Sie sind auf Seite 1von 23

F

locking phenomena is the main subject of the


Finite Elements for Plates and scientific field of finite element technology, an
Shells important research area within the broader field
of computational mechanics.
Manfred Bischoff
Institute for Structural Mechanics, University of
Stuttgart, Stuttgart, Germany
Introduction

This chapter deals with finite element formu-


Synonyms lations for plate and shell models. It provides
the most important technical details of standard
Discretization methods; Thin-walled structures
finite element formulations, highlights the issue
of locking phenomena, and explains methods to
avoid them. Simple numerical benchmark tests
Definitions
are described that help to assess finite element
performance. Details about mechanical model-
The finite element method (FEM) is a discretiza-
ing of thin-walled structures as well as the cor-
tion method to compute approximate solutions
responding mathematical formulations are not
for problems described by partial differential
treated here (see the section on “Cross-Refer-
equations. It is based on weak forms, which can
ences” at the end of this chapter for a list of
either be derived from the underlying differential
related material). However, the most important
equations via the Galerkin method or, in special
equations are provided in order to allow for a self-
cases, from variational principles. For plate
contained reading of this chapter.
and shell theories, one frequently has to deal
The governing equations of plate and shell
with stiff differential equations, posing specific
theories can be constructed directly based on
challenges onto the discretization schemes. In
kinematics of two-dimensional surfaces in space,
particular, an effect known as locking may result
using elements of analytical differential geom-
in severe underestimation of the displacements,
etry and postulating the constitutive equations.
i.e., the finite elements are too stiff. Typically,
Alternatively, they can be derived from the gov-
reduced convergence rates in the pre-asymptotic
erning equations of three-dimensional elasticity
regime (for coarse meshes), oscillations of
via dimensional reduction. The second option
stress resultants, and a strong dependence of
is pursued in the following, with the focus on
the errors on the plate or shell slenderness are
materially linear problems, that is, elastic plates
observed. Analyzing and avoiding the pertinent
and shells.
© Springer-Verlag GmbH Germany 2018
H. Altenbach, A. Öchsner (eds.), Encyclopedia of Continuum Mechanics,
https://doi.org/10.1007/978-3-662-53605-6_14-1
2 Finite Elements for Plates and Shells

Kirchhoff-Love Type Plate and Shell The remaining governing equations, namely, the
Models equilibrium equations as well as the essential and
natural boundary conditions, are not explicitly
Kirchhoff Plate Model needed in the following, and therefore they are
not reproduced herein.
Governing Equations and Weak Form Kirchhoff plate theory (Kirchhoff 1850) pro-
In the context of flat plates, only the geometri- vides an approximation of the three-dimensional
cally linear case is treated herein. For plates un- theory of elasticity for thin, flat plates subject to
dergoing large deformations, membrane effects loading in transverse direction. It is asymptoti-
become important, requiring a shell theory. We cally correct, that, is one can show that in the
start with briefly repeating the most important thin limit h ! 0, h being the plate thickness,
equations from three-dimensional continuum me- the result of the model equations converges to-
chanics. ward the three-dimensional solution. The main
Using the short notation @i .ı/ WD @.ı/ @xi
, the assumptions are:
kinematic equations define the Cartesian com-
ponents of the linearized strain tensor 1. Kinematic assumptions:
(a) Cross-sectional material fibers remain
1  straight during deformation. Thus, in-plane
"ij D @j ui C @i uj : (1)
2 displacements u˛ are linear functions of
the thickness coordinate x3 .
For the case of linearized elasticity, the material (b) Cross-sectional material fibers remain nor-
equations (constitutive law) define the compo- mal to the mid-surface throughout defor-
nents of the Cauchy stress tensor mation. Thus, their rotations are coupled
with the gradients of the transverse dis-
ij D Cij kl "kl : (2) placement, ˛ D @˛ U3 ; see also Fig. 1.
2. Static (kinetic) assumption: Transverse nor-
Here, and in the following, repeated indices im- mal stress is zero, 33 D 0.
ply summation, where Latin indices run from 1 to
3, while Greek indices take on values 1 or 2. For A key ingredient of derivation of plate and shell
isotropic materials, the components Cij kl of the models from three-dimensional elasticity is di-
fourth-order material tensor (also called elasticity mensional reduction, expressing all involved me-
tensor) are implicitly defined by chanical quantities as functions of two (instead of
three) spatial coordinates. For the displacements
E
11 D of the Kirchhoff plate model, this reduction is
.1 C /.1  2/
defined as
Œ.1  /"11  ."22 C "33 / ; (3)
E u˛ .x1 ; x2 ; x3 / D x3 @˛ U3 .x1 ; x2 / ; (7)
22 D
.1 C /.1  2/ u3 .x1 ; x2 ; x3 / D U3 .x1 ; x2 / : (8)
Œ.1  /"22  ."11 C "33 / ; (4)
E that is, the displacements ui .x1 ; x2 ; x3 / of
33 D any point in the three-dimensional continuum
.1 C /.1  2/
are expressed as functions of one single
Œ.1  /"33  ."11 C "22 / ; (5) variable, namely, the mid-surface displacement
E E U3 .x1 ; x2 /, which is a function of the mid-
12 D "12 ; 13 D "13 ;
1C 1C surface coordinates only. Equation (7) effectively
E puts into place the kinematic assumptions 1a
23 D "23 : (6) and 1b.
1C
Finite Elements for Plates and Shells 3

Finite Elements for


Plates and Shells, Fig. 1 u2 u1
u3
Flat plate, definition of f2
displacements and f1
rotations U3

x1
x3
x2
h
F

The static assumption (also called kinetic as- It can be seen that vanishing transverse shear
sumption if inertia effects are taken into account) strains "˛3 are a natural result of the specific
is enforced via modification of the material law. displacement parametrization given in Eqs. (7)
The condition 33 D 0 is introduced into Eq. (5), and (8).
which is then solved for "33 , The linear part of "˛ˇ in x3 is the curvature of
the mid-surface of the plate

"33 D ."11 C "22 / : (9)
1 K˛ˇ D @2˛ˇ U3 : (15)
Equations (3) and (4) are thus transformed into
It will be seen later, when dealing with Mindlin’s
E shear deformable plate model, that the material
11 D ."11 C "22 / ; (10) curvature ˛ˇ , being energetically conjugate to
.1   2 /
the moments M˛ˇ , in general differs from this
E
22 D ."22 C "11 / : (11) geometric curvature K˛ˇ . In the Kirchhoff model
.1   2 / they are identical.
With Eqs. (7) and (8), dimensional reduction Stress resultants are defined as resultant forces
of the kinematic Eq. (1) is achieved, resulting or moments of the transverse shear stress and in-
in separate expressions for in-plane, transverse plane normal stress, respectively,
shear, and transverse normal strains, h h
Z2 Z2
1  Q˛ D ˛3 dx3 ; M˛ˇ D x3 ˛ˇ dx3 :
"˛ˇ D @ˇ .x3 @˛ U3 / C @˛ .x3 @ˇ U3 /
2
h
2 h
2
D x3 @2˛ˇ U3 ; (12) (16)
1 Due to symmetry of the stress tensor, both twist-
"˛3 D .@˛ U3 C @˛ U3 / D 0 ; (13) ing moments are equal, M12 D M21 . Introducing
2
the material, Eqs. (10), (11), and (6)1 as well as
"33 D @3 U3 D 0 : (14) the kinematic Eq. (12) into (16)2 yield
4 Finite Elements for Plates and Shells

h
Z2
E    
M11 D x3 @1 U1  x3 @211 U3 C  @2 U2  x3 @222 U3 dx3
.1   2 /
h
2

h
Z2
E  2   
D x32 @11 U3 C  @222 U3 dx3
.1   2 /
h
2

Eh3 Eh3
D K 11 C K22 ; (17)
12.1   2 / 12.1   2 /
Eh3 Eh3
M22 D K 22 C K11 ; (18)
12.1   2 / 12.1   2 /
Eh3
M12 D K12 : (19)
12.1 C /

 4  q
In the last step in Eq. (17), the definition of @1111 C 2@41122 C @42222 U3 D ; (24)
D
the curvature from Eq. (15) has been used. This
defines the material equations of Kirchhoff plate where the abbreviation
theory, combining stress resultants and kinematic
variables. There is no such expression for the Eh3
D WD (25)
transverse shear forces Q˛ , because kinematic 12.1   2 /
assumption 1b is equivalent to assuming an in-
finite transverse shear stiffness along with zero for the plate bending stiffness has been used.
transverse shear strain, cf. Eq. (13). The virtual work principle (principle of virtual
The equilibrium equations for the case of later- displacements) can be written as
ally loaded plates (no distributed moments) read Z
ıK˛ˇ M˛ˇ d˝
@1 M11 C @2 M12  Q1 D 0 ; (20) ˝
„ ƒ‚ …
@2 M22 C @1 M12  Q2 D 0 ; (21) ıWint
Z Z
@1 Q1 C @2 Q2 C q D 0 : (22)
D ıU3 q d˝ C ıU3 Qn d@˝ ; (26)
˝ @˝
Eliminating Q1 and Q2 from the system of „ ƒ‚ …
Eqs. (20), (21), and (22) yields ıWext

