Beruflich Dokumente
Kultur Dokumente
DISSERTATION
Doktor-Ingenieur
von
Irfan Malik
Kurzfassung
Die numerischen Simulationen von physikalischen Eigenschaften, beispielswei-
se elektromagnetische, struktur- und strömungsmechanische Eigenschaften, sind
in der modernen Produktentwicklung ein wesentlicher Bestandteil. Diese Arbeit
handelt von der numerischen Berechnung innerhalb der Kontaktmechanik. Me-
chanische Kontakte kommen in fast allen Produkten des Maschinenbaus vor. Zahn-
radgetriebe, Rollenlager, Ventile und Pumpen sind nur wenige Beispiele unter ei-
ner Vielzahl von Produkten, in denen die Auslegung mechanischer Kontakte die
Lebensdauer und damit die Qualität wesentlich beeinflussen.
Die Simulation dieser Systeme in Hinblick auf maximale bzw. minimale Kon-
taktspannungen und -dehnungen, insbesondere auch aus tribologischer Perspekti-
ve, stellt jedoch besondere Herausforderungen an die dafür nötige Numerik. Her-
kömliche Verfahren, wie z.B. die Finite Elemente Methode, beschreiben die Kon-
taktflc̈hen nur näherungsweise mit Hilfe von linearen Lagrange Elementen. Da-
mit ist die Lösung des Kontaktproblems, je nach Güte des Netzes, nur unzurei-
chend genau. Darüber hinaus ist die Beschreibung der Oberflächennormalen an
Elementgrenzen wegen der fehlenden stetigen Differenzierbarkeit dort nicht ein-
deutig. Dieser Umstand erschwert die robuste Kontaktsuche, die ein wesentlicher
Bestandteil eines jeden Kontakt-Lösungsalgorithmus ist.
Diese Arbeit stellt eine neuartige Methode für die näherunghsweise Berech-
nung von nichtlinearen mechanischen Kontaktproblemen, die adaptive isogeome-
trische Kontaktformulierung basierend auf Polynomial Splines over hierarchical
T-meshes (PHT-Splines), vor. Sie ist im Vergleich zu herkömlichen Methoden ge-
nauer, robuster und effizienter. Bei ihrer Entwicklung liegt der Fokus auf dem
Lösen des statischen Kontaktproblems ohne Reibung für kleine Verformungen im
zwei- und dreidimensionalen Raum.
Die mathematische Beschreibung des Problems umfasst ein System von parti-
ellen Differentialgleichungen mit Randbedingungen und Nebenbedingungen, die
sogenannten Karush-Kuhn-Tucker Bedingungen. Letzere verhindern das nicht-
physikalische Durchdringen mechanischer Strukturen im Kontakt. Für die nähe-
rungsweise Lösung des Problems ist es nötig, das mathematische Modell in seine
integrale Form zu überführen. Dabei werden die Kontaktbedingungen mittels ei-
nes "Strafterms" (Penalty Verfahren) in den Gleichungssatz aufgenommen.
Die numerische Auswertung des Kontaktintegrals ist, wegen den in der Regel
nicht-konformen Oberflächennetze, eine besondere Herausforderung. Dazu wird
in dieser Arbeit die hoch genaue Mortar Methode verwendet. Die entstandene
ii
Der Kontaktrand ist a priori nicht bekannt. Die lokale Verfeinerung entlang
der Kontaktkante, v.a. bei 3D Problemen, ist daher keineswegs ein trivialer Schritt.
Diese Arbeit zeigt, dass die Verwendung eines bestehenden a posteriori Fehlerschätz-
Verfahrens, der Super Convergent Recovery Solution based Fehlerschätzer, zu-
sammen mit der Dörfler Markierungsmethode für diese Aufgabe gut geeignet ist.
Abstract
Numerical simulation of physical phenomena, like electro-magnetics, structural
and fluid mechanics is essential for the cost- and time-efficient development of
mechanical products at high quality. It allows to investigate the behavior of a
product or a system far before the first prototype of a product is manufactured.
Classical procedures like the Finite Element Method suffer from the non-
smooth representation of contact surfaces with discrete Lagrange elements. On
the one hand, an error due to the approximate description of the surface is intro-
duced. On the other hand it is difficult to attain a robust contact search because
surface normals can not be described in a unique form at element edges.
The mathematical model must be transformed into its integral form for ap-
proximation of the solution. Employing a penalty method, contact constraints are
incorporated by adding the resulting equations in weak form to the overall set of
equations.
For an efficient space discretization of the bulk and especially the contact
iv
One drawback of the Spline approximations of stresses though is that they lack
accuracy at the contact edge where the structures change their boundary from con-
tact to no contact and where the solution features a kink. The approximation with
smooth Spline functions yields numerical artefacts in the form of non-physical os-
cillations. This property of the numerical solution is not only a drawback for the
simulation of e.g. tribological contacts, it also influences the convergence proper-
ties of iterative solution procedures negatively.
The contact boundary is a priori unknown. Local refinement along the con-
tact edge, especially for 3D problems, is for this reason not straight forward. In
this work the use of an a posteriori error estimation procedure, the Super Conver-
gent Recovery Solution Based Error Estimation Scheme, together with the Dörfler
Marking Method is suggested for the spatial search of the contact edge.
Numerical examples show that the developed method improves the quality of
solutions along the contact edge significantly compared to NURBS based approx-
imate solutions. Also, the error in maximum contact pressures, which correlates
with the pressure artefacts, is minimized by the adaptive local refinement.
Acknowledgments
Over the last years many people have been supporting me during the preparation
of this thesis and I would like to acknowledge their efforts and assistance.
I also want to thank my advisors at Bosch Dr. Corinna Hager and Dr. Alexan-
der Menk. Alex initiated my research project and gave me great support during
the first year. Both, Corinna and Alex, guided my work in the right direction and
helped me in improving my skills in various ways. I am extremely thankful for
that.
I want to thank the former head of the department ARF at the Central Research
(CR) division of Bosch Dr. Christian Heinen who put trust in my work and gave
me the chance to pursue my PhD project in an industrial environment.
For the pleasant and inspiring working environment I would like to thank all
my colleagues at Bosch CR/ARF1, especially Dr. Karl Bendel, Helga Merz, Dr.
Heiner Storck, Dr. Sietze Van Buuren, Dr. Bernd Scheufele, Dr. Volker Gross and
Jan Schotte. With all of them I enjoyed discussing a wide variety of topics.
I deeply want to express gratitude towards my fellow PhD candidates at Bosch
Anne Schwenkenberg, Daniel Hipp, Joachim Klima and Fabian Hempert for the
fruitful discussions and their support.
Contents vii
List of figures ix
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 State of the art . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Addressed Questions . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
vii
viii CONTENTS
5 Industrial application 87
5.1 Roller Cam Contact Problem . . . . . . . . . . . . . . . . . . . . 88
5.2 NURBS based solution . . . . . . . . . . . . . . . . . . . . . . . 90
5.3 Adaptive PHT-Spline based solution . . . . . . . . . . . . . . . . 93
6 Conclusions 97
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
List of Figures
3.1 Mechanical parts roller and cam from driving train of a high pres-
sure fuel pump modelled with Non-Uniform Rational B-Splines
(NURBS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 NURBS parameterization of one quarter of an annulus in (a) initial
coarse description and (b) after h-refinement . . . . . . . . . . . . 25
3.3 Spline parameterization of 2D contact, basic principle . . . . . . . 30
3.4 Principle of isogeometric mortar integration method in contact
computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Assembly of global system of equations . . . . . . . . . . . . . . 38
3.6 Contact problem description, Hertz 2D . . . . . . . . . . . . . . . 40
3.7 Coarse NURBS discretization of Hertz 2D problem geometry . . . 42
3.8 Approximation results, Hertz 2D contact pressures with different
penalty parameter values εN and NURBS mesh from figure 3.7
with analytical solution . . . . . . . . . . . . . . . . . . . . . . . 43
3.9 Hertz 2D geometries NURBS discretization, fine mesh . . . . . . 44
3.10 Approximation results, Hertz 2D contact pressures with different
penalty parameter values εN and NURBS mesh from figure 3.9
with analytical solution . . . . . . . . . . . . . . . . . . . . . . . 44
3.11 Effect of knot relocation in the vicinity of contact edge on pressure
oscillations, Hertz 2D . . . . . . . . . . . . . . . . . . . . . . . . 45
3.12 Hertz 2D results from uniform h-refinement of contact zone . . . . 46
3.13 Hertz 2D contact pressures from successive refinement by degree
elevation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.14 Error in Hertz 2D contact pressures from numerical integration,
different number of quadrature points, quadratic splines . . . . . . 48
3.15 Hertz 3D problem description . . . . . . . . . . . . . . . . . . . . 49
3.16 NURBS parameterization of Hertz 3D contacting geometries . . . 50
ix
x LIST OF FIGURES
4.14 Pressure curves pN (r) from refinement steps of the adaptive solu-
tion process of Hertz 3D problem . . . . . . . . . . . . . . . . . . 81
4.15 Pressure contour plots with corresponding T-meshes of slave struc-
ture with zoom on contact zone (bottom view) from refinement
steps of the adaptive solution process of Hertz 3D with (a) #DOF =
25272, (b) #DOF = 36360, (c) #DOF = 82296 and (d) #DOF =
123240. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.16 Overall T-meshes of slave and master from refinement steps of the
adaptive solution process of Hertz 3D problem with (a) #DOF =
25272, (b) #DOF = 36360, (c) #DOF = 82296 and (d) #DOF =
123240 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.17 Comparison of oscillation error convergence from uniform (brown)
and adaptive local refinement (blue) of Hertz 3D approximations . 84
5.1 CAD geometry cross section of high pressure fuel pump CP4 from
Bosch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2 Simplified description of 3D roller cam contact problem . . . . . . 89
5.3 Reduced description of simplified 3D roller cam contact problem
based on rotation and plane symmetry assumptions . . . . . . . . 89
5.4 NURBS TP-meshes of roller and cam . . . . . . . . . . . . . . . 90
5.5 Contact pressure contour plots with corresponding NURBS TP-
meshes of roller cam contact 3D problem from (a) mesh 1, (b)
mesh 2, (c) mesh 3 and (d) mesh 4 (#DOF = 515160) . . . . . . . 91
5.6 3D contact pressure plots from NURBS based computation of
roller cam contact problem with (a) mesh 1, (b) mesh 2, (c) mesh
3 and (d) mesh 4 (#DOF = 515160) . . . . . . . . . . . . . . . . 92
5.7 Comparison of 2D contact pressure curves from NURBS based
computation of roller cam contact problem: (a) pN (x, y = 0), (b)
pN (x = 0, y) , (c) enlarged view of pN (y) at contact edge and (d)
enlarged view of maximum pressures max(pN (y)) . . . . . . . . . 93
5.8 Contact pressure contour plots with corresponding T-meshes (slave
side) of the 3D roller cam contact problem from adaptive refine-
ment process steps: (a) #DOF = 60840, (b) #DOF = 75960, (c)
#DOF = 95400 and (d) #DOF = 314040 . . . . . . . . . . . . . 94
5.9 3D contact pressure plots (slave side) of the 3D roller cam con-
tact problem from adaptive refinement process steps: (a) #DOF =
60840, (b) #DOF = 75960, (c) #DOF = 95400 and (d) #DOF =
314040 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
xii LIST OF FIGURES
Introduction
1.1 Motivation
Mechanical contacts appear in nearly every product of mechanical engineering.
Gearboxes, roller bearings, valves and pumps are only some examples. They play
an important role for the robust mechanical design of products with respect to
lifetime requirements.
