Professional Documents
Culture Documents
Meier2017 PDF
Meier2017 PDF
DOI 10.1007/s11831-017-9232-5
ORIGINAL PAPER
Abstract The present work focuses on geometrically formulations fulfilling these requirements can be found
exact finite elements for highly slender beams. It aims at the in the literature very well. However, it will be argued that
proposal of novel formulations of Kirchhoff–Love type, a the shear-free Kirchhoff–Love formulations can provide
detailed review of existing formulations of Kirchhoff–Love considerable numerical advantages such as lower spatial
and Simo–Reissner type as well as a careful evaluation discretization error levels, improved performance of time
and comparison of the proposed and existing formulations. integration schemes as well as linear and nonlinear solvers
Two different rotation interpolation schemes with strong and smooth geometry representation as compared to shear-
or weak Kirchhoff constraint enforcement, respectively, as deformable Simo–Reissner formulations when applied
well as two different choices of nodal triad parametriza- to highly slender beams. Concretely, several representa-
tions in terms of rotation or tangent vectors are proposed. tive numerical test cases confirm that the proposed Kirch-
The combination of these schemes leads to four novel finite hoff–Love formulations exhibit a lower discretization error
element variants, all of them based on a C1-continuous Her- level as well as a considerably improved nonlinear solver
mite interpolation of the beam centerline. Essential require- performance in the range of high beam slenderness ratios
ments such as representability of general 3D, large defor- as compared to two representative Simo–Reissner element
mation, dynamic problems involving slender beams with formulations from the literature.
arbitrary initial curvatures and anisotropic cross-section
shapes, preservation of objectivity and path-independence,
consistent convergence orders, avoidance of locking effects 1 Introduction
as well as conservation of energy and momentum by the
employed spatial discretization schemes, but also a range Highly slender fiber- or rod-like components represent
of practically relevant secondary aspects will be investi- essential constituents of mechanical systems in countless
gated analytically and verified numerically for the differ- fields of application and scientific disciplines such as civil,
ent formulations. It will be shown that the geometrically mechanical and biomedical engineering, material science
exact Kirchhoff–Love beam elements proposed in this work and bio- or molecular physics. Examples are high-tensile
are the first ones of this type that fulfill all these essen- industrial webbings, fiber-reinforced composite materials,
tial requirements. On the contrary, Simo–Reissner type fibrous materials with tailored porosity, synthetic polymer
materials or also cellulose fibers determining the charac-
* Christoph Meier teristics of paper [49, 57, 93, 94, 122, 152]. On entirely
meier@lnm.mw.tum.de different time and length scales, such slender components
1
are relevant when analyzing the supercoiling process of
Mechanosynthesis Group, Massachusetts Institute
DNA strands, the characteristics of carbon nanotubes or the
of Technology, 77 Massachusetts Ave., Cambridge,
MA 02139, USA Brownian dynamics within the cytoskeleton of biological
2 cells, a biopolymer network of highly slender filaments that
Institute for Computational Mechanics, Technical
University of Munich, Boltzmannstrasse 15, crucially influences biologically relevant processes such
85748 Garching b. München, Germany as cell division and cell migration [45, 106, 113, 114, 154,
13
Vol.:(0123456789)
C. Meier et al.
159]. Often, these slender components can be modeled as equilibrium of forces and moments, conservation of energy
1D Cosserat continua based on a geometrically nonlinear or rather existence of work conjugated stress–strain pairs,
beam theory. In all mentioned cases, mechanical contact observer invariance or path-independence of conserva-
interaction crucially influences the overall system behavior. tive problems. Nevertheless, intrinsic beam theories are
While in the authors’ recent contributions [108, 111, 112] decoupled from the 3D continuum mechanics theory. Typi-
a novel beam contact formulation has been proposed that cally, the postulated constitutive constants relating stress
faces the challenges arising from high beam slenderness and strain measures are determined experimentally, while
ratios and complex 3D contact configurations, the present the constitutive constants of induced beam theories follow
contribution focuses on the development of geometrically directly from the corresponding 3D constitutive laws. Such
exact beam element formulations suitable for an accurate, postulated constitutive laws based on experimentally deter-
efficient and robust modeling of highly slender beams. mined constants are favorable for applications where no 3D
While the vast majority of geometrically exact beam ele- continuum foundation exists: considering the low number
ment formulations available in the literature is based on of discrete molecules distributed over the thickness of mac-
the Simo–Reissner theory of thick rods, the current work romolecules as occurring for example in biopolymer net-
proposes geometrically exact finite elements based on the works, DNA strands or carbon nanotubes—to come back
Kirchhoff–Love theory of thin rods that are tailored for to the applications mentioned above—no 3D continuum
high beam slenderness ratios. theory can be applied in a reasonable manner. Neverthe-
less, it is well-established that these slender components
1.1 Classification of Beam Theories and Fundamental can be described in good approximation by 1D continuum
Computational Approaches theories and associated experimentally determined consti-
tutive constants [132]. A compromise between induced and
Basically, two essential motivations for applying a beam intrinsic theories are so-called semi-induced beam theories,
theory instead of a 3D continuum mechanics theory for the where only the constitutive law is postulated. The remain-
modeling of slender bodies can be identified: in the early ing kinetic and kinematic relations are consistently derived
days of beam theories, it was the accessibility of analytic from the 3D theory [3, 51].
solutions as for example “Euler’s Elastica”, even for large Based on the Bernoulli hypothesis of undeformed
deformation problems, that motivated the development and cross-sections and the work of Euler, the “Kirchhoff
application of one-dimensional theories. Nowadays, it is beam” [89], proposed by Kirchhoff in 1859, was the first
the knowledge that the modeling of highly slender bodies formulation allowing for arbitrary initial curvatures and
via beam theories yields considerably more efficient, but large three-dimensional deformations including the states
also more well-posed numerical formulations as it would of bending and torsion. In 1944, the theory was enhanced
be the case for 3D continuum theories. So-called induced by Love [104] to also account for small axial tension. A
beam theories can be regarded as reduced 1D continuum comprehensive historic overview of these early develop-
theories consistently derived from the 3D theory of con- ments is given in the work of Dill [47]. It was Reissner in
tinuum mechanics under consideration of a basic kinematic 1972 for the two-dimensional case [120] and in 1981 for
constraint that reflects the deformation states expected for the general three-dimensional case [121], who completed
slender bodies in a reasonable manner. Such 1D beam the- the theory by two additional deformation measures repre-
ories typically allow to describe the motion and deforma- senting the shear deformation of the beam. While the 3D
tion of slender bodies in 3D space on the basis of proper problem statement of Reissner was still based on some
kinematic, kinetic and constitutive resultant quantities. In additional approximations, Simo [138] extended the work
the case of induced beam theories, these resultant quanti- of Reissner to yield a formulation that is truly consistent
ties can for example be derived via integration of 3D stress in the sense of a semi-induced beam theory. Thus, starting
measures over the beam cross-section. The 3D stress meas- from a basic kinematic assumption, all kinetic and kine-
ures typically result from the constrained 3D displacement matic quantities and relations are consistently derived from
field as well as standard 3D strain measures and constitu- the 3D continuum theory, while the constitutive law has
tive relations. In this context, the cross-section of a beam been postulated. Originally, this theory has been denoted
represents the collection of all material points sharing the as geometrically exact beam theory. Nowadays, it is also
same beam length coordinate in the stress-free configura- referred to as Simo–Reissner beam theory. According to
tion. On the contrary, so-called intrinsic beam theories the definition of Simo [138], also in this work, a beam the-
directly postulate the 1D resultant quantities. These theo- ory is denoted as geometrically exact, if “the relationships
ries are internally consistent in the sense that the result- between the configuration and the strain measures are con-
ant quantities as well as the 1D relations connecting these sistent with the virtual work principle and the equilibrium
quantities still fulfill essential mechanical principles such as equations at a deformed state regardless of the magnitude
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
of displacements, rotations and strains” ([43], p. 1126). moments of inertia of area I2 = I3) are denoted as nonlin-
For that reason, also the notion “finite-strain beams” has ear Euler–Bernoulli beams. The extension to anisotropic
been applied in the original work [138]. However, as later cross-section shapes (different moments of inertia of area
argued by several authors (see e.g. [43]) and in accordance I2 ≠ I3) and arbitrary initial curvatures 𝜅0 ≠ 0 is referred to
with derivations in the literature (see e.g. [2, 3, 104], but as Kirchhoff–Love theory. In order to simplify the subse-
also Sect. 3.5 of this work), a consistency of the geometri- quent comparison of shear-free beam formulations from the
cally exact beam theory and the 3D theory of continuum literature the more refined nomenclature of (i) isotropic, (ii)
mechanics in the sense of a (fully) induced beam theory straight and (iii) anisotropic Kirchhoff–Love formulations
can only be assumed as long as small strains are consid- is introduced for theories capturing i) neither anisotropic
ered. The fulfillment of the basic kinematic assumption cross-section shapes nor initial curvatures (=nonlinear
of rigid cross-sections underlying the geometrically exact Euler–Bernoulli), (ii) anisotropic cross-section shapes but
beam theory requires pointwise six (translational and rota- no initial curvatures or (iii) both, anisotropic cross-section
tional) degrees of freedom in order to uniquely describe the shapes and initial curvatures. The supplementation by
(centroid) position and orientation of the cross-sections. shear deformation modes is covered by the Simo–Reiss-
Consequently, this beam theory can be identified as 1D ner theory. Euler–Bernoulli or Kirchhoff–Love formula-
Cosserat continuum [38], derived from a 3D Boltzmann tions that neglect the mode of axial tension are denoted as
continuum with pointwise three (translational) degrees of inextensible Euler–Bernoulli or Kirchhoff–Love variants.
freedom. While there exists a variety of beam theories that Recently, shear-deformable and shear-free beam formu-
also consider in-plane as well as out-of-plane cross-section lations have been proposed that also neglect the mode of
distortion, this contribution focuses on geometrically exact torsion [110, 127]. These formulations are capable of accu-
beam formulations on the basis of the rigid cross-section rately modeling ropes and cables (providing a mechanically
assumption as applied by Simo and Reissner. Furthermore, consistent bending stabilization), quasi-continua such as
throughout this work, the notion Simo–Reissner theory chains but also general 1D continua with full bending and
will be preferred since the notion geometrically exact beam torsional stiffness if some restrictions concerning external
theory, when following the definition presented above, also loads (no torsional moment loads) and initial geometry
applies to consistently derived shear-free formulations on (I2 = I3, 𝜅0 = 0) are fulfilled (see also [110]). Eventually,
the basis of the Kirchhoff–Love theory. In the remainder of the restriction of shear-free or shear-deformable theories to
this work, the notion “shear-free” represents the opposite the geometrically linear regime yields the well-known lin-
of “shear-deformable” and thus is equivalent to “vanish- ear Euler–Bernoulli and Timoshenko beam models [150].
ing shear strains”, but, of course, not to “vanishing shear An overview of these different beam models and theories is
stresses”. given in Table 1.
Unfortunately, no absolute consensus concerning nam- By identifying the configuration space underlying the
ing of different beam models can be found in the literature. geometrically exact beam theory as nonlinear, differenti-
For that reason, the following nomenclature, trying to be able manifold with Lie group structure and by pointing
consistent with the most important representatives in the out important algorithmic consequences resulting from
literature, will be applied here: geometrically nonlinear the non-additivity and non-commutativity of the associ-
beam models capturing the modes of axial tension, tor- ated group elements, the original work by Simo [138] and
sion and bending and being appropriate for initially straight the subsequent work by Simo and Vu-Quoc [140] laid the
beams with isotropic cross-section shapes (identical foundation for abundant research work on this topic in the
13
C. Meier et al.
following years. The static beam theory [138, 140] has centerline representation based on an isogeometric collo-
been extended to dynamics by Cardona and Geradin [34, cation scheme is proposed in [155]. Besides these purely
35] and by Simo and Vu-Quoc [141]. The contributions of displacement-based elements, also interpolation schemes
Kondoh et al. [91], Dvorkin et al. [50] as well as Iura and directly acting at the strain level combined with a subse-
Atluri [81] can be regarded as further pioneering works in quent derivation of the position and rotation field via inte-
this field. These contributions mark the starting point for gration (see e.g. [151, 162, 163]) as well as mixed formu-
the development of a large variety of geometrically exact lations [131] have been proposed. Moreoever, a variety
beam element formulations [44, 74, 77, 82, 84, 116, 130, of scientific contributions considering time integration of
143], which basically differ in the type of rotation interpo- rotational variables can be found in this context [14, 21, 26,
lation (e.g. interpolation of incremental, iterative or total 46, 64, 78, 86, 101, 125, 142].
rotational vectors), the choice of nodal rotation parametri- While all the formulations presented above have been
zation (via rotation vectors, quaternions etc.), the type of based on the finite element method (FEM), also discrete
iterative rotation updates (multiplicative or additive), or the representatives of Simo–Reissner beam formulations based
time integration scheme applied to the rotational degrees on finite difference schemes can be found in the litera-
of freedom (e.g. based on additive or multiplicative rota- ture [85, 95, 96, 103]. These are often denoted as discrete
tion increments). Also extensions of the geometrically elastic rods and based on the concept of discrete differential
exact beam theory to arbitrary cross-section shapes with geometry (DDG). In the context of finite element formula-
shear centers differing from the cross-section centroid can tions for geometrically nonlinear beam problems, again, a
be found [68]. An overview of the most important develop- variety of alternatives to the geometrically exact formula-
ments at that time is exemplarily given in the text books of tions considered in the last two paragraphs can be found.
Crisfield [40] as well as Geradin and Cardona [59]. A break The maybe most prominent representatives of these alterna-
in this development is given by the works of Crisfield and tives are the corotational method [39, 40, 42, 54, 58] as well
Jelenić [43, 83], who have shown that none of the rotation as Absolute Nodal Coordinate (ANC) [60, 61, 134, 135]
field discretizations of the formulations existent at that time and solid beam element [11, 56] formulations. The corota-
could preserve both of the important properties of objectiv- tional technique was initially introduced by Wempner [158]
ity and path-independence (see also [79] for a discussion of as well as Belytschko et al. [17, 18] and shows strong simi-
this topic). Furthermore, in [43] and [83], a new, objective larities to the “natural approach” of Argyris et al. [4]. The
and path-independent orthogonal interpolation scheme was basic idea is to split the overall non-linear deformation into
proposed that directly acts on the rotation manifold and not a contribution stemming from large rotations and a part
on any of its rotation vector parametrizations as done in the stemming from local deformations expressed in a local,
works before. This formulation was the starting point for “corotated” frame. Often, the local deformation can be
the development of many alternative rotation interpolation modeled on the basis of first- (or second-) order theories
strategies for geometrically exact beams that also preserve such that the entire degree of nonlinearity is represented
these properties. Among others, orthogonal interpolations by the rotation of the local frame. The basic idea of ANC
of relative rotation vectors (see e.g. [63, 128]) or quaterni- beam element formulations is to employ standard shape
ons (see e.g. [62, 123, 164]), non-orthogonal interpolation functions as known from solid finite element formulations
strategies in combination with modified beam models (see in order to interpolate the 3D displacement field within
e.g. [22, 52, 126]) and non-orthogonal interpolation strat- the beam. Instead of introducing a kinematic constraint
egies with subsequent orthogonalization (see e.g. [123]) and deriving a resultant 1D model, different polynomial
can be identified. As reported in the original work [43, 83], degrees are typically applied for the interpolation in beam
the rotation interpolation scheme proposed by Crisfield length direction and in transverse directions. Numerical
and Jelenić can exactly represent the state of constant cur- comparisons as performed e.g. by Romero [123, 124] and
vature. Thus, it can be interpreted as geodesic, i.e. short- Bauchau et al. [15] advocate geometrically exact beams
est, connection between two configurations on the rotation in general, and orthogonal triad interpolation schemes
manifold. Consequently, these geodesic rotation interpola- (see e.g. [43]) in particular, with regard to computational
tion schemes represent the counterpart to linear interpola- accuracy and efficiency. Especially, the arguments I–III of
tions of translational quantities. In the work of Borri and Sect. 1.3 in terms of possible benefits of shear-free as com-
Bottasso [24, 25] as well as in the more recent contribu- pared to shear-deformable beam element formulations in
tions [128, 144, 145, 151], the spatial discretization of the range of high beam slenderness ratios are pronounced
the rotational and translational primary variable fields is even stronger when comparing geometrically exact Kirch-
based on such a geodesic triad interpolation. A 2D exten- hoff–Love beam element formulations with ANC beam ele-
sion of these so-called helicoidal interpolations to higher- ment formulations, with the latter additionally exhibiting
order elements is given in [48]. A formulation with smooth very stiff in-plane deformation modes when dealing with
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
very slender beams. On the other hand, a clear distinction applied different interpolations (C0-continuous Lagrange
between corotational and geometrically exact formulations polynomials for the axial displacements and C1-continu-
will not always be possible. For example, the interpolation ous Hermite polynomials for the transversal displace-
scheme of the geometrically exact formulation proposed by ments) to the different displacement-components, which
Crisfield and Jelenić [43, 83] is also based on the defini- led to a loss of objectivity [5]. In later works, the objec-
tion of an element-local, corotated reference triad and can tivity could be preserved by employing e.g. trigonometric
consequently be interpreted as corotational formulation shape functions, but the corresponding formulations were
at the same time. All in all, it can be stated that geometri- limited to the investigation of plane circular arches (see
cally exact finite element formulations have become well- e.g. [8] or [9]). A different approach was the develop-
established in the meantime and can arguably be regarded ment of framework or corotational shear-free beams [41,
as state-of-the-art methods for the computational treatment 69–71, 99], a category of formulations, which naturally
of geometrically nonlinear beam problems. An overview preserves the objectivity of the continuous problem. As
of the different FEM discretiation techniques is given in pointed out in [5], these Kirchhoff type formulations
Fig. 1. often exhibit a comparatively poor accuracy; a fact,
In the context of the geometrically nonlinear Kirch- which can directly be traced back to the lack of an exact
hoff–Love beam theory, several discrete realiza- representation of the kinematic quantities. A further criti-
tions based on finite difference schemes have recently cal issue relevant in the context of thin Kirchhoff beams
been proposed [1, 19, 20, 65, 98]. In contrast to the is membrane locking, a locking phenomenon given dis-
Simo–Reissner theory, also several works based on an tinction to by Stolarski and Belytschko [147]. In gen-
analytic treatment of Kirchhoff–Love beam problems eral, membrane locking denotes the inability of curved
exist in the modern literature [97, 136]. Interestingly, structural elements, e.g. beams or shells, to represent the
most approaches of these two categories can be found in inextensibility constraint of vanishing membrane/axial
the field of bio- or molecular physics. Although the theo- strains. The work [55] was one of the first contributions
retical basis of shear-free Kirchhoff–Love beam formula- in which this effect was investigated for Kirchhoff beams
tions has a much longer tradition than the Simo–Reissner by relating the slenderness ratio to the condition number
theory of shear-deformable beams, there are only very of the stiffness matrix, but without explicitly using the
few geometrically nonlinear shear-free finite element rep- term locking. Diverse methods have been proposed in the
resentations, which have not reached the excellent prop- literature in order to avoid membrane locking of Kirch-
erties of geometrically exact Simo–Reissner formula- hoff rods. Amongst others, these are the approaches of
tions so far. In his recent works [5] and [6], Armero and reduced/selective integration (see e.g. [115, 117, 118]),
Valverde gave a historic overview of existing Kirchhoff assumed strains based on the Hu-Washizu functional
finite elements and pointed out their drawbacks. Accord- (see e.g. [36, 88, 100]), assumed stresses based on the
ingly, the first Kirchhoff type element formulations have
Fig. 1 Overview of discretization techniques for nonlinear problems of slender continua based on the finite element method
13
C. Meier et al.
Hellinger–Reissner functional (see e.g. [33, 115]) or pen- contribution of Sansour [129] proposes an energy–momen-
alty relaxation/stabilization techniques in combination tum method for 2D, initially straight geometrically exact
with membrane correction factors (see e.g. [105, 149]). elements based on the nonlinear Euler–Bernoulli theory.
Most of these works are limited to 2D beam problems. A On the other hand, Armero and Valverde developed plane
historic overview on Kirchhoff beam elements in general and three-dimensional locking-free Kirchhoff–Love beam
and the key issues of objectivity and membrane locking element formulations accounting for arbitrarily curved ini-
in particular is given in the recent works of Armero and tial geometries as well as anisotropic cross-section shapes
Valverde [5, 6]. that guarantee the fundamental properties of objectivity
The shear-free beam elements presented up to now and geometrical exactness [5, 6]. However, these beam
are typically based on additional kinematic assumptions, elements only cover the geometrically linear case of infini-
thus not being consistent with the concept of geometri- tesimal deformations, but not the general large deformation
cally exact beams. Besides these contributions considered case.
so far, also finite element and finite difference discretiza- Only a few geometrically exact, 3D, large deformation,
tion approaches for Kirchhoff beams can be found in the shear-free beam elements can be found in the literature.
literature which are in fact geometrically exact, but which Arguably the first formulations of this kind have been pro-
rely on global interpolation strategies [20, 53, 159]. These posed by Boyer [28] and Weiss [156, 157]. In his recent
are typically based on a rotation or curvature interpolation work [27], Boyer extended the original formulation [28] for
strategy and a subsequent integration of the rotation field the modeling of undersea cables. Also the essential require-
along the entire beam length in order to yield an explicit ment of objectivity and path-independence are fulfilled by
beam centerline representation. Unfortunately, these global the approaches of Boyer and Weiss. However, these geo-
approaches typically rely on a serial finite element evalu- metrically nonlinear Euler–Bernoulli formulations only
ation and yield dense system matrices and not the very treat the special case of beams with circular cross-sections
desirable sparse system matrices with small bandwidths as and a straight initial configuration, i.e. the case of beams
typical for standard FEM approaches. Consequently, these with rotationally symmetric reference geometry denoted as
approaches are not suitable for high-performance comput- isotropic case in the following (see first column of Table 2).
ing and will not be considered further in the following. The This limitation simplifies the theory considerably, and
number of existing geometrically exact (locally supported) already the modeling of simple piecewise straight frames is
finite element formulations on the basis of the Kirch- difficult since no variables are available that determine the
hoff–Love theory is very limited. For example, the recent cross-section orientation required for kinematic constraints
Meaning of the symbols: “+” requirement fulfilled, “o” requirement partly fulfilled, “−” requirement not
fulfilled, “?” information not available
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
at beam-to-beam joints. The more recent contributions of 2. Representability of general geometries and external
Zhang and Zhao [160, 161] allow for anisotropic cross- loads The most general case of 3D, large deformation
sections but still focus on initially straight beams (denoted problems of thin beams with (a) anisotropic cross-sec-
as straight case in the second column of Table 2). How- tion shapes as well as (b) arbitrary initial geometries
ever, later it will become clear that the first-order twist and initial curvatures have to be represented. (c) The
angle interpolation underlying these formulations might proposed rotation interpolation schemes have to be
in general not allow for optimal spatial convergence rates capable of representing such general scenarios with-
considering the employed third-order centerline interpola- out exhibiting any singularities for practically relevant
tion. In the very recent works of Greco et al. [66, 67], first configurations.
steps towards geometrically nonlinear isogeometric Kirch- 3. Fulfillment of essential mechanical principles The
hoff–Love beam elements accounting for initial curvature essential mechanical principles of objectivity, i.e.
and anisotropic cross-sections have been made. However, observer or frame invariance, as well as path-inde-
the formulation has only been applied to geometrically lin- pendence have to be preserved by the employed discre-
ear examples. Bauer et al. [16] adapted the ideas of Greco tization schemes.
and facilitated an application to geometrically nonlinear 4. Fulfillment of general requirements on spatial finite
examples. These formulations are denoted as anisotropic element (FE) discretization For all finite element for-
case in Table 2. Unfortunately, important properties such mulations presented in this work, accuracy will be
as objectivity and path-independence have not been consid- verified in terms of convergence towards proper ref-
ered in the works of Greco and Bauer. Also the question erence solutions. Furthermore, the following criteria
of membrane locking has not yet been consistently treated have to be fulfilled: (a) Optimal spatial convergence
within the geometrically nonlinear realm by the isotropic, orders measured in appropriate error norms have to be
straight and anisotropic Kirchhoff–Love formulations con- achieved. (b) Effective methods for the avoidance of
sidered in this paragraph. Nevertheless, they can be con- locking are required such that no remaining locking-
sidered as the most elaborate geometrically exact Kirch- related deterioration of the spatial convergence behav-
hoff–Love beam element formulations in the literature ior will be observed for the finite elements, even in the
since they are the only ones accounting for the treatment range of very high slenderness ratios. (c) Besides these
of general 3D, large deformation problems. Consequently, essential requirements, also conservation properties
these formulations will be in the focus of detailed evalua- such as conservation of linear momentum, conserva-
tions and comparisons performed throughout this work. tion of angular momentum and conservation of energy
for arbitrarily rough spatial discretizations are desir-
able for the proposed spatial interpolation schemes.
1.2 Essential Requirements for Geometrically Exact (d) It can also be regarded as beneficial for beam ele-
Beam Element Formulations ment formulations to fulfill simple patch tests, e.g. to
exactly represent the states of constant axial tension,
All in all, it can be concluded that none of the existing torsion and bending for finite element sizes. While this
geometrically exact shear-free beam elements of Kirch- is not a necessary condition for spatial convergence,
hoff–Love type is comparable to the existing shear-deform- the numerical test cases in this work will show that the
able formulations of Simo–Reissner type in terms of gen- exact representability of the strain energy associated
erality and fulfillment of essential properties so far. Also with these deformation states typically leads to a lower
detailed comparisons of shear-free and shear-deformable discretization error level. (e) The finite elements should
geometrically exact beam elements in the range of high rely on a symmetric discretization in the sense that the
slenderness ratios are still missing. This backlog is the result is independent from the chosen node numbering.
motivation for the development of novel geometrically 5. Fulfillment of general requirements on temporal finite
exact Kirchhoff–Love beam elements fulfilling the follow- difference (FD) discretization The main focus of this
ing essential requirements: work lies on the development and evaluation of spatial
finite element discretizations for geometrically nonlin-
1. Geometrical exactness As already mentioned above, ear beam problems. Nevertheless, it is at least required
the proposed beam element formulations have to that (a) dynamics can be represented in general, and
be geometrically exact in the sense that the derived that (b) energy-stable time integration can be achieved.
deformation measures are consistent with the virtual (c) Often, energy-momentum conserving time integra-
work principle and the equilibrium equations at any tion schemes may be favorable.
deformed state independent of the magnitude of dis- 6. Simple realization of essential boundary conditions
placements, rotations and strains. and joints Choices of nodal primary variables, espe-
13
C. Meier et al.
cially with respect to the rotation parametrization, are formulations from the literature, the author’s recent contri-
demanded that enable the formulation of practically butions [109, 110] as well as the formulations proposed in
relevant Dirichlet boundary conditions but also of the current work. Of course, not all of the stated require-
node-wise joints between several beams without the ments have explicitly been stated in the mentioned refer-
need of additional constraint equations. This is not ences. However, the evaluation results presented in Table 2
standard for most of the existing geometrically exact have been derived on the best of the authors’ investigations
shear-free beam element formulations. and understanding of the respective works. Explanations
7. Avoidance of Lagrange multipliers and saddle point and derivations required for the evaluation of the individ-
systems Existing geometrically exact beam elements of ual properties stated in Table 2 can be found in Sects. 5–10
Simo–Reissner type can be subjected to the Kirchhoff in form of paragraphs beginning with the expression
constraint via additional Lagrange multiplier fields. “Review: ...”. The symbols “+”, “o” and “−” in Table 2
Such a procedure typically results in saddle point sys- mean that a requirement is fully, partly or not fulfilled. The
tems and the need of a special class of linear solvers question mark “?” indicates that the required information
or requires global condensation strategies. The shear- could not definitely be extracted from the literature. For
free beam elements considered here should neither rely clarity, it should be mentioned that the fulfillment of con-
on Lagrange multipliers nor should they yield saddle servation properties is considered twice, once in the con-
point systems. All calculations have to be feasible in an text of the spatial finite element (FE) discretization and
element-local manner. once in the context of the temporal finite difference (FD)
8. Suitability for high-performance computing In the lit- discretization.
erature, several finite difference and finite element dis- According to Table 2, the formulations newly proposed
cretizations of Kirchhoff–Love beams have been pro- in the current work can indeed close most of the gaps left
posed that rely on global strategies for rotation field by existing geometrically exact Kirchhoff–Love formula-
construction. Even though these schemes show oth- tions. However, several existing Simo–Reissner beam ele-
erwise desirable properties, they typically suffer from ment formulations already fulfill most of the stated require-
two elementary drawbacks: mostly, these schemes ments, a fact that might indicate the higher complexity
result in dense discrete system matrices and depend of consistently incorporating the additional constraint of
on a successive, i.e. serial, evaluation of the individual vanishing shear strains in Kirchhoff–Love beam element
finite elements within a discretization. These two prop- formulations. In order to underline this statement, also the
erties make such formulations virtually infeasible for well-known Simo–Reissner type formulation proposed by
very large systems and high-performance computing. Crisfield and Jelenić [43, 83], the first one of this type pre-
From the finite element formulations considered here, serving objectivity and path-independence, is represented
it is required to result in sparse system matrices with as one possible example for Simo–Reissner beam element
small bandwidths and to be suitable for element evalu- formulations in Table 2 (right column).
ation routines in parallel.
1.3 Potential Advantages of Shear‑Free Beam Element
In the authors’ recent contribution [109], the first 3D Formulations
large deformation geometrically exact Kirchhoff–Love
beam element formulation that fulfills the essential prop- Given the excellent properties of existing geometrically
erties of objectivity and path-independence and that is exact shear-deformable finite element formulations, the
capable of representing arbitrary initial curvatures and question arises which benefits can be gained from apply-
anisotropic cross-section shapes has been proposed. In ing Kirchhoff–Love instead of Simo–Reissner beam ele-
the subsequent work [110], also the important question of ment formulations. It is quite obvious, that the range of low
membrane locking has successfully been addressed. The beam slenderness ratios, where shear-deformation is not
current contribution extends these methodologies by pro- negligible, requires the application of beam element for-
viding considerable improvements in terms of accuracy mulations taking this fact into account. Also the underlying
and practical applicability (e.g. concerning the definition of continuum theory, based on an unconstrained 1D Cosserat
Dirichlet boundary conditions and joints), a generalization continuum with pointwise six degrees of freedom, and the
to dynamic problems as well as theoretical and numerical resulting discrete problem statement seem to be easier to
comparisons to existing formulations. In Table 2, the ful- be formulated for the Simo–Reissner case. However, in the
fillment of the essential requirements 1–8 stated above is range of high beam slenderness ratios, thus, exactly in sce-
verified for the existing categories of 3D, large deformation narios prevalent in many of the practically relevant appli-
geometrically exact Kirchhoff–Love beam elements iden- cations mentioned in the beginning of this contribution,
tified so far, i.e. for the isotropic, straight and anisotropic shear-free beam element formulations of Kirchhoff–Love
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
type may exhibit considerable advantages. In the following, an increased computational efficiency. Although this
some of these potential benefits shall be outlined: torsion-free theory does not lie in the focus of the pre-
sent work, it has to be emphasized that the proposed
I. Lower discretization error level The most essential Kirchhoff–Love formulations lie the foundation for
difference between the proposed Kirchhoff–Love and such reduced models.
existing Simo–Reissner beam element formulations
lies in the neglect of shear deformation in the beam These possible benefits will be further detailed in Sect. 8
theory underlying the former category. This property and verified numerically in Sect. 11.
is independent from the chosen discretization strat-
egy. Consequently, if otherwise comparable interpo- 1.4 Original Scientific Contributions and Organization
lation strategies are applied, it can be assumed that of Present Work
the Kirchhoff type formulations require less degrees
of freedom in order to yield the same polynomial After having motivated the intention of the current work,
approximation quality, and eventually the same dis- its main scientific contributions shall briefly be highlighted:
cretization error level, since no additional primary
variables are required in order to represent shear • Two novel rotation interpolation strategies for geometri-
deformation. cally exact Kirchhoff–Love beam element formulations
II. Numerical benefits in the range of high slenderness are proposed: (i) the first one represents an orthonormal
ratios In the range of very high slenderness ratios, interpolation scheme that fulfills the Kirchhoff con-
the influence of the shear modes on the overall beam straint in a strong manner. It can be regarded as gener-
deformation is not only negligible, it can also be very alization of the scheme proposed in [109] that allows for
beneficial to abstain from these high stiffness con- an exact conservation of energy and momentum and for
tributions from a numerical point of view: mechani- alternative sets of nodal degrees of freedom that sim-
cal problems of slender beams typically lead to stiff plify the prescription of essential boundary conditions.
differential equations and ill-conditioned numerical (ii) The space-continuous theory of the second variant
problems deteriorating the performance of time inte- is based on a weak enforcement of the Kirchhoff con-
gration schemes, nonlinear solvers and linear solvers. straint. The discrete realization of the Kirchhoff con-
The avoidance of the stiff shear mode contributions straint relies on a properly chosen collocation strategy.
can considerably improve the situation. Concretely, In combination with the employed smooth centerline
detailed numerical investigations on several numeri- interpolation, also this strategy can completely avoid
cal test cases involving highly slender beams will be any additional Lagrange multipliers. For each of these
considered. These test cases will reveal a considerably two element formulations, two different sets of nodal
improved performance of nonlinear solution schemes rotation parametrizations are proposed. One is based
when Kirchhoff type instead of Reissner type discre- on nodal rotation vectors and one on nodal tangent vec-
tizations are employed. Similar trends will be pre- tors. While these different choices are shown to yield
dicted—at least theoretically—for the behavior of lin- identical FEM solutions, they differ in the resulting per-
ear solvers and time integration schemes. formance of nonlinear solvers and the effort required
III. Smooth geometry representation The proposed Kirch- to prescribe essential boundary conditions and joints.
hoff–Love beam elements are based on C1-continuous The four finite element formulations resulting from a
centerline interpolations. These interpolations will combination of the two interpolation schemes and the
eventually result in smooth beam-to-beam contact two choices of nodal primary variables are subject to
kinematics, a property that is highly desirable in order detailed comparisons with respect to resulting discre-
to yield efficient and robust contact algorithms (see tization error levels and the performance of nonlinear
e.g. [111, 112]). solution schemes.
IV. Derivation of reduced models In [110], a special • The resulting finite element formulations are combined
reduced model, denoted as torsion-free beam theory, with a finite difference time integration scheme for large
has been derived from the general Kirchhoff–Love rotations recently proposed in the literature by Brüls and
theory. This reduced model has been shown to only Cardona [31]. This implicit scheme allows for energy-
be valid for special problem classes concerning beam stable, second-order accurate time integration on the
geometry and external loads, which are, however, basis of optimized numerical dissipation and can be
present in many fields of application. The finite ele- identified as a Lie-group extension of the well-known
ments resulting from such a reduced model typically generalized-𝛼 scheme. Up to the best of the authors’
feature a simplified numerical implementation and knowledge, the current work represents the first appli-
13
C. Meier et al.
cation of a Lie group time integration scheme based on the resulting finite element formulations will be further
optimized numerical dissipation to geometrically exact detailed. Also detailed explanations and derivations for the
Kirchhoff–Love beam elements. evaluation results of Table 2 will be presented in Sects. 5
• Furthermore, the current work intends to review and and 6. In the subsequent Sects. 7–10, different finite ele-
evaluate existing geometrically exact beam element for- ment realizations resulting from the proposed interpolation
mulations and to compare these with the formulations schemes and rotation parametrizations will be presented
newly proposed in this work. Concretely, the fulfillment and the fulfillment of the basic requirements stated above
of the essential requirements according to Table 2 will will be confirmed analytically. Concretely, Sect. 7 presents
be studied analytically for the three identified catego- the basics of the geometrically exact Simo–Reissner type
ries of isotropic, straight and anisotropic formulations, formulation proposed in [43, 83], which will serve as ref-
which can be identified as the most general and elabo- erence for several numerical comparisions performed in
rate 3D, large deformation geometrically exact Kirch- Sect. 11. In the subsequent Sect. 8, the benefits of applying
hoff–Love formulations previously proposed in the Kirchhoff–Love instead of Simo–Reissner element formu-
literature, for one exemplary representative of Simo– lations in the range of high slenderness ratios will be fur-
Reissner formulations from the literature as well as for ther quantified. Afterwards, the two major developments of
the Kirchhoff–Love formulations presented in the cur- this work, a Kirchhoff–Love element formulation based on
rent and former works of the authors. A further original strong constraint enforcement as well as a Kirchhoff–Love
contribution of this work is given by the detailed and element formulation based on weak constraint enforcement,
systematic numerical comparison performed between will be proposed in Sects. 9 and 10. For each formulation,
the proposed geometrically exact Kirchhoff–Love and two different variants of nodal rotation parametrization,
two representative Simo–Reissner beam element formu- one based on nodal rotation vectors and one based nodal
lations from the literature. Specifically, resulting spatial tangent vectors will be derived. Eventually, in Sect. 11,
convergence rates, discretization error levels as well as all the proposed concepts and the resulting finite element
the performance of nonlinear solution schemes are com- formulations will be verified by means of proper numeri-
pared for different beam slenderness ratios. cal test cases. The reader who is primarily interested in the
practical implementation of the newly proposed geometri-
Eventually, the organization of the remainder of this cally exact Kirchhoff–Love beam element formulations is
contribution shall briefly be given. One distinctive property referred to Sects. 5, 9, 10 as well as to the appendices of
of geometrically exact beam formulations is the presence this work.
of large rotations within the associated configuration space.
