You are on page 1of 136

Post Critical Behaviour of Shells:

a Sequential Limit Analysis Approach

Tesi presentata per il


conseguimento del titolo di Dottore di Ricerca
Politecnico di Milano
Dipartimento di Ingegneria Strutturale
Dottorato in Ingegneria delle Strutture - XIII Ciclo

di
Nicola Panzeri

Dicembre, 2001
Post Critical Behaviour of Shells:
a Sequential Limit Analysis Approach

Tesi di Dottorato dell’Ing. Nicola Panzeri


Relatori: Prof. Arch. Leone Corradi
Prof. Ing. Carlo Poggi
Dicembre 2001

Dottorato in Ingegneria delle Strutture


del Politecnico di Milano
Collegio dei Docenti:
Prof L. Biolzi
Prof S. Bittanti
Prof.ssa G. Bolzon
Prof. L. Cedolin
Prof.ssa C. Comi
Prof. R. Contro
Prof. A. Corigliano
Prof. L. Corradi
Prof. M. Di Prisco
Prof. A. Franchi
Prof. P Gambarova
Prof. G. Maier (Coordinatore)
Prof. F. Mola
Prof.ssa A. Pandolfi
Prof. A. Pavan
Prof. U. Perego
Prof. C. Poggi
Prof. A. Quarteroni
Prof. G. Sacchi
Prof. S. Sirtori
Prof. A. Taliercio
Summary

In the present work, a method for the study of the post-collapse behaviour of
shell structures, that buckle in the plastic range, is presented. The material is
assumed to be rigid-plastic, thus the elastic deformations are neglected. The
method consists of a sequence of limit analyses solved by means of the finite ele-
ment method.
The following parts can be identified in the thesis:

• an introductory part (chapter 1) presenting the main argument of the thesis.


Moreover, some hystorical remarks on the limit analysis and on the method
adopted, in order to study the collapse behaviour of shell structures, are
recalled;

• a second part consisting of chapters 2 and 3, where the theoretical arguments


are explained. In particular the elastic plastic constitutive relations and the
limit analysis theorems are presented in chapter 2. The method used, for the
solution of the limit analysis problem, is introduced in the same chapter.
Its implementation, by means of the finite element method, is explained
exhaustively in chapter 3. Firts of all a general description is presented,
therefore two elements are formulated: the first is a two nodes axisymmetric
shell element, based on the Kirchhoff hypotheses, the second is a three nodes
shell element capable to deal with shear strains. This element can be used
with thin and medium thick shells too;
• some analyses performed on different shell structures are presented in the
third part of the thesis (chapter 4 and 5) . In particular, chapter 4 deals
with the validation of the proposed method by comparing different results
obtained for symply supported plates. In chapter 5 some results concerning
a cylinder, a frusta and a square box column will be discussed pointing
out the capabilities of the method when used to predict the post-collapse
behaviour of shell structures;
• in the appendixes, A, B and C, some numerical and theoretical aspects, for
sake of clearness not reported in chapters 2 and 3, are presented in detail.
Sommario
Il presente lavoro propone un metodo per lo studio del collasso di strutture a
guscio in cui le deformazioni elastiche sono trascurabili. Questo metodo si basa
su di una sequenza di analisi limite la cui risoluzione è implementata mediante il
metodo degli elementi finiti.
Nell’ambito della tesi si possono individuare le seguenti parti:

• una parte introduttiva (capitolo 1) in cui viene illustrato l’argomento af-


frontato. È inoltre proposta una breve revisione storica dell’analisi limite e
del metodo utilizzato per studiare il comportamento a collasso di strutture
a guscio;
• una seconda parte, costituita dai capitoli 2 e 3, nella quale vengono affrontati
gli argomenti teorici. In particolare, i fondamenti della teoria della plasticità
e i teoremi dell’analisi limite sono richiamati nel capitolo 2. Sempre nello
stesso capitolo viene introdotto il metodo risolutivo, con approccio cine-
matico, del problema di analisi limite. La sua implementazione, mediante il
metodo degli elementi finiti, è descritta nel capitolo 3. Dopo una trattazione
generale vengono formulati due differenti elementi finiti di guscio: il primo
è un elemento rettilineo a due nodi, assialsimmetrico, con formulazione alla
Kirchhoff, mentre il secondo è un elemento triangolare a tre nodi, formulato
in modo da poter essere applicato anche per la modellazione di gusci medio
spessi;
• nella terza parte, composta dai capitoli 4 e 5, vengono presentate alcune
analisi svolte su strutture a guscio di diversa tipologia. In particolare la vali-
dazione del metodo proposto, illustrata nel capitolo 4, è effettuata mediante
l’analisi del collasso di piastre semplicemente appoggiate. Nel capitolo 5
vengono invece studiati un cilindro, un cono ed un tubo a sezione quadrata,
ponendo in evidenza i vantaggi ottenibili applicando il metodo dell’analisi
limite sequenziale allo studio del collasso di strutture a guscio;
• nelle appendici A, B e C, vengono approfonditi alcuni aspetti teorici e nu-
merici non riportati nei capitoli 2 e 3 per non appesantirne eccessivamente
la lettura.
. . . alla fine delle mie fatiche come studente di dottorato, è giunto il momento di scri-
vere qualche riga di ringraziamento per tutti coloro che hanno accompagnato la mia
crescita scientifica e professionale in questi tre anni. Come dimenticare i momenti di
frenetica attività prima di partire per un congresso, o i momenti di sconforto quando
una soluzione che sembrava a portata di mano si è poi rivelata sbagliata? Indimenticabili
sono soprattutto i momenti trascorsi scherzando e ridendo, come amici di lungo corso
che si ritrovano dopo tanto tempo a discutere del più e del meno.
I miei ringraziamenti vanno dunque al prof. Carlo Poggi, che un pomeriggio di quattro
anni fa mi convinse a proseguire gli studi. In questo periodo ne è stata fatta di strada.
Un doveroso grazie al prof. Leone Corradi al quale, nel corso della stesura della presente
tesi, mi sono sempre potuto rivolgere con la certezza di ricevere qualche utile suggeri-
mento.
Desidero, ovviamente, ringraziare anche tutti i docenti del collegio e il prof. Giulio Maier,
coordinatore del Dottorato di Ricerca in Ingegneria delle Strutture, per le possibilità for-
nitemi di approfondire le mie conoscenze nei vari campi dell’ingegneria strutturale.
Ringrazio, inoltre, l’ing. Valter Carvelli, prodigo di consigli, e l’ing. Stefano Mariani per
avermi risparmiato molta fatica avviandomi alla scrittura di questo lavoro in LATEX.

Mi è impossibile nominare tutti coloro che dovrei ringraziare, cosı̀ come non è mia in-
tenzione ridurre questa pagina ad un semplice elenco di nomi. Vorrei quindi dire, a tutti
coloro che non ho ringraziato esplicitamente, che la mia riconoscenza va anche a loro, al
di là di queste poche righe, scritte in fretta, ma con la consapevolezza di aver ricevuto
più di quanto ho dato.
to my parents . . .
Contents

1 Introduction 1
1.1 Objectives and organization of the thesis . . . . . . . . . . . . . . . 1
1.2 Limit analysis of structures . . . . . . . . . . . . . . . . . . . . . . 2
1.3 The sequential limit analysis . . . . . . . . . . . . . . . . . . . . . 5

2 Limit analysis and computation of the limit load 7


2.1 Theory of plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Limit analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Basic assumptions . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Limit theorems . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 Energy dissipation . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 The limit problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3 Finite element solution of the limit problem of shells 17


3.1 Finite element modelling . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Solution procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Mesh Updating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4 The axisymmetric shell element . . . . . . . . . . . . . . . . . . . . 23
3.4.1 Nodal variables . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4.2 Rigid-body motions and natural modes . . . . . . . . . . . 24
3.4.3 Natural strains in the element . . . . . . . . . . . . . . . . . 26
3.4.4 The element dissipation . . . . . . . . . . . . . . . . . . . . 27
3.5 The general shell triangular element . . . . . . . . . . . . . . . . . 29
3.5.1 Nodal variables . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5.2 Natural and cartesian strains . . . . . . . . . . . . . . . . . 31
3.5.3 Rigid-body motions and natural modes . . . . . . . . . . . 33
3.5.4 Natural strains in the element . . . . . . . . . . . . . . . . . 37
3.5.5 The element dissipation . . . . . . . . . . . . . . . . . . . . 43
3.5.6 Nodal equivalent loads . . . . . . . . . . . . . . . . . . . . . 45
3.5.7 The minimization procedure . . . . . . . . . . . . . . . . . . 46

i
4 Test example: simply supported plate 49
4.1 Geometry and mechanical characteristics . . . . . . . . . . . . . . . 50
4.2 Mechanism model . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.1 Thin plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.2 Thick plate . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Finite element limit analysis . . . . . . . . . . . . . . . . . . . . . . 54
4.3.1 Mesh variation . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.2 Thickness variation . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Post collapse behaviour . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4.1 Simply supported plate . . . . . . . . . . . . . . . . . . . . 59
4.4.2 Simply supported plate with in plane renstraints . . . . . . 62

5 Test examples: some shells 65


5.1 Cylinder S1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.1.1 Axisymmetric mechanism approach . . . . . . . . . . . . . . 68
5.1.2 The present approach . . . . . . . . . . . . . . . . . . . . . 70
5.2 Conical shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.1 The TICC5 cone . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3 Square tube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3.1 Axial load . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

6 Conclusions 91
6.1 A critical survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Future developments . . . . . . . . . . . . . . . . . . . . . . . . . . 92

A Strains from drilling modes 93

B Details on the compilation of the energy dissipation 99

C Details on the minimization procedure 109


C.1 Analytical integration to obtain K . . . . . . . . . . . . . . . . . . 109
C.2 Minimization respect to χ . . . . . . . . . . . . . . . . . . . . . . . 114

ii
Chapter 1

Introduction

Plasticity is an important branch of structural engineering and many researchers


have dealt with it in the twentieth century. For a long time theoretical deve-
lopments, experiments and discussions have been led and now the topic is well
settled. The applicability fields of the limit analysis have been defined, and in
many applications the method is used in order to predict the limit carrying ca-
pacity of structures, or their behaviour respect to a limit state. Extensions to
materials different from steel, such as alluminium alloy or anisotropic materials,
have also been studied.
The knowledge of the collapse load or at least of a more or less good approxi-
mation has been a great step ahead in the fifties, but nowadays many structures
need a design that can not leave out of consideration the behaviour after collapse.
Examples of such structures are bumpers, energy absorber and any structure that
can be subject to impact and develops large deformations in the plastic field. Of-
ten the design of such structures is strictly connected to the safety aspect, a topic
that is becoming always more important. In the last decade many publications
have dealt with this topic, whose field of application is very wide. The reader
can refer to some publications whose refences are reported in the bibliography:
in particular [68] deals with the crashworthiness of vehicles, Tabiei and Wu pre-
sented in [117] a study on the roadside safety structures, while in [101] Ravalard
et al. applied the prototypage virtuel to the crash of railway automotives.

1.1 Objectives and organization of the thesis


The aim of this work is to provide a useful tool for the prediction of the limit load
of shell structures and of their behaviour after the collapse. In order to obtain
this goal a finite element program based on the method of limit analysis has been
developed using the approach introduced by Capsoni and Corradi in 1995 [29].
A sequence of limit analysis is carried-out in order to obtain the post-collapse

1
Chapter 1. Introduction

behaviour of the shell structure.


After a brief review of the limit analysis of structures and of the sequential limit
analysis method presented in this chapter, the fundamentals theorems of plasticity
and the approach used are explained in chapter 2. Chapter 3 deals with the for-
mulation of the two shell elements used: an axisymmetric and a general triangular
shell element based on the TRIC element introduced by Argyris in 1997. Finally
some examples and calculations are presented in chapter 4 for simply supported
plates and in chapter 5 for various shells geometries, compared to experimental
data when available. Some conclusions and future developments are indicated in
chapter 6 while some details of calculations are reported in the appendixes. In
the bibliography many references to articles and books on the subject of plasticity
and limit analysis are reported. According to the author the knowledge of the
plasticity theory development can be very useful in the study of new and more
efficient methods of analysis. These references are not exhaustive of the subject
but can provide a wide background on the topic of plasticity, limit analysis and
on the numerical methods used up to now.

1.2 Limit analysis of structures


In plasticity the yield criterion plays a central role: it is with this tool that the
designer can know if a structure, or a part of it, develops plastic strains. The first
yield condition for metals was proposed in a series of papers from 1864 to 1872
by Tresca, who stated that a metal yields plastically when the maximum shear
stress attains a critical value. In the following years the theory was formulated
by St. Venant and Levy who introduced the basic constitutive relations for rigid
perfectly plastic materials in plane stress and in three dimensions. The flow rule
was also introduced. In 1913 von Mises published a paper where his widely used
pressure-insensitive yield criterion was described. In 1924, Prandtl extended the
St. Venant-Levy-von Mises equation for the plane continuum problem to include
the elastic component of strain, and Reuss in 1930 carried out this extension to
three dimension. In 1928, von Mises generalized his previous work to include a
general yield function and discussed the relation between the direction of plastic
strain rate and the regular yield function, thus introducing the concept of yield
function as a plastic potential. In the same period Prandtl attempted to formu-
late general relations for hardening behaviour, and Melan, in 1938, generalized
the foregoing concepts of perfect plasticity and gave incremental relations for har-
dening solids with regular yield surface.
In an independent way, in 1949, Prager arrived at a general framework, similar to
that proposed by Melan, in which the yield function and the loading-unloading
conditions were precisely formulated. In 1958 this framework was extended by
Prager to include thermal effects by allowing the yield surface to change its shape
with temperature.
In 1951, Drucker proposed the material stability postulate, a significant concept

2
1.2. Limit analysis of structures

of work hardening that permitted to treat in an unified manner the plastic stress-
strain relations and many other fundamental aspects of the subject.
Before these developments, an important contribution to the topic was supplied
by the Steel Structure Research Committee (UK), set up in 1929 in order to bring
some order into the design of steel structures. Two years after the Committee
formation, a Code of Practice was published in the First Report that formed the
basis of the British Standard Specification n. 449. In the 1939 the Committee
concluded that the method of design inherent in the Code of Practice was almost
entirely irrational and therefore incapable of refinement. It has to be noted that
the BS449 rules gave, in many cases, a safe method of design, based upon those
theorems of the plastic theory well known nowadays but not clearly articulated
in the ’30s.
J. Baker, Technical Officer in the Committee until 1936, Professor at Bristol Uni-
versity from 1933 until 1943 and in Cambridge since 1943, was at the head of
one of the more active and important research groups, operating in Cambridge
and that had members like J.W. Roderick and M.R. Horne. Their first complete
experimental analysis of the single-bay rectangular portal frame was published
in 1950, but the accompanying theory had been done on a mechanical approach,
without the guidance of any general principle. The sophisticated, but clear, con-
cept of virtual work had not yet found its application into such analysis. This
changed slowly from the publication in 1948 of a work by Freiberg (in Russian)
who stated the basic principles that are needed. It appears however that the
earliest reference to the theorems of limit analysis was probably due to Gvozdev
(1936). Precise formulations of the two fundamentals theorems of limit analysis
were given by Drucker, Greenberg and Prager in 1951-52 for an elastic-perfectly
plastic material and by Hill (1951-52) from the point of view of rigid-ideally pla-
stic materials.
The application of the upper and lower bound theorems has been for a long time
restricted to very simple structures, in most cases only one-dimensional problems.
The subsequent introduction of the discretization techniques and the simultane-
ous growth of the calculations tools led to an impressive development of numerical
methods for the limit analysis of structures. Mathematical programming [83] was
soon recognized as a suitable tool for the solution of rigid-plastic problems, and
the analysis of beams and frames was formulated for a long time in terms of lin-
ear programming theory [85, 91, 116, 118, 126]. For continuous solids the use of
this method is subject to a piece-wise linearization of the yield condition that
can introduce more or less meaningful errors, and the great number of variables
introduced by linearization reduces the effectivenes of the approach.
On the contrary, the linearization can be avoided and the resulting problems can
be solved with major accuracy by means of the nonlinear programming theory.
Some finite elements based on mixed functionals have been proposed [67] and
the favorable ratio accuracy/computation charge played an important role in the
success of this approach.
In the last decade new methods have been proposed for the solution of limit

3
Chapter 1. Introduction

problems based on the kinematic theorem: Sloan and Kleemen [111] obtained
very good solutions for many engineering interesting cases using the mathemati-
cal programming and finite elements that can account for non-continuous velocity
fields. Jiang has recently proposed a different method that lead the formulation
of the limit problem to the optimization of a regular functional solved using an
iterative procedure already applied to visco-elastic problems [67].
In 1991, Huh and Yang [61] proposed a new method for the calculus of limit solu-
tions of a plane stress problem, but their method can also be applied to the more
general three dimensional problems. Using a duality theorem they defined con-
vergence from above and from below to the exact solution. The non-smoothness
of the original problem was avoided by introducing a small quantity δ such as
when δ → 0 the solution of the original problem is recovered. The robustness and
the rate of the convergence of the computational algorithm has been theoretically
demonstrated [21, 127] and succesfully tested [61].
Finally Capsoni and Corradi [29] and Liu et al. [81] separately proposed two ap-
proaches very similar: their are based on a dissipation function expressed directly
in terms of the plastic strains while the normality rule can be considered as a
costraint between the components. This allows to obtain the collapse multiplier
as the free minimum of a convex function and the solution is obtained by mean
of a iterative procedure.
The approaches proposed by Huh & Yang and by Capsoni & Corradi differ from
the method used to deal with the non-smoothness of the dissipation functional.
In [61] a small positive coefficient δ was introduced so that the original problem
corresponds to δ → 0, while Capsoni & Corradi proposed to set as rigid the e-
lements that do not exhibit plastic flow. This approach avoids the outer iteration
on δ needed in [61] and permits to reduce the problem size when rigid elements
are imposed. Unfortunately an efficient way to impose these constraints has not
yet been found and in some cases the efficiency of the method can be reduced.
In the author’s knowledge these new methods have been used in most cases to
obtain the limit load of simple structures, like plate or plane strain/stress pro-
blems [33]. Recently Corradi and Vena published a paper related to the limit
analysis of anisotropic materials using the Tsai-Wu criterion [34, 44, 119], but
the analyses remain confined to the achievement of the collapse load while the
post-collapse behaviour of the structure remain unknown. An attempt to ob-
tain the post-collapse behaviour of rigid-plastic structures has been proposed by
Seitzberger and Rammerstorfer: in [109, 110] they proposed an older idea [53, 59]
consisting in a sequence of limit analyses applied to a mesh updated, consequently
to the velocity field obtained in the previous solution. Their procedure was applied
to axysimmetric structures only, while for more complex geometries an iterative
technique was used.

4
1.3. The sequential limit analysis

1.3 The sequential limit analysis


Despite its semplicity, sequential limit analysis has not been widely used for the
study of the post-collapse behaviour of structures. The method is based on a
sequence of limit analyses carried out on updated meshes: being the first limit
analysis performed on the original mesh, as a results the limit load and the kine-
matic solution are obtained. The velocity field can be seen as instantaneous nodal
velocities of the structure in a Lagrangian coordinate system, therefore it can be
integrated over a small time interval to obtain a small displacement vector. The
superposition of these displacements to the original mesh provides an updated
geometry that can be used for the next iteration. This process is repeted to form
a sequence leading to the solution of a large displacement problem.
Sequential limit analysis has been originally proposed for the study of frames
by Horne and Merchant [59]. Some applications by Gavarini, Horne and Morris
can be found in [53, 60]. Recently the method has received a growing interest,
it has been applied by Yang [128] in order to study the behaviour of some truss
structures, while Seitzberger and Rammerstofer employed it to simulate the large
deformation crushing behaviour of axisymmetric shells [110]. Since in limit ana-
lysis elastic strains are neglected, only the rigid-plastic collapse curve is obtained
but this is a minor limitation in the study of structures that undergo large defor-
mations and for which the elastic contribution is negligible.
Different methods can be employed to predict the post-collapse behaviour: the
most complete simulation is provided by large displacement, incremental analysis
performed keeping track of all types of informations, elastic and plastic strains.
This method, implemented in various commercial F.E. packages, has high com-
puting cost and often suffers of numerical instability problems. Other methods
like mechanism analysis can be employed to predict the post-collapse curve. They
follow the evolution of elementary mechanism consisting of circumferential hinges
where bending dissipation is concentrated, while the regions in between experience
membrane flow only [2, 3, 43]. However, such a procedure only considers axisym-
metric collapse modes and axisymmetric geometric imperfections. Furthermore,
the change in shape of the mechanism can hardly be followed to great accuracy.
Moreover some structures do not collapse in axisymmetric way but could exhibit
different collapse modes such diamond shape [86–88,90,99,100] or circumferential
waves [105].
At the end, the sequential limit analysis seems to be a good balance between com-
putationally efficiency, numerical stability and accuracy of results. It by-passes
the tedious and computationally demanding task of keeping track of stresses and
strains, typical of the incremental approach, remaining globally stable whatever
material and geometry is considered. The difficulties that arise during the elasto-
plastic incremental analyses, when dealing with imperfection sensitivity structures
or with softening materials, do not affect the stability of the sequential limit ana-
lysis method, allowing it to be more efficient and to obtain the post-collapse curve
with a minor number of increments.

5
Chapter 1. Introduction

6
Chapter 2

Limit analysis and


computation of the limit
load

From the structural point of view strength and ductility are the most important
material characteristics. When the stress in a point of the material reaches limit
strength, then rupture or fracture can occur (brittle materials), or some plastic
strains can develop (ductile materials). Structures built with ductile materials
can deform in a considerable way before collapse and often their crisis is due
to the impossibility to achieve an equilibrium configuration rather than material
rupture. This leads to the important consideration that the collapse load, in many
real situations is independent from the load history and its value can be obtained
in a direct way by means of limit analysis. The aim of this chapter is to present
the theoretical bases and assumptions on which the calculations are performed.
In section 2.1 some hypotheses on the theory of plasticity are recalled and the
main theorems are illustrated. Particularization to the limit analysis of structures
is presented in section 2.2 while in the last section of this chapter a formulation of
the limit problem suitable for a finite element analysis is presented. The solution
strategy and the finite element adopted are discussed in chapter 3.

2.1 Theory of plasticity


Let’s consider a body subject to volume forces F in Ω and surface pressures f on
∂f Ω; the surface ∂u Ω is fixed. The following hypotheses are assumed:

• material is isotropic;

• quasi-static response of the structure is considered so that any dynamic

7
Chapter 2. Limit analysis and computation of the limit load

effects are neglected;


• temperature is ignored;
• elastic properties are not influenced by the load story;
• displacement gradient are small so that kinematic relations are linear.
Under previous assumptions the strains in a body can be regarded as the sum of
the elastic and plastic part while the stresses depends from elastic strains only:

²ij = eij + pij (2.1a)


σij = Dijkl ekl (2.1b)

where Dijkl is the elastic stiffness tensor. Latin subscripts are employed to identify
the tensor components and run over the numbers 1,2,3 related to the cartesian
axes.
Analogous relationships can be written using rate terms:

²̇ij = ėij + ṗij (2.2a)


σ̇ij = Dijkl ėkl (2.2b)

where a dot over the relevant symbol stands for a time derivative. In order to
obtain an analytical formulation of the problem some definitions are needed:
• an elastic instantaneous domain, or yield surface that can recognize stress
states related to elastic or plastic situations;
• a flow rule governing the plastic strain rates.
The elastic domain is defined by means of one or more yield functions as follow:

φα (σij , χh ) ≤ 0 , α = 1, . . . , Y (2.3)

where χh are internal variables governing the hardening of the material.


Strain rates are assumed to be normal to the elastic domain by means of the
following flow rule:
XY
∂φα
ṗij = λ̇α (2.4)
α=1
∂σ ij

The admissible alternatives are:

λ̇α ≥ 0 if φα = 0 and φ̇α = 0 ; λ̇α = 0 otherwise (2.5)

From the above equations it can be concluded that:


• if relations (2.3) are satisfied as strict inequalities, the material behaviour
purely is elastic, and the stress state is described by a point into the elastic
domain;

8
2.1. Theory of plasticity

• plastic strains can develop only when the stress state corresponds to a point
on the boundary of the elastic domain (2.3);
• equations (2.4) and (2.5) describe plastic strain rates normal to the elastic
domain at the stress state point. The direction of the normal is uniquely
defined if only one yield function is equal to zero; otherwise plastic strain
rates ṗij have to be contained in the cone defined by the normals in the
point identifying the stress state.
Relations (2.4) and (2.5) can be obtained assuming Drucker’s inequality:
¡ ∗
¢ ∗
¡ ∗ ¢
σij − σij ṗij ≥ 0 ∀ σij :φα σij , χh ≤ 0 (2.6)
where σij is the stress on the yield surface generating the plastic strain rate ṗij

and σij is an admissible stress state.
Power dissipation can now be defined as:
D (ṗij ) = σij ṗij (2.7)
From simple geometrical considerations it can be seen that the dissipation depends
on the plastic strain rates ṗij only. In fact if the surface (2.4) is stricly convex,
a given normal direction defines uniquely the associated stress point. Otherwise
if the elastic domain has a flat portion of the boundary, then the stress state
associated to the plastic strain rate is not uniquely defined, but his projection
on the normal direction is the same within the whole flat domain and thus the
dissipation is still uniquely defined.
It remains to be decided what yield function to use: in the case of an isotropic
material it depends only on the stress invariants:
φ (I1 , I2 , I3 ) = 0 (2.8)
where Ij are the invariants of the stress tensor defined as:
I1 = σii (2.9a)
¡
1 2 ¢
I2 = I − σij σij (2.9b)
2 1
1 ¡ ¢
I3 = 2σij σjk σki − 3I1 σij σij + I13 (2.9c)
6
In (2.9) the Einstein’ summation convention on repeated indexes is adopted.
Plastic deformations do not produce volume changes in metal materials therefore
equation (2.10) must be satisfied for any admissible velocity field:
ṗii = 0 (2.10)
As a consequence the yield surface has to be independent from the mean hydro-
static tension σii /3 and it is natural to suppose that the yield function depends
on the deviatoric stress tensor sij expressed by:
1
sij = σij − δij I1 (2.11)
3

9
Chapter 2. Limit analysis and computation of the limit load

where δij is the Kronecker’s tensor.


Subject to this hypothesis the yield function for an isotropic material can be
expressed in terms of the deviatoric stress tensor invariants J2 and J3 :
φ (J2 , J3 ) = 0 (2.12)
where
1
J2 = sij sij (2.13a)
2
1
J3 = sij sjk ski (2.13b)
3
One of the most used and accurate yield function for metals is described by the
von Mises
q criterion: plastic strains canqevolve when the octahedral shear stress
2 2
τoct = 3 J2 reaches the limit value 3 τ0 . The von Mises yield function is
expressed by: p
φ (J2 ) = 3J2 − σ0 (2.14)

where σ0 = 3τ0 is the uniaxial yield stress.

