Dieter Dinkler

Ursula Kowalsky

Introduction to
Finite Element Methods
Introduction to Finite Element Methods
Dieter Dinkler · Ursula Kowalsky

Introduction to
Finite Element Methods
Dieter Dinkler Ursula Kowalsky
Institut für Statik und Dynamik Institut für Statik und Dynamik
Technische Universität Braunschweig Technische Universität Braunschweig
Braunschweig, Germany Braunschweig, Germany

ISBN 978-3-658-42741-2 ISBN 978-3-658-42742-9 (eBook)

https://doi.org/10.1007/978-3-658-42742-9

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Fachmedien
Wiesbaden GmbH, part of Springer Nature 2024

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher,
whether the whole or part of the material is concerned, specifically the rights of translation,
reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any
other physical way, and transmission or information storage and retrieval, electronic adaptation,
computer software, or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are
exempt from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in
this book are believed to be true and accurate at the date of publication. Neither the publisher
nor the authors or the editors give a warranty, expressed or implied, with respect to the material
contained herein or for any errors or omissions that may have been made. The publisher remains
neutral with regard to jurisdictional claims in published maps and institutional affiliations.

This Springer Vieweg imprint is published by the registered company Springer Fachmedien
Wiesbaden GmbH, part of Springer Nature.
The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany

Paper in this product is recyclable.

Preface

The development and design of technical systems need a deep understanding of

processes and phenomena to be described, if an analysis with respect to safety,
production effort and resources is required. Nowadays, numerical investigations
are at least of the same quality as experimental studies, but more efficient, if
parametrical studies have to be accomplished or structures of high complexity
have to be optimized. The Finite Element Method is one of the most important
tools for engineers to numerically investigate and design structures and all
kinds of components exposed to constant or time–varying external actions and
excitations. Thus it is an essential need to teach students the fundamentals of
the method and to introduce them to the possibilities of applications and the
variety of different element formulations and their respective advantages.

The textbook presented here gives an introduction to the Finite Element Me-
thod from an engineering point of view and offers various insights into the
background and the details of the procedures to get finite elements of different
characteristics, types of discretization, and the related methodological approa-
ches. It is not the intention to present the most efficient types of elements,
because the efficiency of elements depends on the criteria of evaluation, the
kind of programming, and very often on their applications.

Since the number of publications on the Finite Element Method extraordina-

rily increased over the last decades, the textbook is not able to discuss and
to refer to all elements and all applications presented in the literature, but it
is limited to heat conduction, membrane and bending structures of one and
two dimensions. The extension to three dimensional structures is possible with
little effort, if the presented methods are analogously extended to the third di-
mension. Furthermore, the representation does not deal with nonlinearities of
different types and time–varying deformation behavior, since this would need
further mathematical and mechanical background, and a deep insight into the
phenomena of geometric nonlinear deformation behavior of structures, plastic
deformation of materials and other physical processes, what is beyond the ac-
tual Finite Element Method.
VI

The manuscript of this textbook has gradually evolved from lecture notes pre-
sented by the authors at the Technical University Braunschweig and Stuttgart
University to students from the engineering faculties. Thus the textbook is ad-
dressed to students and engineers, who are dealing with structural mechanics
and are interested to develop elements for their own applications. The authors
like to remind of Hermann Ahrens and to gratefully acknowledge his longtime
commitment in lecturing the fundamentals of the method and his support to
develop this kind of presentation of the Finite Element Method.
Finally we gratefully acknowledge Springer Vieweg publishers for the possibility
to publish this monograph and for the great support until going to press.

Dieter Dinkler and Ursula Kowalsky

 VII

Glossary

Physical Description of Processes

δWd = 0 Principle of virtual Displacements
δWσ = 0 Principle of virtual Forces
( )d displacement
( )σ stress
( )e external action
x, y, z physical cartesian coordinates
s boundary coordinate
( )n normal direction at the boundary
( )t tangential direction at the boundary
ξ, η, ζ normalized element coordinates
λa , λb , λc area–coordinates within a triangle
p vector of external actions – loading, heat sources
u vector of field variables – displacements, temperature
ε vector of strain variables
σ vector of stress variables
E matrix of physical properties – elasticity, thermal conductivity

Matrix– and Vector–Symbols at the Element Level

Ω matrix of shape functions
φi shape functions
v element vector of discrete unknowns – nodal unknowns
s element vector of discrete stress variables – nodal stresses
f element vector of discretized actions – nodal loading, heat sources
D matrix of differentiation rules due to the kinematic conditions
E matrix of physical properties – elasticity, thermal conductivity
B element matrix of strains
K element matrix – stiffness, heat resistance
F element matrix – flexibility
S element matrix of subsequent stress analysis

Matrix– and Vector–Symbols at the System Level

K system matrix – stiffness, heat resistance
v system vector of discrete unknowns – nodal unknowns
f system vector of discretized actions – nodal loads, heat sources
VIII

Denomination of Elements

The denomination of the elements is necessary in order to distinguish the ele-

ments with respect to their properties. To avoid misunderstandings the text-
book by hand uses the following abbreviations:

K Kirchhoff plate theory

RM Reissner–Mindlin plate theory
Θ temperature–based formulation
D displacement–based formulation
M mixed formulation
HD hybrid displacement formulation
HM hybrid mixed formulation
Q quadrilateral element
T triangular element
N-M element with N nodes and M degrees of freedom
RI reduced integration
SRI selectively reduced integration

Heat Conduction Elements

HC-ΘQ-4 temperature–based fomulation, bi–linear shape functions
HC-ΘT-3 temperature–based fomulation, linear shape functions

Plane Stress Elements

P-DQ-4 displacement–based fomulation, bi–linear shape functions
P-DQ-4-SRI displacement–based fomulation, bi–linear shape functions
P-DQ-9 displacement–based fomulation, bi–quadratic shape functions
P-DQ-9-SRI displacement–based fomulation, bi–quadratic shape functions
P-HMQ-4-5 hybrid–mixed fomulation, bi–linear shape functions,
5 stress variables
P-HMQC-4-5 hybrid–mixed fomulation, bi–linear shape functions,
5 stress variables due to Pian–Sumihara [80]
P-HMQ-8-13 hybrid–mixed fomulation, serendipity element,
incomplete cubic shape functions for w, 13 stress variables
P-DT-3 displacement–based fomulation, linear shape functions
P-DT-6 displacement–based fomulation, quadratic shape functions
 IX

Kirchhoff Plate Elements

K-DQ-4-16 displacement–based formulation, bi–cubic shape functions
K-DQ-4-12 displacement–based formulation,
incomplete bi–cubic shape functions
K-MQ-4-16 mixed formulation, bi–linear shape functions
K-HMQ-4-16 hybrid mixed formulation, bi–linear shape functions
K-DT-6-21 displacement–based formulation, 5th order shape functions
K-DT-3-18 displacement–based formulation, 5th order shape functions
K-MT-6-12 mixed formulation, cubic shape functions for w,
λ–constraint due to the slope w,n
K-HDT-6-12 hybrid–displacement formulation, cubic shape functions for w,
w,n –constraint
K-HDT-3-9 hybrid–displacement formulation, cubic shape functions for w,
weak conformity
K-HMT-6-12 hybrid–mixed formulation, cubic shape functions for w,
w,n –constraint, quadratic shape functions for the stresses
K-HMT-3-9 hybrid–mixed formulation, cubic shape functions for w,
quadratic shape functions for the stresses, weak conformity
DKT-3-9 displacement–based formulation,
quadratic shape functions for the rotations,
cubic shape functions for the deflection at the interface

Reissner–Mindlin Plate Elements

RM-DQ-4 displacement–based formulation, bi–linear shape functions
RM-DQ-4-SRI displacement–based formulation, bi–linear shape functions,
selectively reduced integration
RM-DQ-4-HC displacement–based formulation, bi–linear shape functions,
reduced integration and hour–glass stabilization
RM-DQ-4-BHC displacement–based formulation, bi–linear shape functions,
selectively reduced integration and hour–glass stabilization
RM-MQ-4-16 mixed formulation, bi–linear shape functions
RM-HMQ-4-12 hybrid–mixed formulation, bi–linear shape functions,
12 stress variables
RM-HDQ-4-5 hybrid–displacement formulation, bi–linear shape functions
for the deflection and the rotations, 5 internal variables
due to balanced quadratic shape functions for the deflection
RM-DT-6-12 displacement–based formulation, quadratic shape functions
for the deflection, linear shape functions for the rotations
RM-DT-6-15 displacement–based formulation, cubic shape functions for
the deflection, quadratic shape functions for the rotations
Table of contents

FOUNDATIONS 1
1 Introduction 3
1.1 Governing Equations and Approximate Solution . . . . . . . . . . . . . . . . . . . 7
1.2 General Aspects Concerning the Finite Element Method . . . . . . . . . . . 8
1.3 A Comparison of Exact Solution and Approximate Solution . . . . . . . . 11
2 Discretization of the Work Equation 19
2.1 Principle of Virtual Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Principle of Virtual Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 The General Procedure to set up the Element Matrices . . . . . . . . . . . 28
2.4 Shape Functions and Convergence Criteria . . . . . . . . . . . . . . . . . . . . . . . 38
3 Structure and Solution of the System of Equations 51
3.1 Real and Virtual Nodal Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 Equations of Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Structure of the System of Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.4 Assembly of the Stiffness Matrix of the Entire System . . . . . . . . . . . . 55
3.5 The Storage and Solution of the System of Equations . . . . . . . . . . . . . 58
3.6 The Optimization of the Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.7 The Fulfillment of Dirichlet Boundary Conditions . . . . . . . . . . . . . . . . . 61
4 Heat Conduction 63
4.1 Heat Conduction at One–Dimensional Description . . . . . . . . . . . . . . . . 63
4.2 Heat Conduction Regarding Two Spatial Dimensions . . . . . . . . . . . . . . 69
4.3 Example of Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5 Membrane Structures 81
5.1 Rectangular Elements Regarding Plane Stress Situation . . . . . . . . . . . 81
5.2 Rectangular Element Comprising Modified Shear Strains . . . . . . . . . . 96
5.3 Plane Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6 Bending Structures 101
6.1 Element Matrices Regarding Euler–Bernoulli Beams . . . . . . . . . . . . . . . 101
6.2 Kirchhoff Plate Element with 16 DOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.3 Kirchhoff Plate Element with 12 DOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.4 The 12 DOF Element Employing a Weak Conformity . . . . . . . . . . . . . 129
XII Table of contents

TRIANGULAR ELEMENTS 133

7 Triangular Elements – Description of Geometry 135
7.1 Local ξ–η–Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.2 Description Employing Area Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 140
8 Triangular Elements to Describe Heat Conduction 145
8.1 A Linear Approach Related to the ξ–η–Coordinate System . . . . . . . . 145
8.2 Example of Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
9 Triangular Elements for Membrane Structures 153
9.1 Linear Shape Functions with Respect to ξ–η–Coordinates . . . . . . . . . 153
9.2 Description Employing Area Coordinates λa , λb , λc . . . . . . . . . . . . . . . . 158
9.3 Quadratic Approach Employing ξ–η–Coordinates . . . . . . . . . . . . . . . . . 159
9.4 Quadratic Shape Functions Using Area Coordinates . . . . . . . . . . . . . . . 166
9.5 A Comparison of Standard Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
10 Triangular Elements for Kirchhoff Plates 173
10.1 Choice of Shape Functions and Nodal Unknowns . . . . . . . . . . . . . . . . . 173
10.2 Complete Approach of the 5th Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.3 A Plate Element with 18 DOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
10.4 A Comparison of Standard Plate Elements . . . . . . . . . . . . . . . . . . . . . . . 193

ISOPARAMETRIC ELEMENTS 197

11 Numerical Integration 199
11.1 Numerical Integration Using Gauss–Legendre Quadrature . . . . . . . . . . 200
11.2 Numerical Integration Applied to Membrane Elements . . . . . . . . . . . . 204
11.3 Triangular Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
12 Isoparametric Elements 215
12.1 Description of the Element Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
12.2 Isoparametric Elements Regarding Membranes . . . . . . . . . . . . . . . . . . . . 218
12.3 Quadrilateral Plate Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

HYBRID QUADRILATERAL ELEMENTS 229

13 Hybrid Finite Elements 231
13.1 Mixed Formulation of Governing Equations . . . . . . . . . . . . . . . . . . . . . . . 234
13.2 Mixed Formulation of Work Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
13.3 Hybrid Discretization of Work Equations . . . . . . . . . . . . . . . . . . . . . . . . . 242
14 Hybrid–Mixed Plane Stress Elements 249
14.1 Mixed Principle of Work for Plane Stress Structures . . . . . . . . . . . . . . 249
14.2 Work Equations of a Hybrid Plane Stress Element . . . . . . . . . . . . . . . . 252
14.3 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
14.4 Subsequent Stress Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Table of contents XIII

14.5 Linear Shape Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

14.6 Quadratic Shape Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
14.7 Linear Approaches with Coupling of Degrees of Freedom . . . . . . . . . . 264
14.8 Convergence Behavior of the Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
15 Hybrid–Mixed Euler–Bernoulli Beam Elements 273
15.1 Mixed Formulation Employing Forces and Displacements . . . . . . . . . . 273
15.2 Work Equation in Hybrid Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
15.3 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
16 Hybrid–Mixed Kirchhoff Plate Elements 283
16.1 Mixed Principles of Work Concerning Kirchhoff Plates . . . . . . . . . . . . 283
16.2 Hybrid–Mixed Rectangular Plate Element . . . . . . . . . . . . . . . . . . . . . . . . 293

HYBRID TRIANGULAR PLATE ELEMENTS 299

17 Hybrid Triangular Plate Elements 301
17.1 Cubic Approach for Triangular Plate Elements . . . . . . . . . . . . . . . . . . . . 302
17.2 Hybrid–Displacement Elements Employing 10+3 DOF . . . . . . . . . . . . 306
17.3 Displacement–based Element with Weak Conformity . . . . . . . . . . . . . . 314
18 Hybrid–Mixed Triangular Plate Elements 317
18.1 Work Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
18.2 Approaches Related to the Deflection and the Stresses . . . . . . . . . . . . 319
18.3 Element Stiffness Matrix and Load Vector . . . . . . . . . . . . . . . . . . . . . . . . 321
18.4 Fulfillment of the Continuity Conditions at the Interface . . . . . . . . . . 323
18.5 Subsequent Stress Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
18.6 Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
19 Discrete Kirchhoff –Theory Element 327
19.1 The Discrete Kirchhoff Triangular Element . . . . . . . . . . . . . . . . . . . . . . . 328
19.2 Stiffness Matrix, Load Vector and Stress Analysis . . . . . . . . . . . . . . . . . 336
19.3 Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
20 Benchmark Concerning Triangular Plate Elements 339
20.1 Square Plate Subjected to Distributed Loading . . . . . . . . . . . . . . . . . . . 339
20.2 Trapezoidal Plate Subjected to Distributed Loading . . . . . . . . . . . . . . . 343

SHEAR–DEFORMATION BEAM AND PLATE ELEMENTS 347

21 Timoshenko Beam Elements 349
21.1 Elements Employing Displacements as Primary Variables . . . . . . . . . . 350
21.2 Mixed Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
21.3 Hybrid–Mixed Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
21.4 Convergence Behavior Concerning the Beam Elements . . . . . . . . . . . . 357
XIV Table of contents

22 Plate Elements Including Shear Deformations 359

22.1 Kirchhoff and Reissner–Mindlin Theories by Comparison . . . . . . . . . . 359
22.2 Governing Equations of the Reissner–Mindlin Theory . . . . . . . . . . . . . . 362
22.3 Weak Formulation of the Governing Equations . . . . . . . . . . . . . . . . . . . . 363
22.4 Quadrilateral Element Employing a Bi–linear Approach . . . . . . . . . . . . 365
22.5 Hybrid–Displacement Quadrilateral Element . . . . . . . . . . . . . . . . . . . . . . 368
22.6 Comparison of the Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
22.7 Displacement–Based Triangular Element . . . . . . . . . . . . . . . . . . . . . . . . . 380
22.8 Mixed Quadrilateral Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
22.9 Hybrid–Mixed Quadrilateral Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393

EVALUATION OF RESULTS 401

23 Error Estimation 403
23.1 The Least Squares Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
23.2 Error Estimation by Applying the Principle of Virtual Work . . . . . . . . 407
23.3 Mesh–Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
24 Quality of Elements 413
24.1 The Eigenvalue Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
24.2 The Locking Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
24.3 Improvement of Elements Suffering from Locking . . . . . . . . . . . . . . . . . 421
24.4 The Patch–Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425

REFERENCES 427

INDEX 437
FOUNDATIONS
1 Introduction

The detailed investigation of the phenomenology of structures and processes

in engineering, medicine and environmental technology is important, if their
safety has to be guaranteed with respect to limit loads and life cycle. In general
experimental methods are well developed for prototypes as basis for production
in series. If individual structures, large scale structures or systems as well as
processes of higher complexity are to be investigated, experimental methods are
too expensive in many cases. Here, computational methods are more efficient,
if they are able to describe the behavior of structures with sufficient accuracy
with respect to space and time.

Basis of analytical and computational studies are models, which usually deal
with partial differential equations or integral equations. The model equations
have to be developed with respect to the level of accuracy, that is of interest.
Engineering models, for which analytical solutions exist, often suffice for sim-
plifying investigations. However, general processes that are to be investigated
may be so complex, that a proper description has to include multiple dimensi-
ons in space with partially curved boundaries as well as the dimension in time,
and should take into account different nonlinearities. Therefore, exact soluti-
ons are only possible for special cases. Hence, a general methodology must be
available to get approximate solutions in order to investigate and to design
engineering structures in a proper way.

A universally applicable approximation method is the Finite Element Method

(FEM). Originally developed for studying the deformation behavior of plate
and membrane structures, nowadays FEM is applied in all fields of enginee-
ring, wherever the method is transferable. Since the Finite Element Method
is a reliable numerical tool to solve the model equations, it has even become
accepted in areas where investigations have been performed solely experimen-
tally so far. Furthermore, due to its efficiency the FEM offers possibilities for
parametric studies, the optimization of structures, and the consideration of un-
certainties in terms of a probabilistic concept. This is the case, for example,
regarding multi–physical processes, which exhibit strongly non–linear behavior
on different spatial and temporal scales, or at investigations of structures with
scattering material behavior and actions respectively. Thus modern commercial
FEM–codes offer different elements to describe and to optimize physical as well
as chemical processes in space and time domain.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_1
4 1 Introduction

As an example of a complex optimization process, the drive shaft of a water

wheel, presented in Figure 1-1, is the result of the interaction of functionality,
physical modeling, FEM–discretization and stiffness optimization.

Fig. 1-1 FEM–discretization of the drive shaft of a water wheel [51]

The classical Finite Element Method has been developed over decades from 1956
on. The mathematical foundation goes back to Trefftz [98] and Courant [29].
With the development of the first computers the method became of interest
for engineers. First pioneers of the method in the field of engineering were
Clough [26, 99] at UC–Berkeley, Zienkiewicz [105, 106] at Swansea University
and Argyris [4, 5] at Stuttgart University. Further textbooks for students have
been presented by Bathe [12], Hughes [47], Onate [76, 77], and many others.

In many industrial processes, the shape of products is developed by means of

computer–aided design procedures (CAD), which employ B–splines to describe
the geometry with higher order continuity in order to avoid kinks. The iso–
geometric extension of the FEM, developed by Cotrell, Hughes and Basilevs
[49, 31], employs non–uniform rational basis splines (NURBS) to approximate
the deformation of structures, whereat the geometry is designed with the same
kind of B-splines. This approach offers some advantages, if the deformation of
structures, designed by means of CAD employing high continuity conditions, is
of interest, where the geometry and the deformation field need the same order
of approximation, and where the solution needs a high order of continuity as
in the case of wave propagation as an example.
 5

Recent developments of the FEM deal with an extension of the fundamental

method to partly very sophisticated applications on multiple scales. Fundamen-
tals of polygonal elements are discussed by Wachspress [101] and Sukumar and
Tabarraei [94], and may be applied to structures of correspondent geometry as
in the case of metallic micro–structures, if the physical and chemical behavior
on this scale is of interest, cf. Figure 1-2-left. Nonetheless, the application to
macroscopic structures is possible as well, if necessary, cf. 1-2–right.

Fig. 1-2 a) Metallic micro–structure with polygonal grains

b) beam like structure [51]

Applying the idea to the investigation of wood may lead to polygonal elements
with a thin–walled geometry for the solid part and an internal polygonal ele-
ment for the description of fluids, cf. Figure 1-3.

Fig. 1-3 Thin–walled polygonal elements to model wood on the meso–scale [51]

A complete different approach to numerically investigate structures offers the

Discrete Element Method. In contrast to the classical Finite Element Method,
which is developed to investigate continua of different type, the Discrete Ele-
ment Method deals with the dynamics of rigid particles, which interact with
each other. The conditions of interaction can be described by means of springs,
viscous dampers or other rheological elements.
6 1 Introduction

Originally the method is founded on the publications by Cundall [33, 34], who
developed the method to describe the behavior of gravel like solids. The exten-
sion of the method offers the possibility to investigate the micro–structure of
continua even down to the dynamics of molecules.
The discretization of beam like structures with particles of different size and
related sieve line as well as different physical and chemical properties offers the
possibility to investigate the deformation behavior of structures with disconti-
nuities down to the micro–structure, cf. Figure 1-4.

❄
❄
✻

Fig. 1-4 a) Discrete element mesh to describe the micro–structure of concrete

b) Crack pattern due to four point loading [51]

The publications on the development of the Finite Element Method deal with
different approaches as basis of the discretization. The textbook at hand ap-
plies the Principle of virtual Work with respect to virtual displacements and
virtual forces. Due to the variety of elements and applications, which can be
hardly discussed sufficiently in a textbook, this book focuses on the basic me-
thodology of the displacement–based formulation as well as mixed and hybrid
formulations in order to explain the different approaches and the fulfillment
of the governing equations. The method will be presented with respect to civil
engineering structures, since it facilitates clear understanding of the derivation.
Therefore we will limit ourselves to load–carrying and deformation behavior of
one– and two–dimensional structures as well as to heat conduction in one and
two dimensions.

All results of examples, tests, and benchmarks have been computed by original
investigations. Nevertheless, the correctness of numbers cannot be guaranteed.
1.1 Governing Equations and Approximate Solution 7

1.1 Governing Equations and Approximate Solution

In structural mechanics there are two basic approaches available to describe
the load–carrying and deformation behavior of structures:

• Going back to Newton’s synthetic approach, the method of sections is al-

so applied to derive the governing equations at the differential element.
Therefore differential equations concerning the equilibrium and kinema-
tics as well as algebraic equations to describe the material behavior can be
derived regarding all engineering structures. Additionally, the boundary
conditions regarding stress and deformation states must be considered.

• Going back to Leibniz’s analytical approach, integration is performed over

the entire structure without differentiating between boundary and domain
or between different differential elements. This can comprise an integral
statement concerning the work performed at deforming the structure for
example, or the energy stored in the structure after the deformation pro-
cess or similar.

Both approaches are equivalent and are convertible into each other. The exact
solution fulfills the governing equations in the domain as well as at the boun-
daries and at the intersections between different integration regions. Thus all
other equations, to get an integral description of the problem, are fulfilled by
implication.
The exact solution is also named the strong solution. Concerning problems with
more than one dimension in space and arbitrary boundaries, exact solutions
generally turn out to be impossible.
Thus for the analysis of complex structures, such as plates, membranes or
shells, methods are required to approximately solve the governing equations. If
the modeling equation is equivalent to the governing equations, the identified
solution is an approximation of the exact solution. Hereby the equations applied
to get the solution are still fulfilled exactly. However, the solution itself is only
an approximation of the exact solution. Therefore, it no longer exactly fulfills
the governing equations at the differential element. Hence the approximate
solution is also named the weak solution. To get an approximate solution an
approach for the primary variable is chosen, which depends on a normalized
function describing the course of the variable in space, and a multiplier, which
scales the solution.
In the case of a displacement field u(x) the approach could be

u(x) = φ(x) · û .
8 1 Introduction

The function φ(x) is prescribed and is named the shape function, whereat poly-
nomials are usually chosen. The multiplier û is still unknown and is evaluated
by means of the respective modeling equation. This idea, developed by the
physicist Walter Ritz [88] to fulfill the Principle of Minimum Potential Ener-
gy, is the basis of all modern methods that apply integral work equations or
principles of energy to compute an approximate solution.

As an approximation to the exact solution, one may choose shape functions that
are defined regarding the whole structure. However, a closed–form approach on-
ly makes sense if the stiffnesses are constantly distributed with respect to the
domain and if the geometry of the structure is regular. The solutions concer-
ning more general problems exhibiting arbitrary geometries as well as arbitrary
boundaries, stiffnesses and external actions are limited employing an closed–
form approach.

1.2 General Aspects Concerning the Finite Element Method

At increasing complexity of geometry or of material and load–carrying behavi-
or, the requirements with respect to the computational method increase, too.
Regarding the optimization of processes and structures that presently proceeds
in all engineering disciplines and that accompanies the description of the re-
lated systems by increasingly sophisticated models, it is important to have a
computational tool at one’s disposal which is generally applicable.
Due to the fact that the Finite Element Method is able to generalize the proce-
dure for numerical investigations of arbitrary systems, the method has become
more accepted than others. Thus the FEM has currently become the standard
procedure concerning all kinds of structural analysis as limit load design and
vibration analysis of structures as well as the analysis of temperature fields,
investigation of transport processes in porous media and even the analysis of
the material behavior at a microscopic level.

The basic idea of FEM regarding structural mechanics may be described as

follows:

Instead of choosing a closed–form approach to describe the primary va-

riables, the structure is virtually divided up into many finite elements.
Concerning the single elements, shape functions are chosen to describe
the course of the primary variables whereby the related multipliers, still
to be determined, are dedicated to specified spatial positions called ele-
ment nodes. These nodal unknowns are the degrees of freedom (DOF) of
the finite model of the structure. Increasing the number of elements and
1.2 General Aspects Concerning the Finite Element Method 9

therefore also the number of DOF may not only result in an improved
approximate solution but also the geometry may be taken into account
in more detail, e.g. if curved boundaries are to be approximated by po-
lygonal representation.

An advantage of the FEM compared to methods dealing with a closed–form

approach is the polynomial order needed to describe the physical fields inside the
elements. The shape functions can be of lower order, thus, when regarding the
entire structure, the primary variables are approximated by a polygonal shape
or piecemeal parabolically, respectively. In general, the approximate solution
improves the more elements that are applied. To ensure the convergence of the
FE approximate solution against the exact solution, the approach has to fulfill
the convergence criteria at mesh refinement, see Section 2.4.2, e.g. requirements
related to the polynomial order of the shape functions and to the continuity of
the physical fields to be determined at element intersections.
Furthermore, choosing elements with identical approaches concerning the entire
structure allows for the analysis of the stress–deformation behavior of a single
element prior to instructing the computer to assemble the complete system
schematically.

The fundamental advantages of the Finite Element Method are:

• applicability to arbitrary problems regarding the whole field of natural

as well as engineering sciences;
• no restrictions concerning the geometry of the structures to be analysed,
also curvilinear boundaries are allowed;
• linear as well as non–linear modeling equations can be solved approxima-
tely;
• arbitrary external actions;
• variable system characteristics; and the
• schematic computational procedure enables a generalized implementation
of different approximations into coding.

When applying the FEM, the degrees of freedom are evaluated by means of work
equations or principles of energy. Regarding the book at hand, the Principle of
virtual Work is applied because in general, it is also valid for non–conservative
systems. The description of the virtual work can be performed in the following
different ways:

• The Principle of virtual Displacements (PvD) results in a formulation,

which applies the displacements as variables to describe the system be-
havior.
10 1 Introduction

• The Principle of virtual Forces (PvF) yields a formulation, which applies

forces or stresses as variables.
• Mixed Principles of virtual Work (PvD and PvF) utilise displacements as
well as stresses as variables to describe the system behavior.
Other approaches are possible, if the Principle of stationary values of Energy is
applicable or if the governing equations are employed and solved approximately
by means of the Method of weighted Residuals.

Concerning the first part of the book at hand, the PvD provides the basis for the
development of finite elements, which are commonly accepted in engineering
practice. In addition, it is closely related to the displacement method applied
in structural analysis. Concerning sizing, the displacement–based formulation
has the disadvantage that the stresses are solely evaluated from a subsequent
analysis, whereby the differentiation of the displacement approach yields a less
smooth distribution.
Applying the PvD, the sum of the external work due to external actions and
of the internal work due to internal force variables vanishes at regarding the
whole system. Dividing the complete structure into finite elements allows for
the description of the virtual work of the entire system as the sum of the virtual
work of the single elements.

Regarding problems in practice and due to the relatively high amount of data
which is generated, the Finite Element Method cannot be applied without using
FE–programs. FE–programs vary only slightly at employing different types of
elements, therefore they are transferable to diverse problems with just few
modifications.
Even to analyse arbitrary structures comprising different components, element
formulations must be provided employing different types of elements. Such an
element library could differentiate between:
• the physics of the problem as bending, membrane behavior, heat conduc-
tion, fluid flow etc.;
• the geometry of the element as quadrilaterals, rectangles, triangles, bars,
etc.; and
• the approximate approach to describe the displacements as linear, qua-
dratic or higher order functions.
Currently, in addition to the basic FE–programming systems, separate codes
are common concerning pre–processing, what deals with the data generation
and control, as well as post–processing, which is neccessary for the visual re-
presentation of the solution and for the interpretation of results.
1.3 A Comparison of Exact Solution and Approximate Solution 11

1.3 A Comparison of Exact Solution and Approximate Solution

Assuming the exact solution to be unknown one may compute an approximate
solution by applying the FEM by means of the Principle of virtual Displace-
ments. Without referring to the related procedure in detail, the differing results
regarding an approximate solution applying FEM and the exact solution should
be presented concerning the bar referred to in Figure 1-5.

H
case 1 units :
EA
x, u, l [m]
p(x) p(x) [ N/m ]
case 2 H [N]
EA
x,u EA [N]
l

Fig. 1-5 Bar – geometry and loading

1.3.1 Analytically Exact Solutions

In order to describe the respective problem, the exact solution has to fulfill
all governing equations regarding the domain to be investigated as well as the
Dirichlet boundary condition for the displacements and the Neumann boundary
condition for the first derivative of the deformation, which correlates with the
longitudinal force. Regarding the bar depicted in Figure 1-5, this yields:

a) Kinematics
ε = u,x in domain and
(1.1)
u(0) = 0 at the boundary,

b) Equilibrium
N,x = 0 case 1 in domain,
N,x = −p (x) case 2 in domain and
(1.2)
N (ℓ) = H case 1 at the boundary,
N (ℓ) = 0 case 2 at the boundary.

c) Material equation – Hooke’s model

N = EA · ε in domain, (1.3)
12 1 Introduction

The exact solutions follow to

• case 1: external boundary action H
H
u(x) = · x,
EA
H
ε(x) = ,
EA
N (x) = H .

• case 2: constantly distributed external action p(x)

p x2
u(x) = · (ℓ · x − ) ,
EA 2
p
ε(x) = · (ℓ − x) ,
EA
N (x) = p · (ℓ − x) .

To control the analytical solution, the solution is to be introduced into the

governing Equations (1.1) to (1.3), and the boundary conditions are to be
checked at positions x = 0 and x = ℓ respectively.

1.3.2 Approximate Solutions

To start with, one may choose a single element to describe the entire structure,
because the PvD does not make any statements concerning the number of
elements needed. Then an approximate approach is chosen to describe the
displacements in the domain, whereby its unknown multipliers are evaluated
by applying the PvD, which is shown here with one nodal unknown:

u(x) = φ(x) · û .

Here, û is the unknown multiplier and φ(x) is the given shape function, for
which a linear run is feasible, see Figure 1-6.

x,u

Φ(x)

Fig. 1-6 Bar – linear approach to describe the displacement

1.3 A Comparison of Exact Solution and Approximate Solution 13

Even though the analytically exact solution is unknown concerning u(x) in

general, the approximate solution may correspond to the strong solution in a
particular case. With u(x) being an approximation, the equilibrium is integrally
fulfilled for the entire structure applying the PvD, but it is only approximately
fulfilled regarding the differential element.
The other governing equations are still considered exactly, as they are introdu-
ced into the PvD equation. At the subsequent analysis stage, the strains ε(x)
are evaluated from the displacements u(x) applying Equation (1.1). Further,
the longitudinal forces N (x) are evaluated from ε(x) applying Equation (1.3).
If u(x) is an approximation, ε(x) and N (x) are approximations, too.

Case 1: External Boundary Action H

Regarding the entire bar, a computation applying the PvD and employing a
linear approach for u(x) according to Figure 1-6 yields the solution:

u(x) = φ(x) · û ,
φ(x) = x/l ,
H ·l
û = .
EA
In this particular case the approximate solution regarding u(x) corresponds
to the exact solution. ε(x) is evaluated from a subsequent analysis following
Equation (1.1), and ε(x) equals the exact solution, too:
H
ε = u,x = φ(x),x · û = 1/l · û = .
EA
The evaluation of the longitudinal force following Equation (1.3)
H
N = EA · ε = EA · =H
EA
also yields the exact solution. Introducing the longitudinal force N (x) into the
equilibrium condition (1.2) and controlling the boundary condition confirms
the PvD to provide the analytically exact solution in this particular case.

With p(x) = 0 it applies N,x = 0 ,

with the boundary condition N (ℓ) = H .

The Dirichlet boundary condition u(0) = 0 is already fulfilled by the approach.

A computation employing several elements or subdomains also yields the exact
solution regarding this external action.
14 1 Introduction

Case 2: Constantly Distributed External Actions p(x)

Regarding the entire bar, a computation applying the PvD and employing a
linear approach for u(x) according to Figure 1-6 yields the approximate solu-
tion of the displacement:
u(x) = φ(x) · û ,
φ(x) = x/l ,
p · ℓ2
û = .
2EA
Here, the approximate solution yields the exact boundary displacement, what
may be explained by means of the Principle of virtual Forces, but can not be
generalized. Regarding the domain, however, deviations exist with respect to
the exact solution, see Figure 1-7.

p . l2 exact solution
u
2 EA approximation
x

Fig. 1-7 Bar – approximation and exact solution regarding the displacement

The strain ε(x) is evaluated by a subsequent analysis following the kinematic

condition of Equation (1.1):

1 p·ℓ
ε = u,x = φ(x),x · û = · û = .
ℓ 2EA
u(x) being an approximation results in ε(x) being an approximation, too. Only
the mean value of the strains equals the strong solution, see Figure 1-8.

p.l
ε EA p.l exact solution
2 EA approximation
x

Fig. 1-8 Bar – approximation and exact solution regarding the strains

The evaluation of the longitudinal force, following the material Equation (1.3),
also yields an approximate solution
p·ℓ
N = EA · ε = .
2
1.3 A Comparison of Exact Solution and Approximate Solution 15

Here, only the mean value of the approximation equals the exact solution, too,
see Figure 1-9.

N p. l p.l exact solution

2 approximation
x

Fig. 1-9 Bar – approximation and exact solution regarding the longitudinal force

Although, in general, the courses of the state variables are only approximati-
ons of the strong solution, parts of the governing equations are exactly fulfilled.
The kinematics following Equation (1.1) as well as the material behavior follo-
wing Equation (1.3) are implicitly enclosed in the work equation and therefore
are exactly fulfilled. The equilibrium conditions (1.2) are not exactly fulfilled,
because the PvD is only a weak formulation of the equilibrium. Since the longi-
tudinal forces are constant, inside the domain the equilibrium condition yields

N,x + p(x) 6= 0 .
|{z}
0

The Neumann boundary conditions are not exactly fulfilled either

p·ℓ
N (ℓ) = 6= 0 .
2
When evaluating the results, the following is to be taken into account. Alt-
hough the approximation regarding the distribution of the longitudinal forces
seems to be relatively coarse compared to the displacements, the PvD optimally
approximates the equilibrium and thus the longitudinal forces N (x), but not
the displacements u(x).

1.3.3 The FE–Solution Regarding two Elements – Case 2

The bar is subdivided into two identical elements, whereby, regarding the dis-
placements, a linear approach is employed for both, see Figure 1-10. This results
in an approximation of the displacement comprising two degrees of freedom û1
and û2 as well as the shape functions φ1 (x) and φ2 (x)

u(x) = φ1 (x) · û1 + φ2 (x) · û2 .

Although the kink at node 1 seems to pertube the solution, it is nonetheless

admissible, what is rigorously proved later on. Compared to the approximation
16 1 Introduction

employing a linear approach and solely one element, the solution dealing with
two degrees of freedom better approximates the exact solution but needs further
discussion.

Φ1(x)

Φ2(x)

x,u

0 1 2

Fig. 1-10 Bar – approximation employing two degrees of freedom

The principle statements concerning the nature of the approximation remain

valid with regard to two elements. Regarding the nodes 1 and 2 the displace-
ments turn out to be exact, whereat the deviations inside the elements become
smaller in comparison to one element, cf. Figure 1-11.

2
3 p.l
8 EA
p . l2 exact solution
u
2 EA approximation

Fig. 1-11 Bar – approximation and exact solution regarding the displacements

Performing the computation of the other state variables, inside the elements the
kinematic conditions (1.1) and the material equation (1.3) are exactly satisfied
again. This yields constant strains ε(x) as well as longitudinal forces N (x),
whereat the analytical and the approximate solutions turn out to be identical
at the midpoints of the elements, cf. Figure 1-12. Since the shape functions do
not be differentiable, strains and stresses offer a jump at node 1.

3p . l
p.l
N 4 exact solution
p.l
approximation
4

Fig. 1-12 Bar – approximation and exact solution of the longitudinal force
1.3 A Comparison of Exact Solution and Approximate Solution 17

Because of the weak formulation of the equilibrium by means of the Principle

of virtual Displacements, the equilibrium condition and the Neumann bounda-
ry condition, given in Equation (1.2), are not exactly fulfilled, but the error
becomes smaller:

N,x + p(x) 6= 0 ,
|{z}
0
p·ℓ
N (ℓ) = 6= 0 .
4
The solution also offends against the intersection condition concerning the lon-
gitudinal force at node 1, where the jump is an indication of the error as well.

Performing further elements, the approximate solution converges to the exact

solution. The equilibrium as well as the Neumann boundary condition are only
approximately fulfilled in general. In particular cases, the nodal displacements
are identical to the analytical solution. This is only valid if the approximate
approach corresponds to the solution of the homogeneous differential equation.

Remarks
The comparison of the exact solution and the approximation leads to the fol-
lowing fundamental statements regarding the PvD.

• The Principle of virtual Displacements satisfies the governing equations

and the boundary conditions with different accuracy.
• The equilibrium and the Neumann boundary conditions are satisfied in
a weak sense.
• The kinematics, the material equation, and the Dirichlet boundary con-
ditions are satisfied exactly.
• The accuracy of the approximation may be increased by increasing the
number of elements.
• The approximation of the displacement field leads to nodal displacements
of high accuracy, whereby the stresses are approximated by lower order.
2 Discretization of the Work Equation

The main idea of FEM is to subdivide the whole structure into finite elements.
Thereby it is possible to describe the virtual work of the entire structure as
the sum of the virtual works performed at the single finite elements. This is
required to develop the system of equations employing the degrees of freedom
of the approach as unknowns,
X
δW = δWel = 0 . (2.1)
This yields a two–part task to be worked on:

• Computation of work performed at each element;

• Summing up the works performed at all elements to get the work perfor-
med on the entire structure.

This section systematically describes the formal procedure at the element level.
At first, the Principle of virtual Displacements (PvD) will be proven to be an
equivalent formulation of the equilibrium conditions with regards to a bar, as
an example. It is the basis for further derivations. As a second approach, the
Principle of virtual Forces (PvF) will be applied to the bar in order to explain
the equivalence of the principle to the kinematic conditions.

2.1 Principle of Virtual Displacements

Regarding a self–contained system, the sum of internal as well as external
works performed at the entire structure must be zero at the changeover from
the undeformed to the deformed situation.
X X
W = (Wi + We ) = 0 . (2.2)

The static case is described here, where deformations and thus also stresses and
strains slowly increase simultaneously with loads from zero to the final value.

The bar is used to clearly and concisely represent the different parts of the
work previously defined. Here, the work theorem according to Equation (2.2)
is applied to a bar loaded by an external action p(x) as well as by an external
boundary action H(ℓ) as shown in Figure 2-1. Due to the real loadings p(x)
and H(ℓ), the deformation state characterized by u(x) and ε(x) arises as do
the longitudinal forces N (x).

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_2
20 2 Discretization of the Work Equation

If the Dirichlet boundary condition is fulfilled, internal work is performed on

the strains by the longitudinal forces as well as external work is performed on
the displacements by the external actions. As both internal stresses and strains
are counteracting, the internal work is preceded by a negative sign
Z Z
X 1 1 1
W =− N · ε dx + p · u dx + [ H · u ]l = 0 . (2.3)
2 2 2
The deformations are generated by the forces itselves, therefore a 1/2 – coeffi-
cient in Equation (2.3) marks the active work, see also Figure 2-2.

2.1.1 Virtual Work

In addition to the real external action, the system is loaded by a virtual – this
means imagined and arbitrary small – external action δP now. This yields a
total deformation u(x) + δu(x) associated with the strains ε(x) + δε(x) as well
as with the normal forces N (x) + δN (x).

p(x)
H
x
l

δu(x) u
u(x)

δε(x) ε
ε(x)

Fig. 2-1 Real and virtual deformation state of a bar

Regarding this overall state being achieved in two steps, the work theorem also
has to be fulfilled:
Z Z
X 1 1 1
W =− N · ε dx + p · u dx + [ H · u ]x=ℓ
2Z 2 2
1 1
− δN · δε dx + [ δP · δu ]x=ℓ
2Z Z 2
− N · δε dx + p · δu dx + [ H · δu ]x=ℓ = 0 . (2.4)
2.1 Principle of Virtual Displacements 21

Referring to Equation (2.4), the 1/2 – coefficient marks the active work of the
basic state regarded at first, and the additional virtual state added afterwards.
Both parts of the work must be zero individually because of Equation (2.3).
Thus the remaining virtual passive work, which the real stress state performs
on the virtual deformation added subsequently, must be zero, too:
Z Z
δW = − N · δε dx + p · δu dx + [ H · δu ]l = 0 . (2.5)

Figure 2-2 illustrates the meaning of the aforementioned parts of the work
Equation (2.4), assuming a linear correlation between the forces and the re-
lated deformations. The area determined by the straight and the x–axis can
be interpreted as the active work. The horizontally–lined triangles mark the
active work of the basis state as well as the active work of the virtual state, the
crosshatched rectangle marks the passive work according to Equation (2.5).

N δN

δε

Fig. 2-2 Active and passive work – PvD

Referring to Equation (2.5) the virtual state must only fulfill kinematics. Thus,
the virtual external action and the related virtual stress state, primarily in-
troduced to enable a better understanding of the basic principle, both are not
required any longer. Therefore, since only the virtual deformation state is re-
quired, this can be prescribed independently of any virtual load.

The following may explain which assumptions ensure the Principle of virtual
Displacements according to Equation (2.5) being equivalent to the equilibrium
condition. Assuming that kinematics according to Equation (1.1) also apply to
the virtual state
δε = δu,x in domain,
(2.6)
δu(0) = 0 at boundary,

Equation (2.5) turns to

Z Z
δW = − N · δu,x dx + p · δu dx + [ H · δu ]x=ℓ = 0 . (2.7)
22 2 Discretization of the Work Equation

Integration by parts transforms Equation (2.7) into

Z Z
δW = N,x · δu dx − [ N · δu ]ℓ0 + p · δu dx + [ H · δu]x=ℓ
Z
= (N,x + p) · δu dx + [( H − N ) · δu]x=ℓ = 0 . (2.8)

Referring to Equation (2.8) the differential equation related to the equilibrium

is given within the brackets of the integral. The boundary term takes into
account the boundary conditions concerning the real force N (ℓ) = H at position
x = ℓ. The boundary condition concerning the virtual displacements δu(0) = 0
at position x = 0 is fulfilled a priori according to Equation (2.6).
Thus, the following statement applies:

The Principle of virtual Displacements according to Equation (2.5)

is equivalent to the equilibrium conditions (1.2), if the virtual state
fulfills kinematics (2.6) and (1.1) respectively.

2.1.2 Fulfillment of Governing Equations and Boundary Conditions

In addition to the equilibrium condition, kinematics and material equations
must be fulfilled concerning the real state as well, if the Principle of virtual
Displacements shall be the basis of the analysis of the overall state. Hence, the
internal forces N , applying Equation (2.7), may be described by the material
equation (1.3) and kinematics (1.1). The Dirichlet boundary conditions have to
be fulfilled a priori, since they are part of kinematics.
Z Z
δW = − N · δu,x dx + p · δu dx + [ H · δu ]l = 0
Z Z
= − ε · EA · δu,x dx + p · δu dx + [ H · δu ]l = 0
Z Z
= − u,x · EA · δu,x dx + p · δu dx + [ H · δu ]l = 0 .

Now, all governing equations as well as all boundary conditions are taken into
account, and the work equation describes the real problem completely and with-
out any approximation. Since the work is solely described by real and virtual
displacements, it is named displacement–based formulation. For the purpose of
practical analyses it is convenient to multiply the equation by −1. Thus, the
PvD according to the work equation (2.9) is applied to further considerations:
Z Z
− δW = u,x · EA · δu,x dx − p · δu dx − [ H · δu ]l = 0 . (2.9)
2.1 Principle of Virtual Displacements 23

However, it must be pointed out that the PvD integrally fulfills the equilibrium
conditions concerning the entire structure. This is also known as a weak form
of equilibrium. Kinematics and the material equations are still exactly fulfilled
at every point of the structure, because these governing equations are taken
into account implicitly by the PvD. Even the boundary conditions are fulfilled
differently. While the Neumann boundary conditions of the real state, being
equilibrium conditions, are fulfilled integrally by the work equation and named
as natural boundary conditions, the Dirichlet boundary conditions of the real
and of the virtual state must, in contrast, be fulfilled exactly by the displace-
ment field. Therefore, they are named as substantial boundary conditions.

2.1.3 Fulfillment of the Conditions at the Element Intersections

FE–analyses are applied where no analytical solution is known or can be found.
If the unknown deformation state of a structure is evaluated by applying the
work equation (2.9), approximations have to be chosen to describe the real and
the virtual displacements respectively, which could match by chance the exact
solution in special cases of geometry as well as of external actions. However,
the approximations have to satisfy kinematics according to Equation (1.1) and
Equation (2.6) in both cases. Thus the approximations for the real as well as the
virtual displacements must be continuous as well as continuously differentiable
regarding the entire domain.

However, the virtual work regarding the entire structure according to Equation
(2.9) is represented by a scalar quantity, which can also be described as the sum
of the works performed at single integration domains, which are the elements:
X
δW = δWel = 0 . (2.10)

If the work equation is integrated element by element, it might also be con-

venient to approximate the displacement field within all elements by the same
shape functions. Nevertheless, it must be ensured that, when integrating the
work equation regarding one element at a time, all terms of work are considered
and, moreover, the governing equations – including boundary and intersection
conditions – must not get hurt. This can be achieved if the continuity of the
approximations is ensured at the element intersections without any gaps of the
displacements.
As the shape functions are limited to single elements, kinematics might no lon-
ger be fulfilled at the element intersections, if the shape functions do only satisfy
the continuity of displacements. Thus the strains may become discontinuous at
intersections of neighboring elements.
24 2 Discretization of the Work Equation

By means of a simplifying example the fulfillment of kinematics and the equili-

brium conditions at the intersections is explained as follows. Figure 2-3 repre-
sents a bar subdivided into two finite elements.
A B C
1 2
H
p(x)
x
l l

Fig. 2-3 Subdivision into elements

Figure 2-4 applies the method of sections to the elements considered, in order
to visualize the internal forces and displacements at the respective nodes.

N A 1 B N N B 2 C
A B B H
p p

A 1 B B 2 C

uA1 uB1 uB2 uC2

Fig. 2-4 Elements being cut cleanly

Following Equation (2.9), the work equation that applies the PvD to every
single element is comprised of both internal and external work inside the domain
as well as work performed at the element boundaries and yields
Z Z
−δWel = δu,x · EA · u,x dx − δu · p(x) dx − [ δWel−bound. ] (2.11)

with the work performed at the element intersections, cf. Figure 2-4,

[ δWel−bound. ](1) = [ δuB · NB − δuA · NA ](1) ,

(2.12)
[ δWel−bound. ](2) = [ δuC · H − δuB · NB ](2) .

When adding up the elements’ work to the system’s work, the work at the
elements’ intersections is also added up. If the shape functions regarding the
displacements are continuous at the element intersections, the following applies:

uB(1) = uB(2) and δuB(1) = δuB(2) . (2.13)

2.2 Principle of Virtual Forces 25

Furthermore, the internal force variables have to fulfill Equation (2.14)

NB(1) = NB(2) , (2.14)

whereby the respective terms of internal work at the element intersections inside
the structure compensate exactly for each other and can be ignored at the
element level.

In general the polynomial order of the shape functions regarding the displace-
ments results in discontinuous longitudinal forces at the elements’ intersections
and therefore Equation (2.14) might not be fulfilled exactly. However, if the
work is disregarded at element intersections, Equation (2.14) is fulfilled impli-
citly in its weak form by applying the PvD to the entire system in the same
way as all the other equilibrium conditions.

2.2 Principle of Virtual Forces

In analogy to the Principle of virtual Displacements being a weak formulation of
the equilibrium conditions, the Principle of virtual Forces is a weak formulation
of kinematics. Both principles are applied to be the basis of the development
of generalized hybrid finite elements, cf. Section 13. Thus in the following the
equivalence of the Principle of virtual Forces and kinematics is shown in order
to explain the duality of both principles.

Here again, the work theorem is applied to the bar shown in Figure 2-5. The bar
is loaded by a distributed external action p(x) and a prescribed displacement
ue at the boundary.

ue
p(x)
δP
P
x
l δN = δP

Fig. 2-5 A bar loaded by a distributed external action and virtual load

In contrast to the Principle of virtual Displacements the virtual load δP is

applied to the bar at the position x = ℓ first. Thus the work performed by the
virtual stress state on the related deformation yields
Z
X 1 1
W =− δN · δε dx + [ δP · δu ]x=ℓ = 0 . (2.15)
2 2
26 2 Discretization of the Work Equation

The deformations are generated by the forces itselves, therefore a 1/2 – coeffi-
cient in Equation (2.15) marks the active work, see also Figure 2-6.

Virtual Work
In addition to the virtual action, the system is loaded by the real external
action p(x), ue now. This yields a total deformation δu(x) + u(x) associated
with the strains δε(x) + ε(x) as well as with the normal forces δN (x) + N (x).
Regarding this overall state being achieved in two steps, the work theorem also
has to be fulfilled:
Z
X 1 1
W =− δN · δε dx + [ δP · δu ]x=ℓ
2Z 2Z
1 1 1
− N · ε dx + p · u dx + [ N · ue ]x=0
2
Z 2 2
− δN · ε dx + [ δP · u ]x=ℓ + [ δN · ue ]x=0 = 0 . (2.16)

Referring to Equation (2.16), the 1/2 – coefficient marks the active work of the
virtual state regarded at first, and the additional real state added afterwards.
Both parts of the work must be zero individually because of Equation (2.15).
Thus the remaining virtual passive work, which the virtual stress state performs
on the real deformation added subsequently, must be zero, too:
Z
δW = − δN · ε dx + [ δP · u ]x=ℓ + [ δN · ue ]x=0 = 0 . (2.17)

Figure 2-6 illustrates the meaning of the aforementioned parts of the work
Equation (2.16), assuming a linear correlation between the forces and the re-
lated deformations. The area determined by the straight and the x–axis can
be interpreted as the active work of the overall state. The horizontally–lined
triangles mark the active work of the virtual state as well as the active work of
the real state, the crosshatched rectangle marks the passive work according to
Equation (2.17).

δP P

δu

Fig. 2-6 Active and passive work – PvF

2.2 Principle of Virtual Forces 27

Referring to Equation (2.17) the virtual state must only fulfill the equlibrium
conditions. Thus, the virtual deformations and strains, primarily introduced
to enable a better understanding of the basic principle, both are not required
any longer. Therefore, since only the virtual stress state is required, this can
be prescribed independently of any virtual deformation.

Weak Form of Kinematics

The following may explain which assumptions ensure the Principle of virtual
Forces according to Equation (2.17) to be equivalent to kinematics.

Applying kinematics
u,x − ε = 0 in domain,
(2.18)
u − ue = 0 at boundary x = 0 ,
in a weak formulation Equations (2.18) turn to
Z
δW = δN · ( u,x − ε ) dx − [ δN · ( u − ue ) ]x=0 = 0 . (2.19)

Referring to Equation (2.19) kinematics related to the real state is given with-
in the brackets of the integral. The boundary condition concerning the real
displacement u − ue = 0 at position x = 0 is considered in a weak sense as well.

Integration by parts transforms Equation (2.19) into

Z
δW = − ( δN,x · u + δN · ε ) dx + [ δN · u]ℓ0 − [ δN · ( u − ue ) ]x=0 = 0 .

At x = 0, the boundary terms eliminate each other except for the prescribed
displacement. At x = ℓ, the remaining bondary term belongs to the virtual
load δN u|ℓ = δP u|ℓ . Furthermore, if the virtual state fulfills the equilibrium
condition δN,x = 0,
Z
δW = − δN · ε dx + [ δN · ue ]x=0 + [ δP · u]x=ℓ = 0 (2.20)

yields the final statement. Comparing Equation (2.17) and Equation (2.20) it
becomes clear, that the Principle of virtual Forces is a weak formulation of
kinematics and thus may be applied to develop finite elements analogously to
the Principle of virtual Displacements. Thus, the following sentence applies:
The Principle of virtual Forces according to Equation (2.17) is equi-
valent to kinematics (2.18), if the virtual state fulfills the equilibri-
um conditions (1.2).
28 2 Discretization of the Work Equation

2.3 The General Procedure to set up the Element Matrices

The weak form of the equilibrium condition described by the work equation
applying the PvD (2.9) is valid for the entire structure. According to equation
(2.10), the virtual work of the structure can be represented as the sum of the
single finite elements’ work
X
δW = δWel = 0 .
Thereby the element’s work may either be formulated in analogy to the system’s
work equation or may be reassembled at every single element, see Section 2.1.3.

The benefit of the FEM is that every single element’s work can be described
applying a pattern that is identical for all elements. In the following, all indi-
vidual procedural steps are represented in detail up to the discretization of the
work equation, as well as the evaluation of the internal force variables.

2.3.1 Solution Procedure

According to Equation (2.9), the initial equation to compute an approximate
solution is the weak form of the equilibrium regarding the entire structure
Z Z
− δW = u,x · EA · δu,x dx − p · δu dx − [ H · δu ]l = 0 .

After the choice of shape functions and the integration of the work equation
concerning the element domain, one gets the discrete form of the work equation,
which depends on the still unknown nodal displacements. The discretization of
the work equation takes place in three steps:
1. set up the work equation regarding a single element applying a displace-
ment–based formulation,
2. choose shape functions regarding the displacements, and
3. introduce the shape functions into the work equation and perform inte-
gration.
Furthermore the FE–procedure comprises:
4. sum up all elements’ work to get a description of the entire system and
computation of the nodal displacements as presented in Section 3, and
5. subsequent stress analysis to evaluate the internal force variables element
by element.
The general procedure is shown by investigating a bar e.g. and is transferred
to other problems as heat conduction, membranes, beams and plates in the
following sections.
2.3 The General Procedure to set up the Element Matrices 29

Step One: Work Equation Regarding a Single Element

Figure 2-3 shows the bar subdivided into two elements and loaded by a con-
stantly distributed external action as well as by an external boundary action.
Both elements are identical concerning length ℓ and elasticity EA.
According to Section 2.1.3, work performed at the element boundaries does
not have to be considered if the element intersection conditions are fulfilled. At
element level, the work H δuc performed by the external boundary action H
can be dropped, because it is considered at the system level. Therefore, when
adding up the elements’ works to get the system’s work, only
Z Z
−δWel = δu,x · EA · u,x dx − δu · p(x) dx 6= 0 (2.21)

has to be taken into account, whereby this notation equals the work equation
represented at system level. It must be pointed out that the element’s work
given in Equation (2.21) does not include the work performed at the element
intersections. Therefore, the work equation at element level cannot be directly
used to evaluate the displacements.

Step Two: The Shape Functions

If the strong solution is unknown, an approximate approach must be chosen
to describe the displacements at element level. This approximation must fulfill
the continuity conditions of the PvD concerning the displacements inside the
related domain as well as the displacements at the beginning and at the end
of the element. Here, a linear approach with two degrees of freedom will suffice
to describe u(x).
At first, the polynomial approach applies:
u(x) = a0 · x0 + a1 · x1 = a0 + a1 · x . (2.22)
The coefficients a0 , a1 are the degrees of freedom of the approach and x0 , x1
are general monomials of different order, whereat the coordinate 0 ≤ x ≤ ℓel is
defined within each single element.
Regarding this general polynomial approach, the degrees of freedom do not have
any physically descriptive meaning. Furthermore, it might be difficult to fulfill
the element’s intersection conditions concerning the displacements. Hence it
makes more sense to employ the physically meaningful and easily interpretable
nodal displacements uA and uB as the degrees of freedom:
u(x) = φA (x) · uA + φB (x) · uB . (2.23)
The fundamental advantage of this descriptive approach is the fact that the
continuity conditions at the element’s intersections as well as the Dirichlet
30 2 Discretization of the Work Equation

boundary conditions may be met more easily. The following will show how
to transfer the general polynomial approach given in Equation (2.22) into the
descriptive approach given in Equation (2.23), cf. Figure 2-7.

uB
u(x)
uA
ΦA uA ΦB uB
= +
1 1
A B x
x ΦA = 1− xl ΦB = l

l l l

Fig. 2-7 Scaling of the approach with respect to the nodal displacements

An evaluation of the approach in Equation (2.22) according to the element

boundaries results in the two equations

x = 0 : u(0) = a0 + a1 · 0 = uA ,
(2.24)
x = ℓ : u(ℓ) = a0 + a1 · ℓ = uB

yielding the solution

a0 = u A ,
a1 = (uB − uA ) /ℓ .

Thus, the approach according to Equation (2.22) leads to

u(x) = uA + (uB − uA ) · x /ℓ ,

or merging in a more conventional way gives

x x
u(x) = (1 − ) · uA + · uB , (2.25)
ℓ ℓ
where uA and uB are the nodal displacements and φA (x) and φB (x) are the
related shape functions
x x
φA (x) = 1 − and φB (x) = .
ℓ ℓ
These shape functions φi are also employed to describe the virtual displace-
ments δu(x). Due to the fact that two degrees of freedom will need two con-
ditional equations later on, the work equation is arranged according to two
2.3 The General Procedure to set up the Element Matrices 31

independent virtual displacement states. It is advantageous to choose the sha-

pe functions of the virtual displacements to be the same as the shape functions
of the real displacements
δuA (x) = φA (x) · δuA ,
δuB (x) = φB (x) · δuB ,
since this choice leads to a symmetrical system of equations and thus turns out
to be efficient concerning the solution of the overall system of equations.

Step Three: Integration of the Element Work

Following Equation (2.21), the real as well as the virtual displacements are to
be differentiated with respect to x:
u(x),x = φA (x),x · uA + φB (x),x · uB ,
δuA (x),x = φA (x),x · δuA , (2.26)
δuB (x),x = φB (x),x · δuB .
Introducing these approaches into the element’s work equation (2.21) leads to:
Z Z
−δWA = δuA · φA,x · EA · {φA,x · uA + φB,x · uB } dx − δuA · φA · p(x) dx ,
Z Z
−δWB = δuB · φB,x · EA · {φA,x · uA + φB,x · uB } dx − δuB · φB · p(x) dx .

Since the nodal values are independent of the integration, the element’s total
work can be represented in a matrix notation:
− δWel = − δWA − δWB
Z " # " #
φA,x EA · φA,x φA,x EA · φB,x uA
= [ δuA δuB ] · { dx ·
φB,x EA · φA,x φB,x EA · φB,x uB
Z " #
φA · p(x)
− dx } . (2.27)
φB · p(x)
The element matrix is symmetrical only if the shape functions regarding the
real and the virtual displacements are identical.
We may start with
p(x) =p = const. ,
φA (x),x = − 1/ℓ = const. ,
φB (x),x = + 1/ℓ = const. .
32 2 Discretization of the Work Equation

After analytical integration of the element’s work it follows:

" 1 # " #  2

− ℓ EA (− 1ℓ ) x − 1ℓ EA 1ℓ x uA p ( x − x2ℓ ) ℓ
−δW = [ δuA δuB ] · { · − }0
1 1 1 1 uB x2
ℓ
EA (− ℓ
) x ℓ
EA ℓ
x p ( 2ℓ
)
" # " # " #
EA/ℓ −EA/ℓ uA pℓ/2
= [ δuA δuB ] · { · − }. (2.28)
−EA/ℓ EA/ℓ uB pℓ/2

The integrated representation of the element matrix equals the stiffness matrix
of the bar, as it has been established by the displacement method developed
from structural analysis. Therefore the FEM’s element matrix may also be
named as element stiffness matrix.
The entries of the element matrix illustrate the nodal forces, which evolve from
the real nodal displacements and which perform work on their corresponding
virtual nodal displacements.
Every single sum of a row represents the work of the element referring to a rigid
body motion performed with respect to the corresponding virtual displacement
state, wherein the nodal displacements uA and uB are the same size. Due to
the fact that a rigid body motion must not produce any forces, the work and
therefore every sum of a row regarding the element stiffness matrix is equal to
zero. Just as the sum of a column describes the nodal forces’ work performed
with respect to a virtual rigid body displacement.

Step Four: Change over to the Overall System

Adding up the elements’ works to the system’s work provides the system’s
stiffness matrix, the system’s load vector and the system’s vector of degrees of
freedom. The set up of the system of equations to evaluate the nodal displace-
ments is shown in Section 3.

Step Five: Subsequent Analysis of Stresses

The unknown nodal degrees of freedom are evaluated by solving the overall
system of equations. By applying a subsequent analysis regarding every element
individually, the strains may be evaluated from the known u(x) by means of
kinematics

ε = u(x),x
= φA (x),x · uA + φB (x),x · uB
1 1
= − · uA + · uB ,
ℓ ℓ
2.3 The General Procedure to set up the Element Matrices 33

and, furthermore, the longitudinal forces may be evaluated in terms of the

material equation
1 1
N = EA · ε = EA · (− · uA + · uB ) . (2.29)
ℓ ℓ
When applying the matrix notation, the strain may be represented by
   
1 1 uA
[ε]= − ·
ℓ ℓ uB
as well as the longitudinal force
   
1 1 uA
[ N ] = [ EA ] · [ ε ] = [ EA ] · − · .
ℓ ℓ uB
To evaluate the equations element by element, there is no need for integration
or solving a system of equations. Due to the differentiation, the polynomial
order of the courses to describe the distributions of strains and stresses is lower
than the respective order for the description of the displacements.

2.3.2 Example
Section 1.3 gives an approximate solution regarding the example referred to in
Figure 1-5 and Figure 2-8, which is also compared to the analytical solution.

H
case 1
EA
p(x)
case 2
EA
x,u
l

Fig. 2-8 Bar – geometry and loading

According to the procedure presented in Section 2.3.1, the approximate solution

is represented for one element in the following.

Loading Condition: Constantly Distributed External Action p(x)

The discretized form of the PvD is given by
" # " # " #
EA/ℓ −EA/ℓ uA pℓ/2
[ δuA δuB ] · { · − } = 0.
−EA/ℓ EA/ℓ uB pℓ/2
34 2 Discretization of the Work Equation

The boundary conditions uA = 0 and δua = 0, prescribed at x = 0, may be

introduced by eliminating the first row and the first column of the system of
equations. It remains
     
[ δuB ] · { EA/ℓ · uB − pℓ/2 } = 0 .

Thus the displacement follows to

   
uB = pℓ2 /2 EA .

Applying the subsequent stress analysis, the longitudinal force is evaluated to

   
1 1 uA
[ N ] = [ EA ] · − · .
ℓ ℓ uB

Consideration of the boundary condition uA = 0 yields

   
1 EA
[ N ] = [ EA ] · · [ uB ] = uB = [ pℓ/2 ] .
ℓ ℓ

Loading condition: External Boundary Action

The equivalent procedure may be applied to investigate the external boundary
action H. Introducing the load vector
 
0
p= ,
H

the displacement follows to

[ uB ] = [ Hℓ/EA ] ,

and the longitudinal force gives

[ N ] = [ H ].

2.3.3 Matrix Notation

Transferring to matrix notation enables a more general formulation of the work
equation. Therefore, symbols are introduced which are related to the matrices
and vectors given in Section 2.3.1. Their specific contents are represented with
regard to the bar. In the following, matrices are marked with capital bold
letters, while vectors shall be identified by small bold letters.
2.3 The General Procedure to set up the Element Matrices 35

Step One: Work Equation Regarding a Single Element

The work equation regarding the bar without considering any boundary terms
Z Z
−δWel = δu,x · EA · u,x dx − δu · p(x) dx

can be generalized by applying the following notations:

vector of displacements : u = [ u(x) ] ,
virtual displacements : δu = [ δu(x) ] ,
∂
operator matrix : D = [ ∂x
] = [ ∂x ] ,
elasticity matrix : E = [ EA ] ,
load vector : p = [ p(x) ] .
The operator matrix D comprises the rule of differentiation that is to be applied
to the shape functions, while the transposed operator matrix DT is applied to
the function on the left hand side.
The virtual displacements have to appear as transposed vectors regarding the
work equation if both the displacements and the external actions are represen-
ted by applying vector notation. That is the only way the virtual work can be
represented as having a scalar quantity.
Introducing all matrix symbols into the work equation yields
Z Z
− δWel = δuT · DT · E · D · u dx − δuT · p dx . (2.30)

Step Two: Shape Functions

The approach u(x) regarding the displacements of the bar
u(x) = φA (x) · uA + φB (x) · uB (2.31)
is transferred to matrix notation applying the matrix Ω, which incorporates
all shape functions, and the nodal displacement vector v, which comprises the
nodal degrees of freedom, yielding
 
uA
u = [ φA φB ] · or
uB
u = Ω·v. (2.32)
Hence the virtual displacements follow to
δu = Ω · δv , (2.33)
whereby the virtual nodal displacements may be arbitrarily small but finite.
36 2 Discretization of the Work Equation

Step Three: Integration of the Element’s Work

The description of the element’s work applying matrix notation may take place
either following the detailed representation according to Section 2.3.1 or by
applying Equation (2.30) and introducing the chosen symbols for the shape
functions
Z Z
− δWel = [ δvT · ΩT · DT ] · E · [ D · Ω · v ] dx − [ δvT · ΩT ] · p dx .

Since the nodal degrees of freedom v as well as the virtual nodal values δv are
independent of x, they can be excluded from the integral of work
Z Z
− δWel = δvT · { [ ΩT · DT ] · E · [ D · Ω ] dx · v − ΩT · p dx } . (2.34)

Moreover, the auxiliary matrix

B= D·Ω , BT = ( D · Ω )T = ΩT · DT

can be introduced to get a short form describing the field of strains. This yields
Z Z
− δWel = δvT · { BT · E · B dx · v − ΩT · p dx }
| {z } | {z }
K f

with the element stiffness matrix K, that is symmetrical if identical shape

functions u and δu are employed in the same order, the element load vector f
and the element displacement vector v. Assigning the element stiffness entries
as well as the load vector to the respective nodes yields the following pattern
clearly arranged:
" # " #
kAA kAB fA
K= , f= .
kBA kBB fB

The matrix notation of the virtual work offers the big advantage of being ap-
plicable to all physical problems. Only the contents of the respective matrices
and vectors as well as the integration region must be tentatively adjusted to
the particular problem being investigated.

Step Four: Change over to the Overall System

The work of the element is now defined by applying the matrix notation

− δWelement = δvT { K v − f } .
2.3 The General Procedure to set up the Element Matrices 37

After adding up the works regarding all the elements to get the system’s work,
and applying the same matrix symbols at the system level, the work equation
follows to

− δWsystem = δvT { K v − f } = 0 .

Here, K is the stiffness matrix of the system, f is the load vector of the system
and v is the displacement vector of the system. The procedure of adding up
the work with regard to all elements is explained in more detail in Section 3.
According to common practice with respect to the literature, no difference is
made between element matrices and system matrices or between the respective
vectors. In the following vectors will be characterized by small letters, matrices
by capital letters.
In addition to the elements’ works, the boundary conditions as well as the
external boundary actions must be taken into account, but may be considered
at the system’s level afterwards. This is advantegeous, because it does not
perturbe the element by element procedure to get the system’s stiffness matrix
and the load vector.
After assembling the system of equations to compute the system’s displacement
vector, v may be evaluated numerically by means of a standard equation solver,
cf. Section 3.

Step Five: Subsequent Analysis of Stresses

Applying the matrix notation the analysis of the longitudinal force follows from
Equation (2.29)
   
∂ uA
[ N ] = [ EA ] [ φA φB ] .
∂x uB

A symbol notation applies

the vector of strains ǫ = [ ε(x) ] ,

and the vector of stresses σ = [ N (x) ] .

Hereby, the computational rule regarding the strains follows to

ǫ = D·u = D·Ω·v = B·v, (2.35)

and regarding the stresses to

σ = E·ǫ = E·D·Ω·v = E·B·v. (2.36)

38 2 Discretization of the Work Equation

It is advantageous to multiply E and B a priori, what leads to

σ = S·v, (2.37)

whereat S is named the stress matrix.

S incorporates the governing equations and the shape functions for the com-
putation of the force variables within the element. Therefore, to get the longi-
tudinal forces, it only has to be multiplied by the related nodal displacements
v of the element, which must be extracted from the global solution vector.

If special values of stresses are of interest the coordinates of the related positions
have to be taken into account for the computation of the stress matrix S̃. Thus

σ̃ = S̃ · v (2.38)

no longer describes the polynomial approach of the stresses, but the specified
values.

2.4 Shape Functions and Convergence Criteria

The Finite Element Method substantially bases on the shape functions em-
ployed to approximately describe the displacement field inside the element’s
domain. Usually these functions may be polynomials of different orders and
are to be related to the dimensions of the element geometry. To improve the
quality of the approximate solution and to converge against the exact solution
– if it exists – either the order of the shape functions or the number of ele-
ments is to be increased. Thus, shape functions and convergence are strongly
correlated to each other.

2.4.1 Shape Functions

Polynomials may be selected by applying different representations according
to different scaling factors. In principle, all polynomials are equivalent if they
employ the same polynomial order and thus the same number of degrees of
freedom. Thus, the transformation of scaling factors from one set of factors to
another one is possible. An example is discussed in Section 2.3.1 and Equations
(2.22) and (2.25). Nevertheless, according to FEM, representations are only
appropriate when applying physically meaningful nodal degrees of freedom as
scaling factors.
The following groups of shape functions arise among others and may explain
the different representations:
2.4 Shape Functions and Convergence Criteria 39

1. General Polynomials
The most simplified representation of the displacement field consists of power
terms of x, which are named monomials. Thus the complete approach is given
by a general polynomial of specific order, e.g. a quadratic approximation of the
displacement u(x) is given in Figure 2-9.

A B
u(x) = a0 x0 + a1 x1 + a2 x2
l

x0 1 Φ 1(x)

Φ 2 (x)
x1 l

l2 Φ 3 (x)
2
x

Fig. 2-9 General polynomial of second order

Although the displacement field is approximated, the generalized coordinates

ai do not attach any physical meaning and are attributed to the origin of the
coordinate x. In the following, these polynomials are characterized as general
polynomials to make a clear difference to Lagrange and Hermite Polynomials.
Nonetheless, the transformation of the scaling factors ai to scaling factors with
a physical meaning could be done afterwards.

2. Legendre Polynomials
Another type of approximation without any physical meaning of the scaling
factors deals with Legendre Polynomials. Legendre Polynomials are shape func-
tions employing the special feature
Z
1 2 1 für i = j
φi (x) · φj (x) dx = · δij , δij = { ,
ℓ 2i + 1 0 für i 6= j

which means, that they are orthogonal with respect to the integral. This leads
to numerical advantages, when the virtual work equation has to be integrated.
These polynomials, however, do not exhibit any physically meaningful degree of
freedom. In this case the transformation of the scaling factors to scaling factors
with a physical meaning can be done after the integration of the virtual work.
40 2 Discretization of the Work Equation

Regarding the Legendre Polynomials according to Figure 2-10, the following

approach of the displacement field u(x) may be established:

A B
u(x) = a0 · φ0 + a1 · φ1 + a2 · φ2 + . . . ,
x
l l

φ0 (x) = 1 Φ 1(x)
1

1
x Φ 2 (x)
φ1 (x) =
ℓ

Φ 3 (x)
1
1 x2
φ2 (x) = ( 3 2 − 1)
2 ℓ
1
2
Fig. 2-10 Legendre Polynomials

3. Lagrange Polynomials
If shape functions are chosen that employ nodal displacements at designated
coordinates as scaling factors, so–called Lagrange Polynomials are obtained. In
Figure 2-11 quadratic Lagrange Polynomials are depicted.
A C B
u(x) = uA φ1 + uB φ2 + uC φ3 l l
2 2

x x2
φ1 (x) = 1 − 3 +2 2 Φ 1 (x)
ℓ ℓ 1

x x2 1
φ2 (x) = − +2 2 Φ 2 (x)
ℓ ℓ

x x2 1 Φ 3 (x)
φ3 (x) = +4 −4 2
ℓ ℓ

Fig. 2-11 Quadratic Lagrange Polynomials

2.4 Shape Functions and Convergence Criteria 41

The three scaling factors of the quadratic approach of the displacement field are
the displacements uA , uB , uC at the element nodes A, B, C. The shape functions
φ1 , φ2 , φ3 no longer keep single power terms, but functions, that describe the
influence of the related nodal displacement within the element. Thus, Lagrange
Polynomials are functions, which comprise the value of 1 in the node, that
keeps the unknown displacement, and a value of 0 in all other nodes.

4. Hermite Polynomials
Hermite Polynomials employ the nodal displacements as well as their higher
derivatives at the domain’s boundaries and further nodes inside the domain as
scaling factors. They are of advantage, if the intersection conditions between ele-
ments are described by means of higher derivatives of the displacement course.
Applying the cubic Hermite Polynomials according to Figure 2-12, this means:
u(x) = uA φ1 + u,x A φ2 + uB φ3 + u,x B φ4 .
A B
x
l
2 3
x x
φ1 (x) = 1 − 3 +2 3
ℓ2 ℓ 1 Φ1 (x)

x2 x3
φ2 (x) = x − 2 + 2 1 Φ2 (x)
ℓ ℓ

x2 x3
φ3 (x) = 3 −2 3 1 Φ3 (x)
ℓ 2 ℓ

x2 x3
φ4 (x) = − + 2 1 Φ4 (x)
ℓ ℓ

Fig. 2-12 Cubic Hermite Polynomials

Even higher order Hermite Polynomials may be employed if second or higher

order derivatives are chosen as nodal degrees of freedom. Since the nodal degrees
of freedom have different units, the Hermite Ploynomials have different units as
well. Hermite Polynomials are normalised functions, which comprise the value
of 1 with respect to the prevailing unknown, and a value of 0 with respect to
all other unknowns.
42 2 Discretization of the Work Equation

2.4.2 Criteria of Convergence

At mesh refinement the finite element approximation of the course of displa-
cements needs to converge towards the analytical strong solution, assuming it
exists. To ensure this convergence, the shape functions regarding the displa-
cements must fulfill special conditions which have been partially brought up
in Section 2.1.3 already. If these conditions – in the following named as con-
vergence criteria – are fulfilled, the convergence is ensured towards the exact
solution.
The displacement–based formulation of the FEM applies the PvD, therefore
the requirements with respect to the shape functions are correlated to the
work equation developed in Section 2.1.

Criterion of at least Constant Strains

At a minimum, the strains must be approximated constantly by the shape
functions regarding the displacements inside the element. The same applies to
the elongation ε regarding bars, or the curvature κ regarding beams. It is an
important requirement,
R without which the computation of the virtual internal
work δWi = − δεT· σ dV , given here in the most general matrix notation,
and of the stresses when applying a subsequent analysis σ = E · ε would not
be possible.

Concerning bars, the displacements must be approximated at least linearly to

be able to describe constant strains and constant longitudinal forces within the
element. This applies analogously, concerning the displacement fields at mem-
brane structures with respect to coordinates x and y, and for spatial structures
of three dimensions.
Regarding the Euler–Bernoulli beam the curvature κ = w,xx is applied, sin-
ce the effect of shear deformations is neglected. Therefore, at least quadratic
shape functions are required to describe the deflections. This requirement also
analogously applies for the course of the deflection w(x,y) regarding Kirchhoff ’s
plate theory.

Criterion of Conformity
The criterion of conformity requires continuity and moreover continuous diffe-
rentiability of the shape functions within the element up to the strains, which
are required to have a continuous course inside the element’s domain. Otherwi-
se additional work must be taken into account when applying the PvD. At the
element’s intersections the following conditions must be fulfilled and the same
is valid for the system’s boundary conditions:
2.4 Shape Functions and Convergence Criteria 43

• When assembling the elements all substantial intersection conditions re-

garding real displacements as well as virtual displacements must be ful-
filled. Regarding the bar, the continuity of the displacements is required.
Only by this the work performed by the internal force variables cancels
each other out at the element intersections.
Note: Regarding the PvD, the conditions concerning the displacements
are the substantial boundary conditions as well as intersection conditions.
They have to be directly fulfilled by the shape functions. The boundary
as well as intersection conditions concerning the force variables are fulfil-
led in their weak form which implicitly applies the PvD work equation as
shown above.
• If the boundary as well as the intersection conditions are fulfilled regar-
ding the displacements only, which formally means the zero order derivati-
ve of the displacements, this is called a C0 –conformity of shape functions.
Regarding a bar employing linear shape functions to describe u(x) and
δu(x), this condition may be fulfilled by the displacements ( uA , uB ) and
the virtual displacements ( δuA , δuB ), respectively. As a consequence the
strains as well as the longitudinal forces may incorporate some gaps at
the element intersections.
• Regarding the Euler–Bernoulli beam the deflection and its first derivati-
ve are substantial boundary as well as intersection conditions. Therefore
C1 –conformity of the shape functions is required, since one integration
constant corresponding to w and one corresponding to w,x follow from
kinematics. Analogous statements apply concerning the respective virtual
displacement variables. Since the beam element has to fulfill the condi-
tions at both element nodes A and B, the approach regarding w must
comprise of at least four degrees of freedom, e.g. ( wA , w,xA , wB , w,xB ),
which means at least a cubic approach. In contrast, bending moments as
well as shear forces may exhibit discontinuities at the element intersecti-
ons.
• Regarding membrane and plate structures, equivalent demands are made
on the approaches compared with the respective one–dimensional struc-
ture. Thus, membrane structures require C0 –conformity of shape functi-
ons, while plate structures require C1 –conformity.
• Conformity of shape functions regarding shell structures is only achievable
when employing high order shape functions. Therefore, if shape functions
are applied, which are not fulfilling the criterion of conformity, the work
performed at the element intersections has to be explicitly taken into
account.
44 2 Discretization of the Work Equation

To conclude, regarding the shape functions and their respective derivatives at

the element intersections, continuity is required up to the differential order that
is lowered by one compared to the order which appears in the work equation
concerning the element’s domain.

There might be some elements which cannot fulfill all element intersection con-
ditions a priori without increasing the polynomial order of the shape functions
considerably. Thus, when applying the work equation, it might be more mea-
ningful to explicitly consider the work performed by the internal force variables
at the element intersections.

Criterion of Rigid–Body Displacement

A rigid–body motion or rigid–body displacement does not describe a motion of
a rigid body, but a motion without strains and stresses. This means, that an
element may undergo a pure translation or/and rotation, cf. Figures 2-13 and
2-14, whereat strains and curvatures vanish.
A B A* B*
uA
uB
x=l
x
uA = uB ε = u’X = 0

Fig. 2-13 Rigid–body displacement of a bar

Since such displacement states may occur at any area inside a structure, de-
pending on the situation, the shape functions must fulfil this criterion right
from the start and must be able to describe these displacement fields a priori.

displacement
w and rotation

Fig. 2-14 Rigid–body displacement of a part of a beam structure

The demand from the strains and curvatures not to occur in the case of a
rigid–body displacement is generally fulfilled regarding plane structures such
2.4 Shape Functions and Convergence Criteria 45

as bars, membranes, beams or plates, if the shape functions are balanced on

each other, cf. Section 2.4.1.

In contrast, for curved structures like circular arcs and shell structures the
rigid body criterion is hardly to fulfil, because of the special kinematic relations
applying to curved structures. Regarding circular arcs, the strain ε additively
follows from the first derivative of u with respect to the curved coordinate s and
from the zero order derivative of w. If the same shape functions, for example
linear courses, are employed to describe both displacements, the condition of
rigid–body displacement without occuring strains cannot be fulfilled at every
single point, see Figure 2-15. Instead a linear strain and thus a linear stress
field occurs.
exact solution linear shape functions for u and w
πs s s
w(s) = wB · sin , w(s) = (1 − ) · wA + · wB ,
2ℓ ℓ ℓ
πs s s
u(s) = uA · cos , u(s) = (1 − ) · uA + · uB ,
2ℓ ℓ ℓ
w 1 1 s
ε = u,s + = 0. ε = − · uA + · · wB 6= 0 .
R ℓ R ℓ
A uA A*

s wA = 0
w
uB = 0
u R
s=l wB wB = uA
B B*

Fig. 2-15 Rigid–body displacement of a circular arc

Criterion of Coordinate Invariance

Regarding two– and three–dimensional structures, the shape functions must
be coordinate–invariant to avoid any influence of rotation of the coordinate
system on the deformation state both of the element and of the system. This
is achieved by ensuring the shape functions are equivalent with respect to all
directions. The easiest way to check the coordinate–invariance of shape functi-
ons is to employ Pascale’s triangle, see Figure 2-16. Pascale’s triangle indicates
the multiplicative combinations of power terms with respect to the x– and y–
directions, which are possible in principle. If all products are taken into account
up to a selected order, this is called a complete approach.
46 2 Discretization of the Work Equation

For example:

• u(x, y) = a0 + a1 · x + a2 · y
is a complete linear approach and

• u(x, y) = a0 + a1 x + a2 y + a3 x2 + a4 xy + a5 y 2
is a complete quadratic approach.

1
x y

x2 xy y2

x3 x2y xy 2 y3

x4 x3y x2y 2 xy 3 y4

symmetric approach of second degree

complete approach of second degree
complete approach of fourth degree

Fig. 2-16 Pascale’s triangle

Complete approaches are inherently coordinate–invariant. But to choose proper

shape functions the element geometry and the number of possible nodes should
be taken into account.

Triangular elements need complete approaches, because they should have in-
variant properties, since they may have different positions in the x–y–plane.
Furthermore, complete approaches match the number of possible nodes inside
triangles, because three unknowns are correlated to the three corners and six
unknowns are correlated to the corners and the midpoints of the edges.

Rectangular as well as quadrilateral elements usually have four, eight or nine

nodes, which do not match the number of the polynomials of complete approa-
ches. Thus, symmetric approaches with respect to the x– and y–coordinates are
chosen. Symmetric approaches arise if one–dimensional approaches of the same
order with respect to the x– and y–directions are multiplied by each other.
2.4 Shape Functions and Convergence Criteria 47

They are invariant against rotation of the coordinate system by 90o and multi-
ple thereof. Regarding two–dimensional plane structures, the approaches may
be taken from the respective one–dimensional structure therefor.

• Regarding membrane structures, it applies in analogy to bars:

Employing a linear approach with re-
spect to the x–direction and an equi-
valent linear approach with respect to 1
the y–direction, a bi–linear approach is x y
generated to describe the displacements xy
u(x, y) and v(x, y) that occupies 4 adja-
cent combinations in Pascale’s triangle.
• Regarding plate structures, it applies in analogy to beams:
The product of a cubic approach with
respect to the x–direction and the equi- 1
valent cubic approach with respect to x y
the y–direction generates a bi–cubic x2 xy y2
3 2 2
approach to describe the deflection x x y xy y3
w(x, y), which occupies the adjacent x3 y x2 y 2 xy 3
combinations in Pascale’s triangle. This x3 y 2 x2 y 3
3 3
leads to an approach employing 16 de- x y
grees of freedom.

2.4.3 Scaling Matrix

Regarding two– or more–dimensional elements employing higher order approa-
ches, difficulties often arise when trying to directly relate the shape functions to
the physically meaningful degrees of freedom, especially when applying triangu-
lar elements. In such cases, a more general procedure, which is also well–suited
to coding may be chosen to determine the shape functions. To approach the
course of displacements u = [ u(x) ] inside the element, at first the general
polynomial with the notation
u=ψ·a (2.39)
is chosen, where the vector ψ comprises the power terms of the approach cor-
responding to the coordinates x, y and a comprises the related scaling factors
ai , cf. Section 2.3.1 and Equation (2.22). The transformation of the general
polynomial into a physically meaningful one is divided into two steps.

At first, the relationship has to be established between the scaling factors a =

[ a1 a2 a3 . . . ] and the desired physically meaningful nodal degrees of freedom
v = [ uA uB uC . . . ]. Here, it could be necessary to provide the derivatives of
48 2 Discretization of the Work Equation

the general polynomials with respect to the coordinates x, y, if derivatives of

the displacements are selected to be nodal degrees of freedom, e.g. ϕ = w,x in
the case of beams and plates.

Applying the procedure for each physically meaningful nodal degree of freedom,
the respective nodal coordinates are successively introduced into the general
relation (2.39). This yields a system of equations which is comprised of the
relations between the physically meaningful nodal degrees of freedom v and
the scaling factors a of the general polynomial by numbers:
   
uA ψ̃ A
   
 uB  =  ψ̃ B  · a
... ...

and in matrix notation

v = Ψ̃ · a . (2.40)

Thus, the matrix Ψ̃ transforms the scaling factors a to the physically meaning-
ful nodal degrees of freedom v. Reversing Ψ̃ according to Equation (2.40) in a
second step, one obtains
−1
a = Ψ̃ ·v = G·v. (2.41)

Thus the vector of scaling factors a regarding Equation (2.39) may be described
directly as employing the nodal degrees of freedom. It follows that

u = ψ ·G·v = Ω·v. (2.42)

The matrix G is named as scaling matrix and scales the general polynomial
according to Equation (2.39) with respect to the physically meaningful nodal
degrees of freedom. The product ψ · G now describes the physically meaningful
shape functions Ω.

As an example, in the following Table 2.1, the scaling of a quadratic polynomial

with the general coordinates [ a1 a2 a3 ] is represented with respect to the nodal
displacements [ uA uB uC ] regarding a bar. The physically meaningful shape
functions evaluated by this procedure turn out to be the quadratic Lagrange
Polynomials according to Section 2.4.1.
2.4 Shape Functions and Convergence Criteria 49

Table 2.1 Relationship between the general polynomial

and the physically meaningful shape functions

direct approach scaling procedure

u=Ω·v u = ψ · a, a= G·v
shape functions general polynomial
 
Ω = [ φA φB φC ] ψ = 1 x x2
x 2
φA = 1 − 3 ℓ
+ 2 xℓ2
2
φB = − xℓ + 2 xℓ2
2
φC = +4 xℓ − 4 xℓ2

degrees of freedom
v T = [ uA uB uC ] aT = [ a0 a1 a2 ]
scaling matrix
v = Ψ̃ · a
     
uA ψ(x = 0) 1 0 0
 uB  =  ψ(x = ℓ)  · a =  1 ℓ ℓ2  · a
2
uC ψ(x = ℓ/2) 1 ℓ/2 ℓ /4
−1
a = Ψ̃ ·v = G·v
 
1 0 0
−1
G = Ψ̃ =  −3/ℓ −1/ℓ 4/ℓ 
2 2
2/ℓ 2/ℓ −4/ℓ2

shape functions
u=ψ·G·v
 
1 0 0
 
Ω = ψ · G = 1 x x2 ·  −3/ℓ −1/ℓ 4/ℓ 
2/ℓ2 2/ℓ2 −4/ℓ2
h 2 2 2
i
= (1 − 3 x + 2 x2 ) (− x + 2 x2 ) (4 x − 4 x2 )
ℓ ℓ ℓ ℓ ℓ ℓ
50 2 Discretization of the Work Equation

It is an essential fact that the integration of the element stiffness matrix and
of the load vector can be simplified employing the scaling matrix and may
be generally more efficiently established. Therefore, integration of the element
stiffness matrix K and the load vector f may be performed by employing the
general polynomial. After the integration procedure, the stiffness matrix as well
as the load vector are multiplied by the scaling matrix, which now can be done
numerically
Z
δvT · K · v = δvT · GT · ψ T · DT · E · D · ψ dA · G · v , (2.43)
Z
δvT · f = δvT · GT · ψ T · p dA . (2.44)

Even if the scaling matrix appears to be a detour from computing the physically
meaningful shape functions at first, it is often more advantageous as well as
numerically more efficient to follow the procedure given in Equations (2.43)
and (2.44), with regard to the programming of the stiffness matrix.
3 Structure and Solution of the System of Equations

In Section 2 the Principle of virtual Work is discretized at element level. Here-

with, the virtual work is described by the virtual and the real nodal displace-
ments, the stiffness matrix and the load vector.
The work performed at a single element can be incorporated into the work of the
entire system only if the conditions of transition at the element intersections are
considered. These are, as shown in Section 2.1.3, the conditions of equilibrium
and of deformation. The conditions of equilibrium are fulfilled weakly at system
level by the work equations addressing the Principle of virtual Displacements.
The conditions of deformation are fulfilled strongly for the virtual and the real
displacements, for example, shown in Figure 3-1 for a tensile bar.
p(x)

1 2 3 elements
1 2 3 4 nodes

u (x)
possible qualitative final solution

u1 u2 u3 u4 unknown real nodal displacements

δ u1 δ u2 δ u3 δu4 virtual nodal displacements

1 1 22 2 33 3 4
final solution inside single elements

1 1 shape functions for

real and virtual displacements
inside single elements
1 1

Fig. 3-1 Shape functions for the course of displacements at a tensile bar

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_3
52 3 Structure and Solution of the System of Equations

3.1 Real and Virtual Nodal Displacements

The unknown real displacement field of the tensile bar is approximated element–
wise with the shape functions and the related unknown nodal displacements.
In order to satisfy the conditions of the work equation, shape functions and no-
dal displacements must directly fulfill kinematics inside the element and at the
element intersections, cf. convergence criteria. Since the real displacements ha-
ve to be continous at the element boundaries, the nodal displacements of two
neighboring elements cannot be independent variables at the common node.
Therefore, they will be assigned as nodal displacements of the corresponding
node of the system. Thus, being the unknown multipliers of the shape func-
tions, the nodal displacements ui are the unknowns of the given problem and
completely determine the deformation state of the system.