In order to provide the theoretical basis for subsequent
derivations, in the following Sect. 2, the SO(3) group of 2 The Rotation Group SO(3)
large rotations as well as possible parametrizations will be
introduced. In Sect. 3, the most general type of geometri- The category of beam theories considered in this contri-
cally exact beam formulations considered in this work, the bution assumes the beam cross-sections to be rigid. Con-
Simo–Reissner theory of thick (shear-deformable) beams, sequently, the cross-section kinematics are point-wise
will be presented. Subsequently, in Sect. 4, the general the- uniquely defined by six degrees of freedom, three transla-
ory will be restricted to the Kirchhoff–Love theory of thin tional ones representing the position vector of the cross-
(shear-free) beams. There, the different methodologies of section centroid and three rotational ones describing the
imposing the Kirchhoff constraint of vanishing shear strains cross-section orientation. Thereto, an orthonormal triad
in a strong or weak sense will be investigated. Afterwards, consisting of the base vectors 𝐠1 , 𝐠2 , 𝐠3 ∈ ℜ3 is attached on
the space-time-continuous beam problem will be discre- the beam cross-sections. Furthermore, a right-handed iner-
tized. In Sect. 5, a recently proposed extension of the gen- tial Cartesian frame 𝐄1 , 𝐄2 , 𝐄3 ∈ ℜ3 associated with the
eralized-𝛼 method from vector spaces to Lie groups [7, 31, material configuration and a corresponding right-handed
32], will be applied to the beam formulations proposed in inertial Cartesian frame 𝐞1 , 𝐞2 , 𝐞3 ∈ ℜ3 of the spatial con-
this work. In the subsequent Sect. 6, spatial discretization is figuration are introduced. Nevertheless, for simplicity,
performed, which represents the core topic in the develop- it is assumed that both frames coincide, thus 𝐞i = 𝐄i for
ment of geometrically exact Kirchhoff beam elements con- i = 1, 2, 3. Now, the rotation from the global frame 𝐄i onto
sidered in this work. In Sects. 6.1 and 6.2, specific finite the local frame 𝐠i is described via the orthogonal transfor-
element interpolations employed to the translational and mation 𝚲 ∈ SO(3):
rotational primary variable fields considered in this work with 𝚲 = 𝐠j ⊗ 𝐄j = (𝐠1 , 𝐠2 , 𝐠3 )𝐄j
will be proposed. In Sect. 6.3, the already stated essential
𝐠i = 𝚲𝐄i (1)
requirements on the employed spatial discretizations and
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
for i, j = 1, 2, 3. Here and in the following, the summation is presented. The rotation vector parametrization is given by
convention over repeated indices holds. Throughout this the well-known Rodrigues formula:
work, the index near a matrix [for example the index (.)𝐄i in
Eq. (1)] denotes the basis in which the associated tensor is 𝚲(𝝍) = exp(𝐒(𝝍))
[ ] (6)
represented. In the context of geometrically exact beam the- = 𝐈 + sin 𝜓𝐒(e𝝍 ) + (1 − cos 𝜓)𝐒(e𝝍 )𝐒(e𝝍 ) .
ories, the two-point tensor 𝚲 acts as push-forward operator
(see e.g. [107]) between material and spatial objects. From Here, 𝜓 = ||𝝍|| represents the scalar rotation angle and
a rather mathematical point of view, the rotation tensor 𝚲 𝐞𝝍 = 𝝍∕||𝝍|| the axis of rotation. Throughout this work,
can be identified as an element of the Special Orthogonal ||(.)|| denotes the Euclidean norm in ℜ3. As indicated by the
group SO(3) of orthogonal transformations according to notation exp (𝐒(𝝍)), Eq. (6) represents a closed-form repre-
sentation of the exponential map initially introduced in (5)
SO(3): = {𝚲 ∈ ℜ3×3 |𝚲T 𝚲 = 𝐈3 , det(𝚲) = 1}, (2) of the last section. The rotation vector of a given rotation
under the action of non-commutative multiplication tensor can for example be extracted by employing Spur-
SO(3) × SO(3) → SO(3), 𝚲3 = 𝚲1 𝚲2 ≠ 𝚲2 𝚲1 with inverse rier’s algorithm [146]. The infinitesimal spatial quantity 𝛿𝜽,
element 𝚲−1 = 𝚲T and identity element 𝐈3. Here, det(.), (.)T denoted as multiplicative rotation vector variation or spin
and (.)−1 are the determinant, transpose and inverse. 𝐈3 is a vector, allows to express the variation of 𝚲:
3 × 3 identity matrix. The nonlinear manifold SO(3) can be
classified as Lie group with tangent space d |
𝛿𝚲: = d𝜖 | exp (𝜖𝐒(𝛿𝜽))𝚲 = 𝐒(𝛿𝜽)𝚲,
|𝜖=0 (7)
𝛿𝚲 ∈ T𝚲 SO(3): = {𝐒(𝐚)𝚲|𝐒(𝐚) ∈ so(3)}. (3) or 𝛿𝐠i : = 𝛿𝜽 × 𝐠i .
Here, so(3) denotes the set of skew symmetric tensors with
Alternatively, the variation 𝛿𝚲 can be expressed by the
𝐒(𝐚)𝐛 = 𝐚 × 𝐛 for 𝐚, 𝐛 ∈ ℜ3. The isomorphism between
material spin vector 𝛿𝚯 = 𝚲T 𝛿𝜽 via right translation:
so(3) and ℜ3 enables a unique expression of 𝐒(𝐚) ∈ so(3)
by the vector 𝐚 ∈ ℜ3 denoted as axial vector. By insert- d|
ing the special choice 𝚲 = 𝐈3 into (3) it can easily be veri- 𝛿𝚲: = | 𝚲 exp (𝜖𝛿𝚯) = 𝚲𝐒(𝛿𝚯). (8)
d𝜖 |𝜖=0
fied that so(3) represents the tangent space to SO(3) at the
Expressing the spin vectors 𝛿𝚯 and 𝛿𝜽 in the associated
identity:
frames 𝐄i and 𝐠i, respectively, yields:
so(3): = T𝐈 SO(3) = {𝐒(𝐚)|𝐒(𝐚) = −𝐒(𝐚)T ∀ 𝐚 ∈ ℜ3 }. (4)
𝛿𝚯 = 𝛿Θi 𝐄i , 𝛿𝜽 = 𝛿Θi 𝐠i . (9)
The Lie group SO(3) and its Lie algebra so(3) are related
Thus, the components of the spatial spin vector expressed
by the exponential map exp (.):so(3) → SO(3):
in the local basis 𝐠i are identical to the components of the
𝐒(𝐚)2 𝐒(𝐚)3 material spin vector expressed in the global basis 𝐄i. This
exp (𝐒(𝐚)): = 𝐈3 + 𝐒(𝐚) + + + ⋯. (5)
2! 3! relation also holds for all the other pairs of spatial/material
quantities considered in this work (see e.g. [138]). Based
Up to now, the rotation group SO(3) has been introduced
on the defining Eq. (8) and the representation of skew-sym-
without stating any specific parametrization of the rota-
metric tensors by means of the Levi-Civita-Symbol 𝜖ijk, the
tion tensor 𝚲. In the following sections, two possible par-
components of 𝐒(𝛿𝚯) can be determined:
ametrizations, which will be useful in the development of
beam elements according to the Simo–Reissner and Kirch- 𝐒(𝛿𝚯) = 𝚲T 𝛿𝚲 = (𝐄i ⊗ 𝐠i ) ⋅ (𝛿𝐠j ⊗ 𝐄j )
hoff–Love theory, will be presented. = 𝐠Ti 𝛿𝐠j 𝐄i ⊗ 𝐄j with Sij (𝛿𝚯) = −𝜖ijk 𝛿Θk . (10)
���
=:Sij (𝛿𝚯)
2.1 SO(3) Parametrization Via Rotation Vectors
For later use, the components 𝛿Θi shall be expressed by 𝐠i
There exist various parametrizations of the rotation tensor
and 𝛿𝐠i. From the previous Eq. (10), it follows:
such as rotation (pseudo-) vectors, Euler angles or Rodrigues
parameters that are based on a minimal set of three param- 𝛿Θ1 = 𝐠T3 𝛿𝐠2 = −𝐠T2 𝛿𝐠3 ,
eters. Also four-parametric representations of the rotation
tensor such as quaternions have proven to be very useful for
𝛿Θ2 = 𝐠T1 𝛿𝐠3 = −𝐠T3 𝛿𝐠1 , (11)
practical purposes. Within this work, two different parameter- 𝛿Θ3 = 𝐠T2 𝛿𝐠1 = −𝐠T1 𝛿𝐠2 .
izations will be employed: the one presented in this section is
based on rotation vectors 𝝍 ∈ ℜ3. In Sect. 2.2, an alternative A relation between infinitesimal additive and multiplicative
parametrization especially suited for Kirchhoff formulations increments is given by the tangent operator 𝐓 defined via
𝛿𝝍 = :𝐓𝛿𝜽 with
13
C. Meier et al.
𝐓= 1
𝐒(𝝍)𝐒(𝝍)T a discussion of possible singularities of the SR mapping).
𝜓2 ( ) (12) Subsequently to the definition of an intermediate triad 𝚲M
𝜓∕2
+ tan (𝜓∕2) 𝐈 − 𝜓12 𝐒(𝝍)𝐒(𝝍)T − 12 𝐒(𝝍).
according to (15), the cross-section triad 𝐠i can be defined
based on a relative rotation of the intermediate triad 𝚲M
The inversion of the previous Eq. (12), expressing mul- with respect to the tangent 𝐭 by an angle of 𝜑:
tiplicative by means of additive increments, is given by
𝛿𝜽 = 𝐓−1 𝛿𝝍 with: 𝚲 = exp (𝐒(𝜑𝐠1 ))𝚲M (16)
( ) Equations (15) and (16) uniquely define a triad 𝚲 para-
𝐓−1 = 𝜓12 1 − sin𝜓(𝜓) 𝐒(𝝍)𝐒(𝝍)T metrized by the four degrees of freedom (𝐭, 𝜑). Evidently,
(13) one of these four degrees of freedom, namely the norm ||𝐭||
+ sin𝜓(𝜓) 𝐈 + 1−cos (𝜓)
𝐒(𝝍).
𝜓2
of the tangent vector, will not influence the triad orienta-
For details on the derivation of the transformations 𝐓 and tion. However, as it will turn out in the next sections, the
𝐓−1, the interested reader is e.g. referred to [34, 40, 77, non-unit tangent vector 𝐭 is a quantity that directly results
141]. While the rotation vector parametrization presented from the beam centerline description. Thus, the only addi-
so far represents a well-known tool in the formulation of tional degree of freedom introduced in order to describe the
geometrically exact beam elements of Simo–Reissner type, triad orientation is the relative angle 𝜑 and consequently,
in the following section, an alternative parametrization of the proposed type of triad parametrization is not redundant.
large rotations will be proposed which offers some advan- For later use, the spatial spin vector 𝛿𝜽 shall be expressed
tages in the description of Kirchhoff type beam elements. by means of additive increments (𝛿𝐭, 𝛿𝜑) of the four param-
eters (𝐭, 𝜑). Therefore, it is split into a component 𝛿𝜽∥
parallel to 𝐠1 and a component 𝛿𝜽⟂ perpendicular to 𝐠1 as
2.2 SO(3) Parametrization Via “Smallest Rotation” follows:
The alternative parametrization considered in this section 𝛿𝜽 = 𝛿𝜽∥ + 𝛿𝜽⟂ = 𝛿Θ1 𝐠1 + 𝛿𝜽⟂ . (17)
consists of four degrees of freedom (𝐭, 𝜑) with 𝐭 ∈ ℜ3 and Throughout this work, the indices (.)∥ and (.)⟂ of a vector
𝜑 ∈ ℜ. In the context of Kirchhoff beam elements pre- will denote the components of the vector which are paral-
sented later in this work, 𝐭 will be the non-unit tangent vec- lel or perpendicular to the vector 𝐠1, respectively. Taking
tor aligned to the space curve representing the beam cen- advantage of (7), 𝛿𝜽⟂ can be derived (see [110]):
terline. Due to the Kirchhoff constraint of vanishing shear
( )
strains, the first base vector 𝐠𝟏 of the cross-section triad 𝐭 𝐭 𝐭 × 𝛿𝐭
𝛿𝜽⊥ = 𝐠1 × 𝛿𝐠1 = ×𝛿 =
𝚲 = (𝐠1 , 𝐠2 , 𝐠3 )𝐄i, can be expressed by this tangent vector: ||𝐭|| ||𝐭|| ||𝐭||2
(18)
𝐭 𝐠 × 𝛿𝐭
𝐠𝟏 = . (14) = 1 with 𝛿𝐭 ∈ ℜ3 .
||𝐭|| ||𝐭||
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
𝛿t = 𝛿||𝐭|| =
𝐭T
𝛿𝐭 = 𝐠T1 𝛿𝐭. (23) In this section, the fundamentals of the geometrically exact
t Simo–Reissner beam theory based on the work of Reiss-
Combining Eqs. (21) and (23) yields a transformation rule ner [120, 121] as well as Simo and Vu-Quoc [138, 140]
between the two sets (𝛿𝜽, 𝛿t) and (𝛿𝐭, 𝛿𝜑): will be briefly summarized. The results of this section will
provide an essential basis for the subsequent derivation of
( ) ( )( ) Kirchhoff type beam formulations.
𝛿𝜽 𝐓𝜽𝐭 𝐠1 𝛿𝐭
=
𝛿t 𝐠T1 0 𝛿𝜑
⏟⏞⏞⏞⏟⏞⏞⏞⏟ 3.1 Basic Kinematic Assumptions
=:𝐓−1
( ) ( )( )
M
13
C. Meier et al.
the arc-length parameter s. The rotation from the global where 𝐤(s) and 𝐊(s): = 𝚲T (s)𝐤(s) are referred to as spatial
frame 𝐄i onto the initial local frame 𝐠0i (s) is described and material curvature vector defined as:
via the orthogonal transformation s → 𝚲0 (s) ∈ SO(3)
as introduced in (1) leading to the following defini- 𝐒(𝐤(s)) = 𝚲� (s)𝚲T (s), 𝐒(𝐊(s)) = 𝚲T (s)𝚲� (s),
tion s → C0 : = (𝐫0 (s), 𝚲0 (𝝍 0 (s))) ∈ ℜ3 × SO(3) of the (28)
𝐤 = Ki 𝐠i , 𝐊 = Ki 𝐄i .
initial configuration. Correspondingly, the deformed
configuration of the beam at time t ∈ ℜ is given by In a similar manner, the spatial and material angular veloc-
s, t → C: = (𝐫(s, t), 𝚲(𝝍(s, t))) ∈ ℜ3 × SO(3), where the ity vectors 𝐰 and 𝐖 are defined according to:
orthogonal transformation 𝚲(𝝍(s, t)) maps from the global
. .
frame 𝐄i onto the current local frame 𝐠i (s) and the base vec- 𝐒(𝐰(s)) = 𝚲(s)𝚲T (s), 𝐒(𝐖(s)) = 𝚲T (s) 𝚲(s),
tor 𝐠1 (s, t): = 𝐠2 (s, t) × 𝐠3 (s, t) is in general not tangential (29)
𝐰 = Wi 𝐠i , 𝐖 = Wi 𝐄i .
to the deformed centerline anymore due to shear deforma-
tion. According to Sect. 2, 𝚲(s, t) can be represented by For completeness, also the spin vectors, which have been
three rotation parameters (e.g. by a rotation vector 𝝍(s, t)), introduced in Sect. 2, are repeated here:
leading to pointwise six, three translational and three rota-
tional, degrees of freedom. The basic kinematic assumption 𝐒(𝛿𝜽(s)) = 𝛿𝚲(s)𝚲T (s), 𝐒(𝛿𝚯(s)) = 𝚲T (s)𝛿𝚲(s),
(30)
of the geometrically exact Simo–Reissner theory consid- 𝛿𝜽 = 𝛿Θi 𝐠i , 𝛿𝚯 = 𝛿Θi 𝐄i .
ered so far can easily be summarized by the following con-
strained position vector field associated with the initial and Throughout this work, the dot (.) ̇ = d (.) denotes
dt
deformed configuration the derivative with respect to time t. By applying
the Young theorem (𝚲̇ � ) = (𝚲)̇ � and making use of
𝐒(𝐚)𝐒(𝐛) − 𝐒(𝐛)𝐒(𝐚) = 𝐒(𝐒(𝐚)𝐛), the following compatibil-
(26) ity relations can be shown:
. .
𝐤 = 𝐰� + 𝐰 × 𝐤, 𝐊 = 𝐖� − 𝐖 × 𝐊,
𝛿𝐤 = 𝛿𝜽� + 𝛿𝜽 × 𝐤, 𝛿𝐊 = 𝛿𝚯� − 𝛿𝚯 × 𝐊, (31)
where (s2 , s3 ) represent convective coordinates describing . .
the position of an arbitrary material point within the (rigid) 𝛿𝐰 = 𝛿 𝜽 +𝛿𝜽 × 𝐰, 𝛿𝐖 = 𝛿 𝚯 −𝛿𝚯 × 𝐖.
cross-section. In order to simplify the notation for subse-
quent derivations, the convective coordinate vector given In the next sections, stress resultants, mechanical equilib-
by as well as the redundant name s1 : = s rium and constitutive relations will be presented.
for the arc-length s has been introduced. The kinematic
quantities defining the initial and deformed configuration 3.2 Stress Resultants, Mechanical Equilibrium
as introduced so far are illustrated in Fig. 2. In order to sim- and Objective Deformation Measures
plify notation, the time argument t will be dropped in the
following. Later in this section, the arc-length derivative of With ̃𝐟 and 𝐦 ̃ denoting distributed external forces and
the base vectors 𝐠i (s) will be required. Similar to Eq. (7), moments per unit length and 𝐟𝜌 and 𝐦𝜌 representing the
this derivative can be formulated as forces and moments due to inertia effects, the strong form
of equilibrium reads (see e.g. [3, 121, 138]):
𝚲� (s) = 𝐒(𝐤(s))𝚲(s) or 𝚲� (s) = 𝚲(s)𝐒(𝐊(s)), (27)
Fig. 2 Kinematic quantities defining the initial and deformed configuration of the geometrically exact beam
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
𝐟� + ̃
𝐟 + 𝐟𝜌 = 𝟎, 𝛿o 𝜸 = 𝛿𝐫 �(− 𝛿𝜽 × 𝐫 � , ) 𝛿o 𝝎 = 𝛿𝜽� ,
(32) (36)
𝐦 + 𝐫� × 𝐟 + 𝐦
� ̃ + 𝐦𝜌 = 𝟎. 𝛿𝚪 = 𝚲T 𝛿𝐫 � + 𝐫 � × 𝛿𝜽 , 𝛿𝛀 = 𝚲T 𝛿𝜽� .
Here, E and G are the Young’s modulus and the shear mod-
Here, 𝐟𝜎 and 𝐦𝜎 denote external forces and moments at the ulus, A, Ā 2 and Ā 3 are the cross-section and the two reduced
Neumann boundary 𝛤𝜎 of the considered beam. Based on cross-sections, I2 and I3 are the two principal moments of
the principle of virtual work, the following objective spatial inertia and IT is the torsional moment of inertia. Similarly
deformation measures can be identified: to (37), the length-specific kinetic energy Π ̃ kin of the beam
𝜸 = 𝐫 � − 𝐠1 , can be formulated:
13
C. Meier et al.
∫
𝐟ext : = ̃
𝐟ds + [𝐟𝜎 ]Γ𝜎 , theory and the subsequent discretization procedures. Thus,
the reader may alternatively skip these content and proceed
0
l with Sect. 4. Concretely, in Sect. 3.4, the alternative proce-
( ) dure of deriving the weak and strong form of the balance
∫
𝐦ext : = 𝐫 ×̃ ̃ ds + [𝐫 × 𝐟𝜎 + 𝐦𝜎 ]Γ𝜎 ,
𝐟 +𝐦 equations presented in Sect. 3.2 on the basis of a variational
0
(41) problem statement and properly defined Lagrangian is
l l
undertaken. Section 3.5 is meant to confirm the constitutive
∫ ∫
Πint : = ̃ int ds, Πkin : =
Π ̃ kin ds,
Π relations presented above based on a derivation from 3D
0 0 continuum mechanics. There, the consistency of the 1D and
l l 3D continuum model in the sense of a fully induced beam
theory will be verified for the case of locally small strains.
∫ ∫
𝐥a: = ̃𝐥ds, 𝐡: = (̃𝐡 + 𝐫 × ̃𝐥)ds.
0 0
Remark Two possible time integration schemes can be
These definitions will be required subsequently. Based derived from the variants of either employing 𝐖 and 𝐀
on (40), the inertia forces 𝐟𝜌 and moments 𝐦𝜌 yield: directly or expressing them via additive rates (44):
̇ 𝐚 = 𝐓̇ −1 𝝍̇ + 𝐓−1 𝝍.
𝐰 = 𝐓−1 𝝍, ̈ (44)
Finally, the problem setup presented in Sects. 3.1–3.3 3.4 Interlude: Variational Problem Formulation
has to be completed by boundary and initial conditions in of Simo–Reissner Beam Theory
order to end up with a well-defined initial boundary value
problem: For the case that no external forces act on the beam, i.e.
𝐫 = 𝐫u , 𝝍 = 𝝍 u on 𝛤u , 𝐟𝜎 = ̃
𝐟 = 𝐦𝜎 = 𝐦 ̃ = 𝟎, the strong and weak forms (32)
𝐟 = 𝐟𝜎 , 𝐦 = 𝐦𝜎 on𝛤𝜎 , and (33) shall equivalently be formulated on the basis of the
𝛤𝜎 ∩ 𝛤u = ∅, 𝛤𝜎 ∪ 𝛤u = {0, L} (45) well-known Hamilton principle:
𝐫 = 𝐫0 , 𝐫̇ = 𝐯0 , 𝚲 = 𝚲0 , 𝐰 = 𝐰0 at t = 0. T
dt = 0 with
Based on a trial space (𝐫(s, t), 𝚲(𝝍(s, t))) ∈ of func- �
𝛿 [𝛿𝐫 = 𝛿𝜽 = 𝟎]t=T
t=0
. (47)
t=0
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
� �
T
= (𝛿 𝜽̇ ̃𝐡 + 𝛿 𝐫̇ T̃𝐥 − 𝛿𝜽�T 𝐦 (51)
t=0 s=0 (49)
[ ]T
− 𝛿𝐫 � − 𝛿𝜽 × 𝐫 � 𝐟 )ds dt
T l The (non-orthonormal) covariant bases and
have to be determined from (26):
� �
T T �T
= (𝛿𝜽 𝐦𝜌 + 𝛿𝐫 𝐟𝜌 − 𝛿𝜽 𝐦
t=0 s=0
[ ]T
− 𝛿𝐫 � − 𝛿𝜽 × 𝐫 � 𝐟 )ds dt = 0.
(52)
From the first to second line, [ the objective] variations Also the contravariant base vectors can be determined
̇ 𝛿o 𝝎 = 𝛿𝜽� and 𝛿o 𝜸 = 𝛿𝐫 � − 𝛿𝜽 × 𝐫 � [see (31)
𝛿o 𝐰 = 𝛿 𝜽, from the second line of (52) via
and (36)] have been inserted. Furthermore, the additional
relation 𝛿o 𝐫̇ T 𝐫̇ = (𝛿 𝐫̇ − 𝛿𝜽 × 𝐫)
̇ T 𝐫̇ = 𝛿 𝐫̇ T 𝐫̇ has been applied.
From the second to the third line, partial integration of the (53)
inertia terms together with the boundary conditions in (47)
have been applied. Since no external forces and moments
are considered, i.e. 𝐟𝜎 = 𝐦𝜎 = 𝟎, partial integration of (49) Making use of the relation 𝐫 � = 𝐠1 + 𝜸 (34) and inserting
yields: the first line of (52) as well as (53) into (51) yields:
T ⎛ T 1[ ( )]
dt = � 𝛿𝜃 T [𝐦� + 𝐫 � × 𝐟 + 𝐦𝜌 ]
⎜ l 𝓕= 𝐠 + 𝜸 + 𝐤 × s2 𝐠2 + s3 𝐠3 ⊗ 𝐠01
� � s=0 ⎜⎜
𝛿 C 1 [ K s ] [ K01 s2 ] (54)
������������������� + 𝐠2 ⊗ 𝐠02 + 01C 3 𝐠01 + 𝐠3 ⊗ 𝐠03 − C
𝐠01 .
t=0 t=0 ⎝ ̇
=0
(50)
⎞ Finally, by inserting (54) into (51), the individual com-
⎟ ponents of the Cauchy–Green deformation tensor can be
+ 𝛿𝐫 T [ 𝐟 � + 𝐟𝜌 ]⎟dsdt = 0.
��� ⎟ determined. However, in order to gain further insight into
̇
=0 ⎠ the underlying structure of the deformation gradient, the
very ellegant procedure suggested by Geradin and Car-
The arbitrariness of 𝛿𝜽(s, t) and 𝛿𝐫(s, t) directly yields the dona [59] (for initially straight beams) as well as Linn
strong form (32). Inverting the last step from (49) to (50), by et al. [103] (for initially curved beams) is employed by
partial integration of the weighted strong form (32), this time slightly reformulating (54) on the basis of the relative rota-
only along the beam length l, yields the weak form of the bal- tion tensor 𝚲:̃ = 𝚲𝚲T . By applying the auxiliary relation
0
ance equations (33). The terms occurring in this weak form 1∕C = 1 + (1 − C)∕C to the pre-factor of 𝐠1 and solving all
for an unloaded beam can already be identified in the third the products in (54), the deformation gradient can eventu-
line of (49). While conservative external forces could also ally be reformulated according to:
be included into the Lagrangian (48), for a consideration of ( )
non-conservative external forces and 3D external moments, � 𝐈3 + 𝐇 ⊗ 𝐠01
ℱ=𝚲
which are known to be non-conservative (see e.g. [140]), the (55)
� = 𝐠i ⊗ 𝐠0i .
with 𝐇: = Hi 𝐠0i , 𝐈3 = 𝐠0i ⊗ 𝐠0i , 𝚲
starting point has to be the strong form of equilibrium (32).
The components of the vector H identified in (55), denoted
3.5 Interlude: Relation Between 1D and 3D as material strain vector in [59] and [103], read:
Constitutive Laws
1[ ]
H1 = Γ1 + (K2 − K02 )s3 − (K3 − K03 )s2 ,
The aim of the following considerations is to derive the con- C
stitutive laws (37) and (38) in a consistent manner from the 1[ ]
H2 = Γ2 − (K1 − K01 )s3 , (56)
3D continuum theory. Thereto, the deformation gradient 𝓕 of C
the 3D position field subject to the kinematic constraints (26) 1[ ]
H3 = Γ3 − (K1 − K01 )s2 .
C
13
C. Meier et al.
Based on the deformation gradient (55), the Cauchy–Green vanishing in-plane stress components ̄ 22 = ̄ 33 = 0 and of
deformation tensor can be derived according to: a constitutive parameter E in front of the normal stress ̄ 11,
only holds for the special case 𝜈 = 0. This is a consequence
1 [ ] HT H
𝓔= 2C
H ⊗ 𝐠01 + 𝐠01 ⊗ H + 𝐠
2C2 01
⊗ 𝐠01 . (57) of the kinematic assumption of rigid cross-sections, which
requires the existence of in-plane reaction forces in general.
The result (57) has been consistently derived from the basic In order to resolve these two putative contradictions for
kinematic assumptions (26) without any additional approx- general cases 𝜈 ≠ 0, the constraint of rigid cross-sections
imations. However, in order to finally end up with the con- can be weakened by allowing for a uniform lateral contrac-
stitutive laws of the geometrically exact beam theory based tion of the cross-section with in-plane strain components
on the quadratic form (37), the well-known assumption of ℰ̄22 = ℰ̄33 = −𝜈E11 and for a proper in-plane warping field
small strains is made: (see [103] or [156] for further details). Alternatively, the
Γi ≪ 1, R ⋅ Ki ≪ 1, R ⋅ K0i ≪ 1 for i = 1, 2, 3. (58) approximation 𝜈 = 0 can be employed in the first two lines
of (61). In praxis, this slight inconsistency is often taken
The assumptions (58) state that small local axial and shear into account, which is not unusual in the field of structural
strains are considered and that the radii of initial and theories (see e.g. [3] or [90]). In a last step, the material
deformed centerline curvature have to be small as com- force and moment stress resultants are determined by inte-
pared to the cross-section radius R. In the following, a gration of the stress vector acting on a beam
first-order approximation in these small quantities is con- cross-section with material normal vector 𝐠01 according to
sidered by setting C ≈ 1 and neglecting the last, quadratic the definition proposed in [138]:
term in (57). Based on these small-strain assumptions, the
approximated Cauchy–Green deformation tensor eventually
reads
(63)
1[ ]
̄ =
𝓔≈𝓔 2
H ⊗ 𝐠01 + 𝐠01 ⊗ H = :ℰ̄ij 𝐠0i ⊗ 𝐠0j . (59)
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
the components of the alternative material objects, e.g. of freedom (𝐫(s), 𝜑(s)) that fulfills the Kirchhoff con-
̃ = Fi 𝐠0i, when expressed with respect to the “curved”,
𝐅 straint in a strong manner as already derived in [109] will
local basis 𝐠0i are identical to the components of the origi- be briefly repeated and extended to the dynamic case. As
nal material objects, e.g. 𝐅 = Fi 𝐄i, when expressed with already introduced in Sect. 2.2, the scalar-valued quantity
respect to the “straight”, global basis 𝐄i. Via push-forward, 𝜑(s) will describe the relative rotation between the material
also the 1. Piola–Kirchhoff stress tensor and the Cauchy frame 𝐠i (s) and an intermediate frame 𝐠Mi (s) with respect to
stress tensor 𝝈 can be determined. the tangent vector 𝐭(s) according to (16). By means of (15),
one example for a suitable intermediate frame 𝐠Mi (s) has
(65) already been given, the “Smallest Rotation” intermedi-
ate frame. Nevertheless, the following derivations are pre-
Starting with the 1. Piola–Kirchhoff stress tensor or with sented in a general manner, which allows to insert arbitrary
the Cauchy stress tensor, the spatial stress resultants, e.g. alternative intermediate frames 𝐠Mi (s).
𝐟 = Fi 𝐠i, can be derived similarly to (63) and (64). Since
again only first-order terms of the small strains (58) are rel-
4.1 Kinematics
evant, it is sufficient to approximate the deformation gradi-
ent, required for the push-forward, according to:
In the next three sections, the Kirchhoff constraint (67) is
(66) strongly enforced [see (16)] according to
Consequently, as already postulated in the sections before,
𝐭
the relevant pull-back/push-forward operator is given by 𝐠1 = 𝐠M1 = , 𝐭 = 𝐫�,
||𝐭|| (68)
a rotation tensor 𝚲
̃ . If for the derivations above a 3D con-
tinuum formulation with material strain and stress meas- 𝚲 = exp(𝐒(𝜑𝐠1 ))𝚲M , 𝚲M = (𝐠M1 , 𝐠M2 , 𝐠M3 )𝐄i ,
ures based on a straight reference configuration had been
applied, the resulting 1D material objects would be based where the centerline-aligned intermediate triad base vec-
on the global basis 𝐄i and the total rotation tensor 𝚲 tors 𝐠Mi (s) = 𝐠Mi (𝐫 � (s)) are completely defined by the
could be identified as the relevant pull-back/push-forward centerline 𝐫(s) but not further specified for now [an
operator. example is given by (15)]. The Kirchhoff constraint is
incorporated by expressing the current configuration
s, t → C: = (𝐫(s, t), 𝚲(𝐫(s, t), 𝜑(s, t))) ∈ ℜ3 × SO(3) via the
new set of primary variables (𝐫(s, t), 𝜑(s, t)). Inserting (68)
4 Kirchhoff–Love Beam Theory in (28) yields (see [109]):
13
C. Meier et al.
In analogy to Reissner beam formulations, the first com- The inertia forces 𝐟𝜌 as well as the inertia moments 𝐦𝜌 are
ponent 𝛿Θ1 of the spin vector, representing a multiplicative identical to (42) and (43). Like in the Reissner case, the
increment, will directly be employed in the weak form and spatial or material angular velocities 𝐰 and 𝐖 as well as the
not further expressed via additive increments according to spatial or material angular accelerations 𝐚 and 𝐀 can either
𝛿Θ1 = 𝛿ΘM1 + 𝛿𝜑. Consequently, the admissible variations be directly used in the employed time integration scheme or
are s → 𝛿C: = (𝛿𝐫(s), 𝛿𝜽(𝛿𝐫(s), 𝛿Θ1 (s), 𝐫(s, t))) ∈ ℜ3 × ℜ3 they can be expressed via the (additive) rate of the primary
with the new set of variational primary variables variables (𝐫(s), 𝜑(s)). For the latter approach, transforma-
(𝛿𝐫(s), 𝛿Θ1 (s)) defining the Kirchhoff case. Again, compat- tion matrices depending on the definition of the employed
ibility conditions similar to (31) can be stated. Due to the intermediate triad [see e.g. 𝐓−1M
in (24) in the case of the
Kirchhoff constraint, these conditions are only required for SR intermediate triad] as well as their time-derivatives are
the tangential vector components. Left-multiplication of required in order to formulate relations similar to (44).
the second line of (31) with 𝐄T1 = (𝚲T 𝐠1 )T yields:
𝐰 = 𝐓𝜽𝐭 𝐭̇ + 𝜑𝐠
̇ 1, 𝐚 = 𝐓̇ 𝜽𝐭 𝐭̇ + 𝐓𝜽𝐭 𝐭̈ + 𝜑̇ 𝐠̇ 1 + 𝜑𝐠
̈ 1. (76)
K̇ 1 = W1� + 𝐠̇ T1 (𝐠1 × 𝐠�1 ),
𝛿K1 = 𝛿Θ�1 + 𝛿𝐠T1 (𝐠1 × 𝐠�1 ), (72) In the Kirchhoff case, a third variant can be advanta-
𝛿W1 = 𝛿 Θ̇ 1 + 𝛿𝐠T1 (𝐠1 × 𝐠̇ 1 ). geous: similar to the curvature vector 𝐤 [see (69)] and the
spin vector 𝛿𝜽 [see (71) or (25)], also the angular velocity
𝐰 can be split into two components
4.2 Deformation Measures and Stress Resultants
𝐒(𝐫 � )𝐫̇ �
𝐰 = 𝐰∥ + 𝐰⟂ = W1 𝐠1 + 𝐒(𝐠1 )𝐠̇ 1 = W1 𝐠1 + ,
Having defined kinematics that are compatible with the ||𝐫 � ||2 (77)
𝐚 = Ẇ 1 𝐠1 + W1 𝐠̇ 1 + 𝐒(𝐠1 )̈𝐠1 .
