Professional Documents
Culture Documents
Tu 2020 6122300053 13251 13279
Tu 2020 6122300053 13251 13279
BY
MR. DUY VO
THAMMASAT UNIVERSITY
BY
MR. DUY VO
THAMMASAT UNIVERSITY
DISSERTATION
BY
MR.DUY VO
E:--JTITLED
was approved as partial fulfillment of the requirements for the degree of Doctor of
Philosophy (Engineering and Technology)
Chairperson
(Associate Professor Jaroon Rungamomrat, Ph.D.)
Member 'f4-,
(Associate Profess.9r Nakhorn Po?varodom, Ph.D.)
--
Member )l� �.
(Research Assistant �ofc�e)Pakawat Sancharoen, Ph.D.)
Director
(Professor Pruettha Nanakom, D.Eng.)
(1)
ABSTRACT
Kirchhoff stress tensor as the energy conjugate pair. This approach facilitates consideration
of hyperelastic materials in analysis of highly flexible beam structures. Although in the
context of the conventional finite element approach, many beam formulations have been
developed by using the Green-Lagrange strain and second Piola-Kirchhoff stress tensors,
there virtually exist no isogeometric Timoshenko beam formulations that are derived by
using this conjugate pair. Three-dimensional beam configurations are reduced into one-
dimensional structures using the beam axis and director vectors of the cross-sections. The
displacements of the beam axis and the total cross-sectional rotation are considered as
unknown kinematics. The cross-sectional rotation is represented by an orthogonal tensor,
which is parameterized by a vector-like parameter. Updating the cross-sectional rotations
is performed purely through natural exponentiation and superposition of relevant rotational
quantities. This enables the proposed Timoshenko beam formulation to tackle beam
structures undergoing large displacements and rotations without any restriction in
magnitude.
To show the accuracy and efficiency of the proposed beam formulations, some
benchmark and well-established numerical examples with various types of beams, i.e.,
straight, curved, pre-twisted beams, and lattice-like beam structures, are analyzed. The
obtained results are compared with those in the literature, obtained from both analytical
and numerical methods.
ACKNOWLEDGEMENTS
During this research, I spent eight months at Tokyo Institute of Technology, Tokyo,
Japan. I am very thankful to friends and colleagues in HIROSE & BUI Lab for making
memorable moments in Japan.
Last but not least, special thanks are dedicated to my beloved family and my dear
girlfriend, Giang, for their patience and encouragement at all times.
Duy Vo
TABLE OF CONTENTS
Page
ABSTRACT (1)
ACKNOWLEDGEMENTS (4)
CHAPTER 1 INTRODUCTION 1
1.1 General 1
3.1 B-splines 9
5.1.1 Smallest rotation (SR) mapping and director vectors of Euler-Bernoulli beams
52
CHAPTER 6 CONCLUSIONS 81
REFERENCES 83
APPENDICES 90
APPENDIX A 91
APPENDIX B 93
APPENDIX C 95
APPENDIX D 98
BIOGRAPHY 100
LIST OF TABLES
Tables Page
4.1 Rolled-up straight cantilever beam – Single circle test: comparison of the number of
degrees of freedom. 44
5.1 Twisting of a circular ring. Comparison of the number of degrees of freedom for a
half of the ring. 75
LIST OF FIGURES
Figures Page
3.1 (a) Cubic B-spline basis functions with the knot vector 𝛷 =
0, 0, 0, 0, 0.25, 0.5, 0. ,75, 0.75, 1, 1, 1, 1. (b) An associated cubic B-spline curve with
knots and control points. 10
3.2 (a) NURBS and B-spline curves-the associated weights of the NURBS curve are
given in the parentheses, and the unit weights are dropped. (b) Associated ration B-spline
and B-spline basis functions. 12
4.2 Beam axis and director vectors in the reference and current configurations. 16
4.5 Cantilever quadrant subjected to an end moment: (a) Strain energy obtained by
various NURBS curves of different degrees. (b) Contribution of the membrane and shear
strain energy. 37
4.7 Pre-twisted circular arch – Convergence tests for the deformation of the arch axis. 40
4.8 Pre-twisted circular arch: (a) Displacement 𝑢𝑋. (b) Displacement 𝑢𝑌. (c)
Displacement 𝑢𝑍. (d) Reference and deformed configurations. 41
4.10 Rolled-up straight cantilever beam – Single circle test: (a) Convergence tests for the
position of the free end. (b) Normalized load-displacement curves of the free end. (c)
Reference and final deformed configurations. 43
4.11 Rolled-up straight cantilever beam – Double circle test: (a)Convergence tests of the
position of the free end. (b) Normalized load-displacement curves of the free end. (c)
Reference and deformed configurations. 45
4.13 Lateral buckling of a cantilever beam: (a) Convergence tests for the deformation of
the beam axis. (b) Load-displacement curves of the free end. (c) Reference and deformed
configurations. 47
5.2 Square frame subjected to opposite compression forces: (a) Problem set up. (b)
Equivalent model. 65
5.4 Lee’s frame: (a) Problem set up. (b) Normalized load-displacement curves of the
point 𝐷. (c) Deformed frame axes. 67
5.6 A pre-twisted circular arch: (a) Top view. (b) NURBS representations of the arch and
control points. 69
5.8 A pre-twisted circular arch – Convergence test for the relative error in the
deformation of the arch axis. 71
5.9 A pre-twisted circular arch: (a) Displacement 𝑢𝑋. (b) Displacement 𝑢𝑌. (c)
Displacement 𝑢𝑍. (d) Reference and deformed configurations (with a scale factor of 1000
for displacements). 72
5.10 Twisting of a circular ring: (a) Problem set up. (b) Moment-rotation curve of the
point 𝐷. 74
5.12 Compression of a lattice tower: (a) Problem set up. (b) Target surface and embedded
curves. 77
5.13 Compression of a lattice tower: (a) NURBS representation of the generatrix of the
target surface with the axis of revolution. (b) Parent domains of the embedded curves. 78
5.14 Compression of a lattice tower – Comparison of the displacement at the joints 𝐾 and
𝑁. 79
CHAPTER 1
INTRODUCTION
1.1 General
Among various beam theories, both the Euler-Bernoulli and the Timoshenko beam
theories are popular. Both theories assume that the beam cross-sections are rigid. The major
difference between them lies in the orientations of beam cross-sections. While the former
theory assumes that rigid beam cross-sections are always perpendicular to the beam axis
during deformation, the latter theory releases this assumption by including the shear effect.
In practice, the Euler-Bernoulli beam theory is suitable for analysis of thin beams, whereas
the Timoshenko beam theory has a wider range of applications since its ability to deal with
both thin and thick beams. Nevertheless, the performance of finite element formulations
based on the Timoshenko beam theory suffers from the well-known shear locking
phenomenon when accounting for thin beams. Special treatments are thus required to
eliminate the shear locking issue.
Recently, isogeometric analysis (IGA) (Hughes, Cottrell & Bazilevs, 2005) has
become a powerful numerical approach in the field of computational mechanics. The
primary concept of IGA is to fill in the gap between geometric design and computational
analysis by using the same basis functions used in computer-aided design (CAD), e.g.,
rational B-spline basis functions, for both geometric descriptions of domains and
approximation of unknown fields. With this approach, exact representations of complex
reference geometry, generated by CAD, can be used directly in the analysis without any
modifications. In addition, the use of CAD basis functions with high order of continuity
for the approximation of unknown fields results in better accuracy per degree of freedom.
These advantages of IGA have been verified for analysis of structural members, i.e., plates
and shells (Dornisch & Klinkel, 2014; Dornisch, Klinkel & Simeon, 2013; Dornisch,
Müller & Klinkel, 2016; Kiendl, Bazilevs, Hsu, Wüchner & Bletzinger, 2010; Kiendl,
Bletzinger, Linhard & Wüchner, 2009), beams and rods (Fang, Yu, Van Lich & Bui, 2019;
Vo & Nanakorn, 2020; Yu, Hu, Zhang & Bui, 2019; Yu, Zhang, Hu & Bui, 2019), and
other engineering applications (Lai et al., 2017; Yu, Lai & Bui, 2019).
In literature, for shear deformable beams, the geometrically exact beam theory
(Simo, 1985) is intensively considered. However, in the geometrically exact beam theory,
the deformation gradient tensor and first Piola-Kirchhoff stress tensor are used as the
energy conjugate pair in the derivation of finite element equations. This approach is not
convenient when considering complicated material constitutive laws, e.g. hyperelastic
materials or functionally graded materials, where the Green-Lagrange strain tensor is
widely used. In these cases, beam formulations using the Green-Lagrange strain tensor and
second Piola-Kirchhoff stress tensor in the derivation of finite element equations are
preferable.
As the use of Timoshenko beam formulations for analysis of thin beams suffers
from the shear locking effects, the development of Euler-Bernoulli beam formulations is
necessary to get rid of special treatments of the shear locking issue. In fact, the efficiency
and accuracy of Euler-Bernoulli beams with the IGA approach have been shown by many
previous studies. However, due to the lack of rotational degrees of freedom in Euler-
Bernoulli beam formulations with the IGA approach, analyzing multi-patch beams is still
challenging. In this study, this limitation is handled by using the end rotations of patches
as degrees of freedom. Numerical results show the accuracy and efficiency of the
developed beam formulation in modelling beam structures consisting of multi-patches.
This study aims to develop effective beam formulations in association with the IGA
approach using kinematic assumptions of the Timoshenko beam theory and the Euler-
Bernoulli beam theory. The main objectives of this thesis are as follows:
- The kinematic assumptions of the Timoshenko beam theory and the Euler-
Bernoulli beam theory are considered.
CHAPTER 2
REVIEW OF LITERATURE
Spatial flexible beam structures subjected to large displacements and rotations are
often encountered in a wide range of engineering applications, perhaps most notably in
space, marine, and offshore structures. In order to gain insight into mechanical behavior of
these structures, many spatial beam formulations have been proposed (Bathe & Bolourchi,
1979; Cardona & Geradin, 1988; Crisfield, 1990; Ibrahimbegovic, 1997; Ibrahimbegović,
1995; Ibrahimbegović, Frey & Kožar, 1995; Reissner, 1973; Simo, 1985; Simo & Vu-
Quoc, 1986). A thorough discussion on the historical development of spatial beam
formulations is beyond the scope of this work, and curious readers may refer to some
compact reviews, e.g., (Greco & Cuomo, 2015; Meier, Popp & Wall, 2019).
Recently, isogeometric analysis (IGA) (Hughes, Cottrell & Bazilevs, 2005) has
become a powerful numerical approach in the field of computational mechanics. The
primary concept of IGA is to fill in the gap between geometric design and computational
analysis by using the same basis functions used in computer-aided design (CAD), e.g.,
rational B-spline basis functions, for both geometric descriptions of domains and
approximation of unknown fields. With this approach, exact representations of complex
reference geometry, generated by CAD, can be used directly in analysis without any
modifications. In addition, the use of CAD basis functions with high order of continuity
for the approximation of unknown fields results in better accuracy per degree of freedom.
These advantages of IGA have been verified for analysis of structural members, i.e., plates
and shells (Dornisch & Klinkel, 2014; Dornisch, Klinkel & Simeon, 2013; Dornisch,
Müller & Klinkel, 2016; Kiendl, Bazilevs, Hsu, Wüchner & Bletzinger, 2010; Kiendl,
Bletzinger, Linhard & Wüchner, 2009), beams and rods (Fang, Yu, Van Lich & Bui, 2019;
Vo & Nanakorn, 2020; Yu, Hu, Zhang & Bui, 2019; Yu, Zhang, Hu & Bui, 2019), and
other engineering applications (Lai et al., 2017; Yu, Lai & Bui, 2019).
using IGC is performed by Auricchio, Beirão da Veiga, Kiendl, Lovadina and Reali (2013).
The governing equations are given in two forms, i.e., the displacement-based form and the
mixed one, where both displacements and forces are primary unknowns. The locking
effects in the displacement-based formulation can be avoided by using rational B-spline
basis functions of high degree, whereas the mixed formulation is completely locking-free
regardless of the degrees of the basis functions used. The IGC approach is applied to the
geometrically exact beam model (Simo, 1985) for geometrically nonlinear analysis of
beams by Weeger, Yeung and Dunn (2017), and shortly later also by Marino (2016, 2017).
Different parameterizations of cross-sectional rotations are used, namely quaternions in the
former work and vector-like parameters in the latter works. Later, the IGC approach with
the geometrically exact beam model has been further applied to several engineering
problems, i.e., frictionless contacts (Weeger, Narayanan, De Lorenzis, Kiendl & Dunn,
2017), shape optimization (Weeger, Narayanan & Dunn, 2019), and nonlinear dynamic
analysis of beam structures (Marino, Kiendl & De Lorenzis, 2019, 2019). Through all the
works discussed above, the IGC approach has been shown to be effective for nonlinear
analysis of beams where the numerical integration is avoided, and the need for the higher
continuity of displacements required by the governing equations is easily handled by the
smoothness of rational B-spline basis functions. Therefore, the IGC approach permits good
accuracy to be obtained with reduced computational costs. However, for problems with
complex mechanical phenomena, where the governing equations are not available, the IGC
approach cannot be used. In this case, the variational approach can be considered as an
effective alternative. One should also note that variational Timoshenko beam formulations
with the IGA approach have not been widely considered. Most recently, the variational
formulation for the geometrically exact beam model (Simo, 1985) is used with the IGA
approach by Choi and Cho (2019) for design sensitivity analysis of beam structures.
Recently, several isogeometric shell formulations have been proposed with the
energy conjugate pair of the Green-Lagrange strain tensor and the second Piola-Kirchhoff
stress tensor (Dornisch & Klinkel, 2014; Dornisch, Klinkel & Simeon, 2013; Dornisch,
Müller & Klinkel, 2016; Kiendl, Bletzinger, Linhard & Wüchner, 2009). This approach
facilitates analysis of shell structures made of hyperelastic materials (Chen et al., 2014;
Kiendl, Hsu, Wu & Reali, 2015; Tepole, Kabaria, Bletzinger & Kuhl, 2015), whose strain
energy density functions are usually described in terms of the Green-Lagrange strain
tensor. In the conventional finite element approach, many beam formulations have been
developed using the Green-Lagrange strain tensor and the second Piola-Kirchhoff stress
tensor. However, there virtually exist no isogeometric Timoshenko beam formulations
derived with this energy conjugate pair of the strain and stress tensors.
Only a few IGA Euler-Bernoulli beam formulations have been proposed (Bauer,
Breitenberger, Philipp, Wüchner & Bletzinger, 2016; Bauer, Wüchner & Bletzinger, 2019;
Greco & Cuomo, 2013, 2014; Radenković & Borković, 2018; Raknes et al., 2013). By
using only the displacements of the beam axis as unknown kinematics, nonlinear dynamic
analysis of beams is performed by Raknes et al. (2013). Since only displacements are used
as the unknowns, torsional behavior cannot be considered. For analysis of beams with
torsional effects and pre-twisted configurations, the axial rotation angle around the beam
axis is included in the formulations developed by Greco and Cuomo (2013) and
Radenković and Borković (2018) for linear analysis, and by Bauer, Breitenberger, Philipp,
Wüchner and Bletzinger (2016) for geometrically nonlinear analysis. The formulation
developed by Greco and Cuomo (2013) is later extended by Greco and Cuomo (2014) for
assemblies of multi-patch beam structures by using the cross-sectional rotations at the ends
of patches as discrete unknowns. Although the formulations proposed by Greco and Cuomo
(2013, 2014) are derived with finite deformation kinematics, only examples in the linear
regime are shown. Geometrically nonlinear analysis of multi-patch beam structures is
performed by Bauer, Wüchner and Bletzinger (2019). Rigid connections between patches
is considered by using some additional constraints, and the penalty approach is used to
incorporate these constraints into the virtual work principle. In this approach, the accuracy
is affected by the choices of problem-dependent coefficients. Additionally, for beam
structures with many patches, the inclusion of these additional constraints is not practically
convenient.
