Universitat Politècnica de Catalunya-ETSECCPB

Programa de doctorado en Ingenieria Civil

PhD’s thesis

Innovative mathematical and numerical

models for studying the deformation of shells
during industrial forming processes with the
Finite Element Method

Vicente Mataix Ferrándiz

Director: Eugenio Oñate Ibañez de Navarra

Co-director: Riccardo Rossi
Page 2 of 374 Vicente Mataix Ferrándiz
 Ce qui est simple est toujours faux. Ce qui ne lest pas est inutilisable

Paul Valéry (1871 1945 AD, French poet, essayist, and philosopher)

Vicente Mataix Ferrándiz Page 3 of 374

 Empty space

Essentially, all models are wrong, but some are useful

Aphorism, attributed to George E. P. Box (1919 2013 AD, British statistician)

Page 4 of 374 Vicente Mataix Ferrándiz

 Abstract
This document contains the result resulting from the work in the doctoral thesis Innovative mathematical
and numerical models for studying the deformation of shells during industrial forming processes with the
Finite Element Method. The objective of this thesis is to contribute to the development of finite element
methods for the analysis of the stamping processes, an area of problems with a very clear industrial
application. Indeed these kinds of problems involve multiple disciplines and require the understanding
of different mechanical problems, being the most relevant disciplines the continuous mechanics, the
plasticity, contact problems, among others, depending of the problematic of study.
To achieve the proposed goals, attention is focused in the first section of this thesis in the solid-shell
elements, which are an attractive kind of element for the simulation of forming processes. This is due to the
fact that any kind of generic 3D constitutive law can be employed without any kind of additional hypothesis,
besides the thermomechanic problem is formulated without additional assumptions. Additionally this type
of element allows the three-dimensional description of the deformable body, thus contact on both sides of
the element can be treated easily.
This work will present in first place the development of a triangular prism element as a solid-shell,
for the analysis of thin/thick shell, undergoing large deformations. The element is formulated in Total
Lagrangian (TL) formulation, and employs the neighbour (adjacent) elements to perform a local patch
to enrich the displacement field. In the original formulation a modified right CauchyGreen deformation
tensor (C̄) is obtained; in the present work, a modified deformation gradient (F̄) is obtained, which allows
to generalise the methodology and thanks to this we are able to employ the push-forward and pullback
concepts. The push-forward and pullback technique provide a mathematically consistent method for
define the time derivatives of the tensors; and for example it can be employed to work with elasto-plasticity.
The element is based in three modifications: (a) a classical assumed strain approach for transverse
shear strains (b) an assumed strain approach for the in-plane components using information from
neighbour elements and (c) an averaging of the volumetric strain over the element. The objective is to
use this type of element for the simulation of shells avoiding transverse shear locking, improving the
membrane behaviour of the in-plane triangle and to handle quasi-incompressible materials or materials
with isochoric plastic flow. The element considers just one Gauss point in the plane and a chosen number
of integration points along the axis, thanks to this it is possible to consider problems with a significant
non-linearity related to plasticity.
This work will continue presenting the contact formulation developed, which consists in state-of-art
of the numerical contact mechanics formulation for implicit simulations. Consisting in an exact mortar
integration of the contact interface, this allows to obtain the most consistent integration possible between
the integration domains, as well as the most exact solution possible. The implementation also considers
several optimisation algorithms, like the penalty constraint optimisation, but particularly remarkable
the consideration of a Augmented Lagrangian Method (ALM) with dual Lagrange multipliers, a new
contribution of this work. The latter allows to static condensate the system of equations, allowing to remove
the Lagrange Multiplier (LM) of the resolution and therefore permitting the consideration of iterative
solvers. Additionally, the formulation has been properly linearised, ensuring the quadratic convergence of
the problem. In order to solve the system of equations, a semi-smooth Newton method is considered,
consisting in an active set strategy, extensible also in the case of frictional problems. The formulation works
both for frictionless and frictional problems, the later essential for the simulation of forming processes.
This frictional formulation is framed in the traditional friction models, like the Coulomb friction, but the
development presented can be extended to any type of frictional model. This contact formulation is fully
compatible with the solid-shell introduced on this work.

Vicente Mataix Ferrándiz Page 5 of 374

 The missing ingredient necessary in order to successfully perform the industrial processes would
be the constitutive models. This will be materialised in the plasticity formulation considered in this work.
In these models, we will be able to consider large strain deformation plasticity models, with arbitrary
combination of yield surfaces and plastic potentials, the so-called non-associative models. In order to
compute the corresponding tangent tensor for these general laws, numerical implementations based on
perturbation methods.
The final theoretical work of this thesis will consist in the developments of adaptive remeshing
techniques. In here, different approaches will be presented. Starting with the metric based techniques,
including the level-set and Hessian approaches. These techniques are of general purpose, and can
be considered both in application of structural or Computational Fluid Dynamics (CFD) problems. In
addition to this approach, the Super convergent Patch Recovery (SPR) error estimation method is
presented. This approach is more conventional than the former ones. In both cases, in the Hessian and
the SPR error estimation, have been extended in order to apply it to contact mechanics problems, main
contribution of this work in this field.
With all the former developments, we will be ready for the introduction of practical cases focused in
the context stamping processes. The most important point to highlight on this is that these examples will
be compared with the reference solutions available in the literature as a validation of the developments
presented until this point.
The present document is organised as follows. The first chapter introduces the thesis, stating the
objectives and reviews the state-of-the-art on the most transcendental subjects. The second chapter
does a shallow introduction of the continuous mechanics and Finite Element Method (FEM) concepts,
required for following chapters. The third chapter derives the formulation of the solid-shell previously
mentioned and the free-rotation shell, which is the theoretical inspiration of the solid-shell presented,
including several academic examples that are commonly employed in the literature as a benchmark of
shell elements. The fourth chapter will present the contact mechanics formulation developed, consisting
in an implicit mortar ALM formulation, the chapter also includes several examples commonly found in the
literature, which are usually considered for validation. The next chapter presents the plasticity formulation
used, including some technical details in the implementation side, some examples of validation will be
also introduced at the end of this chapter. The final theoretical chapter shows the adaptive remeshing
algorithms developed in the context of this work, and presents several examples, including not only solid
mechanics cases, but also CFD. The next chapter encapsules some validation and application cases for
stamping processes. The final chapter comprises the conclusions, as well as the future works coming
after the presented developments.

Keywords: metal-forming, stamping, shells, solid-shells, contact, plasticity,adaptive remeshing

Page 6 of 374 Vicente Mataix Ferrándiz

 Resumen
La tesis doctoral Modelos matemáticos y numéricos innovadores para el estudio de la deformación de
láminas durante los procesos de conformado industrial por el Método de los Elementos Finitos. pretende
contribuir al desarrollo de métodos de elementos finitos para el análisis de procesos de estampado, un
área problemática con una clara aplicación industrial. De hecho, este tipo de problemas multidisciplinares
requieren el conocimiento de múltiples disciplinas, como la mecánica de medios continuos, la plasticidad,
la termodinámica y los problemas de contacto, entre otros.
Para alcanzar los objetivos propuestos, la primera parte de esta tesis abarca los elementos de sólido-
lámina. Este tipo de elemento resulta atractivo para la simulación de procesos de conformado, dado que
cualquier tipo de ley constitutiva tridimensional puede ser formulada sin necesidad de considerar ninguna
conjetura adicional. Además, este tipo de elementos permite realizar una descripción tridimensional del
cuerpo deformable, por tanto, el contacto de ambas caras puede ser tratado fácilmente.
Este trabajo presenta en primer lugar el desarrollo de un elemento de sólido-lámina prismático
triangular, para el análisis de láminas gruesas y delgadas con capacidad para grandes deformaciones.
Este elemento figura en formulación Lagrangiana total, y emplea los elementos vecinos para poder
computar un campo de desplazamientos cuadráticos. En la formulación original, se obtenía un tensor de
Cauchy derecho modificado (C̄); sin embargo, en este trabajo, la formulación se extiende obteniendo un
gradiente de deformación modificado (F̄), que permite emplear los conceptos de push-forward y pull-back.
Dichos conceptos proveen de un método matemáticamente consistente para la definición de derivadas
temporales de tensores y, por tanto, puede ser usado, por ejemplo, para trabajar con elasto-plasticidad.
El elemento se basa en tres modificaciones: (a) una aproximación clásica de deformaciones transver-
sales de corte mixtas impuestas; (b) una aproximación de deformaciones impuestas para las componentes
en el plano tangente de la lámina; y (c) una aproximación de deformaciones impuestas mejoradas en la
dirección normal a través del espesor, mediante la consideración de un grado de libertad adicional. Los
objetivos son poder utilizar el elemento para la simulación de láminas sin bloquear por cortante, mejorar el
comportamiento membranal del elemento en el plano tangente, eliminar el bloqueo por efecto Poisson y
poder tratar materiales elasto-plásticos con un flujo plástico incompresible, así como materiales elásticos
cuasi-incompresibles o materiales con flujo plástico isocórico. El elemento considera un único punto de
Gauss en el plano, mientras que permite considerar un número cualquiera de puntos de integración en
su eje, con el objetivo de poder considerar problemas con una significativa no linealidad en cuanto a
plasticidad.
Este trabajo continúa con el desarrollo de la formulación de contacto empleada, una metodología que
se encuentra en la bibliografía sobre la mecánica de contacto computacional para simulaciones implícitas.
Dicha formulación consiste en una integración exacta de la interfaz de contacto mediante métodos de
mortero, lo que permite obtener la integración más consistente posible entre los dominios de integración,
así como la solución más exacta posible. La implementación también considera varios algoritmos de
optimización, como la optimización mediante penalización. La contribución más notable de este trabajo es
la consideración de multiplicadores de Lagrange aumentados duales como método de optimización. Estos
permiten condensar estáticamente el sistema de ecuaciones, lo que permite eliminar los multiplicadores
de Lagrange de la resolución y, por lo tanto, permite la consideración de solvers iterativos. Además, la
formulación ha sido adecuadamente linealizada, asegurando la convergencia cuadrática del problema.
Para resolver el sistema de ecuaciones, se considera un método de Newton semi-smooth, que consiste
en una estrategia de set activo, extensible también en el caso de problemas friccionales. La formulación
es funcional tanto para problemas sin fricción como para problemas friccionales, lo que es esencial para
la simulación de procesos de estampado. Esta formulación friccional se enmarca en los modelos de
fricción tradicionales, como la fricción de Coulomb, pero el desarrollo presentado puede extenderse
a cualquier tipo de modelo de fricción. Esta formulación de contacto es totalmente compatible con el
elemento sólido-lámina introducido en este trabajo.
El componente necesario restante para la simulación de procesos industriales son los modelos
constitutivos. En este trabajo, esto se ve materializado en la formulación de plasticidad considerada.
Estos modelos constitutivos se considerarán modelos de plasticidad para grandes deformaciones, con
una combinación arbitraria de superficies de fluencia y potenciales plásticos: los llamados modelos no
asociados. Para calcular el tensor tangente correspondiente a estas leyes generales, se han considerado
implementaciones numéricas basadas en métodos de perturbación.

Vicente Mataix Ferrándiz Page 7 of 374

 Otra contribución fundamental de este trabajo es el desarrollo de técnicas para el remallado adaptativo,
de las que se presentarán distintos enfoques. Por un lado, las técnicas basadas en métricas, incluyendo
los enfoques level-set y Hessiano. Estas técnicas son de propósito general y pueden considerarse tanto
en la aplicación de problemas estructurales como en problemas de mecánica de fluidos. Por otro lado, se
presenta el método de estimación de errores SPR, más convencional que los anteriores. En este ámbito,
la contribución de este trabajo consiste en la estimación de error mediante las técnicas de Hessiano y
SPR para la aplicación a problemas de contacto numérico. Con los desarrollos previamente introducidos,
estaremos en disposición de introducir los casos de aplicación centrados en el contexto de procesos de
estampado. Es relevante destacar que estos ejemplos son comparados con las soluciones de referencia
disponibles en la bibliografía como forma de validar los desarrollos presentados hasta este punto.
El presente documento está organizado de la siguiente manera. El primer capítulo establece los
objetivos y revisa la bibliografía acerca de los temas clave de este trabajo. El segundo capítulo hace una
introducción de la mecánica de medios continuos y los conceptos relativos al Método de los Elementos
Finitos (MEF), necesarios en los desarrollos que se presentarán en los capítulos siguientes. El tercer
capítulo aborda la formulación del elemento sólido-lámina, así como del elemento de lámina sin grados
de libertad de rotación que inspira el sólido-lámina desarrollado. Esta parte muestra varios ejemplos
académicos que son comúnmente empleados en la bibliografía como problemas de referencia de láminas.
El cuarto capítulo presenta la formulación desarrollada para la resolución de problemas de contacto
numérico, consistente en una formulación implícita de integración exacta mediante métodos mortero
y multiplicadores de Lagrange aumentados duales. Este capítulo incluye, asimismo, varios ejemplos
comúnmente encontrados en la bibliografía, que generalmente son considerados para su validación. El
quinto capítulo presenta la formulación de plasticidad empleada, incluyendo algunos detalles técnicos
desde el punto de vista de la implementación, así como varios ejemplos de validación. El sexto capítulo
muestra los algoritmos de remallado adaptativo desarrollados en el contexto de este trabajo, y presenta
varios ejemplos, que incluyen no solo casos estructurales, sino también de mecánica de fluidos. El
séptimo capítulo encapsula algunos casos de validación y aplicación para procesos de estampado. El
capítulo final comprende las conclusiones, así como los trabajos que podrían continuar el presente
estudio.

Palabras clave: Conformado metálico, estampado, láminas, sólido-lámina, contacto, plasticidad,

mallado adaptativo

Page 8 of 374 Vicente Mataix Ferrándiz

 Acknowledgements

I still remember the day I arrived to Barcelona 5 years ago to start my master thesis in FSI. My car, that I was supposed
to take in order to bring all my stuff from Paris to Barcelona, just burn out the week before and I needed to take the
TGV instead, with all my stuff. It is because of that I want to acknowledge in first place Katie, one of my roomies when
I arrived to Barcelona, who helped me that first day that I arrived to Barcelona to settle.
The next person I meet and I want to thank is my director Eugenio Oñate Ibañez de Navarra, who brings me the
opportunity to work in this thesis, and had enough patience to deal with all the delays and problems that I had during
all this time. Especially the patience to deal with my continuous trips between Barcelona and Paris. I hope the result
obtained in this work was worth it. I would like to have a special acknowledge to Merce Alberich, who helped me with
a lot of paper work, and to communicate with Eugenio. In the same line, I would like to notice the administratives in
the doctoral school, Silvia and Rosa, who help me with all the documents related with this thesis. I want to mention
the support of the Generalitat de Catalunya which is funding my PhD with the grant FI AGAUR.
I want to especially thank to my co-director Riccardo Rossi for all the help and support during all this time.
Particularly in the last year of the thesis, where the relationship became more difficult due to the distance. At the
end I hope you accepted volcanic personality. I would like to thank to all the Kratos staff. Special remarks for Carlos
Roig, aka Charlie for help me with all those programmation questions. To Jordi Cotela, who helped me in several
technical issues too, and at the end I hope understood my sense of humour. To Antonia Larese, who gave me several
academic tips. Ilaria Iaconeta, who was a good friend and understands the pain of having a thesis finished, but not
really. And to Pooyan Dadvand for all the technical help and Kratos origin history. Thanks to all the team, to the
prepost duo Enrique Escolano-Miguel Pasenau, to the fantastic duo Anna-Adrià, to the interfacer Javier Garate and to
the doctor mesh Abel Coll. Without you, all this work it would have been much more difficult.
Thanks to everyone I meet these years in Barcelona at the International Center for Numerical Methods in
Engineering (CIMNE), especially I want to thank to Rubén Zorrilla who was my office partner, and we have together
so much metahumour, which is the best type of humour. To the rookies that I left in the office, Riccardo Tosi, Marc
Nuñez and Agustina Giuliodori. To Miguel Ángel Celigueta and Lucia Barbu, they are an adorable couple and I wish
you a lot of happiness together. To Eduardo Soudah, which passion for biomechanics is fascinating. To Guillermo
Casas, with whom I have had the pleasure to share debates about politics from time to time. Miguel Maso, who
is a wonderful person, and I wish him the best. To Pablo Becker and Marcelo Raschi, with whom I have enjoyed
technical discussions. Ignasi Pouplana, who was the first person I meet in CIMNE, and who helped me so much at
the beginning, and with whom I enjoyed working at the end. To Lorenzo Gracia, who gave me such good times, and
enjoyed so many paellas of mine. To Tomás Varona whose friendship I value so much and that I hope will last for
many years. Finally, to Alejandro Cornejo, who helped me so much during the last stages of the thesis, but has also
turned out to be one of my best friends. From Barcelona, I also like to address to Miguel Zatarain, who helped me to
break routine when I was living in Barcelona, and who host me when I was not living there anymore.
To the people from Munich, which contribute so much to the StructuralMechanicsApplication, and
therefore help me directly and indirectly with the developments presented on this work. Remarks for Philipp Bucher,
who controlled a very relevant part of my developments. And to Tobias Teschemacher who is the coolest guy I know,
and I had a great time with him. I would like to mention to Roland Wüchner, who did everything in order I could go to ,
but unfortunately I could not go at the end. Deeply thanks to Mohamed Khalil, who accompanied me during my first
steps in the world of numerical contact. Without you probably this thesis would be right now completely different. Also
coming from , thanks to Elisa Magliozzi who ended being a very good friend, and to Anna Rehr, which master thesis I
had the pleasure to supervise. From Inria I would like to acknowledge to Algiane Froehly, thanks to her, and to the
Mmg team, I could accomplish a very relevant part of my thesis.
Finally, but not for that reason less important, I want to thank to my family, to my father who has proven once
again to have patience with me and my "obsessive perfectionism", to my mother who has given me so much support,
and has suffered so much for me, and to my sister, that we have both had to leave the "terreta" in order to dedicate
ourselves to research. But I want to thank especially to my wife Sara, who has been the co-pilot of this work from
the distance, and in the last stage together, she knows everything I have had to go through and do to get here, and
sometimes she has had to suffer because of this.
Page 10 of 374 Vicente Mataix Ferrándiz
CONTENTS CONTENTS

Contents

I Theoretical framework 21
1 Introduction 23
1.1 Evolution of the stamping technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.1.1 History of the stamping and forming technology . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.1.2 Historical review of Finite Element Analysis (FEA) of formability and forming technology . . . 27
1.2 Most common issues in forming processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.2.1 Springback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.2.2 Wrinkling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Finite Element formulation 33

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.2 Concept of Finite Element (FE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Solid mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.1 Linear form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.2 Non-linear form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.2.1 Deformation gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.2.2 Strain tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4 Finite Element formulation for Computational Structural Dynamics (CSD) . . . . . . . . . . . . . . 38
2.4.1 Weak formulation in solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.2 TL and Updated Lagrangian (UL) formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.3 Solution of the non-linear-equilibrium equations system . . . . . . . . . . . . . . . . . . . . . . 40
2.4.3.1 Newton-Raphson method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.3.2 Line-search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4.3.3 Arc-length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4.4 Time integration schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.4.1 Newmark time scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.4.2 Bossak algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.4.3 α-generalised . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.4.4 Backward Differentiation Formula (BDF) . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.5 Damping matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3 Rotation-free shells and solid-shell elements 49

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.1 Historical outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2 State of the Art in shell formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2.1 Element requisites and stabilisation techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 50

Vicente Mataix Ferrándiz Page 11 of 374

CONTENTS CONTENTS

3.2.2 Shell formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.3 Rotation-free shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3.2 Basic Shell Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3.2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3.2.2 Shell kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.2.3 Constitutive models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.3 Enhanced Basic Shell Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.3.1 Total Lagrangian formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.3.1.1 Computation of the membrane strains . . . . . . . . . . . . . . . . . . . . . 54
3.3.3.1.2 Computation of the bending strains . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.3.1.3 Tangent stiffness matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4 Prismatic solid-shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4.2 Basic kinematics of the standard element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4.3 Modifications of the standard element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.3.1 In plane behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.3.2 Transverse shear behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4.3.3 Transverse normal behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4.3.3.1 Enhanced Assumed Strain (EAS) formulation . . . . . . . . . . . . . . . . 61
3.4.3.3.2 Balance equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.4.3.3.3 Pull-Back and Push-Forward (Extension of the formulation) . . . . . . . . . . 62
3.5 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.1 Patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.1.1 Membrane patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.1.2 Bending patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5.2 Cantilever . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.5.3 Cantilever subjected to end bending moment . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.4 Frequencies test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.5.5 Cook membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.5.6 Scoordelis cylindrical roof test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.5.7 Sphere test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.5.8 Pull-out of an open-ended cylindrical shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.5.9 Pinched semi-cylindrical . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.5.10 Slit test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.5.11 Cylindrical panel test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.5.12 Conical shell test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.5.13 Wrinkling test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.5.14 Fluid Structure Interaction (FSI)-Vein test . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4 Contact mechanics 85
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.1.1 Historical outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.1.2 Contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.2 State of the Art in computational contact mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.2.2 Discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.2.2.1 Node-To-Node (NTN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2.2.2 Node-To-Segment (NTS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2.2.3 Contact Domain Method (CDM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.2.4 Segment-To-Segment (STS) (Mortar methods) . . . . . . . . . . . . . . . . . . . . 91
4.2.2.5 Other alternative methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.2.2.5.1 Isogeometric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.2.2.5.2 Smooth surface approximation . . . . . . . . . . . . . . . . . . . . . . . . . 93

Page 12 of 374 Vicente Mataix Ferrándiz

CONTENTS CONTENTS

4.2.2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.3 Optimisation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.3.1 Penalty Method (Penalty Method (PM)) . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.3.2 Lagrange Multiplier Method (LMM) . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2.3.3 ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2.3.4 Double Lagrange Multiplier Method (DLMM) . . . . . . . . . . . . . . . . . . . . . 94
4.2.3.5 Other alternative methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.3.5.1 Perturbed Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.3.5.2 Nitsche . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.3.5.3 Other minor mentions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.5 Frictional models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3.2 Definition of the problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3.3 Frictionless contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3.3.1 Strong formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3.3.2 Weak formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.3.3.2.1 Scalar Lagrange multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.3.3.2.1.1 LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.3.3.2.1.2 Penalty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3.3.2.1.3 ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3.3.2.2 Vector Lagrange multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3.3.2.2.1 LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3.3.2.2.2 ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3.3.3 Augmented Lagrange multiplier parameters calibration . . . . . . . . . . . . . . . . . 102
4.3.3.4 Discretisation and numerical integration . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.3.3.4.1 Dual Lagrange multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.3.3.4.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.3.3.4.1.2 Graphical representation . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.3.3.4.1.3 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.3.3.4.2 Mortar operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.3.3.4.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.3.3.4.2.2 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3.3.4.3 Algebraic form of the problem . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3.3.4.3.1 Scalar LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3.3.4.3.2 Components LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3.3.4.3.3 Penalty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.3.4.3.4 Scalar ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.3.4.3.5 Components ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.3.4.4 Static condensation of the system in considering of the DLMM . . . . . . . . 110
4.3.3.5 Work-flow. Solution algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.3.3.6 Active set strategy (Semi-smooth Newton Raphson) . . . . . . . . . . . . . . . . . . 112
4.3.4 Frictional contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3.4.1 Strong formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3.4.1.1 Tangential contact condition - Coulomb’s law . . . . . . . . . . . . . . . . . 113
4.3.4.2 Weak formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.3.4.2.1 LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.3.4.2.2 Contact constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.3.4.2.3 Penalty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.3.4.2.4 ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.3.4.3 Discretisation and numerical integration . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3.4.3.1 Discrete contact condition in tangential direction . . . . . . . . . . . . . . . . 118
4.3.4.3.2 Slip definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3.4.3.3 Algebraic form of the problem . . . . . . . . . . . . . . . . . . . . . . . . . . 120

Vicente Mataix Ferrándiz Page 13 of 374

CONTENTS CONTENTS

4.3.4.3.3.1 LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

4.3.4.3.3.2 Penalty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.3.4.3.3.3 ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.3.4.4 Work-flow. Solution algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.3.4.5 Active set strategy (Semi-smooth Newton Raphson) . . . . . . . . . . . . . . . . . . 123
4.4 Contact detection. Search techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.4.2 Bounding volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.4.2.1 Oriented Bounding Box (OBB) implementation . . . . . . . . . . . . . . . . . . . . 124
4.4.2.1.1 Collision detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.4.2.1.2 Collision detection with Separating Axis Theorem (SAT) . . . . . . . . . . 127
4.4.3 Tree structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.4.4 Penetration definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.4.5 Self-contact detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4.5.2 Algorithm considered . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.4.5.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4.5.3.1 Planes detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4.5.3.2 Tubular detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4.5.3.3 Contact example (S-shape profile) . . . . . . . . . . . . . . . . . . . . . . . 134
4.5 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.5.1 Basic patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.5.1.1 Frictionless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.5.1.2 Frictional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.5.2 Taylor patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.5.2.1 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.5.2.2 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.5.3 Friction base test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.5.4 Hertz problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.5.4.1 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.5.4.1.1 Plane-sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.5.4.1.2 Cylinder-cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.5.4.1.2.1 Frictionless case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.5.4.1.2.2 Frictional case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.4.2 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.4.2.1 Plane-sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.4.2.2 Sphere-sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.5.5 Teeth model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.5.6 Energy conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.5.7 Double arc benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.5.7.1 Frictionless . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.5.7.2 Frictional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.5.8 Arc pressing block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.5.9 Hyperelastic tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.5.10 Contacting cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.5.10.1 Horizontal movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.5.10.2 Vertical movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.5.11 Press fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.5.12 Ironing punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.5.12.1 Circular ironing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.5.12.2 Shallow ironing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.6 Derivatives for contact mechanics linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.6.1 Derivatives for 2D contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.6.1.1 Jacobians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.6.1.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

Page 14 of 374 Vicente Mataix Ferrándiz

CONTENTS CONTENTS

4.6.1.1.2 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

4.6.1.2 Shape functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.6.1.2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.6.1.2.2 Integration segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.6.1.2.3 Local coordinates (Gauss points) . . . . . . . . . . . . . . . . . . . . . . . . 165
4.6.1.2.4 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.6.1.3 Dual shape functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.6.1.3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.6.1.3.2 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.6.1.4 Normal and tangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
4.6.1.4.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
4.6.1.4.2 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.6.2 Derivatives for 3D contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.6.2.1 Jacobians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.6.2.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.6.2.1.1.1 Integration segments derivatives . . . . . . . . . . . . . . . . . . . . 170
4.6.2.1.1.2 Jacobian derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.6.2.1.2 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.6.2.2 Shape functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.6.2.2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.6.2.2.2 Local coordinates (Gauss points) . . . . . . . . . . . . . . . . . . . . . . . . 174
4.6.2.2.3 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.6.2.3 Dual shape functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.6.2.3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.6.2.3.2 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.6.2.4 Normal and tangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.6.2.4.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.6.2.4.2 Convergence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

5 Plasticity 189
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.1.1 Historical outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.2 State of the Art in numerical plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.2.1 Historical outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.2.2 Constitutive models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.2.2.1 Yield criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.2.2.2 Yield surface hardening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.3 Finite strain elasto-plastic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.3.1 Fundaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
5.3.2 Plastic loading surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.3.2.2 Isotropic hardening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.3.2.3 Kinematic hardening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.3.2.4 Stress-strain relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.3.2.5 Stability condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.3.2.5.1 Local stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.3.2.5.2 Global stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.3.2.6 Condition of unicity of solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.3.3 Finite strain return mapping operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.4 Large deformation elastic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.4.2 Neo-Hookean material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.4.3 Kirchhoff material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.5 Implementation details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

Vicente Mataix Ferrándiz Page 15 of 374

CONTENTS CONTENTS

5.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

5.5.2 Class structure in Kratos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.5.3 Numerical implementation tangent constitutive tensor . . . . . . . . . . . . . . . . . . . . . . . 200
5.5.3.1 Forward Finite Differences (FD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.5.3.2 Centered FD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.5.3.3 Numerical details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.6 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.6.2 Cube minimal example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.6.3 Tensile test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.6.4 Application in Computational Contact Mechanics (CCM). Gears example . . . . . . . . . . . 204
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

6 Adaptative remeshing 211

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
6.2 State of the Art in mesh refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
6.2.1 Mesh generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
6.2.2 Adaptive finite element refinement techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
6.3 Hessian based remeshing technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6.3.1 Error estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6.3.1.1 Upper bound on the interpolation error . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6.3.1.2 Numerical computation of the interpolation error . . . . . . . . . . . . . . . . . . . . 216
6.3.1.3 Mesh adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.3.1.4 Computation of the relative error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.3.2 Metric based remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
6.3.2.1 Concept of metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
6.3.2.2 Metric intersection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.3.3 Hessian based metric measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.3.3.1 Hessian theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.3.3.2 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.4 Level set based remeshing technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.4.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.4.2 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.5 SPR based remeshing technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.5.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.5.2.1 Error measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.5.2.2 Error estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.5.2.3 Recovery based methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
6.5.2.4 Superconvergence concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
6.5.2.5 SPR calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.5.2.6 Local mesh size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
6.5.2.6.1 Compute h . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
6.5.2.6.2 Compute respective metric . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.5.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.6 Internal values interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.6.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.6.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.7 Integration points values extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
6.7.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
6.7.3 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.8 Adaptive remeshing methods applied on CCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.8.1 Level set metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
6.8.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

Page 16 of 374 Vicente Mataix Ferrándiz

CONTENTS CONTENTS

6.8.1.2 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

6.8.2 Hessian metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
6.8.2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
6.8.2.2 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
6.8.2.2.1 Simplest patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
6.8.2.2.2 Punch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
6.8.3 SPR metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6.8.3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6.8.3.2 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.8.3.2.1 Simplest patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.8.3.2.2 Punch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.9 Remeshing workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
6.9.1 Level set and Hessian remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
6.9.1.1 Standard problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.9.1.2 Contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.9.2 SPR remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.9.2.1 Standard problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.9.2.2 Contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.10 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.10.1 Coarse sphere. Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.10.2 Beam problem. Hessian of displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.10.3 Hertz problem. Hessian of contact and Von Mises (VM) stress . . . . . . . . . . . . . . . . . 243
6.10.4 Hertz problem. SPR error computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.10.5 Contacting cylinders with adaptive remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.11 CFD numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.11.1 Cavity. Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.11.2 Embedded cylinder. Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.11.3 Box isosurface example. Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
6.11.4 Lamborghini. Level set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.11.5 Channel CFD. Hessian of velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.11.5.1 2D cylinder channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
6.11.5.2 3D sphere channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254

II Application 259
7 Application 261
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
7.2 Cylinder punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
7.3 Spherical punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264

III Conclusions 267

8 Final conclusions 269
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
8.2 Rotation free shells and solid-shell elements. Solid-shell . . . . . . . . . . . . . . . . . . . . . . . . . 269
8.3 Contact mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
8.4 Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
8.5 Adaptive remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
8.6 Application cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
8.7 Other remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
8.7.1 Other developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
8.7.2 Implicit approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

Vicente Mataix Ferrándiz Page 17 of 374

CONTENTS CONTENTS

8.7.3 Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

9 Future works 277

9.1 Solid-shell implementations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.2 Contact mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.3 Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
9.4 Adaptive remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
9.5 Metal stamping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
9.6 Other points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278

Appendices 281

Appendices

A Theoretical complements 285

A.1 Pull-Back, Push-Forward fundamental concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
A.2 Mortar segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
A.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
A.2.2 Exact integration vs Collocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
A.2.2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
A.2.2.2 Solution study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
A.2.3 Delaunay vs Convex polygon construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
A.3 Mesh tying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
A.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
A.3.2 Strong formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
A.3.3 Weak formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
A.3.4 Discretisation and numerical integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
A.3.5 Numerical example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

B Implementation 295
B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
B.2 Kratos Multiphysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
B.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
B.2.2 Framework (Application Programming Interface (API)) . . . . . . . . . . . . . . . . . . . . . 296
B.2.3 Introduction. Main classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
B.2.4 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
B.2.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
B.2.4.2 ModelPart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
B.2.4.3 Submodelparts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
B.2.5 Base classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.2.6 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.2.7 Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.2.8 Input-Output (IO) classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.2.9 Testing framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.3 Mmg library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.3.1 What is Mmg and how does it work? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
B.3.2 Integration between Mmg and Kratos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
B.3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
B.3.2.2 Class structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
B.3.2.3 Submodelpart recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302

Page 18 of 374 Vicente Mataix Ferrándiz

CONTENTS CONTENTS

C Automatic differentiation 305

C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
C.2 Mathematical concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
C.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
C.3.1 Principles of Automatic Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
C.3.1.1 Derivation modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
C.3.1.2 Automatic Differentiation exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
C.3.2 Kratos integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310

D Constrained optimisation problems 313

D.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
D.2 Penalty method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
D.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
D.2.2 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
D.2.3 Applicability on contact problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
D.2.3.1 Adapted Penalty Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
D.3 Lagrange Multiplier method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
D.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
D.3.2 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
D.3.3 Applicability on contact problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
D.4 Augmented Lagrange Multiplier method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
D.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
D.4.2 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
D.4.2.1 Standard formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
D.4.2.2 Uzawa iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
D.4.3 Applicability on contact problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
D.4.3.1 Adapted Augmented Lagrangian Method . . . . . . . . . . . . . . . . . . . . . . . . 320
D.5 MultiPoint Constraint (Master-Slave elimination method) . . . . . . . . . . . . . . . . . . . . . . . . . 320
D.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
D.5.2 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
D.5.3 Applicability on contact problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
D.6 Summary of the different methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
D.7 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
D.7.1 Initial spring wall problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
D.7.1.1 Penalty method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
D.7.1.2 Lagrange multiplier method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
D.7.2 Non-linear spring contact problem with wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
D.7.2.1 Initial wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
D.7.2.1.1 Lagrange multiplier method . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
D.7.2.1.1.1 Active set 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
D.7.2.1.1.2 Active set 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
D.7.2.1.1.3 Active set 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
D.7.2.2 Moved wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
D.7.2.2.1 Penalty method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
D.7.2.2.2 Lagrange multiplier method . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
D.7.2.2.2.1 Active set 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
D.7.2.2.2.2 Active set 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
D.7.2.2.2.3 Active set 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
D.7.2.2.3 Augmented Lagrange multiplier method . . . . . . . . . . . . . . . . . . . . 331
D.7.2.2.3.1 ε = 0.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
D.7.2.2.3.2 ε = 1.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
D.7.2.2.3.3 ε = 5.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
D.7.2.2.3.4 ε = 10.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
D.7.3 Over-constrained optimisation problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334

Vicente Mataix Ferrándiz Page 19 of 374

CONTENTS CONTENTS

Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336

E Mortar mapper 339

E.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
E.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
E.2.1 General mapping theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
E.2.2 Dual Lagrange multiplier mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
E.2.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
E.2.2.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
E.2.3 Discontinuous meshes mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
E.2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
E.2.3.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
E.3 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
E.3.1 Non-matching meshes of triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
E.3.2 Non-matching meshes of triangles and quadrilaterals . . . . . . . . . . . . . . . . . . . . . . . 344
E.3.3 Discontinuous meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
Bibliography of chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346

List of Tables 349

List of Figures 351

List of Algorithms 357

Glossaries 359
Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Mathematical symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
Rotation-free shell and solid-shell elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
Constraint enforcement and optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Finite element framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Other terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
Mortar formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
Contact mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Adaptive remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
Plasticity and constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373

Page 20 of 374 Vicente Mataix Ferrándiz

 Part I

Theoretical framework

Vicente Mataix Ferrándiz Page 21 of 374

CHAPTER 1. INTRODUCTION

Chapter 1

Introduction

Figure 1.1: Examples of numerical simulation of forming. Source[OnlShe]

Stamping is a kind of manufacturing technology[BookAT12b; BookHu+13], where a sheet blank that has a simple
shape is plastically formed between tools or dies, to obtain product components with certain shape, size, and
performance (see Figure 1.1). Sheet metal, mould and stamping equipment are three major factors for stamping.
Indeed, sheet metal forming processes usually produce little scraps and generate the final part geometry in a very
short time, usually in one stroke or a few strokes of a press. As a result, sheet forming offers potential savings in
energy and material, especially in medium and large production quantities, where tool costs can be easily amortised.
We can differentiate mainly two different technologies according to the working temperature, the hot stamping and the
cold stamping; the first one is suitable to process a kind of sheet which has high resistance to deformation and low
plasticity, while the second one is employed for metal sheet stamping processes at room temperature.
The concept of formability[BookBan10] can be introduced, which is the capability of sheet metal to undergo plastic
deformation to a given shape without defects, and thus is one of the most relevant fields of study for the stamping
processes. The defects have to be considered separately for the fundamental sheet metal forming procedures of
deep drawing and stretching. The difference between these types of stamping procedures is based on the mechanics
of the forming process. Figure 1.2 illustrates that formability is a complex characteristic.
The increase along the time of the costs of material, energy and manpower require that sheet metal forming
processes and tooling be projected and developed avoiding as much as possible the trial and error methodology, with
the shortest possible lead times. Owing to this reason, the concept of virtual manufacturing has been developed in
order to increase the industrial performances[BookBan10], being the most efficient way to reduce manufacturing times
and improving the final quality of the products. The finite element method is currently the most widely used numerical
procedure for simulating sheet metal forming processes. The structure of an expert system for the analysis of sheet
metal formability is illustrated by Figure 1.3. For the expert system[OnlShe] two broad divisions of methodologies can

Vicente Mataix Ferrándiz Page 23 of 374

 CHAPTER 1. INTRODUCTION

Figure 1.2: Parameters influencing sheet metal formability. Inspired in[BookBan10]

be applied:

• Inverse One-step: In this approach all the deformation is assumed to happen in one increment or step and is
the inverse of the process which the simulation is meant to represent. The mesh initially is considered with the
shape and material characteristics of the finished geometry, and is deformed to the flat pattern blank. Then the
strain computed in this inverse forming operation is then inverted to predict the deformation potential of the flat
blank being deformed into the final part shape.

• Incremental analysis: This method starts with the mesh of the flat blank and simulate the deformation of the
blank inside of tools modelled to represent a proposed manufacturing process. This incremental forming is
computed "forward" from initial shape to finals, and is calculated over a number of time increments for start to
finish.

As the Incremental analysis includes the model of the tooling and allows for the definition of boundary conditions
which more fully replicate the manufacturing proposal, incremental methods are more commonly used for process
validation; in the other hand, Inverse One-step with its lack of tooling and therefore poor representation of process is
limited to geometry based feasibility checks.
The increasing acceptance of these numerical approaches[BookBHS07] within both research and industrial
environments is due to improved awareness, enhanced maturity of computational models and associated algorithms
and, more importantly, dramatic increases in computational power/cost ratios, due to this the cost of the simulations
has dropped drastically in the last years.
We can highlight the fact that during the last years a lot of significant advances have been archived in the
development of finite element tools for the simulation of forming processes. Now the use of commercial codes is widely
used for the design and optimisation of these processes, particularly in highly competitive sectors of manufacturing
industry. The widespread acceptance of such methods stems in their capabilities to predict deformed shapes,
manufacturing defects, forces involved in the process, residual stresses and inelastic strains. Such predictions, which
can be usually obtained through relatively inexpensive numerical simulations, are of considerable assistance to the
designer of forming operations, allowing to optimise the process parameters and manufactured product properties as
well as significantly reducing conception-to-production times. Despite such advances, there are still many complex
problems to solve, some of these problems are related directly with the complexity of the manufacturing process; but
from a mechanical point of view forming problems are complex due to fact they join all the origins of non-linearity in
the mechanical problems, which are[BookWri08]:

Page 24 of 374 Vicente Mataix Ferrándiz

CHAPTER 1. INTRODUCTION

Figure 1.3: Structure of an expert system for the analysis of sheet metal formability. Inspired in [BookBan10]

• Geometrical nonlinearity: Large displacements and rotations have to be considered.

• Finite deformations: Not only the displacements are large, but also the strains.
• Physical nonlinearity: The non linear behaviour of the material, like in the case plasticity, essential to work in
forming simulations.

• Stability problems: Geometrical and material instability. Geometrical instability includes bifurcations like
buckling, or any snap-through behaviour. The material instability could be because of a instability in the
equilibrium equations.

• Nonlinear boundary conditions: Contact between two bodies or deformation dependent of the loading,
phenomena very frequent in the stamping processes.

• Coupled problems: Like thermomechanical problems, which are very frequent in forming processes or FSI,
which appears for example in the hydroforming manufacturing processes.

For the practical[BookAT12b] and efficient use of these technologies, the knowledge of the principal variables of
the sheet metal forming processes and their interactions is required. These variables include:

1. The flow behaviour and formability of the formed sheet material under processing conditions

2. Die geometry, materials and coatings

3. Friction and lubrication

4. The mechanics of deformation, i.e. strains, stresses and forces

Vicente Mataix Ferrándiz Page 25 of 374

1.1 Evolution of the stamping technology CHAPTER 1. INTRODUCTION

5. Characteristics of the sheet metal forming presses and tooling

6. Geometry, tolerances, surface finish and mechanical properties of the formed parts

7. The effects of the process on the environment

1.1 Evolution of the stamping technology

1.1.1 History of the stamping and forming technology
Stamping[BookHu+13] is a kind of metal processing method that has already existed in ancient times. The ancient
struck preformed metal blocks with hammers. Lately, 4000 years ago, in ancient China[ArtKC06] metal plastic
work can be traced back, mainly forging as a processing method. Today, sheet metal forming can be considered a
fully automatic mass production industry. It applications extend from household commodities like pans or cans, to
automotive and aeronautical industry, among other applications. We can consider the rise of the modern stamping
era 200 years ago, during the Industrial Revolution in United Kingdom (UK). During these years, the development of
modern iron technology contributed to the development of stamping procedures and machinery.

The car industry was probably the main responsible of the development of the stamping technology when the price
became a problem in the process of car popularisation. Thus using stamping to produce automotive parts became
popular because of its contribution to lower the production cost; nowadays about 60-65% of the car parts are made by
stamping.

We can enumerate the five stages that the stamping technology has experienced in developed countries[BookHu+13]
during the 20th century as follows:

1. Before the Second World War (WWII), the stamping line was composed of manually loaded double-action
drawing press and several single-action drawing presses. Being manually feed carried as result:

• Low production efficiency.

• Poor security conditions.
• Poor product quality.

2. During the 1960s, still the same technology was considered, but each press was equipped with a robot hand in
order to reduce manual labour.

3. In the 1970s, the automatic stamping line was established and could be controlled only by one or two operators.
This increased productivity, as well as lowered costs and improved quality.

4. Later, in the 1980s, owing to the emergence of multistation presses, a double-action drawing press and a
multi-station press were used for composing a stamping line.

5. Since the 1990s, the double-action drawing press was replaced as the leading equipment of stamping lines.
To replace it, Numerical Controlled (NC) hydraulic cushion was installed into the first station of the large
multi-station press, which made this press form a flexible production unit independently.

Moving forward to the present, the automotive industry is the pillar industry of the national economy in many indus-
trially developed countries or newly industrialising countries. For this reason the development of numerical methods
and design technologies has been crucial; where Computer-Aided Design (CAD)/Computer Aided Engineering
(CAE)/Computer Aided Manufacturing (CAM) integration technology is playing an increasingly important role in
product design, mould design and manufacturing process.

Page 26 of 374 Vicente Mataix Ferrándiz

CHAPTER 1. INTRODUCTION 1.2 Most common issues in forming processes

1.1.2 Historical review of FEA of formability and forming technology

FEM[BookBHS07; BookBan10] has been developed and used for sheet forming simulations since the 1960s and
early 1970s, when the continuum mechanics foundations for problems involving large displacements and large strains
became well established. The establishment of 3D Non-Linear (NL) shell theory made it possible to analyse the
forming process of some complex parts such as car bodies cover, and the algorithm for contact and friction problems
took geometric nonlinearity into consideration.
It was at the beginning of the 1970s when the first theoretically correct FE-formulation of the large deformation
problem was presented by Hibbit et al.[ArtHMR70] in a TL frame. Five years later, in 1975, McMeeking[ArtMR75],
Rice and Bathe[ArtBRW75] presented correct UL formulations of the problem. At that time the simulations were
based in plane strain or axisymmetric formulations, and at that time sometimes the models lack full consistency
from a theoretical point of view. Later the 3D formulation of the sheet forming problem was presented by Wang and
Budiansky in 1978[ArtWB78]. The following decade saw a high activity in the field.
The development took place along several different paths. Methods based on 2D as well as 3D formulations were
developed. Depending on the choice of description of motion, type of constitutive relations, and solution procedures,
the methods were described as the solid approach, the static-implicit approach, the static-explicit approach, the
rigid-plastic approach and the flow approach (which have the disadvantage that any phenomenon related with elasticity
cannot be simulated, as the springback ).
During this period of time (the 1980s) the flow and rigid-plastic approaches were more popular than the static-
implicit one, mainly because they were more stable. The practical application of sheet forming simulations was too
unstable procedures and excessive computing times, even for very small problems.
In 1989 Honecker [ArtHen89] and Mattiasson presented results from a study, in which the dynamic-explicit
approach was evaluated in application to sheet metal stamping, the results from this study were very promising. Since
sheet forming processes usually have much longer durations than what the dynamic, explicit method normally is
intended for, the prerequisites for the problem have to be modified in some way, in order to fully utilise the benefits of
the method, but with additional backwards like virtual inertial forces.
At early 1990s there was significant increase of the practical utilisation of sheet forming simulations within the
industry, and from the middle of this decade most companies within the automotive industry were performing sheet
stamping simulations on a regular basis. Dynamic-explicit codes were dominating the software market, codes like
LS-DYNA and ABAQUS/Explicit appeared in the market. From the side of the static-implicit the code AutoForm
emerged from a research project at Eidgenössische Technische Hochschule (ETH) in Zurich at that time, which
used some innovative algorithms to enhance stability and computational efficiency. Besides this code is very extended
today, the software market is still dominated by various dynamic-explicit codes like LS-DYNA, ABAQUS/Explicit,
PAM-STAMP 2G and STAMPACK.
To promote the research of sheet metal forming simulation and investigate the reliability of numerical analysis
algorithms, international research organisations have designed a series of standard questions, such as Ohio State
University (OSU), Verein Deutscher Ingenieure (VDI) and NUMISHEET. These questions aim to assess the
performance of the existing software to predict splitting, wrinkling, buckling, and springback.

1.2 Most common issues in forming processes

The following are the most common issues that every engineer who want to work with forming simulations should deal
with to perform a realistic and correct simulation of the problem. In addition to the following one the splitting problem
can be mentioned, which as the name indicates this defect appears when the material breaks, producing splits in
it.

1.2.1 Springback
Springback [BookBan07] is a particularly critical aspect of sheet metal forming (See Figure 1.4). Even relatively
small amounts of springback in structures that are formed to a significant depth may cause the blank to distort to
the point that tolerances cannot be held. Such geometrical differences caused by springback mainly occur for three
reasons:

Vicente Mataix Ferrándiz Page 27 of 374

1.2 Most common issues in forming processes CHAPTER 1. INTRODUCTION

Figure 1.4: Springback

• Due to interferences in the residual stress state of the formed parts

• After removal of the external forces
• Due to the Bauschinger [ArtBau86] effect. In a typical sheet metal forming process, a considerable amount of
material undergoes non-proportional loading. It is well known that the mechanical response of metals depends
not only on the current state but also on the previous deformation history. One of the phenomena related to the
change of mechanical properties during a non-proportional loading is the Bauschinger effect[BookBan10]. New
materials such as high-strength steel, aluminium and magnesium are particularly prone to springback.

To compensate for the springback effects the tool must be reworked, which causes further costs and requires
more time. To avoid these additional efforts during die design, the simulation of springback becomes more and more
important in the forming simulation.

1.2.2 Wrinkling

Figure 1.5: Wrinkling phenomena

Compressive stresses are formed in the plane of the sheet results in wrinkling[ArtFO11; ArtCFM03; ArtCW00;
ArtXSZ15], this kind of instability could occur during the forming process, producing an undesired shape and defects
in the final product. The tendency of wrinkling increases with:

• Unsupported length of sheet metal

• Decreasing thickness
• Not uniformity in thickness
• Lubricants trapped can also contribute to wrinkling

At the end this problematic is an instability that occurs due to the same reasons that buckling occurs for one-
dimensional elements like beams, extended for two-dimensional elements. The Figure 3.32 shows us some example
of wrinkling for simple problem of drape simulation.

Page 28 of 374 Vicente Mataix Ferrándiz

CHAPTER 1. INTRODUCTION 1.3 Objectives

1.3 Objectives
The main objective of this thesis is to advance in the development of FEM for the stamping processes, and do it in
such a way that the resulting product of this work will be a robust tool with the capability of solve problems demanded
by the industry, or in other words, a product with an able to compete in the market with the existing commercial
solutions. To reach this aim the following goals are set:

• The implementation of a fully functional prismatic solid-shell element based in the formulation of Flores[ArtFlo13c;
ArtFlo13a; ArtFlo13b], with the convenient modification of the formulation to be able of computing the push-
forward and pull-back. This goal has been reached satisfactory during the current course of doctorate.

• With the aim of a robust methodology to compute forming processes, the computations shown in this work
will focus in implicit methods, in contrast to the extended explicit approaches. The drawbacks of the implicit
method[BookBan07] were the computation speed, the high memory requirements and the frequent convergence
problems with complex processes; but fortunately the power of implicit methods is increasing more rapidly
than that of explicit methods. Implicit[BookBel+14] methods for the treatment of nonlinear constraints, such as
contact and friction, have been improved tremendously. Sparse iterative solvers have also become much more
effective.

• The implementation of a fully functional contact method. In the literature[BookWri06; BookLau10; PhDYas11;
PhDPop12; BookBel+14; ArtOAM08] it exists an extent discussion of which method could be the more adequate
for simulate these kinds of problems. In order to perform the most consistent simulation, the method chosen will
be the mortar method, which will be based in Popp[PhDPop12] work.

• The implementation of the required constitutive laws in order to obtain a good approach of the behaviours that
will experiment the forming problems.

• Strategies of self-refinement[PhDHat03; ArtXSZ15; BookZZT13] will be considered. The objective is to be able

of computing any kind problem, starting from a coarse mesh; and refine whenever is necessary to obtain a
feasible solution of the problem.

• At the end of the development cycle, different routines coming from the know-how[BookBan10; BookVal10;
BookKoc08; BookHu+13; BookAT12a; BookTP07] of the industry will be implemented with the aim of obtain a
fully functional commercial software.

For more details related with the objectives the chapters 8.Final conclusions and 9.Future works contain these
points developed in greater depth.

Vicente Mataix Ferrándiz Page 29 of 374

1.3 Objectives CHAPTER 1. INTRODUCTION

Page 30 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookAT12a] Sheet Metal Forming - Fundamentals. Taylan Altan and A. Erman Tekkaya. ASM International.
2012.
[BookAT12b] Sheet Metal Forming: Processes and Applications. Taylan Altan and Erman Tekkaya. ASM
International. 2012.
[BookBan07] Advanced Methods in Material Forming. Dorel Banabic. 2007.
[BookBan10] Sheet Metal Forming Processes: Constitutive Modelling and Numerical Simulation. Dorel Banabic.
Springer-Verlag Berlin Heidelberg. 1st ed. 2010.
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookBHS07] Encyclopedia of Computational Mechanics (3 Volume Set). René de Borst, T.J.R. Hughes, and
Erwin Stein. Wiley. 1st ed. 2007.
[BookHu+13] Theories, Methods and Numerical Technology of Sheet Metal Cold and Hot Forming: Analysis,
Simulation and Engineering Applications. Ping Hu, Ning Ma, Li-zhong Liu, and Yi-guo Zhu.
Springer-Verlag London. 1st ed. 2013.
[BookKoc08] Hydroforming for Advanced Manufacturing. M. Koc. CRC Press. 2008.
[BookLau10] Computational Contact and Impact Mechanics: Fundamentals of Modeling Interfacial Phenomena
in Nonlinear Finite Element Analysis. Tod A. Laursen. Springer. 2010.
[BookTP07] Mechanics Modeling of Sheet Metal Forming. Sing C. Tang and Jwo Pan. Society of Automotive
Engineers, Inc. 2007.
[BookVal10] Applied Metal Forming: Including FEM Analysis. Henry S. Valberg. Cambridge University Press. 1st
ed. 2010.
[BookWri06] Computational Contact Mechanics. Peter Wriggers. Springer. 2nd. 2006.
[BookWri08] Nonlinear finite element methods. Peter Wriggers. Springer. 1st ed. 2008.
[BookZZT13] The Finite Element Method: its Basis and Fundamentals. O. C. Zienkiewicz, J.Z. Zhu, and
Robert L. Taylor. Butterworth-Heinemann. 7th ed. 2013.

Articles
[ArtBau86] “Über die Veränderung der Elastizitätsgrenze und der Festigkeit des Eisens und Stahls durch
Strecken und Quetschen, durch Erwärmen und Abkühlen und durch oftmals wiederholte
Beanspruchung”. Johann Bauschinger. In: Mitteilungen des mechanisch-technischen
Laboratoriums der Königlich Technischen Hochschule München. No. 1, Vol. 13, 1886,
[ArtBRW75] “Finite element formulations for large deformation dynamic analysis”. Klaus-Jürgen Bathe,
Ekkehard Ramm, and Edward L. Wilson. In: International Journal for Numerical Methods in
Engineering. No. 2, Vol. 9, 1975, pp. 353–386. John Wiley & Sons, Ltd. DOI:
10.1002/nme.1620090207.

Vicente Mataix Ferrándiz Page 31 of 374

PH.D.’S THESIS BIBLIOGRAPHY

[ArtCFM03] “Analytical and numerical investigation of wrinkling for deep-drawn anisotropic metal sheets”.
J. P. de Magalhaes Correia, G. Ferron, and L. P. Moreira. In: International journal of mechanical
sciences. No. 6, Vol. 45, 2003, pp. 1167–1180. Elsevier.
[ArtCW00] “An analytical model for plate wrinkling under tri-axial loading and its application”. Jian Cao and
Xi Wang. In: International journal of mechanical sciences. No. 3, Vol. 42, 2000, pp. 617–633.
Elsevier. DOI: 10.1016/S0020-7403(98)00138-6.
[ArtFlo13a] “A “Prism” solid element for large strain shell analysis”. Fernando G. Flores. In: Computer Methods
in Applied Mechanics and Engineering. Vol. 253, 2013, pp. 274–286. Elsevier. DOI:
10.1016/j.cma.2012.10.001.
[ArtFlo13b] “Development of a non-linear triangular prism solid-shell element using ANS and EAS techniques”.
Fernando G. Flores. In: Computer Methods in Applied Mechanics and Engineering. Vol. 266, 2013,
pp. 81–97. Elsevier. DOI: 10.1016/j.cma.2013.07.014.
[ArtFlo13c] “UN ELEMENTO PRISMA TRIANGULAR DE SÓLIDO-LÁMINA PARA EL ANÁLISIS CON
GRANDES DEFORMACIONES”. Fernando G. Flores. In: Mecánica Computacional. Vol. XXXII,
2013, pp. 63–87.
[ArtFO11] “Wrinkling and folding analysis of elastic membranes using an enhanced rotation-free thin shell
triangular element”. Fernando G. Flores and Eugenio Oñate. In: Finite Elements in Analysis and
Design. No. 9, Vol. 47, 2011, pp. 982–990. Elsevier. DOI: 10.1016/j.finel.2011.03.014.
[ArtHen89] “Finite-Element Procedures for 3D Sheet Forming Simulation”. A Henecker. In: . Vol. 457, 1989,
[ArtHMR70] “A finite element formulation for problems of large strain and large displacement”. Hugh D Hibbitt,
Pedro V Marcal, and James R Rice. In: International Journal of Solids and Structures. No. 8, Vol. 6,
1970, pp. 1069–1086. Elsevier. DOI: 10.1016/0020-7683(70)90048-X.
[ArtKC06] “Achievements of metal plastic working in ancient China”. WANG Kezhi and S Chen. In: J Plast
Eng. Vol. 6, 2006, pp. 114–125.
[ArtMR75] “Finite-element formulations for problems of large elastic-plastic deformation”.
Robert M McMeeking and JR Rice. In: International Journal of Solids and Structures. No. 5, Vol. 11,
1975, pp. 601–616. Elsevier. DOI: 10.1016/0020-7683(75)90033-5.
[ArtOAM08] “Algorithms and strategies for treatment of large deformation frictional contact in the numerical
simulation of deep drawing process”. M. C. Oliveira, J. L. Alves, and L. F. Menezes. In: Archives of
Computational Methods in Engineering. No. 2, Vol. 15, 2008, pp. 113–162. Springer. DOI:
10.1007/s11831-008-9018-x.
[ArtWB78] “Analysis of sheet metal stamping by a finite-element method”. N-M Wang and Bernard Budiansky.
In: Journal of Applied Mechanics. No. 1, Vol. 45, 1978, pp. 73–82. American Society of Mechanical
Engineers. DOI: 10.1115/1.3424276.
[ArtXSZ15] “Drape simulation using solid-shell elements and adaptive mesh subdivision”. Q. Xie, K.Y. Sze, and
Y.X. Zhou. In: Finite Elements in Analysis and Design. Vol. 106, 2015, pp. 85–102. Elsevier. DOI:
10.1016/j.finel.2015.08.001.

Ph.D.’s thesis
[PhDHat03] “Automated modeling and remeshing in metal forming simulation”. Nitin Vasant Hattangady. 2003.
[PhDPop12] “Mortar Methods for Computational Contact Mechanics and General Interface Problems”.
Alexander Popp. 2012.
[PhDYas11] “Computational contact mechanics: geometry, detection and numerical techniques”.
Vladislav A. Yastrebov. 2011.

Online resources
[OnlShe] Wikipedia. Sheet metal forming simulation. URL :
https://en.wikipedia.org/wiki/Sheet_metal_forming_simulation.

Page 32 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION

Chapter 2

Finite Element formulation

The limitations of the human mind are

such that it cannot grasp the behaviour of
its complex surroundings and creations in
one operation. Thus the process of
subdividing all systems into their
individual components or elements,
whose behaviour is readily understood,
and then rebuilding the original system
from such components to study its
behaviour is a natural way in which the
engineer, the scientist, or even the
economist proceeds.

Olgierd C. (Olek)
Zienkiewicz[BookZZT13]
(1921 - 2009 AD, British academic of
Polish descent.FEM pioneer)

2.1 Introduction
This chapter introduces the concepts underlying the nonlinear continuum mechanics for FEA. This is necessary in
order to understand the formulation exposed in following chapters, both for solid elements and for the understanding
of the contact mechanics theory.
In the first section, the FEM theory is introduced (2.2.Finite Element Method), in the second section the solid
mechanics theory is presented (2.3.Solid mechanics), the third and last section presents the application of the FEM in
the resolution of solid mechanics problems (2.4.Finite Element formulation for CSD).

2.2 Finite Element Method

2.2.1 Introduction
Physicists, engineers and mathematicians have been considering the Ordinary Differential Equation (ODE), and
its generalisation in multiple variables the Partial Differential Equation (PDE), as a key tool in their fields since
the calculus was created. Particularly physicists, and by extension engineers too, have been using the PDE as a
tool to describe a wide variety of physical phenomena such sound, heat, diffusion, electrostatics, electrodynamics,
fluid dynamics, elasticity, among others. The description provided by this mathematical tool introduces a continuous
problem with a set of Boundary Conditions (BC), solvable only in a very limited range of cases and with a series of
hypotheses taken in order to simplify the problem. So that we extend the number of problems we are able to solve we

Vicente Mataix Ferrándiz Page 33 of 374

2.2 Finite Element Method CHAPTER 2. FINITE ELEMENT FORMULATION

can discretise the problem, and solve approximately considering a numerical approximation instead of an analytical
solution. With the advent of modern computers discrete problems can generally be solved readily even if the number
of elements is very large.

This discretisation process has been an object of discussion, as well as has brought fruitful developments and
successive works. Each one of the methodologies arose has its advantages and its disadvantages, so when choosing
one it may become a compromise solution or even a solution that takes into consideration several ones at the same
time. Mathematicians have developed general techniques applicable to different types of problems[BookZZT13],
the most significative ones are the FD[ArtRic11], weighted residual procedures[BookFin13], the use of variational
methods[ArtPom69], among others.

On the other hand, engineers with their more practical point of view took this decomposition principle and create
an analogy between the discrete entities and finite portions of the continuum domain[BookZZT13], leading to the
development of the first FE methods by Turner [ArtTur56] in 1956 when working for Boeing group. In any case, the
first one considers the term FEM to refer to the method was Clough[ArtClo60] in 1960. Since then the developments
of the FEM have been staggering1 , and has been seen as a general method able to deal with any kind of continuum
mechanics problem, this means a general method to solve problems defined via PDE. This includes the development
of different codes for the FEM, at the beginning these code developed at Berkeley was nameless[BookBel+14],
engineers developed new applications by modifying and extending these early codes. The next generations became
more sophisticated, name included, and grow and even became a commercial product. After that, many programmes
appear, many commercial, many with educational purpose and even Free and Open Source Software (FOSS)
projects arise.

Other kinds of methods that it is relevant to mention, which follows a similar strategy on dividing the domain in
discrete entities, is the Finite Volume Method (FVM)[BookLL92]. This method allows to solve conservative problem
by applying the divergence theorem in the discrete volumes, then the volume integrals in a PDE that contain a
divergence term are converted to surface integrals. These terms are then evaluated as fluxes at the surfaces of each
finite volume. Due to the latter limitation, only conservative problems can be solved2 , the FEM are a method more
suitable to solve multi-physics and more general problems. This is the reason why this methodology is the chosen
one, to deal with the physics of the different types of problems presented in this work.

2.2.2 Concept of FE
Here the concept of FE early introduced is detailed. Once the continuous problem is formulated in a PDE form, we
can proceed[BookZZT13] recasting the problem in an alternate form, denominated weak form from which we can
approximate the solution. A weak form to a set of PDE is obtained using the following steps[BookZZT13]:

1. Multiply each equation by an appropriate arbitrary function, denominated test function, defined in the domain of
interest (Ωi ). The expression obtained is a function of functions called a functional.

2. Integrate this product over the space domain of the problem. This can be achieved following different methodolo-
gies, the most common use, and the one considered in this work, is the weighted residuals Galerkin[ArtGal15]
method, usually referenced as Galerkin method only. This approach consists in considering the same shape
functions for the test functions than for the unkowns. The method of weighted residuals[BookFin13] is previous
to the FEM, and its consideration as the integration method leads to the Generalized Finite Element Method
(GFEM).

3. Use integration by parts to reduce the order of derivatives to a minimum. Or use more advance techniques as
the Green’s theorem.

4. Introduce BC if possible. This depends on the nature of the BC, which may lead to a modification of the
equations which define the problem, or just simple BC directly imposed over the system.

1 For years journals as Journal of Applied Mechanics rejected papers on the FEM because it was considered no scientifically relevant[Book-

Bel+14].
2 It is possible to deal with not conservative problems with its respective additional treatment, see for example the work of Castro[ArtCGP06].

Page 34 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION 2.2 Finite Element Method

The former procedure can be summarised conceptually as follows. We can take the Figure 2.1 as reference
problems, a stationary heat equation with a non-uniform set of BC. Then we can reduce the FE problem to solve in
the following three steps:

1. We want to solve the Laplacian equation for the heat problem on a certain domain with a certain BC from Figure
2.1a. Both the geometry and the BC considered on this setup are not trivial, meaning is not easy to obtain an
analytical solution. Then the weak problem is formulated. The Left-Hand Side (LHS) and Right-Hand Side
(RHS) is obtained from this weak form. We will solve the resulting system as any standard linear solver.

2. The domain is divided in discrete elements, like triangles as seen on Figure 2.1b. The previous weak form is
integrated in the discrete mesh, obtaining the respective system of equations.

3. After integrating on this discrete domain we are able to solve the PDE with our BC, obtaining as result the
temperatures seen on Figure 2.1c.

(a) Laplacian (∇2 ) problem with a set of (b) Triangular mesh considered to discre- (c) Temperature distribution obtained
Dirichlet BC tise the domain
Figure 2.1: Minimal example of FEA for a Laplacian problem. Resolution using Mathematica[OnlWol]

In order to obtain the system of equations for the FEA problem, we need to derive the respective LHS and RHS.
In order to do so, we need to use the derivative, which is a linear operation, and then can be applied systematically,
and therefore it is a potentially automated task. This can be done in fact in an automatic manner using the Automatic
Differentiation (AD) procedure, which is detailed on the corresponding appendix C.Automatic differentiation.
The FEM can be used in order to solve a large range of problems, including complex NL problems. Without an
understanding of the fundamentals, a finite element program is a black box that provides simulations[BookBel+14].
However nonlinear FEA requires the understanding of the underlying theory, as well as an adequate level of experience.
On this kind of simulation, many choices and pitfalls are taken into account, without it the analyst could provide
an incorrect diagnosis. On NL problems bifurcation or localisation may occur, and many times convergence of the
numerical analysis is not always obtained in nonlinear problems[BookWri08].
Because of this we directly address the literature where this nonlinear FEM is studied on detail. The work
of Zienkiewicz[BookZTF14], Belytschko[BookBel+14], de Borst[BookBor+12] and Wriggers[BookWri08] are good
references. In summary, any NL FEA has the following steps[BookBel+14]:

Vicente Mataix Ferrándiz Page 35 of 374

2.3 Solid mechanics CHAPTER 2. FINITE ELEMENT FORMULATION

1. Development of a model: Which represents the phenomenon from a physical point of view.

2. Formulation of the governing equations: Translate the previous model into a set of PDE and its respective
BC which allows to treat the problem mathematically.

3. Discretisation of the equations: This includes the selection of the most adequate method to solve it and the
corresponding framework.

4. Solution of the equations: This means not only the algebraic resolution of the system of equations, but also
the definition of the discrete mesh and the proper solution strategy.

5. Interpretation of the results: Including its proper representation and post-process and reach the pertinent
conclusions.

We need to differentiate between the origins of the NL in order to properly address its problematic. We can list the
causes of the nonlinearities in a FEA as the following six ones[BookWri08]:

• Geometrical nonlinearity: When large displacements and rotations have to be considered, because the
effects on the change on the geometry cannot be obviated.

• Finite deformations: Here not only the displacements are large but also the strains. Both of this NL are
discussed on the chapter dedicated to the element developed on this work, see 3.Rotation-free shells and
solid-shell elements.

• Physical nonlinearity: Many materials depict nonlinear behaviour, like plasticity or damage. This is studied on
this work on the respective chapter 5.Plasticity.

• Stability problems: We can differentiate between the geometrical and material instabilities. This instability is
complex, and usually require specific strategies in order to be taken into account like the arc-length.

• Nonlinear boundary conditions: Like for example the contact between two bodies. This subject is treated on
deep in this work, for more information see 4.Contact mechanics.

• Coupled problems: When more than one type of problem is combined. Problems like FSI, thermomechanical,
chemical reactions, etc.

2.3 Solid mechanics

2.3.1 Linear form
If we consider an infinitesimal strain-displacement the equations of Navier-Cauchy [BookZZT13] are expressed in
tensor notation as Equation (2.1).
Equation of motion (linear momentum): Which is an expression of Newton’s second law.

(2.1a) ∇ · σ + b = ρü

Strain-displacement equations: For infinitesimal strain (k∇uk ≪ 1 and kuk ≪ 1) E ≈ e ≈ ε.

 
(2.1b) ε= 1
2
∇u + (∇u)T

Constitutive equations: For elastic materials, Hooke’s law represents the material behaviour and relates the
unknown stresses and strains. In this equation C represents the fourth-order constitutive tensor.

(2.1c) σ=C:ε

2.3.2 Non-linear form

Now we consider a NL geometry and kinematics[OnlFin; BookBHS07; BookNPO09] the development for problems
where the solid deals with deformations in which both rotations and strains are arbitrarily large.

Page 36 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION 2.3 Solid mechanics

Figure 2.2: Deformation of a continuum body

2.3.2.1 Deformation gradient

Considering a generic continuum body Ω (Figure 2.2) which occupies a region of the three-dimensional Euclidean
space ℜ3 in its reference configuration, with boundaries ∂Ω or Γ. Let Ω be subjected to a motion φ so that for each
time t, the deformation (2.2a) maps each material particle X of Ω into the position x it occupies at time t. The set
(2.2b) is called the current or deformed configuration. The two-point tensor F defined in (2.2a) by (3.26) is termed
the deformation gradient, where {EI }I=1,2,3 and {ei }i=1,2,3 are fixed orthonormal bases in the reference and deformed
configuration, respectively, typically chosen to be coincident with the standard basis in ℜ3 . The Jacobian (J) of
mapping (2.2a) can be represented as (3.15a).

(2.2a) φ(., t) : Ω → ℜ3
(2.2b) x = φ(X, t)
∂φi
(2.2c) F(X, t) = D φ(X, t) = ei ⊗ EI
∂ XI
(2.2d) J = det(F)

2.3.2.2 Strain tensors

Imposing the standard condition on the deformation gradient det(F) > 0 ensuring that (2.2a) is a one-to-one mapping,
the polar decomposition theorem admits the unique representation of the deformation gradient (3.26) in the form
(2.3a), see Figure 2.3, where:

• U and V: Are positive definite symmetric tensors. These tensors are known as the right and left stretch tensors
respectively, and measure local stretching near X. Since U and V are symmetric, it follows from the spectral
theorem that they admit the spectral decomposition as can be seen in (2.3b).

• R: Is an orthogonal tensor, called rotation tensor and measures the local rigid rotation of points close to X .

In (2.3c) the right and left Cauchy-Green (C and b) tensors are introduced. For the definition of the strain tensors
in (2.3d) we can find the definition of a set of strain measures in function of the parameter m. The GreenLagrange
strain tensor, E(m) , is a family particular member of this (with m = 2). Other commonly employed members of this
family are the Biot (m = 1), Hencky (m = 0) and Almansi (m = −2) strain tensors. It should be emphasised that, in
order to preserve the invariance of the stress work per unit mass, a particular choice of strain measure necessitates
usage of the associated dual stress measure within the constitutive relations. When linearised about the reference

Vicente Mataix Ferrándiz Page 37 of 374

2.4 Finite Element formulation for CSD CHAPTER 2. FINITE ELEMENT FORMULATION

Figure 2.3: Polar decomposition of F

state, all strain measures give the standard small strain tensor.

(2.3a) F=R·U=V·R
3
X 3
X
(2.3b) U= λ i li ⊗ li , V = λi ei ⊗ ei , li = Rei
i=1 i=1

(2.3c) C = F · F = U2 , b = F · FT = V2
T
(
1
(m) m
(Um − I) m 6= 0
(2.3d) E =
ln(U) m = 0

Equivalently, in terms it spectral decomposition can be reformulated as (2.3e).

 P3
(
E m) = ( i=1 f (λi )li ⊗



li
1 m
(2.3e) m
λi − 1 m 6= 0
f (λi ) =

ln(λi ) m = 0

We add to the former that all these strain tensors (E(m) ) for any strain measure can be transformed between them
in considering the push-forward (φ∗ ) and pull-back (φ∗ ) operations. This operation just requires, in addition to the
given strain tensor in a given strain measure, the deformation gradient (F). For example, using the push-forward we
can transform the Green-Lagrange strain tensor into the Almansi strain tensor. With the pull-back operation, we can
proceed in the opposite direction, from Almansi to Green-Lagrange. For more information about the procedure, we
address to a dedicated Appendix section A.1.Pull-Back, Push-Forward fundamental concepts where the operation is
detailed.

2.4 Finite Element formulation for CSD

2.4.1 Weak formulation in solids
Here we introduce a summarise of the Galerkin method[OnlGal] for the deduction of the weak formulation, using as
main reference the work of Zienkiewicz[BookZZT13]. The deduction can be followed in (2.4).

Page 38 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION 2.4 Finite Element formulation for CSD

First we divide the domain Ω in subdomains Ωe (elements), as well as the boundary Γ, (2.4a). In the following we
will represent the discrete form of the former continuous expressions as ˆ.

( P
Ω ≈ Ω̂ = e Ωe
(2.4a) P P P
Γ ≈ Γ̂ = e Γe = et Γte (Neumann) + eu Γue (Dirichlet)

For the displacement (u) we consider an approximation employing the shape functions (Nb ) corresponding to the
elements considered in our formulation, (2.4b).

X
(2.4b) u(ξ , t) = û = Nb (ξ )ub (t) = N(ξ )u(t)
b

With the shape forms we can define the derivatives, needed to define the strain matrix (B), (2.4c). This matrix is
necessary in order to formulate the strain-displacement equation ε = Bu.

∂ Na ∂ xj ∂ Na ∂ Na ∂ Na
(2.4c) B= i
= i
, and in matrix form =J
∂ξ ∂ξ ∂ xj ∂ξ ∂x

Finally, the weak form of the equilibrium’s equation can be expressed as (2.4d). In this case we include the
dynamic effects in order to be a proper CSD. For that we will consider the respective velocity u̇ and acceleration ü.
The damping coefficient will be simplified with the parameter c, a better alternative will be shown at 2.4.5.Damping
matrix. In here the body forces (b) as well as the external forces (t̄) are taken into consideration.

Z Z Z Z Z 
T T T T T T
(2.4d) δ We (u) = δ u N ρNd Ωü + N cNd Ωu̇ + B σd Ω − N bd Ω − N t̄d Γ
Ωe Ωe Ωe Ωe Γ

That can be expressed in a semi-discrete form from (2.4e). In here we can express the stress divergence or
stress force term (P), the mass matrix (M), the damping matrix (C) and the resulting external forces (f). P includes the
resulting internal forces of the system.

(2.4e) Mü + Cu̇ + P(σ ) = f

Where the former components can be expressed as (2.4f). In case of considering a linear problem, the stress
divergence (P) can be obtained directly with the stiffness matrix (K) and its displacements (u).
 P R

 M = e Ωe NT ρNd Ω
 P R T
C = P e R Ωe N cNd Ω



(2.4f) P = e Ω e BT σ d Ω
 (e) (e)


 For linear
R elasticity P R(σ ) = K u

 P
NT Bd Ω + NT t̄d Γ
f =
e Ωe Γte

2.4.2 TL and UL formulation

Extending the previous section, here we introduce the two possible formulation for the mechanic problem, the UL and
TL formulation[BookBel+14]. In the TL form the only variable which depends on the deformation is the nominal stress
(P, see Table 2.1), that is, it is the only variable which varies with time. In the UL form the domain of the element (or

Vicente Mataix Ferrándiz Page 39 of 374

2.4 Finite Element formulation for CSD CHAPTER 2. FINITE ELEMENT FORMULATION

body), the spatial derivatives ∂ NI /∂ xi and the Cauchy stress (σ ) depends on the deformation, and hence on time,
which complicates the derivation of the tangent stiffness.

 R
int
f = ΩR0 B0 PdΩ0 

(2.5a) Kmat T
IJ = Ω0 B0I C
SE
B0J d Ω0

 geo R T
KIJ = I Ω0 B0I [S] B0J d Ω0

 R
int
f = ΩRBσ d Ω

 σT 
(2.5b) Kmat T
IJ = Ω BI C BJ d Ω

 geo R T
KIJ = I Ω BI [σ ] BJ d Ω

The UL forms (2.5b) are generally easier to use than the TL forms (2.5a), since B is more easily constructed than
B0 and many material laws are developed in terms of rates of Cauchy stress (σ ). The material stiffness in TL form
can be combined with the geometric stiffness in UL form or vice versa. In order to compute the material stiffness
matrix (Kmat ), in addition to compute B for the TL and B0 for the UL, we need to compute the consistent constitutive
tensor (C), CSE for TL and CσT for UL. The geometric stiffness matrix (Kgeo ) depends on Second Piola-Kirchhoff
(PK2) (S) on the TL frame, but in the UL it depends on the Cauchy stress (σ ). The element SPRISM (3.4.Prismatic
solid-shell) has been developed in TL formulation, but there is an intention to implement in UL as well.
The following, Table 2.1, will help to understand the different types stress measures and the possible transformation
applying the corresponding considering the push-forward (φ∗ ) and pull-back (φ∗ ) operations.

Cauchy stress σ Nominal stress P Second PK stress S Corotational Cauchy stress σ̂

σ= J −1 F · P J −1 F · S · FT JR · σ̂ · RT
P= JF−1 · σ S · FT JU−1 · σ̂ · RT
S= JF−1 · σ F−T P · F− T JU−1 · σ̂ · U−1
σ̂ = RT · σ · R J −1 U · P · R J −1 U · S · U
τ = Jσ F·P F · S · FT JR · σ̂ · RT
Table 2.1: Transformations of stresses. Source[BookBel+14]
2.4.3 Solution of the non-linear-equilibrium equations system
When we solve non-linear problems[BookOll14] we need to consider efficient methods designed for this purpose.
This means the resolution of ∆ft+i
∆t
= −Ji t+∆t · ∆ui+1
t+∆t
, which can be carried basically by two kinds of methods, a
linearisation based in the standard/modified Newton-Raphson (NR) technique or a Quasi-Newton approximation
technique of the Jacobian matrix (J ). Both methods can be improved with the use of the convergence accelerators
algorithms. In the following we are going to present the NR method, the concept of line search, and the arc-length
method.

2.4.3.1 Newton-Raphson method

This is the fastest convergence method to solve non-linear equations systems using the linearisation technique. This
technique assumes that the solution is convergent in the surroundings of it, being the convergence ratio in this case
quadratic. Its solution comes from the inversion of the Jacobian operator (J ) from Equation (2.6). In our system of
equations J acts as the LHS.

  −1
(2.6) [∆u]t+∆t
i+1 = − Ji
t+∆t
· [∆f]t+
i
∆t

The advantages of the method can be listed as the followings:

• It always needs the tangent operator, which is not always easy to obtain, store and invert (for example in the
case of antisymmetric operators)

Page 40 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION 2.4 Finite Element formulation for CSD

• The convergence velocity depends significantly on the initial guess solution

• The method can get stuck in a local minima, which could be difficult to leave afterwards

This will be the default strategy followed in our developments and problems. In the contact chapter, it will be shown
the strategies followed in order to adapt the NR strategy to the proper resolution of the contact problem.

2.4.3.2 Line-search
Zienkiewicz[BookZTF14] classifies this method in the category of convergence accelerators. The approach[BookZTF14]
is based in the concept of the descent direction. This descent direction can be obtained applying different procedures,
such the gradient descent, NR or quasi-Newton method. Independently of the approach, the method will proceed as
the following Algorithm 1.

Algorithm 1 Line-search
1: procedure Line-search
2: Set iteration counter i = 0, and make an initial guess x0 for the minimum
3: while (rhstotal > toleranceabs and ratio > tolerancerel ) and i < iterationmax do
4: Compute descent (or search) direction dx
5: Choose ηk (step size) to minimize the projection of the residual on the search direction: G(ηk ) = f (xk + ηk dx)
6: Update xk+1 = xk + ηk dx
7: Update i = i + 1 and the RHS

This method is discarded, as cannot be used, at least without additional considerations, for CCM. We will then
focus on the other two methods. In general, it is complex to consider convergence accelerators on contact problems,
as the BC of the problem is continuously changing and this is troublesome for this kind of strategy.

2.4.3.3 Arc-length

Figure 2.4: Graphic representation of the arc-length method

Unstable behaviours[BookOll14] occur for various problems in structural mechanics, for which it is difficult to find
a solution. To avoid this, a system of equilibrium equations with restrictions of type ∆f(u, u̇, ü, λ) are usually used,
where the magnitude of external force λfext is an unknown conditioned by an additional equation c(u, λ). The basic
concepts of the method are presented in (2.7) (see Figure 2.4).

Vicente Mataix Ferrándiz Page 41 of 374

2.4 Finite Element formulation for CSD CHAPTER 2. FINITE ELEMENT FORMULATION

Considering the following relations stating from the equilibrium equations (2.7a).
   
∂∆f ∂∆f
(2.7a) = Ji , = fext
∂u i ∂λ i

It allows to rewrite the equilibrium equations as (2.7b).

(
0 = ∆f (ui , λi ) + Ji · δ ui+1 − fext · δλi+1
(2.7b)
0 = c(u, λ)i+1

The displacement increment can be written as (2.7c), where umax is the total displacement with the last or maximal
value of the external force and δ û is the solution of the system of equations without correction and δλi+1 is the change
of the application factor of the load.

(2.7c) δ ui+1 = δ ûi+1 + umax

i+1 δλi+1

  
int ext
Ji · δ ûi+1 = − Mü + Cu̇ + fi (u̇, u) − λi f

(2.7d) Ji · umax = fext

λi+1 = λi + δλi+1


With the basis developed previously (2.7) it is possible to implement different versions of the method. The version
implemented in Kratos is based in different implementations from the literature[ArtFM93; ArtTL98; ArtCri83].

2.4.4 Time integration schemes

Taking the semi-discrete form of the dynamic principle, Equation (2.8), we obtain the equation governing the solid
dynamics (CSD). We will express in this manner, instead of the proper residual form, Mü+Cu̇+LHSstatic ∆u = RHSstatic ,
for being more illustrative showing the resulting structure when applying a certain time integration scheme.

(2.8) Mü + Cu̇ + Ku + f = 0

We can now choose between the implicit and explicit approach, following[BookOll14] we can define them as the
following. If the response at current time depends completely from the solution in the current step then we are working
in a explicit solution, but in the other hand if the solution depends on velocity and the acceleration at current time, then
we have an implicit solution. Considering the set of equations (2.9), which depend on the α parameter, then we can
say:

• Explicit: It states equilibrium at time t, with α = 0 (Forward Euler ). The displacement in next step is obtained
depending on the velocity and displacement of the previous step. This methodology if simple in terms of structure
requires less memory storage, does not require expensive tangent operators, the algorithms obtained are
reliable and it requires a very small time increment what it can be numerically expensive and unapproachable.

• Implicit: It formulates the equilibrium at time t + ∆t with α = 1 (Backward Euler ). The displacement in the
next step is obtained depending on the current time velocity and on the displacement of the previous step. For
this approach the time increments can be much larger preserving stability, allows more precise solutions, but
requires the computation of the tangent operators (LHS) and large storage demand.

(
ut+∆t = ut + ∆t u̇t+α∆t
(2.9)
u̇t+∆t = (1 − α)u̇t + αu̇t+∆t

Page 42 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION 2.4 Finite Element formulation for CSD

As we have said previously, the aim of this work will focus in an implicit solution; thus we need to discretise (2.8)
with some temporal scheme, the ones that we present corresponds with the most extended in structural dynamics,
the Newmark-β (2.4.4.1.Newmark time scheme) scheme, the Bossak scheme (2.4.4.2.Bossak algorithm) and the
α-generalised (2.4.4.3.α-generalised). For additional literature related with structural dynamics, and in particular
CSD, we recommend to consult the bibliography[BookCP12; BookOll14; BookRR12], where alternative schemes are
explained, and the stability and convergence of each one is deduced.

2.4.4.1 Newmark time scheme

Below with one step method employing Newmark schema we obtain (2.10).


n+1 n+1
(c3 M + c2 C + c1 K)u + f
= 0

n+1 1
(2.10) u = u + ∆t u̇ + 2 − β ∆t 2 ün + β∆t 2 ün+1
n n

 n+1
u̇ = u̇n + (1 − γ )∆t ün + γ∆t ün+1

γ 1
Where the coefficients of the scheme are c1 = 1, c2 = β∆t and c3 = β∆t 2 .

2.4.4.2 Bossak algorithm

The Kratos framework uses the Bossak [ArtWBZ80] algorithm by default for the time integration. This method is based
in the Newmark method, introducing additional parameters to dissipate spurious oscillations in the high frequencies.
The parameters introduced modify the form of the equilibrium equation including an averaging process between the
current and the old-time station so to preserve the consistency requirements. The only variation respect Newmark is
in the equilibrium equation (2.11).

(2.11) Mün+1−αB + Cu̇n+1 + Kun+1 + f = 0

2
Where −31 < αB < 0, considering the Newmark parameters defined as γ = 12 − αB and β = (1−α
4
B)
. If αB = 0 we
obtain the Newmark scheme unconditionally stable. The Bossak method present some implementation advantages
for non-linear problems, owing to that the mass matrix is constant on Lagrangian meshes.

2.4.4.3 α-generalised
The α-generalised (Chung[ArtCH93]) is the generalisation of the time integration schemes, being the Newmark and
Bossak particular versions of him. In this case the method involves a modification of the equilibrium equation, with the
αf the damping and elastic forces are taken into a linear combination and with αm the inertial forces are modified.
With αm = αf = 0 the method obtained corresponds with Newmark. The expression of the method is presented in
(2.12).
The equations from (2.10) corresponding to un+1 and u̇n+1 remain exactly the same, except that the value the
Newmark constants which are γ = 12 + αm − αf and β ≥ 14 + 12 (αf − αm ) for an unconditionally stable solution.

(2.12a) Mün+1−αm + Cu̇n+1−αf + Kun+1−αf + f = 0

Where each component of the displacement and its derivatives is calculated as:


n+1−αm
ü
 = (1 − αf )ün+1 + αf ün
(2.12b) u̇n+1−αf = (1 − αm )u̇n+1 + αm u̇n

 n+1−αf
u = (1 − αm )un+1 + αm un

Vicente Mataix Ferrándiz Page 43 of 374

2.4 Finite Element formulation for CSD CHAPTER 2. FINITE ELEMENT FORMULATION

2.4.4.4 BDF
The BDF is a family of implicit methods for numerical integration of ODE, originally proposed by Curtiss[ArtCH52].
This family of methods is commonly used in stiff 3 ODE. Despite the method can be applied to any type of ODE, if
applied directly to the Newton’s second laws we can obtain the following (2.13) for the BDF1 (backward Euler method)
and BDF2 schemes.
The general formula of the BDF, for a variable depending on time, can be written as:

dy
= f (t, y ), y (t0 ) = y0
dt
(2.13a) s
X
ak yn+k = ∆t β f (tn+s , yn+s )
k =0

In here s corresponds with the desired order of the scheme, and ∆t with the time increment. The ak and β parameters
are obtained via Lagrange interpolation polynomials in order to achieve O(h)s . The BDF1 and BDF2 coefficients
correspond with:

BDF1: yn+1 − yn = ∆tf (tn+1 , yn+1 )

(2.13b) 4 1 2
BDF2: yn+2 − yn+1 + yn = ∆tf (tn+2 , yn+2 )
3 3 3
Applying this to Newton’s second laws, we obtain and to our system of equations (LHS and RHS). The definition of un+1
depends on the BC of the problem, and can be deduced from (2.13b) or by means of a second-order approximation
2
like un+1 = un + ∆t u̇n + ∆2t ün .
s
dy n+1 X
(2.13c) = bdfi y n+1−i
dt
i=0

( (
1 1 3 2 1
ün+1 = ∆t u̇
n+1
− ∆t u̇
n
ün+1 = 2∆t
u̇n+1 − ∆t u̇
n
+ 2∆ t
u̇n−1
(2.13d) BDF1: 1 1
BDF2: 3 2 1
u̇n+1 = ∆t u
n+1
− ∆t u
n
u̇n+1 = 2∆t
un+1 − ∆t u
n
+ 2∆ t
un − 1

( (
1 1 9 3
LHS = K + ∆t 2 M + ∆t C LHS = K + 4∆t 2
M + 2∆ t
C
(2.13e) BDF1: BDF2:
RHS = f − Mü − Cu̇ RHS = f − Mü − Cu̇

In addition, in case an adaptive time step (∆t) is considered we need to update the BDF coefficients. If considering
tn
ρ = ∆∆tn+1 , then:
ρ ρ
BDF1: bdf0 = , bdf1 = −
∆t ∆t
(2.13f) 1
BDF2: tcoeff = , bdf0 = tcoeff (ρ2 + 2ρ), bdf1 = −tcoeff (ρ2 + 2ρ + 1), bdf2 = tcoeff
∆t + ∆t ρ

2.4.5 Damping matrix

The damping matrix presented in (2.4e) depends on the damping coefficient (c) of the given element e, which is
not always available. Usually a critical-damping ratio respect the critical is considered (ξ ). With this critical-damping
ratio we can estimate the damping matrix. In order to do so we will consider the classical Lord Rayleigh[BookRay77]
damping. In this formulation, the damping matrix (C) is assumed to be proportional to the mass (M) and stiffness (K)
matrices as in Equation (2.14a).

(2.14a) C = ηM + δK
3 ODE unstable unless a small time increment is taken into account.

Page 44 of 374 Vicente Mataix Ferrándiz

CHAPTER 2. FINITE ELEMENT FORMULATION 2.4 Finite Element formulation for CSD

Figure 2.5: Graphic representation of Rayleigh damping matrix

In here η is the mass-proportional damping coefficient and δ are the stiffness-proportional damping coefficient. Taking
into consideration that the properties of the modal equations and the orthogonality conditions, we can express (2.14a)
as (2.14b).

1 ωn
(2.14b) ξn = η+ δ
2ωn 2

In this expression we have the relationship between the n critical-damping ratio and the n natural frequency. Consider-
ing two different natural frequencies, we can define a system of equations such as:
" #
  1 
ξi 1 ωi ωi η
(2.14c) = 1
ξj 2 ωj ωj δ

If damping for both frequencies is set to be equal, then the conditions associated with the proportionality factors can
be simplified as:

2ξ
(2.14d) ξi = ξj = ξ therefore δ= and η = ωi ωj δ
ωi + ωj

Vicente Mataix Ferrándiz Page 45 of 374

2.4 Finite Element formulation for CSD CHAPTER 2. FINITE ELEMENT FORMULATION

Page 46 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookBHS07] Encyclopedia of Computational Mechanics (3 Volume Set). René de Borst, T.J.R. Hughes, and
Erwin Stein. Wiley. 1st ed. 2007.
[BookBor+12] Nonlinear Finite Element Analysis of Solids and Structures, 2nd edition. René de Borst,
Mike A. Crisfield, Joris J. C. Remmers, and Clemens V. Verhoosel. Wiley. 2nd Edition. 2012.
[BookCP12] Dynamics of Structures. Ray W. Clough and Joseph Penzien. McGraw-Hill. 3rd ed. 2012. P. 752.
[BookFin13] The method of weighted residuals and variational principles. Bruce A Finlayson. SIAM. 2013.
[BookLL92] Numerical methods for conservation laws. Randall J LeVeque and Randall J Leveque. Springer.
1992.
[BookNPO09] Computational Methods for Plasticity Theory and Applications. E. A. de Souza Neto, D. Periæ, and
D.R.J. Owen. Wiley. 2009.
[BookOll14] Nonlinear Dynamics of Structures. Sergio Oller. Springer International Publishing. 2014.
[BookRay77] Theory of Sound. Lord Rayleigh. 1877. P. 93.
[BookRR12] Elements of Structural Dynamics: A New Perspective. G. Visweswara Rao and Debasish Roy. 2012.
[BookWri08] Nonlinear finite element methods. Peter Wriggers. Springer. 1st ed. 2008.
[BookZTF14] The Finite Element Method for Solid and Structural Mechanics. O. C. Zienkiewicz, Robert L. Taylor,
and David Fox. Butterworth-Heinemann. 7th ed. 2014.
[BookZZT13] The Finite Element Method: its Basis and Fundamentals. O. C. Zienkiewicz, J.Z. Zhu, and
Robert L. Taylor. Butterworth-Heinemann. 7th ed. 2013.

Articles
[ArtCGP06] “High order finite volume schemes based on reconstruction of states for solving hyperbolic systems
with nonconservative products. Applications to shallow-water systems”. Manuel Castro,
José Gallardo, and Carlos Parés. In: Mathematics of computation. No. 255, Vol. 75, 2006,
pp. 1103–1134. DOI: 10.1090/S0025-5718-06-01851-5.
[ArtCH52] “Integration of stiff equations”. CF Curtiss and Joseph O Hirschfelder. In: Proceedings of the
National Academy of Sciences of the United States of America. No. 3, Vol. 38, 1952, p. 235.
National Academy of Sciences. DOI: 10.1073/pnas.38.3.235.
[ArtCH93] “A time integration algorithm for structural dynamics with improved numerical dissipation: the
generalized-α method”. Jintai Chung and GM Hulbert. In: Journal of applied mechanics. No. 2, Vol.
60, 1993, pp. 371–375. American Society of Mechanical Engineers. DOI: 10.1115/1.2900803.
[ArtClo60] “The finite element method in plane stress analysis”. Ray W Clough. In: Proceedings of 2nd ASCE
Conference on Electronic Computation, Pittsburgh Pa., Sept. 8 and 9, 1960. 1960,

Vicente Mataix Ferrándiz Page 47 of 374

ONLINE RESOURCES BIBLIOGRAPHY

[ArtCri83] “An arc-length method including line searches and accelerations”. M. A. Crisfield. In: International
Journal for Numerical Methods in Engineering. No. 9, Vol. 19, 1983, pp. 1269–1289. John Wiley &
Sons, Ltd. DOI: 10.1002/nme.1620190902.
[ArtFM93] “Geometrical interpretation of the arc-length method”. M. Fafard and B. Massicotte. In: Computers
& Structures. No. 4, Vol. 46, 1993, pp. 603–615. DOI: 10.1016/0045-7949(93)90389-U.
[ArtGal15] “Series solution of some problems of elastic equilibrium of rods and plates”.
Boris Grigoryevich Galerkin. In: Vestn. Inzh. Tekh. Vol. 19, 1915, pp. 897–908.
[ArtPom69] “K. Washizu, Variational Methods in Elasticity and Plasticity.(International Series of Monographs in
Aeronautics and Astronautics). X+ 348 S. m. Fig. Oxford/London/Edinburgh/New
York/Toronto/Sydney/Paris/Braunschweig 1968. Preis geb. 120 s. net”. W Pompe. In:
ZAMM-Journal of Applied Mathematics and Mechanics/Zeitschrift für Angewandte Mathematik und
Mechanik . No. 5, Vol. 49, 1969, pp. 319–319. Wiley Online Library. DOI:
10.1002/zamm.19690490535.
[ArtRic11] “IX. The approximate arithmetical solution by finite differences of physical problems involving
differential equations, with an application to the stresses in a masonry dam”. Lewis Fry Richardson.
In: Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a
Mathematical or Physical Character . No. 459-470, Vol. 210, 1911, pp. 307–357. The Royal Society
London.
[ArtTL98] “A user-controlled arc-length method for convergence to predefined deformation states”. J. G. Teng
and Y. F. Luo. In: Communications in Numerical Methods in Engineering. No. 1, Vol. 14, 1998,
pp. 51–58. John Wiley & Sons, Ltd. DOI:
10.1002/(SICI)1099-0887(199801)14:1<51::AID-CNM130>3.0.CO;2-L.
[ArtTur56] “Stiffness and deflection analysis of complex structures”. MJ Turner. In: journal of the Aeronautical
Sciences. No. 9, Vol. 23, 1956, pp. 805–823. DOI: 10.2514/8.3664.
[ArtWBZ80] “An alpha modification of Newmark’s method”. WL Wood, M Bossak, and OC Zienkiewicz. In:
International Journal for Numerical Methods in Engineering. No. 10, Vol. 15, 1980, pp. 1562–1566.
Wiley Online Library. DOI: 10.1002/nme.1620151011.