Kirchhoff constraint according to (67), the deformation meas-
ures, constitutive relations and stress resultants presented in
While 𝐰⟂ is determined by 𝐫(s), W1 is not expressed via 𝐫(s)
Sect. 3 can be adapted to the shear-free case. Inserting the
and 𝜑(s) as it has been done in (76).
constrained curvature vectors from (70) into the deformation
measures 𝛀 and 𝝎 according to (34) and (35) yields
Remark Three possible time integration schemes result
⎛ KM1 + − KM01 −
𝜑� ⎞ 𝜑�0 from the different variants given above:
⎜ ⎟
𝛀 =⎜ 𝐠T2 𝜿 − 𝐠T02 𝜿 0 ⎟ ,
⎜ ⎟ ̇ 𝐫̈ , 𝐰, 𝐚)n+1 = 𝐟 ((𝐫, 𝚲)n+1 , (𝐫, 𝚲, 𝐫,
1) (𝐫, ̇ 𝐰, 𝐫̈ , 𝐚)n ),
⎝ 𝐠T3 𝜿 − 𝐠T03 𝜿 0 ⎠𝐄i
(73) ̇ 𝐫̈ , 𝜑,
2) (𝐫, ̇ 𝜑)
̈ n+1 = 𝐟((𝐫, 𝜑)n+1 , (𝐫, 𝜑, 𝐫, ̇ 𝜑,
̇ 𝐫̈ , 𝜑)
̈ n ),
⎛ KM1 + 𝜑� − KM01 − 𝜑�0 ⎞ ̇ 𝐫̈ , W1 , Ẇ 1 )n+1 = 𝐟((𝐫, 𝚲)n+1 , (𝐫, 𝚲, 𝐫,
3) (𝐫, ̇ W1 , 𝐫̈ , Ẇ 1 )n ).
⎜ ⎟
𝝎 =⎜ 𝐠T2 𝜿 − 𝐠T02 𝜿 0 ⎟ . (78)
⎜ T T ⎟ Remark From the tangent transformations (69), (71)
⎝ 𝐠3 𝜿 − 𝐠03 𝜿 0 ⎠𝐠i
and (77), the following similarities become obvious:
The objective variation of (73) is still given by ( )( 𝛿𝐭 )
1
𝛿𝜽 = ||𝐭|| 𝐒(𝐠 ) 𝐠 ,
𝛿𝛀 = 𝚲T 𝛿𝜽� , 𝛿o 𝝎 = 𝛿𝜽� , but now the spin vector is con- 1 1
𝛿Θ1
strained according to (71). By construction, the shear ( )( 𝐭 � )
1
components of 𝚪 and 𝜸 in (34) and (35), vanish due to the 𝐤 = ||𝐭|| 𝐒(𝐠1 ) 𝐠1
K1
, (79)
Kirchhoff constraint (67): ( )( 𝐭̇ )
1
𝐰 = ||𝐭|| 𝐒(𝐠1 ) 𝐠1 .
W1
𝚪 = 𝜖𝐄1 , 𝜸 = 𝜖𝐠1 with 𝜖: = ||𝐫 � || − 1
𝛿𝚪 = 𝛿𝜖𝐄1 , 𝛿o 𝜸 = 𝛿𝜖𝐠1 with 𝛿𝜖 = 𝛿𝐫 �T 𝐠1 . (74)
Here, the abbreviation 𝜖 has been introduced for the 4.3 Strong and Weak Form
remaining component representing the axial tension. Based
on the force split 𝐟 = 𝐟∥ + 𝐟⟂ = 𝐅1 𝐠1 + 𝐟⟂, also the consti- In this section, the spatial representation of mechanical
tutive relations simply as consequence of the Kirchhoff equilibrium will be considered. In the following, the nota-
constraint: tion will be simplified by summarizing external forces and
moments as well as inertia forces and moments accord-
̃ int (𝛀, 𝜖) = 1 𝛀T 𝐂M 𝛀 + 1 EA𝜖 2 ,
Π ing to ̃ 𝐟 + 𝐟𝜌 and 𝐦
𝐟𝜌 : = ̃ ̃ + 𝐦𝜌. The shear forces
̃ 𝜌: = 𝐦
𝐟⟂ , which provide no work contribution in the Kirchhoff
2 2
̃ int
𝜕Π
𝐌= T
= 𝐂M 𝛀 = 𝚲 𝐦, (75) case, can be eliminated from the strong form of mechanical
𝜕𝛀
̃ int
𝜕Π equilibrium yielding the following set of four differential
F1 = 𝜕𝜖 = EA𝜖.
equations:
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
( )
𝐠T1 𝐦� + 𝐦 ̃ 𝜌 = 0, derivation of reduced beam models, i.e. beam models that
[ � ( )]� are valid under certain restrictions concerning initial geom-
(80)
̃𝜌 + ̃
(F1 𝐠1 )� + ||𝐫𝐫� ||2 × 𝐦� + 𝐦 𝐟𝜌 = 𝟎.
etry and external loads and will eventually yield consider-
ably simplified finite element formulations. On this basis,
The set (80) is sufficient in order to solve for the four pri- a special torsion-free beam element formulation could
mary variables 𝐫 and 𝜑 as soon as the stress resultants 𝐟∥ be derived that completely abstains from any rotational
and 𝐦 are expressed by kinematic and constitutive relations degrees of freedom, a distinct property that drastically sim-
from Sect. 4.2. Multiplying (80) with the admissible trans- plifies many standard procedures in the numerical treat-
lational and rotational variations 𝛿𝐫 and 𝛿Θ1 and integrating ment of geometrically exact beam element formulations.
by parts gives the equivalent weak form of equilibrium:
[ ] 4.4 Weak Enforcement of Kirchhoff Constraint
G=∫
l
𝛿𝜽�T 𝐦 + 𝛿𝐫 �T 𝐠1 F1 −𝛿𝐫 T �
𝐟𝜌 − 𝛿𝜽T 𝐦
� 𝜌 ds
0 ��� �����
𝛿 𝝎T 𝐦 In the last sections, the set of primary variable fields
[ o 𝛿𝜖F]1
𝐫 � ×𝛿𝐫 � (𝐫(s), 𝜑(s)) has been chosen in a way such that the Kirch-
− 𝛿𝐫 T 𝐟𝜎 + 𝛿𝜽T 𝐦𝜎 =̇ 0, 𝛿𝜽 = 𝛿Θ1 𝐠1 + .
Γ𝜎 ||𝐫 � ||2 hoff constraint (67) of vanishing shear strains is strongly
(81) fulfilled by construction. However, more flexibility in the
In (81), the constrained spatial spin vector (71) has been subsequent discretization process (see e.g. Sect. 6) could
identified and already substituted with the symbol 𝛿𝜽. As be gained by formulating a Reissner type beam problem,
indicated by the curly brackets in (81), the pre-factors of the which allows for two independent interpolations for the
stress resultants 𝐦 and F1 are represented by the objective centerline field 𝐫(s) as well as the triad field 𝚲(s), and by
variations 𝛿o 𝝎 and 𝛿𝜖, underlining the geometrical exact- weakly enforcing the Kirchhoff constraint of vanishing
ness of the proposed Kirchhoff beam formulation. Finally, shear strains by means of additional constraint equations:
and 𝜑u the orientation of the cross-section with respect to is required. The Lagrange multiplier fields 𝜆Γ2 (s) and
a rotation around the tangent vector. How these conditions 𝜆Γ3 (s) in (84) can be interpreted as the shear force compo-
can be modeled in practice is shown in “Appendix B”. By
introducing the trial space (𝐫, 𝚲(𝐫, 𝜑)) ∈ satisfying (82)
nents F2 (s) and F3 (s), i.e. reaction forces which enforce the
∫
the specific choice of the intermediate triad definition. 𝛿Π𝜆Γ23 = (𝛿𝜆Γ2 (s)Γ2 (s) + 𝛿𝜆Γ3 (s)Γ3 (s))ds
0
Remark In [109], a special choice of intermediate triad l
(85)
field in (68) has been proposed, the Frenet–Serret triad of
∫
+ (𝜆Γ2 (s)𝛿Γ2 (s) + 𝜆Γ3 (s)𝛿Γ3 (s))ds.
the beam centerline. It has been argued in this work, that
this choice is not favorable for numerical purposes as con- 0
sequence of singularities occurring on straight beam seg- The first term in (85) represents the weak statement of the
ments. However, the Frenet–Serret frame is very beneficial Kirchhoff constraint (83) while the second term can be
for the analytical treatment of Kirchhoff–Love beam prob- interpreted as the work contribution of the shear reaction
lems and has e.g. been applied in order to derive analytical forces. Similar to the displacement primary fields, a proper
reference solutions for the numerical examples of Sect. 11. trial space (𝜆Γ2 , 𝜆Γ3 ) ∈ 𝜆Γ23 and a proper weighting space
(𝛿𝜆Γ2 , 𝛿𝜆Γ3 ) ∈ 𝜆Γ23 have to be introduced which uniquely
Remark As shown in [110], the Kirchhoff–Love beam define the resulting mixed beam formulation. The discrete
theory presented so far provides an ideal basis for the realization of (85) will be presented in Sect. 10.
13
C. Meier et al.
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
In a similar manner, the modified as well as the physical nodal primary variables (e.g. in terms of rotation para-
angular accelerations are related according to: metrization), this time integration scheme can directly be
applied without the need for any adaptions.
(1 − 𝛼m )𝐀md,n+1 + 𝛼m 𝐀md,n = (1 − 𝛼f )𝐀n+1 + 𝛼f 𝐀n
(92) Review: As mentioned in Sect. 5, a large number of sci-
with 𝐀md,0 = 𝐀0 .
entific contributions considers the development of energy-
momentum schemes for Simo–Reissner type formulations.
For later use, it is favorable to express the vectors 𝐫̇ n+1 and
However, this does not hold true for Kirchhoff–Love beam
𝐫̈ n+1 in terms of the primary unknown 𝐫n+1:
element formulations. For example, the formulations of ani-
)( sotropic type [see requirements (5a–c) in Table 2] haven’t
𝛾
𝐫̇ n+1 = �
𝐮 + 1 − 𝛽𝛾 𝐫̇ n
𝛽Δt n+1
( ) considered dynamics so far. In [157], a energy-momentum
+ 𝛾
Δt 1 − 2𝛽 𝐫̈ md,n , scheme is proposed, but only for temporal discretization
𝐫̈ n+1 =
1−𝛼m
� 1−𝛼m
𝐫̇ n of the beam centerline. The torsional problem associated
𝐮n+1 − 𝛽Δt(1−𝛼 (93)
[
𝛽Δt2 (1−𝛼 f ) f )] with the rotational degree of freedom is only considered in
𝛼m (1−𝛼m )(0.5−𝛽)
+ 1−𝛼f
− 𝛽(1−𝛼 ) 𝐫̈ md,n a static manner, i.e. arbitrary dynamic problems contain-
f
𝛼f
− 1−𝛼 𝐫̈ n . ing also rigid body rotations with respect to the beam axis
f are not accessible by this approach. The temporal discre-
tization applied in the present work encompasses both, the
A similar relation can be formulated for the material angu- translational and the rotational fields. As verified numeri-
lar velocity 𝐖n+1 and acceleration 𝐀n+1: cally in Sect. 11, the numerical dissipation provided by the
( ) extended generalized-𝛼 scheme will enable energy-stable
𝛾 ̃
𝐖n+1 = 𝚯 + 1 − 𝛽𝛾 𝐖n
𝛽Δt n+v1 time integration for highly nonlinear problems. Concretely,
( )
+Δt 1 − 2𝛽 𝛾
𝐀md,n , in the well-known ellbow-cantilever example of Sect. 11.7,
1−𝛼m ̃ n+1 − 1−𝛼m 𝐖n a considerably improved energy stability can be observed
𝐀n+1 = 𝛽Δt2 (1−𝛼 ) 𝚯 (94) as compared to [83], where a Lie-group extension of the
[ f 𝛽Δt(1−𝛼f])
𝛼
+ 1−𝛼m − 𝛽(1−𝛼
(1−𝛼m )(0.5−𝛽)
𝐀md,n standard Newmark scheme [141] has been employed. Nev-
)
𝛼
f f
ertheless, energy-momentum schemes for time discretiza-
− 1−𝛼f 𝐀n .
f tion of the proposed Kirchhoff–Love beam elements pro-
vide an interesting direction of future research.
In [31] and [32], it has been proven that the integration
scheme given by Eqs. (88–94) yields the same favorable
properties as the standard generalized-𝛼 method, which 6 Spatial Discretization of Primary Fields
are second-order accuracy, unconditional stability (within
the linear regime), controllable damping of the high-fre- Spatial discretization is exclusively considered in the con-
quency modes and minimized damping of the low-fre- text of finite element methods. It represents the core topic
quency modes. Remarkably, the parameter choice leading in the development of geometrically exact Kirchhoff beam
to this optimal behavior is identical to that of the standard elements within this work. Throughout this work, quanti-
generalized-𝛼 scheme. Furthermore, it has been shown ties evaluated at a node i are marked by a superscript, e.g.
that this scheme can consistently treat non-constant mass ai , and nodal primary variables are distinguished by an
matrix contributions, such as the term 𝚲𝐂𝜌 𝐀 occurring additional hat, e.g. â i .
in geometrically exact Reissner and Kirchhoff type beam
13
C. Meier et al.
The highest derivative of the primary variable 𝐫(s) occurring In an analogous manner, the presented procedure can be
in the weak form (81) of the Kirchhoff beam is given by the extended to Hermite polynomials of higher order.
second derivative 𝐫 �� (s). Consequently, for the interpolation
of the centerline 𝐫(s) shape functions are required that fulfill
C1-continuity at the element boundaries. Besides this require- 6.2 Discretization of Rotation Field
ment, a C1-continuous centerline representation can be very
beneficial for problem class requiring a smooth geometry In Sect. 2.1, two parametrizations of rotation tensors have
representation such as beam-to-beam contact problems (see been investigated: a parametrization via rotation vectors
e.g. [111, 112]). In order to guarantee for C1-continuity, Her- 𝝍 and a parametrization via the SR mapping on the basis
mite shape functions are employed: of the set (𝐭, 𝜑). In the following two sections, these two
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
variants will be employed in order to parametrize the rota- node i coincides with one of the two element boundary
tion tensors 𝚲i at the element nodes i = 1, ..., nΛ. In the sub- nodes employed in the Hermite interpolation (97), this
sequent Sects. 6.2.3 and 6.2.4, also two possible approaches tangent vector indeed represents a nodal primary vari-
for the interpolation of these nodal triads in the elements able, i.e. 𝐭(𝜉 i ) = 𝐭̂ i for i = 1, 2, otherwise the tangent vector
interior, one based on rotation vectors and one based on the simply represents the interpolated centerline derivative at
SR mapping, will be presented. this position, i.e. 𝐭(𝜉 i ) = 𝐫 � (𝜉 i ). Based on (additive) incre-
ments Δ𝜑̂ i,k+1
n of the nodal relative angles, the configura-
6.2.1 Parametrization of Nodal Triads Via Rotation tion update from iteration k to iteration k + 1 of the New-
Vectors ton–Raphson scheme at time step tn reads:
Only for rotations with a magnitude smaller than 180◦ a Remark Within this work, intermediate triads 𝚲M based
unique rotation vector can be extracted from a given triad on the SR mapping are used for two different purposes:
(by applying e.g. Spurrier’s algorithm, see [146]). Within firstly, they are used for the definition of nodal material tri-
this work, rotation vectors are always extracted in a manner ads 𝚲i based on nodal relative angles 𝜑̂ i and associated
such that 𝝍 ∈] − 𝜋;𝜋]. Within this range, the transforma- nodal intermediate triads 𝚲iM [SR mapping “in time” from
tion matrix 𝐓 between additive and multiplicative rotation 𝜑̂
13
C. Meier et al.
6.2.3 Triad Interpolation Based on Local Rotation Vectors from (105) can be written in an abstract manner in the form
𝝍 h (𝜉) = nl(𝝍̂ 1 , ..., 𝝍̂ nΛ , 𝜉). While an explicit interpolation
In this section, a triad interpolation is presentented that rule for the rotation vectors 𝝍 h (𝜉) is not needed for prac-
has originally been proposed by Shoemake [137] and for tical purposes—the triad field is already given by (105)—
the first time employed to geomerically exact beam element the discrete version of the spin vector field 𝛿𝜽h (𝜉) and the
formulations by Crisfield and Jelenić [43, 83]. On each of field of (multiplicative) rotation vector increments Δ𝜽h (𝜉)
the considered nΛ nodes, a triad 𝚲i, with i = 1, 2, ..., nΛ, is will be required in the next sections for the spatially discre-
defined by primary degrees of freedom either according tized weak form of the balance equations and its lineariza-
to Sect. 6.2.1 or to Sect. 6.2.2. The interpolation strategy tion. In [43], these discretized fields have consistently been
presented in this section is independent from the specific derived from the triad interpolation (105), leading to:
choice of nodal primary variables. First, a reference triad
∑
nΛ
�𝐈i (𝜉)Δ𝜽̂ i = :�𝖨Δ𝜽̂ ,
𝚲r based on the triads at nodes I and J is defined: Δ𝜽h (𝜉) =
i=1
∑ �i
nΛ
i
(107)
I
𝚲r = 𝚲 exp(𝐒(𝚽 ∕2)), IJ
𝛿𝜽h (𝜉) = 𝐈 (𝜉)𝛿 𝜽̂ = :�𝖨𝛿𝜽̂ .
(104) i=1
exp(𝐒(𝚽IJ )) = 𝚲IT 𝚲J .
The arc-length derivatives of the interpolations (107) fol-
The nodes I and J are chosen as the two middle triads for
low in a straightforward manner and are given by:
elements with an even number nΛ of nodes and as the one
middle triad (i.e. I = J ) for elements with an odd number ∑nΛ
1 �i i
nΛ of nodes [see also Eq. (6.2) in [43], which is based on a Δ𝜽�h (𝜉) = 𝐈,𝜉 (𝜉)Δ𝜽̂ , (108)
J(𝜉)
sightly different node numbering]. Based on the definition i=1
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
Fig. 3 Employed element-local node numbering scheme for translational and rotational primary fields. a Node numbering applied to Simo–
Reissner elements. b Node numbering applied to Kirchhoff–Love elements
As emphasized in [83], the generalized shape functions ful- Based on the reference triad 𝚲r and the nodal triads 𝚲i the
fill interpolation and completeness properties interpolation procedure is defined according to:
Fig. 3b). The nodal triads are oriented tangential to the beam Review: In [109], it has been argued that the SR mapping
centerline curve. Thus, the first base vectors yield: exhibits a singularity for 𝐠̄ T1 𝐠1 = −1 in (15). For the proposed
𝐫 � (𝜉) 𝐫 � (𝜉 i ) finite element formulations this can only occur for rotation
𝐠1 (𝜉) = → 𝐠i1 = . (112) increments per time step that are larger than 180◦ or large
||𝐫 � (𝜉)|| ||𝐫 � (𝜉 i )||
element deformations exhibiting relative rotations between
Similarly to Sect. 6.2.3, one nodal triad 𝚲I has to be chosen the element center and boundary nodes that are larger than
as reference triad of the interpolation scheme: 180◦ (see [109]). Consequently, these singularities are not
practically relevant for reasonable spatial and temporal
𝚲r = 𝚲I with I ∈ {1, 2, ..., nΛ }. (113)
13
C. Meier et al.
discretizations. In [109], it has also been argued that these sin- vector 𝛿𝚯 ̂ : = (𝛿 Θ ̂ nΛ )T represent assemblies of
̂ 1 , ..., 𝛿 Θ
1 1 1
gularities occurring at relative rotations of 180◦ represent the the Lagrange shape functions Li and the nodal twist com-
optimum that can be achieved for mappings on the tangent ponents 𝛿 Θ̂ i . Alternatively, it can be discretized in a Bub-
1
vector. The approach employed by the anisotropic formula- nov–Galerkin manner based on (114), yielding:
tions from the literature is sightly different [see requirement
(2c) in Table 2]. There, only one spatially fixed reference )𝛿𝐫 � �
𝛿𝜽h (𝜉) = 𝛿Θ1h (𝜉)𝐠1 + 𝐒(𝐫
��𝐫 � ��2
,
triad at the beam endpoint is used for the initial SR mapping 𝛿Θ1h (𝜉) = 𝛿ΘM𝜑 1 (𝜉) + 𝛿𝜑h (𝜉),
in space, and only one temporarily fixed reference triad at (118)
∑
nΛ
every centerline position s for the SR mapping in time. Thus, 𝛿𝜑h (𝜉) = Li (𝜉)𝛿𝜑i .
for these formulations, singularities can occur for practically i=1
relevant configurations in case the relative rotation between The spin vector 𝛿𝜽h (𝜉) in (118) shall be completely
the cross-sections at the beam end point and at an arbitrary expressed via nodal variations 𝛿𝚯 ̂ and 𝛿 𝖽̂ , i.e. via assem-
1
centerline position s exceeds 180◦ or in case the total cross- blies of nodal primary variables. Thereto, 𝛿ΘM𝜑 1 (𝜉) and 𝛿𝜑i
section rotation from the initial to the current configuration are expressed as
exceeds 180◦ at any centerline position s.
𝐠T1 (𝐠I1 × 𝛿𝐠I1 ) − 𝐠I1T (𝐠1 × 𝛿𝐠1 )
̂I +
𝛿ΘM𝜑 1 (𝜉) = 𝛿 Θ ,
Review: In the literature, the smallest rotation mapping 1
1 + 𝐠T1 𝐠I1
defined in (15) of Sect. 2.2 is often denoted as “rotation
̂ i − 𝛿Θi
𝛿𝜑i = 𝛿 Θ (119)
without twist”. Thus, sometimes it is mistakenly assumed 1 M 𝜑1
that an intermediate triad field as employed in (114) would 𝐠i1T (𝐠I1 × 𝛿𝐠I1 ) − 𝐠I1T (𝐠i1 × 𝛿𝐠i1 )
exhibit a vanishing torsion (see e.g. [66], where the slightly 𝛿ΘiM ̂I +
= 𝛿Θ ,
𝜑1 1
1 + 𝐠i1T 𝐠I1
different interpolation scheme (131) has been employed).
However, according to (115), the torsion of the intermedi-
which directly follows from interpolation (114). Inserting
ate triad field (114) constructed via the SR mapping does
the relations (119) into the spin vector (118) yields:
not vanish for general curved 3D configurations. It can eas-
ily be shown that the torsion vanishes in the limit of fine 𝛿Θ1h (𝜉) = 𝖫∥ 𝛿𝚯 ̂ + 𝛿Θ
1 1,diff (𝜉)
discretizations lele : = s(e),2 − s(e),1 → 0. In this limit, the ∑ i 𝐠iT1 (𝐠I1 ×𝛿𝐠I1 )−𝐠IT1 (𝐠i1 ×𝛿𝐠i1 )
nΛ
intermediate triad field becomes identical to an element- 𝛿Θ1,diff (𝜉) = − L 1+𝐠iT 𝐠I (120)
i=1 1 1
wise Bishop frame [23] and the relation 𝐠T1 (𝐠I1 ×𝛿𝐠I1 )−𝐠IT (𝐠1 ×𝛿𝐠1 )
+ 1
1+𝐠T1 𝐠I1
.
� �
𝐯𝜃∥Θ = 𝐋T∥ ⊗ 𝐠T1 , 𝐯𝜃⊥ = −𝐇�T 𝐒(�𝐭),
𝛿𝜽h (𝜉) = 𝛿Θ1h (𝜉)𝐠1 + 𝐒(𝐫 )𝛿𝐫
, (n )
��𝐫 � ��2 ∑ Λ
∑
nΛ
̂ i = :𝖫∥ 𝛿𝚯 ̂ , (117) 𝐯𝜃∥d = Li (𝜉)𝐯1 (𝜉i ) − 𝐯1 (𝜉) ⊗ 𝐠T1 (𝜉), (122)
𝛿Θ1h (𝜉) = Li (𝜉)𝛿 Θ 1 1
i=1 i=1
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
�n �
Comparing Eqs. (117) and (122) leads to the conclusion ∑Λ
that the difference between the Bubnov–Galerkin and the 𝛿𝜽h (𝜉) = L i
(𝜉)𝐠Ti
1 0
𝐰 𝐠1 (𝜉)
i=1
Petrov–Galerkin variant is expressed by one additional +
𝐫 � (𝜉)×(𝐰0 ×𝐫 � (𝜉))
��𝐫 � (𝜉)��2
term involving 𝗏𝜃∥d. The arc-length derivative of (122) � � �� (127)
�
nΛ
(123) i.e. 𝛿𝜽h (𝜉) ≠ 𝐰0. Thus, the Petrov–Galerkin variant (117)
𝐯�𝜃 = 𝐋�T
∥
⊗ 𝐠T1 + 𝐋T∥ ⊗ 𝐠�T
1
,
∥Θ
It shall be investigated if the variants (117) and (118) can Thus, adding the term 𝛿Θ1,diff (𝜉)𝐠1 (𝜉) to the Petrov–Galer-
represent a constant distribution 𝛿𝜽Th (𝜉) = 𝐰0 = const. and kin variant (127) yields the desired result
h (𝜉) = 𝟎, as it is the case for their counterparts (107)
𝛿𝜽�T 𝛿𝜽h (𝜉) = 𝐰0 = const. for the Bubnov–Galerkin spin vector
and (110). Thereto, the nodal primary variable variations interpolation in case the nodal variations are given
are chosen as by (125). Alternatively, this result can be obtained by con-
sidering that (118) represents the consistent variation of the
𝛿 𝐝�j = 𝐰0 × 𝐝�j , objective triad interpolation (114) (see also Sect. 6.3.2).
𝛿�𝐭 j = 𝐰0 × �𝐭 j , (125) Since this interpolation is objective, the variation of the
̂ i = 𝐠T (𝜉 i )𝐰0 discrete internal energy has to vanish for infinitesimal rigid
𝛿Θ for j = 1, 2; i = 1, ..., nΛ .
1 1 body rotations. For an arbitrary stress resultant 𝐦, this is
only possible if 𝛿𝜽�h (𝜉) ≡ 𝟎 and consequently
Inserting 𝛿 𝐝̂j and 𝛿 𝐭̂ j according to (125) into (100) gives 𝛿𝜽h (𝜉) ≡ const. can be displayed exactly [see also the weak
the desired result 𝛿𝐫h (𝜉) = 𝐰0 × 𝐫h (𝜉). With this result and form (81)]. Recapitulatory, the interpolation (118) can rep-
by inserting (125) into the spin vector interpolations (117) resent arbitrary constant spin vector distributions 𝐰0, while
and (118), which both fulfill the interpolation property for for the interpolation (117) this is only possible for 2D prob-
the tangential spin vector components 𝛿 Θ ̂ i , it can be shown
1 lems or in the special case 𝐰0 = 𝟎. This result will be
that 𝐰0 is at least represented correctly at the element important in order to investigate the conservation proper-
nodes: ties of the resulting finite element formulations (see e.g.
Sects. 6.3.5 and 9.4). Finally, the field of (multiplicative)
𝛿𝜽h (𝜉 i ) = (𝐠T1 (𝜉 i )𝐰0 )𝐠1 (𝜉 i )
𝐫 � (𝜉 i )×(𝐰 ×𝐫 � (𝜉 i )) (126) rotation vector increments Δ𝜽h (𝜉) and the derivative
+ ||𝐫 � (𝜉0i )||2 = 𝐰0 , i = 1, ..., nΛ . Δ𝜽�h (𝜉), required for a consistent linearization of the discre-
tized weak form, follow from Eqs. (122) and (123) by sim-
Here, the Grassmann identity for cross-products ply replacing the variations 𝛿(.) by increments Δ(.). The
𝐚 × (𝐛 × 𝐜) = (𝐚T 𝐜)𝐛 − (𝐚T 𝐛)𝐜 with 𝐚, 𝐛, 𝐜 ∈ ℜ3 as well as spin vector considered above is expressed via multiplica-
𝐠1 = 𝐫 � ∕||𝐫 � || have been employed. Next, it will be inves- tive nodal increments 𝛿 Θ ̂ i as nodal primary variables. In
1
tigated for the variants (117) and (118), if the choice (125) case a rotation parametrization of the nodal triads via nodal
leads to a constant spin vector field along the beam ele- rotation vectors according to Sect. 6.2.1 is employed, the
ment. Inserting (125) into (117) and taking advantage of nodal vector of multiplicative iterative rotation increments
the Grassmann identity eventually yields: given by Δ𝚯 ̂ i = (ΔΘ̂ i , ΔΘ ̂ i )Ti can be directly used for
̂ i , ΔΘ
1 2 3 𝐠
13
C. Meier et al.
triad update as shown in (101). However, if a rotation para- At first glance, this interpolation seems to be more straight-
metrization of the nodal triads via the SR mapping and forward than (114) since no nodal triads are required for
nodal relative angles 𝜑̂ i according to Sect. 6.2.2 is constructing the intermediate triad field. However, as shown
employed, the rotation vector increments shall be expressed in [109], an interpolation of this kind is neither objective nor
by means of additive increments Δ𝜑̂ i of the nodal primary path-independent. This non-objective and path-dependent
variables 𝜑̂ i as shown in (103). A relation between ΔΘ ̂ i and
1
interpolation has e.g. been applied by the anisotropic formu-
i can easily be derived on the basis of Eqs. (19) and (20): lations from the literature [see requirement (3) of Table 2].
Δ𝜑̂
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
depend on history values. However, the corresponding nodal The identities 𝛀∗ = 𝛀 and 𝚪∗ = 𝚪 for the original and the
displacements can always arise in a way such that the finite rotated deformation measures (73) and (74) are a direct
element solution is independent from the actual load path in consequence of (134). Thus, also for Kirchhoff beam ele-
case the considered physical problem is path-independent. ments, the requirements (132) are sufficient in order to
In other words, the arising nodal displacements yield the ensure objectivity. In the following, the validity of (132)
path-independent solution to the discrete optimization prob- will be investigated for the interpolations in Sects. 6.1
lem (based on a proberly defined Lagrangian) which is asso- and 6.2.
ciated with the path-independent physical problem. In the Objectivity of centerline interpolations Due to the linear
numerical investigations performed at the end of this chapter, dependence of the centerline interpolations (95) or (97) on
this property will be verified. However, throughout this sec- the nodal vectors, the proof of the first part of (132) is inde-
tion, the fundamental property of objectivity, i.e. the invari- pendent from the intermediate triad field:
ance of the applied deformation measures under rigid body ( )
motions, will be investigated. Thereto, a rigid body transla- 𝐫 ∗ (𝜉) = 𝚲R 𝐫(𝜉) + 𝐫R
(135)
tion 𝐫R and a rigid body rotation 𝚲R are superimposed onto for 𝐝̂∗i = 𝚲R (𝐝̂i + 𝐫R ) and ̂𝐭 ∗i = 𝚲R 𝐝̂i . □
the beam centerline curve 𝐫(𝜉) and the triad field 𝚲(𝜉). A
rigid body motion is characterized by constant fields 𝐫R and Objectivity of triad interpolation via local rotation vectors
𝚲R along the beam, thus 𝐫R′ ≡ 𝟎 and 𝚲′R ≡ 𝟎. In the following, The fulfillment of (132), i.e. the objectivity of interpola-
the subscript (.)∗ denotes quantities that result(from the super-
) tion (105) has been shown in [43]. The interested reader is
imposed rigid body motion, thus 𝐫 ∗ (𝜉) = 𝚲R 𝐫(𝜉) + 𝐫R and referred to this reference.
𝚲∗ (𝜉) = 𝚲R 𝚲(𝜉). A formulation is denoted as being objective Objectivity of triad interpolation via “Smallest Rota-
if such a rigid body motion does not affect the material defor- tion” Based on the relations (133–135) and the strong ful-
mation measures 𝛀 and 𝚪. It is straightforward to show that fillment of the Kirchhoff constraint, the base vector 𝐠∗1 (𝜉)
the space-continuous versions of the deformation measures 𝛀 resulting from the rigid body rotation yields:
and 𝚪 are objective, i.e. 𝛀∗ = 𝛀 and 𝚪∗ = 𝚪 (see e.g. [43]).
The question of interest is if this objectivity is preserved by the 𝐫 ∗� (𝜉) 𝚲 𝐫 � (𝜉)
𝐠∗1 (𝜉) = = R� = 𝚲R 𝐠1 (𝜉). (136)
employed spatial discretization schemes. In [43], it is shown ||𝐫 (𝜉)|| ||𝐫 (𝜉)||
∗�
that the fulfillment of the following requirement guarantees In a next step, the nodal primary variables are chosen such
for the objectivity of a geometrically exact beam formulation that the nodal triads are also rigidly rotated:
based on the Simo–Reissner theory as introduced in Sect. 3:
( ) 𝚲i∗ = 𝚲R 𝚲i . (137)
𝐫h∗ (𝜉)=𝚲
̇ R 𝐫h (𝜉) + 𝐫R , 𝚲∗h (𝜉)=𝚲
̇ R 𝚲h (𝜉) Using (136), (137) and (15), the following relation between
(132)
→ 𝚪h = 𝚪h , 𝛀∗h = 𝛀h .
∗
the vectors 𝐠M𝜑 2 (𝜉) and 𝐠M𝜑 3 (𝜉) of the intermediate triad
The following investigations will exclusively be applied to 𝚲M𝜑 (𝜉) [see (114)] and their counterparts 𝐠∗M 2 (𝜉) and
𝜑
discretized quantities. In order to shorten notation, the sub- 𝐠∗M 3 (𝜉) resulting from the rigid body motion can be
script (.)h will be omitted throughout this section. First, it
𝜑
shall be shown that the validity of (132) is also sufficient for derived:
the invariance of the deformation measures (73) and (74) of 𝐠∗T 𝐠∗ (𝜉) ( ∗ )
the Kirchhoff beam theory. If (132) is valid, it follows: 𝐠∗M i (𝜉) = 𝐠∗ri − ri 1
1+𝐠∗T 𝐠∗ (𝜉)
𝐠 (𝜉) + 𝐠∗r1
1
𝜑 r1 1
𝐠Tri 𝐠1 (𝜉) ( )
𝐫 ∗� (𝜉) = 𝚲R 𝐫 � (𝜉), 𝐫 ∗�� (𝜉) = 𝚲R 𝐫 �� (𝜉), = 𝚲R 𝐠 −
ri 𝚲R 𝐠1 (𝜉) + 𝐠r1 (138)
1+𝐠Tr1 𝐠1 (𝜉)
𝐠∗i (𝜉) = 𝚲R 𝐠i (𝜉), 𝐠∗� (𝜉) = 𝚲R 𝐠�i (𝜉). (133)
i = 𝚲R 𝐠M𝜑 i (𝜉), i = 2, 3.