Although very promising results have been reported (Bauer, Breitenberger, Philipp,
Wüchner & Bletzinger, 2016; Bauer, Wüchner & Bletzinger, 2019; Greco & Cuomo, 2013,
2014), it is noticed that strain measurements are highly nonlinear in terms of the unknown
kinematics since the Euler-Rodriguez formula is used for the description of finite cross-
sectional rotations. Consequently, the linearization of the system equations is
computationally expensive. In the context of geometrically nonlinear analysis, the
enforcement of rigid connections in multi-patch beams still requires an approach that is
more convenient for practical applications.
CHAPTER 3
In IGA, rational B-spline basis functions are used both for the approximations of
the geometry and unknown variable fields. In this section, rational B-spline basis functions
and their properties are briefly introduced. More details about these functions can be found
in the textbook of Piegl and Tiller (1997).
3.1 B-splines
where 𝜂 is the parametric variable and 𝑁𝑖,𝑝 (𝜂) is a B-spline basis function of degree 𝑝. In
addition, 𝐏𝑖 ∈ ℝ𝑑 is a control point while 𝑛 is the number of the control points.
where 𝜂𝑖 ≤ 𝜂𝑖+1 are knot values. A knot vector is called uniform if its elements are equally
spaced. In addition, a knot vector is called open if the multiplicities of knot values at the
ends are both equal to 𝑝 + 1. Hereafter, knot vectors in this paper refer to nonuniform,
open knot vectors.
1 if 𝜂 ∈ [𝜂𝑖 , 𝜂𝑖+1 )
𝑁𝑖,0 (𝜂) = { (3.3)
0 otherwise
Figure 3.1 (a) Cubic B-spline basis functions with the knot vector 𝛷 =
[0, 0, 0, 0, 0.25, 0.5, 0. ,75, 0.75, 1, 1, 1, 1]. (b) An associated cubic B-spline curve with
knots and control points.
𝜂 − 𝜂𝑖 𝜂𝑖+𝑝+1 − 𝜂
𝑁𝑖,𝑝 (𝜂) = 𝑁𝑖,𝑝−1 (𝜂) + 𝑁 (𝜂).
𝜂𝑖+𝑝 − 𝜂𝑖 𝜂𝑖+𝑝+1 − 𝜂𝑖+1 𝑖+1,𝑝−1 (3.4)
B-spline basis functions have the property of the partition of unity, namely
∑𝑛𝑖=1 𝑁𝑖,𝑝 (𝜂) = 1. They are non-negative and locally supported on [𝜂1 , 𝜂𝑛+𝑝+1 ]. At any
point on [𝜂1 , 𝜂𝑛+𝑝+1 ], there are at most 𝑝 + 1 basis functions that are non-zero. In addition,
at an internal knot value, namely a knot value that is not the first or the last one, B-spline
basis functions are 𝐶 𝑝−𝑘 continuous, where 𝑘 is the multiplicity of the knot value and 1 ≤
𝑘 ≤ 𝑝 (Piegl & Tiller, 1997). The continuity of a B-spline curve follows the continuity of
its B-spline basis functions. In general, a B-spline curve does not pass through all its control
points. However, if an internal knot value of a B-spline curve has the multiplicity of 𝑝, the
B-spline curve interpolates the control point at that knot value. In addition, a B-spline curve
having an open knot vector interpolates its first and last control points. A so-called control
polygon is obtained by connecting consecutive control points with linear functions.
B-spline basis functions 𝑁𝑖,𝑝 (𝜂), defined by the Cox-de Boor relation, are in fact
polynomials, written in different forms. Consequently, B-spline curves are polynomial
curves and are not capable of exactly representing Conic sections, such as circles and
ellipses. Another class of curves, called non-uniform rational B-spline (NURBS) curves,
does not have this drawback. NURBS curves generally have non-uniform knot vectors and
are constructed from so-called rational B-spline basis functions. A NURBS curve is defined
by
𝑛
∑𝑛𝑖=1 𝑁𝑖,𝑝 (𝜂)𝑤𝑖 𝐏𝑖
𝐂(𝜂) = = ∑ 𝑅𝑖,𝑝 (𝜂)𝐏𝑖 . (3.5)
∑𝑛𝑖=1 𝑁𝑖,𝑝 (𝜂)𝑤𝑖
𝑖=1
Here, 𝑤𝑖 is a scalar weight coefficient or simply a weight, and 𝑅𝑖,𝑝 (𝜂) is a rational B-spline
basis function defined by
𝑁𝑖,𝑝 (𝜂)𝑤𝑖
𝑅𝑖,𝑝 (𝜂) = . (3.6)
∑𝑛𝑗=1 𝑁𝑗,𝑝 (𝜂)𝑤𝑗
In this study, all 𝑤𝑖 s are restricted to be positive. As a result, the denominators in Eqs. (3.5)
and (3.6) are always positive, and the division by zero is explicitly avoided. It should be
noted that, if all 𝑤𝑖 s in a rational B-spline basis function 𝑅𝑖,𝑝 (𝜂) are identical, the rational
B-spline basis function 𝑅𝑖,𝑝 (𝜂) simply becomes a B-spline basis function 𝑁𝑖,𝑝 (𝜂).
Therefore, B-spline curves are special cases of NURBS curves.
Figure 3.2 (a) NURBS and B-spline curves-the associated weights of the NURBS
curve are given in the parentheses, and the unit weights are dropped. (b) Associated
ration B-spline and B-spline basis functions.
As an example comparison, a NURBS curve and a B-spline curve having the same
control points are shown in Figure 3.2a. The non-unit weights used in the NURBS curve
are given in the parentheses. The associated rational B-spline and B-spline basis functions
are shown in Figure 3.2b. It can be noticed that the NURBS curve exactly represents a
circle.
CHAPTER 4
The kinematic description of a spatial beam based on the Timoshenko beam theory
is presented here. In this study, the unstressed configuration of the beam is treated as the
reference configuration. From now onwards, as a rule for notations, upper-case and lower-
case letters are used to denote the kinematic quantities in the reference and current
configurations, respectively. The convective coordinates of the beam are denoted by
𝜉 (𝑖) , 𝑖 = 1,2,3, where 𝜉 (1) is the convective coordinate along the beam axis. In addition,
𝜉 (2) and 𝜉 (3) denote the convective coordinates along the principal axes of the cross-
sections. The ranges of Roman and Greek indices are from 1 to 3 and 2 to 3, respectively,
unless otherwise stated. In this study, the arc-length of the reference beam axis is used as
𝜉 (1) . For convenience, 𝜉 (1) is interchangeably written as 𝜉.
Any configuration of a spatial beam under the Timoshenko beam theory can be
represented by the beam axis and the cross-sections. Here, director vectors are the unit
vectors that are aligned with the principal axes of the cross-sections. The determination of
the director vectors is given below.
The unit tangent vector of the reference beam axis at 𝑄0 can be identified as
d𝐑 0 (𝜉)
𝐀1 = = 𝐑′0 (𝜉). (4.2)
d𝜉
𝐀′1
𝐍= 𝐁 = 𝐀1 × 𝐍. (4.3)
‖𝐀′1 ‖
The three equations in Eqs. (4.2) and (4.3) are the so-called Frenet-Serret formulas. For
later use, the derivatives of 𝐍 and 𝐁 with respect to 𝜉 are given by
One can see that ‖𝐀′𝟏 ‖ = 0 for a straight beam. Consequently, 𝐍 and 𝐁 are not defined.
Later in subsection 4.4.3, an alternative method is introduced for the determination of 𝐍
and 𝐁 of a straight beam.
𝐀2 cos 𝜙 sin 𝜙 𝐍
[ ]=[ ] [ ]. (4.5)
𝐀3 − sin 𝜙 cos 𝜙 𝐁
Due to the application of the external loads, the cross-section C moves to the cross-
section c in the current configuration, and the point 𝑄0 on the reference beam axis moves
Figure 4.2 Beam axis and director vectors in the reference and current configurations.
to the point 𝑞0 on the current beam axis as shown in Figure 4.2. The position of 𝑞0 is given
by
where 𝐮0 (𝜉) is the translational displacement vector of the beam axis. The tangent vector
of the current beam axis at 𝑞0 can be written as
d𝐫0 (𝜉)
𝐚1 = = 𝐫0′ (𝜉). (4.8)
d𝜉
In the Timoshenko beam theory, the cross-section c is still planar but generally no
longer perpendicular to the current beam axis. As a result, the director vectors 𝐚2 and 𝐚3
of the cross-section c cannot be determined from the current beam axis. In this study, they
are determined from 𝐀 2 and 𝐀 3 as
𝐚𝛼 = 𝚲𝐀 𝛼 (4.9)
Let 𝑄 be an arbitrary material point on C as shown in Figure 4.3. The position vector
of 𝑄 can be determined as
The point 𝑄 moves to the point 𝑞 on c in the current configuration. The position
vector of 𝑞 is expressed as
It can be seen from Eqs. (4.12) and (4.13) that the position of any point of the beam
can be determined by the beam axis and the director vectors of the cross-section containing
the point. However, the beam axis 𝐫0 (𝜉) and the director vectors 𝐚𝛼 of the cross-section in
the current configuration are unknown. Note that 𝐚𝛼 is related to 𝐀 𝛼 through the rotation
tensor 𝚲 as shown in Eq. (4.9). For the proposed Timoshenko beam formulation, the current
beam axis 𝐫0 (𝜉) and the rotation tensor 𝚲(𝜉) are considered as the unknown kinematics.
Covariant base vectors are defined, respectively, for a point in the reference and
current configurations as
𝐆𝛼 = 𝐀 𝛼 𝐠 𝛼 = 𝐚𝛼 . (4.16)
The reciprocal contravariant vectors 𝐆𝑖 are obtained from the following relation as
1 𝑖=𝑗
𝐆𝑖 ∙ 𝐆𝑗 = { (4.17)
0 𝑖 ≠ 𝑗.
𝐆 = 𝐺𝑖𝑗 𝐆𝑖 ⊗ 𝐆 𝑗 (4.18)
where the coefficients 𝐺𝑖𝑗 are the covariant metric components in the reference
configuration, defined as
𝐺𝑖𝑗 = 𝐆𝑖 ∙ 𝐆𝑗 . (4.19)
Similarly, the covariant metric components in the current configuration are defined as
𝑔𝑖𝑗 = 𝐠 𝑖 ∙ 𝐠 𝑗 . (4.20)
𝐅 = 𝐠 𝑖 ⊗ 𝐆𝑖 . (4.21)
1 T 1 1
𝐄= (𝐅 𝐅 − 𝐆) = (𝐠 𝑖 ∙ 𝐠 𝑗 − 𝐆𝑖 ∙ 𝐆𝑗 )𝐆𝑖 ⊗ 𝐆 𝑗 = (𝑔𝑖𝑗 − 𝐺𝑖𝑗 )𝐆𝑖 ⊗ 𝐆 𝑗 . (4.22)
2 2 2
1
𝐸𝑖𝑗 = (𝑔 − 𝐺𝑖𝑗 ). (4.23)
2 𝑖𝑗
To obtain 𝐸𝑖𝑗 , 𝐺𝑖𝑗 and 𝑔𝑖𝑗 are determined, with the help of Eqs. (4.15) and (4.16) as
1 if 𝛼 = 𝛽 1 if 𝛼 = 𝛽
𝐺𝛼𝛽 = { 𝑔𝛼𝛽 = { (4.27)
0 if 𝛼 ≠ 𝛽 0 if 𝛼 ≠ 𝛽.
2
In the above expressions of 𝐺𝑖𝑗 and 𝑔𝑖𝑗 , all the quadratic terms, i.e., 𝛰 [(𝜉 (2) ) ],
2
𝛰 [(𝜉 (3) ) ], and 𝛰(𝜉 (2) 𝜉 (3) ), are ignored. By making use of Eqs. (4.24)-(4.27), the non-
𝛤11
12 𝐾 13 𝐾
⏞
1 (2) ⏞ (3) ⏞
𝐸11 = (𝐚1 ∙ 𝐚1 − 𝐀1 ∙ 𝐀1 ) + 𝜉 (𝐚1 ∙ 𝐚2 − 𝐀1 ∙ 𝐀 2 ) + 𝜉 (𝐚1 ∙ 𝐚3 − 𝐀1 ∙ 𝐀′3 )
′ ′ ′
2
(4.28)
𝛤12 𝐾23
1 ⏞ (3) ⏞
𝐸12 = [(𝐚1 ∙ 𝐚2 − 𝐀1 ∙ 𝐀 2 ) + 𝜉 (𝐚2 ∙ 𝐚′3 − 𝐀 2 ∙ 𝐀′3 )] (4.29)
2
𝛤13 𝐾32
1 ⏞ (2) ⏞
𝐸13 = [(𝐚1 ∙ 𝐚3 − 𝐀1 ∙ 𝐀 3 ) + 𝜉 (𝐚3 ∙ 𝐚′2 − 𝐀 3 ∙ 𝐀′2 )] (4.30)
2
or in compact expressions as
1 1
𝐸12 = (𝛤12 + 𝜉 (3) 𝐾23) 𝐸13 = (𝛤13 + 𝜉 (2) 𝐾32). (4.32)
2 2
The strain component 𝐸11 consists of one constant term and two linear terms with
respect to 𝜉 (2) and 𝜉 (3) , which are the convective coordinates along the principal axes of
the cross-sections. The constant term describes the membrane action while the linear terms
represent the bending actions. The term 𝛤11 in Eq. (4.31) is referred to as the membrane
strain, while the terms 𝜉 (2) 𝐾12 and 𝜉 (3) 𝐾13 are the bending strains. In fact, the terms 𝐾12
and 𝐾13 are terms that are similar to the bending curvatures in the directions of the principal
axes of the cross-sections in the Euler-Bernoulli beam theory. However, because of the
cross-sectional rotations due to the shear deformation in the Timoshenko beam theory, 𝐾12
and 𝐾13 are not exactly curvatures.
Each of the other two strain components, 𝐸12 and 𝐸13 , has one constant term and
one linear term. The constant terms describe the shear actions whereas the linear terms
represent the torsional actions. The terms 𝛤12 /2 and 𝛤13 /2 represent the direct shear strains,
while 𝜉 (3) 𝐾23 /2 and 𝜉 (2) 𝐾32 /2 are the torsional shear strains. The terms 𝐾23 and 𝐾32 are
in fact the torsional curvatures and are related through the following relations, i.e.,
In other words, the two torsional curvatures have the same absolute value.
The second Piola-Kirchhoff stress tensor 𝐒 is the energy conjugate of the Green-
Lagrange strain tensor 𝐄. The second Piola-Kirchhoff stress tensor 𝐒 is expressed as
𝐒 = 𝑆 𝑖𝑗 𝐆𝑖 ⊗ 𝐆𝑗 . (4.35)
As the Green-Lagrange strain tensor 𝐄 has only three non-zero components, it follows that
where 𝐸̅ , and 𝐺̅ are Young’s modulus and the shear modulus measured in the curvilinear
coordinate system 𝐆𝑖 ⊗ 𝐆𝑗 (see Appendix B).
