The edge-based
smoothed finite element method (ES-FEM, [13]) and the DSG technique [2] were
used for the discretization of the problem domain. This technique has been recently
developed by Nguyen-Xuan et al. [17] in order to avoid the transverse shear locking
and to improve the accuracy of the numerical solution of Reissner-Mindlin plates.
Several benchmark plate problems in literature are investigated to test the performance of the method in which the thickness of the plate varies from very thick to
very thin (below the Kirchhoff limit). Numerical examples demonstrate that the proposed method not only exhibits excellent convergence and accuracy of solutions, but
also ensures that the total number of variables in the optimization problem is kept
to a minimum compared with the standard finite element formulation.
Mxx
1
1 Sxx
Myy ,
,
m=
s=
M0 M
S0 Syy
xy
(1)
1 xx
1 xz
yy
,
=
,
=
0 2
0 yz
xy
in which M0 = y t 2 /4, S0 = y t, 0 = y (1 2 )/E, 0 = 40 /t are normalized
quantities. Note that 0 is defined by assuming the relation M0 0 = S0 0 to hold.
The non-dimensional stress and strain resultant vectors are introduced as follows
T
T
= m s ,
= .
(2)
We denote the nodal variables of the plate element by q = [ w x y ]T and loading g = [ 0 0 p ]T acting on the plate where w is the transverse displacement; x , y
are the rotations around the y- and x-axes; and p is the transverse load. Using the
` and M. Staat
T.N. Trn
104
Fig. 1 Plate subjected to
transverse load
stress and strain resultant vectors introduced above, the equilibrium conditions of
plates can be written as follows
R g = 0,
R=
M0 x
M0 y
S0
M0 y
M0 x
S0 x
S0
.
S0 y
(3)
= Lq,
1
L=
0 x
0
1
0 y
1
0
1
0
1
0 x
10
1
0 y
0
.
1
0
(4)
The von Mises yield condition in the space of stress resultants takes the form
f ( ) =
T P 1 = 0,
(5)
in which the matrix P is composed of moment and shear force components Pb and
Ps as follows
0
,
Ps
Pb
P=
0
1
1/2 0
1
0,
Pb = 1/2
0
0
3
3 0
P =
.
0 3
s
(6)
The plastic dissipation function per unit area of the plate is defined as
) = M0 T Qb +
D (,
p
16 T s
T Q ,
=
M
Q
0 0
t2
(7)
where
Qb 0
,
Q=P =
0 Qs
4/3 2/3 0
1
Qb = Pb
= 2/3 4/3 0 ,
0
0 1/3
1
105
1/3
0
1
Qs = Ps
=
.
0 1/3
(8)
i =
Nn
B I (xi )qI ,
(10)
I =1
where Nn(i) is the total number of nodes of elements containing the common edge i.
(i)
(i)
I (xi ) is the
For inner edges (see Fig. 2) Nn = 4, for boundary edges Nn = 3. B
(i)
smoothed strain matrix on the domain which is calculated numerically by an
assembly process similar to the standard FEM
(i)
i) =
B(x
Nn
I =1
(i)
(i)
Ne
1
1
Aj Bj ,
B I (xi ) = (i)
3
A
(11)
j =1
in which Ne , Aj , Bj are the number of elements, the area and the strain matrix
of the j th element around the edge i, respectively. For inner edges (see Fig. 2)
` and M. Staat
T.N. Trn
106
Fig. 2 Division of the
domain into triangular
elements and smoothing cells
(k) connected to edge k of
triangular elements
(i)
(i)
and the global stiffness matrix is then assembled from all domain stiffness matrices
(i) by a similar process as in the FEM.
K
107
and yield function allow us to replace the time-dependent stress field (x, t) by its
values calculated at load vertices (x, Pk ). Let the fictitious elastic stress resultant
subjected to:
R(x)
=0
in ,
(14)
f ( E (x, Pk ) + (x))
0 k = 1, m; in .
The first constraint in (14) ensures that the equilibrium of the residual stress
field is fulfilled. The second constraint means that the actual stress resultant field
= E + does not anywhere violate the yield condition in (5). Using ES-FEM
to discretize the weak form of the equilibrium equations and the yield condition, the
maximization problem (14) can be reformulated as
= max
subjected to:
N
ed
A(i) B Ti i = 0,
(15)
i=1
i ) 0 i = 1, Ned , k = 1, m
f ( E
ik +
in which E
ik is the fictitious elastic stress resultant vector on the smoothing domain
i at load vertex Pk .
k=1
vertex, the kinematical condition may not be satisfied, however the accumulated
Acstrains over a load cycle must be kinematically compatible i.e. = Lq.
cording to the kinematic theorem of shakedown, the upper bound shakedown load
multiplier + may be formulated by the following form
+ = min
m
D p ( k )d
k=1
k = Lq in ,
subjected to:
k=1
m
M0 0 E (x, Pk )T k d = 1,
k=1
(16)
` and M. Staat
T.N. Trn
108
in which the second constraint is the normalization of the external power of loading.
By using ES-FEM and the plastic dissipation defined in (7), the discretized form of
the minimization problem (16) is rewritten as
+ = min
Ned
m
A(i) M0 0 Tik Q ik
k=1 i=1
m
subjected to:
ik = B i q i = 1, Ned ,
(17)
k=1
Ned
m
A(i) M0 0 Tik E
ik = 1.
k=1 i=1
Note that when m = 1, i.e. the loads are constant, the formulations (15) and (17)
reduce to those of limit analysis.
T
Q = Q1/2 Q1/2 .
(19)
Introducing these new notations into Eq. (17) gives (primal problem)
+ = min
Ned
m
M0 0 e Tik e ik
k=1 i=1
m
subjected to:
e ik B i q = 0 i = 1, Ned ,
k=1
(20)
Ned
m
e Tik tik 1 = 0.
k=1 i=1
Now, we can write the Lagrangian for the primal problem (20)
m N
m
m
Ned
ed
e ik B i q
k=1
k=1
k=1 i=1
109
with the Lagrange multipliers i , . The duality between the upper and lower
bounds is proven through the two following propositions.
Proposition 1 If there exists a finite solution + for the upper bound shakedown
load multiplier (20), then + has its dual form as
+ = max
i ,
Ned
T i = 0,
(a)
B
i
subjected to: i=1
i + tik M0 0 . (b)
(22)
Proof By using the Lagrangian introduced in (21), the dual problem of the primal
problem takes the form
0
1
(23)
max min L .
i , e ik ,q
Since a finite solution for the problem (20) exists, the two constraints in (20) are
affine, the objective function in (20) is convex, thus according to the strong duality
theorem, there exists no gap between primal and dual problems. It follows that
min
h(eik ,q)=0
Ned
m
k=1 i=1
0
1
M0 0 e Tik e ik = max min L ,
i , e ik ,q
(24)
T
Ned
Ned
m
M0 0 e ik
i tik e ik +
iT B i q + .
T
e ik e ik
i=1 k=1
i=1
(25)
M0 0 e ik
e ik 0 eik , (a)
i
ik
e Tik e ik
(26)
Ned
(b)
iT B i q = 0 q.
i=1
` and M. Staat
T.N. Trn
110
(27)
which of course violates the strong duality theorem. Taking the system (26) into
(25) leads to the conclusion
g( i , ) = min L = .
e ik ,q
(28)
T i = 0.
B
i
(29)
i=1
Furthermore, we can prove that the condition (26a) is equivalent to the following
inequality constraint
i + tik M0 0
i, k.
(30)
Firstly, if the inequality (30) holds, then the Cauchy-Schwartz inequality permits
us to write
(i + tik )T e ik i + tik eik M0 0 eik = M0 0 e Tik e ik eik , (31)
which is equivalent to the condition (26a). Secondly, if on the contrary we find a set
(i , ) satisfying
i + tik > M0 0 ,
(32)
we can always choose a strain rate vector e ik such that the angle between this vector
and the vector b = i + tik is equal to zero, or bT e ik = b e ik . It follows that
e T e ik
bT e ik = b e ik > M0 0 e ik = M0 0 e Tik e ik = M0 0 ik
,
e Tik e ik
or
M0 0 e ik
i tik
e Tik e ik
e ik < 0,
(33)
(34)
which violates the condition (26a). Equalities (28), (29) and inequality (30) conclude our proof.
Proposition 2 If there exists a finite solution + for the upper bound shakedown
load multiplier (20), then the static formulation (15) is exactly the dual problem of
111
min
Ned
m
h(eik ,q)=0
M0 0 e Tik e ik =
k=1 i=1
2 %Ned
max
(i) T =0
i
i=1 A Bi
i )0
f ( E
ik +
= .
(35)
ik
M0 0
41/2
T (Q1/2 )T i
Q1/2 Q1/2
+ E
ik
M0 0
T
41/2
3 1/2 T
1/2 T
(Q ) i
) i
E
E
1 (Q
= M0 0
+ ik Q
+ ik
. (36)
M0 0
M0 0
Using (5), (7), (8) and (36), the inequality (22b) now becomes
i 0,
f E
ik +
(37)
i by B
i from (18) and then (Q1/2 )T i by
with i = (Q1/2 )T i /M0 0 . Substitute B
M0 0 i in (22a), one has
Ned
i=1
B Ti i =
Ned
A(i) B Ti
Ned
1/2 T
Q
i = M0 0
A(i) B Ti i = 0
i=1
(38)
i=1
or
Ned
A(i) B Ti i = 0.
(39)
i=1
` and M. Staat
T.N. Trn
112
L
e ik
i tik = 0,
= M0 0
e ik
e Tik e ik
L
i = 0,
= iT B
q
m N
ed
L
T
e ik tik 1 = 0.
=
(40)
k=1 i=1
By using the Newtons method to solve the optimality conditions (40) and after
some manipulations, we get the following system of linear equations
Kd q = Kq + f1 + f2 ( + d),
(41)
where
K=
Ned
T E1 B i ,
B
i i
i=1
f1 =
Ned
T E1
B
i i
i=1
f2 =
Ned
i=1
T E1
B
i i
M1
ik ( i + tik )
k=1
m
e Tik e ik + 02
e Tik e ik + 02 tik ,
M1
ik
(42)
Mik = M0 0 I + ( i + tik )
Ei =
e Tik e ik
e Tik
e Tik e ik + 02
e Tik e ik + 02
M1
ik
and 02 is a small positive number, e.g. 1020 . The system (41) can be considered
as the linear system arising in purely elastic computations with the global stiffness
matrix K and the elastic matrix E1
i . At each optimization iteration, by solving this
system we will get the incremental vector of nodal variables d q and then the incre d e ik , d i and d are actually
mental vectors d e ik , d i and d. The vectors d q,
Newton directions which assure that a suitable step along them will lead to a decrease of the objective function of the primal problem (20) and to an increase of
the objective function of the dual problem (22). Iterating these steps a stable set of
e ik , i and is obtained satisfying all conditions in (20) and (22). A detailed
q,
explanation of the primal-dual algorithm can be found in Vu et al. [22].
113
Fig. 3 Geometrical dimensions (a) and FE-mesh (b) of rectangular plate with eccentric square
cutout
6 Validations
In this section, two numerical examples are presented to evaluate the performance
of the proposed method. Two plates that have different slenderness ratios, subjected
either to uniform out-of-plane pressure loading or concentrated load are considered.
In all examples, the structures are made of elastic-perfectly plastic material and 3node triangular element T3 is used to discretize the structural domains. For each test
case, some existing analytical and numerical solutions found in literature are briefly
represented and compared.
114
` and M. Staat
T.N. Trn
Fig. 4 Convergence of
plastic collapse and
shakedown load factors of a
rectangular plate
Fig. 5. In all cases, only one value is presented since the lower bound is identical with the upper one and the results are normalized with pab/M0 . It is seen that
the performance of the proposed method is effective and free of locking when the
slenderness ratio becomes very large.
115
Simply supported
Clamped
Kirchhoff, upper
bound, [3]
Reissner-Mindlin
Kirchhoff, upper
bound, [3]
Reissner-Mindlin
30
5.140
4.807
9.757
9.359
45
5.609
5.403
10.596
10.488
60
5.966
5.827
11.394
11.303
75
6.197
6.019
11.893
11.799
90
6.278
6.278
12.062
11.928
Silva [4] using the Kirchhoff plate assumptions. All edges of the plate are simply
supported or built-in and the uniform load p can be constant or varying within a
range p [0, pmax ].
We consider first thin rhombic plates with the slenderness ratio 2r/t = 20. Numerical limit analysis for the different skewness angles was carried out and the
results are collected in Table 1 and compared with the least upper bounds obtained
by Capsoni and Vicente da Silva [4]. It is seen that our numerical solutions are always smaller than those of Capsoni and Vicente da Silva (the maximum difference
is 6.5 % and 4.1 % for = 30 ) except for the case of simply supported edges with
= 90 , in which they are identical. This difference may be interpreted by the two
following observations: (1) The solutions obtained by Capsoni and Vicente da Silva
are pure upper bounds and (2) for skew plates ( = 90 ), the bending moments at
obtuse corners tend to zero for the Reissner-Mindlin plate while they tend to infinity
for the Kirchhoff plate as pointed out by Hggblad and Bathe [7].
Figure 7 depicts the convergence of the upper and lower bounds of the plastic
collapse and shakedown load multipliers, given in dimensionless form pr 2 /M0 , for
the case of simply supported plate with = 60 .
116
` and M. Staat
T.N. Trn
Fig. 7 Convergence of
plastic collapse and
shakedown load factors of
rhombic plate
Choosing the rhombic plate with the skewness angle = 60 , we study now the
performance of the present method for several plate slenderness ratios. The graph
in Fig. 8 shows the normalized plastic collapse and shakedown loads calculated for
different boundary conditions. We may also see that the results are nearly unchanged
even if the thickness of the plate becomes very small (below the Kirchhoff limit).
References
1. Bischoff M, Bletzinger KU (2001) Stabilized DSG plate and shell elements. In: Wall WA,
Bletzinger K-U, Schweizerhof K (eds) Trends in computational structural mechanics. CIMNE,
Barcelona, pp 253263
117
2. Bletzinger KU, Bischoff M, Ramm E (2000) A unified approach for shear-locking free triangular and rectangular shell finite elements. Comput Struct 75:321334
3. Capsoni A, Corradi L (1999) Limit analysis of platesa finite element formulation. Struct
Eng Mech 8:325341
4. Capsoni A, Vicente da Silva M (2011) A finite element formulation of Mindlin plates for limit
analysis. Int J Numer Methods Biomed Eng 27:143156
5. Cecchi A, Milani G, Tralli A (2007) A Reissner-Mindlin limit analysis model for out-of-plane
loaded running bond masonry walls. Int J Solids Struct 44:14381460
6. Chen JS, Wu CT, Yoon S, You Y (2001) A stabilized conforming nodal integration for
Galerkin mesh-free methods. Int J Numer Methods Eng 50:435466
7. Hggblad B, Bathe KJ (1990) Specifications of boundary conditions for Reissner/Mindlin
plate bending finite elements. Int J Numer Methods Eng 30:9811011
8. Hopkins HG, Wang AJ (1954) Load-carrying capacities for circular plates of perfectly-plastic
material with arbitrary yield condition. J Mech Phys Solids 3:117129
9. Le VC (2009) Novel numerical procedures for limit analysis of structures. PhD thesis, University of Sheffield, UK
10. Le VC, Gilbert M, Askes H (2010) Limit analysis of plates and slabs using a meshless equilibrium formulation. Int J Numer Methods Eng 83:17391758
11. Liu GR, Dai KY, Nguyen TT (2007) A smoothed finite element method for mechanics problem. Comput Mech 39:859877
12. Liu GR, Nguyen-Thoi T, Lam KY (2009) An edge-based smoothed finite element method
(ES-FEM) for static, free and forced vibration analyses of solids. J Sound Vib 320:11001130
13. Liu GR, Nguyen-Thoi T, Nguyen-Xuan H, Lam KY (2009) A node-based smoothed finite
element method (NS-FEM) for upper bound solutions to solid mechanics problems. Comput
Struct 87:1426
14. Mansfield EH (2000) Collapse pressures for rhombic plates. Int J Mech Sci 42:635643
15. Nguyen-Thanh N, Rabczuk T, Nguyen-Xuan H, Bordas S (2008) A smoothed finite element
method for shell analysis. Comput Methods Appl Mech Eng 198:165177
16. Nguyen-Xuan H, Rabczuk T, Bordas S, Debongnie JF (2008) A smoothed finite element
method for plate analysis. Comput Methods Appl Mech Eng 197:11841203
17. Nguyen-Xuan H, Liu GR, Thai-Hoang C, Nguyen-Thoi T (2010) An edge-based smoothed
finite element method (ES-FEM) with stabilized discrete shear gap technique for analysis of
Reissner-Mindlin plates. Comput Methods Appl Mech Eng 199:471489
18. Staat M, Heitzer M (eds) (2003) Numerical methods for limit and shakedown analysis. Deterministic and probabilistic approach. NIC series, vol 15. John von Neumann Institute for
Computing, Jlich. http://hdl.handle.net/2128/2926
19. Tran TN, Kreiig R, Vu DK, Staat M (2008) Upper bound limit and shakedown analysis of
shells using the exact Ilyushin yield surface. Comput Struct 86:16831695
20. Tran TN (2011) A dual algorithm for shakedown analysis of plate bending. Int J Numer Methods Eng 86:862875
` TN, Staat M (2013) An edge-based smoothed finite element method for primal-dual
21. Trn
shakedown analysis of structures under uncertainty. In: de Saxc G, Oueslati A, Charkaluk E,
Tritsch J-B (eds) Limit states of materials and structures: direct methods, vol 2. Springer,
Dordrecht, pp 89102
22. Vu DK, Yan YM, Nguyen-Dang H (2004) A primal-dual algorithm for shakedown analysis of
structure. Comput Methods Appl Mech Eng 193:46634674
Abstract In this paper, the lower-bound of direct methods is applied to fiber reinforced metal matrix periodic composites. Three boundary conditions for the localization problem are discussed and the influence of hardening matrix material is
studied. Furthermore, in combination with homogenization theory, plastic material
parameters are predicted by using yield loci fitting on the macroscopic limit stress
domain. The proposed approach is validated through a numerical example of unidirectional periodic composites with square fiber patterns.
1 Introduction
To predict safe service conditions of structures or structural elements made of heterogeneous materials under variable loads beyond elasticity is a challenging task
in civil and mechanical engineering. There are two major difficulties: to consider
variable loads with unknown evolution in time and to determine the effective material properties. The former difficulty can be overcome by direct methods (DM),
namely limit and shakedown analysis. Limit analysis only requires the load range,
and shakedown analysis needs the envelope of the independent loads [13]. Therefore, the application of DM to composites arose many interests these years. To solve
the latter difficulty, multi-scale modeling method and homogenization theory are involved [27]. Either using the lower-bound [14, 23, 33, 34] or upper-bound [3, 21, 22]
approach, periodic composites are investigated at the representative volume element
(RVE) level with the elastic perfectly plastic material properties of each phase.
DM has been formulated for structures assuming elastic perfectly plastic material
behavior. However, work-hardening occurs most notably for ductile materials, like
metals. Thus DM for plasticity models with hardening has also been investigated for
long time [15, 16, 18, 20, 32], where the studied objects are homogeneous global
119
120
structures [2, 19, 25]. In this work, we applied the lower-bound approach of DM to
periodic composites with the consideration of work-hardening.
As the global effective material properties are concerned, in [7], homogenized
elastic parameters are obtained basing on the homogenization theory and the constitutive laws of elastic materials. The plastic properties are only studied for the
composites under plane stress case by using yield loci fitting. Here, the predictions
for unidirectional periodic composites are discussed.
The numerical tools required for the lower-bound approach are the finite-element
method and non-linear optimization. To reduce the scale of optimization problem,
non-conforming three-dimensional finite elements are used to discretize the RVE,
and the interior-point-algorithm based optimization tool (IPOPT) [29, 31] together
with the pre-programming language AMPL [11] are used.
is a small scale parameter, which determines the size of the representative volume
element (RVE). It plays an important role in studying the heterogeneous material,
especially for non-uniform structures. For a heterogeneous material with periodic
distribution, the smallest possible unit is normally defined as the RVE. The macroscopic strain E and stress are linked to mesoscopic strain and stress by:
5
6
1
E(x) =
( ) dV = ( ) ,
(2)
V V
5
6
1
( ) dV = ( ) .
(3)
(x) =
V V
Here, stands for the averaging operator. In DM for periodic heterogeneous materials, the macroscopic stress is decomposed as [34]:
E
1
1
1
1
+ dV =
dV with
dV = 0
E dV +
=
V V
V V
V V
V V
(4)
where is the safety factor, E is the purely elastic stress field and is the timeindependent residual stress field.
121
on V .
(5)
on V .
(6)
anti-periodic.
(7)
The local strain can be split into an overall strain E and a fluctuating field per ,
where the average of per over RVE vanishes,
(u) = E + uper = E + per ,
(8)
6
per = 0.
(9)
In the numerical implementation, to carry out the strain approach, uniform displacement is imposed on the boundary, see Fig. 2(L). For Stress approach, besides
the uniform stress imposed on the boundary, one degree of freedom of the boundary
is coupled in order to maintain the periodic deformation, as shown in Fig. 2(M).
122
These two approaches are normally suitable for the periodic RVE, while the third
one is more realistic, especially for non-periodic unit cell. Figure 2(R) describes the
constraints for periodicity method:
ui ui + udi = 0
(10)
here, ui and ui are the displacements of the relative opposite periodic node pairs and
udi is the macroscopic displacement. Figure 3 illustrates the deformations of a nonperiodic RVE under pure thermal loading by using strain approach and periodicity
approach, respectively.
The common ground in these three approaches is the consistent deformation of
boundaries. The strain energies, by using different boundary conditions, are ordered
in the following way [27]:
hom : E.
E : d hom : E E : d hom
per : E E : d
(11)
hom are the 4th order tensors of elastic stiffness, under uniform
d hom , d hom
per and d
stress, periodicity and uniform strain, respectively, which depend on the micro variable .
The objective studied in this work is the periodic composites under mechanical
loading, and the strain approach was adopted as the boundary condition. The local-
123
in V ,
div = 0
E
per
= d : (E + ) in V ,
Pstrain = E n
anti-periodic on V ,
uper
periodic
on V ,
= E.
(12)
The residual stress field should satisfy the self-equilibrium condition and periodicity conditions:
div = 0 in V ,
res
Pstrain =
(13)
n
anti-periodic on V .
Anti-periodicity means that either E n or n has opposite values on opposite
sides of V . Periodicity of uper indicates that the displacements at two opposite
points of the boundary are the same. For strain approach, we assume that uper = 0
and per = 0.
Here, is the virtual strain, and u is the virtual displacement. p and f are
surface force and body force, respectively. For periodic heterogeneous materials,
the external loads can be either macroscopic stresses or macroscopic strains E.
The left side of Eq. (14) implies that the discretization here has to be carried out
Since the scale
for the purely elastic stress field E and the residual stress field .
of the optimization problem is mainly determined by the type and the number of
finite elements, the choice of a proper element type is very important. In this work,
a non-conforming solid element was applied in the limit and shakedown analysis of
composites because of its accuracy and efficiency [5].
The lower-bound problem of DM for periodic composites with elastic perfectly
plastic material model can be formulated finally as the following mathematical programming:
= 0,
[C]{}
max
(15)
F (iE (Pk ) + i , Y i ) 0, i [1, NGS].
124
is the load factor, Pk is the vertices of load envelope and N GS is the number of
Gaussian points and Y is the yield strength. F is the von Mises yield criterion.
= 0,
[C]{}
i , Yf i ) 0, i [1, NGSF],
F ( E
(16)
max
i (Pk ) +
E
m
F ( j (Pk ) + j , Uj ) 0, j [1, NGSM]
125
Meaning
NE
Number of elements
NK
Number of nodes
NGE
NGSF
NGSM
NGS
NL
Number of load vertices: NL = 1, limit analysis; NL = 2n , shakedown analysis. n is the number of independent loads
= 0,
[C]{}
i , Yf i ) 0, i [1, NGSF]
F ( E
max
(17)
i (Pk ) +
126
where is the back stress field. It is important to note that, for matrix with unlimited kinematic hardening, there is no optimal solution for limit analysis, since the
ultimate strength of the material is infinite.