@211 M11 C 2@212 M12 C @222 M22 C q D 0 : (23) where Qn denotes prescribed transverse shear
forces at the Neumann part of the boundary @˝ .
Along with the material law and the kinematic Along with the kinematic and material equations
equations, this finally provides the biharmonic of the Kirchhoff plate model, the virtual work
differential equation (strong form) of the Kirch- principle can be written in terms of the primal
hoff plate model, variable U3 only,
Finite Elements for Plates and Shells 5

Z
 2 
@11 ıU3 D@211 U3 C 2@212 ıU3 D.1  /@212 U3 C @222 ıU3 D@222 U3 d˝
˝
Z Z
D ıU3 q d˝ C ıU3 Qn d@˝ 8ıU3 : (27)
˝ @˝

The requirement of this equation to hold for 2. The shape functions have to be C nv - F
arbitrary virtual displacements ıU3 is mathemat- continuous inside the elements.
ically equivalent to the equilibrium equations and 3. The shape functions have to be C nv 1 -
the force boundary conditions. An approximation continuous across element boundaries,
error comes into play as soon as the function
spaces for U3 and ıU3 are restricted, as it is done where a C n -continuous function is defined
in finite element formulations. as a function for which the nth derivative is
continuous. For mixed and hybrid-mixed finite
Finite Element Formulation element formulations, based on multi-field
Here, and in the following, it is assumed that variational principles, different conditions apply.
the reader has some basic knowledge of the For displacement-based elements, conditions 1
finite element method (FEM). The focus of this and 2 are necessary for consistency (and thus,
chapter is on specific aspects of finite element along with stability, for convergence), whereas
formulations for plates and shells. Within FEM, condition 3 is only a sufficient condition. Thus,
the domain ˝ is divided into subdomains ˝e , incompatible formulations may or may not be
S
the finite elements, such that ˝ D ˝e . Each convergent, which has to be checked individually,
element has a set of nodes K, where the ones on for instance, via the patch test.
the element edges are shared with neighboring el- The variational index for Kirchhoff plate
ements and carry the same degrees of freedom. In theory is nv D 2, as second derivatives
a standard Galerkin finite element approach (also show up in the definition of the curvature and
often called standard displacement formulation), twist. Therefore, according to condition 1, at
the primary variable, namely, the displacement least completely quadratic shape functions are
U3 in Kirchhoff plate theory, is approximated required. Condition 2 is easily satisfied, but the
by a local ansatz, consisting of shape functions third condition, asking for C 1 -continuity across
expressed in terms of local element coordinates element edges, poses severe challenges. The
; . The virtual displacements are treated simi- difficulties in formulating corresponding shape
larly. The shape functions cannot be freely chosen functions are one of the major reasons for the
but are subject to a number of conditions in popularity of Reissner-Mindlin-type plate and
order to obtain a consistent formulation. These shell finite elements, for which C 0 -continuity is
conditions crucially depend on the underlying sufficient (see below).
weak form, in particular the variational index In analogy to C 1 -continuous Bernoulli beam
nv . It is defined as the highest derivative of the finite elements, which use cubic Hermite poly-
primal variable showing up in the virtual work nomials to interpolate the transverse deflection
principle (27). from nodal displacements and their derivatives, it
The most important conditions are as fol- is natural to introduce displacements U3 as well
lows: as their directional derivatives @˛ U3 (or, equiv-
alently, rotations ˛ ) as nodal unknowns. Adini
1. The shape functions have to be complete up to and Clough (1961) proposed a corresponding
polynomial order nv . formulation for rectangular four-node elements
using 12 degrees of freedom. The element is
6 Finite Elements for Plates and Shells

known as Adini-Clough-Melosh (ACM) element mentation. Until today, the DKT (discrete Kirch-
today, as Melosh (1961) independently developed hoff triangle) element is one of the most popular
an identical formulation. The ACM element sat- Kirchhoff plate finite elements.
isfies C 1 -continuity at the nodes but not along Since approximately three decades, most of
the edges and is thus incompatible. Moreover, it the literature focuses on general shell elements,
does not pass the patch test for quadratic displace- including plates as a special case, instead of
ments (constant moments). distinctively presenting plate finite element for-
In order to overcome this problem, Bogner mulations. More recent developments in finite
et al. (1965) introduced an additional twisting element formulations with relevance for plates
degree of freedom @212 U3 . The resulting BFS are therefore treated below in the section on
element is C 1 -continuous along the edges for “Kirchhoff-Love Shell Finite Elements”.
the special case of rectangular element shapes. A
similar construction for arbitrary quadrilaterals, Remarks on Modeling and Discretization
however, is not possible. As the aforementioned In this section, some qualities of linear Kirchhoff
continuity condition is sufficient, but not neces- plate theory that are of particular interest in the
sary for convergence, the BFS element can still context of finite element analysis are briefly de-
be used for arbitrarily shaped elements. scribed.
Figure 2 shows the function spaces of the The normality assumption is equivalent to as-
ACM and BFS element, respectively. The suming infinite transverse shear stiffness. As a re-
quadratic pattern of the BFS element on the sult, transverse shear strains vanish, see Eq. (13),
right illustrates the fact that the shape functions and the material law cannot be used to com-
can be obtained in the form of a tensor product of pute transverse shear forces. Instead, they are
the one-dimensional cubic Hermite polynomials computed from derivatives of the moments via
used for Bernoulli beam elements. equilibrium Eqs. (20) and (21)
One of the first fully conforming elements is
the HCT element by Clough and Tocher (1965). Q1 D @1 M11 C @2 M12 ; (28)
It relies on subdivision of a triangular element
Q2 D @2 M22 C @1 M12 : (29)
into subtriangles. Its limited popularity today is
probably due to the cumbersome formulation and
The moments M˛ˇ are obtained from the cur-
implementation and the fact that it is relatively
vatures K˛ˇ via constitutive Eqs. (17), (18), and
stiff.
(19). The curvatures are the second derivatives
An alternative strategy is to circumvent the
of the deflection U3 according to Eq. (15). Thus,
C 1 -requirement using the discrete Kirchhoff con-
Eqs. (28) and (29) involve 3rd derivatives of the
cept, pioneered by Stricklin et al. (1969) and
shape functions. In most C 1 -continuous Kirch-
Dhatt (1970). It relies on the idea of using a
hoff plate finite elements, the shape functions are
shear deformable plate model as basis, for which
3rd order or higher. However, formulations with
C 0 -continuous shape functions are sufficient (see
quadratic shape functions are also possible, for
the section on “Reissner-Mindlin-Type Plate and
instance, when using splines (see the section on
Shell Models” below). The Kirchhoff constraint
Kirchhoff-Love Shell Elements below). In this
of normals remaining normal is then discretely
case, the 3rd derivatives are identically zero, and
satisfied along the edges of the element, such
the transverse shear forces cannot be obtained
that convergence to the Kirchhoff plate solution
directly from (28) and (29). An alternative is
is guaranteed with mesh refinement. More pre-
to use finite differences on patches of adjacent
cisely, the deflection along the edges is linked
elements.
to the nodal rotations like in Bernoulli beam
One of the peculiarities of Kirchhoff plate
elements. The idea did not receive a lot of at-
theory is the mismatch between the order of the
tention until (Batoz 1980) (see also Batoz 1982)
differential equation (4th order) and the num-
presented an in-depth study of its theoretical
ber of degrees of freedom available (only one,
background, performance, and practical imple-
Finite Elements for Plates and Shells 7

Finite Elements for


Plates and Shells, Fig. 2
Function spaces of
rectangular Kirchhoff plate
finite elements by Adini
and Clough (1961) (left)
and Bogner et al. (1965)
(right)

viz., U3 ) to define boundary conditions. If, for ingenious idea of Gustav Kirchhoff to resolve the
example, the transverse deflection U3 is fixed edge problem is the reason that thin plate theory
along an edge parallel to the x1 -coordinate line, is associated with his name until today.
then the rotation 1 D @1 U3 vanishes as well, Kirchhoff plate theory does not involve any
resulting in a twisting moment M21 as support boundary layer effects nor are there any locking
reaction. A Neumann-type boundary condition problems, such that no further remarks on dis-
M21 D 0 cannot be directly realized. At simply cretization are necessary at this point. The most
supported edges, these twisting moments artifi- important issue, namely, dealing with the C 1 -
cially contribute to equilibrium, and as a result, continuity requirement, has already been treated
the transverse shear forces and the actual support in the previous section.
reactions are underestimated.
The problem is resolved by introducing effec- Benchmarking
tive transverse shear forces that include a contri- When assessing the performance of finite ele-
bution from derivatives of the artificial twisting ments, it is first necessary to verify convergence.
moments at the boundary: This can be done by applying Iron’s patch test,
which demands exact representation of the pri-
Q1eff D Q1 C @2 M12 ; Q2eff D Q2 C @1 M21 : mal variables up to a polynomial order equal to
the variational index. For Kirchhoff plates, this
(30)
means that quadratic displacement distributions,
and thus constant moments, have to be repre-
An illustrative way to explain this idea is to
sented exactly for arbitrary meshes. This can be
picture the twisting moments as pairs of forces
tested by analyzing:
on infinitesimal line elements along the edge of
the plate. Adding the forces at the interfaces
of adjacent elements directly leads to the above • A rectangular cantilever plate subject to a con-
definition. It also gives rise to point forces at stant bending moment along the edge opposite
corners of plates. In a finite element implemen- to the clamped edge
tation, computation of the effective shear forces • A rectangular plate that is simply supported at
has to be implemented in addition to the afore- two adjacent edges und subject to a point force
mentioned computation of shear forces from the P at the free corner
equilibrium equations. Note that Q˛eff are really
the physical forces and support reactions, not If Poisson’s ratio is set to zero, the first scenario
just the artificial result of a mathematical trick. produces a state of uniaxial bending with the ex-
In Mindlin’s shear deformable plate theory, the act solution being identical to the corresponding
point forces appear as strong concentrations of Bernoulli beam solution. In the second example,
transverse shear forces, which become singular a constant twisting moment M12 D  P2 occurs in
with vanishing plate thickness h ! 0. This the case of a rectangular plate, independent of its
8 Finite Elements for Plates and Shells

