Applied

Finite Element Analysis

for Engineers

FRANK L. STASA
Florida Institute of Technology

Holt, Rinehart and Winston

New York Chicago San Francisco Philadelphia
Montreal Toronto London Sydney Tokyo
Mexico City Rio de Janeiro Madrid
 V

To Frank H'. and Lena C. Stasa,

Ann R. Holt,
and to the memory of
John A. Holt

HRW
Series in
Mechanical
Engineering

L. S. Fletcher, Series Editor

F L Stasa APPLIED FINITE ELEMENT ANALYSIS FOR ENGINEERS

B J. Torby ADVANCED DYNAMICS FOR ENGINEERS
l til % » uiD&ftk
Af tMt BmnnnTT iOPWf»»

^„

Copyright C
^('Aco.th.

All nghis rcscrscd

1985
wo.
CBS Publishing
Shifty
B? 5
'
'

Address correspondence lo
383 Madison Ascrue. New YorL, NY 10017

Library of Congress Cataloging In Publication Data

Stasa. Frank L
Applied finite element anal) sis for engineers

(HRW senes in mechanical engineering)

Includes bibliographies and index
I Finite element method I Title II Senes
TAM7F5S72 1985 620' 001 '515353 85 742
ISBN D-03-OL2737-0
Pnnted ui the United States of America

5678 OtS 987654321

CBS COLLEGE PUBLISHING

Holt, Rinehart and Winston
The Drsden Press
Saunders College Pubhsh'ng
 Preface

Written for senior-year undergraduates and first-year graduate students with solid back-
grounds in differential and integral calculus, this book is oriented toward engineers and
applied mathematicians. A course in linear algebra is helpful but not essential. Courses in
elasticity, heat transfer, and fluid mechanics should facilitate the student’s understanding of
the applications emphasized in this text. Overall, the author’s approach represents a com-
promise between the purely mathematical and the purely applied developments.
Stress or structural analysis is given about the same level of treatment as thermal and
fluid flow analysis. However, this book is structured so that it may be used in courses in
which the application area is strictly stress analysis or strictly thermal analysis. Moreover,
it is also possible to use this book without covering any of the material related to variational
calculus and variational formulations. This strategy is made possible by emphasizing the
Galerkin weighted-residual method in thermal (and fluid flow) analysis and the principle of
virtual displacements in stress analysis. Consequently, this book should be useful to instruc-
tors teaching a course on the finite clement method to senior undergraduate students. If the
nonvariational path is desired, the instructor should omit the following sections: the last part
of Sec. 4-2, Sec. 4-3, parts of Secs. 4-4, and Secs. 4-8, 5-3, and 8-6.
The two chapters are introductory in nature. Chapter 1 contains a brief survey of
first

what the element method is, as well as a brief history' of the method. Chapter 2 contains
finite

a review of the necessary mathematical concepts of matrices, vectors, and determinants. The
reason for including this review in the text proper was to establish a common ground from
which the finite element method could be developed, regardless of the specific background
of the engineering student.
All aspects of the finite element are explored in Chapter 3 by way of one of the simplest
of all engineering applications — the truss. The so-called direct approach is adopted for this

purpose and each step in the finite element solution process is given in full detail. For this

reason, all students must be exposed to (and indeed should master) Chapter 3, which is the
only structural analysis chapter that must be covered by all users of this book. Both two-
and three-dimensional trusses are covered, but the former is given broader, more compre-
hensive treatment. The student is also introduced to computer programming concepts and a
two-dimensional truss program.
Chapter 4 provides the general framework for the development of nearly all (nonstruc-
tural) finite element models. Here the student is introduced to the concepts of globally based
approximations to the true solution of simple ordinary differential equations. Among the

methods covered are the Ritz, Rayleigh-Ritz (variational), point collocation, subdomain
collocation, least squares, and Galerkin methods. The last four of these methods comprise
the class of approximate solution methods known as the weighted-residua! methods. Both
the Rayleigh-Ritz (variational) and the Galerkin (weighted-residual) methods are extended
to piecewise continuous approximations and, hence, to the finite element method itself. The

student is introduced, of course, to the concept of shape functions at this point. Chapter 4
concludes with an application in the thermal analysis area: a pin fin (a type of extended

iii
J» MtU-ACt

surface). Instructors may wish to have their students review Chapter l before proceeding
with the next chapter.
Chapter 5 is devoted specifically to the development of finite element models in the

stress (or structural) analysts area A brief review of some of the more important concepts
of elasticity is provided and may be skipped by those who have had a prior graduate course

on this subject Depending on the instructor’s preference, either of two developments may

be used the principle of minimum potential energy (variational) or the principle of virtual
displacement with a simple one-dimensional application the uniaxial stress member. The
results from Chapter 5 are used throughout Chapter 7, the primary stress application chapter.

Chapter 6 is. in effect, a catchall chapter, which contains essential material that has
not been covered adequately up to this point. Here the student is introduced more formally
to the concept of parameter functions (such as displacement and temperature functions) and
and completeness requirements that these functions should satisfy. Shape
to the compatibility
functions (C°-contmuous only) are derived and presented for the following types of one-,
two-, and three-dimensional elements two-node lineal (l-D). three-node triangular (2-D),
four-node rectangular (2-D), four-node tetrahedral (3-D), and eight-node brick (3-D). Local,
normalized coordinates, such as length, area, and volume coordinates, as well as serendipity
coordinates, are introduced Axtsymmctnc elements arc also presented Three simple inte-
gration formulas are given in terms of length, area, and volume coordinates for integrations
over lineal, triangular, and tetrahedral elements, respectively Finally, an alternative to the
matrix inversion technique is provided, namely, the active zone equation solver This method

is based on triangular decomposition, forward elimination, and backward substitution, and

takes advantage of the banded, and often symmetric, nature of the assemblage stiffness
matrix. This method requires that the assemblage stiffness matrix be stored as a column
vector (instead of a square matnx)
Chapter 7 is the main stress analysis application chapter Among the topics covered arc
the following two-dimensional stress analysis (plane stress and plane strain), axisymmetne
stress analysis, three-dimensional stress analysis, and the analysis of beams The notion of
substructunng and condensation is also introduced and working equations arc developed
The chapter concludes uuh a brief description of the development of a two-dimensional
stress analysis program The instructor is referred to the Instructor's Solutions manual for a
listing (in FORTRAN) of one version of
this program —
called program STRESS (the user’s
manual program is also included in the Instructor’s Solutions manual)
to the
Chapter 8 is the principal thermal and fluid flow analysis chapter, in which the following
topics arc covered one-, two-, and three-dimensional thermal analysis, and axisymmetne
thermal analysis Material on variational formulations in two-dimensional problems is pro-
vided. but this also may be skipped by those preferring the nonvanalional path Other
application areas in Chapter 8 include convective energy transport, two-dimensional potential
flow, and two-dimensional incompressible fluid flow The chapter concludes with a brief
description of the development of a two-dimensional, steady -state thermal analysis program
Again the instructor is referred to the Instructor's Solutions manual for a listing (in FOR-
TRAN) of one version of this program—called HEAT (the user’s manual to the program is
also included in the Instructor’s Solutions manual)
Chapter 9 introduces higher-order elements and numerical integration (quadrature) The
one-, two-, and three-dimensional elements introduced m
Chapter 6 arc extended to quadratic
and cubic order Subparametnc, isoparametric, and superparametne elements are also in-
troduced Two- and three-dimensional isoparametric formulations arc developed for the
quadrilateral and triangular elements as well as for the brick and tetrahedral dements Special
formulas arc presented, which facilitate greatly the evaluation of the integrals that naturally
arise
 :

PREFACE V

In conclusion, Chapter 10 is devoted to transient thermal analysis and dynamic structural

analysis. The concept of partial discretization is presented and applied to stress analysis and
thermal analysis. Lumped and consistent capacitance and mass matrices are discussed. So-
lution methods are developed based on the finite element method itself (i.e., in time) and
on the finite difference method. The result is two- and three-point recurrence schemes for
transient thermal analysis and dynamic structural analysis, respectively. The chapter con-
cludes with a brief introduction to modal analysis.
Appendix A contains the material property data to be used in the problems, unless
otherwise noted in the problem statements. Appendix B contains a short user’s manual to
the (two-dimensional) truss program along with the program listing (in FORTRAN). Appendix C
contains listings of subroutines ACTCOL and UACTCL (and function DOT), which have
been used with the written permission of McGraw-Hill and which appear in Professor
Zienkiewicz’s third edition of The Finite Element Method. The length of the STRESS and
HEAT programs and their respective user’s manuals precluded their inclusion in this text.
Instructors who would like to have cqpies of all of these programs on floppy disks are
encouraged to write The disk formats available are IBM PC, Apple II series,
to the author.

and 8-inch IBM 3740 standard format for CP/M-based machines. *Other FORTRAN pro-
grams that can be obtained on these disk formats include: beam analysis, a TurboPascal**
version of the two-dimensional truss program, fin analysis, transient one-dimensional thermal
analysis, and transient two-dimensional thermal analysis.
The following suggestions are made to instructors teaching on the quarter and semester
system. For those teaching on the quarter system: Chapters 1 to 4 can be covered comfortably
during the fall quarter. The winter quarter could be devoted to stress analysis with coverage
of Chapters 5, 6, 7, 9, and 10 (omit Secs. 10-4, 10-7, and 10-8). The spring quarter could
be devoted to thermal and fluid flow analysis with coverage of Chapters 6, 8, 9, and 10
(omit Secs. 10-3, 10-9, and 10-10). For students who take all three courses, some of the
material is necessarily repeated. The author has found this repetition not to be a problem,
because these students get a firmer grasp of the material the second time around.
For those teaching on the semester system Chapters 1 to 6 could be covered comfortably
during the first semester, and Chapters 7 to 10 during the second semester. This pace allows
sufficient class time for the discussion of computer programming techniques. A useful project
for the first course is to have the students modify the two-dimensional truss program in

Appendix B so that it could be used in three-dimensional truss applications.

The Instructor’s Solutions manual contains a listing (in FORTRAN) of a two-dimen-
sional, static stress analysis program (for Chapter 8). The manual is available from the

publisher upon proper written request. The instructors may find it useful to distribute copies

of these programs to their classes.

The author wishes to thank all of the students who have used the original notes and
class-tested the text manuscript. Their comments and suggestions were taken seriously, and
their words of encouragement will always be remembered. Deserving of special recognition
is Jay A. Buckman, who did an outstanding job of proofreading the page proofs. A special
debt of gratitude is owed to General Herbert McChrystal, who first suggested this project.
Very special thanks are extended to U. Shripathi Kamath, who had the monumental task of
providing the solutions manual to the text, and to Michael Weaver for converting the FOR-
TRAN version of the truss program into Pascal. The author is also thankful to the editors

*CP/M is registered trademark of Digital Research, Inc.

**TurboPascal is a trademark of Borland International.
 PRO- ACL

John J. Beck, L> nn Coninjcci, and Rachel Hocken for their help and coopcra Jion in publishing
the manuscript. Finally, and most important of all, the author is once again deeply indebted
to his wife, Donna, who meticulously typed every page of the manuscript and without whose
encouragement this book would probably not have been completed, and to Lisa Ann, for
helping to put this project into proper perspective and for providing the perpetual light at

the end of tunnel.

Frank L. Stasa
 CONTENTS

1- Preface iii

1
2- General Concepts 1

1-1 Introduction 1
1-2 What Is the Finite Element Method? 3
1-3 Discretization 5
1-4 Relationship to the Finite-Difference Method 8
1-5 Advantages and Disadvantages of the Finite
Element Method 10
1-6 Brief History of the Finite Element Method 13
7 Remarks 15
References 16

2 Mathematical Preliminaries 19
1 Introduction 19
2-2 Definition of a Matrix 19
2-3 Equality of Two Matrices 21
2-4 Matrix Algebra 21
Matrix Addition and Subtraction 22
Matrix Multiplication 22
The Commutative, Distributive, and Associative
Properties 24
2-5 Transpose of a Matrix 24
2-6 Determinants, Minors, and Cofactors 25

rii
 TAEiLF OF CONTLSTS

2-7 Adjoint and Inverse of a Matrix 27

2-8 Matrix Partitioning 28
2-9 Definition of a Vector 29
2-10
2- Equality of Two Vectors 31
2-11 Length of a Vector 31
2-12 Vector Algebra 32
2-13 Direction Cosines 33
14 Solution to Systems of Linear Algebraic
3-
Equations 35
References 36
Problems 37

3 Truss Analysis: The Direct Approach 43

1 Introduction 43
3-2 Element Formulation: The
Finite Direct
Approach 45
Discretization 45
Element Stiffness Relationship m Local Coordinates 47
Transformation from Local to Global Coordinates 49
Global Element Stiffness Relationship 51
Assemblage 55
Application of Loads 58
Application of Restraints on Nodal Displacements and
Solution 59
Computation of the Element Resultants 61
3-3 Application to a Specific Example 62
3-4 Summary of the Basic Steps 70
Step 1: Discretization 70
Step 2: Determination of the Local Element
Characteristics 71
Step 3: Transformation of the Element
Characteristics 71
Step 4: Assemblage of the Global Element
Characteristics 72
Step 5: Application of the Prescribed Displacements 72
Step 6: Solution 72
Step 7: Calculation of Element Resultants 72
3-5 Three-Dimensional Truss Formulation 73
3-6 Description of a Simple Computer Program:
TRUSS 79
Description of the Program 79
Mesh Generation 83
3-7 Remarks 87
References 90
Problems 90
4 Variational and Weighted Residual
Formulations 101
4-1 Introduction 101
4-2 Some Approximate Solution Methods 102
General Concepts 103
The Ritz Method 104
The Variational Method 107
4-3 Variational Calculus: An Introduction 110
4-4 Some Additional Mathematics 112
Integration by Parts 113
Taylor’s Series 113
The Differential of a Function 113
The Variation of a Functional 114
The Commutative Properties 114
Miscellaneous Rules of Variational Calculus 114
4-5 The Euler-Lagrange Equation, Geometric and
Natural Boundary Conditions 115
4-6 The Method of Weighted Residuals 121
General Concepts 121
Point Collocation 122
Subdomain Collocation 124
Least Squares 126
Galerkin 128
4- Comparison with the Exact Solution 129
4-7 More Mathematics 130
Derivative of a Matrix with Respect to a Scalar 131
Integral of a Matrix with Respect to a Scalar 131
5- Derivative of a Scalar Function with Respect to a
Vector 131
The Sum of Other Integrals
Integral as a 133
4-8 The Rayleigh-Ritz Finite Element Method 133
4-9 The Galerkin Finite Element Method 142
4-10 Application: One-Dimensional Heat Transfer in a
Pin Fin 146
The Element Characteristics 147
Heat Removal Rate 149
Fin Efficiency 150
11 Remarks 155
References 157
Problems 157

5 General Approach to Structural Analysis 185

1 Introduction 185
5-2 Basic Concepts in Elasticity 186
 TABLE OF CONTENTS

Stress at a Point 187

The Equations of Static Equilibrium 189
Principal Stresses 191
Strain at a Point 195
Strain-Displacement Relations 196
The Compatibility Equation 197
A Constitutive Relationship — Hooke's Law 198
Surface Tractions 199
5-3 Principle of Minimum Energy
Potential 200
5-4 Principle of Virtual Displacements 204
5-5 The Finite Element Basis 206
5-6 The Shape Function Matrix 207
5- Two-Node Lineal Element 208
Three-Node Tnangular Element 208
Four-Node Tetrahedral Element 210
Properties of C°-Continuous Shape Functions 211
5-7 The Finite Element Characteristics 213
6-
5-8 Application: Tapered Uniaxial Stress Member 215
The Element Charactenstics 216
The Element Resultants 219
9 Remarks 225
References 226
Problems 227

6 Parameter Functions; C°-Continuous Shape

Functions; Simple Integration Formulas;
Active Zone Equation Solvers 239
1 Introduction 239
6-2 Parameter Functions 239
Definition of a Parameter Function 240
Restrictions on the Parameter Functions 240
6-3 One-Dimensional Element 244
Global Coordinates 244
Serendipity Coordinate 247
Length Coordinates 248
6-4 Two-Dimensional Elements 249
Three-Node Tnangular Element 249
Four-Node Rectangular Element 253
6-5 Three-Dimensional Elements 257
Four-Node Tetrahedral Element 258
Eight-Node Bnck Element 263
6-6 Axisymmetric Elements 265
6-7 Some Simple Integration Formulas 267
Length Coordinates 267
 TABLE OF CONTENTS

Area Coordinates 268

Volume Coordinates 269
6-8 Active Zone Equation Solver 270
Triangular Decomposition 270
Forward Elimination and Backward Substitution 273
Classification of Symmetric Real Matrices 276
Storage Considerations 277
Computer Implementation 279
6-9 Remarks 280
References 281
Problems 282

7 Stress Analysis 287

7-1 Introduction 287
7-2 Two-Dimensional Stress Analysis 287
Definitions of Plane Stress and Plane Strain 288
The Shape Function Matrix N 289
The Strain-Nodal Displacement Matrix B 290
The Constitutive Relationship 291
The Element Stiffness Matrix 294
The Element Nodal Force Vectors 296
Composite Nodal Force Vector 302
Assemblage Step 302
Prescribed Displacements 304
Solution for the Nodal Displacements 305
The Element Resultants 305
Remarks 306
7-3 Axisymmetric Stress Analysis 306
Definitions 306
The Strain-Displacement Relationship 306
The Shape Function and Strain-Nodal Displacement
Matrices 308
The Constitutive Relationship 309
The Element Stiffness Matrix 310
The Nodal Force Vectors 311
Remarks 315
7-4 Three-Dimensional Stress Analysis 316
The Strain-Displacement Relationship 317
The Shape Function and Strain-Nodal Displacement
Matrices 318
The Constitutive Relationship 319
The Element Stiffness Matrix 319
The Element Nodal Force Vectors 320
Remarks 322
xii TABLE OF CONTENTS

7*5 Beams 322

Review of the Elementary Theory 323
C'-Continuous Shape Functions 325
7-
The Finite Element Characteristics 329
Remarks 332
7*6 Substructuring 337
7 Development of a Two-Dimensional Stress
Analysis Program: Program STRESS 339
The Main Program 340
Mesh Generation 340
Data Storage 341
8-
7*8 Remarks 343
References 343
Problems 344

8 Steady-State Thermal and Fluid Flow

Analysis 371
1 Introduction 371
8-2 The Direct Iteration Method 371
8-3 One-Dimensional Heat Conduction 374
The Governing Equation 375
The FEM Formulation 376
8-4 The Green-Gauss Theorem 385
8-5 Simple Two-Dimensional Heat Conduction 388
8-6 Variational Formulations in Two Dimensions 391
The Euler-Lagranee Equation Geometnc and Natural
Boundary Conditions 391
The Finite Element Formulation 396
8-7 The Gaferkin Method in Two Dimensions 397
8-8 General Two-Dimensional Heat Conduction 398
The Governing Equation 399
The FEM Formulation 400
The Element Resultants 409
8-9 Axisymmetric Heat Conduction 421
The Governing Equation 422
The Finite Element Formulation 422
8-10 Three-Dimensional Heat Conduction 430
The Governing Equation 431
The Finite Element Formulation 431
8-11 Two-Dimensional Potential Flow 436
Velocity Potential Formulation 437
The Element Resultants 443
Stream Function Formulation 445
8-12 Convection Dominated Flows 448
 TABLE Of CONTENTS xiii

8-13 Incompressible Viscous Flow 452

8-14 Development of a Two-Dimensional Thermal
Analysis Program 456
The Main Program 457
8- Mesh Generation 457
Data Storage 459
Variable Property Routine: SUBROUTINE
VPROP 460
Point Heat Sources 463
9-
15 Remarks 463
References 464
Problems 465

9 Higher-Order Isoparametric Elements and

Quadrature 509
1 Introduction 509
9-2 One-Dimensional Elements 510
Length Coordinates 510
Serendipity Coordinate 511
9-3 Two-Dimensional Elements 512
The Triangular Element 512
The Rectangular Element 514
9-4 Three-Dimensional Elements 515
The Tetrahedral Element 516
The Brick Element 517
9-5 Subparametric, Isoparametric, and
Superparametric Elements 520
9-6 Two-Dimensional Isoparametric Formulations 525
The Quadrilateral Element 525
The Triangular Element 530
9-7 Three-Dimensional isoparametric Formulations 534
The Brick Element 535
The Tetrahedral Element 539
9-8 Numerical Integration: Rectangular and Brick
Elements 542
9-9 Numerical Integration: Triangular and
Tetrahedral Elements 553
9-10 Numerical Integration: Required Order 558
References 559
Problems 559

10 Transient and Dynamic Analyses 577

10-1 Introduction 577
10-2 Partial Discretization 578
 TABLE OF CONTENTS

10-3 Dynamic Structural Analysis 579

10-4 Transient Thermal Analysis 581
10-5 Lumped Versus Consistent Matrices 584
Consistent Matrices 584
Lumped Mass and Capacitance Matrices 586
10-6 Solution Methods 588
10-7 Two-Point Recurrence Schemes: The Finite
Difference Method 589
Forward Difference Scheme 589
Backward Difference Scheme 591
Central Difference Scheme 591
Summary and Application of the Prescribed
Temperatures 592
10-8 Two-Point Recurrence Schemes: The Finite
Element Method 594
Point Collocation 596
Subdomain Collocation 597
Galerkin 597
10-9 Three-Point Recurrence Schemes 604
Special Cases 608
Initial and Boundary Conditions 611
10-10 Introduction to Modal Analysis 614
References 818
Problems 618

Appendix A Structural and Thermal Properties 623

A-1 Structural Properties 623
A-2 Thermal Properties 624

Appendix B Program TRUSS 627

B-1 Listing of Source Code for Program TRUSS 627
B-2 Description of Input to Program TRUSS 638

Appendix C Active Zone Equation Solvers 643

Index 647
 General Concepts

1-1 INTRODUCTION

The element method of analysis is a very powerful, modem computational

finite

tool.The method has been used almost universally during the past 15 years to solve
very complex structural engineering problems, particularly in the aircraft industry'.
It is now gaining wide acceptance in other disciplines such as thermal analysis,
fluid mechanics, and electromagnetics. The method requires the use of a digital
computer because of the large number of computations involved.
Structures take many different, sometimes very' diverse, forms and utilize many
different materials. Oftentimes a structure itself is composed of a multitude of
materials. For example, bridges and space-deployable antennas are both structures,
but they hardly resemble each other in appearance and materials. A space-deployable
antenna looks like a gigantic umbrella that is closed during launch and opened, or
deployed, once in earth orbit. Figure 1-1 shows an antenna in the deployed con-
figuration. The structural design and analysis of this antenna was done with large
finite element computer programs such as NASTRAN [I] and STARDYNE [2],

In the design of a bridge, engineers are concerned with the stresses developed
in the bridge as a result of the weight of the vehicles riding on it, in addition to

wind loads and its own weight. These stresses are important because all materials
have upper limits on allowable stresses. When these allowable stresses are exceeded,
the structure may no longer perform its intended function. In the extreme case, the
structure may actually collapse. A tragic example of this is the collapse of two 32-
ton sky walks at the Hyatt Regency Hotel in Kansas City on July 27, 1981.
In the case of the space-deployable antenna, excessively large deflections of
its carefully contoured umbrellalike surface will deteriorate the antenna perfor-
 LMJlAt COSOPTS

Figure 1*1 Photograph of a spacc-dcpioyabic antenna tn solar thermal vacuum test cham-
ber ( Courtes\ of Harris Corporation and TRW )

mancc too large, the antenna may not be able to perform its
If the distortions arc

intended function These deflections may be caused by the heating effect of the
sun. Most materials stretch when heated and shrink when cooled. If part or all of
the antenna becomes shaded by the spacecraft, an improperly designed antenna may
distort enough to render itself useless.

In a typical thermal analysis application, uc may need to know the temperatures

within the body being analyzed. For example, for the space-deployable antenna the
amount of distortion as a result of the temperature change cannot be ascertained
until the temperatures themselves arc determined Sometimes in addition to the
temperatures, we need to know how much heat is flowing through a given area.
In fluid mcchantcs problems, the analyst may need to determine the fluid sclocitics
and pressures within or around some device or object The complex flow pattern
in a coal gasifier, for example, may be needed [3| The finite element method is
being used today to solve these and many other more complex problems
It should be obvious that it is not possible to obtain closed-form, analytical

solutions to complex problems such as these Often a skilled engineer can make
simplifying assumptions to obtain a closed-form solution, but the very nature of
 WHAT IS THE FINITE ELEMENT METHOD’ 3

Figure 1-2 Three-noded triangular element.

some problems, such as the bridge and space-deployable antenna, may preclude
such an approach. Complexities may arise because of the irregular and varied
geometry’, mixed boundary conditions, nonlinear material behavior, and nonuniform
loading conditions. The finite element method is particularly well-suited to handle
these and other complications.

1-2 WHAT IS THE FINITE ELEMENT METHOD?

The basic idea behind the finite element method is to divide the structure, body,
or region being analyzed into a large number offinite elements, or simply elements.
These elements may be one. two. or three dimensional. A popular and classical
two-dimensional element is the triangle shown in Fig. 1-2. When a two-dimensional
structure (or heat transfer device, etc.) is divided into hundreds or sometimes
thousands of these nonovcrlapping triangles, we can see that essentially all planar

geometries can be easily accommodated. Note that this particular element has three
nodes (;, j, k) appropriately placed at the vertices of the triangle. Before explaining
the purpose of these nodes or nodal points, let us see how we may use this type
of element to represent an irregularly shaped plate.
Figure 1-3 shows such a plate. By way of illustration, let us say that the

temperature distribution within the plate is sought for a given plate material, spec-
ified boundary conditions, etc. The field variable of interest here is the temperature

Because there is an infinite number of points in the plate, there is also an infinite

number of temperatures to be determined. The problem is therefore said to have

an infinite number of degrees of freedom. It should be recalled that when closcd-

Figure 1-3 Plate with variable thickness and an irregular shape.

6 gi.slrm. coNcrrrs

Figure 1-5 Illustration of the method or exhaustion in obtaining the area under a curve

(a) Exact area (b) Lower bound on estimate b> using inscribed rectangles (c) Upper bound
on estimate bv using circumscribed rectangles

where was actually equivalent to integration tn some cases In one of his extant
it

treatises,Measurement of a Ciritc. Archimedes placed the value of the circum-

ference of a circle between 3'A and 3'%i times its diameter Consequent!) . firm
limits were placed on the value of tr. Although others throughout the ages subse-
quently improved the value of t. it was Eudoxus and Archimedes who were largely
responsible for determining the value of this important constant
Bv wav of illustration, let us determine a sequence of approximate values of
tr bv inscribing and circumscribing regular pol)gons inside and around a circle as
shown in Fig- 1-6 for the case of regular inscribed and circumscribed octagons

The lengths of the sides of the polvgons are rcadtlv measured (because the) are
straight and not curved), and the resulting lower and upper bounds of tr arc rcadil)
 DISCRETIZATION 7

Figure 1-6 Approximation to the circumference of a circle by (a) an inscribed octagon

and (b) a circumscribed octagon.

calculated from Eq. (1-1). The results are given in Table 1-1, first, for an eight-
sided polygon, then, for each successive case doubling the number of sides until
a staggering 1,048,576 sides [6] are used —
hence the name method of exhaustion!
Note lower estimate converges more rapidly to the accepted value for
that the it

of 3.1415926536 (to 10 decimal places).

These simple and antiquated examples illustrate the concept of discretization —
the essence of which is the division of a larger, more difficult problem into a number

Table 1-1 Calculation of the Value of ir by the Method of Exhaustion Using Inscribed
and Circumscribed Regular Polygons

Number of Inscribed Circumscribed

sides polygons polygons

8 3.061467458921 4.959315235374
16 3.121445152258 3.878006734963
32 3.136548490546 3.477392565633
64 3.140331156955 3.302370812490
128 3.141277250933 3.220307554543
256 3.141513801144 3.180543968220
512 3.141572940367 3.160968277095
1024 3.141587725277 3.151255641334
2048 3.141591421511 3.146417964326
4096 3.141592345570 3.144003766021
8192 3.141592576585 3.142797824426
16384 3.141592634339 3.142195142707
32768 3.141592648777 3.141893874079
65536 3.141592652387 3.141743257818
131072 3. 141 592653289 3.141667954200
262144 3.141592653515 3.141630303519
524288 3.141592653571 3.141611478460
1048576 3.141592653585 3.141602066002
H CTM RAL COVCl ITS

of smaller, simpler problems such that ihe sum of the solutions for the simpler
problems, in some sense, approximates the solution to the original problem. The
example involving the computation of the area under a curve utilized a \ery common
finite element —the rectangular element. The example involving the evaluation of
it utilized another finite element — element. The
the lineal the rectangle, like triangle
in Sec. 1-2, is a two-dimensional element, whereas the lineal element is one-
dimensional. Quite appropnately. the two-dimensional clement was used to evaluate
an area, whereas the onc-dimensional element was used to determine the circum-
ference (which is a length and. hence, also one-dimensional) Not surprisingly,
three-dimensional elements such as tetrahedra and rectangular prisms (or bricks)
may be used m the method of exhaustion to determine the volume of three-dimen-
sional objects with complicated boundaries

1-4 RELATIONSHIP TO THE FINITE-DIFFERENCE METHOD

The finite-difference method (FDM) is an alternate solution technique that may be

used to solve the same types of problems for which the finite element method (FEM)
is suited. This is particularly true in the caves of thermal and fluid flow analyses
Both methods require the analyst to discretize the structure, object, or region being
analyzed or modeled However, the way in which the discretization is done is

fundamentally different
For the purpose of this discussion, let us restrict our attention to ihe two-
dimensional case. It should be recalled that the triangular and rectangular elements
may be used in FEM. So as not to obscure the basic point that isto be made here,
let us further restrict the discussion to the case of the rectangular dement In FEM.
the nodes arc placed at the comers of the rectangles Since the rectangle has four
comers, there are four nodes. This is quite different from the way in which the
discretization is done in FDM.
In FDM, the object or region being analyzed is divided into a finite number
-

of lumps. Note that the word “lump.” not the word “dement.' is used In the
baste FDM approach, each of these lumps have a constant value of
is assumed to

the pertinent field variable (this value, however, may be a function of time in
transient problems). For example, in a heat transfer or thermal analysis problem,
the relevant field variable is the temperature. Therefore, each lump would be
isothermal. Since the entire lump is isothermal, it seems natural that the node for
the lump should be associated not w ith any of its corners, but rather at its geometric
center or centroid. Figure 1-7 shows the two different methods of discretization
In Fig. l-7(a), a typical rectangular finite dement is shown with four nodes
It should be recalled from the discussion in Sec 1-2 that each of these four nodes
may have a different value of the field variable (displacements, temperatures,
velocities, pressures, etc.) It should further be recalled that by using interpolation
functions for the field variable within each dement, the field variable is generally

not constant over an dement but vanes in some prescribed manner, depending on
the interpolation polynomial being used
 RELATIONSHIP TO THE FINITE DIFFERENCE METHOD
9

(a) (b)

Figure 1 -7 Two main types of discretization, (a) An element used in finite element analysis,
(b) A lump used in finite-difference analysis. Note the four nodes on the element in (a) and
the one node on the lump in (b).

Figure l-7(b) shows a typical lump used in FDM analysis. The one and only
node appears at the center of the rectangular lump. In this case, the entire lump is
assumed to have a constant value of the field variable. In both FEM and FDM, the
nodes represent the locations at which the field variables are to be determined.
To further illustrate the difference between the FEM and FDM approaches, let

us discretize a rectangular plate into elements and lumps such that the same number
of nodes is used in each case. Figures l-8(a) and (b) clearly show the difference

between the two approaches. The FDM discretization in Fig. l-8(a) requires that
if nodes are to be placed on the plate boundaries (e.g., to impose a prescribed
temperature on this boundary), then the surface lump is essentially one-half the size
of an interior lump. Comer lumps are seen to be one-fourth the size of interior
lumps. This assumes that the nodes are to be equally spaced; if they are not, the
situation is even more complicated. Note that for the FEM discretization in Fig.
l-8(b), each element is of the same size for equally spaced nodes. If it is desirable
to assign nodes that are not equally spaced, the FEM approach is more easily
adapted.
More importantly, in cases in which the geometry of the region being analyzed
is irregular, as shown in Fig. 1 -9(a), the triangular and quadrilateral elements are
routinely used, and it is seen that the plate boundaries are approximated quite well

lumps fora finite-difference analysis and

Figure 1-8 Rectangular plate discretized into (a)

(b) rectangular elements for a finite element analysis.

10 OHM RAL COSCLPTS

with these elements. Note that different element types may be used in the same
model, as illustrated here by the use of the triangular, rectangular, and quadrilateral
elements. On if the same plate is discretized into lumps for an FDM
the other hand,
analysis, as shown in Fig. l-9(b), then the effectne plate boundary becomes jagged
as shown. The alternative is to write special finite-difference equations for these
nodes that explicitly include the effect of the curved boundary (7J. Unlike the
formulation in FEM, the FDM formulation is nontrivial when curved boundaries
arc present.
The different methods of discretization in FEM and FDM, however, are not
the only significant differences between the two types of formulations. In FDM, a
basic law is written for each node, such as the first law of thermodynamics (con-
servation of energy) or Newton's second law of motion. For example, in a heat
transfer formulation, an energy balance is made on each lump. This is relatively
straightforward (except when curved boundaries arc present).
In FEM, although a direct energy balance approach can be used, other more
powerful and more popular approaches are taken such as the principle of virtual

work, variational methods, and weighted-residual methods These various methods

of FEM formulations arc developed in detail throughout this text For structural
and stress analysis problems, the principle of virtual work will usually be used,
whereas for nonstructural applications, the method of weighted residuals will be
used almost exclusively The reasons for favoring the method of weighted residuals
over variational formulations will become evident in Chapters 4 and 8 As the name
implies, the variational formulations require knowledge of variational calculus, an
obvious disadvantage

1-5 ADVANTAGES AND DISADVANTAGES

OF THE FINITE ELEMENT METHOD [8]

As mentioned in the previous section, one of the main advantages of FEM over

most other approximate solution methods, including the popular finite-difference

Figure 1-9 Irregularly shaped plate shown discretized into (a) rectangular, triangular. and
quadrilateral elements and (b) lumps
 ADVANTAGES, DISADVANTAGES OF FINITE ELEMENT METHOD

method, is the fact that FEM can handle irregular geometries routinely. The tri-
angular element in two-dimensional applications is used with no special consid-
erations. Of course, closed-form analytical solutions are invariably nonexistent when
irregular geometries are present.
Another significant advantage of FEM is that a variable spacing of the nodes
is also routinely handled. When a body is discretized using finite elements (in FEM)
or lumps (in FDM), the nodes are said to form a mesh. Typical two-dimensional
meshes were shown in Figs. 1-8 and 1-9. When the nodes are not equally spaced,
the mesh is said to be graded. The finite element method lends itself to the use of
graded meshes. It will be seen in later chapters that special subroutines can be
written that will automatically or semiautomatically generate unequally spaced nodes,
thereby reducing the amount of input required by the user of such a computer
program.
Another advantage of FEM, again over FDM and especially over analytical
solution techniques (as opposed to numerical techniques) is the ease with which
nonhomogeneous and anisotropic materials may be handled. Materials whose prop-
erties are not spacially dependent are said to be homogeneous, whereas materials
with spacially dependent properties are heterogeneous. A special case of a heter-
ogeneous material in a thermal analysis is one for which the thermal conductivity
is temperature-dependent. The reason for this is simple: the thermal conductivity
is a function of temperature, the temperature is a function of spacial coordinates,
and, therefore, the thermal conductivity must also be a function of spacial coor-
dinates. Concrete is an example of a heterogeneous material.
Materials may also be classified as isotropic and anisotropic. An isotropic
material is one whose properties (Young’s modulus, thermal conductivity, etc.) do
not exhibit a direction sensitivity. For example, even though concrete may be
nonhomogeneous, each direction appears to have the same (rather random) variation
in thermal conductivity and, therefore, it is isotropic. Anisotropic materials, on the
other hand, will have one or more properties that are direction-dependent. For
example, a laminated metallic structure quite frequently will have different values
of certain properties, such as Young’s modulus or thermal conductivity, in different

directions. Wood is another example of an anisotropic material; it is generally

stronger in the direction of the grain and hence would have a higher value of
Young’s modulus in this direction. Very little extra effort is required in the FEM
formulation when heterogeneous and/or anisotropic materials are to be modeled,
even when some parts of the structure or body are made of one material and other
parts are made of different materials.
All the various types of boundary conditions that we may encounter in a typical
FEM application except those that require prescribed values of the held variables
themselves, are automatically included in the formulation. Recall that typical field
variables are displacements in structural and stress analysis, temperatures in thermal
analysis, fluid velocities and pressures in fluid flow analysis, etc. In other words,
prescribed displacements, temperatures, velocities, pressures, etc., are not auto-
matically included in the FEM formulation and solution. They are systematically
enforced just before the solution for the nodal values of the unknown field variables
12 OHM RAL COSCI ITS

is obtained, as discussed in subsequent chapters. Among the boundary conditions

that are automatically included are the following:

Structural and stress anal) sis

Concentrated surface forces
Distributed surface forces (tractions)
Thermal analysis (heat conduction)
Convection
Radiation
Applied heat fluxes
Insulation
Fluid flow analysis
Pressure gradients
Velocity gradients

Jn all cases, these conditions need not be constant When these boundary
conditions or other properties arc a function of the (unknown) field variable, the
problem becomes nonlinear Special solution techniques, one of which is discussed
in Chapter 8, must be applied in these cases The basic finite element method is

applicable, however, for both linear and nonlinear problems

Another advantage is that higher-order elements may be implemented with
relative case Several higher-order elements arc shown in Fig. I- 10 Higher-order
elements require the use of higher-order interpolating polynomials Note that ad-
ditional nodes are introduced along the sides of the two-dimensional elements and
between the two end nodes of the one-dimensional clement In fact, by using
isoparametric elements defined in Chapter 9. curved sides ma) actually be used
thereby allow ing very close fits to essentially all irregular geometries Occasionally,
interior nodes are introduced as shown in Fig. 1-10. The use of these nodes requires
special considerations as discussed m Chapter 9
Among the disadvantages of FEN! is the necessity for a digital computer and
fairly extensive software. In fact, the MacNcal-Schwcndler version of NASTRAN
1 1 1 has more than 430,000 lines of FORTRAN source code Of course, such a large
general program need not always be used More specific and significantly smaller
programs may be all that arcneeded in some applications The programs included
in the appendices of this book arc provided mainly for educational purposes, but
on occasion they may be used to solve small practical problems
A
couple of obvious questions have probably arisen The first is whether or
not the finite clement method is more accurate than the finite-difference method
We cannot answer this question without a significant number of qualifications, hut
FEM seems to be more accurate when curved boundaries arc present By referring
to Fig. 1-9 again, we can see why this is not really surprising the finite element
method can usually include the curved boundary more precisely than the finite-
method can
difference
The second question that nuy be asked pertains to the relative execution times
of typical and comparable FDM and FEM models Once again it is difficult to
make comparisons, but FEM seems to have longer execution times. However, it
 BRIEF HISTORY OF FINITE ELEMENT METHOD 13

(a) (c)

Figure 1-10 Representative higher-order elements (a) One-dimensional elements, (b)

Two-dimensional triangular elements with straight sides (c) Two-dimensional triangular
element with curved sides, (d) Two-dimensional rectangular elements with straight sides,
(e) Two-dimensional quadrilateral element with curved sides.

must be said that FDM models will generally require more hours of computer input
preparation by people than will FEM. As will be seen in subsequent chapters, many
of the tedious calculations in FEM are relegated to the computer. For example, in
a thermal application, internal conductances need not be calculated in FEM models
because they are inherently included in the computer software. On the other hand,
FDM thermal models quite frequently require the user to calculate many of the
internal conductances by hand. This will become an even more significant advantage
of FEM as engineering labor costs rise and computational costs decline.

1-6 BRIEF HISTORY OF THE FINITE ELEMENT METHOD

An early but crude application of finite elements was presented in Sec. 1-3 where
the value of -it was determined by the method of exhaustion. However, the finite
14 Cl M RAL CONCTTO

clement method used today was developed to its present state very recently. Ac-
cording to Zienkicwicz [9], the development has occurred along two major paths:
one in mathematics and the other in engineering. Somewhere in between these two
extremes arc the variational and weighted residual methods, both of which require
the use of trial functions to effect a solution. The use of these functions dates back
almost 200 years.
A trial function is an assumed mathematical function that is usually based on
physical intuition, is applied globally to the region being analyzed, and approximates
the expected behavior of the region to some external forces. These functions are
used in various types of integral formulations [10], described in detail in Chapter
4. A qualitative example may help to clarify these abstract points

Let us consider again the heat conduction problem in a rectangular plate. If

the temperature is known to be zero along the boundaries of the plate and a maximum
at the center based on physical reasoning, then one or more trial functions could
be chosen that exhibit this known trend Such a method is rather limited for two
reasons (1) the trial functions must satisfy the boundary conditions exactly, and
(2) the trial functions must be chosen based on physical reasoning In other words,
it is very difficult to apply this method to any but the simplest of problems
As indicated earlier in this chapter, the finite clement method is predicated
along the same lines with an important exception the trial functions arc not applied
globally to the entire region being analyzed but only locally (i e , on an element
basis) This point is of paramount importance, since the method becomes applicable
to real problems with irregular geometries and unknown solutions This important
step did not occur until 1943 when Courant |1 Ij introduced piecewise continuous
trial functions But this is getting ahead of the story, which begins in 1795
As mentioned earlier, the use of trial functions is associated with neither the
purely mathematical nor the engineering developments In a paper by Gauss 1 12}
in 1795, Inal functions (on a global basis) were used in what is now called the
method of weighted residuals Later, these functions were used in variational meth-
ods by Rayleigh [13J in 1870 and by Ritz [ 14 in 1909 In fact, any variational-
j

based approximate solution method is frequently referred to today as the Raylcigh-

Ritz method In a landmark paper in 1915, Galerkin [15] introduced a particular
type of weighted residual method, which to this day bears his name, the Galerkin
method. However, all of these early uses of trial functions were done on a global
basis.
In 1943, Courant [I I] introduced ptcceuise i ontmuous trial /unit ions As the
name suggests, these functions were not applied globally to the entire region being
analyzed but rather to many small regions or elements By using trial functions on
a local basts, Courant greatly extended the applicability of trial functions in obtaining
approximate solutions to real-world problems Many feel that the finite clement
method has its roots in this paper by Courant. although he did not introduce the
terminology finite element
Along more mathematical lines, the finite-difference method was introduced
bv Richardson [16] in 1910 and later improved by Ltebman [17] in 1918 and
Southwell ( 1 8| in 1946 Variational finite differences were developed by Varg3
 REMARKS 15

[19] in 1962. These mathematical developments lead directly to the present-day

finite element method, and it may now be said that the finite-difference method is
a special case of the more general element method [20],
finite

In the early 1940s, aircraft engineers were developing and using analysis meth-
ods that are now recognized as early forms of the finite element method. The first
applications used the so-called force matrix method (also known as the redundant
force method). In this method, the nodal unknowns are the forces, not the dis-
placements. When the displacements of each node are taken as the unknowns, the
method is called the stiffness method.
However, the stiffness method was not developed until 1953 in an important
paper by Levy [21], Other key contributors were Schuerch [22] also in 1953; Argyris
and Kelsey [23] in 1955; and Turner, Clough, Martin, and Topp [24] in 1956. In
a paper in 1960, Clough [25] first introduced the term finite element. Readers
interested in the historical details of the development of the finite element method
along the structural engineering lines are referred to the book by Martin and Carey
[26] in which a fascinating account is given.
The application of the finite element method to nonstructural problems has an
equally interesting history. Among the first to apply the method to general problems
described by Laplace’s and Poisson’s equations were Zienkiewicz and Cheung [27]
and Visser [28] in 1965. The application at this particular time happened to be
conduction heat transfer, but it was immediately recognized that the procedure was
applicable to all problems that could be stated in a variational form.

Other researchers, such asSzabo and Lee [29J, showed how the method of
weighted residuals, particularly the Gaierkin method, could be used in the study
of nonstructural applications to retain the basic finite element process. Zienkiewicz
[30] in 1971, in a second edition of an earlier book, was evidently the first to
include in one book the general applicability of the finite element method to problems
describable by ordinary and partial differential equations, or field problems in
general. Field problems include all the problems normally associated with the
continuum. Zienkiewicz’s third edition [31] of the same book in 1977 is an out-

standing example of the level of maturity that the finite element method now enjoys.
Since the second edition of Zienkiewicz’s book, numerous other books have ap-
peared; those that have not already been cited are included in references 32 to 46.
Interestingly, the present-day finite element method does not have its roots in

any one discipline. The mathematicians continue to put the finite element method
on firm theoretical ground, and the engineers continue to find interesting applications
and extensions in many branches of engineering. These concurrent developments
have made the finite element method one of the most powerful of the approximate
solution methods.

1-7 REMARKS

In this chapter some of more important introductory concepts were developed.

the
introduced by contrasting it to the more familiar
The basic finite element method was
 .

1 6 Cl M-RAL COSCLPTS

finitc-diffcrencc method. The notion of discretization, which is critical to the use

of the finite element method, was introduced. The advantages and disadvantages
of the method were also discussed. The use of trial functions, particularly on a
local (or element) basis, was also introduced.
It cannot be emphasized enough that it is the notions of discretization and the
use of localtrial functions (or interpolating polynomials) that make the finite element

method applicable to real-world problems with irregular geometries, heterogeneous

and anisotropic materials, mixed boundary conditions, and so forth. Although these
concepts may seem abstract to the uninitiated, they will undoubtedly become much
more familiar in subsequent chapters
The reader will be exposed further to discretization beginning with Chapter 3
and to global and local trial functions beginning with Chaptcr4, w here the variational
and weighted residual methods are also formally introduced Because the direct
approach is taken, trial functions arc not needed in Chapter 3 in which the two-
dimensional truss model is developed Our study begins in Chapter 2 with a review
or some of the more important mathematics needed in the development of the truss
model.

REFERENCES

1. MSCINASTRAN User . s Manual. MacNcal-Schwcndlcr Corp ,

Los Angeles. 1979
2. STARDYNE for Sc ope 3 4 Operating Ssstcm. User Information Manual. Cjbcmet Ser-
vices. Control Data Corp . Minneapolis. Minn . 1978
3. Kolbc. R L . and K K Kimble. 'The Development and Use of a Two-dimensional
rmite Element Program for Incompressible-Viscous Flow Through a Complex Internal
Geometry.” Congr Technol Technol Exchange. Pittsburgh. Pa Ma> 3-6. 1982
hit ,

4. Macropaedia. "Eudoxus of Cnidus.' Em u lopaedia Dnnmnii a. sol G.pp 1021-1022,

19X1
5. Macropaedia. "Archimedes.’ Etu \ c lopaecha Britannic a, v ol l.pp 10X7-1088. 1981
FORTRAN SO User s Guide Microsoft FORTRAN -SO for the
6. Demonstration Program.
Computer Microsoft Consumer Products. Bellevue. Wash l> 1980
Apple If ,

7. Mjcrs, G L Anahiual Methods in Conduction Heat Transfer, McGraw-Hill. New

.

York. 1971. pp 310-313

8. Zicnkicwicz, O C The Finite Element Method. McGraw-Hill (UK). London. 1977,
.

pp a 30-432
9. Zicnkicwicz. O C The Finite Element Method. McGraw-Hill (UK). London. 1977.
.

pp 3-4
10. Arpaci. V S Conduction FIeat Transfer. Addison-Wcslej Reading. Mass
. 1906. pp . .

59-83.
11. Courant. R . ’‘Variational Methods for the Solution of Problems of Equilibrium jnd
Am Math Sen vol 49, pp 1-23. 1943
Vibrations,'' Bull

12. Gauss, C. F., Carl Friedrich Gauss H erLs, sol. VII, Gottingen. German), 1871
13. Lord Ra)!cigh, ”On the Thcor> of Resonance.’* Trans Ros Soc (London I vol AI6I.
pp. 77-118. 1870
14. Ritz, W., “Uber erne neuc Mcthode zur Losung gewisscn Variations Problcmc dcr —
mathcmatischen Phvsik," J Retne Anges* Math., vol 135. pp 1-61.1909.
 ,

REFERENCES 17

15. Galerkm, B G , "Series Solution of Some Problems of Elastic Equilibrium of Rods

and Plates" (in Russian), Vestn
pp 897-908, 1915 In:h Tekh vol 19,
16. Richardson, L F "The Approximate Arithmetical Solution by Finite Differences of
,

Physical Problems,” Turns Rox Soc (London), sol A210,

pp 307-357, 1910
17. Liebman. H “Die angenaherte Ernuttlung harmomschen, functionen und konformer
,

Abbildung,” Sitzungsber Math Phwtk Kl Ba\ei Akad IViss Muchen, vol 3, pp

65-75,1918
18. Southwell, R V .Relaxation Methods inTheoi ettcal E/nsics, Clarendon Press, Oxford,
England, 1946
19. Varga, R S , Matitx Itetatne Analxsts, Prentice-Hall. Englewood Cliffs, N J , 1962
20. Zienkiewicz, O C , The Finite Element Method, McGraw-Hill (UK), London, p 569,
1977
21. Levy, S , “Structural Analysis and Influence Coefficients for Delta Wings,” J Aero
nant Sci , vol 20, no pp 449-454, July 1953
7,
22. Schuerch, H U , "Delta Wing Design Analysis,” Nat Aeion Meeting, Sol Automot
Eng , Los Angeles, Paper No 141, September 29-October 3, 1953
23. Argyris, J H , and S Kelsey, Eneigx Theoieim and St) uctmal Analxsts, Butterworth,
1960 (reprinted from An a aft Eng , 1954-1955)
24. Turner, M J , et al , “Stiffness and Deflection Analysis of Complex Structures,” J
Aeronaut Sci pp 805-824, September 1956
, vol 23 ,
no 9,
25. Clough. R W
"The Finite Element in Plane Stress Analysis.” Pioc 2nd ASCE Conf
,

on Electron Computation Pittsburgh, Pa September 1960 ,

26. Martin, H C , and G F Carey, Introduction to Finite Element Analysis Theoix and
Application, McGraw-Hill, New York, 1973, pp 1-13
27. Zienkiewicz, O C ,
and Y K Cheung, “Finite Elements in the Solution of Field
Problems,” The Engmeei, 1965, pp 507-510
28. Visser, , W
“A Finite Element Method for the Determination of Nonstationary' Tern
perature Distribution and Thermal Deformations,” Plot Conf Math Methods Stmc
Mech , Air Force Institute of Technology, Dayton, Ohio, 1965
29. Szabo, B A , and G C Lee, “Derivation of Stiffness Matrices for Problems in Plane
Slasticity by Galerkin’s Method,” hit J Numet Methods Eng ,
vol Lpp 301-310,
1969
30. Zienkiewicz, O C , The Finite Element Method in Engineering Science, McGraw-Hill
(UK), London, 1971
31. Zienkiewicz, O C ,
The Finite Element Method, McGraw Hill (UK), London, 1977
32. Desai, C S , and J Abel, Inti oduction to the Finite Element Method, Van Nostrand-
Reinhold, New York, 1971
33. Oden, J T Finite Elements of Nonlinear Conlinua, McGraw-Hill, New York, 1972
,

34. Gallagher, R H Finite Element Analysis Fundamentals, Prentice-Hall, Englewood

,

N J 1975
Cliffs, ,

35. Nome, D H and G , de Vries, The Finite Element Method, Academic, New York,
1973
36. Robinson, J , Integrated Theorx of Finite Element Methods, Wiley, London. 1973
37. Huebner, K H The Finite Element Method for Engineers, Wiley, New York, 1975
,

38. Bathe, K -J ,andE L Wilson, Numerical Methods in Finite Element Analysis, Ptentice-
Hall, Englewood Cliffs, N J , 1976
39. Segerhnd, L J , Applied Finite Element Analysis, Wiley, New York, 1976
40. Tong, P , and J N Rossettos, Finite-Element Method Basic Technique and Imple
mentation, MIT Press, Cambridge, Mass ,
1977
 .

18 Gl-M-R AL CONCI JTS

4 1 . Norris, D. H.. and G. dc Vries, An Introduction to f mite Element Anahsis, Academic,

New York, 1978
42. Hinton, E.. and D. R. J. Owen. An Introduction to hmte Element Computations,
Pincndge Press Limited. Swansea. UK. 1979.
43. Cheung, Y. K., and M. F. Yeo, A Practical Introduction to Finite Element Anal\sis,
Pitman, London. 1979
44. Desai, C. S , Elementars Finite Element Method, Prentice- Hall. Englewood Cliffs,
N J , 1979.
45. Becker, E. B G , F. Carey, and J. T Oden, Finite Elements- An Introduction, sol 1

Prcnticc-Hall, Englewood Cliffs. N J., 1981.

46. Bathe, K.-J., Finite Element Procedures in Engineering Anahsis, Prentice-Hall, En-
glewood Cliffs, N.j.. 1982
 Mathematical Preliminaries

2-1 INTRODUCTION

Before actually developing the truss mode! in Chapter 3. a review of the necessary'
mathematics is given here. In all practical finite element analyses, matrix and vector
notation must be used because of the large number of variables and computations
involved.
This mathematical review includes the following topics relating to matrices
and vectors: the definitions of matrices and vectors: matrix and vector equalities;
matrix and vector algebra; transpose, determinant, cofactor, minor, adjoint, and
inverse of a matrix: matrix partitioning; length of a vector, direction cosines; and
solutions to linear systems of equations. Additional review material is presented at

appropriate places throughout the text, usually immediately preceding the appli-
cation of the topic. The matrices and vectors we discuss are always assumed to be
composed of only real elements unless explicitly stated otherwise.

2-2 DEFINITION OF A MATRIX

A matrix is a rectangular or square array of elements or numbers. The use of the

word “element” should not be confused with its use in the phrase “finite element.”
The context usually makes the usage clear. An element in a matrix is simply an
entry in the matrix as explained below. The notion of a matrix is more easily
understood by means of a linear system of equations. For example, the system of
equations
20 MVTHtMVnCM.rRLUMI'.MUtS

filial + fl««2 + «|l'» = /’j

^ ^ ( 2-1 )
0»I^| + 022*2 + — ,
0T

could be written in the compact matrix form

Ax = b (2-2)

where by definition

The array in the matnx A is a matrix of coefficients The size (or

of elements
dimension) of A said to be 2 x 3 (or 2 by 3) Note that first the number of rows
is

is number of columns is gi\en The matrix x contains three rows

gisen, then the
and one column, and the matnx b two rows and one column. Hence, the matnees
x and b arc said to be of size 3 x and 2x1. respccnvel>
1

A column vector is defined as a matnx with only one column Therefore, \

and b are also referred to as column sectors A row \ector is a matnx with only
one row An example of a row sector is gisen b> the 1 x 4 matnx

|l 3 -2 0)

These notions arc readily generalized An m x n matnx is a rectangular arras

of elements with m rows and n columns If m is equal to n. the matnx is said to
be square. Quite frequently each entry in a matnx may be an algebraic expression,
.

a denvatise, or possibly an integral, it need not always be a pure number A column

sector is an n x array of elements, whereas a row sector is a
I x n array After I

the transpose of a matrix is introduced, the row sector will always be wntten in
terms of column lector representation (see Sec
its 2-5)
An clement or entryin the nh row and yth column of an m x n matnx A is
written o,
r The
first subsenpf denotes the nh row and the second subsenpt the yth

column, where I t *s m and l « j ?= « The ith entry in an n x I column sector

b may be wntten as h,. where 1 SiS»
Similar remarks hold for a I x n row
sector. In general, matnees (where neither m nor n is unity) are denoted ssuh
uppercase English and Greek boldface letters. Column (and toss ) sectors arc denoted
with lowercase English and Greek boldface letters.

Two special types of square matnees arc the identity matnx and the null matnx
The identity matnx is denoted as I or and has on Us main diagonal and
unity

zero elsewhere For example, the 3 x 3 identity matnx is gisen by

The main diagonal always runs from the upper left to the lower right. The null
 MATRIX ALGEBRA 21

matrix is denoted as 0 or 0„ x „ and, as its name suggests, has only zero elements
or entries. The 2x2 null matrix is given by

0 0
0 2x2
0 0

The equation Ax = b contains the notions of matrix equality and matrix

multiplication. Matrix equality is very easy to define and understand and is reviewed
next.

2-3 EQUALITY OF TWO MATRICES

Two m x n matrices A and B are said to be equal if and only if the corresponding
elements of A and B are equal. Note that A and B must be of the same size. Stated
differently, two m X n matrices A and B are equal if and only if ay — b tJ
for
1 =S / m and l =£ j =£ n.

Example 2-1

Consider the four matrices A, B, C, and D given below and determine which of
the various combinations are equal (taking two at a time):

6' '3
2> 1 f
A = 2 -2 8 B - 2 -2
J 0 7 J
~3 '3
f f
C = 2 -2 D = 8 -2
1 0 J 0

Solution

Only B, C, and D can possibly be equal because they are at least of the same size.
Each element in B is equal to the corresponding element in C; hence B = C.
However, B ^ D because b2l = 2 and dzt = 8. Since these two elements are not
equal, the corresponding matrices B and D cannot be equal. Similarly, C/D.
Therefore, only B and C are equal. I

2-4 MATRIX ALGEBRA

of matrix addition and subtraction are discussed. This is

In this section the notions
followed by a review of the two types of matrix multiplication and finally a summary
of the commutative, distributive, and associative laws for matrices.
22 UATf fCMATICAi. PRl LmiVARII-S

Matrix Addition and Subtraction

Matrix addition (and subtraction) is performed by adding (or subtracting) the cor-
responding elements of two m x n matrices to form a third m x n matrix. Let A
and B be two m x n matnees and let the m X n matrices C and I) be defined by

C = A + B and D = A - B
The matrices A and B are said to be conformable for addition (and subtraction)
because thc> both have the same number of rows and columns. From the above
definitions, these may be written on an element basis as

c4J = a„ + b„
and for I < <m
t and 1 ^j ^ n
‘i,, = «,
- i>,i .

Example 2-2
For A and B given by

determine A + B and A - B

Solution

First, both A and B arc of the same size so the matrices ma> be added and subtracted
The results are

6 + 2 -1+0 2 + 5
0 + (-8) 3 + 3 5 + 4

6-2 -1-0 2-5'

0 - (-8) 3-3 5 - 4

Matrix Multiplication

Two types of multiplication may be defined. (1) multiplication of a matrix b> a

scalar and (2) multiplication of two (or more) matrices The former is simpler and
is defined first.

Multiplication of a Matrix by a Scalar

Let j be an) real scalar and A any m x n matrix The product of r and A. written
jA. is obtained by mu!iipl>ing every element of A b) j. In other words, the elements
 MATRIX ALGEBRA 23

of xA are sa tj . Note that the product is of size m X n (such as A itself). This type
of matrix multiplication is commutative; that is, sA = As.

Example 2-3

For the matrix A given in Example 2-2, determine the matrix C defined by 6A.

Solution

6 - 1 2 36 -6 12
0 3 5 0 18 30

Multiplication of Two Matrices

The product of an m x p matrix A and a q x n matrix B, written AB, is only

defined when p equals q. Furthermore, the matrices A and B are said to be con-
formable for multiplication when p equals q. The matrix A is said to premultiply
B, and the matrix B to postmultiply A.
Let an m X r matrix A premultiply an r X n matrix B to form a third (m X n)
matrix C. The elements of C, or the c,/ s, are given by
r

c t] = 2 a, L b, (2-3)
*= 1

for 1 i =£ m and 1 =£ y =£ n. It can be shown that this type of matrix multiplication

is not commutative in general; that is, AB # BA. In fact, the product BA may not
even be defined if B and A are not conformable for multiplication (even though A
and B, in that order, are conformable). Another way of checking whether or not
two matrices A and B are conformable for multiplication is to check if the number
of columns of A is equal to the number of rows of B; if they are, AB is defined.
A similar test may be stated for the product BA.

Example 2-4
Given the two matrices A and B, determine if AB and BA are defined; if they are,
compute them. The matrices A and B are

0
-4 7
A = 4 and B
5 0
3

Solution

A is of size 3x2 and B is 2 x 2. Therefore, AB is defined and is of size 3x2

by 2 X 2 or 3 x 2. Using Eq. (2-3), we get

— 4) + (2)(7) + (0)(0)
'-8 14'
(2)( (0)(5)
AB = (6)( — 4) + (4)(5) (6)(7) + (4)(0) = -4 42
( — 5)( — -35
4) + 35
<-5)(7) 4- (3)(0)
(3)(5)
24 M ATI It MAT1CAL PKLLIMIN \RI15

On ihc other hand. BA is not defined because the number of columns of B (in
this case two) is not equal to the number of rows of A (in this case three), The
matrices A and B are conformable for multiplication, but B and A (in this order)
arc not.

The Commutative, Distributive, and Associative Properties

Let A, B, and C be matrices and let s be a scalar. In the laws summarized below,
it is assumed that the matrices arc conformable for addition and multiplication as
the situation warrants.

Commutative laws: A + D = B + A (2-4)

s\ = As (2-5)

Distributive laws: s(A + B) = sA + sB (2-6)

A(B + C) = AB + AC (2-7)

(A + B)C = AC + BC (2-8)

Associative laws A - B = A + - B) ( (2-9)

(A + B) + C * A + (B + C) (2-10)

(Afl)C = A(BC) (2-11)

The proofs of these laws are relatively straightforward and arc left to the reader as
exercises (sec Problems 2-18 to 2-20). It should be remembered that the product
of two matrices is not commutative, that is, AB ^ BA

2-5 TRANSPOSE OF A MATRIX

The transpose of an mA with elements is an n x m matrix denoted

x n matrix
as A r (and smed A whose elements are given by
transpose) The switching of
the subscripts should be noted The transpose of a product of moor more matrices
occurs often and an important property is given by

(AB) r « IW (2-12)

(ABC)' = C r B TA r 12-13)

and so forth If A = Ar (or a,, = a„). the matrix A is said to be ssmmelnc If

A= - A' (or <i„ = —0p)* A is skew-ssmmetrtc A skew -symmetric matrix must

have zeros on its mam diagonal Symmetric and skew -symmetric matrices must be
square. The transpose of a column vector is a row sector, and vice versa From
this point on a lower case boldface letter will always represent a column vector
The transpose will always he used to indicate a row sector (sec L sample 2-JO)
 DETERMINANTS. MINORS. AND COFACTORS

Example 2-5

Determine the transpose of

2 - I

A = I 9
6 5

Solution

Since A is 3 x 2. the transpose is 2 X 3 and is given by

2 1 6
AT = -] 9 5

Example 2-6
Determine if the matrices A, B, C defined below are symmetric, skew-symmetric,
or neither:

0 - 1 3 4
-1 6
1 0 -6 -7
A = 2 0 B —
-2 6 5 5
0 5
4 7 -5 0 .

0 1
-6'
C = I 0 -3
6 3 0

Solution

Since a tJ = a for i - 1 to 3 and j =

,,
1 to 3, the matrix A is symmetric. Since
b3 1
b , 3 and b 3x ^ -fc 13 the matrix B ,
is neither symmetric nor skew-symmetric.
Finally, since c,, = -c,, for i = l to 3 and j - 1 to 3, the matrix C is skew-
symmetnc. Symmetric matrices arise quite frequently in finite element analysis.

2-6 DETERMINANTS, MINORS, AND COFACTORS [1]

Consider an n x n matrix A. The determinant of A, written det A, is a scalar

function of A which for n = 1,2, and 3 is given by

det A s , ,
= au (2-14a)

det A 2 yT = «n«2Z ~ a na n (2-1 4b)

detA 3x3 = fl|i fl22 a 33 _ a u a 23 a 32 + a l2 a 23 a 3l

— (121012033 "F a 13032 a 21 ^31^22^13 (2-14c)

26 MVTIH SfATfCAL FRrf IWVARILS

Although this pattern can be generalized, it is not a very practical way to evaluate
determinants for « larger than 3. Fortunately, another method exists called the
Laplace expansion method. This method is most easily explained after the minor

and cofactor of a matrix arc defined.

The minor M
of a square matrix A is the determinant of the n —
tJ
X n — I I

matrix derived from A by eliminating the ith row and the 7 th column of A. Such
as the determinant itself, the minor is also scalar-valued and is only defined for
square matrices An 11 x n matnx A contains ir minors.
The cofaitor C of a matrix A is easily defined in terms of the minor v by
tJ
M
CV = {-IY*'M„ 12-15)

Note that when 1 + j is even, the cofactor and minor are identical, when 1 + j is

odd, the cofactor and minor differ only in the signs An n x n matrix A has ir
cofactors.
The determinant of the original n x n matrix A may now be computed utilizing
the notion of the cofactor by selecting any arbitrary column j and using

Jet A = X !h,Ch 12-16)

An equivalent expression for evaluating the determinant is given by

dot A = X 12-17]
m= I

where row / is selected for the expansion Although these expressions look for-
midable, they are not difficult to apply in practice, as shown in Example 2-7
Equations (2-16) and (2-17) are used repeatedly until determinants of matrices of
size 2x2 need to be evaluated, at which point Eq (2-l4b) is used (see Example
2-7). This method of evaluating the determinant of a matrix is known as the Laplace
expansion method

Example 2-7

Using the Laplace expansion method and expanding with respect to the second row,
find the determinant of A where

Solution

The three minors associated with the second row arc

 1

ADJOINT AND INVERSE OF A M ATRIX 27

Each of these 2x2 determinants may be evaluated with the help of Eq. (2- 14b)
to give

Mzt = 0 M 22 = -12 M Z3 = 8

The corresponding cofactors are

C2I = (-I)' +l (0) = 0 Cn — - l)

2+2 (— = -12
( 12)

and
"3
C23 = (-D 2 (8) = -8
Therefore, the determinant of A is given by

detA = (3)(0) + OX-12) + (~5)(-8) = 28

Try to evaluate the determinant of A by expansion with respect to another row or

column. 0

2-7 ADJOINT AND INVERSE OF A MATRIX [1]

The adjoint (adj) of an n x n matrix A, denoted as adj(A), is easily defined in

terms of the transpose of the cofactor matrix C. Stated mathematically,

adj (A) = Cr (2-18)

The adjoint itself is another n x n matrix and is particularly useful in providing a

concise expression for the inverse of a matrix as described below.
The inverse of an n X n matrix A, denoted as A -1 (and stated A inverse), is

another n x n matrix such that

AA _I = A“'A = f

where I is the n x n identity matrix. In other words, the inverse of an n X n

matrix A is another n x n matrix such that when it is both premultiplied and
postmultiplied by A itself, the result is the n x n identity matrix. In terms of the

adjoint, the inverse of A is given by

adj (A)
= ( 2 - 19 )
det A
If the A is zero, then A -
determinant of does not exist and A is said to be singular.
When the determinant of A is nonzero, the inverse of A is unique and A is said to
be nonsingular. The inverse is only defined for square matrices. When Eq. (2-19)
is applied to a nonsingular 2x2 matrix A, the inverse is given by

1 Cl 22 Cl
A -1 = 12
( 2 - 20 )
det A -021 «!l
28 MATHfM VT1CAL FRTUMIN ARILS

where det A is given by Eq (2- 1 4b). A useful identity is given by

(AB)-‘ = B-'A' 1
{2-211

providing A and B are both square, of the same size, and nonsingular. Similarly,
if in addition C is also square, of the same size, and nonsingular, then

(ABC)-* - C-'B-'A -1 (2-22)

The identities given by Eqs (2-21) and (2-22) may be readily extended to products
-1
involving four or more matrices If A = A r
, then the matrix A is said to be
orthogonal Orthogonal matrices will anse quite natural!} in Chapter 3.
Actually, Eq (2-19) is not a very practical A moreway to evaluate an inverse.
practicalmethod is based on Gauss-Jordan elimination (see Chapter 6) In addition,
computer programs or subroutines abound that may be used effectively to compute
inverses of rather large matrices One such subroutine is given in Appendix B
(subroutine 1NVDET) [2J

Example 2-8

Evaluate the inverse of A where

Solution

By Eq. (2-20), A' 1

is computed as follows

-8
A" 1
= !
1
(2»4» - ( - 3X81 [3 2j

Example 2-9

Compute the inverse of A which is given by

Solution

Since del A = 0. A 1
does not exist The matnx A is singular

2-8 MATRIX PARTITIONING

It is always po->sibIe to subdivide or partition matnees into a number of smaller

submatnees. If the submatnees arc conformable, they may be treated as though
they are scalar elements when adding, subtracting, or multiplying two matnees.
 MATRIX PARTITIONING 29

The order of the products is not arbitrary, however, as it is if the elements are truly
scalars. For example, let us compute the product of A and B to form C in two
different ways:

B,
C = AB = [A, i A2 1
A|B, + AtBj
B,

j
^12 AjjBj + A12B2 c,-
“
A2 i ! Aoi bT A21B1 f A22B2 c2
The subdivisions of A and B must be chosen in order to ensure that the products
such as A,B] and A n B, are meaningful (i.e.. the submatrices must be conformable).
Stated differently, the number of columns in Ai must equal the number of rows in
B,, etc. Matrix partitioning is not introduced here to provide an alternate means of
multiplying two matrices. In Chapter 3, matrix partitioning arises quite naturally
when the element stiffness matrix is developed for a two-dimensional truss.

2-9 DEFINITION OF A VECTOR

Let us initially restrict our definition of a vector to that in a three-dimensional space

in which a vector is a directed line segment with a unique length (magnitude) and
direction as shown in Fig. 2-1. It is convenient to resolve a vector into its three
components along three mutually perpendicular axes. These axes establish a co-
ordinate system, and since they are mutually perpendicular they are said to be
orthogonal. For example, a vector v may be written with respect to a cartesian
reference frame in terms of its three components vt , v,, v. as

v = vri + i\j + v-k

where i, j. and k are the three unit vectors along the x, 3’, and z axes, respectively.

A unit vector has a length of unity (see Sec. 2-1 1 ) and is directed along its associated
coordinate axis. Figure 2-1 shows a cartesian reference frame and the three unit
vectors.
A vector is often written in column matrix form as mentioned in Sec. 2-2. For
example, the vector v may be written as

*’r Vl

V = v> = v2
_V-_ _v3 „

Often numerical subscripts, as shown above, prove to be more convenient than

letter subscripts. In this text, a matrix represented by boldfaced lowercase letters

(e.g., v) willdenote a column matrix (or simply a vector). In order to indicate a

r
row matrix, the transpose will be used. In other words. v represents a row matrix,
whereas v represents a column matrix (or a vector). There are two reasons for this
30 MATHEMATICAL PRELIM] S ARIES

Figure 2-1 Vector v in a cartesian reference frame

convention, the first is that it removes any ambiguity in matrix expressions developed
later, and the second is that it conserves space in writing the vector out, as the next

example illustrates It is important to recognize that all vectors may be written in

matrix form (as a column matrix), but not all matrices have a vector meaning.

Example 2-10
Write v in matrix form where

v = li - 4j + 3k

Solution

In column matnx or column vector form,

[
which in turn may be written as

= (1 -4 3]

or as

v = [I -4 3)
T

Clearly these all convey the same information, but the second two forms require
fewer lines.

Although the notion of a vector was introduced from a geometrical point of

view', vectors need not have only three components In later chapcrs, vectors with
four or more components will naturally arise
 EQUALITY OF TWO VECTORS 3]

2-10 EQUALITY OF TWO VECTORS

Two vectors u and v are equal
if and only if their respective components are equal

and they are both written with respect to the same coordinate system. If u and
if

v are given by

u = u xI + i/J + M-k

and

v = v4 i + vj + v,k

and if v = u, then

ux - v*

U. — V,

U. — V.

since both are assumed to be written with respect to the same xyz coordinate system.

2-11 LENGTH OF A VECTOR

A vector that is directed from the origin of the coordinate system to another point
is said to be a position vector. In Fig. 2-2, the position vectors p and q, which are
directed from the origin to the points P and Q, respectively, are shown. If point P
has coordinates (p,, p 2 p 2 ) and Q has coordinates (q u q 2 , q 2 ), then p and q are
,

given by

p = p,i + p2 j + p3 k

Figure 2-2 Vector v defined in terms of the position vectors p and q.

32 MATH1 VIAT1CAI PfULlVUSARICS

and

<1 = ?|i + <7;j + '/"I'

From Fig 2-2. i( is noted that p + v = q (see Sec 2-12) and so

' = q - P = - />i)i + (</; - Pz )j + (</i

~ Fi)*'
(9i

Based on geometrical considerations, the length of the sector v, denoted as |v|, is

given by

I'l = |q - Pl = V(<h “ Pi)

!
+ <</; - rs)' + (<!> ~ P>)
:
(2-23)

Although Eq (2-23) is based on geometrical considerations in three dimensions,

the notion of vector length may be readily extended to multidimensional space (see
Problems 2-41 and 2-42)

Example 2-11

Compute the length ol the vector that runs from point P at (3. -4,6) to point Q at
(2, 5.-7)

Solution

The vector v is easily determined as

' = <1 - P = 12 - 3]i + |5 — ( — 4)|j + [-7 - 6]k

or

v — - li + 9j - 13k

The length of v is now calculated from Eq (2-23) as

M = V<- D + :
(9)
:
+ (- 13)
2
=15 8

2-12 VECTOR ALGEBRA

Vector algebra includes the notions of vector addition and subtraction, as well as
two types of vector multiplication. In this chapter only one type of vector multi-
plication. the dot product, is needed and thus reviewed The other type is reviewed
in later chapters.
Two vectors arc simply added component by component to obtain the vector
sum, provided the vectors arc written with respect to the same coordinate system
Tire rules for vector addition and subtraction are identical to those presented in Sec
2-4 for matrix addition and subtraction Addition of two vectors is frequently given
a geometrical interpretation by using the parallelogram law However, simple ad-
dition of the respective components is generally more useful in this text
The dot product of two vectors p and q is denoted as p-q and is a scalar defined
by
 VECTOR ALGEBRA 33

p-q = jp||q| cos a (2-24)

where a is the smaller angle included between p and q. Equation (2-24) implies

i-i = j-j = k-k = 1 (2-25)

and

i'j = j'k = ki = 0 (2-26)

With the help of Eqs. (2-25) and (2-26), p-q may be given in terms of the cartesian
components as

p-q = />i<7 , + p2 q2 + Pi<h (2-27)

which in turn can be written as

<7il
p-q = Ipi Pi Pi] <ii
- p7 q (2-28)

Jh.
Therefore, the dot product of p and q is frequently written in matrix form as p r q.
The dot product is also frequently referred to as a scalar product and an inner
product.

Example 2-12
Given the vectors p and q from Example 2-1 1, determine p + q. p - q, and p-q.

Solution

From Example 2-11, we have

p = 3i - 4j + 6k

and

q = 2i + 5j - 7k

Therefore, the required results are

p + q = 5i+ Ij- Ik

p - q = li - 9j + 13k

and

p-q = (3)(2) + (—4}(5) + (6)(-7> = -56

2-13 DIRECTION COSINES

The direction cosines of a vector v are defined as the cosines of the angles formed
by the vector v and the three coordinate axes. Figure 2-3 shows a vector > in two
34 MATHEMATICAL PRO IMIS ARIES

Figure 2-3 The cosines of the angles a, and a. are the direction cosines

dimensions. The ancles that \ makes with the x and v axes arc «! and a 2 , respec-
tively. Prom Fig. 2-3 and the notion of the dot product, it follows that

vi = |v||ij cos a,

But vi = », and |i| = I. and solving for cos a, gi\es

cos a, = r1
. (2 -29a)
M
In a similar fashion, it is readily shown that

cos a,
'
= £
M
(2-29b)

In three dimensions, if a, is the angle between a vector v and the : axis, then in

addition to the above, we have

In general, the direction cosines arc not all independent and are related by the
following

cos 2 a, + cos 2 a 2 + cos 2 a, = I (2-30)

The simple proof is left to the reader (see Problem 2-49)

Example 2-13
Tor the vector v in Example 2-1 1 . determine the three direction cosines and verify
that £q. (2-30) is satisfied.
 DIRECTION COSINES 35

Solution

From Eq. (2-29), we compute

cos a, = —-1- =
15.o
-0.0633

cos a2 = —9— = +0.570

15.o

cos a3 = —-13
— = -0.823
15.o

from which we get

a, = 93.6° a2 = 55.3° a3 = 145.4°

Equation (2-30) is readily verified once a,, a 2 and a 3

, are known. 0

2-14 SOLUTION TO SYSTEMS OF LINEAR ALGEBRAIC EQUATIONS

In all static stress analysis and other steady-state applications, a system of linear
algebraic equations of the form

Ka = f (2-31)

will have to be solved for the vector The system of equations implied in Eq.
a.

(2-31) may be solved by many methods,

but only the matrix inversion method is
reviewed here. Chapter 6 contains a more practical method based on Gaussian
elimination.
The solution to Eq. (2-31) by the matrix inversion method is straightforward.
If both sides of Eq. (2-31) are premultiplied by the inverse of the n X n matrix K
(assuming the inverse exists) and if it is noted that K~'K = I and la = a, then

an explicit expression for a results and is given by

a = K"'f (2-32)

Unfortunately this method is not very' practical in large problems because of the
time consumed (even by a computer) in obtaining the inverse. Nevertheless, it

provides a concise expression for the solution to Eq. (2-31). As mentioned in Sec.

2 - 7 , the inverse of a matrix is most readily performed by a subroutine on a digital

computer. Such a subroutine (subroutine INVDET) may be found in Appendix B.

Equation (2-32) simply states that to obtain a (which may be the nodal displace-
ments, temperatures, velocities, pressures, etc.), we simply invert K and post-
multiply this result by the vector F (which is known).
36 MAU!I_MAT1CAI f R1 LIVIINARILS

Example 2-14
Determine the solution to the following system of linear algebraic equations:

t, - 3t, =5
2r (
+ 4 t2 - 8r3 = 22
-7x + 8v 2 -
t
x3 = 16

Solution

From the matrix representation, we have

1 -3 O' 5
K= 2 4-8 and f = 22
-7 8-1 16

from which K 1
is computed (using subroutine LNVDET) to be

-0 52632 0 02632 -0 21053'

K 1
= -050877 0 00877 -007018
.-0 38596 - 0 11404 - 0 08772.

The solution for jc,, .r 2 and . r, is then easily computed from Eq (2-32) by a simple
matnx multiplication to be

jr, = -5 421

t2 = - 3 474
t, = - 5 842

Tlic reader should verify these by direct substitution into the original system of
equations B

It should be mentioned that K will always have an inverse for well-posed

problems solved by the finite element method Therefore, the questions as to whether
the sjstcm of equations represented by Eq (2-31) has one solution, no solution,
or several solutions do not arise For well-posed problems. Eq (2-31) has only one
solution for a and therefore the solution is unique One solution method applicable
to nonlinear algebraic equations is described in Chapter 8 The reader may wish to
consult references 3 to 1 2 for additional material relating to matrices and vectors.

REFERENCES

1. Brogan, W L Modern
, Control Theory. Quantum. New York. 1974, pp 45 —4ft-
2. Hornbeck, R. W Numerical Methods, Quantum. New York. 1975
3. Bellman. R. E., Introduction to Matnx Anahsn, 2nd cd McGraw-Hill. New York. .

1970.
4. Bronson, R.. Matrix Methods —An Introduction, Academic. New York, I960
 ,

PROBLEMS 37

5. Campbell, H. G., Matrices with Applications , Appleton -Century-Crofts, New York,

1968.
6. Franklin, J. H.. Matrix Theory Prentice-Hall, Englewood Cliffs, N.J., 1968.
7. Fuller, L. E., Basic Matrix Theory Prentice- Hall, Englewood
, Cliffs, N.J.. 1962.
8. Hohn, F. E., Elementary Matrix Algebra, 3rd ed., Macmillan, New York, 1973.
9. Moore, J. T., Elements of Linear Algebra and Matrix Theory, McGraw-Hill, New York,
1968.
10. Pipes, L. A., Matrix Methods for Engineering, Prentice-Hall, Englewood Cliffs, N.J.,
1963.
11. Schneider, H., and G. P. Barker, Matrices and Linear Algebra, Holt, Rinehart and
Winston, New York, 1968.
12. Stein, F. M., Introduction to Matrices and Determinants, Wadsworth. Belmont, Calif.,
1967.

PROBLEMS

2-1 Represent the following systems of equations in the matrix form Ax = b:

a. 2.V, + 3.v 2 = 4

-5*i + 6*2 = —6
b. —8*! + 3*2 + 6*3 = -9
4*2 — 5*3 = 6

*3 = 5

c. 4*1 - 5*2 + 6*3 = 5

2*| + 5*, - *3 = 2

2-2 Express the following systems of equations in matrix form:

a. 3 *i + 4*2 + *3 = 6
2*2 - 7*3 = -2
— 8*i + 6*2 + *3=0
b. 6*| + 5*2 + 3*3 + *j = 10
7*| + 3*2 — 6*4 = —7

2-3 VVhat is the 6 X 6 identity matrix and the 4x4 null matrix?
2-4 For the matrices A and B given below, determine if A and B are equal:

2 - 8 2 8
a. A = B = -8
8 6 6

6 f 6 1

b. A = -2 4 B = -2 4
L
1 8_

8 o' 8 o'
c. A = 0 5 B = 0 5
-3 2 -3 2
38 M AT1HA1 V51CAL PRt UMJS ARll-S

2-5 Which of the matrices gisen below arc equal (taking two at a time)?

2-6 For the matrices in Problem 2-5. determine the sums and differences indicated below.
If the matrices arc not conformable for addition or subtraction, so state ibis.

a. A + B b. A - C c. A - P d. A + E
e. U + D f. C + E g. D + F
2-7 Repeat Problem 2-6 for the following sums and differences

a. B + C b. B - C c. B - D d. C - E
e. C + D f. E-E g. F + F
2-8 For the matrices of Problem 2-5, determine

a. 5A b. -7B c. 8C
d. OD e. - 2E f. 5F

2-9 Repeat Problem 2-8 for

a. -7A b. OB c. -C
d. 8D e. 4E f. OF

2-10 For the matrices of Problem 2-5. determine the following products If the matrices
arc not conformable for multiplication, so stale (his

a. AB b. AC c. D\ d. DF e. FD
2-11 Repeat Problem 2-10 for the following products

a. CB b. BC c. CE d. EF e. FE

2-12 Problem 2-5'’

What is the size (or dimensions) of the matrices gisen in

2*13 What is the size of each of die products in Problem 2-I0 ’

1'1
2-14 What is the size of each of the products in Problem 2- 1

2-15 For the matrices in Problem 2-5. show that

a. A + K * E + A
b. 6D * 1)6
c. F(A + El = FA + re
d. C - E « C + (-E)
 PROBLEMS 39

2-16 For the matrices of Problem 2-5. show that

a. B + D = D + B
b. 8(C + E) = 8C + 8E
c. (B + I))F = BF + 1)F
d. (A + C) + E = A + (C + E)

2-17 For the matrices of Problem 2-5. show that

a. 5(A + C) = 5A + 5C
b. BF r FB
c. (BF)A = B(FA)
d. AE s* EA
2-18 By writing the matrices on an clement basis (e.g.. C is denoted as c ). prove the
v
commutative laws given by Eqs. (2-4) and (2-5).

2-19 Prove the distributive laws given by Eqs. (2-6) to (2-8) by writing each matrix on
an clement basis.

2-20 Prove the associative laws given by Eqs. (2-9) to (2-1 1).

2-21 Consider the matrices given below:

a. Which matrices arc symmetric? Why?

b. Which matrices are skew-symmetric? Why?

2-22 For the matrices of Problem 2-21. determine

a. Ar b. Br c. Fr

2-23 For the matrices of Problem 2-21, determine

a. CT b. Dr c. FJ d. GT
2-24 For the matrices of Problem 2-21, show that

a. (AD) r =D rA r
b. (GBF) = F rB rG r
r

c. Dr = D
 1

40 MM HI MA.TICM. m l IVU'.ARllA

2-25 For the matrices of Problem 2-21, show that

a. (BC)r = C rB r
C = -C T
b.
c. (ADG) r = CW
2-26 Prose for all symmetric matrices M that I’
r
MP is jlso always symmetric, where P
is not necessarily symmetric Ibut conformable)

2-27 Using Eqs. (2-14) and the matrices from Problem 2-21, determine

a. (let A
b. (let 11

2-28 Repeat Problem 2-27 for

a. det I)
b. del E
2-29 For the matrix A ot Problem 2-21. determine the following

a. The matrix formed bv the minors of A

b. The matrix formed by the cofactors of A
c. The adjoint of A
d. The determinant of A
e. The inverse of A

2-30 Repeat all parts of Problem 2 29 for ihc matrix D of Problem 2-2 J

2-31 Repeat all parts of Problem 2 29 for the matrix B of Problem 2 21

2-32 It may be shown in general that if A and H are both square and of the same si/c.

then det(AB) = (del Alldet B) For the matrices A and D of Problem 2-21. show
that dell AD) = (del AH del D)

2-33 Repeat Problem 2-32 for matrices B and C of Problem 2 2

2-34 Write the following sectors as column vectors:

a. p = 3i - 2j + 19k
b. q = -81 + 6j
c. r - 6i + 28k

2-35 Write the column vectors implied in Problem 2-34 as row vcciors <b> using the
transpose).

2-36 For the vectors given below, determine which arc equal (taking two at a time)

p = 3i - 6j + 4k q = 5i + 7k

r = 5i + 7j s = 31 - 6j 4k

u = 71 - 8j + 15k » - 71 — 8j 15k

2-37 For the vectors p. q. and u from Problem 2 36. determine

a. 'p' b. ,q c. u
 PROBLLMS 41

2-38 For the \ colors r. s. and \ from Problem 2-36, deternitne

a- IH b. |s| c. h|

2-39 A vector v is directed from point P at (3, -2.5) to point Q at (7,- 1,-3) in a
cartesian frame of reference

a. Determine the position vectors for both points

b. Determine the vector v

c. Calculate the length of v

2-40 A vector u is directed from point R at { -5.4.2) to point S at (0,6,7) in a cartesian

frame of reference

a. Determine the position vectors for both points.

b. Determine the vector u
c. Determine the length of u.

2-41 I3v a direct extension of Eq. (2-23). show that the length of v may be computed from

M =
for three-dimensional vectors. The result is, in fact, valid for n-dimcnsional vectors.

2-42 By using the result of Problem 2-41, compute the lengths of u and v where

a. u = 1 6 -5 0 4 2]
T

b. v = |-1 2 5 -3 6 1\
T

2-43 For the vectors p. r. and v from Problem 2-36, determine

a. p + r b, p - r c. r + v
d. v - p e. pr f. r-v

2-44 For the vectors q. s. u. and > of Problem 2-36. determine

a. q+s b. s — u c. u + v

d. u - v e. u-v f. q-u

2-45 For the vectors p, r. and v of Problem 2-36, determine the angle between

a. p and r b. r and v

Hint Use Eq (2-24) with the help of Eq (2-28)

2-46 For the vectors q. u. and \ of Problem 2-36. determine the angle between

a. u and v b. q and u

2-47 Determine the direction cosines of the vectors p, q, and r from Problem 2-36 and
Eq (2-30) is satisfied in each case
verify that

2-48 Determine the direction cosines of the vectors s and u from Problem 2-36 and verify
that Eq (2-30) is satisfied in each case

2-49 Prove that the sum of the squares of the direction cosines is unity [Eq (2-30)] Hint
Use Eqs (2-29) and the definition of the length of a vector.
2-50 Solve the following systems of equations by the matrix inversion method.

a. 3*i + 5r, = -6
2t, - 4 tj = 5
b. 6r, - 3r» = 5
-3r, + 5r2 = 3

2-51 Using the matrix inversion method, solve the following systems of equations-

a. 5 r, — 6t, = 7
3r, + 7t; = 8
b. 5r> + 2t> + 3rj = 5
2xj + isj + = 7
3t, + + 5x} = 9
 Truss Analysis:
The Direct Approach

3-1 INTRODUCTION

One of the most important practical types of engineering structures is the truss. A
truss may be defined as a structure that is composed of straight members connected
at the joints by smooth pins. Smooth pin joints cannot support a moment. A typical
truss is shown in Fig. 3-1. As shown in this figure, external loads such as P and
Q may act only at the joints. A truss must be restrained in some manner so that it

does not undergo free-body motion when loaded. The truss in Fig. 3-1 is restrained

by a pin joint at A and a roller joint at B. The reader may wish to consult a statics

textbook, such as Beer and Johnston [ 1 ], for a review of the various types of
supports.