Online resources
[OnlFin] Wikipedia. Finite strain theory. URL :
http://en.wikipedia.org/wiki/Finite_strain_theory.
[OnlGal] Wikipedia. Galerkin method. URL : http://en.wikipedia.org/wiki/Galerkin_method.
[OnlWol] Mathematica, Version 11.3. Wolfram Research Inc.

Page 48 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

Chapter 3

Rotation-free shells and solid-shell

elements

Things should be described as

simply as possible, but no simpler.

Albert Einstein
(1879 - 1955 AD, German-born
theoretical physicist)

3.1 Introduction
In first place we will introduce a simple historical outline in the shell theory. This will be extended in the state-of-the-art
section (3.2.State of the Art in shell formulation), the current state of the standard shell theory. The theories relative to
rotation-free formulation will be introduced in its respective sections.

In the following sections, we will introduce the element solid-shell developed to tackle the main aim of this work,
the prismatic solid-shell element. This element has some resemblance with a previous work, the Enhanced Basic
Shell Triangle (EBST) element, a rotation-free shell, which considers the neighbour elements with the objective of
enrich quadratically the in-plane behaviour.

For that reason we will introduce in first place the rotation-free elements (3.3.Rotation-free shells), and later on we
will present the prismatic solid-shell element (3.4.Prismatic solid-shell). The examples of performance of the prismatic
solid-shell element can be found in the section concluding this chapter to this one, 3.5.Numerical examples.

3.1.1 Historical outline

The base of the shell theories can be found in the plate theory. This theory takes a similar approach to the beam
theory applied in thin geometries. The classic thin plate theory establishes that the normal remains straight and
orthogonal to the middle plane after deformation[BookOña13]. This theory is based in the postulates presented in
the original work of Kirchoff in 1850[ArtKir50], despite some people associate this theory to some previous works
by Marie Sophie Germain[BookBD80; BookBD12] in 1811. We have some analytical solutions for this theory for
rectangular plates, as the one presented by Navier in 1820. In here Navier navier1820memoire introduced a simple
method for finding the displacement and stress when a plate is simply supported. The idea was to express the
applied load in terms of Fourier components, find the solution for a sinusoidal load (a single Fourier component),
and then superimpose the Fourier components to get the solution for an arbitrary load. Another analytical solution
was proposed by Lévy [ArtLev99] in 1899, where the main difference with the Navier solution consists in start with an

Vicente Mataix Ferrándiz Page 49 of 374

3.2 State of the Art in shell formulation CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

assumed form of the displacement and try to fit the parameters so that the governing equation and the boundary
conditions are satisfied.
An evolution to this theory, which can be applied to thick plates, is the thick plate theory proposed by Reiss-
ner [ArtRei45] and Mindlin[ArtMin51]. This theory assumes that normals remain straight, though not necessarily
orthogonal to the middle plane after deformation. These two theories are complementary, and it is relevant to know
when to choose between them depending on the relative thickness of the element (usually we consider as thin
thickness 1
elements when average side
≤ 10 ).
The shell theory will be therefore an extension of the previous postulates, but applied to non-planar sur-
faces[BookOña13]. The non-coplanarity introduces membrane behaviour (axial force) in addition to the plate
flexural resistance (bending and shear forces). The governing equations of a curved shell (equilibrium and kinematic
equations, etc.) are more complex[BookKra67; ArtNio85; ArtNoo90a; BookTW59; ArtNoo90b] than in the case of
plates, mainly due to the curvature of the middle surface. This theory is introduced in first place by Love[ArtLov88] in
1888, applied only for thin shells. This short introduction to the history of shell theory will be expanded in the shell
state-of-the-art section at 3.2.State of the Art in shell formulation, where some of these standard shell models will be
presented.

3.2 State of the Art in shell formulation

3.2.1 Element requisites and stabilisation techniques
As it is presented in the book of Wriggers[BookWri08], but also in the work of Belytschko[BookBel+14], the target
properties of any new element, extensible to shell elements are the following1 :

a) Locking free behaviour for incompressible materials

b) Good bending performance
c) No locking in thin elements
d) No sensitivity against mesh distortions
e) Good coarse mesh accuracy,
f) Simple implementation of nonlinear constitutive equations
g) Efficiency (particularly relevant in explicit simulations)

In order to achieve this objective there are different techniques[BookWri08], here we present a short summarises
of the most relevant:

1. Reduced integration and stabilisation: Consist in the consideration of reduced number of integration points,
with the corresponding reduction in the computational cost and memory, especially in explicit algorithms. This
methodology was developed in order to avoid locking in case of incompressibility. The principle from which
derives this technique is the fact that reduced integration is associated with rank deficiency of the tangent
matrices, which is cured with the consideration of stabilisation techniques. Because of this these elements need
the choice of artificial stabilisation parameters. Its main advantages are:

• Locking free for incompressibility

• Good coarse mesh accuracy
• Not sensitive against mesh distortions
• Can be used for arbitrary constitutive equations

2. Hybrid or mixed variational principles: These elements include additional Degree Of Freedom (DOF)
in order to stabilise the formulation. The main disadvantage is the requirement of adapting the constitutive
relationships.
1 As we will see the element developed in this work, 3.4.Prismatic solid-shell, enjoy these advantages.

Page 50 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.2 State of the Art in shell formulation

3. Enhanced strain elements based on the Hu-Washizu principle: Within the enhanced strain formulations,
nonconforming strain measures are introduced within the Hu-Washizu principle[ArtWas82]. The main elements
developed concerning this principle are the ones present in the work of Simo[ArtSF89; ArtSFR89; ArtSFR90;
ArtSRF90; ArtSK92; ArtSRF92; ArtSim93]. These elements fulfil all points of the previously mentioned
requirement, therefore they are well suited for all applications. Additionally these methods have been proven to
be efficient in order to overcome the hour glassing effect. This method includes techniques such as EAS and
Assumed Normal Strain (ANS).
4. Mixed variational principles for problems with rotational DOF: In here the momentum and moment of
momentum are weakly enforced. This leads to the symmetry of the stress tensor and therefore the rotational
DOF can be introduced as independent field variables.
5. Composite or macro elements: Here elements are constructed from subelements, which use simplified or
special shape functions.
6. Higher order displacement elements: This technique consists in the consideration of higher order geometries
as quadratic or cubic, or even hierarchical shape functions using polynomials or Non-Uniform Rational B-
spline (NURB). In fact, this last option allows a more efficient formulation, being competitive with the other
approaches.

In Wriggers[BookWri08], some other techniques are mentioned, such as nodal based elements or Cosserat point
elements.

3.2.2 Shell formulations

We can differentiate three different approaches[BookOña13] for the definition of a shell theory:

1. Classical shell theories: These formulations can be found summarised in several sources[BookNov59; Art-
Nio85; BookWem82; BookGou12; BookCal89]. We can mention particularly the Discrete-Kirchhoff assumptions,
as an extensively used methodology[ArtWOK68; ArtWem69; ArtDha69; ArtDha70; ArtDMM86; ArtMG86].
2. 3D elements of small thickness: This one it is not practical as it requires the consideration of very small
elements in order to avoid the ill-conditioning coming from a slender aspect ratio of the shell geometry.
3. Shell elements by degeneration of solid elements: These formulations, aka Degenerated Shell (DS),
comes from the concept of enforcing the shell theory assumptions into the general 3D elasticity[ArtAIZ70]. This
denomination can be misleading and it is often used in order to differentiate the formulation from curved shell
elements derived from standard shell theory. This kind of elements have enjoyed a lot of popularity, and several
authors have significant contributions in this formulation[BookBD90; ArtPar79; ArtKan79; ArtHL81; ArtFV82].
These kinds of elements, as a generalisation of the ReissnerMindlin shell theory suffer of membrane locking and
therefore require of some kind of stabilisation as the ones presented in 3.2.1.Element requisites and stabilisation
techniques, particularly reduced integration or assumed strain fields. The main difference between the DS and
the standard formulation is that in the case of the DS the Principle of Virtual Work (PVW) is written in terms
of stresses and strains, as in solid elements, and the integrals are computed in the volume.
• Continuum-Based Resultant (CBR) shell theory: This theory introduces simplifications in DS elements,
so that the PVW can be expressed in terms of resultant stresses and generalised strains, therefore the
integrals can be computed over the shell surface only. It exists a variation of this formulation called CBR-S,
whereas an assumption we eliminate the z-dependence in the surface Jacobian (J).

Buechter and Ramm[ArtBR92] have compared between degenerated shell elements and those based on standard
Reissner-Mindlin/Naghdi type shell theory. In this work, they highlight the standard shell for practical purposes as
their formulation is generally simpler.
In addition, it is relevant to highlight the so-called . As mentioned in the stabilisation techniques in 3.2.1.Element
requisites and stabilisation techniques, this methodology consists in the consideration of the same basis functions for
representing the geometry in CAD, like the NURB[BookHug+14; BookFHK02; BookPT97; BookRog00]. Thanks to
this, the FE analysis works on a geometrically exact model and no meshing is necessary[BookHBC09; ArtBen+10;
ArtHCB05].

Vicente Mataix Ferrándiz Page 51 of 374

3.3 Rotation-free shells CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

3.3 Rotation-free shells

3.3.1 Introduction
The next section introduces the rotation-free plates and shells developed in the last years by Eugenio Oñate et
al.[ArtOF05; ArtFO05; ArtFO01; ArtOZ00; ArtFO11]; that despite the modifications and improvements realised in
all this time, the principle is the same, a plate/shell that just consider displacement as degree of freedom in the
nodes.
The traditional Kirchoff theory is enough to describe the most common cases of plates and shells, but in the cases
where a thick shells approach is interesting to be considered, some difficulties appear associated to the C 12 continuity,
and then just cases which need C 0 continuity have been used; in this sense the rotation free theories have been
presented as a very interesting alternative.

3.3.2 Basic Shell Triangle

3.3.2.1 Basics
Let us consider a patch of four three node triangles from Figure 3.1a, where the main triangle corresponds with the
one with the nodes marked with circles, and the external ones the marked with squares. To develop the formulation
we assume a constant curvature field, then the deflection is interpolated linearly in a standard finite element manner
as (3.1a), and then calling the curvatures as (3.1b).

η
5 3 4
4

1 G2.. G1
3

2
M
1

3
2
. 1
G3
2
ξ

2 3
1
5
6 6
(a) Patch of three nodes triangular elements including the (b) Patch of elements in the isoparametric space
central triangle (M) and three adjacent triangles (1, 2 and 3)
Figure 3.1: Patch of elements for Basic Shell Triangle (BST) and EBST

3
X
(3.1a) w= Lei wie
i=1

Where Lei are the linear shape functions and wie the nodal deflection of each node.
 
 κxx 
(3.1b) κ= κyy = κ̂
 
κxy
Where κ is the curvature vector and κ̂ is the constant curvature field defined in (3.1c).
Z Z  T
1 ∂2w ∂2w ∂2w
(3.1c) κ̂ = − 2 , − 2 , −2 dA
AM AM ∂x ∂y ∂x ∂y
Where AM is the area of the central triangle.
2 The function f is said to be of (differentiability) class C k if the derivatives f ′ , f ′′ , ..., f (k ) exist and are continuous (the continuity is implied by

differentiability for all the derivatives except for f (k )).

Page 52 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.3 Rotation-free shells

3.3.2.2 Shell kinematics

Considering a shell of the initial thickness h0 and with an undeformable middle surface in Ω0 (ℜ3 ) with a boundary Γ0 .
The positions X and x of a point in the undeformed and the deformed configurations can be written as a function of
the coordinates of the middle surface ϕ and the normal t3 at the point as (3.2a), where ξ and η are the curvilinear
principal coordinates and ζ is the distance from the point to the middle surface in the undeformed configuration.

(
X(ξ , η , ζ ) = ϕ0 (ξ , η ) + λt03
(3.2a)
x(ξ , η , ζ ) = ϕ(ξ , η ) + ζλt3

Defining λ as the parameter that relates the thickness at the present and initial configuration (3.2b).

h
(3.2b) λ=
h0
A convective system is computed at each point as (3.3).
 ∂x
gi (ζ ) = ∂ξi i = 1, 2, 3

(3.3) gα (ζ ) = ∂ (ϕ(ξ∂ξ ,η )+ζλt3 )
α
= ϕ′ α + ζ (λt3 )′ α α = 1, 2
 ∂ (ϕ(ξ ,η )+ζλt3 )
g 3 (ζ ) = = λt3

∂ζ

The curvatures of the middle surface are obtained by (3.4a).

1

(3.4a) καβ = ϕ′ α · t3′ β + ϕ′ β · t3′ α = −t3 · ϕ′ αβ α, β = 1, 2
2
The deformation gradient tensor (F) is (3.4b).
 
(3.4b) F = [x′ 1 , x′ 2 , x′ 3 ] = ϕ′ 1 + ζ (λt3 )′ 1 ϕ′ 2 + ζ (λt3 )′ 2 λ t3

The product FT F = U2 = C (where U is the right stretch tensor, and C the right Cauchy-Green deformation tensor) can
be written as (3.4c).
   
a11 + 2κ11 ζλ a12 + 2κ12 ζλ 0 a11 + 2χ11 ζλ a12 + 2χ12 ζλ 0
(3.4c) U2 =  a12 + 2κ12 ζλ a22 + 2κ22 ζλ 0  → if κ0ij = 0 →  a12 + 2χ12 ζλ a22 + 2χ22 ζλ 0 
0 0 λ2 0 0 λ2

It is also useful to compute the eigendecomposition of U as (3.4d), where λα and rα are the eigenvalues and
eigenvectors of U.
3
X
(3.4d) U= λα rα ⊗ rα
α=1

3.3.2.3 Constitutive models

For the constitutive model, we recommend to consult directly the literature[ArtOF05; ArtFO05; ArtFO01; ArtOZ00;
ArtFO11]. It is possible to consider more than one constitutive model, and the ones presented in the references are
focused in cases where the objective is to calculate large strains of metal forming processes.

3.3.3 Enhanced Basic Shell Triangle

The following model corresponds with the enhanced version of the BST, called EBST, being the main features of this
element:

• The geometry of the patch formed by the central element and the three adjacent elements is quadratically
interpolated from the position of the six nodes in the patch.

Vicente Mataix Ferrándiz Page 53 of 374

3.3 Rotation-free shells CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

• Interpolated from the position of the six nodes in the patch. The membrane strains are assumed to vary linearly
within the central triangle and are expressed in terms of the (continuous) values of the deformation gradient at
the mid-side points of the triangle.

• The assumed constant curvature field within the central triangle is obtained using now twice the values of the
(continuous) deformation gradient at the mid-side points.

3.3.3.1 Total Lagrangian formulation

As mentioned above a quadratic approximation of the geometry of the four elements patch is chosen using the position
of the six nodes in the patch. It is useful to define the patch in the isoparametric space using the nodal positions given
in the Table 3.1 and Figure 3.1b. The quadratic interpolation is defined as (3.5a).

6
X
(3.5a) ϕ= Ni ϕi
i=1

N1 = ζ + ξη N4 = ζ2 (ζ − 1)
(3.5b) N2 = ξ + ηζ N5 = ξ2 (ξ − 1) with ζ = 1 − ξ − η
N3 = η + ζξ N6 = η2 (η − 1)

The Cartesian derivatives are defined as (3.5c).

      
Ni,1 −1 Ni,ξ ϕ0′ ξ · t1 ϕ0′ η · t1 Ni,ξ
(3.5c) =J =
Ni,2 Ni,η ϕ0′ ξ · t1 ϕ0′ η · t2 Ni,η

1 2 3 4 5 6
0 1 0 1 -1 1
ξ
0 0 1 1 1 -1
η
Table 3.1: Isoparametric coordinates of the six nodes in the patch of Figure 3.1b

3.3.3.1.1 Computation of the membrane strains :

We define the Green-Lagrange strains as (3.6a), and the virtual membrane strains as (3.6b). Differently from the
original BST element the membrane strains within the EBST element are now a function of the displacements of the
six patch nodes.
 i 
3
X 1  ϕ′ 1 · ϕi′ 1 − 1 
(3.6a) ǫm = N̄i ϕi′ 2 · ϕi′ 2 − 1
2 
2ϕi′ 1 · ϕi′ 2

i=1

 i

3
X  ϕi′ 1 · δ ′ 1 
(3.6b) δǫm = N̄i ϕi′ 2 · δϕi′ 2 = Bm δ ap
δϕi′ 1 · ϕi′ 2 + ϕi′ 1 · δϕi′ 2
 
i=1

3.3.3.1.2 Computation of the bending strains :

We employ the same curvature field assumed in the BST.
 
3 LM 0  
X
 0
i,1
M  t3 · ϕi′ 1
(3.7a) κ=2 Li,2
t3 · ϕi′ 2
i=1 LM
i,2 LM
i,1

Page 54 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.4 Prismatic solid-shell

The gradient is computed as (3.7b).

 
    ϕ1
i i i i
ϕi′ 1 N1,1 N2,1 N3,1 Ni+3,1  ϕ2 
(3.7b) =  
ϕi′ 2 i
N1,2 i
N2,2 i
N3,2 i
Ni+3,2  ϕ3 
ϕi+3

Finally, the variation for curvature is (3.7c).

 (
3 LM 0 3  i

  i
)
X i,1 X Nj,1 t3 · δ u j
Ni+3,1 t3 · δ ui+3

δκ =2  0 LM  i + i
i,2
Nj,2 t3 · δ u j Ni+3,2 t3 · δ ui+3
i=1 LM
i,2 LM
i,1 i=1
(3.7c)  M 1

X 3 Li,1 ρ11 + LM
i,2 ρ11

−  LM 1 M
i,1 ρ22 + Li,2 ρ22

 (t3 · δ ui ) = Bb δ ap
i=1 LM 1 M
i,1 ρ12 + Li,2 ρ12

3.3.3.1.3 Tangent stiffness matrix :

The tangent stiffness matrix is divided into material and geometric components. Being the material tangent matrix
the one showed in (3.8a), and the geometric tangent matrix the one from (3.8b).
Material stiffness matrix:
Z Z
M
(3.8a) K = BT Dep BdA where B = Bm + Bb
A0M

Geometric stiffness matrix:

(3.8b) KG = KG G
m + Kb

Where:
3 6 6     
AM X X X  k  k
N11 k
N12 k
Nj,1
(3.8c) δ uT KG
m ∆u = δ ui Ni,1 k
Ni,2 k k k ∆ uj with Nij = σmij
3 N21 N22 Nj,2
k=1 i=1 j=1

Numerical experiments[ArtFO05; ArtFO11] have shown that the bending part of the geometric stiffness is not so
important and can be disregarded in the iterative process.

3.4 Prismatic solid-shell

3.4.1 Introduction
Solid-shells have been during the last fifteen years[ArtDB84; ArtHS98; ArtHSD00; ArtHau+01; ArtSou+05; ArtKGW06a;
ArtPar+06] an important improvement in the shells simulations, providing reliable simulations and avoiding the prob-
lematic that is associated to the kinematics hypothesis and plane stress constitutive laws related to the use of the
shell element (or any kind of bidimensional elements as plates and membranes). The main advantages when using
solid-shell elements are:

a) General 3D constitutive relations

b) Large transverse shear can be considered, and considering additional elements along the thickness improve
this behaviour

c) There does not need to consider transitions between solid and shell elements (all the elements are solids)

d) Contact forces can be introduced directly in the geometry and in a realistic way without any additional technique,
which is especially important for the consideration of friction

Vicente Mataix Ferrándiz Page 55 of 374

3.4 Prismatic solid-shell CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

e) The element is rotation-free, avoiding the storage and computation of these variables

f) In the case where we have non-parallel boundaries this can be modelled correctly

For the consideration of strongly non-linear problems, problems where the contact-friction, large deformations
and complex constitutive laws are considered, the use of low interpolation order elements is preferred. Most of the
existing solid-shells are linear hexahedron[ArtDB84; ArtHS98; ArtHSD00; ArtHau+01; ArtSou+05; ArtKGW06a;
ArtPar+06; ArtSVR11; ArtAC09; ArtSAV11; ArtSR11; ArtSen+16] (usually trilinear 8-node brick), which have two main
disadvantages; the first one is the hourglass effect, which is called this way due to the characteristics shapes adopted
in the proper modes and a stabilisation is required to reduce these problems; the second problematic is the meshing
of the plane, due to the fact that meshing quadrilateral is less performant than triangles. For this reasons the triangular
prisms (wedges) could be considered an interesting alternative, especially for the second problem mentioned, but this
kind of geometry is not exempt of problematic, owing to the low order of interpolation of the geometry, when a linear
triangular prism is considered. This last problem can be solved with the consideration of the neighbours elements3 , in
consequence the element becomes quadratic in the plane solving this last problem.
In addition to this, it is well known that low interpolation order elements in the standard displacement formulation
(in contract to the mixed formulation) in the consideration of slender structures and incompressible materials suffer
severe locking effects. The transverse shear locking provokes problems in the bending behaviour, especially when
more slender is the element. The membrane locking appears especially in the initially curved shells when bending is
preponderant without middle surface stretching. A curvature thickness locking can appear in problems with initially
curved geometry due to artificial transverse strains and stresses under pure bending. Finally, the volumetric locking
can appear when the material present an incompressible, nearly incompressible behaviour or elastic-plastic materials
with isochoric plastic flow (typical in metals).
The element has been implemented into Kratos[OnlKra], the in-home FEM-Multiphysics open-source code,
implemented in C++ with parallelisation capabilities. The pre/post-process of all the presented examples haven been
processed with GiD[OnlMel+16], the CIMNE software for pre and post processing.

3.4.2 Basic kinematics of the standard element

We can find the standard isoparametric interpolations for the linear 6-node triangular prism (or wedge) in (3.9), where
XI , xI and uI correspond with the original coordinates (or undeformed configuration), the current coordinates (or
deformed configuration) and the displacements in the node I respectively.

6
X
(3.9a) X(ξ ) = N I (ξ )XI
I=1

6
X 6
X

(3.9b) x(ξ ) = N I (ξ )xI = XI + uI
I=1 I=1

The shape function N I (ξ ) from (3.10) are defined in function of the local coordinates (ξ , η , ζ ), where the tho first
define the position in the plane of the triangular base and the third one corresponds with the coordinate along the
prism axis.

N 1 = zL1 , N 2 = ξ L1 , N 3 = η L1
(3.10a)
N 4 = zL2 , N 5 = ξ L2 , N 3 = η L2

In this definition the third triangular coordinate and the axis interpolation is defined as:

(3.10b) z = 1 − ξ − η, L1 = 21 (1 − ζ ), L2 = 12 (1 + ζ )
3 Like in the EBST elements, see 3.3.Rotation-free shells.

Page 56 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.4 Prismatic solid-shell

We can define following the standard formulation the Jacobian matrix at each integration point as (3.15a), in
consequence we can compute the Cartesian derivatives of the shape functions. At each element centre, a local
Cartesian triad can be defined as in (3.11b), that allows to compute the Cartesian derivatives with respect to this local
system following the orthotropic directions.

∂X
(3.11a) J= → NXI = J−1 NξI
∂ξ

 
(3.11b) R= t1 , t2 , t3 → NYI = RT NXI

As it will be shown in the following section, the left Cauchy tensor C is modified into the C̄ using the assumed
strain techniques that in one case include an additional internal degree of freedom α, leading to the improved tensor
C̄. The balance equation to solve in the strong form for a TL4 formulation in large strain hypothesis is (3.12).

( R 1
g1 (u, α) = S(C̄) : δu C̄dV0 + gext = 0
(3.12) RV0 12
g2 (u, α) = V0 2 S(C̄) : δα C̄dV0 + gext = 0

Where S is the second Piola-Kirchoff stress tensor (PK2).

3.4.3 Modifications of the standard element

The prism presented in the previous lines corresponds with the standard, and it needs to be modified for the sake
of large strain elastic-plastic analysis of shells. For this purpose different modifications are introduced in the metric
tensor C. The discretisation to be introduced in the prism solid-shell can be defined in first place with the discretisation
of the triangular middle surface and in second place with the discretisation along the thickness. We will assume that
the upper and lower face are almost parallel and thus the normal direction can be defined in function of ζ . In (3.13)
the right Cauchy-Green tensor is decomposed according to the different behaviours that define classically the shell
elements, this means we decompose in-plane (membrane and bending behaviour), transverse shear and normal
components. The calculation of each one of the different components will be detailed in the next sections.

m
 m s

C11 C12 C13
m m s 
(3.13a) C=  C21 C22 C23
m m n
C31 C32 C33

Where the index m, s and n mean membrane, shear and normal behaviour, respectively. In consequence this tensor
can be decomposed in three components as detailed in (3.13b).

1 1 2 2 1 2
C1 = C11 t ⊗ t + C22 t ⊗ t + C12 (t ⊗ t + t ⊗ t ) (tangent plane)
 2 1

(3.13b) 1 3 3 1 2 3
C = C1 + C2 + C3 → C2 = C13 (t ⊗ t + t ⊗ t ) + C23 (t ⊗ t + t ⊗ t ) (transverse shear)
3 2

C2 = C33 t3 ⊗ t3 (thickness strain)


3.4.3.1 In plane behaviour

The improvement in the in-plane behaviour is the same considered for the EBST element, a rotation-free shell element,
where a four-element patch is considered (Figure 3.2), and the neighbour nodes allow us to work with a quadratic
element in the in-plane behaviour; for more information about this element, we address the reader to[ArtOF05;
ArtFO05]. The same computations from the EBST element can be considered in the upper and lower face as we can
see in the quadratic shape functions from (3.14) and it derivatives in Table 3.2.
4 The formulation can be extended to UL following the standard notation, for more information you can look among others[BookBel+14;

BookBor+12].

Vicente Mataix Ferrándiz Page 57 of 374

3.4 Prismatic solid-shell CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

8 3 7

1 2

9
(a) 2D (b) 3D
Figure 3.2: Patch performed in the element considering the neighbour elements

N 1 = (z + ξη ), N 7 = z2 (z − 1),
(3.14) N 2 = (ξ + η z), N 8 = ξ2 (ξ − 1),
N 3 = (η + z ξ ), N 9 = η2 (η − 1)

We define a local system of coordinates from (3.11b) taking as reference the components f1 and f2 in the tangent
plane and f3 in the normal. In each mid-side point of the element, we compute the in-plane Jacobian as shown in
(3.15a).
 
X ξ · t1 Xη · t 1
(3.15a) J=
X ξ · t2 Xη · t 2

And in combination of the shape function derivatives we can compute the Cartesian derivatives in (3.15b).
 K  K
N1I −1 NξI
(3.15b) = JK
N2I NηI

With the Cartesian derivatives, we can calculate the in-plane deformation gradient components fK1 and fK2 , and
with it CijK which are averaged over each integration point along the thickness with (3.16a). As with the rotation-free
shells, when a neighbour is missing the values from the central node of the face are used for the averaging.

(3.16a) C̄ij (ξ ) = L1 C̄ij1 + L2 C̄ij2

The variations are computed as (3.16b).

 1
  1 1
  1 2
  
C̄
2 11
C̄
2 11
C̄
2 11
Ē11
(3.16b) δ 1
C̄  = δ 1 1
C̄  L1 + δ  1 2
C̄  L2 = δ  Ē22 
2 22 2 22 2 22
1 2
C̄12 C̄12 C̄12 2Ē12

1 2 3 7 8 9
NξI −1 + η 1 − η z − ξ 21 − z ξ − 12 0
NηI −1 + ξ z − η 1 − ξ 21 − z 0 η − 12
Table 3.2: Derivatives of the shape functions from (3.14)
At each face a modified tangent matrix B̄f relating the incremental tensor components with the incremental
displacements δ u can be written as shown in (3.17), where just the nodes from the face and its opposite neighbour is

Page 58 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.4 Prismatic solid-shell

considered.

     J(K ) 
1 f
C̄
2 11
3
1 K
C̄
2 11
3 4 fK1 N1
1 f 1X 1 K 1 XX J(K )  δ uJ(K )
δ C̄
2 22
= δ C̄
2 22
=  fK2 N2
(3.17) f 3 K 3 J(K ) J(K )
C̄12 K =1 C̄12 K =1 J=1 (fK1 N2 + fK2 N1 )

   
= B̄fm 3x18
δ uf → B̄m 3x36
= L1 B̄1m L2 B̄2m

R1
We can obtain the equivalent nodal force vector from the integral (3.18), with −1
Sij Lf Jd ξ = S̄ijf .

 T  T  2 T 
1
Z 1S11  1 1 

 S̄11 S̄11


(3.18) rTm δ u =  S22  L B̄m L2 B̄2m Jd ξ =  1 
S̄22 1  2 
B̄m , S̄22 2
B̄m δ u
−1 S12  S̄ 1

S̄ 2


12 12

In (3.19) the calculation of geometric stiffness for the membrane behaviour is presented, where we sum the
contribution of the nG integration points along the direction ζ .

  1
 T  
Z C̄
2 11
S11
∂  
δ uT KmG ∆u = δ 1
C̄
2 22
  S22  ∆udV
V ∂u
(3.19a) C̄12 S12
nG 2 4 
3 X   K
X VolG X X   S11 S12 N1I
= Lf δ uI N1I N2I
3 S21 S22 N2I
G=1 f =1 K =1 J=1

2 X 4
3 X
( "n  #  K )
G
X
I
 K X VG f S̄11 S̄12 N1I J
δ uKmG ∆u = δu N1I N2I L ∆u
3 S̄21 S̄22 N2I
f =1 K =1 J=1 G=1
(3.19b) ( 3 4 ( ))f
2  f f
 K
X XX
I
 K S̄11 S̄12 N1I J
= δu N1I N2I ∆u
f
S̄21 f
S̄22 N2I
f =1 K =1 J=1

3.4.3.2 Transverse shear behaviour

To avoid the transverse shear locking we introduce an interpolation in natural coordinates of mixed tensorial compo-
nents, a very common practice in the literature. We consider a linear variation of the transverse shear strain tangent
to the side. Here we compute a mixed component of the right Cauchy-Green tensor as (3.20), where the components
relatives to the transverse shear Cη3 and Cξ3 are written with respect to a mixed coordinate system that includes the
in-plane natural coordinates (ξ , η ) and the spatial local coordinate in the transverse direction (y3 ). The components
are written in terms of the transverse shear strain tangent to the side computed in each mid-side point, as seen in
Figure 3.3. Besides, the numerical integration is performed along the axis, which means P( 13 , 31 ).

 √ 1
  √ 
    2Ct3 2f1t · f13
Cξ 3 −η −η 1−η 2
 −Cη3  = P(ξ , η )  −f2η · f23 
(3.20) =
Cη 3 ξ ξ−1 ξ
Cξ33 f3η · f33

The deformation gradient components are ft (natural coordinate derivative) and f3 (local Cartesian coordinate
T
derivative), these are expressed in (3.21a), where j−
3 are third column of the inverse of the transverse of the Jacobian.

 
    f1t · f13
C̄12 Cξ 3
(3.21a) = J−
p
1 −1
= Jp Pa  − f2η · f23 
C̄23 Cη 3
fξ · f33
3

Vicente Mataix Ferrándiz Page 59 of 374

3.4 Prismatic solid-shell CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

(a) 2D (b) 3D
Figure 3.3: Nodes considered in the computation of the transverse shear strains

The modified transverse shear Cartesian components can be obtained from (3.21b), interpolating these values the
components from the right Cauchy-Green tensor can be obtained (3.21c).
 √ 1    √ 2  
2f1t x3 − x2 2f1t x6 − x5 X6
T
(3.21b)  −f2η  =  x1 − x3  ,  −f2η  =  x4 − x6  , f3 = N3I xI = ∇ξ (x)j−
3
3 2 1 3 5 4
fξ x −x fξ x −x I=1

   1  2    
C̄13 C̄12 1 C̄13 2 2E13 C̄13
(3.21c) (ζ ) = L + L and (ζ ) = (ζ )
C̄23 C̄23 C̄23 2E23 C̄23

The tangent matrix B̄s is also obtained by interpolating from both faces (3.22a). In a similar way to the in-plane
behaviour, the internal forces can be obtained from (3.22b). Finally, the geometric stiffness matrix can be obtained
from (3.22c).

 

− 1 δ f1t · f13 + f1t · δ f13
2 2 2 2
(3.22a) B̄s (ζ ) = B̄1s L1 + B̄2s L2 where B̄fs = Jfp Pa B̃fs e
and B̃s δ u =  −δ fη · f3 − fη · δ f3 
δ f3ξ · f33 + f3ξ · δ f33

   
S̄12
Z     1 
Z L1 
S̄12   B̄1s S̄23 
(3.22b) rTs = B̄s dV = Q̄T4x1 where Q̄T4x1 =    Jd ζ
V S̄23 2x18 B̄2s −1
 S̄12 2

4x18 L
S̄23

  T
T B̄1  ′ ′    1
δ u KsG ∆u = ∆ δ u Q̄ considering Q̄1 Q̄2 = Q̄1 Q̄2 J− p →
B2s
  ′ ′
  √ 1 √  
(3.22c) −Q̄1 + Q̄2 2δ ft · ∆f13 + 2∆f1t · δ f13
 
1   ′ ′


Q̄1 Q̄2 ∆ B̄s1 δ u =  −Q̄1 − 2Q̄2 −δ f2η · ∆f23 + ∆f2t · δ f23


3  ′ ′
   
2Q̄1 + Q̄2 δ f3ξ · ∆f33 + ∆f3ξ · δ f33

3.4.3.3 Transverse normal behaviour

As introduced previously, to avoid locking (volumetric locking) in quasi-incompressible problems due to the Poisson
effect the EAS formulation is considered. With this formulation we obtain a modified C3 component.

Page 60 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.4 Prismatic solid-shell

3.4.3.3.1 EAS formulation :

In the standard EAS method the convective strain components are interpolated, in our case as we just want to
improve C3 and some modifications as presented below are considered. At the element centre (ξ = η = 13 and ζ = 0)
the Cartesian deformation gradient, and in consequence the right Cauchy tensor, can be enhanced as shown in
(3.23). So, in this EAS the changes will affect just to the C33 component, and the C13 and C23 are computed as
presented in the previous section.

6
X
(3.23) fC
3 = N3IC XI → f̄3 = fC
3e
αζ 2 2αζ
→ C̄33 = f̄3 · f̄3 = C33 e
I=1

With this enhancement the deformation matrix, internal forces and geometric stiffness can be calculated as shown
in (3.24).

6
!
1 X
(3.24a) δ Ē33 = δ C̄33 = δ fC C 2αζ
3 · f −3 e + C̄33 ζδα = N3IC δ uI · fC3 e2αζ + C̄33 ζδα = e2αζ BC3 δ ue + C̄33 ζδα
2
I=1

Z Z 1 Z 1
(3.24b) rTn = B3 S33 dV = e2αζ BC C
3 S33 Jd ζ = B3 S̄33 with S̄33 = e2αζ S33 Jd ζ
V −1 −1

 
6
X 6
X 1
(3.24c) δ uT KGn ∆u = δ fC3 · ∆fC3 S̄33 = ( δ uI ) T N3IC N3JC S̄33  1  ∆ uJ
I=1 J=1 1

3.4.3.3.2 Balance equation :

In the following lines, we will introduce the balance equation which allows us to obtain the implicit solution of
the problem. Because we will focus in the implicit solution of the problem we will address the reader to[ArtFlo13c;
ArtFlo13a; ArtFlo13b] where the explicit solution of the problem is presented too.
The balance equation related with α DOF from (3.12) can be expressed as (3.25a), where we can obtain the
residue and approximate the solution with a Newton-Rapson technique the residual can be nullified (3.25b).
Z 1 Z 1
(3.25a) δα S33 C̄33 ζ Jd ζ = 0 → rα = S33 C̄33 ζ Jd ζ
−1 −1

  (
Z 1 ∂ S33 C̄33
∂ S33 C̄33 ∂ S33 C̄33 = C̄33 D3 B̄ + 2S33 B̄3
(3.25b) ∆u + ∆α ζ Jd ζ + rα = 0 with ∂ S∂33uC̄33

−1 ∂u ∂α ∂α = C̄33 D33 C̄33 ζ + 2S33 ζ

The previous expression can be condensed in the following expression(3.25c), where after integrate the operators H
and kα are obtained.

rα 1
(3.25c) H1x18 ∆u + kα ∆α + rα = 0 then ∆α = − H∆u
kα kα

Where:
( R1

H1x18 = −1
C̄33 D3 B̄ + 2S33 B̄3 ζ Jd ζ
(3.25d) R1

kα = −1
C̄33 C̄33 D33 + 2S33 ζ 2 Jd ζ

Vicente Mataix Ferrándiz Page 61 of 374

3.4 Prismatic solid-shell CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

If the expression is introduced in the balance equation associated with the displacement in (3.12) we can obtain
(3.25e), which can be linearised in the expression (3.25f).
Z
(3.25e) δ uT B̄T SdV − δ uT Gext = δ uT r(u, α)
V

Z 1     T  
∂S ∂S ∂ B̄ ∂ B̄T
BT ∆u + ∆α + ∆u + ∆α S Jd ζ + r(u) = 0
−1 ∂u ∂α ∂u ∂α
Z 1 

BT DB̄∆u + DT3 C̄33 ζ∆α + SG∆u + 2S33 B̄T3 ζ∆α Jd ζ + r(u) = 0
−1
(3.25f) Z 1 


BT DB̄SG ∆u + B̄T DT3 C̄33 + 2S33 B̄T3 ζ∆α Jd ζ + r(u) = 0
−1
 

rα 1
KT ∆u + HT ∆α + r(u) = 0 → KT ∆u − HT + H∆ u + r(u) = 0
kα kα
Finally, all this can be expressed (3.25g) as a modification of the elemental tangent stiffness matrix and the internal
forces.
  (
1 T rα K̄T = KT − HT k1α H
(3.25g) KT − H H ∆ u − HT + r(u) = 0 →
kα kα r̄ = r − HT krαα

3.4.3.3.3 Pull-Back and Push-Forward (Extension of the formulation) :

The formulation presented until now is broadly the formulation already presented in the works of Fernando
Flores[ArtFlo13c; ArtFlo13a; ArtFlo13b], the main step forward of this work is the extension of the formulation. As we
have seen, the formulation presented allows us to obtain a modified left Cauchy tensor C̄. With this tensor we are
able to obtain strains in a traditional way, g.e with Green-Lagrange, but we are unable to work with the Pull-Back and
Push-Forward operations, these are, furthermore, the underlying operations in the constitutive operations in Kratos.
The fundamental concepts for this are presented in the Appendix section A.1.Pull-Back, Push-Forward fundamental
concepts.
In order to perform these operations, we need the deformation gradient F, or in our case F̄ owing to we are
working with a modified right Cauchy tensor C̄. Obtain one from another is not a trivial operation, and we must
consider additional assumptions to obtain our modified deformation gradient. In a standard formulation to obtain the
deformation gradient F we compute (3.26) from the material displacement gradient tensor ∇X u, C can be obtained
easily from here.

(3.26) F = ∇ X u + I → C = FT F

We will present now (3.27a) the polar decomposition of F, which will be the key idea considered to obtain F̄. In
this decomposition R represents the proper orthogonal tensor and U is the right stretch tensor. The right stretch
tensor can be computed from the square root of the right Cauchy tensor3.27b. The only remaining component
needed to compute the F̄ will be the modified proper orthogonal tensor where we will take the assumption that the
modified rotation tensor (R̄) equals the rotation tensor (R), meaning R̄ = R. The computation of this F̄ is summarised
in (3.27c).

(3.27a) F = RU thus F̄ = R̄Ū

√ p
(3.27b) U= C thus Ū = C̄

( √
C = FT F → U = C → R = F · U−1
(3.27c) F̄ = RŪ √
Ū = C̄

Page 62 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

3.5 Numerical examples

First of all, we consider as reference the results obtained with previous implementations of the element by Flo-
res[ArtFlo13b; ArtFlo13a], as well as the results from [ArtSLL04; ArtKGW06b]. In the following, when "Ref." is
displayed will refer to this; meanwhile when "Cal."5 is displayed the simulation will have been performed with the
in-home implementation in Kratos.

3.5.1 Patch test

Coordinates [mm]
1: (0.04, 0.02)
2: (0.18, 0.03)
3: (0.16, 0.08)
4: (0.08, 0.08)
Figure 3.4: Patch test geometry
The first step in the implementation of an element into a FEM code is the verification of the kinematics, for this
reason the patch test is understood as a necessary condition for the convergence of the element. In the case of
solid elements it is expected that when nodal displacements corresponding to a constant strain gradient (membrane
patch test) are imposed, constant efforts are obtained in all the elements; so in owing to the SPRISM is in fact a solid
element this is the behaviour we should expect. In the case of a solid-shell element clearly, this must satisfy at least
the membrane patch test and, although it may not be necessary, it is highly desirable that the element satisfies the
bending patch test as this will lead to a more robust and reliable element.
Figure 3.4 shows a patch of elements that has been widely used to access quadrilateral shell elements and
hexahedral solid shell elements, like where are working with prisms (wedges), we should split in two these elements.
The size of the largest sides is a = 0.24mm and the size of the shortest side is b = 0.12mm, while the thickness
considered is t = 0.001mm. The lower surface has been located at coordinate z = −t /2. The mechanical properties
of the material are: Youngs modulus E = 106 MPa and Poisson ratio ν = 0.25. Because the problem considered
is linear √
just 2 integration points across the thickness are used located in the usual Gauss quadrature positions
(ζ = ±1/ 3).

3.5.1.1 Membrane patch test

X-DISPLACEMENT Y-DISPLACEMENT
3e-08 2.4e-08
2.6667e-08 2.1333e-08
2.3333e-08 1.8667e-08
2e-08 1.6e-08
1.6667e-08 1.3333e-08
1.3333e-08 1.0667e-08
1e-08 8e-09
y y
6.6667e-09 5.3333e-09
3.3333e-09 2.6667e-09
z x 0 z x 0

(a) Displacement X (b) Displacement Y

Figure 3.5: Solution for the displacement
The prescribed nodal displacements (on the boundary nodes) are defined by the linear functions (3.28), and
uz = 0 only on the nodes in the lower face to allow contraction due to Poisson effect. Due to the pure membrane
strains developed in this test, the internal DOF α obtained is zero and constant in all the elements.

(

y
ux = x + 2

· 10−3
(3.28) x
uy = y + 2
· 10−3
5 Or nothing more added.

Vicente Mataix Ferrándiz Page 63 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

Sxx-CAUCHY STRESS TENSOR Syy-CAUCHY STRESS TENSOR

1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
y y
1.3333e+05 1.3333e+05
1.3333e+05 1.3333e+05
z x 1.3333e+05 z x 1.3333e+05

(a) σxx (b) σyy

Sxy-CAUCHY STRESS TENSOR

40000
40000
40000
40000
40000
40000
40000
y
40000
40000
z x 40000

(c) σxy
Figure 3.6: Solution for the stress

Using the present element SPRISM the correct results are obtained for both the displacements of the interior
nodes according to (3.28) (Figure 3.5)and the element stresses (σxx = σyy = 1333.3MPa and σ = 400Mpa, Figure
3.6).

3.5.1.2 Bending patch test

X-DISPLACEMENT Y-DISPLACEMENT
1.5e-11 1.2e-11
1.1667e-11 9.3333e-12
8.3333e-12 6.6667e-12
5e-12 4e-12
1.6667e-12 1.3333e-12
-1.6667e-12 -1.3333e-12
-5e-12 -4e-12
y y
-8.3333e-12 -6.6667e-12
-1.1667e-11 -9.3333e-12
z x -1.5e-11 z x -1.2e-11

(a) Displacement X (b) Displacement Y

Figure 3.7: Solution for the displacement
In this case the displacement field associated with a constant bending stress state is given by (3.29), that is
prescribed on the exterior nodes of both shell faces.

 y

z
ux = x + 2
·
 · 10−3 2
(3.29) uy = y + x2 · · 10−3 z
2


uz = x + xy + y 2 · 21 · 10−3
2


We obtain a value for the internal DOF α constant in all elements and equal to: α = 0.3333 · 10−8 . The bending
stresses at the integration points are σxx = σyy = ±0.3849 Mpa6 and σxy ± = 0.1155 Mpa7 (Figure 3.8), while the
displacements at the interior nodes correspond exactly with the expression (3.29) (Figure 3.7). So in consequence
the element satisfies this test too.
6 If we interpolate the values to the external faces we get σxx = σyy = ±0.6666 Mpa
7 Interpolated to the most external faces σxy ± = 0.2000Mpa.

Page 64 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

Sxx-CAUCHY STRESS TENSOR Syy-CAUCHY STRESS TENSOR

66.664 66.665
51.85 51.85
37.036 37.036
22.222 22.222
7.4072 7.4073
-7.4071 -7.4071
-22.221 -22.221
y y
-37.036 -37.036
-51.85 -51.85
z x -66.664 z x -66.665

(a) σxx (b) σyy

Sxy-CAUCHY STRESS TENSOR

19.999
15.555
11.111
6.6665
2.2222
-2.2221
-6.6664
y
-11.111
-15.555
z x -19.999

(c) σxy
Figure 3.8: Solution for the stress

3.5.2 Cantilever

Y X F
Z

Figure 3.9: Cantilever with point load

A cantilever plate strip of length L = 10 mm width b = 1 mm and thickness t1 = 0.1 mm is subjected to a
transverse load F = 40 N (Figure 3.9). For the selected Youngs modulus E = 106 MPa the behaviour is one with
large displacements but small strains. Using different values of Poisson ratio (ν = 0.0, ν = 0.3, ν = 0.49, ν = 0.499
and ν = 0.4999 (quasi-incompresible)) it can be assessed if the proposed assumed strain techniques allow to avoid
respectively the transverse shear locking, the Poisson effect locking and the volumetric locking. At the same
time, different geometries, with different numbers of divisions in length have been considered, see Figure 3.10, one
in the width and one across the thickness with two integration points. The final deformed configurations (vertical
displacement is 70% of the length) is achieved considering small load steps.

z z z z

x y x y x y x y

(a) 8 divisions in length (b) 16 divisions in length (c) 24 divisions in length (d) 32 divisions in length
Figure 3.10: Cantilever geometries
The Figure 3.11b shows the maximum vertical displacement versus the load factor (from 0 to 1) for 5 different
values of the Poisson ratio, for the geometry of the 32 divisions in length. The case ν = 0 allows to compare with the

Vicente Mataix Ferrándiz Page 65 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

Z-DISPLACEMENT
0.0070968
0.0063083
0.0055197
0.0047312
0.0039427
0.0031541
0.0023656
0.0015771
z
0.00078853
0

x y

(a) Displacement for ν = 0.0 with 32 subdivisions

1.0 8
Ref.ν = 0.0
Sol.ν = 0.0 7
0.8 Sol.ν = 0.3
Sol.ν = 0.49 6

Displacement z [mm]
Sol.ν = 0.499
Sol.ν = 0.4999 5
Load factor

0.6
Ref.ν= 0.0
4 Ref.ν= 0.3
Ref.ν= 0.499
0.4 Ref.ν= 0.4999
3 Cal.ν= 0.0
Cal.ν= 0.3
2 Cal.ν0.4
0.2 Cal.ν0.499
Cal.ν0.4999
1

0.0 0
0 1 2 3 4 5 6 7 8 0 5 10 15 20 25 30 35
Displacement z [mm] Number of elements
(b) Evolution of displacement with load factor (c) Comparision of results respect ν and the number of longitudinal
divisions
Figure 3.11: Solution

reference value (uz = 7.08mm) and to see if the approach used to cure transverse shear locking is adequate. The
result obtained uz = 7.06mm indicates that effectively the element is free of transverse shear locking. The second
value of Poisson ratio (ν = 0.30) is used to assess if the EAS technique avoids the appearance of locking due to
Poissons effect. In this case the computed displacement is uz = 7.03mm that although it is not exactly the same
value obtained for ν = 0 shows that the proposed method avoids the Poissons effect locking allowing a proper
gradation of the transverse normal strain. Finally, the last three values of Poisson ratio (0.49,0.499 and 0.4999) allow
to observe if the performance of the element deteriorates significantly in the quasi-incompressible range. It can be
seen that although differences grow with Poisson ratio, this is below 4% for the higher value considered.
Besides, Figure 3.11c plots the tip displacement as a function of the mesh density (number of divisions along
the length) for four different Poissons ratio. In the reference results obtained by Fernando G. Flores it can be seen
that convergence deteriorates for Poissons ratio larger than 0.499; as it happens in our in-home implementation, but
slightly less significant in this case.

3.5.3 Cantilever subjected to end bending moment

Figure 3.12 shows a cantilever subjected to end moment M. The problem has been considered in from the
reference[ArtSLL04]. The cantilever forms a circular arc with its radius R given by the classical flexural formula
R = EI = M . Using the formula, the analytical normalised deflections can be derived to be (3.30) where M0 = EI = L.
The maximum end moment Mmax is taken to be M0 at which the beam will be bent into a circle.

    
U M0 M W M0 M
(3.30) = sin −1 , = 1 − cos
L M M0 L M M0

Page 66 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

M=0.35M max

E=1.2×106
ν = 0, L = 12 M
b = 1, h = 0.1 M=0.7M max
sh
M o = EI/L = 25/3 l me
in itia
M max = 2πM o = 50π/3
M=M max

Figure 3.12: Cantilever subjected to end bending moment

Figure 3.13a plots the end moment against the vertical and horizontal tip deflections for different configurations of
the geometry (8, 16 and 32 subdivisions) and Figure 3.13b portrays the deformed cantilever at M = Mmax .

1.0
Ref. Disp. Z
Ref. Disp. X
Disp. Z(8) x y

0.8 Disp. X(8)

Disp. Z(16)
Disp. X(16)
Disp. Z(24)
Disp. X(24)
Moment factor

0.6 Disp. Z(32)

Disp. X(32)

0.4

|DISPLACEMENT|

0.2 12.02
10.684
9.3488
8.0133
6.6777
5.3422
4.0066

0.0 z
2.6711

0 2 4 6 8 10 12 14 16 1.3355
0

Displacement [m] x y
GiD

(a) Initial geometry (b) Deformed geometry

Figure 3.13: Solution for a cantilever subjected to an end bending moment

3.5.4 Frequencies test

90

80

70
Displacement [mm]

60

50

40

30

20 Ref. Disp.
Consistent Disp.
10 Lumped Disp.

0
0.00 0.02 0.04 0.06 0.08 0.10 0.12
Number of elements
Figure 3.14: Frequencies behaviour test

This example (extracted from[ArtFlo13a; ArtOUS04]) considers the dynamic behaviour of a cantilever beam with
length, width and thickness L = 1; b = 0.1 and t = 0.01 respectively. The mechanical properties are Youngs modulus

Vicente Mataix Ferrándiz Page 67 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

E = 100GPa, Poissons ratio µ = 0 and mass density ρ = 1000kg /m3 . The point load applied at the free side (Figure
3.9) has a value of 100N with a Heaviside step time function. As the problem is elastic with µ = 0 there is no Poissons
effect across the thickness nor volumetric locking. The behaviour is purely bending and it is useful to evaluate the
shear locking and assess the proposed cure. The discretisation includes eight uniform divisions along the length, one
in the width and one element through the thickness. The solution shown in Figure 3.14 a very close behaviour to the
reference[ArtFlo13a], both for lumped and consistent mass matrix.

3.5.5 Cook membrane

This example (see Figure 3.15) involves a large amount of shear energy and is commonly used to assess in-plane
bending performance.

16
F

44

48
Figure 3.15: Cook ’s membrane geometry

Plane strain condition will be considered here with two different material behaviour: (a) a quasi-incompressible
elastic material with G = 80.1938GPa and K = 40.1 · 104 GPa corresponding with a Poisson ratio µ = 0.4999 and
(b) an elasticplastic material with elastic properties G = 80.1938GPa and K = 164.21GPa implying a Poisson ratio
µ = 0.29 and J2 plasticity with isotropic hardening as a function of the effective plastic strain ep defined by (3.31) and
the values of the Table 3.3.

8 8

7
7
6
Displacement [mm]

Displacement [mm]

6
5
Ref. Souza Ref. Displ. Flores
5 Ref. Flores nu=0.499 4 Ref. Displ. Q1P0
Ref. nu=0.4999 Ref. Displ. Q1E4
Sol. nu=0.49 3 Displ. Sol.
4 Sol. nu=0.499
Sol. nu=0.4999 2
3
1

2 0
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Number of elements Number of elements
(a) Displacement versus number of elements for the elastic case (b) Displacement versus number of elements for the elastic-plastic case
Figure 3.16: Cook ’s membrane convergence solution

(3.31a) σy∗ = Linear hardening + Exponential hardening

Page 68 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

(
Linear hardening = σy + ep · θ · h
(3.31b) p
Exponential hardening = (σ0 − σ∞ ) · (1 − e−η·e )

Yiled stress σy 0.45GPa Kinematic hardening modulus h 0.12924GPa

Reference hardening modulus σ0 0.45GPa Infinity hardening modulus σ∞ 0.715GPa
Hardeningexponent η 16.93 Pure isotropic hardening θ 1
Table 3.3: Isotropic-kinematic hardening law constans
The applied load is 100kN for the elastic case and 5kN for the elasticplastic material. The plane strain condition
implies coefficient C33 = 1 at all points (α = 0), thus the version without ANS for the in-plane components locks
due to the almost incompressibility constraint in the same way that a constant strain triangle does. Because of that
this example is intended to assess how the improvement in the membrane field collaborates to cure the volumetric
locking.

(a) Vertical displacement for the elastic case (32 elements ν = 0.4999) (b) Vertical displacement for the elastic-plastic case(16 elements)
Figure 3.17: Cook ’s membrane deformed shape solution

Figures 3.16a and 3.16b show a convergence analysis as the mesh is refined, where the vertical displacement of
the point C has been plotted versus the number of divisions per side. In Figures 3.17a and 3.17b the deformed shape
for the elastic and elastic-plastic cases is shown. The results have been compared with the reference [ArtFlo13b], and
other results presented in the same article, where the results that we have obtained present a very good convergence
in comparison.

3.5.6 Scoordelis cylindrical roof test

In this example we proceed with a linear analysis of a cylindrical shell under self-weight, which is free along the one
side as can be seen in the Figure 3.18 and it has been simplified in one quarter of the original size with additional
consideration of the symmetry conditions of the problem. We have considered five different geometric configurations,
with a different number of elements across the side, until the convergence of the solution have been archived. As we
have mentioned previously, the problem is linear, therefore it is possible to solve the problem considering just two
Gauss points across the thickness.

The problem is membrane dominant, that is why could be interesting to observe the relevance of the ANS for
in-plane components in non-isochoric problems. The results obtained are shown in the Figure 3.19, where in Figure
3.19b the results are compared with our reference[ArtFlo13a], having a very close convergence to the reference
solution. Besides the Figure 3.19a shows the deformed shape obtained for the maximum number of elements.

Vicente Mataix Ferrándiz Page 69 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

try sy
me m
sym m
et
ry

Di A
ap
hr
ag
m
free
300
α=40
Z

300
Y X

Figure 3.18: Geometry of the scoordelis cylindrical roof

4.0

3.5

3.0
Displacement [mm]

2.5

2.0

1.5
Ref. Quadratic
1.0 Ref. Non Quadratic
Sol. Quadratic
0.5
0 5 10 15 20 25 30 35
Number of elements
(a) Deformed shape of the scoordelis (b) Displacement versus number of elements
Figure 3.19: Solution for the scoordelis cylindrical roof

3.5.7 Sphere test

(a) Free Z
(b)
Y
Sy

X
mm
etr
y
metr

y
Sym

F Free
Figure 3.20: Geometry of a semi-spherical shell with a hole, both original and deformed geometry
Some interesting kind of problem to check the suitability of the element is the problems where an initially double
curved geometry. This problem (Figure 3.20) is analysed and solved recurrently in the context of large elastic
displacements. The Figure 3.20 presents the geometry considered in the resolution of the problem, where once

Page 70 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

again the symmetry has been considered to simplify the resolution of the problem. This kind of problem is mainly
an inextensional bending problem where Poisson effect has an important role in the behaviour of the structure, in
contrast with the membrane effect that is less significant in this problem, on the other hand, the membrane locking
and the curvature-thickness looking could appear.

(a) Deformed shape of the semi-spherical shell

6 6
Ref. point A Ref. point A
Ref. point B Ref. point B
5 5
Point A (32) Point A (32)
Point B (32) Point B (32)
Point A (16) Point A (24)
Displacement [mm]

Displacement [mm]

4 4
Point B (16) Point B (24)
Point A (8)
Point B (8)
3 3

2 2

1 1

0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Load factor Load factor
(b) Displacement versus load for R /t = 250 (c) Displacement versus load for R /t = 1000
Figure 3.21: Geometry of a semi-spherical shell with a hole solution

Several meshes have been considered, with 8, 16, 24 and 32 elements by side respectively, with a middle radius of
R = 10mm and thickness of t = 0.04mm (R /t = 250). The coarser is the element, the more it could suffer the looking
effect due to the initial curvature, considering this when solutions more differ more than the 5% from the target values.
We are considering the following mechanical properties, E = 6.825 · 104 GPa and ν = 0.3. See Figure 3.21.

3.5.8 Pull-out of an open-ended cylindrical shell

Figure 3.22 shows an open-ended cylinder being pulled by a pair of radial forces P. The problem has been considered
in the Reference [ArtSLL04], among others. Owing to symmetry, just one-eighth of the shell is modelled.

The Figure 3.23a presents the convergence of the solution considering a different number of elements, having a
converged solution very close to the reference one[ArtSLL04]. The final deformation of the one-eight is shown in
Figure 3.23b, showing that the problem involves very large displacements.

Vicente Mataix Ferrándiz Page 71 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

P=40000

A
E = 10.5 MPa

free
ν = 0.3125

e
dge
R = 4.953 m
Z
Y L = 10.35 m
X
h = 0.094 m
R B
C Pmax = 40000 N
fre
ee

L
dg
e

P=40000
P
Figure 3.22: The open-end cylindrical shell subjected to radial pulling forces.

4000
Ref. WA
3500 Ref. UB
Ref. UC
3000 Sol. WA(8)
Sol. UB (8)
2500 Sol. UC (8)
Point load [N]

Sol. WA(16)
2000 Sol. UB (16)
Sol. UC(16)
1500

1000

500

0
0 1 2 3 4 5
Displacement [mm]
(a) Loaddeflection curves (b) The deformed shape
Figure 3.23: Solution of the open-ended cylindrical shell

3.5.9 Pinched semi-cylindrical

P=2000 cla
mp
ed

e A
fre
R

Z
W =0 Y L
X
7
E=2.0685×10 Pa; ν =0.3
R= 1.016; L=3.048; h=0.03
Figure 3.24: The semi-cylindrical shell subjected to an end pinching force

Figure 3.24 shows the semi-cylindrical shell[ArtSLL04] subjected to an end pinching force at the middle of the
free-hanging circumferential periphery. The other circumferential periphery is fully clamped. Along its longitudinal
edges, the vertical deflection and the rotation about the Y -axis are restrained.
The solution obtained is shown in Figure 3.25, where the Figure 3.25a presents the load-deflection curve, the
behaviour in our obtained solution is slightly different from the one presented in the reference[ArtSLL04], this can be

Page 72 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

due to the fact that we are taking the displacement in the point where the load is applied, meanwhile the reference
solutions corresponds to a shell and the solution corresponds with the displacement in the middle surface; a possible
thus to improve our solution way to tackle this problematic is adding an additional layer and plot the solution in the
point belonging to the middle surface. Figure 3.25b presents the deformed shape obtained for the maximum value of
the load.

2000
Reference (Shell)
Sol. SolidShell(32)
Sol. SolidShell(40)
1500
Point load [N]

1000

500

0
0 500 1000 1500 2000
Displacement [mm]
(a) Loaddeflection curves (b) The deformed shape
Figure 3.25: Solution of the semi-cylindrical shell

3.5.10 Slit test

The Figure 3.26a shows the original geometry of an annular plate and it presents a large displacement (Figure
3.26b) due to a load applied in one face while the other face is constrained. This problem is a common benchmark
considered to study the behaviour of shells under large rotations, in our case, as we are considering a solid-shell we
do not have rotations in our element and it could be considered as large displacement problems.

B Z

Ri=6 Re=10
A X Y

6
E=21x10 ν=0

thick. = 0.03 Pmax=0.8

Re
Ri

(a) Initial geometry (b) Deformed geometry

Figure 3.26: Slit annular plate

The solution looks like the solution expected Figure 3.27b. A comparison of the results for the coarse mesh
(Figure 3.27a) allows to observe the influence of the ANS for membrane part that for the maximum load factor indicate
a difference in displacements larger than 4%. The results for the fine mesh are in excellent agreement with those
provided in[ArtFlo13b].

Vicente Mataix Ferrándiz Page 73 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

1.0
Ref. point A
Ref. point B
0.8 Point A (360)
Point B (360)
Point A (710)
0.6 Point B (710)
Load factor

Point A (1440)
Point B (1440)
0.4

0.2

0.0
0 2 4 6 8 10 12 14 16 18
Displacement [mm]
(a) Displacement-load (b) Deformation solution
Figure 3.27: Slit annular plate solution

3.5.11 Cylindrical panel test

This example[ArtSLL04; ArtFlo13b] considers a rectangular cylindrical panel simple supported along the straight
sides and free along the curved sides, that is subjected to a vertical point load in its centre (see Figure ??). The
middle surface geometry is defined by the length of the panel L = 508mm, the radius of the cylinder R = 2540mm and
the half angle θ = 0.1rad. The behaviour of the panel presents a limit point, followed by a strong loss of strength and
a final stiffening once the curvature is inverted. Two different thicknesses for the same mid-surface geometry have
been considered t = 12.7mm and t = 6.35mm that for the thin case leads to a snap back of the loaded point. This
example has been widely used to assess the performance of shell elements and non-linear path-following techniques,
like the arc-length.

free
P
d
ge
hin
C

free

d
θ ge
hin L
R = 2540, L = 254
L θ = 0.1 radian
Z
h = 12.7 or 6.35
Pmax = 3000
Y E = 3202.75 Pa
X ν = 0.3
Figure 3.28: Geometry of the hinged cylindrical panel

For this problem two meshes have been considered with 8 and 16 elements per side. In this case 2 elements in
the thickness direction have been used that allows to introduce the hinge in the middle surface and then to compare
with solutions obtained with shell elements. The vertical displacement of the loaded point C.

Figures 3.29a and 3.29b show the evolution of the load-deflection compared with the reference[ArtSLL04], besides
the Figures 3.29c and 3.29d present the deformed shape of the problem.

Page 74 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

3000 2000
Reference Shell
2500 Sol. Solid-Shell (8)
Sol. Solid-Shell (16)
1500
2000
Point load [N]

Point load [N]

1500
1000
1000

500
500
Reference Shell
0 Sol. Solid-Shell (8)
Sol. Solid-Shell (16)
−500 0
0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035 0.040 0.000 0.005 0.010 0.015 0.020 0.025 0.030
Displacement [m] Displacement [m]
(a) Thin geometry displacement-load (b) Thick geometry displacement-load

(c) Thin geometry displacement (d) Thick geometry displacement

Figure 3.29: Solution of the cylindrical panel

3.5.12 Conical shell test

This example is selected to demonstrate the ability of the developed finite element to deal with strongly non-linear
situations. The geometrical data are taken from[ArtKGW06b]. Here, elasticplastic material behaviour is assumed.
All necessary material and geometrical data are depicted in Figure 3.30. The non-linear behaviour is computed
using an arc-length algorithm with displacement control. The constitutive law considered is a J2 hyperelastic-plastic
model.

p Geometry: Material:
p t r
r=1m E = 206.9 Pa
R=2m ν = 0.29
L=1m L σy = 0.45 Pa
t = 0.1 m σ∞ = 0.715 Pa
Load: p R h = 0.12924 Pa
η = 16.93
Figure 3.30: Geometry of the cone shell

The results are depicted in Figure 3.31a, where w denotes the vertical displacement of the upper edge. The
load deflection diagram demonstrates that our result is close in the order of magnitude to the reference, thus some
improvements are needed in the modelisation, in special the relative to the relative to the arc-length strategy. The
Figures 3.31b and 3.31c present the deformation and plastic strain in the last stages of the simulation.

Vicente Mataix Ferrándiz Page 75 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

Reference Solution (16) Solution (8)

0.05

0.04
Point load [N]
0.03

0.02

0.01

0.00
0.0 0.5 1.0 1.5 2.0
Displacement [m]
(a) Initial geometry

(b) Deformed geometry (c) Plastic strain

Figure 3.31: Cone-shell test

3.5.13 Wrinkling test

The problem consists[ArtTSW05; ArtFlo13a] of a square membrane (see Figure 3.32) with a side a = 229mm made
of a thin film of Mylar with a thickness t = 0.0762mm. The Mylar mechanical properties are E = 3790MPa and
µ = 0.38. The top and bottom edges are clamped and the lateral edges are free. The top edge is subjected to a
uniform horizontal displacement ∆ = 1mm along the edge.

a
Figure 3.32: Square thin film under in-plane shear

Two uniforms structured meshes with 26 − 26 and 51 − 51 nodes, with 1250 and 5000 elements respectively
have been considered. Figure 3.33 plots two out-of-plane displacement profiles along the centre of the square in

Page 76 of 374 Vicente Mataix Ferrándiz

CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS 3.5 Numerical examples

both Cartesian directions, as well as the deformation obtained. The solution that we obtain are very close to those
obtained in the reference [ArtFlo13a].

(a) Deformation with 25 elements per side (b) Deformation with 50 elements per side

4 1.5
Reference 25
3 Reference 50 1.0
25
2 50 0.5
Displacement z [mm]

Displacement z [mm]

1
0.0
0
−0.5
−1
−1.0
−2 Reference 25
Reference 50
−3 −1.5 25
50
−4 −2.0
0.00 0.05 0.10 0.15 0.20 0.25 0.00 0.05 0.10 0.15 0.20 0.25
Coor. X Coor. Y
(c) Transverse displacement profiles along the centre of the square:(d) Transverse displacement profiles along the centre of the square:
y = a /2 x = a /2
Figure 3.33: Solution of the wrinkling simulation

3.5.14 FSI-Vein test

An additional test that has been performed to check the robustness of the element, which consists in a simple FSI
simulation of an elastic vein. This element has clear advantages from the common approaches performed nowadays,
which consist normally in the consideration of multiple layers of solid elements (usually hexahedron), despite the vein
is under "shell stresses". This owing to the need of having a proper and correct definition of the irregular geometry
and thickness that a real vein can show, besides a correct definition of the interface for a proper FSI simulation, which
cannot be correctly modelled with a conventional shell element. So this element has advantages of the solid element,
and like the element is clearly created to compute "shell stresses" there does not need to consider additional layers of
solid to enrich the behaviour of the solid.

This problem was originally proposed in [ArtNob01] and later reproduced in [PhDVMO09]. Its aim is to simulate
the FSI arising in the modelling of blood flow in the human cardiovascular system. As described in [PhDVMO09], the
problem consists of a thin elastic vessel, which in this case has been modelled with the current element considering
an hyperelastic constitutive law, conveying the blood flow, which is modelled as an incompressible fluid using the
Navier-Stokes equations.

Vicente Mataix Ferrándiz Page 77 of 374

3.5 Numerical examples CHAPTER 3. ROTATION-FREE SHELLS AND SOLID-SHELL ELEMENTS

Regarding the geometry, it consists in a straight cylinder of radius r0 = 0.005m which length and thickness
are L = 0.05m and t = 0.001m. The blood physical parameters are ρf = 1000kg /m3 and dynamic viscosity
µf = 0.003kg /ms, yielding a kinematic viscosity νf = 3e − 06m2 /s. Regarding the solid parameters, the density is
ρs = 1200kg /m3 while the Poisson ratio and Young modulus are νs = 0.3 and E = 3e05Pa. Regarding the boundary
conditions, both sides of the vein are clamped (radial displacements allowed) and an overpressure of p = 1333.2Pa
(see (3.32)) is imposed at the inlet boundary for 3ms.



 1333.2 sin (2π t) if t ≤ 0.25ms

 1333.2 if 0.00025 < t ≤ 0.275ms
(3.32) p=

 1333.2 (1 − sin (2π (t − 0.275))) if 0.275ms < t ≤ 0.3ms


0.0 otherwise

l= 0.25 Valdés (2007) l= 0.5 Present work l= 0.25 Valdés (2007) l= 0.5 Present work
l= 0.25 Present work l= 0.75 Valdés (2007) l= 0.25 Present work l= 0.75 Valdés (2007)
l= 0.5 Valdés (2007) l= 0.75 Present work l= 0.5 Valdés (2007) l= 0.75 Present work
0.015
1500

0.010
1000
p [N/ m 2]
ur [cm]

0.005 500

0.000 0

0.000 0.002 0.004 0.006 0.008 0.010 0.000 0.002 0.004 0.006 0.008 0.010
t [s] t [s]
(a) Radial displacement (b) Pressure
Figure 3.34: Solution of the FSI-Vein test
Figures 3.34a and 3.34b collects a comparison between the results in [PhDVMO09] and the obtained ones for
three control points placed at 0.25l, 0.5l and 0.75l, being l the tube length. Regarding the radial displacements
(Figure 3.34a), it can be seen that the obtained results are similar to the reference ones. The major differences appear
after the peak value when the vein section is recovering its shape. Besides, this vein retraction is much clear in the
presented solution and can be clearly noted by the negative radial displacements. This behaviour is more similar to
real hemodynamics and has been also observed in similar problems in the literature [PhDCal06].
On the other hand, the pressure evolution is also assessed in Figure 3.34b. As can be noted, the pressure trend
matches the radial displacement evolution but some oscillations appear in the solution. Regarding the nature of these
oscillations, it can be asserted that they are not numeric, since one oscillation is developed in several time steps, and
this is done due to the fact that any non-reflecting boundary has been considered.

Page 78 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookBD12] Sophie Germain: An essay in the history of the theory of elasticity. Louis L Bucciarelli and
Nancy Dworsky. Springer Science & Business Media. 2012.
[BookBD80] Sophie Germain: An Essay in the History of the Theory of Elasticity. Louis L.. Bucciarelli and
Nancy Dworsky. D. Reidel Publishing Company. 1980.
[BookBD90] Modélisation des structures par éléments finis: Solides élastiques. Jean-Louis Batoz and
Gouri Dhatt. Presses Université Laval. 1990.
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookBor+12] Nonlinear Finite Element Analysis of Solids and Structures, 2nd edition. René de Borst,
Mike A. Crisfield, Joris J. C. Remmers, and Clemens V. Verhoosel. Wiley. 2nd Edition. 2012.
[BookCal89] Theory of shell structures. Chris R Calladine. Cambridge university press. 1989.
[BookFHK02] Handbook of computer aided geometric design. Gerald Farin, Josef Hoschek, and M-S Kim.
Elsevier. 2002.
[BookGou12] Analysis of shells and plates. Phillip L Gould. Springer Science & Business Media. 2012.
[BookHBC09] Isogeometric Analysis: Toward Integration of CAD and FEA. T.J.R. Hughes, Yuri Bazilevs, and
J. Austin Cottrell. Wiley. 1st ed. 2009.
[BookHug+14] Computer graphics: principles and practice. John F Hughes, Andries Van Dam, James D Foley,
Morgan McGuire, Steven K Feiner, and David F Sklar. Pearson Education. 2014.
[BookKra67] Thin elastic shells: an introduction to the theoretical foundations and the analysis of their static and
dynamic behavior. Harry Kraus. John Wiley & Sons. 1967.
[BookNov59] The theory of thin shells. Valentin Valentinovich Novozhilov. P. Noordhoff. 1959.
[BookOña13] Structural Analysis with the Finite Element Method Linear Statics: Volume 2. Beams, Plates and
Shells. Eugenio Oñate. Springer Netherlands. Jointly published with CIMNE, Barcelona,
Spain2013. 2013.
[BookPT97] The NURBS book: monograph in visual communication, 1997. LA Piegl and W Tiller. Springer, New
York. 1997.
[BookRog00] An introduction to NURBS: with historical perspective. David F Rogers. Elsevier. 2000.
[BookTW59] Theory of plates and shells. Stephen P Timoshenko and Sergius Woinowsky-Krieger. McGraw-hill.
1959.
[BookWem82] Mechanics of solids with applications to thin bodies. Gerald Wempner. Springer Science &
Business Media. 1982.
[BookWri08] Nonlinear finite element methods. Peter Wriggers. Springer. 1st ed. 2008.

Vicente Mataix Ferrándiz Page 79 of 374

ARTICLES BIBLIOGRAPHY

Articles
[ArtAC09] “An improved assumed strain solid–shell element formulation with physical stabilization for
geometric non-linear applications and elastic–plastic stability analysis”. Farid Abed-Meraim and
Alain Combescure. In: International Journal for Numerical Methods in Engineering. No. 13, Vol. 80,
2009, pp. 1640–1686. Wiley Online Library.
[ArtAIZ70] “Analysis of thick and thin shell structures by curved finite elements”. Sohrabuddin Ahmad,
Bruce M. Irons, and O. C. Zienkiewicz. In: International Journal for Numerical Methods in
Engineering. No. 3, Vol. 2, 1970, pp. 419–451. DOI: 10.1002/nme.1620020310.
[ArtBen+10] “Isogeometric shell analysis: The Reissner–Mindlin shell”. D.J. Benson, Y. Bazilevs, M.C. Hsu, and
T.J.R. Hughes. In: Computer Methods in Applied Mechanics and Engineering. No. 5–8, Vol. 199,
2010, pp. 276–289. DOI: 10.1016/j.cma.2009.05.011.
[ArtBR92] “Comparison of Shell Theory and Degeneration”. N Büchter and E Ramm. In: . 1992, pp. 15–30.
Springer. DOI: 10.1007/978-3-7091-2604-2_2.
[ArtDB84] “A continuum mechanics based four-node shell element for general non-linear analysis”.
Eduardo N. Dvorkin and Klaus-Jürgen Bathe. In: Engineering computations. No. 1, Vol. 1, 1984,
pp. 77–88. MCB UP Ltd. DOI: 10.1108/eb023562.
[ArtDha69] “Numerical analysis of thin shells by curved triangular elements based on discrete Kirchhoff
hypothesis”. G Dhatt. In: Proc. ASCE, Symp. on Applications of FEM in Civil Engineering,
Vanderbilt Univ., Nashville, Tenn., 1969. 1969, pp. 13–14.
[ArtDha70] “An efficient triangular shell element”. GS Dhatt. In: AIAA Journal. No. 11, Vol. 8, 1970,
pp. 2100–2102. DOI: 10.2514/3.6068.
[ArtDMM86] “A new triangular discrete Kirchhoff plate/shell element”. G. Dhatt, L. Marcotte, and Y. Matte. In:
International Journal for Numerical Methods in Engineering. No. 3, Vol. 23, 1986, pp. 453–470.
DOI : 10.1002/nme.1620230310.

[ArtFlo13a] “A “Prism” solid element for large strain shell analysis”. Fernando G. Flores. In: Computer Methods
in Applied Mechanics and Engineering. Vol. 253, 2013, pp. 274–286. Elsevier. DOI:
10.1016/j.cma.2012.10.001.
[ArtFlo13b] “Development of a non-linear triangular prism solid-shell element using ANS and EAS techniques”.
Fernando G. Flores. In: Computer Methods in Applied Mechanics and Engineering. Vol. 266, 2013,
pp. 81–97. Elsevier. DOI: 10.1016/j.cma.2013.07.014.
[ArtFlo13c] “UN ELEMENTO PRISMA TRIANGULAR DE SÓLIDO-LÁMINA PARA EL ANÁLISIS CON
GRANDES DEFORMACIONES”. Fernando G. Flores. In: Mecánica Computacional. Vol. XXXII,
2013, pp. 63–87.
[ArtFO01] “A basic thin shell triangle with only translational DOFs for large strain plasticity”.
Fernando G. Flores and Eugenio Oñate. In: International Journal for Numerical Methods in
Engineering. No. 1, Vol. 51, 2001, pp. 57–83. John Wiley & Sons, Ltd. DOI: 10.1002/nme.147.
[ArtFO05] “Improvements in the membrane behaviour of the three node rotation-free {BST} shell triangle
using an assumed strain approach”. Fernando G. Flores and Eugenio Oñate. In: Computer
Methods in Applied Mechanics and Engineering. No. 6–8, Vol. 194, 2005, pp. 907–932. DOI:
10.1016/j.cma.2003.08.012.
[ArtFO11] “Wrinkling and folding analysis of elastic membranes using an enhanced rotation-free thin shell
triangular element”. Fernando G. Flores and Eugenio Oñate. In: Finite Elements in Analysis and
Design. No. 9, Vol. 47, 2011, pp. 982–990. Elsevier. DOI: 10.1016/j.finel.2011.03.014.
[ArtFV82] “Some results on the behaviour of Degenerated Shell (D.S.) elements”. G. Fezans and G. Verchery.
In: Nuclear Engineering and Design. No. 1, Vol. 70, 1982, pp. 27–35. DOI:
10.1016/0029-5493(82)90264-3.
[ArtHau+01] “Solid-shell’elements with linear and quadratic shape functions at large deformations with nearly
incompressible materials”. R. Hauptmann, S. Doll, M. Harnau, and K. Schweizerhof. In: Computers
& Structures. No. 18, Vol. 79, 2001, pp. 1671–1685. Elsevier.

Page 80 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY ARTICLES

[ArtHCB05] “Isogeometric analysis: CAD, finite elements, NURBS, exact geometry and mesh refinement”.
T.J.R. Hughes, J.A. Cottrell, and Y. Bazilevs. In: Computer Methods in Applied Mechanics and
Engineering. No. 39, Vol. 194, 2005, pp. 4135–4195. DOI: 10.1016/j.cma.2004.10.008.
[ArtHL81] “Nonlinear finite element analysis of shells: Part I. three-dimensional shells”. Thomas J.R. Hughes
and Wing Kam Liu. In: Computer Methods in Applied Mechanics and Engineering. No. 3, Vol. 26,
1981, pp. 331–362. DOI: 10.1016/0045-7825(81)90121-3.
[ArtHS98] “A systematic development of ‘solid-shell’element formulations for linear and non-linear analyses
employing only displacement degrees of freedom”. R. Hauptmann and K. Schweizerhof. In:
International Journal for Numerical Methods in Engineering. No. 1, Vol. 42, 1998, pp. 49–69. Wiley
Online Library.
[ArtHSD00] “Extension of the’solid-shell’ concept for application to large elastic and large elastoplastic
deformations”. R. Hauptmann, K. Schweizerhof, and S. Doll. In: International Journal for Numerical
Methods in Engineering. No. 9, Vol. 49, 2000, pp. 1121–1141.
[ArtKan79] “A simple and efficient finite element for general shell analysis”. Worsak Kanok-nukulchai. In:
International Journal for Numerical Methods in Engineering. No. 2, Vol. 14, 1979, pp. 179–200.
DOI : 10.1002/nme.1620140204.

[ArtKGW06a] “A robust non-linear solid shell element based on a mixed variational formulation”. Sven Klinkel,
Friedrich Gruttmann, and Werner Wagner. In: Computer Methods in Applied Mechanics and
Engineering. No. 1, Vol. 195, 2006, pp. 179–201. Elsevier. DOI:
10.1016/j.cma.2005.01.013.
[ArtKGW06b] “A robust non-linear solid shell element based on a mixed variational formulation”. Sven Klinkel,
Friedrich Gruttmann, and Werner Wagner. In: Computer Methods in Applied Mechanics and
Engineering. No. 1–3, Vol. 195, 2006, pp. 179–201. Elsevier. DOI:
10.1016/j.cma.2005.01.013.
[ArtKir50] “Über das Gleichgewicht und die Bewegung einer elastischen Scheibe.” Gustav Kirchhoff. In:
Journal für die reine und angewandte Mathematik . No. 40, Vol. 1850, 1850, pp. 51–88. De Gruyter.
[ArtLev99] “Sur l’équilibre élastique d’une plaque rectangulaire”. Maurice Levy. In: Comptes Rendus Acad. Sci.
Paris. Vol. 129, 1899, pp. 535–539.
[ArtLov88] “XVI. The small free vibrations and deformation of a thin elastic shell”.
Augustus Edward Hough Love. In: Philosophical Transactions of the Royal Society of London.(A.).
No. 179, 1888, pp. 491–546. The Royal Society London.
[ArtMG86] “A triangular thin-shell finite element based on discrete Kirchhoff theory”. S.S. Murthy and
R.H. Gallagher. In: Computer Methods in Applied Mechanics and Engineering. No. 2, Vol. 54, 1986,
pp. 197–222. DOI: 10.1016/0045-7825(86)90126-X.
[ArtMin51] “Influence of rotatory inertia and shear flexural motions of isotropic elastic plates”.
Raymond D Mindlin. In: . 1951, DOI: 10.1007/978-1-4613-8865-4_29.
[ArtNio85] “Shell theory, vol. xiv”. Frithiof I Niordson. In: . 1985, Amsterdam, New York, NY, North-Holland:
Elsevier Science Pub Co.
[ArtNob01] “Numerical approximation of fluid-structure interaction problems with application to
haemodynamics”. Fabio Nobile. In: . 2001, EPFL.
[ArtNoo90a] “Bibliography of monographs and surveys on shells”. Ahmed K Noor. In: Applied Mechanics
Reviews;(United States). No. 9, Vol. 43, 1990, DOI: 10.1115/1.3119170.
[ArtNoo90b] “Bibliography of Monographs and Surveys on Shells”. Ahmed K. Noor. In: Applied Mechanics
Reviews. No. 9, Vol. 43, 1990, pp. 223–234. DOI: 10.1115/1.3119170.
[ArtOF05] “Advances in the formulation of the rotation-free basic shell triangle”. Eugenio Oñate and
Fernando G. Flores. In: Computer Methods in Applied Mechanics and Engineering. No. 21–24, Vol.
194, 2005, pp. 2406–2443. DOI: 10.1016/j.cma.2004.07.039.
[ArtOUS04] “Selective mass scaling for thin walled structures modeled with tri-linear solid elements”.
L. Olovsson, M. Unosson, and K. Simonsson. In: Computational Mechanics. No. 2, Vol. 34, 2004,
pp. 134–136. DOI: 10.1007/s00466-004-0560-6.

Vicente Mataix Ferrándiz Page 81 of 374

ARTICLES BIBLIOGRAPHY

[ArtOZ00] “Rotation-free triangular plate and shell elements”. Eugenio Oñate and Francisco Zarate. In:
International Journal for Numerical Methods in Engineering. No. 1-3, Vol. 47, 2000, pp. 557–603.
DOI : 10.1002/(SICI)1097-0207(20000110/30)47:1/3<557::AID-
NME784>3.0.CO;2-9.
[ArtPar+06] “Sheet metal forming simulation using {EAS} solid-shell finite elements”. M.P.L. Parente,
R.A. Fontes Valente, R.M. Natal Jorge, R.P.R. Cardoso, and Ricardo J. Alves de Sousa. In: Finite
Elements in Analysis and Design. No. 13, Vol. 42, 2006, pp. 1137–1149. DOI:
10.1016/j.finel.2006.04.005.
[ArtPar79] “A critical survey of the 9-node degenerated shell element with special emphasis on thin shell
application and reduced integration”. H. Parisch. In: Computer Methods in Applied Mechanics and
Engineering. No. 3, Vol. 20, 1979, pp. 323–350. DOI: 10.1016/0045-7825(79)90007-0.
[ArtRei45] “The effect of transverse shear deformation on the bending of elastic plates”. Eric Reissner. In: J.
appl. Mech.. 1945, A69–A77.
[ArtSAV11] “On the use of EAS solid-shell formulations in the numerical simulation of incremental forming
processes”. JIV Sena, RJ Alves de Sousa, and RAF Valente. In: Engineering Computations. No. 3,
Vol. 28, 2011, pp. 287–313. Emerald Group Publishing Limited. DOI:
10.1108/02644401111118150.
[ArtSen+16] “Single point incremental forming simulation with adaptive remeshing technique using solid-shell
elements”. José IV Sena, Cedric Lequesne, Laurent Duchene, Anne-Marie Habraken,
Robertt AF Valente, and Ricardo J Alves de Sousa. In: Engineering Computations. No. 5, Vol. 33,
2016, pp. 1388–1421. Emerald Group Publishing Limited. DOI: 10.1108/EC-06-2015-0172.
[ArtSF89] “On a stress resultant geometrically exact shell model. Part I: Formulation and optimal
parametrization”. J.C. Simo and D.D. Fox. In: Computer Methods in Applied Mechanics and
Engineering. No. 3, Vol. 72, 1989, pp. 267–304. DOI: 10.1016/0045-7825(89)90002-9.
[ArtSFR89] “On a stress resultant geometrically exact shell model. Part II: The linear theory; Computational
aspects”. J.C. Simo, D.D. Fox, and M.S. Rifai. In: Computer Methods in Applied Mechanics and
Engineering. No. 1, Vol. 73, 1989, pp. 53–92. DOI: 10.1016/0045-7825(89)90098-4.
[ArtSFR90] “On a stress resultant geometrically exact shell model. Part III: Computational aspects of the
nonlinear theory”. J.C. Simo, D.D. Fox, and M.S. Rifai. In: Computer Methods in Applied Mechanics
and Engineering. No. 1, Vol. 79, 1990, pp. 21–70. DOI: 10.1016/0045-7825(90)90094-3.
[ArtSim93] “On a stress resultant geometrically exact shell model. Part VII: Shell intersections with 56-DOF
finite element formulations”. J.C. Simo. In: Computer Methods in Applied Mechanics and
Engineering. No. 3, Vol. 108, 1993, pp. 319–339. DOI: 10.1016/0045-7825(93)90008-L.
[ArtSK92] “On a stress resultant geometrically exact shell model. Part V. Nonlinear plasticity: formulation and
integration algorithms”. J.C. Simo and J.G. Kennedy. In: Computer Methods in Applied Mechanics
and Engineering. No. 2, Vol. 96, 1992, pp. 133–171. DOI: 10.1016/0045-7825(92)90129-8.
[ArtSLL04] “Popular benchmark problems for geometric nonlinear analysis of shells”. K.Y. Sze, X.H. Liu, and
S.H. Lo. In: Finite Elements in Analysis and Design. No. 11, Vol. 40, 2004, pp. 1551–1569. DOI:
10.1016/j.finel.2003.11.001.
[ArtSou+05] “A new one-point quadrature enhanced assumed strain (EAS) solid-shell element with multiple
integration points along thickness: Part I—geometrically linear applications”.
Ricardo J. Alves de Sousa, Rui P. R. Cardoso, Robertt A. Fontes Valente, Jeong-Whan Yoon,
José J. Grácio, and Renato M. Natal Jorge. In: International journal for numerical methods in
engineering. No. 7, Vol. 62, 2005, pp. 952–977. Wiley Online Library.
[ArtSR11] “A reduced integration solid-shell finite element based on the EAS and the ANS concept—Large
deformation problems”. Marco Schwarze and Stefanie Reese. In: International Journal for
Numerical Methods in Engineering. No. 3, Vol. 85, 2011, pp. 289–329. Wiley Online Library. DOI:
10.1002/nme.2966.
[ArtSRF90] “On a stress resultant geometrically exact shell model. Part IV: Variable thickness shells with
through-the-thickness stretching”. J.C. Simo, M.S. Rifai, and D.D. Fox. In: Computer Methods in
Applied Mechanics and Engineering. No. 1, Vol. 81, 1990, pp. 91–126. DOI:
10.1016/0045-7825(90)90143-A.

Page 82 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY PH.D.’S THESIS

[ArtSRF92] “On a stress resultant geometrically exact shell model. Part VI: Conserving algorithms for
non-linear dynamics”. J.C. Simo, M. S. Rifai, and D. D. Fox. In: International Journal for Numerical
Methods in Engineering. No. 1, Vol. 34, 1992, pp. 117–164. DOI: 10.1002/nme.1620340108.
[ArtSVR11] “Sheet metal forming and springback simulation by means of a new reduced integration solid-shell
finite element technology”. Marco Schwarze, Ivaylo N Vladimirov, and Stefanie Reese. In:
Computer Methods in Applied Mechanics and Engineering. No. 5, Vol. 200, 2011, pp. 454–476.
Elsevier. DOI: 10.1016/j.cma.2010.07.020.
[ArtTSW05] “Effective Modeling and Nonlinear Shell Analysis of Thin Membranes Exhibiting Structural
Wrinkling”. Tessler Alexander, Sleight David W., and Wang John T. In: Journal of Spacecraft and
Rockets. No. 2, Vol. 42, 2005, pp. 287–298. American Institute of Aeronautics and Astronautics.
DOI : 10.2514/1.3915.

[ArtWas82] “Variational methods in elasticity and plasticity Pergamon Press”. K Washizu. In: . 1982, Oxford,
England.
[ArtWem69] “Finite elements, finite rotations and small strains of flexible shells”. Gerald Wempner. In:
International Journal of Solids and Structures. No. 2, Vol. 5, 1969, pp. 117–153. Elsevier. DOI :
10.1016/0020-7683(69)90025-0.
[ArtWOK68] “Finite-element analysis of thin shells”. Gerald A Wempner, John T Oden, and Dennis A Kross. In:
Journal of the Engineering Mechanics Division. No. 6, Vol. 94, 1968, pp. 1273–1294. ASCE.

Ph.D.’s thesis
[PhDCal06] “Simulación del flujo sanguneo y su interacción con la pared arterial mediante modelos de
elementos finitos”. Francisco José Calvo Plaza. 2006.
[PhDVMO09] “Nonlinear Finite Element Analysis of Orthotropic and Prestressed Membrane Structures”.
J. G. Valdés, J. Miquel, and Eugenio Oñate. Elsevier Science Publishers B. V. 2009.

Online resources
[OnlKra] Kratos Homepage. URL : http://www.cimne.com/kratos/.
[OnlMel+16] GiD Homepage. URL: http://www.gidhome.com/;%20https:
//www.gidhome.com/support/brand-resources/citing-gid/.

Vicente Mataix Ferrándiz Page 83 of 374

ONLINE RESOURCES BIBLIOGRAPHY

Page 84 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS

Chapter 4

Contact mechanics

Alan Turing is reported as saying

that PDEs are made by God, the
boundary conditions by the Devil!
The situation has changed, Devil has
changed places... We can say that
the main challenges are in the
interfaces, with Devil not far away
from them...

Jacques-Louis Lions
(1928 - 2001 AD, French
mathematician)

4.1 Introduction
When speaking about CCM we need to know that the phenomenon covers an extended range of problems, and
depending of the problem we want to solve we will need to consider a different numerical technique[BookWri06]. The
most general one, and the one we are interest on is the resolution employing FEM, which can be applied in a large
range of problem, from small to large deformations and from linear to NL material behaviour. The second methods
are the Discrete Element Method (DEM) which study the interaction between a large number of particles coming
into contact. Finally, multibody systems, which describe rigid bodies interacting between them, creating a mechanism,
and where contact may influence the behaviour of the system.
Despite the large number of topics covered by CCM, as we already mention, we will focus on the application of
FEM for the resolution of contact problems. This is consistent with the developments of the previous chapters are
classified on this type of method.

4.1.1 Historical outline

History of the contact mechanics is probably as long as the history of civilization. Practically any physical interaction
between object involves contact and friction, making then the invention of the wheel probably the first human invention
involving this problematic. What we know for sure[BookPop10] is that the ancient Egyptians already employed oil
reducing the friction between the wheels and the floor in order to facilitate the transportation of the pyramid blocks
(Figure 4.1).
We then can jump to the modern contributions of Leonardo Da Vinci in the Codex-Madrid I, who experimented
with the friction phenomenon. He studied the influence of the contact area on the friction, testing objects of the
same mass but different contact area, coming to the conclusion that the frictional force is proportional to the weight

Vicente Mataix Ferrándiz Page 85 of 374

4.1 Introduction CHAPTER 4. CONTACT MECHANICS

Figure 4.1: Representation of moving blocks during ancient Egypt. Source[BookPop10]

and independent of the contact area. This was a crucial influence to Charles Augustin Coulomb, French engineer,
to which we must thank many of the expressions used in computational mechanics[ArtPP15], like the extended
expression Fτ = µN, denominated Coulomb friction law. It is in 1785[ArtCou85], when in The theory of simple
machines differentiated for first time between kinetic and static friction.
The first mathematical contributions we have records off are attributed to Euler [ArtEul50a; ArtEul50b]1 , who
studied the friction problem assuming that the roughness of the contact surfaces can be represented with a series of
triangles. This leads him to the conclusion that the static friction coefficient is larger than the dynamic one, which
means we need to apply a larger force in order to starting to move an object than once that the object is moving. In the
same way that we own the use of the Greek letter π to represent the circumference ratio, among other Greek letters
used as mathematical symbols, we own to Euler the use of µ in order to represent the friction coefficient.
During a long time, the contact conditions were modelled in a very experimental manner, as the lack of analytical
results did not allow consider an alternative approach. It was Hertz with his2 On the contact of elastic solids[ArtHer82]
who presented the first analytical solution for a contact problem. In 1882, Hertz solved the contact problem of two
elastic bodies with curved surfaces (Figure 4.2). This problem is so relevant that still today,this classical solution
provides a foundation for modern problems in contact mechanics, and it is extensively used as the main benchmark
for contact mechanics today[ArtSpe], as we will see in following sections.