2.2 Limit analysis


In section 2.1 the relations governing an elastic-plastic problems have been pre-
sented. It is clear that the solution of a such problem requires an incremental
procedure due to the irreversible nature of elastic-plastic behaviour. Such analysis
can be very complicated, and the computational effort is not negligible. In many
engineering practical applications the limit load of the structure under conside-
ration is the most important information required by the designer: its value can
be obtained by means of direct methods such as limit analysis. It provides, in a
direct way, the value of the collapse load avoiding the computationally demanding
procedure of the elasto-plastic incremental method. According to the approach
used, statically or kinematically admissible, the value obtained corresponds to
the lower bound or to the upper bound multiplier of the base loads applied. Be-
cause of the hypotheses assumed, and reported in subsection 2.2.1, the multiplier
calculated by means of the limit analysis is not the actual collapse load of the
real structure but an approximation. Neverthless the limit analysis provides a
useful tool in order to predict the ultimate carrying capacity of a structure and
its collapse mechanism.
In the following the main theorems of the limit analysis are remembered.

2.2.1 Basic assumptions


As mentioned before, the limit or collapse load is defined as the plastic collapse
load of an idealized structure for which some assumptions are made:

10
2.2. Limit analysis

• plastic deformations can increase without limits;


• no hardening or softening is considered (ideal or perfect plasticity);
• changes in the geometry are negligible, hence the geometrical description of
the structure remains unchanged during the deformation at the limit load.
The small deformations assumption allows the use of the virtual work principle
(2.15): Z Z Z
fT u∗ dx + FT u∗ dx = σ T ²∗ dx (2.15)
∂f Ω Ω Ω

where the () quantities form a compatible set while f, F and σ form an equili-
brated set. Engineering notation is used, therefore:
T
σ = {σx , σy , σx , τxy , τyz , τxz } (2.16a)
T
² = {²x , ²y , ²x , γxy , γyz , γxz } (2.16b)

The virtual work principle, and its rate form obtained substituting the finite
quantities with their rates, may be used to obtain the theorems of limit analysis
presented in section 2.2.2

2.2.2 Limit theorems


In the mechanics of deformable solids three basic relations must be satisfied for
a valid solution of a problem, namely the equilibrium equations, the constitutive
relations and the compatibility equations. Three solution sets can be associated to
the previous relations: a statically admissible set S, the constitutively admissible
set C = Cσ ∩ C²̇ and the kinematically admissible set K. The set Cσ contains
all the solutions satisfying equation (2.3) while in C²̇ the solutions have to satisfy
equations (2.4) and (2.5). If the intersection of the sets S, C and K is empty, there
is no solution. If it consists of a single point, the solution is unique. Otherwise
there is a set of feasible solutions, one of which may be the most preferred. The
limit analysis considers only two sets at a time and the search of the optimal
solution is facilitated by an objective function. In the primal formulation of a
limit analysis problem, the optimal solution corresponds to an extreme point in
S ∩C σ . All the non-optimal solutions furnish a multiplier lower than the limit
one. On the contrary, the optimal solution of the dual formulation corresponds
to an extreme point in K ∩C ²̇ . In this case the non-optimal solutions furnish a
multiplier greater than the limit one.

Statically admissible stress field and lower bound theorem


A statically admissible field is characterized by the following properties:
• the equilibrium relations are satisfied;

11
Chapter 2. Limit analysis and computation of the limit load

• the stress boundary conditions are also satisfied;


• no part of the body violates the yield condition.
Load associated to a statically admissible field cannot be greater than the collapse
load, therefore the lower bound theorem may be stated as follows:

Theorem 2.2.1. If an equilibrium distribution of stresses σ E can be found which


balances the body forces αF in Ω and the surface tractions αf in ∂f Ω and it is
¡ E¢
everywhere internal to the elastic domain, φ σij ≤ 0, then the body subject to
loads αF and αf will not collapse.
Obviously α ≤ s where α is the multiplier of the static admissible problem
and s the exact solution of the limit problem.

Kinematically admissible velocity field and upper bound theorem


A kinematically admissible velocity field ²̇∗ is characterized by the following pro-
perties:
• strain rate and velocity compatibility equations are satisfied;
• the homogeneous velocity boundary conditions are also verified;
Equating the external rate of work Π (u̇∗ ) to the internal dissipation power D (²̇∗ )
a base load multiplier β can be found so that applied loads are always greater or
equal to the actual limit load:
βΠ (u̇∗ ) = D (²̇∗ ) (2.17)
where
Z

D (²̇ ) = σ T ²̇∗ dx (2.18a)

Z Z
Π (u̇∗ ) = FT u̇∗ dx + fT u̇∗ dx (2.18b)
Ω ∂f Ω

The upper bound theorem may be stated as follows:

Theorem 2.2.2. If a compatible mechanism of plastic rate deformations ²̇∗ and


velocities u̇∗ is assumed which satisfies the kinematics boundary conditions on
∂u Ω, then the loads βF and βf determined equating the rate of external work to
the rate of internal dissipation will be either greater or equal to the actual limit
load.

12
2.2. Limit analysis

The kinematic multiplier β is always greater than the limit multiplier s except
for the optimal solution that furnishes β = s.

It must be noted that the lower bound theorem considers only the equilibrium and
the yield conditions while the upper bound theorem considers only the velocity
field and the energy dissipation. A suitable choice of stress and velocity fields
allows to obtain sufficiently close bounds of the limit load for the problem under
consideration.

2.2.3 Energy dissipation


The use of the upper bound theorem requires the definition of the energy dissi-
pation of the structure for a given strain rate field. The specific dissipation is
defined as:
D̂ = σ T ²̇ (2.19)
where the stress tensor σ satisfies the yield condition
φ (σ) = f (σ) − σ0 = 0 (2.20)
f and σ0 in (2.20) depend on the yield criterion used while, because of the rigid-
plastic assumption, ²̇ represents the plastic strains rate, normal to the yield bound
by means of the flow rule.
In the following the von Mises criterion is assumed: the yield function is therefore
strictly convex, continuously differentiable and the associated flow rule permits
to describe the plastic strain rates as functions of plastic flow. In particular the
yield function is expressed in terms of the second deviatoric stress invariant (eq.
(2.13a)): r
p 1 T
f (s) = 3J2 = s Rs (2.21)
2
where R is the diagonal constant matrix defined in (2.22) and s the deviatoric
stress tensor.
R = diag {3, 3, 3, 6, 6, 6} (2.22)
Plastic deformations occur only when the stress state point is on the boundary
of the elastic domain (2.3) and the flow rule (2.4) imposes the normality between
the boundary and the plastic strain rate vector. Therefore:
φ (s) = 0 ⇔ f (s) = σ0 (2.23)
and:
∂φ
²̇ = λ̇ = nλ̇ (2.24)
∂s
where ²̇ represent the deviatoric strain tensor.
It is easy to verify using eq. (2.23) that the normal n can be expressed as:
1
n= Rs (2.25)
2σ0

13
Chapter 2. Limit analysis and computation of the limit load

A finite element solution of the rigid-plastic problem by means of the upper bound
theorem requires the expression of the dissipation power in term of plastic strains.
In order to achieve this expression some operations are performed using eq. (2.24).
Let Θ be a symmetric matrix, therefore a quadratic form in the plastic strains
can be written:
λ̇2 T
²̇T Θ²̇ = s RΘRs (2.26)
4σ02
If a matrix Θ can be found satisfying the condition

sT RΘRs = 4σ02 ∀ s such that φ (s) = 0 (2.27)

one obtains:
λ̇2 = ²̇T Θ²̇ (2.28)
T
The condition of plasticity (2.23) implies s Rs = 2σ02 and this is verified for any
matrix Θ such that
RΘR = 2R (2.29)
The solution of equation (2.29) leads to the following relation:
1
Θ = 2R−1 = diag {2, 2, 2, 1, 1, 1} (2.30)
3
and the dissipation power can be expressed in terms of plastic strains only by
virtue of eq. (2.23) and eq. (2.28):
∂φ ∂f p
D̂ = σ T λ̇ = σ T λ̇ = σ0 λ̇ = σ0 ²̇T Θ²̇ (2.31)
∂σ ∂σ
The expression (2.31) for the dissipation is the base for the limit analysis by means
of the finite element method.
The use of the deviatoric stress tensor bypasses the problem of a singular matrix
in the definition of Θ, a problem solved in a different way in [31].

2.3 The limit problem


As seen in section 2.2 the collapse load multiplier can be obtained as the maximum
of the lower bound solutions, or as the minimum of the upper bound multipliers.
In a finite element approach as that used in this work, the upper bound theorem is
more useful than the lower bound one because it is easier to impose a kinematically
admissible velocity field than an equilibrated stress field. Moreover, an expression
of the dissipation power in terms of plastic strain rates has been obtained in
subsection 2.2.3 and this can be used to obtain the multiplier in an upper bound
approach. In this section an alternative form of the upper bound theorem will be
presented and a proof of the statement will be provided.
Considering the upper bound theorem reported in subsection 2.2.2 it can be noted

14
2.3. The limit problem

that both the dissipation power and the rate of work of external load depend
linearly on the mechanism amplitude; hence this is arbitrary and it can be fixed
imposing Π (u̇) = 1.
Moreover plastic deformations develop at constant volume, therefore the limit
problem can be expressed as follows:
Z
s = min D̂ (²̇) dx (2.32a)
˙ u̇
, Ω
subject to ²̇ = ∇s u̇ in Ω (2.32b)
T
µ ²̇ = 0 in Ω (2.32c)
u̇ = 0 on ∂u Ω (2.32d)
Z Z
Π (u̇) = T
F u̇dx + fT u̇dx = 1 (2.32e)
Ω ∂f Ω

where D̂ (²̇) is defined in eq. (2.31) and

T
µ = {1, 1, 1, 0, 0, 0} (2.33)
Equations (2.32b), (2.32c) and (2.32d) represent respectively the compatibility
equations, the incompressibility and the boundary velocity conditions. The pro-
blem (2.32) is convex, but where no plastic strains occur, D̂ (²̇) is not differentiable
and as a consequence the functional is not stationary at solution.
In order to prove the statement (2.32) the Lagrangean functional L of the problem
is introduced:
Z h i
L (²̇, u̇, σ, P, α) = D̂ (²̇) − σ T (²̇ − ∇s u̇) + P µT ²̇ dx − α [Π (u̇) − 1] (2.34)

The functional (2.34) is differentiable respect to all variables but ²̇. Its optimality
condition reads:
δL ≥ 0 subject to u̇ = 0 on ∂u Ω (2.35)
Relation (2.35) must hold an equality when variations respect σ, P and α are
considered as required by (2.32).
Stationarity respect to u̇ leads to the following expression:
Z "Z Z #
δu L = σ T δ (∇s u̇) dx − α FT δ u̇dx + fT δ u̇dx
Ω Ω ∂f Ω (2.36)
= 0 ∀ δ u̇ = 0 on ∂u Ω

If variation respect to the plastic strains rate ²̇ is considered the following relation
is obtained: Z h i
T
δ² L = δ D̂ (²̇) − (σ − P µ) δ ²̇ dx ≥ 0 ∀ δ ²̇ (2.37)

15
Chapter 2. Limit analysis and computation of the limit load

In order to calculate δ D̂ (²̇) it is necessary to distinguish between the region


ΩP , where plastic strains occur, and the rigid part ΩR : in ΩP the dissipation is
differentiable and eq (2.37) holds as an equality leading to

σ T = sT² + P µT (2.38)
n oT
In fact δ D̂ = ∂ D̂/∂ ²̇ δ ²̇ and

( )T
∂ D̂ ²̇T Θ ²̇T Θ
= σ0 p = σ0 = σ0 nT Θ (2.39)
∂ ²̇ T
²̇ Θ²̇ λ̇

and recalling equations (2.25) and (2.30):


( )T
∂ D̂
= sT (2.40)
∂ ²̇

Relation (2.37) therefore holds as an equality if equation (2.38) is satisfied in any


point of Ω.
In ΩR the dissipation is zero and its variation due to a plastic strain rate is equal
to the dissipation calculated in the deformed state:

δ D̂ = D̂ (²̇) = sTδ² δ ²̇ (2.41)

and the following relation is obtained:


T
[sδ² − (σ − P µ)] δ ²̇ ≥ 0 ∀ δ ²̇ such that µT δ ²̇ = 0 in ΩR (2.42)

If the lagrangean multiplier fields σ (x) and P (x) are interpreted as stresses and
hydrostatic pressure distributions, equation (2.36) imposes the equilibrium of the
body subject to the basic loads multiplied by α.
Moreover equation (2.38) has been obtained imposing the normality rule (2.4)
therefore the stress tensor σ is everywhere contained within the limit domain
in ΩP . Equation (2.42) compared with the Drucker’s postulate (2.6) implies
φ (σ) ≤ 0 in ΩR .
Therefore the optimal value of α is both statically and kinematically admissible
being, for eq. (2.32e):
Z Z
α∗ = α∗ Π (u̇) = σ T∗ ²̇∗ dx = D̂ (²̇∗ ) dx = β∗ (2.43)
Ω Ω

where the subscript ∗ refers to the solution values.

16
Chapter 3

Finite element solution of


the limit problem of shells

3.1 Finite element modelling


Let us consider a discretization of the body Ω in ne finite elements and nn nodes.
Each element is connected to nne nodes and in each node nv nodal variables
(displacement or rotations) are unknown. These nodal variables can be defined in
the global reference system (GRS) or in a local reference system (LRS) associated
to the element into consideration.
Let us define:

Definition 1. (O, x, y, z) the global reference system (GRS) of the structure;

Definition 2. (G, x0 , y 0 , z 0 ) the local reference system (LRS) of the finite element
considered.
The displacements can be expressed in either reference system by means of
the following relation:
v0 = Tv (3.1)
where the vectors v0 and v represent the displacements in the LRS and GRS
respectively and T is the transformation matrix. Because of the ortogonality of
the reference systems T−1 = TT and

v = TT v 0 (3.2)

Let us collect the nodal variables of the whole structure in the vector u if referred
to the GRS or in the vector u0 if referred to the LRS. If only nodal variables
related to an element e are considered they are grouped in the vectors ue and u0e

17
Chapter 3. Finite element solution of the limit problem of shells

referred to the GRS and to the LRS respectively.


Let us consider now an element e with all its nodal variables collected in a vector
u0e : the displacement over the element surface (or axis) can be expressed by means
of a matrix of shape functions N (x0 ):

u0e (x0 ) = N (x0 ) u0e (3.3)

Strains over the element are obtained with the hypothesis of small displacement,
so ²ij = 12 (si,j + sj,i ). When the derivation process is applied to the matrix
N (x0 ) one obtains:
²0e (x0 ) = Be (x0 ) u0e (3.4)
Local coordinates have been used because calculations are generally easier in the
LRS than in the GRS.
In computing the dissipation only modes that entail plastic strains are useful,
while rigid body motions have only to be considered from a kinematical point of
view. The natural formulation proposed in the sixties by Argyris is appropriate
because it is based on the separation of rigid-body motions from natural modes.
The sum of rigid-body motions and natural modes has to be equal to the number
of nodal variables of the element and they can be collected in the vector ρe :
µ ¶
ρ0
ρe = (3.5)
ρN

where the sub-vector ρ0 contains the rigid-body motions and the sub-vector ρN
the natural modes.
Normally natural modes are referred to a natural reference system (NRS) defined
accordingly to the element used: for example for a triangular shell element the
natural axes are coincident with the sides of the element it self. If strains are
expressed in the NRS there exist a matrix T so as
−1
²e (x) = Te ²0e (x0 ) (3.6)

and they can be obtained from the natural modes:

²e (x) = be (x) ρN (3.7)

In all the expressions a bar over a symbol denote a quantity in the NRS. Suppose
that the relation between the natural modes ρN and the nodal variables u0e is
known, therefore plastic strains (3.4) can be expressed in terms of the natural
modes only. Using equations (3.6) and (3.7) one obtains:

²0e (x0 ) = be (x0 ) ρN (3.8)

where the vector of natural strains

ρN = C0e u0e (3.9)

18
3.2. Solution procedure

has s = m − nr entries, being m the number of nodal variables and nr the number
of rigid-body motions of an element.
Matrix be (x0 ) can be obtained from the natural counterpart multiplying it by the
transformation matrix T:
be (x0 ) = Te be (x) (3.10)

3.2 Solution procedure


Keeping in mind the formulation (2.32) of the limit problem the power dissipation
can be obtained in terms of natural strains. Recalling equations (2.28) and (3.8)
and observing that the matrix Θ has to be invariant respect to a coordinate
transformation being the material isotropic, one obtains:
q q
T T T
λ̇ = ρ̇TN be (x) Te ΘTe be (x) ρ̇N = ρ̇TN be (x) Γe be (x) ρ̇N (3.11)

where a kinematically admissible velocity field has been considered. The matrix
T
Γe = Te ΘTe depends only on the element position while matrix be , describing
the strains in the element, depends on the coordinates x̄.
The dissipation of the whole structure can be found adding together the energy
dissipation of each element:
ne Z
X q
T
D= σ0 ρ̇TN be (x) Γe be (x) ρ̇N dx (3.12)
e=1 Ωe

Problem (2.32) can now be formulated in terms of finite element quantities:


ne Z
X q
T
s = min σ0 ρ̇TN be (x) Γe be (x) ρ̇N dx (3.13a)
ρ̇N ,u̇ Ωe
e=1
subject to ρ̇N = Ce u̇ e = 1, . . . , ne (3.13b)
T
p u̇ = 1 (3.13c)

where:

• the incompressibility condition has not be written because it will be con-


sidered in the obtaining of the appropriate matrix Θ;

• nodal variables are eliminated on the constrained boundary;

• vector p is the vector of equivalent nodal loads.

As suggested in section 2.3 equations (2.32) define the collapse multiplier as the
minimum of a convex, but not everywhere differentiable function. The finite
element domain has to be split in two part: a region EP where the elements

19
Chapter 3. Finite element solution of the limit problem of shells

develop plastic strain (ρN 6= 0), and a region ER where the elements are only
subject to rigid-body motions. In the first region the elements contribute to the
dissipation power and the function De is differentiable, while in ER the rigidities
impose some constraints to the nodal variables. These constraints can be seen as
a set of linear equations:
u̇ = G0 u̇0 (3.14)
where the vector u̇0 collects all the independent nodal variables in the global
reference system.
Natural straining modes can be obtained as follows:

ρN = C0e u̇0e = C0e Te u̇e = C0e Te G0e u̇0e = C0e u̇0e (3.15)

where the pedix e refers to a single element e. After substitution of ρN in eq.


(3.12) the dissipation power becomes:
XZ q
T
D= σ0 u̇T0e CT0e be (x) Γe be (x) C0e u̇0e dx
e Ωe
XZ q (3.16)
= σ0 u̇T0e Λ0e u̇0e dx
e Ωe

and the discrete form of the limit problem can be written as:

s = min D (u̇0 ) ∀ e in EP (3.17a)


u̇0

subject to pT G0 u̇0 = 1 (3.17b)

The lagrangean function reads:


¡ ¢
L (u̇0 ; α) = D (u̇0 ) − α pT G0 u̇0 − 1 (3.18)

In equations (3.17) and (3.18) all the elements in EP have to be considered.


Between the element nodal variables and all the nodal variables there exist a
relation due to the assemblage operation simbolically written as:

u̇e = Le u̇ (3.19)

For an illustration purpose of the solution procedure let us suppose that all the
elements undergo plastic flow so that ER = 0 and G0 = I. Under this hypothesis
the lagrangean function is differentiable everywhere and must be stationary at
solution. Therefore the following non-linear problem is obtained:
 
X ne Z
∂L Λ0e (x)
= σ0 q dx u̇0 − αGT0 p = 0 (3.20)
∂ u̇0 e=1 Ωe
T
u̇0 Λ0e (x) u̇0

20
3.3. Mesh Updating

Equation (3.20) is solved iteratively with this procedure:


ne Z
X Λ0e (x)
Hj = σ0 q dx (3.21a)
e=1 Ωe u̇T0 Λ0e (x) u̇0
1
u̇∗ = H−1 T
j G0 p ; αj+1 = ; u̇0(j+1) = αj+1 u̇∗ (3.21b)
pT G0 u̇0
This procedure is ended when
¯ ¡ ¢ ¯¯
¯α
¯ j+1 −¡ D u̇¢0(j) ¯
¯ ¯ ≤ kd (3.22)
¯ D u̇0(j) ¯

being kd an appropriate value defined by the user.


In the procedure just illustrated the possibility of one or more rigid elements has
not been considered. If during the solution process one or more elements are
detected as rigid, and this can be done using the following check:
De
≤ ke (3.23)
Ve
the elements are eliminated from the EP domain and the constraints imposed by
the rigidities are applied changing in a suitable way matrix G0 .
In equation (3.23) De and Ve are respectively the dissipation and the volume of
the element e while the constant ke is an appropriate value defined by the user.
It would be important to check the solution, to ensure that the minimum of the
dissipation power is attained. Unfortunately, such a test can hardly be devised,
due to a well-known property of rigid plastic collapse: a corollary to the limit
analysis theorems states that the stress state associated with a mechanism is
defined in the plastic portion only, and no information on stresses is available in
the rigid region. Therefore the solution should be considered as a kinematically
admissible value, bounding from above the plastic multiplier. However, numerical
tests performed with different starting vectors produced the same final mechanism
suggesting the correctness of the procedure implemented.

3.3 Mesh Updating


The method proposed is based on a sequence of limit analyses performed on
updated meshes. This updating is done by multiplying the collapse mechanism,
obtained solving the limit problem, by a constant k such as:

max |vj | ≤ umax ∀j ∈ [1, nn ] (3.24)

where vj is the displacement vector in the node j, umax a value defined by the
user and nn the number of nodes in the discretization.

21
Chapter 3. Finite element solution of the limit problem of shells

Let us collect the nodal coordinates of the node j of the original mesh in the
vector sj , therefore the coordinates of the updated mesh are:

sup
j = sj + kvj ∀j ∈ [1, nn ] (3.25)

A new limit analysis is then performed on the updated mesh and the procedure
iterates up to the maximum number of steps defined by the user.

22
3.4. The axisymmetric shell element

3.4 The axisymmetric shell element


Axisymmetric shell structures, subject to axisymmetric loads and boundary con-
ditions, can be analyzed with axisymmetric elements, but some limitations are
implicit in the procedure. As illustrated in [43, 46, 96] an axisymmetic approach
can only deal with axisymmetric collapse while other collapse mode, such as di-
amond mode, cannot be recognized. Moreover only axisymmetric imperfections
can be considered. In spite of these drawbacks, the use of axisymmetric elements
permits to reduce a shell structure to a mono-dimensional one, decreasing in a
considerable way the number of the elements to be used and as a consequence the
time required to obtain the solution.
Therefore an axisymmetric element, labelled AX2P, has been developed for the
analysis of collapse of rigid plastic shells, and in this section its main characteri-
stics will be presented.

3.4.1 Nodal variables


The AX2P element is a 2 node linear axisymmetric shell element with 3 degrees
of freedom per node formulated in natural coordinates in order to separate the
rigid body motions from the natural modes.
Let us consider a global reference system (O, x, z) and a local reference system
(G, x0 , z 0 ) where G is the centroid of the element, x axis is coincident with the
shell axis and the direction of x0 is the same as the shell meridian (see fig. 3.1).
Let le be the length of the element e and ζ a dimensionless coordinate such that

w 

1
β` 

u 

z`
φ
G
z
x`, ζ x
w 

` l

2
β2`
u 

Figure 3.1: the AX2P element and the coordinate systems

− 12 ≤ ζ ≤ 12 . Due to the simplicity of the element the natural reference system


can be taken coincident with the local reference system.
The three degrees of freedom per node correspond to two displacements (axial

23
Chapter 3. Finite element solution of the limit problem of shells

and radial) and to one rotation as reported in figure 3.2. These nodal variables

w 

β1 u 

x
w

β2 u

Figure 3.2: nodal variables of the AX2P element

are collected in the vector u0e :


T
u0e = {u01 , w10 , β10 , u02 , w20 , β20 } (3.26)
(6×1)

or, if expressed in the global reference system:

ue = {u1 , w1 , β1 , u2 , w2 , β2 } (3.27)
(6×1)

Vectors ue and u0e are connected by the following relation:


· ¸
0 T
ue = u = Te u e (3.28)
T e

where  
cos φ sin φ 0
T = − sin φ cos φ 0 (3.29)
0 0 1

3.4.2 Rigid-body motions and natural modes


The axisymmetric shell element AX2P has six degrees of freedom but only five
modes imply dissipation, while one mode corresponds to the rigid motion in ver-
tical direction. The degrees of freedom are grouped in the vector ρc :
à !
ρ01
ρc = ρN
6×1 5×1

24
3.4. The axisymmetric shell element

where ρ01 is the rigid-body motion and the sub-vector ρN represents the natural
modes.
In the following a mathematical definition of rigid-body motion and natural modes
will be given.

Rigid-body motion

In an axisymmetric problem the rigid body motion corresponds to a displacement


in the vertical direction as pictured in figure 3.3.

u1 + u2 u0 + u02 w0 + w20
ρ01 = = 1 cos φ − 1 sin φ (3.30)
2 2 2

Figure 3.3: rigid-body motion of AX2P element

Natural straining modes

The five natural straining modes collected in the vector ρN are pictured in figure
3.4 and their definition is given by equations (3.31).

w1 + w2 u0 + u02 w0 + w20
q1 = = 1 sin φ + 1 cos φ (3.31a)
2 2 2
−w1 + w2 −u01 + u02 −w10 + w20
q2 = = sin φ + cos φ (3.31b)
2 2 2
−u1 + u2 −u01 + u02 −w10 + w20
q3 = = cos φ − sin φ (3.31c)
2 2 2
β1 + β2
q4 = (3.31d)
2
−β1 + β2
q5 = (3.31e)
2

25
Chapter 3. Finite element solution of the limit problem of shells

(a) natural mode q1 (b) natural mode q2 (c) natural mode q3

(d) natural mode q4 (e) natural mode q5

Figure 3.4: natural straining modes of AX2P element

3.4.3 Natural strains in the element

In order to obtain the strains into the element, Kirchhoff’s hypotheses will be
considered. Therefore the element is intended to be used with thin shells where
shear strains are negligible.
As introduced in subsection 3.4.1 the natural and local reference system are co-
incident, therefore no relation between them has to be provided. Strains in the
element can be easily written if the nodal variables are expressed in the LRS. The
displacements in the elements are:

µ ¶ µ ¶
0 1 1
u (ζ) = u01 − ζ + u2 0
+ζ (3.32a)
2 2
w0 + w20 β 0 − β20 5 β 0 − β20 2
w0 (ζ) = 1 − 1 + (w20 − w10 ) ζ + 1 ζ + (w20 − w10 ) ζ 3
2 2 4 2
(3.32b)

26
3.4. The axisymmetric shell element

The membranal strains and curvatures are obtained as follows:


1 du0 (ζ)
es = (3.33a)
le dζ
w0 (ζ) cos φ + u0 (ζ) sin φ
eθ = (3.33b)
r (ζ)
1 d2 w 0
χs = − 2 (3.33c)
le dζ 2
sin φ dw0 (ζ)
χθ = − (3.33d)
r (ζ) le dζ
Substituting equations (3.32) in (3.33) one obtains:
1 1
es = − u01 + u02 (3.34a)
le le
µ ¶ µ ¶ µ 2 ¶
1 sin φ 0 1 5 cos φ 0 ζ 1 cos φ 0
eθ = −ζ u + − ζ − ζ3 w + − β +
2 r (ζ) 1 2 4 r (ζ) 1 2 2 r (ζ) 1
µ ¶ µ ¶ µ 2 ¶
1 sin φ 0 1 5 3 cos φ 0 ζ 1 cos φ 0
+ζ u + + ζ +ζ w − − β
2 r (ζ) 2 2 4 r (ζ) 2 2 2 r (ζ) 2
(3.34b)
1
χs = − 2 (−6ζw10 + β10 + 6ζw20 − β20 ) (3.34c)
le
·µ ¶ µ ¶ ¸
sin φ 5 5
χθ = − − − 3ζ 2 w10 + ζβ10 + + 3ζ 2 w20 − ζβ20 (3.34d)
r (ζ) le 4 4
or in a more compact way:
² (ζ) = Be (ζ) u0e (3.35)
where © ªT
² (ζ) = es eθ χs χθ (3.36)
Taking into consideration the thickness of the shell the strains can be obtained
from the midplane quantity:

²s (z 0 , ζ) = es + z 0 χs (3.37a)
²θ (z 0 , ζ) = eθ + z 0 χθ (3.37b)

In the following the strains are collected in the vector ² (z 0 , ζ).