Based on (133), the axial tension, the Frenet–Serret curva-
ture vector as well as the total torsion read: From (136) and (138), it can be concluded that the interme-
( )0.5 diate triad field is rigidly rotated:
||𝐫 ∗� || = 𝐫 �T 𝚲TR 𝚲R 𝐫 � = ||𝐫 � ||
→ 𝜖 ∗ = ||𝐫 ∗� || − 1 = ||𝐫 � || − 1 = 𝜖, 𝚲∗M (𝜉) = 𝚲R 𝚲M𝜑 (𝜉). (139)
( )
𝜑
13
C. Meier et al.
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
elements have to be compared. The underlying proposition In (144), 𝜉 i denotes the parameter coordinate and nCP the
is that elements with rh < r (especially rh < 1, which means number of Collocation Points (CPs) where the “original”
that more constraints than degrees of freedom are present) strains are evaluated. Moreover, Li (𝜉) are Lagrange shape
have a tendency to lock, whereas values rh > r of the con- functions of polynomial order nCP − 1. Linearizing (144) at
straint ratio indicate that not enough constraint equations the undeformed, straight configuration C0, yields the B-bar
are available in order to reproduce the constraint in an structure typical for geometrically linear finite elements:
accurate manner. Following this hypothesis, the case rh = r
has to be regarded as the optimal constraint ratio. Through- L(G𝜖 )|C0 : = 𝛿�
𝖽T 𝖧�T 𝐭0 K 𝐭0T 𝖧� Δ�
𝖽,
out this contribution, the relevant locking phenomena will ��� ���
be analyzed based on a mechanical as well as a numerical 𝖡T 𝖡
(145)
perspective. The corresponding concepts are applicable in a T ∑
nCP
L(G𝜖̄ )|C0 : = 𝛿�
𝖽T 𝖡 K𝖡Δ�
𝖽, 𝖡(𝜉) = Li (𝜉)𝖡(𝜉 i ).
straightforward manner. In future work, the stability of the
i=1
most promising element formulations also has to be investi-
gated in a mathematically rigorous manner, either based on
direct analysis or on numerical stability tests. In the last equation, the additional abbreviation K: = EA
Membrane locking In [110], the effect of membrane lock- has been introduced to shorten the notation.
ing in the context of geometrically exact Kirchhoff beam ele-
ments based on a Hermite centerline interpolation according Remark Actually, the residual vector stated in (164) con-
to Sect. 6.1.2 has been investigated in detail. Here, only the sists of functional expressions that cannot be integrated
main results shall be recaptulated. In [110], it has been found exactly by Gauss integration. However, in numerical simu-
that the state of (exactly) vanishing axial strains can only be lations, the deviations in the results for a number of Gauss
represented for straight configurations but not for abritariy points nG ⩾ 4 turned out to be very small as compared to
curved configurations. Furthermore, the amount of parasitic the discretization error. Therefore, the notion “exact inte-
axial strain energy occuring in states of pure pending has gration” will be used whenever four or more Gauss points
been shown to increase quadratically with the beam element are applied.
slenderness ratio leading to a progressively (over-) stiff sys-
tem answer. The constraint ratios of the space-continuous Remark In the last section, it has been shown that the
and discrete 2D Kirchhoff–Love beam problem have been strain field 𝜖(𝜉) is objective and path-independent, meaning
determined to r = 2 > rh = 1, indicating that the considered that the strain field does not change as a consequence of a
element formulation is prone to membrane locking. Differ- rigid body motion and that its value at a certain configu-
ent solution strategies exist in order to get rid of these locking ration is independent from the deformation path leading to
effects. In [110], the approach of Assumed Natural Strains that configuration. Since the assumed strain field 𝜖(𝜉)
̄ repre-
(ANS, see e.g. [73]), Reduced Integration (RI, see e.g. [115]) sents a pure re-interpolation of the original strain field 𝜖(𝜉 i )
and an alternative procedure proposed in [110] and denoted evaluated at fixed collocation points 𝜉 i, the former will also
as Minimally Constrained Strains (MCS), have been com- fulfill objectivity and path-independence.
pared. The proposed MCS scheme can be characterized as an
assumed strain/B-bar approach (see e.g. [139]), and it can be In [110], the number and location of CPs has been cho-
derived in a variationally consistent manner. Concretely, the sen according to nCP = 1 and 𝜉 1 = −1, 𝜉 2 = 1, 𝜉 3 = 0. In
contribution G𝜖 of the axial tension to the weak form of the order to motivate this choice, the constraint ratio resulting
mechanical equilibirum equations is replaced by G𝜖̄: from the MCS method and different sets of CPs as well as
for the alternative methods RI and ANS will be presented
� int,𝜖 = 1 EA𝜖 2 → Π
Π � int,𝜖̄ = 1 EA𝜖̄2 , in the following. In case of a strain re-interpolation such as
2 ( 2 ) given by (144), it is sufficient that the constraint of vanish-
G𝜖 = 𝛿𝐫 �T 𝐠1 EA ||𝐫 � || − 1 → G𝜖̄ = 𝛿 𝜖EA
̄ 𝜖,̄ ing axial strains is fulfilled at the collocation points in order
��� ���������
𝛿𝜖
(144) to end up with a vanishing axial strain energy. Since the
𝜖
∑
nCP
∑
nCP Hermite interpolation (97) provides a C1-continuous center-
𝜖(𝜉)
̄ = Li (𝜉)𝜖(𝜉 i ), 𝛿 𝜖(𝜉)
̄ = Li (𝜉)𝛿𝜖(𝜉 i ). line, the first derivative 𝐫 ′ and consequently also the axial
i=1 k=1
13
C. Meier et al.
tension 𝜖 is C0-continuous at the element boundaries. Thus, the axial tension term of the weak form, yielding the same
exactly one constraint equation ||𝐫 � (𝜉 i )||=1
̇ results from number of constraint equations neq,c = 2nele + 1 as the MCS
each (interior and element boundary) CP. approach. However, within this contribution, the MCS
method will be preferred due to its arguably more consist-
Remark The chosen CPs are motivated by strain-conti- ent variational basis and the uniform integration scheme
nuity: if the axial strains were not C0-continuous at the ele- resulting for the individual work contributions of the weak
ment boundaries, each element boundary node would pro- form.
vide two constraint equations, one for element (e) and one
for the subsequent element (e + 1), i.e. ||𝐫 � (𝜉 (e) = 1)||=1
̇ All in all, it can be concluded that the choice of the ele-
and ||𝐫 � (𝜉 (e+1) = −1)||=1
̇ . ment boundary nodes and the element midpoint as CPs of
the MCS method leads to the minimal possible number of
In Table 3, the constraint ratios resulting from the constraint equations and to an optimal constraint ratio. Con-
methods ANS, RI and MCS are compared for different sequently, a successful avoidance of locking effects could
choices concerning number and locations of collocation or be confirmed for this approach. A similar effectiveness can
Gauss points. As shown in [110], the ANS approach has be expected for a reduced Gauss–Lobatto integration of the
to be based on CPs with vanishing axial strain values for axial tension terms with integration points 𝜉 1 = −1, 𝜉 2 = 1
a pure bending state, which requires four CPs. The index and 𝜉 3 = 0. On the contrary, the minimal number of three
i in Table 3 indicates that all collocation or Gauss points integration points possible for a reduced Gauss–Legendre
lie in the elements interior while variants that are marked integration scheme leads to an increased constraint ratio of
by an index b also employ the element boundary nodes rh = 3 > r and consequently to a suboptimal locking behav-
𝜉 = ±1. According to the statements made above, the vari- ior. While the MCS and RI methodologies aim at a reduc-
ants with rh ∕r = 1 represent the constraints associated with tion of the number of constraint equations, the functional
the axial tension in an optimal manner, whereas variants principle of the ANS approach is different: there, parameter
with rh ∕r < 1 have a tendency to lock. Consequently, at first coordinates have to be determined where the constraint is
glance, the variants MCS-3b, MCS-2i and RI-2i seem to be already correctly fulfilled by the original element formula-
equally suitable. However, as derived in [110], the variants tion (when applying the latter to a representative test case).
MCS-2i and RI-2i will lead to an underconstrained system These parameter coordinates are typically chosen as CPs
of equations, allowing for zero-energy modes and yielding for the ANS approach. Such a procedure can avoid lock-
a singular system of equations and a rank deficient stiffness ing effects in a manner that is independent of the number
matrix in the straight configuration, while the variant MCS- of constraint equations. The drawback of the ANS method
3b exactly provides the minimal number of constraint equa- is that the positions of these points may change for general
tions required in order to avoid such zero-energy modes in deformation states in the geometrically nonlinear regime,
the straight configuration. For that reason, this method has which might considerably deteriorate the effectiveness of
been denoted as method of Minimally Constrained Strains. this approach. All these theoretical considerations recom-
Similarly, an optimal constraint ratio of rh = r = 4 and the mend the proposed MCS approach as method of choice
avoidance of zero-energy modes can be verified for this in order to avoid locking effects for the Kirchhoff beam
choice CPs in the 3D case. elements considered in this contribution. The numerical
results presented in [110] have confirmed this prediction.
Remark Of course, locking would be avoided if the beam
elements could exactly represent the internal energy asso- Review: In all the three categories isotropic, straight and
ciated with pure bending. According to [110], the MCS anisotropic large deformation geometrically exact Kirch-
method fulfills this requirement only for straight configu- hoff–Love finite elements available in the literature [see
rations while arbitrary curved configurations will yield a requirement (4b) in Table 2], the consequences of mem-
slightly over-constrained system of equations. Thus, the brane locking have been observed. However, a rigorous
state of constant curvature and vanishing axial tension can- treatment seems to be missing in these works. For exam-
not be displayed exactly. It has been shown numerically that ple, in [157], oscillations of the membrane forces have been
membrane locking still is avoided. This result is reasonable, observed and cured by means of a special force averaging
since the variant MCS-3i still fulfills the optimal constraint procedure. However, this procedure seems to be rather a
ratio r∕rh = 1 for arbitrarily curved configurations. post-processing step than an invasion in the actual finite
element formulation that would also improve the final dis-
Remark As an alternative to the variant MCS-3i, also a placement solution. In [160], the axial tension term of the
reduced Gauss–Lobatto integration scheme with integra- weak form is replaced by an averaged element-wise con-
tion points 𝜉 1 = −1, 𝜉 2 = 1 and 𝜉 3 = 0 could be applied to stant approximation of 𝜖(𝜉) in order to be able of exactly
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
representing the solution of a straight beam under axial exact beams could for example require the exact represent-
load. However, the influence of this procedure on mem- ability of a state of constant 3D curvature (bending and tor-
brane locking effects in curved configurations, i.e. configu- sion) and constant axial tension (as well as constant shear-
rations that are actually relevant for membrane locking, has deformation in case of Simo–Reissner formulations). The
not been investigated. Finally, [67] proposes a mixed finite beam centerline curve associated with such a configuration
element formulation in combination with a multi patch is given by a helix with constant slope as investigated in
B-spline approach in order to treat the locking effects. Inter- the numerical example of Sect. 11.3. In the following, it
estingly, the third-order variant of these B-spline patches will be distinguished, if only the strain energy associated
could be identified as Hermite interpolation comparable with such a state can exactly be represented or if addition-
to (97), which turned out to be favorable for the avoidance ally also the beam geometry in terms of centerline and triad
of locking as compared to single B-spline patches due to field of such a state can exactly be represented. In Sects. 9
the lower inter-element continuity enforced by the former and 11.3, it will be verified theoretically and numerically
approach. However, the geometrical compatibility equa- that the SK element formulations based on strong enforce-
tions of the multi patch approach as well as the considered ment of the Kirchhoff constraint and proposed in Sect. 9
numerical examples seem only to cover the geometrically can in general neither represent the strain energy nor the
linear regime. In contrast, the numerical examples in [110] beam geometry associated with such a state. Due to the
and Sect. 11 of this work confirm that this gap could be identical measure 𝜅 of centerline curvature and the compa-
closed by the proposed MCS method, which successfully rable centerline discretizations employed by the isotropic,
cures membrane locking in 3D large deformation problems. straight and anisotropic Kirchhoff–Love formulations from
Shear locking The phenomenon of shear locking does the literature, it can directly be concluded that also none of
not lie in the focus of the current work. Nevertheless, the these formulations can fulfill this patch test, neither in the
cause for such locking effects shall at least be briefly com- energetic nor in the geometrical sense (see again Table 2).
pared with the situation already discussed for membrane On the other hand, it will be verified theoretically and
locking. Shear locking denotes the inability of a finite numerically that the WK element formulations based on
element to exactly represent the state of vanishing shear weak enforcement of the Kirchhoff constraint (see Sect. 10)
strains. This situation can again be illustrated by means of a as well as the Simo–Reissner formulation proposed by
pure bending example. It is assumed that the beam center- Crisfield and Jelenić [43, 83] (denoted as CJ element in
line is discretized either by Lagrange polynomials accord- Sect. 7) can represent the exact energy state of the patch
ing to (95) or by Hermite polynomials according to (97) test due to the employed geodesic triad interpolation (and
and that the triad interpolation is given by (105). It has an appropriate collocation scheme such as MCS or reduced
already been stated that the triad interpolation (105) can integration for coupling the centerline and the triad field),
exactly represent constant curvatures. On the other hand, but not the exact geometry. There exist a few very recent
it has been shown that the Hermite centerline representa- 3D Simo–Reissner beam element formulations that can
tion (and also its counterpart based on Lagrange polynomi- exactly represent such a patch test in the energetic and also
als) cannot exactly display the state of constant curvature. in the geometrical sense. An explicite/closed-form analytic
This means, the state 𝐫 �T (𝜉)𝐠2 (𝜉) ≡ 𝐫 �T (𝜉)𝐠3 (𝜉) ≡ 0 cannot representation of the discretization underlying these formu-
be displayed exactly in combination with a constant curva- lations in form of a helicoidal interpolation applied to the
ture. In other words, the interpolation spaces applied to the triad and to the centerline field is typically only possible for
translation (i.e. to the beam centerline) and to the rotation first-order interpolations [144, 145, 151]. As emphasized
field (i.e. to the triad field) do not optimally match in the in [102], such a helicoidal scheme can be identified as one
sense that a state of constant curvature and vanishing shear geodesic interpolation on the semidirect product manifold
deformation cannot be represented exactly. Similar to mem- SE(3): = ℜ3 ⋊ SO(3), denoted as Special Euclidean group,
brane locking, the ratio of the shear stiffness to the bend- while e.g. the formulation of Crisfield and Jelenić [43, 83]
ing stiffness increases quadratically with the beam element consists of two individual geodesic interpolations, a linear
slenderness ratio. Thus, again the element slenderness ratio one with constant slope on the Euclidean vector space ℜ3
represents the key parameter for this locking effect. Further as well as a spherical interpolation with constant curvature
2D investigations on membrane and shear locking effects in on the Special Orthogonal group SO(3), together compos-
geometrically linear and nonlinear shear-deformable beam ing a interpolation scheme on the direct product manifold
element formulations can e.g. be found in [76]. ℜ3 × SO(3). An extension of these helicoidal interpola-
tions to higher-order schemes is e.g. given by [162, 163].
Review: At the end of this section, also a few considera- There, a closed-form representation of the interpolation has
tions shall be made concerning element patch tests [require- not been possible anymore and the triad and centerline field
ment (4d) in Table 2]. Such a patch test for geometrically have been generated out of the strain field in an implicit
13
C. Meier et al.
manner via numerical integration. In contrast to Simo– employed hyperelastic stored energy functions represent
Reissner formulations, Kirchhoff–Love element formula- quadratic forms in the derivatives of these primary variable
tions should typically be of higher-order in order to guar- fields, the convergence rate in the energy error yields:
||e||e = (h2(k−m+1) ).
antee for C1-continuity and reasonable convergence orders
(see also Sect. 6.3.4). It is not clear if the increased numeri- (148)
cal effort required for implicitly defining higher-order heli- Here, h: = lele is the element length and k the polynomial
coidal fields via numerical integration could overcompen- degree completely represented by the trial functions. In
sate possible advantages in terms of higher approximation [148], the convergence rate of the L2-error for the Ritz
quality when developing comparable Kirchhoff–Love ele- solution to a variational problem of order m is shown to be
ment formulations that exactly fulfill the patch test in the given by:
energetic and geometrical sense. It seems to be a promising
direction of future research to compare formulations of this ||e||2 = (hk+1 + h2(k−m+1) ). (149)
kind with the Kirchhoff–Love elements proposed here.
The second term in (149) represents the dependence of the
L2-error convergence rate on the energy error convergence,
reflecting the variational basis of the finite element method.
6.3.4 Optimal Convergence Orders
The first term represents the pure polynomial approxima-
tion of the trial functions with respect to the considered
In order to compare the convergence behavior of different
primary variable field, e.g. 𝐫(𝜉). In most cases, e.g. when
finite element formulations, a well-defined error measure is
displacement-based solid elements are considered, the first
required. Thereto, the following relative L2-error ||e||2rel will
exponent is smaller than the second one and dominates the
be considered in the numerical examples of Sect. 11:
overall discretization error. For this reason, only the first
√ term is considered by many authors. However, in the fol-
l
umax l ∫0
||e||rel =
2 1 1
||𝐫h − 𝐫ref ||2 ds. (146) lowing, it will be shown that especially for Kirchhoff type
beam elements also the second term of (149) is important.
Thereto, the expected convergence rates for Reissner and
In (146), 𝐫h denotes the numerical solution of the beam Kirchhoff type beam elements shall briefly be discussed.
centerline for a certain discretization. For all examples Element formulations of Simo–Reissner type The highest
without analytic solution, the standard choice for the refer- derivative of primary variable fields occurring in the weak
ence solution 𝐫ref is a numerical solution via the WK-TAN form associated with the Simo–Reissner beam problem is
element (see Sect. 10.2) employing a spatial discretization m = 1. Thus, for the third-order (k = 3) Reissner beam ele-
that is by a factor of four finer than the finest discretization ments considered in this work, a convergence rate of six
shown in the convergence plot. The normalization with the is expected in the energy error (147), while the L2-error
element length l makes the error independent of the length is dominated by the first term of (149) leading to a corre-
of the considered beam. The second normalization leads to sponding optimal convergence rate of four.
a more convenient relative error measure, which relates the Element formulations of Kirchhoff–Love type The sub-
L2-error to the maximal displacement umax of the load case. sequently proposed Kirchhoff beam elements will lead
For some examples, also the relative energy error to the values k = 3 for the polynomial degree of the triad
functions and m = 2 for the highest arc-length derivative
Πint,h − Πint,ref
||e||e,rel = , (147) in the weak form. Consequently, the convergence of the
Πint,ref energy error is of order four. Furthermore, the exponents
of both terms in (149) take on a value of four, also lead-
will be considered. Here, Πint,h and Πint,ref represent the ing to an expected convergence rate of four for the L2-error.
stored energy functions (see e.g. 75) associated with a cer- Thus, also the second, energy-related term has to be con-
tain discretization and with the reference solution. Before sidered for Kirchhoff problems of this kind. For that rea-
the convergence plots of selected numerical examples son, at least polynomials of order three should be chosen
will be considered, the optimal convergence rates in the as trial functions for Kirchhoff beam elements: reducing
norms (146) and (147) expected for the different beam ele- the polynomial degree from k = 3 to k = 2 would lead to
ment formulations shall be briefly discussed. The conver- an undesirable decline in the L2-convergence rate from four
gence in the energy error (which is minimized by the finite to two. Thus, the third-order Kirchhoff beam elements pro-
element method) is dominated by the highest derivative m posed in this work can be regarded as approximations of
of the primary variable fields 𝐫(𝜉) and 𝚲(𝜉) occurring in the lowest order that is reasonable from a numerical point
the energy (see e.g. 75) and in the weak form. Since the of view. However, the fact that both exponents of (149)
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
take on the same value for k = 3 also means that in some 6.3.5 Conservation Properties
cases, usually for examples involving complex deformation
states and strain distributions, the second, energy-related Since the finite element solution converges towards the cor-
term might determine the overall error-level. Thus, in order responding analytic solution in the limit of fine spatial discre-
to fully exploit the approximation power of the employed tizations, elementary properties of the analytic solution such
discretization, it can often be sensible to apply trial func- as conservation of linear momentum (or rather equilibrium
tions of increased polynomial degree k > 3. In this case, of forces in statics), conservation of angular momentum (or
the first term of (149) will determine the overall error level rather equilibrium of moments in statics) as well as conser-
for sufficiently fine discretizations (since the second term vation of energy (or rather balance of external and internal
converges faster) and consequently the discretization error work for non-conservative problems) will also be fulfilled by
is exclusively limited by the approximation power of the the numerical solution for lele → 0. However, often it is desir-
applied polynomial order. The extension of the proposed able to provide such properties already for arbitrarily rough
Kirchhoff beam elements to Hermite interpolations of order spatial discretizations. The question, if these properties of the
k > 3 is possible in a straightforward manner and will be space-continuous problem are inherited by the spatially dis-
treated in future research work. In the numerical example cretized problem, will later be investigated for the different
of Sect. 11.2, a first proof of concept will be given for such beam element formulations proposed in subsequent sections.
an extension. Thereto, use will be made of the fact that the discretized weak
form of the balance equations is fulfilled for arbitrary values
Remark Based on these considerations the question of the nodal primary variable variations. Choosing the nodal
arises, which trial function orders kSR and kKL of formula- primary variable variations such that the associated virtual
tions based on the Simo–Reissner and the Kirchhoff–Love motion represents a rigid body translation
(𝛿𝐫h (𝜉), 𝛿𝜽h (𝜉)) ≡ (𝐮0 , 𝟎) ∈ h
theory have to be chosen in numerical examples in order
with 𝐮�0 = 𝟎, (150)
to perform a reasonable comparison of their convergence
behavior. The answer to this question depends on the pri- allows to investigate the conservation of linear momen-
mary interest, which might either lie in the L2- or in the tum. The special choice (150) leads to 𝛿𝐫h� (𝜉) ≡ 𝟎 as well
energy convergence. Here, the third-order Kirchhoff ele- as 𝛿𝜽�h (𝜉) ≡ 𝟎 and to vanishing contributions of the inter-
ments are compared to third-order Reissner elements, lead- nal forces and moments in the discretized weak forms (33)
ing to equal L2- but to different energy error rates. Alter- and (81) associated with the Simo–Reissner and the Kirch-
natively, one could compare the third-order Kirchhoff hoff–Love beam theory. Inserting (150) into these weak
elements with second-order Reissner elements, leading to forms yields
equal energy error rates but to different L2-error rates.
l l
[ ]
∫ ∫
Review: Later in Sect. 9, a quadratic interpolation will 𝐥̇ = 𝐟ext with 𝐥: = �𝐥ds, 𝐟ext : = �
𝐟ds + 𝐟𝜎 , (151)
Γ𝜎
be applied to the relative angle field 𝜑h (𝜉) of the interpo- 0 0
lation scheme (114). Since the orientation of the mate-
rial triad field is determined by the relative angle 𝜑h (𝜉) as and to exact conservation of linear momentum 𝐥 = const.
well as the tangent vector field 𝐫h� (𝜉), with the latter being for the unloaded system, viz. if 𝐟ext = 𝟎. Since reaction
a polynomial of order two, this second-order interpola- forces at Dirichlet supports are also included in 𝐟ext, (151) is
tion is sufficient for triad field discretization. Even a higher equivalent to the equilibrium of forces in the static case, i.e.
polynomial degree for 𝜑h (𝜉) could not further improve the if 𝐥 = 𝟎. Similarly, a choice of the nodal primary variable
exact polynomial representation of rotational strains, which variations representing a rigid body rotation given by
is of first order. On the other hand, if 𝜑h (𝜉) was only inter- (𝛿𝐫h (𝜉), 𝛿𝜽h (𝜉)) ≡ (𝐰0 × 𝐫h (𝜉), 𝐰0 ) ∈ h , 𝐰�0 = 𝟎, (152)
polated linearly, only the constant term of rotational strains
could be exactly represented and the second term in (149) allows to investigate the angular momentum. Relation (152)
would dominate the L2-convergence leading to a decline leads to 𝛿𝐫h� (𝜉) ≡ 𝐰0 × 𝐫h� (𝜉) as well as 𝛿𝜽�h (𝜉) ≡ 𝟎 and
in the expected optimal order to two. In Meier et al. [109], again to vanishing contributions of the internal forces and
these considerations have been confirmed numerically. The moments in the discrete versions of the weak forms (33)
Kirchhoff–Love formulations of straight type [see require- and (81) associated with the Simo–Reissner and the Kirch-
ment (4a) in Table 2] also apply only a first-order interpo- hoff–Love beam theory. Inserting (152) into these weak
lation for the relative angle 𝜑h (𝜉) and a comparable inter- forms yields:
mediate triad definition. Thus, also for these formulations a
decline in the optimal convergence order from four to two
is expected.
13
C. Meier et al.
∫
.
𝐡 = 𝐦ext , 𝐡: = (̃𝐡 + 𝐫 × ̃𝐥)ds, and invariant under rigid body motions. The second pre-
requisite ensures that the corresponding contribution to the
0
l
(153) weak form represents exact increments of this potential,
( ) which fulfills the third requirement (155) per definition, but
∫
𝐦ext : = 𝐫 ×̃ ̃ ds + [𝐫 × 𝐟𝜎 + 𝐦𝜎 ]Γ𝜎 ,
𝐟 +𝐦 also the first and second requirement (151) and (153), since
0 infinitesimal rigid body translations and rotations according
to (150) and (152) will lead to vanishing increments of this
and to exact conservation of the angular momentum 𝐡 = objective potential. However, the opposite conclusion does
const. for the unloaded system, viz. if 𝐦ext = 𝟎. Since obviously not hold true: also Petrov–Galerkin formulations
possible reaction moments at Dirichlet supports are also can at least fulfill the first and second requirement (151)
included in 𝐦ext, relation (153) is equivalent to the equilib- and (153) if the chosen test functions can represent rigid
rium of moments in the static case, i.e. if 𝐡 = 𝟎. Finally, a body translations and rotations (150) and (152) (see e.g. CJ
choice of the nodal primary variable variations according element of Sect. 7 or WK element of Sect. 10).
to
(𝛿𝐫h (𝜉), 𝛿𝜽h (𝜉)) ≡ (𝐫̇ h (𝜉), 𝐰h (𝜉)) ∈ h ,
Review: Following the argumentation from above, it can
(154) be expected that the objective and path-independent Bub-
allows to investigate the mechanical power balance. nov–Galerkin FEM discretizations employed in the formu-
Inserting (154) into the discrete versions of the weak lations of the straight type will in general fulfill the above
forms (33) and (81) associated with the Simo–Reissner conservation properties [see requirement (4c) of Table 2] as
and the Kirchhoff–Love theory, and making use of the rela- long as the variation and discretization process required to
�̇ int = 𝐰�T 𝐦 + (𝐫̇ � − 𝐰 × 𝐫 � )T 𝐟 and Π
tions Π �̇ kin = 𝐰T 𝐜𝜌 𝐚 (see derive the test functions is conducted in a consistent man-
also Sect. 3.4) as well as 𝐰 𝐒(𝐰)𝐜𝜌 𝐰 = 0, yields the fol-
T ner and no additional approximations (e.g. small tension
lowing relation assumptions according to ||𝐫 || ≈ 1 as discussed in [161])
are applied that could spoil this consistency. The situa-
Π̇ kin + Π̇ int = Pext , tion is slightly different for the anisotropic formulations:
l [ ] [ ] the non-objective and path-dependent character of the
(155) employed discrete deformation measures might in general
∫
Pext : = 𝐫̇ T � � ds + 𝐫̇ T 𝐟𝜎 + 𝐰T 𝐦𝜎 ,
𝐟 + 𝐰T 𝐦
Γ𝜎
lead to non-vanishing work/energy increments of the inter-
0
nal forces and moments in case of infinitesimal rigid body
translations and rotations according to (150) and (152). The
and consequently to exact energy conservation exact representation of these non-vanishing energy incre-
Πkin + Πint = const. for the unloaded system, viz. if ments resulting from consistently derived Bubnov–Galer-
Pext = 0. So far, it has been shown that exact conservation kin test functions may in this case exactly be the reason that
of linear momentum, angular momentum and energy [see the conservation properties (151) and (153) will be spoiled.
Eqs. (151), (153) and (155)] can be guaranteed for the spa- Eventually, also the existing Kirchhoff–Love formulations
tially discretized (and time-continuous) problem, provided of isotropic type in Table 2 shall be discussed. There, a
the special choices (150), (152) and (154) for the transla- standard polynomial interpolation based on Lagrange shape
tional and rotational variation fields are contained in the functions is applied to the tangential spin vector com-
discrete weighting space h of the considered finite ele- ponent 𝛿Θ1 similar to the Petrov–Galerkin variant (117).
ment formulation. In the following sections, the question Consequently, as shown in Eq. (127), also these formula-
if (150), (152) and (154) can indeed be represented by the tions cannot represent constant spin vector distributions
discrete weighting functions, will be investigated for the and exact conservation of angular momentum (153). It can
proposed beam element formulations. Of course, also the be shown that these straight formulations indeed represent
time integration scheme influences the conservation prop- Petrov–Galerkin formulations, which in turn can also not
erties of the fully discrete system considerably. However, guarantee for the conservation of energy (155). To verify
the investigation of this factor does not lie within the scope this statement, the following considerations are made: if the
of this work. In general, it can be stated that the three con- discrete spin vector field in [28, 157] is consistent with the
servation properties considered above will be fulfilled by a underlying triad interpolation in a Bubnov–Galerkin sense
finite element formulation if (i) the formulation is objective is not straight-forward to be answered, since no explicit
and path-independent and (ii) the test functions are consist- triad interpolation scheme is given (and required) by the
ently derived in aBubnov–Galerkin manner. Thefirst pre- considered isotropic formulations. However, at least the
requisite ensuresthat a unique potential exists in form of arc-length derivative of the discrete triad field is defined
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
and the relation 𝐠�1h (𝜉) = 𝐫h� (𝜉)∕||𝐫h� (𝜉)|| holds, a consist-
ent Bubnov–Galerkin interpolation of the tangential spin
The subscripts (.)𝐝̂ and (.)𝜽̂ distinguish the residual vector
vector component 𝛿Θ1h (𝜉) fulfilling (156) cannot be based i
contributions associated with the variations 𝛿 𝐝̂i and 𝛿 𝜽̂ .
on a pure polynomial interpolation as done in the men-
Within this work, a linearization Δ𝗋CJ = 𝗄CJ Δ̂𝗑CJ based
tioned references. Consequently, these interpolations are of i
on multiplicative rotation increments Δ𝜽̂ according to
Petrov–Galerkin type. 1T n T
Δ̂𝗑CJ : = (Δ𝐝̂1T , Δ𝜽̂ , ..., Δ𝐝̂nn T , Δ𝜽̂ n )T as given in [83] is
employed. In dynamic problems, the element residual vec-
7 Simo–Reissner Beam Element
tor 𝗋CJ and stiffness matrix 𝗄CJ slightly differ from the origi-
nal work [83] due to the applied time integration scheme.
Throughout this work, the Reissner type beam element
formulation proposed by Crisfield and Jelenić [43, 83], in
7.2 Avoidance of Locking Effects
the following referred to as CJ element, will serve as ref-
erence formulation for numerical comparisons. In the next
In [43] and [83], the authors proposed a reduced Gauss inte-
section, the main constituents required to derive the ele-
gration scheme in order to avoid shear locking and mem-
ment residual vector will be presented. In the subsequent
brane locking in the range of high beam slenderness ratios.
Sects. 7.2 and 7.3, this element formulation will be investi-
Thereto, nn − 1 integration points have been employed
gated with respect to possible locking effects and the fulfill-
for the integration of the internal force contribution of a nn
ment of mechanical conservation properties as introduced
-noded element. The effectiveness of this procedure will
in Sects. 6.3.3 and 6.3.5.
be verified in subsequent numerical examples and shall be
briefly motivated by the following considerations: the 3D
7.1 Element Residual Vector
Simo–Reissner beam problem is based on neq = 6 differential
equations (32) describing the beam problem and pointwise
In this section, the element residual vector of the CJ ele-
neq,c = 3 constraint equations in order to represent the state of
vanishing axial strains 𝐫 ′T 𝐠1 ≡ 1 and vanishing shear strains
ment will be derived from the general, space-continuous
𝐫 ′T 𝐠2 ≡ 𝐫 ′T 𝐠3 ≡ 1 prevalent in a pure bending problem. Con-
Simo–Reissner beam problem. First, all trial and weighting
functions are replaced by their discrete counterparts taken
from the finite-dimensional trial subspace (𝐫h , 𝚲h ) ∈ h ⊂
sequently, the constraint ratio of the space-continuous prob-
and the weighting subspace (𝛿𝐫h , 𝛿𝜽h ) ∈ h ⊂ . In the fol-
lem yields r = neq ∕neq,c = 2. As consequence of the reduced
integration, the discrete number of constraint equations takes
lowing, nn -noded finite elements with the vectors of nodal
on a value of neq,c = 3 ⋅ (nn − 1) ⋅ nele. Given the total num-
primary variables 𝗑̂ CJ : = (𝐝̂1T , 𝝍̂ 1T , ..., 𝐝̂nn T , 𝝍̂ nn T )T and
1T n T ber of equations neq = 6(nn − 1) ⋅ nele after application of
𝛿 𝗑̂ CJ : = (𝛿 𝐝̂1T , 𝛿 𝜽̂ , ..., 𝛿 𝐝̂nn T , 𝛿 𝜽̂ n )T are considered. The
proper Dirichlet conditions, the constraint ratio of the discrete
centerline interpolation is based on Lagrange polynomials
problem setting results in:
of order nn − 1 according to Sect. 6.1.1, i.e. nr = nn. Fur-
thermore, the rotation field interpolation follows Eq. (105) 3 ⋅ (nn − 1) ⋅ nele
of Sect. 6.2.3 based on nΛ = nn nodes, in combination with
rh = lim = 2 = r. (158)
nele →∞ 6(nn − 1) ⋅ nele
a Petrov–Galerkin approach for the spin vector discretiza-
tion given by (110). In contrast to the original works [43, Relation (158) yields the optimal constraint ratio for
83], here the modified generalized-𝛼 scheme of Sect. 5 is this element formulation. Consequently, no locking
employed for time integration, defining the velocities and effects are expected. The investigations made so far can
accelerations required for the inertia forces 𝐟𝜌 and moments be further refined by realizing that the CJ element can
𝐦𝜌 as stated in (42). Inserting these interpolation schemes exactly represent the internal energy associated with a
as presented in the previous sections into the weak form of 3D pure bending state. In order to understand this state-
the balance equations (33) yields the element residual vector ment, the internal energy is split into contributions stem-
contributions 𝗋CJ,𝐝̂ and 𝗋CJ,𝜽̂ according to ming from torsion and bending and into contributions
13
C. Meier et al.
stemming from axial tension and shear deformation, i.e. of angular momentum as well as conservation of energy.
Πint,h = Πint,Ω,h + Πint,Γ,h . For a pure bending state, the The representation of a rigid body translation (150) is triv-
energy contribution Πint,Γ,h has to vanish, and thus the total ial and given by the nodal primary variable variations
internal energy of a pure bending state is given by Πint,Ω,h,
which in turn is uniquely defined by the curvature vector 𝛿 𝐝̂i = 𝐮0 ,
i
𝛿 𝜽̂ = 𝟎 for i = 1, ..., nn . (159)
field 𝐊h (𝜉) = const. In order to represent the desired (con-
stant) distribution of the curvature vector field 𝐊h (𝜉), which Similarly, a rigid body rotation (152) can be displayed by
is possible for the employed triad interpolation (105), only the nodal primary variable variations
(nn − 1) ⋅ nele of the (nn − 1) ⋅ nele + 1 nodal rotation vec-
i
tors have to arise properly, while the one remaining nodal 𝛿 𝐝̂i = 𝐰0 × 𝐝̂i , 𝛿 𝜽̂ = 𝐰0 for i = 1, ..., nn . (160)
rotation vector describes rotational rigid body modes of
It follows from (159) and (160), that conservation of linear
the beam. Since the curvature vector field is defined via the
and angular momentum can be guaranteed. This statement
arc-length derivative of the rotation field, this one remain-
holds for both spin-vector discretizations (107) and (110)
ing nodal rotation vector can also be interpreted as inte-
and for the discretized centerline variation (96) since all of
gration constant resulting from an integration of the cur-
these variants fulfill proper completeness conditions and
vature field. Additionally, the nodal position vectors have
can exactly represent a constant vector field 𝛿𝜽 = 𝐰0. If the
to arise in a way such that Πint,Γ,h = 0. While the require-
ment 𝚪h (𝜉) ≡ 𝟎 cannot be fulfilled exactly for the employed
nodal velocities and angular velocities of the time-continu-
ous problem are chosen as primary variable variations
triad trial function spaces, the reduced Gauss integration
scheme applied for the CJ element yields a finite number i
of 3 ⋅ (nn − 1) ⋅ nele constraint equations in order to sat- 𝛿 𝐝̂i = 𝐝̇ i , 𝛿 𝜽̂ = 𝐰i for i = 1, ..., nn , (161)
isfy Πint,Γ,h = 0. Thus, similar to the rotation field, only only the Bubnov–Galerkin variant (107) leads (per defi-
(nn − 1) ⋅ nele of the (nn − 1) ⋅ nele + 1 nodal position vec- nition) to an exact representation of the rates of the spa-
tors have to arise properly in order to fulfill these constraint tially discretized hyperelastic and kinetic energy and con-
equations, while the one remaining nodal position vector sequently to exact energy conservation for the spatially
describes translational rigid body modes of the beam. All in discretized, time-continuous problem. On the contrary, the
all, the (nn − 1) ⋅ nele + 1 nodal position and rotation vectors Petrov–Galerkin variant (110), which has been employed
can always arise in a way such that a 3D pure bending case in (157), is not variationally consistent with the triad
can be represented, which consists of a constant curvature interpolation (105) occurring in the discrete internal and
vector field 𝐊h (𝜉) =const., a vanishing (reduced integrated) kinetic energies. Consequently, the weak form (157) does
energy contribution Πint,Γ,h of axial and shear strains as well not represent exact energy rates of the spatially discretized
as six superposed rigid body modes. Consequently, the tor- problem.
sion and bending modes represented by 𝛀h (𝜉) as well as the
axial tension and shear values at the Gauss points repre-
sented by 𝚪h (𝜉GP ) are non-competing and no locking effects
8 Motivation for “Shear‑Free” Beam Theories
are expected. These considerations can easily be extended
to arbitrary curvature fields 𝐊h (𝜉) that are representable by
Geometrically exact Simo–Reissner beam elements unify
the employed triad interpolation and arbitrary fields 𝚪h (𝜉)
high computational efficiency and accuracy. In fields of
for which the term 𝚪Th 𝚪h as occurring in the energy integral
application where thick beams are involved and the effect
can be integrated exactly by the reduced Gauss integration
of shear deformation is important, they are favorable as
scheme. In Sects. 11.2 and 11.3, the expected result that the
compared to the Kirchhoff type counterparts. However,
discrete hyperelastic energies associated with pure bending
with increasing beam slenderness ratio 𝜁 = l∕R, the shear
states in 2D and in 3D can be represented exactly by this
contribution to the overall beam deformation decreases.
beam element formulation will be verified by means of cor-
Furthermore, it is exactly the avoidance of the high stiffness
responding numerical test cases.
contributions of the shear modes which makes the Kirch-
hoff–Love theory of thin beams not only applicable, but
7.3 Conservation Properties
also favorable in the range of high beam slenderness ratios.