This simple approach provides for most linear problems, like the linear elas-
ticity problem, good results. For the non-linear contact problem though, classical
finite element procedures based on the isoparametric approach suffer from the ap-
proximate and non-smooth description of contact surfaces. In addition, the quality
1
2 CHAPTER 1. INTRODUCTION
of the numerical solution strongly depends from the chosen mesh. Since systems
for geometry design and physical analysis were developed independent from each
other, the time-consuming step meshing, for the connection of those, makes clas-
sical finite element procedures in practice also inefficient.
• the exact geometry description throughout the analysis steps and the
• How are contact surfaces and variables described in a discrete setting and
what are the possible approaches for the numerical evaluation of contact
contributions?
• What are possible refinement strategies to improve the accuracy of the ap-
proximate solution?
• How can isogeometric contact solution procedures be made adaptive for the
automatic and efficient improvement of the solution?
1.2. STATE OF THE ART 3
Standard monographs on the topics contact mechanics, linear and non-linear Fi-
nite Element Analysis (FEA), Computational Contact Mechanics, NURBS and
IGA are [46], [77], [79], [8], [82], [51], [60] and [16]. An overview of isogeomet-
ric contact implementations and classical contact solution procedures is given in
[19].
Coming to the second question, presented contact publications apply the basic
idea of IGA by describing boundaries in contact with finite elements exactly (no
matter how coarse the mesh) from the Spline structure of geometries [41]. Con-
tact variables are approximated with e.g. NURBS basis functions based on the
isoparametric concept and Galerkins approach (see [17] and [16]). For local con-
tact search, the gap between contacting structures must be determined based on
the evaluation of closest point projections (CPP). Numerical procedures presented
therefor for example in [54] and [55] employ the smooth and continuous NURBS
parameter form of the contact boundaries.
It is important to note that, in general, analysis suitable NURBS construction
of the geometries must be watertight, i.e. no gaps or overlaps are allowed. This
is not a trivial task, especially for complex geometry design. Different techniques
and studies on the construction of Spline geometries suitable for analysis are given
in [63], [53], [75], [83] and [76].
One challenging task, also in isogeometric implementations, is the numerical
evaluation of contact contributions in variational form based on generally non-
conforming meshes of the structures. This task is solved by adopting conven-
tional methods from classical finite element analysis (see [30],[31], [9], [78]) to
the isogeometric setting (see [56], [41], [54], [18]). A detailed discussion of the
available approaches is given in the introduction of section 3.3.
4 CHAPTER 1. INTRODUCTION
Studies, which are not directly related to contact but which are relevant for
modelling interfaces with non-matching discretization in the context of IGA are
given in [61], [39] and [4].
5), the following three open questions are put, to guide through the thesis:
3. What are the possibilities and drawbacks of such adaptive procedures when
applied not only to academic examples, but rather used for industrial appli-
cation?
1.4 Outline
This thesis is structured in 6 chapters. After this introductory chapter, the friction-
less contact problem in which bodies undergo small deformations is stated in its
weak form. Relevant parts from continuum and contact mechanics together with
contact conditions are presented for this purpose.
are then compared to results from chapter 3. This chapter also closes with a short
summary.
Finally, Chapter 6 concludes the presented thesis with a summary and an out-
look on future research directions.
Chapter 2
7
8 CHAPTER 2. CONTINUUM AND CONTACT MECHANICS
2.1.1 Kinematics
A single Point P of a body B in its initial configuration Ω0 ⊂ Rd of spatial dimen-
sion d is described with the position vector X. Another point Q of this structure
has the position X + dX, described with the relative distance dX to P. When the
structure is deformed from external load, both points are displaced to x and x + dx
in the current configuration Ω ⊂ Rd . The displacements u of both positions are
given with the vector operations:
u(X) = X − x
(2.1)
u(X + dX) = (X + dX) − (x + dx).
The amount that both positions are displaced relatively to each other after defor-
mation is given with the vector du and it is computed from the total differential of
u with
∂u
du = dX =: H dX. (2.2)
∂X
In (2.2) the quantity H is called the displacement gradient tensor and when it is
written out, it consists of a symmetric part ε and an anti-symmetric part ω :
1 1
H = [(∇u)T + (∇u)] + [(∇u)T − (∇u)] (2.3)
|2 {z } |2 {z }
ε ω
1
ε = [(∇u)T + (∇u)] (2.4)
2
is a measure of small deformation, i.e. kuk 1 and k∇uk 1, of linear elastic
structures.
2.1.2 Kinetics
A structure is in static balance while it is subjected to surface tractions t dΓ and
body forces ρb dΩ only, when
Z Z
ρb dΩ + t dΓ = 0 (2.5)
Ω Γ
and Z Z
(x × ρb) dΩ + (x × t) dΓ = 0. (2.6)
Ω Γ
2.1. LINEAR STRUCTURAL MECHANICS 9
Equation (2.5) gives in spatial coordinates the balance of linear momentum in the
global form and (2.6) is the balance of angular momentum. Both equations are
for the static case. Applying now the relation
t(x, n) = n · σ (x), (2.7)
between stress tensor σ , which describes with its components σi j with i, j =
1, · · · , d the state of stress at any given point P ∈ Ω, and the traction vector t
together with the Gauss divergence theorem to equation (2.5) yields
Z
(∇ · σ + b) dΩ = 0. (2.8)
Ω
In (2.7), the outer unit normal vector at an arbitrary boundary point x is denoted
by n. This linear mapping is known as the Cauchy formula. Equation (2.8) are the
balance equations of a general continuum and their local form is given by
∇ · σ + b = 0. (2.9)
The set of partial differential equations in (2.1) - (2.9) is the first part for the for-
mulation of boundary value problem of linear elasticity.
The local from of (2.6), derived in a similar way as it was done for (2.5), is
equivalent to the relation
σ = σT. (2.10)
The interested reader is referred to section 4.1 in [2] for details. The symmetry
of σ is a result which is known as Cauchys second equations of motion. It is not
explicitly part of the boundary value problem of linear elasticity, (2.10) is rather
implicitly required.
Representing the stress tensor on the one hand as the fist derivative of φ with
respect to ε and on the other hand with (2.11) gives the relation
∂φ
E:ε = ,
∂ε
which leads after differentiating both sides to the expression
∂ 2φ
E= (2.13)
∂ε∂ε
for the elasticity tensor. Due to the interchangeable sequence of differentiation in
(2.13) the elasticity tensor is symmetric. This reduces its independent coordinates
to 21 in three dimensions (d = 3).
For the contact problem model only isotropic materials are assumed, i.e. their
response is indifferent from the directions. This allows to represent the remaining
components of E with
The two material parameters λ and µ in (2.14), the so-called Lamé constants, are
used for the mathematical description of the model. Their expression with the
elasticity modulus E and poisson ratio ν, well-known in engineering, is given by
νE E
λ= , µ= . (2.15)
(1 + ν)(1 − 2ν) 2(1 + ν)
The index-notation Ei jkl is done with the Kronecker-delta symbol δi j defined by:
(
1 for i = j,
δi j =
0 for i 6= j.
Γu ∪ Γσ = ∂ Ω and Γu ∩ Γσ = 0.
/ (2.17)
2.2. CONTACT MECHANICS OF SMALL DEFORMATIONS 11
The introduced equations of the overall section 2.1 build the setting of bound-
ary value problem (bvp) of linear elasticity.
in case Γl are locally convex. The convective coordinates ζ ∈ [0, 1]d−1 in (2.19)
indicate the parameter form of the master surface description. The coordinates ζ¯
corresponding to the CPP are found, when surface normal and tangential vectors
are perpendicular, i.e. n̄ · x(ζ¯ ),ζ (1) = 0 and n̄ · x(ζ¯ ),ζ (2) = 0 for a three-dimensional
contact.
With x(ζ¯ ) the normal gap function gN is defined as the scalar product of
the difference between position vectors and the normalized surface normal of the
CPP:
gN = (xs − x(ζ¯ )) · n̄. (2.20)
12 CHAPTER 2. CONTINUUM AND CONTACT MECHANICS
In case two structure penetrate and the distance is negative, the linear elas-
tic structural problems of both structure need to be modified with contact terms.
Additional stresses tN normal to the surface are build. For positive distances the
structural equations remain unchanged.
R(u) = ∇σ
σ (u) + b = 0 in Ω (2.24)
These integral equations are completely equivalent to the residuals. The solution
of (2.26) though has to fulfill the field equations only in average over the spatial
domain, while the solution of (2.24) and (2.25) has to be fulfilled point-wise. The
test functions δ u are introduced for gaining more flexibility in finding the average
solution. It is important for the test-functions to fulfill
δu = 0 on Γu , (2.27)
since the solution is prescribed there in the form of the Dirichlet boundary condi-
tion (2.18a). For weaker differentiability requirements to the solution, the weighted
residual is further transformed to
Z Z Z
T T
(∇δ u) E ∇u dΩ − δ u b dΩ − δ uT t̂ dΩ = 0, (2.28)
| Ω {z } |Ω Γσ
{z }
δ Πi δ Πe
n d o
V = δ u : δ u ∈ H 1 (Ω) , u|Γu = 0
(2.30)
With (2.29) and (2.30) vector valued functions of dimension d are defined in so-
d
called Sobolev spaces denoted with H 1 (Ω) . Characteristic for those functions
is that they are square integrable, i.e. Ω u2 dΩ < +∞, and their first derivatives
R
are square integrable as well. Based on (2.28), (2.29) and (2.30) the weak form
of the linear elasticity problem can be now noted in the following compact form:
Find u ∈ S such that
δ Πi − δ Πe = 0, ∀ δu ∈ V . (2.31)
Here balance equations and the Neumann boundary conditions will be fulfilled
in weak sense. Material law, kinematics and Dirichlet boundary conditions are
fulfilled point-wise.
2.3. WEAK FORM OF CONTACT PROBLEM 15
to treat only the active contact zone Γ̃c . The parameter εN in (2.32) can be in-
terpreted in context of the contact problem as the mechanical stiffness of springs
which act with the pressure
∂ Πc (u)
Z Z
εN
δ Πc (u) = δ uT = εN ūN δ ūN dΓ + ū2N δ dΓ . (2.35)
∂u Γ̃sc 2 Γ̃sc
| {z }
≈0
With neglecting the second summand in (2.35) for small displacements, contact
constrains are incorporated in a weak sense.
16 CHAPTER 2. CONTINUUM AND CONTACT MECHANICS
u = u1 , u2 and δ u = δ u1 , δ u2 .
(2.36)
In the same way, requirements for the trial- and test-functions defined by (2.29)
and (2.30) for each structure separately, are summed up in the function spaces
S := S 1 × S 2 , V := V 1 × V 2 , (2.37)
for Ω = 2l=1 Ωl . Now the weak form of the bilateral contact problem is written
S
The contact problem, presented in the previous chapter in a generic form, is now
applied to actual geometry. Typically, geometries are designed in Computer-
Aided-Design (CAD) systems. The underlying mathematical representation in
such programs is most often in parameter form using Non-Uniform Rational B-
Splines (NURBS). Mechanical and geometry model describe the contact problem
in a unique form.
Closed form solutions for contact problems with arbitrary geometry design
are not possible. Finite Element procedures for example, provide an approximate
solution instead. The problem is transformed here into a discrete form. The geom-
etry is divided into a finite number of sub-domains, the Finite Elements, in which
the governing equations are fulfilled locally or element-wise. The error from ap-
proximation of can be reduced when a large number of finite elements is used.
17
18 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
The first Section 3.1 of this chapter gives a brief overview of geometry mod-
elling with NURBS by summarizing relevant definitions and properties.
Using shape function from CAD-based geometry design for approximating the
solution of problems partial differential equations, is the key idea behind isogeo-
metric analysis. In Section 3.2 this idea is described in more detail and relevant
properties are summarized before it is applied to the linear elasticity part of the
contact problem. This section is closed with an overview of refinement strate-
gies for NURBS parametrized models to reduce the approximation error of the
solution.