- Normal force:
- Shear forces:
- Bending moments:
2 2
𝐼2 = ∫ (𝜉 (2) ) d𝐴 𝐼3 = ∫ (𝜉 (3) ) d𝐴 𝐼23 = ∫ 𝜉 (2) 𝜉 (3) d𝐴. (4.42)
𝐴 𝐴 𝐴
- Torsional moment
2 2
𝐽 = ∫ [(𝜉 (2) ) + (𝜉 (3) ) ] d𝐴. (4.44)
𝐴
Recall that the beam axis 𝐫0 (𝜉) and the rotation tensor 𝚲(𝜉) are considered as the
unknown kinematics in this study. The beam axis 𝐫0 (𝜉) is a member of the linear Euclidean
space while the rotation tensor 𝚲(𝜉) belongs to the orthogonal group 𝑆𝑂(3).
Mathematically, a beam configuration is a member of a nonlinear manifold Ω, defined as
Let 𝚲1 and 𝚲2 be two distinct rotation tensors. The summation 𝚲1 + 𝚲2 does not always
result in an orthogonal tensor. Hence, Ω is not a linear space but a nonlinear manifold. Due
to the mathematical complexity intrinsic to the space of beam configurations, the
admissible variations of the kinematics used in the derivation of the system equations are
explicitly defined.
̃).
𝚲 = exp(𝛉 (4.46)
In practice, it is more convenient to work with the axial vector of a skew-symmetric tensor
̃ is defined as
instead of the tensor itself. The axial vector 𝛉 of 𝛉
̃𝐚 = 𝛉 × 𝐚,
𝛉 ∀𝐚 ∈ ℝ3 . (4.47)
By considering a virtual displacement vector 𝛿𝐮0 (𝜉) and a virtual axial vector
𝛿𝐰(𝜉), a varied beam axis 𝐫0𝜀 (𝜉) and a varied rotation tensor 𝚲𝜀 (𝜉) are defined as
Since 𝐫0𝜀 (𝜉) ∈ ℝ3 and 𝚲𝜀 (𝜉) ∈ 𝑆𝑂(3), it follows that a varied beam configuration
(𝐫0𝜀 (𝜉), 𝚲𝜀 (𝜉)) is always a member of Ω, i.e., (𝐫0𝜀 (𝜉), 𝚲𝜀 (𝜉)) ∈ Ω. Therefore, the varied
beam configuration (𝐫0𝜀 (𝜉), 𝚲𝜀 (𝜉)) is admissible. Straightforwardly, the admissible
variation of the unknown beam axis 𝐫0 (𝜉) is as explicitly given in Eq. (4.48), i.e.,
d𝚲𝜀
𝛿𝚲(𝜉) = 𝜀 | = 𝜀𝐪 ̃ (𝜉)𝚲(𝜉).
̃(𝜉)𝚲(𝜉) = 𝛿𝐰 (4.51)
d𝜀 𝜀=0
̃ 𝚲 = (𝐈 + 𝛿𝐰
𝚲𝜀 = 𝚲 + 𝛿𝚲 = 𝚲 + 𝛿𝐰 ̃ )𝚲. (4.52)
It can be seen from Eqs. (4.48) and (4.52) that the vector 𝛿𝐮0 (𝜉) is an infinitesimal virtual
displacement vector superposed onto the current beam axis, and the skew-symmetric tensor
̃ (𝜉) adds an infinitesimal virtual rotation to the current cross-sectional rotation.
𝛿𝐰
However, as aforementioned, it is more convenient to work with the virtual axial vector
̃ (𝜉). Thus, the vectors 𝛿𝐮0 (𝜉) and 𝛿𝐰(𝜉) are considered as the admissible
𝛿𝐰(𝜉) than 𝛿𝐰
variations of the unknown kinematics. In fact, 𝛿𝐰(𝜉) can be interpreted as an infinitesimal
virtual rotation superposed onto the current cross-sectional rotation (Simo & Vu-Quoc,
1986).
Once the variation of 𝚲 has been clearly defined, the variations of the director
vectors 𝐚2 and 𝐚3 are determined as
The weak form of the equilibrium is expressed using the virtual work principle as
where 𝛿Π is the total virtual work. In addition, 𝛿Π𝑖𝑛𝑡 is the internal virtual work, and 𝛿Π𝑒𝑥𝑡
is the external virtual work.
In this study, the internal virtual work 𝛿Π𝑖𝑛𝑡 is expressed by means of the Green-
Lagrange strain tensor 𝐄 and the second Piola-Kirchhoff stress tensor 𝐒 as
where 𝑉 is the initial volume of the beam. By using the expressions of the strain
components in Eqs. (4.31) and (4.32) and the definitions of the cross-sectional resultant
forces in Eqs. (4.37)-(4.44), the internal virtual work 𝛿Π𝑖𝑛𝑡 can be alternatively expressed
as
𝛿Π𝑖𝑛𝑡 = ∫(𝛿𝛤11 𝑁 11 + 𝛿𝛤12 𝑁12 + 𝛿𝛤13 𝑁13 + 𝛿𝐾12 𝑀12 + 𝛿𝐾13 𝑀13 + 𝛿𝐾23 𝑀23 ) d𝜉
0
(4.57)
where 𝐿 is the length of the reference beam axis. The following generalized strain and
stress vectors are introduced as
The generalized strain vector 𝚪 and generalized stress vector 𝐌 are related as follows:
𝐸̅ 𝐴 0 0 0 0 0
0 𝐺̅ 𝐴 0 0 0 0
0 0 𝐺̅ 𝐴 0 0 0
𝐌= 0 0 0 𝐸̅ 𝐼2 𝐸̅ 𝐼23 0 𝚪 = 𝐃𝚪. (4.60)
0 0 0 𝐸̅ 𝐼23 𝐸̅ 𝐼3 0
[0 0 0 0 0 𝐺̅ 𝐽]
Hereafter, all the equations are expressed in matrix form for computational convenience.
(4.62)
where 𝐅 and 𝐐 are the distributed force and distributed moment vectors applied on the
beam axis. Furthermore, 𝐟(0), 𝐟(𝐿), 𝐪(0), and 𝐪(𝐿) are the concentrated force and
concentrated moment vectors applied at 𝜉 = 0 and 𝜉 = 𝐿, respectively.
In this study, rational B-spline basis functions are used for both the geometric
approximation and the discretization of the unknown kinematics. A very brief introduction
to rational B-spline basis functions and their properties has been presented in Chapter 3.
More details can be found in the textbook of Piegl and Tiller (1997).
The reference and current beam axes are represented as NURBS curves as
𝑛 𝑛
One should recall that 𝑅𝑖,𝑝 (𝜂) is a rational B-spline basis function of degree 𝑝, 𝜂 is the
parameter referred to as the knot parameter, and 𝑛 is the number of basis functions. In
addition, 𝐑 𝑖 and 𝐫𝑖 are the position vectors of the 𝑖th control point in the reference and
current configurations. Hereafter, the degree 𝑝 of basis functions is omitted for notational
simplicity. The displacement vector of the beam axis is then obtained as
𝑛 𝑛
𝑛 𝑛
The variation of the displacement 𝛿𝐮0 (𝜉) and the virtual axial vector 𝛿𝐰 are
approximated as
𝑛 𝑛
The variations of the tangent vector 𝐚1 and the director vectors 𝐚2 and 𝐚3 are
expressed, respectively, as
𝑛 𝑛
𝑛 𝑛
𝛿𝐚𝛼 = − ∑ 𝑅𝑖 (𝜂)𝐚̃𝛿𝐰
𝛼 𝑖 𝛿𝐚′𝛼 = − ∑[𝑅𝑖′ (𝜂)𝐚̃𝛿𝐰
𝛼
̃′
𝑖 + 𝑅𝑖 (𝜂)𝐚𝛼 𝛿𝐰𝑖 ].
𝑖=1 𝑖=1
(4.68)
In this study, two different parameterizations of the beam axis, i.e., arc-length
parameterization 𝜉 and the NURBS parameterization 𝜂, are employed. Here, the relation
between the two parameterizations is established.
d𝐑 0 (𝜂)
d𝜉 = ‖ ‖ d𝜂 = ‖𝐑̇ 0 (𝜂)‖d𝜂 = 𝐼d𝜂. (4.69)
d𝜂
Hereafter, the dot notations indicate the derivatives with respect to 𝜂. It follows that
d𝑅𝑖 (𝜂) d𝜂 1
𝑅𝑖′ (𝜂) = = 𝑅̇ (𝜂). (4.70)
d𝜂 d𝜉 𝐼 𝑖
The determination of 𝑅̇𝑖 (𝜂) is available in the textbook of Piegl and Tiller (1997).
𝑅𝑖′ (𝜂)𝐚1T 𝟎T
𝑅𝑖′ (𝜂)𝐚T2 −𝑅𝑖 (𝜂)𝐚1T 𝐚̃2
𝑅𝑖′ (𝜂)𝐚T3 T
−𝑅𝑖 (𝜂)𝐚1 𝐚 ̃3
𝐇𝑖 = 𝐇 = [𝐇1 𝐇2 … 𝐇𝑛 ] (4.72)
𝑅𝑖′ (𝜂)(𝐚′2 )T ′
−𝑅𝑖 (𝜂)𝐚1T 𝐚 ̃′
̃2 − 𝑅𝑖 (𝜂)𝐚1T 𝐚2
𝑅𝑖′ (𝜂)(𝐚′3 )T ′ T
−𝑅 (𝜂)𝐚1 𝐚 T ̃
̃3 − 𝑅𝑖 (𝜂)𝐚1 𝐚′
𝑖 3
[ 𝟎T −𝑅𝑖′ (𝜂)𝐚T2 𝐚
̃3 ]
𝛿𝐩𝑖
𝛿𝐯𝑖 = [ ] 𝛿𝐯 = [𝛿𝐯1T 𝛿𝐯2T … 𝛿𝐯𝑛T ]T . (4.73)
𝛿𝐰𝑖
It follows that
By using Eq. (4.74) in Eq. (4.61), the internal virtual work 𝛿Π𝑖𝑛𝑡 is rewritten as
𝛿Π𝑖𝑛𝑡 = 𝛿𝐯 ∫ 𝐇 T 𝐌 d𝜉.
T
(4.75)
0
To obtain the external virtual work 𝛿Π𝑒𝑥𝑡 in a compact form, the distributed load
vector 𝐝, point load vector 𝐏, and the matrix 𝐋 are defined as
𝐝 = [𝐅 T 𝐐T ]T (4.76)
where 𝐈6 is an identity matrix of size 6 × 6. The external virtual work 𝛿Π𝑒𝑥𝑡 now can be
expressed compactly as
𝛿Π𝑒𝑥𝑡 = 𝛿𝐯 ∫ 𝐋T 𝐝 d𝜉 + 𝛿𝐯 T 𝐏.
T
(4.79)
0
By substituting Eqs. (4.75) and (4.79) into Eq. (4.55), the system equations are
obtained as
𝐿 𝐿
∫ 𝐇 T 𝐌 d𝜉 = ∫ 𝐋T 𝐝 d𝜉 + 𝐏. (4.80)
0 0
The system equations in Eq. (4.80) are nonlinear in terms of the unknown
kinematics. As a result, the linearization of the system equations is required in order that
the Newton-Raphson algorithm can be used to solve them. To linearize the system
equations, first the virtual work equation in Eq. (4.55) is linearized. After that, the
linearized system equations are obtained by discretizing the linearized virtual work
equation.
𝑛 𝑛
The vectors Δ𝐩𝑖 s and Δ𝐰𝑖 s are considered as the degrees of freedom. They are collected in
a vector Δ𝐯 as
Δ𝐩𝑖
Δ𝐯𝑖 = [ ] Δ𝐯 = [Δ𝐯1T Δ𝐯2T … Δ𝐯𝑛T ]T . (4.82)
Δ𝐰𝑖
The virtual work equation in Eq. (4.80) is linearized as (Dornisch, Müller &
Klinkel, 2016)
where D[𝛿Π]Δ𝐯 is the linear increment of the total virtual work 𝛿Π with respect to Δ𝐯 given
by
𝐿 𝐿
By using the above equation, the first integral term in Eq. (4.84) becomes
𝐿 𝐿 𝑛 𝑛 𝐿
T T
∫ 𝛿𝚪 Δ𝐌 d𝜉 = ∫ 𝛿𝚪 𝐃Δ𝚪 d𝜉 = ∑ ∑ 𝛿𝐯𝑖T (∫ 𝐇𝑖T 𝐃𝐇𝑗 d𝜉) Δ𝐯𝑗 . (4.86)
0 0 𝑖=1 𝑗=1 0
𝛿𝐚1T Δ𝐚1
𝛿𝐚1T Δ𝐚2 + 𝛿𝐚T2 Δ𝐚1 + 𝐚1T Δ𝛿𝐚2
𝛿𝐚1T Δ𝐚3 + 𝛿𝐚T𝟑 Δ𝐚1 + 𝐚1T Δ𝛿𝐚3
Δ𝛿𝚪 = . (4.87)
𝛿𝐚1T Δ𝐚′2 + (𝛿𝐚′2 )T Δ𝐚1 + 𝐚1T Δ𝛿𝐚′2
𝛿𝐚1T Δ𝐚′𝟑 + (𝛿𝐚′3 )T Δ𝐚1 + 𝐚1T Δ𝛿𝐚′𝟑
[𝛿𝐚T2 Δ𝐚′3 + (𝐚′3 )T Δ𝛿𝐚2 + (𝛿𝐚′3 )T Δ𝐚2 + 𝐚T2 Δ𝛿𝐚′3 ]
The increments of the kinematics in the above equation can be obtained by replacing, in
Eqs. (4.67) and (4.68), the symbol 𝛿 with the symbol Δ, i.e.,
𝑛 𝑛
Δ𝐚𝛼 = ∑ −𝑅𝑖 (𝜂)𝐚̃Δ𝐰
𝛼 𝑖 Δ𝐚′𝛼 = ∑[−𝑅𝑖′ (𝜂)𝐚̃Δ𝐰
𝛼
̃′
𝑖 − 𝑅𝑖 (𝜂)𝐚𝛼 Δ𝐰𝑖 ].
𝑖=1 𝑖=1
(4.89)
̃ Δ𝐰
Δ𝛿𝐚𝛼 = 𝛿𝐰 ̃ 𝐚𝛼 ̃′ )𝛿𝐰
Δ𝛿𝐚′𝛼 = (𝛿𝐰 ̃ 𝐚𝛼 + 𝛿𝐰 ̃′ )𝐚𝛼 + 𝛿𝐰
̃ (Δ𝐰 ̃ Δ𝐰
̃ 𝐚′𝛼 .