In this brief section, possible benefits of applying Kirchhoff
In the following, it will be investigated if the Simo–Reiss-
type beam elements in the range of high slenderness ratios
ner beam element formulation proposed by Crisfield and
will be illustrated and, at least approximately, quantified. In
Jelenić and repeated in the section above can represent the
Sect. 11, most of these effects will also be investigated and
variational fields (150), (152) and (154) required in order to
verified by means of numerical examples.
guarantee conservation of linear momentum, conservation
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
8.1 Improved Stability of Time Integration Scheme mechanical beam models: the full shear-deformable and
extensible Simo–Reissner beam formulation, a beam for-
The dynamic equations of motion of highly slender beams mulation subject to the Kirchhoff constraint of vanishing
typically result in very stiff Partial Differential Equations shear deformation, and a beam formulation subject to the
(PDEs). With increasing beam slenderness ratio, the ratio Kirchhoff constraint and an additional inextensibility con-
between high eigenfrequencies (associated with shear straint enforcing vanishing axial tension. For the roughest
modes), intermediate eigenfrequencies (associated with discretization considered in this numerical experiment and
axial tension and twisting modes) and low eigenfrequencies an investigated slenderness ratio of 𝜁 = 100, a rather mod-
(associated with bending modes) increases considerably. erate slenderness ratio as compared to many applications as
As a consequence, the stability requirement of explicit time mentioned in Sect. 1, the time step size could be increased
integration schemes leads to very small critical time step by a factor of ≈100 when abstaining from the shear mode
sizes as compared to the large oscillation periods of the and by a further factor of ≈5 when additionally abstain-
bending modes. On the contrary, implicit time integration ing from the axial tension mode. These results indicate the
schemes can provide unconditional stability in the linear potential of the Kirchhoff type beam formulations. Further-
regime of small deformations. However, in the large defor- more, they suggest that the first step towards an extensible
mation regime, also their performance is considerably dete- Kirchhoff beam formulation might already represent the
riorated by such high-frequency contributions. Despite the essential one with respect to numerical savings.
stability aspect, high-frequent modes are strongly affected
by the time discretization error and should be avoided as
long as no high-frequency analysis is required by a spe- 8.2 Improved Performance of (Iterative) Linear Solvers
cific application. In order to illustrate the relevant fre-
quency spectrum, in the following, the proportionalities of According to the previous argumentation, a high ratio
the eigenfrequencies resulting from pure bending 𝜔b, pure of the highest to the lowest dynamical eigenfrequencies,
torsion 𝜔t, pure axial tension 𝜔a and pure shearing 𝜔s are measured by the dynamic spectral radius, deteriorates the
given for the linearized beam problem: performance of time integration schemes. In a similar man-
√ √ √ √ ner, the performance of iterative linear solvers (see e.g.
EI 1 E GIT E [119]) decreases with increasing ratio of the highest to the
𝜔b ∼ ∼ , 𝜔 ∼ ∼ ,
𝜌Al4 𝜁 𝜌l2 t
𝜌IP l2 𝜌l2 lowest eigenvalue of the tangent stiffness matrix, a meas-
√ √ √ (162) ure for the condition number of this matrix. Furthermore,
E GA E even if direct linear solvers are applied, very high condition
𝜔a ∼ , 𝜔s ∼ ∼𝜁 .
𝜌l2 𝜌I 𝜌l2 numbers might considerably limit the achievable numeri-
cal accuracy. Especially in dynamics, where such round-off
According to the relations (162), the ratio of axial and errors tend to accumulate, these effects are undesirable. In
torsional eigenfrequencies to bending eigenfrequencies the following, the influence of the different deformation
increases linearly with increasing slenderness ratio. The modes on the condition number is investigated. For sim-
ratio of shear eigenfrequencies to bending eigenfrequencies plicity, the physical units of the considered beam problem
increases quadratically with increasing slenderness ratio. are chosen such that the element length lies in the range
Thus, from a theoretical point of view, the avoidance of of lele ≈ 1. Since the element length, or better the element
shear modes could already improve the numerical behav- Jacobian, typically enters the element formulation with
ior considerably. Since the numerical examples of Sect. 11 different exponents occurring in different stiffness matrix
mainly focus on static analysis and only present a brief out- entries, an element length of lele ≈ 1 seems to be a reason-
look on possible dynamic investigations, a numerical verifi- able choice with respect to conditioning. In this case, the
cation of these theoretical considerations does not lie within resulting contributions to the element stiffness matrix of a
the scope of this work. However, numerical investigations Simo–Reissner beam element that has been linearized with
of this question available in the literature confirm the pre- respect to the straight configuration, typically obeys the fol-
dicted trend: Lang and Arnold [95] investigated the geo- lowing proportionalities:
metrically nonlinear oscillations of a slender beam, which
kb ∼ EI ∼ ER4 , kt ∼ GIT ∼ ER4 ,
has been modeled by means of the geometrically exact (163)
ka ∼ EA ∼ ER2 , ks ∼ GA ∼ ER2 .
Simo–Reissner theory and discretized via finite differences
(see also [96]). In order to measure the influence of high- Again, kb, kt, ka and ks denote stiffness contributions from
frequency modes on time integration stability, the maximal bending, torsion, axial tension and shear modes. As long
possible time step sizes of the applied explicit time inte- as lele ≈ 1 holds and the discretization is kept fixed, the
gration scheme have been determined for three different cross-section radius R decreases linearly with increasing
13
C. Meier et al.
slenderness ratio 𝜁 . According to (163), it is expected that contact kinematics, a property that is highly desirable in
the ratio of high stiffness contributions (from shear and order to yield efficient and robust contact algorithms (see
axial tension modes) to low stiffness contributions (from e.g. [111, 112]).
torsional and bending modes), and also the condition num-
ber, increases quadratically with the beam slenderness ratio 8.6 Abstaining from of Large Rotations
𝜁 . Furthermore, it is expected that a pure neglect of shear
modes is not sufficient in order to improve conditioning. It has already been mentioned earlier in this work that
Thus, the supplementation of the proposed Kirchhoff–Love the proposed Kirchhoff–Love beam formulations provide
formulations by an additional inextensibility constraint an ideal basis for the derivation of reduced beam models
seems to be beneficial in order to get also rid of the axial which are valid under certain restrictions concerning exter-
stiffness contributions and to improve also the performance nal loads and initial geometry. For example in [110], a tor-
of linear solvers. However, the formulation of inextensibil- sion-free beam element formulation could be derived (and
ity constraints does not lie in the focus of this contribution extended to dynamic problems in [111]) that is based on a
(see the remark at the end of this section). pure centerline representation and can consequently avoid
the treatment of large rotations and associated degrees of
8.3 Improved Performance of Nonlinear Solvers freedom. Thus, many steps within a nonlinear finite ele-
ment algorithm that are typically complicated by the pres-
The relation between the performance of a nonlinear solver, ence of large rotations (e.g. spatial interpolation, time
e.g. of a Newton–Raphson scheme, and the conditioning discretization, non-symmetric tangent stiffness matrix, non-
of the considered problem, e.g. measured via the condition constant and non-symmetric mass matrix, incremental and
number of the tangent stiffness matrix, is not that clear as iterative configuration updates) are comparable to those of
in the case of linear solvers. Nevertheless, typically, it is at standard solid finite elements for this torsion-free formula-
least expected that the performance of tangent-based non- tion. For further details on this topic, the interested reader
linear solvers also deteriorates for ill-conditioned problems is referred to [110, 111].
showing slope differences of the target function by several
orders of magnitude when stepping in different directions 8.7 Conclusion
(e.g. in directions that activate shear and axial tension
modes or in directions that activate bending and torsional These sources of potential benefits were the motivation for
modes). Interestingly, all numerical examples investigated the development of shear-free element formulations based
in this work will confirm the trend that the nonlinear solver on the Kirchhoff–Love theory of thin beams. Different real-
performance of the considered Reissner type beam ele- izations of such element formulations, e.g. based on a weak
ments deteriorates drastically with increasing slenderness or on a strong enforcement of the Kirchhoff constraint,
ratio while the total number of Newton iterations required will be presented in the next sections. The influence of the
by the Kirchhoff formulations remains unchanged. aforementioned aspects on the resulting numerical behavior
will be verified in Sect. 11 via proper test cases.
8.4 Reduced System Size
Remark As discussed above, the performance of itera-
Kirchhoff type beam element formulations do not require tive linear solvers could be improved by supplementing the
any degrees of freedom for representing the mode of shear proposed Kirchhoff type element formulations by an addi-
deformation. It can be expected that, at least as long as tional inextensibility constraint. Unfortunately, in contrast
no convergence deteriorating phenomena such as locking to the Kirchhoff constraint, there is no straightforward way
occur, the same polynomial approximation and the same to enforce the inextensibility constraint directly through a
discretization error level can be guaranteed with fewer special choice of the primary variables or by a collocation
degrees of freedom. This prediction will be confirmed by approach that would allow for Lagrange multiplier elimi-
the numerical examples in Sect. 11. nation on element level, as long as the interpolation prop-
erty 𝐫(xi ) = 𝐝̂i with i = 1, 2 has to be fulfilled for the posi-
8.5 Smooth Geometry Representation tion vector field 𝐫(s) at the element boundary nodes. This
statement can easily be illustrated by considering a straight
The proposed Kirchhoff–Love beam elements will be based beam element of arbitrary order. In order to avoid zero-
on C1-continuous centerline interpolations. These inter- energy modes, inextensibility means in such a case that the
polations will eventually result in smooth beam-to-beam solution for the nodal position vectors at the two boundary
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
13
C. Meier et al.
The element formulation based on the degrees of freedom where all blank entries are zero. The transformation matri-
𝗑̂ TAN and 𝛿 𝗑̂ TAN , the residual (165) and (166) will be denoted ces 𝐓
̃ 1 and 𝐓̃ 2 follow from Eq. (22) as:
as SK-TAN element, which stands for “Strong Kirchhoff ( 1 1 1) ( 2 2 2)
−̂t 𝐒(𝐠1 ) 𝐠1 −̂t 𝐒(𝐠1 ) 𝐠1
constraint enforcement combined with nodal triad para- � 1
𝐓 = 1T
� 2
, 𝐓 = , (170)
𝐠1 0 𝐠2T 0
metrization via nodal TANgents”. Correspondingly, the 1
combination of the degrees of freedom 𝗑̂ TAN and 𝛿 𝗑̂ TAN , ther- with 𝐠i1 = exp (𝐒(𝝍̂ i ))𝐄1. In order to simplify the transfor-
esidual (164) and (166) will be denoted as SK-TAN+CS ele- mation between the sets of degrees of freedom considered
ment, referring to the Consistent Spin vector interpolation in Sects. 9.1 and 9.2, the residual 𝗋TAN,𝐝̂ and 𝗋TAN,𝚯̂ 1 accord-
underlying the Bubnov-Galerkin variant. It has to be empha- ing to (164) are slightly reordered and the sought-after
sized that the replacement of the original axial tension terms residual 𝗋ROT is introduced:
with the corresponding MCS terms according to (166) is
standard for all Kirchhoff type beam element formulations 𝗋TAN,𝐝�: = (𝐫 T , 𝐫T , 𝐫T , 𝐫T )T ,
TAN,𝐝�1 TAN,�𝐭 1 TAN,𝐝�2 TAN,�𝐭 2
considered in this work. Only in examples where for com- T
parison reasons also variants without MCS method are con- 𝗋TAN,𝚯̂ 1 : = (rTAN,Θ̂ 1 , rTAN,Θ̂ 2 , rTAN,Θ̂ 3 ) ,
1 1 1
𝗑TTAN 𝗋TAN = 𝛿�
𝛿� �T 𝗋TAN =𝛿�
𝗑TROT 𝖳 ̇ 𝗑TROT 𝗋ROT
In some scenarios, e.g. applications where complex rota- �𝐱
(172)
tional Dirichlet or coupling conditions should be prescribed �T 𝗋TAN .
→ 𝗋ROT = 𝖳�
𝐱
at the element boundary, it can be beneficial to employ the
alternative parametrization of the triads 𝚲1 and 𝚲2 at the According to (170) and (25), the matrix 𝐓 ̃ and its inverse
element boundary nodes via rotation vectors according to ̃ −1 are well-defined as long as ̂t1 ≠ 0 and ̂t2 ≠ 0. The
𝐓
Sect. 6.2.1. In such a case, an alternative set of nodal primary physical interpretation of ̂ti = 0 is that an arc-segment on
variables given by 𝗑̂ ROT : = (𝐝̂1T , 𝝍̂ 1T , ̂t1 , 𝐝̂2T , 𝝍̂ 2T , ̂t2 , 𝜑̂ 3 )T the beam centerline at the position of the node i with ini-
1T 2T
and 𝛿 𝗑̂ ROT : = (𝛿 𝐝̂1T , 𝛿 𝜽̂ , 𝛿 ̂t1 , 𝛿 𝐝̂2T , 𝛿 𝜽̂ , 𝛿 ̂t2 , 𝛿 Θ
̂ 3 )T can
1 tial length ds would be compressed to a length of zero.
be employed. Here, 𝝍̂ and 𝝍̂ represent the rotation vectors
1 2
Since such a scenario isimpossible from a physical point of
1 2
associated with these boundary triads, 𝛿 𝜽̂ and 𝛿 𝜽̂ are the view, theserequirements are assumed to be fulfilled.Conse-
corresponding spin vectors and ̂t1 and ̂t2 represent the magni- quently, the transformation from the residual vector 𝗋TAN to
tudes of the nodal tangents, i.e. ̂t1 : = ||𝐭 1 || and ̂t2 : = ||𝐭 2 ||, as the residual vector 𝗋ROT is based on a non-singular matrix
introduced in Sect. 2.2. In this case, the nodal tangents of the �𝐱̂ . The same statement holds for the transformation of the
𝖳
Hermite interpolation (97) are no primary variables anymore, global residual vectors 𝖱TAN and 𝖱ROT via the matrix 𝖳̃𝖷 rep-
but have to be expressed by 𝝍̂ 1 and ̂t1 as well as 𝝍̂ 2 and ̂t2 [see resenting an assembly of the matrices 𝖳𝗑̂ . Based on these
�
also (22)]: considerations, the following relation can be established:
𝐭̂ i → 𝐭 i = ̂ti exp (𝐒(𝝍̂ i ))𝐄1 for i = 1, 2. (168) ̃T 𝖱TAN ,
i 𝖱ROT = 𝖳
The transformation between the variations (𝛿 𝜽̂ , 𝛿 ̂ti ) as
𝖷
well as (𝛿 𝐭̂ i , 𝛿 Θ
̂ i ) for i = 1, 2 is given by the transformation ̃𝖷 ∈ ℜnX × ℜnX , rank(𝖳
with 𝖳 ̃ 𝐗 ) = nX (173)
1
matrices 𝐓 and 𝐓
̃ ̃ −1 according to (22) and leads to the fol- → 𝖱TAN = 𝟎 ⇔ 𝖱ROT = 𝟎.
lowing relation:
As long as a unique solution of 𝖱TAN = 𝟎 exists, the solu-
⎛ 𝐈3 ⎞ tion of 𝖱ROT = 𝟎 will lead to the same mechanical equilib-
⎜ 𝐓 �1 ⎟
⎜ ⎟ rium configuration. In other words, the pure re-parametri-
�𝗑̂ 𝛿 𝗑̂ ROT
𝛿 𝗑̂ TAN = 𝖳 �𝐱̂ = ⎜
with 𝖳 𝐈3 ⎟, (169) zation performed from Sect. 9.1 to 9.2 will not change the
⎜ 𝐓 ⎟
� 2
⎜ ⎟ results of the discretized beam problem. Nevertheless, the
⎝ 1⎠
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
transformation matrix 𝖳 ̃𝐗 depends on the primary degrees Moreover, it has also been stated in Sect. 6.3.3 that the dis-
of freedom in a nonlinear manner and has to be considered crete hyperelastic energy Πint,h associated with a pure 2D
within a consistent linearization procedure. bending case cannot be displayed exactly by the SK-TAN
Throughout this work, the element formulation based element. Of course, this statement still holds in 3D. In
on the degrees of freedom 𝗑̂ ROT and 𝛿 𝗑̂ ROT , the resid- Sects. 11.2 and 11.3, the expected result that the discrete
ual (165) transformed via (172) together with the MCS hyperelastic energies associated with pure bending states
approach (166) will be denoted as SK-ROT element, which in 2D and in 3D cannot be represented in an exact manner
stands for “Strong Kirchhoff constraint enforcement com- by this beam element formulation will be verified by means
bined with nodal triad parametrization via nodal ROTation of corresponding numerical test cases. There, it will also
vectors”. Correspondingly, the combination of the degrees be shown that this property leads to a slightly increased
of freedom 𝗑̂ ROT and 𝛿 𝗑̂ ROT , the residual (165) transformed discretization error level as compared to the subsequently
via (172) together with the MCS approach (166) will be derived WK-TAN element. However, this observation is
denoted as SK-ROT+CS element. Since the SK-ROT and independent from the element slenderness ratio and cannot
the SK-TAN element yield the same finite element solution be attributed to membrane locking.
(expressed via different nodal primary variables), the fol-
lowing theoretical investigations concerning locking behav-
9.4 Conservation Properties
ior and conservation properties will only be performed for
the SK-TAN element. For the SK-ROT element, a lineari-
Also for the proposed SK-TAN beam element with strong
zation Δ𝗋SK−ROT = 𝗄SK−ROT Δ̂𝗑ROT based on the increment
1T 2T enforcement of the Kirchhoff constraint, it shall be inves-
vector Δ̂𝗑ROT : = (Δ𝐝̂1T , Δ𝜽̂ , Δ̂t1 , Δ𝐝̂2T , Δ𝜽̂ , Δ̂t2 , Δ𝜑̂ 3 )T
1 2 tigated if the variational fields (150), (152) and (154)
will be employed (see “Appendix E”). Here, Δ𝜽̂ and Δ𝜽̂
required for conservation of linear momentum, conserva-
represent multiplicative increments while Δ𝜑̂ 3 can be iden-
tion of angular momentum and conservation of energy can
tified as an additive increment.
be represented by the corresponding discrete weighting
subspace h. The representation of a rigid body transla-
9.3 Avoidance of Locking Effects tion (150) is given by the nodal primary variable variations
𝛿 𝐝�j = 𝐮0 ,
In order to investigate the locking behavior of the SK-TAN
element, the investigations already made in Sect. 6.3.3 only 𝛿�𝐭 j = 𝟎, (175)
have to be extended from 2D to the general 3D case. In 3D, 𝛿Θ̂ i = 0 for j = 1, 2; and i = 1, 2, 3.
1
the Kirchhoff beam problem is described by neq = 4 differ-
ential equations (80) and constrained by neq,c = 1 constraint This result can be verified by inserting the choices for 𝛿 𝐝̂j
equation in case a pure bending state shall be represented. and 𝛿 𝐭̂ j made in (175) into the Hermite interpolation (100)
Thus, the constraint ratio of the space-continuous prob- and making use of the completeness conditions underly-
lem yields r = neq ∕neq,c = 4. Due to the employed MCS ing the Hermite polynomials (see [109]), which yields
method, the discrete number of constraints takes on a value 𝛿𝐫h (𝜉) = 𝐮0 = const. as well as 𝛿𝐫h� (𝜉) = 𝛿𝐠�1h (𝜉) = 𝟎. Insert-
of neq,c = 2nele + 1. Given the total number of equations ing these relations together with 𝛿 Θ ̂ i = 0 into either (117)
neq = 8nele + 1 after application of Dirichlet boundary con-
1
or (118) results in the required vanishing of the discrete
ditions, the discrete constraint ratio yields: spin vector field. Thus, both the Petrov–Galerkin as well as
8nele + 1 the Bubnov–Galerkin variant for the spin vector interpola-
rh = lim = 4 = r. (174) tion lead to an exact conservation of the linear momentum.
nele →∞ 2nele + 1
Next, a rigid body rotation (152) has to be displayed by the
Relation (174) yields the optimal constraint ratio for this following choice of nodal primary variable variations
element formulation. Consequently, no locking effects are
expected. Furthermore, it has been shown in Sect. 6.3.3 𝛿 𝐝�j = 𝐰0 × 𝐝�j ,
that the requirement of representing a straight beam config- 𝛿�𝐭 j = 𝐰0 × �𝐭 j , (176)
uration with arbitrary distribution of 𝜖̄h (𝜉) yields a number ̂ i = 𝐠T (𝜉 i )𝐰0 for j = 1, 2; and i = 1, 2, 3.
𝛿Θ
of independent equations that equals the number of degrees 1 1
13
C. Meier et al.
exactly represent such a “virtual” rigid body rotation, while strain 𝜖h (𝜉) but also of the shear strains Γ2,h (𝜉) and Γ3,h (𝜉).
the Petrov–Galerkin variant (117) cannot. Thus, only the Applying the constraint of vanishing shear strains in a con-
Bubnov–Galerkin interpolation can guarantee for exact sistent manner directly on the re-interpolated strain fields
conservation of angular momentum. This result will be Γ̄ 2,h (𝜉) and Γ̄ 3,h (𝜉) yields a collocation point type approach
confirmed by subsequent numerical examples. Finally, the of constraint enforcement which does not require additional
conservation of energy has to be investigated. If the nodal Lagrange multipliers (see Sect. 10.2). Throughout this con-
velocities and angular velocities of the time-continuous tribution, this variant will be preferred since it does neither
problem are chosen as nodal primary variable variations, yield additional Lagrange multiplier degrees of freedom nor
a saddle point type system of equations. Also for this element
𝛿 𝐝�j = 𝐝̇ j , formulation, the two variants concerning nodal rotation para-
(177) metrization according to Sects. 6.2.1 and 6.2.2 will be pre-
𝛿�𝐭 j = 𝐭̇ j ,
sented in Sects. 10.2 and 10.3.
𝛿Θ̂ i = 𝐠T (𝜉 i )𝐰i for j = 1, 2; and i = 1, 2, 3,
1 1
10.1 Basis: Hermitian Simo–Reissner Element
again, only the Bubnov–Galerkin variant (118) leads (per
definition) to an exact representation of the rates of the
The Reissner type beam element formulated in this sec-
discrete internal and kinetic energy and to exact energy
tion represents an intermediate step in the derivation of a
conservation for the spatially discretized, time-continuous
corresponding Kirchhoff type beam element formulation
problem. In contrast, the Petrov–Galerkin variant (117)
in the next section. The discrete beam centerline represen-
is not variationally consistent with the triad interpola-
tation is given by the Hermite interpolation (97) based on
tion (114) underlying the discrete energies and cannot
the position and tangent vectors 𝐝̂i and 𝐭̂ i at the two element
guarantee for energy conservation of the time-continuous
boundary nodes. Furthermore, the rotation interpolation is
problem.
given by a three-noded representation of (105) with nodal
triads 𝚲1, 𝚲2 and 𝚲3. Again, a finite element formulation
will be considered on the basis of a strain re-interpolation
10 Kirchhoff–Love Beam Element Based on Weak similar to the MCS method (166). While in the Kirchhoff
Constraint Enforcement case (166), only the axial strain has been treated, now the
entire deformation measure 𝚪 will be re-interpolated from
As alternative to the formulation presented in the last sec- the coordinates 𝜉 1 = −1, 𝜉 2 = 1, 𝜉 3 = 0 in order to avoid
tion, a beam element will now be presented that is based on membrane as well as shear locking:
the weak fulfillment of the Kirchhoff constraint. Thus, the
∑
3
∑
3
basis for the intended element formulation is provided by the ̄
𝚪(𝜉) = Li (𝜉)𝚪(𝜉 i ), ̄
𝛿 𝚪(𝜉) = Li (𝜉)𝛿𝚪(𝜉 i ). (178)
Simo–Reissner beam theory. In a first step, a finite element i=1 i=1
formulation of Simo–Reissner type with a C1-continuous
centerline representation will be derived in Sect. 10.1. After- On the basis of (178), also the hyper-elastic stored energy func-
wards, the Kirchhoff constraint of vanishing shear strains will tion Π ̃ int (𝛀, 𝚪) given in (37) has to be replaced by Π � int (𝛀, 𝚪)̄
be enforced in order to end up with a finite element formula- . Now, one can introduce the following set of degrees of
tion of Kirchhoff type. Following the derivations in Sect. 4.4, freedom 𝗑̂ HSR : = (𝐝̂1T , 𝐭̂ 1T , 𝝍̂ 1T , 𝐝̂2T , 𝐭̂ 2T , 𝝍̂ 2T , 𝝍̂ 3T )T
the weak statement of the Kirchhoff constraint (85) can be as well as the associated variation vector
1T 2T 3T
realized by introducing spatial interpolations for the Lagrange 𝛿 𝗑̂ HSR : = (𝛿 𝐝̂1T , 𝛿 𝐭̂ 1T , 𝛿 𝜽̂ , 𝛿 𝐝̂2T , 𝛿 𝐭̂ 2T , 𝛿 𝜽̂ , 𝛿 𝜽̂ )T . Based
multipliers and their variations, i.e. by choosing a proper dis- on the weak form (33), (178), the definitions (36)
crete trial space (𝜆Γ,2,h , 𝜆Γ,3,h ) ∈ 𝜆Γ23 ,h ∈ 𝜆Γ23 and a proper and (35), (97) and (105), the element residual vector can be
discrete weighting space (𝛿𝜆Γ,2,h , 𝛿𝜆Γ,3,h ) ∈ 𝜆Γ23 ,h ∈ 𝜆Γ23. derived as follows:
The resulting nonlinear system of discrete equilibrium equa-
1
tions will contain discrete Lagrange multipliers as additional ( ) [ ]
∫
nodal primary variables and exhibit a saddle point type struc- 𝗋HSR,𝐝̂ = 𝗏𝚪1 𝐅̄ − 𝖧T �
𝐟𝜌 Jd𝜉 − 𝖧T 𝐟𝜎
𝛤𝜎
ture. In order to avoid the additional effort of solving a large −1
system of equations with saddle point structure, a slightly dif- 1
( ) (179)
∫
ferent approach is chosen here. In the next section, a modified 𝗋HSR,𝜽̂ = 𝖫�T 𝐦 + 𝗏𝚪2 𝐅̄ − 𝖫T 𝐦
� 𝜌 Jd𝜉
Reissner type beam element formulation will be presented, −1
which is based on a smooth Hermite centerline interpolation [ ]
and a MCS type strain re-interpolation not only of the axial − 𝖫T 𝐦𝜎 = 𝟎.
𝛤𝜎
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
In (179), the following definitions have been applied: Similar to the CJ element, only the latter version (183) will
be employed throughout this contribution. The final finite
𝐅 = 𝐂F 𝚪, element residual vector resulting from the discretized coun-
∑
3
( ) terparts of these fields reads:
𝗏𝚪1 = Li (𝜉) 𝖧�T 𝚲 (𝜉 i ) ,
i=1 (180) 1
∫
�
∑ 3
( ) 𝗋WK−TAN,𝐝� = (𝗏𝜃 𝐦 + 𝗏𝜖 F̄ 1 − 𝖧T � � 𝜌 )Jd𝜉
𝐟𝜌 − 𝗏 𝜃 ⊥ 𝐦
⊥
𝗏𝚪2 = L (𝜉) 𝖫T 𝐒(𝐭)𝚲 (𝜉 i ) .
i
−1
i=1
− [𝖧T 𝐟𝜎 ]𝛤𝜎 − [𝗏𝜃⊥ 𝐦𝜎 ]𝜎 ,
This element formulation could e.g. be applied to problems of 1
( )
thick beams with higher continuity requirements (e.g. beam
∫
�
𝗋WK−TAN,𝚯� = � 𝜌 Jd𝜉
𝗏𝜃 𝐦 − 𝗏𝜃∥Θ 𝐦
contact interaction). However, here, the formulation (179) 1 ∥Θ
−1
solely represents an intermediate step in the derivation of
Kirchhoff beam elements with weak enforcement of the − [𝗏𝜃∥Θ 𝐦𝜎 ]𝛤𝜎 , (184)
Kirchhoff constraint as performed in the next two sections. ∑ 3
𝗏𝜃 ⊥ ; = Li (𝜉)𝗏𝜃⊥ (𝜉 i ),
10.2 Tangent Vector‑Based Parametrization i=1
∑
3
𝗏𝜖 ; = Li (𝜉)𝗏𝜖 (𝜉 i ),
Due to (178), the general weak constraint enforcement of
i=1
Sect. 4.4 can be simplified:
∑
Γ̄ 2 (𝜉) ≡ Γ̄ 3 (𝜉) ≡ 0
3
𝗏𝜃∥Θ ; = Li (𝜉)𝗏𝜃∥Θ (𝜉 i ),
(181) i=1
→ 𝐫 �T (𝜉 i )𝐠𝟐 (𝜉 i ) =̇ 𝐫 �T (𝜉 i )𝐠𝟑 (𝜉 i ) =̇ 0 for i = 1, 2, 3.
∑3
According to (181), the Kirchhoff constraint is exactly with 𝗏̄ �... = i=1 (L,𝜉i (𝜉)∕J)𝗏... (𝜉 i ). In the following, the for-
fulfilled at the three collocation points. In the follow- mulation based on the degrees of freedom 𝗑̂ TAN and 𝛿 𝗑̂ TAN
ing, a parametrization is chosen that directly fulfills these and on the residual (184) will be denoted as WK-TAN
constraints without the need for additional Lagrange element, which stands for “Weak Kirchhoff constraint
multipliers. Thereto, the same set of nodal degrees enforcement with nodal triad parametrization via nodal
of freedom 𝗑̂ TAN = (𝐝̂1T , 𝐭̂ 1T , 𝜑̂ 1 , 𝐝̂2T , 𝐭̂ 2T , 𝜑̂ 2 , 𝜑̂ 3 )T as TANgents”. For the WK-TAN element, a linearization
well as the same set of nodal primary variable varia- Δ𝗋WK−TAN = 𝗄WK−TAN Δ̂𝗑TAN based on the increment vec-
tions ̂ 1 , 𝛿 𝐝̂2T , 𝛿 𝐭̂ 2T , 𝛿 Θ
𝛿 𝗑̂ TAN = (𝛿 𝐝̂1T , 𝛿 𝐭̂ 1T , 𝛿 Θ ̂ 2, 𝛿Θ
̂ 3 )T tor Δ̂𝗑TAN : = (Δ𝐝̂1T , Δ𝐭̂ 1T , Δ𝜑̂ 1 , Δ𝐝̂2T , Δ𝐭̂ 2T , Δ𝜑̂ 2 , Δ𝜑̂ 3 )T is
1 1 1
as in Sect. 9.1 are chosen. In case of a Bubnov–Galerkin employed (see “Appendix D”).
approach, the following discrete spin vector field results
from the triad interpolation (105) in combination with con- Remark Actually, a collocation type approach has been
straint (181) (see also (107)): applied in order to enforce the Kirchhoff constraint. Nev-
ertheless, the notion “weak constraint enforcement” is kept
∑
3
throughout this work since such a procedure still represents
𝛿𝜽h (𝜉) = �𝐈i (𝜉)𝛿𝜽i
the basis of the space-continuous formulation. Moreover,
(
i=1
) (182) the difference to the formulation based on “strong con-
� i � i
with 𝛿𝜽i = ̂ i 𝐠1 (𝜉 i ) + 𝐒[𝐫 (𝜉 )]𝛿𝐫 (𝜉 ) .
𝛿Θ straint enforcement” of Sect. 9 shall be emphasized by this
1
||𝐫 � (𝜉 i )||2 naming.
Since the Kirchhoff constraint is exactly fulfilled at the
three element nodes, the constrained variant (71) of the 10.3 Rotation Vector‑Based Parametrization
nodal spin vectors 𝛿𝜽i has been combined with (107).
Based on the alternative Petrov–Galerkin approach (110), Also for the element formulation of Sect. 10.2 based on
the Kirchhoff constraint given by (181) yields the following weak Kirchhoff constraint enforcement, a coordinate
expression for the spin vector field: transformation from 𝗑̂ TAN and 𝛿 𝗑̂ TAN to the alternative pri-
mary variables 𝗑̂ ROT : = (𝐝̂1T , 𝝍̂ 1T , ̂t1 , 𝐝̂2T , 𝝍̂ 2T , ̂t2 , 𝜑̂ 3 )T
∑
3
and
1T 2T
̂ 3 )T
𝛿 𝗑̂ ROT : = (𝛿 𝐝̂1T , 𝛿 𝜽̂ , 𝛿 ̂t1 , 𝛿 𝐝̂2T , 𝛿 𝜽̂ , 𝛿 ̂t1 , 𝛿 Θ
𝛿𝜽h (𝜉) = Li (𝜉)𝛿𝜽i 1
can be performed. The transformation rule for the ele-
(
i=1
) (183) ment residual vector 𝗋ROT = 𝖳 �T 𝗋TAN is identical to (172) in
� i � i 𝗑̂
̂ i 𝐠1 (𝜉 i ) + 𝐒[𝐫 (𝜉 )]𝛿𝐫 (𝜉 ) .
with 𝛿𝜽i = 𝛿 Θ Sect. 9.2. Throughout this work, the element formulation
1
||𝐫 � (𝜉 i )||2
13
C. Meier et al.
based on the degrees of freedom 𝗑̂ ROT and 𝛿 𝗑̂ ROT and on by 𝛀h (𝜉) as well as the axial tension values at the collo-
the element residual vector (184) transformed via (172) cation points represented by 𝜖(𝜉CP ) are non-competing and
will be denoted as WK-ROT element, which stands no locking effects are expected. Again, these considerations
for “Weak Kirchhoff constraint enforcement com- can easily be extended to arbitrary curvature fields 𝐊h (𝜉)
bined with nodal triad parametrization via nodal ROTa- that are representable by the employed triad interpola-
tion vectors”. For the WK-ROT element, a linearization tion and arbitrary second-order polynomials 𝜖(𝜉) ̄ accord-
Δ𝗋WK−ROT = 𝗄WK−ROT Δ̂𝗑ROT based on the increment vector ing to (166). In Sects. 11.2 and 11.3, the expected result
1T 2T
Δ̂𝗑ROT : = (Δ𝐝̂1T , Δ𝜽̂ , Δ̂t1 , Δ𝐝̂2T , Δ𝜽̂ , Δ̂t1 , Δ𝜑̂ 3 )T will be that the discrete hyperelastic energies associated with pure
employed (see “Appendix E” ). bending states in 2D and in 3D can exactly be represented
by the WK-TAN/ROT elements will be verified by means
10.4 Avoidance of Locking Effects of corresponding numerical test cases.
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
Fig. 4 Objectivity test: rigid body rotation of an initially stress-free beam with quarter circle geometry. a Problem setup. b Internal energy due
to imposed rigid body rotation
as the fulfillment of conservation properties will be veri- considered in the following sections, a Newton–Raphson
fied for the different beam element formulations presented scheme based on consistent linearization has been applied
in Sects. 7–10. Finally, based on the arguments given in in order to solve the set of nonlinear equations resulting
Sect. 8, the focus will also lie on detailed comparisons of from the temporally and spatially discretized weak form of
Reissner and Kirchhoff type beam element formulations for the balance equations. As convergence criteria, the Euclid-
example with respect to the resulting discretization error ean norms of the displacement increment vector Δ𝖷k+1 n+1
and
level or the performance of the Newton–Raphson scheme. of the residual vector 𝖱(𝖷k+1n+1
) are checked.