43
44 TRUSS ANALYSIS Tilt DIRtCT APPROACH

Each member in the truss can only support axial loads, which implies that
lateral loads may not act between two joints All loads, therefore, must act at the
joints. Moreover, members in a truss are connected only at their extremities; hence,

members may not be continuous through a joint. In Fig. 3-1, for example, there
arc two distinct members, CD and DE, and not one member, CE. A truss is said
to be rigid if it does not collapse when loaded. A rigid truss, however, may and
will deform slightly under the action of one or more loads. The most common types
of trusses are shown in Fig. 3-2.
In this chapter, the finite element formulation of a two-dimensional truss is

deselopcd The direct approach, which does not require the use of variational
principles, virtual work principles, or weighted residual methods, is taken. The
mam purpose for considering this type of problem is to demonstrate how the direct
approach is used and to provide a step-by-step illustration of the finite element

Stidiutn

Figure 3-2 Typical trusses (a) Typical roof trusses (b) Typical bridge trusses (c) Other
types of trusses (Hum Vector Mechanics for Engineers Statics and Dummies by Bear and
Johnston Copyright C 1962 by McGrav. -Hill. New y ork Used with permission of McGraw-
Hill Book Company.)
 FINITE ELEMENT FORMULATION. THE DIRECT APPROACH 45

method. A specific problem is worked out completely by hand, and the basic steps
are then summarized. A simple computer program called TRUSS is described and
applied to this problem.

3-2 FINITE ELEMENT FORMULATION: THE DIRECT APPROACH

As mentioned in Sec. 3-1, a truss is assumed to be composed of only uniaxial force

members or bar elements. In other words, the joints are all assumed to be pinned
and the members cannot withstand bending or torsional moments. This particular
formulation is restricted to the two-dimensional truss only and will yield the dis-
placements of each joint as well as the axial elongation, strain, stress, and force
for each member. The finite element method is applicable to both statically deter-
minate and indeterminate trusses (and structures in general). The FEM formulation
for the three-dimensional truss is given in Sec. 3-5.

Discretization

The first step in any finite element analysis is discretization. Here the truss must
be discretized into a number of finite elements. The truss is composed of axial

tensile and compressive members, and it seems quite natural to consider each
member as an element. A typical truss element e is shown in Fig. 3-3. Note that

the element has and j, one at either end. It is at these nodes that the
two nodes, i

forces are transmitted from one element to the next. These elements are frequently
referred to as bar elements, as opposed to beam or frame elements. Bar elements
can withstand only axial forces; beam elements allow bending moments; and frame
elements include axial forces, bending moments, and torsion [2].
Each joint in the truss is represented by a node that is given a unique number.
Nodes are typically numbered consecutively from one to the maximum number
present. These node numbers are often referred to as the global node numbers, as
opposed to the local node “numbers” i and j. The numbering scheme has a strong
influence on the computational time to obtain the solution (see Chapter 7). Elements
are also numbered consecutively from one to the maximum number of elements.
However, the numbering scheme for the elements is totally arbitrary', unlike the
nodal numbering scheme. Local node numbers in this book are denoted with low-
ercase letters /, j, k, etc. For example, element e has only two nodes, and they are
denoted as nodes / and j. However, all nodes in a structure have global node numbers
and on element 3, for example, the global node numbers may be 10 (node i) and
7 (node j). The corresponding global node numbers are denoted with capital letters

e
)

Figure 3-3 Typical finite element in truss model.

46 TRUSS ANALYSIS THE niRfCT APPROACH

/, J, K , etc. The point is that every element in the truss has local node “numbers”
i and j with corresponding gtobal node “numbers” I and J. In the example above,
/ is 10 and./ is 7.
An essential part of every finite element analysis and computer program is the
data that defines the coordinates of every node present in the truss and the elements.
Element data are typically composed of nodal connectivity information and material
set flags. Nodal and element data are generally generated within a computer pro-
gram, but the user must still provide some minimum amount of data to describe
the particular problem at hand. This is called mesh generation , and one method for
accomplishing this is presented in Sec. 3-6.

Example 3-1

Figure 3-4(a) shows a cantilever truss with only five members and four joints.
Members A, C, and £ arc composed of a 0 5-in -diameter steel rod (with a modulus
of elasticity of 30 x |0 6 pst), and members B and D are composed of a 0.4-in.-
diameter aluminum rod (with a modulus of x 10 6 psi). Discretize the truss
1 1

showing the nodal coordinate data and the element data in tabular form (hence
suitable for use in a computer program) The coordinates (in inches) of the joints

arc shown in Fig 3-4(a).

Solution

Since the truss has five members, it is natural to discretize the truss into five
elements, each member being an element Each joint can then be taken as a node.
Therefore, four nodes are used The first step in the discretization process is to

Figure 3-4 (a) A cantilever truss with five members and four joints, (b) Truss discretized
into five elements and four nodes.
 FINiri. LLEMCNl FORMULATION' THf. DIRECf APPROACH 47

number each joint or node consecutively from one to the maximum number of
nodes present. For the truss under consideration the nodes are numbered 1 to 4, as
shown in Fig. 3-4(b). Using the coordinates shown in Fig. 3-4(a) we may summarize
the nodal coordinate data as shown below:

Node
number a coordinate, in. i coordinate, in.

1 12 0 0.0
2 12.0 6.0
3 0.0 00
4 0.0 10.0

Let us use material set flags 1 and 2 to denote the 0.5-in. steel and 0.4-in. aluminum,
respectively. With the help of Fig. 3-4<b). the element data may be summarized
as follows:

Element Node Node Material

number 1
J set flag

1 1 2 1

2 3 1 2
3 3 2 1

4 4 2 2
5 3 4 1

ra

Element Stiffness Relationship in Local Coordinates

It will soon prove to be very convenient if a local coordinate system, xV, is chosen
as shown in Fig. 3-5. This x'v' coordinate system is known as a local coordinate

system, since it is defined locally on the element. Note that the a' axis is directed

nodal forces and nodal

Figure 3-5 Truss element in local coordinate system showing
displacements.
4S TRUSS ANALYSIS T1t£ DtRTCT APPROACH

along ihe length of the element from node i to node j. The v' axis must be chosen

to be perpendicular to the t' axis, but its sense is arbitrary Nodal forces in the x' .

\ directions at node i are denoted as U' and V,’. respectively, and nodal
and
displacements as «' and i‘ respectively Similarly, at node j. the nodal forces are
.

U't and l',' and the nodal displacements are n* and v* in the r' and y' directions,
respectively
It should be recalled from elementary strength of materials that the axial elon-
gation 5 is given by

where L is the length of the member (or clement). P the axial force. A the eross-
sccttona) area, and £ the modulus of elasticity t Young's modulus). It is assumed
that the clastic range is noi exceeded and that A is constant Solving Eq (3-1) for
P and interpreting it as the axial force at nixie 1 (£,'). we can wnte

u; = — u,; - I !3-2a)

where the effective elongation 6 is given by the difference between the nodal
displacements along the element In a similar tashion. we may wnte

2W
v; -
y ' l3 '

Note that V, ~ - C‘. which must be the case if element t is to be in static

equilibrium in the x‘ direction Because this particular type of clement cannot support
transverse forces, we also have

r; = 0 (3 -2c)

v; - 0 (3-2d)

Equations (3-2) may be wnlten in matrix form as

"
1 0 - 1 0 u' ~v~
0 0 0 0 v' r;
(3-3)
- 1 0 1 0 u] L"
LOO 0 oj _v; J

Since the global nodal coordinates denoted as and ( .

;
) are specified for
(

any given tniss, the element length may be computed from

l = Y(t, - t,
>-'
+ - \,)
: 13-4)

The other two propenics. A and £. are also specified for every element, usually
with the help of material set flags m a computer program (see Example 3-1 and
Secs 3-3 and 3-6) Equation (3-3) may be wntten concisely as

K' a r = f' (3-5)

 FINITE ELEMENT FORMULATION THE DIRECT APPROACH 49

'

where K' represents the local element stiffness matrix, a e the local element nodal
'
e
displacement vector, and i the local element nodal force vector. The superscript
C) is used throughout this text to denote element and a prime (') to denote the
local coordinate system, unless stated otherwise. Before Eq. (3-5) can be applied
to the truss on the whole, it must be transformed to the global (.yy) coordinate
system. This transformation is developed in the next section.

Example 3-2
Determine the local element stiffness matrix for member C (or element 3) for the
truss in Example 3-1

Solution

From Eqs. (3-3) and (3-5), the local element stiffness matrix K f
for element 3 (or
e — 3) is given by
l

1 0 i 0
I

A <3) £< 3) 0 0 j
0' 0
i

L<3) -1 "o"T~ 7 0 i

0 0 i

i
0 0_

where element 3 has material set flag 1, which corresponds to the 0.50-in. steel or

A0 = 1)
(0 5)
2 = 0.196 in.
2

£ (3) = 30 X 10 6 psi

Lm = V (* 3 - x2 )2 + ()-3 - y2 )
2
= V (0 - 12)
2
+ (0 - 6)
2 = 13.42 in.

Computing £ <3)
gives

438 0 !
-438 (f
' 0 0 00
K <3)
-438 '77
!

438 “o'
Ibf/in.

0 0 *
0 0

for the local element stiffness matrix for element 3 (member C). The 10 3 scalar

multiplier should be noted. B

Transformation from Local to Global Coordinates

A local (x’y') coordinate system was used to derive the stiffness relationship given
by Eq. (3-3). The x' axis was purposely taken to be directed from node i to node
j. The y' axis, which must be chosen to be perpendicular to the _v' axis, is otherwise
arbitrary (with respect to its sense). Figure 3-6 shows a global coordinate system
in relation to a typical element e. The purpose of this discussion is to find a
50 TRUSS ASALV SIS THE DIRECT APPROACH

Figure 3-6 Relationship between local and global coordinate s>stems

transformalion that >v»|] enjbJe us !o write Eq (3-5) nilh respect to a global co-
ordinate system.
First, let us set up a global tv system as shown in Fig 3-7(a) and draw a
position vector r to an arbitrary point P The vector r may be written in terms of
its two cartesian components r, and r, as follows

r = r,i + rj (3-6)

Next, let us define the same vector r in terms ot the rotated coordinate system t'y'

where, with no loss in generality, the x' axis is purposely taken to be in the same
direction as r itself, as shown in Fig 3-7(b) In terms of the cartesian components
(rj and r') in the rotated frame, the vector r becomes

r = r'i' + r'j' (3-7)

where and j' arc the two unit vectors

i’ in the x' and directions, respectively.
From Eqs (3-6) and (3-7), we have

r,i + r .j — rji' + r'j' (3-8)

Figure 3*7 Position vector r (a) m global coordinate system and tb) in local coordinate
system.
 FINITE ELEMENT FORMULATION THE DIRECT APPROACH 51

Taking the dot product ol both sides of this equation with respect to i gives

>\ = '•JH' + r'i*j' (3-9)

where id 1 =
and i*j = 0 have been used. Note that id' represents the cosine of
the angle between the x and ,v' axes, and i *j represents the cosine of the angle '

between the x and y' axes. It should be recalled that these are the direction cosines,
which will be denoted as

nn = cos(.y.a') = id' (3-10a)

71,2 = cos(.r,v') = i-j' (3-10b)

where cos(.v,.v') represents the cosine of the angle between the jc and .v' axes, etc.
Therefore, with the help of Eqs. (3-10), Eq. (3-9) becomes

t\ = n u r[ + 77, 2 /\' (3-1 la)

By forming the dot product of Eq. (3-8) with j in a completely analogous manner
we also have

r i
= ll
2l r 't
+ n 22 r '\
(3-1 1b)

where ;i
2, represents the cosine of the angle between the y and x' axes, and n 22
represents the cosine of the angle between the v and y' axes. Clearly, in the notation
for the direction cosines, the first subscript is associated with the global coordinate
system and the second with the local (rotated) system. Equations (3-11) may be
written in matrix form as

rX _ »11 "12
(3-12a)
t\ fhi /?22

or more concisely as

r = Tr' (3-1 2b)

where T is the so-called transformation matrix that transforms a vector in the local
system to one in the global system. It can be shown that the matrix T is orthogonal
(see Problem 3-21) because T -1 = T 7 (see Sec. 2-7). Therefore, it follows from
Eq. (3- 12b) that

r’ = T rr (3-13)

This last result be used to transform a vector in the global system to one in a
may
local system. These transformations are used next to determine the stiffness rela-
tionship with respect to the global coordinate system.

Global Element Stiffness Relationship

Before the element stiffness matrices IC and nodal force vectors f can be assem-
bled to represent the entire truss, to a global coordinate
T and r ,/e frdve*
system. Interpreting r' as [u't v's ]
OF ippwrys'
&CC. Ko ^^ %^r~~
 ,

52 TRUSS ANALYSIS THE DIRTCT APPROACH

M.'l
= ["n n2 ,
u
13-14}
V'J j_«|2 l>22

where u, and v, are the displacements at node i in the global coordinate directions
x and y, respectively. Similarly, for the displacements at node j, it follows from
Eq. (3-13) that

(3-15)

Figure 3-8(a) shows the relationship between the local and global nodal displace-
ments. Equations (3-14) and (3-15) may be written in one matrix equation as

~u'~ ~»ii "21 0 0 11,

»; »12 Hjj 0 0 V,
(3-16)
ui 0 0 "it "21 Uj

»; 0 0 «I2 "22 V,

or more concisely as

a' = Ra' (3-17)

where, by definition

(3-18)

and a' represents the vector of nodal displacements for element e referred to the
global coordinate system, or

= T
a' In, |
ii, v,) 13-191

In a completely analogous manner we may also write

Figure 3-8 T)pica1 element t showing global and local (a) nodal displacements and (b)
nodal forces. AH forces and displacements arc shown with a positisc sense.
 FINITE ELEMENT FORMULATION. THE DIRECT APPROACH 53

'

t
e = RP (3-20)

where P is the vector of nodal forces for element e referred to the global coordinate
system, or

f- = W, V',
j
UJ vy (3-21)

Figure 3-8(b) shows the relationship between the local and global nodal forces.
With the help of Eqs. (3-17) and (3-20), we may write Eq. (3-5) as
-

K' Ra'' = RP (3-22)

r
which may then be premultiplied by R to give

R rK e Ra* = P -

(3-23)

where R r R = R~‘R = I has been used on the right-hand side. Equation (3-23)
may be written simply as

Kf a‘‘ = P (3-24)

where Kf . the element stiffness matrix referred to global coordinates, is obtained

directly from the element stiffness matrix referred to local coordinates (K f ’) by

K = R r K‘ R
f
(3-25)

The matrix FP is often simply referred to as the global element stiffness matrix (as
opposed to the local element stiffness matrix IP ). If the multiplications indicated

in Eq. (3-25) are carried out, we get

«ll «2I«II !
-n?. -«2I«I1
«!l«2l «2I
•>
~ ,! ,i
21 -"21
i 1I (3-26)
1 1
-«ll -«21«II ! «ii «21«II

-«1I«2I -»2I ! «11«2I n\ j

It should be noted that in the expression for K f

only two of the four possible
direction cosines appear: n u and n 2 \. By definition, /i
n is the cosine of the angle
between the x and x' axes, and n 2i is the cosine of the angle between the y and ,v'

axes. Recall that the x' axis is directed along the element from node / to node/
From Fig. 3-9 it is immediately apparent that

nn = cos 6 r = { 3 ' 27a >

and

n 2i = cos 0, = (3-27b)

where (.r„y,) and for element e are the global coordinates of nodes i and/
ix r yf)
respectively, and L is the element length [easily computed from Eq. (3-4)]. It should
be recalled that the coordinates of each node are a vital part of the input to every
 }

54 TRUSS ANALYSIS THE DIRECT APPRO XCH

Rgure 3-9 Definition of angles 0, and 0, used in obtaining the direction cosines.

finite element analysis land computer program! Consequently, the two needed
direction cosines arc routinely computed with Eqs (3-27). It must be emphasized
that the important result is Eq. (3-26). which enables us to compute the global
element stiffness matrix for every element since A and E are specified, and L. n, t ,
and n ; arc ,
easily computed (for every clement) In the next section, the notion of
assemblage isintroduced It is an easy matter to show that K' is symmetric because
K' is itself symmetric (see Problem 2-26)

Example 3-3
For the truss in Example 3-1. determine the global element stiffness matrix for
element 3.

Solution

The global element stiffness matnx. denoted as K', is given by Eq. (3-26) in terms
of the direction cosines n M and n*,. which are computed from Eqs (3-27) for
element 3 as follows:

V; — y _
~
6.0 -00 = 0 4472
L'" 13 42

where the result for the element length from Example 3-2 has been used The
calculation of K' 5 from Eq (3-26) is now straightforward, and the result is
’

351 176 -351 - 176"

176 88 -176 -88
K°’ = 10'
-351 - 176 351
” 176"
-176 -88 176 8S_

where the cross-sectional area A f, was also taken from Example

‘ 3-2. As expected.
K 4 *’
is symmetric. ®
 FINITE ELEMENT FORMULATION THE DIRECT APPROACH 55

Assemblage

The assemblage step is necessary in all finite element analyses. It is in this step
that the original truss (or in general, any structure, body, or region being analyzed)
is put back together or assembled from the individual elements comprising it. Suffice
it to say now that the assemblage step is based on the principle of compatibility;
that is, the.x and v displacements associated with a particular node on any given

element must be identical to those associated with the same node on each element
that shares this node [3],
Consider a two-dimensional truss with N nodes. Each element e has two nodes
rand j with global node numbers / and J. Since each node has two degrees of
freedom (i.e., two components of displacement) the 4 x 4 global element stiffness
matrix may be thought of as being partitioned into four submatrices, each of size
2 X 2, or

~K(, Kf/
K' ( 3 - 28 )
K
These 2x2 submatrices are easily extracted from Eq. (3-26). Since global node
number / corresponds to local node i and global node number J to local node j,
Eq. (3-28) may be written in terms of the global nodal subscripts as

1
fa./
| 11 ( 3 - 29 )

w! S Kh
1
The assemblage stiffness matrix, denoted as K", is always of the form

Kh K ‘l. Kfj • • Kb
Ki, K£, Kgj . • Kb
Ki, K 3.2 K b ;
• Kb
'
a
1 T, '< *4 2 KX'3 • • K^jy__

Note that the superscript (°) denotes assemblage. Each of these K"/s are 2 X 2
submatrices. To begin the assemblage process, the assemblage stiffness matrix K“
is zeroed out. Then the individual global element stiffness submatrices (i.e., the

Kf/s) are added to the corresponding position in the matrix K a

(i.e., at K "j) for

element 1 , then for element 2, and so forth, up to and including the last element.
The assemblage of each 2x2 submatrix in Eq. (3-30) may be given by the
relationship

M
K fj = 2 Kf j for / = 1 to N and J = 1 to N (3-31)
e— 1

In Eq. (3-31), M is the maximum number of elements, and K£ y is taken as the 2

x 2 null matrix if element e does not have nodes / and J. The next example will
56 TRUSS ANALYSIS TUfc D1RFCT APPROACH

help to clarify the assemblage process for the truss (see Sec. 3-3 for a numerical
example).

Example 3-4
For the truss of Example 3-1, determine the assemblage stiffness matrix in terms
of the 2 x 2 global element stiffness submatrices K eu .

Solution

Element I has global node numbers 1 and 2 (i.c., / and J, respectively), and so
the global element stiffness matrix for element 1, denoted as K ,,
\ is given by

KV>, K\'(
K". |

kv* j
Kvt

Having first zeroed-out the assemblage stiffness matrix Ku , we may then “add”
this result to K“ to get

r kvi HL 0 0

Ki'Jj
0
Hi
0
0
“o'
0
T
0 0 o" 0

This is not the complete form of the assemblage stiftness matrix because four more
elements have yet to be considered Element 2 has global node numbers 3 and 1

(t e , / and J , respectively) and the global element stiffness matrix for element 2
is given by

K ,..,
= [KfliKyr
|k?7[ kb
After combining this element stiffness matrix with the result from element 1, we
get

KB + kr KB 0
KB tF?L_ 0
Kfi r o KJ’i 0
0 0 0 0

Note the position of each entry in K“ carefully After considering clement 3 (with
nodes 3 and 2), element 4 (with nodes 4 and 2). and clement 5 (with nodes 3 3nd
4), the assemblage stiffness matrix becomes

K\'| + Kr KVi KVi : • J

Kp| ! KW + KV’t + KW ! KVl HliriWii
KVI 1
KVI KVI + KVI + KVSI j

1
0 1
K‘4'i K*;j
 FINITE ELEMENT FORMULATION. THE DIRECT APPROACH 57

Several significant observations should be made. First, each of the individual 2 x

2 element stiffness submatrices is symmetric, as an examination of Eq. (3-26)
clearly shows. Mathematically, this may be stated as

KIj = Ky, (3-32)

It then follows that K" itself is symmetric, which has several advantages as delin-
eated in Chapters 6 and 7. Another property of K" is that it is banded, which means
the nonzero entries in Ka are gathered along the main diagonal. In the example
above, this is not very obvious because the truss has only five elements and four
nodes. Because of the symmetric nature of K", we usually use the term half-
bandwidth, denoted as b n in this text. A general expression for the half-bandwidth
is given by

K = AWO + A mM ) (3-33)

where A DO F
r
is the number of degrees of freedom per node and A max is the maximum
of the maximum differences between node numbers on each element considering
all elements one at a time. In the truss model there are two degrees of freedom

associated with each node (the two components of displacement). From Eq. (3-33)
it can be seen that the way in which the nodes are numbered can have a strong
influence on the half-bandwidth. In a computer program, the execution time in-
creases with the square of the bandwidth for the usual solution techniques [4], 0
Example 3-5
Determine the half-bandwidth of the assemblage stiffness matrix from Example 3-
4 by direct examination of K" and by computation using Eq. (3-33).

Solution

Let us write K" in the following form showing each potentially nonzero entry as
an x and each zero entry as a 0:

-I
X x
-Y.-Y
X -Y
.Y

JC
.Y

-Y Jt x X 00 0
0
jc jy^Ly .y .y X X .x

X X X -Y X X X X
K" =
x x x x x X X X
N
0 0
-

-v
v •'

x x
v
'

x \ v
x
Y
-

x
_0 o\ -V .V X -Y \v_

A direct count of the number of nonzero entries in the half-bandwidth above yields
6. Equation (3-33). of course, must give the same result for the half-bandwidth,
but let us compute it. There are two degrees of freedom per node, hence N Dor = .

2. The maximum difference between any two node numbers considering all elements
58 TRUSS ANALYSIS THF DtRLCT APPROACH

one at a time is 2, or = 2. Therefore, Eq. (3-33) readily gives b, =2(1 +

2) = 6 as expected

Application of Loads

u
The assemblage nodal force vector f may be obtained by an assemblage process
similar to the one used to obtain K“ It is simpler, however, because each element
r
nodal force vector, f is a column vector (of size 4 x 1), as opposed to a matrix.
,

The assemblage of each element nodal force vector in this manner is illustrated in
subsequent chapters Because the external loads on the truss can only occur at the
1'
joints, the effective assemblage nodal force vector f can be obtained by a more
direct route as explained below
The assemblage nodal force vector, with 2N entries, is first zeroed-out. where
N is the number of nodes present Let us assume that at global node number K,
there is a point load with components F ,
and These components are referred
to the global coordinate system as shown in Fig 3-10 Because the load is applied
at node K, we must add F, and F„ to the {2K - l)st and 2A'th positions in F',

respectively. Reaction forces are not considered to be loads in finite element anal-
ysis. They are handled routinely as restraints (see Sec 3-3).

Example 3-6

For the loading shown in Fig 3-4(a) for the truss in Example 3-1, determine the
J
assemblage nodal force vector f .

Solution

The 2000-lbf load is the only externa! load and must be resolved into its two cartesian
components, F, and Fy . Note from Example 3-1 that an xy (global) coordinate
system is implied as shown in Fig 3-4(a). The .r and y components of the applied
load arc given by F, = - 2000 cos 60’ = - 1000 lb and F v = - 2000 sin 60° =

Figure 3-ID Portion of truss showing node K with external load (components F, and h ,)
 — ,

FINITE ELEMENT FORMULATION: THE DIRECT APPROACH 59

- 1732 lb. Since this load is applied to node 1, the first [from 2(1) -
1] and second
[from 2(1)] entries in f" must be changed to yield the following result:

-
-
1000'
1732
M
F*J
Node 1

=
0
0
M
F>J
Node 2
f"
0
0
M
PyJ
Node 3

_
0
0 _
M
PyJ
Node 4

If another force also acts atnode l the respective components would be added
,

to those above. The restraints on the nodal displacements are considered next. 0

Application of Restraints on Nodal Displacements and Solution

The assemblage stiffness matrix K" and nodal force vector f" are obtained as
described above. The assemblage system equation is of the form

K“a = f
n
(3-34)

where a is the vector of nodal unknowns or, in this case, the vector of unknown
nodal displacements. The vector a is given by

a = [m, v, i u2 v2 i
j
uN v„f (3-35)

where N is the maximum number of nodes and u,, v, are the x and y components
of the displacement at global node number /. In Eq. (3-35), the transpose is used
to conserve space a itself is a column vector as implied by Eq. (3-34). Equation

(3-34) cannot be solved for the vector a because the restraints on the nodal dis-
placements have not been considered. In fact, the inverse of K" does not exist, in

general, until the restraints on displacements are taken into account. In other words
K" is singular. After the restraints are considered, Eq. (3-34) is written as

Ka = f

where K is no longer singular. Therefore, K -1 exists, which implies a unique

solution for the vector of nodal displacements a. Two methods are commonly used
to impose the prescribed displacement restraints [5]. Each is discussed below.

Method 1

The first method is most easily explained by an example. Consider the system of
equations

k\ 1<?1 + ki2 a2 + ^ 13 a 3 — fi (3-36a)

60 TRUSS ANALYSIS Tilt DIRECT APPROACH

Ajitij — f2 (3-36b)

^3i° i + + A 33 a 3 — f) (3-36c)

Let us impose a restraint on a2 , in other words, that a, is to have a specified value,

e.g., ap (a prescribed displacement in this truss analysis). Let us further assume
that the system of equations is symmetric (such as the truss model) and that the
symmetry is to be preserved Let us replace Eq. (3-36b) with

The system of equations thus becomes

&n<Zi + A (2^2 + Ai 3 a 3 ~f
a2 ~ ap

A 3 |fl, + k J2a 2 + A 33 a 3 = /3
But the symmetry has been destroyed, and to preserve rt we need a zero coefficient
on a 2 in the first equatton We can get that by noting that a 2 ~ ap (which is known)
and writing

A,,tf| + A 13 ct 3 = /i
- k l2ap

Note that the term involving a 2 is conveniently transposed to the right-hand side
(and ap replaces ci 2 ) After performing a similar operation on the third equation and
using matrix notation. we get

-
An 0 A| 3
hi = h “
0
.An 0
1 0
A 33 _
U =
L«3j
K- If3 k,&p .

Note that f2 disappears* The implication is that for degrees of freedom in the truss
model where an unknow n reaction force exists, this unknown force (f2 above) never
really enters the formulation Instead, a known displacement is imposed. In struc-
tural analysis problems, these prescribed displacements arc usually zero, and in this

case ap ~ 0 m Eq. (3-37)

Method 2

A much simpler method exists for imposing prescribed displacement restraints. The
method is based on the concept of penalty functions, and a proof of wh> the method
works is given in Zicnkiewicz [6J. The purpose here is not to prove it but to learn
how to apply it.

Let us again consider the system of equations in Eq (3-36). In this method a

large number p is selected, say six to twelve orders of magnitude larger than the
largest coefficient k tJ , and is added to the coefficient A„ if </, is to be prescribed, fn
addition, the right-hand side of the ith equation is changed to p times the prescribed
 FINITE ELEMENT FORMULATION THE DIRECT APPROACH 61

value (of displacement). In the example above, since a is to be prescribed (i.e.,

2
az ~ aP)-> the second equation becomes

^2i a \ + (k-22 + P)«2 + Ar

2 3«3 = pOp

Note that j3 is chosen to be much larger than the coefficients and, for all practical
purposes, this last equation is really equivalent to

P«2 = P ap
or

Gp

which is the desired result. The other equations are unchanged. In matrix form the
system of equations becomes
’
^12 ^'l 3 «I '/
^“21 kj2 P A'23 a2 = 3fl,,
f

-^31
-A*

1 - a 3_
—
. h _

Note that this simple method automatically preserves symmetry. In a computer

program. Method 2 is extremely easy to apply. Examples that demonstrate Methods
1 and 2 are given in Secs. 4-8 and 3-3, respectively.
Both methods result in the system of equations represented by the matrix
equation

Ka = f (3-39)

(without any superscripts). This equation may be solved for the nodal displacements
contained in the vector a by any method applicable to the solution of linear, algebraic

equations. In fact, some solution techniques, such as the one in Sec. 6-8, take
advantage of the symmetric and banded nature of the stiffness matrix K, but let us
use the matrix inversion method here. If K _I is computed and if Eq. (3-39) is

premultiplied by it, we get

a = K~ !
f (3-40)

This method is used in Sec. 3-3 in the solution to an example problem. A much
more practical method of solution is presented in Sec. 6-8.

Computation of the Element Resultants

By definition, the element resultants for the truss include quantities such as the
axial elongations, strains, stresses, and forces. Each of these is computed more or
less directly from the nodal displacements a that were computed to be the solution
of Ka = f.

The elongation 8 for element e may be computed in terms of the x components

of nodal displacements at nodes / and j by
 1 1

62 TRUSS ANALYSIS Tlif DIRFCT APPROACH

8 = itj — «j (3-41)

But from Eq. (3-14), we have

u\ = «||M, + 021 V, (3-42a)

and from Eq (3-15),

«; = «„«, + n 2 ,Vj (3-42b)

Note that in Eqs. (3-42) only two of the direction cosines appear («u and n2t )'
these are easily computed for every element by Eqs. (3-27). In Eqs. (3-42), u,. \\.

u„ and \’j are known once the nodal displacements are determined (from the solution
of Ka = f) Therefore, for any element e, the elongation 8 is readily computed.

The axial strain t for a typical element e is computed from

where L is the element length computed from Eq (3-4). The material is assumed
to be elastic so that Hooke's law applies Therefore, the axial stress a in the element
is computed from

ct = Ee (3-44)

Finally the axial force F tn the clement is determined by

F ~ aA (3-45)

In Eqs (3-41 ), (3-43), (3-44). and F denote

and (3-45). positive values of 8. e, o\
tension, whereas negative values denote compression The formulation of this sec-
tion is utilized in the next section in an example

3-3 APPLICATION TO A SPECIFIC EXAMPLE

In this section, a simple problem is executed completely by hand in order to illustrate

the basic steps in the finite element method. Let us reconsider the truss in Example
3- show n in Fig 3~Us) It should be tvcalled that members A C. and E arc made .

of a 0.5-in. -diameter steel rod and members B and D of a 0.4-m -diameter aluminum
rod. The modulus of elasticity of steel and aluminum are 30 x 10 and 1 x I06 ft

psi, respeemely The truss is constrained and loaded as show n in Fig 3-4(a). which
also shows the coordinates of each joint
The first step in any finite element anjljsts is discretization. Let us use the
results of Example 3-1 where we took each member of the truss to be an clement.
Itshould be recalled from that example that the discretization step ma> be divided
into two separate but related tasks- (1) specification of the nodal coordinates and
(2) specification of the element data, including material set definitions. Because
so
this step is sitally important, let us summarize the results of the discretization
from Example 3-1.
 APPLICATION TO A SPECIFIC EXAMPLE (j3

Nodal Coordinate Data

Node
number a coordinate, in. y coordinate, in.

1 12.0 0.0
2 12.0 6.0
3 0.0 0.0
4 0.0 10.0

Element Data

Element Node Node Material

number i
i set

1 i 2 1

2 3 1 2
3 3 2 1

4 4 2 2
5 3 4 1

Note that member A is taken as element 5, member B as element 4, and so forth.

In addition, material set 1 is the 0.5-in. steel and material set 2, the 0.4-in. alu-
minum. The nodal numbering scheme used above will result in the smallest possible
bandwidth of the assemblage stiffness matrix.

were needed for each element: the cross-sectional

In Sec. 3-2, three properties
area, the modulus of elasticity, and the length. The length is easily calculated by
Eq. (3-4) from the nodal coordinates. However, the remaining two “material prop-
erties” must be specified:

Material Cross-sectional Modulus of

2
set area, in. elasticity, psi

i 0.1963 30 x 10 6
2 0.1257 11 x 10 6

All of the above information is used frequently throughout the finite element solution
process. In summary, the truss model will be composed of four nodes and five
elements involving two different types of materials. Not surprisingly, this same
information becomes an integral part of the input to the computer program described
in Sec. 3-6.

The next step is to determine the local element stiffness matrix for element 1

to transform it to the global element stiffness matrix, and finally to “add” it into

the assemblage stiffness matrix (which is always zeroed-out before processing the
first element). From Eqs. (3-3) and (3-5), the local element stiffness matrix is given
by
$4 TRLSS ANALYSIS THfc DIRECT APPROACH

1 0 -I o'
_ AE 0 0 0 0
L -1 0 “‘T ‘o'
0 0 0 0_

Element 1 is composed of material 1, and so we have

1
°
i — o
(0 1963)(30 x ID
6
)
r o o i
i
© ©
—
(3-47a)
V02 - 12.)
1
+ (6 - 0.)
; -1 O J —

Loo! o

or

982 0 -982 o'

j

0 0 0 0
K'» = 10’
1
m (3~47b)
-982 "o"j 982 o'
0 o |
0 0

Although only three or four significant digits are shown here, all calculations
were carried out using nine digits and. therefore, numbers may not add exactly as
shown during the assemblage. The reader should note the scalar multiplier of 10'
and \crif> that each entry in K"* has units of pound-force per inch The stiffness
terms in structural models will always have units of force per unit length — similar
K lM
to the units of the stiffness for a spring Although K ,n can be transformed to
by applying Eq (3-25) to element I. it is easier to use the result for K' given by
Eq (3-26) Recall that only two of the direction cosines are needed, which, for
clement 1. are readily computed as follows

13 -48a)
60
6 0 - 0.0
(3-48b)

The reader should examine the orientation of element 1 in Fig 3-4(b) to verify
these direction cosines by inspection With the help of Eq (3-26). the global element
stiffness matrix for element 1 may be verified to be

0 0 0 0
{

0 982 jO -982
K «'»
10
'
Ibf/in (3-49)
0 0 fo o"
0 - 982 » 0 982.

This result must now be added into the assemblage stiffness matrix (which at this
point is simply a null matrix) Since clement I connects nodes I and 2. the assem-
blage stiffness matrix K u becomes
 APPLICATION TO A SPECIFIC EXAMPLE 65

0 0 0 0 i 0 0 0 0
| ;

0 982 j 0 -982 i
0 0 0 0
j

0 0 To 0 ! 0 5 io 0
0 -982 0 982 0 0
K“ =
!
j
Ibf/in. (3-50)
0 0 To 0 0 | 0 ! 0 0
0 0 jO 0 |o o ! o 0
0 0 !o 0 ] 0 0 ! 0 0
0 0 j
0 0 j
0 o ! o 0

The reader should corroborate this result for the assemblage stiffness matrix
(after processing the first element) with Example 3-4.
r ’, 1-
Although the nodal force vectors f f ,
and f" could be considered at this

point for element I , it proves to be more convenient to obtain f" directly as indicated

by Example 3-6. This same comment applies, of course, to the nodal force vectors
for elements 2 through 5.
Element 2 is considered next, and since it is composed of material set 2, K <2>

is given by

1-1 0~
1 0
6
(0.1257)(11 x 10 0 0 i__o_ 0
T
)
K< 2
>'
= (3-51a)
12.0 “-T"o' r j

0 0 1
0 0_

or

115 0 1-115 0
0 0 0 0
(3-51 b)
-115 IT 115 "o'
0 0 0 0

The reader may verify that for element 2 we have nn — 1.0 and « 2 i
= 0-0, both
by inspection and by using Eqs. (3-27). Therefore, we have K <2> = K <2, \ or the
global element stiffness matrix is identical to the local element stiffness matrix [sec

Eq. (3-26)]. Since element 2 is connected by nodes 3 and 1 , the assemblage stiffness

matrix becomes (after processing two elements)

~ 115 0 j
0 0 I
115 0 1 0 0“
0 9S2 ! 0 -982 1
1
4. ...
0 _oj_o_ 0
T ~0
0 0 0 0 I
o ! o
] 1
1

0 -982 ]
0 982 1 0 o 1.0 0
*1
-115 0 [o 0 1
115 o" J o" "o"
1

0 o ! o 0 f 0 0 1 0 0
"5"
0 0 To 0 1

l
0 "ofo”
1

_ 0 0 !
1
0 0 1
1
0 o !
(
o 0_

Again the reader should examine this result in light of Example 3-4.
66 TRUSS ANALYSIS Tlth DIRFCT APPROACH

Element 3 is a little more interesting than the previous two elements because
it is oriented neither horizontally nor vertically. The length of element 3 is easily

computed from Eq. (3-4) to be

V02 - 0)1 + (6. - o.) 3 ;

= 13.42 in.

and K |J>
becomes

1 0 -1 0"|
6
(0. 1963X30 x 10 ) 0 0 0 0
K n> = i

-1 cr 0
(3-53aj
13.42 l

L o 0 0 0J

439 0 _ 439 o 1
0 0 0 0
K<3. (3-53b)
-439 0 439 0 |

0 0 0 0 1

In order to transform K'” into K0 ’, the direction cosines are needed. Recall from
Example 3-3, for element 3 we calculated = 0 894, n 2 i
= 0-447. With the
help of Eq (3-26), the global clement stiffness matrix for element 3 is computed
to be

351 176 -351 -176

176 88 -176 -88
K' 3 >
= !0 J
-351 -176 351 176
-176 -88 176 88

This matrix must now be included in the assemblage matrix K", and since element
3 connects nodes 3 and 2, we get the result

115 0 0 0 1 -115 0 0 0
0 982 0 -982 [
0 0 0_ 0 _

0 0 351 1
76~! -351 "—176 IT' "o'
0 -982 176 1070 1
I_17_6__ -88 o__ 0
- 1 15 o” 1

-351 - 176 466“ 76 o’" "o' i

|

0_ ” 0 -176 176 88 _0 _0_

0 0 0 O 1
0 O o" 'o'
0 0 0 o : 0 0 0 0

Once again the reader should reexamine Example 3-4. This last result is the as-
semblage stiffness matrix after processing three elements
The procedure to be used on element 4 is identical to that used above. The
reader should verify that for element 4 we get

109 0 j
-109 0
0 0 0 0
I

- 109 0 109 'o'

j

0 0 ! 0 0.
 APPLICATION TO A SPECIFIC EXAMPLE 67

and with ?i n = 0.949 and n 2] = -0.316,

98 -33 -98 33
4> = -33 11 33 - 11
K' Ibf/in. (3-57}
-98 33 98 -33
L 33 -11 -33 11_

After processing four elements and noting that element 4 connects nodes 4 and
2,
the assemblage matrix K" becomes

115 0 i 0 o !
-115 0 ! 0 0
0 982 0 -981 0 0 i 0 0
j
j
0 0 450 143 -351 - 176 -98 33
| j
f
0 -981 143 1080 - 176 -88 33 -11
K" = 10 3
! !
{ (3-58)
-115 0 T-35f - 176 T 466 176 ! 0 0
0 0 [
— 176 -88 176 88 ! 0 0
j

0 0 1 oo
33 0 0 98 -33
| J

0 0 i 33 -11 |
0 o !
-33 11
1

The pertinent results for element 5 are now summarized. The local element
stiffness matrix K ,5)
is given by
1
589 0 1 -589 0
1

0 0 1
0 0
=
K< S) '

10
3

-589 o'
J..
1

1
1
589 T (3-59)

0 0 1

I
0 0

and with n u = 0.0 and n 2 i

~ 1-0, the global element stiffness matrix becomes

0 0 0 0

K<5) = io
3 “
0 589J 0 -589
0 1
0” 0
1

(3-60)

_0 -589 0 589_ j'

Since element 5 connects nodes 3 and 4 (in this order), the assemblage stiffness
matrix becomes

115 0 J
0 0 i -115 0 0 0
i

0 982 ! 0 -982 | 0 _„o_J 0 0

0 0 ! 450 143 -351 - 176 i
- 98 33
{

0 -982 i
143 1080 !
-176 -88 j 33 -11
K" = 10 3
- -351 - 176 r 466 176"j~ 0 0
(3-61)
115 0 {

0 0 !
- 176 -88 j
176 677 «
0 -589
0 o !
-98 33 J
0 0 i
98 -33
33 -111 0 -589 - 33 600
0 0 i |

All five elements have now been processed, and this last version of the assemblage

stiffness matrix represents the complete assemblage stiffness matrix for the entire
truss. Note that it is symmetric and slightly banded with a half-bandwidth of 6 as

observed from the third row (also see Example 3-5).

68 TRUSS ANALYSIS TI1L DIRECT APPROACH

The next step is to determine the assemblage nodal force vector. Because the
truss and loading in the present example are the same as those in Example 3-6, we
u
may use the result of that example directly. For completeness, f is restated here:

f' = |
- 1000 - 1732 i
0 0[0 0|0 0]
r (3-62)

u
Note the use of the transpose to denote f , a column vector in row matrix form.
Actually, there are three other loads as a result of the reactions. As mentioned
in Sec 3-2, these reaction forces need not be considered in the construction of f".
However, let us include these reactions in P to review why this is the case. Let us
denote the reaction in the r direction at node 3 as U3 and denote the reactions in
the .r and v directions at node 4 as U
4 and V4 , respectively. If these external forces
arc included in the assemblage nodal force vector, we get

r
P = 1-1000 -1732 0 0 !

0 ;
f/4 V4 |
(3-63)

The assemblage system equation K"a = f" becomes

115 0 0 o -115 0 0 0 -1000

0 982 0 -982 0 0 0 0 -1742
u 0 450 " l-»4 -Tsi - 176 -98 33*'
0
0 -982 144 1080 -176 -88 33 -II 0
--nr 0 --irr 466 7f<r 0 ii "i if,

0 0 - 176 -88 176 677 0 -589 0

0 0 r -98 “"VP d (7 98 "“jr «4 u,
0 0 43 -II 0 -589 - 44 600 .' 4 . L Jv «

(3-64)

At this point, the matrix on the left is singular, that is, its inverse docs not exist
and, therefore, impossible to solve for a, the vector of unknown nodal dis-
it is

placements. However, the restraints on some of the nodal displacements must be

considered, namely u% - 0. u 4 = 0, and v4 = 0 Let us use the second method
presented in Sec 3-2 to impose these restraints (The first method is illustrated
numerically in Chapter 4 ) An arbitrary value of 1 0 x 10
1
* is taken as p, and Eq.
(3-64) becomes
‘ ~
115 0 0 0 - 1 15 0 0 0 u - 1000
0 982 0 -982 0 0 0 0 i, -1742
~
0 O 4*>U “RT '=T3T -176 -98 33 ti 0
0 -982 143 1080 -176 -88 33 -II 0
—
-IIS U -3S1 -T76* IU» 176 l> 0 “i 0 X 10"
0 0__ -176 -88 176 67? 0 “ -589 t. 0 ..

u *“98 33 o U IU
n -34 u4 0 x 10"
0 0 33 -11 0 -589 -33 10” _‘ 4 o x tq^.
_

(3-65)
 APPLICATION TO A SPECIFIC EXAMPLE 69

The 10 15 (and not 10 IS ) is entered on the appropriate diagonal entries because of

the 1 0 scalar multiplier. Note that after the restraints are considered, the right-
hand side is effectively

f = { — 1000 - 1732 ;
0 0 i
0 0 ;
0 Of
which is precisely the same as Eq. (3-62). Therefore, the loads as a result of the
reactions need not be considered —a very fortuitous result.
If Eq. (3-65) is solved by the matrix inversion method presented in Sec. 2-14,
the resulting nodal displacements, in inches, are

"l
= -0.00868 »2 = + 0.00996

I'l
= -0.03528 1*2 = -0.03351

«3 = 0.00000 U4 = 0.00000

v3 = -0.00176 ‘4 = 0.00000

Note that the restrained degrees of freedom all have zero displacements, as expected.
The only remaining task is to determine the so-called element resultants, which
include the axial elongations, strains, stresses, and forces. Element 3 is considered
in detail to illustrate the use of Eqs. (3-41) to (3-45). The final results for all

elements are subsequently summarized below. Recall that element 3 connects nodes
3 and 2 (in this order). From Eqs. (3-41) and (3-42) with « u = 0.894 and n 2l =
0.447. the axial elongation 5 for element 3 becomes

5 <3> = («,,«, + ;i
2l vy)
- (n u itt +
or

5 ,J) = [(0.894)(0.00996) + (0.447)( -0.03351)]

- [(0.894X0.0) + (0.447X- 0.00176)]

= -0.00529 in.

The strain in element 3 is readily computed by Eq. (3-43) as

8 -0.00529
e
<31 — 394 x 10 6
in. /in.
L 13.42

the stress from Eq. (3-44) as

a<3> = e E = (-394 x 10~ 6 )(30 x 10

6
)
= -1 1.800 psi

and finally the force as

= ctA = (-11,800X0.1963) = 2320 lbf compression

The reader should determine the axial elongations, strains, stresses, and forces for
the remainins elements (see Problems 3-53 to 3-56). The results are summarized
below.
70 TRUSS ANALYSIS Tf It DIRLCT APPROACH

Element
number 5, in e, in./in ct, psi F. Ibf

l 0 00176 0 000294 8,820 1730

2 -0 00868 -0 000723 -7,960 -1000
3 -0 00529 -0 000395 - 11,800 -2320
4 0Q2004 0 001585 17,400 2190
5 0 00176 0 000176 5,290 1040

Note that elements 1 , (members £, B, and A) are in tension and elements

4, and 5
2 and 3 (members D and C) The maximum force of 2320
are in compression.
(compression) occurs in element 3 (member C) but the maximum stress of 17,400
psi (tension) occurs in clement 4 (member B). In Problem 3-57 the reader is asked
to do a force balance on each node, which serves as a check on these results and
also yields the (unknown) reaction forces l/ 3 U4 and V4 In Problem 3-58, the
, ,

reader is asked to determine these reaction forces directly with the help of Eq
(3-M)
This completes the two-dimensional truss example. In Sec. 3-4 the basic steps
in all finite element analyses are summarized This is followed by the three-di-
mensional truss formulation and a description of a two-dimensional truss program,
calledTRUSS, which automates in a computer program all of the steps executed
above The program may be applied to the problem here, as well as to many other
more complicated two-dimensional truss problems.

3-4 SUMMARY OF THE BASIC STEPS

The simple problem in Sec 3-3 illustrated all the basic steps in the finite element
method In the steps that are summarized below, it is assumed that the problem is

well-posed, a global coordinate system is defined, and all pertinent geometrical and
material data arc know n

Step 1: Discretization

The structure, body, or region being analyzed must be discretized into a suitable
number of elements Each of the dements has several nodes associated with it. For
example, each clement in the truss model has two nodes —
one at either end of the
element. Discretization results in the specification of the finite dement mesh and
involves two distinct but related tasks, nodal definitions and clement definitions.
 SUMMARY OF THE BASIC STEPS 7J

The nodes are always numbered consecutively from one to the total number
of nodes present. The nodal numbering pattern has a strong influence on execution
time in a computer program (for large problems), as explained in Chapters 7 and
8. The basic idea is to number the nodes in such a way so as to minimize the
bandwidth of the assemblage stiffness matrix. Many
canned finite element large,
programs renumber the nodes automatically to ensure this. Nodal definitions are
complete when the coordinates of each of the nodes are also specified.
To define the elements, we first number them consecutively from one to the
maximum number of elements present. The element numbering scheme is com-
pletely arbitrary. The nodes associated with each element must be specified. In
addition, the material property data to be used for each element are specified with
the help of material set flags. Using material set flags reduces significantly the
amount of material property data to be supplied in any computer program.
It should be mentioned that any practical finite element program must have
provisions for automatic or semiautomatic mesh generation. A simple mesh gen-
erator is discussed in Sec. 3-6.

Step 2: Determination of the Local Element Characteristics

Expressions for the local element characteristics must be derived or taken from a
suitable reference. By element characteristics we mean the element stiffness ma-
trices and nodal force vectors. The word “local” refers to the fact that the element
characteristics are derived in a local reference system, which usually changes from
element to element. The proper expressions for the element characteristics in struc-
tural problems may be derived by the direct approach, the variational approach,

including the principle of minimum potential energy, the virtual-work approach,

or the weighted-residual approach. In the truss problem, the direct approach was
used. The other approaches are covered in detail in several of the following chapters.
In nonstructural problems, the variational and weighted-residual approaches are
used most often. With expressions in hand for the local element characteristics, the
local element stiffness matrices (and nodal force vectors) may be determined nu-
merically for each element.

Step 3: Transformation of the Element Characteristics

The element characteristics from Step 2 must be transformed from the local co-
ordinate systems to the global system. This is accomplished with the help of a
transformation matrix as explained in Sec. 3-2 for the truss model. The transfor-
mation of the local element characteristics needs to be performed only when the
72 TRUSS ANALYSIS THE DIRECT APPROACH

unknown parameter function is a vector, such as the (nodal) displacements in the

truss problem, and then only when a local coordinate system is used.

Step 4: Assemblage of the Global Element Characteristics

The global clement stiffness matrices and global element nodal force vectors must
be assembled to form the assemblage stiffness matrix and assemblage nodal force
vector.The basic idea behind Step 4 is that the unknown parameter function must
have the same value at any given node regardless of the element containing the
node. The assemblage procedure for the truss was illustrated symbolically in Ex-
ample 3-4 and numerically in Sec. 3-3,

Step 5: Application of the Prescribed Displacements

After the assemblage step, the assemblage stiffness matrix ami assemblage nodal
force vector must be modified according to one of the two methods presented in
Sec. 3-2 in order to impose the constraints on the restrained degrees of freedom
In structural and stress analysis problems, these arc typically the nodal displacements
(or deflections and slopes in models involving bending) in thermal analysis prob-
lems, the prescribed nodal temperatures must be imposed in a similar manner In
fluid flow problems, the prescribed nodal velocities and pressures must be similarly
imposed. The resulting stiffness matrix after Step 5 is nonsingular lor well-posed
problems.

Step 6: Solution

The resulting system of equations Ka = f must be solved for the vector of nodal
unknowns a. In the truss problem, the vector a contains the nodal displacements
as given by Eq. (3-35). The solution may be obtained by any of ihe methods suitable
to a system of linear algebraic equations In this chapter, only the matrix inversion
method of solution is used. In Chapter 6, another method is presented
In thermal analysis problems the vector a is generally composed of the nodal

temperatures, and in fluid flow problems it is composed of the nodal velocities and
nodal pressures.

Step 7: Calculation of the Element Resultants

The element resultants, such as axial elongations, strains, stresses, and internal
forces, are computed more or less directly from the nodal displacements from the
solution step. In thermal analysis and fluid flow problems, the element resultants
 ,.

Sf'-IM \RY Ol HU t! WO SA| i". 73

are the internal hcat-flow.s and fluid stresses, which are computed from the nodal
temperatures and nodal velocities, respectively.

3-5 THREE-DIMENSIONAL TRUSS FORMULATION

Let us now extend the development in Sec. 3-2 to the three-dimensional truss.
Consider the global .we coordinate system shown in Fig. 3-1 Ha). Note the position
vector r which defines an arbitrary point /\ Let us write the vector r in terms o!
its three cartesian components as follows

r = r,i 4 rj 4 - r.k (3-661

where i, j, and k arc the three unit vectors in the a. y. and r directions, respectively
as shown on the figure.
As in the two-dimensional truss formulation, let us write the same sector r in
terms of its components in the rotated coordinate system .\'y'z' where, with no loss
of generality, the .v’ axis is purposely taken to be in the same direction a*, r itself,
as shown in Fig. 3-1 1(b). Writing the vector r in terms of its components
(r[, r'. and r’) in the rotated system, we have
r = r\ i' 4- r'j' 4 -
r.k' (3-67)

where i\ j', and k' arc the unit vectors in the a', y'. and z directions, respectively

Again the x'y’z' coordinate system is commonly referred to as a local coordinate

system. From Eqs. (3-66) and (3-67), we have

r,i + r, j + r.k = r,'i' 4 -

r'j’ •*
Kk' (3-68)

Forming the dot product of this last equation with / gives

\
Figure 3-11 Position scvti'i r in a three dimension.’! (.0 ek'’’-'' <

liv.i! eootdm.tle s\ oeiTi

74 TRUSS ANALYSIS THE DIRECT APPROACH

r, = r'i-i’ + r' vj'

}
+ r.'iV (3-69)

But

M* = cos(x.r') = nn (3-70a)

ij' = cos(T.y') = n l2 (3-70b)

and

Ms' = cos(x.r') = n„ (3-70c)

Therefore, Eq. (3-69) becomes

r% = n„rl + n ]2 r‘ + n )y r': (3-71)

In a completely analogous manner, wc also have

r\ ~ n u r\ + naTx + n :i r .- (3-72)

and

r. = + n s2r[ + n }y r[ 13-73)

The last three equations may be written in matrix notation as

k, «,2 «n r\

«2 ,
n2 2 »2 \ r[ (3*74a)
[_n„ n,2 r'

or more concisely as

r = Tr' (3-74b)

Equation (3-74b) transforms the components of an arbitrary sector in the local

coordinate system to the components in the global system Since it can be shown
that the transformation matrix T is orthogonal. »e have

r' = T»r = T rr (3-75)

where

nn «;i n„
n,2 0,2

[n,, 02, o,j.

We are now in a position to develop the global element stiffness matrix from
the focal one for the three-dimensional truss Firstofall. we must extend the notation
to include the r and r' components of the nodal displacements and nodal forces
Let us represent the three components of displacement as u, v. » and the three
components of nodal forces as V. V. W
The subscripts i and j are used as usual
to denote the two respective nodes on each element. A prime (') denotes that
the
 THREE-DIMENSIONAL TRUSS FORMULA! ION 75

variable in question is referenced to the local system (i.e., the x'y'z' coordinate

system), while the absence of a prime refers to the global system (i.e., the xyz
coordinate system). Figure 3-12 may help to clarify the notation; only nodal dis-
placements are shown, since the figure showing nodal forces would look identical
except that uppercaseletters would be used to denote the forces.

As in the development for the two-dimensional truss, the two nodal displace-
ments in the x' direction, h' and uj, are related to the two nodal forces in the same
direction, Uj and Uj , by

AE
U', = — (it’, - uj) (3-77a)

and

Uj = —
AE
(uj - uj) (3-77b)

where A represents the cross-sectional area of the element, E the modulus of

elasticity, and L the element length. Now L may be computed from

L = V(.v, - x,f + (y, - y,f + ( 2j -If (3-78)

For the truss element only axial loads are allowed, and so we have
v; = w; = o (3-77c)

and

Vj = Wj = 0 (3-77d)

Figure 3-12 Nodal displacements in xyz and x'y'z' coordinate systems.

76 TRUSS ANALYSIS THE DIRLCT AfTRO^ClI

In other words, all joints are assumed to be pinned. Let us write Hqs. (3-7?) in
matrix form as

100-100
0 0 0 0 0 0
0 0 0 0 0 0
i o "d r~d""d
ooo ooo
ooo ooo
or much more concisely as

where Kr represents the local element stiffness matrix, a

r
the local element nodal
displacement vector, and F the local element nodal force vector
Let us now determine the global clement stiffness matrix. From the development
in Sec. 3-2, it should be obvious that

a' = Ra' (3-80)

and

r = Rf r (3-81)

where

«ll M;i 0 0 0
«i: M>2 ll\> 0 0 0
_ rjyjLl M|J MjV 0 0 0
“ 0 t’J 0 0 0 Mu
[ j M;i »n
0 0 0 »I2 n 22
0 0 0 «U "2i «»
As usual, a' and F. respectively, denote the vector of nodal displacements and the
vector of nodal forces for element e, referred to the .global system These arc given
explicitly by
r
a' = [ Uj v, »*, ii, Vj |
(3-83)

and
T
r = i u, v, ir, u, v ;
w, \
(3-84)

Equation (3-79b) becomes

K' Ra' = RF (3-85)

which may be prcmultiplied by R 1

(which equals Rr ) to give

K'a r = r (3-86)
 J .

INSCRIPTION Of A Sl'im ( OMM IIPW" Ti" 1

81

to which the user responds "6” if the input file is F0RT06 DAT. “7" the inpui it

file is F0RT07.DAT, etc. If the user enters "0" or presses the RETURN key
the program stops. This gives the capability to run multiple cases This is the only
nortion of the program that is system-dependent and it may need to be modified
for use on other systems. The description given here applies to the Microsoft version
of FORTRAN called FORTRAN-80 [7J. and the program runs without changes on
the Apple II Plus microcomputer with a Z-80 card installed, as well as on other
microcomputers with the CP/M‘ operating system
The main program then reads the 80-column title and the master control data
The mesh is generated by calls to subroutines NODGEN and ELEGEN In sub-
routine NODGEN, Section 2 of the input file is read and the nodal coordinates are
specified either on a node-by-nodc basis or by generation. The x and \ coordinates
of node I are stored in XCOOR(l) and YCOOR(I), respectively In subroutine
ELEGEN, Section 3 of the input file, which is used to define the elements and to
specify the material set flags for each element, is read Again, these data may cither
be given on a elemcnt-by-clcment basis or by generation The method used to
generate the mesh is described later in this section For clement L. the global node
numbers I and NODKL) and NODJ(L). respectively, and the material
J are stored in

set flag is stored in MATFLG(L) The material property data in Input Section 4
are read next by a call to subroutine MATERL For material set number MSNO,
the cross-sectional area and the modulus of elasticity are read and stored in DAT-
MAT(MSNO.l) and DATMAT(MSN0.2), respectively
Subroutine BCOND is then called, it reads the boundary condition flags, the
imposed nodal forces, and the imposed nodal displacements. The boundary' con-
dition flags may be specified cither on a node-bv-node basis or by generation. For
node I, the two boundary condition flags in the .r and y direction arc stored in
NBCX(I) and NBCY(I). respectively For nodal force number NFORCE, the value

of the force is read and stored in FORCE(NFORCE) Similarly, for nodal displace-

ment number NDISP, the value of the displacement is read and stored in DJSP(NDISP)
The use of the boundary condition flags is explained in detail in Appendix B
Suffice it to say now that negative flags are used to denote forces and positive flags
to denote displacements. The input data arc then printed in summary form by a tall
to subroutine SUMMRY
In the next section of the main program, the assemblage nodal force vector
and the assemblage T he respective array names arc
stiffness matrix arc ?croed-out

A NFS' and ASM. Since 20 joints or nodes are allowed and each node has two
degrees of freedom (v and \ components of displacement), the dimensions of V AM
and ASM are 40 x and 40 x 40.1respectively

In the DO “2500” the global clement stiffness matrices are

loop labeled
generated by a call to subroutine ST1FT with the help of prior calls to subroutines
COORDS. LENGTH. PROPTY. and TRANS F Subroutine COORDS simply re-

'CP M n a rr.-i<’c:rl tv 5.—. 4 IJ e Pvx-js 1 -

78 TRUSS AN ALT SIS 11 IF DIRECT APPROACH

M- = = «„

In summary, the three direction cosines n M , n ;i , and n*, are given by

X) — X,
flu = cos(.t.t’) = (3-93a|
L
v, — v,
n zl = cos(vjr') = 13.93b]
z

nJt = cos(r„r') = f3-93c)

L

where L. of course, is the length of the element given by Eq. (3-78). For any
element, the global nodal coordinates are known since the> are an integral part of
the input to any finite element program. Therefore, the global element stiffness
matrix K' is easily computed from Eq. (3-88)
The assemblage process remains the same as described in Sec 3-2 except that
each of the four 2x2 submatnees in K' is now a 3 x 3 submatm. Prescribed
displacement boundary conditions are applied in the usual manner External nodal
forces arc also applied in the same manner as described in Sec. 3-2 for the two-
dimensional truss. The resulting assemblage sjstcm equation

Ka = f (3-94)

is solved for the nodal displacements, which in turn can be used to solve for the
internal axial forces and stresses in each clement The procedure is summarized
below. The elongation 8 of an element is given b>

8 = u) - a', (3-95)

where from Eq (3-75). we have

uj = n 1
(i.
(3-96a)

and

u’ ~ n,,uy + m,» y + rtjjn, (3-96b)

Since the direction cosines are readil) calculated with the help of Eq (3-93). and
since the nodal displacements n„ u,. and are now known (i c . from
the solution for a in Ka = F). the elongation 8 is easilv found for each element.
The axial strains e. stresses a. and forces F within each element can then be
determined by using Eqs. <3-43) to (3-45) Negative values of f , o-. and F denote
compression (when 8 < 0) This completes the formulation of the three-dimensional
truss problem
 ,

unsciumoN oi a simru: comruti-r rrixiram truss 79

3-6 DESCRIPTION OF A SIMPLE COMPUTER PROGRAM: TRUSS

A simple FORTRAN program for the analysis of two-dimensional trusses is pre-

sented in this section. A complete listing of the program is provided in Appendix
B. The program is capable of analyzing trusses with up to 20 joints and 30 members.
No more than 5 different types of “materials" may be present with no more than
15 different nodal forces and 5 different prescribed displacements. Although nearly
20 subroutines are used, the program is quite simple. In fact, the use of subroutines
in this manner should facilitate one's understanding of the program. In general,
each subroutine performs only one task or a few related tasks. Any consistent set
of units may be used.
The program is written in the American National Standards Institute FORTRAN
language as described in the ANSI document X3.9-1966 (approved on March 7.
1966). The newer FORTRAN 77 is not used, but the program should run with no
significant changes on systems using FORTRAN 77. The program is meant to be
run from a console where the variable LCONSL in the main program is the console
device number (c.g., 3 on an Apple II microcomputer, 5 on the VAX 1/780). The 1

user should change this number accordingly on his or her particular system.
The input to the program is described in detail in Appendix B. Table 3-1
contains a ready reference for the input to TRUSS and should be used in conjunction
with Appendix B. The input is divided into seven sections. Section 1 begins with
an 80-column which may be used to document the case being run. This title
title

is followed by the master control data, which includes the number of nodes, the
number of elements, the number of materials, the number of different prescribed

displacements, the number of different point loads, and the output device number.
Each of the remaining six input sections is always preceded by a dummy subtitle

whose only purpose is to make the input file more readable.

Input Sections 2 and 3 are used to define the mesh, including nodal coordinate
and element data, as well as the material set flags for each element. As explained
in Appendix B. the nodal coordinate and clement data may be defined on a one-
by-one basis or they may be generated. Input Section 4 contains the material
set present. For the truss, these properties are the
property data for each material
modulus of elasticity. Input Sections 5. 6, and 7 include
cross-sectional area and
theboundary condition Pay data, the imposed nodal force data, and the imposed
nodal displacement data. Input Sections 2 through 7 must be terminated with a
mandatory blank line.

Description of the Program

A main program is show n in Figure 3- 3 Tins flow chart

simplified flowchart of the 1

is typical of all finite clement programs that are based on the

so-called stiffness

approach. The mam program does not do any of the finite element computations —
these are relegated to the subroutines.
 )

80 TRUSS ANALYSIS THE DIRECT APPROACH

Table 3-1 Summary of Input to TRUSS Program

The input to TRUSS is summarized to proside a read) reference once the detailed input
explanations arc understood (see Appendix B). AH titles are read under 20A4 formats, all

integer input variables arc read with IS formats, and all real input variables arc read with
F8 0 formats Integer v anablcs alwa>s begin with the letters I to N. Microsoft's FORTRAN-
80 allows the user to enter the data in free format st>Ie with commas or tabs (control-1)
separating the input Helds

Section l Input
TITLE (80-column title)
RHODES HELBH HMATLS NPDIS NPLDS LOOT

Section 2 Input
SUBT (c g . NODAL COORDINATE DATA)
HI HG NF
XI YI XF YF
(blank line)

Section 3 Input
SDBT (c g . ELEMENT DATA)
LI MS LG LF NG
HI NJ
(blank line)

Section J Input
SDBT (e g MATERIAL PROPERTY DATA)
.

MSNO AREA ELMOD

(blank line)

Section 5 Input
SUBT (e g BOUNDARY CONDITION FLAG DATA)
.

NI IBCX IBCY HG HF
(blank line)

Section 6 Irput
SDBT <e g NODAL FORCE DATA)
.

NFORCE FORCE
(blank line)

Section 7 Input
SDBT (e g.. NODAL DISPLACEMENT DATA)
NDISP DISP
(blank line)

The main program begins by setting LCONSL to the console device number
and by writing on the screen

INPUT THE NUMBER OF THE INPOT FILE ( t-10

 DESCRIPTION OF A SIMPLE COMPUTER PROGRAM. TRUSS 81

to which the user responds “6” if the input file is F0RT06.DAT, “7” if the input
file is F0RT07.DAT, etc. If the user enters “0" or presses the RETURN key,
the program stops. This gives the capability to run multiple cases. This is the only
Mrtion of the program that is system-dependent and it may need to be modified

for use on other systems. The description given here applies to the Microsoft version
of FORTRAN called FORTRAN-80 [7], and the program runs without changes on
the Apple II Plus microcomputer with a Z-80 card installed, as well as on other
microcomputers with the CP/M 1
operating system.
The main program then reads the 80-column title and the master control data.
The mesh is generated by calls to subroutines NODGEN and ELEGEN. In sub-
routine NODGEN, Section 2 of the input file is read and the nodal coordinates are
specified either on a node-by-node basis or by generation. The x and y coordinates
of node I are stored in XCOOR(I) and YCOOR(I), respectively. In subroutine
ELEGEN, Section 3 of the input file, which is used to define the elements and to
specify the material set flags for each element, is read. Again, these data may either
be given on a element-by-element basis or by generation. The method used to
generate the mesh is described later in this section. For element L, the global node
numbers I and J are stored in NODI(L) and NODJ(L), respectively, and the material
set flag is stored in MATFLG(L). The material property data in Input Section 4
are read next by a call to subroutine MATERL. For material set number MSNO,
the cross-sectional area and the modulus of elasticity are read and stored in DAT-
MAT(MSNO,l) and DATMAT(MSNO,2), respectively.

Subroutine BCOND is then called; it reads the boundary condition flags, the
imposed nodal forces, and the imposed nodal displacements. The boundary con-
dition flags may be specified either on a node-by-node basis or by generation. For
node I, the two boundary condition flags in the x and y direction are stored in
NBCX(I) and NBCY(I), respectively. For nodal force number NFORCE, the value

of the force is read and stored in FORCE(NFORCE). Similarly, for nodal displace-

ment number NDISP, the value of the displacement is read and stored in DISP(NDISP).
The use of the boundary condition flags is explained in detail in Appendix B.
Suffice it to say now that negative flags are used to denote forces and positive flags
to denote displacements. The input data are then printed in summary form by a call
to subroutine SUMMRY.
In the next section of themain program, the assemblage nodal force vector
and the assemblage stiffness matrix are zeroed-out. The respective array names are
ANFV and ASM. Since 20 joints or nodes are allowed and each node has two
degrees of freedom (x and y components of displacement), the dimensions of ANFV
and ASM are 40 x 1 and 40 X 40, respectively.
In the DO loop labeled “2500” the global element stiffness matrices are
generated by a call to subroutine STIFF with the help of prior calls to subroutines
COORDS, LENGTH, PROPTY, and TRANSF. Subroutine COORDS simply re-

CP/M is a registered trademark of Digital Research.

Figure 3-13 Flowchart of program TRUSS
 DESCRIPTION OF A SIMPLE COMPUTER PROGRAM TRUSS
83

.v and v coordinates of nodes I and J for element L and stores them in

turns the
XX(1), YY(1) for node I and XX(2), YY(2) for node J. Subroutine LENGTH
computes the length of a straight line between two points and, of course, is used
to compute the element length. Subroutine PROPTY returns the cross-sectional
area (AREA) and the elastic modulus (ELMOD) to be used for element L from the
DATMAT array. Subroutine TRANSF returns the two direction cosines, and
h 2 i, which are needed to obtain the global element stiffness matrix for element L.
These direction cosines are stored in DIRCOS(l) and DIRCOS(2), respectively.
Subroutine STIFF then generates the global element stiffness matrix directly
by using Eq. (3-26). As in the example problem in Sec. 3-3, it is more expedient
to use Eq. (3-26) rather than Eq. (3-25) since the matrix multiplications result in
the disappearance of «, 2 and n 22 as shown in Sec. 3-2. The 4x4 global element
stiffness matrix, stored in array ESM, for element L is immediately assembled into
the assemblage stiffness matrix ASM by a call to subroutine ASSEMK. In other
words, the element stiffness matrix for element 1 is generated and assembled, then
for element 2, and so forth. This reduces the memory required since the individual
element stiffness matrices are not saved. The reader should examine the listing of
subroutine ASSEMK in Appendix B carefully because it gives the algorithm for
the assemblage step in a very' explicit form.
The assemblage nodal force vector ANFV is generated by subroutine ANFVEC
that applies the nodal forces. Then the assemblage stiffness matrix and the assem-
blage nodal force vector are modified to impose the prescribed displacements by a
call to subroutine PDBC. Method 1 of Sec. 3-2 is used to impose these displace-
ments. The system of equations Ka = f is then solved by calling subroutine
EQSOLV which returns the solution for the nodal displacements in the vector SOLN.
Subroutine EQSOLV obtains the solution by calling the matrix inversion subroutine
INVDET [8] (which computes K _l ), and the matrix-vector multiplication subrou-
tine MATVEC (which computes a by postmultiplying K -1 by the vector f). The
nodal displacements are printed by calling subroutine PRINTN. and the element

resultants are computed and printed via a call to the postprocessor subroutine

POSTPR. The element resultants include the following for every element: the axial

elongation, strain, stress, and force.

The main program then returns to the beginning for a new case. The user may
then enter the input file number for a second case. If no more cases are to be run.

the user simply presses the RETURN key on the console to stop execution. Cases

may also be stacked in one input file.

Mesh Generation

All practical finite element programs must have provisions for some type of node
and element generation — referred to collectively as mesh generation. The TRUSS
program can be used to illustrate one simple mesh generation scheme. Some schemes
generate the nodes and elements in two separate steps (such as the one to be presented
here), whereas others generate nodes and elements in one integrated step. References
84 TOUSS ANALYSIS TlIC DIRLCT APPROACH

19—1 1 j
give other mesh-generating algorithms. The scheme used here is patterned
after that presented by Zienkiewicz [ 12).

Node Generation
The easiest way to introduce the node generation scheme used in the TRUSS program
is to illustrate it with an example. Consider the straight line shown in Fig. 3-14(a)

along which six equally spaced nodes are to be placed. Let us assume that the
nodes arc numbered in such a way that the difference between two successive node
numbers is a constant, as shown in the figure. Let NI be the initial node, NG the
nodal increment, and NF the final node. In the present example, we have NI =
3, NG = 4, and NF = 23 Let us further assume that the coordinates of nodes
NI and NF are also Known and are given by (XI, Yl) and (XF.YF), respectively.
The number of equally spaced divisions, D1V, is given by

NF - NI
DIV
NG
which is always a whole number (providing the nodes are numbered properly). The
increments in the x andy directions. DX and DY, arc then given by

XF ~ XI
DX =
DIV
and

YF ~ Yl
DY
DIV
The first node in this generation sequence is completely specified since its node number

is NI and its coordinates arc (XI, Yl) The second node is defined from the first with
the global nodal number given by NI + NG, the x coordinate by XI + DX, and the

y coordinate as Yl + DY The third node is generated from the second in a similar

fashion. This process is repeated until the last node (whose global number is NF) is

generated. In order to asoid round-off errors, it is prudent to assign the known co-
ordinates (XF.YF) to node NF In tfie present example, DIV = (23 - 3)/4 = 5,

Figure 3-14 Illustration of mesh generation for (a) the nodes and (b) the elements
 DESCRIPTION- OF A SIMPLE COMPUTER PROGRAM TRUSS 85

DX — [3 — (— 2)]'5 = I. and DY = (16 — 6)'5 = 2. The resulting nodal coor-

dinates are summarized as follows.

Node number .r coordinate \ coordinate

3 _2 6.
7 -i S.

n 0 10.
15 I. 12.
19 2. 14.

23 3. 16.

In summary. the following variables must be provided as input: NI. NG. NT.
XI. YI. XF. YF. The reader should compare these variables with those required
in Input Section 2 described in Appendix B and summarized in Table 3-1 In Chapter
.

7. this method is modified slightly so that the nodes are not necessarily equally
spaced. This modification allows the practical use of graded meshes.

Element Generation
Like nodal generation, element generation is explained most easily by an illustration.
Let us consider the elements shown in Fig. 3-14(b). It should be noted that not
only is there a constant difference between two successive node numbers, but there
is also a constant difference between two successive element numbers. Let LI be
the initial element number. LG number increment. LF the final element
the element
number, and NG the nodal number increment. Furthermore, let us assume that
nodes i and j for the initial element LI are given by NT and N'J, respectively. This
is also a very convenient place to specify the material set to be used for all elements
in this generation sequence. Let us denote the material set flag as MS. It should
be apparent that the first element is completely defined because the element number
is LI, nodes i and j are NT and NT. and the material set number is MS. From this

element, it is a simple matter to define the second element; the element number is

LI - LG. node i is NT -r NG. node j is NT NG. and again the material set
number is MS. The third element is defined from the second, and so on. until the
last clement LF is defined. For the example above. LI = 4. LG = 7. LF = 32.

and NG = 4. Let us assume arbitrarily that material set 2 is to be used for these
elements. The resulting element definitions in this element generation sequence are
summarized as follows:

Element Node Node Materia!

number I
J set number
->
4 3 7
11 7 ii

IS ii 15 2
*>
25 15 19
32 19 23 2
86 TRUSS ANALYSIS THE DIRECT APPROACH

In summary. the following variables must be specified for each clement gen-
cration sequence: LI. MS, LG, LF, NG, Nl, and NJ. These variables should be
compared with those required in Input Section 3 (Table 3-1 and Appendix B). The
reader should also examine subroutines NODGEN and ELEGEN in Appendix B
carefully so that this method of mesh generation is thoroughly understood.

Example 3-7

Using the TRUSS program, resolve the example problem in Sec. 3-3. More spe-
cifically, determine the displacements of each joint, and for each member determine
the axial elongation, strain, stress, and internal force. The truss is shown in Fig.
3-4(a).

Solution

For convenience, let us use the discretization (or mesh) shown in Fig 3-4(b). A
complete listing of the input file is given in Table 3-2. The reader should study
this input w'lth frequent reference to Table 3-1 and Appendix B A few hey points
with respect to the input arc made below
First note that follow mg the TITLE are the master control data- four nodes,
five elements,two materials, one prescribed displacement, two different point loads,
and 3 for the output device number (the Apple II console) This is immediately
followed by the NODAL COORDINATE DATA subtitle and the actual input Sole
that nodes l and 2 are defined on a node-by-node basis (NG = 0), while nodes 3
and 4 are defined via the generation feature (with NG = 1). These input lines are
followed by a mandatory blank line.
The ELEMENT DATA subtitle and data follow next Elements 1, 3, and 5
arc defined in the first-generation sequence and elements 2 and 4 in the second-
generation sequence From the first sequence note that clement 3 would be defined
as having nodes 2 and 3 (in this order) (l is desirable to use the discretization from
Example 3-1 and for this reason element 3 is redefined in the input file as having
,

nodes 3 and 2 The TRUSS program always uses the latest nodal coordinate and
element data specified. Again a mandatory blank line terminates this input section
(as well as all the following sections).
The MATERIAL DATA subtitle and material property definitions then follow

Note that the material dat3 can be specified in any order. The BOUNDARY CON-
DITION FLAGS subtitle and data follow. Note that the flags for node 2 are not
defined, so IBCX = 0 and IBCY = 0 are used (by default) for this node The
file is completed by the NODAL FORCE LOADS and CONSTRAINED
input
NODAL DISPLACEMENTS subtitles and data. The material data, force data, and
displacement data may be specified more than once for each condition (like the

nodal coordinate, clement, and boundary condition fl3g data) The program simply
uses the last values read when conflicting input data arc provided
The output from the program is shown in Tabic 3-3 and is self-explanatory.
The nodal displacements and element resultants agree with the results from See.
 DESCRIPTION' OF A SIMPLE COMPUTER PROGRAM. TRUSS 87

Table 3-2 Input Data File for Example 3-7

EXAMPLE 3-7
4 5
NODAL COORDINATE DATA
1
:

15
MODEL OF TRUSS SHOWN IN FIG. 3-4
2 3

15.0 0.0
5
15.0 L.O
3 1 4
0 . 0 . 0 . 10 .

ELEMENT DATA
1 1 5 5 1
1 5
5 5 5 4 1
3 1
3 1 0 0 0
3 2

MATERIAL DATA (5 DIFFERENT MATERIALS)

5 0.1557 11.E+0L
1 0.13L3 30.E+0L

1
3
-1-50
BOUNDARY CONDITION FLAGS

1 0 0
D
D
4 1 1 0 0

NODAL FORCE LOADS

1 - 1000 .

5 -1735.

CONSTRAINED NODAL DISPLACEMENTS

1 0.00
(blank line)

3-7 REMARKS

It is important to put this rather lengthy chapter into perspective. The two-dimen-
sional truss was defined, and the complete finite clement formulation was presented.
A simple problem was solved by hand to illustrate the process. The basic steps
were then summarized and extended to the three-dimensional truss. A simple pro-
gram called TRUSS was described; it may
be used to solve more complicated two-
dimensional truss problems. The program illustrates the basic features of most finite
element programs.
Table 3-3 Output from Program TRUSS for Example 3-7

EXAMPLE 3-7 : MODEL OF TRUSS SHOWN IN FIG. 3--4

NUMBER OF NODES: 4
NUMBER OF ELEMENTS: 5
NUMBER OF MATERIALS: 2
NUMBER OF PRES DISP: 1
NUMBER OF PT LOADS: 2
OUTPUT UNIT NUMBER: 3

NODE NO. IBCX IBCY X-COORD Y-COORD

1 -1 -2 12.000 O.ODDO
2 0 12.000 b . DODD
3 1 0 0.0000 O.DDOD
4 1 1 0 0000
. 10.0000

ELEMENT NO. NODE I NODE J MAT SET FLAG

1 1 2 1
2 3 1 2
3 3 2 1
4 4 2 2
5 3 4 1

MATERIAL AREA ELASTIC MODULUS

1 .19b3 3000E + Qfl
.

2 .125? .1100E+0fl

SUMMARY OF DIFFERENT EXTERNAL LOADS

NUMBER NODAL FORCE
1 -10D0.0
2 -1735.

SUMMARY OF PRESCRIBED NODAL DISPLACEMENTS

NUMBER NODAL DISP.
1 O.OOD

SUMMARY OF NODAL DISPLACEMENTS

NODE NO. X-COMPONEWT JT-COMPONENT
1 - flb7Q7E-Q2 . - 35277E-01 .

5 .99S52E-a2 -. 33513E-01
3 O-ODOO ~.17b4bE-D2
4 D.0000
.

SUMMARY OF ELEMENT RESULTANTS

ELEMENT NO. ELONGATION STRAIN STRESS FORCE
x .i?b4bE-02 .e'KiiE-oa aa23.2 i732.u
2 - flL7fl?E-Q2
. 72322E-D3 -79S5.4 -1000.00
3 -.52939E-02 -.39459E-03 -11338. -2323.7
4 .20D42E-0X .15645E-02 17429. Pl^D.fl
5 .17L4LE-D2 .17b4bE-03 5293.9 1039.2

88
 RI. MARKS 89

Although the approach illustrated in this chapter is referred to as the finite

element method, this is but one type of many finite element methods. The method
presented here and throughout this book is referred to as the stiffness approach,
where the primary unknowns a are related to the nodal forces f by the relationship
Ka = f. Not surprisingly, K is referred to as the stiffness matrix. An alternate
finite element method exists, called the flexibility method or the force matrix method,
where the primary unknowns are the nodal forces. The form of the global system
equation is Lf = a, where a, as usual, represents the (now known) nodal displace-
ments. Obviously, the flexibility matrix L is related to the inverse of the stiffness
matrix K. The stiffness approach is much more powerful and popular than the
flexibility approach. For this reason, the stiffness approach is used exclusively in
this book.
Although the finite element method is generally regarded as an approximate
solution technique, it does give the exact solution for the two-dimensional truss
(and the three-dimensional The reason for this is that the elongation 8 is
truss).

given exactly by PLIAE A and £ are constant in any member). The

(providing that
assumption of smooth pin joints, which is also accommodated exactly in the finite
element formulation, also contributes to the exact solution by the finite clement
method. Some argue that because the truss model is exact it should not be considered
a bona fide application of the finite element method. Needless to say. the author
does not agree with this point of view.
Section 3-2 deserves some special attention. It will become evident in later

chapters that most of the steps delineated in this section are routine and are readily
extended to other applications in structural analysis, as well as to problems in other
disciplines. The reader will soon come to appreciate the following steps as being
routine in structural analysis: (1) discretization, (2) assemblage of the stiffness
matrix and nodal force vector, (3) application of the prescribed displacements, (4)
solution for the nodal displacements, and (5) calculation of clement resultants. The
steps that are different for different models are the ones in which we determine the
local and/or global element stiffness matrices and nodal force vectors conveniently —
referred to as the element characteristics. Any one of several methods may be used
to derive the element characteristics: (I) the direct approach. (2) the variational

approach, (3) the virtual-work approach, or (4) the weighted-residual approach. In

this chapter, only the direct approach is used (although it may not seem to be vers'

direct to the newcomer to the finite element method). Nevertheless, a direct and
well-known relationship, namely.

P =

was used to obtain the element characteristics as given by Eq. (3-3) in the local
coordinate system and by Eq. (3-24) (and Eq. (3-26)] in the global system. The
direct approach is not easy to apply when the elements to be used are not obvious.
In these cases other, more indirect, approaches are taken. These other approaches
are discussed in several of die tollowing chapters.
90 TRUSS ANALYSIS THE DIRECT APPROACH

REFERENCES

1. Beer, F. P., and E. R Johnston, Jr.. Vector Mechanics for Engineers: Statics
Dynamics, McGraw-Hill, New York, 1962, p 111.
2. Cheung. Y. K., and M. F Yeo, A Practical introduction to Finite Element Anal)
Pitman, London, 1979, p I.
3. Dcsai, C. S Elementary' Finite Element Method, Prentice-Hall, Englewood Cl
,

N J., 1979, pp 30-31

4. Zienkiewicz, O. C , The Finite Element Method, McGraw-Hill (UK), London, l(

pp 135-136
5. Huebner. K. H , The Finite Element Method for Engineers, Wiley, New York, H
pp 51-53.
6. Zienkiewicz, O C . The Finite Element Method. McGraw-Hill (UK). London, H
p 85.
7. FORTRAN’80 FORTRAN-80 for the Apple
User's Guide, Microsoft II Computer,
crosoft Consumer Products, © 1980. Bellevue. Wash
8. Hombcck, R. W Numerical Methods, Quantum. New York. 1975.
.

9. Zienkiewicz, O C . The Finite Element Method. McGraw-Hill (UK), London, 19

pp 204-207
10. Zienkiewicz, O C , and D V Phillips. “An Automatic Mesh Generation Scheme
Plane and Cursed Element Domains.” Int J Numer Methods Eng ,
vol 3, pp 5
528, 1971
11. Gordon. W J . and C A Hall. “Construction of Curvilinear Coordinate Systems i

Application to Mesh Generation,” Int J Numer Methods Eng ,

vol 7, pp 461-4
1973
12. Zienkiewicz, O C , The Finite Element Method. McGraw-Hill (UK). London. 19
p 693

PROBLEMS

3-1 Discretize the simple Warren truss shown in Fie P3-1 by making tables similai
those mExample 3-1 Use the global coordinate system shown in the Figure 1
truss composed of equilateral triangles w hose sides arc 3 feet long Members
is

C, E. G. I, and K arc made of 0 5-in -diameter steel rods and members D. D. F,

and J of 0.25-in -diameter aluminum rods

D //

Figure P3-1

3-2 Discretize the Pratt truss shown in Fig P3-2 by making tables similar to those
Example 3-1. Use the global coordinate system shown in the figure All memb
 t’ROUIJ MS <J 1

arc square in cross scciion and arc made of steel. The horizontal, slanted, and vertical
members have dimensions 1 x 1cm. 0.75 x 0.75 cm, and 0.5 x 0.5 cm. re-
spectively.

Figure P3-2

3-3 Discretize the cantilever truss shown in Fig. P3-3 by making tables similar to those
in Example 3-1. Use the global coordinate systemshown in the figure. Members A,
B, and C are made of members J and K are of 0.5-in.-
0. 75-in. -diameter steel rods:

diameter steel rods; and the remaining members are of 0.6-in. -square aluminum bars.

Figure P3-3

3-4 Discretize the K truss shown in Fig. P3-4 by making tables similar to those in Example
3-1. Use the global coordinate system shown in the figure. All horizontal members
are 0.75-cm-diametcr steel rods, and all other members are l-em-square steel bars.

Figure P3-4
92 TRUSS ANALYSIS THE DIRECT APPROACH

3-5 Compute the local element stiffness matrix for element 5 in the discretized Warren
truss shown in Fig. P3-5 The truss is composed of equilateral triangles with sides
of length L, and the members arc steel rods with diameter D. where L = 2 m and
D = 1 cm

© © m ©

Figure P3-5

3-6 Repeat Problem 3-5 for element 8 with L = 6 ft and D = 05 in

3-7 For the discretized Pratt truss shown in Fig P3-7. compute the local element stiffnes'
matrix for element 7 if L = 6 ft and h = 9 ft Element 7 has a 05 x 1.0 in
rectangular cross section and is composed of aluminum

Figure P3-7

3-8 Repeat Problem 3-7 for element 5 if L = 1 m and h

= 5 I m Element 5 has a
I x 2 cm rectangular cross-section and is composed of steel

3-9 A cantilever truss is discretized as shown in Fig P3-9 For element 6. determine the
local element stiffness matrix if L = 5 m and h = 7 m. Element 6 is composed of

a 1-cm-diametcr steel rod

o n © m ®

© m ©
Fiaure P3-9
 PROBLEMS 93

3-10 Repeat Problem 3-9 for element 2. which is composed of 0.5-in.-diameter steel. Take
L - 10 ft and h = 14 ft.

3-11 For the discretized K truss shown in Fig. P3-1 1, compute the local element stiffness
matrix for element 9. which is composed of a 0.75-in.-diameter aluminum rod. Take

L — 4 ft and h = 2 ft.

Figure P3-11

3-12 Repeat Problem 3-1 1 forelement 10, which is composed of a 2-cm-diameter aluminum
rod. Take L - 1.5 m and h - 0.75 m.

3-13 For element 5 of the Warren truss in Problem 3-5, determine the global element
stiffness matrix if element 5 connects

a. Nodes 3 and 4 (in this order)

b. Nodes 4 and 3 (in this order)

3-14 For element 8 of the truss in Problem 3-6. determine the global element stiffness
matrix if element 8 connects

a. Nodes 6 and 7 (in this order)

b. Nodes 7 and 6 (in this order)

3-15 For element 7 of the Pratt truss in Problem 3-7, determine the global element stiffness

matrix if element 7 connects

a. Nodes 4 and 7 (m this order)

b. Nodes 7 and 4 (in this order)

3-16 For element 5 of the Pratt truss in Problem 3-8, determine the global element stiffness

matrix if element 5 connects

a. Nodes 3 and 4 (in this order)

b. Nodes 4 and 3 (in this order)

3-17 For element 6 of the cantilever truss in Problem 3-9, determine the global element
element 6 connects
stiffness matrix if

a. Nodes 4 and 5 (in this order)

b. Nodes 5 and 4 (in this order)

94 TRUSS AS A LYSIS THE DIRECT APPROACH

3-18 For element 2 of the cantilever truss in Problem 3-10. determine the global element
stiffness matm if element 2 connects

a. Nodes 3 and 5 (in this order)

b. Nodes 5 and 3 (in this order)

3-1 9 For element 9 of the K truss in Problem 3-11, determine the global clement stiffness

matrix if clement 9 connects

a. Nodes 6 and 8 (m this order)

b. Nodes 8 and 6 (ui this order)

3-20 For element 10 of the K truss in Problem 3-12. determine the global element stiffness

matrix if element 10 connects

a. Nodes 5 and 8 (in this order)

b. Nodes 8 and S (in this order)

3-21 Show that the matrix T defined in Sec 3-2 is orthogonal, i c , show that T" 1
= Tr .