Since the Principle of virtual Work is applied for each element, the virtual dis-
placements are also described for each element separately. As per Section 2.1.3,
kinematics at the element intersections must also be fulfilled by the virtual
displacements. This is only possible, if the virtual nodal displacements of the
elements are the same at the common nodes of neighboring elements, and thus
become virtual nodal displacements of the system. Hence the virtual work at
the common nodes of neighboring elements can be summarised.

Nonetheless, the polygon–shaped approximation of the course of the real as

well as of the virtual displacements of the tensile bar, as shown in Figure 3-1,
do not yet fulfill the displacement boundary condition. The boundary conditi-
ons will be considered later on in the global system of equations.

3.2 Equations of Condition

The Principle of virtual Displacements is equivalent to the conditions of equi-
librium, if the virtual displacements satisfy the kinematics, cf. Section 2.1.1.
Since the virtual work performed at the bar is the summation of the virtu-
al work performed on all chosen compatible virtual displacements, the virtual
work performed on each single virtual displacement must be zero. This leads
to a set of equations, which may be arranged in matrix–vector notation, em-
ploying the stiffness matrix, the load vector, and the vector of the real nodal
displacements. Choosing the same sequence for the virtual and for the real
displacements yields a symmetric stiffness matrix. Because the real nodal dis-
placements are the only unknowns according to the PvD, they can be computed
solving these equations, presuming the number of independent virtual displa-
cements corresponds to the number of unknown nodal displacements.
3.3 Structure of the System of Equations 53

Each row of the system of equations comprises the virtual work of the real
stress state on the respective virtual displacement, whereat the virtual nodal
displacements occur as multipliers of the corresponding work equations. Thus
the work equation can also be taken as the condition of equilibrium of the nodal
forces acting in the direction of the virtual nodal displacements.
Accordingly, the conditions of equilibrium do not need to be fulfilled strongly
at all differential elements, but only approximately employing shape functions
and nodal displacements.

3.3 Structure of the System of Equations

The treatment of the equations of condition for the global unknowns is repre-
sented for the example of the tensile bar, given in Figure 3-1, and considering a
discretization into three elements. At first, it is shown how to store the element
matrices. Without considering the boundary conditions, the given example em-
ploys four unknowns, u1 to u4 . The four equations of condition, which describe
the work performed by the real stress state on the related virtual displacements
δu1 to δu4 , are now available for their determination.

Assembling the entire system of equations element by element is particularly

convenient for coding, and is universally accepted. Therefore, the schematic
representation of the virtual work performed at a single element i, as shown
in Figure 3-2, can be used. Due to the two element nodes A, B the element
stiffness matrix K consists of 2 × 2 sub–matrices kAA , kAB , kBA , kBB and the
element load vector f of two sub–vectors fA and fB , respectively.
uA uB

i i i
δuA kAA kAB fA
i i
=K = f
δuB i i i
kBA kBB fB

Fig. 3-2 Stiffness matrix and load vector of the element i

Since the virtual work of every single element is computed independently from
each other, the summation of the virtual work of the elements to the total virtu-
al work of the system has to satisfy the kinematic conditions of the Principle of
54 3 Structure and Solution of the System of Equations

virtual Displacements afterwards. This will be performed in two steps according

to the real and the virtual shape functions. At first, the virtual work perfor-
med at every single element is not yet coupled with each other as schematically
represented in Figure 3-3.

u1 u2 u2 u3 u3 u4
1 1
δ u1 K f
δu 2

2 2
δu 2 K f
δu 3

3 3
δu 3 K f
δu 4

Fig. 3-3 The virtual work without considering the transition conditions

The transition conditions of deformation are fulfilled for the final state at the
element intersections, if the actual displacements ui at the common nodes of
the neighboring elements are the same and therefore considered as the nodal
displacements of the system:
u2 is same for elements 1 and 2,
u3 is same for elements 2 and 3.
The virtual work thus exhibits the scheme shown in Figure 3-4.

u1 u2 u3 u4
1 1
δ u1 K f
δu 2

2 2
δu 2 K f
δu 3

3 3
δu 3 K f
δu 4

Fig. 3-4 Conditions of deformation for the actual nodal displacements are fulfilled
3.4 Assembly of the Stiffness Matrix of the Entire System 55

Due to the Principle of virtual Displacements the conditions of deformation

must be fulfilled at every node concerning the virtual displacements δui as well:
δu2 is same for elements 1 and 2, and
δu3 is same for elements 2 and 3.
Thus the virtual work performed on the same virtual displacements is to be
summed up regarding the entire system, cf. Figure 3-5. This condition governs
the overlap at the element stiffness matrices and the load vectors.

u1 u2 u3 u4
1
δ u1 K f1
δu 2 K2 f2
δu 3 K3 f3
δu 4

Fig. 3-5 Fulfilment of conditions of deformation for the virtual nodal

displacements

Now the virtual work fulfills all kinematic conditions at the element intersecti-
ons. The columns of the stiffness matrix are to be multiplied by the unknown
real nodal displacements. Each row is to be multiplied by the respective virtual
displacement, whereupon the values of the variables δu1 to δu4 do not influence
the results, since they can be excluded from the related equation.

3.4 Assembly of the Stiffness Matrix of the Entire System

The work performed by the real stress state on all possible virtual displacements
of the entire system is summarized in the global system of equations. Since the
work performed on the entire system is equal to the sum of the work performed
on the single elements, the work performed at every single element can be stored
additively and sequentially into the respective rows and columns of the system’s
matrix, corresponding to the respective real and virtual nodal displacements.

The node–wise allocation of the work performed on an element is preserved in

the global system of equations, even if more than one unknown exists per node.
For the correct allocation of the work into the rows and columns of the global
stiffness matrix, the node numbers of the entire system must be correlated to
the node numbers of the element as well as to their sequence in the element.
56 3 Structure and Solution of the System of Equations

This information is summarized in the connectivity matrix. For the tensile bar
in Figure 3-1, this yields the connectivity matrix in Table 3.1:

Table 3.1 Connectivity matrix for the bar Figure 3-1

node
element A B
1 1 2
2 2 3
3 3 4

The Connectivity Matrix for 2–Dimensional Structures

For 2–dimensional structures related to heat conduction, membranes and pla-
tes, the sequence of nodes is different at the element– and at the system–level
respectively. Therefore the connectivity matrix is extremely important. Intro-
ducing a connectivity matrix corresponding to a generalized system with more
than two nodes per element, several unknowns per node and arbitrary numbe-
ring of the total system nodes are possible.
In the following, the problem is explained by means of the discretization of a
rectangular area into four rectangular elements each employing four element
nodes A, B, C, D. The partitioning of the rectangular area into four elements
could yield the element– and node–numbers as given in Table 3.2.

Table 3.2 Connectivity matrix for a plane structure

1 2 3 System with four elements,

connectivity matrix
1 2
element node A node B node C node D
4 5 6 1 1 2 5 4
2 2 3 6 5
3 4 3 4 5 8 7
8
4 5 6 9 8
7 9

The element stiffness matrix K of a single element may split up into the sub–
matrices klm , cf. Figure 3-6. The sub–matrices of the element matrix corre-
sponding to the nodal unknowns of the element are square matrices of the size
n×n, corresponding to the n real unknowns per node as well as to the n virtual
unknowns. The real and the virtual unknowns may comprise only one nodal
3.4 Assembly of the Stiffness Matrix of the Entire System 57

unknown as in the case of heat conduction, cf. Section 4, two nodal unknowns
as in the case of membranes, cf. Section 5, or four nodal unknowns as in the
case of plates, cf. Section 6.
Element matrix
vA vB vC vD

A B δvA kAA kAB kAC kAD

δvB kBA kBB kBC kBD
i
δvC kCA kCB kCC kCD
D C
δvD kDA kDB kDC kDD

Fig. 3-6 Correlation of the node–numbers and the element stiffness matrix

By means of the connectivity matrix, the 4 × 4 sub–matrices of the element

matrix will be additively and sequentially stored into the entire system matrix
of 9 × 9 sub–matrices. The storage is shown in Figure 3-7 for the element
number 2. Because the sequence of the node numbers at the element level and
the sequence of the node numbers at the system level do not coincide, the
element matrix will not be stored as a whole, but its sub–matrices are to be
placed carefully into the corresponding rows and columns depending on the
node numbers at the system level.

1 2 3 4 5 6 7 8 9

1
2 2 2 2
2 k AA k AB k AD k AC
2 2 2 2
3 k BA k BB k BD k BC
4
2 2 2 2 2
K = 5 k DA k DB k DD k DC
2 2 2 2
6 k CA k CB k CD k CC
7

Fig. 3-7 System matrix, wherein only element 2 is incorporated

58 3 Structure and Solution of the System of Equations

The procedure of building up the left hand side of the entire system of equa-
tions can be directly implemented into a finite element program, if the storage
considers all elements of the connectivity matrix. It can be transferred to the
system’s matrix and to the load vector as shown in Figure 3-8.

1 2 3 4 5 6 7 8 9

1
X X X X 1 X

2
X X X X 2 X

3 3
X X X X 4 X+
4 + + + +
X X X X X
K= 5 + + + + f= 5 +
6 6

7 + + + + 7 +
8 + + + + 8 +
9 9

X element 1 + element 3

element 2 element 4

Fig. 3-8 The assembly of the system matrix and the load vector

3.5 The Storage and Solution of the System of Equations

The structure of the element matrix and its storage as given in Figure 3-7 and
Figure 3-8 have some properties, which are of advantage regarding the required
store and the solution of the system of equations. Since the element matrices
are symmetric and are stored symmetrically the resulting system matrix is also
symmetric. Thus, just the diagonal and half of the matrix have to be compu-
ted and stored, if computing time and storage capacity have to be reduced.
Furthermore, the system matrix is filled with zero values outside the employed
diagonals, which are of no influence on the solution process. Neglecting the zero
diagonals reduces the required storage capacity as well as the computing time.

The number of the non–zero diagonals depends on the connectivity of the

node–numbers and the unknowns of the nodes, cf. Figure 3-8, and is named
bandwidth (BW). The bandwidth of the system matrix is defined to the number
of diagonals with non–zero elements in the upper triangle, including the main
3.6 The Optimization of the Bandwidth 59

diagonal. Thus the bandwidth can be determined by means of the connectivity

matrix:

BW = (the maximum difference of node numbers in an element + 1)

× (number of unknowns per node).

Because of the symmetry, only the upper– or lower–triangular matrix is needed.

Since the components of the matrix lying outside the bandwidth have zero
values, they do not have to be considered in the solution process. Thus the
system matrix will be stored in such a manner that the main diagonal on the
left hand side is stored in the first column of a rectangular matrix with BW as
the number of columns, whereby only the components of the upper triangular
matrix will be stored at all, as shown in Figure 3-9.

BW = 5 · ( nodal unknowns )
z }| {
   
⊗ ∗ ∗ ∗ ⊗ ∗ ∗ ∗
 ∗ ⊗ ∗ ∗ ∗ ∗   ⊗ ∗ ∗ ∗ ∗ 
   

 ∗ ⊗ ∗ ∗ 
 −→ 
 ⊗ ∗ ∗ 

 ∗ ∗ ⊗ ∗ ∗ ∗  Utilisation of  ⊗ ∗ ∗ ∗ 
   
 ∗ ∗ ∗ ∗ ⊗ ∗ ∗ ∗ ∗ 
 symmetry and ⊗ ∗ ∗ ∗ ∗ 

  
∗ ∗ ∗ ⊗ ∗  bandwidth
∗  ⊗ ∗ ∗
  
  

 ∗ ∗ ⊗ ∗ 
 −→ 
 ⊗ ∗ 

 ∗ ∗ ∗ ∗ ⊗ ∗   ⊗ ∗ 
∗ ∗ ∗ ⊗ ⊗

Fig. 3-9 The storage of the symmetric system matrix

Since a symmetric system of equations does not only reduce the storage ca-
pacity needed but also considerably reduces the computing time, the order of
magnitude of the effort proportionally accounts for n · B W 2 , whereat n is the
number of unknowns of the entire system.

3.6 The Optimization of the Bandwidth

In a few cases the geometry of the structure allows a structured element mesh
employing a node numbering, which can be developed manually. But very often
an unstructured mesh or a mesh refinement is necessary, if strong gradients of
the stress fields locally occur. Thus a situation may be quickly reached, wherein
it no longer makes sense to manually partition a structure into finite elements
including the allocation of element and node numbers.
60 3 Structure and Solution of the System of Equations

Dealing with more complex discretizations, preprocessing is needed wherein

the FE–meshes can be generated automatically including the assembly of the
connectivity matrix. In the literature, different algorithms for the development
of unstructured meshes are offered as the Advancing–front method see for exam-
ple Lo [63, 64], George [39], and Löhner [65, 66], and the Delauny triangulation,
see Ruppert [89] and Shewchuk [90].
If the mesh development is finished, an optimization of the node numbering is
able to rigorously reduce the numerical effort for solving the system of equati-
ons. These algorithms are able to reduce the bandwidth and the storage capaci-
ty by renumbering the system nodes with respect to minimal numerical effort,
see Cuthill and McKee [32] and Liu and Sherman [62]. Here, the optimization
of the node numbering may be part of the preprocessing, where it influences
all kind of data with respect to the node numbers as the connectivity matrix,
the boundary conditions, the loading etc.

Though the disadvantage of an arbitrary node numbering is not evident at first,

it will become quite clear when regarding the allocation of the global stiffness
matrix, cf. Figure 3-10. The stiffness values of the element nodes are incorpo-

arbitrarily node numbering optimized node numbering

Fig. 3-10 Distribution of matrix elements of the system matrix

rated into the system matrix as related to their current system node numbers.
Hence, arbitrary allocation of the node numbers results in the system matrix
without any diagonal structure as shown in Figure 3-10–left. Since the compu-
ting time is related to n · BW 2 and the needed storage capacity with respect
to n · BW a renumbering of the nodes is necessary, which may significantly
3.7 The Fulfillment of Dirichlet Boundary Conditions 61

decrease the storage capacity required as well as the computing time needed.
The system matrix presented in Figure 3-10–right particular clearly shows the
effects of optimization of node numbering with regard to the bandwidth.

3.7 The Fulfillment of Dirichlet Boundary Conditions

In the Finite Element Method developed using the PvD, the Neumann boundary
conditions do not have to be fulfilled separately. They are fulfilled weakly by
the work equation, see Section 2.1.2.
The Dirichlet boundary conditions are to be fulfilled strongly by presetting the
related nodal unknowns to zero, as shown in Figure 3-1. In the given example,
Figure 3-1, the first column with u1 = 0 and the first row with δu1 = 0 are
affected. For the efficiency of the solution procedure, the reduction of the size
of the system matrix as well as of the number of unknowns is not that essential.
It is more important to preserve the structure of the FE–program as simple
as possible. Hence, all nodes with prescribed boundary conditions are formally
kept. Therefor two possibilities shall be looked at:

1. The values of the components of the respective rows and columns are set
to zero which is similar to cancelling out the related rows and columns.
Since this yields a singular matrix, the values on the related main diago-
nals are to be set to ’1’, whereby the related unknown will be formally
computed to zero.
2. A fixed constraint can be considered to be a spring with high stiffness.
The virtual work of the spring’s force can be introduced into the system
matrix by adding the spring stiffness to the value of the respective main
diagonal component.

Inhomogeneous boundary conditions with a certain value of the unknown 6= 0,

for example a preset displacement, can also be easily considered.

1. Before the related row and column are replaced by zero values, the related
column is to be multiplied by the prescribed value of the unknown and
is to be subtracted from the right–hand side. Afterwards, the addressed
row and column are set to zero and the element at the main diagonal is
set to be ’1’. The value of the corresponding element in the load vector
is then replaced by the prescribed value.
2. The product of the prescribed displacement value and the spring stiffness
is to be stored in the load vector.
It should be noted that the boundary conditions does not affect the symmetry
and the storage of the system’s matrix either.
4 Heat Conduction

Heat conduction is a quite another physical phenomenon in comparison to the

bar of Section 2. Concerning the building practice, an analysis of temperature
fields is required according to completely different tasks. Regarding buildings,
thermal bridges are analysed as well as the efficiency of heat exchangers, the
thermal curing of bulky concrete devices, and the heat conduction in the case of
fire etc. Foundation ground freezing is applied to stabilise the soil if the ground
is not sufficiently strong to perform a tunnel excavation or similar actions, or
if partly–saturated soils are present.
Regarding the FEM heat conduction gives a very comprehensive introduction
into the discretization of multi–dimensional field equations. As such, the tasks
to be worked on concern the type of description of the phenomenology as well as
its introduction into the FE–algorithm. For convenience, the terminology will
firstly be introduced regarding one spatial dimension, then it will be extended
to two spatial dimensions and finally transferred to the FE–context. The units
of the quantities describing the heat conduction are [K] = [K elvin] for the
temperature, [W ] = [W att] for the power and [J] = [J oule] for the energy.

4.1 Heat Conduction at One–Dimensional Description

A one–dimensional domain is given, that is heated at a boundary by an external
temperature field Tb or inside the domain by a heat source qv . While the heating
takes place, a part of the heat is conducted into and transported through the
domain, whereby the temperature T [K] within the domain increases.

4.1.1 Governing Equations for One Dimension

The heat transport is designated as heat flux q [W/m2 ] or as q [J/s m2 ] employ-
ing different units, which can be exchanged by means of W = J/s. The greater
the difference in temperature regarding neighboring points in the domain, the
greater the heat flux will occur. Thus in each point of the domain, Fourier’s
model of heat conduction applies

q(x) = −λ · T,x , (4.1)

that can be designated as the first governing equation. The negative sign cap-
tures the heat flux from hot to cold. λ [W/m K] or as [J/s m K] is the material
parameter describing the thermal conductivity, cf. Figure 4-1.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_4
64 4 Heat Conduction

z dx/2 dx/2
y T
dy
dT

x dz q (x) x
x

Fig. 4-1 Temperature gradient and heat flux

If the amount of heat is not totally transported through the domain, a part of
it may be stored inside the material. The stored energy ψ [ J/m3 ] depends on
the temperature as well as on the material density ρ [kg/m3 ], and on the heat
storage capacity or heat capacity c [W s/kg K] or c [J/kg K] respectively. This
relation leads to the second governing equation to describe the stored heat or
the stored energy
ψ(x, t) = ρ · c · T . (4.2)
At this, c describes the amount of heat that is required to warm up 1 kg of the
material by 1 K. The stored energy depends on the position x and the time t.

q-dq/2 q+dq/2
ψ , qv

0 x

dx

Fig. 4-2 Energy balance of stored energy and heat flux

The third governing equation describes the balance of energy, cf. Figure 4-2. If
one part of the heat is transported and one part is stored, the sum of the parts
changed in space and in time needs to be balanced. Considering that the heat
is not only transported from the boundary into the domain but also develops
as qv (x, t) in the domain itself – for example due to chemical reactions as a
result of hydration as the concrete hardens – it applies

ψ̇ + q,x = qv . (4.3)

The balance of energy confirms nothing else but the fact that the sum of heat
storage and heat flux might equal the heat supplied to the system. Introducing
the first and the second governing equation into the balance of energy gives

(ρ · c · T )˙ − (λ · T,x ),x = qv .
4.1 Heat Conduction at One–Dimensional Description 65

In the following, only the stationary situation will be investigated, so the deri-
vatives with respect to the time disappear. Thus it applies

−(λ · T,x ),x = qv . (4.4)

4.1.2 Boundary Conditions

The governing equations comprise two derivatives with respect to the coordi-
nate x, thus two boundary conditions need to be formulated. These are the
Dirichlet boundary condition describing the temperature Tb and the Neumann
boundary condition describing the heat flux qb

T − Tb = 0 , q − qb = 0 . (4.5)

Here, Tb and qb are the prescribed values at the boundary. In the case of tem-
perature and heat flux being combined at the boundary, the Cauchy boundary
condition needs to be fulfilled

q = −h · (T∞ − T ) (4.6)

with h [W/K m2 ] being the coefficient of heat transfer. This includes the influ-
ence of thermal convection at the boundary, when T∞ is the outside tempe-
rature. Here, T and q are not prescribed but the result of the complete heat
conduction process.
If radiation at the surface of a structure is taken into account, the boundary
condition deals with T 4 and needs algorithms to solve the nonlinearity. Thus,
this type of boundary condition is not discussed here.

4.1.3 Weak Form of the Energy Balance Equation

Unlike describing elastic structures, here the Principle of virtual Displacements
can not be applied to develop an alternative to the governing equations. In com-
plete analogy to the bar, a Method of weighted Residuals (WR) may be defined,
which deals with a weak formulation of the describing differential equation. The
differential equation describing the heat conduction

− q,x + qv = 0 (4.7)

applies for each differential element. If the equation is multiplied by a weighting

function, which could be interpreted as virtual temperature, it follows to

δT {−q,x + qv } = 0 . (4.8)
66 4 Heat Conduction

After integration over the entire domain, the weak form of the energy balance
equation follows to Z
δT {−q,x + qv } dV = 0 , (4.9)
which can be designated as weighted residual δR. Respective consideration of
the boundary conditions dealing with virtual heat fluxes and the surface S gives
Z Z
δR = δT {−q,x + qv } dV + {[ δq (T − Tb ) ] + [ δT (q − qb ) ]} dS = 0 . (4.10)

Integration by parts of the first term of the integral yields

Z Z
δR = {δT,x q + δT qv } dV + {[ δq (T − Tb ) ] + [ −δT qb ]} dS = 0 . (4.11)

Taking into account Fourier’s model of heat conduction and presuming that
the temperature (T − Tb ) and the virtual temperature δT exactly fulfill the
boundary conditions, it follows
Z Z
−δR = {δT,xλ T,x − δT qv } dV + δT qb dS = 0 . (4.12)

This equation fully complies with the Principle of virtual Displacement regar-
ding the bar, thus the weak form may be interpreted as Principle of virtual
Temperature. If a Cauchy boundary condition is to be taken into account, the
weak form is to be extended accordingly:
Z Z
−δR = {δT,x λ T,x − δT qv } dV + δT [ −h(T∞ − T ) ] dS = 0 . (4.13)

The discretization, which is now enabled by applying the Finite Element Me-
thod, happens in complete analogy to the bar. However, here the unit of δR is
[W ] and converting by [W = N m/s] may result in [N m/s] only if the unit of
δT is 1.

4.1.4 Example of use

The temperature field regarding a wall may be evaluated as follows. Within
the domain heat production is neglected by assuming qV = 0.
Considering the heat transfer between
the surface of the wall and the ambience 0 l x
results in the Cauchy boundary con-
ditions, which are taken into account as T∞ 0 T0 T1 T∞ 1

x=0 → q0 = −h (T0 − T∞0 ) ,

h h
x=ℓ → qℓ = −h (T∞1 − T1 ) .
4.1 Heat Conduction at One–Dimensional Description 67

The course of the temperature is to be found concerning the domain. Employing

a cross section of 1 m2 and absence of heat conduction in y– and z–direction,
the weak formulation may be simplified to
Z
−δR = {δT,x λ T,x − δT qV } dx + δT [ −h(T∞ − T ) ]10 = 0 ,

whereby the matrix notation is basis for the FE–description

Z Z
− δRel = δuT · DT · E · D · u dx − δuT · p dx .

Hereby, the heat conductivity is described with E = [ λ ], and the differentiation

of the temperature field with D = [ −∂x ], and the heat production within the
domain with p = [ qv ]. The Cauchy boundary conditions are taken into account
at the system’s level later on.
Proper shape functions have to satisfy the convergence criteria, which lead to
linear shape functions to describe the temperature field inside the element.
Scaling the general polynomial

T (x) = a0 + a1 · x

to the nodal temperatures TA and TB leads to

x x
T (x) = TA · ( 1 − ) + TB · ( ) ,
ℓ ℓ
cf. Figure 4-3, what can be transferred to the matrix notation of the tempera-

A B
x
l

1 TA ( 1 - x / l )

1 TB ( x / l )

Fig. 4-3 Shape functions for the description of the temperature field

ture field as
 
x x TA
[ T (x) ] = [ ( 1 − ) ( )] · ,
ℓ ℓ TB
u = Ω·v,
68 4 Heat Conduction

and also of the virtual temperature field

δu = Ω · δv .

Now, the integration of the weighted element residual can be performed ana-
logously to Section 2.3.2. Indroducing the shape functions into the weighted
residual gives
Z Z
− δRel = δvT · { [ ΩT · DT ] · E · [ D · Ω ] dx · v − [ ΩT ] · p dx } ,

what can be simplified by means of the auxiliary matrix

B= D·Ω , BT = ( D · Ω )T = ΩT · DT

as
Z Z
− δRel = δvT · { BT · E · B dx ·v − ΩT · p dx } .
| {z } | {z }
K f

Employing the linear shape functions the residual can be written with element
matrix K, element heat production vector f, and element vector of nodal tem-
peratures v as
" # " # " #
λ/ℓ −λ/ℓ TA qV ℓ/2
− δRel = [ δTA δTB ] · { · − }.
−λ/ℓ λ/ℓ TB qV ℓ/2

If just one element 0 ≤ x ≤ ℓ is taken into account, the element matrix equals
the system matrix. Considering qV = 0 and Cauchy boundary conditions gives
" # " #
λ/ℓ −λ/ℓ T0
− δRsystem = [ δT0 δT1 ] · { ·
−λ/ℓ λ/ℓ T1

+ δT0 [ −h ( T∞0 − T0 ) ] + δT1 [ −h ( T∞1 − T1 ) ] = 0 .

Introducing the boundary conditions into the matrix notation yields

" # " # " #
λ/ℓ + h −λ/ℓ T0 h · T∞0
− δRsystem = [ δT0 δT1 ] · { · − } = 0.
−λ/ℓ λ/ℓ + h T1 h · T∞1
4.2 Heat Conduction Regarding Two Spatial Dimensions 69

Solving the equations towards T0 and

T1 gives the unknown nodal tempera- T∞ 0 T0 T1 T∞ 1
tures
1 hℓ
T0 = {(1 + ) T∞0 + T∞1 } , h h
2 + h ℓ/λ λ
1 hℓ
T1 = {(1 + ) T∞1 + T∞0 } T
2 + h ℓ/λ λ
T0 T∞ 1
and therefore the linear distribution of T∞ 0 T1
the temperature inside the wall – see
the figure on the right – 0 - 0+ 1 - 1+ x

1 hℓ x hℓ x
T (x) = {[1 + (1 − )] T∞0 + [1 + · ] T∞1 } .
2 + h ℓ/λ λ ℓ λ ℓ

4.2 Heat Conduction Regarding Two Spatial Dimensions

Generally, heat conduction is a three–dimensional phenomenon that may be
reduced to a two– or a one–dimensional problem under specific conditions. Here
the two–dimensional heat conduction shall be described up to the discretization
of the weak form applying the FEM.
Regarding two-dimensional heat conduction, all governing equations as well as
the weak form need to be described in the x–y–subspace. Figure 4-4 shows the
sign rule concerning the heat fluxes qx and qy respectively.

y x dq
qy - 2 y
dq
qx - 2 x
qv , ψ
dq
qx + 2 x
dq
qy + 2 y dz
dy
dx

Fig. 4-4 Sign rule

External heat sources can be point– or line–shaped as well as surface–orientated

or volumetric. Regarding practical purposes, these could be e.g. welding anodes,
solar radiation or hydration heat when concrete hardens.
70 4 Heat Conduction

4.2.1 Governing Equations and the Weak Form

The governing equations – energy balance E and Fourier’s model of heat con-
duction F – are given below and, for comparison, the energy balance equation is
given applying the method of weighted residuals as well as the weak form, which
implicitly comprises all governing equations. The temperature field T (x, y) and
the heat fluxes qx (x, y) and qy (x, y) are the related variables. The governing
equations as well as the integral equations, which are basis for the discretiza-
tion, are summarized including the Cauchy boundary conditions. Here it is
postulated that heat conduction does not take place in z–direction. Thus the
description can be reduced to x–y–plane.
a) governing equations without heat storage (ψ̇ = 0)
E : − qx,x − qy,y + qV = 0
F : qx + λ · T,x =0
qy + λ · T,y =0
b) weighted residual
Z Z
δR = δT ( −qx ,x − qy ,y + qV ) dA + δT [−h(T∞ − T )] ds = 0

c) weak form
Z Z Z
− δR = λ ( δT,x T,x + δT,y T,y ) dA − δT qV dA − δT [−h (T∞ − T )] ds .

4.2.2 Matrix Representation of the Weak Form

The element matrix as well as the element load vector are derived regarding the
weak form, applying the matrix representation according to Equation (2.34)
Z Z
− δR = δvT · { ΩT · DT · E · D · Ω dA · v − ΩT · p dA } (4.14)
A A
| {z } | {z }
element matrix element load vector
or regarding the abbreviated form:
Z Z
− δR = δvT · { BT · E · B dA · v − ΩT · p dA } . (4.15)
A A

Due to their clear arrangement, the element matrices are not derived from the
weak form, but, applying the matrix representation, are developed according
to the bar. Therefore, the matrix symbols need to be linked to the context
described by the governing equations.
4.2 Heat Conduction Regarding Two Spatial Dimensions 71

4.2.3 Heat Conduction Matrix E and Operator Matrix D

Fourier’s model of heat conduction describes the relation between the heat
flux and the derivatives of the temperature field. A value corresponding to the
strains regarding elasticity is possible but uncommon. Thus the heat conduction
is directly represented by q and T . Employing the governing equations defined
in Section 4.2.1 it applies
σ = E · D· u,
" # " # " #
qx +λ −∂x  
= · · T . (4.16)
qy +λ −∂y
The vector u now comprises the temperature field T , which is described by
employing the shape functions and the nodal temperatures.

4.2.4 Linear Shape Functions Regarding the Temperature

The choice of the shape functions regarding the temperature follows from the
convergence criteria. The number of nodes employing nodal degrees of freedom
with respect to the temperature and their placing inside the element complies
with the element geometry, since all nodes need to be considered equally, see
Figure 4-5.

ly A lx

D B
x
y C

Fig. 4-5 Geometry and coordinate system regarding the rectangular element

The convergence criteria show that a linear symmetric approach regarding

T (x, y) and δT (x, y) is sufficient. In the following this simplest possible ap-
proach is chosen. Regarding the mathematical formal derivation, the tempe-
rature field is approximated by symmetric generalized polynomials of the first
order according to the x–y–coordinates employing the coefficients ai being ini-
tially unknown:
T (x, y) = a0 + a1 x + a2 y + a3 xy . (4.17)
The symmetry of the approach with respect to the coordinates may be control-
led by applying Pascale’s triangle according to Figure 2-16.
72 4 Heat Conduction

It is possible to transform the general polynomial into the physically meaningful

approach, however it is not shown here. A product approach employing linear,
physically meaningful shape functions and the corresponding nodal degrees of
freedom, well–known from the one–dimensonal heat conduction problem, achie-
ves this purpose more directly. The bi–linear shape functions are illustrated in
Figure 4-6, whereby the nodal temperatures TA , TB , TC , TD are the correlated
physically meaningful unkowns.

1
A
1
B

1
D
1
C

Fig. 4-6 Linear shape functions regarding rectangular elements

Therefore, the approach regarding the temperature fields T and δT is given

with
T (x, y) = φA · TA + φB · TB + φC · TC + φD · TD ,
(4.18)
δT (x, y) = φA · δTA + φB · δTB + φC · δTC + φD · δTD .

If the primary nodal unknowns of the element are arranged in the vector v by
nodes
 
v T = TA TB TC TD , (4.19)
the shape functions are to be merged according to the matrix Ω by the sequence
of the primary nodal unknowns regarding the temperature field T
 
Ω= φA φB φC φD . (4.20)

Table 4.1 explicitly represents the shape functions and their derivatives with
respect to the coordinates.
4.2 Heat Conduction Regarding Two Spatial Dimensions 73

Table 4.1 Bi–linear shape functions regarding rectangular elements

node shape function derivative derivative

i φi φi,x φi,y
x y y x
A (1 − lx
) · (1 − l ) − l1 · (1 − ly
) (1 − lx
) · (− l1 )
y x y

x y 1 y x
B lx
· (1 − ly
) lx
· (1 − ly
) lx
· (− l1 )
y

x y 1 y x 1
C lx
· ly lx
· ly lx
· ly
x y y x
D (1 − lx
)· l − l1 · ly
(1 − lx
) · l1
y x y

Thus the approach regarding the temperature follows to

u = Ω·v, (4.21)

and analogously to the virtual temperature

δu = Ω · δv . (4.22)

4.2.5 Differentiation of the Temperature Field

The auxiliary matrix
B= D·Ω (4.23)
incorporates the differentiated shape functions according to the differentiation
rule D of Equation (4.16) and should be introduced into the weighted residual.
For the chosen bi–linear shape functions matrix B reads as follows
" #
−φA,x −φB,x −φC,x −φD,x
B= . (4.24)
−φA,y −φB,y −φC,y −φD,y
The derivatives of the shape functions applied at B are given in Table 4.1.

4.2.6 Element Matrix

Applying the matrices B and E the element matrix may be computed according
to Equation (4.15) Z Z ℓy ℓx
K= BT · E · B dx dy . (4.25)
0 0
The integrand comprises a quadratic form resulting in a 4 × 4–matrix. The
multiplication scheme is given in Figure 4-7. The columns of the integrand are
74 4 Heat Conduction

assigned to the real nodal temperatures and the rows to the virtual nodal tem-
peratures respectively. The symmetry of the integrand’s matrix entries follows
from its quadratic form and the choice of identical shape functions for the real
and virtual temperatures. Therefore the element matrix is symmetric, too.

* O B
E O *

BT BT E B

Fig. 4-7 Multiplication scheme concerning the element matrix

To get a clear view, the integrand is multiplied beforehand employing the fol-
lowing scheme and is integrated afterwards:

BT · E · B = λ · I . (4.26)

The integration matrix I is given below whereupon the differentiated shape

functions are given in Table 4.1. The products applied regarding the matrix I
are constant according to the respective direction, if both shape functions are
partially differentiated with respect to this direction, and quadratic according
to the other direction since two linear functions are multiplied by each other.
 
φA,x φA,x φA,x φB,x φA,x φC,x φA,x φD,x
 + φA,y φA,y + φA,y φB,y + φA,y φC,y + φA,y φD,y 
 
 
 φ φ φ φ φ φ φ φ 
 B,x A,x B,x B,x B,x C,x B,x D,x 
+φ 
 B,y φA,y + φB,y φB,y + φB,y φC,y + φB,y φD,y 
I=  

 φC,x φA,x φC,x φB,x φC,x φC,x φC,x φD,x 
 
 + φC,y φA,y + φC,y φB,y + φC,y φC,y + φC,y φD,y 
 
 
 φD,x φA,x φD,x φB,x φD,x φC,x φD,x φD,x 
+ φD,y φA,y + φD,y φB,y + φD,y φC,y + φD,y φD,y

According to the Equations (4.25) and (4.26), the element matrix K follows
after the integration of I and is given by the representation below. As already
explained regarding bars, the sums of single columns as well as rows of the
4.2 Heat Conduction Regarding Two Spatial Dimensions 75

element matrix disappear, since a constant temperature field without heat flux
needs to be represented.
 ℓ ℓ ℓ ℓy

2 ℓy + 2 ℓℓx −2 ℓy + ℓℓx − ℓy − ℓℓx ℓx
− 2 ℓx
ℓy 
 x y x y x y
 
ℓy ℓx ℓy ℓx ℓy ℓx ℓy
− ℓ − ℓℓx 
 
 −2 ℓ + ℓ 2 ℓ + 2 ℓ −2 ℓ
λ  x y x y ℓ x y x y 
K= · 
6  ℓy ℓy ℓy ℓy
 − ℓ − ℓℓx ℓx
− 2 ℓx
ℓy
2 ℓx
+ 2 ℓx
ℓy
−2 ℓx
+ ℓx 

ℓy 
 x y
 
ℓy ℓx ℓy ℓx ℓy ℓx ℓy ℓx
ℓ
− 2 ℓ
− ℓ
− ℓ
−2 ℓ
+ ℓ
2 ℓ
+ 2 ℓ
x y x y x y x y

4.2.7 Element Vector of Thermal Action

Due to Equation 4.15 the vector of thermal action follows
Z
f= ΩT · p dA
A
and can be evaluated, if the spatial course of the external action is given. Here,
a constantly distributed heat source qV inside the element is assumed. Applying
 
p = qV = const.
p can be excluded from the integral
Z
f= ΩT dA · p . (4.27)
A
Employing the matrix of shape functions
 
φA
φ 
 B
ΩT = 


 φC 
φD
and the shape functions according to Section 4.2.4 the integration gives the
value ℓx ℓy /4 regarding all shape functions:
   
fA 1
 f  ℓ ℓ  1   
 B  x y  
f = = ·  · qV .
 fC  4  1 
fD 1
A physically meaningful interpretation of the element vector of thermal action
is possible if the integral of the heat source qV is interpreted as four nodal
sources of equal size.
76 4 Heat Conduction

4.2.8 Subsequent Flux Analysis

The temperature field is known from the computation of the nodal tempera-
tures of the entire system, as shown in Section 3. If the heat fluxes are of
interest, a subsequent flux analysis additionally needs to be performed. There-
fore, Fourier’s model of heat conduction

qx + λ · T,x = 0 ,
qy + λ · T,y = 0

is evaluated element by element, applying the temperature field now known.

Since the nodal temperatures are assigned to the shape functions in the related
element, the respective temperature field may be represented as a course of
function.
However, the courses of function of the heat fluxes are not usually of interest,
but discrete values at characteristic points are, where element nodes as well
as other positions may be preferred. This can be achieved by introducing the
respective coordinates into the course of function. The matrix, representing
the heat fluxes at the chosen nodes depending on the nodal temperatures, is
designated as matrix S̃, as known from one–dimensional structures. The course
of the heat flux inside the element is formally given here by σ, which requires
the matrix representation defined in Section 2.3.2 to:

σ = E · ǫ,
σ = E · D· u, (4.28)
σ = E·D·Ω·v = E·B·v = S·v.

If the heat conduction matrix E is constant, the course of function of the heat
flux σ already follows with the course of function regarding B.

The special contents of the matrices required for the evaluation are known
from the derivation of the element matrices. Here the stress matrix S describes
the heat fluxes normalized to the nodal temperatures. Employing the shape
fuctions given in Section 4.2.4 yields
" #
φA,x φB,x φC,x φD,x
S = E · B = −λ · ,
φA,y φB,y φC,y φD,y

whereat the derivatives of the shape functions are included in Table 4.1. The
stress matrix S reveals the course of the heat fluxes inside the element, con-
stantly related to the respective direction and linearly perpendicular to it.
4.3 Example of Use 77

The nodal values σ̃ of the heat fluxes may be evaluated after the nodal coor-
dinates have been introduced into B. Thus the discrete stress matrix S̃ follows
with the discrete auxiliary matrix B̃:

σ̃ = E · B̃ · v ,
σ̃ = S̃ · v . (4.29)

In detail the heat fluxes according to the element nodes may be evaluated to

σ̃ = S̃ · v ,
   
σ(x = 0 , y = 0 ) S(x = 0 , y = 0 )
  
σ(x = ℓx , y = 0 )   S(x = ℓx , y = 0 )
  

 = ·v,
 σ(x = ℓx , y = ℓy )   S(x = ℓx , y = ℓy ) 

σ(x = 0 , y = ℓy ) S(x = 0 , y = ℓy )

and with the chosen bi–linear shape functions to

 
1 1
 " #   − ℓx ℓx 
 
 qx   −1 1 
   ℓy ℓy 
 qy  



A 
 "
 #   −1 1   
 q   ℓx ℓx 
 x    TA
 qy  
 − ℓ1 1
ℓy
 
  
 B
  y   TB 
 = −λ ·  ·
 " #  .
1
− ℓ1
 TC 

 qx 


 ℓx x
 
 
  TD

 qy

  − ℓ1 1 
 " C   y ℓy 
 #   
qx 
1

− ℓ1
   
   ℓx 
x
qy  
D  −1 1 
ℓy ℓy

4.3 Example of Use

In the following, the heat conduction inside a wall is simulated employing rec-
tangular elements as explained above. Figure 4-8 shows the geometry of a
horizontal cut in the region of the entrance of a building, whereat the door is
assumed to be wooden. Symmetry conditions are applied at the left and at the
right edge of the structure and are realized by means of qx = 0.
78 4 Heat Conduction

A constant temperature of 220 C inside the building and of 60 C outside the

building is assumed as boundary conditions for the temperature field inside the
wall. Cauchy boundary conditions are not taken into account.
The different materials are indicated by different shadings. λ1 = 0. 08 W/mK is
valid for brickwork and λ2 = 0. 12 W/mK is valid for wood. The discretization
is performed by means of rectangular elements with bi–linear shape functions
for the temperature. The dimensions [ m ] and FE–discretization chosen for the
investigation are represented in Figure 4-8.

0.50 1.25

0.125
o λ 2 = 0.12 B 0.125
22
0.50
o
6
A λ 1 = 0.08 0.50
0.25
y

x 0.75 0.25 0.75

Fig. 4-8 2–dimensional heat conduction – initial configuration

The heat conduction process results in a stationary temperature field shown in

Figure 4-9. The temperature field is linearly distributed perpendicular to the
wall in regions, where an almost undisturbed heat conduction exists – cross
sections A and B. Near the outer corners the boundary conditions dominate
the temperature field inside the wall, what leads to thermal bridges.
The heat fluxes qx [ W/m2 ] and qy [ W/m2 ] are visualized in Figures 4-10 and
4-11. Due to the bi–linear approximation of the temperature field, the heat
fluxes are constant inside the elements in the direction of the heat flux, and
thus cause the gaps at the intersection of the elements. It is obvious, that the
heat flux is larger in the region of the larger heat conduction coefficient as well
as in the direction of the shorter transport distance. Disturbancies occur at
the inner and outer corners of the wall, where the heat flux is following the
direction of the lowest resistance. It has to be mentioned, that singularities
occur at the inner corners of the wall, which are indicated by high gradients of
the heat fluxes.
4.3 Example of Use 79

22. 0

o
C

6. 0

Fig. 4-9 2–dimensional heat conduction – temperature T (x, y)

6. 3

W/m2

0. 0

Fig. 4-10 2–dimensional heat conduction – heat flux qx

0. 0

W/m2

−15. 4

Fig. 4-11 2–dimensional heat conduction – heat flux qy

5 Membrane Structures

Membrane structures carry their loading along the main plane structure, so
the equilibrium conditions also may be addressed at this structural level. Thus,
instead of 3–dimensional coordinates and volume integrals, all governing equa-
tions as well as the work equations may be represented in the 2–dimensional
x–y–coordinate system, what may be equivalent to an integration with respect
to the third coordinate.

5.1 Rectangular Elements Regarding Plane Stress Situation

Regarding plane stress situations, the stresses rectangular to the main plane
structure are identical to zero, so the respective strains may be determined in
a subsequent analysis if required. Figure 5-1 shows the sign rule concerning the
displacements u(x, y) and v(x, y) as well as the stresses σxx , σyy and σxy with
respect to the x–y–coordinate system.

dx
x,u
px
y,v px σxx
dy
dy py
σxy
py σyx
dx
σyy
Fig. 5-1 Sign rule

5.1.1 Governing Equations and Work Equations

In the following, the governing equations – kinematics K, equilibrium conditions
E, material equations M – and for comparison the weak form of equilibrium
applying the PvD and moreover the work equations are presented wherein the
last ones implicitly comprise all governing equations.

External actions are the distributed loadings px , py [ N/m3 ]. Line–shaped acti-

ons as well as point–shaped loads are possible, too, but will not be considered
here. The variables employed by the governing equations are the displacement

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_5
82 5 Membrane Structures

fields u(x, y), v(x, y) [ m ], the stresses σxx (x, y), σyy (x, y), σxy (x, y), [ N/m2 ] as
well as Green’s strains εxx (x, y), εyy (x, y), εxy (x, y) [ 1 ].

a) Governing equations
K: εxx − u,x =0
εyy − v,y =0
εxy − 21 ( u,y + v,x ) =0
E: σxx ,x + σyx ,y + px =0
σyy ,y + σxy ,x + py =0
σxy − σyx =0
E
M: σxx − 1−ν 2
· (εxx + νεyy ) =0
E
σyy − 1−ν 2
· (εyy + νεxx ) =0
E
σxy − 1+ν
· εxy =0
b) Principle of virtual Displacements
Z
δW = (−δεxx σxx − δεyy σyy − 2 δεxy σxy + δu px + δv py ) dV = 0

c) Work equations: Instead of the integration with respect to the volume the
integration with respect to the z–coordinate usually is conducted a priori. As-
suming the thickness t to 1 m and employing the loadings px , py [ N/m2 ] this
leads to:
Z Z
E 1
−δWu = {δu,x (u, x + ν v,y ) + (1 − ν) δu,y (u, y + v,x )} dA − δu px dA ,
1 − ν2 2
Z Z
E 1
−δWv = {δv, y (v,y + ν u, x ) + (1 − ν) δv, x (u, y + v, x )} dA − δv py dA .
1 − ν2 2

5.1.2 Matrix Notation of the Work Equations

To get a clear arrangement of the element matrices the matrix notation develo-
ped regarding the bar has to be linked with respect to the relations formulated
in the governing equations. The derivation of the element stiffness matrix and
of the element load vector fits with the work equation given in the matrix
notation according to Equation (2.34):
Z Z
−δW = δvT · { ΩT · DT · E · D · Ω dA · v − ΩT · p dA } . (5.1)
A A
| {z } | {z }
element stiffness matrix element load vector
5.1 Rectangular Elements Regarding Plane Stress Situation 83

Or rather, in shortened representation, it follows

Z Z
−δW = δvT · { BT · E · B dA · v − ΩT · p dA } . (5.2)
A A

5.1.3 Elasticity Matrix E and Operator Matrix D

Hooke‘s model describes a linear–elastic material behavior by the elasticity
matrix E. Regarding a plane stress situation, the material equations follow to:

σ =E·ε ,
     
σxx 1 ν εxx
  E    
σ
 yy  = ν 1  ·  εyy  . (5.3)
1 − ν2

σxy 1 2εxy
2 (1 − ν)
Applying the governing equations from Section 5.1.1 kinematics may be repre-
sented by matrix notation, which is applied to virtual strains analogously:

ε = D·u ,
   
εxx ∂x " #
    u
 εyy = ∂y  · . (5.4)
v
2εxy ∂y ∂x

The displacement vector u includes the displacements u(x, y) and v(x, y), which
can be described by means of shape functions and nodal degrees of freedom.

5.1.4 Bi–linear Shape Functions Regarding Displacements

Usually it is non–trivial to choose the shape functions concerning the displace-
ments. The order of the approach follows from the convergence criteria, while
the nodal degrees of freedom and their positions within the element follow from
the element geometry, since all nodes must be handled equally, see Figure 5-2.

ly A lx

D B
x, u
y, v C

Fig. 5-2 Geometry and coordinate system regarding the rectangular element
84 5 Membrane Structures

The convergence criteria yield a symmetrically linear approach. These most

simple shape functions are chosen below. At the mathematically formal de-
rivation, the displacement fields are given as symmetric polynomials of the
first order with respect to the x–y–coordinates employing the so far unknown
coefficients ai and bi :
u(x, y) = a0 + a1 · x + a2 · y + a3 · xy ,
(5.5)
v(x, y) = b0 + b1 · x + b2 · y + b3 · xy .
The symmetry of shape functions may be checked with Pascale’s triangle as
shown in Figure 2-16. It should be noted that the general polynomials may
be transformed into the physically meaningful approach according to Section
2.4.3.

At this point, a product approach, derived from the linear shape functions well–
known from the bar and being physically meaningful, achieves the goal faster.
Thus, the displacements u in x–direction and v in y–direction are approximated
by the unknown nodal displacements at nodes A to D and the bi–linear shape
functions φA (x, y), . . . which are correlated to the unknowns. Therefore, the
physically meaningful approach regarding the displacement fields u(x, y) and
v(x, y) yields:
u(x, y) = φA · uA + φB · uB + φC · uC + φD · uD ,
(5.6)
v(x, y) = φA · vA + φB · vB + φC · vC + φD · vD .
If the nodal degrees of freedom of the respective element are arranged in the
vector v by nodes,
 
vT = uA vA uB vB uC vC uD vD , (5.7)

the shape functions need to be arranged in the matrix Ω in the same order as
the nodal degrees of freedom in both displacement fields u
" #
φA φB φC φD
Ω= . (5.8)
φA φB φC φD

Hence the approach regarding the displacements yields

u = Ω·v. (5.9)

Analogously to the description of the real displacements, the same approach is

also employed regarding the virtual displacements

δu = Ω · δv . (5.10)
5.1 Rectangular Elements Regarding Plane Stress Situation 85

The bi–linear shape functions of the displacement fields are depicted in Figure
5-3. All shape functions φA , . . . are normalized to 1 at the nodes, linear in the
x– and y–direction and quadratic along the diagonal. Table 5.1 presents the

1
A
1
B

1
D
1
C
Fig. 5-3 Bi–linear shape functions regarding displacements at rectangular elements

shape functions and their derivatives with respect to the coordinates.

Table 5.1 Bi–linear shape functions regarding rectangular elements

node shape function derivative derivative

i φi φi,x φi,y
x y y x
A (1 − lx
) · (1 − l ) − l1 · (1 − ly
) (1 − lx
) · (− l1 )
y x y

x y 1 y x
B lx
· (1 − ly
) lx
· (1 − ly
) lx
· (− l1 )
y

x y 1 y x 1
C lx
· ly lx
· ly lx
· ly
x y y x
D (1 − lx
)· l − l1 · ly
(1 − lx
) · l1
y x y

5.1.5 Differentiation of Displacement Fields

The work equation employs the auxiliary matrix B defined according to the
rule of differentiation D, see Equation (5.4),

B = D· Ω, (5.11)
86 5 Membrane Structures

that describes the strain field. Regarding the complete element, it follows
 
φA,x φB,x φC,x φD,x
 
B= φA,y φB,y φC,y φD,y  . (5.12)
φA,y φA,x φB,y φB,x φC,y φC,x φD,y φD,x

The differentiated shape functions are listed in Table 5.1.

5.1.6 Element Stiffness Matrix

Applying the matrices B and E, the element stiffness matrix can be computed
according to Equation (5.2)
Z ℓy Z ℓx
K= BT · E · B dx dy . (5.13)
0 0

The integrand employs a quadratic form that yields an 8 × 8-matrix. The mul-
tiplication scheme is represented in Figure 5-4. While the columns of the inte-

E B
* * 0
* * 0
0 0
*

BT

B TE B TE B

Fig. 5-4 Multiplication–scheme regarding the stiffness matrix

grand are linked to the real nodal degrees of freedom, the rows are linked to
the respective virtual nodal displacements. The symmetry of the integrand as
well as of the element stiffness matrix follows from the quadratic form of the
integrand and from the identical choice of shape functions regarding real and
virtual displacements.
5.1 Rectangular Elements Regarding Plane Stress Situation 87

For the sake of a clear view, the integrand is split into three parts according to
the terms occurring in the elasticity matrix
E νE E
BT · E · B = 2
· I1 + 2
· I2 + · I3 (5.14)
1−ν 1−ν 2(1 + ν)
with
φA,x φA,x φA,x φB,x φA,x φC,x φA,x φD,x
 

 φA,y φA,y φA,y φB,y φA,y φC,y φA,y φD,y

 φB,x φA,x φB,x φB,x φB,x φC,x φB,x φD,x 
 
φB,y φA,y φB,y φB,y φB,y φC,y φB,y φD,y
I1 = 


 φC,x φA,x φC,x φB,x φC,x φC,x φC,x φD,x 
 

 φC,y φA,y φC,y φB,y φC,y φC,y φC,y φD,y

 φD,x φA,x φD,x φB,x φD,x φC,x φD,x φD,x 
φD,y φA,y φD,y φB,y φD,y φC,y φD,y φD,y

φA,x φA,y φA,x φB,y φA,x φC,y φA,x φD,y

 
 φA,y φA,x φA,y φB,x φA,y φC,x φA,y φD,x 
 

 φB,x φA,y φB,x φB,y φB,x φC,y φB,x φD,y

 φB,y φA,x φB,y φB,x φB,y φC,x φB,y φD,x
I2 = 

 
 φC,x φA,y φC,x φB,y φC,x φC,y φC,x φD,y

 φC,y φA,x φC,y φB,x φC,y φC,x φC,y φD,x 
 
 φD,x φA,y φD,x φB,y φD,x φC,y φD,x φD,y
φD,y φA,x φD,y φB,x φD,y φC,x φD,y φD,x

φA,y φA,y φA,y φA,x φA,y φB,y φA,y φB,x φA,y φC,y φA,y φC,x φA,y φD,y φA,y φD,x
 

 φA,x φA,y φA,x φA,x φA,x φB,y φA,x φB,x φA,x φC,y φA,x φC,x φA,x φD,y φA,x φD,x


 φB,y φA,y φB,y φA,x φB,y φB,y φB,y φB,x φB,y φC,y φB,y φC,x φB,y φD,y φB,y φD,x

φB,x φA,y φB,x φA,x φB,x φB,y φB,x φB,x φB,x φC,y φB,x φC,x φB,x φD,y φB,x φD,x
I3 = 



 φC,y φA,y φC,y φA,x φC,y φB,y φC,y φB,x φC,y φC,y φC,y φC,x φC,y φD,y φC,y φD,x


 φC,x φA,y φC,x φA,x φC,x φB,y φC,x φB,x φC,x φC,y φC,x φC,x φC,x φD,y φC,x φD,x

 φD,y φA,y φD,y φA,x φD,y φB,y φD,y φB,x φD,y φC,y φD,y φC,x φD,y φD,y φD,y φD,x
φD,x φA,y φD,x φA,x φD,x φB,y φD,x φB,x φD,x φC,y φD,x φC,x φD,x φD,y φD,x φD,x

The differentiated shape functions are given in Table 5.1. Because of the bi–
linear shape functions regarding the displacements, quadratic functions at ma-
ximum appear in the integrand. According to Equations (5.13) and (5.14), the
element stiffness matrix follows from the sum of the three sub–matrices after
integrating:
K = K1 + K2 + K3 .
In the following the matrices K1 , K2 , K3 are given by numbers.
88 5 Membrane Structures
 
ℓ ℓ ℓ ℓy
2 ℓy −2 ℓy − ℓy ℓx
 x x x 
 
ℓx ℓx
− ℓℓx −2 ℓℓx
 
 2ℓ ℓy 
 y y y 
 
 ℓy ℓy ℓy ℓy 
 −2 ℓx 2ℓ
x ℓx
−ℓ
x

 
 
ℓx ℓx ℓx
− ℓℓx
 
 2ℓ −2 ℓ 
E  ℓy y y y 
K1 =  
6(1 − ν 2 )  ℓy ℓy ℓy ℓy 
 − ℓx ℓx
2ℓ
x
−2 ℓ
x

 
 
− ℓℓx −2 ℓℓx 2 ℓℓx ℓx
 
 y y y ℓy 
 
 
 ℓy ℓy ℓy ℓy 
 ℓ −ℓ −2 ℓ 2ℓ 
 x x x x 
 
ℓx ℓx ℓx
−2 ℓ −ℓ ℓ
2 ℓℓx
y y y y

 
1 1 −1 −1
 1 −1 −1 1 
 

 −1 −1 1 1 

νE  1 −1 −1 1 
K2 = 2
 
4(1 − ν ) 
 −1 −1 1 1 
 −1 1 1 −1 
 
 1 1 −1 −1 
−1 1 1 −1

 
4 ℓℓx 3 2 ℓℓx −3 −2 ℓℓx −3 −4 ℓℓx 3
y y y y
 
 ℓ ℓ ℓℓ 
 3
 4 ℓy 3 −4 ℓy −3 −2 ℓy
2 ℓy  −3
x x x 
x
 
 ℓx ℓx ℓx ℓx 
 2 3 4ℓ −3 −4 ℓ −3 −2 ℓ 3 
 ℓy y y y 
 
 ℓy ℓy ℓy ℓy 
 −3 −4 ℓ −3 4 ℓ 3 2ℓ 3 −2 ℓ 
E  x x x x 
K3 =  
24(1 + ν) 
 −2 ℓx −3 −4 ℓℓx 3 4 ℓℓx 3 2 ℓℓx −3 

 ℓy y y y 
 
 ℓy ℓy ℓy ℓy 
 −3 −2 ℓ −3 2 ℓ 3 4ℓ 3 −4 ℓ 
 x x x x 
 
 
 −4 ℓx −3 −2 ℓ ℓx
3 2ℓℓx
3 4ℓ ℓx
−3 
 ℓy y y y 
 
ℓ ℓ ℓ ℓ
3 2 ℓy 3 −2 ℓy −3 −4 ℓy −3 4 ℓy
x x x x
5.1 Rectangular Elements Regarding Plane Stress Situation 89

5.1.7 Element Load Vector

The computation of the integral
Z
f= ΩT · p dA
A

is possible, if the distribution of the external action is known. At first, only

a special case is discussed regarding an external action constantly distributed
inside the related element. With
" #
px
p= = const.
py
p may be extracted from the integral:
Z
f = ΩT dA · p . (5.15)
A

The matrix of shape functions

 
φA

 φA 

 φB 
 
 φB 
ΩT = 
 φC


 

 φC 

 φD 
φD
and the shape functions φi are given according to Section 5.1.4. The integration
provides the value ℓx ℓy /4 concerning all shape functions. Thereby the element
load vector follows
 
1
  
 1
fA 1  " #
 f  ℓ ℓ  
px
 B  x y   1
f = = · · .
4  1 py
 fC  

fD

 1
1 
1
An interpretation of the load vector is possible by interpreting the integral of
the constantly distributed external action px , or rather py , as four equally sized
nodal actions.
90 5 Membrane Structures

A variable external action can be taken into account if it is linearly approxima-

ted, similar to the description of the displacements. Thus, the external action
may be approached by employing the matrix of shape functions already known
as well as the nodal ordinates of the external action which yields

p = Ω · p̃ (5.16)

with Ω : matrix of shape functions

and p̃ : external action ordinates at the nodes.

A representation in detail gives:

px = φA · pxA + φB · pxB + φC · pxC + φD · pxD ,

py = φA · pyA + φB · pyB + φC · pyC + φD · pyD .

Thus it follows
Z Z Z
f = ΩT · p dA = ΩT · Ω · p̃ dA = ΩT · Ω dA · p̃ . (5.17)

The procedure of computing the integral is similar to the procedure concerning

the element stiffness matrix and yields a symmetric 8 × 8–matrix, see Equation
(5.18)
   
4 2 1 2 p
   xA 

 4 2 1 2   pyA 
   
 2 4 2 1   pxB 


   
  
ℓx · ℓy  2 4 2 1   pyB  
f= 
·
 . (5.18)
36   1 2 4 2   pxC 
   

 1 2 4 2   pyC 
 
  p 
 2 xD 
 1 2 4 
  
pyD
2 1 2 4

5.1.8 Subsequent Stress Analysis

After the calculation of the nodal displacements of the entire system, presen-
ted in Section 3, the deformation state of the structure is well–known and
the problem is all but solved. However, the state of stress must be known for
dimensioning of a structure.
At derivation of the work equation applying the PvD, the stress variables are
described by applying the material equations to the respective strains. In turn,
5.1 Rectangular Elements Regarding Plane Stress Situation 91

these strains are described by kinematics relating to the respective derivatives

of the displacements. Correspondingly, in the subsequent stress analysis, the
stresses may now be evaluated vice versa from the related nodal displacements.
As the nodal displacements are connected to the related shape functions in each
element, the displacements and thus also the stresses may not only be indicated
as nodal values, but also the courses of the function may be evaluated in the
whole element. For dimensioning, however, the courses of the stresses are not
essentially necessary in each element, as discrete values at exclusive positions
are sufficient. This is achieved by introducing the coordinates of the selected
positions into the respective course of the function. The matrix, which evaluates
the stress variables at the selected coordinates dependent on the respective
nodal displacements, is called stress matrix S̃, sometimes simply S.

Applying the matrix notation defined in Section 2.3.2, the computation of the
stresses σ regarding the respective element follows to

σ = E · ǫ,
σ = E · D· u, (5.19)
σ = E·D·Ω·v = E·B·v = S·v.

If the elasticity matrix E is constant, the courses of the stresses σ are already
determined with the course of the function of B.

The particular contents of the matrices required for the computation of stres-
ses are known from the derivation of the element matrices. By means of the
elasticity matrix E and the matrix of the differentiated shape functions B, the
stress matrix S can be determined. Employing the shape functions, cf. Section
5.1.4, the following applies:
 
φ νφA,y φB,x νφB,y φC,x νφC,y φD,x νφD,y
E  A,x
S =E·B= νφA,x φA,y νφB,x φB,y νφC,x φC,y νφD,x φD,y  .
1 − ν2
eφA,y eφA,x eφB,y eφB,x eφC,y eφC,x eφD,y eφD,x

As such, the abbreviation e = (1 − ν)/2 is used here. Table 5.1 comprises the
derivatives of the shape functions. It can be seen from the stress matrix S
that the course of the function inside the element is constant in the respective
direction of differentiation, and is linear perpendicular to it. Assuming that
Poisson’s ratio is not zero, additional linearly–constant terms have to be su-
perposed. Therefore, the course of the function is generally bi–linear, unlike as
at the bar.
92 5 Membrane Structures

The nodal values of the stresses σ̃ are obtained by inserting the coordinates of
the nodes into B. Applying the discrete matrix B̃, the stress matrix S̃ yields:

σ̃ = E · B̃ · v ,
σ̃ = S̃ · v ,
   
σ(x = 0 , y = 0 ) S(x = 0 , y = 0 )
 

 σ(x = ℓx , y = 0 )  
 S(x = ℓx , y = 0 ) 
 = ·v,
 σ(x = ℓx , y = ℓy )    S(x = ℓx , y = ℓy ) 

σ(x = 0 , y = ℓy ) S(x = 0 , y = ℓy )

−ℓ1 − ℓν ν
 1 
x y ℓx ℓy
 ν ν 
   −
 ℓx − ℓ1 ℓx
1
ℓy


y
σxx  e 
   − − ℓe e e 

 σyy    ℓy
 x ℓx ℓy 

   1   


σxy A
 
 −
 ℓx
1
ℓx
− ℓν ν
ℓy
 u

y  
 σxx   ν ν  v A

   −
 ℓx ℓx
− ℓ1 1
ℓy

   
 σyy    y
 u 
− ℓe − ℓe e e
    
 σxy  E  x y ℓx ℓy  v B
 B 
 = 1 − ν2 
 ·   .

 σxx   − ℓν 1 ν
− ℓ1

 u 
y ℓx ℓy x


 σyy
 
 



  v C
ν 1
   − ℓ1 ℓx ℓy
− ℓν 
u
 


 σxy C



 y x 
 

 σxx
 


 − ℓe e e
ℓy ℓx
− ℓe 

v D
y x
    

 σyy   
 − ℓν 1
− ℓ1 ν 

 y ℓx x ℓy 
σxy  ν 
D 
 − ℓ1 ℓx
− ℓν 1
ℓy


y x

−ℓe e
ℓx
e
ℓy
− ℓe
y x

5.1.9 Example of Use

The cantilever according to Figure 5-5 serves as an example of how to evaluate
the quality of the element. It is loaded by a constantly distributed external acti-
on in the y–direction at the free right–hand side boundary. Assuming Poisson’s
ratio is zero, the load–bearing behavior is equivalent to the behavior of a canti-
lever arm loaded at the free end. Local disturbances may occur at the position
of the load application, if it does not correspond to the shear stress distribution
regarding an undisturbed area. Due to kinematic constraints, disturbances in
5.1 Rectangular Elements Regarding Plane Stress Situation 93

shear stress behavior are also present at the clamping cross section.
x = l/2

x, u PY
l B
3
y, v

l
l = 15 m, t = 1 m, E = 100 000 N/m2, ν = 0, PY = 2 N/m2

Fig. 5-5 Cantilever – geometry and loading

In the following, cf. Figure 5-6, the stresses σxx and σxy are depicted regarding a
mesh employing 24 × 8 elements without taking into account the antisymmetry.
The jumps at the element intersections are clearly visible, following from the
subsequent stress analysis executed element by element.
σxx
180. 0

−180. 0
σxy
24. 0

−7. 8

Fig. 5-6 Cantilever – Distribution of stresses σxx and σxy

The approximation of the shear stresses that shows a change of sign at the
element intersections in the x–direction yet is particularly poor. An improve-
ment of these results employing the same number of elements as well as keeping
the order of shape functions unchanged may only be possible when applying a
special procedure e. g. according to Section 5.2.
94 5 Membrane Structures

The jumps in the distribution of stresses are depicted for the undisturbed area
in the center of the structure at x = 7. 50 m regarding meshes differently refi-
ned, see Figure 5-7, where only the upper half of the cantilever is represented.
The convergence of the normal stresses σxx is quite good, whereat the discreti-
zation error can be evaluated directly from the stress jumps. The shear stresses
converge very slowly and show particularly large jumps with changing signs in
some parts, so the results appear altogether questionable.

0.0 50.0 σxx 0.0 20.0 σxy

6 384 6 24 384 24 6
6

neutral axis

x = 7.5 ε
x = 7.5 + ε ε 0

Fig. 5-7 Cantilever – distribution of stresses at position x = 7. 50 m

In Figure 5-8 the convergence behavior of the vertical displacement at the

free boundary and of the stresses regarding position x = 7. 50 m is depicted
dependent on mesh refinement. The values of displacements and of stresses are
normalized with respect to the values regarding 384 elements, which is here
assumed to be the reference values. The discretization employs n = 3 elements
for the entire cantilever and n > 3 elements for the upper half of the cantilever.
The relatively good convergence of the displacements is characteristic applying
the FE–formulation employing displacements as primary variables. The reason
of this phenomenon is the bi–linear shape function for the displacements, which
describes the smooth deformation of the cantilever quite well, compare the
results of the bar in Section 1.3.
To represent the convergence behavior of the stresses, the averaged values of the
stresses are shown at the element intersections. Although the averaged values
are taken into account, the relatively poor convergence of the shear stresses
is obvious. The shape functions do not only approximate the parabolic shear
stress distribution in a poor way regarding the cross section of the structure,
but also result in oscillations inside an element.
5.1 Rectangular Elements Regarding Plane Stress Situation 95

σxx
σe x = 7,5 m
1,0

0,5

3 6 24 96 384 no. of elements

σxy
σe x = 7,5 m
1,0

0,5

3 6 24 96 384 no. of elements

v
ve
x = 15,0 m
1,0

0,5

3 6 24 96 384 no. of elements

Fig. 5-8 Cantilever – convergence behavior of stresses and displacements

Overall, the distribution of stresses regarding the cantilever may only be poorly
approached when employing a rectangular plane stress element with bi–linear
shape functions, even if a relatively fine discretization is applied. Reason for
this is the poor approximation of the shear stiffness inside the element, which
is clearly visible in this example. An improvement of the element behavior is
explained in Section 5.2.
96 5 Membrane Structures

5.2 Rectangular Element Comprising Modified Shear Strains

The results of the rectangular element employing a bi–linear approach are not
that sufficient, although the displacements of the cantilever slab may be appro-
ximated quite well, see Figure 5-5. The reason for the relatively bad courses
of the stresses is the poor approximation of the shear stiffness in the element.
This leads to the large oscillations in the course of the shear stresses beyond
the edges of the elements, which can be seen in the example.
If non–physical oscillations occur in the distribution of stresses or of displace-
ments, this indicates numerical effects which may be motivated by the choice of
shape functions. Here, the shape functions are already determined by the con-
vergence criteria, so an improvement of the results is possible by increasing the
order of the shape functions, cf. the results in Section 9.4 employing quadratic
shape functions.
When employing linear shape functions, an improvement is possible only if the
characteristics of the stiffness matrix are analysed in more detail. Compressive
as well as tensile stresses are well described by the element already, if the lower
order of approximation compared to that regarding the displacements is igno-
red. In contrast, the shear stresses are very poorly approximated - depending
on the displacement field.
In Figure 5-9 the case of pure shear deformation is shown, whereby the shear
strains and thus the shear stresses are constant inside the element and therefore
may be represented exactly by adequately employing linear shape functions.

εyx
A B

εxy
x uA = uB = −uC = −uD ,
εxy
y
vA = −vB = −vC = vD .
D C
εyx

Fig. 5-9 Pure shear deformation

Due to the definition of shear strain,

2 εxy = u,y + v,x ,

multiplicative shape functions are sufficient for u(x, y), if they are constant in
the x–direction, and for v(x, y), if they are constant in the y–direction, in order
to correctly describe the shear strains inside the element.
5.2 Rectangular Element Comprising Modified Shear Strains 97

The second case, as given in Figure 5-10, describes a pure bending deformation
of the element whereby the strains in the x– as well as in the y–direction are
exactly represented, but additional shear deformations γAD and γBC occur,
showing changing signs and being non-physical. They arise because of the linear
shape functions that are employed, which cannot describe the element curvature
at pure bending. The drawback results in activation of shear stiffness in the case
of pure bending and therefore the respective bending deformation converges
more slowly against the exact solution.

pure bending bending described by linear functions

A B A B

x x

y εxy y εxy

D C D C
∆u ∆u

uA = −uB = uC = −uD and vA = vB = vC = vD = 0 .

Fig. 5-10 Deficient description of bending deformation

5.2.1 Selectively Reduced Integration

An elimination of this defect is possible, if the work term
Z
E
2 · δεxy · · 2 · εxy dA
2(1 + ν)

is integrated comprising a constant δεxy as well as εxy according to Figure 5-9,

and, referring to this, the stiffness matrix is modified. In this way, pure shear
deformation can be described but the non–physical shear deformation at pure
bending is not, see Figure 5-10.
This procedure is called selectively reduced integration, since only the constant
terms of the work integral are evaluated and all higher order ones are ignored,
what is equivalent to a constant shear strain field. The results obtained from
this modified element are surprisingly good, and do not show any oscillations
in the shear stress distribution.
This description of the shear state corresponds to Figure 5-9. In the following,
the resulting stiffness matrix K3 is given. When compared to K3 as derived
in Section 5.1.6, the procedure presented here can be interpreted as averaging.
98 5 Membrane Structures

The entries are constant except for the quotients of lengths and the sign.
 ℓx ℓx 
ℓy
1 ℓy
−1 − ℓℓx −1 − ℓℓx 1
y y
 
 ℓy ℓy ℓy ℓy 
 1 1 − ℓ −1 − ℓ −1 ℓ 
 ℓx x x x 
 
 ℓx ℓ x ℓ x ℓ x


 ℓy 1 ℓy
−1 − ℓy
−1 − ℓy
1 

 
 ℓy 
 −1 − ℓy −1 ℓy 1
ℓy
1 − 
3·E  ℓx ℓx ℓx ℓx 
K3 =  .
24(1 + ν) 
 − ℓx −1 − ℓx 1 ℓx
1 ℓx
−1 

 ℓy ℓy ℓy ℓy 
 
 ℓ y ℓ y ℓ y ℓ y

 −1 − −1 ℓ 1 1 −ℓ 
 ℓx x ℓx x 
 
 ℓx ℓx ℓx ℓx 
− −1 − 1 1 −1 
 ℓy ℓy ℓy ℓy 
 
ℓy ℓy ℓy ℓy
1 ℓ
1 − ℓ
−1 − ℓ
−1 ℓ
x x x x

The stress matrix applied at subsequent analysis is given below regarding con-
stant shear stresses. This corresponds to an averaging of the shear stresses
according to Section 5.1.8. As such, e = (1 − ν)/2/2 is applied.
 1 ν 
−ℓ − ℓν 1
ℓx ℓy
x y
 ν ν 1 
   − − 1 
 ℓx ℓy ℓx ℓy 
σxx  e 
  − − e −e e e e e
−ℓe 
 σyy    ℓ y ℓ x ℓ y ℓ x ℓ y ℓ x ℓ y x 
     
   1 1 ν ν 
 σxy A  − −  u
    ℓx ℓx ℓy ℓy   
 σxx   ν ν 1 1   v A 
  − −    
 σ    ℓx ℓx ℓy ℓy   
 yy    e e e e e e e e   u 
  − − − −   
 σxy  E  ℓy ℓx ℓy ℓx ℓy ℓx ℓy ℓx   v B 
 B =   ·   .
 σ  1 − ν 2 ν 1 ν 1   u 


xx 


 − ℓy ℓx ℓy
− ℓx
 
  v



 σyy      
1 ν 1 ν C

 σ



 − ℓy ℓx ℓy
− ℓx
 
  u  

 xy C   e   
  − − e −e e e e e e 
−ℓ  v D
 σxx   ℓy ℓx ℓy ℓx ℓy ℓx ℓy x
   
 σyy    ν 1 1 ν 
  
 − ℓy ℓx
− ℓx ℓy 

σxy D  ν 

 − ℓ1 ℓ
− ℓν ℓ1  
y x x y

−ℓe − ℓe −ℓe e
ℓx
e e
ℓy ℓx
e
ℓy
−ℓe
y x y x
5.2 Rectangular Element Comprising Modified Shear Strains 99

5.2.2 Example of Use

Once again the cantilever, as given in Figure 5-5, serves as a test to investigate
the quality of the element. The increase in quality becomes particularly clear
when compared to the results employing the unmodified shape functions. Even
in case of 3 elements the error of the tip displacement achieves only 4 %.
In the following Figures 5-11, the stresses σxx and σxy are depicted applying
a mesh with 24 × 8 elements, without taking into account the antiysmmetry.
Due to the constant shear strains inside the element, the shear stresses are also
constant inside each element. As a result, the real stress distribution can now
be approximated much better, whereat the averaged stresses at the element in-
tersections are very close to the exact values. Local disturbances at the position
of the load application as well as at the clamping cross section are preserved
because of the reasons mentioned above.
σxx
180. 0

−180. 0
σxy
15. 0

2. 5

Fig. 5-11 Cantilever – distributions of stresses σxx and σxy ,

employing selectively reduced integration

Remarks
Even if the selectively reduced integration allows for very good results here,
the results may be poor at other applications, cf. Section 24.2 and 24.4. Fur-
thermore, if the complete stiffness matrix would be integrated by a reduced
integration scheme, displacement fields, not comprising any strain energy, may
occur, which are named zero–enery modes. These displacement fields may be
recognized by an oscillation of the displacements in terms of an hourglass and
therefore are called hourglass modes.
100 5 Membrane Structures

5.3 Plane Strain

The load–carrying behavior of a 3–dimensional structure, where strains does
not occur with respect to the direction of thickness, is called plane strain. Com-
pared to plane stress, it is a characteristic of plane strain that stresses σzz exist
due to restraints in the direction of thickness. Plane strain is applicable to
many problems in geotechnics. Here many 3–dimensional situations as tunnel
constructions, pit excavations or continuous footing may be investigated by em-
ploying 2–dimensional modeling, if the load–carrying behavior can be assumed
to be invariant in the third direction.
The governing equations regarding plane strain can be derived from those of
plane stress by substituting E/(1 − ν 2 ) in place of E and ν/(1 − ν) in place of
ν with respect to all previous derivations at Section 5.1.
However, it is formally more elegant, and also more advantageous concerning
the description of nonlinear material behavior, to modify the elasticity matrix
E, the operator matrix D as well as the vectors of the stresses σ and the strains
ǫ. The stress σzz and the strain εzz are to be considered additionally regarding
the direction of thickness, whereby εzz is set here to zero explicitly. Thus, the
material equations and kinematics can be applied as given below:

σ = E·ε,
     
σxx 1−ν ν ν εxx

 σyy 
 E 
 ν 1−ν ν  
  εyy 

 =  ·  , (5.20)
 σzz  (1 + ν)(1 − 2ν)  ν ν 1−ν   εzz 
σxy (1−2ν)/2 2εxy

ǫ = D· u,
   
εxx ∂x
 

 εyy  
  ∂y 
 u
 = · . (5.21)
 εzz    v
2εxy ∂y ∂x

Except for the fact that the matrix B is comprised of an additional row employ-
ing zero entries, there are no significant differences in the further derivations,
therefore these are not shown here.
6 Bending Structures

Thin walled structures, which are loaded perpendicular to their main spatial
dimensions, carry their loading by bending. Modeling of bending structures
usually requires kinematic assumptions, which yield either the Euler–Bernoulli
beam theory at investigating 1–dimensional structures or the Kirchhoff theory
of plates for 2–dimensional structures. Investigating the bending behavior of
thin bending structures, shear deformations may be neglected, what requires
special finite elements, dealing with C1 –conformity at the element intersections.

6.1 Element Matrices Regarding Euler–Bernoulli Beams

The kinematic assumption that governs the Euler–Bernoulli beam theory is the
Bernoulli hypothesis neglecting the shear deformations by specifying εxz = 0.
Thus the shear force does not perform any work and the deflection w being the
only independent displacement variable. Employing the beam element descri-
bed below, the definition of signs applies as it is represented in Figure 6-1.
p
z

Me
x e
V
ϕ p
z

z,w
M - dM M + dM
2 x 2
dQ dQ
Q- Q+
2 2

Fig. 6-1 Euler–Bernoulli beam – sign definition

6.1.1 Governing Equations and Work Equations

a) Governing equations
K: ϕ,x − κ =0
or : w,xx − κ = 0
w,x − ϕ =0

E: Q,x + pz =0
or : M,xx + pz = 0
M,x − Q =0

M : M − EI · (−κ) = 0

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_6
102 6 Bending Structures

b) Principle of virtual Displacements

Z
δWd = {δκ M + δw pz } dx = 0

c) Work equation
Z Z
− δWd = δw,xx · EI · w,xx dx − δw · pz dx = 0 .

6.1.2 Matrix Notation of the Work Equation

In order to derive the element stiffness matrix and the load vector regarding
Euler–Bernoulli beams, matrix notation is applied according to Equation (2.34)
Z ℓ Z ℓ
T T T
−δWd = δv { Ω · D · E · D · Ω dx · v − ΩT · p dx } , (6.1)
0 0

or in a short form
Z ℓ Z ℓ
T T
− δWd = δv { B · E · B dx · v − ΩT · p dx } . (6.2)
0 0

6.1.3 Elasticity Matrix E and Operator Matrix D

Concerning Euler–Bernoulli beams, only the bending moment M does perform
work as a stress variable σ. The negative curvature −κ is the corresponding
strain variable here, since the positive moment and the positive curvature are
defined opposed to each other. The bending stiffness EI is the only element of
the elasticity matrix E here,

M = EI · (−κ) ,
σ = E · ǫ. (6.3)

The differentiation rule, necessary to compute the curvature κ from the de-
flection w, is given by the operator matrix D. Applying matrix notation, the
vectors of strains ε and of displacements u, which only comprise the deflection
w(x), yield

−κ = −w,xx = −d2x w → D = [−d2x ] ,

ǫ = D ·u. (6.4)
6.1 Element Matrices Regarding Euler–Bernoulli Beams 103

6.1.4 Shape Functions to Describe the Deflection

According to the convergence criteria, the deflection w(x) is to be described by
a polynomial approach of the 3rd order at minimum, thus the mathematical
formulation yields

w(x) = a0 + a1 · x + a2 · x2 + a3 · x3 . (6.5)

To get a physically meaningful approach, it makes sense not only to use the
displacements wi , but also the rotations ϕ|i = w,x |i as degrees of freedom at
the element nodes, since, regarding convergence criteria, the requirement of
C1 –conformity may be easily fulfilled hereby

w(x) = φ1 · wA + φ2 · ϕA + φ3 · wB + φ4 · ϕB . (6.6)

The related shape functions φi , represented in Figure 6-2, are the Hermite–
Polynomials of the 3rd order. The courses of the functions are given as well
as the respective second derivatives. It is to be emphasized that the approach
is chosen to describe the deflection w(x), therefore the rotations are nodal
displacements variables only.

A B
i φi (x) φi (x),xx x
l

x2 x3
1 1−3 ℓ2
+2 ℓ3
− 6 ℓ12 + 12 ℓx3 1 Φ1 (x)

2 x3
2 x − 2 xℓ + ℓ2
− 4 1ℓ + 6 ℓx2 1 Φ2 (x)

2 3
3 3 xℓ2 − 2 xℓ3 6 ℓ12 − 12 ℓx3 1 Φ3 (x)

2 x3
4 − xℓ + ℓ2
− 2 1ℓ + 6 ℓx2 1 Φ4 (x)

Fig. 6-2 Cubic Hermite–Polynomials

Applying the nodal displacements

vT = [ wA ϕA wB ϕB ]
104 6 Bending Structures

and the shape functions merged into the matrix of shape functions
Ω = [ φ1 φ2 φ3 φ4 ]
the approach to describe the deflection follows to
u=Ω·v (6.7)
and in analogy, regarding the virtual deflection, to
δu = Ω · δv . (6.8)

6.1.5 Differentiation of Shape Functions

The work equation requires the shape functions Ω being differentiated accor-
ding to the differentiation rule D, yielding the matrix B of curvatures:
B = D · Ω = [ −φ1,xx − φ2,xx − φ3,xx − φ4,xx ] . (6.9)
The entries φi,xx given in B are listed in Figure 6-2.

6.1.6 Element Stiffness Matrix

The matrices E and B, required for the computation of the stiffness matrix
Z ℓ
K= BT · E · B dx , (6.10)
0

are already known. The matrix product, being part of the integrand, yields
 
φ1,xx φ1,xx φ1,xx φ2,xx φ1,xx φ3,xx φ1,xx φ4,xx
 
 φ2,xx φ1,xx φ2,xx φ2,xx φ2,xx φ3,xx φ2,xx φ4,xx 
BT · E · B = EI · 

.

 φ3,xx φ1,xx φ3,xx φ2,xx φ3,xx φ3,xx φ3,xx φ4,xx 
φ4,xx φ1,xx φ4,xx φ2,xx φ4,xx φ3,xx φ4,xx φ4,xx
The second derivatives of the shape functions are linear according to Figure
6-2. The products may be analytically integrated, what results in:
 
12 6
ℓ 3 ℓ 2 − 12
ℓ 3
6
ℓ 2
 
 6 4 6 2 


 ℓ2 ℓ
− ℓ2 ℓ 
K = EI ·  12
. (6.11)
 − 3 6 12 6 
 ℓ − ℓ 2 ℓ 3 − 2
ℓ 

 
6 2 6 4
ℓ2 ℓ
− ℓ2 ℓ
6.1 Element Matrices Regarding Euler–Bernoulli Beams 105

Compared to the displacement method, the matrix K corresponds to the exact

stiffness matrix of Euler–Bernoulli beams clamped at both ends. Therefore,
the entries of the first and of the third row can be interpreted as shear forces
at the ends of the beam, which perform work on the respective virtual nodal
deflections, and the entries of the second and of the fourth row as bending
moments, which perform work on the respective virtual nodal rotations.

If the nodal displacement variables are ordered by nodes, the element stiffness
matrix may be clearly arranged applying sub–matrices, which explicitly point
out the allocation of the stiffness entries to the element nodes
" #
kAA kAB
K= .
kBA kBB

6.1.7 Element Load Vector

Regarding constantly distributed external actions p = [ p ] = constant in the
element, p can be extracted from the integral, so that the element load vector
is determined by the integral of the shape functions
Z ℓ Z ℓ
T
f= Ω · p dx = ΩT dx · p . (6.12)
0 0

The integration of the matrix Ω of shape functions gives

 ℓ 
2
ℓ2
" #  
fA  
f=

= 12 
ℓ  · p. (6.13)
fB  
 2 
ℓ2
− 12

The load vector corresponds to the supporting forces and moments of the beam
clamped at both ends and loaded by a constantly distributed external action.
A varying action p(x) may be approximated by a linear course of the function
inside the element. Applying the matrix Ωp
h x x i
Ωp = 1 − = [ φA φB ] ,
ℓ ℓ
employing linear shape functions according to Section 2.3.1 and the related
nodal values p̃ = [ pA pB ], it applies

p = Ωp · p̃ = φA · pA + φB · pB . (6.14)
106 6 Bending Structures

The integral to be computed to get the load vector

Z ℓ
f= ΩT · Ωp dx · p̃
0

comprises the shape functions ΩT , employing polynomials of the third order

to describe the virtual displacements, as well as the linear shape functions Ωp
that approximates the course of external action.

6.1.8 Subsequent Stress Analysis

To evaluate the stress resultants, the formal procedure of the stress analysis,
already known from the bar as well as from the membrane element, is applicable
σ = E · ǫ,
σ = E · D· u, (6.15)
σ = E·D·Ω·v = E·B·v = S·v.
If the elasticity matrix E is constant, the course of stesses σ follows the course
of curvatures given by B. The nodal values σ̃ of the stresses are obtained by
introducing the coordinates of the nodes into B. Then, the stress matrix S̃ can
be evaluated from the matrix of discrete curvatures B̃, which yields
σ̃ = E · B̃ · v ,
σ̃ = S̃ · v .
Regarding Euler–Bernoulli beams, the material model is defined over the ben-
ding moments. The stress matrix follows with
E = [ EI ] and B = [ −φ1,xx − φ2,xx − φ3,xx − φ4,xx ] ,

comprising a linear course of the bending moment inside the element

h i
σ = EI · ℓ62 − 12x ℓ3
4
ℓ
− 6x
ℓ2
− 6
ℓ2
+ 12x
ℓ3
2
ℓ
− 6x
ℓ2
·v.

Evaluation at the element nodes gives the nodal moments MA and MB

" # " # " 6 4 2 #
σ(x = 0) MA ℓ2 ℓ
− ℓ62 ℓ
σ̃ = = = EI · ·v.
σ(x = ℓ) MB − ℓ62 − 2ℓ 6
ℓ2
− 4
ℓ
The shear forces are evaluated applying the equilibrium condition
h i
Q = M,x = dx M = dx σ = EI · − 12 ℓ 3 − 6
ℓ 2
12
ℓ 3 − 6
ℓ 2 ·v.
6.1 Element Matrices Regarding Euler–Bernoulli Beams 107

Thus the shear forces are constantly approximated inside an element, but allo-
cated both element nodes. The stress matrix regarding the shear forces as well
as the bending moments can be summarised and ordered by nodes
   
QA − 12
ℓ 3 − ℓ62 12
ℓ 3 − ℓ62
   
   6 4 6 2 
 MA   ℓ2 ℓ
− ℓ 2 ℓ 
  = EI ·  ·v. (6.16)
   12 6 12 6 
 QB   − ℓ3 − ℓ2 ℓ 3 − ℓ 2 
   
MB − ℓ62 − 2ℓ ℓ
6
2 − 4
ℓ

In this formulation, the stress matrix is identical to the element stiffness matrix
according to Equation (6.11), except of the signs.

6.1.9 Example of Use

The single span beam with a length of l = 8 m and a bending stiffness of EI =
10,000 kN m2 is analysed with regard to the loading and boundary conditions
as given in Figure 6-3.

1 2 3 4
1 2 3 4 5
x
EI = 10,000 kNm2
z
l = 8.0 m
2.0 2.0 2.0 2.0

lc 1 p = 10 kN/m

lc 2 P = 16 kN

lc 3 w1 = 0.04 m

Fig. 6-3 Single span beam – geometry and loading

The discretization is chosen employing four elements and five nodes. The results
concerning the loading conditions two and three represented in Figure 6-4 are
identical to the exact analytical solution, since the approach for the deflection
is the solution of the homogeneous differential equation, and the particular
solution part disappears due to the external actions selected here. The results
108 6 Bending Structures

concerning loading condition one are characteristic for the FEM, being only an
approximate solution.

Regarding the selected discretization, the quality of the approximation can be

identified from the jumps in the courses of the M – and Q–courses and from the
fulfillment of the boundary condition for the bending moment. The jumps in
the M –course disappear at the element intersections, since the error resulting
from the particular solution due to a constantly distributed loading is equal
at the respective left and right hand side of a node. The error present at the
left hand side support ∆M = 3. 33 kN m cannot be clearly seen due to the
measuring unit used. The shear force is determined from the M –course and
therefore is also an approximation.

1 2 3 4 5

1 2 3 4

w1
20 mm
w2
w3

−76.67 = M5
lc 1: M −40.00 = Q 4
Q

−24.00 = M5
lc 2: M −11.00 = Q 4
Q

−18.75 = M5
lc 3: M −2.34 = Q 4
Q

Fig. 6-4 Single span beam – state variables regarding loading conditions 1 to 3
6.2 Kirchhoff Plate Element with 16 DOF 109

6.2 Kirchhoff Plate Element with 16 DOF

The Kirchhoff theory of plates [55] is basis for the investigation of the deforma-
tion behavior of thin plates. Applying the Kirchhoff theory of plates, the defor-
mations caused by shear forces are assumed to be negligible, comparable to the
Euler–Bernoulli beam theory. This can be realized by the kinematic constraints,
that straight lines perpendicular to the mid-surface remain perpendicular to
the mid-surface and straight after the deformation, and the thickness does not
change during the deformation. As a consequence of the Kirchhoff hypothesis,
effective shear forces must be defined and prescribed on the free boundaries of
the plate. However, this is not essential regarding the displacement–based for-
mulation investigated here, since the Neumann boundary conditions are only
fulfilled weakly by the PvD. Employing the rectangular plate element described
below, the definition of signs applies as it is represented in Figure 6-5.
p
x
y p
z,w dx
dy
dy
dx
units :
m xy
p [ N/m2 ] m yy m xx
w [m]
q [ N/m ] qx
m yx
m [ Nm/m ] qy

Fig. 6-5 Stress resultants of plates – definition of signs

In the following subsections, as a first approach, the rectangular element deve-

loped by Bogner, Fox, Schmit [20] will be described in detail.

6.2.1 Governing Equations and Work Equations

a) Governing Equations
K : w,x − ϕx =0
w,y − ϕy =0
ϕx ,x − κxx =0
ϕy ,y − κyy =0
(ϕx ,y + ϕy ,x ) − 2κxy =0
or : w,xx − κxx = 0
w,yy − κyy = 0
w,xy − κxy = 0
110 6 Bending Structures

E: qx ,x + qy ,y + p =0
mxx ,x + mxy ,y − qx =0
myy ,y + mxy ,x − qy =0
or : mxx ,xx + 2mxy ,xy + myy ,yy + p = 0
M : mxx − B (−κxx − νκyy ) = 0
myy − B (−κyy − νκxx ) = 0
mxy − B (1 − ν) (−κxy ) = 0
E t3
with plate bending stiffness B=
12 (1 − ν 2 )
b) Principle of virtual Displacements
Z
− δWd = { δκxx mxx + 2δκxy mxy + δκyy myy + δw p } dA = 0

c) Work equation
Z
− δWd = B { δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
Z
+ δw,xy 2(1 − ν) w,xy } dA − δw p dA .

6.2.2 Matrix Notation of the Work Equation

To derive the element stiffness matrix as well as the load vector regarding
Kirchhoff ’s plate theory, once again the general work equation is applied in the
matrix notation
Z Z
− δW = δvT · { ΩT · DT · E · D · Ω dA · v − ΩT · p dA } (6.17)
A A
or in a short form
Z Z
− δW = δvT · { BT · E · B dA · v − ΩT · p dA } . (6.18)
A A

The matrix symbols are filled with physical content, if the governing equations
are interpreted in a corresponding manner.