(4.90)
By substituting Eqs. (4.67)-(4.68), Eqs. (4.88)-(4.90) into Eq. (4.87), the second
integral term in Eq. (4.84) is then written as
𝐿 𝑛 𝑛 𝐿
T
∫ Δ𝛿𝚪 𝐌 d𝜉 = ∑ ∑ 𝛿𝐯𝑖T (∫ 𝐆𝑖𝑗 d𝜉) Δ𝐯𝑗 , (4.91)
0 𝑖=1 𝑗=1 0
𝐆𝑖𝑗,𝑢𝑢 𝐆𝑖𝑗,𝑢𝑤
𝐆𝑖𝑗 = [ ] (4.92)
𝐆𝑖𝑗,𝑤𝑢 𝐆𝑖𝑗,𝑤𝑤
′ T
+ [𝑅𝑖′ (𝜂)𝑅𝑗′ (𝜂)(𝐚
̃)
2
T
+ 𝑅𝑖′ (𝜂)𝑅𝑗 (𝜂)(𝐚̃
2) ]𝑀
12
(4.94)
T
+ [𝑅𝑖′ (𝜂)𝑅𝑗′ (𝜂)(𝐚
̃)
3
T
+ 𝑅𝑖′ (𝜂)𝑅𝑗 (𝜂)(𝐚̃
′
3) ]𝑀
13
+[𝑅𝑖′ (𝜂)𝑅𝑗 (𝜂)(𝐚2 𝐚1T ) − 𝑅𝑖′ (𝜂)𝑅𝑗 (𝜂)(𝐚1T 𝐚2 )𝐈 + 𝑅𝑖 (𝜂)𝑅𝑗′ (𝜂)(𝐚2 𝐚1T )
+[𝑅𝑖′ (𝜂)𝑅𝑗 (𝜂)(𝐚3 𝐚1T ) − 𝑅𝑖′ (𝜂)𝑅𝑗 (𝜂)(𝐚1T 𝐚3 )𝐈 + 𝑅𝑖 (𝜂)𝑅𝑗′ (𝜂)(𝐚3 𝐚1T )
𝑛 𝑛
where
𝑛 𝑛
The linearized virtual work equation in Eq. (4.83) now yields the linearized system
equations, i.e.,
𝐿 𝐿
𝐊Δ𝐯 = ∫ 𝐋 𝐟 d𝜉 + 𝐏 − ∫ 𝐇 T 𝐌 d𝜉.
T
(4.100)
0 0
𝑛
(𝑘+1) (𝑘) (𝑘) (𝑘+1) (𝑘+1)
𝐫𝑖 = 𝐫𝑖 + Δ𝐩𝑖 𝐫0 (𝜂) = ∑ 𝑅𝑖 (𝜂)𝐫𝑖 . (4.101)
𝑖=1
𝑛 𝑛
(𝑘) (𝑘) (𝑘)
Δ𝐰 = ∑ 𝑅𝑖 (𝜂)Δ𝐰𝑖 Δ𝐰 ′(𝑘)
= ∑ 𝑅𝑖′ (𝜂)Δ𝐰𝑖 . (4.102)
𝑖=1 𝑖=1
The rotation tensor and its derivative at the 𝑗-th Gauss point are then updated as
′ (𝑘)
𝚲𝑗′(𝑘+1) = exp[Δ𝐰
̃ (𝑘) (𝜂(𝑗) )] 𝚲𝑗 ̃ (𝑘) (𝜂(𝑗) )] 𝚲𝑗′(𝑘)
+ exp[Δ𝐰 (4.104)
where 𝜂(𝑗) is the 𝑗-th Gauss point. The determination of the derivative of the exponential
map of a skew-symmetric tensor is detailed in Appendix A. Once the rotation tensors at
the Gauss points are updated, the director vectors of the cross-sections at the Gauss points
are obtained by Eq. (4.9) and their derivatives are determined by Eq. (4.11).
𝑏ℎ3 ℎ𝑏 3
𝐼2 = 𝐼3 = (4.105)
12 12
where ℎ and 𝑏 are, respectively, the thickness and width of the cross-section. Note that, for
a rectangular cross-section, 𝐼23 is zero.
1 3
𝑏 𝑏4
𝐽 = ℎ𝑏 [ − 0.21 (1 − )] for ℎ ≥ 𝑏. (4.106)
3 ℎ 12ℎ4
In case 𝑏 > ℎ, the roles of ℎ and 𝑏 in the above expression are simply interchanged. Note
that, in the Gauss integration, it is found that good results can be obtained with 𝑝 + 1 Gauss
points for each distinct knot span.
The first example involves the examination of the locking effects, i.e., membrane
locking and shear locking, on the numerical results obtained by the present formulation. A
cantilever quadrant in 𝑂𝑋𝑌 plane, shown in Figure 4.4, is considered. A concentrated
moment is applied at the free end in the positive direction of the 𝑍 axis with the magnitude
of
4𝐸𝐼2
𝑀=√ (4.107)
𝜋𝑅
Figure 4.5 Cantilever quadrant subjected to an end moment: (a) Strain energy obtained
by various NURBS curves of different degrees. (b) Contribution of the membrane and
shear strain energy.
where 𝐸 is Young’s modulus. It is obvious that, for the given loading and boundary
conditions, there exists only the bending deformation. Therefore, this example is ideal for
the examination of the locking effects. In this example, only linear analysis is performed
by fully applying the moment 𝑀 in one step with one iteration.
The membrane, shear, and bending strains of the arch can be, respectively,
computed as (Adam, Bouabdallah, Zarroug & Maitournam, 2014)
d𝜃𝐀𝟑 (𝜉)
𝜒𝑏 (𝜉) = (4.110)
d𝜉
where 𝑢𝐀1 (𝜉) and 𝑢𝐀2 (𝜉) are the displacements of the arch axis in the direction of the unit
tangent vector 𝐀1 and the director vector 𝐀 2 , respectively. Moreover, 𝜃𝐀3 (𝜉) is the cross-
sectional rotation around the director vector 𝐀 3 .
In this example, the locking effects are studied by considering the strain energy of
the arch. The numerical strain energy 𝐸𝑛𝑢𝑚 of the arch can be computed as
𝜋 𝜋 𝜋
𝑅 𝑅 𝑅
2 2 2
1 1 1
𝐸𝑛𝑢𝑚 = ∫ 𝜀𝑚 𝐸𝐴𝜀𝑚 d𝜉 + ∫ 𝛾𝑠 𝐺𝐴𝛾𝑠 d𝜉 + ∫ 𝜒𝑏 𝐸𝐼𝜒𝑏 d𝜉 = 𝐸𝑚 + 𝐸𝑠 + 𝐸𝑏
2 2 2
0 0 0
(4.111)
where 𝐸𝑚 , 𝐸𝑠 , and 𝐸𝑏 are the membrane, shear, and bending strain energy, respectively. In
addition, 𝐺 is the shear modulus. The exact solution for the strain energy 𝐸𝑒𝑥𝑎𝑐𝑡 is given
by
𝑀2 𝜋𝑅
𝐸𝑒𝑥𝑎𝑐𝑡 = = 1. (4.112)
4𝐸𝐼2
The arch is analyzed with various values of the ratio 𝑅/ℎ ranging from 1𝑒1
(relatively thick beams) to 1𝑒5 (very thin beams). The beam is also represented using
NURBS curves of different degrees 𝑝 = 2, 3, 4, 5, 6. The number of control points is fixed
at 15 to distinguish between the errors due to discretization and locking. The computed
values of the strain energy 𝐸𝑛𝑢𝑚 are plotted in Figure 4.5a together with the exact solution.
It is observed that, when 𝑅/ℎ = 1𝑒1, the present formulation offers good results for all the
values of 𝑝. When 𝑅/ℎ = 1𝑒5, the quadratic and cubic NURBS curves provide inaccurate
results.
The locking effects can be further shown by examining the ratio of the membrane
and shear strain energy to the total strain energy 𝐸𝑛𝑢𝑚 . This ratio is defined as
𝐸𝑚 + 𝐸𝑠
𝐶𝑚𝑠 = . (4.113)
𝐸𝑛𝑢𝑚
The ratio 𝐶𝑚𝑠 is computed and plotted in Figure 4.5b. Since there is only bending strain in
the beam, it is expected that 𝐶𝑚𝑠 is zero. It can be seen that, when 𝑅/ℎ is in the range of
thin beams, the majority of strain energy is due to the membrane and shear deformations
with the use of the quadratic and cubic NURBS curves. In contrast, for all values of 𝑅/ℎ,
the contribution of the membrane and shear deformations is entirely negligible with the use
of the NURBS curves of higher degrees. These results clearly reveal that the locking effects
happen when the quadratic and cubic NURBS curves are used.
In conclusion, the obtained numerical results show that the present formulation is
not locking-free, but the effects can be effectively reduced by using NURBS curves of high
degree for the approximations of the geometry and unknown kinematics.
The second example deals with the analysis of a quarter circular arch with a pre-
twisted configuration shown in Figure 4.6. In the same figure, the information of the cross-
sectional dimensions and material properties are also given. In the initial configuration, the
Figure 4.7 Pre-twisted circular arch – Convergence tests for the deformation of the
arch axis.
arch axis lies in the 𝑂𝑋𝑌 plane while its cross-section rotates around the arch axis. The
initial twisted configuration is described by the angle 𝜙, defined in Figure 4.2, as
𝜋
𝜙 = (1 − 𝜂) . (4.114)
2
All the translational displacements and the axial rotations at both ends of the arch are fixed.
The arch is subjected to a uniformly distributed force 𝑞𝑍 = 1 kN/m in the negative direction
of the 𝑍 axis. This example is initially introduced by Greco and Cuomo (2013) and recently
considered by Radenković and Borković (2018). Once again, only linear analysis is
performed by fully applying the load in one step with one iteration. The excellent
convergence property of the IGA approach is one of its most prominent features. Here, this
property in the present formulation is verified thoroughly. As indicated by Bazilevs, Beirão
Da Veiga, Cottrell, Hughes and Sangalli (2006) and Greco and Cuomo (2013), when
NURBS curves of degree 𝑝 are used for the discretization of the geometry and unknown
kinematics, the minimum expected order of convergence is 𝑝 + 1 for displacements. As
Figure 4.8 Pre-twisted circular arch: (a) Displacement 𝑢𝑋 . (b) Displacement 𝑢𝑌 . (c)
Displacement 𝑢𝑍 . (d) Reference and deformed configurations.
similar to Greco and Cuomo (2013), in this study, the order of convergence with respect to
the number of control points is considered. To test the convergence, the following relative
error is used, i.e.,
𝐿
∫0 ‖𝐫𝑛𝑢𝑚 − 𝐫𝑟𝑒𝑓 ‖𝑑𝜉
Relative error = 𝐿 (4.115)
∫0 ‖𝐫𝑟𝑒𝑓 ‖𝑑𝜉
where 𝐫𝑛𝑢𝑚 is the deformed arch axis obtained by using NURBS curves of various degrees
to approximate the geometry and the unknown kinematics, and 𝐫𝑟𝑒𝑓 is the deformed arch
axis obtained by using sextic NURBS curves with 100 control points. Figure 4.7 shows the
results of the convergence tests. The horizontal axis indicates the number of control points
while the vertical axis indicates the relative error. Each curve in the plot is associated with
a fixed NURBS degree 𝑝. It can be observed that the minimum expected orders of
convergence of 𝑝 + 1 are achieved by all the considered NURBS degrees.
This example studies a straight cantilever beam subjected to an end moment shown
in Figure 4.9. The problem is considered by many studies as a benchmark problem for
Figure 4.10 Rolled-up straight cantilever beam – Single circle test: (a) Convergence
tests for the position of the free end. (b) Normalized load-displacement curves of the
free end. (c) Reference and final deformed configurations.
testing geometrically nonlinear beam formulations. This is because the beam exhibits very
large displacements and rotation. In addition, only the bending deformation occurs in the
beam and, because of that, the exact solutions can be easily derived. Since the beam axis
is straight and not a non-degenerate curve, the unit normal vector 𝐍 and the unit binormal
vector 𝐁 by the Frenet-Serret formulas are not defined. Subsequently, the director vectors
𝐀 2 and 𝐀 3 cannot be determined from Eq. (4.5). Here, an alternative method is used to
define 𝐍 and 𝐁. There always exists a rotation tensor 𝚲0 , such that (Crisfield, 2000)
𝐀1 = 𝚲0 𝐞1 . (4.116)
Note that 𝐀1 is the unit tangent vector of the reference beam axis. The vectors 𝐍 and 𝐁 are
then determined as
𝐍 = 𝚲 0 𝐞2 𝐁 = 𝚲 0 𝐞3 . (4.117)
The director vectors 𝐀 2 and 𝐀 3 can then be determined from 𝐍 and 𝐁 by rotating them
around 𝐀1 with an angle 𝜙 using Eq. (4.5). In this study, for a straight beam, 𝐞1 is always
set along the initial beam axis and used as 𝐀1 , as shown in Figure 4.9.
The exact solutions for the displacements in the directions of the 𝑋 and 𝑌 axes of
the free end can be obtained, respectively, as
𝑢𝑋 sin 𝜔 𝑢𝑌 1 − cos 𝜔 𝑀𝐿
= −1 = 𝜔= . (4.118)
𝐿 𝜔 𝐿 𝜔 𝐸𝐼2
In this study, the beam is loaded up to two values of 𝑀, i.e., 2𝜋𝐸𝐼2 /𝐿 and 4𝜋𝐸𝐼2 /𝐿. The
first test is referred to as the single circle test while the second test is referred to as the
double circle test. In the single circle test, the beam is perfectly bent into a circle while in
the double circle test, the beam is wound around itself twice.
Table 4.1 Rolled-up straight cantilever beam – Single circle test: comparison of the
number of degrees of freedom.
Present 1 48
Figure 4.11 Rolled-up straight cantilever beam – Double circle test: (a)Convergence
tests of the position of the free end. (b) Normalized load-displacement curves of the
free end. (c) Reference and deformed configurations.
The convergence tests for the Euclidean norm ‖𝐫𝑒𝑛𝑑 ‖ of the free end at 𝑀 =
2𝜋𝐸𝐼2 /𝐿 are performed, and their results are plotted in Figure 4.10a. The orders of
convergence by all the used NURBS degrees are better than the minimum expected orders
of convergence of 𝑝 + 1.
Next, the beam is represented by septic NURBS curves with 8 control points. This
corresponds to 48 degrees of freedom. The displacements of the free end obtained by the
present formulation are plotted in Figure 4.10b together with the exact solutions. The
present results match well with the exact solutions. The deformed configuration of the
beam at the final load obtained in this study is shown in Figure 4.10c. It can be seen that
The results of the convergence tests for the double circle test are depicted in Figure
4.11a. The orders of convergence by all the employed NURBS degrees are found to be
better than the minimum expected orders of convergence of 𝑝 + 1.
Next, the beam is represented by sextic NURBS curves with 15 control points. This
corresponds to 90 degrees of freedom. The obtained displacements of the free end are
plotted in Figure 4.11b in comparison with the exact solutions. Good agreement between
the present results and the exact solutions is found. In addition, the deformed configurations
at some load steps, from this study, are visualized in Figure 4.11c. Very smooth
representations of these configurations are observed.
Figure 4.13 Lateral buckling of a cantilever beam: (a) Convergence tests for the
deformation of the beam axis. (b) Load-displacement curves of the free end. (c)
Reference and deformed configurations.
The accuracy of the present formulation for analysis of beams under very large
displacements is demonstrated in the previous example. However, the previous problem is
a plane problem and involves only a two-dimensional displacement field. In this example,
the accuracy of the present formulation is further illustrated by using the buckling analysis
of a cantilever beam in Figure 4.12, which involves a three-dimensional displacement field.
The beam is subjected to a concentrated force 𝑃 at the free end. The extreme slenderness
of the beam cross-section, i.e., ℎ/𝑏 = 20, must be noted. The lateral buckling of the beam
is induced by a perturbation lateral force of 0.001𝑃 as shown in Figure 4.12.
The convergence tests for the relative error defined in Eq. (4.115) at the final load
of 𝑃 = 2000 are performed. In this example, 𝐫𝑟𝑒𝑓 is the deformed beam axis obtained by
using sextic NURBS curves with 100 control points. The results of the convergence tests
are plotted in Figure 4.13a. It can be observed that the orders of convergence that are better
than the minimum expected orders of convergence of 𝑝 + 1 are obtained by all the used
NURBS degrees.