Figure 4.2: Original Hertz paper in 1882. Source[ArtHer82]

The developments that happened just after that were motivated for its application to railways, reduction gears and
to rolling contact bearing industry when Industrial Revolution was at full steam at the beginning of the XX th century.
The analytical solutions that appeared during this time only allowed to solve a very limited rank of problems, limited to
1 Being widely considered the most prolific mathematician of all time, there was almost no field that he did not study. As Pierre-Simon Laplace

said: "Read Euler, read Euler, he is the master of us all".

2 In German in the original version: Über die berührung fester elastische Körper und über die Harte

Page 86 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.1 Introduction

only certain simple geometries, linear material mainly, and not much more. This contrasted with the industrial needs,
where the solution of complex geometries, not trivial BC and NL materials. Not only that, industrial needs implied
even more complex behaviours, with friction, wear, adhesion, large deformations and large sliding[BookYas13].

In 1933 Signorini[ArtSig33] formulated the general problem of the equilibrium of a linear elastic body in frictionless
contact with a rigid foundation, what is also denominated unilateral contact problem. Later more mathematical
developments came, like the contributions of Fichera[ArtFic63] on the uniqueness of variational inequalities. Later
contributions like the book of Kikuchi and Oden[BookKO88] who extended the prove of Fichera to the Signorini
problem.

In any case, until the existence of modern computation, for industrial applications contact was modelled as a local
problem using the stress and strain field obtained directly from the analysis of a complete structure[BookWri06]. Once
the computational power raised in the second half of XX th century the whole NL constraint of the contact problem
could be considered. At the beginning only some semi-analytical problems could be solved, but after the appearance
of modern FEM with NASA STRucture ANalysis (NASTRAN) more and more developments have emerged. In a
first stage, only Signorini’s problem was solved, limiting the resolution of unilateral contact, in furthers steps friction
was included, later large deformations and finally bilateral multibody contact.

We can also mention some additional problematic that emerged in order to treat the frictional problem. While the
frictionless problem can be easily formulated as minimisation problem with inequality constraint following standard
approaches (barrier, PM, LMM, ALM, ...), see the respective deductions from the annexe D.Constrained optimisation
problems. There is not such associated minimisation principle for the fricional contact problem, as Kikuchi and
Oden[BookKO88] proved. This is due to the dependence of the frictional status with the normal contact pressure,
which at the same time induces additional second order dependencies (such geometrical configuration). Since then
the frictional problem has been proved as a complex problem with many challenges.

In order to tackle these problematic, several approaches to work on these problematic. Some developments came
from the replacement of a variational inequality with a variational equality with a modified contact term, which allowed
to consider the classical optimisation techniques. More techniques have been proposed, Wriggers[BookWri06] list
a wide list of alternatives emerged, like simplex method, parametric quadratic programming, the flexibility method,
Nitsche method, direct elimination, cross constraint, among others.

Additionally to classic contact mechanics developments on the Tribology theory was made. We can mention
as examples the contributions related with the adhesive contact by Johnson et al.[ArtJKR71] and Derjaguin et
al.[BookMau13]. These contributions were in conflict between them, leading to the two most representative models
of adhesive elastic contact, the Johnson, Kendall and Roberts (JKR) and Derjaguin, Muller and Toporov (DMT)
respectively, these two methods are studied and compared in Barthel[ArtBar08]. More recent developments by Talon
and Curnier [ArtTC03] couple the adhesive model with frictional contact.

Furthermore, on the Tribology domain is important to mention the contributions from Bowden and Tabor [ArtBT39;
BookBBT01] in the mid-twentieth century. They were the first to emphasise the importance of surface roughness for
bodies in contact, due to the microstructure of the surfaces in contact, the true contact area between friction partners
is found to be less than the apparent contact area. Additionally we can mention the work of Kragelsky [ArtKra68], who
was one of the first on doing developments related to the wear phenomenon.

4.1.2 Contact problem

Has we already said, we can find contact and friction in almost all kinds of movements, both in nature and in
human-origin ones. We can find this problem[BookWri06] in many engineering applications, p.e. on civil engineering
the foundation problem requires to take into account the contact3 , but traditionally, restricting the solution to small
deformations, several simplifications were taken into account. Other problems related to this field of engineering, more
industrial-related, can be the study of bearing and connections on metallic structures. The applications in mechanical
engineering may even be applicable in more problematic, like gears and bearings, metal forming, cutting processes,
rolling contact of car tyres, crashing among other problems. Extending this field to the domain of the biology, we can
find in biomechanics problematic such as human joints, teeth implants, stents, etc.
3 It is extensively known the solution of Boussinesq[BookBou85], which considers several simplifications in order to represent the elastic

support.

Vicente Mataix Ferrándiz Page 87 of 374

4.1 Introduction CHAPTER 4. CONTACT MECHANICS

The physics of the contact interaction is particularly rich and complicated[PhDYas11], due to the multiscale and
multiphysical nature of the phenomenon. This kind of mechanical problems are classically formulated as a boundary
value problems, where the contact constraints are formulated as sets of inequalities, and the problem becomes even
more complex when the frictional effect is assumed. A whole discussion about the constraints and optimisation can
be found in the Appendix D.Constrained optimisation problems. In fact, a proper treatment of the contact constraint
requires the consideration of the micromechanical approach, not only for the proper consideration of the frictional
behaviour, but also in the normal contact[BookWri06].

These boundary conditions are solution dependent, which naturally leads to difficulties in the formulation of the
frictional contact problem, for example considering the Coulomb’s friction law yields to a non-smooth[BookAnd01]
energy functional. Additionally the contacting bodies may penetrate each other or be separated, where with the FE
discretisation leads to mathematical and numerical difficulties. Another challenge is the boundary discretisation and
methods to surpass this problematic. These points are discussed on the state-of-the-art section (4.2.State of the Art
in computational contact mechanics) and the formulation section (4.3.Formulation).

Finally, the algorithm considered for the detection phase can suppose a significant bottleneck in terms of efficiency.
This is discussed and presented in deeper detail in the corresponding section 4.4.Contact detection. Search
techniques.

First we should define the basic contact problem, Figure 4.3 illustrates a common contact problem. Every contact
problem is defined between two entities (not necessarily different, e.g self-contact) here noted as Ωi for i = 1, 2. The
contact problem occurs on the interface, where we want to link the displacements of the first domain (x1 ) with its
projection over second domain (x̂2 ) in a local reference frame (n, τ 1 and τ 2 ). The contact problem will consist in avoid
the penetration in the normal direction between domains, and respect the tangent movement restriction imposed by
the frictional component of the contact.

Figure 4.3: Basic definition of the contact problem

It is important to distinguish that we can divide two types of contacts, the bilateral and unilateral, where the first
one supposes the contact between two or more deformable bodies, in contrast to the unilateral contact, where the
contact occurs between a rigid solid and a deformable solid. A priori for the problems of metal forming the second one
is enough to formulate the problem, but in order to consider more advanced problems in this field a priori is preferable
to consider the bilateral case for being more general.

On Figure 4.4 the different states of the contact problem are presented. In here we can differentiate each one of

Page 88 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.2 State of the Art in computational contact mechanics

Figure 4.4: Contact states. Based on[PhDYas11]

them, with its corresponding idealised simplification. The illustration shows two different bodies4 , with its corresponding
denominations as master and slave. In the interface between the two bodies there is a local reference frame, defined
by the normal and tangent directions on this interface. The states listed on this image can be listed as:

• a) Frictionless contact: This contact allows the movement in the tangent direction of the local reference frame,
but does not allow the movement toward the opposite body. The idealisation will lead to a perfect rail, no friction,
between the two bodies. This is what is usually denominated as Karush-Kuhn-Tucker (KKT) condition.

• b) Separation: This state is the absence of contact, meaning that the two bodies are not in contact any more
and therefore there is not interaction between them any more.

• c) Stick state: In the case of the frictional contact, before the threshold value is surpassed, whatever model of
friction is considered, the two bodies will be fully tied,acting as one body. The idealisation will be this case will
correspond with a sewing in the interface.

• d) Slip state: Once this threshold is surpassed, the bodies can move freely in the tangent direction. Will be
the same as the case a), but in this case the rail which idealises the interface will not be ideal, and therefore
hindering the movement due to the present friction.

4.2 State of the Art in computational contact mechanics

4.2.1 Introduction
In this section we compare the existing methods, we can consider in order to solve the contact problem. Due
to the huge industrial and engineering interest in this type of problem, many developments exist, and therefore
we will focus on the most relevant ones on the sake of brevity. First, we will introduce the discretisation methods
available, following with the optimisation procedures necessaries to fulfil the contact inequality constraint. Finally, a
brief introduction to existing techniques available to model the frictional behaviours will be presented. We address
to some additional monographs, which will provide a deeper outline of the State of the Art in CCM. We found
particularly relevant to highlight the book of Wriggers[BookWri06], Laursen[BookLau10], Schweizerhof [BookSK13]
and Yastrebov [BookYas13].

4.2.2 Discretisation
In order to solve the contact problem, several approaches can be considered from the point of view of the discretisation.
When we mention discretisation, we refer the procedure followed for the integration of the interface. The following
ones are the main methods that can be considered from this perspective.
4 As previously stated, the contact may occur between more bodies, but here for the sake of simplicity, and in order to present the problem we

will consider only two bodies.

Vicente Mataix Ferrándiz Page 89 of 374

4.2 State of the Art in computational contact mechanics CHAPTER 4. CONTACT MECHANICS

4.2.2.1 NTN

Figure 4.5: NTN. Inspired[PhDYas11]

The discretisation of the contact interface is between a node of the slave domain to a node of the master domain
(Figure 4.5), reason why the method is denominated this way. This method, originally proposed by Francavilla and
Zienkiewicz[ArtFZ75] in 1975 is the oldest and the simplest of all the discretisation methods available. We can list the
pros and cons of the method as follows:

• Pros:
◦ Passes the Taylor test[ArtTP] when the mesh is conforming
◦ Simple conceptually and to implement. As it is node to node, which means that relate directly DOF, it is
possible to consider all kinds of constraint enforcement methodologies.

• Cons:
◦ Small slip and small deformations. The method loses precision when the nodes move across the interface
and the nodes are not coincident any more.
◦ Related with previous method, the mesh must be conforming between interfaces. This constraint introduces
a restriction in mesh generation.

4.2.2.2 NTS

Figure 4.6: NTS. Inspired[PhDYas11]

The discretisation of the contact interface is between a node of the slave domain to the surface of the master
domain (Figure 4.6). Despite of being quite close in time to the NTN, in 1977 Hughes et al.[ArtHTK77] proposed the
NTS, which already allowed to solve problems in large deformations. We can summarise the pros and cons of the
method in:

• Pros:
◦ Simple and robust. The implementation is probably the most extended in FEA software.
◦ Large deformations and large slip can be considered
◦ Mesh independent. The meshes are not required to be conforming in the interface like in the NTN

Page 90 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.2 State of the Art in computational contact mechanics

• Cons:
◦ Fails Taylor test[ArtTP] for nonconforming meshes, except if considering the double pass LMM, solving
the problem twice, swapping master and slave. This problem is solved with the modification suggested by
Zavarise[ArtZL09].
◦ It has several difficulties to calculate the gap in order to compute the contact. In the work of Yastre-
bov [PhDYas11] several alternatives to solve this issue are proposed.

4.2.2.3 CDM

Figure 4.7: CDM. Inspired[PhDYas11]

The discretisation of the contact interface is based on a full triangulation of the zone between contacting surfaces
based on surface nodes (Figure 4.7). This method is in fact a full symmetric NTS discretisation proposed originally
by Oliver [ArtOli+09a; ArtOli+09b] and Hartmann[ArtHar+09]. We can list the main advantages and disadvantages
as:

• Pros:
◦ Passes Taylor test[ArtTP]
◦ Large deformations and large slip can be considered like in the NTS
◦ In-house developed in CIMNE[ArtOli+09a; ArtHar+09]
• Cons:
◦ Mesh dependent, in part related by next point
◦ Triangulation problems may happen for 3D cases[ArtHar+10]

4.2.2.4 STS (Mortar methods)

Figure 4.8: STS. Inspired[PhDYas11]

In the STS discretisation of the contact interface is between a surface of the slave domain to a node of the
master domain (Figure 4.8), it is also denominated Mortar method as a metaphor to the strong union between
bricks in a wall. As a note of curiosity, the method was originally developed on the field of Domain Decomposition
Method (DDM)[BookWoh01; BookTos05], but originally proposed for CCM by Simo[ArtSWT85]. The method can be
summarised as:

• Pros:
◦ Passes Taylor test[ArtTP], even with different types of mesh combinations, i.e. tetrahedra with hexahedra
meshes as shown in the section for numerical examples 4.5.Numerical examples

Vicente Mataix Ferrándiz Page 91 of 374

4.2 State of the Art in computational contact mechanics CHAPTER 4. CONTACT MECHANICS

◦ Correct integration of contact forces. The method is consistent as both meshes are fully integrated.
◦ Large deformations and large slip can be considered[ArtPL04]. The Mortar based formulation leads to a
consistent formulation of the frictional contact problem for large sliding and large deformations.
◦ Additionally the Mortar formulation has been shown as being general enough. As a method coming from
DDM it can be considered in order to couple different types of problems in a Multiphysics way. The method
can be considered on mapping techniques (see our appendix E.Mortar mapper) or to strongly couple
problems with mesh tying techniques[PhDPop12]. The work of Seitz[ArtSWP18] shows a fully integrated
Multiphysics with thermo-elasto-plastic frictional contact, or the FSI implementation in Popp[PhDPop12]
thesis. The consideration of DLMM (4.2.3.4.DLMM) extends the application range of the formulation.

• Cons:

◦ Complex implementation in 3D. We experienced great advances since Puso and Laursen[ArtPL04] and
later thanks to Brunssen[PhDBru08], Popp[PhDPop12] and Gitterle[PhDGit12]. The complexity comes from
the 3D intersection between two flat geometries, Figure 4.9, which requires some additional considerations,
especially when the Gateaux derivatives (see from appendixes 4.6.Derivatives for contact mechanics
linearization).

Figure 4.9: 3D segmentation between two triangles

4.2.2.5 Other alternative methods

The following section introduces some additional integration approaches existing in the state of art of the CCM. Many
of these approaches cannot be directly applied in a standard FEM formulation, but we consider them as interesting to
mention.

4.2.2.5.1 Isogeometric :

In the recent years, the isogeometric analysis has been on the rise. The key concept underlying on this method is
the meshless integration. The integration is done directly on the NURB, avoiding them the necessity of the mesh
step during the preprocess. This method originally developed by Hughes[BookHBC09] has grown significantly on
last years due to the continuous workflow integration between FEA and CAD. Additionally NURB provide an exact
representation of the surfaces and high order of integration during the FE computations.

The method has been considered particularly in order to represent surfaces, which suits to thin objects such as the
ones present during the forming simulations. Additionally to the corresponding shell formulation[ArtBen+11] contact
formulations are needed. We can mention respect this, the works from Lorenzis[ArtLWZ12a; ArtLWZ12b].

Page 92 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.2 State of the Art in computational contact mechanics

4.2.2.5.2 Smooth surface approximation :

In order to overcome the problematic from the discontinuity in the standard FEM interface, we may define a smooth
approach of the boundary[BookBHS07]. In order to do so, we can consider different types of discretisations, such as
Hermite, splines, Bézier surfaces or NURB.

Sometimes this last approach can also be denominated Segment-To-Analytical-Surface (STAS)[BookSK13],

and can be found between the previous method (4.2.2.5.1.Isogeometric) and the standard FEM. There are different
ways to approach analytically the surface. We can define this surface at the beginning of the simulation and then move
it, which implies that the surface is rigid and then limits the contact to unilateral cases[ArtWI93]. This is not necessarily
a limit, if we think that in the case of forming processes the tools considered are de facto rigid. The second approach
can consist in consider standard FEM, with two deformable bodies and update the analytical surface continuously,
this is for example done with Nagata[ArtNag05] patches by Neto[ArtNet+14; ArtNOM17].

4.2.2.6 Conclusion

After all the methods presented, and due to our consideration of standard FEM, we will consider as discretisation
method the STS or Mortar approach (4.2.2.4.STS (Mortar methods)). This method will provide us the best standard
FEM integration possible, despite its technical problems related to implementation details.

4.2.3 Optimisation method

The assumption of a known a priori contact surface allows to replace the variational inequality by a variational equality
with an additional contact term. The form of this contact term depends upon the choice of the optimisation method.
Owing to the fact that we have decided to implement a Mortar approach discretisation based in Popp[PhDPop12], our
initial thought could be to consider as optimisation method the DLMM. In the following section, we will show the most
common approaches that could be taken into consideration in order to solve the CCM problem, justifying this way our
final choice.

The main optimisation methods are studied in detail in the corresponding appendix, we directly address to
D.Constrained optimisation problems for further detail about these methods. In the mentioned appendix not only the
different methods are introduced, but also compared in detail between them, and its suitability for solving different
types of constrained problems.

4.2.3.1 Penalty Method (PM)

This is probably the most extended optimisation method considered on CCM, particularly in explicit approaches. The
contact conditions are fulfilled exactly only in case of the infinite penalty parameter (ε), which results in ill-conditioning.
For more details check D.2.Penalty method.

• Pros:

◦ Very simple and robust. The simplicity of the method can be seen in the corresponding section in the
optimisation appendix D.2.Penalty method, and its robustness in the last numerical example shown
(D.7.3.Over-constrained optimisation problem).
◦ Pure displacement based formulation, no change in the system size.

• Cons:

◦ The solution is inexact, which happens only in case of an infinite value of ε, making the system unsolvable
due to the deterioration on the condition number.
◦ Choice of penalty parameter. If the value is chosen to small the penetration is large, if large as we have
stated will ill-condition the system.

Vicente Mataix Ferrándiz Page 93 of 374

4.2 State of the Art in computational contact mechanics CHAPTER 4. CONTACT MECHANICS

4.2.3.2 LMM
Contact conditions are exactly satisfied by the introduction of an extra DOF called LM, and usually represented by the
Greek letter λ. This and more is analysed in detail at D.3.Lagrange Multiplier method.

• Pros:
◦ Exact solution is obtained when solving the system of equations. This is the main advantage of the LMM,
and the reason of its extensive use in optimisation problems.
◦ User independent. It does not depend, as in the case of the PM of a user-driven decision to choose a
parameter.

• Cons:
◦ Additional DOFs. The LHS of the system grows in ncontact additional DOF for the case of frictionless
contact, and in 3 × ncontact in the case of frictionless with full λ components or for frictional problems. Here
ncontact is the number of nodes in contact.
◦ Moderate convergence rate in the NR. Additionally the condition of the system is affected due to the zero
diagonal terms introduced by the Lagrange multipliers.

4.2.3.3 ALM
It is a LMM regularised by a PM. It yields a smooth energy functional and fully unconstrained problem, resulting
in exact fulfilment of contact constrains with finite value of the penalty parameter (ε). For more details check
D.4.Augmented Lagrange Multiplier method. The method is successfully considered with a Mortar approach by
Cavalieri and Cardona[ArtCFC12; ArtCC12; ArtCC13a; ArtCC15; ArtCC13b].
The method is often considered as a synonym of the Uzawa iteration (see D.4.2.2.Uzawa iteration). Indeed the
Uzawa iteration is always applied to the ALM, but this is just one of the possible approaches possible to be considered
in order to solve the system. In the following when we mention the ALM we will refer to the standard approach
without Uzawa iteration. This is due to the reason that the convergence order of the Lagrange multiplier becomes
linear.

• Pros:
◦ Exact solution as in the case of the LMM. The result is not influenced by the penalty terms (ε).
◦ No additional DOFs with Uzawa iteration algorithms. Otherwise the system of equations is increased in
the same way it is done in the standard LMM.
◦ Smooth functional is obtained. See Figure D.4 from the appendix, compared with the Figure D.3.
◦ Less sensitive to penalty choice compared with the PM. The penalty (ε) only contributes in smooth L, but
does not affect the solution. Once the system has converged, the contribution of ε becomes zero.

• Cons:
◦ In the case of an Uzawa iteration algorithm coming comes at the price of increased computational costs
due to an additional nonlinear iteration loop (augmentation), typically only showing a linear convergence
rate.

4.2.3.4 DLMM
This concept of dual Lagrange multiplier was introduced first by Wohlmuth[ArtWoh02]. The method is an extension of
the LMM, but the Lagrange multiplier (λ) are interpolated considering a different set of shape functions. These shape
functions are locally supported and continuous dual basis functions.

• Pros:
◦ Exact solution as in the LMM case

Page 94 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.2 State of the Art in computational contact mechanics

◦ System can be condensed, reducing the DOF respect the original, the system then becomes pure
displacement.
◦ With regard to linear solvers, DLMM allows for an out-of-the-box application of state-of-the-art iterative
solution and preconditioning techniques, such as the Generalized Minimal RESidual (GMRES) iterative
solver. It can be used with Algebraic MultiGrid (AMG) methods if the proper modifications are taken into
account, like recent works of Wiesner [ArtWie+18].

• Cons: As the disadvantages of the LMM disappear, there is none. We can mention the fact that is slightly more
difficult to implement in comparison to the standard LMM.

4.2.3.5 Other alternative methods

The following are some other additional approaches possible to solve the optimisation problem related with contact.
Here we will add the ones we have considered as most promising after the already presented, more methods are
mentioned in the optimisation appendix (D.1.Introduction).

4.2.3.5.1 Perturbed Lagrangian :

This special formulation can be used to combine both PM and LMM in a mixed formulation, therefore, similar
to the already mentioned ALM. The method was considered originally by Oden[ArtOde81] in 1981. In this method
the contribution of the Lagrange multiplier is regularised with the inclusion of a complementary term. This method
is discarded due to its limitations in order to consider the slip case in frictional simulations, where this is usually
done with the inclusion of an incremental constitutive equation for the frictional behaviour considered[BookWri06].
In addition to that we can mention as an advantage that the method does not deteriorate the conditioning of the
system.

4.2.3.5.2 Nitsche :

Nitsche[ArtNit71] methods are based on a different concept respect the LMM. In here the stress vector on the
interface is computed from the stress field inside the solid body. The formulation usually includes a penalty term
to avoid ill-conditioning, but in the same way as the ALM the constraint is enforced exactly and this term does not
have effect in the final solution. The resulting formulation does not introduce additional DOF, but as it formulates the
stresses from the displacement field, this method becomes complex to formulate on NL cases.

An extended overview of recent advances of the method applied in CCM can be found in the work of Chouly [ArtCho+17].
Here additional references can be found.

4.2.3.5.3 Other minor mentions :

We can mention additionally, some of them came directly from techniques applied on the DDM[BookWoh01]
field. i.e. the Finite Element Tearing and Interconnect (FETI) methods[ArtDFS98] or monotone methods[ArtKor94;
PhDKra01].

Another recent relevant contribution is the work of Hiermeier [ArtHWP18], who introduces a gradient based Mortar
formulation, still based on LMM and ALM, which provides a symmetric system of equations to respect the active-set
contributions. A symmetric LHS has advantages as it allows to consider certain types of iterative solvers such as the
Conjugate Gradient (CG).

4.2.4 Conclusion
After all the methods here introduced, we conclude that we will combine two methodologies. In the Popp’s work[PhD-
Pop12] and derived ones, the DLMM is considered in combination of a Non-Linear Complementary Function
(NCP). The resulting system is close to a ALM solution. We will take the dual shape functions considered on the
DLMM in combination with the ALM approach used by Cavalieri and Cardona[ArtCFC12]. This means that we will
work with ALM displacement only approach due to the consideration of DLMM, but without using the Uzawa iteration
algorithm.

Vicente Mataix Ferrándiz Page 95 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

Additionally we will consider the Adapted Augmented Lagrangian Method (AALM) by Busseta[PhDBus;
ArtBMP] in order to enhance the method. Check D.4.3.1.Adapted Augmented Lagrangian Method for further
detail.

4.2.5 Frictional models

The frictional problem is quite complex by itself, but in this work we will
focus on simple models that will allow to introduce the phenomenon
in our simulations, not taking into account the dynamic effects of the
friction[ArtPPS03]. Basically, the frictional behaviours are addressed in
this work are limited to simple frictional laws as Coulomb and Tresca. The
basic definitions of these frictional laws are such as the follow[BookWP99],
Equation (4.1). In here we can see as the Tresca law depends only in
a constant threshold parameter g, meanwhile in the case of Coulomb it
depends on the normal reaction and the friction coefficient (µ).

(
Tresca: kFT k ≤ g
(4.1)
Coulomb: kFT k ≤ µ |FN |

In addition to these frictional models, there exist more advanced fric-

tional models, like the ones in Figure 4.10. These laws express that the
friction no longer depends on the normal force when the latter surpasses
a certain threshold. It is a kind of saturation of the friction threshold, which
is often considered in metal forming. The Coulomb cone can become
a truncated cone like in the Coulomb-Orowan and Shaw or a cylinder Figure 4.10: Friction cone for variants of
(Tresca). It is also possible to regularise the Coulomb law[BookWP99], the Coulomb law. Inspired [ArtRao16]
this smoothes the law reducing part of the numerical problematics of the original piecewise function. i.e. square root
regularisation, hyperbolic tangent or a piecewise polynomial.
Besides of the pure frictional behaviour, the frictional models can incorporate additional effects, such as the wear,
adhesion or variational evolutions in the friction coefficients[ArtRao16]. For more detailed frictional behaviours, we will
address literature at the end of the section.
It is commonly admitted in the literature an analogy between plasticity and friction. In here the question of
applicability of plasticity principles to frictional contact problems remains open. Indeed, the formulations of several
frictional laws and its respective numerical procedures have been derived, many times taking into account on this
analogy. The work of Antoni[ArtAnt17] analyses this phenomenon in detail.
Finally, we want to address some relevant works in the field of frictional behaviour in order to extend the short
summary presented here. The book of Popov [BookPop10] is a very complete work about the friction in contact. We
can also mention to more forming oriented manuscripts as Boisse[BookBAL03].

4.3 Formulation
4.3.1 Introduction
The following section presents the formulation employed in the derivation of the frictional Mortar contact con-
dition formulation with Augmented Dual Lagrange Multiplier (ADLM), based mainly in the work of Alexander
Popp[PhDPop12; ArtPop+10], but also on Cavalieri and Cardona[ArtCC12; ArtCC13b] and the work of Yastre-
bov [PhDYas11; BookYas13]. In addition to the mentioned work, this author has incorporated its own consideration,
which will be highlighted.
The contact mechanics problems are based on the Initial Boundary Value Problem (IBVP) of non-linear solid
mechanics and the unilateral contact constraints. After recapitulating some basic notation and the strong formulation,
a weak formulation of the contact mechanics problems with two subdomains will be introduced. In contrast to the

Page 96 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

mesh tying case, see A.3.Mesh tying, unilateral contact leads to a constrained minimisation problem with inequality
constraints, or more generally to so-called variational inequalities. It should be mentioned that both frictionless and
frictional contact can either be formulated as variational inequalities with a constrained solution or as saddle point
problems based on Lagrange multipliers, where the focus will be on the latter approach here.

The motivation for dual Lagrange multipliers[BookWoh01; ArtWoh02] lies in the fact that an extension of the master
side basis functions to the slave side of the interface has a global support for standard Lagrange multipliers.

4.3.2 Definition of the problem

The basic fundamentals of a contact problem, featuring potentially large deformations and large sliding[ArtYL08;
PhDPop12; BookWri06] will be formulated in the following lines (Figure 4.11). Reference configurations of two
containing bodies are denoted by open sets Ω1 and Ω2 , and the deformations φ1 and φ2 of these two bodies are
to be found. We call Γu where the Dirichlet boundary conditions are prescribed, Γσ where the Neumann boundary
conditions are prescribed, and Γc where the contact constrains will be defined and enforced. The spatial counterparts
are denoted as γu , γσ ,and γc .

Figure 4.11: Definition of the contact problem

4.3.3 Frictionless contact

Before introduce the formulation employed for the frictional contact, the formulation for the frictionless will be introduced.
Many terms are reused and common in both formulation, so the theoretical introduction for the frictional case will
require a smaller effort.

4.3.3.1 Strong formulation

On each subdomain Ωi0 , the IBVP of finite deformation elastodynamics needs to be satisfied, viz (4.2). In here we are
adding the BC terms to the linear form from the FE chapter, 2.3.1.Linear form, and splitting in two different domains
(i = 1, 2) the problem. This includes the Dirichlet BC in (4.2b), the Neumann BC from (4.2c) and the initial BC defined
in (4.2d) and (4.2e).

Vicente Mataix Ferrándiz Page 97 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

(4.2a) ∇ · σ i + bi = ρi üi in Ωi × [0, T ]

(4.2b) ui = ui on Γiu × [0, T ]
(4.2c) σ i · ni = ti on Γiσ × [0, T ]



(4.2d) ui Xi , 0 = ui0 Xi in Ωi0



(4.2e) u̇i Xi , 0 = u̇i0 Xi in Ωi0

The contact constraints in normal direction are typically

given in form of Hertz-Signorini-Moreau (HSM) conditions
as given in (4.3), and Figure 4.12. In optimisation theory these
conditions are denominated usually as KKT5 .
In the course of deriving a weak formulation, the balance
of linear momentum at the unilateral contact problem for the
interface Γic is typically exploited and a Lagrange multiplier
vector field λn is introduced, thus setting the basis for a mixed
variational approach. Unilateral contact constraints are typi-
cally formulated (and later also numerically evaluated) in the
current configuration.

(4.3) gn ≥ 0 , pn ≤ 0 , pn gn = 0 on Γic × [0, T ]

Figure 4.12: KKT conditions of non-penetration.

4.3.3.2 Weak formulation
We will distinguish two different cases for the weak formulation, the first one considering the LM as a scalar variable.
The second one considering the LM decomposed in its Cartesian components, which means that the contact pressure
can be represented as n · λ.

4.3.3.2.1 Scalar Lagrange multiplier :

In this case the LM corresponds directly with the contact pressure, which is the complementary form of the normal
gap (gn ). We present this form first, as it is simpler than the components form, where we will simply present the parts
of the formulation changing respect the scalar form.

4.3.3.2.1.1 LMM :
The general theory for the LMM is presented and detailed in D.3.Lagrange Multiplier method. In the relative to the
resolution of the contact problem, to start the derivation of a weak formulation of (4.2), appropriate solution spaces U i
and weighting spaces V i need to be defined as (4.4).

( 
U i = ui ∈ H 1 (Ω)kui = ûi on Γiu ,
(4.4) 
V i = δ ui ∈ H 1 (Ω)kδ ui = 0 on Γiu

Additionally the Lagrange multiplier vector λn = λn · n = −t1c , which enforce the unilateral contact constraint(4.3),
represents the negative slave side contact traction t1c , is chosen from a corresponding solution space denoted as
M.
In terms of its classification in functional analysis, this space represents the dual space of the trace space L1 of
V . In the given context, this means that M = H 1/2 (Γc ) and L1 = H 1/2 (Γc ), where M and L1 denote single scalar
1

components of the corresponding vector-valued spaces M and W .

5 Moreau[ArtMor73] formulated this on 1974 expressed as sub-gradients. The expression from Karush[ArtKar39] is previous, from 1939.

Page 98 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

Based on these considerations, a saddle point type weak formulation is derived next. This can be done by
extending the standard weak formulation of non-linear solid mechanics as defined to two subdomains and combining
it with the Lagrange multiplier coupling terms introduced in generic form. Find ui ∈ U i and λn ∈ M such that we
obtain (4.5a), then once derived (4.6).

Z
(4.5a) Lco (u, λn ) = λn · gn dΓico
Γ1c

Where gn is the continuous normal gap, that can be defined as (4.5b).

(4.5b) gn = n1 · u1 − u2

(4.6a) δL(u, λn ) = δLV + δLM

(4.6b) δLV = −δLkin (ui , δ ui ) − δLint,ext (ui , δ ui ) − δLco (λi , δ ui ) = 0 ∀δ ui ∈ V
(4.6c) δLM = − δLλ (ui , δλi ) ≥ 0 ∀δλi ∈ M

Herein, the kinetic contribution δLkin , the internal and external contributions δLint,ext and the unilateral contact
contribution δLco to the overall virtual work on the two subdomains, as well as the weak form of the unilateral contact
constraint δLλ , have been abbreviated as (4.7).

2 Z
X 
(4.7a) − δLkin (u) = ρi üi · δ ui dΩi
i=1 Ωi
2
"Z   Z #
X ∂σ i ∂δ u̇i
(4.7b) − δLint,ext (u) = : δ u̇i + σ i : − b · δ ui i
dΩ − i i i
t · δ u dΓσ
Ωi ∂ xj ∂ xj Γiσ
i=1
Z
(4.7c) − δLco (u, λn ) = λn · δ gn dΓ1co
Γ1c
Z
(4.7d) − δLλ (u, λn ) = δλn · gn dΓ1co
Γ1c

The coupling terms on Γc also allow for a direct interpretation in terms of variational formulations and the principle
of virtual work. Whereas the contribution in (4.7c) represents the virtual work of the unknown interface tractions
λ = t1c = t2c , the contribution in (4.7d) ensures a weak, variationally consistent enforcement of the unilateral contact
constraint (4.3). Nevertheless, the concrete choice of the discrete Lagrange multiplier space Mh in the context of
mortar finite element discretisations is decisive for the stability of the method and for optimal a priori error bounds.
Finally, it is pointed out that the weak formulation (4.6b) and (4.6c) possesses all characteristics of saddle point
problems and Lagrange multiplier methods.
In contrast to the mesh tying case, where this mapping only came into play in the discrete setting, γc1 and γc2
cannot even be guaranteed to be identical in the continuum framework for unilateral contact, because they not only
comprise the actual contact surfaces but the potential contact surfaces.
As compared with the mesh tying case, it is noticeable that the weak formulation contains inequality (4.6c)
conditions for unilateral contact. These require a particular numerical treatment based on active set strategies.

4.3.3.2.1.2 Penalty :
This optimisation method is better detailed in the corresponding section of the appendix, see D.2.Penalty method.
The main advantages of this method is that does not require to add additional DOF to the system of equations,
reducing the system of equations to be solved and that it does not introduce ill-conditioning problems to the system

Vicente Mataix Ferrándiz Page 99 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

of equations resolution due to the introduction of saddle point in the formulation. The main disadvantage is that it
never achieves the exact solution as this would require the consideration of an infinite penalty, which will induce
ill-conditioning in the system. Therefore the main problem of this method is the chosen of the proper penalty value,
enough to solve approximately the problem without the cited problematic.
The main difference from the formulation presented in (4.7), is the lack of (4.7d), as our formulation will no deal
with LM. Additionally (4.7c) will be rewritten as in (4.8b), which comes from (4.8a). In here the εn is a positive penalty
parameter corresponding to the normal contact.

Z
1
(4.8a) − Lco (u) = εn · δ gn2 dΓ1co
2 Γ1c
Z
(4.8b) − δLco (u) = εn · δ gn dΓ1co
Γ1c

4.3.3.2.1.3 ALM :
One of the main disadvantages of the standard Lagrange multiplier is the saddle point problem that appears in
the formulation. For solving that, an Augmented Lagrangian method to solve contact problems with friction was
proposed by Alart and Curnier [ArtAC91] based on a reformulation of the contact and friction laws into a system of
equations without inequalities. More details about ALM can be found in the optimisation appendix D.4.Augmented
Lagrange Multiplier method. The resulting Lagrangian (L) can be seen as a combination of the standard LM
Lagrangian (4.3.3.2.1.1.LMM) and the penalty Lagrangian (4.3.3.2.1.2.Penalty).
Focusing in the functional relative to the contact (Lco (u, λn ) = LV co + LM ), we can rewrite (4.5a) as (4.9).

Z
ε 1
(4.9) Lco (u, λn ) = k λn · gn + gn2 − hk λn + εgn i2 dΓico
Γ1c 2 2ε

Where ε is a positive penalty parameter, k is a positive scale factor, and hi is the Macauley bracket operator, that
is (4.10).

(
xx ≥0
(4.10) hx i
0x <0

This functional is C 1 differentiable saddle-point, as shown in Figure D.4, from D.4.3.Applicability on contact
problems. The solution is obtained as the set of values that render this functional stationary.
The solution does not depend on the value of parameters ε and k . Nevertheless, the convergence rate does
depend on their value. In numerical computations, default values of ε and k are selected in terms of a mean value of
the Young modulus (E) of the bodies in contact and of a mean value of mesh size, as (4.11). Numerical examples
show that this choice gives a better condition number of the iteration matrix than other choices. This is proven later at
4.3.3.3.Augmented Lagrange multiplier parameters calibration, where a numerical experiment is provided, to illustrate
on the influence of these parameters on the convergence properties.

Emean
(4.11) ε = k ≈ 10
hmean

The functional (4.9) can be separated in two different parts, as can be seen in (4.12).

Z (
k λn · gn + ε2 gn2 dΓico if k λn + εgn ≤ 0 (Contact zone)
(4.12) Lco (u, λn ) = dΓ1co
Γ1c − 2kε λ2n if k λn + εgn > 0 (Gap zone)

Page 100 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

Finally, we can derive (4.12) to obtain the variational form from (4.13), where to simplify we define the augmented
normal pressure λ̄n = k λn + εgn .

Z (
λ̄n · δ gn + kgn δλn if λ̄n ≤ 0 (Contact zone)
(4.13) δLco (u, λn ) = 2 dΓ1co
Γ1c − kε λn δλn if λ̄n > 0 (Gap zone)

The functional from (4.13) makes that the system obtained varies in function if the nodes are present in the contact
or the gap zone, so the system is not a priori known and in the following to present the numerical discretisation will
focus in the solution obtained in the gap zone. Once this is derived the solution in the gap zone can be obtained in a
straightforward way.

4.3.3.2.2 Vector Lagrange multiplier :

Here the vectorised, or by components, of frictionless contact formulation is presented. The main reason in
order to define it by this form is that the resulting system of equations can be statically condensed, therefore
removing the corresponding LM DOF from the system of equations, and by this solving a system of equations
purely based on displacement DOF. This is done taking in consideration properties from the dual Lagrange multiplier
which will be presented later on 4.3.3.4.1.Dual Lagrange multipliers. Of course the solution for the penalty method,
4.3.3.2.1.2.Penalty, remains identical, as there is not any LM taking roles on the formulation.

The main modification which takes place on this formulation is the replacement of the contact pressure LM,
or λn , by the consideration of a LM defined in the Cartesian components, represented as λ, where the normal
components (λn ) are different or equal to zero, and the tangential ones (λτ ) are always zero. This can be summarised
in (4.14).

(
λn = n · λ
(4.14)
λτ = λ − n · (n · λ) = 0

4.3.3.2.2.1 LMM :

From the formulation stated at 4.3.3.2.1.1.LMM, we modify the terms from (4.7) related to the LM, resulting (4.15).
(4.7c) will be reformulated as (4.15a) and (4.7d) into (4.15b).

Z


(4.15a) δLco (u, λ) = λ · δ u1 − δ u2 dΓ1co
Γ1c

Where û2 are the displacement from the master side projected on the slave side.
Z
(4.15b) δLλ (u, λ) = δ (n · λ) · gn − (λ − n · λ) (δλ − n · δλ) dΓ1co
Γ1c

4.3.3.2.2.2 ALM :

Taking as base the solution presented on 4.3.3.2.1.3.ALM, and the modifications introduced in the previous
section, we will need to define the augmented LM by components, as defined in (4.16).

(4.16) λ̄ = k λ + εngn

Vicente Mataix Ferrándiz Page 101 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

With this consideration, for the contact contributions of the potential, (4.12) will be reformulated as (4.17a) and
(4.13) into (4.17b) respectively. Additionally,(4.7c), which defines the LM contribution will be rewritten as (4.17c). On
these equations we will consider λn = k (n · λ) + εgn to simplify the expressions.

Z (

λ̄n · u1 − u2 + ε2 gn2 dΓico if λ̄n ≤ 0 (Contact zone)
(4.17a) Lco (u, λ) = dΓ1co
Γ1c − 2kε λ2 if λ̄n > 0 (Gap zone)

Z (

λ̄n · δ u1 − δ u2 + kgn δλ · n if λ̄n ≤ 0 (Contact zone)
(4.17b) δLco (u, λ) = 2 dΓ1co
Γ1c − kε λδλ if λ̄n > 0 (Gap zone)

Z
k2
(4.17c) δLλ (u, λ) = k (n · δλ) · gn − (λ − n · λ) (δλ − n · δλ) dΓ1co
Γ1c ε

4.3.3.3 Augmented Lagrange multiplier parameters calibration

The expression from Equation (4.11) is taken directly from literature[ArtCC12]. In here we present a simple patch
test[ArtTP] example, in order to show the influence of k and ε in the condition number (κ) of the LHS. We will see as
the ε affects always negatively to κ, this is proven in the optimisation appendix at D.7.2.2.3.Augmented Lagrange
multiplier method, meanwhile k can improve or worsen depending on the range value considered.

E Solid 1 ν Solid 1 E Solid 2 ν Solid 2

100Pa 0.3 100Pa 0.3
Table 4.1: Parameters considered for ALM parameters calibration

The properties of the materials considered are listed on Table 4.1, these properties are quite simple numbers,
this will allows later to compute easily the ALM parameters. Taking (4.11) and with the corresponding h≈ 10, our
reference values will correspond with ε = k = 100. In addition to the load considered in the top face of the punch block
is equal to 1Pa.

σmax (A) |λmax (A)|

(4.18a) κ(A) = (4.18b) κ(A) =
σmin (A) |λmin (A)|

(a) Mesh of the patch test (b) Displacement solution for the
Figure 4.13: Condition number study for the ALM

Page 102 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

The condition number (κ) of a function measures how much the output value of the function can change for a
small change in the input argument. A problem with a low condition number is said to be well-conditioned, while
a problem with a high condition number is said to be ill-conditioned. In order to compute the corresponding κ we
should evaluate the Singular Value Decomposition (SVD), in order to compute the maximal and minimal singular
values required in (4.18a). This expression can be simplified if A is normal (A∗ A = AA∗ ), in that case we can simply
compute κ considering the maximal and minimal eqigenvalues as shown in (4.18b)6 . This last approach simplifies
considerably the numerical effort required, as we dispose of optimal methods for the extreme eigenvalues, like the
power iteration for the greatest eigenvalue and the inverse power iteration for the smallest one respectively.

In Figure 4.13 we can see the proposed mesh for the 3D Taylor patch test (Figure 4.13a), as well as the
displacement solution of the problem (Figure 4.13b).

Figure 4.14: Condition number study graphic representation

Finally, the Figure 4.14 represents graphically the results from Table 4.2. We represent in the left the surface plot
without any additional consideration, on the other hand, on the contour plot the k axis considers a logarithmic scale.
The latter is due to the fact that the ε value is always increasing κ and therefore we do not require to consider a more
detailed scale for this variable, but in the case of k it provides us relevant information taking into account that we
modify the values in powers of 10 (Table 4.2). In this continuous representation, we can appreciate several things.
First of all, as previously stated, is that the ε increases always the condition number, and the k may improve or not
the κ depending of the value. Second of all we can see, particularly in the right figure like the value estimated from
Equation (4.11), 100 − 100, provides the best conditioning in overall.

4.3.3.4 Discretisation and numerical integration

4.3.3.4.1 Dual Lagrange multipliers :

4.3.3.4.1.1 Definition :

The discretisations of the displacements correspond with the standard ones in the finite element formulation, for
more information check the literature[BookZZT13]. In addition, an adequate discretisation of the Lagrange multiplier
vector λ is needed, and will be based on a discrete Lagrange multiplier space Mh being an approximation of M.
Thus, we can define the discrete Lagrange multiplier as (4.19), with the shape functions Φj and the discrete nodal

6 We are considering SVD as it is the proper definition of the condition number, but in fact this is a very expensive operation and can be applied

only in small systems. Another alternative if A is symmetric, then it is possible to compute the
√ ratio between the max and min eigenvalues of A. In
the case that A is not symmetric it is possible to compute the max and min eigenvalues of AT A.

Vicente Mataix Ferrándiz Page 103 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

k ε κ k ε κ k ε κ
1 1.00E-12 1.13E+05 10 1000 1.74E+04 1000 10 1.92E+04
1 1.00E-02 1.13E+05 10 10000 2.69E+05 1000 100 1.81E+04
1 1 1.13E+05 100 1.00E-12 1.74E+04 1000 1000 3.14E+04
1 10 1.18E+05 100 1.00E-02 1.74E+04 1000 10000 6.43E+04
1 100 1.63E+05 100 1 1.74E+04 10000 1.00E-12 6.64E+05
1 1000 6.15E+05 100 10 1.74E+04 10000 1.00E-02 6.63E+05
1 10000 2.69E+07 100 100 1.74E+04 10000 1 6.63E+05
10 1.00E-12 1.74E+04 100 1000 1.74E+04 10000 10 6.63E+05
10 1.00E-02 1.74E+04 100 10000 9.38E+04 10000 100 3.53E+05
10 1 1.74E+04 1000 1.00E-12 6.60E+04 10000 1000 3.26E+06
10 10 1.74E+04 1000 1.00E-02 6.60E+04 10000 10000 7.38E+05
10 100 1.74E+04 1000 1 5.75E+05
Table 4.2: Results of numerical experiment for ALM parameters

Lagrange multipliers λh .

1
m
X

(4.19) λh = Φj ξ 1 , η 1 λ j
i=1

Details on how to define dual Lagrange multiplier shape functions Φj using the so-called biorthogonality relationship
with the standard displacement shape functions Nk have first been presented in Wohlmuth[ArtWoh02]. A common
notation of the biorthogonality condition is (4.20), where Γ1co,h represents the discrete contact interface.

Z Z
(4.20) Φj Nk1 dΓico = δjk Nk1 dΓico , j, k = 1, ..., m1
Γ1co,h Γ1co,h

Herein, δjk is the Kronecker delta, and the most common choice m1 = n1 is assumed. For practical reasons, the
biorthogonality condition is typically applied locally on each slave element, represented with the index e, yielding
(4.21), where me1 represents the number of Lagrange multiplier nodes of the considered slave element.

Z Z
(4.21) Φj Nk1 de = δjk Nk1 de , j, k = 1, ..., me1
e e

Combining the biorthogonality condition in (4.21) and the partition of unity property of the dual shape functions, it
follows that (4.22).

Z Z
(4.22) Φj de = Nj1 de , j = 1, ..., me1
e e

It is important to point out that the element-wise biorthogonality condition in (4.21) must be satisfied in the physical
space, and not simply in the finite element parameter space. Consequently, a matrix system of size me1 × me1 must be
solved on each slave element. The first step for doing this is to introduce unknown linear coefficients ajk such that
(4.23).

1 1
(4.23) Φj (ξ , η ) = ajk Nk1 (ξ , η ) , Ae = [ajk ] ∈ Rme ×me

Page 104 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

It can easily be verified that, as a second step, insertion of (4.23) into (4.21) yields the unknown coefficient matrix
Ae as (4.24), where J(ξ , η ) is the slave Jacobian determinant.

1
Ae = De M−
e
Z
1 1
De = [djk ] ∈ Rme ×me , djk = δjk Nk1 (ξ , η )J(ξ , η )de
(4.24) e
Z
me1 ×me1
Me = [mjk ] ∈ R , mjk = Nj1 (ξ , η )Nk1 (ξ , η )J(ξ , η )de
e

4.3.3.4.1.2 Graphical representation :

The dual shape functions presented on the Equation (4.25) for the linear line is presented on the Figure 4.15. In
the case of the linear triangles the dual shape functions from (4.26) is represented on Figure 4.16. Finally, the bilinear
quadrilateral shape functions from (4.27) is shown in Figure 4.17. These graphic representations can be compared
with the standard shape functions, these are presented at the end of this chapter, which defines the shape functions
derivatives, see 4.6.Derivatives for contact mechanics linearization.

   1

Φ1 (1 − ξ )
(4.25) = 2
1
Φ2 2
(1 + ξ )

   
Φ1 1−ξ−η
(4.26)  Φ2  =  ξ 
Φ3 η
Figure 4.15: Dual shape functions for the 2D linear line

   1

Φ1 4
((1 − ξ )(1 − η ))
1
 Φ2   ((1 + ξ )(1 − η )) 
(4.27)  Φ3  = 
   4
1

4
((1 + ξ )(1 + η )) 
1
Φ4 4
((1 − ξ )(1 + η ))

4.3.3.4.1.3 Derivatives :
To define ∆φ it is necessary to define the derivatives from (4.24), which can be obtained with (4.99), which can be
found in the corresponding section of the derivatives section of this chapter, see 4.6.1.3.Dual shape functions.

4.3.3.4.2 Mortar operators :

4.3.3.4.2.1 Definition :
Considering the discrete Lagrange multiplier(4.19) in (4.6b) we obtain (4.28), where χ is the interface mapping.
This mapping, or projection, between domains is shown in the derivatives section, 4.6.Derivatives for contact
mechanics linearization, in the 2D case in the Figure 4.95 and in the 3D case in the Figure 4.100.

n
m X 1 1 Z ! n
m X 1 2 Z !
X X

(4.28) −δLco,h = λTnj Φj Nk1 dΓico δ d1nk − λTnj Φj Nl2 ◦ χh dΓico δ d2nl
j=1 k=1 Γ1c,h j=1 l=1 Γ1c,h

Vicente Mataix Ferrándiz Page 105 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

(a) Φ1 (b) Φ2

(c) Φ3
Figure 4.16: Dual shape functions for the 3D linear triangle

Numerical integration of the mortar coupling terms is exclusively performed on the slave side Γc,h of the interface.
In (4.28), nodal blocks of the two mortar integral matrices commonly denoted as De and Me can be identified. This
leads to the following definitions (4.29).

Z ngp
X
D[j, k ] = Djk Indim = Φj Nk1 dΓico Indim 1
, j = 1, ...m , k = 1, ...n =1 1 1
wg φgj Ngk Jg
Γ1c,h g=1
(4.29) Z ngp

X
M[j, l] = Mjl Indim = Φj Nl2 ◦ χh dΓico Indim 1
, j = 1, ...m , k = 1, ...n =2 2 1
wg φgj Ngk Jg
Γ1c,h g=1

With these matrices we can express the functional (4.28) in the following way (4.30) for the standard Lagrange
multiplier. This definition can be translated into the ALM formulation, but here we present the reasoning for the
standard Lagrange multiplier only, so the we can deduce the algebraic calculation of the discrete nodal weighted gap.
In here, xn corresponds with the normal component of the coordinates of all the nodes, meanwhile xnS and xnM

Page 106 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

(a) Φ1 (b) Φ2

(c) Φ3 (d) Φ4
Figure 4.17: Dual shape functions for the 3D bilinear quadrilateral

correspond with the slave and master domain nodes respectively.

   
  0 0
(4.30) −δLco,h = δ xTnS DT λn − δ xTnM MT λn = δ xnN xnM xnS  −MT  = δ xn  −MT  λn = δ xTn fco (λn )
DT DT
| {z }
BTco

Herein, the discrete mortar unilateral contact operator Bco and the resulting discrete vector of unilateral contact
forces fco (λn ) = Bco λn acting on the slave and the master side of the interface are introduced.
To finalise the discretisation of the considered unilateral contact problem, a closer look needs to be taken at the
weak constraint contribution δLλh in (4.6c). Due to the saddle point characteristics and resulting symmetry of the
mixed variational formulation in (4.6b) and (4.6c), all discrete components of δLλ,h have already been introduced and
the final formulation is given as (4.31), with gn (x) = Bmt x · n representing the discrete unilateral contact constraint at
the coupling interface.

(4.31) −δLλ,h = δλTn DxnS − δλTn MxnM = δλTn Bmt x · n = δλTn gn (x)

Where, this discrete form of gn can be renamed as nodal weighted gap for each node (ḡn ).

Vicente Mataix Ferrándiz Page 107 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

4.3.3.4.2.2 Derivatives :
To obtain a fully quadratic convergence in the computation of the contact problem we should compute the
derivatives of the Mortar operators, in consequence the derivatives of the Mortar operators are defined in its
corresponding section at the end of this chapter, see 4.6.Derivatives for contact mechanics linearization.

4.3.3.4.3 Algebraic form of the problem :

We will consider two cases for the matrix representation of the problem. The first one, corresponding to the scalar
LM, presented in the corresponding section 4.3.3.2.1.Scalar Lagrange multiplier. This one cannot be condensed, and
therefore requires the direct resolution of the problem with all the DOF.
The second case corresponds with the vector representation of the LM, from the formulation presented in
4.3.3.2.2.Vector Lagrange multiplier. This resulting system can be condensed due to the properties of the D,
which is diagonal due to the consideration of the dual Lagrange multipliers, presented in 4.3.3.4.1.Dual Lagrange
multipliers.

4.3.3.4.3.1 Scalar LMM :

The following is the matrix form of the scalar form of the contact pressure, see 4.3.3.2.1.1.LMM. Once computed
the mortar operators, the resulting system for unilateral contact corresponds with (4.32a). In this representation7 , and
in the following ones, the subindex N represents all the DOF but the ones related with the contact problem, M will
represent the master DOF and finally S will identify the slave DOF. At the same time, the slave DOF can be divided
between the active (A) and inactive (I ) DOF. In here the residual for the LM, rλA and rλI , corresponds in an algebraic
way with (4.32b). We call it algebraic way to the expression of the residual in considering the mortar operators (M, D).
The corresponding LHS is defined by the derivatives of the mortar operators from 4.3.3.4.2.2.Derivatives.

    
KN N KN M KN SA KN S I 0 0 ∆ uN rN

 KN N KMM KMSA KMSI − (n · MA )T − (n · MI )T 
 ∆uM   rM 
(n · DAA )T (n · DAI )T
  
 KS A N KSA M K S A SA KSA SI 
∆uSA   rS A 
(4.32a)  = −
 
(n · DIA )T (n · DII )T
  
 KSI N KS I M KSI SA K SI S I  ∆ uS I 


 rSI 

 ∂ rλA ∂ rλ A ∂ rλ A 
 0 ∂ uM ∂ u SA ∂ uS I 0 0  ∆λA   rλ A 
0 0 0 0 0 I ∆λI rλ I
(
rλAn = −n · (Dx1 − Mx2 )
(4.32b)
rλI = λn

4.3.3.4.3.2 Components LMM :

Here, the algebraic system upcoming from components representation of the LMM in the frictionless contact
formulation is presented in (4.33a). The Lagrangian has been previously shown in 4.3.3.2.2.1.LMM. The main
difference is the full consideration of the mortar operators M and D.

    
KN N KN M KN SA KN S I 0 0 ∆ uN rN
 KN N KMM KMSA KMSI −MTA −MTI  ∆ uM   rM 
DTAA DTAI
    
 KS A N KSA M KSA SA KSA SI  ∆ uS A   rSA 
(4.33a) 
KSI N KS I M KSI SA KS I S I DTIA DTII
  = − 



 ∆ uS I 


 rS I 

∂ r λA ∂ rλA ∂ rλA ∂ rλ A
 0 ∂ uM ∂ u SA ∂ u SI ∂λA 0  ∆λA   rλA 
0 0 0 0 0 I ∆λI rλI

In the same way that in the previous definition for the scalar LM, but considering the penalisation of the tangent
components, we define the LM residuals rλA and rλI in the Equation (4.33b). In here, and in the following, we define
7 We will consider a very close notation with the one presented in Popp[PhDPop12].

Page 108 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

λ−n(n·λ)
τ as the tangent direction of the λ, defining τ = kλ−n(n·λ)k .

(
rλA = n · (−n · (Dx1 − Mx2 )) − τ · λ
(4.33b)
rλI = λ

This system, as stated before can be statically condensed, this will be presented later (4.3.3.4.4.Static condensation
of the system in considering of the DLMM) as it can be considered both for LMM and ALM formulations. This
methodology can be considered too in the frictional formulation, which always requires to decompose the LM in
components.

4.3.3.4.3.3 Penalty :

The formulation presented in 4.3.3.2.1.2.Penalty is the simplest of the introduced, as can be seen in (4.35a). It is
relevant to highlight that the inactive slave contact nodes do not add any contribution to the RHS and the LHS. We
want to remark that we do not assume KMS , KSA M and KSI M to be zero, as these terms have contributions from
others mechanical problems.

 
KN N KN M  
KN
S 
   
 ∂ n· M T ∂ n· M T  ∆uN
T
 KMN KMM − ε n · M + ∂ uM xM KMS − ε ∂ uS xM 

 A
 ∆ uM 

T
  
T
    
 ∂ n·DA
T
∂ n· D A  ∆ uS A 
 K
 SA N KSA M + ε ∂ uM xSA KSA SA + ε n · DA + ∂ uS xSA  
∆uSI
A
(4.34)
K SI N KSI M KSI SI
 
rN
 rM − εn · MxM 
= − 
 rSA + εn · DA xSA 
rSI

4.3.3.4.3.4 Scalar ALM :

Here the algebraic version of 4.3.3.2.1.3.ALM si presented in (4.35a), with the corresponding residual for the LM
defined in (4.35b).

 
KN N KN M KN S A KN SI 0 0 ∆uN
 
rN

T

 KN N KMM KMSA KMSI − (k n · MA ) − (k n · MI )T 
 ∆ uM   rM 
(k n · DAA )T (k n · DAI )T
  
 KSA N KS A M KSA SA KS A S I 
∆ uS A   rS A 
(4.35a)  = −
 
(k n · DIA )T (k n · DII )T


 KSI N KSI M KSI SA KSI SI 
 ∆ uS I   rS I 
 ∂ rλA ∂ rλA ∂ rλA    
 0 ∂ uM ∂ uS A ∂ u SI 0 0  ∆λA   rλA 
0 0 0 0 0 k2
I ∆λI rλ I
ε
(
rλAn = −kn · (Dx1 − Mx2 )
(4.35b) k2
rλ I = ε λn

4.3.3.4.3.5 Components ALM :

The final contribution that will be shown for the frictionless contact is the one relative to the ALM implementation for

Vicente Mataix Ferrándiz Page 109 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

the vectorial LM from 4.3.3.2.2.2.ALM. The LHS can be seen in Equation (4.36a), and the LM RHS in (4.36b).

    
KN N KN M KN S A KN SI 0 0 ∆ uN rN

 KN N KMM KMSA KMSI −k MTA −kMTI 
 ∆ uM   rM 
kDTAA kDTAI
  
 KSA N KSA M KSA SA KSA SI 
∆ uS A   rSA 
(4.36a)  = −
 
k DTIA k DTII


 KSI N KSI M KS I S A KSI SI 
 ∆ uS I   rSI 
 ∂ rλA ∂ rλA ∂ rλA ∂ rλ A    
 0 ∂ uM ∂ uS A ∂ uS I ∂λA 0  ∆λA   rλA 
k2 ∆λI rλI
0 0 0 0 0 ε I
( 2
k
rλA = k n · (−n · (Dx1 − Mx2 )) − τ ·λ
(4.36b) k2
ε
rλ I = ε λ

4.3.3.4.4 Static condensation of the system in considering of the DLMM :

As mentioned before, it is possible to statically condensate the system of equations, solving a system in pure
displacements, instead of a system with displacements and LM. We already stated that this is possible to the diagonal
property of the D when considering a DLMM formulation. The following presents the construction considered for
this system of equations that can be applied in any LHS where the LM considered is decomposed in Cartesian
components, both frictionless and frictional formulation. We describe the global LHS as shown in Equation (4.37),
where the matrix is decomposed in 36 × 36.

  
KN N KN M KN SA KN S I 0 0 ∆uN

 KN N KMM KMSA KMSI KMLMA KMLMI 
 ∆ uM 


 KSA N KSA M KS A S A KSA SI KSA LMA KSA LMI 
 ∆ uS A 


 KS I N KSI M KSI SA K SI S I KSI LMA KSI LMI 
 ∆uSI 

 0 KLMA M KLMA SA KLMA SI KLMA LMA 0  ∆λA 
0 0 0 0 0 KLMI LMI ∆λI
(4.37)  
rN

 rM 

 rSA 
= − 

 rS I 

 rλA 
rλI

The last system presented can be simplified if we consider that KSI LMA = 0, KSA LMI = 0 and ∆λI = 0.
Additionally the KLMI LMI is diagonal, so its resolution is trivial, or can be applied by simply imposing λI = 0. With
this we statically condensate (4.37) into (4.38). The resulting system is in pure displacement, instead of mixed.

  
KN N KN M KN SA KN S I ∆ uN
 KN N + PKSA N KMM + PKSA M KMSA + PKSA SA KMSI + PKSA SI   ∆uM 
  
 KS I N KS I M KSI SA KS I S I   ∆uSA 
CKSA N KLMA M + CKSA M KLMA SA + CKSA SA KLMA SI + CKSA SI ∆ uSI
(4.38)  
rN
 rM + PrSA 
= − 
 rLMA + CrSA 
rS I

In this last equation, operators P and C can be computed as follows in (4.39).

1

T
(4.39a) P = K−
SA LMA KMA LMA
1

T
(4.39b) C = K−
SA LMA KLMA LMA

Page 110 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

Finally, we can extract and deduce the expression necessary in order to compute the active LM in a standalone
manner as (4.40). This will be evaluated after computing the displacement DOF, from which it depends. It is important
to note that KSA LMA is a diagonal matrix, as is the result of the global assemble of the mortar operators D. Equation
(4.40a) can be reformulated as Equation (4.40b).

(4.40a) KSA N · rN + KSA M · ∆uM + KSA SA · ∆uSA + KSA SI · ∆uSI + KSA LMA · ∆λA = rSA
1


(4.40b) ∆λA = K−
SA LMA rSA − KSA N · rN − KSA M · ∆uM − KSA SI · ∆uSI − KSA SA · ∆uSA

4.3.3.5 Work-flow. Solution algorithm

The following section introduces the algorithm to be considered in order to solve the frictionless contact problem.
It is relevant to highlight the fact that this algorithm is related with the semi-smooth strategy of the next section,
4.3.3.6.Active set strategy (Semi-smooth Newton Raphson), and the previous one, 4.3.3.4.3.Algebraic form of the
problem. The algorithm is presented in Algorithm 2, here we will explain the differences between each one of the
approaches presented (standard LM, penalty, etc.), the approaches are similar but with slight differences that are
relevant to highlight.
The most significant difference across formulations is the threshold to be considered in the active set computation.
For LM the threshold for each node to be checked is λi+1 i+1
nn+1 and for the ALM is k λnn+1 − εn g̃n . λnn+1 may correspond
with the LM for scalar LM or for the normal components (λn = n · λ) in case of vector LM. In case of penalty formulation
the threshold would be −εn g̃n . Additionally, it is important to separate the residual corresponding to the displacement
solution and the LM solution, in case that the problem considers LM.

Algorithm 2 Algorithm for the frictionless contact problem

1: procedure A LGORITHM FOR THE FRICTIONLESS CONTACT PROBLEM
2: t = 0 and i = 0
3: Initialise the solution for u0 = 0
4: In case of solving LM solution, Initialise the LM solution λ0 = 0 or λ0n = 0
5: Initialise the active set A01 and I10 such that A01 ∪ I10 = S and A01 ∩ I10 = ∅
6: while t < tend do
7: t = t + ∆t and i = i + 1
8: Initialise the increment of solution for ∆ui1 = 0
9: In case of solving LM solution, Initialise the LM increment of solution ∆λi1 = 0 or ∆λin1 = 0
10: Search for potential contact pairs, and if required update the pairs and the active set, respecting step 5
11: We define the problem as not converged conv = false
12: while conv = false do
13: Find the solution corresponding to the system presented in 4.3.3.4.3.Algebraic form of the problem
14: Update the solution, so uin+1 = uin + ∆uin+1
15: In case of LM solution λin+1 = λin + ∆λin+1 or λinn+1 = λinn + ∆λinn+1
16: Update the active set as in (4.41). The threshold for each one of the cases is represented in (4.42).
i+1
 i+1
In+1 := j ∈ S|thresholdn+1 ≥0
(4.41) i+1 i+1
An+1 := j ∈ S|thresholdn+1 <0

(4.42) thresholdLM = λn or n · λ, thresholdPenalty = ǫḡn , thresholdALM = k λn + ǫḡn or kn · λ + ǫḡn

17: Compute the corresponding residual to check (4.43).

(4.43) kru k < toleranceu , krλ k < toleranceλ

18: The solution is converged if Ai+1 i i+1 i

n+1 = An+1 , In+1 = In+1 and the residuals from (4.43) are converged.

The former algorithm, particularly the check of convergence can be appreciated in Figure 4.18, where the current
contact convergence check present in Kratos is shown. In here the different types of residuals are check, as well as

Vicente Mataix Ferrándiz Page 111 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

the active set check.

Figure 4.18: Example of convergence check in the frictionless contact

4.3.3.6 Active set strategy (Semi-smooth Newton Raphson)

As previously stated, the fully discretised unilateral contact problem introduces one significant complexity in the
problem resolution. This issue is the contact specific inequality constraints, which introduces two different sets of
discrete active and inactive constraints, which are unknown a priori. This previous dilemma does not appear for the
mesh tying case, see A.3.Mesh tying, where the subsets are known a priori. From a mathematical point of view, this
introduces an additional source of non-linearity apart from the already existing geometric and material non-linearity.
This issue can be solved with the consideration of an appropriate strategy, a common approach is the Primal-Dual
Active Set Strategies (PDASS), which is already presented in detail in the optimisation appendix (D.Constrained
optimisation problems).

The principle behind any active set strategy for uni-

lateral contact is to iterate looking for the correct subset
of master-slave nodes in contact until there is not vari-
ation in the respective subsets in the given time step. In
the other hand, the contact non-linearity cannot be sim-
ply solved in application of the PDASS. This is because
finding the correct active set A cannot be resolved by a
standard NR type approach.

Can be affirmed that in each one of the non-

converged subsets we can apply an standard NR type
algorithm, in the same manner it is applied to the other
types of non-linearity. These non-converged subsets
are obtained from rearranging the KKT conditions. In
Popp[PhDPop12] the discrete KKT conditions are refor-
mulated within a so-called NCP, this NCP is equivalent
to the Lagrangian (L) contribution for the LM in the ALM, Figure 4.19: Nodal NCP function, or Lagrangian (L) contri-
therefore in case of consider an ALM formulation it is de bution for the LM in the ALM
facto considered in the Lagrangian (L) definition. Equa-
tion (4.44) shows the expression which defines the frictionless NCP. This corresponds with the augmented normal
contact pressure λ̄n presented before (4.13), and the criteria will consist on activate/inactivate the corresponding node
if the augmented contact pressure λ̄n is in compression or traction correspondingly. The graphical representation can
be seen in Figure 4.19, where the equivalence with the KKT conditions is indicated in red colour, and this is drifted
accordingly to the penalty ε contribution.

(4.44) Cλn = k λn − max(0, k λn + εḡn )

Hence, the PDASS considered accommodates derivative information on the subsets, allowing the resolution in
considering NR algorithm also for the contact non-linearity. With the previous consideration, the contact problem can
be solved together with the other non-linearity, therefore we can consider one single iterative scheme. For the penalty
formulation, just in considering the algorithm presented in 4.3.3.5.Work-flow. Solution algorithm the PDASS is taken
into account with a standard NR strategy.

Page 112 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

4.3.4 Frictional contact

In the same manner to the frictionless case before introduce the weak formulation we need to formulate the problem,
we require the definition of the strong formulation. The ALM formulation presented is a combination of the following
references [ArtCC15; PhDGit12; PhDDoc; ArtDPS; ArtYLM].

4.3.4.1 Strong formulation

The part relative to the solutions spaces is exactly the same than the one presented for the frictionless case (4.4) as
well as the balance of the linear momentum (4.2). Particularly the developments and definitions presented for the
frictionless contact for the LM by components, see 4.3.3.2.2.Vector Lagrange multiplier, will be applied here. The
KKT (4.3.3.1) condition is still into play for the frictional problem.

4.3.4.1.1 Tangential contact condition - Coulomb’s law :

Friction is a complex physical phenomenon. The science that studies the friction, the tribology (see Tribology),
has concluded the many origins of the physics of the frictional phenomenons. This combines the interactions of elastic
and plastic deformations at the contact interface, interaction with wear particles, microfractures, excitation of electrons,
etcetera. There exist different friction models as we have already introduced in the state-of-the-art of this chapter, see
4.2.5.Frictional models.
In continuum mechanics, the most common description is the phenomenological, or macroscopic, law of Coulomb
(Figure 4.21), which with Tresca8 are the most extended and yet simple models of friction. In our models we can
identity all these models with F . In general we will take into consideration the Coulomb’s law, as it is the most
extended and commonly used.

Figure 4.20: Coulomb’s schematic depiction of frictional contact conditions in tangential direction

The Coulomb’s law can be defined as (4.45).

(4.45a) φco := ktτco k − µkpn k ≤ 0

(4.45b) vτ ,rel (X1 , t) + β tτco = 0
(4.45c) β≥0
(4.45d) φco β = 0

Being µ the friction coefficient, and β the velocity-traction ratio. Equation (4.45a) requires the magnitude of the
tangential stress vector to not exceed the product of the coefficient of friction and the normal contact pressure. When
the tangential stress is less than the Coulomb limit (φco < 0), the continuity equation (4.45d) forces the β to be zero
8 Which is simpler because does not depend on contact normal pressure, it is just a threshold value.

Vicente Mataix Ferrándiz Page 113 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

and, accordingly, the tangential relative velocity to be zero, this is called stick state. When the tangential stress is at
the Coulomb limit (φco = 0)), β may be greater than zero in (4.45d) and therefore the tangential stress is forced to
oppose the relative tangential velocity in (4.45b), this is called the slip state.
The Figure 4.21 presents the graphical representation of Coulombs frictional conditions for 3D contact problems.
The sub-Figure 4.21a introduces relation between the norm of the tangential velocity and the components of the
tangential stress vector. In here the admissible points are situated either in the blue circle (β = 0, k tτco k≤ µ k pn k |),
which corresponds with the stick state; or on the surface of the semi-infinite cylinder (β ≥ 0, k tτco k≤ µ k pn k |)
marked with red colour, associated with the slip state. On the other hand, the sub-Figure 4.21b shows relation
between the contact pressure and the components of the contact tangential stress vector. We can see in blue the
interior of the Coulombs cone ( k tτco k≤ µ k pn k |), which represents the stick state. Contrarily the surface of the
cone in red ( k tτco k= µ k pn k |) exemplify the slip state.

(a) Relation between the norm of the tangential velocity (b) Relation between the contact pressure and the compo-
and the components of the tangential stress vector nents of the contact tangential stress vector
Figure 4.21: Graphical representation of Coulombs frictional conditions for 3D contact problem. Inspired [PhDYas11]
Additionally, the Table 4.3 shows the analogy mentioned in 4.2.5.Frictional models between the friction and the
plasticity phenomenons. The relationship here corresponds with the one presented by Yastrebov [PhDYas11], a more
detailed analysis can be found in Antoni[ArtAnt17].

Friction Plasticity
Stick state Elastic deformation
Slip state Plastic flow
Coulombs’s cone ∂ C(pn ) Yield surface
Maximal frictional stress k tτco k= µ|pn | Yield strength
Table 4.3: Analogies between friction and plasticity

Finally, the definition of the tangent direction is needed. We use the complementary direction to the normal,
so calling the normal as n we can define the tangent as (4.46a). With this we can define for example, the tangent
Lagrange multiplier as (4.46b).

(4.46a) τ =I−n⊗n
(4.46b) λτ = λ − nλn

4.3.4.2 Weak formulation

4.3.4.2.1 LMM :

Page 114 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

The weak formulation of these equations can be obtained again by multiplying (4.2) with the test function wi ∈ V i
and integrating in each domain i as (4.47).

Z Z Z
     
(4.47) Li (ui ) = ∇ · σ i + bi · wi dΩi + ti − σ i · ni · wi dΓiσ + tico − σ i · ni · wi dΓico = 0
Ωi Γiσ Γico

From (4.47) the part corresponding to the frictional contact would the shown in (4.48).

Z
 
(4.48) Lco (u, λ) = tico − σ i · ni dΓico
Γ1c

If we consider the test function wi as virtual displacements δ u, and with the application of the Gauss divergence
theorem it is possible to determine the equivalent expression for the virtual work (4.49).

Z Z Z Z
∂δ u̇i i
(4.49) δLi (ui , δ ui ) = σi : dΩ − bi · δ ui dΩi − ti · δ ui dΓiσ − tico · δ ui dΓico = 0∀δ ui ∈ V i
Ωi ∂ xj Ωi Σiσ Γico

Equation (4.49) can be regrouped on contributions as in (4.7), obtaining (4.52).

2 Z
X 
(4.50a) − δLkin (u) = ρi üi · δ ui dΩi
i=1 Ωi
2
"Z   Z #
X ∂δ u̇i
i
(4.50b) − δLint,ext (u) = σ : − b · δ ui i
dΩ − i
t · δ u ddΓσ i i
Ωi ∂ xj Γiσ
i=1
Z
(4.50c) − δLco (u, λ) = t1co · δ t1c dΓ1co
Γ1co

Where tco represents the traction on the interface, the balance between the contact interface implies (4.51)

(4.51) t1co dγ 1 = −t2co dγ 2

As in the frictionless case, we will take the slave surface as reference, this allows to rewrite the virtual work integral
for the contact as (4.52a).

Z
(4.52a) δLco (u, λ) = − t1co · δ gdΓ1co
Γ1co

Where g represents the general gap (4.52b) (instead of just normal gap).


(4.52b) g = u 1 − χ · u2

The Lagrange multipliers are introduced as additional unknowns on the slave contact surface and are identified as the
negative contact traction as (4.52c).

(4.52c) λ = −t1co

Vicente Mataix Ferrándiz Page 115 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

4.3.4.2.2 Contact constraints :

The Equation (4.52a) is reformulated. Using the Lagrange multiplier in (4.52c) we can decompose into normal
and tangential components λn and λτ (4.53).

(4.53a) λn ∈ R+0 : g(δλn − λn ) ≥ 0∀δλn ∈ R+0

(4.53b) λτ ∈ B (µλn ) : vτ ,rel · (δλτ − λτ ) ≤ 0∀δλτ ∈ B (µλn )

Where B (µλn ) is a (n1)-dimensional sphere with centre 0 and radius µ · λn and λτ is a trial force in the tangential
plane. It is the so-called principle of maximal dissipation representing Coulomb’s law of friction. Then (4.52a) can be
rewritten and the weak form of both normal and tangential contact conditions is obtained as (4.54).

Z
(4.54a) λ ∈ M (λ ) : g(δλn − λn )d λ ≥ 0,
λ1co
Z
(4.54b) vτ ,rel (δλτ − λτ )d λ ≤ 0∀δλ ∈ M(λ)
λ1co

Where M(λ) is the admissible solution space for the Lagrange multiplier and the test space for the trial forces δλ.
It is a convex subset M(λ) ⊂ M which accommodates the restrictions of λ and δλ in (4.53). The subspace can be
rewritten as (4.55).


(4.55) M(λ) := δλ ∈ M : hδλ, ηi ≤ hµλn , kη τ ki, η ∈ L1 with ηn ≤ 0

Here h·, ·i is the duality pairing of the spaces M and V 1 on γco

1
given by (4.56).

Z
(4.56) hδλ, ηi := δλη d γ
1
γco

4.3.4.2.3 Penalty :
Before define the ALM frictional formulation, we introduce the penalty formulation for the frictional problem. The
ALM formulation will result from the combination of this with the previous LMM. The Lagrangian (L) for the problem is
defined on (4.57a), separating the normal (4.57b) and tangent contributions (4.57c). In these expressions, tnco = εn gn
defines the normal contact traction and tτco = ετ vτ ,rel corresponds to the tangent contact traction.

Z
(4.57a) Lco (u) = ln + lτ dΓico
Γ1c
(
− ε2n gn2 , tnco ≤ 0, (Contact zone)
(4.57b) ln (gn ) =
0 , tnco > 0, (Gap zone)
( ε
n
 − 2 vτ ,rel · vτ ,rel , ktco k ≤ −µtco , stick
τ τ

µ n 2 vτ ,rel , tnco ≤ 0, (Contact zone)
(4.57c) lτ (vτ ,rel ) = − ετ tco kvτ ,rel k , ktτco k > −µtnco , slip

, tnco > 0, (Gap zone)

0

With the Lagrangian (L) we can define the derivative in (4.58).


n
Z  tco · δ gn + tco · δ vτ ,rel
τ
if ktτco k ≤ −µtnco (Contact stick zone)
tnco · δ gn − µtnco kvττ ,rel if ktτco k > −µtnco (Contact slip zone) dΓico
v ,rel
(4.58) δLco (u) = k δ vτ ,rel
1 
Γc 
0 if tnco > 0 (Gap zone)

Page 116 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

4.3.4.2.4 ALM :
In the very same way, we have defined an Augmented Lagrangian for the frictionless method we can define
the same for the frictional case. The ALM method to solve contact problems with friction was proposed by Alart
and Curnier [ArtAC91], additionally to these references we have considered the work of Cardona[ArtCC15] and
Yastrebov [PhDYas11]. The solution obtained for the ALM results in a combination of the former LMM (4.3.4.2.1.LMM)
and penalty (4.3.4.2.3.Penalty) solution.
Focusing in the functional relative to the contact (Lco (u, λ) = LV co + LM ), we can rewrite (4.48) as (4.59).

Z
(4.59) Lco (u, λ) = ln + lτ dΓico
Γ1c

Being ln and lτ (4.60) the corresponding parts of the Augmented Lagrangian formulation for the normal and
tangent contributions respectively.

(
λ̄n gn − ε2n gn2 , λ̄n ≤ 0, (Contact zone)
(4.60a) ln (gn , λn ) = 2
− 2kεn λ2n , λ̄n > 0, (Gap zone)

With λ̄n = k λn + εn gn being the augmented Lagrange multiplier for the normal direction.
lτ (vτ ,rel , λτ ) =
(
ετ
 λ̄τ · vτ ,rel − 2 vτ ,rel · vτ ,rel
 , kλ̄τ k ≤ −µλ̄n , stick
(4.60b) 1 2 2 2

, λ̄n ≤ 0, (Contact zone)
− 2ε k λτ · λτ + 2µλ̄n kλτ k + µ λ̄n , kλ̄τ k > −µλ̄n , slip
 k2 τ

− 2ετ λτ · λτ , λ̄n > 0, (Gap zone)

With λ̄τ = k λτ + ετ vτ ,rel being the augmented Lagrange multiplier for the tangent direction.
Where εn and ετ are a positive penalty parameter, both for normal and tangent direction, k is a positive scale
factor. Using the hi is the Macauley bracket operator, from (4.10), we can express everything as (4.61).

1

(4.61a) ln (gn , λn ) = k 2 λ2n − hλ̄n i2
εn
1

(4.61b) lτ (vτ ,rel , λτ ) = k 2 λτ · λτ − kλ̄τ k2 − hkλ̄τ k − µk − λ̄n ki2
ετ

This functional is C 1 differentiable saddle-point, as shown in Figure 4.22. The solution is obtained as the set of
values that render this functional stationary. The locus which corresponds to the solution in the normal direction is
presented in the optimisation appendix, D.4.3.Applicability on contact problems, in the Figure D.4.
As in the frictionless case, the solution does not depend on the value of parameters ε and k. Nevertheless,
the convergence rate does depend on their value. Finally, we can derive (4.59) to obtain the variational form from
(4.62).

δLco (u, λ) =

λ̄ · δ gn + kgn δλn + λ̄τ · δ vτ ,rel + vτ ,rel · δ λ̄τ if kλ̄τ k ≤ −µλ̄n (Contact stick zone)
(4.62)
Z   n

λ̄τ
k λτ +µλ̄n kλ̄ i
λ̄ n · δ gn + kg n δλ n − µ λ̄ n
λ̄τ
kλ̄τ k
δ vτ ,rel − ετ
τk
δλτ if kλ̄τ k > −µλ̄n (Contact slip zone) dΓco
Γ1c 

 k 2 2
k
− εn λn δλn − ετ λτ δλτ if λ̄n > 0 (Gap zone)

The functional from (4.62) makes that the system obtained varies in function if the nodes are present in the contact
(slip or stick) or the gap zone, the system is not a priori known like in the frictionless case but with an additional
configuration.

Vicente Mataix Ferrándiz Page 117 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

Figure 4.22: Augmented Lagrangian function for the frictional contact problem. Corresponding to (4.61b)

4.3.4.3 Discretisation and numerical integration

Many of the things defined for the frictionless case, like the definition of the dual Lagrange multipliers or the definition
of the mortar operators remain exactly the same. In this case we will focus in the kinematic definitions for the frictional
case using the mortar operators, basically the definition of the discrete contact condition in the tangential direction
and the slip in the discrete using the mortar operators.

4.3.4.3.1 Discrete contact condition in tangential direction :

Following the same procedure for the normal direction we can obtain the part equivalent for the tangential direction.
The most relevant thing to take into account before any definition is the concept of the relative velocity in the tangential
direction vτ ,rel , where we will use for our definition the discrete form of the material velocity field ẋi , which uses the
same shape functions for interpolation as the xi . We can then define (4.63).

Z
vτ ,vel · (δλτ − λτ ) d γ
γc1
(4.63) nX
slaves
"Z nX
master Z #
T


≈ (δλτ − λτ ) τ j Φj Nj1 d γ ẋ1j − Φj Nl2 ·ξ d γ ẋ2l ≥ 0∀δλ ∈ M(λ)
j=1 γc1 l=1 γc1

We can express this equation using the mortar operators, see 4.3.3.4.2, what will give us the following expression
(4.64). Where ṽτ j is the weighted relative velocity.

Z nX
" nX
# nX
slaves master slaves
T
(4.64) vτ ,vel · (δλτ − λτ ) d γ ≈ (δλτ − λτ ) τ j Dj ẋ1j − Ml ẋ2l = (δλτ − λτ )T ṽτ j ≥ 0
γc1 j=1 l=1 j=1

4.3.4.3.2 Slip definition :

An important aspect of a proper formulation of frictional laws in the finite sliding context is framed indiffer-
ence[PhDGit12; ArtYLM] of the rate measures involved. This affects the tangential relative velocity of the contacting
bodies (vτ ,rel ) in the considered case of frictional contact. This assures that this quantity is unaffected by any rigid

Page 118 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

body motion which the two contacting bodies might experience at the instant of the question. Mathematically, this can
be tested with formulating the tangential relative velocity vτ ,rel in an alternative reference frame. Then in the current
(mortar projected) instance, we must ensure frame indifference.

Working in the time continuous case first, one may readily show that the tangential component of the mortar
projected tangential velocity is not frame indifferent (4.65a). Frame indifference is assessed by viewing the motion
from another reference frame, denotes in the following by superscripts c(t), which can be related to the original spatial
frame via (4.65b). Where c(t) is the relative rigid body translation between the original spatial frame and observer c(t),
while a relative rotation is produced by the proper orthogonal tensor , (4.65c). The frame indifferent relative tangential
velocity should satisfy. However, by considering the effect of the transformation (4.65b) on (4.65a), it is readily seen
that (4.65d).

" nX
#
master
nononbj
(4.65a) ṽτ = τj Dj ẋ1j − Ml ẋ2l
l=1
(1∗)
(4.65b) ẋl = c(t) + Q(t)ẋ1l
(4.65c) ṽ∗τ = Q(t)ṽτ
" nX
#
master
nonobj ∗ nonobj
(4.65d) ṽτ = Q(t)ṽτ − Q̇(t) Dj x1j − Ml x2l · τj
l=1

h Pnmaster i
Because the term Dj x1j − l=1 Ml x2l 6= 0 in general ṽnonobj
τ
∗
does not satisfy the equation (4.65c), and thus
some modifications are required to this relative velocity measure to assure material frame indifference. It is possible
to restore the objectivity with the inclusion of the rate of a mortar projected distance between the two bodies, denoted
as g. Then in consequence (4.66) is obtained.

" nX
#
master

(4.66) ṽτ = τ j Dj ẋ1j − Ml ẋ2l − ġ

l=1

We obtain an expression which retains the interpretation of the tangential relative velocity in the case where
perfect sliding occurs (i.e. when ġ = 0), but which contains the modification necessary to make the velocity measure
objectives under all conditions of contact. This is readily seen by using direct calculation to exactly reexpress (4.66)
as (4.67b), considering (4.67a).

" nX
# " nX
# " nX
#
master master master
d
(4.67a) ġ = Ḋj x1j − Ṁl x2l = Ḋj ẋ1j − Ṁl ẋ2l + Ḋj x1j − Ṁl x2l
dt
l=1 l=1 l=1
" nX
#
master

(4.67b) ṽτ = τ j Ḋj x1j − Ṁl x2l

l=1

The time derivatives of the mortar operators can be defined using any desired scheme, for example using the
backward Euler (4.68) scheme as time discretisation.

d (· ) (·)t+∆t − (·)t
(4.68a) ≈
dt ∆t
t+∆t
dD Dl − Dtj dM Mt+∆t − Mtl
(4.68b) ≈ , ≈ l
dt ∆t dt ∆t

Vicente Mataix Ferrándiz Page 119 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

With this we can define tangential relative velocity ṽτ as (4.69a), that multiplies by ∆t gives us the nodal slip
increment ũτ (4.69b).

" nX
#
Dt+
j
∆t
− Dtj master
Mt+ ∆t
− Mtl 2
(4.69a) ṽτ = τ j x1j − l
ẋl
∆t ∆t
l=1
" nX
#
master



(4.69b) ũτ = τ j Dt+
j
∆t
− Dtj x1j − Mt+
l
∆t
− Mtl ẋ2l
l=1

4.3.4.3.3 Algebraic form of the problem :

The resulting systems of equations presented in this section will consider always a vectorial LM in the cases for
the LMM and ALM, as opposed to the solutions shown in 4.3.3.4.3.Algebraic form of the problem. Because of this
reason, we will be able to consider always the static condensation presented in 4.3.3.4.4.Static condensation of the
system in considering of the DLMM for the LM approaches. It is also relevant to spotlight that in the following sections
the symbols representing the slip and stick states in addition to the states already presented in 4.3.3.4.3.1.Scalar
LMM. These states are represented respectively as sl and st subindexes.

4.3.4.3.3.1 LMM :
Equation (4.70a) presents the system of equations relative to the standard LMM for the frictional problem. We can
see that, we have an additional block to block presented for the frictionless solution in 4.3.3.4.3.2.Components LMM.
This is due to the fact that now the active DOF relative to the LM can be split into two different groups. The group
relative to the slip state (sl) and the stick state (st). The residual of the contact solution is therefore also split into
these two subgroups, as see in Equation (4.70b).

 
KN N KN M KN SA KN SI 0 0 0    
 KN N KMM KMSA KMSI −MTAsl −MTAst −MTI  ∆ uN rN
  rM
 KSA N KSA M KS A S A KSA SI DTAAsl DTAAst DTAI  ∆uM 





  rS A
 KS I N KSI M KSI SA KSI SI DTIAsl DTIAst DTII  ∆uSA 





(4.70a)  = −
  rS I
 ∂ rλ A
sl
∂ rλA
sl
∂ rλA
sl
∂ rλA
sl
∂ rλA
sl
 ∆ uSI  



 0 ∂ uM ∂ uS A ∂ uS I ∂λAsl ∂λAst 0 
 ∆λAsl   rλAsl 
sl   
 ∂ rλA ∂ rλA ∂ rλA ∂ rλ A ∂ rλA 
 0 st st st st sl
0  ∆λAst   rλAst 
 ∂ uM ∂ uS A ∂ uS I ∂λAsl ∂λAsl 
sl ∆λI rλ I
0 0 0 0 0 0 I

Equation (4.70b) shows the different residuals associated to the LM, and presented in Equation (4.70a). The
respective derived terms in the LHS, as in the frictionless case, require the mortar operators derivatives defined in
4.3.3.4.2.2.Derivatives.


rλAsl = n · (−n · (Dx1 − Mx2 )) − (τ · λ − F )

(4.70b) rλAst = n · (−n · (Dx1 − Mx2 )) + ũτ

rλ I = λ


(4.70c) F = −µλn τ

The F is the generic representation of the frictional threshold, we have already mentioned this in the section
dedicated to the Coulomb’s frictional law (4.3.4.1.1.Tangential contact condition - Coulomb’s law). We will consider this
frictional law, as our theoretical developments have departed from this hypothesis. In any case, the formulation can
be adapted to consider a different frictional criterion changing F from (4.70b), as well as in the corresponding active
set computation (see 4.3.4.4.Work-flow. Solution algorithm). The corresponding expression for F for the Coulomb’s
frictional law is presented in (4.70c), and it will be considered in the following for the standard LM. Additionally, ũτ ,
corresponds with the expression from Equation (4.69b).

Page 120 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.3 Formulation

4.3.4.3.3.2 Penalty :
For the frictional solution, as we must split solution between the slip and the stick state and we lack of LM in our
system, we separate the contact DOF between slip and stick in order to separate the contact contributions added
to the displacement DOF. Because of this, the system in Equation (4.71a) does not look as an square LHS, 6 × 3
instead of 6 × 6, but it is actually a square system of equations.

(4.71a)
 
KN N KN M 
K
 N S
 slT

 ∂ n· M ∂ n·MslT 
 Ksl
MN Ksl
MM − εn n · MslT + ∂ uM xsl
M + ∂FM
∂ uM Ksl
MS − εn ∂ uSA xsl
M+
∂FM
∂ uS A
 

      

 ∆ uN
 ∂ n·DslT ∂ n·DslT  ∆uslM 
Ksl Ksl sl
Ksl n · DslT xsl
A ∂FS A ∂FS
SA M + ε n ∂ uM x S A + SA S A + ε n A + +
  


SA N
 
∂ uM
 
∂ uS A SA ∂ uS A 
 ∆uslSA 

 

st st stT
∂ n·MstT
st st
∂ n·MstT  ∆ust 

 KMN KMM − εn n · M + ∂ uM xM − εt ∂∂uũM
τ
KMS − εn ∂ u SA xst ∂ ũτ
M − ε t ∂ uS


M
∆ust


SA
      A 
 ∂ n·DstT ∂ n·DstT  ∆ uS I
Kst Kst xsl Kst n · DstT xsl
A ∂ ũτ A
SA M + ε n SA + εt ∂ uM SA S A + ε n A + + εt ∂∂uũSτ
 
 SA N ∂ uM ∂ u SA SA A

KSI N KSI M KSI SI
 
rN

 rsl
M − ε n n · Msl xsl
M + FM


 rSA + εn n · Dsl
sl sl
A S + FS
x 
= − st st st


 rM − εn n · M xM − εt ũτ 

 rst st st
SA + εn n · DA xS + εt ũτ

rSI

The frictional threshold (F ) for the penalty formulation, considering the Coulomb’s frictional law, corresponds with
F = −µεn ḡn τ . It consists in take the penalty approach for the normal contact pressure (εn ḡn ), and multiply by the
friction coefficient (µ) in the tangent direction (τ ). In this case, as we lack the LM solution in order to determine the
tangent direction τ we must consider the direction of the slip increment (ũτ ), defined as τ = kũũττ k .

4.3.4.3.3.3 ALM :
As we have already pointed in 4.3.4.2.4.ALM we have to consider two different penalties, one in the normal
direction (εn ) and the other one for the tangent direction (ετ ). This increases a little bit the complexity of the expressions
in comparison with the LMM one. In (4.72a) the resulting system of equations is shown.

 
KN N KN M KN SA KN SI 0 0 0    
 KN N KMM KMSA KMSI −k MTAsl −kMTAst −k MTI  ∆ uN rN
 

 KSA N KSA M KS A S A KSA SI kDTAAsl k DTAAst k DTAI 

∆uM 


 rM 

 KS I N KSI M KSI SA KSI SI kDTIAsl kDTIAst k DTII  ∆uSA 


 rS A 

(4.72a)  = −
  rS I
 ∂ rλA
sl
∂ rλA
sl
∂ rλA
sl
∂ rλA
sl
∂ rλA
sl
 ∆ uS I 
 0 0    
 ∂ uM ∂ uS A
sl
∂ u SI ∂λAsl ∂λAst  ∆λAsl 


 rλAsl 

 ∂ rλA ∂ rλA ∂ rλA ∂ rλA ∂ rλA 
 0 st st st st sl
0  ∆λAst   rλAst 
 ∂ uM ∂ uS A ∂ u SI ∂λAsl ∂λAsl 
sl
∂ rλI
∆λI rλ I
0 0 0 0 0 0 λI

As in the LMM case, we separate the LM residual in three parts, for inactive, active slip and active stick DOF. The
resulting residuals in (4.72b). In the case of the ALM, the F for the Coulomb’s frictional law is computed considering
the augmented LM (λ̄n ) instead of the standard one (λn ), this can be expressed as F = −µλ̄n τ .



k2 F
rλAsl = k n · (−n · (Dx1 − Mx2 )) − εn τ · λ −
 k
(4.72b) rλAst = k n · (−n · (Dx1 − Mx2 )) + k ũτ
 2 2
rλI = kεn n · λ + εkτ τ · λ


Vicente Mataix Ferrándiz Page 121 of 374

4.3 Formulation CHAPTER 4. CONTACT MECHANICS

4.3.4.4 Work-flow. Solution algorithm

The main changes relative to the solution loop presented in 4.3.3.5.Work-flow. Solution algorithm for the frictionless
solution is the need to compute a more complex active set, which includes the slip and stick state, as well as the
computation of the corresponding residuals for these states. The last stated, means that the residual corresponding
to the LM is divided into three different components, the relative to the normal direction, the relative to the tangent
direction associated to the slip state, and finally the one obtained from the stick state. It is relevant to separate the
residuals from the slip/stick states, as the magnitude orders from these components change greatly and it may be
difficult to achieve a convergence if mixing them. We remark the need to do this search at each time step in order to
adapt to the evolution of the geometry.