3.4.4 The element dissipation


The dissipation can be computed in terms of strains as obtained in equation
(2.31):
X XZ
D= De = σ0 λ̇dx (3.38)
e e Ωe

27
Chapter 3. Finite element solution of the limit problem of shells

where: q
λ̇ = ²̇0T (z 0 , ζ) Θ²̇0 (z 0 , ζ) (3.39)
Matrix Θ can be obtained from (2.30) by neglecting the terms related to shear
2×2
strains and considering the incompressibility condition (2.32c):
· ¸
1 4 2
Θ= (3.40)
3 2 4

Substituting equations (3.34) and (3.40) in (3.39) one obtains:


p
λ̇ = a + bz 0 + cz 02 (3.41)

where the coefficients a, b and c are:


4¡ 2 ¢
a= ės + ės ėθ + ė2θ (3.42a)
3
4
b = (2ės χ̇s + ės χ̇θ + ėθ χ̇s + 2ėθ χ̇θ ) (3.42b)
3
4¡ 2 ¢
c= χ̇s + χ̇s χ̇θ + χ̇2θ (3.42c)
3
The dissipation can now be integrated analytically on the thickness and by nume-
rical quadrature over the midplane of the element. The procedure is exhaustively
explained in appendix B.

28
3.5. The general shell triangular element

3.5 The general shell triangular element


The analysis of shell structures more general than axisymmetric ones requires the
use of a different element able both to mesh every shell geometry and to accurately
describe plastic strains in all directions. As seen in section 3.1 a formulation in
natural coordinates can be very useful and it has been verified that such elements
have a good behaviour in describing also complicated phenomena [15, 19]. More-
over, in order to have the widest field of application, the shell element has to be
able to describe every shell geometry and this has led to the choice of a triangular
element.
The element presented in this section is an evolution of the TRIC element al-
ready formulated in natural coordinates by the Argyris work group. The TRIC is
a 3-node shear deformable flat shell element developed in the late ’90s by Argyris
and his co-workers to simulate arbitrary isotropic and laminated composite shells.
It represents the natural evolution of three previous elements, the TRUMP, the
TRUNC and the LACOT [7–16, 18, 19]. The main characteristics of the TRIC
are the formulation based on the natural approach, the elimination of the locking
phenomena and its robustness and computational effectiveness compared to con-
ventional isoparametric elements. Recently the convergence requirements have
been theoretically proved be means of the non-consistent formulation proposed
by Bergan and co-workers [22, 24, 25, 51]. The original description of the element
can be found in [18], while in the following the modified formulation used to adapt
the TRIC element to the rigid-plastic analysis will be presented accordingly to
the symbols introduced in sections 3.1 and 3.2.

3.5.1 Nodal variables


The TRIC element is a triangular shell element with 3 nodes and 6 degrees of
freedom per node. The particular formulation allows the separation of the six rigid
body motions from the twelve natural modes that involve constant and higher
order deformations. These modes are the only responsible of the dissipation.
Let us consider a triangular element, its vertex are numerated from one to three
while the sides are called α, β, γ. Side α is opposite to vertex number 1, side β
to vertex number 2 and side γ to vertex number 3. The length of each side is
indicated by li where i = α, β, γ. Accordingly to section 3.1 the global reference
system is referred to as (O, x, y, z) while the local reference system, associated
to every element is referred to as (G, x0 , y 0 , z 0 ) where G is the centroide of the
element. The z 0 axis is normal to the element surface
The six degrees of freedom per node correspond to three displacements and three
rotations as reported in figure 3.6. These nodal variables are collected in the
vector u0e :
T
u0e = {u01 , v10 , w10 , θ10 , φ01 , ψ10 , u02 , v20 , w20 , θ20 , φ02 , ψ20 , u03 , v30 , w30 , θ30 , φ03 , ψ30 } (3.43)
(18×1)

29
Chapter 3. Finite element solution of the limit problem of shells

α
3
y’
∠γ
z’ x’ z
y
G
1 ∠α
∠β
β γ x
2

Figure 3.5: the TRIC element and the local coordinate system



  
  φ
  
ψ

 θ


Figure 3.6: nodal variables of the TRIC element

or in the global reference system:


T
ue = {u1 , v1 , w1 , θ1 , φ1 , ψ1 , u2 , v2 , w2 , θ2 , φ2 , ψ2 , u3 , v3 , w3 , θ3 , φ3 , ψ3 } (3.44)
(18×1)

Vectors ue and u0e are connected by the following relation:


 
T
 T 
 
 T 
u0e = 

 ue = Te ue
 (3.45)
 T 
 T 
T

where  
cos xc0 x cos xc
0y cos xc
0z
 c c 0z 
T = cos y x cos y y
0 0 cos yc  (3.46)
cos zc
0x cos zc
0y cos zc
0z

30
3.5. The general shell triangular element

3.5.2 Natural and cartesian strains


In this section the relations between natural and cartesian strains will be provided.
The natural reference system is referred to as (α, β, γ) and it is defined by the
three sides of the corresponding element.
Strains in the global reference system are defined as:
1
²ij = (vi,j + vj,i ) i, j = x, y, z (3.47)
2
where v is the cartesian displacements vector. The consistent strain and stress
vectors are:
² = {²xx , ²yy , ²zz , γxy , γxz , γyz } (3.48)
(6×1)

σ = {σxx , σyy , σzz , σxy , σxz , σyz } (3.49)


(6×1)

where γij are the engineering shear strains.

Membrane strains

Definition 3. Natural membrane strains ²m are defined as the three strains


measured in the natural coordinate system.
Membrane strains defined in the local and natural reference system are grouped
in the vectors ²0m and ²m :
²0m = {²x0 x0 , ²y0 y0 , γx0 y0 } (3.50)
(3×1)

²m = {²mα , ²mβ , ²mγ } (3.51)


(3×1)

In order to obtain the relations between cartesian and natural strains we get a
triangle in the non deformed and deformed configuration:
∆s2 (1 + ²mα )2 =∆x02 (1 + ²x0 x0 )2 + ∆y 02 (1 + ²y0 y0 )2
π
− 2∆x0 (1 + ²x0 x0 )∆y 0 (1 + ²y0 y0 ) cos(
+ γx 0 y 0 )
2
but ∆x0 = ∆s cos α and ∆y 0 = ∆s sin α, therefore with some simplifications:
(1 + ²mα )2 =(1 + ²x0 x0 )2 cos2 α + (1 + ²y0 y0 )2 sin2 α
π
− 2(1 + ²x0 x0 )(1 + ²y0 y0 ) cos α sin α cos(
+ γx 0 y 0 )
2
Subject to hypotheses of small strains the value of γxy is small and it can be
assumed that cos( π2 + γx0 y0 ) w −γx0 y0 . The above relation becomes:
(1 + ²mα )2 =(1 + ²x0 x0 )2 cos2 α + (1 + ²y0 y0 )2 sin2 α
+ 2γx0 y0 (1 + ²x0 x0 )(1 + ²y0 y0 ) cos α sin α

31
Chapter 3. Finite element solution of the limit problem of shells
 

∆   
ε α
∆ ∆  
ε  
 ∆  γ !
 
α  

∆  ∆   
ε    

Figure 3.7: the TRIC element and the local coordinate system

If the second and third order terms are neglected a simple relation between carte-
sian and natural strains can be obtained:

1 + 2²mα = cos2 α + 2 cos2 α²x0 x0 + sin2 α + 2 sin2 α²y0 y0 + 2γx0 y0 cos α sin α

Following the same procedure for the β and γ sides one obtains:
  c2 
s2αx0 cαx0 sαx0  ²x0 x0 
²mα αx0
²mβ  =   −1 0
cβx0 sβx0 cβx0 sβx0   ²y0 y0  ⇔ ²m = Tm ²m
2 2
(3.52)
²mγ c2 0 s2 0 cγx0 sγx0 γx0 y0
γx γx

where cαx0 = cos(αxd0 ) and sαx0 = sin(αx d0 ) are the cosine and sine of the angle
between the side α and the local axis x0 . Matrix Tm corresponds to the membranal
part of the transformation matrix introduced in equation (3.6):
 sβx0 sγx0 sγx0 sαx0 sαx0 sβx0 
d1 d2 d3
 cβx0 cγx0 cγx0 cαx0 cαx0 cβx0 
Tm = 
 s c d1+c s d2 d3

 (3.53)
0 0 0 0 s 0 c 0 +c 0 s 0 s 0 c 0 +c 0 s 0
− βx γx d1 βx γx − γx αx d2 γx αx − αx βx d3 αx βx

where:

d1 = (cβx0 sαx0 − cαx0 sβx0 ) (cγx0 sαx0 − cαx0 sγx0 )


d2 = (cγx0 sβx0 − cβx0 sγx0 ) (cαx0 sβx0 − cβx0 sαx0 )
d3 = (cαx0 sγx0 − cγx0 sαx0 ) (cβx0 sγx0 − cγx0 sβx0 )

Transversal strains

Definition 4. Natural transversal strains ²s are defined as the three strains


measured transversally to the natural coordinate system.

32
3.5. The general shell triangular element

y’

εsα εsα εx’z’

α
α εy’x’ x’

Figure 3.8: the transverse strains

It is supposed that a face of the hypothetical triangular base prism deforms


but all the other faces’ angles remain right angles. From simple trigonometric
considerations the following relations can be obtained:
   
γsα cαx0 sαx0 µ ¶
γsβ  = cβx0 sβx0  γx0 z0 ⇔ ²s = T−1 0
s ²s (3.54)
γy 0 z 0
γsγ cγx0 sγx0

It is important to observe that three transverse natural strains depend on only two
local shear strains and that the superscript −1 does not denote an inverse matrix,
but it has been reported only for congruency with the symbology introduced in
section 3.1.

3.5.3 Rigid-body motions and natural modes


The shell element TRIC has 18 degrees of freedom but only 12 modes imply
dissipation. The degrees of freedom are grouped in vector ρc :
 
ρ0
ρc =  6×1 
18×1 ρN
12×1

where the sub-vector ρ0 contains the rigid-body motions and the sub-vector ρN
the natural modes.
In the following a brief mathematical definition of rigid-body motions and natural
modes will be given, for a more detailed definition see [16].

Rigid-body motions
In the 3D-space three translations and three rotations are allowed.
¡ ¢T
ρ0 = ρ01 ρ02 ρ03 ρ04 ρ05 ρ06
6×1

In the local reference system the translation modes are easily detected as the

33
Chapter 3. Finite element solution of the limit problem of shells

ρ ρ  ρ 

   
     


  


ρ  ρ  ρ 
φ
  ψ 
     
θ

Figure 3.9: rigid-body motions of TRIC element

centroid displacement:

u01 + u02 + u03 v 0 + v20 + v30 w0 + w20 + w30


ρ01 = ρ02 = 1 ρ03 = 1 (3.55a)
3 3 3

Rotation motions in x0 and y 0 can be obtained writing the equation of the plane
for three points and using partial derivatives. For the definition of the rotation
in the z 0 axis see [17].

−x02 w10 + x03 w10 + x01 w20 − x03 w20 − x01 w30 + x02 w30
ρ04 = (3.56a)
2Ω
−y20 w10 + y30 w10 + y10 w20 − y30 w20 − y10 w30 + y20 w30
ρ05 = (3.56b)
2Ω
u01 x0α + u02 x0β + u03 x0γ + v10 yα0 + v20 yβ0 + v30 yγ0
ρ06 = − (3.56c)
4Ω

where Ω is the area of the element, u0i , vi0 , wi0 the displacements of vertex i in the
local reference system and x0i , yi0 the projections of the side i onto the local axes
x0 and y 0 .

34
3.5. The general shell triangular element

Natural straining modes


The 12 natural straining modes are subdivided into four categories:
¡ ¢
ρN = q 0m q S q A q d
12×1

Axial modes The axial modes are grouped in the sub-vector q 0m and they are
due to the stretching of the corresponding side (see fig. 3.10). The straining is
calculated as the elongation of the side scaled by its length. If u0i and vi0 are the
qmα qmβ qmγ
1/2 ∆lβ
1/2 ∆lα

1/2 ∆lβ

1/2 ∆lγ
1/2 ∆lα
1/2 ∆lγ

Figure 3.10: natural axial straining modes of TRIC element

local displacement in x0 and y 0 directions at node i the axial straining modes are:
 
 −cαx0 −sαx0 cαx0 sαx0
 u01
 0  0 0  v10 
qmα  cβx0
lα lα lα lα
  
 sβx0 −cβx0 −sβx0  u0 
qmβ
0 
=  lβ l 0 0 lβ lβ  
2
(3.57)
0  −c 0 −sβ 0  v20 

qmγ γx γx cγx 0 s γx 0
0 0 u3 
0
lγ lγ lγ lγ
v30

Symmetric bending modes The symmetric bending modes are grouped in


the vector q S and are calculated from the symmetric part of the sides bending
(see fig. 3.11). The nodal rotations are projected on the direction perpendicular
to the side and then the symmetric part is evaluated.
If θi0 and ϕ0i are the rotations of the node i in the x0 and y 0 directions the
straining modes are:
 0
θ1
    ϕ01 
qSα 0 0 sαx0 −cαx0 −sαx0 cαx0  0  
θ2 
qSβ  = −sβx0 cβx0 023 024 sβx0 −cβx0  
ϕ02 
 (3.58)
qSγ sγx0 −cγx0 −sγx0 cγx0 0 0  
 θ30 
ϕ03

35
Chapter 3. Finite element solution of the limit problem of shells

qSa qSb qSg

Figure 3.11: natural symmetric bending straining modes of TRIC element

Antisymmetric bending and shearing modes The antisymmetric bending


and shearing modes are grouped in the vector q A and they are calculated both
from the antisymmetric part of the bending and from the shearing part of the
sides. The nodal rotations are projected on the direction perpendicular to the
side and then the antisymmetric part is evaluated eliminating the rigid rotations
ρ04 and ρ05 .

b b b
q Aa q Ab q Ag

Figure 3.12: natural antisymmetric bending straining modes of TRIC element

 
θ10 − ρ04
    ϕ01 − ρ05 
qAα 0 0 sαx0 −cαx0 sαx0 −cαx0  
 θ20 − ρ04 
qAβ  = sβx0 −cβx0 0 0 sβx0 
−cβx  0  (3.59)
ϕ2 − ρ05 
0

qAγ sγx0 −cγx0 sγx0 −cγx0 0 0  


 θ30 − ρ04 
ϕ03 − ρ05

As pictured in figures 3.12 and 3.13 it can be seen that the antisymmetric modes
are due to the bending q bA and shearing q sA part. Their value will be obtained
minimizing the dissipation energy of the structure.

36
3.5. The general shell triangular element


q Αα q Αβ
q Αγ

1/2 Αγ
1/2  Αα

1/2   Αβ

Figure 3.13: natural antisymmetric shearing straining modes of TRIC element

Azimuthal or drilling modes The drilling straining modes are grouped in


the vector q d and are calculated from the nodal rotations perpendicular to the
element plane (see fig. 3.14).

qdγ

qdα

qdβ

Figure 3.14: natural drilling straining modes of TRIC element

   0 
qdα ψ1 − ρ06
qdβ  = ψ20 − ρ06  (3.60)
qdγ ψ30 − ρ06

3.5.4 Natural strains in the element


As obtained in section 3.2 the energy dissipation can be calculated integrating
the specific dissipation over the element. The specific dissipation D̂ depends on
strains, expressed either in local or natural coordinates. In this section a relation
between natural modes and natural strains is obtained.
Internal coordinates of the element are now introduced: these are useful in the
description of the position of a point over the element.
Yα Yβ Yγ
ηα = ηβ = ηγ =
lα lβ lγ
Zα Zβ Zγ
ζα = ζβ = ζγ =
hα hβ hγ

37
Chapter 3. Finite element solution of the limit problem of shells

ηβ,


β β

ζβ,Ζβ
ζα,Ζα
η α,


ζγ,Ζγ α

ηγ,


γ


Figure 3.15: internal coordinate of TRIC element

where the relation ζα + ζβ + ζγ = 1 holds. The jacobian matrix of the transfor-


mation between ζ and η coordinates is here reported:
 
0 1 −1
∂ζi
= −1 0 1 (3.61)
∂ηj
1 −1 0

Strains from axial and bending modes


If a section perpendicular to the mid-plane of the shell in the undeformed state
remains perpendicular after the deformation occurs then their displacements in
natural coordinates will be given by:
X ∂wj0
ui = u0i − z i, j = α, β, γ (3.62)
j
∂Yi

where ui are the displacements parallel to the side i and wj0 the displacements
perpendicular to the element plane due to the symmetric bending mode qSj .
Therefore the membranal strains in the natural reference system are expressed by
the equation 3.63:
X ∂ 2 wj0
0
²mi = qmi −z i, j = α, β, γ (3.63)
j
∂Yi2

where the subscript mi refers to the membrane strains in the i direction.


The displacements in z 0 direction are due to the symmetric and antisymmetric
bending modes while the drilling modes impose only in-plane deformations. In

38
3.5. The general shell triangular element

the following, symmetric and antisymmetric bending mode are treated separately
for the sake of simplicity.

Symmetric bending modes Displacements in z 0 direction due to the sym-


metric bending modes q S are described by the following expression:
1 1 1
wS0 = lα ζβ ζγ qSα + lβ ζγ ζα qSβ + lγ ζα ζβ qSγ (3.64)
2 2 2
The boundary conditions are verified, for example considering only the natural
mode qSα :
wS0 |ζβ =0 = 0 wS0 |ζγ =0 = 0 (3.65)
∂wS0 1 ∂wS0 1
|ζα =ζβ =0 = − qSα |ζα =ζγ =0 = qSα (3.66)
∂Yα 2 ∂Yα 2
Along the sides β = 0 and γ = 0 the rotations are not constant but go to 0 as
(1 − ζα ). The displacements in the plane of the element are obtained by means
of the equation (3.62) and a subsequent derivation along Yj leads to the natural
strains:
∂wS0
uSj = −z (3.67)
∂Yj
∂ 2 wS0
²mSj = −z (3.68)
∂Yj2

and in explicit form, using the jacobian reported in equation (3.61):


qSα
²mSα = z (3.69)

qSβ
²mSβ =z (3.70)

qSγ
²mSγ =z (3.71)

Antisymmetric bending modes As introduced in subsection 3.5.3 the ben-


ding antisymmetric modes are a part of the whole antisymmetric mode. In order
to identify the shear and the bending part a bending parameter χ is introduced
so that:
q bA = χq A
q sA = (1 − χ)q A
Therefore the displacements in z 0 direction due to the antisymmetric bending
modes q bA are described by the following expression:

0 1 b 1 b 1 b
wA = lα ζβ ζγ (ζβ − ζγ )qAα + lβ ζγ ζα (ζγ − ζα )qAβ + lγ ζα ζβ (ζα − ζβ )qAγ (3.72)
2 2 2

39
Chapter 3. Finite element solution of the limit problem of shells

The boundary conditions are verified, for example considering only qAα as in the
symmetric case:
0 0
wA |ζβ =0 = 0 wA |ζγ =0 = 0 (3.73)
0 0
∂wA ∂wA 1 b
|ζα =ζβ =0 = |ζα =ζγ =0 = q (3.74)
∂Yα ∂Yα 2 Aα

Along the sides β = 0 and γ = 0 the rotations are not constant but go to 0 as
(1 − ζα )2 . As in the symmetric case the displacements in the plane of the element
are obtained by means of the equation (3.62) and a subsequent derivation along
Yj leads to the natural strains:

∂ 2 wA
0
²mAj = −z 2 (3.75)
∂Yj

If computations are performed, the following expressions for the strains due to
the antisymmetric bending modes are obtained:
1 £ b b b
¤
²mAα = −z 2
3lα (ζγ − ζβ )qAα + lβ ζα qAβ − lγ ζα qAγ (3.76)

1 £ b b b
¤
²mAβ = −z 2 −lα ζβ qAα + 3lβ (ζα − ζγ )qAβ + lγ ζβ qAγ (3.77)

1 £ b b b
¤
²mAγ = −z 2 lα ζγ qAα − lβ ζγ qAβ + 3lγ (ζβ − ζα )qAγ (3.78)

Strains from antisymmetric shearing modes


The antisymmetric shearing modes involve only natural transverse strains. From
figure 3.13 it can be observed that the variation of strains over the elements is a
linear one and it is expressed by the following relation:
qAi s
γsi = (1 − ζi ) (3.79)
2

Strains from azimuthal modes


In the assembly of flat elements the azimuthal degrees of freedom are often ne-
glected and substituted with fictitious couple Mz . Many works deal with this
argument and the reader can refer to some publications listed in the bibliogra-
phy [4, 5, 23, 65, 84, 112, 130]. In the present formulation of the TRIC element the
drilling modes effect has been completely considered, differently from the original
formulation where fictitious terms were employed in order to eliminate the singu-
larity in the stiffness matrix [18].
The azimuthal modes give rise to in-plane deformations like the one pictured in
figure 3.16. It is clear that only natural membrane strains are involved. In order

40
3.5. The general shell triangular element

qda
3 y’

r R

x’
1 vga(hg)
2

Figure 3.16: strains due to the drilling mode qdα

to calculate these strains the displacements over a linear segment ρ, linking a ver-
tex to a point R lying on the opposite side, are assumed similar to those imposed
at the sides. If we refer to the drilling mode qdα the displacements on the segment
ρ are: ¡ ¢
vρα (ηρ ) = ηρ − 2ηρ2 + ηρ3 lρ1 qdα (3.80)
where lρ1 is the segment length and 0 < ηρ < 1 the dimensionless coordinate.
The displacements direction is perpendicular to the segment considered and some
further developments lead to explicit relations. Let P be a point on the segment
ρ with internal coordinate (ζ1 , ζ2 , ζ3 ). ei is the unit vector in the direction of side
i and (x0j , yj0 ) the coordinate of point j in the local reference system. Therefore
the following relations hold:

x03 − x02 y 0 − y20


eα = i+ 3 j = cαx0 i + sαx0 j (3.81)
lα lα
x0 − x03 y 0 − y30
eβ = 1 i+ 1 j = cβx0 i + sβx0 j (3.82)
lβ lβ
x0 − x01 y 0 − y10
eγ = 2 i+ 2 j = cγx0 i + sγx0 j (3.83)
lγ lγ
OP = O1 + 1P = x01 i + y10 j + ζβ lγ eγ − ζγ lβ eβ (3.84)

and the coordinates of P in the local reference system are:

x0P = x01 (1 − ζβ − ζγ ) + ζβ x02 + ζγ x03


yP0 = y10 (1 − ζβ − ζγ ) + ζβ y20 + ζγ y30

The unit vector in the direction of the segment ρ, from vertex 1 to point R, is:

x0P − x01 y 0 − y10


eρ1 = i+ P j = x̃0P 1 i + ỹP0 1 j
|1P | |1P |

41
Chapter 3. Finite element solution of the limit problem of shells

It is now possible to obtain the unit vector normal to the segment ρ and directed
as the displacements:
nρj = k ∧ eρj = −ỹP0 j i + x̃0P j j j = 1, 2, 3 (3.85)
In order to find the displacement of point P it is now necessary to know the
dimensionless coordinate ηρ and this is possible if the length of segment ρ is
known. It can be observed that the point R has internal coordinates (0, ζβR , ζγR )
where ζβR and ζγR are related by the following relation of proportionality (see figure
3.17:
ζγP lγ ζβP lβ ζβP ζβR
1P = = ⇒ P = R
sin χ sin τ ζγ ζγ
It is now possible to calculate ζβR and ζγR :

3
lb

P R
t
c
1

lg 2

Figure 3.17: scheme used to find the position of the point R

½ζ R + ζ R = 1 ½ζ R = ζβP
β γ β P
ζβ +ζγP
ζP → ζγP
ζβR = ζγR ζβP ζγR = P
γ ζβ +ζγP

Therefore the segment length and the dimensionless coordinate are:


r³ ´2 ³ ´
lρ1 = ζβR x02 + ζγR x03 − x01 + ζβR y20 + ζγR y30 − y10 (3.86)
¯ ¯
|1P | ¯¯ x0P − x01 ¯
¯
ηρ1 = =¯ R 0 ¯ = ζβP + ζγP = 1 − ζα (3.87)
lρ1 ¯ ζβ x2 + ζγR x03 − x01 ¯

and the displacement vector for a point P on the segment 1R is:


vρα = (ηρ − 2ηρ2 + ηρ3 )lρ1 qdα nρ = (ζα2 − ζα3 )lρ1 qdα (−ỹP0 1 i + x̃0P 1 j) (3.88)
vρα = ζα2 qdα (−(y10 (ζα − 1) + ζβ y20 + ζγ y30 )i + (x01 (ζα − 1) + ζβ x02 + ζγ x03 )j)
(3.89)

42
3.5. The general shell triangular element

In order to calculate the strains in the natural system the displacement vector is
projected on the element sides.
vραki = vρα • ei i = α, β, γ (3.90)
The strains are calculated with the following formula:
X ∂vρjki
²mi = i, j = α, β, γ (3.91)
j
∂Yj

where the index j refers to the drilling mode and the index i to the strain direction.
Omitting the calculus (which can be found in appendix A) the contribute of
drilling modes to natural strains is:

²mα = 4 ζα (ζβ qdβ − ζγ qdγ ) (3.92)
lα2

²mβ = 4 2 ζβ (ζγ qdγ − ζα qdα ) (3.93)


²mγ = 4 ζγ (ζα qdα − ζβ qdβ ) (3.94)
lγ2

Total natural strains


It is now possible to write the natural strains ²̄ as a function of natural modes
ρ̃N by means of two matrixes bm and bs . The vector ρ̃N contains the so called
extended natural modes where the antisymmetric modes are split in the bending
and shearing parts. · ¸
b 0
²̄ = m ρ̃ (3.95)
0 bs N
where:
¡ ¢T
ρ̃N = q 0m qS q bA qd q sA (3.96)
15×1
 −3z(ζγ −ζβ ) −zlβ zlγ

z
100 lα 0 0 lα l2
ζα l2
ζα 0 4 lΩ
2 ζα ζβ −4 lΩ
2 ζα ζγ
 α
−3z(ζα −ζγ )
α
−zlγ
α α

= 
z zlα Ω Ω
bm 010 0 0 ζβ ζβ −4 ζα ζβ 0 4 ζβ ζγ
l2 l2 l2 2
 lβ
β

β β
l
β 
z −zlα zlβ −3z(ζβ −ζα )
001 0 0 lγ 2

ζγ 2

ζγ lγ 4 lΩ2 ζα ζγ −4 lΩ2 ζβ ζγ
γ γ

(3.97)
" 1−ζα 0 0
#
2
1−ζβ
bs = 0 2 0 (3.98)
1−ζγ
0 0 2

3.5.5 The element dissipation


The rigid-plastic problem requires the definition of the dissipation. This is done
in terms of natural straining modes. As seen in subsection 2.2.3 the dissipation
can be expressed in terms of strains:

43
Chapter 3. Finite element solution of the limit problem of shells

X XZ
D= De = σ0 λ̇dx (3.99)
e e Ωe

where: p
λ̇ = ²̇0T Θ²̇0 (3.100)
The contribution of an element to the dissipation is split in two parts: the con-
tribution of membrane strains and the contribution of transversal shear strains.
This can be done because there is no coupling between membrane strains and the
transverse natural modes (see eq. 3.95). Equation (3.101) becomes:
XZ q
D= σ0 λ̇2m + λ̇2s dx (3.101)
e Ωe

where
λ̇2m = ²̇0T 0
m Θm ²̇m (3.102a)
λ̇2s = ²̇0T 0
s Θs ²̇s (3.102b)
Matrix Θm and Θs are obtained by equation (2.30) keeping in mind the incom-
pressible relation (2.32c): ²̇z0 z0 = −²̇x0 x0 − ²̇y0 y0 .
 