Finite Elements for Plates and Shells, Fig. 3 Undeformed and deformed configuration of shell body

size. The point force corresponds to the effective marized. Convective, curvilinear coordinates  i
shear force at the edge, emanating from the non- are used, where the mid-surface of the shell
zero twisting moments along the free edge. Both is parametrized by  ˛ and  3 2 Œ h2 ; h2  is
tests ought to be performed with regular and the thickness coordinate; see Fig. 3. The nota-
irregular meshes. tion to express partial derivatives is redefined as
Point forces in the interior of plates, some- @i .ı/ WD @.ı/
@ i
. Capital letters refer to quantities in
times used in benchmark problems in the lit- the undeformed configuration; lowercase letters
erature, give rise to singularities in all stress are used for the deformed configuration.
resultants, i.e., transverse shear forces, bending Dimensional reduction of the three-dimensional
moments, and the twisting moment become infi- equation is realized by expressing the position
nite. The displacements under the load, however, vector X to a point in the undeformed shell body
remain finite. by the position vector R to its projection on the
mid-surface and the shell normal (the so-called
Kirchhoff-Love Shell Model director) A3 ,

Kinematic Equations X. 1 ;  2 ;  3 / D R. 1 ;  2 / C  3 A3 . 1 ;  2 / :


This section focuses on the kinematic equations (31)
of thin shell theory, because those are the most A similar expression applies for the deformed
important ones when discussing finite element configuration
formulations. Shell formulations based on the
normality assumption, neglecting transverse x. 1 ;  2 ;  3 / D r. 1 ;  2 / C  3 a3 . 1 ;  2 / ; (32)
shear strains, can be traced back to the landmark
contribution of Love (1888). In contrast to plate where r is related to R via the mid-surface dis-
theory, there are numerous different possibilities placement v,
to formulate shell models, in particular regarding r D R C v: (33)
simplifying assumptions concerning geometry
The displacement u of a point in the shell body is
and curvature. A corresponding overview, with
given by
an emphasis on shell modeling in the context of
finite element formulations, is given in Bischoff
u D v C  3 .a3  A3 / : (34)
et al. (2004).
In the following, the governing equations for
the geometrically nonlinear case are briefly sum-
Finite Elements for Plates and Shells 9

The shell directors in the undeformed and de-


1 
formed configuration are obtained from the co- Eij D gij  Gij
2
variant base vectors A˛ D @˛ R and a˛ D @˛ r,
respectively, as 1 
D @i u  Gj C @j u  Gi C @i u  @j u ;
2
A1  A2 a1  a2 (38)
A3 D ; a3 D : (35)
jA1  A2 j ja1  a2 j
with gij D gi  gj and Gij D Gi  Gj , can
With the covariant base vectors in the shell body be expressed in terms of mid-surface quantities
F
only. Neglecting the parts that are quadratic in the
G˛ D @˛ X D @˛ R C  3 @˛ A3 D A˛ C  3 @˛ A3 ; thickness coordinate  3 , the kinematic equations
0
are separated into a membrane part E˛ˇ and
G3 D @3 X D A3 ; (36) curvatures K˛ˇ
3
g˛ D a˛ C  @˛ a3 ; g3 D a3 ; (37)
0
E˛ˇ D E˛ˇ C  3 K˛ˇ ; (39)
the components of the Green-Lagrange strain
tensor E D Eij Gi ˝ Gj , with

0 1  1 
E˛ˇ D @˛ v  Aˇ C @ˇ v  A˛ C @˛ v  @ˇ v D a˛ˇ  A˛ˇ ; (40)
2 2
1 
K˛ˇ D @˛ a3  aˇ C @ˇ a3  a˛  @˛ A3  Aˇ  @ˇ A3  A˛ ; (41)
2

where a˛ˇ D a˛  aˇ and A˛ˇ D A˛  Aˇ . The @ˇ a˛  a3 D @ˇ @˛ r  a3 D @˛ @ˇ r  a3


out-of-plane components Ei3 are zero.
D @˛ aˇ  a3 DW b˛ˇ : (46)
From the orthogonality of A˛ and A3 (a˛ and
a3 , respectively) in the undeformed and deformed
B˛ˇ and b˛ˇ are the second fundamental forms
configuration, it follows
of the undeformed and the deformed mid-
surface, respectively. The curvatures can thus
@ˇ .A˛  A3 / D 0 D @ˇ A˛  A3 C A˛  @ˇ A3
be expressed as
(42)
K˛ˇ D b˛ˇ C B˛ˇ : (47)
) A˛  @ˇ A3 D @ˇ A˛  A3 ; (43)
analogously: a˛  @ˇ a3 D @ˇ a˛  a3 : (44) For thin shells, the metric can be assumed to
be constant through the thickness (G˛  A˛ )
For a practical implementation, expressions of in the constitutive equations. This is also known
the form @ˇ a˛  a3 are preferable to the equiva- as Love’s first approximation. The contravariant
lent a˛ @ˇ a3 , because they are easier to linearize. components of the membrane forces N ˛ˇ and
Furthermore, because derivatives commute, moments M ˛ˇ are thus given by
we have

@ˇ A˛  A3 D @ˇ @˛ R  A3 D @˛ @ˇ R  A3
D @˛ Aˇ  A3 DW B˛ˇ ; (45)
10 Finite Elements for Plates and Shells

 0   0 
N 11 D Eh E11 0
C E22 ; N 22 D Eh E22 0
C E11 ; (48)
Eh
N 12 D E0 ; (49)
2.1 C / 12
Eh3 Eh3
M 11 D .K11 C K22 / ; M 22 D .K22 C K11 / ; (50)
12.1   2 / 12.1   2 /
Eh3
M 12 D K12 : (51)
12.1 C /

Finite Element Formulation quadratic basis functions. The group around Kai-
Like in the case of Kirchhoff plate elements, Uwe Bletzinger in Munich was the first to present
the main challenge in the formulation of finite a corresponding Kirchhoff-Love shell element in
elements for Kirchhoff-Love shell theory is Kiendl et al. (2009).
the requirement of C 1 -continuity. Essentially, Kirchhoff-Love shell finite elements do not
Kirchhoff-Love shell finite elements can be play a dominant role today in commercial finite
formulated as a combination of concepts for element codes. However, the introduction of IGA
Kirchhoff plates for the bending part and two- has triggered a renaissance of the Kirchhoff-Love
dimensional solid elements for the membrane shell model, and discussion of corresponding
part. For the latter, C 0 -continuity of the formulations has reentered the focus of research
displacement approximation is sufficient. Until in computational mechanics.
today, classical concepts developed in the early
days of finite element analysis, like the SHEBA Membrane Locking and Methods to Avoid It
family of shell finite elements (Argyris and Besides the C 1 -continuity requirement, the
Scharpf 1968), belong to the most popular main challenge in finite element technology for
formulations. A relatively simple, yet very Kirchhoff-Love shell theory is the problem of
successful, thin shell element can be obtained membrane locking. Generally, the term locking
from the combination of the DKT element, describes the phenomenon that finite elements
described above, with the CST element (constant behave too stiff in certain situations. Locking can
strain triangle) for the membrane part; see Bathe be defined phenomenologically by the following
and Ho (1981). Also the BFS and ACM element symptoms:
technologies, described above, can be extended
to shells. 1. Displacements are severely underestimated
Fully conforming Kirchhoff-Love shell finite for coarse meshes. In the pre-asymptotic
elements based on the technology of subdivi- range, the order of convergence is below the
sion surfaces have been proposed by Cirak et al. optimal one.
(2000). C 1 -continuous function spaces may also 2. For a given mesh, the error depends on a cer-
be constructed using splines. Although spline- tain model parameter, and also the size of the
based finite elements are known since the 1970s, pre-asymptotic range with suboptimal conver-
(Höllig 2003), corresponding formulations, had gence depends on this parameter. In the case
their breakthrough only after Hughes et al. (2005) of thin-walled structures, this parameter is the
proposed the technology of isogeometric analy- slenderness; for general elasticity problems, it
sis (IGA). Using B-splines or NURBS (nonuni- is the bulk modulus (volumetric locking, also
form rational B-splines) allows construction of called Poisson locking, not treated herein).
fully C 1 -continuous function spaces already with
Finite Elements for Plates and Shells 11