= 0,
[C]{}
i , Yf i ) 0, i [1, NGSF],
F ( E
i (Pk ) +
m ) 0, j [1, NGSM],
j j , Yj
F ( E
max
(18)
j (Pk ) +
E
m
F ( j (Pk ) + j , Uj ) 0, (a)
or F ( , m m ) 0. (b)
j
Uj
Yj
As shown in Fig. 7, Eq. (18a) indicates that the subsequent yield surfaces stays
always inside the bounded loading surface [32]. Equation (18b) means that motion
of the origin center of the subsequent yield surface is bounded by the back stress
surface [26]. The mathematical equality of both conditions is proved in [19]. Nevertheless, the obtained optimized value of back stresses under the two conditions are
slightly different [6].
127
No. Var
No. Eq
No. Ineq.
Elastic-perfectly plastic
6NGS + 1
3NK + 9NE
NL NGS
Isotropic hardening
6NGS + 1
3NK + 9NE
NL NGS
6NGS + 6NGSM + 1
3NK + 9NE
NL NGS
6NGS + 6NGSM + 1
3NK + 9NE
NL (NGS + NGSM)
6NGS + 6NGSM + 1
3NK + 9NE
NL NGS + NGSM
With the assumption that the material model of fiber is elastic, the material model
of matrix is elastic-perfectly, isotropic hardening, unlimited kinematic hardening
and limited kinematic hardening, respectively, the sizes of static shakedown problem
are shown in Table 2.
(19)
128
In our former work, the yield criterion of periodic composites under plane stress
case are discussed [7]. Firstly, three states during the failure are defined [8, 12]:
Onset of plasticity (Homogenized elastic stress): EL = EL E .
Shakedown state (Homogenized shakedown stress): SD = SD E .
Limit state (Homogenized limit stress): LM = LM E .
E is the homogenized macroscopic stress of purely elastic stress field. Based
on limit homogenized macroscopic stress domain, there are two possibilities to derive the yield criteria:
To find the best fitted mathematical formulation;
To identify the related parameters by using existing yield criteria.
The former approach seems quite difficult because of the uncertainty of the number
of parameters. Therefore, the latter is adopted to seek a feasible solution. Since
the studied object in the numerical example is the unidirectional fiber reinforced
periodic composites, whose global material behavior can be treated as an orthotropic
one, Hills yield criterion is hypothesized to fit the limit domain.
However, the stress components, either in micro-/mesoscropic level ij or in
macroscopic level ij depend on the orientation of the coordinate system. Nevertheless, there are certain invariants associated with every tensor. The three principle
stresses are calculated through the characteristic equation:
3 I1 2 + I2 + I3 = 0,
(20)
I1 , I2 and I3 are the first, second, and third stress invariants, respectively.
(21)
1 1
1
1
,
+
2 Y 2 Z2 X2
1 1
1
1
,
G=
+
2 Z2 X2 Y 2
1 1
1
1
.
H=
+
2 X2 Y 2 Z2
(22)
with:
F=
129
(23)
Tx1 y1 z1
cos
= sin
0
sin
cos
0
0
cos
0 sin
1
0
sin
cos
0
0
cos
0 sin
1
0
sin
cos
0
0
0 . (24)
1
130
(26)
If the von Mises criterion is projected to plane, Eq. (25) can be written as:
12 + 22 = C
with C =
2
1
= Y2 .
3F
3
(27)
The projection of the von Mises yield criterion on (1 , 2 )-plane and -plane is
shown in Fig. 9, and the ellipse from Hills criterion becomes a circle.
1 2
1 1
1
F=
;
G=H =
.
2 Y 2 X2
2 X2
(28)
(29)
131
1
Y=
.
F +H
(30)
(31)
Here, (xc , yc ) is the center of the ellipse, is the angle between the X-axis and the
major axis of the ellipse, parameter t varies from 0 to 2 , a and b are major and
minor radii, respectively.
In the case of transversely homogeneous material, xc = yc = 0 and = 4 . Substitute the known value into Eq. (31), we get:
x y 2
= 1 =
2b
1
1
1
1
1
1
2
2
x
y
xy = 1.
+
+
+
+
2a 2 2b2
2a 2 2b2
a 2 b2
x+y
2a
1
1
F + 2H = 2 + 2 ,
2a
2b
=
1
1
H F =
2a 2 2b2
H = 2 ,
3a
1
1
F =
2.
2
2b
6a
(32)
(33)
(34)
In our case, c4 = c5 = 0 and c6 = 1. Comparing Eqs. (25) and (34), we get the
following three equations:
F + 1.866H + 0.134G = c1 ,
F + 0.134H + 1.866G = c2 ,
(35)
G 2F + H = c3 .
In conclusion, the methodology to determine the yield criterion of unidirectional
fiber reinforced periodic metal matrix composites is:
132
1.
2.
3.
4.
5.
5 Numerical Examples
Take the square patterned unidirectional fiber reinforced periodic metal matrix composites as an example, the RVE is shown in Fig. 10, with perfect interface. The fiber
ratio is 40 %. Because of symmetry, the quarter of the RVE is used for the finite
element analysis (see Fig. 11), with the dimension lx = ly = 50 mm. The RVE is
subjected to two independent uniform displacement loadings, U 1 = U 2 = U 0 =
0.02 mm.
The material properties of each phase are shown in Table 3, with the assumption
that each phase is isotropic.
8-node non-conforming elements are applied for the calculation of purely elastic
stress field E and the self-equilibrated constant matrix [C]. Since the unidirectional fiber reinforced composites can be treated as plane strain case, all degrees of
freedoms in the fiber direction are fixed, i.e. there is no displacement deformation
in the fiber direction. AMPL+IPOPT are used as the optimization tool.
Besides the limit load factor LM and the shakedown load factor SD , the elastic
load factor EL and the alternating plastic load factor AP are also calculated for
the comparison in this work. The elastic load factor EL under two independent
loads L1 and L2 is defined as:
max EL F iE (Pk ), Y i 0, i [1, NGS], k = 1
(36)
where the load vertex P1 indicates the combination of L1 and L2 .
The alternating plasticity load factor AP under two independent loads L1 and
L2 can be simplified as follows:
1
E
F ( 2 [ E
i (L1 ) + i (L2 )], Y i ) 0,
max AP
(37)
E
F ( 12 [ E
i (L1 ) i (L2 )], Y i ) 0, i [1, NGS].
Figure 12 shows the different load domains of the considered example, assuming
that both phases have elastic perfectly plastic material behavior.
Considering the hardening, the shakedown domains for different matrix material models are shown in Fig. 13. We observe that the shakedown domain with
isotropic hardening is only enlarged compared to the elastic perfectly plastic model.
The shakedown domain with unlimited kinematic hardening is bounded by the alternating plasticity load domain.
If the RVE is under loading U 1 = U 2, the shakedown load factors of elastic
perfectly plastic, limited kinematic hardening and unlimited kinematic hardening
133
E (GPa)
Y (MPa)
U (MPa)
Matrix(Al)
70
0.3
80
120
Fiber(Al2 O3 )
370
0.3
2000
Fig. 12 Load domains of periodic composites with the fiber volume fraction 40 % under plane
strain case, with EL: elastic load factor; SD: shakedown load factor; LM: limit load factor and AP:
alternating plasticity load factor. (L) Displacement domain; (R) Macroscopic stress domain
have the same value, as shown Fig. 13. However, the mechanisms are obviously
different. For example, under loading U 1 = U 2 = SD U 0, the equivalent stress
fields obtained under different material models are shown in Fig. 14.
134
Although based on the same elastic stress field E , the residual stress fields
and back stress fields , obtained from the optimization programming, are quite
different. For the elastic perfectly plastic material model, obviously there is no back
stress field. Nevertheless, the total equivalent stress fields of these three different
material models are similar.
Figure 15 shows us merely the limit domains with elastic perfectly plastic and
limited kinematic hardening material models, since there is no bound for limit load
of matrix with unlimited kinematic hardening.
According to the limit displacement domain of the elastic perfectly plastic material model, with the aid of homogenization approach and stress invariant theory, the
macroscopic principle stresses domain is obtained, as shown in Fig. 16(L), which is
projected into -plane afterwards, see Fig. 16(R). Based on the least square fitting
method, the obtained parameters of Hills criterion are as follow:
Major axis of ellipse: a = 241.8306;
From Eqs. (33) and (30), we get the axial strength of the unidirectional periodic
composites:
X = 296.1808 MPa = 3.70Ym ;
(38)
Indices f and m represent fiber and matrix, respectively. c is the critical strain
defined by:
3 m f4
(39)
c = min Y , Y .
Em Ef
135
Fig. 14 Different stress fields under different material models: (L) Elastic perfectly plastic material model; (M) Limited kinematic hardening material model; (R) Unlimited kinematic hardening
material model
1 m
(mt rm Em )
k (1 + m )(1 2m )
(40)
136
Fig. 16 Yield criterion fitting: (L) Homogenized stresses (1 -2 ) domain; (R) Hills yield criterion fitting based on the projection into plane
where, index t means the transverse direction. rm is the radial maximum residual
strain, which is approximated to zero in our case. k is the stress concentration
factor, with the definition:
max
,
(41)
k =
p
p is the outer force that is applied on the RVE or mesoscopic components. According to the numerical result, k is around 1.15. Therefore, the effective transverse
yield strength based on micromechanics is:
YY = YZ = 93.65 MPa = 1.17Ym .
When compared with the analytical results from microscopic mechanics, we observe that the predicted strength in transverse direction matches well. However, the
137
numerical value of the strength in fiber direction is bigger than the analytical one.
The possible reason lies in the too restrict constraints in the numerical analysis: the
degrees of freedom in fiber direction are completely fixed which leads to a greater
homogenized elastic stress in fiber direction. Nevertheless, the advantages of numerical methods are obvious, which consist in the possibility to consider the fiber
distributions, imperfect bounded interfaces, or other types of composites, instead of
the unidirectional one.
6 Conclusions
In this paper, the lower-bound approach of DM is applied on periodic composites,
including hardening material models for the matrix. As expected, for isotropic hardening, the shakedown domain is enlarged with the same shape as elastic perfectly
plastic one. For unlimited kinematic hardening, there is no bound for the limit load
and shakedown domain is bounded by alternating plasticity. Two yield surfaces of
limited kinematic hardening model provide more realistic solutions. Furthermore, in
combination with homogenization theory, plastic material parameters are predicted
by using yield surface fitting on the macroscopic limit homogenized stress domain.
However, the present work is based on the assumption that fiber and matrix have
perfect interfaces. The debonding failure will be studied in the further work.
Acknowledgements
References
1. Kassem GA (2009) Micromechanical material models for polymer composites through advanced numerical simulation techniques. PhD thesis, Germany: RWTH-Aachen
2. Bouby C, de Saxc G, Tritsch JB (2009) Shakedown analysis: comparison between models
with linear unlimited, linear limited and non-linear kinematic hardening. Mech Res Commun
36:556562
3. Carvelli V, Maier G, Taliercio A (1999) Shakedown analysis of periodic heterogeneous materials by a kinematic approach. J Mech Eng 50:229240
4. Chen M (2011) Shakedown and optimization analysis of composite materials. PhD thesis,
China: Southeast University
5. Chen M, Hachemi A, Weichert D (2010) Non-conforming element for limit analysis of periodic composites. Proc Appl Math Mech 10:405406
6. Chen M, Hachemi A, Weichert D (2012) Shakedown analysis of periodic composites with
kinematic hardening. Proc Appl Math Mech 12:271272
7. Chen M, Hachemi A, Weichert D (2012) Shakedown and optimization analysis of periodic
composites. In: Saxc G, Oueslati A, Charkaluk E, Tritsch J-B (eds) Limit state of materials
and structures, direct methods 2. Springer, Berlin, pp 4569
8. Chen M, Zhang LL, Weichert D, Tang WC (2009) Shakedown and limit analysis of periodic
composites. Proc Appl Math Mech 9:415416
9. Conn AR, Gould NIM, Toint PL (1992) LANCELOTa Fortran package for large-scale nonlinear optimization. Springer, Berlin
138
10. Daniel IM, Ishai O (2006) Engineering mechanics of composite materials, 2nd edn. Oxford
University Press, New York
11. Fourer R, Gay DM, Kernighan BW (2003) In: AMPL: a modeling language for mathematical
programming, 2nd edn. Duxbury, N. Scituate
12. Hachemi A, Chen M, Weichert D (2009) Plastic design of composites by direct methods.
In: Simos TE, Psihoyios G, Tsitouras Ch (eds) Numerical analysis and applied mathematics,
international conference, pp 319323
13. Knig JA (1987) Shakedown of elastic-plastic structures. Elsevier, Amsterdam
14. Magoariec H, Bourgeois S, Dbordes O (2004) Elastic plastic shakedown of 3d periodic heterogenous media: a direct numerical approach. Int J Plast 20:16551675
15. Maier G (1973) Shakedown matrix theory allowing for work hardening and second-order
geometric effects. In: Sawczuk A (ed) Foundations of plasticity, pp 417433
16. Melan E (1938) Zur Plastizitt des rumlichen Kontinuums. Ing-Arch 8:116126
17. Mouthamid S (2008) Anwendung direkter Methoden zur industriellen Berechnung von Grenzlasten mechanischer Komponenten. PhD thesis, RWTH-Aachen, Germany
18. Pham DC (2005) Shakedown static and kinematic theorems for elastic-plastic limited linear
kinematic-hardening solids. Eur J Mech A, Solids 24:3545
19. Pham PT (2011) Upper bound limit and shakedown analysis of elastic-plastic bounded linearly
kinematic hardening structures. PhD thesis, RWTH-Aachen, Germany
20. Ponter ARS (1975) A general shakedown theorem for elastic plastic bodies with work hardening. In: Proc SMIRT-3, p L5/2
21. Ponter ARS, Leckie FA (1998) Bounding properties of metal-matrix composites subjected to
cyclic thermal loading. J Mech Phys Solids 46:697717
22. Ponter ARS, Leckie FA (1998) On the behaviour of metal matrix composites subjected to
cyclic thermal loading. J Mech Phys Solids 46:21832199
23. Schwabe F (2000) Einspieluntersuchungen von Verbundwerkstoffen mit periodischer
Mikrostruktur. PhD thesis, RWTH-Aachen, Germany
24. Simon JW, Weichert D (2011) Numerical lower bound shakedown analysis of engineering
structures. Comput Methods Appl Mech Eng 200:28282839
25. Simon WJ (2011) Numerische Einspielunguntersuchungen mechanischer Komponenten mit
begrenzt kinematisch verfestigendem Materialverhalten. PhD thesis, RWTH-Aachen, Germany
26. Stein E, Zhang G, Knig JA (1992) Shakedown with nonlinear hardening including structural
computation using finite element method. Int J Plast 8:131
27. Suquet P (1982) Plasticit et homognisation. PhD thesis, Universit Pierre et Marie Curie,
France
28. Talreja R, Singh CV (2012) Damage and failure of composite materials. Cambridge University
Press, Cambridge
29. Wchter A (2002) An Interior Point Algorithm for Large-Scale Nonlinear Optimization with
Applications in Process Engineering. PhD thesis, Chemical Engineering, Carnegie Mellon
University, Pennsylvania:
30. Wchter A (2009) Short tutorial: getting started with ipopt in 90 minutes. Technical report,
IBM Research Report
31. Wchter A, Biegler LT (2006) On the implementation of a primal-dual interior-point filter
line-search algorithm for large-scale nonlinear programming. Math Program 106:2557
32. Weichert D, Gross-Weege J (1988) The numerical assessment of elastic-plastic sheets under
variable mechanical and thermal loads using a simplified tow-surface yield condition. Int J
Mech Sci 30:757767
33. Weichert D, Hachemi A, Schwabe F (1999) Application of shakedown analysis to the plastic
design of composites. Arch Appl Mech 69:623633
34. Weichert D, Hachemi A, Schwabe F (1999) Shakedown analysis of composites. Mech Res
Commun 26:309318
1 Introduction
Nowadays structures are continuously designed to withstand repeated thermomechanical loading that forces them to enter the plastic regime. Such loading conditions are encountered either in civil or mechanical engineering. Typical examples
of such structures are buildings and bridges under seismic loading on the one hand
and nuclear reactors and aircraft gas propulsion engines on the other.
K.V. Spiliopoulos (B) K.D. Panagiotou
Department of Civil Engineering, Institute of Structural Analysis & Antiseismic Research,
National Technical University of Athens, Zografou Campus, Zografos 157-80, Athens, Greece
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_7,
Springer Science+Business Media Dordrecht 2014
139
140
The life cycle assessment of such a structure constitutes an important task for a
structural engineer. However, the long term response of a structure, subjected to a
given thermo-mechanical loading which exhibits inelastic time independent plastic
strains, is quite complex, because of the need to perform lengthy and expensive incremental calculations, especially for structures with a high degree of redundancy.
In the case that the long term response turns out to be a stabilized state there are procedures called direct methods, which may lead directly to these states. Scleronomic
or rheonomic stable materials guarantee the existence of such states [1]. Thus direct
methods search for this asymptotic state right from the start of the calculations.
The most well known cyclic state is the elastic shakedown. The search for this
state is based on the lower [2] and upper bound [3] shakedown theorems of plasticity.
The formulation of these problems is normally done using mathematical programming (MP). One may refer to various such procedures like a nonlinear Newton-type
algorithm [4] or the interior point methods (e.g. [5, 6]).
There are also very few approaches that are not based on MP. Internal parameters are introduced in [7] which characterize local inelastic mechanisms. Another
procedure is the Linear Matching Method (LMM) [8] which is a generalization of
the elastic compensation method [9] and is based on matching a linear problem to a
plasticity problem. A sequence of linear solutions, with spatially varying moduli, is
generated that provide upper bounds that monotonically converge to the least upper
bound.
The method was further extended beyond shakedown, for loadings that can be
decomposed into constant and time varying components, so as to provide an upper
bound estimation of the ratchet boundary [10].
Besides the knowledge of safety margins, it is important to be able to determine
the long-term effects on a structure for a given cyclic loading. For this purpose, an
alternative to the cumbersome incremental procedure is a method called Direct Cycle Analysis (DCA) originally suggested in [11] and implemented in the commercial
program Abaqus [12]. The main assumption of the method is that the displacements
at the steady cycle will become cyclic. The method is quite involved and appears
to be a mixture of an incremental and an iterative procedure. The displacements
are decomposed into Fourier series whose coefficients are evaluated in an iterative
way by linking them with the coefficients of the Fourier series of the out-of-balance
load vector. This vector is evaluated as in an incremental procedure, and static admissibility is enforced by leading it to zero. The procedure seems to be suited for
the cases of alternating plasticity but fails to converge for loadings that are close to
ratcheting as also mentioned in [12] since, because of its main assumption, it cant
predict such a case.
A new direct method to predict any long-term cyclic state of an elastic-perfectly
structure under a given cyclic loading was quite recently suggested [13]. The method
focuses on the cyclic nature of the residual stresses at the steady state. The method
has been called Residual Stress Decomposition Method (RSDM) and is based on
decomposing the residual stresses in Fourier series inside a cycle of loading. The
decomposition of the residual stresses, so as to find a simplified way to predict
creep cyclic steady stress states, was originally proposed in [14].
141
In the RSDM the coefficients of the Fourier series are evaluated in an iterative
way by integrating the residual stress rates over the cycle. By satisfying equilibrium
and compatibility at time points inside the cycle one may evaluate these rates. Plastic
effects are accounted for by adding the elastic and the residual stress at the cycle
points. If the sum exceeds the yield surface, the plastic strain rate may be represented
by the stress in excess of the yield surface. This excess stress provides then input for
iteration. If the plastic strain rates stabilize, in the form of a converged residual stress
vector, the procedure stops. One can easily distinguish any of the three different
cases, shakedown, alternating plasticity or ratcheting. A one-dimensional truss and
a two-dimensional plate with a hole under plane stress have demonstrated in [13] the
application of the procedure. For the plate, results are, in the present work, extended
to include plane strain conditions. The whole approach is shown to be stable and
computationally efficient, with uniform convergence.
2 Cyclic Steady-States
Let us consider a body of volume V and surface S. On one part of S we have zero
displacement conditions and on the other part of the surface a cyclic loading of the
form (1) is applied.
P(t) = P(t + nT )
(1)
where P(t) is the set of loads that act on S; t is the time point inside the cycle, T is
the period of the cycle, n = 1, 2, . . . , denotes the number of full cycles. Bold letters
are used, herein, to denote vectors and matrices.
Let us suppose that our structure is made of an elastic-perfectly plastic material.
At any time point = t/T inside the cycle the structure will develop a stress field
( ) which may be decomposed into an elastic part el ( ), that equilibrates the external loading P( ) assuming a completely elastic behavior, and a self-equilibrating
residual stress part ( ) that is due to inelasticity. Therefore:
( ) = el ( ) + ( ).
(2)
(3)
The residual strain rate itself may be decomposed into an elastic and a plastic
part [15]. Thus the final compatibility equation is expressed as:
pl ( ).
= e el ( ) + el
r ( ) +
(4)
The elastic strain rates are related to the stress rates through the elasticity matrix D, whereas the plastic strain rate vector through the gradient of the flow rule:
el ( ) = D e ,
) = D el
(
r ,
f
pl =
(5)
142
1
( ) = a0 +
cos(2k ) ak + sin(2k ) bk .
2
(6)
k=1
Differentiating the above with respect to one may write the following expression for the derivative:
) = 2
k sin(2k ) ak + k cos(2k ) bk .
(7)
(
k=1
Making use of (7) and the orthogonality properties of the trigonometric functions
one may get expressions for the Fourier coefficients of the cosine and sine series in
terms of the residual stress derivatives:
1
1
)d,
ak =
sin(2k ) (
k 0
(8)
1
1
)d.
bk =
cos(2k ) (
k 0
A more involved formula proves to be needed for the constant term, which uses
the information at the beginning and at the end of the cycle [13]:
1
1
1
)d
a0,e =
a0,b +
(
(9)
ak,b
ak,e +
2
2
0
k=1
k=1
where the subscripts b and e denote the beginning and the end of the cycle respectively. As seen from Eqs. (8) and (9), because of (7), there is an implicit dependence
of the Fourier coefficients and thus an iterative scheme may be used to estimate
them, once the residual stress derivatives are calculated.
143
(10)
(11)
Since the strain rates are kinematically admissible, the residual stress rates are
self-equilibrated, and fixed supports have been assumed, one may write, for a virtual
strain field , using the Principle of Virtual Work (PVW):
T dV
= 0.
(12)
V
T
T
el
B D BdV r =
B dV +
BT D pl dV
V
or equivalently:
+
K r = R
(13)
BT D pl dV
(14)
according to the closest point projection [17], along the vector CB, with the plastic
strain rate pl directed along BC. We use, instead, the vector CA which is p . This
vector is a radial return type of mapping along the known OC. It may be easily
determined, especially for a von Mises yield surface. It is an equivalent measure for
the plastic straining in the sense that they either both exist or not.
144
1 ()
()
()
() ( ) = a0 +
cos(2k ) ak + sin(2k ) bk .
2
(15)
k=1
(16)
()
() ( ) Y
()
( ).
()
p ( ) =
() ( )
(17)
( ), which
4. Assemble for the whole structure the rate vector of the nodal forces R
is the r.h.s. of Eqs. (13)(14):
)+
BT ()
(18)
R ( ) = R(
p ( )dV .
V
(19)
(20)
1
a0
a
(+1)
()
=
() ( ) d.
ak
+ 0 +
ak +
2
2
0
k=1
k=1
145
9. From the updated Fourier coefficients evaluate the updated distribution of the
residual stresses, at all the Gauss points, using (15), and check the convergence
through their norms at the end of the cycle:
(+1) (1)2 () (1)2
tol
(+1) (1)2
where tol is a specified tolerance.
(22)
If (22) holds, the procedure stops as we have reached a cyclic stress state (cs),
and () = (+1) = cs ; otherwise we go back to step 1 and repeat the process.
()
(+1)
,
Once a cyclic stress state has been attained, we look at cs
p = p = p
which was evaluated during the last iteration. We may determine the nature of the
obtained solution, for each GP, by evaluating the following integral over the cycle:
1
cs
i =
p,i
( )d
(23)
0
4 Application Examples
The method is applied to two structures one being a one dimensional and the other a
two dimensional plate element with a hole under plane stress or plane strain conditions. A value of 104 for the tolerance proved quite accurate to stop the iterations.
146
Element
Areas (cm2 )
1.806
30.825
24.940
34.583
27.908
F0 = 400 kN.