4 2 0
1
Θm = 2 4 0 (3.103)
3
0 0 1
· ¸
1 1 0
Θs = (3.104)
3 0 1
Relations (3.102) have to be expressed in terms of natural strains. This can be
done immediately for the membrane part while for the contribution of transverse
modes a short manipulation is necessary.
T
λ̇2m = ²̄˙ Tm T̄ m Θm T̄ m ²̄˙ m (3.105)
or in a more compact way:
T
λ̇2m = ²̇m Γm ²̇m (3.106)
Regarding the contribution of transverse strains, from eq. (3.54) it can be seen
that the two cartesian transversal strains in local coordinates have to be expressed
in terms of the three natural transversal strains. This can be done by solving only
two equations at time and this leads to three different solutions:
µ ¶ · ¸µ ¶
γx 0 z 0 1 lα yβ0 −lβ yα0 γsα
= (3.107)
γy0 z0 2Ω −lα x0β lβ x0α γsβ
µ ¶ · ¸ µ ¶
γx0 z0 1 lβ yγ0 −lγ yβ0 γsβ
= 0 0 (3.108)
γy 0 z 0 2Ω −l x
β γ l x
γ β γsγ
µ ¶ · ¸ µ ¶
γx 0 z 0 1 lγ yα0 −lα yγ0 γsγ
= 0 0 (3.109)
γy0 z0 2Ω −l x
γ α l x
α γ γ sα

44
3.5. The general shell triangular element

where Ω is the area of the element, lµ the side µ length and x0µ , yµ0 the projections
of side µ on local reference axes. Now three different values of λ̇2s can be obtained
and their mean value is used in order to calculate the dissipation due to the shear
strains:
 T " # 
γ̇sα 2
lα ( 2
lβ 2
+lγ ) −lα lβ (yα
0 0
yβ +x0α x0β ) −lγ lα (yγ0 yα0
+x0γ x0α ) γ̇sα
1   
2
λ̇s = γ̇ sβ −l α β( α β
l y 0 0
y +x α β)
0 0
x l β( γ
2
l 2
+l α)
2
−l β γ( β γ
l y 0 0
y +x β γ)
0 0
x γ̇sβ 
36Ω2 −lγ lα (yγ0 yα +x0γ x0α ) −lβ lγ (yβ yγ +x0β x0γ ) (lα2 +lβ2 )
γ̇sγ 0 0 0 2
lγ γ̇sγ
(3.110)
or in a more compact way:
T
λ̇2s = ²̇s Γs ²̇s (3.111)

Plastic flow, and subsequently the dissipation power, can now be expressed in
terms of natural strains:

T T
λ̇2 = ²̇m Γm ²̇m + ²̇s Γs ²̇s (3.112)

The symmetric matrixes Γm and Γs are functions of the element attitude only
while the natural strains depend on the position where they are calculated.
The procedure adopted in order to calculate the energy dissipation of the element
is reported in appendix B.

3.5.6 Nodal equivalent loads


When an element is subject to a pressure load, normal to its surface, the effects
have to be expressed by means of appropriate nodal forces and moments applied
to the nodes. Let us consider a uniform pressure p in the direction of z 0 axis (see
fig. 3.18) and the displacements of the element surface:

w0 (ζα , ζβ , ζγ ) = ρ03 + ρ04 y 0 − ρ05 x0 + wS0 + wA


0
(3.113)

where single terms of equation (3.113) have been defined in the previous subsec-
tions. The work equality is now imposed:
Z
pw0 (ζα , ζβ , ζγ ) dA = p0T 0
e ue (3.114)
Ae

where the vector p0 collects the nodal equivalent loads expressed in the local
reference system.
Without loss of generality the local axis x0 can be imposed parallel to the side α.

45
Chapter 3. Finite element solution of the limit problem of shells

3 z'
y'

x'
G
1
2

Figure 3.18: the pressure load

Therefore, if integrations are performed one obtains:


 
0
 0 
 Ω 
 
 Ω 3 
 (lγ sγx0 − lβ sβx0 ) 
 Ω24 
 (lβ cβx0 − lγ cγβx0 )
 24 
 0 
 
 0 
 
 0 
 
 Ω 
p0 = 
 Ω
3 
 (3.115)
 Ω − 24 lγ sγx0 
 
 24 (lγ cγx0 − lα ) 
 0 
 
 0 
 
 0 
 
 Ω 
 3 
 Ω 
 24 lβ sβx 
0
 Ω (lα − lβ cβx0 ) 
24
0

Between displacements in local and global reference system relation 3.45 holds,
therefore:
pTe = p0T
e Te (3.116)

3.5.7 The minimization procedure


In section 3.2 the solution procedure has been described in a general way and the
method proposed can be applied to any element without notable modifications.
Actually, for the AX2P element presented in section 3.4 the procedure applies
inalterate and only the matrix Λ0e , typical of the element, has to be calculated.

46
3.5. The general shell triangular element

On the contrary, the natural modes of the TRIC element depends on the di-
splacements ue and on the bending coefficients χe . Therefore both have to be
considered by the minimization procedure. In particular the bending coefficient
is assumed constant in the respective element and its effects are restricted to the
element it self. In the following, the necessary modifications will be described,
deferring to the appendixes for the details of the procedure.
In the original work [18], the antisymmetric modes were split in the bending
and shear part by considering their stiffnesses: a closed form was obtained and
in a subsequent publication the benefits of the procedure were explained [14].
In particular, the introduction of shear modes permits to avoid the locking phe-
nomenon, without having recourse to different expedients needed when using other
elements [31, 32, 39, 77, 114].
In the present work the TRIC element has been modified in a suitable way to
adapt it to the limit analysis. As a consequence the bending coefficients χe cannot
be obtained in a closed form but they have to be considered in the minimization
procedure. This permits to obtain the vector χ of bending coefficients that as-
sure the lower dissipation. In this way the shear effects are accounted for, thus
avoiding locking phenomena.
Let us consider the expanded natural modes grouped in the vector ρ̃N defined in
equation (3.96). Being χe constant over the element e the shear and bending part
of the antysimmetric modes are obtained in a simple way:
qbA = χe qA (3.117a)
qsA = (1 − χe ) qA (3.117b)
where qA is the vector of the antisymmetric modes obtained as explained in
equation (3.59).
The minimization problem (3.13) becomes:
s = min D (u̇, χ)
u̇,χ
ne Z
X q (3.118a)
= min σ0 u̇T LTe TTe CTe ΛTe (χ) Πe Λe (χ) Ce Te Le u̇dx
u̇,χ Ωe
e=1
subject to pT u̇ = 1 (3.118b)
where:
· ¸ · ¸T · ¸· ¸
Πm 0 b 0 Γm 0 bm 0
Πe = = m (3.119)
0 Πs 0 bs 0 Γs 0 bs
 
I
 I 
 

Λe (χ) =  χe I  (3.120)

15×12  I
(1 − χe ) I
and

47
Chapter 3. Finite element solution of the limit problem of shells

• u is the vector of nodal variables of the whole structure in the global re-
ference system;
• Le is the connectivity matrix of the element e;
• Te is the coordinate transformation matrix defined in (3.45);
• χ is the vector of bending coefficient χe ;

• Λe (χ) is the matrix relating the natural modes and the extended natural
modes.

The problem can be solved minimizing the following Lagrangean function:


¡ ¢
L (u̇, χ, α) = D (u̇, χ) − α pT u̇ − 1 (3.121)

At solution the following equalities holds:


∂D
− αp = 0 (3.122a)
∂ u̇
∂D
=0 (3.122b)
∂χ
Stationarity respect to the displacements leads to the following non-linear expres-
sion:
XZ Z
σ0 dx u̇ − αp = 0 ⇒ Ku̇ − αp = 0 (3.123)
e Ωe λ̇ e

where
Z = LTe TTe CTe ΛTe (χ) Πe Λe (χ) Ce Te Le (3.124)
while the bending coefficient of the element e that leads to a minimum of the
lagrangean function can be obtained as follows:
R
ρ̇T B ρ̇ dx
Ωe N λ̇e N
χe = − R (3.125)
q̇T A q̇A dx
Ωe A λ̇ e

A more detailed dealing of the argument can be found in appendix C. Solution


of the non-linear equation (3.123) is provided by means of an iterative procedure
similar to that reported in equation (3.21b). If some elements are found to dis-
sipate less than a given tollerance value (see equation (3.23)) they are set rigid.
Proper constraints are imposed to the solution procedure by means of the matrix
G0 introduced in equation (3.14).

48
Chapter 4

Test example: simply


supported plate

In this chapter some simply supported plates subject to transversal pressure are
studied in order to validate the approach presented in chapter 2 and the imple-
mentation by means of the finite element method developed in chapter 3.
Plates and stiffened plates are typically found in ship hulls, platform decks, civil
buildings and many other structures where they have to withstand normally static
loads and in some cases dynamic forces and explosions, with possible impact of
dropped objects. In this case test experience shows that the use of static plastic
displacement models, combined with momentum and energy considerations, could
be useful. In the formulation of the present approach these aspects have not been
considered, but according to the author they could be included with little effort.
The analyses here presented are all concerned with a simply supported plate sub-
ject to a static transversal pressure load. The limit load furnished by the method
proposed in this work, will be verified by comparison with some upper bound va-
lues available in the literature. Moreover some incremental analyses, performed
by means of the commercial code ABAQUS, will be used in order to validate the
collapse behaviour foreseen by the sequential limit analysis. The investigation is
not restricted to the thin plate only, but thick plates will also be considered in
order to check the capabilities of the proposed element.
The geometry and the mechanical characteristics of the studied plates are re-
ported in section 4.1. Different plates are considered, with slenderness β = 2a/h
changing from β = 1 up to β = 100. An estimate of the limit load, by means of a
mechanism model, is presented in section 4.2 for the two extreme cases. Finally
in sections 4.3 and 4.4 the limit load and the post-collapse behaviour, obtained
by means of the sequential limit analysis, will be discussed and compared with
the results available in literature and with the elasto-plastic analyses performed
by the commercial code ABAQUS.

49
Chapter 4. Test example: simply supported plate

4.1 Geometry and mechanical characteristics


All the studied plates are of square shape and simply supported at the four sides.
Their geometric and mechanical characteristics are reported in table 4.1 while a
sketch is pictured in figure 4.1. The slenderness β is defined as:




  


Figure 4.1: simply supported plate

Side 2a (mm) 20
Thickness h (mm) 0.2÷20
Slenderness β 100÷1
Yield stress σ0 (MPa) 200
Young modulus E (GPa) 200

Table 4.1: Geometric and mechanics characteristics of plates

2a
β=
h
Simply supported plates subject to transversal pressure represent one of the most
studied case. In literature many results are available and they are obtained refer-
ring to a dead pressure p0 defined as:
M0
p0 = (4.1)
a2

50
4.2. Mechanism model

where M0 is the limit bending moment


h2
M0 = σ0 (4.2)
4
and h the plate thickness. If equation (4.2) is subsituted in equation (4.1) the
dead pressure can be expressed as a function of the slenderness:
σ0
p0 = (4.3)
β2
Therefore the dead pressure p0 considered in the analyses is not constant but
depends on the slenderness of the plate considered. Thus the limit pressure can
be expressed as p = αp0 where α is the plastic multiplier obtained by the limit
analysis.

4.2 Mechanism model


In this section the limit load of two different plates will be obtained by applying a
suitable mechanism collapse. In particular, a thin plate with β = 100 and a very
thick plate with β = 1 will be considered.

4.2.1 Thin plate


Let us consider a thin plate, with slenderness β = 100, subject to a transversal
pressure p. Being the plate thin, a mechanism like that pictured in figure 4.2(a) is
supposed to well describes the collapse of the plate. Kirchhoff’s hypotheses are as-
sumed. The dotted diagonals represent the hinge lines: these have to be regarded
as the limit case of narrow strips exhibiting cylindrical curvature. Hence defor-
mations vanish everywhere except in the hinge lines where plastic deformations
concentrate. Due to the presence of the uniform pressure, deformations occur in a
rather large zone, but, in order to simplify the computations, the aforementioned
hypothesis is considered. If a virtual displacement δ, normal to the plate surface,
is applied in the point E, the rotations of the four rigid parts are:
δ
• part 1: θ̇x = a θ̇y = 0

• part 2: θ̇x = 0 θ̇y = − aδ

• part 3: θ̇x = − aδ θ̇y = 0


δ
• part 4: θ̇x = 0 θ̇y = a

Along the hinge lines the relative rotation is:


√ δ
θ̇r = 2
a

51
Chapter 4. Test example: simply supported plate

y y

A B A B
q1
Yield lines
1
vertical displacement x
x constant over the plate
4 2 q2
q4
E
3
Yield lines
q3
D C D C

(a) bending collapse mechanism (b) punch collapse mechanism

Figure 4.2: two different collapse mechanism for a square plate

The limit load is obtained by equating the internal and the external work:
Z
4 2
Le = pwdA =
pa δ
A 3
2 √ δ√ 16
Li = 4 √ M 0 2 2a = √ M0 δ
3 a 3
M0
Le = Li ⇒ p = 6.92 2
a

Thus, if p0 represents the dead pressure, the plastic multiplier α is constant:

M0
p = 6.92 = 6.92p0 ⇒ α = 6.92 (4.4)
a2

This simple mechanism, adopted to introduce the calculation of the limit load of a
thin plate, is the same mechanism adopted by Baldacci et al. [20], but many other
collapse mechanisms have been presented in the past, each one leading to different
multipliers. For sake of comparison some of these plastic multipliers are reported
in table 4.2. In the same table, where available, the lower bound multipliers
obtained by equilibrated approaches are also indicated. In the following sections,
the more accurate plastic multiplier obtained by Iliouchine [66] will be used as
the analytical reference value for thin plates.

(1) corners are allowed to lift up.

52
4.2. Mechanism model

M0
dead pressure, p0 = a2

Reference Lower bound Upper bound Method


Baldacci et al. [20] 5.15 6.92 Analytical
Capsoni, Corradi [33] — 6.256 Numerical
(1)
Del Rio Cabrera [28] — 6.348 Analytical
Hodge [57] 5.16 6.98 Analytical
Hodge, Belytschko [58] 6.216 6.42 Numerical
Iliouchine [66] — 6.6 Analytical

Table 4.2: Upper and lower bounds for a simply supported plate subject to
transversal pressure load

4.2.2 Thick plate


Let us now consider a thick plate, with slenderness β = 1, subject to a transversal
pressure p. From the geometrical point of view a slenderness β = 1 corresponds to
a cube, therefore the theory of plates cannot be applied. Actually, the proposed
finite element is not suitable for analyses on such geometry, but some characte-
ristics can be highlighted with this example.
The collapse is supposed to happen by means of a punch mechanism such as the
one pictured in figure 4.2(b) where plastic strains develop near the support. If a
virtual displacement δ normal to the plate surface is applied, the collapse load can
be calculated by equating the internal and the external work in the hypothesis
that only shear strains are involved:
Z
Le = pwdA = 4pa2 δ
A
8 16a2
Li = 4τ0 2ahδ = √ ahσ0 δ = √ σ0 δ
3 3β
4 σ0
Le = Li ⇒ p = √
3 β
Thus, if p0 represents the dead pressure, the plastic multiplier α become:
4 σ0 4 4
p= √ = √ βp0 ⇒ α = √ β (4.5)
3 β 3 3
The plastic multipliers obtained both for the thin and the thick plates are reported
in figure 4.3 where, for a better illustration, the logarithmic scale has been used
for the β-axis.

53
Chapter 4. Test example: simply supported plate

4.3 Finite element limit analysis


In this section the approach developed in the present work will be used in order to
obtain the limit load of simply supported plates subject to transversal pressure.
Mesh accuracy and thickness variation will be considered. The latter investigation
will bear out the good behaviour of the shell element also for medium thick plates.
Being the problem symmetric, only a quarter of plate has been considered in the
analyses.
As done in section 4.2 the plates are considered subject to a dead pressure p0 that
depends on the plate slenderness β:
M0 σ0
p0 = 2
= 2
a β

4.3.1 Mesh variation


In order to study the accuracy and effectiveness of the element developed in section
3.5, some analyses have been performed on a thin plate, with slenderness β = 100,
using different meshes. The models characteristics are indicated in table 4.3. The
plastic multiplier α and the bending parameter χ, introduced in subsection 3.5.4,
are also reported.
Being the TRIC element non-conforming, the convergence to the exact solution
can be approached from lower values despite the kinematic approach used. This
happens for the plate under consideration as it is evident from figure 4.3. This
is not a drawback, as demonstrated by [] where a mixed formulation was used
to achieve better results. For sake of comparison the upper bound multiplier,

Nodes Elements Degrees of freedom Collapse multiplier χ


9 8 24 5.93 0.997
15 16 49 6.11 0.988
27 36 109 6.16 0.971
49 72 217 6.15 0.932
64 98 295 6.16 0.908
91 148 445 6.22 0.862

Table 4.3: results for a simply supported plate subject to transverse pressure for
β = 100

furnished by Iliouchine [66], and the results obtained in [33,102], using a Kirchhoff
finite element, are reported in figure 4.3. Both the numerical analyses clearly
converge to an upper bound multiplier lower than the analytical one. Moreover,

54
4.3. Finite element limit analysis

being the Kirchhoff element compatible, the value of 6.256 reported in [33] can
be considered as a reference multiplier that is well approximated by the analyses
performed with the TRIC element.
Unlike the quadrilateral finite element used in [33], the TRIC element can also
deal with shear strains by means of the bending parameter χ. Its mean value
reported in table 4.3, seems to be influenced by the element dimension, but the
minimization procedure always guarantees the most accurate result. In conclusion
the comparison reveals a good behaviour of the element and of the approach used
when applied to thin plates.

7
Collapse multiplier

6. 5


 
  
5. 5 [33, 102]
     

0 2 5 50 75 100

Elements

Figure 4.3: collapse multiplier obtained using different meshes

4.3.2 Thickness variation


Among the characteristics of the original TRIC element [18], one of the most
notable is the ability to deal with medium thick shells. The analysis of medium
thick plates requires the use of elements based on Mindlin’s theory, which can
account for the shear effect over the thickness. Unfortunately, locking phenom-
ena occur when these elements are used for the discretization of thin plates, and
some tricks have to be used in order to solve this problem [33, 107]. On the other
hand the TRIC element introduces the antisymmetric shear modes and, in its
original formulation, the locking problem is avoided in a very simple but efficient
way [18]. In the present formulation the antisymmetric shear modes are obtained
by introducing a parameter χ and minimizing the dissipation with respect to it
(see section 3.5.7). This approach allows the study of thin and medium thick
shells avoiding locking problems occuring with classic Mindlin’s formulation.

55
Chapter 4. Test example: simply supported plate

In order to check the ability of the element in the study of thin and thick struc-
tures some analysis have been performed on plates of slenderness varying between
β = 1 and β = 100. The 98 elements mesh has been chosen for the analyses. The

Collapse multiplier 6


 
 
[33, 107]
2


 
 !"  

0

1 10 100

β
Figure 4.4: Collapse multiplier for different slendernesses

two plastic mechanism presented in section 4.2 give an idea of what happens in
the extreme cases: the folding mechanism reported in figure 4.2(a) well describes
the collapse phenomena associated to very thin plates, while the punch mecha-
nism (see figure 4.2(b)) can be used with very stocky plates (in this case β = 1
refers to a cube). The simple calculations performed in section 4.2 permit to draw
the two curves reported in figure 4.4, which identify an approximate upper bound
envelope of the plastic multiplier for plates with varying slenderness. The results
obtained by finite element limit analysis are also reported and compared with the
analytical solutions proposed.
From the values obtained it can be stated that the TRIC element performs very
well in the limit analysis of thin and medium thick plates, where results are com-
parable with the ones obtained by Capsoni and Corradi using Mindlin’s elements
with mixed formulation; nevertheless, for very stocky plates the TRIC element
furnishes the lowest values but these can not be considered a better approxima-
tion of the exact plastic multiplier because of the non-conforming formulation of
the element.
The good behaviour of the procedure, introduced in order to deal with shear
strains, is confirmed by the graph of figure 4.5, where the values assumed by the
bending parameter χ are plotted vs. the slenderness parameter β. As expected
when the slenderness decreases the bending parameter reduce it self and this cor-
responds to a greater influence of shear strains. The velocity fields related to

56
4.3. Finite element limit analysis

0. 8

0. 6

χ 0. 4

0. 2

1 10 100

Figure 4.5: the values assumed by the bending coefficient χ for different slender-
nesses

the plastic load of the plates considered are reported in figures 4.6 and 4.7; the
likeness with the simple collapse mechanisms reported in figure 4.2 is manifest:
in particular figure 4.6 shows how the energy is dissipated in a major way by the
elements over the diagonal, while in the punch case (figure 4.7) the whole dissi-
pation is concentrated on the elements lying on the boundaries. The examples
just reported show that the TRIC element is suitable both for limit analysis of
thin and medium thick structures subject to bending and shear strains. During
the analyses, all the element showed in plane rigidity and the dissipation was due
to the bending natural modes only. As explained in appendix C, the axial modes
were set to zero, and the relevant constraints were taken in account during the
solution process. This avoid the infinity that occurs in the computation of the
matrix K when only bending modes are present.
The following examples will demonstrate the good behaviour of the element when
also in plane strains are involved.

57
Chapter 4. Test example: simply supported plate

(a) dissipation density (b) collapse mode

Gi

Figure 4.6: simply supported plate with slenderness β = 100

(a) dissipation density (b) collapse mode

Figure 4.7: simply supported plate with slenderness β = 1

58
4.4. Post collapse behaviour

4.4 Post collapse behaviour


The use of the sequential limit analysis permits to study the post collapse be-
haviour of a shell structure assumed to be rigid-plastic. In this section some
results concerning the collapse evolution of a plate subject to a transverse pres-
sure will be presented, either with or without in plane constraints. In order to
verify the accuracy of the approach used, some comparisons with elasto-plastic
analyses performed by means of the code ABAQUS, will be shown.

4.4.1 Simply supported plate


The first example refers to a simply supported plate with slenderness β = 100
and thickness h = 0.2 mm (see figure 4.1). Its geometric and mechanical char-
acteristics are reported in table 4.4 The analyses have been executed on three

a h E σ0
β ν
(mm) (mm) (GPa) (MPa)
10 0.2 100 200 200 0.3

Table 4.4: geometric and mechanical characteristics of the considered plate

different models whose most important characteristics are reported in table 4.5.
The meshes M1 and M2 are pictured in figure 4.8. Although the post-collapse
behaviour is expected to be stable, the modified Riks algorithm, implemented in
the code ABAQUS, has been used for the incremental analyses. The automatic
step increment, used by default, allows to deal with any nonlinearity. On the con-
trary the sequential limit analysis requires to specify a maximum displacement
between the steps: it has been imposed to be umax = 0.3 mm, an acceptable
value if compared with the plate thickness. The parameters used to predict the
collapse behaviour of the plate under consideration are summarized in table 4.6.

Element Degrees of
Model Mesh Nodes Elements
type freedom
A M1 TRIC 64 98 295
B M1 S3R 64 98 384
C M2 S9R5 289 64 1734

Table 4.5: characteristics of the models used in the analyses

59
Chapter 4. Test example: simply supported plate

G iD

(a) mesh M1 (b) mesh M2

Figure 4.8: meshes used for the post collapse analysis of simply supported plates
with slenderness β = 100

The A and B model are suitable for a correct comparison of the results because of

initial arc maximum


model umax kd ke
length arc length
A 0.3 mm 0.001 0.001 — —
B-C — — — 1 1.5

Table 4.6: parameters used in the sequential limit analysis

the same mesh used. Actually the CPU time required by the method proposed in
this work is shorter than that needed by the incremental analyses (see table 4.7).
This is mainly due to the major stability of the present approach that, in order to
achieve the same solution, requires fewer step than those necessary to the incre-
mental analisys. A significant difference between the post collapse behaviour is
clear observing the curves reported in figure 4.9(a), moreover it has to be pointed
out that, the second order effect being stiffening, this example does not involve
particular difficulties to the incremental analyses and the accuracy depends on
the element used. Therefore a second incremental analysis has been performed
adopting the C model where a more accurate element and a different mesh have
been used. The results obtained are reported in figure 4.9(b) and compared with

60
4.4. Post collapse behaviour

16 0
 16 0




Collapse multiplier

Collapse multiplier
120 120

8 0 8 0

40 40

β   β  
0 0

0 1 2 3 4 5 0 1 2 3 4 5

Displacement (mm) Displacement (mm)

(a) model A and B (b) model A and C

Figure 4.9: comparison between the results obtained by the sequential limit anal-
ysis and the incremental elasto-plastic approach

the load-displacement curve furnished by the sequential limit analysis: in this


case there are only minor discrepancies due to the elastic part of the solution not
considered by the limit analysis.
The deformed configurations corresponding to a maximum displacement of 4.2
mm are reported in figure 4.10(a) for the sequential limit analysis, and in figure
4.10(b) for the elasto-plastic incremental analysis. The two collapse modes are
very similar and exhibit a major energy dissipation near the corner. It seems
that the TRIC element has a good accuracy while the classic triangular shell
element performs in a more stiffer way. Comparable results between the sequen-

Number of
Model Method CPU time
iterations
A S.L.A. 15 23”
B Incremental 47 29”
C Incremental 58 1’24”

Table 4.7: comparison of solution times for a displacement of 4 mm. The analyses
have been performed on a PC-Athlon 1200 MHz

61
Chapter 4. Test example: simply supported plate

(a) Sequential limit analysis (b) Incremental elasto-plastic analysis

Figure 4.10: deformed state at a maximum displacement of 4.2mm

tial limit analysis and the elasto-plastic incremental approach can be achieved by
means of a more accurated element like the S9R5. In this case the solution time
grows but the stability of the problem does not involve particular problems to the
convergence of the incremental analysis.