Algorithm 3 Algorithm for the frictional contact problem

1: procedure A LGORITHM FOR THE FRICTIONAL CONTACT PROBLEM
2: t = 0 and i = 0
3: Initialise the solution for u0 = 0
4: In case of solving LM solution, Initialise the LM solution λ0 = 0
5: Initialise the active set A01 and I10 such that A01 ∪ I10 = S and A01 ∩ I10 = ∅
6: Initialise the slip/stick set A0sl1 and A0st1 such that A0sl1 ∪ A0st1 = A01 and A0sl1 ∩ A0st1 = ∅
7: while t < tend do
8: t = t + ∆t and i = i + 1
9: Initialise the increment of solution for ∆ui1 = 0
10: In case of solving LM solution, Initialise the LM increment of solution ∆λi1 = 0
11: Search for potential contact pairs, and if required update the pairs and the active set, respecting step 5
12: We define the problem as not converged conv = false
13: while conv = false do
14: Find the solution corresponding to the system presented in 4.3.4.3.3.Algebraic form of the problem
15: Update the solution, so uin+1 = uin + ∆uin+1 and in case of LM solution λin+1 = λin + ∆λin+1
16: Update the active set as in (4.73). The threshold for each one of the cases is represented in (4.74).
i+1
 i+1
In+1 := j ∈ S|thresholdn+1 ≥0
(4.73) i+1 i+1
An+1 := j ∈ S|thresholdn+1 <0

(4.74) thresholdLM = λn or n · λ, thresholdPenalty = ǫḡn , thresholdALM = k λn + ǫḡn or kn · λ + ǫḡn

17: Update the slip/stick set as in (4.75). To evaluate this we require the frictional threshold F , considering
Coulomb’s law, for the cases represented in (4.76). We also require the tangent contact stress (tτco ) (4.77).

Ai+1 i+1
sln+1 := j ∈ A|tco ≥ Fn+1
τ
(4.75) i+1
Astn+1 := {j ∈ A|tco < Fn+1 }
τ

(4.76) FLM = µn · λ, FPenalty = µεn ḡn , FALM = µ (k n · λ + εn ḡn )

(4.77) tτLMco = kτ · λk, tτPenaltyco = kετ ũτ k, tτALMco = kk τ · λ + ετ ũτ k

18: Compute the corresponding residual to check (4.78).

(4.78) kru k < toleranceu , krλn k < toleranceλn , krλslτ k < toleranceλslτ , krλstτ k < toleranceλstτ

19: Solution converged if Ai+1 i i+1 i i+1 i

sln+1 = Asln+1 , Astn+1 = Astn+1 , In+1 = In+1 and residuals (4.78) are converged.

As well as we did in the frictionless case, we present on Figure 4.23 the output in Kratos representing the NL
convergence check of a time step. We can appreciate the four different residuals which are check, in addition to the

Page 122 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.4 Contact detection. Search techniques

two different active set checked.

Figure 4.23: Example of convergence check in the frictional contact

4.3.4.5 Active set strategy (Semi-smooth Newton Raphson)

The reformulation of frictional contact conditions is simi-
lar to the frictionless case. As well as in the former case,
the LM contribution of the Lagrangian (L) in the ALM
(4.60) replaces the NCP presented in Popp[PhDPop12]
and Gitterle[PhDGit12]. This NCP takes the form of a
two component vector equation and is written as (4.79),
the visual representation can be seen in Figure 4.24.
This representation shows a distinctive behaviour for the
slip/stick states in the similar manner to the one pre-
sented in the Figure 4.22 for the frictional Lagrangian
(L) contribution. As in the frictionless case, for the fric-
tional problem the PDASS is taken into consideration in
considering the workflow presented in 4.3.4.4.Work-flow.
Solution algorithm.
Figure 4.24: NCP for the frictional contact problem.

(4.79) Cτ (λτ , ũτ ) = max(µλ̃n , kλ̃τ k)λτ − µmax(0, λn )λ̃τ

4.4 Contact detection. Search techniques

4.4.1 Introduction
Detection phase it is quite different depending on if we are considering an implicit or an explicit approach[PhDYas11]:

• Explicit approach: Here, as the time step is usually very small, we need to detect the solids that actually have
penetrated each other, and apply the corresponding forces in order to repulse them. There are different ways to
apply these repulsive forces, but the point is that we focus on real penetration, not estimations.

• Implicit approach: As the solution of the system depends on the values that have not been completed yet,
we need to estimate potential pairings. With this potential pairing, we include the corresponding DOF, with
the corresponding additional Lagrange multiplier, and assemble the LHS and RHS related to these potential
pairings.

As the target of this work is the consideration of an implicit approach, we need to take care of the problematic of
the former. The detection phase is crucial, as the additional DOF and assembling is not only numerically expensive,
but the condition of the system is significantly affected by them.
Additionally, for a problem with n slave elements and m master elements, the cost of a naive brute force searching
algorithms (searching all master elements for valid intersections with each slave element) is O(nm), which is
unacceptable for a problem with a large amount of contact surface elements. The ultimate goal of a contact searching
algorithm is to find all proximate slave-master element pairs. In conclusion, this phase may be as important as the
formulation considered as it influences the whole resolution of the problem.
These methodologies have been particularly developed in recent years, thanks not only for FE requirements,
but also computer graphics(physically based modelling, animation)[ArtZK12; ArtRKC02], robotics (motion planning,
collision avoidance)[ArtTK14], industrial applications (virtual prototyping, assembly tests)[ArtFig10; ArtFF03] and

Vicente Mataix Ferrándiz Page 123 of 374

4.4 Contact detection. Search techniques CHAPTER 4. CONTACT MECHANICS

Figure 4.25: Conventional penetration estimations. Symmetric: a segment intersection; b volume intersection.
Asymmetric: c,d segment in volume; e,f node in volume; g,h node under surface. Source[PhDYas11]

videogames[BookEri04; ArtCai+14], where the Collision Detection (CD) plays an important role in their respective
fields.
The penetration is another aspect to take into account when detecting potential contact pairs. There is a whole
science in the contact penetration or gap estimation. From the more conventional methods, Figure 4.25, through more
advanced or recent methods such as the estimation of penetration by ray-tracing[ArtPR15a]. In this work we present
an approach consistent with the Mortar formulation considered for the contact computation.
In what follows, a contact searching algorithm with Bounding Volume Hierarchies (BVH) can be applied
(4.4.3.Tree structures), furthermore, there are different kinds of bounding volumes that can be used to improve this
search, see 4.4.2.Bounding volumes. Finally, the penetration estimation is an important step too, and the proposed
approach will be presented at 4.4.4.Penetration definition.

4.4.2 Bounding volumes

There are many kinds of DOP that can be used to define the bounding volumes, Figure 4.26a illustrates some of them.
We address to the literature, particularly to the work of Dinas[ArtDB15], which summarises each methodology and
makes a comparison between them. There are several different ways to define a bounding volume for a geographical
region.
The simplest DOP probably will be to consider the nodes inside a radius, a radius search or BS. Another very
simple solution, define a cuboid whose surfaces are parallel to the x − y , y − z and x − z surfaces, is called AABB,
but these two methods have the drawback of the lack if tightness, and many slave and master surfaces are likely to be
tested for proximity even though they are not so close (Figure 4.26b). Another approach is the OBB, which is much
tighter and therefore giving a better fit, but with an additional computational cost and memory consumption. Finally, it
can be considered the k-DOP (convex hull) to define the bounding volumes, which are always tighter than AABB and
may be tighter or looser than OBB depending of the geometries of the objects. However the k-DOP are much simpler,
cheaper and more efficient for the detection of k-DOP overlapping, which is quite relevant for our applications since
deformation of contact bodies requires frequent updating of these volumes[ArtYL08].
In any case the Figure 4.26a as the decision to be taken results in a compromise solution between numerical cost,
complexity and precision. For this reason the OBB has been chosen among all the options, as a intermediate solution
that will satisfy the requirements. Originally a BS approach was chosen, but it was insufficient, as in order to detect all
the contact pairs, a large enough radius factor must be taken into account (see Figure 4.27a), returning many false
potential pairs, which impoverished the convergence of the problem. Additionally, this approach is very dependent on
the radius r chosen, as can be seen in Figure 4.27b.

4.4.2.1 OBB implementation

Fortunately, we have plenty of documentation on how to implement this type of object thanks to its wide use in
video game graphics engines. Our implementation is based on the one considered in the documentation of the
GeometricTools[BookEbe99] library and the tutorials found for game development[OnlCho].

Page 124 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.4 Contact detection. Search techniques

(a) Types of Discretised Orientation Polytopes (DOP)

(b) Proximity tests for DOP

Figure 4.26: Types of bounding volumes. a) The radius search (Bounding Spheres (BS)). b) Axis-Aligned Bound-
ing Box (AABB). c) OBB. d) Discrete Orientation Polytopes (k-DOP) (convex hull). Inspired on [ArtYL08]

In order to define the collision algorithms, we can consider between two OBB we need to define the geometry
which defines it properly. Considering we are in ℜ3 . An OBB is defined[BookEbe99] by a centre C, a set of right-
handed orthogonal axes A0 , A1 and A2 , and a set of real positive extends a0 , a1 and a2 . With this we can define the
OBB as solid box following Equation (4.80a). Additionally Equation (4.80b) represents the eight vertices of the box,
where |σi | = 1 for all i.

2
X
(4.80a) C+ xi Ai : |xi | ≤ |ai | for all i
i=0

2
X
(4.80b) C+ σi ai Ai
i=0

Our interest would be to build the OBB from a provided geometry in order to automate its construction. We have
two options in 3D for surface geometries. The first approach takes the AABB, the lowest and highest point, and builds
and axes and defines an orthonormal base. The second approach will be based on the normal which defines that
surface geometry. Taking this normal, we can define a base, rotate the points to this base and get the point at the
maximum distance from the geometry centre. Combining the tangent base in order to define a vector in the direction
to the furthest point, then the third axis is taken as the cross product of the normal and this vector. This last method
defines a less arbitrary bounding box than the first one. The difference can be seen in Figure 4.29, where we build
the OBB for 4 triangles, 2 in the top and 2 over the first ones. The first one, Figure 4.29a, may look more regular, but
just takes an arbitrary direction as all the nodes are in the same plane. The one from Figure 4.29b defines a larger
OBB, and it is oriented to the furthest point from the centre, which is less arbitrary.

4.4.2.1.1 Collision detection :

Vicente Mataix Ferrándiz Page 125 of 374

4.4 Contact detection. Search techniques CHAPTER 4. CONTACT MECHANICS

(a) Poor search results obtained with the BS approach (b) Influence of radius in BS. Source[PhDYas11]
Figure 4.27: BS approach

Figure 4.28: OBB. Source[ArtRom11]

The simplest collision detection consists in two steps, a first stage where we check that the points of one OBB are
inside the other, the second stage consists in check if there is an intersection between the faces of the OBB. For
the first step, we need to define the inside operations for a OBB. In order to do, the simplest procedure consists in
rotating the points so these points became oriented with the Cartesian axis, Equation (4.81) and (4.82), and therefore
consider the standard AABB inside algorithm.
For 2D we consider the following signs:
 
signx2D = −1.0 1.0 1.0 −1.0
(4.81a)
signy2D = −1.0 −1.0 1.0 1.0

For each point i of the destination OBB:

(4.81b) xaabb
i = Cdest + signxi2D Adest
0 a0
dest
+ signyi2D Adest
1 a1
dest

We check if is inside the origin OBB:

(4.81c) kxaabb
xi − Cxorig k <= a0orig and kxaabb
yi − Cyorig k <= a1orig

For 3D we consider the following signs:

 
signx3D = −1.0 1.0 1.0 −1.0 −1.0 1.0 1.0 −1.0
(4.82a) signy3D = −1.0 −1.0 1.0 1.0 −1.0 −1.0 1.0 1.0
signz3D = −1.0 −1.0 −1.0 −1.0 1.0 1.0 1.0 1.0

For each point i of the destination OBB:

(4.82b) xaabb
i = Cdest + signxi3D Adest
0 a0
dest
+ signyi3D Adest
1 a1
dest
+ signzi3D Adest
2 a2
dest

We check if is inside the origin OBB:

(4.82c) kxaabb
xi − Cxorig k <= a0orig and kxaabb
yi − Cyorig k <= a1orig and kxaabb
zi − Czorig k <= a2orig

Page 126 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.4 Contact detection. Search techniques

(a) AABB aligned OBB (b) Normal plane aligned

Figure 4.29: OBB from surfaces build comparison

For the intersection of the faces, the very same algorithms considered for the intersections of lines in 2D and
quadrilaterals in 3D can be considered here. In total, the number of checks necessaries in order to check the
intersection is 24 in 2D and 52 in 3D. Of course, not all checks are required. In case of one positive check we already
know that there is at least one intersection.

4.4.2.1.2 Collision detection with SAT :

(a) Illustration of the separating axis theorem (I) (b) Illustration of the separating axis theorem (II)
Figure 4.30: Separating axis theorem

This algorithm is more refined than the previous one.

In this case instead of direct check each one of the
nodes and faces, we apply a very simple geometrical
concept. SAT[ArtRom11] is based on the separating axis
theorem, Figure 4.30, which states that it is sufficient
to find one axis that separates A and B to be sure that
they are disjoint and so do not overlap. Therefore, if we
are able to draw a line to separate two polygons, then
they do not collide (Figure 4.31). This concept can be
used not only between OBB but also between general
polygons or even a polygon and a circle. The definition
of the SAT consists on two steps, during the first step we
define each one of the potential planes to be checked,
the definition of each plane is found in Equation (4.83), Figure 4.31: SAT concept. Source[OnlCho]
and secondly we check if there is a separation plane between the two OBB with the one presented on (4.84).

The normal vectors to be considered in 2D correspond with the right-handed orthogonal axes A0 and A1 from

Vicente Mataix Ferrándiz Page 127 of 374

4.4 Contact detection. Search techniques CHAPTER 4. CONTACT MECHANICS

both OBB, and the following cross-products:


orig orig
 Adest
 dest
0 , A1 , A0 , A1
orig
(4.83a) n Adest
0 × A0 , Adest0 × Aorig
1
 Adest × Aorig , Adest × Aorig

1 0 1 1

The normal vectors to be considered in 3D correspond with the right-handed orthogonal axes A0 , A1 and A2 from
both OBB, and the following cross-products:

orig orig orig

 Adest dest dest
0 , A1 , A2 , A0 , A1 , A2
 orig orig
× Aorig
 dest
A0 × A0 , Adest 0 × A1 , Adest 0 2
(4.83b) n orig orig orig
 Adest
 1 × A 0 , Adest
1 × A 1 , A dest
1 × A2
 orig
× Aorig × Aorig
 dest
A2 × A0 , Adest 2 1 , A2
dest
2

In 2D the algorithm for checking if there is a separation plane, being n the normal which defines the plane:

k(Cdest − Corig ) · nk >

(4.84a) k(Adest
0 a0
dest
) · nk + k(Adest
1 a1
dest
) · nk +
k(Aorig orig orig orig
0 a0 ) · nk + k(A1 a1 ) · nk

In 3D the algorithm for checking if there is a separation plane:

k(Cdest − Corig ) · nk >

(4.84b) k(Adest
0 a0
dest
) · nk + k(Adest
1 a1
dest
) · nk + k(Adest
2 a2
dest
) · nk +
k(Aorig orig orig orig orig orig
0 a0 ) · nk + k(A1 a1 ) · nk + k(A2 a2 ) · nk

The resulting number of checks therefore to be evaluated in this case would be of 8 for the 2D case and 15 in
3D.

4.4.3 Tree structures

Figure 4.32: Space subdivision approaches

The tree structures are computer data structures that emulate a tree structure with a set of binary nodes[Book-
FSS12], see Figure 4.33a. A node that is upper in the hierarchy respect to other nodes which is connected with him is
called parent node, and node below is denominated child node. The node in the top of the tree is called root node,
while the nodes in the bottom are called leaf nodes.

Page 128 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.4 Contact detection. Search techniques

A data structure which stores the data points with respect to their position in space is needed to perform a search
for neighbours. In two-dimensional spaces, the standard data structure is a quadtree, and the octree is analogue in
three-dimensional space.

The kd-tree data structure is based on a recursive subdivision of space into disjoint hyperrectangular regions
called cells (see Figure 4.33b). Each node of the tree is associated with a region called box, and is associated with a
set of data points that lie within this box. The root node of the tree is associated with a bounding box that contains all
the data points.

In addition to the data points themselves, a kd-tree is specified by two additional parameters, the bucket size
(threshold of the number of data points associated with a node) and a splitting rule (which determine how a
hyperplane is selected). For more details look up the literature[ArtYL08; BookPhD06; OnlRtr].

(a) Binary tree (b) A kd-tree of bucket size one and the corresponding spacial decomposition
Figure 4.33: Tree structures

The binary structure tree that will be considered in this work will be therefore the kd-tree. This tree is already
available in the Kratos code.

4.4.4 Penetration definition

As we have already mentioned, the main difference in explicit and implicit simulation detection is that in the former
case we predict possible contact phenomena, and we define potential contact pairs before penetration happens. This
means that slave nodes approaching master surfaces have to be detected at a certain distance, which is usually
denoted as Maximal Detection Distance (MDD). The MDD is a significative parameter of the detection procedure.
This MDD should be considered as small as possible in order to avoid the creation of non-necessary contact pairs. In
order to do that we should define an automatic procedure which defines this value in function of certain parameters,
such as the element size (h). Once this is defined we need to estimate the penetration, so we can compare it with the
MDD.

For that, we highlight that contact penetration estimation has its own advantages and disadvantages, and usually
these techniques are uncoupled of the contact formulation considered. In order to compute a consistent penetration,
or gap (gn ), with the weighted gap (ḡn ) computed during the contact contributions on 4.3.3.4.2.Mortar operators we
will consider this very same Mortar formulation as the base concept.

This is done using the Mortar mapper developed in this work. The concept is to map the coordinates of one
domain to the other, and once these coordinates have been mapped into a destination domain we can compute
the consistent gap node-by-node applying the gn formula, gn = −n · (x1 − χh x2 ). The obtained gap is equal to the
weighted gap divided by the total Mortar integration area which involves the node. This value has length units and
therefore can be compared with some reference length. This very same concept can be applied in order to compute a
consistent slip in case it is of interest (mostly for post-process motives).

We can illustrate this former concept with an example. The Figure 4.34 shows a full example, consistent in a
circular surface and a plane surface, Figure 4.34a. As one of the sides is plane, it is easy to estimate the gap as the
y coordinate of the circular side. The nodal area corresponds with the integrated region in each node, which is the
whole section until the circular section does not intersect with the plane any more, see Figure 4.34b. With this we can

Vicente Mataix Ferrándiz Page 129 of 374

4.4 Contact detection. Search techniques CHAPTER 4. CONTACT MECHANICS

(a) Mesh of the example (b) Nodal area

(c) Consistent gap (d) Weighted gap

Figure 4.34: Consistent gap example. Circular surface vs plane

see as if dividing ḡn (Figure 4.34d) by the normal area, the consistent normal gap (gn ) is obtained (Figure 4.34c). This
one coincides with the predicted values (the height to the circular region).
A more complex case its illustrated in the Figure 4.35, where the double curvature considered to test the Mortar
mapper in E.Mortar mapper is considered to compute a consistent gap. The first figure shows the mesh, Figure 4.35a,
and the second one in Figure 4.35b. In this last image, we can see like the gap varies according to the curvature and
following the "parallel lines" that define this double curvature geometry. Some peaks appear in the nodes that belong
to coarse mesh regions.
The details of the implementation of this mapper can be found in the respective appendix, E.Mortar mapper. The
definition of the consistent gap is detailed in the algorithm 8.

4.4.5 Self-contact detection

4.4.5.1 Introduction
There are mechanical problems for which the determination of the contact pairs is not trivial, and especially complex
are the problems where we do not enforce any master-slave9 (or mortar and non-mortar ) relationship. For two or
more independent bodies a priori we can assign these relationships, but this is not the case for self-contact problems,
where there is only one body which contact with itself as the name indicates. This class of contact problems needs a
particular contact detection procedure.
9 We will consider the master-slave criteria considered in the formulation, as done by Popp[PhDPop12], where the slave conditions correspond

with the one where the integration is performed.

Page 130 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.4 Contact detection. Search techniques

(a) Mesh of the example (b) Consistent gap

Figure 4.35: Consistent gap example. Double curvature surfaces

There is not much in the literature detailing the treatment of the self-contact problem, and the more detailed
are probably the contributions of Yastrebov [PhDYas11; ArtYCF11; BookYas13], which are designed for the NTS
integration. There are indeed a lot of publications[ArtSN17; ArtDem+12; ArtCam15] detailing this kind of problematic
for fabric or very slender elements, as rods, where the self-contact is a common issue. A self-contact is more
probable for thin or oblong solids, for which one or two dimensions are much smaller than others than for solids with
all dimensions of the same order. In any case, the work presented here is more focused for solids than the slender
geometries mentioned before.

4.4.5.2 Algorithm considered

The detection time is higher than for the contact of the same order with known a priori master-slave discretisation,
because the preliminary stage requires several steps hardly parallelisable. This detection times vary in function of
several parameters, like the normal discrimination, the initially assigned pairs, and in general all the steps involved in
the presented algorithm.
A relevant, but also a big limitation when automatically assign
contact pairs is that in principle a node cannot be master and
slave at the same time, and that situation should be avoided. We
will deactivate a condition when assigning a master/slave relation
to the nodes and that condition holds master and slave nodes
at the same time. This for example can be seen in the tubular
example from 4.4.5.3.2.Tubular detection, where the conditions
that share master and slave nodes will be deactivated. This
is a big drawback, but in large meshes with small sizes, these
conditions appear isolated and does not represent a significative
amount respect the total. Figure 4.36: Self-contact normals descrimination
Before starting our algorithm for automatically pair conditions that may contact between them, we perform an initial
search, where we potentially pair all conditions against all conditions. As this can be very expensive, numerically
speaking, not only to perform, but also to filter later with our approach, we should pre-filter as many pairs as it is

Vicente Mataix Ferrándiz Page 131 of 374

4.4 Contact detection. Search techniques CHAPTER 4. CONTACT MECHANICS

possible for us. The method which provides the best results, and the simplest, is probably to check the similarity
between the normals of the conditions. If the normals are close, not necessarily equal, therefore we can consider that
these are pointing in the same direction and therefore will not deal with contact. This can be appreciated in Figure
4.36, where normals 1 and 3 are quite similar, so we can easily discriminate the pair 1 − 3. On the other hand, the
difference between 1 and 2 is more subjective and it will depend on the threshold we consider to discriminate the pair
1 − 2.

Our proposal takes as a main hypothesis that the master/slave conditions are neighbours between them. This
assumption is based on the fact that when we define manually certain conditions as the master or slave usually
we define a region for each domain. Therefore, our goal is to replicate or mimic the behaviour that a human would
consider when assigning the domains. Therefore, we will create a list of potential slave conditions and we will fill
this list following a neighbourhood criterion. The idea is to create a list10 of conditions, that we will fill based on the
proximity respect to the condition previously added to the list, except if we have already assigned that condition as a
master, in that case we will jump to the next candidate.

Algorithm 4 Self-contact detection algorithm

1: procedure S ELF - CONTACT DETECTION ALGORITHM
2: We perform an initial search with all conditions againts all conditions
3: for all node ∈ ContactMeshnodes do
4: Reset SLAVE flag
5: Reset MASTER flag
6: for all condition ∈ ContactMeshconditions do
7: Reset SLAVE flag
8: Reset MASTER flag
9: Reset ACTIVE flag
10: We create a set ordered_conditions that will contain all conditions in order of proximity between them
11: We add as first condition to this set one choosen arbitrarily
12: We create a map containing the boundaries of all conditions
13: for all condition ∈ ContactMeshconditions do
14: for all boundary ∈ conditionboundaries do
15: Insert the connectivity of the boundary and the geometry where it belongs
16: for all condition ∈ ordered_conditions do
17: for all boundary ∈ conditionboundaries do
18: if A neighbour is found then
19: Add that neighbour to ordered_conditions
20: else
21: Add arbitrary condition not added previously to ordered_conditions
22: for all condition ∈ ordered_conditions do
23: for all pair ∈ conditionpotential − pairs do
24: if Condition are neighbours then
25: Skip to the next pair
26: else
27: if Check the nodes of the pair are not SLAVE then
28: Assign MASTER flag to condition and nodes in the pair condition
29: Clear pair condition database
30: if At least one pair is MASTER then
31: Assign SLAVE flag to condition and nodes
32: for all condition ∈ ContactMeshconditions do
33: if condition shares MASTER/SLAVE nodes then
34: Unset ACTIVE flag to condition
35: else
36: Set ACTIVE flag to condition

10 We have considered std::unordered_set in particular.

Page 132 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.4 Contact detection. Search techniques

The previous list will be filled at first by an arbitrary condition, and starting from this first condition the rest of the
list will be filled. In order to determine the neighbours of each condition, we will create a map11 containing all the
boundaries nodes and the original conditions where it belongs. Later, with this list of boundaries we will fill the set of
conditions progressively, adding a neighbour condition to the last one added. If no neighbour is found we will add
other arbitrary conditions, not added previously.
Once we have filled the set of conditions, we proceed with the potential master/slave assignment. We iterate over
this set of ordering conditions, if the condition has been previously marked as masters we skip it, if for we check their
potential pairs. If the conditions are neighbours, we skip to the next potential pair. In the opposite case, we must
check that no node of the potential pair is already assigned as slaves, if this is the case we can assume this pair is a
master surface, and we will clear its potential pairs, simplifying our assigning process. If the condition has any master
condition assigned, we can assume is a slave condition. Finally, we will check if the conditions share master/slave
nodes, if that is the case we deactivate the condition, if not we can consider is an active one.
All the previously stated procedure is summarised in the Algorithm 4, which additionally includes some additional
implementation details, in any case we present it in the most pseudo-code way possible.

4.4.5.3 Examples
4.4.5.3.1 Planes detection :
The first example we will show is a very simple case where there is no node shared between the domains,
and therefore we can easily manually assign the master/slave roles. If the procedure works properly will return the
expected result, which means that one plane will be fully master, and the other one will be fully slaves. This simple
case is part of the ContactStructuralMechanicsApplication test suite and it is located in the following
file.

(a) Mesh of the example (b) Master flag (c) Slave flag
Figure 4.37: Self-contact detection simplest case, two parallel planes

The result can be seen in Figure 4.37, where we can appreciate like the lower plane is fully master (Figure 4.37b),
and the upper plane fully slave (Figure 4.37c).

4.4.5.3.2 Tubular detection :

The case consists of a very simplified ring, formed on quadrilaterals. The complexity comes from the fact that
there are shared nodes between the potential domains, which means that as we confront a pure self-contact problem.
In this case we assign as initial potential pairs everything with everything, our procedure will filter this lately. This case
can be found as a test in here.
The first example is the case considering only one element of thickness. See Figure 4.38, were due to the initial
assignment of everything with everything, the number of master conditions (Figure 4.38b) is larger than the number
of the slave ones (Figure 4.38c). This is due to the fact that as we follow the procedure, we set as master all the
potential pairs, before jump to the next potential slave condition. Additionally can be seen like those two conditions
will be inactive, as they share master and slave nodes simultaneously.
The second example is a little bit more complex, in here we consider two elements of thickness. This increases
the complexity, as the neighbourhood is not as simple as in the first case. The results obtained, Figure 4.39, look like
an extension of the former case. In that case, nothing else to add to the comments previously stated.
11 We do this at the programming level considering std::unordered_map in this case.

Vicente Mataix Ferrándiz Page 133 of 374

4.4 Contact detection. Search techniques CHAPTER 4. CONTACT MECHANICS

(a) Mesh of the example (b) Master flag (c) Slave flag
Figure 4.38: Self-contact detection for simple tubular case

(a) Mesh of the example (b) Master flag (c) Slave flag
Figure 4.39: Self-contact detection for tubular case

4.4.5.3.3 Contact example (S-shape profile) :

The problem consists in a S-shape profile which is enforced to self-contact, where we impose a negative vertical
displacement on the top face. The geometry and the mesh of the problem can be seen in Figure 4.40. It is relatively
easy, but a bit tedious, to manually assign the master/slave pairs, at least for the initial deformations. Because of this,
we can consider it as a good example to test the procedure. In any case the automatic assignment is not trivial as the
number of triangular faces is large and the neighbourhood must be properly detected. The example can be found in
the Kratos Examples repository, in here.

The results can be seen in Figure 4.41. Where on Figure 4.41a the resulting displacement can be seen, with the
corresponding expected contact in the corners. Additionally the Figure 4.41b shows the master nodes, which are the
vast majority and the Figure 4.41c the slave nodes, where the featured nodes correspond to the corners in contact.
In order to better understand the result obtained it is relevant to remember the fact the domain considered as the
one where to perform the integration is the slave side, therefore the default domain corresponds with the master one.
Because of that, the most predominant domain the master side.

Page 134 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

(a) Mesh seen in front (b) Mesh seen in perspective

Figure 4.40: S-shape profile mesh (tetrahedra)

(a) Displacement (b) Master flag (c) Slave flag

Figure 4.41: S-shape profile resulting self-contact simulation

4.5 Numerical examples

The following section presents several examples commonly used in order to check the proper functioning of the CCM
implementations. The tests presented will go from less complex to more complex, starting from the most basic patch
tests to the most common and complex contact benchmarks. The examples are solved considering ALM.

4.5.1 Basic patch test

(a) Straight interface (b) Slope interface

Figure 4.42: Geometries of the simplest patch test

Vicente Mataix Ferrándiz Page 135 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

Here the most basic of all possible patch test is checked. These consist it to block sharing a clear interface between
the domains. The result expected is a continuous gradient in the displacement in the normal direction of the interface
for the frictionless contact, a much more extensive casuistry for frictional case which will be explained in detail
below. Here we present two different cases, the first one from Figure 4.42a where the interface between two 1 × 1
squares is straight. The second case, from Figure 4.42b, where the interface is a slope between two non-regular
quadrilaterals.

Body E ν ρ
Die 2.069 × 1011 Pa 0.29 1000
Block 2.069 × 1011 Pa 0.29 1000

Table 4.4: Parameters considered for simple patch test

4.5.1.1 Frictionless
For both problems a vertical displacement of 0.1 meters
will be imposed in the top face of the upper quadrilateral.
The properties of the materials correspond with the ones
shown in Table 4.6. With this given BC, the solution ob-
tained for the first problem, Figure 4.43a, is the expected
one, a continuous gradient of the displacement in the
vertical direction. On the other hand, the result of the
second one, Figure 4.43b, also corresponds with the
expected behaviour, a continuous solution in the vertical
direction.
One additional remark to the first case is the fact that
we are considering the same Poisson’s ratio (ν ) for both
bodies, the deformation of the interface is symmetric in
the interface, giving the impression that the displacement
is also tied in the tangential direction. But for example
if one of the ν is taken as zero, this body will deform just
vertically and the other one will experiment Poisson’s
effect. (a) Straight interface (b) Slope interface
Figure 4.43: Solution for the frictionless simplest patch test
4.5.1.2 Frictional

(a) Displacement solution slip (b) Slip solution slip (c) Displacement solution (d) Slip solution stick
stick
Figure 4.44: Geometry and solution for the frictional simplest patch test

Page 136 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

Here for the sake of simplicity only the problem with straight interface, from Figure 4.42a, will be studied. In the
study we need to differentiate between two possible states in the frictional case, the slip and the stick state. In here, in
contrast with the frictionless case, we set a vertical load in the top face of the upper element, additionally a tangent
load in order to trigger the tangent behaviour. This problem thanks to its simplicity allows us to check the correctness
of the slip/stick state detection, where just adjusting the friction coefficient µ it is possible to change between one state
and the other. The first case corresponds with the slip states, where in Figures 4.44a and 4.44b we can appreciate
like a relative drift appears on the interface. On the other hand, on the Figures 4.44c and 4.44d, the solution for
the stick state is shown, and can be seen as the two blocks move in solidarity. Notice that Figures 4.44b and 4.44d
represent the SLIP flag, the value can be only 0 and 1, being 1 slip state, and 0 stick state.

4.5.2 Taylor patch test

The Taylor [ArtTP] patch test is slightly more complex problem than the former patch test. In this case the interface is
not coincident between th two domains, additionally a distributed load is considered in the upper face of the domains.
The load considered is p = 10Pa, and the properties of the materials correspond with the ones shown in the Table 4.5.
We will present the solution for the 2D and 3D problem.

Body E ν ρ
Die 3 × 103 Pa 0.4 1000
Block 3 × 103 Pa 0.4 1000

Table 4.5: Parameters considered for Taylor [ArtTP] patch test

4.5.2.1 2D
The Figure 4.45a illustrate the setup of the Taylor test in 2D. The solution obtained in the other hand corresponds
with a continuous gradient of vertical displacement as shown in Figure 4.45b and a continuous vertical stress Figure
4.45c.

(a) Setup (b) Displacement solution (c) Stress solution

Figure 4.45: Solution for the Taylor patch test in 2D

4.5.2.2 3D

(a) Setup (b) Displacement solution (c) Stress solution

Figure 4.46: Solution for the Taylor patch test in 3D

Vicente Mataix Ferrándiz Page 137 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

For the 3D case, see Figure 4.46a, the same conclusions than in the 2D case can be applied here. Continuous
gradient of displacement (Figure 4.46b) and continuous stress field (Figure 4.46c) in the vertical direction.

4.5.3 Friction base test

The following test, extracted from the work of Dong[ArtDon99], is a test which allows to study the effects in the contact
frictional behaviour with respect to different values of the friction coefficient µ. The properties of the bodies are the
ones shown in the Table 4.6, where µ has been considered with three different values, 0.25, 0.5 and 1. The geometry
and mesh considered are the ones from Figure 4.47 and the load q = 20000KN /m. The solution is compared with the
one in reference[ArtDon99] in Figure 4.48b, with a very good agreement. The solution from Figure 4.48a shows the
detachment that appears in the interface for µ = 0.25.

Body E ν µ
Die 2.1 × 1011 Pa 0.29 1, 0.5, 0.25
Block 2.1 × 1011 Pa 0.29 1, 0.5, 0.25

Table 4.6: Parameters considered for simple patch test

(a) Geometry (b) Mesh

Figure 4.47: Friction problem from Dong[ArtDon99]

(a) Displacement solution for µ = 0.25 (b) Solution compared with reference
Figure 4.48: Solution for pure friction problem

4.5.4 Hertz problem

We have already highlighted the relevance of this test in the former sections. This is the most commonly used and
extended benchmark for contact mechanics, originally published by Hertz[ArtHer82] in 1882. This benchmark has the
advantage that the original solution is analytical and therefore we can compare with a solution that we have known a

Page 138 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

priori for different combinations of material and geometries. Using as main reference the solution of Zhu[ArtZhu12]
for the analytical solutions of the different Hertz problems. We will study basically two cases, the rigid plane-sphere
contact and sphere-sphere12 contact problem, both for 2D and 3D problems. It will be assumed we work under the
hypothesis that we work in the domain of infinitesimal deformations.

4.5.4.1 2D
4.5.4.1.1 Plane-sphere :
The plane-sphere configuration from Figure 4.49 corresponds with the problem to solve, which requires in 2D
to be solved considering axisymmetric formulation. We specialise the resolution of the problem for a radius equal
to 6.1237 meters, and pressure of P = 5 × 105 Pa. The properties considered for the two solids are represented on
Table 4.7, where the plane can be considered de facto rigid. We have considered different mesh sizes, and we have
compared the results between then and the respective error.

Body E ν
Sphere 1 × 108 Pa 0.29
Block 1 × 1026 Pa 0.29

Table 4.7: Parameters considered for 2D sphere-plane Hertz benchmark

(a) Setup (b) Mesh

Figure 4.49: Setup for the 2D sphere-plane Hertz benchmark
The analytical solution is presented in Equation (4.85), taken from the reference[ArtZhu12]. The Equation (4.85c)
gives the analytical solution for the contact pressure in all the domain b, then the Equation (4.85d) provides the
vertical displacement of the contact interface.

1 p
(4.85a) Eeff = 1−ν12 1−ν22
,r = x2 + y2
E1
+ E2
s s
2 2
 4
3P π R 3 (1 ν2) R b
q
− 3
Rd03
3
(4.85b) b= , p0 = 3P , d0 = , f0 = Eeff
4Eeff 2b2 R
r
r2
(4.85c) pn = p0 1−
b2
s
 3
2 r2
(4.85d) y = − p0 π b2 1−
3 b2

As previously stated, the solution obtained corresponds with the solution for different mesh sizes. Figure 4.50a
presents the different solutions for the vertical displacement for different mesh sizes, and Figure 4.50b the same but
12 In 3D, in 2D a contact cylinder-cylinder will be considered

Vicente Mataix Ferrándiz Page 139 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

for the contact pressure. It is notorious that the displacement solution converges even for very coarse meshes and
the pressure solution it is more difficult to be converged, particularly in the contact frontier. In fact, it can be noticed
that the finer mesh does not provide necessarily the better solution, and more intermediate meshes present better
results.

(a) Displacement solution (b) Stress σyy solution

Figure 4.50: Solution compared for different mesh sizes for the 2D Hertz plane-sphere contact

This last statement can be seen in an easier manner in the solutions presented on Figure 4.51, where the error for
the displacement (Figure 4.51a) and the pressure (Figure 4.51b) are compared for different mesh sizes respect the
analytical solutions.

(a) Error in displacement solution (b) Error in stress σyy solution

Figure 4.51: Error compared for different mesh sizes for the 2D Hertz plane-sphere contact

4.5.4.1.2 Cylinder-cylinder :
The configuration for this problem is shown in Figure 4.52. This consists in two infinite cylinders, not to be confused
with spheres, that become into contact. We will consider the parameters from Table 4.8. Additionally the following
values have been taken into account as BC and geometrical values: q = 0.05851Pa, p = 0.625Pa and R = 8m.

4.5.4.1.2.1 Frictionless case :

Page 140 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

Body E ν µ
Upper cylinder 200Pa 0.3 0.2
Lower cylinder 200Pa 0.3 0.2

Table 4.8: Parameters considered for 2D cylinder-cylinder Hertz benchmark

(a) Setup (b) Mesh

Figure 4.52: Setup for the two cylinders Hertz benchmark

First we present the solution for the frictionless part. Considering µ = 0, the normal contact pressure corresponds
with (4.86b). This is what is represented in Figure 4.53. The comparison with the analytical solution will be presented
all together with the frictional solution.

s

2R 2 p 1 − ν 2
(4.86a) b=2
Eπ
4Rp p
(4.86b) pn = b2 − x 2
π b2

(a) Displacement solution (b) Stress σyy solution

Figure 4.53: Solution for the two cylinders frictionless Hertz benchmark

Vicente Mataix Ferrándiz Page 141 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

4.5.4.1.2.2 Frictional case :

The solution considered has been taken from Wang[ArtWZ13] and Gitterle[PhDGit12], and the resulting is
compared with normal (4.86b) and tangential pressure (4.87b) of reference. Figure 4.54 presents and compares the
solution with the reference, showing a very good agreement with it.

r
q
(4.87a) c=b 1−
µp
√ √
pt = µ 4Rp 2 2
π b2 (√b − x − c2 − x 2) if |x | ≤ c
(4.87b) 4Rp
pt = µ πh2 ( b − x 2 )
2 if c < |x | ≤ b

Figure 4.54: Solution for the two cylinders frictional Hertz benchmark

4.5.4.2 3D

4.5.4.2.1 Plane-sphere :

The description and solution for this problem corre-

spond with the previously introduced at 4.5.4.1.1.Plane-
sphere, as this was solved considering an axisymmetry.
We will consider a sphere of 12.2474 meters of diameter
with a load of p = 5 × 105 Pa and the material properties
from Table 4.9. We consider two different mesh sizes,
Figure 4.55 shows the finer mesh. The calculation is
done in just one static step. This problem can be found
in the following link.

We compare the resolution of both meshes with the

reference solution. Figure 4.56 exposes the displace-
ment and VM stress solution. For the displacement we
got a very good agreement with the analytical solution
in both cases, but particularly good in the case of the
refined mesh, Figure 4.57a, in the case of the contact
pressure we got a better approach within the finer case,
Figure 4.58a.
Figure 4.55: Mesh considered for 3D Hertz plane-sphere
Page 142 of 374 Vicente Mataix Ferrándiz
CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

Body E ν
Sphere 1 × 108 Pa 0.29
Plane 2.1 × 1011 Pa 0.29

Table 4.9: Parameters considered for 3D Hertz plane-sphere problem

(a) Displacement solution (b) VM stress solution

Figure 4.56: Solution for the 3D Hertz plane-sphere contact

(a) Displacement solution (b) Stress σyy solution

Figure 4.57: Solution compared for different mesh sizes for the 3D Hertz plane-sphere contact

(a) Error in displacement solution (b) Error in stress σyy solution

Figure 4.58: Error compared for different mesh sizes for the 3D Hertz plane-sphere contact

Vicente Mataix Ferrándiz Page 143 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

4.5.4.2.2 Sphere-sphere :

This problem, which comes from here consists in two contacting

spheres as seen in the reference solution from Figure 4.59. We con-
sider two hemispheres of 12.2474 meters of diameter, and considering
an applied load of q = 1.0e3Pa. We can transform that given pressure
2
to the corresponding force of the reference solution with F = q π4D . The
characteristic parameters of the spheres are the ones present in Table
4.10. The calculation is performed in just one static step.

As well as the cooperation with the reference solution, we will compare

the radius of the contact area and the maximum contact pressure, corre-
sponding with the analytical solution from (4.88). The obtained solution
is compared with the reference one in (4.89), in here the maximal error
is around a 1%. The displacement and VM stress field from Figure 4.60
shows the obtained solution.

Figure 4.59: Reference solution

3F for two spheres Hertz contact.
(4.88a) Pmax = Source[ArtZhu12]
2π a2
v h i
2 2
u 3F 1−v1 + 1−v2
u
u E1 E2
(4.88b) a=t
3
 
1 1
4 R1
+ R2

12.24742
(4.89a) F = 1.0e3 · π · = 150000/4 · π = 117808.787N
4
(4.89b) a = 0.6301m vs 0.627m → 0.5% error
(4.89c) Pmax = 1.41641 × 105 Pa vs 1.435467 × 105 Pa → 1.3% error

Body E ν
Upper body 1 × 108 Pa 0.29
Lower body 1 × 106 Pa 0.29

Table 4.10: Parameters considered for 3D Hertz sphere-sphere problem

(a) Displacement solution (b) VM stress solution

Figure 4.60: Solution for two hemispheres Hertz contact

Page 144 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

4.5.5 Teeth model

The problem consists in a simplified model of a tooth with different types of layers, as the following: the first model with
enamel-composite (Figure 4.61a) and the second one formed of enamel-dentine-composite (Figure 4.61b). The idea
underlying this study is to compare the solutions between the two models and determine the benefits of the additional
layer of dentine as a reinforcement of the composed structure. The properties of each one of the layers of the model
appear on Table 4.11. It can be found here.

(a) Enamel-composite (b) Enamel-dentine-composite

Figure 4.61: Teeth layers model
Body E ν
Punch 2.069 × 1011 Pa 0.29
Enamel (color 3) 8 × 1010 Pa 0.3
Dentine (color 4) 2 × 1010 Pa 0.3
Composite (color 2) 1.03 × 1010 Pa 0.3

Table 4.11: Parameters considered for teeth model

(a) Enamel-composite displacement (b) Enamel-composite VM stress

(c) Enamel-dentine-composite displacement (d) Enamel-dentine-composite VM stress

Figure 4.62: Solution for teeth layers model

Figure 4.62 the solutions compared of both alternatives allows us to see the advantages of the additional layer of
dentine to the model.

Vicente Mataix Ferrándiz Page 145 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

4.5.6 Energy conservation

The following is an example to show the energy conservation in a frictionless contact simulation. In order to do so we
will present the geometry shown in Figure 4.63. In this pictures what we see is a ring on 2m outer diameter and an
inner diameter of 1.8m. Inside the ring we have a cylinder of 0.4m. The problem is a fully 3D simulation with 0.1m
thickness in the Z axis. In this problem the cylinder is only subjected to its own weight. The inner cylinder is softer
than the outer ring, which we can consider rigid from a practical point of view, see Table 4.12.

(a) Mesh seen from the front (b) Mesh seen on perspective
Figure 4.63: Energy conservation test. Cylinder inside ring. The example can be found in here

Body E ν ρ
Ring 2.069 × 1011 Pa 0.29 7850
Cylinder 2 × 108 Pa 0.29 1000

Table 4.12: Parameters considered for energy conservation test

With the given information, we can predict the exact movement of the body the total energy and the velocity in any
given time. For that we formulate the total energy as the total between the kinematic and the potential energy from
Equation (4.90a). In here h is the height of the cylinder in a given time, m the total mass of the cylinder , g the gravity
and v its velocity. If we define as only DOF the angle between the cylinder and the centre, being θ = 0 then h = 0.7m
and v = 0m/s as initial values. The total mass is as easy to calculate as m = 0.22 × π × 0.1 × 1000 = 4π kg. The total
energy is then, taking into account the initial BC previously mentioned then (4.90b). With this we can easily estimate
the maximum velocity as Equation (4.90c).

1
(4.90a) Etot = Ekin + Epot = hmg + mv 2
2
(4.90b) Etot = 0.7 × 4π × 9.81 = 86.2632J
r
2Etot
(4.90c) vmax = = 3.7m/s
m

Then taking as DOF θ we must express then the velocity in terms of θ̇, considering the radius of the given trajectory
as R = 0.7 then v = R θ̇. h can be represented with h = R(1 − sin(θ)). With this we can define the equation which
defines analytically our movement as (4.91a). Replacing the operations properly, we obtain the following expression

Page 146 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

(4.91b). Solving the corresponding ODE the expression for θ results on Equation (4.91c).

1
(4.91a) Etot = R(1 − sin(θ))mg + m(R θ̇)2
2
r
2g
(4.91b) θ̇ = sin(θ)
R
r r !! !!
1 1 2g 2g
(4.91c) θ= π − 4am −c1 +t − 2
2 4 R R

Where am is Jacobi elliptical function amplitude[OnlJac]. The constant c1 can be obtained replacing the BC
previously stated, giving us as result c1 = −0.990539. On Figure 4.64 we can see the comparison on the solution
obtained with the analytical reference. The error on the solution obtained is less than a 1%.

(a) Displacement evolution (b) Energy evolution

Figure 4.64: Evolution of displacement and energy compared with analytical solution

(a) t = 0.0s (b) t = 0.99s (c) t = 1.98s

(d) t = 2.97s (e) t = 3.96s

Figure 4.65: Solution of the problem at certains times. The time is specified in each subfigure

Vicente Mataix Ferrándiz Page 147 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

The Figure 4.65 shows the energy conservation in our simulation, and we can see as at the predicted time the
cylinder returns into its original position, then the loop starts over. The obtained maximum velocity coincides with the
predicted in (4.91c). Something that is relevant to remark is the fact as expected in a frictionless case, and therefore
there is no rotation movement, the contact point between the cylinder and the ring is always the same. Besides, the
energy is preserved despite the numerical dissipation due to Bossak scheme (see 2.4.4.2.Bossak algorithm).

4.5.7 Double arc benchmark

In this test case, we will consider the crushing of a hyperelastic bi-material half-ring on a hyperelastic base. This test
makes it possible to take into account large displacements, large deformations and large landslides coupled to contact
/ non-contact transitions. The reference solutions have been taken from G. Drouet[PhDDro15] and K.Poulios and
Y.Renard[ArtPR15a]. The problem geometry as well as the boundary conditions are sketched below on Figure 4.66,
both 2D and 3D. It can be found at the Examples repository.

(a) Geometry 2D (b) Geometry 3D

Figure 4.66: Double arc benchmark

(a) Hexahedra structured (b) Tetrahedra unstructured

Figure 4.67: Double arc meshes

Body E ν
First arc 3 × 108 Pa 0.32
Second arc 1 × 109 Pa 0.32
Support material 1 × 1011 Pa 0.3

Table 4.13: Parameters considered for double arc benchmark

We consider a hyperelastic half-ring composed of two materials (external 190mm and inner 170mm diameters)
and a hyperelastic base of length 250mm and height 50mm, see Figure. A vertical displacement of −90mm is
imposed at each end of the half-ring and the base of the base is recessed. Two variables of interest are considered
mainly: the displacement of the middle of the half-ring as a function of the loading pitch (60 steps of loading from the
contact is used, ie about 1.16mm per loading step) and the contact pressure. On the other hand, an hyperelastic
Neo-Hookean constitutive law is considered in both materials. The structure characteristic parameters are defined in

Page 148 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

Table 4.13. The time step is 0.0005seconds, while the total simulation time is 0.2seconds. With an increment of the
imposed displacement equal to 0.4t. For the frictional case, the friction coefficient considered is µ = 0.5.

4.5.7.1 Frictionless

The problem stated above has been solved using an unstructured mesh of tetrahedra and a structured mesh of
hexahedron. The resulting deformation for the former case can be seen in the Figure 4.68.

(a) Solution 1 (b) Solution 2

(c) Solution 3 (d) Solution 4

Figure 4.68: Displacement solution for frictionless in double arc benchmark

Figure 4.69: Compared solution for frictionless in double arc benchmark

Vicente Mataix Ferrándiz Page 149 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

We compare with the reference solution, Figure 4.69, where we have for the hexahedrical meshes a deviation
from the reference solution for the last steps, where we have a larger deformation and finally a stabilisation of the
deformation. For the unstructured meshes of the tetrahedron, the results present more differences, respect the
reference solution, higher deformation in general except for the last stages of the problem. In any case the results
obtained are in very good agreement with the reference solution.

4.5.7.2 Frictional

(a) Solution 1 (b) Solution 2

(c) Solution 3 (d) Solution 4

Figure 4.70: Displacement solution for frictional in double arc benchmark

Figure 4.71: Compared solution for frictional in double arc benchmark

The solution for the frictional case can be seen in the following images, Figure 4.70, and the compared solution
with the literature references (Figure 4.71), where the differences respect the reference are aligned with the ones

Page 150 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

already appreciated on the frictionless case. Besides, the effects of the friction can be noted in the way the arc
deforms in comparison of the frictionless case.

4.5.8 Arc pressing block

The problem consists in an arc and a block, in here the arc is pressed against the block. These two deformable solids
are considered as hyperelastic (Neo-Hookean), and the solid is formulated in UL framework. We consider three
cases, depending of the relationship of stiffness between the arc and the block, the first one with a rigid block, the
second one with a deformable block and the final one for deformable block with the arc being rigid. These properties
are specified on Table 4.14. The BC consists on a displacement imposed is uy = t, with t = [0.0, 1.0] for the two first
cases, t = [0.0, 1.775] for the rigid arc. The example can be found here.

Body Constitutive Law (CL) E ν

Arc (Rigid block) Neo-Hookean 68.96 × 108 Pa 0.32
Block (Rigid block) Neo-Hookean 68.96 × 107 Pa 0.32
Arc (Deformable block) Neo-Hookean 68.96 × 108 Pa 0.32
Block (Deformable block) Neo-Hookean 68.96 × 105 Pa 0.32
Arc (Deformable block-Rigid arc) Neo-Hookean 68.96 × 109 Pa 0.32
Block (Deformable block-Rigid arc) Neo-Hookean 68.96 × 105 Pa 0.32

Table 4.14: Parameters considered for arc pressing blocks

(a) Mesh 2D (b) Mesh 3D

Figure 4.72: Arc pressing block

Figure 4.73 summarises the solution obtained for each one of the material properties previously presented. It
can be seen the effects of the relative stiffness in each one of the cases. Particularly in the case where the block is
defined and the arc is rigid (Figure 4.73b), and the case where the block is rigid (Figure 4.73c) and all the deformation
lies in the arc.

(a) Deformable block (b) Rigid arc (c) Rigid block

Figure 4.73: Solution for different stiffness between the arc and the block

Vicente Mataix Ferrándiz Page 151 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

4.5.9 Hyperelastic tubes

The problem[ArtPR15a] consists in two cylin-
ders, Figure 4.74, with hyperelastic be-
haviour, as shown in the Table 4.15. We
impose a vertical displacement uz = −0.01t
in the upper cylinder, during a time interval of
t = [0.0, 4], in 100 steps. The two cylinders
deal with finite deformations and are defined
in a UL framework.

The two cylinders become into contact

and deal with self-contact in the inner ring of
the upper cylinder. For further details about
this example, the following link expands and
presents the necessary files.
Figure 4.74: Mesh for hyperelastic tubes
Body CL E ν
Upper cylinder Neo-Hookean 10000Pa 0.3
Lower cylinder Neo-Hookean 100000Pa 0.3

Table 4.15: Parameters considered for hyperelastic tubes

(a) t = 1s (b) t = 2s

(c) t = 3s (d) t = 4s
Figure 4.75: Displacement solution for hyperelastic tubes

Page 152 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

The Figure 4.75 presents the solution across different time steps, t = [0, 4]s. Additionally, Figure 4.76 shows a
split of the final configuration, t = 4s, so the deformation at the end of the simulation can be appreciated at Figure
4.76a and the VM stress at Figure 4.76b. In this configuration the phenomenon of self-contact arises.

(a) Displacement (b) VM stress

Figure 4.76: Slice solution for hyperelastic tubes

4.5.10 Contacting cylinders

The following case consists in two deformable hemicylinders (Figure 4.77) which experiment large deformations with
Neo-Hookean hyperelastic behaviour (Table 4.16). We applied two different movements in the base of the upper
cylinder, first a horizontal movement (4.5.10.1.Horizontal movement) and later a vertical movement (4.5.10.2.Vertical
movement). The relative position of the hemicylinders changes slightly between the two configurations. The problem
is studied again in the test cases from the adaptive refinement chapter, see 6.10.5.Contacting cylinders with adaptive
remeshing. These problems can be found here.

(a) Mesh 2D (b) Mesh 3D

Figure 4.77: Contacting cylinders mesh

Body CL E ν
Upper cylinder Neo-Hookean 2.1 × 1011 Pa 0.29
Lower cylinder Neo-Hookean 2.1 × 1011 Pa 0.29

Table 4.16: Parameters considered for contacting cylinders

4.5.10.1 Horizontal movement
A horizontal displacement is imposed ux = 0.2t, for a time interval of t = [0, 2.5]s in 1000 steps. Figure 4.78 shows the
displacement solution in different time steps, resulting in the reaction at the bottom of the lower hemicylinder from
Figure 4.79. The problem has been solved both considering frictionless (Figures 4.78a, 4.78b, 4.78c) and frictional
(µ = 1) (Figures 4.78d, 4.78e, 4.78f)cases. The difference of behaviour between the two cases is notorious.

Vicente Mataix Ferrándiz Page 153 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

(a) Frictionless t = 0.5s (b) Frictionless t = 1.0s (c) Frictionless t = 1.5s

(d) Frictional t = 0.5s (e) Frictional t = 1.0s (f) Frictional t = 1.5s

Figure 4.78: Solution for contacting cylinders with horizontal movement

Figure 4.79: Frictionless reaction solution in the lower cylinder for the horizontal movement

4.5.10.2 Vertical movement

(a) t = 0.5s (b) t = 1s

Figure 4.80: Solution for contacting cylinders with vertical movement

Page 154 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

For the vertical displacement is imposed uy = 0.1t, for a time interval of t = [0, 1]s in 200 steps. At Figure 4.80 the
deformation experimented is presented, as well as the reaction (Figure 4.81) in the base of the still hemicylinder.

Figure 4.81: Reaction solution in the lower cylinder for the vertical movement

4.5.11 Press fit

(a) Geometry

(b) Mesh 2D (c) Mesh 3D

Figure 4.82: Press fit problem

The problem consists of the numerical simulation of a press-fit process of a block in a channel according
to[ArtFW06; ArtDSB15]. The geometric and material parameters are shown in the Figures 4.82, where the symmetry
of the problem is used and only half of the domain is simulated. It is considered Neo-Hookean compressible
hyperelastic material for both components with the values from Table 4.17. With this problem, it is possible to evaluate
the behaviour of the contact element in a friction contact problem with large deformation and sliding.

Vicente Mataix Ferrándiz Page 155 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

Body E ν µ
Die 68.96 × 108 Pa 0.32 0.1
Block 68.96 × 107 Pa 0.32 0.1

Table 4.17: Parameters considered for press fit

Figure 4.83: Solution for 2D press fit compared with reference

The process is modelled by applying a non-homogeneous boundary condition u = 1000mm in the left face of the
block. The height of the block is greater than the channel, imposing to the problem an initial penetration ∆initial = 1mm
and consequently an initial contact stress. The first time step uses u = 0mm and the program generate the normal
stress necessary to the non-penetration condition, separating the bodies in contact. After this step, we employed a
non-homogeneous boundary condition u = 1000mm. It is considered plane strain state and UL formulation.

Additionally a 3D setup is also evaluated, extruding 250mm, but this one cannot be compared with any reference.
This can be found in together with the other case studies shown. Comparing the results with the reference, we got
slightly higher values than in the reference as seen in Figure 4.83. In the other hand, the evolution in the solution is
presented on the Figures 4.84.

(a) Solution t = 0.28s (b) Solution t = 0.56s (c) Solution t = 0.84s

(d) Solution t = 1.12s (e) Solution t = 1.4s

Figure 4.84: Press fit 2D solution

For the 3D case, where we do not have a reference solution to compare, therefore we can see the deformation
evolution in the first five figures from Figure 4.85. The reaction in the base support most western points is shown in
the last image of the set, the Figure 4.85f.

Page 156 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

(a) Solution t = 0.28s (b) Solution t = 0.56s (c) Solution t = 0.84s

(d) Solution t = 1.12s (e) Solution t = 1.4s (f) Horizontal reaction in the base for 3D case
Figure 4.85: Press fit 3D solution

4.5.12 Ironing punch

The following are two examples extensively considered in the literature[ArtFW06; ArtHar+09; ArtPR15b], which
are the ironing tests. These tests consist in moving die along a solid block, undergoing large deformation in the
process.

1. This case cannot be found in literature, and it is a little bit more challenging as the curvature of the contacting
die is higher than in the previous case, we will call this case as circular ironing (4.5.12.1.Circular ironing).

2. We will show the results obtained with the standard ironing test, called shallow ironing, which is the one
commonly referred on the literature (4.5.12.2.Shallow ironing). In this second case, we will consider both a
frictionless and a frictional case.

4.5.12.1 Circular ironing

In this case we will run the simulation just considering the frictionless case as our interest is to compare the behaviour
between the two different geometries. The geometry of the problem is fully defined in the Figure 4.86a, and the
hexahedral structured mesh considered in Figures 4.86b and 4.86c. The properties of the materials are identical for
the shallow ironing problem, with a Neo-Hookean material behaviour (5.4.2.Neo-Hookean material). We impose a
vertical displacement as uy = −t during the time interval t = [0, 1]s and ux = t − 1 at t = [1, 10]s keeping the previous
vertical displacement. The example is found in the repository.

Body E ν
Die 68.96 × 108 Pa 0.32
Block 68.96 × 107 Pa 0.32

Table 4.18: Parameters considered for circular ironing

Vicente Mataix Ferrándiz Page 157 of 374
4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

(a) Geometry of the problem

(b) Mesh (c) Perspective of the Mesh

Figure 4.86: Circular ironing test

(a) t = 2.5s (b) t = 5s

(c) t = 7.5s (d) t = 10s

Figure 4.87: Circular ironing test solution

Figures 4.87 present the deformation evolution for this case in the interval going from t = [0, 10]s. The Figure 4.88
shows the different solution obtained with respect the shallow ironing example.

Page 158 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.5 Numerical examples

Figure 4.88: Comparison between the frictionless solutions of the shallow and the circular ironing

4.5.12.2 Shallow ironing

(a) Mesh seen from the front (b) Mesh seen on perspective
Figure 4.89: Shallow ironing test
On this case an indenter or die with a circular arc shaped bottom edge is pressed against a rectangular block
and is forced to slide along the block length, see Figure 4.89 where the geometry and mesh considered can be
appreciated. The contacting bodies exhibit a Neo-Hookean material behaviour (5.4.2.Neo-Hookean material), with the
parameters from Table 4.19. The problem solved on this case is fully 3D with a deep of 1m, but in many references
can be found as a 2D case with plane strain hypothesis. The example can be found in here.

Body CL E ν µ
Die Neo-Hookean 68.96 × 108 Pa 0.32 0.3
Block Neo-Hookean 68.96 × 107 Pa 0.32 0.3

Table 4.19: Parameters considered for shallow ironing

Although the performed simulation is quasi-static, the load steps are defined as a function of time for the sake
of presentation of the results. From t = [0, 1]s, the indenter is moved vertically towards the block until the vertical
displacement reaches 1m. From t = [1, 12]s, the die is displaced horizontally by a total distance of 11m. This
loading differs sometimes in the literature, because of that the results obtained must the translated to the equivalent
displacement evolution process.

The Figures 4.90 compare the solution to the shallow ironing test between the friction and the frictional solution at
different times. We can appreciate that the frictional case indeed opposes the movement of the die. Figures 4.91

Vicente Mataix Ferrándiz Page 159 of 374

4.5 Numerical examples CHAPTER 4. CONTACT MECHANICS

compares the solution obtained with the reference[ArtPR15b]. In general the agreement is good with [ArtPR15b], and
then the solution can be considered as valid.

(a) Frictionless t = 2.5s (b) Frictionless t = 5s

(c) Frictional t = 2.5s (d) Frictional t = 5s

Figure 4.90: Shallow ironing test solution, comparing frictional and frictionless cases

(a) Frictionless solution (b) Frictional solution

Figure 4.91: Reaction solution for shallow ironing test

Page 160 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

4.6 Derivatives for contact mechanics linearization

On this section we can find in extension the definition of the directional derivatives, also called Gateaux deriva-
tives[BookBel+14], associated with the mortar operators (4.92) and the slip and gap (4.93). These derivatives will
be noted with ∆. With the components shown on (4.92), in addition to the derivatives of the normal and tangent
(4.93), it is possible to define consistently the proper consistent linearisation of the tangent operator for the contact
problem. The derivatives of the normal and tangent must be taken into account in order to define the derivatives of
the gap and the slip. Similar developments can be found on other works, as Popp[PhDPop12; ArtPGW09; ArtPop+10],
Yang[ArtYLM], Hammer [PhDHam13] and Cavalieri[ArtCFC12; ArtCC15; ArtCC13b] work.
First the definitions concerning the 2D problem will be presented (4.6.1.Derivatives for 2D contact), later the
concerning to the 3D problem (4.6.2.Derivatives for 3D contact). The presenting work will focus particularly on the
linear shape functions, as long as or work focuses on linear geometries.

ngp ngp ngp

X X X
1 1 1 1 1
(4.92a) ∆D[j, k] = wg ∆φgj Ngk Jg + wg φgj ∆Ngk Jg + wg φgj Ngk ∆Jg1
g=1 g=1 g=1

ngp ngp ngp

X X X
2 1 2 1 2
(4.92b) ∆M[j, l] = wg ∆φgj Ngk Jg + wg φgj ∆Ngk Jg + wg φgj Ngk ∆Jg1
g=1 g=1 g=1

Expressing the computation of the weighted gap in an algebraic way, based on the mortar operators definition.

(4.93a) g̃i = ni · (Dn x1 − Mn x2 )i

Being i the node index and n the current time step, then

(4.93b) ∆g̃i = ∆ni · (Dn x1 − Mn x2 )i + ni · (∆Dn x1 − ∆Mn x2 + Dn ∆x1 − Mn ∆x2 )i

In the same way, expressing the slip on an algebraic way, the derivative of the slip would be:

(4.93c) ũτ i = τ i · ((Dn − Dn−1 ) x1 − (Mn − Mn−1 ) x2 )i

Being i the node index and n, n − 1 the current and previous time steps, then

∆ũτ i = ∆τ i · ((Dn − Dn−1 ) x1 − (Mn − Mn−1 ) x2 )i

(4.93d) + τ i · ((∆Dn − Dn−1 ) x1 − (∆Mn − Mn−1 ) x2 )i
+ τ i · ((Dn − Dn−1 ) ∆x1 − (Mn − Mn−1 ) ∆x2 )i

After the definition of the following directional derivatives (∆), it is necessary it is algebraic representation of
them on an equivalent matrix-vector format. This process is quite standard for FE formulations, and in our case it is
done in an automatic way using the AD, for more details we address to the corresponding Appendix C.Automatic
differentiation. The code for the tests for the quadratic convergence shown on the following sections are accessible
in the public repository of Kratos, implemented on Kratos C++ unittest format. The following link guides toward
it.

4.6.1 Derivatives for 2D contact

In this section the derivatives concerning 2D linear lines, necessaries in order to properly compute the contact are
presented, see (4.92) and (4.93). The derivatives are tested using a set of pairs following the Figure 4.92 structure,
the Table 4.20 contains the coordinates of the nodes of the actual configurations tested. The perturbations applied on
each pair are defined on Table 4.21, which lists the nodes perturbed and its amplitude.

Vicente Mataix Ferrándiz Page 161 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

Figure 4.92: Line validation geometries

Pair  P11   P2 1
 P12    P22 
1 −1.0 0.0 0.0  1.0 0.0 0.0
 1.2 0.0−3 0.0   −0.8 0.0−3 0.0 
2 −1.0 0.0 0.0  1.0 0.0 −30.0   1.2 1.0E −3 0.0  −0.8 1.0E −4 0.0
3 −1.0 0.0 0.0 −1.0 5.0E 0.0 −1.0 1.0E 0.0 −1.0 7.0E 0.0
Table 4.20: 2D lines pairs geometrical configuration

Case Node perturbed Perturbation amplitude

1 1 −5.0e − 2
2 1 −5.0e − 2
3 1 −1.0e − 1
Table 4.21: 2D lines pertubartion amplitudes

4.6.1.1 Jacobians

4.6.1.1.1 Theory :

As the integration of the mortar formulation is performed only in the slave (mortar) domain, then the derivative of
the Jacobian is defined only on the slave domain. Taking into consideration the definition of the Jacobian as (4.94a), it
is possible to define the directional derivative of the slave Jacobian determinant at each Gauss point in a 2D linear
line as (4.94b). In order to simplify and due to its extended use, we will call det(J) as J.

n
X
(4.94a) J (ξ ) = Nk,ξ (ξ ) xk
k=1

Pn
n
!

Nk,ξ ξg1 xk X

(4.94b) ∆J ξg1 = Pnk=1
· Nk,ξ ξg1 ∆xk
k=1 Nk,ξ ξg1 xk k=1

4.6.1.1.2 Convergence study :

Finally, in the Figure 4.93, the convergence plot of the Jacobian derivatives for the 3 different geometrical pairs
from Table 4.20 is shown. In this figure, and the followings, L2 represents the Euclidean norm of the error, difference
between the estimated value and the exact one. The plot is created considering in the x-axis the amplitude of the
perturbation, and in the y the logarithm of L2 of the error; therefore the slope indicates the convergence rate of the
error in function of the perturbation.

Page 162 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

Figure 4.93: Convergence plot for the Jacobian derivatives for the 2D linear line

4.6.1.2 Shape functions

4.6.1.2.1 Theory :
The shape functions of both domains depend on the integration segments between the two domains on contact. In
order to compute the shape functions of a 2D line, we do as (4.95a), Figure 4.94, where ξ corresponds with the local
coordinate of the line. With this, its derivatives depend only on the derivative of the local coordinate ξ (4.95b). Despite
the fact that lines in Figure 4.94 are not in contact, due to its relative position, will be affected by the perturbations,
allowing to measure the derivative convergence.

   1

N1 (1 − ξ )
(4.95a) = 2
1
N2 2
(1 + ξ )

   
∆N1 − 12 ∆ξ
(4.95b) = 1
∆N2 2
∆ξ

Figure 4.94: Shape functions for the 2D linear line

For this reason we need to compute on first place the derivatives of the local coordinates, which depends on the
derivatives of the integration segments. We will separate in two small sections those derivations. The derivation of
these terms is based on the work of Laursen and Yang[ArtYLM] and Popp[PhDPop12].

Vicente Mataix Ferrándiz Page 163 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

4.6.1.2.2 Integration segments :

Figure 4.95: Integration segment for a linear line

The derivation of the integration points require the derivation of the integration segments. As previously presented
this corresponds with the current and effective section of the line on contact, see Figure 4.95. As can be seen on this
picture we have two possible scenarios, one where the projected slave node (ξa1 ) corresponds with the original slave
node and the second case corresponds with the counterpart scenario for the master side (ξa2 ). The same logic is
applied for the end integration segment coordinates (ξb1 and ξb2 ). If the integrated node corresponds with the original
geometry the derivative therefore will be 0.

The derivative ∆ for this first scenario is presented on (4.96), where the sub-index a for the coordinates (x1a ) and
normal (na ) represent the first node on the slave side. On this case (4.96b) the shape functions and shape function
gradients are computed on ξa2 .

(4.96a) ∆ξa1 = 0

1
∆ξa2 = − Pnc  P c
nm
 ·
2 2 y
l=1 Nl,ξ xl
m
na− l=1 Nl,2ξ yl2 nax
 c
  c

nm nm
X
X

(4.96b)
 Nl2 ∆xl 2
− ∆xa1  nay − Nl2 ∆yl 2
− ∆ya1  nax
l=1 l=1
 c   c  
nm nm
X
y
X

+ Nl2 xl2 − xa1  ∆na −  Nl2 yl2 − ya1  ∆nax 
l=1 l=1

On the second case (4.97), as the slave side, which is our reference domain, does not correspond with its
projection the n considered then is the one interpolated for the projected point. This means that the interpolation
of the nodal normal will be computed. On this case (4.97a) the shape functions and shape function gradients are
computed on ξa1 .

num
(4.97a) ∆ξa1 =
denom

 c
 c
  c
 c

ns ns ns ns
X X y
X X
1 1 
denom = −  Nk,ξ xk Nk1 nk  +  1
Nk, 1 
ξ yk Nk1 nkx 
k=1 k=1 k=1 k=1
(4.97b)      
nsc nsc nsc c
ns
X
X y
X
X
− Nk1 xk − xa2  
1 1
Nk, ξ nk + Nk1 yk − ya2  
1 1
Nk, x
ξ nk
k=1 k=1 k=1 k=1

Page 164 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

 c
 c
  c
 c

ns ns ns ns
X
X y
X
X
num =  Nk1 ∆xk − ∆xa2  
1
Nk1 nk  −  Nk1 ∆yk − ∆ya2  
1
Nk1 nkx 
k=1 k=1 k=1 k =1
(4.97c)      
nsc nsc nsc nsc
X
X y
X
X
+ Nk1 xk − xa2  
1
Nk1 ∆nk  −  Nk1 yk − ya2  
1
Nk1 ∆nkx 
k=1 k=1 k=1 k=1

(4.97d) ∆ξa2 = 0

4.6.1.2.3 Local coordinates (Gauss points) :

With the values obtained in the previous expression, it is possible to evaluate the ∆ of the Gauss Point (GP)
coordinates, and therefore calculate the values for the shape function derivatives. We evaluate those derivatives both
on the slave side (ξg1 , (4.98a)) and the master side (ξg2 , (4.98b)).

For the slave side, we just need to apply the definition for the GP and derive it. As it consists on a simple sum, the
derivative is straightforward.

1
1

(4.98a) ∆ξg1 = 1 − ξg ∆ξa1 + 1 + ξg ∆ξb1
2 2

For the master side, as it depends on a projection, the linearisation requires to take that projection into consideration,
resulting the following expression. All the shape functions and shape function gradients are computed on ξg2 .

∆ξg2 = − Pnmc 2 2  y 1 Pnmc 2 2  x

k=1 Nk,ξ xk ng − k =1 Nk ,ξ yk ng
hP c  P c 
nm 2 2

1 y nm 2 2

1
(4.98b) · N ∆ x − ∆ x ng − N ∆ y − ∆ y nx
P k=1
c
k k
 g P c k=1 k
k  g i g
nm 2 2

1 y nm 2 2 1 x
+ k=1 Nk xk − xg ∆ng − k=1 Nk yk − yg ∆ng

4.6.1.2.4 Convergence study :

Finally, in the Figure 4.96, the convergence plot of the shape function derivatives for the 3 different geometrical
pairs from Table 4.20 is shown.

Figure 4.96: Convergence plot for the shape function derivatives for the 2D linear line

Vicente Mataix Ferrándiz Page 165 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

4.6.1.3 Dual shape functions

4.6.1.3.1 Theory :
On standard FEM shape functions defined on the reference configuration, by definition these are not dependant of
the deformation of the domain. This is not the case for dual shape functions, this is due to its intrinsic dependence on
the J of the slave side, see (4.99a). This means that our dual shape functions are deformation-dependant. In addition
we shall consider its dependence with the standard shape function derivatives. All this together give us the following
expression (4.99b). The Ae definition is presented in the corresponding section, 4.3.3.4.1.Dual Lagrange multipliers,
on the Equation (4.24).

(
Ae (N, J)
(4.99a) Φ = Ae N 1
N(N)

(4.99b) ∆Φ = Ae ∆N 1 + ∆Ae N 1

So in addition to the terms from the 4.6.1.2.Shape functions we need to evaluate the corresponding terms
associated to Ae (4.100).

1 −1
∆Ae = ∆De M−
e − De ∆Me Me
ngp
1 1 X
∆De = ∆[djk ] ∈ Rme ×me , ∆djk = δjk 1
wg Ngk ∆Jg1
(4.100) g=1
ngp ngp
1 1 X X
∆Me = ∆[mjk ] ∈ Rme ×me , ∆mjk = wg wg Ngj1 Ngk
1
∆Jg1
g=1 g=1

4.6.1.3.2 Convergence study :

Finally, in the Figure 4.97, the convergence plot of the dual shape function derivatives for the 3 different geometrical
pairs from Table 4.20 is shown.

Figure 4.97: Convergence plot for the dual shape function derivatives for the 2D linear line

Page 166 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

4.6.1.4 Normal and tangent vectors

4.6.1.4.1 Theory :
The linearisation of the tangent vectors is associated, as its definition depends on, on the linearisation of the normal
vector. On these lines only the definition of the normal derivative, it is included as the definition of the tangent derivative
is defined as the cross product between the normal derivative and the out of plane vector (∆τ = vz × ∆n).

Figure 4.98: Normal average for a 2D line. Proposed by Taylor and Papadopoulos[ArtPT92]
The definition of the normal vector (n) for a 2D line is defined on (4.101a), where xi , yi correspond with the
coordinates of the i node of the line (1, 2). The unit normal corresponding to (4.101a) can be obtained with
(4.101b).
Then the average normal[ArtPT92] (n̄) is defined as (4.101c), this definition is general and can be extended to 3D.
Figure 4.98 represents the normal average for a 2D line. It is relevant to mention that the average normal (4.101c)
is already computed over unitary normals, in contrast with the average area normal proposed by Popp[PhDPop12],
this is done due to practical reasons, as this approach was significantly more robust (better convergence) in several
numeric studies, particularly when coarse meshes were considered.

   
y2 y
(4.101a) narea = − 1
x1 x2

narea
(4.101b) n=
knarea k

In order to obtain the average normal we sum the normals of the neighbours entities (n) to the node, then we make
unitary dividing by its norm.
Pnjneigh
c=1 nc
(4.101c) n̄ =
Pnjneigh
c=1 nc

So, once defined the normal (4.102) and its average value (4.103) we can define its directional derivative.

∆narea knarea k − narea ∆ knarea k

(4.102a) ∆n = 2
knarea k

For the 2D line the resulting expression for the narea derivative will be (4.102b), which corresponds with taken the
corresponding unitary vector depending of the derivative evaluated.
   
y2 y
(4.102b) ∆narea = ∆ −∆ 1
x1 x2

Vicente Mataix Ferrándiz Page 167 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

For the directional derivative of the average normal, the resulting expression is decomposed in its different
components.

Pnjneigh Pnjneigh Pnjneigh Pnjneigh

∆ c=1 nc c=1 nc − c=1 nc ∆ c=1 nc
(4.103a) ∆n̄ = 2
Pnjneigh
c=1 nc

The first components are easy to obtain, as the derivative of a sum is the sum of its derivatives.
neigh neigh
nj nj
X X
c
(4.103b) ∆ n = ∆ nc
c=1 c=1

The second components require to calculate the derivative of a norm, proceeding the standard manner the expression
obtained is (4.103c).

neigh neigh
nj Pnjneigh Pnjneigh nj
X nc ∆ nc X
(4.103c) ∆ nc = c=1 c=1
= n̄∆ nc
Pnjneigh
c=1
c=1 nc c=1

Additionally in order to evaluate the value of the normal derivative on a specific GP, calculated on the corresponding
local coordinates ξg1 . The procedure to obtain is straightforward as only requires to apply the chain rule on the shape
function interpolation of the normal evaluated on the GP (4.104).

c c
ns ns
X
X

1 1 1
(4.104) ∆ng = Nk, ξ ξg ∆ξg nk + Nk1 ξg1 ∆nk
k=1 k=1

4.6.1.4.2 Convergence study :

Finally, in the Figure 4.99, the convergence plot of the normal derivatives for the 3 different geometrical pairs from
Table 4.20 is shown.

Figure 4.99: Convergence plot for the normal derivatives for the 2D linear line

Page 168 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

4.6.2 Derivatives for 3D contact

In this section the derivatives concerning 3D linear triangles and bilinear quadrilaterals, as part of (4.92) and (4.93)
are presented. The main difference between this approach and the one provided for 2D cases is the segmentation of
the surfaces, where the actual integration is done. The derivatives are tested using a set of pairs following the Figure
4.100a structure and the Table 4.22 for the case of the triangle. Again, the perturbations applied on each pair are
defined on Table 4.23.

(a) Triangle validation geometries (b) Quadrilateral validation geometries

Pair  P11   P21   P31   P12   P22   P32 

1  0.0 0.0 0.0 −3  1.0 0.0 0.0 0.0 1.0 0.0  0.0 1.0 1.0E −3  0.0 0.0 1.0E −3  1.0 0.0 1.0E −3 
−3 −3 −3
2 −0.1 0.1 1.0E −3  1.1 0.2 0.0 0.1 1.0 0.0 −0.1 1.3 1.0E  0.1 0.2 1.0E  1.2 0.2 2.0E 
3 −0.1
 0.1 1.0E 1.1 0.2 0.0 0.1 1.0 0.0 −0.1 1.3 1.0E −3 0.1 0.2 1.0E −3  1.2 0.2 2.0E −3 
4  0.0 0.0 0.0 −3  1.0 0.0 0.0 0.0 1.0 0.0  0.0 1.0 1.0E −3  0.0 0.0 1.0E −3  1.0 0.0 1.0E −3 
5 −0.1
 0.1 1.0E 1.1 0.2 0.0 0.1 1.0 0.0 −0.1 1.3 1.0E −3  0.1 0.2 1.0E −3  1.2 0.2 2.0E −3 
6 0.0 0.0 0.0 1.0 0.0 0.0 0.0 1.0 0.0 −0.1 1.0 1.0E −3 0.0 0.0 1.0E −3 1.0 0.0 1.0E −3
Table 4.22: 3D triangles pairs geometrical configuration
Case Node perturbed (I) Perturbation amplitude (I) Node perturbed (II) Perturbation amplitude (II)
1 4 -5.0e-1 - -
2 4 -5.0e-3 - -
3 4 1.0e-1 5 1.0e-1
4 4 1.0e-1 5 5.0e-2
5 1 5.0e-2 - -
6 4 -5.0e-2 - -
Table 4.23: 3D triangles pertubartion amplitudes

The derivative tests for quadrilateral pairs are defined following Figure 4.100b and Table 4.24. The Table 4.25
defines the perturbations.
Pair  P11   P21   P3 1
  P4 1
 
2
P1   P2 2
  P3 2
  P4 2

−3 −3 −3 −3 −3 −3
1 0.0
 0.2 1.0E  1.0
 0.2 1.0E  1.1 1.1 0.0 0.2 1.0 0.0 −0.1 1.0 1.0E −3  1.0 1.1 1.0E −3  1.0 0.1 2.0E −3  0.0 0.1 2.0E −3 
2  0.0 0.0 0.0   1.0 0.0 0.0   1.0 1.0 0.0  0.0 1.0 0.0  − 0.1 1.0 1.0E   1.0 1.0 1.0E   1.0 0.0 1.0E   0.0 0.0 1.0E 
3 0.0 0.3 2.0E −3 1.0 0.2 1.0E −3 1.2 1.1 0.0 0.2 1.1 0.0 −0.1 1.0 2.0E −3 1.2 1.1 −3
2.0E 1.0 0.1 −3
3.0E 0.1 0.1 −3
3.0E
Table 4.24: 3D quadrilaterals pairs geometrical configuration
Case Node perturbed Perturbation amplitude
1 5 -5.0e-3
2 5 -5.0e-3
3 5 -5.0e-3
Table 4.25: 3D quadrilaterals pertubartion amplitudes

Vicente Mataix Ferrándiz Page 169 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

4.6.2.1 Jacobians
4.6.2.1.1 Theory :

4.6.2.1.1.1 Integration segments derivatives :

Before introduce the theory that defines the directional derivatives of the J, we need to introduce the derivatives of
the segments obtained during the integration. The resulting operation is presented on the Figure 4.100, and detailed
on Figure 4.101. For more details about the method followed in order to evaluate the mortar integration, we address
the corresponding section A.2.Mortar segmentation, where we debate the two different integration methods studied,
and reason why we decided to consider a segment base approach.
The Figure 4.100 just represents the intersection and clipping technique, where the intersected points generate
the triangles considered during the integration. The later, Figure 4.101, shows in deeper detail how this clipping
technique works. The relevant part on this method is the consideration of the clipping coordinates, the resulting point
coordinates (xclip ) depend on the coordinates of the points from the slave segment (x̂11 , x̂12 ) and master segment (x̂21 ,
x̂22 ). This means that in order to compute properly the derivatives it is necessary to identify the corresponding nodes
on the master and slave geometries when computing the clipping algorithm[BookSE02; BookHug+14].

Figure 4.100: Intersection and clipping procedure during mortar segmentation

Figure 4.101: Detail on intersection on mortar segmentation

We need to differentiate two different scenarios when obtaining the intersections:

• The point of the integration triangle is part of the original master/slave geometry. This case is less complex to
compute, as it takes directly the components of the master/slave side derivatives without direct dependence on
the counterpart domain.

Page 170 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

• The point which integrates the integration triangle belongs to an intersection (most general case). All points
show on Figure 4.101 belong to this category.

For the first case, we need to differentiate if the node is a slave node (4.105a) or a master one (4.105b). Then its
derivatives are computed respectively according to (4.106), (4.106a) for the slave and (4.106b) for the master.



(4.105a) xclip = x1 − x1 − x1plane · nplane nplane


(4.105b) xclip = x2 − x2 − x1plane · nplane nplane




∆xclip = ∆ x1 − x1 − x1plane · nplane nplane
(4.106a) 

 

= ∆x1 − ∆x1 − ∆x1plane · nplane + x1 − x1plane · ∆nplane nplane − x1 − x1plane · nplane ∆nplane



∆xclip = ∆ x2 − x2 − x1plane · nplane nplane
(4.106b) 

 

= ∆x2 − ∆x2 − ∆x1plane · nplane + x2 − x1plane · ∆nplane nplane − x2 − x1plane · nplane ∆nplane

For the intersection case, we need to define the clip algorithm in order to obtain the corresponding coordinates.
We will consider the Foley [BookHug+14] clipping algorithm can be defined as follows (4.107), an alternative approach
can be to project x̂11 , x̂12 , x̂21 and x̂22 into the auxiliary plane defined by nplane prior to compute the clipping coordinates.
The corresponding derivative, originally deduced by Puso and Laursen[ArtPL04], can be found in (4.108), where we
need to decompose into its different components due to long expression deduced.

x̂11 − x̂21 × x̂22 − x̂21 · nplane

(4.107) xclip = x̂11 −

x̂12 − x̂11
x̂12 − x̂1 × x̂22 − x̂1 · nplane
1 2

!
x̂11 − x̂21 × x̂22 − x̂21 · nplane

∆xclip = ∆x̂11 −∆

x̂12 − x̂11
x̂12 − x̂1 × x̂22 − x̂1 · nplane
1 2
(4.108a)
num
∆num · denom − num · ∆denom 1

= ∆x̂11 − ∆x̂12 − ∆x̂11 − x̂2 − x̂11
denom denom2
The second term is quite complex, that’s why we have decomposed into different components. The expression for the
numerator corresponds with (4.108d), and (4.108e) for the denominator.



(4.108b) num = x̂11 − x̂21 × x̂22 − x̂21 · nplane



(4.108c) denom = x̂12 − x̂11 × x̂22 − x̂21 · nplane

Then the respective derivatives are defined as:

∆num = ∆ x̂11 − x̂21 × x̂22 − x̂21 · nplane
(4.108d)








= x̂11 − x̂21 × x̂22 − x̂21
· ∆nplane + ∆x̂11 − ∆x̂21 × x̂22 − x̂21 + x̂11 − x̂21 × ∆x̂22 − ∆x̂21 · nplane




∆denom = ∆ x̂12 − x̂11 × x̂22 − x̂21 · nplane
(4.108e)








= x̂12 − x̂11 × x̂22 − x̂21 · ∆nplane + ∆x̂12 − ∆x̂11 × x̂22 − x̂21 + x̂12 − x̂11 × ∆x̂22 − ∆x̂21 · nplane

4.6.2.1.1.2 Jacobian derivatives :

Once obtained the integration cell vertex derivatives we can define the corresponding derivative of J. As the
integration occurs on the cell vertex the proper Jacobian to consider is the J on the integration cell, Jclip , see (4.109a).

Vicente Mataix Ferrándiz Page 171 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

Considering the three vertices of the integration triangle as x1clip , x2clip and x3clip , and its respective derivatives, then
derivative of Jclip can be calculated using (4.109b).

(4.109a) Jclip = x2clip − x1clip × x3clip − x1clip



∆Jclip = ∆ x2clip − x1clip × x3clip − x1clip
   
x2clip − x1clip × x3clip − x1clip 


=     · ∆x2clip − ∆x1clip × x3clip − x1clip
2 1 3 1
(4.109b) xclip − xclip × xclip − xclip
   
x2clip − x1clip × x3clip − x1clip 


+     · x2clip − x1clip × ∆x3clip − ∆x1clip
x2clip − x1clip × x3clip − x1clip

4.6.2.1.2 Convergence study :

Figure 4.102: Convergence plot for the Jacobian derivatives for the 3D linear triangle
The convergence plot of the Jacobian derivatives for the 6 different geometrical pairs of the 3D linear triangles
from Table 4.22 can be found in the Figure 4.102.

Figure 4.103: Convergence plot for the Jacobian derivatives for the 3D bilinear quadrilateral

Page 172 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

The convergence plot of the Jacobian derivatives for the 3 different geometrical pairs of the 3D bilinear quadrilater-
als from Table 4.24 can be found in the Figure 4.103.

4.6.2.2 Shape functions

4.6.2.2.1 Theory :
The same reasoning applied on the 2D case 4.6.1.2.Shape functions is applied here. This means we need to
define first the shape functions and its derivatives. Those shape functions depend on its local coordinates (ξ , η ), then
the derivatives depend on the derivatives of the local coordinates. Those derivatives will be obtained in a independent
subsection. First, we define the shape functions of the 3D linear triangle (4.110a), and its derivatives (4.110b). On
the Figure 4.104 we represent on 3D the shape functions according to ξ and η .

   
N1 1−ξ−η
(4.110a)  N2  =  ξ 
N3 η
   
∆N1 −∆ξ −∆η
(4.110b)  ∆N2  =  ∆ξ 0 
∆N3 0 ∆η

(a) N1 (b) N2

(c) N3
Figure 4.104: Shape functions for the 3D linear triangle

Vicente Mataix Ferrándiz Page 173 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

Second, we define the shape functions of the 3D bilinear quadrilateral and its derivatives at (4.111). The graphic
representation of the shape functions are shown in the Figure 4.105.

   1

N1 4
((1 − ξ )(1 − η ))
1
 N2   ((1 + ξ )(1 − η )) 
(4.111a)  N3  = 
   4
1

4
((1 + ξ )(1 + η )) 
1
N4 4
((1 − ξ )(1 + η ))
   ∆ξ 
∆N1 − 4 − ∆η
4
 ∆N2   ∆ξ − ∆η 
(4.111b)  ∆N3  =  ∆ξ
   4 4 
∆η 
4 4
∆N4 − ∆ξ ∆η
4 4

(a) N1 (b) N2

(c) N3 (d) N4
Figure 4.105: Shape functions for the 3D bilinear quadrilateral

4.6.2.2.2 Local coordinates (Gauss points) :

Here we will define the derivatives of the corresponding GP local coordinates of the integration triangles, Figure
4.101. Important to highlight the fact that the complexity of this task came from the fact that the inner GP from the
integration triangles must be projected into the master and slave domains. This, in the most general case requires the
resolution of a NL problem with a NR iterative process.

Page 174 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

To obtain the local coordinates ξ and η we will consider the expression proposed by Popp[ArtPGW09], but
simplified as our geometries are linear and consequently reduces the complexity of the system of equations. In order
to obtain the coordinates of the GP we use (4.112a), where we can see as the coordinates of the GP in the integration
triangle (represented with N̄) has an equivalent on the master/slave geometries. With this expression we can define
this as a residual (4.112b) of a linear system of equations, which converges in one NL iteration. The complete system
of equations (4.112c) require to compute the corresponding Jacobian (J) (4.112d).

Pnsc P3
x1g = 1 1
k =1 Nk xk = N̄i xclip
(4.112a) Pnmc 2 2 P3i=1
x2g = l=1 Nl xl = i=1 N̄i xclip

RHS1 = x1g
(4.112b)
RHS2 = x2g

 
ξi
(4.112c) LHSi = RHSi ,i = 1, 2
ηi

Where LHSi corresponds with the Jacobian:

" Pn c #
i i
k=1 Nk,ξ xk
s

(4.112d) LHSi = J i = Pnsc ,i = 1, 2

i i
k=1 Nk,η xk

With this we can define the equivalent expressions for the derivatives of the local coordinates (ξ and η ), we
redefine the RHS (4.113a), and using the same LHS (J) (4.112d), we can solve (4.113b) the new system of equations
in order to obtain our target derivatives.

P3
RHS1 = ∆x1g = i=1 N̄i ∆xclip
(4.113a) P3
RHS2 = ∆x2g = i=1 N̄i ∆xclip

 
∆ξi 1
(4.113b) = LHS−
i RHSi ,i = 1, 2
∆ηi

4.6.2.2.3 Convergence study :

Figure 4.106: Convergence plot for the shape function derivatives for the 3D linear triangle

Vicente Mataix Ferrándiz Page 175 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

The convergence plot of the shape function derivatives for the 6 different geometrical pairs of the 3D linear
triangles from Table 4.22 can be found in the Figure 4.106.

Figure 4.107: Convergence plot for the shape function derivatives for the 3D bilinear quadrilateral

The convergence plot of the shape function derivatives for the 3 different geometrical pairs of the 3D bilinear
quadrilaterals from Table 4.24 can be found in the Figure 4.107.

4.6.2.3 Dual shape functions

4.6.2.3.1 Theory :

The procedure to follow is identical to the one presented for the 2D case in 4.6.1.3.Dual shape functions. No
additional modification is needed, except, of course, the corresponding computation on the integration triangles, which
means of the Jclip (4.6.2.1.Jacobians) and the use of the shape functions and derivatives already presented on the
previous section 4.6.2.2.Shape functions. Then we proceed with the convergence study.

4.6.2.3.2 Convergence study :

The convergence plot of the dual shape function derivatives for the 6 different geometrical pairs of the 3D linear
triangles from Table 4.22 can be found in the Figure 4.108.

Figure 4.108: Convergence plot for the dual shape function derivatives for the 3D linear triangle

Page 176 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

The convergence plot of the dual shape function derivatives for the 3 different geometrical pairs of the 3D bilinear
quadrilaterals from Table 4.24 can be found in the Figure 4.109.

Figure 4.109: Convergence plot for the dual shape function derivatives for the 3D bilinear quadrilateral

4.6.2.4 Normal and tangent vectors

4.6.2.4.1 Theory :

Figure 4.110: Normal average for a 3D geometry

As previously introduced, the tangent vector is taken as the complementary direction to the normal (4.114a), then
in order to define its derivative we just need to compute the normal derivative, then the computation of the tangent
derivative will be automatically taken into account with the AD procedure. The definition of the normal for 3D surface
geometries corresponds with (4.114b), where can be defined as the cross product between the rows of the Jacobian
matrix. Then we compute the unit normal the same way we do for the 2d case (4.101b).
The average normal behaves exactly the same to the definition presented for the 2D case (4.101c), the main

Vicente Mataix Ferrándiz Page 177 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

difference in the practical case came with the fact that the number of neighbours in 3D is not bounded, see Figure
4.110, in contrast with the 2D case where there are only two potential neighbours.