Since the Kirchhoff type beam element formulations For convergence, these norms have to fall below pre-
based on a tangent-based triad parametrization and the scribed tolerances 𝛿𝖱 and 𝛿𝖷, i.e. ||𝖱(𝖷k+1 n+1
)|| < 𝛿𝖱 and
formulations based on a rotation vector-based triad para- ||Δ𝖷k+1n+1
|| < 𝛿 𝖷. Typical convergence tolerances chosen
metrization (compare e.g. Sects. 9.1 and 9.2 or Sects. 10.2 for the subsequent examples are in the range of 𝛿𝖷 = 10−8
and 10.3) have been shown to yield identical FEM solu- as well as 𝛿𝖱 = 10−7 , 10−9 , 10−11 , 10−13 for the slender-
tions, only the former category will be investigated with ness ratios 𝜁 = 10, 100, 1000, 10000. For the (quasi-)static
respect to spatial discretization errors. Furthermore, for problems presented in the following sections, the external
all examples without analytic solution, the standard choice loads are applied on the basis of an incremental procedure,
for the reference solution 𝐫ref (see also (146) of Sect. 6.3.4) where N shall denote the number of load steps and Δt the
is a numerical solution via the WK-TAN element (see load step size. As long as nothing is stated to the contrary,
Sect. 10.2) employing a spatial discretization that is by a the following simple procedure is applied in order to adapt
factor of four finer than the finest discretization shown in the load step size during the static simulation in an efficient
the corresponding convergence plot. In order to achieve a manner: initially, a comparatively small load step size Δt0 is
good comparability among the different geometries and chosen, e.g. according to N0 = 1. If the Newton–Raphson
load cases, a standard set of geometrical and constitutive scheme has not converged within a prescribed number of
parameters has been applied in all simulations unless stated niter,max iterations, the step size is halved and the load step
otherwise. This standard set consists of a beam with initial is repeated. This procedure is repeated until convergence
length l = 1000 and square cross-section with side length can be achieved. Then, after four converging load steps on
R. These parameters lead to a cross-section area of A = R2 the low step size level, the step size is doubled again. Also
and to moments of inertia of area of I2 = I3 = :I = R4 ∕12 this procedure of successively doubling the step size after
and IT = R4 ∕6. Different beam slenderness ratios 𝜁: = l∕R four converging load steps at the current step size level is
are generated by varying the value R = 0.1, 1.0, 10, 100 of repeated until the original step size Δt0 is reached again.
the cross-section side length. The standard choice for the This procedure will not only drastically increase the overall
constitutive parameters is E = 1.0 and G = 0.5, thus lead- computational efficiency, it also allows for comparatively
ing to EI = GIP = R4 ∕12. For all numerical examples objective and fair comparisons of the performance of the
13
C. Meier et al.
Newton–Raphson scheme for different element formula- formulation investigated in [109] will be considered (see
tions. In subsequent numerical examples, such comparisons also the last remark at the end of Sect. 6.2.4). For clear-
will be made on the basis of the accumulated number of ness, the internal energy Πint is normalized by the factor
Newton iterations Πint,r = 0.5EI𝜋 2 ∕(4l), which is equal to the mechanical
work that is required to bend the initially stress-free quarter
∑
N
niter,tot : = niter,n , (185) circle into a straight beam by means of a discrete, external
n=1 end-moment. Of course, the internal energy should vanish
required to solve the entire problem. Here, niter,n is the for a beam that is merely rotated out of its stress-free initial
number of iterations required for load step tn. In “load step configuration in the sense of a rigid body rotation.
adaption scheme” above, non-converging steps are consid- From Fig. 4b, it becomes obvious, however, that the
ered in the total number of iterations with niter,n = niter,max. internal energy of the SR formulation increases over the
number of rotations, which is a clear indication for the
already theoretically predicted non-objectivity. Within
11.1 Example 1: Verification of Objectivity 10 rotations, the normalized energy reaches a value of
almost Πint,r ∕4, which results in a clearly visible deforma-
The objectivity of the Kirchhoff beam element formula- tion of the initial quarter circle. On the contrary, the inter-
tions proposed in Sects. 9 and 10 has already been proven nal energy of all the other investigated element formula-
theoretically. In order to verify these results numerically, tions results in a value that is zero up to machine precision.
the following test case will be investigated (see Fig. 4a): Finite element formulations based on such interpolation
at the clamped end of an initially curved beam with slen- schemes as the investigated SR element might show rea-
derness ratio 𝜁 = 10, whose stress-free centerline con- sonable results for static test cases (see e.g. [109]). How-
figuration equals a quarter circle, a Dirichlet rotation with ever, especially in dynamic problems involving consider-
respect to the global x-axis is imposed. For the presented able rigid body motions, non-physical results as well as a
quasi-static example, a total rotation angle of 20𝜋, increas- drastic deterioration of the conservation properties investi-
ing linearly over 100 load steps, is prescribed. In order to gated in Sect. 6.3.5 can follow from the application of such
investigate objectivity, the normalized internal (hyperelas- element formulations.
tic) energy is plotted over the total number of rotations (see
Fig. 4b) for the WK-TAN and SK-TAN element as well as 11.2 Example 2: Pure Bending in 2D
for the Bubnov–Galerkin variant of the SK-TAN element
formulation with Consistent Spin (SK-TAN+CS) vec- The examples shown in this section exclusively focus on
tor interpolation according to (164). For comparison rea- 2D geometries and load cases. The section is sub-divided
sons, also the non-objective SR Kirchhoff beam element into two subsections: in Sect. 11.2.1, two load cases, a pure
Fig. 5 Initial and deformed configuration of an initially straight beam for two different load cases M and M+F. a Straight beam bent by end-
moment. b Straight beam bent by end-moment and -force
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
0 0
10 10
-2 -2
10 10
-4 -4
10 10
rel
rel
2
2
-6 -6
10 10
||e||
||e||
10 -8 SK-TAN-FI, ζ=10 10 -8 SK-TAN-MCS, ζ=10
SK-TAN-FI, ζ=100 SK-TAN-MCS, ζ=100
-10 SK-TAN-FI, ζ=1000 -10 SK-TAN-MCS, ζ=1000
10 SK-TAN-FI, ζ=10000 10 SK-TAN-MCS, ζ=10000
th
4 order 4 th order
-12 -12
10 10
10 1 10 2 10 3 10 4 10 1 10 2 10 3 10 4
element length element length
(a) (b)
10 0 10 0
-2 -2
10 10
10 -4 10 -4
rel
rel
2
-6 -6
10 10
||e||
||e||
ζ=10000, SK-TAN-MCS
-8 -8
10 ζ=10000, SK-TAN-MCS 10 ζ=10000, SK-TAN-ANS
ζ=10000, SK-TAN-ANS ζ=10000, SK-TAN-RI
ζ=10000, SK-TAN-RI ζ=10000, SK-TAN-FI
10 -10 ζ=10000, WK-TAN-MCS 10 -10 ζ=10000, WK-TAN-MCS
4 th order 4 th order
-12 -12
10 10
10 1 10 2 10 3 10 4 10 1 10 2 10 3 10 4
element length element length
(c) (d)
Fig. 6 Straight beam subject to the load cases M and M+F: relative L2-error plotted over the element length. a Reference: analytic, different
slendernes ratios. b Reference: analytic, different slenderness ratios. c Reference: analytic ζ = 10,000. d Reference: WK-TAN-MCS, ζ = 10,000
pending case as well as a combined moment-and-force load 11.2.1 Comparison of Different Anti‑Locking Methods
case, yielding geometrically nonlinear, but still moderate
centerline deformations, are considered. This section aims An initially straight beam is clamped at one end. Two dif-
at the investigation of membrane locking effects and at the ferent load cases will be analyzed: the first load case M is
comparison of different anti-locking tools, especially of identical to the example analyzed in Sect. 6.3.3 and solely
the MCS method proposed in Sect. 6.3.3. In Sect. 11.2.2, consists of a discrete end-moment 𝐌 = (0, 0, M)T applied
again a pure bending and a combined moment-and-force in one load step. The moment M = EI𝜋∕(2l) exactly bends
load case will be considered. However, due to higher load the beam into a quarter-circle shaped arc. In the second
factors, the resulting degree of deformation is further load case M+F, the end-moment and an additional tip force
increased as compared to the examples of Sect. 11.2.1. 𝐅 = (0, F, 0)T in global y-direction are applied in one load
This higher degree of deformation reveals clear differences step. The initial and deformed geometries for these load
in the approximation quality of the WK and SK Kirchhoff cases are illustrated in Fig. 5.
beam element variants. Besides the comparison of these While the standard parameters and the slenderness ratios
two variants, also a first proof of concept for the devel- 𝜁 = 10, 100, 1000, 10000 have been chosen for the load
opment of higher-order Hermitian Kirchhoff elements is case M, only the highest and therefore most critical slen-
given. derness ratio 𝜁 = 10000 in combination with an external
force of F = 1.0 ⋅ 10−10 ≈ 10M∕l has been investigated
13
C. Meier et al.
for the load case M+F. In a first step, the SK-TAN element error increases almost by two orders of magnitude when
formulation according to (165) is applied in combination enhancing the slenderness ratio from 𝜁 = 10 to 𝜁 = 10000.
with a full Gauss–Legendre integration (“SK-TAN-FI”) However, Fig. 6a also reveals that this effect decreases
with nG = 4, a reduced Gauss–Legendre integration (“SK- with decreasing element sizes and almost completely dis-
TAN-RI”) with nG = 3, a classical assumed natural strain appears for discretizations with more than 32 elements.
approach (“SK-TAN-ANS”) as well as the MCS method The reason for this behavior lies in the fact that the ele-
according to (166) based on nG = 4 integration points ment slenderness ratio 𝜁ele = lele ∕R is the key-parameter
(“SK-TAN-MCS”) (see Sect. 6.3.3 and [110] for further for the observed locking effect and that the latter also
details on these variants). In Fig. 6a, the relative L2-error decreases with decreasing element sizes. However, for typi-
of the load case M and different slenderness ratios is plot- cal engineering applications with relative error bounds in
ted with respect to an analytic reference solution. For spa- the range of 1%, the effect is by no means negligible. For
tial discretization, the variant SK-TAN-FI based on meshes sufficiently fine discretizations the expected convergence
with 1, 2, 4, 8, 16, 32 and 64 elements has been applied. order of four is reached. In Fig. 6b, the relative L2-error is
Accordingly, the convergence is slowed down dramatically plotted for the same slenderness ratios as before, but for
with increasing slenderness ratio. If the beam is e.g. dis- the variant SK-TAN-MCS, which is supplemented by the
cretized by one finite element ((lele = 1000), the relative MCS method according to (166). As expected, the locking
Fig. 7 Initial and deformed configuration of an initially straight beam for two load cases M and M+F at higher loads. a Straight beam bent by
end-moment. b Straight beam bent by end-moment and -force
0 0
10 10
-2 -2
10 10
10 -4 10 -4
e,rel
rel
2
-6 -6
10 10
||e||
||e||
(a) (b)
̃ : relative L2- and energy error plotted over element length. a Reference: analytic, L2-error. b
Fig. 8 Initially straight beam under load case M
Reference: analytic, energy error.
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
effect completely disappears for all investigated slenderness behavior will change with increasing deformation. In the
ratios. However, as shown in Fig. 6c, for the load case M following, the MCS method will be employed per default
and the highest investigated slenderness ratio of 𝜁 = 10000, and the additional abbreviation ...-...-MCS omitted.
the same effect can alternatively be achieved by applying a
simple reduced integration procedure (variant SK-TAN-RI) 11.2.2 Comparison of Different Element Formulations
or a classical ANS approach (variant SK-TAN-ANS). On
the contrary to load case M, Fig. 6d reveals that no distinc- In the convergence plots investigated for the two load
tive improvement of the locking behavior can be obtained cases M and M+F, no noteworthy differences between
by these alternative methods for the load case M+F: both the SK and WK element could be observed. In order to
the ANS approach as well as the reduced integration investigate the difference in the two general approaches
scheme can only slightly alleviate the locking effect in the of enforcing the Kirchhoff constraint in a strong or in a
range of rather coarse discretizations as compared to the weak manner further in detail and to perform first com-
variant SK-TAN-FI. The MCS approach, however, com- parisons with geometrically exact beam element formu-
pletely eliminates the error offset due to membrane lock- lations of Simo–Reissner type, two additional load cases
ing also for this load case. The explanation for this obser- will be considered (see Fig. 7): the first load case con-
vation is obvious and in agreement with the statements of sidered in this section, in the following denoted as M ̃,
Sect. 6.3.3: similar to the working principle of the MCS simply increases the magnitude of the external moment
method, the reduced integration scheme can alleviate lock- by a factor of eight as compared to the previous load
ing by reducing the number of constraint equations. case M, i.e. M ̃ = 8M , thus leading to a deformed geom-
Yet, as shown in Sect. 6.3.3, the MCS method leads to etry that is represented by a double circle (see Fig. 7a).
a lower number of constraint equations as compared to the Since the contribution of Simo and Vu-Quoc [140], this
simple reduced integration scheme, which makes the lat- load case has been established as a standard test case for
ter method less effective. While the reduced integration geometrically exact beam element formulations. Finally,
scheme seems to be sufficient for the load case M, a spe- in a fourth load case, denoted as M ̃ +F
̃ , the end-moment
cial case yielding symmetric curvature distributions within M = 8M is supplemented by a tip force 𝐅 = (0, F,
̃ ̃ 0)T in
the elements, the more general deformed configurations global y−direction, whose magnitude is this time exactly
resulting from the load case F+M already demonstrate the chosen as F ̃ = 0.08M . The initial geometry as
̃ = 10M∕l
limits of this simple method. On the other hand, the work- well as the final configuration of this last load case is
ing principle of the standard ANS method does not aim illustrated are Fig. 7b.
at a reduction of the number of constraint equations, but In Fig. 8a, the L2-error resulting from the load case
rather at an evaluation of the critical axial tension term at M̃ has been plotted for the WK-TAN and the SK-TAN
selected collocation points with vanishing parasitic strains. element as well as for the Reissner type CJ beam ele-
In the geometrically nonlinear regime of large deforma- ment formulation proposed by Crisfield and Jelinic [43,
tions, the parameter space positions of these optimal col- 83] and presented in Sect. 7. Here, discretizations with
location points are deformation-dependent. Obviously, the 8, 16, 32, 64, 128 and 256 elements have been employed.
load case M+F already leads to a change in these positions These discretizations are comparable to the load case
up to an extent that almost completely destroys the working M since similarly to that example, also here, the rough-
principle and impact of the ANS method. Summing up, it is est discretization is based on one finite element per 90◦ -
to say that based on these two examples the proposed MCS arc segment of the analytic solution. In order to enable a
method seems to be superior to standard methods such as reasonable comparison of different element formulations,
reduced integration or ANS in terms of locking avoidance here and in the following, the discretization error will
when combined with the considered geometrically exact be plotted over the total number of degrees of freedom
Kirchhoff beam elements. In example 8 of Sect. 11.6, a fur- resulting from the respective finite element discretiza-
ther comparison of these different anti-locking methodolo- tion. Since no shear deformation is present for this exam-
gies on the basis of a very general problem setting involv- ple, the Reissner and Kirchhoff type elements converge
ing 3D deformation states and an initially curved geometry towards the same analytic solution. All element formula-
will be presented, which will confirm this result. For com- tions exhibit the expected optimal convergence order of
pleteness, in Fig. 6c, d, also the L2-error of the WK-TAN- four, indicated by the black dashed line. The WK-TAN
MCS element based on a weak enforcement of the Kirch- element shows the expected result that Kirchhoff element
hoff constraint according to (184) has been plotted. For the formulations can represent the same discretization error
load cases M and M+F, this formulation yields a compara- level with less degrees of freedom as compared to the
ble convergence behavior and discretization error level as Reissner type element formulation (see Sect. 8). Further-
the SK-TAN-MCS element. Next, it will be shown that this more, for this example, it can even be shown that the lines
13
C. Meier et al.
0 0
10 10
Crisfield & Jelenic Crisfield & Jelenic
-2 SK-TAN -2 SK-TAN
10 4 th order
10 4 th order
6 th order 6 th order
-4 -4
10 10
rel
rel
2
2
-6 -6
10 10
||e||
||e||
-8 -8
10 10
-10 -10
10 10
-12 -12
10 10
1 2 3 4 1 2 3 4
10 10 10 10 10 10 10 10
degrees of freedom degrees of freedom
(a) (b)
representing the discretization error of the WK-TAN ele- in (149) vanishes, the finite element problem degener-
ment and of the CJ element formulation would be almost ates to a pure problem of polynomial curve approxima-
identical if the discretization error was plotted solely over tion represented by the first term in (149) and, thus, the
the degrees of freedom associated with the centerline discretization error plotted over the number of centerline
interpolation. Thus, the observable difference in Fig. 8a DoFs yields similar results for the Lagrange centerline
is a pure result of the additional rotational degrees of interpolation of the Reissner type element and the Her-
freedom required for Reissner type element formulations mite centerline interpolation of the Kirchhoff type ele-
in order to represent shear deformation. Such a behavior ment. The situation is completely different for the SK-
is expected for this pure bending example since the two TAN element, which cannot exactly represent the internal
considered element formulations can exactly represent energy associated with a pure bending state. A closer
the internal energy associated with a pure bending state investigation would confirm the expected result that
(see Sects. 7.2 and 10.4). Consequently, the discretiza- the SK-TAN element exhibits a remaining error in the
tion error contribution stemming from the second term
0 0
10 10
-2 -2
10 10
10 -4 10 -4
rel
rel
2
-6 -6
10 10
||e||
||e||
-8 -8
10 10
Crisfield & Jelenic Crisfield & Jelenic
-10
WK-TAN -10
WK-TAN
10 SK-TAN 10 SK-TAN
4 th order 4 th order
10 -12 10 -12
10 1 10 2 10 3 10 4 10 1 10 2 10 3 10 4
degrees of freedom degrees of freedom
(a) (b)
Fig. 10 Initially straight beam under load case M ̃ : L2-error for different element formulations. a Reference: Crisfield and Jelenić, ζ = 100. b
̃ +F
Reference: Crisfield and Jelenić, ζ = 10,000
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
length-specific hyperelastic stored energy (75), which is From a rather mathematical point of view, the increased
more or less constant along the beam length. discretization error level of the SK-TAN element can be
Based on this finding, it can easily be answered why the explained by a high level of the second, energy-related term
discretization error level of the SK-TAN element applied to in (149) that dominates the overall discretization error. This
the load case M ̃ is considerably increased as compared to behavior, in turn, is a pure consequence of the fact that the
the first load case M (while an identical level of the length- two exponents k + 1 and 2(k − m + 1) in (149) are identi-
specific L2-error has been observed for the WK-TAN and cal for trial functions of polynomial degree k = 3. However,
the CJ element): by the FEM, solely the distribution of for polynomial degrees k > 3, the second term in (149) is
the second centerline derivative 𝐫 ′′ is optimized in order expected to converge with a higher rate and consequently,
to yield a minimal energy error within the beam domain for sufficiently fine discretizations, the first term reflect-
Ωl , while the centerline field 𝐫 itself is only constrained at ing the pure polynomial approximation power will deter-
the clamped end of the beam. Thus, with increasing dis- mine the overall discretization error level. In this range, a
tance from the clamped end, the discretization error in the lower discretization error per DoF can be expected for the
centerline field 𝐫, resulting from a two-fold integration of Kirchhoff type beam element formulations independently
the (more or less) constant error in the second derivative of the beam length, the complexity of the deformation
𝐫 ′′ along an increasing arc-length segment, also increases. state or the type of boundary conditions. For a first proof
Consequently, by assuming comparable errors in the of principle, in Fig. 9b, the L2-error resulting from the CJ
length-specific energy for comparable discretizations (i.e. element with fifth-order Lagrange interpolation as well as
the same number of finite elements representing the same from the SK-TAN element based on a fifth-order Hermite
angle segment of the analytic solution), a higher length- interpolation are depicted: while for very rough discretiza-
specific discretization error is expected for the load case tions, the energy-related error contribution with higher
M̃ as compared to the load case M. Fig. 8b confirms the convergence rate still seems to dominated the overall dis-
expected result that the energy error of the SK-TAN ele- cretization error of the SK-TAN element, the expected opti-
ment does not vanish for this example and exhibits a con- mal gap between the Reissner discretization error (higher
vergence order of four. Furthermore, it is shown that the level) and the Kirchhoff discretization error (lower level)
length-specific energy error “averaged” along the entire can be observed. For comparison reasons, in Fig. 9a, the
beam length is identical to the length-specific energy error results of the corresponding third-order variants of Fig. 8a
“averaged” only along the first eighth of the beam (repre- are repeated. Since this work focuses on the development
senting a quarter circle). As consequence of the error accu- of third-order Kirchhoff beam elements, no further details
mulation described above, the length-specific L2-error is on the construction of higher-order Hermite polynomials
lower (and similar to load case M) if it is only “averaged” (either by introducing additional nodes or by considering
along the first eighth of the beam (see Fig. 8a). higher-order derivatives) will be given at this point. How-
ever, it is expected that a comparable behavior as illustrated
10 0 10 0
10 -2 10 -2
10 -4 10
-4
e,rel
e,rel
-6 -6
10 10
||e||
||e||
-8
10 10 -8
-10
WK-TAN WK-TAN
10 SK-TAN 10 -10 SK-TAN
4 th order 4 th order
-12 -12
10 10
1 2 3 4 1 2 3 4
10 10 10 10 10 10 10 10
degrees of freedom degrees of freedom
(a) (b)
Fig. 11 Initially straight beam under load case M ̃ : energy-error for different element formulations. a Reference: WK-TAN, ζ = 100. b Refer-
̃ +F
ence: WK-TAN, ζ = 10,000
13
C. Meier et al.
2000 2000
Crisfield & Jelenic Crisfield & Jelenic
WK-TAN WK-TAN
SK-TAN SK-TAN
1500 WK-ROT 1500 WK-ROT
SK-ROT SK-ROT
Newton iterations
Newton iterations
1000 1000
500 500
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
number of elements number of elements
(a) (b)
Fig. 12 Initially straight beam under load case M ̃ : total number of Newton iterations. a Moderate slenderness ratio: ζ = 100. b High slender-
̃ +F
ness ratio: ζ = 10,000
in Fig. 9b, can also be achieved for the other test cases con- for the solution of the geometrically linear theory. This
sidered in the following, if fifth-order SK and WK elements result is remarkable for this highly nonlinear example.
are employed. A detailed investigation of general geometri- From Fig. 10a, b, it can again be observed that all ele-
cally exact beam element formulations of Kirchhoff–Love ment formulations exhibit the expected convergence rate
type with polynomial degree k > 3 will be considered in of four, that the error level of the SK-TAN element lies
future research work. slightly above and that the error level of the WK-TAN ele-
Eventually, also the convergence behavior of the fourth ment lies below the error level of the CJ element. Despite
load case shall be investigated. In Fig. 10a, b the L2-error the fact that the Simo–Reissner formulations yield the more
of the CJ, the WK-TAN and the SK-TAN element is plotted general solutions, which also contain the effects of shear
for the two beam slenderness ratios 𝜁 = 100 and 𝜁 = 10000. deformation, the Kirchhoff type WK-TAN element formu-
Since no closed-form analytic solution has been available lation will be considered as numerical reference solution
for this example, a numerical reference solution based on for all remaining examples throughout this work. Such a
the element formulation of Crisfield and Jelenić has been procedure seems to be sensible, since within this contribu-
employed. As a consequence of shear deformation induced tion, the convergence behavior of the Kirchhoff type ele-
by the tip force F̃ , the result derived from the Reissner and ments and not of the Reissner type elements shall be stud-
Kirchhoff type beam element formulations will differ in the ied. Nevertheless, the model error of the Kirchhoff type
limit of very fine discretizations h → 0. This “model error” beam elements is still observable, but this time in form of a
of the shear-free Kirchhoff elements becomes visible in kink and a remaining cutoff error level in the convergence
form of a kink in the convergence diagram and a certain plots of the Reissner type formulation.
cutoff error level that remains constant even for arbitrar- While the WK-TAN element was able to exactly represent
ily fine discretizations. As expected, the model difference the internal energy of the load case M ̃ , now, for the load case
between the Simo–Reissner and the Kirchhoff–Love beam M̃ +F̃ , the energy convergence of the formulations WK-TAN
theory decreases with increasing beam slenderness ratio, a and SK-TAN can be compared (see Fig. 11a, b for the two
property that is reflected by a lower cutoff error level for slenderness ratios 𝜁 = 100 and 𝜁 = 10000). Both element
the higher slenderness ratio 𝜁 = 10000. For the lower slen- formulations exhibit the expected convergence order of four,
derness ratio 𝜁 = 100, the relative error distinguishing the and similar to the load case M ̃ , the WK-TAN element yields
Kirchhoff from the Reissner solution lies below 10−3, which a better approximation of the internal energy than the SK-
can be assumed as reasonable approximation for many TAN element. This is observable in form of a lower energy
engineering applications. For the high slenderness ratio error level in Fig. 11a, b and the reason for the lower L2-error
𝜁 = 10000, the relative error between these two models is level visible in Fig. 10a, b. A possible explanation for the
smaller than 10−7. For the investigated cases of 𝜁 = 100 and better performance of the WK-TAN element may be found
𝜁 = 10000, the cutoff error scales almost quadratically with by considering the interaction of the employed translational
the slenderness ratio, which would be the expected result and rotational interpolation schemes: as shown in Sect. 10.4,
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
0
10
-2
10
-4
10
rel
2
-6 Cri. & Jel., ζ=10 2
10
||e||
Cri. & Jel., ζ=10 4
-8 WK-TAN, ζ=10 2
10
WK-TAN, ζ=10 4
2
SK-TAN, ζ=10
-10
10 SK-TAN, ζ=10 4
4 th order
-12
10
1 2 3 4
10 10 10 10
degrees of freedom
(a) (b)
Fig. 13 Initially straight beam clamped at one end and loaded with a three-dimensional discrete end-moment. a Initial and deformed geometry.
b Relative L2-error, reference: analytic
the number of unknowns equals the number of equations no difference between the slenderness ratios 𝜁 = 100 and
required for the WK-TAN element to (energetically) repre- 𝜁 = 10000 is evident, which again underlines the success-
sent a pure bending state. This means that an exact repre- ful avoidance of membrane locking and the effectiveness of
sentation of the internal energy associated with states of the MCS method. Finally, also the performance of the New-
constant axial tension, bending curvature and torsion is pos- ton–Raphson scheme shall be investigated and compared
sible. This property does not hold for the SK-TAN element between Kirchhoff and Reissner type element formulations
(see e.g. Sect. 9.3), where the corresponding system of equa- (see Fig. 12). Since the computationally expensive steps of
tions that has to be fulfilled for representing a pure bend- solving a nonlinear system of equations and evaluating the
ing state is slightly over-constrained. While the difference tangent stiffness matrix have to be conducted in every New-
between the WK and the SK elements in the L2-error level ton iteration, a reduction in the total number of Newton itera-
is expected to vanish with higher-order trial functions, such tions niter,tot as defined in (185) would considerably increase
an effect can in general not be predicted for the energy error. the overall efficiency of the numerical algorithm. In Fig. 12a,
Eventually, it has to be mentioned that for both the L2-error b, the total number of Newton iterations niter,tot of the load
as well as the energy error plots of the load case M ̃ +F̃, case M̃ +F̃ in combination with slenderness ratios of 𝜁 = 100
15000 15000
Crisfield & Jelenic
Simo & Vu Quoc
WK-TAN
SK-TAN
WK-ROT
Newton iterations
10000
Newton iterations
10000
SK-ROT Crisfield & Jelenic
Simo & Vu Quoc
WK-TAN
SK-TAN
WK-ROT
5000 5000 SK-ROT
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
number of elements number of elements
(a) (b)
Fig. 14 Initially straight beam loaded with a three-dimensional discrete end-moment: total number of Newton iterations. a Moderate slender-
ness ratio: ζ = 100. b High slenderness ratio: ζ = 10,000
13
C. Meier et al.
and 𝜁 = 10000 has been plotted for the element formulations the two slenderness ratios 𝜁 = 100 and 𝜁 = 10000. How-
CJ, WK-TAN, SK-TAN, WK-ROT and SK-ROT and differ- ever, this time, the beam is loaded by a 3D end-moment
ent spatial discretizations. While the final FEM solutions 𝐌1 : = (M, 0, M)T , with M = 10 for 𝜁 = 100 and M = 10−7
have been shown to be independent from the choice of nodal for 𝜁 = 10000, which contains an additional moment com-
rotation parametrization, the number of Newton iterations ponent in beam length direction inducing torsion. The ini-
required for the SK/WK-ROT and SK/WK-TAN variants tial and deformed configuration are illustrated in Fig. 13a.
might differ considerably. Therefore, also the Newton per- As argued in [110], the analytic solution of this example is
formance of these variants has been investigated. For solv- given by the following space curve representation:
ing the highly nonlinear beam problem, the load step adap-
tion scheme mentioned above based on an initial number of ⎛ √1 (sin 𝛽 + 𝛽) ⎞
⎜ 2 ⎟ EI s
N0 = 2 load steps has been employed. 𝐫(s) = R0 ⎜ (1 − cos 𝛽) ⎟, R0 = , 𝛽=√ . (186)
⎜ √ (𝛽 − sin 𝛽) ⎟ 2M
By comparing Fig. 12a, b, one realizes that the Newton 1 2R0
⎝ 2 ⎠
performance of the Kirchhoff type element formulations is
rarely influenced by the considered slenderness ratio, while
the number of Newton iterations required by the Reissner This solution represents a helix whose mid-axis points into
element increases drastically with increasing slenderness the (1, 0, 1)T -direction, viz. in the direction of the applied
ratio. Furthermore, it seems that the SK/WK-TAN variants external moment. The special parameter choice of this
require fewer Newton iterations than the SK/WK-ROT var- example leads to a radius R0 of the enveloping cylinder
iants. These trends will be confirmed, and even more pro- that is identical to the slope of the helix. In Fig. 13b, the
nounced, in more general 3D examples presented in sub- relative L2-error resulting from the two investigated slen-
sequent sections. In summary, the following conclusions derness ratios is plotted for the element formulations CJ,
can be drawn from the 2D examples considered in this sec- WK-TAN and SK-TAN as well as spatial discretizations
tion: (1) The proposed Kirchhoff elements yield accurate based on 8, 16, 32, 64, 128 and 256 elements. Again, all
results with acceptable model errors for slenderness ratios element formulations exhibit the expected convergence
of 𝜁 ≥ 100 and a model error that decreases quadratically order of four and the discretization error level of the WK-
with increasing beam slenderness ratio. (2) The expected TAN element is lower, whereas the discretization error of
convergence orders of four in the L2- as well as in the the SK-TAN element is slightly higher than for the CJ ele-
energy error could be confirmed for all investigated Kirch- ment. Furthermore, no visible difference can be observed
hoff elements. (3) In combination with the MCS method, between the discretization error levels associated with
none of the considered element formulations exhibited an the two different slenderness ratios. Due to the choice
influence of the element slenderness ratio on the resulting GIT = EI2 = EI3, it can easily be verified that this example
discretization error. This result confirms the effectiveness results in an analytic solution exhibiting vanishing axial
of the MCS method in the avoidance of membrane lock- tension and shear deformation as well as a constant spatial
ing. (4) The discretization error level of the WK-TAN ele- and material curvature vector along the entire beam point-
ment lies below the error level of the Reissner type element ing into the direction of the external moment vector, i.e.
and also below the error level of the SK-TAN element. 𝐤 = 𝐊 = M∕(EI)(1, 0, 1)T = const.
The increased error level of the SK-TAN element has been Thus, already the roughest discretizations of the CJ and
shown to vanish with higher polynomial degree k > 3 of the WK-TAN elements can exactly represent the hyperelastic
trial functions. (5) While the total number of Newton itera- stored energy function for this pure bending case, which
tions required by the Reissner type element formulations can be interpreted as a simple 3D patch test for geometri-
considerably increases with increasing beam slenderness cally exact beams. Finally, also the number of Newton iter-
ratio, the number of iterations remains more or less con- ations shall be investigated (see Fig. 14). In order to enable
stant for the Kirchhoff type formulations. The conclusions more general conclusions, this time, a second Reissner type
from these 2D tests will be confirmed by the 3Dl examples beam element formulation, which is based on a completely
investigated in the following sections. different triad interpolation scheme, has additionally been
included in the comparison. Concretely, this element repre-
sents an “interpretation” formulated by Crisfield [40] (see
11.3 Example 3: Pure Bending in 3D Chapter 17.2) of the original variant proposed by Simo and
Vu-Quoc [140], in the following denoted as SV element.
In this section, the 3D extension of the pure bending This time, the load step adaption scheme presented above
examples (load cases M and M̃ ) of the last section will be based on an initial number of N0 = 10 load steps has been
considered. Again, the focus lies on an initially straight, employed.
clamped beam of standard length l = 1000 investigated for
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
Similarly to the 2D case considered in the last section, parametrization of nodal triads via tangent vectors seems
the Newton performance of the Reissner type element for- to be more good-natured than the parametrization based on
mulations drastically deteriorates with increasing slender- nodal rotation vectors. This effect already gets visible as
ness ratio whereas the performance of the Kirchhoff type difference between the WK/SK-TAN and the WK/SK-ROT
elements remains unchanged (or is even slightly improved variants and seems to be more or less independent from the
in case of the WK/SK-TAN variants). Concretely, for the beam slenderness ratio. Secondly, the high stiffness contri-
slenderness ratio 𝜁 = 10000, all investigated discretizations butions resulting from the shear mode seem to considerably
of the WK/SK-TAN elements exhibit a remarkably constant deteriorate the Newton convergence in the range of high
number of niter,tot = 140 ± 4 iterations, while the total num- slenderness ratios: this effect becomes obvious as differ-
ber of iterations required by the WK/SK-ROT elements ence between the WK/SK-ROT elements and the Reissner
increases from niter,tot ≈ 500 to niter,tot ≈ 2000 with increas- type elements. The linearizations of all of these four ele-
ing number of elements and the total number of iterations ments are based on multiplicative updates of nodal rotation
required by the Reissner type formulations is almost by two vectors. This observation is emphasized by the two ele-
orders of magnitude higher than for the WK/SK-TAN ele- ments types WK-ROT and CJ, which additionally exhibit
ments and lies constantly above a value of niter,tot > 12000. the same triad interpolation. These two elements only differ
Seemingly, the considerable difference between the WK/ in the centerline interpolation (based on Lagrange or Her-
SK-TAN elements and the Reissner type element formula- mite polynomials), which is not expected to influence the
tions can be attributed to two different effects: firstly, the Newton convergence in such a drastic manner, and the fact
that the WK-ROT element additionally enforces the con-
straint of vanishing shear strains. Consequently, the avoid-
ance of shear modes seems to be the main reason for the
considerably improved performance of the Kirchhoff type
element formulations. Finally, the observation that the total
number of Newton iterations required by the WK/SK-ROT
variants increases with increasing number of elements is
only of secondary practical interest since the discretizations
Fig. 15 Initially straight beam loaded by a discrete end-moment and relevant for practical applications are located in the range
end-force: initial and deformed configuration of small element numbers (on the left of Fig. 14b). The
Fig. 16 Deformed configurations for simultaneous (sim) and succes- (sim). e Step t = 0.25 (suc). f Step t = 0.50 (suc). g Step t = 0.75
sive (suc) loading with a moment M and a force F. a Step t = 0.25 (suc). h Step t = 1.0 (suc)
(sim). b Step t = 0.50 (sim). c Step t = 0.75 (sim). d Step t = 1.0
13
C. Meier et al.
0 0
10 10
-2 Crisfield & Jelenic -2 Crisfield & Jelenic
10 WK-TAN 10 WK-TAN
-4 SK-TAN -4 SK-TAN
10 10
10 -6 10 -6
10 -8 10 -8
rel
rel
2
2
||e||
||e||
-10 -10
10 10
10 -12 10 -12
10 -14 10 -14
10 -16 10 -16
10 -18 10 -18
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
number of elements number of elements
(a) (b)
Fig. 17 Verification of path-independence: L2-error between the load cases “simultaneously” and “successively”. a Moderate slenderness ratio:
ζ = 100. b High slenderness ratio: ζ = 10,000
0
10
10 -2
10 -4
rel
2
-6
10
||e||
Fig. 18 Verification of path-independence: L2-error plotted over element length for different element formulations. a Reference: WK-TAN, ζ =
100. b Reference: WK-TAN, ζ = 10,000
observations made so far, will be confirmed by the subse- adaption scheme, an evaluation and comparison process
quent 3D examples. is intended that is as fair and objective as possible. Nev-
ertheless, this example shows that an absolute statement
Remark For some of the discretizations investigated in concerning the robustness of the nonlinear solution scheme
Fig. 14, solutions of the SK-TAN element could already based on a single example / discretization and a deliberately
be found in one load step. However, since in these cases chosen “good-natured” step size, as sometimes done in the
no convergence could be achieved for simulations based literature, is questionable. Here, the degree of arbitrariness
on two or three load steps, the solution of the problem by is intended to be minimized by employing an automated
means of one load step can rather be regarded as a “lucky scheme for determining the optimal load step size, by com-
shot” than as a representative convergence behavior. In paring the results of different test cases, different discretiza-
order to avoid a biased comparison resulting from such tions, different element types (here Reissner and Kirchhoff
effects, the initial number of load steps has been increased type beam elements) as well as different representatives
to N0 = 10. By this means and the mentioned load step for each element type. In order to avoid biased results as
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
consequence of incorrect linearizations, the results of the “suc”). In the latter case, the external moment is increased
Reissner type elements derived on the basis of an analytic linearly from zero to M within the pseudo-time interval
representation of the consistent tangent stiffness matrix t ∈ [0;0.5], whereas the external force is increased linearly
have been verified by simulations on the basis of a consist- from zero to F within the pseudo-time interval t ∈ [0.5;1.0].
ent tangent stiffness matrix derived via an automatic differ- The deformed shapes of both load cases have been plot-
entiation tool. ted for the pseudo-time steps t = 0.25, t = 0.5, t = 0.75 and
t = 1.0 in Fig. 16. Apparently, the two load cases lead to
11.4 Example 4: Verification of Path‑Independence different deformation paths, but to an identical final con-
figuration. It contradicts intuition that this final deformed
In Sect. 6.3.2, the fundamental property of objectiv- configuration lies completely in the half space with z ≤ 0,
ity has already been verified for the proposed Kirchhoff although the tip force points into the positive z-direction.
beam elements. In this section, it will be shown that For the case of small forces F, this observation has been
these element formulations are also path-independent, verified by deriving an analytical solution based on a lin-
i.e. for beam problems whose analytic solution is inde- earization of the equilibrium equations with respect to the
pendent from a specific loading path, these beam element double-circle configuration resulting from the end-moment
formulations also yield a discrete solution that is inde- M (see also [110]). Furthermore, this observation is in
pendent from a specific loading path. As numerical test agreement with the results obtained in [75] and [13], where
case for path-independence, an initially straight clamped a similar example based on a slightly modified parameter
beam with initial length l = 1000 and slenderness choice has been analyzed.
ratio 𝜁 = 100 (thus R = 10) is considered that is loaded In order to investigate possible path-dependence effects
by an end-moment 𝐌 = (0, 0, M)T , with the moment also in a quantitative manner, the relative L2-error has
M = 4EI𝜋∕l ≈ 10.47 being defined such that it exactly been calculated between the solution 𝐫h,suc of the load case
bends the beam into a “double-circle”, and an additional “suc” for a certain discretization and the solution 𝐫h,sim of
end-force 𝐅 = (0, 0, F)T , with F = 0.01 ≈ M∕l. the load case “sim” for the same centerline discretization.