4.4.2 Simply supported plate with in plane renstraints


In practical situations, plate elements are often not simply supported, but in plane
constraints whose effect is to increase the stiffness of the structure when subject
to large displacements, can exist at the boundaries. In order to evaluate this effect
some analyses have been performed on the same plate with slenderness β = 100
just presented introducing in plane constraints at the boundaries. The sequential
limit analysis and the incremental approach have been applied to models A and
C respectively.
As for the case of simply supported plate, the present approach well describes the
post-collapse behaviour of such structures, even though the load-displacement
curve is not so smooth (see figure 4.11(a)). The deformed state obtained using
the limit analysis for a maximum displacement of 2.5 mm is pictured in figure
4.11(b).
The stiffness increment is clearly shown in figure 4.4.2 where the curves ob-
tained both with and without in plane constraints are reported.

62
4.4. Post collapse behaviour

16 0



Collapse multiplier

120

8 0

4 0

β  
0

0 0.5 1 1.5 2 2.5

Displacement (mm)
(b) Deformed state

(a) Pressure-displacement comparison

Figure 4.11: Results obtained for a simply supported plate with in plane restraints.
Slenderness β = 100

16 0 
Collapse multiplier

120   


    
8 0

40

β
 
0

0 1 2 3 4 5

Displacement (mm)

Figure 4.12: Comparison between simply supported plates with or without in


plane constraints.

63
Chapter 4. Test example: simply supported plate

64
Chapter 5

Test examples: some shells

In chapter 4 some examples related to the post-collapse behaviour of simply sup-


ported plates have been presented. It has been shown that the TRIC element,
modified in a suitable way, performs better than classical triangular element and
it can deal with medium thick shells too. Even though plate structures are very
common in civil, naval and other relevant engineering fields, some applications
require the use of parts typologically different, capable of developing relevant
strains at constant or decreasing applied force. Example of such parts can be
found in crash problems, where the energy absorption capabilities of structural
components are of prominent interest for the assessment of safety in a wide range
of applications, such as design of vehicles and, in general, components that have
to withstand collisions. The research in this field has experienced a considerable
growth in recent years, made possible by the ever increasing computer power and
stimulated by the introduction of codes more and more concerned with safety in
every aspect. In several industrial applications, the design involves the definition
of suitable zones at planned deformation, able to absorb significant impact [75].
Such zones are usually rather localized and deform by developing large strains, so
that the consequent deceleration is acceptable. As an example, such a solution
has been recently adopted in the design of rail vehicles to enhance safety [76].
Thick or medium thick shells are suitable as energy absorbers, since they buckle
in the plastic range, permitting considerable dissipation and significant displace-
ments at decreasing force. Axisymmetric shells, like cylinders or cones, are often
employed to this purpose because of high effectiveness and low cost. A first, yet
meaningful indication on their crashworthiness is provided by static analyses as-
sessing the behavior of the structure after collapse and measuring the energy that
it can dissipate [72, 74].
Different methods may be employed to predict the post-collapse curve. The most
complete simulations, although computationally demanding, are provided by large
displacement, incremental elastic plastic analyses. This approach furnishes a lot
of information sometime unnecessary to the designer and some cases can be solved

65
Chapter 5. Test examples: some shells

only using advanced solver not always implemented. Even simple structures like
cylinder parts require the use of these solver, being their post-collapse behaviour
unstable. Simpler and faster alternatives, such as mechanism analysis, which fol-
lows the post-collapse evolution of elementary mechanisms consisting of circum-
ferential hinges where bending dissipation is concentrated, while the regions in
between experience membrane flow only, are also available [3,43,69,89]. However,
such procedures often consider only axisymmetric collapse modes and axisymme-
tric geometric imperfections. Furthermore, the change in shape of the mechanism
during the post-collapse evolution can hardly be followed to great accuracy.
A good balance between computational efficiency and accuracy of results is pro-
vided by sequential limit analysis adopted in the formulation of the present ap-
proach. In the last decade this method has received an incresing interest and
has been applied to different problems where plastic deformations are predomi-
nant [46, 62, 63, 110]. Its effectiveness and stability have been proved in many
situations, even where incremental methods exhibit difficulties to follow the post-
collapse path.
The advantages of the approach presented in this work and its ability to describe
also unstable post-collapse behaviour will be shown studying the collapse of three
different shells often used as energy absorber and here listed:

• a thick cylindrical shell subject either to axial load or external pressure;

• a medium thick frusta subject both to axial load and external pressure;

• a square cross section tube subject to axial load.

5.1 Cylinder S1
Despite their simplicity, cylindrical structures are often used as a part of energy
absorber due to their chashworthiness and low cost, therefore many studies have
been conducted in the past in order to predict the collapse behaviour and the
absorbed energy of such elements. Alexander [3] proposed a simple collapse me-
chanism, which became universally known as concertina mode, that was adopted
and modified by many researchers in order to predict the mean collapse load of
cylindrical and conical structures [43,88–90]. This method can be applied only to
axisymmetric structures that collapse in axisymmetric way and if imperfections
have to be considered their shape can only be axisymmetric.
The concertina mode is not the only collapse mechanism that a cylindrical shell
can exhibit, but for elements axially compressed with R/t > 40 ÷ 50 [120] the
collapse entails the formation of axial and circumferential waves to make a dia-
mond mode also known as the Yoshimura pattern [129]. This more complicated
collapse mechanism has been widely investigated both theoretically [69, 94] and
experimentally [86, 87] and the interest in the method is shown by recent publi-
cations on the topic [52].

66
5.1. Cylinder S1

The limitations of the concertina and diamond mode approaches can be overcome
by means of finite element analysis, which permits to consider any imperfection
shape and to detect any collapse mode. In particular the approach proposed in
this work and based on the sequential limit analysis is able to supply in an efficient
and stable way the post-collapse behaviour of any shell part, proposing itself as
an alternative tool to the study of such structures.
In this section a cylindrical shell labelled S1 subject either to axial compression or
external pressure will be considered. The geometric and mechanics characteristics
of the shell are reported in table 5.1, while a sketch is pictured in figure 5.1. Due
to geometric characteristics the collapse is expected to be axisymmetric.
The values of the squash load P0 and of the elastic buckling load PE for the
axially
p compressed cylinder are also indicated. The dimensionless parameter
λ̄ = P0 /PE gives an indication on the slenderness of the shell and the value
of 0.18 states that it is very stocky. Cylinder S1 was previously studied by

R th H σ0 E P0 PE
λ̄
(mm) (mm) (mm) (MPa) (GPa) (kN) (kN)
19.5 1.0 25.0 200 200 24.5 760 0.18

Table 5.1: geometrical and mechanical characteristics of cylinder S1

th
H

Figure 5.1: the S1 cylinder

Seitzberger and Rammerstofer, who furnished also experimental data that were
used for validation of the proposed method. In [110] the limit analysis problem
was formulated on the basis of Ilyushin yield surface [27, 66, 104] describing the
yield criterion in terms of generalized stresses. With respect to the work above,
the approach proposed exhibits two main differences. First, the definition of the
shell yield surface is avoided and the dissipation power is directly defined by closed

67
Chapter 5. Test examples: some shells

form integration over the thickness, which allows for considering more general ma-
terials, such as those governed by Hill or Tsai-Wu criterion [34,44,119]. Secondly,
the limit analysis procedure proposed in [30], which has the distinctive feature
of detecting and eliminating from the problem the finite elements that are not
involved in the collapse mechanism, is employed with significant computational
saving when plastic flow affects limited zones of the structure only, as typically
occurs for energy absorbers.

5.1.1 Axisymmetric mechanism approach


In order to obtain a first estimate of the collapse behaviour, the approach de-
veloped by Andronicou and Walker [6] has been applied to the S1 cylinder. The
mechanism shape is assumed to be in the form of a concertina having three hinges
with straight segments between them as shown in figure 5.2(a). A rigid-plastic
idealisation of the material behaviour is assumed, in which elastic deformations
and strain hardening are ignored. The work done in bending at three hinges
during a virtual change of deformation is:

dW1 = 4M π (D + h sin θ) dθ (5.1)

where the limit moment M depends on the axial stress:


" µ ¶2 #
σ 0 t2 P
M= 1− (5.2)
4 σ0 t

The material between the hinges stretches longitudinally and the related work is:

dW2 = 2πσ2 th2 cos θdθ (5.3)

where the circumferential stress σ2 is assumed constant over the mechanism. Its
value can be obtained if a yield criterion is assumed, for example for the von Mises
criterion: µ q ¶
2 2 1 2 2
σ1 − σ1 σ2 + σ2 = σ0 ⇒ σ2 = σ1 − 4σ0 − 3σ1 (5.4)
2
The last contribution to the internal work is due to the longitudinal stretch in
the mechanism that, being the whole mechanism plasticized, can be found by
imposing the flow rule:

2σ1 − σ2 h2
dW3 = πDP cos θdθ (5.5)
2σ2 − σ1 R
Finally the external work is:
· ¸
2σ1 − σ2 h
dWe = 2πDP h sin θ + cos θ (5.6)
2σ2 − σ1 2R

68
5.1. Cylinder S1

Equating the external and internal virtual work terms the equilibrium equation
is given by:
2 ¡ ¢ ³ p ´
√ t 1 − P̄ 2 (D + h sin θ) − h2 cos θ P̄ − 4 − 3P̄2 − 2P̄ hD sin θ = 0 (5.7)
3
P
where P̄ = σ0 t Andronicou and Walker considered the length of the mechanism

P
25
   
    
  
  ! 
20
D "   
"  # $  %

Force (kN)
t
15

h 10

θ
5
θ
0

P
0 2 4 6 8 10

Shortening (mm)

(a) axisymmetric mechanism (b) the load shortening curve

Figure 5.2: axisymmetric mechanism approach


variable with the deformation θ and the load P in such a manner as to minimise
the required load for each deformed state. The minimisation of equation (5.7)
with respect to h gives:

3DP̄ + tP̄ 2 − t
h= √ ³√ ´ (5.8)
3 cot θ 4 − 3P̄ 2 − P̄

Substituting equation (5.8) in (5.7) and solving it by an appropriate method


(for example Newton-Raphson) the collapse load P can be found as a function
of the evolution variable θ. The obtained load-shortening curve is reported in
figure 5.2(b) where it is compared with experimental data and axisymmetric finite
element limit analysis. The shortening may be obtained by:
2σ1 − σ2 h2
∆ = 2h (1 − cos θ) + sin θ (5.9)
2σ2 − σ1 R

69
Chapter 5. Test examples: some shells

5.1.2 The present approach


In order to get a deeper knowledge of the collapse evolution the proposed ap-
proach was used to analyze the cylinder S1 subject either to axial load or external
pressure. In this subsection the results obtained will be presented and compared
with the elastic-plastic counterpart.
Different models (see table 5.2 and figures 5.3) were used: the first two models
labelled A and B are of axisymmetric type and they are described by elements
AX2P and SAX1 for limit and incremental analysis respectively; models C, D
and E describe a quarter of the cylinder, while models F and G describe the whole
cylinder surface.

Element Degrees of
Model Mesh Nodes Elements Notes
type freedom
A M1 AX2P 51 50 148 Axisym.
B M1 SAX1 51 50 148 Axisym.
C M2 TRIC 159 272 751 Quarter
D M2 S3R 159 272 960 Quarter
E M3 S9R5 525 120 3156 Quarter
F M4 TRIC 220 400 1081 Whole cyl.
G M5 S9R5 841 200 5046 Whole cyl.

Table 5.2: characteristics of the models used in the analyses

Axial load
The post-collapse behaviour of a cylindrical shell subject to axial compression
load is expected to be unstable because of the second order geometric effects. In
these situations a step to step approach can deal with convergency problems and
the time needed to obtain the required solution can be greater than using direct
methods. Some recent studies showed that direct methods can be more efficient
than traditional ones [62] in the post-collapse analysis of structures that exhibit
an unstable behaviour.
For the case under consideration experimental results are available and they are
used as validation of the procedure adopted. The post-collapse curves obtained
by sequential limit analysis are reported in figure 5.4(a) and compared with the

70
5.1. Cylinder S1

25 node 51

20 node 41
Height (mm)

15 node 31

10 node 21

5 node 11

node 1
0

18 19

Radius (mm)

(a) mesh M1 (b) mesh M2 (c) mesh M3

(d) mesh M4 (e) mesh M5

Figure 5.3: the different meshes used for the analysis of S1 cylinder

experimental results described in [110] and referring to a considerably longer spe-


cimen. Because of this fact, comparatively large displacements are present prior
the collapse, but the process zone is very nearly the same and, as far as the
post-collapse response is concerned, comparison is meaningful. In figure 5.4(a)
the post-critical experimental curve was shifted toward left, so as to depart from
the collapse load. Results of incremental analyses are reported in figure 5.4(b):
comparison shows that present results are in good agreement with all alternative
solutions, while the element S3R seems to be too stiffer as pointed out in sec-

71
Chapter 5. Test examples: some shells

Pmax Number of
Model Method CPU Time
(MPa) iterations
Exper. 24.50 — —
A S.L.A 24.67 29 15”
B Incremental 24.50 1098 4’02”
C S.L.A 24.75 18 54”
D Incremental 24.50 1529 23’12”
E Incremental 24.45 1475 25’16”
F S.L.A 24.85 18 1’31”
G Incremental 24.43 1477 40’09”

Table 5.3: results obtained for the S1 cylinder subject to axial load. The CPU
times refer to a shortening of 8 mm. The analyses have been performed on a
PC-Athlon 1200 MHz

tion 4.4.
As experimentally obtained, the collapse of the cylinder shell is axisymmetric
and the evolution of the mechanism up to contact is shown in figure 5.5(a) while
in figure 5.5(b) is reported a deformed configuration corresponding to an end-
shortening of 4.2 mm.
For such analysis where axisymmetric collapse occurs the AX2P element give the
best result in shorter time (see table 5.3).

72
5.1. Cylinder S1

25
   
      25   
   
20   
  !    
"   20
  

"     

Force (kN)

15 "  #

Force (kN)
1 5
  

  
10
1 0

5
5

0
0

0 2 4 6 8 10 0 2 4 6 8 1 0

Shortening (mm) Shortening (mm)

(a) comparison with experimental data (b) comparison with incremental analyses

Figure 5.4: load-shortening curves for the S1 cylinder subject to axial load

2 5

S1

2 0


 

15
 

 



 
10
 

2 7

46
4 4
0 335
1 .9 6
18 2 2 2 6 32. 9
          

i D3
(a) Mesh M1 (b) Mesh M2

Figure 5.5: evolution of the collapse obtained by sequential limit analysis for the
S1 cylinder subject to axial load

73
Chapter 5. Test examples: some shells

Pressure load
Although no experimental tests were available for comparison, the investigation
was extended to the case of external pressure. The resulting collapse modes show
a particular aspect: while the axisymmetric analysis imposes an axisymmetric
collapse mechanism, the general shell analysis permits to recognize more complex
collapses. Indeed, this happens both for limit and incremental analyses where,
after a beginning axisymmetric collapse, subsequent deformations concentrate in
few narrow zones of the shell (figure 5.6) and the radial displacement becomes
meaningless. For this reason the curves reported in figure 5.7(a) refer to the
axisymmetric meshes only, while the other analyses are compared in terms of the
top displacements as reported in figure 5.7(b). This can be done because the
nodes on the top edge are constrained to have the same vertical displacement.

Number of
Model Method Plim (kN) CPU Time
iterations
A S.L.A 10.86 8 4”
B Incremental — 83 15”
C S.L.A 11.43 8 13”
D Incremental — 190 1’49”
E Incremental — 130 1’55”
F S.L.A 11.52 19 56”
G Incremental — 130 3’30”

Table 5.4: results obtained for the S1 cylinder subject to external pressure. The
CPU times refer to a top displacement of 0.2 mm. The analyses have been per-
formed on a PC-Athlon 1200 MHz

The axisymmetric curve reported in figure 5.7(a) is in good qualitative agreement


with that produced by incremental analysis, but transverse displacements are
somewhat larger. This may be caused by the fact that the mechanism involves
the entire cylinder (see figure 5.8) and, the material being assumed as rigid-plastic,
only the membrane stresses in the deformed configuration contrast the increasing
pressure. In this situation the role played by the elastic stiffness, even if not
dramatic, appears of engineering significance.
The curves reported in figure 5.7(b) are due to an initial axisymmetric behaviour,
even in the models from C to G where shell elements have been used: indeed, the
curves are similar and the vertical shift is merely due to the coarser mesh used in
the F model. After this initial behaviour something happens and the deformations

74
5.1. Cylinder S1

d is s -d e n s ity

0 .2 6 3 0 2
0 .2 3 5 1
0 .2 31
17946

Figure 5.6: deformed mesh obtained by the model C analysis

1 6
1 6    

  
1 2
1 4  !#"
Pressure (MPa)

Pressure (MPa)

%$&#'( $

8 1 2
  
 
  !
"$#
 %'&
4
"$#
 %'( 1 0

0 8

0 2 4 6 8 0 0.2 0.4 0.6 0.8

Transversal displacement (mm) Top edge displacement (mm)

(a) pressure - radial displacement (b) pressure - top base displacement

Figure 5.7: results obtained for the S1 cylinder subject to external pressure

develop in few small zones detectable in figure 5.6: the limit pressure drops and
a local collapse occurs.

75
Chapter 5. Test examples: some shells

3 0
S1

2 5


 2 0

 

 15


 10 !#"%$ 

10 15 2 0
         

Figure 5.8: evolution of the collapse obtained by the A model analysis

76
5.2. Conical shells

5.2 Conical shells


Conical shells are structural elements often used in civil and marine applications.
In the former case they may be parts of tanks and silos while in marine and
offshore structures they appear as parts of underwater housings or as transition
elements between cylindrical parts. Loading conditions are often complex and
may include bending and torsion, but axial compression and external pressure
are considered the main actions to be taken into consideration in determining the
stability and strength of these structures.
Studies on elastic buckling of perfect conical shells were developed both analyti-
cally [79, 108], and experimentally [49, 50], and originated mainly from aeronau-
tical research. Other studies dealt with the effects of pre-buckling nonlinearities
and boundary conditions, or were oriented to producing imperfection sensitivity
curves by means of numerical models [98]. In most current codes dealing with this
problem [48], the critical buckling load of conical shells is reduced to the study of
an equivalent cylinder, while different approaches are proposed for elastic-plastic
cases [54].
When slenderness of conical shells is not high, yield stress is usually achieved
in a section of the shell and the buckling phenomenon becomes influenced by
the plastic effects. The analysis of the plastic buckling process is usually tack-
led by means of finite element models [42], that may become burdensome and
uneconomic in the design phase. Mechanism models such as that presented in
subsection 5.1.1 have been developed and validated [43, 96] for conical shells too,
but with the same limitations of the cylindrical counterpart.
Being based on a finite element discretization the method proposed in this work
overcome the mechanism models’ limitations proposing itself as an efficient tool
for the prediction of the collapse load and the post-collapse behaviour of such
structures. The capabilities of the sequential limit analysis are investigated by
simulating an experimental test made by Ross et al. whose particulars can be
found in [105]. Some medium thick conical shells labelled TICC4, TICC5 and
TICC6 were subject to external pressure and to axial load due to the pressure
over the small cap. In all the tested specimen the collapse mode was asymmetric
with the formation of some circumferential waves and subsequent rupture near the
bottom edge. Circumferential deformations were recorded during the experiments
by means of ten strain gauges applied to the bottom internal part of the shell.
Unfortunately shortening data are not available and the comparison between the
analyses and experimental test will be done in terms of the maximum achieved
load and the corrispondent collapse mode.

5.2.1 The TICC5 cone


The cone labelled TICC5 is here analyzed. Its geometric properties are reported
in table 5.5 while a sketch of the shell can be found in figure 5.9. Due to the asym-

77
Chapter 5. Test examples: some shells

metric collapse experimentally obtained by Ross et al. the mechanism model is


not applied to the case under consideration, therefore the sequential limit anal-
ysis and the incremental elasto-plastic approach will be presented using general
shell elements. The mechanical properties adopted in the analyses are indi-

Ri Rs th H
α
(mm) (mm) (mm) (mm)
52.75 14.75 3.5 100 20.8◦

Table 5.5: geometric characteristics of cone T5

t h
H

R

Figure 5.9: the TICC5 cone

cated in table 5.6. In the same table are also reported the squash load and the
elastic
p buckling load due to the axial compression. The dimensionless parameter
λ̄ = P0 /PE gives an indication on the slenderness of the cone and the value of
0.17 states that it is very stocky. Specimen TICC5 was machined from a solid
billet of aluminium alloy in a very precisely way so that the maximum initial
imperfection was of 0.0096 mm. Thus, imperfections were considered negligible.
Furthermore, in order to avoid secondary bending stresses, during the test the
model was hung vertically from its bigger base while the pressure was applied on
the lateral surface and on the small cap. Therefore the cone was subject both
to external pressure and axial load. The strain gauges applied to the internal
surface recorded a linear behaviour until a pressure of about 220 bar, when some
circumferential waves formed, while the rupture occurred for an external pressure

78
5.2. Conical shells

σ0 E P0 PE
ν λ̄
(MPa) (GPa) (kN) (kN)
250 61.9 0.3 75.8 2516 0.17

Table 5.6: mechanical characteristics of cone T5

of 275.9 bar. Obviously this value was influenced by the hardening of the ma-
terial not considered in the numerical analyses. It is important to observe that
a descending solution path can not be followed by the experimental equipment
and only the maximum load can be used as comparison. However, from the engi-
neering point of view, the descending path represents an important information
in order to understand the dangerousness of every structure.
From the analytical point of view the post-collapse behaviour of cone TICC5 was
studied both by the sequential limit analysis and by elastic-plastic incremental
approach. Two models labelled H and I have been used and their characteristics
can be found in table 5.7. In figure 5.10 the meshes M6 and M7 used for the dis-
cretization are reported. The post-collapse behaviour of the cone was obtained

Element Degree of
Model Mesh Nodes Elements Notes
type freedom
H M6 TRIC 390 720 1981 Whole cone
I M7 S9R5 882 210 5298 Whole cone

Table 5.7: characteristics of the models used in the analyses

by the sequential limit analysis in 30 step with a maximum displacement of 2.0


mm. On the contrary the elastic-plastic incremental analysis was performed using
the modified Riks method implemented in the commercial code ABAQUS with
automatic evaluation of the step amplitude. This procedure is strongly recom-
mended when the analysis is expected to be highly non-linear as in this case. The
load-shortening curve reported in figure 5.11(a) is not meaningful because, for the
particular load case under consideration, the area under the curve does not rep-
resent the dissipated energy. Moreover the top displacements are very smalls and
the elastic strains can become the relevant part of the whole shortening. Indeed
the elastic effects, present prior the collapse, can be recognized from the dashed
curve pictured in figure 5.11(a) where it is manifest that their amplitude is not
negligible. Therefore a comparison in terms of the dissipated energy, which is an
always growing quantity, is also reported in figure 5.11(b). Although the curves
are not coincident their shape is similar: both exhibit a drop in the pressure

79
Chapter 5. Test examples: some shells

(a) mesh M6 (b) mesh M7

Figure 5.10: the different meshes used for the analysis of the TICC5 cone

followed by an horizontal path where the shell proceeds to dissipate at constant


pressure. The horizontal gap between the two curves is probably due to the elastic
displacements that influence the collapse mechanisms detected by the methods.
As reported in figures 5.12 and 5.13, the post-collapse behaviour, foreseen by the
numerical analyses, is rather different: the incremental analysis identify three
waves (see figure 5.13) as experimentally obtained [105], while the limit method
proposed in this work enhance the deformations present in one wave only (figure
5.12). The maximum loads obtained are reported in table 5.8 and compared with
the experimental collapse value. The latter is sensibly higher than the values
numerically obtained and this is probably due to the hardening in the material.
Being impossible to compare the two analyses for the same top displacement, the
CPU solution times, reported in table 5.8, refer to a dissipated energy of 2 MJ.
In spite of the fact that the same dissipation can be obtained by the sequential
limit analysis in fewer step than those required by the incremental solution, the
needed CPU time is five times higher. This is in clearly disagreement with the
results obtained for the plate and the cylindrical shell where the sequential limit
analysis has always been faster than the incremental one.

80
5.2. Conical shells

3 0
3 0

Experimental
25
rupture 25
Pressure (MPa)

Pressure (MPa)
20
20

1 5
1 5

1 0 
 1 0 

5
   
5
     
0 0

0 0.05 0.1 0.1 5 0.2 0.25 0 1 000000 2000000

Shortening (mm) Dissipation (J)

(a) load-shortening curve (b) load-dissipation curve

Figure 5.11: shortening and dissipation of the cone TICC5. The dots refer to the
deformated states reported in figures 5.12 and 5.13

Number of
Model Method Pmax (MPa) CPU Time
iterations
Exper. 27.59 — —
H S.L.A 24.00 30 12’05”
I Incremental 22.00 59 2’40”

Table 5.8: results obtained for the TICC5 cone subject to axial load and external
pressure. The CPU times refer to a dissipation of 2 MJ. The analyses have been
performed on a PC-Athlon 1200 MHz

81
Chapter 5. Test examples: some shells

3 c

17 s
. 12

5
3
4
5
1. 3

(a) dissipation = 703.6 kJ (b) dissipation = 1286.3 kJ

s G 6 5

1
2 | . 2
. 2
|
2 .i 4 1 n
5 5 68
66 D

0 .
8

(c) dissipation = 1605.2 kJ (d) dissipation = 1874.5 kJ

Figure 5.12: some steps of the collapse foreseen by the sequential limit analysis

82
5.2. Conical shells

Viewport: 1 ODB: d:/nicola/Dottorato/tesi/...82-210.abq/t5-882-210.odb


Viewport: 1 ODB: d:/nicola/Dottorato/tesi/...82-210.abq/t5-882-210.odb

PEEQ
fraction = -1.0) SNEG, (fraction = -1.0)
rit.: 75%) (Ave. Crit.: 75%)
01e-02 +1.703e-01
20e-02 +1.559e-01
39e-02 +1.416e-01
59e-02 +1.273e-01
78e-02 +1.129e-01
97e-02 +9.861e-02
17e-02 +8.428e-02
60e-03 +6.995e-02
54e-03 +5.562e-02
47e-03 +4.129e-02
40e-03 +2.696e-02
33e-03 +1.263e-02
34e-04 -1.705e-03

COLLAPSE ANALYSIS OF A CONE 3 COLLAPSE ANALYSIS OF A CONE


ODB: t5-882-210.odb ABAQUS/Standard 6.1-12 Tue Nov ODB: 27 t5-882-210.odb
14:19:20 ora solare ABAQUS/Standard
Europa occidentale
6.1-1 2001Tue Nov 27 14:19:20 ora solare Europa occidentale 2001
Viewport: 1 ODB: d:/nicola/Dottorato/tesi/...82-210.abq/t5-882-210.odb
Viewport: 1 ODB: d:/nicola/Dottorato/tesi/...82-210.abq/t5-882-210.odb
1 Step: Step 1 Increment
(a) dissipation
13: Step Time = 1 =
100.4
74.8
Step: Step
kJ 1 Increment
(b) dissipation
20: Step Time =
= 235.5
177.0
kJ
Primary Var: PEEQ PEEQ Primary Var: PEEQ
Deformed Var: U Deformation Scale Factor: +1.000e+00Deformed Var: U Deformation Scale Factor: +1.000e+00
fraction = -1.0) SNEG, (fraction = -1.0)
rit.: 75%) (Ave. Crit.: 75%)
55e-01 +8.829e-01
42e-01 +8.046e-01
29e-01 +7.264e-01
16e-01 +6.481e-01
03e-01 +5.699e-01
90e-01 +4.917e-01
77e-01 +4.134e-01
64e-01 +3.352e-01
51e-01 +2.569e-01
38e-01 +1.787e-01
48e-02 +1.004e-01
23e-03 +2.220e-02
12e-02 -5.605e-02

COLLAPSE ANALYSIS OF A CONE 3 COLLAPSE ANALYSIS OF A CONE


ODB: t5-882-210.odb ABAQUS/Standard 6.1-12 Tue Nov
ODB:
27 t5-882-210.odb
14:19:20 ora solare
ABAQUS/Standard
Europa occidentale
6.1-1 2001Tue Nov 27 14:19:20 ora solare Europa occidentale 2001

1 Step: Step 1 Increment


(c) dissipation =Step:
37: Step Time = 1 612.0
770.6Step
kJ 1 Increment
(d) dissipation = 1267.0
50: Step Time = 1253.
kJ
Primary Var: PEEQ Primary Var: PEEQ
Deformed Var: U Deformation Scale Factor: +1.000e+00Deformed Var: U Deformation Scale Factor: +1.000e+00

Figure 5.13: some steps of the collapse foreseen by the elasto-plastic incremental
analysis

83
Chapter 5. Test examples: some shells

5.3 Square tube

In previous sections the method proposed has been applied to the study of the
post-collapse behaviour of cylindrical shells and frusta. An accurate investigation
of Wirsching and Slater [124] permitted to underline that circular tubes provide
one of the best energy-absorption capability. Nevertheless, in many application
such as cars, railways coaches and ships, tubular frames are rarely used and
occasional impacts are absorbed by sheet metal or plastic structures. In order
to understand the post-collapse behaviour of such structures many experiments
were conducted over square tubes [2,69,113,121–123] and small model coaches [82].
Dynamic effects were also investigated [56,71] and the crumpling load was pointed
out to be generally higher than the corrisponding static value.
Referring to the static analyses some advances in the mechanics of crumpling
of square tubes have been made, working out some theoretical models which
include the plastic work done by travelling plastic hinges. The latter concept
has been introduced [73, 97, 125] for experimental indication of the rolling radius
and [69, 123] for theoretical predictions, but has also been applied with success to
the analysis of cylindrical shells subject to axial load. Obviously these models are
not as general as possible and they can be applied only in particular situations:
for example initial imperfections or not classic boundary conditions are difficult
to deal with. Recent advances in numerical methods applied to limit analysis
suggest to use the finite element method for these problem too.
Huh et al. [62] developed a degenerated four-node shell element suitable for limit
analysis and checked their approach by simulating some square tubes. In parti-
cular the capacity of dealing with strain-hardening materials was introduced into
the formulation of the limit problem, simply by tracking the effective plastic strain
calculated from successive iterations. Nevertheless this is not a challenge problem
and this capability could be easily implemented in the present work. The main
difference between this method and the work published by Huh et al. is due to
the element used. While for the four-node element employed in [62] the reduced
integration technique was necessary to avoid locking phenomena and the zero-
energy modes were eliminated by using phisical stabilisation, the TRIC element,
here modified in a suitable way, do not need particular care and the locking
problem is avoided by its natural formulation and by minimizing respect to the
bending parameter χ.
In order to check the ability of the TRIC element to describe also the post-
collapse behaviour of square tubes, an analysis is performed on the same specimen
studied by Huh et al. [62], by means of the approach present in this work. Since
the possibility to consider strain-hardening materials was not implemented the
results are expected to be lower than that obtained in the original work.