6.2.3 Elasticity Matrix E and Operator Matrix D

According to Kirchhoff ’s plate theory, only bending and twisting moments, em-
ployed as stress variables σ, perform work on the respective virtual curvatures
δκ, employed as generalized strain variables δε.
6.2 Kirchhoff Plate Element with 16 DOF 111

The material equations link stresses σ and strains ǫ to each other:

σ = E ·ε,
     
mxx 1 ν −κxx
   ν 1   
 myy  = B ·   ·  −κyy  . (6.19)
1
mxy 2
(1 − ν) −2κxy

The differentiation rule D to compute the strains ǫ from the deflection is defined
by the kinematic conditions

ε = D · u,
   
−κxx −∂xx
   
 −κyy  =  −∂yy  · [ w ] . (6.20)
−2κxy −2∂xy

The only independent state variable is the deflection surface w(x, y), which is
approximated by shape functions and nodal displacement variables

w(x, y) = Ω(x, y) · v . (6.21)

6.2.4 Shape Functions to Describe the Deflection

The choice of the shape functions to describe the deformation behavior of pla-
tes is more difficult than regarding membrane structures because the conver-
gence criteria require a C1 –conformity analogous to that of beams. This means
that the deflection must be continuous and continuously differentiable parallel
and rectangular to common element intersections. In analogy to the beam this
requires at least cubic polynomials with respect to the x– and y–coordinate
employing deflections and derivatives of the deflection surface as nodal dis-
placement variables at the element boundaries.

The shape functions assigned to the nodal displacement variables can be deve-
loped as a product approach employing Hermite–Polynomials that have been
already introduced regarding bending at Euler–Bernoulli beams. The combina-
tion of the respective four shape functions of the beam concerning the x– and
y–directions as a product approach yields 16 shape functions in total to describe
the deflection. If the nodal displacement variables vnode = [ w w,x w,y w,xy ]node
are chosen as degrees of freedom in the four corners of the rectangular element,
112 6 Bending Structures

the element intersection conditions can be fulfilled. Overall, the symmetric ap-
proach employs 16 degrees of freedom, ordered by the nodes here
vT = { [ w w,x w,y w,xy ]A
[ w w,x w,y w,xy ]B
[ w w,x w,y w,xy ]C
[ w w,x w,y w,xy ]D } . (6.22)

The shape functions assigned to the displacement variables of a single node are
depicted in Figure 6-6.

w w,x

w,y w,xy

Fig. 6-6 Shape functions regarding displacement variables related to node D

The shape functions related to the displacement variables of the other nodes can
be obtained analogously, if those are combined with each other appropriately.
The order of the nodal displacement variables also defines the order of the
shape functions in the vector
Ω ={[ φ1 (x) · φ1 (y) φ2 (x) · φ1 (y) φ1 (x) · φ2 (y) φ2 (x) · φ2 (y) ]A
[ φ3 (x) · φ1 (y) φ4 (x) · φ1 (y) φ3 (x) · φ2 (y) φ4 (x) · φ2 (y) ]B
[ φ3 (x) · φ3 (y) φ4 (x) · φ3 (y) φ3 (x) · φ4 (y) φ4 (x) · φ4 (y) ]C
[ φ1 (x) · φ3 (y) φ2 (x) · φ3 (y) φ1 (x) · φ4 (y) φ2 (x) · φ4 (y) ]D }
or written in a short form related to the products of the shape functions
Ω = [ φA1 φA2 φA3 φA4
φB1 φB2 φB3 φB4
φC1 φC2 φC3 φC4
φD1 φD2 φD3 φD4 ]. (6.23)
6.2 Kirchhoff Plate Element with 16 DOF 113

Thereby the deflection w(x, y) follows in a short form to

u = [ w(x, y) ] = Ω(x, y) · v . (6.24)

If the more formal way is chosen, that is also correct and applies the scaling
matrix, at first the symmetric general polynomial approach is chosen:

w(x, y) = a1 + a2 x + a3 y
+ a4 x2 + a5 xy + a6 y 2
+ a7 x3 + a8 x2 y + a9 xy 2 + a10 y 3
+ a11 x3 y + a12 x2 y 2 + a13 xy 3
+ a14 x3 y 2 + a15 x2 y 3 + a16 x3 y 3 . (6.25)

The symmetry of this approach can be proved by Pascale’s triangle, cf. Figure
2-16. The transformation of the general polynomial into an approach employing
physically meaningful degrees of freedom is possible according to Section 2.4.3,
when the scaling is related to the nodal displacement variables according to
Equation (6.22). With the general polynomial

[ w(x, y) ] = ψ · a

the physically meaningful degrees of freedom of Equation (6.22) can be repre-

sented by

v = Ψ̃ · a ,

whereat the 16 × 16–matrix ψ̃ may be computed by introducing the coordi-

nates of the related nodes either into the general polynomial itself or into the
respective derivatives. Now, the scaling matrix G is obtained from the solution
of the system of equations with respect to a
−1
a = Ψ̃ ·v = G·v.

Hereby, the general polynomial ψ can be transformed into the approach Ω

employing physically meaningful nodal displacement variables

Ω = ψ · G.

The scaling matrix G is given below.

 114

1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 

0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
 

2 3 1
 

0 0 2 0 0 0 0 0 0 0 0 0 0 w
 

a1 − − lx lx
− lx
 


0 0 1 0 0 0 0 0 0 0 0 0 0 0

   3    

3
 

0 0 0 0 0 0 0 0 0
 lx2

y y
 

ly2
    w,x 
0 − l2 0 − l1
0 

1
 a2   0   

0 0 0 0 0 0 0 0 0 0 w
w,y  

2
     

lx x x
0 − l23 l12
 a3   3
0 

3
 a4   − ly2 

0 0 0 0 0 0 0 0 0


2
   2  w,xy A 

x x lx x
 

0 − l32 − l2 − l1
 a5   0 

3 2 3
   lx3  

0 0 0 0 0 0 0 0 0
y y


ly2
 a6    

0 −l
   0 0   w,x   

1 2 1
w
   w,y  

0 0 0 0 0 0 0 0 0 0 0 0
 

2 3 2


y ly ly ly
 a7   1 

−


2 1
 a8   0 − ly2 0 −l 


0 0 0 0 0 0 0 0 0
 = 2   w,xy B 

3 2


lx lx x x
   

0 − l23 l12
 a9   l 3  
 

6 6 4 9 3 6 2 9 3 3 1 9 6 3
 a10     w,x 
   0 0    

x y x y x y x y x y x y x y x y x y x y x y


w
 a11     w,y  
   

2 1 2
  
a12   9

0 0 0 0 0 0 0 0 0 0 0
2 


ly3 ly2 ly3

w,xy C 

−
  l2 l2 l l2 l2 l l l − l2 l2 l l2 − l2 l l l l2 l2 − l l2 − l2 l l l − l2 l2 − l l2 l2 l l l   
 a13   xy xy xy xy x y 

a15 w,y


3 4 2 6 3 4 2 6 3 2 1 6 3 2 1
    
   1  

a16


w,xy D
 a14   0
ly2 

− − − − − −
w,x  
    
  
 6  
− 3 2 − 2 2 − 3 − 2 

x y x y x y x y x y x y x y x y x y x y
 lx ly lx ly lx ly lx ly lx3 ly2 lx2 ly2 lx3 ly lx2 ly lx3 ly2 lx2 ly2 lx3 ly lx2 ly lx3 ly2 lx2 ly2 lx3 ly lx2 ly 
 

4 2 2 1
 6 

l3 l3 l2 l3 l3 l2 l2 l2
− l2 l3 − l 4l3 − l23l2 − l 2l2 l26l3 − l 2l3 l23l2 − l 1l2 − l26l3 l 2l3 l23l2 − l 1l2 l26l3 l 4l3 − l23l2 − l 2l2 

x y x y x y x y x y x y x y x y x y x y x y x y x y x y x y x y
− l34l3 l22l3 − l32l2 l21l2 l34l3 − l22l3 − l32l2 l21l2 − l34l3 − l22l3 l32l2 l21l2
 xy xy xy xy xy x y
 
6 Bending Structures
6.2 Kirchhoff Plate Element with 16 DOF 115

6.2.5 Differentiation of Shape Functions

The differentiation rule D is applied to the shape functions Ω regarding the
deflection surface and yields the matrix

B = D ·Ω. (6.26)

The physical meanings of the components of B turn out to be the curvatures

with respect to the related shape functions. To differentiate the entries of the
matrix of shape functions, the matrix scheme may be applied, given in Figure
6-7. Here, the matrix B is arranged with respect to the nodes of the element.
Thus, each sub–matrix has the same structure but with respect to the shape
functions related to the node.

node A node B node C node D

Ω
D

Fig. 6-7 Multiplication scheme B = D · Ω

The submatrix BA comprising the shape functions φA1 , φA2 , φA3 , φA4 applies
for the node A, e.g., as follows

φA1 φA2 φA3 φA4

−∂xx −φA1 ,xx −φA2 ,xx −φA3 ,xx −φA4 ,xx
.
−∂yy −φA1 ,yy −φA2 ,yy −φA3 ,yy −φA4 ,yy
−2∂xy −2φA1 ,xy −2φA2 ,xy −2φA3 ,xy −2φA4 ,xy

The differentiation of the shape functions with respect to the coordinates may
be seen more clearly applying product notation
 
−φ1 (x),xx φ1 (y) −φ2 (x),xx φ1 (y) −φ1 (x),xx φ2 (y) −φ2 (x),xx φ2 (y)
 
BA = −φ1 (x) φ1 (y),yy −φ2 (x) φ1 (y),yy −φ1 (x) φ2 (y),yy −φ2 (x) φ2 (y),yy  .
−2φ1 (x),x φ1 (y),y −2φ2 (x),x φ1 (y),y −2φ1 (x),x φ2 (y),y −2φ2 (x),x φ2 (y),y A
116 6 Bending Structures

6.2.6 Element Stiffness Matrix

The element stiffness matrix follows from the integration of
Z lxZ ly
K= BT · E · B dx dy . (6.27)
0 0

The integrand may be computed applying the scheme according to Figure 6-8
and is assigned to the corresponding element nodes.

A B C D
0
E ** ** 0 B
0 0 *

BT BT E B

Fig. 6-8 Multiplication scheme BT · E · B

The 16 × 16 stiffness matrix may be also computed and clearly arranged by

applying the 4 × 4 submatrices kij that are assigned to the nodes
 
kAA kAB kAC kAD
 kBA kBB kBC kBD 
 
K= .
 kCA kCB kCC kCD 
kDA kDB kDC kDD

All submatrices kAA , kAB , . . . are similarly arranged; only the shape functions
applied to build them are different.
6.2 Kirchhoff Plate Element with 16 DOF 117

The submatrix kAA , e.g., follows to

Z
kAA = BTA · E · BA dA
A
Z Z Z
=B IAA1 dA + νB IAA2 dA + 2B(1 − ν) IAA3 dA ,
A A A

whereat the multipliers of the integrals result from the elasticity matrix E. In
detail this yields
 
φA1 ,xx φA1 ,xx φA1 ,xx φA2 ,xx φA1 ,xx φA3 ,xx φA1 ,xx φA4 ,xx
 +φA1 ,yy φA1 ,yy +φA1 ,yy φA2 ,yy +φA1 ,yy φA3 ,yy +φA1 ,yy φA4 ,yy 
 
 φ , φ , φA2 ,xx φA2 ,xx φA2 ,xx φA3 ,xx φA2 ,xx φA4 ,xx 
 A2 xx A1 xx 
 +φ , φ , +φ , φ , +φ , φ , +φ , φ , 
 A2 yy A1 yy A2 yy A2 yy A2 yy A3 yy A2 yy A4 yy 
IAA1 =  ,
 φA3 ,xx φA1 ,xx φA3 ,xx φA2 ,xx φA3 ,xx φA3 ,xx φA3 ,xx φA4 ,xx 
 
 +φA3 ,yy φA1 ,yy +φA3 ,yy φA2 ,yy +φA3 ,yy φA3 ,yy +φA3 ,yy φA4 ,yy 
 
 φ , φ , φA4 ,xx φA2 ,xx φA4 ,xx φA3 ,xx φA4 ,xx φA4 ,xx 
A4 xx A1 xx
+φA4 ,yy φA1 ,yy +φA4 ,yy φA2 ,yy +φA4 ,yy φA3 ,yy +φA4 ,yy φA4 ,yy

 
φA1 ,xx φA1 ,yy φA1 ,xx φA2 ,yy φA1 ,xx φA3 ,yy φA1 ,xx φA4 ,yy
 +φA1 ,yy φA1 ,xx +φA1 ,yy φA2 ,xx +φA1 ,yy φA3 ,xx +φA1 ,yy φA4 ,xx 
 
 φ , φ , φA2 ,xx φA2 ,yy φA2 ,xx φA3 ,yy φA2 ,xx φA4 ,yy 
 A2 xx A1 yy 
 +φ , φ , +φ , φ , +φA2 ,yy φA3 ,xx +φA2 ,yy φA4 ,xx 
 A2 yy A1 xx A2 yy A2 xx 
IAA2 = ,
 φA3 ,xx φA1 ,yy φA3 ,xx φA2 ,yy φA3 ,xx φA3 ,yy φA3 ,xx φA4 ,yy 
 
 +φA3 ,yy φA1 ,xx +φA3 ,yy φA2 ,xx +φA3 ,yy φA3 ,xx +φA3 ,yy φA4 ,xx 
 
 φ , φ , φA4 ,xx φA2 ,yy φA4 ,xx φA3 ,yy φA4 ,xx φA4 ,yy 
A4 xx A1 yy
+φA4 ,yy φA1 ,xx +φA4 ,yy φA2 ,xx +φA4 ,yy φA3 ,xx +φA4 ,yy φA4 ,xx

 
φA1 ,xy φA1 ,xy φA1 ,xy φA2 ,xy φA1 ,xy φA3 ,xy φA1 ,xy φA4 ,xy

 φA2 ,xy φA1 ,xy φA2 ,xy φA2 ,xy φA2 ,xy φA3 ,xy φA2 ,xy φA4 ,xy 

IAA3 =  .
 φA3 ,xy φA1 ,xy φA3 ,xy φA2 ,xy φA3 ,xy φA3 ,xy φA3 ,xy φA4 ,xy 
φA4 ,xy φA1 ,xy φA4 ,xy φA2 ,xy φA4 ,xy φA3 ,xy φA4 ,xy φA4 ,xy
Thus the stiffness matrix follows with three parts as well

kAA = kAA1 + kAA2 + kAA3 ,

where k1 is assigned to the integral I1 , k2 to the integral I2 and k3 to the

integral I3 .
118 6 Bending Structures

The representation of the integrals may be clarified by introducing the coeffi-

cients
α = ℓy /ℓ3x , β = ℓx /ℓ3y and γ = 1/ℓxℓy ,
which are related to the differentiation of the shape functions and to the inte-
gration with respect to the element area.
The numbers of the entries of the submatrices resulting from integration are
listed below for the submatrix kAA . All submatrices are symmetric. Therefore,
only the upper triangle matrix is represented. The rows and the columns of the
matrices have to be multiplied by the specified coefficients. These coefficients
show that the units of the different virtual as well as real nodal displacement
variables are different. It is applied

1 ℓx ℓy ℓx ℓy
28080 14040 3960 1980 ·1
B·α
kAA1α = · 9360 1980 1320 ·ℓx ,
6300
720 360 ·ℓy
240 ·ℓx ℓy

1 ℓx ℓy ℓx ℓy
28080 3960 14040 1980 ·1
B·β
kAA1β = · 720 1980 360 ·ℓx ,
6300
9360 1320 ·ℓy
240 ·ℓx ℓy

1 ℓx ℓy ℓx ℓy
18144 9072 9072 1386 ·1
ν ·B·γ
kAA2 = · 2016 7686 1008 ·ℓx ,
6300
2016 1008 ·ℓy
224 ·ℓx ℓy

1 ℓx ℓy ℓx ℓy
9072 756 756 63 ·1
2 (1 − ν) · B · γ
kAA3 = · 1008 63 84 ·ℓx .
6300
1008 84 ·ℓy
112 ·ℓx ℓy
6.2 Kirchhoff Plate Element with 16 DOF 119

6.2.7 Element Load Vector

Assuming a constantly distributed external action p = [ p ] the element load
vector follows in matrix notation to
Z Z
f= ΩT · p dA = ΩT dA · p . (6.28)
A A
Integration yields
f T = [ 36 6ℓx 6ℓy ℓx ℓy (6.29)
36 −6ℓx 6ℓy −ℓx ℓy
36 −6ℓx −6ℓy ℓx ℓy
ℓx ℓy
36 6ℓx −6ℓy −ℓx ℓy ] · 144
·p.
Under arbitrarily distributed external surface actions, the loading can be ap-
proximated employing bi–linear shape functions and nodal values, in analogy
to a membrane element. With
p = Ωp · p̃
describing the course of the surface action regarding the values p̃ at the element
nodes it applies
Z Z
f= ΩT · p dA = ΩT · Ωp dA · p̃ . (6.30)
A A
The integrand is comprised of a 16 × 4 matrix, which is multiplied by the vector
of nodal action values after integration. Usually this procedure is avoided, since
it hardly influence the results, if small elements are needed for a sufficient
approximation of the deformation behavior.

6.2.8 Subsequent Stress Analysis

The bending and the twisting moments are computed from the material equa-
tions introduced at Section 6.2.1. The elasticity matrix is given by
 
1 ν
 
E=B· ν 1 
1
2
(1 − ν)
and the matrix of curvatures by
  
−φA1 ,xx −φA2 ,xx −φA3 ,xx −φA4 ,xx " # " # " #
  
B =   −φA1 ,yy −φA2 ,yy −φA3 ,yy −φA4 ,yy  .
−2φA1 ,xy −2φA2 ,xy −2φA3 ,xy −2φA4 ,xy B C D
120 6 Bending Structures

In this way, the stress matrix Sm = E · B and thus the courses of the plate’s
moments may be computed with
  
−φA1 ,xx − νφA1 ,yy "# "# "#
  
σ = E · B · v = B ·   −νφA1 ,xx − φA1 ,yy  ·v.
B C D
−(1 − ν)φA1 ,xy A2 A3 A4

The stress analysis concerning the moments is performed at the corner nodes of
the element yielding the nodal moments, which can be evaluated by applying
the stress matrix S̃m explicitly given on the next page.

σ̃ = S̃m · v .

The shear forces may be computed analogously to the procedure presented for
the Euler–Bernoulli beam by means of the equilibrium conditions
 
" # " # " # mxx
qx mxx ,x + mxy ,y ∂x 0 ∂y  
= = ·  myy  = Dq · σ .
qy myy ,y + mxy ,x 0 ∂y ∂x
mxy

Employing the shape functions regarding the deflection it explicitly applies

" # "" #   #
qx −φA1 ,xxx − φA1 ,yyx
=B· ·v
qy −φA1 ,yyy − φA1 ,xxy A2 A3 A4 B C D

or in matrix notation

q = Dq · σ = Dq · ( E · B ) · v
= Sq · v .

The stress analysis concerning the shear forces is performed at the corner nodes
of the element as well. It applies

q̃ = S̃q · v ,

whereat S̃q is the stress matrix related to the shear forces at the element nodes.
It should be mentioned, that the shear forces may also be computed by means
of the equilibrium conditions and the first derivatives of the moments, if the
moments are described with a bi–linear approximation with respect to the cor-
ner values.
The moments, regarding the element nodes A to D, are evaluated applying the following stress matrix S̃m
including the abbreviation f = −(1 − ν)

6 4 4ν 2 2ν
ℓ2x
+ 6ν
ℓ2y ℓx ℓy
0 ℓx
0 0 0 0 0 0 ℓ2
0 ℓy
0
x y
− ℓ62 − 6ν
4 2ν 2
 

0 ℓ2 ℓx
0 0 0 0 0 0 0 ℓy
y x y
− 6ν − ℓ62
 

0 0 f 0 0 0 0 0 0 0 0 0 0 0
 6ν 6 4ν 

mxx
 ℓ2 + ℓ2 ℓx ℓy 0 
 x 

myy 6 4ν 2ν
 

0 + 6ν 0 0 0 0 0 0
  

ℓx x ℓy ℓ2 ℓy
 0 0 

mxy ℓ2x ℓ2x ℓ2y y

− ℓ4 − 6ν
 

6ν 2ν 6ν 4 2
    

mxx 0 0 0 0 0 0 0 0
 −6 −2 0 0 

+ ℓ62 ℓ
 

ℓx ℓx ℓ2x y x ℓy y ℓy
 

− 4ν − ℓ62
 
6.2 Kirchhoff Plate Element with 16 DOF

myy
  
A   

0 0 0 0 0 0 f 0 0 0 0 0 0 0
 

mxy
   − 2 − 0 
 

6
    

mxx
 

0 0 0 0 + 6ν 0 0
 0 0 

ℓ2 y ℓ y ℓ2x ℓ2y x ℓ y x x
 

− 6ν 0 − 2ν − ℓ4 − 4ν − ℓ62 − ℓ2
 

myy
 B  ·v.

6ν
 =B·
 

mxy 0 0 0 0 0
 

+ ℓ62 ℓ ℓ2 ℓ
y x y x
 0 0 

y ℓ2x y x
− ℓ62 − ℓ2 0 − 4ν − ℓ4 0 − 6ν − 2ν

    

mxx
   

0 0 0 0 0 0 0 0 0 0 f 0 0 0
   0 0 

myy
 C   
 

2 6 6µ 4
 

mxy D
 

0 0 0 0 0 0 0
 

ℓy ℓx
+ ℓ2 ℓx ℓ y
 0 0 

ℓ2y x ℓ2x y
0 − 2ν − ℓ62 − 4ν
    

6 2ν 6ν 4ν
 

0 0 0 0 0 0 0
 − 6ν 0 

+ ℓ62
ℓy y ℓ2x ℓx ℓ2x y ℓx y
 

0 − ℓ2 − 6ν − ℓ4
 
 

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 f
 − 2 0 
 
121
 122

The shear forces, regarding the element nodes, are evaluated applying the related stress matrix S̃q

6 6 4 12 2
ℓ3 ℓ2 ℓ2
0 ℓy ℓ3x
0 0 0 0 0 0 0 0 ℓy
x y
− 12 − − ℓ62 − ℓ62
2 12
 

ℓx ℓy ℓx
0 0 ℓx
0 0 0 0 0
x ℓ3y y
− ℓ62 − ℓ62
qx
 x y x 

12 6 6 4 2
 12 

qy A 0 0 0 0 0 0 0 0
    − 3 0 62− 62 4 0 

ℓ3x ℓ2y ℓ2x ℓy y ℓy

−6 − − ℓ62
 ℓy 

qx
 

6 6 12
   

0 0 0 0 0 0
 

qy x ℓ3
 − 12 0 

x y ℓ2x ℓ2y ℓx ℓ3y y

0 − ℓ62 − ℓ2 − 12 − −4 − ℓ62

    ℓ3x ℓ2x 
 

12 6 6
 

qx
 0 0 

0 0 0 0 0 0 0
 

y y ℓ3x ℓ2y ℓ2x y ℓ3 x x

 

− ℓ62 − ℓ2 − − ℓ4 − 12 − ℓ62
 B

qy
 

12 6 6
   ·v.
  = B ·

0 0 0 0 0 0 0 0
 0 0 

ℓ3
 

qx y y ℓ3y ℓ2x ℓ2y ℓx x

− 12 − ℓ62 − −4 − ℓ62
   
   

qy 12 6 6
 

D
C

0 0 0 0 0 0 0
  0 − ℓ2 

ℓy y ℓ3x x ℓ3x ℓ2y ℓ2x

  

0 − ℓ2 − ℓ62 − 12 −
   x
 

12 6 2 12 6 6 4
 

0 0 0 0 0 0 ℓx
0 ℓx
 0 − 62 − ℓ4 

y y x ℓ3y ℓ2x ℓ2y

− ℓ3 0 − ℓ2 0 − ℓ62 −
 y
 
6 Bending Structures
6.2 Kirchhoff Plate Element with 16 DOF 123

6.2.9 Example of Use

As an example of use the square plate loaded by a constantly distributed ex-
ternal action according to Figure 6-9 is investigated. The boundaries at their
respective opposite sides are clamped or simply supported. Thus, only a quar-
ter of the plate may be discretized, if symmetry conditions are considered. The
discretization employing 2 × 2 elements is depicted in Figure 6-9.

x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
7 2
E = 3.0 . 10 kN/m
T
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 6-9 Quadratic plate – geometry and loading

Representation of Results
The course of the deflection and the moments is depicted in Figure 6-10 regar-
ding the 16 DOF element at selected sections applying a discretization of 2 × 2
and 4 × 4 elements respectively. The deflection gives only little changes due to
mesh refinement, which indicates that the chosen approach may approximate
the deformation behavior very well.
The course of the bending moments is comprised of linear parts as well as of
cubic parts superposed because of lateral curvatures. The respective linear part
is represented only, which still shows significant changes in the case of mesh
refinement due to the second derivatives of the deflection surface. The quality
of the approximate solution of the respective discretization may be identified
from the jumps of the bending moments at the element intersections as well as
from the fulfillment of the boundary condition mxx = 0 at position G.
The twisting moment is directly computed employing the nodal displacement
variable w,xy , thus it is approximated as well as the deflection surface itself.
The course of the twisting moment is parabolic along the element boundaries
and given here as a polygonal line.
124 6 Bending Structures

section G − M: G M
0

w [mm]

G M
0
+ +
1.0

m xx [kNm/m]

section M − E: M E
0
5

w [mm]

M
0
+ E
3.0

m yy [kNm/m]

M
0
E
1.0 +

m xx [kNm/m]

section G − GE: G GE
0
− 1.0

m xy [kNm/m]

Fig. 6-10 Square plate – state variables regarding different discretizations

6.2 Kirchhoff Plate Element with 16 DOF 125

Investigation of Convergence
Table 6.1 represents the convergence behavior of the most important state va-
riables regarding the 16 DOF element, cf. Figure 6-11 and the comparison with
other elements is given in Section 10.4. As shown in Section 6.1.9, the deflection

Table 6.1 16 DOF plate element – convergence behavior of state variables

element pattern 1×1 2×2 4×4 8×8 16 × 16

wM [cm] 0. 74627 0. 73603 0. 73616 0. 73618 0. 73618
mxyT [kNm/m] ÷ −1. 516 −1. 536 −1. 537 −1. 537
mxxM [kNm/m] 2. 814 2. 219 2. 164 2. 156 2. 154
myyM [kNm/m] 5. 040 3. 481 3. 236 3. 183 3. 170
myyE [kNm/m] −4. 664 −6. 194 −6. 760 −6. 925 −6. 969
qxG [kN/m] 1. 87 1. 63 1. 93 2. 16 2. 29
qyE [kN/m] 1. 87 3. 15 4. 06 4. 59 4. 87

exhibit only little errors, even when employing a small number of elements. Due
to the chosen shape functions the convergence behavior concerning the twisting
moment mxy is just as effective, because mxy may be directly computed from
the twisting w,xy , which is a nodal degree of freedom, but not from the deri-
vative of the deflection surface. The bending moments converge more slowly
against the exact solution, since the second derivative of the deflection surface
causes a roughening of the course of the corresponding state variable. The shear
forces converge even less effectively, since they are computed by applying the
third derivative of the deflection surface. Figure 6-11 shows the development
of the state variables z with respect to mesh refinement, whereat the reference
variables zexact are chosen from the 16 × 16 mesh.
z
z exact myy M
mxx M
w
1.0 mxy T
myy E
0.5

1 4 16 64 elements
16 36 100 324 unknowns
Fig. 6-11 Square plate – convergence behavior regarding the state variables
126 6 Bending Structures

6.3 Kirchhoff Plate Element with 12 DOF

Due to the twistings w,xy being nodal displacement variables, the approach
is no longer coordinate–invariant. To get a coordinate–invariant approach, no-
dal displacement variables w,xx and w,yy would also be required to uniquely
transform the curvatures.
As an alternative to the 16 DOF element, an approach may be chosen with-
out comprising any twisting w,xy , which consequently employs only 12 nodal
displacement variables, [ w w,x w,y ] at each node. Although this approach is
not conform, it still converges to the correct solution, cf. Section 6.2.9. A direct
derivation of the approach comprising 12 nodal displacement variables is not
possible, but it may be derived by means of a general polynomial comprising 12
degrees of freedom ai . 10 degrees of freedom are assigned to a complete cubic
approach according to Pascale’s triangle and employing 10 power terms. In
addition, the power terms x y 3 and x3 y are related to the remaining two degrees
of freedom. This approach originally was developed by Adini and Clough [1]
and Melosh [71]. It comprises
w(x, y) = a1 + a2 x + a3 y
+ a4 x2 + a5 xy + a6 y 2
+ a7 x3 + a8 x2 y + a9 xy 2 + a10 y 3 (6.31)
+ a11 x3 y + a13 xy 3 ,
or u = ψ 12 · a12 .
The scaling matrix of the 12 DOF element can now be derived from the ap-
proach of the 16 DOF element, as follows. According to Equation (6.25), the
degrees of freedom a12 , a14 , a15 , a16 are assigned to the power terms, which are
no longer taken into account. Thus the corresponding columns of the scaling
matrix Ψ̃ related to the 16 DOF element must be eliminated. Moreover, the
fourth, eighth, twelfth and sixteenth row is to be eliminated, since the twisting
nodal displacement variables are omitted. It applies
v12 = Ψ̃12 · a12
and subsequently
−1
a12 = Ψ̃12 · v12 = G12 · v12 .
In this way, the shape functions employing physically meaningful nodal degrees
of freedom may be computed, applying the scaling matrix, to
Ω12 = ψ 12 · G12 .
All further steps to compute the element matrices are the same with regard to
the 16 DOF element.
 1 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0
 

0 0 1 0 0 0 0 0 0 0 0 0
 

3
 

a1 0 2 0 0 0 0 0 0 0 w
x x
 

x lx
− l32 − l2 − l1
 

a2 w,x
1 1 1 1
       

0 0 0 0
 

a3 x y y x lx ly lx x y lx ly ly w,y
 

− l 1l − l1 − l1 − l 1l


w
      

a4 3
     

0 0 0 0 0 0 0 0


y y ly2 y w,x
    

a5 − l32 − l2 − l1
      A 


2 1 1 w,y
    

a6


3 2 0 2 0 0 0 0 0 0 0
  
  
6.3 Kirchhoff Plate Element with 12 DOF

lx lx x lx w
− l23
      

a7
     

3 2 1 3


w,x
    

a8 0 0
2l 2l
  =    

y lx ly lx ly y x y x y
 B

lx x y lx x y


0 − l23l − l 1l 0 − l23l − l 2l
  · 

a9 w,y
     

3 2 3 1


w
      

a10 0 0


lx ly2 lx ly x y x y lx ly2 x y x y lx ly
    

− l 3l2 0 − l 2l 0 − l 1l − l 3l2
     

a11 w,x


2 1 1
  
   C 

0 0 0 0 0 0 0 0
    

a13 w,y


ly3 ly2 y ly2 D

− l23
  
  
      

2 2 1
 

0 0 0 0
 

3l 3l 2l
x y x y lx y x y x y x y lx y lx y
 

− l32l − l21l − l21l − l32l l21l

 

2 1 1 2
 

0 0
 

x y x y lx ly3 lx ly2 x y lx ly2 lx ly3 x y

− l 2l3 0 − l 1l2 − l 2l3 0 − l 1l2
 
 
127
128 6 Bending Structures

The shape functions regarding the 12 DOF element are depicted in Figure
6-12. The breach of the element intersection conditions is obvious, since the
shape functions related to the rotations become linear in one direction, and
thus exhibit a kink at the intersection to neighboring elements.

Fig. 6-12 Shape functions of the 12 DOF element

Example of Use
For comparison the following Table 6.2 shows the convergence behavior of the
12 DOF element. The deflection converges significantly less effectively in com-
parison to the 16 DOF element, since the clamping and the element intersecti-
on conditions with respect to the rotations cannot be exactly described, which
yields a more flexible structural behavior. In contrast, the moments as well as
the shear forces are relatively well approximated, because the omitted twisting
degrees of freedom w,xy does not influence the bending moments mxx , myy and
the shear forces qx , qy that much. It should be mentioned that the twisting
moment mxy now has to be computed at the element level with poorer conver-
gence, since the corresponding curvature w,xy is no nodal degree of freedom as
it is in the case of the 16 DOF element.

Table 6.2 12 DOF element – convergence behavior of state variables

element pattern 1×1 2×2 4×4 8×8 16 × 16

wM [cm] 0. 9702 0. 8017 0. 7527 0. 7403 0. 7372
mxyT [kNm/m] −1. 751 −1. 570 −1. 546 −1. 539
−1. 524 −1. 553 −1. 542 −1. 538
mxxM [kNm/m] 3. 107 2. 385 2. 207 2. 167 2. 157
myyM [kNm/m] 6. 443 3. 777 3. 308 3. 201 3. 175
myyE [kNm/m] −6. 064 −6. 721 −6. 899 −6. 960 −6. 977
qxG [kN/m] 1. 14 1. 60 1. 92 2. 16 2. 29
qyE [kN/m] 2. 80 3. 75 4. 36 4. 74 4. 94
6.4 The 12 DOF Element Employing a Weak Conformity 129

6.4 The 12 DOF Element Employing a Weak Conformity

The plate element with 12 DOF employs the deflection w and the rotations w,x
and w,y as nodal unknowns and hence does not have the difficulties recognized
in plate elements with 16 DOF. Nonetheless, the element does not satisfy the
conditions of conformity at the intersections of neighboring elements. Instead
of introducing additional variables at the intersections to satisfy the conditions,
the work equation may be extended by the work of the stress variables at the
respective intersection. This leads to a weak formulation of the conformity with
respect to the slope w,n and improves the properties of the element remarkably.

6.4.1 Stiffness matrix

The work equation of the Kirchhoff theory of plates is already given explicitly
in Section 6.1 with
Z
−δW = B {δw,xx · (w,xx + νw,yy ) + δw,yy · (νw,xx + w,yy )
Z
1−ν
+2 δw,xy · w,xy } dxdy − δw · p dxdy . (6.32)
2

Because of the non–conforming approach the work equation must be completed

by the work performed by the moment on the conjugated rotation along the
boundary.
The nodal unknowns w w,x w,y , transformed to the boundary coordinates,
are available on each boundary as w w,s w,n , with s being the coordinate
in the direction of the boundary and n normal to it. The DOF w and w,s
suffice to fulfill the element boundary conditions for w and w,s as a 3rd order
Hermite–polynomial. The two remaining nodal DOF w,n suffice to describe
the slope w,n as a linear polynomial along the boundary and thus may reflect
the corresponding stiffness. Hereby the virtual work with respect to Equation
(6.32) satisfies the continuity condition at the intersection up to a linear course
of w,n but neglect the quadratic part, which is necessary for the cubic approach,
see Figure 6-13.

quadratic part

linear part w,n

w,n B
A

Fig. 6-13 Course of w,n at the element boundary

130 6 Bending Structures

In order to satisfy the continuity with respect to the rotations w,n , the work
equation has to be extended by additional terms at the element boundaries,
which take into account the work perfomed by the bending moments mn on the
corresponding rotations. The bending moments are replaced by the material
equations, thus leading to a pure displacement–based formulation with
Z Z
− δWel.b. = δmn · (w,n quad − w,n lin )} ds + mn · (δw,n quad − δw,n lin )} ds
Z
= B · (δw,nn + ν δw,ss ) · (w,n quad − w,n lin )} ds
Z
+ B · (w,nn + ν w,ss ) · (δw,n quad − δw,n lin )} ds .

Applying the extended work equation, the continuity conditions for w,n and
δw,n can be fulfilled weakly at the system level, too.

6.4.2 Example of Use

The example deals with the square plate with clamped and hinged bounda-
ries which is already investigated in Section 6.2.9 with respect to a constantly
distributed loading, cf. Figure 6-14.
x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
7 2
E = 3.0 . 10 kN/m
T
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 6-14 Quadratic plate – system and loading, see Section 6.2.9

The deflection at the centre of the plate is given in Table 6.3 concerning the
element with 12 DOF, taking into account boundary integrals of work, and
applying it to different meshes. The results obtained by the modifications are
excellent, cf. Figure 6-15, where the results are compared to the different rectan-
gular elements with 16 DOF and 12 DOF respectively. The deflection converges
almost as well as for the element with 16 DOF.
6.4 The 12 DOF Element Employing a Weak Conformity 131

The convergence of the bending moments is comparably good. This is confirmed

for the bending moments mxx and myy at the centre of the plate M , see Table
6.3. Since the twisting w,xy is no longer a nodal degree of freedom, the twisting
moment has to be computed inside the elements by means of the differentiated
shape functions. Thus the twisting moment mxyT comprises the two values of
the neighboring elements.

Table 6.3 12 DOF element – convergence of state variables

element pattern 1×1 2×2 4×4 8×8 16 × 16

wM [cm] 0. 7692 0. 7449 0. 7408 0. 7376 0. 7365
mxyT [kNm/m] −1. 648 −1. 723 −1. 566 −1. 543
−1. 740 −1. 587 −1. 556 −1. 542
mxxM [kNm/m] 2. 564 1. 869 2. 097 2. 140 2. 150
myyM [kNm/m] 5. 128 3. 204 3. 181 3. 170 3. 167
myyE [kNm/m] −4. 808 −6. 036 −6. 705 −6. 910 −6. 965
qxG [kN/m] 0. 96 0. 46 0. 96 1. 22 1. 36
qyE [kN/m] 2. 24 3. 27 4. 16 4. 65 4. 91

1.3
normalized centre deflection

1.2 16 DOF element

12 DOF element
12 DOF element, weak conformity
1.1

1.0

1 4 16 64 256
number of elements

Fig. 6-15 Quadratic plate – convergence of the centre deflection wM

TRIANGULAR ELEMENTS
7 Triangular Elements – Description of Geometry

A disadvantage of rectangular elements is undeniably the restricted field of

application, which, according to Section 4, 5 and 6, is limited to rectangular
geometries. Furthermore the mesh refinement, often required, is very inefficent
if it covers the whole domain.
An element geometry allowing the discretization of arbitrary geometries as well
as an efficent mesh refinement is the triangle. Triangular elements seem to have
the ideal geometry for the FEM, but the derivation of the element matrices is
slightly more difficult when compared to rectangular elements.
The rectangular elements previously studied are described in the orthogonal
x–y–coordinate system, so that the edges of the elements are parallel to the
direction of the coordinates. Therefore, the choice of shape functions and the
integration of the element matrices as well as the load vectors turn out to be
relatively straight forward. This only holds for right–angled triangles if the coor-
dinate system forms the right angle. In contrast, an arbitrary triangle would
have to be divided into multiple integration domains when integrated in x– and
y–coordinates, which is very time–consuming. The definition of shape functi-
ons related to triangular geometries in x–y–coordinates would become very
complex, too, because in general, distinctions between different cases would be
necessary. Thus when employing triangular elements, it is more advantageous
to apply a local element coordinate system.
Since the work equation
Z Z
−δWel = δvT · { ΩT · DT · E · D · Ω dxdy · v − ΩT · p dxdy} (7.1)

is related to the x–y–coordinates it turns out, that applying a local coordinate

system needs a transformation of the coordinates, a transformation of the deri-
vatives with respect to the coordinates, and a transformation of the integration
rules.

7.1 Local ξ–η–Coordinate System

In addition to the global x–y–coordinate system, in which the structure is lo-
cated and in which the governing equations as well as the weak formulation are
described, local ξ–η–coordinates are defined with respect to two of the element
edges, see Figure 7-1. Each of the three element corner nodes is equivalent to
the others and may be chosen as the origin of the local coordinate system.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_7
136 7 Triangular Elements – Description of Geometry

Because of the coordinate invariance, this must not influence the results. The
ξ–η–coordinates are normalised to the length of the element edges ℓAB and
ℓAC , so that the conditions
0≤ξ≤1 and 0≤η≤1
are fulfilled.
xc xa xb x

A
ya ( ξ = 0; η = 0 )

η ξ

yc C
( 0;1 )
yb B
( 1;0 )
y

Fig. 7-1 The coordinate system for triangular elements

With regard to the element domain, the shape functions and the integration
limits can be clearly formulated by applying local coordinates.

7.1.1 Transformation of Coordinates

The transformation from one coordinate system into the other can be performed
by means of the corner coordinates, see Figure 7-1, applying
x = xa + (xb − xa ) · ξ + (xc − xa ) · η ,
y = ya + (yb − ya ) · ξ + (yc − ya ) · η .
Thus it might be useful to employ the following abbreviations concerning the
coordinate differences of the related element nodes
aa = xc − xb ,
ab = xa − xc , with aa + ab + ac = 0 ,
ac = xb − xa
and
ba = yb − yc ,
bb = yc − ya , with ba + bb + bc = 0 .
bc = ya − yb
7.1 Local ξ–η–Coordinate System 137

Hence the matrix notation of the transformation of coordinates gives

" # " # " # " #
x xa ac −ab ξ
= + · . (7.2)
y ya −bc bb η

The inverse representation is needed with respect to the re–transformation.

The matrix notation yields
" # " # " # " #
ξ 1 b b ab x xa
= · ·{ − } (7.3)
η det b c ac y ya

applying det = ac · bb − ab · bc .

7.1.2 Transformation of Partial Derivatives

The work equation comprises the partial derivatives of the displacement field
u(x, y) with respect to the coordinates x and y. If the displacement field is
described in relation to local coordinates ξ and η, the derivatives of the displa-
cement field with respect to the coordinates ξ and η need to be transformed in
accordance with the chain rule
∂u ∂u ∂ξ ∂u ∂η
= · + · ,
∂x ∂ξ ∂x ∂η ∂x
∂u ∂u ∂ξ ∂u ∂η
= · + · .
∂y ∂ξ ∂y ∂η ∂y
The derivatives of the local coordinates with respect to the global coordinates
follow Equation (7.3), e.g.
∂ξ ∂ 1 bb
= { · [ bb (x − xa ) + ab (y − ya ) ]} = .
∂x ∂x det det
Concerning all derivatives, the matrix notation yields
   ∂ξ ∂η    " # 
∂u ∂u ∂u
∂x ∂x ∂x ∂ξ 1 bb bc ∂ξ
 =  = ·  . (7.4)
∂u ∂ξ ∂η ∂u det ab ac ∂u
∂y ∂y ∂y ∂η ∂η

The inverse representation gives

  " # 
∂u ∂u
∂ξ ac −bc ∂x
 =  . (7.5)
∂u −ab bb ∂u
∂η ∂y
138 7 Triangular Elements – Description of Geometry

The coefficient matrix of the inverse representation is named Jacobian matrix

J. The determinant of the Jacobian matrix gives

det J = ac · bb − ab · bc .

Thus " #
ac −bc
J= (7.6)
−ab bb
holds as well as " #
−1 1 bb bc
J = .
det J ab ac

7.1.3 The Integration of the Element Area

To prepare the element stiffness matrix, at integration of the work performed
in the respective element, the differential domain dA = dx · dy needs to be
described by means of an equivalent expression in dξ and dη. Using vector
calculus, an area can be computed by the vector product of the vectors spanning
the domain. As shown in Figure 7-2, the differential vectors in the direction of
the local coordinates are

ξ = 0, η = 0

   
(xb − xa ) dξ ac dξ
ξ=0 η=0 dvξ =  (yb − ya ) dξ  =  −bc dξ  ,
A
0 0
   
(xc − xa ) dη −ab dη
η A dvη =  (yc − ya ) dη  =  bb dη  .
η=1 0 0
ξ=1 ξ

Fig. 7-2 Description of the triangle’s area

The vector product provides a third vector whereby its norm is related to the
value of the area
   
ex ey ez 0
dvξ × dvη = det  ac dξ −bc dξ 0  =  0 .
−ab dη bb dη 0 [ ac · bb − ab · bc ] dξ dη
7.1 Local ξ–η–Coordinate System 139

The component in z–direction gives the area of the differential parallelogram

spanned by the coordinates dξ and dη, cf. Figure 7-2,

dA = [ ac · bb − ab · bc ] dξ dη = det J dξ dη . (7.7)

At integration of the work related to the domain of the triangle, the integration
limits need to be adapted to the triangular geometry. Because of the crooked
boundaries, the upper integration limit is variable in the ξ–direction,
Z Z Z 1 Z 1−η
Integrand dx dy = Integrand · det J dξ dη .
y x 0 0

Thus the area of the triangular domain can be evaluated

Z 1 Z 1−η Z 1
1
A∆ = det J dξ dη = (1−η) det J dη = [ η−η 2 /2 ]10 det J = det J .
0 0 0 2
When regarding integrals employing polynomials of an arbitrary order, the
solution can be analytically obtained by
Z 1 Z 1−η
i! k!
ξ i · η k dξdη = . (7.8)
0 0 (i + k + 2)!

7.1.4 Linear Shape Functions and Nodal Degrees of Freedom

The linear shape functions employed to describe the displacement field are
chosen as shown in Figure 7-3. Each shape function describes a plane tilted

1 1

A B A B A B
1

C C C

Fig. 7-3 Linear shape functions regarding triangular elements

over one edge of the element wherein its value is equal to one regarding the
opposite node

u = [ u(x, y) ] = Ω · v
 
uA
 
= [ φA φB φC ]  uB  ,
uC
140 7 Triangular Elements – Description of Geometry

with the nodal degrees of freedom

v T = [ uA uB uC ]

and the shape functions

φa = 1 − ξ − η, φb = ξ , φc = η .

Such, all shape functions are equivalent regarding the corner nodes A, B and
C, see Figure 7-3.

7.2 Description Employing Area Coordinates

As an alternative, the stiffness matrix can be derived employing area coordi-
nates. In comparison to the local ξ–η–coordinates, area coordinates have the
advantage that the shape functions are clearly recognized ab initio as being
symmetric concerning all element nodes and thus the element stiffness matrix
will be.
The position of an arbitrary point P within a triangular element can be de-
scribed by means of the triangles formed by the lines connecting P with the
element corners, cf. Figure 7-4.
xp x

A (
λ a = 1, λ b = 0, λ c = 0 )

Ab Ac
yp P
C Aa
( 0, 0, 1 )
B
y ( 0, 1, 0 )

Fig. 7-4 The definition of area coordinates

The triangle is subdivided into the three triangles Aa , Ab and Ac . The indices
correspond to the opposite corner points. The arbitrary position of P within
the triangle is set by means of the area coordinates
Aa Ab Ac
λa = , λb = , λc = , (7.9)
A△ A△ A△
7.2 Description Employing Area Coordinates 141

whereby, because of Aa + Ab + Ac = A△ , it holds additionally

λa + λb + λc = 1 . (7.10)
The area coordinates fulfill the constraint (7.10), so only two independent coor-
dinates act here as well. The area coordinates λa , λb and λc are normalized
and thus are independent of the size of the element as well as of its position in
the x–y–coordinate system.

7.2.1 Transformation between Cartesian and Area Coordinates

The coordinates x and y of an arbitrary point P (x, y) inside the triangle can
be described by means of area coordinates as follows:
x = xa · λa + xb · λb + xc · λc ,
(7.11)
y = ya · λa + yb · λb + yc · λc .
This can be controlled very easily with respect to the corner nodes regarding
Figure 7-4, as the value for the related λ has to equal 1, and the two others must
equal 0. At the midpoints of the element edges, two values for λ equal 1/2, and
the third one equals 0. Concerning the centre of the triangle, all coordinates
equal 1/3.
Taking into account the constraint (7.10) with respect to λi the transformation
instruction applies in the matrix notation
     
1 1 1 1 λa
 x  =  xa xb xc  ·  λb  . (7.12)
y ya yb yc λc
The re–transformation may be performed by applying the inverse representa-
tion
     
λa xb yc − xc yb yb − yc xc − xb 1
  1    
 λb  =  xc ya − xa yc yc − ya xa − xc  ·  x  .
det
λc xa yb − xb ya ya − yb xb − xa y
The determinant of the coefficient matrix equals twice the triangular area, cf.
Section 7.1.3
1 1 1
det xa xb xc = (xb yc − xc yb ) − (xa yc − xc ya ) + (xa yb − xb ya )
ya yb yc
= (xb − xa )(yc − ya ) − (xc − xa )(yb − ya )
= ac · bb − ab · bc = 2A△ .
142 7 Triangular Elements – Description of Geometry

The re–transformation can be shortened to

     
λa xb yc − xc yb ba aa 1
  1    
 λb  =  xc ya − xa yc bb ab  ·  x  (7.13)
2A∆
λc xa yb − xb ya bc ac y
by employing the abbreviations defined above. Subsequently, the re–transfor-
mation is required to perform the transformation of the derivatives.

7.2.2 Transformation of Partial Derivatives

The differentiation of a displacemente field u(x, y), defined with respect to the
Cartesian coordinates x and y, can be performed by applying the chain rule, if
the shape functions for the description of the displacement are given employing
area coordinates. Therefore it holds that
∂u ∂u ∂λa ∂u ∂λb ∂u ∂λc
= · + · + · ,
∂x ∂λa ∂x ∂λb ∂x ∂λc ∂x
(7.14)
∂u ∂u ∂λa ∂u ∂λb ∂u ∂λc
= · + · + · .
∂y ∂λa ∂y ∂λb ∂y ∂λc ∂y
The derivatives of the area coordinates can be computed with respect to the
Cartesian coordinates regarding Equation (7.13)
∂λa 1 1
= (yb − yc ) = ba ,
∂x 2A△ 2A△
∂λb 1 1
= (yc − ya ) = bb ,
∂x 2A△ 2A△
∂λc 1 1
= (ya − yb ) = bc ,
∂x 2A△ 2A△
∂λa 1 1
= (xc − xb ) = aa ,
∂y 2A△ 2A△
∂λb 1 1
= (xa − xc ) = ab ,
∂y 2A△ 2A△
∂λc 1 1
= (xb − xa ) = ac .
∂y 2A△ 2A△
Thus it follows from Equation (7.14)
 
∂u 1 ∂u ∂u ∂u
= ba + bb + bc ,
∂x 2A△ ∂λa ∂λb ∂λc
  (7.15)
∂u 1 ∂u ∂u ∂u
= aa + ab + ac .
∂y 2A△ ∂λa ∂λb ∂λc
7.2 Description Employing Area Coordinates 143

A Jacobian matrix can be formally given, as with ξ–η–coordinates, as a 2 × 3–

matrix, but this is of no further relevance.

7.2.3 Integration with Respect to Area Coordinates

The integration of the work with respect to the area coordinates has to take into
account, that the three coordinates λa , λb , λc must satisfy the condition (7.10).
Thus, the integration has to be performed with respect to two area coordinates
only. Analogously to Figure 7-2 the differential area dx dy = detJ dξ dη may
also be transformed to the area coordinates, if dξ is replaced by dλa and dη is
replaced by dλb .
Z Z Z 1 Z 1−λb
Integrand dx dy = Integrand · det J dλa dλb .
y x 0 0

Hence, the area of the triangular domain can be evaluated with

Z 1 Z 1−λb Z 1
A∆ = det J dλa dλb = (1 − λb ) det J dλb
0 0 0
1
= [ λb − λ2b /2 ]10 det J = det J .
2
The integration of the individual terms is performed by applying the integration
rule related to the product terms of the area coordinates
ZZ
i! j! k!
λia · λjb · λkc dλa dλb = . (7.16)
(i + j + k + 2)!
It should be noted that λc is dependent on λa and λb for example, therefore
the integration has to be performed with respect to two area coordinates only.

7.2.4 Shape Functions Employing Area Coordinates

Considering the same nodal displacements in the same order as in Section
7.1.2, the general shape of the matrix comprising the shape functions remains
the same, see Figure 7-3. The representation of the shape functions employing
area coordinates and their respective derivatives are symmetric according to
the area coordinates:

φA = λa , φB = λb , φC = λc .

The comparison with the shape functions depicted in Figure 7-3 shows that
they are identical.
144 7 Triangular Elements – Description of Geometry

Shape functions of higer order

The development of complete shape functions of higher order employing area
coordinates happens differently from the procedure concerning ξ–η–coordi-
nates, whereby there is no representation comparable to Pascale’s triangle.
However, the number of the independent shape functions remains the same
as in Pascale’s triangle. Thus a complete approach of n–th order includes all
(n + 1) · (n + 2)/2 possible products of n–th order of the area coordinates.
As such, a linear approach employs three independent functions that depend
linearly on the area coordinates

λa , λb , λc .

A quadratic approach comprises six independent quadratic shape functions

λ2a , λ2b , λ2c , λa · λb , λa · λc , λb · λc .

A scaling to the corner and the mid points of the edges leads to, cf. Figure 7-5,

φA = λa (2λa −1) , φB = λb (2λb −1) , φC = λc (2λc −1) ,

φD = 4 λa λb , φE = 4 λb λc , φF = 4 λa λc .

A 1
C

1
B

1
1/4
1 1/4 F
1 1/4

D
E

Fig. 7-5 Quadratic shape functions

A cubic approach comprises ten independent cubic shape functions

λ3a , λ3b , λ3c , λ2a · λb , λ2a · λc , λ2b · λa , λ2b · λc , λ2c · λa , λ2c · λb , λa · λb · λc .

Shape functions of higher orders can be developed appropriately.

8 Triangular Elements to Describe Heat Conduction

The basic idea when performing the element matrix as well as the load vector is
still the weak formulation of the energy balance equation applying the matrix
notation
Z Z
−δRel = δv · { Ω · D · E · D · Ω dxdy · v − ΩT · p dxdy} .
T T T
(8.1)

At first, the element matrices regarding elements with a linear temperature ap-
proach are derived, describing constant heat fluxes inside the element. Elements
employing a quadratic or even higher polynomial approach may be derived but
are not looked at here. Concerning linear shape functions, all convergence cri-
teria are fulfilled. The derivation of the element matrices does not essentially
change compared with rectangular elements, thus the presentation is focused
on the application of the local element coordinate system.

8.1 A Linear Approach Related to the ξ–η–Coordinate System

The approach to describe the temperature inside the element domain can be
chosen in relation to the local coordinate system, since all necessary transfor-
mations are known between the coordinate systems. The convergence criteria
need to be fulfilled and checked as with any other type of element.
With regard to heat conduction, at least Co –conform shape functions are to be
employed, which have to be complete shape functions to fulfill the criterion of
coordinate invariance in the case of triangular elements.

8.1.1 Linear Shape Functions and Nodal Unknowns

The linear shape functions employed to describe the temperature field are cho-
sen as shown in Figure 8-1. Such, all shape functions are equivalent concerning
the corner nodes A, B and C. Each linear function describes a plane tilted over
one edge of the element wherein its value is equal to one regarding the opposite
node

u = [T ] = Ω·v

with the nodal degrees of freedom

v T = [ TA TB TC ] .

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_8
146 8 Triangular Elements to Describe Heat Conduction

The shape functions chosen with Ω and their respective derivatives are:

φA = 1 − ξ − η φB = ξ φC = η
φA ,ξ = −1 φB ,ξ = 1 φC ,ξ = 0
φA ,η = −1 φB ,η = 0 φC ,η = 1

1 1

A B A B A B
1

C C C

Fig. 8-1 Linear shape functions regarding triangular elements

Hence it applies explicitly to

 
TA
    
T (x, y) = φA φB φC  TB  .
TC

8.1.2 Differentiation of the Temperature Field

In the weak form of the balance equation, the differentiation rule related to
the computation of the heat fluxes from the temperatures is comprised of the
derivatives with respect to the x–y–coordinates. If the shape functions are
described in local ξ–η–coordinates, the derivatives need to be transformed with
respect to the local coordinates. Thus the operator matrix D is transformed
according to Section 7.1.2 to
" #  ∂ξ ∂η
 " #
−∂x ∂x
∂ξ + ∂x
∂η 1 bb ∂ξ + bc ∂η
D= = −  ∂ξ ∂η
=− , (8.2)
−∂y ∂ξ + ∂η det J ab ∂ξ + ac ∂η
∂y ∂y

whereby the negative sign is positioned in front of the matrix for convenience
here. Thus the matrix B, comprising the derivatives of the shape functions,
yields

B = D·Ω
"" # " # " # #
1 bb φA ,ξ + bc φA ,η
=− ·
det J ab φA ,ξ + ac φA ,η
A B C
8.1 A Linear Approach Related to the ξ–η–Coordinate System 147
      
1 −bb − bc bb bc
=− ·
det J −ab − ac A
ab B
ac C
" #
1 ba bb bc
=− · . (8.3)
det J aa ab ac

The matrix B turns out to be constant. Because the heat flux is also computed
by means of B, the heat flux is also constantly distributed inside the element.
Considering the chosen abbreviations for the coordinate differences, B may be
recognized now to be symmetric with respect to all element nodes.
It is remarkable that the order of the shape functions is reduced by one relating
to the whole element, not only in the respective direction like it does concerning
rectangular elements employing symmetric shape functions.

8.1.3 Element Matrix

The element matrix K follows analogously to Section 4
Z
K= BT · E · B dA
A
Z 1 Z 1−η
= BT · E · B · det J dξ dη .
0 0

In addition to the matrix of heat conductivity E

" #
λ
E= ,
λ

B is also constant here. When applying det J = 2A△ , the whole integrand
turns out to be constant. Thus, in this case, the element matrix is performed
by matrix multiplication BT · E · B analogously to the multiplication scheme
given in Section 4.2 and finally by multiplying the product by the factor
1 1
det J · = 2 A△ · = A△ .
2 2
Comparing to Section 4.2, the element matrix comprises three rows and three
columns, that are related to the virtual as well as to the real nodal temperatu-
res. This yields
 
b a · b a + aa · aa b a · b b + aa · ab b a · b c + aa · ac
λ
K= ·  b b · b a + ab · aa b b · b b + ab · ab b b · b c + ab · ac  .
4A△
b c · b a + ac · aa b c · b b + ac · ab b c · b c + ac · ac
148 8 Triangular Elements to Describe Heat Conduction

8.1.4 Vector of Thermal Action

At first, the vector of thermal action is computed for the special case of element
oriented, constantly distributed external energy input, cf. Section 4.2. With

p = [ qV ] = constant ,

this yields
Z Z
f= ΩT · p dA = ΩT dA · p .
A A

In the case of linear shape functions, the integral can be solved independently
of the chosen coordinate system. The external energy input is therefore equally
distributed to the three nodes by 1/3 each, because the shape functions are
equivalent
 
A△
 3 
 A△

f =

 · [ qV ] .

 3 
A△
3

A fluctuating external energy input may be described and taken into account
at performing the integration by means of the matrix of shape functions and
the nodal values of qV .

8.1.5 Subsequent Flux Analysis

Employing the chosen linear shape functions for the description of the tempe-
rature field, the matrix of the derivatives B is constant in terms of each entry,
resulting in constant heat fluxes. Therefore, the heat fluxes qx , qy are evaluated
only once for the whole element and can be assigned, for example, to the centre
of the element. At first, this holds in general
" #
qx
σ= = E·B·v = S·v.
qy

The stress matrix S follows, regarding the chosen linear shape functions to
" #
λ ba bb bc
S=E·B=− · .
2F△ aa ab ac
8.2 Example of Use 149

8.2 Example of Use

The following example is already investigated applying rectangular elements
in Section 4.3. Figure 8-2 shows the geometry and the dimensions [ m ] of a
horizontal cut in the region of the entrance of a building. A constant tempera-
ture of 220 C within the building and 60 C outside the building are assumed as
Dirichlet boundary conditions for the temperature field inside the wall. Cauchy
boundary conditions are not taken into account. The different materials are in-
dicated by different gray scale values. λ1 = 0. 08 W/mK is valid for brickwork
and λ2 = 0. 12 W/mK is valid for wood. The discretization with triangular
elements is represented in Figure 8-2 as well.

0.50 1.25

0.125
λ 2 = 0.12 B 0.125
22 o
0.50
6o
A λ 1 = 0.08 0.50
0.25
y

x 0.75 0.25 0.75

Fig. 8-2 Brickwork – Geometry and heat conduction coefficients λ [ W/mK ]

The results for linear shape functions, cf. Figures 8-3, 8-4 and 8-5, and for
quadratic shape functions, cf. Figures 8-6, 8-7 and 8-8, are plotted inside the
elements, whereat the heat fluxes are discontinous at the element intersections.
Figures 8-9, 8-10 and 8-11 represent the results after averaging them at the
nodes of neighboring elements.
The heat conduction process results in a stationary temperature field. The tem-
perature field is nearly linear in regions, where an almost linear heat conduction
without disturbance exists – see cross sections A and B. Thermal bridges de-
velop close to the outer corners, where the boundary conditions dominate the
temperature field inside the wall in two directions.
The heat fluxes qx [ W/m2 ] and qy [ W/m2 ] are correlated to the gradients of
the temperature field. Thus the heat flux is larger in the direction of the shorter
transport distance, cf. qy close to the cross sections A and B, and larger in the
region of the larger heat conduction coefficient – right part of the structure.
150 8 Triangular Elements to Describe Heat Conduction

22. 0

o
C

6. 0

Fig. 8-3 Brickwork, linear shape functions – temperature T (x, y)

5. 8

W/m2

0. 0

Fig. 8-4 Brickwork, linear shape functions – heat flux qx (x, y)

−0. 7

W/m2

−15. 4

Fig. 8-5 Brickwork, linear shape functions – heat flux qy (x, y)

8.2 Example of Use 151

22. 0

o
C

6. 0

Fig. 8-6 Brickwork, quadratic shape functions – temperature T (x, y)

8. 4

W/m2

−1. 2

Fig. 8-7 Brickwork, quadratic shape functions – heat flux qx (x, y)

0. 2

W/m2

−17. 2

Fig. 8-8 Brickwork, quadratic shape functions – heat flux qy (x, y)

152 8 Triangular Elements to Describe Heat Conduction

22. 0

o
C

6. 0

Fig. 8-9 Brickwork, quadratic shape functions – temperature T (x, y)

8. 4

W/m2

−1. 2

Fig. 8-10 Brickwork, quadratic shape functions – averaged heat flux qx (x, y)

0. 2

W/m2

−17. 2

Fig. 8-11 Brickwork, quadratic shape functions – averaged heat flux qy (x, y)
9 Triangular Elements for Membrane Structures

In this section different approaches concerning the choice of the applied coor-
dinate system are represented when employing triangular elements for the de-
scription of plane stress linear elasticity. The procedure to derive triangular
elements for the description of linear elasticity is almost the same as for the
description of stationary heat conduction. Only the transformation of the par-
tial derivatives has to be extended to a vector notation regarding the primary
variables.

Again, basis of the development of the element stiffness matrix and the load
vector is the matrix notation of the work equation
Z Z
−δWel = δvT · { ΩT · DT · E · D · Ω dxdy · v − ΩT · p dxdy} . (9.1)

The description of the geometry, the transformation of coordinates, and the

transformation of derivatives follow Section 7 for the ξ–η–coordinates as well
as for the area coordinates λa , λb , λc . Thus the following sections describe trian-
gular element formulations for both coordinate systems and for linear as well as
quadratic shape functions. Basis of the development of the stiffness matrix and
the load vector is the choice of the approximation of the displacement fields, in
matrix notation
u = Ω·v,
which have to be applied to the shape functions and the coordinate system.

9.1 Linear Shape Functions with Respect to ξ–η–Coordinates

The approach to describe the displacements u(x, y), v(x, y) inside the element
domain is chosen in relation to the local coordinate system. The convergence
criteria need to be fulfilled and checked as with any other type of element. With
regard to membrane elements, at least Co −conform shape functions are to be
employed, which have to be complete shape functions to fulfill the criterion of
coordinate invariance in the case of triangular elements.

9.1.1 Shape Functions and Nodal Unknowns

The linear shape functions employed to describe the displacements are chosen
as shown in Figure 9-1. Such, all shape functions are equivalent concerning the

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_9
154 9 Triangular Elements for Membrane Structures

corner nodes A, B and C. Each linear function describes a plane tilted over one
edge of the element wherein its value is equal to one regarding the opposite
node. As represented in Section 8.1.1, the shape functions and their respective
derivatives are given with:

φA = 1 − ξ − η φB = ξ φC = η
φA ,ξ = −1 φB ,ξ = 1 φC ,ξ = 0
φA ,η = −1 φB ,η = 0 φC ,η = 1

1 1

A B A B A B
1

C C C

Fig. 9-1 Linear shape functions regarding triangular elements

According to the plane stress situation, cf. Section 5.1, the displacements u(x, y)
and v(x, y) have to be approximated, which yields the nodal unknowns v of
linear shape functions

vT = [ uA vA uB vB uC vC ] .

Hence the matrix notation of the displacement fields applies explicitly to

 
uA
 vA 
" # " #



u φA φB φC  uB 
=  .
v φA φB φC 
 vB 

 
 uC 
vC

9.1.2 Differentiation of Displacement Fields

In the work equation, the differentiation rules related to the computation of the
strains from the displacements are comprised of the derivatives with respect to
the x–y–coordinates. Because the shape functions are described in local ξ–η–
coordinates, their derivatives need to be transformed with respect to the local
coordinates. Thus the operator matrix D of the plane stress is transformed
9.1 Linear Shape Functions with Respect to ξ–η–Coordinates 155

according to Section 7.1.1 to

   ∂ξ ∂η 
∂x ∂
∂x ξ
+ ∂
∂x η
   
   ∂ξ ∂η 
D= ∂y  =  ∂
∂y ξ
+ ∂
∂y η 
   
∂y ∂x ∂ξ ∂η ∂ξ ∂η
∂
∂y ξ
+ ∂
∂y η
∂
∂x ξ
+ ∂
∂x η
 
bb ∂ξ + bc ∂η
1  
=  ab ∂ξ + ac ∂η  . (9.2)
det J
ab ∂ξ + ac ∂η bb ∂ξ + bc ∂η

Applying the differentiation rules to the shape functions yields the matrix B,
which is needed to compute the stiffness matrix,

B = D·Ω
      
bb φA ,ξ + bc φA ,η
1       
= ·  ab φA ,ξ + ac φA ,η      
det J
ab φA ,ξ + ac φA ,η bb φA ,ξ + bc φA ,η A B C
      
−bb − bc bb bc
1       
= ·  −ab − ac   ab   ac  
det J
−ab − ac −bb − bc A ab b b B ac b c C
 
ba bb bc
1  
= · aa ab ac . (9.3)
det J 
aa b a ab b b ac b c

The matrix B turns out to be constant. Because the strains and therefore also
the stresses are computed by means of B, they are constant inside the element,
too. Considering the chosen abbreviations for the coordinate differences, B may
be recognized now to be symmetric with respect to all element nodes.

9.1.3 Element Stiffness Matrix

The matrix formulation of the element stiffness matrix K follows analogously
to Section 5
Z Z 1 Z 1−η
K= BT · E · B dA = BT · E · B · det J dξ dη .
A 0 0
156 9 Triangular Elements for Membrane Structures

In addition to the elasticity matrix E, here, B is also constant. When applying

det J = 2A△ , the whole integrand turns out to be constant. Thus in this case,
the element stiffness matrix is performed by matrix multiplication BT · E · B
analogously to the multiplication scheme given in Section 5.1 and finally by
multiplying the product by the factor

1 1
det J · = 2 A△ · = A△ .
2 2

The element stiffness matrix may be summed up by three parts, which can be
assigned to the three different components of the elasticity matrix, as shown in
Section 5.1. With

K = K1 + K2 + K3

this yields
 
ba · ba ba · bb ba · bc

 aa · aa aa · ab aa · ac 

E  bb · ba bb · bb bb · bc 
K1 = · ,
4A△ (1 − ν 2 ) 
 ab · aa ab · ab ab · ac 

 bc · ba bc · bb bc · bc 
ac · aa ac · ab ac · ac
 
b a · aa b a · ab b a · ac
 aa · b a aa · b b aa · b c 
 
νE  b b · aa b b · ab b b · ac 
K2 = · ,
4A△ (1 − ν 2 )  ab · b a
 ab · b b ab · b c 

 b c · aa b c · ab b c · ac 
ac · b a ac · b b ac · b c
 
aa · aa aa · b a aa · ab aa · b b aa · ac aa · b c

 b a · aa ba · ba b a · ab ba · bb b a · ac ba · bc 

E  ab · aa ab · b a ab · ab ab · b b ab · ac ab · b c 
K3 = ·  .
8A△ (1 + ν)  b b · aa bb · ba b b · ab bb · bb b b · ac bb · bc 

 ac · aa ac · b a ac · ab ac · b b ac · ac ac · b c 
b c · aa bc · ba b c · ab bc · bb b c · ac bc · bc

For the same reasons as regarding rectangular elements, the sums over the row’s
components as well as over the column’s components of the element stiffness
matrix must turn out to be identical to zero.
9.1 Linear Shape Functions with Respect to ξ–η–Coordinates 157

9.1.4 Load Vector

At first, the load vector is computed for the special case of constantly distri-
buted external actions, cf. Section 5.1. With
" #
px
p= = constant ,
py
this yields
Z Z
f= ΩT · p dA = ΩT dA · p .
A A
In the case of linear shape functions, the integral can be solved independently
of the chosen coordinate system. The load is therefore equally distributed to
the three nodes by 1/3 each, because the shape functions are equivalent
 
F△
 3 
 F△ 
 3  " #
 F△ 
  px
f = 3 · .
F△



 py
 3 
 F△ 
 3 
F△
3
A fluctuating external action may be described and taken into account ana-
logously to Section 5.1.7, when performing the integration by means of the
matrix of shape functions and the nodal values of the loading.

9.1.5 Subsequent Stress Analysis

Concerning the chosen linear shape functions for the description of the displa-
cements, the matrix of strains B is constant in terms of each element, resulting
in constant stresses. Therefore, the stresses are evaluated only once for the
whole element and can be assigned, for example, to the centre of the element.
At first, this holds in general
σ = E·B·v = S·v.
The stress matrix S follows, employing the chosen linear shape functions and
the abbreviation e = (1 − ν)/2, to
 
ba νaa bb νab bc νac
E
S= E·B= ·  νba aa νbb ab νbc ac  .
2A△ (1 − ν 2 )
eaa eba eab ebb eac ebc
158 9 Triangular Elements for Membrane Structures

9.2 Description Employing Area Coordinates λa , λb , λc

As an alternative, the stiffness matrix can be derived employing area coordi-
nates. In comparison to the local ξ–η–coordinates, area coordinates have the
advantage that the shape functions and thus the element stiffness matrix are
clearly recognized ab initio as being symmetric concerning all element nodes.

9.2.1 Linear Shape Functions Employing Area Coordinates

The shape functions employing area coordinates are symmetric according to the
area coordinates, cf. Figure 9-2, and the comparison with the shape functions
depicted in Figure 9-1 shows that they are of the same shape.
φA = λa φB = λb φC = λc
φA ,λa =1 φB ,λa =0 φC ,λa =0
φA ,λb =0 φB ,λb =1 φC ,λb =0
φA ,λc =0 φB ,λc =0 φC ,λc =1

1 1

A B A B A B
1

C C C

Fig. 9-2 Linear shape functions regarding area coordinates

Analogously to Section 9.1.2 and taking into account Section 7.2.2, the operator
matrix D results in
 P c 
∂λi
  ∂λ
 i=a ∂x i

∂
 x
 c

  P ∂λi 
D=  ∂y  = 
  ∂ 
∂y λi 
 i=a 
∂y ∂x  P c c
P 
∂λi ∂λi
∂
∂y λi
∂
∂x λi
i=a i=a

 Pc 
b ·∂
 i=a i λi 
 c

1  P 
=  ai ·∂λi .
2A△ 
 i=a


 Pc c
P 
ai ·∂λi bi ·∂λi
i=a i=a
9.3 Quadratic Approach Employing ξ–η–Coordinates 159

Therefore the derivative of the shape functions can be represented by

 
ba bb bc
1  
B=D·Ω= · aa ab ac  .
2A△
aa b a ab b b ac b c
The matrix B is constant and of course can be derived identically to Section
9.1.2. Hence the element stiffness matrix K, as well as the element load vector
p and the stress matrix S, are also performed identically to the matrices pre-
sented in Sections 9.1.3, 9.1.4 and 9.1.5. By employing area coordinates, the
integration is performed with respect to the chosen notation as shown expli-
citly in Sections 9.1.3 and 9.1.4. The formulation employing area coordinates
offers the advantage of shape functions being clearly arranged as well as being
equivalent for all nodal displacements.

9.3 Quadratic Approach Employing ξ–η–Coordinates

The previously chosen linear approach to describe the displacements has the
disadvantage of only constantly approximating the stresses inside an element.
A quadratic approach to describe the displacements allows for a linear distri-
bution of strains and stresses. The complete quadratic approach is comprised
of six DOF regarding triangles and, when related to u(x, y) for example, the
mathematical representation holds that

u(x,y) = a0 + a1 x + a2 y + a3 x2 + a4 xy + a5 y 2 ,

which can be checked by means of Pascale’s triangle, see Section 2.4.2. The
displacements of the corner nodes A to C and of the midpoints of the edges D
to F are chosen to represent the DOF of the physically meaningful approach,
cf. Figure 9-3.
x

A ( ξ = 0, η = 0 )
F ( 0, 0.5 )

C ( 0, 1 ) D ( 0.5, 0 )

η
E ( 0.5, 0.5 ) B ( 1, 0 )

y ξ

Fig. 9-3 Nodal coordinates concerning the 6–nodes triangular element

160 9 Triangular Elements for Membrane Structures

9.3.1 Quadratic Shape Functions

At first, the local ξ–η–coordinate system is also chosen for the 6–nodes element,
cf. Figure 9-3. The shape functions are chosen according to the physically
meaningful degrees of freedom and are depicted in Figure 9-4.

φA = (1−ξ−η) · (1−2ξ−2η) φB = ξ (2ξ − 1) φC = η (2η−1)

φA ,ξ = −3 + 4ξ + 4η φB ,ξ = 4ξ − 1 φC ,ξ = 0
φA ,η = −3 + 4ξ + 4η φB ,η = 0 φC ,η = 4η − 1

A 1
C

1
B

1
1/4
1 1/4 F
1 1/4

D
E

φD = 4ξ (1 − ξ − η) φE = 4ξη φF = 4η (1 − ξ − η)
φD ,ξ = 4 − 8ξ − 4η φE ,ξ = 4η φF ,ξ = −4η
φD ,η = −4ξ φE ,η = 4ξ φF ,η = 4 − 4ξ − 8η

Fig. 9-4 Quadratic shape functions concerning triangular elements

The approach of the displacements and analogously of the virtual displacements

is given with
" #
u(ξ, η)
u= = Ω·v,
v(ξ, η)

whereat the nodal displacments v as well as the matrix of shape functions Ω

9.3 Quadratic Approach Employing ξ–η–Coordinates 161

need to be extended to the 6 nodes A to F according to Figure 9-3

" #
φA φB φC φD φE φF
Ω= ,
φA φB φC φD φE φF
 
vT = uA vA uB vB uC vC uD vD uE vE uF vF .

The transformation, the partial derivatives and the notes related to the inte-
gration given in Section 9.1 are also valid here and are independent of the shape
functions. Thus the operator matrix D remains unchanged when compared to
Section 9.1.2. Because the number and particularly the courses of the shape
functions φi in Ω are changed, the differentiation rule D leads to different
derivatives regarding the matrix B:
 
bb φA ,ξ bb φB ,ξ ...
 + bc φA ,η + bc φB ,η ... 
 
 
1 
 a φ
b A ξ, a φ ,
b B ξ . . . 

B=D·Ω= · .
det J  + ac φA ,η + ac φB ,η . . . 
 
 
 ab φA ,ξ bb φA ,ξ ab φB ,ξ bb φB ,ξ . . . 
+ ac φA ,η + bc φA ,η + ac φB ,η + bc φB ,η ...

The matrix B is represented here only concerning the nodes A and B, for to the
nodes C to F the procedure has to be performed analogously. The derivatives
of the shape functions are represented explicitly in Figure 9-4. Since the matrix
B comprises linear functions, the subsequent stress analysis yields linear strains
and stresses inside the element domain.

9.3.2 Element Stiffness Matrix

Regarding the integral to perform the element stiffness matrix
Z 1 Z 1−η Z 1 Z 1−η
K= BT · E · B det J dξdη = BT · E · B · (2A△ ) dξdη ,
0 0 0 0

the integrand BT · E · B is not constant but is of the second order in parts,

since the differentiation rules lower the order of the shape functions all over by
’one’. The multiplication of the matrices equals the scheme given in Figure 5-4
in Section 5.1.6 and leads to a stiffness matrix of the size 12 × 12.
162 9 Triangular Elements for Membrane Structures

As an example the integration is shown concerning the entry K[10,7] of the

element stiffness matrix:

 
... bb φD ,ξ + bc φD ,η ...
1  
B= · ... 0 ...  7th column ,
2A△
... ab φD ,ξ + ac φD ,η ...
 
... bb φD ,ξ + bc φD ,η ...
E  ...
E·B = · ν(bb φD ,ξ + bc φD ,η ) ...  7th column ,
2A△ (1 − ν 2 )
 
1−ν
... 2
(a b φD , ξ + a c φD , η ) . . .
 
.. .. ..
 . . . 
1
BT =
 
· 0 ab φE ,ξ + ac φE ,η bb φE ,ξ + bc φE ,η  10th row .
2A△  
.. .. ..
. . .
The product of the 10th row and the 7th column gives
Z 1Z 1−η
1 E 
K[10,7] = · ν(bb φD ,ξ + bc φD ,η )(ab φE ,ξ + ac φE ,η )
2A△ 2A△ (1 − ν 2 ) 0 0
1−ν 
+ (ab φD ,ξ + ac φD ,η )(bb φE ,ξ + bc φE ,η ) · 2A△ dξdη .
2
When employing the derivatives of the shape functions given in Figure 9-4, it
follows
Z 1Z 1−η
E
K[10,7] = { ν [ bb (4 − 8ξ − 4η) + bc (−4ξ) ] [ ab 4η + ac 4ξ ]
2A△ (1−ν 2 ) 0 0
1−ν
+ [ ab (4 − 8ξ − 4η) + ac (−4ξ) ] [ bb 4η + bc 4ξ ] } dξdη .
2
Concerning integrals employing polynomials of an arbitrary order, the solution
can be analytically obtained by applying
Z 1Z 1−η
i! k!
ξ i · η k dξdη = .
0 0 (i + k + 2)!
Again, the element stiffness matrix K is divided according to the three different
parts of the elasticity matrix

K = K1 + K2 + K3 .
 E
K1 = ·
4A△ (1 − ν 2 )

ba ba
0 aa aa
 

0 bb bb
 

1
 

ab ab . . . symmetric. . .
 1 

0 − 3 ab aa 0
 − bb ba 

bc bc
 3 

3
 

1 1
 
 1 

0 − 3 ac aa 0 − 3 ac ab 0 ac ac
 − bc ba 0 − 1 bc bb 0 

8/3(bbbb +
 3 

4
0 b a b b 0 0 0
 

3 b b +b b )
b a a a
 
 

8/3(a a
b b +
 4 

4 4
0 0 0 0 0
 bb ba 

3 ab aa 3 aa ab a a +a a )
b a a a
 3 
 
9.3 Quadratic Approach Employing ξ–η–Coordinates

8/3(b c b c +
 

4 4 8
0 0 b b 0 b b 0 b b 0
 

3 c b 3 b c 3 a c bc bb +bb bb )
 
 

8/3(ac ac +
 

4 4 8
0 0 0 a c a b 0 a b a c 0 a a a c 0
 

3 3 3 a a +a a )
c b b b
 
 

8/3(b b
c c +
 

4 8 8
0 0 0 0
 

0 3 ba bc 0 3 bb bc 3 ba bb b c b a +b a b a )
 
 

8/3(a c a c +
 4 

4 4 8 8
0 a a 0 0 0 a a 0 a a 0 a a 0
 bc ba 

3 c a 3 a c 3 b c 3 a b ac aa +aa aa )
 3 
 
 
163
 164

νE
K2 = ·
4A△ (1 − ν 2 )

0
0
 

1
 

0 − 3 b b aa 0
 aa b a 

ab b b 0 . . . symmetric. . .
 

−
 

1 1
 
 1 

0 − 3 b c aa 0 − 3 b c ab 0
 3 ab b a 0 

ac b c 0
 
 

4 4
0 b b a a 0 b a a b 0 0 0
 

3 3
 1 

4/3(a a ba +
 − 3 ac ba 0 − 31 ac bb 0 

4
0 0 0 0 0
 

3 aa b b a b +a
b b c bc )
 
 

4/3(aa bc
 4 

4
0 0 0 0 0
 ab b a 

3 b c ab 0 43 bb ac
+ac ba )
 3 
 

4/3(aabc 4/3(aa ba +
 

4
0 0 0 0
 

3 ac b b 0 43 ab bc 0
+a b )
c a ab bb +acbc )
 
 

4/3(ab bc 4/3(ab ba
 

4
0 0 0 0 0 0
 

3 b c aa 0 43 ba ac
+a b )
c b +aa bb )
 
 

4/3(ab bc 4/3(aabb 4/3(aa ba +

 

4
0 0 0 0 0
 

3 ac b a 0 3 aa b c 0
+ac bb ) +ab ba ) ab bb )+ac bc )
 
 
 4 
9 Triangular Elements for Membrane Structures
 E
K3 = ·
8A△ (1 + ν)

aa aa
ba ba
 
 
 b a aa 
 

bb bb . . . symmetric. . .
 1 
 − 3 ab aa − 13 ab ba ab ab 

1 1
 

ac ac
 1 
 − 3 bb aa − 13 bb ba bb ab 
 

b c ac bc bc
 1 
 − 3 ac aa − 3 ac ba − 3 ac ab − 31 ac bb 
 

8/3(ab ab +
 1 

0 0
 − 3 bc aa − 13 bc ba − 31 bc ab − 31 bc bb 

ab aa +aa aa )
 
 

4/3(aa ba + 8/3(bb bb +
 4 

0 0
 3 ab aa 43 ab ba 43 aa ab 43 aa bb 

ab ba +acbc ) bb ba +ba ba )
 
9.3 Quadratic Approach Employing ξ–η–Coordinates

 

4 4 4 4 8 4/3(ac ba 8/3(ac ac +
 4 

0 0
 3 bb aa 43 ba ba 43 ba ab 43 ba bb 

3 ac ab 3 ac b b 3 ab ac 3 ab b c 3 aa ac +aa bc ) ac ab +ab ab )
 
 

4 4 4 4 4/3(a b
a c 8 4/3(aa ba + 8/3(bcbc +
 

0 0
 

3 b c ab 3 bc bb 3 b b ac 3 b b b c +ac ba ) 3 ba bc ab bb +ac bc ) bc bb +bb bb )

 
 

4 4 8 4/3(acbb 8 4/3(aa bb 8/3(acac +

 

0 0
 

3 aa ac 3 aa b c 3 ab ac +ab bc ) 3 aa ab +ab ba ) ac aa +aa aa )

 
 
 4 

4 4 4 4/3(ab bc 8 4/3(ab ba 8 4/3(aa ba + (bc bc +

 3 ac aa 43 ac ba 

3 b c aa 3 bc ba 0 0 3 b a ac 3 b a b c 3 bb bc 3 ba bb
 

+ac bb ) +aa bb ) ab bb +ac bc ) bc ba +ba ba )

 
165

 4 
166 9 Triangular Elements for Membrane Structures

9.4 Quadratic Shape Functions Using Area Coordinates

Following Section 7.2, every point of the triangle is defined by the area coordi-
nates λa , λb and λc . In particular, Figure 9-5 shows the values related to the
six nodes of the element.
x

A ( λ a = 1, λ b = 0, λ c = 0 )
F ( 0.5, 0, 0.5 )

D ( 0.5, 0.5, 0 )
C ( 0, 0, 1)

y E ( 0, 0.5, 0.5 ) B ( 0, 1, 0 )

Fig. 9-5 The area coordinates of a 6–node triangular element

Employing area coordinates the quadratic shape functions to describe the dis-
placements given in Figure 9-4 and their respective derivatives follow to

φA = λa (2λa −1) , φA,λa = 4λa −1 , φA,λb = 0 , φA,λc = 0 ,

φB = λb (2λb −1) , φB,λa = 0 , φB,λb = 4λb −1 , φB,λc = 0 ,
φC = λc (2λc −1) , φC,λa = 0 , φC,λb = 0 , φC,λc = 4λc −1 ,
φD = 4 λa λb , φD,λa = 4 λb , φD,λb = 4 λa , φD,λc = 0 ,
φE = 4 λb λc , φE,λa = 0 , φE,λb = 4 λc , φE,λc = 4 λb ,
φF = 4 λa λc , φF,λc = 4 λc , φF,λb = 0 , φF,λa = 4 λa .
Here again, the shape functions and their derivatives turn out to be symmetric
according to the respective area coordinate.

9.4.1 Stiffness Matrix

Following Section 9.2.1, the operator matrix D can be put up to
 Pc 
bi ·∂λi
 i=a 
 c

1   P 
D= a i ·∂λi  .

2A△   i=a 
 Pc c
P 
ai ·∂λi bi ·∂λi
i=a i=a
9.4 Quadratic Shape Functions Using Area Coordinates 167

In this way, the differentiation of the shape functions yields

B= D·Ω

b (4λa −1) bb (4λb −1) bc (4λc −1)
 a
1 
= aa (4λa −1) ab (4λb −1) ac (4λc −1)
2A△ 

aa (4λa −1) ba (4λa −1) ab (4λb −1) bb (4λb −1) ac (4λc −1) bc (4λc −1)

4ba λb 4bb λc 4ba λc
+4bb λa +4bcλb +4bc λa 



4aa λb 4ab λc 4aa λc .
+4ab λa +4ac λb +4ac λa 


4aa λb 4ba λb 4ab λc 4bb λc 4aa λc 4ba λc 
+4ab λa +4bb λa +4ac λb +4bc λb +4ac λa +4bc λa
When applying the matrix B, the integration of the element stiffness matrix
can be performed again following this repeated scheme:
ZZ
K= BT · E · B · (2A△ ) dλa dλb .

The integration of the individual terms follows the integration rule related to
the product terms of the area coordinates as shown in Section 7.2.4
ZZ
i! j! k!
λia · λjb · λkc dλa dλb = .
(i + j + k + 2)!
Thus the evaluation of the entry K[10,7] regarding the stiffness matrix yields
ZZ
E
K[10,7] = [ ν (4ba λb + 4bb λa )(4ab λc + 4ac λb )
2A△ (1−ν 2 )
1−ν
+ (4aa λb + 4ab λa )(4bb λc + 4bc λb ) ] · dλa dλb
2
νE 1 2 1 1
= (16ba ab + 16ba ac + 16bb ab + 16bb ac )
2A△ (1−ν 2 ) 24 24 24 24
E 1 2 1 1
+ (16aa bb + 16aa bc + 16ab bb + 16ab bc )
4A△ (1+ν) 24 24 24 24
νE E
= 2
(aa bc + ac ba ) + (aa bc + ac ba ) .
3A△ (1−ν ) 6A△ (1+ν)
Of course, the stiffness matrix K equals the stiffness matrix employing the
ξ–η–coordinates derived in Section 9.3.1.
168 9 Triangular Elements for Membrane Structures

9.4.2 Load Vector

Here, the computation of the load vector is performed by employing area coor-
dinates and considering the special case of a constantly distributed external
action
Z Z ZZ
f = ΩT · p dA = ΩT dA · p = 2A△ ΩT dλa dλb · p .

Because of the nodal symmetry, the integration of the shape functions is to be

performed for only one corner node and one midpoint node of the respective
edge. For node A, this holds
ZZ ZZ  
2 2 1
2A△ φA dλa dλb = 2A△ (2λa − λa ) dλa dλb = 2A△ 2 · − = 0,
24 6
and for node D, it yields
ZZ ZZ  
1 A△
2A△ φD dλa dλb = 2A△ (4λa λb ) dλa dλb = 2A△ 4 · = .
24 3
Thus the corner nodes A, B and C do not take any amount of the external
action defined above, and only the midpoints of the edges D, E and F take
A△ /3 each.

9.4.3 Subsequent Stress Analysis

Employing quadratic shape functions to describe the displacements in a tri-
angular element results in linearly distributed strains, because the previously
computed matrix B fluctuates linearly. Therefore, the stresses in the element
are also linearly distributed. It gives sense to only compute the stresses related
to the corner nodes A, B and C to be able to describe a linear course in the
element. Thus the stress matrix S̃ of the corner nodes is given by

S̃ = E · B̃ .

The coordinates related to the nodes A, B and C need to be introduced into the
matrix B, as given in Section 9.4.1. Applying the abbreviation e = (1 − ν)/2,
the stress matrix explicitly follows to
 σxxa 3ba 3νaa −bb −νab −bc −νac 4bb 4νab 0 0 4bc 4νac
σyya 3νb a 3aa 4νbb 4ab 0 0 4νbc
   

−νbb −ab −νbc −ac

σxya 3eba 4eab 4ebb 0 0 4eac
   

−eab −ebb −eac −ebc

   4ac 
   

0
   3eaa 4ebc 

σxxb 3bb 3νab 4ba 4νaa 4bc 4νac

   

−νaa −bc −νac

E
   

σyyb 3νbb 3ab 4νba 4aa 4νbc 4ac 0

   

−aa −νbc −ac

   −ba 0 

σxyb 3eab 3ebb 4eaa 4eba 4eac 4ebc 0

   
   

−eba −eac −ebc

 =  −νba 0 ·v .
  2A△ (1 − ν 2 )  

σxxc 3bc 3νac 0 0 4bb 4νab 4ba

   −eaa 0 
   

−νaa −bb −νab

   

σyyc 3νbc 3ac 0 0 4νbb 4ab 4νba

   

−aa −νbb −ab

   −ba 4νaa 
9.4 Quadratic Shape Functions Using Area Coordinates

σxyc 0 0 4eab 4ebb 4eaa 4eba

   
  
−eaa −eba −eab −ebb −3eac −3ebc

   −νba 4aa 
169
170 9 Triangular Elements for Membrane Structures

9.5 A Comparison of Standard Elements

The cantilever–like structure depicted in Figure 9-6 is investigated employing
linear and quadratic shape functions respectively related each to triangular and
rectangular elements. Because of the antisymmetry of the problem, u = 0 is
to be predefined with respect to the center line. Therefore, only the upper half
of the structure needs to be discretized. The external action at the right end
of the structure is defined as a constantly distributed boundary action. The
following element types are to be compared:

1. The triangular element employing a linear approach,

2. the triangular element employing a quadratic approach,
3. the rectangular element employing a linear approach but performing se-
lectively reduced integration – SRI),
4. the rectangular element employing a quadratic approach related to 9 no-
des, which is introduced in Section 11.2.1.

x = l/2

x, u PY
l B
3
y, v

l
l = 15 m, t = 1 m, E = 100 000 N/m2, ν = 0, PY = 2 N/m2

Fig. 9-6 Cantilever – geometry and loading

Investigations are performed applying two different discretizations regarding

each element type. With regard to the coarser mesh, the upper half of the
structure is discretized by means of 39 nodes (78 unknowns), which yields either
two rows with 12 rectangular elements each employing a linear approach, or one
row with six rectangular elements employing a quadratic approach, see Figure
9-7. Concerning a discretization employing triangular elements, the respective
rectangular element is divided by a diagonal rising from left to right.
Regarding the more refined mesh with 250 unknowns, the number of elements
in the x– and y–directions are each doubled. When evaluating the accuracy of
the results, both the number of unknowns of the system of equations and the
9.5 A Comparison of Standard Elements 171

bandwidth should be taken into account. As stated previously in Section 3, the

computing time needed to solve the system of equations increases depending
on the square of the bandwidth.