3. Nonphysical stresses occur mostly in an oscil- The phenomenon can also be explained via
lating manner. the unbalance of derivatives in the kinematic
equations. Assume that equal order interpolation
Membrane locking results from the inability with polynomials of order p is used for both u
to represent pure bending deformations, also and w. Then, " D 0 can only be fulfilled if the
called inextensional deformations in the context order of w is p  1, because @1 u is of order p  1.
of shells, if the elements are curved. Due to Furthermore, with Eq. (52)2 , a constant requires
the complexity of the differential equations for the balance of @1 u and @211 w. From the previous
curved beams and shells, the origin of membrane condition, @211 w is of order p 3, so u would have F
locking is not as easily explained as transverse to be of order p  2. Reintroducing this into (52)
shear locking, which occurs in the Mindlin plate for " D 0 now imposes w to be of order p  3 and
model (see below). One way to shed some light so on. No matter how high the polynomial order
on this phenomenon is to consider a finite element of the involved finite element approximation is,
formulation based on degrees of freedom aligned eventually the entire approximation power is lost.
with the curvilinear coordinate system. This is Standard displacement-based Kirchhoff-Love
not typically done in FEM, but it can be done shell elements with equal order interpolation of
and serves as a vehicle to explain the origin of all variables therefore suffer from membrane
membrane locking in the following. locking, regardless of the type of shape functions
We consider a circular arch with radius R, used, e.g., Lagrange polynomials, B-splines,
fixed at one end and modeled by geometrically or NURBS. However, as already mentioned,
linear Bernoulli beam theory. Defining u and w as membrane locking shows up only for curved
the tangential and outward radial displacement, elements. Thus, linear triangles are always free
respectively, and  as the axial coordinate of the from membrane locking. Bilinear quadrilaterals
curved beam, the kinematic equations are exhibit locking only for warped geometry. In
many practical situations, the elements are
w u
" D @1 u C ; D @1  @211 w ; (52) flat or almost flat, if the structure is meshed
R R accordingly. For instance, a structured mesh of
where " is the normal strain (corresponding to a cylindrical shell consists of flat elements only.
the membrane strain in the context of shells) and Therefore, removing membrane locking for low-
is the curvature. A pure bending deformation order elements is not extensively treated in the
requires the membrane strain to vanish, " D 0, literature.
such that One way to avoid membrane locking is selec-
tive reduced integration of the membrane part.
w w The drawback of this simple method is that zero
@1 u D  ) D  @211 w : (53)
R R2 energy modes (hourglass modes) can spoil the
solution. Another option is to enrich the function
If polynomials or splines are used as shape func-
space of the membrane strains by the enhanced
tions, as typical for FEM and IGA, a constant
assumed strain (EAS) method by Simo and Rifai
curvature D 0 (constant bending moment)
(1990) or, equivalently, by the method of incom-
together with (53)2 implies the condition that w is
patible modes (Taylor et al. 1976). This does
constant. Due to the applied boundary condition,
not completely remove membrane locking in the
this constant must be zero and consequently w D
case of distorted meshes, but it alleviates it to an
0 in the entire beam. From Eq. (53)2 , it then
amount that satisfies practical needs.
follows that also D 0. Therefore, the condition
of vanishing membrane strains D 0 implies
Benchmarking
zero curvature and thus zero bending moments,
As plates are a subset of shells, the same bench-
effectively excluding inextensional deformations.
marks proposed for Kirchhoff plate elements
should also be passed by Kirchhoff-Love shell
12 Finite Elements for Plates and Shells

Finite Elements for


Plates and Shells, Fig. 4
Cylindrical shell strip,
problem setup

finite elements. In addition, benchmarks to check lent results for displacements, while at the same
whether a given element is free from membrane time, the stress resultants exhibit huge errors.
locking are needed. A simple, yet meaningful, The benchmark test ought to be performed
test is to compute a cylindrical shell strip subject also with an irregular mesh. If, for instance, four-
to uniaxial bending, as shown in Fig. 4; see also node elements are used, the element edges must
Echter et al. (2013). If Poisson’s ratio is set to not be aligned with the yaxis, such that inside
zero, a purely uniaxial state of bending occurs the domain, the elements have a warped paral-
and the Bernoulli beam solution can be taken as lelogram shape. For regular meshes, four-node
reference. Two load cases are considered, and in elements usually perform well without special
both cases, the magnitude of the load is scaled treatment of the membrane part, but for deformed
with h3 such that the displacement is independent meshes, locking may occur. This issue has been
of the thickness in the thin limit. studied in some detail in the doctoral thesis of the
In load case 1, a bending moment is applied at author, (Bischoff 1999).
the tip of the cantilever. It gives rise to a constant Numerous benchmarks for Kirchhoff-Love
bending moment M 11 in the entire structure, shell elements may be found in the literature.
while all other stress resultants are identically However, according to the author’s experience,
zero. The radial displacement is w D 1:2, in- an element that passes the aforementioned simple
dependent of the thickness. For a given finite locking test will also perform well in other
element mesh, the error will be zero or at least benchmarks.
independent of h if the element formulation is
locking-free. If the error increases with h, this is
a hint toward membrane locking. In load case 2, Reissner-Mindlin-Type Plate and
a force is applied at the free end, resulting in Shell Models
sinusoidally varying bending moments M 11 and
membrane forces N 11 . The exact  3 solution forthe Mindlin Plate Model
2 1
radial displacement is w D 10
C h 4000

0:9425C0:0007854h2 . Besides the displacement, Governing Equations and Weak Form
the membrane force N 11 should be checked. The Shear deformable plate theory is frequently re-
exact result can be easily computed from equi- lated to the contribution of Mindlin (1951), who
librium. An element that suffers from membrane established a shear deformable plate model with
locking will show oscillations in N 11 . This test is the aim to study the influence of rotatory iner-
particularly important for elements with higher- tia on the bending vibration of plates, although
order shape functions, which often provide excel- there have been earlier works, like the one of
Finite Elements for Plates and Shells 13

Reissner (1945). The differences are subtle, and As in this section, Cartesian coordinates are used;
this chapter mostly follows Mindlin’s derivation the symbol for the partial derivative is redefined
as it is simple and straightforward. accordingly, @i .ı/ WD @.ı/
@xi
. The linear part of the
Like in the case of the Kirchhoff plate model, in-plane strain is called the material curvature
there are kinematic assumptions and static as-
sumptions. However, the normality condition is 1 
˛ˇ D @ˇ ˛ C @˛ ˇ ; (59)
dropped, leaving: 2

as it is the energetically conjugate quantity to the


1. Kinematic assumption: Cross-sectional ma- F
moment M˛ˇ and shows up in the corresponding
terial fibers remain straight during deforma-
material equation. It differs from the geometric
tion. Thus, in-plane displacements u˛ are lin-
curvature K˛ˇ D @2˛ˇ U3 but converges toward
ear functions of the thickness coordinate x3 .
it with decreasing thickness, lim ˛ˇ D K˛ˇ .
2. Static (kinetic) assumption: Transverse nor- h!0
mal stress is zero, 33 D 0. Transverse shear strain is non-zero in the
Mindlin plate model. The definition of the shear
In the Mindlin plate model, the displacements are angle
given as ˛ D ˛ C @˛ U3 D 2"˛3 (60)
completes the kinematic equations. With the ab-
u˛ .x1 ; x2 ; x3 / D x3 ˛ .x1 ; x2 / ; (54) breviation D for the bending stiffness, introduced
u3 .x1 ; x2 ; x3 / D U3 .x1 ; x2 / : (55) in Eq. (25), the material equations can be writ-
ten as
that is, the displacements ui .x1 ; x2 ; x3 / of
any point in the three-dimensional continuum M11 D D11 C D22 ; (61)
are expressed as functions of three variables M22 D D22 C D11 ; (62)
(as opposed to only one in the Kirchhoff
model), namely, the mid-surface displacement M12 D D.1  /12 ; (63)
U3 .x1 ; x2 / and the rotations of cross-sectional Q˛ D ˛s Gh ˛ ; (64)
fibers ˛ .x1 ; x2 /. Equation (54) realizes the
kinematic assumption. The decisive difference where the first three equations are similar to
to the Kirchhoff model, Eq. (7), is that rotations those used in Kirchhoff plate theory, for which
are independent of the displacements. In order K˛ˇ D ˛ˇ . The parameter ˛s is the shear
to enforce the static assumption, the same correction factor that takes into account that the
modification of the material law as in Kirchhoff through-the-thickness distribution of "˛3 is not
plate theory is carried out, cf. Eq. (9). constant, as implied by the kinematic equations,
From Eqs. (54) and (55), the strain compo- but approximately quadratic. It is usually set to
nents are obtained as ˛s D 56 .
1   The equilibrium conditions in terms of the
"˛ˇ D x3 @ˇ ˛ C @˛ ˇ ; (56) stress resultants are identical to those of the
2
Kirchhoff model, Eqs. (20), (21), and (22). Intro-
1
"˛3 D .˛ C @˛ U3 / D 0 ; (57) ducing the material law, Eqs. (61), (62), (63), and
2 (64) and kinematic Eqs. (59) and (60) provide a
"33 D @3 U3 D 0 : (58) set of three second-order differential equations
14 Finite Elements for Plates and Shells