84
5.3. Square tube

5.3.1 Axial load


The considered specimen is a square tube whose dimensions are indicated in
table 5.9. The symbols used are evident from figure 5.14(a) where a sketch of the
tube is reported. An indication of the specimen slenderness may be useful to

a h t
(mm) (mm) (mm)
50 50 1.4

Table 5.9: geometrics characteristics of the tube considered



t h

a
a

(a) the specimen geometry (b) the initial imperfection

Figure 5.14: specimen geometry and imperfections assumed for the analyses

understand which type of collapse can occur. During the tests the bottom and
top edges were welded to a plate, therefore they can be considered clamped and
the elastic buckling load can be found as the buckling load of four square plates
with two sides simply supported and the others clamped. From [26] the buckling
load is:
π2 D
PE = 4th 2 6.7
h
where the coefficient 6.7, depending on the boundary conditions, can be found in
graphs or tables from the literature [26]. The squash load is calculated neglecting

85
Chapter 5. Test examples: some shells

the boundary conditions effects with the following expression:

P0 = 4thσ0

The mechanical properties, the squash and the buckling loads are reported in table
5.10. A value of 0.45 for the slenderness parameter λ̄ states that the tube is stocky,
therefore the collapse can be studied by means of the limit analysis. For sake of

σ0 E P0 PE
ν λ̄
(MPa) (GPa) (kN) (kN)
270 205 0.3 75.6 381.5 0.45

Table 5.10: mechanical characteristics of the tube considered

comparison two analyses have been performed by the limit method proposed in
this work and by the classic incremental approach. The models adopted, whose
characteristics are reported in table 5.11, have been labelled J and K, being the
first used for the sequential limit analysis while the second has been used for the
elasto-plastic method. The corresponding meshes M8 and M9 are pictured in
figure 5.3.1: in particular the mesh M8 consists of 441 nodes and 800 elements
for a total of 2281 degrees of freedom while the M9 consists of 1089 nodes and
256 elements for a total of 6534 degrees of freedom. Being the tube symmetric
only a quarter has been considered and proper boundary conditions are imposed
at the vertical edges. Moreove, clamped boundary conditions are applied at the
top and bottom edges as imposed by the test equipment. The post-collapse of

Element Degrees of
Model Mesh Nodes Elements Notes
type freedom
J M8 TRIC 441 800 2281 Quarter
K M9 S9R5 1089 256 6534 Quarter

Table 5.11: characteristics of the models used in the analyses

the square tube was obtained by the sequential limit analysis in 14 steps: the
first five with a maximum displacement of 0.3 mm and the following steps with
a maximum displacement of 2 mm. This permits to better follow the collapse in
the beginning, reducing the approximation errors of the subsequent steps. On the
contrary, the modified Riks algorithm has been used in the incremental analysis
where the step amplitude was automatically setted by the solver. As reported
in table 5.12 the incremental simulation need much more steps than the method

86
5.3. Square tube

(a) M8 mesh i
(b) M9 mesh
D

Figure 5.15: the mesh used in the analyses

proposed to achieve a vertical displacement of 12 mm after which the contact


occurs. This leads to a notable difference in solution times as can be seen from
the values reported. The load-shortening curves just obtained are reported in
figure 5.16(a) and compared with experimental data obtained by Huh et al. For
sake of comparison the experimental curve was traslated towards left in order
to do not consider the elastic effects. The limit load and the related collapse
mode detected by the present approach agree well with experimental results and
incremental analysis. On the contrary the post collapse behaviour is influenced
by strain hardening effects which are not considered in the present work: this is
the reason because experimental results are higher than the values here obtained.
Anyway the sequential limit analysis is a stable and very efficient tool to predict
the behaviour of structures that undergo plastic flow during the collapse. In
particular this effectiveness is much more manifest when dealing with decreasing
collapse evolution.
The collapse mode analytically obtained, whose evolution is reported in figure
5.17, is similar with the experimental deformation reported in figure 5.16.

87
Chapter 5. Test examples: some shells

Pmax Number of
Model Method CPU Time
(MPa) iterations
Exper. 80.0 — —
J S.L.A 76.57 14 7’56”
K Incremental 24.50 970 37’30”

Table 5.12: results obtained for the square tube subject to axial load. The CPU
times refer to a shortening of 12 mm. The analyses have been performed on a
PC-Athlon 1200 MHz

100

 

80     


 
 !
Force (kN)

60

40

20

0 2 4 6 8 10 12

Shortening (mm)

(a) load-shortening curves (b) during the experiment [62]

Figure 5.16: square tube subject to axial load

88
5.3. Square tube

s ity

15 88
3
0 i
5
1 9 4 77
1 3 38 3
5 1 .8 7 3 26
. . 91
.9 5 6 1

(a) shortening = 0.6 mm (b) shortening = 1.2 mm

d is s -d e n i d is s - n s it

00 00
2 6 7 .0 4 67.
34 9 6 11 2
2 2 0 1.
1
3
20 9
2 1
3 6 .3 8 i 3
3 9 3 1

(c) shortening = 3.1 mm (d) shortening = 9.7 mm


.

Figure 5.17: some steps of the collapse

89
Chapter 5. Test examples: some shells

90
Chapter 6

Conclusions

6.1 A critical survey


A method to follow the post-collapse behaviour of shells that buckle in the plastic
range has been proposed. The method is set on the theoretycal basis of limit
analysis. In particular the kinematic approach is formulated, in conjunction with
finite element modelling, so as to reduce the problem to the search of the essen-
tially free minimum of a convex function. Such function is not differentiable where
plastic strain rates vanish, namely in the rigid parts of the body, therefore, in the
solution procedure the elements detected as rigid are eliminated and appropriate
constraints equations are set. Two shell elements, formulated accordingly to the
natural approach, have been proposed:
• an axisymmetric one, based on Kirchhoff’s hypotheses, suitable for the ana-
lysis of thin shell structures;
• a triangular shell element whose natural modes depends on nodal displace-
ments and on bending parameter χe . This element is suitable both for the
analysis of thin and medium thick shells.
Regarding the triangular shell element, the solution procedure has been modified
to account for the effects of the bending parameter, thus avoiding the locking
phenomenon typical of Mindlin’s elements.
The method proposed has been applied to study the post-collapse behaviour of
some shell structures; the results obtained permit to draw the following conclu-
sions:
• the method is much more stable than the incremental approach and bigger
steps can be used in order to obtain the desidered solution;

91
Chapter 6. Conclusions

• the analyses performed with the approach proposed are generally faster than
the incremental counterparts;
• the triangular element proposed is more accurate than classic triangular
element and it performs well with thick shells too.
In conclusion, the method proposed yields some important benefits to the
collapse analysis, proposing itself as an efficient tool for the prediction of the
collapse load and the post-collapse behaviour of structures that buckle in the
plastic range.

6.2 Future developments


Some work has to be done in order to increase the capabilities of the method. First
of all, the procedure implemented to set as rigid the elements that do not exhibit
plastic strains has to be modified. Actually the effectiveness of the whole method
can be negatively affected when the related constraints are applied, furthermore,
improvement in this topic could be exploited by a future contact algorithm.
In order to deal with crash problems, the method has to be modified to account
for inertial forces rising from high acceleration, as well as a contact algorithm has
to be implemented. Although only consistency conditions have to be satisfied,
this is always a challenge problem, especially when dealing with contact and self-
contact situations.
Another interesting characteristic we are dealing with, is the ability to consider
two different sets of load. A set composed of constant load, and a second set
where loads, instead, are affected by a load multiplier.

92
Appendix A

Strains from drilling modes

As seen in section 3.5.4 the displacement vector of a point P on a generic element


subject to a drilling mode qdα is:

vρα = ζα2 qdα (−(y10 (ζα − 1) + ζβ y20 + ζγ y30 )i) + (x01 (ζα − 1) + ζβ x02 + ζγ x03 )j) (A.1)

Natural strains due to the drilling modes are calculated from the displacements
in P . They are found to be:

X ∂vρjki
²mi = i, j = α, β, γ (A.2)
j
∂Yi

where vρjki is the displacement in P due to the drilling mode j projected on


the side i.
In the following the calculations necessary to expand the relations A.2 are re-
ported.

vραkα = vρα • eα = ζα2 qdα ( − ((ζα − 1)y10 + ζβ y20 + ζγ y30 )i+


(A.3)
(x01 (ζα − 1) + ζβ x02 + ζγ x03 )j) • (cαx0 isαx0 j)
vρβkα = vρβ • eα = ζβ2 qdβ ( − (ζα y10 + (ζβ − 1)y20 + ζγ y30 )i+
(A.4)
(ζα x01 + (ζβ − 1)x02 + ζγ x03 )j) • (cαx0 isαx0 j)
vργkα = vργ • eα = ζγ2 qdγ ( − (ζα y10 + ζβ y20 + (ζγ − 1)y30 )i+
(A.5)
(ζα x01 + ζβ x02 + (ζγ − 1)x03 )j) • (cαx0 isαx0 j)

If the projections of side j on local axes x0 , y 0 are called x0j , yj0 the equation
A.5 can be written in a more simplified way:

93
Appendix A. Strains from drilling modes

vραkα = ζα2 qdα (−(ζβ yγ0 − ζγ yβ0 )cαx0 + (ζβ x0γ − ζγ x0β )sαx0 ) (A.6)
vρβkα = ζβ2 qdβ ((ζα yγ0 − ζγ yα0 )cαx0 − (ζα x0γ − ζγ x0α )sαx0 ) (A.7)
vργkα = ζγ2 ψδ (−(ζα yβ0 − ζβ yα0 )cαx0 + (ζα x0β − ζβ x0α )sαx0 ) (A.8)

The calculus of the components parallel to sides β and γ can be done changing
the direction cosines. The strains are derived from the definition:

dvρkα 1 dvρkα
²mα = =
dYα lα dηα
µ ¶ (A.9)
1 X ∂vρjkα ∂ζα ∂vρjkα ∂ζβ ∂vρjkα ∂ζγ
= + +
lα j ∂ζα ∂ηα ∂ζβ ∂ηα ∂ζγ ∂ηα
dvρkβ 1 dvρkβ
²mβ = =
dYβ lβ dηβ
X µ ¶ (A.10)
1 ∂v ρjkβ ∂ζα ∂vρjkβ ∂ζβ ∂vρjkβ ∂ζγ
= + +
lβ j ∂ζα ∂ηβ ∂ζβ ∂ηβ ∂ζγ ∂ηβ
dvρkγ 1 dvρkγ
²mγ = =
dYγ lγ dηγ
µ ¶ (A.11)
1 X ∂vρjkγ ∂ζα ∂vρjkγ ∂ζβ ∂vρjkγ ∂ζγ
= + +
lγ j ∂ζα ∂ηγ ∂ζβ ∂ηγ ∂ζγ ∂ηγ

From the derivatives of ζ respect to η (equation 3.61):

µ ¶
1 X ∂vρjkα ∂vρjkα
²mα = − + (A.12a)
lα j ∂ζβ ∂ζγ
µ ¶
1 X ∂vρjkβ ∂vρjkβ
²mβ = − (A.12b)
lβ j ∂ζα ∂ζγ
µ ¶
1 X ∂vρjkγ ∂vρjkγ
²mγ = − + (A.12c)
lγ j ∂ζα ∂ζβ
(A.12d)

Partial derivatives of v respect to the internal coordinate ζ are now calculated:

∂vραkα
= 2ζα qdα ((−ζβ yγ0 + ζγ yβ0 )cαx0 + (ζβ x0γ − ζγ x0β )sαx0 )
∂ζα (A.13)
= 2ζα qdα (ζβ (−yγ0 cαx0 + x0γ sαx0 ) + ζγ (yβ0 cαx0 − x0β sαx0 ))

94
It can be observed that the terms in the internal parentheses are the projec-
tions of sides β and γ on the direction perpendicular to side α, therefore they are
the height hα of the triangle. Introducing the element area Ω the derivatives can
be written in a more simplified way:

∂vραkα 2Ω Ω
= 2ζα qdα (ζβ + ζγ ) = 4 ζα (1 − ζα )qdα
∂ζα lα lα
∂vραkα Ω
= ζα2 qdα (−cαx0 yγ0 + sαx0 x0γ ) = 2 ζα2 qdα
∂ζβ lα
∂vραkα Ω
= ζα2 qdα (cαx0 yβ0 − sαx0 x0β ) = 2 ζα2 qdα
∂ζγ lα
∂vρβkα Ω
= ζβ2 qdβ (cαx0 yγ0 − sαx0 x0γ ) = −2 ζα2 qdα
∂ζα lα
∂vρβkα
= 2ζβ qdβ ((ζα yγ0 − ζγ yα0 )cαx0 + (−ζα x0γ + ζγ x0α )sαx0 )
∂ζβ
= 2ζβ qdβ (ζα (−cαx0 yγ0 + sαx0 x0γ ) − ζγ (cαx0 yα0 − sαx0 x0α ))

= −4 ζα ζβ qdβ

∂vρβkα
= ζβ2 qdβ (−cαx0 yα0 + sαx0 x0α ) = 0
∂ζγ
∂vργkα Ω 2
= ζγ2 qdγ (−cαx0 yβ0 + sαx0 x0β ) = −2 ζ qdγ
∂ζα lα γ
∂vργkα
= ζγ2 qdγ (cαx0 yα0 − sαx0 x0α ) = 0
∂ζβ
∂vργkα
= 2ζγ qdγ ((−ζα yβ0 + ζβ yα0 )cαx0 + (ζα x0β − ζβ x0α )sαx0 )
∂ζγ
= 2ζγ qdγ (−ζα (cαx0 yβ0 − sαx0 x0β ) + ζβ (cαx0 yα0 − sαx0 x0α ))

= −4 ζα ζγ qdγ

∂vραkβ
= 2ζα qdα (ζβ (−cβx0 yγ0 + sβx0 x0γ ) + ζγ (cβx0 yβ0 − sβx0 x0β ))
∂ζα

= −4 ζα ζβ qdα

∂vραkβ Ω 2
= ζα2 qdα (−cβx0 yγ0 + sβx0 x0γ ) = −2 ζ qdα
∂ζβ lβ α
∂vραkβ
= ζα2 qdα (cβx0 yβ0 − sβx0 x0β ) = 0
∂ζγ
∂vρβkβ Ω 2
= ζβ2 qdβ (cβx0 yγ0 − sβx0 x0γ ) = 2 ζ qdβ
∂ζα lβ β

95
Appendix A. Strains from drilling modes

∂vρβkβ
= 2ζβ qdβ (ζα (cβx0 yγ0 − sβx0 x0γ ) − ζγ (cβx0 yα0 − sβx0 x0α ))
∂ζβ

= 4 ζβ (1 − ζβ )qdβ

∂vρβkβ Ω 2
= ζβ2 qdβ (−cβx0 yα0 + sβx0 x0α ) = 2 ζ qdβ
∂ζγ lβ β
∂vργkβ
= ζγ2 qdγ (−cβx0 yβ0 + sβx0 x0β ) = 0
∂ζα
∂vργkβ Ω 2
= ζγ2 qdγ (cβx0 yα0 − sβx0 x0α ) = −2 ζ qdγ
∂ζβ lβ γ
∂vργkβ
= 2ζγ qdγ (−ζα (cβx0 yβ0 − sβx0 x0β ) + ζβ (cβx0 yα0 − sβx0 x0α ))
∂ζγ

= −4 ζβ ζγ qdγ

∂vραkγ
= 2ζα qdα (ζβ (−cγx0 yγ0 + sγx0 x0γ ) + ζγ (cγx0 yβ0 − sγx0 x0β ))
∂ζα

= −4 ζα ζγ qdα

∂vραkγ
= ζα2 qdα (−cγx0 yγ0 + sγx0 x0γ ) = 0
∂ζβ
∂vραkγ Ω 2
= ζα2 qdα (cγx0 yβ0 − sγx0 x0β ) = −2 ζ qdα
∂ζγ lγ α
∂vρβkγ
= ζβ2 qdβ (cγx0 yγ0 − sγx0 x0γ ) = 0
∂ζα
∂vρβkγ Ω
= 2ζβ qdβ (ζα (cγx0 yγ0 − sγx0 x0γ ) − ζγ (cγx0 yα0 − sγx0 x0α )) = −4 ζβ ζγ qdβ
∂ζβ lγ
∂vρβkγ Ω 2
= ζβ2 qdβ (−cγx0 yα0 + sγx0 x0α ) = −2 ζ qdβ
∂ζγ lγ β
∂vργkγ Ω 2
= ζγ2 qdγ (−cγx0 yβ0 + sγx0 x0β ) = 2 ζ qdγ
∂ζα lγ γ
∂vργkγ Ω
= ζγ2 qdγ (cγx0 yα0 − sγx0 x0α ) = 2 ζγ2 qdγ
∂ζβ lγ
∂vργkγ
= 2ζγ qdγ (−ζα (cγx0 yβ0 − sγx0 x0β ) + ζβ (cγx0 yα0 − sγx0 x0α ))
∂ζγ

= 4 ζγ (1 − ζγ )qdγ

From equationa A.12a, A.12b and A.12c the strains due to the drilling modes
are now calculated:

96
µ ¶ 
1 Ω 2 Ω 2
²α
mα = −2 ζα qdα + 2 ζα qdα = 0



lα lα lα 

µ ¶ 

1 Ω Ω
²βmα = 4 ζα ζβ qdβ ²mα = 4 2 ζα (ζβ qdβ − ζγ qdγ ) (A.14)
lα lα 
 lα
µ ¶ 

1 Ω 

²γmα = −4 ζα ζγ qdγ 

lα lα

µ ¶ 
1 Ω 
²α
mβ = −4 ζα ζβ qdα 


lβ lβ 

µ ¶ 

1 Ω 2 Ω 2 Ω
²βmβ = 2 ζβ qdβ + 2 ζβ qdβ = 0 ²mβ = 4 2 ζβ (ζγ qdγ − ζα qdα ) (A.15)
lβ lβ lβ 
 l β
µ ¶ 

1 Ω 

²γmβ = 4 ζβ ζγ qdγ 

lβ lβ

µ ¶ 
1 Ω 
²α
mγ = 4 ζα ζγ qdα 


lγ lγ 

µ ¶ 

1 Ω Ω
²βmγ = −4 ζβ ζγ qdβ ²mγ = 4 2 ζγ (ζα qdα − ζβ qdβ ) (A.16)
lγ lγ 
 lγ
µ ¶ 

1 Ω 2 Ω 2 

²γmγ = 2 ζγ qdγ + 2 ζγ qdγ = 0

lγ lγ lγ

and the equations 3.92, 3.93 and 3.94 are demonstrated.

97
Appendix A. Strains from drilling modes

98
Appendix B

Details on the compilation


of the energy dissipation

As seen in section 3.5.4 the strains in the element can be formulated in terms
of natural modes. The expressions are here reported and in the following some
calculations will be done in order to obtain a more simple and easily implementable
form.

0 qSα 1 £ b b b
¤
²mα =qmα +z − z 2 3lα (ζγ − ζβ )qAα + lβ ζα qAβ − lγ ζα qAγ
lα lα
(B.1a)

+4 ζ (ζ q − ζγ qγ )
2 α β β

0 qSβ 1 £ b b b
¤
²mβ =qmβ +z − z 2 −lα ζβ qAα + 3lβ (ζα − ζγ )qAβ + lγ ζβ qAγ
lβ lβ
(B.1b)

+4 ζβ (ζγ qγ − ζα qα )
lβ2

0 qSγ 1 £ b b b
¤
²mγ =qmγ +z − z 2 lα ζγ qAα − lβ ζγ qAβ + 3lγ (ζβ − ζα )qAγ
lγ lγ
(B.1c)

+4 ζγ (ζα qα − ζβ qβ )
lγ2

s (1 − ζα )
²sα = qAα (B.1d)
2
s (1 − ζβ )
²sβ = qAβ (B.1e)
2
s (1 − ζγ )
²sγ = qAγ (B.1f)
2

99
Appendix B. Details on the compilation of the energy dissipation

where q bA + q sA = q A .
The dissipated energy in each element can be calculated by means of the following
equation:
Z q
De = σ0 ρ̃˙ Tm Πm ρ̃˙ m + ρ̃˙ Ts Πs ρ̃˙ s dx (B.2)
Ωe

where for convenience the transversal shear modes are separated from the
other ones:

¡ ¢T
ρ̃m = q 0m q S q bA qd (B.3)
12×1
¡ ¢
ρ̃s = q sA (B.4)
3×1

The terms of symmetric matrixes Πm and Πs , defined in equation 3.119, are


12×12 3×3
now evaluated

Πm (1, 1) = Γm11
Πm (1, 2) = Γm12
Πm (1, 3) = Γm13
z0
Πm (1, 4) = Γm11

z0
Πm (1, 5) = Γm12

z0
Πm (1, 6) = Γm13

ζβ − ζγ 0 lα ζβ lα ζγ
Πm (1, 7) = 3 z Γm11 + 2 z 0 Γm12 − 2 z 0 Γm13
lα lβ lγ
lβ ζα 0 ζα − ζγ 0 lβ ζγ
Πm (1, 8) = − z Γm11 − 3 z Γm12 + 2 z 0 Γm13
lα2 lβ lγ
lγ ζα 0 lγ ζβ (ζα − ζβ ) 0
Πm (1, 9) = z Γm11 − 2 z 0 Γm12 + 3 z Γm13
lα2 lβ lγ
4Ω 4Ω
Πm (1, 10) = − 2 ζα ζβ Γm12 + 2 ζα ζγ Γm13
lβ lγ
4Ω 4Ω
Πm (1, 11) = 2
ζα ζβ Γm11 − 2 ζβ ζγ Γm13
lα lγ

100
4Ω 4Ω
Πm (1, 12) = − ζ ζ Γ
2 α γ m11
+ 2 ζβ ζγ Γm12
lα lβ

Πm (2, 2) = Γm22
Πm (2, 3) = Γm23
z0
Πm (2, 4) = Γm21

z0
Πm (2, 5) = Γm22

z0
Πm (2, 6) = Γm23

(ζβ − ζγ ) 0 lα ζβ lα ζγ
Πm (2, 7) = 3 z Γm21 + 2 z 0 Γm22 − 2 z 0 Γm23
lα lβ lγ
lβ ζα 0 (ζα − ζγ ) 0 lβ ζγ
Πm (2, 8) = − z Γm21 − 3 z Γm22 + 2 z 0 Γm23
lα2 lβ lγ
lγ ζα 0 lγ ζβ (ζα − ζβ ) 0
Πm (2, 9) = 2
z Γm21 − 2 z 0 Γm22 + 3 z Γm23
lα lβ lγ
4Ω 4Ω
Πm (2, 10) = − 2 ζα ζβ Γm22 + 2 ζα ζγ Γm23
lβ lγ
4Ω 4Ω
Πm (2, 11) = ζα ζβ Γm21 − 2 ζβ ζγ Γm23
lα2 lγ
4Ω 4Ω
Πm (2, 12) = − ζ ζ Γ
2 α γ m21
+ 2 ζβ ζγ Γm22
lα lβ