1 4 7 10 13 16 19 21 24 27 30 34 37

2 5 8 11 14 17 20 22 25 28 31 35 38

3 6 9 12 15 18 21 23 26 29 32 36 39

Fig. 9-7 Cantilever – meshes employing triangular and rectangular elements

The vertical displacement at the end of the cantilever is normalized by the

external action as well as by Young’s modulus, and is compared in Table 9.1
with respect to the different element types. It is obvious that higher order
shape functions lead to better results when requiring approximately the same
numerical effort. The selectively reduced integration (SRI) also leads to better
results than expected.

Table 9.1 Cantilever – displacement vB for different element types and meshes

Elements Unknowns Bandwidth vB · E/P

1. Triangle 78 8 94. 9
linear approach 250 12 109. 2
2. Triangle 78 14 114. 8
quadratic approach 250 22 115. 0
3. a Rectangle 78 10 111. 2
bi–linear approach 250 14 114. 0
3. b Rectangle 78 10 114. 5
bi–linear approach – SRI 250 14 114. 9
4. Rectangle 78 18 114. 9
quadratic approach 250 26 115. 0
172 9 Triangular Elements for Membrane Structures

The improvement of the results by employing quadratic shape functions as well

as by performing selectively reduced integration is especially obvious with re-
spect to the stresses. In the following, the stresses to the left as well as to the
right of the cut in x = ℓ/2 are displayed for ℓ = 15 m, Py = 2 N/m2 and
for the more refined discretization. The results regarding triangular elements
employing the linear as well as the quadratic approach are depicted on the
left, while the results regarding rectangular elements employing the linear ap-
proach as well as the quadratic approach combined with the selectively reduced
integration scheme are depicted on the right.

σxx

σxy

Fig. 9-8 Cantilever – stress distributions for different element types at x = 7. 50 m

10 Triangular Elements for Kirchhoff Plates

The governing equations as well as the work equation concerning the Principle
of virtual Displacements are given in Section 6.2 regarding Kirchhoff ’s plate
theory. They are independent of the element formulation, so that they can
be adopted unchanged when considering triangular elements. The coordinate
system for the description of the triangular elements for plates can be taken to
be the same as the one introduced in Section 7.
Because of the requirements necessitated by C1 –conformity and the demand of
coordinate invariance, it becomes much more difficult to choose shape functions
and nodal unknowns concerning the triangular elements for Kirchhoff ’s plate
theory when compared to the membrane elements.

10.1 Choice of Shape Functions and Nodal Unknowns

The required order n of the shape functions for the description of the deflection
surface w depends on the convergence criteria to be fulfilled, cf. Section 2.4.2:

• Constant deformations – curvature or twisting – require at least quadra-

tic shape functions for the bending surface w(x, y).
• Among others the coordinate invariance requires all comparable element
nodes to employ the same degrees of freedom. Therefore the DOF of an
element node must be clearly transformable. Thus in addition to [ w ],
two first order derivatives [ w,x w,y ] and possibly three second order de-
rivatives [ w,xx w,yy w,xy ] must be available at each element node.
• C1 –conformity requests the deflection inside the element domain to be
described by a complete polynomial of nth–order. The chosen bending
surface w(x, y) must result in a continuous polynomial of nth–order at
the element intersections to all neighboring elements, and the same must
hold for the slopes w,s (s) and w,n (s) with polynomials of the (n − 1)th–
order along the local intersection coordinate s.
This means that at all element intersections, (n + 1) nodal degrees of
freedom are needed for the description of w(s) and w,s (s) and additional
n independent degrees of freedom for the description of w,n are to be
defined. Thus each element edge requires (2n + 1) independent DOF. If
n DOF are chosen for each element corner, an additional DOF is needed
in the centre of each edge.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_10
174 10 Triangular Elements for Kirchhoff Plates

The C1 –conformity is achievable, but requires high effort and provides con-
forming elements with restricted applications. Thus, in the following, various
shape functions with ascending polynomial order are investigated, whereat li-
near shape functions are not able to describe constant curvatures, since these
are 2nd order derivatives of the bending surface w(x, y).
A complete quadratic approach (n = 2) employs 6 degrees of freedom and has
been applied first by Morley [73]. As shown in Figure 10-1, there are 3 DOF
per element edge, since [ w ] is chosen in the corner nodes and the slope [ w,n ]
is considered in the centre of the edges. Although 2 · 2 + 1 = 5 DOF are
formally needed at each edge, the solution converges towards the correct value
in the case of uniform meshes. Because of this restriction, the element cannot
be generally applied. The convergence is poor because the curvature is only
constantly approached inside the element domain.
n
w
s
A
w,n w,n
F D

C E B
w w
w,n

Fig. 10-1 Triangular plate element with non–conforming quadratic shape functions

Complete cubic shape functions (n = 3) employ 10 degrees of freedom. There-

fore, 2 · 3 + 1 = 7 DOF are required at each element edge. As shown in Figure
10-2, the DOF [ w w,x w,y ] are chosen in the corner nodes and [ w ] in the
triangle’s centre of gravity, thus [ w ] is approximated as a cubic polynomial
on the element boundary, and subsequently w,s is approached quadratically.
w,n is, at most, linearly describable, concerning the two remaining degrees of
freedom per edge. Thus a unique transformation of the slope is not possible.
w w,x w,y
A

M
C B
w w,x w,y w w,x w,y

Fig. 10-2 Triangular plate element with non–conforming cubic shape functions
10.1 Choice of Shape Functions and Nodal Unknowns 175

Complete shape functions of the 4th order employ 15 degrees of freedom. There-
fore, 2 · 4 + 1 = 9 DOF are required at each element edge. If [ w w,x w,y ] are
chosen in the corner nodes and [ w w,n ] in the centres of the edges, as shown in
Figure 10-3, a maximum of 8 degrees of freedom is obtained. Thus w could be
described as a polynomial of the 4th order along the element boundary and thus
w,s would subsequently be approached as a 3rd order polynomial. The nodal
degrees of freedom w,n relating to the corners and to the centres of the edges,
enable, at most, a description of the slope w,n with a quadratic polynomial.
n
w w, x w, y
s
A
w w, n w w, n
F D

C E B
w w, x w, y w w, x w, y
w w, n
Fig. 10-3 Triangular plate element with non–conforming 4th order shape functions

Complete shape functions of the 5th order employ 21 degrees of freedom. Thus
at each element edge 2 · 5 + 1 = 11 nodal unknowns are needed to satisfy the
compatibility conditions. If [ w w,x w,y w,xx w,xy w,yy ] are chosen as physically
meaningful unknowns in the corner nodes as well as [ w,n ] in the centres of the
edges, 13 DOF are obtained per element boundary and therefore all convergence
criteria can be satisfied.
n
A v={ [ w w,x w,y w,xx w,xy w,yy ]A
s
[ w w,x w,y w,xx w,xy w,yy ]B
F [ w w,x w,y w,xx w,xy w,yy ]C
D

C E [ w,n ]d [ w,n ]e [ w,n ]f }

Fig. 10-4 Triangular plate element with conforming 5th order shape functions

Despite of being a valuable approximation of the bending surface, the possibili-

ties of an application are limited. At first, the displacement boundary conditions
can not be easily fulfilled in all cases when considering the 2nd order deriva-
tives of w. Further, the element will not fulfill the moment equilibrium at the
176 10 Triangular Elements for Kirchhoff Plates

element intersections, if the plate bending stiffness changes between neighbo-

ring elements. Nevertheless, triangular elements with complete shape functions
of the 5th order are introduced in detail to explain the general procedure for
coping with triangular elements.

Shape functions of an even higher order can be chosen, but are not efficient
as the effort needed to compute the stiffness matrix increases sharply and,
moreover, the corresponding choice of unknowns creates further difficulties.
Because of the difficulties in the fulfillment of the C1 –conformity concerning
Kirchhoff ’s plate theory, triangular plate elements have been developed, which
go beyond the standard displacement–based formulation of the Principle of
virtual Displacements, cf. Sections 17, 18, 19.

10.2 Complete Approach of the 5th Order

The triangular plate element with a complete 5th order approach was originally
investigated by Bosshard [21] and Argyris, Fried and Scharpf [5]. The complete
5th order approach employs 21 degrees of freedom, cf. Figure 10-4. Therefore,
the bending surface w(x, y) may be described as a complete polynomial of the
5th order depending on the coordinate s along the element edge. The course
of w(s) is defined as a Hermite Polynomial of the 5th order with the degrees
of freedom [ w w,s w,ss ], transformed to the boundary coordinates of both
corner nodes of the respective edge. Thus w,s is automatically approached as
a polynomial of the 4th order.
Subsequently, the slope w,n is also a combination of the slopes w,x and w,y .
Therefore, it is described as a polynomial of 4th order in s, too. Regarding
w,n and w,ns in the corner nodes and w,n in the centres of the edges, this
polynomial is also complete. Thus the requirements of the element intersection
conditions are fulfilled for w, w,s and w,n and therefore the conformity criterion
for Kirchhoff ’s plate theory is satisfied as well. The additional unknowns w,nn
take care of the unique transformation of the curvatures in the corners.

10.2.1 The Process to Assemble the Element Stiffness Matrix

In order to develop the element stiffness matrix, it is advantageous to utilize the
symmetry of the area coordinates, cf. Section 7.2. Additionally, it is difficult
to directly assign shape functions according to every individual unknown in
the case of a 5th order polynomial, as it has been done in Section 6.2 for the
respective rectangular element. Hence the general polynomial is chosen here
as a starting point for the formulation, which is also numerically much more
10.2 Complete Approach of the 5th Order 177

efficient. The deflection in the element is approached to

w = ψ · a. (10.1)

The vector ψ contains the monomials of the chosen general polynomial in area
coordinates. The vector a contains the coefficients a1 −a21 of the polynomial. At
first the relationship between the coefficients ai and the physically meaningful
nodal unknowns v is to be formulated. This results in

v = Ψ̃ · a . (10.2)

The inverse matrix of Ψ̃ defines the scaling matrix G. Thus

a= G·v (10.3)

is obtained. The vector of the physically meaningful shape functions Ω is now

obtained by employing a in Equation (10.1)

w = ψ ·G·v = Ω·v. (10.4)

Nevertheless, the computation of the element stiffness matrix, following Section

10.2.5, applies the general polynomial here, since this procedure has substantial
numerical advantages.

10.2.2 General Polynomial Employing Area Coordinates

The general polynomial approach for the deflection is

w (λa , λb , λc ) = ψ · a . (10.5)

For an approach of the 5th order, and following the definition of area coordi-
nates given in Section 7.2, this yields

ψ = [ λ5a λ4a λb λ4a λc λ3a λ2b λ3a λb λc λ3a λ2c λ2b λc λ2a
λ5b λ4b λc λ4b λa λ3b λ2c λ3b λa λc λ3b λ2a λ2c λa λ2b (10.6)
λ5c λ4c λa λ4c λb λ3c λ2a λ3c λa λb λ3c λ2b λ2a λb λ2c ].

To derive the scaling matrix, the derivatives of the deflection surface must be
evaluated with respect to the area coordinates.
178 10 Triangular Elements for Kirchhoff Plates

This results in

w,λa = [ 5λ4a 4λ3a λb 4λ3a λc 3λ2a λ2b 3λ2a λb λc 3λ2a λ2c 2λ2b λc λa
0 0 λ4b 0 λ3b λc 2λ3b λa λ2c λ2b
0 λ4c 0 3
2λc λa 3
λc λb 0 2λa λb λ2c ] · a ,

w,λa λa = [ 20λ3a 12λ2a λb 12λ2a λc 6λa λ2b 6λa λb λc 6λa λ2c 2λ2b λc
0 0 0 0 0 2λ3b 0
0 0 0 2λ3c 0 0 2λb λ2c ] · a,

w,λa λb = [ 0 4λ3a 0 6λ2a λb 3λ2a λc 0 4λb λc λa

0 0 4λ3b 0 3λ2b λc 6λb λa 2λ2c λb
2

0 0 0 0 λ3c 0 2λa λ2c ] · a ,

w,λa λc = [ 0 0 4λ3a 0 3λ2a λb 6λ2a λc 2λ2b λa

0 0 0 0 λ3b 0 2λc λ2b
0 4λ3c 0 6λ2c λa 3λ2c λb 0 4λa λb λc ] · a ,

w,λb =[ 0 λ4a 0 2λ3a λb λ3a λc 0 2λb λc λ2a

5λ4b 4λ3b λc 4λ3b λa 3λ2b λ2c 3λ2b λa λc 3λ2b λ2a 2λ2c λa λb
0 0 λ4c 0 λ3c λa 2λ3c λb λ2a λ2c ] · a,

w,λb λb = [ 0 0 0 2λ3a 0 0 2λc λ2a

20λb 12λb λc 12λb λa 6λb λ2c 6λb λa λc 6λb λ2a
3 2 2
2λ2c λa
0 0 0 0 0 2λ3c 0 ] · a,

w,λb λc = [ 0 0 0 0 λ3a 0 2λb λ2a

0 4λ3b 0 6λ2b λc 3λ2b λa 0 4λc λa λb
0 0 4λ3c 0 3λ2c λa 6λ2c λb 2λ2a λc ] · a ,

w,λc =[ 0 0 λ4a 0 λ3a λb 2λ3a λc λ2b λ2a

0 λ4b 0 2λ3b λc λ3b λa 0 2λc λa λ2b
5λ4c 4λ3c λa 4λ3c λb 3λc λa 3λc λa λb 3λc λb 2λ2a λb λc ] · a ,
2 2 2 2 2

w,λc λc = [ 0 0 0 0 0 2λ3a 0
3
0 0 0 2λb 0 0 2λa λ2b
20λ3c 12λ2c λa 12λ2c λb 6λc λ2a 6λc λa λb 6λc λ2b 2λ2a λb ] · a.
10.2 Complete Approach of the 5th Order 179

10.2.3 Derivatives with Respect to x–y–Coordinates

The transformation of the 1st order derivatives of the deflection with respect to
the area coordinates λi to the respective derivatives with respect to coordinates
x and y follows Section 7.2.2. Analogously the complete transformation of the
1st as well as the 2nd order derivatives is provided by the following scheme in
matrix notation
  
∂w ∂λa ∂λb ∂λc
0 0 0
 ∂x   ∂x ∂x ∂x
  
 ∂w   ∂λa ∂λb ∂λc
 ∂y   ∂y ∂y ∂y
0 0 0
  
 2   2
 ∂ w   ∂ λa ∂ 2 λb ∂ 2 λc ( ∂λa )2 ( ∂λb )2 ( ∂λc )2
 ∂x2  =  ∂x2 ∂x2 ∂x2 ∂x ∂x ∂x
  
 2   2 2 2
 ∂ w   ∂ λa ∂ λb ∂ λc ( ∂λa )2 ( ∂λb )2 ( ∂λc )2
 ∂y2   ∂y2 ∂y 2 ∂y 2 ∂y ∂y ∂y
  
 ∂2w   2 2 2
∂ λa ∂ λb ∂ λc ∂λa ∂λa ∂λb ∂λb ∂λc ∂λc
∂x∂y ∂x∂y ∂x∂y ∂x∂y ∂x ∂y ∂x ∂y ∂x ∂y
 
∂w
 ∂λa 
 ∂w

 
  ∂λb 
 
0 0 0  ∂w 
  ∂λc 
  
0 0 0   ∂2w 
  ∂λ2a

  
∂λb ∂λb ∂λc   ∂2w

2 ∂λa
2 ∂λa ∂λc
2 · 
.
∂x ∂x ∂x ∂x ∂x ∂x   ∂λ2b
  
∂λb ∂λb ∂λc   ∂2w 
2 ∂λ
∂y
a
∂y
2 ∂λ a ∂λc
∂y ∂y
2 ∂y ∂y   ∂λ2c

  
  ∂2w

∂λa ∂λb ∂λa ∂λb ∂λa ∂λc ∂λa ∂λc ∂λb ∂λc ∂λb ∂λc  
∂x ∂y
+ ∂y ∂x ∂x ∂y
+ ∂y ∂x ∂x ∂y
+ ∂y ∂x  ∂λa λb 
 
 ∂2w 
 ∂λa λc

 
 ∂2w 
∂λb λc

The first and the second row describe the transformation of the 1st order deri-
vatives as developed in Section 7.2.2. The rows three, four and five describe the
transformation of the 2nd order derivatives, whereat the 2nd order derivatives
of the area coordinates with respect to the x–y–coordinates occur, which de-
pend on the element geometry. In the case of a linear transformation as taken
into account at hand and given in Section 7.2.1 the 2nd order derivatives of the
area coordinates disappear and lead to a simplification of the scheme as follows.
After introducing the abbreviations for the derivatives of the area coordinates,
180 10 Triangular Elements for Kirchhoff Plates

following Section 7.2.2, this yields

 
w,x
 w,y 
 
= 1 ·
 

 w,xx  2A∆

 w,yy 

w,xy
 
w,λa
 w, 
 λb 
  
ba bb bc  w,λc 
 
a a a  w, 
 a b c  λa λa 
 2 2 2  
 ba bb bc 2 aa ab 2 aa ac 2 ab ac  w,λb λb . (10.7)
  

 a2a a2b a2c 2 ba bb 2 ba bc 2 bb bc  w,λc λc 
 
 
ba aa bb ab bc ac ba ab +aa bb ba ac +aa bc bb ac +ab bc  w,λa λb 
 
 w, 
 λa λc 
w,λb λc

10.2.4 Scaling Matrix

The computation of the scaling matrix Ψ̃ is separately performed row by row
for each nodal unknown, as the corresponding nodal coordinates are introdu-
ced into the approach and its derivatives respectively, following Section 10.2.2.
Such, the deflection in node A is computed as follows

w|A = w (λa = 1, λb = 0, λc = 0)
= [ 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ] · a.

Accordingly, the slope of the deflection surface in x–direction, see Equation

(10.7), follows to

1
w,x = · [ ba · w,λa + bb · w,λb + bc · w,λc ] ,
2A∆

and, introducing the derivatives of the approach, is evaluated for node A ap-
10.2 Complete Approach of the 5th Order 181

plying the coordinates (λa = 1, λb = 0, λc = 0), to

ba
w,x|A = ·[ 5 0 0 00 00 0 00 00 00 0 00 00 00 ]·a
2A∆
bb
+ ·[ 0 1 0 00 00 0 00 00 00 0 00 00 00 ]·a
2A∆
bc
+ ·[ 0 0 1 00 00 0 00 00 00 0 00 00 00 ]·a
2A∆
1
= · [ 5ba bb bc 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ] · a.
2A∆
This procedure is transferred to the other unknowns with respect to the corner
nodes, which addresses the first 18 DOF of the vector v.

The scaling of the approach with respect to the slopes w,n in the centres of the
edges is more time–consuming, as the derivatives are defined with respect to
the global x–y–coordinates, but these must be transformed with respect to the
local element boundary coordinates. To identically define the unknowns w,n
in neighboring elements at assembling the system, the direction of the element
boundary coordinates n, s should be defined globally. The positive direction
of the normal with respect to the edge of the triangular element considered,
will be defined such that it encloses an angle of 0 ≤ β < π with the x–axis, as
shown in Figure 10-5. The angle between the normal and the x–axis is explicitly
evaluable when considering the differences in the coordinates of the respective
corner nodes.
x

A π yb − ya bc
tan( βd − )= =−
2 xb − xa ac
F D π yc − yb ba
βf tan( βe − )= =−
2 xc − xb aa
E βd
C π ya − yc bb
βe B tan( βf − )= =−
2 xa − xc ab
y

Fig. 10-5 The definiton of the positive normal direction

The transformation of the slopes w,x and w,y into the slopes w,s and w,n can
be performed with respect to the element boundary, as shown in Figure 10-6,
182 10 Triangular Elements for Kirchhoff Plates

by

π π
w,s = + cos( β − ) · w,x + sin( β − ) · w,y ,
2 2
π π
w,n = − sin( β − ) · w,x + cos( β − ) · w,y .
2 2

x
w,s
w, n
D
s w,y s
(β − π)
2
td td
y
nd w,x nd

Fig. 10-6 Definition of w,n and w,s according to w,x and w,y

By means of the derivatives w,x and w,y the last three rows of the matrix Ψ̃
can be computed as follows. Due to the transformation of coordinates the slope
w,n|D satisfies

w,n|D =
π 1
− sin( β − )· [ ba · w,λa + bb · w,λb + bc · w,λc ](λa = 21 , λb = 12 , λc =0)
2 2A∆
π 1
+ cos( β − ) · [ aa · w,λa + ab · w,λb + ac · w,λc ](λa = 21 , λb = 12 , λc =0)
2 2A∆

and may be evaluated analogously to w,x|A . The same transformation leads to

the slopes w,n|E and w,n|F .

The entire Ψ̃–matrix employs 21 rows and columns and links the coefficients of
the general polynomial with the physically meaningful nodal degrees of freedom

vT = { [ w w,x w,y w,xx w,xy w,yy ]A

[ w w,x w,y w,xx w,xy w,yy ]B
(10.8)
[ w w,x w,y w,xx w,xy w,yy ]C
[ [ w,n ]D [ w,n ]E [ w,n ]F ] }.
10.2 Complete Approach of the 5th Order 183

The asymmetric matrix Ψ̃ is structured as follows

 
Ψ̃11 0 0
 
 0 Ψ̃22 0 
Ψ̃ =   0
. (10.9)
 0 Ψ̃33 

← Ψ̃n →
The sub–matrices at the main diagonal Ψ̃ii are rectangular matrices of the size
6 × 7. The factors positioned behind the Ψ̃ii –matrices are arranged for each
respective row to be multiplied by that factor:
 
1 0 0 0 0 0 0
  1
 5b bb bc 0 0 0 0 
 a  · 2A∆
  1
 ·
 5aa a b a c 0 0 0 0  2A
  ∆
Ψ̃11 =   1
 20b2 8ba bb 8ba bc 2bb2
2bb bc 2bc2
0  4A2
 ·
 a
  1∆
 
 20aa ba 4(aa bb +ab ba ) 4(aa bc +ac ba ) 2ab bb ab bc +ac bb 2ac bc 0  · 4A2
  ∆
1
2 2 2
20aa 8aa ab 8aa ac 2ab 2ab ac 2ac 0 ·
4A2∆
 
1 0 0 0 0 0 0
  1
 5b bc ba 0 0 0 0 
 b  · 2A∆
  1
 ·
 5ab a c a a 0 0 0 0  2A
  ∆
Ψ̃22 =   1
 20b2 8bb bc 8ba bb 2bc2
2ba bc 2
2ba 0  4A2
 ·
 b
  1∆
 
 20ab bb 4(ac bb +ab bc ) 4(ab ba +aa bb ) 2ac bc ac ba +aa bc 2aa ba 0  · 4A2
  ∆
1
20a2b 8ab ac 8aa ab 2a2c 2aa ac 2a2a 0 · 4A2
∆
 
1 0 0 0 0 0 0
  1
 5b ba bb 0 0 0 0 
 c  · 2A∆
  1
 ·
 5ac a a a b 0 0 0 0  2A
  ∆
Ψ̃33 =   1
 20b2 8ba bc 8bb bc 2ba2
2ba bb 2bb2
0  · 4A2

 c
  1∆
 
 20ac bc 4(ac ba +aa bc ) 4(ac bb +ab bc ) 2aa ba ab ba +aa bb 2ab bb 0  · 4A2
  ∆
1
20a2c 8aa ac 8ab ac 2a2a 2aa ab 2a2b 0 · 4A2
∆
184 10 Triangular Elements for Kirchhoff Plates

The Ψ̃n –matrix is completely taken and has 3 rows and 21 columns. The in-
troduction of the abbreviations
π π 1
a = {−ba · sin(βd − ) + aa · cos(βd − )} · ,
2 2 32A∆
π π 1
b = {−bb · sin(βe − ) + ab · cos(βe − )} · ,
2 2 32A∆
π π 1
c = {−bc · sin(βf − ) + ac · cos(βf − )} · ,
2 2 32A∆
and consideration of the derivatives given in Section 10.2.2 yields the rows 19,
20 and 21:

5a 4a + b c 3a + 2b c 0 c

Ψ̃n =  0 0 0 0 0 0 0
5a b c + 4a 0 b 2c + 3a 0
5b c a + 4b 0 c 2a + 3b 0
5b 4b + c a 3b + 2c a 0 a
0 0 0 0 0 0 0

0 0 0 0 0 0 0

5c a b + 4c 0 a 2b + 3c 0  .
5c 4c + a b 3c + 2a b 0 b

To get the scaling matrix, the inversion

−1
G = Ψ̃

can be computed numerically for each element in the FE–program when apply-
ing its respective coordinates. Here, the scaling matrix is also explicitly specified
with
 
G11 0 0 0
 0 G22 0 0 
 
 
 0 0 G 33 0 
G= .
 ← g7 → 
 
 ← g → 
14
← g21 →

Although the vectors gi are computed as rows 7, 14, 21 of G they are embed-
ded in the scaling matrix as rows 19, 20, 21 for simplicity. By employing the
10.2 Complete Approach of the 5th Order 185

abbreviations a, b and c the vectors gi are determined to

 
g7 · c
 
 g14 · a  =
g21 · b

−5(11a + 5b + 5c) −ac (16a + 9b + 4c) + 5cab bc (16a + 9b + 4c) − 5cbb

 0 0 0
−5(11a + 5c + 5b) ab (16a + 9c + 4b) − 5bac −bb (16a + 9c + 4b) + 5bbc

cab ac −a2c (1. 5a + b) bc ac (3a + 2b)−c(ab bc + ac bb ) cbb bc − b2c (1. 5a + b)

0 0 0
bab ac −a2b (1. 5a + c) bb ab (3a + 2c)−b(ab bc + ac bb ) bbb bc − b2b (1. 5a + c)

−5(11b + 5a + 5c) ac (16b + 9a + 4c) − 5caa −bc (16b + 9a + 4c) + 5cba

−5(11b + 5c + 5a) −aa (16b + 9c + 4a) + 5aac ba (16b + 9c + 4a) − 5abc
0 0 0

caa ac −a2c (1. 5b + a) bc ac (3b + 2a)−c(aa bc + ac ba ) cbc ba − b2c (1. 5b + a)

aaa ac −a2a (1. 5b + c) ba aa (3b + 2c)−a(acba + aa bc ) aba bc − b2a (1. 5b + c)
0 0 0

0 0 0
−5(11c + 5b + 5a) aa (16c + 9b + 4a) − 5aab −ba (16c + 9b + 4a) + 5abb
−5(11c + 5a + 5b) −ab (16c + 9a + 4b) + 5baa bb (16c + 9a + 4b) − 5bba

0 0 0
aaa ab −a2a (1. 5c + b) ba aa (3c + 2b)−a(abba + aa bb ) aba bb − b2a (1. 5c + b)
baa ab −a2b (1. 5c + a) bb ab (3c + 2a)−b(abba + aa bb ) bbb ba − b2b (1. 5c + a)

1 0 0

0 1 0 .
0 0 1
186 10 Triangular Elements for Kirchhoff Plates

The sub–matrices at the main–diagonal are computed as 6 × 6 matrices

 
1 0 0 0 0 0
 5 ac −bc 0 0 0 
 
 
 5 −ab bb 0 0 0 
G11 = ,

 10 4ac −4bc 0,5a2c −bc ac 0,5b2c 

 
 20 4(ac − ab ) −4(bc − bb ) −ab ac ab b c + ac b b −bb bc 
10 −4ab 4bb 0,5a2b −bb ab 0,5b2b

 
1 0 0 0 0 0

 5 aa −ba 0 0 0 

 
 5 −ac bc 0 0 0 
G22 = ,

 10 4aa −4ba 0,5a2a −ba aa 0,5b2a 

 
 20 4(aa − ac ) −4(ba − bc ) −ac aa ac b a + aa b c −bc ba 
10 −4ac 4bc 0,5a2c −bc ac 0,5b2c

 
1 0 0 0 0 0

 5 ab −bb 0 0 0 

 
 5 −aa ba 0 0 0 
G33 = .

 10 4ab −4bb 0,5a2b −bb ab 0,5b2b 

 
 20 4(ab − aa ) −4(bb − ba ) −aa ab aa b b + ab b a −ba bb 
10 −4aa 4ba 0,5a2a −ba aa 0,5b2a

10.2.5 Element Stiffness Matrix

The element stiffness matrix K is computed with
Z
K = G · ψ T · DT · E · D · ψ dA · G = GT · Kgen · G .
T
(10.10)

Here, Kgen incorporates the coefficients of the general polynomial and will be
transformed into the element stiffness matrix K related to the physically mea-
ningful nodal unknowns by applying the scaling matrix G. Thus integration
is performed with regard to the general polynomial. This results in simpli-
fied programs and therefore in a substantial reduction in numerical cost. The
computation of the integrand is demonstrated for two different possibilities.
10.2 Complete Approach of the 5th Order 187

1st Approach
Here, the B–matrix is not evaluated with the shape functions Ω but with the
general polynomial given with respect to the area coordinates

B = D·ψ. (10.11)

The differentiation with respect to x and y can be replaced by the differen-

tiation with respect to the area coordinates λi following Equation (7.14). The
multiplication and the integration procedure involved when computing the ele-
ment stiffness matrix Kgen employing the general polynomial described by area
coordinates are performed here, along the lines of Section 7.2, yielding
Z Z
Kgen = BT · E · B · 2A∆ dλa dλb .

Figure 10-7 shows the related multiplication scheme.

E B
3 21
3

21 3 3 21 21

K = . dA .
21

21
21

21
21

GT BT G

Fig. 10-7 Matrix scheme for the computation of K

The method of explicit integration to get the stiffness matrix is not performed
here, since the second approach is more efficient and clearly arranged.

2nd Approach
Here, to get the element stiffness matrix, the matrix operations in the integrand
are performed in advance with D and E, as per Section 6.2,

DT · E · D =
1−ν
B · { xx ∂∂xx + yy ∂∂yy + νxx ∂∂yy + νyy ∂∂xx + 4 xy ∂∂yx } . (10.12)
2
The multiplication yields a 1×1–matrix. For example, xx ∂ describes the partial
double differentiation with respect to x as related to the shape function in
front of it, and ∂xx describes the application of partial double differentiation
188 10 Triangular Elements for Kirchhoff Plates

with respect to x as related to the shape function behind it. When applying
this representation, it is possible to compute each term in the stiffness matrix
separately by applying the related differential operator. Such, the entry in the
8th row and the 10th column of the stiffness matrix Kgen can be performed, for
example, by applying the 8th shape function from Equation (10.6) in front of
the differential operator and by applying the 10th shape function from Equation
(10.6) behind the differential operator:
8th function : λ5b
10th function : λ4b λa
Z Z
Kgen [ 8, 10 ] = λ5b · DT · E · D · λ4b · λa · 2A∆ dλa dλb .

The differentiation with respect to x and y must be replaced by the differen-

tiation with respect to the area coordinates. To get the entries of the stiffness
matrix, the integration with respect to the area coordinates can be performed
analogously to Section 7.2. The final stiffness matrix K, as it is related to the
degrees of freedom, is transformed numerically using the scaling matrix.

10.2.6 Element Load Vector

The integration of the virtual external work is performed here concerning a
constantly distributed action. For the integration the general polynomial is
applied for convenience, which gives
Z Z Z Z
f = ΩT · p dA = ΩT dA · p = GT ψ T · 2A∆ dλa dλb · p . (10.13)
A A

After the integration, multiplication by the scaling matrix G is performed with

f = GT · fgen .
The load vector fgen , which is not scaled yet, is integrated analytically
T 1 1 1 2 1 2 2
fgen = 2A∆ · { [ 6·7 5·6·7 5·6·7 4·5·6·7 4·5·6·7 4·5·6·7 3·4·5·6·7
]

1 1 1 2 1 2 2
[ 6·7 5·6·7 5·6·7 4·5·6·7 4·5·6·7 4·5·6·7 3·4·5·6·7
]

1 1 1 2 1 2 2
[ 6·7 5·6·7 5·6·7 4·5·6·7 4·5·6·7 4·5·6·7 3·4·5·6·7
] }·p .

The computation of the element load vector concerning linearly varying actions
is not commonly accepted and hence is not shown here.
10.3 A Plate Element with 18 DOF 189

10.2.7 Subsequent Analysis for Stress Resultants

The instruction to compute the stresses, cf. Equation (2.39),

σ =E·D·Ω·v = E·B·v

is given here considering the general polynomial related to area coordinates,

σ (λi ) = E · D · ψ (λi ) · G · v ,

whereby σ stores the moments. At first, similar to the computation of the

element stiffness matrix, the differential operator

D · ψ (λi )

is to be performed if the moments’ magnitudes should be computed at any

arbitrary position in the element. The distribution of the moments’ magnitudes
inside the element and along its edges is described cubically. If the magnitudes
are to be computed at the corner nodes, the nodal unknowns of the curvatures
can be transformed according to the material equations in order to directly
compute the values.
The computation of the shear forces is performed similarly to that in Section
6.2.8, employing the 3rd order derivative of the deflection surface. At first the
matrix notation gives

q = Dq · σ = Dq · (E · D · ψ (λi ) · G · v) = Sq · v ,

with Sq representing the stress matrix for shear forces. Explicitly introducing
the shape functions for the deflections yields
" # " #
qx ψ (λi ) ,xxx + ψ (λi ) ,yyx
= −B · ·G·v.
qy ψ (λ ) ,yyy + ψ (λ ) ,xxy
i i

10.3 A Plate Element with 18 DOF

By eliminating the unknowns w,n related to the centres of the edges, one gets
an element which is substantially easier to implement into a program than
the element with 21 DOF. The elimination reduces the number of independent
unknowns, but the approach of 5th order remains unchanged. Very early, Butlin
and Ford [25], Cowper [30] and Bell [16] have independently developed elements
without the unknowns w,n .
190 10 Triangular Elements for Kirchhoff Plates

The starting point of this derivation is the 21 DOF element previously described
with the unknowns
T
v21 ={ [ w w,x w,y w,xx w,xy w,yy ]A
[ w w,x w,y w,xx w,xy w,yy ]B
[ w w,x w,y w,xx w,xy w,yy ]C
[ [ w,n ]D [ w,n ]E [ w,n ]F ] }.
Concerning the 18 DOF element, the slopes in the normal directions w,n at
the centres of the edges are described as a linear combination of the nodal
unknowns at the corner points. Therefore, the nodal unknowns
T
v18 ={ [ w w,x w,y w,xx w,xy w,yy ]A
[ w w,x w,y w,xx w,xy w,yy ]B
[ w w,x w,y w,xx w,xy w,yy ]C }
suffice. The elimination is possible only if the remaining unknowns are able to
consistently fulfill the element intersection conditions along the element edges,
i.e. without a gap in w and in w,n .

The 5th order approach of the 21 DOF element fulfills all element intersection
conditions, i.e. for the slope w,n as a polynomial of the 4th order employing 5
DOF at each edge. Eliminating w,n at the centres of the edges, 4 DOF of the
quality [ w,n w,ns ] are left at the element corners, and a Hermite Polynomial of
the 3rd order can be defined at each edge relating to w,n . Therefor, the slopes
w,n at the centres of the edges are described by a linear combination of the
nodal unknowns [ w,x w,y w,xx w,yy w,xy ] at the corner points, which have
to be transformed to the edge coordinates n, s. Thus the element intersection
conditions for [ w w,s ], described as a polynomial of the 5th order, are still
fulfilled exactly. However, w,n is only described with a polynomial of the 3rd
order instead of the 4th order, what causes a small loss in accuracy but does
not provide any other disadvantage.
At first, the general polynomial gives
w,n = a0 + a1 · s + a2 · s2 + a3 · s3 ,
which is adapted to the nodal unknowns [ w,n w,ns ] at the corner points of the
edges. In this way, the slopes at the corner points w,s and w,n are described by
means of w,x and w,y , cf. Equation (10.8). Similarly w,ns may be computed.
In general, the relationship between the nodal unknowns of the elements with
18 DOF and 21 DOF is defined by
v21 = T · v18
10.3 A Plate Element with 18 DOF 191

with the transformation matrix

" #
I
T= .
Tn

The first 18 rows describe the identities of the nodal unknowns at the element
corners. The relationship between the slopes at the centres of the edges and
the nodal unknowns at the corners is established only by the last three rows.
Applying the transformation matrix T, the transformation rule to get the ele-
ment with 18 DOF from the element with 21 DOF yields

stiffness matrix: K18 = TT · K21 · T ,

stress matrix: S18 = S21 · T ,
load vector: f18 = TT · f21 .

Concerning the FE–program, it is convenient to multiply, once and in advance,

the scaling matrix G21 of the element with 21 DOF by the transformation
matrix T from the right. The product G18 can be assumed to be the scaling
matrix of the element with 18 DOF:
 
G11 0 0
 0 G22 0 
 
 
 0 0 G33 
G18 = G21 · T =  .

 ← g7 → 

 
 ← g14 → 
← g21 →

It is given explicitly here and is employed after the integration of the general
polynomial instead of the scaling matrix G21 with the main diagonal sub–
matrices G11 , G22 , G33 as given in Section 10.2.4. The vectors gi comprise 18
columns and are summarised into a matrix as follows. Here ℓa , ℓb , ℓc mean the
lengths of the respective element side and β holds for the angle between the
side and the x–axis, see Figure 10-5.
 
g7 · c
 
 g14 · a  =
g21 · b
192 10 Triangular Elements for Kirchhoff Plates


−5(11a + 5b + 5c) −ac (16a + 9b + 4c) bc (16a + 9b + 4c)

 +5cab − sin β/2 −5cbb + cos β/2


 0 0 0
 −5(11a + 5c + 5b) ab (16a + 9c + 4b) −bb (16a + 9c + 4b)
−5bac − sin β/2 +5bbc + cos β/2

cab ac −a2c (1,5a + b) bc ac (3a + 2b) − cab bc cbb bc −b2c (1,5a + b)

+ sin β cos β lc /8 −cac bb −(1−2 sin2 β)lc /8 − sin β cos β lc /8
0 0 0
bab ac −a2b (1,5a + c) −bab bc + bb ab (3a + 2c) bbb bc −b2b (1,5a + c)
− sin β cos β lb /8 −bac bb +(1−2 sin2 β)lb /8 + sin β cos β lb /8

−5(11b + 5a + 5c) ac (16b + 9a + 4c) −bc (16b + 9a + 4c)

−5caa − sin β/2 +5cba + cos β/2
−5(11b + 5c + 5a) −aa (16b + 9c + 4a) ba (16b + 9c + 4a)
+5aac − sin β/2 −5abc + cos β/2
0 0 0

caa ac− a2c (1,5b + a) bc ac (3b + 2a) − caa bc cbc ba −b2c (1,5b + a)
− sin β cos β lc /8 −cac ba +(1−2 sin2 β)lc /8 + sin β cos β lc /8
aaa ac− a2a (1,5b + c) ba aa (3b + 2c) − aac ba aba bc −b2a (1,5b + c)
+ sin β cos β la /8 −aaa bc −(1−2 sin2 β)la /8 − sin β cos β la /8
0 0 0

0 0 0
−5(11c + 5b + 5a) aa (16c + 9b + 4a) −ba (16c + 9b + 4a)
−5aab − sin β/2 +5abb + cos β/2
−5(11c + 5a + 5b) −ab (16c + 9a + 4b) bb (16c + 9a + 4b)
+5baa − sin β/2 −5bba + cos β/2


0 0 0
2 2 
aaa ab −aa (1,5c + b) ba aa (3c + 2b) − aab ba aba bb −ba (1,5c + b) 
−aaa bb +(1−2 sin2 β)la /8

− sin β cos β la /8 + sin β cos β la /8 .

baa ab −a2b (1,5c + a) bb ab (3c + 2a) − bab ba bbb ba −b2b (1,5c + a) 
+ sin β cos β lb /8 −baa bb −(1−2 sin2 β)lb /8 − sin β cos β lb /8
10.4 A Comparison of Standard Plate Elements 193

10.4 A Comparison of Standard Plate Elements

In the following, a square plate is investigated which is loaded by a constantly
distributed loading p = 1. 0 N/m2 . The support is partially defined by clamping
and bearing respectively, as shown in Figure 10-8, where only one fourth of the
plate is discretized due to its symmetry. The investigation of convergence, cf.
Figures 10-10, 10-11 and 10-12, is limited to a few standard elements.
x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
7 2
E = 3.0 . 10 kN/m
T
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 10-8 Square plate – geometry and loading

The element meshes are depicted in Figure 10-9, whereat two triangles cover
the same area as one rectangular element.

Fig. 10-9 Square plate – meshes for rectangular and triangular elements

Since the solution employing the element comprised of 18 DOF only deviates
insignificantly from the solution employing the element with 21 DOF, just one
solution is presented here.
Concerning the general polynomial the difference between the rectangular ele-
ments comprised of 12 DOF and the rectangular element comprised of 16 DOF
is explained in Section 6.2.
194 10 Triangular Elements for Kirchhoff Plates

The quality of an element may be evaluated by means of the convergence and

the numerical effort to get the solution. The computing time required to solve
the system of equations is to be regarded as well as the bandwidth and the
number of unknowns. Additionally, the computational cost must be taken into
account, which is afforded to generate the stiffness matrix. The discretizations
for rectangular and triangular elements are given in Table 10.1.

Table 10.1 Square plate – mesh size, unknowns and bandwidth

12 16 △18

elements 1 4 16 64 1 4 16 64 2 8 32 128
unknowns 12 27 48 243 16 36 100 324 24 54 150 486
bandwidth 12 15 21 33 16 20 28 44 18 24 36 60

The following figures show the convergence behavior of state variables, which
indicate the quality of the elements. It is obvious, that due to the higher po-
lynomial degree the triangular element with 18 DOF yields superior results,
whereat the quality of the moments is of the same order as of the deflection.
Nonetheless, the rectangular elements converge to the right results as well but
very slowly regarding the moments.
As mentioned above the elements with 16 DOF respectively 18 DOF employ
curvatures as nodal DOF, what restricts their application to plates without
stiffeners and without discontinuities in stiffness distribution.
% error of wM
∆ 18
* 16
* 12 - weak conformity
* 12
5

0
20 100 500 dof

Fig. 10-10 Square plate – the convergence behavior of the deflection wM

10.4 A Comparison of Standard Plate Elements 195

% error of mxxM

40
∆ 18
* 16
30 * 12 - weak conformity
* 12
20

10

0
20 100 500 dof

-10

-20

Fig. 10-11 Square plate – the convergence behavior of the moment mxxM

% error in m yyE

20 100 500 dof

0

-5

-10

-15
∆ 18
* 16
-20
* 12 - weak conformity
* 12
-25

-30

Fig. 10-12 Square plate – the convergence behavior of the moment myyE
ISOPARAMETRIC ELEMENTS
11 Numerical Integration

When computing the element stiffness matrix and the load vector, an important
part is the solution of integrals like
Z Z Z
f (x) dx , f (x,y) dA and f (x,y,z) dV .
x A V

The integrand g(x) · h(x) or other products of functions are also included along
with the integrand f (x).
Concerning approaches of lower order, analytical integration can still be reaso-
nably managed, but for approaches of higher order, for example those shown
in Section 10, the effort increases enormously. For different problems as descri-
bed in Section 12 the analytical integration is not possible, and thus numerical
integration must be performed. Therefor different methods can be applied. Be-
sides the Newton–Cotes method, the Gauss–Legendre quadrature is also used
prevalently, since it offers the best possible solution for minimal effort. The
derivation of the Gauss–Legendre quadrature will not be shown in detail here
but its application in the FEM context. Basis for the numerical integration
scheme is the application of a normalized element coordinate system, as has
been described previously for the triangular elements in Section 7.

xA x0 xB

l/2 l/2

ξ = -1 ξ= 0 ξ = +1
A ξ B

Fig. 11-1 Global and local coordinate systems

Considering one dimension, the transformation of coordinates may be described

by means of Figure 11-1 with
ℓ
x = x0 + ·ξ,
2
where x is the global coordinate, x0 is the center of the element, ℓ is the
length of the element and −1. 0 < ξ < + 1. 0 is the local normalized coordinate.
Thus all integrals of the work equations can be transformed from the global

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_11
200 11 Numerical Integration

coordinates to the local one, if x is exchanged by ξ. Assuming f (x) and g(ξ) are
normalized shape functions to approach the displacement field, the integration
can be changed to local coordinates with
x=x
Z B ξ=1
Z
ℓ
f (x) dx = g(ξ) dξ ,
2
x=xA ξ=−1

where the integration with respect to local coordinates can be separated from
the element geometry, if the shape functions are described in local coordinates,
too.
In the following, the rules for numerical integration with respect to cartesi-
an coordinates of rectangular elements and to area coordinates of triangular
elements are explained.

11.1 Numerical Integration Using Gauss–Legendre Quadrature

In general, the representation in normalized element coordinates allows for the
following integral to be solved numerically. In the case of continuous polyno-
mials the integration can be performed exactly as the sum of the weighted
function values at the given supporting points ξi , which yields
Z1 n
X
f (ξ) dξ = f (ξi ) · wi . (11.1)
−1 i=1

Here, each function value f (ξi ) is to be multiplied by a fixed weighting factor

wi . As such, it is fundamental that the integration is performed with respect
to the normalized coordinate ξ, which can be transformed to any arbitrary real
length. As an example, the positions of the origin of coordinates and supporting
points are shown in Fig. 11-2.
f (ξ )

f (ξ i ) f ( ξ i+1)

ξ
−1 i 0 i +1 +1

Fig. 11-2 Gauss–Legendre quadrature – supporting points of a function

At numerical integration, the weighting factors wi may define the domain of

influence with respect to the corresponding supporting points. Related to the
11.1 Numerical Integration Using Gauss–Legendre Quadrature 201

supporting points ranging from n = 1 to n = 6, the coordinates of the suppor-

ting points as well as the weighting factors are given in Table 11.1. The number
of supporting points n depends on the order of the function to be integrated.
Considering n supporting points, the integration is exact up to an order of
m = 2n − 1.

Table 11.1 One dimension – supporting points ξi and weighting factors wi

n m = 2n − 1 ξi wi
1 1 0. 0 2. 0
√
2 3 ±1/ 3 1. 0
3 5 p 0. 0 8/9
± 3/5 5/9
4 7 ±0. 3399 8104 3585 0. 6521 4515 4863
±0. 8611 3631 1594 0. 3478 5484 5137
5 9 0. 0 0. 5688 8888 8889
±0. 5384 6931 0106 0. 4786 2867 0499
±0. 9061 7984 5939 0. 2369 2688 5056
6 11 ±0. 2386 1918 6083 0. 4679 1393 4573
±0. 6612 0938 6466 0. 3607 6157 3048
±0. 9324 6951 4203 0. 1713 2449 2379

An analogous procedure can be passed if the integration of a product of multiple

functions is performed, e.g. the product of virtual and real shape functions:
Z1 n
X
f (ξ) · g(ξ) dξ = f (ξi ) · g(ξi ) · wi .
−1 i=1

This is true, since the product of two functions gives a new function, which has
to be integrated numerically by the same integration rule.

Example
Analytical Integration:
Z+1 Z+1
1 1 2
f (ξ) · g(ξ) dξ = (1 − ξ) · (ξ 2 ) dξ = [ ξ 3 − ξ 4 ]+1
−1 = .
3 4 3
−1 −1
202 11 Numerical Integration

√
Numerical integration with n = 2, ξi = ±1/ 3, wi = 1. 0:
Z+1
1 1 1 1 2
(1 − ξ) · (ξ 2 ) dξ = 1 · [ (− √ )2 − (− √ )3 ] + 1 · [ (+ √ )2 − (+ √ )3 ] = .
3 3 3 3 3
−1

Multi–dimensional integration
Similarly, integrals can be solved regarding coordinates in two directions. The
positions of the supporting points are shown in Figure 11-3 for n = 3 in ξ– and
η–directions. Here the summation is to be performed in both directions
Z1 Z1 n X
X n
f (ξ, η) dξ dη = f (ξi , ηj ) · wi · wj . (11.2)
−1−1 i=1 j=1

1 1

1 3/5
ξ
1 3/5

3/5 3/5

η
Fig. 11-3 Rectangular domains – supporting points regarding a square at n = 3

Similarly, the numerical integration can be extended to volume integrals:

Z1 Z1 Z1 n X
X n X
n
f (ξ, η, ζ) dξ dη dζ = f (ξi , ηj , ζk ) · wi · wj · wk . (11.3)
−1−1−1 i=1 j=1 k=1

Triangles
Concerning the triangular elements, the integration is done using a single sum-
mation regarding the supporting points and the weighting factors, as shown in
Table 11.2
Z1 1−η
Z n
1 X
f (ξ, η) dξ dη = · f (ξi , ηi ) · wi . (11.4)
2 i=1
0 0
11.1 Numerical Integration Using Gauss–Legendre Quadrature 203

Table 11.2 Triangular domains – supporting points ξi and weighting factors wi

exact till position of the supporting points weighting-

the power supporting points ξ η factors
m λa λb λc wi
A

1 1 1
1 a a 3 3 3
1
C
B

A 1 1 1
a 2 2
0 3
c a 1 1 1
2 b 0 2 2 3
C c 1
0 1 1
b B 2 2 3

1 1 1
A
a 3 3 3
− 27
48
b 25
b 0. 6 0. 2 0. 2 48
3 a 25
c c 0. 2 0. 6 0. 2 48
C 25
d B d 0. 2 0. 2 0. 6 48

1 1 1
a 3 3 3
0. 225 000 000 0

b α1 β1 β1
A
e c β1 α1 β1 0. 132 394 152 7
c
5 a d d β1 β1 α1
f
C g e α2 β2 β2
b B
f β2 α2 β2 0. 125 939 180 5
g β2 β2 α2

α1 = 0. 059 715 871 7 , β1 0. 470 142 064 1

α2 = 0. 797 426 985 3 , β2 0. 101 286 507 3
204 11 Numerical Integration

11.2 Numerical Integration Applied to Membrane Elements

To perform numerical integration, the element coordinate system, cf. Figure
5-2, Section 5.1.4, is to be transformed into a normalized coordinate system as
given in Figure 11-4. Concerning the transformation between the global x–y–
coordinate system and the local ξ–η–coordinate system, the respective rule is
defined as
x = x0 + lx · ξ and
(11.5)
y = y0 + ly · η .

xo x

2 lx
A (−1;−1) B(1;−1)

yo 2 ly ξ

D (−1;1) C ( ξ = 1; η = 1)
η
y

Fig. 11-4 Local coordinates related to a rectangular element

Starting with the Principle of virtual Displacements the matrix formulation in

global coordinates is given with
Z Z
−δWd = δvT ΩT · DT · E · D · Ω dA v − δvT ΩT · p dA . (11.6)
A A

The transformation to local coordinates has to consider the following steps:

• Shape functions with respect to local coordinates.

• Derivatives of shape functions with respect to local coordinates.
• Integration of the virtual work.

So far, the transformation of the coordinates has been formally prepared. To

perform the differentation of the shape functions with respect to the ξ–η–
coordinates as well as the integration of the element work, the shape functions
11.2 Numerical Integration Applied to Membrane Elements 205

must also be given in local coordinates. In the following, symmetric linear and
quadratic shape functions and its derivatives are given with respect to local
coordinates.

11.2.1 Shape Functions Employing Local Coordinates

The linear shape functions are given in local coordinates as already covered in
Section 5.1.4:

φA = 14 (1 − ξ)(1 − η) φA,ξ = − 14 (1 − η) φA,η = − 14 (1 − ξ)

φB = 14 (1 + ξ)(1 − η) φB,ξ = 1
4
(1 − η) φB,η = − 14 (1 + ξ)
φC = 14 (1 + ξ)(1 + η) φC,ξ = 14 (1 + η) φC,η = 1
4
(1 + ξ)
φD = 14 (1 − ξ)(1 + η) φD,ξ = − 14 (1 + η) φD,η = 1
4
(1 − ξ)

A 1
2

1
4 1
D B

η ξ
C

Fig. 11-5 Linear shape function φB of the rectangular membrane element

The symmetric quadratic approach is a product of one of three quadratic shape

functions both in the x– and in the y–direction. Apart from the displacements
at the corners, the nine DOF require physically meaningful displacements at
the centres of the edges and at the centre point, see the nodes A...I in Figure
11-6. The Lagrange Polynomials assigned to the nodal displacements

vT = [ uA uB . . . uH uI | vA vB . . . vH vI ]

are subsequently given for the element employing quadratic shape functions.
The vector of the shape functions

φ = [ φA φB φC φD φE φF φG φH φI ]

incorporates the following shape functions in local coordinates

206 11 Numerical Integration

1 1
φA = 4
ξ (1 − ξ) · η · (1 − η) φA,ξ = 4
(1 − 2ξ) · η · (1 − η)
φA,η = 14 ξ (1 − ξ)(1 − 2η)
φB = − 14 ξ (1 + ξ) · η · (1 − η) φB,ξ = − 14 (1 + 2ξ) · η · (1 − η)
φB,η = − 14 ξ (1 + ξ)(1 − 2η)
1
φC = 4
ξ (1 + ξ) · η · (1 + η) φC,ξ = 14 (1 + 2ξ) · η · (1 + η)
φC,η = 14 ξ (1 + ξ)(1 + 2η)
φD = − 14 ξ (1 − ξ) · η · (1 + η) φD,ξ = − 14 (1 − 2ξ) · η · (1 + η)
φD,η = − 14 ξ (1 − ξ)(1 + 2η)
φE = − 12 (1 − ξ 2 ) · η · (1 − η) φE,ξ = − 12 (−2ξ) · η · (1 − η)
φE,η = − 12 (1 − ξ 2 )(1 − 2η)
1
φF = 2
ξ (1 + ξ)(1 − η 2 ) φF,ξ = 12 (1 + 2ξ)(1 − η 2 )
φF,η = 12 ξ (1 + ξ)(−2η)
1
φG = 2
(1 − ξ 2 ) · η · (1 + η) φG,ξ = 12 (−2ξ) · η · (1 + η)
φG,η = 12 (1 − ξ 2 )(1 + 2η)
φH = − 12 ξ (1 − ξ)(1 − η 2 ) φH,ξ = − 12 (1 − 2ξ)(1 − η 2 )
φH,η = − 12 ξ (1 − ξ)(−2η)
φI = (1 − ξ 2 )(1 − η 2 ) φI,ξ = − 2 ξ (1 − η 2 )
φI,η = (1 − ξ 2 )(−2η) .
A

H E
1
I
D B

G F
η ξ
C

Fig. 11-6 Quadratic shape function φB of the rectangular membrane element

11.2 Numerical Integration Applied to Membrane Elements 207

11.2.2 Derivatives with respect to local coordinates

The transformation of the partial derivatives with respect to the global coor-
dinates into partial derivatives with respect to the local coordinates is done
accordingly to Section 11.1 with Equation (11.7).
" #  ∂x ∂y  " # " # " #
∂ξ ∂ξ ∂ξ ∂x ℓx 0 ∂x
= ∂y
· = · . (11.7)
∂η ∂x ∂y 0 ℓy ∂y
∂η ∂η
| {z }
J

∂x ∂y ∂y ∂x
Here it holds det J = ∂ξ
· ∂η
− ∂ξ
· ∂η
= ℓx · ℓy .

The inverse representation yields

" # " # " 1 # " #
∂x  −1  ∂ξ 0 ∂ξ
= J · = ℓx 1 · . (11.8)
∂y ∂η 0 ℓy
∂η

11.2.3 Element Stiffness Matrix and Load Vector

The operator matrix D, as defined in Section 5.1.3, can be processed to build the
derivatives with respect to local ξ–η–coordinates. Therefor it is advantageous
to represent the operator matrix D as a product of two auxiliary matrices, to
enable a decoupling between the derivatives with respect to x and y
 
    ∂x 0
∂x 0 1 0 0 0
  ∂y 0 
 
  
D =  0 ∂y  =  0 0 0 1  ·  .
 0 ∂x 
∂y ∂x 0 1 1 0
0 ∂y
Next, the derivatives with respect to x–y–coordinates will be replaced by the
derivatives with respect to ξ–η–coordinates, using Equation (11.8)
 
  ∂ξ 0
1 0 0 0 " #
  J−1 0  ∂
 η 0 

D= 0 0 0 1 · −1
· . (11.9)
0 J  0 ∂ξ 
0 1 1 0
0 ∂η
Thereby, differentiation of the shape functions can be conducted in sub–steps,
if the nodal unknowns are not listed in the order of the nodes as before, but in
the order of the displacements. With
vT = [ uA uB uC uD | vA vB vC vD ]
208 11 Numerical Integration

the matrix of linear shape functions results in

" #
φ 0
Ω= with φ = [ φA φB φC φD ]
0 φ

and thus
 
  φ,ξ 0
1 0 0 0 " #
  J−1 0 
 φ,η 0 

B=D·Ω=  0 0 0 1 · ·  (11.10)
0 J−1  0 φ,ξ 
0 1 1 0
0 φ,η

or, briefly,

B = D · Ω = B1 · B2 · B3 . (11.11)

The elasticity matrix E, given in Section 5.1.3, and the matrix B, subdivi-
ded into three parts, yield the element stiffness matrix with respect to local
coordinates
Z Z 1Z 1
K= BT · E · B · dA = BT · E · B · det J dξdη . (11.12)
A −1 −1

The integral may be transformed by means of Equation (11.11) to

Z 1 Z 1
K= BT3 BT2 BT1 · E · B1 B2 B3 · det J dξ dη . (11.13)
−1 −1

B2 , B1 and E as well as det J = ℓx ·ℓy were discussed in Section 11.2.2. All the
terms are constant inside an element and hence they have the same numerical
value at each supporting point of the numerical integration. Thus the matrix
multiplication

Ê = BT2 · BT1 · E · B1 · B2 · det J

can be performed in advance, what yields the element related constant matrix
 ℓy 
ℓ
0 0 ν
x
 
E  0 1−ν · ℓx 1−ν
0 
 2 ℓy 2 
Ê = · . (11.14)
1 − ν2  0 1−ν 1−ν ℓy
· 0 
 2 2 ℓx 
ℓx
ν 0 0 ℓ y
11.2 Numerical Integration Applied to Membrane Elements 209

It simplifies the integrand in Equation (11.13), which gives

Z 1Z 1
K= BT3 · Ê · B3 dξ dη . (11.15)
−1 −1

The matrix B3 is comprised of the shape functions, differentiated with respect

to the local coordinates. For a linear approach this results in
 
φA,ξ φB,ξ φC,ξ φD,ξ 0
 
 φA,η φB,η φC,η φD,η 0 
B3 = 
.

 0 φA,ξ φB,ξ φC,ξ φD,ξ 
0 φA,η φB,η φC,η φD,η

The sequence of the nodal unknowns yields a clear partition of the matrix B3
and the stiffness matrix K into 4 sub–matrices. Table 11.3 illustrates the arran-
gement of the integrand of Equation (11.15) employing linear shape functions.

To compute the element stiffness matrix by numerical integration, the matrix

multiplication BT3 · Ê · B3 is to be performed considering the numerical values
of the shape functions at the supporting points. Thereby, the zero entries in Ê
will be omitted immediately.
When employing bi–linear shape functions, n = 2 supporting points are needed
in each direction. Since the matrix Ê is constant and the matrix B3 is still linear
in one direction because of the partial differentiation in the other direction, then
the integrand in Equation (11.15) will employ power terms at a maximum order
of m = 2 in each direction.
The selectively reduced integration of the stiffness matrix, given in Section 5.2,
is performed by integrating the terms multiplied by ê22 , ê23 , ê32 , ê33 using a
lower order of integration. If numerical integration is to be performed using
2 × 2 supporting points for example, then the terms defined above will be
evaluated using only a 1–point integration. The 1–point integration matches a
constant distribution of shear stresses.

Employing bi–quadratic shape functions to approach the displacements, 9 no-

des are available per element, so that B3 is to be extended to the shape functions
φA until φI . In this case n = 3 supporting points are needed to numerically
integrate the stiffness matrix, since the integrand in Equation (11.15) incorpo-
rates power terms at a maximum order of m = 4 in each direction. In this case,
a reduced integration with respect to each direction does not resolve the issue.
Here, the shape functions for the shear terms are to be chosen according to the
shear deformations being linearly distributed inside the element.
 210

B3
φA,ξ φB,ξ φC,ξ φD,ξ 0 0 0 0
φA,η φB,η φC,η φD,η 0 0 0 0
0 0 0 0 φA,ξ φB,ξ φC,ξ φD,ξ
0 0 0 0 φA,η φB,η φC,η φD,η

ê11 0 0 ê14 ê11 φA,ξ ê11 φB,ξ . . . ê14 φA,η ê14 φB,η . . .
0 ê22 ê23 0 ê22 φA,η ê22 φB,η . . . ê23 φA,ξ ê23 φB,ξ . . .
0 ê32 ê33 0 ê32 φA,η ê32 φB,η . . . ê33 φA,ξ ê33 φB,ξ . . .
ê41 0 0 ê44 ê41 φA,ξ ê41 φB,ξ . . . ê44 φA,η ê44 φB,η . . .
Ê Ê · B3

φA,ξ φA,η 0 0
φB,ξ φB,η 0 0 (BT3 · Ê · B3 )δu u (BT3 · Ê · B3 )δu v
φC,ξ φC,η 0 0
φD,ξ φD,η 0 0
0 0 φA,ξ φA,η
0 0 φB,ξ φB,η (BT3 · Ê · B3 )δv u (BT3 · Ê · B3 )δv v
0 0 φC,ξ φC,η
0 0 φD,ξ φD,η
BT3 BT3 · Ê · B3
Table 11.3 Matrix scheme of the integrand to compute the stiffness matrix
11 Numerical Integration
11.2 Numerical Integration Applied to Membrane Elements 211

11.2.4 Element Load Vector

A constantly distributed external action yields an integral in local coordinates
Z Z Z 1Z 1
f = ΩT · p dA = ΩT dA · p = ΩT (ξ, η) · det J dξdη · p .
−1 −1

In the case of a rectangular element, the determinant of the Jacobian matrix

is constant, thus only the shape functions are to be integrated numerically.
Since these functions employ lower order terms compared to the terms to be
integrated when evaluating the stiffness matrix, less supporting points are to
be considered at numerical integration.

11.2.5 The Subsequent Stress Analysis

Although numerical integration is not required for the computation of the stress
matrix, it is given here for completeness. The preparation shows some simila-
rities compared to the computation of the stiffness matrix. The stress matrix
yields, see Section 5.1.8,

σ̃ = E · B̃ · v = S̃ · v

or, after transformation,

S̃ = E · B̃ = E · B1 · B2 · B̃3 = Ẽ · B̃3 .

Here, Ẽ is also computed once in advance.

 1 ν

0 0
E  ℓx ℓy

 ν 0 0 1 
Ẽ = · .
1 − ν 2  ℓx ℓy 
0 1−ν
2ℓ
1−ν
2ℓx
0
y

The zero entries in Ẽ will be immediately omitted in the program. At the

multiplication of the term Ẽ · B̃3 , the fact that the sequence of unknowns is
fixed to uA , uB , uC . . . , vA , vB , vC . . . in B̃3 should be taken care of. However,
the sequence uA , vA , uB , vB , . . . should be applied in the program code, cf. the
procedure to compute the element stiffness matrix.
In the case of the corresponding local coordinates being given, the stresses
can be evaluated either at the element nodes or at the supporting points of
the Gauss–Legendre quadrature. It can be shown that the stresses are best
approximated at the supporting points. This can be explained by the fact that
the integral of the internal work δε·σ is also evaluated at the supporting points
at the numerical integration.
212 11 Numerical Integration

11.3 Triangular Elements

Triangular elements for membranes applying local coordinates are explained
in Section 9 and for plates in Section 10. The shape functions as well as the
integration are represented in both local ξ–η–coordinates and surface coordi-
nates λa , λb , λc , but integration is performed analytically so far. To switch to
numerical integration, only the supporting points need to be taken from Sec-
tion 11.1. It should be mentioned that in the case of triangular elements, the
directions of integration are not distinguished from each other, instead a simple
single–summation is performed over all the supporting points.

11.3.1 Triangular Elements for Membranes

Considering a linear approach for the displacements and therefore constant
strains, the numerical integration of the element stiffness matrix requires only
a single supporting point, since the function to be integrated is constant.
Employing a quadratic approach for the displacements and thus linear strains
three supporting points are to be considered, since the integrand is quadratic.

Similarly to Section 11.2.3, the following notation is applied at integration. For

triangular elements the inverse of the Jacobian matrix is represented by
 
−1 1 bb bc
J = ·
det J ab ac

with det J = ac · bb − ab · bc being constant within the element. Thus the

multiplication of the matrices E, B1 , B2 yields the following Ê–matrix due to
Equation (11.14)

Ê = BT2 · BT1 · E · B1 · B2 · det J ,

which may be coded directly. As an abbreviation, e = (1 − ν)/2 is used.

E 1
Ê = · ·
1 − ν 2 det J
 
b2b + e a2b b b b c + e ac ab (e + ν) bb ab e bc ab + ν ac bb
 
 b b b c + e ac ab
 e a2c + b2c e bb ac + ν bc ab (e + ν) ac bc 

· .
 (e + ν) bb ab e bb ac + ν bc ab
 e b2b + a2b e b b b c + ac ab 

e bc ab + ν ac bb (e + ν) ac bc e bb bc + ac ab a2c + e b2c
11.3 Triangular Elements 213

When assembling B3 , three shape functions for the linear approach and six
shape functions for the quadratic approach are to be considered for u and v
respectively. Therefore, the integrand of the stiffness matrix

BT3 · Ê · B3

is comprised of the same terms as in Section 11.2.3. At the numerical integration

the factor 1/2 in Equation (11.4) is to be taken into account.
The load vector is computed analogously to the case of rectangular elements.

To perform the subsequent stress analysis the matrices, Ẽ and B3 are to be

adapted. Thus the stress matrix, analogously taken from Section 9.4.3, yields

S̃ = E · B̃ = E · B1 · B2 · B̃3 = Ẽ · B̃3 ,

with
Ẽ = E · B1 · B2 .
Here, Ẽ is computed for the whole element to be evaluated
 
bb bc ν ab ν ac
E 1  
Ẽ = · ·  ν bb ν bc ab ac . (11.16)
1 − ν 2 det J  1−ν 1−ν 1−ν 1−ν

2
ab 2
ac 2
bb 2
bc

11.3.2 Triangular Elements for Plates

Triangular plate elements employing 18 or 21 DOF are derived in Section 10.
Because of the employment of the general polynomial the numerical integration
of the virtual work may be focussed on the integrals
Z Z
Kgen = ψ T DT E D ψ dA and fgen = ψ T p dA .

The stiffness matrix regarding the general polynomial and the corresponding
load vector have to be transformed to the physically meaningful nodal unknows
by means of the scaling matrices G18 and G21 , which are not affected by the
numerical integration. The computation of the scaling matrix and the deter-
mination of the globally positive direction of the normal vector with respect
to each edge of the element is presented explicitly in Sections 10.2.4 and 10.3.
The matrix multiplications

K = GT
18 · Kgen · G18 as well as f = GT
18 · fgen
214 11 Numerical Integration

are performed after the numerical integration of the general polynomial has
been executed. With constantly distributed external actions, the integration
can also be performed explicitly, see Section 10.2.6.

The following should be taken care of at numerical integration:

A complete approach of the 5th order is presented for the element employing
21 DOF. Since the general polynomial is differentiated twice, it leads to an
integrand of the element stiffness matrix which comprises a polynomial of 6th
order.
For an exact integration of the element A
stiffness matrix, 12 supporting points
are to be considered together with the
corresponding weighting factors, see the 2

figure on the right. The coordinates of 10 12

the supporting points are given at 15
places behind the decimal point in Ta- 1
11 5 4 9
ble 11.2. The integration is performed,
as it is the rule for triangular elements, 7 3 8 6
using a single–summation over all sup-
C B
porting points.
The stress analysis is performed similarly to the other elements. The bending–
and twisting–moments at the corner nodes are computed formally using the
stress matrix, whereat in this case the stress matrix comprises a direct corre-
lation of the moments and the curvature DOF at the corner nodes.
12 Isoparametric Elements

In Section 11.2, to map a rectangular element to a unit square, a constant

transformation rule is applied element by element. Now the same (= iso) shape
functions φ, which are chosen to describe the physical behavior of an element,
may also be used to describe the element geometry. This idea is first published
by Taig [95] and later on introduced by Irons and Zienkiewicz [52] as isopara-
metric element concept.

rectangle quadrilateral
x x
(1;−1)
(−1;−1) (1;−1) linear (−1;−1)
functions

y ξ y ξ

(−1;1) (1;1) (−1;1)

η (1;1)
η

x xa xe xb x
quadr.
y E B
functions a A

I F
yh H ξ
y ξ

yd D C
G
η
η
y

Fig. 12-1 Geometry variations employing isoparametric elements

The most significant advantage of an isoparametric element is its adaptabili-

ty to any existing geometry, e.g. angular or curvilinear boundaries, see Figure
12-1. Sometimes super– or sub–parametric elements are also employed. With
the former ones, the order of the shape functions used to describe the element
geometry is higher than the order used to describe physics and with the lat-
ter ones, the reverse is true. An example of a sub–parametric element is the
membrane element employing 9 DOF and thus quadratic shape functions to
describe the displacements u(x, y) and v(x, y) while describing the geometry
by only employing linear shape functions.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_12
216 12 Isoparametric Elements

Generally, isoparametric elements considering both four or nine nodes are to be

applied carefully. Since the typically used multiplicative approach of the displa-
cements is not coordinate–invariant, the results become increasingly inferior as
the geometry deviates more and more from a rectangle. Hence the mesh needs
to be inspected, for example, the nodal angles are to be looked at. This could
be evaluated by

90o − ε ≤ ϕ ≤ 90o + ε,

when ε is a measurement of tolerance, e.g. of 10o .

In principle the idea of an isoparametric approach can also be applied to tri-
angular elements. A triangular element, arbitrarily lying in the x–y–plane, see
Section 7.1, may be interpreted as a triangular isoparametric element. In this
case, the description of the geometry is also performed using a linear trans-
formation from local ξ–η–coordinates to global x–y–coordinates. Thereby the
employment of complete shape functions is advantageous, because this auto-
matically results in a coordinate–invariant approach.

12.1 Description of the Element Geometry

The additional step from a rectangular element as given in Section 11.2 to an
isoparametric quadrilateral element consists of performing the changed trans-
formation rule between the ξ–η–coordinates and the x–y–coordinates. Inside
an element, each arbitrary pair of coordinates (x, y) is described now by shape
functions related to scaled coordinates

x = φ(ξ, η) · x̃ ,
(12.1)
y = φ(ξ, η) · ỹ .

Thereby, the shape functions are the same as before

φ(ξ, η) = [ φA φB φC φD ... ]

and x̃ as well as ỹ represent the vectors of the coordinates of the element nodes
   
xa ya
 xb   yb 
   
 xc 
ỹ =  yc  .
 
x̃ =  ,
 xd   yd 
   
.. ..
. .
12.1 Description of the Element Geometry 217

Since the coordinates x and y now depend on ξ and η, the transformation of

the partial derivatives is performed by
" #  ∂x ∂y  " #
∂ξ ∂ξ ∂ξ ∂x
= ∂y
· .
∂η ∂x ∂y
∂η ∂η
| {z }
J

The matrix entries of the Jacobian matrix J may be computed applying Equa-
tions (12.1):
∂x ∂y
= φ,ξ · x̃ , = φ,ξ · ỹ ,
∂ξ ∂ξ
∂x ∂y
= φ,η · x̃ , = φ,η · ỹ .
∂η ∂η
In general the Jacobian matrix follows with
 
φA,ξ xa φA,ξ ya
 + φB,ξ xb + φB,ξ yb 
" #  
φ,ξ · x̃ φ,ξ · ỹ  + ... + ... 
 
J= = , (12.2)
φ,η · x̃ φ,η · ỹ  φA,η xa φA,η ya 
 
 + φB,η xb + φB,η yb 
+ ... + ...
which yields the inverse of the Jacobian matrix
" # " #
i11 i12 −1 1 i22 −i12
J= , J = ·
i21 i22 det J −i21 i11

with det J = i11 · i22 − i12 · i21

P P
and i11 = φk,ξ · xk , i12 = φk,ξ · yk ,
k k
P P
i21 = φk,η · xk , i22 = φk,η · yk ,
k k

where k runs over all nodes and shape functions inside the element respectively.
In principle, the Jacobian matrix which is still dependent on ξ and η is not
constant. Hence, J−1 and det J cannot be evaluated for the entire element
but only for each separate coordinate pair ξi and ηj . Thus the integration
cannot be performed analytically any longer, and numerical integration needs
to be applied, e.g. the Gauss–Legendre quadrature. Normally, the number of
supporting points is chosen as in the case of a rectangular element.
218 12 Isoparametric Elements

12.2 Isoparametric Elements Regarding Membranes

The procedure of how to develop isoparametric elements related to membranes
shall now be demonstrated, whereby the respective rectangular element has al-
ready been prepared for numerical integration, cf. Section 11.2.3. The extension
to non–rectangular elements needs the following supplement.

12.2.1 The Stiffness Matrix and the Stress Matrix

Following Section 11.2.3 the element stiffness matrix of a rectangular element
can be computed with Equation (12.3).
Z 1Z 1
K= BT3 BT2 BT1 · E · B1 B2 B3 · det J dξ dη . (12.3)
−1 −1

Applying the transformation of coordinates as represented in Section 12.1 yields

the stiffness matrix of an quadrilateral isoparametric element. Hence, perfor-
ming the integration to get the element stiffness matrix, the matrices

J, det J , J−1 , Ê

need to be computed at every supporting point, since the transformation of the

derivatives may vary inside the element. This gives
 
1 ν
 1−ν 1−ν 
T E  2 2 
E 1 = B1 · E · B1 = ·  .
1 − ν2  1−ν 1−ν 

2 2
ν 1

It is followed by the numerical evaluation of Ê at each supporting point. Ap-

plying
" #
J−1 0
B2 =
0 J−1

yields

Ê = BT2 · E1 · B2 · det J .

In contrast to Section 11.2.3, Ê is now completely filled, since B2 is not a

diagonal matrix. Ê can be computed analytically and can be evaluated for
each supporting point.
12.2 Isoparametric Elements Regarding Membranes 219

The multiplication of the matrices yields

E 1
Ê = 2
· ·
1 − ν det J
 
i222 + e i221 −i22 i12 − e i11 i21 −(e + ν)i22 i21 e i12 i21 + νi11 i22
 
 −i22 i12 − e i11 i21
 e i211 + i212 e i22 i11 + ν i12 i21 −(e + ν)i11 i12  
· .
 −(e + ν)i22 i21 e i22 i11 + ν i12 i21 2 2
 e i22 + i21 −e i22 i12 − i11 i21 

e i12 i21 + νi11 i22 −(e + ν)i11 i12 −e i22 i12 − i11 i21 i211 + ei212

As an abbreviation, e = (1 − ν)/2 is used.

Similarly, at the computation of the stress matrix, only part of the matrix
multiplications to get Ẽ can be performed independently of the coordinates, at
which the stresses are to be evaluated. Again, the stresses may be determined
at both the supporting points or at the element nodes, cf. Section 11.2.3. Thus
the stress matrix, analogously taken from Section 11.2.3, yields

S̃ = E · B̃ = E · B1 · B2 · B̃3 = Ẽ · B̃3 . (12.4)

Ẽ is computed for each point to be evaluated

 
i22 −i12 −ν i21 ν i11
E 1  νi 
Ẽ = · · 22 −ν i12 −i21 i11 . (12.5)
1 − ν 2 det J

1−ν 1−ν 1−ν 1−ν
− 2 i21 i
2 11
i
2 22
− i
2 12

12.2.2 The Selectively Reduced Integration

Detrimental load–carrying behavior concerning bending of the elements em-
ploying linear shape functions has already been discussed in Sections 5.2, and
a procedure using modified shear deformations has been suggested as a way
out. For isoparametric membrane elements, the modification of the element
stiffness matrix is possible by means of selectively reduced integration, too.

When employing a linear approach for the displacements, this can be achieved
if the terms, resulting from the entries (1,2), (1,3), (2,1), (2,2), (2,3), (2,4), (3,1),
(3,2), (3,3), (3,4), (4,2) and (4,3) of the Ê–matrix, are integrated numerically
applying a single supporting point. It means that the parts of the integrand
with respect to the terms u,y and v,x must be described to be constant inside
the element. The results given in Section 12.2.3 illustrate the improvements
concerning accuracy.
220 12 Isoparametric Elements

By employing quadratic shape functions for the displacements, reduced inte-

gration can be performed by applying two supporting points in each direction,
concerning the parts of the integrand relating to the terms u,y and v,x . This
procedure provides very good results although it cannot be compared directly
to that regarding linear shape functions. Nevertheless, in this case, integration
should correspond to a linear distribution of u,y and v,x respectively. There-
by the integrands would arise quadratic at maximum, which does not correlate
with integration when applying two supporting points, whereby third order po-
lynomials may also be exactly integrated. Hence, at the integration procedure
chosen here, unwanted parts of a higher order are also integrated. Nevertheless,
the results are excellent, see Section 12.2.3.
As an alternative, the terms u,y and v,x could be described directly using
linear shape functions, so that the higher order terms disappear. Therefore,
the numerical integration can be performed applying the well–known scheme
and the order of integration with respect to the other parts of the integral.

The reduced integration does not need to be considered at the subsequent stress
analysis, since correct nodal displacements are automatically followed by the
evaluation of correct stresses.

12.2.3 Comparison of Standard Membrane Elements

The quality of isoparametric quadrilateral elements can be checked, if the re-
sults are compared to the results obtained from triangular elements, which are
comprised of complete shape functions, cf. Section 9. A benchmark often given
in the literature, is the cantilever first analysed by Cook [28], shown in Figure
12-2.
c
x
B
y D 1N
b

C
E = 1.0 N/mm 2
ν = 0.33
a

A a = 44 mm
b = 16 mm
c = 48 mm

Fig. 12-2 Cook’s cantilever – geometry and loading

12.2 Isoparametric Elements Regarding Membranes 221

In the following, the cantilever is investigated by employing 4– and 9–node qua-

drilateral elements as well as 3– und 6–node triangular elements. The meshes
related to triangular elements develop from the division of the quadrilaterals
related to the smaller diagonal. The results definitely are of inferior quality if
the quadrilaterals are divided along the longer diagonal.
The following Table 12.1 shows the convergence of the vertical end displace-
ment in the centre line of the cantilever regarding different types of elements.
Employing the same number of unknowns, the results obtained from the qua-
dratic approach are definetely of a superior quality compared to those obtained
from the linear approach. Applying the selectively reduced integration of the
quadrilateral elements as presented in Section 12.2.2 improves the results sub-
stantially, especially regarding the coarse element mesh.

Table 12.1 Cook’s cantilever – vertical displacement vD

El. mesh 3×3 5×5 9×9 17 × 17 33 × 33 65 × 65

1 iso - 4 11. 845 18. 299 22. 079 23. 430 23. 817 23. 924
2 iso - 4 - SRI 20. 043 22. 648 23. 569 23. 846 23. 927 23. 953
3 iso - 9 19. 644 23. 289 23. 839 23. 925 23. 949 23. 960
4 iso - 9 - SRI 22. 626 23. 653 23. 885 23. 937 23. 954 23. 962
5 tri - 3 11. 991 18. 283 22. 022 23. 411 23. 815 23. 924
6 tri - 6 18. 358 23. 301 23. 856 23. 935 23. 951 23. 961

Tables 12.2 and 12.3 present the convergence of the stresses σxx at the positions
A and B. The convergence behavior already noted regarding the displacements
is more clearly detected here. The best convergence can be observed for the
9–node element by applying selectively reduced integration, cf. Section 12.2.2.

Table 12.2 Cook’s cantilever – compressive stresses σxx · (−1) at position A

9×9 17 × 17 33 × 33 65 × 65
El left right left right left right left right
1 0. 1907 0. 0947 0. 1676 0. 1149 0. 1499 0. 1228 0. 1397 0. 1261
2 0. 1497 0. 1619 0. 1411 0. 1468 0. 1353 0. 1382 0. 1321 0. 1335
3 0. 1349 0. 1438 0. 1298 0. 1333 0. 1289 0. 1300 0. 1287 0. 1291
4 0. 1546 0. 1412 0. 1354 0. 1328 0. 1302 0. 1299 0. 1291 0. 1290
5 0. 0987 0. 0960 0. 1126 0. 1154 0. 1204 0. 1230 0. 1245 0. 1261
6 0. 1207 0. 1435 0. 1263 0. 1331 0. 1281 0. 1299 0. 1285 0. 1290
222 12 Isoparametric Elements

Table 12.3 Cook’s cantilever – tensile stresses σxx at position B

9×9 17 × 17 33 × 33 65 × 65
El. left right left right left right left right
1 0. 1610 0. 1852 0. 1759 0. 1917 0. 1808 0. 1893 0. 1824 0. 1867
2 0. 1779 0. 1609 0. 1805 0. 1736 0. 1816 0. 1786 0. 1825 0. 1810
3 0. 1808 0. 1894 0. 1830 0. 1846 0. 1832 0. 1835 0. 1832 0. 1832
4 0. 1819 0. 1921 0. 1833 0. 1844 0. 1833 0. 1834 0. 1832 0. 1832
5 0. 1619 0. 1452 0. 1759 0. 1694 0. 1808 0. 1779 0. 1823 0. 1810
6 0. 1790 0. 1848 0. 1828 0. 1839 0. 1831 0. 1834 0. 1832 0. 1832

As known from rectangular elements the jump of the stresses at the intersection
of neighboring elements is an indication of the discretization error and the
convergence behavior. As expected, the convergence of stresses develops poorly
compared to the convergence of displacements. Because of the non–rectangular
element geometry, which normally yields poor shear stress approximations, the
distributions of the other stresses are also badly influenced. Furthermore, the
fulfillment of the Neumann boundary conditions needs a combination of all
stresses, what includes a transformation of the equilibrium conditions at the
boundary.
The course of the stresses σxx at the cross–section A–B is shown in Figure 12-3
for the mesh sized with 65 × 65 nodes. Due to the oblique boundaries at the
upper and the lower edge of the cantilever the course is not linear but slightly
curved.

B
48
x
B
16

y
44

sxx
- 0.1

- 0.2
0.2

0.1

0.0

Fig. 12-3 Cook’s cantilever – stress σxx at section A–B regarding 65 × 65 nodes
12.3 Quadrilateral Plate Elements 223

Figure 12-4 gives the distributions of the stresses σxx , σyy , σxy over the whole
structure.
Stress σxx Stress σyy Stress σxy

−0. 20 N/mm2 + 0. 20

Fig. 12-4 Cook’s cantilever – the stresses regarding 65 × 65 nodes

12.3 Quadrilateral Plate Elements

An isoparametric approach equivalent to the rectangular plate element employ-
ing 16 DOF is subject to restrictions, so that other types of elements should be
chosen in order to enable arbitrary mesh geometries, cf. Figure 12-5. The plate
element employing 12 DOF, cf. Section 6.3, does not, in fact, behave conform
but can be applied as a sub–parametric element.

A
y x (-1,-1)
z, w (1,-1)
B

(-1,1) η
D ξ
w

local coordinates (ξ, η)

(1,1)

C
Fig. 12-5 Local coordinates of a quadrilateral plate element
224 12 Isoparametric Elements

At both the development of the stiffness matrix and the scaling matrix, and in
contrast to the membrane elements, the 1st as well as the 2nd order derivatives
of the deflection need to be transformed because of the curvature. The 1st and
the 2nd order derivatives with respect to the local coordinates ξ, η are trans-
formed into the respective derivatives with respect to the global coordinates x
and y altogether by matrix operation, cf. the following scheme
   ∂x   ∂w 
∂w ∂y
 ∂ξ   ∂ξ ∂ξ  ∂x 
    
 ∂w   ∂x ∂y  ∂w 
    
   
 ∂η   ∂η ∂η   ∂y
 
 2   2 2   ∂2w 
 ∂ w   ∂ x ∂ y ∂x 2 ∂y 2 ∂x ∂y  

 ∂ξ 2
=
  ∂ξ 2 ( ) ( ) 2( ) .
∂ξ 2   ∂x2

   ∂ξ ∂ξ ∂ξ ∂ξ 

 ∂2w   ∂2x 2  2
   ∂ y ∂x 2 ∂y 2 ∂x ∂y  ∂ w 

( ) ( ) 2( ) 
 ∂η 2 2 2   ∂y 2
   
  ∂η ∂η ∂η ∂η ∂η ∂η  
 2   2 2

 ∂ w  ∂ x ∂ y ∂x ∂x ∂y ∂y ∂x ∂y ∂y ∂x   ∂ 2 w 
( ) ( ) ( + )
∂ξ∂η ∂ξ∂η ∂ξ∂η ∂ξ ∂η ∂ξ ∂η ∂ξ ∂η ∂ξ ∂η ∂x∂y

Employing a linear approach to describe the coordinate transformation x(ξ, η)

and y(ξ, η), the second derivatives with respect to ξ and η disappear.
The matrix of coefficients equals the Jacobian matrix. The inverse of the Ja-
cobian matrix is to be introduced into the integrand of the Principle of virtual
Work, similar to the way by which membrane elements were handled. Addi-
tionally, the Jacobian matrix is needed to compute the scaling matrix and to
perform the subsequent stress analysis.

12.3.1 The 16 DOF Plate Element

The approach employing 16 DOF is not complete regarding 4–node Kirchoff
plate elements, cf. Section 6.2. Hence, it must inevitably run into problems with
respect to the isoparametric concept. It must also be noted that the shape func-
tions are related to the physically meaningful unknowns, the nodal rotations
w,x and w,y , and partially the curvature w,xy .
It is essential that the rotations w,x and w,y are transformed consistently from
local to global coordinates, but the curvature w,xy is not, as the curvatures w,xx
and w,yy are missing. Thus w,xy remains solely an element variable, which is
coupled with the respective local coordinate system. The curvature w,xy bears
the same meaning for two neighboring elements only if the local coordinates
pass from one element to the next without a kink at the intersection line, see
Figure 12-6. Therefore, the plate element employing 16 DOF is only satisfac-
12.3 Quadrilateral Plate Elements 225

torily applicable, if regular element meshes are generated without any kinks at
the mesh lines.

regular mesh lines − irregular mesh lines −

possible not possible

Fig. 12-6 Sub–parametric plate elements – meshes employing quadrilaterals

When performing numerical integration of the elment matrices, the following

needs to be considered. The deflection of the element employing 16 DOF is
described by means of a bi–cubic approach. Although the shape functions are
double–differentiated in one direction, the integrand is comprised of a poly-
nomial of the 6th order with respect to the other direction. Therefore, nume-
rical integration is to be performed considering four supporting points in each
direction.
Because the origin of the coordinate system now lies in the centre of the related
element, the Hermite Polynomials of Section 6.2 need to be transformed to the
local coordinate system. This yields

φ1 (ξ) = (2 − 3ξ + ξ 3 )/4 φ1 (η) = 2 − 3η + η 3 )/4

2 3
φ2 (ξ) = (1 − ξ − ξ + ξ ) · ℓx /8 φ2 (η) = (1 − η − η 2 + η 3 ) · ℓy /8
φ3 (ξ) = (2 + 3ξ − ξ 3 )/4 φ3 (η) = (2 + 3η − η 3 )/4
2 3
φ4 (ξ) = (−1 − ξ + ξ + ξ ) · ℓx /8 φ4 (η) = (−1 − η + η 2 + η 3 ) · ℓy /8 .

12.3.2 The 12 DOF Plate Element

The rectangular plate element employing 12 DOF exhibits the deflection w and
the rotations w,x and w,y at the corner nodes as unknowns, and thus does not
have the difficulties mentioned above.
The choice of the shape functions for the element with 12 DOF is possible by
means of a general polynomial. Ten DOF ai are assigned to a complete cubic
226 12 Isoparametric Elements

approach as per Pascal’s triangle. Additionally the exponents ξ · η 3 and ξ 3 · η

are chosen for the remaining two DOF. This is done similar to Equation (6.25)

w(x,y) = a1 + a2 ξ + a3 η
+ a4 ξ 2 + a5 ξη + a6 η 2
+ a7 ξ 3 + a8 ξ 2 η + a9 ξη 2 + a10 η 3
+ a11 ξ 3 η + a13 ξη 3 . (12.6)

In a first step the stiffness matrix and the load vector should be integrated
concerning the general polynomial. Four supporting points are needed in each
direction, because of the cubic order of the approach. The scaling of the co-
efficients of the general polynomial to the physically meaningful unknowns is
performed afterwards by means of the multiplication of the stiffness matrix as
well as of the load vector by the scaling matrix G12 .

The scaling matrix for the element with 12 DOF can be derived for the isopa-
rametric quadrilateral element with

[ w(x,y) ] = ψ 12 · a12
v12 = Ψ̃12 · a12

and
−1
a12 = Ψ̃12 · v12 = G12 · v12 .

Thus the approach for the element with 12 DOF is established by the scaling
matrix developed on the next pages. Since the continuity conditions of the rec-
tangular element with 12 DOF are not fulfilled as discussed in Section 6.3, it is
obvious, that the continuity conditions at the intersections of the isoparametric
element are not satisfied as well. Nonetheless, the element converges against the
correct solution.

Fig. 12-7 Shape functions for the element with 12 DOF

12.3 Quadrilateral Plate Elements 227

The scaling matrix G12 with respect to the nodal degrees of freedom
 
vT = [ w w,x w,y ]A [ w w,x w,y ]B [ w w,x w,y ]C [ w w,x w,y ]D

may be developed by means of the Jacobian matrix and the following procedure.
Taking into account a quadrilateral element as represented in Figure 12-5 the
element geometry is described by the differences of the coordinates

a1 = 0. 25 · (−xA + xB + xC − xD )
a2 = 0. 25 · (+xA − xB + xC − xD )
a3 = 0. 25 · (−xA − xB + xC + xD )
b1 = 0. 25 · (−yA + yB + yC − yD )
b2 = 0. 25 · (+yA − yB + yC − yD )
b3 = 0. 25 · (−yA − yB + yC + yD ) .

Therewith, considering the linearly fluctuating geometry, the Jacobian matrix

of the element
" 4 # " #
Σi=1 φi ,ξ xi Σ4i=1 φi ,ξ yi a1 + a2 · η b 1 + b 2 · η
J= =
Σ4i=1 φi ,η xi Σ4i=1 φi ,η yi a3 + a2 · ξ b 3 + b 2 · ξ
depends just on the quantities ai , bi . The inversion of the Jacobian matrix yields
" # " #
i11 i12 1 b3 + b2 · ξ −b1 − b2 · η
−1
J = = ·
i21 i22 det J −a3 − a2 · ξ a1 + a2 · η
whereat the determinant is given with

det J = (a3 + a2 · ξ) · (b1 + b2 · η) − (b1 + b2 · η) · (a3 + a2 · ξ) .