Again, for comparison reasons, also the case of an Thus, basically the relative L2-error definition of equation
increased slenderness ratio 𝜁 = 10000 with correspond- (146) has been applied, with 𝐫h = 𝐫h,suc and 𝐫ref = 𝐫h,sim.
ingly adapted loads M = 4EI𝜋∕l ≈ 1.047 ⋅ 10−7 and The results obtained for the two different slenderness ratios
F = 10−10 ≈ M∕l will be investigated. The problem setup and the investigated element formulations CJ, SK-TAN and
as well as the deformed configuration for this example are WK-TAN are illustrated in Fig. 17. Accordingly, for all
shown in Fig. 15. In the following, two different possibili- investigated element types, discretizations and slenderness
ties how to apply these tip loads are investigated: in a first ratios, this error vanishes up to machine precision, which
load case, the moment and the force are applied simultane- verifies the path-independence of these formulations.
ously (load case “sim”), while, in a second load case, the For completeness, Fig. 18 represents the discretiza-
moment and the force are applied successively (load case tion error resulting from the CJ, SK-TAN and WK-TAN
8000 8000
Crisfield & Jelenic
7000 Simo & Vu Quoc 7000
WK-TAN
6000 SK-TAN 6000
WK-ROT
Newton iterations
Newton iterations
1000 1000
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
number of elements number of elements
(a) (b)
Fig. 19 Verification of path-independence—load case “simultaneously”: total number of Newton iterations. a Moderate slenderness ratio: ζ =
100. b High slenderness ratio: ζ = 10,000
13
C. Meier et al.
Table 4 Arc-segment loaded by out-of-plane force—case 𝜁̃ = 102: arises in a way such that the total torsion is represented
tip displacement and relative error for different formulations exactly, which in turn results in an inconsistent triad orien-
Formulation # Elements ux (l) uy (l) uz (l) tation. However, since for isotropic beams, only the torsion,
but not the triad orientation, enters the weak form, the final
Crisfield and Jelenić 32 47.15044 15.68480 53.47486 result for the beam centerline is correct. Later in Sect. 11.5,
Crisfield and Jelenić 8 47.15044 15.68480 53.47486 it will be shown that the situation changes for anisotropic
Simo and Vu-Quoc 8 47.14634 15.69146 53.47362 beams, i.e. beams with initial curvature or with anisotropic
WK-TAN 32 47.15215 15.68535 53.47176 cross-section shapes. There, the neglect of the intermediate
WK-TAN 8 47.15178 15.68510 53.47225 triad torsion will indeed lead to an inconsistent centerline
SK-TAN 8 47.15201 15.68557 53.47216 solution resulting in a decreased spatial convergence rate.
Furthermore, this investigation explains why certain Kirch-
hoff element formulations available in the literature, which
Table 5 Arc-segment loaded by out-of-plane force—case 𝜁̃ = 104: accidentally neglect this torsion term, nevertheless produce
tip displacement and relative error for different formulations correct results and consistent convergence rates for the
Formulation # Elements ux (l) uy (l) uz (l) centerline solution as long as isotropic beam problems are
considered.
Crisfield and Jelenić 32 47.15129 15.68508 53.46860 Finally, in Fig. 19, the total number of Newton iterations
Crisfield and Jelenić 8 47.15129 15.68508 53.46860 required by the different finite element formulations is plot-
Simo and Vu-Quoc 8 47.14719 15.69174 53.46736 ted for the load case “sim” and the two investigated slen-
WK-TAN 32 47.15129 15.68508 53.46860 derness ratios. Here, the load step adaption scheme based
WK-TAN 8 47.15093 15.68482 53.46908 on N0 = 10 has been employed. The obtained results are
SK-TAN 8 47.15115 15.68530 53.46900 very similar to the last section: the Newton performance of
the Reissner type element formulations drastically deterio-
rates with increasing slenderness ratio whereas the perfor-
elements for the considered slenderness ratios of 𝜁 = 100 mance of the Kirchhoff type elements remains unchanged.
and 𝜁 = 10000. There, the observations already made in For the slenderness ratio 𝜁 = 10000, all investigated discre-
earlier examples with respect to convergence rate, discre- tizations of the WK/SK-TAN elements exhibit a remarkably
tization error level and cutoff error between Kirchhoff constant number of niter,tot = 107 ± 1 iterations, while the
and Reissner type element formulations are confirmed. total number of iterations required by the WK/SK-ROT ele-
For comparison reasons, also the L2-error resulting from ments increases from niter,tot ≈ 800 to niter,tot ≈ 2500 with
the reduced isotropic beam element formulation proposed increasing number of elements. Again, the total number of
in [110] has been plotted. The result is very similar to the iterations required by the Reissner type beam element for-
SK-TAN element. The only reason why the discretiza- mulations is almost by two orders of magnitude higher than
tion error level is slightly lower for the isotropic than for for the WK/SK-TAN elements and lies constantly above a
the SK-TAN element lies in the twist interpolation, which value of niter,tot > 6000 for the CJ element formulation and
only requires two DoFs for the isotropic element but three niter,tot > 7000 for the SV beam element formulation.
DoFs for the SK-TAN element. In Sect. 6.2.4, the impor-
tance of a consistent torsion (115) of the intermediate triad 11.5 Example 5: Arc‑Segment with Out‑of‑Plane Load
field has been emphasized. Now, in Fig. 18b, the discretiza-
tion error for a variant (SK-TAN no KM1) has been plotted, Besides the objectivity test in Sect. 11.1, all examples
where exactly this torsion term has been neglected. Sur- investigated in the previous sections were based on iso-
prisingly, the resulting discretization error level is identi- tropic geometries, i.e. straight beams with quasi-circular
cal to the “correct” SK-TAN element formulation. In this cross-sections EI2 = EI3. Now, an initially curved beam
context, the question might arise how this contradiction can will be considered. The initial geometry is represented by
be explained. In order to answer this question, it has to be a 45◦-degree circular arc-segment with curvature radius
realized that the actual triad orientation is not important for r0 = 100 that lies completely in the global x-y-plane and
isotropic examples in order to yield a consistent centerline that is clamped at one end.
convergence. It can easily be verified that only the mechan- The section constitutive parameters of the beam result
ical torsion has to be represented correctly, which is also from a quadratic cross-section shape with side length R = 1
the functional principle of the isotropic beam element. If and a Young’s modulus of E = 107 as well as a shear mod-
the torsion of the intermediate triad field is neglected, the ulus of G = 0.5 ⋅ 107. This initial geometry is loaded by
total torsion is solely represented by the derivative of the an out-of-plane force 𝐟 = (0, 0, fz )T in global z-direction
relative angle field 𝜑(𝜉). Consequently, the relative angle with magnitude fz = 600. This example has initially been
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
13
C. Meier et al.
0
10
-2
10
10 -4
rel
2
-6
10
||e||
-8
10
Crisfield & Jelenic
WK-TAN
10 -10 SK-TAN
4 th order
-12
10
10 1 10 2 10 3 10 4
degrees of freedom
(a) (b)
Fig. 21 Plane arc-segment loaded by out-of-plane force: L2-error plotted over element length for different formulations. a Reference: WK-TAN,
𝜁̃ = 100. b Reference: WK-TAN, 𝜁̃ = 10,000
500 500
Crisfield & Jelenic
Simo & Vu Quoc
400 WK-TAN 400
SK-TAN
WK-ROT
Newton iterations
Newton iterations
0 0
0 5 10 15 0 5 10 15
number of elements number of elements
(a) (b)
(c) (d)
Fig. 22 Plane arc-segment loaded by out-of-plane force: total number of Newton iterations and load steps. a Number of Newton iterations for 𝜁̃
= 100. b Number of Newton iterations for 𝜁̃ = 10,000. c Number of load steps for 𝜁̃ = 100. d Number of load steps for 𝜁̃ = 10,000
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
0
10 10 0
10 -2 10 -2
-4
10 10 -4
rel
rel
2
-6
10 10 -6
||e||
||e||
10 -8 10 -8
Crisfield & Jelenic Crisfield & Jelenic
-10
WK-TAN -10 WK-TAN
10 SK-TAN 10 SK-TAN
4 th order 4 th order
-12 -12
10 10
1 2 3 4
10 1 10 2 10 3 10 4 10 10 10 10
degrees of freedom degrees of freedom
(a) (b)
Fig. 24 Helix with varying slope loaded with axial force: relative L2-error plotted over element length
13
C. Meier et al.
⎛ sin 𝛽 ⎞
𝐫0 (𝛽) = R0 ⎜ cos 𝛽 − 1 ⎟,
⎜ 6 2 ⎟
⎝ 81𝜋 2 𝛽 ⎠
l
� �
(187)
R0 = � � �2 � � �2
3𝜋 2
6 +1+ 27𝜋 ln 4
+ 4
1+ 3𝜋
Fig. 25 Helix with varying slope and twisted cross-section loaded
4 8 3𝜋
with axial force: initial and final shape for 𝜁 = 100 ≈ 34.36.
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
10 0 10 0
10 -2 10 -2
10
-4
10 -4
rel
rel
2
2
-6
10 -6 10
||e||
||e||
10
-8
Crisfield & Jelenic 10 -8 Crisfield & Jelenic
WK-TAN WK-TAN
SK-TAN -10
SK-TAN
10 -10 SK-TAN+CS 10 SK-TAN+CS
4 th
order 4 th order
-12 -12
10 10
1 2 3 4
10 1 10 2 10 3 10 4 10 10 10 10
degrees of freedom degrees of freedom
(a) (b)
Fig. 26 Helix with varying slope and twisted cross-section loaded with axial force: L2-error for different formulations. a Reference: WK-TAN, ζ
= 100. b Reference: WK-TAN, ζ = 10,000
Fig. 27 Free oscillations of an elbow cantilever: initial and deformed configurations at different time steps. a Step 0. b Step 5. c Step 10. d Step
15. e Step 20. f Step 25. g Step 30. h Step 35. i Step 40. j Step 45. (Color figure online)
The radius R0 of the enveloping cylinder of the helix stiffnesses of the cases 𝜁1 = 100 and 𝜁2 = 10000 leading
is chosen such that the helix exactly consists of 4.5 loops, to the comparable values uz,max,1 ≈ 267 and uz,max,2 ≈ 266
i.e. 𝛽 ∈ [0;9𝜋], along the standard length of l = 1000. for the maximal tip-displacement in z-direction. In
Also this example is investigated for two different slender- Fig. 24, the resulting relative L2-error of the CJ, SK-TAN
ness ratios 𝜁 = 100 and 𝜁 = 10000 with associated axial and WK-TAN element is plotted for discretizations with
forces F = 2 ⋅ 10−1 as well as F = 2 ⋅ 10−9. The ratio of 16, 32, 64, 128, 256 and 512 elements. All element for-
these forces is chosen identical to the ratio of the bending mulations show the expected convergence rate of four,
13
C. Meier et al.
Fig. 28 Free oscillations of an elbow cantilever: system energy plot- and Jel., 2 elements, Δt = 0:25. d WK-ROT, 2 elements, Δt = 0:25.
ted over time for different beam element formulations. a WK-ROT, e SK-ROT, 2 elements, Δt = 0:25. f SK-ROT+CS, 2 elements, Δt =
16 elements, Δt = 0:01. b SK-ROT, 16 elements, Δt = 0:01. c Cri. 0:25
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
the discretization error level of the WK-TAN element is The first segment of the cantilever is clamped at
slightly lower than the discretization error level of the SK- the position s = 0. In the original work, the cross-sec-
TAN element. However, both discretization error levels lie tion of the beam is described by the section parameters
below the error level of the Reissner type CJ element. In EA = GA2 = GA3 = 106, GIT = EI2 = EI3 = 103 as well as
the authors’ former contribution, also the different anti- 𝜌A = 1 and 𝜌IP ∕2 = 𝜌I2 = 𝜌I3 = 10. It can easily be veri-
locking approaches already investigated for the 2D case in fied that these section parameters can for example be rep-
Sect. 11.2.1 have been compared for this 3D helix example. resented by a quadratic cross-section with radius R ≈ 0.1,
The derived results and the drawn conclusion are similar as thus resulting in a slenderness ratio of 𝜁 ≈ 100 for each of
for this 2D example. For further details on this comparison, the two segments, for which the rotational inertia contribu-
the interested reader is referred to this reference. tions are additionally scaled by a factor of 104. This arti-
Finally, the helix problem shall be even more generalized ficial scaling has been applied in order to emphasize (and
by not only accounting for 3D initial curvatures but also for properly verify) the rotational inertia contributions, which
anisotropic cross-section shapes with initial twist as illus- would otherwise be comparatively small for the chosen
trated in Fig. 25. Again, the two slenderness ratios 𝜁 = 100 slenderness ratio. This cantilever beam is loaded by a dis-
and 𝜁 = 10000 have been investigated. The square cross- crete force 𝐅 = (0, 0, Fz )T pointing in global z-direction
section of the last example is extended to a rectangular and acting at the rigid corner of the elbow. The magnitude
cross-section shape with dimensions b = 10 and h = 5 and of the force is linearly increased from Fz = 0 at t = 0 to
an assumed torsional moment of inertia IT ≈ 3.2875 ⋅ 10−2 Fz = 50 at t = 1 and then linearly decreased to Fz = 0 at
for the case 𝜁 = 100. The case 𝜁 = 10000 is defined by t = 2. In the remaining simulation time until T = 50, the
b = 0.1 and h = 0.05 and an assumed torsional moment of cantilever executes geometrically nonlinear free oscilla-
inertia IT ≈ 3.2875 ⋅ 10−6 . The external forces have been tions in 3D space with amplitudes being in the range of the
chosen as F = 5 ⋅ 10−2 for 𝜁 = 100 and F = 5 ⋅ 10−10 for segment lengths. The initial and deformed configurations
𝜁 = 10000 and the initial twist as one twist rotation per at different time steps are illustrated in Fig. 27. For time
helix loop. The resulting L2-error plotted in Fig. 26 again integration, the modified generalized-𝛼 scheme of Sect. 5
shows a consistent convergence behavior similar to Fig. 24. with 𝜌∞ = 0.95 has been employed. As spatial discretiza-
Additionally, in Fig. 25, also the Bubnov–Galerkin vari- tions, the CJ element, the WK-ROT element as well as the
ant (SK-TAN+CS) of the SK-TAN element with consistent Petrov–Galerkin variant SK-ROT and the Bubnov–Galer-
spin vector interpolation has been plotted. Accordingly, no kin variant SK-ROT+CS with consistent spin vector inter-
visible difference compared to the Petrov–Galerkin (SK- polation according to (164) have been applied. This exam-
TAN) variant can be observed. In a last step, also the bal- ple yields a further complexity that has not been present in
ances of forces and moments are investigated for this most the previous test cases: the modeling of a rigid beam con-
general example of Fig. 25. In Table 7, the reaction forces nection occurring at the corner of the elbow. As already
and moments at the clamped end of the helix at s = 0 and mentioned in earlier sections, the modeling of such kinks
the force and moment contributions (with respect to the in the beam centerline is easier to be realized by the SK/
point s = 0) resulting from the external load applied at s = l WK-ROT variants. Nevertheless, the resulting solutions are
are plotted for discretizations with eight elements. It can identical as for the SK/WK-TAN variants. In Fig. 28, the
easily be verified that the balance of forces and moments internal, kinetic and total system energy is plotted for dif-
is exactly fulfilled by the variants CJ, WK-TAN and SK- ferent time step sizes and spatial discretizations. For discre-
TAN+CS, while the Petrov–Galerkin variant SK-TAN tizations with 8 WK-ROT and SK-ROT elements per elbow
only fulfills the balance of forces but not the balances of segment and a time step size of Δt = 0.01 as illustrated
moments. This confirms the prediction made in Sect. 9.4. in Fig. 28a, b, no visible oscillations in the total system
energy and no visible differences in the energy contribu-
11.7 Example 7: Free Oscillations of Elbow Cantilever tions of the Reissner and Kirchhoff type elements can be
observed. In Fig. 28c–f, the energy contributions result-
The final example represents a dynamic test case. The ing from a larger time step size of Δt = 0.25 and a rougher
example has initially been investigated in [141] and subse- spatial discretization based on one beam element per elbow
quently been considered in several contributions in the field segment has been plotted. Accordingly, also for these rough
of geometrically nonlinear beam element formulations (see discretizations, the overall system energy is preserved very
e.g. [32, 83]). A right-angle elbow cantilever beam consists well. The fact that the total system energy approximation
of two straight beam segments of length l = 10 being rig- resulting from the (energetically consistent) Bubnov–Galer-
idly connected at one of their ends. In the initial configura- kin discretization SK-ROT+CS is rarely improved as com-
tion, the first segment points into global y-direction and the pared to the Petrov–Galerkin variants CJ, WK-ROT and
second segment into global x-direction. SK-ROT indicates that for the chosen spatial and temporal
13
C. Meier et al.
discretizations, the influence of the temporal discretization has employed the well-known orthonormal, geodesic triad
error might dominate the error in the total system energy. interpolation scheme proposed in [43]. Furthermore, for
As compared to reference [83], where an identical time step each of these two interpolation schemes, two different sets
size of Δt = 0.25 and a comparable spatial discretization of nodal rotation parametrizations are proposed, one based
consisting of one second-order CJ element per elbow seg- on nodal rotation vectors (ROT) and one on nodal tangent
ment has been applied, the oscillations of the total system vectors (TAN). While these different choices have been
energy could be considerably decreased and the instability shown to yield identical FEM solutions, they differ in the
observed there at the end of the considered time interval resulting performance of nonlinear solvers and in the effort
could be completely avoided. This improvement can be required for prescribing essential boundary and transition
attributed to the applied Lie group extension of the general- conditions such as rigid joints. The four finite element for-
ized-𝛼 scheme of Sect. 5, whereas in reference [83] the Lie mulations resulting from a combination of the two interpo-
group extension of a Newmark time integration scheme as lation schemes and the two choices of nodal primary vari-
proposed in [140] has been considered. Based on (simpli- ables have been denoted as SK-TAN, SK-ROT, WK-TAN
fied) Reissner type beam element formulation (see [101]), and WK-ROT elements, respectively. Taking advantage of
similar observations have already been made in refer- a third-order Hermite interpolation, both element formu-
ence [32], where the Lie group extension of the general- lations provide a C1-continuous centerline representation.
ized-𝛼 scheme has been proposed. In order to avoid membrane locking effects in the regime
of high beam slenderness ratios, the concept of Minimally
Constrained Strains (MCS) recently proposed in [110] has
12 Conclusion been employed for re-interpolation of the axial tension
field. Eventually, the proposed beam elements are supple-
The current work focused on geometrically exact finite ele- mented by an implicit, second-order accurate time integra-
ments for highly slender beams. It aimed at the proposal tion scheme recently proposed in the literature for time dis-
of novel formulations of Kirchhoff–Love type, a detailed cretization of large rotations. This integration scheme can
review of existing formulations of Kirchhoff–Love and be identified as Lie group extension of the well-known gen-
Simo–Reissner type and a careful evaluation and com- eralized-𝛼 method with comparable properties. The gener-
parison of the proposed and existing formulations. In the ality and flexibility of this scheme allow for a straightfor-
authors’ recent contribution [109], the first 3D large defor- ward combination with the different element formulations
mation geometrically exact Kirchhoff–Love beam element considered in this work.
formulation that fulfills the essential properties of objec- The review of existing geometrically exact Kirch-
tivity and path-independence and that is capable of repre- hoff–Love beam elements revealed that there are only a
senting arbitrary initial curvatures and anisotropic cross- few approaches of this kind that are suitable for general 3D,
section shapes has been proposed. The developed beam large deformation problems. These approaches have been
element formulation incorporates the modes of axial ten- categorized in isotropic, straight and anisotropic Kirch-
sion, torsion and anisotropic bending. In the subsequent hoff–Love formulations. A detailed evaluation of these
work [110], also the important question of membrane lock- formulations led to the result that typically only a few of
ing has successfully been addressed. The current contribu- the essential requirements summarized in Table 2 are ful-
tion has extended these methodologies by providing con- fillled. On the contrary, for the finite element formulations
siderable improvements in terms of accuracy and practical proposed in this work, the fulfillment of these essential
applicability as well as a generalization to dynamic prob- properties such as objectivity and path-independence, con-
lems. Thereto, two alternative interpolation schemes have sistent spatial convergence behavior, the avoidance of lock-
been proposed: (i) The first scheme is based on a strong ing in the high slenderness regime or the conservation of
enforcement of the Kirchhoff constraint (SK) enabled by a energy and momentum by the spatial discretization scheme
novel orthonormal rotation interpolation scheme. (ii) The have been shown theoretically and verified by means of
space-continuous theory of the second scheme is based on representative numerical examples. In the context of con-
a weak enforcement of the Kirchhoff constraint (WK). The servation properties, especially, the influence of applying
discrete realization of the Kirchhoff constraint relies on a either Bubnov–Galerkin or Petrov–Galerkin discretiza-
properly chosen collocation strategy which can entirely tions has been in the focus. Concerning locking behavior,
abstain from the use of additional Lagrange multipliers. the recently proposed MCS method has been compared
While this second formulation allows for arbitrary rota- with alternative methods known from the literature such
tion interpolations, the investigated numerical realization as Assumed Natural Strains (ANS) or Reduced Integration
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
(RI) schemes (see also [110]). In contrast to these alterna- degree k = 3 of the employed trial functions and predicted
tive methods, the MCS method could effectively avoid any to vanish for higher-order elements with k > 3. This pre-
evidence of membrane locking for all investigated load diction has been confirmed by means of a first numerical
cases and slenderness ratios. In contrast to previously exist- test case employing Hermite polynomials of order k = 5
ing Kirchhoff–Love formulations, Simo–Reissner formula- which resulted in the expected optimal discretization error
tions fulfilling the requirements of Table 2 can be found in level of the SK elements lying below the error level of the
the literature very well. However, it has been argued that Simo–Reissner reference formulation. While most of the
the shear-free Kirchhoff–Love formulations can provide investigated examples have been of quasi-static nature, also
considerable numerical advantages such as lower spatial one dynamic benchmark test from the literature has been
discretization error level, improved performance of time conducted. There, the accuracy of the inertia contributions
integration schemes as well as linear and nonlinear solvers as well as the energy stability of the employed time integra-
or smooth geometry representation as compared to shear- tion scheme could be confirmed.
deformable Simo–Reissner formulations when applied to Besides the resulting discretization error level, also
highly slender beams. On the basis of several numerical the total number of Newton–Raphson iterations required
examples, detailed and systematic numerical comparisons to solve the considered test cases by means of the differ-
of the resulting discretization error levels and performance ent element formulations and for different slenderness
of the nonlinear solver have been performed between the ratios has been analyzed in a systematic manner. For all
four variants of the proposed geometrically exact Kirch- investigated examples and slenderness ratios, the proposed
hoff–Love beam elements and two well-established geo- SK-TAN and WK-TAN elements required less Newton
metrically exact Simo–Reissner beam element formula- iterations to solve the problem as compared to the two well-
tions known from the literature. Most of the examples established Simo–Reissner formulations chosen as refer-
have been investigated for the two different slenderness ence. In the small slenderness range 𝜁 = 100, the results
ratios 𝜁 = 100 and 𝜁 = 10000. For the low slenderness of the four proposed Kirchhoff–Love variants and the two
ratio 𝜁 = 100, the general model difference between the investigated Simo–Reissner formulations lay at least in the
shear-free Kirchhoff–Love theory and the Simo–Reiss- same order of magnitude. While the behavior of the Kirch-
ner theory of shear-deformable beams, measured in form hoff–Love formulations remained more or less unchanged,
of the relative L2-error remaining in the limit of arbitrar- the number of Newton iterations required by the two dif-
ily fine spatial discretizations, typically lay below 0.1%. ferent Simo–Reissner formulations increased considerably
Also the quadratic decrease of this model difference with with increasing slenderness ratio. In the investigated exam-
increasing slenderness ratio could be confirmed numeri- ples with slenderness ratio 𝜁 = 10000, this number was up
cally. In all investigated examples, the proposed WK ele- to two orders of magnitude higher for the Simo–Reissner
ments have shown a lower discretization error level than elements as compared to the proposed SK-TAN and WK-
the investigated Simo–Reissner beam element formula- TAN elements. Also the number of iterations required by
tion. These results confirm the theoretical prediction that the SK-ROT and WK-ROT elements, which are based on
Kirchhoff type formulations can achieve the same discre- nodal rotation vectors for triad parametrization, has been
tization error level as Reissner type formulations with less independent from the considered slenderness ratio, higher
degrees of freedom, since no shear deformation has to be as for the SK-TAN and WK-TAN elements but still consid-
represented. Compared to the excellent results of the WK erably lower as for the Reissner type elements. In summary,
elements, the SK elements showed an increased discretiza- out of the four proposed Kirchhoff variants, the WK-TAN
tion error level in the L2-norm of the displacement error element, based on a weak enforcement of the Kirchhoff
but also in the error of the strain energy. This observation constraint and a triad parametrization via nodal tangent
might be traced back to the fact that the WK elements can vectors, can be recommended in terms of a low discretiza-
exactly represent the strain energies associated with the tion error level and an excellent performance of the New-
states of spatially constant deformation measures, i.e. con- ton–Raphson scheme. Of course, further factors could be
stant axial tension, torsion and bending along the beam ele- considered in a comprehensive comparison. For example,
ment, while the SK element cannot. In contrast to the WK the SK-ROT and WK-ROT elements based on nodal rota-
elements, the L2-error level resulting from the SK elements tion vectors simplify the prescription of Dirichlet condi-
was in some examples even higher as for the Simo–Reiss- tions. The flexibility of the proposed beam element vari-
ner elements. Based on the underlying convergence theory, ants allows to combine the advantages of the two different
this phenomenon could be attributed to the polynomial rotation parametrizations by choosing e.g. the WK-TAN
13
C. Meier et al.
element as basic formulation that provides an excellent ̃𝐈i� (𝜉) = Li� (𝜉)𝚲r 𝐓−1 (𝚽lh (𝜉))𝐓(𝚽i )𝚲T
l r
Newton Raphson performance and by replacing the nodal + Li (𝜉)𝚲r 𝐓−1 (𝚽lh (𝜉))𝐓(𝚽il )𝚲Tr
tangents by nodal rotation vectors at nodes where complex � ,s �n �
∑Λ j j
boundary or coupling conditions have to be prescribed. iI −1
− 𝛿 𝚲r 𝐓,s (𝚽lh (𝜉)) L (𝜉)𝐓(𝚽l )
This can be realized by a simple transformation applied to �n j=1
��
∑Λ j�
the residual and stiffness contributions of the relevant node. + 𝐓−1 (𝚽lh (𝜉)) L (𝜉)𝐓(𝚽l )
j
𝐯I 𝚲Tr (191)
The abstaining from stiff shear mode contributions under- � j=1
�n �
lying the proposed Kirchhoff–Love element formulations iJ −1 ∑Λ j j
− 𝛿 𝚲r 𝐓,s (𝚽lh (𝜉)) L (𝜉)𝐓(𝚽l )
may not only yield an improved Newton–Raphson perfor- �n j=1
��
mance. Also the highest eigenfrequency band of slender ∑ Λ
j
−1
+ 𝐓 (𝚽lh (𝜉)) j�
L (𝜉)𝐓(𝚽l ) 𝐯J 𝚲Tr .
beams, which is associated with the shear modes, can be j=1
avoided by this means. The theoretical considerations made
in this work give hope for considerably improved stabil- Finally, the required arc-length derivative 𝐓−1
,s (𝚽lh (𝜉))
is
ity properties of numerical time integration schemes when given by (see also [40, 82]):
combined with the developed shear-free elements. A future
Φlh sin Φlh − 2(1 − cos Φlh )
numerical investigation of this topic seems to provide a 𝐓−1 T �
,s (𝚽lh (𝜉)) = 𝚽lh 𝚽lh 𝐒(𝚽lh )
considerable scientific potential. Φ4lh
1 − cos Φlh
+ 𝐒(𝚽�lh )
Compliance with Ethical Standards Φ2lh
( )
1 sin Φlh ( )
Conflict of interest On behalf of all authors, the corresponding au- + 2 1− 𝐒(𝚽lh )𝐒(𝚽�lh ) + 𝐒(𝚽�lh )𝐒(𝚽lh )
Φlh Φlh
thor states that there is no conflict of interest.
3 sin Φlh − Φlh (2 + cos Φlh )
+ 𝚽Tlh 𝚽�lh 𝐒(𝚽lh )𝐒(𝚽lh ).
Φ5lh
(192)
Appendix A: Rotational Shape Function Matrices Here, the abbreviations 𝚽lh = 𝚽lh (𝜉) and Φlh = ||𝚽lh (𝜉)||
have been applied. The limit 𝐓−1 �
,s (𝚽lh (𝜉)) → 0.5𝐒(𝚽lh (𝜉))
In this appendix, the shape functions ̃𝐈i (𝜉) required for the can be derived for small angles 𝚽lh (𝜉) → 𝟎 (see [83]).
multiplicative rotation increments
∑
nΛ
∑
nΛ
Appendix B: Dirichlet Boundary Conditions
Δ𝜽(𝜉) = �𝐈i (𝜉)Δ𝜽̂ i , �
Δ𝜽 (𝜉) = �𝐈i� (𝜉)Δ𝜽̂ i . (188)
i=1 i=1
and Joints
associated with the triad interpolation (105) and originally For many applications, the formulation of proper Dirichlet
derived in [83] shall be presented: boundary conditions or joints between the nodes of differ-
ent beam elements are of a high practical relevance. This
̃𝐈i (𝜉) = Li (𝜉)𝚲r 𝐓−1 (𝚽lh (𝜉))𝐓(𝚽i )𝚲T appendix represents a brief summary, where the possibility of
� l � r ��
∑
nΛ
j formulating some basic constraint conditions will be investi-
iI
+ 𝛿 𝚲r 𝐈𝟑 − 𝐓 (𝚽lh (𝜉)) −1
L 𝐯I 𝚲Tr
j
(𝜉)𝐓(𝚽l )
gated for the SK-ROT and the SK-TAN element.
� �j=1
��
∑
nΛ
j
iJ −1
+ 𝛿 𝚲r 𝐈𝟑 − 𝐓 (𝚽lh (𝜉)) j
L (𝜉)𝐓(𝚽l ) 𝐯J 𝚲Tr . Appendix B.1: SK‑ROT Element
j=1
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
13
C. Meier et al.
rotated coordinate system. Again, the first component T̂ 1 of and of the global tangent stiffness matrix properly and to
the tangent vector, when expressed in the material frame, eliminate the degrees of freedom (𝐭̂ b , 𝜑̂ b ) from the global
represents its magnitude and must not be prescribed. If the system of equations. Again, the magnitude of the tangent
Dirichlet conditions are time-dependent, the prescribed vector ̂tb is not influenced by the rigid joint and enters the
evolution of the relative angle has to be adapted, since now system of equations as new degree of freedom. While in the
the intermediate frame 𝚲aM might change in time: last section, the motion of the rigid joint was completely
determined by the set (𝐝̂a , 𝝍̂ a , ̂ta , ̂tb ), in this section the
𝜑̂
exp(𝐒[𝜑̂ an+1 𝐠a1,n+1 ]) = 𝚲an+1 𝚲aT alternative set (𝐝̂a , 𝐭̂ a , 𝜑̂ a , ̂tb ) is employed.
All in all, it can be concluded that the realization of
M ,n+1 𝜑̂
(203)
with 𝚲aM = sr(𝚲aM ,n , 𝐠a1,n+1 ). clamped ends with arbitrary orientation or of rigid joints
𝜑̂ ,n+1 𝜑̂
between beams is simpler for the SK-ROT formulation based
Thus, the required value 𝜑̂ an+1 has to be determined based on nodal rotation vectors. While for these elements such
on the prescribed current triad 𝚲an+1 and the intermediate conditions can directly be formulated in the global coordi-
triad 𝚲aM ,n of the last step (see Sect. 6.2.2). The remaining nate system, the tangent vector-based SK-TAN formulation
𝜑̂
requires an additional transformation of the corresponding
conditions remain unchanged as compared to (202).
lines and columns of the global residual vector and stiffness
Connections and joints Based on (197), (198) and (25),
matrix. In Sect. 11, some properties of the tangent vector-
the relations between (𝐭̂ a , 𝜑̂ a ) and (𝐭̂ b , 𝜑̂ b ) can be stated:
based variant will become apparent which make this type
𝛿 𝐭̂ b = −tb 𝐒(𝐠b1 ) 𝛿𝜽b +𝐠b1 𝛿tb of formulation favorable for many applications. If certain
���
element nodes require Dirichlet conditions of the type con-
( =𝛿𝜽
̇ a )
̂ a + 𝐠b 𝛿 ̂tb , sidered here, it is still possible to apply a hybrid approach,
= −tb 𝐒(𝐠1 ) ta 𝐒(𝐠a1 )𝛿 𝐭̂ a + 𝐠a1 𝛿 Θ
b 1
(204)
( 1 1
) and to replace the nodal tangents by nodal rotation vectors as
̂b
𝛿Θ bT b bT 1 a ̂a
= 𝐠1 𝛿𝜽 = 𝐠1 ta 𝐒(𝐠1 )𝛿 𝐭 + 𝐠1 𝛿 Θ1 . a ̂a
primary variables only at the specific nodes where such con-
1 ���
̇ a
=𝛿𝜽
ditions are required. All the results derived so far apply in a
similar manner to the WK-TAN and WK-ROT elements.