The isogeometric paradigm is applied to the contact part of the problem in
section 3.3. For this the parameter form of the NURBS geometry description is
inherited for the discrete representation of the contact boundary. Normal gap func-
tion and its variation are transformed to a discrete setting by using smooth shape
functions from the this very same surface representation. For the general case
of non-matching isogeometric finite element meshes of the contacting partners,
the mortar method is employed for the numerical evaluation of the contact inte-
gral. Due to the non-linear character of the problem, stemming from the unknown
contact surface, the discrete form of the overall problem is solved iteratively by
applying the Newton-Raphson method.
In section 3.4, computational examples in from 2- and 3-D are presented to
show the impact of polynomial order, mesh size, penalty parameter and the num-
ber of integration points on the approximated contact pressures. From these results
chapter 4 is motivated.
This chapter closes with a summary.
Figure 3.1: Mechanical parts roller and cam from driving train of a high pressure
fuel pump modelled with Non-Uniform Rational B-Splines (NURBS)
for all ξ = (ξ (1) , ξ (2) , ξ (3) ) with the multi-indices n = (n1 , n2 , n3 ), p = (p1 , p2 , p3 )
and i = (i1 , i2 , i3 ) for number, degree and index. The overall set of n1 n2 n3 basis
functions span the space
The single knot-vectors Ξ(1) , Ξ(2) and Ξ(3) , used for defining (3.6), build the tensor
product mesh
in which a single element Ek has the global index k, given by k := (k3 − 1)m1 m2 +
(k2 − 1) · m1 + k1 .
The definition of bivariate basis functions is done in an analog way.
3.1.2 Geometries
Based on section 3.1.1 the rational piecewise parametric representation of NURBS
geometries is given here.
(α) (α)
The components of the control points are Pi /wi , with Pi being the spatial co-
ordinates and wi being the corresponding weights.
(α) (α) d
h i
Surfaces and volumes x(ξξ ) : ξ1 , ξmα → Rd with d = 2 resp. d = 3 are
defined by
n
x(ξξ ) = ∑ Rpi (ξξ )Pi . (3.10)
i=1
Similar to (3.6) the multi-index notations i, n and p are employed in (3.10). The
(α)
control point components are Pi /(∏dα=1 wid ). Connecting control points of
curves, surfaces and volumes with straight lines, control polygons, control nets
resp. control meshes are developed.
weights. The overall set of m finite elements, describing the geometry exactly,
i.e. Ω = Ωh = m
S
e=1 Ωe , is defined in parameter space Ω0 by the tensor product
mesh T (see (3.8)) and mapped to physical space by a global geometry function
G(ξξ ) : Ωh0 → Ωhe . Conventionally G(ξξ ) is the geometry description with NURBS.
To ensure a bijective or invertible mapping between parameter and physical space,
the NURBS construction of the geometry has to be waterproof, i.e. no gaps or
overlaps are allowed.
Based on (3.11), trial- and test functions u(x), δ u(x) with x ∈ Ω of the un-
known displacement field are approximated by
n n
uh (x) = ∑ Rpi ◦ G−1 (x) qi , δ uh (x) = ∑ Rpi ◦ G−1 (x) δ qi . (3.12)
i=1 i=1
In (3.13) coefficients of the algebraic equations stemming from the bilinear form
of the problem, build the global stiffness matrix K ∈ RN×N with N = d · n be-
ing the product of spatial dimension d and the overall number of basis function
n = ∏dα=1 nα . The load vector entries in f ∈ RN result from discretization of the
problems linear form. The unknown control point displacements and their varia-
tions form the vectors q and δ q. Homogeneous Dirichlet boundary conditions are
considered by zeroing out rows and columns corresponding to the non-zeros basis
functions.
Kq = f. (3.14)
Solving (3.14) and displacing the control points of the spline geometry from initial
configuration with the corresponding values from q, yields the deformed spline
geometry for the given structural problem.
In (3.15) the matrices Bi are defined by partial derivatives of rational basis func-
tions in the form
Ri,x(1) 0
Bi = 0 Ri,x(2) . (3.16)
Ri,x(2) Ri,x(1)
The derivatives with respect to parameter coordinates are transformed to deriva-
tives with respect to corresponding physical coordinates by the rule ∇x Ri (x) =
∇ξ Ri (ξξ ) · DG(ξξ )−1 . In this rule DG is the Jacobian of the geometry function
G(ξξ ).
More details are given in [7].
3.2. ISOGEOMETRIC DISCRETIZATION CONCEPT 25
(a) (b)
The possibilities for approximating the deformation field based on this rela-
tively coarse geometry description are limited when the structure is subjected to
external forces and displacements. Thus the initial geometry is refined (see fig-
ure 3.2 (b)). Several approach for refining the geometry respectively for including
additional basis functions to trial and test function spaces are possible. All meth-
ods for refining NURBS geometries have in common, that the geometrical shape
of the problem is not changed. This is explicitly beneficial for the purpose of do-
ing analysis.
where
1,
1 ≤ A ≤ k− p
¯ A
αA = ξ ξ −ξ−ξ , k − p + 1 ≤ A ≤ k, (3.18)
A+p A
0, A ≥ k + 1.
The knot insertion in 2D and 3D has a global effect on the mesh of one patch
geometries due to the tensor product structure of NURBS. For this reason, local
refinement is not simply possibly with this procedure.
Classical finite element procedures for the solution of highly non-linear con-
tact problems suffer from the non-smooth representation of surfaces with La-
grange Elements. The robustness of contact search is difficult to attain because
the surface normal representation at the element edges is not unique due to the
3.3. NURBS DISCRETIZATION OF CONTACT 27
• contact detection,
Generally multiple NURBS patches are required for the exact representation
of complex geometries which again requires patch-coupling techniques presented
for example in [27], [83] and [61]. The geometry examples in this work are mod-
elled with single NURBS patches. Details on patch coupling techniques are omit-
ted here for this reason. One benefit of single patch geometry representations is
the robust contact search. For the iterative determination of closest point projec-
tions with Newtons Method continuous second derivatives are required. This can
only be assured with single patches.
Numerical evaluation of the contact part in isogeometric setting has been done
already with
28 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
following in principle the classical finite element counterparts. A review and dis-
cussion of the listed approaches is given in [19].
The error coming from the numerical integration based on this approach is fur-
ther reduced by implementing the two-half-pass version GPTS-2hp, which com-
putes the contact part twice, switching the role of slave and master. Both integra-
tion results are then averaged.
For an accurate and robust integration, in this work the isogeometric mortar
method is applied for the numerical evaluation of the contact part. Constrains
are fulfilled here in a weak sense by projecting them to control-points of the slave
3.3. NURBS DISCRETIZATION OF CONTACT 29
In the first part of this section 3.3.1 contact surfaces and variables are repre-
sented in a discrete form based on the isogeometric setting for 2- and 3D NURBS
geometries. The second part 3.3.2 contains details on the local contact search. The
isogeometric Mortar Method for the numerical evaluation of the contact integral
resulting in a stiffness matrix and residual force vector is presented in the third
part 3.3.3. The overall system is then solved in section 3.3.4 after assembling the
discrete sets of equations from the linear elasticity parts and the contact part to
a global non-linear system of algebraic equations. The section is closed with the
presentation of numerical examples and a summary.
for the presented example. Their parameter form is derived from the NURBS
(2)
geometry description by keeping the parameter ξ (2) := ξ0 fix. Non-zero basis
functions
(1) (2)
R− −
i (ξ ) := Ri1 ,i2 =1 (ξ , ξ0 )
in (3.19) are the i = 1, · · · , p1 + 1 subset of bivariate basis functions Ri1 ,i2 (ξξ )
which describe the corresponding surface element. Control-points, associated
with the contact curve are denoted by P− i1 ⊂ Pi1 ,i2 .
(1)
Here, the parameter ζ (1) := ζm1 is defined fix and the j = 1, · · · , p2 + 1 non-zero
basis functions are
(1)
R+j (ζ ) := R+j1 =m1 +p1 +1, j2 (ζm1 , ζ (2) )
d−1
In 3D, contact surface elements are described by i = 1, · · · , ∏α=1 (pα + 1)
(1) (1)
control-points and the corresponding basis functions Ri (ξξ ) with e.g. ξ ∈ [ξ1 , ξm1 ]×
(3) (3) (2) (2) (3) (3)
[ξ1 , ξm3 ], respectively R(ζζ ) with e.g. ζ ∈ [ζ1 , ζm2 ] × [ζ1 , ζm3 ].
Discretization of variables
l,h
After the NURBS representation of the boundary elements Γc,e , the contact
integrand is now transformed to its discrete form:
εN δ uN uN → εN δ uhN uhN
Contact element control points from the initial configuration of salve and master
are denoted here with P− +
i,0 and Pi,0 .
The overall discrete form δ Πhc,e of the contact part from one element in pen-
etration is derived by inserting the single contact quantities from (3.21), (3.22)
and (3.23) into (2.35):
!
Z p1 +1 p2 +1
δ Πh =
c,e R− (ξ )δ q− −
∑ i R+ (ζ¯ )δ q+ n̄
i ∑ j j
Γsc,e i=1 j=1
!
p1 +1 p2 +1
∑ R− −
i (ξ )qi − ∑ R+j (ζ¯ )q+j n̄ + · · ·
i=1 j=1
!
p1 +1 p2 +1
···+ ∑ R− −
i (ξ )Pi,0 − ∑ R+j (ζ¯ )P+j,0 n̄ dΓ. (3.24)
i=1 j=1
The surface normal evaluated at the CPP on master side is denoted herein with n̄.
For a more concise notation of (3.24), the single contributions are sorted into
matrices:
− − − −
P1,0
R1 (ξ ) 1d q1 δ q1
.
.. .
. .
. ..
. . .
− − − −
Rnsc,e (ξ ) 1d qnsc,e δ qnsc,e e P sc,e ,0
Ce = qc,e = + δ qc,e = + P = n+
+ ¯ P (3.25)
−R (ζ ) 1d q δq c,0
1 1 1 1,0
.. .. ..
..
. . . .
+ ¯ + +
−Rnmc,e (ζ ) 1d qnsc,e δ qnsc,e P+
nm ,0 c,e
where 1d is the identity matrix of dimension d and are number of non- nsc,e , nm
c,e
zero basis functions. Now (3.24) can be written in the form
Z
δ Πhc,e = δ qTc,e CTe n̄n̄T Ce qc,e − CTe n̄n̄T Ce Pec,0 dΓ = fc,e
h
(ξ , ζ¯ ). (3.26)
Γsc,e
contact search, are therefore an important part of the overall contact solution pro-
cedure.
The aim of a local search is to find the parameter ζ¯ corresponding to the CPP
x̄ ∈ Γm,h s h
c on the master side for a given discrete point x ∈ Γc and evaluate the sign
of uN . With the local search and the penetration function (2.33), active parts, i.e.
the parts in penetration, of the contacting boundaries are identified for evaluation
of the contact integral (see section 3.3.3).
In 2D, the implementation of local search is done with the extremal condition
f (ζ ) = xm m s
,ζ (ζ ) · (x (ζ ) − x ) = 0. (3.27)
The initial guess ζ0 builds the start of Newton’s procedure. Iterating then with
the steps
The initial parameter ζ0 plays an important role for the convergence of New-
tons procedure. Dependent from this parameter, solutions from two iteration steps
can start oscillating between each other and no convergence is attained. In the con-
sequence, it influences the overall robustness of a contact algorithm. Following
[60], in this work the initial guess is determined in two steps. First, those master
contact elements are identified from a bucket search ([72]) which are candidates
for containing the initial guess. In a second step, grids of discrete points (guess
points) in the determined master elements are chosen for a finer search for the
shortest distance. This second step can be omitted for a better performance, in
case of contact surfaces with high regularity.
Partial derivatives xm
,ζ
and xm
,ζ
in (3.28) are determined by applying the prod-
uct rule (see e.g. [7]).