For simplicity the entries of the inverse are named by ikl , cf. Section 12.1.
Since the Jacobian matrix is developed with respect to the transformation of
derivatives
 ∂    ∂ 
∂x i11 i12
 =  ∂ξ 
∂ ∂
∂y
i21 i22 ∂η

it can be applied to describe the nodal degrees of freedom w,x and w,y by means
of the unknowns a12 of the general polynomial (12.6). The complete Ψ̃–matrix
is represented on the next page, whereat the rows have to be evaluated with
respect to the local coordinates of the respective node.
 228

v = Ψ̃12 · a ,

vT = [ w w,x w,y ]A [ w w,x w,y ]B [ w w,x w,y ]C [ w w,x w,y ]D ,

 

1 −1 −1 1 1 1 −1 −1 −1 −1 1 1
0 i11 i12 3i11 2i11 + i12 i11 + 2i12 3i12
 

−2i11 −i11 − i12 −2i12 −3i11 − i12 −i11 − 3i12

0 i21 i22 −2i21 −i21 − i22 −2i22 3i21 2i21 + i22 i21 + 2i22 3i22 −3i21 − i22 −i21 − 3i22
 
 

1 1 1 1
 

1 1 −1 −1 −1 −1 −1 −1
 
  node A

0 i11 i12 2i11 3i12

 

−i11 + i12 −2i12 3i11 −2i11 + i12 i11 − 2i12 −3i11 + i12 −i11 + 3i12
 

0 i21 i22 2i21 3i22



−i21 + i22 −2i22 3i21 −2i21 + i22 i21 − 2i22 −3i21 + i22 −i21 + 3i22

 

Ψ̃12
 

1 1 1 1 1 1 1 1 1 1 1 1
 
 


0 i11 i12 2i11 i11 + i12 2i12 3i11 2i11 + i12 i11 + 2i12 +3i12 3i11 + i12 i11 + 3i12
 node B
= 
 

0 i21 i22 2i21 i21 + i22 2i22 3i21 2i21 + i22 i21 + 2i22 3i22 3i21 + i22 i21 + 3i22
 
 
 

1 1 1 1
 

1 −1 1 −1 −1 −1 −1 −1
 


0 i11 i12 2i12 3i12

 node C
 

−2i11 i11 − i12 3i11 −2i11 + i12 i11 − 2i12 3i11 − i12 i11 − 3i12
 

0 i21 i22 2i22 3i22



node D


−2i21 i21 − i22 3i21 −2i21 + i22 i21 − 2i22 3i21 − i22 i21 − 3i22
 
 
12 Isoparametric Elements
HYBRID QUADRILATERAL ELEMENTS
13 Hybrid Finite Elements

At finite element formulations employing the PvD the equilibrium condition is

satisfied in a weak sense, whereas the other governing equations are fulfilled
exactly by introducing these equations into the work equation of the PvD, cf.
Section 2.1. As a consequence, the displacements persist as primary variables,
whereas the stresses or stress resultants, which are the more important state
variables for the structural design, are only computed by the subsequent stress
analysis. Since the displacements have to be differentiated with respect to ki-
nematics, this procedure yields stress distributions, which are generally more
roughened compared to the approach of the displacements.
More general approaches deal with displacement and stress variables, whereat
the Principle of virtual Displacements and the Principle of virtual Forces are
applied to fulfill the equilibrium and the condition of deformation in a weak
sense. This idea leads to the Mixed Finite Element Method and the Hybrid
Finite Element Method, which are applied in the following sections to develop
finite elements for membranes and plates. The most general formulation is
established by Hu and Washizu [44, 102] and thus known as the Hu–Washizu
Principle. Generalized variational principles are applied to finite elements very
early by Pian and Tong [80], Greene et al. [41] and McLay [67].

Mixed Finite Element Method

The Mixed Finite Element Method deals with the equilibrium applying the
PvD incorporating the stresses and kinematics applying the PvF incorporating
displacements u and strains ε. Replacing the strains ε at the PvF by the ma-
terial equation yields the condition of deformation in weak form, described by
displacement and stress variables respectively. The integral representations of
both principles PvD and PvF to discretize the equilibrium and the condition
of deformation is named mixed formulation of virtual work.
The idea of Mixed Principles of Work is originally developed as variation of a
potential and first published by Hellinger [43]. Prange [82] describes an equiva-
lent formulation for frame structures transferring the force–based method and
the displacement–based method into each other by a Legendre transformation.
Later on Reissner [87] develops the today’s common formulation. In literature,
the mixed formulation often is named as Hellinger–Reissner Principle. The
mixed formulation of the finite element method is of advantage, if nonlineari-
ties have to be taken into account, cf. Kröplin and Dinkler [58, 59, 60], but is
restricted to structures without structural bifurcations.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_13
232 13 Hybrid Finite Elements

Applying the mixed formulation of the virtual work the displacements and the
stresses are employed both as primary variables. Thus shape functions are to
be chosen for the displacement variables as well as for the stress variables.
Thereby it is advantageous that the stress variables are directly computed and
no subsequent analysis is needed. Moreover, in particular regarding bending
problems, approaches of less polynomial order can be chosen concerning the
displacements likewise fulfilling the convergence criteria. A drawback might re-
sult from a higher number of degrees of freedom at every node.

Hybrid Finite Element Method

The disadvantages of mixed elements can be avoided, if the stress variables are
still defined inside the element, but are eliminated at the element level after the
discretization. This procedure leads to an element dealing with displacement
variables at the system level. Pian and Tong [80] and Pian and Sumihara [81]
have been the first to propose this method as well as to apply it for the develop-
ment of plane stress elements. Noor and Peters [74, 75] develop hybrid elements
to investigate the nonlinear behavior of shells. The following is substantial to
the procedure:

• Parts of the shape functions as well as of the related degrees of freedom

are defined inside the element. This part of the approach does not fulfill
all conditions of continuity at the element interfaces.
• Solely at the element interfaces additional degrees of freedom are defined.
Thereby further conditions of continuity can be fulfilled either weakly or
exactly related to the entire interface length.

x A interface
y z
a

B
d element domain b
D

c element boundary

C
Fig. 13-1 Element domain and interface

A 2–dimensional element is taken as an example to clarify the partioning of

the element into the element domain, wherein the virtual work is discretized
employing continuous shape functions, and the element interface, whereat the
conditions of continuity are to be fulfilled, cf. Figure 13-1.
 233

The classification into element domain and element interface results in multiple
possibilities at formulating and discretizing the work equations. Displacement
as well as stress fields are defined relating to the element domain and to the
element interfaces, therewith the conditions of equilibrium and of deformation
are to be fulfilled. Due to the different parts of the element, which are adressed
as domain and as interfaces, the formulation is named as hybrid element me-
thod. In total three different hybrid formulations are defined in literature, cf.
Figure 13-2:
1. HMM The hybrid–mixed model, whereby conditions of equilibrium as
well as of deformation are weakly fulfilled in the element domain. The
hybrid–mixed model comprises displacement as well as stress variables as
variables inside the element.
2. HSM The hybrid–stress model, whereby the conditions of equilibrium
are exactly fulfilled inside the element domain. This is equivalent to the
force method in structural analysis.
3. HDM The hybrid–displacement model, whereby the conditions of defor-
mation are exactly fulfilled inside the element domain. This is equivalent
to the displacement method in structural analysis.
The denotation goes by the degrees of freedom defined inside the element do-
main. Concerning the element interfaces normally displacement variables are
defined as primary variables of the system.

Finite Element Methods

✥ ❵❵❵
✥✥✥ ❵❵❵
✥✥✥
✥ ❵❵
hybrid elements mixed elements displacement elements

hybrid–mixed equilibrium and def. condition weakly fulfilled

model elem. DOF: displacement and stress variables

hybrid–stress equilibrium exactly, def. condition weakly fulfilled

model elem. DOF: displacement and stress variables

hybrid–displacement equilibrium weakly, kinematics exactly fulfilled

model elem. DOF: displacement variables

Fig. 13-2 Overview of the different finite element methods

234 13 Hybrid Finite Elements

At first the hybrid–mixed model HMM shall be illustrated by means of exam-

ples of a bar and a rectangular plane stress structure. These both examples
of use demonstrate the principles of the procedure. Due to C0 –conformity of
the formulations the conditions at the interface are fulfilled already by the
approach. Thus advantages at the choice of the shape functions to describe
the behavior inside the element domain as well as at the interfaces exist in
comparison to the model employing only displacements as primary variables,
which may even increase at investigating nonlinear load–carrying behavior.
Global system variables are the displacement variables at the element interfaces.
Regarding bars and plane stress structures, these are the displacements at the
nodes of the system.
Further applications are Euler–Bernoulli beams and Kirchhoff plates, whereby
C1 –conformity is to be ensured, that is, concerning mixed models with approa-
ches related to w and M , strongly fulfilled for w and weakly fulfilled for ϕ.
Concerning hybrid models additional arrangements are to be set up to fulfill
all conditions at the interface.
Hereinafter the principle procedure at deriving either mixed or hybrid–mixed
formulations is developed relating to the example of the bar.

13.1 Mixed Formulation of Governing Equations

In Section 2.1 the PvD is derived from the work equation and is shown to be
equivalent to the differential equation of equilibrium. In the succeeding sections
this procedure is also chosen to derive the mixed principle of work applying the
PvD and the PvF to the bar depicted in Figure 13-3.

p(x)
He
c, EA

x, u
l

Fig. 13-3 Bar loaded by constantly distributed and concentrated external action

The governing equations are given already in Section 1.3.1, whereby hereinafter
the condition of equilibrium is extended by a tangential bedding force Fc . The
material equations describe the bedding force with the bedding modulus c and
the impressed strains εT from heating by T0 with the coefficient of thermal
expansion αT .
13.1 Mixed Formulation of Governing Equations 235

a) Kinematics

u,x − ε = 0 in the domain, (13.1)

u − ue = 0 at fixed boundary: ue = 0 .

b) Equilibrium

−fc + N,x + p(x) = 0 in the domain, (13.2)

e e
N −H =0 at free boundary: H = 0 .

c) Material Equations

fc − c · u = 0 , (13.3)
εel − N/EA = 0 , (13.4)
εT − αT · T0 = 0 . (13.5)

Condition of Deformation
Applying the formulation employing the displacements as primary variables,
the material equations as well as the governing equations related to kinematics
are introduced into the PvD and hence into the condition of equilibrium. Re-
garding the mixed formulation only the material equation related to bedding
is introduced into the condition of equilibrium.
The condition, which enforces the equality of strains due to heating and elasti-
city and due to kinematics, is named condition of deformation:

ε = εel + εT . (13.6)

Introducing the kinematics, Equation (13.1), and the material equations (13.4,
13.5) yield the condition of deformation in the domain, which is employed
to connect the displacements with the stress variables, and the displacement
boundary conditions:
1
u,x − · N − αT · T0 = 0 and u − ue = 0 . (13.7)
EA
Equilibrium (13.2) and the condition of deformation (13.7) are summarized
applying the matrix vector notation:
" # " # " # " #
−c ∂x u p 0
1 · + = . (13.8)
∂x − EA N −αT · T0 0
236 13 Hybrid Finite Elements

13.2 Mixed Formulation of Work Equations

To derive the finite element method in mixed formulation Equation (13.8) as
well as Equation (13.2) and Equation (13.6) are formulated equivalently em-
ploying the principle of virtual work.

The Principle of Virtual Displacements – PvD

The Principle of virtual Displacements represents a weak formulation of the
conditions of equilibrium. Thereby the real forces perform work on the virtual
displacements
Z ℓ
(−δu · Fc − δε · N + δu · p ) dx + [ δu · H e ]bound. = 0 .
0

If the virtual strains fulfill kinematics δu,x − δε = 0, integration by parts of the

second term yields
Z ℓ
δu · (−Fc + N,x + p ) dx − [ δu · (N − H e ) ]bound. = 0 .
0

It becomes clear, that the PvD is equivalent to the conditions of equilibrium in

the domain and at the boundary.
Introducing the material equation related to the bedding force into the PvD
yields the formulation, which bases the discretization of the finite element me-
thod
Z ℓ Z ℓ Z ℓ
− δu · c · u dx − δu,x · N dx + δu · p dx + [ δu · H e ]bound. = 0 . (13.9)
0 0 0

This formulation of the PvD completely corresponds to the displacement–based

finite element formulation. Only by employing the Principle of virtual Forces to
formulate the condition of deformation in a weak representation a completely
independent method arises.

The Principle of Virtual Forces – PvF

Along the lines of the Principle of virtual Displacements the Principle of virtual
Forces is a weak representation of the conditions of deformation related to the
domain as well as to the boundary, cf. Section 2.2. Regarding the PvF the
virtual longitudinal force performs work on the real strains and displacements
Z ℓ
− δN · (u,x − ε) dx + [ δN · (u − ue ) ]bound. = 0 .
0
13.2 Mixed Formulation of Work Equations 237

The internal work terms get a negative sign here, since the internal virtual force
variables and the strains act against each other, cf. Equation (13.9). Replacing
the real strains by the material equation, yields in a first step
Z ℓ
N
− δN · (u,x − − αT · T0 ) dx + [ δN · (u − ue ) ]bound. = 0 .
0 | EA {z }
−ε

Regarding the integral of the internal work the condition of deformation (13.7)
can be recognized now, including the displacement boundary condition in weak
form. The request to exactly fulfill the displacement boundary condition by the
approach itself yields the work equation of the PvF in the following represen-
tation, which is applied for the discretization:
Z ℓ
N
− δN · (u,x − − αT · T0 ) dx = 0 . (13.10)
0 EA
Concerning the work equations of the PvD and of the PvF, the displacements
and the forces are independent variables. Thus for both independent approaches
can be chosen in the context of finite element methods. Nonetheless, the boun-
dary conditions at the boundary of the structure and the interface conditions
between two elements are fulfilled in a different manner.

13.2.1 Shape Functions, Element Matrix, Load Vector

Applying the convergence criteria analogously to Section 2.4.2 to mixed displa-
cement and force variables formulation yields linear shape functions to describe
u(x) and constant shape functions to describe N (x), following the criterion of
at least constant strains. The same is valid for the virtual displacements as well
as for the virtual forces.
The criterion of conformity requires linear shape functions to describe u(x).
To approach N (x) constant shape functions are adequate, since, concerning N ,
system boundary conditions as well as element interface conditions are weakly
fulfilled by the work equation. Thus an element is derived, which incorporates
different degrees of freedom relating to the element nodes and induces special
indexing and control procedures at programming plane stress elements.
In general linear shape functions are chosen to describe the longitudinal forces
N (x), too. This might be advantageous at investigating geometric as well as
physical nonlinearities.

Choosing linear shape functions to describe the displacements and the forces,
already known from the formulation solely employing displacements as primary
238 13 Hybrid Finite Elements

variables, cf. Section 2.3.2 and Figure 13-4, u(x) and N (x) may be scaled to
nodal values as follows
u(x) = φA (x) · uA + φB (x) · uB ,
N (x) = φA (x) · NA + φB (x) · NB .

x x

φA =1-x/l φB =x/l

Fig. 13-4 Linear shape functions

Concerning the virtual states δu, δN , the same shape functions are chosen
δu(x) = φA (x) · δuA + φB (x) · δuB ,
δN (x) = φA (x) · δNA + φB (x) · δNB .
Employing linear shape functions to describe N (x) according to the nodal de-
grees of freedom NA und NB , the conditions at the element interfaces Ni =
Ni+1 result in the fact, that concentrated external loadings must not act at
element interface nodes but are only reasonable at system boundaries.

13.2.2 Matrix Notation of the Work Equations

To derive a general matrix representation the notation following Section 2.3.3 is
applied, whereby the contents of the matrices related to bars may be adopted,
too. To distinguish the discrete displacement variables from the stress variables,
the indices ( )s relating to stresses or stress resultants and ( )v relating to
displacements are introduced. Applying the matrix notation of Table 13.1 the
work equations (13.20) can be generally represented relating to an arbitrary
element in mixed formulation.
At element level the work performed at the element interfaces as well as the
work performed at the system boundaries are not taken into account. Concer-
ning the PvD it follows
Z Z Z
− δWd = δvT · { ΩTv C Ωv dx · v + (ΩTv DT ) Ωs dx · s − ΩTv p dx}. (13.11)

Accordingly, the work performed on virtual stresses PvK is formulated to

Z Z Z
− δWσ = δsT ·{ ΩTs (D Ωv ) dx·v− ΩTs E−1 Ωs dx·s− ΩTs ǫT dx}. (13.12)
13.2 Mixed Formulation of Work Equations 239

in general related to the bar

displacements u = Ωv · v u = [ u(x) ]
Ωv = [ φA φB ]
vT = [ uA uB ]
δu = Ωv · δv δvT = [ δuA δuB ]

stresses σ = Ωs · s σ = [ N (x) ]
Ωs = [ φA φB ]
sT = [ NA NB ]
δσ = Ωs · δs δsT = [ δNA δNB ]

kinematics ǫ =D·u D = [ ∂x ]
strains ǫ = ǫel + ǫT ǫ = [ǫ]
 1 
material equation ǫel = E−1 · σ E−1 = EA
strains due to heating ǫT = [ αT · T0 ]
bedding C = [c]
external action p = [ px ]

Table 13.1 Matrix notation of the governing equations of bars

Along the lines of the displacement–based formulation the following abbrevia-

tions may be introduced

Bv = D · Ωv ,
BTv = (D · Ωv )T = ΩTv · DT ,

and may be applied in the same meaning. The work equations (13.11) and
(13.12) may be summarized applying the matrix notation, line by line extrac-
ting the virtual displacements and the virtual stresses:
Z " T # " # Z " T #
 T T Ωv C Ωv ΩTv DT Ωs v Ωv p
− δW = δv δs { T T −1
dx · − T dx} .
Ωs D Ωv −Ωs E Ωs s Ω s ǫT
| {z } | {z }
element matrix load vector
Since the heat conduction is not included as part of the work equations, the
strains εT are considered as external action in the load vector.
240 13 Hybrid Finite Elements

Element Matrix
Concerning a generally applicable notation the representation arranging stres-
ses behind displacements, is convenient. However, applying the notation to
special element formulations, the arrangement with respect to nodal degrees of
freedom is more reasonable. Therewith the element matrix related to the bar
and employing linear shape functions follows to:
   
δuA : c·ℓ 1 c·ℓ 1 uA
 3 − 2 6
− 2   
   
δNA :  1 ℓ 1 ℓ   N 
 − 2 − 3EA 2 − 6EA   A 
 · . (13.13)
 c·ℓ 1 c·ℓ 1   
δuB :  6

2 3 2

 
 u B 

   
1 ℓ 1 ℓ
δNB : − 2 − 6EA 2 − 3EA NB

In contrast to the displacement–based formulation the matrix cannot be un-

derstood as a stiffness matrix.
Load Vector
Regarding element by element constantly distributed external actions p as well
as constant heating T0 the load vector results from the integral of external work
 
ℓ/2 0
" # " #
fA  0 ℓ/2  p
 
= · . (13.14)
fB  ℓ/2 0  εT
0 ℓ/2
Linearly varying actions may be considered employing linear shape functions
and related nodal values.
13.2.3 Test
In order to evaluate the quality of the mixed formulation the following test
is performed. Figure 13-5 shows the bar as given in Section 1.3.2 loaded by
constantly distributed external actions.
p(x)
EA = 6 kN
EA
x,u l = 5m
l p(x) = 1 N/m

1 2 3 4 5

Fig. 13-5 Bar – geometry, loading, and discretization with four elements
13.2 Mixed Formulation of Work Equations 241

Employing linear shape functions for displacements and stresses the discreti-
zations with four, five, six, and seven elements yield the study of convergence
represented in Table 13.2, whereat the boundary condition of the force N |ℓ is
fulfilled in a weak sense.

Table 13.2 Bar – convergence of displacements and stresses

number of nodes 1 2 3 4 5 6 7 8

N [ kN ] 5. 00 3. 75 2. 50 1. 25 0. 00
u [m] 0. 00 0. 95 1. 56 1. 99 2. 083
N [ kN ] 4. 90 4. 10 2. 90 2. 10 0. 90 0. 10
u [m] 0. 00 0. 77 1. 34 1. 76 2. 02 2. 083
N [ kN ] 5. 00 4. 16 3. 33 2. 50 1. 66 0. 83 0. 00
u [m] 0. 00 0. 65 1. 15 1. 58 1. 85 2. 04 2. 083
N [ kN ] 4. 95 4. 34 3. 52 2. 91 2. 09 1. 48 0. 66 0. 05
u [m] 0. 00 0. 56 1. 02 1. 41 1. 71 1. 92 2. 05 2. 083

The displacements at the nodes become partly the same values as at employing
the displacement-based element, whereby the displacements by chance partly
match the exact values.
The stresses approach the linear course of the exact solution quite well, depen-
ding on the number of elements. The stresses match the exact values in the
case of an equal number of elements but oscillate around the exact solution in
the case of an uneven number of elements, since the condition of deformation
Z
N
δN (u,x − ) dx = 0
EA
is fulfilled in a weak sense. This indicates some inconsistencies at the element
formulation. Nonetheless the element converges against the exact solution.

Oscillations may be avoided, if the approaches for the displacements and the
stresses are adapted to each other. This means, that a linear approach of the
displacements needs a constant approach for the stresses, and a quadratic ap-
proach of the displacements needs a linear approach of the stresses. In these
cases the results have the same quality as at employing the displacement–based
elements. Nonetheless, if the order of the shape functions for N is lower than the
242 13 Hybrid Finite Elements

order of the shape functions for the displacements, the number of element no-
des does not coincide with the number of stress variables. This means, that the
stresses can be dealt with as internal variables as present in the displacement–
based formulation.

The disadvantages concerning the bar element in the mixed formulation may
be summarized as follows:

• The mixed formulation has got twice as much unknowns in comparison to

the displacement–based element, without any improvement of the results.
• If N is chosen as nodal variable, an external nodal action is impossible,
since this would effect a jump with respect to the Force N .
• If stresses are nodal degrees of freedom, due to the equilibrium conditions
two elements may be connected at a node at most, because more elements
would generate different stresses at the node with respect to each element.
Thus trusses can not be analysed applying the mixed element.
• If the approach of the stresses is of lower order than the approach of
the displacements, the degrees of freedom related to the stresses are only
element variables, without any connection to neighboring elements. In this
case the element is comparable to the displacement–based formulation.

In order to overcome the disadvantages of the mixed formulation of the work

equations, the hybrid formulation has been developed as discussed in the fol-
lowing sections.

13.3 Hybrid Discretization of Work Equations

The procedure to discretize the work equations employing hybrid elements is
developed regarding the example of the bar, see Figure 13-6.
HeA p(x) HeB

A B
x

Fig. 13-6 Element of a bar loaded by distributed and nodal external actions

The Principle of virtual work in mixed displacement–force formulation is given

with PvD and PvF, see Section 13.2. Applying the work equation related to
the PvD (13.9) the equilibrium is weakly fulfilled in the domain as well as at
13.3 Hybrid Discretization of Work Equations 243

the system boundary. The interface condition related to the forces is fulfilled
strongly or weakly depending on the approach. Applying the work equation
related to the PvF (13.10) the conditions of deformation are weakly fulfilled
with respect to the domain but the boundary as well as the interface conditions
are strongly fulfilled concerning the displacements.
Applying hybrid elements both types of work equations are discretized in the
domain employing shape functions without fulfillment of the element interface
conditions. The element interface conditions related to kinematics and equili-
brium are separately looked at.

13.3.1 Work Equation at Element Level

Regarding hybrid elements the element domain and the interface are to be
distinguished. Applying the method of sections the segmentation to element
domain and interface is possible, thus differing conditions may be formulated
for both parts. Figure 13-7 shows the bar element with the element domain
a − b and the interfaces A and B, which are identical to the system nodes,
where the transition to the neighboring elements is to be ensured.

HeA p(x) HeB

A Na a b Nb B
u E(x), NE(x)

Fig. 13-7 State variables at the cut cleanly bar element

The cut cleanly of the interfaces, cf. Figure 13-7, yields:

• the element domain ( )E between a and b,

• the element boundaries a and b,
• the interfaces A and B,
• the cut cleanly longitudinal forces Na , Nb .

At employing hybrid–mixed elements the work related to the element domain

is described by longitudinal forces and displacements. However, at the element
interfaces – these are the system nodes A and B – only displacements are
defined.
Relating to the element the work equations may be specified according to the
indices ( )E for the element domain, for the element boundaries a, b and for the
e e
interfaces A, B. The work performed by the external actions HA , HB is taken
into account, when the systems’ equations are assembled.
244 13 Hybrid Finite Elements

Applying the PvD to the element domain including element boundaries yields
Z
δWd = {−δεE · NE − δuE · c · uE + δuE · p(x)} dx |domain
+ [−δua · Na + δub · Nb ]el.bound. = 0 . (13.15)
At the element intersection A between the elements j − 1, j and at the ele-
ment intersection B between the elements j, j + 1 the PvD describes the work
performed by the forces Na , Nb on the virtual displacements δuA , δuB of the
intersections, which are independent of δua , δub so far:
h i
δWd, A = δuA · −Nbj−1 + Naj + HA e
= 0, (13.16)
h i
δWd, B = δuB · −Nbj + Naj+1 + HB e
= 0. (13.17)

Equations (13.16) and (13.17) fulfill the equlibrium at the system level in a
weak sense, if the forces Na , Nb are replaced by the variables at the element
domain of the neighboring elements employing Equation (13.15). Due to the
elimination of the forces Na , Nb no system variables exist, which are related
to longitudinal forces at the interfaces. Thus, at the element level, introducing
kinematics into the PvD yields
Z
− δWd = {δu,x E · NE + δuE · c · uE − δuE · p(x)} dx 6= 0 , (13.18)

Applying the PvF kinematics is given in a weak formulation with:

Z
1
δWσ = {−δNE · εE + δNE · · NE + δNE · αT · T0 } dx |domain
EA
− δNa · [ua − uA ]trans.A−a + δNb · [ub − uB ]trans.B−b = 0 . (13.19)
e
Gaps ∆u are possible between element boundary and interface, but are not
taken into account here. Strong fulfillment of the kinematic conditions at the
interface related to the real as well as to the virtual displacements between
element boundaries and interfaces,
uA = ua , uB = ub , δuA = δua , δuB = δub ,
yields in a first step
Z
1
− δWσ = {δNE · εE − δNE · · NE − δNE · αT · T0 } dx = 0
EA
and introducing kinematics into the work equation succeeds in
Z
1
− δWσ = {δNE · u,x E − δNE · · NE − δNE · αT · T0 } dx = 0 . (13.20)
EA
13.3 Hybrid Discretization of Work Equations 245

Applying this formulation of the virtual work, the element interface conditions
related to the longitudinal forces can only be fulfilled weakly, since the longi-
tudinal forces are defined solely at the element level and not at the interfaces,
cf. Equations (13.16) and (13.17). Thus no direct coupling to the neighboring
elements is present concerning N and δN , hence the PvF is to be arranged at
the element level only.

13.3.2 Stiffness Matrix and Load Vector of the Hybrid Element

The discretization of the work equation is related to the element domain, em-
ploying linear shape functions as for the mixed formulation, cf. Section 13.2.
Applying matrix notation the integrals of the work equations related to the
domain (13.18) and (13.20) can be given as follows:
Z " ΩT C Ω ΩTv DT Ωs
# " # Z " #
δv : v v v ΩTv p
dx − dx (13.21)
δsE : ΩTs D Ωv −ΩTs E−1 Ωs sE ΩTs εT
with the degrees of freedom, slightly changed in comparison to Section 13.2,
v T = [ uA uB ] at system level
and sTE = [ Na Nb ] at element level.
The integrals related to Equation (13.21) are directly comparable with the
mixed formulation corresponding to Equation (13.13). Thereby the element
matrix and the load vector are already derived, see Equation (13.14).
At element level the PvF is valid
Z
−δWσ = δsTE {ΩTs D Ωv · v − ΩTs E−1 Ωs · sE − ΩTs εT } dx = 0 . (13.22)

Applying Equation (13.22) the longitudinal forces sE can be described depen-

ding on the nodal displacements v
Z Z
sE = { Ωs E Ωs dx } · { [ ΩTs D Ωv · v − ΩTs εT ] dx } .
T −1 −1
(13.23)

In this form the equation can be used to eliminate sE from the equation apply-
ing the PvD. Out of it follows a representation, only incorporating the degrees
of freedom at the system level v:
Z
T

− δWd = δv ΩTv C Ωv dx
Z Z Z

+ ΩTv DT Ωs dx · ΩTs E−1 Ωs dx }−1 · { [ ΩTs D Ωv · v − ΩTs εT ] dx
Z
− δvT ΩTv p dx . (13.24)
246 13 Hybrid Finite Elements

According to programming it might be advantageous to apply matrix notati-

on to Equation (13.21) and to the elimination procedure subject to Equation
(13.23), whereat the virtual variables are extracted again:
" # " # " #
δv : Hvv Hvs v fp 6= 0 at element level,
· − dx (13.25)
δsE : Hsv −F sE fT = 0 at element level.

The following abbreviations are introduced:

Z
Hvv = ΩTv C Ωv dx ,
Z
Hvs = ΩTv DT Ωs dx = HTsv ,
Z
F= ΩTs E−1 Ωs dx ,
Z
fp = ΩTv p dx ,
Z
fT = ΩTs εT dx .

Following the integration of the virtual work terms the computation of sE is

numerically performed solving

sE = F−1 [ Hsv · v − fT ] . (13.26)

Introducing sE into the PvD yields

−δWd = δvT { [ Hvv + HTsv F−1 Hsv ] · v − fp − HTsv F−1 fT } 6= 0 . (13.27)

The expression in the square brackets of Equation (13.27) corresponds to the

stiffness matrix of the displacement–based method.

Linear shape functions

Applying linear shape functions, cf. Section 13.2, and employing the nodal
degrees of freedom

v T = [ uA uB ] at system level,
and sTE = [ Na Nb ] at element level,
13.3 Hybrid Discretization of Work Equations 247

which satisfy the continuity of the displacements but not of the stresses, gives
the element matrix according to Equation (13.21)
Hvv HTsv
   
cℓ cℓ
δuA : 3 6
− 12 − 12 uA
   
δuB :
 cℓ cℓ 1 1   
 6 3 2 2
  uB 
   
  · . (13.28)
 −1 1 ℓ ℓ   
δNa :  2 2
− 3EA − 6EA   Na 
   
ℓ ℓ
δNb : − 12 1
2
− 6EA − 3EA Nb

Hsv −F
The elimination of the longitudinal forces sE at the element level is performed
following Equations (13.25) and (13.26) with
" # " # " # " # " #
1 1
ℓ 3 6 Na − 12 12 uA 1 εT ℓ
− · 1 1 · + · = . (13.29)
EA
6 3
Nb − 12 12 uB 1 2
Solving the equations (13.29) with respect to the stresses gives in a first step
" # " # " # " # " #
Na EA 4 −2 − 12 12 uA 1 εT ℓ
= · ·{ · − }
Nb ℓ −2 4 − 21 12 uB 1 2
and after multiplication the final result
" # " # " # " #
Na EA −1 1 uA 1
= · · − EA εT . (13.30)
Nb ℓ −1 1 uB 1
Thus comparable to the element employing displacements only as primary va-
riables and linear shape functions, it follows Na = Nb . Hence, concerning N , a
constant approach is possible, leading to the same result here.

Introducing Equation (13.30) into the PvD, the stiffness matrix follows, accor-
ding to Equation (13.27), to
" # " # " #
1 1
T 3 6 EA 1 −1 uA
δv ·K·v = [ δuA δuB ]·{c·ℓ 1 1 + ℓ }· . (13.31)
6 3
−1 1 uB
In this case, the stiffness matrix of the hybrid–mixed method is identical by
chance to the matrix according to displacement–based elements, cf. Section
2.3.1, and thus leads to the same results applying the element to structural
analysis. Nonetheless, the hybrid–mixed method gives more freedom at the
choice of shape functions for displacements and stresses inside the element.
14 Hybrid–Mixed Plane Stress Elements

The advantages of the hybrid–mixed formulation become more obvious at trans-

ferring the concept to plane stress structures. The freedom at the choice of shape
functions and at the fulfillment of conditions of continuity is visible only now.
Once again Figure 14-1 distinguishes between element domain and interface.
At the element domain the shape functions to describe the displacements and
the stresses can be chosen. So far, they do not have to fulfill the conditions at
the element interfaces, due to the fact that the conditions of continuity only
have to be fulfilled at the interfaces themselves.

x A interface
y z
a

B
d element domain b
D

c element boundary

Fig. 14-1 Hybrid–mixed plane stress element

14.1 Mixed Principle of Work for Plane Stress Structures

Succeeding, the governing equations as well as the work equations concerning
PvD and PvF are prepared by analogy with the bar. To discretize the work
equations a rectangular element employing linear shape functions is chosen.
The matrix notation comprises the same symbols as already given in Section
13.3.

14.1.1 Governing Equations

The governing equations are already represented in matrix notation, cf. Section
5.1. Thus, incorporating the matrix symbols into kinematics, it follows
ǫ = Du · u ,

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_14
250 14 Hybrid–Mixed Plane Stress Elements
   
εxx ∂x 0  
    u
 yy   0
ε = ∂y  · . (14.1)
v
2εxy ∂y ∂x
The conditions of equilibrium can be given with
Dσ · σ + p = 0 ,
 
" # σxx " #
∂x 0 ∂y   px
·  σyy  + = 0, (14.2)
0 ∂y ∂x py
σxy
as well as the material equations according to the plane stress state with
σ = E · ǫ,
     
σxx 1 ν εxx
  E    
 σyy  =  ν 1  ·  εyy  , (14.3)
1 − ν2 1
σxy 2 (1 − ν) 2εxy
and in inverse representation
ǫ = E−1 · σ ,
     
εxx 1 −ν σxx
  1    
 εyy  =  −ν 1  ·  σyy  . (14.4)
E
2εxy 2(1 + ν) σxy
Expressions concerning heating or bedding, as given for the bar, are not consi-
dered here. By analogy with the bar and to eliminate the strains, the material
equation (14.4) is introduced into kinematics (14.1), what leads to the condition
of deformation:
Du · u − E−1 · σ = 0 . (14.5)
Here, the differential operator Du correlates with the differential operator Dσ ,
since stresses and strains are energetically conjugated variables. Summarizing
the governing equations, the conditions of equilibrium as well as of deformation
may lead to the matrix formulation
" # " # " #
0 Dσ u p
· = . (14.6)
Du E−1 σ 0
In addition to the governing equations, which are to be considered in the do-
main, the respective boundary conditions are to be taken into account. However
they are not especially addressed here, cf. Section 5.
14.1 Mixed Principle of Work for Plane Stress Structures 251

14.1.2 Work Equations in Mixed Formulation

By analogy with the bar the governing equations are formulated equivalently
applying the Principle of virtual Work now.

The Principle of Virtual Displacements

The Principle of virtual Displacements is equivalent to the conditions of equi-
librium (14.2), if the virtual displacements fulfill kinematics. As a start, the
following is valid
Z
δWd = {−δεxx · σxx − δεyy · σyy − 2δεxy · σxy + δu · px + δv · py } dA
A Z
+ {δun σn + δus σt } ds = 0 .
s

At the boundary ds the indices n and t indicate the normal and the tangential
direction respectively. Replacing the virtual strains by kinematics yields
Z
δWd = {−δu,x · σxx − δv,y · σyy − (δu,y + δv,x ) · σxy + δu · px + δv · py } dA
A Z
+ {δun σn + δut σt } ds = 0
s

or in matrix notation
Z Z Z
T T
− δWd = ( δu · D ) · σ dA − δu · p dA − δuT · σ e ds = 0 .
T
(14.7)
A A s

At each boundary, the stress vector σ e only comprises the tangential and the
normal component, which are related to the displacements un and ut . The
brackets in Equation (14.7) imply to which state variable the differential ope-
rator D is to be applied.

The Principle of Virtual Forces

The Principle of virtual Forces is equivalent to the condition of deformation.
After have been introducing the material equations the weak form of the con-
dition of deformation results in
Z
1 ν ν 1
δWσ = {− δσxx · (u,x − σxx + σyy ) − δσyy · (v,y + σxx − σyy )
A E E E E
2(1 + ν)
− δσxy · (u,y + v,x − σxy } dA
E
Z
+ {δσn · ( un − uen ) + δσt · ( ut − uet )} ds = 0 .
s
252 14 Hybrid–Mixed Plane Stress Elements

Requiring the displacement boundary conditions as well as the displacement

continuity conditions to be fulfilled concerning the real displacements by the
shape functions with un − uen = 0 and ut − uet = 0, yields
Z
1 ν ν 1
δWσ = {− δσxx · (u,x − σxx + σyy ) − δσyy · (v,y + σxx − σyy )
A E E E E
2(1 + ν)
− δσxy · (u,y + v,x − σxy } dA = 0 .
E

Thus the conditions of deformation are fulfilled weakly with the PvF in the
domain and are fulfilled strongly at the boundary. It follows in matrix notation
Z
− δWσ = δσ T · ( D · u − E−1 · σ ) dA = 0 . (14.8)
A

Both principles of virtual work may be summarized by analogy with Equation

(14.6) by matrix formulation
Z " # " #
 T T0 DT u
δu δσ · · dA
A D −E−1 σ
Z " # Z " #
 T T
 p  T T
 σe
= δu δσ · dA + δu δσ · ds . (14.9)
A 0 s 0

14.2 Work Equations of a Hybrid Plane Stress Element

Regarding hybrid elements the element domain is distinguished from the ele-
ment boundary and the interface. In the mixed formulation the virtual work is
arranged according to the Principle of virtual Work given by the PvD and the
PvF, see Section 14.1 – without contour integrals. In addition the virtual work
at the element boundaries and at the interfaces is to be described, in a first
step without fulfillment of the conditions of continuity between the elements.
Figure 14-2 identifies the stress variables σn , σt , that perform work at the ele-
ment boundary b–c and at the interface B–C, as an example.

With the PvD and the PvF the displacements and the stressess are identified
to be the variables of description in the element domain and at the interfaces.
Hence, concerning the following discretization, different procedures are possible
at choosing adequate shape functions and at fulfilling the element continuity
conditions.
14.2 Work Equations of a Hybrid Plane Stress Element 253

x A
y z

a
B
d b
σn
D σt
c σt
σn

C
Fig. 14-2 Normal and tangential stresses at element boundary and interface

14.2.1 Work Equations at Element Level

At element level the PvD comprises the same expressions as the mixed formu-
lation of the Principle of virtual Work following Equation (14.7). In addition
the work is to be considered, that the stresses perform on the virtual displa-
cements at the element boundaries as well as at the interfaces. Subsequently
the work equations are formulated for a rectangular element to particularly
illustrate the conditions of continuity related to the displacements between
the element boundaries and the respective interfaces. The Principle of virtual
Displacements yields
Z
δWd = {−δεxx · σxx − δεyy · σyy − 2δεxy · σxy + δu · px + δv · py } dA
A
Z
 
+ δun,a/b − δun,A/B · σn ds |transitionA/B−a/b
Z
 
+ δun,b/c − δun,B/C · σn ds |transitionB/C−b/c
Z
 
− δun,d/c − δun,D/C · σn ds |transitionD/C−d/c
Z
 
− δun,a/d − δun,A/D · σn ds |transitionA/D−a/d
Z
 
+ δut,a/b − δut,A/B · σt ds |transitionA/B−a/b
Z
 
+ δut,b/c − δut,B/C · σt ds |transitionB/C−b/c
Z
 
− δut,d/c − δut,D/C · σt ds |transitionD/C−d/c
Z
 
− δut,a/d − δut,A/D · σt ds |transitionA/D−a/d . (14.10)
254 14 Hybrid–Mixed Plane Stress Elements

Concerning the Principle of virtual Forces, in addition to the mixed principle

of work, the work performed by the virtual stresses on the real displacements
at the element boundaries and at the interfaces is to be considered, since the
related courses do not have to be chosen identically:
Z
1 ν
δWσxx = −δσxx · {εxx − · σxx + · σyy } dA
A Z E E
 
− δσn · un,a/d − un,A/D ds |transitionA/D−a/d
Z
 
− δσn · un,b/c − un,B/C ds |transitionB/C−b/c , (14.11)
Z
ν 1
δWσyy = −δσyy · {εyy + · σxx − · σyy } dA
A Z E E
 
− δσn · un,a/b − un,A/B ds |transitionA/B−a/b
Z
 
− δσn · un,d/c − un,D/C ds |transitionD/C−d/c , (14.12)
Z
2(1 + ν)
δWσxy = −δσxy · {2εxy + · σxy } dA
A E
Z
 
− δσt · ut,a/b − ut,A/B ds |transitionA/B−a/b
Z
 
− δσt · ut,b/c − ut,B/C ds |transitionB/C−b/c
Z
 
− δσt · ut,d/c − ut,D/C ds |transitionD/C−d/c
Z
 
− δσt · ut,a/d − ut,A/D ds |transitionA/D−a/d . (14.13)

14.2.2 Conditions of Deformation at the Element Interface

The condtions of continuity related to the displacements between two ele-
ments are fulfilled weakly by applying the contour integrals (14.11), (14.12)
and (14.13). Thus, the approaches to describe the displacements may differ
concerning the element boundary and the interface. The polynomial order of
the shape functions with respect to the displacements at the element boundary
is to be chosen equally or higher compared to the polynomial order at the in-
terface, in order to be able to transfer sufficient information from the element
domain to the interface. Tolerating small gaps between element boundary and
interface and strongly fulfilling the respective conditions of continuity related
14.2 Work Equations of a Hybrid Plane Stress Element 255

to the real displacements

un,interf ace = un,element boundary , ut,interf ace = ut,element boundary ,

results in disappearing of all boundary terms from PvF in Equations (14.11),

(14.12) and (14.13). At choosing the same conditions related to the virtual
displacements

δun,interf ace = δun,element boundary , δut,interf ace = δut,element boundary ,

the contour integrals with respect to PvD are omitted, too, cf. Equation (14.10).
After having introduced kinematics into the work equations concerning real as
well as virtual strains, the work performed at element level follows to
Z
δWd = {−δu,x · σxx − δv,y · σyy − (δu,y + δv,x ) · σxy

+ δu · px + δv · py } dA |domain 6= 0 , (14.14)
Z
1 ν
δWσxx = − δσxx · {u,x − · σxx + · σyy } dA |domain = 0 , (14.15)
E E
Z
ν 1
δWσyy = − δσyy · {v,y + · σxx − · σyy } dA |domain = 0 , (14.16)
E E
Z
2 (1 + ν )
δWσxy = − δσxy · {(u,y + v,x ) − · σxy } dA |domain = 0 . (14.17)
E
Thereby it is essential that the displacements are defined with respect to the
element domain and to the interface and that the stresses are only defined with
respect to the element domain.

14.2.3 Stress Conditions at the Element Interface

The element continuity conditions related to the stresses are only to be fulfilled
weakly, since the stresses are only defined in the element domain as wells as
at the section between element boundary and interface but are not defined at
the interface itself. Presuming the equality of the virtual displacements at the
interface between two elements j and j + 1 yields omitting of the respective
work integrals at system level. With
Z
δun · [ −σn |jb−c + σn |j+1
a−d ] ds = 0 ,
Z
δut · [ −σt |jb−c + σt |j+1
a−d ] ds = 0
256 14 Hybrid–Mixed Plane Stress Elements

the equilibrium is fulfilled weakly at the interface between the neighboring

elements j and j + 1 applying the PvD at system level. The same is valid
concerning the respective work expressions at the element boundary and the
interface. Thus, they do not have to be considered any further. When con-
centrated or line–shaped external loadings occur, the respective external work
expressions have to be additionally taken into account at the interface.
Hence, the real as well as the virtual stresses are not directly coupled to the
neighboring elements and the PvF is employed at element level only.

14.3 Element Stiffness Matrix

Developing the element stiffness matrix, the approach for the displacements
may be chosen as a general polynomial or as shape functions dealing with
physically meaningful degrees of freedom

u = Ωv · v , or u = Ψu · au = Ψu · G · v ,

whereat G represents the scaling matrix. The approach for the sresses is chosen
as a general polynomial with respect to the element coordinates

σ = Ψs · as .

To start with, the work equation representing the PvD comprises the degrees
of freedom related to the system level v as well as the degrees of freedom as at
element level
Z Z

−δWd = δvT (ΩTv DT ) Ψs dA · as − ΩTv p dA . (14.18)

Applying the PvF, Equations (14.15) to (14.17), the following is valid just as
well
Z
−δWσ = δaTs { ΨTs (D Ωv ) · v − ΨTs E−1 Ψs · as } dA = 0 . (14.19)

Thus the degrees of freedom as may be computed from the nodal displacements
v solving
Z Z
as = { ΨTs E−1 Ψs dA }−1 · { ΨTs (D Ωv ) dA} · v . (14.20)

The solution can be applied to eliminate as from the PvD equation.

14.4 Subsequent Stress Analysis 257

Concerning programming it is convenient to formulate the Equations (14.18)

and (14.19) as well as the subsequently following elimination process according
to Equation (14.20) in matrix notation. Extracting the virtual variables yields
 
−δW = δvT δaTs ·
Z " # " # Z " #
0 (ΩTv DT ) Ψs v ΩTv p
{ dA − dA} (14.21)
ΨTs (D Ωv ) −ΨTs E−1 Ψs as 0
comprising the degrees of freedom
v at system level and
as at element level
corresponding to the approaches according to Sections 14.5, 14.6 and 14.7. The
integrals according to Equation (14.21) may be numerically evaluated follow-
ing Gauss. Thus the degrees of freedom as can be numerically eliminated. By
analogy with the bar,
" # " # " #
δv : 0 HTsv v fp 6= 0 at element level,
· − (14.22)
δas : Hsv −F as 0 = 0 at element level
is valid. The computation of as can be performed now with
as = F−1 · Hsv · v . (14.23)
Introducing as into the conditions of equilibrium yields

−δWd = δv · [ 0 + HTsv · F−1 · Hsv ] · v − fp . (14.24)
The expression given in square brackets at Equation (14.24) corresponds to the
stiffness matrix of the displacement–based method.

14.4 Subsequent Stress Analysis

Alternative A
The calculation of the stresses related to the element domain can be performed
by evaluating the governing equations equivalently to the procedure at applying
displacement–based elements, cf. Section 12. This procedure is advantageous,
since the approaches related to the displacements may be directly evaluated
applying kinematics. Thus as a start the following is valid:
σ = E · ǫ,
σ = E · D· u, (14.25)
σ = E·D·Ω·v = E·B·v = S·v.
258 14 Hybrid–Mixed Plane Stress Elements

The general formulation of the stress matrix S is performed by analogy with

Section 12.2.2, and the evaluation is conducted preferably at the corner nodes
of the element or the supporting points of the numerical integration with

S̃ = E · B̃ = E · B1 · B2 · B̃3 = Ẽ · B̃3 . (14.26)

Ẽ is computed for each point to be evaluated with

 
i22 −i12 −ν i21 ν i11
E 1  
Ẽ = · · ν i22 −ν i12 −i21 i11  . (14.27)
1 − ν 2 det J 1−ν 1−ν 1−ν 1−ν
− 2 i21 i
2 11
i
2 22
− i
2 12
Concerning the element employing linear approaches to describe the displace-
ments the product of matrix B̃ with the columns 2 and 3 of matrix Ẽ should
be evaluated in the centre of the element and with the columns 1 and 4 in the
corner nodes of the element respectively. Thus physically non–desirable oscil-
lations in the courses of stresses may be avoided, whereby this procedure is
comparable to applying the selectively reduced integration according, cf. Sec-
tion 5.

Alternative B
The calculation of the stresses, which is performed here by employing the shape
functions to describe the stresses and the PvF according to
 
σxx
 σyy  = σ = Ψs · as and as = F−1 · Hsv · v , (14.28)
σxy
yields more accurate results and is more efficient on top, if, after the discretiza-
tion of the work equation concerning the PvF, the product F−1 · Hsv is stored
element by element and is availabe for the subsequent stress analysis. After
the computation of as the stresses may be evaluated by means of the general
polynomials Ψs .

14.5 Linear Shape Functions

The shape functions to describe the geometry of the quadrilateral element and
to describe the displacements in the element domain may be taken from the
displacement–based formulation, cf. Sections 5.1 and 12.2. Applying linear sha-
pe functions the element coordinates and element nodes are used as described
in Figure 14-3. The following approach employ bi–linear shape functions to
describe the displacements and three balanced polynomials with five general
14.5 Linear Shape Functions 259

degrees of freedom to describe the stresses yielding an element, which is named

P-HMQ-4-5.

x A
y z local coordinates (ξ, η)
(-1,-1)
a
B
(-1,1)
d η ξ b (1,-1)
D
c
(1,1)

Fig. 14-3 Element nodes and coordinates – linear shape functions

Linear Approaches Concerning Geometry and Displacements

At applying isoparametric elements the shape functions to describe the geome-
try are given as product expressions. Regarding bi–linear approaches employing
the shape functions φ(ξ, η)i , this yields
1
x = Σi (1 + ξi · ξ)(1 + ηi · η) xi ,
4
1
y = Σi (1 + ξi · ξ)(1 + ηi · η) yi .
4
Thereby xi , yi , ξi and ηi are the global and the local coordinates of the respec-
tive element nodes. By complete analogy the approach related to the displace-
ments is chosen to
1
u(ξ, η) = Σi (1 + ξi · ξ)(1 + ηi · η) ui ,
4
1
v(ξ, η) = Σi (1 + ξi · ξ)(1 + ηi · η) vi .
4
Thus, concerning the real displacements, it follows

u = Ωv · v

with uT = [ u v ] and the nodal displacements vT = [ uA vA uB vB uC vC uD vD ]

as well as with the same related to the virtual displacements

δu = Ωv · δv .
260 14 Hybrid–Mixed Plane Stress Elements

Linear Approaches Concerning the Stresses

The element continuity conditions are only fulfilled weakly by the stresses.
Therefore, the approaches related to the stresses inside the element domain
may be chosen such, so as to adapt best their courses to the distribution of
the displacements and strains respectively. In general, the number of degrees of
freedom of the approaches to describe the stresses must be equal or larger than
the number of degreees of freedom to describe the strains, which yields a regular
stiffness matrix. Mathematical evidence related to this statement can be found
at Babuska [6, 7], Brezzi [22] and Ladyzhenskaya [61] and is denoted as LBB–
Condition. Thereby, as a mechanical interpretation, each possible displacement
field can be connected to a specific strain-related stiffness.

The bi–linear approaches to describe the displacements u(ξ, η) and v(ξ, η) in-
corporate four nodal displacement variables each. Thereby, in two dimensions,
three rigid body motions – two translations and a rotation – may be described
as well as five displacement fields, which are available to describe the strains
εxx εyy εxy . Therefor an approach which balances stresses and strains should in-
corporate at least five degrees of freedom related to the stresses. At choosing the
approaches it should be claimed that the courses of the strain εxx = u(x, y),x
and thus also of the stress σxx are constant in x–direction. Concerning the
strain εyy = v(x, y),y and the shear strain 2εxy = u(x, y),y + v(x, y),x the
corresponding facts are valid. This is reached by employing the following non–
scaled approaches
σxx = a1 + a2 · η constant in ξ–direction,
σyy = a3 + a4 · ξ constant in η–direction,
σxy = a5 constant.
Therewith the approach to describe the stresses is given by
σ = Ψs · as
and aTs= [ a1 a2 a3 a4 a5 ]. By analogy the approach to describe the virtual
stresses might be chosen.

There is a small discrepancy in assuming the strains in direction of the coordi-

nates x, y to be constant, but inside the element to be constantly arranged in
the direction of the loacal coordinates ξ, η. This drawback may be avoided at
employing isoparametric quadrilateral elements including an additional trans-
formation. However, no substantial improvement concerning the quality of the
results is achieved. A comparison with respect to other elements is represented
in Section 14.8 by means of two benchmarks.
14.6 Quadratic Shape Functions 261

14.6 Quadratic Shape Functions

At deriving the formulation employing the displacements as primary variables
it has turned out, that the convergence behavior of quadratic approaches is
significant better in comparison to linear approaches, cf. Section 9.5. Thus it
seems to be natural to employ equivalent approaches to derive hybrid elements,
too. Applying incomplete approaches of third order to describe the displace-
ments and approaches with 13 degrees of freedom to describe the stresses yields
an element, which is named P-HMQ-8-13.

x A local coordinates (ξ, η)

y z
(-1,-1) E
H a
(0,-1)
(-1,0)
h e B
(-1,1)
d η ξ b (1,-1)
D g f
(1,0)
(0,1) F
c
(1,1)
G

C
Fig. 14-4 Element nodes and coordinates – quadratic approaches

Concerning non–rectangular quadrilateral elements, the shape functions to de-

scribe the geometry are still given with
1
x = Σi (1 + ξi · ξ)(1 + ηi · η) xi ,
4
1
y = Σi (1 + ξi · ξ)(1 + ηi · η) yi .
4
Hereby xi , yi , ξi and ηi are the coordinates of the respective element nodes, cf.
Figure 14-4.

Bi–Quadratic Approaches Concerning the Displacements

To describe the displacements u, v incomplete cubic approaches are chosen, that
employ eight degrees of freedom each and that are related to the corners as well
as to the midpoints of the edges. Since the element midpoint is neglected, the
approaches do not comprise all parts of a symmetrical approach, because the
term ξ 2 · η 2 is neglected. This fact turns out to be no drawback, but avoids
unphysical oscillations of the solution inside the element domain. Employing
262 14 Hybrid–Mixed Plane Stress Elements

the approach for the displacements u(ξ, η)

u(ξ, η) = a1 + a2 ξ + a3 η + a4 ξ 2 + a5 ξη + a6 η 2 + a7 ξ 2 η + a8 ξη 2

yields the matrix notation

u(ξ, η) = Ψ8 · au .

The scaling of u(ξ, η) with respect to the nodal displacements

vu T = [ uA uB uC uD uE uF uG uH ]

results in
 
1 −1 −1 1 1 1 −1 −1
 

 1 1 −1 1 −1 1 −1 1 

 1 1 1 1 1 1 1 1 
 
 
 1 −1 1 1 −1 1 1 −1 
Ψ̃8 = 


 1 0 −1 0 0 1 0 0 

 
 1 1 0 1 0 0 0 0 
 

 1 0 1 0 0 1 0 0 

1 −1 0 1 0 0 0 0

−1
with the inverse representation G8 = Ψ̃8
 
−0,25 −0,25 −0,25 −0,25 0,50 0,50 0,50 0,50
 

 0 0 0 0 0 0,50 0 −0,50 

 0 0 0 0 −0,50 0 0,50 0 
 
 
−1  0,25 0,25 0,25 0,25 −0,50 0 −0,50 0 
Ψ̃8 =  .
 0,25 −0,25 0,25 −0,25 0 0 0 0 

 
 0,25 0,25 0,25 0,25 0 −0,50 0 −0,50 
 
 −0,25 −0,25 0,25 0,25 0,50 0 −0,50 0 
 
−0,25 0,25 0,25 −0,25 0 −0,50 0 0,50

Choosing an equivalent description for the displacement v(ξ, η) yields the ma-
trix notation concerning quadratic approaches for uT = [ u(ξ, η) v(ξ, η) ]

u = Ψ16 · au,v .
14.6 Quadratic Shape Functions 263

Scaling au,v with respect to the nodal displacements

vT = [ uA vA uB vB uC vC uD vD uE vE uF vF uG vG uH vH ]

yields the physically meaningful shape functions Ωv

u = Ψ16 · G16 · v ,
= Ωv · v .

The approach to describe the virtual displacements is equivalently chosen.

Quadratic Approaches Concerning the Stresses

The incomplete cubic approach to describe the displacements u(ξ, η) and v(ξ, η)
comprises 2 × 8 nodal displacements. Therewith, in two dimensions, three rigid
body motions – two translations and one rotation – may be described as well
as 13 displacement fields, that are available for the description of the strains
[ εxx εyy εxy ]. Therefore, an approach to describe the stresses and being balan-
ced to the strains comprises 13 degrees of freedom as well.
By analogy with the linear approach and with respect to the strains, one may
request on εxx = u(x, y),x and thus also on the stresses σxx being linear in
x–direction but quadratic in y–direction. Correspondingly, requests are to be
made with respect to εyy = v(x, y),y and, regarding the shear strains, to 2·εxy =
u(x, y),y + v(x, y),x . These demands are fulfilled employing the following non–
scaled approaches

σxx = a1 + a2 · ξ + a3 · ξη + a4 · η + a5 · η 2 linear in ξ–direction,

2
σyy = a6 + a7 · η + a8 · ηξ + a9 · ξ + a10 · ξ linear in η–direction,
σxy = a11 + a12 · ξ + a13 · η bi–linear.

Hence the approach to describe the stresses is given with

σ = Ψs · as

and aTs = [ a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13 ], an analogous approach is

chosen to describe the virtual stresses. Equivalent to the linear approach, small
discrepancies arise at assuming the courses of strains being linear in direction
of the coordinates x, y, but being addressed in direction of the coordinates ξ, η
inside the element. Introducing a modification at the approaches related to the
stresses, these deviations may be considered, cf. Section 14.7. The benchmarks
represented in Section 14.8 illustrate the high quality of the element.
264 14 Hybrid–Mixed Plane Stress Elements

14.7 Linear Approaches with Coupling of Degrees of Freedom

The approaches to describe the stresses are chosen with respect to the local ξ–η–
coordinate system so far, cf. Sections 14.5 and 14.6. In the case of geometrically
deformed elements this may result in undesirable oscillations as well as bad
convergence. These drawbacks may be partially eliminated by a modification
of the approaches, that takes into account the variation of the components
of the Jacobian matrix with respect to the element coordinates. Figure 14-5
depicts the angles ϕξ and ϕη of the local ξ–η–axes with respect to the global
x–y–coordinates as well as the modification of the element geometry along the
coordinates.

xo x

1 B
A

4 ϕξ
yo
2 ξ
ϕη
D
3
η C
y

Fig. 14-5 Isoparametric quadrilateral element

The element geometry is described by the differences of the coordinates

a1 = 0. 25 · (−xA + xB + xC − xD )
a2 = 0. 25 · (+xA − xB + xC − xD )
a3 = 0. 25 · (−xA − xB + xC + xD )
b1 = 0. 25 · (−yA + yB + yC − yD )
b2 = 0. 25 · (+yA − yB + yC − yD )
b3 = 0. 25 · (−yA − yB + yC + yD ) .

Therewith, considering the linearly fluctuating geometry, the Jacobian matrix

14.7 Linear Approaches with Coupling of Degrees of Freedom 265

of the element may be computed by means of ai , bi

" 4 # " #
Σi=1 φi ,ξ xi Σ4i=1 φi ,ξ yi a1 + a2 · η b1 + b2 · η
J= = .
Σ4i=1 φi ,η xi Σ4i=1 φi ,η yi a3 + a2 · ξ b3 + b2 · ξ
Concerning the origin of the local coordinate system, this yields
det J|0 = det J|ξ=0, η=0 = b3 · a1 − b1 · a3 .
The angles between the coordinate axes are given with
cos ϕη = 2 · b3 /ℓη ,
sin ϕη = −2 · a3 /ℓη ,
cos ϕξ = 2 · a1 /ℓξ ,
sin ϕξ = 2 · b1 /ℓξ ,
employing the differences of the coordinates of the midpoints of the edges
p
ℓξ = 2 · (a1 )2 + (b1 )2 ,
p
ℓη = 2 · (a3 )2 + (b3 )2 .

14.7.1 Approach Following Pian and Sumihara

By analogy with Section 14.5, Pian and Sumihara [81] develop a 4–node ele-
ment comprising bi–linear approaches to describe the displacements as well as
linear approaches employing five degrees of freedom βi to describe the stres-
ses. Deviating from Section 14.5 the approaches to describe the stresses are
not independent from each other, but are coupled by the degrees of freedom
β4 , β5 . This element is named P-HMQC-4-5, because of the coupling of the
shape functions. The coefficients ai , bi , which are employed in the approaches
 
    β1
σxx 1. 0 0 0 a1 · a1 · η a3 · a3 · ξ  β2 
 
 σyy  =  0 1. 0 0 b1 · b1 · η b3 · b3 · ξ  β3  ,
·
 
σxy 0 0 1. 0 a1 · b1 · η a3 · b3 · ξ  β4 
β5
take into account the geometry of the element. The approach is deviated by
employing additional internal quadratic polynomials concerning the displace-
ments, see [81]. The dependencies on units, that exist in the approaches concer-
ning the degrees of freedom β4 , β5 , could be eliminated by a convenient scaling
procedure. Employing these approaches yields computational results of high
quality, which is proven by comparison to the quadratic approach presented in
Section 14.6, cf. the study of convergence at Section 14.8.
266 14 Hybrid–Mixed Plane Stress Elements

14.7.2 Approach with Transformation of Coordinates

The coupled linear approaches comprising five degrees of freedom may be devia-
ted completely different by means of a transformation of the stresses between
the global x–y coordinate system and the local ξ–η coordinates. Figure 14-6
illustrates the transformation with two sectional views.

σηη x x
σyy
σηξ σyx
1
5 σξξ
5
6
σηξ σξη
ϕξ σηη 6

σxy
σξξ ξ
σxx σξη
ϕξ
3 η ϕη
ξ
η ϕη
y y

Fig. 14-6 Coordinate transformation from x–y to ξ–η

The lengths of the edges, that are required for the transformation procedure,
are identified from the corner nodes of the sections. The coordinates of the
midpoint of the edges 1 and 3 are given with Figure 14-5. Thus the lengths

ℓ13 = ℓη ,
ℓ15 = ℓη sin ϕη = −2 · a3 ,
ℓ35 = ℓη cos ϕη = 2 · b3 ,

may be computed directly, since the coordinates of the corner node 5 are de-
termined by the midpoints of the edges 1 and 3. The coordinates of the node
6 may be determined with respect to corner node 5 by
∆y56 b1
tan ϕξ = = ,
∆x56 a1
∆x56 −a3
tan ϕη = = .
b3 − ∆y56 b3
Introducing

∆x56 = −a1 · a3 · b3 / det J0 ,

∆y56 = −b1 · a3 · b3 / det J0
14.7 Linear Approaches with Coupling of Degrees of Freedom 267

the still missing lengths follow to

p
ℓ56 = (∆x56 )2 + (∆y56 )2 = −a3 · b3 · ℓξ / det J0 ,
ℓ36 = (b3 − ∆y56 )/ cos ϕη = a1 · b3 · ℓη / det J0 ,
ℓ16 = ℓ13 − ℓ36 = −b1 · a3 · ℓη / det J0 .

In a first step the conditions of equilibrium related to the x– and the y–direction
are assembled following Figure 14-6, and they are rearranged to transform the
stresses now. Thus matrix notation yields
 
ℓ36 ℓ56

σxx

ℓ
cos ϕξ ℓ
sin ϕη − ℓℓ36 sin ϕη − ℓℓ56 cos ϕξ  σξξ 
 35 35 35 35 
 σ   ℓ16 sin ϕξ ℓ56
cos ϕη ℓ16
cos ϕη ℓ56
sin ϕξ 
  σηη 
 
 yy   ℓ15 ℓ15 ℓ15 ℓ15
  =  ℓ36 ℓ 56 ℓ 36 ℓ 56
 .
 σxy    σξη 
 ℓ35 sin ϕξ − ℓ35 cos ϕη ℓ35
cos ϕη − ℓ sin ϕξ 
35 
σyx ℓ16
cos ϕξ − ℓ56 sin ϕη − ℓ16 sin ϕη ℓ56
cos ϕξ σηξ
ℓ15 ℓ15 ℓ15 ℓ15

Replacing the trigonometric functions as well as the length of the edges by the
differences of the coordinates and taking into account σxy = σyx results in
 ℓ ℓ
 
  a1 a1 ℓη a3 a3 ℓ ξ a1 a3 a1 a3 σξξ
σxx ξ η
  1 
 ℓη ℓξ

 σηη 

 σyy  = ·  b1 b1 ℓ b3 b3 ℓ b1 b3 b1 b3   .
det J0  ξ η  σξη 
σxy ℓ ℓ
a1 b1 ℓη a3 b 3 ℓ ξ a1 b 3 b 1 a3 σηξ
ξ η

In a second step and following Section 14.5 the linear approach to describe the
local stresses is chosen to σξη = σηξ employing the 5 degrees of freedom βi
   
σξξ   β1
1 0 0 η 0 
 σηη   β2 
   0 1 0 0 ξ   
 =  β3  (14.29)
 σξη  0 0 1 0 0  
 β4 
σηξ 0 0 1 0 0
β5

and is applied to transform the stresses

 ℓ ℓ ℓ ℓ
 β1


σxx
 a1 a1 ℓη a3 a3 ℓ ξ 2a1 a3 a1 a1 ℓη ·η a3 a3 ℓ ξ ·ξ 
 ξ η ξ η  β2 
1 ℓ ℓ ℓ ℓ 
b1 b1 ℓη b3 b3 ℓ ξ b1 b1 ℓη b3 b3 ℓ ξ
   
 σyy  =  2b1 b3 ·η · ξ  β3 .
det J0  ξ η ξ η  
σxy ℓ ℓ ℓ ℓ  β4 
a1 b1 ℓη a3 b 3 ℓ ξ a1 b3 + b1 a3 a1 b1 ℓη ·η a3 b 3 ℓ ξ ·ξ
ξ η ξ η β5
268 14 Hybrid–Mixed Plane Stress Elements

Transforming the degrees of freedom βi with

ℓ ℓ
β̃1 = β1 ℓη /det J0 , β̃2 = β2 ℓξ /det J0 , β̃3 = β3 /det J0 ,
ξ η

ℓ ℓ
β̃4 = β4 ℓη /det J0 , β̃5 = β5 ℓξ /det J0
ξ η

the approach can be represented by

 
    β̃1
σxx a1 a1 a3 a3 2a1 a3 a1 a1 · η a3 a3 · ξ  β̃2 
     
 σyy  =  b1 b1 b3 b3 2b1 b3 b1 b1 · η b3 b3 · ξ  · 
 β̃3 .

σxy a1 b 1 a3 b 3 a1 b 3 + b 1 a3 a1 b 1 · η a3 b 3 · ξ  β̃4 
β̃5

The appproach presented here is equivalent to the approach developed by Pian

und Sumihara following a completely different path. It can be transferred ap-
plying a scaling procedure to the degrees of freedom β̃1 , β̃2 , β̃3 . The advantage
of the formulation presented here is the possibility to apply it to higher order
polynomials by analogy with Equation (14.29).

14.8 Convergence Behavior of the Elements

The hybrid–mixed plane stress elements partially yield very good results. Due
to the special adaptation of the courses of stresses to the strain fields the
physically non–reasonable zero energy modes are avoided.

Example of Use 1
As a first example of use a benchmark is investigated at first published by Pian
and Sumihara, cf. Figure 14-7.

P2

2 2 1 1 4 P1
P2
A B C
y, v 2
x, u D P1
P2
1 1 2 3 3

E = 1500 N/m 2 , ν = 0.25, P1 = 1000 N, P2 = 150 N, l x = 10 m, l y = 2 m

Fig. 14-7 Cantilever published by Pian and Sumihara [81] – geometry and loading
14.8 Convergence Behavior of the Elements 269

The cantilever is examplarily investigated employing linear approaches to des-

cribe the displacements and relating linear approaches to describe the stresses
according to Sections 14.5 and 14.7 as well as employing quadratic approaches
according to Section 14.6. Due to the complex element geometry and the chosen
approaches the different element formulations approximate the displacements
as well as the stresses with different accuracy.

Applying the finite element mesh according to Figure 14-7 the subsequent
Table 14.1 comprises the displacement results related to the loading case P1 =
1000 N . The three hybrid element formulations are compared to the analytical
solution, that is evaluated with respect to a linearly distributed loading in
y–direction p1 (y) = 2000 · ( y − 1 ) N/m.

Table 14.1 Loading case bending P1 – comparison of different elements

vA vB vC vD

reference solution 0. 0000 25. 000 100. 000 100. 000

P-HMQ-4-5 – Section 14.5 −0. 3847 21. 508 76. 818 77. 542
P-HMQ-8-13 – Section 14.6 −0. 00825 24. 847 99. 991 100. 026
P-HMQC-4-5 – Section 14.7 −0. 552 25. 538 94. 012 96. 184

The stresses σxx related to P1 are represented in Figure 14-8. The results il-
lustrate the influence of the element geometry on the approximation of the
stresses. The quadratic approaches P-HMQ-8-13 yield the best results as ex-
pected. The stresses evaluated from the element P-HMQC-4-5 published by
Pian and Sumihara yield good results although an approach of low order is
employed. In contrast the bi–linear approach P-HMQ-4-5 is not adequate.

N/m2 σxx
-3000 P1 = 1000 N
analytical solution
P-HMQ-4-5
P-HMQ-8-13
P-HMQC-4-5
-2000
x

Fig. 14-8 Stress σxx at the upper boundary of the cantilever for P1
270 14 Hybrid–Mixed Plane Stress Elements

The second loading case is more difficult to be described by the given elements,
since the stresses σxx develop linearly along the x–axis, whereat the stresses
σxy are quadratically distributed along the y–axis. Due to the parabolic distri-
bution of the stresses σxy in y–direction the load is transferred to concentrated
nodal actions according to the PvD. The subsequent Table 14.2 comprises the
displacement results concerning the loading case P2 = 150 N . As in loading case
P1 the quadratic approach gives the best results, whereat the linear approach,
according to Section 14.5, is hardly acceptable.

Table 14.2 Loading case bending P2 – comparison of different elements

vA vB vC vD

reference solution 0. 0000 32. 600 102. 750 102. 750

P-HMQ-4-5 – Section 14.5 −0. 509 27. 765 81. 714 82. 023
P-HMQ-8-13 – Section 14.6 −0. 0251 32. 472 102. 439 102. 421
P-HMQC-4-5 – Section 14.7 −0. 809 32. 138 97. 533 98. 188

Applying the mesh according to Figure 14-7 the stresses σxx are represented
in Figure 14-9 related to the loading case P2 . As in loading case P1 the results
illustrate the strong influence of the element geometry on the approximation of
the stresses. The quadratic approach yields the best results and describes the
stresses σxx to be linearly distributed within the element without a larger gap at
the interface of neighboring elements. The stresses evaluated from the element
published by Pian and Sumihara yield good results, although an approach
of low order is employed. In contrast the bi–linear approach is not adequate,
since it leads to a stepwise approximation with larger gaps at the interfaces of
neighboring elements.

N/m2 σxx
-4000 P2 = 150 N
reference solution
P-HMQ-4-5
P-HMQ-8-13
P-HMQC-4-5
0
x

Fig. 14-9 Stress σxx at the upper boundary of the cantilever for P2
14.8 Convergence Behavior of the Elements 271

Example of Use 2
As a further example of use Cook’s can- c
tilever [28] is chosen, since comparable x
results are available. Without comple- B
y D 1N

b
tely representing the study of conver-
gence according to Section 12.2.3, the
results for different element formulati- C
ons are compared in Table 14.3 apply- E = 1.0 N/mm 2
ing meshes with 3 × 3 and 5 × 5 no- ν = 0.33

a
des. Concerning the hybrid elements the A a = 44 mm
respective subsequent stress analysis is b = 16 mm
performed each with alternative B. c = 48 mm

Table 14.3 Cook’s cantilever – vertical displacements in D, stresses in A and B

nodes 3×3 5×5 5×5 5×5

v|D v|D σxx |B σxx |A
2
mm mm N/mm N/mm2
element left right left right
1 iso - 4 11. 845 18. 299 0. 1319 0. 1474 −0. 2005 −0. 0474
2 iso - 4 - SRI 20. 038 22. 648 0. 1775 0. 1305 −0. 1491 −0. 1854
3 iso - 9 19. 644 23. 289 0. 1635 0. 2066 −0. 1350 −0. 1757
4 iso - 9 - SRI 22. 626 23. 653 0. 1596 0. 2220 −0. 1738 −0. 1808
5 tri - 3 11. 991 18. 283 0. 1323 0. 0887 −0. 0776 −0. 0482
6 tri - 6 18. 358 23. 301 0. 1567 0. 1878 −0. 1086 −0. 1766
7 hybrid, 14.5 17. 756 21. 935 0. 1909 0. 1660 −0. 1375 −0. 1936
8 hybrid, 14.6 21. 063 23. 411 0. 1985 0. 1565 −0. 1153 −0. 1570
9 hybrid, 14.7 21. 128 23. 021 0. 1783 0. 1612 −0. 1127 −0. 1451

Regarding the hybrid–mixed element formulations and due to the especially

chosen approaches to describe the stresses the displacements are approximated
very well even at applying few elements. The stresses σxx |B = 0,1832 N/mm2
and σxx |A = −0,1290 N/mm2, evaluated according to Section 12.2.3 and chosen
as a reference here, are also considerably better approximated in comparison
to other element formulations employing an approach of a comparable order.
Due to the excellent behavior of convergence the hybrid elements employing
linear approaches according to Pian and Sumihara as well as those employing
quadratic approaches according to Section 14.6 might be preferred compared
to other element formulations.
15 Hybrid–Mixed Euler–Bernoulli Beam Elements

The derivation of a hybrid–mixed element formulation for the Euler–Bernoulli

beam takes place by analogy with the bar.

15.1 Mixed Formulation Employing Forces and Displacements

Concerning FEM formulations that employ the displacements as primary va-
riables the choice of the approach to describe the deflection has been discussed
already with respect to C1 –conformity. Applying the mixed formulation this
specific characteristic can be avoided by incorporating w and M as primary
variables, whereat the definition of signs is given in Figure 15-1.
p
z

Me
e
x V
ϕ p
z

z,w dM dM
M- M+
2 x 2
dQ dQ
Q- Q+
2 2

Fig. 15-1 Euler–Bernoulli beam – coordinates and definition of signs

15.1.1 Governing Equations

The governing equations related to beams according to the Euler–Bernoulli
theory have been represented already at Section 6.1, whereat the description
dealing with differential equations of second order is used:

a) Kinematics
in the domain w,xx − κ = 0 ,
at the boundary ϕ − ϕe = 0 , (15.1)
w − we = 0 .
b) Equilibrium
in the domain M,xx + pz = 0 ,
at the boundary Q−Ve = 0, (15.2)
M + Me = 0 .

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_15
274 15 Hybrid–Mixed Euler–Bernoulli Beam Elements

c) The material equation

EI · κ + M = 0 (15.3)
connects the bending moment to the curvature. Regarding Eq. (15.1) and
replacing the curvature by the bending moment incorporating the material
equation (15.3) yields the condition of deformation
1
w,xx + · M = 0. (15.4)
EI
The equations (15.2) and (15.4) may be given in matrix notation with
" # " # " #
0 ∂x2 w pz
1 · + = 0. (15.5)
∂x2 EI M 0

15.1.2 Work Equations

The governing equations as well as the related boundary conditions may be
formulated equivalently applying the Principle of virtual Work.

The Principle of Virtual Displacements – PvD

The Principle of virtual Displacements describes the virtual work that the ben-
ding moment and the external actions pz (x), V e , M e perform on a virtual state
of deformation.
Z
−δWd = {δκ · M + δw · pz } dx + [ δw · V e + δϕ · M e ]bound. = 0 .

The PvD is equivalent to the conditions of equilibrium, if the virtual displa-

cements strongly fulfill kinematics. This becomes obvious when describing the
virtual curvature by the deflection line,
Z
{δw,xx · M + δw · pz } dx + [ δw · V e + δϕ · M e ]bound. = 0 . (15.6)

Considering Eq. (15.6) the boundary conditions related to the force variables
are naturally fulfilled. Integrating by parts of the first expression and taking
into account δw,x = δϕ yields
Z
{−δw,x · M,x + δw · pz } dx + [δw · V e + δϕ · (M e + M )]bound. = 0 . (15.7)

Hereby the boundary condition concerning the bending moment occurs as an

substantial condition. Integrating by parts once again results in
Z
{δw ·M,xx +δw ·pz } dx+[ δw ·(V e −M,x )+δϕ·(M e +M ) ]bound. = 0 . (15.8)
15.1 Mixed Formulation Employing Forces and Displacements 275

Applying the integral formulation the conditions of equilibrium are fulfilled

weakly with respect to the domain as well as to the boundary. The boundary
conditions related to the force variables both are fulfilled exactly by the contour
expressions, if corresponding variables are available and the contour expressions
disappear.

The Principle of virtual Forces – PvF

The Principle of virtual Forces describes kinematics of the real state of defor-
mation in a weak formulation, if the virtual forces strongly fulfill the conditions
of equilibrium. As a start
Z
−δWσ = δM · (−w,xx + κ) dx − [ δQ · (w − we ) + δM · (ϕ − ϕe ) ]bound. = 0

is valid. Replacing the curvature by the material equation and thus by the
bending moment yields
Z
M
δM · (−w,xx − ) dx − [ δQ · (w − we ) + δM · (ϕ − ϕe ) ]bound. = 0 . (15.9)
EI
Concerning the domain the condition of deformation is weakly fulfilled by the
integral. The contour expressions take into account the displacement boun-
dary conditions. Integrating the first expression in the integral by parts and
introducing w,x = ϕ yields
Z
M
{δM,x · w,x − δM } dx − [ δQ · (w − we ) − δM · ϕe ]bound. = 0 . (15.10)
EI
A second integration by parts of the same term and replacing δM,x = δQ
results in
Z
M
{−δM,xx · w − δM } dx + [ δQ · we + δM · ϕe ) ]bound. = 0 . (15.11)
EI
In Eq. (15.9) the contour expressions comprise the boundary conditions related
to the displacement variables w and ϕ. After integration by parts the boundary
condition related to the rotation can only be fulfilled weakly applying Eq.
(15.10). In contrast the boundary condition related to the deflection may be
fulfilled strongly with w − we = 0. Repeated integration by parts yields weak
fulfillment of both boundary conditions applying Eq. (15.11).

Remarks
With both principles of work, applying the Equations (15.6), (15.7) and (15.8)
as well as (15.9), (15.10) and (15.11), different but equivalent formulations are
availabe, which may be chosen as a basis concerning the discretization.
276 15 Hybrid–Mixed Euler–Bernoulli Beam Elements

Applying Equations (15.6) and (15.9) requires quadratic approaches to describe

w and δw as well as at least constant approaches to describe M and δM . Since
the boundary conditions related to w and δw and also those related to w,x and
δw,x are to be fulfilled, even cubic approaches must be employed with respect
to w and δw. In comparison to the displacement–based formulation this line
of action yields no advantages and, on top, it also requires cubic approaches
to describe the bending moments, since otherwise the system matrix would
become singular, cf. the conditions according to Babuŝka [6] and Brezzi [22].
Choosing Equations (15.8) and (15.11) at least constant approaches are requi-
red to describe w and δw, at least cubic approaches are to be employed to
describe M and δM . Here the boundary conditions related to the displace-
ment variables are to be fulfilled naturally with the PvF and are to be fulfilled
strongly related to M and M,x as well as to δM and δM,x .
If only first order derivatives of the state variables occur, which means that
Equations (15.7) and (15.10) are applied, it is adequate to employ linear ap-
proaches for all state variables. Therewith the boundary conditions related to
w and δw as well as related to M and δM can be fulfilled strongly, which is
simply possible according to the approaches for w and M . Thus it remains
Z
{δw,x · M,x − δw · pz }dx − [ δw · V e ]bound. = 0 , (15.12)
Z
M
{δM,x · w,x − δM }dx + [ δM · ϕe ]bound. = 0 . (15.13)
EI
The boundary conditions related to ϕ and Q are fulfilled naturally according
to the work equations.

15.1.3 Boundary and Element Interface Conditions

Applying the PvD according to the formulation given with (15.12), the boun-
dary conditions related to δw and M are fulfilled strongly. In contrast the
conditions related to Q = M,x as well as to δϕ = δw,x are fulfilled weakly.
At the element interfaces the virtual displacements of neighboring element
boundaries need to be continuous, thus δwl = δwr = δw is fulfilled strongly by
the approaches. The continuity condition related to the shear force Q = M,x
is fulfilled naturally, if the vertical equilibrium at the internal node

M,x l − M,x r = V e ,

here additionally comprising a concentrated external action, is formulated ap-

plying the PvD
δwl · M,x l − δwr · M,x r = δw · V e .
15.2 Work Equation in Hybrid Formulation 277

Both terms, occuring at the left hand side, are implicitly incorporated in the
work equation. The external work δw · V e is taken into account by the load
vector.

Applying the PvF according to Eq. (15.13) the boundary conditions related to
δM and w are fulfilled strongly and the conditions related to δQ = δM,x as
well as to ϕ = w,x are fulfilled weakly. At a simply supported system boundary
the contour expression vanishes, since δM = 0 is valid. Regarding a clamped
system boundary a rotation of the support ϕe is to be considered with δM · ϕe
in the load vector. Concerning a fixed clamping it follows ϕe = 0.
At the element interfaces the virtual bending moments are continuous. Here,
the condition δMl = δMr = δM is fulfilled strongly, in contrast, the continuity
condition related to the rotation

w,x l − w,x r = ϕl − ϕr = ∆ϕe

is fulfilled naturally. ∆ϕe corresponds to a sharp kink at the internal node here,
which is to be considered in the load vector by −δM · ∆ϕe , comparable to the
procedure at the system boundary.

Numerical results are represented in Section 21.2 and Section 21.3 with an
extension to shear deformations by means of simplified structures.

15.2 Work Equation in Hybrid Formulation

The derivation of the hybrid–mixed formulation for Euler–Bernoulli beams ap-
plies the Equations (15.12) and (15.13). The development aims at getting an
element employing the variables w, M at element level, whereby bending mo-
ments are to be numerically eliminated in a second step. Besides the work
performed by distributed loadings p(x), the work equations related to the sys-
tem level also comprise the work performed by concentrated actions V e as
well as by single bending moments M e , that act at system nodes. Additionally
bedding of the beam is taken into account to illustrate the differences of the
formulation compared to elements employing only displacements as primary
variables.

In comparison to bars and plane stress structures a significant difference arises

from the choice of the element as well as of the system variables. Concerning
the beam w und M are defined at element level. At the system level w and
in addition, to ensure C1 –conformity, the rotation ϕ are introduced as system
variables, and the virtual rotation δϕ is introduced correspondingly. Thus va-
278 15 Hybrid–Mixed Euler–Bernoulli Beam Elements

riables are defined at the interfaces, which are not available at the element
domain.
At Figure 15-2 a beam element is represented together with the interfaces A
and B, that are identical with the system nodes. Possible gaps ∆we between
element boundaries and interfaces are not considered.

Vei pz Vei+1

A B
i a b i+1
wi Ma Qa wE ME Mb Qb wi+1
ϕi ϕ i+1

Fig. 15-2 Hybrid–mixed beam element – element domain and interface

Cutting cleanly of the element domain including the element boundaries a and
b from the interfaces A and B yields the bending moments Ma and Mb as well
as the shear forces Qa = (M,x )a and Qb = (M,x )b . The bending moments are
defined only related to the element domain and to the cut, but not related to
the interface as primary variables. As kinematic variable the deflection wE is
chosen. The nodal displacements and the nodal rotations are defined at the
system level to ensure the element continuity conditions with C1 –conformity.

Applying the PvD by analogy with Section 13.1 the equilibrium of forces is
formulated concerning the element. In addition the equilibrium of moments at
the interfaces – which means the system nodes i and i + 1 – is to be ensured
by employing the virtual rotations.
Z
δWd = {δκE · ME − δwE · c · wE + δwE · pz } dx
+ Ma · (−δϕA + δw,x |a ) − Mb · (−δϕB + δw,x |b )
+ Qa · (δwA − δwa ) − Qb · (δwB − δwb ) .

Applying the PvF at element level the conditions of deformation are fullfilled
yielding
Z
1
δWσ = {δME · κE + δME · · ME } dx
EI
+ δMa · (−ϕA + w,x |a ) − δMb · (−ϕB + w,x |b )
+ δQa · (wA − wa ) − δQb · (wB − wb ) = 0 .

If the continuity of the real displacements as well as of the virtual displacements

15.3 Element Stiffness Matrix 279

is required with

wA = wa , wB = wb , δwA = δwa , δwB = δwb ,

the corresponding contour expressions of the principles of work disappear. Due

to the independence of the virtual displacements δw from the virtual rotations
δϕ the virtual work performed on δw and δϕ may be separately looked at. Thus
it remains
Z
δw : δWw = {δκE · ME − δwE · c · wE + δwE · pz } dx
+ {−(δw,x )b · Mb + (δw,x )a · Ma }el.bound. ,
δϕ : δWϕ = {−δϕA · Ma + δϕB · Mb }interf ace ,
Z
1
δME : δWM = {δME · κE + δME · · ME } dx
EI
+ {−δMb · (w,x )b + δMa · (w,x )a }el.bound.
+ {−δMa · ϕA + δMb · ϕB }interf ace = 0 .

Applying kinematics κE = wE ,,xx and accordingly δκE = δwE ,xx as well as in-
tegration by parts of the respective work expressions yields the work equations,
that are introduced into the discretization further on
Z
δw : − δWw = {δwE,x · ME,x + δwE · c · wE − δwE · pz } dx , (15.14)

δϕ : − δWϕ = − {−δϕA · Ma + δϕB · Mb }interf ace , (15.15)

Z
1
δME : − δWM = {δME,x · wE ,x − δME · · ME } dx
EI
− {−δMa · ϕA + δMb · ϕB }interf ace = 0 . (15.16)

As at the procedure concerning bars and plane stress structures the virtual
work related to the PvF disappears at element level, thus Equation (15.16) can
be taken to eliminate the bending moments ME .

15.3 Element Stiffness Matrix

Concerning the element domain, approaches are chosen to describe wE and
ME , which are given in matrix notation by

[ wE ] = Ωv · w und wT = [ wa wb ] ,
T
[ M E ] = Ωs · s und s = [ Ma Mb ] .
280 15 Hybrid–Mixed Euler–Bernoulli Beam Elements

Along the lines of Section 13.2 and after the discretization of the virtual work
performed at the element domain the matrix notation of Equations (15.14),
(15.15) and (15.16) gives
" R T R T T # " # "R T #
δw : Ωv C Ωv dx Ωv D · D Ωs dx w Ωv p dx
R T T R · − .
δs : Ωs D · D Ωv dx − ΩTs E−1 Ωs dx s 0
Since the continuity conditions wA = wa , wB = wb , δwA = δwa , δwB = δwb
are fulfilled, the deflections as well as the rotations at the interfaces are defined
as nodal displacement variables
vT = [ wA wB ϕA ϕB ] .
Thereby the work performed by the nodal rotations can be introduced into the
work equation applying matrix notation. Employing linear shape functions to
describe wE and ME as well as the rotations ϕA , ϕB only being defined at the
interfaces and integrating the work performed at the element domain yields
Hvv HTsv
 cℓ cℓ 1
 
δwA : 3 6 ℓ
− 1ℓ wA
  
cℓ cℓ
− 1ℓ 1
  
δwB :  6 3 ℓ
 wB 
  
  
δϕA : 
 +1 
 ϕA 



  . (15.17)
δϕB :  −1  


ϕB 
  
  
1 ℓ ℓ 
δMa : 
 ℓ − 1ℓ +1 − 3EI − 6EI   =0
Ma 
  
ℓ ℓ
δMb : − 1ℓ 1
ℓ
−1 − 6EI − 3EI Mb =0

Hsv −F
By analogy with the bar again, the calculation of the element stress resultants
s takes place applying the deformation condition
s = F−1 · Hsv · v. (15.18)
After introduction the stress resultants into the PvD the element stiffness ma-
trix and the load vector follow from eliminating the bending moments

−δWd = δvT [ Hvv + HTsv · F−1 · Hsv ] · v − fp . (15.19)
Replacing " #
6 −6 4ℓ 2ℓ EI
−1
F · Hsv = · (15.20)
−6 6 −2ℓ −4ℓ ℓ2
15.3 Element Stiffness Matrix 281

yields the part of the stiffness matrix without bedding

 
12 −12 6ℓ 6ℓ
 
 −12 12 −6ℓ −6ℓ  EI
T −1  
Hsv · F · Hsv =  · 3 . (15.21)
2 2 
 6ℓ
 −6ℓ 4ℓ 2ℓ  ℓ
6ℓ −6ℓ 2ℓ2 4ℓ2

Assuming constantly distributed loading p(x) = const. as well as concentrated

e
actions Vie , Vi+1 e
and Mie , Mi+1 at the interfaces the load vector yields
 
pℓ/2 + Vie
 e

 pℓ/2 + Vi+1 
fp = 
 e
.
 (15.22)
 Mi 
e
Mi+1

Equation (15.21) corresponds to the stiffness matrix of the displacement me-

thod, though not sorted with respect to the nodes yet. In addition to the
element matrix given by Equation (15.21) and considering Equation (15.19),
the work performed due to bedding Hvv related to Equation (15.17) is to be
summed up. Differences to the formulation employing displacements as prima-
ry variables become obvious only when R taking into account bedding, since at
the hybrid–mixed model the integral wE · c · δwE dx is evaluated employing
linear approaches, whereby cubic approaches are to be employed to describe
wE at the displacement–based formulation. The different approaches regarding
the hybrid–mixed model and the formulation employing displacements as pri-
mary variables also result in different formulations concerning the load vector
Equation (15.22). Identical results may be only achieved in special cases.

Numerical results are represented in Section 21.2 and Section 21.3 with an
extension to shear deformations by means of simplified structures.
16 Hybrid–Mixed Kirchhoff Plate Elements

The formulation of the hybrid–mixed quadrilateral plate elements is derived

in analogy to the Euler–Bernoulli beam. Again, the element is split into the
element domain with the element boundaries and the interfaces, which are
identical with the system grid lines being cut cleanly.

x A
y z,w

ϕn(A,B)
ϕn(D,A)
a

p B
d b
qn
D qn
mt mt
mn mn
mt c mt
ϕn(B,C)
qn qn
ϕn(C,D)
C

Fig. 16-1 Hybrid plate element – element domain and interfaces

16.1 Mixed Principles of Work Concerning Kirchhoff Plates

To develop the mixed plate element Kirchhoff ’s plate theory is chosen, to illu-
strate the analogy with the mixed Euler–Bernoulli beam element.