3-22 Show that the matrix R defined b> Eq (3-IS) is orthoeonal, i e.. show that R _l =*

Rr
3-23 Indicate how the global element stiffness matm is added to the assemblage stiffness

matrix for element 5 of the Warren truss in Problem 3-13 if element 5 connecis
a. Nodes 3 and 4 (in this order)

b. Nodes 4 and 3 (in this order)

3-24 Indicate how the global element stiffness matrix is added to the assemblage stiffness

matrix for clement 8 of the Warren truss m Problem 3-14 if element 8 connects

a. Nodes 6 and 7 (in this order)

b. Nodes 7 and 6 (in this order)

3-25 Indicate how the global element stiffness matrix is added to the assemblage stiffness

matrix for element 7 of the Pratt truss in Problem 3 15 if dement 7 connects

a. Nodes 4 and 7 (in this order)

b. Nodes 7 and 4 (in this order)

3-26 Indicate how the global clement stiffness matrix is added to the assemblage stiffness

matrix for element 5 of Problem 3-16 if clement 5 connects

a. Nodes 3 and 4 (in this order)

b. Nodes 4 and 3 (in this order)

6 of Problem 3-17 added to

3-27 How is the global element stiffness mainx for element
the assemblage stiffness matrix tf element 6 connects

a. Nodes 4 and 5 (in this order)

b. Nodes 5 and 4 (in this order)

3-28 Indicate how the global element stiffness matrix for dement 2 of Problem 3-18 is

added to the assemblage stiffness matrix if dement 2 connects

a. Nodes 3 and 5 (in this order)

b. Nodes 5 and 3 (in this order)
 PROBLEMS 95

3-29 How does the global element stiffness matrix for element 9 of Problem 3-19 contribute
to the assemblage stiffness matrix if element 9 connects

a. Nodes 6 and 8 (in this order)

b. Nodes S and 6 (in this order)

3-30 Indicate how the global element stiffness matrix for element 10 of Problem 3-20 is

added to the assemblage stiffness matrix if element 10 connects

a. Nodes 5 and 8 (in this order)

b. Nodes 8 and 5 (in this order)

3-31 Determine the half-bandwidth of the assemblage stiffness matrix for the discretized
Warren truss in Problem 3-5:

a. By direct examination of Ka
b. By Eq. (3-33)

3-32 What is the half-bandwidth of the assemblage stiffness matrix for the discretized Pratt
truss of Problem 3-7:

a. By direct examination of K"

b. By Eq. (3-33)

3-33 Determine the half-bandwidth of the assemblage stiffness matrix for the discretized

cantilever truss of Problem 3-9:

a. By direct examination of K"

b. By Eq. (3-33)
3-34 What is the half-bandwidth of the assemblage stiffness matrix for the discretized K
truss of Problem 3-11:

a. By direct examination of K“
b. By Eq. (3-33)

3-35 A stadium truss is loaded as shown in Fig. P3-35. The global node numbers are also
shown. What is the assemblage nodal force vector if there are no other loads present?

Figure P 3-35
96 TRUSS ANALYSIS THE D1RTCT APPROACH

3*36 A Fink truss to be used in a roof is loaded as shown in Fig. P3-36 The global node
numbers are also shown Determine the assemblage nodal force vector if there are
no other loads present Note that two loads act on node 8.

Figure P3-36

3*37 A cantilever truss is loaded as shown in Fig F3-37 The global node numbers arc
also shown Determine the assemblage nodal force vectorNote the two loads acting
at node 7
 PROBLEMS 97

3-38 A Howe truss is loaded as shown in Fig. P3-38. The global node numbers are also
shown. Determine the assemblage nodal force vector.

Figure P3-38

3-39 For the stadium truss in Problem 3-35:

a. Specify which degrees of freedom are restrained.

b. Indicate how the assemblage system equation K“a = f
a
is modified to impose
these prescribed displacements.

3-40 For the Fink truss in Problem 3-36:

a. Specify which degrees of freedom arc restrained.

b. Indicate how the assemblage system equation K“a = f“ is modified to impose

these prescribed displacements.

3-41 For the cantilever truss in Problem 3-37:

a. Specify which degrees of freedom arc restrained.

b. Indicate how the assemblage system equation K"a = f“ is modified to impose

these prescribed displacements.

3-42 For the Howe truss in Problem 3-38:

a. Specify which degrees of freedom are restrained.

b. Indicate how the assemblage system equation K“a
= f“ is modified to impose

these prescribed displacements.

3-43 Consider the following system of equations:

3.r, - 4.t; - 5.r, = 7

-4r, + Ixi - 2x, = 2

— 5x, — Zx2 + .v 3 = 5

equations to
Using Method 1 of Sec. 3-2, modify (but do not solve) the system of
impose .v
3
= -2. Do not destroy the symmetry.
98 TRUSS ANALYSIS THE DIRECT APPROACH

3-44 Repeat Problem 3-43 using Method 2 of Sec 3-2.

3-45 Consider the following system of four equations and four unknowns

(tti + 2t» + 4tj + 5r4 = 8

2r, + 5t2 - 7r, + 2r4 = 7

4ri — 7 v2 + &tj - 7r4 = 2

5t| + 2c 2 ~ 7jcj + 3t 4 = 5

Using Method 1 of Sec 3-2. modify (but do not solve) the system of equations to
impose X| = 5 and .tj = 3 (concurrently) Do not destroy the symmetry.

3-46 Repeat Problem 3-45 using Method 2 of Sec 3-2

3-47 Sotve the resulting system of equations from Problem 3-43 for the unknowns (after

imposing Xj = -2) Use the matrix-inversion method

3-48 Solve the resulting system of equations from Problem 3-45 for the unknowns (after

imposing x, = 5 and x } = 3) Use the matrix-inversion method

3-49 Reconsider the Warren truss in Problem 3-5 If nodes 3 and 4 arc found to have
displacements

u, = 0 0100 cm u4 — -0 0150 cm
vj =* 0 0200 cm v4 = 0 0015 cm
for some loading condition, determine for element 5 the axial

a. Elongation b. Strain c. Stress d. Force

3-50 Reconsider the Pratt truss in Problem 3-7 If for some loading condition, nodes 4

and 7 arc found to have displacements

«4 = 0 0100 in r/ 7 = -0 0020 m
v 4 = 0 0300 in v, = 0 0045 in

Determine for element 7 the axial

a. Elongation b. Strain c. Stress d. Force

3-51 For some loading condition, nodes 4 and 5 for the cantilever truss in Problem 3-9
arc known to have the following displacements.

«4 = 0 0100 cm uj = -0 0315 cm

v4 = 0 0000 cm V, «= -0 0126 cm
Determine for element 6 the axial

a. Elongation b. Strain e. Stress d. Force

3-52 K truss in Problem

For some particular loading condition, nodes 6 and 8 for the
3-1 1 are known to have the following displacements

Ut,
= 0 0000 in w* = — 0 0269 in

v*- -00158 m », = +00105 in

 PROBLEMS 99

Determine for element 9 the axial

a. Elongation b. Strain c. Stress d. Force

3-53 For the truss described in Sec. 3-3 and shown in Fig. 3-4, determine the element
resultants for element 1 by using the results in Sec. 3-3 (show the calculations).

3-54 For the truss described in Sec. 3-3 and shown in Fig. 3-4, determine the element
resultants for element 2 by using the results in Sec. 3-3 (show the calculations).

3-55 For the truss described in Sec. 3-3 and shown in Fig. 3-4, determine the element
resultants for element 4 by using the results in Sec. 3-3 (show the calculations).

3-56 For the truss described in Sec. 3-3 and shown in Fig. 3-4, determine the element
resultants for element 5 by using the results in Sec. 3-3 (show the calculations).

3-57 Verify the values obtained for the forces in each member of the truss in Sec. 3-3 by
doing a force balance on each node. Remember to include the unknown reaction
force Uj at node 3 and forces U4 and V4 at node 4. From the force balances, determine
the values of U3 U4, , and VA for static equilibrium.

3-58 With the help of Eq. (3-64) and the results for the nodal displacements, determine
the unknown reaction forces U U4
3 , ,
and V4 for the truss in Sec. 3-3. Are these results
consistent with those from static equilibrium considerations?

3-59 Reconsider the stadium truss shown in Fig. P3-35. The nodal coordinates may be
obtained from the superimposed grid. All horizontal members are made of 1-cm-
diameter steel rods and all remaining members of 0.75-cm-diameter aluminum rods.
The truss is loaded and restrained as shown in the figure. Using the TRUSS program,
determine the nodal displacements and the element resultants (for all nodes and
elements, respectively).

3-60 A Fink truss is loaded and restrained as shown in Fig. P3-36, AH horizontal members
are made of 0.5-in. -diameter steel rods and all other members of 0.75-in. square

aluminum bars. The nodal coordinates may be taken directly from the Figure with
the help of the superimposed grid. Using the TRUSS program, determine the nodal
displacements and element resultants.

3-61 Reconsider the cantilevertruss shown in Fig. P3-37. The horizontal, slanted, and

vertical members made of 0.75-in.-, 0.5-in.-, and .0-in. -diameter steel rods.
are 1

The nodal coordinates may be taken directly from the figure with the help of the
superimposed grid. The truss is loaded and restrained as shown. Using the TRUSS
program, determine the nodal displacements and the element resultants.

3-62 The Howe truss shown is to be analyzed. The truss is loaded and
in Fig. P3-38
restrained as shown. All verticalmembers are made of I.5-cm square steel bars and
all other members of 2.0-cm-diameter aluminum rods. The nodal coordinates may
be obtained directly from the figure. With the help of the TRUSS program, determine
the nodal displacements and the element resultants.

3-63 Modify the TRUSS program in Appendix B so that it can be used to analyze three-
dimensional trusses. Make use of the formulation m Sec. 3-5.
 Variational and Weighted
Residual Formulations

4-1 INTRODUCTION

In Chapter 3 the direct approach was introduced when the two-dimensional truss
problem was formulated. was pointed out that alternate, more indirect approaches
It

exist that may be used to develop finite element models. Some of these methods
of formulation are introduced in this chapter. The reader will soon come to appreciate
that virtually all problems that are describable by ordinary and partial differential

equations can be solved by the finite element method. Moreover, linear and non-
linear problems may be solved in nearly the same way, although nonlinear problems
generally require an iterative solution.
In this chapter only steady-state problems are studied. By steady -state we mean
that the field variable is a function of spacial coordinates only and not a function
of time. Problems that allow for time-varying field variables are said to be unsteady,
transient, dynamic, or time-dependent. In some texts transient problems are referred
to as propagation problems [1]. In structural and stress analysis, time-independent
problems are referred to as static or equilibrium problems, whereas time-dependent
problems are almost exclusively referred to as dynamic.
In all steady-state, static, or equilibrium problems, a system of algebraic equa-
tions results that is always in the form

Ka = f

It should be recalled that the truss model resulted in a system of equations in exactly

this form [see Eq. (3-39)]. Furthermore, the vector or column matrix a always
contains the nodal unknowns, which are really the values of the field variables at
the nodal points. It cannot be emphasized enough that in structural models (without
102 VARIATION XL AND WEIGHTED RESIDUAL FORMULATIONS

bending), these unknowns arc the nodal displacements: in thermal models, they arc
the nodal temperatures: and in fluid flow, problems, they are the nodal \elocities
and pressures. When a fluid mechanics problem is formulated in terms of velocities
and pressures, the formulation is said to involve the primitive variables. This is in
contrast to stream function and vorticity formulations that arc less direct [2J. The
main purpose of a finite element analysis is to determine the \alues of the field
variable(s) at the node points. Other quantities such as the stresses or heat-flows
may then be determined in subsequent calculations.
This chapter is further restricted to the study of only one-dimensional problems.
This means that the field \anable of interest is a function of only one variable,
e.g., .r. Two- and three-dimensional problems are covered in subsequent chapters.
In addition, variational calculus is introduced in this chapter to familiarize the reader
vv one of the most popular types of finite element formulations.
ith

The basic difference between ordinary differential calculus and variational

calculus may be explained as follows. If we want to determine the value of .r that
maximizes or minimizes some function v = fi 0, we simpl> take the derivative of
y with respect to x, set the result equal to zero, and solve for t. The sign of the
second derivative at this value of x indicates whether a minimum, maximum, or
point of inflection has been found On the other hand, if we want to know what
function results in a certain definite integral taking on a minimum or maximum
value, then the calculus of variations is required For example, if we want to obtain
the equation of the “curve” that results tn the shortest distance between two points
:n a plane, the calculus of variations may be used. This very example is illustrated

later in this chapter and gives the expected result, the equation is that of a straight
line that connects the two points

4-2 SOME APPROXIMATE SOLUTION METHODS

Oftentimes exact solutions to differential equations are not easilv obtained In fact,

quite frequent!}, it is riot even possible to obtain an exact solution with current!}
available mathematical techniques. This is particular!} true when the geometry is
irrrgnfar. or the properties v ary spacratty. or perhaps, the profxfrties are a function
of the field variable Problems of the latter variety are said to be nonlinear En-
gineers. mathematicians, and applied scientists find approximate solution method*
indispensable in these cases.
Among the approximate solution techniques that arc useful to a thorough un-
derstanding of how the finite element method works are the following, rhe Ruz
method, the variational, or Ravletgh-Ritz. method, and the weighted-residual method.
Each of these is said to be an integral method because we work with an integral
form of the problem statement instead of the governing differential equation directly.
The Ritz and variational methods are illustrated in this section
The Ritz method is quite simple, requires no additional mathematics bevond
calculus, and is, in fact, a special case of 3 particular weighted-residual method
(see Sec. 4-6). The variational method, on the other hand, requires some know ledge
 SOME APPROXIMATE SOLUTION METHODS
103

of variational calculus— a definite disadvantage. For this

reason, Secs. 4-3 to 4-5
are devoted to the development of some of the more basic
concepts in variational
calculus. The weighted residual methods are deferred
until Sec. 4-6. The reader is
referred to the book by Arpaci [3] for a review of the approximate
solution methods
based on integral formulations to problems in heat conduction.
In some books the Ritz method is referred to as the Rayleigh-Ritz method.
Unfortunately, the variational method is also referred to as the Rayleigh-Ritz method.
In this text, the Ritz method always refers to the nonvariational
integral formulation
and the Rayleigh-Ritz method always refers to the variational formulation.

General Concepts

Before actually illustrating the Ritz and Rayleigh-Ritz methods with a specific
problem, let us establish some general concepts. So as not to obscure the basic
ideas behind the approximate solution methods, let us restrict the present discussion
to a single governing differential equation with only one independent variable. Let
us represent the governing equation as

f[T(x)] = 0 in 0, (4-1)

where T represents the function sought (e.g., temperature) that is a function of x

only. The symbol Cl represents the domain of the region governed by Eq. (4-1).
In addition, let us specify the boundary conditions symbolically in the form

gdT(x)} = 0 on r, (4-2a)

SiPix)) = o on r2 (4-2b)

etc., where T ,
and T? include only those parts of the domain fl that are on the
boundary. Figure 4-1 may help to clarify the notation.
Let us approximate the solution to Eqs. (4-1) and (4-2) with the approximate
function T where
n
T - T'(x; o2, . . . , a„) ~ 2 a,N,(x) (4-3)
i=l

which has one or more unknown (but constant) parameters a , a 2 t , . . . , a„ and

boundary conditions given by Eqs. (4-2) exactly. Note
that satisfies the that the

Figure 4-1 Schematic of one-dimensional problem domain fl with two global boundaries
T, and r\.
104 VARIATIONAL AND WLIGIITLD RLSIDUAL IORMULATIOSS

prime (’) denotes an approximate solution (not a derivative). The functions N,(x)
arc referred to as trial funt lions. The problem reduces to having to make somewhat
judicious choices for these trial functions and solving for the parameters o,, <i ,
2
.... fl„. In general, if a sequence of approximations could be made, such as

T = T(t;a,) = £7 ,JV,(t) (4-4a)

T ~ 7”(T;tf|,flj) = a,N,(x) + a 2 N2 ( t) (4-4b)

T - 7"(r.n,,fl2 ,<»,) = a,N t

(x) + a 2 N 2 (x) + a3 N3 (x) (4-4c)

then presumably better accuracy could be obtained w ith each successive higher-
order approximation Equations (4-4) are referred to as the first-order, second-order,
and third-order approximations, respectively The reader is referred to Becker,
Carey, and Oden (4) for a discussion on the necessary properties of the trial func-
tions. Suffice it to say here that they must be continuous and differentiable up to
the highest order present in the integral form of the governing equation
It should not be surprising that if T' given in Eq (4-3) is substituted for T in
Eq (4-1), the governing equation will not be satisfied exactly. Instead of getting
f{T') equal to zero, we get a residual R Mathematically, we may write

<i|, <i 2 , , o„)| = R(x, a,. a2 , , a„) (4-5)

where the notation is supposed to indicate that the residual R is a function of x and
the parameters ri|, a2 , etc The exact solution results when the residual R is zero
for all points in the domain H For the approximate solution methods, the residual
is not in general zero everywhere in 0. although it may be zero at some selected
points

The Ritz Method

Only the first-order Ritz approximation

is considered here Refer to the book by
Arpaci |5) for an of the second-order Ritz approximation. The basic
illustration

Ritz method for the first-order approximation simply requires that the integral of

the residual /?(r,«i) with respect to x be zero over the domain fl. or

t:o,)<ir = 0 M-61
fg<

Note that since .x is essentially a dummy variable in the integral. Eq (4-6) results
m an algebraic equation in the unknown parameter For well-posed problems
and a wcll-bch3ved trial function A',(t). this equation may be solved for a,. Well-
behaved trial functions include those included m the set of polynomials, circular
functions, and other continuous and differentiable functions This
simple method

is best illustrated by an example.

 x

SOME APPROXIMATE SOLUTION METHODS JQ5

Example 4-1

Solve the ordinary linear differential equation

—
d T
dx~
+r lOOO.v
2 = 0 0 < .r < 1 (4-7)

subject to the boundary conditions

T( 0) = 0 (4-8a)

and

7(1) = 0 (4-8b)

by using the Ritz method with the trial function

W,(x) = 41 - X2 ) (4-9)

This particular trial function was chosen because it allows T to satisfy the boundary
conditions exactly, and it does not grossly violate the physics of the problem as
explained later.

Solution

The first-order approximation T from Eq. (4-4a) becomes

T - — a x( 1
{
- x2 ) (4-10)

An expression for the residual R is needed, but first verify that T given by Eq.
(4-10) satisfies the boundary conditions exactly (see Problem 4-1). The residual R
by definition may be computed from

R = —Tr +
d2
ax 2
lOOO.v
2

And since

d 2T'
— 6a ix
dx 2

we have
R(x\a } )
— —6a x
x + lOOOx 2 (4-11)

Note that as the notation R(x\a t ) indicates, the residual is indeed a function ofx
and the parameter a,. By Eq. (4-6), the first-order Ritz method requires

— 6a x + t
lOOO.r
2
) dx = 0
J^(

which may be readily integrated to give

— 3a, 2
+ 1000 :
o
106 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

Evaluating and solving forcii gives

a i
= %
,0<1
(4-121

Therefore, an approximate solution is given by

T(x) = '«% r(l - x2) (4- 13)

fti See. 4-6, this approximate solution is compared with the other approximate
solutions as well as with the exact solution.

Let us try to give some physical significance to the governing differential

equation given by Eq (4-7) The reader may recognize this as a special form of
the heat conduction equation. More specifically, Eq. (4-7) represents onc-dimen
sional heat conduction in a bar of unit length insulated around the periphery with
an internal heat source (internal energy generation) that is proportional to the square
of jt as depicted in Fig 4-2 Therefore, higher temperatures are expected near the
end x ~ although the temperatures at both ends must be zero by Eqs. (4-8) In
other words, the maximum temperature T is expected to occur somewhere between
x ~ 0.5 and x = 1 0. It is tor this reason that the trial function given by Eq
(4-9) is most appropriate, for it too shows a maximum temperature between x =
0.5 and x = 1 0 (in fact, the reader may wish to show that the maximum occurs
at x = 0 577). In short, the trial function (and hence, the approximate solution)
used m Example 4-1 is said to satisfy, albeit approximately, the physics of the
problem. This may be stated differently as follows the assumed temperature dis-

tribution does not grossly violate the physics of the problem

It is emphasized that two requirements must be met by the assumed trial

function(s). (1) the physics of the problem must not be grossly violated and (2) the
boundary conditions must be satisfied exactly Later it will be seen that the finite

Schematic of problem posed Example 4-1

Figure 4-2 in
 SOME APPROXIMATE SOLUTION METHODS 107

element method completely eliminates these rather restrictive conditions, which are
very difficult to satisfy in most real-world problems.

The Variational Method (Rayleigh-Ritz)

Although variational calculus is covered in the next several sections, it is illustrative

to consider the main idea behind the variational method here in a numerical example.
A typical one-dimensional problem in variational calculus is one in which we trv
to find a function T that minimizes or maximizes integrals of the form

/ = f
Ja
F[.r.F(x).Ft (.r)] dx (4-14a)

where Tx denotes the first derivative of T with respect to .r. The processes of
minimization and maximization are frequently referred to collectively as extremi-
zation. and the integral is said to be extremized or to be made stationary. This
integral / is referred to as a functional: sometimes the integrand F is also referred
to as the functional. This is not terribly surprising because the word “functional”
means function of functions. Since F is a function of .v. T. and 7\. and T is a
function of x itself, clearly / is also a function of functions. It should be emphasized
that the notation Tx denotes the derivative of T with respect to .r. The prime (') is

used to represent an approximation, not the derivative (i.e. . T' is the approximation
to T, not the derivative of T). The subscript notation for the derivatives proves to
be very convenient in two- and three-dimensional variational formulations (see
Chapter 8).
It be shown in Sec. 4-5
will that the functional F that corresponds to Eq.
(4-7) from Example 4-1 is

UdT
F = I000.r
2
F (4-15)
2\dx
Consequently, the variational formulation of the problem in Example (4-1) becomes

- dT
IOOO.wF dx {4-1 4b)
dx

The idea is to find the function T{x) that extremizes l. Before illustrating how Eq.
(4- [4b) is used in an approximate solution, several key points must be made.
For a weli-posed problem such as that in Example 4-1. the function Tlx) that
extremizes / in Eq. (4- 14b) is exactly the same as that which satisfies the original
differential equation and boundary conditions. This implies that the solution T(x)
is unique, which alwavs the case for well-posed problems. A second observation
is

is that the original differential equation [Eq. (4-7)] contains a second-order derivative

of T. whereas the variational formulation given by Eq. (4-14b) contains only a first-
order derivative! Therefore, the variational form may be used to obtain solutions
to problems that are not readily admitted by the differential formulation. An example
108 VARIATIONAL AND V, LIGHT! D R LSI DUAL IORMULATIONS

of this is a body with two different materials and, hence, thermal conductivities.

At the point where the two materials meet, the second derivative of the temperature
required by the differential formulation may not exist. A variational formulation of
the problem would readily yield the correct solution, since the second-derivative
in thisexample is not needed in the formulation. For this reason, the variational
formulation of a physical problem is often referred to as the weak formulation.
Although the discussion in the previous paragraph alluded to an exact solution
of Eq. (4-14b) for the function T(x) that extremizes the functional /, we shall use
this equation to obtain an approximate solution for T(x), as shown in the next
example. In this text we will only be concerned with the use of variational for-

mulations in approximate solutions, except for a simple illustrative problem in See.

4-3.

Example 4-2
Solve the problem posed in Example 4-1 by the variational method by extremizing
/ given by Eq (4-14b) with respect to parameter a in the approximation T' (to T) x

given by

r< t) = <i, AM*) = - t-’) ( 4 - 16 )

Note that this same form of the approximate solution was assumed m Example
4-1.

Solution

In terms of the approximation 7", Eq. (4- 14b) may be written

/(<!,) = J^IOOOrT’ - dx (4-171

j j

Note that / is a function of n, (as indicated by the notation /(«[)! because T is a

function of x and the parameter a lt and the .r is integrated-out The idea is to

determine a suchx
that

—=
da x
0 (4-181

or, in words, such that the functional / is extremized (or made stationary) To
determine whether a minimum or maximum has been found, we may check the
the
second derivative of / with respect to a x at the value of a x that extremizes / in
first place. If d'I!da\ >
0, a minimum has been found, if d'lhla ? 0. a maximum <
derivative of T{x) needed Eq (4-17). which is
has been found. The first is in

readily computed from Eq (4-16) to be

r, = = a,(l - 3r-’)
(4-19)
(lx
 SOME APPROXIMATE SOLUTION' METHODS

Therefore, Eq. (4-17) becomes

/(«,) = £{100ar 2 [fl,.v(l - x 2 )] - - 3a 2 )] 2 } dx

Integration and evaluation readily yields

/(a i )
= ,mo/i2a
l
— Vsaj (4-20)

from which we may compute

—=
da ,
I«X)/
12 _ 4/sa

Setting the right-hand side to zero and solving for «, gives

a, = 5000
/48 (4-21)

An approximate solution to Eq. (4-7) by the variational method subject to the

boundary conditions given by Eqs. (4-8) is

T'(x) = 500
%sa(1 - a2) (4-22)

The reader should compare this approximate result with that from the Ritz method
of Example 4-1. The approximate solutions from Examples 4-1 and 4-2 are shown
for comparison in Fig. 4-3. These approximate solutions are compared to the exact

solution later in this chapter. 13

Figure 4-3 Comparison of approximate solutions to problem posed in Example 4-1

110 VARIATIONAL AND WEIGHTED RESIDUAL KJRMULATIONS

4-3 VARIATIONAL CALCULUS: AN INTRODUCTION

The calculus of variations is introduced in this section for functions that are a
function of one independent variable only, e.g., v = f(x). The basic approach that
is taken utilizes only ordinary differential and integral calculus. The result is then
interpreted in light of the calculus of variations. In what follows, when a derivative
is needed the subscript notation from Sec. 4-2 is used; e.g., ilfldx is denoted as
Tx (not as V) and dyfdx as y,.
In general, we usually know the differential equation that describes a phenom-
enon. However, the variational formulation is usually not obvious. For example,
is it obvious that Eq. (4- 14b) is the variational formulation that corresponds to the
differential formulation given by Eqs. (4-7) and (4-8)? The idea then is to develop
a systematic procedure by which we may derive the variational form of a problem
from the differential formulation.
It is very important to put the material in this section into proper perspective.
The author is of the opinion that although the variational formulation allows us to
obtain a firmer grasp of the underlying concepts behind the finite element method,
it is by no means absolutely essential to the application of the method to practical
problems. It will be learned later that the weighted-residual methods arc far easier

to apply and can be used even when no variational formulation or principle exists.
This last point should be clarified All variational formulations have a corresponding
differential formulation, but, unfortunately, the converse is not true some differ-

ential formulations have no classical variational principle The reader is referred

to Zienk lewicz [6] for a procedure that yields a nontlassical variational formulation
to some of these problems. This does not mean that the finite element method
cannot be used in these cases, because we simply resort to the more powerful (and
simpler) weighted-residual approach.
Before a systematic procedure is developed that will allow us to obtain the

variational form of a problem from the differential formulation, let us examine one
of the classical problems in variational calculus, that of finding the equation of the
curve that has the shortest length between two points in a plane Obviously, the

curve is a straight line joining the two points, but let us show how the calculus of

variations may be used to prove this intuitively obvious result First, we must set

up an integral / that represents the length of the curve. Figure 4-4 shows a family
of curves that connects points A and B in the plane of the figure By minimizing
the integral l, we will obtain the equation of the line that results in the shortest

possible length.
An elemental arc length tls from the Pythagorean theorem is given by

4 ’ 231
<ls = Vdh? + (<fv)' <

which may be written as

* = i +
(*)*
(4-24)

J
 v \r.i *t:on \l calclix s an introdlchov m

Figure 4-4 Far".:'. cf eunea that corr.cc is. po : rts A ar.d S m the plane of the paper.
 )

112 VARIATIONAL AND WEIGHT! D RESIDUAL fORVIULATIONS

from which it may be concluded that yx itself is (another) constant, or

vx = C,

But y, = dyfdx and so d\ULx = C2 , which may be integrated to give

y = C2 x + C 3 (4-30)

The integration constants C2 and C3 could then be determined for a particular

problem by using the boundary conditions

j(o) = v, (4-3 la

>tf>) = y2 (4-31 b)

In any event, Eq. (4-30) represents a straight line as expected

Figure 4-5 shows the function \(r) that actually results in the shortest length
between the two points A and D In addition, the variation of y, denoted as 5\, is
also shown. If v(v) and its variation 8v are added, we get other possible curves \(t)
for which the integral / may be evaluated It is emphasized that in this case only
one particular curve, namely that labeled \(r), actually results in the shortest length.

4-4 SOME ADDITIONAL MATHEMATICS

Several concepts from ordinary calculus are reviewed in this section An extension
of these concepts to variational calculus is also provided. Since our goal is to
develop an operational understanding of these concepts, no attempt is made to prove
them here The reader may wish to consult the references at the end of this chapter
for additional information

Figure 4-5 The curve that results in the shortest length connecting A and B is shown along
with the variation 8* and the curve 1(1)
 )

SOME ADDITIONAL MATHEMATICS H3

Integration by Parts [7]

If dit/dx and dvtdx are continuous functions of x over a suitable interval, e.g., from
x = a to x = b, then the formula for integration by parts applies and is given by
dv rb du
~dx — UY
r- dx
L
-f Ja
(4-32)

The first term on the right-hand side is frequently referred to as the integrated term.

Taylor’s Series [8]

A function fix) that is continuous and differentiable over an interval that includes
x = a may be written in terms of an infinite series such as

f{x) = fid) + + • •
(4-33)

This infinite series is the Taylor series expansion for the interval. Note that f(a)
denotes that the first derivative of / with respect to x is to be evaluated at .t = a,
that f^{a) denotes that the second derivative of / with respect to x is likewise to be
evaluated at x = and so on. Oftentimes the terms involving the factors (x - a)V
a,
nl for « & 2 are small in comparison to the first-order term and the truncated series
given by

fix) = fid) + fr(a)(x - a) (4-34)

is often referred to as a first-order Taylor expansion (of the function / about the

point x = a). It is emphasized that Eq. (4-34) may be used only for values of x
very' close to a.

The Differential of a Function

Consider a continuous and differentiable function f(x,y that has two independent
variables, x and v. The differential df is then given by

df= —dr + — rfv (4-35)

fix dv

For a function f(x,y,z) with three independent variables x, y, and z, we have

df
J =—dx + —dv + -d: (4-36)
dx dv '
dz

and so on. Note that we may take the differential of a function but not of a functional,
which is a function of functions. Instead we speak of a variation, as discussed next.
1 1 4 VARIATIONAL AND W EIGIITED RESIDUAL FORMULATIONS

The Variation of a Functional

Lei us now consider a functional F{x,\,yx ). It should be obvious from the example
depicted in Fig. 4-5 that the variation is not taken on the variable x but rather on
the function j(t) and possibly some of its den vat in es. Therefore, the variation of
x, wntten 8.t, is identically zero, and the vanation of F is given by

— 8y + — 5v
__ dF c flF„
SF = x (4-37)
fly fly.

It is hinted here that the usual rules of ordinary calculus apply in the calculus of
variations, and in most cases they do. For example, Eq. (4-37) follows from Eq.
(4-36) if the differential operator d is replaced by the variational operator 8 and. if
further, it is recognized that 5.r s 0 Of course,/ in Eq (4-36) is a true function,

whereas F in Eq. (4-37) is a function of functions; i.e., a functional. One other

subtle difference exists in the expressions for the differential: the variables x, y,
and z are truly independent, whereas in the expression for the variation, y and y\
are treated as though they are independent (they are actually related by y, dyf =
dx). Equation (4-37) may be readily extended to cases in which the functional F
is a function of more than two functions (see Problem 4-15)

The Commutative Properties

One of the most important properties of variational calculus is the commutative

property that states that the differential of the variation of a function v is identical

to the variation of the differential of the same function. The commutative property
may be wntten mathematically as

d( 8v) = 8 (d\) (4-381

A simple proof of this relationship is provided in reference (9) It is assumed here

that the function y is differentiable
Another commutative property is the one that states that the vanation of the
integral of a functional F is the same as the integral of the vanation of the same
functional, or mathematically as

6 JV <ir = fl F dx H-331

Note that the two integrals must be evaluated between the same two limits

Miscellaneous Rules of Variational Calculus

the definitions of differentiation and vanation. it can be shown that all the
From
z be any continuous
rulesof differentiation hai e vanational counterparts. Let > and
and differentiable functions. Then we may wntc
 THE EULER-LAGRANGE EQUATION 115

8( v + z) = 83’ + Sz (4-40)

8( vz) = y 8z + z 83’ (4-41)

_ z 83- - v 8z
(4-42)

8(y") = /iy"
_l
8y (4-43)

and so forth. The similarity between these and each of their counterparts in dif-
ferential calculus should be obvious.

4-5 THE EULER-LAGRANGE EQUATION: GEOMETRIC

AND NATURAL BOUNDARY CONDITIONS

In this section itwill be seen how we may obtain the variational formulation from
the differential. As mentioned earlier, a systematic procedure to accomplish this is
desirable because, more often than not, the differential equation that describes a
particular problem is easily derived and the variational form is not. Obtaining the

variational form from the differential allows us to take advantage of some of the
approximate solution techniques — including the finite element method itself (see
Sec. 4 - 8 ).
The necessary condition for the existence of an extremum of the functional

/ = f F(x,y,y x ) dx (4-44)
Ja

is that its first variation 8/ must be zero, or

8/ = 8 [ F(x, y,)\) dx = 0 (4-45)

Ja

provided that

= 0 (4-46)

is satisfied [ 9 ]. We now want to find the condition(s) that y(.v) must satisfy such

that the integral / is extremized or made stationary. Let us proceed by noting that

by Eq. ( 4 39 ) the variation of the

- integral is the same as the integral of the variation,

and so Eq. ( 4 - 45 ) becomes

8/ = p8F(A,y,y,) dx = 0 (4-47)
Ja

Now with the help of Eq. ( 4 - 37 ), we may write

b/dF
8/ = 8v + dx (4-48)
dv
1 6 VARIATIONAL AND \S LIGHTED RESIDUAL FORMULATIONS

Let us rewrite 8yx in a more convenient form by using Eq. (4-38):

. , _ Htfy) _ d(Sr)
( 4 -49 )

Therefore, Eq. (4-48) becomes

8' = (4-50)
£l By dyM dx \

Let us now integrate by pans the second term in the integral (see Eq. (4-32)) with

u = —
dF
dy.
and
dv
dx
m d( 8v)

cZr
(4-51)

from which it follows that

du _ d /1
=
dx dx \o 3 and v 8> (4-52)

Therefore, Eq. (4-50) becomes

61 =
I 8
>{
+
r[s -*©]»*-• (4-53)

One way for Eq (4-53) to be true is if

—9F.
ty,
8 .v 1
= 0 (4-54)
|

for then we have

(4-55)
io [a.v dx'

Note that the condition expressed by Eq. (4-54) is identical to that already stipulated
by Eq. (4-46). Since Eq. (4-55) is to hold for all arbitrary vanations 5>. the only
way for Eq. (4-55) to hold in general is if the expression in the brackets itself is

zero or

_ 0 14 - 56 )
By dx\dyx J

This last result is really a differential equation with y and y, as the

pseudoin-

dependent variables. In fact. Eq (4-56) is precisely the same differential equation

that results from the differential formulation in the first place. Actually F
is a

functional because it is a function of two functions, >• and \ x . Equation (4-56)

is

referred to as the Euler or Euler-Lagrange equation for the problem w hose

functional

is given by Eq. (4-44).

 THE EULER-LAGRANGE EQUATION H7

Before interpreting these results, let us show that the functional to be extremized
for the problem in Example 4-1 is, in fact, given by Eq. (4-14b).

Example 4-3
For the problem posed in Example 4-1, show that the functional to be extremized
is given by Eq. (4-14b). Recall that the problem in Example 4-1 was stated as
follows:

—
d-T
dx -
+ lOOOv 2 = 0 0 *£ .t =£ ] (4-7)

7X0) = 0 (4-8a)
7(1) = 0 (4-8 b)

Solution

First, let us check to see whether or not the condition expressed in Eq. (4-46) [or
Eq. (4-54)] is met. In terms of the notation in this problem, Eqs. (4-56) and
(4-46) may be written

£- 4(^.0 (4-57a)
ST dx\dTj
and
i

= 0 (4-57b)
o

Since the value of T is to be held fixed at either end of the interval, 0 =s x 1,

the variation of T or bT must be zero at these two points. Therefore, Eq. (4-57b)
is satisfied by virtue of the conditions T( 0) = 0 and T( 1) = 0, since 87” = 0 at
these points. This type of boundary condition is referred to as a geometric boundary
condition [10]. When SF/ST, = 0 on the boundary, we have a natural boundary
condition (see Example 4-4). Let us now turn to Eq. (4-57a) and make a term-by-

term comparison with the governing equation expressed in Eq. (4-7), which yields

dF_ 2
1000.x (4-58)
dT

and

_±(dF\ = _±(_dT\ = dizIA

dx \STJ dx\ dx) d\

A further comparison of the latter yields

118 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

The reader may \erify that Eq. (4-58) implies

F = lOOQr-T + f(Tx ) (4-60)

and Eq. (4-59) implies

F = - 'AT} + g{T) (4-61)

by direct substitutions. A further comparison of Eq. (4-60) with Eq. (4-61) yields

f(T t ) = - ViT~

and

g(T) = 100Qt-r

so that

F = 100Qr 2 7 - 'AT} (4-15)

Actually a constant term may be added to the right-hand side since the resulting
expression for F would still satisfy Eq. (4-56), however, this constant in no way
affects the extremum of the functional/ From Eqs (4-44) and (4-15), the functional
is given by

i4
- i4bi

This last result is identical to that presented in Eq (4-l4b) for this same problem
It is emphasized that the function T(r) that cxtremizes the functional / given in Eq
(4-14b) is the same one that satisfies the governing differential equation and bound-
ary conditions given by Eqs (4-7) and (4-8). respectively

In Example 4-4, Eq (4- 1 4b) is derived in an alternate but completely equivalent

manner

Example 4-4
By starting with Eq. (4-55) derive Eq (4-!4b) for the problem stated in Example
4-1

Solution

We now recognize the bracketed term in Eq (4-55). When it is equated to zero,

Eq
it becomes the differential equation for the problem. We m3y write from

(4-55)

(4-62)
SI = /O'[iooa<-- + £(f)]
 J

THE EULER-LAGRANGE EQUATION’

119

This may be broken into two separate integrals to give

= f lOOOx 2 8 7 dx + =
8/
Jo
f
Jo ' T (
dx \ dx j
87 dx 0 (4-63)

Integrating the second integral by parts with

u = 8T and

yields

r'dTd(bT)
-i

87 1000.V
2
87 r/.v + -— 87 dx (4-64)
-O dx Jo dx dx

Working with the second integral and using Eq. (4-38), we have
{'dTd(hT) ,
['dT/dr\ f ( cirV
(4-65)
Jo dx dx Jo dx \dx / Jo \dx
If in Eq. (4-64) we require that

dT,
-87 = 0 (4-66)
dx

we are left with

dT
8/ 1000.v
2
87 - IS dx (4-67)
dx
2
Since the variation of .v is zero (recall that only 7 and T x
have a variation), Eq.
(4-67) may be written with the help of Eq. (4-39) as

8/ = 8 r I000.r
2
7 - dx (4-68)
Jo 2\dx
from which we conclude that the functional I itseif is given by

1000.T
,
2
7 - -
( dT 1
— dx (4-69)
2 \dx

which again is identical to Eq. (4- 14b).

Let us examine the condition expressed by Eq. (4-66) a little more closely.
For the geometric boundary conditions in Example 4-1 we have 87 = 0 at .v = 0 ,

and x = 1. Another waywhich Eq. (4-66) could be satisfied is when dTtdx is

in

zero on the boundaries, i.e., at .v = 0 and x = 1 in this problem. This type of

boundary condition is the so-called natural boundary condition. If 7 represents

temperature, then dTtdx is zero when insulation is present. In a similar fashion, in

Eq. (4-57b) the condition that 3F/dTv = 0 is also referred to as a natural boundary

condition.
120 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

If in Eq. (4-64) we do not require the integrated term to be zero, the reader
may show that Eq. (4-68) becomes

#L +
i:[— Kf)H M 70 -
'

which introduces a mechanismi by which nonzero

n< values of dTldx on the boundaries
may be accommodated. In Eq1 (4-70) it is implied that
(tr

(£-)['= -r-S7
dx .

or that the variation is made on T only and n 0 on dTldx t (in the integrated term
only).
In this section, we have learned how to obtain the variational formulation from
the differential in two different but equivalent ways. The class of problems has
been restricted to a single second-order differential equation in a single independent
variable x, i.e., one-dimensional problems The Euler-Lagrange equation was in-

troduced for the class of problems whose functional F is given by Eq. (444),
providing we also satisfy

= 0 (4-72)

In Problem 4-23, the reader is asked to show that for the functional

(4-73}

the corresponding Euler-Lagrange equation is

df _ d/ar\ d 2 f dF\
+ (4-74)
dy dx\dyj dx 2

where y„ denotes the second derivative of y(r) with respect to x The following
conditions are assumed to exist in this case:

(4-75a)

and

(4-75b)

Equations (4-74) and (4-75) are readily derived from Eq. (4-73)
by following the
indicated earlier in this section and by integrating
by
first of the two procedures
of natural boundary conditions must be
parts twice. From Eq (4-75a), one set
given by
 F
THE METHOD OF WEIGHTED RESIDUALS 121

*
~9F _ rf/aF\l
(4-76)
_dy x dx \dy„) J o

while from Eq. (4-75b), we must have

a
(4-77)
dy xx

as another set of natural boundary conditions. The geometric boundary conditions

are given by

(4-78)

and

(4-79)

Equation (4-78) is equivalent to the statements that y(a) is prescribed and y(b) is

prescribed, i.e., the function y(.v) is prescribed on the boundaries. Equation (4-79)
implies that dy/dx is prescribed at x = a and x = b as well. What order differential
equation does Eq. (4-74) correspond to? What is the highest-order derivative present
in the functional? Try to generalize this observation.
In Problems 4-26 and 4-27, the reader will begin to appreciate why classical
variational formulations exist for only those problems whose differential formula-

tions do not contain odd-ordered derivatives.

4-6 THE METHOD OF WEIGHTED RESIDUALS

The method of weighted residuals provides an analyst with a very powerful ap-
proximate solution procedure that is applicable to a wide variety of problems. It is

the existence of the various weighted-residual methods that makes it unnecessary

to search for variational formulations to nonstructural problems in order to apply
the finite element method to these problems. Before presenting four of the most
popular weighted-residual methods, let us present a general framework around which
the various methods are developed. In the process of doing this, it is necessary to
repeat some of the general concepts given in Sec. 4-2.

General Concepts

Let us restrict the present discussion to a single governing equation with only one
independent variable. Let us represent the governing equation as

f{T(x)\ = 0 in fl (4-80)
 .

\ AR1ATI0N XL AND WEIGHTED RESIDl AL FOR ML LATIOSS

where T represents the function sought (sa\ temperature), which is a function of

only x, and w here fl is the domain of the region go\ emed by Eq. (4-SO). In addition,
letus specify the boundary conditions in the form

gilTU)] = 0 on T,
{4 ^ 1}
S;ITCr)] = 0 on I\

etc.,where r ( and T; include onl\ those parts of fi that are on the boundary. Let
us again approximate the solution to Eqs. (4-SO) and (4-81) with an approximate
function V
that is given b\

r = 7*(r; o„ a2 , . .</„) = S a,St ix) (4-82)

j=I

which has one or more unknown (but constant) parameters a (

. a} . . . . ,a„ and
that satisfies the boundar) conditions given b} Eq. (4-SJ) exact)}. As before, the
functions A’( t) are referred to as the trial functions If this approximate solution T
is substituted into Eq. (4-80) for T(x), it should not be surprising that it will not
necessarily satisfv this equation exactlv, that is. some residual error R{x, o,, ...
o„) results Therefore, we write

a,. a2 . . . o„)| = /?(r. <i|, o: . . a„) (4-83)

The method of weighted residuals requires that the parameters a,. <i
:. .

be determined b> satisfv mg

/n»’,(0 /?(t: U|. .u B )dt = 0 i= 1.2. .n (4-84)

where the functions w,U) are the n arbitrary » et gluing functions. The choice of the
weighting functions is left largel) up to personal preference, but four particular
functions are used most often The most popular weighted-residual methods are
referred to as (1) point collocation, (2) subdomain collocation. (3) least squares,
and (4) Galerkin. Each of these is developed in turn, and the problem stated in
Example 4-1 is sohed in Examples 4-5 to 4-8 w ith each of the methods, respective!}
Of these four methods, the Galerkin method has the widest use in finite element
analysis for reasons that will become apparent later. The least-squares method is
also used but to a much lesser extent. Some of these methods are used in Chapter
10 when FEM formulations to unsteady problems arc considered.

Point Collocation

In the point collocation method, the weighting functions ii ,(x) are denoted as 8(r - »,)

and defined such that

8{r — x,) dx = I for t = x,

j
(4-851
and
8{t — x.) t£r = 0 for x ^ x,
j
 THE METHOD OE WEIGHTED RESIDUALS 123

The .y,’s are referred to as the collocation points and are selected arbitrarily by the
analyst. The 8 here has nothing to do with the variational operator but rather is

defined by Eq. (4-85). Substitution of 5 (.y - a,) for in Eq. (4-84) gives

/a8(.v - a-,) R( x; a,, a 2 , . . .

, a„) dx = 0 for / = I, 2 n (4-86)

If Eq. (4-86) is evaluated at n collocation points, .v,, .y 2 , .... ,y„, then n algebraic
equations in n unknowns result:

R{ -Vj ; a t , a 2 , .... a n ) = 0

R{x2 \ «i, ci
2, .... a„) — 0
(4-87)

f?(.Y„; a,, «2 , . . . , a„ ) = 0

It is emphasized that the n unknowns in Eq. (4-87) are the constant parameters rq,
a2 , , a„. Once these are determined, the approximate solution is given by Eq.
(4-82) and the problem is solved, at least approximately. The constants o ( , «2 ,
. . .
, a„ are dependent on the choices made for the n trial functions, N ,(.y), N 2 (.y).

.... N„(.x), and on the choice made for the n collocation points. For well-posed
problems and well-behaved trial functions, the solution of the system of equations
implied in Eq. (4-87) will yield the numerical values for the a,' s. Equation (4-87)
may be written more concisely as

R(x,\ a t , a 2 ...
, , a„) =0 for / = 1, 2, .... it (4-88)

where ,y, is the /th collocation point. This method is illustrated numerically in
Example 4-5.

Example 4-5

Using the trial function assumed in Example 4-1 determine an approximate solution .

to the problem posed in that example by the point collocation method.

Solution

Note that since only one trial function is used, only one parameter (namely, a,)
may be introduced. This implies that only one collocation point may be selected.
It is as good a choice choose the (one and only) collocation point .y,
as any to to

be in the middle of the domain, or A = Vi. Recall that the residual R(x:a ) t x
is

given by Eq. (4-11) or

R(x:a t )
= — 6o,.y + 1000.Y" (4-1 1

By Eq. (4-S8), we must evaluate this residual at the collocation point or at a, =

'/j and set the result to zero, or

— 6rt,( '/:) + 1000('/;) : = 0

 (

124 VARIATIONAL AND V. LIGHTED RESIDUAL FORMULATIONS

Solving fora yields |

a, = ,oon
/i2 (4-89)

Therefore, an approximate solution to the problem by the point collocation method

is given by

Tp, = '«"/i2Jt(l - 2
X ) (4-90)

The reader should compare this approximate solution with those from Examples
4-1 and 4-2.

Subdomain Collocation

In subdomain collocation, the weighting functions are chosen to be unity oxer a

portion of the domain and zero elsewhere. Mathematically, the h*,(x) are given by

«•,» =

"'iM - (o

. 11 for x in Q„
H" X _

x
'

[0 for not in fl„

where fl, is a portion of the domain not included in ft 2 , fl 3 , . .ft*. It is instructive

to show these particular weighting functions in graphical form, for example, see
Fig. 4-6 for n = 3.
Let us return now to our general discussion with n unknown parameters. Sub-
-
stitution of iv, jc) given by Eq. (4-9!) into Eq. (4-84) gives n integral equations

Jnj /?(x; fl|, a2 , . . . , a„) dx - 0

o„ a2 a,) dx = 0
( 4 . 92 |

fn R(x; a,, a2 , ... a„) dx =0

The different integration domains should be noted. Since the residual error R
is a

in Eq
known function of x, a,, a2 , ... a„ [see Eq. (4-83)]. the integrations
in n unknowns. Again
it
(4-92) can be carried out, resulting again in n equations
these
are a u a 2 c*. The values obtained [or
is emphasized that the unknow ns
 THE METHOD OF WEIGHTED RESIDUALS J 25

Figure 4-6 Possible subdomains and weighting functions for the subdomain collocation
weighted-residual method (for three subdomains).

constants are, in general, dependent on the n subdomains chosen as well as on the

choice of the n trial functions.

Example 4-6
Repeat Example 4-1 fay using the subdomain collocation method with the same trial

function.

Solution

Since there is only one trial function and one unknown parameter ct\, only one

integral of the residual (over the entire domain from ,v = 0 to x = 1 ) needs to be

evaluated:

R(x;a ) ci\ = 0 (4-93)

J t

Substituting R{.\,ci{) given by Eq. (4-1 1) into Eq. (4-93) gives

'( — 6a,.t + 1000a 2 ) dx = 0

|o
Integrating, evaluating, and solving for a, yields

a i
— ,00
% (4-94)
126 VARIATIONAL AND W E1G1ITED RESIDUAL TORMUUVTIONS

Therefore, the approximate solution to the problem by the subdomain collocation

method is given by

K= ,ot
”/941 - x
2
) (4-95)

The reader will undoubtedly note that the same value for o, was obtained from the
Ritz method in Example 4-1 This is not just a coincidence: the first-order Ritz
method is identical to the first-order subdomain collocation weighted-residual
method

Least Squares

The method of least squares requires that the integral I of the square of the residual
R be minimized. The integral l is given by
/ = /ll(K(*»- 0i. o2 , •
, o n )) 2 iLx (4-96)

The problem isreduced to the determination of the parameters a,, a 2 o„

such that the value of the integral / is minimized Therefore, we must take the
partial derivative of /, first, with respect to a and set the result to zero, then, with
{

respect to a2 and set that result to zero, and so forth Mathematically, this may be
written as

JnlW*.- , o n )] 3 dx = 0

~ ^ fl|, a2 , , o„)] 2 tlx = 0

«|. a2 , , a„))
2
dx = 0
da„ da n

where fl includes the entire .r domain. Because the limits on the integral are not a
function of the a,' s. the order of the integration and differentiation may be inter-
changed to give

r ^ .

J T~
a da a f a> . . n„)] 2 dx 0
i

r d
- , aj) : dx = 0

and so forth. Carrying out the differentiations and dividing by the constant factor

2 (and dropping the arguments on R for now) gives

128 VARIATIONAL AND W LIGHTED RCSIDUAL LORMULATIONS

dx = 0 (4--I00I
L
J0 0(1 1

Computing dRlda gives x

dR
~ — - — A6t
Sa x

and Eq. (4-100) becomes

(-6x)(-6a,x + 1000r : ) dx = 0 14-101)

(a
Simplifying the integrand, integrating, evaluating, and solving for <1 , yields

o, = ' at
y»

from which we may write the approximate solution from the least-squares method
as

Ti = loo
%.t(l - x2 ) {4-102}

The reader should note that still another value for is obtained.

Galerkin

In the Galerkin weighted-residual method, the trial functions N,(x) themselves arc
used as the weighting functions, or

»f,(r) = N,(x) {4-103}

For the Galerkin method Eq (4-84) becomes

a2 , . ,a„)dx = 0 fori = 1 son (4-104)

Because there is one trial function for each unknown parameter, Eq. (4-104) really
gives n such equations that, when solved, yield the values of the unknown param-
eters, fl|, «2 , . . . , a„. It should be obvious that again the values obtained for the
n.’s are dependent on the choice of trial functions. The Galerkin method is seen to

be quite simple.

Example 4-8

Resolve the problem in Example 4-1 by ustng the Galerkin method and the same
trial function.

Solution

The trial function N (x) ist

given by Eq (4-9), or

AM*) = -t(l - t
2
)
(4-9)
 THE METHOD OF WEIGHTED RESIDUALS 129

Substituting R from Eq. (4-1 1) and A ,(.t) from

r
Eq. (4-9) into Eq. (4-104) for i = 1

only gives

tfl ~ x 2 )( — 6a,x + lOOOx 2 ) dx = 0 (4-105)

fQ

Simplifying the integrand, integrating, evaluating, and solving for a, yields

a i
= 50C
%s (4-106)

Therefore, an approximate solution by the Galerkin method is given by

T'c = 500
% s t(1
. - x2 ) (4-107)

The reader should note that exactly the same value for «, was obtained in Example
4-2. There an approximate solution was obtained by the Rayleigh-Ritz method,
which is a variational method. This is not a coincidence for it can be shown that
the variational and Galerkin methods must give identical results, providing the
problem has a classical variational statement (or principle) in the first place [11].
This profound result carries over into the application of these methods to FEM, as
shown in Secs. 4-8 and 4-9 for the same problem. E

Comparison with the Exact Solution

If the problem posed in Example 4-1 is solved exactly, we get

T{x) = I000
/nx(l - x3 ) (4-108)

Note the exponent of 3 on the x in the parentheses; our approximate solution was
assumed to be of the form

T'(x) = a,x(I - X2 ) (4-109)

which is slightly different than the exact solution above. The exact solution from

Eq. (4-108) compared to the four approximate solutions by the various weighted-
is

residual methods in Fig. 4-7. The approximate solution from the Ritz method
(Example 4-1) is identical to that obtained from subdomain collocation. Similarly,
the Rayleigh-Ritz (variational) and Galerkin solutions are identical.
It is seen that none of the methods seems to do a particularly outstanding job

of approximating the exact solution. However, in Secs. 4-8 and 4-9, the variational
and Galerkin methods are adapted to a Finite element formulation for a solution to
the same problem. The fundamental difference between what we have done so far
and what we will be doing next lies in the type of trial functions that are assumed.

Recall that trial functions were chosen that applied globally to the entire domain
beins analyzed. From this point onward, trial functions will be chosen that are to
be applied locally (over each and every finite element). These trial functions will
130 VARIATIONAI AND V. LIGHTED RESIDUAL hORML'LATIONS

Figure 4-7 Comparison of approximate solutions to problem posed in Example 4-1 with

the exact solution

be said to be piecewise continuous Piecewise continuous trial functions in this text

will be referred to as shape functions
This important step is fundamental to the finite element method and has as-
sociated with it two important implications: (1) the piecewise continuous trial func-
tions or shape functions need not satisfy the physics of the problem since they are
to be applied locally in every element, not globally, and (2) the global boundary'
conditions need not be satisfied. This is not to say that the shape functions have
no restrictions placed on them. These restrictions are delineated in Chapter 6. Before
actually illustrating this procedure, we need to review some additional mathematics
with which the reader may not be familiar

4-7 MORE MATHEMATICS

In this section we will review the following- (1) the denvame of a matrix with
respect to a scalar, (2) the integral of a matrix with respect to a scalar. (3) the
derivative of scalar with respect to a sector, and (4) the integral as a sum of other
integrals. Examples of the use of these notions are provided throughout the text m
general and in the next two sections in particular
 MORI. MAUI! MA'rirs 131

Derivative of a Matrix with Respect to a Scalar

Consider an in x n matrix M whose elements are functions of .v. The derivative

of M with respect to .v is obtained by taking the derivative of each of the entries
(or elements) in M with respect to x. or
dMu MI 12 Min,
<Lx (lx dx

rfM(.v) (IM 2 ,
dM22 dhUn
dx (lx dx dx (4-110)

dMn„ dM,n„
dx dx (lx

The instances when M is a column matrix (// = 1) or when M is a row matrix

( in
- 1) are seen to be special cases.

Integral of a Matrix with Respect to a Scalar

Let us again consider an inx matrix M whose elements arc functions of a.

ii It

follows that the integral of M with respect to may be found by taking the integral
.v

of each element in M. or

Mi dx /A/); (IX .
M» dx

f M 2l dx f 22 (l\ . .
M„ dx

J M(x) dx = 14-111)

M»I dx f M m2 dx fM,„„ dx

Although Eq. (4-1 1 1) is written with indefinite integrals, the same idea holds for

definite integrals (providing the same limits arc used on all integrals). Like differ-
entiation, integration of a column matrix and a row matrix are seen to be special
cases of Eq. (4-1 1 1 ).

Derivative of a Scalar Function with Respect to a Vector

Let us now consider the following scalar function:

i(\) = «,A’,(-') r/-,V : (A) - • • • - v l (4-112)

and determine < ly/ila,. or

132 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

JV,(r) (4-113a)

N 2 (x) (4-1 13b)

— = NM
da„
{4-1 13c)

We may write

~=
8y
da,
N,(x) for i = 1.2 n (4-114)

If we define N and a as

N = MU) N2 U) •
'
N,(x) 1 (4-115)

and

a = [n, a2 (4-116)

then Eq. (4-1 12) may be concisely written as

\(r) = Na (4-117)

where it is implied that N = N(r) Note the use of the transpose in Eq. (4-116)
that is used simply to conserve space since a is really a column matrix. From Eqs.
{4-! 13) it follows that

dy_

dci]
N,(x)

iL N2 (x) = Nr
da (4*118)

N„(x)

Therefore, it may be concluded for N and a defined by Eqs (4-115) and (4-1 16),
respectively, that

t»(Na)
 THE RAYLEIGH-R1TZ FINITE ELEMENT METHOD

This result may not be what we would intuitively expect because of the need to
transpose N. It is particularly useful in the next section where the Rayleigh-Ritz
method is adapted to FEM. This result is not needed in the Galerkin method.

The Integral as a Sum of Other Integrals

Itfollows directly from the definition of integration that an integral may be evaluated
by dividing the domain f1 into M
nonoveriapping subdomains IT and summing the
integrals over each subdomain, or

M
fafW dx = 2 fn rf(x) dx (4-120)
e=\

Each nonoverlapping subdomain £T will correspond to one finite element in this

e
text. The superscript ( ) will be interpreted to be the element number.

4-8 THE RAYLEIGH-RITZ FINITE ELEMENT METHOD

In Sec. 4-2 the Rayleigh-Ritz method was used to find an approximate solution to
the problem posed in Example 4-1. In this section, we wish to obtain another
approximate solution but now in the context of the finite element method. As in
Example 4-2, the variational statement will be the starting point. However, the

fundamental difference in the treatment here compared to that in Sec. 4-2 is the
following: piecewise continuous trial functions will be used that apply only to a
small portion of the entire domain, i.e., over a finite element. The reader should
recall that the trial functions used in Secs. 4-2 to 4-6 were applied globally, i.e.,

to the entire domain. Now the trial functions will be taken to be interpolation
polynomials as described below.
Consider the domain that includes the interval a *£ x b as shown in Fig.
4-8. Let us divide this interval into M
subintervals as shown. Actually, this process
is called discretization and a typical subinterval from .r, to xk is really a finite element
and simply denoted as element e. The ends of the subinterval are the locations
is

of the two nodes, nodes j and k. Since the element e has two nodes, a linear
interpolating polynomial may be used to describe the behavior of the field variable,
e.g., y(x) ,
over the element, or

y
e = nix + b (4-121)

The superscript) on the

(
efield variable serves to remind us that the approximation

applies only over element e and not globally. But at = -V,, we must have y r = y} ,

e =
and at x = xk , we must have y yk Therefore, applying these two conditions
.

to Eq. (4-121) yields

v. = niXj + b (4-122a)

\\ = mxk + b (4-1 22b)

Figure 4-8 Typical one-dimensional problem domain discretized into M elements w
nodes with an assumed linear variation of the field variable \(r) over a typical elem
w ith nodes j and t

that may be solved for mi and b to give

6 = -
'<

Substituting these expressions for m and b in Eq (4-121) and some rearrangen

gives the following

Ut - V \T< - v
It is observed that Eq (4-124) is of the form

/' y = + a 2 N 2 [x)

y
r = ajNj(x) + ak Nk ( t)

NM t
= AT,(v) = • (4-t2
 THE RAYLE1GH-R1TZ FINITE ELEMENT METHOD ]35

Figure 4-9 Typical element e showing the two shape functions and the resulting variation
of the field variable y e (x) (linear between nodes j and k).

and

N2(x) = N k {x)
= — (4-127b)
xk ~ *j

The functions Nj(x) and N k (.r) are referred to as the shape functions for the element.
We see that instead of using higher-order polynomials that apply globally to
the entire domain, we will now use two first-order shape functions that apply only
to a small subinterval or element. These particular shape functions are shown in

Fig. 4-9. Note that they are both linear and at.v = xJt we have/V, = 1 andA^ = 0,
and x = xk we have
at ,
N }
= 0 and N = k 1 . Equation (4- 125b) may be written in
matrix form as follows:

y
e = Na c (4-128)

where by definition

N = [Nj{x) N l (x)\
(4-129a)

and

(4-1 29b)

It should be emphasized that the vector a‘ contains the values of the field variable
at thetwo nodes. The idea is to use Eq. (4-128) to represent y(.v) over each element
in the discretized domain and to find the values of the a/s by extremizing the

functional / with respect to the a/s. The set of all a/s for the entire domain represents
the finite element solution to the original problem.
J36 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

Because the function y(.t) itself given by Eq. (4-121) is continuous and the
first derivatives are not necessarily continuous (from element to element), the
function j(.t) is said to be C°-continuous. A function with continuous first derivatives
is said to be C'-continuous. etc. In a similar fashion, the two shape functions given
by Eq. (4-127) are also said to be C°-continuous or to possess C°-continuity. Some
C'-continuous shape functions are derived in Chapter 7 for the beam model.
These rather abstract ideas are best illustrated by tackling a numerical example,
which again, for comparison purposes, is taken to be the problem posed in Example
4-1.

Example 4-9
Using the Raylcigh-Ritz finite element method, find an approximate solution to the
problem posed in Example 4-1 by determining the value of the field variable, 7\

at six equally spaced node points.

Solution

Before the Rayleigh-Ritz method can be used, the variational statement of the
problem is needed. Because of the two prescribed temperature boundary conditions
at x 0 and x = l (i.e on the global boundary), we could start with the functional
*= .

given by Eq (4-69) from Example 4-4 However, let us use the functional implied
in Eq (4-70) so as to show what happens to the integrated term Therefore, the
functional to be used here is

{— r-
Wr
"

„
> L
"
+ f [iooat :7-' - U—
A*/ VI )
J 1
(4-130)

Note that Eq (4-120) has been used to represent / as a sum of the integrals over
each element The integrated term above will be appreciated later to be completely
equivalent to that implied in Eq (4-70) because the internal contributions will
cancel during the assemblage step and onl> the end conditions will survive The
summation sign in Eq. (4-130) unnecessarily clutters the equations, therefore, let
us drop the summation and write the functional for a typical clement c as

/ = S /< (4-1321
r= I

and where Af is the number of elements used. This summation is really considered
f
at the assemblage step later The o,*s in the approximation y [to y(t)J in element

e arc to be determined by differentiating / with respect to a and setting the result

to zero, or
 —
da
=
e
2
~ !
—
da e
(4-133)

For now, let us extremize I e in Eq. (4-131) with respect to the vector a'' where for
T e we take

Te = NV (4-134)

as implied by Eq. (4-128), or

dl e
_ dT dT
e
2
dr 1 d dT e
+ 1000.V dx = 0 (4-135)
da e dx da e
L da e
2 da e
\ dx

But from Eq. (4-119) we have

dT e d(Na e )
= Nr (4-136)
da e
da e

and it also follows that

l_d_(dr -i d(dT e \dTe d N r dN

e c c
ae
2 da \ dx 2 dx da \ dx da \ dx / dx dx dx

(4-137)

Using the results from Eqs. (4-136) and (4-137) in Eq. (4-135) yields

~N r + lOOO.v
2
N r dx
-lv
« dN r d N
a e dv = 0 (4-138)
dx dx d\

But the vector a e is not a function of x so it can be removed from the right (why
not from the left?) to give

dNr
f«
J*j dx
dN
dx
dx a c = —
dT
dx
N
_
r
+
Jx,
I’'
1000.v 2 N r dr (4- 139)

which is of the form

K ea = e
t
e
(4-140)

c
where may be referred to as the element stiffness matrix, f as the element nodal
force vector, and a' as the vector of nodal unknowns (i.e.. T; and Tk ). Before Eq.
(4-140) can be solved, however, two more important tasks must be completed: (1)
assemblage of the K^’s and P’s to form K" and f" and (2) application of the
geometric boundary' conditions (prescribed temperatures) at .r = 0 and .r = The 1 .

matrix K" is referred to as the assemblage stiffness matrix and the vector f“ is
referred to as the assemblage nodal force vector (before considering the geometric
boundary conditions). After the geometric boundary conditions arc applied, we have
a system of equations represented by

Ka = f
 140 VARIATIONAL AND WEIGHTED RESIDUAL rORMULATJONS

Obviously, we also have

K® = K<» = K<*> = K® = (4-1 46a)

£
For F 1 ’, we have

- - [;]^ * [-;]=¥
IMP [<0.2)* - 4(0.2)(0 0) 1 + 3(0 0)*]
+
(I2)(0 2) (_3(0 2)* - 4(0.0)(0.2) J + (0
0)*J

Similarly, we And

The assemblage of the element stiffness matrices to form K“ follows the same
line of reasoning presented m Chapter 3 where the truss was studied However,
the situation here is even simpler because the submatnces in the Kr 's are of size
1 x I , or simply scalars. The reader is reminded that the primary factor in deciding
where the element stiffness matrix K' for element e is added into the assemblage
stiffness matrix K Q
is the two global node numbers (/ and J ) associated with the
element. This assemblage step was summarized in Chapter 3 for elements containing
only two nodes (such as the present problem) m Eq (3-31). The details of the
assemblage of the stiffness matrices are omitted below, but the reader should show
the missing steps if there is still some doubt as to how this important step is

accomplished.
a
The assemblage of the nodal force vectors to form done in a completely f is

analogous manner, except that it is even simpler because only a column matrix is
involved. If the element nodal force vector is imagined to be partitioned as

(4-147a)
 THE RAYLE1GH-R1TZ FINITE ELEMENT METHOD 141

then the assemblage of these to form fa may be summarized by the following

equation:

M
f* = 2 f,f (4-147b)
e— 1

where M is the maximum number of elements used in the model and f,f is taken
to be zero if element e does not contain node n. With this algorithm, the reader
should be able to show that the assemblage nodal force vector in the following
matrix equation results:

Note how the unknown derivatives ( clT/dx ) for the interior nodes have canceled.
Physically this may be interpreted as follows: although there is a temperature
gradient and hence a heat flux at each node and hence between two neighboring
elements, these internal heat fluxes cancel. The only heat fluxes that survive are
those on the global boundary, and then only if there is actually an imposed heat
flux there. If the temperature gradients (or heat fluxes) at either or both ends were
specified, the value would simply be entered as indicated above. That case would
correspond to a prescribed heat flux into or from the ends. In any event, in the

problem at hand we have prescribed temperatures at both ends, and by using Method
1 (Method 2 may also be used), Eq. (4-148) is modified to impose T =
x
0 and
T6 = 0, which results in

10 -5
0 0 0
0 10 0 0
0 -5 10 -5 0
0 0 -5 10 -5
0 0 0 -5 10
0 0 0 0 0

The disappearance of the (unknown) temperature gradients at nodes 1 and 6

should be noted and thoroughly understood: in thermal problems, in general, the

boundary nodes must have either a prescribed temperature or a prescribed gradient

(heat flux), convective, and/or radiative boundary conditions. It should be noted
further that the case of insulation is a special case of a specified gradient since dT/
dx = 0. This zero-gradient case is the so-called natural boundary condition. In any
event, solving Eq. (4-149) for the nodal temperatures yields
142 VAKlAllOMAt. AND Wt IGIIU 1) HI SUUIAL l ORMULAHONS

7 =
,
0

7: « 16.5

r, = 31.2

Ta = 39.2

7, = 32.5

r6 = o

The reader will note that these values lie right on the curve for the exact solution
in Fig. 4-7! The significant point here is that two linear shape functions (or trial
functions on a local or piecewise basis) have given results that are extremely close
to the exact sotution, whereas the use of cubic-order, global trial functions in the
Rayfcigh-Ritr method resulted in an error as large as 2lVc (at x = 0 2) Except

for the Galcrhin method, the other global approximate solution methods resulted
in even larger errors.

So far it has been seen dial the variational (Raylcigh-Ritz) and Galcrkin weighted-
residual methods perform the best out of all the approximate solution techniques
considered here The variational FEM solution presented in this section resulted in
a solution very close to the exact one (with only five elements) The next section
will illustrate how the Galcrkin method is adapted to FEM

4-9 THE GALERKIN FINITE ELEMENT METHOD

In See 4-6. four different weighted-residual methods were introduced In addition,

four approximate solutions to the same problem were obtained by each of these
methods Global trial functions were used, these functions applied to the entire
domain In this section we wish to repeat the solution only for the Galcrkin method,
but this time in the context of the finite element method. The implication is that
piecewise continuous trial tunclions or shape functions will be employed for each
element In fact, as in the Raylcigh-Ritz FEM (sec Sec. 4-8). these shape functions
will actually be interpolation polynomials.
As in See. 4-8, the problem domain is divided or discretized into a number of
elements Figure 4-8 is again applicable to this formulation, and the interval a <
x ’C b <i e., the problem domain) is divided into M such elements. A typical element
e has two nodes, j and A, with coordinates r; and r4 respectively The field variable
,

y at nodes j and A has values and >», respectively. As before, since each element
has two nodes and since a unique straight line may be drawn between two points,
the interpolation polynomial to be used to describe the field variable, c g , ytr).

must be of the form

v' = mx + b (4-1 50a)

 THE GALERK1N UNITE ELEMENT METHOD

which is identical to Eq. (4-121). Therefore, the results of the first part of Sec.
4-8 are directly applicable here. In other words, we wish to express
y%\) in the
form

y‘(A) = yjNj(x) + yk Nk (.v) (4-150b)

From Eqs. (4-125) to (4-127). we see that at x = x} we have y e(Xj) = yJt and at
x = x, we have y %\ k
) yk = . As in the Rayleigh-Ritz FEM formulation in Sec.
4-8, two piecewise continuous linear trial functions are to be used. The functions
Nj(x) and Nk {x) are refemed to as shape functions.

Example 4-10
Resolve the problem posed in Example 4-1 (approximately) by the Galerkin finite
element method by using six equally spaced nodes.

Solution

Recall that the Galerkin method requires that the trial functions themselves be used
as the weighting functions. The integral of the weighted residual R in this case is

written as

JnN,(x)R(x, fl
2 , . . . , a„) dx = 0 (4-151)

But instead of using trial functions N,(x) that are applied globally to the entire x
domain and that satisfy the boundary conditions automatically (and exactly), let us
try to represent the solution Te over a small interval (really a Finite element), say
from Xj to x k as
,

Te = NjixWj + N {x)T
k k (4-152)

where T} and Tk are analogous to the unknown parameters, and a 2 , in the general,

weighted-residual method. Moreover, if N/x) and Nk (x), in addition to being linear,

are such that when a = a,we have Nj = and Nk — 0, and when a = xk we have
1

Nj = 0 and N = k 1 , then in fact we have appropriate shape functions. The reader

should verify that for N/x) and N (x)
k given in Eq. (4-127) this is in fact the case.

In summary, in one-dimensional C°-continuous problems with two nodes per ele-

ment, the shape functions must be linear because there are two nodal points per
element and they must satisfy the following conditions:

JV/a,) = Nfixk ) = 1 (4-1 53a)

and

N/xk ) = N k ( a,)
= 0 (4-1 53b)

In the Galerkin method, the two shape functions jV, and Nk are the weighting

functions. From Eq. (4-104), we may write the following weighted-residual equa-
tions:
144 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

«,(<)[«'(,; Tj.T,)] dx = 0 (4-154a)

2^ J_"

and

2 Ndx)\R r{x\Tj,Tt )\ dx = 0 (4-154b|

]'

where e denotes the clement and A/ is the number of elements (or intervals into
which the domain is divided). A typical finite element goes from x (where the ,

temperature is 7}) to .r* (where the temperature is Tk ). If matrix notation is used

and N r is defined as

then Eqs (4-154) may be written in one equation as

" f«
2 (£t = 0 (4-155)

But the residual R' for a typical element e in the problem at hand is, by definition,

R' = + (OOOt 2 (4-156)

dx 7

where the approximate solution Tr , applicable only to element e, has been substituted
for the exact temperature T in the governing equation and the result set equal to

the residual Rr for the element Therefore. Eq (4-155) becomes

2 N T (~- + 100Qr ; dx = 0 (4-1571

J“ j

Before breaking this integral info two separate integrals. Jet us agree to drop the
summation sign. The reason for this is simple it represents the assemblage step
that it is hoped is becoming routine In other words, it is preferred to work with a
typical clement (clement e) and at some later stage we will put the pieces back
together. Breaking Eq (4-157) into two separate integrals gives

Nr d< + (“ N r ( |0(Xlr ! ) tlx = 0 (4-1581

Jt, dx‘

If the first integral is integrated by parts we get

Nr —f
lir
-
dx dx
+ pN r (100Qr-)<b = o (4-159)
 THE GALERKIN FINITE ELEMENT METHOD 145

But we have already agreed to write T e in terms of two (linear) shape functions,
Nj and Nk and the two nodal temperatures, T and 7), as
,
}

T e = Na f (4-160)

where

N = [Nj N k (4-161)

and

(4-162)

Using Eq. (4-160) in the integral terms only in Eq. (4-159) gives

dr fn dN
— T
dN f«
Nr —a dx+e
N r (1000x 2 ) dx — 0 (4-163)
dx jx, dx dx jxi

The reason for not using Eq. (4-160) in the integrated term is that by not doing so,
we have allowed a mechanism for imposing gradient boundary conditions (see
Problem 4-56). But a c is not a function of x and may be pulled out of the integral
from the right to give a rather simple equation of the form

K e ne = t
e
(4-164)

where

dN T dN
K‘ dx (4-165)
-L x, dx dx

and
dT
f
e = Nr
dx
+ pN r (1000x
Jxi
2
)dx (4-166)

For all practical purposes, Eqs. (4-165) and (4-166) from the Galerkin method are
identical to Eqs. (4-141) and (4-142), respectively, from the variational method!
As mentioned before, this is not surprising because when a classical variational

principle exists, it must give the same FEM formulation as the Galerkin method.
Again the element stiffness matrix is seen to be symmetric.
Equation (4- 165) for the element stiffness matrix and Eq .
(4- 1 66) for the element
nodal force vector are the so-called element characteristics. Note that it is not

necessary to distinguish between the local and global element stiffness matrices
was not used in the development of these equations.
since a local coordinate system
Moreover, the function T(x) is a scalar function (such as temperature) and hence
no transformation from a local to a global coordinate system would be necessary'
before the assemblage step. It should be obvious that the remaining part of the
146 variational and wriGim d rlsidual formulations

solution by the Galerkin finite element method is identical to that in Example

4-9.

4-10 APPLICATION: ONE-DIMENSIONAL HEAT TRANSFER

IN A PIN FIN

Consider a circular pm fin of length L, and varying cross-sectional area A and

perimeter P as shown in Fig 4-1 1 Fins such as these are used to increase the heat
transfer rate, particularly to gases Gases generally have low convective heat transfer
coefficients By increasing the surface area, such as in the pin fin, we can increase
the heat transfer rate significantly The heat transfer rate is also referred to as the
heal removal rate in this text Depending on the circumstances, fins arc also
commonly used in liquids
The temperature distribution in the fin is needed for the case when the base is

held at a temperature Tb . the tip is insulated, and the fin itself convccts to a fluid
at a temperature Tu
with a heat transfer coefficient h. The thermal conductivity is
denoted as k, and the temperature at any point along the fin is denoted as T. The
heat removal rate and fin efficiency (defined below-) arc to be determined also
The governing equation may be shown to be (see Problem 4-66)

£(
M dr) “ kP(T ~ Ta) = ° for 0 is x ^ L, (4-167)

subject to the two boundary conditions

T(0) = Tb (4*1 68a)

tfT
iLf ) = 0 (4-168W
dx

The intent is to obtain an approximate solution using the finite element method. In
particular the finite clement characteristics will be derived using the Galerkin method.
 APPLICATION ONE-DIMENSIONAL HEAT TRANSFER IN A PIN FIN

The Element Characteristics

The derivation begins by forming the integral of the weighted residual and setting
the result to zero, or

- hP(T - Ta ) dx = 0 (4-169)

where a typical element e is assumed to connect node j (at .r = x.) to node k (at

x - xk ) as shown in Fig. 4-12. The superscript (

e
) on the field variable, T in this
case, is no longer shown. Note that the weighting function is taken to be the transpose
of the shape function matrix, since the Galerkin method used. At nodes j and k,
is

the temperatures are 7j and Tk ,

respectively. Note also that each node represents a
planar surface and not just a point.
Let us integrate by parts the term with the second-order derivative as follows:

U r dJ
f
Jy
N T -(kA^]
dx \ dx J
dx = N TkA~
dx A,
f*£/N
dx
,
kA
x
—dx
dx
,
(4-170)

As illustrated in Example 4-9, the so-called integrated term cancels for all interior

nodes during the assemblage step. Moreover, at the base of the fin, the temperature
is prescribed. As shown in Example 4-9, the unknown temperature gradient at the
corresponding node is eventually eliminated when the prescribed temperature is
imposed on this node. Also, at x = Lf we have dT(Lf )/dx =
,
0. Therefore, this

integrated term need no longer be considered here. (See Sec. 8-3 for the case of a
noninsulated tip.) Therefore, Eq. (4-169) becomes

-
fw
Jy
—
//

dx
r dT
kA—dx -
dx Jxj
N ThPTdx+\Jx N ThPTa
fn

j
dx = 0 (4-171)

Figure 4-12 One-dimensional element with nodes j and k.

148 VARIATIONAL AND WEIGHTTD RESIDUAL FORMULATIONS

But using T = Na' and noting that only N is a function of .r, we get

K'a r = V (4-172)

where

K' = KJ + Kfv (4-173)

and

f'* r/N r r/N

kf = (
——kA—— d\ (4-1741
J »/ dx dx

K '* = yT,' pNdr (4-175)

f“

and

[- - J ‘n t /iPTu dr (4-176)

The vector of nodal unknowns a c is given by

T
a' = [7; Tt )
(4-177)

and the shape function matnx N by

N = [NjM Nk W 1
(4-178)

where N, and /V4 are, in turn, given by Eqs (4-127)

In the most general case, k, h. A P. and Ta may all be functions of t However,
,

let us perform the integrations m Eqs (4-174) to (4-176) approximately by taking

the following approach Let us treat these parameters as constants in any given
element by using die values at the midpoint of the element where r = x and by
denoting them as k, h. A, P and Tu It can be shown that (sec Problem 4-69)
,

-\
l[-i
1

'll
Ij
(4-179)

mh il (4-180)
6 L1 2j

WLT,\ ll (4-181)
2 L<J

where L is the element length, or L = xk - xr

for
The element stiffness matrices and nodal force vectors may be determined
these
every element with the help of Eqs (4-179) to (4-181) The assemblage of
2x2 matrices and 2 x 1 vectors is performed in the usual manner to
give a
 APPLICATION ONE-DIMENSIONAL III.AT 1 RANSH.R IN A PIN HN 149

system of /V algebraic equations in the N unknou n nodal temperatures. This assumes

the fin is discretized into M elements with N nodes, where M is equal to N — 1.
The result is K"a = f
a
. The geometric (or prescribed temperature) boundary con-
ditions are applied at this point to yield Ka = f. This matrix equation may be
solved for the nodal temperatures in the vector a. Therefore, it is assumed that the
nodal temperatures are known in what follows.

Heat Removal Rate

The heat removal rate can be determined by two different methods, once the nodal
temperatures are known. The first method is based on differentiation, and the second
method is based on integration. Not surprisingly, the second method is significantly
more accurate, as illustrated numerically in Example 4-1 below. 1

In the first method, the heat removal rate QR from the fin is evaluated by
computing the heat flux from conduction at the base of the fin (where .v = 0), and
by multiplying this value by the cross-sectional area A. This may be summarized
as

[
c

Qr " M dx = -kA
dx
'
.in (4-182)
t =0

where it is assumed that element 1 is at the base of the fin as shown in Fig. 4-13.

It follows (see Problem 4-70) that

<2*
= - t2 )
(4-183)

some elements near the base

Figure 4-13 Portion of the discretized fin showing
150 VARIATIONAL AND WEIGHTED RESIDUAL FOR ML LATOVS

where it is further assumed that element I connects nodes 1 and 2, and where L is

the element length. More accurate results are usually obtained if the cross-sectional
area at the base is used instead of the area element centroid.
at the If QK is negative,
heat is conducted toward the base of the fin.

In the second method, the heat removal rate is determined by integrating the
heat loss from the exposed surfaces of the fin that undergo convection. Let us
assume the tip of the fin is insulated, so that only the fin periphery needs to be
considered. Therefore, we have

Q, = JT liP(T - TJ dr (4-1841

However, it is preferred to vvnte this as a sum of M separate integrals over each

of the M elements, or

q, = s lxi"u‘a-T.tdx (4-185)
r-l >

r
It can be shown (see Problem 4-70) that if T = Na is used, and if h, T„ . and^P
are evaluated at x = .r (i.e.. at the centroid of the element) and denoted as h. T,,
and P . then Eq (4-IS5) becomes

/T + T, _ \
(?* = 2 hPL\^ '
^
-
r,J
14-1 861

where Tt and Tt are the temperatures at nodes j and L for element e These nodal
temperatures are known from the solution of Ka = f It is emphasized that the
heat removal rate from Eq (4-186) is more accurate than that from Eq (4-183)

Fin Efficiency

The fin efficiency ij, may be defined in several ways, but only one definition is

given here. This definition is based on the hypothetical condition that the entire fin

is at the base temperature, because this would result in the maximum possible heat
removal rate. This corresponds to an infinite thermal conductivity and. therefore,
is not physically realizable Nonetheless, this definition of fin efficiency docs give
an indication of the relative effectiveness of the fin in increasing the heat removal
rate. Denoting this maximum rate as we may vvnte

(4-187)

where

(?™ = jT*nr. - r.idr (4-183)

 APPLICATION ONE-DIMENSIONAL HEAT TRANSFER IN A PIN FIN

If average values of h, P, and Ta are used for each element, Eq. (4-188) becomes

M
Qnax = 2 hPL{Tb - Ta ) (4-189)
e= I

Again it has been assumed that the tip of the fin is insulated.

Example 4-11

Determine the temperatures, heat removal rate, and efficiency of the circular pin
fin shown The fin is made of pure copper and has a thermal
in Fig. 4-14(a).

conductivity k of 400 W/m-°C. The base is held at Tb = 85°C and the ambient
fluid temperature Ta is maintained at 25°C. The fin length Lr is 2 cm, and the

diameter D is 0.4 cm. The convective heat transfer coefficient h is 150 W/m 2 -°C
and the tip at x = Lf is insulated.

Solution

The first step in the finite element solution is discretization of the fin. Let us
arbitrarily assume only two elements (and three nodes) as shown in Fig. 4-14(b).
It is convenient to summarize the node and element data in tabular form as shown

in Table 4-3. The calculations are summarized below in a form that may be readily
implemented in a computer program (see Problem 4-78).

Insulated
tip

Figure 4-14 Circular pin fin (a) analyzed in Example 4-11 and (b) discretized into two
elements and three nodes.
1 52 VARIATIOVAL AND W UGI tTFD RESIDUAL FORMULATIONS

Table 4-3 Node and Element Data for Example 4-M

Node number t, m
1 00
2 001
3 0 02

Nodes connected
Element number J A

1 1 2
2 2 3

Element I: Node j is 1 . Node A is 2

ij = 00m and xk = 0.01 m

L = xk - x,
~ 0 01 - 0.00 = 0 01 m
D = 0.004 m

A = —
ttD 1
=
7t(0

-l—
004 2

1 2566 x 10" 5 m 2

P = vD = ir(0 004) = 1 2566 x I0' 2

m
(400)(1 2566 x 10 i>r 1 -i]
0 01 L-i ij

0 50265 -0 50265] \v/°c

[ v
[-0 50265 0 50265]
_2
(150)(1 2566 x I0 )(0.01) [2 ll
K“ = 2
6 Ll J

[0 00628 0.00314]
[0 00314 0 00628

50893 -0 4995l]
•
= K',“ + -[-*
k;
49951 0 50893]
wrc

0.50893 0 49951 0
K° = -0.49951 0 50893 0
0 0 0

(150)0.2566 X 10~ 2 )(0.01)(25)

25) [l] _ [0 23561
2
2:
~ ij [0 23561
2:
W
23561
0.23f
f° = 0 23561
23i W
0
 application one-dimensional heat transfer in a pin fin

Element 2: Node j is 2. Node k is 3.

Xj = 0.01 m and = 0.02 m

L ~ xL - Xj = 0.02 - 0.01 = 0.01 m
D = 0.004 m
A = 1.2566 X 10“ 5 m2
P = 1.2566 X 10~ 2 m
K< 2) = K<r i)
!

0.50265 -0.50265
w/°c
-0.50265 0.50265

K« 2 > = K<!>

0.00628 0.00314
0.00314 0.00628
WPC
K< 2) = K(2> + K' 2> = K <! >

0.50893 -0.49951
W/°C
-0.49951 0.50893

0.50893 -0.49951 0
Ka -0.49951 1.01786 -0.49951 YV/°C
0 -0.49951 0.50893

0.23561
f(2> _ yd) _ W
0.23561

0.23561
f“ = 0.47122 W
0.23561

Before application of the prescribed temperatures we have K°a = f", or

~
0.50893 -0.49951 0 0.23561
-0.49951 1.01786 -0.49951 r. = 0.47122
0 -0.49951 0.50893 t3 _ 0.2356 1 _

Note that the assemblage stiffness or conductance matrix K“ is symmetric and

banded with a half-bandwidth of 2 [this also follows from Eq. (3-33)]. The pre-
scribed temperature boundary condition at the base of the fin will be imposed by
using Method 1 from Sec. 3-2. The first step is to modify the first row (since T ]

is to be imposed) as shown below:

1 0 0 85
-0.49951 1.01786 -0.49951 t2 — 0.47122
0 -0.49951 0.50893_ LJjJ 0.23561
154 VARIATIONAL AND W EIGHTH© RESIDUAL IORML LATIONS

In order to presene symmetry, the Af;) entry must be multiplied by the prescribed
temperature and transposed to the right-hand side (in the same row). Therefore, />
must be replaced by 0.47122 — (— 6.4995 1 ){85.) or 42.930. This is the only term
that needs to be modified because K is now symmetric. The system of equations
to be solved is gi\en by

’l 0 0 >' 85
0 I. 01786 -0.49951 T~ = 42.930
0 - 0.49951 0 50893 .
0.23561

Solving this system of linear, algebraic equations by the matrix inversion method
yields the following nodal temperatures:

T, = 85 0°C T2 = 81.8°C r, = S0.8°C

The heat removal rates may now be computed as summarized below.

5 81 8
= 150rt 0 0041(0 0 n ~
.
-, - 25
^ ^

81 8 * 50 8
+ ISOrlOIXMKOOl)^ -
25^

= . 10 + I 06 = 2 16 YV

If the alternate (less accurate) method is used, we get

This is quite different than the result from the method based on integration It is

shown below that the exact value for the heat removal rate is 2.155 W Clearly,

the method based on integration is significantly more accurate.

Finally , fin tKvcvttvcy is readily computed to be

2J*
^ = Q™ =
150— (0 004)(0.02)(85 - 25)

= ^
2.26
« 0.956 or 95.6^

The results from Example 4-1 1 are compared in Table 4-4 to the exact solution
gi\en by (13)

cosh XlL, - x)
Hr) = T„ + {T„ - Ta )
cosh XL,

and
 REMARKS J55

Qr = KkA(Tb - T„) tanh KLf

where

k -

Note that the results for 4 and 8 elements are also given in this table. These results
were obtained by the computer program described in Problem 4-78. The nodal
temperatures, heat removal rates, and efficiencies in Table 4-4 converge to the
exact solution as the number of elements is increased. Five significant digits are
shown for the purpose of comparison only.
A detailed discussion of error predictions and convergence is beyond the scope
of this text. Introductory' material on these topics may be found in the book by
Becker, Carey, and Oden [!4J.

4-11 REMARKS

This chapter began with an introduction to some of the more important integral
methods that lead to the present-day finite element method in nonstructural appli-
cations. Only one-dimensional problems were considered, but the basic groundwork

Table 4-4 Summary of Results for Example 4-11

Temperatures, °C

x cm, Exact 2 Elements 4 Elements 8 Elements

0.0 85.000 85.000 85.000 85.000

0.25 83.998 83.998
0.50 83.134 83.133 83.134
0.75 82.407 82.407
1.00 81.814 81.807 81.812 81.814
1.25 81.354 81.354
1.50 81.027 81.024 81.026
1.75 80.830 80.830
2.00 80.765 80.756 80.762 80.765

Heat removal rate, W

Case Eq. (4-183) Eq. (4-186) Fin efficiency, %*

2 Elements 1.6050 2.1618 95.6

4 Elements 1.8771 2.1569 95.4
8 Elements 2.0149 2.1557 95.3
Exact 2.1552 2.1552 95.3

*Based on Eq (4-187) with Q,, from Eq. (4-186).