(4.114a) τ =I−n×n
(4.114b) narea = x,ξ × x,η

Then in order to evaluate ∆ for the normal on 3D surfaces we can apply the chain rule to the previous definition
on (4.114b) and (4.101b). The resulting derivative (4.115) results as an addition.

∆narea knarea k − narea ∆ knarea k ∆ x,ξ × x,η kx,ξ × x,η k − x,ξ × x,η ∆ kx,ξ × x,η k
(4.115a) ∆n = 2
= 2
knarea k kx,ξ × x,η k

Where the first derivative is:

 c   c   c   c 
ns ns ns ns

X X X X
(4.115b) ∆ x,ξ × x,η = Nk,ξ ∆xk  ×  Nk,η xk  +  Nk,ξ xk  ×  Nk,η ∆xk 
k=1 k=1 k =1 k=1

And the second term:

x,ξ × x,η ∆ x,ξ × x,η

(4.115c) ∆ x,ξ × x,η =
= n∆ x,ξ × x,η
x,ξ × x,η

Finally, in order to obtain the derivative of the average normal we proceed following the same equation presented
for the 2D case (4.103). The normal derivative for the GP also behaves the same way as in 2D (4.104).

4.6.2.4.2 Convergence study :

The convergence plot of the normal vector derivatives for the 6 different geometrical pairs of the 3D linear
triangles from Table 4.22 can be found in the Figure 4.111. We can appreciate that the convergence is greater than
cubic.

Figure 4.111: Convergence plot for the normal vector derivatives for the 3D linear triangle

The convergence plot of the normal vector derivatives for the 3 different geometrical pairs of the 3D bilinear
quadrilaterals from Table 4.24 can be found in the Figure 4.112.

Page 178 of 374 Vicente Mataix Ferrándiz

CHAPTER 4. CONTACT MECHANICS 4.6 Derivatives for contact mechanics linearization

Figure 4.112: Convergence plot for the normal vector derivatives for the 3D bilinear quadrilateral

Vicente Mataix Ferrándiz Page 179 of 374

4.6 Derivatives for contact mechanics linearization CHAPTER 4. CONTACT MECHANICS

Page 180 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookAnd01] From Convexity to Nonconvexity. Anders Klarbring (auth.) R. P. Gilbert P. D. Panagiotopoulos
P. M. Pardalos (eds.) Lars-Erik Andersson. Springer US. 1st ed. 2001.
[BookBAL03] Friction & flow stress in forming & cutting. Philippe Boisse, Taylan Altan, and Kees van Luttervelt.
Sterling, VA: Kogan Page. 2003.
[BookBBT01] The friction and lubrication of solids. Frank Philip Bowden, Frank Philip Bowden, and David Tabor.
Oxford university press. 2001.
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookBHS07] Encyclopedia of Computational Mechanics (3 Volume Set). René de Borst, T.J.R. Hughes, and
Erwin Stein. Wiley. 1st ed. 2007.
[BookBou85] Application des potentiels à l’étude de l’équilibre et du mouvement des solides élastiques.
Joseph Boussinesq. Gauthier-Villars, Imprimeur-Libraire. 1885.
[BookEbe99] Dynamic collision detection using oriented bounding boxes. David Eberly. 1999.
[BookEri04] Real-time collision detection. Christer Ericson. CRC Press. 2004.
[BookFSS12] OpenDSA: a creative commons active-ebook (abstract only). Eric Fouh, Maoyuan Sun, and
Clifford Shaffer. 2012. Pp. 721–721.
[BookHBC09] Isogeometric Analysis: Toward Integration of CAD and FEA. T.J.R. Hughes, Yuri Bazilevs, and
J. Austin Cottrell. Wiley. 1st ed. 2009.
[BookHug+14] Computer graphics: principles and practice. John F Hughes, Andries Van Dam, James D Foley,
Morgan McGuire, Steven K Feiner, and David F Sklar. Pearson Education. 2014.
[BookKO88] Contact problems in elasticity: a study of variational inequalities and finite element methods.
Noboru Kikuchi and John Tinsley Oden. siam. 1988.
[BookLau10] Computational Contact and Impact Mechanics: Fundamentals of Modeling Interfacial Phenomena
in Nonlinear Finite Element Analysis. Tod A. Laursen. Springer. 2010.
[BookMau13] Contact, adhesion and rupture of elastic solids. Daniel Maugis. Springer Science & Business
Media. 2013.
[BookPhD06] R-Trees: Theory and Applications. Alexandros Nanopoulos PhD Apostolos N. Papadopoulos PhD
Yannis Theodoridis PhD (auth.) Yannis Manolopoulos PhD. Springer-Verlag London. 1st ed. 2006.
[BookPop10] Contact Mechanics and Friction: Physical Principles and Applications. Valentin L. Popov.
Springer-Verlag Berlin Heidelberg. 1st ed. 2010.
[BookSE02] Geometric tools for computer graphics. Philip Schneider and David H Eberly. Elsevier. 2002.
[BookSK13] Computational Contact Mechanics: Geometrically Exact Theory for Arbitrary Shaped Bodies.
Karl Schweizerhof and Alexander Konyukhov. Springer-Verlag Berlin Heidelberg. 1st ed. 2013.
[BookTos05] Domain Decomposition Methods — Algorithms and Theory. Andrea Toselli Olof B. Widlund (auth.)
Springer-Verlag Berlin Heidelberg. 1st ed. 2005.

Vicente Mataix Ferrándiz Page 181 of 374

ARTICLES BIBLIOGRAPHY

[BookWoh01] Discretization Methods and Iterative Solvers Based on Domain Decomposition.

Barbara I. Wohlmuth. Springer-Verlag Berlin Heidelberg. 1st ed. 2001.
[BookWP99] New Developments in Contact Problems. Peter Wriggers and Panagiotis Panagiotopoulos.
Springer-Verlag Wien. 1st ed. 1999.
[BookWri06] Computational Contact Mechanics. Peter Wriggers. Springer. 2nd. 2006.
[BookYas13] Numerical Methods in Contact Mechanics. Vladislav A. Yastrebov. Wiley-ISTE. 1st ed. 2013.
[BookZZT13] The Finite Element Method: its Basis and Fundamentals. O. C. Zienkiewicz, J.Z. Zhu, and
Robert L. Taylor. Butterworth-Heinemann. 7th ed. 2013.

Articles
[ArtAC91] “A mixed formulation for frictional contact problems prone to Newton like solution methods”. P. Alart
and A. Curnier. In: Computer Methods in Applied Mechanics and Engineering. No. 3, Vol. 92, 1991,
pp. 353–375. DOI: 10.1016/0045-7825(91)90022-X.
[ArtAnt17] “A further analysis on the analogy between friction and plasticity in Solid Mechanics”.
Nicolas Antoni. In: International Journal of Engineering Science. Vol. 121, 2017, pp. 34–51.
Elsevier. DOI: 10.1016/j.ijengsci.2017.08.012.
[ArtBar08] “Adhesive elastic contacts – JKR and more”. Etienne Barthel. In: Journal of Physics D: Applied
Physics. Vol. 41, 2008, p. 163001. IOP Publishing. DOI:
10.1088/0022-3727/41/16/163001.
[ArtBen+11] “A large deformation, rotation-free, isogeometric shell”. D.J. Benson, Y. Bazilevs, M.-C. Hsu, and
T.J.R. Hughes. In: Computer Methods in Applied Mechanics and Engineering. No. 13, Vol. 200,
2011, pp. 1367–1378. Elsevier. DOI: 10.1016/j.cma.2010.12.003.
[ArtBMP] “The adapted augmented Lagrangian method: a new method for the resolution of the mechanical
frictional contact problem”. Philippe Bussetta, Daniel Marceau, and Jean-Philippe Ponthot. In:
Computational Mechanics. No. 2, , pp. 259–275. DOI: 10.1007/s00466-011-0644-z.
[ArtBT39] “The area of contact between stationary and moving surfaces”. Frank Philip Bowden and
David Tabor. In: Proceedings of the Royal Society of London. Series A. Mathematical and Physical
Sciences. No. 938, Vol. 169, 1939, pp. 391–413. The Royal Society London.
[ArtCai+14] “Collision detection using axis aligned bounding boxes”. Panpan Cai, Chandrasekaran Indhumathi,
Yiyu Cai, Jianmin Zheng, Yi Gong, Teng Sam Lim, and Peng Wong. In: . 2014, pp. 1–14. Springer.
DOI : 10.1007/978-981-4560-32-0_1.

[ArtCam15] “Computational modeling and simulation of rupture of membranes and thin films”.
Eduardo MB Campello. In: Journal of the Brazilian Society of Mechanical Sciences and
Engineering. No. 6, Vol. 37, 2015, pp. 1793–1809. Springer. DOI:
10.1007/s40430-014-0273-5.
[ArtCC12] “An augumented lagrangian method to solve three-dimensional nonlinear contact problems”.
FJ Calavalieri and Alberto Cardona. In: Latin American applied research. No. 3, Vol. 42, 2012,
pp. 281–289. SciELO Argentina.
[ArtCC13a] “Three-dimensional numerical solution for wear prediction using a mortar contact algorithm”.
Federico José Cavalieri and Alberto Cardona. In: International Journal for Numerical Methods in
Engineering. No. 8, Vol. 96, 2013, pp. 467–486. Wiley Online Library. DOI: 10.1002/nme.4556.
[ArtCC13b] “An augmented Lagrangian technique combined with a mortar algorithm for modelling mechanical
contact problems”. FJ Cavalieri and A Cardona. In: International Journal for Numerical Methods in
Engineering. No. 4, Vol. 93, 2013, pp. 420–442. Wiley Online Library. DOI: 10.1002/nme.4391.
[ArtCC15] “Numerical solution of frictional contact problems based on a mortar algorithm with an augmented
Lagrangian technique”. F. J. Cavalieri and A. Cardona. In: Multibody System Dynamics. No. 4, Vol.
35, 2015, pp. 353–375. DOI: 10.1007/s11044-015-9449-8.

Page 182 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY ARTICLES

[ArtCFC12] “Un algoritmo de contacto mortar aplicable a problemas tridimensionales”. F.J. Cavalieri,
V.D. Fachinotti, and A. Cardona. In: Revista Internacional de Métodos Numéricos para Cálculo y
Diseño en Ingeniería. No. 2, Vol. 28, 2012, pp. 80–92. DOI:
10.1016/j.rimni.2012.03.001.
[ArtCho+17] “An overview of recent results on Nitsche’s method for contact problems”. Franz Chouly,
Mathieu Fabre, Patrick Hild, Rabii Mlika, Jerome Pousin, and Yves Renard. In: . 2017, pp. 93–141.
Springer.
[ArtCou85] “The theory of simple machines”. CA Coulomb. In: Mem. Math. Phys. Acad. Sci. No. 161-331, Vol.
10, 1785, p. 4.
[ArtDB15] “A literature review of bounding volumes hierarchy focused on collision detection”. Simena Dinas
and José M Bañón. In: Ingeniería y competitividad. No. 1, Vol. 17, 2015, pp. 49–62. Universidad
del Vall.
[ArtDem+12] “Computational comparison of the bending behavior of aortic stent-grafts”. Nicolas Demanget,
Stéphane Avril, Pierre Badel, Laurent Orgéas, Christian Geindreau, Jean-Noël Albertini, and
Jean-Pierre Favre. In: Journal of the mechanical behavior of biomedical materials. No. 1, Vol. 5,
2012, pp. 272–282. Elsevier. DOI: 10.1016/j.jmbbm.2011.09.006.
[ArtDFS98] “Solution of coercive and semicoercive contact problems by FETI domain decomposition”.
Zdenek Dostál, Ana Friedlander, and Sandra A Santos. In: Contemporary Mathematics. Vol. 218,
1998, pp. 82–93. Providence, RI: American Mathematical Society. DOI:
10.1090/conm/218/03003.
[ArtDon99] “A simple benchmark problem to test frictional contact”. CY Dong. In: Computer methods in applied
mechanics and engineering. No. 1-2, Vol. 177, 1999, pp. 153–162. Elsevier. DOI:
10.1016/S0045-7825(98)00377-6.
[ArtDPS] “A frictional mortar contact approach for the analysis of large inelastic deformation problems”.
T Doca, FM Andrade Pires, and JMA Cesar de Sa. In: International Journal of Solids and
Structures. No. 9, , pp. 1697–1715. Elsevier. DOI: 10.1016/j.ijsolstr.2014.01.013.
[ArtDSB15] “High-order mortar-based element applied to nonlinear analysis of structural contact mechanics”.
APC Dias, AL Serpa, and ML Bittencourt. In: Computer Methods in Applied Mechanics and
Engineering. Vol. 294, 2015, pp. 19–55. Elsevier. DOI: 10.1016/j.cma.2015.05.013.
[ArtEul50a] “Sur la diminution de la resistance du frottement”. Leonhard Euler. In: Mémoires de l’académie des
sciences de Berlin. 1750, pp. 133–148.
[ArtEul50b] “Sur le frottement des corps solides”. Leonhard Euler. In: Mémoires de l’Académie des Sciences
de Berlin. 1750, pp. 122–132.
[ArtFF03] “Collision Detection for Virtual Prototyping Environments”. Mauro Figueiredo and T Fernando. In: .
2003, pp. 157–166.
[ArtFic63] “Sul problema elastostatico di Signorini con ambigue condizioni al contorno”. Gaetano Fichera. In:
Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Natur.(8). Vol. 34, 1963, pp. 138–142.
[ArtFig10] “S-cd: surface collision detection toolkit for virtual prototyping”. Mauro Figueiredo. In: IADIS
International Journal on Computer Science and Information Systems. Vol. 5, 2010, pp. 72–86.
[ArtFW06] “Mortar based frictional contact formulation for higher order interpolations using the moving friction
cone”. Kathrin A Fischer and Peter Wriggers. In: Computer methods in applied mechanics and
engineering. No. 37-40, Vol. 195, 2006, pp. 5020–5036. Elsevier. DOI:
10.1016/j.cma.2005.09.025.
[ArtFZ75] “A note on numerical computation of elastic contact problems”. A Francavilla and OC Zienkiewicz.
In: International Journal for Numerical Methods in Engineering. No. 4, Vol. 9, 1975, pp. 913–924.
Wiley Online Library. DOI: 10.1002/nme.1620090410.
[ArtHar+09] “A contact domain method for large deformation frictional contact problems. Part 2: Numerical
aspects”. S. Hartmann, J. Oliver, R. Weyler, J.C. Cante, and J.A. Hernández. In: Computer
Methods in Applied Mechanics and Engineering. No. 33–36, Vol. 198, 2009, pp. 2607–2631. DOI:
10.1016/j.cma.2009.03.009.

Vicente Mataix Ferrándiz Page 183 of 374

ARTICLES BIBLIOGRAPHY

[ArtHar+10] “A 3D frictionless contact domain method for large deformation problems”. S Hartmann, R Weyler,
J Oliver, JC Cante, and JA Hernández. In: Computer Modeling in Engineering and Sciences
(CMES). No. 3, Vol. 55, 2010, p. 211.
[ArtHer82] “Über die berührung fester elastische Körper und über die Harte”. H Hertz. In: Verhandlungen des
Vereins zur Beförderung des Gewerbefleisses. 1882,
[ArtHTK77] “A finite element method for large displacement contact and impact problems”. TJR Hughes,
RL Taylor, and W Kanoknukulchai. In: Formulations and Computational Algorithms in FE Analysis.
1977, pp. 468–495. MIT Press: Boston.
[ArtHWP18] “A truly variationally consistent and symmetric mortar-based contact formulation for finite
deformation solid mechanics”. Michael Hiermeier, Wolfgang A. Wall, and Alexander Popp. In:
Computer Methods in Applied Mechanics and Engineering. Vol. 342, 2018, pp. 532–560. DOI:
10.1016/j.cma.2018.07.020.
[ArtJKR71] “Surface energy and the contact of elastic solids”. Kenneth Langstreth Johnson, Kevin Kendall, and
AD Roberts. In: Proceedings of the royal society of London. A. mathematical and physical sciences.
No. 1558, Vol. 324, 1971, pp. 301–313. The Royal Society London.
[ArtKar39] “Minima of functions of several variables with inequalities as side constraints”. William Karush. In:
M. Sc. Dissertation. Dept. of Mathematics, Univ. of Chicago. 1939,
[ArtKor94] “Monotone multigrid methods for elliptic variational inequalities I”. Ralf Kornhuber. In: Numerische
Mathematik . No. 2, Vol. 69, 1994, pp. 167–184. Springer. DOI: 10.1007/s002110050178.
[ArtKra68] “Friction and wear”. IV Kragelsky. In: Mechanical Engineering, Moscow. 1968, Elsevier.
[ArtLWZ12a] “A mortar formulation for 3D large deformation contact using NURBS-based isogeometric analysis
and the augmented Lagrangian method”. L. de Lorenzis, Peter Wriggers, and G. Zavarise. In:
Computational Mechanics. No. 1, Vol. 49, 2012, pp. 1–20. Springer. DOI:
10.1007/s00466-011-0623-4.
[ArtLWZ12b] “A mortar formulation for 3D large deformation contact using NURBS-based isogeometric analysis
and the augmented Lagrangian method”. L. De Lorenzis, Peter Wriggers, and G. Zavarise. In:
Computational Mechanics. No. 1, Vol. 49, 2012, pp. 1–20. Springer-Verlag. DOI:
10.1007/s00466-011-0623-4.
[ArtMor73] “Systèmes élastoplastiques de liberté finie”. JEAN JACQUES Moreau. In: Travaux du Séminaire
d’Analyse Convexe. Vol. 3, 1973, p. 33.
[ArtNag05] “Simple local interpolation of surfaces using normal vectors”. Takashi Nagata. In: Computer Aided
Geometric Design. No. 4, Vol. 22, 2005, pp. 327–347. DOI: 10.1016/j.cagd.2005.01.004.
[ArtNet+14] “Applying Nagata patches to smooth discretized surfaces used in 3D frictional contact problems”.
D. M. Neto, M. C. Oliveira, L. F. Menezes, and J. L. Alves. In: Computer Methods in Applied
Mechanics and Engineering. Vol. 271, 2014, pp. 296–320. Elsevier. DOI:
10.1016/j.cma.2013.12.008.
[ArtNit71] “Über ein Variationsprinzip zur Lösung von Dirichlet-Problemen bei Verwendung von Teilräumen,
die keinen Randbedingungen unterworfen sind”. Joachim Nitsche. In: . No. 1, Vol. 36, 1971,
pp. 9–15.
[ArtNOM17] “Surface Smoothing Procedures in Computational Contact Mechanics”. D. M. Neto, M. C. Oliveira,
and L. F. Menezes. In: Archives of Computational Methods in Engineering. No. 1, Vol. 24, 2017,
pp. 37–87. DOI: 10.1007/s11831-015-9159-7.
[ArtOde81] “Exterior penalty methods for contact problems in elasticity”. JT Oden. In: . 1981, pp. 655–665.
Springer. DOI: 10.1007/978-3-642-81589-8_33.
[ArtOli+09a] “A contact domain method for large deformation frictional contact problems. Part 1: Theoretical
basis”. J. Oliver, S. Hartmann, J. C. Cante, R. Weyler, and J. A. Hernández. In: Computer Methods
in Applied Mechanics and Engineering. No. 33–36, Vol. 198, 2009, pp. 2591–2606. DOI:
10.1016/j.cma.2009.03.006.
[ArtOli+09b] “A New Approach in Computational Contact Mechanics: The Contact Domain Method”. X Oliver,
S Hartmann, JC Cante, R Weyler, and JA Hernández. In: CIMNE, Barcelona. 2009,

Page 184 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY ARTICLES

[ArtPGW09] “A finite deformation mortar contact formulation using a primal–dual active set strategy”.
Alexander Popp, Michael W. Gee, and Wolfgang A. Wall. In: International Journal for Numerical
Methods in Engineering. No. 11, Vol. 79, 2009, pp. 1354–1391. John Wiley & Sons, Ltd. DOI:
10.1002/nme.2614.
[ArtPL04] “A mortar segment-to-segment contact method for large deformation solid mechanics”.
Michael A. Puso and Tod A. Laursen. In: Computer Methods in Applied Mechanics and Engineering.
No. 6–8, Vol. 193, 2004, pp. 601–629. Elsevier. DOI: 10.1016/j.cma.2003.10.010.
[ArtPop+10] “A dual mortar approach for 3D finite deformation contact with consistent linearization”.
Alexander Popp, Markus Gitterle, Michael W. Gee, and Wolfgang A. Wall. In: International Journal
for Numerical Methods in Engineering. No. 11, Vol. 83, 2010, pp. 1428–1465. John Wiley & Sons,
Ltd. DOI: 10.1002/nme.2866.
[ArtPP15] “The research works of Coulomb and Amontons and generalized laws of friction”. Elena Popova
and Valentin L Popov. In: Friction. No. 2, Vol. 3, 2015, pp. 183–190. Springer. DOI:
10.1007/s40544-015-0074-6.
[ArtPPS03] “Vibration damping by friction forces: theory and applications”. Karl Popp, Lars Panning, and
Walter Sextro. In: Modal Analysis. No. 3-4, Vol. 9, 2003, pp. 419–448. Sage Publications Sage CA:
Thousand Oaks, CA. DOI: 10.1177/107754603030780.
[ArtPR15a] “An unconstrained integral approximation of large sliding frictional contact between deformable
solids”. Konstantinos Poulios and Yves Renard. In: Computers & Structures. Vol. 153, 2015,
pp. 75–90. DOI: 10.1016/j.compstruc.2015.02.027.
[ArtPR15b] “An unconstrained integral approximation of large sliding frictional contact between deformable
solids”. Konstantinos Poulios and Yves Renard. In: Computers & Structures. Vol. 153, 2015,
pp. 75–90. Elsevier. DOI: 10.1016/j.compstruc.2015.02.027.
[ArtPT92] “A mixed formulation for the finite element solution of contact problems”. Panayiotis Papadopoulos
and Robert L Taylor. In: Computer Methods in Applied Mechanics and Engineering. No. 3, Vol. 94,
1992, pp. 373–389. Elsevier. DOI: 10.1016/0045-7825(92)90061-N.
[ArtRao16] “Art of Modeling in Contact Mechanics”. Michel Raous. In: . Vol. 570, 2016, Springer. DOI :
10.1007/978-3-319-40256-7_4.
[ArtRKC02] “Fast Continuous Collision Detection between Rigid Bodies”. Stephane Redon,
Abderrahmane Kheddar, and Sabine Coquillart. In: Computer Graphics Forum. No. 3, Vol. 21, 2002,
pp. 279–287. DOI: 10.1111/1467-8659.t01-1-00587.
[ArtRom11] “Advanced methods for robot-environment interaction towards an industrial robot aware of its
volume”. Fabrizio Romanelli. In: Journal of Robotics. Vol. 2011, 2011, Hindawi. DOI:
10.1155/2011/389158.
[ArtSig33] “Sopra alcune questioni di elastostatica”. Antonio Signorini. In: Atti della Societa Italiana per il
Progresso delle Scienze. No. II, Vol. 21, 1933, pp. 143–148.
[ArtSN17] “Woven fabrics computational simulation using beam-to-beam contacts formulation”.
Mauro Takayama Saito and Alfredo Gay Neto. In: Revista Interdisciplinar De Pesquisa Em
Engenharia. No. 22, Vol. 2, 2017, pp. 09–25.
[ArtSpe] “The Hertz contact problem with finite friction”. DA Spence. In: Journal of elasticity . No. 3-4, ,
pp. 297–319. Springer. DOI: 10.1007/BF00126993.
[ArtSWP18] “A computational approach for thermo-elasto-plastic frictional contact based on a monolithic
formulation using non-smooth nonlinear complementarity functions”. Alexander Seitz,
Wolfgang A Wall, and Alexander Popp. In: Advanced Modeling and Simulation in Engineering
Sciences. No. 1, Vol. 5, 2018, p. 5. Springer. DOI: 10.1186/s40323-018-0098-3.
[ArtSWT85] “A perturbed Lagrangian formulation for the finite element solution of contact problems”.
Juan C Simo, Peter Wriggers, and Robert L Taylor. In: Computer methods in applied mechanics
and engineering. No. 2, Vol. 50, 1985, pp. 163–180. Elsevier. DOI:
10.1016/0045-7825(85)90088-X.

Vicente Mataix Ferrándiz Page 185 of 374

PH.D.’S THESIS BIBLIOGRAPHY

[ArtTC03] “A model of adhesion coupled to contact and friction”. C. Talon and A. Curnier. In: European
Journal of Mechanics - A/Solids. No. 4, Vol. 22, 2003, pp. 545–565. DOI:
10.1016/S0997-7538(03)00046-9.
[ArtTK14] “Real-Time Hybrid Virtuality for Prevention of Excavation Related Utility Strikes”. Sanat Talmaki and
Vineet R. Kamat. In: Journal of Computing in Civil Engineering. No. 3, Vol. 28, 2014, p. 04014001.
DOI : 10.1061/(ASCE)CP.1943-5487.0000269.

[ArtTP] “On a patch test for contact problems in two dimensions”. Robert L. Taylor and
Panagiotis Panagiotopoulos. In: . ,
[ArtWI93] “On the treatment of nonlinear unilateral contact problems”. P Wriggers and M Imhof. In: Archive of
Applied Mechanics. No. 2, Vol. 63, 1993, pp. 116–129. Springer. DOI: 10.1007/BF00788917.
[ArtWie+18] “Algebraic multigrid methods for dual mortar finite element formulations in contact mechanics”.
T. A. Wiesner, A. Popp, M. W. Gee, and W. A. Wall. In: International Journal for Numerical Methods
in Engineering. No. 4, Vol. 114, 2018, pp. 399–430. DOI: 10.1002/nme.5748.
[ArtWoh02] “A Comparison of Dual Lagrange Multiplier Spaces for Mortar Finite Element Discretizations”.
Barbara I. Wohlmuth. In: ESAIM: Mathematical Modelling and Numerical Analysis. No. 6, Vol. 36,
2002, pp. 995–1012. EDP Sciences. DOI: 10.1051/m2an:2003002.
[ArtWZ13] “Hertz Theory: Contact of Spherical Surfaces”. Q. Jane Wang and Dong Zhu. In: . 2013,
pp. 1654–1662. Springer US. Boston, MA. DOI: 10.1007/978-0-387-92897-5_492.
[ArtYCF11] “A local contact detection technique for very large contact and self-contact problems: sequential
and parallel implementations”. VA Yastrebov, Georges Cailletaud, and Frédéric Feyel. In: . 2011,
pp. 227–251. Springer. DOI: 10.1007/978-3-642-22167-5_13.
[ArtYL08] “A contact searching algorithm including bounding volume trees applied to finite sliding mortar
formulations”. Bin Yang and TodA. Laursen. In: Computational Mechanics. No. 2, Vol. 41, 2008,
pp. 189–205. Springer-Verlag. DOI: 10.1007/s00466-006-0116-z.
[ArtYLM] “Two dimensional mortar contact methods for large deformation frictional sliding”. Bin Yang,
Tod A. Laursen, and Xiaonong Meng. In: International Journal for Numerical Methods in
Engineering. No. 9, , pp. 1183–1225. Chichester, New York, Wiley [etc.] 1969-. DOI:
10.1002/nme.1222.
[ArtZhu12] “Tutorial on hertz contact stress”. Xiaoyin Zhu. In: . Vol. 521, 2012, pp. 1–8.
[ArtZK12] “k-IOS: Intersection of spheres for efficient proximity query”. Xinyu Zhang and Young J. Kim. In:
2012 IEEE International Conference on Robotics and Automation. 2012, pp. 354–359. DOI:
10.1109/ICRA.2012.6224889.
[ArtZL09] “A modified node-to-segment algorithm passing the contact patch test”. Giorgio Zavarise and
Laura De Lorenzis. In: International journal for numerical methods in engineering. No. 4, Vol. 79,
2009, p. 379. DOI: 10.1002/nme.2559.

Ph.D.’s thesis
[PhDBru08] “Contact analysis and overlapping domain decomposition methods for dynamic and nonlinear
problems”. Stephan BrunSSen. 2008.
[PhDBus] “Modélisation et résolution du problème de contact mécanique et son application dans un contexte
multiphysique”. Philippe Bussetta.
[PhDDoc] “Energy wear methods for dual-mortar contact analysis of frictional problems at finite inelastic
strains”. Thiago de Carvalho Rodrigues Doca.
[PhDDro15] “Local mortar method for contact problems with high order non-matching meshes”.
Guillaume Drouet. 2015.
[PhDGit12] “A dual mortar formulation for finite deformation frictional contact problems including wear and
thermal coupling”. Markus Gitterle. 2012.

Page 186 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY ONLINE RESOURCES

[PhDHam13] “Frictional mortar contact for finite deformation problems with synthetic contact kinematics”.
Michael E. Hammer. 2013.
[PhDKra01] “Monotone multigrid methods for Signorini’s problem with friction”. Rolf H Krause. 2001.
[PhDPop12] “Mortar Methods for Computational Contact Mechanics and General Interface Problems”.
Alexander Popp. 2012.
[PhDYas11] “Computational contact mechanics: geometry, detection and numerical techniques”.
Vladislav A. Yastrebov. 2011.

Online resources
[OnlCho] Collision Detection Using the Separating Axis Theorem. URL :
https://gamedevelopment.tutsplus.com/tutorials/collision-detection-
using-the-separating-axis-theorem-gamedev-169.
[OnlJac] Jacobi elliptic functions. URL: https:
//www.encyclopediaofmath.org/index.php/Jacobi_elliptic_functions.
[OnlRtr] Wikipedia. R-tree. URL : https://en.wikipedia.org/wiki/R-tree.

Vicente Mataix Ferrándiz Page 187 of 374

ONLINE RESOURCES BIBLIOGRAPHY

Page 188 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY

Chapter 5

Plasticity

As the water shapes itself to the

vessel that contains it, so a wise man
adapts himself to circumstances.

Confucius
(551 - 479 BC, Chinese philosopher)

5.1 Introduction
The mathematical modelling of plasticity in solids and structures are essential for a good modulation of the forming
processes[BookBan10]. As well as other industrial processes and applications, like the design of steel and concrete
civil structures, or the understanding of the response of the soil or rocks in geotechnical problems[BookBHS07].
We can shortly introduce[BookNPO09] that in contrast with the elastic constitutive laws, the elasticplastic laws are
path-dependent and dissipative[BookZTF14], this means that large part of the work expended in plastically deforming
the material is irreversibly converted to other forms of energy, generally heat[BookHil98]. The stress depends on the
entire history of the deformation[BookHS98], and cannot be written as a single-valued function of the strain; rather it
can only be specified as a relation between rates of stress and strain[BookBel+14].

5.1.1 Historical outline

Plasticity is one of the most developed theories for describing material NL behaviour. The history of the studies in
this scientific discipline, which involves mathematics, physics and engineering applications is rich. For a complete
overview of the history of plasticity, we address directly to the overview presented on the book of Hill[BookHil98] and
Zyczkowski[BookZyc81], much more complete than the one presented here.

Having said that, the first developments came from the Coulomb[ArtCou76] contributions in 1776, when he
presented a work stating the dependence of the sliding resistance on a plane between two bodies, being a function of
the adhesion and the frictional properties1 . This work, together with the later developments published by Poncelet
and Rankine, was useful for the calculation of retaining walls.

We can identify as the firsts[BookWei] properly speaking works on plasticity the contributions of Tresca[BookTre65]
in 1864 on the extrusion of metals, this was the later denominated Tresca yield criterion, which states that the metal
yields when the maximal shear stress surpasses a threshold. During that time, the also-French mathematician St.
Venant[ArtSai70] introduced basic constitutive relations for perfectly plastic materials, where it related the direction on
the strain increase and the principal stresses. These developments were later extended to 3D by Lévy.
1 We discuss in the contact chapter 4.Contact mechanics, the similarities between the frictional contact theory and the plasticity models.

Vicente Mataix Ferrándiz Page 189 of 374

5.2 State of the Art in numerical plasticity CHAPTER 5. PLASTICITY

In 1886 Bauschinger [ArtBau86] introduced the effect that now carries its name, which it consists in the fact
that by deforming a metal in one direction until its elasticity limit has been exceeded, and then deforming it in the
opposite direction, its proportionality limit in the latter direction is lower due to the material imperfections. It was before
the First World War (WWI), in 1913, when Von Mises[ArtMis13] introduced the known as J2 theory. Later in 1924
Prandtl[ArtPra24] extended the St. Venant work for elastic-plastic behaviour, which Reuss extended later to 3D. It
was not until 1928 when Von Mises[ArtMis28] extended his previous theory considering the yield function as a plastic
potential in the incremental stressstrain relations of the flow theory. That very same year Prandtl tried to formulate
general relations for hardening behaviour.

In the period between wars, and especially during the WWII,

many relevant developments were done on this field. The war stim-
ulated research through problems such as the calculation of the
stresses in autofrettaged gun barrels and of the forces resisting a
shot penetrating armour plate[BookHil98].

Additionally to this, it is particularly known and interesting to

highlight the problems originated during the production of the Liberty
ships[ArtZha16]. These were cargo ships built in the United States
(U.S.) during WWII in order to provide goods and carry troops. The
initial design was modified by the U.S. Maritime Commission to
conform to make it quicker and cheaper to build, this new design
consisted on replace part of the riveting operations with welding.
This was due because riveting represented the 30% of the workforce Figure 5.1: Liberty ships suddenly broke in half
costs. Unfortunately, no further studies were carried to understand while moored at the dock
to how welding would affect the structure[ArtGer12]. The rapid and
massive construction requirements originated several issues that many notes and trigger the alarm about a study
the reasons that induce these problems. Almost 1500 instances of significant brittle fractures were recorded (Figure
5.1). Usually this problem is studied as a fracture phenomenon, but studies from Altstetter et al.[ArtAA95] proposed
an alternative approach that enhanced localised plasticity might nucleate either ductile fracture or quasi-cleavage.
In any case, these problems lead deeper studies on the NL behaviour of the materials and benefited therefore the
understanding of the plasticity.

So in 1949 Prager [ArtPra49] proposed a framework for the plastic constitutive relations for hardening materials
with smooth yield functions. Two years later Drucker his material stability postulate. Together, they presented in
1952[ArtDP52] its criterion, which tries to represent the plastic deformation of soils. This is a pressure-dependent
model that determines whether a material has exceeded the elastic limit. On 1953 Koiter [ArtKoi53] generalised the
plastic stress-strain relations for non-smooth yield functions.

The following developments, due to the rise of the computational power were more focus on the research on the
numerical analysis, especially after the 1970s, so it will be detailed in the next section 5.2.State of the Art in numerical
plasticity.

5.2 State of the Art in numerical plasticity

5.2.1 Historical outline
We want to convey that the developments that will be shown in this work are within the classical theory, and we will
not be innovative in this matter. The work done in this field was needed in order to be able to address the forming
process problem. So the first works that can be considered as reference, the foundations for the numerical plasticity
including infinitesimal theory, can be found in the book from the 1950s of Hill[BookHil98] and later in the article of
Koiter [ArtKoi60].

Having said that, we can attribute[BookWei] to Duvaut and Lions[ArtDL76] the first study of the Boundary Value
Problem (BVP) of elasto-plasticity, who defined the problem as variational inequality2 . Later Johnson[ArtJoh76] split
2 See the Appendix D.Constrained optimisation problems for further detail. On this Appendix the inequality is treated in order to introduce the

mathematical concept necessaries to understand the CCM, but the theory can be applied on plasticity too.

Page 190 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.2 State of the Art in numerical plasticity

this variational inequality in two steps, one to solve the steady state problem after removing the velocity and a second
step to solve the velocity of the problem.
In 1985 J.C. Simo and R.L. Taylor [ArtST85], derived the consistent elasto-plastic tangent tensors. Cep . This result
can be seen in the corresponding section at 5.5.3.Numerical implementation tangent constitutive tensor.
Later, incremental finite strain elasto-plasticity for stable states of equilibrium and the coercivity of internal
energy is presented by using multiplicative decomposition of strain, Frechet-derivatives and the chain rule, yielding
corresponding consistent tangent operators with the same formal structure as for linearised strains, published by
J.C. Simo in 1988[ArtSim88a; ArtSim88b]. The theory related to this point is shown at 5.3.Finite strain elasto-plastic
models. Additionally, in the range of large deformations, the original proposition for the multiplicative decomposition of
the deformation gradient considered on this work can be attributed to Lee and Liu[ArtLL67].
In the last 20 years significant advances have been done, particularly on the development of finite deformation theo-
ries in order to solve practical and industrial problems which involve large strains. In order to applied the corresponding
NR strategy, we need the proper linearisation of the problem. In general these problems require to work with discrete
equations, where only a few cases as the J2 flow theory can provide a closed-form solution. So general algorithms for
the return-mapping algorithms are needed. Denominating these procedures as Closest Point Projection Method
(CPPM), complete monographs about this subject can be found in the work of Lubliner [BookLub08] and Simo and
Hughes[BookHS98].
The main works we want to address for a deeper understanding for the reader, which analize from a extended point
of view and with a general purpose, are the books of Neto[BookNPO09], Lubliner [BookLub08], Simo[BookSim93] and
Simo and Hughes[BookHS98].

5.2.2 Constitutive models

The following is a brief summary of the most commonly employed yield criteria (5.2.2.1.Yield criteria) and yield surface
hardening (5.2.2.2.Yield surface hardening).

5.2.2.1 Yield criteria

The discontinuity or yield criterion is a scalar function of tensor arguments that defines the elastic domain. It is
regularly presented in the form Φ(σ ,q), where σ is the Cauchy stress tensor, but can be represented with any stress
tensor, and q are the (stress-like) internal variables, grouped as "back-stress". The surface limits the stress state, in
such a way that is not admissible any state outside the defined surface.
The following[BookNPO09; BookOll14] summarises some of the most common yield criteria used in engineering
practice. We briefly summarise them, but for deeper understanding we recommend to attend to the literature.

• Tresca yield criterion: Originally presented in 1864[BookTre65]. It assumes that plastic yielding begins when
the maximum shear stress reaches a critical value.

• Barret de Saint Venant yield criterion: Barret de Saint Venant, published in 1871[ArtSai70] a constitutive
equation for inelastic of an elastically rigid, perfectly plastifying solid material in plane stress state with the
hypothesis of isotropic deformations, stating that the main axes of strain coincide with the main axes of stress.

• Von-Mises yield criterion: According to it the plastic yielding begins when the J2 stress deviator reaches a
critical value. Where J2 cam is defined as (5.1).

1
(5.1) J2 = tr (σdev )2
2
This is probably the most extended yield criterion, especially for metals.

• Mohr-Coulomb yield criterion: It is based on the assumption that the phenomenon of macroscopic plastic
yielding is the result of frictional sliding between material particles.

• Drucker-Prager yield criterion: It states that plastic yielding begins when the J2 invariant of the deviatoric
stress and the hydrostatic stress, p, reach a critical combination.

Vicente Mataix Ferrándiz Page 191 of 374

5.3 Finite strain elasto-plastic models CHAPTER 5. PLASTICITY

It is relevant to introduce here the concept of yield surface (Φ) and plastic potential (Γ) so we can define what
is an associated plastic flow rule and a non-associated plastic flow rule. When both potentials coincide, we
have an associated plastic flow rule, and a non-associated plastic flow rule otherwise. It can be interpreted as
saying that the plastic strain increment vector is normal to the yield surface (Φ), this is denominated the normality
rule[BookOll14].
All the yield criteria presented here, except for Barret de Saint Venant, are available in Kratos at the module
StructuralMechanicsApplication. This includes all the non-associated plastic flow rule combinations.
See B.2.Kratos Multiphysics for further details of the Kratos structure and 5.5.Implementation details for the imple-
mentation details of the constitutive models. For more detailed description on the yield criteria, we address directly to
the literature[BookOll14; BookNPO09].

5.2.2.2 Yield surface hardening

Essentially[BookNPO09], hardening is characterised by a dependence of yield stress level upon the history of plastic
straining to which the body has been subjected. The following are the main hardening behaviours (Figure 5.2
summarises the two last types of hardening behaviours):

• Perfect plasticity: A material model is said to be perfectly plastic if no hardening is allowed, that is, the yield
stress level does not depend in any way on the degree of plastification. In this case, the yield surface remains
fixed regardless of any deformation process the material may experience.

• Isotropic hardening: A plasticity model is said to be isotropic hardening if the evolution of the yield surface is
such that, at any state of hardening, it corresponds to a uniform (isotropic) expansion/contraction of the initial
yield surface, without translation. The movements of the yield surface can be:

◦ Positive: When the surface experiences an expansion of the initial yield surface. This is also denominated
an isotropic hardening elasto-plastic process.
◦ Null: When the plastic loading surface does not experience any evolution. Also referred as isotropic
perfectly elasto-plastic process.
◦ Negative: When there is a contraction in the initial yield surface. This corresponds with an isotropic
softening elasto-plastic process.

• Kinematic hardening: Appears when the yield surfaces preserve their shape and size but translate in the
stress space as a rigid body. It is frequently observed in experiments that, after being loaded (and hardened)
in one direction, many materials show a decreased resistance to plastic yielding in the opposite direction.
This phenomenon is known as the Bauschinger effect and can be modelled with the introduction of kinematic
hardening.

Figure 5.2: Types of hardening. Left: Isotropic hardening. Right: Kinematic hardening

5.3 Finite strain elasto-plastic models

In many industrial applications, such as the metal forming processes, we cannot take into account the infinitesimal
deformation hypothesis for the plasticity theory, without a significant loss in accuracy or even completely wrong

Page 192 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.3 Finite strain elasto-plastic models

results[BookNPO09]. So, in order to tackle this kind of problem we need to formulate the problem in finite deformations.
The following introduces the basic theoretical fundaments and applied concepts in order to formulate the finite
deformation elasto-plastic model.

5.3.1 Fundaments
The following introduces the base theoretical fundaments required to formulate the finite strain elasto-plastic prob-
lem. The Equation (5.2) introduces the base formulation required for a finite strain elasto-plasticity[BookBHS07;
BookNPO09] models.

Figure 5.3: Multiplicative decomposition of the deformation gradient F

1. Multiplicative decomposition of the deformation gradient F, being Fe and Fp the elastic and plastic strains (Figure
5.3) originally proposed in [ArtLL67].

(5.2a) F = Fe · Fp

2. The free-energy potential Ψ̄(Ce ,α) and elastic strain-stress relation for the stresses (S̄) in the intermediate
configuration. In here Ce correspond with the elastic right Cauchy-Green tensor and the (strain-like) internal
variables (α) of the current constitutive model. With the free-energy potential it is possible to derive the
hyperelastic law considered.
(
S̄ = 2 ∂∂Ψ
Ce
(5.2b)
S̄ := Fe−1 τ Fe−T = Fp SFpT

3. Yield surface Φ(T, q) for the stress-like hardening variables q= −∂Ψ/∂α and the Mandel stress T= FeT τ Fe−T =
Ce S̄. The yield function will define the onset of the plastic yielding.

4. Plastic evolution equations, or dissipation potential, from which the flow rule and hardening law (evolution laws)
for internal variables are derived. In order to define that we require of a plastic potential (Γ). Being Lp the plastic
velocity gradient and the general flow functions, Mp the plastic flow function and Mh the general hardening flow
function.
(
T
Lp = γ̇ R e ∂Γ e
∂τ R = γ̇ Mp (T, q) (This expression is also known as the normality rule)
(5.2c)
α̇ = γ̇ Mh (T, q)

For Lp = Ḟp Fp−1 and the plastic multiplier (γ̇ ), with Mp = ∂Γ

∂T and Mh = ∂Γ
∂q for the associative case.

5. Loading/unloading conditions, for the elasto-plastic model the KKT complementary conditions and the consis-
tence condition during plastic loading.
(
γ̇ ≥ 0, Φ ≤ 0, γ̇Φ = 0
(5.2d)
γ Φ̇ = 0 (Consistency condition)

Vicente Mataix Ferrándiz Page 193 of 374

5.3 Finite strain elasto-plastic models CHAPTER 5. PLASTICITY

From these KKT conditions we can deduce[BookOll14]:


 Φ < 0 ⇒ γ̇ = 0 Elastic behaviour or unloading,
γ̇ > 0
(5.2e)  Φ=0⇒
γ̇ = 0 Neutral plastic load
Φ > 0 ⇒ Incompatible state

Additionally in the case of numerical implementation it is necessary to implement return-mapping algorithm

to obtain the current state of the material. In our case we take into consideration the implicit Backward Euler
algorithm, well known in the literature[ArtMT94]. More details concerning the return-mapping algorithm in the
following section,5.3.3.Finite strain return mapping operations.

5.3.2 Plastic loading surface

5.3.2.1 Introduction
In order to evaluate the plastic multiplier (γ̇ ), we will adopt the formulation from Oller [BookOll14; PhDOll88]. This will
be evaluated considering the plastic loading function in addition to the yield surface (Φ). In order to understand the
difference between these two functions, we need to define the influence in the constitutive behaviour by each one of
the functions[BookOll14]. The limit between the elastic and the plastic areas is set through the yield surface (Φ) and
from such limit this surface can move in the stress space, follow the evolution of the plastic process and transform
itself into the so-called plastic loading surface. This function representing the plastic loading surface is simply the
discontinuity or yield limit function updated for each of the (stress-like) internal variables value (q) at every moment of
the pseudo-time (t) of the elasto-plastic process. The phenomenon governing the yield surface change of position in
the stress space is known as plastic hardening, where some models have been introduced in the state-of-the-art
section, 5.2.2.2.Yield surface hardening, and now we detail the formulation for isotropic (5.3.2.2.Isotropic hardening)
and kinematic hardening (5.3.2.3.Kinematic hardening).

5.3.2.2 Isotropic hardening

We define the isotropic hardening considering a plastic hardening function (K), which we generally define with the
(stress-like) internal variables (q). In classic plasticity, it is defined depending on the internal variable of plastic
hardening, Equation (5.3).

(5.3) K κp = f κp

Defining the hardening function (K) as an internal variable of the plastic process, we can obtain a more general
formulation, as seen in (5.4). In here the tensor function (hκ (σ , q)) and the scalar function (hκ (σ , q)) depends on the
updated tensor state and the internal variables.

 
∂Γ (σ ; κp )
κ̇p = γ̇ Hκ (σ , q) = γ̇ hκ (σ , q) :
(5.4) ∂σ
K̇ = γ̇ HK (σ , q) = hK (σ , q)κ̇p

5.3.2.3 Kinematic hardening

As previously stated, the kinematic hardening consists in a translational motion of the plastic loading surface. The
motion depends on the kinematic plastic hardening internal variable (η ), which defines the stress space origin, or
back stress. The continuous change of position of this coordinate origin during the elasto-plastic process induces a
translational movement of the yield surface (Φ), which can be combined with an isotropic movement. We can express
this in a general form as in the Equation (5.5).

(5.5) Φ(σ , q) = f (σ − η ) − K = 0

Page 194 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.3 Finite strain elasto-plastic models

√ p
In here the plastic hardening can be defined according to Prager and Melan as η̇ = β κ̇p , with β = ck ε̇ε̇p . ck
depends on the type of plastic potential considered, being ck = 32 hk for VM and Equation (5.6) in general.

 s 
1 σij εpij
(5.6) ck =  p p ·  · hk
ε̇rs ε̇rs f (σkl − ηkl )

5.3.2.4 Stress-strain relation

This can be obtained by the plastic yield general criterion and Prager s consistency condition, see Equation
(5.7).

Φ(σ , q) = f (σ − η ) − K = 0 
∂Φ ∂Φ
(5.7a) ∂Φ ∂Φ ∂Φ ⇒ : σ̇ + : η̇ − K̇ = 0
Φ̇ = : σ̇ + : η̇ + K̇ = 0 ∂σ ∂η
∂σ ∂η ∂K

From this latter expression, and replacing η̇ = β κ̇p = ck ε̇P then we can derive Equation (5.7b).

∂Φ ∂Φ P

: C : ε̇e + ck : ε̇ − hK hκ : ε̇P = 0
∂σ ∂η
(5.7b)    
∂Φ ∂Φ ∂Γ ∂Φ ∂Γ ∂Γ
: C : ε̇ − γ̇ :C: − ck : + hK hκ : =0
∂σ ∂σ ∂σ ∂η ∂σ ∂σ
From the latter expression, the plastic consistency factor (γ̇ ). The plastic multiplier defines a factor which evaluates
the distance between an inadmissible tensor state outside the domain and the plastic loading surface. With this we
define γ̇ as in Equation (5.7c).
∂Φ
: C : ε̇
∂Γ
(5.7c) γ̇ =   being γ̇ ≥ 0
∂Φ ∂Γ ∂Γ ∂Φ ∂Γ


−ck : + hκ hκ : + ∂σ :C: ∂σ
∂ n ∂σ ∂σ
| {z }
A

In here we have isolated A. If we consider a zero kinematic hardening (ck = 0) then A can be defined as:

∂Φ ∂Γ
(5.7d) A=− σ = hK hκ :
∂κp ∂σ

5.3.2.5 Stability condition

It is relevant to introduce here the concept of local stability condition, or Druckers stability condition. This concept is
only meaningful for the stability behaviour of a point in a solid, therefore in order to ensure the stability of the whole
solid we require to check the stability in each point if using this criterion. Besides, if the stability is loss in some points
this does not necessarily mean that the global stability has been lost. For example, in softening materials, in which
the local stability condition may not be satisfied in some points, but globally may not be the case. The stability of the
whole solid can be proved by a weaker condition known as global stability condition.

5.3.2.5.1 Local stability :

The Druckers second postulates requires (5.8), and due to the fact that γ̇ is always positive and defined in (5.7c),
therefore can be ensured if the convexity of Γ, plastic potential, is required. One simple way to ensure γ̇ > 0 and the
∂Γ ∂Φ
convexity of Γ is to impose an associated plastic flow rule ( ∂σ ∝ ∂σ ).

 
∂Γ
(5.8) ε̇ : C : ε̇p = ε̇ : C : γ̇ ≥0
∂σ

Vicente Mataix Ferrándiz Page 195 of 374

5.3 Finite strain elasto-plastic models CHAPTER 5. PLASTICITY

5.3.2.5.2 Global stability :

For softening materials the previous postulate is sufficient, but not necessarily, stability condition. In order to
formulate a global stability condition we will consider a weak form, as we do in order to formulate FEM. Defining Π
as the total energy in the system in the final configuration, Π∗ the total energy in the initial configuration and δ u the
virtual displacement from the initial configuration to the final one. With this we define Equation (5.9), which defines
the equilibrium while applying the virtual displacement. In here, the variation of the total potential energy developed is
null.

1 2 1
(5.9) Π = Π∗ + δΠ + δ Π + ··· ⇒ ∆Π = Π − Π∗ ∼ δΠ + δ 2 Π + · · ·
= |{z}
2! 2!
0

As δΠ ≈ 0, the total increase of the virtual work is equal to the second variation of the functional and it is, therefore,
the concave or convex stability condition of the functional. With this if ∆Π > 0 the original configuration is stable for
any δ u, and
R ∆Π < 0 is unstable. With this calling Ω0 the initial configuration and Ωp the plastic configuration, and as
∆Π ∼ = 21 Ω δσ : δεd Ω we can define the global stability condition as (5.10).

Z " Z Z #
1 1 1  
(5.10) ∆Π ∼
= δσ : δεd Ω = δσ : δεd Ω + δσ : δεd Ω ∼
= ∆ΠΩ0 + ∆ΠΩp > 0
2 Ω 2 Ω0 2 Ωp

5.3.2.6 Condition of unicity of solution

Considering two different virtual displacements, δ u1 and δ u2 , the potential energy difference is equivalent to the
potential energy for a virtual displacement equivalent to the difference of those virtual displacements (∆(δ u) =
δ u2 − δ u1 ), with the corresponding increment on the strain (∆(δε)) and stress (∆(δσ )). The corresponding increment
in potential energy is Equation (5.11).

Z 

=0 There is no unicity of the solution
(5.11) ∆ δ2 Π = ∆(δσ ) : ∆(δε)d Ω
Ω 6= 0 There is unicity of the solution

If ∆ δ 2 Π = 0 during the virtual displacement change ∆(δ u), it means that the stress in both final configurations
are the same δσ 1 = δσ 2 ; in consequence, ∆(δσ ) = δσ 1 − δσ 2 = 0. Therefore, there are two admissible and
independent of each other kinematic states δ u1 = 6 δ u2 , but the stress increment is identical for both configurations
δσ 1 = δσ 2 , which implies that there

is not a single solution, but there exists a bifurcation. Considering the previously
stated kinematic states, if ∆ δ 2 Π 6= 0, the unicity of the solution is guaranteed.

5.3.3 Finite strain return mapping operations

In addition to the formerly presented, we need to consider some additional concepts. Besides the multiplicative
decomposition between the elastic (Fe ) and plastic (Fp ) deformation gradient from (5.2a), we must decompose these
deformation gradient in its stretch (Ue , Ve , Up , Vp ) and rotation tensors (Re , Rp ) as seen in (5.12).

Fe = Re Ue = Ve Re
(5.12)
Fp = Rp Up = Vp Rp

With this decomposition we are able to define the general elastic predictor/return-mapping algorithm. The crucial
difference between the discretisation of the large strain problem and the infinitesimal one lies in the numerical
approximation of the plastic flow equation, see (5.2c). In order to be consistent with the presented multiplicative

Page 196 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.4 Large deformation elastic models

decomposition, we will consider an exponential map backward discretisation[BookNPO09]. Considering an implicit

scheme algorithm, the resulting update formula for the plastic deformation gradient (Fp ) results in (5.13).

 
p ∂Γ
(5.13a) Fn+1 = exp Ren+1
T
Ren+1 Fpn
∂τ n+1

In view of the isotropy of the tensor exponential function can be simplified as:
 
p ∂Γ
(5.13b) Fn+1 = Ren+1
T
exp Ren+1 Fpn
∂τ n+1

The incompressibility of the plastic flow for pressure insensitive flow potentials is carried over exactly to the incremental
rule (5.13b).
As we have adopted, a standard Backward Euler difference scheme to discretise the plastic flow equation, the
updating formula for the plastic deformation gradient will be refactored to (5.14). This formula in general is not
volume-preserving, therefore this solution will result in an accuracy loss in the resolution of elasto-plastic constitutive
equations of plastically incompressible models.

 − 1
p ∂Γ
(5.14) Fn+1 = I− ReT
n+1 Ren+1 Fpn
∂τ n+1

With this, in order to update the elastic deformation gradient (Fe ) we need to combine the definition of the plastic
deformation gradient (Fp ) with the multiplicative elasto-plastic split, we obtain the following equivalent kinematic update
expression in terms of the elastic deformation gradient (5.15).

 
∂Γ
(5.15a) Fen+1 = F∆ Fen ReT
n+1 exp Ren+1
∂τ n+1

The definition of the incremental deformation gradient is the following:

(5.15b) F∆ ≡ Fn+1 (Fn )−1

With this last expression we can compute the trial elastic deformation gradient, if we save the deformation gradient
from the previously converged time step. This will help us to compte a consistent update of the elastic deformation
gradient.

(5.15c) Fetrial = F∆ Fn

In order to be consistent with the expression from (5.14), therefore we calculate the update of the elastic deformation
gradient as:
 
∂Γ
(5.15d) Fen+1 = Fetrial I+ ReT
n+1 Ren+1
∂τ n+1

5.4 Large deformation elastic models

5.4.1 Introduction
In order to consider elastic-plastic models with large deformations, we need to consider elastic models in large
deformations that will act as the predictors of our plastic behaviour, derived from the free-energy potential from (5.2b).
The most commonly used models for that purpose are the hyperelastic ones. We will present the two hyperelastic
models considered on this work, the Neo-Hookean and Kirchhoff material. Before introduce these models, we
will present the common theory that defines the hyperelastic materials. Hyperelastic materials[BookBel+14] are

Vicente Mataix Ferrándiz Page 197 of 374

5.5 Implementation details CHAPTER 5. PLASTICITY

characterised by the existence of a stored (or strain) energy function that is a potential for the stress (S) (5.16a). In
here Ψ corresponds with the stored energy potential, on the other hand, w is a potential expressed in terms of the
Green-Lagrange strain.

∂Ψ(C) ∂ w(E)
(5.16a) S=2 =
∂C ∂E
We can obtain in a similar manner the consistent tangent moduli CSE as (5.16b).
∂ 2 ψ (C) ∂ 2 w(E)
(5.16b) CSE = 4 =
∂ C∂ C ∂ E∂ E

5.4.2 Neo-Hookean material

This model is an extension of the isotropic linear Hookes law to finite strain. The material can be used to model
materials such as plastics and rubber-like substances. The corresponding potential is shown in the Equation (5.17a).
With this and considering (5.16a) we can obtain the corresponding stresses (5.17b). In a similar manner, we can
νE E
obtain with (5.16b) the consistent C can be obtained with (5.17c). In here λ0 = (1+ν )(1 −2ν ) and µ0 = 2(1+ν ) are the
Lamé constants.

1 1
(5.17a) Ψ(C) = λ0 (ln J)2 − µ0 ln J + µ0 (trace C − 3)
2 2


S =λ0 ln JC−1 + µ0 I − C−1
(5.17b)
τ =λ0 ln JI + µ0 (B − I)
 
SE
Cijkl =λCij−1 Ckl−1 + µ Cik−1 Cjl−1 + Cil−1 Ckj−1
(5.17c)

τ
Cijkl =λδij δkl + µ δik δjl + δil δkj

5.4.3 Kirchhoff material

The so-called Saint Venant-Kirchhoff material or just Kirchhoff material is probably the simplest NL elastic model. It is
basically an extension of the . In cases where we have a problem in small strains and large rotations, this material
provides a proper NL behaviour. In this case the potential w can be expressed as (5.18a). With this we can obtain the
corresponding CSE (5.18b) and the S (5.18c).

Z Z
1 1
(5.18a) w= Sij dEij = Cijkl Ekl dEij = Cijkl Eij Ekl = E · C : E
2 2
(5.18b) CSE = λI ⊗ I + 2µI
(5.18c) S = λ trace(E)I + 2µE = C : E

5.5 Implementation details

5.5.1 Introduction
The following section will include a series of numerical details related with the implementation of the constitutive
models presented in this chapter. In a first subsection, 5.5.2.Class structure in Kratos, we will detail the modular
structure followed in order to implement this numerical models inside Kratos. In this section only the isotropic
hardening constitutive laws are detailed, but the very same principle has been considered for the kinematic hardening
constitutive laws. For further detail about Kratos structure, we directly address to the corresponding appendix,
B.2.Kratos Multiphysics.
The last subsection will introduce the numerical methods considered in order to obtain the tangent operators of
the respective constitutive models, 5.5.3.Numerical implementation tangent constitutive tensor.

Page 198 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.5 Implementation details

5.5.2 Class structure in Kratos

Figure 5.4: Elastic-plastic CL class structure

Vicente Mataix Ferrándiz Page 199 of 374

5.5 Implementation details CHAPTER 5. PLASTICITY

The class structure of the CL implemented in Kratos can be summarised as shown in Figure 5.4. This provides a
modular interface similar to a Matrioshka, where the classes are recursively inside the others. In this class structure,
we can appreciate like the templates are an essential part in the design. The main components of the CL are:

• PlasticPotential: Corresponds with Γ. It depends on the Voigt size (TVoigtSize). This class defines
the plastic potential derivative.

• YieldSurface: Corresponds with Φ. This depends on the PlasticPotential. It is the class responsible
of the equivalent stress and its derivative.

• GenericConstitutiveLawIntegratorPlasticity: With a derived class for Finite Strain (FS). De-

pends on the YieldSurface. This is the class responsible of computing and integrate the stress vector.

• GenericSmallStrainIsotropicPlasticity: Depends on GenericConstitutiveLawIntegratorPlastici

This is the CL which actually computes the response, for Small Strain (SS).

• GenericSmallStrainIsotropicPlasticity: The same as the previous case, but for FS. Additionally
to the GenericFiniteStrainConstitutiveLawIntegratorPlasticity also depends on the FS
elastic CL from 5.4.Large deformation elastic models.

• ConstitutiveLaw: Finally, everything it also derives from the base CL, which defines the base API for
constitutive models.

• Parameters: The CL operates with the Parameters class which is extensively used as input class.

In conclusion, this is a generic interface for constructing elastic-plastic constitutive laws, which allows an extensive
combination. The drawback to this is a significative increase at the compilation time for each combination.

5.5.3 Numerical implementation tangent constitutive tensor

It is not always possible to obtain the consistent tangent constitutive tensor in an analytical expression, or if possible
this one is not trivial. The class structure shown in the previous section 5.5.2.Class structure in Kratos allows to
obtain a large number of elastic-plastic non-associated CL combinations, and the corresponding C which obtainment
is not as simple as in the associated cases. In order to obtain the consistent C we proceed with a numerical procedure
based in the stress vector perturbation method[ArtMar+08; ArtMOB11]. This procedure is based in the concept
from Equation (5.19) from Barbu[PhDBar16] and Cornejo[PhDCor], where the components of C are associated to
the increases in the stresses and the strain[ArtMOB11]. An alternative method to the previously stated would be
considering the AD, as the method is applied with a Forward Euler scheme is denominated sometimes as Forward
Automatic Differentiation (FAD)[ArtCOH14; ArtRH15].

t ∂σij
(5.19) C = Cijkl =
∂εekl

Here εekl corresponds with the elastic components of the strain tensor.

Particularly for the elasto-plastic tangent operator (Cep ) is given by the expression of (5.20). This expression is not
easy to compute in general, due to this, it is interesting to have a numerical alternative that provides us a general
method for computing the tangent tensor.

   ∂Φ 
ep Cp : ∂Γ
∂σ ⊗ ∂σ :C
(5.20) C =C− ∂Γ ∂Φ ∂Γ


hc hκ : ∂σ + ∂σ : Cp : ∂σ

Page 200 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.5 Implementation details

5.5.3.1 Forward FD
This is the approach followed by Martinez[ArtMar+08; ArtMOB11]. The approach takes only the first-order term of the
Taylor expansion of (5.19). This leads to the forward FD from Equation (5.21a). Applying this FD to (5.19) we obtain
the expression of Equation (5.21b). We can appreciate that this method implies to probate and calculate the stress 6
times in case of a 3D Voigt notation strain tensor is considered, therefore this is an expensive procedure, but can be
acceptable in case of an implicit approach as ours.
The Forward difference can be defined as follows:
 
∂u ui+1 − ui
(5.21a) ≈
∂x i ∆x
Applying to our problem:
εe +∆εe εe
t ∆σij σij kl kl − σij kl
(5.21b) C= Cijkl ≈ ≈
∆εekl ∆εekl

Unfortunately, this approach is a mere first order O(h), Figure 5.5, so in order to ensure the quadratic convergence
of the NR we recommend to consider the approach of the following section (5.5.3.2.Centered FD).

5.5.3.2 Centered FD
Applying a Taylor expansion to the criteria previously shown, and considering central difference approach (5.22a),
we can obtain a second order approach O(h)2 [ArtPRH00]. The rate of convergence can be proved to be quadratic in
this case, see Figure 5.5. In this case the drawback is that the number of perturbations required is double than the
previous scheme, so the computational cost increase must be collated with the convergence boost obtained; therefore
this will depend of the type of problem solved and the number of DOF of the system.
The definition of a central difference is the following:
 
∂u ui+1 − ui −1
(5.22a) ≈
∂x i 2∆x
We can apply this to the former (5.19) and then obtain:
εe +∆εekl εe −∆εekl
t ∆σij σij kl − σij kl
(5.22b) C= Cijkl ≈ ≈
∆εekl 2∆εekl

This method is considered by default for constitutive relations in the software MFront developed by Helfer [ArtHPF15].
This software is specialised in the modelling of generic constitutive behaviours, showing up the generality and power
of the method.

5.5.3.3 Numerical details

In Kratos, as any standard FE code, the constitutive equations of the material provide for a certain deformational
state (ε or F), the associated the stress tensor and its internal variables (α). From the resulting tensile-deformation
state, a small disturbance is applied to ε or F, and the stresses associated with this new deformation are computed.
These stresses, together with the perturbation applied, will be used to calculate C by means of the equations shown.
This procedure will be repeated for each component of the corresponding strain tensor (ε) or the deformation gradient
(F).
From the mathematical point of view, the ideal perturbation should be ∆ε → 0 to obtain the exact C. Of course
from a numerical point of view this is not possible, and we must compute finite value such that the approximation is
numerically close to the analytical value but free of numerical issues. For this, we will consider the Martinez[ArtMOB11]
criteria, shown in Equation (5.23).

6 0 → ∆εj = εj · 10−5
if εj =
(5.23)
if εj = 0 → ∆εj = min {|εk |} · 10−5 ∀k = 1, n εk 6= 0

Vicente Mataix Ferrándiz Page 201 of 374

5.6 Numerical examples CHAPTER 5. PLASTICITY

Figure 5.5: Consistent constitutive tensor perturbation method convergence. The code which tests the convergence
rate can be found here

Applying this, we ensure that the increase will be always close to the initial configuration. The only drawback of
the process described is that it can provide values of the perturbation near zero. i.e., when one of the components of
the ε is close to zero. In this situation, the Equations (5.21b) or (5.22b) may lead to an indeterminacy. In order to
avoid this situation, we need to ensure that the value of the perturbation is large enough, for that we can impose the
condition[ArtMOB11] from Equation (5.24).

(5.24) ∆εj > max {εk } · 10−10 ∀k = 1, n

5.6 Numerical examples

5.6.1 Introduction
The following section introduces three minimal examples to show the results obtained with the implementation
presented. First a minimal cube, one hexahedron element, is tested considering the proposed class structure
approach and compared with the exact consistent Von Mises yield criterion. The second example is a tensile test,
comparing the solution obtained with different types of elements. The third example is the application of the CL to a
contact problem.

5.6.2 Cube minimal example

The problem here presented consists in a very simple mesh, consisting in only one
hexahedron element, Figure 6.17. The size of the cube is 1 × 1 × 1. In this example
we will compare the solution obtained with an analytical consistent elastic-plastic
law with an associative Von Mises yield criterion (J2 ) with the CL derived of our
class structure previously presented.
So, this example will compare the solution obtained for an elastic-perfectly
plastic CL with the following parameters Table 5.1. In addition to the former, for
our implementation, a large value for the fracture energy must be considered in
order to properly model the perfect plasticity (i.e. 1016 ).
The nodes from the base (1, 2, 3, 4 in Figure 6.17) will be considered fixed, the
rest of the nodes (5, 6, 7, 8 in Figure 6.17) will be moved according to the following
pattern (5.25), where ω is the frequency of the movement and will be considered
to be ω = 5Hz. Figure 5.6: Simple cube geome-
try. Only one 8-node hexahedron

Page 202 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.6 Numerical examples

E ν γc
2 · 1011 Pa 0.3 9Pa

Table 5.1: Parameters considered for the cube minimal example

(5.25a) u = Ax1 + Bx2

Where:

(5.25b) x1 = sin(ω t)X, x2 = sin(2ω t)X

   
10−10 −2 · 10−10 0 0 7 · 10−10 0
(5.25c) A =  5 · 10−11 7 · 10−11 10−11  , B = 2.5 · 10−10 1.7 · 10−10 10−11 
−2 · 10−11 0 −3 · 10−10 0 10−10 −3 · 10−10

Figure 5.7: Von Mises stresses compared with the reference solution. The test can be found here

Finally, Figure 5.7 shows the solution comparison. We can see that the obtained solution is coincident with the
reference solution. We can also appreciate that as expected in a elastic-perfectly plastic case the γc acts as the
asymptotic value of the equivalent stress, Von Mises in this case.

5.6.3 Tensile test

(b) Mesh of hexahedra

(c) Mesh of tetrahedra

(a) Perspective of the geometry (d) Mesh of wedges

Figure 5.8: Mesh comparisons. It can be found here
The problem consists on a tensile test. Three different meshes have been created, see Figure 5.8, in order to
validate the computation of the internal energy dissipation, which depends on the element length as previously shown.
The test in general also shows the proper work of the implemented CL.

Vicente Mataix Ferrándiz Page 203 of 374

5.6 Numerical examples CHAPTER 5. PLASTICITY

The geometry can be appreciated in Figure 5.8a. The meshes of hexahedra and wedges are structured (Figures
5.8b and 5.8d), on the other hand, the mesh of tetrahedra is unstructured (Figure 5.8c). We are in tensile bar with
the following sizes 0.2 × 0.5 × 5.39 and the material properties shown in Table 5.2. Additionally in all cases, an
associative Von Mises model is considered, with an exponential softening as hardening curve.

E ν γc Gf Max. stress position Maximum stress Friction angle Dilatancy angle

2 · 1011 Pa 0.29 2 · 108 Pa 4 · 105 0.1 210 · 106 Pa 32 32
Table 5.2: Parameters considered for the tensile test

(a) Mesh of hexahedra

(b) Mesh of tetrahedra

(c) Mesh of wedges

Figure 5.9: Resulting plastic dissipation

The left side is blocked and the right side of the tensile bar is moved with a pacing of ux = 0.003t, being t the time
variable which goes from 0.847 to 20. The resulting plastic dissipation at the end of the simulation (Figure 5.10) for
each mesh corresponds with Figure 5.9.

Figure 5.10: Resulting displacement for the mesh of hexahedra

From the Figure 5.9 we can appreciate the effect of the mesh element in the results obtained. The mesh of
hexahedra from Figure 5.9a shows us a homogeneous distribution of the values. From the mesh of tetrahedra, we
obtain a distribution with higher peaks and narrower than the previous case. Finally, in the case of the wedges,
where the mesh is structured but non-symmetric, we can appreciate a very significative influence of the mesh. The
distribution obtained depends on the mesh orientation.

With this we can conclude that the mesh has a significative role on the results obtained from the point of view
of the CL considered. Because of this we must pay attention to the values considered and the mesh and element
type considered.

5.6.4 Application in CCM. Gears example

Page 204 of 374 Vicente Mataix Ferrándiz

CHAPTER 5. PLASTICITY 5.6 Numerical examples

The following is an example of a con-

tact problem considering a plasticity
law. The problem consists in two gears,
one is fixed, the other has a rotational
movement imposed. The rotational
movement imposed in the smaller gear
corresponds with Equation (5.26). The
movement is such as the two gears
come into contact.

(5.26)
    
ux = x0 − R cos atan y0 − ω t
x0
   
uy = y0 − R sin atan y0 − ω t
x0

The contact area is a small region

compared with the rest of the geometry
of the problem as seen in Figure 5.11.
Because of this, the contacting gear
does not experiences any significative
movement or increase of stresses in a
overall view, see Figure 5.13. Is on the
details of Figure 5.12 where we can
Figure 5.11: Gears example mesh
appreciate that the larger gear is the
one which experiences highest stresses. Because of this, it is the larger gear which exceeds γc and then yields, as
seen in Fugure 5.12c.

This example can be found in Kratos examples repository. Additionally to the current setup linear elastic case is
also present.

E ν γc Gf
2 · 1011 Pa 0.29 525106 Pa 1 · 108 Pa

Table 5.3: Parameters considered for gear example

The parameters are the ones considered in Table 5.3. As an elastic-perfectly plastic in an associative VM yield
criterion (J2 ) the only plasticity parameters we need to define are γc and Gf . In order to compute the contact in
considering this NL behaviour, the only precaution needed is to adjust the initial contact pair detection to avoid initial
overpressures due to contact pairs in tension that are not supposed to come into contact.

(a) Displacement solution (b) VM stress (c) Plastic dissipation

Figure 5.12: Detail solution for the displacement, VM stress and plastic dissipation

Vicente Mataix Ferrándiz Page 205 of 374

5.6 Numerical examples CHAPTER 5. PLASTICITY

(a) Displacement solution (b) VM stress

Figure 5.13: Overall solution for the displacement and the VM stress

Page 206 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookBan10] Sheet Metal Forming Processes: Constitutive Modelling and Numerical Simulation. Dorel Banabic.
Springer-Verlag Berlin Heidelberg. 1st ed. 2010.
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookBHS07] Encyclopedia of Computational Mechanics (3 Volume Set). René de Borst, T.J.R. Hughes, and
Erwin Stein. Wiley. 1st ed. 2007.
[BookHil98] The mathematical theory of plasticity. R. Hill. Clarendon Press; Oxford University Press. 1998.
[BookHS98] Computational Inelasticity. T.J.R. Hughes and J.C. Simo. Springer-Verlag New York. 1st ed. 1998.
[BookLub08] Plasticity theory. Jacob Lubliner. Dover Publications. web draft. 2008.
[BookNPO09] Computational Methods for Plasticity Theory and Applications. E. A. de Souza Neto, D. Periæ, and
D.R.J. Owen. Wiley. 2009.
[BookOll14] Nonlinear Dynamics of Structures. Sergio Oller. Springer International Publishing. 2014.
[BookSim93] Topics in the numerical analysis and simulation of classical plasticity. JC Simo. Elsevier Amsterdam.
1993.
[BookTre65] Mémoire sur l’écoulement des corps solides, par MH Tresca. Henri Tresca. Noblet et Baudry. 1865.
[BookWei] Plasticity: Mathematical Theory and Numerical Analysis. B. Daya Reddy (auth.) Weimin Han.
Springer-Verlag New York. 2nd ed.
[BookZTF14] The Finite Element Method for Solid and Structural Mechanics. O. C. Zienkiewicz, Robert L. Taylor,
and David Fox. Butterworth-Heinemann. 7th ed. 2014.
[BookZyc81] Combined loadings in the theory of plasticity. Michal Zyczkowski. Springer Science & Business
Media. 1981.

Articles
[ArtAA95] “Hydrogen-enhanced localization of plasticity in an austenitic stainless steel”. Daniel P Abraham
and Carl J Altstetter. In: Metallurgical and Materials transactions A. No. 11, Vol. 26, 1995,
pp. 2859–2871. Springer. DOI: 10.1007/BF02669644.
[ArtBau86] “Über die Veränderung der Elastizitätsgrenze und der Festigkeit des Eisens und Stahls durch
Strecken und Quetschen, durch Erwärmen und Abkühlen und durch oftmals wiederholte
Beanspruchung”. Johann Bauschinger. In: Mitteilungen des mechanisch-technischen
Laboratoriums der Königlich Technischen Hochschule München. No. 1, Vol. 13, 1886,
[ArtCOH14] “Automatic differentiation for numerically exact computation of tangent operators in small-and
large-deformation computational inelasticity”. Qiushi Chen, Jakob T Ostien, and Glen Hansen. In: .
2014, pp. 289–296. DOI: 10.1002/9781118889879.ch38.

Vicente Mataix Ferrándiz Page 207 of 374

ARTICLES BIBLIOGRAPHY

[ArtCou76] “An attempt to apply the rules of maxima and minima to several problems of stability related to
architecture”. CA Coulomb. In: Mémoires de l’Académie Royale des Sciences. Vol. 7, 1776,
pp. 343–382.
[ArtDL76] “Inequalities in Mechanics and Physics Springer Verlag Berlin”. G Duvaut and JL Lions. In: . 1976,
Heidelberg New York.
[ArtDP52] “Soil mechanics and plastic analysis or limit design”. Daniel Charles Drucker and William Prager. In:
Quarterly of applied mathematics. No. 2, Vol. 10, 1952, pp. 157–165. DOI:
10.1090/qam/48291.
[ArtGer12] “8 - Modeling hydrogen induced damage mechanisms in metals”. W. Gerberich. In: . Vol. 1, 2012,
pp. 209–246. Woodhead Publishing. DOI: 10.1533/9780857095374.2.209.
[ArtHPF15] “Implantation de lois de comportement mécanique à l’aide de MFront: simplicité, efficacité,
robustesse et portabilité”. Thomas Helfer, Jean-Michel Proix, and Olivier Fandeur. In: . 2015, Giens,
France.
[ArtJoh76] “Existence theorems for plasticity problems”. Claes Johnson. In: Journal de Mathématiques Pures
et Appliquées. No. 4, Vol. 55, 1976, pp. 431–444. GAUTHIER-VILLARS 120 BLVD
SAINT-GERMAIN, 75280 PARIS, FRANCE.
[ArtKoi53] “Stress-strain relations, uniqueness and variational theorems for elastic-plastic materials with a
singular yield surface”. Warner Tjardus Koiter. In: Quarterly of applied mathematics. No. 3, Vol. 11,
1953, pp. 350–354. DOI: 10.1090/qam/59769.
[ArtKoi60] “General theorems for elastic plastic solids”. Warner Tjardus Koiter. In: Progress of Solid
Mechanics. 1960, pp. 167–221. North Holland Press.
[ArtLL67] “Finite-strain elastic—plastic theory with application to plane-wave analysis”. EH Lee and DT Liu.
In: Journal of applied physics. No. 1, Vol. 38, 1967, pp. 19–27. AIP. DOI: 10.1063/1.1708953.
[ArtMar+08] “A numerical procedure simulating RC structures reinforced with FRP using the serial/parallel
mixing theory”. Xavier Martinez, Sergio Oller, Fernando Rastellini, and Alex H Barbat. In:
Computers & Structures. No. 15-16, Vol. 86, 2008, pp. 1604–1618. Elsevier. DOI:
10.1016/j.compstruc.2008.01.007.
[ArtMis13] “Mechanik der festen Körper im plastisch-deformablen Zustand”. R v Mises. In: Nachrichten von
der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse. Vol. 1913,
1913, pp. 582–592.
[ArtMis28] “Mechanik der plastischen Formänderung von Kristallen”. R von Mises. In: ZAMM-Journal of
Applied Mathematics and Mechanics/Zeitschrift für Angewandte Mathematik und Mechanik . No. 3,
Vol. 8, 1928, pp. 161–185. Wiley Online Library.
[ArtMOB11] “Caracterización de la delaminación en materiales compuestos mediante la teora de mezclas
serie/paralelo”. Xavier Martnez, Sergio Oller, and E Barbero. In: Revista internacional de métodos
numéricos para cálculo y diseño en ingeniera. No. 3, Vol. 27, 2011, pp. 189–199. Elsevier.
[ArtMT94] “A return mapping algorithm for isotropic elastoplasticity”. Anton Matzenmiller and Robert L Taylor.
In: International journal for numerical methods in engineering. No. 5, Vol. 37, 1994, pp. 813–826.
Wiley Online Library. DOI: 10.1002/nme.1620370507.
[ArtPra24] “Spannungsverteilung in plastischen Körpern”. Ludwig Prandtl. In: . 1924, pp. 43–54.
[ArtPra49] “Recent developments in the mathematical theory of plasticity”. William Prager. In: Journal of
applied physics. No. 3, Vol. 20, 1949, pp. 235–241. AIP. DOI: 10.1063/1.1698348.
[ArtPRH00] “Numerical differentiation for local and global tangent operators in computational plasticity”.
Agustín Pérez-Foguet, Antonio Rodríguez-Ferran, and Antonio Huerta. In: Computer Methods in
Applied Mechanics and Engineering. No. 1, Vol. 189, 2000, pp. 277–296. Elsevier. DOI:
10.1016/S0045-7825(99)00296-0.
[ArtRH15] “Automatic differentiation for stress and consistent tangent computation”. Steffen Rothe and
Stefan Hartmann. In: Archive of Applied Mechanics. No. 8, Vol. 85, 2015, pp. 1103–1125. Springer.
DOI : 10.1007/s00419-014-0939-6.

Page 208 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY PH.D.’S THESIS

[ArtSai70] “Memóire sur l’établissement des équations differéntielles des mouvements intérieurs opérés dans
les corps solides ductiles au delá des limites oú l’élasticité pourrait les ramener á leur premier état”.
Barre de Saint-Venant. In: Compt. Rend. Vol. 70, 1870, pp. 473–480.
[ArtSim88a] “A framework for finite strain elastoplasticity based on maximum plastic dissipation and the
multiplicative decomposition: Part I. Continuum formulation”. Juan C Simo. In: Computer methods
in applied mechanics and engineering. No. 2, Vol. 66, 1988, pp. 199–219. Elsevier. DOI:
10.1016/0045-7825(88)90076-X.
[ArtSim88b] “A framework for finite strain elastoplasticity based on maximum plastic dissipation and the
multiplicative decomposition. Part II: computational aspects”. Juan C Simo. In: Computer methods
in applied mechanics and engineering. No. 1, Vol. 68, 1988, pp. 1–31. Elsevier. DOI:
10.1016/0045-7825(88)90104-1.
[ArtST85] “Consistent tangent operators for rate-independent elastoplasticity”. Juan C Simo and
Robert Leroy Taylor. In: Computer methods in applied mechanics and engineering. No. 1, Vol. 48,
1985, pp. 101–118. Elsevier. DOI: 10.1016/0045-7825(85)90070-2.
[ArtZha16] “Technical problem identification for the failures of the liberty ships”. Wei Zhang. In: Challenges. No.
2, Vol. 7, 2016, p. 20. Multidisciplinary Digital Publishing Institute. DOI:
10.3390/challe7020020.

Ph.D.’s thesis
[PhDBar16] “Numerical simulation of fatigue processes: application to steel and composite structures”.
Lucia Gratiela Barbu. Universitat Politècnica de Catalunya. 2016.
[PhDCor] “Coupled formulation between the finite element method and the discrete element method for
studying multi-fracture processes in solids and structures”. Alejandro Cornejo.
[PhDOll88] “Un modelo de daño continuo para materiales-friccionales”. Sergio Oller. 1988.

Vicente Mataix Ferrándiz Page 209 of 374

PH.D.’S THESIS BIBLIOGRAPHY

Page 210 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING

Chapter 6

Adaptative remeshing

The measure of intelligence is

the ability to change.

Albert Einstein
(1879 - 1955 AD, German-born
theoretical physicist)

6.1 Introduction
Adaptive remeshing of FEM solutions refers to improving the quality of the solutions by enriching the approximation in
some manner so as to achieve the best solution for a given computational effort. The concept of adapting the mesh to
improve the quality of the solution is summarised in Figure 6.1 and the conceptual steps on the Algorithm 5.

(a) Contact patch test before remeshing (b) Contact patch test after remesh
Figure 6.1: Adaptive mesh techniques applied on a contact patch test

Vicente Mataix Ferrándiz Page 211 of 374

6.2 State of the Art in mesh refinement CHAPTER 6. ADAPTATIVE REMESHING

Thus the objective of adaptive remeshing is to obtain a mesh which is optimal, in the sense that the computational
costs involved are minimal under the constraint that the error in the FE solution is below a certain limit and, if possible,
distributed over the entire mesh. In order to do that we need to estimate the magnitude of the error of the FE solutions.
This is not trivial, as we do not know a priori the solution of the problem. Which means we need to develop techniques
which can estimate the error of the finite element solution from the known data[BookWri08], this means, on the
geometry, material data and our current approximate solution.

Algorithm 5 Adaptive refinement process

1: procedure A DAPTIVE REFINEMENT PROCESS
2: Select an initial mesh, which approximates accurately the geometry
3: while error > tolerance do
4: Solve the discrete problem
5: Compute error estimators or error indicators
6: if error > tolerance then
7: A new mesh has to be constructed
8: The already computed deformations and internal variables have to be projected onto the new mesh

This chapter introduces the adaptive remeshing techniques developed in order to improve the quality of the
solution. First the state-of-the-art of the respective techniques will be introduced on 6.2.State of the Art in mesh
refinement, later the different approaches followed to define the mesh metrics, each one with its respective section:
Hessian metric (6.3.Hessian based remeshing technique), level set metric (6.4.Level set based remeshing technique),
next the SPR method is presented (6.5.SPR based remeshing technique), later a short section dedicated to the
internal values interpolation techniques (6.6.Internal values interpolation) and finally an adaptation of the former
methods applied on CCM (6.8.Adaptive remeshing methods applied on CCM). Part of this work has been published
by Cornejo and Mataix[ArtCor+19].

6.2 State of the Art in mesh refinement

6.2.1 Mesh generation
During the last years, many techniques have emerged for this purpose. The early works from O.C. Zienkiewicz[ArtZP71]
on 1970s, which was only based on the geometries boundaries of the domain size and the required distribution of
element size1 . Since then many different technologies have arise, including mapping techniques, semi-automatic
remeshing methods, where the domain had to be subdivided manually in an initial stage into simpler subdomains[Art-
TWM82]. This methodology then deals with know-how dependency from the user, and that is generally limited for
structured meshes (quadrilaterals and hexahedra).
Alternatively to the structured meshes algorithms can find the family of unstructured meshes techniques. In
contrast with the previous ones, these methods deal with the called simplex geometries, corresponding to triangles
in two dimensions and tetrahedra in three dimensions. These techniques are based mainly in four families of
algorithms[BookZZT13; BookWri08]:

• Delaunay triangulation methods[ArtBow81].

• The advancing front method[ArtLo85].

• Tree methods, quadtrees for two dimensions and octrees for the three-dimensional cases[ArtYS84].

• The algorithm of recursive region splitting[ArtRRS95] places nodes on the boundary of the region to be meshed
by considering a certain density distribution.

These techniques can be used additionally for the remesh of unstructured quadrilateral meshes with the
combination of subsequent triangles, and the use of octree meshes for the automatic generation of hexahedric
meshes[BookZZT13; ArtPJR17; ArtPVR18].
1 Isoparametric mapping method

Page 212 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.2 State of the Art in mesh refinement

6.2.2 Adaptive finite element refinement techniques

Mesh adaptation is now widely used in numerical simulations to improve the accuracy of the solutions as well
as to capture the behaviour of physical phenomena[ArtFA05]. This technique allows to reduce considerably the
computational cost, associated with the corresponding DOF reduction, but obtaining an accurate solution neverthe-
less[ArtAF03].
This allows to compute complex problems with good results in three dimensions without dealing with the problematic
of the initial remeshing during pre-process, which can be a time consuming and error-prone[BookZZT13] task.
Additionally, allowing to adapt the computation during the simulation evolution, which alternative implies the creation
during the first stage a mesh that fits all the problem evolution, which can be a priori not known.
The process is called adaptation due to the existing dependence nature, the process depends on the previous
results[BookZZT13]; these methodologies were introduced originally by Babuka and Rheinboldt[ArtBR78][ArtBR79]
on the late 70s. The reasons for doing these adaptive procedures can be summarised, for solid mechanics problems,
displacements and stress errors should be contained in a certain tolerance. A very common approach of this is the
computation of the error energy norm and prescribing a threshold in order to not exceed a certain percentage of the
total energy norm.
In order to doing so several techniques exist, like the Recovery by Equilibrium of Patches (REP) techniques.
One of the most extended techniques based on REP is the SPR by Zienkiewicz and Zhu [ArtZZ92]. Of course the
methodologies based on REP are not the only ones available to measure the error, also from Zienkiewicz and Zhu
we can find techniques as the recovery methods[ArtZZ87]. From Babuka and Rheinboldt[ArtBR78] we can find the
residual-based methods, these methodologies as the name suggest estimate the error from the residual obtained on
the finite element approximation. We can also find relevant contributions in the field of mesh adaptability based on the
solution error in the work of Oñate and Bugeda[ArtOB93; ArtOña+06; ArtBO94]. To study on detail all the techniques
previously mentioned we address to the book of E.Stein et al.[BookRam01], a complete compendium of works from
different authors addressing a set of complex NL problems.
Recently techniques based on the Hessian have emerged[ArtWes+03], this requires that the variable to be used
as error estimation to be twice continuously differentiable in the case we consider an analytical solution, or otherwise
requires to us to be able to numerically estimate this second derivative. This methodology has additionally the
advantage of giving a proper measure in order to create an anisotropic new mesh, anisotropic meshes are those
that employ extremely narrow mesh elements. This type of meshes have the advantage to reduce the number of
new elements necessary on regions where the minimal size required would imply a high number of elements in case
of considering an isotropic mesh. These last techniques are the ones consider on this study and the theoretical
background will be expanded on the theory section (6.3.Hessian based remeshing technique).

Figure 6.2: Concept of h-refinement and p-refinement

Finally, we can identify several procedures for the finite element mesh refinement[BookWri08], the most common

Vicente Mataix Ferrándiz Page 213 of 374

6.3 Hessian based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

are the following four:

• h-refinement: The same type of elements is used but their size is changed. See Figure 6.2. This method can
be divided at the same time on:
◦ Enrichment: Also called element division. At the same time, we can distinguish between hanging nodes
and transition elements methods.
◦ Remeshing: A complete new mesh is generated.
• p-refinement: Where the same size of element is considered, but the polynomial order of the elements is
altered hierarchically. See Figure 6.2. Here the size of the local matrices of the elements is increased, with it
corresponding computational cost in computing and assembling.
• hp-refinement: This last technique consists in a combination of the two previous methods.
• r-refinement: Replaces finite element nodes such that the element sizes are optimised within a FE mesh in
order to reduce the overall error of the FE solution. The convergence to the exact solution for this method is not
guaranteed[BookZZT13].

The methodology considered on this work corresponds with the h-refinement, in particular we will consider the
remeshing technique, which means as already introduced, that we will create a whole new mesh. In order to do so we
will consider the Mmg library, an anisotropic metric based remeshing library. For more details we will address the
corresponding section in the appendixes B.3.Mmg library.

6.3 Hessian based remeshing technique

In this section we analyse on detail the techniques considered for remeshing. In order to understand the concepts
considered on the contact applied remeshing (6.8.2.Hessian metric), we need to introduce first the concepts of error
applied to a general PDE (6.3.1.Error estimation), later the concept of metrics is introduced (6.3.2.Metric based
remeshing) and general Hessian based metric measure (6.3.3.Hessian based metric measure). The section will end
with the introduction of the transfer operators used for the damage.

6.3.1 Error estimation

6.3.1.1 Upper bound on the interpolation error
We consider a general pde defined in a bounded domain Ω of ℜ3 where u denotes the exact solution of the problem.
We will call to uh the FE solution obtained in a certain mesh Th , and therefore we can define the solution error eu as
the difference between these two values, (6.1). Using this error, we desire to generate a new mesh Th on which the
error eu computed on this mesh new mesh, is bounded by a given tolerance value.

(6.1) eu = u − uh

Usually, the approximation error is usually significantly difficult to quantify, and depends intrinsically of the nature of
the PDE which define the problem. Because of this is interesting to define an indirect approach[ArtFA05] which allows
us to estimate eu . If we define Πh u as the interpolation of the solution u in the mesh Th , therefore the interpolation
error (ẽu ) will be defined as the difference between the exact solution and this interpolated one, (6.2).

(6.2) ẽu = u − Πh u

The Cea’s lemma[ArtCea64] it is an important tool for proving error estimates for the FEM applied to elliptic PDE,
although it is possible to prove that this approach also works well for hyperbolic PDE.

(6.3) ku − uh k ≤ c ku − Πh uk

Page 214 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.3 Hessian based remeshing technique

Where c is a constant independent of the current mesh Th . With this it is possible to control the approximation
erroreu by controlling the interpolation error ẽu .The solution uh converges towards u as h is reduced. In any case
the final objective is to reduce the number of DOF while preserving the desired level of accuracy for the FE solution.
Therefore, the objective of the following approach concern[ArtFA05]:

• The construction of a suitable ẽu estimate

• The definition of an anisotropic metric map

• The adaptation of a mesh to the sizing and stretching constraints of the metric map

Defining K as one element  in the given mesh Th , with its corresponding

  vertices of the given geometry. For
example, for the triangle K = a, b, c and for a tetrahedron K = a, b, c, d . In here, the interpolated solution Πh u
coincides with u in K . This interpolation is performed considering the standard FE shape functions2 . In the following
the deduction will assume that we work with linear tetrahedra. In order to bound ẽu = u − Πh u on K , we consider a
Taylor expansion with integral rest of the function e at a vertex of K (in here we consider the vertex a as an example)
with respect to any interior point x in K in (6.4).

Z 1
−
→ −
→ −
→−→
(6.4) (u − Πh u) (a) = (u − Πh u) (x) + xa, ∇ (u − Πh u) (x) + (1 − t) ax, H(x + t xa)ax dt
0

Where ∇u(x) denotes the gradient of the variable u, and H(x) denotes the Hessian, at the point x. Assuming that
the maximal error is reached at the point x (closer to a than to the rest of the vertex K , a, b, c for tetrahedra), see
(6.5a), and this is equivalent to (6.5b).

(6.5a) ∇ (u − Πh u) (x) = 0
(6.5b) h~v , ∇ (u − Πh u) (x)i = 0, ∀~v ⊂ K

Defining e(x) as the error, defined as (6.5c). Considering a′ as the point corresponding to the intersection of the
line ax with the face opposite to a. Considering the local coordinate ξ such that ~ax = ξ~aa′ . As a is closest to x than
any other vertex of K , then ξ ≤ 3/43 . Therefor (6.5c) can be rewritten as (6.5d).

Z 1
−
→ −
→−→
(6.5c) |e(x)| = (1 − t) ax, H(x + t xa)ax dt
0

Z 1
−
→ −
→
|e(x)| = (1 − t)λ2 aa′ , H(a + t xa)~a′ dt
0
Z 1
(6.5d) 9
6 max |h~aa′ , H(y )~a~a′ i| (1 − t)dt
16 y ∈aa′ 0

9
6 max |h~aa′ , H(y )~a′ i|
32 y ∈K

We can now consider the infinity norm (L∞ ) of the interpolation error ẽu , we obtain (6.5e). From this, the bound
(6.5f) can be deduced.
2 In case of a triangle, these shape functions can be defined as Π u = (1 − ξ − η )u(a) + ξ u(b) + η u(c), with 0 6 ξ + η 6 1. For the tetrahedron
h
the shape functions are defined as Πh u = (1 − ξ − η − ζ )u(a) + ξ u(b) + η u(c) + ζ u(d), with 0 6 ξ + η + ζ 6 1.
3 Remark, we are considering a tetrahedron in this deduction.