16.1.1 Governing Equations

The governing equations are represented at full length in Section 6.2. Figure 16-
2 shows the differential element and the related stress resultants at the positive
cutting line.
By analogy with the Euler–Bernoulli beam the conditions of equilibrium are
summarized by a differential equation of 2nd order

mxx ,xx + myy ,yy + 2mxy ,xy + p = 0 . (16.1)

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_16
284 16 Hybrid–Mixed Kirchhoff Plate Elements

p
x
y p
z,w dx
dy
dy
dx

m xy
m yy m xx

qx
m yx
qy

Fig. 16-2 Kirchhoff ’s plate theory – coordinates and definition of signs

Thus, applying matrix notation, it follows

 
mxx
 
[ ∂xx ∂yy 2∂xy ] ·  myy  + [ pz ] = [ 0 ]. (16.2)
mxy

Again, the condition of equilibrium (16.2) may be represented with the matrix
symbols already known
−DT · σ + pz = 0 . (16.3)
Kinematics is described as a 2nd order system, too:

ǫ =D·u
   
−κxx −∂xx
   
 −κyy  =  −∂yy  · [ w ] . (16.4)
−2κxy −2∂xy

The material equation is applied in its inverse representation, thereby the fle-
xibility F = 1/B(1 − ν 2 ) = 12/Et3 is introduced as a material parameter. In
matrix notation this yields

ǫ = E−1 · σ,
     
−κxx F −νF mxx
     
 −κyy  =  −νF F  ·  myy  . (16.5)
−2κxy 2(1 + ν)F mxy
16.1 Mixed Principles of Work Concerning Kirchhoff Plates 285

Incoporating the material equation into kinematics results in the deformation

condition

D · u − E−1 · σ = 0 ,
       
−∂xx F −νF mxx 0
       
 −∂yy  · [ w ] −  −νF F  ·  myy  =  0  . (16.6)
−2∂xy 2(1 + ν)F mxy 0

The matrix symbols of the mixed formulation are determined from the gover-
ning equations, hence the work equations may be described correspondingly.

16.1.2 Work Equations

Here again, the governing equations are formulated equivalently to the Principle
of virtual Work.

The Principle of Virtual Displacements – PvD

Concerning Kirchhoff ’s plate theory the Principle of virtual Displacements com-
prises the internal work performed by the moments and the external work per-
formed by distributed external actions pz (x, y)
Z
δWd = {mxx · δκxx + myy · δκyy + 2 · mxy · δκxy + pz · δw} dx = 0 . (16.7)

Introducing kinematics to replace the virtual curvature yields

Z
δWd = {mxx · δw,xx + myy · δw,yy + 2 · mxy · δw,xy + pz · δw} dx = 0 . (16.8)

Choosing a notation by analogy with the Euler–Bernoulli beam in Section

15.1.2, which only comprises first order derivatives of the deflection w and
the moments, the work performed by the moments on the curvature is to be
integrated by parts once. Thereby the first as well as the second expression is to
be integrated by parts with respect to x and y respectively. The third expres-
sion is processed correspondingly, but split up one half each related to x and y.
The contour expressions, resulting from integration, become line integrals now
Z Z
δWd = − δw,x · mxx ,x dA + δw,x · mx dy|x = konst.
ZA Zy
− δw,y · myy ,y dA + δw,y · my dx|y = konst.
A x
286 16 Hybrid–Mixed Kirchhoff Plate Elements
Z Z
− δw,x · mxy ,y dA + δw,x · mx dx|y = konst.
ZA Zx
− δw,y · mxy ,x dA + δw,y · my dy|x = konst.
ZA y

+ δw · p dA = 0 . (16.9)
A

At a system boundary, the contour expressions concerning Equation (16.9)

may be interpreted as follows, furthermore employing the coordinate describing
the contour ds and the indices t, n to indicate the tangential and the normal
direction respectively. The contour integral
Z
δWbound. = δw,n · mn ds (16.10)
s

disappears, if, regarding a clamped boundary, the virtual rotations and, at a

simply supported or free boundary, the conjugated real moments are equal to
zero. Regarding a flexible clamping the contour integral is to be evaluated, since
the clamping moments can be described by means of a corresponding material
equation.
If the element continuity conditions related to the moments are strongly fulfilled
with
mn l − mn r = 0 ,
the element continuity conditions related to the virtual slopes are weakly ful-
filled with
mn ( δw,n l − δw,n r ) = 0 .
Thus the respective expressions of the work equation also disappear at the
element interface.
The contour integral
Z
δWbound. = δw,s · mt ds (16.11)
s

disappears regarding a boundary supported by δw,s = 0. At a free boundary the

boundary conditions related to the twisting moment as well as to Kirchhoff ’s
effective shear force are fulfilled weakly, if the contour integral is evaluated.
Starting at Equation (16.11) the contour integral may be interpreted as follows.
A second integration by parts initially yields
Z
δWbound. = [−δw · mt ,s + (δw · mt ),s ] ds .
s
16.1 Mixed Principles of Work Concerning Kirchhoff Plates 287

The first term at the integral corresponds to the part of the work performed
by Kirchhoff ’s effective shear force, that is not captured by the shear force yet.
The last term may be integrated following the respective boundary. This yields
Z
δWbound. = −δw · mt ,s ds + δw · mt |corner .
s

The term comprising the work performed at the corner disappears, since ei-
ther the twisting moment at the free corner or the virtual displacement at
the supported corner is equal to zero. The remaining integrals are summarized
to Kirchhoff ’s effective shear force qKn employing the shear forces qx , qy and
qn implicitly incorporated in the work equation. Thus, the element continuity
conditions related to the effective shear forces following the contour coordinate
s of the element are also fulfilled weakly at the system level with
Z
δw[ (qn + mt ,s )l − (qn + mt ,s )r ] ds = 0 .
s

With Equation (16.9) the work equation related to the PvD follows to
Z
− δWd = {δw,x · mxx ,x + δw,y · myy ,y + δw,x · mxy ,y + δw,y · mxy ,x } dA
ZA Z
− δw,s · mt ds − δw · p dA = 0 (16.12)
s A

forming the basis of the discretization. Thus, besides the equilibrium in the
domain, the boundary conditions qKn = 0 at the free boundary as well as
mt = 0 at the free corner are fulfilled naturally. The virtual work performed
by line–shaped external actions in s–direction as well as work performed by
concentrated actions may be additively taken into account summing up the
external work at the interfaces in the load vector.

Regarding the PvD, the boundary conditions related to the bending moments
as well as to the virtual displacements are substantial ones and therefore are
to be fulfilled by the approaches.

The Principle of Virtual Forces – PvF

The PvF describes the deformation conditions in the domain as well as at the
boundary in a weak form by
Z
δWσ = {− δmxx [ w,xx + F · ( mxx − ν myy ) ]
− δmyy [ w,yy + F · ( ν mxx − myy ) ]
288 16 Hybrid–Mixed Kirchhoff Plate Elements

− δmxy [ 2w,xy + 2(1 + ν)F · mxy ] } dA

Z Z
+ δqKn ( w − we ) ds + δmn ( ϕn − ϕen ) ds = 0 .

Integrating the terms related to the domain by parts again, yields

Z
δWσ = {δmxx ,x w,x − δmxx · F · (mxx − ν myy )
+ δmyy ,y w,y − δmyy · F · ( ν mxx − myy )
+ (δmxy ,y w,x + δmxy ,x w,y ) − δmxy · 2(1 + ν)F · mxy } dA
Z Z Z
+ δqKn ( w − we ) ds + δmn (−w,n + ϕn − ϕen ) ds − δmt w,s ds = 0 .

The boundary conditions concerning the real deflection w − we = 0 are sub-

stantial and are to be fulfilled by the approach describing w. Hence, the first
contour integral disappears. If the deformation conditions related to the slo-
pes ϕn = w,n are fulfilled naturally – which means by the work equation –
, the second contour integral is reduced to the expression δmn · ϕen . Thereby
impressed kinks at the element interfaces or impressed rotations at the plate
boundaries may be realized at incorporating the expression in the load vector.
Together with the part of Kirchhoff ’s effective shear force that is related to the
twisting moment, the third contour integral describes the work performed by
the virtual corner force δmt on the deflection w at a non–supported corner of
the plate, which has to disappear to δmt |corner = 0 :
Z
(δmt ,s · w + δmt · w,s ) ds = [ δmt · w ]|corner = 0 .

Due to the first part being implictly considered already by the first contour
integral, the third contour integral of the work equation is to be evaluated
explicitly.
Thus the work equation related to the PvF yields
Z
δWσ = {δmxx ,x w,x − δmxx · F · ( mxx − ν myy )
+ δmyy ,y w,y − δmyy · F · ( ν mxx − myy )
+ (δmxy ,y w,x + δmxy ,x w,y ) − δmxy · 2(1 + ν) · F · mxy } dA
Z Z
− δmn ϕen ds − δmt w,s ds = 0 . (16.13)

Comparing the contour integrals to the corresponding terms being derived at

the PvD, the symmetry as well as the duality of both of the principles of work
become obvious.
16.1 Mixed Principles of Work Concerning Kirchhoff Plates 289

Matrix Notation of PvD and PvF

To build the subsequently following element matrix it is convenient to represent
the work expressions related to the PvD and to the PvF in matrix notation.
Without considering the load vector it is valid
Z
 
−δW = δw δmxx δmyy δmxy ·
A
   
x∂
∂x ∂y (x ∂ ∂y + y ∂ ∂x )∗
y∂ w
   
 x∂ ∂x −F νF   mxx 
·
  m  dA . (16.14)
 
y ∂ ∂y νF −F

   yy 
(x ∂ ∂y + y ∂ ∂x )∗ −2(1 + ν)F mxy

At the positions indicated by ∗ and besides the domain integrals given here,
the contour integrals related to Eqns. (16.13) and (16.12) are to be evaluated.

16.1.3 Element Matrix and Load Vector

Due to the integration by parts the PvD and the PvF only comprise first order
derivatives of the variables of description, thus linear approaches to describe
[ w mx my mxy ] may be chosen at discretization. The shape functions related
to the virtual as well as to the real state variables exhibit the same courses as
at plane stress structures:

u = Ωv · v with u = [ w(x, y) ] ,
Ωv = [ φA φB φC φD ] ,
vT = [ wA wB wC wD ] ,
 
mxx (x, y)
 
σ = Ωs · s with σ =  myy (x, y)  ,
mxy (x, y)
 
φ φ φ φ
 A B C D 
Ωs = 
 φA φB φC φD ,

φA φB φC φD

sT =
[ mxxA mxxB mxxC mxxD | myyA myyB myyC myyD | mxyA mxyB mxyC mxyD ] .
290 16 Hybrid–Mixed Kirchhoff Plate Elements

Applying the approach for the deflection and the moments to Equation (16.14)
directly leads to the element matrix. Therefore, corresponding to Equation
(16.14), the discretized representation of the virtual work is given summarizing
the terms to be integrated for the element matrix with φT = [ φA φB φC φD ]:
 
0 φT,x φ,x φT,y φ,y (φT,x φ,y + φT,y φ,x )∗
Z  
 φT,x φ,x −F φT φ νF φT φ 0 
  dA .
 T T T 
A φ,y φ,y νF φ φ −F φ φ 0 
T T
(φ,x φ,y + φ,y φ,x )∗ 0 0 −2(1 + ν)F φ φ T

The domain integrals may be integrated analogously to the plane stress element
according to Section 14. Moreover, besides the domain integrals, the contour
integrals are to be evaluated, which are indicated by ( )∗ . Although the contour
integral is only to be evaluated at the free boundary, regarding the computa-
tion in practice, it may be more convenient to evaluate the integral in general
related to each element and to each element boundary. This results in an all–
purpose element matrix, that can be applied to all different possibilities of
support. Disregarding the free boundary the expressions cancel out each other
at assembling the elements to the entire structure.
Assembling the nodal variables in the vector z with
zT = [ wA mxxA myyA mxyA |
wB mxxB myyB mxyB |
wC mxxC myyC mxyC |
wD mxxD myyD mxyD ]
yields the element matrix A, which is given on the next page. Thus the Principle
of virtual Work is discretized on the element level as
−δW = δzT {A · z − f } .

Load Vector
Due to the linear approaches the computation of the load vector takes place
by analogy with the plane stress elements according to Section 14 and thus is
taken over. At the PvD distributed external actions are taken into account by
Z
δWd = δvT ΩT · p dA . (16.15)
A

At the PvF no curvatures are considered arising from non–uniform heating

αT ∆T /d, however, by analogy with plane stress structures, they may be pre-
pared in x– as well as in y–direction.
 lx ly lx ly lx ly ly lx
Abbreviations: F1 = , , , A1 = , A2 = ,
Et3 /12 Et3 /12 Et3 /12
F2 = −ν · F3 = 2(1 + ν) ·
lx ly

0
A1
 

3
− F91
A2
 
 

− F92 − F91
 

0 0
 
 

− F93
 

A2 1
 1 3 

0 3
0+ 2
0
 

− A31
 2 −1 

2 A1
 

3 18 18
0 3
 

−F − F91
 

A2 F2 1 A2
0 . . . symmetric. . .
 

6 18 18 3
 − A1 − F 1 

− −F − F92 − F91
 

3
0 0 0 0
 

2 18
−F − 12 + 1 − F93
 
 

A2 A1 A2
 

0 3
0 6 3
0
 0− 1 

− A61 − 12 + 0 0 − 12
 

2 A1 A1
0 0
A= 

6 36 36 6 3
 

−F − F181 − F182 − F91

 

1 A2
0 0
 

6 36 36 3
−F − A32 − F182 − F181 − F92 − F91
 − A1 − F 1 

3 1
 

0 0 0 0 0
16.1 Mixed Principles of Work Concerning Kirchhoff Plates

 

36
−F 0 + 12 − F183 2
−1 − F93
 − A2 − F 2 
 

A1 A2 1 A2 1 A2 1
0 0+ 0 +0 0 0+ 0
 1 

6 3 2 3 2 3 2
 −2 + 0 0 

− A61 − A31
 

A1 F1 2 A1 A1 2 A1
0 0 0
 

6 18 18 6 3 18 3
− −F − F361 − F362 − F181 − F − F91
 

F1 A2 F2 F1 A2
 

0 0 0
 

6 3
− 18 − A62 − F362 − F361 − 18 − 18 − F92 − F91
 

3 1 1 3
 

0 0 18 2
+0 0 0 2
0 0 18
0 0
 

0 − 12 −F − F363 0− −F − 12 + 1 − F93
 − A32 − F182 
291

 
292 16 Hybrid–Mixed Kirchhoff Plate Elements

16.1.4 Convergence Behavior Concerning the Mixed Element

As an example to test the formulation, again the square plate is investigated
according to Section 6.2, cf. Figure 16-3. By analogy with Section 6.2 and
x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
7 2
E = 3.0 . 10 kN/m
T
t = 0.1 m
ly
G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 16-3 Square plate – system and external actions

concerning different variables of state, the convergence behavior is summarized

in the Table 16.1. In comparison to displacement–based elements an inferior
convergence can be observed regarding the deflection and a partially better
convergence regarding the moments.

Table 16.1 Convergence Behavior of mixed plate elements

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 0. 9524 0. 8364 0. 7640 0. 7435 0. 7381
mxxM [ kN m/m ] 4. 167 2. 627 2. 262 2. 181 2. 161
myyM [ kN m/m ] 6. 548 4. 100 3. 367 3. 216 3. 178
myyE [ kN m/m ] −5. 952 −6. 731 −6. 913 −6. 966 −6. 979
mxyT [ kN m/m ] ÷ −1. 388 −1. 633 −1. 548 −1. 539

Although the element converges against the accurate solution, the mixed plate
element comprises some properties, which may limit the applications.
• Investigating oblique plates leads to problems with the fomulation of the
boundary conditions with respect to the moment mn at hinged or free
boundaries, since a transformation of the moment is needed there.
• The computation of stiffened plates is impossible, if the stiffeners have to
be discretized as well. In these cases the bending moment can not be a
nodal degree of freedom.
16.2 Hybrid–Mixed Rectangular Plate Element 293

16.2 Hybrid–Mixed Rectangular Plate Element

The mixed formulation dealing with the variables [ w mxx myy mxy ] concerning
the element domain and already known from Section 16.1, is adopted. At the
interfaces the deflection related to the corner nodes are defined as degrees of
freedom of the system. To link the corner nodes, linear approaches are chosen
to describe the deflection. To ensure the C1 –conformity and by analogy with
the Euler–Bernoulli beam additional rotations with respect to the interfaces
are defined. The following degrees of freedom are possible:
• at the nodes of the system A, B, C, D :
w, ϕx , ϕy → 12 degrees of freedom
This formulation is comparable to the displacement–based formulation
comprising 12 DOF, cf. Section 6.3.
In contrast to the Euler–Bernoulli beam different approaches can be chosen
here to describe the degrees of freedom at the interfaces:
• w, ϕx , ϕy each empploying a linear course along the interface. The shape
functions are only defined at the interface, see Figure 16-4.
• ϕn , ϕt employing a constant or linear course along the interface between
the nodes of the system. ϕn and ϕt are not defined according to the
corners.

x A
y
interface ’I’
z
a

B
D d element domain ’E’ b
1
1
c

C
1
1

Fig. 16-4 Linear approaches concerning w, ϕx , ϕy at the interfaces

Subsequently, the general procedure at computing the element matrices is

shown. The element exactly fulfills the element continuity conditions related to
w. The element continuity conditions concerning w,n and w,s are ensured by
the rotations ϕx and ϕy at the interfaces. In the following section this element
is referred to as K-HM-4-16.
294 16 Hybrid–Mixed Kirchhoff Plate Elements

16.2.1 Work Equations

The formulation of the equilibrium applying the PvD takes place, corresponding
to the mixed principle of work presented in Section 16.1, considering kinematics

κxx = w,xx , κyy = w,yy , κxy = w,xy in the element domain.

Cutting cleanly the interfaces yields the shear forces and the moments at the
element boundaries E and at the interfaces I according to Figure 16-1. Incor-
porating the work performed at the transition between element boundary and
interface results in:
Z
δWd = {δw,xx · mxx + δw,yy · myy + 2δw,xy · mxy + δw · p} dA
AZ Z
− mn · (δw,n E − δϕn I ) ds − mt · (δw,s E − δw,s I ) ds
Z
+ qn · (δwE − δwI ) ds .

Enforcing continuity of real as well as virtual displacements wE = wI and

δwE = δwI and, by analogy with the beam, integrating the internal work by
parts, yields the virtual work performed on the virtual deflection surface
Z
− δWd,w = {δw,x · mxx,x + δw,y · myy,y + δw,y · mxy,x + δw,x · myx,y } dA
A Z Z
− δw · p dA + mt · δw,s ds (16.16)
A I

as well as performed on the virtual rotations at the interfaces

Z
δWd,ϕ = − mn · δϕn ds . (16.17)
I

The formulation of the deformation conditions takes place by analogy with

Section 16.1 applying the PvF, whereby the work δmn ϕn is to be taken into
account.
Z
12
− δWσ,mxx = {δmxx ,x w,x − δmxx 3 (mxx − ν · myy )} dA
A Z Et
− δmn ϕn ds = 0 , (16.18)
I
Z
12
− δWσ,myy = {δmyy ,y w,y − δmyy (myy − ν · mxx )} dA
A Z Et3
− δmn ϕn ds = 0 , (16.19)
I
16.2 Hybrid–Mixed Rectangular Plate Element 295
Z
12
− δWσ,mxy = {δmxy ,y w,x + δmxy ,x w,y − 2δmxy 3 (1 + ν) · mxy }E dA
A Z Et
− δmt w,s ds = 0 . (16.20)
I

All work expressions related to the interfaces are

>0 at the positive cut

<0 at the negative cut.

16.2.2 Element Stiffness Matrix

Employing the approaches to describe the deflection and the moments
[ w ] = Ωv · w ,
σ = Ωs · s , σ T = [ mxx myy mxy ] ,
as well as the degrees of freedom related to the nodes

wT = [ wA wB wC wD ]

and the degrees of freedom related to the moments

sT = { [ mxx myy mxy ]a [. . .]b [. . .]c [. . .]d }

the work performed with respect to the element domain yields

   
δw : Z 0 (ΩTv DT ) (DΩs ) w
  dA ·  .
δs : A (ΩTs DT ) (DΩv ) −ΩTs E−1 Ωs s
When preparing the work equation with consideration of the parts related to
the interface the rotations are to be taken into account as additional degrees
of freedom. With the nodal displacement variables

vT = [ wT ϕT ] = [ wA wB wC wD ϕxA ϕxB ϕxC ϕxD ϕyA ϕyB ϕyC ϕyD ]

the matrix notation of the work equations follows – in a first step without the
parts related to external work –
 R T T   
δw : 0 0 Ωv D D Ωs dA + ( )∗ w
   
 R T   
   
δϕ :  0 0 Ωϕ Ωs ds  ·  ϕ .
   
R R R   
δs : ΩTs DT D Ωv dA + ( )∗ ΩTs Ωϕ ds − ΩTs E−1 Ωs dA s
296 16 Hybrid–Mixed Kirchhoff Plate Elements

Here, the expression ( )∗ alludes to the contour integrals, that are to be consi-
dered at the element boundary. The domain integrals
Z Z
(ΩTv DT ) (D Ωs ) dA and ΩTs E−1 Ωs dA

as well as the contour integrals ( )∗ may be directly taken from the mixed
formulation.
R
Integrating the additional work expressions at the interface ΩTϕ · Ωs ds and
employing linear approaches to describe ϕx , ϕy yields, regarding rectangular
elements, the matrix
A B C D
mxx myy mxy mxx myy mxy mxx myy mxy mxx myy mxy
 ly ly 
δϕxA : 3 6
 
 lx lx 
δϕyA : 3 6 
 
 −ly −ly 
δϕxB : 3 6


 
lx lx
δϕyB : 6 3


 .
 −ly −ly 
δϕxC : 6 3

 
 −lx −lx

δϕyC : 3 6


 
 ly ly 
δϕxD : 6 3 
 
−lx −lx
δϕyD : 6 3

To eliminate the moments related to the element domain the matrix notation
of the element matrix fits best:
" # " # " #
δv : 0 HTsv ∗ v f 6= 0 at element domain
· −
δs : Hsv ∗
−F s 0 = 0 at element domain

Thereby the contour integrals related to the element boundaries are taken into
account at the positions ∗. The elimination of the moments related to the
element domain yields the virtual work described with the PvD – including the
external work –

− δWd = δv { (0 + HTsv · F−1 · Hsv ) · v − f } . (16.21)

16.2 Hybrid–Mixed Rectangular Plate Element 297

16.2.3 Convergence Behavior of the Hybrid–Mixed Element

As an example to test the formulation the square plate loaded by a constantly
distributed external action according to Section 6.2 is investigated again, see
Figure 16-5. Employing symmetry the behavior of convergence with respect to
the midpoint deflection as well as to especially chosen values of moments is
investigated at mesh refinement. The discretizations are defined according to
the element mesh for the quarter plate represented in the figure.

x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
E = 3.0 . 107 kN/m2
T
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 16-5 Kirchhoff plate – geometry and loading

Table 16.2 summarizes the results concerning the state variables related to
the hybrid–mixed element employing 12 degrees of freedom. Comparing the
results to the results of the plate elements according to Sections 6.2, 6.3 as
well as Section 16.1.4 illustrates, that the convergence concerning the state
variables of the hybrid–mixed element is only marginally inferior despite of the
considerably lower polynomial order of the shape functions.

Table 16.2 Convergence behavior of the hybrid–mixed plate element

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 0. 8572 0. 8080 0. 7556 0. 7411 0. 7375
mxxM [ kN m/m ] 3. 750 2. 512 2. 237 2. 174 2. 159
myyM [ kN m/m ] 5. 893 3. 798 3. 320 3. 204 3. 147
myyE [ kN m/m ] −5. 357 −6. 666 −6. 911 −6. 966 −6. 979
mxyT [ kN m/m ] −0. 714 −1. 149 −1. 569 −1. 597 −1. 578
−1. 225 −1. 225 −1. 404 −1. 478
298 16 Hybrid–Mixed Kirchhoff Plate Elements

Employing the same number of degrees of freedom the results of the displace-
ment–based element comprising 12 degrees of freedom are slightly more accu-
rate, because of the higher order shape functions to describe w. This can be
observed explicitly at incorporating distributed external actions and bedding
respectively. Further differences occur regarding the expressions related to ben-
ding, too. But an advantage of the K-HMQ-4-16 element is the less effort at
integration of the element stiffness matrix due to the lower order approaches.
HYBRID TRIANGULAR PLATE ELEMENTS
17 Hybrid Triangular Plate Elements

In Section 11.1 different triangular plate elements are investigated according

to Kirchhoff ’s plate theory. As a result it can be stated that a triangular ele-
ment is possible employing a conform approach of 5th order with 21 degrees of
freedom, which has some disadvantages with respect to applications. Hence the
question is raised whether it is possible in general to also receive conformity
with approaches of less order and without curvatures as nodal degrees of free-
dom. It seems to be meaningful to investigate this regarding, as an example,
the cubic approach also being conform, when concerning rectangular elements.

From Section 11.1 it is known, that the triangular plate element employing a
cubic approach to describe w and therefor including 10 degrees of freedom –
as represented in Figure 17-1 – is not conform and converges to an incorrect
solution in particular cases.

w
x w,x
y z w,y w,s
A
w,n
D+w w
B w,x
C w,y
w
w,x
w,y

Fig. 17-1 Displacement–based 10 degrees of freedom plate element

Concerning this plate element, at each boundary 6 degrees of freedom are availa-
ble. Due to the shape functions required to be cubic polynomials along the
boundary with respect to w employing four degress of freedom and to be qua-
dratic polynomials with respect to w,n with three degrees of freedom to fulfill
the continuity conditions by the approach itself, one degree of freedom is mis-
sing to ensure the C1 –conformity. Therefore, the slopes w,n are not continuous,
which causes a bad behavior of convergence. Subsequently different possibilities
are presented to ensure C1 –conformity, whereupon the cubic approach to des-
cribe the deflection provides a basis for the discretization in each case according
to Section 17.1.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_17
302 17 Hybrid Triangular Plate Elements

One possibility deals with elements employing 10+3 degrees of freedom, of

which three degrees of freedom are defined related to each corner, one degree is
related to the center of the triangle and one degree of freedom to each midpoint
of an edge, cf. Section 17.2 and Section 18. Another possibility aims for elements
employing nine degrees of freedom, relating to the deflection as well as to the
rotations [ w w,x w,y ] with respect to the corners of the element, cf. Section
17.3 and Section 19.

A complete different way to overcome this problem has been developed by

Hsieh, Clough and Toucher and is published as the HCT element by Clough
and Tocher [27]. The HCT element is a triangle dealing with nine degrees
of freedom employing [ w w,x w,y ] as unknowns at the three corners of the
element, and describing the slope normal to the edges of the element as a linear
course. The triangular element consists of three sub–triangles, each employing
an incomplete cubic approach with nine degrees of freedom. Thus, the element
overall employs 27 degrees of freedom. The continuity conditions between the
sub–elements result in 18 kinematic constraints, which are used to eliminate the
internal variables by means of static condensation. Due to its special structure
the element is not discussed here.

17.1 Cubic Approach for Triangular Plate Elements

Concerning triangular elements, the approach to describe the deflection is cho-
sen according to the displacement–based formulation

u = [w] = ψ · a, (17.1)

whereat, regarding area coordinates, the complete cubic approach is given by

ψ = [ λ3a λ2a λb λ2a λc λ3b λ2b λc λ2b λa λ3c λ2c λa λ2c λb λa λb λc ] .

The scaling of the approach with respect to physically meaningful degrees of

freedom takes place by analogy with Section 11. The deflection and the rotati-
ons related to the element nodes as well as the deflection at the barycenter are
chosen as physically meaningful nodal degrees of freedom. Applying the scaling
matrix Ψ̃ comprising the shape functions being evaluated with respect to the
nodal displacement variables v it follows

v = Ψ̃ · a .
17.1 Cubic Approach for Triangular Plate Elements 303

The nodal displacement variables are defined with

vT = { [ w w,x w,y ]a [ w w,x w,y ]b [ w w,x w,y ]c [ w ]d }

while the matrix comprising the approach Ψ̃ is given by

 
2A 0 0
 3b b b 0 0 0 
 a b c 
 3a a ac 
 a b 
 
 2A 0 0 
 
 0 3bb bc ba 0 0 
  1
Ψ̃ =  3ab ac aa ·
 2A .
 
 
 2A 0 0 
 
 0 0 3bc ba bb 0 
 
 3ac aa ab 
 
2 2 2 2 2 2 2 2 2 2
27 A 27 A 27 A 27 A 27 A 27 A 27 A 27 A 27 A 27 A

Here, A represents the domain of the triangular element. The scaling matrix
G is defined as the reversed representation of Ψ̃
−1
G = Ψ̃ ,
 
1 0 0
 3 ac −bc 0 0 0 
 
 3 −ab bb 
 
 
 
 1 0 0 
 
 0 3 aa −ba 0 0 
G=
 3 −ac bc
 . (17.2)

 
 
 1 0 0 
 
 0 0 3 ab −bb 0 
 
 3 −aa ba 
 
−7 ab −ac bc −bb −7 ac −aa ba −bc −7 aa −ab bb −ba 27

Thus, the approach scaled to physically meaningful degrees of freedom follows

to
−1
u = [ w ] = ψ · Ψ̃ ·v = ψ ·G·v.
304 17 Hybrid Triangular Plate Elements

17.1.1 Elimination of wd at Element Level

The computation of the element stiffness matrix K may be followed by the
elimination of the nodal degree of freedom wd applying static condensation,
since it is defined only in the element domain and is solely dependent on the
other degrees of freedom of the respective element. Considering i, j = 1, . . . 9
and d = 10 the discretized work equation at element level follows to
" # " # " #
kji kjd vi fj 6= 0 ,
· −
kdi kdd wd fd = 0.

Subscripting the equation to be eliminated by d, relating to the last row yields

[ kdd ] · wd + [ kdi ] · vi = [ fd ] ,
wd = [ kdd ]−1 · { [ fd ] − [ kdi ] · vi } .

Replacing wd in the first row, the stiffness matrix is lowered by one line and
one column to

{ [ kji ] − [ kjd ] · [ kdd ]−1 · [ kdi ] } · vi = [ fj ] − [ kjd ] · [ kdd ]−1 · [ fd ] .

If necessary the eliminated midpoint deflection wd may be computed in a sub-

sequent analysis.

17.1.2 Subsequent Analysis Concerning the Stress Resultants

Following the computation of the nodal displacement variables the moments
[ mxx myy mxy ] and the shear forces [ qx qy ] may be computed at the element
level in a subsequent analysis by analogy with Section 11.2. Therefore, the
material equations and the equations of equilibrium are applied as follows:

mxx = −B (w,xx + ν · w,yy ) ,

myy = −B (w,yy + ν · w,xx ) ,
mxy = −B (1 − ν) w,xy ,
qx = mxx ,x + mxy ,y = −B (w,xxx + w,xyy ) ,
qy = mxy ,x + myy ,y = −B (w,yyy + w,xxy ) .

Independent of the chosen element formulation, these equations are applied in

the subsequent sections, each comprising the cubic approach according to the
element domain.
17.1 Cubic Approach for Triangular Plate Elements 305

17.1.3 Hybrid Elements to Ensure C1 - Conformity

At the element domain the followig 10 degrees of freedom
T
vE = { [ w w,x w,y ]a, b, c [ w ]d }E

are defined related to the scaled cubic approach. The element continuity con-
ditions concerning w and w,t are fulfilled by the approach, if [ w w,x w,y ]a, b, c
and [ w w,x wy ]A, B, C are equal, see Figure 17-2.

b
x A
y z a d +
a
D
x
F x c
x x d + b
x B
x
c E
C

Fig. 17-2 Hybrid triangular element – element domain and interface

Due to w being cubic, the first order derivative with respect to the boundary
coordinate s is described quadratically and is directly defined by w. Thus in
total four nodal displacements are allocated related to the tails of the element
boundary concerning the cubic course. Two degrees of freedom are left to li-
nearly describe the course of w,n along the boundary and at the interface. To
create an element continuity ensuring conformity a quadratic course also is nee-
ded with respect to w,n . The fulfillment of the element continuity conditions
with respect to w,n succeeds applying a hybrid formulation of the work related
to the element.
To ensure the element continuity conditions two principle methods of resolution
are possible:

1. Derivating elements employing additional degrees of freedom at the in-

terface between neighboring elements, that may exhibit different physical
meanings. Thus the element continuity conditions are fulfilled strongly.
2. Derivating elements without additional degrees of freedom at the element
interfaces, but taking into account additional work performed between
element boundary and interface. Thus the element continuity conditions
are fulfilled weakly applying the work equation.
306 17 Hybrid Triangular Plate Elements

17.2 Hybrid–Displacement Elements Employing 10+3 DOF

By analogy with other Kirchhoff plate elements at system level the work equa-
tion
Z
−δWd = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA = 0 (17.3)
A

is valid. The terms of virtual work that vanish at assembling the element for-
mulation, are not considered concerning Equation (17.3).

quadratic part

linear part w,n

w,n B
A

Fig. 17-3 Course of rotation ϕn along interface

Applying the cubic approach according to Equation (17.1) the element conti-
nuity conditions related to the rotations w,n and related to the virtual rotations
δw,n cannot be fulfilled, since at the element boundary w,n is only described
linearly by the slopes w,n related to the corners. Therefore additional conditi-
ons and degrees of freedom are to be defined at the interface, that consider the
higher polynomial parts, cf. Figure 17-3. This goal can be reached following
different paths.

17.2.1 Version A Employing Lagrange Multipliers

This idea to fulfill the element continuity conditions has been proposed by
Harvey and Kelsey in [42] – in the following it is referred to as K-MT-6-12.
At system level the work equation applying the PvD is arranged by analogy
with the preparing of element formulations that ensure conformity. In addition
the constraints according to the element interface are arranged employing the
Method of Lagrange multipliers. Thereby the line of action is not motivated by
virtual work but by a kinematic constraint only.
At the interfaces and in addition to the displacement variables at the element
corners [ w w,x w,y ] the Lagrange multipliers λ are chosen related to the mid-
points of the edges, cf. Figure 17-2, and it follows

vT = { [ w w,x w,y ]A, B, C [λ]D, E, F } .

17.2 Hybrid–Displacement Elements Employing 10+3 DOF 307

The Lagrange multiplier λI is defined at the interface I and is connecting

the slopes w,n of the neighboring elements i and i + 1. Thus it exhibits the
physical meaning of a bending moment mn at the midpoint of an edge. The
work equation applying the PvD is extended to the constraints now:
Z
−δW = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA
Z A

+ ΣI {(δw,n |i − δw,n |i+1 ) · λI + δλI · (w,n |i − w,n |i+1 )} ds . (17.4)

I
Concerning the slope w,n and according to Figure 17-4
w,n = −w,x · sin β + w,y · cos β
is valid, introducing β to be the angle between the x–axis and the corresponding
edge of the triangle. The trigonometric functions may be computed introducing
the differences of the coordinates related to the corner nodes of the respective
edge, whereat i, i + 1 has to be counted clockwise. Therefor it is stated
p
∆x = xi+1 − xi , ∆y = yi+1 − yi and ℓi+1
i = ∆x2 + ∆y 2 ,
∆y ∆x
sin β = and cos β = .
ℓ ℓ
x
w,s
w, n
D
s w,y s
β
td td
y
nd w,x nd

Fig. 17-4 The transformation of w,s and w,n with respect to w,x and w,y

It seems to be meaningful to evaluate the contour integrals at element level, so

that an additional degree of freedom λI is assigned to each element boundary.
Thus at element level the following terms of work are to be evaluated:
Z
−δW = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA
Z A

+ ΣI=1,2,3 {δw,n · λI + δλI · w,n )} ds . (17.5)

I
308 17 Hybrid Triangular Plate Elements

After integration of the work employing the cubic approach to describe w ac-
cording to Section 17.1 matrix notation of the element stiffness matrix yields
" #
kww kwλ
K= .
kλw 0

Applying the complete element matrix and the respective load vector the bary-
centric–related deflection wd may be eliminated according to Section 17.1.2.
After allocation of the elements to the entire system of equations the element
continuity conditions related to the slopes arise as
Z
−δWλ|I = {δλI · (w,n |i − w,n |i+1 )} ds = 0 (17.6)
I

and hence are fulfilled strongly at applying the element.

At performing analyses it has to be taken notice of the fact, that Lagrange
multipliers are to be introduced to be zero at all simply–supported as well as
free system boundaries, since they exhibit the physical meaning of the bending
moment mn . This denotes, that the element corresponds to a plate element in
mixed force–displacement formulation.
The only drawback of the element is the zero–stiffness entries at the main dia-
gonal with respect to the Lagrange multipliers. This results in a semi–definite
system of equations, that requires a special equation solver concerning indivi-
dual cases.

Test – Version A
As an example to test the formulation given above the square plate according
to Section 6.2.9 is investigated, cf. Figure 17-5.
x clamped

E constantly distributed loading: p = 1 kN/m2

y
T E = 3.0 . 107 kN/m2
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 17-5 Square plate – system and loading

17.2 Hybrid–Displacement Elements Employing 10+3 DOF 309

In the subsequent Table 17.1 and Table 17.2 the behavior of convergence concer-
ning the element employing displacements as primary variables and additional
Lagrange multipliers is summarized with respect to different state variables.
Relating to the nodes, that are belonging to multiple elements and at differing
values, the results are given for the respective neighboring elements.

Table 17.1 Square plate – results related to distributed action p = 1 kN/m2

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 4000 0. 6597 0. 7195 0. 7322 0. 7352
mxxM [ kN m/m ] 0. 5000 2. 384 2. 498 2. 374 2. 268
myyM [ kN m/m ] 2. 500 3. 632 3. 587 3. 412 3. 293
myyE [ kN m/m ] −2. 500 −3. 741 −5. 119 −6. 051 −6. 533
mxyT [ kN m/m ] −2. 000 −1. 069 −1. 346 −1. 479 −1. 515
0. 000 −1. 294 −1. 493 −1. 544 −1. 546
−1. 069 −1. 346 −1. 479 −1. 515

Table 17.2 Square plate – results related to concentrated load P = 1 kN in M

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 0400 0. 0874 0. 1023 0. 1064 0. 1076
mxxM [ kN m/m ] 0. 0500 0. 3753 0. 6586 0. 9248 1. 188
myyM [ kN m/m ] 0. 2500 0. 5918 0. 8375 1. 071 1. 314
myyE [ kN m/m ] −0. 2500 −0. 3798 −0. 5281 −0. 6039 −0. 6374
mxyT [ kN m/m ] −0. 2000 −0. 1006 −0. 1219 −0. 1348 −0. 1415
−0. 0000 −0. 2378 −0. 1880 −0. 1669 −0. 1575
−0. 1006 −0. 1219 −0. 1348 −0. 1415

The comparison of the results applying the plate elements according to Sec-
tion 6.2.9 as well as to Sections 10.2 and 10.3 demonstrates, that the results
related to the hybrid element employing displacements as primary variables
and additional Lagrange multipliers exhibit worse convergence due to the lower
polynomial degree of the respective shape functions. In particular, the conver-
gence of the stress resultants is not sufficient. Nevertheless the results illustrate
that the method of Lagrange multipliers is appropriate to ensure the element
continuity conditions in principle requiring little effort.
310 17 Hybrid Triangular Plate Elements

17.2.2 Version B Employing Rotational Degrees of Freedom

In Section 17.2.1, applying the method of Lagrangian multipliers, a kinematic
constraint has been formulated to fulfill the inter–element continuity concer-
ning the slope w,n , yielding a hybrid–mixed formulation of the work equations.
A completeley different formulation is possible, when arranging the equilibrium
of moments mn related to neighboring elements employing the PvD at the in-
terface. This leads to a hybrid–displacement formulation of the work equations
and the K-HDT-6-12 triangular element may be derived.

Here, the work equation applying the PvD is extended to further terms of work
concerning the element boundaries and the interfaces. By analogy with the
hybrid–mixed plate element the rotations ϕn at the interfaces are chosen as
additional variables to ensure the equilibrium of moments mn at the element
interfaces. Thus the moments mn perform virtual work on the slopes w,n at
the element boundaries and on the rotations ϕn at the interfaces respectively.
Z
−δWd = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA
Z Z A

+ δw,n · mn ds − δϕn · mn ds . (17.7)

E I

The terms related to the virtual work, that vanishes at assembling the elements,
are not taken into account at Equation (17.7). This affects the work performed
by the twisting moment mt and by the shear forces qn or rather Kirchhoff ’s
effective shear force qKn = qn + mt ,s on the deflection w and on the rotation
w,s respectively.
At assembling the vitual work of the elements to the entire structure or rather
to the system of equations the condition of equilibrium related to the bending
moments at the element interface
Z
δϕn · ( mn (i) − mn (i+1) ) ds = 0
I

directly arises as a constraint in the system of equations.

The continuity condition related to w,n between the element boundary and the
interface can be fulfilled applying the PvF at element level, when considering
the work performed by the virtual moment δmn on the rotation of the element
boundary as well as of the interface. Due to the linear part of w,n being fulfilled
already by the cubic approach to describe w, the rotation ϕn only has to fulfill
17.2 Hybrid–Displacement Elements Employing 10+3 DOF 311

the element continuity condition related to the quadratic part of w,n . This is
to be caught at formulating the element continuity condition with
Z Z
−δWσ = + δmn · ( w,n |quadr. − w,n |linear ) ds − δmn · ϕn ds . (17.8)
E I

and respective discretization.

In addition to the rotations ϕn at the interfaces the work equations (17.7)

and (17.8) comprise the cutted cleanly bending moments mn . These are trans-
formed together with the moments [ mxx myy mxy ] into the x–y–coordinate
system. Afterwards they are described by means of the material equations and
thus by the curvatures [ w,xx w,yy w,xy ], so that an element is generated pu-
rely employing displacements as primary variables. As a first step it follows,
regarding the element boundary,

mn = mxx · sin2 β + myy · cos2 β − 2mxy · cos β · sin β

Here, β is defined as the angle between the x–axis and the respective edge of
the triangle, cf. Figure 17-4, whereat sin β and cos β are computed by means
of the coodinate differences of the corners as given in Section 17.2.1:
p
∆x = xi+1 − xi , ∆y = yi+1 − yi and ℓi+1
i = ∆x2 + ∆y 2 ,
∆y ∆x
sin β = and cos β = .
ℓ ℓ
Introducing the material equations yields

mn = −B (w,xx + νw,yy ) sin2 β + (w,yy + νw,xx ) cos2 β

− 2w,xy (1 − ν) cos β sin β element boundary

.

As an alternative the curvatures at the element boundary may be applied to

evaluate the moments mn

mn = −B ( w,nn + νw,ss ) ,

which results, at introducing the transformation rule for the derivatives, in the
same formulation related to the x–y–coordinates.
Furthermore the slope w,n is to be transformed at the element boundary. Ac-
cording to Section 17.2.1 it is obtained

w,n = −w,x · sin β + w,y · cos β .

312 17 Hybrid Triangular Plate Elements

Introducing mn and w,n into the work Equations (17.7) and (17.8), employing
the cubic approach to describe w, and a quadratic approach to describe ϕn
yields the stiffness matrix according to Section 17.1
" #
kww kwϕ
K=
kϕw 0

and the vector of nodal displacement variables at the interface

vT = { [w w,x w,y ]A, B, C [ϕn ]D, E, F } .

Thereby kww (size 10 × 10) represents the work performed at the element
domain and at the element boundary, kwϕ = kTϕw (size 3 × 10) represents the
work performed at the interface. To numerically integrate the work concerning
the element domain a 3–points Gauss–Legendre integration is sufficient, cf.
Section 12.1.

Test – Version B
As an example to test the formulation given above the square plate according
to Section 6.2.9 is investigated, see Figure 17-6.
x clamped

E constantly distributed loading: p = 1 kN/m2

y
T E = 3.0 . 107 kN/m2
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 17-6 Square plate – system and loading

In Table 17.3 and Table 17.4 the behavior of convergence concerning the hy-
brid element employing displacements as primary variables and 10+3 element
degrees of freedom is summarized with respect to different state variables. Re-
lating to the nodes M and T , that are belonging to multiple elements, the
results are given for the respective neighboring elements. The gap of the mo-
ments between different elements is an indication of the discretization error.
17.2 Hybrid–Displacement Elements Employing 10+3 DOF 313

Comparing the results to those related to plate elements according to Sec-

tion 11 illustrates, that the convergence for the hybrid–displacement element
is slightly inferior due to the lower polynomial order of the shape functions.
In comparison to the displacement–based element employing approaches of 5th
order, however the element discussed here is more generally applicable due
to the nodal displacement variables [ w w,x w,y ϕn ], since it incorporates no
curvatures as degrees of freedom. Comparing to the element according to Sec-
tion 17.2.1, also the moments converge quite fast against the reference solution.

Table 17.3 Square plate – results related to distributed action p = 1 kN/m2

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 6785 0. 7231 0. 7358 0. 7361 0. 7361
mxxM [ kN m/m ] 3. 169 2. 291 2. 258 2. 168 2. 180
2. 651 2. 201 2. 248 2. 149 2. 179
myyM [ kN m/m ] 4. 705 3. 935 3. 350 3. 197 3. 201
2. 758 3. 882 3. 315 3. 179 3. 190
myyE [ kN m/m ] −4. 241 −6. 066 −6. 746 −6. 917 −6. 971
mxyT [ kN m/m ] −1. 750 −0. 7656 −1. 546 −1. 536 −1. 545
+0. 107 −2. 082 −1. 608 −1. 567 −1. 538
−1. 861 −1. 595 −1. 561 −1. 524

Table 17.4 Square plate – results related to concentrated load P = 1 kN in M

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 0742 0. 1025 0. 1069 0. 1078 0. 1080
mxxM [ kN m/m ] 0. 3786 0. 6528 0. 9428 1. 207 1. 395
0. 3071 0. 6008 0. 9175 1. 170 1. 362
myyM [ kN m/m ] 0. 5214 0. 7649 1. 017 1. 276 1. 447
0. 3357 0. 7593 1. 036 1. 288 1. 494
myyE [ kN m/m ] −0. 4643 −0. 5502 −0. 6338 −0. 6572 −0. 6520
mxyT [ kN m/m ] −0. 2000 −0. 1334 −0. 1355 −0. 1547 −0. 1803
+0. 0286 −0. 2214 −0. 1656 −0. 1506 −0. 1496
−0. 1698 −0. 1469 −0. 1518 −0. 1626
314 17 Hybrid Triangular Plate Elements

17.3 Displacement–based Element with Weak Conformity

As suggested in Section 6.4, in the case of non–conform approaches the element

continuity conditions also may be fulfilled weakly without employing additional
degrees of freedom by taking into account the work performed at the element
interfaces. Subsequently this path is described, that results in the formulation
of the element K-HDT-3-9 employing solely displacements as primary variables
and that incorporates nine degrees of freedom. By analogy with other Kirchhoff
plate elements the work equation applying the PvD is valid at system level
Z
−δWd = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA = 0 . (17.9)
A

At the element domain the work equation applying the PvD is extended to
the work at the element boundaries as well as at the interfaces. In contrast to
Section 17.2.2, at the element boundary the moment mn performs work on the
total rotation w,n |quadr. and at the interface the resultant moment mn performs
work on the rotations w,n |linear existing at the interface, without requiring an
additional variable. Thus the work equation applying the PvD yields
Z
−δWd = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA
Z Z A

+ δw,n |quadr. · mn dsE − δw,n |linear · mn dsI . (17.10)

E I

The equilibrium condition related to mn is fulfilled weakly hereby. The conti-

nuity condition, related to w,n at the element boundary and at the interface,
is fulfilled applying the PvF. By analogy with Section 17.2.2 it is obtained now
Z Z
−δWσ = + δmn · w,n |quadr. dsE − δmn · w,n |linear dsI = 0 . (17.11)
E I

The additional terms at the element boundary E as well as at the interface I

describe the work performed by the virtual moments δmn on the slopes w,n
at the element boundaries and at the interfaces. Subsequently and by analogy
with Section 17.2.2 the resultant moment mn and the resultant virtual moment
δmn are described by the curvatures, so that the element only exhibits the nine
degrees of freedom related to the element corners.
17.3 Displacement–based Element with Weak Conformity 315

Test
As an example to test the formulation given above the square plate according to
Section 6.2.9 and Figure 17-7 is investigated. In Table 17.5 and Table 17.6 the
convergence behavior is summarized with respect to different state variables.
x clamped

E constantly distributed loading: p = 1 kN/m2

y
T E = 3.0 . 107 kN/m2
t = 0.1 m
G M ly ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 17-7 Square plate – system and loading

It is substantial, that the results converge against the reference solution, al-
though only the displacement variables [ w w,x w,y ] related to the corner nodes
are defined as degrees of freedom and the element continuity condition is only
fulfilled weakly concerning w,n . Comparison with the results obtained by the
plate element according to Section 17.2 demonstrates, that the results similarly
good converge against the reference solution. Only the clamping moment at
position E shows a slightly inferior convergence.

Table 17.5 Square plate – results related to distributed action p = 1 kN/m2

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 6896 0. 7289 0. 7343 0. 7359 0. 7361
mxxM [ kN m/m ] 1. 482 2. 411 2. 145 2. 147 2. 150
3. 319 2. 317 2. 170 2. 152 2. 152
myyM [ kN m/m ] 1. 206 2. 975 3. 030 3. 120 3. 152
4. 801 3. 105 3. 135 3. 146 3. 158
myyE [ kN m/m ] −4. 310 −5. 456 −6. 096 −6. 507 −6. 737
mxyT [ kN m/m ] −1. 586 −1. 322 −1. 532 −1. 541 −1. 537
−0. 827 −1. 459 −1. 517 −1. 537 −1. 539
−1. 862 −1. 646 −1. 570 −1. 544
316 17 Hybrid Triangular Plate Elements

Table 17.6 Square plate – results related to concentrated load P = 1 kN in M

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512

wM [ cm ] 0. 0744 0. 1005 0. 1060 0. 1075 0. 1079

mxxM [ kN m/m ] 0. 4034 0. 6316 0. 8838 1. 151 1. 417
0. 2241 0. 5940 0. 8494 1. 113 1. 378
myyM [ kN m/m ] 0. 5276 0. 6481 0. 9253 1. 199 1. 467
0. 2103 0. 6900 0. 9716 1. 240 1. 507
myyE [ kN m/m ] −0. 4655 −0. 5699 −0. 6025 −0. 6303 −0. 6463
mxyT [ kN m/m ] −0. 1793 −0. 1057 −0. 1435 −0. 1496 −0. 1498
−0. 0414 −0. 1602 −0. 1445 −0. 1468 −0. 1479
−0. 1984 −0. 1624 −0. 1538 −0. 1508

Although the element offers a very good performance regarding the presented
example, a general application is restricted to element geometries, where the
internal angles are limited to about α, β, γ < 100o.
18 Hybrid–Mixed Triangular Plate Elements

The governing equations related to Kirchhoff ’s plate theory are given in Sec-
tion 6. The weak formulation of the governing equations is developed applying
the hybrid–mixed principle of work according to Section 16, where the discreti-
zation employing rectangular elements is represented. Subsequently the work
equations are picked up again and are extended with respect to the demands
on triangular elements. By analogy with Section 17 the element is split into
element domain, element boundary and interface, cf. Figure 18-1.

b
x A
y z a d +
a
D
x
F x c
x x d + b
x B
x
c E
C

Fig. 18-1 Hybrid triangular element

18.1 Work Equations

By analogy with Section 17 the governing equations are arranged equivalently
applying the principle of virtual work.

The Principle of Virtual Displacements – PvD

Regarding Kirchhoff plates the Principle of virtual Displacements comprises the
internal work performed by the moments and the external work performed by
the distributed actions p(x, y) as well as the work performed on the boundaries.
At element level this yields
Z
δWd = {mxx · δκxx + myy · δκyy + 2 · mxy · δκxy + p · δw} dA
A Z

+ {mn · δw,n + mt · δw,s + qn · δw} ds

ZE
− {mn · δw,n + mt · δw,s + qn · δw} ds ,
I

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_18
318 18 Hybrid–Mixed Triangular Plate Elements

whereupon the work at the contour is performed by the moments mn and mt

on the conjugated rotations, see Figure 18-2. When required the PvD may

x
y z w ,s
w ,n
w ,s w ,n
s s
s

w ,n
w ,s

Fig. 18-2 Slopes at element boundaries

be extended to the work performed by concentrated and line–shaped external

actions respectively. Introducing kinematics related to the virtual curvatures
and integrating the domain terms by parts yields
Z
δWd = − { δw,x · mxx,x + δw,y · myy,y + δw,x · mxy,y + δw,y · mxy,x } dA
A
Z Z
+ qn · δw ds − {mn · δw,n + mt · δw,s + qn · δw} ds .
E I

In the case the virtual displacements are continuous between the element boun-
dary and the interface, it follows
Z Z
qn · δw ds − qn · δw ds = 0
E I

and thus
Z
δWd = − { δw,x · mxx,x + δw,y · myy,y + δw,x · mxy,y + δw,y · mxy,x } dA
A
Z
− {mn · δw,n + mt · δw,s } ds . (18.1)
I

The Principle of Virtual Forces – PvF

The PvF describes the conditions of deformation related to the element do-
main and to the transition from element boundary to interface by an integral
formulation. Incorporating the flexibility F = 12/E · t3 the work equation
18.2 Approaches Related to the Deflection and the Stresses 319

yields
Z
δWσ = {− δmxx [ κxx + F · ( mxx − ν myy ) ]
A
− δmyy [ κyy + F · ( ν mxx − myy ) ]

− δmxy [ 2κxy + 2(1 + ν)F · mxy ] } dA

Z
+ {δqn · w + δmn · w,n + δmt · w,s } ds
ZE
− {δqn · w + δmn · w,n + δmt · w,s } ds = 0 .
I
Here as well, exactly fulfilling kinematics, integrating by parts the domain in-
tegrals, and implying continuity of the deflection w between element boundary
and interface
Z Z
δqn · w ds = δqn · w ds
E I
yields the work equation, that provides the basis of the discretization:
Z
δWσ = {+ δmxx ,x w,x − δmxx · F · (mxx − ν myy )
A
+ δmyy ,y w,y − δmyy · F · ( ν mxx − myy )

+ (δmxy ,y w,x + δmxy ,x w,y ) − δmxy · 2(1 + ν)F · mxy } dA

Z
− {δmn w,n + δmt w,s } ds = 0 . (18.2)
I
Comparing the work equation applying the PvF to the one applying the PvD,
the symmetry and the duality of both of the work principles become obvious.

18.2 Approaches Related to the Deflection and the Stresses

Cubic Approaches to Describe the Displacements
To describe the displacements complete cubic approaches employing 10 degrees
of freedom are chosen, which, by analogy with Section 17, are assigned to the
corner nodes as well as to the barycentric node. The scaling of the approach
related to the deflection w(λa , λb , λc )
u = [ w ] = ψu · a ,
ψ u = [ λ3a λ2a λb λ2a λc λ3b λ2b λc λ2b λa λ3c λ2c λa λ2c λb λa λb λc ] ,
a = [ a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 ]
320 18 Hybrid–Mixed Triangular Plate Elements

to the nodal displacement variables

vT = { [ w w,x w,y ]a [ w w,x w,y ]b [ w w,x w,y ]c [ w ]d }

is processed applying the scaling matrix G10 according to Equation (17.2).

Thus, matrix notation related to the deflection w follows to

u = [ w ] = ψ u · G10 · v = Ωv · v

and the same may be arranged with respect to the virtual deflection δw.

Quadratic Approaches Concerning the Moments

The complete cubic approach to describe the deflection w exhibits 10 nodal dis-
placement variables. Hereby three rigid body motions – one translation and two
rotations – may be described as well as seven displacement states, being availa-
ble to describe the curvatures [ κxx κyy κxy ]. Hence, an approach to describe the
moments and being balanced with respect to the displacement variables must
exhibit at least seven degrees of freedom related to the moments. However, due
to the derivatives with respect to x and y, at least linear approaches employing
nine degrees of freedom are required. Nevertheless, linear approaches result in
vanishing stiffness according to the barycentric deflection, so that the system
stiffness matrix becomes singular. Thus quadratic approaches to describe the
moments are chosen here. Introducing the following non–scaled approaches

mxx = b1 · λ2a + b2 · λ2b + b3 · λ2c + b4 · λa λb + b5 · λa λc + b6 · λb λc ,

myy = b7 · λ2a + b8 · λ2b + b9 · λ2c + b10 · λa λb + b11 · λa λc + b12 · λb λc ,
mxy = b13 · λ2a + b14 · λ2b + b15 · λ2c + b16 · λa λb + b17 · λa λc + b18 · λb λc

and the respective approaches to describe the virtual moments result in 18

available degrees of freedom bi , to fulfill weakly the conditions of deformation
in the domain. Applying matrix notation the approach is summarized with

σ = Ψσ · b ,
T
σ = [ mxx myy mxy ] ,
 2 2 2 
λa λb λc λa λb λa λc λb λc 0 0
λ2a λ2b λ2c λa λb λa λc λb λc
 
Ψσ =  0 0 ,
2 2 2
0 0 λa λb λc λa λb λa λc λb λc
bT = [ b1 b2 b3 b4 b5 b6 | b7 b8 b9 b10 b11 b12 | b13 b14 b15 b16 b17 b18 ] .
18.3 Element Stiffness Matrix and Load Vector 321

18.3 Element Stiffness Matrix and Load Vector

Regarding matrix notation the work equation applying the PvD according to
Equation (18.1) comprises the degrees of freedom related to the element boun-
daries v as well as the degrees of freedom b at element level:
Z Z
T

−δWd = δv (Ωv Dv ) (Dσ Ψσ ) dA · b − ΩTv p dA .
T T
(18.3)
A A
Applying the PvF according to Equation (18.2) it can also be stated
Z
−δWσ = δbT { (ΨTσ DTσ )(Dv Ωv ) · v − ΨTσ E−1 Ψσ · b } dA = 0 . (18.4)
A
Thus the degrees of freedom b may be described by the nodal displacements
v, to eliminate b the equation applying the PvD may be used. Matrix notation
of Equations (18.3) and (18.4) yields
Z " # " # Z " #
δv : 0 (ΩTv DTv ) (Dσ Ψσ ) v ΩTv p
dA − dA .
δb : A (ΨTσ DTσ )(Dv Ωv ) −ΨTσ E−1 Ψσ b A 0
Integration may be processed applying numerical integration according to Gauss–
Legendre, hence the element matrix and the load vector are numerically given.

Integration of Contour Integrals

The virtual work performed by the moments on the rotations of the interfaces
is given applying Equations (18.1) and (18.2) with
Z
−δWd |I = ( δw,n mn + δw,s mt ) ds ,
ZI
−δWσ |I = ( δmn w,n + δmt w,s ) ds .
I
The description of the slope w,n is discussed in Section 17.2.1 and Section 17.2.2
at fulfillment of the element continuity conditions. The derivatives w,n and w,s
of the deflection surface are transformed with the angle β of the respective edge
with respect to the x–coordinate, cf. Figure 18-3,
w,n = − w,x · sin β + w,y · cos β ,
w,s = + w,x · cos β + w,y · sin β .
Again, sin β and cos β are computed by means of the coordinate differences of
the corner nodes:
p
∆x = xi+1 − xi , ∆y = yi+1 − yi and ℓi+1
i = ∆x2 + ∆y 2 ,
∆y ∆x
sin β = and cos β = .
ℓ ℓ
322 18 Hybrid–Mixed Triangular Plate Elements

x
w,s
w, n
D
s w,y s
β
td td
y
nd w,x nd

Fig. 18-3 Transformation of slopes related to the element boundary

The transformation of the moments [ mxx myy mxy ] into the moments [ mn mt ]
with respect to the contour coordinate yields
mn = mxx · sin2 β + myy · cos2 β − 2mxy · cos β · sin β ,
mt = (−mxx + myy ) · sin β · cos β + mxy · (cos2 β − sin2 β) .
Applying the transformation matrix Tv concerning the derivatives of the de-
flection surface and the transformation matrix Tσ concerning the moments,
matrix notation yields
Z " T T
# " #
 T  0 (Ω v Tv ) (Tσ Ψσ ) v
−δW |I = δv δbT T T
ds .
G (Ψ T
σ σ )(T v Ω v ) 0 b

Elimination of Moment–DOF
Having integrated the work terms the degrees of freedom b may be numerically
eliminated. Introducing the abbreviations
Z Z
Hσv = (ΨTσ DTσ )(Dv Ωv ) dA + (ΨTσ TTσ )(Tv Ωv ) ds ,
ZA I
T −1
F= Ψσ E Ψσ dA ,
ZA
fp = ΩTv p dA ,
A
matrix notation gives
" # " # " #
δv : 0 HTσv v fp 6= 0 at element level
· −
δb : Hσv −F b 0 = 0 at element level
and the computation of b yields
b = F−1 Hσv · v . (18.5)
18.4 Fulfillment of the Continuity Conditions at the Interface 323

After incorporating b into the conditions of equilibrium the PvD follows to

−δWd = δvT { HTσv F−1 Hσv } · v − δvT fp . (18.6)

The expression inside the curly bracket of Equation (18.6) states the stiffness
matrix with respect to the element domain.

18.4 Fulfillment of the Continuity Conditions at the Interface

By analogy with Section 18.3 the element continuity conditions related to w,n
may be fulfilled employing additional degrees of freedom at the interfaces –
version A – or weakly applying the work equation – version B.

Version A – Element K–HMT–6–12

The fulfillment of the element continuity conditions related to w,n may happen
by introducing additional rotational degrees of freedom at the interface, when
describing the slopes w,n as well as δw,n by employing quadratic approaches
in the work equations (18.1) of the PvD and (18.2) of the PvF
Z
δWd = . . . − {mn δw,n + mt δw,s } ds
I

as well as
Z
δWσ = . . . − {δmn w,n + δmt w,s } ds .
I

By analogy with Section 17.2 the slopes w,n are described by a linear course
w,n |linear , given by the nodal displacement variables at the interface, as well as
by an additional quadratic part w,n |quadr. , which is assigned to the rotational
degree of freedom ϕn related to the midpoint of the interface. Thus the work
equations have to be transformed applying

w,n = w,n |linear + ϕn |quadr.

δw,n = δw,n |linear + δϕn |quadr. .

Hence, at derivation of the Equations (18.5) and (18.6), the vector of the no-
dal displacement variables comprises the 10 degrees of freedom of the cubic
approach and the three rotational degrees of freedom related to the interface.
324 18 Hybrid–Mixed Triangular Plate Elements

Version B – Element K–HMT–3–9

At deriving the work equations (18.1) related to the PvD and (18.2) related to
the PvF the terms of work
Z
δWd = . . . − {mn δw,n + mt δw,s } ds
I

as well as
Z
δWσ = . . . − {δmn w,n + δmt w,s } ds
I

are to be evaluated at the interface, whereat w,n again comprises linear and
quadratic parts. At evaluating the integrals related to the element domain and
to the interface the element continuity condition concerning w,n is fulfilled
weakly without employing further degrees of freedom. This effects the triangu-
lar element to comprise three nodal displacement variables [ w w,x w,y ] each
related to the corners of the element.

18.5 Subsequent Stress Analysis

As an alternative the matrix Hσv , that has been evaluated at computing the
stiffness matrix, may be stored element by element and may be introduced to
the subsequent stress analysis, cf. Section 14.7 – version B.

18.6 Test
To test the elements the square plate is investigated according to Section 6.2.9.
Taking into account the symmetry, a quarter of the plate is investigated ap-
plying different discretizations to illustrate the convergence of the results.
x clamped

E constantly distributed loading: p = 1 kN/m2

y
T E = 3.0 . 107 kN/m2
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 18-4 Square plate – system and loading

18.6 Test 325

Version A – Element K–HMT–6–12

The subsequent Tables 18.1 and 18.2 summarizes the behavior of convergence of
different variables of state. Regarding nodes, that belong to multiple elements,
the results are presented each with respect to the neighboring elements. The
comparison of the results applying the plate elements according to Section
11 and to Section 17 demonstrates the extraordinarily good convergence of
the hybrid–mixed element. The deflection as well as the moments comparably
converge quite fast against the reference solution.

Table 18.1 Square plate – results related to distributed action p = 1 kN/m2

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
unknowns 17 43 131 451 1667
wM [ cm ] 0. 5843 0. 7165 0. 7346 0. 7360 0. 7361
mxxM [ kN m/m ] 2. 600 2. 228 2. 268 2. 197 2. 165
1. 462 2. 311 2. 298 2. 208 2. 168
myyM [ kN m/m ] 4. 026 3. 556 3. 380 3. 236 3. 184
3. 013 3. 471 3. 337 3. 223 3. 181
myyE [ kN m/m ] −3. 652 −5. 505 −6. 575 −6. 883 −6. 959
mxyT [ kN m/m ] −2. 236 −1. 301 −1. 470 −1. 520 −1. 532
+0. 030 −1. 634 −1. 629 −1. 570 −1. 546
−1. 728 −1. 627 −1. 561 −1. 543

Table 18.2 Square plate – results related to concentrated action P = 1 kN in M

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
unknowns 17 43 131 451 1667
wM [ cm ] 0. 0621 0. 1002 0. 1066 0. 1078 0. 1080
mxxM [ kN m/m ] 0. 2455 0. 6203 0. 9297 1. 199 1. 464
0. 1956 0. 5931 0. 9190 1. 189 1. 454
myyM [ kN m/m ] 0. 4220 0. 7298 1. 016 1. 280 1. 544
0. 3009 0. 7294 1. 025 1. 290 1. 554
myyE [ kN m/m ] −0. 3884 −0. 5158 −0. 6133 −0. 6495 −0. 6602
mxyT [ kN m/m ] −0. 2000 −0. 1175 −0. 1416 −0. 1473 −0. 1486
0. 0101 −0. 2042 −0. 1667 −0. 1538 −0. 1503
−0. 1336 −0. 1480 −0. 1491 −0. 1491
326 18 Hybrid–Mixed Triangular Plate Elements

Version B – Element K–HMT–3–9

Table 18.3 and Table 18.4 summarize the behavior of convergence concerning
the hybrid–mixed plate element employing 9 degrees of freedom. Comparing the
results to those according to the plate element employing 12 degrees of freedom
– version A – demonstrates, that, due to the weak formulation of the continuity
condition concerning w,n , the convergence is somewhat inferior. Furthermore
it is worth to take notice of the fact that the element incorporates considerably
less degrees of freedom at applying the identical finite element mesh.

Table 18.3 Square plate – results related to distributed action p = 1 kN/m2

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
unknowns 12 27 75 243 867
wM [ cm ] 0. 4869 0. 6749 0. 7208 0. 7323 0. 7354
mxxM [ kN m/m ] 1. 270 2. 289 2. 139 2. 149 2. 152
3. 253 2. 433 2. 209 2. 165 2. 156
myyM [ kN m/m ] 1. 635 3. 142 3. 086 3. 136 3. 157
3. 572 3. 124 3. 125 3. 145 3. 159
myyE [ kN m/m ] −3. 043 −4. 481 −5. 549 −6. 229 −6. 599
mxyT [ kN m/m ] −1. 399 −1. 191 −1. 499 −1. 535 −1. 537
−0. 248 −1. 323 −1. 487 −1. 524 −1. 533
−1. 842 −1. 657 −1. 577 −1. 550

Table 18.4 Square plate – results related to concentrated action P = 1 kN in M

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
unknowns 12 27 75 243 867
wM [ cm ] 0. 0524 0. 0933 0. 1040 0. 1070 0. 1078
mxxM [ kN m/m ] 0. 1605 0. 5611 0. 8269 1. 097 1. 364
0. 3025 0. 5417 0. 8093 1. 076 1. 342
myyM [ kN m/m ] 0. 1517 0. 6022 0. 8873 1. 163 1. 431
0. 3750 0. 6274 0. 9145 1. 186 1. 453
myyE [ kN m/m ] −0. 3276 −0. 4937 −0. 5634 −0. 6102 −0. 6361
mxyT [ kN m/m ] −0. 1267 −0. 1042 −0. 1426 −0. 1490 −0. 1497
−0. 0270 −0. 1347 −0. 1408 −0. 1460 −0. 1478
−0. 1910 −0. 1612 −0. 1534 −0. 1506
19 Discrete Kirchhoff –Theory Element

The elements presented in the sections above and related to the Kirchhoff ’s
plate theory apply the PvD in the representation
Z
−δWd = B · {δw,xx (w,xx + νw,yy ) + δw,yy (w,yy + νw,xx )
A Z
+ 2 (1 − ν) δw,xy w,xy } dA − δw pz dA = 0 . (19.1)
A

Here the internal work is described introducing the curvatures of the deflection
surface.
Nevertheless, the natural formulation of the work due to bending employs the
gradients of the rotations, since the curvatures are defined herewith. Thus,
without consideration of the Kirchhoff–hypothesis, it can be stated
Z
−δWd = B · {δϕx ,x (ϕx ,x + νϕy ,y ) + δϕy ,y (ϕy ,y + νϕx ,x )
A Z
1
+ (1 − ν) (δϕx ,y + δϕy ,x ) (ϕx ,y + ϕy ,x )} dA − δw pz dA . (19.2)
2 A

Applying this formulation the virtual work is completely described. The Kirch-
hoff–hypothesis is missing, without taking it into account, the constraints of
the rotations ϕx , ϕy and the deflection w are not considered, and the boundary
condition related to w cannot be fulfilled. To take into account the Kirchhoff–
hypothesis different paths of solution may be followed. These are in essence:

1. Introducing the kinematic relations w,x − ϕx = 0 as well as w,y − ϕy = 0

into the work equation (19.2) yields Equation (19.1).
2. Extending the work equation (19.2) by means of the method of Lagrangi-
an multipliers and with respect to the virtual work due to the constraints
of the Kirchhoff–hypothesis yields
Z
−δWd = −δWϕx , ϕy + {δλx (w,x − ϕx ) + (δw,x − δϕx )λx dA
ZA
+ {δλy (w,y − ϕy ) + (δw,y − δϕy )λy dA
A

and thus it follows the weak formulation of the Kirchhoff–hypothesis. Here

the Lagrangian multipliers correspond to the shear forces qx , qy concer-
ning their physical meaning.
© The Author(s), under exclusive license to
Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_19
328 19 Discrete Kirchhoff –Theory Element

3. At discrete fulfillment of the Kirchhoff–hypothesis at special positions, the

domain integrals and thus corresponding elaborate approaches to describe
the real as well as the virtual deflections w, δw are avoided.

With the idea to discretely fulfill the Kirchhoff–hypothesis Stricklin, Haisler,

Tisdale and Gunderson [93] developed a very efficient triangular element, which
has been discussed later on by Batoz, Bathe and Ho [15] comparing it to other
triangular elements. Furthermore it will be denoted as DKT element, or DKT-
3-9 element respectively.
The idea aims to a triangular element employing three nodes inside the element,
choosing the nodal deflections w as well as the nodal rotations w,x and w,y to
be the degrees of freedom, see Figure 19-1.
w w,x w,y
A

C B
w w,x w,y w w,x w,y

Fig. 19-1 Nodal degrees of freedom related to the deflection and the rotations

19.1 The Discrete Kirchhoff Triangular Element

According to the formulation given with Equation (19.2) the PvD is taken
as a basis, being discretized with respect to the element domain employing
quadratic approaches to describe the rotations ϕx , ϕy . The rotations related to
the element nodes, cf. Figure 19-2, are chosen as degrees of freedom concerning
the internal work.
x

A ( λ a = 1, λ b = 0, λ c = 0 )
F ( 0.5, 0, 0.5 )

D ( 0.5, 0.5, 0 )
C ( 0, 0, 1)

y E ( 0, 0.5, 0.5 ) B ( 0, 1, 0 )

Fig. 19-2 Element nodes employing a quadratic approach of the rotations

19.1 The Discrete Kirchhoff Triangular Element 329

Employing a polynomial approach in a general notation with respect to area

coordinates yields
" #
ϕx
uϕ = = Ψϕ · aϕ ,
ϕy
" #
λ2a λ2b λ2c λa λb λa λc λb λc 0 0 0 0 0 0
Ψϕ = ,
0 0 0 0 0 0 λ2a λ2b λ2c λa λb λa λc λb λc

aϕ = [ a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 ]

and a respective approach to describe the virtual rotations. Hereby, the inte-
gration of the virtual work can be evaluated with respect to the coefficients
a. The scaling of the general polynomial to the physically meaningful nodal
rotations [ ϕx ϕy ]A, B, C takes place by evaluating the approach with respect
to the coordinates of the element nodes. In matrix notation follows

vϕ = Ψ̃ϕ · aϕ

and in detail
" #     
ϕ a
 x   1   1 
 ϕ     a 
 y   1   2 
" #A     
 ϕ     a3 
 x   1   
     
 ϕy   1   a4 
" # B     
     a 
 ϕx   1   5 
     
 ϕy     a6 
   1   
" #C  =  · .
 ϕx     a7 
   1/4 1/4 1/4   
     
 ϕy   1/4 1/4 1/4   a8 
" # D    
     a 
 ϕx   1/4 1/4 1/4   9 
     
1/4   a10
 ϕy     
" # E   1/4 1/4 
     
 ϕx     a11 
   1/4 1/4 1/4   
ϕy 1/4 1/4 1/4 a12
F

Inverse representation of the equation

−1
aϕ = Ψ̃ϕ · uϕ
330 19 Discrete Kirchhoff –Theory Element

yields the scaling matrix

 
 +1 
 
 +1 
 
 
 +1 
 
 −1 −1 +4 
 
 
 −1 −1 +4 
 
 
−1  −1 −1 +4 
Gϕ = Ψ̃ϕ =  
 
 +1 
 

 +1 

 

 +1 

 
 −1 −1 +4 
 
 
 −1 −1 +4 
−1 −1 +4

and the physically meaningful shape functions related to the rotations

" #
ϕx
uϕ = = Ψϕ · Gϕ · vϕ = Ωϕ · vϕ .
ϕy

Since, applying the PvD and the approaches to describe the rotations, the
Kirchhoff–hypothesis is not taken into account yet, it needs a second step to
fulfill the Kirchhoff–hypothesis. Therefor a cubic Hermite–polynomial is chosen
to describe w at the element boundaries, which is assigned to the deflections w
and to the rotations w,s according to the element nodes of the respective boun-
dary, see Figure 19-3. Thus it is feasible to describe the Kirchhoff–hypothesis
along the boundary employing the displacements w and the rotations ϕx , ϕy
and to fulfill the element continuity conditions related to w.
A
s
[ w w,s ] AB
[ w w,s ] CA

s
B
C [ w w,s ] BC s

Fig. 19-3 Element edges with deflections and rotations

19.1 The Discrete Kirchhoff Triangular Element 331

Concerning the boundary A − B described with the coordinate s the course of

the deflection w(s) is chosen to

w(s) = φ1 · w|A + φ2 · w,s |A + φ3 · w|B + φ4 · w,s |B , (19.3)

whereat the shape functions φi are given in Figure 6-2. Employing this approach
the discrete fulfillment of the Kirchhoff–hypothesis takes place at the corners of
the element as well as at the midpoints of the edges, if it is possible to describe
the previous degrees of freedom related to the element rotations [ ϕxi ϕyi ] with
the new degrees of freedom [ w w,x w,y ] related to the displacement variables.
The evaluation of the Kirchhoff–hypothesis concerning the corner nodes yields

ϕx |i = w,x |i and ϕy |i = w,y |i .

The evaluation of the Kirchhoff–hypothesis at the midpoints of the edges gives

ϕt |0.5ℓ = − 1.5
ℓ
· w|i − 0. 25 · w,s |i + 1.5
ℓ
· w|i+1 − 0. 25 · w,s |i+1 (19.4)

concerning the tangential rotation ϕt with respect to the contour coordinate

s, when assigning the corner nodes of the respective boundaries by i, i + 1.
Simplifying, the rotation ϕn normal to the boundary is described by a linear
course of the slope w,n along the respective edge with
ϕn |0.5ℓ = 0. 5 · w,n |i + 0. 5 · w,n |i+1 , (19.5)

hence w,n is non–conform. Applying the transformation

" # " # " #
w,s |i cos β sin β w,x |i
= · , (19.6)
w,n |i − sin β cos β w,y |i

the slopes w,n and w,s are connected to the slopes w,x and w,y of the nodal
degrees of freedom, see Figure 19-4.

x
w,s
w, n
D
s w,y s
β
td td
y
nd w,x nd

Fig. 19-4 The transformation of w,s and w,n with respect to w,x and w,y
332 19 Discrete Kirchhoff –Theory Element

The trigonometric functions may be computed by means of the differences of

coordinates related to the corner nodes as given in Section 17.2.1:
p
∆x = xi+1 − xi , ∆y = yi+1 − yi and ℓi+1
i = ∆x2 + ∆y 2 ,
∆y ∆x
sin β = and cos β = .
ℓ ℓ
Applying the Eqns. (19.4), (19.5) and (19.6) yields
 
" # " # w
ϕt − 1.5
ℓ
−0. 25 cos β −0. 25 sin β  
= ·  w,x 
ϕn 0 −0. 5 sin β 0. 5 cos β
0.5ℓ w,y i
 
" # w
+ 1.5
ℓ
−0. 25 cos β −0. 25 sin β  
+ ·  w,x  .
0 −0. 5 sin β 0. 5 cos β
w,y i+1

The transformation of the rotations ϕt , ϕn into the rotations ϕx , ϕy related to

the midpoints of the edges takes place according to Figure 19-4 with
" # " # " #
ϕx cos β − sin β ϕt
= · .
ϕy sin β cos β ϕn

Thereby the fulfillment of the Kirchhoff–hypothesis is possible at the midpoints

of the edges employing the degrees of freedom related to the corner nodes:
 
" # " 1.5
2 2
# w
ϕx − ℓ cos −0. 25 cos +0. 5 sin −0. 75 cos sin  
= 1.5 2 2
·  w,x 
ϕy − sin −0. 75 cos sin −0. 25 sin +0. 5 cos
0.5ℓ ℓ w,y i
 
" 1.5
2 2
# w
+ ℓ cos −0. 25 cos +0. 5 sin −0. 75 cos sin  
+ ·  w,x  .
1.5
+ ℓ sin −0. 75 cos sin −0. 25 sin2 +0. 5 cos2
w,y i+1

The scaling of the nodal rotations with respect to the nodal degrees of freedom
of the deflection surface w is summarized in the scaling matrix Gw with

vϕ = Gw · vw .

The scaling employs the twelve nodal rotations vϕ and the nine degrees of
freedom vw at the three corner nodes A, B, C, which describe the deflection at
the element boundaries.
19.1 The Discrete Kirchhoff Triangular Element 333

The scaling matrix may be divided into submatrices with respect to the degrees
of freedom at the nodes A, B, C
" # 
 ϕx 

 ϕy 

" #A 

 ϕx 

 
 ϕy 
"
 #B 

 ϕx   
  vwA

 ϕy  
   
" C
#  = GwA GwB GwC ·  vwB  .
 ϕx 



 vwC
 ϕy 
"
 #D 

 ϕx 
 
 ϕy 
"
 #E 


 ϕx 

ϕy
F

The submatrices are given in detail with the abbreviations cIJ = cos β, and
sIJ = sin β, and the length ℓ of the edge IJ
 
 0 1 0 
 
 0 0 1 
 

 0 0 0 

 
 0 0 0 
 

 0 0 0  
 
  w
 0 0 0  
GwA · vwA =  ·  w,x 
 ,
 1.5 
−
 ℓ
cAB −0. 25 c2AB + 0. 5 s2AB −0. 75 cAB sAB 
 w,y A
 1.5
−0. 25 s2AB + 0. 5 c2AB

 − ℓ sAB −0. 75 cAB sAB 
 

 0 0 0 

 
 0 0 0 
 
 + 1.5 c 2 2
AC −0. 25 cAC + 0. 5 sAC −0. 75 cAC sAC 
 ℓ 
1.5
+ ℓ sAC −0. 75 cAC sAC −0. 25 s2AC + 0. 5 c2AC
334 19 Discrete Kirchhoff –Theory Element
 
 0 0 0 
 0 0 0 
 
 

 0 1 0 

 0 0 1 
 
 
 0 0 0   



 w
 0 0 0   
GwB · vwB =  ·  w,x  ,
 + 1.5 cAB −0. 25 c2AB + 0. 5 s2AB −0. 75 cAB sAB 
 ℓ  w,y B
+ 1.5 −0. 25 s2AB + 0. 5 c2AB
 
 ℓ
sAB −0. 75 cAB sAB 
 
− 1.5 cBC −0. 25 c2BC + 0. 5 s2BC
 
 ℓ
−0. 75 cBC sBC 
 

 − 1.5
ℓ
sBC −0. 75 cBC sBC −0. 25 s2BC + 0. 5 c2BC 

 
 0 0 0 
0 0 0

 
 0 0 0 
 0 0 0 
 
 

 0 0 0 

 0 0 0 
 
 
 0 1 0   
  w

 0 0 1 
  
GwC · vwC =  ·  w,x  .
 0 0 0 
  w,y C

 0 0 0 

 1.5 
+
 ℓ
cBC −0. 25 c2BC + 0. 5 s2BC −0. 75 cBC sBC 

 1.5
−0. 25 s2BC + 0. 5 c2BC

+ ℓ sBC −0. 75 cBC sBC 
 
 − 1.5 cAC −0. 25 c2AC + 0. 5 s2AC −0. 75 cAC sAC 
 ℓ 
− 1.5
ℓ
sAC −0. 75 cAC sAC −0. 25 s2AC + 0. 5 c2AC

Simplifying, the scaling matrices Gϕ and Gw may be multiplicated by each

other, so that the coefficients ai , assigned to the approach to describe ϕ, are
directly scaled to the nodal displacement variables [ w w,x w,y ] by

a = Gϕ · Gw · vw = Gϕw · vw .

As in the case of Gw the scaling matrix Gϕw is divided with respect to the
19.1 The Discrete Kirchhoff Triangular Element 335

degrees of freedom of the corner nodes

 
vwA
   
a= GϕwA GϕwB GϕwC ·  vwB  ,
vwC

 
 0 1 0 
 0 0 0 
 
 

 0 0 0 

− 6ℓ cAB −1 − c2AB + 2 s2AB
 
 −3 cAB sAB 
 

 + 6ℓ cAC −1 − c2AC + 2 s2AC −3 cAC sAC  
 w

 
GϕwA · vwA

= 0 0 0   
 ·  w,x  ,

 0 0 1 
 w,y A
 
 0 0 0 
 
− 6ℓ sAB −1 − s2AB + 2 c2AB
 
 −3 cAB sAB 
 

 0 0 0 

+ 6ℓ sAC −1 − s2AC + 2 c2AC
 
 −3 cAC sAC 
0 0 0

 
 0 0 0 
 0 1 0 
 
 

 0 0 0 

 6
 + ℓ cAB −1 − c2AB + 2 s2AB

−3 cAB sAB 
 
 0 0 0   

 6
 w
2 2 
=  − ℓ cBC −1 − cBC + 2 sBC −3 cBC sBC
   
GϕwB · vwB  ·  w,x  ,

 0 0 0 
 w,y B
 
 0 0 1 
 

 0 0 0 

 6 
 + sAB
 ℓ −3 cAB sAB −1 − sAB + 2 c2AB
2 

 0 0 0 
 
6
− ℓ sBC −3 cBC sBC −1 − sBC + 2 c2BC
2
336 19 Discrete Kirchhoff –Theory Element

 
 0 0 0 
 0 0 0 
 
 

 0 1 0 

 0 0 0 
 
 6 
 − cAC −1 − c2 + 2 s2 −3 cAC sAC   
 ℓ AC AC  w
 6
 + cBC −1 − c2BC + 2 s2BC

−3 cBC sBC   
GϕwC · vwC = ℓ  ·  w,x  .

 0 0 0 
 w,y C
 
 0 0 0 
 

 0 0 1 

 
 0 0 0 
 
 −6 s −3 c s −1 − s 2 2 
 ℓ AC AC AC AC + 2 cAC 
6
+ ℓ sBC −3 cBC sBC −1 − sBC + 2 c2BC
2

19.2 Stiffness Matrix, Load Vector and Stress Analysis

Applying the scaling matrix Gϕw it is possible to arrange the approaches being
scaled to the physically meaningful degrees of freedom and to perform the
integration herewith:
uϕ = Ψϕ · Gϕw · vw = Ωw · vw .
Thus the computation of the stiffness matrix becomes very efficient, since the
integration with respect to Ψϕ can be conducted analytically with less effort
Z
K = Gϕw · ΨTϕ · DT · E · D · Ψϕ dA · Gϕw .
T

The computation of the load vector with respect to distributed loadings p(x, y)
has to be performed by direct integration – without employing the shape func-
tions – and it has to be allocated to the deflections of the corner nodes. This is
necessary, because the deflection is not approached inside the element domain.
Thus the element load vector is given with
A
fT = [ 1 0 0 1 0 0 1 0 0 ] · ·p .
3
The subsequent analysis concerning the moments σ T = [ mxx myy mxy ]A, B, C
also takes place applying the scaling matrix Gϕw , since the stress matrix is
formulated employing the general approaches to describe the rotations
σ = E · D · Ψϕ · Gϕw · vw = Sϕ · Gϕw · vw .
19.3 Test 337

Evaluating the stress matrix at the element nodes yields

B
S̃ϕ = ·
2A∆

2ba 0 0 bb bc 0 ν · 2aa 0 0 ν · ab ν · ac 0
 
 ν · 2ba 0 0 ν · bb ν · bc 0 2aa 0 0 ab ac 0 
 
 e · 2aa 0 0 e · ab e · ac 0 e · 2ba 0 0 e · bb e · bc 0 
 
 
0 2bb 0 ba 0 bc 0 ν · 2ab 0 ν · aa 0 ν · ac
 
 
0 ν · 2bb 0 ν · ba 0 ν · bc 0 2ab 0 aa 0 ac ,
 

0 e · 2ab 0 e · aa 0 e · ac 0 e · 2bb 0 e · ba 0 e · bc
 
 
 
 

 0 0 2bc 0 ba bb 0 0 ν · 2ac 0 ν · aa ν · ab 

 0 0 ν · 2bc 0 ν · ba ν · bb 0 0 2ac 0 aa ab 
0 0 e · 2ac 0 e · aa e · ab 0 0 e · 2bc 0 e · ba e · bb

e = (1 − ν)/2 , B = E · t3 /12 (1 − ν 2 ) .

19.3 Test
As an example to test the formulation the square plate is taken according
to Section 6.2.9 and Figure 19-5. Exploiting the symmetry, only a quarter
of the plate is investigated applying different discretizations to illustrate the
convergence of the results.
x clamped

E constantly distributed loading: p = 1 kN/m2

y
T E = 3.0 . 107 kN/m2
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 19-5 Square plate - system and loading

The subsequent Tables 19.1 and 19.2 demonstrate the behavior of convergence
of different variables of state concerning the DKT element employing 9 degrees
of freedom related to the element. Regarding nodes, that belong to multiple
elements, the results are presented each with respect to the neighboring ele-
ments.
338 19 Discrete Kirchhoff –Theory Element

Table 19.1 Square plate – results related to distributed action p = 1 kN/m2

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 9922 0. 8300 0. 7629 0. 7432 0. 7379
mxxM [ kN m/m ] 3. 720 2. 780 2. 320 2. 197 2. 165
4. 341 3. 244 2. 458 2. 234 2. 174
myyM [ kN m/m ] 3. 720 3. 447 3. 273 3. 195 3. 174
6. 821 3. 862 3. 366 3. 220 3. 180
myyE [ kN m/m ] −6. 201 −7. 265 −7. 195 −7. 086 −7. 032
mxyT [ kN m/m ] −1. 329 −1. 549 −1. 572 −1. 561
−1. 240 −1. 347 −1. 470 −1. 523 −1. 534
−1. 587 −1. 518 −1. 506 −1. 513

Table 19.2 Square plate – results related to concentrated action P = 1 kN in M

mesh 1×1 2×2 4×4 8×8 16 × 16

elements 2 8 32 128 512
wM [ cm ] 0. 1190 0. 1122 0. 1095 0. 1085 0. 1082
mxxM [ kN m/m ] 0. 4465 0. 7991 1. 063 1. 328 1. 593
0. 5209 0. 7411 0. 9716 1. 228 1. 491
myyM [ kN m/m ] 0. 4465 0. 7579 1. 044 1. 314 1. 580
0. 8186 0. 9094 1. 157 1. 419 1. 683
myyE [ kN m/m ] −0. 7442 −0. 7130 −0. 6815 −0. 6723 −0. 6679
mxyT [ kN m/m ] −0. 2233 −0. 1322 −0. 1466 −0. 1495 −0. 1496
−0. 0744 −0. 1412 −0. 1395 −0. 1454 −0. 1475
−0. 1911 −0. 1612 −0. 1527 −0. 1502

Comparing the results applying the plate elements according to Sections 17.2
and 18 illustrates, that the results obtained by the DKT element converge
somewhat inferior against the reference solution. In particular, to obtain suf-
ficiently accurate stress resultants, a more refined discretization is required.
Moreover the displacements converge against the reference values approaching
from above. The numerical efficiency can be seen as an advantage, since the
computation of the stiffness matrix may follow a 3–point–integration and an
analytical integration respectively.
20 Benchmark Concerning Triangular Plate Elements

Although all elements, that are described in Sections 17, 18, and 19, converge
against the reference solution, it is useful to compare the convergence behavior
and to discuss advantages and disadvantages. Therefor two benchmarks are
defined, which are applied throughout the textbook.

20.1 Square Plate Subjected to Distributed Loading

The first benchmark deals with the square plate, which is investigated as a test
example, cf. Figure 20-1. Instead of analysing a simply supported plate or an
all over clamped plate, the square plate combines both boundary conditions,
what effects a more complex twisting field within the plate. The material data
are chosen related to concrete and the bending stiffness is computed within the
element subroutines numerically with B = E · t3 /12/(1 − ν 2 ).
x clamped

E constantly distributed loading: p = 1 kN/m2

y
7 2
T E = 3.0 . 10 kN/m
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 20-1 Square plate – system and loading

The comparison of the numerical effort to get the results is partly possible by
taking into account the number of the degrees of freedom and the bandwith.
The effort to compute the element matrices is a second indicator, but it de-
pends strongly on kind as well as preparation of programming, if analytical
integration or numerical integration is applied. Thus, Table 20.1 compares the
element formulations with 9 and 9+3 degrees of freedom to the element for-
mulation given in Section 11 employing 18 degrees of freedom. The number of
elements, the number of unknowns, and the bandwidth are chosen for different
discretizations such, that the mesh lines are kept constant.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_20
340 20 Benchmark Concerning Triangular Plate Elements

Table 20.1 Mesh size – unknowns and bandwidth

△9 △ 9 (+3) △ 18
mesh lines 2 3 5 9 2 3 5 9 2 3 5 9
elements 2 8 32 128 2 8 32 128 2 8 32 128
unknowns 12 27 75 243 17 43 131 451 24 54 150 486
bandwidth 12 15 21 33 17 23 35 59 24 30 42 66

The convergence behavior is represented by means of the Figures 20-2, 20-3,

and 20-4 instead of numbers, which are tabled within the respective sections.
Figure 20-2 illustrates the convergence of the midpoint deflection wM against
the number of unknowns, Figure 20-3 shows the midpoint bending moment
mxxM and Figure 20-4 the clamping moment myyE .