156 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

has been laid for extension to two and three dimensions in later chapters. Among
the integral methods introduced here were the Ritz method, the variational or
Rayleigh-Ritz method, and the weighted-residual method. All these approximate
solution methods may be used to solve differential equations. However, we generally
work with an integral form of the governing equation. Each of these methods was
introduced on a global basis, which meant that the trial functions were assumed to
apply globally to the entire problem domain.
The Ritz method is the simplest and least powerful of the methods. A specific
example problem was considered. Following this, the Rayleigh-Ritz method was
introduced and applied to the same problem. This method requires a rudimentary
knowledge of variational calculus, which is the mam disadvantage of this approach.
Some of the more important concepts in the calculus of variations were reviewed.
Then four of the most popular weighted-residual methods were introduced, which
included the point collocation method, the subdomain collocation method, the least-
squares method, and the Galerkin method. These weighted-residual methods do not
require any advanced mathematics beyond ordinary calculus (i.e., variational cal-
culus is not needed in these methods). Moreover,
it was stated that the weighted-

residual methods could be used even when a classical variational principle docs not
exist. Each of these methods was illustrated by solving the same example problem,
and the results of each of the approximate solution methods were compared with
the exact solution None of them did a particularly good job of matching the exact
solution, but the Galerkin and Rayleigh-Ritz methods were among the best.
After some additional mathematics was reviewed, the Rayleigh-Ritz and Gal-
erkin methods were cast into a form that is directly useful in the finite clement
method. The same example problem was solved, approximately, by using trial
functions that no longer applied globally to the entire problem domain but rather
applied locally over each element. These piecewise continuous trial functions gave
rise to two other functions shape functions These particular
that are referred to as
shape functions allow the be continuous within each element and
field variable to

at the ends of each element However, the derivatives at the ends of each clement
are not necessanly continuous Therefore, these particular shape functions arc said
to be C°-continuous, or to possess C°-continuity. In Chapter 7, shape functions
with continuous first derivatives will be introduced; sh*pe functions with continuous
firstderivatives are said to be C '-continuous, or to possess C -continuity. It w'as
1

seen that accurate results could be obtained with the use of the piecewise continuous
trial functions, which are really interpolating polynomials The reader is referred

to the book by Myers [12] for additional material relating to the variational finite
clement method as it applies to one-dimensional problems in heat transfer
the Rayleigh-Ritz method is applied to a second-order differential equa-
When
tion with no first-order derivative present, it results in an integral formulation that
has only a first-order derivative Therefore, the integral formulation is said to be
a
a u eok formulation. !n a similar fashion, the Galerkin method also results in
integrated by
similar result if the term containing the second-order derivative is
parts. This is a fundamental step in the finite element method and one that we will
sec throughout the book. The implication is element method can be
that the finite
 , .

PROBLEMS 157

used to obtain solutions to problems that possess real material discontinuities, such
as composite materials with two or more thermal conductivities.
The emphasis in this chapter has been the solution of one-dimensional, second-
order differential equations. In the next chapter, as in Chapter 3. we turn to the
structural analysis area where the general finite element formulation to problems
in stress analysis is presented. Two additional methods of finite element formula-
tions. which are more readily applied to these problems, are introduced.

REFERENCES

1. Hinton. E., and D. R. J. Owen. An Introduction to Finite Element Computations,

Pineridge Press. Swansea. UK. 1979. pp. 3-5.
2. Huebner, K. H.. The Finite Element Method for Engineers. Wiley, New York. 1975,
pp. 339-357.
3. Arpaci, V. S.. Conduction Heat Transfer. Addison-Wesley. Reading. Mass., 1966.
4. Becker. E. B.. G. F. Carey, and J. T. Oden. Finite Elements: An Introduction, vol. I
Prentice-Hall. Englewood Cliffs. N.J.. 1981, pp. 6-9.
5. Arpaci, V. S.. Conduction Heat Transfer. Addison-Wesley. Reading. Mass., 1966. pp.
163-164
6. Zienkiewicz, O. C., The Finite Element Method. McGraw-Hill (UK). London. 1977,
pp. 74-75.
7. Schwartz. A.. Calculus and Analytic Geometry Holt. Rinehart and Winston, New York.
1967. p. 357.
8. Schw artz. A. Calculus and Analytic Geometry .Holt.
, Rinehart and Winston . New York.
1967. pp. 798-799.
9. Arpaci, V. S., Conduction Heat Transfer. Addison-Wesley. Reading. Mass.. 1966. pp.
439-440.
10. Huebner. K. H.. The Finite Element Method for Engineers. Wiley. New York, 1975.
p. 445.
11. Zienkiewicz. O. C.. The Finite Element Method. McGraw-Hill (UK). London, 1977,
pp. 70-72.
12. Myers. G. E.. Analytical Methods in Conduction Heat Transfer, McGraw-Hill. New
York. 1971, pp. 320-362.
13. Incropera. F. P . and D. P. DeWitt. Fundamentals of Heat Transfer. Wiley, New York.
1981. pp. 105-117.
14. Becker, E. B.. G. F Carey, and J. T. Oden. Finite Elements. An Introduction, vol. 1,

Prentice-Hall. Englewood Cliffs. N.J . 1981. pp. 36-38.

PROBLEMS

4-1 For the problem posed in Example 4-1. venf} that the approximate solution given by
Eq. (4-10) satisfies the boundary conditions given by Eq f-t-S) exactlv

4-2 Reconsider the problem posed in Example 4-1 It is desired to obtain another approx-
imate solution by the first-order Rnz method bv using the following trial function.

jV|t v) = T(1 - XJ )
 1 . 0

160 VARIATIONAL AVO V, LIGHTED RESIDUAL FORMULATION'S

4*13 Reconsider the problem posed in Problem 4-5.

a. Determine the approximate solution by the second-order Raylcigh-Ritz method

by assuming

The functional to be extremized is given in Problem 4- 1

b. Compare the result from part (a) with the approximate solution from Problem
4-12 and with the exact solution
c. Make an educated guess as to what would happen if a scries such as

(2k — Dirt
no = k
2
-
n t sm
10

were used in the Raylcigh-Ritz method for increasingly larger values of n.

4*14 Reconsider the problem posed in Example 4-1

a. Obtain the second-order Raylcigh-Ritz solution that corresponds to the following

trial functions

N,(x) = 41 - x 2 )
N,(x) = 41 - t
4
)

b. Compare the results from part (a) with the approximate solution from Example
4-2 and with the exact solution
c. Compare the results from part (a) with the approximate solution from Problem
4-7

4-15 Extend Eq (4-37) to the ease when the functional F is a function of x, it*). >,(.*).

and y„(jr) where the subscripts are used to denote derivatives

4-16 Consider the following differential equation

~
dx-
+ 6i = lOt 0 < x rs 2

subjcct to the boundary conditions

3(0) « I

3 (2 )
= 0

a. By using the method in Example 4-3, verify that the functional to be extremized
is given by

/ =

7
b. Arc the boundary' conditions given above geometric or natural
c. Do the boundary conditions satisfy Eq (4-46)? Why or why not?
 PROBLEMS 161

4-17 Reconsider the differential equation given in Problem 4-16. State whether each of
the following sets of boundary conditions is geometric, natural, or neither:

a- v(0) = 5 b. yt(0) = 5 c. v(0) = 6

M2) = 3 v(2) = 3 v,(2) = 0
4-18 Reconsider the problem posed in Problem 4-16. Repeat part (a) by using the method
illustrated in Example 4-4.

4-19 Nonlinear differential equations may also have a variational formulation. For example,
consider the differential equation

~
dz T
dx~
+ 97 3 = 5.v’ 0 < .r < 5

subject to

7(0) = 9
7(5) = 0

a. Which term (or terms) makes the differential equation nonlinear?

b. Show that the corresponding variational formulation is given by

1 (drV\
/ = f 373 — 5.x
3
T - dx
2\dx

if the method illustrated in Example 4-3 is used.

4-20 Repeat part (b) of Problem 4-19 by using the method in Example 4-4.

4-21 Consider the following nonlinear differential equation:

d_(dT
+ 57 =0 0<.t<4
dx \ dx

subject to the boundary conditions

dm = 0
dx

and

7(4) = 100

a. Classify the boundary conditions: i.e., are they geometric, natural, or neither?

b. Show' that the corresponding functional / is given by

= /iT2
S
- \(‘ dx
I
f
J0 \ J\

by using the method in Example 4-3.

4-22 Repeat Part b of Problem 4-21 by using the method in Example 4-4.

4-23 The necessary' condition for the existence of an extremum of the functional

/ = f F(.t y,y„yxt ) dx ( 4 - 73 )
*
162 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

is thai us first variation 5/ must be zero pros nlcd that

[m©M~ (4-75a)

(4-75b)

a. Show that the corresponding Euler-Lagrangc equation is given by Eq (4-74).

b. What order differential equation does Eq (4-74) correspond to"! Hint. See Problem
4-25.
c. What is the htghcst-ordcr derivative present in the functional FI
d. Try to generalize the results from parts (b) and (c) Explain why the variational
formulation may be called the it eal formulation

4-24 In Problem 4-23, identify

a. The geometric boundary conditions and explain physically what these imply.
b. The natural boundary conditions and explain physically what these imply

4-25 Consider the following fourth-order differential equation

— + 8—
d*\

dx*
d~\
dx-
;
4- 4v = to 0S,S5
subject to the boundary conditions

v(0) = 0 v<5) = 100

v,<0) = 1 \,(5) = 0

a. is geometric, natural, or neither

Indicate whether each of the boundary conditions
Hint See Eqs (4-73) to (4-791 and Problems 4-23 and 4-24
b. Show that the corresponding functional / is given by

+2,; ~
"I B(S) -•(£) H*
C. Do the boundary conditions meet the conditions expressed in Eqs (4-75)'’ Which
ones?
1
d. What is the highest -order derivative present in the governing differential equation'
In the variational formulation (i.c., in the functional / p
c. From the results of part (d). explain why the variational formulation may be called
the Meal formulation?

4-26 It isnot possible to obtain a classical variational formulation to problems whose

governing differential equation contains an odd-ordered derivative For example,
consider
dz \ d\
+ 4 + 5v = 3
dx : dx

Using the method illustrated in Example 4-3. m to obtain an expression for the

functional /. Describe clearly the difficulties that arc encountered

 PROBLEMS 163

4-27 Reconsider the dilemma posed in Problem 4-26. Using the method illustrated in
Example 4-4. try to obtain an expression for the functional I. Describe clearly the
difficulties that are encountered.

4-28 Reconsider the problem posed in Example 4-1 and obtain additional approximate
solutions by using the trial function

A', (a) = -v(l - a-

4
)

and the following weighted-residua! methods:

a. Point collocation b. Subdomain collocation

c. Least squares d. Galerkin

4-29 Reconsider the problem posed in Example 4-1 and obtain additional approximate
solutions by using the trial function

At, (a) = sin ha

and the following weighted-residual methods:

a. Point collocation b. Subdomain collocation

C. Least squares d. Galerkin

4-30 Recall that the exact solution to the problem posed in Example 4-1 is given by

T(A) = ,000
/l2A(l — A3 )

Consider the following trial function

JV,( a) = a(1 - A3 )

a. Obtain the corresponding approximate solution by using each of the four weighted-
residual methods.
b. How do the approximate solutions from part (a) compare with the exact solution?

Try to generalize this result.

4-31 Solve the problem posed in Example 4-1 by using the point collocation method if

the following two trial functions are used in a second-order approximation:

At,(x) = a(1 - a2 )

At,(a) = a(1 - a 4
)

Take the two collocation points at a = 'A and a = 2

A.

4-32 Repeat Problem 4-31 by using the subdomain collocation method. Use two equally
spaced intervals.

4-33 Repeat Problem 4-31 by using the least-squares method.

4-34 Repeat Problem 4-31 by using the Galerkin method.

4-35 Solve the problem posed in Example 4-1 by using the point collocation method if

the following two trial functions are used in a second-order approximation:

W, (a) = sin tta

N2(\) = sin 3-rr.v

164 VARIATIONAL AND WEIGIITFD RESIDUAL FORMULATIONS

4-36 Repeat Problem 4-35 by using the subdomain collocation method Use two equally
spaced inters als

4-37 Repeat Problem 4-35 by using the least-squares method

4-38 Repeat Problem 4-35 by using the Galerkin method.

4-39 Reconsider the problem posed in Example 4-1 and obtain the first-order approximate
solution for the point collocation method by using the tnal function

N U)t
= i(l - 1
t )

a. Assume the follow mg ‘Collocation points, x = 0 2. 0.4, 0 6, and 0.8

b. Plot the four approximate solutions from part (a) on a graph that also includes
the exact solution How do these results compare to that from Example 4-5?
Which is the best result 1

4-40 Consider the differential equation

~+
at*
100 = 0 Osis 10

subject to the boundary conditions

710) = 0
TOO) = 0

Assume the trial function

TTl
A'i(0 = sm
-J5

Obtain the corresponding approximate solution from the point collocation method
and compare it to the exact solution by showing both on a graph Take the collocation
point at r = 5

4-41 Repeat Problem 4-JO by using the subdomain collocation method

4-42 Repeat Problem 4-40 b> using the least-squares method

4-43 Repeat Problem 4-40 by using the Galerkin method

4-44 Consider the differential equation

~
dx 1
- T + JOftt = 0 0sr<10

subject to the boundary conditions

7X0) = 0
7(10) = 0

a. Assume the tnal function

iV|<0 = x-’UOO - x 2 )

and obtain the corresponding approximate solution by using the Rit?

method
 PROBLEMS 165

b. Obtain the exact solution and compare it with the result from part (a) by showing
both on a graph.

4-45 Repeat Problem 4-44 by using the Rayleigh-Ritz method.

4-46 Repeat Problem 4-44 by using the point collocation method. Take the collocation
point at x = 5.

4-47 Repeat Problem 4-44 by using the subdomain collocation method.

4-48 Repeat Problem 4-44 by using the least-squares method.

4-49 Repeat Problem 4-44 by using the Galerkin method.

4-50 Consider the following differential equation for the temperature T in a bar undergoing
convection to ambient at 25°C:

“—
clx-
3(7' - 25) = 0 0 <.t<2

subject to the boundary' conditions

T(0) = 150

^ dx
= 0

a. Are the boundary conditions geometric, natural, or neither?

b. What does the second boundary condition physically represent?
c. Derive the integral that represents the functional for the problem. Use either of
the two methods illustrated in the text.

d. Explain clearly and concisely why the following approximate solution is appro-
priate.

r (x) = 150 - fl,x(4 - x)

e. Obtain the approximate solution corresponding to the Rayleigh-Ritz method.

f. Obtain the approximate solutions corresponding to the point collocation, subdo-
main collocation. least-squares, and Galerkin methods.

4-51 Verify that the element stiffness matrix for Example 4-9 [which is given by Eq.
(4-141)] evaluates to the result given by Eq. (4-143).

4-52 Verify Eq. (4-145) by performing the exact integration.

4-53 In arriving at Eq. (4-145), an exact integration was performed. Because integrals

such as these arise in essentially all finite element formulations, very sophisticated
numerical integration methods have been developed. These techniques are illustrated
later in the text. One simple method can, however, be introduced here. Let x denote

the value of the x coordinate at the centroid of the element, orx = (x, + xt )/2. The
integral then evaluated approximately by evaluating the integrand at the centroid,
is

pulling the result through the integral and evaluating the trivial integral that remains.
Therefore, an alternate (but approximate) way to evaluate

f I00Qx2N r dx
•'X.
166 VARIATIONAL AND W LIGHTED RESIDUAL FORMULATIONS

from Example 4-9 (and Example 4-10) is given by

looaAN 7 * = iooo (^'

* (t, - x)
JT
^ t

a. Verify that N r is given by

"-[$]
when evaluated at x — (t, + r t )/2

b. Redo Example 4-9 by using this approximate integration method. Repeat only
those parts of the finite element solution that arc different from those in Example
4-9
C. How do the resulting nodal temperatures compare to those obtained from the exact
integration 7
d. Why is it reasonable to expect this approximate integration to yield more accurate
nodal temperatures as the number of elements is increased?

4-54 Consider the differential equation

+ 1000c 2 = 0 0 s x s I

dx-

subjeef to the boundary conditions

7(0) = 25

7X1) = 50

By making use of as much of Example 4-9 as is possible, determine the vatucs of 7

at six equally spaced nodal points

4-55 Repeat Problem 4-54 for the following boundary conditions

7X0) = 25

4-56 Repeat Prebiem 4-54 for the following boundary conditions

7(0) = 50

4-57 Let us determine the governing equation for the uniaxial stress member as follows
Consider the tapered uniaxial stress member shown in Fig P4-5?(a) The cross-
sectional area vanes with x and may be denoted as A(x). or simply A The axial stress
is denoted as a, where a is positive for tensile stresses and negative otherwise.
The
axial force <rA is assumed to vary* according to a first-order Taylor expansion
as

shown in Fig. P4-57(b) Also shown is the body force bA dx as a result of gravity,

where b is the body force (or weight) per unit volume

 -

PROBLEMS 167

oA

bA -5
\l ifx

1 —

(a A
aA -
^ ) dx

(b)

Load
(a)

Figure P4-57

a. By considering the elemental volume shown in Fig. P4-57(b) to be in static

equilibrium in the x direction, show that

+ bA = 0 (4-190a)

b. Rewrite Eq. (4-190a) in terms of the axial strain e by invoking Hooke’s law (which
is a constitutive relationship) or cr = £e. Note that this implies that we are assuming
the material to be linear elastic.
c. Next, eliminate the strain € in the result from part (b) by invoking the strain-

displacement relationship, e = dutch, where u is the axial displacement that is,

in general, a function of .r. In other words, show that Eq. (4- ] 90a) becomes

d_
+ bA = 0 (4-1 90b)
dx

Equation (4- 190b) is now in a suitable form fora Ritz, Raylcigh-Ritz, or weighted-
residual solution. It is the governing equation for the problem.

4-58 Reconsider Problem 4-57 in general and Eq. (4- 190b) in particular. It is desired to
obtain the corresponding functional. Let us proceed as follows: from Eq. (4-55) it

follows that

8 k dx = 0

In the usual manner (by integrating by parts, etc.), show that this may be written as

8/ = AE-~ Sa -8 Jo
[ ^Ae(^)
\dxj
d\ + S [ ubA dx
J
= 0 (4-191a)
dx o 2 0
168 VARIATION AL AND w ElGItTtD RESIDUAL FORMULATIONS

But E du'dx = a and cr4 = P, where P is the axial force, therefore, show that /
itself is given b>

/«=(>» -If <k +

/
*1* (4-191 b)
|

4-59 Reconsider Problems 4-57 and 4-58 in general and Eq (4- 19 lb) in particular. Let
us now obtain the finite clement characteristics (ic„ K' and P) by the Ravlcigh-Ritz
method. We proceed in the following manner

a. Rcwnte Eq (4- 19 lb) on an element basis as follows.

d,+
'-H.-3WS) (
ifbA dx (4-1 92a)

where a typical element r connects node j at r = x, and node k at r = xt as

show n in Fie P4-59 Note that a tj pical node actually represents a plane and not
just a point as shown
b. Since the shape functions from Secs 4-8 and 4-9 arc applicable here (whj 1 ). the
axial displacement if oxer a l> pical clement e can be represented as

if = Na' (4-1 92b)

where N is given by

N -
[ 7T7 f^r] I4-192C)

L -
Figure P4-59 Note that L is defined here to be the clement length, or ** Ji
 k

PROBLEMS 169

and a' by

ae (4-192d)

where, in turn, ut and uk are the axial displacements at nodes j and k, respectively
(positive in the +x direction). Substitute for u r in Eq. (4-192a) with u e given by
c
Eq. (4- 192b), compute cll'lda , set the result to zero, and rearrange the terms to
show that we get

K'a f = F (4-1 92e)

where

dN r dN
K'
-r ~^ AE lu dx (4-1 92f)

and

f' = P NT N TbA dx (4-192g)

c. Treating A, b. and E as constants in any given element by using the values at the

midpoint of the element where .t = x and denoting them as A, b, and £, respec-

tively, show that

1 -1
(4-1 92h)
1 1

and

'-p? bAL i
+ (4-1 92i)
p A
2 i

where L is the element length, or L — .rj —x and Pj and are the axial forces
at nodes j and k, respectively (positive in the +.v direction),
d. After the assemblage step and the geometric boundary conditions (prescribed
displacements) have been imposed, the following matrix equation may be used
to represent the entire uniaxial stress member:

Ka = f

This equation is solved in the usual manner for the vector of nodal unknowns a.
Therefore, the axial displacement at each nodal point is now known. Show that

the average strain I within the element is given by

e = (4-1 92j)

the average axial stress u by

E(u k - Uj)
u — (4-1 92k)
L
170 VARIATIONAL AND V. LIGHTED RESIDUAL tORMULATIONS

and the av cragc axial force F by

- _ AE(ut - »,)
14-1 92m)

These average axial strains, stresses, and forces arc generally associated with the
centroid of the element, i e , at x *= x = {xt + ,r4 )/2.

4-60 Reconsider Problem 4-57 in general and Eq (4- 190b) in particular. Let us now obtain
the finite element characteristics (i.c . K' and P) by the Galerkin weighted-residual
method We proceed in the following manner.

a. Form the weighted-residual equation for a typical element in the usual manner
by w ritmg

n
'[i
ae
S) +m ]* =o
( (

where a typical element e connects node j at x = x, and node A at x ~ xt as

shown in Fig P4-59 Note that a typical node actually represents a plane and not
a point as shown Integrate by parts the term in Eq (4- 193a) containing the second

derivative and show that

N rAE ^1-
<Zr 1
J,
f
(
‘

—dx
,1E ^
dx
,i< + [‘
J,
N'M dx - 0 |4-193b)

b. Note that

AE 7- = 4E« = /to = P
dx

where e is the axial strain, a is the axial stress, and P is the axial force Use this

result only in the integrated term in Eq (4-|93b) to give

PN'I' -
| ,
f
' ~ ~ dx
AL
dx
dx + •
f
J,
t
' N'M dx - 0 |4-193c)

Note that superscript (') has now been added to u to give u' because the assumed
displacement function applies only on an element basis, not globally
C. Since the shape functions from Secs 4-8 and 4-9 arc applicable here (why >). »c
can represent the axial displacement oxer a typical element e as

if = NV (4-1 93d)

where .V is given by Eqs (4-l29a) and (4-127) and a' is given by

(4-193e)

where u, and ut are the axial displacements at nodes j and A. respectively (positive

in the + x direction) Show that Eq (4-l93c) can be written in the form

K'a' = r (4-1930
 PROBLEMS 171

Jx
“"T

:
i

— dx
t r
AE
*

dx
dx (4-1 93g)

|*(

and f
f = P r
+ N TbA dx (4-1 93h)
N’
I
1
r
/

Note that these expressions for the finite element characteristics are identical to
those obtained by the Rayleigh-Ritz method in Problem 4-59.
d. Do part (c) in Problem 4-59.
e. Do part (d) in Problem 4-59.

4-61 Consider the tapered uniaxial stress member shown in Fig. P4-6I(a). The top of the
member is fixed, and a load P is applied to the bottom end. It is desired to obtain
the displacements and the internal strains, stresses, and forces. Although it is possible
to obtain the exact solution, a finite element solution is sought as described below.
The member has a circular cross section with D ~ 1.0 in., d = 0.5 in., H — 12
in., and P = 5000 Ibf. In addition, the material is 0.65c carbon steel with E —
30 x 10
6
psi and b = 0.283 lbf/in.
3
. Using the discretization shown in Fig.

P4-6I(b) with five equally spaced nodes and making use of the results of Problem
4-59 (or Problem 4-60). do the following.

a. Determine the displacement at each nodal point.

b. Determine the element resultants that are defined here to include the strains,
stresses, and forces within each element.
c. Show that the body force contribution to the assemblage nodal force vector may
be neglected with very little error,

d. Compare these results with those from the exact solution.

4-62 Solve Problem 4-61 if the member has a circular cross section with D — 2 cm,
d = 1 cm, H = 25 cm, and P = 18 kN. In addition, the material is 0.6% carbon
E = 21 x 10'° N/nf and b = 77 kN/m
3
steel with .

F
D *]

Figure P4-61
172 VAKIVlUfSU SI M*\« tAltONS

4-63 Sohc Problem 4 61 if the criws vevtnm of the member is square

4-64 SoKv Prs'Wcm 4 62 if live cross section of the member is square

4-65 \Vhat is the fuff bandwidth of the assemblage Milfnrss ma'm for the discretized
umasial stress member in Problem 4 6! Docs 1 q (3-33) foe the eom\t result?

4-66 |jct us dense 7(0 in a circular pin fin of

the governing equation for the temjvrjJurc
sarswgcross sccttonaJareaAanJpeftmctcr/’ asshowninl if P4 6Wa) Weprivced
b) takine an infinitesimal slice of the fin with dullness </« and represent the heat
transfer rate to the slice from condition as q,\, where «/, t> the heat flu* A* shown
in I ic 1*4 With). t)ic corresponding beat transfer rate leasing llie slice is oNaircdh)
assuming a first order Ta)Ior expansion In adJition, tfie conscctisc heat transfer
from the cxpncsl surface of the fm is jroen hs h/’if - Tt \ i/r. where h is the
c«*nvcetoc heat transfer coefficient and T. is the ambient fluid temperature far removed
from the fin The length of the fm is /.,

8. I mma \tead)-state cncrps balance on the inlimtcsmnl slice shown in lip

P4 With), show that

(4-1948)

Figure P4 66
 PROBLEMS 173

b. Next, invoke Fourier’s law of heat conduction

qx = -k— (4-1 94b)

dx

where k is the thermal conductivity. Equation (4- 194b) is referred to as a con-

stitutive relationship . The minus sign is necessary’ because q x is assumed to be
positive in the direction of decreasing temperature. With the help of Eq. (4- 194b),
show that Eq. (4- 194a) becomes

- ItP (T - Ta) = 0 (4-1 94c)

Equation (4-194c) is now in a suitable form for a Ritz, Rayleigh-Ritz, or weighted-

residual solution and represents the governing equation for the problem.

4-67 Reconsider Problem 4-66 in general and Eq. (4- 194c) in particular. It is desired to
obtain the corresponding functional. Let us proceed as follows. From Eq. (4-55) it

follows that

8/ = f — ( kA
clx )
— hP(T - T„) 87' dx = 0
1_« V

In the usual manner (by integrating by parts, etc.) show that this may be written as

Lr L/
f i r
- 5 - hPT2 dx + 5 hPTJdx = 0 (4-1 95a)
Jo 2 Jo

But k clT/tlx - -qx where

, qx is the heat flux from conduction. Therefore, show
that / itself is given by

Lf Lf
f
-- 1
hPT2
(
dx+\ hPTJdx (4-1 95b)
2 Jo Jo

4-68 Reconsider Problems 4-66 and 4-67 in general and Eq. (4- 195b) in particular. Let
us now obtain the finite element characteristics (i.e., K‘ and f‘) by the Rayleigh-
Ritz method by proceeding in the following manner:

a. Rewrite Eq. (4- 195b) on an element basis as

hP(Te ) 2 dx + f ItPTJ'dx (4-1 96a)

•S

where a typical element e connects node j at = x; and node k at .v = xh as

shown in Fig. 4-12. Each node actually represents a planar surface, not just a

point.
174 VARIATIONAL AND tt EIGMTED RESIDUAL FORMULATIONS

b. Since the shape functions from Secs. 4-8 and 4-9 are applicable here (why?), the
temperature T* over a typical element e can be represented as

Tf = NV (4-196b)

where N is gisen by

N
L r*
_ xj
—
*t-
-!
tj
(4-196c)

and a' by

H (4*196d)

where Tj and Tk are the temperatures at nodes j and k. respectisely. Substitute for
7* in Eq (4-196a) with T given by Eq (4- 196b), compute dl'ldaf, set the result
to zero, and rearrange the terms to y leld

K'a' = T (4*1 96e)

where

K' = K' + (4-196fl

and

v = r; + tu (4-196g)

and. further.

(4-196h)

K' = v K ThPS dx (4-196i)

J

r; = (4-196j)

and

N ThPTa dx (4-1 96k)

fj. = J

Note that K' and K[, 2 x 2 element stiffness or conductance matrices, and
arc
the nodal force sectors and f^ are both of size 2 x 1
c. It should be noted that h. A. P. and T„ may all be functions of x Howeser,
it.

let us perform the integrations in Eqs (4-t96h) to (4-19614 approximately by

taking the following approach Let us treat these parameters as constants in 3ny
gi\en element by using the salues at the midpoint of the element where r = x
and by denoting them as I, A. 4. P. and Ta Show that

(4-196m)
 PROBLEMS 175

K CV (4-196n)

P
*9
= (4-196p)

hPT0 L 1
(4-1 96q)
2 1

where qs and qt represent the heat fluxes from conduction at nodes j and k; Aj
and Ak
are the values of the cross-sectional areas at nodes
j and k; and L is the
element length (i.e., L =
x k — rj). In Chapter 8 it will be seen how convection,
radiation, and prescribed heat flux boundary conditions may be included. In the
fin model, heat fluxes are not usually imposed on the ends of the fin; therefore,
is identically zero for all elements.

4-69 For the pin fin problem formulated in Sec. 4-10, begin with the expressions for the
element characteristics [given by Eqs. (4-174) to (4-176)) and derive Eqs. (4-179)
to (4-181).

4-70 The heat removal rate from a fin may be computed by Eq. (4-183) or by Eq. (4-186)
once the nodal temperatures are known.

a. Derive Eq. (4-183) by beginning with Eq. (4-182).

b. Derive Eq. (4-186) by beginning with Eq. (4-185).
c. Which of the two methods gives more accurate results? Why?
4-71 Consider the fin shown in Fig. P4-71(a). The base of the fin (at .t = 0) is held at a
fixed temperature Tb , the surface of the fin (except the tip) is exposed to a fluid at

temperature Ta , and the tip of the fin is insulated. It is desired to obtain the steady-
state temperatures within the fin and the heat removal rate. Although it may be
possible to obtain an exact solution, a finite element solution is sought. The fin has
a circular cross section with D = 0.5 cm, d = 0.2 cm, and Lf = 3 cm. It undergoes
convection to a fluid at T„ = 25°C with h = 75 W/m 2 -°C. The fin is made of cast
aluminum (4.5% copper) with k = 168 W/m-°C and has a base temperature Tb of
90°C. Using five equally spaced nodes as shown in Fig. P4-71(b) and making use
of the results of Sec. 4-10 or Problem 4-68:

a. Determine the temperature at each nodal point.

b. Determine the heat removal rate from the fin using the two methods given in Sec.
4-10. Which of these is more accurate? Determine the fin efficiency.

4-72 Solve Problem 4-71 for a fin of circular cross section with D - 0.20 in., d = 0.10
in., and L{ — 1.5 in. The fin is made of mild steel (1% carbon) that has a thermal
conductivity of 26 Btu/hr-ft-°F and undergoes convection to a fluid at 70°F with a
heat transfer coefficient of 50 Btu/hr-ft 2 -°F. The base temperature is 250°F.

4-73 Consider the straight fin of a rectangular cross section shown in Fig. P4-73. The fin

is made of mild steel (1% carbon) that has a thermal conductivity of 26 Btu/hr-ft-
°F. Both the upper and lower surfaces undergo convection to a fluid at 35°F with a
2
heat transfer coefficient of 30 Btu/hr-ft -°F. The base temperature is 120°F. Take
Lf = 1 .625 in. and r = 0.375 in. The fin width U' is much greater than Lf and so a
one-dimensional temperature distribution is sought.
 176 VARIATIOS'AL AND V. LIGHTED RESIDUAL FORMULATIONS

Figure P4-71

Figure P4-73
 PROBLEMS J77

a. Discretize the fin into four equal-length elements with five nodes.
b. Using the results of Sec. 4-10 or Problem 4-68, determine the temperature at each
nodal point.
C. Determine the heat removal rate in two different ways. Which of these two results

ismore accurate? What is the fin efficiency?

d. Compare the results from parts (b) and (c) with those from the exact solution.

4-74 Solve Problem 4-73 made of cast aluminum (4.5% copper) with a thermal
if the fin is

conductivity of 168 W/m-°C. The convective heat transfer coefficient is 85 W/nr-

°C on both the upper and lower surfaces to a fluid at 5°C. The base temperature is

at 95°C. Take = 3 cm and t = 0.75 cm.

4-75 Consider the straight fin of triangular cross section shown in Fig. P4-75. The fin
material is aluminum alloy 2024-T6, and has a thermal conductivity of 177 W/m-
°C. Both the upper and lower surfaces convect to a fluid at 35°C through a heat
transfer coefficient of 150 W/nr-°C. Assume a base temperature of 100°C and take
Ly = 1.0 cm and H — 0.4 cm. Since the fin width H7 is much greater than Lt . a one-
dimensional temperature distribution is expected.

a. Discretize the fin into four equal-length elements with five nodes.
b. Use the results of Sec. 4-10 or Problem 4-68, and determine the temperature at

each nodal point.

c. Determine the heat removal rate in two different ways. Which of these two results

is more accurate? What is the fin efficiency?

4-76 Solve all parts of Problem 4-75 if the fin is made of mild steel with k = 26 Btu/hr-
ft-°F, h = 2
150 Btu/hr-ft -°F, Tb = 200°F. Ta = 75°F, H = 0.25 in., and L, = 1.0

in.

Figure P4-75
 178 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

4-77 Convert the TRUSS program m Appendix B into one that could be used to fine

displacements, strains, stresses, and forces within a uniaxial stress member, A

up to 30 elements (and 31 nodes) with up to five different materials. The proj

should be able to handle prescribed displacements as well as imposed nodal foi

Take advantage of the developments in Problem 4-59 (or 4-60) A uniform taper
be assumed

4-78 Convert the TRUSS program in Appendix B into one that could be used to fini

temperature distribution within, the heat removal rate from, and the efficiency (

arbitrary fin Allow up to 30 elements (and 31 nodes) with up to five diffi

materials The program needs to handle only prescribed temperature boundary

ditions (and insulation, which, being a natural boundary condition, is automatn
included) Take advantage of the development m Sec. 4-10 Allow for spac
dependent thermal conductivities, cross-sectional areas, heat transfer coefficv
perimeters, and ambient fluid temperatures by taking the following approach. Ch
SUBROUTINE PROPTY to the following

SUBROUTINE PROPTY (L, MATFLG, DATMAT,

1 THERCN, AREA, HTC, PERIM, TABB)

C-. . DEFINES THE THERMAL CONDUCTIVITY, CROSS-SECTK

C AREA, HEAT TRANSFER COEFFICIENT, PERIMETER, ANI
C AMBIENT TEMPERATURE FOR ELEMENT L

DIMENSION MATFLGOO), DATMAT(5,S>

NFLAG - MATFLG(L)

THERCN - DATMATf NFLAG, 1)

IF (THERCN .LT. 0. ) CALL VPR0P (THERCN)

AREA = DATMAT(NFLAG, 2)
IF (AREA .LT. 0. CALL VPROP) ( AREA )

HTC = DATMAT(NFLAG, 3)
IF (HTC .LT. 0. CALL VPROP (HTC)
)

PERIM = DATMAT(NFLAG, 4)
IF (PERIM -LT. D. CALL VPROP )
( PERIM

TAMB = DATMAT NFLAG, S) (

IF (TAM8 -LT. 0. CALL VPROP) ( TAMB

RETURN
END

Note that if a negative value is read for any property, then SUBROUTINE VPROP is cal
which is gucn by

SUBROUTINE VPROP (PROP)

COMMON /CONST/ C( 10)
COMMON /VPROPS/X
 .

PROBLEMS 179

C HATHEHRTICftL CONSTANTS:
PI = 3.141513

IPROP = -PROP
GO TO (1, 3, 3, 4, 5) , IPROP

1 PROP — definition of first variable property

as a function of x and the C(i)'s

RETURN

3 PROP - definition of second variable

property and so forth.
RETURN

3 CONTINUE
RETURN

4 CONTINUE
RETURN

5 CONTINUE
RETURN

END

Note that if PROP is negative, it is converted to a positive integer IPROP. The computed
go to then transfers control to the line whose label is numerically equal to IPROP. Thus by
simply including the proper FORTRAN statements in SUBROUTINE VPROP, variable
properties arc easily accommodated. The C(i)’s are user-defined constants that should be
read in Section 1 of the input file (after NNODES, NELEM, etc.) dnd may be used in

SUBROUTINE VPROP. By way of example, let us say that we want to use this technique

to allow for the varying cross-sectional area for the fin in Problem 4-75. Let us define
C( 1 ) — 0.004 (for H in meters) and C(2) = 0.01 (foriyin meters). If we use a “-3” as
the input for the area A in Section 4 of the input file (sec Appendix B for a description of

the input to the TRUSS program), the statement with label “3” should read

3 PROP = C(l) * <1. -X/C(5))

since this in effect gives

A = Wll( 1 .
- jc/L/i

which represents the correct area variation with x (W is unity since a unit width is assumed).
Note that the variable X represents the x coordinate and is passed to VPROP via the labeled
common (namely, VPROPS).

4-79 Use the computer program from Problem 4-77 (or one furnished by the instructor)
to solve Problem 4-61

a. Run the program for 5, 9, and 17 equally spaced nodes.

b. What is concluded from the results of part (a)?

4-80 Use the computer program from Problem 4-77 (or one furnished by the instructor)
to solve Problem 4-63.
 VARIATIONAL AND WEIGHTED RESIDUAL FORMULATIONS

a. Run the program for 5, 9. and 17 equally spaced nodes,

b. What is concluded from the results of part (a)?

4-81 Use the computer program from Problem 4-78 (or one furnished by the instructor)
Use SUBROUTINE VPROP to include the spacially varying
to solve Problem 4-71.
and perimeter.
cross-sectional area

a. Run the program for 5, 9. and 17 equally spaced nodes

b, What is concluded from the results of part (a) 9

4-82 Solve Problem 4-72 by using the computer program from Problem 4-78 (or one
furnished by the instructor) to solve Problem 4-72 Use SUBROUTINE VPROP to
include the spacially varying cross-sectional area and perimeter

a. Run the program for 5, 9, and 17 equally spaced nodes

b. Is there a significant difference in the results from part (a)? Explain

4-83 Solve Problem 4-73 by using the computer program from Problem 4-78 (or one
furnished by the instructor) Use 5. 9, and 17 nodes to demonstrate increased accuracy
as the clement size decreases

4-84 Solve Problem 4-74 by using the computer program from Problem 4-78 (or one
furnished by the instructor) Use S. 9. and 17 nodes to demonstrate increased accuracy
as the element size decreases

4-85 Solve Problem 4-75 by using the computer program from Problem 4-78 (or one
furnished by the instructor) Use 5. 9. and 17 nodes and SUBROUTINE VPROP to
model the cross-sectional area variation with x

4-86 Solve Problem 4-76 by using the computer program from Problem 4-78 (or one
furnished by the instructor) Assume 5. 9, and 17 nodes, and use SUBROUTINE
VPROP to model the cross-scctional area variation with x

4-87 Consider the annular fin of constant thickness as shown in Fig P4-87 The outer
edge at radius R„ is insulated The inner edge at radius/?, is held at the base temperature
Tb Both top and bottom surfaces convect to a fluid at temperature Ta through a heat
transfer coefficient h

a. Show that the governing equation is given by

£(**£)
- zw - r-> = 0

and the boundary conditions by

7XR.) = Tb

f(M-o
dr

b. Explain why the finite clement formulation from Sec. 4- 10 or Problem 4-68 is

applicable providing that we use

A = -~rt
 PROBLEMS 181

Figure P4-87

and

P = 4-r

4-88 Use the program from Problem 4-78 (or one furnished by the instructor) to determine
the temperature distribution within, the heat removal rate from, and the efficiency
of the brass annular fin shown in Fig. P4-87. Take k = 60 Btu/hr-ft-°F, = 0.125
t

in., R, = 0.625 in., R„ = 1.5 in.,Y = 300°F. h

fc
= 20 Btu/hr-ft 2-°F. and Ta =
70'F. Use 30 nodes.
Hint: See Problems 4-78 and 4-87 and make use of SUBROUTINE VPROP to
take the variable cross-sectional area A and perimeter P into account. The variable
X in SUBROUTINE VPROP should be interpreted to be the radial coordinate r.

4-89 Do Problem 4-88 for an annular fin made of A1SI 1010 carbon steel with k = 64

\V/m-°C. t = 5 mm, R, = 1.0 cm. R„ = 2.0 cm. Th = 160°C, h = 15 \V/m 2 -°C,

and Ta — 40’C. Use 30 nodes. (See Problem 4-88 for a hint.)

4-90 Consider the differential equation and boundary conditions in Problem 4-5.

a. Derive the integral expressions for the element characteristics. Use the variational
method.
b. Evaluate the integrals from part (a) by using the shape functions given by Eqs.
(4-127) for the two-node one-dimensional element.
C. Use four equally spaced nodes (three elements) and determine the value of T at

these nodes.

4-91 Solve all parts of Problem 4-90 by using the Galerkin method.

4-92 Consider the differential equation and boundary conditions in Problem 4-16.

a. Derive the integral expressions for the element characteristics. Use the variational

method.
 •

182 VARtATIOVAL AND VV LIGHTED RESIDUAL FORMULATIONS

b. Evaluate the integrals from part (a) by using the shape functions given by Eqs.
(4-127) for the two-node one-dimensional element.
c. Use four equally spaced nodes (three elements) and determine the value of \ at

these nodes

1-93 Solve all parts of Problem 4-92 by using the Galerkin method

1-94 Consider the differential equation and boundary conditions in Problem 4-25. Try to
derive the integral expressions for the element characteristics Use the variational
method Can the shape functions for the two-node one-dimensional element be used
to evaluate the integrals 9 Why or why not 9 Him Integration by parts must be used
twice.

1-95 Do Problem 4-94 by using the Galerkin method

1-96 The differential equation given in Problem 4-26 docs not have a classical variational
formulation

a. Use the Galerkin method and derive the integral expressions for the finite clement
characteristicsAllow for only natural and geometric boundary conditions
b. Is each of the stiffness matrices symmetric ’

c. What can be concluded when the differential equation contains an odd-order

derivative 9

1-97 Consider the differential equation and boundary conditions in Problem 4-44

a. Derive the integral expressions for the element characteristics Use the variational
method
b. Evaluate the integrals from part (a) by using the shape functions given by Eqs
(4-127) for the two-node one-dimensional clement
C. Use four equally spaced nodes (three elements) and determine the value of T at

these nodes

4-98 Solve all parts of Problem 4 97 by using the Galerkin method

4-99 Consider the differential equation and boundary conditions in Problem 4-50

a. Derive the integral expressions for the element characteristics Use the variational
method
b. Evjhutc the mtegtak from fan (a) by using the shape functions given by Eqs
(4-127) for the two-node one-dimensional element
c. Use four equally spaced nodes (three elements) and determine the value of T at
these nodes

4-100 Solve all parts of Problem 4-99 by using the Galerkin method

4-101 Consider the differential equation and boundary conditions in Problem 4-87

a. Derive the integral expressions for the clement characteristics Use the variational

method Note that the elements m this case arc concentric annuli with nodes /
and k at radii r, and rt respectively
.

b. Evaluate the integrals from part (a) by using the shape functions given by Eqs
(4-127) for the two-node onc-dimcnsional clement Note that r must be replaced
by r, by r and xt by r* Do not cv aluatc the integrals exactly Instead, evaluate
r
the integrand at the clement centroid, and proceed
 PROBLEMS 183

4-102 Solve al! parts of Problem 4-101 by using the Galerkin method.

4-103 Solve Problem 4-8S by hand using two elements and the formulation from Problem
4-101 or Problem 4-102.

4-104 Solve Problem 4-89 by hand using two elements and the formulation from Problem
4-101 or Problem 4-102.

4-105 Use the variational approach to determine the integral expressions for the element
characteristics for the differential equation and boundary conditions given in Problem
4-19. Note that the differential equation is nonlinear. Comment on how this problem
might be solved numerically with the finite element method.

4-106 Solve Problem 4-105 by using the Galerkin method to derive the integral expressions
for the element characteristics.
 General Approach
to Structural Analysis

5-1 INTRODUCTION

One of the most important and challenging steps in ail finite element analyses is

the derivation of the finite element characteristics. It should be recalled that these
-

are defined to be the expressions for the element stiffness matrix K' and nodal
force vector ff The superscript (') serves as a reminder that the expressions are
derived on an element basis. In the last chapter these finite element characteristics
were derived based on the Rayleigh-Ritz and Galerkin methods. The former is a
variational method, whereas the latter is based on the weighted-residual method.
These methods, particularly the Galerkin method, are easily applied to nonstructura!
two- and three-dimensional analyses, as will be seen in later chapters. However,
in stress and structural analysis applications, the finite element characteristics are
more easily derived by invoking one of the following: (1) the principle of minimum
potential energy, (2) the principle of virtual displacements, (3) the minimum com-
plementary energy principle. (4) the principle of virtual stress. (5) Reissner’s prin-

ciple. or (6) Hamilton’s principle flj.

The principles of minimum potential energy and of virtual displacements are
used in this chapter. Both approaches result in the so-called stiffness (or displace-

ment) method, wherein the primary unknowns are the nodal displacements (as
opposed to the nodal forces). The finite element formulation based on these two
principles is presented in Sec. 5-5 after each is introduced. The four remaining
methods are outside the scope of this book. This chapter is further restricted to

static stress analysis.

The chapter begins with a review of the basic concepts of elasticity, which
include the notions of stress and strain at a point, principal stresses, and constitutive
1 86 GENERAL APPROACH TO 5TRLCTURAL AVALS SIS

relationships (generalized Hooke’s law); the equations of static equilibrium and

compatibility; and the strain-displacement relationships. This is followed by the
development of the principles of minimum potential energy and virtual displace-
ments for direct use in the finite element method. The piecewise continuous trial

functions that are used to represent the displacements within an element will be
referred to as displacement functions. The entena that these functions must satisfy
to ensure convergence are delineated in Chapter 6 Although the approach is illus-

trated with a simple one-dimensional example, it should be emphasized that the

expressions to be derived for the finite element characteristics are applicable to
one-, two-, and three-dimensional stress analysis problems, as well as to those with
axisymmetry. In fact, these results can even be used in the analysis of plates and
shells.

5-2 BASIC CONCEPTS IN ELASTICITY

The terms stress and strain as used below are defined quantitatively in subsequent
sections. For now let us define stress as force per unit area and strain as elongation
per unit length. In this book only linear elastic materials are considered This
particular class of materials behaves in such a way that when loaded and subse-
quently unloaded, the stress-strain relationship is linear, as shown in Fig 5- 1(a),

and the material returns to its original undeformed state A material that returns to
its original shape with a stress-strain relationship, as depicted in Fig. 5- 1(b). is said
to be nonlinear elastic The elastic limit of the material is the stress at which
permanent deformation occurs The proportional limit is the same as the elastic
limit for linear elastic materials. The analysis of structures outside the linear elastic
range is beyond the scope of this book.
For a more formal development of the material in this section, the reader is
referred to references 12—4] for several excellent books on the subject In keeping

<al (b)

Figure 5-1 Stress-strain curves for (a) linear clastic and (b) nonlinear elastic materials
188 GENERAL APPROACH TO STRUCTURAL ANALYSIS

when the body is cut along some plane in general, a force R is required to maintain
static equilibrium.Such a plane is shown in Fig. 5-2(b), where the force R has
been resolved into a normal component R„ and two tangential and orthogonal
(perpendicular) components R„ and R l2 The latter two components are in the plane
.

of the of working with the entire area A,

cut. Instead let us work with a much
smaller area AA
such that the respective forces are AR„. AR , ,, and AR, 2 as shown
in Fig. 5-2(c). The normal stress o„ is then defined as

AR„
tr„ lim (5-1
AA— AA
and the two shear (or tangential) stresses as

AR ,,
——
AR,-,
cr,i = ,
lim and
,
tr,
2 = lim * (5-2)
AA-0 AA AA-0 AA
In other words, as the area over which the force acts is reduced to zero at a point,
three different stresses arise, one normal to the plane containing the point and the
other two tangential The two tangential stresses are referred to as shear stresses.
The shear stresses are orthogonal to each other as well as to the normal stress
In the United States where the English engineering system of units is still being
used, the customary unit of stress is pounds per square inch (psi) or kilopounds per
square inch (ksi) In the International System of Units (SI), the unit of stress is
newtons per square meter (N/m 2 ) For conversion purposes, we may use

1 psi = 6895 N/m 2 (5-3)

The trend in the United States in engineering has been and will continue to be
toward the SI system of units It is for this reason that both sets of units are used
in this text.
Let us now consider an infinitesimal volume element of a body positioned at
a point in a global xyz-coordinate system as shown in Fig 5-3(a) There are six

faces to the cube, and on each face one normal and two shear stresses act as shown
Note that the two subscripts are identical for normal stresses, whereas they are
different for shear stresses Thus, no ambiguity should result in using the same
symbol to represent both types of stresses This nomenclature makes use of the
following convention The subscript is associated with the outward normal to
first

the plane over which the stress acts, and the second subscript is associated with
the direction of the stress Positive faces are defined to have outward normals in
the same direction as the coordinate axes, negative faces have outward normals
opposite the coordinate axes The three visible faces in Fig 5-3(a) happen to be
positive faces, while the three hidden faces are negative. An infinitesimal two-

dimensional element is shown in Fig. 5-3(b), where the top and right sides are

positive faces and the bottom and left sides are negative faces.
sign
With the notions of positive and negative faces in hand, the following
convention is adopted Normal and shear stresses arc considered to be positive tf

directed in the positive coordinate direction on a positive face.

Normal and shear
negative coordinate
stresses are also considered to be positive if directed in the
 llLnLKAt. ArfKUftUl II
J

Figure 5-4 Infinitesimal two dimensional element with stresses and body forces shown

some continuous manner Consequently, a first-order Taylor expansion is used to

represent the stresses Each of the stresses in Fig 5-4 is shown with a positive
sense The two components of the body force per unit volume (b, and b are also
shown. The reader may show that the following equations result if static equilibrium
is to be assured (see Problem 5-1)

—
do.,
— +
dor\,
—— + fc, = 0 (5-4a)
ax dv

do.. da. 4
+ -r + 1 ,

b, = 0 (5-4b)
dx d\

cr„ — o\, (5-5)

These equations can be extended to the three-dimensional case by taking an ele-

mental volume dx dy dz and applying the equations of static equilibrium in a similar
manner (see Problem 5-2) The final result is given by

do„
dr
+
do\,

dv
+ ^
dr
+ b, = 0 (5-6a)

d(T n

dx
do, 4
dl
+ ^
dz
= 0 (5-6b)

tkjj- do,. do..

+ + (5-6c)
~dx "aT "aT
 BASIC CONCEPTS IN ELASTICITY 191

with

cr,.. = cr, r
cr,. = a., and or, = cr t - (5-7)

Obviously. b. is die r component of the body force per unit volume. Equations
(5-4) and (5-6) are known as the equilibrium equations in two and three dimensions,
respectively.

Principal Stresses

The reader will recall that Mohr's circle may be used to obtain the two-dimensional
stresses at a point on a plane whose normal makes an angle 0 with the axis. Such
a plane is shown in Fig. 5-5. Because a computer program implementation is our
ultimate goal, the transformation equations from ct„. cr,, , and an to or rV . or,.,-,

and ay,- are given here:

cr tV = ’/a (ay t cr,,) + '/’(ay - a,,) cos 20 -f cr n sin 20 (5-8a)

cr,-,. = Zi (cr„ f d„) - '/: (ct„ - cr,.,) cos 20 - a,, sin 20 (5-8b)

cr,-,. = - '/2 (cr„ - a-,,) sin 20 4- a,, cos 20 (5-8c)

Note that oy,- is not shown in Fig. 5-5 and is the norma) stress on the plane with
a normal in the y' direction.
Now let us summarize the equation for the principal stresses in two dimensions.
Recall that the principal stresses occur on the so-called principal planes whose

Figure 5-5 Element for transformation of stress at a point.

1 92 GENERAL APPROACI TO STRUCTURAL AN ALV S IS
£

outward normals are rotated through an angle 0,, from the .r and v axis, respectively,
where

tan 20,, (5-9)

The two principle stresses, Oj and or3 , are given by

By definttton cr, is taken to be the algebraically larger principal stress, as shown

on the Mohr’s circle in Fig 5-6 The reader should recall from solid mechanics
that Mohr’s circle of stress is a circle on the a-T plane, where cr and t denote
general normal and shear stresses, respectively Therefore, the maximum principal
stress is CTj and the minimum principal stress is a 2 In terms of these principal .

stresses, the maximum shear stress T ra , is given by

g| ~ <T2
(5-11)
2

and occurs on a plane onented 45° from the principal planes. For the purpose of
plotting Mohr’s circle or using Eqs (5-8), a shear stress is positive if it would
cause a counterclockwise rotation about a point within the element and is negative
otherwise
The principal stresses for the general three-dimensional state of stress are given
by the solution of the following cubic equation

cr
3 - /, cr
2
+ I2 a - l3 = 0 (5-1 2a)

Figure 5-6 Mohr’s circle for the two-dimensional state of stress

 .

BASIC CONCEPTS IN ELASTICITY 193

where and /, are known as the stress tensor invariants that are defined by

/. = OTr + + Or_ (5-12b)

I2 II
b s tf -t-
0\,'O'-. + O’.-rO’u “ CTn - 2
O', . - cr
2
(5-1 2c)

0 c, On <>u-

h = det (5-12d)
_o-., O’.-, a.._

Since Eq. (5- 12a) is third order in a. three roots may be obtained. These roots
represent the three principal stresses.

Example 5-1

For the two-dimensional state of stress shown in Fig. 5-7:

a. Determine the principle stresses.

b. Show the orientation of the element that results in the stresses from part (a).
c. Determine the maximum shear stress and the associated normal stresses.
d. Show these results on Mohr’s circle of stress.

Figure 5-7 State of stress for Example 5-1

194 GCNLRAL APPROACH TO STRUCTURAL ANALYSIS

or

CT| = 1828 psi and cr 2 = -3828 psi

Note that o, is taken to be the algebraically larger stress [see the Mohr’s circle in
part (d)]
b. The angle 0,, is easily computed from Eq. (5-9) to be

f — 2000)
2(
0^ * V> tan
[lOOO - ( — 3000)_

or

0 ;)
= -22 5°

Therefore, the principal stresses from part (a) result if the clement is rotated — 22.5°
(or 22.5° clockwise) as shown in Fig 5-8(a) Note that the shear stress is zero in
(his state of stress

c. The maximum shear stress is given by

o, - 02 1828 ~ (-3828)

Figure 5-8 Element orientations for (j) principal stress state and (b) maxrmum shear stress
state, (c) Mohr's circle of stress
 BASIC CONCEPTS IN ELASTICITY 195

and occurs on an element rotated 45° counterclockwise from the principal stress
= 45° + (-22.5°) = 22.5° as shown in Fig. 5-8(b). From
orientation, or at 0
Eqs. (5-8),

= 1000 + (-3000) + 1000 - (-3000)

OVr> cos [2(22.5°))
j
+ (-2000) sin [2(22.5°)) = - 1000 psi

OVv’ = o-jV = ~ 1000 psi

1000 - (-3000)
avo = ~ ^ sin [2(22.5°))

+ (-2000) cos [2(22.5°)] = -2828 psi

The negative sign of cr rV means that the shear stress on the x'x' face is directed
as shown in Fig. 5-8(b), since this would cause the element to rotate clockwise
about some internal point.
d. Mohr’s circle is shown in Fig. 5-8(c) and clearly indicates that the shear
stress is zero on the principal planes and that the maximum shear stress results
when the element is rotated 45° from the principal stress orientation. Recall that a

counterclockwise element rotation 0 is represented on Mohr’s circle with a coun-

terclockwise rotation 20. H

Strain at a Point

Strain like stress generally varies from one point to another in a body under load.

Consider a square element whose sides are of unit length as shown in Fig. 5-9(a).
Under the action of some external loading, the element will deform such that the
sides of the square are no longer perpendicular. In Fig. 5-9(b) the element is shown
after such a deformation. The normal strains e„ and are defined with the help

of Fig. 5-9(b) as

= x'a ~ xa (5-13a)

and

= y'u - y„ (5-13b)

Recall that these definitions are based on the fact that OA and OB are both of unit
length before the deformation. Also, from this figure it is seen that the shear strain
en is defined as

= p! + y2 (5-13c)

These normal and shear strains are all shown with a positive sense. Hence, the
shear strain e u is positive when the angle BOA becomes smaller than rr/2. Note
that two subscripts are again used: for normal strains the two subscripts arc identical.

It is seen, therefore, that positive normal and shear stresses will give positive norma!
198 GCNCRAL APPROACH TO STRUCTURAL ANALYSIS

The relationship stems from the fact that there are two displacements (in two
dimensions) and three strains Therefore, the strains must be related to the dis-
placements with the relationship given by Eq (5-22)
In Problem 5-10, the reader is asked to show that Eq (5-22) holds and in
Problem 5-11 to state the six compatibility equations that must hold m three di-
mensions Since these equations are automatically satisfied in the stiffness approach
to FEM, the three-dimensional form of these equations is not explicitly stated here.

A Constitutive Relationship —Hooke’s Law

As mentioned, only linear elastic materials that undergo small deformations are
considered in this text It is assumed that the deformations do not result in strains

that would cause the material to exceed the proportional limit; that is, the material
is assumed to remain within the elastic limits of Hooke's law It should be recalled

that in the most simple form, Hooke's law states that the normal stress tr is pro-
portional to the normal strain e in a uniaxial state of stress, or

ff = Ee (5-23)

where £, the proportionality constant, is known as Young’s modulus, the elastic

modulus, or the modulus of elasticity The reader will recall that this important
material constant was used in the two-dimensional truss model developed in Chapter
3. Note that since the strain e is dimensionless, E has units of newtons per square
meter (N/m 2 ) or pounds per square inch (psi), which are the same as the units of
stress.
Equation (5-23) is more appropriately referred to as a constitutive relationship
as opposed to a law. A lan in engineering typically holds for all conditions and
under all Examples of laws of nature
circumstances (barring relativistic effects)
are the conservation of mass, the and second laws of thermodynamics, and
first

Newton’s laws of motion Examples of constitutive relationships are Fourier's law

of heat conduction, Newton's viscosity law. Pick’s law of diffusion, and Hooke’s
law. In some texts constitutive relationships are referred to as particular laws Note
that these relationships only hold for a certain class of materials operating in some
(imited range. However, without these re/ationships it would not 6c possible to
ascertain stress distributions, temperature distributions, velocity and pressure dis-
tributions, and so forth The reason for this is simple: constitutive relationships arc
used to bring the total number of equations used in a given model up to the total
number of unknowns.
For example, in three-dimensional stress analysis, there are fifteen unknowns
three displacements, six stresses, and six strains. With the constitutive relationship,

there are a total of fifteen equations, three equilibrium equations, six strain-dis-
placement equations, and six constitutive (stress-strain) relationships The com-
patibility equations are derived from the strain-displacement relations and, therefore,
are automatically satisfied if the other fifteen equations and boundary conditions
are satisfied. In other words, statically indeterminate problems
become solvable
 BASIC CONCEPTS IN ELASTICITY 199

largely because of the use of some relevant constitutive relationship that describes
the material behavior. The importance of constitutive relationships cannot be over-
emphasized.
Hooke's law in the form of Eq. (5-23) is of rather limited use. A much more
general form is given by [61

= D(e - e0 ) - cr 0 (5-24)

where D is referred to as the material property matrix. <x the stress vector, e the
strain vector, eo the self-strain vector, and tr0 the initial (or residual) stress vector.
In three dimensions. D is a 6 x 6 matrix and each of the remaining column vectors
is of size 6x1. Specific D matrices are given in Chapter 7 for plane stress, plane
strain, axisymmetric stress, and three-dimensional stress analyses. In three-dimen-
sional analysis, we define cr. e, <r 0 . and Eq as follows:

a = CT- 0"rv or Oyrf (5-25)

e = [e„ e_ (5-26)

£ II cT
n o\ v
O
cr_
“0 a/
ct (5-27)

H e_ r
iff W fcvx
0 “"0
e
W si (5-28)

We have already reviewed the stress and strain components contained in a and e.
However. cr0 and %
deserve comment.
The initial stress vector <rQ represents prestresses that are known to exist in a
material before it is loaded. The finite element method (or any method) cannot
predict these. They must be specified by the analyst.
The self-strain vector may be a result of crystal growth, shrinkage, or. most
commonly, temperature changes. Various forms of €q are presented in Chapter 7
for the different types of stress analysis. The material property' matrix D is symmetric
for both isotropic and anisotropic materials. In the isotropic case, the material
property matrix will be seen to contain only two independent material properties
(such as Young's modulus E and Poisson's ratio p).

Surface Tractions

A surface traction, or simply traction, is defined to be a distnbuted external or

surface loading per unit area. In this text, the surface traction vector is denoted by
s. which may be written in vector form in three dimensions as
r
s = [s r 5, 5. )
(5-29)

where 5 ,. s,_. and s : are the forces per unit area in the x. y. and r directions,
respectively. In two-dimensional applications, we simply take s. = 0.An example
of a surface traction is the hydrostatic pressure (a force per unit area) on the water-
filled sideof a dam. Note that in Eq. (5-29) only a single subscript is used to denote
the component of the traction.
200 GENERAL APPROACH TO STRUCTURAL ANALYSIS

For the two-dimensional situation depicted in Fig. 5- 10(a), note that s, = tr ()

and j, = cr n because the plane over which the surface traction acts happens to be
parallel to the v axis In a similar fashion, in Fig 5-10(b), we have = cr,, and
s, = tr,i. On nonplanar surfaces or surfaces that are not parallel to one of the
coordinate axes, no such simple relationships exist between the surface tractions
and the stresses acting on the surface.

5-3 PRINCIPLE OF MINIMUM POTENTIAL ENERGY

The principle of minimum total potential energy |7] may be stated as follows: out
of all the possible displacement fields that satisfy the geometric boundary conditions
(l e , the prescribed displacements), the one that also satisfies the equations of static
equilibrium results in the minimum total potential energy of the structure (or body)
This has been proved rigorously by Sokolnikoff |8| The total potential energy II

is defined here to be the sum of two diflercnt types of potential energy, that
associated with the internal potential energy U, (i e , the so-called strain energy)
and that associated with the external potential energy Uc from the external forces
that act on the system, or

It = U, + U, (5-30)

But for conservative force systems the loss in the external potential energy during
the loading process must be equal to the work done. IV,. on the system by the
external forces, or -Ur = +1V,. and Eq (5-30) becomes

U = U, - IV, {5-311

The principle of minimum total potential energy requires that II be a minimum for

stable equilibrium Furthermore, since II is a (unction of functions (the strain and

displacement functions), it is a functional The minimization of the total potential

Figure 5-10 Relationship between surface tractions and stresses on surface of two-
to
dimensional body for two special cases (a) surface parallel to i axis, (b) surface parallel

x axis.
 \

PRINCIPLE OF MINIMUM POTENTIAL ENERGY 201

energy is a problem in variational calculus. In order to minimize II, we insist that

the first variation of the total potential energy be zero, that is,

§n = W, - 8WC = 0 (5-32)

or

8U, = 8U', (5-33)

For convenience, a cartesian reference frame is assumed with \, v, and : axes. The
term 8{7, represents the first variation in the strain energy (as a result of a variation
in the strain) or

5 U, — /v'(o’ u 8e„ + cr„ 8e„ + cr.-Se.-

+ cr T , 8e„ + cr,-8e,. + a. x 8e /x ) dV (5-34a)

Equation (5-34a) can be written much more concisely using the matrix notation
from Sec. 5-2:

8U, = f v (8e) T adV (5-34b)

where 8e represents the first variation of the strain vector. The integral in Eqs. (5-
34)is said to be a volume integral because the integration is to be performed over

the volume of the body or structure being analyzed.

In considering the variation in the work done, 8U',,, by the external forces, we
must allow for the following possibilities: a body force b (per unit volume), a
surface traction s, and up to A' point loads fp The first variation of the work done
.

by these external forces (as a result of a variation in the displacements) is given

by

5'E<. = / v(6 t 8h + b 8v + b- Sir) dV + f$(s x 8 it + j, Sr

iV

+ j. Sir) dS + 2 (/p< 8n + f 5v + f Sir) |5-35a)

P= 1

where the first term represents a volume integral and the second term a surface
integral. As the name suggests, a surface integral must be evaluated only on the
surfaceS of the body or structure being analyzed. With the help of matrix notation,
Eq. (5-35a) may be written as
N
SU^ = J\/(Su) b dV + fs r
m)TsdS + 2 r
(8u) fp (5-35b)
p= 1

where the body force vector b is given by

T
b = [b x bx b._ \ 15-36)

the surface traction vector s by

'

s = [s x sx X-]
7
( 5 - 37 )
202 GENERAL APPROACH TO STRUCTURAL ANALYSIS

and the point load sector f; ,

by

fp = \fPK fr fP T (5-38)

Obviously the vector (8u) r represents the fnst variation in the displacement vector,
or

(8u) r = |
Sir 8v 6ir| (5-39)

where u, v, and w are the displacements in the v, v, and c directions, respectively.

With the help of Eqs. (5-34) and (5-35), we may write Eq (5-33) as

N
SvIStVtrdV = jV(&u T )l> dV + J s (»u)’sdS + 2 (5u)% 1540)
/»“ I

Let us now assume a linear elastic material jnd eliminate a by invoking the
constitutive (or stress-strain) relationship given by Eq (5-24) and rearranging the
result (sec Problem 5-18) to get

/ytSel'De dV = JV(8e)'De 0 dV - JV(8e>'«r0 <IV

N
T
+ J V'(8u) b (IV + /j(8u) 7
s c/5 + £ (8u )% (5-41)
P= l

This equation looks formidable but the reader need not despair This result will be
the starting point in Sec 5-7. where it is used to obtain the finite element char-
acteristics for essentially all problems in static, linear elasticity

In Problem 5-19 the reader is asked to show that the total potential energy TI

must be given by

II = kiJVe'Der/l' - JVt'De,, </V + f v e'v {

>clV

N
- Jyu'bdV - J s u sdS - r
£ u r f,, (5-42)
P- l

The first three terms represent the strain energy as a result ot the strains, the self-
and the prestresses, respectively The last three terms represent the work
strains,
done on the structure by the body torccs. the surface tractions, and the point loads,
respectively

Example 5-2

Use the principle of minimum total potential energy to determine the relationship
between the clongatton A and an axial force P in a uniaxial stress member of
uniform ccoss-secuonal area A. length L. and modulus of elasticity E One end of
the bar is restrained and the load P is applied to the free end. as shown in Fig
5-11.
 PRINCIPLE OF MINIMUM POTENTIAL ENERGY
203

Figure 5-11 Uniaxial stress member of length L in (a) unloaded condition and (b) loaded
with axial force P.

Solution

Although it is possible to start with Eq. (5-41), it is easier in this example to use

Eq. (5-42) directly by taking

eo = 0 Oo = 0 b — 0 .9 = 0

or

n = '/zEe-AL - PA

since D is simply E for the state of uniaxial stress. The first term in the expression
for FI represents the strain energy and the second, the work done by the point load
P in moving a distance A. Noting that c = A/L gives

Since we want to determine the elongation A for equilibrium and hence the A that

minimizes fl. we compute dUklA, set the result to zero, solve for A, and get
 1

204 GENERAL APPROACH TO STRUCTURAL ANALYSIS

which is a very well-known result. In fact, this result was used in the truss for-

mulation in Chapter 3 where the direct approach was illustrated. In Problem 5-20,
the reader is asked to show that the potential energy has in fact been minimized
by examining the sign on d 2 fl/r/A 2 .

5-4 PRINCIPLE OF VIRTUAL DISPLACEMENTS

The is actually a special case of the more general

principle of virtual displacements
principle of virtual work. Recall from elementary mechanics that u ork is the product
of a displacement and the component of the force in the direction of the displacement.
Virtual work is imagined to occur in one of two ways: ( l ) when the forces are real
and the displacements arc virtual (imagined) or (2) when the displacements arc real
and the forces arc virtual The principle of \irtuat displacements applies to the
former, and the principle of irnial fore es to the latter One statement of the principle
\

of virtual displacements is the following If the work done by the external forces
on a structural system is equal to the increase in strain energy of the system for
any set of admissible virtual displacements, then the system is in equilibrium. By
admtssable virtual displacements we mean those that satisfy the prescribed dis-
placement boundary conditions but are otherwise arbitrary Before actually devel-
oping this principle, let us illustrate it in a simple example

Example 5-3

Use the principle of virtual displacements to derive an expression for the elongation
A in a uniaxial stress member of constant cross-sectional area A, length L, and
clastic modulus £ if a load P is applied to the free end while the other end is

completely restrained as shown in Fig 5-1

Solution

From Fig. 5-11 it is seen that the work done by the force P is £A Note that this

is not given by £A/2 because the load P is assumed to be at its full value during
the loading (and elongation) process. Note further that the virtual displacement is

entirely arbitrary purposely (and conveniently) taken to be equal to the actual

and is

elongation A' Similarly, the increase in the strain energy is given by oeV, or

creV = (Ee)tAL = AEL

where € = A!L has been used The principle ot virtual displacements states that
 PRINCIPLE OF VIRTUAL DISPLACEMENTS 205

whereupon solving for A, we get

which again is the desired and well-known result. As mentioned in Example 5-2,
this result was the basis for the direct approach illustrated by the two-dimensional
truss model in Chapter 3. jg

Let us now develop the appropriate mathematical form of the principle of

virtual displacements. For reasons that will become evident at the end of this section,
letus denote the virtual displacements in the x, y, and z directions as bit, Sv, and
8w, respectively. It should be emphasized that bu does not represent the variation
in it but instead is the virtual displacement as defined above. We must recognize
that the virtual displacements will cause virtual straining, with the virtual strains
to be denoted as 8e tv 8e„ , , etc. Let us further consider the external loads as a result
of body force b, surface tractions s, and N point loads f , defined as
;

b = [b x />, b: ] T (5-36)

r
s = [j t x.] (5-37)

f, = f/„. fp. fr f (5-38)

The principle of virtual displacements may now be stated mathematically [9] as

/y(o-„ 8e„ + cr„ 8e„. + gt.. 5e_. + cr v , 8e n + o\. 8e, ; + o. t 8e_. t ) dV

= fy(br bit + b, Sv + b. Sic) dV + f^(s { 8 it + j, 8v + sz 8u') dS
A'

+ 2 (fPy bit + fpy 8v + fp, 8vv)

P= 1

or, more concisely, in matrix form as

N
f v (b€)
r
cr dV = Jy(8u) b dV + /S (8u) rs dS +
r
2 (8u) f;
r
,
(5-43)
P= 1

It is seen that this is identical to the result given by Eq. (5-40), which was obtained
by the principle of minimum potential energy. Equation (5-43) [or (5-40)] is known
as the weak form of the equilibrium equations, because the equilibrium equations
are seen to contain second derivatives of the displacements (after Hooke’s law is

invoked), whereas Eq. (5-43) [or (5-40)] contains only first derivatives. Further-

more, these two equations are valid for nonlinear as well as linear stress-strain

(constitutive) relationships. Obviously, if Eq. (5-43) is applied to a linear elastic

material with the constitutive relationship given by Eq. (5-24), then Eq. (5-41)
results.
206 GENERAL APPROACH TO STRUCTURAL ANALYSIS

It is seen that the principles of minimum potential energy and of virtual dis-
placements give identical results if the virtual displacements (and strains) are as-
sociated with the variations in the displacements (and strains), and vice versa. The
reason for representing the first variation of each displacement in the same manner
as a virtual displacement should now be obvious. In effect, we have proven the
equivalence of the principles of minimum potential energy and of virtual displace-
ments.

5-5 THE FINITE ELEMENT BASIS

In Secs. 5-3and 5-4 the principles of minimum total potential energy and of virtual
displacements were used to develop a very important relationship among the internal
stresses (and strains) and the external loads. Both approaches gave virtually the
same result. In any event, volume and surface integrals arose that looked rather
formidable. However, by discretizing the region to be analyzed into a number of
suitable finite elements, as shown in Fig 5-12, the integrations may be performed
with relative ease, for example,

v
r
/,.(8el a dV = 2 Jy.lS tfadV (5^14)
t = I

Figure 5-12 Discretization in one. two. and three dimensions with (a) the two-node lineal
clement
element, (b) the three-node triangular element, and (c) the four-node tetrahedral
 THE FINITE ELEMENT BASIS 207

/5 (8u)
r
s dS = 2 /s ,(8ii) rs dS (5-45)
e— I

where V e and S r serve as a reminder that the integrations are to be performed over
the element volume and element boundary, respectively. Note that the results from
the M nonoverlapping elements are. in effect, added together to obtain the integrals
over the original body or region. As we have seen several times before, the sum-
mation is not usually carried in the mathematical manipulations because we simply
consider it to represent the assemblage step. If Eqs. (5-40) and (5-43) hold over
the entire body or region and the region is arbitrarily selected by the analyst (such
as the selection of a free body), then these equations must also hold on an element
basis, or

r o- d\' = r
JV(5e) JV<-(8u) b d\
r
+ fS c( 8u) rs dS + S(8u) r f ;,
(5-46)

where fp must be interpreted to be the point loads acting at each node of the element
in addition to any other imposed loads acting within the element. The notations V 1
’

and S e represent the element volume and element boundary' or surface, respectively.
On internal nodes where no external loading is to be applied, fp from one element
will cancel (during the assemblage step) that from the adjacent element(s) that
shares the same nodes. These same comments hold for the surface traction term
(see Sec. 5-8). In other words, although the integral over Se is nonzero in general
on the element boundaries for element e. the contribution to the assemblage nodal
force vector will be seen to be effectively zero. The reason for this is that the

contribution from the neighboring element will always be of the same magnitude
but of the opposite sign (by Newton's third law) unless an external load is explicitly

applied there.
Returning to Fig. 5-12, it is seen that the two-node lineal element may be used
effectively in one-dimensional problems, the three-node triangular element in two-
dimensional (and axisymmetric) problems, and the four-node tetrahedral element
in three-dimensional problems. In the next section, the shape function matrix N is

developed in terms of the shape functions ;V,. A ,, 7

etc., associated with each of the
nodes.

5-6 THE SHAPE FUNCTION MATRIX

Recall from Chapter 4 that the field variable in the finite element method is routinely

represented by an interpolating polynomial on a piecewise continuous basis. Let

us restrict the present discussion to applications that require only C°-continuity such
as plane stress, plane strain, axisymmetric stress, and three-dimensional stress (but
not beam or plate bending or shell analysis). The field variables in the stiffness-
based finite element method are always the displacements it. v. and w.
208 GENERAL APPROACH TO STRUCTURAL ANAL* SIS

Two-Node Lineal Element

It is instructive to review what happened in Chapter 4 where a scalar field \ariablc

was represented in one-dimension with the lineal element. If we now interpret this
scalar field \anable to be the displacement u in the case of one-dimensional stress
analysis* we may take the interpolating polynomial

ft = Ci + <st (5-471

to represent this displacement within a typical element. In fact. Eq (5-47) is referred

to as a displacement function in particular and as a parameter function in general.
If we insist that at node i we have «(t,) = u, and at node j . u(Xj) = uJt then we
can find N,{x) and Nt (x) such that

« = N,{x)u, + A/ tty (5-481

where u, and u, are the displacements at nodes / and j, respectively. In Chapter 4

we found it convenient to represent this last result in the matrix form

ii = Na' (5-49)

where N. the shape function matrix, is defined as

N = (N,(x) N y
(x)J (5-50)

and a r , the vector of nodal unknowns, as

r
a' = [«, «,) (5-51)

Because the assumed displacement function is linear in t. the shape functions arc
also linear in v Moreover, they must satisfy the following conditions

N,{x,) = I /V,(t,) = 0
(5-52)
N,U,) = 0 Nfx,) = I

Three-Node Triangular Element

In a similar fashion, it seems reasonable to represent the x component of displace-

ment in two-dimensional (and axisymmetric) problems as

u - c, + c2 r + Cy\ (5-53)

if the three-node triangular element from Fig. 5-12 used Note that the triangle is

has three nodes and, not accidentally, the parameter function for the displacement
(i.e., the displacement function) h3s three constants Therefore, three shape func-

tions should be expected for this element, or

u = A',(r,\)«, + Nj{x.\)ttj + Aj(t,v)i/* (5-54)

210 GENERAL APPROACH TO STRUCTURAL ANALYSIS

Figure 5-13 Plot of shape function for node i for the three-node triangular element (with
C^-continuity)

Four-Node Tetrahedral Element

Finally, it seems reasonable to represent the x component of displacement in three-

dimensional problems by

u — f| + or + + t4r (5-62)

if the four-node tetrahedral element from Fig 5-12 is used Note now that the

tetrahedral element has four nodes and the assumed displacement function has four
constants. Therefore, four shape functions should be expected for this element, and
the displacement function for the x component may be written as

n = N,(x,\,:)u, + + Nkix,\.:)ut + NJx,\,z)um (5-63)

where u,, ur uk and nm

, x components of the nodal displacements, and N„
are the
Nr Nk , and Nm are the four shape functions Again we insist that «(r,,\„r,) =
tilxj.Vj.Zj) = u and so forth It should be obvious by now that we must ha\e
r
W*. 0 Nk!x , 0 = 0

N {x ,yr z,) =
t f
0 Nfix,.\ r :,) = 1 N {xr yr z,)
L = 0 NJ r \r :,)
= 0 (5-64)

N,{xk .y t .ik ) « 0 Nfixt.xt.ztt = 0 = I

= 0

AK'«A)-0 Sj(xm .y m .zm ) = 0 Ni(xm ,y M ,:m ) = 0 NJxm ,\ m .:„) = I

In a similar fashion, if we assume the displacement functions for the y and :

components of displacement to be

v = ts + t<rr + < 7s + (% z
(5-65)

and
if = r* + ( i„r + c ,, y + c i2 : (5-66)
 THE SHAPE FUNCTION MATRIX 211

we can in turn use the same shape functions to write these as

v = N,(x,y,z)v, + Nj(x,y,z)Vj + N {x,y,z)v + Nm(x,y,z)vm

k k (5-67)

and

u> = N,(x,y,z)w, + Nj(x,y,z)Wj + N (x,y,z)w +

k k NJx,y,z)wm (5-68)

where v,, y,, vt , and v,„ are the v components of the nodal displacements, etc. If
we now define u and a 1'
as

u = [u v wf (5-69)

and
ae = [h, v, iv, i
Uj Vj Wj i uk n wk I
v,„ wm ]
T
(5-70)

we may write

u = Na f (5-71)

where
"
N, 0 0 !
nj 0 0 i
Nk 0 0 \
Nn 0 0
N = 0 N, 0 !
0 Nj o !
0 0 !
o Nm 0 (5-72)
0 0 N -
!
0 0 Nj\ 0 0 Nk !
o 0 Nm _
Now the shape function matrix is seen to be of size 3 x 12, because each node
has three degrees of freedom and the element has four nodes. A typical shape
function, e.g., N, will be seen in Chapter 6 to vary linearly from unity at the node
in question (node /) to zero at each of the remaining nodes.

Properties of C°-Continuous Shape Functions

For the one-, two-, and three-dimensional elements in Fig. 5-12, the shape functions
satisfy the following three properties:

1. A shape function associated with a particular node (e.g., node /) must evaluate
to unity if evaluated at the coordinates of this node (node i); all other shape
functions must evaluate to zero at this same node. Actually, this property is

contained in the next property.

2. A shape function associated with a particular node (e.g., node /) varies linearly
from unity at the node in question (node /) to zero at each of the remaining

nodes.
3. The sum of all the shape functions at any point within the element or on the
element boundaries must sum to unity, or

2/Vp = 1 (5-73)

where the element is assumed to have n nodes and (3 takes on i, j, k, etc., as

appropriate. In this text, whenever a shape function for an arbitrary node is
referenced, the subscript (3 is used and the shape function is denoted as p N .
212 GtNtRAL APPROACH TO STRUCTURAL ANALYSIS

The shape functions for the lineal element were derived in Chapter 4; in terms
of the nomenclature in this chapter they are

(5-74a)

and

A',(r) = (5-74b)

The shape functions for the three-node triangular and four-node tetrahedral dements
are derived m Chapter 6.
Example 5-4
Verify that the shape functions for the one-dimensional, two-node lineal element
given by Eqs (5-74) have the three properties given above

Solution

we have

N,U,) - = I Nj{ x,)

Xj ~ X,

X. ~ X,
N,(t,) = = 1

x. - X,

which venfies the first property In fact, these conditions were summarized earlier

in Eq (5-52)
Next, note that N,i t) is unity at x = x, (i e . at node i) and decreases linearly

to zero at x = x (i.e
} . at node j). Similarly, note that N,(x) is unity at \ = x, (t c.,

at node j) and decreases linearly to zero at t = r, (i.e . at node 0 Thus the second
property is verified
Finally, note that

T; - r i - x, x - :
X + x — X,
N, + Nj = I

x, - X, x, - X, Xj - X,

for all x. Actually, because the shape functions apply to an element that connects
node i at x ~ x, to nodey at r = Xj , we should say that the above condition strictly

holds for x, StSt, (assuming t, is less than .t ).

/
Thus the third and final property

is verified. ®
Throughout this text, the shape function matrix is referred to quite often B)
gaining an appreciation of where the shape functions themselves come from, it
is

hoped that the development tn the next section will be more readily understood
now be
The next section picks up where we left off with Eq (5-46), which will
 THE FINITE ELEMENT CHARACTERISTICS 213

the starting point for the derivation of the finite element characteristics for all
problems in static stress analysis. The reader may find it helpful to reread Sec.
5-2. with special attention given to the definitions of cr, e, <r 0 , e 0 and L.
,

5-7 THE FINITE ELEMENT CHARACTERISTICS

We are now in a position to formulate the finite element characteristics for all

problems in static, linear stress analysis. Equation (5-46) provides the starting point
and is restated here for convenience:

(Su) r b r
7
Jy, (8e) tr dV = f ye dV + fs , (5u) s dS + 2 (8u)% (5-46)

Recall that this equation was derived in two seemingly different, but completely
equivalent, ways: first from the principle of minimum total potential energy and
then from the principle of virtual displacements. In what follows, the symbol 8u
may represent either the variation on the displacements or the virtual displacements,
depending on the reader’s preference. Similar comments apply to Se and 8a‘\
Let us begin by relating the strain vector e to the vector of nodal unknowns
ae (i.e., the nodal displacements) by using Eqs. (5-16) and (5-60) [or (5-71)], or

e = Lu = LNa c (5-75)

Let us refer to the matrix LN as the strain-nodal displacement matrix and denote
it by B, or

B = LN (5-76)

Therefore, Eq. (5-75) may be written as

e = Ba‘' (5-77)

Recall from Sec. 5-2 that L is a linear operator matrix that contains first derivatives

with respect to only a and y in problems requiring only C°-continuity. If the shape
function matrix N contains only linear functions (in „v and v) such as those in Sec.
5-6, then the matrix B contains only constants. In any event, B does not contain
any of the nodal displacements, which implies that

Se = S(Ba‘) = B Sa
1,
(5-78)

and

= fB 7
(Se)
7
(B 8a<y = (8a‘') (5-79)

Again depending on the reader’s preference, 8a may denote

1'
either the variation

in the nodal displacements or the virtual nodal displacements. Similarly, we have

Su = 8(Na‘ )
= N ba c (5-80)

and

(8u)
r = (N 8a‘ )
r = (Sa‘
,

)
r Nr (5-81)
214 GENERAL APPROACH TO STRUCTURAL ANALYSIS

Substituting the results from Eqs. (5-79) anti (5-81) into Eq. (5-46), after transposing
all the terms to the left-hand side and pulling the ( ha r ) T through the integrals and
the summation, yields

(5a')
r r
{fv B <J dV - JV, N T b dV - fs N r s , dS - 2 N%} = 0 (5-82)

But the term in braces must be zero because (8a ) r t-

is not necessarily zero and is

quite arbitrary. If the principle of minimum potential energy path is followed, then
Sa' represents the first variation of nodal displacements. This variation is quite
arbitrary, and so the term in braces must be zero. If the principle of virtual dis-
r
placements is the basis for Eq. (5-46), then 5a is arbitrary because it represents
arbitrary virtual displacements Actually, if the statements of these principles from
Secs. 5*3 and 5-4 are read carefully, the variations in the displacements and the
virtual displacements are not completely arbitrary because they must satisfy the
prescribed displacement boundary conditions. In any event, 8a' is nonzero, in
general, and again it is concluded that the term in braces is zero, or

fv B r <r dV - f v, N r b JV - Js N r s dS - , 2 N'T, - 0 (5-83)

Equation (5-83) is still quite general because a constitutive relationship has not
yet been invoked However, Jet us now consider only linear elastic materials for
which the constitutive (or stress-strain) relationship from Sec 5-2 applies, or

fJ
~ D(€ - c0) + <r 0 (5-24)

where the material property matrix D contains pertinent material properties such as
the elastic modulus and Poisson’s ratio Specific examples of this matrix as well
as the self-strain vector € 0 and the prestress vector a 0 are presented in Sec 5-8
and Chapter 7 With the help of Eqs (5-77) and (5-24), we may write Eq (5-83)
as

{ fv B r DB dv] a' - J v , B r Dc 0 dV + B T a0 dV
- fv N rb </!' - Jj, N'sfi - SN'f, » 0 15-84)

Note how the vector a r was pulled out of the integral (to the right) because it is
not a function of the spacial coordinates Finally we get an equation of the form

K'a' = P (5-85}

where

'•-t-C.+ H + C + lfL l5 861

’

and where we define the element stiffness matrix K by f

K' = fv B r DB dV 15-Sfl

and the five element nodal force vectors by

r;„ = f v . B r De 0 dV (5-881

C. = / B r <b) dV (5-831
 APPLICATION': TAPERED UNIAXIAL STRESS NUMBER 215

n = !v N r b dv (5-90)

= fs , N r
fj s dS (5-91)

and

fk = 2 N% (5-92)

For the uniaxial stress member, discretized with the two-node lineal element,
the matrix B is of size I x 2 because L is a scalar and N is 1 x 2 (recall that
B = LN). The matrix D is actually a scalar (or a I x 1 matrix). The element
stiffness matrix K*" is of size 2 x 2 as expected, since there are only two nodes
on each element and each node has only one degree of freedom. In a similar laslnon.
it may be shown that each of the five nodal force vectors is of size 2 x 1 (see
Problem 5-24).
Similarly, for problems in two-dimensional stress analysis, analyzed with the
three-node triangular element, the matrix B is of size 3x6 because L is 3 x 2
and N is 2 x 6. The matrix D is of size 3x3. The element stiffness matrix Kr
is 6 x 6 as expected, because there are three nodes per element and each node
has two degrees of freedom. Not surprisingly, each of the nodal force vectors must
be of size 6 x 1 (see Problem 5-25).
Finally, for three-dimensional problems in stress analysis analyzed with the
four-node tetrahedral element, it can be determined (see Problem 5-26) that K' and
f' are of sizes 12 x 12 and 12 x 1, respectively.
In the next section these rather abstract notions are illustrated with a problem
in one-dimensional stress analysis.

5-8 APPLICATION: TAPERED UNIAXIAL STRESS MEMBER

In this section the general expressions for the Finite element characteristics given
by Eqs. (5-87) to (5-92) are applied to a simple stress analysis problem — a tapered

uniaxial stress member analyzed with the two-node lineal element First the for-

mulation given for a general problem in this class and then it is illustrated
is

numerically m an example for a specific problem.

Consider the tapered uniaxial stress member shown in Fig 5- 14(a) Note that
the conditions on the ends of the member are quite arbitrary In other words, the
ends of the member max both be restrained (prescribed displacements), both loaded
(via surface tractions and'or point loads), or a combination of both
The member is discretized into a number of tvv o-node lineal elements as show n
in Fis 5- 14(b) with node i (at r =
that a node
v,f and node j (at v = v,) Note is

associated with a planar surface and not a single point The nodal displacement it,

occurs at \ - v,. while the nodal displacement it, occurs at a = x,

A typical element e is shown in Fig 5- 14(c) with point loads f, and f and or
surface tractions and v, acting at the nodes and w ith a body force /> = y as a
r,

result of eravitv, where y is the weight density of the material In addition, this
216 GCNI KAI. APPROACH TO SfRUCTURAL ANALYSIS

Liut restrained
or loaded

fcnd restrained
or loaded
(a) (b) <c»

Figure 5-14 General onc-Uimcnsional stress problem (a) uniaxial stress member, (b)

discretized into several two-node lineal elements, and (c) typical element with positive point
loads and tractions shown on nodal planes

element is assumed to undergo a temperature change A7" and to be prestressed to

a stress of <r„

The Element Characteristics

From Eq. (5-50), the shape 1 unction matrix N is ot size I x 2 where the shape

functions N,(x) and Nf{x) are given by Eqs (5-7*1) and N itself by

But the strain-nodal displacement matrix B is related to N through the linear operator
matrix L by B = LN, where L = (Mr. This follows from the fact that the strain-
displacement relationship for this class of problems is given by e = duULx and from
the definition of L, namely, e = Lu Therefore.