The procedure predicts that the structure will shakedown. The constant in time
steady state residual stress may be seen in Fig. 3(a). In Fig. 3(b) one may also see
the distribution of the total stress, for bar 1, inside the cycle, where nowhere the
yield stress is exceeded. Analogous behavior is observed, of course, for all the other
bars.
147
(b) The second cyclic loading case has the following variation with time:
Fc (t) = 200 sin(2t/T ),
F0 = 200 kN.
For this loading case the RSDM predicts that the structure is going to suffer
from alternating plasticity. In Fig. 4 one may see the uniform convergence of the
procedure towards the final steady state.
The distribution of the cyclic residual stress predicted for the middle bar 1 inside
the steady cycle may be seen in Fig. 5. The procedure shows that in the steady
state the middle bar suffers plastic strain rates, of alternating nature. These strains
spread within the time intervals [0.169, 0.362] and [0.638, 0.851], inside the cycle,
rendering the total plastic strain over the cycle (parameter 1 expression (23), also
equal to the total area under the curve, Fig. 6) equal to zero.
The results for both the two loading cases agree well with those in [18].
148
at the edges of the plate. Due to the symmetry of the structure and the loading, only
one quarter of the plate is considered.
The boundary conditions as well as its finite element mesh discretization are
shown in Fig. 7. The ratio between the diameter d of the hole and the length L of
the plate is equal to 0.2. Also the ratio of the depth of the plate and its length is
equal to 0.05. Ninety-eight, eight-noded, isoparametric elements with 3 3 Gauss
integration points were used.
The material data used was: Youngs modulus E = 0.21 105 kN/cm2 , Poissons ratio = 0.3 and yield stress y = 36 kN/cm2 .
Both plane stress and plain strain conditions have been examined as the procedure 3.1 may be applied to both of them.
149
Px (t) = 0.
The predicted by the procedure behavior for the structure is a shakedown state
and this complies with the fact that this loading is below the shakedown boundary
estimated in [19]. In Fig. 8 the computed by the RSDM steady-state residual stress
distribution is plotted for the GP 2, for both plane stress and plane strain conditions.
150
This residual stress distribution is unique and will be the same with the one that
would be predicted from an incremental step-by-step analysis, e.g. [12] (see also
examples in [13]). The total stress distribution for this point is plotted in Fig. 9. In
Fig. 10 one may also see the convergence towards the final steady states.
(b) The second cyclic loading case has the following variation with time:
Py (t) = 0.7223y sin2 (t/T ),
Px (t) = 0.
The value of this load, at many cycle points, is in excess of the shakedown limit
computed by using a plane stress modeling, and below the shakedown limit assuming a plane strain condition [19]. The present numerical procedure (RSDM) also
shows that this loading will lead the plate to shakedown for plane strain, but assuming plane stress conditions the loading leads some GPs to reverse plasticity. So,
for the plane strain case, in Figs. 11, 12 one may see the computed by the RSDM
steady-state residual stress distribution for the GP 2 and its effective total stress
distribution, respectively.
On the other hand, for plane stress modelling, we plot, for the most strained
point GP 2, the variation of the yy component of the excess stress vector cs
p , which
has the biggest values from the three components (Fig. 13). We may see that plastic straining occurs, alternately, inside the time intervals [0, 0.04], [0.45, 0.55] and
[0.96, 1] at the steady cycle.
151
(c) The third cyclic loading case involves two loads, one constant in time and one
varying with time:
Py (t) = 0.5y sin2 (t/T ),
Px (t) = 0.93y .
This loading, at many cycle points, is above the ratcheting boundary. The results
for GP 1, assuming plane strain conditions, may be seen in Fig. 14, where plastic
straining of the same positive sign inside the cycle intervals [0, 0.11] and [0.89, 1]
at the steady cycle is observed. On the other hand, with a plane stress modeling,
one may observe that plastic strains of the same positive sign appear during the
whole cycle (Fig. 15). For both cases the xx direction of the component of the excess
152
stress vector cs
p , which has the biggest values from the three components, is plotted.
This ratcheting behavior is observed also for quite a few GPs around the structure,
which definitely constitutes incremental collapse mechanisms for both plane strain
and plane stress conditions that may be seen in Figs. 16, 17 respectively. We may
observe that we have a much more spreading of ratcheting for the case of plane
stress than for the case of plane strain.
In Fig. 18 one may see the convergence of the RSDM for this loading case, for
both plane stress and plain strain conditions.
Reviewing the examples considered herein, we note that, within the adopted tolerance, the number of iterations ranged from a minimum of 80 for the case of reverse
plasticity of the truss example, to a maximum number of 740 for the case of ratcheting of the plate example, under plane strain. The total CPU-time required to solve
this last example was just 260 s, using an Intel Core i7 at 2.93 GHz with 4096 MB
RAM.
The number of time points inside the cycle should be enough so that it may adequately represent the applied loading. Fifty time points inside the cycle were used
for all the examples considered herein. Three terms of the Fourier series were found
enough to represent the residual stress decomposition. The RSDM procedure proved
to be quite stable, no matter which asymptotic behavior was reached. Another important fact of the computational efficiency of the approach is that the stiffness matrix needs to be decomposed once and for all at the start of the calculations.
Fig. 17 Ratcheting
mechanismRSDM (plane
stress condition)
153
154
5 Concluding Remarks
The Residual Stress Decomposition Method (RSDM) is a direct method that proves
to be a simple and efficient procedure to estimate the long-term effects of the cyclic
loading on a structure. For a given time history of this loading it may equally predict
any possible steady state either it is elastic shakedown or alternating plasticity or
ratcheting.
The method, although currently developed for elastic-perfectly plastic material
and the von Mises yield surface, has the potential of extension to other types of
behavior and yield surfaces.
It also appears to have the potential to provide safety margins for a cyclic loading
the exact history of which is not known, but only its variation ranges, and work is
being done towards this direction.
References
1. Drucker DC (1959) A definition of stable inelastic material. J Appl Mech 26:101106
2. Melan E (1938) Zur plastizitt des rumlichen Kontinuums. Ing-Arch 9:116126
3. Koiter W (1960) In: Sneddon IN, Hill R (eds) General theorems for elastic-plastic solids.
North-Holland, Amsterdam
4. Zouain N, Borges L, Silveira JL (2002) An algorithm for shakedown analysis with nonlinear
yield function. Comput Methods Appl Mech Eng 191:24632481
5. Simon J-W, Weichert D (2011) Numerical lower bound shakedown analysis of engineering
structures. Comput Methods Appl Mech Eng 200:28282839
6. Vu DK, Yan AM, Nguyen-Dang H (2004) A primal-dual algorithm for shakedown analysis of
structures. Comput Methods Appl Mech Eng 193:46634674
7. Zarka J, Frelat J, Inglebert G, Kasmai-Navidi P (1990) A new approach to inelastic analysis
of structures. Nijhoff, Dordrecht
8. Ponter ARS, Carter KF (1997) Shakedown state simulation techniques based on linear elastic
solutions. Comput Methods Appl Mech Eng 140:259279
9. Mackenzie D, Boyle T (1993) A method of estimating limit loads by iterative elastic analysis,
I: simple examples. Int J Press Vessels Piping 53:7795
10. Ponter ARS, Chen H (2001) A minimum theorem for cyclic load in excess of shakedown, with
application to the evaluation of a ratchet limit. Eur J Mech A, Solids 20:539553
11. Maitournam MH, Pommier B, Thomas J-J (2002) Dtermination de la rponse asymptotique
dune structure anlastique sous chargement thermomcanique cyclique. C R, Mc 330:703
708
12. Abaqus 6.10, theory & users manual, Dassault systemes (2010)
13. Spiliopoulos KV, Panagiotou KD (2012) A direct method to predict cyclic steady states of
elastoplastic structures. Comput Methods Appl Mech Eng 223224:186198
14. Spiliopoulos KV (2000) Simplified methods for the steady state inelastic analysis of cyclically
loaded structures. In: Weichert D, Maier G (eds) Inelastic analysis of structures under variable
loads. Kluwer Academic, Dordrecht, pp 213232
15. Gokhfeld DA, Cherniavsky OF (1980) Limit analysis of structures at thermal cycling. Sijthoff
& Noordhoff, Rockwille
155
16. Frederick CO, Armstrong PJ (1966) Convergent internal stresses and steady cyclic states of
stress. J Strain Anal 1:154169
17. Simo JC, Hughes TJR (1998) Computational inelasticity. Springer, Berlin
18. Palizzolo L (2004) Optimal design of trusses according to a plastic shakedown criterion.
J Appl Mech 71:240246
19. Chen HF, Ponter ARS (2001) Shakedown and limit analyses for 3-D structures using linear
matching method. Int J Press Vessels Piping 78:443451
Abstract Direct methods can be used to rapidly verify the safety of solid bodies
and structures against collapse, and also to assist engineers wishing to rapidly identify structurally efficient designs for a specified load carrying capacity. Layout optimization is a direct method that can be used to solve very large-scale problems when
adaptive solution schemes are employed, and the same underlying mathematical formulation is applicable to both analysis and design problems. Here the truss layout
optimization formulation is applied to various benchmark design problems and the
discontinuity layout optimization formulation applied to various plane strain limit
analysis problems. It is observed that highly accurate solutions can be obtained,
close to known analytical solutions. Finally future directions in the field of layout
optimization are briefly considered.
1 Introduction
To verify the safety of solid bodies and structures against collapse, engineers have
traditionally had to rely either on simplistic hand type calculations, or on significantly more complex computational tools which identify the collapse state in an
indirect, iterative, manner. Such computational tools can be unreliable and often
require large amounts of computer time and/or high levels of operator expertise.
Additionally, in many engineering disciplines the initial design stage is carried out
in an ad-hoc manner, with engineering intuition often used to identify structurally
M. Gilbert (B) C.C. Smith S.J. Hawksbee A. Tyas
Department of Civil and Structural Engineering, University of Sheffield, Mappin Street,
Sheffield S1 3JD, UK
e-mail: [email protected]
C.C. Smith
e-mail: [email protected]
S.J. Hawksbee
e-mail: [email protected]
A. Tyas
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_8,
Springer Science+Business Media Dordrecht 2014
157
158
M. Gilbert et al.
efficient designs. Direct analysis and design methods can potentially address both
these issues, and similarities between analysis and design formulations can be exploited. Thus for example, a general integrated analysis and design framework has
been recently developed at the University of Sheffield, allowing a range of potential
applications to be explored. Highly efficient interior point linear programming (LP)
solvers are used in conjunction with an adaptive solution scheme to allow large-scale
problems to be solved.
A focus on both analysis and design problems also permits similarities in the underlying mathematical formulations to be studied in more detail. This led the authors
to realize that the truss layout optimization design technique, developed almost half
a century ago [5], could be modified so as to provide a powerful new direct analysis method for plasticity problems: discontinuity layout optimization [27]. In this
chapter, both these direct methods will be described, with a simple adaptive solution
scheme used to allow large-scale problems to be treated. The research has led to industrial applications (e.g. [6]), and the highly accurate solutions that can be obtained
has allowed new analytical solutions to be identified for old problems.
159
4. Sort list to identify largest violations, and use these to refine the reduced problem.
5. Repeat from step 2 until the violation list is empty.
The violations listed in step 3 may be violations of a yield surface or perhaps
violation of a virtual strain limit in the case of a truss layout optimization problem.
(1)
N q g,
T
where B is a (3n 3m) equilibrium matrix and q and f are respectively 3m and 3n
vectors of forces at discontinuities and block loads. Also qT = {N1 , M1 , S1 , N2 , M2 ,
S2 , . . . , Nm , Sm , Mm } and where Ni , Mi and Si are respectively the normal force,
shear force and moment acting at discontinuity i (i = 1 . . . m), f = fD + fL , where
fD and fL are respectively vectors of dead and live loads acting on blocks. N and g
depend on the adopted yield criteria, which for an individual discontinuity i can be
expressed as follows:
NTi qi gi
(2)
160
M. Gilbert et al.
or, in expanded form for a joint where friction and/or rocking failure may occur as:
1 tan i 0
0
1 tan i 0 Si
0
Ni
(3)
0
0
li /2
1
Mi
0
0
li /2 1
where li represents the length of each discontinuity, which is assumed to have angle
of friction i . The LP variables are therefore the discontinuity forces Si , Ni and Mi
and the live load factor .
161
Fig. 2 Adaptive solution example: hinged failure mechanism of three span masonry viaduct
solution scheme. In this case three iterations were required to obtain a solution of
the required accuracy, and the peak number of yield constraints required was less
than 500. In contrast had 24 fixed planes been used to circumscribe the yield surface
the number of yield constraints involved would have exceeded 3000, and the accuracy of the solution would have been lower. (The analysis was undertaken using the
LimitState:RING software [18].)
162
M. Gilbert et al.
Fig. 4 Stages in truss layout optimization procedure: (a) starting problem (point load applied
remote from support); (b) discretization of design domain using nodes; (c) interconnection of nodes
with potential truss bars; (d) identification of optimal subset of potential bars using optimization
(giving the optimal truss layout)
163
(4)
q 0,
},
where V is the total volume of the structure, qT = {q1+ , q1 , q2+ , q2 , . . . , qm
+
and qi , qi are the tensile and compressive internal forces in bar i (i = 1 . . . m);
cT = {l1 /1 , l1 /1 , l2 /2 , l2 /2 , . . . , lm /m }, where li and i are respectively the
length and yield stress of bar i. B is a suitable (2n 2m) equilibrium matrix and
y
y
y
y
fT = {f1x , f1 , f2x , f2 , . . . , fn } where fjx and fj are the x and y components of
the external load applied to node j (j = 1 . . . n). The presence of supports at nodes
can be accounted for by omitting the relevant terms from f, together with the corresponding rows from B. This problem is in a form which can be solved using LP,
with the bar forces in q being the LP variables.
An equivalent kinematic formulation can be derived using duality principles:
max V = fT u
(5)
subject to:
BT u c
uT
y
y
y
= {ux1 , u1 , ux2 , u2 , . . . , un },
(6)
uxj
y
uj
and where
and
are the x and y compowhere
nents of the virtual displacements of node j . In this case the nodal displacements in
u are the LP variables.
164
M. Gilbert et al.
Fig. 5 Simple truss layout optimization problem (after [7]): (a) design domain [1 unit 2 units],
loading and support conditions; (b) minimal connectivity ground structure containing 11 bars;
(c) optimal solution for ground structure [b], volume = 3.36603; (d) full connectivity ground
structure containing 15 bars, including overlapping bars; (e) optimal solution for ground structure
[d], volume = 2.63397 (limiting compressive and tensile strengths taken as unity)
of approx. 100 million LP variables when using the formulation given in Eq. (4),
which would be difficult to solve directly using a standard desktop PC.
Thus referring to the basic adaptive algorithm described in Sect. 2.1, it is desirable to commence with an initial problem which has limited initial connectivity
(Step 1), and to then use violation of the dual constraint, Eq. (6), in Step 3 as a
means of producing a list of additional bars likely to improve the solution (Step 4).
The procedure continues (Step 4Step 2) until the violation list is empty. The final
solution is provably equal in terms of volume to that which would have been obtained had the corresponding fully connected ground structure problem been solved
instead; further details are provided in [7].
165
Fig. 6 Point load between pin/roller supports example: optimal layouts at each iteration in the
adaptive solution procedure (after [8])
when an initially fully connected ground structure is used but the computational effort required is considerably less. Additionally, when adaptivity was used, the peak
number of LP variables present was 5766, compared with 245,520 when an initially
fully connected ground structure was employed.
166
M. Gilbert et al.
Fig. 7 Point load between fixed pins example: 180 180 nodal division solution (after [8])
over 526 million potential members. The reported optimal volume of 2.57176 [8] is
just 0.037 % higher than the known exact value of 1 + /2. The solution is shown
in Fig. 7; this problem took approximately 15 hours to solve when using hardware
available in 2003 [8], and is still computationally intractable today when using a
fully connected ground structure.
167
Fig. 8 Uniform load between fixed pins example: (a) numerical solution for + = case;
(b) exact analytical solution for + = 0.417785 case (limiting ratio for vertical hangers) (after
[23])
168
M. Gilbert et al.
Fig. 9 Uniform transmissible load between fixed pins example: analytical solution [30]
(7)
169
subject to:
Bd = 0,
(8)
Np d = 0,
(9)
fTL d = 1,
(10)
p0
(11)
where is the unknown load factor at collapse, fD and fL are vectors containing
respectively specified dead and live loads acting at discontinuities, d contains displacements along the discontinuities, p is a vector of plastic multipliers and g contains the corresponding dissipation coefficients. B is a suitable compatibility matrix
containing direction cosines and N is a suitable flow matrix.
Though this formulation has similarities with the corresponding truss optimization formulation, it also includes additional terms that are not directly analogous.
These include live and dead loads in Eq. (7), a flow rule in Eq. (9) that permits modelling of frictional problems, and a term to exclude trivial zero solutions, Eq. (10).
For a typical translational cohesive-frictional plane-strain plastic analysis problem: dT = {dT1 , dT2 , . . . , dTm } = {s1 , n1 , s2 , n2 , . . . , nm }, where si and ni are the relative shear and normal displacements at discontinuity i; gT = {c1 l1 , c1 l1 , c2 l2 , . . . ,
cm lm }, where li and ci are respectively the length and cohesive shear strength of
discontinuity i.
In the same way that force equilibrium is enforced at nodes for the truss optimization problem, kinematic compatibility is enforced at the nodes in DLO as
illustrated in Figs. 10 and 11, where i and i are respectively x-axis and y-axis
direction cosines for truss bar or discontinuity i. Thus the contribution of a given
discontinuity i to the global compatibility constraint Eq. (8) can be written as:
i i
i
i
si
Bi di =
(12)
i i ni .
i i
Similarly, when the Mohr-Coulomb failure criteria is used the local plastic flow
constraint for discontinuity i can be written as:
1
pi
1
1
s
Ni pi di =
(13)
i =0
2
n
tan i tan i
i
pi
where Ni is a local plastic flow matrix, pi is a vector containing plastic multipliers
pi1 , pi2 , where pi1 , pi2 0, and where i is the angle of friction of the material.
The discontinuity displacements in d and the plastic multipliers in p are the variables in the optimization problem, which can be solved using LP when the Tresca
or Mohr-Coulomb failure criteria are used.
Stages in the DLO procedure are outlined diagrammatically in Fig. 12. In the
kinematic formulation compatibility at nodes is explicitly enforced. However, when
discontinuities crossover one another at non-nodal points, it can be shown that compatibility is implicitly enforced.
170
M. Gilbert et al.
Fig. 10 Equilibrium at
(unloaded) node
% in a truss
(see Eq. (4)). 5i=1 i qi = 0,
%5
i=1 i qi = 0
Fig. 11 Compatibility at a
node in a sliding mechanism
(translation only, see Eq. (8)).
%5
i si = 0,
%i=1
5
i=1 i si = 0
171
Fig. 12 Stages in DLO procedure: (a) starting problem (surcharge applied to block of soil close
to a vertical cut); (b) discretization of soil using nodes; (c) interconnection of nodes with potential
discontinuities; (d) identification of critical subset of potential discontinuities using optimization
(giving the layout of slip-lines in the critical failure mechanism) (after [10])
(15)
BT t + fL q = fD ,
(16)
NT q g
(17)
subject to:
y
y
y
{t1x , t1 , t2x , t2 , . . . , tn }
y
tj
where tT =
and where tjx and can be interpreted as x and
y direction equivalent nodal forces acting at node j (j = 1 . . . n), corresponding in
y
a work sense to uxj and uj respectively, and where q is here a vector of shear and
normal forces acting on discontinuities, i.e. qT = {S1 , N1 , S2 , N2 , . . . , Nm }, where
Si and Ni represent respectively the shear and normal force acting on discontinuity
y
i (i = 1 . . . m). The LP variables are therefore tjx , tj , Si , Ni and the live load factor
. The objective is thus to maximize whilst ensuring that the yield condition is not
violated along any potential discontinuity.
The required equilibrium constraint can alternatively be written for a potential
discontinuity i interconnecting nodes A and B as follows:
BTi ti + fLi qi = fDi
(18)
172
M. Gilbert et al.
i
i
i
i
i
i
i
i
tAx
y
tA
fs
fs
+ Li Si = Di .
tx
n
n
Ni
fLi
fDi
B
y
tB
(19)
The required yield constraint can also be written for a potential discontinuity i as
follows:
NTi qi gi
or, in expanded form for the Mohr-Coulomb yield condition as:
1 tan i
Si
c i li
1 tan i
Ni
c i li
(20)
(21)
y
s
s
fLi
Si
fDi
tA
i i i i
+
=
+
(22)
x
n
n
i i i i
fLi
fDi
N i
tB
y
tB
where Si and N i are identical to Si and Ni respectively, except that they are not
LP variables. Using the newly computed values of Si and N i , the yield constraint
Eq. (21) can be checked for violation for the potential discontinuity. When using an
adaptive solution scheme, this checking process can be repeated at each iteration for
all potential discontinuities, with any violating discontinuities becoming candidates
for admission to the expanded LP problem at the next iteration (i.e. in step 3 of the
adaptive solution scheme presented in Sect. 2.1).
173
174
M. Gilbert et al.
175
Punch position
Nodal divisions
DLO (q/k)
Diff (%)
Central
45 15
5.17344
0.425
90 30
5.15408
0.243
180 60
5.14719
0.109
30 15
5.17344
0.425
60 30
5.15408
0.243
120 60
5.14719
0.109
180 180a
5.14352a
0.037a
Edge
a solution
176
M. Gilbert et al.
177
relatively recently. This means that there are still many potentially fruitful avenues
to explore. For example, the problems described in this chapter all assume the presence of straight-line connections (e.g. truss bars or slip-line discontinuities) between
nodes. However, different kinds of connections can potentially be used. Thus in twodimensions curved connections could instead be used, perhaps to model truss bars
optimally shaped to carry their own self-weight, or to model the curved slip-lines
associated with rotational collapse mechanisms. Another possibility is to consider
surface connections, which may represent shell elements in a design context, or potential surface discontinuities in a three-dimensional limit analysis context. Initial
investigations into the latter have already begun and this will therefore be considered further in the next section.
178
M. Gilbert et al.
Fig. 18 Deformed mechanism, obtained using DLO, for rough based square punch (showing a
quarter of the square punch)
constraint, sir
2 + s 2 ; where s is the resultant shear change in velocity
si1
ir
i2
6 Conclusions
Layout optimization techniques were developed almost half a century ago, but are
now the subject of renewed interest. When suitable adaptive solution schemes are
employed, layout optimization techniques can be applied to large-scale problems.
This means that very accurate solutions can be obtained relatively rapidly, potentially allowing new-insights into the optimal forms of structures to be obtained.
The inherent similarities between limit analysis and limit design problem formulations also means that methods originally developed for the design of trusses can
be adapted to identify the critical layout of discontinuities in solids (discontinuity
layout optimization, DLO). DLO is a powerful and flexible technique which is now
finding use in industry, clearly demonstrating its practical usefulness.