% error of wM
K-DT-3-18
DKT-3-9
10
K-HMT-3-9
K-HMT-6-12
K-HDT-3-9
K-HDT-6-12
5 K-MT-6-12

0
20 100 500 dof

-5

-10

Fig. 20-2 Square plate – convergence of midpoint deflection

20.1 Square Plate Subjected to Distributed Loading 341

The 18 DOF element shows superior convergence with respect to all state
variables due to the employed shape functions of 5th order. Nonetheless, due to
the kinematic variables of second order, it is limited to benchmark applications.
The elements K-HDT-3-9 and K-HDT-6-12 converge very good as well, but the
last one employs 12 DOF and thus needs more effort to compute the stiffness
matrix and to solve the system of equations.
The convergence of the hybrid–mixed elements K-HMT-3-9 and K-HMT-6-12
is not as good with respect to the deflection, but convergence is better with
respect to the moments, since the formulations employ quadratic polynomials
to desrcibe the moments inside the element before the elimination process takes
place.
The DKT element and the K-MT-6-12 element converge a little inferior, but in
comparison to the other elements they need less effort to compute the element
matrix.
% error of mxxM

70

60 K-DT-3-18
DKT-3-9
50 K-HMT-3-9
K-HMT-6-12
40 K-HDT-3-9
K-HDT-6-12
K-MT-6-12
30

20

10

0
20 100 500 dof

-10

-20

Fig. 20-3 Square plate – convergence of moment mxxM at the midpoint

342 20 Benchmark Concerning Triangular Plate Elements

% error in m yyE

20 100 500 dof

0

-5

-10

-15

-20

-25
K-DT-3-18
DKT-3-9
-30 K-HMT-3-9
K-HMT-6-12
-35 K-HDT-3-9
K-HDT-6-12
-40 K-MT-6-12

Fig. 20-4 Square plate – convergence of clamping moment myyE

Remarks
• All elements converge against the same solution.
• The element employing 18 DOF is not applicable to general problems.
• The convergence of the K-MT-6-12 published by Harvey–Kelsey [42] is
somewhat inferior in comparison to other elements.
• All elements with 9+3 DOF do not converge much better compared to
the correlated elements with 9 DOF and a weak formulation of the ele-
ment continuity conditions.
• Although the convergence is not equal with respect to all state variables,
the DKT element and the hybrid–mixed element K-HMT-3-9 show a good
overall performance with respect to convergence and effort.
20.2 Trapezoidal Plate Subjected to Distributed Loading 343

20.2 Trapezoidal Plate Subjected to Distributed Loading

As a second example to test different element formulations the rhombus plate
is investigated, clamped at one side, according to [27] and Figure 20-5. The
units inch, psi, that have been addressed in the original publication, are taken
over without converting them into SI-units. The plate exhibits a singularity at
the nook S concerning the moments, that strongly influences the results and,
at mesh refinement, leads to a heavy increase of the moments.

constantly distributed loading: p = 0.26066 psi free M

E = 1.05 .107 psi
t = 0.125 in

e
fre

ly
ν = 0.3

e
fre
lx = 12.0 in
ly = 8.48 in y S
x E clamped
lx

Fig. 20-5 Cantilever plate as 45o –rhomboid – system and loading

The plate is analyzed employing different triangular elements as well as different

mesh sizes. The elements are arranged such that their longest edge is pointed
at the direction of the longer diagonal of the rhombus. This arrangement is
extremely inappropriate, nevertheless it demonstrates the differences in the
convergence behavior of the elements.
The element according to Section 17.2.2 – version B is not presented here, since
it converges only slowly concerning this example. The results are represented
according to the following elements:
• K-DT-6-21 — Element employing displacements as primary variables
with 6 nodes, 21 nodal displacement variables – Section 10.2. Due to
the nodal degrees of freedom including curvatures at the corners of the
plate, the moments are evaluated directly employing the nodal displace-
ment variables.
• K-DT-3-18 — Element employing displacements as primary variables
with 3 nodes, 18 nodal displacement variables – Section 10.3. Due to
the nodal degrees of freedom including curvatures at the corners of the
plate, the moments are evaluated directly.
• K-MT-6-12 — Hybrid element employing displacements as primary va-
riables with 6 nodes, 9+1 nodal displacement variables related to the
344 20 Benchmark Concerning Triangular Plate Elements

element domain and 3 nodal displacement variables related to the inter-

faces – Section 17.2.1 – version A.
• DKT-3-9 — Hybrid element employing displacements as primary varia-
bles with 6 internal nodes, 12 nodal rotations related to the element
domain and 3 nodes with 3 × 3 nodal displacement variables related to
the interfaces – Section 19.
• K-HMT-6-12 — Hybrid–mixed element with 6 nodes, 9+1 nodal displa-
cement variables related to the element domain and 3 nodal displacement
variables related to the interfaces as well as 3 × 6 degrees of freedom as-
signed to the stresses related to the element domain – Section 18.4 –
version A.
• K-HMT-3-9 — Hybrid–mixed element with 3 nodes, 9+1 nodal displace-
ment variables as well as 3 × 6 degrees of freedom assigned to the stresses
related to the element domain – Section 18.4 – version B.
The subsequent Table 20.2 summarizes the convergence behavior of the deflec-
tion wM and the clamping moment myyE concerning triangular elements with
different approaches and degrees of freedom related to the element, whereat
Table 20.3 gives a hint to the numerical effort. Regarding nodes, that belong to
multiple elements, the clamping moments myyE are presented each with respect
to the neighboring elements. The bending moments located in the both first
lines are evaluated each from the elements lying at the boundary, the bending
moment related to the third line is evaluated from the element lying in be-
tween, whereat one corner is located at the boundary. The clamping moments
of the elements K-DT-6-21 and K-DT-3-18 are directly computed applying the
curvatures of the corner nodes, so that they exhibit the same value referring to
neighboring elements. Thus they are only given once.

Remarks
• The elements incorporating midpoints at the edges significantly better
converge than the respective elements without midpoints at the edges.
• The deflection wM converges to values 0. 28 in ≤ wM ≤ 0. 29 in, never-
theless which element formulation is addressed. The deflection measured
from the experiment wM ≈ 0. 297 in, according to [27], is slightly larger
than the computed values, which may be ascribed to modeling inaccuracy.
• The clamping moments related to the elements K-DT-6-21, K-DT-3-18,
K-HMT-6-12, K-HMT-3-9 converge against values −7. 10 lbs. in/in ≤
myE ≤ −6. 90 lbs. in/in. The clamping moments related to the element
K-MT-6-12 slower converge, which also is regarded to some extent at the
tests represented in Section 17.
20.2 Trapezoidal Plate Subjected to Distributed Loading 345

Table 20.2 Rhombus plate – results for distributed loading p = 0. 26066 psi
units: w [ in ], m [ lbs.in/in ]

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 2 8 32 128 512 2048
K-DT-6-21 wM 0. 2685 0. 2841 0. 2895 0. 2894 0. 2886 0. 2880
myyE −7. 793 −6. 570 −7. 154 −7. 105 −7. 063
K-DT-3-18 wM 0. 2923 0. 2993 0. 3022 0. 2975 0. 2929 0. 2901
myyE −9. 642 −6. 764 −6. 508 −6. 688 −6. 847
K-MT-6-12 wM 0. 1886 0. 2400 0. 2652 0. 2768 0. 2830 0. 2856
myyE −9. 384 −7. 938 −6. 381 −6. 278 −6. 611
−9. 384 −7. 938 −6. 381 −6. 278 −6. 611
−9. 384 −7. 938 −6. 381 −6. 278 −6. 611
DKT-3-9 wM 0. 6856 0. 3892 0. 3037 0. 2859 0. 2847 0. 2861
myyE −12. 44 −10. 48 −9. 062 −8. 050 −7. 532
−12. 25 −9. 770 −8. 213 −7. 615 −7. 314
−11. 86 −10. 44 −9. 649 −8. 356 −7. 690
K-HMT-6-12 wM 0. 2125 0. 2568 0. 2762 0. 2841 0. 2866 0. 2872
myyE −8. 781 −7. 625 −6. 838 −6. 904 −6. 995
−6. 228 −8. 209 −7. 506 −7. 034 −7. 018
−13. 02 −8. 781 −6. 859 −6. 992 −7. 020
K-HMT-3-9 wM 0. 1190 0. 1803 0. 2275 0. 2605 0. 2772 0. 2840
myyE −6. 443 −7. 766 −7. 564 −7. 266 −7. 155
−6. 185 −7. 510 −7. 277 −7. 074 −7. 038
−9. 334 −12. 13 −10. 67 −8. 933 −7. 956

Table 20.3 Number of nodal displacement variables for the chosen discretizations

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 2 8 32 128 512 2048
9 degrees of freedom 12 27 75 243 867 3267
12 degrees of freedom 17 43 131 451 1667 6403
18 degrees of freedom 24 54 150 486 1734 6534
21 degrees of freedom 29 70 206 694 2534 9670
346 20 Benchmark Concerning Triangular Plate Elements

Figures 20-6 and 20-7 represent the convergence behavior by charts. The mo-
notonic convergence of the deflections is obvious, they tend to the same final
value. Due to the approach of fifth order the element K-DT-6-21 is the best,
nevertheless it employs kinematic degrees of freedom relating to second order
derivatives. The hybrid–mixed elements exhibit a very good convergence be-
havior that corresponds to the respective approach, in particular at regarding
the convergence of the clamping moments.

% error in w

20 100 500 1000 2000 dof

0.0

-10.0

-20.0 K-DT-6-21
K-MT-6-12
-30.0
DKT-3-9
-40.0 K-HMT-6-12
K-HMT-3-9
-50.0

Fig. 20-6 Rhombus plate – convergence of the deflection wM

reference wM ≈ 0. 288 inch

% error in m yyE

+10.0
20 100 500 1000 2000 dof
0.0

-10.0

-20.0 K-DT-6-21
K-MT-6-12
-30.0 DKT-3-9
K-HMT-6-12
K-HMT-3-9

Fig. 20-7 Rhombus plate – convergence of the averaged values of myyE

reference myyE ≈ 7. 01 lbs. in/in

In contrast, the clamping moments do not uniformly converge, although the

averaged values are represented related to the nodes. One reason is the fact, that
the singularity in S particularly strongly influences the clamping moment myyE
at the center of the clamping when employing few degrees of freedom, since the
peak of the moment is decreased via few elements. The poor convergence of
the K-MT-6-12 element as well as the DKT-3-9 element is clearly visible, too.
SHEAR–DEFORMATION BEAM AND PLATE ELEMENTS
21 Timoshenko Beam Elements

The influence of shear deformations on the deformation states of beams and

plates has been neglected so far, since, concerning slender devices, it is of in-
ferior significance. The fundamental definitions and the governing equations
have been originally published by Timoshenko [97] concerning beam structures
with isotropic material behavior. Later on, the model has been extended to
the description of composite materials in different publications, higher order
descriptions of cross–sectional deformations have been added to some extent.

In the following section the fundamental changes concerning the Euler–Bernoulli

beam theory are described taking into account the shear deformations. The es-
sential difference becomes visible regarding kinematics as well as the material
equation related to the shear force Q, since the cross section is not enforced
any longer to be perpendicular to the beam axis, see Figure 21-1. Thereby the
cross–sectional rotation ϕ and the deflection w become completely independent
variables. Thus, concerning elements employing displacements as primary va-
riables, at least linear approaches are to be chosen to describe w as well as to
describe ϕ, that exhibit C0 –conformity. Hence, due to the admissibility of lower
order approaches to describe the deflection w and the more simple fulfillment
of the condition of conformity, this more accurate theory is chosen even if the
influence of shear is minor. Due to the rotation ϕ of the cross section also mo-
j dj

x
w,x x
me
g

Fig. 21-1 Differential beam element including shear deformations

ments me [N m/m] may act along the beam axis, which can be interpreted as
couple of forces, see Figure 21-1–right hand side. Thus the governing equations
concerning beams including shear deformations γ are summarized as follows.

a) Kinematics

w,x − ϕ − γ = 0 ,
ϕ,x − κ = 0 .

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_21
350 21 Timoshenko Beam Elements

b) Equilibrium
Q,x + p = 0 ,
M,x − Q − me = 0 .
c) Material Equations
GAS · γ − Q = 0 ,
EI · κ + M = 0 .
Shear deformations cause fundamental changes at developing a finite element
formulation, since the work equation applying the PvD is to be extended by
additional terms and the discretization of the work equation may effect non–
physical results to some extent, that are to be eliminated applying special
arrangements.

21.1 Elements Employing Displacements as Primary Variables

Detailed derivation of element matrices is disclaimed, since it is comprehensi-
ble in detail by means of the preceding sections. At formulation of the work
equation attention should be paid to the fact, that, because of the shear defor-
mations, also the shear forces perform work and are to be taken into account
applying the PvD. Hence, the work conjugated state variables are the bending
moment M and the curvature κ as well as the shear force Q and the shear
strain γ. Thus the PvD follows
Z Z
−δWd = {−δκ M + δγ Q} dx − δw · p dx = 0 .

Introducing kinematics the primary variables concerning the PvD are determi-
ned with w and ϕ, which are independent from each other now and therefore
are to be described accordingly when choosing the shape functions.
Z
−δWd = {δϕ,x EI ϕ,x + (δw,x − δϕ) GAS (w,x − ϕ)} dx
Z
− δw · p dx = 0 . (21.1)

Due to the first derivatives of the variables the criteria of convergence offer
linear shape functions for both the deflection and the rotation employing the
nodal degrees of freedom [ wA ϕA wB ϕB ] inside the element, cf. Section 2.4.2
and Equation 21.2.
x x
w(x) = ( 1 − ) wA + wB ,
ℓ ℓ (21.2)
x x
ϕ(x) = ( 1 − ) ϕA + ϕB .
ℓ ℓ
21.1 Elements Employing Displacements as Primary Variables 351

The displacement–based method yields, employing linear approaches to descri-

be w and ϕ, the following element stiffness matrix:
 
δwA : GAS GAS
−
GAS GAS
 ℓ 2 ℓ 2 
 
δϕA :  GA S EI GA S ℓ GA S EI GAS ℓ 
2 ℓ
+ 3
− 2
− ℓ
+ 6 

 GA

 (21.3)
δwB :  − S GA S GA S GA S

ℓ
− 2 ℓ
− 2


 
GAS EI GAS ℓ GAS EI GAS ℓ
δϕB : 2
− ℓ
+ 6
− 2 ℓ
+ 3

Therein the shear stiffness is defined to GAS = κQ ·GA , whereby κQ represents

a correction factor concerning the effective cross–sectional area.

The cantilever according to Figure 21-2 is investigated, applying a discretiza-

tion with one single element here.

x EI, GAS , l x
j
z,w

Fig. 21-2 Cantilever including shear deformations – system and loading

Due to the fact, that the rotation ϕ describes the rotation of the cross section
and not the inclination of the deflection line here, the realization of a clamping
is only possible with ϕ = 0, but is normally linked to a kink out of shear strain
because of w,x = γ + ϕ. Considering the boundary conditions wA = 0, δwA = 0
as well as ϕA = 0, δϕA = 0 yields the both equations
GAS GAS
δwB : + · wB − · ϕB =P, (21.4)
ℓ 2 
GAS EI GAS · ℓ
δϕB : − · wB + + · ϕB = 0 , (21.5)
2 ℓ 3
with the solutions
!
Pℓ 1
ϕB = EI GAS ·ℓ
, (21.6)
2 ℓ + 12
!
1 ℓ2
wB = P ℓ + 2 . (21.7)
GAS 4EI + GA3S ·ℓ
352 21 Timoshenko Beam Elements

The Euler–Bernoulli beam theory describing the shear–rigid beam should be

incorporated by GAS → ∞. Hence, applying Equation (21.7), this yields a
disappearing deflection wB at the end of the cantilever, since the shear stiffness
in the denominator covers the influence of the bending stiffness. The physically
mindless result achieved here is numerically originated and also exists at a
higher number of elements. This numerically caused phenomenon is referred
to as locking. The phenomenon arises, when employing approaches of identical
order to describe w and ϕ. Because of the kinematics

γ = w,x − ϕ (21.8)

this may lead to an unbalanced approach of the shear strains and the respective
virtual work. At choosing a linear approach to describe w and ϕ and due to the
shear strains being described employing a constant term w,x and a linear term
ϕ, this results in a serrated approximation of γ and thus of the shear forces.
One possibility to remedy this, is to choose an approach of one power more
to describe w than that to describe ϕ, whereby both parts of Equation (21.8)
comprise the same polynomial order.
Another possibility still is the choice of linear approaches to describe the varia-
bles w and ϕ, and though, instead of applying a linear course of ϕ to describe
the work performed on shear strains, only to take a constant average, see Figure
21-3.

w,x j
g
=
x
i i

Fig. 21-3 Interpretation of the reduced integration at the supporting point i

At numerical integration, this can be reached by choosing a 1–point–integration

for the respective term of work instead of a numerically exact integration em-
ploying two Gaussian supporting points. By analogy with the integration of
the work performed on shear strains concerning 2D plane elements according
to Section 5.2, the procedure is referred to as selectively reduced integration.

The selectively reduced integration eliminates the locking phenomenon. Hence,

the element is not only applicable at rigid shear behavior with GAS → ∞ now,
but also faster converges against the correct solution.
21.1 Elements Employing Displacements as Primary Variables 353

At this case at hand the element matrix comprising selectively reduced integra-
tion differs from the element applying exact integration according to Equation
(21.3) only with regard to four components:
 
δwA : GAS GAS GAS GAS
ℓ 2
− ℓ 2
 
 
GAS EI GAS ℓ GA GAS ℓ
δϕA :  − EI

 2 ℓ
+ 4
− 2S ℓ
+ 4 
 . (21.9)
 GA GAS GAS GAS

δwB : 
 − ℓ
S
− − 
2 ℓ 2 
 
GAS GAS ℓ GAS GAS ℓ
δϕB :
2
− EI
ℓ
+ 4
− 2
EI
ℓ
+ 4

After having introduced the boundary conditions, the computation of the dis-
placement variables of the cantilever according to Figure 21-2 yields the equa-
tions
GAS GAS
δwB : + · wB − · ϕB =P, (21.10)
ℓ 2
 
GAS EI GAS · ℓ
δϕB : − · wB + + · ϕB = 0 , (21.11)
2 ℓ 4

with the solutions

P ℓ2 1
ϕB = · , (21.12)
2 EI
 
1 ℓ2
wB = P ℓ · + . (21.13)
GAS 4EI

Concerning rigid shear behavior GAS → ∞, the part of the deflection related
to shear strains vanishes. The locking phenomenon does not exist. The element
stiffness matrices correctly describe the case of pure bending with respect to
the chosen approaches.

However, at solving the system of equations simulating rigid shear behavior

GAS → ∞, numerical problems might occur, if the bending stiffness exhibits
the same order of magnitude as the numerically representable accuracy of the
shear stiffness – which means that solely the last digit of the shear stiffness is
changed, see e.g. the component [ 4, 4 ] of the element stiffness matrix (21.9).
In this case, the element is not able to represent the bending behavior any
longer, so that the shear stiffness is to be chosen correspondingly to the bending
stiffness.
354 21 Timoshenko Beam Elements

Example
Subsequently the results are represented referring to the example of a cantilever
beam loaded by a concentrated external action, cf. Firgure 21-4. It is essential
that, employing linear approaches, the course of the bending moment is con-
stant inside the element due to M = −EI · ϕ,x and also the course of the shear
force is constant because of the reduced integration of the work terms related
to shear behavior though Q = GAS · (w,x − ϕ) is valid.
P
EI = 4,000 kNm 2
x T
GA S = 200 kN
j
lx = 5m
z,w
P = 2 kN

M [kNm]
-8.75 -6.25 -3.75 -1.25

2.0 Q [kN]

Fig. 21-4 Cantilever, displacement–based element – courses of stress resultants

21.2 Mixed Elements

The work equations applying the PvD and the PvF are formulated incorpo-
rating all state variables [ w ϕ M Q ], following the mixed displacement–force
formulation. Without explicitly deriving the work equations related to PvD and
PvF follow to
Z Z
−δWd = {−δϕ,x M + (δw,x − δϕ) Q } dx − δwp dx

+ [ δϕM e − δwP ] |bound. = 0 , (21.14)

Z
1 1
−δWσ = {−δM (ϕ,x + M ) + δQ [ (w,x − ϕ) − Q ] } dx
EI GAS
+ [ δM (ϕ − ϕe ) + δQ (w − we ) ] |bound. = 0 . (21.15)
If the boundary conditions related to the displacement variables w and ϕ are
exactly fulfilled by the approaches, the contour integrals at the PvF vanish.
Concerning mixed elements, linear approaches to describe the four state varia-
bles [ w ϕ M Q ] cause oscillating shear forces. The oscillating originates from
21.2 Mixed Elements 355

the different orders of the terms being introduced into the condition of equili-
brium M,x − Q = 0. A constant approach to describe Q and accordingly δQ
inside the element eliminates the oscillations, moreover concentrated external
actions P may be put on all nodes of the element.

There is a further possibility to decrease the number of unknowns related to

the element and to save on storage capacity and computing time as well as
to accelerate the convergence of the bending moments and of the deflections
against the exact solution.
Applying Q = M,x and integrating the terms δϕ,x M +δϕM,x at PvD as well as
δM ϕ,x + δM,x ϕ at PvF simplifies the work equations, so that solely w and M
remain as primary variables and the corresponding terms at the boundary are
exactly fulfilled by the approaches. Hereby the shear force and the rotation are
eliminated as variables, which results in describing the virtual work performed
on shear deformations by the derivatives of the bending moments.
Z Z
−δWd = δw,x M,x dx − δw · p dx − [ δwP ] |bound. = 0 , (21.16)
Z
1 1
−δWσ = {δM,x w,x − δM M − δM,x M,x } dx
EI GAS
−[ δM ϕe ] |bound. = 0 . (21.17)

Employing linear approaches to describe w and M the element matrix (21.18)

follows to
 
1 1
δwA : ℓ
− ℓ
 
 
 1 ℓ 1 1 ℓ 1 
δMA :  ℓ
− 3EI
− GAS ℓ
− ℓ
− 6EI
+ GAS ℓ 
 
  (21.18)

δwB :  1 1 
−ℓ ℓ

 
 
δMB : 1 ℓ 1 1 ℓ 1
−ℓ − 6EI + GA ℓ ℓ
− 3EI − GA ℓ
S S

Comparing to the element matrix according to Section 15.3 demonstrates, that

the element matrix, concerning GAS → ∞, describes the Euler–Bernoulli beam
even without any additional procedure.

Example
At testing the mixed element gives excellent results, which are represented in
Figure 21-5. Subsequently the results are represented concerning a cantilever
beam loaded by a concentrated external action. Due to the linear approaches
356 21 Timoshenko Beam Elements

to describe the deflection w and the bending moment M the mixed element
yields the exact result applying the element matrix (21.18) and arbitrary shear
stiffnesses GAS > 0.
P
EI = 4,000 kNm 2
x T
GA S = 200 kN
j
lx = 5m
z,w
P = 2 kN

-10.0
M [kNm]

2.0 Q [kN]

Fig. 21-5 Cantilever, mixed element – courses of stress resultants

21.3 Hybrid–Mixed Elements

By analogy with Section 15 also hybrid–mixed elements may be developed con-
cerning beams that incorporate shear deformations. Thereby the work equati-
ons with respect to the element domain may be formulated by analogy with
Sections 15.2 and 21.2. The procedure to eliminate the shear forces applying
Q = M,x and δQ = δM,x , that has been chosen concerning the mixed method,
may also be established here, so that the deflection w and the bending moment
M are variables related to the element domain. At the interface the moment
equilibrium is formulated applying the virtual rotation δϕ. The supplement
of the work equations, required due to shear deformations, follows by analogy
with Section 21.2 incorporating the term δM,x M,x /GAS in the PvF to
Z
− δWw = {δwE,x · ME,x + δwE · c · wE − δwE · pz } dx , (21.19)

− δWϕ = − {−δϕA · Ma + δϕB · Mb }interf ace , (21.20)

Z
1 1
− δWσ = {δME,x · w,x − δME · · ME − δME ,x ME ,x } dx
EI GAS
− {−δMa · ϕA + δMb · ϕB }interf ace = 0 . (21.21)

By analogy with Section 15.3, the elimination of the bending moments yields
the deflection w and the rotation ϕ being system variables related to the in-
terface. Thus, instead of the boundary conditions of the mixed formulation
21.4 Convergence Behavior Concerning the Beam Elements 357

related to w and M , the boundary conditions related to w and ϕ are now the
substantial ones and are exactly fulfilled by the approaches at the interface.

21.4 Convergence Behavior Concerning the Beam Elements

Subsequently, regarding the simplifying example according to Figure 21-6, the
convergence is investigated. Referring to a bending stiffness EI = 4,000 kN m2
the shear stiffness is varied with GAS = 200 kN, 2,000 kN, 20,000 kN .

p EI = 4,000 kNm 2
T GA S = 200 kN
x
j lx = 5m
z,w p = 2 kN/m

Fig. 21-6 Cantilever incorporating shear deformations – system and loading

Figure 21-7 shows the moment as well as the shear force distribution concerning
the element employing displacements as primary variables and the mixed as well
as the hybrid–mixed element.

M d [kNm] M hm [kNm]
-19.531 -10.156 -3.906 -0.781 -25.00 -14.06 -6.25 -1.56 0.0

Q d [kN] Q hm [kN]
8.75 6.25 3.75 1.25 8.75 6.25 3.75 1.25

left: displacement–based element right: mixed element

hybrid–mixed element

Fig. 21-7 Cantilever incorporating shear deformations – results for 4 elements

Applying the displacement–based element the bending moments are computed

at the element domain employing the first order derivatives of the linear ap-
proch to describe the rotation ϕ, thus the course is constant. The shear force
is computed to be constant with the shear strains γ = w,x − ϕ evaluated at the
barycenter of the element, despite the linear course of ϕ in the element.
Applying the mixed element the deflection and the bending moment are the
primary variables and thus are approximated to have a linear course. The shear
358 21 Timoshenko Beam Elements

force is evaluated from the first order derivative of the bending moment in the
element, hence the courses are constant element by element.
Applying the hybrid–mixed element the bending moment is linearly approxi-
mated in the element domain and the shear force is computed from the first
order derivative of the bending moment.

Regarding the deflection the results are represented in Table 21.1. The ele-
ment employing displacements as primary variables yields, independent of the
number of elements and of the shear stiffness, the exact results concerning the
deflection wT . The mixed and the hybrid–mixed element yield identical results
but only an approximation concerning the deflection wT . Nevertheless, the de-
flection converge against the exact values at an increasing number of elements.
Introducing high values for the shear stiffness, all elements presented here con-
verge against the solution regarding the Euler–Bernoulli beam theory.

Table 21.1 Elements with varying shear stiffness – deflection wT [ m ]

EI/GAS [ m2 ] 20 2 0,2
displacement–based elements 1 0. 1640625 0. 0515625 0. 0403125
2 0. 1640625 0. 0515625 0. 0403125
4 0. 1640625 0. 0515625 0. 0403125
8 0. 1640625 0. 0515625 0. 0403125
mixed elements 1 0. 1770833 0. 0645833 0. 0533333
2 0. 1673177 0. 0548177 0. 0435677
4 0. 1648763 0. 0523763 0. 0411263
8 0. 1642660 0. 0517660 0. 0405160
hybrid–mixed elements 1 0. 1770833 0. 0645833 0. 0533333
2 0. 1673177 0. 0548177 0. 0435677
4 0. 1648763 0. 0523763 0. 0411263
8 0. 1642660 0. 0517660 0. 0405160
22 Plate Elements Including Shear Deformations

A more accurate approximation of the deformation behavior of plates, compa-

ring to Kirchhoff ’s plate theory, also takes into account the shear deformations
acting in direction of thickness of the plate. Therefor kinematics is extended
to the description of the deformation behavior of moderately thick plates via
deformations of the mid place of the plate, too. The extension of the plate
theory has been published by Reissner [86] and Mindlin [72] first.

22.1 Kirchhoff and Reissner–Mindlin Theories by Comparison

The example represented in Figure 22-1 is numerically investigated with respect
to Kirchhoff ’s plate theory and the Reissner–Mindlin plate theory. The results
illustrate the differences between both theories and give guidance concerning
the discretization at the boundary region of the plate.

hing constantly distributed loading: p = 1.0 kN/m2

ed
z x E = 3.0 .107 kN/m2
ed
ng

t = 0.1 m
hi

p v = 0.2
GAS = 26,040 kN/m
e
fre

l x = 10.0 m
y hing
ed l y = 10.0 m

Fig. 22-1 Reinforced concrete plate – system and loading

According to Kirchhoff ’s plate theory the twisting moments are unequal zero
at a simply–supported boundary. Regarding a simply–supported plate they
increase along the boundary up to the corner, where they effect a lift–off force.
Applying the Reissner–Mindlin plate theory the twisting moment at the simply–
supported boundary is equal zero, if the conjugated rotation can be reached
unrestrictedly. This is denoted as a soft bearing. At increasing the shear stiffness
the gradient of the twisting moment continuously rises in the direct neighbor-
hood of the simply–supported boundary, till the twisting moment approaches
Kirchhoff ’s solution at GAS → ∞. This vividly justified limit becomes visible,
when the twisting moment is equal to zero at the boundary and rises to a finite
value at a small distance to the boundary. This implies at the same time, that
a corresponding fine discretization is necessary in the boundary region.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_22
360 22 Plate Elements Including Shear Deformations

In Figure 22-2 the bending and the twisting moments according to Kirchhoff ’s
plate theory are opposed to the results according to the Reissner–Mindlin plate
theory. Symmetries and antisymmetries of the stress resultants can be observed
for both plate theories. However, at the Reissner–Mindlin theory, the bending
moments are mxx 6= 0, myy 6= 0 along the boundaries and, due to the chosen
soft bearing, the twisting moments are mxy = 0.

mxx [ kN m/m ]

myy [ kN m/m ]

mxy [ kN m/m ]

Fig. 22-2 Reinforced concrete plate – bending and twisting moments,

a) left according to Kirchhoff’s plate theory,
b) right according to the Reissner–Mindlin plate theory

Concerning the Reissner–Mindlin theory, perturbations of the bending mo-

ments mxx und myy are existing at the corners related to the simply–supported
as well as to the free boundary, which are caused by the gradients of the conju-
gated rotations. Due to the simply–supported boundaries the bending moments
are zero there. However, regarding the corner, the twistability is disabled be-
cause of the stiffness along the respective rectangular boundaries.
22.1 Kirchhoff and Reissner–Mindlin Theories by Comparison 361

Also the course of the shear forces mainly differs at the boundary region of the
plate, cf. Figure 22-3. Instead of the corner force according to Kirchhoff ’s plate
theory the shear forces qx , qy according to the Reissner–Mindlin theoriy reach
corresponding high values. Applying the Reissner–Mindlin theory, the boun-
dary conditions related to the shear forces qn = 0 and related to the moments
mn = 0, mt = 0 are ensured at the free boundary. Applying Kirchhoff ’s plate
theory, only the conditions related to mn = 0 and related to the Kirchhoff ’s
effective shear force qKn = 0 are fulfilled.

qx [ kN/m ]

qy [ kN/m ]

w [m]

Fig. 22-3 Reinforced concrete plate – bending surface and shear forces,
a.) left according to Kirchhoff’s plate theory,
b.) right according to the Reissner–Mindlin plate theory
362 22 Plate Elements Including Shear Deformations

22.2 Governing Equations of the Reissner–Mindlin Theory

The extension of the governing equations according to Kirchhoff ’s plate theory
acts on the assumption, that the transverse shear deformations in the direction
of thickness are constant and therefor the Kirchhoff hypothesis and subsequent-
ly the Kirchhoff ’s effective shear force are not to be postulated any longer.
Therefore, the twisting moment and the conjugated rotation become indepen-
dent variables and have to be taken into account as boundary conditions if
necessary, cf. Figure 22-4. Thus in contrast to Kirchhoff ’s plate theory, apply-
ing the Reisnsner–Mindlin plate theory three independent boundary conditions
are to be set at each boundary now.

y x
z

mxy
ϕy

myx
ϕx

Fig. 22-4 Twisting moment and corresponding rotation at the boundaries

Furthermore, by employing the transverse shear deformations the work perfor-

med by the shear forces on the conjugated deformations has to be taken into
account, too, so that the work equations are to be extended correspondingly.

a) Kinematics

κxx − ϕx,x = 0
κyy − ϕy,y = 0
2κxy − ϕx,y − ϕy,x = 0
γxz − w,x + ϕx = 0
γyz − w,y + ϕy = 0

Introducing the additional equations related to the shear strains, the rotations
ϕn , ϕt are independent of w and, accordingly, are to be regarded as independent
kinematic boundary conditions. Furthermore, since the shear strains do not
vanish, they need the shear forces as conjugated variables.
22.3 Weak Formulation of the Governing Equations 363

b) Equilibrium
qx,x + qy,y + p = 0
mxx,x + mxy,y − qx + mex = 0
myy,y + mxy,x − qy + mey = 0
Nevertheless the eqilibrium conditions are identical to the ones related to Kirch-
hoff ’s plate theory but analogously to the Timoshenko beam extended by ex-
ternal actions mex , mey . Three independent boundary conditions according to
mn , mt , qn may be forced at each boundary now, whereat external actions at
the boundaries could be applied as well.

c) Material equations
mxx + B(κxx + ν · κyy ) = 0
myy + B(κyy + ν · κxx ) = 0
mxy + B(1 − ν)κxy = 0
qx − GAS γxz = 0
qy − GAS γyz = 0
Hereby, the plate bending stiffness B and the shear stiffness GAS , that, accor-
ding to the Reissner–Mindlin theory, is modified by the shear correction factor
κQ compared to the complete cross section area, are determined as follows:
E · t3 E ·t
B= , GAS = κQ · .
12(1 − ν 2 ) 2(1 + ν)
The shear correction factor κQ is proposed by Reissner to be 5/6, but is chosen
differently, depending on the derivation and on the material.

22.3 Weak Formulation of the Governing Equations

Subsequently, the determination of the element stiffness matrix is presented
concerning elements employing displacements as primary variables. The weak
formulation of the conditions of equilibrium can be declared by the PvD, if the
virtual work performed by the shear forces and the moments on the conjugated
virtual displacements is balanced with
Z
δW = {(δκxx mxx + δκyy myy + 2δκxy mxy ) − (δγxz qx + δγyz qy )} dA
Z
+ {δwp + δϕx mex + δϕy mey } dA = 0 . (22.1)
364 22 Plate Elements Including Shear Deformations

The PvD can be transformed into the work equation, that bases the finite
element formulation, by incorporating the material equations and kinematics.
First of all, introducing the material equations yields
Z
δW = − B{δκxx (κxx + νκyy ) + δκyy (κyy + νκxx ) + 2δκxy (1 − ν)κxy )} dA
Z
− {GAS (δγxz γxz + δγyz γyz )} dA
Z
+ {δwp + δϕx mex + δϕy mey } dA = 0 .

In a second step the equations describing kinematics are introduced replacing

the real as well as the virtual strains:
Z
δW = − B{δϕx ,x (ϕx ,x + νϕy ,y ) + δϕy ,y (ϕy ,y + νϕx ,x )

+ 2 (1 − ν)(δϕx ,y + δϕy ,x )(ϕx ,y + ϕy ,x )} dA

Z
− GAS {(δw,x − δϕx )(w,x − ϕx ) + (δw,y − δϕy )(w,y − ϕy )} dA
Z
+ {δwp + δϕx mex + δϕy mey } dA = 0 .

Since the deflection w and the rotations ϕx , ϕy as well as the respective virtual
displacement variables are independent from each other, the work equation
must be fulfilled separately for all virtual displacement variables.
Z
δWw = − {GAS δw,x (w,x − ϕx ) + GAS δw,y (w,y − ϕy )} dA
Z
+ δwp dA = 0 , (22.2)
Z
δWϕx = − B{δϕx ,x (ϕx ,x + νϕy ,y ) + 2 (1 − ν) δϕx ,y (ϕx ,y + ϕy ,x )} dA
Z Z
+ GAS δϕx (w,x − ϕx ) dA + δϕx mex dA = 0 , (22.3)
Z
δWϕy = − B{δϕy ,y (ϕy ,y + νϕx ,x ) + 2 (1 − ν) δϕy ,x (ϕx ,y + ϕy ,x )} dA
Z Z
− GAS δϕy (w,y − ϕy ) dA + δϕy mey dA = 0 . (22.4)

In the following, the work equations (22.2), (22.3) and (22.4) provide the basis
for the discretization with the finite element method. Therefor the respecti-
ve approaches are to be chosen for the independent displacement variables
[ w ϕx ϕy ].
22.4 Quadrilateral Element Employing a Bi–linear Approach 365

22.4 Quadrilateral Element Employing a Bi–linear Approach

Comparing Reissner–Mindlin plates to the Timoshenko beam it is obvious that
the locking phenomenon may occur as well, what can be avoided by a modifica-
tion of the stiffness matrix. Nevertheless, the procedure to generate the stiffness
matrix should be the same as in the case of beams. The simplest approach deals
with bi–linear shape functions and is referred to as RM-DQ-4 element, what
means Reissner–Mindlin, displacement, quadrilateral element, four nodes.

22.4.1 Approaches to Describe the Deflection and Rotations

The criteria of convergence require, that the approaches to describe the deflec-
tion and the rotations are at least linear, conform and coordinate invariant and
additionally ensure the criterion of rigid body motion. Thus, the bi–linear ap-
proaches already chosen to desribe 2–dimensional structures may be employed.
The bi–linear approaches are given with
 
wA
 wB 
 
 wC 
 
 wD 
 
     ϕ 
w φA φB φC φD 0 0 0 0 0 0 0 0  xA 
 
  ϕ
u =  ϕx  =  0 0 0 0 φA φB φC φD 0 0 0 0  ·  xB 
   
 ϕxC 
ϕy 0 0 0 0 0 0 0 0 φA φB φC φD 
 ϕxD 

 
 ϕy A 
 
 ϕy B 
 
 ϕy C 
ϕy D

and in matrix–vector notation with

u = Ω·v. (22.5)

Analogously the respective virtual displacement fields are described by

δu = Ω · δv . (22.6)

22.4.2 Element Stiffness Matrix and Load Vector

The work equations applying the PvD comprise derivatives of the shape func-
tions with respect to the global x–y–coordinates, that may be summarized by
366 22 Plate Elements Including Shear Deformations

the abbreviations related to the differential operator

Dx = [ ∂x ] , Dy = [ ∂y ] ,
   
Bx = Dx · φA φB φC φD , By = Dy · φA φB φC φD ,
 
φ = φA φB φC φD ,
applying matrix–vector notation. The sequence of the nodal displacement va-
riables is determined by Equation (22.5). Thus, regarding the bending stiffness
as well as the shear stiffness, the element stiffness matrix follows to
K = Kshear + Kbending =
 
T T
 Bx Bx + By By −BTx φ −BTy φ 
Z  
 
GAS 
 −φT Bx φT φ  dA

 
 
−φT By φT φ
 

Z  
 
 1−ν T 1−ν T 
+ B BTx Bx + 2 By By νBTx By + 2 By Bx  dA
 
 
1−ν T 1−ν T
νBTy Bx + 2 Bx By BTy By + 2 Bx Bx

However, employing the linear approaches chosen here, by analogy with the
Timoshenko beam element, the numerically caused locking–phenomenon arises,
that is effected by the transverse shear strains. With
γxz = w,x − ϕx bzw. γyz = w,y − ϕy
and employing linear approaches to describe w and ϕx a constant course con-
cerning w,x and a linear course concerning ϕx are added up, which causes
non–physically oscillations at integration and, as a consequence, an unwanted
stiffening of the element. This phenomenon may by avoided by selectively re-
duced integration of the respective terms of work.

Hence, the numerical integration of the work at element level is performed ap-
plying a 2 × 2 integration concerning the bending parts and a 1 × 1 reduced in-
tegration concerning all shear parts, what leads to an element, which is referred
to as RM-DQ-4-SRI. However, due to the selectively reduced integration, zero–
energy modes may exist. This means, that deformations may arise, that require
22.4 Quadrilateral Element Employing a Bi–linear Approach 367

no stiffness and hence no energy would be stored related to this deformation

mode. Basically the deflection surface w(x, y) is affected, because, due to the
one–point integration, the displacement variables [ wA wB wC wD ] are solely
connected to the shear stiffness related to the center of the element. Therefore,
the deformation w,xy = constant may be freely arranged, if not prevented by
the boundary conditions of the plate. In contrast, the rotations ϕx , ϕy are fixed
by the bending stiffness of the plate. The deformation w,xy = constant may be
stabilized by assigning a substitute stiffness correlated to the deformation.

The basic idea to stabilize the stiffness matrix, is denoted as hour–glass con-
trol referring to Kavangh and Key [54], Kosloff et al. [57], see also Belytschko
et al. [36, 18, 17]. Nonetheless, the choice of the stabilization matrix is diffi-
cult, depends on the model equations, and governs the convergence against the
reference solution.
One possibility consists in weighting the stiffness matrix K1×1 related to the
reduced integration of all parts by the matrix K2×2 related to the complete
integration of all parts. In the following this element is referred to as RM-DQ-
4-HC. With
K = K1×1 + ε · (K2×2 − K1×1 ) (22.7)
the results may be influenced substantially. The multiplier ε has to be adjusted
to the problem to be investigated. Employing ε < 0. 01 already leads to a signi-
ficantly slower convergence of the displacements against the reference solution.
Nevertheless, it is advantageous, that unwanted oscillations of moments and
shear forces are highly damped.
Belytschko et al. [17] develop another stabilization matrix applying only a
weighting of the shear stiffness and propose to connect the weighting to the
element geometry, they implement ε = r · t2 /A, whereat 0. 01 < ε < 0. 05 may
yield feasible results,
K = KB,2×2 + {KS,1×1 + ε · (KS,2×2 − KS,1×1 )} . (22.8)
In the following this formulation is named RM-DQ-BHC.

Element Load Vector

The element load vector comprises the external work performed by surface
loadings p(x, y) as well as mex (x, y), mey (x, y):
 
Z Z p(x, y)
T T 
Ω ·  mex (x, y)

f = Ω · p dA =  dA .
A A
mey (x, y)
368 22 Plate Elements Including Shear Deformations

Subsequent Stress Analysis

Applying matrix notation, first of all, it is valid
σ = E · D· u,
u =Ω·v
and in detail
     
mxx B νB 0 0 0 0 ∂x 0
   νB B 0 0 0     
 myy     0 0 ∂y  w
   1−ν
   
 mxy  = −  0 0 B 0 0 · 0 ∂y ∂x  ·  ϕx 
,
   2   
     
 qx   0 0 0 −GAS 0   ∂x −1 0  ϕy
qy 0 0 0 0 −GAS ∂y 0 −1
 
wA

 wB 


 wC 


 wD 

     ϕxA

w φA φB φC φD 0 0 0 0 0 0 0 0 



     ϕxB 
 x  0 0 0 0
ϕ = φA φB φC φD 0 0 0 0 ·
ϕxC
.
 
ϕy 0 0 0 0 0 0 0 0 φA φB φC φD 
 ϕxD


 

 ϕy A 


 ϕy B 

 ϕy C 
ϕy D
Concerning the bending moments, the shape functions are evaluated at the cor-
ner nodes, whereas, concerning the shear forces, due to the reduced integration
of the stiffness matrix, the evaluation is performed at the center of the element.

22.5 Hybrid–Displacement Quadrilateral Element

Concerning quadrilateral elements applying the Reissner–Mindlin theory with
linear approaches, the locking phenomenon can be avoided by a reduced integra-
tion of shear terms of work. However, thereby zero–energy modes of deformation
may arise, that can be prevented employing hour–glass control. The method
of hour–glass control implies the drawback, that satisfying results may not be
obtained for all configurations and element geometries, but from case to case
either too high stiffnesses may be generated or non–physical deformations may
be only partially prevented.
22.5 Hybrid–Displacement Quadrilateral Element 369

A far more effective method is the application of modified approaches for the
deflection surface w(x, y) as well as for the rotations ϕx (x, y), ϕy (x, y), which
are aligned to kinematics. In Section 22.4.1 bi–linear approaches are chosen
to describe the deflections and the rotations, that result in the phenomena
discussed before. Taking into account that, by analogy with the beam, both
terms describing the shear strains γxz = w,x − ϕx and γyz = w,y − ϕy should
comprise the same polynomial order, then, the terms w,x and w,y should be
described employing bi–linear approaches.
This can be managed, if a hybrid–displacement element is applied, that is
referred to as RM-HDQ-4-5. In addition to the bi–linear approaches to describe
[ w ϕx ϕy ] with respect to the element nodes a−d , which match the approaches
onto the interfaces A − D , additional quadratic approaches of w are chosen
within the element domain, that are assigned to the element nodes e − i subject
to Figure 22-5. This approach is referred to as method of incompatible modes
and discussed by Wilson et al. [104], Taylor et al. [96], Ibrahimbegovic and
Wilson [50], and Simo and Rifai [92].

x A
y
a

h e
i B
d b
D
g f

c
η ξ
C

Fig. 22-5 Element nodes related to the element boundaries and the interfaces

Subsequently the approaches to describe w(x, y) are given, that fulfill this con-
dition. As before, the bi–linear approaches are assigned to the degrees of free-
dom related to the element corner nodes:
1
φa = · ( 1. 0 − ξ ) · ( 1. 0 − η ) ,
4
1
φb = · ( 1. 0 + ξ ) · ( 1. 0 − η ) ,
4
1
φc = · ( 1. 0 + ξ ) · ( 1. 0 + η ) ,
4
1
φd = · ( 1. 0 − ξ ) · ( 1. 0 + η ) .
4
370 22 Plate Elements Including Shear Deformations

The quadratic approaches are hierarchically developed and are assigned to the
degrees of freedom related to the midpoints of the edges and to the center of the
element. It is essential, that the approaches φe , φi are quadratic in ξ–direction
and are linear in η–direction. In contrast, the approaches φf , φh are quadratic
in η–direction and linear in ξ–direction. The approach φi concerning the degree
of freedom related to the center node of the element is bi–quadratic. This yields
1
φe = · ( 1. 0 − ξ 2 ) · ( 1. 0 − η ) ,
2
1
φf = · ( 1. 0 + ξ ) · ( 1. 0 − η 2 ) ,
2
1
φg = · ( 1. 0 − ξ 2 ) · ( 1. 0 + η ) ,
2
1
φh = · ( 1. 0 − ξ ) · ( 1. 0 − η 2 ) ,
2
φi = ( 1. 0 − ξ 2 ) · ( 1. 0 − η 2 ) .

Hence the element comprises 9 degrees of freedom concerning w(x, y) and 4 de-
grees of freedom each to employ bi–linear approaches concerning the rotations
ϕx (x, y) and ϕy (x, y) inside the element domain, but 4 degrees of freedom on
the interfaces for w(s), ϕx (s), ϕy (s) each.

22.5.1 Element Stiffness Matrix

In order to derive the element stiffness matrix the following work equations
have to be taken into account, which are related to Figure 22-6.

A
y x
z

p B
d b
D
σ σ
ϕx
σ c σ
ϕx ϕy
w
w ϕy
C

Fig. 22-6 Displacement variables and stress resultants at the hybrid element
22.5 Hybrid–Displacement Quadrilateral Element 371

The Principle of virtual Displacements has to be applied inside the element

domain a − b − c − d and at the interface A − B − C − D with respect to
the chosen shape functions. Between the interface and the element domain the
Principle of virtual Forces has to be applied in order to satisfy the continuity
conditions with respect to the displacements. Thus the equations
Z
δWd, I = − δuTI, lin. · σ ds 6= 0
I
Z
δWd, E = + (δuTE, lin.+ δuTE, quadr. ) · σ ds
E
Z
(δuTE, lin.+ δuTE, quadr. ) · DT · E · D · (uE, lin.+ uE, quadr. ) dA = 0
−
E
Z Z
δWσ = − δσ T · uI, lin. ds + δσ T · (uE, lin.+ uE, quadr. ) ds = 0
I E

have to be discretized. The first equation ensures equilibrium at the system’s

level between neighboring elements. The second equation describes the virtual
work performed at the domain of the element. The third equation leads to the
orthogonality condition
Z
δσ T · uE, quadr. ds = 0 ,
E

if the shape functions at the interface ui, lin. as well as at the element domain
uE, lin. employ identical courses and variables. Because of the orthogonality
condition also
Z
δuTE, quadr. · σ ds = 0
E

is valid and has to be taken into account with respect to δWd, E = 0. This
yields
Z Z
δuTE, lin. · σ ds = δuTE, lin. · DT · E · D · (uE, lin. + uE, quadr. ) dA ,
E E
Z
0= δuTE, quadr. · DT · E · D · (uE, lin. + uE, quadr. ) dA ,
E

whereat the first integral describes the work at the interface

Z
δWd, I = δuTE, lin. · DT · E · D · (uE, lin. + uE, quadr. ) dA 6= 0
E

and the second equation can be applied to eliminate the variables of the qua-
dratic shape functions at the element level.
372 22 Plate Elements Including Shear Deformations

The computation of the element stiffness matrix comprises two steps.

In a first step the work terms concerning bending as well as shear are evaluated
for all approaches according to the scheme presented in Section 22.4.2. Hereby
it seems to be reasonable to process the integration regarding the bi–linear
approaches by analogy with Section 22.4.2, so that the sub–matrix K12, 12 is
related to the respective 12 degrees of freedom of the corner nodes of the
interface. The work terms concerning the quadratic approaches are stored in the
sub–matrices K12, 5 , K5, 12 , K5, 5 , so that the stiffness matrix gets the following
scheme:
" #
K12, 12 K12, 5
K17, 17 = .
K5, 12 K5, 5

In a second step the degrees of freedom related to the quadratic approaches are
eliminated, so that the remaining element stiffness matrix K is solely assigned
to the degrees of freedom related to the corner nodes. With
" # " # " #
K12, 12 K12, 5 v12 f12 6= 0 at element level
· −
K5, 12 K5, 5 v5 05 = 0 at element level

and applying static condensation it follows at element level

( K12, 12 + K12, 5 · K−1

5, 5 · K5, 12 ) · v12 − f12 6= 0 . (22.9)

Due to the final representation of the work equations the elimination of the
degrees of freedom v5 is possible without taking into account the respective
contour integrals. Thus the element continuity conditions may be ensured only
concerning the linear approaches.

22.5.2 Subsequent Stress Analysis

The subsequent stress analysis is provided following Section 22.4.2. Applying
matrix–vector notation yields

σ = E · D· u,
u =Ω·v

The bending moments and the shear forces are computed inside the element
domain by means of the bi–linear shape functions only, without considering the
quadratic parts. The evaluation of the shape functions takes place concerning
the bending moments at the corner nodes and concerning the shear forces at
the center of the element in order to avoid unphysical oscillations.
22.6 Comparison of the Elements 373

22.6 Comparison of the Elements

Although the different elements satisfy the fundamental requirements regarding
the criteria of convergence, their ability to describe the bending and the twisting
behavior has to be investigated, since the reduced integration, the stabilization
and the hybrid–dispalcement approach may effect undesired phenomena caused
by numerical inconsistencies. The results obtained from employing different ele-
ments are presented in Tables 22.1 and 22.2, comparing the following elements:
22.4 : RM-DQ-4 completely integrated
22.4 : RM-DQ-4-SRI selectively reduced integrated
22.4 : RM-DQ-4-HC reduced integrated,
including stabilization ε = 0. 01/A
22.4 : RM-DQ-4-BHC, selective reduced integration,
including stabilization ε = 0. 05 · t2 /A
22.5 : RM-HDQ-4-5 hybrid–displacement element

22.6.1 Test 1
At first a cantilever plate loaded by a twisting moment is investigated accor-
ding to Figure 22-7. The main deformation behavior is related to twisting, but
in the vicinity of the load application a small part of the loading is carried via
bending. The shear deformations are negligible. As a reference concerning the
displacement variables and the twisting moment the values of the correspon-
ding torsion bar may be chosen:

GIT = (E/2) · (b · t3 /3) = 33,333 kN m2 ,

ϑ = −1. 451 · 10−4 [ 1 ] at load application,
w = ± 0. 728 · 10−4 m at load application,
MT = 1. 0 kN m at load application.

5m
P
P
1m

x EI = 16,666 kNm2
y GA S = 10 7 kN
z, w
P = 1 kN

Fig. 22-7 Cantilever plate – system and torsional loading

374 22 Plate Elements Including Shear Deformations

The cantilever plate is investigated employing the thickness t = 0. 1 m and the

material parameters E = 2. 0 · 108 kN/m2 , and G = 1. 0 · 108 kN/m2 . Table
22.1 represents the convergence behavior concerning the essential state varia-
bles. The study illustrates that not all elements converge against the correct
solution, but numerical results partly lead to large deviations with respect to
the reference values at the torsion bar. It should be noted, that the given value
of the twisting moment is mxy = 0. 5 kN m/m, since the concentrated loadings
act as corner forces on the plate.

Table 22.1 Cantilever plate – results concerning the load of P = ±1. 0 kN

elements 1×1 1×2 1×4 1×8 1 × 16

acc. to 22.4 10000 · ϑ −0. 212 −0. 568 −0. 984 −1. 242 −1. 367
RM-DQ-4 10000 · w ±0. 106 ±0. 284 ±0. 493 ±0. 622 ±0. 686
mxy 0. 035 0. 233 0. 431 0. 493 0. 488
0. 087 0. 263 0. 439 0. 493 0. 489
acc. to 22.4 10000 · ϑ −22. 66 −74. 18 −171. 7 −256. 0 −291. 7
RM-DQ-4-SRI 10000 · w ±11. 33 ±37. 09 ±85. 90 ±128. 0 ±145. 9
mxy 3. 778 3. 273 2. 316 1. 490 1. 140
−1. 778 −1. 273 −0. 316 0. 510 0. 860
acc. to 22.4 10000 · ϑ −8. 634 −5. 176 −3. 376 −2. 442 −1. 968
RM-DQ-4-HC 10000 · w ±4. 317 ±2. 588 ±1. 688 ±1. 221 ±0. 984
mxy 1. 439 0. 295 0. 489 0. 472 0. 447
−0. 378 0. 679 0. 508 0. 516 0. 511
acc. to 22.4 10000 · ϑ −20. 24 −36. 67 −29. 69 −17. 81 −10. 20
RM-DQ-4-BHC 10000 · w ±10. 12 ±18. 33 ±14. 85 ±8. 918 ±5. 126
mxy 3. 375 1. 541 0. 588 0. 497 0. 471
−1. 491 −0. 161 −0. 461 0. 500 0. 483
acc. to 22.5 10000 · ϑ −1. 394 −1. 408 −1. 435 −1. 449 −1. 451
RM-HDQ-4-5 10000 · w ±0. 697 ±0. 704 ±0. 718 ±0. 726 ±0. 728
mxy 0. 232 0. 660 0. 519 0. 498 0. 491
0. 686 0. 398 0. 491 0. 500 0. 495

22.6.2 Test 2
As a second example the cantilever plate loaded by pure bending is investigated
according to Figure 22-8. The results may be calibrated to the reference values
of a corresponding beam, whereby the shear deformations are negligible here.
Concerning the beam and employing E = 2. 0·108 kN/m2 , G = 1. 0·108 kN/m2 ,
22.6 Comparison of the Elements 375

and t = 0. 1 m the reference values follow to

w = wB + wS = 0. 0050 + 0. 000001 ≈ 0. 005 m at load application,
mxx = −10. 0 kN m/m at clamping,
qx = 2. 0 kN/m along the entire beam.
5m
P
P
1m

x EI = 16,666 kNm2
y GA S = 10 7 kN
z, w
P = 1 kN

Fig. 22-8 Cantilever plate – system and external bending action

The results of this loading case are represented in Table 22.2. As in the case
of torsional loading not all elements converge against the correct solution. In
contrast, here, the element RM-DQ-4 of Section 22.4 hardly converges due
to locking phenomena, comparable to the behavior of the Timoshenko beam
element, whereat the other elements converge well against the correct solution.

Table 22.2 Cantilever plate – results concerning the load of P = +1. 0 kN

elements 1×1 1×2 1×4 1×8 1 × 16

acc. to 22.4 100 · w 0. 0004 0. 00160 0. 00632 0. 0243 0. 0850
RM-DQ-4 mxx −0. 004 −0. 024 −0. 111 −0. 457 −1. 647
qx 2. 00 2. 00 2. 00 2. 00 2. 00
acc. to 22.4 100 · w 0. 3751 0. 4688 0. 4923 0. 4981 0. 4996
RM-DQ-4-SRI mxx −5. 00 −7. 50 −8. 75 −9. 375 −9. 687
qx 2. 00 2. 00 2. 00 2. 00 2. 00
acc. to 22.5 100 · w 0. 1074 0. 2084 0. 3030 0. 3796 0. 4321
RM-DQ-4-HC mxx −1. 429 −3. 333 −5. 385 −7. 143 −8. 378
qx 2. 00 2. 00 2. 00 2. 00 2. 00
acc. to 22.5 100 · w 0. 3334 0. 4413 0. 4774 0. 4905 0. 4957
RM-DQ-4-BHC mxx −4. 444 −7. 059 −8. 485 −9. 231 −9. 612
qx 2. 00 2. 00 2. 00 2. 00 2. 00
acc. to 22.5 100 · w 0. 3751 0. 4688 0. 4923 0. 4981 0. 4996
RM-HDQ-4-5 mxx −5. 00 −7. 50 −8. 75 −9. 375 −9. 687
qx 2. 00 2. 00 2. 00 2. 00 2. 00
376 22 Plate Elements Including Shear Deformations

22.6.3 Remarks
The chosen structure is statically determined, so that, applying the PvD, the
equilibrium is ensured and the resultant stress variables converge against the
correct values in almost all cases. The displacements naturally depend on the
stiffnesses, that strongly differ from each other in parts according to the element
formulations shown here, hence the following comments may be given.

RM-DQ-4 – Complete integration according to Section 22.4:

• The element employing bi–linear approaches according to Section 22.4
shows at pure bending the locking phenomenon, since the shear stiffness
is very high compared to the bending stiffness. At pure bending the ele-
ment converges against the wrong values.
• Regarding torsional loading the element slowly converges against the re-
ference values.

RM-DQ-4-SRI – Selectively reduced integration according to Section 22.4:

• At pure bending the element comprising a 1 × 1–integration concerning
the work terms related to shear deformation behavior yields the same
values as the hybrid–displacement element, regarding this loading case it
converges against the reference values.
• Regarding torsional loading the element diverges concerning the displa-
cement variables. This is caused by the zero–energy mode related to w,xy ,
that is directly attracted by the loading here. The averaged twisting mo-
ment converges against mxy = 1. 0 kN m/m.

RM-DQ-4-HC – Completely reduced integration and stabilization comprising

hour–glass control according to Section 22.4 employing ε = 0. 01/A:
• Regarding bending the element comprising hour–glass control very slowly
converges against the reference values. The convergence behavior of the
element depends on the choice of the stabilization parameter ε.
• At twisting the convergence is very slowly, too. Choosing a constant ε
gives wrong results, which are not presented here.

RM-DQ-4-BHC – Selectively reduced integration and stabilization comprising

hour–glass control according to Section 22.4 employing ε = 0. 05 · t2 /A:
• Regarding bending the element comprising hour–glass control slowly con-
verges against the reference values. The convergence behavior of the ele-
ment depends on the choice of the stabilization parameter ε.
• At twisting the convergence cannot be trusted.
22.6 Comparison of the Elements 377

RM-HDQ-4-5 – Hybrid–displacement element according to Section 22.5:

• Concerning bending and twisting the element converges fast and reliably
against the analytical values regarding a respective beam loaded by ben-
ding as well as twisting.
• Inside the element at the position of load application, the twisting mo-
ment converges against the reference value mxy = 0. 5 kN m/m.

Hence it becomes obvious, that only the hybrid–displacement element according

to Section 22.5 reliably converges against the reference values concerning both
loading cases.

22.6.4 Benchmark
As a benchmark to test the element, again, by analogy with the triangular
Kirchhoff plate elements according to Section 20, the unilaterally clamped
rhombus plate in Figure 22-9 is investigated referring to [103, 27]. All calculati-
ons are performed with the element employing modified approaches according
to Section 22.5.

constantly distributed loading: p = 0.26066 psi free M

E = 1.05 .107 psi
t = 0.125 in
e
fre

ly
ν = 0.3
e
fre

lx = 12.0 in
ly = 8.48 in y S
x E clamped
GA S variable
lx

Fig. 22-9 Rhombus plate – system and loading

The bending and the transverse shear stiffnesses are specified to

E · t3
B= = 1,878 lbf. in2/in ,
12 · (1 − ν 2 )
E·t
GAS = κQ · = κQ · 505,000 lbf ,
2 · (1 + ν)
whereby κQ acts as a parameter to scale the shear stiffness. Due to the slender-
ness of the plate, exhibiting a ratio ℓ / t = 12 / 0. 125 = 96, the shear deformati-
ons are negligible at realistic values of QAS . Thus, subsequently the parameters
378 22 Plate Elements Including Shear Deformations

κQ = 0. 1−1. 0−10. 0 as well as the thickness of the plate t = 0. 125−1. 25 in are

varied. Applying t = 0. 125 in the shear deformations are negligible, since the
plate is extremely slender. Applying t = 1. 25 in the plate is moderately slen-
der, so that the bending stiffness is relatively high and the shear deformations
are to be taken into account.
Figure 22-10 represents the study applying t = 1. 25 in. The behavior of con-
vergence of the deflection illustrates a monotonic approximation to limit values
depending on the shear stiffness. The triangular element w21 is given for compa-
rison. Regarding high values of the shear stiffness the Reissner–Mindlin element
converges against the value according to Kirchhoff ’s plate theory.

100 wM [inch]

0.70

0.60

20 100 500 1000 2000 dof

0.50

0.40

0.30

K-D-6-21 Kirchhoff
0.20 GA s = 50500 lbf
GA s = 505000 lbf
GA s = 5050000 lbf

Fig. 22-10 Rhombus plate – convergence behavior of the deflection wM

The subsequent Tables 22.3 and 22.4 comprise the results concerning the de-
flections wM and the clamping moment myyE . In accordance with this study
the following statements can be made:

1. Applying small bending stiffnesses the bending deformation is dominant,

so that, already by employing a relatively small number of elements, the
results concerning the stress resultants lie within the range of the refe-
rence solution.
2. Applying a higher bending stiffness (t = 1. 25 in) the shear deformations
become dominant, so that the results converge against other solutions.
22.6 Comparison of the Elements 379

3. Regarding increasing shear stiffnesses, the element converges against the

solution applying Kirchhoff ’s plate theory.
4. In the case of increasing shear stiffnesses and a small bending stiffness
(t = 0. 125 in) the load carrying behavior changes, whereat the clam-
ping moment myyE only slowly converges against the solution applying
Kirchhoff ’s plate theory.

Table 22.3 Rhombus plate – results concerning p = 0. 26066 psi, t = 0. 125 in

units: w [ in ], m [ lbf.in/in ]

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 1 4 16 64 256 1024
unknowns 12 27 75 243 867 3267
κQ

0. 1 wM 0. 4545 0. 2911 0. 2785 0. 2834 0. 2870 0. 2891

myyE −6. 704 −7. 723 −7. 688 −7. 465 −7. 312

1. 0 wM 0. 4539 0. 2901 0. 2774 0. 2823 0. 2856 0. 2872

myyE −6. 697 −7. 718 −7. 689 −7. 426 −7. 253

10. 0 wM 0. 4539 0. 2900 0. 2737 0. 2822 0. 2854 0. 2868

myyE −6. 696 −7. 717 −7. 689 −7. 421 −7. 236

Table 22.4 Rhombus plate – results concerning p = 0. 26066 psi, t = 1. 25 in

units: w [ in ], m [ lbf.in/in ]

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 1 4 16 64 256 1024
unknowns 12 27 75 243 867 3267
κQ

0. 1 1000 · wM 0. 9686 0. 7702 0. 7388 0. 7362 0. 7376 0. 7387

myyE −6. 545 −7. 720 −8. 430 −8. 684 −8. 765

1. 0 1000 · wM 0. 5111 0. 3677 0. 3489 0. 3510 0. 3542 0. 3562

myyE −6. 867 −7. 903 −8. 151 −8. 096 −8. 026

10. 0 1000 · wM 0. 4598 0. 3004 0. 2873 0. 2920 0. 2961 0. 2985

myyE −6. 758 −7. 786 −7. 776 −7. 607 −7. 480
380 22 Plate Elements Including Shear Deformations

22.7 Displacement–Based Triangular Element

Subsequently the derivation of the element stiffness matrix is represented con-
cerning triangular elements employing displacements as primary variables. The
weak formulation of the equilibrium conditions can be aligned applying the
PvD. Since the deflection w and the rotations ϕx , ϕy as well as the respec-
tive virtual displacements are independent, the work equation is to be fulfilled
separately for all virtual displacement variables:
Z
δWw = − {GAS δw,x (w,x − ϕx ) + GAS δw,y (w,y − ϕy )} dA
Z
+ δwpdA = 0 , (22.10)
Z
δWϕx = − B{δϕx ,x (ϕx ,x + νϕy ,y ) + 2 (1 − ν) δϕx ,y (ϕx ,y + ϕy ,x )} dA
Z Z
+ GAS δϕx (w,x − ϕx ) dA + δϕx mex dA = 0 , (22.11)
Z
δWϕy = − B{δϕy ,y (ϕy ,y + νϕx ,x ) + 2 (1 − ν) δϕy ,x (ϕx ,y + ϕy ,x )} dA
Z Z
− GAS δϕy (w,y − ϕy ) dA + δϕy mey dA = 0 . (22.12)

As stated in Section 22.4, the work equations (22.10), (22.11) and (22.12) are
basis for the discretization employing triangular elements. Therefor the respec-
tive approaches are to be chosen to describe the independent displacement
fields w, ϕx , ϕy .

22.7.1 Approaches to Describe the Displacement Variables

The criteria of convergence claim the approaches for the deflection and the
rotations to be at least linear, conform and coordinate invariant and in addition
to fulfill the criterion of rigid body motion.

Employing triangular elements it seems to be reasonable to choose complete

approaches in order to ensure the coordinate invariance from the very begin-
ning. Furthermore the approaches for w as well as for ϕx and ϕy should be
chosen thus being able to describe the shear strains w,x − ϕx and w,y − ϕy
without numerically caused but unphysical oscillations. This can be achieved
by employing quadratic approaches to describe w and linear approaches to des-
cribe ϕx , ϕy or respective approaches of higher polynomial order. This element
is referred to as RM-DT-6-12.
22.7 Displacement–Based Triangular Element 381

w ϕx ϕy
A

w w
F D

C E B
w ϕx ϕy w w ϕx ϕy

Fig. 22-11 Reissner–Mindlin theory – displacement variables for w and for ϕx , ϕy

Quadratic approaches to describe w require six element nodes, linear approa-

ches to describe ϕx , ϕy only comprise three nodes as represented in Figure
22-11. At the end an element is developed incorporating varying degrees of
freedom with respect to the element nodes. This combination of element nodes
and variables is comparable to the elements presented in Sections 10 and 18.
The approaches to describe w and ϕx , ϕy are marked by ϕ2i concerning qua-
dratic shape functions and by ϕ1i concerning linear shape functions to simplify
the notation. Thereby, in total, the element comprises 12 degrees of freedom.
Applying matrix–vector notation the approaches are given with
 
wA
 wB 
 
 wC 
 
 wD 
 
   2 2 2 2 2 2   wE 
w φA φB φC φD φE φF 0 0 0 0 0 0  
  wF 
 
   1 1 1
u =  ϕx  =  0 0 0 0 0 0 φA φB φC 0 0 0  ·  
 ϕxA 
ϕy 0 0 0 0 0 0 0 0 0 φ1A φ1B φ1C 
 ϕxB 

 
 ϕxC 
 
 ϕy A 
 
 ϕy B 
ϕy C
and introducing matrix symbols with
u = Ω·v. (22.13)
Correspondingly the virtual displacemt fields are described by
δu = Ω · δv . (22.14)
By complete analogy the element employing cubic approaches to describe w as
well as quadratic approaches to describe the rotations ϕx , ϕy may be developed
with overall 21 nodal degrees of freedom.
382 22 Plate Elements Including Shear Deformations

22.7.2 Element Stiffness Matrix and Load Vector

The element stiffness matrix is developed according to Section 22.4. Correspon-
ding to the different approaches the derivatives related to w are defined:

Dx = [ ∂x ] , Dy = [ ∂y ] ,
   
Bxw = Dx · φ2A
φ2B φ2C φ2D φ2E φ2F , Byw = Dy · φ2A φ2B φ2C φ2D φ2E φ2F ,
 2 2 2 2 2 2 
φw = φA φB φC φD φE φF .

The derivatives of the shape functions related to the rotations may also be
summarized in matrix notation by
   
Bxϕ = Dx · φ1A φ1B φ1C , Byϕ = Dy · φ1A φ1B φ1C ,
 
φϕ = φ1A φ1B φ1C .

The sequence of the nodal displacement variables is determined according to

Equation (22.13). Therefor the element stiffness matrix can be assembled con-
cerning the bending stiffesses as well as the shear stiffnesses, which yields

K=
 
 BTxw Bxw + BTyw Byw −BTxw φϕ −BTyw φϕ 
Z  
 
GAS 
 −φTϕ Bxw φTϕ φϕ  dA

 
 
−φTϕ Byw φTϕ φϕ
 

Z  
 
 1−ν T 1−ν T 
+ B BTxϕ Bxϕ + 2 Byϕ Byϕ νBTxϕ Byϕ + 2 Byϕ Bxϕ  dA .
 
 
1−ν T 1−ν T
νBTyϕ Bxϕ + 2 Bxϕ Byϕ BTyϕ Byϕ + 2 Bxϕ Bxϕ

Element Load Vector

The element load vector comprises the work performed by the external surface
action, given by p(x, y) as well as by mex (x, y), mey (x, y),
 
Z Z p(x, y)
ΩT · p dA = ΩT ·  mex (x, y)  dA .
 
f =
A A
mey (x, y)
22.7 Displacement–Based Triangular Element 383

22.7.3 Subsequent Stress Analysis

The computation of moments and shear forces takes place by analogy with
Section 22.4.2. Employing linear approaches to describe the rotations ϕx , ϕy
the course of the moments is constant inside the element, and correspondingly,
employing quadratic approaches the course is linear. The subsequent stress
analysis concerning the moments is carried out with the stress matrix Sm , that
is formulated comprising the approaches for the rotations:

σ m = Em · D · Ωϕ · vϕ = Sm · vϕ .

Employing linear approaches to describe the rotations the moments can be

evaluated to
 
ϕxA
     ϕ 
mxx ba bb b c ν · aa ν · ab ν · ac  xB 
  B    ϕxC 
 myy  = ·  ν · b a ν · b b ν · b c aa aa aa  ·  
2A∆  ϕy A
 

mxy e · aa e · ab e · ac e · b a e · b a e · b a  ϕy B 
ϕy C

with e = (1 − ν)/2 , B = E · t3 /12 (1 − ν 2 ) .

The subsequent stress analysis concerning the shear forces takes place applying
kinematics γxz = w,x − ϕx and γyz = w,y − ϕy . Matrix–vector notation yields

σ q = E q · D · Ω · v = Sq · v .

Due to the chosen approaches the courses of the shear forces are linear inside
the element, and in detail given by
 
wA
 wB 
 
     wC 
qxA 3ba −bb −bc 4bb 0 4bc −1 0 0 0 0 0  
  wD 
 
  
 qyA   3aa −ab −ac 4ab 0 4ac 0 0 0 −1 0 0   wE 
     
 qxB  GA  −b 3b −b 4b 4b 0 0 −1 0 0 0 0   wF 
  S  a b c a c   
 = · · .
 qyB  2A∆  −aa 3ab −ac 4aa 4ac 0 0 0 0 0 −1 0   ϕxA 
     ϕ 
 qxC   −b −b 3b 0 4b 4b 0 0 −1 0 0 0   xB 
   a b c b a   ϕ 
 xC 
qyC −aa −ab 3ac 0 4ab 4aa 0 0 0 0 0 −1  
 ϕy A 
 
 ϕy B 
ϕy C
384 22 Plate Elements Including Shear Deformations

22.7.4 Behabior of Convergence Concerning the Triangular Element

As an example to test the formulations and by analogy with the triangular
elements related to Kirchhoff ’s plate theory presented in Section 20, the unila-
terally clamped rhombus–plate is investigated referring to [27], cf. Figure 22-12.
The bending and the transverse shear stiffnesses are defined to
E · t3
B= = 1,878 lbf. in2/in ,
12 · (1 − ν 2 )
E·t
GAS = κQ · = κQ · 505,000 lbf ,
2 · (1 + ν)
whereby κQ serves as a parameter to scale the shear stiffness. Subsequently,
for κ = 10. 0 and t = 0. 125 in the linear / quadratic approaches to describe
ϕx , ϕy , w are compared to quadratic / cubic approaches for ϕx , ϕy , w. This
study makes possible the following statements:
1. Due to the larger effort, the approaches of higher order converge signi-
ficantly faster and show an acceptable solution though the number of
elements is comparatively low.
2. The approaches of less order converge against the correct solution, too,
but are particularly non–satisfying concerning the stress resultants.
3. Regarding the chosen geometries and shear stiffnesses the results converge
against the solutions related to Kirchhoff ’s plate theory.

constantly distributed loading: p = 0.26066 psi free M

E = 1.05 .107 psi
t = 0.125 in
e
fre

ly

ν = 0.3
e
fre

lx = 12.0 in
ly = 8.48 in y S
x E clamped
GA S = 5.05 .105 lbf
lx ly

Fig. 22-12 Rhombus plate – system and loading

The subsequently given Table 22.5 illustrates the behavior of convergence re-
lated to the deflection wM and the clamping moment myyE employing w–
quadratic / ϕ–linear approaches and w–cubic / ϕ–quadratic approaches in
comparison. The clamping moment is evaluated separately concerning the three
elements meeting at the respective corner at positon E.
22.7 Displacement–Based Triangular Element 385

Table 22.5 Rhombus plate – p = 0. 26066 psi, t = 0. 125 in, κQ = 10. 0

units: w [ in ], myyE [ lbf.in/in ]

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 2 8 32 128 512 2048
wqua. wM 0. 0938 0. 0997 0. 1070 0. 1435 0. 2234 0. 2731
myyE −6. 089 −7. 590 −8. 738 −8. 358 −7. 291
−6. 073 −7. 490 −8. 259 −7. 599 −6. 931
−6. 109 −7. 728 −9. 415 −9. 441 −7. 801
wcub. wM 0. 1886 0. 2401 0. 2660 0. 2784 0. 2850 0. 2871
myyE −9. 362 −7. 915 −6. 522 −6. 776 −7. 006
−9. 325 −7. 821 −6. 728 −6. 844 −7. 011
−9. 410 −8. 030 −6. 262 −6. 684 −6. 998

Subsequently, employing the w–cubic / ϕ–quadratic approaches, the results

are represented in Tables 22.6 and 22.7, whereat the shear stiffness is varied
with κQ = 0. 1, κQ = 1. 0, κQ = 10. 0 as well as the thickness is varied with
t = 0. 125 in, t = 1. 25 in.

Table 22.6 Rhombus plate – p = 0. 26066 psi, t = 0. 125 in

units: w [ in ], myyE [ lbf.in/in ]

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 2 8 32 128 512 2048
unknowns 17 43 131 451 1667 6403
κQ
0. 1 wM 0. 1894 0. 2444 0. 2706 0. 2841 0. 2885 0. 2901
myyE −8. 843 −7. 769 −7. 094 −7. 102 −7. 118
−7. 764 −7. 591 −7. 208 −7. 126 −7. 124
−10. 23 −7. 906 −6. 893 −7. 062 −7. 109
1. 0 wM 0. 1887 0. 2412 0. 2676 0. 2809 0. 2864 0. 2879
myyE −9. 212 −7. 901 −6. 813 −7. 008 −7. 050
−8. 914 −7. 709 −7. 136 −7. 038 −7. 056
−9. 597 −8. 112 −6. 375 −6. 960 −7. 041
10. 0 wM 0. 1886 0. 2401 0. 2660 0. 2784 0. 2850 0. 2871
myyE −9. 362 −7. 915 −6. 522 −6. 776 −7. 006
−9. 325 −7. 821 −6. 728 −6. 844 −7. 011
−9. 410 −8. 030 −6. 262 −6. 684 −6. 998
386 22 Plate Elements Including Shear Deformations

Table 22.7 Rhombus plate – p = 0. 26066 psi, t = 1. 25 in

units: w [ in ], myyE [ lbf.in/in ]

mesh 1×1 2×2 4×4 8×8 16 × 16 32 × 32

elements 2 8 32 128 512 2048
unknowns 17 43 131 451 1667 6403
κQ
0. 1 1000 · wM 0. 5111 0. 6161 0. 6873 0. 7169 0. 7294 0. 7349
myyE −8. 789 −8. 666 −8. 732 −8. 762 −8. 786
−7. 771 −8. 506 −8. 636 −8. 740 −8. 782
−10. 29 −8. 883 −8. 821 −8. 775 −8. 782
1. 0 1000 · wM 0. 2374 0. 2965 0. 3333 0. 3480 0. 3538 0. 3563
myyE −8. 787 −8. 040 −7. 842 −7. 861 −7. 894
−6. 955 −7. 747 −7. 777 −7. 850 −7. 892
−11. 20 −8. 376 −7. 886 −7. 859 −7. 886
10. 0 1000 · wM 0. 1959 0. 2503 0. 2806 0. 2930 0. 2975 0. 2995
myyE −8. 765 −7. 741 −7. 280 −7. 275 −7. 299
−6. 994 −7. 414 −7. 318 −7. 291 −7. 302
−11. 00 −8. 045 −7. 167 −7. 236 −7. 284

Remarks
Comparing the quadrilateral elements according to Section 22.4.2 to the trian-
gular element employing a quadratic / cubic approach it may be stated:
• Both types of elements converge against the reference solution.
• The triangular element discretizes the deflection employing a complete cu-
bic approach and the rotations employing a complete quadratic approach.
• The quadrilateral element discretizes the deflection as well as the rota-
tions employing bi–linear approaches whereby the deflection comprises
additional quadratic approaches being adapted to the rotations.
• At applying the same element mesh the triangular element exhibits round
about twice the number of degrees of freedom.
• Concerning the triangular element the displacement and the stress resul-
tant variables faster converge against the reference solution at taking into
account the same number of degrees of freedom.
• Concerning the triangular element the stress resultant variables converge
bottom–up against the reference solution. Concerning the quadrilateral
element the stress resultant variables converge top–down against the re-
ference solution.
22.8 Mixed Quadrilateral Element 387

22.8 Mixed Quadrilateral Element

Concerning the mixed formulation of the governing equations both forces and
moments as well as deflection and rotations may be taken into account as
variables. The number of variables can be reduced to the primary variables
[ w mxx myy mxy ], if a transformation of the work equations is performed com-
parable to the procedure, that has been established concerning the Timoshenko
beam. In the following the weak formulation of the equilibrium conditions ap-
plies the Principle of virtual Displacements whereas the weak formulation of
the conditions of deformation applies the Principle of virtual Forces. In the
following sections a mixed element referred to as RM-MQ-4-16 is developed.

Conditions of Equilibrium
The work equation applying the PvD describes the internal work of bending
and twisting moments as well as shear forces on the conjugated virtual strains
and curvatures. The external work takes into account the pressure p, and the
bending moments mex , mey at the boundary of the plate. In contrast to Kirch-
hoff ’s plate theory the twisting moment mexy can act on the conjugated virtual
rotations δϕx , δϕy at the boundary. Thus the Principle of virtual Displacements
includes the follwing terms
Z
δWd = {(δκxx mxx + δκyy myy + 2δκxy mxy ) − (δγxz qx + δγyz qy ) + δwp}dA
Z Z
+ {δϕx mex + δϕy mexy } dy + {δϕy mey + δϕx mexy } dx = 0 . (22.15)

The PvD may be extended by further external actions like concentrated loa-
ding P and line shaped loading qxe , qye if necessary. Furthermore, areal moments
mex , mey could be considered as well, but are neglected in the presentation for
simplicity. The PvD may be transformed by employing
Z Z
{δκxx mxx − δγxz qx }dA + δϕx mex dy
Z Z
= {δϕx ,x mxx − (δw,x − δϕx )(mxx ,x + mxy ,y )}dA + δϕx mex dy
Z Z
= {−δw,x (mxx ,x + mxy ,y ) + (δϕx mxx ),x + δϕx mxy ,y }dA + δϕx mex dy
Z Z
= {−δw,x (mxx ,x + mxy ,y ) + δϕx mxy ,y }dA + δϕx (mx + mex ) dy

and accordingly the work performed by the bending moments myy and by the
shear forces qy might be transformed. Thus only the moments [ mxx myy mxy ]
388 22 Plate Elements Including Shear Deformations

remain as primary variables of the actual stress state, whereas the virtual de-
flection δw and the virtual rotations δϕx , δϕy describe the virtual deformations:
Z
δWd = {− δw,x (mxx ,x + mxy ,y ) − δw,y (myy ,y + mxy ,x )

+ (δϕx mxy ,y + δϕy mxy ,x ) + (δϕx ,y + δϕy ,x ) mxy + δwp} dA

Z Z
+ δϕy mexy dy + δϕx mexy dx
Z Z
+ δϕx (mx + mex ) dy + δϕy (my + mey ) dx = 0 . (22.16)

The work equation may be further simplified by integrating the work performed
on the rotations by the twisting moments
Z
{(δϕx mxy ,y + δϕy mxy ,x ) + (δϕx ,y + δϕy ,x ) mxy } dA
Z Z
= δϕx mxy dx + δϕy mxy dy .

Replacing the corresponding terms in the PvD leads to

Z
δWd = {− δw,x (mxx ,x + mxy ,y ) − δw,y (myy ,y + mxy ,x ) + δwp} dA
Z Z
+ δϕx (mxy + mexy ) dx + δϕy (mxy + mexy ) dy
Z Z
+ δϕx (mx + mex ) dy + δϕy (my + mey ) dx = 0 . (22.17)

In contrast to Kirchhoff ’s plate theory the boundary conditions of the bending

and the twisting moments are described by the PvD in a weak sense, if the
rotations are primary variables. Strongly fulfilling the boundary conditions re-
lated to the bending moments with mx + mex = 0 and my + mey = 0 as well
as the twisting moments with mxy + mexy = 0 drops the contour integrals and
yields the final formulation of the PvD
Z
δWd = {−δw,x (mxx ,x + mxy ,y )−δw,y (myy ,y + mxy ,x ) +δwp}dA = 0. (22.18)

This means, that all moments have to be treated as essential boundary conditi-
ons. It is remarkable, that the PvD only employs the moments [ mxx myy mxy ]
as primary variables and the virtual deflection to describe the weak formulation
of the equilibrium as presented by Kirchhoff ’s plate theory. This is valid, since
the shear deformations only modify the kinematic conditions and the material
equations.
22.8 Mixed Quadrilateral Element 389

Conditions of Deformation
The conditions of deformation related to the Reissner–Mindlin plate theory are
weakly formulated applying the Principle of virtual Forces. Incorporating the
bending flexibility FB and the shear flexibility FS and without the boundary
conditions concerning the deflection w, that are strongly fulfilled, it follows
Z
δWσ = + {δmxx (κxx + FBxxxx · mxx + FBxxyy · myy )} dA
Z
+ {δmyy (κyy + FByyxx · mxx + FByyyy · myy )} dA
Z
+ {2 δmxy (κxy + FBxyxy · mxy )} dA
Z Z
− {δqx (γxz − FSxx · qx )} dA − {δqy (γyz − FSyy · qy )} dA
Z
− {δmy (ϕy − ϕey ) + δmxy (ϕx − ϕex )} dx
Z
− {δmx (ϕx − ϕex ) + δmxy (ϕy − ϕey )} dy = 0 .

In contrast to Kirchhoff ’s plate theory, at the boundary the virtual twisting mo-
ments δmxy act on the conjugated rotations ϕx , ϕy , which can be constrained
by external rotations ϕex , ϕey .
Incorporating the governing equations concerning kinematics κxx = ϕx ,x and
κyy = ϕy ,y as well as κxy = (ϕx ,y + ϕy ,x )/2 and after integration by parts
the work performed on the rotations ϕx , ϕy , different contour integrals are
eliminated. It remains the work equation
Z
δWσ = − {δmxx ,x ϕx − δmxx (FBxxxx · mxx + FBxxyy · myy )} dA
Z
− {δmyy ,y ϕy − δmyy (FByyxx · mxx + FByyyy · myy )} dA
Z
− {(δmxy ,y ϕx + δmxy ,x ϕy ) − 2δmxy · FBxyxy · mxy } dA
Z Z
− {δqx (γxz − FSxx · qx )} dA − {δqy (γyz − FSyy · qy )} dA
Z Z
+ {δmy ϕey + δmxy ϕex } dx + {δmx ϕex + δmxy ϕey } dy = 0 .

Applying the equilibrium qx = mxx ,x + mxy ,y and qy = myy ,y + mxy ,x for the
virtual and the actual shear forces as well as kinematics γxz = w,x − ϕx and
390 22 Plate Elements Including Shear Deformations

γyz = w,y − ϕy , the primary variables [ w mxx myy mxy ] persist, which are
used to formulate the PvF in equivalence to the PvD
Z
δWσ = − {δmxx,x w,x − δmxx (FBxxxx · mxx + FBxxyy · myy )} dA
Z
− {δmyy ,y w,y − δmyy (FByyxx · mxx + FByyyy · myy )} dA
Z
− {δmxy ,x w,y + δmxy ,y w,x − 2δmxy · FBxyxy · mxy } dA
Z
+ {(δmxx ,x + δmxy ,y ) · FSxx · (mxx ,x + mxy ,y )

+ (δmyy ,y + δmxy ,x ) · FSyy · (myy ,y + mxy ,x )} dA

Z Z
+ {δmy ϕey + δmxy ϕex } dx + {δmx ϕex + δmxy ϕey } dy = 0 . (22.19)

Boundary Conditions
The work equations (22.18) applying the PvD and (22.19) applying the PvF are
formulated employing the primary variables [ w mxx myy mxy ], whereby the
respective boundary conditions are to be fulfilled strongly by the approach. In
contrast to Kirchhoff ’s plate theory, this means, that the twisting moment along
the boundary is to be set to mxy = −mexy , whereas the boundary conditions
concerning the conjugated rotations ϕx , ϕy are fulfilled weakly, if the twisting
moment can freely evolve.
In the case, where the boundaries do not coincide with the x– and y–coordinates,
the moments have to be transformed to the n– and t–directions of the boun-
dary, if mn or mt are prescribed. This fact is a major drawback, since the
transformation has to be conducted after assembling all elements by means of
the system’s equations.

Approaches to Describe the Primary Variables

At comparing the work equations of the PvD (22.18) and of the PvF (22.19) to
the respective work equations related to Kirchhoff ’s plate theory according to
Section 16.1, it becomes obvious, that the work performed on the shear strains
may be completely described by the bending and twisting moments of the re-
spective integrals in Equation (22.19) indicated by the shear–flexibility FS . This
means, that, by analogy with Kirchhoff ’s plate theory, bi–linear approaches to
describe the primary variables are sufficient, the corresponding discretization
of the work equations may be adopted.
22.8 Mixed Quadrilateral Element 391

Benchmark
As an example to test the formulation the square plate is investigated again
according to Section 6.2.9. Utilizing the symmetry, the quarter plate is investi-
gated in turn for different FE–discretizations, cf. Figure 22-13.
x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
E = 3.0 . 107 kN/m2
T
ly t = 0.1 m
G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 22-13 Square plate – system and loading

By reason of the Reissner–Mindlin theory the twisting moments and the con-
jugated rotations related to the boundaries GE–E as well as GE–G are to be
provided case–dependent. Here the rotation in direction of the boundaries is
set to zero, so that the twisting moment takes finite values at the boundaries,
what is named a strong support, cf. Table 22.8. Therefore, the boundary and
accordingly the symmetry conditions are provided as follows:

Table 22.8 Square plate – boundary conditions, strong support

boundary mxx myy mxy w comment

GE–E ∗ ∗ ∗ 0 clamped
GE–G 0 ∗ ∗ 0 hinged
G–M ∗ ∗ 0 ∗ symmetry
E–M ∗ ∗ 0 ∗ symmetry

Subsequently the behavior of convergence of several state variables is investi-

gated and represented in Tables 22.9, 22.10, and 22.11. In comparison to the
formulations employing displacements as primary variables the displacements
converge slightly inferior, the moments better to some extent.
392 22 Plate Elements Including Shear Deformations

Table 22.9 Square plate incorporating shear deformations – EI/GFQ = 5. 0

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 2. 9861 2. 5548 2. 4373 2. 4088 2. 4017

mxxM [ kN m/m ] 6. 041 4. 077 3. 661 3. 578 3. 557
myyM [ kN m/m ] 5. 802 3. 447 2. 975 2. 865 2. 838
myyE [ kN m/m ] −4. 783 −4. 859 −4. 967 −4. 997 −5. 004
mxyT [ kN m/m ] ÷ −1. 410 −1. 447 −1. 439 −1. 435
qxG [ kN/m ] 1. 600 1. 926 2. 280 2. 512 2. 648
qyE [ kN/m ] 2. 509 3. 151 3. 618 3. 930 4. 031

Table 22.10 Square plate incorporating shear deformations – EI/GFQ = 0. 5

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 1. 1720 1. 0265 0. 9543 0. 9349 0. 9300

mxxM [ kN m/m ] 4. 480 2. 876 2. 500 2. 421 2. 402
myyM [ kN m/m ] 6. 420 3. 999 3. 326 3. 185 3. 150
myyE [ kN m/m ] −5. 750 −6. 382 −6. 585 −6. 639 −6. 653
mxyT [ kN m/m ] ÷ −1. 336 −1. 521 −1. 481 −1. 475
qxG [ kN/m ] 1. 263 1. 572 2. 005 2. 224 2. 356
qyE [ kN/m ] 2. 801 3. 620 4. 177 4. 553 4. 748

Table 22.11 Square plate incorporating shear deformations – EI/GFQ = 0. 05

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 0. 9747 0. 8558 0. 7837 0. 7634 0. 7582

mxxM [ kN m/m ] 4. 200 2. 654 2. 288 2. 207 2. 187
myyM [ kN m/m ] 6. 534 4. 090 3. 363 3. 213 3. 177
myyE [ kN m/m ] −5. 931 −6. 693 −6. 880 −6. 934 −6. 948
mxyT [ kN m/m ] ÷ −1. 380 −1. 616 −1. 538 −1. 531
qxG [ kN/m ] 1. 198 1. 533 1. 953 2. 175 2. 304
qyE [ kN/m ] 2. 851 3. 695 4. 247 4. 656 4. 889
22.9 Hybrid–Mixed Quadrilateral Element 393

22.9 Hybrid–Mixed Quadrilateral Element

The hybrid–mixed formulation concerning the Reissner–Mindlin plate theory
is developed by analogy with the Timoshenko beam theory, whereby the ro-
tational degrees of freedom related to the interfaces are chosen equivalently
to the hybrid–mixed element related to Kirchhoff ’s plate theory according to
Section 16.2. According to Section 22.8, the mixed formulation is chosen in-
corporating the variables [ w mxx myy mxy ] in the element domain, so that
the work equations can be transferred. This type of element is referred to as
RM-HM-4-16.