This beam is previously analyzed by Smoleński (1999), where it is split into ten
two-noded beam elements. To compare with the results obtained by Smoleński (1999), the
beam is represented in this study by sextic NURBS curves with 20 control points. The free-
end displacements obtained by this study and the work of Smoleński (1999) are compared
in Figure 4.13b. The comparison shows good agreement between the two solutions. The
deformations of the beam at some load steps, from this study, are also shown in Figure
4.13c.
The examples considered so far involve the beam structures which can be
represented by only one NURBS patch. In fact, the presence of the rotational degrees of
freedom in the present formulation allows analysis of multi-patch beam structures to be
performed easily. This capability is shown in this example where twisting of a circular ring,
shown in Figure 4.14, is considered. The ring is fixed at the point 𝐵 and subjected a twisting
moment 𝑀 at the point 𝐶. The relation between the angle 𝜃 around the 𝑋 axis and the
moment 𝑀 at 𝐶 is investigated. The ring is first introduced by Goto, Watanabe, Kasugai
and Obata (1992). It is numerically solved by Pai and Palazotto (1996) using the multiple
shooting method and by Smoleński (1999) using the finite element method, among others.
In this study, the whole ring is analyzed. The ring is split into four NURBS patches,
and each patch is represented by septic NURBS curves. The total number of control points
used in this study for the whole ring is 28, which is equivalent to 15 control points if only
half of the ring is analyzed. Note that 15 control points correspond to 90 degrees of
freedom. Goto, Watanabe, Kasugai and Obata (1992) simulate half of the ring with two
hundred beam elements with 1206 degrees of freedom, and Smoleński (1999) model half
of the ring using forty eight beam elements with 294 degrees of freedom. Figure 4.15 shows
the numerical results for the twisting angle 𝜃 at 𝐶, obtained by the present formulation and
the works of Goto, Watanabe, Kasugai and Obata (1992) and Smoleński (1999). The figure
also shows the numerical results obtained by Pai and Palazotto (1996) as reference results.
The numerical results from this study and Smoleński (1999) are virtually identical with the
reference results. Although many beam elements are used by Goto, Watanabe, Kasugai and
Obata (1992), the difference with the reference results is still remarkably noticeable. This
is probably due to the fact that the initial curvature of the ring is not well represented by
the geometry of the beam elements.
The deformed configurations of the ring are visualized in Figure 4.16 at every 90
degrees of the twisting angle 𝜃. It can be observed that the ring undergoes very large three-
dimensional deformations. Indeed, at the final equilibrium state considered in this study,
the ring is transformed into a smaller ring with a diameter of one third of its original size.
With this example, the excellent efficiency and accuracy of the present formulation for
analysis of multi-patch beam structures subjected to very large three-dimensional
deformations are validated.
CHAPTER 5
It is well-known that shear effects are neglected in the Euler-Bernoulli beam theory,
and as a result, beam cross-sections are always orthogonal to the beam axis during
deformation. To derive a new Euler-Bernoulli beam formulation, some modifications are
performed on the kinematic descriptions of Timoshenko beams in Chapter 4 to impose the
orthogonality between the cross-sections and the beam axis.
Mathematically, the orthogonality between the cross-sections and the beam axis
can be expressed as
𝐀 𝛼 ⋅ 𝐀1 = 0 𝐚𝛼 ⋅ 𝐚1 = 0 (5.1)
and consequently,
By definition, the orthogonality between the cross-sections and the beam axis in the
reference configuration is automatically met. In the current configuration, an alternative
parametrization of the rotation tensor is presented to fulfil this orthogonality requirement.
5.1.1 Smallest rotation (SR) mapping and director vectors of Euler-Bernoulli beams
In the current configuration, the unit tangent vector of the beam axis is computed
as
𝐚1 𝐚1
𝐭= = . (5.3)
‖𝐚1 ‖ 𝑎1
̅̅̅̅
𝐀2 ⋅ 𝐭 ̅̅̅̅
𝐀3 ⋅ 𝐭
𝐧 = ̅̅̅̅
𝐀2 − (𝐭 + ̅̅
𝐀̅̅1 ) 𝐛 = ̅̅̅̅
𝐀3 − (𝐭 + ̅̅
𝐀̅̅1 ). (5.4)
1+𝐀 ̅̅̅̅1 ⋅ 𝐭 1+𝐀 ̅̅̅̅1 ⋅ 𝐭
Since ̅̅
𝐀̅̅1 , ̅̅̅̅
𝐀 2 , and ̅̅̅̅
𝐀 3 are rotated with the same rotational angle and direction, the vectors
𝐭, 𝐧, and 𝐛 in the above equations are also orthonormal. In analogous to the expressions of
the director vectors 𝐀 2 and 𝐀 3 in the reference configuration, the director vectors 𝐚2 and
𝐚3 are determined by a quasi-2D rotation of the vectors 𝐧 and 𝐛 around the vector 𝐭 with a
correction angle 𝜃. The resulting vectors of this rotation are given by
By denoting the rotation tensor associated with the mapping expressed via Eqs. (5.3)-(5.5)
as 𝚲𝐸𝐵 , the following compact relations can be written as
𝐭 = 𝚲𝐸𝐵 ̅̅̅̅
𝐀1 𝐚𝛼 = 𝚲𝐸𝐵 ̅̅̅̅
𝐀𝛼 . (5.7)
It can be easily verified that the constraint for the current configuration in Eq. (5.1)
̅̅̅̅1 , ̅̅̅̅
is satisfied by 𝚲𝐸𝐵 . Since the set of vectors {𝐀 𝐀 2 , ̅̅̅̅
𝐀 3 } is given in advance, 𝚲𝐸𝐵 is entirely
identified by the vector 𝐭 and the axial rotation angle 𝜃. Recall that the vector 𝐭 can be
determined by the current beam axis 𝐫0 (𝜉) via Eq. (5.3). Therefore, for the proposed Euler-
Bernoulli beam formulation, the current beam axis 𝐫0 (𝜉) and the axial rotation angle 𝜃 are
considered as the primary kinematics.
̅̅̅̅1 , ̅̅̅̅
The method for obtaining suitable choices of {𝐀 𝐀 2 , ̅̅̅̅
𝐀 3 }, proposed by Meier,
Popp and Wall (2014), is adopted. In this study, the undeformed configuration is
considered as the reference configuration. Generally, at a point on the current beam axis
with the convective coordinate 𝜉, the unit tangent vector 𝐀1 and the director vectors 𝐀 2
and 𝐀 3 , determined at the same 𝜉 on the reference beam axis, can be chosen as the reference
̅̅̅̅1 , ̅̅̅̅
vectors {𝐀 𝐀 2 , ̅̅̅̅
𝐀 3 }. However, in some cases when the vectors 𝐀1 and 𝐭 are antiparallel,
̅̅̅̅1 , ̅̅̅̅
this choice of {𝐀 𝐀 2 , ̅̅̅̅
𝐀 3 } results in a singularity because 1 + 𝐀1 ⋅ 𝐭 = 0 and the relations
in Eq. (5.4) are thus undefined. This situation may happen when significantly large
deformations occur between the reference and current configurations. For these cases, the
unit tangent vector 𝐚1 and the director vectors 𝐚2 and 𝐚3 of a known configuration at a few
̅̅̅̅1 , ̅̅̅̅
loading steps prior to the current one can be used as {𝐀 𝐀 2 , ̅̅̅̅
𝐀 3 }, instead of {𝐀1 , 𝐀 2 , 𝐀 3 }.
The variations of the beam axis 𝐫0 (𝜉) and the rotation tensor 𝚲𝐸𝐵 are determined
as
𝛿𝐭 = 𝛿𝐰 × 𝐭 (5.9)
For later use, a relation between the vector 𝛿𝐰 and the variations of the primary kinematics
is given by (Meier, Popp & Wall, 2014)
𝐭 ⊗ ̅̅
𝐀̅̅1 𝐭 × 𝛿𝐚1
𝛿𝐰 = 𝐭𝛿𝜃 + (𝐈 − ) . (5.11)
1 + ̅̅
𝐀̅̅1 ⋅ 𝐭 𝑎1
(𝐭 + ̅̅̅̅
𝐀1 ) ⊗ ̅̅̅̅
𝐀2 + (𝐀 ̅̅̅̅2 ⋅ 𝐭)𝐈 (𝐀
̅̅̅̅2 ⋅ 𝐭)(𝐭 + ̅̅̅̅
𝐀1 ) ⊗ ̅̅̅̅
𝐀1
𝛿𝐧 = [− + ̅̅̅̅2 , 𝐭)𝛿𝐭
] 𝛿𝐭 = 𝐎(𝐀 (5.12)
̅̅̅̅
1 + 𝐀1 ⋅ 𝐭 ̅̅̅̅
(1 + 𝐀1 ⋅ 𝐭) 2
(𝐭 + ̅̅̅̅
𝐀1 ) ⊗ ̅̅̅̅
𝐀3 + (𝐀 ̅̅̅̅3 ⋅ 𝐭)𝐈 (𝐀
̅̅̅̅3 ⋅ 𝐭)(𝐭 + ̅̅̅̅
𝐀1 ) ⊗ ̅̅̅̅
𝐀1
𝛿𝐛 = [− + ̅̅̅̅3 , 𝐭)𝛿𝐭.
] 𝛿𝐭 = 𝐎(𝐀 (5.13)
̅̅̅̅
1 + 𝐀1 ⋅ 𝐭 ̅̅̅̅
(1 + 𝐀1 ⋅ 𝐭) 2
𝐾12 = 𝐀T2 𝐀′1 − 𝐚T2 𝐚1′ 𝐾13 = 𝐀T3 𝐀′1 − 𝐚T3 𝐚1′ . (5.14)
Since the direct shear strains 𝛤12 and 𝛤13 vanish, the shear forces 𝑁12 and 𝑁13 also vanish.
A generalized strain vector 𝚪 and a generalized stress vector 𝐌 are defined for Euler-
Bernoulli beams as
Their relation can be obtained by using Eqs. (4.37), (4.40), (4.41), and (4.43) as
𝐸̅ 𝐴 0 0 0
0 𝐸̅ 𝐼2 𝐸̅ 𝐼23 0
𝐌 = 𝐃𝚪 𝐃= . (5.16)
0 𝐸̅ 𝐼23 𝐸̅ 𝐼3 0
[0 0 0 𝐺̅ 𝐽]
where 𝛿Π𝑖𝑛𝑡 and 𝛿Π𝑒𝑥𝑡 are the internal and external virtual works, respectively. The
internal virtual work 𝛿Π𝑖𝑛𝑡 can be expressed as
where 𝑉 and 𝐿 are the volume and length of the reference beam configuration, respectively.
(5.19)
where 𝐅 and 𝑀 are the distributed force and torsional moment. In addition, 𝐟(0) and 𝐟(𝐿)
are the concentrated forces applied at 𝜉 = 0 and 𝜉 = 𝐿, respectively. The same
interpretation holds for the concentrated torsional moments 𝑚(0) and 𝑚(𝐿).
The approximations of the beam axis in the reference and current configurations
are expressed as
𝑛 𝑛
𝑛 𝑛
𝑛 𝑛
The axial rotation angle 𝜃 is approximated in the same manner as the beam axis as
𝑛 𝑛
Eventually, there are four control variables at each control point, i.e., the displacement
vector 𝐩𝑖 having three components and the axial rotation angle 𝜃𝑖 . Vectors of control
variables are defined as
𝑛 𝑛
𝑛 𝑛
As discussed in Chapter 4, the two parameterizations of the beam axis, i.e., arc-
length parameter 𝜉 and knot parameter 𝜂, can be related by using Eq. (4.69) as
d𝐑 0 (𝜂)
d𝜉 = ‖ ‖ d𝜂 = ‖𝐑̇ 0 (𝜂)‖d𝜂 = 𝐼d𝜂. (5.27)
d𝜂
It follows that
(5.28)
By using Eq. (5.11), the vector 𝛿𝐰 can be determined from the primary kinematics and
their variations. Substituting Eqs. (5.11), (5.25), and (5.26) into Eq. (5.29) yields
𝑅𝑖′ (𝜂)𝐚1T 0
′ ′ T
𝑅𝑖 (𝜂)(𝐚1 ) 𝐚 ̃𝐗(𝐀
2
̅̅̅̅1 , 𝐭) − 𝑅𝑖 (𝜂)𝐚2
′′ T
𝑅𝑖 (𝜂)(𝐚1′ )T 𝐚̃𝐭
2
𝐇𝑖 = (5.31)
′ ′ T
𝑅𝑖 (𝜂)(𝐚1 ) 𝐚 ̃𝐗(𝐀
3
̅̅̅̅
1 , 𝐭) − 𝑅 ′′
𝑖 (𝜂)𝐚 T
3 𝑅𝑖 (𝜂)(𝐚 ′
1 ) T
̃𝐭
𝐚3
′
[−𝑅𝑖 (𝜂)𝐭 ̅̅
T ′ (𝐀 ̅̅
𝐗 1 , 𝐭) − 𝑅𝑖 ′′
(𝜂)𝐭 T ̅̅̅̅
𝐗(𝐀1 , 𝐭) −𝑅𝑖′
(𝜂) ]
𝐇 = [𝐇1 𝐇2 … 𝐇𝑛 ]. (5.32)
Regarding the external virtual work 𝛿Π𝑒𝑥𝑡 , the following matrices are introduced
as
𝐝 = [𝐅 T 𝑀]T (5.34)
𝛿Π𝑒𝑥𝑡 = 𝛿𝐯 T ∫ 𝐂 T 𝐝 d𝜉 + 𝛿𝐯 T 𝐏. (5.37)
0
Substituting Eqs. (5.33) and (5.37) into Eq. (5.17) yields the system equations as
𝐿 𝐿
∫ 𝐇 𝐌 d𝜉 = ∫ 𝐂 T 𝐝 d𝜉 + 𝐏.
T
(5.38)
0 0
The system equations are nonlinear in terms of the control variables, and they are solved
using the Newton-Raphson algorithm. Their linearization is given below.
Recall that the external loads are assumed to be deformation independent in this
study, the linearization of the total virtual work 𝛿Π is expressed as
where D[𝛿Π]Δ𝐯 is the linear increment of the total virtual work 𝛿Π with respect to Δ𝐯. This
increment can be determined as
𝐿 𝐿
In the first term of the right-hand side of Eq. (5.40), Δ𝐌 can be obtained as
It follows that
𝐿 𝐿
∫ 𝛿𝚪 Δ𝐌 d𝜉 = 𝛿𝐯 (∫ 𝐇 T 𝐃𝐇 d𝜉) Δ𝐯.
T T
(5.42)
0 0
In the second term of the right-hand side of Eq. (5.40), Δ𝛿𝚪 can be expressed as
(5.43)
The matrices 𝐋11, 𝐋12, 𝐋13, and 𝐋23 are given in Appendix D. Using these matrices, the
second term in the right-hand side of Eq. (5.40) can be written as
𝐿 𝐿
∫ Δ𝛿𝚪 T 𝐌 d𝜉 = 𝛿𝐯 T [∫(𝐋11𝑁11 + 𝐋12 𝑀12 + 𝐋13 𝑀13 + 𝐋23 𝑀23 ) d𝜉] Δ𝐯. (5.46)
0 0
Substituting Eqs. (5.42) and (5.46) into Eq. (5.40) yields an alternative expression
of the linear increment D[𝛿Π]Δ𝐯 as
D[𝛿Π]Δ𝐯 = 𝛿𝐯 [∫(𝐇 T 𝐃𝐇 + 𝐋11 𝑁11 + 𝐋12 𝑀12 + 𝐋13 𝑀13 + 𝐋23 𝑀23 ) d𝜉] Δ𝐯 = 𝛿𝐯 T 𝐊Δ𝐯
T
(5.47)
where 𝐊 is referred to as the tangent stiffness matrix. Substituting Eqs. (5.33), (5.37), and
(5.47) into Eq. (5.39) finally gives
𝐿 𝐿
With the help of Eqs. (5.9), (5.10), (5.12), and (5.13), Eqs. (5.49) and (5.50) yield
(5.53)
It will be shown shortly that this relation allows the consideration of the total cross-
sectional rotations at the ends of patches as discrete unknowns.