179
References
1. Andersen KD, Christiansen E, Overton ML (1998) Computing limit loads by minimizing a
sum of norms. SIAM J Sci Comput 19(3):10461062
2. Bazaraa MS, Jarvis JJ, Sherali HD (2009) Linear programming and network flows, 4th edn.
Wiley, New York
3. Chakrabarty J (2006) Theory of plasticity, 3rd edn. Butterworth-Heinemann, Oxford
4. Darwich W, Gilbert M, Tyas A (2010) Optimum structure to carry a uniform load between
pinned supports. Struct Multidiscip Optim 42(1):3342
5. Dorn WS, Gomory RE, Greenberg HJ (1964) Automatic design of optimal structures. J Mech
3:2552
6. Gilbert M, Smith CC (2010) Soil structure revolution. Struct Eng 88(3):1822
7. Gilbert M, Tyas A (2003) Layout optimisation of large-scale pin-jointed frames. Eng Comput
20(8):10441064
8. Gilbert M, Tyas A, Pritchard T (2003) An efficient linear programming based layout optimization method for pin-jointed frames. In: Proceedings of the 5th world congress on structural and
multidisciplinary optimization, Venice. CD-ROM proceedings paper no 67
9. Gilbert M, Casapulla C, Ahmed HM (2006) Limit analysis of masonry block structures with
non-associative frictional joints using linear programming. Comput Struct 84(13):873887
10. Gilbert M, Smith CC, Haslam IW, Pritchard TJ (2010) Application of discontinuity layout
optimization to geotechnical limit analysis problems. In: Proceedings of the 7th European
conference on numerical methods in geotechnical engineering, Trondheim, Norway, pp 169
174
11. Gilbert M, Smith CC, Pritchard TJ (2010) Masonry arch analysis using discontinuity layout
optimisation. Proc Inst Civ Eng, Eng Comput Mech 163(3):155166
12. Hawksbee SJ, Smith CC, Gilbert M (2013) Application of discontinuity layout optimization
to three-dimensional plasticity problems. Proc R Soc A 469(2155):20130009
13. Hemp WS (1974) Michell framework for uniform load between fixed supports. Eng Optim
1(1):6169
14. Heyman J (1966) The stone skeleton. Int J Solids Struct 2(2):249279
15. Heyman J (1972) Coulombs memoir on statics. Cambridge University Press, Cambridge
16. Johnson W, Mellor PB, Woo DM (1958) Extrusion through single hole staggered and unequal
dies. J Mech Phys Solids 6:203222
17. Lewinski T, Zhou M, Rozvany GIN (1994) Extended exact solutions for least-weight truss
layouts, part I: cantilever with a horizontal axis of symmetry. Int J Mech Sci 36:375398
18. LimitState (2011) In: LimitState:RING masonry arch bridge analysis software. http://
www.limitstate.com/ring
19. LimitState (2012) In: LimitState:GEO geotechnical analysis software. http://www.limitstate.
com/geo
20. Livesley RK (1978) Limit analysis of structures formed from rigid blocks. Int J Numer Methods Eng 12(12):18531871
21. Makrodimopoulos A, Martin CM (2006) Lower bound limit analysis of cohesive-frictional
materials using second-order cone programming. Int J Numer Methods Eng 66(4):604634
22. Michell AGM (1904) The limits of economy of material in frame-structures. Philos Mag
8(47):589597
180
M. Gilbert et al.
23. Pichugin AV, Tyas A, Gilbert M (2012) On the optimality of Hemps arch with vertical hangers. Struct Multidiscip Optim 46(1):1725
24. Shiau JS, Lyamin AV, Sloan SW (2003) Bearing capacity of a sand layer on clay by finite
element limit analysis. Can Geotech J 40:900915
25. Vicente da Silva M, Anto AN (2007) A non-linear programming method approach for upper
bound limit analysis. Int J Numer Methods Eng 72:11921218
26. Smith CC, Cubrinovskii M (2011) Pseudo-static limit analysis by discontinuity layout optimization: application to seismic analysis of retaining walls. Soil Dyn Earthq Eng 31(10):1311
1323
27. Smith C, Gilbert M (2007) Application of discontinuity layout optimization to plane plasticity
problems. Proc R Soc A 463(2086):24612484
28. Sok T, Lewinski T (2010) On the solution of the three forces problem and its application in
optimal designing of a class of symmetric plane frameworks of least weight. Struct Multidiscip
Optim 42(6):835853
29. Tyas A, Gilbert M, Pritchard TJ (2006) Practical plastic layout optimization of trusses incorporating stability considerations. Comput Struct 84(3):115126
30. Tyas A, Pichugin A, Gilbert M (2011) Optimum structure to carry a uniform load between
pinned supports: exact analytical solution. Proc R Soc A 467(2128):11011120
31. Wright SJ (1987) Primal-dual interior-point methods, vol 54. SIAM, Philadelphia
Abstract The aim of this work is to propose a macroscopic plastic model for
Porous nonassociated Drucker-Prager-type materials, using homogenization techniques and the hollow sphere model proposed by Gurson (J Eng Mater Technol
99:215, 1977) for von Mises solid matrix. In the first part, we determine analytically the plastic limit state of a hollow sphere with a Drucker-Prager matrix and subjected to hydrostatic loading. For the associated case, the collapse is complete with
a unique regime. For the nonassociated cases, we consider weaker solutions (partial
collapse and regime change). Nevertheless, we show that the collapse is complete
and exhibits a single regime. Consequently, the collapse stress field and the limit
load do not depend on the value of the dilation angle, as confirmed by numerical
simulations. This result has been already obtained by Maghous et al. (Eur J Mech A,
Solids 28:179188, 2009) by means of a modified second moduli approach. In Gursons footsteps, Guo et al. (J Mech Phys Solids 56:21882212, 2008) proposed a
macroscopic model for porous solid with pressure-sensitive dilatant matrix obeying
to the normality law (associated material). The second part of the paper is a first
attempt to extend Guos model to the nonassociated materials. Using the concept of
bipotential, we proposed a two-fields variational approach to deduce a macroscopic
model.
181
182
L. Cheng et al.
1 Introduction
The present work must be considered as a first attempt to propose a macroscopic
plastic model for Porous nonassociated Drucker-Prager-type materials, using homogenization techniques. In a famous paper [35], Gurson derived an upper bound
limit analysis approach of a hollow sphere and a hollow cylinder having a von Mises
solid matrix. Several extensions of Gursons model have been further proposed in
the literature, the probably most important developments being those accounting
for void shape effects [32, 33, 54]. Other extensions are concerned by the plastic anisotropy [2, 55] or take into account the plastic compressibility of the matrix
through associated Drucker-Prager model in the perspective of applications to polymer and cohesive geomaterials [1, 41, 42, 52].
In the first part (Sect. 2), we consider the hollow sphere subjected to the particular case of hydrostatic loading. This part is a direct extension of a recent paper [60]
dedicated to the same problem but here with the nonassociated flow rule. Classical bound limit analysis theorems have been generalized to the class of implicit
standard materials, i.e. with a nonassociated flow rule represented with a bipotential [9, 24]. In the spirit of Gursons paper, a trial velocity field is built by adding
linear terms to the exact one for hydrostatic loading. The goal of this part is to determine such an exact field. In limit analysis, the most simple solutions are smooth
with a single plastic regime covering the whole body but, generally speaking, it
is a priori expected that limit state solutions may involve some field discontinuities compatible with the continuum mechanics principles [17, 58]. In particular,
the collapse may be incomplete and/or exhibits distinct regimes in subdomains. For
nonassociated Drucker-Prager model, the collapse stress field is statically and plastically admissible. Then the limit load for the nonassociated model is a priori less
than the limit load for the corresponding associated model (i.e. with the normality
rule and the same friction angle). As a matter of fact, although exact solutions do
not exist up to now for such class of problems, numerical simulations show that
for classical soil mechanics applications (bearing capacity of a strip footing, stability of foundations and tunnels), the limit load of the nonassociated case is really strictly less than the one of the corresponding associated case [3, 18, 39]. In
the hollow sphere problem, we consider the event of such weaker solutions but
we conclude to the impossibility of incomplete collapse and more than one plastic regime. The paradoxical consequence is the nonsensitivity of the limit load to
the dilatancy angle. This theoretical result which has been already obtained by [52]
through a non-linear homogenization technique is confirmed by numerical simulations.
In the second part, we tackle the general case of combined hydrostatic and deviatoric loadings. Roughly speaking, a constitutive law in Mechanics is a relationship
between dual variables. The constitutive laws of the materials can be represented,
as in Elasticity, by a univalued mapping or, as in Plasticity, can be generalized
in the form of a multivalued mapping. But this representation is not necessarily
convenient. When the graph is maximal and cyclically monotone, we can model it
thanks to a convex and lower semi-continuous function , called a superpotential (or
183
pseudo-potential), such that the graph is the one of its subdifferential . The function and its Fenchel conjugate one verifies for any couple of dual variables
Fenchels inequality. The dissipative materials admitting a superpotential of dissipation are often qualified as standard [36] and the law is said to be a normality law,
a subnormality law or an associated law.
However, several models proposed these last decades, particularly in Plasticity,
are nonassociated. For such laws, the second author introduced in [20, 21] a suitable
modelization based on the bipotential, a function b of both dual variables, convex
and lower semicontinuous in each argument and satisfying a cornerstone inequality
saying that for any couple of dual variables the value of the bipotential is greater than
or equal to their duality pairing. When equality holds, the couple is said extremal.
In a mechanical view point, the extremal couples are the ones satisfying the constitutive law. Materials admitting a bipotential are called implicit standard materials
(ISM) because the constitutive law is a subnormality law but the relation between
the dual variables is implicit. The classical standard materials correspond to the particular event of the bipotential being separated as the sum of a superpotential and
its conjugate one. In this sense, the cornerstone inequality of the bipotential generalizes Fenchels one. The existence and construction of a bipotential for a given
constitutive law is discussed in [11, 12, 14].
Linked to the structural mechanics and in particular with the Calculus of Variation, the bipotential theory offers an elegant framework to model a broad spectrum
of nonassociated laws. Examples of such nonassociated constitutive laws are:
in soil mechanics, non-associated Drucker-Prager [3, 8, 22, 24, 39] and Cam-Clay
models [23, 62],
the nonlinear kinematical hardening rule for cyclic Plasticity [57, 21, 53] and
Viscoplasticity [37],
Lemaitres coupled damage law [4],
the coaxial laws [26, 61],
Coulombs friction law [9, 21, 22, 24, 25, 2831, 38, 40, 44],
the blurred constitutive laws [13, 15].
A complete survey can be found in [26]. In the previous works, robust numerical
algorithms were proposed to solve structural mechanics problems.
The limit analysis is a general method to determine the plastic collapse of structures under proportional loading [58], particularly in soil mechanics [17, 18], but it
is restricted to associated plasticity (with normality law). The classical presentation
of the nonassociated plasticity is based on a yield function and a plastic potential.
The bipotential offers an alternative formulation opening naturally into a variational
formulation, paving the way to an extension of limit analysis techniques to nonassociated laws [8, 9, 16, 24, 62]. Extension of this method to the repeated variable
loading is called shakedown theory and its extension to the ISM by the bipotential
approach was also considered in [68, 10, 26].
184
L. Cheng et al.
rr
drr
+2
= 0,
dr
r
with static boundary conditions:
rr (a) = p,
rr (b) = q.
(2)
(3)
(4)
where e is the equivalent stress of Von Mises, m the mean stress, 0 > 0 the
cohesion stress of the material and the pressure sensitivity factor related to the
friction angle by:
tan = 3.
The nonassociated flow rule:
d =
G
,
(5)
185
(6)
Of course, for the particular event = hence F = G, the normality rule is recovered and the plasticity model is associated. Without loss of generality, we can
assume that:
1
(7)
0 < ,
2
or equivalently 0 < 56 18 . In practice, these conditions are fulfilled by
the geomaterials and other pressure sensitive dilatant materials. Experimental data
can be found for polymers, high strength steels and aluminium in [34].
1
(2rr ) ,
drr = +
2e
1
(2 rr ) .
d = +
2e
(8)
(9)
.
drr = ( ),
d = d = +
2
(10)
186
L. Cheng et al.
Eliminating the velocity between the two equations (1), one obtains:
d
(rd ).
dr
Eliminating the strain rate components between the previous relations leads to:
drr =
d
3
+
= 0,
dr
s r
where we put for convenience s = 1 + 2. The general solution is:
3
(r) = r s .
(11)
where
vr (r) = K r 1 s ,
(12)
.
K = +
2
(13)
As usual, due to the nature of the limit load problem, the kinematical fields are
defined up to a positive factor.
Let us now discuss the existence of solutions with discontinuities:
Suppose that there exist adjoining spherical shells which have distinct plastic
regimes corresponding respectively to and . The continuity of the velocity
field at the interface of radius r0 entails:
3( s1 s 1 )
K = K r0
Assuming that K and K do not vanish, they should have the same sign,
and the plastic multiplier should be positive. However, under the condition (7),
= K /( + 2 ) and = K /( 2 ) have opposite signs, which is absurd
because, accounting for (11), condition (6) would be violated in one of these
regimes. Thus no change of regime is allowed at the limit state.
Finally, let us suppose that the collapse is not complete. The plastic multiplier
field is identically null in a nonplastified spherical shell, hence so is the velocity
field because of (11), (12) and (13). It is absurd to assume the existence of a
plastic yielding adjoining shell because the continuity of the velocity field at the
interface would force the velocity field to vanish in the yielding shell. Hence the
collapsed must be complete.
Next, the corresponding statical solution is presented. Considering (8), (9) and once
again = , the yield function reads:
F ( ) = ( rr ) + (rr + 2 ) 0 = 0.
It follows:
2( rr ) = 3 (H rr ),
(14)
187
(15)
(16)
It can be verified that , rr and (q + p) have the same sign . In short, the solution
is defined by the limit load (16) and, in the interval a r b, by the collapse fields
of plastic multiplier (11), velocity (12) and radial stress (15). The stress field and
limit loads do not depend on the dilatancy angle and they are identical to the ones of
the associated case with same friction angle, previously obtained in [60] when the
pressure vanishes. Only the collapse mechanism is dilatancy angle dependent. This
insensitivity of the limit load to the dilatancy angle agrees with the model recently
proposed by [52].
For the variational model that will be presented in the second part, it is convenient
to determine also the mean stress field (8). From (14), we obtain:
2 = (2 3 )rr + 3 H.
Taking into account this last relation, = and (15), it comes:
1
f b 3
.
0 (0 + 3p)
m (r) =
3
s r
(17)
In the second part, to lighten the notations, the subscript will be erased.
188
L. Cheng et al.
Fig. 1 Hollow sphere model: Geometry of the elementary cell and boundary conditions
A reference case, denoted Case 1, is firstly defined in which the associated flow
rule is applied with the following parameters: a = 0.585 m, b = 1 m (f 0.2002),
= = 30 , E = 500 MPa, = 0.2 and 0 = 1 MPa. Then, in order to verify
the precision of the proposed model in nonassociated cases, two other simulations
(denoted Cases 2 and 3) are performed with two different values of (15 and 0 ),
both under compression and traction conditions.
Figure 2 displays the FEM results of Cases 1 to 3, where the materials possess
the same porosity (f 0.2002) but different dilation angles. (r) denotes the radial displacement and the limit stress is the asymptotic value. The limit loads of
associated case ( = = 30 ) and nonassociated ones ( = 30 , = 15 and 0 ),
as expected, have almost the same value with very small differences of the order
1 %. The differences between the reference analytical solution and the finite element ones are rather small and can be attributed to numerical errors due to the
discretization. However, the FEM points of these three lines in this figure do not
coincide entirely. In other words, as the displacements imposed upon the external
boundary being the same, the ones at the internal boundary are not. Therefore, the
limit load of nonassociated Drucker-Prager porous material does not depend on the
dilation angle, whereas the collapse mechanism does. More details can be found
in [19] concerning the sensitivity of the proposed model with respect to the porosity
and Youngs modulus.
p G
p 3s
=
d =
+ 1 ,
(18)
2e
189
where is Cauchy stress tensor, s the deviatoric stress, 1 the unit tensor. the scalar
p is defined as:
1/2
2
p = e : e
3
with e the deviatoric part of d. The plastic dilatancy is given by:
tr d = 3 p .
(19)
190
L. Cheng et al.
0 p
0
(20)
,
b(d, ) = dm + ( ) 3m
(22)
(23)
191
It is worth remarking that, with respect to the previous minimization problems, the
bipotential has the required convexity properties. Let us show how to recover simply the plastic yielding condition F ( ) = 0 by the bipotential formalism. The first
minimization problem becomes:
min b0 (d, ) d : = 0,
H (d)0
where b0 is the finite part of the bipotential, given by (20). Relaxing the kinematical
condition H (d) 0 by use of Lagranges multiplier , this constrained minimization problem is transformed into an equivalent saddle-point problem
max min L(d, , ) = b0 (d, ) d : + H (d) = 0,
0
0
0 p p
dm + ( ) 3m
e + 3dm m + p dm .
0
+ ,
e = ( ) 3m
0
.
3m =
L
d,
=
b
max
min
(d,
d
:
F
(
)
= 0,
0
0
0 p
L d, , = dm + ( ) 3m
(e : s + 3dm m )
+ (e + 3m 0 ).
3s
,
2e
( ) p dm + = 0.
(25)
(26)
192
L. Cheng et al.
p 3s
d = e + dm 1 =
+ 1 .
2e
For the treatment of the apex, the reader is referred to [39].
The set of kinematical admissible velocity fields is defined in the following sense:
Ka = v s.t. v(x) = D.x on S .
The associated strain rate field is d(v) = grads v. The set of statically admissible
stress fields is:
Sa = { s.t. div = 0 in VM and = p1 in Vf }.
The set of admissible couples is the product A = Ka Sa and the set of extremal
ones is:
E = (v, ) s.t. d(v), is extremal in VM .
The homogenization problem consists in determining the set A E of admissible
and extremal fields. Accounting for its strong nonlinear nature, this problem has
in general no closed analytical solution. We present now an equivalent variational
formulation, more appropriate for simple approximations, thanks to relevant choice
of trial fields and minimization procedure. Let us consider an admissible couple
(v, ). Thus, by Hills lemma, one has:
d(v) : dV = V 1
d(v)dV : p1 + V 1
d(v) : dV ,
D : = V 1
V
Vf
VM
D : = V 1
d(v)dV : p1 + V 1
193
d(v) : ( + p1)dV .
VM
d(v) : ( + p1)dV .
(27)
VM
Sa
In particular, this occurs for admissible couples (v , ), (v, ), (v, ) and moreover,
in the last case, because of (23):
B(v, ) = 0.
In short, one has for all admissible fields v Ka and Sa :
B v , B(v, ) = 0 and B v, B(v, ) = 0,
which proves (28).
Now, let us discuss some relevant aspects of the variational principles for rigid
perfectly plastic materials such as the one described in the previous sections. The
set of plastically admissible velocity and stress fields are respectively defined as:
Kp = v s.t. H d(v) 0 in VM ,
Sp = s.t. F ( ) 0 in VM .
The sets of licit velocity and stress fields are respectively Kl = Ka Kp and Sl =
Sa Sp while we considered the finite valued functional:
B0 v , = V 1
b0 d v , + d v : p1 dV D : ( + p1).
VM
Sl
(29)
194
L. Cheng et al.
The limit analysis approach consists in finding nontrivial solutions qualified as collapse mechanisms. It is expected that there exist only under an equality condition on
that can be interpreted as the equation of the yielding surface in the macroscopic
model.
It is worth noting that if both D and are chosen arbitrarily, there is in general
no solution to the problem (29). In a practical point of view, it is more convenient
for instance to fix only and to find D and v satisfying the first minimization
problem in (29). Introducing Lagranges multiplier field x (x), this constrained
minimization problem is transformed into an equivalent saddle-point problem
1
max min L (v, , ) = B0 (v, ) + V
H (d)dV .
0 vKa
VM
1
max min L (v, , ) = B0 (v, ) + V
H (d)dV ,
0 vKa
VM
Satisfying the kinematical condition only in an average sense but not locally anywhere in VM is a strong approximation but leading to easier calculations. As consequence of the approximation, it is crucial to remark that the minimum of B0 may
not be expected to be zero. Nevertheless, in the spirit of Ladevzes method of the
error on the constitutive law [4551], its value for the minimizer can be used as a
variational error estimator [30]. The minimum principle allows obtaining the bettersolution within the framework imposed by the approximations.
Introducing the macroscopic strain rate of the void and equivalent strain rate,
respectively defined by:
1
1
D void = V
ddV ,
(v) = V
p dV ,
Vf
VM
T=
+ p1
,
0 + 3p
1
B0 (v, )
=
tr(D D void ) + 1
(v,
) D : T ,
B0 (v, ) =
0 + 3p 3
where:
(v,
) = V 1
VM
0 3m p
dV .
0 + 3p
195
(31)
Introducing the normalized multiplier = /(0 + 3p), the normalized Lagrangian is:
L (v, , )
L (v, , ) =
,
0 + 3p
1
1
L (v, , )
+ 1
tr(D D void ) + (v)
(v,
) D : T .
(32)
(33)
D void = C0 f 1 + f C1 (e e + e e ) + C2 ez ez ,
(34)
2C0
,
s De
196
L. Cheng et al.
with
1
J ( ) =
2
1+
1
3 cos2 1 sin d,
2
where
2x
1/s
sign(C1 C2 ), | | 1.
1 + 2 x 2/s
On the other hand, the bifunctional depends on the stress field only through m .
Under condition (7), the following trial stress field is chosen as the solution (17) of
the pure hydrostatic case, without additional terms:
1
f b 3
m (r) =
,
0 (0 + 3p)
3
s r
=
I ( ),
p dV =
(v, ) = V
s VM r
s
where
I ( ) = De
x J ( ) 1 + 2 x 2/s dx.
(35)
1
=
(1 f )Dm f f C0
L (v, , )
+ (v) + 1
(v,
) (De Te + 3Dm Tm ),
,De ,
(37)
Te = ,De + 1
197
1
(1 f ),
1
,C0 + 1
f f = 0.
,C0
3Tm =
(38)
(39)
(f
,C0
f ) + ( 1)
f f + ,C0
,De ] + (1 )(,C0
,De ,De
,C0 )
(f f )[ ,De + (1 )
f f + ,C0
3Tm = (1 f )
(40)
,C0
,C0 + (1 )
f f + ,C0
(41)
C0
1 f ,
(v,
) =
C0
f f ,
f f + ,C0 = 1 f.
Finally, the macro-stress is given by (40) and (41):
1
Te = 0,
(42)
3Tm = 1 f ,
that was expected with respect to the exact result (16) determined in the first part.
The solution is the same as for the corresponding associated case [34].
Purely deviatoric case: = 0 C0 = 0. The integral (36) is:
1
1 f 1
I ( ) = De
.
x dx = De
1
f
Then, it holds:
(v) = I (0) = De (1 f ),
(v,
) = De f f ,
198
L. Cheng et al.
f f + ,C0 = f f.
Finally, the macro-stress is given by (40) and (41):
f f ,
Te = 1 f + 1
Tm = 0.
(43)
When the flow rule is associated, we recover the result obtained in [34]. Nevertheless, the accuracy of (43) is poor for strong lack of associativity. The weakness
of this model is that we use the expression (17) of the mean stress deriving from
the solution in the pure hydrostatic case. We are currently working to improve
the trial stress field by introducing additional terms into (17) and working on the
stress principle.
4 Conclusion
Unlike currently observed in other problems, in the one of the hydrostatically loaded
hollow sphere the limit load and collapse stress field for the nonassociated cases
are the same as for the corresponding associated case. This event may appear at
first glance paradoxical. The key point is the strong condition of central symmetry
which is very restrictive and prevents field discontinuities generally allowed by the
continuum mechanics. Thus only complete solution with a unique plastic regime
is considered and it is necessarily identical to the one of the associated case. The
general case of combined hydrostatic and deviatoric loadings is in progress. The
first elements of the corresponding theory are exposed in Sect. 3.