(5-94)
 APPLICATION': TAPERED UNTAXIAL STRESS NUMBER 217

The material property matrix D is really a scalar here as explained below. The
stress-strain relationship for the state of uniaxial stress is given bv

cr = E(e - €,))
- cr0 {5-95)

Comparing Eqs. (5-95) and (5-24) yields D - E. Note that each term in Eq. (5-
95) is a scalar. This constitutive relationship is a somewhat more general form of
Hooke's law (a = Ee) because
includes the possibilities of both self-strains e 0
it

(e.g.. as a result of a temperature change) and prestresses or0 The self-strain e0 is .

related to the temperature change AT by

eo = a,A T (5-96)

where a. is the coefficient of thermal expansion, in units of inches per inch per
c
degree Fahrenheit (m..'in.- F) or meters per meter per degree Celsius (m/m-°C).
We are now in a position to evaluate the element stiffness matrix K e‘
by
beginning with Eq. (5-S7) and writing

K" = JV, BTDB d\ r

(5-977
£[-l 1] A(.t) dx

where dV — dx has been used and A is the cross-sectional area of the elsrecr
A(.r)

at a specified value of .r. The notation xf is used to represent xy — x, in a rrne

,

concise manner and really represents the element length. If the constants
~ -re-
integral are pulled through the integration, we get

One of the simplest ways to evaluate the remaining integral is to e-’aiure r-r:.
at x = x. where

_ ,r, + .v.

denote it as A. and treat A as a constant to get

1
-1
K' 1
I

The bar ( ) on the area A is dropped with the c nuirm." “

sectional area varies with x. a constant value may vl cr r — - "

value at .t = x. Equation (5-99) provides an erg?cr-,--

- — ... .-

matrix for the uniaxial stress member. Next v e t— . ^ -

218 CLNERAL AJ’I’ROACJ TO STRUCTURAL ANALYSIS
J

first determine the element nodal force vector as a result of the self-
Let us
strain thathappens to be caused by a temperature change A7\ With the help of
Eqs. (5-94), (5-96), and D - C, it can be shown from Eq, (5-88) that

r;„ = Sv n r D£„ civ

=
jr"-T [~il
* l5 ' 1001

= AEa, bT

In Eq. (5-100) it is assumed that a, and AT are constant in any one element. If

they vary as a function of x, they may simply be evaluated at x =x and treated

as constants any one element.
in In a similar fashion, it can be shown that (see
Problems 5-27 and 5-28)
_
R. = !v B r CT„ clV = 15-1011
[
J

and

t( = JV.N'Ix/V = 15-1021

where again o0 and y arc assumed to be constant in the element, or suitable average
values (i.e , evaluated at x = x) arc used The nodal force vector as a result of
the surface tractions \, and s, may be determined from Eq. (5-91) as follows

r/ = IS f N r s ds
= + N'(r; ) fAl s, dA
( 5 - 103 )

where A, and A, are the cross-sectional areas at nodes / and j, respectively Finally,
the nodal force vector f|fL as a result of the point loads at nodes and j is determined
/

from Eq (5-92) to be

fp L = 2 N% = + N r
(r,)/y

( 5 - 104 )

Note that if a point load ft, is imposed within the element, c g . at x ~ r then

f,f
L is evaluated as follows

ffl = lN'lV„>y„ jjij

+

For example, if xr — t, then half of fp is allocated to node i and half to node j

(see Problem 5-29).
 APPLICATION: TAPERED UNIAXIAL STRESS NUMBER 219

This completes the evaluation of the element characteristics for the two-node
linealelement in one-dimensional stress analysis. It should be emphasized that
s„
Sj,f, and fj, are positive if directed in the positive direction.

The Element Resultants

The element stiffness matrices and nodal force vectors may be determined for every
element with the help of Eqs. (5-99) to (5-104). The assemblage of each of these
2x2 matrices and 2 x vectors is performed in the usual manner to give a
1

system of N
linear algebraic equations in (V unknown nodal displacements assuming
the member is discretized into N - elements with N nodes. The result is K“a = f“.
1

The geometric (or prescribed displacement) boundary conditions are applied at this
point to yield Ka =
which may be solved for the nodal displacements contained
f,

in the vector a. Therefore, in

what follows it is assumed that all the nodal dis-
placements are now known. Each element nodal displacement vector a is also 1'

assumed to be known.
The element strains and stresses may be computed from Eq. (5-77) as

e = Ba 1
(5-105)

and from Eq. (5-24) [or Eq. (5-95)) as

a = £[Ba f — a, AT] + tr0 (5-106)

Note and stress within the element are denoted as e and a,

that the axial strain
respectively, where the bars ( ) imply average strain and average stress. The
strains and stresses within an element are not necessarily constant. However, Eqs.

(5-105) and (5-106) can only result in one strain and one stress for each element
(both constant and not a function of x) because B is a constant matrix in this case.

Therefore, the strain e and stress <r are generally assumed to be the local values of
e and cr at .r = T. It follows that the average force F within an element is given
by F = crA.
The use of the equations developed in this section is illustrated numerically in

the next example.

Example 5-5

Determine the displacements, and stresses within the tapered circular, uni-
strains,

axial stress member loaded as shown in Fig. 5-1 5(a), which also shows the di-
mensions of the rod. The rod is fabricated from carbon steel with a modulus of
6
elasticity of 30 x 10 psi (21 x 10 10 N/m 2 ).

Solution

The first step in any finite element analysis is discretization of (he region under
consideration into a suitablenumber of finite elements. For the purpose of illustrating
the basic approach, let us take only two elements and three nodes as shown in Fig.
5-I5(b). It is convenient to summarize the node and element data in a tabular
form
220 GENERAL APPROACH TO STRUCTURAL ANALYSIS

Figure 5*15 Uniaxial stress member in Example 5-5 (a) Dimensions and loads and (b)

discretization into two elements and three nodes

as shown in Table 5-1 Note that the radius of the rod is supplied as part of the

node data In a computer program, it is somewhat more expedient to include the

cross-sectional area in the material property data This was in fact done in the

TRUSS program The variable area could be taken into account with the help of a
variable property routine (see Problem 5-39)

Table 5-1 Node and Element Data lor Example 5-5

Node
number r, in r, in

1 00 0 500
2 60 0 375
3 12 0 0 250

Element Node Node

number '
J

1 1 2

2 2 3
 APPLICATION: TAPERED UNIAXIAL STRESS NUMBER 221

The radius of the bar at x = x is denoted as r and is easily computed from

_ _ l D\Lh - (D, - D,)x

r [
— L L

where D (
is the diameter of the rod at the base. D z is the diameter at the tip. and
Lb is the length of the bar.
The calculations are summanzed below in a form that may be readily imple-
mented in a computer program.

Element 1

Node i is 1 . Node j is 2.

x, — 0.0 in. xt = 6.0 in.

xr — xt — x, = 6 — 0 = 6.0 in.

E = 30 x 10
6
psi

.1
= 3.0 in.

1 (1.0 - 0-5K3.0)
r 0.4375 in.
12.0

A ~r 2 = (3. 14)(0.4375)
: = 0.601 in.
:

(0.601)(50 x 10'’) I
-1
it
K’
6.0 -I 1

3007 -3007 3
x 10 Ibf/in.
— 3007 3007

3007 -3007 0
3
K" -3007 3007 0 x !0 Ibf/in.

0 0 0

l I _ 0
Ibf
0

0
f° = 0 Ibf

Element 2
Node / is 2. Node j is 3.

,r, = 6.0 in. xf = 12.0 in.

Xy = 12.0 - 6.0 = 6.0 in.

222 GFNERAL APPROACI TO STRUCTUR AL ANALYSIS
I

= 30 x 10 6 psi

6+12 =
9 0 in
2

- 0
:
1 I"
_ (10 5 )(9 0)
= 0.3
2 [ 12.0

A = (3 14)(0 3125)
2 = 0 307 in

A, ir(0 25 )
2 = 0 196 in
2

X 6
(0 307)(30 IQ
K (2> = )

:]
60 [-:
1

1534 - 1534 3
] X 10
I0 lb
lbf/in
-1534 I534J
-3007
3007 °1
K" = * -3007 4541 - 534
1534 x 10 3 lbf/in
0 -1534 1 534
534j I

f
<2;
= f&' +
0 0 ^ 0
[
3000 [(0 196K 10,200) [5000
0
0
5000

Before the application of the prescribed displacement we have K°a = f". or

3007 -3007 0 «i 'o'

— 3007 4541 -1534 «2 - 0
0 -1534 1534. _«3_ .5

where both sides have been divided by 10 3 Note that the assemblage stiffness
matrix is symmetric and banded, with a half-bandwidth of two [this also follows

from Eq (3-33)]. If the prescribed displacement boundary condition is applied by

using Method 1 from Sec 3-2. we get
1 °i
’l 0 0
0 4541 -1534 _ 0
"2
.0 - 1534 1534. Jb _ L5J

Solving this system of linear, algebraic equations (c g., by the matrix inversion
method) yields the following nodal displacements

«, = 0.0 in. it2 = 0 00166 in. uy = 0 00492 in.

 APPLICATION': TAPERED UNIAXIAL STRESS NUMBER 223

The element resultants are now easily determined as summarized below:

Element 1

B = iq-i 1] in.
1

e = Ba'" % -
[ 1
0.00166
= 0.0002772 in. /in.

= x I0 5 )(0.0002772) =
cr £(e - ^o) - / 0 = (30 8320 psi

F - UA = (8320X0.601) = 5000 ibf

Element 2
-1 1

B = _ 1] in.
1

12 - 6 12 -

= = 0.00166
e Ba' 2 ’
>/o [-1 1
]
0.00492
= 0.0005-133 in. /in.

= £e = (30 x 10 6 )(0.0005433) = 16.300

5? psi

£ = m I = (16.300X0.307) = 5000 Ibf

A few subtle points regarding the calculation of the element nodal force vectors
in Example 5-5 need to be discussed. The reader may observe that on an individual
element there are surface tractions (or point loads) acting at the nodes as a result
of the forces exerted by the adjacent elements or by the restrained end. For example,
let us consider a bar discretized into three elements as shown in Fig. 5-16. It should
be apparent that the internal surface tractions and'or point loads cancel during the
assemblage step. Moreover, if a node is restrained (e.g., node 1), then there is in

effect an unknown force at this node. The reader should be convinced, however,
that this unknown force never really enters into the formulation because of the
application of the prescribed displacement at this node. This same observation was
made Chapter 3 where the two-dimensional truss model was developed (and in
in

Chapter 4 where the pin fin was modeled). The second point to be made is that
the stress distribution near the point of application of point loads (and nonuniform
surface tractions) is never attained in a one-dimensional formulation, no matter how-
many elements are used. Predicting such stress distributions requires at least a two-
dimensional analysis (see Chapter 7).

The from Example 5-5 are compared to the exact solution (see Problem
results
5-37) in Table 5-2. where the finite element solutions for four and eight elements
are also given. These FE.M results were obtained by the computer program described
in Problem 5-39. Stresses are reported with live significant digits in order to show
224 GbNfcRAL APPROACH TO STRUCTURAL ANALYSIS

Figure 5-16 Graphical explanation of the cancellation of (a) internal surface tractions and
(b) internal point forces that occurs during the assemblage step

the remarkable accuracy of the results. The nodal displacements Table 5-2 m
demonstrate convergence to the exact solution as the number of elements is in-
creased —
here the maximum error decreases from -3 3% for two elements to less

than —0.3% for eight elements

A detailed discussion of error predictions and convergence is beyond the scope
of this text. The interested reader may find introductory material on this important
subject in the book by Becker, Carey, and Oden 1 10].
 L ™
REMARKS
Table 5-2 225

5_ 5

REMARKS

£»* ,„ piK from lhc

based finite IT P Seen t0

Equivalent °
? ctura l analysis
prob-
,
me a b°
°d "'herein tfl >’ ie,d ed the
Piacement T (
*e stiffness-
the equStaw"
and n 01
forc E
MoreovTLh
prini
T*
are the nod al
«fis-

appear * Z CqUat,0ns ,n that

Z t,ons (after ,he « most only fi^ 5*"?** in a weak
fom, of
implication
" ls ^^*
is that
lower-order
generalized form ofH^T^
displacement aVV S a
fuTcUnn
functions
P
^
° f thc d,s Jace

may
ments
'

be used
'
W>««». The
in the finite
226 GENERAL APPROACH TO STRUCTURAL ANALYSIS

element solution. For example, first-order displacement functions w ere used throughout
this chapter, whereas the equilibrium equations would require at least second-order
functions. In the stiffness-based finite element method, the compatibility equations
are satisfied exactly by these displacement functions (see Problems 5-49 and 5-50).
The shape function matrix was introduced for the three-node triangular and
four-node tetrahedral elements. Although the shape functions themselves were not
derived in this chapter, the basic properties that these functions must satisfy for
problems requiring only C°-continuity were presented. In the next chapter, explicit
expressions for these shape functions will be derived.
Following this, the finite element characteristics were derived for all problems
in linear, static stress or structural analysis It should be recalled that by finite
element characteristics we mean expressions for the element stiffness matrix Kr
and the element nodal force vectors V. These expressions are given by Eq. (5-87)
and Eqs. (5-88) to (5-92), respectively. Their importance cannot be overemphasized
because they may be used to formulate all problems in linear, static stress analysts

The general approach was illustrated with the simple application of a uniaxial
stress member. Admittedly, this example is of little practical importance, but it
should help to clarify some of the concepts developed in this chapter. When applied
to a numerical example, convergence to the exact solution was demonstrated as the
number of elements was increased The stresses from the finite element solution
also compared very favorably with those from the exact solution.
Chapter 6 is devoted to the derivations of the shape functions that we will need
in our study of two- and three-dimensional stress analysis in Chapter 7 It will be

seen in Chapter 8 that these same shape functions may be used in one-, two-, and
three-dimensional thermal analysis and fluid mechanics Indeed, they also may be
used in the analysis of problems in lubrication, electromagnetics, etc. (1 1]

REFERENCES

1. Hucbncr. K H . The Finite Element Method for Engineers. Wiley, New York, 1975,
pp 457-462
2. Fitzgerald. R W . Strength of Materials. Addison-Wcslcy. Reading. Mass . 1967
3. Timoshenko, S . and D H Young. Elements of Strength of Materials. 5th cd . Van
Nostrand, New York, 1968
4. Popov, E P . Introduction to Mechanics of Solids, Prcnticc-Hall, Englewood Cliffs.
N.J., J968
5. Popov, E P , Introduction to MiVnamcs of Solids, Prentice-Hall. Englewood Cliffs.

N J . I96S. pp 308-306
6. Zienkicwicz, O. C., The Finite Element Method, McGraw-Hill (UK). London, 1977.
p. 28.
A Treatise on the Mathematical Theory of Elasticity. Doscr, New
7. Lose, A. E. H.,
York, 1944.
S Mathematical Theory of Elasticity, 2nd cd , McGraw-Hill, New
8. Sokolnikoff, I. ,

York, 1956.
 PROBLEMS 227
9.

Ugural. A. C.
and S. K. Fenster, Advanced Strength and Applied Elasticity, American
.

Elsevier, New
York. 1975, pp. 315-316.
10. Becker, E. B., G. F. Carey, and J. T. Oden, Finite Elements: An Introduction, vol.
1,
Prentice-Hall, Englewood Cliffs, N.J., 1981, pp. 36-38.
11. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977
pp. 423-449.

PROBLEMS

Note: The properties in Appendix A should be used unless stated otherwise in the problem
statement.

5-1 With the help of Fig. 5-4, show that the equations of static equilibrium in two
dimensions are given by Eq. (5-4) by doing force balances in the a and y directions.
Show that Eq. (5-5) holds by doing a moment balance about the center of the
infinitesimal element.

'
5-2 Derive the three-dimensional form of the equilibrium equations given by Eqs. (5-6).
Show that Eq. (5-7) holds by doing a moment balance about the center of an infin-
itesimal cubic element.

5-3 Consider the two-dimensional state of stress shown in Fig. P5-3.

r 4

A 1500 N/cm 2

-* 2500 N/cm 2

2000 N/cm 2

Figure P5-3

a. Plot Mohr’s circle of stress.

b. Determine the principal stresses and show the corresponding element orientation.

c. Determine the maximum shear stress and the associated normal stress and element
orientation.
d. Use Eqs. (5-9) to (5-11) to corroborate the results from Mohr’s circle.

5-4 Consider the two-dimensional state of stress shown in Fig. P5-4.

a. Plot Mohr’s circle of stress.

b. Determine the principal stresses and show the corresponding element orientation.

c. Determine the maximum shear stress and the associated normal stress and element
orientation.
d. Use Eqs. (5-9) to (5-11) to corroborate the results from Mohr's circle.
228 GENERAL APPROACH TO STRUCTURAL ANALV SIS

Figure P5-4

5-5 Consider the two-dimensional state of stress shown in Fig P5-5

a. PlotMohr’s circle of stress

b. Determine the principal stresses and show the corresponding clement orientation
c. Determine the maximum shear stress and the associated normal stress and clement

orientation
d. With help of Eqs (5-12), verify the principal stresses from part (b) Assume that
a,., aw , and cr. arc zero What is the value of the third principal stress in this
case’’

Figure P5-5

5-6 Consider the two-dimensional state of stress shown in Fig P5-6

a. Plot Mohr's circle of stress

b. Determine the principal stresses and show the corresponding element orientation
c. Determine the maximum shear stress and the associated normal stress and element
orientation.

d. With help of Eqs (5-12), verify the principal stresses from part (b)
Assume that
(Tj., ot_. and a- are zero What is the value of the third principal stress in this
case?
 PROBLEMS 229

>

2000 N/cm 2

3500 N/cm 2

“SOOO n/cm 2

Figure P5-6

5-7 In Chapter 7, the state of plane stress is defined as that state of stress such that the
stresses ct,., cr,., and ar~ are identically zero (see Sec. 7-2).

a. What can you conclude about one of the principal stresses? Hint: Equations
(5-12) must be used.
b. Derive Eq. (5-10) by beginning with Eqs. (5-12) and using the fact that <r n ,
or,.,

and cr.. are zero.

5-8 What are the four strain-displacement or kinematic relationships for small defor-
mations in axisymmetric problems? it is not necessary to derive them. Instead, use
a suitable reference.

5-9 What is the linear operator matrix L in three-dimensional stress analysis? The matrix
L is defined by e = Lu, where e and u are given by Eqs. (5-20) and (5-21),
respectively.

5-10 Show that the compatibility equation in two dimensions is given by Eq. (5-22).

5-11 By using a suitable reference, state the six compatibility equations that must hold in
three dimensions. Why are there six relationships?

5-12 A portion of the boundary of a two-dimensional body to be analyzed is shown in

Fig. P5-12. On this boundary, a uniform surface traction acts as shown.

Figure P5-12
230 GENERAL APPROACH TO STRUCTURAL ANALYSIS

a. Determine s, and sy on this surface.

b. Suggest an application that could result in this type of surface traction

5-13 Part of the boundary of a two-dimensional body to be analyzed is shown in Fig.

P5-13. On this boundary, a uniform surface traction acts as shown
a. Determine s2 and sy on this surface.

b. Suggest an application that could result in this type of surface traction

Uniform surface
traction acting
on the global
boundary

Figure P5-13

5-14 Part of the boundary of a two-dimensional body to be analyzed is shown in Fig

P5-14 On boundary, a uniform surface traction acts as shown
this

a. Determine s, and sY on this surface

b. Suggest an application that could result in this type of surface traction

Uniform surface traction

acting on the global boundary

Figure P5-14

shown to
portion of the boundary of a two-dimensional body to be analyzed
is
5-15 A
Fis. P5-15 n 'h |S boundary, a uniform surface traction acts as shown.
O
 PROBLEMS 231

Figure P5-15

a. Determine and sx on this surface

b. Suggest an application that could result m this type of surface traction

5-16 A portion of the boundary of a two-dimensional body to be analyzed is shown in

Fig P5-16 On this boundary', a uniform surface traction acts as shown

Figure P5-16

a. Determine .v, and s, on this surface

b. Suggest an application that could result in this type of surface traction

5-17 Part of the boundary of a two-dimensional body to be analyzed is shown in Fig

P5-17 On this boundary, a uniform surface traction acts as shown

a. Determine j, and s% on this surface

b. Suggest an application that could result in this type of surface traction
 )

232 GENERAL APPROACH TO STRUCTURAL ANALYSIS

Figure P5-17

5-18 Starting with Eq (5-40) and using the constitutive relationship for a linear clastic
material, show that Eq (5-41) holds

5-19 Show that the total potential energy ot a structural system in static equilibrium is

given by Eq (5 42). given that the first variation 811 is given by bq (5 32) and that
Eqs (5-34b) and (5 35b) hold

5-20 Reconsider Example 5 2 and show that potential energy II has indeed been minimized
by computing ttUVeiA 1 Hint What is the significance of the sign of this result ’

5-21 Show the two shape functions graphically for the two-node lineal element tor C'"-
contmuity given that they arc linear and must satisfy Eq (5 52) Clearly label each
of the shape functions

5-22 For the three-node triangular element, show ihc shape functions N
,( i * ) and x. \

in a manner similar to that in Fig 5-13

5-23 Dense Eq (5-84) from Eq (5 83) with the help of Eqs (5-77) and (5-24)

5-24 Clearly show why each of the element nodal force vectors given by Eqs (5 88) to
(5-92) is of size 2 x I for the two-node lineal element

5-25 Show why each of the element nodal force vectors given by Eqs (5-88) to (5 92) is

of size 6 x I for the three-node triangular element In two-dimensional problems in

stress analysis, how many degrees of freedom are there per node What do the degrees ’

of freedom represent'’

5-26 For three-dimensional problems in stress analysis anal) zed with the four node tetrj-
hcdral element, show that the element stillness matrix and nodal force sectors given
by Eq (5-87) and Eqs (5-88) to (5 92) arc of sizes 12 x 12 and 12 x I. rcvpcc
lively . In three-dimensional problems in stress jnal>sis. how many degrees of freedom
-7
arc there per node * What do the degrees of freedom represent

5-27 Show that the element nodal lorce vector from a prestress in a uniaxial stress member
is given by the result in Lq (5-101) if the two-node lineal element is used What
7
assumptions have to be made in arriving at this result
 PROBLEMS 233

5-28 Show that the clement nodal force vector from a body force in a uniaxial stress
member is given by the result in Eq. (5-102) if the two-node lineal element is used.
What assumptions have to be made in arriving at this result?

5-29 Consider the one-dimensional element shown in Fig. P5-29. This element has been
extracted from a discretized brass bar in a state of uniaxial stress. The bar has a
circular cross section with a constant diameter D and undergoes a temperature change
AT. A load P is applied as shown at point p. A surface traction also acts as shown.
.v

If -V/ = 2 cm. xj = 5 cm, xp - 3.5 cm. D = cm, AT = + 15°C, P - 1300 N,

1

and s = 3000 N/cnr:

1

Figure P5-29

a. Determine the element stiffness matrix.

b. Determine the relevant element nodal force vectors. The weight may be neglected.
c. Determine the composite element nodal force vector.

5-30 Solve all parts of Problem 5-29 if .v, = 2 in...v, = 5 in.,.v,, = 3.5 in ..D = 0.5 in.,

AT = + 10°F, P - 1500 lbf, and r = 1800 Ibf/in.

2
.

5-31 Consider the one-dimensional element shown in Fig. P5-31. This element has been
extracted from a discretized bar and is in a state of uniaxial stress. The bar is made
of hard drawn copper and has a rectangular cross section with dimensions w and h.
Two point loads T, and P2 are applied as shown. A surface traction s also acts as
shown. The bar undergoes a temperature change AT and is initially under a prestress
ct0 If .r, = 3 in., x, = 5 in., xp = 4.5 in., w
.
= 0.75 in., h = 0.60 in., AT =
-20°F, P = 1000 lbf, P 2 - 200 lbf, x = 600 lbf/in. and cr0 = -400 Ibf/in.
2 2
:
,
|

Figure P5-31
234 GENERAL APPROACH TO STRLCTVRAL AN ALT "SIS

a. Determine the demert stiffness matrix

b. Determine the relevant element Dedal force sectors. The weight mas he reelected
c. Determine the composite clement nodal force sector.

5-32 Solse all parts of RxsKem 5-31

ifx, «= cm.x, = 3 cm. j, — 2J cm. w- =
I cm. U
h =I Ocro.AT = — 25*C.P, = SOON .Pz ~ ICO N. j = 600 N cm% and o, =

—350 N crir.

5-33 Consider the one-dimensional dement shown in Fig P5-33. This dement has been
extracted from a discretized circular bar and ts in a state of uniaxial stress The bar
is made of aluminum alios 6061 and i» tapered such that at nodes i andj the diameters
of the bar are D, and D,. re«pectisds Two
jsw-r loads P, and P- are applied as
shown A shown The bar undergoes a teraprrarere
surface traction i also acts as
dunce AT and ts trenails o . I f t, = 23 cm. r, = 20 cm. sf =
under a prestress
22 era. D. = 2 0 cm. D, = 1 50 cm. AT = -30'C. P, = 200 N. P: = 150 N.
j — 200 S cm', and «y„, = — 200 X err.

Figure P5-33

a. Determine the dement stiffness minx

b. Determine the relevant dement nodal toree sectors The weigh.! mas be neglected

C- Dcterm-ne the compos i*e clement ncslal force sector

5-34 Sols call parts of Problem 5-33 if r = 10 in s = 7 m . xf = 9 in D = 0 75 n

D, =0 50 in . AT = -dO F P = 500 Ibf. P =
;
200 IM. f = a00 Ifcf in . and
<Tj = — 700 Esfin.-

5-35 Consider the uniaxial stress member shown in Fig P5-35

a. Solve for the displacements and the demert resuhanis for D = cm D- =

I

0.75 cm. L>

- 6 cm F = 1500 N and j = 5000 N err Take £ = 21*10*
N nr. Use onls two elements
b. Compare the results from parr iai with the exact <o.’itoi gt\ c*s tn Problem ---' J
 PROBLEMS 235

Figure P5-35

5-36 Solve both parts of Problem 5-35 if £>, = 0.75 in.,D 2 = 0.25 in., Lb = 5 in.,
s = -8000 Ibf/in.
2
, E - II X 10 6 lbf/in. : . and F = 1500 Ibf.
5-37 Show that the exact solution for the displacement to the problem posed in Example
5-5 is given by

4 UP 1 1
_
it(.x) =
ttE(D\ - D 2) l ~ [(Di — Di)’ Li , jv Dj

where P is the sum of the point load and the force from the surface traction acting
on the tip of the rod.

5-38 The exact solution for the displacement at the tip of an untapered uniaxial stress
member is given by u = FLh /AE, where F is the axial force, Lh is the length of the
member, A is the cross-sectional area (assumed to be constant), and E is the elastic
modulus. Show that the exact solution for the displacements in a tapered uniaxial
stress member gives this same result if £>, approaches D 2 The exact . solution is given
in Problem 5-37.

5-39 Convert the TRUSS program in Appendix B into one that may be used to solve
problems in one-dimensional stress analysis. The program should allow for self-

strains (from a temperature change), prestresses, body forces, surface tractions, and
point loads. Assume no more than 30 elements (and 31 nodes). The following pa-

rameters should be taken as the material properties. E, A, a,, A T, o0 , and b. Allow

up to five different types of materials and make use of material set flags. Allow for
two different prescribed displacements (using positive boundary condition flags) and
236 GENERAL APPRO ACt! TO STRUCTURAL AN ALT SIS

up to five different tractions (using negative boundary condition Hags), such as the
scheme used in the TRUSS program Allow for point loads by specifying the node
number that corresponds to the location of the toad. Also specify on the same input
line the magnitude of the point load (positive m the posime x direction). Use the

variable property subroutine VPROP and subroutine PROPTY, both given below, to
allow forspaciallv varying properties.

SUBROUTINE PROPTT <L, MATFLG, DS7MAT, ELMOD,

1 AREA, ALPHA, DELTAT. SIGSAO. BODTF)
C
C DEFINES THE ELASTIC MODULUS, CROSS-SECTIONAL AREA,
C COEFFICIENT OF THERMAL EXPANSION, TEMPERATURE CHANGE,
C PRESTRESS, AND BODY FORCE (PER UNIT VOLUME).
DIMENSION HATFLG(30) , DATKAT(S,L)
NFLAG - RATFLG(L)
C
ELMOD - DATHAT( NFLAG, 1)
IF (ELSOD -LT. 0.) CALL VPROP (ELfiOD)
C
AREA = DATHAT(NFLAG, 5)
IF (AREA .LT 0.) CALL VPROP (AREA)
C
ALPHA - DATMAT( NFLAG, 3)
IF (ALPHA .LT. D.) CALL VPROP (ALPHA)
C
DELTAT = DATHAT< NFLAG, A)
IF (DELTAT LT. 0.) CALL VPP.OP (DELTAT)
C
SIGHAO = DATKAT( NFLAG, 5)
IF (SIGBAO .LT. 0.) CALL VPROP (SIGMAO)
C
BODTF - DATMAT NFLAG b)
( ,

IF (BODTF .LT. 0.) CALL VPROP (BODTF)

C
RETURN
END

SUBROUTINE VPROP (PROP)

COMMON CONST C(10)
COMMON VPROPS X

THE C(I)'S ABE TEN USER-DEFINED CONSTANTS READ IN

THE FIRST INPUT SECTION (SEE THE TRUSS PROGRAM).
NOTE THAT OTHER CONSTANTS, SUCH AS PI, MUST BF
DEFINED BEFORE THE COMPUTED GO TO. THE 'X IS 1

THE VALUE OF THE X COORDINATE VBEN THIS SUBROUTINE

IS CALLED.

pi = a.Kis**
1PBOP = -PROP
GO TO (1, ?, 3, A. S). IPPOP

PROP « sose function of X, C(l), C(8), etc.

RETURN

PROP = soae otter function of X, C(l)» etc.

RETURN
 PROBLEMS 237

C
3 CONTINUE
RETURN
C
4 CONTINUE
RETURN
c
5 CONTINUE
RETURN
C
END

Note that if PROP is negative, it is converted to a positive integer 1PROP. The computed
go to then transfers control to the line whose label is numerically equal to 1PROP. Thus by
simply including the proper FORTRAN statements in SUBROUTINE VPROP, variable
properties are easily accommodated. The C(/)’s are user-defined constants that should be
read in Section 1 of the input file (after NNODES. NELEM. etc.) and may be used in
SUBROUTINE VPROP. By way of example, let us say that we want to use this technique
to allow for the varying cross-sectional area in the problem posed in Example 5-5. Let us
define C{!) = 1.0 (for £>, in inches), C(2) = 0.5 (for D2 in inches), and C(3) = 12.0 (for
Lb in inches). If we use a “-4” as the input for the area A in Section 4 of the input file
(see Appendix B for a description of the input to the TRUSS program), the statements
beginning with label "4” should read

4 CONTINUE
DIfift = (C(1)*(C(3)-X) + C( 2 ) *X )
/ C(3 )

PROP = PI * DI AM* *2 ( )
. 4.
RETURN

since this in effect gives

DU)

ttPU) 2
AU)
4

where the expressions for DLr) and A(.v) represent the correct diameter and area variations
with .v. Note that the variable X in the subroutine represents the .r coordinate and is passed
to VPROP via the labeled common (namely. VPROPS). Use the program to verify the results

'n Example 5-5 and Table 5-2.

5-40 Use the program developed in Problem 5-39 (or one furnished hy the instructor) to

solve Problem 5-35 with two. four, and eight elements. What happens to the results

as the number of elements is increased?

5-41 Use the proaram developed in Problem 5-39 (or one furnished by the instructor) to
solve Problem 5-36 with two. four, and eight elements What happens to the results
as the number of elements is increased ?

5-42 Reconsider Example 5-5. Instead of the point load F and traction i acting on the rod.

the free end isalso complete!} restrained. In other words, both ends ol the bar
now
stress-free state. The bar
are restrained and the bar is assumed to be initially in a
238 GENERAL APPROACH TO STRUCTURAL ANALYSIS

— 6 -6
then undergoes a temperature decrease of 150°F. Assuming a, 5 X I0 in /
in -°F, determine the nodal displacements and element resultants if only two elements
are used Do not use a computer program.

5-43 Determine an expression for the exact solution for Problem 5-42. Gi\e the result in

terms of the pertinent variables and then apply it to the situation in Problem 5-42.

5-44 Solve Problem 5-42 with the help of the computer program from Problem 5-39 (or
one furnished by the instructor) for two, four, and eight elements What happens to
the results as the number of elements is increased?

5-45 Consider the problem posed in Problem 5-35. Instead of the point load F and traction
s acting, the free end is now also completely restrained. In other words, both ends
of the bar arc restrained and the bar is assumed to be initially in a stress-free state.
The bar then undergoes a temperature increase of 15°C Assuming a, = 11.7 x
10 6 m/m'°C, determine the nodal displacements and element resultants if only two
elements are used

5-46 Solve Problem 5-45 with the help of the computer program from Problem 5-39 (or
one furnished by the instructor) for two. four, and eight dements What happens to
the results as the number of elements is increased 7

5-47 Consider the problem posed in Problem 5-36 Instead of the point load F and traction
s acting, the free end is now also completely restrained In other words, both ends
of the bar are restrained and the bar is assumed to be initially in a stress-free state

The bar then undergoes a temperature dei reuse of 120°F Assuming o, = 13 X I0"‘
in /in -°F, determine the nodal displacements and element resultants if only two

elements are used

5-48 Solve Problem 5-47 with the help of the computer program from Problem 5-39 (or
one furnished by the instructor) for two. four, and eight elements What happens to
the results as the number of elements is increased’’

5-49 Show that the trial functions given by Eqs (5-53) and (5-56) for the thrcc-node
triangular element satisfy exactly the compatibility equation given by Eq (5-22)

5-50 Show that the trial functions given by Eqs (5-62), (5-65), and (5-66) for the four-
node tetrahedral element satisfy exactls the six compatibility equations
 Parameter Functions;
C°-Continuous Shape Functions;
Simple Integration Formulas; Active
Zone Equation Solvers

6-1 INTRODUCTION

In this chapter, the concept of the parameter function is formalized. This is followed
by a review of what meant by C°- and C‘-continuous problems. The conditions
is

that the assumed parameter functions must satisfy are given and are discussed in
some detail. The shape functions are then derived for some of the more popular
one-, two-, and three-dimensional dements. In addition, normalized coordinates,
which facilitate the use of these shape functions, are introduced. Axisymmetric
elements are also considered. Three special integration formulas, which may be
used to evaluate the integrals when the lineal, triangular, and tetrahedral elements
are used, are presented. The chapter concludes with an alternate equation solver.
Up to now method has been used to solve
the rather inefficient matrix inversion
the resulting systems of algebraic equations. The equation solver presented at the
end of this chapter takes advantage of the banded and symmetric nature of the
assemblage stiffness matrix.

6-2 PARAMETER FUNCTIONS

In this section, parameter functions are defined in the context of problems in stress
analysis, thermal analysis, and fluid flow analysis. Following this, the requirements
that the assumed parameter functions must satisfy are given and discussed.

239
240 Parameter and shape functions. integration formulas

Definition of a Parameter Function

By now « should te evident that in stress analysis problems to be formulated using

the stiffness approach to FEM, the field variables of interest are the displacements.
From a knowledge of the displacements, the element resultants such as the stresses
and strains may be easily computed. The functions to be used to represent the
displacements on an element basis are referred to as displacement functions in
and as parameterfunctions in general For example, a typical displacement
particular
function m
one -dimensional stress analysis is given by Eq. (5-17) and in two-
dimensional stress analvsis b> Eqs (5-53) and (5-56).
In thermal analvsis problems, the field variable of interest is the temperature.
On an element basis, the temperature within a typical element is approximated by
an assumed temperature function Again, in general terms, this function is referred
to as a parameter function This situation is slightly simpler than the corresponding
stress analvsis problem because the temperature field is a scalar, whereas the dis-
placement field is a vector. Not surprisingly, for the three-node triangular element,

a suitable parameter function for the temperature is of the same form as that given
by Eq (5-53) for a typical displacement
In fluid flow problems, the field variables of interest are the fluid velocities
and pressure If a finite element analysis is to be performed with these so-called
primitive variables, the parameter functions to be used may be referred to more
specifically as lelocm and pressure functions Because the velocities are vectors,
this situation is more closely related to the situation in stress analysis However,
we usually take the Eulenan approach whereby we observe the Quid velocities and
pressure at fixed points This is in contrast to the Lagiangian point of view whereby
an individual particle is followed throughout its motion The reader will recall that

in stress analysis, the Lagrangian point of view is adopted

Let us now summarize By parameter function we mean any suitable function
that is used to represent the field variable in a rypjcal element Parameter functions
are most often taken to be polynomials because of the ease with which they can
be manipulated, as well as for the reasons delineated below In stress analysis, the
parameter function* are really the displacement functions, in thermal analysis, the
temperature function, and in fluid flow analysis, the velocity and pressure functions
(if the formulation is based on the primitive variables). The reader should be able
to extend these notions to other disciplines, such as electromagnetics and mass
transfer.

Restrictions on the Parameter Functions

As explained below, the assumed parameter functions must meet two primary
requirements: compatibility and completeness The primary reason for satisfying
these requirements is to ensure com ergence as the element size is reduced or. more
specifically, as the mesh is refined in a regular fashion Before actually discussing
 PARAMETER FUNCTION'S 241

the meaning of these requirements, may be

it instructive to review what is meant
by a weak formulation.
It should be recalled that a weak formulation to a problem is an integral
formulation that contains in its integrand derivatives of the field variable that are
ol a lower order than those in the original governing differential equation. For an
example, the reader may wish to compare the
differential equation given by Eq.
(4-7) with corresponding weak formulations: Eq. (4-130) from the variational
its

approach and Eq. (4-159) from the Galerkin method.

Furthermore, it should be recalled that a C°-continuous problem is one whose
weak formulation contains at most only first-order derivatives. In a similar fashion,
a C-continuous problem is one whose weak formulation contains at most only
second-order derivatives. This may be generalized as follows: a C"~ ‘-continuous
problem is one \\ hose weak formulation contains at most only nth order derivatives.
We are now in a position to present the compatibility requirement.

Compatibility

The compatibility’ requirement may be stated as follows. For C°-continuous prob-

lems. the parameter function itself (not its derivatives) must be continuous along
the boundaries of the element. For C'-continuous problems, the parameter function
and its first derivative must be continuous, and not necessarily zero, along the
boundaries of the element. This may be generalized as follows. In C"-continuous
problems, the parameter function and its first n derivatives are continuous, and not
necessarily zero, along the boundaries of the element.
It is instructive to discuss this requirement as it relates to one-dimensional
problems, as shown in Fig. 6-1. In particular, consider Fig. 6- 1(a), Note that for
the C°-continuous problem, such as the uniaxial stress member or one-dimensional
heat transfer in a fin, the field variable itself is continuous at the interface between
any’ two elements (in this case, at each node). It should be recalled that the weak
formulation to these problems involves at most only first-order derivatives.

Figure 6-1 One-dimensional element with (a) C-continuous parameter functions and (b)

C-continuous parameter functions. In (a) only the parameter function itself is continuous at

each element interface: in (b) the parameter function and its slope are continuous at each
element interface.
242 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

On the other hand, in C'-continuous problems, the parameter function itself

and its first den vative must be continuous on the interface between any t\\ o elements.
This situation anses in the case of the beam element shown in Fig. 6- 1(b) where
both deflection continuity and slope continuity must be guaranteed so that the beam
does not develop kinks. Recall from elementary solid mechanics that the slope of
the beam at any point is simply the derivative of the deflection at this point (see

Sec. 7-5).
The reader should be thoroughly convinced that problems in two-dimensional
stress (plane stress and plane strain), axisymmetric stress, and three-dimensional
0
stress are all C
-contmuous problems The reason for this is that the weak for-
mulation to these problems involves at most first-order derivatives of the displace-

ments). Therefore, only the displacement function itself (not any of its derivatives)
needs to be continuous along the element boundaries. In Example 6-1 (in Sec.
6-3), the two-node lineal elementis shown to satisfy the compatibility requirement

for C°-contmuous problems. The reader should be further convinced that the pa-
rameter functions assumed in Sec 5-6 for the triangular and tetrahedral elements

do in fact satisfy the compatibility requirement (see Problems 6-3 and 6-4).
On the other hand, in problems involving the bending of plates and shells, the
deflections and slopes should be continuous along the element boundaries This
ensures that the plate or shell does not kink Problems of this type are C'-continuous
because the weak formulation contains at most only second-order derivatives of the
deflections
Elements that obey the compatibility requirement are said to be conforming,
and those that do not obey this requirement are said to be nonconforming Each of
the elements presented in Sec 5-6 is conforming for problems with C°-continuity
In the analysis of plate bending with the three-node triangular element, slope con-
tinuity along the element boundaries cannot be guaranteed (the slopes at the nodes
are continuous). Therefore, this element is said to be nonconforming However,
this element is used in such analyses because it gives results of acceptable accuracy
(particularly if special precautions are taken), even though it violates the compat-
ibility requirement (!)
Let us now turn to the second requirement, that of completeness

Completeness
For C°-continuous problems, the completeness requirement may be stated as fol-
lows. The parameter function must be capable of representing both a constant value
of the field variable and constant first partial derivatives as the element size decreases
to a point. For C'-continuous problems, the parameter function must be capable of
representing both a constant value of the field variable and constant first and second
partial derivatives as the may be generalized
element size decreases to a point. This
as follows. For C "-continuous problems, the parameter function must be capable
of yielding a constant value of the field variable as w ell as constant partial derivatives

of up to order n + 1 as the element size decreases to a point.

 PARAMETER FUNCTIONS 243

Let us examine what this requirement implies for C°-continuous problems in

stress analysis. Consider, for example, the state of uniaxial stress. Constant values
ot both the displacement itself and the derivative of the displacement (i.e., the
strain) must be possible in the assumed form of the displacement function. For the
two-node lineal element, we have been assuming a function of the form

u = r, + ct.v {6-1)

Note that if c 2 we have it = r,, which is a constant as required. This is

is zero,
body mode since the body should be able to undergo a rigid
referred to as the rigid
body displacement without straining (when c 2 takes on a value of zero). In addition,
du/dx = c 2 which is also a constant as required (regardless of the value of c,).
.

Since the derivative of the displacement is the strain, this second condition requires
the displacement function to allow a constant strain in the element. This notion is

readily extended to two- and three-dimensional analyses.

Thermal analysis problems involving heat conduction require only ^-contin-
uous parameter functions because the weak formulations contain at most only first-
order derivatives of the temperature. The completeness requirement is equivalent
to allowing for constant temperature and constant derivative of the temperature.
Since the derivatives of the temperature [with respect to the spatial coordinate(s)]

are proportional to the heat fluxes as a result of conduction, the second condition
is equivalent to the requirement that constant heat fluxes be possible within the
element. Not surprisingly, the assumed forms of the parameter functions in Sec.

5-6 for the displacements are applicable to problems in thermal analysis (see Chapter
8 )-
Problems in viscous fluid flow analysis that are formulated in terms of the
primitive variables are C°-continuous because the weak formulation is seen to
contain derivatives of the velocities of no order greater than one. The completeness
requirement requires that it be possible to have both constant velocities and stresses
within the element. Potential flow problems formulated in terms of velocity potential
function (or stream function) will also be seen to be C°-continuous. In this case,
the completeness requirement is equivalent to the requirements that it be possible
to have constant values of the velocity potential function (or stream function) and
of the velocity components within the element.
This may be generalized in the case of all C°-continuous problems as follows.
The assumed parameter function must contain a pure constant term in addition to
terms that are First order in the spatial coordinates. The reader should review the

assumed form of the parameter functions in Sec. 5-6 in order to be thoroughly

convinced that these assumed parameter functions satisfy the completeness require-
ment (see Problems 6-5 and 6-6). Elements that satisfy the completeness requirement

are said to be complete.

Further discussion of the compatibility and completeness requirements may be
found in several other books on the finite element method. The reader may wish
to consult the books by Huebner [2], Desai [3], and Martin and Carey [4],
as well

as other general books on this subject (see references 38 to 46 at the end of Chapter

1 ).
 1

ONE -DIMENSION -\L ELEMENT 245

Solving for the vector of constants yields

-
C\

c2
1

.*
A'/

v
M (6-5)

Substituting this into Eq. (6-2bi yields

~ 1

1 x, [V ( 6- 6 )
l
.
*/. 1

If the 2x2 matrix is inverted and if the matrix multiplications are carried out,
we get

(6-7)

which is of the form

(b = + A' U')d>
; ;
(6-8)

The shape functions are given by

A' ( U) = — (6-9a)
r
-
/
- *.

A'.(.r) = (6-9b)
xi ~ x ‘

which the reader will recall. In Fig. 6-2 these shape functions are shown super-
imposed on a typical element. It should be recalled further that these shape functions
satisfy the three properties stated in Sec. 5-6, as shown in Example 5-4.

Example 6-1

Show that the following parameter function d> = c, + c 2 .r for the two-node lineal
element satisfies the compatibility requirement for C°-continuous problems.

Figure 6-2 Two-node lineal element w ith shape functions shown for C-contmuous prob-
lems.
246 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

Solution

Consider two adjacent elements with global node numbers I, 2. and 3 as shown
in Fig. 6-3 The compatibility requirement requires that, in effect, the value of the
parameter function at node 2 (t.e., must be the same regardless of whether
clement l or element 2 isconsidered. The assumed form of the parameter function
results in the equivalent form given by Eq. (6-7). or

A + .
r*-x\
|

U-J !<p.
U-v
\

In terms of the global coordinates, for element 1 at r = x2 we ha\e

«t»(t 2 ) = + = ^2

and for element 2 at r = r; we ha\e

4>(t2 )
=
^
— j
= 6;

Stnce <Mr : ) from element I is the same as <5>(r0 from element 2 (i c , both are
equal to 4>»|. the compatibility requirement i\ satisfied Obviouslv. every other
shared node in problems with more than two onc-dtmensional elements can be
shown to satisfy the compatibihtv requirement in a similar fashion H
Note that the element has two nodes and the assumed parameter function has
two constants, c, and c 2 Therefore, two shape functions should be expected, which
was, in fact, what resulted, one for each node All of this is predicated on the
assumption, however, that we are dealing onl> with (""-continuous problems In

an) event. Eqs. (6-9) provide us with the shape functions for this simple clement
in terms of the global coordinate (i e . t)

Next a new coordinate s>stem is introduced that allows us to perform the

more easil). cspeciall) when numerical integrations are per-
required integrations
formed (see Chapter 9)

Figure 6-3 Two adjacent two-node lineal elements used in Example 6-1 to show clement
compatibilit)
 ONE-DIMENSIONAL ELEMENT

Serendipity Coordinate

The so-called serendipity coordinate r is a local,

normalized coordinate defined in
Fig. 6-4 relative to the global coordinate system.
The reason for the name serendipity
is given in Sec. 6-4. Note that r = 0 at.v = x, where x = (x, + x,)/2. Note also
that r varies from - I at .r = .t, to+ at .t = xr that is, —
1
r ^ 1 I . It should
be obvious that r and .r are related by

( 6 - 10 )

In terms of this new coordinate, the shape functions become

N =
t
'/2(1 - r) (6-1 la)

and

Nj = Vi(\ + r) (6-11 b)

as shown in Example 6-2 and Problem 6-9.

Example 6-2

Show that the shape function for node i is given by Eq. (6-1 la) for the two-node
lineal element.

Solution

2(.r - .rA
x
j ~ T /

x, - x - lx -t- 2[(.r, + jc,)/2]

£ l

2 x, - -ri

2\x, - A, / Xj - .r,

r = ~t r = +l

1
o| -r

i 4 I /

’ -t,

Figure 6-4 Two-node lineal element showing global coordinate r and serendipity coor-
dinate r.
248 PARAMETER AND SHAPE FUN CHONS. INTEGRATION FORME LAS

which is the desired result [see Eq. (6-9a)]_ The reader should be cominced that
the shape functions given by Eqs. (6-1!) have the three properties delineated in
Sec. 5-6 and that, when plotted on a typical element, the result is the same as Fic.
6-2 .

Length Coordinates

The main reason for introducing length coordinates is to take advantage of the
simple integration formulas given in Sec. 6-7. Unlike the single global coordinate
.t and the single serendipity coordinate r, two length coordinates are associated with
the lineal element However, these two length coordinates are not independent, as
will be seen shortly. Consider the element shown in Fig 6-5(a). An internal point

p is also shown (point p is not a node) Let us now define the so-called length
coordinates L, and Lf as follows.

length pj
(6-1 2a)
length ij

and
length ip
(6-12b)
length ij

Obviousl>. we have 0 « L, I and 0 *5 L, *£ 1 Moreover, we must also have

L, + L, = l (6-13)

because the sum of the fractional lengths must equal unity The two length coor-
dinates are shown on the element in Fig 6-5(b) If the point p is located at the
general global coordinate x. then Eqs (6-12) arc equivalent to

(6-1 4a)

and

(6-1 4b)

(bi

Figure 6-5 Two-node lineal element showing (a) interior point p and (bl the two length

coordinates L, and Lr Ante; Point p is not a node

 TWO-DIMENSIONAL ELEMENTS 249

The reader should recognize the expressions on the right-hand sides as N, and /V,,

and we have the fortuitous result that the shape functions, in terms of the length
coordinates, are given by

N, = L, (6-15a)

and

Nj = Lj (6-1 5b)

Once again, these shape functions are seen to have all three properties stated in
Sec. 5-6 and can be plotted as shown in Fig. 6-2. Like the serendipity coordinate,
the length coordinates are also referred to as the local, normalized coordinates (or
simply normalized coordinates) because they are defined locally in each element
and are normalized to unity.

6-4 TWO-DIMENSIONAL ELEMENTS

In this section two of the most popular two-dimensional elements are considered:
the three-node triangular element and the four-node rectangular element. In both
cases, the appropriate shape functions are derived in terms of the global coordinates
and then cast into forms that utilize suitable normalized coordinates. In both cases,
only the shape functions for C°-continuous problems are considered.

Three-Node Triangular Element

A typical triangular element is shown in Fig. 6-6(a) with nodes i, j, and k. By

convention, the nodes associated with a particular element are always given in a
counterclockwise order. For example, the element shown in Fig. 6-6(b) with global

coordinates in general and

Figure 6-6 Three-node triangular element showing (a) nodal

(b) global node numbers for a typical element.

 1 x
250 PARAMETER ASD SHAPE aACIKAS. INTEGRATIOS KAMI LAS

node numbers 5. 17. and 21 may be defined as hasing nodes 5. 21. 17 or 21. 17.
5 or 17, 5. 21 because each of these designations is In accordance with the con-
\ention just given.
It is com enient to set up a cartesian reference frame and to define the coordinates

of the nodes in terms of these coordinates. In other words, the coordinates of nodes
i,j, and 4 are (r ._i ). and (xA .yJ. respectively. Let us again represent the
y y
parameter function as 6 such
that at nodes i. j, and 4 we have d»,. 6 . and 6*. In
y
other words, the parameter function 6 must evaluate to 6, at (x„v,). etc. Once
again <ti may be the x or
component of displacement, the temperature, etc.
v

We are now in a position to dense the shape functions for the three-node
triangular element m terms of the global (x.v) coordinates.

Global Coordinates

We begin by assuming some type of parameter function From Secs. 5-6 and 6-2
it would appear that the form

6 = fj C; Cj» (6-16a)

or

is appropnate because it meets both the compatibilit> and the completeness re-
quirements for C^-continuous problems At node i where x ~ x, and v = \,. we
require 6 = at node j where x = x and\ = \ we require 6 = 6r andfinall>
«!»,.
y r
at node 4 where .r = xA and v = v A we require 6 = 6*. From this and fcq
.

(6-16a). it follows that

6, ® <ii
*" + c,v, (6-17a)

6, = i,1
-* + C 5 Vy (6-17bl

6 = <,i
c; xi + c»v A (6-17c)

Wnting these in matnx form gives

fil- f!

Solving for the vector of constants and substituting the result into Eq (6- 1 6b) vield>
! m (6-18)

i x, v (
r'[V
I I X, \, *,
.i j-k lA.
Equation (6-19) is of the form
= v)6A 16-201
£> A',(x,v )<5, -i- <V (r,i)6 <VA (x.
j /
 A

TWO-DIMENSIONAL ELEMENTS 251

where it can be shown (see Problem 6-10) that the shape functions are given by

N, lx ,v) = =
mu + m 2 |.v + "fn v (6-21 a)

Njlx,y) = -
mn + "'22-c + "'32 V (6-21 b)

N k (.x,y)
-
= w n + m 2i x + «'33.V (6-21 c)

and in turn

JS II 1 K> m 2\ = 0, --
y»)/2 a "fn = (A; - \jV2A
m l2
= foy, - x,yk )!2A m 22 = : (y»
-
- y,)/2A ?!
11 \ V? C'l
(6-21 d)

"63 = (-r,Vj - Xj\,)/2A m 22> = o,

--
y,)/2A «' i3 = <*, ~ \V2

and
~
I
-h V,
1

A - '/: det 1 XJ >j

= area of triangle ijk (6-21 e)

It should be noted that these /« ’s are strictly

y
a function of the nodal coordinates.
Since the area A
of the triangular element is never zero, the inverse indicated in
Eq. (6-19) will always exist. This in turn implies unique values of the m ,/ s for
every element. The reader should show that these shape functions satisfy ihe three
properties given in Sec. 5-6 (see Problem 6-11).
Note that the element has three nodes and the assumed parameter function has
three constants, c,. c 2 , and c 3 Therefore, three shape functions are to be expected
.

and, indeed, three such functions were found, as given by Eqs. (6-21). Next a new
type of normalized coordinate is introduced that allows us to evaluate easily the
various integrals that arise.

Area Coordinates
Consider the three-node triangular element shown in Fig. 6-7. The point p is an
arbitrary internal point, not a node. The area coordinates L,. Lr and Lk are defined

as follows:

area pjk
(6-22a)
'
area ijk

area pki
(6-22b)
1
area ijk

area pij
,
— (6-22c)
area ijk

follows that the area coordinates are not all independent

From these definitions it

but are related by

16-23)
L, + Lj + Lk — 1
252 PAKAMfc I Ui AW 111AM- rLMTI ICTVS. LVIUiXATHTS KRML1AS

Figure &-7 Three-node triangular element with interior point p Sole- Point p is rot 2
cods

because tbs sum of the fractional areas must equal units . It should also be obsiocs
that ssbsn point p coincides with nods 1, L, = I 2nd Lt — Lk = 0 This Iasi

observation is generalized and summarized below

L,(x,.y,t = I L,(xr \ji = 0 L,lxk ,\ k ) = 0

Lt ix,.y,\ = 0 Lt ixr \,) = 1 L) ixl .\ l ) = 0 (6-24}

L k ix,.x,) = 0 Lt ixr \ji = 0 £. i (xi .s 1 ) = I

Now consider points p, and p 2 which happen to be on a straight line that is

.

parallel to leg jk as shown in Fig 6-fcla) Because the area of a triangle is given
by one-half the product of the base and the altitude, it should be clear that lines of
constant area coordinate, eg . L,. are parallel to the opposite leg, in this case leg
jk Moreover, as the line containing points p, and p 2 is moved parallel to leg jk.
the area pjk vanes linear!) Therefore, the area coordinates L, must vary linearly

Figure 6-8 Three-node tnarzular clement show mg lines of constant area coordinates In
(3) triangles pjk 2nd p*jk base the same areas and hence the dashed line represents 2 line
cf constant L In tb| specific lines of cue steer L, zzz shown.
 TWO-DIMENSIONAL ELEMENTS 253

from zero on leg jk to unity at node i. These observations are summarized in Fia.
6-8(b), which shows several lines of constant L , . Similar conclusions may be drawn
about the behavior of and L k .

The reader may have noted by nowthat the area coordinates possess the three
properties of C°-continuous shape functions given in Sec. 5-6 for the triangular
element. Consequently, the shape functions for nodes /, j, and k may be given in
terms of the area coordinates as

N, = L, N -
)
L ,
and N = k Lk (6-25)

The implication of this result will be appreciated when integrals must be evaluated
that contain the shape functions, as shown in Example 6-8 in Sec. 6-7. Example
6-3 below shows a little more rigorously why N, = Ln and the reader may show
that Nj = Lj and N =
k Lk in a similar fashion (see Problems 6-13 and 6-14).

Example 6-3

From the definition of L, given by Eq. (6-22a), show that L, is the same as the
shape function N, given in Eqs. (6-21), and hence N, = L,.

Solution

From the definition of L, we have

"
'1 .V y
Vi det 1 Xj v
- r

area pjk _
I
-C

area ijk A
(-V,.C
- -Vr.Vy) + (Vy ~ >'Q X + (Xt - -Vy)V
_
2/1

= in I,
+ m 2 i-v + m 3! y = N,

where /«,,, mi 2 i, and /«,, are defined in Eq. (6-2 d). 1

Four-Node Rectangular Element

Figure 6-9 shows a typical rectangular element with nodes /, j. k. and in in a

cartesian reference frame. The element must be oriented such that sides ij and km
are parallel to the and sides jk and mi parallel to the a axis. This clement
y axis,

is useful in situations in which the geometry is regular. In this text, the nodes

associated with a given element must be given by starting at node i and proceeding
counterclockwise to node m. For example, the element shown in Fig. 6-9 (b) must
given
be defined as having nodes 18, 5, 10. 7, in this order, if the shape functions
below are used. In Chapter 9. these restrictions are lifted at the expense of
a little

more mathematics.
256 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

hence the name serendipity, coordinates. Note that r = 0 and s = 0 at t = r and

v = T, respectively, where
r, + t, + t„,
_ (6-31 a)
2 “ 2

and

(6-31 b)

The serendipity coordinates r and s are normalized by

(6-32a)

(6-32b)

where a and b are the element half-lengths as shown in Fig 6-10 It also follows
that - 1 *£ r 1 and - 1 ^s« 1
Since the element has four nodes and each node has one shape function as-
sociated wath it, four shape functions must be found Although these functions can
be derived based on the results of Eqs (6-29). (6-31). and (6-32) after much
algebraic manipulation, let us simply present the shape functions as

N, = </4(l + r)( 1 - s) Nk = !4(1 - r)(] + s) ,,,

Nj = •/«(! + r)( i + s) N„ = Va(1 - r)(l - s)

It should be noted that each of these shape functions has the same form as the
assumed parameter function [see Eq (6-26a)J Moreover, the shape functions given
in Eq. (6-33) have the following properties

1. The shape function associated with a particular node evaluates to unity at this
wade, ad other shape functions evaluate to zero at this, node
2. The shape function associated with a given node varies from unit) at the node
in question to zero at each of the remaining nodes
3. The sum of the shape functions is identically equal to unity

The parameter function may be given in terms of these shape functions and
the nodal values <J>( , 4>7 , etc , as

<!> = '/4(1 + r)(l - s)<{>, + '/4(1 + r)(l + s)4>,

+ '/4<1 — r)(I + j)$ x + 'A{\ ~ r)(l - 5)d>,„ (6-34)

This form of the parameter function meets the compatibility and completeness
requirements stated in Sec. 6-2.
 TWO-DIMENSIONAL ELEMENTS 257

Example 6-4
Integrals arise routinely in the finite element method. In Chapter 8. for example,
the integral for the stiffness matrix as a result of convection from a surface will be
seen to contain the shape functions and the convective heat transfer coefficient h.
A typical entry in the matrix is given by the integral fA N,Njh dx dy. With the help
of the shape functions given by Eq. (6-33), evaluate this integral for the four-node
rectangular element. Assume h to be constant.

Solution

First, from Eqs. (6-32) we note that

dx = a dr

and

dy - b ds

The integral is then evaluated as follows:

N.Njhdxdv = + ’1(1 - s)](!4(l + /)(! + s)]h ab dr ds

$ f 1

abh P p
= —— j
(1 + 2r + 2
r )(l - s
2
) dr ds

= ~ [/_,(• + 2r + r
2
) dr][ f_
t
(1
- s
2
) ds]

+ -1 + -
r -7
=
abh 17
(
r -f r-
, ^ — 17
I
5“
*7
—
16 Lv 3 ) - \ - LV 3 / - 1 _

abh 8\/4\ _ 2 ,, ~ hA
ab '

16 \3J\3J 9 18

Note that the area A of the element is given by (2a)(2b) and, in fact, has been used
above. It will be shown in Chapter 9 how integrals such as this may be evaluated

using Gauss-Legendre quadrature (a particular type of numerical integration).

6-5 THREE-DIMENSIONAL ELEMENTS

Two of the most popular three-dimensional elements are now considered: the four-
node tetrahedral element and the eight-node brick element. In the case of the former,
the appropriate shape functions are derived in terms of the global coordinates
and

in terms of suitable normalized coordinates. In the case of

the brick
then given
element, the shape functions are simply given terms of the serendipity coordinates,
in

In both cases, only the shape functions for C°-continuous

problems
now r, s, and t.

are considered.
258 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

Four-Node Tetrahedral Element

A typical tetrahedral element with nodes i ,j, L, and m is shown in Fig. 6-11. The
nodes associated with a particular element must be given m the order specified by
the following convention

1. Any of the nodes may be specified to be the first node.

2. The next three nodes must be taken in a counterclockwise direction as viewed
from the first node.

In effect, this gives twelve different ways in which an element may be defined, as
summarized in Table 6-1 Zicnkiewicz [61 gives a systematicway of splitting an
eight-cornered brick into six tetrahedra. Such a process could be relegated to a
subroutine and obviously makes the tetrahedral element much more practical because
it is easier to picture a three-dimensional body discretized into bricks.

Table 6-1 Proper Element Definitions for

the Four-Node Tetrahedra! Elements

12 Nodes
3 4

Example 6-5
Consider the element shown in Fig. 6-12. Which of the following element definitions
satisfies the ordering convention?

a. 5, 9, 18, 21
b. 9, 5, 18, 21

Solution

The order 5, 9, 18, 21 does satisfy the convention because nodes 9, 18,
and 21
occur in a counterclockwise order when viewed from node 5. The order 9, 5, 18,
 three-dimensional elements 259

Figure 6-11 Four-node tetrahedral element showing nodal coordinates in general.

Figure 6-12 Tetrahedral element with global node numbers shown for a typical element.

21 does not satisfy the convention because nodes 5, 18, and 21 do not occur in a
counterclockwise direction when viewed from node 9. B
Let us set up a cartesian reference frame and define the coordinates of the
nodes in terms of these global coordinates. In other words, the coordinates of nodes
/, j, k, and m are ( x„yn z,), (.v,,y,,z,), etc. Let us again represent the parameter
function as c() such that at nodes i,j, k, and m, we have 4>, , <j>
y,
4>A , and Stated
differently, the parameter function 4> must evaluate to 4>, at (,Y,,_v,,r,), etc. Of course,
(f> may be the .v, y, or z components of displacements, the temperature, etc.

The shape functions for the four-node tetrahedral element are now derived in
terms of the global (x,y ,z) coordinates by using the procedure illustrated in Secs.

6-2 and 6-3.

260 PARAMETER AND SHAPE FUNCTIONS INTEGRATION FORMULAS

Global Coordinates

From Secs. 5-6 and 6-2, it may be concluded that the parameter function

<}) = Cj + c2 x + c3y + c4 z (6-35a)

or

4> = [1 (6-35b)

Lc 4 J
is appropriate because it meets both the compatibility and the completeness re-
quirements for C°*continuous problems At node /, where x - x„ y = y„ and
z = z„ we insist that <|> = <J>,, or

<j>, = C| + c2x, + i 3 y, + t 4 r, (6-36a)

In a completely analogous manner, we have

<$),
= £,+ c 2 x} + t 3 \j + c 4 Zj (6-36b)

<j>* = Ci + c2 xk + rji* + c4 z k (6-36c)

and

4»„, = c, 4- c 7 x„, + c 3) m + c4 z„, (6-36d)

Writing Eqs (6-36) in matrix form yields

<f>, 1 C|

4>, 1 A, Zj C2
4>* 1 xk 1* Zk <3
_1 x,„ v„, z,„_ L^J
Solving for the vector of constants and substituting the result into Eq. (6-35b) gives

1 X, V, z, 4>,

1 *, y, h 4>,
(6-38)
1 ** y\ zt

_ 1 xm >'m zm. [_4>,J

which is of the form

4» = N,(x,y,z)$, + + N k (x,y,z)fyk + N„{x,y,z)§,„ (6-39)

It can be shown that the shape functions are given by

N,(x,y,z) = m,, + m2 iX + m M y + m4 ^z (6-40a)

Nj{x,y,z) = mi 1 2 + + m 32 y + m42 z
MI 22 *
(6-40b)

Nk(x,v,z) = mj3 + m23 x + m33 v + m43 z (6-40c)

Nm (x,y,z) = m i4 + m 24 x + in 34 y + m44 z (6-40d)

 — s

THREE-DIMENSIONAL ELEMENTS 261

where

v
- , -j

»'n .u 111 2 1
det v* zk

(6-40e)

n h\ del _1_
"'ll det
6V 6K

and so forth. In these expressions. V is the volume of the tetrahedron and is given
in terms of the nodal coordinates by

AJ V, *7
det - J = volume ijkm (6-40f)
xk }\ :k

I
1
'V/l 3 m -/it.

The reader may want to review Sec. 2-7 in order to see more clearly how these
results were obtained. Since the volume of the tetrahedral element is never zero,
the indicated inverse in Eq. (6-38) always exists and. therefore, each of the m ,/
may be computed for each element as indicated above. It can be shown that these
shape functions satisfy the three properties given in Sec. 5-6 (see Problems 6-30
to 6-32).
Finally, it is noted that the element has four nodes and the assumed parameter
function has four constants, which in turn implies four shape functions. This is. in

fact, seen to be the case. In effect, this assures us that the four-node tetrahedral
element meets the compatibility requirement for C°-continuous problems.
Next another normalized coordinate is introduced— the volume coordinate. In

Sec. 6-7 (and Chapter 9), the reader will come to appreciate these coordinates.

Volume Coordinates
Consider the four-node tetrahedral element shown in Fig. 6-13, where the internal

point not to be confused with a node. In effect, point p is used to create four
p is

dement with interior point p. Note: Point p is not a node.

Figure 6-13 Tetrahedral
262 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

other tetrahedra, each contained in the onginal tetrahedron The volume coordinates
L,, Lj, Ll and L„
. are defined as follows:

volume pjkm volume pijm

volume ijkm volume ijkm (6-411
volume pkmi \olume pijk

volume ijkm volume ijkm

Because the sum of these fractional volumes must equal unity, the four volume
coordinates are not independent but rather are related by

L, + L, + Lk + L„, - 1 (6-421

Obviously, if the point p coincides w ith one of the nodes, the volume coordinate
associated with that node has a value of unit) and all other volume coordinates are
zero Since the volume of a tetrahedron is one-third the product of the area of any
face and the height normal to the face, it follows that the volume coordinates vary
linearly throughout the tetrahedron It also follows that planes of constant volume
coordinate for a given node are parallel to the face of the tetrahedron that is opposite
the node
Like the length and area coordinates, the volume coordinates possess the same
three properties that the shape functions have for C°-contmuous problems There-
fore, the shape function for a particular node is equal to the corresponding volume
coordinate, or

N, = L, N, = Lj Nk = Lk and Nm = Lm (6-43)

Example 6-6 below shows more rigorously why N, = L, from which the reader
may show in a similar fashion the validity of the remaining equalities in Eq (6-

43) (see Problems 6-33 to 6-35)

Example 6-6

With the help of the definition of L, given in Eq (6-41), show that L, is the same
as the shape function N, given in Eqs (6-40), and hence N = L

Solution

Beginning with the definition of L„ we write

l x y z
i x, y,
~1
~det
6 i xi yi
1 xm
volume pjkm
L,
volume ijkm V
 THREE-DIMENSIONAL ELEMENTS

or

"l yt h
w*
.V
L = ia h
'

w _-Y„,
V*

v,„
Zk -
J
1
Vi

y
Zk

' -V, 1 "l xj

S**
Z

6V
_
1
-h
-v„,
L
“w de '
j
l -Vi

-V„,
)’k

y,„.

or

L, = + m2 i-v + wi 3 ,y -F w,,r = /V,

where mn ,
;h 21 ,
»j 31 , and w4 ,
are defined in Eqs. (6-40). El

Eight-Node Brick Element

The eight-node brick element, shown in Fig. 6- 14(a) is also known as the rectangular
prismatic element. Note that the nodes are conveniently numbered 1, 2, . . . ,
8.

The letter designation (/, j, etc.) for this element is cumbersome and, therefore, is

abandoned.
Each element must be defined by its eight global node numbers given in the

order 1, 2, . . . ,
8. For example, the brick element in Fig. 6- 14(b) must be defined
as having nodes 15, 18, 14, 10, 21, 62, 81, and 45 (in this order) if the shape
functions given below are used. The element must be oriented such that its

1-2-3-4 face is parallel to the x-y plane, the 1 -5-6-2 face is parallel to the y-z plane,

and so forth. In Chapter 9 these overly restrictive requirements are relaxed.

Each node is typically given a unique global node number and is defined by
its nodal coordinates. For example, the coordinates of node 1 are (.v, , Vi ,r,). Let
us represent the parameter function as <5. It follows that at nodes 1, 2, . . . , 8,

Figure 6-14 (a) Eight-node brick element showing nodal coordinates in general (b) Eight

node brick clement showing global node numbers lor a t\ picul element.
264 PARAMETER AND SHAPE FUNCTIONS, INTEGRATION FORMULAS

the parameter function evaluates to 4> t , <j>

2, . . , 4*s- As usual, cj> may represent
one of the components of the displacement, the temperature, and so forth.

Global Coordinates

The shape functions for the brick element may be derived in terms of the global
coordinates by following the standard procedure that has been illustrated in this
chapter for each of the other elements As a starting point, the following parameter
function may be assumed

<t>
= c, + c 2x + c3 y + c4 r + t 5 .rv + t b yz + c 7 zr + c 6 xyz (6-44)

The reader is assured that this form of the parameter function satisfies both the
compatibility and completeness requirements for C°-contmuous problems Although
the standard procedure would yield the shape functions (after much algebra), it is

far more convenient to use the shape functions for this element if they are given
in terms of the serendipity coordinates as given below

Serendipity Coordinates

Figure 6-15 shows the four-node brick element and a local, normalized coordinate
system (r,j,r) with its origin at the centroid (t,v,z) of the brick The coordinates
of the centroid are given by

X )
+ X4 x2 + -ti
(6-45a)
2 2

>’i + yi V
"t \e.c (6-45b)
2

s«+l /»-l

Figure 6-15 Brick element showing the serendipity coordinates r. s and t Note —
—J
r £ +1, — <j s +1,
1 and - 1 < ^ +1 Aho note
r r = (r - x)!a, s = (>
— v)/h,

and t = (z — ~i)Ic
 AXISYMMETRIC ELEMENTS 265

2
z i + fs z4 + Zg
, etc. (6-45c)
2 2

The so-called serendipity coordinates are defined in terms of global

coordinates by

-V — x
(6-46)
a

where a, b, and c are the element half-lengths in the x, y, and z directions, re-
spectively. It follows that each serendipity coordinate has a value between - 1 and
+ l,or — +1, — s + |, and — 1 ^ t s£ + 1 . The shape functions
for the eight-node brick element are given below (for C°-continuous problems):

(V, = 14(1 + r)( 1 - r)(l + t) N5 = 'A( 1 + r)(l - s)(l - t)

N2 = %(l + r)( 1 + s)(l + t) N6 = '4(1 + r)(l + s)(l - r)

(6 . 47)

N3 = 14(1 - r)(l + s)( 1 + /) N =

7 14(1 - r)(l + s)(l - t)

M, = 14(1 - r)(l - s)(l + t) Ns = >4(1 - r)(l - s)(l - /)

Each of these shape functions has the same form as the assumed parameter function
[i.e., Eq. (6-44)]. In addition, the shape functions given by Eqs. (6-47) have the
same three properties that the two-dimensional serendipity shape functions possess.
Also, the compatibility requirement is met by these shape functions on each face
of the element (i.e., on the element boundaries).
Special numerical integration methods make this element quite practical as the
reader will come to appreciate in Chapter 9 where Gauss-Legendre quadrature is

introduced.

6-6 AXISYMMETRIC ELEMENTS

Many practical problems are axisymmetric in nature. Such problems must meet
two conditions to be classified as axisymmetric. The the body must be
first is that

a body of revolution as shown in Fig. 6- 1 6(a). Consequently, a cylindrical coordinate

system is usually adopted with global coordinates r, 0, and z. However, because
of the geometric axisymmetry, the geometry is invariant with respect to the 0
coordinate. This means that, in effect, only the two remaining coordinates (r,z)
need be considered. The second requirement is that the material properties, the
to
external effects, boundary conditions, and so forth, may not be a function of 0.
External effects include surface tractions, body forces, prestresses, and heat sources.
Despite these restrictions, many practical problems fall into this class of problems.
The axisymmetry, in effect, allows us to analyze a three-dimensional problem
as though were a two-dimensional problem. This implies that the three-node
it

triangular element and the four-node rectangular element may be used provided
that they are rotated through 2 -tt radians as shown in Fig. 6-16. The
elemental
actually
volume is then given by dV = 2tt r dr dz. Note that these elements are
 .

266 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

Figure 6-16 (a) Bod) of re\olutton discretized into (b) triangular and (c) rectangular
donut shaped elements

tnangular and rectangular donuts or toroids How e\ er. because of the axisj mmetiy
only half of the planar domain needs to be analyzed, as shown for the axisymmetric
body in Fig 6-17 The planar domain is am plane that goes through and is parallel

to the centerline of the bods The analysis for all practical purposes is thus fwo
dimensional
The shape functions for the three-node tnangular element are gi\en by Eqs.
(6-21) lor Eq (6-25)] presided that t is replaced by r (or :) and \ is replaced by
r (or r). Of course, each of the nodal coordinates must be similarly reinterpreted.
The shape functions for the four-node rectangular element are also readily adapted
to this situation.

Figure 6-17 (a) Axisymmetnc body with us half-plane discretized into (b) tnangular
elements and (c) rectangular elements
 SOME SIMPLE INTEGRATION FORMULAS 267

6-7 SOME SIMPLE INTEGRATION FORMULAS

When the length, area, and volume coordinates were introduced, it was stated that
the main reason for doing so was to allow the use of simple integration formulas
in evaluating the integrals that routinely arise. These integration formulas are pre-
sented in this section and are illustrated in several examples. In each of these
formulas the factorial operator is used, where n! = (>;)(» - l)(n — 2) • •
•
( 2 )( 1 )

and 0 ! = 1 .

Length Coordinates

The appropriate integration formula in terms of the length coordinates L, and Lt is

given by

gift!
JlL?Lf dl / (6-48)
(a 4- p + 1)!

where dl is an elemental length between nodes i and j and / is the length of line
between nodes i and j, as shown in Fig. 6-18. The exponents a and p must be
positive integers.

Example 6-7

With the help of Eq. (6-48) evaluate the element nodal force vector for the body
force (that is, f*) as given by Eq. (5-102) for the uniaxial stress member.

Solution

The nodal force vector f(, is given by

fl = fVbA dx
J X,
= r
Jji
yA dx

Let us assume that 7 the weight per unit volume, and A, the cross-sectional area,
,

are both constant, so that the product 7 A may be removed from the integral.

One-dimensional element showing infinitesimal length cl/ as well as the actual

Figure 6-18
element length I.
268 PARAMETER AND SHAPE FUNCTIONS, INTEGRATION FORMULAS

Moreover, if the shape functions N, and are given in terms of the length coor-
dinates [see Eqs. (6-15)1, we have

r l!0!
/ 1
(1+0+1 )!
U= 7^// 0!1'

L(0 + 1 + 1)!
'-1

[ is the element length in this case

Xj„ we get

f> " 2 \
[:]
which is the expected result [see Eq. (5-102)].

Area Coordinates

The appropriate integration formula in terms of the area coordinates is given by

pipy
SA L?LfLl dA 2A (6-49)
(a + P + y + 2)'

where dA is an elemental area of the element (usually dx dy or dr dz) and A is the

area of the triangle formed by nodes i, j, and k The area A is easily computed
from Eq. (6-21e) since the nodal coordinates are always known The exponents a,
P, and y must be positive integers.

Example 6-8

A typical entry in the stiffness matrix as a result of convection from a thin two-
dimensional body is given by fA N,Nt h dx dy (as shown in Chapter 8) Evaluate
this integral if the three-node triangular element is used

Solution

With the help of Eqs. (6-25) and (6-49), we evaluate the integral as follows

J^N,Njhdxdy = j^L,Ljh dA

,
1 ! 1
'
0 !

+
2A = —
hA
(I + 1 + 0 2)! 12
 SOME SIMPLE INTEGRATION FORMULAS 269

Volume Coordinates

The appropriate integration formula in terms of the volume coordinates L L

n ; , Lk ,

and L m is given by

aipi-ylS!
hLr^LJLl dV = 6V (6-50)
(a + 3 + y + 8 4- 3)!

where clV is volume (usually dx dy dz) and V is the volume of the

the elemental
tetrahedron formed by nodes /, j, k. and in. This volume is readily computed from

Eq. (6-40f). The exponents a, (3, y, and 5 must be positive integers.

Example 6-9

A typical entry in the element nodal force vector as a result of a constant volumetric
heat generation rate Q (per unit volume) is seen in Chapter 8 to be given by
$ v NjQ dx dy dz. Assuming a four-node tetrahedral element, evaluate this integral.

Solution

With the help of the second equation in Eq. (6-43), we get

SvNjQ dx dy dz = fvLjQ dV
0 ! 1 ! 0 0
! !

- Q 6V = 'AQV
(0 + 1 + 0 + 0 + 3)!

assuming Q is constant over the element. From this result it is concluded that one-
fourth of the heat generation is allocated to node j. In a similar fashion, it may be
shown that one-fourth of QV is allocated to each of the remaining three nodes. The
total amount is, therefore, given by QV. The validity of this result should be
intuitively obvious. ®
Example 6-10
Redo Example 6-9 with the following spaciaily dependent volumetric heat gener-
ation rate:

Q = Qox
where Q0 is a constant.

Solution

The global coordinate a can be represented in the terms of the i coordinates of each

of the nodes by writing

A k t
.\
t
"t" Ty.ly ' I'i.X I
 , J
270 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

where the volume coordinates have been used. The integral may now be evaluated
in the following manner

fvfyQ dx dv dz = Qo fyLj(L,x, + LjXj + L kxk + L mx„) dV

- Q0 (f yL,L,x, dV + /vLJXj dV + / yL,Lt xk dV + $yL,Lm xm dv)

l't'0'0' O'2'O'O’
x 6V + x, 6V
.
(1 + I +0 + 0 + 3)' •
(0 + 2 + 0 + 0 + 3)'

O'in'O 1
0'1'Q’l' 1
+ xk 6 + X"
(0 + 1 + 1 + 0 + 3)' (0 + 1 + 0 + 1 + 3)' J

= + ZlO Xj + V20 X k + V20 V

X^Jq 0
Q„V
m (x + Ixj + xk +

Unlike the previous example, this result is not at all intuitively obvious.

6-8 ACTIVE ZONE EQUATION SOLVER

Up to now we have used only the matrix inversion method to solve the system of
algebraic equations implied in Ka = f This approach is not very practical in large
finite element models because the computation of the inverse of a large matrix can
result in excessive computer execution time A much more efficient solution method
is discussed in this section The basic idea is to decompose the matrix K into lower
and upper triangular matrices L and U such that K = LU It is shown below how
this results in a straightforward solution for the vector a, once this triangular

decomposition (or LU decomposition) has been performed. Note The matrix L in

this section is in no way related to the linear operator matrix (also L) from Chapter

Triangular Decomposition

The triangular decomposition of K into L and U such that K — LU is readily

summarized in the form of a short algorithm Let us first illustrate the basis for the
algorithm by working on a 3 x 3 matrix K It follows from K = LU that
‘

A'„ *12 *13 ^11 0 0 v tl t/12 t/,3

*21 *22 *2, = ^21 ^22 0 0 U22 ^23

Uji *32 *31 -f'JI ^32 ^33. 0 0 U 33_

The number of unknowns on the right-hand side of Eq (6-51) is seen to be greater

than the number of implied equations, that is, twelve unknowns and nine equations
The number of unknowns may always be reduced to the number of implied equations
 ACTIVE ZONE EQUATION SOLVER 271

by requiring each diagonal entry in L to be unity, or Ln = 1. Therefore. Eq.

(6-51) becomes

(D (D (D
“ 0'
K 12 Ku 1 0 "c/„ «/.3 Uu
V, ATn A23 = ^2 J
1 0 0 u22 Uu
LAT, AT: AT, J^3! l ,2 1
_ 0 6 Uu
Note that the matrix K has been divided into (three) zones. In general, an n X n
matrix K would be divided into n such zones. If the matrices on the right-hand
side are multiplied out and equated with the left-hand side, entry by entry, the
following nine equations in nine unknowns (L 2 ,. L,,. L, 2 . U u U i2
- . etc.) are

obtained:

Zone 1: AT, = y„
Zone 2: K t
r = C /,2
AT, = U 2 U\ K

AT j = Co,C/,.
Zone 3: A', = C /,3
AT, = A-, // ,

AT, = C. 3 ,C/, :

K2S = + t/2 ,

K2i = L 3I 13 + L 32 U22 + U3i

[/

These equations may in turn be solved for the nine unknowns to gi\c the following:

Zone 1: Uu A'„

Zone 2: Vn Kn

Ln hi
Un
U22 AT, A 21 (/ 12

Zone 3: t/.3

Ui hi
Uu
K ,r - Ly,U n
u 2
:

Urr

U2 3 = AT, - L 2l U n

Uu = A', - - L
A few subtle points should be made here. Note how the calculations in each

zone use only the entries in the K

matrix that are in this zone and only those
entries

no! from subsequent

from the L and U matrices from previous and present zones (bui
 1 1 1
1

272 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

zones). The present zone here is more commonly referred to as the acthe zone;
hence, the name active zone equation solver.
The above equations may be generalized for an n x n matrix K as follows
17] For the first active zone, we have
Uu = A',, (6-53)

Ln = 1 (6-54)

and for each subsequent active zone j from 2 to », we have

U Xj = K tj
(6-56)

and

K„ - S L„U„,
Ljx ~ for i = 2, 3. .
J - (6-57)
U„
I -
U„ = Kv - 2 L, m Umi for t - 2. 3. J - (6-58)
m= 1

and finally

= 1
(6-59)

j~ 1

a" H i 2 L im Unj (6-60)

For a symmetric K matrix, the entries in the lower triangular matrix L can be
obtained directly from the upper triangular matrix U from

L„ - for ] 2 i
(6-61)

In Problem 6-44. the reader is asked to show why Eq (6-61) holds when K is

symmetric.
The use of these equations is best illustrated by an example

Example 6-1

Perform the triangular decomposition of the following sjmmetric matrix

 ACTIVE ZONE EQUATION SOLVER

Solution

The calculations are summarized below for each of the three zones. Althouah K
is symmetric. Eq. (6-61 ) is not used in order to illustrate the more general equations.

Zone 1:

©
~8j 4 2"
© ©
4 S 4 =
_2 4 S_

Note: L u = 1 and Uu = 8
Zone 2:

© © ©
rJ
4
4
8
2
4 =
LU 0.5 1
'8 4
6
2 4 8_

Note: L 2! = = 0.5. U l2 = 4. L 22 — 1. and 6/^ = 8 - (0.5)(4) = 6.

Zone 3:

© © ©
2 I 8 4 2
4 0.5 1 6 3
2 4 8_ 0.25 0.5 ]_ 6_

Note: I,, = * = 0.25. C/, 3 = 2. Z. J; = [4-(0.25)(4)|/6 = 0.5. U23 =

4-{0.5)(2) = 3. L 33 = 1. and d/33 - 8-(0.25)(2) - (0.5)(3) = 6.
It is concluded that

1 0 0 "s 4 2
0.5 1 0 and U = 0 6 3
0.25 0.5 1
_ 0 0 6_

The reader should verify that the product of L and U (in this order) does indeed
gi%-e the original K matrix. The L matrix could also be obtained with Eq. (6-61)
since K is symmetric. EB

Forward Elimination and Backward Substitution

Let us now reconsider the system of algebraic equations Ka = f and try' to obtain
the solution for the vector a by taking advantage of the material in the previous
section. Beginning with Ka = f and using K = LU. we may write
LUa = f ( 6 - 62 )
274 PARAMETER AND SHAM FUNCTIONS. INTEGRATION I ORMULAS

Denoting Ua as the vector z, we have, in eflect, two systems of equations:

Lz = f {6-63}

and

Ua = z (6-64)

Let us write these out explicitly for the case of a 3 x 3 K matrix in order to gain
some insight into the general algorithm to be presented. Equation (6-63) implies

= f,

L 2i z, + z2 = f2 (6-65)

L„ri + L n z, + 2, = f,
 ACTIVE ZONE EQUATION SOLVER 275

and

a» ~ jr (6-70a)
u nn

a, = .
i = n - 1 , « - 2, . . . , 1 (6-70b)
V ll

For rather obvious reasons, the steps in Eqs. (6-69) and (6-70) are referred to as
the forward elimination and backward substitution steps, respectively. A numerical
example should help to clarify the procedure.

Example 6-12
Solve the following system of algebraic equations by using forward elimination and
backward substitution:

Sfl| + 4o j + 2a — 34

4 a, + 8 a2 + 4 = 56

2 + 4 a2 4- 8a 3 = 46

Solution

Since the matrix K is the same as that in Example 6-11, the results of the triangular
decomposition from that example may be used. The forward elimination step gives

- 34

:2 = 56 - (0.5)(34) = 39

r3 = 46 - (0.25)(34) - (0.5)(39) = 18

and the backward substitution step gives

n3 = '% = 3

39 - (3)(3)
a-> = - c
->

34 - (2)(3) - (4)(5)

Therefore, the solution to the original system of equations is given by

a = 5
3

A quick check by direct substitution reveals that this is indeed the solution.
276 PARAMDTLR AND SIIAPL FUNCTIONS. INTLGRATION FORMULAS

When the K matrix is very large, the triangular decomposition requires much
more execution time than the forward elimination and backward substitution steps.
Therefore, once the triangular decomposition step has been performed, the system
Ka = f may be solved for several difierent fs rather economically. This capability
is very important tn all practical finite clement programs and is referred to as
tesolution capability
Next we turn to the question of when the active zone solution method is
guaranteed to give valid results In order to answer such a question, we must first

define what is meant by a symmetric, positive (or negative) definite matrix.

Classification of Symmetric Real Matrices [8]

First, the principal Illinois of a matrix K must be defined If K is an n x « matrix,

then the n principal minors of K are denoted as A,, A2 , . , A„ and are given by

and so forth, up to A„ = det K

With the help of the principal minors, symmetric real matrices can then be
classified as (l) positive definite, (2) negative definite, (3) positive semidefimte,
(4) negative semidefimte. and (5) indefinite, as summarized in Table 6-2 Note that
a positive definite, symmetric real matrix is one whose principal minors are all
strictly positive On the other hand, note how the principal minors alternate between
strictly negative and strictly positive for negative definite matrices
Fortunately, the finite element formulations in this text result in positive (or
negative) definite K
matrices, except in a few cases as noted later. This is fortunate
because the active zone equation solver presents no unusual difficulties in these
cases and may be used as presented When the K matrix is unsynimetnc (see
Chapter 8) or indefinite, some additional checks may be necessary to ensure that
the equations can be solved In these cases, certain rows and columns of may K

Table 6-2 Classification of Symmetric Real Matrices*

Classification Condi (ions

of symmetric real matrix on the principal minors

Positive definite A, > 0. A. > 0. . A„ >0

Negative definite A, < 0, A-> > 0. A, < 0. etc
Positive semidefimte A, a 0. A, >0. . A„ s0
Negative semidefimte A| s0,i-> 0. A, < 0. etc
Indefinite None of the above patterns

*Fram Brogan, VV L , Modern Control Theon Quantum

, Publishers, New York. 1 97-1
 ACTIVE ZONE EQUATION SOLVER 277

have to be interchanged to effect a solution. Treatment of such cases is beyond the

scope of this text; the interested reader may wish to consult references [9-1 1], In
what follows it is tacitly assumed that interchanges of rows and columns are not
necessary.

Storage Considerations

It should be recalled that the assemblage stiffness matrix K is generally banded

and symmetric. This is quite significant because it becomes possible to reduce
significantly the storage requirements of such a matrix. If it is symmetric, then the
entries below the main diagonal need not be stored. This alone reduces the storage
requirement by a factor of nearly one-half. If the matrix is banded or if there are

many leading zeros in the upper triangular portion of the matrix, then the storage
requirements are reduced further.
Consider, for example, the 7 x 7 matrix K that is given by

*,, K \2 *, 3 0 0 0 0
K->-> 0 0 K-, 6 0
*33 K 34 0 0
KiA KA5 0 0 (6-72)

*ss k57
symmetric AT66 A"67

*77 _
The half-bandwidth bH in this case is readily seen to be five, as dictated by the

second row. One way to store the coefficients in this matrix is by the so-called
banded storage method.