The solution ζ¯ to (3.27) is found, when either both points xs and xm (ζi ) co-
incide with a given tolerance ε1 or when the cosine of both vectors xm (ζ ) and
,ζ i
m s
x (ζi ) − x is in a given tolerance ε2 (see (3.29)).
For the 3-dimensional case, the pair of parameters (ξξ , ζ¯ ) is found, when the
extremal conditions
f (ζζ ) = xm
,ζ (1)
(ζζ ) · (xm (ζζ ) − xs ) = 0 and (3.30a)
g(ζζ ) = xm
,ζ (2)
(ζζ ) · (xm (ζζ ) − xs ) = 0 (3.30b)
are fulfilled. These two coupled algebraic equations with the surface basis vectors
xm
,ζ (1)
and xm
,ζ (2)
and the distance vector d := xm (ζζ ) − xs are again nonlinear.
(1) (1) (2) (2)
Their solution ζ¯ ∈ [ζ1 , ζm1 ] × [ζ1 , ζm2 ] is found iteratively starting with
the initial guess ζ 0 . The iteration steps are defined by
f (ζζ i ) −1
ζ i+1 = ζ i − J (3.31)
g(ζζ i ) i
|xm
,ζ (1)
(ζζ i ) · (xm (ζζ i ) − xs )| |xm
,ζ (2)
(ζζ i ) · (xm (ζζ i ) − xs )|
≤ ε2 , ≤ ε2 . (3.33)
|xm
,ζ (1)
(ζζ i )| · |(xm (ζζ i ) − xs )| |xm
,ζ (2)
(ζζ i )| · |(xm (ζζ i ) − xs )|
3.3. NURBS DISCRETIZATION OF CONTACT 35
The initial guess yields from a two-step approach described above for the 2D case.
Herein the first, second order and mixed partial derivatives are denoted with xm ,ζ (i)
,
m m
xζ (i) ,ζ (i) and xζ (i) ,ζ ( j) with i, j ∈ {1, 2}. Their computation formula for NURBS
parameter description of xm (ζζ ) can be found for example in [16]. Note that for
positive definite Jacobian matrices, the iterations will converge to a unique solu-
tion ζ¯ .
Similar to the 2D case, the convergence criterion (3.32) checks if slave and
master points xs and xm (ζζ i ) coincide with a given tolerance ε1 in iteration step
i. The criteria given in (3.33) check on the other hand if basis and distance
vectors xm
,ζ (1)
(ζζ i ), xm
,ζ (2)
(ζζ i ) and (xm (ζζ i ) − xs ) are orthogonal to each with a given
tolerance ε2 .
A special property of contact integrals is that two structures are involved with
their own surface parameterizations. During integration, this must be considered
at the same time. In addition, contact meshes result from independent geometry
parameterizations and therefore they do not match in general. For this reason, a
simple numerical integration of the contact integral is not possible. The evaluation
must be done in a piecewise form.
Z n∗c,GP
h
fc,e (ξ , ζ¯ ) dΓ ≈ fch (ξi , ζ¯ )wi ˆ ˜
det DF(ξi ) · det DF̂(ξi ) . (3.34)
Γc,e
∑
i=1
In (3.34) fch (ξ , ζ¯ ) represents the discrete form of the contact integrand. The num-
ber of quadrature points nc,GP is chosen with regard nc,GP ≥ (p + 1)d , where p
is the polynomial degree of slave basis functions (influence of differently chosen
nc,GP on numerical solutions is addressed section 3.4.1), and d is the number of
physical dimensions.
3.3. NURBS DISCRETIZATION OF CONTACT 37
Elements, which are not in penetration over the whole length, contribute only
with gauss-points i = 1, · · · , n∗c,GP in penetration to the numerical integration (for
more details, see [30]). In figure 3.4 they are represented with red stars. Gauss-
points not in penetration are shown with gray stars.
Positions of the integrations points ξi are defined on the interval [−1, 1] for
each parametric coordinate and are computed together with the corresponding
weights wi from the roots of Legendre polynomials. The quadrature rules are
listed for example in [85] or in [43].
The discrete representation of the contact integrand (see section 3.3.1) and the
numerical integration presented here, yield the algebraic equations
δ Πhc,e = δ qc,e Kec qc,e + δ qTc,e Kec Pec,0 . (3.35)
These equations represent the linearized contact part of the non-linear problem for
one element Γhc,e .
The set of equations (3.35) from all contact elements e = 1, · · · , n∗c in penetra-
tion are assembled, i.e.
∗
nc
[
δ Πhc = δ Πhc,e ,
e=1
and variational contact displacements in δ qc are set to one, yielding the overall
linear set of equations
δ Πhc = Kc qc + Kc Pc,0 . (3.36)
| {z }
fc
In (3.36) the global contact stiffness matrix is denoted with Kc and the contact
residual force vector is denoted with fc . In a global solution procedure for non-
linear contact problems, Kc and fc must be determined in each iteration step from
scratch and added to the global system which consists of the linear elasticity parts
of slave and master (see next section).
In literature, methods like the Fixed-Point Method, the Modified or Quasi New-
ton Raphson Method or the Arc-Length Continuation Method are suggested (see
[79], [8] and [72]). In this work the classical Newton Raphson Method is applied
for the approximate solution of the static structural contact problem.
The starting point of Newton‘s method is the residual form of the problem
given by
r(q) = (Ks + Km + Kc (q)) · q − (fs + fm ) + fc (q) = 0. (3.37)
| {z } | {z }
fint (q) fext (q)
In a second step, r(q) is linearized in terms of q:
∂r
r̃(q) = r(qi ) + ∆q + · · · (3.38)
∂q
|{z}
KT
The initial guess q0 plays, as described already in sec. 3.3.2, an important role
for the convergence behavior of Newton methods. For the computation of numer-
ical examples in the next sections, control-point displacements q0 are derived by
displacing the slave structure from its initial configuration until it penetrates the
master side.
The approximate solutions to the non-linear problem with a given load step
is found, when the solutions qi do not change significantly by ∆qi , i.e. k∆q ik
kqi k is
smaller a given parameter εq , or when the difference between internal and exter-
kf k
nal forces f int,i is smaller than a given parameter εr .
k ext,i k
Based on the found solution q and the initial configurations of the contacting
structures, the displaced NURBS geometries in contact are computed. The com-
putation of strains and stresses is done in an additional post processing step based
on equations from 2.1.
40 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
• the number of integration points for the numerical evaluation of the mortar
integral
are varied to investigate their effect on the accuracy of approximate results and
robustness of the solution procedure.
3.4.1 Hertz 2D
The 2D Hertz problem is derived from contact situation of an infinitely long cylin-
der resting on a rigid plane and subjected to a line load F by assuming plane strain
in the cross section of the cylinder. Figure 3.6 depicts the simplified problem.
u(1) = 0 on Γsu,1 , Γm
u,1 ,
u(1) = u(2) = 0 on Γm
u,2 .
The load F is applied with the traction force vector t̂, with
F
Z
= t̂ dΓ,
2 Γsσ
Analytic solution
The analytical solution of the contact pressure distribution, taken from [46], is
described with the material constants E and ν, the radius R of the slave structure
and the applied force F (see Figure 3.6). For the contact radius a the formula is
given by r
FR
a = 2· , (3.41)
πE ∗
with E ∗ = E
(1−ν 2 )
. The analytic maximum pressure p0 is defined by
2F
p0 = . (3.42)
πa
Both, (3.41) and (3.42) define then the quadratic pressure distribution
s
(1) 2
(1) x
p(x ) = p0 1 − (3.43)
a
NURBS discretization
The NURBS model of the problem is depicted in figure 3.7 with its tensor
product bi-quadratic mesh. For salve and master, contact parts of the domains are
0.2
0.1
0 0.2 0.4
meshed with a higher elements density. Due to the tensor product structure of
NURBS this refines also parts of which do not necessarily contribute to a more
accurate approximate solution. Due to the fact that numerical evaluation of the
contact integral is performed on the slave side, the mesh of master side is chosen
coarser. This increases efficiency of the contact search.
σ (x) n) · n
pN (x) = (σ (3.44)
3.4. NUMERICAL RESULTS 43
and normalized with p0 and a, are shown in figure 3.8. The analytical solution is
represented with the blue solid curve.
N
= 10 7
1
N
= 10 10
0.8 = 10 15
N
0.6
pN /p0
0.4
0.2
-0.2
0 0.5 1 1.5 2
x/a
Computation results though, depicted in figure 3.10 and gained from the same
three penalty parameter values but with a finer mesh for the contact region of
the slave structure (see figure 3.9), show significant differences. Only parameter
εN = 1010 provides for the chosen mesh acceptable results.
44 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
0.2
0.1
0 0.2 0.4
analytical solution
1
N
= 10 7
N
= 10 10
N
= 10 15
0.5
pN /p0
-0.5
0 0.5 1 1.5 2
x/a
These results show that also for Spline discretized contact computation, there
is a clear dependency of the chosen mesh and the penalty parameter. It is also no-
ticeable that all approximate solutions shown above feature oscillatory behavior
in vicinity of the contact end, i.e. for x/a > 1. These oscillations differ only in
frequency and amplitude. Following the explanations given in [32], this artificial
behavior is due to the inability of smooth basis functions, like Spline basis func-
tions, to capture the local C0 -continuous feature of the analytical solution.
1.2
0.8
0.2
0.6
pN /p0
0.1
0.4 0
0.2 -0.1
1 1.5 2
0
-0.2
0 0.5 1 1.5 2
x/a
Figure 3.11: Effect of knot relocation in the vicinity of contact edge on pressure
oscillations, Hertz 2D
10-2
1.2
0.8
0.1 10-3
0.6
0.4 0
0.2
-0.1
1 1.5 2
0 10-4
-0.2
0 0.5 1 1.5 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
104
It is also interesting to see in figure 3.12 (b) that the relative local error from
oscillations, quantified by
R 2a h
rel |pN | dx
eh = a (3.45)
p0
reduces monotonously with each refinement step.
1.2 1.2
1 1
0.8 0.8
0.1 0.1
0.6 0.6
0.4 0 0.4 0
0.2 0.2
-0.1 -0.1
1 1.5 2 1 1.5 2
0 0
-0.2 -0.2
(a) degree elevation results mesh 1 (coarse) (b) degree elevation results mesh 2 (fine)
For both meshes, the approximate pressures improve with each refinement
step. Especially the steep character of the analytic solution is captured more ac-
curately with the enriched approximation space. Different from e.g. h-refinement
result tough, the oscillatory behavior of the solutions does not improve. The fre-
quency and also the amplitudes increase with the order. This observation is also
known as the Gibbs phenomenon (see also [42], sec. 3).
At last, the error from chosen number of gauss points nGPc for numerical inte-
gration of the mortar integral is investigated. Generally, this error is assessed by
the so-called contact patch test with plane interfaces (see [28]). Here, the Hertz
2D problem with its curved interface is chosen for this purpose in the two bi-
quadratic Spline discretizations (coarse and fine mesh). Then, contact pressure
are computed with varying number of Gauss-points, i.e. nGPc = 2, · · · , 6.
In figures 3.14 (a) and (b) the approximated pressure curves are depicted.
48 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
1.2 1.2
1 1
0.8 0.8
0.1 0.1
0.6 0.6
0.4 0 0.4 0
0.2 0.2
-0.1 -0.1
1 1.5 2 1 1.5 2
0 0
-0.2 -0.2
(a) contact pressures, mesh 1 (coarse) (b) contact pressures, mesh 2 (fine)
0.12 0.12
0.1 0.1
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
-0.04 -0.04
-0.06 -0.06
-0.08 -0.08
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(c) errors from different number of gauss (d) errors from different number of gauss
points for mesh 1 (coarse) points for mesh 2 (fine)
Figure 3.14: Error in Hertz 2D contact pressures from numerical integration, dif-
ferent number of quadrature points, quadratic splines
While pressure curves computed from the coarse discretization do not differ
significantly from computations with nGPc > 2, results gained from mesh 2 vary
for each nGPc . The figures 3.14 (c) and (d) show the relative error curves
with p∗N = pN,nGPc =6 . From these curves it can be seen that the last significant
improvement of the error is gained from increasing nGPc from 3 to 4. This is the
case for both meshes. The numerical integration of the mortar integral for the
upcoming examples will be performed therefore with nGPc = p + 2 Gauss-points
per contact element, where p is the Spline degree.