1 1 ˛s Gh
@211 1 C .1  /@222 1 C .1 C /@212 2  .1 C @1 U3 / D 0 ; (65)
2 2 D
1 1 ˛s Gh
@222 2 C .1  /@211 2 C .1 C /@212 1  .2 C @2 U3 / D 0 ; (66)
2 2 D
q
@211 U3 C @222 U3 D  : (67)
˛s Gh

It requires definition of three essential boundary The virtual work principle, which can serve as
conditions at each edge, which matches the num- a weak form for formulation of finite elements,
ber of degrees of freedom (U3 , 1 , and 2 ). Thus, can be written as
in contrast to Kirchhoff plate theory, no extra Z Z
 
treatment is needed at the boundaries. ı˛ˇ M˛ˇ C ı ˛ Q˛ d˝ D ıU3 q d˝ :
The coefficient of the last term in the first two
˝ ˝
equilibrium Eqs. (65) and (66) ˛sDGh D 6˛s h.1/2 (68)
tends to infinity with decreasing thickness, while The boundary terms, introducing prescribed mo-
the other coefficents are independent of h. There- ments and transverse shear forces at the edges,
fore, for thin plates we are dealing with a set are omitted for the sake of brevity. Introducing
of stiff differential equations, which is one of the kinematic and material equations yields
the reasons for the problem of transverse shear
locking, described below.
Z h
@1 ı1 D@1 1 C @1 ı1 D@2 2 C @2 ı2 D@1 1 C @2 ı2 D@2 2 (69)
˝

1
C .@1 ı2 C @2 ı1 / D.1  / .@1 2 C @2 1 / (70)
2
i Z
C .ı˛ C @˛ ıU3 / ˛s Gh .˛ C @˛ U3 / d˝ D ıU3 q d˝ : (71)
˝

As only first derivatives of all primal variables quadrilateral finite elements with equal order
show up, the variational index of this weak form interpolation of U3 and ˛ ,
is nv D 1, and thus bilinear, C 0 -continuous shape 2 3 2 K3
functions are sufficient for consistency. This dra- U3 n
Xnode U3
matically simplifies finite element formulation in 4 1 5  NK 4 1K 5 (72)
comparison to Kirchhoff elements. According to 2 KD1 2K
the “no free lunch theorem,” this comes along
with trouble, namely, transverse shear locking, defining three degrees of freedom .ı/K at each
which is described below. node. If bilinear shape functions

Finite Element Formulation 1


NK D .1 C K / .1 C K / (73)
The simplest option to formulate finite elements 4
for the Mindlin plate model is to use Lagrange
are used, this results in a four-node element
shape functions NK for the formulation of
(nnode D 4) with a total of 12 degrees of freedom.
Finite Elements for Plates and Shells 15

In Eq. (73),  and  are the coordinates in the number of Gauss points is equal to the number
parametric space of the element, with 1  of nodes. Their performance, however, is strongly
f; g  1 and K ; K denote the nodal coordi- deteriorated by the problem of transverse shear
nates in this local coordinate system. locking, described below, particularly when low-
The discretized kinematic equations in Voigt order polynomials are used. In the context of
notation IGA, B-splines or NURBS are often used as
shape functions. The overall structure of the finite
2 3K 2 3
11 0 @1 NK 0 element formulation, however, is the same.
2 3K
6 22 7 6 0 0 @2 NK 7 F
6 7 n
Xnode
6 7 U3
6212 7 D 6 0 @2 NK @1 NK 74 1 5 Remarks on Modeling and Discretization
6 7 6 7
4 1 5 KD1 4@1 NK NK 0 5 2 In contrast to Kirchhoff plate theory, applying
2 @2 NK 0 NK proper boundary conditions does not involve par-
„ ƒ‚ … ticular challenges from a mathematical point of
BK
(74) view in shear deformable plate models, because
defines the nodal strain displacement matrix BK . the order of the governing differential equations
The element strain displacement matrix, refer- matches the number of degrees of freedom. How-
ring to all degrees of freedom, is defined as ever, boundary layers occurring in certain sit-
B D ŒB1 B2 B3 : : : Bnnode . Along with the ma- uations require special attention. In a nutshell,
terial matrix C in Voigt notation and the Jacobian boundary layers have to be expected at simply
matrix J, mapping element coordinates to the supported and free edges but not at clamped
global Cartesian coordinate system edges. Arnold and Falk (1989) provide an in-
depth study of this issue.
2 3
D D 0 0 0 For example, for a simply supported edge par-
6D D 0 0 0 7 allel to the x1 -direction, the essential and natural
6 7
CD6
6 0 0 D.1  / 0 0 7;
7
boundary conditions are U3 D 0, M22 D 0,
4 0 0 0 ˛s Gh 0 5 and M12 D 0. Both rotations ˛ are free (non-
0 0 0 0 ˛s Gh zero). In the Kirchhoff model, 1 D @1 U3 is
  zero, because U3 D 0, and thus M12 is non-zero.
@ x @ x
JD  1  1 ; (75) Within the domain, the Kirchhoff solution and the
@ x2 @ x2 Mindlin solution are approximately equal, with
the difference diminishing with thickness h going
the element stiffness matrix can be computed as to zero. At the boundary, however, there is a sig-
nificant difference: while in the Kirchhoff model,
Z1
large twisting moments M12 occur, those have
kD BT CB jJj dd : (76)
to be zero in the Mindlin model. Therefore, M12
1 varies from large values near the boundary to zero
The strain displacement matrix B contains deriva- at the boundary in a very narrow region, the so-
tives of the shape functions with respect to Carte- called boundary layer. Due to equilibrium, in the
sian coordinates, which have to be computed via same region, transverse shear forces build up. In
the chain rule fact, these are equivalent to the extra contribution
in the effective shear forces introduced in Eq. (30)
@˛ NK D @ NK @˛  C @ NK @˛  ; (77) for the Kirchhoff model to solve the edge prob-
lem. The point forces in the corner, showing up in
where @˛  and @˛  are the components of the the Kirchhoff theory, comply with concentrated
inverse of the Jacobian. transverse shear forces in the Mindlin model. For
For the Lagrangian family, Gauss integration h ! 0, these become singular.
provides a stable and consistent element if the In order to resolve the high gradient of the
solution, it is practical to introduce a small layer
16 Finite Elements for Plates and Shells

of narrow elements along edges with boundary locking in Kirchhoff-Love shell elements, the
layers. The high aspect ratios of the elements following can be observed:
usually do not introduce problems, although com-
mercial finite element codes might issue “shape 1. Displacements and rotations are underes-
warnings” in such cases. The size of the boundary timated, and the pre-asymptotic order of
layer is proportional to the thickness, such that for convergence is below the optimal one. As a
very thin plates, it is extremely small. result, also the bending and twisting moments
Boundary layers can be avoided by introduc- are underestimated.
ing the artificial boundary condition ˛ D 0 2. The error depends on the thickness and in-
at an edge parallel to x˛ , that is, fixing the creases with decreasing thickness. More pre-
rotation normal to the edge. This is called hard cisely, the length-to-thickness ratio of the indi-
support in the literature, whereas the original, vidual finite elements is the crucial parameter.
physically more realistic version is denoted as As a rule of thumb, it can be said that for
soft support. The hard support option is of mostly the standard displacement elements described
academic interest. First of all, it mimics the artifi- above, reasonable results cannot be expected
cial Kirchhoff constraint at the edge and allows a unless the element size is smaller than the
direct comparison of solutions, either analytical thickness. For plates with high aspect ratios,
or numerical. Moreover, closed-form solutions extremely fine meshes are required, resulting
can be easily found for some special cases with in prohibitive numerical expense.
hard support, whereas exact analytical solutions 3. Transverse shear forces show strong oscilla-
for problems with soft support are rare. tions, with the maximum and minimum values
Another important issue in shear deformable being much larger than the exact ones.
plate models is the singularity with respect to
point forces. In contrast to Kirchhoff theory, the Like in the case of membrane locking, the ori-
exact solution for the displacement under a point gin of transverse shear locking can be explained
force is infinity. For the same reason, the stiffness from the kinematic equations, exemplified here
of a point support is zero. For example, a uni- for elements with bilinear shape functions. For
formly loaded plate supported by single bearings a case of pure bending, transverse shear forces
at the corners has infinite displacements in the are zero, and thus also the shear angles have
entire domain. The singularity is rather weak, to vanish, ˛ D 0. This condition also applies
such that fine meshes are required to approxi- for arbitrary load cases in the thin limit (the
mately resolve it. However, quantifying displace- “Kirchhoff limit”). From kinematic Eq. (60), it
ments in such situations is pointless, because follows
any finite value is wrong. This is an issue also
when combining beam elements and plate (or ˛ D ˛ C@˛ U3 D 0 ) ˛ D @˛ U3 ;
shell) elements in industrial applications. A plate (78)
supported by columns that are modeled with which is identical to the normality constraint
beam elements shows the same singularities as from the Kirchhoff model. If linear shape func-
described above, and the computational results tions are used, @˛ U3 is constant, and thus ˛
have to be interpreted with care. also has to be constant in order to comply with the
constraint. From the definition of the curvature
Transverse Shear Locking and Methods to in (59), it then follows
Avoid It
The most important issue in the context of FEM 1
˛ˇ D @ˇ ˛ C@˛ ˇ D 0; (79)
for shear deformable plates (and shells) is the 2 „ƒ‚… „ƒ‚…
Dconst: Dconst:
phenomenon of transverse shear locking. Along
the lines of the three typical symptoms of locking meaning that the curvature is identically zero,
outlined above in the section about membrane effectively excluding bending and twisting. Zero
Finite Elements for Plates and Shells 17