References
1. Barthlmy JF, Dormieux L (2003) Dtermination du critre de rupture macroscopique dun
milieux poreux par homognisation nonlinaire. C R, Mc 331:271276
2. Benzerga AA, Besson J (2001) Plastic potentials for anisotropic porous solids. Eur J Mech A,
Solids 20:397434
3. Berga A, de Saxc G (1994) Elastoplastic finite element analysis of soil problems with implicit
standard material constitutive laws. Rev Eur lments Finis 3:411456
4. Bodovill G (1999) On damage and implicit standard materials. C R Acad Sci Paris, Sr II,
Fasc B, Mc Phys Astron 327:715720
5. Bodovill G, de Saxc G (2001) Plasticity with nonlinear kinematic hardening: modelling and
shakedown analysis by the bipotential approach. Eur J Mech A, Solids 20:99112
6. Bouby C, de Saxc G, Tritsch JB (2006) A comparison between analytical calculations of the
shakedown load by the bipotential approach and step-by-step computations for elastoplastic
materials with nonlinear kinematic hardening. Int J Solids Struct 43:26702692
7. Bouby C, de Saxc G, Tritsch JB (2009) Shakedown analysis: comparison between models
with the linear unlimited, linear limited and nonlinear kinematic hardening. Mech Res Commun 36:556562
8. Bousshine L, Chaaba A, de Saxc G (2001) Softening in stress-strain curve for Drucker-Prager
non-associated plasticity. Int J Plast 17:2146
199
9. Bousshine L, Chaaba A, de Saxc G (2002) Plastic limit load of plane frames with frictional
contact supports. Int J Mech Sci 44:21892216
10. Bousshine L, Chaaba A, de Saxc G (2003) A new approach to shakedown analysis for nonstandard elastoplastic material by the bipotential. Int J Plast 19:583598
11. Buliga M, de Saxc G, Valle C (2008) Existence and construction of bipotential for graphs
of multivalued laws. J Convex Anal 15:87104
12. Buliga M, de Saxc G, Valle C (2009) Bipotentials for non-monotone multivalued operators:
fundamental results and applications. Acta Appl Math 110:955972
13. Buliga M, de Saxc G, Valle C (2011) Blurred constitutive laws and bipotential convex covers. Math Mech Solids 16:161171
14. Buliga M, de Saxc G, Valle C (2010) Nonmaximal cyclically monotone graphs and construction of a bipotential for the Coulombs dry friction law. J Convex Anal 17:8194
15. Buliga M, de Saxc G, Valle C (2010) Blurred maximal cyclically monotone sets and bipotentials. Anal Appl 8:323336
16. Chaaba A, Bousshine L, de Saxc G (2010) Kinematic limit analysis of nonassociated perfectly plastic material by the bipotential approach and finite element method. J Appl Mech
77:1631
17. Chen WF (1975) Limit analysis and soil plasticity. Elsevier, New York
18. Chen WF, Liu XL (1990) Limit analysis in soil mechanics. Developments in geotechnical
engineering, vol 52
19. Cheng L, Jia Y, Oueslati A, de Saxce G, Kondo D (2012) Plastic limit state of the hollow
sphere model with nonassociated Drucker-Prager material under isotropic loading. Comput
Mater Sci 62:210215
20. de Saxc G, Feng ZQ (1991) New inequality and functional for contact friction: the implicit
standard material approach. Mech Struct Mach 19:301325
21. de Saxc G (1992) Une gnralisation de lingalit de Fenchel et ses applications aux lois
constitutives. C R Acad Sci Paris, Sr II 314:125129
22. de Saxc G, Bousshine L (1993) On the extension of limit analysis theorems to the nonassociated flow rules in soils and to the contact with Coulombs friction. In: XI Polish conference
on computer methods in mechanics, Kielce, Poland, pp 815822
23. de Saxc G (1995) The bipotential method, a new variational and numerical treatment of the
dissipative laws of materials. In: Proceedings of the 10th international conference on mathematical and computer modelling and scientific computing, Boston, Massachusetts
24. de Saxc G, Bousshine L (1998) Limit analysis theorems for the implicit standard materials:
application to the unilateral contact with dry friction and the non-associated flow rules in soils
and rocks. Int J Mech Sci 40:387398
25. de Saxc G, Feng ZQ (1998) The bi-potential method: a constructive approach to design the
complete contact law with friction and improved numerical algorithms. Math Comput Model
6:225245
26. de Saxc G, Bousshine L (2002) Implicit standard materials. In: Weichert D, Maier G (eds)
Inelastic behaviour of structures under variable repeated loads. CISM courses and lectures,
vol 432. Springer, Wien
27. Ekeland I, Temam R (1975) Convex analysis and variational problems. North Holland, Amsterdam
28. Feng ZQ, Hjiaj M, de Saxc G, Mrz Z (2006) Effect of frictional anisotropy on the quasistatic
motion of a deformable solid sliding on a planar surface. Comput Mech 37:349361
29. Feng ZQ, Hjiaj M, de Saxc G, Mrz Z (2006) Influence of frictional anisotropy on contacting
surfaces during loading/unloading cycles. Int J Non-Linear Mech 41:936948
30. Fortin J, de Saxc G (1999) Modlisation numrique des milieux granulaires par lapproche
du bipotentiel. C R Acad Sci, Sr IIb 327:721724
31. Fortin J, Hjiaj M, de Saxc G (2002) An improved discrete element method based on a variational formulation of the frictional contact law. Comput Geotech 29:609640
32. Garajeu M, Suquet P (1997) Effective properties of porous ideally plastic or viscoplastic materials containing rigid particles. J Mech Phys Solids 45:873902
200
L. Cheng et al.
33. Gologanu M, Leblond JB, Perrin G, Devaux J (1997) Recent extensions of Gursons model
for porous ductile metals. In: Suquet P (ed) Continuum micromechanics. Springer, Berlin
34. Guo TF, Faleskog J, Shih CF (2008) Continuum modeling of a porous solid with pressuresensitive dilatant matrix. J Mech Phys Solids 56:21882212
35. Gurson AL (1977) Continuum theory of ductile rupture by void nucleation and growth, part I:
yield criteria and flow rules for porous ductile media. J Eng Mater Technol 99:215
36. Halphen B, Nguyen Quoc S (1975) Sur les matriaux standard gnraliss. C R Acad Sci Paris
14:3963
37. Hjiaj M, Bodovill G, de Saxc G (2000) Matriaux viscoplastiques et loi de normalit implicites. C R Acad Sci Paris, Sr II, Fasc B, Mc Phys Astron 328:519524
38. Hjiaj M, de Saxc G, Mrz Z (2002) A variational-inequality based formulation of the frictional contact law with a non-associated sliding rule. Eur J Mech A, Solids 21:4959
39. Hjiaj M, Fortin J, de Saxc G (2003) A complete stress update algorithm for the non-associated
Drucker-Prager model including treatment of the apex. Int J Eng Sci 41:11091143
40. Hjiaj M, Feng ZQ, de Saxc G, Mrz Z (2004) Three dimensional finite element computations
for frictional contact problems with on-associated sliding rule. Int J Numer Methods Eng
60:20452076
41. Jeong HY, Pan J (1995) A macroscopic constitutive law for porous solids with pressuresensitive matrices and its applications to plastic flow localization. J Mech Phys Solids
39:13851403
42. Jeong HY (2002) A new yield function and a hydrostatic stress-controlled model for porous
solids with pressure-sensitive matrices. J Mech Phys Solids 32:36693691
43. Jia Y (2006) Contribution la modlisation thermo-hydro-mcanique des roches partiellement
satures: application au stockage des dchets radioactifs. Thesis of University of Lille 1
44. Laborde P, Renard Y (2008) Fixed points strategies for elastostatic frictional contact problems.
Math Methods Appl Sci 31:415441
45. Ladevze P (1975) Comparaisons de modles de milieux continus. Thse dEtat, Universit
Pierre et Marie Curie, Paris
46. Ladevze P, Coffignal G, Pelle JP (1986) Accuracy of elastoplastic and dynamic analysis.
In: Babuska I, Gago J, Oliveira E, Zienkiewicz OC (eds) Accuracy estimates and adaptive
refinements in finite element computations. Wiley, New York, pp 181203
47. Ladevze P, Pelle JP, Rougeot P (1991) Error estimation and mesh optimization for classical
finite element. Eng Comput 8:6980
48. Ladevze P, Moes N (1997) A new a posteriori error estimation for nonlinear time-dependent
finite element analysis. Comput Methods Appl Mech Eng 157:4568
49. Ladevze P, Pelle JP (2001) La matrise du calcul en mcanique linaire et non linaire. Hermes Science, Paris
50. Ladevze P, Florentin E (2006) Verification of stochastic models in uncertain environments
using the constitutive relation error method. Comput Methods Appl Mech Eng 196:224225
51. Ladevze P, Puel G, Deraemaeker A, Romeuf T (2006) Validation of structural dynamics
models containing uncertainties. Comput Methods Appl Mech Eng 195:373393
52. Maghous S, Dormieux L, Barthlmy JF (2009) Micromechanical approach to the strength
properties of frictional geomaterials. Eur J Mech A, Solids 28:179188
53. Magnier V, Charkaluk E, Bouby C, de Saxc G (2006) Bipotential versus return mapping algorithms: implementation of non-associated flow rules. In: Topping BHV, Montero G, Montenegro R (eds) Proceedings of the eighth international conference on computational structures
technology, Las Palmas de Gran Canaria, September 1215, 2006. Civil-Comp Press, Stirlingshire, paper 68
54. Monchiet V, Charkaluk E, Kondo D (2007) An improvement of Gurson-type models of porous
materials by Eshelby-like trial velocity fields. C R, Mc 335:3241
55. Monchiet V, Cazacu O, Kondo D (2008) Macroscopic yield criteria for plastic anisotropic
materials containing spheroidal voids. Int J Plast 24:11581189
56. Moreau JJ (2003) Fonctionnelles convexes. Istituto Poligrafico e Zecca dello Stato, Rome
57. Rockafellar RT (1970) Convex analysis. Princeton University Press, Princeton
201
58. Save MA, Massonnet CE, de Saxc G (1997) Plastic limit analysis of plates, shells and disks.
Elsevier, New York
59. Shao JF, Jia Y, Kondo D, Chiarelli AS (2006) A coupled elastoplastic damage model for
semi-brittle materials and extension to unsaturated conditions. Mech Mater 38:218232
60. Thor P, Pastor F, Pastor J, Kondo D (2009) Closed-form solutions for the hollow sphere
model with Coulomb and Drucker-Prager materials under isotropic loadings. C R, Mc
337:260267
61. Valle C, Lerintiu C, Fortun D, Ban M, de Saxc G (2005) Hills bipotential. In: MihailescuSuliciu M (ed) New trends in continuum mechanics. Theta series in advanced mathematics.
Theta Foundation, Bucharest, pp 339351
62. Zouain N, Pontes Filho I, Borges L, Mouta da Costa L (2007) Plastic collapse in nonassociated hardening materials with application to Cam-clay. Int J Solids Struct 44:4382
4398
1 Introduction
A global evaluation of the structural collapse safety of slender elastic structures
should consider all possible loadings, including the deviations due to load imperfections and geometrical defects. Standard path-following approaches, aimed at recovering the equilibrium path for a single loading case and assigned imperfections,
are not suitable for this purpose. In fact in order to perform a reliable structural
safety assessment the nonlinear analysis should be performed with respect to all
possible imperfection shapes. The consequent computational burden can be very
G. Garcea (B) A. Bilotta A. Madeo R. Casciaro
Dipartimento di Modellistica per lIngegneria, Universit della Calabria, Rende, Italy
e-mail: [email protected]
A. Bilotta
e-mail: [email protected]
A. Madeo
e-mail: [email protected]
R. Casciaro
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_10,
Springer Science+Business Media Dordrecht 2014
203
204
G. Garcea et al.
2 Local Analysis
Adopting the same functional notation as the paper of Budiansky [7] we consider a
hyperelastic structure subjected to an assigned load, p[], which is linear with a
parameter (p[] = p)
and described by its potential energy [u] in terms of stress
and displacement here denoted by u. Equilibrium implies the condition
= 0,
u := [u]u pu
u T
(1)
205
where [u] is the strain energy, p[]u the external work, U the manifold of the admissible configurations and T its tangent space (the prime stands for the Frechts
differentiation with respect to u). Usually (and conveniently) the configuration is
described making U a linear manifold, so T becomes independent from u.
Equation (1) defines a curve (it may be composed of several separate branches)
in the space (u, ), called equilibrium path that can be expressed in a parametric
form in terms of a suitable abscissa = g[u, ]
u = u[ ],
(2)
= [ ]
selected so that = 0 gives the known equilibrium point (u0 , 0 ). The aim of
the asymptotic analysis is the evaluation of the equilibrium path starting from this
known configuration usually assumed corresponding to 0 = 0, exploiting an implementation of the Koiter approach to elastic stability [25] in a FEM context.
1
1 ... 3
u[ ] = u0 + u 0 + u 0 + u 0 + ,
2
6
(3)
... 3
1
1
[ ] = 0 + 0 + 0 + 0 +
2
6
where a subscript denotes the point in which the quantities are evaluated and
(u 0 , 0 ), (u 0 , 0 ), etc are the tangent, curvature etc. of the equilibrium path in = 0.
Expressing Eq. (1) using the parametrization in Eq. (2) we obtain a function of the
real variable , its Taylor expansion in = 0 furnishing
u[ ] [ ]p u = 0 0 p u
+ 0 u 0 0 p u
1
+ 2 0 u 0 + 0 u 20 0 p u
2
...
1 ...
+ 3 0 u 0 + 30 u 0 u 0 + 0 u 30 0 p u +
6
(4)
206
G. Garcea et al.
0 u 0 0 p u = 0,
2
0 u 0 0 p u = 0 u 0 u,
...
...
0 u 0 0 p u = 30 u 0 u 0 + 0 u 30 u,
... = ...
u T
(5)
with the perturbation equation of order zero not reported because it is verified a
priori by the equilibrium point (u0 , 0 ).
In order to solve (5) in a point where the operator [u0 ] is not singular, i.e. a
regular point, we have to add to Eqs. (5) the normalization condition, that is the definition of the abscissa introduced to parametrize the equilibrium curve, := g[u, ],
whose asymptotic expansion gives
g0 u 0 + g 0 0 = 1,
g0 u 0 + g 0 0 = g0 u 20 g 0 20 2g 0 0 u 0 ,
...
...
3g 0 u 0 0 3g 0 0 0 3g 0 u 0 20 3g 0 u 20 0
(6)
where a hat denotes the derivative with respect to . Equations (5) and (6) allow us
to define the necessary asymptotic quantities unequivocally.
The solution of the initial nonlinear problem is then reduced to the solution of a
sequence of linear problems that define, at each order, the asymptotic terms of that
order
(u 0 , 0 )
(u 0 , 0 )
= 0,
0 u 0 u 0 pu
u,
g0 u 0 + g 0 0 = 1,
0 u 0 u + 0 u 20 u 0 pu
= 0,
u,
(7)
g0 u 0 + g 0 0 = g0 u 20 g 0 20 2g 0 0 u 0
...
This is an important feature of the method which decouples each order from the next
one and allows the asymptotic expansion to be stopped at the desired order.
207
u T .
(8)
Let us now consider the single mode case, i.e. dim(ker(c )) = 1. The perturbation
equations (7) with respect to the critical point can be written, as well as the regular
point case. Due to the singularity of c in order to verify the first of Eq. (7) for
u = vc , the satisfaction of the following Fredholm orthogonality condition on the
first order perturbation equation is required:
c p vc = 0.
(9)
Equation (9) is verified in both cases of c = 0, i.e. a limit point for the loads, and
of p vc = 0 i.e. a bifurcation point (see Fig. 1).
2.2.1 Limit Point
In this case the tangent to the equilibrium path, (u c , c ), is unequivocally defined
from the first condition (7). In particular the solution assumes the following expression
u c = vc ,
c = 0
(10)
208
G. Garcea et al.
(11)
u T
(12)
7 89 :
789:
c u c vc + c u 2c vc c p vc = 0
c u 2c vc = 0.
(13)
After substitution of u c from (11) we have the second order polynomial equation
2c c u 2c vc + 2 c c u c vc2 + 2 c vc3 = 0
that, together with the normalization condition linear in c and
c gc u c + g c + gc vc = 1,
(14)
(15)
gives the two values of the tangent. The condition p vc = 0 then detects a bifurcation
point characterized by two different equilibrium paths with two different tangents.
For a given value of we can define, from Eq. (14) two values of c . A particular
case occurs when, with an appropriate selection of the parametrization, one of the
tangents is characterized by the value = 0. In this case Eq. (14) and condition
c = 0, require that:
c u 2c vc = 0.
The two tangents are then
u (1)
c,
c =u
(1)
c = 1,
(2)
(2)
u c = c u c + vc ,
1 c vc3
(2)
c =
2 c u c vc
with defined by the normalization condition. For the first tangent we have
[ ]
c =
=1
=0
(16)
209
that is a path parameterized with = and for which all the higher order derivatives of with respect to are zero. The discussion can be easily extended to the
curvature (see [8]).
2.2.3 Path Representation on a Bifurcation Point
The previous discussion allows a different rewriting of the two paths passing through
a simple bifurcation point. From a simple extension of the first order asymptotic
expansion we use, the two solutions can be expressed as
[ ] = c + c + ,
(17)
u[, ] = uf [] + v[ ]
with the fundamental path, that is the initial path followed by the structure starting
from = 0, defined as
uf [] = uc + ( c )u c
parameterized in and obtained from (17) by setting = 0 while the bifurcated
path,
1
v[ ] = vc + 2 vc +
2
intersects uf in the bifurcation point. With this choice measures the distance between u and uf and can be selected as
1
= u uf , vc vc , vc + 2 vc , vc +
2
(18)
where , represents a given scalar product. Equation (18) at the first order furnishes a normalization condition for vc while at the second order an orthogonality
condition between vc and vc .
210
G. Garcea et al.
0 u + 0 u 2 u = 0,
u T .
0 u + 30 u u + 0 u 3 u = 0,
(19)
(22)
V W = {0}
(23a)
where denotes the direct sum and we select V so that c is definite positive
on W . This means that V could be generically selected but it is necessary to include
211
all the directions of singularity of c . The natural choice for V is to use all the
buckling modes vi obtained by means of Eq. (21)
n
V := v : v =
i vi , i R
(23b)
i=1
where i are scalar coefficients. From now on we also assume the following definition for W
f
W = w : c u c vi w = 0, vi .
(23c)
Equations (23a) ensure that any u T can be decomposed as
u = v + w
with v V , w V .
(23d)
The equilibrium path is obtained by adding the fundamental path and the bifurcated one. Using the decomposition in Eqs. (23a)(23d) we obtain
u[k , ] = uf [] + v[k ] + w[k , ]
with w W , v V
(24)
where the orthogonal correctives w are functions of and k . We also assume that
in uf [c ] we have k = 0.
Introducing the admissible configurations (24) into the equilibrium equation we
obtain a scalar function of k and
r[k , ] := u[k , ] p u
(25)
that using the decomposition in Eqs. (23a)(23d) becomes
u0 W ,
()
r [k , ] u[k , ] p u = 0
uk = vk ,
k = 1...m
(26)
where from now on Greek indexes go from 0 to m while the Roman ones from 1
to m and the index zero denotes quantities in W . The residual equation for = 0,
expressed in asymptotic form, will be used to evaluate a polynomial expression for
w[, i ] that is then substituted in the remaining equations for = 1 . . . m, these
ones also being expressed in asymptotic form, to obtain the equilibrium equations.
212
G. Garcea et al.
1 2 r ()
1 2 r ()
2
(
)
+
i j + = 0
c
2 2 c
2 i j c
(27)
where rc = r [uf [c ]] = 0.
Denoting with a pedex k the derivative with respect to k we have
r ()
r ()
2 r ()
i
2 r ()
2 r ()
i j
= c (vi + w i )u ,
f
= c u c + w c p u ,
f
= c w i + c u c + w c (vi + w i ) u ,
(28)
f
2
f
= c w c + u c + c u c + w c u ,
= c w ij + c (vi + w i )(vj + w j ) u .
Due to Eqs. (19) we obtain from the second and the fourth equation of (28) for = 0
that
v0 W
c wu
0=0
w = 0,
0 = 0
c wu
w = 0
(29)
that furnishes w k = 0 due to the definition of W . The same occurs for w c = 0 while
we obtain
c w ij + c vi vj u0 = 0 v0 W .
(30)
We obtain then
w[, k ] =
1
i j w ij
2
i,j
that can be introduced in Eq. (27) for = 1 . . . m to evaluate the equilibrium path.
213
(31)
Letting
p[]u
= [uf ] p u
1 2 2
1 2
3
b u vb + ( 3b )b u u (32)
2
6
be the equilibrium residual along the fundamental path and performing again an
asymptotic expansion of the remaining terms of (31) starting now from uc we obtain
the following function of and k
r () [k , ] = p[]
+ i b vi + b w[k , ] + ( b )i b u vi
1
2 2
+ ( b )b uw[
vi
k , ] + i ( b ) b u
2
1
+ ( b )2 b u 2 w[k , ]
2
1
+ b i j vi vj + w[k , ]2 + 2i vi w[k , ]
2
1
+ ( b )b u i j vi vj + w[k , ]2 + 2i vi w[k , ]
2
3
1
(33)
+ b i vi + w[k , ] + u = 0
6
where the convention of summing on the repeated index has been used.
214
G. Garcea et al.
From the condition r (0) [k , ] = 0 and deleting high order terms in the expansion
we obtain the following expression for w[, k ]
m
1 2
w[k , ] =
i k w ij
(34)
w +
2
i,j =1
where
c w + c u 2 u0 = 0,
c w ij + c vi vj u0 = 0,
u0 W .
(35)
Substituting the definition in Eq. (34) in Eqs. (33) and recalling Eq. (21) we
obtain the asymptotic expression for the equilibrium equations for = 1 . . . m
m
m
b
1
k [] + (k )k b
i Cik +
i j Aij k
2
2
i=1
i,j =1
1
+ ( b )2
i B00ik
2
m
i=1
m
m
1
1
+ ( b )
i j B0ij k +
i j h Bij hk = 0,
2
6
i,j =1
k = 1 . . . m (36)
i,j,h=1
1
1
k [] = 2 b u 2 vk + 2 ( 3b )b u 3 vk ,
2
6
A = v v v ,
ij k
b i j k
B
ij w hk + w ih w j k + w ik w j h ),
ij hk = b v i v j v h v k b (w
B
= b u 2 vi vk b w00 w ik ,
00ik
B0ij k = b u vi vj vk ,
Cik = b w w ik
(37)
215
presence of these, while preserving the general behaviour of the structure, changes
some aspects of its response and often causes a reduction in the carrying capacity
(imperfection sensitivity).
and/or
The presence of small additional imperfections expressed by a load q q[]
an initial displacement u u can be easily considered in the asymptotic analysis.
Once introduced in the equilibrium equation (26) and performing the expansion
they only affect Eq. (36). In particular we only need to add the additional terms (see
[8, 27, 28])
vk ,
lk [] := q[]
k [] := c u u vk
g
(38)
(39)
216
G. Garcea et al.
for the limit load multiplier and all the other useful statistical information. This solution process, we call full analysis, can be considered as a standard approach for
imperfection sensitivity analysis (see [10]). The number of repetitions needed to obtain statistically reliable results increases (quite) exponentially with the number m
of the buckling modes and for large m can become very expensive. The imperfection
sensitivity analysis can however be performed in a simple and efficient way when a
criterion for defining the (few) significant imperfection forms exists.
i = 1 . . . m, t R
(41)
m
1
Aij h i i h = min
,
2
(k )
i,j,h=1
i i = 1
(42)
i=1
m
1
Bij hk i j h k = min
,
3
(k )
i,j =1
i i = 1
(43)
i=1
217
(44)
where u is the initial path tangent, solution of the linear vectorial equation
K0 u = p
(45)
K0 being the stiffness matrix evaluated for u = u0 and p the unitary load
vector, defined by the energy equivalencies
uT K[u]u = [u]uu,
uT p = pu.
K[] := K[u0 + u]
(46)
This corresponds to a nonlinear eigenvalue problem which can be linearized and solved using standard algorithms and exploiting matrix K1
0 ,
already available from the previous step, to perform the iterations [9].
(iii) Letting c be an appropriate reference value for the cluster, e.g. the
smallest of i or their mean value, the asymptotic approximation for the
218
G. Garcea et al.
m
i=0
m
1
i v i +
i j wij
2
(47)
i,j =0
where the quadratic corrections wij W are obtained by the linear orthogonal equations
wT (Kc wij + pij ) = 0,
w W
(48)
219
the case of symmetric buckling, problem (43) can be transformed into a non-convex
Quadratic Problem subject to linear constraints and solved using the strategy presented in [10].
220
G. Garcea et al.
221
guaranteed in the structural modeling. Small inaccuracies, deriving from geometrical incoherencies in the higher-order terms of the expansion of the kinematical
laws or in its finite element representation, significantly affect the accuracy of the
solution and can make it unreliable. Structural models not affected by rigid body
motions or by changes in observer are then required. This aspect is more important
with respect to the path-following case where only the first variation needs to be
correctly evaluated.
With this aim the Implicit Corotational Method (ICM) has been proposed as a
tool to obtain geometrically exact nonlinear models for structural elements, such as
beams or shells, undergoing finite rotations and small strains starting from the solutions for the 3D Cauchy continuum used in the corresponding linear modelings
(see [15, 16]). The main idea is to associate a corotational frame to each point of
the 3D continuum so allowing the motion in the neighbor of the point to be split
in a pure stretch followed by a pure rotation, according to the decomposition theorem. It is possible to show how, using the small strain hypothesis and rotation
algebra, the linear stress and linear strain fields, when viewed in this corotational
frame, can provide accurate approximations for the Biot nonlinear stress and strain
fields. Once the corotational rotation is appropriately defined, the local statics and
kinematics of the model are recovered from the linear solution as a function of the
stress/displacement resultants. Stress and strain fields are then introduced within a
mixed variational principle in order to obtain the constitutive laws directly in terms
of stress/strain resultants. This completes the ICM definition of the nonlinear model.