3.4. NUMERICAL RESULTS 49
3.4.2 Hertz 3D
In this section, the 3D Hertz problem of a sphere in contact with a rigid plate
and subjected to a point load F is considered from [46] for verification of the 3D
Solver implemented in Matlab.
Problem description
Figure 3.15 depicts the problem in a simplified form. Due to rotational sym-
metry, it is represented with one eights of the sphere with radius R, the Dirichlet
boundary conditions
u(1) = 0 on Γs,m
u,1
u(2) = 0 on Γs,m
u,2
u(3) = 0 on Γm
u,3
Analytical solution
50 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
3P
p0 = . (3.48)
2πa2
Both parameters (3.47) and (3.48) define the quadratic pressure distribution
r
r
p(r) = p0 1 − ( )2 (3.49)
a
p
along the radius r := x2 + y2 , with x = x(1) and y = x(2) .
Assumptions made in sec. 3.4.1 for the analytical solution of the 2D Hertz
problem apply also to the 3D case.
NURBS model
Figure 3.16 shows the NURBS geometries of the 3D Hertz contact problem
with the tensor product (TP)-meshes.
The slave structure is modeled with cubic splines and the master body is repre-
sented with quadratic splines. The sphere geometry is modeled with a inner radius,
to avoid distorted elements ([52]). Similar to the 2D model, the TP-meshes of both
3.4. NUMERICAL RESULTS 51
structures resolve the contact zone with a higher element density compared to rest
of the domains.
Results
(a) (b)
(c) (d)
(e) (f)
Figure 3.17: Hertz 3D pressure contour plots of slave structure with zoom on
contact zone (bottom view) for meshes with (a) 16848, (b) 44616, (c) 137592
and (d) 473304 DOF. (e) contact pressure curves for the different meshes from
cut in radial direction of rotation symmetric pressure distribution with analytical
solution. (f) oscillation error convergence w.r.t. number of DOF (see also 3.12)
3.5. SUMMARY 53
From successive h-refinement of the contact zone in radial direction, the error
of approximated maximum pressures is reduced to an acceptable limit. The cor-
responding results are depicted in the figures 3.17.
3.5 Summary
In this chapter, the isogeometric contact analysis of the frictionless problem in 2-
and 3D was presented based on NURBS.
First, the geometry modelling with NURBS, the basic idea of the isogeomet-
ric concept, its application to the linear elasticity part of the problem and classical
refinement strategies for NURBS based IGA were shown.
In the final part, Hertz contact problems were modelled and computed in 2D
54 CHAPTER 3. ISOGEOMETRIC CONTACT ANALYSIS WITH NURBS
and 3D for verification of the implemented NURBS based contact solution algo-
rithms. The summary from computational results is listed as follows:
• From both, h-refinement and degree elevation, the approximate solution was
improved.
• The error from numerical integration of the mortar integral did not decrease
significantly, when the number of integrations points was chose in parame-
ter direction grater than the degree of basis function plus one.
In the previous chapter, contact tractions computed from the NURBS based anal-
ysis showed non-physical oscillations in the vicinity of contact edges, where the
exact solution features a kink. Strategies for the elimination of this artefact, like
knot-relocation or uniform h-refinement, for a higher accuracy and a more robust
solution method, are too expensive and inefficient, especially for contact compu-
tations in 3D.
Solve
One of the first possibilities, investigated for the purpose of local refinement in
isogeometric setting were T-Splines presented in [6], [25] and [65]. Their appli-
cation in the context of contact computation was presented in [22], [23] and [24].
Other possibilities are for example LR splines ([44], [11]) or Hierarchical Splines
([33],[34],[50]) which have been applied to contact in [88] and [38]. In this work,
polynomial splines over hierarchical T-meshes (PHT-Splines) (see [21],[20]) are
applied for local refinement of the contact solution. They have been successfully
used for different applications of IGA in [84], [70], [57], [74] and [14].
Estimation
Based on error estimation, it is possible to mark and refine parts of the do-
55
56 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
main in which the approximated solution is to rough from a given mesh. In the
case of NURBS based contact analysis, these parts lie in vicinity of the contact
edge. While [88] employ the penetration status of elements for local refinement
respectively coarsening with LR-Splines, in this work, the classical tool of error
estimation is investigated.
Classically, two basic possibilities of error estimation, the residual based and
projection or defect correction method (rely on super-convergence properties),
are distinguished. The first technique has been used for elastic problems in solid
mechanics for example in [5] or [45] and the later in [86] and [87]. A review of
a posteriori error estimation techniques (also for elasticity problems) is given in
[69] or [1].
For contact problems in linear elasticity, error estimation techniques for clas-
sical FEA are presented e.g. in [48], [40], [81] and [80].
Error estimation techniques in IGA are not as mature as in classical FEM. In
this work, the a posteriori error estimation technique from [3] based on recov-
ery solutions and super-convergence theory for PHT-Splines is employed (Matlab
code from ISM, Bauhaus University Weimar).
Marking
Error estimation techniques are not the only important building blocks in the
construction of adaptive refinement procedures. Also the marking strategy plays
a significant role in finding the optimal mesh (mesh with as few elements as pos-
sible) with respect to a given error tolerance.
Refining for example only few elements in each refinement step yields on the
one hand an optimal mesh, on the other hand computational costs increase from
solving the non-linear contact problem in each iteration. Also, the presence of
singularities in case of coarse meshes results in errors some distance away which
lead to unnecessary refinement. These so-called pollution errors however become
smaller as the area around the singularity is refined.
With these things in mind, one simple marking strategy is to select those el-
ements in which the estimated error exceeds some absolute threshold (see [34]).
This strategy though, does not consider the overall distribution of the error. There-
fore, the Dörfler mariking strategy (see [26]) is preferably used in this work.
Scope
introduced with the concept of local knot-vectors and the truncation of basis func-
tions.
The third part focuses on the steps error estimation and the marking strategy
of the overall adaptive procedure.
In the final part effectiveness of the developed procedure for the solution of
contact problems in 2D and 3D is shown. For this purpose, the numerical exam-
ples form section 3.4 are computed and the results from PHT-Spline and NURBS
discretization are compared.
PHT splines result from the Bezier representation of NURBS (see e.g. [10]).
In this transformation process, the underlying tensor product mesh structure of
NURBS is broken up based on Bezier decomposition. The resulting C0 -elements
make PHT-splines capable for local refinement, similar to classical Finite Element
Analysis. One single element from the T-mesh is refined successively on different
hierarchy levels. New basis functions with local support are added to the element
basis by the cross-insertion procedure. However, priorly the existing basis needs
to be truncated for global linear independence of the refined basis and a sparse
system matrix for efficiency.
In the following subsections, PHT-Splines with their capability for local re-
finement are described in more detail mainly from [3].
58 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
In the refinement step, the bottom right element Ek=4,0 ∈ T0 is then subdi-
vided into four children elements E j1 ,l+1 , E j2 ,l+1 , E j3 ,l+1 and E j4 ,l+1 . This is done
by adding another level l = l + 1 in the hierarchy tree and by connecting the mid-
dle points of the opposite edges of Ek=4,0 by straight lines. Refined elements are
represented in figure 4.1 as filled leaves of the hierarchy tree.
with α = 1, · · · , d and control points P = {PA }nA=1 with n = ∏dα=1 nα (see also
sec. 3.1). For the Bèzier representation of NURBS, their tensor product meshes T
are decomposed into Bèzier elements.
Re = Ce Be . (4.5)
Local knot-vectors
can be defined and associated to these basis functions. It is important that all other
functions and their derivatives vanish at ξi .
When this concept is extended to bi-cubic Splines, four basis functions are as-
(1) (2) (1) (1) (2) (2)
sociated to the knot [ξi1 , ξi2 ]. They have support on [ξi1 −1 , ξi1 +1 ] × [ξi2 , ξi2 ].
Function values, first order and mixed partial derivatives of other basis functions
(1) (2)
vanish at [ξi1 , ξi2 ]. For the general case of Cα,α -continuous bivariate Splines,
the multiplicity must be p − α for interior knots and p + 1 for boundary knots (as
for open knot vectors).
Truncation
Practically, this step is done by zero out certain Bézier ordinates of the chil-
dren elements. Basis functions are modified this way in regions close to the new
vertexes resulting from the subdivision, i.e. the support is reduced in the corners
of children elements. Basis functions, which could be linear dependent from the
new basis functions, are deactivated by zero out their Bézier ordinates completely.
From this step, the original basis {RkA (ξξ )}N
A=1 is transformed to the truncated basis
The second part of the refined basis are the (α + 1)d basis functions
N+(α+1)d
{R̃Bk+1 (ξξ )}B=1
62 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
k+1
associate to the newly inserted knot ξ˜ i for the case p = 2α + 1. In a first step
these are standard non-truncated B-spline basis functions, computed from local
knot vector information. In a second step, corresponding Bézier ordinates must
be determined and global basis indexes must be assigned. The latter is done by
using the index from basis-functions which were removed during truncation or
incrementally new indexes are assigned.
N+(α+1)d
Both, {R̄Ak+1 (ξξ )}N k+1 ξ
A=1 and {R̃B (ξ )}B=1 form the basis of a refined ele-
ment.
It is important to note here that different from NURBS or B-Splines over uni-
form meshes, PHT-Splines have a variable number of basis functions with support
over given refined elements. Despite of the truncation step, this can lead to the sit-
uation that basis functions are locally linear dependent. Globally though, the basis
remains linear independent (see [3]).
l,k=0
Single points xl,k=0 from elements Γc,e of initial hierarchy level k = 0 are
described by
N
s−1 (1) (2)
xs,k=0 (ξ (1) ) = ∑ Bi (ξˆ )C−,0
e ◦ F̂e (ξ , ξ0 ) P−,0 (4.7)
i=1 | {z } i
φi− (ξ (1) )
and
M
−1 (1)
xm,k=0 (ζ (2) ) = ∑ |B j (ζˆ )Ce+,0 ◦ F̂{zme (ζm1 , ζ (2) ) P+,0 (4.8)
j=1 } j
φ+
j (ζ
(2) )
for slave and master side. Herein Bi (ξˆ ) and B j (ζˆ ) are the univariate Bernstein
basis functions with ξˆ , ζˆ ∈ [−1, 1], C−,0
e and Ce+,0 are the reduced element ex-
traction operators and P−,0i and P+,0
j are control-points corresponding to the N
64 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
l,0
respectively M non-zero basis functions in Γc,e . The functions F̂l linearly map the
corresponding Bernstein basis from the reference element [−1, 1] to an element
l,0
Γc,e .
l,k+1
Physical points xl,k+1 from a refined contact or a child element Γc,e are
defined by
N N+(α s +1)d −1
xs,k+1
(ξ (1)
) = ∑ Bi (ξ (1)
)C̄e−,k+1 P−
i + ∑ B ∗ (ξ (1) )C̃e−,k+1 P̃−
∗ (4.9)
i=1
| {z } ∗
i =N+1
|i {z } i
φ̄i− (ξ (1) ) φ̃i−
∗ (ξ
(1) )
and
M M+(α m +1)d −1
(2) (2)
x m,k+1
(ζ )= ∑ B| j (ζ )C̄+,k+1
e P+j + ∑ B j∗ (ζ (2) )C̃e+,k+1 P̃+j∗
j=1 {z } ∗
j =M+1 | {z }
φ̄ +
j (ζ
(2) ) φ̃ +
j∗
(ζ (2) )
(4.10)
for slave and master. In (4.9) and (4.10) the truncated contact element Bezier op-
erator are denoted with C̄e−,k+1 and C̄e+,k+1 , the control-points and corresponding
l,k+1 −,k+1
Bezier operator of new non-zero basis functions in Γc,e are P̃− +
i∗ , P̃ j∗ , C̃e and
C̃e+,k+1 for slave and master.