Πm (3, 3) = Γm33
z0
Πm (3, 4) = Γm31

z0
Πm (3, 5) = Γm32

z0
Πm (3, 6) = Γm33

(ζβ − ζγ ) 0 lα ζβ lα ζγ
Πm (3, 7) = 3 z Γm31 + 2 z 0 Γm32 − 2 z 0 Γm33
lα lβ lγ

101
Appendix B. Details on the compilation of the energy dissipation

lβ ζα 0 (ζα − ζγ ) 0 lβ ζγ
Πm (3, 8) = − z Γm31 − 3 z Γm32 + 2 z 0 Γm33
lα2 lβ lγ
lγ ζα 0 lγ ζβ (ζα − ζβ ) 0
Πm (3, 9) = z Γm31 − 2 z 0 Γm32 + 3 z Γm33
lα2 lβ lγ
4Ω 4Ω
Πm (3, 10) = − 2 ζα ζβ Γm32 + 2 ζα ζγ Γm33
lβ lγ
4Ω 4Ω
Πm (3, 11) = 2
ζα ζβ Γm31 − 2 ζβ ζγ Γm33
lα lγ
4Ω 4Ω
Πm (3, 12) = − 2
ζα ζγ Γm31 + 2 ζβ ζγ Γm32
lα lβ

z 02
Πm (4, 4) = Γm11
lα2
z 02
Πm (4, 5) = Γm12
lα lβ
z 02
Πm (4, 6) = Γm13
lα lγ
(ζβ − ζγ ) 02 ζβ ζγ
Πm (4, 7) = 3 2
z Γm11 + 2 z 02 Γm12 − 2 z 02 Γ13
lα lβ lγ
lβ ζα 02 (ζα − ζγ ) 02 lβ ζγ 02
Πm (4, 8) = − z Γm11 − 3 z Γm12 + z Γm13
lα3 lα lβ lα lγ2
lγ ζα 02 lγ ζβ 02 ζα − ζβ 02
Πm (4, 9) = z Γm11 − z Γm12 + 3 z Γm13
lα3 lα lβ2 lα lγ
4Ω 0 4Ω 0
Πm (4, 10) = − z ζα ζβ Γm12 + z ζα ζγ Γm13
lα lβ2 lα lγ2
4Ω 0 4Ω 0
Πm (4, 11) = 3
z ζα ζβ Γm11 − z ζβ ζγ Γm13
lα lα lγ2
4Ω 0 4Ω 0
Πm (4, 12) = − z ζα ζγ Γm11 + z ζβ ζγ Γm12
3
lα lα lβ2

z 02
Πm (5, 5) = Γm22
lβ2

z 02
Πm (5, 6) = Γm23
lβ lγ

102
(ζβ − ζγ ) 02 lα ζβ lα ζγ 02
Πm (5, 7) = 3 z Γm21 + 3 z 02 Γm22 − z Γm23
lα lβ lβ lβ lγ2
ζα 02 (ζα − ζγ ) 02 ζγ
Πm (5, 8) = − z Γm21 − 3 z Γm22 + 2 z 02 Γm23
2
lα lβ2 lγ
lγ ζα 02 lγ ζβ (ζα − ζβ ) 02
Πm (5, 9) = 2
z Γm21 − 3 z 02 Γm22 + 3 z Γm23
lα lβ lβ lβ lγ
4Ω 0 4Ω
Πm (5, 10) = − z ζα ζβ Γm22 + 2 z 0 ζα ζγ Γm23
lβ3 lβ lγ
4Ω 0 4Ω
Πm (5, 11) = 2l
z ζα ζβ Γm21 − 2 z 0 ζβ ζγ Γm23
lα β lβ lγ

4Ω 0 4Ω
Πm (5, 12) = − z ζα ζγ Γm21 + 3 z 0 ζβ ζγ Γm22
lα2 lβ lβ

z 02
Πm (6, 6) = Γm33
lγ2
(ζβ − ζγ ) 02 lα ζβ lα ζγ
Πm (6, 7) = 3 z Γm31 + 2 z 02 Γm32 − 3 z 02 Γm33
lα lγ lβ lγ lγ
lβ ζα 02 (ζα − ζγ ) 02 lβ ζγ
Πm (6, 8) = − 2
z Γm31 − 3 z Γm32 + 3 z 02 Γm33
lα lγ lβ lγ lγ
ζα 02 ζβ (ζα − ζβ ) 02
Πm (6, 9) = z Γm31 − 2 z 02 Γm32 + 3 z Γm33
lα2 lβ lγ2
4Ω 0 4Ω
Πm (6, 10) = − z ζα ζβ Γm32 + 3 z 0 ζα ζγ Γm33
lβ2 lγ lγ
4Ω 0 4Ω
Πm (6, 11) = 2l
z ζα ζβ Γm31 − 3 z 0 ζβ ζγ Γm33
lα γ lγ
4Ω 0 4Ω
Πm (6, 12) = − z ζα ζγ Γm31 + 2 z 0 ζβ ζγ Γm32
lα2 lγ lβ lγ
2 2
(ζβ − ζγ )2 02 ζβ (ζβ − ζγ ) 02 lα ζβ 02
Πm (7, 7) = 9 2
z Γ m11 + 3 2 z (Γ m21 + Γ m12 ) + 4 z Γm22
lα lβ lβ
ζγ (ζβ − ζγ ) 02 2
lα ζβ ζγ 02 lα2 ζγ2 02
−3 z (Γ m31 + Γ m13 ) − 2 z (Γ m32 + Γ m23 ) + z Γm33
lγ2 lβ lγ2 lγ4

103
Appendix B. Details on the compilation of the energy dissipation

lβ ζα (ζβ − ζγ ) 02 lα ζγ (ζα − ζγ ) 02 lβ ζα ζγ 02
Πm (7, 8) = −3 3
z Γm11 + 3 z Γm32 + z Γm31
lα lβ lγ2 lα lγ2
(ζα − ζγ )(ζβ − ζγ ) 02 lα ζβ ζγ 02 ζα ζβ 02
−9 z Γm12 + z Γm23 − z Γm21
lα lβ lβ lγ2 lα lβ
lβ (ζβ − ζγ )ζγ 02 lα ζβ (ζα − ζγ ) 02 lα lβ ζγ2 02
+3 z Γm13 − 3 z Γm22 − z Γm33
lα lγ2 lβ3 lγ4
lγ ζα (ζβ − ζγ ) 02 lγ ζα ζβ 02 lα (ζα − ζβ )ζγ 02
Πm (7, 9) = 3 z Γm11 + z Γm21 − 3 z Γm33
lα3 lα lβ2 lγ3
lγ ζβ (ζβ − ζγ ) 02 lα lγ ζβ2 02 lα (ζα − ζβ )ζβ 02
−3 2 z Γm12 − z Γm22 + 3 z Γm23
lα lβ lβ4 lβ2 lγ
(ζα − ζβ )(ζβ − ζγ )ζγ 02 lα ζβ ζγ 02 ζα ζγ 02
+9 z Γm13 + 2 z Γm32 − z Γm31
lα lγ lβ lγ lα lγ

12Ωζα ζβ (ζβ − ζγ ) 0 4Ωlα ζα ζβ2 0 4Ωlα ζα ζβ ζγ 0


Πm (7, 10) = − 2 z Γm12 − z Γm22 + z Γm32
lα lβ lβ4 lβ2 lγ2
12Ωζα ζγ (ζβ − ζγ ) 0 4Ωlα ζα ζβ ζγ 0 4Ωlα ζα ζγ2 0
+ 2
z Γm13 + 2 2
z Γ23 − z Γm33
lα lγ lβ lγ lγ4

12Ωζα ζβ (ζβ − ζγ ) 0 4Ωζα ζβ2 0 4Ωζα ζβ ζγ 0


Πm (7, 11) = z Γm11 + z Γm21 − z Γm31
lα3 lα lβ2 lα lγ2
12Ωζβ ζγ (ζβ − ζγ ) 0 4Ωlα ζβ2 ζγ 0 4Ωlα ζβ ζγ2 0
− z Γm13 − z Γ m23 + z Γm33
lα lγ2 lβ2 lγ2 lγ4

12Ωζα ζγ (ζβ − ζγ ) 0 4Ωζα ζβ ζγ 0 4Ωζα ζγ2 0


Πm (7, 12) = − z Γ m11 − z Γm21 + z Γm31
lα3 lα lβ2 lα lγ2
12Ωζβ ζγ (ζβ − ζγ ) 0 4Ωlα ζβ2 ζγ 0 4Ωlα ζβ ζγ2 0
+ 2 z Γm12 + 4 z Γm22 − z Γm32
lα lβ lβ lβ2 lγ2

lβ2 ζα2 02 ζα (ζα − ζγ ) 02


Πm (8, 8) = 4
z Γm11 + 3 2
z (Γm21 + Γm12 )
lα lα
lβ2 ζα ζγ (ζα − ζγ )2 02
− 2 2 z 02 (Γm31 + Γm13 ) + 9 z Γm22
lα lγ lβ2
(ζα − ζγ )ζγ 02 lβ2 ζγ2 0
−3 z (Γ 32 + Γ 23 ) + z Γm33
lγ2 lγ4

104
lβ lγ ζα2 02 lγ ζα (ζα − ζγ ) 02 lβ (ζα − ζβ )ζγ 02
Πm (8, 9) = − z Γm11 − 3 z Γm21 + 3 z Γm33
lα4 2l
lα β lγ3
lγ ζα ζβ 02 lγ ζβ (ζα − ζγ ) 02 lβ ζα (ζα − ζβ ) 02
+ 2
z Γm12 + 3 3 z Γm22 − 3 z Γm13
lα lβ lβ lα2 lγ
ζβ ζγ 02 (ζα − ζβ )(ζα − ζγ ) 02 lβ ζα ζγ 02
− 2 z Γm32 − 9 z Γ23 + 2 z Γm31
lβ lγ lβ lγ lα lγ

ζα2 ζβ 0 ζα ζβ (ζα − ζγ ) 0 ζα ζβ ζγ 0
Πm (8, 10) = 4Ω 2l
z Γm12 + 12Ω 3 z Γm22 − 4Ω z Γm32
lα β l β lβ lγ2
lβ ζα2 ζγ 0 ζα ζγ (ζα − ζγ ) 0 lβ ζα ζγ2
− 4Ω 2 2
z Γm13 − 12Ω 2
z Γm23 + 4Ω 4 z 0 Γm33
lα lγ lβ lγ lγ
lβ ζα2 ζβ 0 ζα ζβ (ζα − ζγ ) 0 lβ ζα ζβ ζγ
Πm (8, 11) = −4Ω 4
z Γm11 − 12Ω z Γm21 + 4Ω 2 2 z 0 Γm31
lα lα2 lβ lα lγ
lβ ζα ζβ ζγ 0 ζβ ζγ (ζα − ζγ ) 0 lβ ζβ ζγ2
+ 4Ω 2 2
z Γm13 + 12Ω 2
z Γm23 − 4Ω 4 z 0 Γm33
lα lγ lβ lγ lγ

lβ ζα2 ζγ 0 ζα ζγ (ζα − ζγ ) 0 lβ ζα ζγ2 0


Πm (8, 12) = 4Ω z Γ m11 + 12Ω z Γm21 − 4Ω z Γm31
lα4 lα2 lβ 2 l2
lα γ

ζα ζβ ζγ 0 ζβ ζγ (ζα − ζγ ) 0 ζβ ζγ2 0
− 4Ω z Γ m12 − 12Ω 3 z Γ m22 + 4Ω z Γm32
lα2 lβ lβ lβ lγ2

lγ2 ζα2 02 lγ2 ζα ζβ 02 ζα (ζα − ζβ ) 02


Πm (9, 9) = z Γ m11 − 2 l2
z (Γm21 + Γm12 ) + 3 z Γm31
lα4 lα β lα2
lγ2 ζβ2 02 ζβ (ζα − ζβ ) 02
+ z Γm22 − 3 z (Γm32 + Γ23 )
lβ4 lβ2
ζα (ζα − ζβ ) 02 (ζα − ζβ )2 02
+3 2
z (Γm31 + Γm13 ) + 9 z Γm33
lα lγ2

lγ ζα2 ζβ 0 lγ ζα ζβ2 0 ζα ζβ (ζα − ζβ ) 0


Πm (9, 10) = −4Ω z Γ m12 + 4Ω 4 z Γ22 − 12Ω z Γm32
2
lα lβ 2 lβ lβ2 lγ
ζα2 ζγ 0 ζα ζβ ζγ 0 ζα ζγ (ζα − ζβ ) 0
+ 4Ω z Γm13 − 4Ω 2 z Γm23 + 12Ω z Γm33
lα2 lγ lβ lγ lγ3

lγ ζα2 ζβ 0 lγ ζα ζβ2 0 ζα ζβ (ζα − ζβ ) 0


Πm (9, 11) = 4Ω 4
z Γ m11 − 4Ω 2 2 z Γm21 + 12Ω z Γm31
lα lα lβ lα2 lγ
ζα ζβ ζγ 0 ζβ2 ζγ 0 ζβ ζγ (ζα − ζβ ) 0
− 4Ω 2
z Γ m13 + 4Ω 2 z Γm23 − 12Ω z Γm33
lα lγ lβ lγ lγ3

105
Appendix B. Details on the compilation of the energy dissipation

lγ ζα2 ζγ 0 lγ ζα ζβ ζγ ζα ζγ (ζα − ζβ ) 0
Πm (9, 12) = −4Ω 4
z Γm11 + 4Ω 2 2 z 0 Γm21 − 12Ω 2l
z Γm31
lα lα lβ lα γ

lγ ζα ζβ ζγ 0 lγ ζβ2 ζγ ζβ ζγ (ζα − ζβ ) 0
+ 4Ω 2 z Γm12 − 4Ω 4 z 0 Γm22 + 12Ω z Γm32
2
lα lβ lβ lβ2 lγ

ζα2 ζβ2 2
2 ζα ζβ ζγ
ζ2 ζ2
2 α γ
Πm (10, 10) = 16Ω2 Γm22 − 16Ω (Γ m32 + Γm23 ) + 16Ω Γm33
lβ4 lβ2 lγ2 lγ4

ζα2 ζβ2 2
2 ζα ζβ ζγ
ζ ζ2ζ
2 α β γ
Πm (10, 11) = −16Ω2 2 Γ m21 + 16Ω Γm31 + 16Ω Γm23
lα2 lβ lα2 l2
γ lβ2 lγ2
ζα ζβ ζγ2
− 16Ω2 Γm33
lγ4

ζα2 ζβ ζγ ζ2 ζ2
2 α γ
ζ ζ2ζ
2 α β γ
Πm (10, 12) = 16Ω2 2 l2
Γ m21 − 16Ω Γ m31 − 16Ω Γm22
lα β
2 l2
lα γ lβ4
ζα ζβ ζγ2
+ 16Ω2 Γm32
lβ2 lγ2

ζα2 ζβ2 ζ ζ2ζ


2 α β γ
ζ2ζ2
2 β γ
Πm (11, 11) = 16Ω2 4
Γm11 − 16Ω 2 l2
(Γ m31 + Γ13 ) + 16Ω Γm33
lα lα γ lγ4

ζα2 ζβ ζγ ζ ζ ζ2
2 α β γ
ζ ζ2ζ
2 α β γ
Πm (11, 12) = −16Ω2 Γ m11 + 16Ω Γm31 + 16Ω Γm12
4
lα lα2 lγ2 lα2 lβ2
ζβ2 ζγ2
− 16Ω2 Γm32
lβ2 lγ2

ζα2 ζγ2 ζ ζ ζ2
2 α β γ
ζ2ζ2
2 β γ
Πm (12, 12) = 16Ω2 Γm11 − 16Ω (Γm12 + Γ m21 ) + 16Ω Γm22
lα4 lα2 lβ2 lβ4

(1 − ζα )2
Πs (1, 1) = Γs11
4
(1 − ζα )(1 − ζβ )
Πs (1, 2) = Γs12
4
(1 − ζα )(1 − ζγ )
Πs (1, 3) = Γs13
4
(1 − ζβ )2
Πs (2, 2) = Γs22
4
(1 − ζβ )(1 − ζγ )
Πs (2, 3) = Γs23
4

106
(1 − ζγ )2
Πs (3, 3) = Γs33
4

The integral (B.2) can be evaluated analitically respect to the thickness while
the integration over the mid-surface of the element has to be performed numeri-
cally. In order to solve the integration over the thickness the dependence of Πm
and Πs on z 0 is exploited:
 
Πmm zΠmS zΠbmA Πmd
 zΠSm z 2 ΠSS z 2 ΠbSA zΠSd 
Πm =
zΠb

Am z 2 ΠbAS z 2 ΠbAA zΠb Ad
Πdm zΠdS zΠbdA zΠdd
£ s ¤
Πs = ΠAA

where all the sub-matrixes are of shape 3x3.


The integral (B.2) can be threfore written in the following way:
Z Z h
2 p
De = σ0 am + as + bz 0 + cz 02 dz 0 dS (B.5)
Se −h
2

where h is the thickness of the element and:

am = qTm Πmm qm + 2qTm Πmd qd + qTd Πdd qd (B.6a)


as = qsA T ΠsAA qsA (B.6b)
T
b= 2qTm Πms qS + 2qTm ΠbmA qbA
+ 2qTS ΠSd qd + 2qbA ΠbAd qd (B.6c)
T
c= qTS ΠSS qS + 2qTS ΠbSA qbA + qbA ΠbAA qbA (B.6d)

It is pointed out that being Πm and Πs positive definite in EP therefore b2 −4ac ≤


0.
The following cases have been examinated separately:

• c = 0 and b = 0 (corresponding to strains due to axial, shear and drilling


modes)
Z h
p
2 √
am + as + bz 0 + cz 02 dz 0 = am + as h (B.7)
−h
2

• am = 0 and b = 0 (corresponding to strains due to bending and shear modes


only)
Let us indicate with r1 the following square root:
p
r1 = 4as + ch2

107
Appendix B. Details on the compilation of the energy dissipation

therefore:
Z h2 p √
h as r1 + ch
am + as + bz 0 + cz 02 dz 0 = r1 + √ ln √ (B.8)
−h
2
4 2 c r1 − ch

• am = 0, as = 0 and b = 0 (corresponding to strains due to bending modes


only)
Z h
p
2 √ h2
am + as + bz 0 + cz 02 dz 0 = c (B.9)
−h
2
4

• b2 − 4ac = 0
Z h2 p
2bh + ch2
a + bz 0 + cz 02 dz 0 = √ sign (b + ch) +
−h 8 c
2 (B.10)
2bh − ch2
√ sign (b − ch)
8 c

• b2 − 4ac < 0
Let us indicate with r2 and r3 the following square roots:
p
r2 = 4a + 2bh + ch2
p
r3 = 4a − 2bh + ch2

Therefore:
Z p
h
2
−h
a + bz 0 + cz 02 dz 0 =
2
√ 

 c [(b + ch) r2 + (ch − b) r3 ] + 

1 · µ ¶ µ ¶¸
3 ¡ 2
¢ b √ b √
8c 2 
 b − 4ac ln √ − ch + r3 − ln √ + ch + r2  
c c
(B.11)

The integration over the mid-plane can be performed by mean of a classical


quadrature procedure, in particular for the present element three gauss points
have been considered at the following coordinate:
¡ ¢
• G1: ζα = 16 ζβ = 23 ζγ = 16
¡ ¢
• G2: ζα = 16 ζβ = 16 ζγ = 23
¡ ¢
• G3: ζα = 23 ζβ = 16 ζγ = 16

108
Appendix C

Details on the minimization


procedure

The minimization procedure implemented for the limit problems to be solved


with the general shell triangular element has been presented in subsection 3.5.7
skipping some important aspects for the sake of simplicity. In this appendix
both the analytical integration over the thickness of the matrix K introduced
in equation (3.123) and the theoretical derivation of equation (3.125) will be
presented.

C.1 Analytical integration to obtain K


The integration of the matrix K over the thickness presents much more problems
than the integration of the dissipation due to the presence of the denominator.
Using the notation introduced in appendix B and equation (3.124) the matrix K
can be written as:

X Z Z h
2 Πz (χ)
K= LTe TTe CTe σ0 √ dz 0 dSCe Te Le (C.1)
e Se −h
2
am + as + bz 0 + cz 02

where
 Πmm zΠmS zχe ΠbmA Πmd

2 2 b
 zΠ z Π z χ Π zΠ 
Πz = ΛTe (χ) Πe Λe (χ) = Sm SS e SA Sd
zχe ΠbAm z 2 χ2e ΠbAS z 2 χ2e ΠbAA +(1−χe )2 ΠsAA zχe ΠbAd
Πdm zΠdS zχe ΠbdA Πdd
(C.2)
As for the integration of the power dissipation the values of the coefficient am ,
as , b and c lead to the following cases:

109
Appendix C. Details on the minimization procedure

• c = 0, b = 0
This case happens only when no bending deformations occur.
Z h
2 1 h
√ dz 0 = √ (C.3a)
−h2
a a
Z h
2 z
√ dz 0 = 0 (C.3b)
−h
2
a
Z h
2 z2 h3
√ dz 0 = √ (C.3c)
−h
2
a 12 a

• am = 0, as = 0, b = 0
This case can occur only when the axial (qm ) and drilling (qd ) modes are
nulls and have to be applied as constraints. The element is therefore mem-
branally rigid and this allows to avoid the infinity of the first integral.
Z h
2 1
√ dz 0 = ∞ (C.4a)
−h
2
cz 02
Z h
2 z
√ dz 0 = 0 (C.4b)
−h
2
cz 02
Z h
2 z2 h2
√ dz 0 = √ (C.4c)
−h
2
cz 02 4 c

• am = 0, b = 0
Also in this case the axial (qm ) and drilling (qd ) modes are nulls and have
to be applied as constraints. The positive term as allows to integrate the
following terms:

Z h √
2 1 0 1 r1 + ch
√ dz = √ ln √ (C.5a)
−h
2
as + cz 02 c r1 − ch
Z h
2 z
√ dz 0 = 0 (C.5b)
−h2
as + cz 02
Z h √
2 z2 0 h as r1 + ch
√ dz = r1 − √ ln √ (C.5c)
−h
2
as + cz 02 4c 2c c r1 − ch

• b2 − 4ac = 0
From the physics point of view there exist a point on the section where
λ̇2 = 0, therefore in that point no dissipation occurs.This problem has to
be specifically dealed imposing some constraints rising from the variation

110
C.1. Analytical integration to obtain K

¯b¯
of the coefficient a. In particular if ¯ ch ¯ ≤ 1 the point is into the section
otherwise it is external and no problem occurs.
Let us recalling the definition of the power dissipation:
XZ
D= σ0 λ̇dx (C.6)
e Ωe

The matrix K is obtained as a consequence of the derivation process:


Z
∂D X ∂ λ̇
= σ0 dx = Ku̇ (C.7)
∂ u̇ e Ωe ∂ u̇

but p
λ̇ = a (u̇) + b (u̇) z 0 + c (u̇) z 02 (C.8)
and therefore:
XZ ∂a
+ ∂b 0 ∂c 02
σ0 ∂ u̇ ∂ u̇ z + ∂ u̇ z
Ku̇ = dx (C.9)
e Ωe 2 λ̇

b2
But being a = 4c its variation depends on the variation of b and c:

∂a 2bc ∂∂bu̇ − b2 ∂∂cu̇


= (C.10)
∂ u̇ 4c2
while the plastic flow can be expressed as follows being c > 0:
r
b2 1
λ̇ = + bz + cz 2 = √ |b + 2cz| (C.11)
4c 2 c

The variation of the power dissipation become:

XZ √ ∂a + ∂b z 0 + ∂∂cu̇ z 02
Ku̇ = σ0 c ∂ u̇ ∂ u̇ dx
e Ωe |b + 2cz|
¡ ¢ (C.12)
XZ √ 1 (b + 2cz) ∂∂bu̇ − 4c12 b2 − 4c2 z 2 ∂∂cu̇
= σ0 c 2c dx
e Ωe |b + 2cz|

Without loss of generality σ0 can be supposed constant over the element


therefore the expression of the variation become:
 1 Z b + 2cz ∂b 
dx−
X √  2c Ωe |b + 2cz| ∂ u̇ 
Ku̇ = σ0 c  1 Z (b − 2cz) (b + 2cz) ∂c 
 (C.13)
e dx
4c2 Ωe |b + 2cz| ∂ u̇

111
Appendix C. Details on the minimization procedure

The variation of a, b and c can be easily found if the coefficient are expressed
in the following way:
a
a =ρ̃˙ TN Π̃ ρ̃˙ N = u̇T LTe TTe CTe Πa Ce Te Le u̇ (C.14a)
b
b =ρ̃˙ TN Π̃ ρ̃˙ N = u̇ T
LTe TTe CTe Πb Ce Te Le u̇ (C.14b)
c
c =ρ̃˙ TN Π̃ ρ̃˙ N = u̇T LTe TTe CTe Πc Ce Te Le u̇ (C.14c)

where Le is the connectivity matrix, Te the coordinate transformation ma-


trix, Ce the matrix of the natural modes while the Πi matrix are here
reported:
 
Πmm · · Πmd
a  · · · · 
Πa =ΛTe Π̃ Λ =   ·
 (C.15a)
· 
2
12×12 · (1 − χe ) ΠsAA
Πdm · · Πdd
 b 
· ΠmS χe ΠmA ·
b  ΠSm · · ΠSd 
Πb =ΛTe Π̃ Λ =  
 (C.15b)
12×12 χe ΠAmb
· · χe ΠbAd 
· ΠdS χe ΠbdA ·
 
· · · ·
c · Π χ Π b
·
Π =Λe Π̃ Λ = 
c T
· χe Π
SS e SA  (C.15c)
12×12
b
AS χe ΠAA ·
2 b

· · · ·

Therefore the variations of b and c are:


∂b
= 2LTe TTe CTe Πb Ce Te Le u̇ (C.16a)
∂ u̇
∂c
= 2LTe TTe CTe Πc Ce Te Le u̇ (C.16b)
∂ u̇
and do not depend on z 0 . Now the integrations over the thickness can be
performed:
¯b¯
– ¯ ch ¯≤1

Z h Z b
− 2c Z h
2 b + 2cz 0 2 b
dz = −dz 0 + dz 0 = (C.17a)
−h
2
|b + 2cz| −h
2
b
− 2c c
Z h Z − 2cb
2 (b − 2cz) (b + 2cz) 0
dz = − (b − 2cz) dz 0 +
−h
2
|b + 2cz| h
−2
Z h2 (C.17b)
3b2 ch2
(b − 2cz) dz 0 = −
b
− 2c 2c 2

112
C.1. Analytical integration to obtain K

Therefore:
X √ Z · µ ¶ ¸
T T T b b 3b2 h2 c
K= σ0 cLe Te Ce Π − − Π dSCe Te Le
e Se c2 4c3 4c
(C.18)
¯b¯
– ¯ ch ¯>1

Z h ¶ µ
2 b + 2cz 0 b
dz = sign h (C.19a)
−h2
|b + 2cz| ch
Z h µ ¶
2 (b − 2cz) (b + 2cz) 0 b
dz = sign bh (C.19b)
−h
2
|b + 2cz| ch
Therefore:
X √ µ ¶Z · ¸
T T T b h b bh c
K= σ0 cLe Te Ce sign Π − 2 Π dSCe Te Le
e
ch Se c 2c
(C.20)
In order to avoid possibly indefiniteness of the matrix K the constraints
due to the equation (C.10) have to be imposed. From equations (C.14)
and (C.10) one obtains the following relation:
· ¸
bc b2
LTe TTe CTe Πa − Πb + 2 Πc Ce Te Le u̇ = 0 (C.21)
2 4c
The constraints (C.21) can be imposed using the same procedure adopted
for the rigid elements.
• b2 − 4ac < 0
Z h
2 1 1 x + 1 + r3
√ dz 0 = √ ln (C.22a)
−h2
0
a + bz + cz 02 c x − 1 + r4
Z h · ¸
2 z0 h x + 1 + r3
√ dz 0 = √ −x ln + r3 − r4 (C.22b)
−h
2
a + bz 0 + cz 02 2 c x − 1 + r4
 
Z h
02 2 (3x + 1) r4 − (3x − 1) r3 +
2 z h
√ dz 0 = √  ¡ 2 ¢ x + 1 + r3  (C.22c)
−h a + bz 0 + cz 02 8 c 3x − 4y ln
2
x − 1 + r4
where
b
x= (C.23a)
ch
4a
y= 2 (C.23b)
p ch
r3 = y + 2x + 1 (C.23c)
p
r4 = y − 2x + 1 (C.23d)

113
Appendix C. Details on the minimization procedure

C.2 Minimization respect to χ


At solution the equation (3.122b) must be satisfied and this condition allow to
obtain the correct bending factor.
Let us suppose the bending factor χe constant over a whole element without any
correlation with the bending factor of adjacent elements. Subject to this hypo-
thesis the minimization can be performed separately on any element. Pointing
out the dependance on χe one obtains:
Z q Z q
De = σ0 ρ̇T ΛTe (χe ) Πe Λe ρ̇dx = σ0 λ̇2 (χe )dx (C.24)
Ωe Ωe

The variation respect to χe is:


Z ∂ λ̇2
∂De ∂χ
= σ0 dx (C.25)
∂χe Ωe 2λ̇
If the calculations are performed:

λ̇2 (χe ) =q̇Tm Πmm q̇m + 2z q̇Tm ΠmS q̇S + 2zχe q̇Tm ΠbmA q̇A +
2q̇Tm Πmd q̇d + z 2 q̇TS ΠSS q̇S + 2z 2 χe q̇TS ΠbSA q̇A +
h i (C.26)
2
2z q̇TS ΠSd q̇d + q̇TA z 2 χ2e ΠbAA + (1 − χe ) ΠsAA q̇A +
2zχe q̇TA ΠbAd q̇d + q̇Td Πdd q̇d

and

∂ λ̇2 (χe ) h i
= 2z q̇Tm ΠbmA q̇A + 2z 2 q̇TS ΠbSA q̇A + 2χe q̇TA z 2 ΠbAA + ΠsAA q̇A
∂χe (C.27)
− 2q̇TA ΠsAA q̇A + 2z q̇TA ΠbAd q̇d

Expression (C.27) can be written in a more simple way if two matrix, A and B,
are introduced:
∂ λ̇2 (χe )
= q̇TA Aq̇A χe + ρ̇TN Bρ̇N (C.28)
∂χe
Substituting equation (C.28) in (C.25) and using the condition (3.122b) lead to:
R ρ̇T
N Bρ̇N
Ωe λ̇
dx
χe = − R q̇T A (C.29)
A q̇A
Ωe λ̇
dx

and the equation (3.125) is demonstrated.