Banded Storage Method

If an n x n matrix K has a half-bandwidth , then the nonzero coefficients (and
the imbedded zeros by row) are stored in a compacted n X b n matrix. The matrix
in Eq. (6-72), for example, would be stored in a 7 x 5 matrix as follows:

*«.
*22
*33
*44
*S 5
*66
*77

Note how each row from the diagonal entry is slid to the left, with imbedded zeros
in the row also necessarily being stored. If the original matrix K is not banded, no

reduction in storage requirements occurs. Nevertheless, in this example, the original

278 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

matrix would require 49 (or 7 x 7) storage locations, whereas the banded storage
would require 35 (or 7x5) locations. A more efficient storage method is presented
next.

Skyline Storage Method

It is more efficient to store the K matrix in a column vector form as illustrated
far
below. The method is most easily understood by way of an example. Reconsider
the 7x7 matrix given by Eq (6-72) Let us store the nonzero coefficients (and
imbedded zeros by column) in a column vector A(i) as shown in Fig. 6-19. Note
how each column in the K matrix is stacked in A(i). Note further that leading zeros
in the column need not be stored Zeros that are imbedded in a column, however,
must be stored.

/ JD1AGO)

2 3

3 6

4 8

5 10

6 15

7 18

Figure 6-19 Skyline storage method for K matrix m Eq (6-72)

 ACTIVE ZONE EQUATION SOLVER 279

The JDIAG(/) array in Fig. 6-19 is used to find the location of each nonzero
(or imbedded zero) entry in the column vector. Physically, for a given value of
j
(1 j s= n ), JDIAG(/) represents the number of entries stored in A(i) up to and
including the diagonal entry (he., K ;j
). For example, K55 is the tenth entry in A(i)
and, hence, JDIAG(5) is 10. The location of Kmn in A(i) is most easily found by
using the following simple formula:

i = JDIAG(n) - |(n - m)\ (6-73)

For example, let us locate the position in the A(i) array that is reserved for K36 .

Here n is 6 and m is 3. Therefore, i = JDIAG(6) - |(6 - 3)| = 15 - 3 = 12.

The reader should note in Fig. 6-19 that K36 is indeed the twelfth entry in the A(i)
array. Note: n should always be the larger subscript (not necessarily the second
subscript).
This same method can be used to store unsymmetric matrices with a symmetric
profile. In this case, the lower coefficients are stored by row, with unity assigned
to each diagonal entry in another column vector, e.g., the C(i') array. Because the
profile is assumed to be symmetric, the same JDIAG(/) array may be used. For
obvious reasons, the JDIAG(/) array is frequently referred to as the pointer array
of the diagonal pivots, or simply the pointer array.

Computer Implementation

Subroutines are readily available that incorporate the active zone solution and skyline
storage methods. Two such FORTRAN subroutines are introduced in this section
subroutine ACTCOL symmetric matrices and subroutine UACTCL for unsym-
for
metric matrices [12]. These subroutines are given in Appendix C. In Chapter 8, it
will be seen how an unsymmetric K matrix may arise. In such cases, subroutine
UACTCL must be used. The subprogram header lines are given by

SUBROUTINE ACTCOL (A, B, JDIAG, NEQ, AFAC, BACK)

and

SUBROUTINE UACTCL (A, C, B, JDIAG, NEQ, AFAC, BACK)

where the parameters A, B, C, JDIAG, NEQ, AFAC, and BACK are defined in

Table 6-3. In subroutine UACTCL, the diagonal entries in the C array (see Table
in the A array). The
6-3) are set to unity (the actual diagonal entries are stored
original reference for further details on the use
reader may wish to consult the [12]
subroutines require
of these subroutines. It should also be remarked that these
which performs a vector dot product.
function DOT (also given in Appendix C).
280 PARAMETER AND SHAPE FUNCTIONS INTEGRATION FORMULAS

Table 6-3 Variables Used in Equation Solution Subprograms ACTCOL and UACTCL*
A(NAD) Upper triangular coefficients when called, replaced by If on return

B(NEQ) Right-hand side vector at call (i e . f). solution vector (i e., a) upon
return to calling program

C(NAD) Lower triangular coefficients at call. L at return (used in UACTCL

only)

JDIAG(NEQ) Pointer array to determine location in A or C of diagonal pivots

NEQ Number of equations to be solved

NAD Length of A or C arrays equal to JDIAG(NEQ)

AFAC Logical variable if true, triangular decomposition performed

BACK Logical variable if true, forward elimination and backward

substitution performed

Copyright© 1977 McGraw Hill (UK) London From Zicnkicwiez O C The finite tlement Method
3d cd Reproduced by permission of the publisher

6-9 REMARKS

In this chapter, the compatibility and completeness requirements have been intro-
duced If the assumed form of the parameter function satisfies these requirements,
convergence is guaranteed as the elements are reduced in size in some regular
fashion
The one-dimensional lineal element was reintroduced along with two new one-
dtmensional normalized coordinates the serendipity coordinate and the length co-
ordinates. The shape functions for this element were derived in terms of all three
coordinate systems
Among the two-dimensional elements considered in this chapter were the three-
node triangular element and the four node rectangular element The shape functions
were derived for the triangular element m terms of the global coordinates Then
the so-called area coordinates were introduced, and each shape function was seen
to be equal to the corresponding area coordinate This fact has important implications
m the evaluation of the resulting integrals because special integration formulas may
be used The shape functions for the rectangular element were given in terms of
the serendipity coordinates
a similar fashion, the shape functions for two three-dimensional elements
In
were presented For the four-node tetrahedral element the shape functions were
derived in terms of the global coordinates using the standard procedure Volume
coordinates, which facilitate the evaluation of the volume integrals that will naturally
 REMARKS 281

arise, were introduced. The eight-comered brick element was introduced along with
its associated shape functions in the so-called serendipity coordinates.
A special type of problem was discussed that is not strictly one, two, or three
dimensional. These problems are referrred to as axisymmetric problems and must
meet two basic conditions: both the geometry and the external effects must be
axisymmetric. It was shown how the two-dimensional elements may be used in
such situations. This point will be demonstrated in more detail in Chapters 7 and
8 .

As mentioned above, special integration formulas may be used when line, area,
and volume integrals need to be evaluated. Such integrals may be readily evaluated
if the lineal, triangular, and tetrahedral elements are used with the shape functions
given in terms of the length, area, and volume coordinates. Examples were given,
but the reader may expect to see many more applications of these formulas in the
next several chapters.
The chapter was concluded with a more efficient method for the solution of
the system of algebraic equations — the active zone equation solver. This method
may be divided into three steps: (1) triangular decomposition, (2) forward elimi-
nation, and (3) backward substitution. Up to now, we have been using the relatively
inefficient matrix inversion method. The active zone method takes advantage of
the banded (and symmetric) nature of the assemblage system equations. Two special
FORTRAN subroutines were introduced that very effectively perform the necessary
tasks, thereby paving the way to our obtaining solutions to larger, more practical
problems.
In Chapters 7 and 8, specific application areas are considered: stress analysis
in Chapter 7 and thermal and fluid flow analysis in Chapter 8. These two chapters

will pull together everything that has been covered up to this point. The reader may
wish to review Chapter 5 before tackling the next chapter.

REFERENCES

1. Zienkiewicz. O. C.. The Finite Element Method. McGraw-Hill (UK). London. 1977.
pp. 287-288.
2. Huebner. K. H., The Finite Element Method for Engineers. Wiley. New' York. 1975.
pp. 79-81.
3. Desai, C. S., Elementary Finite Element Method. Prentice-Hall, Englewood Cliffs.

N.J., 1979, pp. 43-45.

4. Martin, H. C., and G. F. Carey. Introduction to Finite Element Analysts. Theon and
Application, McGraw-Hill. New
York, pp. 52-53.
5. Micropaedia. "Serendib.” Eiu \clopaedta Britanmca. vol IX. p 67. 1981.

6. Zienkiewicz. O. C., The Finite Element Method. McGraw-Hill (UK). London. 1977.
pp. 141-144.
1977.
7. Zienkiewicz. O. C.. The Finite Element Method. McGraw-Hill (UK). London.
p. 719.
8. Brogan, W. L.. Modern Control Theory. Quantum Publishers, New 4ork. 1974. p

134?
282 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

9.

Wilkinson, J H ,
and C. Reinsch, Linear Algebra Handbook for Automatic Compu-
tation II, Springer-Verlag. New York, 197 1.
10. Ralston, A A First Course in Numerical Analyses,
, McGraw-Hill, New York, 1965
11. Fox, L An
, Introduction to Numerical Linear Algebra, Oxford University Press, New
York, 1965
12. Zienkiewicz, O C , The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp 740-746.

PROBLEMS

6-1 In amass transfer model, we typically try to find the concentration distribution of
some chemical species in a fluid or a solid What specific name could be suggested
for the parameter funciton in this case?

6-2 In electromagnetics, what could be suggested for specific name(s) of the parameter
function(s) 7

6-3 Show that the three-node triangular element satisfies the compatibility requirement

if a parameter function of the form «J>

= c, + + Cjy is assumed for C°-contmuous
problems

6-4 Show that the four-node tetrahedral element satisfies the compatibility requirement

if a parameter function of the form <J>

= cj + c*r + Cjy + c4 z is assumed for C°-
continuous problems

6-5 Show that the three-node triangular element satisfies the completeness requirement
if a parameter function of the form <J>
= c, + Cjt + fjy is assumed for C°-continuous
problems

6-6 Show that the four-node tetrahedral element satisfies the completeness requirement
if a parameter function of the form = c, + <J>
+ cy + c4 z is assumed for C°-
continuous problems

6-7 In Problem 6-5, which term (or terms) represents the rigid-body mode and which
term (or terms) represents the constant strain 9

6-8 In Problem 6-6, which term (or terms) represents the rigid-body mode and which
ttm (os terms) represents the constant strain9

6-9 Show that the shape function for node j for the two node lineal element (with C°-
continuity) is given by Eq (6-1 lb) m terms of the serendipity coordinate r

6-10 three-
With the help of Eqs (6-19) and (6-20), show that the shape functions for the
node triangular element are given by Eqs (6-21) for C° continuous problems Hint
See Sec 2-7 for the method to be used in obtaining the matrix inverse Denote the
entries in the inverted matrix as m„, m i2 ,
etc

satisfy
6-1 1 Show that the shape functions given by Eqs (6 21) for the triangular element
the three properties delineated in Sec 5 6

6-12 Show the lines of constant L, and Lt on the three-node triangular element
for Lt -
0, 'A, 'A, Vi, and 1 and forL* = 0, /*, etc.
 PROBLEMS 283

6-13 From the definition of Ly given by Eq. (6-22b), show that L} is the same as the shape
function A', given in Eqs. (6-21). and hence Aj = Ly
6-14 From the definition of Lk given by Eq. (6-22c), show that is the same as the shape
function A\ given in Eqs. (6-21). and hence N =
k L*.

6-15 Determine the explicit form of the shape function for node i for the rectangular
element in terms of the global coordinates (.v.v) and the nodal coordinates. Hint: See
Eqs. (6-29) and (6-30).

6-16 Repeat Problem 6-15 for node j.

6-17 Repeat Problem 6-15 for node k.

6-18 Repeat Problem 6-15 for node in.

6-19 For the four-node rectangular C°-continuous element.

a. Plot N, on face mi.

b. Plot A', on face ij.
c. Plot N, along the straight line connecting nodes i and k.

6-20 For the four-node rectangular C°-continuous element,

a. Plot Nj on face ij.

b. Plot Nj on face jk.

c. Plot A'j along the straight line connecting nodes j and m.

6-21 For the four-node rectangular C°-continuous element.

a. Plot A',on face jk.

b. Plot Aj on face km.
C. Plot A t along
r
the straight line connecting nodes k and i.

6-22 For the four-node rectangular C°-continuous element,

a. Plot A'm on face mi.

b. Plot Nm on face km.
c. Plot A'm along the straight line connecting nodes in and j.

6-23 Evaluate the following integral for the four-node rectangular element:

fA A’~h dx dy

where A denotes the area of the rectangle and h is a constant.

6-24 Evaluate the following integral for the four-node rectangular element:

I 'V'V /' dx d>

where A denotes the area of the rectangle and h is a constant.

6-25 Evaluate the following integral for the four-node rectangular element:

fA AjNJi dx dy

where A is the area of the rectangle and h is a constant.

284 PARAMETER AND SHAPE FUNCTIONS. INTEGRATION FORMULAS

6-26 For the four-node tetrahedral element shown in Fig 6-12, give all twelve possible
element definitions in terms of the four global node numbers

6-27 Determine expressions similar to those m Eq (6-40c) for m v m J2

, . and niy, for

the four-node tetrahedral element

6-28 Determine expressions similar to those in Eq. (6-40e) for m 13 m2S m3i
, . , and «t4J for

the four-node tetrahedral element

6-29 Determine expressions similar to those

the four-node tetrahedral clement
in Eq. (6-40e) for m, 4 . »ii4 , and % for

6-30 Show that N, evaluates to unity at node i for the four-node tetrahedral element and
that N, is zero if evaluated at the coordinates of node j, node A. and node m
6-31 From Problem 6 30 and Eqs (6-40), give a plausible argument for why N, must
decrease linearly from unity at node i to zero at each of the remaining nodes

6-32 Show that 2 N$ = I for the four shape functions given by Eqs (6-40) for the four-
node tetrahedral clement

6-33 With the help of the definition of L given by Eq (6-41), show that L} is the same
as the shape function N, given in Eqs (6 40) for the four-node tetrahedral element

6-34 With the help of the definition of Lk given by Eq (6-4|), show same
that Lk is the

as the shape function given in Eqs (6 40) for the four-node tetrahedral element

6-35 With the help of the definition of L m given by Eq (6-41 ). show that L,„ is the same
as the shape function \'m given in Eqs (6 40) for the four node tetrahedral element

6-36 What two conditions must a problem satisfy in order to be considered axisymmetnc ’

6-37 For the axisymmetnc body shown in Fig 6- 1 7(a) and discretized in Fig 6-17(b).
what are the shape functions in terms of the radial and axial coordinates r and :

6-38 Evaluate the following line integrals for the two-node lineal element

a. fiN*dl b. J N,Nt dl
i

c. ftN*N]dl d. // NfNj dl

6-39 Evaluate the following line integrals for the two-node lineal dement

a. fi N,tdl b. J,N,Nt rdl

6-40 Evaluate the following area integrals for the three-node triangular element

a. IA N.N, dA b. fA n; dA
c -
Ja N,N, dA d. Ja NlN';Nt dA
6-41 Evaluate the following area integrals for the three-node triangular etement

a.
JA N,xdA b. fA NjV dA

6-42 Evaluate the following volume integrals for the four-node tetrahedrat element

a. J v N,dV b. / v N,Nt dV
c. JV N,N} Ni Nm dV d. JV NfNi dV
286 PARAMETER AND SHAPE FUNCTIONS, INTEGRATION FORMULAS

6-51 Consider the following symmetric matrix.

0~~
20 10 5 0 0
40 20 10 0 0
40 20 10 0
40 20 10
symmetric 40 20
20 __

a. What is the half-bandwidth?

b. How many storage locations are required if this matrix is stored by the banded
storage method9 What is the resulting banded storage matrix?
c. How many storage locations are required if this matrix is storedby the skyline
storage method (i e , in column-vector form)'' What is the column vector [i e ,

the4(0 array in ACTCOL1 9

d. What is the JDIAG(/) array for part (c) 9

6-52 Consider the following symmetric matrix

50 20 10 0 0 0 0 O'
75 50 0 25 0 0 0
80 20 0 50 0 0
30 0 0 0 0
40 20 0 0
symmetric 50 5 10
2 5

1 __

a. What is the half-bandwidth 9

b. How many storage locations are required if this matrix is stored by the banded
storage method 9 What is the resulting banded storage matrix 9
c. How many storage locations are required if this matrix is by
stored the skyline

storage method (i e in column-vector form ) 9 What

,
is the column vector [l e ,

the 4(0 anay in ACTCOL] 9

d. What is the JDIAG(/) array for part (c) 9

6-53 Consider the unsymmetric matrix

— 50 20 0 4 O 0 0 0 0“
30 100 80 20 5 0 0 0 0
5 20 130 40 20 10 0 0 0
0 90 100 150 100 80 40 0 0
K = 0 50 75 80 180 100 60 40 0
0 0 0 90 150 200 90 85 40
0 0 O 10 75 80 210 215 35
0 0 0 0 5 50 100 220 190
_0 0 0 0 0 6 10 55 160

a. What is the half-bandwidth 9

b. What are the 4(0 and C(i) arrays for use in subroutine UACTCL9
c. What is the JDIAG(/) array for part (b) 9
 Stress Analysis

7-1 INTRODUCTION

In this chapter, the finite element method is applied to several different classes of
problems in stress analysis. The chapter begins with two-dimensional problems,
since one-dimensional stress analysis was introduced in Chapter 5. This is followed
by the finite element formulations to axisymmetric and three-dimensional problems.
In each case, expressions for the element stiffness matrix and element nodal force
vectors are derived from Eqs. (5-87) to (5-92). However, an alternate approach is

taken in the study of the beam model. More specifically, the finite element char-
acteristics for the beam are derived with the help of the Galerkin method (one of
the weighted-residual methods from Chapter 4). Appropriate working expressions
are then derived for the element stiffness matrix and element nodal force vectors
for the beam element. The concept of substructuring is described, which makes the
development of extremely large finite element models feasible. The chapter is
concluded with a brief description of a two-dimensional stress analysis program.

7-2 TWO-DIMENSIONAL STRESS ANALYSIS

In this section, the plane stress and plane strain models are defined. The finite

element characteristics are derived from Eqs. (5-87) to (5-92) for the case of a
specific two-dimensional element. Each of the basic steps in the FEM solution to

such problems is discussed.

287
288 STRESS ANALYSIS

Definitions of Plane Stress and Plane Strain

Plane stress and plane strain models are both two-dimensional idealizations of three-
dimensional problems The situations in which each is appropriate are discussed
below

Plane Stress

The plane model is appropriate when a thin plate is loaded uniformly across
stress
its and in a direction parallel to the lateral surfaces of the plate as shown
thickness t

in Fig 7-1 Note that the thickness t of the plate need not be constant The loading
.

around the plate periphery may be a result of tensile, compressive, or shear forces
The definition of plane stress implies

O’- = 0 O’, = 0 and a,. = 0 (7-1)

both within and on the faces of the plate Also, the body force b_ must be zero
The body forces bx and b s may each be functions of v and \ Therefore, there are
three nonzero stress components o„, cr„, and which are functions of* and v ,

and which remain constant through the thickness of the plate (in the z direction) at

any point It should be pointed out that the strain in the z direction, e~, is not
necessarily zero for the state of plane stress |see Eq (7-16)] In addition, the body
may have self-strains e„ (1 ,
e l(U , and e..,, possibly because of a temperature
change A7" Also, initial residual stresses a Mo , a,,,,, and o’,,,, may exist

Figure 7-1 Plane stress (all loads are m the *-> plane only).
 two-dimensional stress analysis
289

In summary, a problem may be classified as plane stress when the

body to be
analyzed is a relatively thin plate loaded only in
the plane of the plate.

Plane Strain

The plain strain model is appropriate when the normal strain in the r direction,
and the shear strains, e,. and e,.. may be assumed to be zero, or
e~ ~ 0 e ..- = 0 and e,_. = 0 (7-2(

This situation is most likely to arise when

a long prismatic member of constant
ctoss section held between two smooth fixed rigid planes
is
as shown in Fie. 7-2.
Although this is not a very' practical situation, St. Venant’s
principle [1] allows us
to use the plane strain model in the regions
far from the ends that may not meet
the above conditions.
Note the coordinate system shown in Fig. 7-2. External forces
may have only
-V and
\ components and may be a
function of ,v and v only. All cross sections are
expected to have the same deformations. Body forces. b and
t
b per unit volume x ,

may each be functions of a and y only, and in the z direction we must have b. ~
0.
As in plane stress, the body may have self-strains e
no e„„, and possibly because ,

of a temperature change AT and a uo and cr.. Note that

initial stresses cr
rill ,
cr lUj , ,
(1
.

in general a., is not necessarily zero in plane strain and, therefore, we must allow
for the possibility of nonzero rr..„ [see Eq. (7-21)].

The Shape Function Matrix N

Before the shape function matrix can be derived, the two-dimensional region to be
analyzed must be discretized into a suitable number of elements. From Chapter 6,
two elements in particular can be used in this situation: the three-node triangular
element and the four-node rectangular element. Because the triangular element can
accommodate practically all two-dimensional regions, let us restrict the finite ele-
ment formulation to only these elements, as shown in Fig. 7-3. In both plane stress
and plane strain, each node has two degrees of freedom: the x and y components

End view
Smooth fixed Smooth fixed
rigid plane rigid plane

Figure 7-2 Plane strain (all loads are in the x-y plane only).
290 STRESS ANALYSIS

Figure 7-3 Typical two-dimensional region discretized into triangular elements.

of the displacements Since each element has three nodes and each node has two
degrees of freedom, the shape function matrix N is given by Eq. (5-61), which is
repeated here for easy reference

\N, 0 : N, 0 i N, 0 1
(5-611

[
0 /V, |
0 NJ 1
0

The three shape functions are given by Eqs (6-2 ) The shape function matrix N
1

is, therefore, the same for plane stress and plane strain and is given by Eqs.
(5-61) and (6-21)

The Strain-Nodal Displacement Matrix B

Recall that the strain-nodal displacement matrix B is given by Eq (5-76) or B = LN,

where L is the linear operator matrix. It should further be recalled from Eq. (5-75)
that e = I,u, where

t - |e„ t„ I 7 ' 31

and
r 17-41
u = |« v]

where « and v are the x and y components of the displacement field. The strain-
are
displacement (or kinematic) relationships for both plane stress and plane strain
given by Eq. (5-14) for small deformations

= *! *!
+ (5-14)
" dx " dy " Hy Bx
 TWO-DIMENSIONAL STRESS ANALYSIS 291

It follows that the linear operator matrix L is given by

B_
0
Bx

L = 0
3_
(7-5)
By

a_

By dx

The matrix B is then easily determined to be

1i§N o &dx
0 ^1^ o 1

o
By
o ^
By
0
„ —
BN,
By

BN, BN, BN, BN, Mi My

By dx By dx By 3x

But the shape functions are given by Eqs. (6-21), so the matrix B is seen to be
composed of the w ’s and is given by
y
"

;«2i 0 111 22 0 m 23 0
B = 0 »hi o’ m 32 6 m33
_'«31 m2 1
m 32 m22 ">33 ' ?!
23_

Since the m,/ s are known functions of the nodal coordinates [see Eq. (6-2 Id)], the
B matrix is determined. The fact that this matrix is composed of only constant
entries (for the three-node triangular element) has an important implication when
the integrals defining the finite element characteristics are to be evaluated, as will
be seen shortly.

The Constitutive Relationship

Recall that the constitutive relationship for a linear, elastic material is given by Eq.
(5-24). which is repeated here for convenience:

ct = D(e - €0 ) + a0 (5-24)

The material property matrix D depends on whether the material is isotropic or

anisotropic. Only the isotropic case is considered here. The interested reader should
consult Zienkiewicz’s book [2] for the anisotropic form of the material property
matrix. The matrix and the vector e 0 are different for plane stress and plane
D
strain as indicated below.
292 STRESS ANALYSIS

Plane Stress

For an isotropic material in plane stress, we have by the generalized Hooke’s law.

o ir M-<T"
(7-8a)
£

(7-8b)

1
(7*8c)

where p is Poisson’s ratio, £ is the modulus of elasticity, and G is the shear modulus
(one of the Lame constants) related to p and £ by

£
G = (7-9)
2(1 + p)

If we solve Eqs (7-8) for the three stresses and recognize that the initial stresses
not shown in Eqs (7-8) are simply additive to u n , etc we get ,

£
- e«a) + P(*„ “ + (7-1 0a)
1 - p il(e« 2 ,)]

= E
~ e„„) + (e„ - + aVTO (7-1 Ob)
- p ilp(e„
or,, £„„)]
z
1

(7-1 Oc)
>

Equations (7-10) may now be cast into the form of Eq (5-24) by defining the
following [in addition to e from Eq (7-3)1

r
a = [a„ a,, o\, l
(7-11)

<*o = K tu a\,„ o\ 10 ) r (7-12)

eo = U„„ r (7-13)
«,» «» w]
and
rl p 0 “I

If the self-strain vector e 0 is a result of a temperature change A7\ then « 0 IS

given by

AT (7-15)
e0 = [a, A7’ a, 0]
:
 TWO-DIMENSIONAL STRESS ANALYSIS 293

In Eq. (7-15), a, is the coefficient of thermal expansion. Note that no shear strain
is induced by a thermal excursion in an isotropic material. This is not the case in
anisotropic materials.
It is noted that e_- is not necessarih zero and is given by

e ----
= “«r u ~ «-„> + (7-16)

where e ;;„ is taken to be a,lT in the case of a thermal strain. However, this strain
component need not be included in Eq. (7-3): e; . is simply determined from Eq.
(7-16) after cr„ and a n have been found.

Plane Strain

The material property matrix D for the case of plane strain proves to be different
from that for plane stress. For a linear, elastic isotropic material in plane strain,

the appropriate form of Hooke’s law is

a tT = (2 G + \)(e„ - e lro ) + \(e n - e„ 0 ) + cr„0 (7-17a)

cr„ = \(e a - e no ) + (2 G + L)(e,, - e„„) + o- n„ (7-17b)

o-.n = G(e„ - e nX) ) 4- CT ntl (7-1 7c)

where the shear modulus G is given by Eq. (7-9) and X is another one of the Lame
constants defined by

\ = (7-18)
(1 -f |x)(l - 2fc)

If <t. e. cr 0 . and e 0 are defined as above, it follows from Eqs. (7-17) that the
material property matrix D is given by

1 - M- E 0
£
D = -
(J- 1 - p. 0 (7-19)
(1 + |t)(l 2(x) I - 2n
0 0

Therefore, the material propern matrix for an isotropic material in plane strain is

known in terms of the material constants £ and p.-

If the initial or self-stramx are a result of a temperature change \T. wc have

r
Cq — [Cl ^ p)a, \T (1 - p)a, IT 0] (7-20)

Note the additional factor 1 -

p. w hich was not present for the case of plane stress
[see Eq. (7-15)). In Eq. (7-20). a, is the coefficient of thermal expansion. Again
it is pointed out that the shear
strains as a result of a temperature change cannot

develop in an isotropic material.

294 STRESS ANALYSIS

It should be noted that tr„ is not necessarily zero for the case of plane strain
and is given by

a.. — |a(ct„ + ff„) + cr-^, — £e.-0 (7-21}

where is taken to be a, AT in the case of a thermal strain.

This completes the determination of the constitutive relationship for the cases
of plane stress and plane strain. It is emphasized that only linear, elastic, isotropic

materials are considered here The reader is referred to Zienkiewicz’s book [2] for

the case of anisotropic materials.

The Element Stiffness Matrix

The element stiffness matrix is given by Eq (5-87) for all problems in stress analysis
and is repeated here for convenience.

Kf = clV 15-87)

Both B and D are composed only of constants Moreover, the elemental volume
dV is given by

dV = / dx dy (7-22)

where / is the thickness of the plate for plane stress or the length of the long
prismatic member in plane strain The thickness t is frequently taken to be unity
for the plane strain case Therefore, Eq (5-87) becomes
r
K' = B DB fA ,t dx dy (7-23)

For reasonably smatl elements, the thickness t may be taken as a constant average
value (perhaps the value at the centroid or the average of the three values at the

nodes). It then follows that t may be taken through the integral We are left with

/ dx dy, which is really the area A of the triangle and is given by Eq (6-2 le)

Therefore, the element stiffness matrix is given by

K' = B r DBrA (7-24)

It should be noted that K' is a 6 x 6 matrix because B T is 6 x 3, D is 3 x 3,

and Bis 3 x 6 This is consistent with the fact that there are six nodal displacements
(two at each of the three nodes of the triangle)

Example 7-1

Consider the element shown in Fig. 7-4. Determine the element stiffness matrix if
nodes i, j, k are 2, 1,3, respectively. The material is steel, which has an elastic
modulus of 30 x 10s psi and a Poisson’s ratio of 0.3. The thickness may be
assumed to be 0.25 in (constant) The nodal coordinates (in inches) are shown on
the figure. All loads (not shown) may be assumed to be in the plane of the element,
hence the plane stress case applies.
 TWO-DIMENSIONAL STRESS ANALYSIS 295

Solution

From Eq. (7-7), we see that we need six of the nine m,/ s, which are easily calculated
as shown below. The area A of the element is needed in the calculation of the m ,’s
and is determined from Eq. (6-2 le) as follows:

1 4 2
A '/’ det 1 0 2 = 4 in.
2

1 0 0

This result checks with the value of the area from the simple formula that is given
by the product of one-half the base and height. The in ,/ s are calculated as follows:

>’j “ )\ Vi - y3 2-0 =
m 2i = 0.25 in.-'
2A 2A 2(4)

-Vi
- Xj -V3 - -Vl o 1 0
m3 =,
= 0
2A 2A 2(4)

^ n - y, v3 — y2 0-2 = -1
nzoo -0.25 in.
2A 2A 2(4)

~ xk x2 - 1 o
x, .r 3
1 = -1
m 32 = 0.50 in.
2A 2A 2(4)

}\ - )'j
>2
— V 2-2 =
™23 = - I
0
2A 2A 2(4)

Xj ~ x , -Vi
~ x2 0-4 = -0.50
m 33 = in.' 1

2A 2A 2(4)

It then follows that the B matrix is given by

'0.25 0 -0.25 0 0 0
B = 0 0 0 0.50 0 -0.50
0 0.25 0.50 -0.25 -0.50 0
296 STRESS ANALYSIS

The material property matrix D (for plane stress) is given by

33 0 9 89 0
D = 9.89 33 0 0 x 10
f’
lbf/in.-

0 0 115 .

The reader maj now verify that product B r DBM gives the following stiffness
malax for this element.

2 06 0 -2 06 1 24 0 — 1.24"
0 0 72 1 44 -0 72 - 1 44 0
-206 l 44 4 94 -2 68 —2 8S 1.24
1 24 -0 72 -2 68 S 97 1 44 -S 25
0 -I 44 -2 88 1 44 2 8S 0
-1 24 0 1 24 -8 25 0 S 25.

The sjmmetiy of K r
should be noted The 2x2 submatrices, however, are not
necessanl) svmmetric Partitioning the element stiffness matrix in this wav greatly

facilitates the assemblage step ®

The Element Nodal Force Vectors

Each of the five nodal force vectors given b> Eqs (5-88) to (5-92) ma> now be

evaluated for plane stress and plane strain problems.

Self-Strain

The element nodal force vector as a result of the self-strains is given by Eq

(5-88), or

fto = JV'B r De 0 d\' = /^»B 7 De 0 r dx d\ (7-25)

where, for thermal strains. e0 is given by Eq (7-15) for plane stress and bv Eq
(7-20) for plane strain If we limit the present discussion to thermal strains and take
the coefficient of thermal expansion a, and the temperature change as constants
within an element (and hence constant €«). then Eq (7-25) becomes

(7-261
= B r Dc 0 rt
 TWO-DIMENSIONAL STRESS ANALYSIS

Note that the thickness / was also assumed to be constant. Whenever a property
varies over the element, it may be evaluated at the centroid and subsequently treated
as a constant. More sophisticated integration schemes arc discussed in Chapter 9.
r
Note that f.f, is6 x 1 because is 6 x
15 3. D is 3 X 3. and e 0 is 3 x 1. It is

emphasized that 15 and I) are known and arc given by Eqs. (7-7) and (7-14) [or
Eq. (7-19)]. respectively.

Example 7-2

For the element in Example 7- 1 . determine the vector ff if the coefficient of thermal(;

6
expansion is 6.0 x 10 in./in.-°F and the temperature increases by 150°F.

Solution

First we calculate e 0 from Eq. (7-15) since the plane stress case is applicable:

6 )(150)'
'a, AT '(6.0 x 10 900*
€0 — «, AT (6.0 x 1 0
r>
)( 150) = 900 x 10~ 6 in. /in.

0 o . 0 .

Then, using the results for the 15 and D matrices from Example 7-1. we have

0.25 0 0
0 0 0.25 '900'
33.0 9.89 0
-0.25 0 0.50
f.f,
= 9. 89 33.0 0 900 (0.25K4.0)
0 0.50 -0.25
0 0 11.5 0
0 0 -0.50
0 -0.50 0

or

9.650
0„._
”-9*650
f' = Ibf
19.300
o'""
- 19.300

Prestresses

The element nodal force vector as a result of the prestresses a uo , cr uo . and cr U(1 is

easily determined from Eq. (5-89) as follows:

PCTll
T
= / v<- B 7 cr 0 dV = fA <B tr0 t dx dy = B T cr0 tA (7-27)

Note that the prestresses are assumed to be constant within the element (or the
values at the centroid are used). Once again it is noted that is of size 6x1.
298 STRESS ANALYSIS

Body Forces
It follows from Eq. (5-90) that the nodal force vector as a result of the body forces
bx and f>
v is given by

” N, 0
“
0 N,
N 0
= JV # N T b dV }
17-28)
0 Nj
N* 0
0 N _ k

Noting that each shape function is given by its respective area coordinate [see Eqs.
(6-25)], it follows that

/ L,bx t dx dy
/ L,by t dx dy
/ Ljbx t dx dy {7-29)
S Ljbyt dx dy
/ ~Lk bx t dx dy
S L t by t dx dy _

Integrating each term with the help of the integration formula given by Eq. (6-49)
gives the very simple (and expected) result below A typical term is evaluated m
the following manner

l'Q'Q 1

IA L,bx t dx dy bx t2A Abx tA

]

(1 + 0 + 0 + 2 )!

The complete result is given by

bx
by

bx
by

which says, in body forces in the t and y directions are allocated

effect, that the
equally to each node. This would not necessarily be the case if bx , b v , or t were
not constant within the element
 TWO-DIMENSIONAL STRESS ANALYSIS 299

Surface Tractions

The element nodal force vector as a result of the surface tractions s x and as given
by Eq. (5-91) may be evaluated by first noting that

~N, 0
-
0 N,

r 0
n = JVN S dS = Sse
0
(7-31)

n l 0
- 0
Note that the integrals are to be evaluated around the element boundaries (in a
counterclockwise direction) as denoted by the integration limit S e . As we have seen
in the examples presented thus far. the evaluation of such integrals for elements
totally within the body never contributes to the assemblage nodal force vector. In
this case, each internal force is equal and opposite on adjacent sides of the element,
therefore, the net contribution of these internal forces to the assemblage nodal force
vector is zero.
Figure 7-5(a) shows an element e with nodes i and j (but not k) on the global
boundary. A surface traction is assumed to act on leg ij. Let us denote the length
of . The integral may be evaluated rather easily if the shape functions
this leg as /
y
are written in terms of the area coordinates [see Eq. (6-25)] and if dS is replaced
by t dl. In addition, on leg ij, we have i\\ — L k — 0. and hence Eq. (7-31) becomes

rsh L,s xt dl

//„ L,s t
t dl

h LjS x t dl
(7-32)
u LjS x
t dl

0
O L

Figure 7-5 (a) Portion of the global boundary showing a surface traction on leg ij of

element e and (b) a variable surface traction.

300 STRISS ANALYSIS

With the help of the integral formula for the length coordinates (i.e., Eq (6-48)],
and assuming s,, s,, and t to be constant over leg ij, the nodal force vector as a
result of the surface tractions becomes

for leg ij on the global boundary {7-33}

Expressions may be denved in a similar fashion, or set up by inspection if leg jA

or A/ happen to be on the portion of the global boundary over which a surface
traction acts If the surface traction is distributed as shown in Fig. 7-5(b), the

integrals in Eq (7-32) could be evaluated as illustrated in the problems ora numerical

integration method from Chapter 9 may be used

Example 7-3
Reconsider the element in Example 7-1 A distributed surface traction acts on the
leg connecting nodes l and 3 (local nodes./ and A), as shown in Fig 7-6. Determine
the corresponding clement nodjl force vector

Solution

First, wc need to determine r, and r, Since the traction is directed only in the x

direction, we have a, = 0 The surface traction s, is not constant over IcgyA and,
2
therefore, the effective a, is taken to be the average of J 600 and 2000 lbf/in or,

=
1800 lbt/m 2 Noting that IcgyA (and not leg ij) in this case is on the global
boundary, we write the following using Eq (7-33) as a general guide

Figure 7-6 Element with imposed surfjee traction for Example 7-3
 TWO-DIMENSIONAL STRESS ANALYSIS 3Q1

Because an average value for the surface traction is used, the allocation to nodes
1 and 3 is equal. gj

Point Loads

The nodal force vector as a result of a point load (PL) at (.\ 0 ,v0 ) is given by Eq.
(5-92), or

N, 0
0 A7 ,

A 0 Njix0 ,y0 )fpt

fpL ~ SN 7
f
;,
— 2
',

(7-34)
0 tyUoO’o )fp,
N, 0
0 N k Ndx0 ,y0 )fPi

where f., x and L are the magnitudes of the point load in the and y directions,
respectively. Note that if the point load is applied at, e.g., node /, then the cor-
responding shape function (A7,) is unity and the others are zero, thus giving the
obvious intuitive result. Although the summation sign has been dropped above, it

is implied. In other words, each point load acting within an element has a corre-
sponding nodal force vector as given above. Given the coordinates of the point of
application of the point loadi.e. (.io,y„)J, we can determine the element in which
( ,

the load occurs by evaluating N„ A ,, and Nk at (.v 0 ,v0 ). When all three shape
7

functions yield values between zero and unity (i.e.. nonnegative and less than one),
the element containing the point load is found.

Example 7-4
A point load acts at (0.6, 1.6) for the element in Example 7-1. The .v and y
components of the point load are + 1500 and -2300 lbf, respectively. Determine

the corresponding nodal force vector.

Solution

The shape functions are given by Eqs. (6-21). Since they make use of all nine of

the m '
s, we must compute m u m l2
, , and m l3 (which were not needed in Example
7-1):

VU ~ ~ - (0X2)
AO'/ _ My i A'O'i _ (0)(0)
W|, ~ 2A
2A 2(4)

- _ r,v 2 - x2y3 _ (0)(4) - (4)(0)

in |2
v, -V,

~ ~
2A 2A 2(4)

-Vv
_ ~ a )'2 (4)(2) - (0)(2)
/ -V« _ -TD’i i
_
mn ~
2 A 2A 2(4)
302 STRESS ANALYSIS

Therefore, the shape functions may be evaluated at x = 0.6 and v = 1.6 as follows:

A', = 0 + (0.25)(0.6) + (0)0.6) = 0.15

AT, = 0 + (-0.25X0 6) + (0.50)(1.6) = 0.65

Nk = + (0)(0.6) + (-0.50)0.6) = 0.20

1

Note that the sum of the three shape functions is unity as expected. It follows from
Eq (7-34) that f^, is given by

(0.15)(1500) 225
(0 15)( — 2300) -345
(0.65)(ISdSj 975
(0.65)( — 2300) -1495
(0.20X1500) 300
(0 20X-2300) -460

Note most of the point load is allocated to the node j (global node number 1)
that
because node is the closest to the point load In a similar fashion, node i (global
this

node number 2) receives the lowest allocation since it is farthest from the point
load B

Composite Nodal Force Vector

The reader is that the contributions from each of the nodal force vectors
reminded
for a given element must be combined according to Eq. (5-86), which is repeated
here for convenience

p- r«, - e. + is + n + «ti s 861

-

The vector {' (w ithout subscripts) may be referred to as the composite nodal force
\ector. We are now in a position to discuss the assemblage step.

Assemblage Step

The assemblage of the element stiffness matrices

r
K
and the composite nodal force
vector fr follows the standard procedure illustrated in Chapters 3 and 4. Example
7-5 should help to clarify further the basic idea.

Example 7-5
Consider the mesh shown in Fig. 7-7, which contains only two element!.. The
elements are defined in terms of the global node numbers, as indicated below the
 TWO-DIMENSIONAL STRESS ANALYSIS 303

3 4 2

Figure 7-7 Mesh used in Example 7-5.

mesh. The 6x6 element stiffness matrices for the two elements may be written
symbolically in terms of nine 2x2 submatrices as follows:

Kg my Kg
K (n = Kg KV> K'g
K*
IS.32
1 '
Kg Kg

and

K33 Kg Kg
r( 2
K< 2> = Kg Kg *^42
)

Kg Kg Kg

Show the assemblage stiffness matrix in terms of the 2 X 2 submatrices.

Solution

The assemblage stiffness matrix K" is of size' 8x8 and is given by

Kg Kg 0

Kg Kg + Kg Kg + Kg ir< 2 )
**•24

Kg Kg + Kg Kg + Kg Kg
t?(2) jr{2) t>(2
0 IV42 **43 **44

stiffness matrix is
Although it is not obvious from this result, the assemblage
is not too obvious
symmetric. Moreover, it is banded, although the bandedness

should be recalled that the

because only two elements were considered here. It

assemblage stiffness matrix K" is zeroed-out before the element stiffness matrix

for the first element is added into it.

304 STRESS ANALYSIS

Example 7-6
Reconsider the mesh enen m Example 7-5. The 6 x 1 composite nodal force

vectors for both elements mav be written svmbolicallv in terms of three 2 x I

subvectors as follows

r," r,-'

r>* = fi" and r2 ’ = r,

2'

r,“ 7F
Show the assemblage nodal force vector in terms of the 2 xl subvectors.

Solution

The assemblace nodal force vector

LU
is of size 8 x I and is given by

L i?”J
In the two previous examples, the mesh was comprised of onlv two elements. In
general, the assemblage step is performed for even element This results in an
assemblage stiffness m3tnx and an assemblage nodal force vector of sizes 2A x
r

2.Y and 2.V x 1. respectively, where A' is the number of nodes used in the entire
mesh The reader is reminded that the half-bandwidth of K" is given b> Eq (3-
33) Like the assemblage stiffness matrix, the assemblage nodal force vector is

zeroed-out before the nodal force \ector for the first element is added into it

Prescribed Displacements

The prescribed displacements are imposed according to either of the two methods
illustrated in Sec. 3-2. Method I is preferred because it results in a better conditioned
stiffness matnx m large problems After application of the prescnbed displacements,
the s\ stem of equations to be solved is given by

It should be recalled that the vector a is defined by

r 17361
a = (u, v, | u2 in \ i «\ »\I

where u, and
\ are the displacements in the x and v directions at node i, and where
,

A’ is the maximum number of nodes used.

 TWO-DIMENSIONAL STRESS ANALYSIS

Solution for the Nodal Displacements

Before application of the prescribed displacements, K" is singular and does not
possess an inverse. After the proper application of the prescribed displacements,
K is nonsingular for well-posed problems. Thus, a unique solution for the nodal
displacements is ensured. Subroutine ACTCOL from Sec. 6-8 may be used quite

effectively to obtain the solution. It should be recalled that the subroutine utilizes
a three-step procedure: triangular decomposition, forward elimination, and back-
ward substitution.

The Element Resultants

Let us now indicate how we could obtain the strains and stresses within a typical
element once the nodal displacements in the vector a have been found. Recall that
the strains within the region are related to the nodal displacements on an element
basis by

e = Ba‘ (5-77)

Since both the matrix B for any element e and a*' are known (because a is known),
it is clear that Eq. (5-77) can be used to determine the strains (e u , en , and e, v )

within element e. It should be noted that the resulting strains are really averages
for the element and are generally associated with the centroid of the element. In
terms of the in,/ s (defined by Eqs. (6-2 )j and the now known nodal displacements,
1

the average strains (denoted by e„ ? u and 5 U ) within element e

, , are given explicitly

by

ev, = m2 ]ii, + + '"23 "a (7-37a)

e„ = m 3 ,v, + m n Vj + (7-37b)

en = m 3i u, + + m 32 «, + m 22 y, + m 3} uk + m 23 \\ (7-37c)

The average stresses (5\ ( , 5,„ and <r„) within element e may be evaluated from

a = D[Ba‘' - e0 ] + or 0 (7-38)

where each matrix or vector on the right-hand side is known. These average stresses

are also generally associated with the centroid of the element. It should be recalled

that in the case of plane stress, the strains in the z direction are not necessarily zero
and can be computed from Eq. (7-16). Similarly, in the case of plane strain, the
stresses in the z direction are not necessarily zero and may be determined from Eq.
(7-21). Like the other stresses and strains, these too are average
values and are

usually associated with the centroid.

306 STRESS ANALYSIS

Remarks

The displacement function for the three-node triangle was taken to be linear [see
Eq (6- 16a)]. Hence, the displacements are assumed to vary linearly within each
element The strains and stresses, on the other hand, being related to the derivatives
of the displacements, must then be constant. Hence, this element is often referred
to as the constant strain triangle.
Although the expressions for the finite element characteristics have been eval-

uated for the triangular element, the reader should have little difficulty in extending

thisdevelopment to the rectangular element. In Chapter 9, other, more powerful,

two-dimensional elements are introduced.

7-3 AXISYMMETRIC STRESS ANALYSIS

In this section, the axisymmetnc stress model is defined The finite element char-
acteristics are derived from Eqs (5-87) to (5-92) for the case of the axisymmetnc
triangular element Each of the basic steps in the FEM formulation and solution is
discussed

Definitions

An axisymmetnc problem m stress analysis represents another example in which

a three-dimensional problem idealized as a two-dimensional problem Such
may be
problems must meet the two basic conditions delineated in Sec 6-6. It should
further be recalled that these problems lend themselves to a cylindrical coordinate
system with coordinates (r,0,z) as opposed to a cartesian (x,y,z) system However,
because of the axisymmetry, the 0 coordinate need not be included In other words,
the geometry, material properties, loadings, etc may be a function of r and z only
,

(and not a function of 0) For example, in Fig. 7-8, note how the line load acts in
an axisymmetnc manner about the z axis The displacements m
the r and z directions
are denoted by u and v, respectively In this section, the triangular element discussed
in Sec. 6-6 is used for illustrative purposes The element is really a triangular donut
or toroid with the nodes being represented as circles formed about the z axis, as
shown in Fig 7-9

The Strain-Displacement Relationship

of
A fundamental difference exists between axisymmetnc stress analysis and that

plane stress and plane strain; namely, a fourth component of the strain e 96 (an
hence stress tre9 ) must be explicitly considered in addition to the other three strains,
strain at a
err , e22 , and e rjr (and stresses, ct z,, and ct^). The circumferential
 nAuuniHCiWV. Ar<AL,loli oU7

/
/

Figure 7-8 Axisymmetric body with an axisymmetric line load.

point within the axisymmetric body is caused by the radial displacement u at this

same point. Therefore, the strain vector e has four components and is defined by

e = [e„ € O0 T (7-39)
e.. t r: ]

These strains are related to the radial displacement it and the axial displacement r
by the following strain-displacement relationships [3 J:

dll
e rr (7-40a)
dr

Figure 7-9 Axisymmetric body showing the axisymmetric triangular clement (i c.. a tri-

angular toroid).
 u
e 90 (7 -40b)
r

dv
(7-40c)
dz

~
dll
+ —
dv
(7-40d)
dz dr

T
Defining the displacement field vector u by u = \u v] , we may wnte Eqs
(7-40) in a very compact form as

€ = Lu (7-41)

where the linear operator matrix L is defined by

_a
0
dr

0
L - r (7-42)
d_
0
dz
d_ d_

dz dr

The Shape Function and Strain-Nodal Displacement Matrices

Let us define the nodal displacement vector a' by

T 5 ‘ 59 *
ae = [w, v, Uj v, ti
k Vl \
^

since each of the three nodes i,j, k has two components of displacement It follows

from Sec 5 6 that the shape function matrix is given by

0 Nj 0 \Nk 0 (5-61)
N =
N t
0 Nt |
0

where the shape functions themselves are given by Eqs (6-2 1 ) with x and ) replace
to
by r and z, respectively. Hence, the displacement field vector u may be related
the nodal displacement vector a r in the usual manner

(7-43)
u = Na'
 /VA.1D I IWiVUl 1 Kxv- ANALYSIS 309

It then follows from e = Lu = LNa' = Ba e that

—
dN,
dr
0 —
dN
dr
,

0
d N
dr
k
0

K 0 5 0
Nl
0
r /•
r ( 7 - 44 )
B = LN = dN, d/v, dN,
0 0 0
dz dr dz
dN, dN, d/V, dN, dN, m.
dz dr dr dr dz dr

whereupon computing the derivatives we get

m2 \
0 m 22 0 m 23 0
N, n7
lj N,
0 0 0
r r r ( 7 - 45 )

0 Wjl 0 m i2 0 '”33
in 3 , »hi mn m 22 '"33 '”23

Note that N,lr, NJr. and NJr are not constant but rather are functions of /• and r.

Nevertheless, the shape function matrix N and the strain-nodal displacement matrix
B are both determined.

The Constitutive Relationship

As mentioned earlier, four stress components need to be considered. Hence, let us
define

r
o' = [o- rr cr e9 a.. cr r -] ( 7 - 46 )

Note the addition of the stress cr fje , which acts in the 0 direction (i.e. , circumfer-
entially). This definition is consistent with that given for the strain vector e in Eq.
(7-39). Restricting the present discussion to isotropic materials only and assuming
the self-strain is a result of a temperature change AT, we define

^m> a, AT
€ 09o a, AT
€--0 a, AT
— ^rzn — 0

where a, is the coefficient of thermal expansion. Note that a thermal shear strain

cannot exist in an isotropic material. Let us also define the residual stress vector
as

o- 0 = foy ra I L'i 1 1
; . ^rro I ( 7 - 48 )
310 STRESS ANALV SIS

For a linear, elastic material, the constitute e equation is given by Eq. (5-24). or

<r = D(c - e0 ) + Wo (5-24}

Zienkievvicz (4] gives the corresponding material property matrix (for an isotropic
material) as

1 - (t p. p. 0
H- 1 - P ft 0
(1 + H)(l - 2(1.) ft p. 1 — M- 0
0 0 0 K(l - 2p)_
(749)

where E and p. are the modulus of elasticity and Poisson's ratio, respectively. The
materia! property matrix is seen once again to be sj mmetric.

The Element Stiffness Matrix

The general expression for the stiffness matnx is giv en by Eq (5-87). If we substitute
2-ir r dr d: for the elemental volume <l\', we get

K r = $v< B r DB dV = 277 /4 , B r DBr dr dz (7-50)

Note that the volume

integration reduces to an integration over the cross-sectional
area of the triangular donut In other words, we must evaluate the latter integral as
an area integral In effect, the half-plane of the body has been discretized with a

number of triangular elements as shown in Fig 7-10

Figure 7-10 Axisymmetnc body discretized into a number of three-node triangular ele-

ments
 1

AXISYMMETRIC STRESS ANALYSIS 31

The area integral to be evaluated in Eq. (7-50) is not as simple as the plane
stress or plane strain counterpart because the integrand now
is a function of r and
r. The integration in this case is no longer trivial. At this point we could revert to
quadrature or numerical integration to evaluate the area integral in Eq.
(7-50), as
explained in Chapter 9. However, let us evaluate the integral approximately by
evaluating the integrand at the centroid of the triangle and then treating B and r as
though they were constants. The coordinates of the area centroid are denoted by r
and 5 and are given in terms of the nodal coordinates by

r,
+ >) +
r ( 7 - 51 )
3

and
z.

( 7 - 52 )
3

The resulting B matrix is denoted as B, where B = B(r,z), denotes that B is

evaluated at the coordinates of the centroid r and z. Therefore, Eq. (7-50) may be
written as

= 27rB r DB rfA edrdz

-

K*

Clearly, the remaining integral is simply the area A of the triangle given by Eq.
(6-2 le) with r and r replacing .v and v. respectively. The resulting element stiffness

matrix is given by

K c = I'X&m'rA ( 7 - 53 )

Note that K r is of size 6 x 6 and is symmetric. The matrix and scalar multiplications
in Eq. (7-53) would actually be performed in a computer program by calls to

appropriate subroutines.
This simple approach is known to give convergent and valid results il it gives

the exact value for the volume. For example, the volume of a triangular toroid is

given exactly by 2rr FA, which is what this approximate integration method yields.
Therefore, this approximate integration method is guaranteed to give good results.
The accuracy deteriorates as the elements increase in size.

The Nodal Force Vectors

Useful expressions for each of the five nodal force vectors are now derived tor
axisymmetric stress analysis performed with the three-node triangular element.

Self-Strain

The nodal force vector as a result of the self-strain is given by Eq. (5-88) and may
be evaluated here by first noting that

rDe r dr dz 7 - 54 )
ff0 = JV,B rDe0 dV = 2-irJ A ,B 0 (
312 STRESS ANALYSIS

where the integrand is seen once again to be a function of r and z (because B is

actually a function of r and z). Let us, therefore, take the approximate approach
first by evaluating the integrand at the coordinates of the centroid (?,z) and then
by treating the integrand as a constant to give

f' = 2irB r De 0 rA (7-55)

where B = B(?,z), as in the evaluation of the stiffness matrix. This nodal force
vector is of size 6 x las expected If the self-strain is due to a temperature change,
then we simply use Eq (7-47), which explicitly includes the coefficient of thermal
expansion a, and the temperature change A T. If AT varies over the element in
question, then the value at the centroid may be used and thus treated as a constant
This simple approach is consistent with the approximation made above

Residual Stresses

The element nodal force vector as a result of the residual stresses is determined in

the following manner beginning with Eq (5-89)

f!xo
= JV'B r cr0 dV = 2ir/A r r cr r dr
dz = 2irB 7
a 0 r4 (7-56)
0

where tr0 is given by Eq (7-48) If the element in question is not under a prestress,
then a0 is simply taken to be zero

Body Forces
The body forces per unit volume in the radial and axial directions are denoted as
b r and b., respectively The corresponding nodal force vector is evaluated by starting
with Eq (5-90) and noting that

15 = JV,N rb dV = 2T/4 ,N r br dr dz 17-571

where the body force vector b is defined by

b = [br bz ] T < 7' 58)

for thisaxisymmetnc analysis Recall that N is a function of r and z If the body

forces are assumed to be constant over the element in question and the integrand
is evaluated at r and z, then Eq (7-57) may be evaluated to give

allocated to
Thus it is seen that one-third of the body force in each direction is

each node. This result should be intuitively obvious since br and b. were assumed
to be constant
 AXJSYMMETRIC STRESS ANALYSIS 313

If the axisymmetric body is oriented such that the r axis is collinear with
direction of gravity, then b. — p° may be used, where p is the mass density and
g is the acceleration due to gravity. If the body is rotating at an angular speed to

about the r axis, then we may take br = pror. In this case, b, is seen to be a
function of r and we may, in effect, use b, = pror.

Surface Tractions

The integral representing the nodal force vector as a result of the surface tractions
is not a volume integral, but rather a surface integral. Equation (5-91) provides the
starting point, which may be rewritten in terms of the elemental surface area on
the boundary of the element as

= J^N r s dS = 2ir
r
J,N sr dl (7-60)

where s is defined in terms of the surface tractions s r and s. by

s = [sr
r (7-61)
s-]

and dl is the elemental length around the boundary of the element. If the leg of the
triangular element in question is internal and not on the global boundary, then the
corresponding nodal force vector as a result of the internal loads do not contribute
to the assemblage nodal force vector. The reader will recall that these internal

surface tractions are always equal and opposite and. hence, contribute nothing to
the assemblage nodal force vector. In other words, only those legs on the global
boundary over which surface tractions are imposed need to be considered.

For example, consider leg ij for the element shown in Fig. 7-11. One way to

evaluate the integral in Eq. (7-60) is to evaluate the integrand at the midpoint of

Figure 7-11 Typical triangular element with leg on the global boundary' on which a
ij

axisymmetric
surface traction is imposed. Note The surface traction must be
 31 4 STRESS AVALS SIS

leg ij and then treat the integrand as a constant. Let us define r(/ as the radial

coordinate at the point halfway between nodes / and j. Let us also note that N t is

zero on leg ij. With these assumptions, it can be show n that Eq (7-60) simplifies
to
sr /2
s ./2

s’fi
£ = 2ir fv ltJ
sJ2
for leg ij on the global boundary (7-62)

“o
0 J
where l
(J
is the length of leg ij. A
more sophisticated way of evaluating this integral
is illustrated where the special integration formula [see Eq. (6—48)]
in the problems
involving length coordinates is used to simplify the integration. The reader should
be able to write, by inspection, the expression for when legjA or ki is on the
global boundary.
It is interesting to note that if a body of revolution is loaded in a nonaxisymmetne
*
manner, it may still be possible to pertorm an axisjmmctric anal) sis by using a
one-term Fourier series of sine and cosine functions to represent the loads and the
displacements The loads, however, must still be symmetric about a plane through
the r axis An example of such a case would be the wind load on a circular smoke
stack. Hucbner [5] and Wilson [6) give further details Treatment of this situation
is beyond the scope of this text

Line Loads

For axis)mmetnc bodies the so-called point loads arc actually line loads around
the circumference of the body as shown in Fig 7-S. It is customary to define the
line load in the radial and axial directions on a per unit circumference basis. Let
us define the line load per unit circumference in the radial and axial directions as
Fr and F.. respectively The equivalent point loads are then given bv
~
fr = 2-xrF, l ? - 63al

and
(7-63b)
fp = 271 rF.

Let us also assume the load acts at coordinates (r0 ,c0 ) Beginning with Eq (5-92).

we can write

A ,(r0 .r0 )
r
0
0 A',(r0 ,r0 )

A ,(i*o*-o)
r
B 2-r0 F f
17 - 64 )
0 A',(r(i.=o) 2 ~r0 F:

A^(/o,r0 ) 0
0 'V*( ro--o)-
 AX1SYMMETRIC STRESS ANALYSIS 315

If leg ij is the leg on which the load acts as shown in Fig. 7-12, then N (r0 ,z0
k ) is
identically zero, and the nodal force vector as a result of a line load is given by

N,(r0 .z0 )F:

NjirQ .Zo)Fr
fpj
PL — 2.tt r0 (7-65)
-N,(ra ,:a )Fz

0
0

In Eq. (7-65) the summation sign has been dropped because it is understood
that all line load contributions would be added in the usual manner. Similar expres-
sions can be set up by inspection if leg jk or ki is on the global boundary. It should
be obvious that the shape functions in Eq. (7-65) allocate the intuitively expected
amounts of r0 Fr and Itt^F. to each of the nodes. For example, if leg ij is on
the global boundary and if the line load acts one-third of the distance from node /
to node j. then Eq. (7-65) would give
r
fp L = AF
2-n rQ [ 2 r -AF: j
'AFr AF
l
: |
0 0]

as expected. In these cases, however, the stress distribution in the immediate vicinity
of the load would be very poorly represented unless hundreds of tiny elements were
used in this region. It is generally recommended that nodes be placed at the locations
of these line loads for improved accuracy. In this case. Eq. (7-65) still gives the

correct result.

Remarks

The assemblage of the element stiffness matrices and nodal force vectors is per-

formed in the usual manner. The discussion on the assemblage step in Sec. 7-2 is

Figure 7-12 Typical triangular element with leg ij on the global boundary' on which an
axisymmetric line load is applied.
316 STRESS ANALYSIS

applicable here as well. Again Method 1 is recommended for the application of the
prescribed displacements Finally, the resulting system of equations embodied in
Ka = f may be solved with the help of subroutine ACTCOL, as explained in Sec.
6-S Thus, the nodal displacements for every node are known and may be used to
obtain the element strains and stresses as follows. The average strains within an
element may be computed with the help of € — Ba r or

€ = Ba r (7-66)

and the average stresses from

or = D(Ba — r
€0 ) + cr 0 (7-67J

These average stresses and strains are generally associated with the centroid of the
element

7-4 THREE-DIMENSIONAL STRESS ANALYSIS

Three-dimensional stress analysis encompasses all practical structural engineering

problems since no approximations or simplifying assumptions are inherently made
The pnce paid for this generality is that of much larger computer storage require-
ments and obviously longer computational times, as shown in Table 7-1 where two
problems are compared Note that the computational time is roughly proportional
to the number of unknowns and to the square of the half-bandwidth. Table 7-1
illustrates the enormity of the three-dimensional problem and has motivated the
development of higher-order (distorted) elements, some of which are presented in
Chapter 9
The simplest three-dimensional element is the four-node tetrahedral element
with nodes i. j, k, m as shown in Fig 7-13 The finite element formulation of

Table 7-1 Comparison of Computational Times for Typical Two- and

Three dimensional Problems*

Two dimensional Three-dimensional

problem problem

Number of nodes 30 x 30 = 900 30 x 30 X 30 = 27.000

Number of equations (and un-

knowns) 2 x 900 = 1800 3 X 27.000 - 81.000

Variables in bandwidth 2 X 30 = 60 3 x 30 X 30 = 2700

Relative computational time 1 unit 90.000 units

*The computational time is roughly proportional to the number of unknowns and to the square
of the

half bandwidth
 THREE-DIMENSIONAL STRESS ANALYSIS

Figure 7-13 The four-node tetrahedral element.

problems in three-dimensional stress analysis is illustrated quite v. ell with this

element. Recall from Sec. 6-5 that two rules must be followed when specifyin Cij

which nodes are associated with a particular element. The reader may want t c
review the pertinent materia! in Sec. 6-5 in general and Example 6-5 in particular.

If the reader understands the basic approach taken here, little difficult} will be
encountered if an alternate formulation in terms of the eight-node brick is desired.
The integrations in this case would be best performed numericalh as illustrated in
Chapter 9. where other, more practical, elements are introduced.

The Strain-Displacement Relationship

The strain-displacement or kinematic relationship for small deformations in three

dimensions was given by Eq. (5-15) and is repeated here for easy reference:

hu dr cl\v

~ OZ
ox ()\ (5-15)

OK
^
OV
e s: =
bv
— -*- —
b\v
—
bn'
"

OX
— —
bit

bx OX hz by hz

where it. v. and w* are the displacements in the a. and r directions, respective!}

Let us define the displacement field vector u as

r
u = \u i »i I
(5-69)

and the strain vector as

r
e = [e„ e, e.. e, e, £-,i (5-26)

Equation (5-15) may then be written in the usual compact form e = Lu. where
318 STRESS ANALYSIS

*_3
0 0
3r

0 A 0
d\
d_
0 0
dz (7-68)
_3 _3
0
ay 3t

o £ _3

3: 3v
d_ 3^
0
dz 3r

The Shape Function and Strain-Nodaf Displacement Matrices

Let us define the nodal displacement vector a' as follows

r
a' = [u, v, w, ‘
u} v, u, 1 1

ut vt n*4 um \„, u,„] (5-70)

since each of the four nodes i, j, L, m has three components of displacement it

follows from Sec 5-6 that the shape function matrix N is given by

IV, 0 0 Nj 0 0 Nk 0 0 !
0 0
N = 0 N, 0 0 N, 0 0 N k
0 i 0 Nm 0 (5-72)

.0 0 N, 0 0 N, 0 0 Nx 1
0 0 N,„_

where the shape functions themselves are given by Eqs (6-40) It then follows that
the displacement field vector u is related to the nodal displacmeent vector a' in the
usual manner by

u Na f (7-69)

and from B = LN that

0 0 0 0 0 0 J 0
Wll 0 0 ut,, 0 0 0 o 0
0 , 0 0 m 4i 0 0 0 (7-70)
-n,, 0 0 mu
0 Wn 0 mji 0 0 '"44

0 m2 i m4 , 0 0 0

where the in / s are known for any given element and may be computed from Eq
(6-40e). Clearly, the strain-nodal displacement matrix B is known for any given
 THREE-DIMENSIONAL STRESS ANALYSIS 319

The Constitutive Relationship

Assuming a linear, elastic material, we can write the stress-strain relationship in

the usual form:

or = D(e - e 0 ) + or0 (5-24)

where in this case, in addition to e from Eq. (5-26). we define

ct = [cr„ o\, (T
:: on a,. a.,J
/
(5-25)

«o = K,„ e,.„ e.,,,]

7 "

(5-27)

Oo = o-.,„ cr fui o\.„ cr.J

r
(5-28)

The corresponding material property matrix for linear, elastic isotropic materials is

given by Zienkieudcz 17 1
as

If the self-strains e 0 stem from a temperature change A7\ then we have

7
€0 = [«, AT a, AT «, AT 0 0 0] (7-72)

where a, is the coefficient of thermal expansion for the material comprising the
element. Once again it is noted that only normal strains are induced by a temperature
change in an isotropic material.

The Element Stiffness Matrix

Starting with Eq. (5-87) and recognizing that B and D are constant, we get

K' = * B 7 DIU' (7-73)

where V. the volume of the tetrahedron, is given by Eq. <6-400 and is a function
of the nodal coordinates only. For the four-node tetrahedral element, the element
stiffness matrix is of size 12 x 12 and is symmetric
320 STRESS ANALYSIS

The Element Nodal Force Vectors

The clement nodal force vector as a result of the self-strains is easily determined
from Eq (5-88).

Ito = fy,B r De„,n' - B r De 0 V' (7-741

If the self-strain is duea temperature change A7" in a material with a coefficient

to
of thermal expansion a,, then Eq (7-72) should be used for e 0 This element nodal
force vector is of size 12 x 1 since the element has four nodes and three degrees
of freedom per node
The element nodal force sector as a result of the prestresses is computed from
Eq. (5-89) to give

!S» = f v .B'o 0 ,tV = B ra„V (7-751

which also is of size 12 x |

The element nodal torce sector as a result ot the body lorces is computed by
starting with Eq (5-90) Writing the shape Junctions in terms ot the volume co-
ordinates and integrating the result with the help ot hq (6-50). wc get

L, 0 o' ~b~
0 L, 0 ft,

0 0 L, b:
L, 0 0
b<
0 Lj 0 by
ft/

fb
0
L, 0
0 L,

0
ft,

b.
-4 V b:
by
(7-76)

by
0 Lt 0
ft
0 0 Lt

Lm 0 0 by

0 ~Lm o" by

0 0 Lm b:

Note that the body forces per unit volume (ft,. ft,, and ft.) were assumed to be

constant. As expected, one-tourth ot ft, is allocated to the v degrees of freedom of

each of the four nodes, etc
The nodal force vector as a result ot the surface tractions needs to be evaluated
for only those dements that have one or more faces on the global boundary'
Moreover, only those exposed laces with imposed surface tractions acting on them
7-14,
need to be considered. For example, consider the tetrahedron shown in Fig.
imposed surface
which is assumed to have face tjk on the global boundary and an
 THREE-DIMENSIONAL STRESS ANALYSIS 32]

Figure 7-14 Tetrahedral element with face ijk on the global boundary with an imposed
surface traction.

traction with components sy, j,, and t.. On this face N,„ is identically zero, and
Eq. (5-91) simplifies to

N, 0 0
0 N, 0 N,s ,

0 0 N, N,s:
0 0 A
0 Nj 0
0 0 N, NjS.
n = /s< N7s ds = lA,ji dA = Ia,,, dA (7-77)
*r 0 0 A7**.
0 K 0
0 0 N, NiS.
0 0 0 0
0 0 0 0

_0 0 0 0

where A x is the area of the triangle ijk that comprises face ijk of the tetrahedron.
Equation (7-77) may be integrated with the help of the integration formula for area
coordinates (Eq. (6-49) j providing the shape functions are written in terms of the
area coordinates (see Eq. (6-25)} to give the intuitively obvious result

= r
f' s. s. s, s. j
0 0 0( (7-78)

If any of the remaining three faces were on the global boundary, the corresponding
nodal force vector could easily be set up using Eq. (7-78) as a guide.

The element nodal force vector as a result of a point load mas be computed
from Eq. (5-92) as follows. First, note that the external load must necessarily act
322 STRESS ANALYSIS

on one of the four faces of the tetrahedron (or at a node), but only if the face (or
node) is on the global boundary. For example, consider a point load acting some-

where on face ijk of the tetrahedral element in Fig. 7-14. Let the location of the
point load be represented by the coordinates (x0 ,y0 ,r0 ). Obviously, N)H (r0 ,j 0 ,z0 )
is identically zero on this face Therefore, in this case Eq (5-92) reduces to

N, 0 0“
V/..

0 N, 0 A'Ao.Vo.Zo )/,.
0 0 N, (V,(ro,>0' 2 0 )/;>

N, 0 0 ^(vn ,yo,z0)/;,.
0 N, 0 NjUo,\ 0 ,z0 )flh
//>,
0 0 N, NAo.Jo.ZiiV,,
nl 0 0
0 0 N,(Vo.Vo.:o V,:
0 0 N k

0 0 0 6
0 0 0 0
t = *0
0 0 0
°J

Similar expressions for fp L can be set up by inspection if the point load acts on any
of the other three faces

Remarks

The assemblage, application of prescribed displacements, and solution of Ka = f

yield the three nodal displacements for every node These displacements may be
used to obtain strains and stresses for any element e from

? - Ba' (7-80)

and

-- D(Ba f - € 0 ) + tr 0 (7*811

These strains and stresses are average values for the element and are usually as-
sociated with the volume centroid of the tetrahedron Principal stresses can be
computed with the help of Eqs. (5-12).

7-5 BEAMS

The finite element characteristics for the beam element based on the elementary
theory of beams are derived in this section This particular model proves to be very
 BEAMS 323

useful when
statically indeterminate beams are encountered, although the model is

equally applicable to the statically determinate case. It should be pointed out that
in some cases the plane stress finite element characteristics may be used to model

certain types of beams with greater accuracy and detail than the elementary theory'
can provide. For example, if a thin flat plate is cantilevered and loaded as shown
in Fig. 7-15, a rather detailed two-dimensional stress distribution may be found
within the beam using the plane stress formulation described in Sec. 7-2. In Fig.
7-15. the point loads P,, P 2 and P3 and
, the distributed load p(.\) must act in the
plane of the plate for the plane stress model to be valid.

Review of the Elementary Theory

The following assumptions are consistent with the elementary theory of beams.
Hooke’s law applies in that the stresses are related to the strains in the usual linear

fashion. The deflections are assumed to be small in comparison to the beam di-

mensions. Loadings (forces and moments) must be transverse to the beam, and
longitudinal strains are assumed to be negligible.
A typical beam is shown in Fig. 7-16, where the point loads Pj and P2 act on
the beam in the plane of the paper. In addition, a distributed load p(x) and the
externally applied moments M, and ML are shown. The distributed load p(x) may
be a function of The left end of the beam is said to be cantilevered while the
.r.

right end is simply supported. The deflection and slope at the cantilevered end are
zero, whereas the deflection and moment at the simply supported end are zero.
With the help of Fig. 7-17, the following sign conventions are adopted: A
positive moment M results in compressive stresses for positive values ofy; otherwise

Figure 7-15 Beam for which the plane stress model is valid.
324 STRESS ANALYSIS

Figure 7-16 Typical beam with a distributed load per unit length p(x), point load:* P, and
P2 , and applied moments M, and M ;

the moment is negative A shear force l' is positive if directed in the positive
direction on a positive face (a positive tace is one in which the outward normal to

the face is in the positive coordinate direction) or if directed in (he negative direction
on a negative face, otherwise the shear force is negative This sign convention for
shear force is consistent with that defined in Sec 5-2. A distributed lateral force

p is positive if directed in the positive \ direction Note that p(x) is shown to be

constant in Fig 7-17 because as tlx approaches zero. p(x) approaches a constant
from -x to .v + <lx
With these sign conventions, (he following equations |8] relate the shear force
V', the moment M
and the lateral load p
,

Figure 7-17 Infinitesimal beam element showing positive shear forces, moments, and

distributed loading
 BEAMS 325

dM
-V (7-83)
dx

M (7-84)

where w represents the deflection in the v direction and El is the product of the

modulus of elasticity £ and the moment of inertia I of the cross section about the
axis going through the neutral plane of the beam. The stresses as a result of bending
are zero in the neutral plane. The product El is often referred to as the flexural
rigidity. Recall that the moment of inertia is defined by

I = Sa r dA (7-85)

where A is the cross-sectional area of the beam. Furthermore, the slope 0 of the
beam is related to the lateral deflection ir by

dw
0 = (7-86)
lx

If Eqs. (7-82) and (7-83) are combined, we obtain

d~M
= P (7-87)
dx 2

If the last result is combined with Eq. (7-84), we get

d2 d 2w
2
E! = p(x) (7-88)
dx \ dx 2

which is the governing equation for the deflection of the beam. Equation (7-88)
allows for the fact that the flexural rigidity El may vary along the length of the
beam.

C-Continuous Shape Functions

The element characteristics for the beam model can be derived from Eq. (5-
Finite

87), etc.,provided that the generalized displacements are taken to be the deflections
and slopes (w and dw/dx ); the strains to be the curvatures of the neutral plane (cPwl
dx2); and the stresses as the bending moments (M). However, let us take an alternate
but completely equivalent approach by deriving the finite element characteristics
by using the Galerkin weighted-residual method presented in Chapter 4.
Recall that the Galerkin method (in the Finite element context) requires that we
choose a suitable trial or basis function that is applied locally over only a portion
of the complete x domain; i.e., over a typical finite element. Let us denote this
trial displacement function or parameter function by n\ Figure 7-18 shows a typical

Finite element e with nodes i and j over which w is to apply. However, we must

ensure that we have not only interelement deflection continuity but slope continuity
326 STRESS ANALYSIS

Figure 7-18 Typical beam dement e with nodes i and j

as well In other words, at each element boundary (or node in this case) we insist
that both the deflection and slope be continuous, or that the parameter function be
C'-continuous
Recall further that problems were defined to be C'-continuous when the weak
formulation contains at most only second-order derivatives. The weak formulation
corresponding to Eq (7-88) may be obtained by either the variational approach or
by the Galerkin weighted-residual method The latter is illustrated in this section
and indeed yields an integral equation that contains no derivatives of higher order
than two Therefore, the beam model requires C -continuous parameter functions, 1

which in turn requires C'-continuous shape functions as seen below

Each element has two nodes, but four requirements must be met. the deflections
and slopes at the two nodes must be continuous Stated differently, at node i, we
must have a deflection *r, and slope 6,. while at node j we must have deflection it,
and slope 0,. Therefore, four constants must be introduced in the assumed form of
the parameter function. Not surprisingly, the following function meets the com-
patibility and completeness requirements delineated in Sec 6-2

iv = c, + c2 x + <>t 2 + c4 r 3 (7-89)

From Eq (7-86) we see that the slope within an element must be taken to be

2
+ 2ot +
e = ^= c, 3c, t 17-90)

These last two equations provide the starting point for the derivation of the shape
functions.
For a typical element e, the vector ae is taken as

e = r (7-91)
a [ u, 0, |
Wj 0y ]

From Eqs (7-89) and (7-90) it follows that

w, = C] + C2-X, + c+x; + Cvx?

0, = c2 + 2ox, + 3C4-I7

Wj = C, + CyXj + Cyxf + C+X^

0,
= c2 + 2cy\j + 3 c+rf
 BEAMS 327

Let us rewrite these last four equations in matrix form as

i
l -V A,'l <4
«, 0 1 2\, 3at cz
1 A, -v/ Cl
-V
LoJ 0 1 _C 4 _

If we solve for the vector of coefficients we get

- I
3
Cl 1 A, AT A, M7
C; 0 1 2\, 3.\f 0,
(7-93)
C} 1 -V, Af x) u ’y

_C.J_ _0 I lXj 3.y~_ UJ

Note that the inverse of the 4x4 matrix of nodal coordinates is indicated, but let

us postpone getting this inverse for now. Instead let us rewrite Eq. (7-89) in matrix
form as

C|

c2
1 .1 A' .v''
] (7-94)
c-i

L r4.
If Eqs. (7-93) and (7-94) are combined by eliminating the vector of coefficients in

the usual manner, we get

- 1

i A. AT Af" »,
0 1 -V, 3 at 0,
»• = 1 (7-95)
[
1
-'r
-'7 wj
_0 1 3a; J UJ
It is very convenient from a mathematical point of view to work in terms of a local
coordinate £ defined by

£ = (7-96)

w’here L (not to be confused with the linear operator matrix L, which is not used
in this section) is defined by

L — — x, (7-97)

Clearly, L represents the length of the element.

Furthermore, let us write the parameter function if in terms of the shape
functions and nodal deflections and slopes as follow-s:

if = Na" = N„,%v, + N e ,0, + N„jWj + A'^O, (7-98)

Note that each node has two shape functions associated with it: one for the deflection
and one for the slope. If the right-hand side of Eq. (7-95) is multiplied out (after
328 STRESS ANALYSIS

performing the indicated inversion) and the variable r is eliminated in favor of the
local coordinate £ by using Eq (7-96), we get

iv » [ 1 - 2
3£ + 2£
3
K+ [ L(£ - 2£* + £
3
) ]0,

+ [ 3£
2 - 2£3 } Wj + \ L(-i2 + £
3
) ] 0, ( 7 - 99 )

Comparing Eqs. (7-99) and (7-98) reseals that the shape function matrix N in terms
of the local coordinate £ is given by
r r
"
Af„~ 3p + 2p
1 -

N„ Ui - 2p + p)
3? - 2?
L"d *•
. H-p + p) _

The reader should venfy that these shape functions satisfy the following conditions,
at x = x, (where £ = 0).we ha\e N„, = and N„, = N MJ = N9) = 0, and at 1

x - Xj (where £ = 1), we have = and A'„, = N9l = Ne = 0 In addition,

1
,

by Eq (7-86) we must have

d\\ _ d\\ </£ _ I du I (TS

at (7-101)
dx d% dx L dl L

Therefore, the slope 0 within an element is given by

A (\ /l dS9 \ (\ dNu .\ (l d,\\

9 + 9' + + ,7 - 1021
\inrh U^r) (z^j -<
(z «?'
Again the reader should verify that the denvati\es of the shape functions satisfy
the follow'ing conditions at x = x, (where £ = 0), we ha\e

1
L rf£
at £ - 0 ( 7 - 103 )
1 riV,,

L di L </£

and at x = xt (where £ = 1), we ha\e

1 dNe ,
= 1
L d£
at £ = 1 ( 7 - 104 )
1 tLWhl
_ 1 dN6 ,
1 dN„
L di ~ L d£ L d£

With these shape functions in hand, we are now

r
m a position to determine the finite
element characteristics for the beam element
 ?
BEAMS 329

The Finite Element Characteristics

As stated earlier, the Galerkin weighted-residual

method will be used to obtain the
finiteelement characteristics for the beam element. Recall from Chapter 4 that the
Galerkin method requires that

( 7 - 105 )

Note that the term in the braces is really the residual that results when the assumed
parameter function for the element is substituted into the governing equation, Eq.
(7-88). The so-called weighting functions are precisely the shape functions as
embodied in the shape function matrix N.
Again it is more convenient to work in terms of the local coordinate ^ defined
by Eq. (7-96). Therefore, d\ - L and Eq. (7-105) may be written in terms of
the local coordinate E, as follows.

d£ = 0 ( 7 - 106 )

where the summation over the n elements has been intentionally dropped because
we are seeking the element stiffness matrix and the element nodal force vectors. It

should be a well-known fact by now that the summation represents the assemblage
step, which is considered later.

Equation (7-106) may be written as follows:

\_
Nr
L i: d

If we integrate the term on the left-hand side of this last equation by parts, we get

l
L
Wl d'c \ l- di1 )
'.if 40. ± (“M *
l)o d( dl \L' dir !
= f NM 17.1071

Writing Eqs. (7-83) and (7-84) in terms of the local coordinate we get

\_dM
-V ( 7 - 108 )
L ^
and

El d 2 w
( 7 - 109 )
de
Combining these last two results gives

l d ( El d 2 w
330 STRESS ANALYSIS

Therefore, with the help of Eq (7-1 10), the integrated term in Eq (7-107) becomes

Therefore, Eq (7-107) becomes

Integrating the integral on the left-hand side of this last equation by parts again
gives

1 dX T El if-n I ' d 2 N r El d 2 >v

f dt
L dt Lr d? L Jo d? L 2 d? r
(7-111)

Evaluating the integrated term in this last equation with the help of Eqs (7-109)
and (7-100), \se get

Therefore, Eq (7-111) becomes

A7(l
-vtr,)
<)
+ 1
L'
['
Jo d£
^
dZ2
5
= P N TpL di
JO
(7-11

-A/(t,)_
 j )

beams 331

N\ ith the help of Ik). (7-95) and noting ;f is not a function off. we may write l£q.
(7-112) as follows;

-V(x,)
f I
f tf
:
S7 r/'.N -Mix,)
*1

N'/lA r/f ( 7 - 113 )

• Vix,) .0

lliis last equation may tv written conciseh as

Kkr = f ( 7 - 114 )

where P is defined by

r = fi., + f;: ( 7 - 115 )

and K 1-

is the element stiffness matrix given by

r
...
K .if rfvP f* r/=N r/-N
^ ( 7 - 116 )
A' in f /t-
'
df :

In liq. (7-1 15) p, ,, js the clement nodal force vector as a result of shear forces and
moments and is defined b\

- V(.\,)

-Mix,)
fh ( 7 - 117 )
Tr.rr

and f£ is the element nodal force vector as a result of the distributed load p and is

defined by

0. ( 7 - 118 )

Let us now evaluate Kr and ff„ Working on Kr first, vve note that

-6 !2£
<t
2
ST At-4 -r 6£)
( 7 - 119 )
r/£
:
6 - 12f
At - 2 + 6f

If the right-hand side of liq. (7-1 19) is substituted into Eq. (7-1 16). the integrand
multiplied out. and the integral integrated and evaluated, vve get

~ 12 6A - 12 6A
;

AY 6A 4A- -6A 2A 2
1
K r
( 7 - 120 )
A' - 12 ga r
I
12 bL
_ 6A 2A : i -6A 4A- J

Note that El was assumed to be constant for the element. If El is variable, a suitable

average value could be used.

332 STRESS ANALYSIS

Next the element nodal force vector fjj as a result of the distributed lateral
loading p is evaluated. Let us take an average value of the distributed load (written

p) for the element, which is defined as

_ = P ( *.) + />(*,)
;> (7-121]

Then Eq. (7-118) becomes

'(l-3g 2 +-2? J )</«

Jo
pi/ 2
2 3

{p Pi N r
= pL
f\«-25 + i )</S pi 3 / 12
</{ (7-122)
/0
pi/2
<3{
2
-2£ J
M«
2
-pi / 12
°
i (-| 2 + ? 3 ></4
j

Remarks

The assemblage step is routine and is performed in the manner illustrated in Chapter
3 for the two-dimensional truss model Imposed external moments and point loads
may be readily applied via ffo, providing that a node is placed at such points The
point loads are applied via the shear force V(x,) or V{ r,) Application of the prescribed
deflections and slopes is also routine, and Method 1 is again recommended The
solution of Ka = f for the nodal deflections and slopes may be obtained either by
the matrix inversion method or with the help of subroutine ACTCOL |9) given in

Appendix C These nodal deflections and slopes may be used to obtain the element
resultants, as explained below
In this case, the element resultants include the internal moments M , shear forces
V, maximum longitudinal stresses a m „, and maximum shear stresses Tmax at each
point along the beam The internal moments M may be evaluated from

EI(T N ,
a (7-123)
l 2 de

and the shear forces V from

(7-124)
Ld% \L 2 d?)

Note that these last two equations imply that M

vanes linearly over the element e,
whereas V is constant The maximum longitudinal stress (or bending stress) occurs
 BEAMS 333

in the outermost fibers of the beam (i.e., the upper or lower surface) and is given
by

<W = Me — (7-125)

where c is the distance from the neutral plane to the outermost fibers of the beam
as shown in Fig. 7-17. The maximum shear stress T max occurs in the neutral plane
and is given by

V y dA
Tu.ix
= —^ I'

lb
(7-126)

where b is the width of the beam in the neutral plane. Note that in these equations,
we would use the results from Eqs. (7-123) and (7-124) to provide M and V at any
value of .v.

Example 7-7 below illustrates the use of the material in this section and intro-
duces a new simple element — the linear spring element. In this example, one point
on the beam is supported by a spring with a known spring constant ks . Up to now,
only one type of element has been used in any one problem or model. Example
7-7 shows how we can easily model systems with a variety of elements. Large
structural analysis programs such as NASTRAN [10] and STARDYNE [11] have
hundreds of different types of built-in elements.

Example 7-7

Solve for the deflections and slopes at the ends and midspan of the beam shown
in Fig. 7- 19(a). The distributed load p is 4800 N/m and the point load P is 3000
N. The left end is cantilevered and the right end is attached to a linear, elastic
spring with a spring constant ks of 200 kN/m. The beam has a rectangular cross

section with a width b of 3 cm and a height h of 4.31 cm and is made of steel with
an elastic modulus E of 2.0 x 10” N/m 2 Also determine the moment, shear force,
.

and maximum bending and shear stresses at the middle of the distributed load.
Assume L, = Z, 2 = 1 m.

Solution

First, we must determine the moment of inertia / for the beam:

l = Znbh 3 = »/ia (0.03)(0.0431)

3

or

I = 2.00 x 10^ 7 mJ
Therefore, the flexural rigidity is given by

3
El = (2 x 10")(2 x 10' 7 ) = 40.0 x 10 Nnr
334 STRESS ANALYSIS

The element computations are summarized below. Note that only two equal-length
beam elements are used (here L — ] m) in addition to the spring element as shown
in Fig. 7- 19(b).

Element 1

From Eq (7-120), we may compute

- 12
40 X
<1)
!
JO 1 r
m
12

-n
6(1)
40 )2
-6(1)
-60)
12
6(1)
20) 1
-6(1)
[_6(D 2 <n3 -6(1) 4(1)
2

480 240 -480 240

240 160 -240 80
- 480 x I0 3
-240 480 -240
L 240 80 -240 160

and from Eq (7-122),

2.4“
(4800X1 )/2
(4800)(
2
04
f i) = ) / 1
1
3
x 10
(4800)0 )/2 2 4
_-(4800)(l)’/12_ _
-0 4_

Figure 7-19 Beam used in Example 7-7 (a) Loading and restraints and (b) discretization
for the FEM solution
 1

BEAMS 335

if the point load P is considered in the computation of ff,Ar for element we set
1,

r o i

Lo J
Therefore, the composite nodal force vector for element 1 is given by

2.4

x 10
3

5.4
-0.4

Element 2
The stiffness computation for element 2 is identical to that of element There is 1 .

no distributed load acting on element 2. so fj* is zero. Moreover, the point load (

was considered in element 1 . so f$, is zero as well. (How would these computations
be modified if the point load is considered in the computations for element 2? Is
the same result obtained for the assemblage nodal force vector?)

Element 3
The stiffness relationship for the spring element is given by

1 -I 200 -200
K' J>
- 1 -200 200
L

This follows immediately from the direct approach illustrated in Chapter 3 for the
truss element (AEIL for the truss element is analogous to the spring constant ks ).
The assemblage of these element stiffness matrices may be more readily un-
derstood if it is given symbolically as follows if we let k = £7/Z.
3
:

12k 6 kL - Ilk 6 kL 0 0 0
6kL 4 kL z -6 kL 2kC- 0 0 0
- 12 k -6kL 24k 0 - 12 kL 6 kL 0
6kL 2kLr 0 8 kL
2 -6kL IkL 2 0
6 0 - 12 k — 6kL 12 k + k. -6 kL
0 0 6kL IkL 2 -6 kL 4kL 2 0
0 0 0 0 -A, 0 A, J
where the vector of nodal unknowns a is implied to be

2
a = [
H'j 0j ;
U*2 02 j
U’3 03 1 IV4 J
336 STRESS ANALYSIS

Note how the element stiffness matrix for the spring element is incorporated into
the assemblage stiffness matrix. Numerically, we have

480 240 - 480 240 0 0 0 24

240 160 - 240 80 0 0 0 04
-480* -240 " " 0~ 240’ ' -*5*4
965 ’-480 ’
U*“
240 „ „80 - 0 320 -240 80 0 -04
....
-240' -200*
“htjo -24b
0 0 240 80 - 240 160 0 0
'"5 0* 0 0 -"200' ”0 200 0

But W| = 0, 0 (
— 0, and »v4 = 0 must be imposed, and if Method I is used from
Chapter 3, the result is

0 0 “| 0 “1
0 0 0 0
r 1 IV,

0 1 0 0 0 El 0, 0

1
0 0 960 0 240 0 w7 54
0 0 0 El 8? = fW'fH
0 0 -480 -240 fa WJ
0
0
0
0
mrv.m
0
HI 0 0 0
Efl
>_ _
8j
U4
’
0
_ _ J
Solving for the nodal unknowns yields

w’i =0 iv2 = 0 01166 m u>, = 0 00680 m u’4 = 0

6, = 0 e2 = 0 00648 rad 0} = -0 01052 rad

The moment in the beam at the middle of the distributed load (where = Vi

for element 1) is determined from Eqs (7-123) and (7-1 19) as follows

El d N 2 7
El
M= LJ
7 — 2
'l

1
a <0 - 7Y (0»' -
L
r
LB, + 0»-2 + LB2 )
1/2

- (40 x 10 3 )(0 00648 - 0

= gtg2 Q|) _ )
~
L (1)

M = 259 N-m
Therefore, the maximum bending stress at this location is

_ Me _ 059X0 0431/2)
-7 , 2?g x totNM ,
/ 2 X 10

The shear force V at this point is computed from Eq (7- 1 24) as follows

1 d El d 2N
d? {=1 ,
 —
BEAMS 337

-El
V = —jj~(\2w + x
6L0, - 12w2 + 6L0 2 )

= -40 x 10 3 [12(0) + 6(1 )(0) - 12(0.01166) + (6)(0.00648)J

or

V = 4040 N
Therefore, the maximum shear stress at this location is

Vi/2

V y dA 2
ha
Jo Vb (y !2) Vlr
lb lb 8/

or
2

Tm
max
= —
(4040X0.043
-- 1)
t
-7
1 =
-
4.7 X 10 6 N/m 2 E3
“ 8 2 x 0 )

7-6 SUBSTRUCTURIIMG

Occasionally, the structure to be analyzed results in a model that is too large to be

solved even on the largest computers. In this section we will see how it is possible
to analyze such a structure by breaking it into several substructures. For example,
the Boeing 747 airliner was analyzed by breaking the airplane into four parts, or
substructures [12]. This approach is referred to as substructuring and is explained
below. Substructuring can also make it feasible to solve practical structural analysis
problems on microcomputers with limited memory.
In an effort to make this discussion a little less abstract, consider the two-
dimensional region shown The ultimate goal is to find the nodal
in Fig. 7-20(a).

displacements and the element resultants. The concept of substructuring requires

that we divide the original region into two or more parts as shown in Fig. 7-20(b).