N M
us,k
e =∑ φi− (ξ )q−
i and um,k
e = ∑ φ +j (ζ¯ )q+j , (4.11)
i=1 j=1
N M
δ us,k − −
e = ∑ φi (ξ )δ qi and δ um,k
e = ∑ φ +j (ζ¯ )δ q+j (4.12)
i=1 j=1
and
N M
−,k
Xs,k
e =∑ φi− (ξ )Pi,0 and Xem,k = ∑ φ +j (ζ¯ )P0,+,kj . (4.13)
i=1 j=1
In (4.11), (4.12) and (4.13) the definitions ξ (1) := ξ and ζ (2) := ζ are inserted
and evaluation of the quantities at CPP on master side is indicated by ζ¯ .
4.2. CONTACT COMPUTATION WITH PHT-SPLINES 65
l,k+1
In the same way contact quantities are represented in a refined element Γc,e
by
N N+(α s +1)d
us,k+1
e =∑ φ̄i− (ξ )q−
i + ∑ φ̃i−∗ (ξ )q̃−
i , (4.14)
i=1 ∗
i =N+1
M M+(α m +1)d
um,k+1 = φ̄ + ¯ + φ̃ + ¯ +
e ∑ j (ζ )q j + ∑ j∗ (ζ )q̃ j , (4.15)
j=1 ∗
j =M+1
N N+(α s +1)d
δ us,k+1
e =∑ φ̄i− (ξ )δ q−
i + ∑ φ̃i−∗ (ξ )δ q̃−
i∗ , (4.16)
i=1 i∗ =N+1
M M+(α m +1)d
δ um,k+1 = φ̄ + ¯ +
φ̃ + ¯ +
e ∑ j (ζ )δ q j + ∑ j∗ (ζ )δ q̃ j∗ , (4.17)
j=1 ∗
j =M+1
N N+(α s +1)d
Xes,k+1 =∑ φ̄i− (ξ )P−
i,0 + ∑ φ̃i−∗ (ξ )P̃−
i∗ ,0 , (4.18)
i=1 ∗
i =N+1
and
M M+(α m +1)d
Xem,k+1 = φ̄ + ¯ + φ̃ + ¯ +
∑ j (ζ )P j,0 + ∑ j∗ (ζ )P̃ j∗ ,0 . (4.19)
j=1 ∗
j =M+1
The numerical degrees of freedom and control-points in initial configuration which
are added from refinement of contact element Γl,k −
c,e are denoted here with q̃i∗ ,
q̃+j∗ , δ q̃− + − + ∗ s
i∗ , δ q̃ j∗ , P̃i∗ ,0 , and P̃ j∗ ,0 , with i = N + 1, · · · , (α + 1)
d−1 and j ∗ = M +
1, · · · , (α m + 1)d−1 .
The discrete forms given in (4.11), (4.12) and 4.13 defined on Γl,k c,e are now
inserted into the variational contact potential and sorted into matrices, similar to
(3.25). Then, the discrete contact element potential defined on hierarchy level k is
given by
Z
h,k
δ Πc,e = s,k
(δ qkc,e )T (Cke )T n̄n̄T Cke qkc,e − · · ·
Γc,e
Figure 4.3 shows the mortar integration principle applied to PHT-Spline struc-
tures in 2D.
to master side. Based on the quadrature rule given in (3.34), the contact element
integrals for different hierarchy levels (see (4.20) and (4.21)) are transformed to
systems of algebraic equations (see (3.35)).
The overall set of equations (see (3.36)), representing the linearized contact
part, are then assembled from elements e = 1, · · · , n∗c in penetration. Note, ele-
ments can be from different hierarchy levels dependent from the refinement status.
For practically implementation, the hierarchical T-mesh structure shown in figure
4.1, is used to organize element data (see also [62]).
Global solution
defined through x∗i,k . Here, the polynomials ϕ ∗j,k are B-Splines of degree
p∗ ≥ p and continuity α ∗ ≥ α. The corresponding numerical degrees of
freedom are denoted with c∗j,k .
4.3. ADAPTIVE CONTACT FORMULATION 69
4.3.2 Marking
Once, the error in stresses E(σ σ h , Ω) is determined based on the steps described
in the previous section, a strategy of marking elements for local refinement is re-
quired in the overall adaptive contact procedure. Here, it is important to find the
optimal trade-off between the number of refinement steps (required to reach cer-
tain estimated accuracy) and the number of elements in the final mesh.
Similar to [3] or [12] the Dörfler marking strategy is employed in this work,
in which elements with the largest contribution to the total error are selected for
refinement.
In other words, for domain Ωh and the approximate solution σ h a subset M ⊂
h
Ω of elements is marked with the condition
σ h , ΩhM ) ≤ θ E(σ
E(σ σ h , Ωh ). (4.26)
In (4.26) parts of the domain marked for refinement are denoted with ΩhM and
the parameter θ ∈ (0, 1] is constant value. When θ = 1 each element from Ωh
is refined, while θ < 1 means less refinement per step. In practice, a reasonable
compromise between the number of refinement steps and the optimality of meshes
is found when
• θ = 0.75 for problems with smooth solutions and
• θ = 0.5 for problems with singularities.
70 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
Figure 4.4: Overall solution procedure with adaptive local refinement for the con-
tact problem (2D and 3D)
In this procedure, coarse T-meshes Tlk=0 are defined for slave and master with
the hierarchy level k = 0. The effect of this initial discretization on the process of
adaptive local refinement is investigated in numerical examples presented in sec.
4.4. The parameter θ is set to 0.5 (see sec. 4.3.2) due to the weak discontinuity
of stress solutions at the end of contact zone and limits εs , εm > 0 are set for the
relative errors defined in (4.27).
4.4. NUMERICAL RESULTS 71
In the next step, the contact problem is solved for Tkl based on the procedure
given in Algorithm 1.
σ hl , Ω hl )
Form the resulting approximated stresses σ hl , single error values E(σ
are computed numerically for Ωhl based on (4.22). For the decision of continuing
the process of local refinement, relative errors defined by
σ hl , Ω hl )
E(σ
erel
l = (4.27)
σ ∗l ||
||σ
are determined. Herein ||σl∗ || is the energy norm of the recovered stresses (see
sec. 4.3.1).
In this next step, those parts of the domains ΩhMl ⊂ Ωhl with the elements
Ml are marked for refinement which fulfil the inequality given in (4.26). These
elements are then refined with cross-insertion which adds an additional hierarchy
level k = k + 1 to the T-meshes, Tkl := Tlk+1 . With the newly refined meshes, the
procedure starts anew from solving the contact problem.
4.4.1 Hertz 2D
Starting with Hertz 2D problem, initial PHT-Spline parametrizations with the T-
meshes Tlk=0 for slave and master are depicted in figure 4.5.
72 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
Figure 4.5: Initial meshes T0s and T0m of slave and master, Hertz 2D
These meshes have a uniform elements distribution. i.e. contact zones are not
considered with higher resolution. Both Spline geometries are parametrized with
bi-cubic basis functions and the penalty parameter εN is chosen constant through-
out the refinement process. The condition for optimal meshes in this example is
given by
−3 −2
erel
s ≤ 2 ∗ 10 ∧ erel
m ≤ 5 ∗ 10 .
Results
Figure 4.6 shows the errors in maximum pressures from each refinement step
k computed by
k p0 − max(ph,k
N (x))
e p0 = (4.28)
p0
and the errors from pressure oscillations in the vicinity of contact radius a defined
by (3.45) with respect to the number of degrees of freedom for slave and master
side. In (4.28) the analytical solution for the maximum pressure is denoted with
p0 .
4.4. NUMERICAL RESULTS 73
100
10-1
10-2
10-3
10-4
10-5
0 2000 4000 6000 8000 10000
(a)
0
10
-1
10
10-2
10-3
0 2000 4000 6000 8000 10000
(b)
Figure 4.6: Convergence of relative errors in maximum pressures (blue curve) and
from pressure oscillations (brown curve) w.r.t. the number of DOF for (a) slave
and (b) master side
These results show that both errors correlate and they significantly drop al-
ready after the first few refinement steps for slave and maser side. While on the
slave side both errors converge monotonously (except from some oscillation in
the beginning) to a certain value, the error in maximum pressure even increases
on the master side during the adaptive process before it also converges to a fixed
value. Successive refinement steps, in which errors do not decrease (horizontal
line), other parts than the contact zone are refined (see figure 4.8). In this exam-
ple, elements from the curved Neumann boundary Γsσ are also refined.
74 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
For plausibility check of the results in figure 4.6, approximated contact pres-
h,l
sures with analytical solution, PHT-meshes and computed stresses σyy /p0 in con-
tour plots from refinement steps in which a significant drop of oscillation error
occurred, are shown in the figures 4.7, 4.8 and 4.9 for slave and master.
1.2
0.8
0.6 0.2
0.1
0.4
0
0.2 -0.1
1 1.5 2
0
-0.2
-0.4
0 0.5 1 1.5 2
(a)
1.2
0.8
0.6 0.2
0.1
0.4
0
0.2 -0.1
1 1.5 2
0
-0.2
-0.4
0 0.5 1 1.5 2
(b)
0.2 0.2
0.1 0.1
0 0
(a) (b)
0.2 0.2
0.1 0.1
0 0
(c) (d)
(a) (b)
(c) (d)
Figure 4.9: Approximated stresses σyy (x, y)/p0 and hierarchical T-meshes from
adaptive refinement process of Hertz 2D problem with (a) #DOF = 2176, (b)
#DOF = 2368, (c) #DOF = 2848 and (d) #DOF = 3680.
4.4. NUMERICAL RESULTS 77
Figure 4.6 shows, how the oscillatory behaviour of the computed contact
stresses reduces with each refinement step. Especially the solution with 3680
degrees of freedom (green dashed line) of slave side shows nearly a perfect match
with analytical solution, also around x/a = 1. Despite the coarse meshes of master
side compared to slave side, the approximations show a good quality which also
improves with the ongoing local refinement.
The corresponding T-meshes to the solutions in figure 4.8 show the effective
hierarchical mesh-refinement at the end of contact zone which results the approx-
imation qualities presented before.
From the stress distributions σyy (x, y) normalized with p0 shown together with
the T-meshes in figure 4.9, it is obvious that refinement along the contact edge
alone is not enough, to improve the approximation quality of normal stresses. It
shows, how the gradual change of the stress field resulting from contact must be
captured also with the appropriate element distributions.
10-3
-4
10
-5
10
0 0.5 1 1.5 2
4
10
Conclusions
• Special care has to be taken, when traction boundary conditions are applied
to avoid unnecessary refinement
4.4.2 Hertz 3D
Now, the adaptive refinement procedure is tested for the 3D case. For this purpose,
initial meshes Tlk=0 , l ∈ {s, m} of slave and master from the 3D Hertz problem
description in sec.3.4.2 are defined (see figure 4.11).
Figure 4.11: Initial meshes T0l with l ∈ {s, m} of slave and master for the adaptive
solution of Hertz 3D problem
The chosen PHT-Splines are cubic in all three parameter directions for sphere
and cube. The meshes here are defined with a finer resolution of the contact zone
compared to rest of the domains.
Results
Figure 4.12 depicts the errors in maximum pressures and from oscillatory be-
havior computed from the successive refinement steps on the slave side with the
corresponding number of degrees of freedom.