Consider the second and second-last control points, i.e., 𝐫2 and 𝐫𝑛−1. Their position
vectors can be determined as
Here, 𝐭1 and 𝐭 𝑛 are, respectively, the unit tangent vectors at the first and last ends of the
patch. In addition, the lengths of the first and last segments of the control polygon, 𝑙1 and
𝑙𝑛−1, are given by
𝛿𝐩2 = 𝛿𝐩1 + 𝐭1 𝛿𝑙1 − 𝑙1 𝐭̃1 𝛿𝐰1 𝛿𝐩𝑛−1 = 𝛿𝐩𝑛 − 𝐭 𝑛 𝛿𝑙𝑛−1 + 𝑙𝑛−1 𝐭̃𝑛 𝛿𝐰𝑛 .
(5.56)
By using Eqs. (5.53) and (5.56), the following transformations are obtained, i.e.,
𝛿𝐩1 𝐈 𝟎3 𝟎 𝟎 𝛿𝐩1
𝛿𝜃 𝟎T 𝐕(0) 0 0 𝛿𝐰1
[ 1] = [ ] (5.57)
𝛿𝐩2 𝐈 −𝑙1 𝐭̃1 𝐭1 0 𝛿𝑙1
𝛿𝜃2 [𝟎T 𝟎T 0 1] 𝛿𝜃2
Define a new vector of degrees of freedom 𝐱, via its variation, and a new point load
vector 𝐏𝐱 as
𝛿𝐱
= [𝛿𝐩1T 𝛿𝐰1T 𝛿𝑙1 𝛿𝜃2 𝛿𝐩T3 𝛿𝜃3 … 𝛿𝐩𝑛−2 𝛿𝜃𝑛−2 𝛿𝜃𝑛−1 𝛿𝑙𝑛−1 𝛿𝐩T𝑛 𝛿𝐰𝑛T ]T
(5.59)
where 𝐦(0) and 𝐦(𝐿) are the concentrated moments applied at 𝜉 = 0 and 𝜉 = 𝐿,
respectively. With the help of Eqs. (5.57) and (5.58), the following transformations
between the control variables and the degrees of freedom can be obtained as
The internal and external virtual works, i.e., 𝛿Π𝑖𝑛𝑡 and 𝛿Π𝑒𝑥𝑡 , can be alternatively
expressed in terms of the degrees of freedom as
𝐿 𝐿
𝛿Π𝑖𝑛𝑡 = 𝛿𝐯 T ∫ 𝐇 T 𝐌 d𝜉 = 𝛿𝐱 T 𝐓 T ∫ 𝐇 T 𝐌 d𝜉 (5.62)
0 0
𝐿 𝐿
𝛿Π𝑒𝑥𝑡 = 𝛿𝐯 T ∫ 𝐂 T 𝐝 d𝜉 + 𝛿𝐯 T 𝐏 = 𝛿𝐱 T 𝐓 T ∫ 𝐂 T 𝐝 d𝜉 + 𝛿𝐱 T 𝐏𝐱 . (5.63)
0 0
𝐿 𝐿
𝐓 ∫ 𝐇 𝐌 d𝜉 = 𝐓 ∫ 𝐂 T 𝐝 d𝜉 + 𝐏𝐱
T T T
(5.64)
0 0
𝐿 𝐿
Figure 5.2 Square frame subjected to opposite compression forces: (a) Problem set up.
(b) Equivalent model.
whose moments of inertia and polar moment of inertia are, respectively, determined by
using Eqs. (4.105) and (4.106).
All the examples in this section are analyzed using NURBS curves with full inner-
element continuity.
In the first example, the convenience of using the proposed formulation to deal with
rotational boundary conditions is illustrated. Figure 5.2a shows a square frame subjected
to a pair of opposite compression forces. Due to the symmetry, only a quarter of the frame
is analyzed and the equivalent model used is visualized in Figure 5.2b. In the equivalent
model, two sextic NURBS curves with 13 control points, i.e., 𝐑1 to 𝐑13 , are used to
represent the two members.
Figure 5.3 Square frame: (a) Normalized load-displacement curves. (b) Deformed
configurations.
the 𝑖th control point. At 𝐵 and 𝐶, only the horizontal displacement 𝑢𝑋1 and the vertical
displacement 𝑢𝑌13 are not restrained. Since the displacements of 𝐑1 and 𝐑13 are considered
as degrees of freedom, the restrained displacements at 𝐵 and 𝐶 can be straightforwardly
enforced as
On the contrary, the enforcement of the rotational boundary conditions at 𝐵 and 𝐶 is more
involved. In fact, due to the lack of rotational degrees of freedom, the following relations
should be used, i.e.,
These constraints are determined based on the conditions that the tangent vectors at 𝐵 and
𝐶 must retain their original directions. The Lagrange multiplier approach can be used to
implement these constraints. However, it should be noticed that the above relations are
determined based on the initial geometry of the frame. In other words, the constraints used
to prescribe the rotational boundary conditions are unfortunately problem-dependent.
Figure 5.4 Lee’s frame: (a) Problem set up. (b) Normalized load-displacement curves
of the point 𝐷. (c) Deformed frame axes.
In this example, the accuracy and robustness of the proposed formulation is further
examined by analyzing Lee's frame shown in Figure 5.4a. This example is widely
considered as a benchmark problem for geometrically nonlinear analysis of frames, and the
exact solutions are derived by Lee, Manuel and Rossow (1968). The frame is pinned at 𝐵
and 𝐹, and a concentrated force is prescribed at point 𝐷. The horizontal and downward
displacements at the point 𝐷, i.e., 𝑢 and 𝑣, are considered for result comparisons.
The frame consists of three segments, i.e., 𝐵𝐶, 𝐶𝐷, and 𝐷𝐹. In order to maintain
the right angle between the segments 𝐵𝐶 and 𝐶𝐷 and thereby enforce the rigid connection
at 𝐶, some special treatments must be used for isogeometric Euler-Bernoulli beam
formulations that do not include rotational degrees of freedom, e.g., adding a bending strip
at 𝐶 (Huang, He, Jiang, Qiao & Wang, 2016). With the proposed formulation, no special
treatments are required. The frame is analyzed by Simo and Vu-Quoc (1986) using ten
quadratic beam elements with 63 in-plane degrees of freedom. Most recently, the frame is
modeled by Yuan, Wang and Kardomateas (2019) using two ten-noded co-rotational
quadrature beam elements with 57 degrees of freedom.
Figure 5.6 A pre-twisted circular arch: (a) Top view. (b) NURBS representations of
the arch and control points.
In this study, each segment is represented by a septic NURBS curve with 22 control
points or 42 in-plane degrees of freedom in total. A comparison between the results
obtained by the proposed formulation and the exact solutions is given in Figure 5.4b. It can
be observed that the normalized load-displacement curves exhibit severe snap-though and
snap-back behaviors. The obtained excellent agreement between the present results and the
exact solutions validates the accuracy and robustness of the proposed formulation for
stability analysis of beam structures. Deformed configurations of the frame at some loading
steps are also visualized in Figure 5.4c, showing the tangent vectors at 𝐶 and 𝐷 that are
determined separately from each segment. Even with very large displacements, the rigid
connections of patches are perfectly enforced.
𝛼 𝛼 − 130
𝜙=− × 90, (0 ≤ 𝛼 ≤ 65) 𝜙= × 90, (65 ≤ 𝛼 ≤ 130) (5.68)
65 65
where 𝛼 is the angular coordinate of the arch (see Figure 5.6a). Here, the correction angles
are computed in degrees. By using NURBS representations for the axes and the above
correction angles, the exact geometric representation of the arch is easily obtained. On the
contrary, the simulation of this arch is not a trivial task when the conventional finite
element approach is used. In fact, this arch can be simulated by using many straight beam
elements, each with a constant cross-section. To reduce the errors of the geometric
approximation and to ensure the result accuracy, a dense mesh of finite elements should be
used. Figure 5.7 shows a screenshot of the arch in Abaqus with 600 B33 beam elements.
However, specifying the correct orientations of a large number of elements is highly time-
consuming. Additionally, a smooth geometric representation is not obtained, which can be
clearly seen in the magnified figure in Figure 5.7.
Along its whole axis, the arch is subjected to a uniformly distributed force in the
negative direction of the 𝑍-axis. The arch is also fixed at both ends, i.e., at 𝐵 and 𝐷. As
Figure 5.8 A pre-twisted circular arch – Convergence test for the relative error in the
deformation of the arch axis.
discussed in the first example, the boundary conditions relating the displacements at the
arch ends can be straightforwardly prescribed. However, without rotational degrees of
freedom, some additional constraints are required to enforce the rotational boundary
conditions at 𝐵 and 𝐷. Figure 5.6b shows a NURBS representation of the arch axis with
Figure 5.9 A pre-twisted circular arch: (a) Displacement 𝑢𝑋 . (b) Displacement 𝑢𝑌 . (c)
Displacement 𝑢𝑍 . (d) Reference and deformed configurations (with a scale factor of
1000 for displacements).
11 control points, i.e., 𝐑1 to 𝐑11 . The displacements along the coordinate axes of the 𝑖th
control point are denoted by 𝑢𝑋𝑖 , 𝑢𝑌𝑖 , and 𝑢𝑍𝑖 , respectively. If rotational degrees of freedom
are not used, the additional constraints for the arch in this example are given by
Once again, the constraints above are problem-dependent since they are determined based
on the initial geometry of the arch. For spatial beam structures with several complex
130
𝜋 𝑟𝑒𝑓
∫0180 ‖𝐫0𝑛𝑢𝑚 − 𝐫0 ‖ d𝜉
𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒𝐸𝑟𝑟 = 130 . (5.71)
𝜋 𝑟𝑒𝑓
∫0180 ‖𝐫0 ‖ d𝜉
𝑟𝑒𝑓
Here, 𝐫0 denotes the deformed axis, obtained by using two septic NURBS curves with
100 control points, and 𝐫0𝑛𝑢𝑚 denotes the deformed axis obtained by the NURBS curve
being tested. In this test, the relative error is considered to be dependent on the number of
control points. The resulting relative errors are plotted in Figure 5.8. In the figure,
quadratic, cubic, and quartic NURBS curves, i.e., 𝑝 = 2, 3, and 4, are examined. It is
observed that the minimum expected order of 𝑝 + 1 is obtained by using cubic and quartic
NURBS curves.
Lastly, the accuracy of the proposed formulation is verified. Two septic NURBS
curves with 15 control points or 58 degrees of freedom are used for the analysis. Since no
reference results are found in the literature, the arch is analyzed with the beam formulation
presented by Vo, Nanakorn and Bui (2020) using two septic NURBS curves and 100
control points. Furthermore, the arch is also simulated in Abaqus with 600 B33 beam
elements. The displacements of the arch axis, i.e., 𝑢𝑋 , 𝑢𝑌 , and 𝑢𝑍 , are plotted in Figure
5.9a-c. Very good agreement is observed between the results obtained by the proposed
formulation and those using the formulation proposed by Vo, Nanakorn and Bui (2020)
and Abaqus. Additionally, the reference and deformed configurations of the arch are
Figure 5.10 Twisting of a circular ring: (a) Problem set up. (b) Moment-rotation curve
of the point 𝐷.
visualized in Figure 5.9d. Both the reference and deformed configurations are represented
with very smooth geometric representations.
This example is devoted to illustrating the accuracy and efficiency of the proposed
formulation for analysis of beam structures subjected to very large three-dimensional
displacements and rotations. A circular ring shown in Figure 5.10a is considered. The ring
is fixed at point 𝐵 and subjected to a concentrated moment 𝑀 around the 𝑋-axis at point
𝐷. A solution for the relation between the moment 𝑀 and the rotation angle 𝛼 is derived
by Pai and Palazotto (1996) using shooting methods, and this solution is used as the
reference. For an easy comparison, the same geometrical and material properties with those
in the work of Pai and Palazotto (1996) are used in this study, as provided in Figure 5.10a.
Additionally, for a better discussion, the term “connection” is used to denote a connection
between patches in IGA and between elements in the conventional finite element approach.
The ring has been analyzed in several published works, e.g., (Goto, Watanabe,
Kasugai & Obata, 1992; Meier, Popp & Wall, 2014; Smoleński, 1999; Vo, Nanakorn &
Bui, 2020) using different sets of discrete unknowns. The formulations proposed by Goto,
Watanabe, Kasugai and Obata (1992), Smoleński (1999), and Vo, Nanakorn and Bui
(2020) are Timoshenko beam formulations in which the displacements of the beam axis
and the total cross-sectional rotation are considered as unknown kinematics. Therefore,
rigid connections and rotational boundary conditions can be easily considered. Meier, Popp
and Wall (2014) use similar unknown kinematics to those considered in the present study,
i.e., the translational displacements of the beam axis and the axial rotation. However, an
alternative approach is used to handle rigid connections. It can be seen from Figure 5.10a
that the unit tangent vectors at the point 𝐶, determined from the segments 𝐵𝐶 and 𝐶𝐷, are
initially identical. In the work of Meier, Popp and Wall (2014), the connection at this point
is handled by considering the tangent vectors as additional discrete unknowns. Clearly, this
approach cannot be directly used to consider connections of beams with kinks. In addition,
applications of concentrated moments require additional treatments. In the work of Meier,
Popp and Wall (2014), an accurate geometric representation of the beam axis is constructed
with a problem-dependent parameter that requires more computational efforts.
Table 5.1 Twisting of a circular ring. Comparison of the number of degrees of freedom
for a half of the ring.
Present 58
In this study, the ring is split into four segments, i.e., 𝐵𝐶, 𝐶𝐷, 𝐷𝐹, and 𝐹𝐵. Exact
geometric representations of the segments are easily obtained using four septic NURBS
curves. No additional treatments are required for the supports at 𝐵, and the connections of
patches at 𝐶, 𝐷 and 𝐹. Figure 5.10a shows the results of the rotation angle 𝛼 obtained by
the proposed formulation and those in literature (Goto, Watanabe, Kasugai & Obata, 1992;
Pai & Palazotto, 1996; Smoleński, 1999; Vo, Nanakorn & Bui, 2020). The results obtained
by Smoleński (1999) and Vo, Nanakorn and Bui (2020) are identical with the reference
result (Pai & Palazotto, 1996). A comparison of the numbers of degrees of freedom used
in the literature is reported in Table 5.1. One can observe excellent agreement between the
present result, where the least number of degrees of freedom is employed, and reference
result. With this obtained result, the accuracy and efficiency of the proposed formulation
are verified. Moreover, the ring exhibits very large deformations, and the use of Euler-
Bernoulli beam formulations, e.g., presented by Meier, Popp and Wall (2014) and the
proposed formulation, provides high accuracy with fewer degrees of freedom in
comparison with Timoshenko beam formulations, e.g., (Goto, Watanabe, Kasugai &
Figure 5.12 Compression of a lattice tower: (a) Problem set up. (b) Target surface and
embedded curves.