transverse shear strains thus imply zero moments which the parasitic part is zero. For the four-node
in the thin limit. Differential Eq. (65) and (66) elements, these are the centers of the element
illustrate why the problem gets worse with de- edges, such that
creasing thickness. The coefficient of the last
term ˛sDGh tends to infinity with h ! 0, and thus A
 D B .0; 1/d ; B
 D B .0; 1/d ; (81)
the parasitic, oscillating part of the transverse C D
shear strain dominates the entire solution.  D B .1; 0/d ;  D B .1; 0/d : (82)
For elements with higher-order shape func-
tions, this constraint is relaxed. Mindlin plate Here, d is the element displacement vector, and F
elements with quadratic shape functions are able B .; / and B .; / are the row vectors within
to exactly represent constant bending and twist- the strain displacement matrix B related to the
ing moments. However, transverse shear force transverse shear part, transformed into the ele-
oscillations still occur in general applications, ment coordinate system according to Eq. (80).
and the quality of the results also depends on the Modified transverse shear strains are then ob-
slenderness. This means that also these elements tained from linear interpolation of the strain com-
are not free from locking. Selective reduced in- ponents at the sampling points within the element
tegration of transverse shear is a simple option to 1 1
ANS A B
remove locking, but it needs stabilization to avoid  D .1  /  C .1 C /  ;
2 2
zero energy modes.
The literature on methods to avoid transverse ANS 1 C 1 D
 D .1  /  C .1 C /  : (83)
shear locking in plate and shell finite elements 2 2
is extremely rich and cannot be reviewed here.
Thus, ANS is constant in  by construction and
The focus is put onto the assumed natural strain
the parasitic part is removed. The same applies
(ANS) method, which is one of the most suc-
analogously for ANS . Introducing (81) and (82)
cessful and popular remedies. Inspired by the
into (83) defines modified entries
QUAD4 element of MacNeal (1978), it has been
proposed by Hughes and Tezduyar (1981). The 1 1
successful MITC (mixed interpolation of tenso- BN  D .1/B .0; 1/ C .1C/B .0; 1/ ;
2 2
rial components) family of finite elements (see
(84)
Bathe 2006 for a comprehensive overview) fol-
lows a similar concept. It is presented here in a 1
BN  D .1  /B .1; 0/
simplified and digested form that allows straight- 2
forward implementation. 1
C .1 C /B .1; 0/ ; (85)
In order to ensure optimal performance for 2
arbitrary finite element meshes, involving non-
rectangular element shapes, transverse shear for the strain displacement matrix. The modified
strains are expressed in the natural element strain displacement matrix BN contains BN  and
coordinate system, BN  , transformed back into the Cartesian coordi-
nate system, as well as the unmodified entries for
D the curvatures. The stiffness matrix can then be
 1 @ x1 C 2 @ x2 ;
computed in the usual way via Eq. (76).
 D 1 @ x1 C 2 @ x2 : (80) It is one of the elegant features of the ANS
method that relies on the standard nodal displace-
Transverse shear locking can be avoided by re- ment and rotation degrees of freedom and does
moving the parasitic linear parts of the trans- not introduce any additional unknowns. Defini-
verse shear strains that are responsible for the tions of boundary conditions, consistent nodal
oscillations. The idea is to collocate the natural forces for the loads, as well as the assembly
strain components at specific sampling points, at
18 Finite Elements for Plates and Shells

procedure are the same as in the standard dis- of the element shape functions, obtained with
placement finite element formulation. mesh refinement?
• (How) does the solution deteriorate when us-
Benchmarking ing non-rectangular element shapes (distorted
The uniaxial bending test described as patch test meshes)?
for Kirchhoff plate theory above is also useful • (How) does accuracy for a given mesh depend
for Mindlin plate finite elements. However, pro- on the slenderness?
viding the exact result is not a requirement for
consistency. Thus, strictly speaking this is not a It is important to take into account not only dis-
“patch test” for Mindlin plate finite elements. The placements but also bending moments and shear
patch test as a numerical consistency test requires forces in the above considerations. While exam-
exact representation of the primal variables up ining the sensitivity to mesh distortions, trape-
to the order of the variational index nv . This zoidal element shapes turn out to be particularly
means that problems in which U3 and/or ˛ are demanding according to the author’s experience,
linear have to be represented exactly to ensure (Bischoff 1999). A popular benchmark for plate
consistency. These are the rigid body modes and elements that naturally involves non-rectangular
constant shear modes. Pure bending modes in- elements is a skew plate under uniform load,
volve quadratic displacement distributions, and often referred to as the Morley plate problem
their exact representation is thus not needed for (Morley 1962).
consistency. Patch tests for pure shear defor- Particularly for finite elements relying on re-
mation with zero moments can be realized by duced integration, it is important to check for
applying distributed moments. Both the standard artificial zero energy modes. This can be ac-
displacement element and the ANS element, de- complished by verifying the correct rank of the
scribed above, pass these tests. system stiffness matrix by an eigenvalue analysis.
More important, however, are bending tests to If the finite element program does not include
assess element performance in view of transverse this feature, a plate on point supports, subject to
shear locking. A Mindlin plate finite element that a point force, which triggers the corresponding
fails the uniaxial bending test may well be con- modes, can be computed. Otherwise, benchmarks
vergent, but it will not provide satisfactory perfor- with point loads and point supports are not rec-
mance. The ANS element provides exact results ommended, because of the related singularities.
in this case even for distorted meshes, but the The exact solution for the displacement under a
standard displacement element fails already for point force is infinity in the Mindlin plate model.
regular meshes with rectangular element shapes, For the same reason, the stiffness of a point
because of transverse shear locking. support is zero.
Analytical solutions for the academic case of
rectangular plates with hard support and also Reissner-Mindlin Shell Model
for circular plates can be found, for instance, in
Reddy (1999). They can be used to assess the Kinematic Equations
performance of Mindlin plate elements. While Like in the section on the Kirchhoff-Love shell
doing so, the following questions can be used as model, also here the focus is on the kinematic
a guideline to compare different finite element equations, which are the most important ones
formulations: for discussing finite element formulations and
locking. There is no such thing as “the” Reissner-
• Which element size is required to obtain a Mindlin shell theory, but numerous different
predefined accuracy? formulations exist, mostly distinguished by the
• Is the correct order of convergence, which particular kinematic assumptions and other
depends on the error norm used and the order approximations. They are also sometimes
Finite Elements for Plates and Shells 19

denoted as first-order shear deformation theories 1


E˛3 D .a˛  a3  A˛  A3 /
(FOSDT). For finite element formulations, 2 „ ƒ‚ …
however, these differences are only of secondary D0

importance. It is thus sufficient here to regard 1


D .A˛  a3 C @˛ v  a3 /
shear deformable shell theories as an extension 2
of shear deformable plate theory to curved 1
structures, augmented by a membrane part. D .A˛  ˚  A3 C @˛ v  a3 / : (89)
2
The governing equations of a shear de-
formable, Reissner-Mindlin-type shell formu- The linear part in  3 of the transverse shear strain F
lation can be derived from the ones of Kirchhoff- is zero.
Love shell theory by introducing rotations of Assuming a constant metric through the thick-
the director as independent variables. With the ness (Love’s first approximation) is no longer
definition of a rotation tensor ˚ , the director in valid for moderately thick shells. One way to deal
the deformed configuration is obtained as with this problem is to define so-called effective
stress resultants, referring to the metric of the
a3 D ˚  A3 : (86) mid-surface (see Simo and Fox (1989), Bischoff
et al. (2004) for details).
This replaces the construction of a3 from the nor-
mality assumption (35)2 in the Kirchhoff-Love Finite Element Formulation
model. The nontrivial parametrization of ˚ in the Formulation of Reissner-Mindlin shell elements
presence of large rotations, for instance, using is technically much simpler than formulation
Euler angles, Rodrigues parameters, or quater- of Kirchhoff-Love shell elements. Kinematic
nions, is not further discussed here. Eqs. (87), (88), and (89) contain only first
Like in the Kirchhoff-Love model, the strains derivatives of displacements and rotations, and
are split into a membrane part and a bending part, thus the variational index is nv D 1, allowing for
cf. Eq. (39). The expression for the membrane C 0 -continuous shape functions. This simplicity,
strain components however, comes along with the problem of
transverse shear locking. The last expression
0 1  in (89) reveals the same source of transverse
E˛ˇ D @˛ v  Aˇ C @ˇ v  A˛ C @˛ v  @ˇ v
2 shear locking that could already be identified
1  in Eq. (60) in the Mindlin plate model. While
D a˛ˇ  A˛ˇ (87)
2 the displacement is subject to a derivative, the
rotation is not, resulting in an unbalance of the
is identical to the ones from Kirchhoff-Love shell function spaces in the representation of E˛3 .
theory, Eq. (40). The curvatures In addition, the problem of membrane locking,
already discussed in the context of Kirchhoff-
˛ˇ D @˛ aˇ  a3 C B˛ˇ (88) Love shell finite elements, also exists in shear
deformable shell elements.
can formally also be obtained from the expression Historically, there are two apparently com-
derived in the context of Kirchhoff-Love shell peting strategies to obtain shell finite elements
theory, Eq. (41), but they are not identical. The including transverse shear effects:
first expression is not equal to the first fundamen-
tal form, @˛ aˇ a3 ¤ b˛ˇ because a3 is not normal 1. Directly discretizing the weak form obtained
to the deformed surface. As a consequence, the from a given shell theory
material curvature differs from the geometric 2. Applying the shell assumptions and dimen-
curvature ˛ˇ ¤ K˛ˇ , cf. Eq. (47). Finally, the sional reduction to a given three-dimensional
transverse shear strain components are given as discretization of the shell continuum, using
linear shape functions through the thickness
20 Finite Elements for Plates and Shells