4.4. NUMERICAL RESULTS 79
10-1
error from oscillations
error in maximum pressure
10-2
10-3
10-4
0 0.5 1 1.5 2 2.5
105
The errors are computed with (4.28) and (3.45) from 2D curves which yield
from the rotation symmetric solutions of the pressure distributions (see figure 4.14).
While the oscillation error decreases with each refinement step monotonously, the
error in maximum pressures decreases only in the first 4 steps and stagnates in
the subsequent steps. A possible reason for the instable decrease of the maximum
pressure error is the constant penalty parameter throughout the refinement process.
(a) (b)
(c) (d)
From the first plot (a) in figure 4.13 it can be seen that also the 3D solution
possesses oscillatory behavior in vicinity of the circular contact edge. This artifi-
cial property if the solution improves with each refinement step together with the
error in maximum pressures.
0.8
0.2
0.6 0.1
0
0.4
-0.1
Figure 4.14: Pressure curves pN (r) from refinement steps of the adaptive solution
process of Hertz 3D problem
The closer look at PHT-meshes presented in figure 4.15 with pressure distribu-
tion contour plots of the contact zone (bottom view of slave side) from the solution
steps discussed above verifies additionally the intended local refinement along the
circular contact edge.
82 CHAPTER 4. ADAPTIVE ISOGEOMETRIC CONTACT
(a) (b)
(c) (d)
Figure 4.15: Pressure contour plots with corresponding T-meshes of slave struc-
ture with zoom on contact zone (bottom view) from refinement steps of the adap-
tive solution process of Hertz 3D with (a) #DOF = 25272, (b) #DOF = 36360,
(c) #DOF = 82296 and (d) #DOF = 123240.
The overall T-meshes of slave and master in the chosen refinement steps are
shown in figure 4.16.
4.4. NUMERICAL RESULTS 83
(a) (b)
(c) (d)
Figure 4.16: Overall T-meshes of slave and master from refinement steps of
the adaptive solution process of Hertz 3D problem with (a) #DOF = 25272, (b)
#DOF = 36360, (c) #DOF = 82296 and (d) #DOF = 123240
Here, like in the 2D Hertz example results, it is can be seen that the refine-
ment gradually increases to the contact zone for slave and master to decrease the
errors there. Due to the applied traction forces to the curved surface with an inner
radius ri of the slave structure, the adaptive process also identifies relevant errors
in stresses there. This yield additional unintentional refinement.
4.5 Summary
In this chapter, the developed adaptive procedure for contact analysis in the iso-
geometric setting was presented. It is based on the parameter description of the
contacting geometries with PHT-Splines which are capable of an efficient local
refinement. The adaptivity in this approach is achieved by the combination of a
super-convergent recovery-based error estimation procedure and the Dörfler mark-
ing strategy. The theoretical part of this chapter was closed with explanation of
the overall solution strategy. Conclusions from numerical examples of the Hertz
contact are:
Industrial application
Figure 5.1: CAD geometry cross section of high pressure fuel pump CP4 from
Bosch
Functional principle
Fuel enters the pump at the inlet valve. The pump-plunger compresses with
vertical movements fuel in the compression chamber until a required pressure is
reached. The compressed fuel leaves then the pump from the outlet valve.
87
88 CHAPTER 5. INDUSTRIAL APPLICATION
The vertical movement of the plunger is realized from a driving train which
consists of the driving shaft with cam, the roller, the roller shoe, the plunger spring
and the pump plunger itself. Rotational movement (external drive) of the shaft is
transformed through cam-geometry into translational movement of the plunger
which is supported by the roller and roller-shoe.
The highly loaded mechanical contact between roller and cam plays an impor-
tant role in designing a robust pump of high quality. Among different parameters
which describe the large sliding contact problem, the geometry is an important
degree of freedom for an optimal design of the mechanical contact with respect to
minimal stresses.
In this chapter, roller and cam are therefore modelled with Spline functions
and contact stresses are computed with the developed isogeometric procedures
(NURBS based and adaptive PHT-Spline based) to show and test their practica-
bility in industrial environment.
The cam geometry is represented locally as a plate, due to its much larger
radius compared to the roller.
5.1. ROLLER CAM CONTACT PROBLEM 89
The roller geometry however, which is the degree of freedom for an optimal
contact design, is modeled as a whole, in a first step. In a second step, only one
eight of the roller volume is considered due to rotational and planes (x(1) , x(2) -
plane and x(2) , x(3) -plane) symmetries.
Necessary Dirichlet boundary conditions to the problem in figure 5.3 are sum-
90 CHAPTER 5. INDUSTRIAL APPLICATION
marized by
u(1) = 0 on Γsu,2 , Γm m
u,2 and Γu,3 (5.1)
u(2) = 0 on Γsu,1 , Γm m
u,1 and Γu,3 (5.2)
(3)
u =0 on Γm
u,3 (5.3)
The mechanical load which is transmitted from plunger to roller is model as
a point load F acting on the top edge of the free roller (see figure 5.2). It is
transformed to a surface traction t̂ on Γsσ with the condition:
F
Z
= t̂ dΓ. (5.4)
4 Γsσ
The roller material is modelled linear elastic, while the cam is considered rigid.
(a) (b)
Both bodies are described with cubic Splines. The roller geometry is modeled
with a small inner radius to avoid distorted elements.
Results
Figure 5.5 shows contour plots of resulting contact pressures pN (x, y) with TP-
meshes of the contact zone from the initial mesh (shown in figure 5.4) and from
meshes of three successive h-refinement steps of the slave side (roller).
5.2. NURBS BASED SOLUTION 91
(a) (b)
(c) (d)
Figure 5.5: Contact pressure contour plots with corresponding NURBS TP-
meshes of roller cam contact 3D problem from (a) mesh 1, (b) mesh 2, (c) mesh 3
and (d) mesh 4 (#DOF = 515160)
Due to the missing analytical solution to this problem, the solution from mesh
4 is assumed to be the "pseudo analytical solution" of the roller cam contact prob-
lem. From this reference solution, the maximum pressure value p0 := max(pN (x, y))
and the contact zone parameter ax and ay with pN (ax , y = 0) = 0 and pN (x =
0, ay ) = 0 are used for the normalized representation of the results with pN /p0 ,
x/ax and y/ay .
(a) (b)
(c) (d)
Figure 5.6: 3D contact pressure plots from NURBS based computation of roller
cam contact problem with (a) mesh 1, (b) mesh 2, (c) mesh 3 and (d) mesh 4
(#DOF = 515160)
Here, single solutions are assembled to the solution of the original problem
defined in figure 5.2. From these plots, it can be seen already that the solutions
also show the oscillatory behavior at the contact edge.
For a better assessment of this artifact and its correlation with the maximum
pressures, the 2D curves pN (x, y = 0) and pN (x, y = 0) are compared in figure 5.7
for the different meshes.
5.3. ADAPTIVE PHT-SPLINE BASED SOLUTION 93
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2 0.2
0 0
-0.2 -0.2
-1.5 -1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1
(a) (b)
0.2
1
0.15
0.95
0.1
0.05
0.9
0
-0.05 0.85
-0.1
0.8
-0.15
-0.2 0.75
0.95 1 1.05 1.1 1.15 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
(c) (d)
Figure 5.7: Comparison of 2D contact pressure curves from NURBS based com-
putation of roller cam contact problem: (a) pN (x, y = 0), (b) pN (x = 0, y) , (c)
enlarged view of pN (y) at contact edge and (d) enlarged view of maximum pres-
sures max(pN (y))
(a) (b)
(c) (d)
Figure 5.8: Contact pressure contour plots with corresponding T-meshes (slave
side) of the 3D roller cam contact problem from adaptive refinement process steps:
(a) #DOF = 60840, (b) #DOF = 75960, (c) #DOF = 95400 and (d) #DOF =
314040
These plots show the successful local refinement along the contact edge in the
sequential steps.
(a) (b)
(c) (d)
Figure 5.9: 3D contact pressure plots (slave side) of the 3D roller cam contact
problem from adaptive refinement process steps: (a) #DOF = 60840, (b) #DOF =
75960, (c) #DOF = 95400 and (d) #DOF = 314040
Here, it can be seen that the wavy character of pressures around the contact
edge reduces with each shown refinement step.
This impression becomes solid, when the 2D pressure curves are derived in
analogy to sec. 5.2 and compared for the different refinement steps in figure 5.10.
96 CHAPTER 5. INDUSTRIAL APPLICATION
1
1
0.8
0.8
0.6
0.6
0.4
0.4 # DOF = 60840
# DOF = 60840
# DOF = 75960 # DOF = 75960
0.2 # DOF = 95400 # DOF = 95400
0.2 # DOF = 314040
# DOF = 314040
0 0
-0.2 -0.2
-1.5 -1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1
(a) (b)
0.4 0.2
0.1
-0.05
-0.1
0
-0.15
-0.1 -0.2
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 0.95 1 1.05 1.1 1.15
(c) (d)
Figure 5.10: Comparison of contact pressure curves (slave side) derived from the
3D pressures in figure reffig: Roller Cam, PHT, 3D Pressure distributions: (a)
pN (x, y = 0)/p0 , (b) pN (x = 0, y)/p0 , (c) enlarged view of pN (x, y = 0)/p0 close
to contact edge and (d) enlarged view of pN (x = 0, y)/p0 close to contact edge
It is also noted that the number of numerical DOF remains far below the
needed number for the NUBBS based computation of the reference solution.
Chapter 6
Conclusions
6.1 Summary
Spline based contact formulations have been shown to be advantageous compared
to classical finite element analysis with Lagrange polynomials in recent publica-
tions on computational contact mechanics. Due to higher inter-element continuity
of spline-based discretization, the contact search is more robust and from the ex-
act description of the contact boundary, the accuracy of solutions improves. Also,
mesh-refinement of contact does not require an additional communication with
the geometry models.
This thesis addresses the open question about effective refinement procedures
within the isogeometric setting to improve locally the approximate contact so-
lution in an efficient way. Although, different Spline structures (T-Splines, LR-
Splines, etc.) for local refinement have been applied to contact in the isogeomet-
ric setting, in this work PHT-Splines were employed the first time together with
the penalty based mortar method to improve especially the solution in vicinity
of the contact edge. Here, Spline approximated solutions feature an oscillatory
behaviour due to their inability to capture C0 -continuity of the analytical solu-
tion. PHT-Splines break up the tensor product structure of NURBS with Bèzier
extraction and local refinement is possible on a hierarchical T-mesh structure with
cross-insertions.
97
98 CHAPTER 6. CONCLUSIONS
Numerical results from the examples of Hertz contact problems and the roller
cam contact problem showed that the developed adaptive procedure possesses an
increased efficiency and effectiveness in solving the contact problem in 2D and
3D compared to the NURBS based contact solution approach (also implemented
in this work). The error in maximum pressures and the error from pressure oscilla-
tions improved with significantly less numerical degrees of freedom with adaptive
local refinement when compared to uniform h-refinement of the contact zone in
the NURBS based solution. This result was possible due to the effective identifi-
cation of elements resolving the contact end with the applied error estimator. By
this approach, an expensive local contact search within each refinement step has
been avoided.
At last, a Matlab tool was developed for the optimization of contacting ge-
ometries of roller and cam from a high-pressure fuel pump with respect to min-
imal stresses. Here, the implemented NURBS based contact solution procedure
and the developed PHT-Spline based adaptive approach were used with a Spline-
geometry modelling tool for roller and cam. The geometry tool (the subject of a
master thesis instructed during this work) generates an analysis suitable geometry
for a given set of geometrical parameters (angles, radii, length, etc.).
6.2 Outlook
The PHT-Spline based adaptive procedure was developed and tested only for the
frictionless contact problem with linear elastic structures. An application of it
to frictional large sliding contact problems will certainly increase its field of ap-
plication. From numerical point of view, it is also interesting to investigate the
procedures performance when additional non-linearities (geometrical, material)
are included to the problems description.