The nonlinear model so obtained retains all the details of the 3D linear solution,
including torsion/shear warping, while its objectivity is ensured implicitly. Furthermore, the use of the mixed formulation and the greater accuracy with which the ICM
recovers the stress field, allows an accurate description of the constitutive laws in
terms of resultants. ICM does not require any ad-hoc assumption about the structural
model at hand, nor depends on any particular parametrization of the rotation tensor,
but actually behaves as a black-box tool able to translate known linear models into
the corresponding nonlinear ones. Moreover, the direct use of a mixed (stress/strain)
description provides an automatic and implicitly coherent methodology for generating models free of the nonlinear locking effects previously discussed, in a format
directly suitable for use in FEM implementations.
The method was applied in [15] to derive 3D beam and plate nonlinear models
starting from the Saint Vennt rod and Kirchhoff and Mindlin-Reissner plate linear
theories, respectively. Some results obtained in [16] will be also given.
5 Numerical Results
Some results regarding the analysis of both 3D beams and plates are reported
and compared with particular reference to accuracy as previously discussed. In the
monomodal buckling tests, to compare the accuracy with known solutions, the following quantities, defining the postcritical tangent and curvature to the bifurcated
222
G. Garcea et al.
N.elem.
b
2b
Out plane
LC
LS
2D Beam()
PM
In plane
LC
LS
PM
16
32
64
9.901
9.877
9.872
9.901
9.877
9.872
9.901
9.877
9.871
9.918
9.870
9.867
9.918
9.870
9.870
9.918
9.870
9.870
9.870
16
32
64
0.354
0.375
0.375
0.020
0.000
0.000
0.145
0.125
0.125
0.166
0.126
0.125
1.03
1.00
1.00
0.166
0.126
0.125
0.125
Fig. 3 Euler beam: problem description and buckling and post-buckling parameters
1 A111
,
2 A011
b =
The results are compared with known analytical solutions and with the ones obtained using the LC (Complete Lagrangian) and LS (Simplified Lagrangian) technical plate models already implemented in the code named KASP [13, 20]. An independent analysis has also been made using the commercial code ABAQUS.
223
tion [16]. In Fig. 3 the values of the buckling loads and post-critical curvatures are
compared with the values obtained by using the Antman beam model and exact
interpolation functions. The ICM model recovers the analytical solution for sufficiently fine grids exactly. The LC and LS models provide a correct answer for the
buckling load, but have a different post-buckling behavior in the in-plane or outof-plane analysis: LC agrees perfectly with the exact solution in the in-plane case,
b = 2, which is eight times greater; conwhereas LS provides the wrong result /
versely, LS behaves better in the out-of-plane case, by providing the approximation
b /b = 0, while LC gives a completely erroneous unstable postbuckling curvature
/b = 0.75. The resulting paths in Fig. 4 show a good agreement with those
computed by path-following analyses.
224
G. Garcea et al.
used is that reported in [17]. The equilibrium paths reported in Fig. 7, compared
with the results of a path-following analysis with ABAQUS, show the accuracy of
the asymptotic analysis.
The second test is the thin-walled beam in Fig. 8 modelled as a plate assemblage.
The model is that proposed in [15, 16] on the basis of the ICM and is denoted as
MP in the results. The results are compared with those of an ABAQUS analysis
using a path-following approach and of the technical plate models [13]. The greater
accuracy of the objective structural model is evident in Fig. 8 where the equilibrium
paths are depicted.
The third case regards a C-shaped cantilever beam subjected to a single force at
the free end reported in Fig. 9. In this case the strong modal interaction between
non near buckling loads also produce a highly unstable behaviour as shown by the
equilibrium path.
Finally the last test regards the Geodetic Dome modelled through a 3D truss
as proposed in [10]. In this case many locally coincident buckling modes interact and the structure exhibits a very marked unstable behaviour. In Fig. 10, in the
modal space (k ) different equilibrium paths, clearly converging along only one of
the minimum directions, are reported. The test shows how it is possible to perform
the sensitivity analysis in a simplified way along the predetermined quartic form
minimum directions.
225
226
G. Garcea et al.
6 Conclusions
The paper deals with the numerical implementation of the Koiter asymptotic approach to directly evaluate the buckling and postbuckling behaviour of geometrically nonlinear structures also in the case of multiple coincident buckling loads and
random loading and/or geometrical imperfections. Standard techniques, based on repeated path-following analyses, are useful for a thorough investigation of the structural behaviour with a single imperfection shape, but cannot be considered effective
tools to predict the safety factor for geometrically nonlinear problems. The asymptotic method, instead, appears to be an attractive alternative as it allows a reliable
analysis with computational costs similar to those required by a standard load buckling prediction, while subsequent analyses for different imperfections are possible
only solving a nonlinear system of m equations, m being the number of buckling
loads considered. In this sense it can be considered as a direct method for predicting
the safety factor of geometrically nonlinear elastic structures. The method furnishes
227
accurate results and also information about the worst imperfection shape if a series
of modelling and implementation aspects are carefully tuned. In particular it was
shown how the effects of the use of geometrically exact structural models and their
coherent finite element implementation are very relevant, while a mixed formulation
eliminates both interpolation and extrapolation locking phenomena.
References
1. Abichou H, Zahrouni H, Potier-Ferry M (2002) Asymptotic numerical method for problems
coupling several nonlinearities. Comput Methods Appl Mech Eng 191 (51-52):57955810
2. Ali MA, Sridharan S (1988) A versatile model for interactive buckling of columns and beam
columns. Int J Solids Struct 24:481496
3. Arbocz J, Hol JMAM (1990) Koiters stability theory in a computer-aided engineering (CAE)
environment. Int J Solids Struct 26:945973
4. Barbero EJ, Godoy LA, Raftoyiannis IG (1996) Finite elements for three-mode interaction in
buckling analysis. Int J Numer Methods Eng 39:469488
5. Boutyour EH, Zahrouni H, Potier-Ferry M, Boudi M (2004) Asymptotic-numerical method
for buckling analysis of shell structures with large rotations. J Comput Appl Math 168(1
2):7785
6. Brezzi F, Cornalba M, Di Carlo A (1986) How to get around a simple quadratic fold. Numer
Math 48:417427
7. Budiansky B (1974) Theory of buckling and post-buckling of elastic structures. Advances in
applied mechanics, vol 14. Academic Press, New York
8. Casciaro R (2005) Computational asymptotic post-buckling analysis of slender elastic structures. CISM courses and lectures, vol 470. Springer, New York
9. Casciaro R, Garcea G, Attanasio G, Giordano F (1998) Perturbation approach to elastic postbuckling analysis. Comput Struct 66:585595
10. Casciaro R, Mancusi G (2006) Imperfection sensitivity due to coupled local instability: a nonconvex QP solution algorithm. Int J Numer Methods Eng 67:815840
11. Chen H, Virgin LN (2006) Finite element analysis of postbuckling dynamics in plates, part I:
an asymptotic approach. Int J Solids Struct 43:39834007
12. Flores FG, Godoy LA (1992) Elastic postbuckling analysis via finite element and perturbation
techniques, part 1: formulation. Int J Numer Methods Eng 33:17751794
13. Garcea G (2001) Mixed formulation in Koiter analysis of thin-walled beam. Comput Methods
Appl Mech Eng 190:33693399
14. Garcea G, Formica G, Casciaro R (2005) A numerical analysis of infinitesimal mechanisms.
Int J Numer Methods Eng 62:9791012
15. Garcea G, Madeo A, Casciaro R (2012) The implicit corotational method and its use in the
derivation of nonlinear structural models for beams and plates. J Mech Mater Struct 7(6):509
538
16. Garcea G, Madeo A, Casciaro R (2012) Nonlinear FEM analysis for beams and plate assemblages based on the implicit corotational method. J Mech Mater Struct 7(6):539574
17. Garcea G, Madeo A, Zagari G, Casciaro R (2009) Asymptotic post-buckling FEM analysis
using a corotational formulation. Int J Solids Struct 46(2):523532
18. Garcea G, Salerno G, Casciaro R (1999) Extrapolation locking and its sanitization in Koiter
asymptotic analysis. Comput Methods Appl Mech Eng 180(12):137167
19. Garcea G, Trunfio GA, Casciaro R (1998) Mixed formulation and locking in path-following
nonlinear analysis. Comput Methods Appl Mech Eng 165(14):247272
20. Garcea G, Trunfio GA, Casciaro R (2002) Path-following analysis of thin-walled structures
and comparison with asymptotic post-critical solutions. Int J Numer Methods Eng 55:73100
228
G. Garcea et al.
21. Godoy LA, Banchio EG (2001) Singular perturbations for sensitivity analysis in symmetric
bifurcation buckling. Int J Numer Methods Eng 52:14651485
22. Goltermann P, Mllmann H (1989) Interactive buckling in thin-walled beams, I: theory. Int J
Solids Struct 25:715749
23. Ho D (1972) The influence of imperfections on systems with coincident buckling loads. Int J
Non-Linear Mech 7:311321
24. Ho D (1974) Buckling load of non-linear systems with multiple eigenvalues. Int J Solids Struct
10:13151330
25. Koiter WT (1945) On the stability of elastic equilibrium. Thesis, Delft. English translation:
NASA TT-F10, 883 (1967) and AFFDL-TR70-25 (1970)
26. Koiter WT (1976) Some properties of (completely) symmetric multilinear forms with an application to Hos theorem for multi-mode buckling. Technologic University of Delft, Lab Rep
587
27. Lanzo AD, Garcea G (1996) Koiter analysis of thin-walled structures by a finite element approach. Int J Numer Methods Eng 39:30073031
28. Lanzo AD, Garcea G, Casciaro R (1995) Koiter post-buckling analysis of elastic plates. Int J
Numer Methods Eng 38:23252345
29. Peek R, Kheyrkhahan M (1993) Postbuckling behavior and imperfection sensitivity of elastic structures by the Lyapunov-Schmidt-Koiter approach. Comput Methods Appl Mech Eng
108(34):261279
30. Poulsen PN, Damkilde L (1998) Direct determination of asymptotic structural postbuckling
behaviour by the finite element method. Int J Numer Methods Eng 42(4):685702
31. Salerno G, Casciaro R (1997) Mode jumping and attractive paths in multimode elastic buckling. Int J Numer Methods Eng 40:833861
32. Schafer BW, Graham-Brady L (2006) Stochastic post-buckling of frames using Koiters methods. Int J Struct Stab Dyn 6(3):333358
33. Silvestre N, Camotim D (2005) Asymptotic-numerical method to analyze the postbuckling
behavior, imperfection-sensitivity, and mode interaction in frames. J Eng Mech 131(6):617
632
34. Sridharan S (1983) Doubly symmetric interactive buckling of plate structures. Int J Solids
Struct 19:625641
35. van Erp GM, Menken CM (1991) Initial post-buckling analysis with the spline finite-strip
method. Comput Struct 40:11931201
36. Wu B, Wang Z (1997) A perturbation method for the determination of the buckling strength
of imperfection-sensitive structures. Comput Methods Appl Mech Eng 145(34):203215
Abstract The behavior of soils reinforced by micropile networks is still not fully
understood due to the lack of accurate modelling capabilities. Particularly, the complex geometry of large soil-micropile systems makes accurate calculation of the
bearing capacity of the reinforced soil a computational challenge. This complexity
requires highly detailed and finely discretized models to achieve reasonable accuracy using direct numerical methods. Such models lead to large scale numerical
optimization problems that are hardly tractable using a personal computer.
In the present paper a model reduction method is made capable of solving the
numerical static limit analysis problem of soil reinforced by a group of micropiles
according to a 2D plane strain model. The method has been successfully applied to
the limit analysis problem of a soil reinforced by a large group of micropiles when
resources did not permit solution of the full model.
1 Introduction
A micropile is a pile with a small diameter (generally in the range 75 to 200 mm) and
high aspect ratio. Micropiles are used in soil reinforcement and foundation works
beneath existing buildings. The micropile technique was developed as early as 1952
by the Fondedile company under the authority of F. Lizzi [1]. Micropiles were used
for the first time in Italy in soil reinforcement of existing buildings and were then
named root piles (pali radice). Within the timeframe of half a century, the technique
Z. Kammoun H. Smaoui
Ecole Nationale dIngnieurs de Tunis, LR11ES16 Laboratoire de Matriaux, Optimisation
et Energie pour la Durabilit, Universit de Tunis El Manar, 1002, Tunis, Tunisia
Z. Kammoun
e-mail: [email protected]
H. Smaoui
e-mail: [email protected]
J. Pastor (B)
LOCIE Laboratory, Polytech Annecy Chambry, University of Savoie, Chambry, France
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_11,
Springer Science+Business Media Dordrecht 2014
229
230
Z. Kammoun et al.
has been applied all over the world [4] with micropile groups counting as many as
1100 micropiles in a landmark example in Neuchtel, Switzerland.
Nevertheless, the behaviour of micropile groups is still not well understood, particularly because of the complex geometry of large soil-micropile systems that challenged the development of accurate modeling methods.
Various approaches are used for predicting the bearing capacity of micropile
groups. Simplified analytical methods [3] are commonly used in engineering practice whereas elastoplastic analysis ([4] and [1]) is often applied in special application and in research. Another alternative is limit analysis [7] by direct methods. The
merit of Limit Analysis (LA) is the rigorous underlying theoretical basis and the
high level of accuracy that may be achieved.
Because of their complex geometry, raisonably accurate prediction for micropile
groups of practical size by Limit Analysis requires finely discretized finite element
models leading to numerical optimization problems that are too large to be directly
tractable by available algorithms [7].
In an attempt to circumvent the problem size difficulty, different techniques have
been devised to reduce the size of the numerical Limit Analysis problem to be
solved. Among these techniques, homogenization methods [2] have been proposed.
While successful in reducing the computational effort these methods do not provide
a realistic description of the stress and strain fields in the heterogeneous medium,
especially near the boundaries of the reinforced zone. Domain decomposition is another approach that is developed for solving large size LA problems. It converts
the original numerical LA problem into a sequence of smaller LA like subproblems
that are solved iteratively. This approach has proven to be successfull in solving
problems that are untractable when solved directly ([8] and [10]).
In this work, an alternative technique is presented that aims at reducing the size
of the numerical LA problem for uniformly spaced micropile groups by taking advantage of the periodicity of the geometry and structure of the reinforced zone. It
is inspired from the case of fiber reinforced composites which is suited to modeling
using periodic homogenization [5].
In this study, a two dimensional representation of the reinforced soil will be
adopted to reduce the numerical problem dimension. Extension to the three dimensional problem will be possible because it is conceptually equivalent to the two
dimensional problem.
The paper begins with a brief presentation of limit analysis followed by a description of the proposed periodic reduction method. The method is then tested and
assessed by applying it to examples of soil-micropile systems.
231
f
,
f ( ) = 0, 0.
(1)
The so-called associated flow rule (1) (or normality law) characterizes the standard
material of LA. Equivalently, a standard material satisfies Hills maximum work
principle (MWP) [6], which states that:
: v 0
PA .
(2)
(3a)
(3b)
232
Z. Kammoun et al.
233
problem. It is inspired from the case of fiber reinforced composites for which a
successful periodic homogenization approach was developed in [5].
Figure 1 shows a typical soil-micropile group configuration with a reinforced
zone supporting a rigid foundation and the natural soil extending on its sides and
beneath, all the way down to a rigid substrate. To apply the reduction procedure the
domain is subdivided into three parts. The first is the central reinforced zone where
the behavior is assumed to be periodic, denoted zone I. The second, the edge zone,
denoted zone II, is a part of the reinforced soil separating the periodic zone from
the domain occupied by the natural soil. Finally, the rest of the soil represents the
zone III. Although geometrically and materialwise periodic, zone II (the transition
zone) is treated as non periodic.
A representative volume element (RVE) is constituted by a micropile and half
the width of soil on each side in addition to the underlying volume of soil.
Regardless of the number of micropiles it includes, the periodic zone is replaced
by a single periodic representative volume element (PRVE) fulfilling built-in periodicity and inter-RVE continuity constraints.
The periodicity conditions imposed on the stress field are
lef t .n = right .n,
(5)
where n is the normal to the right (or the left) side of the PRVE. As the npp periodic
RVEs are replaced by a PRVE, the loading FR of the reduced problem, equivalent
to the original load F (Figs. 2 and 3) is given by
FR = FT + npp .FP
(6)
where FT is the load supported by zone II and FP is the load supported by the PRVE
in the reduced problem.
This results in a considerable reduction in problem size at the cost of an approximation error. Interestingly, the error is on the conservative side, preserving the lower
bound nature of the solution of the static problem. Edge zones are defined by a few
RVEs on each side. The finite element mesh corresponding to these edge zones and
the natural soil (Zone III) remains unchanged. Furthermore, the detailed modeling
of the soil-micropile composite at the RVE level, both in the horizontal and vertical
directions, has the merit of accounting for the toe and lateral effects on the bearing
234
Z. Kammoun et al.
capacity. This is partly because the interaction effect between micropiles can be captured as a result of the full consideration of the soil layers underlying the reinforced
zone.
5 Numerical Examples
The method is applied to some examples of soil reinforcement by groups of micropiles. The optimization problem is solved using the conic programming code
MOSEK for both the direct problem, when possible, and the reduced problem, and
performance is compared.
The LA problem considered (Fig. 4) is that of a Tresca soil reinforced by a group
of np micropiles to support a weighless foundation slab of width b loaded at its
middle by a force F . The soil cohesion is C = 10 kPa and depth is H = 30 m.
The micropiles length is h = 20 m and width d = 0.2 m. The bearing capacity of
the foundation is determined as the maximum load F that, together with a stress
field , form a statically and plastically admissible pair. The associated numerical
optimization problem is denoted P0 .
235
In the reduced problem (Fig. 5), denoted P , Zone I is modeled using a single
PRVE and the load is defined as the sum of the loads supported by the RVEs in
Zone II and the load supported by the PRVE scaled up by the number of micropiles
belonging to Zone I. Since the solution for the reduced problem is admissible for
P0 , it provides a lower bound for the original LA problem. By limiting the number
of transition micropiles to one on each side, the number of micropiles in the model
decreased from np to only 3.
236
Table 1 Number of elements
for different values of
micropile spacing
Z. Kammoun et al.
Spacing
1.8
2.8
3.8
4.8
5.8
6.8
7.8 8.8
149
Fig. 6 Influence of spacing on bearing capacity. Direct and reduced model solutions
micropiles. The limit load is calculated for different values of micropile spacing.
Figure 6 shows the limit load of the reinforced soil, calculated with different methods, as a function of spacing. The blue line represents the kinematic solution (upper
bound) of the reference problem. The red line represents the static solution (lower
bound) of the reference problem. The yellow line corresponds to the solution of the
reduced model problem.
The results for e < 6.8 were all obtained with the same degree of discretization (elements size) which did not permit the direct solution beyond that spacing.
Therefore, the problems with e > 6.8 were solved with fewer, larger elements. It
should however be stated here that for the case e = 7.8 the results were not shown
because the case was simply not treated. The reason was that it was not possible to
create a regular mesh with the large element size because of the particular geometrical dimensions of the reinforced soil. The number of elements for each spacing is
indicated in Table 1.
It may be noted from the results that:
The reinforced soil bearing capacity increases with spacing for spacing under
6.8 m. Beyond this value the bearing capacity saturates and remains nearly indifferent to spacing. The saturation should reflect the vanishing of the interaction
among micropiles which tend to behave as isolated inclusions.
The error between the direct static and the reduced model solution is relatively
small (less than 4.2 %).
237
The solution stress field is visualized in Fig. 7 for selected values of spacing. It
can be seen that, for small spacing, the behavior of the reinforced area is reminiscent of a block mechanism (Fig. 7(a)). For large spacing (Fig. 7(c)), the elementary
volumes tend to behave independently as if the micropiles were isolated. From the
stress field in Fig. 7(c) a pattern can be seen that is characterized by a localization
of the failure zone in a thin volume of soil surrounding the micropile.
238
Z. Kammoun et al.
For large spacing, the behavior does not clearly reflect the assumption of periodicity. The stress distribution for e = 8.8 in Fig. 9(c) looks more like that in Fig. 7(b)
(for e = 5) than the nearly periodic stress field shown in Fig. 7(c), obtained for
the same spacing e = 9 when the load is supported by the micropiles only.
239
240
Z. Kammoun et al.
95760
419.7
249
98100
432.9
1078
103320
1263.4
232
114300
1304.7
1998
133560
2957.5
354
126000
2938.4
315
0.65
146700
3059.9
1845
148680
3800.3
204
126000
3779.2
273
0.56
162900
3944.1
1699
11
163800
4651.4
360
126000
4619.6
350
0.68
167160
4809.5
1948
15
194040
6342.8
402
126000
6301.0
352
0.66
197400
6580.7
1833
19
224280
8037.6
459
126000
7982.4
281
0.69
227640
8320.8
2299
21
239400
8884.6
446
126000
8822.9
257
0.69
31
126000
13026
327
The boundary conditions are such that the load is carried by the micropiles only.
The same degree of discretization, i.e. in terms of size of finite elements, is used in
all the models. The reduced model counts 126,000 finite elements regardless of the
number of micropiles.
From the results, summarized in Table 2, it is seen that, as expected, the CPU
time required by the reduced model solution has no clear tendency to increase with
the number of micropiles, whereas the CPU time of the Direct solution increases
with it nearly proportionally.
For a reinforcement with 21 micropiles (Table 3) the mesh of the Direct problem
model amounts to 239,400 finite elements, nearly twice the number of elements in
the reduced model, and the consumed CPU time is almost double the CPU of the
reduced model solution for an accuracy gain of 0.7 %. This is the largest problem
for which the Direct solution was possible with the Mac Pro 3 GHz machine used
in this work.
For the same number of micropiles and using a finer mesh with 277200 elements
in the Direct problem model, the Direct solution fails to converge whereas the reduced model solution converges in twice the CPU time and improves the reduced
lower bound by 0.15 %.
The relative error between the Direct and the Reduced Model solutions is between 0.5 and 0.7 % and does not appear to increase with the number of micropiles.
Method
Nbr of elements
F (kN)
241
CPU (s)
Static
239400
8884.6
446
Quasi-periodic
126000
8822.9
257
Quasi-periodic
277200
8835.9
530
Consequently, it may be concluded that the reduction method provides a fairly accurate estimate for an unlimited number of micropiles within a nearly constant computational effort.
6 Conclusion
A model reduction method is proposed to solve the numerical static limit analysis
problem of a composite medium, characterized by periodic reinforcement, embedded in a homogeneous domain, while preserving the fineness of the Finite Element
description of the Representative Volume Element. The reduction method has been
successfully applied to the Limit Analysis of a soil reinforced by a large group
of micropiles when resources do not permit solution of the full model problem.
Numerical results demonstrate that the reduction method provides a fairly accurate
estimate of the limit load for an unlimited number of micropiles within a nearly constant computational time. Significant differences in behavior and bearing capacity
are observed depending on the way the applied load is distributed between the soil
and the micropiles. When the load is supported solely by the micropiles the reduced
model results in terms of limit load of individual micropiles are very close to the
reference solution (in confirmation of the periodicity assumption). When the load is
supported by both the soil and the micropiles the error is larger than when only the
micropiles carry the load. In a future work the reduction method will be extended to
more general periodic representative volume elements by relaxing the symmetry requirement and allowing some forms of controlled variability of the stress field in the
PRVEs. This will lead to more accurate solutions at the cost of a little extra computational effort. Another extension that might improve the accuracy of the reduction
method, consists in limiting the length of the RVE to the height of the micropile
allowing more degrees of freedom in the soil beneath the reinforced zone.