Combining these two relations eventually yields the follow- Appendix C: Linearization of SK‑TAN Element
ing total transformation matrix:
� � ⎛ 𝛿�𝐭 a ⎞ Before deriving the linearization of the SK-TAN element,
𝛿�𝐭 b some former definitions are repeated:
̂b = 𝐓RC ⎜ 𝛿 Θ
̂ a ⎟,
𝛿Θ ⎜ b1 ⎟
1 ⎝ 𝛿 ̂t ⎠ 𝐫� 𝐫�
� � (205) 𝐭: = 𝐫 � , 𝐠1 : = ||𝐫 � ||
, ̃𝐭: =
||𝐫 � ||2
,
−tb 𝐒(𝐠b1 ) t1a 𝐒(𝐠a1 ) −tb 𝐒(𝐠b1 )𝐠a1 𝐠b1 𝐫 �� (𝐫 �T 𝐫 �� )𝐫 � 𝐫 �� �T �� � (207)
𝐓RC : = . 𝐠�1 = ||𝐫 � ||
− ||𝐫 � ||3
, ̃𝐭 � =
||𝐫 � ||2
− 2(𝐫||𝐫 �𝐫||4)𝐫 .
1
𝐠bT
1 ta
𝐒(𝐠a1 ) 𝐠bT 𝐠a
1 1
0
These quantities will be required for later derivations. Lin-
earization of (207) yields:
A similar relation can also be formulated for the iterative
increments. Since the multiplicative rotation increment 1 ( )
Δ𝐠1 = 𝐈3 − 𝐠1 ⊗ 𝐠T1 𝖧� Δ� 𝖽,
components ΔΘ1 = 𝐓ΘM1 𝐭 Δ𝐭 + Δ𝜑 [see (20)] have to be ||𝐫 ||
�
pared to (205): (𝐫 �T 𝐫 �� ) ( )
Δ𝐠�1 = − � 3 𝐈3 − 𝐠1 ⊗ 𝐠T1 𝖧� Δ� 𝖽
||𝐫 ||
� � ⎛ Δ�𝐭 a ⎞ 1 ( ) �
Δ�𝐭 b � RC2 ⎜ Δ𝜑̂ a ⎟, − � 𝐠�1 ⊗ 𝐠T1 + 𝐠1 ⊗ 𝐠�T �
� RC1 𝐓RC 𝐓
=𝐓 1 𝖧 Δ𝖽
Δ𝜑̂ b ⎜ b1 ⎟ ||𝐫 ||
⎝ Δ̂t ⎠ 1 ( ) (208)
(206) + � 𝐈3 − 𝐠1 ⊗ 𝐠T1 𝖧�� Δ� 𝖽,
� � ⎛ 𝐈3 𝟎 𝟎 ⎞ ||𝐫 ||
𝐈3 𝟎 � RC2 : = ⎜ 𝐓Θ 𝐭 1 0 ⎟.
� RC1 : =
𝐓 ,𝐓 2(𝐫 �T 𝐫 �� ) ( ) �
−𝐓ΘM1 𝐭 1 ⎜ TM1 ⎟ Δ�𝐭 � = − 𝐈 − 2𝐠 ⊗ 𝐠 T
𝖧 Δ� 𝖽
⎝ 𝟎 0 1⎠ ||𝐫 � ||4 3 1 1
2 ( ) �
�
− � 2 𝐠�1 ⊗ 𝐠T1 + 𝐠1 ⊗ 𝐠�T 1 𝖧 Δ
𝖽
||𝐫 ||
Equations (205) and (206) allow to transform the corre- 1 ( )
sponding lines and columns of the global residual vector + � 2 𝐈3 − 2𝐠1 ⊗ 𝐠T1 𝖧�� Δ� 𝖽.
||𝐫 ||
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
In the following, the linearization of the SK-TAN element Δ𝜽� = 𝗏�T � 1 + (𝗏�T + 𝗏�T )Δ�
Δ𝚯 𝖽,
𝜃 𝜃 𝜃
based on the residual vector (165) will be derived. The lin- ∥Θ ⊥ ∥d
1
( ) 𝗏�𝜃 = −𝖧��T 𝐒(�𝐭) − 𝖧�T 𝐒(𝐭�� ),
⊥
(n )
∫
Δ𝗋𝐝� = Δ𝗏�𝜃 𝐦 + 𝗏�𝜃 Δ𝐦 + Δ𝗏𝜖 F̄ 1 + 𝗏𝜖 ΔF̄ 1 Jd𝜉 ∑Λ
⊥ ⊥
−1 𝗏�𝜃 = i� �
L 𝗏1 (𝜉i ) − 𝗏1 (𝜉) ⊗ 𝐠T1
∥d
i=1
1
( ) (n )
∫
− T
� 𝜌 + 𝗏𝜃⊥ Δ𝐦𝜌 Jd𝜉 − [Δ𝗏𝜃⊥ 𝐦𝜎 ]𝛤𝜎
𝖧 Δ𝐟𝜌 + Δ𝗏𝜃⊥ 𝐦 ∑ Λ
(213)
+ Li 𝗏1 (𝜉i ) − 𝗏1 (𝜉) ⊗ 𝐠�T
1
,
−1 i=1
1 ( ) ( )
𝖧�T (𝜉) 𝐠I1 × 𝐭�� (𝜉) + 𝖧��T (𝜉) 𝐠I1 × �𝐭(𝜉)
∫
Δ𝗋𝚯̂ 1 = (Δ𝗏�𝜃 𝐦 + 𝗏�𝜃 Δ𝐦 �𝜌
− Δ𝗏𝜃∥Θ 𝐦
∥Θ ∥Θ 𝗏�1 (𝜉) =
−1 1 + 𝐠T1 (𝜉)𝐠I1
( ) ( )
− 𝗏𝜃∥Θ Δ𝐦𝜌 )Jd𝜉 − [Δ𝗏𝜃∥Θ 𝐦𝜎 ]𝛤𝜎 .
𝖧�T (𝜉I ) 𝐠�1 (𝜉) × �𝐭 I + 𝐠�T1
(𝜉)𝐠I1 𝐯1 (𝜉)
(209) − ,
In order to identify the element stiffness matrix 1 + 𝐠T1 (𝜉)𝐠I1
𝗄SK−TAN , (209) has to be brought in the form
with ̃𝐭 � = 𝐫 �� ∕||𝐫 � ||2 − 2(𝐫 �T 𝐫 �� )𝐫 � ∕||𝐫 � ||4. The remaining
Δ𝗋SK−TAN = :𝗄SK−TAN Δ̂𝗑TAN . (210) linearizations in (211) have already been derived in (208).
The vector Δ̂𝗑TAN has already been defined in Sect. 9.1. The In contrast to the spin vector field 𝛿𝜽, the increment field
moment-related terms yield: Δ𝜽 has to be expressed via additive increments Δ𝜑̂ i. The
required relation is given by (129) and repeated here:
Δ𝗏�𝜃 𝐦 = 𝖧��T 𝐒(𝐦)Δ�𝐭 + 𝖧�T 𝐒(𝐦)Δ�𝐭 � ,
⊥
Δ𝐦 = −𝐒(𝐦)Δ𝜽 + 𝐜m Δ𝜽� , ̂ 1, Θ
� 1 = (Θ
Δ𝚯 ̂ 2, Θ
̂ 3 )T ,
1 1 1
iT
Δ𝗏𝜃⊥ 𝐦 � 𝜌 )Δ�𝐭,
� 𝜌 = 𝖧�T 𝐒(𝐦 𝐠1 𝐒(𝐠i1 ) 𝖧� (𝜉 i )Δ�𝖽 (214)
̂i = −
ΔΘ + Δ𝜑̂ i .
Δ𝗏𝜃⊥ 𝐦𝜎 = 𝖧 𝐒(𝐦𝜎 )Δ�𝐭, �T
(211)
1
1+𝐠 𝐠iT i ||𝐭 i ||
1 1
�𝜌 =
Δ𝗏𝜃∥Θ 𝐦 (𝖫T∥ ⊗ 𝐦
� T𝜌 )Δ𝐠1 ,
Δ𝗏𝜃∥Θ 𝐦𝜎 = (𝖫T∥ ⊗ 𝐦T𝜎 )Δ𝐠1 , The linearization of the element residual terms associated
with axial tension results in:
Δ𝗏�𝜃 𝐦 = (𝖫�T
∥
⊗ 𝐦T )Δ𝐠1 + (𝖫T∥ ⊗ 𝐦T )Δ𝐠�1 .
∥Θ
13
C. Meier et al.
∫
� �
Δ𝗋𝚯̂ 1 = �𝜌
(Δ𝗏𝜃 𝐦 + 𝗏𝜃 Δ𝐦 − Δ𝗏𝜃∥Θ 𝐦 In a similar manner, the set of iterative nodal primary vari-
∥Θ ∥Θ
−1 able increments have been defined as:
− 𝗏𝜃∥Θ Δ𝐦𝜌 )Jd𝜉 − [Δ𝗏𝜃∥Θ 𝐦𝜎 ]𝛤𝜎 . 𝗑TAN : = (Δ𝐝�1T , Δ�𝐭 1T , Δ𝜑̂ 1 , Δ𝐝�2T , Δ�𝐭 2T , Δ𝜑̂ 2 , Δ𝜑̂ 3 )T ,
(219) Δ�
1T 2T (226)
In order to identify the element stiffness matrix Δ�𝗑ROT : = (Δ𝐝�1T , Δ𝜽� , Δ̂t1 , Δ𝐝�2T , Δ𝜽 � , Δ̂t2 , Δ𝜑̂ 3 )T .
𝗄WK−TAN , (219) has to be brought in the form
The transformations between these primary variable varia-
Δ𝗋WK−TAN = :𝗄WK−TAN Δ̂𝗑TAN . (220) tions and increments is given by:
The linearization of vectors of the form 𝗏̄ ... and 𝗏̄ ′... as origi-
nally defined in (184) yields:
�𝗑̂ 𝛿 𝐱̂ ROT
𝛿 𝗑̂ TAN = 𝖳 and Δ̂𝗑TAN = 𝖳M𝗑̂ Δ𝐱̂ ROT . (227)
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
𝐫 T �2 , 𝐫 T �2 , rTAN,Θ̂ 2 , rTAN,Θ̂ 3 )T ,
TAN,𝐝 TAN,𝐭 1 1
These two different matrices are required, since the pri- (236)
𝗋ROT : = (𝗋T , 𝗋T , rROT,̂t1 ,
ROT,𝐝�1 ROT,𝜃�1
mary variable variations of the SK/WK-TAN elements are
based on the multiplicative quantities 𝛿 Θ
̂ i , whereas the cor- 𝗋T , 𝗋T , rROT,̂t2 , rROT,Θ̂ 3 )T .
1 ROT,𝐝�2 ROT,𝜃�2 1
13
C. Meier et al.
16. Bauer AM, Breitenberger M, Philipp B, Wüchner R, Bletz- 39. Crisfield MA (1990) A consistent co-rotational formulation for
inger K-U (2016) Nonlinear isogeometric spatial Bernoulli non-linear, three-dimensional, beam-elements. Comput Meth-
beam. Comput Methods Appl Mech Eng 303:101–127 ods Appl Mech Eng 81(2):131–150
17. Belytschko T, Hseih BJ (1973) Nonlinear transient finite ele- 40. Crisfield MA (1997) Non-linear finite element analysis of solids
ment analysis with convected coordinates. Int J Numer Meth- and structures: advanced topics. Wiley, New York
ods Eng 7:255–271 41. Crisfield MA (2003) Non-linear finite element analysis of solids
18. Belytschko T, Lawrence WG (1979) Applications of higher and structures: essentials. Wiley, New York
order corotational stretch theories to nonlinear finite element 42. Crisfield MA, Galvanetto U, Jelenić G (1997) Dynamics of 3-D
analysis. Comput Struc 10(1):175–182 co-rotational beams. Comput Mech 20(6):507–519
19. Bergou M, Wardetzky M, Robinson S, Audoly B, Grinspun E 43. Crisfield MA, Jeleni G (1999) Objectivity of strain measures
(2008) Discrete elastic rods. ACM Trans Graph 27(3):1–63 in the geometrically exact three-dimensional beam theory
20. Bertails F, Audoly B, Cani M-P, Querleux B, Leroy F, and its finite-element implementation. Proc R Soc London
Lévêque J-L (2006) Super-helices for predicting the dynamics 455(1983):1125–1147
of natural hair. ACM Trans Graph 25(3):1180–1187 44. Crivelli LA, Felippa CA (1993) A three-dimensional non-linear
21. Betsch P, Steinmann P (2001) Constrained integration of Timoshenko beam based on the core-congruential formulation.
rigid body dynamics. Comput Methods Appl Mech Eng Int J Numer Methods Eng 36(21):3647–3673
191(35):467–488 45. Cyron CJ, Wall WA (2012) Numerical method for the simula-
22. Betsch P, Steinmann P (2002) Frame-indifferent beam finite tion of the Brownian dynamics of rod-like microstructures with
elements based upon the geometrically exact beam theory. Int three-dimensional nonlinear beam elements. Int J Numer Meth-
J Numer Methods Eng 54(12):1775–1788 ods Eng 90(8):955–987
23. Bishop RL (1975) There is more than one way to frame a 46. Demoures F, Gay-Balmaz F, Leyendecker S, Ober-Blöbaum S,
curve. Am Math Month 82(3):246–251 Ratiu TS, Weinand Y (2015) Discrete variational Lie group for-
24. Borri M, Bottasso C (1994) An intrinsic beam model based mulation of geometrically exact beam dynamics. Numer Math
on a helicoidal approximation—Part I: formulation. Int J 130(1):73–123
Numer Methods Eng 37(13):2267–2289 47. Dill EH (1992) Kirchhoff’s theory of rods. Arch Hist Exact Sci
25. Borri M, Bottasso C (1994) An intrinsic beam model based 44(1):1–23
on a helicoidal approximation—Part II: linearization and 48. Dukić EP, Jelenić G, Gaćeša M (2014) Configuration-depend-
finite element implementation. Int J Numer Methods Eng ent interpolation in higher-order 2D beam finite elements. Finite
37(13):2291–2309 Elem Anal Design 78:47–61
26. Bottasso CL, Borri M (1998) Integrating finite rotations. 49. Durville D (2010) Simulation of the mechanical behav-
Comput Methods Appl Mech Eng 164(34):307–331 iour of woven fabrics at the scale of fibers. Int J Mater Form
27. Boyer F, De Nayer G, Leroyer A, Visonneau M (2011) Geo- 3(2):1241–1251
metrically exact Kirchhoff beam theory: application to cable 50. Dvorkin EN, Oñate E, Oliver J (1988) On a non-linear formu-
dynamics. J Comput Nonlinear Dynam 6:1–14 lation for curved Timoshenko beam elements considering large
28. Boyer F, Primault D (2004) Finite element of slender beams displacement/rotation increments. Int J Numer Methods Eng
in finite transformations: a geometrically exact approach. Int J 26(7):1597–1613
Numer Methods Eng 59(5):669–702 51. Eugster S (2015) Geometric continuum mechanics and induced
29. Brezzi F (1974) On the existence, uniqueness and approxima- beam theories. Springer, New York
tion of saddle-point problems arising from lagrangian multi- 52. Eugster SR, Hesch C, Betsch P, Glocker Ch (2014) Director-
pliers. ESAIM 8(R2):129–151 based beam finite elements relying on the geometrically exact
30. Brezzi F, Fortin M (1991) Mixed and hybrid finite elements. beam theory formulated in skew coordinates. Int J Numer
Springer, New York Methods Eng 97(2):111–129
31. Brüls O, Cardona A (2010) On the use of Lie group time inte- 53. Fan WW, Zhu WD (2016) An accurate singularity-free for-
grators in multibody dynamics. J Comput Nonlinear Dynam mulation of a three-dimensional curved Euler–Bernoulli beam
5:031002 for flexible multibody dynamic analysis. J Vibrat Acoust
32. Brüls O, Cardona A, Arnold M (2012) Lie group general- 138(5):051001
ized-𝛼 time integration of constrained flexible multibody sys- 54. Felippa CA, Haugen B (2005) A unified formulation of small-
tems. Mech Mach Theor 48:121–137 strain corotational finite elements: I. Theory. Comput Methods
33. Cannarozzi M, Molari L (2008) A mixed stress model for lin- Appl Mech Eng 194(21–24):2285–2335
ear elastodynamics of arbitrarily curved beams. Int J Numer 55. Fried I (1973) Shape functions and the accuracy of arch finite
Methods Eng 74(1):116–137 elements. Am Inst Aeronaut Astronaut J 11:287–291
34. Cardona A, Géradin M (1988) A beam finite element non- 56. Frischkorn J, Reese S (2013) A solid-beam finite element and
linear theory with finite rotations. Int J Numer Methods Eng non-linear constitutive modelling. Comput Methods Appl Mech
26(11):2403–2438 Eng 265:195–212
35. Cardona A, Géradin M, Doan DB (1991) Rigid and flexible 57. Gadot B, Martinez OR, du Roscoat SR, Bouvard D, Rodney
joint modelling in multibody dynamics using finite elements. D, Orgéas L (2015) Entangled single-wire NiTi material: a
Comput Methods Appl Mech Eng 89(1):395–418 porous metal with tunable superelastic and shape memory
36. Choit JK, Lim JK (1995) General curved beam elements based properties. Acta Mater 96:311–323
on the assumed strain fields. Comput Struc 55(3):379–386 58. Géradin M, Cardona A (1989) Kinematics and dynamics of
37. Chung J, Hulbert GM (1993) A time integration algorithm for rigid and flexible mechanisms using finite elements and qua-
structural dynamics with improved numerical dissipation: the ternion algebra. Comput Mech 4(2):115–135
generalized-𝛼 method. J Appl Mech 60:371–375 59. Géradin M, Cardona A (2001) Flexible multibody dynamics:
38. Cosserat E, Cosserat F (1909) Théorie des Corps Déformables, a finite element approach. Wiley, New York
2nd edn. Traité de Physique, Paris 60. Gerstmayr J, Shabana AA (2006) Analysis of thin beams and
cables using the absolute nodal co-ordinate formulation. Non-
linear Dynam 45(1):109–130
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
61. Gerstmayr J, Sugiyama H, Mikkola A (2013) Review on the 82. Jelenić G, Crisfield MA (1998) Interpolation of rotational vari-
absolute nodal coordinate formulation for large deformation ables in nonlinear dynamics of 3D beams. Int J Numer Methods
analysis of multibody systems. J Comput Nonlinear Dynam Eng 43(7):1193–1222
8(3):1–12 83. Jelenić G, Crisfield MA (1999) Geometrically exact 3D beam
62. Ghosh S, Roy D (2008) Consistent quaternion interpola- theory: implementation of a strain-invariant finite element
tion for objective finite element approximation of geo- for statics and dynamics. Comput Methods Appl Mech Eng
metrically exact beam. Comput Methods Appl Mech Eng 171(1–2):141–171
198(3–4):555–571 84. Jelenić G, Saje M (1995) A kinematically exact space finite
63. Ghosh S, Roy D (2009) A frame-invariant scheme for the geo- strain beam model: finite element formulation by general-
metrically exact beam using rotation vector parametrization. ized virtual work principle. Comput Methods Appl Mech Eng
Comput Mech 44(1):103–118 120(12):131–161
64. Gonzalez O (1996) Time integration and discrete Hamiltonian 85. Jung P, Leyendecker S, Linn J, Ortiz M (2011) A discrete
systems. J Nonlinear Sci 6(5):449–467 mechanics approach to the Cosserat rod theory—Part 1: Static
65. Goyal S, Perkins NC, Lee CL (2005) Nonlinear dynamics equilibria. Int J Numer Methods Eng 85(1):31–60
and loop formation in Kirchhoff rods with implications to the 86. Kane C, Marsden JE, Ortiz M, West M (2000) Variational
mechanics of DNA and cables. J Comput Phys 209(1):371–389 integrators and the Newmark algorithm for conservative and
66. Greco L, Cuomo M (2013) B-Spline interpolation of Kirch- dissipative mechanical systems. Int J Numer Methods Eng
hoff–Love space rods. Comput Methods Appl Mech Eng 49(10):1295–1325
256:251–269 87. Kapania RK, Li J (2003) A formulation and implementation of
67. Greco L, Cuomo M (2016) An isogeometric implicit mixed geometrically exact curved beam elements incorporating finite
finite element for Kirchhoff space rods. Comput Methods Appl strains and finite rotations. Comput Mech 30(5):444–459
Mech Eng 298:325–349 88. Karamanlidis D, Jasti R (1987) Curved mixed beam elements
68. Gruttmann F, Sauer R, Wagner W (1998) A geometrical non- for the analysis of thin-walled free-form arches. Ingenieur Arch
linear eccentric 3D-beam element with arbitrary cross-sections. 57(5):361–367
Comput Methods Appl Mech Eng 160(34):383–400 89. Kirchhoff G (1859) Ueber das Gleichgewicht und die Bewe-
69. Hsiao KM, Lin JY, Lin WY (1999) A consistent co-rotational gung eines unendlich dünnen elastischen Stabes. J für die reine
finite element formulation for geometrically nonlinear dynamic und angewandte Math 56:285–313
analysis of 3-D beams. Comput Methods Appl Mech Eng 90. Koiter WT (1966) On the nonlinear theory of thin elastic shells.
169(1–2):1–18 In: Proceedings of the Koninklijke Nederlandse Akademie van
70. Hsiao KM, Yang RT (1995) A co-rotational formulation for Wetenschappen, vol 69.
nonlinear dynamic analysis of curved Euler beam. Comput 91. Kondoh K, Tanaka K, Atluri SN (1986) An explicit expres-
Struc 54(6):1091–1097 sion for the tangent-stiffness of a finitely deformed 3-D beam
71. Hsiao KM, Yang RT, Lee AC (1994) A consistent finite element and its use in the analysis of space frames. Comput Struc
formulation for non-linear dynamic analysis of planar beam. Int 24(2):253–271
J Numer Methods Eng 37(1):75–89 92. Koschnick F (2004) Geometrische locking-effekte bei Finiten
72. Hughes TJR (2000) The finite element method: linear static and Elementen und ein allgemeines Konzept zu ihrer Vermeidung.
dynamic finite element analysis. Dover, New York Ph.D thesis, Lehrstuhl für Statik, Technische Universität
73. Hughes TJR, Taylor RL, Kanoknukulchai W (1977) A simple München
and efficient finite element for plate bending. Int J Numer Meth- 93. Kulachenko A, Denoyelle T, Galland S, Lindström SB
ods Eng 11(10):1529–1543 (2012) Elastic properties of cellulose nanopaper. Cellulose
74. Ibrahimbegović A (1995) On finite element implementation of 19(3):793–807
geometrically nonlinear Reissner’s beam theory: three-dimen- 94. Kulachenko A, Uesaka T (2012) Direct simulations of fiber net-
sional curved beam elements. Comput Methods Appl Mech Eng work deformation and failure. Mech Mater 51:1–14
122(1–2):11–26 95. Lang H, Arnold M (2012) Numerical aspects in the dynamic
75. Ibrahimbegović A (1997) On the choice of finite rotation simulation of geometrically exact rods. Appl Numer Math
parameters. Comput Methods Appl Mech Eng 149(1–4):49–71 62(10):1411–1427
76. Ibrahimbegović A, Frey F (1993) Finite element analysis of 96. Lang H, Linn J, Arnold M (2010) Multi-body dynamics simu-
linear and non-linear planar deformations of elastic initially lation of geometrically exact Cosserat rods. Multibody Syst
curved beams. Int J Numer Methods Eng 36(19):3239–3258 Dynam 25(3):285–312
77. Ibrahimbegović A, Frey F, Kozar I (1995) Computational 97. Langer J, Singer DA (1996) Lagrangian aspects of the Kirch-
aspects of vector-like parametrization of three-dimensional hoff elastic rod. SIAM Rev 38(4):605–618
finite rotations. Int J Numer Methods Eng 38(21):3653–3673 98. Lazarus A, Miller JJT, Reis PM (2013) Continuation of equi-
78. Ibrahimbegović A, Mamouri S (2002) Energy conserving/ libria and stability of slender elastic rods using an asymptotic
decaying implicit time-stepping scheme for nonlinear dynamics numerical method. J Mech Phys Solids 61(8):1712–1736
of three-dimensional beams undergoing finite rotations. Com- 99. Le T-N, Battini J-M, Hjiaj M (2014) A consistent 3D corota-
put Methods Appl Mech Eng 191(3738):4241–4258 tional beam element for nonlinear dynamic analysis of flexible
79. Ibrahimbegović A, Taylor RL (2002) On the role of frame- structures. Comput Methods Appl Mech Eng 269:538–565
invariance in structural mechanics models at finite rotations. 100. Lee PG, Sin HC (1993) Locking-free straight beam ele-
Comput Methods Appl Mech Eng 191(45):5159–5176 ment based on curvature. Commun Numer Methods Eng
80. Iosilevich A, Bathe K-J, Brezzi F (1997) On evaluating the inf- 9(12):1005–1011
sup condition for plate bending elements. Int J Numer Methods 101. Lens EV, Cardona A (2008) A nonlinear beam element formu-
Eng 40(19):3639–3663 lation in the framework of an energy preserving time integration
81. Iura M, Atluri SN (1988) Dynamic analysis of finitely stretched scheme for constrained multibody systems dynamics. Comput
and rotated three-dimensional space-curved beams. Comput Struc 86(12):47–63
Struc 29:875–889 102. Linn J (2016) Discrete kinematics of Cosserat rods based on
the difference geometry of framed curves. In: The 4th joint
13
C. Meier et al.
international conference on multibody system dynamics, Mon- 123. Romero I (2004) The interpolation of rotations and its applica-
tréal, Canada tion to finite element models of geometrically exact rods. Com-
103. Linn J, Lang H, Tuganov A (2013) Geometrically exact put Mech 34:121–133
Cosserat rods with Kelvin–Voigt type viscous damping. Mech 124. Romero I (2008) A comparison of finite elements for nonlinear
Sci 4(1):79–96 beams: the absolute nodal coordinate and geometrically exact
104. Love AEH (1944) A treatise on the mathematical theory of formulations. Multibody Syst Dynam 20(1):51–68
elasticity. Dover, New York 125. Romero I (2008) Formulation and performance of variational
105. Lyly M, Stenberg R, Vihinen T (1993) A stable bilinear element integrators for rotating bodies. Comput Mech 42(6):825–836
for the Reissner–Mindlin plate model. Comput Methods Appl 126. Romero I, Armero F (2002) An objective finite element approx-
Mech Eng 110(3–4):343–357 imation of the kinematics of geometrically exact rods and its use
106. Maier M, Müller KW, Heussinger C, Köhler S, Wall WA, in the formulation of an energy-momentum conserving scheme
Bausch AR, Lieleg O (2015) A single charge in the actin bind- in dynamics. Int J Numer Methods Eng 54(12):1683–1716
ing domain of fascin can independently tune the linear and 127. Romero I, Urrecha M, Cyron CJ (2014) A torsion-free non-lin-
non-linear response of an actin bundle network. Eur Phys J E ear beam model. Int J Non-Linear Mech 58:1–10
38(5):50 128. Sander O (2010) Geodesic finite elements for Cosserat rods. Int
107. Marsden JE, Hughes TJR (1994) Mathematical foundations of J Numer Methods Eng 82(13):1645–1670
elasticity. Dover, New York 129. Sansour C, Nguyen TL, Hjiaj M (2015) An energy-momentum
108. Meier C, Grill MJ, Wall WA, Popp A (2016) Geometrically method for in-plane geometrically exact Euler–Bernoulli beam
exact beam elements and smooth contact schemes for the mod- dynamics. Int J Numer Methods Eng 102(2):99–134
eling of fiber-based materials and structures. arXiv preprint, 130. Sansour C, Wagner W (2003) Multiplicative updating of the
arXiv:1611.06436 rotation tensor in the finite element analysis of rods and shells: a
109. Meier C, Popp A, Wall WA (2014) An objective 3D large path independent approach. Comput Mech 31(1):153–162
deformation finite element formulation for geometrically exact 131. Santos HAFA, Pimenta PM, Almeida JPM (2011) A hybrid-
curved Kirchhoff rods. Comput Methods Appl Mech Eng mixed finite element formulation for the geometrically exact
278:445–478 analysis of three-dimensional framed structures. Comput Mech
110. Meier C, Popp A, Wall WA (2015) A locking-free finite ele- 48(5):591–613
ment formulation and reduced models for geometrically exact 132. Schmidt MG, Ismail AE, Sauer RA (2015) A continuum
Kirchhoff rods. Comput Methods Appl Mech Eng 290:314–341 mechanical surrogate model for atomic beam structures. Int J
111. Meier C, Popp A, Wall WA (2016) A finite element approach Numer Methods Eng 13(5):413–442
for the line-to-line contact interaction of thin beams with 133. Schulz M, Filippou FC (2001) Non-linear spatial Timoshenko
arbitrary orientation. Comput Methods Appl Mech Eng beam element with curvature interpolation. Int J Numer Meth-
308:377–413 ods Eng 50(4):761–785
112. Meier C, Wall WA, Popp A (2017) A unified approach for 134. Shabana AA, Hussien HA, Escalona JL (1998) Application of
beam-to-beam contact. Comput Methods Appl Mech Eng the absolute nodal coordinate formulation to large rotation and
315:972–1010 large deformation problems. J Mech Design 120(2):188–195
113. Müller KW, Bruinsma RF, Lieleg O, Bausch AR, Wall WA, 135. Shabana AA, Yakoub RY (2001) Three dimensional absolute
Levine AJ (2014) Rheology of semiflexible bundle networks nodal coordinate formulation for beam elements: theory. J Mech
with transient linkers. Phys Rev Lett 112:238102 Design 123(4):606–613
114. Müller KW, Meier C, Wall WA (2015) Resolution of sub- 136. Shi Y, Hearst JE (1994) The Kirchhoff elastic rod, the nonlin-
element length scales in Brownian dynamics simulations of ear Schrödinger equation, and DNA supercoiling. J Chem Phys
biopolymer networks with geometrically exact beam finite ele- 101(6):5186–5200
ments. J Comput Phys 303:185–202 137. Shoemake K (1985) Animating rotation with quaternion curves.
115. Noor AK, Peters JM (1981) Mixed models and reduced/selec- ACM SIGGRAPH Comput Graph 19(3):245–254
tive integration displacement models for nonlinear analysis of 138. Simo JC (1985) A finite strain beam formulation. The three-
curved beams. Int J Numer Methods Eng 17(4):615–631 dimensional dynamic problem. Part I. Comput Methods Appl
116. Petrov E, Géradin M (1998) Finite element theory for curved Mech Eng 49:55–70
and twisted beams based on exact solutions for three-dimen- 139. Simo JC, Hughes TJR (1986) On the variational foundations of
sional solids Part 1: beam concept and geometrically exact assumed strain methods. J Appl Mech 53:51–54
nonlinear formulation. Comput Methods Appl Mech Eng 140. Simo JC, Vu-Quoc L (1986) A three-dimensional finite strain
165(1–4):43–92 rod model. Part II: computational aspects. Comput Methods
117. Prathap G (1985) The curved beam/deep arch/finite ring ele- Appl Mech Eng 58:79–116
ment revisited. Int J Numer Methods Eng 21(3):389–407 141. Simo JC, Vu-Quoc L (1988) On the dynamics in space of rods
118. Prathap G, Naganarayana BP (1990) Analysis of locking and undergoing large motions: a geometrically exact approach.
stress oscillations in a general curved beam element. Int J Comput Methods Appl Mech Eng 66(2):125–161
Numer Methods Eng 30(1):177–200 142. Simo JC, Wong KK (1991) Unconditionally stable algorithms
119. Quarteroni A, Sacco R, Fausto S (2000) Numerical mathemat- for rigid body dynamics that exactly preserve energy and
ics. Springer, New York momentum. Int J Numer Methods Eng 31(1):19–52
120. Reissner E (1972) On one-dimensional finite-strain beam the- 143. Smolenski WM (1998) Statically and kinematically exact non-
ory: the plane problem. Zeitschrift für Angewandte Mathematik linear theory of rods and its numerical verification. Comput
und Physik (ZAMP) 23(5):795–804 Methods Appl Mech Eng 178(1–2):89–113
121. Reissner E (1981) On finite deformations of space-curved 144. Sonneville V, Cardona A, Brüls O (2014) Geometric interpreta-
beams. Zeitschrift für Angewandte Mathematik und Physik tion of a non-linear beam finite element on the Lie group SE(3).
(ZAMP) 32(6):734–744 Arch Mech Eng 61:305–329
122. Rodney D, Gadot B, Martinez OR, Roscoat SR, Orgéas L 145. Sonneville V, Cardona A, Brüls O (2014) Geometrically exact
(2016) Reversible dilatancy in entangled single-wire materials. beam finite element formulated on the special Euclidean group.
Nat Mater 15:72–77 Comput Methods Appl Mech Eng 268:451–474
13
Geometrically Exact Finite Element Formulations for Slender Beams: Kirchhoff–Love Theory…
146. Spurrier RA (1978) Comment on “singularity-free extraction 156. Weiss H (2002) Dynamics of geometrically nonlinear rods: I.
of a quaternion from a direction-cosine matrix”. J Spacecraft Mechanical models and equations of motion. Nonlinear Dynam
Rockets 15:255–255 30(4):357–381
147. Stolarski H, Belytschko T (1982) Membrane locking and 157. Weiss H (2002) Dynamics of geometrically nonlinear rods: II.
reduced integration for curved elements. J Appl Mech Numerical methods and computational examples. Nonlinear
49:172–176 Dynam 30(4):383–415
148. Strang G, Fix G (2008) An analysis of the finite elment method. 158. Wempner G (1969) Finite elements, finite rotations and small
Wellesley-Cambrigde Press, Cambrigde strains of flexible shells. Int J Solids Struc 5(2):117–153
149. Tessler A, Spiridigliozzi L (1986) Curved beam elements with 159. Yang Y, Tobias I, Olson WK (1993) Finite element analysis of
penalty relaxation. Int J Numer Methods Eng 23(12):2245–2262 DNA supercoiling. J Chem Phys 98(2):1673–1686
150. Timoshenko SP (1921) On the correction for shear of the dif- 160. Zhang Z, Qi Z, Wu Z, Fang H (2015) A spatial Euler–Bernoulli
ferential equation for transverse vibrations of prismatic bars. beam element for rigid-flexible coupling dynamic analysis of
Philos Magn Series 41(245):744–746 flexible structures. Shock Vibrat 2015:208127
151. Češarek P, Saje M, Zupan D (2012) Kinematically exact 161. Zhao Z, Ren G (2012) A quaternion-based formulation of
curved and twisted strain-based beam. Int J Solids Struc Euler–Bernoulli beam without singularity. Nonlinear Dynam
49(13):1802–1817 67(3):1825–1835
152. Vu TD, Durville D, Davies P (2015) Finite element simulation 162. Zupan D, Saje M (2003) Finite-element formulation of geomet-
of the mechanical behavior of synthetic braided ropes and vali- rically exact three-dimensional beam theories based on inter-
dation on a tensile test. Int J Solids Struc 58:106–116 polation of strain measures. Comput Methods Appl Mech Eng
153. Wall WA (2017) BACI: a multiphysics simulation environment. 192(49–50):5209–5248
Technical report. Technical University of Munich, Munich 163. Zupan D, Saje M (2006) The linearized three-dimensional
154. Wang Q, Wang CM (2007) The constitutive relation and small beam theory of naturally curved and twisted beams: the
scale parameter of nonlocal continuum mechanics for modelling strain vectors formulation. Comput Methods Appl Mech Eng
carbon nanotubes. Nanotechnology 18(7):075702 195(33–36):4557–4578
155. Weeger O, Yeung S-K, Dunn ML (2016) Isogeometric col- 164. Zupan E, Saje M, Zupan D (2013) On a virtual work consistent
location methods for Cosserat rods and rod structures. three-dimensional Reissner–Simo beam formulation using the
Comput Methods Appl Mech Eng 316:100. doi:10.1016/j. quaternion algebra. Acta Mech 224(8):1709–1729
cma.2016.05.009
13