Vicente Mataix Ferrándiz Page 215 of 374

6.3 Hessian based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

9 D−→ −→E
(6.5e) ku − Πh uk∞,K 6 max aa′ , H(y )aa′
32 y ∈K
9 D−→ −→E
(6.5f) ku − Πh uk∞,K 6 max aa′ , |H(y)| aa′
32 y ∈K

As in (6.5f) the bound depends on the extremum x, which is not known a priori, we can reformulate it as (6.5g).
The former equation defines a bound in the case where the maximum error is achieved inside the element K . In
the case where the maximum error is obtained on the element face, then the corresponding expression coincides
with (6.5h). In the case where the maximum value is obtained along an element edge, the expression changes to
(6.5i).

9
(6.5g) ku − Πh uk∞,K 6 max max h~v , |H(y )| ~v i
32 y ∈K ~v ⊂K
2
(6.5h) ku − Πh uk∞,K 6 max max h~v , |H(y)| ~v i
9 y ∈[a,b,c] ~v ⊂[a,b,c]
1 D−→ −
→E
(6.5i) ku − Πh uk∞,K 6 max ab, |H(y)| ab
8 y ∈ab

As a summary, that will be taken into consideration as a base for deductions in the next section, relation (6.5f)
provides a proper bound for the interpolation error on an element K .

6.3.1.2 Numerical computation of the interpolation error

Taking (6.5f), where the upper bound on the interpolation error (ẽu ), we can reformulate it as (6.6a), in here cd is a
constant related to the space dimension. The former expression has some drawbacks, as the two maximum values
required are complex to compute.

(6.6a) ku − Πh uk∞,K 6 cd max max h~v , |H(x)| ~v i

x ∈K ~
v ⊂K

Because of that, we introduce the edges of K to replace the maximum value associated with all vectors included
in K . As any vector ~v of K can be written as linear combination of the edges of K , it yields (6.6b), where EK is the set
of edges of K .

(6.6b) ∀~v ⊂ K , k~v k|H(x)| 6 max k~ek|H(x)|

e∈EK

This allows us to rewrite the upper bound from (6.6a) as (6.6c). In this expression the RHS term is not trivial to
numerically evaluate. So we must assume the existence of a metric tensor M¯(K ) as defined in (6.6d). This metric M¯
is such that the region defined by {h~v , M¯(K )~v i|∀~v ⊂ K } is minimal in volume.

(6.6c) ku − Πh uk∞,K 6 cd max max h~e, |H(x)| ~ei

x ∈K e∈EK

(6.6d) max h~e, |H(x)| ~ei 6 h~e, M¯(K )~ei, ∀ e ∈ EK

x ∈K

Page 216 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.3 Hessian based remeshing technique

Finally, the interpolation error εK on an element K is given by the following relationship (6.6e).

(6.6e) εK = c maxh~e, M¯(K )~ei

e∈EK

The former relation relates the interpolation error εK to the square of the largest edge length in K with respect to
the metric M¯. This implies that it is possible to control the interpolation error on the mesh Th by controlling the length
of the mesh edges in K .

6.3.1.3 Mesh adaptation

From (6.6e), which defines εK , we can deduce that if the εK is fixed, the only remaining variable left to change is the
edge length of K . The objective will be adapt the mesh edges in order that the interpolation error will be equidistributed
over the adapted mesh, meaning that we aim to obtain the desired error level with a minimum number of DOF under
fixed constraints.
Considering ε the maximum level of error tolerated on the mesh elements. The mesh edges must be such that
(6.7a).

−
→
(6.7a) ε = c h~e, A (K )~ei, ∀ e ∈ EK

Considering (6.7b) the target metric tensor, this expression leads to (6.7c).

c
(6.7b) M (K ) = M¯(K )
ε
(6.7c) h~e, M (K )~ei = 1, ∀e ∈ EK

(6.7c) is equivalent to define the edge length with respect to the metric M (K ). In order to evaluate the length, we
need to consider the Euclidean norm of a vector ~u (L2 ) for a metric M is defined as (6.7d), and the distance between
two points given by (6.7e).

q √
(6.7d) k~u kM = h~u , ~u iM = t ~u M ~u
q
−→ −→ −→ −→
(6.7e) dM (A, B) = lM (AB) = kAB kM = t AB M AB

In consequence, the relation (6.7b) is equivalent to (6.7f), which prescribes edges of unit length for any edge of K
in order to bound the interpolation error εK on K by a value ε.

2
(6.7f) lM (K ) (~e) =1

The equation (6.7f) introduces the notion of unit mesh as the target optimal mesh. An optimal mesh is the mesh
on which the interpolation error εK is distributed uniformly and bounded by a tolerance value ε.

6.3.1.4 Computation of the relative error

Finally, the relation (6.6c) gives us an absolute bound on the interpolation error εK . [ArtFA05] considers a relative
bound on this error and therefore defining an estimate of the relative error. Doing this will provide a metric independent
of the nature of the physical problem, this allows to combine different variables together. Equation (6.8a) introduces

Vicente Mataix Ferrándiz Page 217 of 374

6.3 Hessian based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

how to compute dimensionless variables and introduces a relative error. We will present our own normalisation
procedures in the following sections.

 
u − Πh u |H(x)|
(6.8a) 6 cd max max ~e, ~e
| u| ǫ ∞,K x ∈K ~
e ∈ EK |u(x)|ǫ

In here |u|ǫ = max |u|, ǫkuk∞,Ω . Additionally we must take into account that the solutions vary from several orders
of magnitude. It is often difficult to capture the weakest phenomena via mesh adaptation. [ArtFA05] proposes as a
potential solution the consideration of a local error estimation in order to overcome this problem. Additionally[ArtCIA07;
ArtCas+97], the error estimate can also be normalised using the local value of the gradient (∇) norm (L2 ) of the
variable u, weak phenomena can be captured even in cases where the variables deal with sudden variations, like
shocks or BC dependent variables. In [ArtFA05] the following error estimate (6.8b) is proposed. Where h is the
element size on the mesh Th and 0 < α < 1.

 
u − Πh u |H(x)|
(6.8b) 6 c max max ~e, ~e
|α|u|ǫ + hk∇uk2 ∞,K
x ∈K ē∈EK α|u(x)|ǫ + hk∇u(x)k2

6.3.2 Metric based remeshing

With the concepts introduced in 6.3.1.Error estimation we are now ready to introduce the Hessian metric[ArtFA05]. In
here, we will introduce in first place the concept of metric (6.3.2.1.Concept of metric), then we will show the intersection
operations needed in case than more than one metric is taken into consideration (6.3.2.2.Metric intersection).

(a) Metric analogy (b) Intersection

Figure 6.3: Metric analogies. Images from [ArtFA05]

6.3.2.1 Concept of metric

The metric represents the operator which is equivalent to a geometric interpolation law on h on the edge[ArtFA05].
Let M be a discrete metric field defined at the vertices of a mesh Th of a domain Ω. Thanks to the interpolation
operation, we have a continuous metric field in the whole domain, i.e. a Riemannian metric space (M (x))x ∈Ω . This
representation of the metric field depends on Th as the interpolation law is applied at the element level.
In order to compute the length, we can consider two points in the domain, a and b. The length of the segment
between this two points can be computed as Gaussian integration as seen in (6.9), where αi corresponds with the

Page 218 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.3 Hessian based remeshing technique

barycentric coefficients and ωi the integration weight. In here where M (a + αi ab) is the metric at the i th Gauss point.
With the former expression, the concept of metric can be understood.

Z 1 k
X
p p
(6.9) ℓM (ab) = tab M (a + tab)abdt ≈ ωi tab M (a + αi ab) ab
0 i=1

The notion of length in a metric space is related to the notion of metric[ArtAF03] and therefore to an adequate
definition of the scalar product in the vector space considered. Define a metric tensor at a point P, respect an element
K , from a mesh Th it considering a matrix M (d × d) symmetric positive defined and not degenerated. In three
dimensions the we can consider (6.10), which can be assimilated to the analogy of an ellipsoid (Figure 6.3a).

 
a b c
M = b d e such that a > 0, d > 0, f > 0
(6.10) c e f
and det(M ) > 0, considering a, b, c, d, e ∈ R

The tensor M can be diagonalised because it is symmetrical. Then, M can be written M = RΛR −1 , where R
is the matrix of the eigenvectors and Λ the matrix of the eigenvalues of M . It should be noted that this decomposition
is not unique, indeed one can take any eigenvector of the subspace associated with one of the eigenvalues or it is
enough to invert columns of R and Λ.
To illustrate the effect of the metric on the actual mesh, we can observe the Figure 6.4, where the metrics present
on the nodes sketch the tetrahedra accordingly.

Figure 6.4: Effects of the metric on a tetrahedra

6.3.2.2 Metric intersection

Frequently, one is brought to work with several metrics. In the case where several metrics are specified at the same
point of the mesh, we try to define a single metric that takes into account all these metrics, that is to say to define the
metric given by the intersection of all these metrics.
To define the intersection of two metrics, we use the fact that a metric tensor is represented geometrically by a
two-dimensional ellipse or a three-dimensional ellipsoid. The metric intersection consists then in keeping the most
restrictive size constraint in all directions imposed by this set of metrics[ArtFA05] (see Figure 6.8b).
The simultaneous reduction enables to find a common basis (e1 , e2 and e3 ) such that M1 and M2 are congruent

Vicente Mataix Ferrándiz Page 219 of 374

6.3 Hessian based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

to a diagonal matrix, in this basis then N is introduced:

(6.11a) N = M1−1 M2

N can be diagonalised in R , because it is symmetrical in the metric M1 . The base in question is given by the
normalised eigenvectors of N that we note e1 , e2 and e3 (they form a base because N is diagonalisable). The
eigenvalues of M1 and M2 are found in this base using the Rayleigh quotient:

(6.11b) λi = eti M1 ei and µi = eti M2 ei

Considering P = (e1 e2 e3 ) be the matrix the columns of which are the eigenvectors of N , common basis of both
metrics, so it can diagonalise at the same time M1 and M2 .
 
λ1 0 0
M1 = P − t  0 λ2 0  P −1
0 0 λ3
(6.11c)  
µ1 0 0
M2 = P − t  0 µ2 0  P −1
0 0 µ3

Computing the metric intersection as:

 
max(λ1 , µ1 ) 0 0
(6.11d) M1∩2 = M1 ∩ M2 = P − t  0 max(λ2 , µ2 ) 0  P −1
0 0 max(λ3 , µ3 )

6.3.3 Hessian based metric measure

Taking as starting point the concepts introduced in 6.3.1.Error estimation, and the concepts of metrics from previous
sections, we are now ready to introduce the Hessian based metric (M ). We can summarise the following properties
introduced by Alauzet[ArtAF03]:

• The analysis and results obtained are not asymptotic, which means that the size of the mesh h is not tending to
zero. This solves some potential errors.

• It is based in the Hessian (H) of the solution.

• The obtained metric is anisotropic. This leads to reduce the number of degrees of freedom to solve the
problem as compared to the equivalent isotropic mesh, for the same level of accuracy[ArtFA05].

• It is independent of the nature of the operator, so it can be used with any type of equation.

6.3.3.1 Hessian theory

We can compute the Hessian[ArtWes+03] matrix (H) of a scalar variable f as proposed in Equation (6.12).

 
∂2f ∂2f
 ∂x 2 ···
 1 ∂ x1 ∂ xn 

 
 .. .. .. 
H=
 . . .  or just:

(6.12)
 
 
 ∂2f ∂2f 
···
∂ xn ∂ x1 ∂ xn2
∂2f
Hi,j =
∂ xi ∂ xj

Page 220 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.3 Hessian based remeshing technique

We remark the fact that the mathematical concept underlying in the consideration of the Hessian in mesh adaptation
has been already introduced in 6.3.1.3.Mesh adaptation. Once the Hessian matrix (H) has been computed, we can
compute the corresponding anisotropic metric by the following (6.13)[ArtFA05]. We highlight that for anisotropic
remeshing the resulting metric M , no additional operations are needed as the Hessian (H) already provides an
anisotropic mesh.

(6.13a) M = R t Λ̃t R with

(6.13b) Λ̃ = diag(λ̃i )

Being the eigenvalues from Equation (6.13) equivalent to the expression from Equation (6.13c). In here we will
consider both the maximum (hmax ) and minimum (hmin ) desired mesh sizes.

   
cd |λi | 1 1
(6.13c) λ̃i = min max , 2 , 2
ǫ hmax hmin

Being ǫ the error threshold and cd a constant ratio of a mesh constant and the interpolation ratio4 . For an isotropic
mesh the metric (Miso ) will be a diagonal matrix, where each element of the diagonal corresponds with the maximum
eigenvalue as seen in Equation (6.13d), where R corresponds with the eigenvectors matrix of M .

 
max(λ̃i ) 0 0
(6.13d) Miso = diag(max(λ̃i )) =  0 max(λ̃i ) 0 
0 0 max(λ̃i )

On the other hand, we can enforce a higher level of anisotropy using the anisotropic ratio (ρ) on the mesh in
consideration of a relative anisotropic radius (Rλrel ) as seen in the metric (Maniso ) in Equation (6.13e).

(6.13e) Maniso = R t Λ̃aniso R

Being the matrix Λ̃aniso equivalent to Equation (6.13f). In here we consider the minimum between the maximal
eigenvalue and the corresponding eigenvalue. This value is at the same time compared with the relative radius, where
the maximum between the two values is taken into consideration.

 
max(min(λ̃1 , λ̃max ), Rλrel ) 0 0
(6.13f) Λ̃aniso = 0 max(min(λ̃2 , λ̃max ), Rλrel ) 0 
0 0 max(min(λ̃3 , λ̃max ), Rλrel )

The relative radius then can be calculated using Equation (6.13g).

Rλrel = |λ̃max − Rλ | where:

(6.13g)
Rλ = (1 − ρ)|λ̃max − λ̃min |
4ǫ 2 9
can be taken as 10−6 as reference value. In the other hand cd can be considered as 9
for 2D cases and 32
for 3D cases.

Vicente Mataix Ferrándiz Page 221 of 374

6.3 Hessian based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

6.3.3.2 Numerical example

Figure 6.5: Initial mesh

The objective is to remesh the structured 1 mesh with the scalar function from Figure 6.5 and (6.14). The original
mesh has a high number of elements structured, our objective is to obtain an unstructured mesh where the smaller
elements will be around our objective function. The χ shaped function (6.14):

f (x, y ) = tanh(−100(y − 0.5 − 0.25 sin(2π x)))

(6.14)
+ tanh(100(y − x))

The results obtained can be seen on Figure 6.6b, here the smaller elements are around the χ shape that can be
seen in the Figure 6.6a showing the scalar function representation.

(a) Scalar function (b) New mesh

Figure 6.6: Solution for the example

Page 222 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.4 Level set based remeshing technique

6.4 Level set based remeshing technique

6.4.1 Theory
On the following section, the fundamentals of the level set technique for building a mesh metric are presented. The
concepts of the metric already presented in the previous section (6.3.2.Metric based remeshing) are taken into
account. We compute the gradient (6.15) of a scalar variable f in order to compute an anisotropic metric to remesh,
using the procedure from (6.16). On Figure 6.7 the gradient of a distance function is presented as an example. The
former example is shown in detail on the section 6.4.2.Numerical example.

∂f ∂f ∂f
(6.15) ∇f = i+ j+ k
∂x ∂y ∂z

Figure 6.7: Scalar and its gradient

We compute the following auxiliary coefficients (6.16), with them we can calculate the metric for the 2D cases
(6.16c) and for 3D cases (6.16d). Calling h the element size, ρ the anisotropic ratio, the metric M , f the scalar value
and ∇f the gradient from that scalar.

(
1.0
c0 = h2
Isotropic metric
(6.16a) c0
c1 = ρ2 Applying anisotropic ratio

∇f
(6.16b) M = c0 n ⊗ n + c1 (I − n ⊗ n), with n =
k∇f k

Then for 2D:

 
c0 (1 − ∇fx2 ) + c1 ∇fx2 (c1 − c0 )∇fx ∇fy
(6.16c) M =
(c1 − c0 )∇fx ∇fy c0 (1 − ∇fy2 ) + c1 ∇fy2

And for 3D:

 
c0 (1 − ∇fx2 ) + c1 ∇fx2 (c1 − c0 )∇fx ∇fy (c1 − c0 )∇fx ∇fz
(6.16d) M =  (c1 − c0 )∇fx ∇fy c0 (1 − ∇fy2 ) + c1 ∇fy2 (c1 − c0 )∇fy ∇fz 
2 2
(c1 − c0 )∇fx ∇fz (c1 − c0 )∇fy ∇fz c0 (1 − ∇fz ) + c1 ∇fz

Vicente Mataix Ferrándiz Page 223 of 374

6.4 Level set based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

6.4.2 Numerical example

The challenge consists in meshing anisotropically the geometry using as metric measure the gradient of the distance,
measured previously with an octree mesher GiD[OnlMel+16]. The geometry chosen is taken from the Stanford 3D
Scanning Repository, and it denominated Stanford bunny [BookRH12], see Figure 6.8a.

The mesh corresponding before remeshing corresponds with Figure 6.8b, the surrounding mesh to the isosurface
is relatively coarse and does not fit the isosurface geometry.

(a) Stanford’s bunny (b) Original octree mesh

Figure 6.8: Stanford’s Bunny example

On Figure 6.9 the solution obtained is presented. In Figure 6.9a the anisotropy distribution (ρ) is represented in
isosurfaces around the bunny. The resulting mesh, which fits the zero distance isosurface of the bunny, can be seen
at Figure 6.9b.

(a) Anisotropic ratio (b) Mesh after remeshing

Figure 6.9: Resulting mesh after remeshing

Page 224 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.5 SPR based remeshing technique

6.5 SPR based remeshing technique

6.5.1 Introduction
As we already mentioned previously, the SPR method was proposed by Zienkiewicz and Zhu in 1992[ArtZZ92], and
continued with later contributions[ArtZZ95]. In order to present the method, the underlying concepts related with error
measure and error estimation will be introduced. We remark that the basic concept of error estimation has been
already introduced in 6.3.1.Error estimation, here we will extend those definitions.

6.5.2 Theory
6.5.2.1 Error measures
The most basic error measure (evar ) would be to simply sustract the obtained solution in the FEA (uh ) to the exact
solution (u), this can be extended to any value we want to use as reference like stresses or strains (6.17). Nonetheless
these local error measures are in general not applicable for a refinement strategy as many singularities, like zero-size
elements, may occur. Because of that various error norms (kek2E ) representing an integral quantity are chosen to
measure the error[BookZZT13], Equation (6.18) presents the energy norms as an example. It is relevant to mention
that according to this concept, it is not always necessary for the mesh in the entire model to be refined. This can be
seen once we consider the Saint-Venants Principle, which enforces that the local stresses in one region do not affect
the stresses elsewhere.

e u = u − uh
(6.17) eε = ε − ε h
eσ = σ − σ h

The global energy norm can be defined as (6.18a).

Z
(6.18a) kek2E = (σ − σ h )T C−1 (σ − σ h ) d Ω
Ω

The sum of the local energy norm of each element (K ) contributions (6.18b).
X Z  X
(6.18b) kek2E = (σ − σ h )T C−1 (σ − σ h ) d Ω = kek2E,K
K ΩK K

6.5.2.2 Error estimation

Unfortunately, these errors measure is not possible to obtain in practice, as we do not know the exact solution most of
the time, so we need to calculate an estimation of the error (êvar ), in such a way that kêvar k ≈ kevar k. Usually an a
posteriori error estimators are used. These types of error estimators are calculated based on a reference solution
of the problem which is usually obtained by a computation on a coarse mesh. On the other hand, an a priori error
estimators estimate the error before a solution is calculated. Due to the nature of the present work, an a posteriori
error estimator will be considered.
The a posteriori error estimators can be divided on[BookAO11]:

• Residual based method: These methods can be divided on:

◦ Explicit residual methods: Explicit schemes involve a direct computation using available data. The
explicit residual method employs the residuals in the current approximation.
◦ Implicit residual methods: Implicit schemes involve the solution of an algebraic system of equations.
The error is estimated via the element residual method indirectly, it generally involves the solution of a
small linear algebraic system.

Vicente Mataix Ferrándiz Page 225 of 374

6.5 SPR based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

(a) FE approximation of uh (b) Gradient of uh (c) Recovered gradient

Figure 6.10: Construction of a recovered gradient in 1D. Inspired[BookAO11]

• Recovery based methods: In linear FE the stresses and strains are constant over the element, meaning that
the stresses field will be discontinous between elements. Taking these lower order approximations, recovery
methods generate a higher-order solution (see Figure 6.11). SPR method falls inside this category.

Each of the presented method has its own strength and weaknesses[BookAO11]. Explicit residual methods
require low computational effort, but requires the estimation of an unknown constant. Implicit residual methods require
the resolution of an auxiliary problem, therefore higher computational cost, and additionally the Neumman BC are
complex to take into account. On the other hand, recovery methods require low to very low computational costs, and
the only weakness is that no upper and lower error bounds exist.

6.5.2.3 Recovery based methods

These methods were originally based on postprocessing procedures. We want to compute the nodal recovered stress
values (σ̃ ∗ ) from the overall recovered stresses (σ ∗ ) using the standard shape functions (N) (6.19).

(6.19) σ ∗ = Nσ̃ ∗

It is assumed that the exact solution is approximated by the recovered solution sufficiently. In consequence we
can evaluate the energy norm error from 6.18 replacing the exact stress with σ ∗ . The resulting expression (6.20)
gives us an approximation of the energy norm.

Z
T
(6.20a) kek2E = (σ ∗ − σ h ) C−1 (σ ∗ − σ h ) d Ω
Ω

X Z T
 X
(6.20b) kek2E = (σ ∗ − σ h ) C−1 (σ ∗ − σ h ) d Ω = kek2E,K
K ΩK K

6.5.2.4 Superconvergence concept

The idea to derive these error indicators is based on the fact that many FE meshes have superconvergence
properties[BookRam01]. Superconvergent points are points of the solution field
where the error of the computed
values decreases faster than elsewhere and those errors have an order O hp+1 . This concept can be illustrated on
Figure 6.11, where the FE solution of the primary variable u is actually exact at the nodes while the FE solution of the
first derivative du
dx
is exact at the centre of the element. In this 1D case, the order of convergence is at least one order
higher (order p + 1) at the nodes for the primary variable and at the GP for secondary variables (order p) like stresses
or strains[ArtZZ92]. This property cannot be applied on general, but can be shown that for problems of the form
BT CBu = f with a differential deformation operator B and a symmetric constitutive tensor C this property holds.
By using a projection method, an improved stress field can be obtained from the superconvergent points. With
the superconvergent values, it is possible to smoothing of such values by a polynomial of order p within a patch of

Page 226 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.5 SPR based remeshing technique

(a) Optimal sampling points for the function (b) Optimal sampling points for the gradient
Figure 6.11: Optimal sampling points in 1D linear elements. Source[BookZZT13]

Figure 6.12: SPR example for linear triangles. Source[BookZZT13]

elements for which the number of sampling points can be taken as greater than the number of parameters in the
polynomial. For simplex linear elements5 , which are used in this work, it has been shown that the average value
of the derivatives at mid sides of adjacent elements are superconvergent[ArtChu80; ArtLev82] (see Figure 6.12).
Additionally in order to avoid the singularity on the LHS of the system we need to solve for the stress recovery, some
additional precautions must be taken into account on the corner elements, with an incomplete patch. One possible
way to deal with this issue is to calculate the recovered solution from a valid neighbour patch to the original patch. It is
relevant to highlight that the optimal sampling points are considered the centroidal points for the construction of a
recovered stress field[ArtZZ92], for further detail see[BookZZT13].

6.5.2.5 SPR calculation

Based on the procedure described in the previous section, we want to construct a solution field with superconvergent
properties in the whole domain from the superconvergent sampling points. In order to do so, we will apply a smoothing
of the stress values within an element patch considering a least-square polynomial fit. In the following, the concepts
and procedures important for SPR-based error estimation are presented in (6.21).

Defining the polynomial of order p that will be considered on the recover. Being ai is a set of unknown parameters
and p contains appropriate polynomial terms. x represents the coordinates respect the interior vertex node of the
patch. Then we can deduce (6.21a).

σ ∗i = pai 
(6.21a) p= 1 x y z ... zp
 T
ai = a1 a2 ... am

5 Linear 2D triangular elements, and the linear tetrahedra in extension for 3D.

Vicente Mataix Ferrándiz Page 227 of 374

6.5 SPR based remeshing technique CHAPTER 6. ADAPTATIVE REMESHING

Our objective is to minimise the following (6.21b).

n n
1 X 2 1 X 2
(6.21b) min Π = min σ h,i − σ ∗i = min σ h,i − pai
ai ai 2 a i 2
k=1 k=1

This minimisation defines the following system of equations with LHS and RHS.

LHSai =
PRHS i
n
(6.21c) LHS = Pk=1 p ⊗ p
n
RHSi = k=1 pσ h,i

The superconvergent nodal stresses may be computed by different patches adding and averaging its corresponding
contribution to the shared nodes.
Finally, it is important to analyse the singularities on the patch, like the corner nodes, where the LHS from (6.21c)
becomes singular. In order to solve that issue in [ArtZZ92] it is proposed to calculate the nodal superconvergent
stress values for boundary nodes from an interior patch.

6.5.2.6 Local mesh size

6.5.2.6.1 Compute h :
With the previous error computed, we are now able to compute the element size h. In order to do so we apply the
methodology from Zienkiewicz[BookZZT13], summarised in Equation (6.22).
The relative energy norm error (η ) should be below a certain threshold value (ηthreshold ). We can consider the error
estimated previously in all the following expressions. It is considered[BookZZT13] to be necessary to remesh in every
entity where η ≤ 1.

kekE kêkE
(6.22a) η= ≈ ≤ ηthreshold
kukE kukE

In the former expression, the total energy of deformation (kukE ) is expressed as (6.22b)6 .
Z
(6.22b) kuk2E = σC−1 σ d Ω
Ω

With this we can define the threshold error (kekthreshold ) as (6.22c).

q
2
(6.22c) kekthreshold = ηthreshold kukE = ηthreshold kuh kE + kek2E

We are interested on distribute this error uniformly across all the nK elements, then the threshold error for each
element (kekthreshold,K ) would be (6.22d).
s s
2 2
kuh k + kek2 kuh k + kêk2
(6.22d) kekthreshold,K = ηthreshold ≈ ηthreshold := eK
nK nK

With this we can define the refinement ratio for each element (ξK ) as in (6.22e). We can assume that the error per
element (kekK ) is proportional to the mesh size (h) to the polynomial power (6.22f). With the previous expression, and
considering the refinement ratio for each element (ξK ) the new element size (hnew ) can be defined as (6.22g).

keK k
(6.22e) ξK =
eK
(6.22f) kekK ∝ hp
h
(6.22g) hnew = √
p
ξK
6 C−1 implies that depending on the material considered, the method may not be applicable.

Page 228 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.6 Internal values interpolation

6.5.2.6.2 Compute respective metric :

The previously calculated element size must be transformed into the metric format, 6.3.2.Metric based remeshing,
in order to be used with the Mmg library. In order to do so, as the size obtained corresponds with an isotropic mesh it
will be as easy as a diagonal tensor, see (6.13d), such that we obtain the final expression from Equation (6.23).

1
(6.23) M = δij 2
hnew

6.5.3 Example
We can illustrate the method with the following example, Figure 6.13. In here we have a simple cantilever beam with a
distributed load. We start with the mesh from Figure 6.13a. Computing the energy error norm with the SPR the final
mesh from Figure 6.13b is obtained. We can appreciate that the error is reduced significantly from Figures 6.13c to
6.13d. Additionally the displacement changes significantly too from Figures 6.13e to 6.13f with the finer mesh (the
deformation factor is tweaked in order to fit better on these pages).

(a) Initial mesh (b) Final mesh

(c) Error (Initial mesh) (d) Error (Final mesh)

(e) Displacement (Initial mesh) (f) Displacement (Final mesh)

Figure 6.13: SPR example

6.6 Internal values interpolation

6.6.1 Theory
In order to recover the internal values information after remeshing, necessary in order to be able to work with
constitutive models that depend on historical values, as the plasticity models used on this work. The Figure 6.14
shows graphically how each one of the transfer operators work[ArtGJ06], we can be listed as (all of them available on
Kratos):

• CPT (Figure 6.14a): It just takes the value from the closest point. It provides acceptable results at low cost.

• SFT (Figure 6.14c): It interpolates the values using the standard FEM shape functions. Leads to an artificial
diffusion of the value, but preserves the original shape of the value profile.

Vicente Mataix Ferrándiz Page 229 of 374

6.7 Integration points values extrapolation CHAPTER 6. ADAPTATIVE REMESHING

(a) Closest Point Transfer (CPT) (b) Least-Square projection Transfer (LST)

(c) Shape Function projection Transfer (SFT)

Figure 6.14: Transfer operators. Source[ArtGJ06]

• LST (Figure 6.14b): It considers a least-square transfer across the closest points. Probably the most accurate
technique but computationally more expensive.

From the methodologies listed above the used in our simulations is the CPT, due to the fact that is the most
commonly used on the literature[ArtChi+12]. There are more available techniques, we address the work of Bus-
setta[ArtBBP] for alternative mesh transfers operators.

6.6.2 Example

(a) The problem of reference

(b) Original mesh (c) Resulting mesh

Figure 6.15: The beam problem of reference for the internal interpolation
The problem presented consists in a very simple beam of 10 × 1 as seen on Figure 6.15a. On this mesh we
consider a simple plasticity law, and we wan to see if we are able to preserve the plastic strain between the initial
mesh (Figure 6.15b) and the final mesh (Figure 6.15c).

The results we wan to interpolate are shown in Figure 6.16a. Considering the CPT method (Figure 6.16b) the
resulting plastic strain profile is quite close to the original one. On the other hand, we have the results for the LST
from Figure 6.16c, which solution is much smoother, but as previously stated this may lead to an artificial diffusion of
the target value.

6.7 Integration points values extrapolation

Page 230 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.7 Integration points values extrapolation

(a) Resulting values at the original mesh

(b) Interpolated results with CPT (c) Interpolated results with LST
Figure 6.16: The beam problem of reference for the internal interpolation

6.7.1 Introduction
In order to be able to consider some GP as a candidate variable to apply the level-set (6.4.Level set based remeshing
technique) or the Hessian (6.3.Hessian based remeshing technique) approach we need to transfer and extrapolate
the values from the integration points to the nodes of the mesh. The technique required to do so is presented in the
following.

6.7.2 Theory
Applying the same concepts considered for stress recovery[BookCA01], we can extrapolate the values from the
integration points. Taking into account the concept of the integration values can be defined with the standard shape
function definition and nodal values, Equation (6.24a). With this definition we can define a matrix operation which
will allow to extrapolate locally the values from the GP, as seen in Equation (6.24b). In here n corresponds with the
number of nodes in the element and m with the number of GP.

 
N1
  . 
(6.24a) value (ξ , η , ζ ) = value1 ··· valuen  .. 
Nn

    gp 
value1 N11 ··· N1m value1
..   .. .. ..   ..
(6.24b) = .
 
 . . .  . 
gp
valuen N1n ··· Nnm valuem

The recovered values from the integration points locally calculated for each element, generally exhibit jumps
between elements. In order to smooth these jumps, we can compute averaged nodal integration values. The two
main methods in order to do so are:

1. Unweighted averaging: Where we assign the same weight to all elements that meet at a node

2. Weighted averaging: The weight assigned to element contributions depends on a certain value. For example,
the integration weight of each GP, the element geometry or the element type

From these methods we will consider a weighted averaging based on the integration weights of each GP. In order
to do so, the corresponding average value at each node will be the one shown in Algorithm 6.

Vicente Mataix Ferrándiz Page 231 of 374

6.8 Adaptive remeshing methods applied on CCM CHAPTER 6. ADAPTATIVE REMESHING

Algorithm 6 Compute of nodal average value

1: procedure C OMPUTE OF NODAL AVERAGE VALUE
2: Reset the nodal average value on the mesh wnode
3: for all node ∈ Meshnodes do
4: wnode = 0
5: We add the contribution of each GP
6: for all element ∈ Meshelements do
7: for all GP ∈ element do
8: Compute Jacobian J, integration weight wgp and shape functions N
9: for all node ∈ element do
10: wnode + = Nnode Jwgp

In order to compute the smooth result, we just need to sum the corresponding contribution of each GP multiplied
by wnode , as seen in Equation (6.25).

elem gp
XX Nnode Jwgp
(6.25) valuenode = valuegp
wnode
i=0 j=0

6.7.3 Numerical example

The following example consists in a
very simple structural problem. I will
consist in a variable height cantilever
considering linear elastic plain strain
constitutive model. The parameters
considered are standard steel, see Ta-
ble 6.1. Figure 6.18 shows the solu-
tion obtained. On Figure 6.18a we
can see the VM stress before being
recovered, and in Figure 6.18b the so-
lution after consider the algorithm pre-
sented. Figure 6.17: Cantilever mesh and geometry

E ν ρ
2.069 · 1011 Pa 0.29 7850kg /m3

Table 6.1: Parameters considered for cantilever example

(a) VM stress non-extrapolated (b) VM stress extrapolated

Figure 6.18: Resulting VM stress before and after recover

6.8 Adaptive remeshing methods applied on CCM

In this section we will introduce the particularities necessaries in order to evaluate the remeshing techniques on the
contact problems. The discontinuity and continuous evolution of the boundary, the nonlinearity of the problem itself,

Page 232 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.8 Adaptive remeshing methods applied on CCM

introduce several challenges that must be taken into account in order to properly evaluate the problem.

Especially in the case of CCM, an adaptive refinement procedure is very beneficial: Particularly in the transition
between the contact zone and the non-contact zone, the exact solution might not be covered appropriately by the
FE-solution, for example due to a C 0 -continuity in the contact pressure. A fine mesh in this area gives the possibility to
cover the contact behaviour appropriately while the computational costs are kept low due to a coarse mesh in areas
where the solution behaves nicely and can be covered by fewer elements.

6.8.1 Level set metric

6.8.1.1 Theory

One first approach we can think in order to remesh the contact problem is the consideration of a level set, see
6.4.Level set based remeshing technique. This approach is adequate for fluid simulations were with this technique,
it is possible to remesh in the boundary layer of the fluid . This is commonly used in CFD[ArtSAS13], particularly
in embedded/immersed formulations, which is known to be crucial for accuracy. An equivalent to this last approach
could be considering the equivalent to the distance gradient in CCM, which would be the gap gradient. In order to
compute the consistent gap, we will consider the formulation proposed in 4.4.4.Penetration definition. In the following
numerical example, we will study if this approach fits our needs.

6.8.1.2 Numerical example

(a) Original mesh (b) Normal gap original

Figure 6.19: Hertz geometry considered. Example here
In order to study the suitability of the method, we can consider a relatively simple geometry so we can see if this
solution fulfils our problematic. We will consider an Hertz-like geometry from Figure 6.19. This geometry is commonly
considered as a benchmark in CCM, but additionally we can even compute analytically the gap (gn ). The initial mesh
is on Figure 6.19a and the gap representation in the Figure 6.19b.

(a) Resulting mesh (b) Normal gap solution

Figure 6.20: Solution obtained

Vicente Mataix Ferrándiz Page 233 of 374

6.8 Adaptive remeshing methods applied on CCM CHAPTER 6. ADAPTATIVE REMESHING

Applying the level-set technique the resulting mesh, Figure 6.20, shows us a finer mesh around the contacting
zone, but with an anisotropy that a priori does not benefit us. Therefore the we can conclude that the level-set
approach does not provide a good mesh estimation for contact problems.

6.8.2 Hessian metric

6.8.2.1 Theory

The next approach we want to consider in the CCM is the Hessian based remeshing procedure. This technique is
described in detail in the previous section 6.3.Hessian based remeshing technique. We want to consider several
variables which provide an adequate description of the contact problem so the mesh will be finer in the interest regions,
especially the boundary region that deals with the contact problem.

The variables that can be taken into account can be many, one of them may be the displacement, but this choice
does not provide a good information. In a generic CSD may be a good approach7 , as the displacement may be a
relevant variable which describes in a good manner the problem itself. But, for example, in a contact problem where
the both domains are already in contact this variable will not be very helpful to describe our problem. A variable more
associated to CCM is the contact pressure, but this one it is available only in the contacting interface, therefore no
affecting the whole domain, and not providing a full information of the problem. Due to the intersection capabilities
of the metric-based technique, see 6.3.2.2.Metric intersection, we can consider one or two additional variables
which will provide a full information of the domain. We think in two different variables which are natural to consider
on this problem, the VM equivalent stress and the strain-energy. They are natural in the sense that many CCM
problems involve plasticity too, and in consequence are descriptive of the whole problem. In summary, the contact
stress will provide us information about the contact boundary, and this information will be complemented with the VM
equivalent stress and the strain-energy. The example shown in 6.8.2.2.2.Punch test explores the consideration or not
consideration of the strain-energy in combination of the VM stress.

So we can consider the contact stress on both sides of the problem, not just in the slave domain, we must map the
values from the slave domain to the master domain. In order to do so, and trying to be the most consistent possible
with the formulation considered, we will apply the Mortar mapper developed in this work, see E.Mortar mapper. With
this, the domain will be remeshed equally, or proportionally, in each one of the contact domains. Additionally, the
VM equivalent stress must be extrapolated to the nodes in considering the method present in 6.7.Integration points
values extrapolation. In order to normalise these values, so the metric obtained is not so restrictive, we will consider a
normalisation factor which will depend on the Young modulus (E) and the Poisson ratio (ν ). These values are taken
into account in order to have a magnitude order to the stresses. The Equation (6.26) shows the normalisation factor
(fnorm ) considered in our formulations. This value has been obtained via numerical experimentation to get a proportion
of factors providing a reasonable value.

20
(6.26) fnorm =
ν2E

6.8.2.2 Numerical example

6.8.2.2.1 Simplest patch test :

The following is the simplest contact patch test possible, with only two blocks, and the displacements in the outer
lines being fixed in the x axis. The geometry and the mesh of the problem considered can be seen in Figure 6.21,
using triangular elements so we can remesh with Mmg. The continuous distribution of displacements in this problem
can be seen in Figure 6.21b. A contact patch test is a problem where the distribution of the stresses is constant. The
constant stresses can be observed on Figure 6.21c. The solution presented is interesting on this contest due to the
computed Hessian will be zero, as there is no variation on the stresses, Figure 6.21d, and therefore in theory the
obtained mesh with this method will be the original one, or the minimal sizes we impose on the resolution. Therefore
this is just an illustrative case to show in which cases the method will preserve the original mesh.

7 This is not the case for 6.10.2.Beam problem. Hessian of displacement, where the frequency corresponds with a highly oscillating beam.

Page 234 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.8 Adaptive remeshing methods applied on CCM

(a) Mesh (b) Displacement (c) VM stress (d) Hessian 2D

Figure 6.21: Simplest patch test

6.8.2.2.2 Punch test :

The numerical example will consist in a simple punch test in plane strain with a very coarse mesh. The problem is
remeshed considering an Hessian metric. The metric intersects the contact pressure, VM stress and additionally the
strain energy. We will evaluate 4 steps of ∆t = 0.5s, with a vertical displacement in the top face of uy = −0.01t. The
remesh will be performed each 2 steps. The material considered is steel for both domains (Table 6.2).

Domain E ν
Upper domain 2 · 1011 Pa 0.29
Lower domain 2 · 1011 Pa 0.29

Table 6.2: Parameters considered for the Hessian remesh punch test

(a) Original mesh (b) Displacement original (c) Strain energy (d) VM stress
Figure 6.22: Punch test geometry considered. Solution t = 0.5s. Example here
We will compare the solution between taking into account the strain energy and without it. As the initial mesh is
very coarse it does not provide much information, but in the later steps we can see as the mesh is refined around critics
points. The original mesh, as well as the displacement, strain energy and VM stress can be found in 6.22.
The Figure 6.23 shows the solution at t = 1.0s considering the Hessian metric for the contact stress and the VM
stress. This solution can be compared with the solution provided with the metric that also considers the strain energy
in the Figure 6.24. Both solutions are very similar, this is in part due to the fact that the first mesh is very coarse and
the information which provides is limited. Anyhow, there are some small differences in the distributions obtained, and
the ones which consider the strain energy provides a more detailed distribution in the solution achieved.
Solutions from Figures 6.25 and 6.26 show the results at t = 2.0s. In this case, as the previous mesh provided
much more information than the original one, the final meshes are better defined.

Vicente Mataix Ferrándiz Page 235 of 374

6.8 Adaptive remeshing methods applied on CCM CHAPTER 6. ADAPTATIVE REMESHING

(a) Solution mesh (b) Displacement solution (c) Strain energy (d) VM stress
Figure 6.23: Punch test solution for VM stress and contact stress Hessian at t = 1.0s

(a) Solution mesh (b) Displacement solution (c) Strain energy (d) VM stress
Figure 6.24: Punch test solution for strain energy, VM stress and contact stress Hessian at t = 1.0s

(a) Solution mesh (b) Displacement solution (c) Strain energy (d) VM stress
Figure 6.26: Punch test solution for strain energy, VM stress and contact stress Hessian at t = 2.0s

In both cases, Figures 6.25a and 6.26a show meshes which are finer in the contact boundary, as we must expect,
as well as the points where the displacement is imposed, and therefore the reactions are concentrated in that region.
Both meshes are quite similar, but the solutions obtained differ significantly, in the case of considering the strain
energy too, the solution profile of the VM stress and the energy is more detailed and closer to the expected solution.
In the other hand, the displacement solution, Figures 6.25b and 6.26b, converges quite fast in comparison with the
other variables, as expected. This faster convergence was mentioned previously in the SPR section, in Figure 6.11
from 6.5.2.2.Error estimation.
Here, we remember that the intersection concepts have always provided the most restrictive of all the provides
metrics, so the consideration of the additional variable always provides additional information without worsening the
previous solution. Of course, this does not mean we can intersect all the possible variables, as our interest is to define

Page 236 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.8 Adaptive remeshing methods applied on CCM

(a) Solution mesh (b) Displacement solution (c) Strain energy (d) VM stress
Figure 6.25: Punch test solution for VM stress and contact stress Hessian at t = 2.0s

a good mesh to solve our problem of interest, and so we should focus on those variables that are representative
of the phenomenon of study, as we have shown in this example. We could for example, consider in addition the
displacement, but as the displacement field is relatively homogeneous in this problem, and as the initial surfaces are
already in contact, the information which is provided is not very relevant.

6.8.3 SPR metric

6.8.3.1 Theory

The adapation of the method for CCM was first introduced by Wriggers[ArtWS98]. In order to do so, the stress
recovery on contact boundaries needs some special consideration[BookRam01]. This approach states that each
node on the slave contact boundary Γ1c (standard patch) is associated with the closest node in the master contact
boundary Γ2c (extended patches), as seen in Figure 6.27. The continuity requirements inside the patch system are
given by Equation (6.27). On these expressions (·)e and (·)s represent the external patch and the standard patch
respectively. (·)Γ represents that it is computed on the boundary.

Figure 6.27: SPR contact interface example for linear triangles. Inspired[ArtWS98; BookZZT13]

The normal component:

(6.27a) N PΓs as − PΓe ae = 0 on ΓC

The tangent component:

(6.27b) T PΓs as = 0 and T PΓe ae = 0 on ΓC

Vicente Mataix Ferrándiz Page 237 of 374

6.8 Adaptive remeshing methods applied on CCM CHAPTER 6. ADAPTATIVE REMESHING

The normal and tangent vectors, are a the transposition in Voigt notation8 of the following (6.27c).
   
nx2 nx ny nx nz τx nx τx ny + nx τy τx nz + nx τz
(6.27c) Nexpanded =  ny nx ny2 ny nz  and Texpanded =  τy nx + ny τx τ y ny τy nz + ny τz 
nz nx nz ny nz2 τz nx + nz τx τz ny + nz τy τ z nx

As a new contribution adapting to our formulation Wriggers[ArtWS98] work, the contact constraint is enforced with
the ALM method. Due to this the method SPR presented in Wriggers[ArtWS98] must be adapted in order to consider
Lagrange multipliers instead of the penalty method. The main difference is that the contact pressure is an unknown
of the system of equations instead of an estimated value from the current geometrical configuration. With this, the
contact boundary conditions are enforced at each node l of the contact boundary for the normal direction (6.28), for
the tangent direction is the same as in Equation (6.27b). In here λ̄n is the augmented contact pressure obtained from
the ALM formulation.
For the slave domain (standard patch):
(6.28a) N (xl ) PΓs (xl ) as − λ̄n (xl ) = 0 on Γ1C
For the master domain (extended patch):
(6.28b) N (xl ) PΓe (xl ) ae − (λ̄n ◦ χ) (xl ) = 0. on Γ2C
On this expression χ represents the mapping operation. This is done considering the mortar mapper developed (see
corresponding Appendix E.Mortar mapper) for a consistent computation of the normal gap (gn ) during the contact
search (4.4.Contact detection. Search techniques).
We must now minimise the expressions from Equation (6.29). It should be noticed that the patches on the standard
and extended side are not coupled. On these expressions, εn and ετ are the normal and tangent penalties, which do
not necessarily coincide with the ones taken into consideration on the ALM formulation.
For the slave domain (standard patch):
n m 
X
2
X  2  2 
(6.29a) min Π = min [σ h (xk ) − Ps (xk ) a] + εn NPΓs (xl ) a − λ̄n (xl ) + ετ TPΓs (xl ) a
ai ai
k=1 l=1

For the master domain (extended patch):

n m 
X X  2  2 
(6.29b) min Π = min [σ h (xk ) − Pe (xk ) a]2 + εn NPΓe (xl ) a − (λ̄n ◦ χ) (xl ) + ετ TPΓe (xl ) a
ai ai
k =1 l=1

The former, leads to the following equation system on Equation (6.30).

For the slave domain (standard patch):

LHSslave a = RHSslave
n
Xslave m
X slave


LHSslave = PTs (xk ) Ps (xk ) + εn PΓs T (xl ) NT NPΓs (xl ) + ετ PΓs T (xl ) TT TPΓs (xl )
(6.30a) k=1 l=1
n
Xslave m
X slave
 
RHSslave = Ps (xk ) σ h (xk ) + εn λ̄n (xl )
k=1 l=1

For the master domain (extended patch):

LHSmaster a = RHSmaster
nX
master mX
master


LHSmaster = PTe (xk ) Pe (xk ) + εn PΓe T (xl ) NT NPΓe (xl ) + ετ PΓe T (xl ) TT TPΓe (xl )
(6.30b) k=1 l=1
nX
master mX
master
 
RHSmaster = Pe (xk ) σ h (xk ) + εn (λ̄n ◦ χ) (xl )
k=1 l=1
8 As the Voigt notation is not standard we present the whole matrix form.

Page 238 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.8 Adaptive remeshing methods applied on CCM

6.8.3.2 Numerical examples

6.8.3.2.1 Simplest patch test :
We will consider the same example from 6.8.2.2.1.Simplest patch test in order
to show that with the proposed method, where the resulting mesh corresponds with
the original mesh. Following the same reasoning from that previous case, we can
calculate the corresponding error in the integration points. We expect that it will be
zero, Figure 6.28, as the stress is equal in all the points of the problem.
In conclusion, the mesh that will be obtained with this procedure will coincide again
with the original one. One more time, the patch test allows us to show the relationship
between stress distribution and the resulting metrics.

6.8.3.2.2 Punch test :

The following is a very simple punch test, the geometry is exactly the same shown
in the Hessian example (6.8.2.2.2.Punch test). The problem consists in one 2 × 1
lower block and one 1 × 1 upper block, with a very coarse mesh, as seen in Figure
6.29a. We impose a small vertical displacement of 0.1 m in the top of the upper block.
The Figure 6.29 shows the evolution of the mesh in each NL iteration, where the third
and fourth are almost identical, when the error has converged. The second mesh
(Figure 6.29b) is a very bad estimation, very fine a uniformly distributed, as the error
from the first coarser mesh (Figure 6.31a) does not provide much information. In the
following iterations (Figure 6.29c and 6.29d) we can appreciate as the mesh is finer
across the points with higher demand, as expected. Figure 6.28: Error obtained

(a) Initial mesh (b) First NL iteration (c) Second NL iteration (d) Third NL iteration
Figure 6.29: Mesh evolution

(a) Initial mesh (b) First NL iteration (c) Second NL iteration (d) Third NL iteration
Figure 6.30: Error solution
Additionally the error, Figure 6.30, on these NL iterations is significantly smaller than in the previous ones. But
more importantly, the mesh is more homogeneous, which is one of the objectives of the methodology to distribute the

Vicente Mataix Ferrándiz Page 239 of 374

6.9 Remeshing workflow CHAPTER 6. ADAPTATIVE REMESHING

error equally between the elements of the mesh.

(a) Initial mesh (b) First NL iteration (c) Second NL iteration (d) Third NL iteration
Figure 6.31: Displacement solution
Finally, as in the Hessian case from 6.8.2.2.2.Punch test, we see as the displacement converges faster than the
previous values (Figure 6.31). As we have already stated, this is expected, as seen in 6.5.2.2.Error estimation.

6.9 Remeshing workflow

The following summarises the simulation workflow followed during the contact simulations. We will differentiate
between the standard remeshing process considered during the Level set and the Hessian metrics, and the workflow
followed for the SPR metric computation, which are slightly different.

6.9.1 Level set and Hessian remeshing

(a) Standard problem (b) Contact problem

Figure 6.32: Workflow for Level set/Hessian metric remesh

Page 240 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.9 Remeshing workflow

We can differentiate between the standard problem and the contact problem.

6.9.1.1 Standard problem

For the standard case, we will compute the metric once the NL loop is converged, and the metric and the remesh will
be computed independently of the previous solution, this means that we will not check any value is below a certain
threshold. After remesh the information, the nodal and integration solutions, must be recovered. Once the solution
stage is computed we will need to re-initialise the solvers and processes if the problem has been remeshed in order
to restore the problem before execute the next time step.

6.9.1.2 Contact problem

The main difference between the standard problem and the contact one is that to avoid problematic and duplicated
conditions the associated conditions are removed and cleared, this will be recovered after re-initialise the processes.
In addition to that we will remap the contact stresses between domains and extrapolate the VM and the strain energy.
After that some additional information must be cleared, as the contact flags and the contact model parts.

6.9.2 SPR remeshing

(a) Standard problem (b) Contact problem

Figure 6.33: Workflow for SPR metric remesh
For the error estimation case the procedure is slightly different, not only because the error must be computed, but
also because if the error is below a certain threshold the remeshing is not executed.

6.9.2.1 Standard problem

The main difference respect 6.9.1.1.Standard problem is that once the NL loop has converged, we will need to
compute the recovered stress in the nodes, to later compute the estimated stress on the nodes. With this we will

Vicente Mataix Ferrándiz Page 241 of 374

6.10 Numerical examples CHAPTER 6. ADAPTATIVE REMESHING

compute the estimated error, if the error is below a certain threshold the time solution stage will be finished, if not with
the previous values we will estimate the size of the elements, and with it, the corresponding metrics.

6.9.2.2 Contact problem

Additionally to the previous consideration for the SPR error metric, we will consider the modifications from 6.9.1.2.Con-
tact problem, except that the extrapolation of integration values is not required and the mapping of the contact stresses
is performed before computing the error on the mesh.

6.10 Numerical examples

This section will show more advanced cases than the previously presented. The ones formerly introduced acted
as proof of concept to the concepts that were introduced in the respective sections. In this section we will focus in
structural cases, but we will dedicate an additional section to CFD cases, 6.11.CFD numerical examples, in order to
show up that this adaptive remeshing techniques are generic and can be considered in problems of different physical
nature.

6.10.1 Coarse sphere. Level set

In this problem we remesh using the gradient of the distance function,
which is the distance to the plane contained in the sphere centre.
The function can be seen in the Figure 6.34, which corresponds with
a linear distance and a constant gradient.
The solution obtained can be seen in Figure 6.35, where despite
the original coarse configuration, the final one is smoothed due to
the inner Bezier cubic surface considered by the Mmg library (see
B.3.1.What is Mmg and how does it work?).
It can be seen like the elements in the centre, where the dis-
tance function is zero, are the smaller ones, and the ones more
anisotropic.

Figure 6.34: Distance function

(a) Original mesh (b) Resulting mesh

Figure 6.35: Mesh before and after remesh for coarse mesh. Example

Page 242 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.10 Numerical examples

6.10.2 Beam problem. Hessian of displacement

The problem corresponds with a steel cantilever 10 × 1 and
considering as only load self-weight, see Figure 6.36. The
initial mesh considered is the coarse mesh from Figure 6.36.
We consider a simulation of 1s with ∆t = 0.01s. The whole
example can be found in here.
The final configuration, displacement and mesh can be Geometry of beam example
seen in Figure 6.37. Due to the high natural frequency of
the problem, the displacement is not a good measure for the
Hessian metric, as the high number of oscillations induce the
Mesh of beam example
resulting mesh in Figure 6.37. Where the distribution is not ho- Figure 6.36: Coarse mesh distance function
mogeneous, and does not follow any particular pattern.

Figure 6.37: Final configuration of the beam example

6.10.3 Hertz problem. Hessian of contact and VM stress

The problem consists in a Hertz contact problem remeshing considering the Hessian of the VM stress and the contact
pressure. We will consider a significant top pressure in order to maximise the contact area, which will help us to
evaluate the quality of the solution achieved, reason why the solution may look so deformed. We will not take into
consideration the strain energy,as the effect of this variable have been already taken into account in the punch test
example from 6.8.2.2.2.Punch test. The materials of this problem can be found in Table 6.3, where the base can be
considered rigid respect to the die.

Domain E ν
Upper domain 2 · 108 Pa 0.35
Lower domain 2 · 1011 Pa 0.29

Table 6.3: Parameters considered for the Hessian remesh Hertz example
There are 10 steps ∆t = 0.1s, and the remesh is executed each 3 steps starting in the 4th . The load applied
depends on time, and it is equal to q = 107 t Pa. The example can be located in Kratos repository.
The first step, Figure 6.38, is a very coarse mesh, but dense enough to provide the information needed in order to
estimate properly the required mesh. Looking at the VM stress from Figure 6.38c we can predict where the mesh will
become denser, this corresponds mostly with the contact interface. Particularly taking into account that we consider
the contact pressure in addition to the VM stress, this area will be especially enhanced. Anyhow, despite not being
terrible, the mesh provides an improved information about the VM equivalent stress.

(a) Mesh (b) Displacement (c) VM stress

Figure 6.38: Hertz solution step 1

Vicente Mataix Ferrándiz Page 243 of 374

6.10 Numerical examples CHAPTER 6. ADAPTATIVE REMESHING

The solution in the second mesh, Figure 6.39, refines the mesh in the required regions predicted previously. In
addition to the contact interface, the loaded face is also enhanced, this is expected as is the region of the problem
where the loads are concentrated.

(a) Mesh (b) Displacement (c) VM stress

Figure 6.39: Hertz solution step 4

The third mesh, Figure 6.40, is more defined than the previous one, with less anisotropic elements near the load
face. Additionally, as the contact zone is wider than in the previous case, the refined contact zone is extended.

(a) Mesh (b) Displacement (c) VM stress

Figure 6.40: Hertz solution step 7
Finally, we can notice that the final mesh it almost identical to the previous one, Figure 6.41. This is an indicative
that the current mesh already fits our requirements. In these two last meshes the VM stress is well represented after
the refinement.

(a) Mesh (b) Displacement (c) VM stress

Figure 6.41: Hertz solution step 10

6.10.4 Hertz problem. SPR error computing

Now we take the same problem from 6.10.3.Hertz problem. Hessian of contact and VM stress, but we will consider
the SPR metric instead of the Hessian. This will allow us to compare the solution obtained with both methods in
addition to the previous punch examples. On the first step, we can observe as the error is mainly concentrated on the
contact boundary, Figure 6.42c, this is what is expected like in the former Hessian case.
With the former error estimation, the obtained mesh is the one from Figure 6.43a. The elements are relatively
small in the contact zone, as the error was concentrated due to the initial coarse mesh. In this new step, the error is

Page 244 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.10 Numerical examples

(a) Mesh (b) Displacement (c) Error GP

Figure 6.42: Hertz solution step 1

quite uniform on the sphere, Figure 6.43c. Therefore, we can expect that the elements will become more uniform, and
become bigger, as the current mesh provides more information than the initial one.

(a) Mesh (b) Displacement (c) Error GP

Figure 6.43: Hertz solution step 4

(a) Mesh (b) Displacement (c) Error GP

Figure 6.44: Hertz solution step 7

(a) Mesh (b) Displacement (c) Error GP

Figure 6.45: Hertz solution step 10

For the third step, the mesh becomes more uniform, Figure 6.44a, as previously mentioned. In any case it
preserves the smaller elements in the contact interface and in the loaded face as it is supposed to be.

Vicente Mataix Ferrándiz Page 245 of 374

6.10 Numerical examples CHAPTER 6. ADAPTATIVE REMESHING

The last mesh, Figure 6.45, is quite similar to the previous step, as it is similar to the resulting mesh obtained
with the Hessian metric, Figure 6.41. So, even if the approaches are quite different both methods provide reasonable
results a relatively similar outcome.
One important difference with the Hessian solution is that the rigid base preserves the same mesh for all steps.
This makes sense, as the stress distribution is constant for a rigid media. This doe snot happens in the Hessian case
as we map the contact stress generating a variation on the contact boundary.

6.10.5 Contacting cylinders with adaptive remeshing

(a) 2D (b) 3D
Figure 6.46: Initial mesh in contacting cylinders

The problem consists in a two cylinders with a very coarse mesh, Figure 6.46, where we impose a horizontal
movement in the upper cylinder. This case is already presented in the contact chapter, 4.5.10.Contacting cylinders,
but preserving the same mesh. The cylinders are formulated as solids in TL framework with hyperelastic behaviour,
with the properties from Table 6.4. We impose a horizontal movement x = 0.2t at the upper cylinder, so the two
cylinders contact between them. The problem has been remeshed each 2.5e − 2s, and the total simulation time is
1.5s. In order to define the metric, the Hessian of the contact stress and the VM stress has been considered. This
example is located at the Kratos repository.

Domain E ν
Upper domain 2 · 1010 Pa 0.35
Lower domain 2 · 1010 Pa 0.35

Table 6.4: Parameters considered for the Hessian remesh contacting cylinders example
We present both the results from a 2D slice and a 3D perspective. The sliced section allows to better appreciate
the element size evolution, as well as the value distribution inside the geometry. We complement with the 3D
representation which allows us to picture the whole problem and the spatial variation in the values. In the following
we represent the solution each ∆t = 0.35s, so we have 4 steps fully represented. The values presented are the
displacement of the configuration, the element size and the VM stress.

(a) Displacement (b) Element size (c) VM stress

Figure 6.47: Contacting cylinders solution at t = 0.35s

Page 246 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.10 Numerical examples

(a) Displacement (b) Element size (c) VM stress

Figure 6.48: Contacting cylinders perspective solution at t = 0.35s

The solution obtained at t = 0.35s, Figures 6.47 and 6.48, shows like the region with the smaller elements is the
expected one, the contact region, where the values of the contact pressure and the VM stresses are concentrated.
This is even more notorious at t = 0.7s, Figures 6.49 and 6.50, where the contact zone is expanded and therefore
there are more elements reduced in the remeshing process. The last steps, from Figures 6.51, 6.52, 6.53 and 6.54,
are basically the opposite deformation from the previous ones, and the conclusions from these steps can be extended
here.

(a) Displacement (b) Element size (c) VM stress

Figure 6.49: Contacting cylinders solution at t = 0.70s

(a) Displacement (b) Element size (c) VM stress

Figure 6.50: Contacting cylinders perspective solution at t = 0.70s

Vicente Mataix Ferrándiz Page 247 of 374

6.10 Numerical examples CHAPTER 6. ADAPTATIVE REMESHING

(a) Displacement (b) Element size (c) VM stress

Figure 6.51: Contacting cylinders solution at t = 1.05s

(a) Displacement (b) Element size (c) VM stress

Figure 6.52: Contacting cylinders perspective solution at t = 1.05s

(a) Displacement (b) Element size (c) VM stress

Figure 6.53: Contacting cylinders solution at t = 1.4s

(a) Displacement (b) Element size (c) VM stress

Figure 6.54: Contacting cylinders perspective solution at t = 1.4s

Page 248 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.11 CFD numerical examples

6.11 CFD numerical examples

6.11.1 Cavity. Level set

This is a very simple remeshing example, where we remesh a cavity 1 × 1 uses the distance to the centre as
destination function. In Figure 6.55 it can be seen as the original mesh is homogeneous (Figure 6.55a) and the size
of the mesh is equal for all elements (Figure 6.55b). The resulting mesh has the smaller elements around the centre
of the square (Figure 6.55c) and the size is distributed depending on the closeness to the centre (Figure 6.55d). The
example can be found here.

(a) Original mesh (b) Original nodal size (c) Resulting mesh (d) Resulting nodal size
Figure 6.55: Mesh before and after remesh cavity case

6.11.2 Embedded cylinder. Level set

The problem consists in an anisotropic remeshing of a 2D fluid channel with a cylinder, using as level-set the distance
function. It consists in a channel 5 × 1, a cylinder of 0.3m diameter, Figure 6.56.

Figure 6.56: Embedded cylinder geometry

We start with a very refined mesh of 40000 nodes and it is simplified to a mesh of just 2500 nodes, centred around
the cylinder of interest, Figure 6.57. The example can be found in the Kratos Examples repository.

Vicente Mataix Ferrándiz Page 249 of 374

6.11 CFD numerical examples CHAPTER 6. ADAPTATIVE REMESHING

(a) Original mesh

(b) Resulting mesh

Figure 6.57: Mesh before and after remesh embedded case

6.11.3 Box isosurface example. Level set

In this case the challenge consists in meshing anisotropically the geometry using again the gradient of the distance.
The initial mesh has a large number of elements (Figure 6.58a), we want to reduce them (Figure 6.58b). Additionally
the remesher will consider the isosurface and we will remove the elements with a negative distance (see Figure 6.59).
The example can be found in the following link.

(a) Original mesh (b) Resulting mesh

Figure 6.58: Base mesh before and after remesh boxes with isosurface

(a) Original mesh (b) Resulting mesh

Figure 6.59: Mesh before and after remesh boxes with isosurface

Page 250 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.11 CFD numerical examples

6.11.4 Lamborghini. Level set

(a) Lamborghini (I) (b) Lamborghini (II)

Figure 6.60: Standard Triangle Language (STL) geometry for the Lamborghini
The problem corresponds with the Figure 6.60, the geometry is quite more complex than the previous cases
considered. The challenge consists in meshing anisotropically the geometry using the gradient of the distance,
previously measured with an octree mesher (GiD[OnlMel+16]). The idea behind this is to have a boundary layer
around the car, so in case of a CFD simulation the details on the aerodynamics study can be captured. The STL file
used and the rest of required files can be found here. The mesh corresponding before remeshing corresponds with
the one shown in Figure 6.61. In here we can appreciate like the mesh elements are isotropic, and does not fit the
profile of the car.

(a) Detail on original mesh (b) Distance function before remesh

Figure 6.61: Original mesh for the Lamborghini
The resulting mesh can be found in the Figure 6.62, where we see that the mesh follows the profile of the car, and
it does in an anisotropic way. Besides, it preserves the distance map from the original mesh.

(a) Detail on resulting mesh (b) Distance function resulting remesh

Figure 6.62: Resulting mesh for the Lamborghini

6.11.5 Channel CFD. Hessian of velocity

The following examples consist in two channels, one in 2D and the next one in 3D. In both cases it has been considered
the Hessian of the velocity to define the metric of the problem.

Vicente Mataix Ferrándiz Page 251 of 374

6.11 CFD numerical examples CHAPTER 6. ADAPTATIVE REMESHING

6.11.5.1 2D cylinder channel

This case consists in an adaptive remeshing of a 2D fluid channel with cylinder using Hessian of velocity as metric
measure. It consists in a channel 3 × 1, a cylinder of 0.5m diameter, see Figure 6.63a. Additionally the BC are the
following: An inlet velocity of 1m/s, wall conditions in the upper and lower boundary and a zero pressure in the outlet.
The total time of the simulation is 5s with a time step of ∆t = 0.01s. The remeshing is performed each 20 steps. The
mesh corresponding before remeshing corresponds with Figure 6.64a.

(a) Geometry (b) Mesh

Figure 6.63: Geometry and initial mesh in 2D channel. The example can be found here
The results obtained correspond with the following, Figure 6.64. In these figures we can appreciate like the mesh
fits the movement of the fluid, and particularly have refined the contour of the cylinder.

(a) Step 101 (b) Step 181

(c) Step 261 (d) Step 341

(e) Step 421 (f) Step 501

Figure 6.64: Solution in 2D channel
The mesh also captures the whirlpools of the flow and its adapt to them. In the other hand the inlet, which is a BC
and therefore the velocity is imposed, preserves practically the same mesh in each of the steps.

Page 252 of 374 Vicente Mataix Ferrándiz

CHAPTER 6. ADAPTATIVE REMESHING 6.11 CFD numerical examples

6.11.5.2 3D sphere channel

This problem stands as in the previous case in an adaptive remeshing

fluid channel with sphere using Hessian of velocity as metric measure,
but on 3D this time. It consists in a channel 3 × 1 × 1, a sphere of 0.5m
diameter and with a velocity of 1m/s in the inlet an zero pressure in
the outlet, the same as Figure 6.63a. The total time of the simulation is
1.2s with a time step of ∆t = 0.01s. The remeshing is performed each
20 steps. The initial mesh considered is the one from Figure 6.65. The
example can be found in the Kratos example repository.

The solution obtained is the one shown in Figure 6.66. On these

figures we represent the size of the elements as well as the contour
flow of the sphere. In the figures shown, we jump by 20 steps, except
in the last three pictures where we increase by 10.
Figure 6.65: Initial mesh in 3D channel

(a) Step 1 (b) Step 21 (c) Step 41

(d) Step 61 (e) Step 81 (f) Step 101

(g) Step 111 (h) Step 121 (i) Step 131

Figure 6.66: Solution in 3D channel

We can see that in the first steps, the size of the elements is similar and bigger than during the rest of the

Vicente Mataix Ferrándiz Page 253 of 374

6.11 CFD numerical examples CHAPTER 6. ADAPTATIVE REMESHING

simulation, especially in the downstream where the velocity where the fluid has not started to flow yet. Once the flow
starts to evolve, the elements start to be reduced. But it is significative that the stall of the flow keeps a relative small
size of the elements. The smaller elements are always the only surrounding the sphere. Finally, reached certain point
the flow around the sphere is fully formed.

Page 254 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookAO11] A posteriori error estimation in finite element analysis. Mark Ainsworth and J Tinsley Oden. John
Wiley & Sons. 2011.
[BookCA01] Introduction to finite element methods. Felippa C.A. 2001.
[BookRam01] Error-controlled adaptive finite elements in solid mechanics.
Ekkehard Ramm E. Rank R. Rannacher K. Schweizerhof E. Stein W. Wendland G. Wittum Peter
Wriggers Walter Wunderlich Erwin Stein. John Wiley & Sons. 2001.
[BookRH12] Virtual reality in medicine. Robert Riener and Matthias Harders. Springer Science & Business
Media. 2012.
[BookWri08] Nonlinear finite element methods. Peter Wriggers. Springer. 1st ed. 2008.
[BookZZT13] The Finite Element Method: its Basis and Fundamentals. O. C. Zienkiewicz, J.Z. Zhu, and
Robert L. Taylor. Butterworth-Heinemann. 7th ed. 2013.

Articles
[ArtAF03] “Estimateur d’erreur géométrique et métriques anisotropes pour l’adaptation de maillage. Partie I :
aspects théoriques”. Frédéric Alauzet and Pascal Frey. In: . No. RR-4759, 2003,
[ArtBBP] “Efficient 3D data transfer operators based on numerical integration”. Philippe Bussetta,
Romain Boman, and Jean-Philippe Ponthot. In: International Journal for Numerical Methods in
Engineering. No. 3-4, , pp. 892–929. Wiley Online Library. DOI: 10.1002/nme.4821.
[ArtBO94] “A methodology for adaptive mesh refinement in optimum shape design problems”. G. Bugeda and
E. Oñate. In: Computing Systems in Engineering. No. 1, Vol. 5, 1994, pp. 91–102. DOI:
10.1016/0956-0521(94)90040-X.
[ArtBow81] “Computing dirichlet tessellations”. Adrian Bowyer. In: The computer journal. No. 2, Vol. 24, 1981,
pp. 162–166. Oxford University Press. DOI: 10.1093/comjnl/24.2.162.
[ArtBR78] “A-posteriori error estimates for the finite element method”. Ivo Babuka and Werner C Rheinboldt.
In: International Journal for Numerical Methods in Engineering. No. 10, Vol. 12, 1978,
pp. 1597–1615. Wiley Online Library.
[ArtBR79] “Adaptive approaches and reliability estimations in finite element analysis”. I Babuka and
WC Rheinboldt. In: Computer Methods in Applied Mechanics and Engineering. Vol. 17, 1979,
pp. 519–540. Elsevier.
[ArtCas+97] “Anisotropic unstructured mesh adaption for flow simulations”. M. J. Castro-Daz, F. Hecht,
B. Mohammadi, and O. Pironneau. In: International Journal for Numerical Methods in Fluids. No. 4,
Vol. 25, 1997, pp. 475–491. DOI:
10.1002/(SICI)1097-0363(19970830)25:4<475::AID-FLD575>3.0.CO;2-6.
[ArtCea64] “Approximation variationnelle des problèmes aux limites”. Jean Cea. In: Annales de l’Institut
Fourier . No. 2, Vol. 14, 1964, pp. 345–444. Imprimerie Durand. 28 - Luisant. DOI:
10.5802/aif.181.

Vicente Mataix Ferrándiz Page 255 of 374

ARTICLES BIBLIOGRAPHY

[ArtChi+12] “An adaptive algorithm for cohesive zone model and arbitrary crack propagation”.
Vincent Chiaruttini, Dominique Geoffroy, Vincent Riolo, and Marc Bonnet. In: European Journal of
Computational Mechanics. No. 3-6, Vol. 21, 2012, pp. 208–218. Taylor & Francis. DOI:
10.1080/17797179.2012.744544.
[ArtChu80] “OPTIMAL POINTS OF THE STRESSES FOR TRIANGULAR LINEAR ELEMENT [J]”.
Chen Chuan-miao. In: Numerical Mathematics A Journal of Chinese Universities. Vol. 2, 1980,
[ArtCIA07] “Introduction à l’analyse numérique matricielle et à l’optimisation”. Philippe G CIARLET. In: . 2007,
DUNOD.
[ArtCor+19] “Combination of an adaptive remeshing technique with a coupled FEM–DEM approach for analysis
of crack propagation problems”. Alejandro Cornejo, Vicente Mataix, Francisco Zárate, and
Eugenio Oñate. In: Computational Particle Mechanics. 2019, pp. 1–18. Springer. DOI:
10.1007/s40571-019-00306-4.
[ArtFA05] “Anisotropic mesh adaptation for {CFD} computations”. P.J. Frey and F. Alauzet. In: Computer
Methods in Applied Mechanics and Engineering. No. 48–49, Vol. 194, 2005, pp. 5068–5082. DOI:
10.1016/j.cma.2004.11.025.
[ArtGJ06] “Plastic model with non-local damage applied to concrete”. Peter Grassl and Milan Jirásek. In:
International Journal for Numerical and Analytical Methods in Geomechanics. No. 1, Vol. 30, 2006,
pp. 71–90. Wiley Online Library. DOI: 10.1002/nag.479.
[ArtLev82] “Stress sampling points for linear triangles in the finite element method”. N Levine. In: Numerical
analysis report. Vol. 10, 1982, p. 82.
[ArtLo85] “A new mesh generation scheme for arbitrary planar domains”. SH Lo. In: International Journal for
Numerical Methods in Engineering. No. 8, Vol. 21, 1985, pp. 1403–1426. Wiley Online Library. DOI:
10.1002/nme.1620210805.
[ArtOB93] “A study of mesh optimality criteria in adaptive finite element analysis”. Eugenio Oñate and
G. Bugeda. In: Engineering Computations: Int J for Computer-Aided Engineering. No. 4, Vol. 10,
1993, pp. 307–321. DOI: 10.1108/eb023910.
[ArtOña+06] “Error estimation and mesh adaptivity in incompressible viscous flows using a residual power
approach”. Eugenio Oñate, Joaquin Arteaga, Julio Garca, and Roberto Flores. In: Computer
Methods in Applied Mechanics and Engineering. No. 4, Vol. 195, 2006, pp. 339–362. DOI:
10.1016/j.cma.2004.07.054.
[ArtPJR17] “Identifying combinations of tetrahedra into hexahedra: a vertex based strategy”. Jeanne Pellerin,
Amaury Johnen, and Jean-Francois Remacle. In: Procedia engineering. Vol. 203, 2017, pp. 2–13.
Elsevier. DOI: 10.1016/j.proeng.2017.09.779.
[ArtPVR18] “There are 174 Subdivisions of the Hexahedron into Tetrahedra”. Jeanne Pellerin, Kilian Verhetsel,
and Jean-Francois Remacle. In: . 2018, p. 266. DOI: 10.1145/3272127.3275037.
[ArtRRS95] “Automatic Mesh Generation for Plates and Shells”. Ernst Rank, Martin Rücker, and
Manfred Schweingruber. In: Proc. of 6th ICCCBE, Berlin. 1995,
[ArtSAS13] “A novel level set-based immersed-boundary method for CFD simulation of moving-boundary
problems”. Mukul Shrivastava, Amit Agrawal, and Atul Sharma. In: Numerical Heat Transfer, Part B:
Fundamentals. No. 4, Vol. 63, 2013, pp. 304–326. Taylor & Francis. DOI:
10.1080/10407790.2013.756258.
[ArtTWM82] “Boundary-fitted coordinate systems for numerical solution of partial differential equations—a
review”. Joe F Thompson, Zahir UA Warsi, and C Wayne Mastin. In: Journal of computational
Physics. No. 1, Vol. 47, 1982, pp. 1–108. Elsevier. DOI: 10.1016/0021-9991(82)90066-3.
[ArtWes+03] “Anisotropic mesh adaption Governed by a Hessian Matrix Metric”. Wilfried Wessner, Hajdin Ceric,
Clemens Heitzinger, Andreas Hossinger, and Siegfried Selberherr. In: . 2003,
[ArtWS98] “Different a posteriori error estimators and indicators for contact problems”. P Wriggers and
O Scherf. In: Mathematical and computer modelling. No. 4-8, Vol. 28, 1998, pp. 437–447. Elsevier.
DOI : 10.1016/S0895-7177(98)00133-2.

Page 256 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY ONLINE RESOURCES

[ArtYS84] “Automatic three-dimensional mesh generation by the modified-octree technique”. Mark A Yerry
and Mark S Shephard. In: International Journal for Numerical Methods in Engineering. No. 11, Vol.
20, 1984, pp. 1965–1990. Wiley Online Library. DOI: 10.1002/nme.1620201103.
[ArtZP71] “An automatic mesh generation scheme for plane and curved surfaces by
‘isoparametric’co-ordinates”. OC Zienkiewicz and DV Phillips. In: International Journal for
Numerical Methods in Engineering. No. 4, Vol. 3, 1971, pp. 519–528. Wiley Online Library.
[ArtZZ87] “A simple error estimator and adaptive procedure for practical engineerng analysis”.
Olgierd C Zienkiewicz and Jian Z Zhu. In: International journal for numerical methods in
engineering. No. 2, Vol. 24, 1987, pp. 337–357. Wiley Online Library. DOI:
10.1002/nme.1620240206.
[ArtZZ92] “The superconvergent patch recovery (SPR) and adaptive finite element refinement”.
OC Zienkiewicz and JZ Zhu. In: Computer Methods in Applied Mechanics and Engineering. No.
1-3, Vol. 101, 1992, pp. 207–224. Elsevier. DOI: 10.1016/0045-7825(92)90023-D.
[ArtZZ95] “Superconvergence and the superconvergent patch recovery”. OC Zienkiewicz and JZ Zhu. In:
Finite elements in analysis and design. No. 1-2, Vol. 19, 1995, pp. 11–23. Elsevier. DOI:
10.1016/0168-874X(94)00054-J.

Online resources
[OnlMel+16] GiD Homepage. URL: http://www.gidhome.com/;%20https:
//www.gidhome.com/support/brand-resources/citing-gid/.

Vicente Mataix Ferrándiz Page 257 of 374

ONLINE RESOURCES BIBLIOGRAPHY

Page 258 of 374 Vicente Mataix Ferrándiz

 Part II

Application

Vicente Mataix Ferrándiz Page 259 of 374

CHAPTER 7. APPLICATION

Chapter 7

Application

The hour has arrived to abandon

theories and go directly to what is
practical.

Samael Aun Weor

(1917 - 1977 AD, Writer)

7.1 Introduction
This chapter introduces several application test cases, integrating all the formulation and parts developed and
presented in this work. The cases presented are mainly taken from the work of Oñate[ArtOZ83]. From this work the
cases considered consist in a case with a cylinder punch and a spherical punch. In the original work, the solution
was obtained considering an axisymmetric viscous shell formulation when we will consider a full 3D J2 elasto-plastic
model simulation. The solution obtained will be compared with the reference presented in this work.

7.2 Cylinder punch

The geometry of this test is represented in the Figure 7.1. We have transformed the dimensions from the original work
in Imperial System to International System. The problem consists in a cylindrical punch, with a blank holder, a die and
a sheet of 8.96mm of thickness. In this case the frictional coefficient (µ) is the same in all the contactong interfaces,
and equal to 0.2.

Figure 7.1: Cylinder punch case

Figure 7.2 shows the mesh considered for the simulation of the cylinder punch case. In this case we have

Vicente Mataix Ferrándiz Page 261 of 374

7.3 Spherical punch CHAPTER 7. APPLICATION

considered the full geometry, instead of considering some simplification taking into account the symmetry of the
problem.

(a) Mesh (b) Mesh cut

Figure 7.2: Mesh considered in the cylinder punch case
In the Figure 7.3 we compare the solution obtained with the original experimental results and the numerical
solution obtained in the reference. Our solution obtained does not coincide with the experimental solution, but is
closer in comparison to the reference numerical solution.

Figure 7.3: Solution for cylinder punch test

7.3 Spherical punch

Figure 7.4 shows the geometry of the spherical punch case. The case consists in a spherical punch, with the
corresponding die, a blank holder and a metal sheet. In the original work, the problem was evaluated with three
different µ between the sheet and the punch, µ1 = 0.04, µ1 = 0.2 and µ1 = 0.5 in order to be compared with the

Page 262 of 374 Vicente Mataix Ferrándiz

CHAPTER 7. APPLICATION 7.3 Spherical punch

experimental result. Here we will just consider only µ1 = µ2 = 0.04, as in the original work the best solution compared
with the experimental was obtained for this case.

(a) 2D sketch with measurements (b) 3D representation

Figure 7.4: Spherical punch case
In this case, instead of a full 3D case, we have considered a quarter of the full geometry, as seen in the Figure
7.5a. The Figure 7.5 also shows the solution obtained at some points of the punch travel.

(a) Mesh (b) Solution uz = 11mm

(c) Solution uz = 16mm (d) Solution uz = 21mm

Figure 7.5: 3D solution for the case µ1 = 0.04 and µ2 = 0.04

The Figure 7.6 shows a comparison of the solution obtained with the original experimental results and the

Vicente Mataix Ferrándiz Page 263 of 374

7.3 Spherical punch CHAPTER 7. APPLICATION

numerical solution obtained in the reference for µ = 0.04. Our solution in certain points is closer to the experimental
result, but in some of the last steps the difference is greater.

Figure 7.6: Solution for spherical punch test

Page 264 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Articles
[ArtOZ83] “A viscous shell formulation for the analysis of thin sheet metal forming”. E. Onate and
O.C. Zienkiewicz. In: International Journal of Mechanical Sciences. No. 5, Vol. 25, 1983,
pp. 305–335. DOI: 10.1016/0020-7403(83)90011-5.

Vicente Mataix Ferrándiz Page 265 of 374

ARTICLES BIBLIOGRAPHY

Page 266 of 374 Vicente Mataix Ferrándiz

 Part III

Conclusions

Vicente Mataix Ferrándiz Page 267 of 374

CHAPTER 8. FINAL CONCLUSIONS

Chapter 8

Final conclusions

Both if you think you can do it

and if you dont, you are in the right

Henry T. Ford
(1863 - 1947 AD, Engineer,
Industrialist and Philanthropist)

8.1 Introduction
As can be seen in the previous chapters, the main objectives delimited for this doctorate work have been reached.
This means the understanding of the concepts necessary in order to simulate forming processes[BookBan10]. This
includes the implementation of a prismatic solid-shell element[ArtOF05; ArtFlo13a; ArtFlo13b] in order to properly
simulate the sheet behaviour without the drawbacks of a standard shell element, the proper implementation of a
computational contact mechanics algorithm[BookLau10; BookWri06; BookBel+14; ArtYL08; PhDPop12; PhDYas11],
the development of elastoplastic CL[BookNPO09] which allow to simulate the forming processes, and the addition
of adaptive remeshing techniques to the formula in order to perform the most accurate simulation possible while
preserving the minimal computational cost. All of this is not possible without a deep understanding of the FEM applied
in the resolution of highly non-linear problems[BookZZT13; BookZTF14; BookBHS07; BookWri08; BookOll14], and
how to tackle and couple all these non-linearities between them. In addition, reaching until this point this also means
that the structure of the Kratos code and C++ programming language have been mastered.
The objectives not reached in the present thesis, as well as other restlessness, are presented in the next chapter
9.Future works. Many significant left open after this work, as the metal forming processes and other similar industrial
processes involve problem of a very diverse typology, being very difficult to tackle all of them in just work.

8.2 Rotation free shells and solid-shell elements. Solid-shell

The prismatic solid-shell element implemented (3.4.Prismatic solid-shell) is fully functional, as can be seen in the
test cases shown in 3.5.Numerical examples. The element is basically based in the formulation of Flores[ArtFlo13c;
ArtFlo13a; ArtFlo13b], but modified in such a way that the methods of push-forward and pull-back can be used, see
A.1.Pull-Back, Push-Forward fundamental concepts, making the element fully compatible with the constitutive laws
implemented in Kratos. Focusing in the test cases from each one of them, we can conclude that:

• Cook’s membrane test: Figures 3.17 represent the behaviour of the Cook’s membrane, respectively elastic
and elasto-plastic, both obtained considering an implicit scheme. The results present even less than the
reference[ArtFlo13b] locking for the elastic case and similar behaviour for the elasto-plastic case.

Vicente Mataix Ferrándiz Page 269 of 374

8.3 Contact mechanics CHAPTER 8. FINAL CONCLUSIONS

• Open ended cylindrical shell test: The results (Figure 3.23) agree with the solution from [ArtSLL04].

• Slit test: Can be seen (Figure 3.27) that with the refinement of the mesh the solution tends to the correct one
[ArtSLL04].

• Panel test: The panel test is a common test which presents a highly non-linear behaviour, and it is necessary
an arc-length to compute. The results obtained, Figures 3.29, show a clear agreement with the reference
[ArtSLL04].

• Sphere test: The Figure 3.21 presents the results obtained for the semi-spherical shell with a ratio R /t = 1000,
which agrees with the results from [ArtSLL04].

• Cone-shell test: This problem is commonly considered to study the combination both plasticity and geometrical
non-linearity, the results shown in Figure 3.31 differ from [ArtKGW06], but with a close behaviour.

• Wrinkling test: This behaviour is originated owing to an instability, and needs the consideration of dynamic
(kinematic) forces for a correct simulation. The results obtained (Figure 3.33) are in agreement with the
reference[ArtFlo13b].

Additionally the element has been proved to be a good alternative to the traditional shells for a FSI simulation,
in 3.5.14.FSI-Vein test a simulation of an elastic vein is performed. The absence of rotational DOF as well as the
existence of two facets instead of one, differencing the inner and outer face of the structure interacting with the fluid,
have proved the suitability of the element for this kind of simulation.

8.3 Contact mechanics

The chapter 4.Contact mechanics contains the developments done in this thesis concerning the CCM. This chapter
constitutes the most relevant piece of work of all the existing document. The chapter starts introducing an historical
outline of the CCM, as well an introduction on the state-of-art of the contact techniques, compared within them, in
order to chose the one finally considered.
This implementation consists in state-of-art of the numerical contact mechanics formulation for implicit simulations,
combining the contributions of Alexander Popp[PhDPop12; ArtPop+10], Cavalieri and Cardona[ArtCC12; ArtCC13]
and the work of Yastrebov [PhDYas11; BookYas13], with off course several contributions of the author.
The formulation, 4.3.Formulation, rests in an exact mortar integration of the contact interface. This exact integration
in the FE geometry allows to obtain the most consistent integration possible between the integration domains, and
therefore the most exact solution possible. We have considered different optimisation algorithms, mainly the LMM,
the PM and particularly the consideration of a ALM with dual Lagrange multipliers, a new contribution of this work,
combining the contributions of Popp and Cavalieri. This dual LM allows us to condensate statically the system of
equations, taking off the LM of the resolution and therefore permitting the consideration of iterative solvers. The
formulation has been properly linearised, ensuring the quadratic convergence of the problem. As resolution strategy
we have considered a semi-smooth Newton method, which consists in an active set strategy, extensible also in the
case of frictional problems. The formulation works both for frictionless and frictional problems, the later essential for
the simulation of forming processes. In here we have presented the developments corresponding to Coulomb friction,
but giving the possibility to be extended to any type of frictional model.
In order to properly model the contact simulations, the algorithms of contact detection are detailed. Contact
detection is a very relevant subject in the relative of the convergence and resolution of the problem. In 4.4.Contact
detection. Search techniques, in first place the bounding volumes are presenting, highlighting the OBB in 4.4.2.1.OBB
implementation which is the technique mainly considered. The corresponding section continues with the introduction
of the tree structures (4.4.3.Tree structures) that can be considered in order to perform a search algorithm. In here
also the techniques considered in order to define a consistent gap (4.4.4.Penetration definition) and a self-contact
algorithm 4.4.5.Self-contact detection, both of them newly contributions of this work. Both techniques have been
proved successful with a series of tests included in the respective sections.
Respecting the tests, we have evaluated several test, most of them reference in the literature. The most relevant
tests can be summarised as follows:

Page 270 of 374 Vicente Mataix Ferrándiz

CHAPTER 8. FINAL CONCLUSIONS 8.4 Plasticity

• A basic patch test, 4.5.1.Basic patch test which allows us to evaluate both if the frictionless and frictional
formulation work as it supposed. This is expanded with the Taylor [ArtTP] patch test 4.5.2.Taylor patch test, and
it’s frictional equivalent 4.5.3.Friction base test.

• Then the Hertz[ArtHer82] test, the main contact reference problem is tested, 4.5.4.Hertz problem, both
frictionless and frictional cases. In here, the analytical solution is obtained in both cases.

• The following test is a new contribution of this work. It consists in a pure frictionless case where the energy must
be conserved. In here we deduce the analytical solution with the energy conservation principle and compare
with the solution obtained. The results agree with the analytical solution, with a very small error, that can be
attributed to the numerical dissipation of the time integration scheme. See 4.5.6.Energy conservation.

• The double arc benchmark 4.5.7.Double arc benchmark, is also a very extended solution in the literature. The
solution we obtained corresponds with the expected one, both in the frictionless and in the frictional case.

• The hyperelastic tubes, 4.5.9.Hyperelastic tubes are an interesting solution that includes self-contact. The
solution obtained is the expected one.

• The contacting cylinders are another literature reference problem. In 4.5.10.Contacting cylinders the solution
obtained is presented.

• The press fit is a frictional validation test, where we obtain almost the same solution as the reference. See
4.5.11.Press fit.

• Finally, the ironing tests from 4.5.12.Ironing punch, are commonly considered in the literature as validation tests.
Our solution corresponds with the one found in the literature.

The chapter concludes with the deduction of the directional derivatives, which are necessary for a proper
linearisation of the problem and obtain the desired quadratic convergence of the problem. The derivatives obtained
have been shown to converge quadratically as expected and necessary.

8.4 Plasticity
In 5.Plasticity the developments concerning the elasto-plastic models considered is presented. This chapter also
starts with a short introduction of the state-of-art of these constitutive models, as well as a very short historical
outline.

The main contribution of this work has been, the introduction a modular large strain elasto plastic model. This
is noted as modular, as it allows the combination of arbitrary of yield surfaces and plastic potential, the so-called
non-associated elasto-plastic CL. The implementation details of these modular classes are shown in 5.5.2.Class
structure in Kratos. This implementation takes as base the developments from Cornejo[PhDCor], and extends it in
order to deal with large strain behaviours.

In order to deal with the computation of the consistent tangent tensor on these non-associate constitutive laws, a
numerical approach has been considered in 5.5.3.Numerical implementation tangent constitutive tensor. In order
to obtain this consistent C we proceed with a numerical procedure based in the stress vector perturbation method.
Two different approaches are shown, one of first order and another one of second order, with its corresponding
computational cost, precision and convergence rate.

Additionally, the chapter presents some numerical examples in order to prove that the methodology works. Starting
from very simple cases of validation as the minimal cube considered in 5.6.2.Cube minimal example, where the CL
has been compared with a perfect J2 of reference, providing the correct solution. The next case tested consisted in a
more complex tensile test, from 5.6.3.Tensile test, where the solution obtained corresponds with the expected one,
but also showing a slightly mesh dependency in the solution. The final test consisted in a CCM test, 5.6.4.Application
in CCM. Gears example, consisting in two gears dealing with contact in a very small contact region, showing the
possibility of coupling these two kinds of phenomena simultaneously.

Vicente Mataix Ferrándiz Page 271 of 374

8.5 Adaptive remeshing CHAPTER 8. FINAL CONCLUSIONS

8.5 Adaptive remeshing

The chapter 6.Adaptative remeshing contains the developments done in the field of the adaptive remeshing. The
chapter includes an initial introduction and state-of-art of the field, that can result very practical in anyone been
introduced in the topic. Aside of this, we have shown different approaches, from more conventional SPR error
measures with isotropic remeshing, as seen in 6.5.SPR based remeshing technique; to approaches based in the
computation of the Hessian error measure, which in contrast provides an anisotropic mesh, seen in 6.3.Hessian
based remeshing technique. Both techniques have been compared, and have been proven to be very capable in
order to work with structural problems, but in the case of Hessian metric we are able to consider in a larger range of
cases.

Additionally both techniques have been adapted to the contact mechanics problem 6.8.Adaptive remeshing
methods applied on CCM, which is the main contribution of this work. The case of the SPR consists in the adaptation
of the previous work of Wriggers[ArtWS98] to the ALM contact formulation developed. The real novelty consists in the
Hessian error measure for CCM, where we have considered the intersection of the different contact related metrics,
like the contact pressure and the VM stress. Both techniques have ben shown effective, but the results got from the
Hessian technique show to be more promising.

In addition to the methods previously mentioned, techniques able to deal with both interpolation and extrapolation
of integration values have been implemented in order to accommodate some needs during the remeshing process.
These techniques are detailed respectively in 6.6.Internal values interpolation and 6.7.Integration points values
extrapolation.

In the relative of the test cases, the chapter includes an extensive list of tests, this includes structural problems from
6.10.Numerical examples and the CFD application examples from 6.11.CFD numerical examples. This last category
shows how the Hessian metric is general enough to deal with a different type of physical problems, particularly
interesting the channel cases from 6.11.5.Channel CFD. Hessian of velocity. The example more characteristic,
and the pride of this work is the adaptive remeshing case of the contacting cylinders in large deformations from
6.10.5.Contacting cylinders with adaptive remeshing.

The library considered in order to implement these remeshing techniques, Mmg, is introduced in its corresponding
Appendix B.3.Mmg library.

8.6 Application cases

In the final chapter 7.Application, which additionally corresponds with the final part of this work. What this chapter
shows is that an initial capability to perform metal forming processes is proven. The combination of the former points
have allowed to perform these simulations. Additionally, as short summary of the chapter, the results obtained differ
slightly with the results present in the main reference[ArtOZ83], but in the comparison with the experimental results
the solution presented resembles more to the latter.

8.7 Other remarks

In here we will present others remarks, that not constitute the main subject of this work, but we consider important to
append to what have been said until this point, in order to have a more accurate picture of what has been done and
achieved.

8.7.1 Other developments

The appendices include different developments tangent to main developments already presented. For example,
from the mortar formulation considered CCM problems can be considered in order to compute mesh tying problems
A.3.Mesh tying, or an explicit mapping algorithm (E.Mortar mapper). This mapping algorithm has allowed to compute
a consistent gap with the mortar formulation considered during the contact pair search. Also the details of the mortar
segmentation algorithms considered can be found in A.2.Mortar segmentation.

Page 272 of 374 Vicente Mataix Ferrándiz

CHAPTER 8. FINAL CONCLUSIONS 8.7 Other remarks

In order to obtain a consistent formulation for the contact mechanics problems, meaning with quadratic conver-
gence, the AD[ArtGru82; ArtKor97; ArtKor02; BookWri08] methodology has been considered. This methodology is
presented in an independent Appendix, see C.Automatic differentiation.
Also, as a complement, which helps in a proper understanding of the constraint optimisation considered in the
contact mechanics formulations, a complete Appendix has been considered, D.Constrained optimisation problems.
Where the mathematical concepts and details of these methodologies are presented.

8.7.2 Implicit approach

All the simulations and tests have been performed with an implicit approach, as one of the mains aims of this research
is to focus in this kind of solution. Some of the problems computed in this work are highly nonlinear, especially
due to the challenging nonlinearities associated with the computational contact and elastoplastic constitutive laws.
This has also implied the necessity of improving the implementation of algorithms related with the implicit resolution
of the system of equations. This includes the improvement of the NR with the inclusion of active-set strategies,
necessaries in order to properly consider the contact problem in an implicit manner. Other developments not included
in this work, but that have helped to improve the implicit approach include the implementation of different constraints
imposing techniques with the consideration of MultiPoint Constraint (MPC), as well as different systems of equations
assemblers.