Obata, 1992; Smoleński, 1999; Vo, Nanakorn & Bui, 2020). This feature proves the
efficiency of Euler-Bernoulli beam formulations in analysis of highly flexible beam
structures.
The deformations of the ring at every increment of 90 degrees of the rotation angle
𝛼 are plotted in Figure 5.11a-d. It is apparent that the ring undergoes very large three-
dimensional displacements and rotations. Indeed, at the final equilibrium state, the ring is
deformed into a smaller one with the radius of one-third of the original radius. With this
example, the accuracy and efficiency of the proposed formulation for analysis of spatial
multi-patch beam structures undergoing large displacements and rotations are clearly
illustrated.
A four-step procedure described by Choi and Cho (2018) is used to construct the
tower. The necessary quantities for the construction of the tower are given here. The beam
segments of the tower are in fact curves embedded on a target surface. Figure 5.12b shows
the target surface and the embedded curves, i.e., red solid lines, used for this example. The
target surface is a revolution surface of a given generatrix. The generatrix and the axis of
revolution are shown in Figure 5.13a. The embedded curves are mapped onto the target
surface from lines in a parent domain shown in Figure 5.13b. With the given target surface
and lines in the parent domain, the mapping procedure presented by Choi and Cho (2018)
is performed to construct the tower considered in this example.
Since there exist no reference solutions for the downward displacements at the
joints 𝐾 and 𝑁, the beam formulation presented by Vo, Nanakorn and Bui (2020) is used
to analyze the same tower, and the obtained results are considered as the reference. The
result comparisons are shown in Figure 5.14. Good agreement between the results validates
the accuracy of the proposed formulation. Furthermore, to show the enforcement of the
rigid joints, the deformed configurations obtained by using the beam formulation
developed by Vo, Nanakorn and Bui (2020) and the proposed one are visualized in Figure
5.15a-b, respectively. Identical deformed configurations are observed.
CHAPTER 6
CONCLUSIONS
The first part of this study is devoted to the Timoshenko beam formulation. This
study is the first effort to derive such a formulation with the aforementioned energy
conjugate pair in the context of isogeometric analysis. This approach facilitates
consideration of hyperelastic materials in analysis of highly flexible beam structures. The
employed unknown kinematics include the displacements of the beam axis and the cross-
sectional rotation. The cross-sectional rotation is represented in terms of an orthogonal
tensor, which is parameterized by a vector-like parameter with the aid of Rodrigues’
rotation formula. Natural exponentiation and superposition are employed for updating the
cross-sectional rotations. This methodology enables the developed Timoshenko beam
formulation to simulate beams subjected to arbitrary large displacements and rotations.
The second part focuses on the Euler-Bernoulli beam formulation for analysis of
multi-patch beam structures. The displacements of the beam axis and the axial cross-
sectional rotation are considered as the unknown kinematics. Finite cross-sectional
rotations are described by using the smallest rotation mapping. This essentially reduces the
nonlinearity of the strain measurements in terms of the unknown kinematics, and
consequently lessens the computational efforts for the required linearization. To deal with
rigid connections in multi-patch beam structures, a novel nonlinear transformation between
the axial cross-sectional rotation and the total cross-sectional rotation is derived. This
transformation allows the use of the total cross-sectional rotations at the ends of beam
patches as discrete unknowns. The accuracy of this approach in the nonlinear regime is
investigated for the first time. Additionally, with the end total cross-sectional rotations, the
difficulties found in enforcing rotational boundary conditions are removed.
The accuracy and efficiency of the proposed beam formulations are assessed via
several numerical examples. Regarding the locking phenomena of the Timoshenko beam
formulation, i.e., membrane and shear locking, the obtained numerical results indicate that
the locking effects can be significantly reduced by using NURBS curves of high degree for
the approximations of the geometry and unknown kinematics. However, the membrane
locking effect in the Euler-Bernoulli beam formulation is not yet elaborated in this study.
Beams with complex initial geometry, e.g., pre-twisted configurations, are conveniently
handled with the proposed method with the use of NURBS beam axes and director vectors.
The excellent convergence property of the present formulations is clearly observed in the
considered examples, where better orders of convergence than the theoretical expected
orders of convergence of 𝑝 + 1 are achieved. On the other hand, the capability of the
present Euler-Bernoulli beam formulation for analysis of complex multi-patch beam
structures undergoing large three-dimensional deformations is also validated. The obtained
results also show good agreement with those available in the literature.
In all the numerical examples, sextic and septic NURBS curves, i.e., 𝑝 = 6, 7, are
mainly used for the analysis to avoid the locking effects. Although acceptable results can
be obtained with the use of either quartic or quintic NURBS curves, i.e., 𝑝 = 4, 5, more
distinct knot spans are required to obtain the same accuracy. As a result, the total number
of Gauss points required for the numerical integration can in fact be increased when quartic
or quintic NURBS curves are used. This is obviously not computationally effective.
Therefore, sextic and septic NURBS curves are employed in this work to balance the
accuracy and computational efficiency. For future works, locking-free beam formulations
for geometrically nonlinear analysis are worth being developed.
REFERENCES
Adam, C., Bouabdallah, S., Zarroug, M., & Maitournam, H. (2014). Improved numerical
integration for locking treatment in isogeometric structural elements, part i: Beams.
Computer Methods in Applied Mechanics and Engineering, 279, 1-28.
10.1016/j.cma.2014.06.023
Argyris, J. (1982). An excursion into large rotations. Computer Methods in Applied
Mechanics and Engineering, 32(1), 85-155. https://doi.org/10.1016/0045-
7825(82)90069-X
Auricchio, F., Beirão da Veiga, L., Kiendl, J., Lovadina, C., & Reali, A. (2013). Locking-
free isogeometric collocation methods for spatial Timoshenko rods. Computer
Methods in Applied Mechanics and Engineering, 263, 113-126.
https://doi.org/10.1016/j.cma.2013.03.009
Bathe, K.-J., & Bolourchi, S. (1979). Large displacement analysis of three-dimensional
beam structures. International Journal for Numerical Methods in Engineering,
14(7), 961-986. 10.1002/nme.1620140703
Bauer, A. M., Breitenberger, M., Philipp, B., Wüchner, R., & Bletzinger, K. U. (2016).
Nonlinear isogeometric spatial Bernoulli beam. Computer Methods in Applied
Mechanics and Engineering, 303, 101-127.
https://doi.org/10.1016/j.cma.2015.12.027
Bauer, A. M., Wüchner, R., & Bletzinger, K. U. (2019). Weak coupling of nonlinear
isogeometric spatial Bernoulli beams. Computer Methods in Applied Mechanics
and Engineering, 112747. https://doi.org/10.1016/j.cma.2019.112747
Bazilevs, Y., Beirão Da Veiga, L., Cottrell, J. A., Hughes, T. J. R., & Sangalli, G. (2006).
Isogeometric analysis: Approximation, stability and error estimates for h-refined
meshes. Mathematical Models and Methods in Applied Sciences, 16(07), 1031-
1090. 10.1142/S0218202506001455
Cardona, A., & Geradin, M. (1988). A beam finite element non-linear theory with finite
rotations. International Journal for Numerical Methods in Engineering, 26(11),
2403-2438. 10.1002/nme.1620261105
Chen, L., Nguyen-Thanh, N., Nguyen-Xuan, H., Rabczuk, T., Bordas, S. P. A., & Limbert,
G. (2014). Explicit finite deformation analysis of isogeometric membranes.
Computer Methods in Applied Mechanics and Engineering, 277, 104-130.
https://doi.org/10.1016/j.cma.2014.04.015
Choi, M.-J., & Cho, S. (2018). Constrained isogeometric design optimization of lattice
structures on curved surfaces: Computation of design velocity field. Structural and
Multidisciplinary Optimization, 58(1), 17-34. 10.1007/s00158-018-2000-9
Choi, M.-J., & Cho, S. (2019). Isogeometric configuration design sensitivity analysis of
geometrically exact shear-deformable beam structures. Computer Methods in
Applied Mechanics and Engineering, 351, 153-183.
https://doi.org/10.1016/j.cma.2019.03.032
Crisfield, M. A. (1990). A consistent co-rotational formulation for non-linear, three-
dimensional, beam-elements. Computer Methods in Applied Mechanics and
Engineering, 81(2), 131-150. https://doi.org/10.1016/0045-7825(90)90106-V
Crisfield, M. A. (2000). Non-linear finite element analysis of solids and structures. Wiley.
Dornisch, W., & Klinkel, S. (2014). Treatment of Reissner–Mindlin shells with kinks
without the need for drilling rotation stabilization in an isogeometric framework.
Computer Methods in Applied Mechanics and Engineering, 276, 35-66.
http://dx.doi.org/10.1016/j.cma.2014.03.017
Dornisch, W., Klinkel, S., & Simeon, B. (2013). Isogeometric Reissner–Mindlin shell
analysis with exactly calculated director vectors. Computer Methods in Applied
Mechanics and Engineering, 253, 491-504.
http://dx.doi.org/10.1016/j.cma.2012.09.010
Dornisch, W., Müller, R., & Klinkel, S. (2016). An efficient and robust rotational
formulation for isogeometric Reissner–Mindlin shell elements. Computer Methods
in Applied Mechanics and Engineering, 303, 1-34.
http://dx.doi.org/10.1016/j.cma.2016.01.018
Fang, W., Yu, T., Van Lich, L., & Bui, T. Q. (2019). Analysis of thick porous beams by a
quasi-3D theory and isogeometric analysis. Composite Structures, 221, 110890.
https://doi.org/10.1016/j.compstruct.2019.04.062
Goto, Y., Watanabe, Y., Kasugai, T., & Obata, M. (1992). Elastic buckling phenomenon
applicable to deployable rings. International Journal of Solids and Structures,
29(7), 893-909. https://doi.org/10.1016/0020-7683(92)90024-N
Greco, L., & Cuomo, M. (2013). B-spline interpolation of Kirchhoff-Love space rods.
Computer Methods in Applied Mechanics and Engineering, 256, 251-269.
http://dx.doi.org/10.1016/j.cma.2012.11.017
Greco, L., & Cuomo, M. (2014). An implicit g1 multi patch b-spline interpolation for
Kirchhoff–Love space rod. Computer Methods in Applied Mechanics and
Engineering, 269, 173-197. https://doi.org/10.1016/j.cma.2013.09.018
Greco, L., & Cuomo, M. (2015). Consistent tangent operator for an exact Kirchhoff rod
model. Continuum Mechanics and Thermodynamics, 27(4), 861-877.
10.1007/s00161-014-0361-x
Huang, Z., He, Z., Jiang, W., Qiao, H., & Wang, H. (2016). Isogeometric analysis of the
nonlinear deformation of planar flexible beams with snap-back. Acta Mechanica
Solida Sinica, 29(4), 379-390. 10.1016/s0894-9166(16)30241-5
Hughes, T. J. R., Cottrell, J. A., & Bazilevs, Y. (2005). Isogeometric analysis: CAD, finite
elements, NURBS, exact geometry and mesh refinement. Computer Methods in
Applied Mechanics and Engineering, 194(39–41), 4135-4195.
http://dx.doi.org/10.1016/j.cma.2004.10.008
Ibrahimbegovic, A. (1997). On the choice of finite rotation parameters. Computer Methods
in Applied Mechanics and Engineering, 149(1), 49-71.
https://doi.org/10.1016/S0045-7825(97)00059-5
Ibrahimbegović, A. (1995). On finite element implementation of geometrically nonlinear
reissner's beam theory: Three-dimensional curved beam elements. Computer
Methods in Applied Mechanics and Engineering, 122(1), 11-26.
https://doi.org/10.1016/0045-7825(95)00724-F
Ibrahimbegović, A., Frey, F., & Kožar, I. (1995). Computational aspects of vector-like
parametrization of three-dimensional finite rotations. International Journal for
Numerical Methods in Engineering, 38(21), 3653-3673. 10.1002/nme.1620382107
Kiendl, J., Bazilevs, Y., Hsu, M. C., Wüchner, R., & Bletzinger, K. U. (2010). The bending
strip method for isogeometric analysis of Kirchhoff–Love shell structures
Marino, E., Kiendl, J., & De Lorenzis, L. (2019). Isogeometric collocation for implicit
dynamics of three-dimensional beams undergoing finite motions. Computer
Methods in Applied Mechanics and Engineering, 356, 548-570.
https://doi.org/10.1016/j.cma.2019.07.013
Mattiasson, K. (1981). Numerical results from large deflection beam and frame problems
analysed by means of elliptic integrals. International Journal for Numerical
Methods in Engineering, 17(1), 145-153. 10.1002/nme.1620170113
Meier, C., Popp, A., & Wall, W. A. (2014). An objective 3D large deformation finite
element formulation for geometrically exact curved Kirchhoff rods. Computer
Methods in Applied Mechanics and Engineering, 278, 445-478.
https://doi.org/10.1016/j.cma.2014.05.017
Meier, C., Popp, A., & Wall, W. A. (2019). Geometrically exact finite element
formulations for slender beams: Kirchhoff–Love theory versus Simo–Reissner
theory. Archives of Computational Methods in Engineering, 26(1), 163-243.
10.1007/s11831-017-9232-5
Pai, P. F., & Palazotto, A. N. (1996). Large-deformation analysis of flexible beams.
International Journal of Solids and Structures, 33(9), 1335-1353.
https://doi.org/10.1016/0020-7683(95)00090-9
Piegl, L., & Tiller, W. (1997). The NURBS book. New York, United States: Springer.
Radenković, G., & Borković, A. (2018). Linear static isogeometric analysis of an
arbitrarily curved spatial Bernoulli–Euler beam. Computer Methods in Applied
Mechanics and Engineering, 341, 360-396.
https://doi.org/10.1016/j.cma.2018.07.010
Raknes, S. B., Deng, X., Bazilevs, Y., Benson, D. J., Mathisen, K. M., & Kvamsdal, T.
(2013). Isogeometric rotation-free bending-stabilized cables: Statics, dynamics,
bending strips and coupling with shells. Computer Methods in Applied Mechanics
and Engineering, 263, 127-143. https://doi.org/10.1016/j.cma.2013.05.005
Reissner, E. (1973). On one-dimensional large-displacement finite-strain beam theory.
Studies in Applied Mathematics, 52(2), 87-95. 10.1002/sapm197352287
Young, W. C., Budynas, R. G., & Sadegh, A. M. (2012). Roak's fomulas for stress and
strain. New York, United States: McGraw-Hill.
Yu, T., Hu, H., Zhang, J., & Bui, T. Q. (2019). Isogeometric analysis of size-dependent
effects for functionally graded microbeams by a non-classical quasi-3D theory.
Thin-Walled Structures, 138, 1-14. https://doi.org/10.1016/j.tws.2018.12.006
Yu, T., Lai, W., & Bui, T. Q. (2019). Three-dimensional elastoplastic solids simulation by
an effective iga based on bézier extraction of NURBS. International Journal of
Mechanics and Materials in Design, 15(1), 175-197. 10.1007/s10999-018-9405-x
Yu, T., Zhang, J., Hu, H., & Bui, T. Q. (2019). A novel size-dependent quasi-3D
isogeometric beam model for two-directional FG microbeams analysis. Composite
Structures, 211, 76-88. https://doi.org/10.1016/j.compstruct.2018.12.014
Yuan, Z., Wang, X., & Kardomateas, G. A. (2019). A co-rotational weak-form quadrature
planar beam element for geometric nonlinear static and dynamic analysis.