While the first option seems straightforward, it theories, it can help to avoid locking already
is frequently regarded as more complicated and on the level of their mathematical formulation.
mathematically more demanding than the second Corresponding formulations hierarchically add
one. This second possibility has first been pro- the transverse shear contribution to the Kirchhoff-
posed by Ahmad et al. (1968). The idea is known Love equations. The resulting elements are then
as concept of degeneration or – somewhat less intrinsically free from transverse shear locking
discriminating – continuum-based approach (not (see Long et al. (2012) for a formulation using
to be confused with continuum shell elements, subdivision surfaces, as well as Echter et al.
see below). Although the technical derivation (2013), and Oesterle et al. (2016) in the context
significantly differs between both cases, the re- of IGA). An equally elegant method to remove
sulting finite elements are identical, if identical membrane locking is still unknown.
assumptions are made and the same finite ele-
ment technology is applied, (Büchter and Ramm Remarks on Modeling and Discretization
1992b). For Reissner-Mindlin shell elements, the same
Reissner-Mindlin-type shell elements can be remarks apply as for Mindlin plate elements,
formulated as an extension of Mindlin plate el- particularly regarding boundary layers and the
ements to curved geometry, supplemented by singularities associated with point forces and
a membrane part. Concepts to avoid transverse point supports. Moreover, the curved geometry
shear locking, described above in the context of introduces an additional issue. C 0 -continuous
Mindlin plates, can be combined with concepts discretization naturally involves non-smooth
to avoid membrane locking, already addressed in representation of the shell geometry. Conse-
the context of Kirchhoff-Love shell elements. A quently, the normal direction and thus definition
popular combination, realized in many scientific of the director A3 are not unique at the nodes.
and commercial finite element codes, is to use the The rotation tensor ˚ used for the update of
ANS (or MITC) method to avoid transverse shear the normal in Eq. (86) contains two independent
locking and the EAS method to alleviate mem- rotations about axes lying in the tangent plane
brane locking, (Andelfinger and Ramm 1993). to the shell’s mid-surface. As the normal is not
The ANS method can also be used to remove unique, there is also no distinct tangent plane
membrane locking. However, in the case of four- at the nodes. Coupling of rotational degrees of
node elements, these do not pass the patch test freedom between adjacent elements is therefore
in the membrane part. Another popular method not well defined.
that can be used to avoid both transverse shear One option to resolve this problem is to use
locking and membrane locking is the discrete the average of the element normals as director
strain gap (DSG) method, (Bletzinger et al. 2000; and to define the tangent plane as the normal
Koschnick et al. 2005), which exhibits compara- plane to this vector. However, in practical ap-
ble performance to the MITC family of elements. plications, also modeling of folded plates, shells
A very successful shear deformable shell ele- with discrete kinks, and shells with intersections
ment based on reduced integration and stabiliza- is required. Therefore, in commercial finite ele-
tion has been proposed by Belytschko and Tsay ment codes, tailored for industrial applications,
(1981). Its coarse mesh accuracy is not very good, an additional, third rotation is introduced, which
but due to the technology of reduced integration resolves the problem. For C 1 -continuous dis-
and its specific algorithmic setup, it is extremely cretizations or the special case of flat plates,
efficient and fast. The Belytschko-Tsay element this rotation complies with a rotation of the di-
(and modified versions of it) is the work horse in rector about its axes, which is not associated
explicit dynamic analyses, for instance, for crash with any deformation. Consequently, no physi-
simulations in automotive industry. cal stiffness can be assigned to this “drilling”
Although C 1 -continuous discretization is not rotation. Assigning zero stiffness to this non-
required for the case of shear deformable shell physical rotational degree of freedom may result
Finite Elements for Plates and Shells 21

in artificial zero energy modes. A small penalty of Mindlin plates, this is not recommendable
stiffness is often applied to remove this problem. because of the involved singularities.
Alternatively, so-called Allman rotations, (All-
man 1984), see also Bergan and Felippa (1985),
Hughes and Brezzi (1989), can be used. They Three-Dimensional and Solid Shell
link the rotations to in-plane displacements via Finite Elements
particular shape functions.
It is important to realize that, regardless of Since the beginning of the 1990s, three-
the specific technology used to introduce drilling dimensional shell formulations, including F
rotations, the associated stiffness is essentially transverse normal strain and stress, entered the
nonphysical. With mesh refinement, it tends to stage, (Simo et al. 1990; Büchter and Ramm
zero. If, as an example, rotational degrees of free- 1992a; Sansour 1996). The corresponding
dom of a beam element, connected with the shell elements typically involve linearly varying
element, are coupled with this drilling rotation, transverse normal strains and thus allow for
the associated connecting stiffness is wrong. implementation of arbitrary three-dimensional
constitutive laws without any modifications,
Benchmarking needed in classical shell formulations to enforce
The benchmark problems proposed above for the zero transverse normal stress condition,
Mindlin plate elements and Kirchhoff-Love shell cf. Eq. (9). Still, they rely on the concept
elements are also sensible for Reissner-Mindlin of dimensional reduction, i.e., all degrees of
shell elements. Analyzing bending of flat plates freedom are located on the mid-surface.
effectively isolates the effect of transverse shear This steps back into three dimensions giving
locking and allows to study the corresponding rise to an additional locking phenomenon, associ-
element performance separately from membrane ated with parasitic transverse normal stress in the
locking. The cylindrical shell strip (Fig. 4) can case of pure bending of curved structures. It has
be used as a test for membrane locking. Three- been first described in Ramm et al. (1994); later
node triangles are free from membrane locking, the term curvature thickness locking has been
because they are always flat. Ironically, six-node coined in Bischoff and Ramm (1997). It is closely
triangles with quadratic shape functions therefore related to the phenomenon of trapezoidal locking
often perform much worse in this case than linear in the context of two-dimensional and three-
elements, provided these are free from transverse dimensional solid elements. A method to avoid
shear locking. Four-node elements will perform this locking effect, based on the ANS concept,
well for a regular mesh, because in this case they has been proposed in parallel by Betsch et al.
are also flat. Therefore, a distorted mesh with (1996) and Bischoff and Ramm (1997). It can
warped element geometry must be used to assess also be removed using the DSG method, which is
the tendency of an element to membrane locking. a general method to avoid all geometric locking
Out of the multitude of benchmark problems effects, (Bischoff 1999).
proposed in the literature, the Scordelis-Lo roof As an alternative, three-dimensional finite
problem, (Scordelis and Lo 1964), ought to be elements can be formulated in a way that makes
mentioned here for two reasons. First, reference them suitable for the analysis of both thick
solutions can be found in many papers, and sec- and thin shells, (Parisch 1995; Hauptmann and
ond, it is a well-suited problem not only for Schweizerhof 1998). Already earlier, Schoop
investigating shear and membrane locking but (1986) had described a corresponding surface-
also to test the ability of a given finite element to oriented shell theory, anticipating this idea.
represent complex states of membrane strain and Formulation of these solid shell finite elements
stress. Many popular benchmark problems, such (also sometimes denoted as continuum shell
as the pinched cylinder and pinched hemisphere elements or surface-oriented shell elements)
problem, involve point forces. Like in the case requires in particular the implementation of
22 Finite Elements for Plates and Shells