The analytical evaluation of the integrals over the thickness can be done as for
the matrix K (see section C.1).

114
Bibliography

[1] Wlodzimierz Abramowicz. The effective crushing distance in axially com-


pressed thin-walled metal columns. 1(3):309–317, 1983.

[2] W. Abramowiez and T. Wierzbicki. A kinematic approach to crushing of


shell structures. Technical Report 790992, SAE, 1979.

[3] J. M. Alexander. An approximate analysis of the collapse of thin cylin-


drical shells under axial loading. Quart. Journal Mech. and Applied Math,
XIII(1):10–15, 1960.

[4] D. J. Allman. A compatible triangular element including vertex rotations


for plane elasticity analysis. Computers and Structures, 19(1-2):1–8, 1984.

[5] D. J. Allman. Evaluation of the constant strain triangle with drilling rota-
tions. International Journal of Numerical Methods in Engineering, 26:2645–
2655, 1988.

[6] A. Andronicou and A. C. Walker. A plastic collapse mechanism for cylinders


under uniaxial end compression. Journal of Constructional Steel Research,
1(4):23–34, September 1981.

[7] J. H. Argyris. Continua and discontinua. In Proceedings of the Interna-


tional Conference on Matrix Methods in Structural Mechanics, pages 1–198,
Wright-Patterson Air Force Base, Dayton, Ohio, October 1965.

[8] J. H. Argyris. Three-dimensional anisotropic and inhomogeneous elastic


media matrix analysis for small and large displacements. Ingenieur archiv,
XXXIV:33–55, 1965.

[9] J. H. Argyris, H. Balmer, J. St. Doltsinis, P. C. Dunne, M. Haase,


M. Kleiber, G. A. Malejannakis, H. P. Mlejnek, M. Muller, and D.W.
Scharpf. Finite element method - the natural approach. Computer Methods
in Applied Mechanics and Engineering, 17/18:1–106, 1979.

115
Bibliography

[10] J. H. Argyris and O. E. Bronlund. The natural factor formulation of the


stiffness for the matrix displacement method. Computer Methods in Applied
Mechanics and Engineering, 5:97–119, 1975.
[11] J. H. Argyris, P. C. Dunne, G. A. Malejannakis, and E. Schelkle. A simple
triangular facet shell element with applications to linear and non-linear
equilibrium and elastic stability problems. Computer Methods in Applied
Mechanics and Engineering, 10:371–403, 1977.
[12] J. H. Argyris, M. Haase, and H. P. Mlejnek. On an unconventional but natu-
ral formation of a stiffness matrix. Computer Methods in Applied Mechanics
and Engineering, 22:1–22, 1980.
[13] J. H. Argyris, M. Haase, and H. P. Mlejnek. Some considerations on the nat-
ural approach. Computer Methods in Applied Mechanics and Engineering,
30:335–346, 1982.
[14] J. H. Argyris, M. Papadrakakis, C. Apostolopoulou, and S. Koutsourelakis.
The tric shell element: theoretical and numerical investigation. Computer
Methods in Applied Mechanics and Engineering, 182:217–245, 2000.
[15] J. H. Argyris, M. Papadrakakis, and L. Karapitta. Elastoplastic analysis of
shells with the triangular element tric. In M. Papadrakakis, A. Samartin,
and E. Onate, editors, Computational methods for shell and spatial struc-
tures, Athens, Greece, June 2000. IASS-IACM 2000, ISASR-NTUA.
[16] J. H. Argyris and L. Tenek. An efficient and locking-free flat anisotropic
plate and shell triangular element. Computer Methods in Applied Mechanics
and Engineering, 118:63–119, 1994.
[17] J. H Argyris and L. Tenek. Linear and geometrically non-linear bending of
isotropic and multilayered composite plates by the natural mode method.
Computer Methods in Applied Mechanics and Engineering, 113:207–251,
1994.
[18] J. H. Argyris, L. Tenek, and L. Olofsson. Tric: a simple but sophisticated
3-node triangular element based on 6 rigid-body and 12 straining modes for
fast computational simulations of arbitrary isotropic and laminated com-
posite shells. Computer Methods in Applied Mechanics and Engineering,
145:11–85, 1997.
[19] J. H. Argyris, L. Tenek, M. Papadrakakis, and C. Apostolopoulou. Post-
buckling performance of the tric natural mode triangular element for
isotropic and laminated composite shells. Computer Methods in Applied
Mechanics and Engineering, 166:211–231, 1998.
[20] R. Baldacci, G. Ceradini, and E. Giangreco. Plasticitá. CISIA - Milano,
1974.

116
Bibliography

[21] A. Ben-Tal, M. Teboulle, and W. H. Yang. A least-squares-based method for


a class of nonsmooth minimization problems with applications in plasticity.
Journal of Applied Mathematics and Optimization, 24:273–288, 1991.

[22] P. G. Bergan. Finite element based on energy orthogonal functions. Nu-


merical Methods in Engineering, 15:1541–1555, 1980.

[23] P. G. Bergan and C. A. Felippa. A triangular membrane element with


rotational degrees of freedom. Computer Methods in Applied Mechanics
and Engineering, 50:25–69, 1985.

[24] P. G. Bergan and L. Hanssen. A new approach for deriving ’good’ elements.
In Brunel University, editor, Proceedings of the Conference on Mathematics
of Finite Element and Applications, 1975.

[25] P. G. Bernard and M. K. Nygard. Finite elemens with increased freedom in


choosing shape functions. Numerical Methods in Engineering, 20:643–663,
1984.

[26] Don O. Brush and Bo O. ALmroth. Buckling of Bars, Plates and Shells.
Mc-GRAW-HILL, 1975.

[27] C. J. Burgoyne and M. G. Brennan. Exact ilyushin yield surface. Interna-


tional Journal of Solids and Structures, 30(8):1113–1131, 1993.

[28] L. Del Rio Cabrera. Analyse limite des plaques rectangulaires. PhD thesis,
Faculté Polytechnique de Mons, 1970.

[29] A. Capsoni and L. Corradi. A plane strain formulation of the elastic-plastic


constitutive law for hardening von mises materials. International Journal
of Solids and Structures, 32:3515–3531, 1995.

[30] A. Capsoni and L. Corradi. A finite element formulation of the rigid-plastic


limit analysis problem. International Journal of Numerical Methods in En-
gineering, 40:2063–2086, 1997.

[31] A. Capsoni and L. Corradi. A mixed finite element model for plane strain
elastic-plastic analysis. part i. formulation and assessment of the overall be-
haviour. Computer Methods in Applied Mechanics and Engineering, 141:67–
79, 1997.

[32] A. Capsoni and L. Corradi. A mixed finite element model for plane strain
elastic-plastic analysis. part ii. application to the 4-node bilinear element.
Computer Methods in Applied Mechanics and Engineering, 141:81–93, 1997.

[33] A. Capsoni and L. Corradi. Limit analysis of plates - a finite element for-
mulation. Structural Engineering and Mechanics, 8:325–341, 1999.

117
Bibliography

[34] A. Capsoni, L. Corradi, and P. Vena. Limit analysis of orthotropic struc-


tures based on hill’s yield condition. International Journal of Solids and
Structures, 38:3945–3963, 2001.

[35] Michele Capurso. Sulla determinazione del carico di collasso delle superfici
a doppia curvatura - nota 2. Costruzioni Metalliche, 6:404–412, 1965.

[36] Michele Capurso. Sulla determinazione del carico di collasso delle superfici
a doppia curvatura - nota 1. Costruzioni Metalliche, 5:339–346, 1965.

[37] Michele Capurso. Sul calcolo elasto-plastico delle piastre circolari e delle
volte di rivoluzione ribassate in regime di grandi spostamenti. Costruzioni
Metalliche, 5:386–403, 1969.

[38] Michele Capurso. Sul collasso elasto-plastico delle piastre metalliche com-
presse al perimetro. Costruzioni Metalliche, 2:134–141, 1969.

[39] Raffaele Casciaro and Leonardo Cascini. A mixed formulation and mixed fi-
nite elements for limit analysis. International Journal of Numerical Methods
in Engineering, 18:211–243, 1982.

[40] Shen-Yeh Chen. An approach for impact structure optimization using the
robust genetic algorithm. Finite Element in Analysis and Design, 37:431–
446, 2001.

[41] Edmund Christiansen and Søren Larsen. Computations in limit analysis for
plastic plates. International Journal of Numerical Methods in Engineering,
19:169–184, 1983.

[42] M. K. Chryssanthopoulos, Carlo Poggi, and A. Spagnoli. Buckling design of


conical shells based on validated numerical models. Thin Walled Structures,
31:257–270, 1998.

[43] M. K. Chryssanthopoulus and Carlo Poggi. Collapse strength of unstiffened


conical shells under axial compression. Journal of Constructional Steel Re-
search, 57:165–184, 2001.

[44] L. Corradi and P. Vena. Limit analysis of orthotropic plates. International


Journal of Plasticity, Accepted for pubblication, June 2001.

[45] Leone Corradi and Nicola Panzeri. Post-collapse analysis of rigid-plastic


shells by sequential limit analysis. In Proc. XV Congresso AIMETA di
Meccanica Teorica e Applicata, 2001.

[46] Leone Corradi, Nicola Panzeri, and Carlo Poggi. Post-critical behaviour of
moderately thick axisymmetric shells: a sequential limit analysis approach.
International Journal of Structural Stability and Dynamics, 1(3):293–311,
2001.

118
Bibliography

[47] J. G. A. Croll. Lower bound elasto-plastic buckling of cylinders. Proc. Instn.


Civ. Engrs, 71:235–261, 1981.

[48] ECCS. Buckling of steel shells, european recommendations. Technical Re-


port 4th edn, European Convention for Constructional Steelwork, Brussels,
1988.

[49] H. El-Sobky, A. A. Singace, and M. Petsios. Mode of collapse and energy


absorption characteristics of constrained frusta under axial impact loading.
International Journal of Mechanical Sciences, 43:743–757, 2001.

[50] C. P. Ellinas, W. J. Supple, and A. C. Walker. Buckling of Offshore Struc-


tures. Granada, London, 1984.

[51] C. A. Felippa, B. Haugen, and C. Militello. From the individual element test
to finite element templates: evolution of the patch test. Numerical Methods
in Engineering, 38:199–239, 1995.

[52] A. Fortunato. Folding instabilities of thin walled structures. In AIMETA


’01 - XV Congresso AIMETA di Meccanica Teorica e Applicata, 2001.

[53] C. Gavarini. Impostazione di un metodo generale per lo studio del com-


portamento delle strutture rigido-plstiche nella fase susseguente al collasso.
Atti dell’Istituto di Scienza delle Costruzioni dell’Universitá di Pisa, 1968.

[54] R. Greiner, J. M. Rotter, and H. Schmidt. The new eurocode on strength


and stability of steel shell structures. Nordic Steel Construction Conference
’98, Bergen, Norway, september 1998.

[55] S. R. Guillow, G. Lu, and R. H. Grzebieta. Quasi static axial compression of


thin-walled circular aluiminium tubes. International Journal of Mechanical
Sciences, 43:2103–2123, 2001.

[56] A. G. Hanssen, M. Langseth, and O. S. Hopperstad. Optimum design for


energy absorption of square aluminium columns with aluminium foam filler.
International Journal of Mechanical Sciences, 43:153–176, 2001.

[57] P. G. Hodge. Plastic analysis of Structures. Mc-Graw Hill, 1959.

[58] P. G. jr. Hodge and T. Belytschko. Numerical methods for the limit analysis
of plates. Journal of Applied Mechanics - ASME, 35:796–802, December
1967.

[59] M. R. Horne and W. Merchant. The stability of frames. Robert Maxwell,


1965.

[60] M. R. Horne and L. J. Morris. Plastic design of low-rise frames. Granada


Publishing Limited, 1981.

119
Bibliography

[61] H. Huh and Wei H. Yang. A general algorithm for limit solutions of plane
stress problems. International Journal of Solids and Structures, 28(6):727–
738, 1991.
[62] Hoon Huh, Kee-Poong Kim, and Hyun Sup Kim. Collapse simulation of
tubular structures using a finite element limit analysis approach and shell el-
ements. International Journal of Mechanical Sciences, 43:2171–2187, 2001.
[63] Hoon Huh, Choong Ho Lee, and Wei H. Yang. A general algorithm for
plastic flow simulation by finite element limit analysis. International Journal
of Solids and Structures, 36:1193–1207, 1999.
[64] Nguyen Dang Hung, Marc Trapletti, and Daniel Ransart. Bornes quasi-
inferieures et bornes superieures de la pression de ruine des coques de revolu-
tion par la methode des elements finis et par la programmation non-lineaire.
International Journal of Non-Linear Mechanics, (13):79–102, 1978.
[65] Adnan Ibrahimbegovic, Robert L. Taylor, and Edward L. Wilson. A robust
quadrilateral membrane finite element with drilling degrees of freedom. In-
ternational Journal of Numerical Methods in Engineering, 30:445–457, 1990.
[66] A. A. Iliouchine. Plasticité. Eyrolles, Paris, 1956.
[67] G. L. Jiang. Nonlinear finite element formulation of kinematic limit analysis.
International Journal of Numerical Methods in Engineering, 38:2775–2807,
1995.
[68] W. Johnson and A. G. Mamalis. Crashworthiness of vehicles. MEP, London,
UK, 1978.
[69] W. Johnson, P. D. Soden, and S. T. S. Al-Hassani. Inextensional collapse of
thin-walled tubes under axial compression. J. of Strain Analysis, 12(4):317–
330, 1977.
[70] Norman Jones. Some recent developments and future trends in thin-walled
sections for structural crashworthiness. Thin Walled Structures, 32:231–233,
1998.
[71] D. Karagiozova and Norman Jones. Dynamic effects on buckling and energy
absorption of cylindrical shells under axial impact. Thin Walled Structures,
39:583–610, 2001.
[72] D. Kecman. An engineering approach to crashworthiness of thin-walled
beams and joints in vehicle structures. Thin Walled Structures, 28:309–320,
1997.
[73] Dusan Kecman. Bending collapse of rectangular section tubes in relation
to the box roll over problem. PhD thesis, Cranfield Institute of Technology,
1979.

120
Bibliography

[74] C. M. Kinderwater and H. Georgi. Composite strength and energy absorp-


tion as an expect of structural crash resistance, chapter 6, pages 189–235.
Elsevier, London, 1993.
[75] C. M. Kinderwater, D. Kohlgruber, and A. Johnson. Composite vehicle
structural crashworthiness - a status of design methodology and numer-
icl simulation techniques. In IJCRASH’98, International Crashworthiness
Conference, pages 444–460, Deaburn (USA), September 1998.
[76] N. E. Kirk, A. F. Kalton, E. G. Candy, G. C. Nicholson, and C. Wil-
son. Modifications to existing rolling stocks to improve crashworthiness.
In IJCRAsh’98, International Crashworthiness Conference, pages 239–250,
Deaburn (USA), September 1998 1998.
[77] Brad L. Koziey and Farooque A. Mirza. Consistent thick shell element.
Computers and Structures, 65(4):531–549, 1997.
[78] J. H. Ku, K. S. Seo, S. W. Park, and Y. J. Kim. Axial crushing behaviour
of the intermittent tack-welded cylindrical tubes. International Journal of
Mechanical Sciences, 43:521–542, 2001.
[79] L. Lackman and J. Penzien. Buckling of circular cones under axial com-
pression. Journal of Applied Mechanics, 27:458–460, 1960.
[80] R. H. Lance and E. T. Onat. Analysis of plastic shallow shells. Journal of
Applied Mechanics, pages 199–209, 1962.
[81] Y. H. Liu, Z. Z. Cen, and B. Y. Xu. A numerical method for plastic limit
analysis of 3-d structures. International Journal of Solids and Structures,
32(12):1645–1658, 1995.
[82] W. T. Lowe, S. T. S. Al-Hassani, and W. Johnson. Impact behaviour of
small scale model motor coaches. Proc. Instit. Mech. Engrs, 186:36–72,
1972.
[83] D. G. Luenberger. Linear and Nonlinear programming. Addison Wesley,
1984.
[84] Richard H. MacNeal and Robert L. Harder. A refined four-noded membrane
element with rotational degrees of freedom. Computers and Structures,
28(1):75–84, 1988.
[85] G. Maier, A. Zavelani, and D. Beneditti. A finite element approach to
optimal design of plastic structures in plane stress. International Journal
of Numerical Methods in Engineering, 4:455–473, 1972.
[86] A. G. Mamalis and W. Johnson. The quasi-static crumpling of thin-walled
circular cylinders and frusta under axial compression. International Journal
of Mechanical Sciences, 25(9):713–732, 1983.

121
Bibliography

[87] A. G. Mamalis, W. Johnson, and G. L. Viegelahn. The crumpling of steel


thin-walled tubes and frusta under axial compression at elevated strain-
rates: some experimental results. International Journal of Mechanical Sci-
ences, 26(11/12):537–547, 1984.

[88] A. G. Mamalis, D. E. Manolakos, S. Saigal, G. L. Viegerlahn, and W. John-


son. Extensible plastic collapse of thin-wall frusta as energy absorbers.
International Journal of Mechanical Sciences, 28(4):219–229, 1986.

[89] A. G. Mamalis, D. E. Manolakos, G. L. Viegelahn, and W. Johnson. The


modelling of the progressive extensible plastic collapse of thin-wall shells.
International Journal of Mechanical Sciences, 30(3/4):249–261, 1988.

[90] A. G. Mamalis, D. E. Manolakos, G. L. Viegerlahn, and N. M. Vaxevanidis.


On the inextensional axial collapse of thin pvc conical shells. International
Journal of Mechanical Sciences, 28(5):323–335, 1986.

[91] J. B. Martin. Plasticity: fundamentals and general results. MIT Press, MA,
1975.

[92] Q. Meng, S. T. S. Al-Hassani, and P. D. Soden. Axial crushing of square


tubes. International Journal of Mechanical Sciences, 25(9-10):747–773,
1983.

[93] J. C. Nagtegaal, D. M. Parks, and J. R. Rice. On numerically accurate finite


element solutions in the fully plastic range. Computer Methods in Applied
Mechanics and Engineering, 4:153–177, 1974.

[94] R. P. Nimmer and J. Mayers. Limit point buckling loads of axially com-
pressed, circular cylindrical shells. the effect of nonlinear behavior. Journal
of Applied Mechanics, 46:386–391, 1979.

[95] E. T. Onat. Plastic analysis of shallow conical shells. Journal of the Engi-
neering Mechanics Division, pages 1–12, December 1960.

[96] Nicola Panzeri and Carlo Poggi. Stability and strength of conical shells
subject to axial load and external pressure. In S.L. Chan and J.G. Teng,
editors, Advances in Steel Structures, ICASS 99, pages 621–630, Hong Kong,
December 1999. Elsevier.

[97] T. H. H. Pian. Dynamic response of thin shell structures. In Proocedings of


the 2nd Symposium on Naval Structural Mechanics, pages 443–452, Oxford,
1960. Pergamon Press.

[98] Carlo Poggi. Numerical analysis of imperfect conical shells. In Krupka and
Schneideer, editors, proc. of Carrying Capacity of Steel Shell Structures,
pages 317–323, Brno, 1997.

122
Bibliography

[99] A. G. Pugsley. On the crumpling of thin tubular struts. Quart. Journ.


Mech. and Appl. Math, XXXII(1):1–7, 1979.

[100] Alfred Pugsley and M. Macaulay. The large-scale crumpling of thin cylin-
drical columns. Quart. Journ. Mech. and Applied Math., XIII(1):1–9, 1960.

[101] Yves Ravalard, Pascal Drazetic, and Eric Markiewicz. Prototypage virtuel:
exemple de crash de véhicules ferroviaires. Mecanique & Industries,
1(4):383–395, July 2000.

[102] Angelo Ravasi and Alessandro Vecchi. Analisi limite rigido-plastica ad ele-
menti finiti di piastre alla kirchhoff. Master’s thesis, Politecnico di Milano,
Milano, 1996.

[103] J. D. Reid. Crashworthiness of automotive steel midrails: Thickness and


material sensitivity. Thin Walled Structures, 26(2):83–103, 1996.

[104] M. Robinson. A comparison of yield surfaces for thin shells a comparison of


yield surfaces for thin shells. International Journal of Mechanical Sciences,
13:345–354, 1971.

[105] Carl T. F. Ross, David Sawkins, and Terry Johns. Inelastic buckling of thick-
walled circular conical shells under external hydrostatic pressure. Ocean
Engineering, 26:1297–1310, 1999.

[106] M. Save. Limit analysis of plates and shells: research over two decades.
13(3-4):343–370, 1985.

[107] Emanuele Scalvini and Paolo Scorza. Analisi limite rigido plastica ad ele-
menti finiti di piastre alla reissner mindlin. Master’s thesis, Politecnico di
Milano, dip. di Ingegneria Strutturale, 1997.

[108] P. Seide. Axisymmetrical buckling of circular cones under axial compression.


Journal of Applied Mechanics, 23:625–628, 1956.

[109] M. Seitzberger. Contributions to an efficient numerical analysis of the plas-


tic collapse behaviour of thin-walled structures. PhD thesis, Vienna Institute
of Technology, VDI Reihe 18, 2000.

[110] Markus Seitzberger and Franz G. Rammerstorfer. On the application of the


exact ilyushin yield surface for plastic collapse analysis of shell structures.
In IASS-IACM 2000 Fourth International Colloquium on Computation of
Shell & Spatial Structures, Chania - Crete, Greece, June 2000.

[111] S. W. Sloan and P. W. Kleeman. Upper bound limit analysis using dis-
continuous velocity fields. Computer Methods in Applied Mechanics and
Engineering, 127:293–314, 1995.

123
Bibliography

[112] Ian M. Smith and William Duncan. The effectiveness of excessive nodal
continuities in the finite element analysis of thin rectangular and skew plates
in bending. International Journal of Numerical Methods in Engineering,
2:253–257, 1970.

[113] P. D. Soden, S. T. S. Al-Hassani, and W. Johnson. The crumpling of


polyvinylchloride tubes under static and dynamic axial loads. In J. Harding,
editor, Mechanical Properties at High Rates of Strain, number 21 in Conf.
Series, pages 327–338. Inst. of Physics, 1974.

[114] Henryk Stolarski and Ted Belytschko. Limitation principles for mixed fi-
nite elements based on the hu-washizu variational formulation. Computer
Methods in Applied Mechanics and Engineering, 60:195–216, 1987.

[115] Henryk Stolarski, Ted Belytschko, and Nicholas Carpenter. Bending and
shear mode decomposition in c0 structural element. 11(2):153–176, 1983.

[116] G. Strang, H. Mattles, and R. Teman. Mathematical and computational


methods in plasticity. In S. Nemat-Nasser, editor, IUTAM, Symposium on
Variational Methods in the Mechanics of Solids, pages 20–28. Pergamon,
Oxford, 1978.

[117] Ala Tabiei and Jin Wu. Roadmap for crashworthiness finite element simu-
lation of roadside safety structures. Finite Element in Analysis and Design,
34:145–157, 2000.

[118] R. Temam. Mathematical problems in plasticity. Gauthier Villard, Paris,


1984.

[119] S.W. Tsai and E.M. Wu. A general theory of strength for anisotropic ma-
terials. Journal of Composite Materials, 8:58–80, 1971.

[120] Viggo Tvergaard. On the transition from a diamond mode to an axisym-


metric mode of collapse in cylindrical shells. International Journal of Solids
and Structures, 19(10):845–856, 1983.

[121] T. Wierzbicki. On the formation and growth of folding modes in compressed


thin-walled structures. In The buckling of structures in theory and practice.
University College London, September 1982. Paper offered for the IUTAM
Symposium on Collapse.

[122] T. Wierzbicki and W. Abramowiez. On the crushing mechanics of thin-


walled structures. Journal of Applied Mechanics, 50:727–734, December
1983.

[123] T. Wierzbicki and T. Akerstrom. Dynamic crushing of strain rate sensitive


box columns. In Proc. 2nd Int. Conf. Vehicle Structural Mech., Southfield,
Michigan, April 1977.

124
Bibliography

[124] P. H. Wirsching and R. C. Slater. The beer can as a shock absorber. Journal
of Engineering Materials and Technology, 95:244, 1973.
[125] Ohokubo Y, T. Akamatsu, and K. Shirasawa. Mean crushing strength of
closed-hat sections members. Technical Report 740040, SAE.
[126] W. H. Yang. A duality theorem for plastic plates. Acta Mechanica, 69:177–
193, 1987.
[127] W. H. Yang. On generalized hölder inequality. Nonlinear Analysis,
16(5):489–498, 1991.
[128] W. H. Yang. Large deformation of structures by sequential limit analysis.
International Journal of Solids and Structures, 30(7):1001–1013, 1993.
[129] Y. Yoshimura. On the mechanism of buckling of a circular cylindrical shell
under axial compression. Technical Report 5, Institute of Science and Tech-
nology of the University of Tokyo, 1951.
[130] O. C. Zienkiewicz. The finite element method. McGraw-Hill, Scarborough,
CA, 1977.

125
. . . and to my brothers

You might also like