References
1. FOREVER (2004) Synthse des rsultats et recommandations du projet national sur les micropieux. Presses de lEcole Nationale des Ponts et Chausses, Paris
2. Abdi R, de Buhan P, Pastor J (1994) Calculation of the critical height of a homogenized
reinforced soil wall: a numerical approach. Int J Numer Anal Methods Geomech 18:485505
3. Ministre de lquipement, du logement et des transports (1993) Fascicule no 62. Titre v:
Rgles techniques de conception et de calcul des fondations des ouvrages de gnie civil
242
Z. Kammoun et al.
4. Estephan R (2003) Contributions aux mthodes de calcul des groupes et des rseaux de micropieux. Thse de doctorat, Ecole Nationale des Ponts et Chausses, 25 juin
5. Turgeman S, Pastor J (1987) Comparaison des charges limites dune structure htrogne et
homognise. J Mc Thor Appl 6(1):121143
6. Hill R (1950) The mathematical theory of plasticity. Oxford engineering science series.
Clarendon, Oxford
7. Kammoun Z (2010) Prvision de la charge limite des sols renforcs par rseaux de micropieux.
Thse de doctorat, Universit de Savoie et Ecole Nationale dIngnieurs de Tunis
8. Kammoun Z, Pastor F, Smaoui H, Pastor J (2010) Large static problem in numerical limit
analysis: a decomposition approach. Int J Numer Anal Methods Geomech 34:19601980
9. MOSEK ApS (2002) C/O Symbion Science Park, Fruebjergvej 3, Box 16, 2100 Copenhagen , Denmark. www.mosek.com.
10. Pastor F, Loute E, Pastor J (2009) Limit analysis and convex programming: a decomposition
approach of the kinematical mixed method. Int J Numer Methods Eng 78(3):213219
11. Pastor J (1978) Analyse limite: dtermination numrique de solutions statiques compltes.
Application au talus vertical. J Mc Thor Appl 2:167196
12. Salenon J (1967) Thorie des charges limites: poinonnement dune plaque par deux
poinons symtriques en dformation plane. C R Acad Sci Paris 265:869872
13. Salenon J (1974) Thorie de la plasticit pour les applications la mcanique des sols. Eyrolles, Paris
14. Trillat M, Pastor J (2005) Limit analysis and Gursons model. Eur J Mech A, Solids 24:800
819
1 Introduction
During the earlier phases of development of civil nuclear power in the UK, a number of phenomena were identified that arose from the unique and extreme conditions
materials and structures were required to withstand. For those with interest in Direct
Methods the best known example is Brees complete solution for a thermally cycled
tube [13]. Brees solution illustrates the nature of structural ratchetting whereas it
was observed within the earliest reactors that polycrystalline uranium exhibited two
interesting modes of strain growth that related to mechanisms occurring on the material micro-scale. The first was Cottrell Creep resulting from neutron irradiation
and the second was cyclic strain growth due to temperature variations within the
243
244
Fig. 1 Steel test specimen repeatedly temperature cycles through a phase change. From Greenwood [5]
temperature range from room temperature to 668 C, the range for which uranium
has the alpha crystal structure.
As the original work was carried out as part of the development of civil nuclear
reactors, the resulting reports [1, 2] were subject to the UK Official Secret Act and
received very limited circulation at the time. The symposium paper [3] summarised
their results, but the details of the analysis were not included. Subsequent workers
in the field seem to have relied predominantly on the brief appendix to Greenwood
and Johnsons [4] paper on the adaptation of the theory for metals that suffer a phase
change during thermal cycling.
The reports [1, 2] were declassified in the early 1980s and may now be accessed
at the UK National Archive at Kew, London. The Archive has agreed to their general
availability and they may be obtained from the web-site listed in the foot-note to the
references.
In the following sections we discuss the two modes of behaviour discuss by Anderson and Bishop and the nature of the analysis used.
I
y
(1)
245
1 0 0
g
ij = g 0 1 0
(2)
0 0 0
where g is proportional to I and the axis (x, y, z) = (b, a, c) with respect to the
crystal structure. In a polycrystal, the random orientation of adjacent crystals results
in an overall strain growth of zero as there is no preferred direction, provided the
b directions are completely randomly orientated. The strains resulting from Eq. (2)
p
g
p
are offset by plastic strain rate ij , so that in a single crystal ij = ij ij = 0 and
there is zero overall strain growth. The polycrystal is a seething mass of plastically
straining crystals. In their analysis Anderson and Bishop adopted the von Mises
yield condition with the associated Prantl-Reuss flow rule, following the same asg
sumptions as previously adoptd by Bishop and Hill [9]. Hence, with ij prescribed
for each crystal, the state of stress at yield is also prescribed and assumed constant
within each crystal. Consider now applying an external small macro stress ij ,
1 0 0
ij = 0 0 0 X
(3)
0 0 0 X
i
in this case referred to fixed external axis (X, Y, Z). The resulting macro strain rate,
as the polycrystal is isotropic, is given by
1
0
0
0 E X .
E ij = 0 1/2
(4)
0
0
1/2 X
i
Anderson and Bishop assumed that all crystals experienced this same total strain
rate and that the stress in each crystal was constant within each crystal but varies
246
Fig. 3 Schematic
representation of source of
strain growth in Cottrell
Creep. The imposition of the
macro strain rate E ij causes a
movement of the yield point
by stress ij
from crystal to crystal. Hence the circumstance within an individual crystal may be
g
understood from Fig. 3. The plastic strain ij , Eq. (2) and the strain growth Eq. (4)
will not be consistent with each other when transformed to a common set of axes
(X, Y, Z). The imposition of E ij causes the stress at yield to move around the yield
surface by an amount ij defined by the flow law. The analysis consists of using the Prandl-Reuss flow rule to compare the states of stress resulting from plastic
p
g
p
g
strain rates ij = ij and ij = ij + E ij and then expanding the difference for
small E x compared with g to obtain the stress difference at yield, ij . The average of ij over-all crystals then gives x . Anderson and Bishop [1] complete
this analysis and obtain
5 x
E x =
g
2 3 y
(5)
for x small compared with y . Although the authors do not consider conditions of equilibrium formally, relying entirely on the kinematics of the situation,
conditions of equilibrium consistent with the kinematic assumptions may be easily
applied to the analysis using the Principle of Virtual Work yielding the identical
result (5) (see Ponter and Cocks [19]). Hence, within the limitations of the model,
result (5) may be regarded as correct, as it has been for the last 60 years.
247
ever, if a macro-stress X , Eq. (3), is applied, a cyclic strain growth Ex per
cycle occurs, given by the general form
T Ts
x
(6)
Ex = B
E
Ts
where B is a numerical constant and E denotes Youngs modulus. Such behaviour
was also shown in test on helical springs suspended under weights. Although the
correlation was by no means exact the linearity with stress and the existence of a
critical temperature difference threshold was confirmed. Cottrell predicted the form
of Eq. (6) in a Note, now lost, on the basis of the behaviour of a two bar structure
(see below). Although the arguments he used were clearly suspect from the point
of view of mechanics, Anderson and Bishop were not entirely surprised to discover
that, on detailed analysis, Alan Cottrell was, again, correct.
The property of the crystal that gives rise to this phenomenon is the strongly
anisotropic thermal expansion coefficients. In terms of axis (x, y, z) = (b, a, c) local
to the crystal, the coefficients of thermal expansion are given by
a = c = 30 108 C1 ,
b = 5 108 C1 .
(7)
0
0
1
0
ij = (2/3) 0 1/2
,
0
0
1/2
= (a b ) = 35 108 C1 .
(8a)
(8b)
(8c)
0
0
1
0
ij = t 0 1/2
,
(9)
0
0
1/2
where t = 2ET /(1 + ). The elastic limit is then given by the von Mises
yield condition as 3/2t = y . A small amount of plastic strain then allows T to
increase to a reverse plasticity limit where the elastic stress varies over a range of
twice yield and T = TS where,
t = 2ETs /(1 + ) = 4/3y .
(10)
248
For T > Ts , a state of reverse plasticity is set up in each crystal and the polycrystal shows no cyclic strain growth. Each crystal suffers equal and opposite plastic
strains
0
0
1
p
0
ij = 0 1/2
p .
(11)
0
0
1/2
The question now arises: what is the growth of strain per cycle if a small external
constant stress x is applied? Anderson and Bishop [1, 3] adopted a form of
analysis that closely follows that used for Cottrell Creep. They assumed that equal
strain increments occur at each extreme of the cycle, given by replacing Eq. (2)
by Eq. (11), but following the same averaging process. Again, the arguments are
entirely kinematic and the details of the internal stress are not considered. Their
analysis produced the answer
x 32 T Ts
Ex =
.
(12)
E
27
Ts
This solution, and various adaptations, has been widely used over the intervening
years for a variety of other systems where internal thermal expansion strains occur.
Greenwood and Johnson [4] adapted this theory for cycling through phase changes,
and it is the Appendix to this paper that appears to be the main source for subsequent
authors in material science. The Anderson Bishop results are much quoted in the
literature (e.g. [58, 1012]) and have been refined in terms of micro-models, see,
for example Leblond [6, 7] and Taleb and Sideroff [8]. Recently Greenwood [5]
has given a review of subsequent work in this area. Ratchetting in metal matrix
composites was identified by Daehn and Gonzalez-Doncel [11] as being a related
phenomenon and the functional form of the Anderson Bishop analysis was adopted.
Other related work on metal matrix composites is summarised by Clyne and Withers
[12]. There are certainly in excess of 100 papers in the literature which uses this
result and also use Andersons extension of the work to creep [2].
It should be emphasized that Anderson and Bishop did not regard their analysis as a detailed realistic micro-model of the polycrystal. The assumption of a von
Mises yield condition is certainly appropriate for the overall behaviour of a polycrystal, but not necessarily appropriate for a single crystal. The model is essentially phenomenologicaldescribing in general terms the dependence of the strain
on governing quantities but probably not reliable in detail. Hence details of noncompliance of the theory with experimental data has not been regarded as a barrier
to its use.
Those readers familiar with thermal loading problem of this nature may well have
already wondered if the Anderson Bishop analysis is the complete answer. Ratchetting in structures due to thermal cycling is common. Ponter and Cocks [14, 15]
looked at ratchet rates for such problems and derived relationships of a similar form
to (12). But the source of ratcheting in the structural cases came from a very different mode of behaviour. The kinematic restrain argument of Anderson and Bishop
do not apply but ratcheting arises from changes in residual stress during the cycle,
249
allowing cyclic strain growth due to transfer of load from one part of the structure
to another. It is worth asking the question if both sources of ratchet strain growth
is present in this problem, that considered by Anderson and Bishop and the Breelike ratchetting of structures. A particular situation may easily be considered where
the Anderson Bishop strain growth does not apply, a two bar structure, and this is
discussed in detail in the next section.
2 = p + t ,
where t = ET /2
(13)
Elastic Limit.
(14)
The full analysis for this problem is summarised in Table 1 where the complete solution is shown for conditions of elastic behaviour, shakedown, reverse plasticity and
ratchetting. The overall behaviour is summarised in the interaction diagram shown
250
Constant i Varying ir
Total i
ps + t
t /2
ps + t /2 = y
T = T0 + T
2
ps t
t /2
y ps t /2 y
T = T0
1
ps
t /2
y ps t /2 y
T = T0
2
ps
t /2
ps + t /2 = y
ps + p + t t /2
p
ps + t /2 = y
T = T0 + T
2
ps + p t t /2
+p
y ps + p t /2 y
T = T0
1
ps + p
t /2
+p
y ps + p t /2 y
T = T0
2
ps + p
t /2
p
ps + t /2 = y
+t
t /2
+(y t /2)
+y
T = T0 + T
2
t /2
(y t /2)
T = T0
1
t /2
(y t /2)
T = T0
2
t /2
+(y t /2)
+y
p + t
t /2
+(y t /2) p +y
T = T0 + T
2
p t
t /2
(y t /2) + p y y + 2p y
T = T0
1
p
t /2
(y t /2) + p y y + 2p y
T = T0
2
p
t /2
+(y t /2) p +y
(15)
251
(16)
where Ts is the temperature difference corresponding to t = 2y . Note the primary difference between (15) and (16). We find, for the polycrystalline problem,
the solution consist of a combination of the two modes of behaviour exemplified by
Eqs. (15) and (16).
3 Polycrystalline Models
The two bar structure, despite its simplicity, shows all the characteristic modes of
behaviour of all such problems, but excludes the precise phenomenon identified by
Anderson and Bishop as the cause of ratchetting in alpha-uranium. As a result, the
behaviour of a polycrystal may be expected to show both forms of ratcheting. The
associated analysis is, therefore, rather more complex and will not be discussed in
detail here. Effectively the problem may be solved by combining the arguments of
Anderson and Bishop [1, 3] but taking into account the fact that the residual stress
within each crystal is not constant but varies. Plastic strains in reverse plasticity
will always occur at both ends of the cycle, but the additional plastic strains that
contribute to the ratchet mechanism when stress is applied may only occur at one
252
instant in the cycle. A detailed analysis (Ponter and Cocks [19]) of the structure
of the problem shows this to be the case. Two models are considered. The first
simplifies the crystal structure by assuming that all crystals have their b direction
orientated along each of the axis of a fixed set of axis, a third in each direction. In
the following we refer to this as the three element model. This simplification allows
a more complete analytic solution than in the fully anisotropic case.
The full interaction diagram, for comparison with Fig. 5, is shown in Fig. 6 for the
three element model. There are many similarity, except that the shakedown boundary at T /TS = 1, the reverse plasticity limit, meets the x /y = 0 axis at a right
angle. This is associated with a rate of increase of ratchet strain with increases in
load beyond shakedown becoming zero. This is unlike the two bar structure, Fig. 5
where this angle is obtuse and the ratchet rate remains constant.
For T /TS > 1 both the three element model and the polycrystal model may
be evaluated for comparison with Anderson and Bishop. Although Anderson and
Bishop did not consider the three element model, the analysis with their assumptions
may easily be carried out. A comparison between all the solutions is shown in Fig. 7.
There a number of features of Fig. 7 of immediate interest. The solutions for
the three element and the fully isotropic models are very similar. This indicates that
analysis of the much simpler three element model provides a close approximation
to the more complex case, providing a useful route to future studies using more
complex behaviour, such as creep. The other, perhaps surprising, result is that the
solution that includes both ratcheting phenomenon gives lower ratchet rates than
the Anderson and Bishop analysis, which only includes one. The reason for this is
that Anderson and Bishop assumed that a contribution to the ratchet strain occurred
throughout the polycrystal at both extremes of the temperature cycle whereas, due
to changing residual stress fields, it only occurs at one extreme. Hence Anderson
and Bishop in this respect overestimated the ratchet strain but underestimated it by
leaving out conventional structural ratcheting. Note that at T /TS = 1the ratchet
rate is zero, whereas for the two bar structure it has a high value.
253
Fig. 7 Comparison of
Anderson Bishop and present
analysis. Solidpolycrystal,
dashedthree element model
4 Conclusion
The Anderson Bishop [13] model for the linear plastic ratchetting of polycrystals
with anisotropic crystal thermal expansion properties when subjected to cyclic temperature and Greenwoods [4, 5] adaptation of the model for alloys cycled through
a phase change have remained unchallenged since the early 1960s. Neither model
was put forward as an exact model of behaviour and various assumptions may be
challenged. Crystals were assumed to be isotropic in both their elastic and plastic
behaviour and satisfy the von Mises yield condition and associated perfectly plastic
flow condition. The total microstrain was assumed to remain the same as a uniform
macrostrain, although this has remained an assumption in subsequent work. The
material constants are assumed independent of temperature. The models should,
perhaps, be sufficient to demonstrate the relationship between the observed superplasticity behaviour and the micromechanical phenomena. Anderson and Bishop
[13] demonstrated that the phenomenon may be explained through the movement
of the yield point around the yield surface, constrained by overall kinematic conditions and the associated flow law. They give a resulting equation (12) where the
accumulation of plastic strain for each cycle of temperature increases linearly with
applied stress and temperature variation. The purpose of this paper is to show that
a separate micromechanics phenomena also makes a contribution, the redistribution
of stress through a varying residual stress. This results in plastic ratchet strains, in
an individual crystal, occuring only at one extreme of the thermal cycle. Hence, at
the two extremes a proportion of the crystals suffer plastic strain growth while the
remainder remain elastic. At the opposing end of the cycle the reciprocal set of crystals experience plastic ratchet strains. This behaviour may be demonstrated in a two
bar structure. The effect on the resulting prediction is to reduce the ratchet rate significantly compared to the Anderson Bishop analysis as shown in Fig. 7. The new
solution shows a lower rate of ratchetting.
254
The Anderson Bishop solutions were not expected to give close predictions of
behaviour and, of course, neither does this new solution Eq. (17), below (Ponter
and Cocks [19]). The primary purposes of this paper bring to the notice of those
workers in Direct Methods that this neglected ratchet phenomenon exists and remind
future workers in that if more detailed micromechanics is attempted of this class of
problems then phenomena associated with changing residual stress field exists. In
the three cases discussed here, the two bar structure, the three element problem and
the polycrystal problem, all the solutions have the same overall form (Ponter and
Cocks [19]):
(T Ts )/Ts
x
(17)
Ex =
E
(a + b(T Ts )/Ts )
where for:
Two Bar structure: a = 0, b = 1/4.
Three Element Model: a = 27/32, b = 1/4.
Polycrystal Model: a = 4/5, b = 0.244.
The Anderson Bishop model corresponds to b = 0.
The analysis does not include the effects of high temperature creep, the subject
of Andersons second report [2]. This problem is more complex and has not been
discussed in this paper. Creep strain rates vary very rapidly with temperature and it
seems likely that most of such problems will involve both plastic and creep strains
within the cycle of temperature, an aspect not considered by Anderson[2], indicating
that the creep problem may now be usefully looked at again, following the results
of this paper
Acknowledgements The first author first became aware of the Anderson and Bishop solutions
when working with Roger Anderson on the modelling of a similar phenomena in the irradiation
induced swelling of structural steels [20]. In recent years, following an interest in the thermal
properties of metal matrix composites it became clear that the solutions have been extensively
applied to problems in material science.
The authors wish to thanks Roger Anderson for assistance in this work. The work for this
paper was part support by a Leverhulme Emeritus Fellowship to the first named author, which is
gratefully acknowledges.
References
1. Anderson RG, Bishop JFW (1957) United Kingdom Atomic Energy Authority Industrial
Group report TN/C681, UK National Archive reference AB 7/7441. Copy is available
with the permission of the National Archive for downloading from http://www.le.ac.uk/
departments/engineering/people/academic-staff/alan-ponter/papers
2. Anderson RG (1958) United Kingdom Atomic Energy Authority Industrial Group report TN/C681, UK National Archive reference AB 7/ 6213. Copy is available with
the permission of the National Archive for downloading from http://www.le.ac.uk/
departments/engineering/people/academic-staff/alan-ponter/papers
255
3. Anderson RG, Bishop JFW (1962) The effect of neutron irradiation and thermal cycling on permanent deformations in uranium under load. In: Proceedings of the institute of metals symposium on uranium and graphite, Institute of Metals, London, paper 3.
Copy is available with the permission of the National Archive for downloading from
http://www.le.ac.uk/departments/engineering/people/academic-staff/alan-ponter/papers
4. Greenwood GW, Johnson RH (1964) The deformation of metals under stresses during phase
transformation. Proc R Soc A 283:403422
5. Greenwood GW (2010) Generation of internal stress and its effects. Mater Sci Technol
26(4):398403
6. Leblond JB, Devaux J, Devaux JC (1989) Mathematical modelling of transformation plasticity
in steels, I: case of ideal-plastic phases. Int J Plast 5:551572
7. Leblond JB (1989) Mathematical modelling of transformation plasticity in steels, II: coupling
with strain hardening phenomena. Int J Plast 5:573591
8. Taleb L, Sidoroff F (2003) A micromechanical modeling of the GreenwoodJohnson mechanism in transformation induced plasticity. Int J Plast 19(10):18211842
9. Bishop JFW, Hill R (1951) A theory of the plastic distortion of a polycrystalline aggregate
under combined stress. Philos Mag 42:1298
10. Hutchinson JM, McCrum NG (1974) Microstress mechanisms for the time dependence of
the modulus of crystalline polymers following an imposed change in volume. Nat Phys Sci
270:295296
11. Daehn GS, Gonzalez-Doncel G (1989) Deformation of whisker-reinforced metal matrix composites under changing temperature conditions. Metall Trans 20A:23552368
12. Clyne TW, Withers PJ (1993) An introduction to metal matrix composites. Cambridge University Press, Cambridge, Chap 5
13. Bree J (1967) Elasto-plastic behaviour of thin tubes subjected to internal pressure and intermittent high-heat fluxes with applications to fast reactor fuel elements. J Strain Anal 2(3):226
238
14. Ponter ARS, Cocks ACF (1984) The incremental strain growth of an elastic-plastic body
loaded in excess of the shakedown limit. J Appl Mech 51(3):465469
15. Ponter ARS, Cocks ACF (1984) The incremental strain growth of elastic-plastic bodies subjected to high levels of cyclic thermal loading. J Appl Mech 51(3):470474
16. Goodall IW, Goodman AM, Chell GC, Ainsworth RA, Williams JA (1991) R5: an assessment procedure for the high temperature response of structures. Report, Nuclear Electric Ltd.,
Barnwood, Gloucester
17. Chen HF, Ponter ARS, Ainsworth RA (2006) The linear matching method applied to the high
temperature life assessment of structures, part 1: assessments involving constant residual stress
fields. Int J Press Vessels Piping 83:123135
18. Chen HF, Ponter ARS, Ainsworth RA (2006) The linear matching method applied to the high
temperature life assessment of structures, part 2: assessments beyond shakedown involving
changing residual stress fields. Int J Press Vessels Piping 83:136147
19. Ponter ARS, Cocks ACF (2013) Thermal ratchetting of polycrystalline metals with inhomogeneous thermal properties. Philos Mag, accepted May 2013, available online
20. Anderson RG, Ponter ARS (1972) An estimate of the Cottrell creep of a metal during swelling.
In: British nuclear energy society conference on irradiation embrittlement and creep in fuel
cladding and core components, London
1 Introduction
Imperfections in structures can arise in the initial production process, or during the
heat-treatment of the component, particularly during welding processes. These defects or flaws are unavoidable within structure components, and they do not generally lead to an immediate failure. Failure modes occurring from these structures
are different from industry to industry, but mostly such failures result from the application of cyclic loading with high temperature. In general, the lifetimes of these
components, operating at elevated temperatures, depend on the nature of plastic
H. Chen (B) W. Chen
Department of Mechanical and Aerospace Engineering, University of Strathclyde,
Glasgow G1 1XJ, UK
e-mail: [email protected]
K. Spiliopoulos, D. Weichert (eds.), Direct Methods for Limit States in Structures and
Materials, DOI 10.1007/978-94-007-6827-7_13,
Springer Science+Business Media Dordrecht 2014
257
258
and creep deformation they experience. The first failure mode is mainly concerned
with the excessive plastic deformations associated with the phenomenon of plastic
collapse, shakedown and ratchetting, while the second is concerned with the creep
fatigue interaction. The ability to accurately model these behaviours of component
subjected to cyclic and variable loading conditions would provide a means of assessing the remaining life of the structural components. Thus, the elastic-plastic-creep
response of a structure needs to be well understood when using it as a design condition.
The determination of these design limits has attracted the attentions of many researchers. The phenomena of shakedown and ratchetting associated with the steady
cyclic response have been researched and modeled extensively by plasticity theorists, materials scientists, mathematicians and engineers. Since closed form solutions of these design limits are very limited due to the complexity of the problem,
the numerical approaches play a key role for the assessment of these design limits
in plasticity.
One approach is to simulate the detailed elastic-plastic response of the structure for a specified cyclic load history, most commonly by the incremental Finite
Element Analysis (FEA) [1]. However, this method requires significant computer
time for complex structures, due to the reason of its investigation of any load cycle. A relatively new cyclic analysis method, Direct Cyclic Analysis (DCA) [2], has
been developed to avoid excessive numerical expense associated with the incremental FEA. It has been recently incorporated into ABAQUS to evaluate the stabilized
cyclic behaviour directly. However, both the incremental FEA and DCA do not predict shakedown or ratchet limits directly. It can only be used to show whether elastic
shakedown, plastic shakedown or ratchetting occurs [3].
To define the shakedown and ratchet limits, alternative approach has been developed. It involves the application of numerical methods [49] for addressing the
structural response in structures subjected to both severe mechanical and thermal
loads. The assessments, provided from these new methods, have the potential of
providing results that combine the accuracy of non-linear FEA simulation methods [10, 11] with the efficiency of rules-based methods [12, 13]. These are direct
methods based upon a programming technique. Direct methods were incorporated
into finite element analysis in order to evaluate the shakedown limit. The material
model is considered to be elastic perfectly plastic, and the load domain including all
the possible load paths eliminates the necessity to know the detailed load history.