8.7.3 Programming
As a final remark, and non-necessarily related with the subject treated in this work, we find relevant to highlight that in
order to achieve the results and solutions presented the author has learnt to deal with collaborative development for the
Kratos project. The structure and fundaments of the Kratos code has been understood, but also the understanding
the workflow of a version control software as git is. The main characteristics of Kratos are introduced in its respective
Appendix B.2.Kratos Multiphysics. The author has become an important member of the Kratos community, and we
feel the need to mention.

Vicente Mataix Ferrándiz Page 273 of 374

8.7 Other remarks CHAPTER 8. FINAL CONCLUSIONS

Page 274 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookBan10] Sheet Metal Forming Processes: Constitutive Modelling and Numerical Simulation. Dorel Banabic.
Springer-Verlag Berlin Heidelberg. 1st ed. 2010.
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookBHS07] Encyclopedia of Computational Mechanics (3 Volume Set). René de Borst, T.J.R. Hughes, and
Erwin Stein. Wiley. 1st ed. 2007.
[BookLau10] Computational Contact and Impact Mechanics: Fundamentals of Modeling Interfacial Phenomena
in Nonlinear Finite Element Analysis. Tod A. Laursen. Springer. 2010.
[BookNPO09] Computational Methods for Plasticity Theory and Applications. E. A. de Souza Neto, D. Periæ, and
D.R.J. Owen. Wiley. 2009.
[BookOll14] Nonlinear Dynamics of Structures. Sergio Oller. Springer International Publishing. 2014.
[BookWri06] Computational Contact Mechanics. Peter Wriggers. Springer. 2nd. 2006.
[BookWri08] Nonlinear finite element methods. Peter Wriggers. Springer. 1st ed. 2008.
[BookYas13] Numerical Methods in Contact Mechanics. Vladislav A. Yastrebov. Wiley-ISTE. 1st ed. 2013.
[BookZTF14] The Finite Element Method for Solid and Structural Mechanics. O. C. Zienkiewicz, Robert L. Taylor,
and David Fox. Butterworth-Heinemann. 7th ed. 2014.
[BookZZT13] The Finite Element Method: its Basis and Fundamentals. O. C. Zienkiewicz, J.Z. Zhu, and
Robert L. Taylor. Butterworth-Heinemann. 7th ed. 2013.

Articles
[ArtCC12] “An augumented lagrangian method to solve three-dimensional nonlinear contact problems”.
FJ Calavalieri and Alberto Cardona. In: Latin American applied research. No. 3, Vol. 42, 2012,
pp. 281–289. SciELO Argentina.
[ArtCC13] “An augmented Lagrangian technique combined with a mortar algorithm for modelling mechanical
contact problems”. FJ Cavalieri and A Cardona. In: International Journal for Numerical Methods in
Engineering. No. 4, Vol. 93, 2013, pp. 420–442. Wiley Online Library. DOI: 10.1002/nme.4391.
[ArtFlo13a] “A “Prism” solid element for large strain shell analysis”. Fernando G. Flores. In: Computer Methods
in Applied Mechanics and Engineering. Vol. 253, 2013, pp. 274–286. Elsevier. DOI:
10.1016/j.cma.2012.10.001.
[ArtFlo13b] “Development of a non-linear triangular prism solid-shell element using ANS and EAS techniques”.
Fernando G. Flores. In: Computer Methods in Applied Mechanics and Engineering. Vol. 266, 2013,
pp. 81–97. Elsevier. DOI: 10.1016/j.cma.2013.07.014.
[ArtFlo13c] “UN ELEMENTO PRISMA TRIANGULAR DE SÓLIDO-LÁMINA PARA EL ANÁLISIS CON
GRANDES DEFORMACIONES”. Fernando G. Flores. In: Mecánica Computacional. Vol. XXXII,
2013, pp. 63–87.

Vicente Mataix Ferrándiz Page 275 of 374

PH.D.’S THESIS BIBLIOGRAPHY

[ArtGru82] “Automatic Differentiation: Techniques and Applications. Lecture Notes in Computer Science 120.”
F. Grund. In: ZAMM - Journal of Applied Mathematics and Mechanics / Zeitschrift für Angewandte
Mathematik und Mechanik . No. 7, Vol. 62, 1982, pp. 355–355. DOI:
10.1002/zamm.19820620735.
[ArtHer82] “Über die berührung fester elastische Körper und über die Harte”. H Hertz. In: Verhandlungen des
Vereins zur Beförderung des Gewerbefleisses. 1882,
[ArtKGW06] “A robust non-linear solid shell element based on a mixed variational formulation”. Sven Klinkel,
Friedrich Gruttmann, and Werner Wagner. In: Computer Methods in Applied Mechanics and
Engineering. No. 1–3, Vol. 195, 2006, pp. 179–201. Elsevier. DOI:
10.1016/j.cma.2005.01.013.
[ArtKor02] “Multi-language and Multi-environment Generation of Nonlinear Finite Element Codes”. J. Korelc.
In: Engineering with Computers. No. 4, Vol. 18, 2002, pp. 312–327. DOI:
10.1007/s003660200028.
[ArtKor97] “Automatic generation of finite-element code by simultaneous optimization of expressions”.
Joe Korelc. In: Theoretical Computer Science. No. 1, Vol. 187, 1997, pp. 231–248. DOI:
10.1016/S0304-3975(97)00067-4.
[ArtOF05] “Advances in the formulation of the rotation-free basic shell triangle”. Eugenio Oñate and
Fernando G. Flores. In: Computer Methods in Applied Mechanics and Engineering. No. 21–24, Vol.
194, 2005, pp. 2406–2443. DOI: 10.1016/j.cma.2004.07.039.
[ArtOZ83] “A viscous shell formulation for the analysis of thin sheet metal forming”. E. Onate and
O.C. Zienkiewicz. In: International Journal of Mechanical Sciences. No. 5, Vol. 25, 1983,
pp. 305–335. DOI: 10.1016/0020-7403(83)90011-5.
[ArtPop+10] “A dual mortar approach for 3D finite deformation contact with consistent linearization”.
Alexander Popp, Markus Gitterle, Michael W. Gee, and Wolfgang A. Wall. In: International Journal
for Numerical Methods in Engineering. No. 11, Vol. 83, 2010, pp. 1428–1465. John Wiley & Sons,
Ltd. DOI: 10.1002/nme.2866.
[ArtSLL04] “Popular benchmark problems for geometric nonlinear analysis of shells”. K.Y. Sze, X.H. Liu, and
S.H. Lo. In: Finite Elements in Analysis and Design. No. 11, Vol. 40, 2004, pp. 1551–1569. DOI:
10.1016/j.finel.2003.11.001.
[ArtTP] “On a patch test for contact problems in two dimensions”. Robert L. Taylor and
Panagiotis Panagiotopoulos. In: . ,
[ArtWS98] “Different a posteriori error estimators and indicators for contact problems”. P Wriggers and
O Scherf. In: Mathematical and computer modelling. No. 4-8, Vol. 28, 1998, pp. 437–447. Elsevier.
DOI : 10.1016/S0895-7177(98)00133-2.

[ArtYL08] “A contact searching algorithm including bounding volume trees applied to finite sliding mortar
formulations”. Bin Yang and TodA. Laursen. In: Computational Mechanics. No. 2, Vol. 41, 2008,
pp. 189–205. Springer-Verlag. DOI: 10.1007/s00466-006-0116-z.

Ph.D.’s thesis
[PhDCor] “Coupled formulation between the finite element method and the discrete element method for
studying multi-fracture processes in solids and structures”. Alejandro Cornejo.
[PhDPop12] “Mortar Methods for Computational Contact Mechanics and General Interface Problems”.
Alexander Popp. 2012.
[PhDYas11] “Computational contact mechanics: geometry, detection and numerical techniques”.
Vladislav A. Yastrebov. 2011.

Page 276 of 374 Vicente Mataix Ferrándiz

CHAPTER 9. FUTURE WORKS

Chapter 9

Future works

The future belongs to those who

believe in the beauty of their dreams

Eleanor Roosevelt
(1884 - 1962 AD, American political
figure, diplomat and activist)

In this chapter we present the future developments that will follow to this PhD’ thesis. The following are ideas that
due to the time constraints, and to limit the study of this work, have not been included in this final work. We will split,
the future works, relating with each one of the subjects tackled in these pages.

9.1 Solid-shell implementations

• Adapt the element from 3.4.Prismatic solid-shell, in order to properly support the UL framework.

• The reduced integration hexahedral solid-shell element from Flores[ArtFlo16] could be a good element in order
to complement the element presented on this work in 3.4.Prismatic solid-shell.

9.2 Contact mechanics

• The implementations presented on this work are mainly focused on serial implementations and only OpenMP.
In order to make these implementations truly High Performance Computing (HPC) it is necessary to extend
the formulation to Message Passing Interface (MPI), particularly using the Trilinos API exposed in Kratos.

• Improve the robustness of the method with iterative solvers, particularly with AMG solvers, that Kratos makes
use of.

• Trying to bring to Kratos the latest implementations emerged in the literature, like the matrix-symmetric contact
problem presented in [ArtHWP18].

• Extend the formulations in order to consider different formulations. Like the Material Point Method (MPM) or
the Iso-Geometric Analysis (IGA)[ArtLWZ12a; ArtLWZ12b; ArtSei+16] formulation. These formulations have
already initial implementations of contact constraints in the current Kratos.

• Extend the contact formulation to other types of elements, like shells or beams[ArtSN17], and combinations of
these elements between them.

• Clean up, and refactor of the code in Kratos. Particularly reduce the compilation and linking times.

Vicente Mataix Ferrándiz Page 277 of 374

9.3 Plasticity CHAPTER 9. FUTURE WORKS

• Consideration of the contact formulation, and couple it with problems of another nature like FSI.
• Extension and modularisation of the frictional models, in order to be able to consider general frictional models
of any type.

• Application to optimisation problems[Bookaut; BookAll07].

9.3 Plasticity
• The definition of thermomechanical behaviour. The definition of thermic deformations, as well as others
thermomechanical behaviours and couplings. Particularly focused in large deformation formulations[ArtLub04].

• The consideration, and extension, of general and large deformation anisotropic material behaviours. This is
necessary in order to properly simulate som metal stamping phenomena[ArtCJL02; ArtGW02].

• Coupling of Kratos with the code MFront[ArtHPF15].

• Extend the testing and validation of the CL in the Kratos framework.

9.4 Adaptive remeshing

• Extension of the adaptive remeshing techniques to HPC and MPI.
• Application of the adaptive remeshing techniques developed on this work in problems of different nature and
couplings. For example, consider it in order to remesh the background mesh present in the MPM formulation.

• Enhance current implementations in order to remesh element types not supported currently, like hexahedra.

9.5 Metal stamping

• Improve and extend the implementations focused on metal stamping simulations.
• Validate the code considering more extended benchmarks.

9.6 Other points

• Refactor, clean up and improvement of the existing code of Kratos in all levels.
• Extension of the current AD[ArtKor97; ArtKor02] implementations, in order to simplify its use and its generality.
It would be nice to apply this technique, not only in the derivation of element, but the definition of CL or frictional
laws.

• Extend and improve the explicit formulation that was developed and dropped during the development of this
work.

• Extend the current Graphic User Interface (GUI) implementations.

• Extend Kratos integration with new libraries and features.

Page 278 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookAll07] Conception optimale de structures (Mathématiques et Applications). Grégoire Allaire. Springer. 1st
ed. 2007.
[Bookaut] Shape Optimization by the Homogenization Method. Grégoire Allaire (auth.) Springer-Verlag New
York. 1st ed.

Articles
[ArtCJL02] “Modeling the Bauschinger effect for sheet metals, part I: theory”. B.K. Chun, J.T. Jinn, and
J.K. Lee. In: International Journal of Plasticity . No. 5–6, Vol. 18, 2002, pp. 571–595. DOI:
10.1016/S0749-6419(01)00046-8.
[ArtFlo16] “A simple reduced integration hexahedral solid-shell element for large strains”. Fernando G. Flores.
In: Computer Methods in Applied Mechanics and Engineering. Vol. 303, 2016, pp. 260–287. DOI:
10.1016/j.cma.2016.01.013.
[ArtGW02] “Role of plastic anisotropy and its evolution on springback”. Lumin Geng and R.H. Wagoner. In:
International Journal of Mechanical Sciences. No. 1, Vol. 44, 2002, pp. 123–148. DOI:
10.1016/S0020-7403(01)00085-6.
[ArtHPF15] “Implantation de lois de comportement mécanique à l’aide de MFront: simplicité, efficacité,
robustesse et portabilité”. Thomas Helfer, Jean-Michel Proix, and Olivier Fandeur. In: . 2015, Giens,
France.
[ArtHWP18] “A truly variationally consistent and symmetric mortar-based contact formulation for finite
deformation solid mechanics”. Michael Hiermeier, Wolfgang A. Wall, and Alexander Popp. In:
Computer Methods in Applied Mechanics and Engineering. Vol. 342, 2018, pp. 532–560. DOI:
10.1016/j.cma.2018.07.020.
[ArtKor02] “Multi-language and Multi-environment Generation of Nonlinear Finite Element Codes”. J. Korelc.
In: Engineering with Computers. No. 4, Vol. 18, 2002, pp. 312–327. DOI:
10.1007/s003660200028.
[ArtKor97] “Automatic generation of finite-element code by simultaneous optimization of expressions”.
Joe Korelc. In: Theoretical Computer Science. No. 1, Vol. 187, 1997, pp. 231–248. DOI:
10.1016/S0304-3975(97)00067-4.
[ArtLub04] “Constitutive theories based on the multiplicative decomposition of deformation gradient:
Thermoelasticity, elastoplasticity, and biomechanics”. Vlado A Lubarda. In: Applied Mechanics
Reviews. No. 2, Vol. 57, 2004, pp. 95–108. DOI: 10.1115/1.1591000.
[ArtLWZ12a] “A mortar formulation for 3D large deformation contact using NURBS-based isogeometric analysis
and the augmented Lagrangian method”. L. de Lorenzis, Peter Wriggers, and G. Zavarise. In:
Computational Mechanics. No. 1, Vol. 49, 2012, pp. 1–20. Springer. DOI:
10.1007/s00466-011-0623-4.

Vicente Mataix Ferrándiz Page 279 of 374

ARTICLES BIBLIOGRAPHY

[ArtLWZ12b] “A mortar formulation for 3D large deformation contact using NURBS-based isogeometric analysis
and the augmented Lagrangian method”. L. De Lorenzis, Peter Wriggers, and G. Zavarise. In:
Computational Mechanics. No. 1, Vol. 49, 2012, pp. 1–20. Springer-Verlag. DOI:
10.1007/s00466-011-0623-4.
[ArtSei+16] “Isogeometric dual mortar methods for computational contact mechanics”. Alexander Seitz,
Philipp Farah, Johannes Kremheller, Barbara I. Wohlmuth, Wolfgang A. Wall, and Alexander Popp.
In: Computer Methods in Applied Mechanics and Engineering. Vol. 301, 2016, pp. 259–280. DOI:
10.1016/j.cma.2015.12.018.
[ArtSN17] “Woven fabrics computational simulation using beam-to-beam contacts formulation”.
Mauro Takayama Saito and Alfredo Gay Neto. In: Revista Interdisciplinar De Pesquisa Em
Engenharia. No. 22, Vol. 2, 2017, pp. 09–25.

Page 280 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY ARTICLES

Vicente Mataix Ferrándiz Page 281 of 374

ARTICLES BIBLIOGRAPHY

Page 282 of 374 Vicente Mataix Ferrándiz

 Appendices

Vicente Mataix Ferrándiz Page 283 of 374

APPENDIX A. THEORETICAL COMPLEMENTS

Appendix A

Theoretical complements

The devil is in the detail.

Ludwig Mies van der Rohe

(1886 - 1969 AD, German-born
architect)

A.1 Pull-Back, Push-Forward fundamental concepts

In this appendix we introduce the concepts of pull-back and push-forward necessaries to understand the concepts
introduced previously, the main reference for this has been taken from [BookBel+14]. These operations allow us
to have a unified description between the Eulerian and Lagrangian tensors. In (A.1) we can appreciate some
examples.
A push-forward by F of the Lagrangian vector dX to the current configuration gives the Eulerian vector dx:

(A.1a) dx = F · dX ≡ φ∗ dX

The pull-back by F−1 of the Eulerian vector dx to the reference configuration gives dX.

(A.1b) dX = F−1 · dx ≡ φ∗ dx

Where φ∗ and φ∗ represent the push-forward and pull-back operations respectively.

If we extend these operations, we can consider the pull-back and push-forward operations on second-order
tensors to obtain the relationships between these tensors in the deformed and undeformed configurations. These
concepts provide us a mathematically consistent method for define the pseudo-time derivatives of the tensors,
called Lie-derivatives. Considering all this, the operations can be considered for example to obtain a formulation of
hyperelastic-plastic constitutive model based on the multiplicative decomposition of the deformation gradient (F), for
more information we address[BookBel+14].

A.2 Mortar segmentation

A.2.1 Introduction
In this section we analyse in detail, the segmentation method chooses in order to compute the mortar contact
constraints. In a first section, the discussion will be focused around the integration method considered, in a second
section the triangulation methods considered on 3D will be presented.

Vicente Mataix Ferrándiz Page 285 of 374

A.2 Mortar segmentation APPENDIX A. THEORETICAL COMPLEMENTS

A.2.2 Exact integration vs Collocation

A.2.2.1 Theory
In this section we study the two different approaches for the mortar method, the exact integration (also denominated
segment base) and the collocation method (usually called element base). The first one as the method indicates,
consists on the decomposition and exact integration of the master and slave segments (Figure A.1), and is the
approach follows in Popp[PhDPop12]. The second alternative (Figure A.2) consists in consider a big number of GP
on the integration of the pair, the GP inside the opposite pair are considered into integration, and the ones outside are
discarded.

Figure A.1: Exact segmentation method for integration. Figure inspired on Popp[PhDPop12]

On the latter method, the integration points are not distributed following any Gauss quadrature, but distributed
uniformly in all the geometry, thats why the method is denominated collocation method. If the number of collocation
points is big enough, the result will be close to the exact solution. Apart from the mortar methods, this approach is
extensively considering on the isogeometric analysis[ArtAur+10; ArtSch+13].

Figure A.2: Collocation method for integration. Figure inspired on Popp[PhDPop12]

From the point of view of the implementation, the element base is a significantly easier method to implement, as
there are not many modifications to a standard FE implementation, only a check for the inside/outside GP. Additionally
the values and derivatives definitions are exactly equal in 2D and 3D, additionally do not depend on the segmentation,
being considerably easier.
On the other hand, the exact integration, particularly complex on 3D, which requires to property implement the
clipping procedure (4.105) and triangulation (see A.2.3.Delaunay vs Convex polygon construct), in addition to the
corresponding derivatives from 4.6.2.Derivatives for 3D contact. This is quire more complex, but as it will be shown
(A.2.2.2.Solution study), the costs languish before the advantages.
The results of the study here presented comes from the experience as both approaches have been considered in

Page 286 of 374 Vicente Mataix Ferrándiz

APPENDIX A. THEORETICAL COMPLEMENTS A.2 Mortar segmentation

the mortar implementation for the contact mechanics. The results obtained with the collocation method drive us take
as final approach the exact integration method, and the results and developments present on this work come from this
integration method. Similar results were obtained by Farah[ArtFPW14], where both methods were tested and the
conclusions obtained were similar.

A.2.2.2 Solution study

Figure A.3: Taylor patch test[ArtTP]

Here we study how the solution obtained with both methods, on Farah[ArtFPW14] work the comparison is done in a
more complex setup, but we will show a simpler case which analytic solution, that even in that case does not fulfil the
analytic solution. We will consider a classical problem which is used on benchmarking the contact formulation, this
method was particularly extended for testing the NTS formulations. Figure A.3 shows this test, called Taylor patch
test[ArtTP].

(a) Vertical displacement (b) Detail displacement on the interface

Figure A.4: Displacement on Taylor test

The problem consists on two solid blocks contacting on the interface, with a distributed load on the top block.
The solution expected is a continuous field of displacements plus a constant contact pressure which coincides

Vicente Mataix Ferrándiz Page 287 of 374

A.2 Mortar segmentation APPENDIX A. THEORETICAL COMPLEMENTS

with the applied load in the top. We will consider for this problem E = 1000, ν = 0.4 and the distributed load as
p = 10Pa.
In our study we analyse the convergence of λ̄n in the resolution of the problem with a different number of GP
on the collocation. We do so, because λ̄n takes longer to converge to the analytical solution. On the other hand,
displacements converge even considering a low number of GP. Farah[ArtFPW14] also studies the convergence of λ̄n ,
but additionally checks the convergence of the energy error.

(a) Augmented contact pressure with mortar segmentation (b) Augmented contact pressure with mortar collocation with 200 GP
Figure A.5: Compared solution for the augmented contact pressure

On Figure A.4 we can see the solution of the vertical displacements of the problem, particularly on the subfigure
A.4a. Then in the subfigure A.4b the detail on the interface is appreciated, here we can see that in the deformed
configuration the interface does not match any more. This loss of matching across of the interfaces is the reason why
the collocation method has problems to converge.

(a) Vertical displacement convergence for different number of GP (b) Augmented contact pressure convergence for different number of GP
Figure A.6: Convergence of the solution for different number of GP

The next Figure shows the solution for the augmented contact pressure, where Figure A.5a shows the exact
solution, corresponding with λ̄n = −10Pa. The Figure A.5b presents the solution for 200 GP with the collocation
method. Both solutions look very close, for more details we will address a more detailed plot next.

Page 288 of 374 Vicente Mataix Ferrándiz

APPENDIX A. THEORETICAL COMPLEMENTS A.3 Mesh tying

The last figure, Figure A.6, tries to represent with more detail the convergence of the solution for the vertical
displacement and for λ̄n . On the first one, Figure A.6a, it can be appreciated that the vertical displacement converges
and gives the correct solution with a very small number of GP. On the other hand, Figure A.6b shows the convergence
of the augmented contact pressure, and even for a large number of GP the exact solution is not obtained, despite of
being close and getting closer with the grow of the number of GP.

A.2.3 Delaunay vs Convex polygon construct

On this section we will study the three different approaches, we can consider in order to build the integration triangles
considered on the mortar segmentation from a cloud of points obtained during the clipping procedure. These methods
are:

• The Delaunay triangulation[Onl]: The procedure (Figure A.7a) generates a set of triangles from a set of points.
The method maximises the minimum angle of all the angles of the triangles in the triangulation. In the case
of convex polygons (see next point) the number of triangles generated corresponds with (n − 2), being n the
number of vertices. The method must not be confused with Constrained Delaunay Triangulation (CDT),
which is an improved version of the method allowing to fix segments.
• The convex polygon construct: Also called fan triangulation, due to the resulting aspect of this method (Figure
A.7b). This method it is only applicable in the case that the points configure a convex polygon, which means only
in polygons in which no line segment between two points on the boundary ever goes outside the polygon[Onl].
The most important property of this kind of polygon is that every convex polygon admits a fan triangulation in
(n − 2) triangles1 .
• The third method is the so-called centre-based triangulation, see Figure A.7c. This method is discarded
because generates n triangles from n vertices (then more expensive than the previous alternatives). The only
advantage compared with the previous alternative is the fact that the aspect ratio of the generated triangles is,
in general, better than the one obtained with the fan triangulation.

Comparing three methods, the Delaunay triangulation it is for sure the most general method, but algorithmically
more complex to implement and significantly more expensive in computational cost. This method is in fact implemented
as optional in Kratos, considering the robust and extensively use triangle library. On the other side, the fan
triangulation is restricted to fewer cases (convex polygons), but the implementation is much simpler, easier and less
error-prone. The third method, centre-based triangulation, despite of giving a better aspect ratio triangulation, has
as the main disadvantage that, as previously mentioned, generates a higher number of triangles, being then from a
computational point of view, more expensive.
It is important to highlight the fact that choose between these two methods is easier if we take into consideration that
the geometries considered for the contact problems are linear all of them. This implies that the resulting segmentation
(Figure A.1) will always provide us a convex polygon, which means we can apply the fan triangulation method.
For this reason, because for a convex polygon the advantages outweigh the disadvantages.

A.3 Mesh tying

A.3.1 Introduction
The following section presents the formulation employed in the derivation of the mesh tying condition formulation
with dual Lagrange multiplier, based in the work of Alexander Popp[PhDPop12; ArtPop+10]. The aim of this section
is to shortly introduce the mesh tying formulation, which can be considered as the base formulation for the contact
formulation from 4.3.Formulation. This formulation is simpler and allows to understand and check how the mortar
integration works.
The mesh tying problem is based on the IBVP of non-linear solid mechanics and the tied contact constraints. After
recapitulating some basic notation and the strong formulation, a weak formulation of the mesh tying problem with
two subdomains will be introduced. Here, only the interpretation as constrained minimisation problem is considered,
leading to an indefinite saddle point formulation based on Lagrange multipliers.
1 This is determined by the results obtained by Catalan, particularly in application of the Catalan number[ArtKG11]

Vicente Mataix Ferrándiz Page 289 of 374

A.3 Mesh tying APPENDIX A. THEORETICAL COMPLEMENTS

(a) Delaunay triangulation

(b) Fan triangulation. Inspired on[Onl] (c) Center-based triangulation

A.3.2 Strong formulation

On each subdomain Ωi0 , the initial boundary value problem of finite deformation elastodynamics needs to be satisfied,
this means to follow (A.2). The tied contact constraint, also formulated in the reference configuration, is given as
(A.2f).

(A.2a) ∇ · Pi + b̂i0 = ρi0 üi in Ωi0 × [0, T ]

(A.2b) ui = ûi on Γiu × [0, T ]
(A.2c) Pi · Ni = t̂i0 on Γiσ × [0, T ]



(A.2d) ui Xi , 0 = ûi0 Xi in Ωi0



(A.2e) u̇i Xi , 0 = u̇ˆ i0 X i
in Ωi0
(A.2f) u1 = u2 on Γic × [0, T ]

In the course of deriving a weak formulation, the balance of linear momentum at the mesh tying interface Γic is
typically exploited and a LM vector field λ is introduced, thus setting the basis for a mixed variational approach.

A.3.3 Weak formulation

To start the derivation of a weak formulation of (A.2), appropriate solution spaces U i and weighting spaces V i need
to be defined as (A.3).

( 
U i = ui ∈ H 1 (Ω)kui = ûi on Γiu ,
(A.3) 
V i = δ ui ∈ H 1 (Ω)kδ ui = 0 on Γiu

Page 290 of 374 Vicente Mataix Ferrándiz

APPENDIX A. THEORETICAL COMPLEMENTS A.3 Mesh tying

Additionally the Lagrange multiplier vector λ = −t1c , which enforces the mesh tying constraint(??), represents the
negative slave side contact traction t1c , is chosen from a corresponding solution space denoted as M. In terms of its
classification in functional analysis, this space represents the dual space of the trace space W 1 of V 1 . In the given
context, this means that M = H 1/2 (Γc ) and L1 = H 1/2 (Γc ), where M and L1 denote single scalar components of the
corresponding vector-valued spaces M and W .
Based on these considerations, a saddle point type weak formulation is derived next. This can be done by
extending the standard weak formulation of NL solid mechanics as defined to two subdomains and combining it with
the LM coupling terms introduced in generic form. Find ui ∈ U i and λ ∈ M such that we obtain (A.4).

(A.4a) −δLkin (ui , δ ui ) − δLint,ext (ui , δ ui ) − δLmnt (λi , δ ui ) = 0 ∀δ ui ∈ V

(A.4b) − δLλ (ui , δλi ) = 0 ∀δλi ∈ M

Herein, the kinetic contribution δLkin , the internal and external contributions δLint,ext and the mesh tying interface
contribution δLmnt to the overall virtual work on the two subdomains, as well as the weak form of the mesh tying
constraint δLλ , have been abbreviated as (A.5).

2
"Z #
X
(A.5a) − δLkin = ρi0 üi i
· δ u dV0
i=1 Ωi0
2
"Z Z #
X

i i i
(A.5b) − δLint,ext = S : δ E − b̂ · δ u dV0 − t̂i0 i
· δ u dA0
i=1 Ωi0 Γiσ
2
"Z #
X

1 2
(A.5c) − δLmnt = λ · δu − δu dA0
i=1 Γic
2
"Z #
X

(A.5d) − δLλ = δλ · u1 − u 2
dA0
i=1 Γic

The coupling terms on Γc also allow for a direct interpretation in terms of variational formulations and the principle
of virtual work. Whereas the contribution in (A.5c) represents the virtual work of the unknown interface traction
λ = t1c = t2c , the contribution in (A.5d) ensures a weak, variational consistent enforcement of the tied contact constraint
(??). Nevertheless, the concrete choice of the discrete LM space Mh in the context of mortar finite element
discretisations is decisive for the stability of the method and for optimal a priori error bounds. Finally, it is pointed
out that the weak formulation (A.4a) and (A.4b) possesses all characteristics of saddle point problems and Lagrange
multiplier methods.

A.3.4 Discretisation and numerical integration

The part relative to the dual LM it is detailed in the respective section 4.3.3.4.1.Dual Lagrange multipliers, and the
Mortar operators from 4.3.3.4.2.Mortar operators. This development is identical for the mesh tying formulation, and
will not be repeated.
Finally, once computed the mortar operators, the resulting system for mesh tying corresponds with (A.6). This can
be statically condensed, as Alexander Popp shows in [PhDPop12]. With this the LM can be removed of the system
thanks to the diagonal property of D when considering the dual LM, making a pure displacement formulation (in case
the tied DOF corresponds with the displacements).

    
KN N KN M KN S 0 ∆dN rN
 KN N KMM 0 − MT   ∆ dM   rM 
(A.6)  T
  = − 
 KSN 0 KSS D   ∆ dS   rS 
0 −M D 0 ∆λ rλ

Vicente Mataix Ferrándiz Page 291 of 374

A.3 Mesh tying APPENDIX A. THEORETICAL COMPLEMENTS

A.3.5 Numerical example

E Solid 1 ν Solid 1 E Solid 2 ν Solid 2

2 · 108 Pa 0.35 2 · 108 Pa 0.35
Table A.1: Parameters considered for the mesh tying example
The following example, with the mesh presented on
Figure A.7, consists in a L shaped solid, where we con-
sider as a different body the circular section from the
angled corner. The mesh interface between the solids
does not match. The problem is formulated as solid
formulated on TL framework, considering Neo-Hookean
hyperelastic material with the properties from the Table
A.1. We apply in the upper corner a vertical displace-
ment equal to t. with t going from [0, 2]s, the lower corner
is fixed.
The solution obtained is shown in Figure A.8, it can
be seen that the continuity in the interface is preserved,
despite the large displacement solution. From a practical
point of view is like considering a continuous element
of the whole domain. This is basically the type of for-
mulation originally considered in the original Mortar for
(a) Mesh front (b) Mesh perspective
DDM[BookWoh01; BookTos05]. Figure A.7: Mesh tying example

(a) Solution at t = 0.5s (b) Solution at t = 1.0s (c) Solution at t = 1.5s (d) Solution at t = 2.0s
Figure A.8: Solution for mesh tying example

Page 292 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookBel+14] Nonlinear Finite Elements for Continua and Structures. Ted Belytschko, Wing Kam Liu,
Brian Moran, and Khalil Elkhodary. Wiley. 2nd ed. 2014.
[BookTos05] Domain Decomposition Methods — Algorithms and Theory. Andrea Toselli Olof B. Widlund (auth.)
Springer-Verlag Berlin Heidelberg. 1st ed. 2005.
[BookWoh01] Discretization Methods and Iterative Solvers Based on Domain Decomposition.
Barbara I. Wohlmuth. Springer-Verlag Berlin Heidelberg. 1st ed. 2001.

Articles
[ArtAur+10] “Isogeometric collocation methods”. F Auricchio, L Beirao Da Veiga, TJR Hughes, A_ Reali, and
G Sangalli. In: Mathematical Models and Methods in Applied Sciences. No. 11, Vol. 20, 2010,
pp. 2075–2107. World Scientific. DOI: 10.1142/S0218202510004878.
[ArtFPW14] “Segment-based vs. element-based integration for mortar methods in computational contact
mechanics”. Philipp Farah, Alexander Popp, and Wolfgang A. Wall. In: Computational Mechanics.
No. 1, Vol. 55, 2014, pp. 209–228. DOI: 10.1007/s00466-014-1093-2.
[ArtKG11] “Some divisibility properties of Catalan numbers”. Thomas Koshy and Zhenguang Gao. In: The
Mathematical Gazette. Vol. 95, 2011, pp. 96–102. DOI: 10.1017/S002555720000245X.
[ArtPop+10] “A dual mortar approach for 3D finite deformation contact with consistent linearization”.
Alexander Popp, Markus Gitterle, Michael W. Gee, and Wolfgang A. Wall. In: International Journal
for Numerical Methods in Engineering. No. 11, Vol. 83, 2010, pp. 1428–1465. John Wiley & Sons,
Ltd. DOI: 10.1002/nme.2866.
[ArtSch+13] “Isogeometric collocation: Cost comparison with Galerkin methods and extension to adaptive
hierarchical NURBS discretizations”. Dominik Schillinger, John A Evans, Alessandro Reali,
Michael A Scott, and Thomas JR Hughes. In: Computer Methods in Applied Mechanics and
Engineering. Vol. 267, 2013, pp. 170–232. Elsevier. DOI: 10.1002/pamm.201310049.
[ArtTP] “On a patch test for contact problems in two dimensions”. Robert L. Taylor and
Panagiotis Panagiotopoulos. In: . ,

Ph.D.’s thesis
[PhDPop12] “Mortar Methods for Computational Contact Mechanics and General Interface Problems”.
Alexander Popp. 2012.

Online resources
[Onl] Wikipedia. Convex polygon. URL : https://en.wikipedia.org/wiki/Convex_polygon.

Vicente Mataix Ferrándiz Page 293 of 374

ONLINE RESOURCES BIBLIOGRAPHY

[Onl] Wikipedia. Delaunay triangulation. URL :

https://en.wikipedia.org/wiki/Delaunay_triangulation.
[Onl] Wikipedia. Fan triangulation. URL :
https://en.wikipedia.org/wiki/Fan_triangulation.

Page 294 of 374 Vicente Mataix Ferrándiz

APPENDIX B. IMPLEMENTATION

Appendix B

Implementation

If I have seen further, it is by

standing on the shoulders of Giants.

Isaac Newton
(1643 - 1727 AD, English
mathematician, physicist,
astronomer, theologian)

B.1 Introduction
The following appendix introduces the main codes used on the developments of this work. The first one is the
multiphysics framework Kratos (B.2.Kratos Multiphysics), where most of the developments were done. The next
section is the remeshing library denominated Mmg (B.3.Mmg library) considered on the adaptive remeshing chapter
(6.Adaptative remeshing).

B.2 Kratos Multiphysics

B.2.1 Introduction

Figure B.1: Kratos Multiphysics logo. Image from Kratos web

The implementation shown in this research has been done on the Kratos multi-physics framework. Kratos[PhDDad07;
ArtDRO10] is a framework for building multidisciplinary finite element programs. We can summarise the following
features:

• Kernel: The kernel and application approach is used to reduce the possible conflicts arising between developers
of different fields.

• Object Oriented (OO): The modular design, hierarchy and abstraction of these approaches fit to the generality,
flexibility and reusability required for the current and future challenges in numerical methods. Summarised as
Multi-Layer Design Approach (MLDA). The main code is developed in C++ and the Python language is used
for scripting. The C++ practises are focus on modern guidelines[BookAle01; BookMey14]

Vicente Mataix Ferrándiz Page 295 of 374

B.2 Kratos Multiphysics APPENDIX B. IMPLEMENTATION

• Open source: The Berkeley Software Distribution (BSD) licence allows to use and distribute the existing
code without any restriction, but with the possibility to develop new parts of the code on an open or close basis
depending on the developers. Additionally Kratos can be obtained for free.

• HPC: Kratos is build-in as a powerful multiphysics HPC[ArtDad+13]. Kratos is based on Trilinos MPI and
OpenMP.

B.2.2 Framework (API)

B.2.3 Introduction. Main classes
The OO design from Figure B.2a provides different classes nec-
essary for a FE simulation which ensure the reusability and ex-
tensibility of the code. Here, basic numerical concepts are rep-
resented by classes like Vector, Matrix and Quadrature.
Node, Element, Condition, and DoF are defined directly
from FE concepts. Model, Mesh and Properties are from
the practical methodology used in FE modelling complemented by
ModelPart, and SpatialContainer, for organising better
all data necessary for analysis. IO, LinearSolver, Process,
and Strategy represent the different steps of a FE program
flow. Finally, Kernel and Application are defined for library
management and its interface definition. (a) Kratos classes
In the MLDA each object only interfaces with other objects in
its layer or in layers below its layer (Figure B.2b). This approach
is used to reduce the dependency of different parts which is
beneficial for the reduction of conflicts between different parts of
the code[ArtDRO10].
The Applications considered and developed for this work
are the following:

• Core: Where the main classes are declared, Figure

B.2a. It also includes several algorithms and utilities
common to all the Applications, like the NR or the
BuilderAndSolver.

• StructuralMechanicsApplication: Where the el-

ements, conditions and CL are implemented. Also other
utilities and strategies are implemented here. The elements
from 3.Rotation-free shells and solid-shell elements and
the CL from 5.Plasticity can be found here.

• ContactStructuralMechanicsApplication: Where
the conditions, strategies and algorithms presented on
4.Contact mechanics can be found.

• MeshingApplication: All the developments relative

with Mmg and the metric and remeshing techniques1 from
6.Adaptative remeshing. The details of the Mmg integra-
tion can be found in B.3.2.Integration between Mmg and
Kratos. (b) Kratos MLDA
Figure B.2: Source[ArtDRO10], updated with the
recent classes added
The following sections will summarise the main classes existing in Kratos.
1 The SPR error algorithms are located in the StructuralMechanicsApplication and ContactStructuralMechanicsApplication.

Page 296 of 374 Vicente Mataix Ferrándiz

APPENDIX B. IMPLEMENTATION B.2 Kratos Multiphysics

B.2.4 Model
B.2.4.1 Introduction
On Figure B.3 an example of the data structure of the Model. The Model stores the whole model to be analysed,
and manages the different ModePart used in the simulation.

Figure B.3: Model data structure

B.2.4.2 ModelPart
The ModelPart holds all data related to an arbitrary part of model. It stores all existing components and data
like Nodes, Properties, Elements, MasterSlaveConstraint, Conditions and solution data related to a
part of the Model. The entities stored on the ModelPart are:

• Node It is a point with additional facilities. Stores the nodal data, historical nodal data, and list of DoF.
• Condition encapsulates data and operations necessary for calculating the local contributions of Condition
to the global system of equations.

• Element encapsulates the elemental formulation in one object and provides an interface for calculating the
local matrices and vectors necessary for assembling the global system of equations. It holds its geometry that
meanwhile is its array of Nodes.

• MasterSlaveConstraint encapsulates the MPC that are used in order to impose relationships between
DOF.

• Properties encapsulates data shared by different Elements or Conditions. It stores any type of data.

B.2.4.3 Submodelparts
In our implementations we use processes (Figure B.4a) to set the BC (both Neumann or Dirichlet), as idealised on
Figure B.4b. In order to be able to assign this BC we need to subdivide the ModelPart into different pieces that we
can call submodelparts.

(a) Example of BC in json format (b) Submodelpart BC

Figure B.4: Concept of submodelpart for BC

Vicente Mataix Ferrándiz Page 297 of 374

B.3 Mmg library APPENDIX B. IMPLEMENTATION

B.2.5 Base classes

In here we detail some relevant classes which define the basic behaviour of Kratos. The classes used in the FEM
are already defined in B.2.4.2.ModelPart. Apart of these classes, we have another relevant base class, as the base
Process, which defines the common interface of the processes. The Process is a class which evaluates different
types of algorithms, like searches, interpolate values, etc.

B.2.6 Geometry
The Geometry is the base class which defines all the common methods of interest in a geometry, such as volume,
area, shape functions, etc. The standard geometries store an array of nodes or points.

B.2.7 Strategy
The strategies are responsible on solving the system of equations. There are 4 types of strategies:

• BuilderAndSolver: This is the base class which defines the necessary methods to assemble the system of
equations. I order to do so it asks the Element, Condition and MasterSlaveConstraint for its local
contribution and assemble them in a global system depending of the DoF order.

• Scheme: This class provides the implementation of the basic tasks that are needed by the solution strategy in
order to integrate the solution over the time. Check algorithms on 2.4.4.Time integration schemes.

• ConvergenceCriteria: This is the base class to define the different convergence criterion considered.
Here the residual/solution increment or any other criteria is taken into account in order to check if converged.

• SolvingStrategy: This is the base class from which we will derive all the strategies (line-search, NR, etc.).
See 2.4.3.Solution of the non-linear-equilibrium equations system. It is the class responsible of actually solving
the problem. It integrates all the former classes.

B.2.8 IO classes
In this class we can include the classes dedicated to read external files, such as the *.mdpa files, default input for
Kratos. We can include the utilities used to read *.json files. These are the default file types used to define the
parameters of the problem and the material Properties. Additionally, in the IO category, we can include classes
such the Logger which is used to show/register/save the simulated information.

B.2.9 Testing framework

The testing framework is very relevant in Kratos, and have allowed a faster in Kratos than the former to the
introduction of this technology. Unittest are extensibility used on Information and Communications Technology
(ICT) to ensure code quality. An unittest is a software testing method by which individual units of the source code,
hence the name comes, are tested to determine whether they behave as intended. We can differentiate in Kratos
two different approaches for the unittest:

• Python tests: These tests are based in the Python unittest framework, which was originally inspired by
JUnit. With this framework we can test and check methods exported to Python and run full short problems in
order to check that the results obtained are the expected.

• C++ tests: This is a custom Kratos testing framework where the test is fully written in C++.

B.3 Mmg library

B.3.1 What is Mmg and how does it work?
Mmg is an open source software for anisotropic automatic remeshing for unstructured meshes based on Delaunay
triangulation. It is licenced under a GNU Lesser General Public License (LGPL) licence and it has been integrated

Page 298 of 374 Vicente Mataix Ferrándiz

APPENDIX B. IMPLEMENTATION B.3 Mmg library

Figure B.5: Mmg logo. Image from Mmg web

in Kratos[ArtDRO10] via the mmg_process.h in the MeshingApplication. It provides 3 applications and 4

libraries:

• The mmg2d application and the libmmg2d library: adaptation and optimisation of a two-dimensional triangula-
tion and generation of a triangulation from a set of points or from given boundary edges.

• The mmgs application and the libmmgs library: adaptation and optimisation of a surface triangulation and
isovalue discretisation.

• The mmg3d application and the libmmg3d library: adaptation and optimisation of a tetrahedral mesh and
implicit domain meshing.

• The libmmg library gathering the libmmg2d, libmmgs and libmmg3d libraries.

The Mmg remeshing process modifies the mesh[BookDob12][ArtDDF14] iteratively until be in agreement with the
prescribed sizes on the idealised (Figure B.6) contour (and directions in case of anisotropic mesh). The software
reads the mesh and the metric, then the mesh is modified using local mesh modifications of which an intersubsection
procedure based on anisotropic Delaunay kernel.

(a) A piece of parametric Bézier cubic surface associated to triangle T (b) The resulting configuration of the vertex relocation procedure
Figure B.6: Mmg idealised geometry. Source[ArtDDF14]
We can resume the algorithm in following steps:

1. Mmg tries to have a good approximation of the surface (with respect to the Hausdorff parameter).

2. It remeshes on geometric criteria. Mmg scans the surface tetrahedra and splits the tetrahedra using prede-
fined patterns if the Hausdorff distance[BookRW09] between the surface triangle of the tetra and its curve
representation doesn’t respect the Hausdorff parameter.

3. The library scans again the surface tetrahedra and collapses all the edges at a Hausdorff distance smaller than
a threshold defined in function of the Hausdorff parameter.

4. Next it intersects the provided metric and a surface metric computed at each point from the Hausdorff parameter
and the curvature tensor at the point.

Vicente Mataix Ferrándiz Page 299 of 374

B.3 Mmg library APPENDIX B. IMPLEMENTATION

5. Then Mmg smooths the metric to respect the gradation parameter. The metrics are iteratively propagated until
the respect of the gradation everywhere.

6. Following it remeshes the surface tetrahedra in order to respect the new metric.

7. Finally, it remeshes both the volume and surface to have edges between 0.6 and 1.3 (in the metric). The long
edges are cut and short ones are deleted (collapsed).

B.3.2 Integration between Mmg and Kratos

B.3.2.1 Introduction
In order to understand the integration between Kratos and Mmg is important to understand the data structure
of Kratos. We address in first place to the corresponding section about Model structure on the Kratos API
(B.2.4.Model).

B.3.2.2 Class structure

The implementation of Mmg into Kratos is
done in the MeshingApplication. The
Unified Modeling Language (UML) dia-
gram presented on Figure B.7 shows how
the implementation is done.
In here we see like there is an auxil-
iary class called MMGUtilities, which is
called by the MMGProcess, which is the
process called for remeshing and MMGIO
which can be used to read write Mmg IO
files.
MMGUtilities is the class responsi-
ble to interact with the Mmg API. Due to
structure in Mmg library, a template ar-
gument must be considered in order to
call the respective methods for each one
of the specialised includes (libmmg2d.h,
libmmg3d.h and libmmgs.h).
Additionally the MMGProcess integrates
all the steps during the remesh process, in-
cluding the value interpolation and the in-
ternal variables recovery from 6.6.Internal
values interpolation. After its execution all Figure B.7: Mmg-Kratos integration
the solvers and the processes must be ini-
tialised.

B.3.2.3 Submodelpart recovery

As we have previously mentioned, on B.2.4.3.Submodel-
parts we need to subdivide the mesh into submodelparts
so we can apply BC. In order to preserve that informa-
tion after remeshing we need to create an identification
system, so we are able to create a unique ID that will
allow us to reconstruct the submodelpart structure after
remeshing, this methodology is commonly called colour
identification. This is thanks to the API of Mmg which al-
lows to preserve a reference number, which works as an
ID. In Figure B.8 we see the concept of this idea. Figure B.8: Concept of colours

Page 300 of 374 Vicente Mataix Ferrándiz

 APPENDIX B. IMPLEMENTATION B.3 Mmg library

In Kratos this is implemented in the AssignUniqueModelPartCollectionTagUtility

class, that can be found here. This class computes a map of collections of submodelparts, in order to do that, a tag is
assigned to each node, condition and element in order to get the collections where it belongs. The main ModelPart
tag is 0, then each submodelpart has 1, 2... Finally, a combination of these submodelparts has another tag, greater
than the former ones.
The class make use of C++ std::unordered_map. First the list of tags of each ModelPart is created, with
the corresponding entities belonging to it. Then with this list map, the submodelparts combination tags are computed.
Finally, with this, the list containing the tags of each submodelparts combination and the entities contained in it. The
following block showns the way it works:

1 void AssignUniqueModelPartCollectionTagUtility::ComputeTags(
2 IndexIndexMapType& rNodeTags,
3 IndexIndexMapType& rCondTags,
4 IndexIndexMapType& rElemTags,
5 IndexStringMapType& rCollections
6 )
7 {
8 // Initialize and create the auxiliary maps
9 IndexIndexSetMapType aux_node_tags, aux_cond_tags, aux_elem_tags;
10
11 // We compute the list of submodelparts and subsubmodelparts
12 const StringVectorType& r_model_part_names = GetRecursiveSubModelPartNames(mrModelPart);
13
14 // Initialize the collections
15 IndexType tag = 0;
16 for (IndexType i_sub_model_part = 0; i_sub_model_part < r_model_part_names.size(); ++i_sub_model_part) {
17 rCollections[i_sub_model_part].push_back(r_model_part_names[i_sub_model_part]);
18
19 if (tag > 0) {
20 ModelPart& r_sub_model_part = GetRecursiveSubModelPart(mrModelPart, r_model_part_names[i_sub_model_part]);
21
22 /* Nodes */
23 NodesArrayType& r_nodes_array = r_sub_model_part.Nodes();
24 const auto it_node_begin = r_nodes_array.begin();
25 for(IndexType i_node = 0; i_node < r_nodes_array.size(); ++i_node)
26 aux_node_tags[(it_node_begin + i_node)->Id()].insert(tag);
27
28 /* Conditions */
29 ConditionsArrayType& r_conditions_array = r_sub_model_part.Conditions();
30 const auto it_cond_begin = r_conditions_array.begin();
31 for(IndexType i_cond = 0; i_cond < r_conditions_array.size(); ++i_cond)
32 aux_cond_tags[(it_cond_begin + i_cond)->Id()].insert(tag);
33
34 /* Elements */
35 ElementsArrayType& r_elements_array = r_sub_model_part.Elements();
36 const auto it_elem_begin = r_elements_array.begin();
37 for(IndexType i_elem = 0; i_elem < r_elements_array.size(); ++i_elem)
38 aux_elem_tags[(it_elem_begin + i_elem)->Id()].insert(tag);
39 }
40
41 ++tag;
42 }
43
44 // Now detect all the cases in which a node or a cond belongs to more than one part simultaneously
45 std::unordered_map<std::set<IndexType>, IndexType, KeyHasherRange<std::set<IndexType», KeyComparorRange<std::set<IndexType» > combinations;
46
47 /* Nodes */
48 for(auto& r_aux_node_tag : aux_node_tags) {
49 const std::set<IndexType>& r_value = r_aux_node_tag.second;
50 if (r_value.size() > 1) combinations[r_value] = 0;
51 }
52
53 /* Conditions */
54 for(auto& r_aux_cond_tag : aux_cond_tags) {
55 const std::set<IndexType>& r_value = r_aux_cond_tag.second;
56 if (r_value.size() > 1) combinations[r_value] = 0;
57 }
58
59 /* Elements */
60 for(auto& r_aux_elem_tag : aux_elem_tags) {
61 const std::set<IndexType>& r_value = r_aux_elem_tag.second;
62 if (r_value.size() > 1) combinations[r_value] = 0;
63 }
64
65 /* Combinations */
66 for(auto& combination : combinations) {
67 const std::set<IndexType>& r_key_set = combination.first;

Vicente Mataix Ferrándiz Page 301 of 374

 B.3 Mmg library APPENDIX B. IMPLEMENTATION

68 for(IndexType it : r_key_set)
69 rCollections[tag].push_back(rCollections[it][0]);
70 combinations[r_key_set] = tag;
71 ++tag;
72 }
73
74 // The final maps are created
75 /* Nodes */
76 for(auto& r_aux_node_tag : aux_node_tags) {
77 const IndexType key = r_aux_node_tag.first;
78 const std::set<IndexType>& r_value = r_aux_node_tag.second;
79
80 if (r_value.size() == 0)
81 rNodeTags[key] = 0; // Main Model Part
82 else if (r_value.size() == 1) // A Sub Model Part
83 rNodeTags[key] = *r_value.begin();
84 else // There is a combination
85 rNodeTags[key] = combinations[r_value];
86 }
87
88 /* Conditions */
89 for(auto& r_aux_cond_tag : aux_cond_tags) {
90 const IndexType key = r_aux_cond_tag.first;
91 const std::set<IndexType>& r_value = r_aux_cond_tag.second;
92
93 if (r_value.size() == 0)
94 rCondTags[key] = 0; // Main Model Part
95 else if (r_value.size() == 1) // A Sub Model Part
96 rCondTags[key] = *r_value.begin();
97 else // There is a combination
98 rCondTags[key] = combinations[r_value];
99 }
100
101 /* Elements */
102 for(auto& r_aux_elem_tag : aux_elem_tags) {
103 const IndexType key = r_aux_elem_tag.first;
104 const std::set<IndexType>& r_value = r_aux_elem_tag.second;
105
106 if (r_value.size() == 0)
107 rElemTags[key] = 0; // Main Model Part
108 else if (r_value.size() == 1) // A Sub Model Part
109 rElemTags[key] = *r_value.begin();
110 else // There is a combination
111 rElemTags[key] = combinations[r_value];
112 }
113
114 // Clean up the collections
115 for (auto& r_collection : rCollections) {
116 std::unordered_set<std::string> aux_set;
117 for (auto& r_name : r_collection.second) {
118 aux_set.insert(r_name);
119 }
120 std::vector<std::string> aux_vector;
121 for (auto& r_name : aux_set) {
122 aux_vector.push_back(r_name);
123 }
124 r_collection.second = aux_vector;
125 }
126 }

Page 302 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookAle01] Modern C++ design: generic programming and design patterns applied. Andrei Alexandrescu.
Addison-Wesley. 2001.
[BookDob12] MMG3D: user guide. Cécile Dobrzynski. 2012.
[BookMey14] Effective modern C++: 42 specific ways to improve your use of C++ 11 and C++ 14. Scott Meyers.
O’Reilly Media, Inc. 2014.
[BookRW09] Variational analysis. R Tyrrell Rockafellar and Roger J-B Wets. Springer Science & Business Media.
2009.

Articles
[ArtDad+13] “Migration of a generic multi-physics framework to HPC environments”. Pooyan Dadvand,
Riccardo Rossi, Marisa Gil, Xavier Martorell, J Cotela, Edgar Juanpere, Sergio R Idelsohn, and
Eugenio Oñate. In: Computers & Fluids. Vol. 80, 2013, pp. 301–309. Elsevier.
[ArtDDF14] “Three-dimensional adaptive domain remeshing, implicit domain meshing, and applications to free
and moving boundary problems”. C. Dapogny, C. Dobrzynski, and P. Frey. In: Journal of
Computational Physics. Vol. 262, 2014, pp. 358–378. DOI: 10.1016/j.jcp.2014.01.005.
[ArtDRO10] “An object-oriented environment for developing finite element codes for multi-disciplinary
applications”. Pooyan Dadvand, Riccardo Rossi, and Eugenio Oñate. In: Archives of computational
methods in engineering. No. 3, Vol. 17, 2010, pp. 253–297. Springer.

Ph.D.’s thesis
[PhDDad07] “A framework for developing finite element codes for multi-disciplinary applications.”
Pooyan Dadvand. Universitat Politècnica de Catalunya. 2007.

Vicente Mataix Ferrándiz Page 303 of 374

PH.D.’S THESIS BIBLIOGRAPHY

Page 304 of 374 Vicente Mataix Ferrándiz

APPENDIX C. AUTOMATIC DIFFERENTIATION

Appendix C

Automatic differentiation

God does not care about our

mathematical difficulties - he
integrates empirically.

Albert Einstein
(1879 - 1955 AD, German-born
theoretical physicist)

C.1 Introduction
The AD[ArtGru82; ArtKor97; ArtKor02; BookWri08], also called algorithmic differentiation or computational
differentiation. It consists in a set of techniques to numerically evaluate the derivative of a function specified by
a computer program. This is possible to the main principle which defines the derivative operation, which is the
differentiation of an algorithm by the use of the chain rule. This operation is completely linear, so it can be easily
adapted to any problem, nevertheless of the complexity. As said Korelc[BookJo16] says, the main advantages of the
AD are:

• Automatic generated codes are highly efficient if the software employed does the corresponding optimisations
properly.

• The symbolic formulation is more compressed, for example we only work on the definition of the energy potential,
and thus gives fewer possibilities of an error.

• The code generated can be adjusted to specific problems, leading to on-demand numerical code generation.
• The code can be generated for different environments and multi-languages from the same symbolic description.
• Some complex and error-prone operations, such algebraic operations are done automatically.
• Simplifies the implementation of multiphysics problems, as the symbolic definitions are purely mathematical and
not related to any specific problem.

Of course the methodology is not free of controversy or drawbacks, the quantity of code generated can be
enormous and less general compared with the manual counterpart, this implies bigger source code files and higher
compilation and linking times.
In order to compute the AD we need a Symbolic and Algebraic Computational Systems (SAC) for the manip-
ulation of mathematical expressions in symbolic form. Some know commercial SAC systems are Mathematica or
Maple. From the Open Source Software (OSS) side, we can find the open-source symbolic Python based package,

Vicente Mataix Ferrándiz Page 305 of 374

C.2 Mathematical concepts APPENDIX C. AUTOMATIC DIFFERENTIATION

Sympy [Onl]. SAC systems cannot be used in directly to define the solution of the problem, as in the case of complex
engineering problems, the uncontrollable growth of expressions and consequently redundant operations leads to
inefficient codes.
In addition we can mention the Hybrid Object-Oriented (HOO)[BookJo16] approach, an intermediate solution.
This approach has brought a new perspective for the development of complex software and several OO FE environ-
ments have been developed. The HOO systems are in general restricted to a particular type of formulations where
the strong form to the element equations can be defined. With this the expression growth problem mentioned for SAC
is reduced, since the symbolic code derivation is used on the resolution of simpler sub-problems. The most relevant
examples of this approach can be found on the OSS environments of FreeFem++ (Hecht[ArtHec+09]) and FEniCS
(Logg[ArtLog07]).

C.2 Mathematical concepts

Figure C.1: AD concept

The NR method (C.1) gives us quadratic convergence(C.2).



 We start with u0

 ∂R
R(ui + ∆ui ) ≈ R(ui ) + ∂ u
(ui )∆ui = 0


−1
(C.1) Solving → ∆ui = − ∂∂Ru R(ui )




 ui+1 = u i + ∆ u i

Convergence when: k∆ui k < ǫu or/and kRk < ǫr


kxk+1 − xsol k
(C.2) lim = M, with M > 0
k →∞ kxk − xsol k2

Then (C.1) gives us the definition of the RHS and the LHS (C.3). In consequence it is possible to use the AD to
obtain the exact definition of the system and achieve quadratic convergence

∂δW ∂ RHS
(C.3) RHS = , LHS = −
∂δ u ∂u

Sometimes the LHS matrix is called tangent matrix, this is due to the mathematical concept underlying the
derivation of the RHS vector necessary to obtain the consistent LHS from (C.3), where in a Single Degree Of
Freedom (SDOF) problem the graphic representation of derivatives corresponds with the tangent in the given
functional (See Figure C.2).
From a developer point of view the main advantages of this technique is the fact that it reduces the development
time, reduces the computation time, and gives us robust and consistent code where the step of consistent linearisation

Page 306 of 374 Vicente Mataix Ferrándiz

APPENDIX C. AUTOMATIC DIFFERENTIATION C.3 Implementation

Figure C.2: NR representation for a SDOF

is considerably simplified. Furthermore, the time that will be required modify the code will be only the needed to
modify the function to derive with AD, or the only thing that is actually debugged is the functional, not the generated
code. This technique has been already used successfully for contact problems by Jakub Lengiewicz[ArtLKS11;
ArtSLK10].

C.3 Implementation
In the implementations presented in this work for the AD derivations, we have used Sympy [Onl]. The quality of the
code obtained is lesser than in the case of the code developed by Korelc[ArtKor97; ArtKor02] AceGen[Onl], but we
have a full control of the whole workflow and the implementations actually done.

In the following section, we will introduce the principles of AD (C.3.1.Principles of Automatic Differentiation), which
will help us to understand the design choices taken of the last section. The last section (C.3.2.Kratos integration) we
introduce the integration between Kratos and Sympy and the modular design followed in order to consider properly
the derivative definition without explicit definition of the values involved and get a fully quadratic convergence.

C.3.1 Principles of Automatic Differentiation

C.3.1.1 Derivation modes

As we said previously, the AD is based on the execution of elementary operations with known derivatives, each one of
them can be evaluated exactly with the chain rule, independently of the complexity of the formulation. We can define
two procedures[BookJo16], the forward mode and the backward mode. We can show this using the example shown
in [BookJo16], taking as reference the equation from (C.4a).

n
X
(C.4a) f = bc with b = ai2 and c = sin(b)
i=1

Vicente Mataix Ferrándiz Page 307 of 374

C.3 Implementation APPENDIX C. AUTOMATIC DIFFERENTIATION

Considering ai as n independent variables. The forward mode (C.4b) accumulates the derivatives of intermediate
variables with respect to the independent variables.

db
∇b = = 2xi for i = 1, 2, · · · , n
dai
dc
(C.4b) ∇c = = cos(b)∇bi for i = 1, 2, · · · , n
dai
df
∇f = = ∇bi c + b∇ci for i = 1, 2, · · · , n
dai
∂f
On the other hand, the backward mode, (C.4c), propagates adjoints x̄ = ∂x , which are the derivatives of the final
values, with respect to intermediate variables.

df
f̄ = =1 1
df
df ∂f
c̄ = = f̄ = bf̄ 1
dc ∂ c
(C.4c)
df ∂f ∂c
b̄ = = f̄ + c̄ = c f̄ + cos(c)c̄ 1
db ∂ b ∂b
∂b
∇f = āi = b̄ = 2ai b̄ for i = 1, 2, · · · , n
∂ ai

The implementations on C.3.2.Kratos integration consider basically the forward mode. A proper implementa-
tion[BookJo16] makes use of both approaches and combine them in order to obtain the most efficient approach
according to the estimated work ratio (C.5). On this equation the work ratio of the forward mode is proportional to the
number of independent variables, and in the case of backward mode to the number of scalar-valued functions.

∂f
cost(f (a), ∂a )
(C.5) wratio (f (a)) =
cost(f (a))

C.3.1.2 Automatic Differentiation exceptions

Here we introduce the concept of AD exceptions, these exceptions appear on the derivative of an implicit dependence.
For example, let’s consider the following nonlinear equation (C.6a), which we will solve via the NR method. If we
assume that (C.6a) depends on an additional independent variable a, as shown on (C.6b). If u depends on a, this will
introduce an additional implicit dependence on R (C.6c). This kind of implicit dependency appears in many non-linear
problems, for example on sensibility analysis or in order to obtain a fully consistent tangent on elasto-plasticity
problems or in contact problems in large deformations and the boundary moves significantly. In order to compute
the derivatives of this implicit dependence, we can apply the chain rule as in (C.6d). This can be solved by a direct
application of the AD, but an alternative approach is to take into account the AD exceptions.

(C.6a) R(u) = 0

(C.6b) R(u, a) = 0

(C.6c) R(u(a), a) = 0

DR ∂ R ∂ u ∂ R
(C.6d) = + =0
Da ∂u ∂a ∂a

Page 308 of 374 Vicente Mataix Ferrándiz

APPENDIX C. AUTOMATIC DIFFERENTIATION C.3 Implementation

 − 1
∂u ∂R ∂R
(C.6e) =−
∂a ∂u ∂a

If R is defined by an algorithm then we can obtain the tangent matrix needed in the NR scheme can be obtained
by AD. Let’s define the computational derivative concept (C.7a). Then we can define the computational derivative of
(C.6b) on (C.7b). It is here where we can proceed in two ways, or directly deriving and solving in each step (C.6e),
which can be very costly, or we can apply the AD exceptions.

δ̂ (·)
(C.7a)
δ̂ (·)

∂R δ̂ R ∂R δ̂ R
(C.7b) := , and :=
∂u δ̂ u ∂a δ̂ a

There are two types of AD exceptions, the local exception (C.8a) or the global exception (C.8b). The difference
between both exceptions is basically as the name indicates that in the case global exception the exceptions are
defined globally, hence valid for every AD procedure call, while it applies only for the specific call. The approach
followed in our implementation would be closer to this last exception, as the derivatives are defined in a modular way.
For more details follow the next section.

Df δ̂ f (a, u(a))
(C.8a) :=
Da δ̂ a Du
=−

δ̂ R
−1
δ̂ R
Da δu δa

u := u| Du 
δ̂ R
 −1
δ̂ R
Da
=− δ̂ u δ̂ a

···
(C.8b)
···
Df δ̂ f (a, u(a))
:=
Da δ̂ a

C.3.2 Kratos integration

The classical way of optimising expressions[BookJo16] in Computer Algebra Systems (CAS) are searching for
common sub-expressions before the generation of code is done. This can be insufficient when applied to general
non-linear problems. This means that in the case of using Sympy environment directly, it is not possible to obtain
complex or highly non-linear FE with this approach.
The design of our AD implementation has in mind these restrictions, and because of this, several utilities which
integrate the Sympy environment has been implemented into Kratos. These utilities introduce simpler techniques
than the ones developed Korelc for AceGen1 , but enough in order to obtain a relatively complex formulation.
The approach followed in order to take into account the AD exceptions (C.3.1.2.Automatic Differentiation ex-
ceptions) is the definition of the derivatives (4.6.Derivatives for contact mechanics linearization) as independent
computations.
In order to do that, on the Sympy implementations we have defined an implicit dependency respect of the DOF of
the terms with derivatives, defined in a standalone way. Those algebraic expressions are carried during all the AD
procedures and finally replaced with the corresponding exceptions on the last stage of the code generation.
1 Like the techniques of Simultaneous Stochastic Simplification[BookJo16].

Vicente Mataix Ferrándiz Page 309 of 374

C.3 Implementation APPENDIX C. AUTOMATIC DIFFERENTIATION

This integration can be represented in a UML diagram, Figure C.3. On this Figure we can distinguish the previously
stated steps. We define a base template in C++ code, which is empty, and the AD generator created based on
Sympy fills the template based in a given Galerkin functional. The derivatives are computed a priori, and the code
generated includes them implicitly until the C++ code conversion steps, where are replaced with the externally
computed values.

Figure C.3: AD workflow in a UML diagram

Page 310 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookJo16] Automation of Finite Element Methods. Peter Wriggers (auth.) Joe Korelc. Springer International
Publishing. 1st ed. 2016.
[BookWri08] Nonlinear finite element methods. Peter Wriggers. Springer. 1st ed. 2008.

Articles
[ArtGru82] “Automatic Differentiation: Techniques and Applications. Lecture Notes in Computer Science 120.”
F. Grund. In: ZAMM - Journal of Applied Mathematics and Mechanics / Zeitschrift für Angewandte
Mathematik und Mechanik . No. 7, Vol. 62, 1982, pp. 355–355. DOI:
10.1002/zamm.19820620735.
[ArtHec+09] “Freefem++, ver. 3.7”. F Hecht, O Pironneau, A Le Hyaric, and K Ohtsuka. In: . 2009,
[ArtKor02] “Multi-language and Multi-environment Generation of Nonlinear Finite Element Codes”. J. Korelc.
In: Engineering with Computers. No. 4, Vol. 18, 2002, pp. 312–327. DOI:
10.1007/s003660200028.
[ArtKor97] “Automatic generation of finite-element code by simultaneous optimization of expressions”.
Joe Korelc. In: Theoretical Computer Science. No. 1, Vol. 187, 1997, pp. 231–248. DOI:
10.1016/S0304-3975(97)00067-4.
[ArtLKS11] “Automation of finite element formulations for large deformation contact problems”.
Jakub Lengiewicz, Joe Korelc, and Stanisaw Stupkiewicz. In: International Journal for Numerical
Methods in Engineering. No. 10, Vol. 85, 2011, pp. 1252–1279. Wiley Online Library. DOI:
10.1002/nme.3009.
[ArtLog07] “Automating the finite element method”. Anders Logg. In: Archives of Computational Methods in
Engineering. No. 2, Vol. 14, 2007, pp. 93–138. Springer. DOI: 10.1007/s11831-007-9003-9.
[ArtSLK10] “Sensitivity analysis for frictional contact problems in the augmented Lagrangian formulation”.
Stanisaw Stupkiewicz, Jakub Lengiewicz, and Joe Korelc. In: Computer Methods in Applied
Mechanics and Engineering. No. 33-36, Vol. 199, 2010, pp. 2165–2176. Elsevier. DOI:
10.1016/j.cma.2010.03.021.

Online resources
[Onl] AceGen. Multi-language, Multi-environment Numerical Code Generation. URL :
https://www.wolfram.com/products/applications/acegen/.
[Onl] SymPy is a Python library for symbolic mathematics. URL : https://www.sympy.org/.

Vicente Mataix Ferrándiz Page 311 of 374

ONLINE RESOURCES BIBLIOGRAPHY

Page 312 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Appendix D

Constrained optimisation problems

The mathematical sciences

particularly exhibit order, symmetry,
and limitation; and these are the
greatest forms of the beautiful.

Aristotle
(384 - 322 BC, Greek philosopher)

D.1 Introduction
The following appendix introduces the concepts concerning the constraint optimisation needed in the resolution of
the contact problems. These optimisation methods are of general use from a mathematical point of view and can be
found extensively on the literature[BookLei04; ArtQS99], especially to the kind of problems we are interested to solve
that are the contact problems[BookAnd01; BookYas13]. The main problematic motivating us to study these constraint
optimisation methodologies is the existence of many physical phenomena such as the impact, friction, among others;
which can be studied with mathematical models with some kind of discontinuous or non-smooth behaviour. In detail,
this problem suffers of bifurcation points, therefore, may lead to non-smooth dynamic systems, where the dynamic in
each mode is associated with a different set of smooth differential equations[BookLei04].
The methods we are going to present in this annexe are the following ones:

• Penalty method (also known as exterior point method[PhDYas11]). This is probably the simplest and probably
most extended method. The last statement comes particularly true when the method can deal with explicit
contributions, then used on explicit simulations. For more details and deeper explanation see D.2.Penalty
method.

• Lagrange multiplier method. The LM can be given a rigorous justification within the context of the variational
calculus[BookCA01], and in contrast to penalty method gives an exact solution. How it is derived and several
additional details are presented on D.3.Lagrange Multiplier method.

• Augmented Lagrange Multiplier method. The ALM basically combines the former methods, for a deeper
understanding see D.4.Augmented Lagrange Multiplier method.

There is another method that we consider is important to highlight, this is the resolution of MPC (see D.5.MultiPoint
Constraint (Master-Slave elimination method)) with the master-slave elimination method. This method is not an
optimisation method by itself, but can be used in order to impose constraints of interest in the resolution of mechanical
problems. The main reason to mention it in addition to the previous methods, it is due to the fact is available on

Vicente Mataix Ferrándiz Page 313 of 374

D.2 Penalty method APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

B.2.Kratos Multiphysics natively. Unfortunately, it is not particularly good to deal with complex contact constraints as it
will be shown later.
There are many methods omitted on this list1 , as we said we will present and study just the methods used in the
developments concerning this work.

D.2 Penalty method

D.2.1 Introduction
The PM method can be idealised as the presence of fictitious elastic structural elements (Figure D.1) which enforce
the constraint approximately for mechanical problems[BookCA01]. In the most general case, it can be defined as
penalty functional appended to our functional of interest, which increases accordingly to how severely the constraint is
violated[BookKO88].

Figure D.1: Analogy between penalty method and springs. Image from [BookYas13]

D.2.2 Formulation
In order to formulate a generic constrained problem and solve it with the PM we just need to add the following to our
functional (D.1).
We start with the following base functional and the following constraint. It can be extended to any number of
constraints, in fact, at contrary to other optimisation methods LM can deal with overconstrained problems (this does
not mean it will fulfil all of them satisfactory).
(
f (w, x) = Base functional
(D.1a)
constraint ≤ 0

Where w and x are the test function and DOF respectively. Then we add the following penalty functional to our base
functional. The objective of this penalty term is precisely to penalise the infringement of this constraint.
ε
(D.1b) fp (w, x) = f (w, x) + max(0, constraint)2
2
Where ε is our penalty parameter. Simple, yet powerful and generic.
As in any functional, in order to deduce the corresponding LHS and RHS we do the following (D.2) (as follows
it will be the same for the rest of methods). This is already stated on the appendix concerning AD (C.Automatic
differentiation).

∂ fp (w, x)
(D.2a) RHS(x) =
∂w

∂ RHS(x)
(D.2b) LHS(x) = −
∂x
1 In [PhDYas11; BookWri06] we can see the following methods as an extension of the list presented: Barrier method (also known as interior

points method), perturbed Lagrangian[ArtSWT85a], Nitsche[ArtNit71] method, cross-constraint method.

Page 314 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.2 Penalty method

D.2.3 Applicability on contact problems

The penalty method is general and powerful enough to be applied on contact problems. So in case we want to
introduce it into a contact problem we can do the following (D.12).
We define the non-penetration condition, together with the HSM.
(D.3a) g ≥ 0, σn ≤ 0, g σn = 0, σt = 0
In order to fulfil these conditions, we can define the contact pressure as a continuous function of the penetration.


g ≥ 0, σn = 0, g σn = 0
(D.3b)
g < 0, σn = εn (−g) < 0, g σn 6= 0
This approximation implies that the non-penetration condition is not respected, but the penetration movement is
resisted, the deeper is the penetration the stronger is the reaction. Then the energy accumulated on these continuous
linear springs is:
Z −h−g i Z −h−g i
1
(D.3c) fp (x) = f (x) − εn h−g ′ i dg ′ = εn g ′ dg ′ = εn h−g i2
0 0 2
Where hi are the Macaulay brackets. Integrating over the boundary, over the normal direction, the contribution to the
balance of virtual works writes as:
Z Z


(D.3d) δ fp (x) = δ f (x) + εn (−gn ) δ gn d Γ1c = εn h−gn i δ gn d Γ1c
Γ1c Γ1c

D.2.3.1 Adapted Penalty Method

On (D.12) the ε is constant, in contrast we can adapt dynamically the value of ε in order to improve the convergence
of the system. Bussetta[ArtBMP] presents the Adapted Penalty Method (APM) as a plausible method from doing
so. The concept is summarised on the Figure D.2, where the function F that defines the ε varies in function of the
penetration (gn ), see (D.4).

Figure D.2: Representation of F . Image from [ArtBMP]

(D.4a) εni+1 = F |gi | , gmin , gmax εni


| gi |

 gmax
if |gi | > gmax

| gi |
(D.4b) F |gi | , gmin , gmax = if |gi | < gmin
 gmin
1 else


This is important because as we have previously stated, the choice of the penalty coefficients is of the utmost
importance in order to get an effective solution. The smaller these coefficients are, the result obtained may not
respect the imposed contact constraint, allowing penetrations. On the other hand, a large value of these coefficients
may induce numerical oscillations and ill-conditioning of the system of equations, preventing the convergence of the
algorithm. There are other alternative proposals to this in order to improve the PM, with more layers of complexity, like
for example the consideration of a double penalty[PhDHet14].

Vicente Mataix Ferrándiz Page 315 of 374

D.3 Lagrange Multiplier method APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

D.3 Lagrange Multiplier method

D.3.1 Introduction
The LMM[BookIK08], named after Joseph-Louis Lagrange, is used on optimisation theory to find the extremum, saddle
point, of a functional subjected to equality constraints[BookAkh88]. The basic idea is to convert a constrained
problem into a form such that the derivative test of an unconstrained problem can still be applied[OnlLag], the obtained
functional is called Lagrangian (L(x, λ)). On this Lagrangian functional, we add a new unknown called Lagrange
multiplier, usually referred as λ. The great advantage of this method is that it allows the optimisation to be solved
without explicit parameterisation in terms of the constraints.

D.3.2 Formulation
The Lagrangian is constructed in the following manner (D.5b), this for problems where the constraint is not limited by
any inequality. For problems where we deal with inequality constraints, we need to formulate the problem following
some kind of active set strategies[PhDYas11], we can define three types of PDASS(D.5d), graphical examples for a
better understanding of the problem will be presented on the section D.7.2.Non-linear spring contact problem with
wall.
We want to minimize f subjected to g.

(D.5a) min f (x) → ∇Lλ (x, λ) = 0

g(x)=0

Where the Lagrangian is defined as follows:

(D.5b) Lλ (x, λ) = f (x) + λg(x)

The gradient (∇) of L can be defined as:

   
∂Lλ ∂ f (x)
+ λ ∂∂g(x)
(D.5c) ∇Lλ (x, λ) = ∂x
∂Lλ = ∂x x =0
∂λ
g(x)

In case we wan to to consider to solve a problem with inequality constraint we can consider the following three active
set strategies (PDASS), where hi denotes the Macaulay brackets, used to describe the ramp function:

Active set strategy 1: g(x) > 0 : f (x) g(x) ≤ 0 : Lλ (x, λ)

(D.5d) Active set strategy 2: λ > 0 : f (x) λ ≤ 0 : Lλ (x, λ)

Active set strategy 3: g(x) > 0 and λ > 0 : f (x) g(x) ≤ 0 or λ ≤ 0 : Lλ (x, λ)


These strategies, combined with (D.5b) can be expressed the following way, with the Macaulay bracket.

Active set strategy 1: Lλ (x, λ) = f (x) − h−λi g(x)

(D.5e) Active set strategy 2: Lλ (x, λ) = f (x) − λ h−g(x)i

Active set strategy 3: Lλ (x, λ) = f (x) − h−λi g(x) − λ h−g(x)i − hλi hg(x)i


The influence of each of the PDASS can be summarised as:

1. This one is based on the check of the violation of the g, it is the most commonly used due to its robust-
ness[PhDYas11], but with a higher number of NL iterations in order to converge.

2. This one checks the positivity of λ. This one may lead to the continuous switch between the base functional (f )
and the Lagrangian Lλ , lacking the robustness provided by the first strategy, even despite of having a faster
rate of convergence on ideal conditions.

3. The third alternative provides the robustness of the first strategy and the rate of convergence of the second one,
but also may diverge faster in case the initial solution is far from the final one. This last point forces to consider
a slow loading of the problem, or a fine-tuning in the BC.

Page 316 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.3 Lagrange Multiplier method

The solution of our minimisation problem is a stationary point in (D.5b), but usually not all stationary problems
of the Lagrangian(D.5b) are solutions of the initial minimisation problem. The resulting system of equations has a
higher number of unknowns with the corresponding additional computational cost. Additionally if we compute the
Hessian (H) of the Lagrangian functional, we will obtain the corresponding LHS of our optimisation problem, being the
gradient equivalent to minus RHS.
First define H as the derivative of the gradient
 
∇g1 (x)T
 ∇g2 (x)T 
(D.6a) H(x) = ∇g(x) = 
 
.. 
 . 
∇gn (x)T
Then we can compute the H from the gradient (D.5c).
" # " #
∂ 2 Lλ ∂ 2 Lλ ∂ 2 f (x) ∂ g(x)
(D.6b) H(Lλ (x, λ)) = ∂x 2 ∂λ∂ x = ∂x 2 ∂x
∂ 2 Lλ ∂ 2 Lλ ∂ g(x)
∂ x ∂λ ∂λ 2 ∂x 0

On (D.6) the H of L is shown, and at (D.6b) we can see the typical LHS structure for a problem solved with LMM.
In here the equations have a zero diagonal for each multiplier term. Thus, special care is needed in the solution
process to avoid division by the zero diagonal[BookZTF14]. This issue is partially solved with the use of ALM.

D.3.3 Applicability on contact problems

Figure D.3: Lagrangian function for the contact problem

Following the same reasoning of D.2.3.Applicability on contact problems, we can define a functional for contact
mechanics using the LMM. The HSM conditions (D.3a) are still applied here, then we can obtain the corresponding
Lagrangian (D.7).
Defining λn as the normal Lagrange multiplier which defines the normal contact pressure.
Z
(D.7a) Lλ (x, λ) = f (x) + λgn (x)d Γ1c
Γ1l

The variation of the Lagrangian then can be expressed as:

Z
(D.7b) δLλ (x, λ) = δ f (x) + gn (x)δλ + λδ gn (x)d Γc = 0
Γc

Vicente Mataix Ferrándiz Page 317 of 374

D.4 Augmented Lagrange Multiplier method APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Where λn and gn have to be fulfilled the minimisation problem with inequality constraints (D.3a). This means that from
(D.5d) the preferable strategy would be the third one.
We can represent (D.7) on the locus2 of Figure D.3, where (D.7) is represented on function of gn and λn . On this
figure we can appreciate the fact that on tension state (λn > 0) there is no solution to the functional (zero solutions),
this generates a discontinuity which affects the resolution the resulting system of equations. This last problematic can
be solved precisely with the application of ALM.

D.4 Augmented Lagrange Multiplier method

D.4.1 Introduction
The ALM was originally introduced by Arrow and Solow[ArtAS58], and later improved by Powell and Hestenes[ArtPow69;
ArtHes69], reason why was originally named as the multiplier method of Hestenes and Powell. The method consists
on a regularised version of LMM[PhDYas11]. This regulazations occurs via the inclusion of penalty parameter, but
once the problem has reached convergence the influence of the penalty parameters disappears, resulting in exact
fulfilment of the constraint.

D.4.2 Formulation
D.4.2.1 Standard formulation
The Lagrangian is constructed in the following manner (D.8b), this is basically the combination of the PM and the
LMM (D.8a). This method was generalised by Rockafellar [ArtRoc73b; ArtRoc73a] for inequality constraints (D.8c), in
this particular case g(x) ≥ 0.
The Lagrangians for the PM and LMM as previously redefined are:

1
(D.8a) fp (x) = f (x) + εg(x)2 ; Lλ (x, λ) = f (x) + λg(x)
2
Combination of previous methods, we obtain the augmented Lagrangian.

1 1
(D.8b) Lλ̄ (x, λ) = Lλ (x, λ) + εg(x)2 = f (x) + λg(x) + εg(x)2
2 2
Expressed as inequality constraints g(x) ≥ 0:

1  2 2

(D.8c) Lλ̄ (x, λ) = f (x) − λ − h− (λ + εg(x))i
2ε
In an expanded form, it rewrites as:

λg(x) + 21 εg(x)2 , λ + εg(x) ≤ 0
(D.8d) Lλ̄ (x, λ) = f (x) +
− 21ε λ2 , λ + εg(x) > 0

We can call λ + εg(x) as λ̄, or augmented Lagrangian. The gradient (∇) of (D.8d) can be expressed as:
  
(λ + εg(x)) ∂∂g(x)
, λ̄ ≤ 0

 x

(D.8e) ∇Lλ̄ (x, λ) = ∇f (x) +  g(x)
 0

 , λ̄ > 0
− λε

Equivalent to what is done in (D.6) we can obtain the Hessian (H) of (D.8b). In (D.9) we see that in this case, we
have partially solved the issue concerning to the zero terms appearing on the diagonal, compared to the previously
obtained H. The problematic have been solved partially because only affects to the inactive set (λ̄ > 0), in the case
2A locus is the set of all points (usually forming a curve or surface) satisfying some condition.

Page 318 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.4 Augmented Lagrange Multiplier method

of the active set we still have to deal with this issue. A Generalized Newton-Raphson method (GNM) was proposed
by Alart and Curnier for nonsmooth potential for the ALM[ArtAla88; ArtAC91; ArtAla97].

 "  2 #
∂ 2 f (x) ∂ 2 g(x) ∂ g(x) ∂ g(x)
 + (λ + εg(x)) + ε
"
∂ 2 Lλ̄ ∂ 2 Lλ̄
# 
 ∂x 2 ∂x 2 ∂x ∂x , λ̄ ≤ 0
 ∂ g(x)
(D.9) H(Lλ̄ (x, λ)) = ∂x 2 ∂λ∂ x = " # ∂x 0
∂ 2 Lλ̄ ∂ 2 Lλ̄ ∂ 2 f (x)
0

∂ x ∂λ ∂λ2 

 ∂x 2
1 , λ̄ > 0
0

ε

D.4.2.2 Uzawa iteration

So as we have seen, we still have some problematic from the standard LMM, like the zero diagonal terms or the non-
positivity of the Lagrange multiplier. In order to solve that, on 1958 the Japanese economist Uzawa presented[ArtAS58;
ArtAHU58] an alternative approach in order to solve this problematic following an iterative approach, known as Uzawa
algorithm[Onl].
Let’s decompose the increment of the Lagrangian into two different components, one for the x DOF and the other
for the update of the Lagrange multiplier (D.10).
Calling λi and xi the respective solution for the NL iteration number i. Then we can define the augmented
Lagrangian in a given increment of a NL iteration as:
∂Lλ̄ (x, λi )

(D.10a) Lλ̄ (xi + ∆xi , λi + ∆λi ) ≈ Lλ̄ (xi + ∆xi , λi ) + ∆λi + O λ2i = 0
∂λ λi

We can do the same for the ∆x term, and expand it.

Lλ̄ (xi + ∆xi , λi ) ≈ Lλ̄ (xi , λi ) + ∂Lλ̄∂(x,

x
λi )
∆ xi = 0
(D.10b) h i h i xi

f (xi ) + λi g (xi ) + 12 εg (xi )2 + ∂∂f (x)
x
+ [λi + εg(x)] ∂∂g(x) x
]|xi ∆xi + O xi2 = 0

Finally, concerning the Lagrange multiplier we apply an update procedure:

(D.10c) λi+1 = λi + εg (xi ) , ∆λi = εg (xi )
Convergence of this method is linear for the part relative to the Lagrange multiplier (O (λi )), as we have removed the
second order part of (D.10a). Besides the advantage of the method is that the L is smooth so that a standard NR is
applicable.
For the present work, the standard approach has been taken into consideration. On the contact chapter of this
work the details concerning why the standard ALM approach has been chosen instead of the Uzawa iteration, check
the corresponding section for further details, 4.3.3.2.1.3.ALM.

D.4.3 Applicability on contact problems

Following a similar criteria to the one employed on (D.7), the equivalent formulation can be obtained for the ALM , see
(D.11). In contrast to the locus for the LMM, the corresponding locus for the ALM (Figure D.4) is C 1 differentiable
saddle-point, with the corresponding advantages for the resolution of the problem. Like the existence of a system of
equations contributions (LHS, RHS) even for λ̄n > 0.
Again, considering λn as the normal contact stress, and calling λ̄n the augmented contact stress, computed as
λ̄ = λ + εgn (x).
Z 
λgn (x) + 12 εgn (x)2 , λ̄ ≤ 0
(D.11a) Lλ̄ (x, λ) = f (x) + d Γ1c
Γ1l − 21ε λ2 , λ̄ > 0

The variation of the augmented Lagrangian then can be expressed as:

Z 
δλgn (x) + (λ + εgn (x)) δ gn (x) , λ̄ ≤ 0
(D.11b) δLλ̄ (x, λ) = δ f (x) + d Γc = 0
Γc − λε δλ , λ̄ > 0

Vicente Mataix Ferrándiz Page 319 of 374

D.5 MultiPoint Constraint (Master-Slave elimination method)
APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Figure D.4: Augmented Lagrangian function for the contact problem

D.4.3.1 Adapted Augmented Lagrangian Method

Algorithm 7 Adaptation of normal penalty coefficient[ArtBMP]

Require: εn , gi and gi −1
1: procedure A DAPTATION OF NORMAL PENALTY COEFFICIENT
2: if gi × gi −1 < 0 then
3: if gi × gi −1 < 0 then

4: εn = (εn gi −1 ) /gi × |gi | + gmax / (gi − gi −1 )
5: else
6: εn = |εn gi −1 / (10gi )|
7: else if gi > gmax then

8: if |gi − gi −1 | > max gi /10, gi −1 /10, 5gmax then
9: εn = 2 ε n
10: else if |gi | = |gi −1 | ± 1% < 10gmax then
q
2
11: εn = ε n |gi | /gmax − 1 + 1
12: else if gi > gmax then

13: εn = 2εn gi −1 /gi
14: else q 


15: εn = ε n |gi | /gmax − 1 + 1
16: else
17: εn = εn

On the expressions from (D.11) it is assumed that the ε is constant, but there are techniques which adapt
dynamically the value of ε in order to improve the convergence of the system. We can mention in particular the
work of Bussetta[PhDBus; ArtBMP] who created AALM, an ALM with a corresponding algorithm in order to update
the ε in function of the current penetration (gn ). On Algorithm 7 the procedure is detailed, here three cases can be
distinguished: The sign of gn changes (gi × gi −1 < 0), the absolute value of gn is more relevant than the defined limit
(gi > gmax ), or the last case where the absolute value of gn is smaller than the limit (gi > gmax ).

D.5 MultiPoint Constraint (Master-Slave elimination method)

Page 320 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS
D.5 MultiPoint Constraint (Master-Slave elimination method)

D.5.1 Introduction
The MPC is a type of MultiFreedom Constraint (MFC)3 where each several displacements (or DOF) per node[BookCA01].
The nature of these constrains can be linear if all the displacement components appear linearly on the LHS, and non-
linear otherwise. These relationships can be solved using the former optimisation methods, but in D.5.2.Formulation
we will present the master-slave method from Felippa[BookCA01] book.

D.5.2 Formulation
Following the notation is taken from [BookCA01], and can be summarised in the following way (D.12).
Our system of equations looks like (D.12a).

(D.12a) LHS∆x = RHS onwards LHS = A and RHS = b

If we define relation matrix T as the following:

(D.12b) x = Tx̄ + g

That in a incremental approach leaves the following:

(D.12c) xi = xi −1 + ∆x = Tx̄ + g → xi −1 + ∆x = T (x̄i −1 + ∆x̄) + g

If the previous time step is converged then xi −1 = Tx̄i −1 + g then simply:

(
i = 0, g
(D.12d) ∆xi = T∆x̄i +
i 6= 0, 0

This brings us to the following eliminated system:

( !
i = 0, Ag
(D.12e) Ā∆x̄ = b̄ in which Ā = TT AT, b̄ = TT b−
i 6= 0, 0

D.5.3 Applicability on contact problems

(a) Patch test not passing (b) Not matching meshes giving bad quality results
Figure D.5: Patch tests solutions obtained with Master -Slave elimination method
In order to consider the MPC to compute the contact constraint, we need to deactivate them once the residuals
concerning the corresponding DOF indicates the contact constraint is on tension, and them not respecting the
unidirectionality of the contact constraint which requires compression.
3 This can be defined as functional equation where two or more DOF are written in a functional which connects them.

Vicente Mataix Ferrándiz Page 321 of 374

D.6 Summary of the different methods APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Unfortunately, the MPC method presented with Master -Slave elimination method does not give a good solution
once the meshes are not matching (similar to a NTN method) for a problem of deformable domains. The solutions
presented on Figure D.5 represent the solution obtained with this method. The weights have been obtained with a
mortar segmentation, therefore exact, and the results obtained are not good.
The first case, Figure D.5a, represents a 3D patch test, with not matching meshes, in this case the constraint
is fulfilled, but as both domains are deformable the resulting solution is not continuous. The solution from Figure
D.5b deals with the same problem, but with a higher number of DOF, then the bad quality of the results is even more
noticeable.
In order to obtain good results with the methods, the master domain must be rigid, or quasi-rigid, or the meshes to
match between the interface of the domains.

D.6 Summary of the different methods

The presented methods can be summarised in the following table, see D.1, inspired and expanded from [BookCA01].
ALM and LMM are virtually identical except for the dependence of the ALM of the penalty (not affecting the final
solution). Additionally the ALM retains partially the positive definiteness due to the fact that adds relatively small
terms on the diagonal of the LHS matrix.

Method Master-Slave Elimination PM LMM ALM

Generality Fair Excellent Excellent Excellent
Ease of implementation Poor to fair Good Fair Fair
Sensitivity to user decisions High High Small to none Small
Accuracy Variable Variable Excellent Excellent
Sensitivity as regards constraint dependence High None High High
Retains positive definiteness Yes Yes No Partially
Table D.1: Comparison of different methods presented

Apart from ALM there are other methods that try to tackle the issues presented by LMM. In short we can list the
following two methods:

• Double Lagrange multiplier: The Lagrange doubles4 (used in the Castem2000 code) avoids the issues
from the LMM provoked by the non-positive LHS matrix obtained with the method. This method duplicates
the number of Lagrange multipliers and adds some dummy terms to the LHS matrix, which increases the
computational cost but makes the LHS positive defined. For further details see [Onl].

• Perturbed Lagrangian: Simo, Wriggers and Taylor proposed a perturbed Lagrangian formulation for the
solution of contact problems. That formulation can be classified as a stabilised method, preserving the stability
of the discretised problem if the penalty parameter, ε, is small enough. For further details see [ArtSWT85b].

D.7 Numerical examples

In this section we compare the different approaches introduced (except for D.5.MultiPoint Constraint (Master-Slave
elimination method)). The problems vary on complexity, and helps us to understand the way these different methods
behave when computing the corresponding contribution of the constraints.

D.7.1 Initial spring wall problem

We start with a very simple case from[PhDYas11]. This problem consists on a SDOF problem, where we idealised a
single spring with a wall constraint that does not allow a penetration, see Figure D.6, which means that x ≥ 0. The
function can be defined as follows (D.13), where k represents the stiffness of the spring. The base functional can be
defined as f (x) = 21 k (x + 1)2 . Taking k = 2.
4 Not to be confused with the dual Lagrange multiplier presented on this work.

Page 322 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

Figure D.6: Simplified contact problem within spring and a wall

(
f (x) = 21 k (x + 1)2
(D.13)
Subjectted to x ≥ 0

D.7.1.1 Penalty method

Figure D.7: Solution for a SDOF solved with PM, comparing different values of ε
This can be formulated as follows for PM.
1 ε
2
(D.14) fp (x) = k (x + 1)2 + max {0, x }
2 2

Taking (D.14) and applying (D.2) iteratively with a NR we can solve the problem. On Figure D.7 we can appreciate
the solutions obtained for different values of ε. It can be seen as ε grows the solution comes closer to the actual

Vicente Mataix Ferrándiz Page 323 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Iteration x fp
1 1 4
2 −1 106
3 −2 × 10−6 8.40465 × 10−11
Table D.2: Convergence on fp with PM for (D.13)

solution of the problem. This problem takes in total three NL iterations to converge starting with x = 1 for a ε = 106 ,
see Table D.2, and even with this value the constraint is not completely satisfied.

D.7.1.2 Lagrange multiplier method

Now we can formulate the corresponding Lagrangian of (D.13) for the LMM. The (D.15) shows the proper potential of
our problem with LMM. In this example we will only show one of the possible alternatives in order to compute the λ
constraint, the next section D.7.2.Non-linear spring contact problem with wall will show the three possible alternatives
available when computing the LMM contribution.