International Journal for Numerical Methods in Engineering, 0(0),
10.1002/nme.6183
APPENDICES
APPENDIX A
There are several choices for the parameterization of the rotation tensor 𝚲, e.g.,
Euler angles, quaternion parameters, and Rodrigues’ rotation formula (Crisfield, 2000). In
this study, Rodrigues’ rotation formula is used.
Any rotation can be described by an axis of rotation and a rotation angle about that
axis. Let 𝛉 = 𝜃𝑖 𝐞𝑖 be a vector-like parameter whose direction is the axis of rotation and the
magnitude 𝜃 = ‖𝛉‖ is the rotation angle about 𝛉. Unless the rotation angle 𝜃 is small, 𝜃𝑖
̃
should not be interpreted as the component rotation about 𝐞𝑖 . A skew-symmetric tensor 𝛉
is defined using the components of 𝛉 as
0 −𝜃3 𝜃2
̃ = [ 𝜃3
𝛉 0 −𝜃1 ]. (A.1)
−𝜃2 𝜃1 0
sin 𝜃 1 − cos 𝜃
̃) = 𝐈 +
𝚲 = exp(𝛉 ̃+
𝛉 ̃𝛉
𝛉 ̃. (A.2)
𝜃 𝜃2
This formulation is known as Rodrigues’ rotation formula, which is the most robust
formulation for finite rotations in three-dimensional spaces. For a relatively small 𝜃, the
following properties should be used for stability, i.e.,
sin 𝜃 1 − cos 𝜃 1
lim ( )=1 lim ( ) = . (A.3)
𝜃→0 𝜃 𝜃→0 𝜃2 2
(A.4)
𝛉′ = 𝜃𝑖′ 𝐞𝑖 (A.5)
𝛉′ ⋅ 𝛉
𝜃′ = (A.6)
𝜃
0 −𝜃3′ 𝜃2′
̃
𝛉 = [ 𝜃3′
′ 0 −𝜃1′ ]. (A.9)
−𝜃2′ 𝜃1′ 0
For a relatively small 𝜃, the following properties should be used for stability, i.e.,
sin 𝜃 ′ 1 − cos 𝜃 ′
lim (𝜃 ′ ) = lim [( ) ] = lim [( ) ] = 0. (A.10)
𝜃→0 𝜃→0 𝜃 𝜃→0 𝜃2
APPENDIX B
The Saint Venant-Kirchhoff material model is used in this study. This model is
considered as an extension of the isotropic linear elastic material model to analysis of
problems with large strains. The relation between the second Piola-Kirchhoff stress tensor
𝐒 and the Green-Lagrange strain tensor 𝐄 can be expressed as
where 𝐶 𝑖𝑗𝑘𝑙 are the material moduli expressed in the curvilinear coordinate system 𝐆𝑖 ⊗
𝐆𝑗 as
𝜈𝐸 𝐸
𝜆= 𝜇 = 𝑘𝑠 =𝐺 (B.3)
(1 + 𝜈)(1 − 2𝜈) 2(1 + 𝜈)
where 𝐸 is Young’s modulus and 𝜈 is Poisson’s ratio. In addition, 𝐺 is the shear modulus
and 𝑘𝑠 is the shear correction factor which varies with the shape of the cross-section. For
rectangular cross-sections, 𝑘𝑠 = 5/6 can be used.
For the Timoshenko beam theory, the stress components 𝑆 22 , 𝑆 33 , and 𝑆 23 are
eliminated by static condensation (Klinkel & Govindjee, 2002). It then follows that
𝐸 𝐺
𝐸̅ = 𝐺̅ = . (B.5)
(𝐺11 )2 𝐺11
Furthermore, the following relation is assumed for relatively thick beams without causing
significant errors to the numerical results, i.e.,
𝐺11 ≈ 𝐀1 . 𝐀1 = 1. (B.6)
With the above assumption, 𝐸̅ and 𝐺̅ are constant over the cross-section. The assumption
simplifies the computation of the cross-sectional resultant forces.
APPENDIX C
The variation of the unit tangent vector 𝐭 and its derivative are given by
(C.2)
𝐭 ⊗ ̅̅
𝐀̅̅1 𝐭 × 𝛿𝐚1
𝛿𝐰 = 𝐭𝛿𝜃 + (𝐈 − ) ̅̅̅̅1 , 𝐭)𝛿𝐚1
= 𝐭𝛿𝜃 + 𝐗(𝐀 (C.3)
1 + ̅̅
𝐀̅̅1 ⋅ 𝐭 𝑎1
̅̅̅̅1 T
𝐭𝐀 𝐭̃
̅̅̅̅1 , 𝐭) = (𝐈 −
𝐗(𝐀 ) (C.5)
T
̅̅̅̅1 𝐭 𝑎1
1+𝐀
T
T
𝐭 ′ ̅̅̅̅ ̅̅̅̅1 )
𝐀1 + 𝐭(𝐀
′ T ̅̅̅̅1 T [(𝐀
𝐭𝐀 ̅̅̅̅1 ′ ) 𝐭 + ̅̅̅̅T
𝐀1 𝐭 ′ ] 𝐭̃
̅̅̅̅1 , 𝐭) = {−
𝐗 ′ (𝐀 + }
T 2 𝑎1
1 + ̅̅
𝐀̅̅1 𝐭 (1 + ̅̅
𝐀̅̅1 𝐭)
T
(C.6)
T
̅̅̅̅1
𝐭𝐀 𝐭̃′ (𝐚1T 𝐚1′ )𝐭̃
+ (𝐈 − T
)[ − ].
1 + ̅̅
𝐀̅̅1 𝐭 𝑎1 (𝑎1 )3
In the increment of 𝛿𝚪, the vector 𝛿𝐰 always appears as 𝐀T Δ𝛿𝐰 and 𝐀T Δ𝛿𝐰 ′ with
an arbitrary vector 𝐀. The lengthy derivation is skipped, and the following expressions are
provided as
1 𝐀 ̃
̅̅̅̅1 𝐭𝐀T 𝐙 ̃
1 ̅̅
𝐀 ̅̅1 𝐭𝐀T 𝐭𝐀̅̅̅̅1 T 𝐙 𝐀 ̃𝐙 𝐀̃ 𝐭𝐚1T ̃
̅̅
1 𝐀T 𝐭𝐀 ̅̅1 𝐙
𝐘(𝐀) = − T + 2 + − − T
𝑎1 1 + ̅̅̅̅
𝐀1 𝐭 𝑎1 (1 + 𝐀 ̅̅̅̅1 T 𝐭) 𝑎1 (𝑎1 )3 𝑎1 1 + ̅̅𝐀̅̅1 𝐭
(C.9)
̃ 𝐭𝐀T 𝐭𝐚T
1 ̅̅̅̅
𝐀 1 1
+ T
(𝑎1 )3 1 + ̅̅
𝐀̅̅ 𝐭 1
′ (𝐀)
̃
̅̅
𝐚1T 𝐚1′ 𝐀 ̅̅1 𝐭𝐀T 𝐙 𝐀̃
1 ̅̅̅̅1 ′ 𝐭𝐀T 𝐙 + ̅̅̅̅
̃ 𝐭 ′ 𝐀T 𝐙 + ̅̅
𝐀 1
̃
𝐀 ̅̅1 𝐭𝐀T 𝐙′
𝐘 ={ T − T
(𝑎1 )3 1 + ̅̅ 𝐀̅̅ 𝐭 𝑎1 1 + ̅̅
𝐀̅̅ 𝐭
1 1
T
1 ̅̅̅̅
𝐀 1
̅̅̅̅1 ′ ) 𝐭 + ̅̅̅̅
̃ 𝐭𝐀T 𝐙 [(𝐀 T
𝐀1 𝐭 ′ ]
+ }
𝑎1 T 2
̅̅
(1 + 𝐀1 𝐭) ̅̅
(C.10)
̃ T
𝐚1T 𝐚1′ ̅̅
𝐀 ̅̅1 𝐭𝐀T 𝐭𝐀
̅̅̅̅1 𝐙
+ {−
(𝑎1 )3 T 2
(1 + ̅̅̅̅
𝐀1 𝐭)
𝐀̃1 𝐭𝐀T 𝐭𝐀
′
1 ̅̅̅̅ ̅̅̅̅1 T 𝐙 + ̅̅
̃
𝐀 ̅̅1 𝐭 ′ 𝐀T 𝐭𝐀̅̅̅̅1 T 𝐙 + ̅̅
̃
𝐀 ̅̅1 𝐭𝐀T 𝐭 ′ ̅̅̅̅
𝐀1 𝐙
T
+
𝑎1 T 2
(1 + ̅̅ 𝐀̅̅1 𝐭)
T
̃
1 ̅̅
𝐀 ̅̅̅̅1 ′ ) 𝐙 + ̅̅
̅̅1 𝐭𝐀T 𝐭(𝐀 ̃
𝐀 ̅̅̅̅1 T 𝐙′
̅̅1 𝐭𝐀T 𝐭𝐀
+
𝑎1 T 2
(1 + ̅̅ 𝐀̅̅1 𝐭)
T
̃
̅̅
1 2𝐀 ̅̅̅̅1 T 𝐙 [(𝐀
̅̅1 𝐭𝐀T 𝐭𝐀 ̅̅̅̅1 ′ ) 𝐭 + ̅̅̅̅ T
𝐀1 𝐭 ′ ]
− }
𝑎1 T 3
̅̅
(1 + 𝐀1 𝐭) ̅̅
̃𝐙 𝐀
𝐚1T 𝐚1′ 𝐀 ̃𝐙′ ̃ 𝐭𝐚1T 𝐀
3𝐚1T 𝐚1′ 𝐀 ̃ 𝐭 ′ 𝐚1T + 𝐀
̃ 𝐭(𝐚1′ )T
+ {− + }+{ − }
(𝑎1 )3 𝑎1 (𝑎1)5 (𝑎1)3
̃𝐙
̅̅̅̅
𝐚1T 𝐚1′ 𝐀T 𝐭𝐀 ̃
1 𝐀T 𝐭 ′ ̅̅
𝐀 ̅̅1 𝐙
1
+{ T − T
(𝑎1 )3 1 + ̅̅
𝐀̅̅ 𝐭 𝑎1 1 + ̅̅ 𝐀̅̅ 𝐭
1 1
T
̅̅̃
1 𝐀T 𝐭𝐀 ̅̅1 ′ 𝐙 + 𝐀T 𝐭𝐀 ̃
̅̅
̃ 𝐙′ 1 𝐀T 𝐭𝐀
̅̅̅̅ ̅̅̅̅1′ ) 𝐭 + ̅̅
̅̅1 𝐙 [(𝐀 T
𝐀̅̅1 𝐭 ′ ]
1
− + }
𝑎1 T 𝑎1 T 2
1 + ̅̅𝐀̅̅1 𝐭 ̅̅
(1 + 𝐀 𝐭) ̅̅ 1
̃
𝐚1T 𝐚1′ ̅̅
𝐀 ̅̅1 𝐭𝐀T 𝐭𝐚1T
+ {− T
(𝑎1 )5 1 + ̅̅ 𝐀̅̅ 𝐭
1
APPENDIX D
Before expressing the increment of the variation of 𝚪, two matrices are introduced
as
𝑛 𝑛
𝑖𝑗
Δ𝛿𝛤11 = 𝛿𝐯 𝐋11 Δ𝐯 = ∑ ∑ 𝛿𝐯𝑖T 𝐋11 Δ𝐯𝑗
T
(D.3)
𝑖=1 𝑗=1
𝑛 𝑛
𝑖𝑗
Δ𝛿𝐾12 = 𝛿𝐯 𝐋12 Δ𝐯 = ∑ ∑ 𝛿𝐯𝑖T 𝐋12 Δ𝐯𝑗
T
(D.4)
𝑖=1 𝑗=1
𝑛 𝑛
𝑖𝑗
Δ𝛿𝐾13 = 𝛿𝐯 𝐋13 Δ𝐯 = ∑ ∑ 𝛿𝐯𝑖T 𝐋13 Δ𝐯𝑗
T
(D.5)
𝑖=1 𝑗=1
𝑛 𝑛
𝑖𝑗
Δ𝛿𝐾23 = 𝛿𝐯 𝐋23 Δ𝐯 = ∑ ∑ 𝛿𝐯𝑖T 𝐋23 Δ𝐯𝑗
T
(D.6)
𝑖=1 𝑗=1
𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗
where the matrices 𝐋11, 𝐋12, 𝐋13, and 𝐋23 are defined as
𝑖𝑗
𝐋23
̅̅̅̅1, 𝐭)]T 𝐙 + 𝐘 ′ (𝐭)} − 𝑅𝑖′′ (𝜂)𝑅𝑗′(𝜂)[𝐗T (𝐀
−𝑅𝑖′ (𝜂)𝑅𝑗′ (𝜂){[𝐗 ′ (𝐀 ̅̅̅̅1 , 𝐭)𝐙 + 𝐘(𝐭)] − 𝑅𝑖′ (𝜂)𝑅𝑗′′ (𝜂)𝐘(𝐭) 𝟎
=[ ]
−𝑅𝑖 (𝜉)𝑅𝑗′ (𝜉)[(𝐭′ )T 𝐙 + 𝐭T 𝐙 ′ ] − 2𝑅𝑖′ (𝜂)𝑅𝑗′(𝜂)𝐭T 𝐙 − 𝑅𝑖 (𝜂)𝑅𝑗′′ (𝜂)𝐭T 𝐙 0
(D.10)
BIOGRAPHY
Publications
International journals
Vo, D., & Nanakorn, P. (2020). Geometrically nonlinear multi-patch isogeometric analysis
of planar curved Euler–Bernoulli beams. Computer Methods in Applied
Mechanics and Engineering, 366, 113078.
https://doi.org/10.1016/j.cma.2020.113078.
Vo, D., & Nanakorn, P. (2020). A total Lagrangian Timoshenko beam formulation for
geometrically nonlinear isogeometric analysis of planar curved beams. Acta
Mechanica, 231(7), 2827-2847. https://doi.org/10.1007/s00707-020-02675-x.
Vo, D., Nanakorn, P., & Bui, T. Q. (2020). A total Lagrangian Timoshenko beam
formulation for geometrically nonlinear isogeometric analysis of spatial beam
structures. Acta Mechanica, 231(9), 3673-3701. https://doi.org/10.1007/s00707-
020-02723-6.
Vo, D., Borković, A., Nanakorn, P., & Bui, T. Q. (2020). Dynamic multi-patch
isogeometric analysis of planar Euler-Bernoulli beams. Computer Methods in
Applied Mechanics and Engineering, 372, 113435,
https://doi.org/10.1016/j.cma.2020.113435
Suttakul, P., Nanakorn, P., & Vo, D. (2019). Effective out-of-plane rigidities of 2D lattices
with different unit cell topologies. Archive of Applied Mechanics, 89(9), 1837-
1860. https://doi.org/10.1007/s00419-019-01547-8.
International conferences
Vo, D., Nanakorn, P. (2017) . Large displacement analysis of 2D beams using field-
consistent beam elements with a higher-order axial-displacement interpolation.
Proceeding of 15th East Asia-Pacific Conference on Structural Engineering and
Construction (EASEC-15) (pp.1672-1678). Xi’an, P.R. China.
Book chapters
Vo, D., Nanakorn, P., & Bui, T. Q. (2020). Multi-patch geometrically nonlinear
isogeometric analysis of spatial beams with additive rotation updates. Reddy, J.
N., Wang, C. M., Luong, V. H., & Le, A. T., Proceeding of ICSCEA 2019 (pp.
1129-1136). Singapore: Springer Singapore.