The penalty approach used in this work for the incorporation of contact condi-
tions might not be the best choice, despite the simple implementation and the low
computational costs, within the adaptive procedure. It was shown in this work that
the accuracy of Spline approximated contact solutions and robustness of the New-
ton method depend on the combination of chosen mesh and penalty parameter.
6.2. OUTLOOK 99
Geometries from the numerical examples presented in this work were mod-
elled with single patches only. The application of multi-patch geometry discretiza-
tion, especially for the simulation of large sliding contacts, within the presented
adaptive contact solution approach would be of great practical use.
100 CHAPTER 6. CONCLUSIONS
Bibliography
[1] Mark Ainsworth and J Tinsley Oden. A posteriori error estimation in finite
element analysis, volume 37. John Wiley & Sons, 2011.
[4] Andreas Apostolatos, Robert Schmidt, Roland Wüchner, and Kai-Uwe Blet-
zinger. A nitsche-type formulation and comparison of the most common do-
main decomposition methods in isogeometric analysis. International Jour-
nal for Numerical Methods in Engineering, 97(7):473–504, 2014.
[5] I Babuvška and Werner C Rheinboldt. Error estimates for adaptive finite
element computations. SIAM Journal on Numerical Analysis, 15(4):736–
754, 1978.
[7] Gernot Beer, Stéphane Bordas, et al. Isogeometric methods for numerical
simulation, volume 240. Springer, 2015.
[8] Ted Belytschko, Wing Kam Liu, Brian Moran, and Khalil Elkhodary. Non-
linear finite elements for continua and structures. John wiley & sons, 2013.
101
102 BIBLIOGRAPHY
[10] Michael J Borden, Michael A Scott, John A Evans, and Thomas JR Hughes.
Isogeometric finite element data structures based on bézier extraction of
nurbs. International Journal for Numerical Methods in Engineering, 87(1-
5):15–47, 2011.
[13] C. Carstensen, O. Scherf, and P. Wriggers. Adaptive finite elements for elas-
tic bodies in contact. SIAM Journal on Scientific Computing, 20(5):1605–
1626, 1999. doi: 10.1137/S1064827595295350. URL https://doi.org/
10.1137/S1064827595295350.
[14] Chiu Ling Chan, Cosmin Anitescu, and Timon Rabczuk. Volumetric
parametrization from a level set boundary representation with pht-splines.
Computer-Aided Design, 82:29–41, 2017.
[16] J Austin Cottrell, Thomas JR Hughes, and Yuri Bazilevs. Isogeometric anal-
ysis: toward integration of CAD and FEA. John Wiley & Sons, 2009.
[19] Laura De Lorenzis, Peter Wriggers, and Thomas J.R. Hughes. Isogeometric
contact: a review, 2014-07. ISSN 0936-7195.
BIBLIOGRAPHY 103
[20] Jiansong Deng, Falai Chen, and Yuyu Feng. Dimensions of spline spaces
over t-meshes. Journal of Computational and Applied Mathematics, 194(2):
267–283, 2006.
[21] Jiansong Deng, Falai Chen, Xin Li, Changqi Hu, Weihua Tong, Zhouwang
Yang, and Yuyu Feng. Polynomial splines over hierarchical t-meshes.
Graphical Models, 70(4):76 – 86, 2008. ISSN 1524-0703. doi: https://
doi.org/10.1016/j.gmod.2008.03.001. URL http://www.sciencedirect.
com/science/article/pii/S1524070308000039.
[24] Rossana Dimitri and Giorgio Zavarise. T-splines discretizations for large
deformation contact problems. PAMM, 15(1):183–184, 2015.
[25] Michael R Dörfel, Bert Jüttler, and Bernd Simeon. Adaptive isogeometric
analysis by local h-refinement with t-splines. Computer methods in applied
mechanics and engineering, 199(5-8):264–275, 2010.
[28] Nagi El-Abbasi and Klaus-Jürgen Bathe. Stability and patch test perfor-
mance of contact discretizations and a new solution algorithm. Computers
& Structures, 79(16):1473–1486, 2001.
[31] Kathrin A. Fischer and Peter Wriggers. Mortar based frictional contact
formulation for higher order interpolations using the moving friction cone.
Computer Methods in Applied Mechanics and Engineering, 2006. ISSN
00457825. doi: 10.1016/j.cma.2005.09.025.
[33] Carlotta Giannelli, Bert Jüttler, and Hendrik Speleers. THB-splines: The
truncated basis for hierarchical splines. In Computer Aided Geometric De-
sign, 2012. ISBN 0167-8396. doi: 10.1016/j.cagd.2012.03.025.
[35] Dietmar Gross, Werner Hauger, and Peter Wriggers. Hydromechanik, Ele-
mente der Höheren Mechanik, Numerische Methoden. Springer, 2002.
[36] Corinna Hager, Patrice Hauret, Patrick Le Tallec, and Barbara I Wohlmuth.
Solving dynamic contact problems with local refinement in space and time.
Computer Methods in Applied Mechanics and Engineering, 201:25–41,
2012.
[37] Paul Hennig, Markus Kästner, Philipp Morgenstern, and Daniel Peterseim.
Adaptive mesh refinement strategies in isogeometric analysis - a computa-
tional comparison. Computer Methods in Applied Mechanics and Engineer-
ing, 316:424–448, 2017.
[39] Christian Hesch and Peter Betsch. Isogeometric analysis and domain de-
composition methods. Computer Methods in Applied Mechanics and En-
gineering, 213-216:104 – 112, 2012. ISSN 0045-7825. doi: https://doi.
org/10.1016/j.cma.2011.12.003. URL http://www.sciencedirect.com/
science/article/pii/S0045782511003823.
[40] Stefan Hüeber and Barbara I Wohlmuth. An optimal a priori error esti-
mate for nonlinear multibody contact problems. SIAM Journal on Numerical
Analysis, 43(1):156–173, 2005.
[43] Fish Jacob and Belytschko Ted. A first course in finite elements. Wiley,
2007.
[44] Kjetil André Johannessen, Trond Kvamsdal, and Tor Dokken. Isogeometric
analysis using LR B-splines. Computer Methods in Applied Mechanics and
Engineering, 2014. ISSN 00457825. doi: 10.1016/j.cma.2013.09.014.
[45] Claes Johnson and Peter Hansbo. Adaptive finite element methods in com-
putational mechanics. Computer methods in applied mechanics and engi-
neering, 101(1-3):143–181, 1992.
[47] Reinhold Kienzler and Roland Schröder. Einführung in die höhere Fes-
tigkeitslehre. Springer, 2009.
[48] Noboru Kikuchi and John Tinsley Oden. Contact problems in elasticity: a
study of variational inequalities and finite element methods. SIAM, 1988.
[49] Ji Yeon Kim and Sung Kie Youn. Isogeometric contact analysis using mortar
method. International Journal for Numerical Methods in Engineering, 2012.
ISSN 00295981. doi: 10.1002/nme.3300.
106 BIBLIOGRAPHY
[50] Gábor Kiss, Carlotta Giannelli, and Bert Jüttler. Algorithms and data struc-
tures for truncated hierarchical b–splines. In International Conference on
Mathematical Methods for Curves and Surfaces, pages 304–323. Springer,
2012.
[53] Lei Liu, Yongjie Zhang, Thomas J. R. Hughes, Michael A. Scott, and
Thomas W. Sederberg. Volumetric t-spline construction using boolean oper-
ations. Engineering with Computers, 30(4):425–439, October 2014. ISSN
1435-5663. URL https://doi.org/10.1007/s00366-013-0346-6.
[54] Jia Lu. Isogeometric contact analysis: Geometric basis and formulation for
frictionless contact. Computer Methods in Applied Mechanics and Engi-
neering, 2011. ISSN 00457825. doi: 10.1016/j.cma.2010.10.001.
[55] Jia Lu and Chao Zheng. Dynamic cloth simulation by isogeometric analysis.
Computer Methods in Applied Mechanics and Engineering, 2014. ISSN
00457825. doi: 10.1016/j.cma.2013.09.016.
[60] Les A. Piegl and Wayne Tiller. The NURBS book. Monographs in
visual communications. Springer, Berlin ; Heidelberg [u.a.], 2. ed. edi-
tion, 1997. ISBN 3-540-61545-8. URL http://swbplus.bsz-bw.de/
bsz055729541cov.htm. UB Vaihingen.
[61] Martin Ruess, Dominik Schillinger, Ali I. Özcan, and Ernst Rank. Weak cou-
pling for isogeometric analysis of non-matching and trimmed multi-patch
geometries. Computer Methods in Applied Mechanics and Engineering,
2014. ISSN 00457825. doi: 10.1016/j.cma.2013.10.009.
[65] Michael A Scott, Xin Li, Thomas W Sederberg, and Thomas JR Hughes. Lo-
cal refinement of analysis-suitable t-splines. Computer Methods in Applied
Mechanics and Engineering, 213:206–222, 2012.
[69] Rüdiger Verfürth. A posteriori error estimation techniques for finite element
methods. OUP Oxford, 2013.
[70] Javier Videla, Cosmin Anitescu, Tahsin Khajah, Stéphane PA Bordas, and
Elena Atroshchenko. h-and p-adaptivity driven by recovery and residual-
based error estimators for pht-splines applied to time-harmonic acoustics.
Computers & Mathematics with Applications, 77(9):2369–2395, 2019.
[74] Ping Wang, Jinlan Xu, Jiansong Deng, and Falai Chen. Adaptive isogeo-
metric analysis using rational pht-splines. Computer-Aided Design, 43(11):
1438–1448, 2011.
[75] Wenyan Wang, Yongjie Zhang, Lei Liu, and Thomas J.R. Hughes.
Trivariate solid t-spline construction from boundary triangulations with
arbitrary genus topology. Computer-Aided Design, 45(2):351 – 360,
2013. ISSN 0010-4485. doi: https://doi.org/10.1016/j.cad.2012.10.
018. URL http://www.sciencedirect.com/science/article/pii/
S0010448512002230. Solid and Physical Modeling 2012.
[80] P Wriggers and A Reiger. Comparison of different error measures for adap-
tive finite element techniques applied to contact problems involving large
elastic strains. WIT Transactions on Engineering Sciences, 24, 1970.
[81] P Wriggers, O Scherf, and C Carstensen. Adaptive techniques for the contact
of elastic bodies. Recent Developments in Finite Element Analysis, pages
78–86, 1994.
[84] P. Yu, C. Anitescu, S. Tomar, S.P.A. Bordas, and P. Kerfriden. Adaptive Iso-
geometric analysis for plate vibrations: An efficient approach of local refine-
ment based on hierarchical a posteriori error estimation. Computer Methods
in Applied Mechanics and Engineering, 342, 2018. ISSN 00457825. doi:
10.1016/j.cma.2018.08.010.
[85] O.C. Zienkiewicz, R.L. Taylor, and J.Z. Zhu. The Finite Element Method: Its
Basis and Fundamentals, Sixth edition. International Journal for Numerical
Methods in Engineering, 2005. ISSN 978-0-7506-6320-5. doi: 10.1016/
B978-1-85617-633-0.00015-0.
[86] Olgierd Cecil Zienkiewicz and Jian Zhong Zhu. The superconvergent patch
recovery and a posteriori error estimates. part 1: The recovery technique.
International Journal for Numerical Methods in Engineering, 33(7):1331–
1364, 1992.
110 BIBLIOGRAPHY
[87] Olgierd Cecil Zienkiewicz and Jian Zhong Zhu. The superconvergent patch
recovery and a posteriori error estimates. part 2: Error estimates and adap-
tivity. International Journal for Numerical Methods in Engineering, 33(7):
1365–1382, 1992.
[88] Christopher Zimmermann and Roger A Sauer. Adaptive local surface re-
finement based on lr nurbs and its application to contact. Computational
Mechanics, 60(6):1011–1031, 2017.