suitable methods to remove transverse shear References


locking, membrane locking, and curvature
thickness locking and to ensure convergence Adini A, Clough RW (1961) Analysis of plate bending
to the shell solution in the thin limit. These by the finite element method. Report to the National
Science Foundation, USA, G7337
technologies are applied in an anisotropic Ahmad D, Irons B, Zienkiewicz O (1968) Curved thick
manner, specifically assigning in-plane and shell and membrane elements with particular reference
transverse directions to the solid finite element. to axi-symmetric problems. In: Berke L, Bader RM,
Solid shell elements can therefore be regarded Mykytow WJ, Przemieniecki JS, Shirk MH (eds) Pro-
ceedings of the 2nd conference on matrix methods in
as three-dimensional shell elements with the structural mechanics
nodes located at the upper and lower surface Allman DJ (1984) A compatible triangular element in-
of the shell instead of its mid-surface. They are cluding vertex rotations for plane elasticity analyses.
Comput Struct 19:1–8
available in many commercial finite element
Andelfinger U, Ramm E (1993) EAS-elements for two-
codes today. There are several advantages over dimensional, three-dimensional, plate and shell struc-
three-dimensional shell elements relying on a tures and their equivalence to HR-elements. Int J Nu-
mid-surface description. Solid shells do not mer Methods Eng 36:1311–1337
Argyris JH, Scharpf DW (1968) The SHEBA family of
involve rotational degrees of freedom, and
shell elements for the matrix displacement method.
they provide flexibility in modeling of complex Aeronaut J 72:873–883
structures involving both thin and thick regions. Arnold D, Falk R (1989) Edge effects in the Reissner-
Transition to standard solid elements for bulk Mindlin plate theory. In: Noor A, Belytschko T, Simo
J (eds) Analytical and computational models for shells.
regions is straightforward. In the context of
ASME, New York, pp 71–90
explicit dynamics, the large transverse normal Bathe K (2006) Finite element procedures. Prentice Hall,
stiffness has a negative effect on the critical time Englewood Cliffs
step. This problem can, for instance, be tackled Bathe KJ, Ho LW (1981) A simple and effective element
for analysis of general shell structures. Comput Struct
by appropriate mass scaling, Confalonieri et al.
13:673–681
(2015). Batoz JL (1980) A study of three-node triangular plate
bending elements. Int J Numer Methods Eng 15:1771–
1812
Batoz JL (1982) An explicit formulation for an efficient
Cross-References triangular plate bending element. Int J Numer Methods
Eng 18:1077–1089
 3D Derivations of Static Plate Theories Belytschko TB, Tsay CS (1981) Explicit algorithms for
 Classical Plate Problems nonlinear dynamics of shells. In: Hughes TJR, Pifko A,
Jay A (eds) Nonlinear finite element analysis of plates
 Direct Derivation of Plate Theories
and shells, AMD-48. ASME, pp 209–231
 Elastic Shells, Linear Shear-Deformable Bergan PG, Felippa CA (1985) A triangular membrane
Theory element with rotational degrees of freedom. Comput
 Elastic Shells, Resultant Non-linear Theory Methods Appl Mech Eng 50:25–69
Betsch P, Gruttmann F, Stein E (1996) A 4-node finite
 Finite Element Methods
shell element for the implementation of general hyper-
 Isogeometric Finite Element Analysis elastic 3D-elasticity at finite strains. Comput Methods
 Laminated Composite Elastic Shells, Non-lin- Appl Mech Eng 130:57–79
ear Theory Bischoff M (1999) Theorie und Numerik einer dreidimen-
sionalen Schalenformulierung. Ph.D. thesis, Institut für
 Limit Analysis of Plates
Baustatik, Universität Stuttgart
 Mathematical Justifications of Plate Models Bischoff M, Ramm R (1997) Shear deformable shell
 On Treatment of Finite Rotations in FEM Anal- elements for large strains and rotations. Int J Numer
yses of Irregular Shell Structures Methods Eng 40:4427–4449
Bischoff M, Bletzinger KU, Wall WA, Ramm E (2004)
 Thin Elastic Shells, Lagrangian Geometrically Models and finite elements for thin-walled structures.
Non-linear Theory In: Encyclopedia of computational mechanics. Wiley,
 Thin Elastic Shells, Linear Theory Chichester, pp 59–137
Bletzinger KU, Bischoff M, Ramm E (2000) A unified
approach for shear-locking-free triangular and rect-
Finite Elements for Plates and Shells 23

angular shell finite elements. Comput Struct 75(3): brane locking. Comput Methods Appl Mech Eng 194:
321–334 2444–2463
Bogner FK, Fox RL, Schmit LA Jr (1965) The generation Long Q, Bornemann B, Cirak F (2012) Shear-flexible
of interelement compatible stiffness and mass matrices subdivision shells. Int J Numer Methods Eng 90:
by the use of interpolation formulas. In: Bader R et al 1549–1577
(ed) Proceeding of 1st conference on matrix methods in Love AEH (1888) On the small vibrations and deforma-
structural mechanics, Airforce Institute of Technology, tions of thin elastic shells. Philos Trans R Soc 179:
Dayton, pp 397–444 491–546
Büchter N, Ramm E (1992a) 3D-extension of nonlinear MacNeal R (1978) A simple quadrilateral shell element.
shell equations based on the enhanced assumed strain Comput Struct 8:175–183
concept. In: Hirsch C (ed) Computational methods Melosh RJ (1961) A stiffness matrix for the analysis of F
in applied sciences. Elsevier, Amsterdam/New York, thin plates in bending. J Aerosp Sci 28:34–40
pp 39–59 Mindlin RD (1951) Influence of rotatory inertia and shear
Büchter N, Ramm E (1992b) Shell theory versus degen- on flexural motions of isotropic elastic plates. J Appl
eration – a comparison in large rotation finite element Mech 18:31–38
analysis. Int J Numer Methods Eng 37:55–62 Morley LSD (1962) Bending of a simply supported rhom-
Cirak F, Ortiz M, Schröder P (2000) Subdivision surfaces: bic plate under uniform normal loading. Q J Mech Appl
a new paradigm for thin-shell finite-element analysis. Math 15:413–426
Int J Numer Methods Eng 47:2039–2072 Oesterle B, Ramm E, Bischoff M (2016) A shear de-
Clough RW, Tocher JL (1965) Finite element stiffness formable, rotation-free isogeometric shell formulation.
matrices for analysis of plate bending. In: Proceedings Comput Methods Appl Mech Eng 307:235–255
of the conference on matrix methods in structural Parisch H (1995) A continuum-based shell theory for
mechanics, WPAFB, Ohio, pp 515–545 non-linear applications. Int J Numer Methods Eng 38:
Confalonieri F, Ghisi A, Perego U (2015) 8-node solid 1855–1883
shell elements selective mass scaling for explicit dy- Ramm E, Bischoff M, Braun M (1994) Higher order
namic analysis of layered thin-walled structures. Com- nonlinear shell formulation – a step back into three
put Mech 56:585–599 dimensions. In: Bell K (ed) From finite elements
Dhatt G (1970) An efficient triangular shell element. to the Ttroll platform. Ivar Holand 70th Anniver-
AIAA J 8:2100–2102 sary, Norwegian Institute of Technology, Trondheim,
Echter R, Oesterle B, Bischoff M (2013) A hierarchic pp 65–88
family of isogeometric shell finite elements. Comput Reddy JN (1999) Theory and analysis of elastic plates.
Methods Appl Mech Eng 254:170–180 Taylor and Francis, Boca Raton
Hauptmann R, Schweizerhof K (1998) A systematic de- Reissner E (1945) The effect of transverse shear defor-
velopment of ‘solid-shell’ element formulations for mation on the bending of elastic plates. J Appl Mech
linear and non-linear analyses employing only dis- 12:69–77
placement degrees of freedom. Int J Numer Methods Sansour C (1996) A theory and finite element formulation
Eng 42:49–69 of shells at finite deformations involving thickness
Höllig K (2003) Finite element methods with B-splines. change. Arch Appl Mech 65:194–216
Frontiers in Applied Mathematics, vol 26. SIAM, Schoop H (1986) Oberflächenorientierte Schalentheo-
Philadelphia rien endlicher Verschiebungen. Ingenieur-Archiv 56:
Hughes TJR, Brezzi F (1989) On drilling degrees 427–437
of freedom. Comput Methods Appl Mech Eng 72: Scordelis AC, Lo KS (1964) Computer analysis of cylin-
105–121 drical shells. J Am Concr Inst 61:539–562
Hughes TJR, Tezduyar TE (1981) Finite elements based Simo JC, Fox DD (1989) On a stress resultant geo-
upon Mindlin plate theory with particular reference to metrically exact shell model. Part I: formulation and
the four-node bilinear isoparametric element. J Appl optimal parametrization. Comput Methods Appl Mech
Mech 48:587–596 Eng 72:267–304
Hughes TJR, Cottrel JA, Bazilevs Y (2005) Isogeometric Simo JC, Rifai MS (1990) A class of mixed assumed strain
analysis: CAD, finite elements, NURBS, exact geome- methods and the method of incompatible modes. Int J
try and mesh refinement. Comput Methods Appl Mech Numer Methods Eng 29:1595–1638
Eng 194:4135–4195 Simo JC, Rifai S, Fox D (1990) On a stress resultant
Kiendl J, Bletzinger KU, Linhard J, Wüchner R (2009) geometrically exact shell model. Part IV: variable
Isogeometric shell analysis with Kirchhoff-Love el- thickness shells with through-the-thickness stretching.
ements. Comput Methods Appl Mech Eng 198: Comput Methods Appl Mech Eng 81:91–126
3902–3914 Stricklin JA, Haisler W, Tisdale P, Gunderson R (1969)
Kirchhoff G (1850) Über das Gleichgewicht und die A rapidly converging triangular plate element. AIAA J
Bewegung einer elastischen Scheibe. J für die reine und 7:180–181
angewandte Mathematik 40:51–88 Taylor RL, Beresford PJ, Wilson EL (1976) A non-
Koschnick F, Bischoff M, Camprubí N, Bletzinger KU conforming element for stress analysis. Int J Numer
(2005) The discrete strain gap method and mem- Methods Eng 10:1211–1219

Das könnte Ihnen auch gefallen