Let us concentrate on the first substructure and temporarily disregard the second.
The element characteristics for this substructure are determined and assembled in
the usual manner to form K"a" = f" for this substructure only. Actually, the
assemblage step is not performed in exactly the usual manner; instead, we assemble
K" and f" such that in partitioned form we have

Kfe
(7-127)
K", Kfe a"

where a',' contains the displacements of the nodes that are on the boundary of the
substructure, and a" contains the displacements of the internal nodes. The displace-

ments of the internal nodes may be eliminated as described below.

If the second matrix equation in Eq. (7-127) is written out and solved lor

a?, we get
338 STRESS ANALYSIS

Figure 7-20 Two-dimensional region illustrating the concept of substructurmg (a) Original

mesh, (b) divided into two substructures, and (c) creating two supplements

a? = (K&r'ff? - Kf.af! (7-128)

But the first matrix equation in Eq (7-127) implies

Kf,af + Kf2 aJ = f|* (7-129)

Eliminating a 2 with the help of Eq (7-128) and rearranging the result gives

[Kf, - Kf,(K<y-'K?,ja? = tf - Kf,<K&r'ff 17-1301

which is of the form

K £a e = f
E (7-131)

where the superscript used to denote the superelement created by eliminating

E is
£ symmetric. The
the internal degrees of freedom It can be shown that K is still
 STRESS ANALYSIS PROGRAM
339

process summarized by Eqs (7-127) to (7-131) is referred to as condensation and

may be performed on each substructure The resulting superelements shown in Fig.
7-20(c) ha\e nodes only on the boundaries The stiffness matrix K £ and nodal force
vector f E for each superelement E are assembled in the usual manner to form the
assemblage stiffness matrix K ' and assemblage nodal force vector f A The pre-
scribed displacements are also imposed in the usual manner (Method 1 from Chapter
3 is The resulting system of equations in Ka = f is solved for the nodal
preferred)
displacements However, the vector a contains onlv the displacements for nodes
on the boundaries of the superelements In order to obtain the nodal displacements
of the internal nodes, we mav use Eq. (7-128) for each substructure, where af is
now known from the solution of Ka = f With the nodal displacements now' known
for each of the original nodes, the element resultants may be obtained in the usual
manner
Note that Eq (7-130) requires the inverse of KN. which can be a rather large
matrix in practical problems Subroutine ACTCOL [9] may be used to obtain this
inverse bv successively solving K"2 x = 8. where 8 is taken to be [1 0 0 ]
r

to get the first column in (K"?)' 1

. then [0100 ]
r
to get the second
column in (K 22 )' and so forth The triangular decomposition of K 22 needs to be
performed only once (which is accomplished by setting AFAC to FALSE after
the first decomposition) Since a large portion of the solution time is spent on the
triangular decomposition phase, obtaining the inverse in this way is quite practical
In addition to the obvious advantages of substructuring just mentioned, there
are others For example, the analysis of large structures can be performed by several
different groups of engineers, with each group responsible for developing the finite
element model for one of the substructures Obv lously. each group must use the
same number of nodes at the same locations along the substructure interfaces
Moreover, those portions of the design, and hence the model, that are finalized
need not be regenerated during each computer run. Only those portions of the design
that are being changed need to be modeled during each run. Tins can result in an
obvious economic adv antage w hen large and complex models are used
Substructunng may also be performed quite readilv in all nonstructural appli-
cations. such as in thermal and fluid flow analyses In addition, substructunng is

not limited to static or steady-state analyses After mastering Chapter 10. the reader
should have little difficulty in applying substructunng to dynamic structural and
transient thermal problems

7-7 DEVELOPMENT OF A TWO-DIMENSIONAL STRESS ANALYSIS

PROGRAM: PROGRAM STRESS

Some helpful hints and comments are given m this section so that the reader can

develop a two-dimensional stress analysis program with further from

instructions

the instructor
1
Specific comments are made with respect to the main program,

mesh generation, and data storage

'A two dimensional stress analjsts program and a user s manual are included m the lnstructor/Solution s

manual for this text

340 STRESS ANALYSIS

The Main Program

The mam program should be kept as brief as possible. An example is the following
seven lines of code:

PROGRAM STRESS
COMMON XL 5500 )
(

MAX = ESOD
LCONSL = 3
CALL DRIVER (MAX, LCONSL)
CALL EXIT
END

The vanable MAX represents the length of the XL array, the contents of which
are described below This vanable should be assigned the value of the dimension
of XL in the unlabeled COMMON Much larger problems may be solved with the
program by increasing this parameter to the memory limit of the computer being
used.The vanable LCONSL (as in the TRUSS program) represents the logical unit
number for the console. Control is then transferred to subroutine DRIVER.

Mesh Generation

The same type of mesh generator discussed in Sec 3-6 may be used in this program
except that the nodes no longer need to be equally spaced This is accomplished
with the help of two spacing factors t and f (FX and FY in the program) These
factors are used as described below
Consider the starting node NI and the final node NF shown m Fig 7-2 (a) 1

Additional nodes can be generated between these two end nodes if a nonzero value
of NG is used The nodes need not be equally spaced, however Without loss of
generality, let us number the nodes 1,2, n, as shown in Fig 7-21(b) and ,

refer to the x and y coordinates of node I by jr,, y lt etc Then the spacing factors

fx and f may be defined by

~ ~ ~ Xn ~'
fx =
*3 Xl
= x* .h _ . . . _ _ "
— Xn _2
- (7-132)
x2 “ Xy X3 — r2 X„_l

and

y. - -VrJ_
L = = . . . = ( 7- 133)
^2 - Jx >3 - >2 y„-\ -
'
y n -2

Concentrating on the x direction for now, we may wnte

X2 ~ *1 = *2 ~ *1
(7-134a)

*3 - *2 = fx {\2 ~ *|)
(7-134b)
 1

STRESS ANALYSIS PROGRAM 34

lb>

Figure 7-21 (;t) Starting notlc XI and ending node XI -

are uved to generate additional
nodev. <h) l.ine along w Inch n mxles are generated (not neccssarih equallv spaced).

A4 ~ li = /,(' 1 ~ •»’) = /,'('; — A,) (7-134c)

A/* - A..-1 ~ /.<•'„ 1

- A„ = ) ~ A,) (7-134d)

Adding these results gives

a,. - i, = 1
i •*•/,•* f: - •
•*+/,' " a,) (7-135)

from which it follows that

a„ - A j

n -2 (7-136)

I + S f-
i=l

Since a, and i„ may be input (as XI and XF. respectively) and since /, may be
input (as FX). then ,\
;
can be found front Eq. (7 -13b). Note that » is the total
number of nodes in the generating sequence (including NI and NF). Obviously ,\
3
can be computed from Fq. (7- 134b). or

A, = A; T /,(a I - .1,) (7-137)

and so forth. It should also be obvious that in they direction the same results hold

except that each x, is replaced by and f t

is replaced by/,. Note that if /, (or f)
is greater than unity, the nodes arc spaced farther apart in moving from NI to NF;
if /, (or /) is less than unity (but greater than zero), the spacing between two
consecutive nodes decreases in moving from NI to NF.

Data Storage

The data for the nodal coordinates, elements, boundary conditions, material prop-
erties, surface tractions, point loads, and so forth, should be stored in the XL array.
In addition, the assemblage stiffness matrix (in column vector form), the assemblage
nodal force vector, and the diagonal pointer array (JDIAG from Sec. 6-8) should
342 STRESS ANALYSIS

also be stored in the XL array. The partitions between each of these sections should
float,and if fewer nodes are used, for example, more materials may be present.
The number of storage locations required must not exceed MAX. If it does, an
error message should be displayed on the console and execution should be termi-
nated.
The numbering of the nodes is critical if the program does not renumber the
nodes to minimize the bandwidth of the assemblage stiffness matrix. However, the
program should store only the banded portion of these matrices, which reduces the
storage requirements drastically over storing the full matrices Nodes on the object
being analyzed are always numbered consecutively, from 1 to the maximum number
of nodes. The bandwidth is minimized when the maximum difference between any
two nodes on each element is minimized Figure 7-22 shows that this is accomplished
quite simply by numbering the nodes m the direction of fewer nodes
Note that in Fig 7-22(a) the nodes are numbered in the direction of the smaller
number of nodes In Fig 7-22(b) the nodes are numbered to the right and in the
direction of the greater number of nodes. The storage requirements for the mesh
in the latter are higher than for the mesh in the former In Fig 7-22(c) the nodes
are numbered in an alternating sweeping fashion, which more than doubles the
storage requirements over that required in Fig 7-22(a) Finally, in Fig 7-22(d) the
element on the lower left has nodes 1 , 2, and 2 1 , the implication is that the stiffness
matrix no longer banded, which in turn means higher storage requirements and
is

increased execution times Although the program need not store the zero terms in

19 ,18 17 16

Proper node numbering scheme (b) Less-desirable node numbering

scheme
Figure 7-22 (a)

(c)Nodes should never be numbered in an alternating sweeping fashion (d) Worst node
—
numbering scheme results in an unhanded stiffness matrix
 .

REFERENCES 343

Figure 7-23 (a) Preferred way of dividing a quadrilateral into two triangles and (b) less-
desirable way of forming two triangles.

the stiffness matrix outside the bandwidth, it should store everything within the
bandwidth (including leading zeros).
Numbering the elements is not critical, but some sort of regular numbering
scheme usually allows the use of an automatic element generation feature. Note
that in forming the elements shown in the lower right-hand corner of Fig. 7-22(a),
the quadrilateral formed by nodes 17, 20, 21, and 18 is divided into two triangles
by using the shorter diagonal (nodes 17 and 21) as opposed to the longer diagonal
(nodes 18 and 20). This is summarized by Fig. 7-23. Regular triangles generally
give better results than obtuse or needle-shaped triangles.

7-8 REMARKS

This chapter illustrates how the finite element method is used in static, linear stress

analysis. The formulations for plate bending and nonlinear problems are conspic-
uously absent. For these applications, the reader is referred to more advanced books
on this subject by Zienkiewicz [13], Bathe [14], and Ugural [15],

REFERENCES

1. Ugural, A. C.and S. K. Fenster, Advanced Strength and Applied Elasticity, American

,

Elsevier, New
York, 1975, pp. 57-58.
2. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977,

pp. 99-101.
3. Ugural, A. C., and S. K. Fenster, Advanced Strength and Applied Elasticity, American
Elsevier, New York, 1975, pp. 75-77
4. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977,
p. 125.
5. Huebner, K. H., The Finite Element Method for Engineers, Wiley, New York, 1975,

pp. 218-219.
6. Wilson, E. L., “Structural Analysis of Axisymmetric Solids,” AlAA Journal, vol. 3,

no. 12, pp. 2267-2274, December 1965.

344 STRESS ANALYSIS

7. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977,
p 140.
8. Popov, E P., Introduction to Mechanics of Solids, Prentice-Hall, Englewood Cliffs,
N.J., 1968, pp 32-36
9. Zienkiewicz, O C., The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp. 740-741
10. MSC/NASTRAN, User's Manual, MacNeal-Schwcndler Corporation, Los Angeles, 1979.
11. STARDYNE Scope 3 4 Operating System, User Information Manual, Cybernet
for
Services, Control Data Corporation, Minneapolis, 1978
12. Hansen, S D , et al , “Analysis of the 747 Aircraft Wing-Body Intersection,” Pro-
ceedings of the 2nd Conference on Matrix Methods in Structural Mechanics, Wnght-
PatterSon Atr Force Base, Dayton, Ohio, October 15-17, 1968
13. Zienkiewicz, O. C . The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp 226-267,450-526
14. Bathe, K , Finite Element Procedures in Engineering Analssis, Prentice-Hall, Engle-
wood Cliffs, NJ , 1982, pp 251-260
15. Ugural, A C , Stresses in Plates and Shells McGraw-Hill, New York, 1981 pp 126-
. ,

136

PROBLEMS

Note. The properties in Appendix A should be used unless stated otherwise.

7-1 Show that the B matrix for the three node triangular element in plane stress or plane
strain is given by Eq (7-7)

7-2 Dense the B matrix for the cases of plane stress and plane strain for the four node
rectangular element. How does this matnx differ from that for the three-node tn-

angular element 0

7-3 For an isotropic matenal, how is the shear modulus related to the modulus of
elasticity and Poisson’s ratio 0

0
7-4 Why is the normal strain e_ not necessarily zero for a thin plate in plane stress
Give a specific example
 PROBLEMS 345

7-5 Why is the normal stress ov. not necessarily zero in a long bar in plane strain? Give
a specific example.

7-6 Repeat Example 7- 1 for the case of plane strain. Since the member is now assumed
to be long in the longitudinal direction, take t to be unity. Useas much of Example
7-1 as possible.

7-7 Consider the triangular element shown in Fig. P7-7. The plate from which the
element is extracted is made of cast iron and has a thickness of 0.5 in. The nodal
coordinates are .V, = 2.0, v, = 1.5 .a, = 1.7, v, = 3.0, xL = 0.6, andv* = 1.8 in.
Determine the element stiffness matrix if all external forces act m the plane of the
plate (and hence in the plane of the element).

Figure P7-7

7-8 Repeat Problem 7-7 for the case of plane strain. Take i to be unity,

7-9 Consider the triangular element shown in Fig. P7-9. The plate from which the
element is extracted is made of brass and has a thickness of 1 cm. The nodal
coordinates are x, — 5, v, = 6, \j = 4. y; = 4. ay = 6, and \\ = 4 cm. Determine
the element stiffness matrix if all external forces act in the plane of the plate (and

hence in the plane of the element)

7-10 Repeat Problem 7-9 for the case of plane strain. Take / to be unity.
346 STRESS ANALYSIS

7-11 Consider the triangular element shown in Fig P7- i I . The element is extracted from
a thin plate of thickness 0 5 cm The material is hot rolled, low carbon steel The
nodal coordinates are x, ~ 0, y, — 0, jc = 0, \ = — 1 , rt = 2, and y* *= — 1 cm
y }

Determine the clement stiffness matrix if all external forces act in the plane of the
plate (and hence in the plane of the element)

Figure P7-11

7-12 Repeat Problem 7-1 1 for the case of plane strain Take t to be unity

7-13 Consider the triangular clement shown in Fig P7-13 The element is extracted from
a thin plate of thickness 0 75 in The material is hard drawn copper The nodal
coordinates are x, = O.y, = 0, x,
- J.j, = 2, xt - -1, andj t = 2 in Determine
the element stiffness matrix if all external forces act in the plane of the plate (and
hence in the plane of the element)

Figure P7-13

7-14 Repeat Problem 7-13 for the case of plane strain Take t to be unity

7-15 Show that the element stiffness matrix for plane stress and plane strain as given by
Eq (7-24) for the three-node triangular element is always symmetric

7-1 6 Show that the element stiffness matrix for plane stress and plane strain for the four-

node rectangular element is always symmetric

7-17 What is the size of the element stiffness matrix for the case of plane stress (or
plane

strain) if the four-node rectangular element used 0 Explain your answer

is

The
7-18 Determine the element stiffness matrix for the element shown in Fig P7-18.
thickness o
etement is extracted from an aluminum (6061 alloy) plate with a
 PROBLEMS 347

1.25 cm. All external forces are in the plane of the plate. The coordinates of the
nodes are x, = 4. \, = 2. ,v
;
= 4. = 3. v ; = 2. v t = 3. .v,_ = 2, and =
2 cm. Perform the integrations by evaluating the integrands at the element centroid
and treating the integrand as a constant

Figure P7-18

7-1S Repeat Problem 7-18 for the case of plane strain. Take / to be unity.

7-20 Determine the element stiffness matrix for the element shown in Fig. P7-20. The
element is extracted from a brass plate with a thickness of 0.375 in. All external

forces arc in the plane of the plate. The coordinates of the nodes are x, - 5. y, =
2. Xj = 5. v. = 4, a, = 2. \\ = 4, ,r,_ = 2. and y„ = 2 in. Perform the integrations
by evaluating the integrands at the element centroid and treating the integrand as a

constant.

Figure P7-20

7-21 Repeat Problem 7-20 for the case of plane strain. Take i to be unity.

7-22 What size the element nodal force vector from a self-strain if the four-node
is

rectangular element is used in plane stress or plane strain formulations?

uniform
7-23 Determine a general expression for the element nodal force vector from a
rectangular element in plane stress. Perform the integrations by
self-strain for the
integrands as though
evaluating the integrand at the element centroid and treating the
they were constants.

7-24 Repeat Problem 7-23 for the case of plane strain.

7-25 For the element in Problem 7-7, determine the element nodal force vector as a result

the temperature increases by 50'F. Assume

plane stress.
of a self-strain if
348 STRESS ANALYSIS

7-26 For the element in Problem 7-7, determine the element nodal force vector as a result
of a self-strain if the temperature increases by 50°F. Assume plane strain and take
I to be unity.

7-27 For the element in Problem 7-9, determine the element nodal force vector as a result
of a self-strain if the temperature increases by 30°C. Assume plane stress.

7-28 For the clement m Problem 7-9, determine the clement nodal force vector as a result
of a self-strain if the temperature increases by 30°C Assume plane strain and take
l to be unity

7-29 For the element in Problem 7 11. determine the element nodal force vector as a
resultof a self-strain if the temperature decreases by 40°C Assume plane stress

7-30 For the element in Problem 7 11. determine the clement nodal force vector as a
result of a self-strain if the temperature decreases by 40°C. Assume plane strain
and take Mo be unity
7-31 For the element in Problem 7 13. determine the element nodal force vector as a
result of a self-strain if the temperature decreases by 75°F Assume plane stress.

7-32 For the clement inProblem 7-13, determine the element nodal force vector as a
6
result of a self-strain if the temperature decreases by 75 F Assume plane strain and
take l to be unity

7-33 For the element in Problem 7 18. determine the element nodal force vector as a
6
result of a self-strain if the temperature decreases by 35 C Assume plane stress
and perform the integrations by evaluating the integrands at the element centroid
and treating the integrands as though they were constants

7-34 For the element in Problem 7 18, determine the element nodal force vector as a
result of a self-strain if the temperature decreases by 35°C Assume plane strain
with t taken as unity and perform the integrations by evaluating the integrands at

the element centroid and treating the integrands as though they w ere constants

7-35 For the element in Problem 7-20, determine the element nodal force vector as a
result of a by 80°F Assume plane stress and
self-strain if the temperature decreases

perform the integrations by evaluating the integrands at the element centroid and

treating the integrands as though they were constants

7-36 For the element in Problem 7-20. determine the element nodal force vector as a
result of a by 80°F. Assume plane strain
self-strain if the temperature decreases
integrands at
with t taken as unity and perform the integrations by evaluating the

the clement centroid and treating the integrands as though they w ere constants

7-37 Determine a general expression for the element nodal force \ ector from a uniform
prestress for the rectangular element in plane stress or plane strain Perform the
the
integrations by evaluating the integrand at the element centroid and treating
integrands as though they were constants

7-38 For the clement in Problem 7-7. determine the element nodal force vector as a result
of the following prestresses- crxn = 1000,
,
= -750. and o- no = 500 psi
 PROBLEMS 349

7-39 For the element in Problem 7-9, determine the element nodal force vector as a result
of the following prestresses: <r no = -2000, <r„ 0 = 1200, and o tTO = -1500
N/cm 2 .

7-40 For the element in Problem 7-11, determine the element nodal force vector as a
result of the following prestresses: cr ira = 1300, cr
u„
= -800, and <r
Kn = 0
N/cm 2 .

7-41 For the element in Problem 7-13, determine the element nodal force vector as a
result of the following prestresses: or„
0 = 4200, a ut) = -2800, and cr
lv,
=
- 1400 psi.

7-42 For the clement in Problem 7-18, determine the clement nodal force vector as a
result of the following prestresses: o-„
0 = -3400, <rufl = 1800, and o n0 -
-2200 N/cm 2 . Perform the integrations by evaluating the integrands at the element
centroid and treating the integrands as though they were constants.

7-43 For the element in Problem 7-20, determine the element nodal force vector as a
result of the following prestresses: ct u0 = - 400, aU() = 800, and o ml = - 1200 psi.

Perform the integrations by evaluating the integrands at the element centroid and
treating the integrands as though they were constants.

7-44 Determine a general expression for the element nodal force vector from a uniform
body force for the rectangular element in plane stress or plane strain. Perform the
integrations by evaluating the integrand at the element centroid and treating the
integrands as though they were constants.

7-45 For the element in Problem 7-7, determine the element nodal force vector as a result
of a body force with the following components: b — 100 and bs = 0 lbf/in 3
x
.

7-46 For the element in Problem 7-9, determine the element nodal force vector as a result

of a body force with the following components: b y = 500 and b x

= 700 N/cm’.

7-47 Assume that the element in Problem 7-11 is oriented such that the y axis is in the
direction opposite that of gravity. If there are no other body forces present, determine
the element nodal force vector. The acceleration due to gravity is 9.81 m/s 2 .

7-48 Assume that the element in Problem 7-13 is oriented such that the x axis is in the

same direction as that of gravity. If there are no other body forces present, determine
2
the element nodal force vector. The acceleration due to gravity is 32.2 ft/s .

7-49 For the elementin Problem 7-18, determine the clement nodal force vector as a

body force with the following components: b x = 500 and

result of a = 0
N/cm\ Perform the integrations by evaluating the integrands at the element centroid
and treating the integrands as though they were constants.

7-50 Assume that the element in Problem 7-20 is oriented such that the y axis is in the

direction opposite that of gravity. If there are no other body forces present, determine
2
the element nodal force vector. The acceleration due to gravity is 32.2 ft/sec .

Perform the integrations by evaluating the integrands at the element centroid and

treating the integrands as though they were constants.

7-51 Determine a general expression for the element nodal force vector from a uniform
surface traction on face ij for the rectangular clement in plane stress or plane strain.
350 STRESS ANALYSIS

Perform the integrations bs evaluating the integrand at the center of face ij and
treating the integrand* a* though thes were constants

7-52 For the element in Problem 7-7. determine the clement nodal force sector as a result
of a surface traction on leg ij with the following components: s, - 1000 and s, =
200 psi

7-53 For the element in Problem 7-9. determine the clement nodal force s ector as a result
of a surface traction on leg jL with the follow mg components: i, = 1500 and t, =
800 N’cnr
7-54 For the element m Problem 7-11. determine the element nodal force sector as a
result of a surface traction on leg Li with the following components- s, = 1200 and

j,
- 750 N cm 1
7-55 For the element in Problem 7-13. determine the element nodal force sector as a
result of a surface traction on leg jl with the follow mg components s, ~ 4200 and

j, = ISOOpsi

7-56 For the element in Problem 7-18. determine the element nodal force sector as a
result of a surface traction on face ij ssuh the following components, s, = 1500

and s, = 500 N cm ; Perform the integrations b> evaluating the integrands at the

centroid of face ij and treating the integrands as though thes w ere constants

7-57 For the element in Problem 7-20. determine the element nodal force sector as a
result of a surface traction on face mi with the following components s, = 1100
and r, = 1 500 psi Perform the integrations bs es aluating the integrands at the

centroid of face mi and treating the integrands as though thes were constants

7-58 It has been assumed in the derivation leading to Eq (7-33) in the test that the
surface traction is uniform oserthe leg of the tnangle in question Dense an alternate
expression for the nodal force sector for a linearis varying surface traction on leg
ij of the ihrec node tnangular element shown m Fig P7 5S Proceed bs assuming
the following forms for s, and s

s, = A -» V,s n

5. = NA, Vy
where s, is the r component of the surface traction at node i. etc Write the shape
functums in terns of the length -eoooiwtates t defined on leg yl and use Eq (6-4S)
to perform the integrations

Figure P7-58
 PROBLEMS 351

7-59 Repeat Problem 7-58 for a uniformly varying

surface traction on leg jk At nodes
j and k. the a and y components of the surface
traction are s and .v,, and j and
„ v
A*, respectively.

7-60 Repeat Problem 7-59 for a uniformly varying surface

traction on leg ki. At nodes
k and t, the .t and
y components of the surface traction are s xk and .v,„ and and ,v
'
,

su respectively.
,

7-61 For the element in Problem 7-7 determine the element

, nodal force vector as a result
of a point load that has the following components:
pt ~ 2000 and/, = 1200 f lbf
The point load is located at ,r0 = 1.5 and y„ = 2.0 in.

7-62 For the element in Problem 7-9, determine the element

nodal force vector as a result
of a point load with the following components:
fpx = 1000 and /,,, = 500 N. The
point load is located atxg = 5 andy = 5 cm.
0

7-63 For the element in Problem 7-11, determine the element nodal
force vector as a
of a point load with the following components:/,, ~ 2400 and/,, = —
result
1500 N.
The point load is located at x0 = 2 and = - 1 cm. y0
7-64 For the element Problem 7-13, determine the element nodal force vector as a
in
result of a point load with the following components:/,, =
-1250 and/, =
- 1800 lbf. The point load is located at ,v = and
0 y0 = 2 in. I

7-65 For the element in Problem 7-18, determine the element nodal force vector
as a
of a point load with the following components:
/, = 2330 and / = - 1500 N.
result

The point load is located at,r0 = 3.0 and 0 = 2.5 cm.

y
7-66 For the element in Problem 7-18, determine the clement nodal force vector as a
resultof a point load with the following components:
fpx = - 2330 and /, =
1500 N. The point load is located at x0 = 4 and 0 = 2 cm.
y
7-67 For the element Problem 7-20, determine the element nodal force vector as a
in

of a point load with the following components: fpx ~ - 2000 and

/ =
result
1000 lbf. The point load is located at x0 = 4 and y0 = 3 in.

7-68 For the element in Problem 7-20, determine the clement nodal force vector as a
result of a point load with the following components: f
p = 2500 and/,, = -3000 lbf.
,

The point load is located at x0 = 2 and y0 = 4 in.

7-69 Using the symbolic notation from Example 7-5, give the expression for the assem-
blage stiffness matrix for the discretized two-dimensional region in Fig. P7-69.
What is the half-bandwidth? Can the half-bandwidth be reduced? If so, explain
how.

Element Nodes
number i 1 k

1 4 1 2
2 1 4 3

Figure P7-69
352 STRESS ANALYSIS

7-70 Using the symbolic notation from Example 7-6, give the expression for the assem-
blage nodal force vector for the discretized two dimensional region in Fig P7-69

7-71 Using the symbolic notation from Example 7-5, give the expression for the assem-
blage stiffness matrix for the discretized two-dimensional region in Fig P7-7I.
What is the half-bandwidth"* Can the half bandwidth be reduced *
1

If so, explain
how

Figure P7-71

7-72 Using the symbolic notation from Example 7 6, give the expression for the assem-
blage nodat force vector for the discretized two-dimensional region in Fig P7-71

7-73 Using the symbolic notation from Example 7-5, give the expression for the assem-
blage stiffness matrix for the discretized two dimensional region in Fig P7-73
What is the half-bandwidth 7 Can the half-bandwidth be reduced 7 If so, explain
how

Figure P7-73

7-74 Using the symbolic notation from Example 7-6, give the expression for the assem-
blage nodal force vector for the discretized two dimensional region in Fig P7-73

7-75 By extending the symbolic notation from Example 7-5, give the expression for the
assemblage stiffness matrix for the discretized two dimensional region in Fig
P7-
7 explain
75 What is the half-bandwidth 7 Can the half-bandwidth be reduced If so,

how
 PROBLEMS 353

Element Nodes
number / / k m
i 3 4 2 i

2 5 6 4 3

Figure P7-75

7-76 Be extending the symbolic notation from Example 7-6, give the expression for the
assemblage nodal force vector for the discretized two-dimensional region in Fig.

P7-75.

7-77 By extending the symbolic notation from Example 7-5, give the expression for the
assemblage stiffness matrix for the discretized two-dimensional region in Fig.
P7-77. What is the half-bandwidth? Can the half-bandwidth be reduced? If so,

explain how.

Element Nodes
number /
/ k m
1 5 4 3 2
2 6 5 2 1

Figure P7-77

7-78 By extending the symbolic notation from Example 7-6, give the expression for the
assemblage nodal force vector for the discretized two-dimensional region in Fig.
P7-77.

7-79 By extending the symbolic notation from Example 7-5, give the expression for the
assemblage stiffness matrix for the discretized two-dimensional region in Fig.

P7-79. Note the use of two different types of elements. What is the half-bandwidth?

Can the half-bandwidth be reduced? If so, explain how.

•534 ANAL I S13

7-80 By extending the symbolic notation from Example 7-6, give the expression for the
assemblage nodal force vector for the discretized two-dimensional region in Fig.
P7-79 Note the use of two different types of elements

7-81 For the element in Problem 7-7 for a particular loading condition, the following
values of nodal displacements arc obtained «,
= 0 00105, v, = 0 0, it, = -0 00055,
v, =0 00200, «* =0 00150. and v*
- -0 00405 in Determine the element strains
and stresses that correspond to these displacements if the clement is in a state of
plane stress Assume both the self-strains and prestresses are zero

7-82 Repeat Problem 7 81 for the case of plane strain Take t to be unity

7-83 For the element in Problem 7 9 for a particular loading condition, the following
values of nodal displacements are obtained u, = 0 00215, v,= -0 00300, it, =
0 0, v, = 0 00560. «*
= -0 00520. and »* = 0 00650 cm Determine the clement
strains and stresses that correspond to these displacements if the element is in a

state of plane stress Assume both the self-strains and prestresses are zero

7-84 Repeat Problem 7 83 for the case of plane strain Take t to be unity

7-85 For the element m Problem 7-11 for a particular loading condition, the following
values of nodal displacement are obtained - —0 = 0 02300, u,
—
it, 01250, v,

0 01500. \, = -0 01550. «* = 0 01750. and v* = 0 0 cm Determine theelement

strains and stresses that correspond to these displacements if the element is in a

state of plane stress Assume the self-strains and prestresses from Problems 1-29
and 7-40

7-86 Repeat Problem 7 85 for the case of plane strain Take t to be unity Assume the
self-strains and prestresses from Problems 7-30 and 7-40

7-87 For the element inProblem 7-13 for a particular loading condition, the following
u, —
values of nodal displacements arc obtained u, = 0 02150. v, = 0 02350,
-0 01575, vy = 0 02550, «* = 0 02500, and i> = 0 0 in Determine the element
strains and stresses that correspond to these displacements if the element is in a
state of plane stress. Assume the self strains and prestresses from Problems 7-31
and 7-41.

7-88 Repeat Problem 7-87 for the case of plane strain Take t to be unity. Assume the
self-strains and prestresses from Problems 7-32 and 7-41
 PROBLEMS 355

7-89 For the element in Problem 7-18 for a particular loading condition, the following
values of nodal displacements are obtained: it, = 0.02250, v, = -0.01250, u =
0.02555, Vj = 0.02350, «t = 0.02500. vk = -0.02340, i, m = 0.01235, and vm
= - 0.02500 cm. Determine the element strains and stresses at the element centroid
that correspond to these displacements if the element is in a state of plane stress.
Assume both the self-strains and prestresses are zero.

7-90 Repeat Problem 7-89 for the case of plane strain. Take t to be unity.

7-91 For the element in Problem 7-20 for a particular loading condition, the following
values of nodal displacements are obtained: u, — -0.01250, v, = -0.03050, iij

= 0.02500, v, = 0.03150, uk = -0.02500, vt = 0.03350, m ti = 0.0. and v,„ =

0.03500 in. Determine the element strains and stresses at the element centroid that
correspond to these displacements if the element is in a state of plane stress. Assume
both the self-strains and prestresses are zero.

7-92 Repeat Problem 7-91 for the case of plane strain. Take t to be unity.

7-93 Derive the B matrix for the case of axisymmetric stress analysis for the four-node
rectangular element.

7-94 Consider the triangular element shown in Fig. P7-7. The body from which the

element is extracted is made of hard drawn copper. The nodal coordinates are r, =
2.0. r, = 1.5, rt = 1.7, z, = 3.0, rL = 0.6, and rt = 1.8 in. Determine the
element stiffness matrix assuming the body is a body of revolution and is loaded
axisymmetrically.

7-95 Consider the triangular element shown inFig. P7-9. The body from which the

element is extracted is made of cast iron. The nodal coordinates are r, = 5, z, =

6, rt = 4, z, = 4, rk = 6, and z k = 4 cm. Assuming the body is a body of
revolution loaded axisymmetrically, determine the element stiffness matrix.

7-96 Consider the triangular element shown in Fig. P7- 1 1 . The element is extracted from

a body of revolution and is loaded axisymmetrically. The material is aluminum

(6061 alloy). The nodal coordinates are r, = 10, r, = 10, r, = 10, e, = 9, rt =
12, and zk = 9 cm. Determine the element stiffness matrix.

7-97 Consider the triangular element shown in Fig. P7-I3- The element is extracted from

a body of revolution and is loaded axisymmetrically. The material is cast iron. The
nodal coordinates are r, = 20, z, = 20, r, = 21, z, = 22, r, — 19, and rt =
22 in. Determine the element stiffness matrix.

7-98 Show that the element stiffness matrix for axisymmetric stress analysis as given by
Eq. (7-53) for the three-node triangular element is always symmetric.

7-99 Show that the element stiffness matrix for axisymmetric stress analysis for the four-

node rectangular element is always symmetric.

7-100 What is the size of the element stiffness matrix for the case of axisymmetric stress
analysis if the four-node rectangular element is used? Justify your answer.

7-101 Determine the element stiffness matrix for the element shown in Fig. P7-18. The
element is extracted from a cast iron body of revolution that is loaded axisym-
metrically. The coordinates of the nodes are r, = 34, z, = 32, r,
= 34, zt = 33.
356 STRESS ANALYSIS

rk = 32, 2k = 33, r„ = 32, and z„ — 32 cm Perform the integrations by evaluating

the integrands at the element centroid and treating the integrands as constants

7-102 Determine the element stiffness matrix for the element shown in Fig P7-20 The
element is extracted from a hot rolled, low carbon steel body of revolution that is

loaded axisymmetncally The coordinates of the nodes are r, = 55, ?, = 52, -

55, 2j = 54, rk = 52, zk = 54, r„ — 52, and z,„ = 52 in Perform the integrations
by evaluating the integrands at the element centroid and treating the integrands as
constants.

7-103 What size is the element nodal force vector from a self-strain if the four-node
rectangular element is used in axisymmetnc stress analysis?

7-104 Determine a general expression for the element nodal force vector from a uniform
self-strain for the rectangular element in axisymmetnc stress formulations Perform
the integrations by evaluating the integrand at the element centroid and treating the
integrands as though they were constants

7-105 For the element in Problem 7-94, determine the element nodal force vector as a

result of a self-strain if the temperature increases by 70°F.

7-106 For the element in Problem 7-95, determine the element nodal force vector as a
result of a self-strain if the temperature increases by 25°C

7-107 For the element in Problem 7-96, determine the element nodal force vector as a
result of a self-strain if the temperature decreases by 60°C

7-108 For the element in Problem 7-97. determine the element nodal force vector as a
result of a self strain if the temperature increases by 55°F

7-109 For the element in Problem 7-101, determine the element nodal force vector as a
result of a self-strain if the temperature increases by 35°C Perform the integrations
by evaluating the integrands at the element centroid and treating the integrands as
though they were constants

7-110 For the element in Problem 7-102, determine the element nodal force vector as a
result of a self-strain if the temperature decreases by 72°F Perform the integrations
by evaluating the integrands at the element centroid and treating the integrands as
though they were constants

7-1 1 1 Determine a general expression for the element nodal force vector from a uniform
prestress for the triangular element in axisymmetnc stress analysis by performing
the exact integrations Hint Use area coordinates to represent r m the following

manner

r = L,r, + L/j + Lfy

and then apply Eq (6-49)

7-1 12 Determine a general expression for the element nodal force vector from a uniform
prestress for the rectangular element inaxisymmetnc stress analysis. Perform the
the
integrations by evaluating the integrand at the element centroid and treating
integrands as though they were constants
 PROBLEMS 357

7-113 For the element in Problem 7-94, determine the element nodal force vector as a
result of the following prestresses: orrr(
,
= -800, cr e9o = 1750, cr..
0 = 1000, and
°V.- 0
= -500 psi.

7-114 For the element in Problem 7-95, determine the element nodal force vector as a
result of the following prestresses: arw = 2000. o-8eo = -3200, o\,0 = 1500, and
y na
i = - 1800 N/cm 2 .

7-115 For the element in Problem 7-96, determine the element nodal force vector as a
result of the following prestresses: vrt0 = 2300, o-
00o = 0, cr..
0 = 1550, and oy. 0
= -2100 N/cm 2 .

7-116 For the element in Problem 7-97, determine the element nodal force vector as a
result of the following prestresses: <srro = 1200, <x fJi)0 = -800, o,.a = 900, oy_.
0
= 400 psi.

7-117 For the element in Problem 7-101, determine the element nodal force vector as a
result of the following prestresses: nrm = —2400, croeo
= -1500, cr.
;0 = 2000,
and cr,_ = 500 N/cm 2 Perform . the integrations by evaluating the integrands at the

element centroid and treating the integrands as though they were constants.

7-118 For the element in Problem 7-102, determine the element nodal force vector from
the following prestresses: cr„„ = 1400, = -2800, o-..
0
= 900 and ay- =
o

1925 psi. Perform the integrations by evaluating the integrands at the element cen-
troid and treating the integrands as though they were constants.

7-119 Determine a general expression for the element nodal force vector from a uniform
body force for the triangular element in axisymmetric stress problems by performing
the exact integrations. Him: See Problem 7-111.

7-120 Determine a general expression for the element nodal force vector from a uniform
body force for the rectangular element in axisymmetric stress problems. Perform
the integrations by evaluating the integrand at the element centroid and treating the
integrands as though they were constants.

7-121 For the element in Problem 7-94, determine the element nodal force vector as a

result of a body force with the following components: br = 500 and b. = 125
3
lbf/in .

7-122 For the element in Problem 7-95, determine the element nodal force vector as a
result of a body force with the following components: br = 1500 and b. - 500

N/cm 3 .

7-123 Assume that the element in Problem 7-96 is oriented such that the r axis is in the

direction opposite that of gravity. If the body is rotated at 12 rad/sec about the z

axis, determine the element nodal force vector. The acceleration due to gravity is

9.81 m/s 2 State all assumptions made.

.

7-124 Assume that the element in Problem 7-97 is oriented such that the z axis is in the
same direction as that of gravity. If the body is rotated at 15 rad/sec about the z
axis, determine the element nodal force vector. The acceleration due to gravity is

2 assumptions made.
32.2 ft/s . State all
 358 STRESS ANALYSIS

7*125 For the element in Problem 7-101, determine the element nodal force vector from
a bodj force with the following components: b, = 1500 and b = 450 N"cm J.
Perform the integrations by evaluating the integrands at the element centroid and
treating the integrands as though thev were constants.

7-126 Assume that the element in Problem 7-102 is oriented such that the z axis is in the
direction opposite that of gravitj If the body is rotated at 20 rad'sec about the z
axis, determine the element nodal force vector. The acceleration due to gravit) is

32 2 ft'sec’. Perform the integrations by evaluating the integrands at the element

centroid and treating the integrands as though they w ere constarts.

7-127 Determine a general expression for the element nodal force vector from a uniform
Surface traction on face ij for the rectangular element in a state of axisjmmetnc
stress Perform the integrations b> evaluating the integrand at the center of face ij

and treating the integrands as though the> were constants

7*128 Determine a general expression for the element nodal force vector from a uniform
on face km for the rectangular element in a state of axisymmetnc
surface traction
stress Perform the integrations by evaluating the integrand at the center of face km
and treating the integrands as though they were constants

7-129 For the element in Problem 7-94, determine the dement nodal force vector as a
result of a surface traction on leg ij with the following components s, = - 1000
and s. = 500 psi

7-130 For the element in Problem 7-95. determine the dement nodal force vector as a
result of a surface traction on leg ki with the following components j, = - 1500
and s = -800 N 'cm 2
7-131 For the dement in Problem 7-96. determine the element nodal force vector as a
result of a surface on leg jk with the following components s, = - 1200
traction
and s = 925 N/cm 2
7-132 For the dement in Problem 7-97, determine the element nodal force vector as a
result of a surface traction on leg ki with the following components i,
= -4200
and j = 0 psi

7*133 For the dement m Problem 7-101. determine the dement nodal force vector as a
result of a surface traction on face ij w uh the follow mg components s,
= — 2350
and j. = 500 N/cm 2 Perform the integrationsby evaluating the integrands at the

center of face g and treating the integrands as though the} were constants

7-134 For the element in Problem 7-102, determine the dement nodal force vector as a
resultof a surface traction on face mi with the following components sr = - 1600
and s = 800 psi. Perform the integrations b} ev aluatmg die integrands at the center
of face mi and treating the integrands as though thev were constants.

7-135 It has been assumed in the derivation leading to Eq (7-62) in the text that the
surface traction is uniform over the leg of the mangle in question Derive an alternate
expression for the nodal force vector for a linearly xnmng surface traction on lee
ij of the three-node triangular element, shown m Fig P7-I35 Proceed by assuming
the follow mg forms for sr and j..
 PROBLEMS 359

s, - A’,.v„ + NjSrJ

v. = A',.t
;,
+ A'^v.,

wlKrc ,r
r, is the /•
component of (he surface traction at node /, etc. In addition,
represent the variable r in the integrands in terms of the nodal values of r (i.e., r,
and ry ) and the length coordinates and L r Finally, write the shape functions in
L,
terms of the length coordinates (defined on leg ij) and use Eq. (6-48) to
perform
the integrations.

Figure P7-135

7-136 Repeat Problem 7-135 for a uniformly varying surface traction on leg jk. At nodes
j and A, the r and r components of the surface traction are s rj and jr( ,
and and
.r.j, respectively.

7-137 Repeat Problem 7-136 for a uniformly varying surface traction on leg At. At nodes
A and t. the r and ; components of the surface traction arc srl and sn and s:k and
.

j.,, respectively.

7-138 For the clement in Problem 7-94, determine the clement nodal force vector as a
result of a point load that has the following components: fpr = 1200 and fp; =
-750 Ibf/in. The point load is located at r0 = 1 .2 and z0 = 2.2 in.

7-139 For the clement in Problem 7-95, determine the clement nodal force vector as a
result of a point load with the following components: fpr = - 1000 and fpl =
500 N/cm. The point load is located at r0 = 5.5 and z 0 = 4.3 cm.

7-140 For the element in Problem 7-96, determine the element nodal force vector as a

of a point load with the following components:^ = 2400 and fp: = 1500
result

N/cm. The point load is located at r0 - and z0 = 9 cm. 1 1

7-141 For the element in Problem 7-97, determine the element nodal force vector as a
result of a point load with the following components: fpr = - 2250 and fp . =
-2800 Ibf/in. The point load is located at r0 = 20 and z0 = 21.5 in.

7-142 For the element in Problem 7-101, determine the element nodal force vector as a
result of a point load with the following components:^,, = -3150 and fpl =
1500 N/cm. The point load is located at r0 = 33.0 and z 0 = 32.5 cm.
 =

360 STRESS ANALYSIS

7-143 For the element m

Problem 7-101, determine the element nodal force vector as a
result of a point load with the following components ./^, = —3150 and f
p ~
-

1500 N'cm. The point load is located at r0 = 34 and r0 = 32 cm

7-144 For the element in Problem 7-102. determine the element nodal force vector as a
result of a point load with the following components.^, = —2575 and f ~ r
1260 Ibfhn The point load is located at r0 = 54 and Zq = 53 in.

7-145 For the clement Problem 7-102. determine the element nodal force vector as a
in
result of a point load with the following components.^ = —3500 and fr ~
800 Ibf'm The point load is located at r0 = 52 and :0 = 54 in

7-146 Using the sy mbolic notation from Example 7 5. give the expression for the assem-

blage stiffness matrix for the discretized axisymmetnc region in Fig. P7-69.

7-147 Using the symbolic notation from Example 7 6. give the expression for the assem-
blage nodal force vector for the discretized axisymmeiric region in Fig P7 69

7-148 Using the s> mbolic notation from Example 7 5. give the expression for the assem-
blage stiffness matrix for the discretized axisvmmetnc region in Fig P7 71

7-149 Using the symbolic notation from Example 7-6. give the expression for the assem-
blage nodal force vector for the discretized axisvmmetnc region in Fig P7-7I

7-150 Using the symbolic notation from Example 7-5. give the expression for the assem-
blage stiffness matnx for the discretized axisvmmetnc region in Fig P7-73

7-151 Using the symbolic notation from Example 7 6. give the expression for the assem-
blage nodal force vector for the discretized axisymmetnc region in Fig P7 73

7-152 By extending the sy mbolic notation from Example 7 5. give the expression for the
assemblage stiffness matnx for the discretized axisymmetnc region in Fig P7 75

7-153 By extending the svmbolic notation from Example 7-6. give the expression for the
assemblage nodal force vector for the discretized axisvmmetnc region in Fig
P7-75

7-154 Bv extending the symbolic notation from Example 7-5. give the expression for the
assemblage stiffness matnx for the discretized axisvmmetnc region in Fig P7 77

7-155 By extending the sv mbolic notation from Example 7-6, give the expression for the
assemblage nodal force vector for the discretized axisvmmetnc region in Fig
P7-77

7-156 By extending the svmbolic notation from Example 7-5, give the expression for the
assemblage stiffness matnx for the discretized axisymmetnc region in
Fig

P7-79 Note the use of two different types of elements

7-157 for the

By extending the svmbolic notation from Example 7-6. give the expression
assemblage nodal force vector for the discretized axisymmetnc region in Fig
P7-79. Note the use of two different types of elements

7-158 For the element in Problem 7-94 for a particular loading condition, the following
values of nodal displacements are obtained u, = 0 00235. i, = 00.b,=
—0 00555.
\j = 0 00260, = 0 00150, and -0 00615 in. Determine the element strains
 PROBLEMS 361

and stresses that correspond to these displacements. Assume that both the element
self-strains and prestresses arc zero.

7-159 For the element in Problem 7-95 for a particular loading condition, the following

values of nodal displacements arc obtained: it,= 0.01215, v, = -0 01300, =

0.0. Vj — 0.00560. uk = —0.00555, and vt — —0.00750 cm Determine the
dement strains and stresses that correspond to these displacements. Assume that

both the self-strains and prestresses are zero.

7-160 For the clement in Problem 7-96 for a particular loading condition, the following

values of nodal displacements are obtained: — -0.01250, v, = -0.02350, u,

u,

= 0.03500. Vj — 0.01550, uk = —0.01775, and vt = 0.01010 cm. Determine

the clement strains and stresses that correspond to these displacements. Assume that

both the self-strains and prestresscs arc zero.

7-161 For the element in Problem 7-97 for a particular loading condition, the following

values of nodal displacements arc obtained: n = —0.02150, v, = -0.02350, h,

= 0.01875, Vj = 0.02550, u k - 0.02500, and vk = 0.0 in. Determine the element

strains and stresses that correspond to these displacements. Assume that both the
self-strains and prestresscs arc zero.

7-162 For the element in Problem 7-10! for a particular loading condition, the following

values of nodal displacements are obtained: it, = -0.02750. v, = 0.01250. «y ~

0.04555, r;, = 0.03350, ii
L = 0.02555, v, = -0.02350, = -0 01235, and
rm = 0.02500 cm. Determine the clement strains and stresses at the element centroid
that correspond to these displacements. Assume that both the self-strains and pre-
stresses arc zero.

7-163 For the element in Problem 7-102 for a particular loading condition, the following

values of nodal displacements are obtained: = -0 02250, v, = -0.02050, u,

«,
= 0.01500, v, = -0.03150, uk = 0.02530, vk = -0.03350, = 0.0. and vm

= 0.03500 in. Determine the element strains and stresses at the element centroid
that correspond to these displacements. Assume that both the self-strains
and pre-

stresses are zero.

7-164 made of bronze. The

The body from which a tetrahedral element is extracted is

nodal coordinates of the element are x, — - -5, z, = 0.0, x, = 1 7, y, -

2.0, y, 1

= = = = and zm =
3.0, z, -0.2, xk 1.5, yk = 2.0, zk = 1.7. \ m 0.6, v„, 1 8,

0.1 in. Determine the element stiffness matrix.

7-165 made of hard draun

The body from which a tetrahedral element is extracted is

- = 6, z,
- 0, - .
copper. The nodal coordinates of the element are 5, y,

= = = 4, and r„, = 0 cm.

3
.
4, z, 0, xk = 5, Vj = 5. zk = 4. ,r„,
= 6. ym
Determine the element stiffness matrix. f

stress analysis as
given
7-166 Show that the element stiffness matrix for three-dimensional
by Eq. (7-73) for the four-node tetrahedral element is always symmetric.

stress anal) sis for

the
7-167 Show element stiffness matrix for three-dimensional
that the
eight-node brick element is always symmetric.

of three-dimensional
7-168 What is the size of the element stiffness matrix for the ease
stress analysis if the eight-node brick element is used?
362 STRESS ANALYSIS

7-169 Determine the element stiffness matrix for the eight-node brick element defined
below. The element is extracted from a hot rolled, high carbon steel structure. The
coordinates of the nodes are = 34, 3 ,
= 32, rj = 20, x2 = 34, \2 = 33, z>
= a '=
20, x, 32, > 3 = 33. e3 = 20. r4 32, v4 = 32. z4 = 20. .r5 = 34, y s =
32, z5 = 18, x6 = 34. 3 # = 33, z6 = 18, r7 = 32, 37 = 33, z7 = 18, rg = 32.

3g = 32, and rg = 18m Perform the integrations bj evaluating the integrands at

the element centroid and treating the integrands as constants

7-170 Determine the element stiffness matrix for the eight node brick element defined
below The element is extracted from an aluminum (6061 alloy) structure. The
coordinates of the nodes are t| = 45. i| = 32, z t — 12, x> — 45, \j = 34, z2
= 12. r, => 42. 33 = 34, c, = 12. r4 = 42. \4 = 32. r4 = 12, r, = 45. 35 -
32, 2S = 9. r6 = 45. i6 * 34, r6 =
9, r7 = 42. i 7 34, r7 =
9. r8 - 42.3* =
= 32, and ;g = 9 cm Perform the integrations by evaluating the integrands at the
element centroid and treating the integrands as constants

7-171 What size is the element nodal force sector from a self-strain if the eight node brick
element is used in three dimensional stress anal) sis'*

7-172 Determine a general expression for the element nodal force sector from a uniform
self strain for the brick element in three dimensional stress formulations Perform
the integrations by evaluating the integrands at the element centroid and treating
the integrands as though they were constants

7-173 For the element in Problem 7-164. determine the element nodal force sector as a
result of a self strain if the temperature increases by 50°F

7-174 For the element in Problem 7-165 determine the element nodal force sector as a
result of a self-strain if the temperature decreases by 72°C

7-175 For the element in Problem 7 169. determine the element nodal force sector as a
result of a self strain if the temperature decreases by 48°F Perform the integrations
by evaluating the integrands at the element centroid and treating the integrands as
though they were constants

7-176 For the element in Problem 7-170. determine Ihe element nodal force sector as
a

resultof a self-strain if the temperature increases b) 37°C Perform the integrations

by evaluating the integrands at the element centroid and treating the integrands as
though they v,ere constants

7-177 Determine a general expression for the element nodal force sector from a uniform
prestress for the brick element in three-dimensional stress an 3 l>sis Perform the
integrations by evaluating the integrand at the element centroid and treating the
integrands as though they w ere constants.

7-178 Determine a general expression for the element nodal force sector from a uniform
the
body force for the brick element in three-dimensional stress problems Perform
treating the
integrations by evaluating the integrand at the element centroid and
integrands as though they were constants

as a
7-179 For the element in Problem 7-164. determine the element nodal force sector
result of a body force with the following components b, = 100, b,
= 125. an -

b, = 75 lbf/in .
3
 PROBLEMS 363

7-180 For the element in Problem 7-165, determine the element nodal force vector as
a
result of a body force with the following components- b t ~ 70, b, = 100, and
b.
= 40 N/cm 3 .

7-181 For the element in Problem 7-169, determine the element nodal force vector from
a body force with components b r = 30, b
the following ,
= 50, and b. = 20
3
lbf/in. . Perform the integrations by evaluating the integrands at the element centroid
and treating the integrands as though they were constants.

7-182 Assume that the element in Problem 7-170 is oriented such that the z axis is in the
direction opposite that of gravity, if no other body forces exist, determine the
corresponding element nodal force vector. The acceleration due to gravity is 9.81
m/s 2 Perform the integrations by evaluating the integrands
. at the element centroid
and treating the integrands as though they were constants.

7-183 Derive a procedure that could be used to evaluate the area of a typical face of the
tetrahedral element. This area is need for use in Eq. (7-78).

7-184 Determine a general expression for the element nodal force vector from a uniform
surface tractions on face 1 -2-3-4 for the brick element in a state of three-dimensional
stress. Perform the integrations by evaluating the integrand at the center of the face
and treating the integrands as though they were constants.

7-185 Determine a general expression for the element nodal force vector from a uniform
surface traction on face 5-6-2- 1 for the brick element in a state of three-dimensional
stress. Perform the integrations by evaluating the integrand at the center of the face

and treating the integrands as though they were constants.

7-186 For the element in Problem 7-164, determine the element nodal force vector as a
result of a surface traction on face ijk with the following components: s x = — 1000,

x, = 500, and s. = 750 psi.

7-187 For the element Problem 7-165, determine the element nodal force vector as a
in

result of a surface traction on face ikm with the following components: s x —

— 1500,
v, = —800, and s._
— 300 N/cm 2 ,

7-188 For the element in Problem 7-169, determine the element nodal force vector as a
result of a surface traction on face 5-6-7-8 with the following components: s r =
1600, sx = -925, and ,v.
= -400 psi. Perform the integrations by evaluating the

integrands at the centroid of the face and treating the integrands as though they were
constants.

7-189 For the element in Problem 7-170, determine the element nodal force vector as a
on face 8-7-3-4 with the following components: s,
=
result of a surface traction
2 by
0, and ,r. = 2500 N/cm
Perform integrations evaluating the
-3200, j = the .

centroid of the face and treating the integrands as though they

were
integrands at the
constants.

7-190 in the derivation leading to

£cj. (7-78) in the text that the
It has been assumed
surface traction is uniform over the face of the tetradedra! clement in question.
vector for a linearly varying
Derive an alternate expression for the nodal force
surface traction on face ijk of a four-node tetrahedral clement Proceed by assuming
the following forms forv,, J,. and s.:
 , ,

364 STRESS ANALYSIS

sx = N,s„ + NjSX] + N s,
t t

5, = N,s„ + NjSv + Nks:k

s. = N,i + NjS_j + Nt s *
where sx, is the x component of the surface traction at node etc. Write the shape

functions in terms of the area coordinates (defined on face yi) and use Eq (6-49)
to perform the integrations

7-191 Repeat Problem 7-190 for a linearly varying surface traction on leg /7m

7-192 For the element m Problem 7-165, determine the element nodal force vector as a
result of a point load that has the following components f , = 600, fpi = —750,
p
andj^., = 500 N The point load is located at Jo = 5, ) 0 = 5, and z0 = 4 cm

7-193 For the element in Problem 7-169, determine the element nodal force vector as a
result of a point load with the following components fpx = - 1 500, fps - 500,
and fp, = 880 lbf The point load is located at = 33 5, > 0 = 32 5, and z0 =
20 0 in

7-194 For the element in Problem 7-164 for a particular loading condition, the following

values of nodal displacements are obtained u, = -0 01250, v, = -0 02350, w, =

0 02220, u, = 0 03500, v, = 0 01550, w, = -0 01245, w* - -0 01775, v* =
0 01010, iv* = 0 0. u„ = -0 01 135. v„, = 0 00990, and wm = 0 0 in Determine
the element strains and stresses that correspond to these displacements Assume that
both the self-strains and prestresscs are zero

7-195 For the element in Problem 7 165 for a particular loading condition, the following
values of nodal displacements are obtained u, = 0 02250, v, = 0 02050, iv,
=
-0 05520, Uj = -0 03575, v, = 0 01750, w, = 0 03245, ut = 0 01775, v* =
-0 01750, w* = 0 07850, um = 0 02435, vm = -0 01990. and = 0 05450
cm Determine the element strains and stresses that correspond to these displace-
ments Assume that both the self-strains and prestresses are zero

7-196 With the help of Fig 7 17. verify Eq (7-82) by doing a force balance on the
infinitesimal beam element in the y direction
7-197 With the help of Fig 7-17, verify Eq (7 83) by doing a moment balance about
some convenient point on the infinitesimal beam element

7-198 Denve the variational principle that corresponds to Eq (7-88) What is the highest
order derivative present in the functional? Please explain Hint Refer to Chapter 4
(and use integration by parts twice).

7-199 beam element satisfy

Verify that the shape functions given by Eq (7-100) for the
the following conditions.
a. At x = x (where £ = 0)

W,, = 1

Afl, = N„ = N9l
= 0
b. At x = x} (where fj = 1).

Nv = 1

N„ = N = N9j = 0 8l
 PROBLEMS 365

7-200 Verify that the shape functions given by Eq. (7-100) for the
beam element satisfy
the conditions given by Eqs. (7-103) and (7-104).

7-201 Derive the finite element characteristics for the beam model
[i.e., Eqs. (7-1 16) to
(7- 1 1 8)] from the variational principle from Problem 7-198. Are the resulting integral
expressions the same as those derived with the Galerkin method? Please
explain.
It is not necessary to evaluate the integrals.

7-202 Consider the beam element shown in Fig. P7-202. The beam from which the element
is extracted is made of hot rolled, low carbon steel. The beam is rectangular in
cross-section with a width w of 4 cm and a height h of 8 cm. The coordinates of
nodes i and j arc x, — 5 and ,t = 6 cm. Determine the element stiffness matrix.
;

K-H

Figure P7-202
7-203 Consider the beam element shown in Fig. P7-202. The beam from which the element
is extracted is made of aluminum (6061 alloy). The beam is rectangular in cross-
section with a width w of 3 in. and a height h of 4 in. The coordinates of nodes i

and j are .v, = 32 and a, = 36 in. Determine the element stiffness matrix.

7-204 Consider the beam element shown in Fig. P7-204. The I-beam from which the
element is extracted is made of hot rolled, high carbon steel.
The cross-section of
the beam is also shown in Fig. P7-204 with the dimensions tv = 4 in., h = 8 in.,
and t = 1 in. The coordinates of nodes / and j are .v, = 42 and .t, = 48 in. Determine
the element stiffness matrix.

Figure P7-204

7-205 Consider the beam element shown in Fig. P7-204. The I-beam from which the
element is extracted is made of cast iron. The cross-section of the beam is shown
in Fig. P7-204 with the dimensions tv = 6 cm. It = 10 cm, and t = 2 cm. The
coordinates of nodes i and j are .v, = 20 and .r, = 24 cm. Determine the element

stiffness matrix.

7-206 A solid circular bar made from brass is to be used as a beam in a certain application.
The diameter of the bar is 3 cm. The coordinates of nodes / and j are x, = 32 and
Xj = 36 in. Determine the element stiffness matrix.

7-207 A solid circular bar made from hot rolled, low carbon steel is to be used as a beam
in a certain application. The diameter of the bar is 1 in. The coordinates of nodes
i and j are x, = 16 and xt = 18 in. Determine the element stiffness matrix.
366 STRESS ANALYSIS

Consider the beam element from Problem 7-202 Determine the nodal force vector
for the element if a downward force of 1000 N and a counterclockwise moment of
800 N-cm act at node i

Consider the beam element from Problem 7-203 Determine the nodal force vector
for the element if a downward force of 1200 lbf and a counterclockwise moment
of 800 lbf m act at node i

Consider the beam element from Problem 7-204 Determine the nodal force vector
for the element if an upward force of 1500 lbf and a clockwise moment of 1800
lbf in act at node i

Consider the beam element from Problem 7-205 Determine the nodal force vector
for the dement if an upward force of 500 N and a clockwise moment of 750 N-cm
act at node j

Consider the beam element from Problem 7-206 If an upward force of 350 N and
a counterclockwise moment of 400 N cm act at node i, and if a downward force
of 425 N acts at node j, determine the nodal force vector for the element

Consider the beam element from Problem 7 207 If an upward force of 425 lbf and
a clockwise moment of 675 lbf in act at node /, and if a counterclockwise moment
of 425 lbf in acts at node j, determine ihe nodal force vector for the element

For the beam clement from Problem 7 202, determine the nodal force vector for
the element if a uniformly distributed load of 1500 N/cm acts in the downward
direction

For the beam element from Problem 7 203, determine the nodal force vector for
the element if a uniformly distributed load of 1200 lbf/in acts in the downward
direction

For the beam element from Problem 7-204, determine the nodal force vector for
the element if a uniformly distributed load of 1650 lbf/in acts in the upward
direction

For the beam element from Problem 7 205, determine the nodal force vector for
the element if a uniformly distributed load of 850 N/cm acts in the upward direction

For the beam element from Problem 7-206, determine the nodal force vector for
the element if a uniformly distributed load of 785 N/cm acts in the downward
direction

For the beam element from Problem 7-207, determine the nodal force vector for
the element if a uniformly distributed load of 685 lbf/in acts in the upward direction

Using as much of Example 7-7 and

is resolve for the nodal deflections
as possible,
that the right end
slopes if the spring is removed Assume only two elements and
of the beam is simply supported

possible,
Assuming only two elements and using as much of Example 7-7 as is

resolve for the nodal deflections and slopes the left end of the beam is simply
if

supported

deflections and
Using as much of Example 7-7 as is possible, resolve for the nodal
removed)
slopes if the both ends of the beam are simply supported (the spring is
Assume only two elements
 PROBLEMS 367

7-223 Assuming only two elements and using as much of Example 7-7 as is possible,
resolve for the nodal deflections and slopes if both ends of the beam are cantilevered
(the spring is removed).

7-224 Reconsider the beam in Example 7-7. The distributed load p is now 3000 lbf/ft and
the point load P is 1000 lbf. The spring is removed (so Ay should be taken as zero).

The beam is rectangular in cross-section with a width b of 1 .5 in. and a height h

of 3 in. The beam is made of hot rolled, high carbon steel and is 3 ft long. Using
only two equal-length elements, determine the deflections and slopes at the ends
and midspan of the beam. In addition, determine the moment, shear force, maximum
bending stress, and maximum shear stress at the middle of the distributed load.

7-225 Modify the TRUSS program in Appendix B so that it may be used to solve for the
nodal deflections and slopes of a beam. Allow for up to 30 elements with up to 5
different materials. The input to the program should be modeled after the input to
the TRUSS program. For each material, read in the following parameters: the elastic
modulus £, the moment of inertia /, the distributed loading p, and the distance cmax
from the neutral plane to the outermost fibers (needed in the stress calculations).
Treat the prescribed deflections and slopes with positive boundary condition flags,
and the imposed forces and moments with negative flags (see the TRUSS program).
At the center of each element, the following results should be given in the output:
the bending moment, the shear force, the maximum bending stress, and the maximum
shear stress. These results should be calculated in a postprocessor subroutine.

7-226 Using the computer program from Problem 7-225 or one furnished by the instructor,
solve for the nodal deflections and slopes for the beam from Problem 7-220. In
addition, determine the element resultants at the midpoint of each element. Use 2,

4, and 8 elements.

7-227 Using the computer program from Problem 7-225 or one furnished by the instructor,
solve for the nodal deflections and slopes for the beam from Problem 7-222. In
addition, determine the element resultants at the midpoint of each element. Use 2,

4, and 8 elements.

7-228 Using the computer program from Problem 7-225 or one furnished by the instructor,
beam from Problem 7-223. In
solve for the nodal deflections and slopes for the
addition, determine the element resultants at the midpoint of each element. Use 2,

4, and 8 elements.

7-229 Using the computer program from Problem 7-225 or one furnished by the instructor,
solve for the nodal deflections and slopes for the beam in Problem 7-224. In addition,
determine the element resultants at the midpoint of each element. Use 2, 4, and 8
elements.

7-230 Explain in your own words what is meant by substructuring and why is it so vital

in large finite element models. Do not hesitate to use some equations in your
explanation.

7-231 Using the stress analysis program furnished by the instructor, solve for the two-

dimensional stress distribution in the thin plate shown in Fig. P7-231. Note that the
plate has a circular hole in it, and therefore a stress concentration exists in the
for this condition.
vicinity of the hole. Verify that the stress concentration factor is 3
Assume the following: D = 0.5 in., w — 6 in., h — 8 in., and r = 1000 psi.
368 STRESS ANALYSIS

Figure P7-231

The plate is made of hot rolled, high carbon steel and has a thickness of 0.375 in
Note that only one-fourth of the plate needs to be modeled because of the two-axis
symmetry Use at least 100 elements

7-232 Using the stress analysis program furnished by the instructor, solve for the two-
dimensional stress distribution in the thin plate shown in Fig P7-23 1 . Note that the
plate has a circular hole in it, and therefore a stress concentration exists in the
vicinity of the hole. Venfy that the stress concentration factor is 3 for this condition.
Assume the following 0 = 2 cm, w = 12 cm, h = 16 cm, and s = 1500 N/
cm 2 The plate is made of hot rolled, high carbon steel and has a thickness of 0 75
cm Note that only one-fouith of the plate needs to be modeled because of the two-
axis symmetry Use at least 100 elements

Figure P7-233
 PROBLEMS 369

7-233 Using the stress analysis program furnished by the instructor, solve for the two-
dimensional stress distribution in the thin plate shown in Fig. P7-233. Note that the
plate has an elliptical hole in it, and therefore a stress concentration exists in the
vicinity of the hole. Determine the value of the stress concentration factor for this
condition. Assume the following: a = 2 cm, b = 1 cm, u- = 10 cm, h - 20 cm,
and j = 1800 N/cm 2 The plate is made of brass and has a thickness of 0.5 cm.
.

Note only one-fourth of the plate needs to be modeled because of the two-axis
that

symmetry. Use at least 100 elements.

7-234 Using the stress analysis program furnished by the instructor, solve for the two-
dimensional stress distribution in the thin plate shown in Fig. P7-233. Note that the
plate has an elliptical hole in it. and therefore a stress concentration exists in the
vicinity of the hole. Determine the value of the stress concentration factor for this
condition. Assume the following: a — 1 in., b — 0.5 in., w = 5 in., h = 10 in.,
and x = 1400 psi. The plate is made of brass and has a thickness of 0.25 in. Note
that only one-fourth of the plate needs to be modeled because of the two-axis
symmetry. Use at least 100 elements.

7-235 Using the stress analysis program furnished by the instructor, solve for the two-
dimensional stress distribution in the thin plate shown in Fig. P7-235. Note that the
plate has a square hole in it, and therefore a stress concentration exists in the vicinity

of the hole. Determine the value of the stress concentration factor for this condition.

Assume the following: d = 0.5 in., »• = 10 in., h = 10 in., and s = 2000 psi.

The plate is made of cast iron and has a thickness of 0.5 in. Note that only one-
fourth of the plate needs to be modeled because of the two-axis symmetry. Use at

least 100 elements.

Figure P7-235

7-236 Using the stress analysis program furnished by the instructor, solve for the two-

dimensional stress distribution in the thin plate shown in Fig. P7-235. Note that the
plate has a square hole in it, and therefore a stress concentration exists in the vicinity

of the hole. Determine the value of the stress concentration factor for this condition.

Assume the following: d — 1 cm, w = 20 cm, h = 20 cm, and s = 2500 N/cm

2
.

The plate is made of cast iron and has a thickness of 0.5 cm. Note that only one-

fourth of the plate needs to be modeled because of the two-axis symmetry. Use at

least 100 elements.

 Steady-State Thermal and Fluid
Flow Analysis

8-1 INTRODUCTION

In this chapter several nonstructural applications are considered. Heat conduction

is given fairly extensive treatment. One-, two-, and three-dimensional problems are
formulated; these include convection, thermal radiation to a large enclosure, pre-
scribed heat fluxes, insulation, and prescribed temperatures. Axisymmetric prob-
lems are also formulated. In all cases, the thermal conductivity may be dependent
on temperature. Only isotropic materials, however, are considered in the formal
development. Several problems involving fluid flow are also formulated. The first

is a problem involving convective energy transport. The second is that of two-

dimensional potential flow. This is followed by a general formulation to steady,
two-dimensional, incompressible viscous fluid flow. The chapter is concluded with
a description of a steady, two-dimensional heat transfer program.
Before these formulations are developed, however, it is necessary to introduce
the reader to one of the simpler nonlinear solution methods. This is necessitated
by the fact that if any of the thermal properties is temperature-dependent, the
resulting system of equations for the nodal temperatures is nonlinear. It will also
be seen that the system of equations in some fluid flow problems is nonlinear as
well.

8-2 THE DIRECT ITERATION METHOD

From both implementation and mathematical points of view, the direct iteration
method is the simplest of the nonlinear solution methods. However, this method
371
372 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

has two main drawbacks The fust is that it does not always converge, and the
second is that the system matrix equation must be in the special form

[K(a)]a = f ( 8-11

as opposed to the more general form

g(a) = f (8-2)

Nonetheless, this method is worth studying because of the simplicity it affords.

To solve Eq (8-1) for the nodal unknowns a, we start the solution process by
guessing appropriate values for a These values are then used to compute K. The
system Ka = f is then solved for the vector a If the new a, denoted by a, +J ,

agrees with the old a, denoted by a,, to within an acceptable tolerance, then the
solution process is stopped, otherwise, a new K matrix isformed using the newly
calculated values m the vector a and the process is repeated This method is
summarized in Fig 8-1 Let us now illustrate this method with two examples.

Figure 8-1 The direct iteration method.

 THE DIRECT ITERATION METHOD 373

Example 8-1

Use the direct iteration method to solve the following nonlinear equation:

a
2
+ 6a - 5 = 0

Solution

The equation must be cast into the form given by Eq. (8-1), or

(a + 6)a = 5

In terms of the indices / and / + I , we have

(a, + 6)a, + 1
= 5

For an initial guess of a0 = 10, the iterations are as follows:

A, A, 4 |

0 10. 0.3125
I 0.3125 0.7921
2 0.7921 0.7362
3 0.7362 0.7423
4 0.7423 0.7416
5 0.7416 0.7417
6 0.7417 0.7417

It is seen that convergence to four-place accuracy is attained in six iterations. The

reader should take note that the second root is - 6.7416 (obtained from the quadratic

formula). In nonlinear problems, the root that is taken to be the solution is generally
based on physical reasoning (e.g., perhaps a cannot be negative).

Example 8-2

Use the direct iteration method to obtain a set of roots to the following system of
nonlinear equations:

a2 + 6a + Ay 2 = 23

6a + xy = 10

Solution

These equations must be cast into the form of Eq. (8-1), or

A + 6 xy X 23
6 X y_ 10
3?4 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

Using the indices * and i + 1 , we have

x, + 6
- [s]
6
?fc:]
which may be solved by the matnx-inversion method once an initial guess for x
and y is made. Alternatively, such a system could be solved with the active zone
equation solver from Sec. 6-8. For initial guesses of x0 = 2 and y0 — 2, the
iterations are summarized as follows:

1 r, *,+ ! >'i + i

0 2 000 2000 - 7500 7 250

1 - 7500 7.250 1 294 -2.980
2 1 294 -2 980 2 097 -1.997
3 2 097 -1 997 2 140 -1.354
4 2 140 -1 354 2 247 -1.626
5 2 247 -1 626 2 181 -1 373
6 2 181 -1 373 2.237 -1 569
7 2 237 -1 569 2 192 -1 409
8 2 192 -1 409 2 228 -1 537
9 2 228 -1 537 2 199 -1.433
10 2 199 -1 433 2 222 -1 517
11 2 222 -1 517 2 203 -1 449
12 2 203 -1 449 2 219 -1503
13 2 219 -1 503 2 206 -1 459
14 2 206 -1.459 2 216 -1 495
15 2 216 -1 495 2 208 -1 466
16 2 208 -1 466 2 215 -I 489
17 2.215 -1 489 2 209 -1 470
18 2 209 -1 470 2 214 - 1 486

The iterations were stopped when two successive iterations were within 1% of each
other. H
emphasized that nonlinear equations in general have more than one solution.
It is

For example, try to verify that r = l and y = 4 is also a solution to the problem
posed in Example 8-2 Caution must be exercised when attempting to obtain the
solutions to nonlinear problems. For all problems formulated in this chapter, the
pro-
direct iteration method tends to give only the physically realizable solution,
viding the initial guesses are reasonable.

8-3 ONE-DIMENSIONAL HEAT CONDUCTION

Recall that a one-dimensional finite element formulation to a specific heat con-

con-
duction problem was presented in Sec. 4-10. In the problem presented there,
 ONE-DIMENSIONAL HEAT CONDUCTION 375

vection from the lateral surface of the body was included, but not thermal radiation.
In this section, the one-dimensional heat conduction problem is extended to allow
for radiation from the lateral surface of the body. Convection, radiation, and heat
fluxes are also included on the ends of the body, i.e., on the boundaries. One
application of such a problem is an extended surface (or fin) convecting to a fluid
and/or radiating to some other large body or to space. In the interest of completeness,
heat generation is also included. A schematic of the one-dimensional heat conduction
model is shown in Fig. 8-2(a).

The Governing Equation

The governing equation for the temperature T is given by

—
ax
f kA-~\
(Lx j
- hP(T - Ta ) - zvP(T4 - Tj) + QA = 0 (8-3)
\

where k is the thermal conductivity of the materials, h is the convective heat transfer
coefficient, e is the emissivity of the surface of the body, P is the perimeter, A is

the cross-sectional area. Ta is the ambient fluid temperature far removed from the

oj -x

(a!

!bl

typical element.
Figure 8-2 (a)Schematic of one-dimensional heat transfer problem and (b)
Note that the boundary heat fluxes may act at a distance (such as the sun).
376 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

body, Tr is the receiver temperature for the radiation, Q is the heat generation rate
per unit volume, and a is the Stefan-Boltzmann constant, given by

a = 5.670 x 10" 8 W/m 2 -K 4

or
“8
cr = 0.1712 x 10 Btu/hr-ft 2 -°R
4

Except for a, each of these parameters may be a function of x. In addition, the

thermal properties (k, h, and e) may be temperature-dependent. Note that an absolute
temperature scale must be used if radiation is present.

The FEM Formulation

Recall from Chapter 4 that two different approaches may be taken to obtain the
finite element formulations once the governing differential equation is known: the
variational and weighted-residual approaches. The former is not as convenient as
the latter since it requires the intermediate step of determining the corresponding
functional The Galerkin weighted residual method is particularly well-suited to
nonstructural problems and is used here.
Recall that if an approximation to the temperature T on an element basis is

substituted into Eq (8-3), the equation will not be satisfied exactly. In general, a
e
residual R results The weighted residual method is stated mathematically by

m ,
7 ** dV = 0
2 J^W (8-41

where M
elements are assumed and the matnx W
r is the transpose of the so-called
weighting function matnx. For the specific case of the Galerkin method, Eq
(8-4) becomes
m ,

2 J*
N 7*' dV = 0 (8-5}
t=\

where N represents the shape function matnx. Since the terms in Eq. (8-3) represent
rate of energy transfer per unit length, the form of Eq. (8-5) to be applied to Eq
(8-3) is

2 f N TR e dx « 0 ( 8 ‘ 6)
Jl'

However, let us agree to drop the summation sign because we are really seeking
the finite element characteristics for a typical element. The summation represents
the assemblage process and is no longer delineated since this step is routine.
In what follows, the lineal element from Sec 6-3 is used. As shown in Fig.
8-2(b), a typical element e connects nodes i and j with coordinates x and xj and t
 ONE-DIMENSIONAL HEAT CONDUCTION 377

temperatures T, and Tj, respectively. This implies a linear temperature distribution

in each element. Therefore, Eq. (8-6) may be written as

- hP(T - TJ - e <rP(T* - T 4 ) + QA dx = 0 (8-7)

If the first term is integrated by parts, we get

N T kA — f dN T
——kA—dx
dT
-
f
N rhPTdx +
f
N r/i/T dx
dx dx dx Ji‘ h'
fl

r T
- J^eaPT4 dx + f N eaPT? dx + J N QA dx = 0 (8-8)

The first term in Eq. (8-8) is related to the heat flux qx from conduction in the x
direction. By Fourier’s law of heat conduction, we have

dx = (8-9)

This heat flux represents the heat transfer rate per unit cross-sectional area in the

x direction. The minus sign is used to give a positive heat flux in the direction of
decreasing temperature. The first term in Eq. (8-8) will contribute to the assemblage
equations for only the two elements at each end of the body; all internal contributions

to the assemblage equations cancel each other during the assemblage step. Therefore,
this term needs to be evaluated for elements 1 and only, assuming the elements M
are numbered consecutively from one end of the body to the other.

Let us assume that the heat transfer to or from the ends is given by any
combination of the following three conditions: (1) convection with heat transfer
coefficients h, and h to a fluid at ambient temperature
}
Tai and Taj , (2) radiation

with surface emissivities e, and e ; and receiver temperatures Tn and T rj ,

and (3) heat
fluxes q, and q (possibly acting at a distance such as sun). The subscripts / and j
}
are used to denote the local node numbers i and j. Let us denote the heat flux from
convection and radiation as q cx and q r respectively. The imposed heat flux will
, ,

be denoted as q s These boundary conditions are shown in Fig. 8-3 for one end ol
.

the body. An energy balance on this end of the body gives

Figure 8-3 Typical boundary in the one-dimensional heat conduclion

problem showing

the heat fluxes considered in the finite element formulation.

378 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

+ qt ~ 9rv + <7r (8-1 Oa)

or

<lx = + q, “ <7, {8-1 Ob)

Note that is defined to be positive if the heat flux is imposed toward the surface.

From Newton’s law of cooling at node j, we have

qn = h/T ~ T OJ )
(8-11)

and from the Stefan-Boltzmann law,

qt = tjtriT
4 - Trj) (8-12)

Since qs = q} at node j, we have for the integrated term at x = x} ,

N?M = Nr<-M>u,
i|
= - q, + q,)\ x -„

= ( - WhjAjT + N r*,V„ - N rt AJ l
aT>

+ N r c,AjaT* + NM^)|,.» (S-131

A similar expression results for the integrated term at node i, namely,

-N rfc4? - (-N r /i,A,r + N T /i,A,r„

ax

- N^e.A.oT 4 + N T t.,A,uT* + (8-14)

Note the sign of the term on the left, with the minus sign included here, the integrated
term is now given by the sum of the nght-hand sides of Eqs (8-13) and (8-14)
At this point, the parameter function T must be related to the nodal temperatures
in the usual manner by

T = Na' (8- 15)

For convenience, let us write T4 as follows.

T* = T T = (Na')W
3 (S' 161

With the help of Eqs (8-13) to (8-16), we may write Eq. (8-8) in the form
(8 ' 17)
K'n' = r
stiffness
The composite element stiffness matrix K' is comprised of five element
or conductance matrices, or

+ K' s + K& 8 ' 181

K' = KJ + K' + Kf <

e
compnsed of six other element
and the composite element nodal force vector f is

nodal force vectors, or

18-19)
v = t' + f; + ig + f'„ + r;„ + r;„
 ONE-DIMENSIONAL HEAT CONDUCTION 379

Note the use of the B in the subscripts on those terms that arise from the boundary
conditions. The expressions for the element characteristics are given by
r
dN dN
1 ——kA—dx
f ,
k<; = (8-20a)
dx dx

T
Kf, = 1N hPN dx (8-20b)

Kf -1 N T
ecrP(Na l
') i
N dx (8-20c)

Kf, fl = N rM,N| l=l + ,

{8-20d|

Kfc = N r £,A a(Na 1

<-

)''Nl, =i „ + N‘r 8J A jl
CT(Na'') 3 N| t= t, (8-20e)

f/; = f N rhPT„ dx (8-21 a)

r
f; = | N eo7T;? dx (8-21 b)

l
Q = [n t QA dx (8-21c)

l
c\Li
= N’Wn.U, + N r/i//ty t=v |
(8-21 d)

f
l
C
rB
= N r M,orT 3 U v,
+ N (8-21 e)

fc = N T 7 ,A,| <
t = (l
+ N TqJ AJ \ x ^ Xi (8-21 f)

Although it has been stated at the outset that the two-node lineal element from
Sec. 6-3 is to be used here, the above expressions for the element characteristics
are quite general. They can be applied readily to all other one-dimensional elements,
some of which are presented in Chapter 9. Several of these expressions will now
be evaluated by way of examples.

Example 8-3

Evaluate the element stiffness matrix for the lineal element from Sec. 6-3.

Solution

Recognizing that we need the derivatives of the shape functions in the expression
for Kf given by Eq. (8-20a), we first write

dN, d ( x, - x \ = _1
_
dx dx \Xj — xj L

and

dN, _ d ( x - x, \
= +1
dx dx\Xj - xj L
380 STEADY-STATE THERMAL AND FLUID FLOW ANALYS IS

where L is the element length Substituting these results into Eq. (8-20a) gives

L
or

kA
K' = ( 8 - 22 )
L
where it has been assumed that k and A are constant. If this is not the case, then
suitable average values may be used [e.g., values atx = (x, + Xj)!2\. H

Example 8-4
Try to evaluate the element stiffness matrix K; Use the lineal element from Sec.
6-3

Solution

For mathematical convenience, let us use the serendipity form of the shape functions
given by Eqs (6-tl), or

N, = Vi(\ - r) {6-1 la)

and

Nj = '/2 + r) (6-1 1b)

(1

where

2(x - x) x - x (6-10)
Xj - x, L/2

and x = (x, + xs )f2 defines the centroid of the element With these definitions,

we attempt to evaluate K; as follows.

(
N r eaP(Na') N- 3
dr

+ l
mPL~ + r]
(8-23)
r.'T[:;;]t
8
1

3
['/*(! - r)T, + 'A{\ + r)7)] dr

This integral is most easily evaluated using Gauss-Legendre quadrature, discussed

in Chapter 9. For now, we shall leave K' in this form.
*
 ONI DIMINSION'AL IIKAT CONDUCTION 381

From Example 8-4, it should he noted that the matrix Kf contains the (unknown)
nodal temperatures (7’ and Tt ) and. therefore, the resulting assemblage system
equations will be nonlinear The direct iteration method from Sec. 8-2 may be used
to obtain the solution. However, if the major stiffness contribution is from the
matrix Kf, the direct iteration method is likely to be divergent. In this case, ov-
crrelaxation or undcrrelaxation may be needed to obtain convergence. A detailed
discussion of these refinements to the direct iteration method is beyond the intended
scope of this text.

Example 8-5
Evaluate the clement stiffness matrix Kf, ;/ if the lineal element from Sec. 6-3 is

used.

Solution

Recall that one of the properties of the shape functions is given by

)
= 1 A', (a,) = 0

A'/.v,) = 0 A',(t,) = I

Therefore, Kfn/( may be evaluated as

+ °
M,|l 0) hjAjlO 1]
o

or

lt,A, 0
0 hjA,

It is emphasized that the stiffness matrix given by Eq. (8-25) will contribute

to the assemblage stiffness matrix only if the element is on the boundary (that is,

atone of the ends). Naturally, convection must be present at this end also. For
example, if element 1 connects nodes and 2, then 1

Similarly for clement M that connects nodes n - 1 and n we have

All other K£, fl ’s may be taken as the 2 x 2 null matrix.

Example 8-6

Evaluate the element nodal force vector ff,. Use the lineal element from Sec. 6-3.
382 STEADY STATE THERMAL AND FLUID FLOW ANALYS IS

Solution

It is convenient to evaluate the integral in Eq (8-2 la) by using length coordinates

and the integration formula given by Eq. (6-48).

f' 7-.*
l
hPTa dx
J L,
/r

[iy hPTa dx

1 ’O'

(1 + 0 + 1)'

(8-26)
1
O'l .

.(0 + 1 + I )»

or

hPLT„
(8-27)
2

Not surprisingly, one-half of the total hPLTa is allocated to each node Note that

L (without subscripts) represents the element length and is not to be confused with
the length coordinates B
Example 8-7

Evaluate the element nodal force vector f'lfl Use the two-node lineal element

Solution

Referring back to Example 8-5 and using Eq (8-2 Id), it follows that

wil
Again it is emphasized that this particular element nodal force vector
contribute to the assemblage stiffness matrix only has a node on the
if the element
element connects
global boundary (assuming convection there). For example, if 1

nodes I and 2, then

 ONE-DIMENSIONAL HEAT CONDUCTION 383

f<n _ h\AiTa \

(8-29a)
[
0

Similarly for element M that connects nodes n - 1 and n we have

f(A/) _ 0
B (8-29b)
C '
AnT
h n ri
,, I
an

In effect, all other f^ B ’s may be taken to be the 2 x 1 null vector.

Example 8-8
Reevaluate the element stiffness matrix by assuming the cross-sectional area A
varies linearly from node i to node j over the lineal element.

Solution

It proves to be very convenient to write the linearly varying cross-sectional area A

as

A(x) = N,(.x)A, + Nj(x)Aj

where A, and A, are the cross-sectional areas at nodes i and j. Note that the shape
functions from Sec. 6-3 are linear themselves and, in effect, provide a convenient
interpolation polynomial. Therefore, we have

k(N,A, + j- dx

where L is the element length. In terms of the length coordinates L, and Lr this

becomes

K? =
£ JjiLA + W dx

Using the integration formula given by Eq. (6-48), this reduces to the following
simple result:

1
-
(8-30a)
-1 1

where A is defined by

_ A, + Aj
A = (8-30b)
2
384 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

The reader is cautioned about generalizing this result. For example, if Kfy is

evaluated m a similar fashion by assuming the perimeter varies Linearly over the

lineal element, the result is

K'“' =
K — 3P +
P,
‘

+ P}
P) P + Pj
P,
'

+ 3 Pj
18 31
-
)

12

This is quite different from the result obtained when the perimeter is evaluated at

the element centroid

Example 8-9
Reconsider Example 4-1 1 from Sec 4-10. Recall that the tip of the fin was assumed
to be insulated. Resolve for the temperature distribution for the case of convection
at the tip Assume that the tip is in contact with a boiling fluid at 10°C with a heat

transfer coefficient of 4000 W/m 2 -°C

Solution

The element stiffness matrix

° ° i
K& 0 hjAj]
= r° 5
(4000X1.2566 x 10“ )
[0
needs to be added to K t21
in Example 4- 1 1 ,
or

0 50893 -0 49951] 0 0
K <2 > =
-0 49951 0 0.05026
50893 0

0 50893 -0 49951
-0 49951 0 55919

The assemblage stiffness matrix K" becomes

0 50893 -0 49951 0
-0 49951 1 01786 -0 49951
0 -0 49951 0 55919

In addition, the element nodal force vector

=>

[v>/J [<(4000)(1 2566 X 10- 5 )(10)

needs to be added to f t2) in Example 4-11, or

fB ,
ro.2356l] r 0 = fo
1 _ To 23561
23561]
"J

[0 23561 J [0 50264 J [o 73825 J

The assemblage nodal force vector f" becomes
 THE GREEN-GAUSS THEOREM 385

'0.2356 f
f
a = 0.47122
_0.73825_

After application of the prescribed temperature of 85°C at node 1 (using Method 1

from Sec. 3-2), we get

~io
0 1.01786 - 0.49951
o Ifr,"]
T2 =
[
85
42.930
_0 - 0.49951 0.559i9j[_r3 J L°-73825_

Solving for the nodal temperatures gives

T, = 85°C T2 = 76.25°C T2 = 69.43°C

As expected, the tip temperature is much lower than that obtained in Example
4-11, where the tip was assumed to be insulated. Section 4-10 should be consulted
for a discussion of how the element resultants may be obtained.