Such direct methods include; the mathematical programming methods [1416], the
Generalized Local Stress Strain (GLOSS) r-node method [17], the Elastic Compensation Method (ECM) [18], and the Linear Matching Method (LMM) [7, 19, 20].
Among these direct methods, the LMM is counted to be the method most amenable
to practical engineering applications involving complicated thermomechanical load
history. The LMM has been extensively applied to a range of problems [8, 19],
through various adaptations, extended to the calculation required for the UK assessment procedure R5 [21] for the high temperature response of structures. The
LMM describes non-linear inelastic material behaviour by linear solutions where
the material coefficients vary both spatially and in time, which makes the method
259
particularly flexible. The LMM has been regarded as an efficient and effective upper
bound programming method for which, in many circumstances, strict convergence
proofs may be constructed. In the past two years, the LMM has been further developed to account for the lower bound shakedown and ratchets limits, and investigate
more complicated cyclic problems. Moreover, the extensions of LMM have resulted
in the application of the method to high temperature creep behavior including the
effect of elastic follow-up [20], i.e. the evaluation of local creep damage due to the
relaxation of stress during creep dwell times.
There are four objectives leading to using the basics of these methods in this
study where limit in plasticity and creep are designed. The first objective is to obtain a LMM approach used for more general purposes. This LMM must be able to
address to a wider class of problems and prospective procedures for lower and upper bound design limits. The second objective is to assess the cyclic response under
creep fatigue conditions by presenting a new numerical procedure. The third is the
examination and the improvement of convergence problems existing in the iterative approach and the last objective is to validate the efficiency and effectiveness
of LMM while designing the limits in plasticity and creep. This validation is performed on three typical practical problems. The first problem is; a defective pipeline
subjected to cyclic thermal loading and constant operating pressure. In the second
problem the effects of drilling holes on the ratchet limit and crack tip plastic strain
range for a central cracked plate subjected to constant tensile loading and cyclic
bending moment are investigated, and in the last one the cyclic structural responses
of a cruciform weldment under creep fatigue interaction is being addressed.
In the following sections, a general cyclic minimum theorem for perfect plasticity and the application of the LMM for a particular class of problems for the design
limits in plasticity and creep will be described. This is followed by the discussion
of convergence and the application of three practical examples with numerical verifications of the proposed methods.
2 Cyclic Behaviour
2.1 General Cyclic Problem
Consider a body with volume V and surface S, where the material is isotropic,
elastic-plastic and satisfies the von Mises yield condition. A cyclic history of temperature (x, t) occurs within volume V . A cyclic load history P (x, t) is applied
over part of S, namely ST . Here denotes a scalar load parameter. On the remainder
of S, namely Su , zero displacements are maintained. Both load and temperature histories have the same cycle time t and, in the following, we are concerned with the
behaviour of the body in a typical cycle 0 t t in a cycle state. For the problem
defined above the stresses and strain rates will asymptote to a cyclic state where
ij (t) = ij (t + t),
ij (t) = ij (t + t).
(1)
260
This arbitrary asymptotic cyclic history may be expressed in terms of three components, the elastic solution, a transient solution accumulated up to the beginning
of the cycle and a residual solution that represents the remaining changes within the
cycle. The linear elastic stress solution is denoted by ij . The general form of the
stress solution is given by
ij (x, t) = ij (x, t) + ij (x) + ijr (x, t)
(2)
where ij denotes a constant residual stress field in equilibrium with zero surface
traction on ST and corresponds to the residual state of stress at the beginning and
end of the cycle. The history ijr is the change in the residual stress during the cycle
and satisfies;
ijr (x, 0) = ijr (x, t).
(3)
It is worth noting that the arguments in this section do not explicitly call on
the properties of perfect plasticity and are therefore common to all cyclic states
associated with inelastic material behaviour.
261
Elastic Shakedown Region In this region, the stresses are exceeding the yield
stress at the first few load cycles, which give rise to constant residual stress in the
structure such that in subsequent load cycles only elastic deformation occurs. The
constant residual stress field has caused the redistribution of the stresses within the
structure. This effectively has the effect of pulling the stress fields, the sum of the
elastic and residual stresses in to the yield surface.
Reverse Plasticity/Plastic Shakedown Region The transition to this region occurs when the effective elastic stresses exceed twice the yield stress. This was made
possible with the accommodation of the time-varying residual stress field, causing
the stress distribution at the outer fibre of the plate, to exceed twice the yield stress.
When the structure exhibits reverse plasticity over each cycle, the positive plastic
strain in the first half of the load cycle followed by equal magnitude negative strain
in the second half, such that there is no accumulation of plastic strain during load
cycle. And the failure mechanism for plastic shakedown is low-cycle fatigue.
Ratchetting Region This region is best characterized by the breakdown of the
elastic, shakedown and reverse-plasticity conditions. In each cycle, plastic strains
accumulate over a significant volume of the plate, leading to structural failure from
the unlimited accumulation of plastic deformation and eventually incremental plastic collapse.
262
n
, =
I n ij
ij (tn ) = ij0 +
(4a)
n n
n
ij ij ij (tn ) + ij (tn ) + ij ij
dV ,
V
n
l=1
ij (tl ),
(4b)
(4c)
263
ij0 M
ij n1
n=1
ij n2
+ +
n=1
ij nM1
(5)
n=1
where M represents the total number of cycles. In this section, the linear matching
c , ) are summarized for both the shakedown and
processes for minimization of I (ij
ratchet limits.
(7)
(8)
where the upper dash refers to deviatoric components. Summing over the cycle
%
(k+1)
n(k+1)
produces a relationship between the compatible strainij
= n ij
and
the constant residual stress ijk+1 with an initial stress state;
(k+1)
ij
1 initial
ij
+ ijk+1 ,
k
2
(k+1)
kk
=0
(9a)
where
1
1
=
k
nk
n
and ijinitial = k
ij (tn )
n
nk
(9b)
The solution of the continuum problem corresponding to Eqs. (9a), (9b) has the
(k+1)
k , ), which is proved by [6].
property that I (ij , ) I (ij
264
1
n(k+1)
n(k+1)
+ ij
,
2 ij
1
T n(k+1)
n(k+1)
kk
=
,
3K kk
1
n(k+1)
n(k+1)
ij
(tn ) + ij (tn1 ) + ij
=
,
2 nk ij
T n(k+1)
ij
(10a)
(10b)
(10c)
ij (t0 ) = ij0 .
(11)
The entire iterative procedure requires a number of cycles, where each cycle contains N iterations associated with N load instances. The first iteration is to evaluate
265
the changing residual stress ij1 associated with the elastic solution ij (t1 ) at the
first load instance. Define ij nm as the evaluated changing residual stress for nth
load instance at mth cycle of iterations, where n = 1, 2, . . . , N and m = 1, 2, . . . , M.
At each iteration, the above changing residual stress ij nm is calculated. When the
convergence occurs at the Mth cycle of iterations, the summation
of changing resid%
n = 0) due to the
ual stresses at N time points must approach to zero ( N
ij M
n=1
stable cyclic response. Hence the constant residual stress ij (t0 ) = ij0 over the cycle
can also be determined by
ij0 =
ij n1 +
n=1
ij n2 + +
n=1
ij nM .
(12)
n=1
1
ij (tn ) + ij (tn )
n
2
(13)
where n is the iterative shear modulus and ij (tn ) is the converged accumulated
residual stress at the time instant tn , i.e.
ij (tn ) = ij0 +
ij kM .
(14)
k=1
I ij , S =
N
n n
n
ij ij S ij (tn )ij
dV 0,
(15a)
V n=1
i.e.
%N
n
n
n=1 ij ij )dV
%N
n )dV
ij (tn )ij
V ( n=1
V(
%N
n
n=1 (ij ))dV
V (y
%N
n )dV
ij (tn )ij
V ( n=1
= SU B .
(15b)
266
For the upper bound ratchet limit, the numerical technique can be accommodated
within the existing methods of shakedown analysis where the linear elastic solution
is augmented by the changing residual stress field, i.e.
(16)
where the history of the residual stress field ij (tn ) associated with the cyclic component of the load history has been calculated by an incremental minimization process (Sect. 3.2.1).
For the von Mises yield condition and the associated flow rule, an upper bound
ratchet limit multiplier can be obtained by
%N
%N
n )dV
n dV
(
ij (tn ) + ij (tn ))ij
y
n=1
n=1 (
ij
V
V
R
(17)
F %N
UB =
n )dV
ij ( n=1 ij
V
which gives the capacity of the body subjected to a predefined cyclic load history
ij (tn ) to withstand an additional constant load ijF before ratchetting takes place.
(18a)
(18b)
(19a)
(19b)
267
creep load time instance, 0 in Eq. (6) needs to be replaced by the creep flow stress
0 = c . The creep flow stress is an implicit function of creep strain c and residual stress c during the creep dwell period. The detailed numerical procedures for
the evaluation of creep strain and flow stress are described in [27, 28] and these
processes are summarized as follows:
We assume a time hardening creep constitutive relation:
c
= B n t m
(20)
where is the effective creep strain rate, is the effective von-Mises stress, t is the
dwell time, and B, m and n are the creep constants of the material. When m = 0,
the time hardening constitutive equation becomes the Nortons law.
During the relaxation process we assume, at each point in space that an elastic
follow up factor Z exists:
Z
c
=
E
(21)
Z(m + 1)
n 1 ( c )n 1 ( s )n 1
where s is the effective value of the start of dwell stress, c is the effective value
of the creep flow stress, and c = (sij + Cij ). Integrating Eq. (21) gives the
effective creep strain during the dwell period t as,
Z
( c s ).
E
Combining Eqs. (22) and (23) and eliminating Z/E gives
=
c =
B(n 1)t m
(
cn
+1
sn
( s c )
)(m + 1)
(23)
(24)
c =
B sn t m +1
.
m + 1
(25)
F
The creep strain rate at the end of dwell time t is calculated by Eqs. (22)
and (24):
1
c (m + 1) cn
1
F
.
(26)
= B( c )n t m =
t (n 1) ( s c ) cn 1 sn 1
For the pure creep where s = c , the creep strain rate becomes:
F
F
= B( s )n t m .
(27)
268
Hence in the iterative process, we begin with current estimated ci , si and use
f
Eqs. (24), (26) or (27) to compute a new value of the creep stress c = c from
Eq. (28) to replace 0 (tn ) in the linear matching condition Eq. (6).
F
1
n
c =
.
(28)
m
Bt
6 Convergence Considerations
The necessary condition for convergence and the exact proof for upper bounds are
provided by [5, 6]. According to this study, in order to get convergent minimum
upper bound limits three conditions must be fulfilled as follows: (1) The material
yield surface must be convex; (2) The class of strain rates and the associated strain
increments guarantee that the minimum upper bound is limited with this class; (3)
The class of selected compatible strain distributions must be adequately extensive
to guarantee a satisfactory upper bound.
The first two conditions can be easily satisfied by an appropriate choice of a class
of linear materials. Condition (3) is vital to the implementation of the LMM within
a finite element scheme. Within the LMM, the equilibrium of the residual stress
field ui from which ij is derived,
field ij relies on the class of displacement
i.e. ij is in equilibrium if and only if V ij ij dV = 0. Hence, for a given finite
element mesh, the process will converge to the least upper bound associated with
the FE mesh and within this class of displacement field ui . However, during the
FE implementation, the volume integration is not exact but usually depends upon
the Gaussian integration to give an exact integral. Hence a point-wise condition is
used to replace above equilibrium condition;
k
wk ijk ij
=0
(29)
el
where wk are the Gaussian weighting factors at the Gauss integration points.
According to the lower and upper bound theorems, the LMM ensures that the
maximum lower bound will be less than the least upper bound. However, unlike
the strict convergence of the upper bound, the magnitude of lower bound may not
always increase monotonically with iterations. But upon convergence, the maximum
lower bound will equal to the least upper bound, where by equilibrium condition
(Eq. (29)) the matching condition is applied at Gauss points.
Due to the point-wise condition of equilibrium (Eq. (29)), whereas the deviation
from convergence at a few Gauss points has little effect on the upper bound which
is determined by volume integrals, the convergence of the upper bound in terms of
a particular number of significant figures may allow some deviation from convergence locally. Hence the convergence of lower bound may be affected significantly
as it is determined by single Gauss point. Generally the upper bound converges
(monotonically) more quickly than the lower bound and the rate of convergence for
lower bound depends upon the characteristic of the problem and also the adopted FE
269
model, such as the complexity of the geometry and boundary conditions, the mesh
arrangement, etc. For some cases where the lower bound converges very slowly, the
convergence is usually judged entirely in terms of the upper bound. Further investigation of the convergence of the LMM iterative algorithms has been carried out and
a separate paper is being prepared for this context.
7 Examples of Applications
In this section, three practical examples of the LMM for differing applications are
provided to confirm the efficiency and effectiveness of the method; the behaviour
of a defective pipeline subjected to cyclic thermal loading and constant operating
pressure, the effects of drilling holes on the ratchet limit and crack tip plastic strain
range for a centre cracked plate subjected to constant tensile loading and cyclic
bending moment, and the cyclic structural responses of a cruciform weldment under
creep fatigue interaction.
270
Fig. 3 The finite element mesh for a pipeline with part-through slot: (a) small slot; (b) circumferential slot; (c) axial slot and (d) large area slot
271
Fig. 5 Centre cracked plate with symmetric holes subjected to reversed bending moment range
M and constant tension p
272
As shown in Fig. 7, the accuracy of the lower and upper bound limit load boundary
obtained by the LMM has been verified by ABAQUS RIKS analysis. For the verification of LMM lower and upper bound ratchet limit boundary the cyclic load points
D (M = 1.6M0 , p = p0 ), and E (M = 1.6M0 , p = 1.1 p0 ), which are
just below and above the calculated upper bound ratchet limit boundary (Fig. 7),
respectively, are chosen for the step-by-step analysis in ABAQUS.
Figure 8 shows the plastic strain history at the crack tip for the cyclic loading D
and E calculated by ABAQUS step-by-step analysis. The calculated plastic strain
for the load case D settles to a stable cycle after about 5 load cycles showing a
reverse plasticity mechanism, and the load case E shows a strong ratchetting mechanism, with the plastic strain increasing at every cycle. This directly confirms the
accuracy of the predicted LMM lower and upper bound ratchet limits.
Optimization studies were performed further involving holes with different diameters drilled at different locations. The study shown that the most significant
decrease in crack tip plastic strain range with least reduction in the ratchet limit
is identified for the hole size D = 150 mm at the optimum location X0 /a = 0.1,
Y0 /a = 0.3, which gives a 72 % reduction in the plastic strain range and does not
reduce the ratchet limit.
273
Fig. 9 Geometrical and analysis parameters of the cruciform weld specimens: (a) dimensions and
applied loading; (b) FE-mesh with designation of different materials, boundary conditions and
mechanical loading
274
A Ramberg-Osgood formulation was adopted to simulate cyclic stress strain relationship, and a time hardening creep constitutive model was used to characterise
creep behaviour. In this creep fatigue damage assessment, the LMM was adopted to
evaluate a steady-state cyclic behaviour and to construct a saturated hysteresis loop.
Then obtained total strain range during the cycle was used to assess fatigue damage
combining R6 fatigue endurance curves [31]. The evaluated creep strain and stress
relaxation data were adopted to evaluate creep damage considering time fraction
rule and using the experimental creep rupture data. The final lifetime of the cruciform weldment was then obtained based on the calculated fatigue and creep damage
under creep-fatigue interaction conditions.
The detailed results of cruciform weldment creep fatigue assessment by the
LMM and comparisons with experimental solutions [32] are presented in Fig. 10.
Visual comparison of the observed and predicted in Fig. 10 for 3 variants of dwell
period t shows that 9 of the 11 simulations accurately predict the experimental results. Therefore, it can be used for the formulation of an analytic assessment
model suitable for the fast estimation of lifetime for a variety of loading conditions. The low computational effort required by the LMM compared to other
computational techniques makes it possible and relatively easy to extrapolate numerical predictions for loading conditions not captured by the available experiments.
A full discussion of the solutions and validations with experimental results are
given by Gorash and Chen [33]. This example demonstrates that, for such complex
industrial problems, the LMM is capable of providing lifetime related solutions that
are much more illuminating than conventional analysis.
275
Fig. 11 Other applications (a) a heat exchanger tube plane (b) rolling contact problem (c) 90 pipe
bends (d) composite cylinder with a crosshole
276
8 Other Applications
The stable and accurate results of the mentioned LMM on shakedown and ratchet
analysis have been confirmed in many industrial applications, including the problem of a heat exchanger tube plate subjected to severe cyclic thermal loading and
constant operating pressure (Fig. 11(a)) [8], the application to rolling contact problem (Fig. 11(b)) [34], shakedown and limit analysis of 90 pipe bends under internal
pressure, cyclic in-plane bending and cyclic thermal loading (Fig. 11(c)) [35], and
the shakedown analysis of a composite cylinder with a cross hole (Fig. 11(d)) [36]
and etc.
9 Conclusions
This study focuses on the performance of an elastic plastic body subjected to cyclic
loading. The design limits in plasticity and creep including shakedown limit, ratchet
limit, cyclic response under creep-fatigue interaction and plastic strain range regarding the fatigue crack initiation have been addressed in this study. The analysis
is performed by describing the steady cyclic state employing a general cyclic minimum theorem. In order to estimate the class of kinematically allowable strain rate
histories, the Linear Matching Method is used for obtaining the minimum of the
functional for these design limits. Three practical examples of the LMM are provided to confirm the efficiency and effectiveness of the method and demonstrate
that the LMM may be applied to a much wider range of circumstances than have
hitherto been possible.
Acknowledgements The author gratefully acknowledges the support of the Engineering and
Physical Sciences Research Council (EP/G038880/1) of the United Kingdom, and the University
of Strathclyde during the course of this work. The author would also like to thank Prof Alan
Ponter of the Department of Engineering, Leicester University, for his advice and discussion on
the theoretical development of the LMM.
References
1. ABAQUS (2007) Users manual. Version 6.7
2. Nguyen-Tajan TMI, Pommier B, Maitournam H, Houari M, Verger L, Du ZZ et al (2003)
Determination of the stabilized response of a structure undergoing cyclic thermalmechanical
loads by a direct cyclic method. In: ABAQUS users conference proceedings
3. Carter P (2005) Analysis of cyclic creep and rupture, part 2: calculation of cyclic reference
stresses and ratcheting interaction diagrams. Int J Press Vessels Piping 1(82):2733
4. Ponter ARS, Carter KF (1997) Limit state solutions, based upon linear elastic solutions with
a spatially varying elastic modulus. Comput Methods Appl Mech Eng 140:237258
5. Ponter ARS, Engelhardt M (2000) Shakedown limits for a general yield condition: Implementation and application for a von-Mises yield condition. Eur J Mech Appl Solids 19:423445
6. Ponter ARS, Chen HF (2001) A minimum theorem for cyclic load in excess of shakedown,
with application to the evaluation of a ratchet limit. Eur J Mech Appl Solids 20:539553
277
7. Chen HF, Ponter ARS (2001) A methodfor the evaluation of a ratchet limit and the amplitude
ofplastic strain for bodies subjected to cyclic loading. Eur J Mech Appl Solids 20:555571
8. Chen HF, Ponter ARS (2005) Interity assessment of a 3D tube plate using the linear matching
method, creep relaxation and reverse plasticity. J Press Vessel Technol 82:85104
9. Chen HF, Ponter ARS (2003) Methods for the evaluation of creep relaxation and the amplitude of strains for bodies reverse-plastic strains for bodies reverse-plastic subjected to cyclic
loading. In: Pressure vessels and piping, transaction of the ASME, Cleveland, OH, July 2003
10. ANSYS Inc. Southpointe, 275 Technology Drive, Canonsburg, PA 15317. www.ansys.com
11. Hibbit, Karlson and Sorensen Inc (1997) ABAQUSIStandard users manual, version 6.1,
vols 1, 2 & 3. USA
12. ASME (1990) Boiler and pressure vessel code. Code case: nuclear components, case N-47-29,
class I components in elevated temperature service. Sect 11, division I
13. AFCEN (1985) Design and construction rules for mechanical components of FBR nuclear
islands. RCC-MR, AFCEN, Paris
14. Liu YH, Carvelli V, Maier G (1997) Integrity assessment of defective pressurized pipelines
by direct simplified methods. Int J Press Vessels Piping 74:4957
15. Vu DK, Yan AM, Nguyen DH (2004) A primaldual algorithm for shakedown analysis of
structures. Comput Methods Appl Mech Eng 193:46634674
16. Staat M, Heitzer M (2001) LISA a European project for FEM-based limit and shakedown
analysis. Nucl Eng Des 206:151166
17. Seshadri R (2005) Inelastic evaluation of mechanical and structural components using the
generalized local stress strain method of analysis. Nucl Eng Des 153:203287
18. Mackenzie D, Boyle JT, Hamilton R, Shi J (1996) Elastic compensation method in shell-based
design by analysis. In: Proceedings of the 1996 ASME pressure vessels and piping conference,
vol 338, pp 203208
19. Chen HF, Ponter ARS, Ainsworth RA (2006) The linear matching method applied to the high
temperature life integrity of structures, part 1: assessments involving constant residual stress
fields. Int J Press Vessels Piping 83(2):123135
20. Chen HF, Ponter ARS, Ainsworth RA (2006) The linear matching method applied to the
high temperature life integrity of structures, part 2: assessments beyond shakedown involving
changing residual stress fields. Int J Press Vessels Piping 83(2):136147
21. Ainsworth RA (ed) (2003) R5: assessment procedure for the high temperature response of
structures. British Energy Generation Ltd, London, p 3
22. Bree J (1967) Elastic-plastic behaviour of thin tubes subjected to internal pressure and intermittent high-heat fluxes with application to fast-nuclear-reactor fuel elements. J Strain Anal
2:226238
23. Bree J (1989) Plastic deformation of a closed tube due to interaction of pressure stresses and
cyclic thermal stresses. Int J Mech Sci 31(1112):865892
24. Chen HF, Ponter ARS (2010) A direct method on the evaluation of ratchet limit. J Press Vessel
Technol 132:041202
25. Koiter WT (1960) General theorems for elastic plastic soloids. In: Sneddon JN, Hill R (eds)
Progress in solid mechanics, vol 1. North Holland, Amsterdam, pp 167221
26. Melan E (1936) Theorie statisch unbestimmter Systeme aus ideal-plastischem Bastoff.
Sitzungsber Akad Wiss Wien, Abtiia 145:195218
27. Chen HF, Ponter ARS (2006) Linear matching method on the evaluation of plastic and creep
behaviours for bodies subjected to cyclic thermal and mechanical loading. Int J Numer Methods Eng 68:1332
28. Chen H, Chen W (2012) A direct method on the evaluation of cyclic behaviour with creep
effect. In: Proceedings of the ASME 2012 pressure vessels & piping division conference,
Toronto, ON, Canada, 1519 July 2012
29. Chen H, Chen W, Li T, Ure J (2011) On ratchet, shakedown and limit analyses of defective
pipeline. J Press Vessel Technol
30. Chen H, Chen W, Li T, Ure J (2011) Effect of circular holes on the ratchet limit and crack tip
plastic strain range in a centre cracked plate. Int J Fract Mech 78:23102324
278
31. Bate SK, Hayes JP, Hooton DG, Smith NG (2005) Further analyses to validate the R5 volume
2/3 procedure for the assessment of austenitic weldments. Report for British Energy Generation Ltd No SA/EIG/11890/R002, Serco Assurance, Warrington, UK
32. Bretherton I, Knowles G, Hayes JP, Bate SK, Austin CJ (2004) Final report on the fatigue and
creep fatigue behaviour of welded cruciform joints. Report for British Energy Generation Ltd
No RJCB/RD01186/R01, Serco Assurance, Warrington, UK
33. Gorash Y, Chen HF (2012) Creep fatigue life assessment of cruciform weldments using the
linear matching method. Int J Press Vessels Piping 104:113
34. Chen H, Ponter ARS (2005) The linear matching method for shakedown and limit analysis
applied to rolling and sliding point contact. Road Mater Pavement Des 6:930
35. Chen H, Ure J, Li T, Chen W (2011) Shakedown and limit analysis of 90 pipe bends under
internal pressure, cyclic in-plane bending and cyclic thermal loading. Int J Press Vessels Piping
88:213222
36. Chen H, Chen W, Li T, Ure J (2011) Shakedown analysis of composite cylinders with cross
hole. J Press Vessel Technol 133:031206