Equilibrium
By analogy with the mixed formulation the work equation applying the PvD
at element level follows to
Z
δWd = + {(δκxx mxx + δκyy myy + 2δκxy mxy ) − (δγxz qx + δγyz qy ) + δwp} dA
E
Z Z
− {δϕx mxx + δϕy mxy } dy − {δϕy myy + δϕx mxy } dx
E E
Z Z Z Z
+ δwqx dy + δwqy dx − δw qx dy − δw qy dx
E E I I
Z
e e
+ {δϕx (mxxE + mxxI ) + δϕy (mxyE + mxyI )} dy
I
Z
+ {δϕy (myyE + meyyI ) + δϕx (mxyE + mexyI )} dx .
I

The index E indicates the element domain and the index I denotes the inter-
face. In comparison to the mixed formulation, here the cutted cleanly shear
forces and moments perform work on the displacements related to the element
boundary E and to the interface I. Furthermore the external moments me
perform virtual work on the interface I.
By analogy with Section 22.8 the domain as well as the contour integrals may
be transformed introducing
Z Z
{δκxxmxx − δγxz qx }dA − δϕx mxx dy
Z Z
= {−δw,x (mxx ,x + mxy ,y ) + δϕx mxy ,y }dA + δϕx (mxx − mxx ) dy ,

so that only the deflection and the moments [ w mxx myy mxy ] as well as the
virtual deflection δw and the virtual rotations δϕx and δϕy are variables at
394 22 Plate Elements Including Shear Deformations

the element domain. If, moreover, the virtual deflections δw are continuous
at the transition between element boundary and interface, the Principle of
virtual Displacements comprises the following work terms related to the element
domain E and to the interface I:
Z
δWd = {−δw,x (mxx ,x + mxy ,y ) − δw,y (myy ,y + mxy ,x ) + δwp}dA
E
Z
+ {δϕx (mxxE + mexxI ) + δϕy (mxyE + mexyI )} dy
I
Z
+ {δϕy (myyE + meyyI ) + δϕx (mxyE + mexyI )} dx . (22.20)
I
It should be noted that the integrals at the boundary, which deal with the
rotations δϕx , δϕy and the twisting moments mxy , can be transferred with
qx
ϕx = w,x − γxz = w,x − = w,x − ( mxx ,x + mxy ,y )
GAs
and analogously with ϕy . Hereby the variables ϕ perpendicular to the bounda-
ry can be avoided.

Conditions of Deformation
By analogy with the mixed formulation the conditions of deformation are for-
mulated applying the PvF. Here, it has to be taken notice of the fact, that
the transition conditions for the displacements between element boundary and
interface are formulated with the virtual stress resultants. By analogy with
Section 22.8 and without impressed kink angles, that are not possible here, the
following is valid
Z
δWσ = + {δmxx (κxx + FBxxxx · mxx + FBxxyy · myy )} dA
E
Z
+ {δmyy (κyy + FByyxx · mxx + FByyyy · myy )} dA
E
Z
+ {2 δmxy (κxy + FBxyxy · mxy )} dA
E
Z Z
− {δqx (γxz − FSxx · qx )} dA − {δqy (γyz − FSyy · qy )} dA
E E
Z Z
− {δmyy ϕy + δmxy ϕx − δqy w} dx − {δmxx ϕx + δmxy ϕy − δqx w} dy
E E
Z Z
+ {δmyy ϕy + δmxy ϕx − δqy w} dx + {δmxx ϕx + δmxy ϕy − δqx w} dy
I I
= 0.
22.9 Hybrid–Mixed Quadrilateral Element 395

With κxx = ϕx ,x , κyy = ϕy ,y , κxy = (ϕx ,y + ϕy ,x )/2 as well as wE − wI = 0

and performing integration by parts different contour integrals are eliminated.
It remains
Z
δWσ = − {δmxx ,x ϕx − δmxx (FBxxxx · mxx + FBxxyy · myy )} dA
E
Z
− {δmyy ,y ϕy − δmyy (FByyxx · mxx + FByyyy · myy )} dA
E
Z
− {(δmxy ,y ϕx + δmxy ,x ϕy ) − 2δmxy · FBxyxy · mxy } dA
E
Z Z
− {δqx (γxz − FSxx · qx )} dA − {δqy (γyz − FSyy · qy )} dA
E E
Z Z
+ {δmyy ϕy + δmxy ϕx } dx + {δmxx ϕx + δmxy ϕy } dy = 0 .
I I

Moreover, introducing δqx = δmxx ,x + δmxy ,y and δqy = δmyy ,y + δmxy ,x as

well as γxz = w,x − ϕx and γyz = w,y − ϕy yields
Z
δWσ = − {δmxx ,x w,x − δmxx (FBxxxx · mxx + FBxxyy · myy )} dA
E
Z
− {δmyy ,y w,y − δmyy (FByyxx · mxx + FByyyy · myy )} dA
E
Z
− {δmxy ,x w,y + δmxy ,y w,x − 2δmxy · FBxyxy · mxy } dA
E
Z
+ {(δmxx ,x + δmxy ,y ) · FSxx · (mxx ,x + mxy ,y )
E

+ (δmyy ,y + δmxy ,x ) · FSyy · (myy ,y + mxy ,x )} dA

Z Z
+ {δmyy ϕy + δmxy ϕx } dx + {δmxx ϕx + δmxy ϕy } dy = 0. (22.21)
I I

For GAS → ∞ the element matrix is comparable to the matrix related to

Kirchhoff ’s plate theory according to Section 16.2. The results are correspon-
dingly good. Nevertheless, at decreasing shear stiffness, the results deteriorate,
if the shear flexibility approaches FS → ∞. This behavior corresponds to the
numerical trouble that appears at applying the displacement–based beam ele-
ment, since the bending flexibility achieves the range of computational accuracy
concerning the shear flexibility.
396 22 Plate Elements Including Shear Deformations

Boundary Conditions
The work equations (22.20) applying the PvD and (22.21) applying the PvF
employ the variables [ w mxx myy mxy ]E as well as the rotations [ ϕx ϕy ]I .
After discretization has happened related to the element domain, the moments
are eliminated numerically, so that a formulation is derived employing the dis-
placement variables [ w ϕx ϕy ] of the respective element nodes. This means,
that the boundary conditions may be set correspondingly to a formulation em-
ploying only displacements as variables. Nevertheless, it has to be taken into
account, that, due to the Reissner–Mindlin theory, the rotations in direction
of the boundary are independent variables.

Approaches to Describe the Primary Variables

At comparing the work equations applying the PvD (22.20) and the PvF (22.21)
to the respective equations related to Kirchhoff ’s plate theory according to Sec-
tion 16.2, it becomes obvious, that the work performed on the shear deforma-
tions is completely described by the respective integrals in Equation (22.21).
This means, that bi–linear approaches to describe the primary variables are
sufficient, and that the respective discretization of the work equations may be
transferred. The rotations at the interfaces are described by linear shape func-
tions as well, whereby the integrals of the twisting moments are integrated with
a one point integration and those of the bending moments with a two point
integration.

Test
As an example to test the formulation the square plate according to Section
6.2.9 and Section 22.8 is investigated again. Utilizing the symmetry the quarter
plate is investigated in turn for varying FE–discretizations.

x clamped

GE E constantly distributed loading: p = 1 kN/m2

y
E = 3.0 . 107 kN/m2
T
t = 0.1 m
ly

G M ν = 0.2
hinged

l x = 10.0 m
l y = 10.0 m

lx

Fig. 22-14 Square plate – system and loading

22.9 Hybrid–Mixed Quadrilateral Element 397

By reason of the Reissner–Mindlin theory the twisting moments and the con-
jugated rotations related to the boundaries GE–E as well as GE–G are to be
provided case–dependent.
In the case of strong support the boundary conditions are given in Table 22.12,
where the rotations at the hinged boundary GE–G and the clamped boundary
GE–E are set to be zero, so that the twisting moments take finite values at
the hinged boundary. Therefore, the boundary and accordingly the symmetry
conditions are provided as follows:

Table 22.12 Boundary Conditions for strong support – Tables 22.14, 22.15, 22.16

boundary w ϕx ϕy comment
GE–E 0 0 0 clamped
GE–G 0 ∗ 0 hinged
G–M ∗ ∗ 0 symmetry
E–M ∗ 0 ∗ symmetry

In comparison to the strong support, cf. Table 22.12, weak boundary conditions
are investigated, too. Table 22.13 gives the boundary conditions for a weak
support at the hinged boundary.

Table 22.13 Boundary Conditions for weak support – Tables 22.17, 22.18, 22.19

boundary w ϕx ϕy comment
GE–E 0 0 0 clamped
GE–G 0 ∗ ∗ hinged
G–M ∗ ∗ 0 symmetry
E–M ∗ 0 ∗ symmetry

Subsequently the behavior of convergence is illustrated regarding several state

variables, different shear stiffnesses and different support at the hinged boun-
dary.
The comparison of the strong and the weak support regarding different shear
stiffnesses shows, that the influence of the boundary conditions on the entire
state variables is restricted to the vicinity of the support, where the twisting
moments have to converge to zero in the case of a weak support. Nonetheless,
dependent on the shear stiffness, the state variables may differ up to 20 %.
398 22 Plate Elements Including Shear Deformations

Table 22.14 Square plate with strong support – EI/GFQ = 5. 0

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 2. 9409 2. 5448 2. 4346 2. 4081 2. 4015

mxxM [ kN m/m ] 5. 895 4. 122 3. 684 3. 583 3. 559
myyM [ kN m/m ] 5. 497 3. 455 2. 980 2. 866 2. 838
myyE [ kN m/m ] −4. 496 −4. 605 −4. 729 −4. 848 −4. 923
mxyT [ kN m/m ] ÷ −1. 629 −1. 433 −1. 412 −1. 417
÷ −1. 265 −1. 436 −1. 459 −1. 454
qxG [ kN/m ] 1. 390 1. 908 2. 269 2. 509 2. 647
qyE [ kN/m ] 2. 314 3. 102 3. 589 3. 871 4. 021

Table 22.15 Square plate with strong support – EI/GFQ = 0. 5

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 1. 1039 1. 0113 0. 9489 0. 9334 0. 9296

mxxM [ kN m/m ] 4. 199 3. 041 2. 534 2. 429 2. 404
myyM [ kN m/m ] 5. 934 3. 968 3. 334 3. 186 3. 150
myyE [ kN m/m ] −5. 297 −5. 853 −6. 094 −6. 330 −6. 484
mxyT [ kN m/m ] ÷ −1. 419 −1. 429 −1. 445 −1. 455
÷ −1. 179 −1. 453 −1. 487 −1. 487
qxG [ kN/m ] 0. 967 1. 775 2. 002 2. 224 2. 356
qyE [ kN/m ] 2. 655 3. 664 4. 241 4. 578 4. 752

Table 22.16 Square plate with strong support – EI/GFQ = 0. 05

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 0. 8936 0. 8373 0. 7769 0. 7614 0. 7576

mxxM [ kN m/m ] 3. 848 2. 867 2. 321 2. 214 2. 189
myyM [ kN m/m ] 5. 975 4. 055 3. 371 3. 215 3. 177
myyE [ kN m/m ] −5. 421 −6. 061 −6. 314 −6. 575 −6. 748
mxyT [ kN m/m ] ÷ −1. 400 −1. 529 −1. 513 −1. 517
÷ −1. 209 −1. 498 −1. 530 −1. 536
qxG [ kN/m ] 0. 880 1. 779 1. 955 2. 172 2. 304
qyE [ kN/m ] 2. 711 3. 751 4. 344 4. 716 4. 915
22.9 Hybrid–Mixed Quadrilateral Element 399

Table 22.17 Square plate with weak support – EI/GFQ = 5. 0

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 2. 9409 2. 5811 2. 4933 2. 4754 2. 4714

mxxM [ kN m/m ] 5. 895 4. 301 3. 877 3. 792 3. 773
myyM [ kN m/m ] 5. 497 3. 525 3. 064 2. 965 2. 943
myyE [ kN m/m ] −4. 496 −4. 667 −4. 815 −4. 953 −5. 035
mxyT [ kN m/m ] ÷ −1. 375 −0. 604 −0. 263 −0. 121
÷ −1. 069 −0. 840 −0. 553 −0. 291
qxG [ kN/m ] 1. 390 1. 857 2. 258 2. 706 2. 725
qyE [ kN/m ] 2. 314 3. 140 3. 651 3. 939 4. 091

Table 22.18 Square plate with weak support – EI/GFQ = 0. 5

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 1. 1039 1. 0259 0. 9694 0. 9579 0. 9559

mxxM [ kN m/m ] 4. 199 3. 022 2. 544 2. 446 2. 424
myyM [ kN m/m ] 5. 934 3. 977 3. 410 3. 279 3. 250
myyE [ kN m/m ] −5. 297 −5. 906 −6. 209 −6. 468 −6. 633
mxyT [ kN m/m ] ÷ −1. 193 −0. 849 −0. 562 −0. 320
÷ −0. 973 −0. 998 −0. 719 −0. 436
qxG [ kN/m ] 0. 967 1. 759 2. 021 2. 379 2. 635
qyE [ kN/m ] 2. 655 3. 668 4. 302 4. 648 4. 826

Table 22.19 Square plate with weak support – EI/GFQ = 0. 05

mesh 1×1 2×2 4×4 8×8 16 × 16

wM [ cm ] 0. 8937 0. 8474 0. 7850 0. 7688 0. 7655

mxxM [ kN m/m ] 3. 848 2. 814 2. 320 2. 215 2. 190
myyM [ kN m/m ] 5. 975 3. 986 3. 402 3. 246 3. 211
myyE [ kN m/m ] −5. 421 −6. 050 −6. 365 −6. 625 −6. 804
mxyT [ kN m/m ] ÷ −1. 166 −1. 082 −0. 942 −0. 722
÷ −0. 992 −1. 147 −1. 035 −0. 802
qxG [ kN/m ] 0. 880 1. 811 2. 038 2. 294 2. 529
qyE [ kN/m ] 2. 711 3. 708 4. 364 4. 740 4. 941
EVALUATION OF RESULTS
23 Error Estimation

FEM is a method of approximately solving the model equations governing a

specific problem. The approximate solution obtained from the FEM can be im-
proved by a refinement of the discretization. The investigations of convergence
concerning the solutions represented here show that, for plates and membranes,
the approximate solution may converge relatively quickly towards the exact so-
lution. However, this depends on the complexity of the distribution of stresses
throughout the investigated structure as well as on the polynomial order of the
shape functions.

If a quality check of the approximate solution is to be performed on specific

tasks in practice, in general, the problem can be resolved using a mesh twice
as fine as was previously used. The error in discretization can be inferred from
the difference between the two FE–solutions, although a final solution obeying
a specified tolerance limit cannot be deduced in individual cases. In the FE–
context, two solutions may be evaluated and compared to each other. These
can be, for example,
1. two solutions with different mesh sizes, or
2. two solutions with the same mesh size but a different polynomial order.
From the difference in the solutions compared and the order of the error of
the discretizations chosen, the actual error can be deduced. Normally, this
procedure is very costly, hence it is often better to perform an error indication
or an error estimation than a rigorous error analysis.

A much less expensive method is to estimate the error after a one–time analysis
and to report its distribution. In mathematics, many possibilities are known
to estimate or indicate discretization errors. Concerning the FE–analysis two
types may be distinguished, these are the a priori and the a posteriori error
estimations [2, 40, 8, 100]. The benefit of the a priori error estimation is more
theoretical, since it investigates the convergence of various norms with respect
to the element sizes and the shape functions, in order to evaluate how fast an
error will decrease. At applying the FEM the a posteriori error estimation is
more useful, since it gives an evaluation of a specific FE–solution, what could
be the basis for a mesh refinement. In constrast to the error estimation an error
indication does not state the size of the error, but only indicates where an error
occurs what is sufficient in many applications.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_23
404 23 Error Estimation

In general, global and local errors must be distinguished. Normally, global error
indicators are worthless, since locating the error is not possible thereby. Thus,
no information is provided concerning the area where the mesh needs to be
refined. As an example, this is the case if the error concerning the deformation
energy is determined using two different meshes. If the total potential must
fulfill a minimum condition, the strain energy converges towards an absolute
minimum when performing mesh refinement.

Local error indicators work at an element level, so that information about the
parts of the domain where mesh adaptation is required can be immediately
received. Nevertheless, no absolute statements can be predicated upon local
error indicators, as the existing local FE–solution usually also depends on the
solutions at other parts of the domain.
The quality check of the approximate solution performed by a local indicator
yields poor information, if, for example at re–entrant angles of membranes, va-
lues of the solution towards infinity are to be approximated. Here, in general,
a significant local error develops, even though the solution might be approxi-
mated very well in the overall structure.

The literature offers many approaches to develop and to apply an error esti-
mation, what can not be discussed in detail here [2, 3, 6, 12, 40, 8, 100]. Thus
in order to give an introduction of the problem the following options may be
discussed:
3. The least squares method applies the available FE–solution for example
to identify the error in equilibrium.
4. By applying the Principle of virtual Displacements the available FE–
solution can be taken into account to improve stresses at element level,
which can be compared to the stresses from applying the material equa-
tions.

23.1 The Least Squares Method

Concerning the displacement formulation of the FEM, the equilibrium equation
is weakly fulfilled, so that the error should also only be recognised in the equi-
librium. Following the derivation of the PvD and its application to the FEM,
the equilibrium conditions are to be fulfilled in the entire domain as well as
the Neumann boundary conditions are to be fulfilled employing the PvD. The
error cannot be assigned to places where Dirichlet boundary conditions need
to be fulfilled. This makes it clear how an error in an FE–solution could be
quantified.
23.1 The Least Squares Method 405

The development of the error indicator can be discussed by means of the sim-
plifying structure of Section 1.3, where a bar is loaded by a distributed external
action p and has to fulfill the Neumann boundary condition with respect to H.
Choosing a formulation of the error geared to least squares, this yields,
Z
E=ℓ [N,x + p ]2 dx + [ N − H ]2boundary .
ℓ

This corresponds to the global error of the FE–solution, which, however, has low
significance, as shown above. Alternatively, the integral related to the domain
can be defined as the addition of the errors related to the element domains and
element intersections, respectively.
∆ℓ−ε
Z Z+ε
E = Σel ∆ℓ [N,x + p ] dx + Σintersect. ∆ℓ [N,x + p ]2 dx + [ N − H ]2boundary .
2

ε −ε

Regarding ε → 0 one can simplify the second integral, since the axial external
action is not defined at the jump discontinuity. Apart from that, the derivatives
of the jumps ∆N at the element intersections can be described mathematically
exactly employing the Dirac–Delta function δ(x), see Figure 23-1.
N

x
Δl Δl Δl Δl
N,x

Δ N i · δ (x) <0

Fig. 23-1 Derivatives at jump discontinuities

This yields
∆ℓ−ε
Z Z+ε
E = Σel ∆ℓ [N,x + p ] dx + Σintersect. δ(x)[ ∆N ]2 dx + [ N − H ]2boundary .
2

ε −ε
406 23 Error Estimation

δ(x) Z+ε
δ(x) dx = 1
−ε
-ε x 0 +ε x

Fig. 23-2 Dirac–Delta Funktion

Applying the definition of the Dirac–Delta function δ(x), see Figure 23-2,
further gives
∆ℓ−ε
Z
E = Σel ∆ℓ [N,x + p ]2 dx + Σintersect. ∆N 2 + [ N − H ]2boundary .
ε

Regarding the single element domains, or intersections, one may obtain the
amount of the local error. Thus the local error can be directly recognised from
the stress jumps at both the respective element intersections as well as from the
integral of the error of the differential equation regarding the element domain.
Simplification gives

Ei = [N−i − N+i ]2intersection .

The procedure may be extended to two or three dimensions, therefore the local
error can also be indicated for membrane elements. Regarding the element mesh
of a membrane given in Figure 23-3, jumps occur wih respect to σyy and σyx
related to the T–T line and with respect to σxx and σxy related to the V–V
line, which must disappear in the case of the exact solution.

σxy

M
T T
σyy σyx

σxx

Fig. 23-3 Stress jumps in a membrane

23.2 Error Estimation by Applying the Principle of Virtual Work 407

However one should take into account, that jumps of normal and tangenti-
al stresses can emerge at every element boundary, which must be considered
separately. If the least squares method is separately applied to the equilibri-
um in the x–and in the y–direction, the points may be recognised, where the
stress jumps indicate an error. Regarding the x–direction, without considering
a system boundary, this yields
∆A−ε
Z
Ex = Σel ∆ℓx ∆ℓy [σxx ,x + σyx ,y + px ]2 dxdy
ε
Z Z
2 2
+ Σintersection ∆ℓx ∆σxx dy + Σintersection ∆ℓy ∆σyx dx ,
y x

and regarding the y-direction it follows

∆A−ε
Z
Ey = Σel ∆ℓx ∆ℓy [σyy ,y + σxy ,x + py ]2 dxdy
ε
Z Z
2 2
+ Σintersection ∆ℓx ∆σxy dy + Σintersection ∆ℓy ∆σyy dx .
y x

It should be remarked that the jumps in x–direction can be used regarding

the refinement in the x–direction and those in the y–direction regarding the
refinement in the y–direction. A similar procedure can be applied by employing
isoparametric elements, if the stress components are known at every intersection
in both the x–and y–directions.

23.2 Error Estimation by Applying the Principle of Virtual Work

Another error estimation method where the neighbouring elements do not need
to be considered is inherent, when FEM is developed by applying the PvD. In
structural analysis, the PvD is applied to compute the internal force variables
of frame structures and has provided excellent results so far. The transfer of
this idea to the element level makes it possible to compute more exact internal
force variables, e. g. regarding a bar, NA and NB at the element intersections.
A comparison of these more exact values of the internal force variables and the
values from stress analysis may yield a measurement of the error.

Concerning bars loaded by dead weight, the linear approach is accompanied by

a step function for N (x) with constant longitudinal forces NF EM element–by–
element, as shown in Figure 23-4.
408 23 Error Estimation

entire system
pX

1 2 3 4 5
x, u

NFEM

1 2 3 4 5

single element
pX
N A B N N N
A B A FEM
= B FEM
= NFEM
i i+1

δuA 1

1 δ uB

Fig. 23-4 FE–approximation regarding the bar loaded by constant px

By applying the PvD, the internal force variables NA and NB may be directly
computed at the element intersections. In accordance with Equations (2.11)
and (2.12), it holds that
Z
{N δu,x − p δu} dx + NA δuA − NB δuB = 0 .

Employing linear shape functions corresponding to Equation (2.28), yields

" # " #   " #
EA/ℓ −EA/ℓ uA pℓ/2 NA
− δWel. = [ δuA δuB ] · { · − + }.
−EA/ℓ EA/ℓ uB pℓ/2 −NB

Introducing the well–known matrix notation as well as the vector N related to

the internal force variables gives

− δWel. = δvT {K · v − f + N} = 0 . (23.1)

The product K · v corresponds here to the longitudinal force ± NF EM from the

23.2 Error Estimation by Applying the Principle of Virtual Work 409

stress analysis. Thus the internal force variables can be computed directly to

NA = NF EM + pℓ/2 ,
NB = NF EM − pℓ/2 .

Hence, the local error is given as the difference between the internal force
variables NA and NB and the longitudinal forces NF EM from stress analysis,

FA = NA − NF EM = + pℓ/2 ,
FB = NB − NF EM = − pℓ/2 .

It should be noted that the error evaluated by this procedure is not an absolute
error but a relative one, as the FE–approximation applied at the element–level
may be totally inaccurate.

Transfering Equation (23.1) to the membrane elements, the nodal forces with
respect to the x–and y–directions are directly obtained, and can be compared
to the stresses from applying the material equations. Analogously to the PvD,
it makes sense to integrate the stresses at the element intersections, taken from
stress analysis, to get the nodal forces.
Comparing the nodal forces from the stress analysis and from applying the
PvD, the following should be taken care of. If the equilibrium is separately
fulfilled for every node, see Figure 23-3, the error obtained from the stresses
σxx and σyy as well as σxy is more or less the same but with opposite signs, so
they cancel out each other with regard to the entire system. Therefore it seems
to be reasonable to divide the nodal forces into parts related to σxx , σyy and
σxy , and then to separately introduce the respective parts into the comparison.
This procedure may lead to an even better estimation of the local error.
410 23 Error Estimation

23.3 Mesh–Adaptation
Applying the error estimation to a FE-solution offers the opportunity to impro-
ve the solution by means of an adaptation of the FE-mesh. In general, a new
discretization can be performed introducing locally distinctive and solution–
matching mesh refinement respectively. This may considerably reduce the com-
putational cost. Modern computer–program systems allow for the indication of
errors and the making of suggestions for a re–discretization related to the error
distribution inside the structure. This procedure, named as mesh–adaptation,
may be performed by applying different strategies, changing the element sizes
by means of the h–adaptation or raising the polynomial order of the approach
by means of the p–adaptation in specified parts of the mesh without breaching
the conformity criterion with respect to the shape functions.

Regarding different cases, the strong solutions of the governing equations exhi-
bit singularities and high gradients with respect to the distribution of stresses,
which effect large stress discontinuities between neighboring elements and, as
a result, local error concentrations.
Singularities are, for example, jumps in the stress distribution, which may be
easily described by applying the PvD, if they occur at the element intersecti-
ons. Singularities also occur as non–real points of infinity inside the distribu-
tion of stresses, for example at nooks or due to concentrated external actions
regarding membranes, plates, and shell structures. In such cases, a proper des-
cription of the stress distribution around the points of infinity should be taken
care of. However, one should not try to directly approximate points of infinity,
as this neither makes sense regarding physics, nor is required when following
St.Venant’s principle.
Nonetheless, since singularities with high stress gradients are part of the model-
ling of the structural behavior, a proper finite element approach in the vicinity
of the singularities is necessary in order to minimize the discretization error. In
general this is possible with a refinement of the finite element mesh or with an
extension of the order of the shape functions. Thus it depends on the elements,
which are applied to solve the problem.

The property of the FEM to improve the solution by means of mesh–refinement

or mesh–adaptation is shown with regard to a 2–dimensional membrane struc-
ture, where singularities occur at the corners of the opening, see Figure 23-5. By
means of this simplifying example a general approach is represented to refine
the mesh with respect to triangular elements.
23.3 Mesh–Adaptation 411

Though a discretization is allowed, employing identically–sized elements and

using the same order of the approach for all of them, what needs less effort with
regard to data processing, this might not be optimal inherently. To approach
the deformation with equal accuracy in the whole structure and thereby also
the stresses, the mesh has to be refined only in the region of the opening due to
the heavy stress gradients expected there. If the same type of element shall be
employed all over the structure, triangular elements would be a good choice,
since they offer the possibility to refine a mesh at desired areas without any
disadvantages. Dealing with triangles Figure 23-5 compares an overall mesh
refinement to a local one. This type of mesh refinement is named h–adaptation,
because the size h of the elements is adapted to regions with large gradients of
the stress state.

Fig. 23-5 Mesh refinement employing triangular elements

When applying the h–method the adaptation of the mesh deals with the ele-
ment size as explained above. Hereby, the mesh is refined by means of very
small elements at the regions where high stress gradients are observed. Gene-
rally these regions need a high number of elements to aproximate the solution
sufficiently, since shape functions of low order are employed.

In general elements of a high polynomial order lead to better results than

elements of lower order, even if the number of unknowns is the same. Applying
the p–method to adapt the mesh the order of the shape functions has to be
specifically increased where high stress gradients occur, while leaving the mesh
unchanged.
412 23 Error Estimation

Both methods to adapt the mesh require special modifications of the elements
at the intersections. At employing Lagrange Polynomials of different order,
the additional DOF of the higher order approach must be attached to the
lower order approach at the respective intersections, as they are no longer
independent DOF. Thus additional nodes are required regarding the element
employing the lower order approach. They are denoted as hanging nodes, see
Figure 23-6 in the case of rectangular elements. Hanging nodes are also needed
in the case of mesh refinement, concerning elements with the same polynomial
order but with discontinuing mesh lines, see Figure 23-6. Here the displacements
related to the additional nodes of the smaller element must be attached to the
displacements related to the respective edge of the larger element.

Fig. 23-6 Mesh refinement employing hanging nodes

The h–p–method applies both adaptation procedures and is occasionally im-

plemented in commercial FE–program systems. It requires a relatively time–
consuming algorithm, since different criteria must be applied in order to adapt
both the mesh and the polynomial order.

Applying quadrilateral isoparametric elements, which are explained in Section

12 in detail, mesh adaptation techniques are possible as well, but effect comple-
tely irregular meshes. Here, efficient mesh generators have to be applied, which
are able to check the elements’ geometry with respect to the mesh refinement
at desired regions of the structure, because trapezoidal elements or even single
triangular elements may occur. Since the physical properties of the elements
usually change with respect to geometrical irregularities, undesired and unphy-
sical solutions could be the result and may be reason of larger discretization
errors, cf. Section 24.
24 Quality of Elements

The quality of an element can be evaluated with respect to different properties.

In Section 2.4 the fundamental convergence criteria have been discussed with
respect to the choice of the shape function. Nonetheless, the applications of the
criteria to the different physical problems have shown, that these criteria are
not able to avoid unphysical oscillations or locking phenomena in some cases, cf.
Sections 5.2 and 21 as examples. Further problems may arise, if isoparametric
elements have to be applied as investigated in Section 14.8. Thus several proce-
dures have been developed in order to evaluate the properties and the behavior
of elements with respect to mesh distortions. The eigenvalue analysis of the
stiffness matrix of a single element is a mathematical tool to control the ability
to describe rigid body motions. The patch test deals with an investigation of the
element behavior as part of a distorted mesh. Furthermore locking phenomena
have to be investigated, because they may affect convergence against a wrong
solution or may affect stiffnesses to tend to infinity.

24.1 The Eigenvalue Analysis

The eigenvalue analysis of the stiffness matrix is a mathematical tool to inve-
stigate the ability of an element to describe the strain fields with respect to
the chosen shape functions and the nodal degrees of freedom. Especially the
rigid body motions can be identified exactly. Taking into account the stiffness
matrix K of an element and the identity matrix I, the eigenvalue problem

( K − λi · I ) φi = 0

yields the eigenvalues λi and the eigenvectors φi . Eigenvalues and eigenvectors

are correlated to each other. If the size of the stiffness matrix is n × n exactly n
eigenvalues and n eigenvectors exist. In the mathematical sense the eigenvalue
problem describes a diagonal transformation of the stiffness matrix, whereat
the eigenvalues λi are a measure of the stiffness of the element with respect
to the deformation field described by the respective eigenvectors φi . Since the
eigenvalues are related to the zero values of the determinant of the coefficient
matrix, an iteration procedure is needed to compute eigenvalues and eigenvec-
tors, which sometimes may effect rounding errors of the eigenvalues and the
eigenvectors. The following results are computed by means of a simultanous
vector iteration.

© The Author(s), under exclusive license to

Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9_24
414 24 Quality of Elements

As an example the regular deformation modes of a membrane element with

bi–linear shape functions and 8 DOF are represented in Figure 24-1. The de-
formation modes 1, 2, 3 are correlated to the three rigid body modes and
the other five deformation modes are activating the stiffness of the element.
Although the rigid body modes can be described by all elements, which are in-
vestigated in the following, the non–zero eigenvalues might be quite diffferent
with respect to the elements.
1 2 3 4

5 6 7 8

Fig. 24-1 Regular deformation modes of a square membrane element

A comparison of the eigenvalues of the different quadrilateral membrane ele-

ments described in this textbook is represented in Table 24.1 for a square
membrane element and in Table 24.2 for an element comprising ℓx /ℓy = 1000.

l x = 1.0 m, l y = 1.0 m, E = 1.0 N/m 2 , ν = 0.0

ly

0.001.l y

lx lx

Fig. 24-2 Rectangular membrane elements for the eigenvalue analysis

Table 24.1 represents the eigenvalues of the stiffness matrix of the square ele-
ment in Figure 24-2. Some of the eigenvalues occur as pairs, since the related
deformation modes activate same stiffnesses with respect to the x– and the
y–coordinate. Regarding the elements P-DQ-4, P-DQ-4-SRI, P-HMQC-4-5
dealing with bi–linear shape functions the eigenvalues 6, 7, 8 describe the lon-
gitudinal and the shear stiffnesses. The eigenvalues 4, 5 are related to the
trapezoidal deformation modes, whereat the element P-DQ-4 is stiffer due to
the unphysical shear stiffnesses as described in Section 5.2.
24.1 The Eigenvalue Analysis 415

The elements P-DQ-9, P-DQ-9-SRI, P-HMQ-8-13 are dealing with quadratic

shape functions. Thus the number of eigenvalues increases and the eigenvectors
describe partly more complex deformation modes activating smaller and higher
stiffnesses. Parts of the deformation modes of the three elements are equal,
but some of them are different with respect to the reduced integration or the
balanced shape functions of the hybrid–mixed element.

Table 24.1 Eigenvalues of membrane elements ℓx /ℓy = 1

no. P-DQ-4 P-DQ-4-SRI P-HMQC-4-5 P-DQ-9 P-DQ-9-SRI P-HMQ-8-13

1. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
2. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
3. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
4. 0. 5 0. 3333 0. 3333 0. 16075 0. 16075 0. 16075
5. 0. 5 0. 3333 0. 3333 0. 28083 0. 27423 0. 16075
6. 1. 0 1. 0 1. 0 0. 28083 0. 27423 0. 29919
7. 1. 0 1. 0 1. 0 0. 49622 0. 47948 0. 29919
8. 1. 0 1. 0 1. 0 0. 51049 0. 50320 0. 47948
9. − − − 0. 68035 0. 50318 1. 1059
10. − − − 0. 68035 0. 51047 1. 1059
11. − − − 1. 1059 1. 1059 1. 3333
12. − − − 1. 4667 1. 3333 1. 8004
13. − − − 1. 8914 1. 8561 1. 8004
14. − − − 1. 8914 1. 8568 1. 8538
15. − − − 1. 9707 1. 8566 4. 4004
16. − − − 2. 0892 2. 0895 4. 4004
17. − − − 5. 2474 4. 9314 −
18. − − − 5. 2474 4. 9314 −

Due to the comparison of the eigenvalues, it has to be noted, that the bi–
linear deformation fields of the elements P-DQ-4, P-DQ-4-SRI, P-HMQC-4-5
are not represented by single modes of the elements P-DQ-9, P-DQ-9-SRI,
P-HMQ-8-13, but can be represented by a combination of modes. Thus the
corresponding eigenvalues do not coincide in all cases. For example the eigen-
value λ8 = 1. 0 of the bi–linear elements is correlated to the shear deformation
φ8 of Figure 24-1. The correspondig eigenvalue of the element P-HMQ-8-13
dealing with quadratic shape functions is λ8 = 0. 47948. This eigenvalue is cor-
related to φ8 superposed by a quadratic deformation field, which reduces the
416 24 Quality of Elements

shear–constraints at the corners of the deformation field. On the other side the
eigenvalue λ14 = 1. 8538 is correlated to φ8 as well, but is superposed by ano-
ther quadratic deformation field, which increases the constraints at the corners.

Table 24.2 represents the eigenvalues of the rectangular element in Figure 24-2,
whereby the lengths of the edges have a relation of ℓx /ℓy = 1000. Such elements
are applied, if beams are investigated without taking into account the beam
theory, or in the three dimensional case, if thin plates have to be investigated
as a three dimensional continuum.

Table 24.2 Eigenvalues of membrane elements ℓx /ℓy = 1000

no. P-DQ-4 P-DQ-4-SRI P-HMQC-4-5 P-DQ-9 P-DQ-9-SRI P-HMQ-8-13

1. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
2. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
3. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
4. 0. 001 0. 00033 0. 00033 2. 66 · 10−9 2. 66 · 10−9 3. 55 · 10−9
5. 166. 6 0. 0010 0. 0010 0. 00066 0. 00050 0. 00016075
6. 333. 3 333. 3 333. 3 0. 00266 0. 00066 0. 00088
7. 500. 0 500. 0 500. 0 80. 377 0. 00088 0. 00110
8. 1000. 0 1000. 0 1000. 0 160. 75 0. 00376 0. 00355
9. − − − 166. 67 160. 75 160. 75
10. − − − 321. 50 166. 67 166. 67
11. − − − 333. 33 333. 33 333. 33
12. − − − 552. 96 500. 00 500. 00
13. − − − 643. 01 643. 01 1105. 9
14. − − − 666. 67 666. 67 1333. 3
15. − − − 1105. 9 1105. 9 2000. 0
16. − − − 1333. 3 1333. 3 4000. 0
17. − − − 2211. 8 2000. 0 −
18. − − − 4423. 7 4423. 7 −

In this case the eigenvalues λi = 0. 001 and λi = 1000. 0 of the bi–linear appro-
ximation describe the stiffnesses in the directions of the x– and y–coordinates,
which are correlated to ℓy /ℓx and ℓx /ℓy respectively. The eigenvalue λi = 500. 0
describes the shear stiffness, and λi = 333. 3 as well as λi = 0. 00033 characte-
rize the trapezoidal modes in both directions.
24.2 The Locking Phenomena 417

As in the case of the square element bending is described only by means of

the trapezoidal modes, whereat the eigenvalue λ5 = 166. 6 of the trapezoidal
mode of the P-DQ-4 element is extremely large and thus indicates a locking
phenomenon of this element, cf. Section 24.2.
Bending is described much better by means of a quadratic approximation of the
displacement fields, because it is able to describe the curvature a priori thereby.
Assuming a quadratic displacement course of v(x, y) = ( ξ −ξ 2 )·4. 0·v|0.5 along
the x–coordinate, the virtual work of a beam regarding this deformation mode
yields
Z
−δW = δv,xx EIv,xx dx
Z
= δv0.5 2. 0 · 4. 0/(ℓx)2 · ( Eℓ3y /12 ) · 2. 0 · 4. 0/(ℓx)2 dx · v0.5

= δv0.5 · (5. 33 · 10−9 ) · v0.5 ,

what is correlated to the first non–zero eigenvalue of the elements employing

bi–quadratic shape functions. The element P-HMQ-8-13 is a little stiffer, since
the approach deals with incomplete bi–quadratic shape functions. The element
P-DQ-9 includes locking phenonomena with respect to the 7th eigenvalue.

24.2 The Locking Phenomena

In general the shape functions have to satisfy the convergence criteria in order
to ensure the convergence against the exact solution of the governing equations.
Nonetheless, applying these criteria to the choice of the polynomial order of the
approach and the correlated nodal degrees of freedom can effect deficiencies of
the element properties, which may cause unphysical solutions due to distortions
of the stiffness matrix. Since some of these problems can lead to artificial stiff-
nesses, which prevent the proper deformation of the element, these phenomena
are named locking phenomena. The literature [9, 68, 106] offers different types
of locking, which can be summarized as follows.

Locking due to incompatible kinematic approaches

Section 21 deals with the development of an element for the Timoshenko beam
theory. Applying linear shape functions to the deflection w and to the rotation
ϕ leads to an incompatible description of kinematics. Applying linear shape
functions for the deflection and the rotation, the shear strain

γ = w,x − ϕ
418 24 Quality of Elements

comprises a constant as well as a linear part, what effects unphysical stiffnes-

ses and wrong solutions. Assuming a constant rotation ϕ avoids the problem
and yields a proper approach of the deformation fields. Analogously Reissner–
Mindlin plates may be treated, cf. Section 22.
A similar problem arises, if curved structures are taken into account. The mem-
brane strain of a curved beam is defined with
ε = u,s + w/R ,
whereat s describes the curved coordinate along the arc and R describes the
radius of curvature. Applying a linear approach for both displacements u(s) and
w(s) the strain comprises a constant and a linear part, what effects unphysical
oscillations of the solution. Assuming a constant w/R avoids these problems
and provides accurate solutions. The same problems can be avoided in the case
of shell structures.
Another incompatibility arises with the definition of shear strains of membra-
nes. Applying linear shape functions the shear strain
1
εxy = · ( u,y + v,x )
2
comprises constant and linear parts in both directions what leads to unphy-
sical oscillations, cf. Section 5.1. Assuming constant shear strains avoids the
oscillations and provides accurate solutions, cf. Section 5.2. This phenomenon
may be treated by means of the selectively reduced integration technique, which
improves the element properties remarkably. Although the selectively reduced
integration helps to avoid unphysical oscillations, it is not sufficient in all cases
as discussed by means of the eigenvalues of the stiffness matrices.

Locking due to non–regular element geometries

Considering two or three dimensional problems elements of square or cubic
geometry usually provide a sufficient convergence behavior. If very thin struc-
tures have to be investigated the lengths of the element edges usually become
very different, what influences the properties of the stiffness matrices in an un-
desired manner. Here, with regard to the element coordinates, the stiffnesses
become extremely different and may block some of the deformation fields, what
is indicated with the eigenvalue 5 of the P-DQ-4 element and the eigenvalues
7 and 10 of the P-DQ-9 element, cf. Table 24.2. Thus the created phenomenon
is referred to as thickness locking.

Another problem occurs, if the element’s geometry is no longer rectangular,

but has got an arbitrary quadrilateral shape. As an example, Figure 24-3 repre-
sents trapezoidal membrane elements of different thicknesses. In these cases the
24.2 The Locking Phenomena 419

stiffnesses vary inside the elements and thus strongly influence their properties
with respect to the deformation modes. This may be discussed by means of
the eigenvalues of the case ℓx /ℓy = 1 given in Table 24.3 and those of the case
ℓx /ℓy = 1000 given in Table 24.4.

e e e = lx / 4
l x = 1.0 m, l y = 1.0 m, E = 1.0 N/m 2 , ν = 0.0

ly e e

0.001.l y

e lx e e lx e

Fig. 24-3 Trapezoidal membrane elements for the eigenvalue analysis

Table 24.3 indicates, that the trapezoidal shape only effects a moderate varia-
tion of the eigenvalues in comparison to a square. However, the deviations from
the accurate values may be caused by additional stiffnesses, which are reason
of undesired stresses with respect to the deformation modes.

Table 24.3 Eigenvalues of membrane elements ℓx /ℓy = 1 and ℓx1 /ℓx2 = 3

no. P-DQ-4 P-DQ-4-SRI P-HMQC-4-5 P-DQ-9 P-DQ-9-SRI P-HMQ-8-13

1. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
2. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
3. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
4. 0. 43882 0. 33384 0. 32071 0. 09829 0. 13487 0. 09787
5. 0. 53514 0. 33357 0. 35233 0. 24789 0. 24251 0. 18985
6. 1. 0 1. 0 1. 0 0. 39631 0. 33787 0. 25247
7. 1. 1467 1. 0 1. 0643 0. 44791 0. 40286 0. 40585
8. 1. 2430 1. 1811 1. 1338 0. 42820 0. 50320 0. 55484
9. − − − 0. 66544 0. 50905 0. 72046
10. − − − 0. 77285 0. 64020 1. 1779
11. − − − 1. 3130 1. 1095 1. 2329
12. − − − 1. 6342 1. 5357 1. 7388
13. − − − 1. 9943 1. 7454 2. 1470
14. − − − 2. 0481 1. 8963 2. 5822
15. − − − 2. 2380 2. 1690 4. 8891
16. − − − 2. 6966 2. 6115 5. 1498
17. − − − 5. 8556 5. 4818 −
18. − − − 5. 8389 5. 5479 −
420 24 Quality of Elements

But, if the element is very thin in one direction, the eigenvalues change much
more and indicate the thickness locking for the elements P-DQ-4, P-DQ-4-SRI
with respect to the eigenvalues 4 and 5, for the element P-DQ-9 with respect
to the eigenvalues 8, 9 ,11, and for the element P-DQ-9-SRI with respect to
the eigenvalues 6, 7 and 8, cf. Table 24.4. Thus these elements are not able to
accurately describe the deformation and may lock the deformation modes, if
they are effected by external loads.
The only elements, which give reasonable results, are the P-HMQC-4-5 element
and the P-HMQ-8-13 element, whereat the linear approach is hardly able to
describe bending.

Table 24.4 Eigenvalues of membrane elements ℓx /ℓy = 1000 and ℓx1 /ℓx2 = 3

no. P-DQ-4 P-DQ-4-SRI P-HMQC-4-5 P-DQ-9 P-DQ-9-SRI P-HMQ-8-13

1. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
2. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
3. 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0
4. 0. 008 13. 363 0. 00032 7. 18 · 10−10 6. 61 · 10−6 6. 17 · 10−10
5. 227. 27 47. 244 0. 00113 0. 00057 0. 00057 0. 00026
6. 454. 54 333. 3 416. 67 0. 00098 18. 422 0. 00069
7. 500. 0 500. 0 500. 0 207. 71 58. 347 0. 00080
8. 1000. 0 1000. 0 1000. 0 256. 08 98. 993 0. 00310
9. − − − 277. 57 171. 27 319. 44
10. − − − 415. 41 207. 71 410. 75
11. − − − 512. 15 410. 75 431. 27
12. − − − 555. 14 546. 65 589. 64
13. − − − 697. 52 649. 39 1187. 5
14. − − − 737. 70 697. 52 1377. 1
15. − − − 1395. 0 1339. 2 2152. 8
16. − − − 1475. 4 1377. 1 4399. 4
17. − − − 2413. 0 2192. 5 −
18. − − − 4825. 9 4761. 0 −
24.3 Improvement of Elements Suffering from Locking 421

Locking due to incompressibility

The numerical investigation of 3–dimensional continua needs volume elements
and the respective theory of elasticity. Since the elastic material behavior is
described by means of the elasticity matrix
E
E= · [. . .]
( 1 + ν ) · ( 1 − 2ν )
an artificial stiffening occurs, if incompressible materials are taken into account.
Here, Poisson’s ratio ν → 0. 5, what means, that the multiplier tends to in-
finity, whereat the stiffness matrix locks any deformation. This phenomenon,
describing the incompressibility, is referred to as Poisson locking and has to be
taken into account with respect to 3–dimensional structures, plain strain situa-
tions and plastic volume constant deformation. An introduction to the problem
is published by Hughes [47].

24.3 Improvement of Elements Suffering from Locking

In general the quality of finite elements depends on the shape functions, that
are chosen to describe the primary variables. In many cases this may lead to
locking phenomena as described in Section 24.2. Locking is a limiting property,
which can be identified by an eigenvalue analysis, a patch test or convergence
studies with respect to those parameters, that govern the phenomenon. To
avoid the numerical problems induced by locking different methods have been
developed, which change the properties of the elements. The following methods
have been developed in order to impove the quality of low order elements and
to offer numerically efficient formulations.

Assumed Natural Strain Method

Very early the basis of the Assumed Natural Strain Method (ANS) has been
developed and published by MacNeal [69, 70] and Hughes et al. [45, 48]. Later
on Bathe et al. have generalized the method, developed a family of elements
called Mixed Interpolation of Tensorial Components (MITC) and published
quadrilateral and triangular elements as MITC-4, MITC-7, . . . [13, 14, 23, 24].
Park et al. [78, 79] discuss the ANS method with respect to C0 –conform shell
elements.
The general idea of the method deals with the assumption, that the strain
fields have to satisfy certain kinematic relations a priori, what means, that
the independence of the strain fields has to be restricted following special as-
sumptions with respect to the displacement approach. This means for example,
422 24 Quality of Elements

that in the case of Reissner–Mindlin plates the conditions γxz = ϕx − w,x and
γyz = ϕy − w,y have to be satisfied without numerical oscillations. Instead of a
selectively reduced integration this demand can be achieved employing shape
functions of different order for the description of the state variables w, ϕx , ϕy .
This idea is applied with a similar intention at Section 22.7 and 22.9, but
without a rigorous mathematical development as presented in [24].
As an example the shear strain γxz may be described in a quadrilateral element
by means of bi–linear shape functions for w(x, y) and ϕx (x, y) with
γxz (x, y) = w(x, y),x − ϕx (x, y)
X X
= φi (x, y),x · wi − φi (x, y) · ϕx i .
Thus the shear strain comprises a constant part and a linear part in x–direction,
what effects the shear locking phenomenon. Assuming a constant shear strain
with respect to the x–coordinate, it follows
X
γxz (y) = φ̄i (y) · γxz i ,

whereat φ̄i (y) are scaled to the nodal strains γxz i . Figure 24-4 illustrates the
course of the shear strain γxz (y), that is constant in x–direction but linear
in y–direction, and normalized with respect to the shear strains γ(xz)E and
γ(xz)G . The nodes E, G are defined as those nodes, where the shear strains
related to the bi–linear shape functions of the primary variables would become
the best value, which usually are the Gaussian supporting points. In the case
of bi–linear shape functions these are the midpoints of the related edges. Thus
the shear strains γxz i can be described with the nodal degrees of freedom of
the primary variables w, ϕx , ϕy of the corresponding element edge.

x γxzE
y z A A

E E

D B D B

G
γxzG G

C C

Fig. 24-4 Course φ̄E (y), φ̄G (y) of the assumed shear strain γxz

The procedure can be applied to higher order polynomials as well, if additional

degrees of freedom with respect to the shear strains are defined at each element
edge, and may be analogously developed for other locking phenomena.
24.3 Improvement of Elements Suffering from Locking 423

Enhanced Assumed Strain Method

A complete different approach has been developed by Simo and Rifai [92] and is
named Enhanced Assumed Strain Method (EAS). Instead of a restriction of the
strain fields or an extension of the approach to describe of the displacements,
the method deals with an extension of the strain fields by additional internal
variables, which modify the strain fields with respect to the undesired numerical
parts. After the integration of the stiffness matrix the internal variables have
to be eliminated at the element level as shown in Section 22.5 for a hybrid–
displacement element or in Section 22.9 for a hybrid–mixed element.
Taking into account the strains of membranes discretized with bi–linear shape
functions
X
εxx = { φi (ξ, η · ui },x
X
εyy = { φi (ξ, η) · vi },x
X X
2εxx = { φi (ξ, η) · vi },x + { φi (ξ, η) · ui },y

the element behavior shows undesired strain distributions as explained in Sec-

tion 5.1 and 5.2. In order to avoid the oscillations of the in–plane shear strains
the strains may be extended by functions, which are able to cancel the undesi-
red oscillations out. This is possible, if additional internal degrees of freedom βi
deal as a correction of the original approach. Thus the extensions of the strains
given by
X
εxx = { φi (ξ, η) · ui },x + ξ · β1
X
εyy = { φi (ξ, η) · vi },y + η · β2
X X
2εxx = { φi (ξ, η) · vi },x + { φi (ξ, η) · ui },y + ξ · β3 + η · β4

are able to eliminate the defiencies of the original approach and lead to similar
results as the selectively reduced intergration, but can be applied to other
locking phenomena, too.

Discrete Shear Strain Gap Method

A third remarkable idea has been developed by Bletzinger et al. [19], has been
refined by Koschnick et al. [56] and is referred to as Discrete Shear Gap me-
thod (DSG). However, it can be applied to other strain fields, too. It may be
explained as follows: The Timoshenko beam theory defines the shear strains as

γ = w,x − ϕ .
424 24 Quality of Elements

Employing linear shape functions φi for w and ϕ results into the problems
discussed in Section 21, due to the different orders of w,x and ϕ. Thus seve-
ral approaches may be applied to this problem, e. g. the selectively reduced
integration technique, the ANS–method, or the EAS–method.
The DSG–method deals with the integration of w(x),x concerning kinematics
Z
w(x) = (γ + ϕ) dx = wγ + wϕ .

Thereby wγ describes the difference between the total deflection w(x) and the
deflection wϕ due to bending and thus it is named shear strain gap.
In order to compute the element stiffness matrix an assumption has to be made
with respect to the course of the shear strains within the element. However,
the accumulated shear strain gap between the element nodes A and B can be
computed to
Z B
B
wγ A = (w,x − ϕ) dx .
A

Since the integral only depends on the values at the boundaries, employing
linear shape functions it follows
wB − wA
wγ B
A = { − (ϕB − ϕA )} · ℓ ,
ℓ
whereat both parts of wγ B
A have got the same polynomial order. Thus the course
of the shear strain can be described by means of a constant shape function
φγ = 1/ℓ with

γ = φγ · wγ B
A

and is applied to the integration of the element stiffness matrix.

The application of the method leads to a stiffness matrix, where parts of the
components have the same numbers as the other approaches. However, the
method can be applied with great success to other locking phenomena, too.
24.4 The Patch–Test 425

24.4 The Patch–Test

The patch test is an engineering tool to investigate the ability of an element to
describe constant strain fields. This means, that the test is more or less related
to the first convergence criterion described in Section 3. The test is developed by
Irons [53] in order to check an element with respect to deficiencies. MacNeal [68]
presents different patch tests for different elements as for membranes, beams,
plates and shells. Although the test helps to understand the element behavior
with respect to mesh distortions, the test is neither necessary nor sufficient to
ensure the convergence against the exact solution.
Figure 24-5 represents an assembly of five elements, which is affected by displa-
cements of the boundaries. The displacements of the boundaries should result
y

7 8 x/l x y/l y

1 0.000 0.000
➃
2 1.000 0.000
5 6
3 0.166 0.166
➁ ➄ 4 0.750 0.250
➂ 5 0.333 0.666
4 6 0.666 0.666
3
7 0.000 1.000
1 ➀ 2
8 1.000 1.000
x

l x = 6.0 m, l y = 3.0 m, E = 1.0 N/m2 , ν = 0.25

Fig. 24-5 Patch test of the membrane elements

in constant strain fields in both directions and constant shear strains within
the patch. If these assumptions become true, the test is fulfilled. As an exam-
ple, Table 24.5 represents the boundary conditions with respect to strain in
y–direction as well as shear in x–direction.

Table 24.5 Boundary conditions of the patch–test

strain in y shear in x
node u[m] v[m] u[m] v[m]

1 0. 0 0. 0 0. 0 0. 0
2 — 0. 0 — 0. 0
7 — 1. 0 1. 0 0. 0
8 — 1. 0 1. 0 0. 0
426 24 Quality of Elements

As an example the elements P-DQ-4-SRI and P-HMQC-4-5 are compared. The

accurate results of the test are σyy = 0. 3333 N/m2 in the case of strain, and
σxy = 0. 13333 N/m2 in the case of shear, whereat the other stresses are zero.
The results of the P-DQ-4-SRI element are listed in Table 24.6. The strains and
thus the stresses are not constant but distorted all over the patch and inside
each element. Nonetheless, the results are quite well, since the deviations from
the exact solution are moderate, and the test is satisfied in average.

Table 24.6 Results of the P-DQ-4-SRI element

strain in y shear in x
ele. node σxx σyy σxy σxx σyy σxy
1 1 0. 0089 0. 3373 0. 0749 −0. 0110 −0. 0112 0. 1234
2 0. 0083 0. 3347 −0. 0445 −0. 0980 −0. 0065 0. 1244
4 0. 0175 0. 2959 −0. 0376 0. 0455 0. 0076 0. 1651
3 0. 0240 0. 3963 0. 0636 0. 0608 0. 0054 0. 0706
2 1 −0. 0113 0. 3430 −0. 0968 0. 1629 0. 0415 0. 1367
3 −0. 0186 0. 5013 −0. 0913 0. 2126 0. 0509 0. 0993
5 0. 0117 0. 2621 0. 0327 −0. 0561 −0. 0147 0. 1657
7 0. 0168 0. 3500 0. 0366 −0. 0948 −0. 0229 0. 1367
3 3 0. 0204 0. 3471 0. 0112 0. 0475 0. 0097 0. 1227
4 0. 0234 0. 3392 −0. 0163 0. 0425 0. 0076 0. 1480
6 0. 0270 0. 3497 −0. 0065 −0. 0574 −0. 0175 0. 1329
5 0. 0224 0. 3312 0. 0194 −0. 0571 −0. 0164 0. 1488
4 5 0. 0259 0. 3472 −0. 0484 −0. 0378 −0. 0021 0. 1024
6 0. 0204 0. 3252 0. 0490 −0. 0379 −0. 0026 0. 1620
8 0. 0150 0. 3238 0. 0531 0. 0018 0. 0074 0. 1322
7 0. 0205 0. 3459 −0. 0525 0. 0020 0. 0078 0. 1322
5 2 0. 0126 0. 3375 −0. 0752 0. 1987 0. 0488 0. 1351
8 0. 0025 0. 3350 0. 0546 −0. 1569 −0. 0401 0. 1357
6 0. 0090 0. 3899 0. 0498 −0. 0977 −0. 0242 0. 0913
4 0. 0048 0. 2532 −0. 0811 0. 1995 0. 0500 0. 1357

A complete different result offers the P-HMQC-4-5 element published by Pian–

Sumihara [81], where the stresses match the accurate solution and are constant
all over the patch, cf. Table 24.7.

Table 24.7 Results of the P-HMQC-4-5 element due to Pian - Sumihara [81]

strain in y shear in x
ele. node σxx σyy σxy σxx σyy σxy
all all 0. 0 0. 3333 0. 0 0. 0 0. 0 0. 1333
References

[1] ADINI, A., CLOUGH, R. W.: Analysis of PLate Bending by the Finite Ele-
ment Method. National Science Foundation, Grant G7337 (1960)

[2] AINSWORTH, M., ODEN, J.T.: A posterior error estimation in finite element
analysis. New York, John Wiley & Sons (2000)

[3] AINSWORTH, M., ODEN, J.T.: A unified approach to a posteriori error esti-
mation using element residual methods. Numerische Mathematik 65, pp 23–
50(1993)

[4] ARGYRIS, J. H.: Die Matrizentheorie der Statik. Ing.–Archiv 25 (1957)

[5] ARGYRIS, J. H., FRIED, I. and SCHARPF, D.: The TUBA family of plate
elements for the matrix displacement method. Aeronautical Journal 72 (1968)

[6] BABUŜKA, I.: Error bounds for finite element method. Numerische Mathe-
matik 16, Springer Verlag (1971)

[7] BABUŜKA, I.: The Finite Element Method with Lagrangian Multipliers. Nu-
merische Mathematik 20, Springer Verlag (1973)

[8] BABUŜKA, I., STROUBOULIS, T.: The finite element method and its relia-
bility. Oxford, Oxford Science Publications (2001)

[9] BABUŜKA, I., SURI, M.: Locking effects in the finite element approximation
of elasticity problems. Numerische Mathematik 62, pp. 439–463 (1992)

[10] BABUŜKA, I.; SZABÓ, B. A.: On the Rates of Convergence of the Finite
Element Method. International Journal of Numerical Methods in Engneering
18, pp. 323-341 (1982)

[11] BABUŜKA, I.: The p- and hp-Versions of the Finite Element Method: The
State of the Art. in: Finite Elements: Theory and Applications, edtrs. D. L.
Dwoyer, M. Y. Hussaini and R. G. Voigt. Proceedings of the ICASE Finite
Element Theory and Application Workshop 1986, Springer-Verlag, New York
(1988)

[12] BATHE, K. J.: Finite Elemente Methoden. Springer Verlag, Berlin (2002)

[13] BATHE, K. J., DVORKIN, E. N.: A Four-Node Plate Bending Element Based
on Mindlin–Reissner Theory and a Mixed Interpolation. International Journal
for Numerical Methods in Engineering 21, pp. 367-383 (1985)

© The Editor(s) (if applicable) and The Author(s), under exclusive license to
Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9
428 References

[14] BATHE, K. J., DVORKIN, E. N.: A formulation of general shell elements –

the use of mixed interpolation of tensorial components. International Journal
for Numerical Methods in Engineering 22, pp. 697-722 (1986)

[15] BATOZ, J. L., BATHE, K. J., HO, L. W. : A Study of Three-Node Triangu-

lar Plate Bending Elements. International Journal for Numerical Methods in
Engineering 15, John Wiley&Sons (1980)

[16] BELL, K.: A refined triangular plate bending finite element. International
Journal for Numerical Methods in Engineering 1 (1969)

[17] BELYTSCHKO, T., TSAY, C.S., LIU, W.K. : A Stabilization Matrix for the
Bilinear Mindlin Plate Element. Computer Methods in Applied Mechanics and
Engineering, 29, pp. 313-327 (1981)

[18] BELYTSCHKO, T., LIU, W. K., ENGELMANN, E.: The gamma–elements

and related concepts. In: T. J. R. HUGHES und E. HINTON (Edtrs.): Finite
Element Methods for Plate and Shell Structures, Vol. 1: Element Technology,
pp. 316347, Pineridge Press International (1986)

[19] BLETZINGER, K. U., BISCHOFF, M., RAMM, E.: A unified approach for
shear–locking–free triangular and rectangular shell finite elements. Computers
& Structures 75, pp. 321-334 (2000)

[20] BOGNER, F. K., FOX, R. L. and SCHMIT, L. A.: The generation of interele-
ment, compatibie stiffness and mass matrices by use of interpolation formulae.
Proc. 1st Conference on Matrix Methods in Structural Mechanics. Wright Pat-
terson Air Force Base, Ohio, AFFDL–TR–66–80, (1965)

[21] BOSSHARD, W.: Ein neues, vollverträgliches endliches Element für Platten-
biegung. International Association for Bridge and Structural Engineering, Bul-
letin 28, 27-40 (1968)

[22] BREZZI, F.: On the existence, uniqueness and approximation of saddle-point

problems arising from Lagrange Multipliers. R.A.I.R.O Anal. Numer. 8 (1974)

[23] BUCALEM, M. L., BATHE, K. J.: Higher-order MITC general shell elements.
International Journal for Numerical Methods in Engineering 36, pp. 3729-3754
(1993)

[24] BUCALEM, M. L., BATHE, K. J.: Finite Element Analysis of Shell Structures.
Archive of Computational Methods in Engineering 4 pp. 3–61 (1997)

[25] BUTLIN, G. and FORD, R.: A compatible triangular plate bending finite ele-
ment. University of Leicester, Engineering Department, Report 68–15 (1968)
References 429

[26] CLOUGH, R. W.: The finite element method in plane stress analysis. In:
Proceedings, 2nd Conference on Electronic Computation, A.S.C.E. Structural
Division. Pittsburgh, Pennsylvania (1960)

[27] CLOUGH, R. W., TOCHER, J. L.: Finite element etiffness matrices for ana-
lysis of plate bending. Proc. 1st Conference on Matrix Methods in Structural
Mechanics, Wright Patterson Air Force Base, Ohio, AFFD-TR-66-80, pp. 515–
545 (1965)

[28] COOK, R. D.: Improved two-dimensional finite element. A.S.C.E., Journal of

the Structural Division, ST 9, 1851–1863 (1974)

[29] COURANT, R.: Variational Methods for the Solution of Problems of Equi-
librium and Vibrations. Bulletin of the American Mathematical Society 49
(1943)

[30] COWPER, G. R., KOSKO, E., LINDBERG, G. M. and Olson, M. D.: Sta-
tic and Dynamic Applications of a High–Precision Triangular Plate Bending
Element. AIAA Journal 7, 1957–1965 (1969)

[31] COTRELL, J. A.; HUGHES, T. J. R.; BASILEVS, Y.: Isogeometric Analysis:

Toward Integration of CAD and FEA. John Wiley & Sons (2009)

[32] CUTHILL, E., McKEE, J.: Reducing the bandwidth of sparse symmetric ma-
trices. ACM P-69, Proceedings of ACM National Conference, New York, pp.
157172 (1969)

[33] CUNDALL, P. A.: A computer model for simulating progressive large scale
movements in blocky rock systems. In: Proceedings of the Symposium of the
International Society of Rock Mechanics. Vol 1, II–8, France (1971)

[34] CUNDALL, P. A. and STRACK, Q. D. L.: A Discrete Numerical Model for

Granular Assemblies. Geotechnique, 29, 47–65 (1971).

[35] DOHERTY, W. P., WILSON E. L. and TAYLOR, R. L.: Stress Analysis of

Axisymmetric Solids Utilizing Higher Order Quadrilateral Finite Elements.
UCB/SESM Report No. 69/03, University of California, Berkeley (1969)

[36] FLANAGAN, D. F., BELYTSCHKO, T.: A uniform strain hexahedron and

quadrilateral with orthogonal hourglass controL. International Journal for Nu-
merical Methods in Engineering 17, pp. 679–706 (1981)

[37] GALLAGHER, R. H.: Finite Element Analysis. Springer Verlag, Berlin (1976)

[38] GIRKMANN, K.: Flächentragwerke, 6. Auflage, Springer-Verlag, Wien (1963)

430 References

[39] GEORGE, P.: The advancing–front mesh generation method revisited. Inter-
national Journal for Numerical Methods in Engineering 58, pp. 1061–1089
(1994)

[40] GRÄTSCH, T., BATHE, K.–J.: A posterior error estimation techniques in

practicl finite element analysis. Computers & Structures 83, Science Direct,
Elsevier, pp. 235–265 (2005)

[41] GREENE, B. E., JONES, R. E., McLAY, R. W., STROME, D. R.: Generalized
variational principles in the finite-element method. AIAA Journal 6, pp. 1254–
1260 (1968)

[42] HARVEY, J. W., KELSEY, S.: Triangular Plate Bending Elements with En-
forced Compatibility. AIAA Journal 9, 1023–1026 (1971)

[43] HELLINGER, E.: Die allgemeinen Ansätze der Mechanik der Kontinua. In:
Die Encyklopädie der mathematischen Wissenschaften, Vol. IV/2, ed. F. Klein
und C. H. Müller, B. G. Teubner Verlag, Leipzig (1913)

[44] HU, H.C. Some Variational Principles in Elasticity and Plasticity. Acta Phy-
sica Sinica 10, pp. 259-290 (1954)

[45] HUGHES, T. J. R., TAYLOR, R. L., KANOKNUKULCHAI, W.: A simple

and efficient finite element for plate bending. International Journal of Nume-
rical Methods in Engineering 11, pp. 1529–1543 (1977)

[46] HUGHES, T. J. R., COHEN, M., HAROUN, M.: Reduced and selective inte-
gration techniques in the Finite Element Analysis of plates. Nuclear Enginee-
ring and Design 46, 203–222 (1978)

[47] HUGHES, T. J. R.,: The Finite Element Method: Linear Static and Dynamic
Finite Element Analysis. Prentice Hall (1987), Dover (2000, ... 2012)

[48] HUGHES, T. J. R., TEZDUYAR, T. E.: Finite elements based upon mindlin
plate theory with particular reference to four-node bilinear isoparametric ele-
ment. In: T. J. R. HUGHES, D. GARTLING and R. L. SPILKER Edtrs: New
concepts in finite element analysis 44, The American Society of Mechanical
Engineers, pp. 81-106 (1981)

[49] HUGHES, T. J. R.; COTRELL, J. A.; BASILEVS, Y.: Isogeometric Analysis:

CAD, finite elements, NURBS, exact geometry and mesh refinement. Compu-
ter Methods in Applied Mechanics and Engineering 194, pp. 4135–4195 (2005)

[50] IBRAHIMBEGOVIC, A., WILSON, E. L.: A modified method of incompatible

modes. International Journal for Numerical Methods in Biomedical Enginee-
ring 7, pp. 187–194 (1991)
References 431

[51] Institute of Structural Analysis, Technische Universität Braunschweig, internal

reports, unpublished

[52] IRONS, B. M. and ZIENKIEWICZ, O. C.: The Isoparametric Finite Element

System A New Concept in Finite Element Analysis. in: Proceedings of the
Conference on Recent Advances in Stress Analysis. Royal Aeronautical Society,
London (1968)

[53] IRONS, B. M., RAZZAQUE, A.: Experience with the patch test for con-
vergence of finite elements. In: The Mathematical Foundations of the Finite
Element Method with Applications to Partial Differential Equations, ed. A.
K. Aziz, Academic Press, New York, pp. 557–587 (1972)

[54] KAVANAGH, K. T., KEY, S. W.: A note on selective and reduced integration
techniques in the Finite Element Method. International Journal of Numerical
Methods in Engineering 4, pp. 148 (1972)

[55] KIRCHHOFF, G.: Über das Gleichgewicht und die Bewegung einer elastischen
Scheibe. Journal fuür die reine und angewandte Mathematik 4, de Gruyter,
Berlin (1850)

[56] KOSCHNICK, F., BISCHOFF,M., CAMPRUB I, N., BLETZINGER, K. U.:

The Discrete Strain Gap Method and Membrane Locking. Computer Methods
in Applied Mechanics and Engineering 194, pp. 2444–2463 (2005)

[57] KOSLOFF, D., FRAZIER, G. A. : Treatment of hourglass patterns in low

order finite element codes. International Journal for Numerical and Analytical
Methods in Geomechanics 2, pp. 57–72 (1978)

[58] KRÖPLIN, B. H., DINKLER, D.: A creep type strategy used for tracing the
load path in elastoplastic post buckling analysis. Computer Methods in Ap-
plied Mechanics and Engineering 32 (1982)

[59] KRÖPLIN, B. H., DINKLER, D., HILLMANN, J.: An energy perturbation

applied to nonlinear structural analysis. Computer Methods in Applied Me-
chanics and Engineering 52 (1984)

[60] KRÖPLIN, B. H., DINKLER D.: Dynamic versus static buckling analysis
of thin walled shell structures. Finite Element Methods for Plate and Shell
Structures - Vol 2: Formulation and Algorithm, eds. T.J.R. Hughes and E.
Hinton, Pineridge Press (1986)

[61] LADYZHENSKAYA, O. A., SOLONNIKOV, V. A.: Unique solvability of an

initial- and boundary-value problem for viscous incompressible nonhomoge-
neous fluids. Journal of Mathematical Sciences 9, pp. 697–749 (1978)
432 References

[62] LIU, W.-H., SHERMAN, A. H.: Comparative Analysis of the Cuthill-McKee

and the Reverse Cuthill-McKee Ordering Algorithms for Sparse Matrices.
SIAM Journal on Numerical Analysis 13, pp. 198-213 (1976)

[63] LO, S. H.: A New Mesh Generation Scheme for Arbitrary Planar Domains
International Journal of Numerical Methods in Engineering 21, pp. 1403–1426
(1985)

[64] LO, S. H.: Finite Element Mesh Generation. CRC Press, Taylor & francis
Group (2015)

[65] LÖHNER, R., PARIKH, P.: Three-dimensional grid generation by the advan-
cing front method. International Journal of Numerical Methods in Engineering
26, pp. 11351149 (1988)

[66] LÖHNER, R.: A 2nd Generation Parallel Advancing Front Grid Generator. in:
Proceedings of the International Meshing Roundtable Conference, Computer
Science (2013)

[67] McLAY, R. W.: A special variational principle for the finite element method.
AIAA Journal 8, pp. 533–534 (1969)

[68] MACNEAL, R. H., HARDER, R. L.: A proposed Standard Set of Problems

to Test Finite Element Accuracy. Finite Elements in Analysis and Design 1,
pp. 3–20 (1985)

[69] MACNEAL, R.: A simple quadrilateral shell element. Computers & Structu-
res, pp. 175-183 (1978)

[70] MACNEAL, R.: Derivation of element stiffness matrices by assumed strain

distributions. Nuclear Engineering and Design 70, pp. 3-12 (1982)

[71] MELOSH, R. J.: A stiffness matrix for the analysis in plate bending. Journal
of the Aerospace Sciences 28, pp. 34–42 (1961)

[72] MINDLIN, R. D.: Influence of Rotatory Inertia and Shear Deformation on

Flexural Motion of Isotropic, Elastic Plates. Journal of Applied Mechanics 18,
pp. 31–38 (1951)

[73] MORLEY, L. S. D.: A trianguiar equilibrium element with Iinearly varying

bending moments for plate bending problems. Journal of the Royal Aeronau-
tical Society 71, pp. 715 (1967)

[74] NOOR, A. K., PETERS, J.: Nonlinear analysis via global-local mixed finite
element approach. International Journal for Numerical Methods in Enginee-
ring 15, pp. 1363-1380 (1980)
References 433

[75] NOOR, A. K., PETERS, J.: Mixed models and reduced/selective integrati-
on displacement models for nonlinear analysis of curved beams. International
Journal for Numerical Methods in Engineering 17, pp. 615-631 (1981)

[76] ONATE, E.: Structural Analysis with the Finite Element Method. Linear Sta-
tics: Volume 1: Basis and Solids. Lecture Notes on Numerical Methods in
Engineering and Sciences, Springer Verlag (2009)

[77] ONATE, E.: Structural Analysis with the Finite Element Method. Linear Sta-
tics: Volume 2: Beams, Plates and Shells, Lecture Notes on Numerical Methods
in Engineering and Sciences, Springer Verlag (2013)

[78] PARK, K. C., STANLEY, G. M.: A curved C0 shell element based on assumed
natural–coordinate strains. Journal of Applied Mechanics 53, Transactions of
ASME, pp. 278-290 (1986)

[79] PARK, K. C., PRAMONO, E., STANLEY, G. M., CABINESS, H. A.: The
ANS shell elements: earlier developments and recent improvements ((assu-
med natural–coordinate strain)). In: Analytical and Computational Models of
Shells, A. K. NOOR (Edtr.) ASME Special Publication CED–3, pp. 217-239
(1989)

[80] PIAN,T. H. H., TONG, P.: Basis of finite element methods for solid continua.
International Journal for Numerical Methods in Engineering 1, pp. 3–28 (1969)

[81] PIAN,T. H. H., SUMIHARA, K.: Rational approach for assumed stress finite
elements. International Journal for Numerical Methods in Engineering 20, pp.
1685–1695 (1984)

[82] PRANGE, G.: Das Extremum der Formänderungsarbeit. Habilitationsschrift,

Technische Hochschule Hannover (1916)

[83] PUGH, E. D. L., HINTON, E. and ZIENKIEWICZ, O. C,: A study of quadri-

lateral plate bending elements with reduced integration. International Journal
for Numerical Methods in Engineering 12, pp. 1059–1079 (1978)

[84] REDDY, J. N.: On Complementary Variational Principles for the Linear Theo-
ry of Plates. Journal of Structural Mechanics 4, pp. 417-436 (1976)

[85] REISSNER, E.: A new Derivation of the Equations for the Deformation of
Elastic Shells. American Journal of Mathematics 63 (1941)

[86] REISSNER, E.: On the Theory of Bending of Elastic Plate. Journal of Ma-
thematics and Physics 23, pp. 184–191 (1944)

[87] REISSNER, E.: On a variational theorem for finite elastic deformations. Jour-
nal of Mathematical Physics 32, pp. 129–135 (1953)
434 References

[88] RITZ, W.: Über eine neue Methode zur Lösung gewisser Variationsprobleme
der mathematischen Physik. Journal für die reine und angewandte Mathematik
35 (1908)

[89] RUPPERT, J.: A Delaunay refinement algorithm for quality 2-dimensional

mesh generation. Journal of Algorithms 18, pp. 548-585 (1995)

[90] SHEWCHUK, J. R.: Delaunay Refinement Algorithms for Triangular Mesh

Generation Department of Electrical Engineering and Computer Science, Uni-
versity of California at Berkeley 2001

[91] SIMO, J. C.; HUGHES, T. J. R.: On the variational foundations of assumed

strain methods. Journal of Applied Mechanics 53, pp. 5154 (1987)

[92] SIMO, J. C., RIFAI, S.: A Class of Mixed Assumed Strain Methods and the
Method of Incompatible Modes. International Journal for Numerical Methods
in Engineering 29, pp. 1595-1638 (1990)

[93] STRICKLIN, J. A., HAISLER, W., TISDALE, P., GUNDERSON, R.: A ra-
pidly converging triangular plate element. AIAA Journal 7, pp. 180–181 (1969)

[94] SUKUMAR, N., TABARRAEI, A.: Conforming polygonal finite elements. In-
ternational Journal for Numerical Methods in Engineering 61, pp. 2045–2066
(2004)

[95] TAIG, I.C.: Structural analysis by the matrix-displacement method. Englisch

Electrical Aviation Limited, Rep. No. SO 17 (1962)

[96] TAYLOR, R.L., BERENSFORD, P. J. und WILSON, E. L.: Non-conforming

element for stress analysis. International Journal for Numerical Methods in
Engineering 10, pp. 1211-1219 (1976) 5

[97] TIMOSHENKO, S., WOINOWSKY–KRIEGER, S.: Theory of Plates and

Shells. 2nd Edition, McGraw–Hill Books (1959)

[98] TREFFTZ, E.: Ein Gegenst ück zum Ritzschen Verfahren. Proceedings of the
2nd International Congress for Applied Mechanics, Zrich, pp. 131-137 (1926)

[99] TURNER, M. J., CLOUGH, R. W., MARTIN, H. C., TOPP, L. J.: Stiffness
and Deflection Analysis of Complex Structures. Journal of the Aeronautical
Sciences 23 (1956)

[100] VERFÜHRT, R.: A review of a posterior error estimation and adaptive mesh
refinement techniques. Chichester, John Wiley & Sons (1996)

[101] WACHSPRESS, E. L.: A Rational Finite Element Basis. Academic Press, New
York (1975)
References 435

[102] WASHIZU, K.: On the variational principles of elasticity and plasticity. Aero-
elastic and Structures Research Laboratory, Massachuetts Institute of Tech-
nology, Technical Report 25-18 (1955)

[103] WILLIAMS, M. L.: Theoretical and experimental effect of sweep upon stress
and deflection in aircraft wings of high solidity – Part 5. California Institute
of Technology (1950)

[104] WILSON, E. L., TAYLOR, R. L., DOHERTY, W. and GHABOUSSI, J.:

Incompatible Displacement Models. Proceedings, ONR Symposium on Nume-
rical and Computer Methods in Structural Mechanics, University of Illinois,
Urbana, (1971)

[105] ZIENKIEWICZ, O. C., CHEUNG, Y. K.: The Finite Element Method in

Structural and Continuum Mechanics. McGraw–Hill, London (1967)

[106] ZIENKIEWICZ, O. C., TAYLOR, R. L.: The Finite Element Method. Vol-
ume I: Basis and Fundamentals. 6th Edition, Butterworth–Heinemann, Oxford
(2005)

[107] ZIENKIEWICZ, O. C., TAYLOR, R. L. and TOO, J. M.: Reduced integration

techniques in general analysis of plates and shells. International Journal of
Numerical Methods in Engineering 3, pp. 275–290 (1971)
Index

assumed natural strain method, 421 hybrid–mixed FEM, 356

hybrid–mixed FEM, results, 357
bandwidth, 58, 59 mixed FEM, 354
bar mixed FEM, results, 355
approximate solution, 12 reduced integration, 352
boundary action, 13 benchmark
distributed load, 14 Kirchhoff – triangular elements
two elements, 15 remarks, 344
exact solution, 12 square plate, 339
governing equation, 11 trapezoidal plate, 343
hybrid–mixed FEM, 242 membrane structure
intersection Cook’s cantilever, 220, 271
virtual work, 24 Reissner–Mindlin plate
mixed FEM, 234 cantilever plate, 373
shape function, 29 trapezoidal plate, 377
stresses, 32 boundary condition
subsequent analysis, 32, 37 Dirichlet, 11, 17, 23, 61
virtual work, 20, 26 Neumann, 11, 17, 23
virtual work integration, 31
work equation, 19, 25, 29 condition of deformation
beam mixed FEM
Bernoulli beam bar, 235
load vector, 105 Euler–Bernoulli beam, 274
subsequent stress analysis, 106 Kirchhoff plate, 285
Euler–Bernoulli beam, 101 membrane, 250
governing equations, 101 connectivity matrix, 55, 58
Hermite–Polynomials, 103 convergence
matrix notation, 102 criteria, 9, 38, 42
mixed FEM, 273 conformity, 42
shape functions, 103 coordinate invariance, 45
stiffness matrix, 104 least constant strains, 42
weak formulation, 102 rigid–body displacement, 44
Timoshenko beam, 349 Discrete Kirchhoff Theory, 337
convergence behavior, 358 hybrid FEM
displacement formulation, results, Kirchhoff 10+3 DOF triangle,
354 308, 312, 315
governing equations, 349 hybrid–mixed FEM

© The Editor(s) (if applicable) and The Author(s), under exclusive license to
Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2024
D. Dinkler und U. Kowalsky, Introduction to Finite Element Methods,
https://doi.org/10.1007/978-3-658-42742-9
438 Index

Kirchhoff – triangles, 325, 326 nodal displacement, 9, 52

Kirchhoff plate, 297 real, 22
membrane, 268, 271 virtual, 22
Kirchhoff – rectangular element
12 DOF – bi–cubic approach, eigenvalue analysis
128 deformation modes, 414
12 DOF – weak compliance, 131 quality of elements, 413
16 DOF – bi–cubic approach, rectangular membranes, 415
125 rigid body modes, 414
Kirchhoff plate trapezoidal membranes, 419
comparison of elements, 194 element
membrane element library, 10
comparison of elements, 220 triangles, 135
isoparametric element, 220 work equation, 29
rectangular element, 95 enhanced assumed strain method, 423
mixed FEM error estimation, 403
bar, 240 least squares method, 404
Kirchhoff plate, 292 Principle of virtual Work, 407
Reissner–Mindlin plate singularities, 410
benchmark, 377
displacement elements, 373 force
hybrid–mixed rectangular element, virtual, 27
396
mixed element, 391 heat conduction, 63, 145
triangular element, 384 one dimension
Timoshenko beam element, 358 boundary conditions, 65
governing equations, 64
degree of freedom, 9 weak formulation, 65
Discrete Kirchhoff Theory, 327 rectangular element, 69
convergence, 337 element matrix, 75
Kirchhoff–Hypothesis, 327, 332 governing equations, 70
scaling matrix, 334 matrix notation, 70
scaling matrix of deflection, 332 subsequent flux analysis, 77
scaling matrix of rotations, 330 vector of thermal action, 75
shape functions of rotations, 328 weak formulation, 70
subsequent stress analysis, 336 triangular element, 145
discrete shear gap method, 423 element matrix, 147
discretization, 28 shape functions, 145
work equation, 19 subsequent flux analysis, 148
displacements vector of thermal action, 148
formulation, 22 hybrid elements, 231, 301
Index 439

Kirchhoff plate subsequent stress analysis, 257

rectangular elements, 283 transformation of coordinates, 266
triangles, 301 work equation, 253
hybrid–mixed FEM, 232, 242, 249
bar, 242 interface
linear shape functions, 246 hybrid FEM
stiffness matrix, 245, 246 Kirchhoff 10+3 DOF triangle,
work equation, 243 306, 310
Euler–Bernoulli beam, 277 hybrid–mixed FEM
continuity conditions, 280 bar, 243, 244
elimination of stresses, 280 Euler–Bernoulli beam, 280
stiffness matrix, 279 Kirchhoff – triangles, 321
work equation, 279 membrane, 254
Kirchhoff – triangles, 317 mixed FEM
10 DOF – convergence, 326 Euler–Bernoulli beam, 276
10 DOF – interface, 324 intersection
10+3 DOF – convergence, 325 condition of deformation, 53
10+3 DOF – interface, 323 isoparametric elements, 215
continuity conditions, 321 element geometry, 216
cubic Displacements, 319 Jacobian, 217
elimination of stresses, 322 Kirchhoff plate
quadratic stress resultants, 320 12 DOF approach, 225
stiffness matrix, 321 16 DOF approach, 224
subsequent stress analysis, 324 quadrangle, 226
Kirchhoff plate quadrilateral elements, 223
convergence behavior, 297 membrane, 218
displacement variables, 293 hybrid–mixed FEM, 264
elimination of stresses, 296 reduced integration, 219
interface, 294, 296 stiffness matrix, 218
interface variables, 293
stiffness matrix, 295 Kirchhoff – isoparametric element
work equation, 294 12 DOF approach, 225
membrane, 249, 252 16 DOF approach, 224
comparison of elements, 268, 271 Kirchhoff – mixed element, 283
continuity conditions, 254, 255 Kirchhoff – rectangular element
convergence behavior, 268, 271 12 DOF – bi–cubic approach
isoparametric element, 264 convergence behavior, 128
linear shape functions, 258 scaling matrix, 127
Pian–Sumihara element, 264 shape functions, 126
quadratic shape functions, 261 12 DOF – weak compliance
stiffness matrix, 256 convergence, 131
440 Index

intersection condition, 130 subsequent stress analysis, 304

stiffness matrix, 129 Kirchhoff plate, 109
16 DOF – bi–cubic approach comparison of elements, 193
convergence behavior, 125 convergence, 194
load vector, 119 Discrete Kirchhoff Theory, 327
nodal displacements, 111 governing equations, 109, 283
scaling matrix, 114 hybrid triangles, 301
shape functions, 111, 113 hybrid–mixed quadrilateral, 293
stiffness matrix, 116 hybrid–mixed triangles, 317
stress matrix, 121 matrix notation, 110
subsequent stress analysis, 119 mixed principle of work, 283
Kirchhoff – triangular element, 173, quadrangle
302 isoparametric element, 226
10+3 DOF – cubic approach, 302 weak formulation, 110
convergence, equilibrium–constraint,
312 locking phenomena, 417
convergence, kinematic–constraint, kinematic locking, 417
308 Poisson locking, 421
convergence, weak conformity, 315 thickness locking, 420
elimination of the centre deflec-
tion, 304 matrix notation
hybrid elements, 305, 306 load vector, 36
interface, equilibrium–constraint, scaling matrix, 47
310 shape function, 35
interface, kinematic–constraint, stiffness matrix, 36
306 subsequent analysis, 37
scaling matrix, 303 work equation, 36
weak conformity, 314 membrane – hybrid–mixed element,
18 DOF – quintic approach, 189 249
scaling matrix, 191 membrane – isoparametric element,
21 DOF – quintic approach, 176 218
general polynomial, 177 reduced integration, 219
Jacobian, 179 stiffness matrix, 218
load vector, 188 subsequent stress analysis, 219
scaling matrix, 180, 184 membrane – rectangular element, 81
stiffness matrix, 186 convergence behavior, 95
subsequent stress analysis, 189 load vector, 89
9 DOF – Discrete Kirchhoff Theo- numerical integration, 204
ry, 327 reduced integration, 97
choice of shape functions, 173 shape functions, 83
numerical integration, 213 stiffness matrix, 87
Index 441

subsequent stress analysis, 90 Kirchhoff plate, 283

membrane – triangular element, 153 condition of deformation, 285
area coordinates, 158 convergence behavior, 292
linear shape functions, 158 element matrix, 289, 291
quadratic shape functions, 166 governing equations, 283
linear shape functions linear shape functions, 289
load vector, 157 load vector, 290
stiffness matrix, 155 matrix notation, 289
stress analysis, 157 nodal unknowns, 289
local coordinates membrane, 251
linear shape functions, 153 condition of deformation, 250
quadratic shape functions, 159 Mixed Principle of virtual Work, 10
numerical integration, 212
quadratic shape functions numerical integration, 199
load vector, 168 Gauss–Legendre quadrature, 199
stiffness matrix, 163 one dimension, 201
stress analysis, 169 rectangular membrane element, 204
membrane structure, 81, 153 bi–linear shape functions, 209
benchmark bi–quadratic shape functions, 209
Cook’s cantilever, 220, 271 reduced integration, 209
comparison of elements, 170, 220 triangles, 202
governing equations, 82, 249 triangular membrane element, 212
plain strain situation, 100 triangular plate element, 213
plane stress situation, 81 two and three dimensions, 202
weak formulation, 82
mesh adaptation, 410 patch–test, 425
hanging nodes, 411 Principle of virtual Displacements,
isoparametric elements, 412 9, 19, 23
triangular elements, 411 definition, 22
mixed FEM, 231 Discrete Kirchhoff Theory, 327
bar, 234 element intersection, 24
condition of deformation, 235 governing equations, 22
convergence behavior, 240 hybrid–mixed FEM
element matrix, 240 bar, 244
load vector, 240 Euler–Bernoulli beam, 278
matrix notation, 238 Kirchhoff–rectangle, 294
shape functions, 237 Kirchhoff–triangles, 317
Euler–Bernoulli beam, 273 membrane, 255
condition of deformation, 274 Reissner–Mindlin plate, 393
continuity conditions, 276 Timoshenko beam, 356
governing equations, 273 mixed FEM
442 Index

bar, 236 stiffness matrix, 370

Euler–Bernoulli beam, 274 subsequent stress analysis, 372
Kirchhoff–rectangle, 285 hybrid–mixed rectangular element,
membrane, 251 393
Reissner–Mindlin plate, 387 boundary conditions, 396
Timoshenko beam, 354 convergence behavior, 396
Reissner–Mindlin plate, 363 shape functions, 396
Timoshenko beam, 350 mixed rectangular element, 387
Principle of virtual Forces, 10, 25 boundary conditions, 390
definition, 26 convergence behavior, 391
governing equations, 27 shape functions, 390
hybrid–mixed FEM quadrilateral element, 365
bar, 245 hour–glass-control, 367
Euler–Bernoulli beam, 278 linear shape functions, 365
Kirchhoff–rectangle, 294 load vector, 367
Kirchhoff–triangles, 318 reduced integration, 366
membrane, 255 stiffness matrix, 365
Reissner–Mindlin plate, 394 subsequent stress analysis, 368
Timoshenko beam, 356 triangular element, 380
kinematics, 27 benchmark – convergence, 384
mixed FEM comparison of elements, 384
bar, 236 load vector, 382
Euler–Bernoulli beam, 275 shape functions, 380
Kirchhoff–rectangle, 287 stiffness matrix, 382
membrane, 251 subsequent stress analysis, 383
Reissner–Mindlin plate, 389 work equation, 380
Timoshenko beam, 354 Reissner–Mindlin plate, 359
weak form of kinematics, 27 benchmark – convergence, 377
comparison of element
reduced integration remarks, 376
isoparametric membrane element, comparison of elements, 373
219 comparison with Kirchhoff plate,
membrane 359
numerical integration, 209 convergence of beam like deforma-
rectangular element, 97 tions, 373
Reissner–Mindlin plate governing equations, 362
quadrilateral element, 366 trapezoidal plate – results, 377
Timoshenko beam, 352 weak formulation, 364
Reissner–Mindlin
hybrid–displacement element, 368 scaling matrix, 47
shape functions, 370 load vector, 50
Index 443

matrix notation, 47
stiffness matrix, 50
shape function, 8, 38
bar, 29
complete approach, 46
convergence, 42
convergence criteria, 38
general polynomials, 39
Hermite Polynomials, 41
Lagrange Polynomials, 40
Legendre Polynomials, 39
matrix notation, 35
Pascal’s triangle, 46
scaling matrix, 47
symmetric approach, 46
singularities, 410
solution
approximate solution, 7, 12
bandwidth, 58
exact solution, 12
storage of equations, 58
strong solution, 7
weak solution, 7
systems’ equations, 55, 58

triangular elements, 135

area coordinates, 140
derivatives, 142
integration, 143
shape functions, 143
transformation, 141
local coordinates, 135
derivatives, 137
integration, 138
Jacobian matrix, 137
shape functions, 139
transformation, 136

virtual work, 28
matrix notation, 34