1
(D.15) Lλ (x) = k(x + 1)2 + min {0, λ} x
2

Figure D.8: Solution for a SDOF solved with LMM

Iteration x λ Lλ
1 1 0 4.12311
2 0 −2 0
Table D.3: Convergence on Lλ with LMM for (D.13)
On the other hand, the convergence of (D.13) with LMM is achieved in only 2 iterations, see Table D.3, reaching
additionally the exact solution and satisfying exactly the constraint imposed. On the Figure D.8 we can see the
full space of solutions x − λ, including the respective constraint and appreciate the convergence in only two NL
iterations.

D.7.2 Non-linear spring contact problem with wall

The following example[PhDYas11; BookYas13], again as seen on Figure D.6, is a simple unidimensional constrained
optimisation problem, which can show us the influence of each one of the methodologies when solving these kinds of

Page 324 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

problems. In this case the base functional is replaced as (D.16). Additionally we will consider the auxiliary variable
x = u + 1, this will help us to simplify the some expressions.

1 4 1
(D.16) f (x) = x = (u + 1)4
4 4

D.7.2.1 Initial wall

We start with a simple case where the wall is fixed to its initial position, then having as constraint (D.17).

(D.17) Subject to: u ≤ 0 or x ≤ 1

D.7.2.1.1 Lagrange multiplier method :

Here we can analyse each one of the strategies we can consider for the LMM, as presented in D.3.2.Formulation.
In summary, the three options consisted on check the positivity/negativity of the constraint (g), the λ or a combination
of both former methods.

D.7.2.1.1.1 Active set 1 :

Figure D.9: Solution for a NL SDOF solved with LMM. Active set 1
As stated on D.3.2.Formulation, the first strategy consists on checks g, as seen on (D.18). The convergence of
this method is shown on Table D.4, this is presented graphically on the Figure D.9, where the arrows represent each
iteration.

1 4
(D.18a) L(x, λ) = x + λu, x ≤1
4
Expressed with the Macaulay bracket (hi)

1 4
(D.18b) L(x, λ) = x − λh−u i
4

Vicente Mataix Ferrándiz Page 325 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Iteration x λ Error
1 0.0 0.0 1.0
2 1.0 0.0 1.0
3 1.0 1.0 2.2204510−16
Table D.4: First active set strategy convergence on Lλ with LMM for (D.16)

D.7.2.1.1.2 Active set 2 :

In this second case, what is checked is the negativity of λ (D.19). For this case, still quite simple, the NL iterations
give the same result, see Table D.5, which coincides with Table D.4. The convergence path is presented on Figure
D.10.

1 4
(D.19a) L(x, λ) = x + λu, λ≤0
4
Expressed with the Macaulay bracket (hi)
1 4
(D.19b) L(x, λ) = x − h−λiu
4

Iteration x λ Error
1 0.0 0.0 1.0
2 1.0 0.0 1.0
3 1.0 1.0 2.2204510−16
Table D.5: Second active set strategy convergence on Lλ with LMM for (D.16)

Figure D.10: Solution for a NL SDOF solved with LMM. Active set 2

D.7.2.1.1.3 Active set 3 :

In the last active set case, which checks simultaneously both g and λ (D.20). Again, for this simple case, the NL
iterations give the same convergence path, see Table D.6, which coincides with Table D.4 and Table D.5. The path
coincides, see Figure D.11.

1 4
(D.20a) L(x, λ) = x + λu, λ≤0 x ≤1
4

Page 326 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

Expressed with the Macaulay bracket (hi)

1 4
(D.20b) L(x, λ) = x − h−λiu − λh−u i + h−λih−u i
4

Iteration x λ Error
1 0.0 0.0 1.0
2 1.0 0.0 1.0
3 1.0 1.0 2.2204510−16
Table D.6: Third active set strategy convergence on Lλ with LMM for (D.16)

Figure D.11: Solution for a NL SDOF solved with LMM. Active set 3

D.7.2.2 Moved wall

In order to make the problem more difficult and appreciate better the differences across methodologies, we will
partially move the wall which defines our constraint. This example is taken from Yastrebov [PhDYas11], where other
moving configurations are presented apart of the following.

(D.21) Subject to: u + 0.9 ≤ 0.0

Moving the wall 0.9m, the resulting constraint will be (D.21). The resulting case is significantly more complex, and
as it will be shown in the next sections, the convergence rate changes from one method to others, some of them even
diverging the solution.

D.7.2.2.1 Penalty method :

In the case of the PM no tuning is required and the convergence is obtained in a straightforward manner. The only
thing that affects the solution is ε value. On Figure D.12 different solutions are presented for different values of ε, the
trend to the exact solution can be appreciated.

For a value of ε = 106 the correct solution can be achieved, see Table D.7. In here it can be appreciated that on
the iteration 7 we sudden increase on the error, this is due to entering on the penalised region (u > 0.9). In any case
the number of NL iteration needed is quite high, 9 iterations, for a SDOF problem.

Vicente Mataix Ferrándiz Page 327 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Iteration u Error
1 0.0 1.0
2 −0.333333 0.296296
3 −0.555556 0.0877915
4 −0.703704 0.0260123
5 −0.802469 0.00770735
6 −0.868313 0.00228366
7 −0.912209 12208.5
8 −0.9 0.0000410749
9 −0.9 1.7180710−12
Table D.7: Convergence with PM for (D.21)

Figure D.12: Solution for a NL SDOF problem solved with PM, comparing different values of ε. Moved wall

D.7.2.2.2 Lagrange multiplier method :

Following the same reasoning from D.7.2.1.1.Lagrange multiplier method, and applying it for (D.21). On this
section the L to be derived are not detailed, as it follows the same principle. The exact solution of the problem is
u = −0.9 and λ = −0.001.

D.7.2.2.2.1 Active set 1 :

Iteration u λ Error
1 − 31 0 8
27
2 −0.555556 0 0.0877915
3 −0.703704 0 0.0260123
4 −0.802469 0 0.00770735
5 −0.868313 0 0.00228366
6 −0.912209 0 0.0122272
7 −0.9 −0.000958925 0.000041075
8 −0.9 −0.001 2.168410−19
Table D.8: First active set strategy convergence on Lλ with LMM for (D.16) with moved wall

The Table D.8 presents the convergence path that is also represented on Figure D.13. Compared with the
unmoved wall, the problem requires a higher number of iterations in order to converge, and it basically moves across

Page 328 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

the u axis, but at the end it finally converges.

Figure D.13: Solution for a NL SDOF solved with LMM. Active set 1. Moved wall

D.7.2.2.2.2 Active set 2 :

Figure D.14: Solution for a NL SDOF solved with LMM. Active set 2. Moved wall
On this case the problem doesn’t even converge to the right solution, as can be seen on Table D.9, and from
Figure D.14 we can notice as it doesn’t move arrived certain point. If we extend the number of NL iterations to 20 we
don’t get the proper solution either, the NL loop remains around the same values.
On Figure D.14 another relevant point is the fact that the functional is almost flat, explaining part of the difficulty
coming from this active set strategy, it is not possible to find a local minima on these cases.
In the last active set we will see as this problematic is no more. Other initial conditions can be tried for the L, but the
idea is to compare the presented methods across them, and showing the problematic presented on D.3.2.Formulation,
where we introduced that the second active set converges faster than the first one, but in case we are far from the
solution we can deal with the divergence on the NR iteration.

Vicente Mataix Ferrándiz Page 329 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

Iteration u λ Error
1 − 31 0 0.639455
2 −0.9 0.459259 0.001
3 −0.933333 0.459259 0.000296296
4 −0.955556 0.459259 0.0000877915
5 −0.97037 0.459259 0.0000260123
6 −0.980247 0.459259 7.70735−6
7 −0.986831 0.459259 2.28366−6
8 −0.991221 0.459259 6.76639−7
9 −0.994147 0.459259 2.00486−7
10 −0.996098 0.459259 5.94032−8
Table D.9: Second active set strategy convergence on Lλ with LMM for (D.16) with moved wall

D.7.2.2.2.3 Active set 3 :

On the case of the third active set strategy on Table D.10 we have exactly the same convergence path as Table
D.8, despite of that the graphical representation differs, Figure D.15.

Iteration u λ Error
1 − 31 0 8
27
2 −0.555556 0 0.0877915
3 −0.703704 0 0.0260123
4 −0.802469 0 0.00770735
5 −0.868313 0 0.00228366
6 −0.912209 0 0.0122272
7 −0.9 −0.000958925 0.000041075
8 −0.9 −0.001 2.168410−19
Table D.10: Third active set strategy convergence on Lλ with LMM for (D.16) with moved wall

Figure D.15: Solution for a NL SDOF solved with LMM. Active set 3. Moved wall

We can infer in a simplified manner that the third active set strategy will behave as the best option of the two other
alternative strategies, despite this is not always true as already stated on D.3.2.Formulation. The following section will
introduce the ALM resolution, so it will be possible to compare with the presented here for LMM, which in addition to
the convergence of the problem improves the conditioning of the system of equations.

Page 330 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

D.7.2.2.3 Augmented Lagrange multiplier method :

The equation (D.22) shows the resulting L for the ALM method for this problem.

 1
k (u + 1)4 + λ(u + 0.9) + 21 ε(u + 0.9)2 , λ + ε(u + 0.9) ≤ 0
(D.22) Lλ̄ (u, λ) = 2
1
2
k (u + 1)4 − 21ε λ2 λ + ε(u + 0.9) > 0

We will see like the different values of ε give us a similar convergence rate, but the condition number (κ) of the
LHS varies in function of ε. In order to properly prove that we need the definition of κ (D.23), the proper definition of κ
defines it as the ratio of the largest to smallest singular value in the singular value decomposition of a matrix, this
means it requires to compute the SVD of the LHS, which is an expensive operation and complex. We can simplify
and estimate κ as the ratio between the largest to smallest eigenvalue, which are not necessarily coincident.

|λmax |
(D.23) κ(A) ≈
|λmin |

In order to do so, we need then to define the LHS from (D.22), considering (D.2), we can deduce (D.24), recalling
the augmented Lagrange multiplier as λ̄ = λ + ε(u + 0.9).

  T  
 2k (u + 1)3 + λ̄ δu
= 0, λ̄ ≤ 0 (Active)


u + 0.9

δλ
(D.24a) RHS(u, λ) = δLλ̄ (u, λ) =   T  
 2k (u + 1)3 δu
= 0, λ̄ > 0 (Inactive)


 λ
−ε δλ

  T   
δu 6k(u + 1)2 + ε 1 δu
, λ̄ ≤ 0 (Active)


1 0

δλ δλ
(D.24b) LHS(x) = ∆δLλ̄ (u, λ) =  T    
 δu 6k(u + 1)2 0 δu
, λ̄ > 0 (Inactive)


δλ 0 − ε1 δλ

From (D.24b) we can calculate κ, in the case of inactive set is trivial as the LHS is diagonal (D.25a). On this
resulting expression, κ grows linearly with ε. For the active set case, we obtain (D.25b) as expression defining κ. On
this case the condition number of the LHS grows quadratically to the ε. In conclusion, for a high ε, in comparison with
the stiffness of the problem, κ become very high which means we have an ill-conditioned problem, affecting the
precision of the solution and its convergence; but, on the other hand, the energy functional becomes smooth, as we
will state next.

 1
6k εu 2 , ≤ 6ku 2 1
(D.25a) κinactive (LHS) ≈ 1
ε
1 assuming ≤ 6ku 2 then ∼ k ε
6ku 2 ε
, ε > 6ku 2 ε

q 
1

2

(D.25b) κactive (LHS) ≈ 6k(u + 1)2 + ε 6k(u + 1)2 +ε 2
+ 4 + 6k(u + 1) + ε + 1 ∼ (k + ε)2
2

Another important point to highlight is that all the graphical representation of the L shows a continuous field
without flat areas (See Figures D.16, D.17, D.18, D.19). This already gives us an idea how the method already solves
some problems presented on the previous section (D.7.2.2.2.Lagrange multiplier method) with the LMM, avoiding the
need of choose between different active set strategies.

Vicente Mataix Ferrándiz Page 331 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

D.7.2.2.3.1 ε = 0.5 :

Our first case with ε = 0.5 show a very similar convergence path to the one presented on D.7.2.2.2.1.Active set 1,
Table D.11, but the functional from Figure D.16 is smoother than the one from Figure D.13.

Iteration x λ Error
1 − 31 0 0.296296
2 −0.555556 0 0.0877915
3 −0.703704 0 0.0260123
4 −0.802469 0 0.00770735
5 −0.868313 0 0.00228366
6 −0.912209 0 0.0122272
7 −0.9 −0.000958925 0.000041075
8 −0.9 −0.001 4.33681−19
Table D.11: Convergence on Lλ̄ with ALM for (D.22) with ε = 0.5

Figure D.16: Solution for a NL SDOF solved with ALM. ε = 0.5. Moved wall

D.7.2.2.3.2 ε = 1.0 :

On this second case with ε = 1.0 the Table D.12 presents practically the same path as Table D.11, with the
exception that once the solution enters in the inactive region the error changes slightly. The Figure D.17 is even
smoother than Figure D.16.

Iteration x λ Error
1 − 31 0 0.296296
2 −0.555556 0 0.0877915
3 −0.703704 0 0.0260123
4 −0.802469 0 0.00770735
5 −0.868313 0 0.00228366
6 −0.912209 0 0.0167938
7 −0.9 −0.000958925 0.000041075
8 −0.9 −0.001 1.0842−18
Table D.12: Convergence on Lλ̄ with ALM for (D.22) with ε = 1.0

Page 332 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

Figure D.17: Solution for a NL SDOF solved with ALM. ε = 1.0. Moved wall

D.7.2.2.3.3 ε = 5.0 :

Iteration x λ Error
1 − 31 0 0.296296
2 −0.555556 0 0.0877915
3 −0.703704 0 0.0260123
4 −0.802469 0 0.00770735
5 −0.868313 0 0.00228366
6 −0.912209 0 0.061588
7 −0.9 −0.000958925 0.000041075
8 −0.9 −0.001 2.60209−18
Table D.13: Convergence on Lλ̄ with ALM for (D.22) with ε = 5.0

Figure D.18: Solution for a NL SDOF solved with ALM. ε = 5.0. Moved wall

Vicente Mataix Ferrándiz Page 333 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

For the case with ε = 5.0, we have again a very similar NL loop, Table D.13, converging again in 8 NL iterations.
The only value that changes again is the error when crossing the constraint threshold. On the other side, the Figure
D.13 shows a significant aspect on the L representation.

D.7.2.2.3.4 ε = 10.0 :
With ε = 10.0 we have again a very similar NL convergence, Table D.14, but the graphical representation of L
fathom the change of aspect on L already seen in the previous case, see Figure D.19.

Iteration x λ Error
1 − 31 0 0.296296
2 −0.555556 0 0.0877915
3 −0.703704 0 0.0260123
4 −0.802469 0 0.00770735
5 −0.868313 0 0.00228366
6 −0.912209 0 0.122021
7 −0.9 −0.000958925 0.000041075
8 −0.9 −0.001 3.90313−18
Table D.14: Convergence on Lλ̄ with ALM for (D.22) with ε = 10.0

Figure D.19: Solution for a NL SDOF solved with ALM. ε = 10.0. Moved wall

D.7.3 Over-constrained optimisation problem

The next example[Onl], and last one on this section, concerns an optimisation problem over-constrained, that will be
shown as an example of problem that can be solved with the PM, but not with the most advanced methodologies as
the LMM or the ALM. The problem to solve of interest is shown in (D.26).

Minimize f (x) = (x1 − 6)2 + (x2 − 7)2

subject to g1 (x) = −3x1 − 2x2 + 6 ≤ 0
(D.26) g2 (x) = −x1 + x2 − 3 ≤ 0
g3 (x) = x1 + x2 − 7 ≤ 0
2
g4 (x) = x − x2 − 34 ≤ 0
3 1

This function and constrains can be seen visually on Figure D.20, here we can appreciate the constraints, as the
limits on the functional representation.

Page 334 of 374 Vicente Mataix Ferrándiz

APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS D.7 Numerical examples

Figure D.20: Graphic representation of (D.26)

This equation becomes (D.27), where ε is the penalty parameter.

fp (ε, x)) = (x1 − 6)2 + (x2 − 7)2

2
+ ε max {0, −3x1 − 2x2 + 6}

2
+ ε max {0, −x1 + x2 − 3}
(D.27)
2
+ ε max {0, x1 + x2 − 7}
  2
2 4
+ ε max 0, x1 − x2 −
3 3

Considering (D.2) and solving applying a NR iterative resolution we can obtain the correct solution in 4 steps
starting from x = 0, y = 0 with ε = 106 , as can be seen on Table D.15.

Iteration x y fp
√
1 0 0 2 468000384000085
2 0 3 14.4222
3 6 7 1.69706 × 107
4 3 4 1.97134 × 10−9
Table D.15: Convergence on fp with penalty method for D.26
On the other hand, if we try to solve this problem using a LMM, we will find problems to solve due to the

Vicente Mataix Ferrándiz Page 335 of 374

D.7 Numerical examples APPENDIX D. CONSTRAINED OPTIMISATION PROBLEMS

overconstraint of the problem. If we formulate (D.26) considering for example the second active set strategy from
(D.5e), with x = 6, y = 7, λi = 0 for i = 1, 2, 3, 4, the resulting LHS will be (D.28). This is a matrix is rank 2, being a
6 × 6 matrix, which means that the system of equations is not solvable.

 
−2 0 0 0 0 0

 0 −2 0 0 0 0 

 0 0 0 0 0 0 
(D.28) LHS =  

 0 0 0 0 0 0 

 0 0 0 0 0 0 
0 0 0 0 0 0

Same happens considering the first active set strategy, which is more robust. The underlying problem is the
overconstrained problem, which we have already shown, is not a problem for a PM. This is a small example that
shows that LMM despite being a powerful and generic technique it is not applicable for all cases.

Page 336 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Books
[BookAkh88] The calculus of variations. Naum Ilich Akhiezer. CRC Press. 1988.
[BookAnd01] From Convexity to Nonconvexity. Anders Klarbring (auth.) R. P. Gilbert P. D. Panagiotopoulos
P. M. Pardalos (eds.) Lars-Erik Andersson. Springer US. 1st ed. 2001.
[BookCA01] Introduction to finite element methods. Felippa C.A. 2001.
[BookIK08] Lagrange multiplier approach to variational problems and applications. Kazufumi Ito and
Karl Kunisch. Siam. 2008.
[BookKO88] Contact problems in elasticity: a study of variational inequalities and finite element methods.
Noboru Kikuchi and John Tinsley Oden. siam. 1988.
[BookLei04] Dynamics and Bifurcations of Non-Smooth Mechanical Systems.
Professor Dr. Henk Nijmeijer (auth.) Dr. Remco I. Leine. Springer-Verlag Berlin Heidelberg. 1st ed.
2004.
[BookWri06] Computational Contact Mechanics. Peter Wriggers. Springer. 2nd. 2006.
[BookYas13] Numerical Methods in Contact Mechanics. Vladislav A. Yastrebov. Wiley-ISTE. 1st ed. 2013.
[BookZTF14] The Finite Element Method for Solid and Structural Mechanics. O. C. Zienkiewicz, Robert L. Taylor,
and David Fox. Butterworth-Heinemann. 7th ed. 2014.

Articles
[ArtAC91] “A mixed formulation for frictional contact problems prone to Newton like solution methods”.
Pierre Alart and Alain Curnier. In: Computer methods in applied mechanics and engineering. No. 3,
Vol. 92, 1991, pp. 353–375. Elsevier. DOI: 10.1016/0045-7825(91)90022-X.
[ArtAHU58] “Studies in linear and non-linear programming”. Kenneth J Arrow, Leonid Hurwicz, and
Hirofumi Uzawa. In: . 1958, Cambridge Univ. Press.
[ArtAla88] “Multiplicateurs “augmentés” et méthode de Newton généralisée pour contact avec frottement”.
P Alart. In: Report, Document LMA-DME-EPFL, Lausanne. 1988,
[ArtAla97] “Méthode de Newton généralisée en mécanique du contact”. Pierre Alart. In: Journal de
mathématiques pures et appliquées. No. 1, Vol. 76, 1997, pp. 83–108. Elsevier.
[ArtAS58] “Gradient methods for constrained maxima, with weakened assumptions”. KENNETH J Arrow and
ROBERT M Solow. In: Studies in linear and nonlinear programming. 1958, pp. 166–176. Stanford
University Press, Stanford, CA.
[ArtBMP] “The adapted augmented Lagrangian method: a new method for the resolution of the mechanical
frictional contact problem”. Philippe Bussetta, Daniel Marceau, and Jean-Philippe Ponthot. In:
Computational Mechanics. No. 2, , pp. 259–275. DOI: 10.1007/s00466-011-0644-z.
[ArtHes69] “Multiplier and gradient methods”. Magnus R Hestenes. In: Journal of optimization theory and
applications. No. 5, Vol. 4, 1969, pp. 303–320. Springer. DOI: 10.1007/BF00927673.

Vicente Mataix Ferrándiz Page 337 of 374

PH.D.’S THESIS BIBLIOGRAPHY

[ArtNit71] “Über ein Variationsprinzip zur Lösung von Dirichlet-Problemen bei Verwendung von Teilräumen,
die keinen Randbedingungen unterworfen sind”. Joachim Nitsche. In: . No. 1, Vol. 36, 1971,
pp. 9–15.
[ArtPow69] “A method for nonlinear constraints in minimization problems”. Michael JD Powell. In: Optimization.
1969, pp. 283–298. Academic Press.
[ArtQS99] “A survey of some nonsmooth equations and smoothing Newton methods”. Liqun Qi and
Defeng Sun. In: . 1999, pp. 121–146. Springer. DOI: 10.1007/978-1-4613-3285-5_7.
[ArtRoc73a] “The multiplier method of Hestenes and Powell applied to convex programming”.
R Tyrell Rockafellar. In: Journal of Optimization Theory and applications. No. 6, Vol. 12, 1973,
pp. 555–562. Springer. DOI: 10.1007/BF00934777.
[ArtRoc73b] “A dual approach to solving nonlinear programming problems by unconstrained optimization”.
R Tyrrell Rockafellar. In: Mathematical programming. No. 1, Vol. 5, 1973, pp. 354–373. Springer.
DOI : 10.1007/BF01580138.

[ArtSWT85a] “A perturbed Lagrangian formulation for the finite element solution of contact problems”.
Juan C Simo, Peter Wriggers, and Robert L Taylor. In: Computer methods in applied mechanics
and engineering. No. 2, Vol. 50, 1985, pp. 163–180. Elsevier. DOI:
10.1016/0045-7825(85)90088-X.
[ArtSWT85b] “A perturbed Lagrangian formulation for the finite element solution of contact problems”.
Juan C. Simo, Peter Wriggers, and Robert L. Taylor. In: Computer Methods in Applied Mechanics
and Engineering. No. 2, Vol. 50, 1985, pp. 163–180. DOI: 10.1016/0045-7825(85)90088-X.

Ph.D.’s thesis
[PhDBus] “Modélisation et résolution du problème de contact mécanique et son application dans un contexte
multiphysique”. Philippe Bussetta.
[PhDHet14] “The Bipenalty Method for Explicit Structural Dynamics”. Jack Hetherington. 2014.
[PhDYas11] “Computational contact mechanics: geometry, detection and numerical techniques”.
Vladislav A. Yastrebov. 2011.

Online resources
[Onl] Algorithms for Constrained Optimization. URL: https://www.me.utexas.edu/~jensen/
ORMM/supplements/units/nlp_methods/const_opt.pdf.
[Onl] Dualisation of the boundary conditions. URL :
https://www.code-aster.org/V2/doc/v11/en/man_r/r3/r3.03.01.pdf.
[Onl] Wikipedia. Uzawa iteration. URL : https://en.wikipedia.org/wiki/Uzawa_iteration.
[OnlLag] Wikipedia. Lagrange multiplier. URL :
http://en.wikipedia.org/wiki/Lagrange_multiplier.

Page 338 of 374 Vicente Mataix Ferrándiz

APPENDIX E. MORTAR MAPPER

Appendix E

Mortar mapper

Projections are just bullshit.

They’re just guesses.

Jason Fried
(Software entrepreneur)

E.1 Introduction
This appendix introduces the mapper developed using the Mortar formulation[ArtDB06; ArtND08]. The goal of a
map is the distribution of the variable of interest from the origin mesh to the destination mesh, see Figure E.1a. This
method provides an exact decomposition of the target mesh (Figure E.1b). With this, the procedure of mapping
has the advantage of conserving, in a weak sense at least, the important physical quantities such as mass, energy,
etc. Additionally this method satisfies the equilibrium in a weak manner, being then consistent with the FEM
formulation.
The mapping is used on this work in order to compute a consistent gap between meshes during the search
process, see Algorithm 8, in order to discard poor contact candidates. Poor candidates can be defined as the contact
candidates that will not activate during the whole NL iteration process. The inclusion of these candidates may be
difficult the convergence of the active-set strategy, then increasing the number of NL iterations computed during the
NR strategy.

Algorithm 8 Consistent gap computation

1: procedure C ONSISTENT GAP COMPUTATION
2: Reset auxiliar values for nodal coordinates on origin mesh xaux
3: for all node ∈ OriginMeshnodes do
4: xaux = xnode
5: for all node ∈ DestinationMeshnodes do
6: xaux = 0
7: Map xaux from OriginMeshnodes → DestinationMeshnodes
8: for all node ∈ DestinationMeshnodes do
9: From node get the normal (n)
10: gconsistent = −n · (xnode − xaux )

The reason why the very same Mortar formulation has been taken into consideration are, first the developments
are shared between the two frameworks; second the consistency between the results obtained between the system of
equations and the search process; but primarily the quality of the results. On this last point, the results that can be

Vicente Mataix Ferrándiz Page 339 of 374

E.2 Theory APPENDIX E. MORTAR MAPPER

obtained with this method are between the best options available[ArtBBP]. Precisely on Bussetta[ArtBBP] the three
types of data transfer procedures are discussed:

• Element Transfer Method (ELM): The simplest one consists on map the value depending only on the location of
the destination mesh on the origin mesh. Methods on this category we can find the nearest element[MasBuc17]
or nearest neighbour [MasBuc17] mappers.

• Procedures based on the weak conservation of the data field. The presented method (sometimes called ELM)
would we encompassed on this category. Other techniques defined on this category can be Finite Volume
Transfer Method (FVTM)[ArtBBP], which consists on decide the meshes on Finite Volumes (FV) elements
and the transfer is computed between FV meshes.

• These are the procedures based on reconstruction of the field in the neighbourhood of this point. On this
category we can find the modern technique Moving Least Square (MLS)[ArtCho+05; ArtLIC07]. Additionally
on [ArtYu+08; ArtHC90; ArtDHC92] we can find more examples of this kind of methodology. This concept is
considered on the method presented on E.2.3.Discontinuous meshes mapping, where the nodes close to the
discontinuity are taken into consideration to reconstruct the field.

(a) Concept of mapper between two meshes (b) Mortar integration mapping between two meshes
Figure E.1: Mapping concepts
The Mortar formulation employed, based on the concept of dual[ArtWoh02; ArtWoh11] LM, which allows to
evaluate in a simplified manner the data transference between meshes. This method has been already presented on
the corresponding contact sections (4.2.3.4.DLMM). Instead of solving a whole assembled system of the transfer
operators, the dual LM allows to assemble it in a diagonal manner, which can be solved in a trivial manner. There are
alternatives methodologies for using the Mortar formulation without solving directly the system of equations, such the
Jacobian-Free Newton Krylov (JFNK) approach by Hansen[ArtHan11].

E.2 Theory
E.2.1 General mapping theory
Based on the concepts introduced on the work of Ute[MasIsr06] and Jaiman[ArtJai+06], we can start defining the
consistent transfer operation between meshes (E.1).

(E.1a) f = Mc u

Where the matrices and vectors can be defined as:

( R
Mc = Ndi Ndi dΓd
(E.1b) R i
Γd
j j
f= N N u dΓ d
Γd d o o

Page 340 of 374 Vicente Mataix Ferrándiz

APPENDIX E. MORTAR MAPPER E.2 Theory

And u = udi the nodal values of the destination mesh. The index i represents the i th NL iteration. We can define the
RHS and RHS of our respective system of equations.

RHS = f − Mc ui
(E.1c)
LHS = − ∂ RHS
∂ u = Mc

With the RHS and the LHS we can obtain the update of the update of u, ∆u.

(E.1d) ui+1 = ui + ∆u

Where:

(E.1e) ∆u = LHS−1 RHS = M− 1 −1

c (f − Mc ui ) = Mc f − ui

This methodology implies the whole system integration, and resolution of the whole system of equations. On
E.2.2.Dual Lagrange multiplier mapping we will introduce how with the consideration of dual Lagrange multiplier the
Mc becomes diagonal, MD , with the respective advantages.

E.2.2 Dual Lagrange multiplier mapping

E.2.2.1 Introduction
As previously stated, using the dual Lagrange multiplier allows to solve the exact system of equations with the
inversion of a diagonal matrix, trivial operation that can be parallelised. The theory introducing the concept of dual
Lagrange multiplier was first introduced by Wohlmuth[ArtWoh02], who used as an alternative method for DDM. These
nonstandard Lagrange multiplier spaces provided optimal discretisation schemes and a locally supported basis for
the associated constrained mortar spaces.

E.2.2.2 Theory

Algorithm 9 Explicit contribution of the pairs during dual λ mortar mapping

1: procedure E XPLICIT CONTRIBUTION OF THE PAIRS
2: while (rhstotal > toleranceabs and ratio > tolerancerel ) and i < iterationmax do
3: Reset global residual norm rhstotal
4: Reset auxiliar values for nodal residual rhsnode
5: Reset weight areas wnode
6: for all elem ∈ DestinationMeshelements do
7: if Areamortar segmentation > tolerance then
8: Calculate D and M
9: Calculate local residual (E.2a) and the local weight area (E.2b)
10: for all node ∈ DestinationMeshnodes do
11: Compute ud = ud + rhs node
wnode
2
12: Compute rhstotal = rhstotal + rhsnode
rhstotal
13: Compute the residual norm: rhstotal = ndof
14: if i = 0 then
15: rhstotal0 = rhstotal
rhstotal
16: Compute ratio = rhs total0
17: i =i +1

We address the theory of the dual Lagrange multiplier previously presented, 4.3.3.4.1.Dual Lagrange multipliers,
which means that we can focus on the required operations starting from the theory already presented in E.2.1.General
mapping theory, and modify some specific operations with our alternative λ.
On Algorithm 9 we present the solution designed to obtain the explicit contributions of the pairs, for them compute
the mapping. From Algorithm 9 we need to define some missing concepts (E.2). Being known D and M in every pair
of entities, where D is assumed to be diagonal as previously stated, then the explicit contribution of errors and weight

Vicente Mataix Ferrándiz Page 341 of 374

E.2 Theory APPENDIX E. MORTAR MAPPER

areas are (E.2a) and (E.2b) respectively. Where the sub-index node refers to the pair entities which share certain
nodes.

N
Xnode

(E.2a) rhsnode = Melem uelem

o − Delem uelem
d
elem=1

N
Xnode

(E.2b) wnode = Delem (inode , inode )

elem=1

E.2.3 Discontinuous meshes mapping

E.2.3.1 Introduction
This is an innovative contribution of this work, which allows to transfer data between two discontinuos meshes. We
will call a discontinuos mesh to the mesh that is not watertight, which means that the nodes are not necessarily
shared between the entities which conform the mesh. This means that following the metaphor of a container, in case
of being filled with water, the water would leak across the gaps between the nodes of the mesh, reason why is called
watertight.
This development can be used on embedded CFD simulations, where the condition of watertight mesh is not
guaranteed, especially from meshes imported from STL files, which usually undergo these problematic.

E.2.3.2 Theory

Algorithm 10 Explicit contribution of the pairs during dual λ mortar discontinuous mapping
1: procedure E XPLICIT CONTRIBUTION OF THE PAIRS ON DISCONTINUOUS MAPPING
2: Create an inverse database (pairing destination-origin entities and vice-versa)
3: while (rhstotal > toleranceabs and ratio > tolerancerel ) and i < iterationmax do
4: Reset global residual norm rhstotal
5: Reset auxiliar values for nodal residual rhsnode
6: Reset weight areas wnode
7: for all elem ∈ DestinationMeshelements do
8: if Areamortar segmentation > tolerance then
9: Calculate D and M
10: Calculate local residual (E.2a) and the local weight area (E.2b)
11: for all node ∈ elem do
12: Local wlocal = D(node, node) contribution
13: Local h equivalent to length of elem
14: for all pairinv ∈ eleminverse paired do
15: for all nodeaux ∈ pairinv do
16: Compute distance d between node and nodeaux
17: Compute local weight area discontinuous contribution (E.3)
18: for all node ∈ DestinationMeshnodes do
rhsnode
19: Compute ud = ud + w discontinuous
node
2
20: Compute rhstotal = rhstotal + rhsnode
rhstotal
21: Compute the residual norm: rhstotal = ndof
22: if i = 0 then
23: rhstotal0 = rhstotal
rhstotal
24: Compute ratio = rhs total0
25: i =i +1

This method preserves the properties presented on E.2.2.Dual Lagrange multiplier mapping, which means
that the corresponding system of equations keeps the possibility to be solved in a simplified manner. The main

Page 342 of 374 Vicente Mataix Ferrándiz

APPENDIX E. MORTAR MAPPER E.3 Numerical examples

difference between the previous approach and the current one is the need to create an inverse mapping of the
interface. This means to create a database which inverts the roles origin-destination meshes. Additionally a coefficient
of contributions is considered to take into account the contribution of close nodes, this contribution is computed
penalising the distance between the destination node and the one evaluated on the discontinuous interface. Algorithm
10 presents the modifications needed on the Algorithm 9 in order to compute the mapping of values for a discontinuous
interface.
On (E.3) the additional contribution of the discontinuous interface is specified. Being k ≈ 1.0e − 4, being k called
discontinuous interface factor. This value has been estimated doing several evaluations in different cases with different
levels of discontinuity on the interface, being this value the best compromise for the different cases studied.

discontinuous wlocal
(E.3) wnode = wnode + d 2
(1 + kh
)

E.3 Numerical examples

In all the cases presented below we map between two meshes with an irregular double-curvature, which makes the
problem difficult to map for simpler methods as the nearest neighbour mapping. The functions (scalar and vector) we
want to map between these two meshes is presented on (E.4). On the scalar case everything depends only on the z
coordinate, meanwhile the vector case is equivalent to the direct imposition of the coordinates of the nodes.
The scalar function:

(E.4a) f (x, y, z) = z

The vector function:

 
(E.4b) f(x, y, z) = x y z

E.3.1 Non-matching meshes of triangles

Figure E.2: Non-matching mesh of triangles

The first case from Figure E.2 present us the previously mentioned example with irregular double-curvature
meshes. On this case, both meshes are conformed by triangular elements. The results obtained for a scalar value
is shown in Figure E.9, meanwhile Figure E.10 shows the results for a vector value. Both solutions provide a very
good approach of the values, this can be appreciated with particular detail on the wireframe results (Figures E.6a and
E.7a), where it is easier to appreciate the resulting values on both meshes.

Vicente Mataix Ferrándiz Page 343 of 374

E.3 Numerical examples APPENDIX E. MORTAR MAPPER

(a) Scalar solution as a wireframe (b) Scalar solution as a continuous field

Figure E.3: Scalar solution for non-matching mesh of triangles

(a) Vector solution as a wireframe (b) Vector solution as a continuous field

Figure E.4: Vector solution for non-matching mesh of triangles

E.3.2 Non-matching meshes of triangles and quadrilaterals

The case from Figure E.5 instead of being fully meshed on triangles as Figure E.2, or even fully meshed with
quadrilaterals, presents both types of geometries. The origin mesh consists on triangular entities, meanwhile the
destination mesh is integrated by quadrilateral entities. This case shows how it is possible not only map values across
non-conformant meshes, but also between different types of geometries, with the additional complexity.

The results obtained on scalar and vector fields (Figures E.9 and E.10) show again a good matching between the
origin and destination values.In the same way as the previous example, the wireframe (Figures E.6a and E.7a) and
continuous field solutions (Figures E.6b and E.7b) are presented.

E.3.3 Discontinuous meshes

Figure E.8: Detail of the non-matching mesh discontinuous of triangles

Page 344 of 374 Vicente Mataix Ferrándiz

APPENDIX E. MORTAR MAPPER E.3 Numerical examples

Figure E.5: Non-matching mesh of triangles and quadrilaterals

(a) Scalar solution as a wireframe (b) Scalar solution as a continuous field

Figure E.6: Scalar solution for non-matching mesh of triangles and quadrilaterals

The following case may a priori look similar to Figure E.2, that’s why Figure E.8 zooms into the detail of the mesh, so
the discontinuity of the meshes is appreciated. Due to this the standard mapper cannot be applied, and the method
presented on E.2.3.Discontinuous meshes mapping must be taken into consideration.

The results obtained are again being of good quality despite of the existence of this discontinuity on the mesh,
both for scalar and vector fields (Figures E.9 and E.10 respectively).

(a) Scalar solution as a wireframe (b) Scalar solution as a continuous field

Figure E.9: Scalar solution for non-matching discontinuous mesh of triangles

Vicente Mataix Ferrándiz Page 345 of 374

E.3 Numerical examples APPENDIX E. MORTAR MAPPER

(a) Vector solution as a wireframe (b) Vector solution as a continuous field

Figure E.7: Vector solution for non-matching mesh of triangles and quadrilaterals

(a) Vector solution as a wireframe (b) Vector solution as a continuous field

Figure E.10: Vector solution for non-matching discontinuous mesh of triangles

Page 346 of 374 Vicente Mataix Ferrándiz

BIBLIOGRAPHY BIBLIOGRAPHY

Bibliography

Articles
[ArtBBP] “Efficient 3D data transfer operators based on numerical integration”. Philippe Bussetta,
Romain Boman, and Jean-Philippe Ponthot. In: International Journal for Numerical Methods in
Engineering. No. 3-4, , pp. 892–929. Wiley Online Library. DOI: 10.1002/nme.4821.
[ArtCho+05] “An improved interface element with variable nodes for non-matching finite element meshes”.
Young-Sam Cho, Sukky Jun, Seyoung Im, and Hyun-Gyu Kim. In: Computer Methods in Applied
Mechanics and Engineering. No. 27-29, Vol. 194, 2005, pp. 3022–3046. Elsevier. DOI:
10.1016/j.cma.2004.08.002.
[ArtDB06] “Information transfer between incompatible finite element meshes: application to coupled
thermo-viscoelasticity”. David Dureisseix and Henri Bavestrello. In: Computer Methods in Applied
Mechanics and Engineering. No. 44-47, Vol. 195, 2006, pp. 6523–6541. Elsevier. DOI:
10.1016/j.cma.2006.02.003.
[ArtDHC92] “Automatic adaptive remeshing for numerical simulations of metalforming”. M Dyduch,
AM Habraken, and Serge Cescotto. In: Computer Methods in Applied Mechanics and Engineering.
No. 1-3, Vol. 101, 1992, pp. 283–298. Elsevier. DOI: 10.1016/0045-7825(92)90026-G.
[ArtHan11] “A Jacobian-free Newton Krylov method for mortar-discretized thermomechanical contact
problems”. Glen Hansen. In: Journal of Computational Physics. No. 17, Vol. 230, 2011,
pp. 6546–6562. Elsevier. DOI: 10.1016/j.jcp.2011.04.038.
[ArtHC90] “An automatic remeshing technique for finite element simulation of forming processes”.
AM Habraken and Serge Cescotto. In: International Journal for Numerical Methods in Engineering.
No. 8, Vol. 30, 1990, pp. 1503–1525. Wiley Online Library. DOI: 10.1002/nme.1620300811.
[ArtJai+06] “Conservative load transfer along curved fluid–solid interface with non-matching meshes”.
Rajeev K Jaiman, Xiangmin Jiao, Philippe H Geubelle, and Eric Loth. In: Journal of Computational
Physics. No. 1, Vol. 218, 2006, pp. 372–397. Elsevier. DOI: 10.1016/j.jcp.2006.02.016.
[ArtLIC07] “Variable-node elements for non-matching meshes by means of MLS (moving least-square)
scheme”. Jae Hyuk Lim, Seyoung Im, and Young-Sam Cho. In: International Journal for Numerical
Methods in Engineering. No. 7, Vol. 72, 2007, pp. 835–857. Wiley Online Library. DOI:
10.1002/nme.1988.
[ArtND08] “A computational strategy for thermo-poroelastic structures with a time–space interface coupling”.
David Néron and David Dureisseix. In: International Journal for Numerical Methods in Engineering.
No. 9, Vol. 75, 2008, pp. 1053–1084. Wiley Online Library. DOI: 10.1002/nme.2283.
[ArtWoh02] “A Comparison of Dual Lagrange Multiplier Spaces for Mortar Finite Element Discretizations”.
Barbara I. Wohlmuth. In: ESAIM: Mathematical Modelling and Numerical Analysis. No. 6, Vol. 36,
2002, pp. 995–1012. EDP Sciences. DOI: 10.1051/m2an:2003002.
[ArtWoh11] “Variationally consistent discretization schemes and numerical algorithms for contact problems”.
Barbara Wohlmuth. In: Acta Numerica. Vol. 20, 2011, pp. 569–734. DOI:
10.1017/S0962492911000079.

Vicente Mataix Ferrándiz Page 347 of 374

MASTER’S THESIS. BIBLIOGRAPHY

[ArtYu+08] “Three-dimensional RITSS large displacement finite element method for penetration of foundations
into soil”. L Yu, J Liu, XJ Kong, and Y Hu. In: Computers and Geotechnics. No. 3, Vol. 35, 2008,
pp. 372–382. Elsevier. DOI: 10.1016/j.compgeo.2007.08.007.

Master’s thesis.
[MasBuc17] Development and Implementation of a Parallel Framework for Non-Matching Grid Mapping.
Philipp Bucher. 2017.
[MasIsr06] Implementation of an Algorithm for Data Transfer on the Fluid-Structure Interface between
Non-Matching Meshes in Kratos and an Algorithm for the Generation. Ute Israel. 2006.

Page 348 of 374 Vicente Mataix Ferrándiz

LIST OF TABLES LIST OF TABLES

List of Tables

2.1 Transformations of stresses. Source[BookBel+14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.1 Isoparametric coordinates of the six nodes in the patch of Figure 3.1b . . . . . . . . . . . . . . . . . . 54
3.2 Derivatives of the shape functions from (3.14) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Isotropic-kinematic hardening law constans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.1 Parameters considered for ALM parameters calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4.2 Results of numerical experiment for ALM parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.3 Analogies between friction and plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.4 Parameters considered for simple patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.5 Parameters considered for Taylor [ArtTP] patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.6 Parameters considered for simple patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.7 Parameters considered for 2D sphere-plane Hertz benchmark . . . . . . . . . . . . . . . . . . . . . . 139
4.8 Parameters considered for 2D cylinder-cylinder Hertz benchmark . . . . . . . . . . . . . . . . . . . . 141
4.9 Parameters considered for 3D Hertz plane-sphere problem . . . . . . . . . . . . . . . . . . . . . . . 143
4.10 Parameters considered for 3D Hertz sphere-sphere problem . . . . . . . . . . . . . . . . . . . . . . . 144
4.11 Parameters considered for teeth model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.12 Parameters considered for energy conservation test . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.13 Parameters considered for double arc benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.14 Parameters considered for arc pressing blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.15 Parameters considered for hyperelastic tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.16 Parameters considered for contacting cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.17 Parameters considered for press fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.18 Parameters considered for circular ironing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.19 Parameters considered for shallow ironing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.20 2D lines pairs geometrical configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.21 2D lines pertubartion amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.22 3D triangles pairs geometrical configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.23 3D triangles pertubartion amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.24 3D quadrilaterals pairs geometrical configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.25 3D quadrilaterals pertubartion amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

5.1 Parameters considered for the cube minimal example . . . . . . . . . . . . . . . . . . . . . . . . . . 203

5.2 Parameters considered for the tensile test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5.3 Parameters considered for gear example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

6.1 Parameters considered for cantilever example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

6.2 Parameters considered for the Hessian remesh punch test . . . . . . . . . . . . . . . . . . . . . . . . 235
6.3 Parameters considered for the Hessian remesh Hertz example . . . . . . . . . . . . . . . . . . . . . 243
6.4 Parameters considered for the Hessian remesh contacting cylinders example . . . . . . . . . . . . . . 246

A.1 Parameters considered for the mesh tying example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

D.1 Comparison of different methods presented . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322

Vicente Mataix Ferrándiz Page 349 of 374

LIST OF TABLES LIST OF TABLES

D.2 Convergence on fp with PM for (D.13) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324

D.3 Convergence on Lλ with LMM for (D.13) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
D.4 First active set strategy convergence on Lλ with LMM for (D.16) . . . . . . . . . . . . . . . . . . . . . 326
D.5 Second active set strategy convergence on Lλ with LMM for (D.16) . . . . . . . . . . . . . . . . . . . 326
D.6 Third active set strategy convergence on Lλ with LMM for (D.16) . . . . . . . . . . . . . . . . . . . . 327
D.7 Convergence with PM for (D.21) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
D.8 First active set strategy convergence on Lλ with LMM for (D.16) with moved wall . . . . . . . . . . . . 328
D.9 Second active set strategy convergence on Lλ with LMM for (D.16) with moved wall . . . . . . . . . . 330
D.10 Third active set strategy convergence on Lλ with LMM for (D.16) with moved wall . . . . . . . . . . . 330
D.11 Convergence on Lλ̄ with ALM for (D.22) with ε = 0.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
D.12 Convergence on Lλ̄ with ALM for (D.22) with ε = 1.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
D.13 Convergence on Lλ̄ with ALM for (D.22) with ε = 5.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
D.14 Convergence on Lλ̄ with ALM for (D.22) with ε = 10.0 . . . . . . . . . . . . . . . . . . . . . . . . . . 334
D.15 Convergence on fp with penalty method for D.26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

Page 350 of 374 Vicente Mataix Ferrándiz

LIST OF FIGURES LIST OF FIGURES

List of Figures

1.1 Examples of numerical simulation of forming. Source[OnlShe] . . . . . . . . . . . . . . . . . . . . . . 23

1.2 Parameters influencing sheet metal formability. Inspired in[BookBan10] . . . . . . . . . . . . . . . . . 24
1.3 Structure of an expert system for the analysis of sheet metal formability. Inspired in [BookBan10] . . . 25
1.4 Springback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.5 Wrinkling phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.1 Minimal example of FEA for a Laplacian problem. Resolution using Mathematica[OnlWol] . . . . . . . 35
2.2 Deformation of a continuum body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Polar decomposition of F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4 Graphic representation of the arc-length method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.5 Graphic representation of Rayleigh damping matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.1 Patch of elements for BST and EBST . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.2 Patch performed in the element considering the neighbour elements . . . . . . . . . . . . . . . . . . 58
3.3 Nodes considered in the computation of the transverse shear strains . . . . . . . . . . . . . . . . . . 60
3.4 Patch test geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5 Solution for the displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.6 Solution for the stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.7 Solution for the displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.8 Solution for the stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.9 Cantilever with point load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.10 Cantilever geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.11 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.12 Cantilever subjected to end bending moment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.13 Solution for a cantilever subjected to an end bending moment . . . . . . . . . . . . . . . . . . . . . . 67
3.14 Frequencies behaviour test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.15 Cook ’s membrane geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.16 Cook ’s membrane convergence solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.17 Cook ’s membrane deformed shape solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.18 Geometry of the scoordelis cylindrical roof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.19 Solution for the scoordelis cylindrical roof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.20 Geometry of a semi-spherical shell with a hole, both original and deformed geometry . . . . . . . . . 70
3.21 Geometry of a semi-spherical shell with a hole solution . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.22 The open-end cylindrical shell subjected to radial pulling forces. . . . . . . . . . . . . . . . . . . . . . 72
3.23 Solution of the open-ended cylindrical shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.24 The semi-cylindrical shell subjected to an end pinching force . . . . . . . . . . . . . . . . . . . . . . . 72
3.25 Solution of the semi-cylindrical shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.26 Slit annular plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.27 Slit annular plate solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.28 Geometry of the hinged cylindrical panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.29 Solution of the cylindrical panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.30 Geometry of the cone shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.31 Cone-shell test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

Vicente Mataix Ferrándiz Page 351 of 374

LIST OF FIGURES LIST OF FIGURES

3.32 Square thin film under in-plane shear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

3.33 Solution of the wrinkling simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.34 Solution of the FSI-Vein test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.1 Representation of moving blocks during ancient Egypt. Source[BookPop10] . . . . . . . . . . . . . . 86

4.2 Original Hertz paper in 1882. Source[ArtHer82] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.3 Basic definition of the contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.4 Contact states. Based on[PhDYas11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.5 NTN. Inspired[PhDYas11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.6 NTS. Inspired[PhDYas11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.7 CDM. Inspired[PhDYas11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.8 STS. Inspired[PhDYas11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.9 3D segmentation between two triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.10 Friction cone for variants of the Coulomb law. Inspired [ArtRao16] . . . . . . . . . . . . . . . . . . . . 96
4.11 Definition of the contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.12 KKT conditions of non-penetration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.13 Condition number study for the ALM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.14 Condition number study graphic representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.15 Dual shape functions for the 2D linear line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.16 Dual shape functions for the 3D linear triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.17 Dual shape functions for the 3D bilinear quadrilateral . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.18 Example of convergence check in the frictionless contact . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.19 Nodal NCP function, or Lagrangian (L) contribution for the LM in the ALM . . . . . . . . . . . . . . . 112
4.20 Coulomb’s schematic depiction of frictional contact conditions in tangential direction . . . . . . . . . . 113
4.21 Graphical representation of Coulombs frictional conditions for 3D contact problem. Inspired [PhDYas11]114
4.22 Augmented Lagrangian function for the frictional contact problem. Corresponding to (4.61b) . . . . . . 118
4.23 Example of convergence check in the frictional contact . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.24 NCP for the frictional contact problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.25 Conventional penetration estimations. Symmetric: a segment intersection; b volume intersection.
Asymmetric: c,d segment in volume; e,f node in volume; g,h node under surface. Source[PhDYas11] 124
4.26 Types of bounding volumes. a) The radius search (BS). b) AABB. c) OBB. d) k-DOP (convex hull).
Inspired on [ArtYL08] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.27 BS approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.28 OBB. Source[ArtRom11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.29 OBB from surfaces build comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.30 Separating axis theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.31 SAT concept. Source[OnlCho] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.32 Space subdivision approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.33 Tree structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.34 Consistent gap example. Circular surface vs plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.35 Consistent gap example. Double curvature surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.36 Self-contact normals descrimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.37 Self-contact detection simplest case, two parallel planes . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.38 Self-contact detection for simple tubular case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.39 Self-contact detection for tubular case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.40 S-shape profile mesh (tetrahedra) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.41 S-shape profile resulting self-contact simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.42 Geometries of the simplest patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.43 Solution for the frictionless simplest patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.44 Geometry and solution for the frictional simplest patch test . . . . . . . . . . . . . . . . . . . . . . . . 136
4.45 Solution for the Taylor patch test in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.46 Solution for the Taylor patch test in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.47 Friction problem from Dong[ArtDon99] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.48 Solution for pure friction problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.49 Setup for the 2D sphere-plane Hertz benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

Page 352 of 374 Vicente Mataix Ferrándiz

LIST OF FIGURES LIST OF FIGURES

4.50 Solution compared for different mesh sizes for the 2D Hertz plane-sphere contact . . . . . . . . . . . 140
4.51 Error compared for different mesh sizes for the 2D Hertz plane-sphere contact . . . . . . . . . . . . . 140
4.52 Setup for the two cylinders Hertz benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.53 Solution for the two cylinders frictionless Hertz benchmark . . . . . . . . . . . . . . . . . . . . . . . . 141
4.54 Solution for the two cylinders frictional Hertz benchmark . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.55 Mesh considered for 3D Hertz plane-sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.56 Solution for the 3D Hertz plane-sphere contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.57 Solution compared for different mesh sizes for the 3D Hertz plane-sphere contact . . . . . . . . . . . 143
4.58 Error compared for different mesh sizes for the 3D Hertz plane-sphere contact . . . . . . . . . . . . . 143
4.59 Reference solution for two spheres Hertz contact. Source[ArtZhu12] . . . . . . . . . . . . . . . . . . 144
4.60 Solution for two hemispheres Hertz contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.61 Teeth layers model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.62 Solution for teeth layers model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.63 Energy conservation test. Cylinder inside ring. The example can be found in here . . . . . . . . . . . 146
4.64 Evolution of displacement and energy compared with analytical solution . . . . . . . . . . . . . . . . 147
4.65 Solution of the problem at certains times. The time is specified in each subfigure . . . . . . . . . . . . 147
4.66 Double arc benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.67 Double arc meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.68 Displacement solution for frictionless in double arc benchmark . . . . . . . . . . . . . . . . . . . . . . 149
4.69 Compared solution for frictionless in double arc benchmark . . . . . . . . . . . . . . . . . . . . . . . 149
4.70 Displacement solution for frictional in double arc benchmark . . . . . . . . . . . . . . . . . . . . . . . 150
4.71 Compared solution for frictional in double arc benchmark . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.72 Arc pressing block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.73 Solution for different stiffness between the arc and the block . . . . . . . . . . . . . . . . . . . . . . . 151
4.74 Mesh for hyperelastic tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.75 Displacement solution for hyperelastic tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.76 Slice solution for hyperelastic tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.77 Contacting cylinders mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.78 Solution for contacting cylinders with horizontal movement . . . . . . . . . . . . . . . . . . . . . . . . 154
4.79 Frictionless reaction solution in the lower cylinder for the horizontal movement . . . . . . . . . . . . . 154
4.80 Solution for contacting cylinders with vertical movement . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.81 Reaction solution in the lower cylinder for the vertical movement . . . . . . . . . . . . . . . . . . . . . 155
4.82 Press fit problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.83 Solution for 2D press fit compared with reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.84 Press fit 2D solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.85 Press fit 3D solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.86 Circular ironing test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.87 Circular ironing test solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.88 Comparison between the frictionless solutions of the shallow and the circular ironing . . . . . . . . . . 159
4.89 Shallow ironing test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.90 Shallow ironing test solution, comparing frictional and frictionless cases . . . . . . . . . . . . . . . . . 160
4.91 Reaction solution for shallow ironing test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.92 Line validation geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.93 Convergence plot for the Jacobian derivatives for the 2D linear line . . . . . . . . . . . . . . . . . . . 163
4.94 Shape functions for the 2D linear line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.95 Integration segment for a linear line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.96 Convergence plot for the shape function derivatives for the 2D linear line . . . . . . . . . . . . . . . . 165
4.97 Convergence plot for the dual shape function derivatives for the 2D linear line . . . . . . . . . . . . . 166
4.98 Normal average for a 2D line. Proposed by Taylor and Papadopoulos[ArtPT92] . . . . . . . . . . . . . 167
4.99 Convergence plot for the normal derivatives for the 2D linear line . . . . . . . . . . . . . . . . . . . . 168
4.100Intersection and clipping procedure during mortar segmentation . . . . . . . . . . . . . . . . . . . . 170
4.101Detail on intersection on mortar segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.102Convergence plot for the Jacobian derivatives for the 3D linear triangle . . . . . . . . . . . . . . . . . 172
4.103Convergence plot for the Jacobian derivatives for the 3D bilinear quadrilateral . . . . . . . . . . . . . 172
4.104Shape functions for the 3D linear triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

Vicente Mataix Ferrándiz Page 353 of 374

LIST OF FIGURES LIST OF FIGURES

4.105Shape functions for the 3D bilinear quadrilateral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

4.106Convergence plot for the shape function derivatives for the 3D linear triangle . . . . . . . . . . . . . . 175
4.107Convergence plot for the shape function derivatives for the 3D bilinear quadrilateral . . . . . . . . . . 176
4.108Convergence plot for the dual shape function derivatives for the 3D linear triangle . . . . . . . . . . . 176
4.109Convergence plot for the dual shape function derivatives for the 3D bilinear quadrilateral . . . . . . . . 177
4.110Normal average for a 3D geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.111Convergence plot for the normal vector derivatives for the 3D linear triangle . . . . . . . . . . . . . . . 178
4.112Convergence plot for the normal vector derivatives for the 3D bilinear quadrilateral . . . . . . . . . . . 179

5.1 Liberty ships suddenly broke in half while moored at the dock . . . . . . . . . . . . . . . . . . . . . . 190
5.2 Types of hardening. Left: Isotropic hardening. Right: Kinematic hardening . . . . . . . . . . . . . . . 192
5.3 Multiplicative decomposition of the deformation gradient F . . . . . . . . . . . . . . . . . . . . . . . . 193
5.4 Elastic-plastic CL class structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.5 Consistent constitutive tensor perturbation method convergence. The code which tests the convergence
rate can be found here . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.6 Simple cube geometry. Only one 8-node hexahedron . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.7 Von Mises stresses compared with the reference solution. The test can be found here . . . . . . . . . 203
5.8 Mesh comparisons. It can be found here . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.9 Resulting plastic dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5.10 Resulting displacement for the mesh of hexahedra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5.11 Gears example mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.12 Detail solution for the displacement, VM stress and plastic dissipation . . . . . . . . . . . . . . . . . . 205
5.13 Overall solution for the displacement and the VM stress . . . . . . . . . . . . . . . . . . . . . . . . . 206

6.1 Adaptive mesh techniques applied on a contact patch test . . . . . . . . . . . . . . . . . . . . . . . . 211

6.2 Concept of h-refinement and p-refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
6.3 Metric analogies. Images from [ArtFA05] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
6.4 Effects of the metric on a tetrahedra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.5 Initial mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.6 Solution for the example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.7 Scalar and its gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.8 Stanford’s Bunny example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.9 Resulting mesh after remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.10 Construction of a recovered gradient in 1D. Inspired[BookAO11] . . . . . . . . . . . . . . . . . . . . . 226
6.11 Optimal sampling points in 1D linear elements. Source[BookZZT13] . . . . . . . . . . . . . . . . . . . 227
6.12 SPR example for linear triangles. Source[BookZZT13] . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.13 SPR example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.14 Transfer operators. Source[ArtGJ06] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.15 The beam problem of reference for the internal interpolation . . . . . . . . . . . . . . . . . . . . . . . 230
6.16 The beam problem of reference for the internal interpolation . . . . . . . . . . . . . . . . . . . . . . . 231
6.17 Cantilever mesh and geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.18 Resulting VM stress before and after recover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.19 Hertz geometry considered. Example here . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
6.20 Solution obtained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
6.21 Simplest patch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
6.22 Punch test geometry considered. Solution t = 0.5s. Example here . . . . . . . . . . . . . . . . . . . . 235
6.23 Punch test solution for VM stress and contact stress Hessian at t = 1.0s . . . . . . . . . . . . . . . . 236
6.24 Punch test solution for strain energy, VM stress and contact stress Hessian at t = 1.0s . . . . . . . . 236
6.26 Punch test solution for strain energy, VM stress and contact stress Hessian at t = 2.0s . . . . . . . . 236
6.25 Punch test solution for VM stress and contact stress Hessian at t = 2.0s . . . . . . . . . . . . . . . . 237
6.27 SPR contact interface example for linear triangles. Inspired[ArtWS98; BookZZT13] . . . . . . . . . . 237
6.28 Error obtained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.29 Mesh evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.30 Error solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.31 Displacement solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

Page 354 of 374 Vicente Mataix Ferrándiz

LIST OF FIGURES LIST OF FIGURES

6.32 Workflow for Level set/Hessian metric remesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

6.33 Workflow for SPR metric remesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.34 Distance function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.35 Mesh before and after remesh for coarse mesh. Example . . . . . . . . . . . . . . . . . . . . . . . . 242
6.36 Coarse mesh distance function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.37 Final configuration of the beam example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.38 Hertz solution step 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.39 Hertz solution step 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.40 Hertz solution step 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.41 Hertz solution step 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.42 Hertz solution step 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.43 Hertz solution step 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.44 Hertz solution step 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.45 Hertz solution step 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.46 Initial mesh in contacting cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.47 Contacting cylinders solution at t = 0.35s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.48 Contacting cylinders perspective solution at t = 0.35s . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.49 Contacting cylinders solution at t = 0.70s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.50 Contacting cylinders perspective solution at t = 0.70s . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.51 Contacting cylinders solution at t = 1.05s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
6.52 Contacting cylinders perspective solution at t = 1.05s . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
6.53 Contacting cylinders solution at t = 1.4s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
6.54 Contacting cylinders perspective solution at t = 1.4s . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
6.55 Mesh before and after remesh cavity case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.56 Embedded cylinder geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.57 Mesh before and after remesh embedded case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
6.58 Base mesh before and after remesh boxes with isosurface . . . . . . . . . . . . . . . . . . . . . . . . 250
6.59 Mesh before and after remesh boxes with isosurface . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
6.60 STL geometry for the Lamborghini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.61 Original mesh for the Lamborghini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.62 Resulting mesh for the Lamborghini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.63 Geometry and initial mesh in 2D channel. The example can be found here . . . . . . . . . . . . . . . 252
6.64 Solution in 2D channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
6.65 Initial mesh in 3D channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.66 Solution in 3D channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253

7.1 Cylinder punch case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

7.2 Mesh considered in the cylinder punch case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
7.3 Solution for cylinder punch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
7.4 Spherical punch case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
7.5 3D solution for the case µ1 = 0.04 and µ2 = 0.04 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
7.6 Solution for spherical punch test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264

A.1 Exact segmentation method for integration. Figure inspired on Popp[PhDPop12] . . . . . . . . . . . . 286
A.2 Collocation method for integration. Figure inspired on Popp[PhDPop12] . . . . . . . . . . . . . . . . . 286
A.3 Taylor patch test[ArtTP] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
A.4 Displacement on Taylor test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
A.5 Compared solution for the augmented contact pressure . . . . . . . . . . . . . . . . . . . . . . . . . 288
A.6 Convergence of the solution for different number of GP . . . . . . . . . . . . . . . . . . . . . . . . . . 288
A.7 Mesh tying example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
A.8 Solution for mesh tying example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

B.1 Kratos Multiphysics logo. Image from Kratos web . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

B.2 Source[ArtDRO10], updated with the recent classes added . . . . . . . . . . . . . . . . . . . . . . . 296
B.3 Model data structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
B.4 Concept of submodelpart for BC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297

Vicente Mataix Ferrándiz Page 355 of 374

LIST OF FIGURES LIST OF FIGURES

B.5 Mmg logo. Image from Mmg web . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

B.6 Mmg idealised geometry. Source[ArtDDF14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
B.7 Mmg-Kratos integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
B.8 Concept of colours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

C.1 AD concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306

C.2 NR representation for a SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
C.3 AD workflow in a UML diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310

D.1 Analogy between penalty method and springs. Image from [BookYas13] . . . . . . . . . . . . . . . . 314
D.2 Representation of F . Image from [ArtBMP] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
D.3 Lagrangian function for the contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
D.4 Augmented Lagrangian function for the contact problem . . . . . . . . . . . . . . . . . . . . . . . . . 320
D.5 Patch tests solutions obtained with Master -Slave elimination method . . . . . . . . . . . . . . . . . . 321
D.6 Simplified contact problem within spring and a wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
D.7 Solution for a SDOF solved with PM, comparing different values of ε . . . . . . . . . . . . . . . . . . 323
D.8 Solution for a SDOF solved with LMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
D.9 Solution for a NL SDOF solved with LMM. Active set 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 325
D.10 Solution for a NL SDOF solved with LMM. Active set 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 326
D.11 Solution for a NL SDOF solved with LMM. Active set 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 327
D.12 Solution for a NL SDOF problem solved with PM, comparing different values of ε. Moved wall . . . . . 328
D.13 Solution for a NL SDOF solved with LMM. Active set 1. Moved wall . . . . . . . . . . . . . . . . . . . 329
D.14 Solution for a NL SDOF solved with LMM. Active set 2. Moved wall . . . . . . . . . . . . . . . . . . . 329
D.15 Solution for a NL SDOF solved with LMM. Active set 3. Moved wall . . . . . . . . . . . . . . . . . . . 330
D.16 Solution for a NL SDOF solved with ALM. ε = 0.5. Moved wall . . . . . . . . . . . . . . . . . . . . . . 332
D.17 Solution for a NL SDOF solved with ALM. ε = 1.0. Moved wall . . . . . . . . . . . . . . . . . . . . . . 333
D.18 Solution for a NL SDOF solved with ALM. ε = 5.0. Moved wall . . . . . . . . . . . . . . . . . . . . . . 333
D.19 Solution for a NL SDOF solved with ALM. ε = 10.0. Moved wall . . . . . . . . . . . . . . . . . . . . . 334
D.20 Graphic representation of (D.26) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

E.1 Mapping concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340

E.2 Non-matching mesh of triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
E.3 Scalar solution for non-matching mesh of triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
E.4 Vector solution for non-matching mesh of triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
E.8 Detail of the non-matching mesh discontinuous of triangles . . . . . . . . . . . . . . . . . . . . . . . 344
E.5 Non-matching mesh of triangles and quadrilaterals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
E.6 Scalar solution for non-matching mesh of triangles and quadrilaterals . . . . . . . . . . . . . . . . . . 345
E.9 Scalar solution for non-matching discontinuous mesh of triangles . . . . . . . . . . . . . . . . . . . . 345
E.7 Vector solution for non-matching mesh of triangles and quadrilaterals . . . . . . . . . . . . . . . . . . 346
E.10 Vector solution for non-matching discontinuous mesh of triangles . . . . . . . . . . . . . . . . . . . . 346

Page 356 of 374 Vicente Mataix Ferrándiz

LIST OF ALGORITHMS LIST OF ALGORITHMS

List of Algorithms

1 Line-search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2 Algorithm for the frictionless contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3 Algorithm for the frictional contact problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4 Self-contact detection algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5 Adaptive refinement process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
6 Compute of nodal average value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
7 Adaptation of normal penalty coefficient[ArtBMP] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
8 Consistent gap computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
9 Explicit contribution of the pairs during dual λ mortar mapping . . . . . . . . . . . . . . . . . . . . . . 341
10 Explicit contribution of the pairs during dual λ mortar discontinuous mapping . . . . . . . . . . . . . . 342

Vicente Mataix Ferrándiz Page 357 of 374

LIST OF ALGORITHMS LIST OF ALGORITHMS

Page 358 of 374 Vicente Mataix Ferrándiz

Acronyms Acronyms

Glossaries

Acronyms
AABB Axis-Aligned Bounding Box.

AALM Adapted Augmented Lagrangian Method.

AD Automatic Differentiation.

ADLM Augmented Dual Lagrange Multiplier.

AGAUR Agència de Gestió d’Ajuts Universitaris i de Recerca.

ALM Augmented Lagrangian Method.

AMG Algebraic MultiGrid.

ANS Assumed Normal Strain.

API Application Programming Interface.

APM Adapted Penalty Method.

BC Boundary Conditions.

BDF Backward Differentiation Formula.

BS Bounding Spheres.

BSD Berkeley Software Distribution.

BST Basic Shell Triangle.

BVH Bounding Volume Hierarchies.

BVP Boundary Value Problem.

CAD Computer-Aided Design.

CAE Computer Aided Engineering.

CAM Computer Aided Manufacturing.

CAS Computer Algebra Systems.

CBR Continuum-Based Resultant.

CCM Computational Contact Mechanics.

Vicente Mataix Ferrándiz Page 359 of 374

Acronyms Acronyms

CD Collision Detection.

CDM Contact Domain Method.

CDT Constrained Delaunay Triangulation.

CFD Computational Fluid Dynamics.

CG Conjugate Gradient.

CIMNE International Center for Numerical Methods in Engineering.

CL Constitutive Law.

COF Coefficient Of Friction.

CPPM Closest Point Projection Method.

CPT Closest Point Transfer.

CSD Computational Structural Dynamics.

DDM Domain Decomposition Method.

DEM Discrete Element Method.

DLMM Double Lagrange Multiplier Method.

DMT Derjaguin, Muller and Toporov.

DOF Degree Of Freedom.

DOP Discretised Orientation Polytopes.

DS Degenerated Shell.

EAS Enhanced Assumed Strain.

EBST Enhanced Basic Shell Triangle.

ELM Element Transfer Method.

ETH Eidgenössische Technische Hochschule.

FAD Forward Automatic Differentiation.

FD Finite Differences.

FE Finite Element.

FEA Finite Element Analysis.

FEM Finite Element Method.

FETI Finite Element Tearing and Interconnect.

FOSS Free and Open Source Software.

FS Finite Strain.

FSI Fluid Structure Interaction.

FV Finite Volumes.

FVM Finite Volume Method.

Page 360 of 374 Vicente Mataix Ferrándiz

Acronyms Acronyms

FVTM Finite Volume Transfer Method.

GFEM Generalized Finite Element Method.

GMRES Generalized Minimal RESidual.

GNM Generalized Newton-Raphson method.

GP Gauss Point.

GUI Graphic User Interface.

HOO Hybrid Object-Oriented.

HPC High Performance Computing.

HSM Hertz-Signorini-Moreau.

IBVP Initial Boundary Value Problem.

ICT Information and Communications Technology.

IGA Iso-Geometric Analysis.

IO Input-Output.

JFNK Jacobian-Free Newton Krylov.

JKR Johnson, Kendall and Roberts.

k-DOP Discrete Orientation Polytopes.

KKT Karush-Kuhn-Tucker.

LBB Ladykaja-Babuska-Brezzi.

LGPL GNU Lesser General Public License.

LHS Left-Hand Side.

LM Lagrange Multiplier.

LMM Lagrange Multiplier Method.

LST Least-Square projection Transfer.

MDD Maximal Detection Distance.

MDOF Multiple Degrees Of Freedom.

MEF Método de los Elementos Finitos.

MFC MultiFreedom Constraint.

MLDA Multi-Layer Design Approach.

MLS Moving Least Square.

MPC MultiPoint Constraint.

MPI Message Passing Interface.

MPM Material Point Method.

Vicente Mataix Ferrándiz Page 361 of 374

Acronyms Acronyms

NASTRAN NASA STRucture ANalysis.

NC Numerical Controlled.

NCP Non-Linear Complementary Function.

NL Non-Linear.

NR Newton-Raphson.

NTN Node-To-Node.

NTS Node-To-Segment.

NURB Non-Uniform Rational B-spline.

OBB Oriented Bounding Box.

ODE Ordinary Differential Equation.

OO Object Oriented.

OOP Object Oriented Programming.

OSS Open Source Software.

OSU Ohio State University.

PDASS Primal-Dual Active Set Strategies.

PDE Partial Differential Equation.

PK1 First Piola-Kirchhoff .

PK2 Second Piola-Kirchhoff .

PM Penalty Method.

PVW Principle of Virtual Work.

REP Recovery by Equilibrium of Patches.

RHS Right-Hand Side.

SAC Symbolic and Algebraic Computational Systems.

SAT Separating Axis Theorem.

SDOF Single Degree Of Freedom.

SFT Shape Function projection Transfer.

SPR Super convergent Patch Recovery.

SS Small Strain.

STAS Segment-To-Analytical-Surface.

STL Standard Triangle Language.

STS Segment-To-Segment.

SVD Singular Value Decomposition.

Page 362 of 374 Vicente Mataix Ferrándiz

Mathematical symbols Mathematical symbols

TL Total Lagrangian.

UK United Kingdom.

UL Updated Lagrangian.

UML Unified Modeling Language.

U.S. United States.

VDI Verein Deutscher Ingenieure.

VM Von Mises.

WWI First World War.

WWII Second World War.

Mathematical symbols
C 0 The class C 0 consists of all continuous functions.

C 1 The class C 1 consists of all differentiable functions whose derivative is continuous; such functions are called
continuously differentiable.

C k When the function f is said to be of k order differentiability.

∆ Symbol for directional derivative, corresponding with the partial derivative respect the DOF.

H It is a square matrix of second-order partial derivatives of a scalar-valued function or scalar field.

J For the Jacobian. When the matrix is a square matrix, both the matrix and its determinant are referred to as the
Jacobian. Except explicitly stated the latter is the case.

κ The condition number of a function measures how much the output value of the function can change for a small
change in the input argument. A problem with a low condition number is said to be well-conditioned, while a
problem with a high condition number is said to be ill-conditioned.

hi Macaulay brackets, these are a notation used to describe the ramp function.
qP
2
L2 The Euclidean norm. As p = 2 in Lp . L2 = i=1 kxi k2 .

L∞ The infinity norm. As p approaches ∞ the Lp approaches the infinity norm or maximum norm. L∞ = maxi kxi k.
Pn
p1
Lp The p norm. For p = 1 we get the taxicab norm, for p = 2 we get the Euclidean norm. Lp = i=1 kxi kp .

∇ The vector derivative of a scalar field f .

∇2 Laplace operator or Laplacian is a differential operator given by the divergence of the gradient of a function on
Euclidean space.

π Ratio of the circumference of a circle to its diameter.

Second order PDE Second order linear PDE are classified as either elliptic, hyperbolic, or parabolic. Any second
order linear PDE in two variables can be written in the form: Auxx + 2Buxy + Cuyy + Dux + Euy + Fu + G = 0.

Vicente Mataix Ferrándiz Page 363 of 374

Rotation-free shell and solid-shell elements Rotation-free shell and solid-shell elements

elliptic PDE Second order PDE is elliptic when: B 2 − AC < 0.

hyperbolic PDE Second order PDE is hyperbolic when: B 2 − AC > 0.

Taylor expansion Taylor series is a representation of a function as an infinite sum of terms that are calculated from
′ ′′ ′′′
the values of the function’s derivatives at a single point. f (a) + f 1!(a) (x − a) + f 2!(a) (x − a)2 + f 3!(a) (x − a)3 + · · · .

Rotation-free shell and solid-shell elements

α The additional internal degree of freedom considered in the stabilisation.
kα The stiffness contribution by the α DOF in the SPRISM element.

AM The area of the central triangle of the EBST element.

C13 The mixed component in axial-normal direction of the left Cauchy tensor in the SPRISM element.

C23 The mixed component in shear-normal direction of the left Cauchy tensor in the SPRISM element.

C3 The third row of the left Cauchy tensor in the SPRISM element.

C33 The normal direction component of the left Cauchy tensor in the SPRISM element.

Cη3 The transverse shear component in η direction of the left Cauchy tensor in the SPRISM element.

Cξ3 The transverse shear component in ξ direction of the left Cauchy tensor in the SPRISM element.

I The node I index in the SPRISM element.

N I (ξ ) The shape functions in the node I in the SPRISM element.

uI The displacements in the node I in the SPRISM element.

xI The current coordinates (or deformed configuration) in the node I in the SPRISM element.

XI The original coordinates (or undeformed configuration) in the node I in the SPRISM element.

κ̂ The constant curvature field in the EBST element.

κ The curvature vector in the EBST element.

C̄ The improved left Cauchy tensor in the SPRISM element.

C̄33 The normal direction component of the improved left Cauchy tensor in the SPRISM element.

F̄ The improved deformation gradient tensor (F) in the SPRISM element.

R̄ The improved rotation tensor (R) in the SPRISM element.

Ū The improved right stretch tensor (U) in the SPRISM element.

λα The eigenvalues of U in the EBST element.

rα The eigenvectors of U in the EBST element.

f1 The deformation gradient components in the first in-plane direction (ξ ) in the SPRISM element.

f2 The deformation gradient components in the second in-plane direction (η ) in the SPRISM element.

f3 The deformation gradient components in the normal direction (ζ ) in the SPRISM element.

ft The deformation gradient components in natural coordinate derivative in the SPRISM element.

Page 364 of 374 Vicente Mataix Ferrándiz

Constraint enforcement and optimization Constraint enforcement and optimization

h0 The initial thickness of the shell.

Lei The linear shape functions of each node in the EBST element.

ϕ The coordinates of the middle surface in the EBST element.

t3 The normal in the middle surface in the EBST element.

ζ The distance from the point to the middle surface in the undeformed configuration in the EBST element.

nG The number of integration points along the direction ζ in the SPRISM element.

wie The nodal deflection of each node in the EBST element.

H This is an auxiliary relation operator between the α DOF and the displacement DOF derived from the formulation
in the SPRISM element.

rα The residual contribution by the α DOF in the SPRISM element.

B̄m The membrane tangent matrix contribution in the SPRISM element.

B̄f The modified tangent matrix in the SPRISM element.

B̄3 The normal direction tangent matrix contribution in the SPRISM element.

B̄s The shear tangent matrix contribution in the SPRISM element.

λ The parameter that relates the thickness at the present and initial configuration in the EBST element.

y3 The spatial local coordinate in the transverse direction in the SPRISM element.

Constraint enforcement and optimization

λ̄n The augmented contact stress vector, used on the ALM for contact problems. Corresponds with nλ̄n .

L This is the symbol used to represent the Lagrangian functional. Different Lagrangian can be obtained depending of
the formulation considered.

Lλ̄ This is the Lagrangian resulting of the ALM method.

Lλ This is the Lagrangian resulting of the LMM method.
λ Lagrange multiplier symbols. The Greek letter λ is equivalent to the Latin letter l, representing then the name of the
mathematician that gives the name to the method.

λ̄ On a ALM formulation, we call augmented Lagrange multiplier the value obtained after adding to our λ the
contribution of the constraint multiplied by a ε value λ + εg(x).

λn Lagrange multiplier used on normal contact. It corresponds with the normal contact pressure.
λ̄n The augmented contact stress, used on the ALM for contact problems, computed as λ¯n = λn + εg(x), if the scale
factor k is taken into consideration then can be evaluated as λ¯n = k λn + εg(x).

λτ Lagrange multiplier used on frictional contact. It corresponds with the frictional contact pressure.
λ Lagrange multiplier vector field.

g Constraint, subjects to certain restrictions a function during optimisation problems.

Vicente Mataix Ferrándiz Page 365 of 374

Finite element framework Finite element framework

f The symbol considered to represent the function to optimise on optimisation problems.

fp This is the function augmented with the PM.

ε The penalty symbol, used to represent a value considered in order to impose a certain restriction a optimisation
problem resolution. The value must be relatively big to the system of equation considered in order to take effect
over the problem resolution. An infinite value would provide a exact fulfill of the constraint, but at the same time
numerically impossible to be solved.

εn The penalty ε on the normal direction.

ετ The penalty ε on the tangent direction.

Finite element framework

αf The damping introduced on the elastic forces in the α-generalised scheme.
αm The damping introduced on the inertial forces in the α-generalised scheme.
α The parameter considered in the conceptual definition of the explicit and implicit schemes on the equilibrium
equation, being explicit when α = 0 and implicit when α = 1.

B The strain matrix, which gives the strain-displacement equation ε = Bu.

b The left Cauchy-Green tensor.

Γ Represents the boundary in a generic solid.

Γ0 Represents the boundary in a generic solid in the initial configuration.

C The right Cauchy-Green tensor.

O(h)s Represents an order s approximation in an integration scheme.

c The damping coefficient.

ccr The critical damping coefficient. Critical damping occurs when the damping coefficient is equal to the undamped
resonant frequency of the oscillator.

ξ Critical-damping ratio, usually expressed in % of ccr .

δ Stiffness-proportional damping coefficient.
η Mass-proportional damping coefficient.
F The deformation gradient is the derivative of each component of the deformed x vector with respect to each
component of the reference X vector.

R It is the rotation tensor.

U It is the right stretch tensor.

V It is the left stretch tensor.

δ u The infinitesimal incremental of displacements.

C 0 It consists of all continuous functions.
C 1 It consists of all differentiable functions whose derivative is continuous; such functions are called continuously
differentiable.

Page 366 of 374 Vicente Mataix Ferrándiz

Finite element framework Finite element framework

ˆ Represents a discretisation of the continuous form.

wgp The integration weight in a certain GP.

C The damping matrix.

K The stiffness matrix.

Kgeo The geometric stiffness matrix.

Kmat The material stiffness matrix.

M The mass matrix.

P Stress divergence or stress force term.

ξ The first local coordinate for the surface geometries, also corresponds with the local derivative on the line.

η The second local coordinate for the surface and volume geometries.

ζ The third local coordinate for the volume geometries.

∇X u The material displacement gradient tensor.

c
nm Number of nodes in the given master condition.

nsc Number of nodes in the given slave condition.

J The Jacobian operator during the NR procedure. It corresponds with the LHS of the system of equations.

φ An arbitrary motion for a given solid.

u The increment of position or displacement.

u̇ The velocity of a certain material point.

ü The acceleration of a certain material point.

X The initial position of each material particle.

x The current position of each material particle.

φ∗ It represents the pull-back operation.

φ∗ It represents the push-forward operation.

ℜ3 Represents the three-dimensional Euclidean space.

ρ The density of a given material.

f The resulting vector of forces.

N The shape function is the function which interpolates the solution between the discrete values obtained at the
mesh nodes in the FEM.

Nb The shape functions corresponding to the elements considered.

Ni The shape function value on the node i in a certain GP.

Ω Represents a generic solid for our problem.

Ω0 Represent the solid in the initial configuration.

Vicente Mataix Ferrándiz Page 367 of 374

Other terms Mortar formulation

Ωe Represent the discrete elements e which conform the domain.

Ωi Represent the i solid of the contact problem.
Ωi0 Represent the i solid of the contact problem in the initial configuration.
m The parameter used to define a certain strain measure.

E(m) The strain tensor for a given strain measure in function of the parameter m.

σ Cauchy stress.
σ̂ Corotational Cauchy stress.
τ First Piola-Kirchhoff (PK1) stress.
P Nominal stress (transpose of PK1 stress).

S PK2 stress.

t Represents the time on our problem.

∆t Represents the time increment ina given step.

Other terms
git Git is a free and open source distributed version control system designed to handle everything from small to very
large projects with speed and efficiency. See link.

C++ It is a general-purpose programming language created as an extension of the C programming language, but
Object Oriented Programming (OOP) designed.

Kratos Kratos Multiphysics (A.K.A Kratos) is a framework for building parallel multidisciplinary simulation software.
Modularity, extensibility and HPC are the main objectives. Kratos has BSD license and is written in C++ with
extensive Python interface. See B.2.Kratos Multiphysics.

Mmg Mmg is an open source software for simplicial remeshing. See B.3.Mmg library.

OpenMP Open Multi-Processing is an API that supports multi-platform shared memory multiprocessing programming.

Trilinos An OO software framework for the solution of large-scale, complex multi-physics engineering and scientific
problems.

Mortar formulation
Ae The coefficient matrix for dual λ shape functions.

De Auxiliar matrix necessary in order to compute Ae . The D is used to represent that represents the slave side.

Me Auxiliar matrix necessary in order to compute Ae . The M is used to represent that represents the master side.

ξa1 On the mortar integrated line corresponds with the local coordinate of the first segment point on the slave side.
ξb1 On the mortar integrated line corresponds with the local coordinate of the second segment point on the slave side.
ξa2 On the mortar integrated line corresponds with the local coordinate of the first segment point on the master side.
ξb2 On the mortar integrated line corresponds with the local coordinate of the second segment point on the master
side.

Page 368 of 374 Vicente Mataix Ferrándiz

Contact mechanics Contact mechanics

Jclip The J corresponding to the integration triangle during 3D mortar segmentation.

mortar Mortar finite element methods (segment-to-segment) allow for a variationally consistent treatment of contact
conditions despite the fact that the underlying contact surface meshes are non-matching or/and non-conforming.

D Mortar operator corresponding with the slave side. When using DLMM the matrix is diagonal. It is a matrix of n × n
size.

M Mortar operator corresponding with the master side. It is a matrix of m × n size.

nplane Normal which defines the auxiliary plane considered during the mortar segmentation procedure.

N̄ The shape functions in the integration triangles during the 3D surface-mortar segmentation.

xclip The resulting point coordinates of the clipping algorithm between the segments from the master and slave side.

x̂11 The first node coordinates on the slave side during a clipping procedure.

x̂12 The second node coordinates on the slave side during a clipping procedure.

x̂21 The first node coordinates on the master side during a clipping procedure.

x̂22 The second node coordinates on the master side during a clipping procedure.

Contact mechanics
Bco The discrete mortar unilateral contact operator.

bilateral Contact between two or more deformable solids.

unilateral Contact between a deformable and a rigid solid.

Tribology The science that covers the interfacial behaviour related to frictional response. Tribology covers topics like
adhesion, friction, wear, lubrication, thermal contact or electric contact.

W This space represents the dual space of the trace space of V.

β The velocity-traction ratio in friction theory.

µ The friction coefficient, also denominated Coefficient Of Friction (COF), is a dimensionless scalar value which
describes the ratio of the force of friction between two bodies and the force pressing them together.

g The Tresca law constant threshold parameter.

F Represents the friction threshold, corresponding with µN in the case of the Coulomb frictional law.

Γc The contact boundary conditions in the initial configuration.

γc The contact boundary conditions in the current configuration.
Γ1c The contact boundary conditions in the initial configuration in the domain first domain.
Γ2c The contact boundary conditions in the initial configuration in the domain second domain.
Γic The contact boundary conditions in the initial configuration in the domain i.
Γσ The Neumann boundary conditions in the initial configuration.
Γu The Dirichlet boundary conditions in the initial configuration.

Vicente Mataix Ferrándiz Page 369 of 374

Contact mechanics Contact mechanics

gn The normal gap considered during the contact computations. Can be defined as the closest distance to a certain
object in the normal direction.

tco It represents the traction on the interface.

tnco It represents the contact normal stress on the interface.

tτco It represents the contact tangent stress on the interface.

χ Represents the contact interface mapping. This means that the values from the domain 1 are projected into the
domain 2.

χh Represents the discrete contact interface mapping. Discrete version of χ.

M The Lagrange multiplier solution space.

M The vector Lagrange multiplier solution space.

Mh The discrete vector Lagrange multiplier solution space.

ln The normal contribution for the ALM frictional contact.

ξg2 The local coordinates corresponding to the GP on the master side.

ξg1 The local coordinates corresponding to the GP on the slave side.

n Represents the unitary normal vector for a given point.

τ Represents the complementary direction to the normal vector n.

τ 1 Represents the first unitary tangent vector for a given point.

τ 2 Represents the second unitary tangent vector for a given point.

lτ The frictional (tangent) contribution for the ALM frictional contact.

narea Represents the area normal vector for a given point. The area normal corresponds with the vector perpendicular
to the surface at a given point.

n̄ Represents the averaged normal vector between adjacent entities.

n The normal of a neighbour entity to a given node.

c(t) It is the relative rigid body translation between the original spatial frame and observer.

k The scale factor used in order to define the contact L contribution. Helps to improve the condition number (κ) of
the system of equations.

t1c The slave side contact traction.

Q(t) An orthogonal tensor considered in the slip frame indifferent formulation.

vτ ,rel It represents the tangential relative velocity of the contacting bodies.

ũτ It represents the nodal slip increment.

ṽτ It represents the tangential relative velocity.

U i The solution spaces defined on the contact problem.

V i The weighting spaces defined on the contact problem.

Page 370 of 374 Vicente Mataix Ferrándiz

Adaptive remeshing Adaptive remeshing

A This symbol is used in order to represent the slave active DOF in the algebraic representations of the contact
problem.

I This symbol is used in order to represent the slave inactive DOF in the algebraic representations of the contact
problem.

M This symbol is used in order to represent the master DOF in the algebraic representations of the contact problem.
N This symbol is used in order to represent the DOF which are not involved in the contact problem.
S This symbol is used in order to represent the slave DOF in the algebraic representations of the contact problem.
sl This symbol is used in order to represent the slave active slip DOF in the algebraic representations of the contact
problem.

st This symbol is used in order to represent the slave active stick DOF in the algebraic representations of the contact
problem.

û2 The displacements of the second (master) domain projected on the first (slave) domain on contact simulations.

δLco The unilateral contact contribution to the energy potential.

ḡn It represents the nodal weighted gap.

ṽτ j It represents the weighted relative velocity.

δLint,ext The internal and external contributions to the energy potential.

δLkin The kinematic contribution to the energy potential.
δLλ The unilateral contact constraint contribution to the energy potential.
δLmnt The mesh tying interface contribution to the energy potential.

x1 The coordinates on the first (slave) domain on contact simulations.

x2 The coordinates on the second (master) domain on contact simulations.

x̂2 The coordinates of the second (master) domain projected on the first (slave) domain on contact simulations.

xn The normal components of the coordinates of all nodes.

xnM The normal components of the coordinates of the master nodes.

xnS The normal components of the coordinates of the slave nodes.

Adaptive remeshing
Rλrel This anisotropic relative radius is considered in order to enforce a higher level of anisotropy on the Hessian
metric tensor.

ρ The anisotropic ratio of the mesh, which indicates the anisotropic shape factor of the element, with 1 being isotropic
mesh and 0 a collapsed geometry.

anisotropic Anisotropic meshes are those that employ extremely narrow mesh elements.

e1 The first component of the common basis in a metric (M ).

e2 The second component of the common basis in a metric (M ).

e3 The third component of the common basis in a metric (M ).

Vicente Mataix Ferrándiz Page 371 of 374

Adaptive remeshing Adaptive remeshing

f Scalar function used on the remeshing chapter to introduce concepts.

h The element size on the mesh.

hmax The maximum element size on the mesh.

hmin The minimum element size on the mesh.

hnew The new element size on the mesh.

εK This interpolation error defined in the Hessian metric remeshing.

ε The maximum level of error tolerated on the mesh elements.

Πh u The linear interpolation of u in the mesh Th in considering the FE shape functions (N).

K This identifies the element number K on the mesh Th .

M The metric which defines the size of the elements in a mesh in all the directions.

M1 The first metric considered in a metric intersection.

M2 The second metric considered in a metric intersection.

Maniso The metric considered in anisotropic mesh.

M¯ This metric defines a region that is minimal in volume.

c A constant independent of the current mesh Th .

cd A constant ratio of a mesh constant.

e(x) The vertices of a given geometry.

EK It is the set of edges of K .

ǫ The error threshold for the Hessian metric.

Miso The metric considered in isotropic mesh.

Λ Eigenvalues matrix of M .

N The common basis between two different metrics (M ).

P Eigenvectors matrix of N .

R Eigenvectors matrix of M .

Th It represents a certain mesh.

The normalisation factor considered during the Hessian metric computation to adequate to the magnitude order of
the problem.

P The reference point considered in the construction of a metric.

ai The set of unknown parameters during the SPR.

evar The error measure for a certain variable var .

êvar The error estimation for a certain variable var .

kek2E The energy norm considered as an error measure.

Page 372 of 374 Vicente Mataix Ferrándiz

Plasticity and constitutive relations Plasticity and constitutive relations

kekK The actual error in the element K .

η The relative energy norm error.
ηthreshold The relative energy norm error limit, or threshold, or sometimes also denominated permissible value.
σ̃ ∗ The nodal recovered stress values.
p The polynomial terms used to recover the stresses during the SPR.

σ ∗ The overall recovered stress values.

ξK The refinement ratio for each element when considering an adaptive remeshing.
kekthreshold The threshold error considered on the adaptive remeshing.
kekthreshold,K The permissible error in the element K .
kukE The total energy of deformation.
u The exact solution of the PDE problem.

eu The FE error, the difference between the exact solution u and the FE solution uh .

ẽu The interpolation error, the difference between the exact solution u and the interpolated solution Πh u.

uh The FE-solution.

x It represents the coordinates respect the interior vertex node of the patch x − xc .

Plasticity and constitutive relations

associated plastic flow rule An associated flow rule can be interpreted as saying that the plastic strain increment
vector is normal to the yield surface (Φ). This is denominated the normality rule.

non-associated plastic flow rule A non-associated flow rule is the normality rule is not fulfilled.

Ω0 The initial configuration.

Ωp The plastic configuration.
C The constitutive tensor.
CσT Material tangent moduli relating Truesdell rate of Cauchy stress tensor to the rate of deformation tensor.

Fe The elastic deformation gradient.

Fp The plastic deformation gradient.

Rp The plastic rotation tensor.

Re The elastic rotation tensor.

Ue , Ve The elastic stretch tensors (right and left).

Up , Vp The plastic stretch tensor (right and left).

εekl The elastic components of the strain tensor.

T
Ce The elastic right Cauchy-Green tensor. Calculated as Ce := (Fe ) · Fe .

Cep The elasto-plastic tangent operator.

Vicente Mataix Ferrándiz Page 373 of 374

Plasticity and constitutive relations Plasticity and constitutive relations

Ψ̄ The free-energy potential.

α The (strain-like) internal variables of the constitutive model.

J2 The VM criterion. Denominated that way as corresponds with the second deviatoric stress invariant. It also
represents the distortion strain energy.

η The kinematic plastic hardening internal variable.

T The Mandel stress tensor. The Mandel stress tensor is a convenient stress measure for finite plasticity.

Gf Fracture energy, the energy that has to be dissipated to open a fracture in a unitary area of the material.

ν The Poisson ratio.

γc The yield stress.
E The Young modulus of the material.

p The hydrostatic stress.

Mp The plastic flow functions.

Mh The general hardening flow function.

K The plastic hardening function, which depends on the mechanical process.

A The plastic hardening parameter.

hκ (σ , q) The scalar function of plastic hardening.

hκ (σ , q) The tensor function of plastic hardening.

κp The internal variable of plastic hardening.

γ̇ The plastic multiplier. Also denominated plastic consistency factor, it is a non-negative scalar which is determined
by the Prager consistency condition.

Γ The plastic potential.

Lp The plastic velocity gradient.

q The (stress-like) internal variables of the constitutive model, usually the hardening variables.

S̄ The stresses in the intermediate configuration.

Ψ The stored energy potential for an hyperelastic material.

w The stored energy potential for an hyperelastic material expressed in terms of the Green-Lagrange strain.

ε The strain tensor in Voigt notation.

σ The stress tensor in Voigt notation.

Π The total energy in the system in the final configuration.

Π∗ The total energy in the system in the initial configuration.

Φ The yield surface or discontinuity surface. Also denominated potential plastic surface.

Page 374 of 374 Vicente Mataix Ferrándiz