8-4 THE GREEN-GAUSS THEOREM

The Green-Gauss theorem is essentially a multidimensional version of integration

by parts, the latter of which is given by Eq. (4-32). We found this equation to be
useful when we developed the finite element characteristics in one-dimensional
nonstructural problems. When dealing with two- and three-dimensional problems,
we will again need to use integration by parts. However, the Green-Gauss theorem
is far easier to apply and is mathematically more rigorous.
The Green-Gauss theorem may be derived from the divergence theorem. It

may be recalled that the divergence theorem states that the integral of the divergence
of a vector over a volume V is precisely equal to the integral of the flux of the
same vector through the closed surface S that bounds the volume V. Stated math-
ematically, the three-dimensional form of the divergence theorem is given by

JV q dV = J q-n dS (8-32)

where q is any vector, d\' is an infinitesimal volume element, dS is an infinitesimal

surface element bounding the volume V. and n is a unit vector that is always normal
to the closed surface S and always directed outward from the body, as shown in
Fig. 8-4{a). The divergence of q is written V-q and in cartesian coordinates is given
by

where q x and q.
. are the x. y, and r components of q. As Eq (8-33) implies,

the divergence of a vector is a scalar.

 I LADY-STATE THERMAL AND FLUID FLOW ANALYSIS

(W
8-4 Definitl0n of terrns used in the divergence theorem in (a) three dimensions
and (b) two dimensions.
..

. ^ uatl0n (8-32)
e divergence theorem if V
may also be used to represent the two-dimensional form of
is interpreted to be an
area A, S to be the path C
enc osing the area, and n to
be the normal unit vector pointing outward from the
ounding path in the plane of the
area A. In this case Eq. (8-32) is written

[v-q dA = £q-n dC (8-34)

The integration around the path

C must be performed in the counterclockwise
direction. Figure 8-4(b)
should help to clarify this notation.
e are now in a position to derive the Green-Gauss
theorem. Let us write the
vector q as the product of a scalar
p and another vector p. or

q = Pp (8-351

Substituting this expression for

q in Eq (8-32) yields

/,V-(Pp) itv = jfpp-n dS (8-361

 1 ' p
THE GREEN -G\ CSS THEOREM 387

The following identity [ I ] proves to be useful:

T-(pp) = pV-p Vp-p {8-37}

where Vp denotes the gradient of the scalar p. In cartesian coordinates. Tp is given

in three dimensions by

VP
«P. _ dp. ^ dP.
(8-38a)
ax dy ^ 3r

and in two dimensions by

6p . dP .

VP + J (8-38b)
i? *
Note that the gradient of a scalar is a vector. Returning to Eq. (8-36) and using
Eq. (8-37), we may write the following:

pV-p eft' = j^pp-n dS - vp-p dV (8-39)

[ [

for the three-dimensional case and

pV-p dA = f_pp-n dC - J vp-pdA (8-40)

[

for the two-dimensional case. Since px is independent of p % and p: and

. vice versa,
it follows that

Jf
P~<A' = fpP<ns dS - f r
led.
*P'dV {8-41 a)
dx

Ji
1
p—
dy
= fp -
t
n, dS -
- 1 ay
(8-41 b)

JV (8-4 1c)
[ dz
Jc fp-
In the two-dimensional case. px is independent of /y . and vice versa: so Eq. (8-40)
implies

= Bp x nx dC - (8-42a}
f P °-^dA Jc
j [
JA
*z~-px dA
-M ox dx

and

Ja
p^ dA = f
Jc
PPx n% dC - Ja
f
&p ; dA (8-42b)
ox ay

In the above. d\
r
is generally taken to be dx dy dz and dA to be dx dy; px . p,-
the direction
and p. are the .r. y. and r components of p: and nx . n... and n- are
coordinate axes. Figure
cosines of the outw ard normal unit vector w ith respect to the
388 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

8-5 shows that nx is the cosine of the angle a, between the vector n (the outward
normal unit vector) and the x axis. Similarly, n, is the cosine of the angle a2 .

Hence, n r and n, are the respective direction cosines.

Equations (8-42) and (8-41) are the two- and three-dimensional forms of the
Green-Gauss theorem. Specific examples that illustrate the use of Eqs. (8-41) and
(8-42) are not given here because several applications are forthcoming in the re-
mainder of this chapter.

8-5 SIMPLE TWO-DIMENSIONAL HEAT CONDUCTION

Consider a long bar of uniform cross section as shown in Fig 8-6(a). It is assumed
that each cross section through the bar is no different from any other both geo-
metrically and thermally. Note that only imposed heat fluxes and prescribed tem-
peratures are allowed. Perfect insulation is a special case of zero heat flux. Let us
allow for a heat source or internal energy generation Q per unit volume and unit
time that is most a function of .r and y only This three-dimensional body may
at
therefore be analyzed as though it were two-dimensional The governing equation

for the steady-state temperature T for a typical cross section far from the ends is
given by

f(*?) + T^t) + Q = 0 (8-43)

3t\ dx) dy\
dy /

(perpendicular to the
Figure 8-5 Two-dimensional region illustrating the unit normal n
boundary C) and the angles a, and a 2 Note that cos a,
.
= n, and cos a 2 " ny
 '>'«>!! SSj.ISM III \1 (OMU ( J|OS

t C *

Figure 8-6 S.l ;m>n. of simp'c two-dimension tl lie.--! conduction problem (a) infinitely
lone KkIj .>!,») (to if on pHe with inMifilaS I.ucr.i! faces

Note tli.it this s.imc equation describes the plate of const. int thickness shown in log.

fvhthj, providing the lateral f.iccs of the plate are well-insulated.

In the next several sections, the iieat flux in a direction normal to some suiface,
usual)} the global boundary. suit be needed A heat flux is generally a vector and
represents the heat transfer rate per unit area In isotropic materials, the heat llux
is always normal to isothermal surfaces A surlacc is isothermal il the temperature
is constant on that surface In any exent. the net heat flux through a suiface from
conduction must be taken in (he direction normal to the surface Clearly, no con-
tribution can he made by the tangential heat flux (if any exists) Let us denote the
net (normal) heat flux vector as q„. the magnitude ol winch is simply q„. In terms
of the outward normal unit vector n. q„ max he written

<I„
= </„" ( 8 - 44 )
390 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

In terms of the heat fluxes in the global coordinate directions, this same net (normal)
heat flux from conduction may be written

q* = ?,* + qyj (8-45)

where q x and q y are the heat fluxes in the x and y directions. This is illustrated in
Fig. 8-7. Invoking Fourier’s law of heat conduction for heterogeneous (but isotropic)
materials, we have

qx =
— a,bT
— and q. = —k
bT
— (8-46)
bx by

It should be recalled from elementary heat transfer that the minus signs are included
so that a positive heat flux vector always points in the direction of decreasing
temperature. From Eqs (8-44) to (8-46), we conclude that

bT bT
q„
= -k—i -
dx
k —
by
j (8-47)

If the dot product of both sides of this last result is taken with the vector n, we get

qn = -k~-n x ~ k^fn (8-48)

dx by

where n x = i*n and tt

y = j*n are the cosines of the angles formed by the x and y
axes with the vector n, that nx and ny are the direction cosines Equation (8-48)
is,

is very important. It what follows and may be extended

will be used frequently in
by inspection to the three-dimensional case. It is emphasized that q„ represents the
net (normal) heat flux from conduction leaving a surface
In the next section some of the concepts of variational calculus introduced in
Chapter 4 are extended to two dimensions In Sec. 8-6, these concepts are applied
to Eq. (8-43) in order to derive the expressions for the finite element characteristics

heat flux q„
Figure 8-7 Boundary of two-dimensional region showing net (normal)
the components in the x and y directions
 V ARI ATIONA1. FORMULATIONS IN TWO DIMENSIONS 39 1

This followed by the derivation of these expressions with the Galerkin method
is

in Sec. S-7. The latter will be seen to be much simpler and applicable to a wider
variety of problems. Therefore, after Sec. 8-6, the Galerkin method will be used
throughout the remainder of the book.

8-6 VARIATIONAL FORMULATIONS IN TWO DIMENSIONS

Recall from Chapter 4 that if the governing differential equation contains second-
order derivatives, the functional contains only first-order derivatives. For Eq.
(8-43) it seems reasonable to assume a functional F of the form

F = F{x.y.T,Tx .Tx ) (8-49)

where the subscripts on T indicate derivatives, i.e..

dT
Tx = — and T =
x
dT
— (8-50)
d.r oy

The integral to be extremized is then given by

/ = J^F(.\,y,T.Tr .Tx ) dx dv (8-51)

Using a procedure that is completely analogous to that used in Sec. 4-5, let us
derive the corresponding Euler-Lagrange equation. It should further be recalled that
the Euler-Lagrange equation is really the governing differential equation for the
problem.

The Euler-Lagrange Equation: Geometric and Natural Boundary

Conditions
392 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

and, similarly

—
dF
dT x
87;y
BF a

BTy dy
(871 (8-55fa)

Therefore, Eq. (8-52) becomes

8/ = -m dx dy = 0 (8-56)
1 [:ar dTx dx dTy dy

Normally, we would use integration by parts at this point. However, since the
problem is two-dimensional, we use the Green-Gauss theorem instead. For example,
let us examine the second term which is in effect an integral of the form

(y p,**
J *dx

where p x is analogous to dFldTx and p to 8 T From Eq. (8-42a), it follows that

3F d dF fa/ dF\
L wj, mdx “y - Lw:- &TdC ~
( [

U[wJ STdrd > 18571

In a similar fashion, the third term m Eq (8-56) may be written as

-
if * udc -
i|(f)
,r ** l8
- 58 ’

Therefore, Eq (8-56) becomes

7
J-»[ar dx\dTj dy\BT /] x

+ f I
J^L dT,
Urn, + — njs TdC =
J
0 (8-591

Continuing further with the analogies from Chapter 4, we conclude that the Euler-
Lagrange equation is given by

dF
(8-60)
3T

and the natural boundary condition by

BF_
( 8 - 61 )

ar/'
temperature
This last result will be shown to be related to the condition of a zero
is
gradient; hence no conduction on that part of the boundary can occur (i e., it
is zero on those
portions
perfectly insulated). Also, the second integral in Eq. (8-59)
 VARIATIONAL FORMULATIONS IN TWO DIMENSIONS 393

of the global boundary where the temperature is prescribed because if 7" is prescribed,
we have 8 T = 0. This last condition is the so-called geometric boundary condition.

Example 8-10
For the problem described by the governing equation given by Eq. (8-43):
(a) Determine the functional F if the method illustrated in Example 4-3 is used.

(b) Determine the variational principle if the method illustrated in Example 4-4 is
used, and if an imposed heat flux q sb is assumed to act on the boundary.

Solution

(a) We first write Eq. (8-43) in a form where a one-to-one correspondence of terms
can be made more readily with the Euler-Lagrange equation, or

(8-62}

from which it is concluded that

dF
= 0 (8-63a)
dT
dF dT -
k - kTx (8-63b)
dTx ~ dx

dF dT _
~ k - kTy >
(8-63c)
37\ by

If Eqs. (8-63) are integrated, we get

F = QT + f(Tx Jy ) (8-64a)

F = - Zi kTx2 + g(T,Ty ) (8-64b)

and
F = - 'AkTy + h{T,Tx ) (8-64c)

where /, g, and h must be such that F itself is given by

F = QT - VikTl - ZzfcTy (8-65a)

or

F = QT - Vzk (8-65b)

may be added to the expression for the functional F. However,

Actually a constant
would have no effect whatsoever on the extremization process to be
illus-
since it

trated below, it is not included in Eqs. (8-65).

(b) From Example 4-4, we begin by writing

ST
8/ + 8 T dx dy = 0 ( 8 - 66 )
dy V" by
 1

394 STEADY STATE THERMAL AND FLUID FLOW ANALV SIS

The two terms containing second derivatives may be rewritten with the help of the
Green-Gauss theorem [see Eqs. (8-42)] if we note that
dT
(3 = 8T p' = l Px =
T.

For example.

/.£(€) s™'*' - I ( 87'"c -

S"-
( 8 - 68 )

and

(8-69)

and

A“(S
at at
T) =
at \at/
= */:A8
\at/

A similar result is obtained for the other term The final result is given by

+ f (k^fn, + lj-n\iTitC = 0 (8-71)

Jc\ at d\ /

This implies that / itself is given by

I =
I [ er
- ,/!
*
(£)
- *** * - [«- r rfc |8 - 7
(£) ]

Integrations around the boundary C are always performed in a counterclockwise

direction as shown in Figure 8-8. If we assume that an imposed heat flux q# acts

on the boundary (transferring energy toward the two-dimensional region) as shown

in Fig. 8-9, then an energy balance at this point on the boundary gives

/ = J[|er ]
** + l c
<i* T <i c i 8 ’731
-

This result is used as a starting point in the next section in the finite element

formulation to the problem ®

Example 8-1

Using the functional from Example 8-10, determine the so-called natural boundary
condition for the problem described by the governing equation given by
Eq.

(8-43).
 VARIATIONAL FORMULA! IONS IN TWO DIMENSIONS

Direction of
integration
around the
boundary

Figure 8-8 The boundary integrals for a two-dimensional region are evaluated by inte-
grating in a counterclockwise direction as shown.

Solution

With the help of Eq. (S-65a), the natural boundary condition expressed in Eq.
(8-61)becomes

BT BT
—k —
Bx
n,
- k—n,> =
By
0 (8-74)

From Eq. (8-48) it is concluded that the part of the boundary C over which Eq.
(8-74) holds is insulated. The net heat flux q„ from conduction is zero in this
case. B
From Examples 8-10 and 8-1 1 and Eqs. (8-73) and (8-74) it is concluded that
the variational principle corresponding to Eq. (8-43) is given by

2 2 ^
f
BT\ ...(BfX
dx dy (8-75)
-I dyJ J

*,B

Figure 8-9 Two-dimensional region showing the heat fluxes from conduction and that

which is imposed on a typical part of the global boundary

396 STEADY -STATE THERM KL AVD FLLID FLO" AN ALYSIS

providing the boundary conditions are either natural or geometric on various parts
of the global boundary.

The Finite Element Formulation

The functional / gi\en by Eq. (8-73) holds over the entire area A . It is convenient
to perform the integrations on an element basis by noting that

(8-76)

"here if elements are assumed Instead of evtremrzing the original integral /, we

take the derivative of l r with respect to the nodal unknowns in the vector a'. Recall
that the parameter function for the temperature T may be written in terms of the
nodal temperatures by writing

T = NV (8-77)

where N is the sh3pe function matrix For the three-node triangular element, for
example. N is given by

X = [N, N, i\J IS-78)

"here the shape functions themselves are given bv Eqs. (6-21). In this case, a'
contains the three nodal temperatures or

(8-79)

Taking the derivative of /' with respect to a' and remembering that Eq (4-119)
must be used, v\ e hav e

dlr
-—=11
d&e
QS t
Jv'
-
Ar
—
.SX dN
L—
7

At
a' - t
k
fiX

ch
7
fiN

Av
a'
]
c

^ J

S r dC = 0
-
fc , a
This last result may be written in the standard form

K'a r = r (8-81)

by defining
Kr — K£, - Kf,
(8-82)

and
(8-83)
r= f6 +

where in turn
 THE GXLERKIN METHOD IN TWO DIMENSIONS 397

6N 0N r
= f
— k— dx dx (8-84a)
J x dx dx

r 3N r 3N
K' =
Jx dx
k—
dy
dx dx (8-84b)

5? II
f
J\<
N T0 dx dx (8-85a)

Fe __
fc
N Tg* dC (8-85b)

In Sec. 8-8, several of these integrals are evaluated for the triangular element.

8-7 THE GALERKIIM METHOD IN TWO DIMENSIONS

In direct contrast to the previous section, the Galerkin weighted-residual method
is applied to Eq. (8-43) in a very straightforward manner. We begin the formulation
on an element basis from the outset by forming the integral of the weighted residual
and setting that result to zero where the weighting functions are the shape functions,
or

( 8 - 86 )

Applying the Green-Gauss theorem to the first two terms by noting

dT
(3 = Nr P^ = k
dy

we get

3N r dT
f
Jc'
N Tk~,i dC x
L dx
k —
dx
dx dy
J
+ [r
Jc
JC'
N Tk~n, dC
dr
dx '

T dT
dN
-lw —k—dxdy
r
[ N Q
,
+ dx dy = 0
dx dx r

Combining the two boundary' integrals into one integral and using Eq. (8-48) gives

f»
Ja
^E
dx
k
dx
dxdv + f
J'V
~k Y dx dy =
dx
(

dy fv
J
N r Qdxdy - f
JC'
N Tq n dC

If a heat flux q sB is assumed to be imposed as shown in Fig. S-9. we have

q„ = -q.56 f 8 ' 87 >

from an energy balance at the point of application. Furthermore, using T - Nae

c
(and noting that a is independent of .v and y) gives
398 STEADY -STATE THERMAL AND FLUID FLOW AVAL1 ! SIS

K'a' = r (8-88)

where

Kr = K', + K' (8-89)

and

f' = r$ + fj. (8-90)

where in turn

K;, _ f
Jv At
-Cx"**
At
(8-84a)

KS - f
Jv 6v
M<Wv dv
(8-84b)

f$ - j N TQdxd\ (8-85a)

II
5 (8-85b)

These expressions are identical to those demed via the variational approach Ob-
viously, the Galerkin method requires significantly fewer steps. Several of these
integrals are evaluated in the next section for the triangular element for which the
shape function matrix N is given by

N = [N. N, A'*]

where the shape functions themselves are given by Eqs (6-2 1 ). The vector of nodal
-

unknowns a* is comprised of the three nodal temperatures in this case

Since the Galerkin method is simple, the variational approach is abandoned in

the remaining part of the chapter (and the book) The Galerkin method w ill be used
exclusively from now on

8-8 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION

In this section the simple heat conduction problem defined in Sec 8-5 is generalized
as shown in Fig. 8- 10 Note that the two-dtmensional region is shown as a relatively

thin plate of variable thickness As in thesimple model, a volumetric heat source

is also to be included Both lateral and boundary heat transfer effects are to be
modeled including convection, simple radiation, and imposed heat fluxes Simple
radiation is defined to be radiation to or from a large enclosure The imposed heat
fluxes may act at a distance such as the sun and. therefore, may be present in
addition to convection and radiation Prescribed temperatures on the global boundary
are also allowed.
The intent in this section is to deme the expressions for the element charac-

teristics in general, i e., without regard to the type of element. As mentioned in

 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION
399

Boundary (may act at a

radiation distance)

Figure 8-10 Schematic of general two-dimensional heat conduction problem

Sec. 8-7, the Galerkin method will be used. Then several of these expressions will
be evaluated by way of examples for the three-node triangular element. However,
before these expressions can be derived, the governing equation that describes this
general problem is needed.

The Governing Equation

The governing differential equation for the temperature T is derived quite easily if

an energy balance is done on an elemental volume t dx dy, where t is the thickness

of the plate. It can be shown that the governing equation is given by

f (kt^) + ~ (kt^) - her - Ta ) - eafr4 - 7?) + q, + Qt = 0 (8-91)

dx \ dx) dy \ dy)

where k is the thermal conductivity, It is the sum of the convective heat transfer
coefficients on the two lateral faces, e is the sum of the two surface emissivities
on the two lateral faces, qs is the sum of the heat fluxes imposed on the two lateral
faces, Q is the heat generation per unit time and volume, Ta is the ambient tem-
perature of the fluid in contact with the lateral faces, Tr is the temperature of the
enclosure receiving the radiation from the lateral surfaces, and ct is the Stefan-
Boltzmann constant (given in Sec. 8-3). Note that q s is positive if it is directed as
shown in Fig. 8-10.
As mentioned earlier, boundary convection, radiation, and heat fluxes are to

be included also. These boundary conditions are considered in the next section
where we derive the element characteristics. The reason for this is that it is far

easier to apply these conditions on an element basis. Prescribed temperatures will

also be handled quite routinely after the assemblage step.
 ,

400 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

It should be noted that if a particular heat transfer effect is negligible or not

present, the corresponding term in Eq. (8-91) is simply omitted. For example, let

us say that in a given problem we have negligible lateral radiation and no imposed
lateral heat flux. In this case, we may taken e and q s both to be zero in a genera!
computer program. It will be seen in the next section that this will eliminate the
corresponding contributions to the element stiffness matrices and/or nodal force
vectors.

The FEM Formulation

The expressions for the finite element characteristics may be derived without actually
specifying the type of element at this point However, later in this section the
emphasis will be on the three-node triangular element from Sec. 6-4. Recall that
in this case the shape function matrix N is given by

N = [N, Nj N k]
(8-92)

where the shape functions themselves are given by Eqs (6-21) A typical element
e has nodes i, j, and k, specified in a counterclockwise order with temperatures T„
Tj and Tk On an element basis, the Galerkin method requires
.

Ta
— /
kl —
dT\
+ —
fl
(,
kt —
dT\
- h(T - Ta )
I

Ldu\ SxJ dy s>7

- £0(7* - T*) + q s + dx dy = 0 (8-93)

2/J
It isemphasized that this integral applies to a typical element e and the integrations
e
are to be performed over the area A of the element Note that each term in Eq
(8-93) has units of energy per unit time If the Green-Gauss theorem is applied to
the two terms containing second-order derivatives, we get

Jc* * f
J*
*£*£* 4 ,
dx dx dx

+ - f
f
Jc' dy * J* fly fly

- £nrhTdxdy + j^N r W a dx dy

— N^oT4 dx dy + J^N T eaT4 dx dy

+ N Tq s dx dy + j^Qtdxdy = 0 (8-94)

However, the two integrals around the element boundary imy be combined with
the help of Eq. (8-48) to give
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION 401

where qn represents the heat flux from conduction in the direction of the outward
normal to the boundary as shown in Fig. 8-7.
At this point it is convenient to consider the nongeometric boundary conditions
(i.e., all except the prescribed temperatures). As mentioned earlier, convection,
radiation,and imposed heat fluxes are to be included. Perfect insulation is a special
case of zero heat flux. Let us perform an energy balance (on an area basis) on the
global boundary shown in Fig. 8-11 such that we may write

In + <7 sB
= + <?rfl (8-96)

where q^ B q rB and qsB represent the convective, radiation, and imposed heat
, ,

fluxes. Using this result to eliminate q n in Eq. (8-95) gives

fc
N r (~q„)t dC = fC'N r(q sS - q nB - q rB )t dC (8-97)

But from Newton’s law of cooling, we have

<7ob
= ^b(T ~~ TaB ) (8-98)

and by application of the Stefan-Boltzmann law, we have

q rB = e s o-(r
4 - T;b ) (8-99)

where hB is the convective heat transfer coefficient on the element boundary', e B is

the surface emissivity of the boundary', TaH is the ambient fluid temperature, and

Tr8 is the temperature of the enclosure receiving the radiation from the boundary.
Strictly speaking, the T in Eqs. (8-98) and (8-99) should be the temperature at the

element boundary', but the shape function matrix in Eq. (8-95) will automatically
take care of this as shown later.
With the help of Eqs. (8-98) and (8-99), we may write Eq. (8-97) as

fc
N r(-q n )tdC = \c N Tq sB tdC - \c WhB iTdC

conduction
Figure 8-11 Typical part of the global boundary in the two-dimensional heat
formulation.
problem showing the heat fluxes considered in the finite element
402 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

+ |ci N
r
;i,lT„jrfC - ^N T Ee atT 4
dC

+ j_N'r,„mT;a dC 18-100)

Recalling that

T = Na f (8-101)

where N is the shape function matrix and a e is the vector of the nodal temperatures
for element e. As in Sec. 8-3, it is convenient to write T4 as follows:

T* = T r = 3
(Na') 3 Na e (8-102)

It follows that Eq. (8-94) may be written as

K'a' = f
e
(8-103)

where

K' = K£x + K„
e
+ Kf v + K' + K %B + k;„ (8-104)

and

p = K + f; + K + fG + Kb + Kb + 1;,
(8-105)

The element stiffness matrices are in turn given by

f aN r 5N
Kae = —kt—dxdy
f
(8-10Ba)
Ja* dx dx

dN T dN
Ln %
f ,
k dxdy (8-106b)

K;„ = J^N rfcN dx dy (8-106c)

K • = J NT E<r(Na') N dx dy ! (8-1 06d)

r (8-1 06e)
K;vS = cr N ft„(N dC
|

K;„ = N r E„<w(Na') , N dC (8-1061)

£ i

and the element nodal force vectors by

07a)
x;, = j^mr.dxdy (8-1

(8-107b)
!; = N t£(tT? dx iy

(8-1 07c)
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION 4Q3

i
Pe
Q - N TQt dx dy (8-1 07d)

fe —
fc
m B tTaB dc (8-1 07e)

Cm 11
N T £ B crrr?s dC (8-107f)
Jc
—
-
v
Pe
jC
'N rq sB r dC (8-107g)

For the three-node triangular element, the stiffness matrices and nodal force
vectors are each of sizes 3x3
and 3x1, respectively. It can be seen almost by
inspection that each of these stiffness matrices is symmetric. Note the use of the
subscript 6 on those terms that arise from the boundary conditions. In these cases,
the integrations are to be performed around the boundaries of each element. How-
ever, if all legs of the element are internal (i.e. , within the body) and not on the
global boundary, the corresponding stiffness matrices (i.e., and KrB
e
) and the
nodal force vectors (i.e., f£.s , f,B ,
and f'B ) are simply taken to be null matrices
and vectors, respectively. This is illustrated in Example 8-17.

Example 8-12
Evaluate the stiffness matrix from conduction in the x direction by using the three-
node triangular element from Sec. 6-4.

Solution

Recall that the shape function matrix is given by Eq. (8-92) for the triangular
element. From the definition of K£r given by Eq. (8- 106a) we see that the first

derivatives of the shape functions are needed with respect to ,v. From Eqs. (6-21),
we note that

m, - dN,
=
ay*
HI 21 H'22 HI 23
dx
’
“ dX dx

Therefore, we may evaluate as follows:

m2 1

k; m 22 m22 dx dy
-L L'«23j
kt[m 2 1
hi 23 ]

mh HI21HI22 m 2l m 23
= ktA < n 22 m 2\ HI22 m22 m23 (8-108)

HI23HI21 HI 23 11*22 11*23 J

The m,j's and area A are a function only of the nodal coordinates and may be
computed with the help of Eqs. (6-21). In Eq. (8-108), the integrations were
performed by treating k and t as constants. If these parameters are not constant,
the above expression still holds, but k and t then represent the values at the element
 1

404 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

centroid. The accuracy of this approximation improves as the number of elements

used in the discretization is increased.
The reader may show in a completely analogous manner that the stiffness
matrix from conduction in the y direction is given by

mil m ii m n m3i m3i

Kjy = ktA m 32 m 31 ml 2 m32 m3i {8-109}
_m 33 m3| m33m32 m33 _
for the triangular element. Note that both K[, and K£>, are symmetric as
expected.

Example 8-13
Evaluate the element stiffness matrix from lateral convection for the three-node
triangular element

Solution

From the definition of Kf v given by Eq (8- 106c), we see that the integrand contains
the shape functions directly (not the derivatives) Therefore, it is more convenient
to replace each shape function Np with its corresponding area coordinate Lp from
Eq. (6-25). The integrals may then be evaluated with the help of the special inte-
gration formula given by Eq. (6-49) From Eqs (8- 106c) and (6-25), we have

~L~ Lj L.Lj L,Lk

1
K ‘
Lk ]h dxdy = hdxdy
-L L,
Lk_
[L, L,
J *'
LjL,
Lk L, Lk Lj
Lj LjL,
Ll J

Using Eq. (8-49), we may evaluate a typical integral as follows

-

J
Mj dxdy = hf^L? dx dy

= h
2 0 0 ,

—
' '

—2 A = / hA
[2 + 0 + 0 + 2))
Another integral is evaluated as

'^hL,Lj dx dy = hj^L.Lj dx dy
J

1 |
ini
= h :2 A = Yah
(1 + 1 + 0 + 2)!

and so forth. Clearly K'v must be given by

f2 1

^12
,
A 1

1 ( 8 - 110 )
12
[l 1 2.
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION 405

Recall that h is the sum of the two convective heat transfer coefficients on the two
lateral faces of the plate. Moreover, if h varies with x and y, the value of h at the

centroid of element should be used in Eq. (8-1 10). E3

Example 8-14
Try to evaluate the element stiffness matrix as a result of radiation from the lateral
faces. Use the three-node triangular element.

Solution

Inspection of the expression for Kf given in Eq. (8-1 06d) reveals that (Na e ) is

needed. Using area coordinates for the shape functions and noting that

a' = [T, Tj Tk f
we have
(NaU 3 = (L,T, + LJ TJ + Lk Tk f ( 8 - 111 )

Since this is a scalar, we may write

r i,
Kf = ect Lj [L, Lj Lk ](L,T, + LjTj + Lk Tk f dx dy ( 8 - 112 )
Ja<
M.
The indicated multiplications could be carried out and Eq. (6-49) used to integrate
the various terms. However, this approach is not very practical, and a numerical
integration should be performed. Such integration methods are covered in detail in

Chapter 9. The unknown temperatures appear in this matrix and the resulting
problem is nonlinear. The direct iteration method from Sec. 8-2 may be used in

this case. B
Examples 8-12 to 8-14 have illustrated how the integrals are evaluated over

the area of the element (i.e., over A e ). In the expressions for Kflfl and Kf fl ,
we
note that the integrations are to be performed around the element boundaries (in a
counterclockwise direction). However, there is no convection or radiation (assuming
an opaque body) between two internal and adjacent elements. Therefore, these two
matrices need to be evaluated only for those elements with at least one leg on the
global boundary B as shown in Fig. 8-12 for the triangular element. Note that leg
ijhappens to be on the global boundary. The next example illustrates how the

corresponding stiffness matrix from convection is evaluated in this case.

Example 8-15
Evaluate K£ B for the element shown in Fig. 8-12 with leg ij on the global boundary.
406 STEADY-STATC THERMAL AND FLUID FLOW ANALYStS

Solution

From Eq. (8-106e), we note that the shape functions themselves are needed in the
integrand. This generally means that area coordinates should be used to facilitate
the evaluation of the integral in this two-dimensional application. Therefore, Eq,
(8-106e) becomes

K L k dC
]

But on leg (/, we have Lk = 0 and dC - dl, or

Lf L,Lj
LjL, Lj
0 0

Note that the integration is to be performed only on leg ij, the length of which is
denoted as ltJ The area coordinates have in effect degenerated to length coordinates.
.

Equation (6-48) may now be used to perform the integrations For example,

2 0
’ ’
2 hB tl„
hgtLf dl = hB l
1, (2 + 0 + I) 1
,J
6

l W '
<" = *•'(, + ‘i +
1
’

!,/»
-

The complete result is given by

2 I 0
for leg ij on the
1 2 0 {8-1 13aJ
6 global boundary
0 0 0

The results from Example 8-15 can be generalized. If nodes j and k are on the
global boundary, we have
0 0 ol
for leg jk on the
K Us = 0 2 1
global boundary
{8-1 13b)

0 1
2J
whereas if nodes k and 1 are on the global boundary,
be we have

2 0 ‘1
"J 0 for leg ki on the (8-1 13c)
r
6
Li
0
o
0
2J
global boundary

Strictly speaking,
where ljk and /*, denote the lengths of legs jk and ki, respectively.
of the element along
the thickness t in Eqs. (8-1 13) represents the average thickness
the leg in question.
 GENERAL TWO DIMENSIONAL HEAT CONDUCTION 4Q7

The reader should try to show that the stiffness matrix given by Eq. (8-106f)
may be written as

L,
K rB e B crt [L, L, 0 )(L,T, + LjTj)
3
dl (8-114)
"I
for the element shown in Fig. 8-12. This integral may be evaluated with the help
of Eq. (6-48). It may also be evaluated numerically using one of the techniques
presented in Chapter 9.
Let us now turn to the nodal force vectors given by Eqs. (8-107). Note that
fcf, , ff, f*, and f
q are to be evaluated by integrating over the element area A'\
whereas f£ B frB and f'B are
, , to be evaluated on the element boundaries. The former
are readily evaluated if area coordinates are used with the following results:

ff. = hATn (8-115)

EdATf
ff = (8-116)

f«
sA 1 (8-117)
3
1.

1
QAt
ft-
l 1 (8-118)
Q
3
1

The result for f§ in Eq. (8-118) deserves special attention. This nodal force
vector results from a volumetric heat source, generally considered to be distributed

Figure 8-12 Typical element with at least one leg of the triangular element on the global
boundary.
408 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

within the body. If Q is not uniform, then the value at the element centroid is to
be used in Eq. (8-118). Let us derive an expression for the nodal force vector if
Q
happens to represent a point source at (x0 ,y0). With the help of the 8-function
defined by Eq. (4-85), we may represent the nodal force vector for the point source
as

-L Q't 8(x - x0 ) 5(y - y0) dx dy (8-119)

where Q' is the energy generation per unit time and per unit thickness. From the
definition of the 8 function, the above integral evaluates to zero everywhere within
the element except at (x0 ,>’o) Therefore, it follows that f£ is given by

WXoJo)
*6 = Q'i N,{x0 ,y0 ) ( 8 - 120 )
.A'iUoO'o).

If more than one point source is present in the element, additional nodal force
vectors similar to Eq. (8-120) would simply be included in the analysis
From a practical point of view, it is best to situate a node at each point source
Not only does this give more accurate results, but it also allows us to consider the
point sources after the assemblage is complete For example, if a point source of
Q’ is located at global node /, then Q't is simply added to the /th position in the

assemblage nodal force vector If the point source is withm an element (not at a
node), we must use Eq (8-120) at the element level Section 8-14 discusses one
method of implementing Eq (8-120) in a computer program
Let us now turn to the nodal force vectors that are to be evaluated on the
element boundary The results are

for leg ij on the

( 8 - 121 )
global boundary

E B (T tl,.T*B
T for leg ij on the
( 8 - 122 )
2 global boundary
_o_

"l"
<7 for leg ij on the (8-123)
I
2 global boundary
.0.

Similar expressions may be set up by inspection if leg jk or ki is on the global

boundary.
The assemblage of the element stiffness matrices and nodal force vectors is

conditions, if
routine (see Example 8-17) The prescribed temperature boundary
illustrated in Sec. 3-2. It has
any exist, are applied with either of the two methods
nonlinear.
been noted that if radiation heat transfer is included, the problem becomes
to be used.
The problem is nonlinear also if temperature-dependent properties are
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION

In these cases, the direct iteration method from Sec. 8-2 may be used to solve the
system of equations for the nodal temperatures.

The Element Resultants

The heat fluxes q x and q in the .t and v directions, respectively, may he computed
y
for every' element by applying Fourier’s law of heat conduction and writing

<?r
= ~k dT _
~ -k
dx
3N e
a
,
—
dx
(8-1 24a)

and

CI\ = -k
dT _
dv
~ —k — af (8-1 24b)
dx

If the triangular element is used, we get

= -k(m 2 J, + m 22 Tj + (8-125a)

= -k(m 3 ,T, + m n Tj + m i3 Tk ) (8-125b)

where the m,/ s are calculated by Eqs. (6-21) and T„ Tp and Tk represent the nodal
temperatures. Recall that these temperatures are known from the solution of Ka = f.

Note that a bar (~) is used on q x and q x

to indicate that these are actually average

heat fluxes for the element. This is a consequence of the fact that the shape functions
are linear (and the derivatives are therefore constants). These average heat fluxes
are usually associated with the centroid of the element.

Example 8-16
Consider the element shown in Fig. 8-13. Note that a point heat source with a

strength of 500 Btu/hr per inch of thickness is located at .r

0 = 1 in. and y0 =
0.2 in. The nodal coordinates are shown in Fig. 8-13. The plate thickness is 0.5 in.

The element is defined as having nodes 1.3. and 4, in this order.

Figure 8-13 Element in Example 8-16 with the nodal coordinates shown in inches and a

point heat source (denoted as x) at (1. 0.2).

410 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

Solution

From Eq. (8-120) we see that the shape functions need to be evaluated at (*0 ,.y0 ).
The shape functions for the triangular element are given by Eqs. (6-21). Each of
the m,j s needs to be computed with the help of Eq. (6-21) with the results sum-
-
marized as follows

mu = 1 0 m 12 = 0. m 13 = 0.

m 21 = —0.50 m 22 = 0 5625 m23 = —0.0625

m 3t = 0. m 32 — — 1 25 m }} — 1.25

In these computations the area A of the element was needed and was computed to
2
be 0.8 in from Eq (6-21e) The shape functions may now be evaluated at the
location of the heat source as follows

N, ~ m,, + m2 iar0 + "» 3 ,y0

= 0 + (-0 50)(1 0) +
1 (0 00)(0.2) - 0.5000

Nj = m n + m 22 0 + m32 y0 jr

= 00 + (0 5625)(1 0) + (- 1 25)(0 2) = 0 3125

N* = m, 3 + m 23 x0 + m 33 y0
= 00 + (-0.0625)(1 0) + (1 25)(0 2) * 0 1875

Note that the sum of the shape functions evaluated at (x0 ,y0 ) is unity From Eq
(8-120) we compute f£ as follows

Note that nodes 1 and 3 are closer to the source than node 4 and, therefore, most
of the source is automatically allocated to nodes 1 and 3 It should also be noted

that the sum of the values allocated to each node is 250 Btu/hr, the total amount
available. ®

Example 8-17
The temperature distribution is needed in the device shown in Fig. 8- 14(a) This
device very long in the direction normal to the plane of the paper The upper
is

and lower surfaces convect to fluids at 36 and 24°C, through convective heat transfer
held
coefficients of 500 and 1000 W/m -°C, respectively The outside boundary is
2

of 35°C The device comprised of two materials that are

at a fixed temperature is

fused together at x = ± 3 cm The materials are bronze and aluminum with thermal
source with
conductivities of 52 and 186 W/m-°C, respectively. A volumetric heat
temperature
a strength of 5 W/cm3 is present in the bronze only. Determine the
 1

GENERAL TWO-DIMENSIONAL HEAT CONDUCTION 41

distribution within and the heat fluxes through the device. Neglect the effects of
thermal radiation.

Solution

Because of the symmetry about the centerline, only one-half of the region needs
to bemodeled as shown in Fig. S- 14(b). The boundary condition on this line of
symmetry is equivalent to that of perfect insulation (i.e., the natural boundary
condition). A problem such as this one is usually analyzed with hundreds of ele-
ments. However, in order to show each step in the finite element solution process.

h B - 500 w/m J •
°c

__
Prescribed
temperature

Example 8-17. (b) Region to be analyzed with

Figure 8-14 (a) Schematic of device in

mechanisms shown, (c) Discretization of the two-dimensional region.

heat transfer
412 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

let us use only four triangular elements as shown in Fig. 8- 14(c). The nodal co-
ordinates are obtained from a scaled drawing of the cross section and are summarized
in Table 8-1 along with the element definitions. The bandwidth of the assemblage
stiffness matrix is minimized by numbering the nodes as shown. Note that two
material set flags are used, one for each material. The element stiffness matrices

and nodal force vectors are calculated as shown below. The assemblage stiffness
matrix and assemblage nodal force vector are also determined after each element
is processed.

Element 1

Element I is defined as having nodes 1 , 3, and 2 (in this order) and material set

1. The nodal coordinates are summarized in Table 8-1. Recall that the element
stiffness matrix for conduction in the x direction is given by Eq. (8-108). The area
of the element is computed from Eq. (6-2 le) as follows:

"l "l 0.000 -0.020"

yil
A = Videt 1 -*3 y3 - '/2 det 1 0.030 -0.017
.1 *2 yi\ .1 0.000 0.020_

OT

A = 0.00060 m 2

The three mv ’s that are needed are computed from Eq (6-21d) as follows-

~ ~ ^
m" _— y» y* _
~ y-i

2A ~1T~

Table 8-1 Node and Element Data for Example

8-17

Node number x, cm >. cm

1 00 -20
2 00 2.0
3 3 0 -1.7
4 30 1 7
5 5 0 -1 0
6 5 0 1 0

Element
number
— k set
Material
number
i J

1 1 3 2 1

2 2 3 4 1

3 3 5 4 2
4 5 6 4 2
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION 413

-0.017 - 0.020
~ -30.833 m
2(0.00060)

37 ~ -
HI 22
y,
= j’2 ji
2A 2A

0.020 - (- 0 . 020 )
33.333 nr 1

2(0.00060)

v.
~ Vi ~ )’3
'”23 =
2A 2A
-0.020 - (-.017)
-2.500 m' 1

2(0.00060)

The thermal conductivity k for the material comprising this element is 52 W/m-°C.
Therefore, for Kf t we get

(-30.833) 2 (-30 833)(33.333) (-30 833X-2 500)

K !!
1
(52 )(1 0)(0.0006) (33 333)( — 30 833) (33 333)' (33.333X- 2.500)
( ~ 2.500)( — 30 833) (-2 500) 2
(-2.500X33 333)

or

29.662 -32.067 2.405

K™ = -32.067 34.667 -2.600 w/°c
2.405 -2.600 0.195

Note that the thickness is taken to be I m since the body is very long in the direction
perpendicular to the two-dimensional region being analyzed. Any value for the
thickness t may be used as long as the same value is used in each element. Similarly,
in the y direction we need
xk - .V, x2 ~ *3 0.0 - 0.030
2A 2A 2(0.00060)
— 25.000 m~ i

- xk
*.r, x, “ -v 2 0.0 - 0.0
~ 2(0.00060)
2A 2A
0.

x, - x, x3 - x i
0.030 - 0.0

2A 2A 2(0.00060)

= 25.000 m' 1

The element stiffness matrix for conduction in the y direction is calculated from

Eq. (8-109) as follows:

(-25.0) 2 (-25.0X0.0) (-25.0X25.0)

K<>> = (52.)(1.0)(0.0006) — 25.0)
(0.0)( (0 0
)-
. (0.0)(25.0)

(25.0X — 25.0)
2
(25.0X0.0) (25. 0)
414 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

or

19.500 0.000 -19.500

K<[> = 0.000 0.000 0.000 W/°C
- 19.500 0.000 19.500

Leg ij or 1-3 is on the global boundary and undergoes convection to a fluid at

a temperature of 24°C. The convective heat transfer coefficient hB is 1000
W/m2-°C. From Eq (8-1 !3a), we see that we also need the length of leg 1-3. This
length is easily computed from

l,j = V(Xj - r,)

2
+7), - y,)
2 = V(jt3 “ *i)
2
+ (}j “ Vj)
2

= V(0.030 - + [-0.017 - (-0.020)] 2

J
0.0)

or

/„
= 0.03015 m
The element stiffness matrix from comection from leg ij is calculated as follows:

2 1 0
(1000 )(1 0)(Q.0301S)
K& = 1 2 0
0 0 0

or

‘10.050 5 025 0000

K& - 5 025 10.050 0.000 \v/°c
0.000 0000 0 000

There are no other contributions to the element stiffness matrix because the
boundary radiation is neglected and the lateral effects are zero (because a two-

dimensional slice out of a long body is being analyzed). Therefore, the composite
stiffness matrix for element 1 is given by Eq. (8-104) or, m this case, as

K (1> = + KU» + K<1),

or

59 212 - 27.042 -17.095'

Km -27.042 44.717 - 2.600
-17.095 - 2.600 19.695_

The assemblage stiffness matrix after processing one element is given by

59.212 - 17.095 -27.042 0.000 0.000 0.000

-17.095 19.695 -2.600 0 000 0.000 0 000
-27.042 -2.600 44 717 0 000 0.000 0000
0.000 0000 0.000 oooo 0.000 0.000
0000 0.000 0.000 0.000 0.000
0.000
0.000 0.000 0.000 0.000 0 000 0000 J
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION 415

Recall that Ku is zeroed out before adding the results from the first element.
The element nodal force vectors are now calculated as follows. The volumetric
6
heat generation Q of 5 W/cm 3 or 5 x 10 \V/m 3 is present in clement From Eq.
1 .

(8-1 18) we get

(5 x 6
10 )(1.0)(0.0006)
T 'lOOO.'

W =
3
i

i
=: 1000.
1000.
W
Leg ij or 1-3 undergoes convection and the corresponding nodal force vector is

given by Eq. (8-121) or

(1000)(1 .0)(0.03015)(24)
V '361.8'
=
o
1 = 361.8 W
0 0.0

All other contributions to the composite nodal force vector are zero, and from Eq.
(8-105) we have
f'" = ffe" +
or

1361.8
f(!)
_ 1361.8 W
1000.0

The assemblage nodal force vector after processing one element is given by

1361.8
1000.0
1361.8
f" = \V
0.0
0.0
0.0

Element 2
Element 2 defined as having nodes 2. 3. and 4 (in this order) and material set
is
I

with k = 52 \V/m-°C and Q = 5 x 10 W/m Moreover, leg ki or

6 3 4-2 is on the
.

and Ter =
c
global boundary and undergoes convection with h p = 500 \V/nr- C ,

36°C. The results are summarized below.

x2 y:
A Vz det •^3 V?
-

L >4.

0.000 0.020'

Vz det 0.030 -0.017 = 0.00051 nr

0.030 0.017_
41 6 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

m 2\ = -33.333 m 22 - -2.941 m 23 = 36.274

m3 =
,
0.000 m i2 = -29.412 m i3 = 29.412

29.467 2.600 -32 067

K<« = 2.600 0 229 -2.829
.-32.067 - 2.829 34.896.
*0.000 0.000 oooo"
K<« = 0.000 22 941 -22 941
Lo.ooo - 22 941 22 941.

l h = 0 0315 m
‘5 025 0 000 2 512"
= 0 000 0 000 0 000
.2 512 0 000 5 025.

k < 2) = K g> + K<*> + K&

34.492 2 600 - 29 554'
[
- 2 600 23.171 -25 771
_ -29 554 -25.771 62 862

59 212 - 17 095 - 27 042 0 000 0.000 0 000

- 17 095 54 187 0 000 - 29 554 0 000 0 000
-27.042 0000 67 887 -25 771 0.000 0 000
0 000 - 29 554 - 25 771 62 862 0 000 0 000
0 000 0.000 0 000 0 000 0 000 0.000
0 000 0 000 0 000 0 000 0 000 0 000

850 O'
2)
f£, = 850 0
|_ 850.0.
 GENERAL TWO DIMENSIONAL HEAT CONDUCTION 4J7

Element 3
Element 3 is defined as having nodes 3. 5, and 4 (in this order) and material set 2
with k — 186 W/m-°C and 0 = 0. Leg ij or 3-5 is on the global boundary and
undergoes convection with hB = 1000 W/m 2 -°C and TaI) = 24°C. The results are
summarized below.

A = 0.00034 m 3

mn = -39.706 m 22 ~ 50.000 m23 = - 10.294

»i 3! = -29.412 mn - 0.000 m 3} = 29.412

99.701 -125.550 25.849'

K< 3> = -125.550 158.100 -32.550
25.849 -32.550 6.701

54.706 0.000 - 54.706

K< 3 > = 0.000 0.000 0.000
_- 54.706 0.000 54.706_

/
v = 0.02119 m
'7.063 3.532 o.ooo'
K& = 3.532 7.063 0.000
0.000 0.000 0.000

K (3) = K' 3) + K< 3) + Kg),

161.471 -122.018 -28.857

-122.018 165.163 -32.550
-28.857 -32.550 61.407

59.212 - 17.095 -27.042 0.000 0.000 0.000

17.095 54.187 0.000 -29.554 0.000 0.000
27.042 0.000 229.358 -54.628 -122.018 0.000
K° -29.554 -54.628 124 270 -32.550 0.000
0.000
0.000 0.000 -122.018 -32.550 165.163 0.000
0.000 0.000 0.000 0.000 0.000 0.000

0
3)
fb 0
0

254.3
tSi 254.3
0.0

254 3
f<
3> = ig) + = 254.3
0.0
 J

418 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

1361.8
2121.3
2466.1
1121.3
254.3
0 0_

Element 4
Element 4 is defined as having nodes 5, 6, and 4 (in this order) and material set 2
with k = 186 W/m-°C and Q = 0 Leg jk or 6-4 is on the global boundary and
undergoes convection with h B = 500 W/m 2 -°C and TaB = 36°C. The results are
summarized below

A = 0 00020 m2
m 21 17.500 m 22 = 67 500 "*23 -50.000

m 3l = - 50 000 m 32 = 50 000 mi3 = 0 000

11 393 -43 943 32 550

K£> = -43 943 169 492 -125 550
32 550 - 125 550 93 000.

000 - 93 000 0 000

000 93 000 0.000
000 0 000 0 000

ljk = 0 02119 m
[0 000 0 000 0 ooo'
0 000 3 532 1 766
0 000 1 766 3 532

Kw = Kw +
104 392 -136 942 32 550
= -136 942 266 024 - 123 784
L 32 550 - 123 784 96 532

59 212- 17 095 -27.042 0.000

0.000 0 000
- 17 095 54 187 0.000 -29.5540.000 0.000
-27.042 0 000 229.358 -54.628--122.018 0.000
K fl

0.000 -123.784
0 000- 29.554 -54.628 220 801
0.000 0.000 - 122.018 0 000 269 556 - 136 942
0.000 0 000 0.000 123 784 -136 942 266 024

0
4>
$ = 0
0.
 GENERAL TWO-DIMENSIONAL HEAT CONDUCTION
419

"
0 0 .

fW 190.7
.190.7

0.0
f(4) = + %% = 190.7
190.7

1361.8
2121.3
2466.1
fa
1312.0
254.3
190.7

Before applying the prescribed temperature boundary conditions on nodes 5 and 6,

we have 'a = f3 or
K ,

59.212 - 17.095 -27.042 0.000 0.000 0.000

- 17.095 54. 187 0.000 -29.554 0.000 0.000
-27.042 0.000 229.358 -54.628 -122.018 0.000
0.000 -29.554 -54.628 220.801 0.000 -123.784
0.000 0.000 -122.018 0.000 269.556 -136.942
0.000 0.000 0.000 - 123.784 -136.942 266.024.

1361.8
2121.3
2466.1
1312.0
254.3
_ i 90.7

After imposing the two prescribed temperatures with Method 1 from Sec. 3-2, we
get

59.212 -17.095 -27.042 0.000 0.000 0.000 r,~ 1361 .8

- 17.095 54.187 0.000 -29.554 0.000 0.000 T, 2121.3

-27.042 0.000 229.358 -54.628 0.000 0.000 r, 6736.7
0.000 -29.554 -54.628 220.801 0.000 0.000 TA 5644.5
0.000 0.000 0.000 0.000 1.000 0.000 r5 35.0

_ 0.000 0.000 0.000 0.000 0.000 1.000_ _T,._ 35.0_

The solution is given by

T, = 71.3°C Ti = 49.6°C Ts = 35.0°C

T2 = 88.8°C r4 = 49.7°C T6 = 35.0°C

The average heat fluxes through the elements may be computed from Eqs. (8-125).
For example, for element 1 we have
9<» = -52. [(-30.833X71.3) + (33.333)(49.6) + (-2.500)(88.S)J

= 39,900 W/m 2
420 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

and

5<'* = -52. [(-25.000X71.3) + (0.0)(49.6) + (25.000)(88.8)]

= -22,800 W/m !
or

5i'> = +3.99 W/cm 3

and

q?' = -2.28 W/cm 2

The negative sign indicates that the heat is flowing in the negative y direction. Both
of these signs seem to be intuitively correct (why?). The results for the other three
elements are as follows
~ 6.76 W/cm 2 q
,3) = 13.6 W/cm 2 4) = 13.7 W/cm 2
qjf* q^

5y
2) - —0 01 W/cm 2 q (3) = - 0.05 W/cm 2 q
l4) ~ 0.0 W/cm 2
Recall that these average heat fluxes are generally taken to be the local values at
the respective element centroids. Because of the crude mesh used, these results are
very approximate

One way to improve the results for the element resultants such as the heat
fluxes is to average the results of two adjacent triangular elements as shown in

Fig. 8-15 For example, if the results for elements I and 2 are averaged, we get

3.99 + 6.76
5.38 W/cm 2
2

and

_
qy =
5i
l)
+
2
~
V?' (-2.28) + (-0,01)
2
-1.14 W/cm 2

Figure 8-15 The method of quadrilateral averages Note how each pair of triangles forms
a quadrilateral. The clement resultants for each pair of triangles are averaged and assigned
to the centroid of the quadrilateral. Although this is illustrated here in a heat transfer ap-

plication, the same technique is used in stress analysis.

 AXISYMMETRIC HEAT CONDUCTION 421

whereas for elements 3 and 4, we get

13.6 + 13.7
<7t
= 13.65 W/cm 2

and

-0.05 + 0.0
<
7,
= -0.025 W/cm 2

This method of combining the element resultants for two adjacent triangular elements
is known as the method of quadrilateral averages. Note that a quadrilateral is

formed by the two triangles. The combined element resultants from two triangles
that form a quadrilateral are then associated with the centroid of the resulting
quadrilateral as shown in Fig. 8-15.

8-9 AXISYMMETRIC HEAT CONDUCTION

Three-dimensional heat conduction may be modeled as a two-dimensional ideali-

zation if axisymmetry exists with respect to both the geometry and the thermal
loads imposed on the body. A body of revolution is obviously geometrically axi-
symmetric as shown in Fig. 8-16. Each thermal load must also be axisymmetric,
however, in order to take the temperature T as a function of only the radial coordinate
r and the axial coordinate r. When conditions such as these exist, the problem is

said to be one of axisymmetric heat conduction.

Boundary
heat flux

that the body

Figure 8-16 Schematic of axisymmetric heat conduction problem. Note
is

axis, and the thermal loading is axisymmetric as well.

a body of revolution about the r
422 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

Only the isotropic case is considered in the formal development; that is, the
thermal conductivity k may be a function of r and z (for a heterogeneous body)
and even the temperature T, but it may not be dependent on direction. Clearly,
axisymmetnc heat transfer cannot occur in a fully anisotropic body. Only if the
body is stratified in such a way that the principal values of thermal conductivity
occur in the r and z directions (with thermal conductivities kr and kz , respectively)
can the problem be considered axisymmetnc The formulation to be presented could
easily be extended to include this special anisotropic case.
Three types of boundary conditions are considered, convection, imposed heat
fluxes, and prescribed temperatures. It should be apparent by now that convection
and imposed heat fluxes are inherently included in the FEM formulation. Prescnbed
temperature boundary conditions, on the other hand, are handled m the usual manner
after the assemblage step Thermal radiation from the surface of the axisymmetric
body is not included, but could easily be added by following the approach taken
in Secs 8-3 and 8-8

The Governing Equation

The governing equation for axisymmetnc steady-state heat conduction in an iso-

tropic, heterogeneous body with a volumetric heat source is given by

= °
£(‘f) +e
k + ,8 - ,26 >
-ri(' f)
where T is the temperature, k is the thermal conductivity, and Q is the heat generation
per unit time and per unit volume The global coordinates are denoted as r and z
in the radial and axial directions, respectively Note that Eq (8-126) is actually the

heat conduction equation in cylmdncal coordinates with the term representing the
circumferential conduction set to zero
The mathematical statements of the boundary conditions for convection and
imposed heat fluxes are more conveniently given later during the FEM formulation

The Finite Element Formulation

It is not really necessary at this point to choose the particular type of axisymmetnc
element to be used in the discretization However, after the general expressions for
the element charactenstics have been derived, the emphasis will be on the tnangular
donut-shaped element. This element and the rectangular donut-shaped element are
shown in Fig. 8-17 for a typical axisymmetnc body Additional axisymmetnc ele-
ments are presented in Chapter 9
The Galerkin method will be used on an element basis to denve the expressions
integral of the
for the element characteristics We begin by forming the volume
product of the residual and transpose of the shape function matnx
N and setting
the result to zero, or
 ' k

AXISYMMETRIC HEAT CONDUCTION

423

Figure 8-17 Typical axisymmetric elements: (a) the triangular donut-shaped element and
(b) the rectangular donut-shaped element.

fid/ dT\
—d dT
f
Jv
Nr --Ur-
rdr\ dr)
+
d: V
(
(
,
— \
d=)
1 Q (8-127)
L J

where it is appropriate to take

dV = 2-rr r dr dz (8-128)

for the elemental volume dV. It is emphasized that the expression in the brackets

in Eq. (8-127) is the residual that results when the approximation for the temperature
T on an element basis (i.e., T — Na r ) is substituted into Eq. (8-126). Note that
each term in Eq. (8-127) has dimensions of energy per unit time. If Eqs. (8-127)
and (8-128) are combined and if the result is split into three separate integrals, we
get

The 2t7 may be cancelled from each term, but we will continue to carry' it. Note
that the integration limit has been changed from Ve to Ae . Recall from Sec. 6-6
that we be analyzing the half-plane as shown in Fig. 6-17 for a
will, in effect,
typical axisymmetric body. For all practical purposes, the problem has been made

two-dimensional with the thickness t (in the two-dimensional problem) being re-
placed by 2-rr rl However, let us explicitly show how the expressions for the element
424 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

characteristics can he derived when the integrands are functions of one or more of
the global coordinates. Because r is independent of z, we should note that

d_
18-130)
dz

Furthermore, if we apply the Green-Gauss theorem to the two terms in Eq. (8-129)
that involve second-order derivatives, we get

r
8N dT
nr dC — 2it
Ja'
f
1

dr
kr —
dr
dr dz

+ 2ir
[
Jc*
N Tkr — n dC 1
dz

dN T
— 2tt I
Ja'
f

dz
.
kr — dr dz
dT
dz

+ 2tt N TQr dr dz = 0 (8-131)

!a<

where Ce denotes that the integrations are to be performed around the boundary of
the element (in a counterclockwise direction) The direction cosines nr and n. are
shown in Fig 8- 8(a) for a typical triangular element
1 In a manner completely
analogous to Eq (8-48), we may write

9« = (8-132)

where qn represents the net normal heat flux from conduction from within the
element to the boundary of the element as shown m Fig. 8- 18(b) As mentioned
earlier both convection and imposed heat fluxes are to be considered It is convenient
at this point to perform an energy balance on the global boundary as shown in

Fig. 8- 18(b) If qsB and q CiB denote the imposed and convective heat fluxes,
respectively, then we may write

= (8-1 33a)
+ 9,8 9ofl

or

9»
_ 9ob “ (8-133b)

We must now invoke Newton’s law of cooling, or

qor = hgfX ~ Trf)

(8 ’ 134)

where hB is the convective heat transfer coefficient and TaB is the temperature of

the fluid far removed from the boundary If Eqs (8-13 1) to (8-134) are combined,

we get

T
-2u NT B liT
l, - TaB dC + ) 2tt
£ N q„r dC
 AXISWMfTRIC HEAT CONDUCTION 425

< bl

Figure 8-18 (a) No-ma' unit itrctcr/j shown fora typical triangular axisxmmetnc clement
(b) Tvp ca! part o r the g!oba< boundary in tha axisymmatnc haat conduction problem showing
tr.a haat fluxc< considared m tba finite elaman; formulation

r st ax 7 ar
— ,
2 -rr
u
f
I
j '
<3N

or
f
Lr —
or
J
Jr Jr
.
— 2tr
r
I

M' dr
,
kr —
or
,
Jr Jr
,
(8-135)

- 2m |
y rQr dr Jr = 0

But the temperature 7 uithm an element e is gixen by

7 = Na r (8-136)

r
where N is the shape function matrix and a is the sector of nodal temperatures
for element e For the three-node triangular element, N is gixen by

X = [A', A, r
NJ (8-137)

and

3' = r (8-138)
(7, Tj 7J
where the shape functions themselves are given by Eqs (6-21) with rand \ replaced
by r and r If the temperature 7 in Eq (8-135) is eliminated by using Eq (8-136),
xx e get
426 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

K ea' = f
r
(8-139)

where K* may be referred to as the composite element stiffness matrix and F as

the composite element nodal force vector, both defined as

K' = + Kf. + K'„ b (8-140)

and

f
e = *'q
+ ft* + ft* (8-141)

The element stiffness matrices themselves are defined by

8N r
K' = 2ir
f
I

•M* dr
kr —
5N
dr
dr dz (8-1 42a)

K' = 2ir
(
Ja»
—— kr
dz
— dz
dr dz {8-142M

and

= 2tt N rA r N dC (8-142cl
f fl

and the element nodal force vectors by

fj - 2ir
f _
N r2r dr dz !8-143»l

V„ = 2zz _N Tq,B rdC (8-143W

jc

and

- 2*
j^
N ThB rT^ iC (8-1430

It is emphasized that the element characteristics given by Eqs. (8-142) and

(8-143) are quite general and may be applied to any suitable axisymmetnc element.
Let us illustrate theuse of the triangular axisymmetric element by way of several
examples.

Example 8-18
Evaluate the element stiffness matrix from conduction in the radial direction for

the triangular axisymmetnc element.

Solution

From Eq. (8- 142a), we see that the denvatives of each of the shape functions are
needed with respect to r. Recall that the shape functions for this element are given

by Eqs. (6-21) with r and z replacing x and v, respectively, or

+ m3 (8-1 44a)
N, = m,i + m2 i
r i
z

= (8-144b)
Nj m, 2 + m22r + "' 32 -
 AXISYMMETRIC HEAT CONDUCTION 427

A'x = "hi + m23 r + m 33 z {8-1 44c)

where the m t}
s are given by Eq. (6-2 Id) with r, and z, replacing x, and v,, etc.
Therefore. Eq. (8- 142a) becomes

in 2 1

L m 22 [/n 2l m 22 w,5 ] kr dr d: (8-145)

in 22

Since the m,/ s are constants and the thermal conductivity is assumed to be constant
(in each element). Eq. (8-145) may be written

m 2l
K'rr
lv
-

= —
1-JT41 m 22 ("Li in 22 r dr d: (8-146)
ni2 3

It proves to be very convenient at this point to represent the radial coordinate r

over the element as follows:

r = N,r, + N/j 4- j\\rK (8-147a)

where r,. rr and rk denote the radial coordinates of nodes i, j, and k. If the shape
functions are written in terms of the area coordinates [see Eq. (6-25)]. we get

r = L,r, 4- L/j 4- Lk rk (8-1 47b)

Note that at node i. the radial coordinate r becomes r,. etc. Moreover, the area
coordinates (like the shape functions in this case) vary linearly with r over the
element. It seems quite reasonable, therefore, to use Eqs. (8- 147b) to represent r.

The integral in Eq. (8-146) may be written as

r dr dz = r. L, dr dz 4- r,
>
I L,1 dr dz
'
Ja< Ja< (8-148)

4- rk Lk dr dz
J
Using the integration formula given by Eq. (6-49). we may evaluate a typical

integral as follows:

1 !0!0!
dr dz = 2A = >M
L L.
(1 4-0 4- 0 4- 2)!

It should now be apparent that

f r dr dz = 'Air, 4- r. 4- rk )A (8-149)
}A‘

where A. the area of the element, may be calculated with the help of Eq. (6-21e)
if r, and z, are used in place of x, and _v,, etc. Let us defined r as

r = '/3 (r, + rj 4- rk ) (8-150)

so that the final result for K.% may be given as

428 STEADY-STATE THERMAE AND FLUID J LOW ANALYSIS

m2 \
m2i m22 m 2l m 23
K'r = 2irKrA tn 22 tn 2] mj2 m 22m 23 {8-151}
m 23m 2] m23m22 mj3 _

The reader should compare the approach taken here with that used in Sec. 7-3 for
axisymmetric stress analysis

It can be shown in a similar manner that the element stiffness matrix from
conduction in the axial direction is given by

f»3l fft3 ,m 32 m 3 im 33
K' 2-nAM m 32 m 3l ml2 m i2m n (8-152)
,m 33 m 3l m 3im 32 mj3 _

for the triangular clement In Eqs. (8-151) and (8-152) k is interpreted to be the
average value of the thermal conductivity over the element if it varies from point
to point The clement stiffness matrix from boundary convection is evaluated in
the next example for the triangular element

Example 8-19
Evaluate K?,* given by Eq. (8-142c) for the element shown m Fig 8-19

Solution

From Fig. 8-19 we note that leg ij of element e is on the global boundary B. In
order to facilitate the evaluation of Eq. (8- 142c), we should represent the shape
functions in terms of the area coordinates. In this case, the integral is to be evaluated
on leg ij where Lk = 0 (and N =
k 0) Therefore, on leg ij the shape function matrix
may be written

N * [L, L, 0]
(8*153)

and the matrix K‘ vB may be written

Figure 8-19 Typuial element with at least one leg of the axisymmetric triangular element
on the global boundary.
 !

\MMM\tnKir in \t rovmriioN

1/., L, 0) h l: r ,\l (8-154)

Let us represent r on leg ij as follows:

r = A',r, 4- A',/) = L,r, 4- L,r, (8-155)

Thus, Eq. (8-154) becomes

o'
Kf,g = 2"//,; 0 (L,r, + ill
1 _ 0
At
0 0
/./,)

where /i
B is assumed to be constant over leg ij. It should be noted that them me
two different types of terms that need to be evaluated. The first is evaluated as

f L;(L,r,) tU = r, f L' ,11

Jill Jly

3!0!
” '' 0
(3 4 « 4 I >!

(8-15Gn)

and the second as

L,Lj{L,r,) ,11 = r, f Lfl., (1!

[
Jly

2! I

= /*.

'(2 4 14 I)!

(8-ir»f.b|

where /„ represents the length of leg //. The final result lot K;,„ is given by

3r, 4- ry r, >j

K cfl8 = 2"V„ r, 4- r, r, 4- 3r, (lor leg ii oil //) (8-157)

12
0 0

The reader should be able to derive the corresponding tesult il leg /A oi A/ happens
to be on the global boundary. ®
The three element nodal force vectors may be evaluated m a similar fashion

with the results

2r, 4- '•/
*
2nQA (ft-lbfl)
f
Q = r, 4- 2 r; 4
12
/> 4
r, 4- 2r,

2r 4 t
-

2-nli„Tf,n l, l
ft ~ r, 4- 2r, (for leg // on //) (0-150)
*c\B
430 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

Equations (8-159) and (8-160) are valid if leg ij is on the global boundary. Similar
expressions could be written by inspection or easily den\ed for legs jk and ki. The
reader should dense the expressions for f£ in the case of a line source Q' at (r^Hj),
where Q‘ is the heat generation rate per unit circumference (i e., at r = r0 and
Z = 5o).
This completes the evaluation of each of the element stiffness matrices and
each of the nodal force sectors. The assemblage equations are formed in the usual
manner. The prescribed temperature boundary conditions, if any are present, would
be applied next The solution for the temperatures would then follow and the heat
flows through the body could be evaluated If the thermal conductivity k or con-
vective heat transfer coefficient h B is dependent on the temperature, the direct
iteration method from Sec 8-2 may be used to sohe for the nodal temperatures.
If a nonsymmetncal thermal load imposed on the body, it may be possible
is

to perform an axisy mmetnc analysis by using a one-term Fourier senes of sine and
cosine functions to represent the thermal loads and temperatures The loads, how-
ever. must still be symmetric about a plane through the r axis Huebner J2J and
Wilson (3J gi\e additional details Treatment of this aspect of axisymmetnc heat
conduction is beyond the scope of this book

8-10 THREE-DIMENSIONAL HEAT CONDUCTION

In this section the three-dimensional heat conduction problem is formulated Figure

8-20 shows an arbitrary three-dimensional body in an nr coordinate system The
boundary conditions that will be considered on the surface of the body are the
following, convection, imposed heat fluxes (possibly acting at a distance such as
the sun), and prescribed temperatures Recall that perfect msuiation is a special
case of zero heat flax. A volumetric heat source will also be included. Thermal
radiation from the surface is not considered but could easily be added
In Fig. 8-20, the heat fluxes from conduction in the x. y. and : directions are
denoted by q,. qy , and q.. These heat fluxes may be related to the
temperature T
within the body by Fourier's law of heat conduction for an isotropic body in three
dimensions, or

where k is the thermal conductivity The minus signs in Eq. (8-161) are necessary
decreasing
to ensure that each heat flux is positive if directed in the direction of

temperature-
 THREE DIMENSIONAL HEAT CONDUCTION
431

Boundary
heat flux
(may act at Boundary
a distance) convection

Figure 8-20 Schematic of three-dimensional heat conduction problem

The Governing Equation

The governing equation for general three-dimensional heat conduction in a heter-

ogeneous, isotropic body is given by

£ 0 ( 8 - 162 )
dx

where Q is the heat generation rate per unit volume The thermal conductivity A
may vary throughout the body and may even be a function of temperature. The
mathematical form of the boundary conditions is given during the FEM formulation

The Finite Element Formulation

As in the other heat conduction formulations, it is not necessary to choose the t\pe
of element to be used in the discretization in order to determine the expressions for
the element charactenstics The reader may recall that in Sec 6-5 two particular
three-dimensional elements were presented, namely, the four-node tetrahedral and
eight-node brick elements. After the general expressions for the element charac-
teristics have been derived, the emphasis will be on the tetrahedral clement Ad-
ditional three-dimensional elements are presented m Chapter 9
The Galerkin method will be used on an element basis to dense the expressions
for the element charactenstics Wc begin by writing
432 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

dxdydz = 0 (8-163}

Note that each term in Eq. {8-163) has dimensions of energy per unit time. Using
the Green-Gauss theorem on the three terms involving second-order derivatives
gives

JS' dx
x
Jv* dx dx
+ f
Js' dy >

T
Jv dy + f
J5‘
N k—„.dS
dy dz

T
* dz dz
+ J N Q drdytk = 0 (8-164)

Note that three surface integrations arise. These integrals are identified by the S'
on the integrals and must be evaluated over the face (of each element) that is on
the global boundary. However, at this point it is convenient to combine these three
surface integrals into one integral by noting that

dT dT dT
<tn = »r
- k-r-n k, n- (8-165)
dx dy dz

where q„ is the net (normal) heat flux from conduction, i e in the direction of the ,

unit normal n to the surface as shown in Fig 8-21. Equation (8-165) is the three-
dimensional form of Eq (8-48)
Now it is convenient to consider the convective and imposed heat flux boundary
conditions An energy balance on the surface of the body in Fig 8-21 gives

q„ + q,B — qC xB 18 - 1 66a)

or

66b)
qn = <lcx B ~ QsB (8-1

conduction
Figure 8-21 Typical part of the global boundary in the three-dimensional heat

problem showing the heat fluxes considered in the finite element formulation.
 THRFJ-.-DlMLNSIONAt. HUT CONDLCTION 433

Recall that the heat flux qtlB is proportional to the temperature difference between
the body and the ambient or

cI<nB ~ hB (T - TaB ) (8-167)

w'here hB is the convective heat transfer coefficient and TaB is the temperature of
fluid far removed from the body. If Eqs. (8-164) to (8-167) are combined, we get

- fy N%r dS + fy N %TlJ} dS + fy N V dS

3Nr bT
'
r
— f —— k — a dz
dx dx
i , , — f
flN
k,
hT
— ,
d\ dx d: , ,

Jl' dx dx ' JX' dv cfy

dT
- f
—— k — dx dv dz +
f
N r0 dx dv dz = 0 (8-168)
Jv' dz dz ' Jv

The temperature T within an element may be represented by

T = NV (8-169)

where the shape function matrix N for the four-node tetrahedral element is given
bv

N = \N, N, Nl Nn \
(8-170)

f
and the vector of nodal temperatures a by

a*- = (7*. T, Tk T„Y (8-171)

The shape functions themselves are given by Eqs. (6-40) in this case. If the tem-
perature T in Eq. (8-168) is eliminated by using Eq. (8-169). we get

K'a'" = P (8-172)

where Kr is the composite element stiffness matrix and P is the composite clement

nodal force vector, or

K' = K„ + KU + Kf + K'* r
(8-173)

and

P = f& + fjs + fT.fi

(8-174)

The clement stiffness matrices are defined as

r
K(, =
h'
f 8N
— —k
d.x
,
—
ON
rn
,
dx d\ dz
, ,
(8-1 75a)

ON T ON
"
Kf, =
Ji'
f

dv
k —
dv
dx dx dz (8-175b)

Kr. = (
J V'
—
<^ T
dz
k
on,
—dz
dx dy dz
, ,
(8-1 75c)
434 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

and

K,'„ = js,
m,s N ,IS (8-1 75d)

whereas the element nodal force vectors are defined as

f
6 = IwQdxdydz (8-1 76a)

fi = JS,N ?qj,dS (8-176b)

and

N Th,TaB dS (8-176c)

The next step is to evaluate these integrals for the particular three-dimensional
element used in the discretization For example, if the four-node tetrahedral element
is used, it can be shown that the results for the three element stiffness matrices
from conduction ar6 given by

'«!i m2\ m 22 m m23 m ,m 24

2l 2

K', = kV
nt 22 m2l ™22 m22 m23 m22 m24 (8-177)
m 2i m 2l m 23 m 22 mji m 2i m 24
_m 24m 2l m 2i m 22 m 2A m 2i w»24 _

Wl?l m M m i2 m 3 ,m 33 m 31 m 34
m}2 m3A
K* = kV (8-1 781
WjjOTj, m 33»h2 m 33 m 33/« 34
_m 34 m 3l m J4 W 32 m M m 3} '«34 _

and

m4l m4l m41

Kr. = kv m42m4l m42 m42 m43 m 42 m 4, (8-179)
n h) m Ai m A3 m 42 m 43
m 44 m 43 mli -

where the m tJ
' s and volume V are given by Eqs. (6-40).
The element stiffness matrix from boundary convection deserves special atten-
tion. Note that the integral for K^,s is a surface integral, not a volume integral.

The surface referred to here is the surface of the element. The tetrahedral element

has four surfaces and, strictly speaking, K^ fl

should be evaluated over each of these
faces. However, if all faces are internal and not on the global boundary, we must
take K'vS to be the 4x4
null matrix since there is no convection on the internal
faces. Moreover, if a face is on the global boundary but no convection is present,
then again K£ g is taken to be the 4x4
null matrix. A nontrivial case is shown
boundary
in Fig. 8-22, where face ijk of the element e is on the part of the global
that undergoes convection.
 THREE-DIMENSIONAL HEAT CONDUCTION

Figure 8-22 Typical element with at least one face of the tetrahedral element on the global
boundary.

Example 8-20
Evaluate the element stiffness matrix Kf,„ for the element shown in Fig. 8-22.

Solution

From Eq. (8-175d) we see that the shape functions themselves appear in the inte-
grand. It is convenient to represent the shape functions in terms of the volume
coordinates [see Eq. (6-43)]. Note that L,„ = 0 on face ijk. Therefore, Eq. (8- 1 75d)
becomes

LJL, LU 0

K
Lj LM 0
hB dS (8-180)
L,Lj Ll 0
0 0 0

where A ljk denotes the area of face ijk. Note that this face is a triangle and the
volume coordinates have degenerated to the three area coordinates. It seems ap-
propriate, therefore, to use the special integration formula given by Eq. (6-49). For
example, a typical term in the matrix may be evaluated as follows:

1 ! 1 !0!
2A,jk
(1 + 1+0 + 2 )! 12

The Final result is given by

2 1 1 0
hpA^t. 1 2 1 0 for face ijk on the
KciB
e

2 0 global boundary B
(8-181)
12 1 I

0 0 0 0

It is emphasized that this result is valid if face ijk of the tetrahedral element is on
the global boundary' B. The reader should be able to evaluate Kflg if any' of the
other faces happen to be on the global boundary. I
The nodal force vector f'Q may be evaluated with the help of Eq. (6-50) if the
shape functions are written in terms of the volume coordinates. The result is
438 STEADY-STATE THERMAL AND FLUID FLOW ANALYS IS

(8-187a)

and

(8-1 87b)

[f these expressions for the velocity components are substituted into the continuity
equation given by Eq (8-185), the result is the two-dimensional form of Laplace's

equation, or
2 2
8 <J)
<3 <f>
+ (8-188)
a? ay
2

The Galerkm method on an element basis gives

Nr
l [0 + 0]*^ = o l8 - 18SI

where the shape function matrix N is dependent on the type of element used. We
will derive the expressions for the element characteristics in general and then apply
these results to the triangular element in particular.
If the Green-Gauss theorem is used, we get

Nr &
dx
n, dC
•M' dx dx

+ f
Jc' ay
y
)Ia ‘
—
ay ay
= 0 (8-190)

Let us use Eqs. (8-187) to introduce the two velocity components into the for-

mulation or

-fJC' N'„„,* rfC- f

JA'
«£***dx dx

- [
Jc
N r viiy dC - [
)a'
— — dWy
dy dy
= 0 (8-1911

At this point it is convenient to represent the velocity potential 4> on an element

basis by writing

<f)
= Na' (8-192)

the values of the potential at the nodes for element

e.
where the vector a' contains
to
Since a' is not a function of x or y, we can combine Eqs. (8-191) and (8-192)

where the composite element stiffness matrix K* is defined by

(8-194)
K' = K£ + t
 TWO-DIMENSIONAL POTENTIAL FLOW

and the composite element nodal force vector F is defined by

f
1
e-
= O'
l uB
j. ft
•
he (8-195)
The element stiftness matrices K£ and
t are defined bv
r
d\’ AN
K» = dx dx (8-1 96a)
Iw i)X dx

and
’
aN 7 f}N
[ rfx rfv (8-1 96b)
-M r
-A' fly Ay
r)v

whereas the element nodal force vectors f£g and ffB are defined by

fIb = -j X T un dC r (8-197a)
Ct

and

ffa = - N rvn, dC (8-1 97b)

|C' *

Equations (8-196) and (8-197) are quite general in that they may be applied to any
two-dimensional element. For example, if the three-node triangular element is used.
K£ t
and Kf. evaluate to

MI J, m2\ m22 m2\ m23

= A m22m 2l MI ?2 m22m 23 (8-198)
m 22 m22 nh J >

and

my, m3l m32 MI 5 PM 33

Kf, = A MI 32 MI 3 |
mh MI 32 MI 33 (8-199)

L m 33 n, 31 MI 33 MI 32 "'53 .

The integrals for ffjJ} and ffg in Eqs. (8-197) need to be evaluated on the element
boundaries only. However, each direction cosine takes on opposite signs on legs
shared by two elements. In effect, the vectors f£g and ffg on an internal leg for one
element cancel the corresponding vectors on the same leg for the adjacent element.
The cancellation occurs at the assemblage step. Therefore, Eqs. (8-197) need to be
evaluated only for those elements with at least one leg on the global boundary. For
example, consider the triangular element shown in Fig. 8-24(a). Note that leg ij is
e
on the global boundary'. In Example 8-21, iuB is evaluated for the case when u is

assumed to vary' linearly over the leg as shown in Fig. 8-24(b).

Example 8-21

Evaluate for the element shown in Fig. 8-24(b). Note that u is assumed to vary'

linearly over the leg such that at nodes i and j. the x components of velocity are
t(, and ut respectively.
.
440 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

(0

Figure 8-24 (a) Typical element with at least one leg of the triangular element on the

global boundary (b) Linearly varying x component of velocity on leg ij of triangle, (c) Linearly

varying y component of velocity on leg ij of triangle

 TWO-DIMENSIONAL POTENTIAL FLOW 44]

where /
y
is the length of leg ij. If L, and Ly are interpreted to be length coordinates,
Eq. (6-48) may be used to integrate Eq. (8-201 ) to give

2m ,
+ Uj

u, + 2iij ( 8 - 202 )

The reader should show that if v is assumed to vary linearly over leg ij as
shown in Fig. 8-24(c), the result is

2v, + v,

fit?
— v, + 2v, (8-203)
0

where v,- and v} are the y components of velocity at nodes / and j, respectively.
Similar expressions could be derived or written by inspection if leg jk or ki happens
to be on the global boundary.
One method for determining the two direction cosines n x and n y for any leg of
the triangular element is now presented. Let us arbitrarily concentrate on leg ij and
define the vector r as the vector running from node
y
i to node j, as shown in Fig.

8-25. Therefore, we have

rv = (Xj - .v,) i + 0) - >’,) j (8-204)

where x, and y, are the coordinates of node i; x} and y} are the coordinates of node
j; and i and j are the unit vectors along the jcand y axes, respectively. The unit
vector n is defined to be the outward normal to leg ij such that

n = nj + n, j (8-205)

where the direction cosines n x and ny are to be determined. The reader may recall
that the cross product of two vectors results in another vector whose direction is

determined by the right-hand screw rule moving from the first vector to the second

direction cosines.
Figure 8-25 Vectors r y k and n needed to determine nx and ny the
442 STEADY-STATE THERMAL AND FLUID FLOW AN A LI SIS

through the smaller angle. If this is applied to the situation shown in Fig. 8-25, we
may write

r„ x k
n (8-206}
K x k|

where k is the unit sector in the z direction. The cross product of two vectors a
and b is most easily evaluated from

I °T A,

\_b, b, b.j
= — a .by) i + ( ajy x nj}:) j + (aj)' — ay bj) k

where a t ay , and a. are the x, v. and z components of a, etc. When Eqs. (8-205)
,

and (8-206) are compared after the cross products are evaluated with the help of
Eq. (8-207), the result is

(8-208a)

and

M, - (8-208b)

where l
tJ
is the length of leg ij and is given by

l„ = Vlt, - t,)= + (y, - >,)' (8-2091

Similar results may be obtained if leg jk or ki is on the global boundary

Example 8-22
Determine the direction cosines n, and «, on leg tj for the element shown in Fig.

8-26

Figure 8*26 Triangular element in Example 8-22

 TWO-DIMENSIONAL POTENTIAL FLOW 443

Solution

Using the nodal coordinates shown in Fig. 8-26, we get

nx =
8-5
===== — = - = 06 3

V (3 - 7)
2
+ (8 - 5)
2
5

Note that the sum of the squares of the direction cosines is unity as it should be.
gj

On occasion itproves to be convenient to combine ff# and ff# into one term
by noting that the normal velocity V„ may be written in terms of u and v as

V„ = mi, + v/i, (8-210)

Equations (8-197) may be combined and ffo may be defined as

ft® = f;;« + ?;*=- N r u„ dc (8-21 u

\cr
It also follows that

<3(j>

V„ = (8-212)
dn

where it represents the coordinate in the direction of the outward normal.

The Element Resultants

This completes the velocity potential formulation of two-dimensional, steady po-

tential flow problems. The assemblage of each element stiffness matrix and nodal

force vector is routine. Note that imposed velocities are not considered to be

geometric boundary' conditions; instead, nodal velocities are imposed via Eqs.
(8-202) and (8-203) for the triangular element and by Eqs. (8-197) in general. The
geometric boundary' conditions are the prescribed velocity' potentials. These may
be applied with either Method or 2 from Sec. 3-2. The resulting system of linear
1

equations solved either by the matnx inversion method or by the active

may be
zone equation solver. The average velocities u and f for a typical triangular element
e could then be determined from

u = ac = ~w->A - w->A - (8-213a)

dx <3.1

V =
dy
= - — By
a*- = -w 3 A - m 3 A-w 3 3<K
(8-213b)

where the my ’s are computed from Eqs. (6-21) and d>,, 4*,, and 4l. are the now
known nodal values of the potential. These average velocities are generally asso-
444 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

dated with the centroid of the triangular element. The method of quadrilateral
averages could be used to improve these results.

Application: Flow around a Long Cylinder

Let us consider a long cylinder positioned transversely in a flow field between two

flat plates as shown in Fig. 8-27(a). Because the cylinder is assumed to be long,
end effects may be neglected; hence, the flow field is two-dimensional. Only the
region outside of the boundary layer is considered so that viscous effects may be
neglected. Let us also assume the fluid to be incompressible. Clearly, the potential
flow formulation applies and the resulting streamlines and potential lines are shown
in Fig. 8-27(b).
Because of the two-axis symmetry, only one-fourth of the region needs to be
modeled as shown in Fig. 8-28 Note that the inlet velocity U enters the finite

(a)

Figure 8-27 Application of two-dimensional, incompressible potential flow (a)

Long
the
cylinder placed transversely in a uniform flow field between two flat plates and (b)
perpendicular
associated flow field around the cylinder Lines of constant potential are alw ay s
to the streamlines
 TWO-DIMENSIONAL POTENTIAL FLOW 445

element formulation via f,‘

B (or ffB ) for those elements with a leg along side a-e.
Also note that iflB and (or fvB ) are identically zero along sides a-b and d-e. Since
the flow cannot penetrate the solid cylinder, V„ must be zero along the cylinder be
and hence is zero here. Finally, along side c-d, an arbitrary' value of d> (a
constant) must be chosen. This last boundary' condition is not explicitly included
in the finite element formulation and is imposed in the usual manner after the
assemblage equations are formed.

Stream Function Formulation

The continuity equation given by Eq. (8-185) is satisfied exactly if the stream

function 4» is defined by

(8-214a)

and

(8-214b)

If these expressions for the velocity components are substituted in the irrotational

flow condition given by Eq. (8-186). the result is the two-dimensional form of

Laplace’s equation or

8 2 4i 3
2
iJi

+ (8-215)
2 2
8x dy

the double
Figure 8-28 Only one-quarter of the flow field needs to be modeled because of
potential formulation.
symmetry'- The boundary conditions shown apply to the velocity
446 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

It should be noted that fines of constant iff are reafly streamlines, and lines of
constant potential must always be orthogonal or perpendicular to these
tf> lines.
Using the Galerkin method on an element basis gives

18-216)

where the shape function matrix N is dependent on the type of element used.
Following the approach taken in the velocity potential formulation, the reader should
be able to show that Eq. (8-216) can be written

K'a' = fe (8-217)

where

K' = K{x + KJ, (8-218)

and

f' = Kb + Kb (8-219)

The element stiffness matrices are now defined by

K", =
f

-M'
——
dN 7 dN
dr dr
dx tty (8-220a)

f dN 7 dN
(8-220b)

and the element nodal force vectors by

Kb = N r nn, (1C (8-221 a)

fc
or

Kb = - N7 v/r_, dC (8-22 1b)

jcr
The evaluation of these element stiffness matrices and nodal force vectors for
typical two-dimensional problems is left as an exercise For the triangular element,
the direction cosines can be evaluated with the help of Eqs (8-208)

Application: Flow around a Long Cylinder

Let us reconsider the example from the velocity potential formulation a long
cylinder placed transversely in a flow field between two flat plates as shown in

velocity
Fig. 8-27(a). Figure 8-27(b) the streamlines and lines of constant
shows
potential. The streamlines actually represent lines of constant stream function 4*-
Because of the two-axis symmetry, only one-fourth of the flow field needs to be
modeled as shown in Fig. 8-29 The line a-b-c represents one streamline and hence
4* is a constant on this line. Let us denote this constant value of ifr as tj/,. The
line
d-e represents a different streamline. Let us denote the value of the stream function
along d-e as 4L- Only one of these values is arbitrary, with the other value to be
determined from physical considerations as shown below. Let us assume that the
inlet velocity U on face ci-e is constant. The velocity U can be related to ij/ by Eq.
(8-2 14a) or

3v|i
U = (8-222a )
3)’

Since U is constant, we may write

4*2 ~ 4h
v = (8-222b)
3’, - 3’,.

If t|/, is taken arbitrarily to be zero, then t|/

2 is given as

4*2 = (3’, - yJV (8-222c)

where ye and ya denote the y coordinates of points e and a respectively. Since ,

there is no tangential velocity along faces a-e and c-d, it follows that

f = 0 (8-223a)

on these faces. Furthermore, since ny = 0 on these faces, we also have

ffi,
= 0 (8-223b)

The conditions expressed by Eqs. (8-223) are equivalent to the condition that the

streamlines are normal to the boundary. Hence

constant
potential

double
Figure 8-29 Only onc-quartcr of the flow field needs to be modeled because of the
shown apply to the stream function formulation.
symmetry'. The boundary conditions
448 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

chjl
0 (8-224}
dn

where n is the coordinate in the direction normal to the boundary. Clearly we have
natural boundary conditions along faces a-e and c-d.

8-12 CONVECTION DOMINATED FLOWS

The governing equations for heat conduction that we have seen in previous sections
did not allow for any type of fluid motion within the region being analyzed. In
general, fluid motion is very effective in transporting energy, usually much more
effective than conduction alone. The purpose of this section is to provide an intro-
duction on how the finite element method may be used to model such problems.
Although a particular case is described and the FEM formulation given, the same
basic approach may be applied to other similar problems In this section, a particular
velocity profile will be assumed for the purpose of illustrating how the temperature
distribution may be obtained
Let us consider the case of forced flow of a Newtonian, constant property,
incompressible fluid between two flat plates as shown in Fig. 8-30. It is assumed
that the flow is laminar and hydrodynamically fully developed, so that a parabolic
velocity profile results In this case the thermal energy equation alone needs to be
solved for the temperature distribution. Various boundary conditions may be im-
posed as shown in Fig. 8-30 The governing equation to such a problem is given
by

Imposed heat flux

Figu re 8-30 Thermal entrance region of a hydrodynamically, fully developed laminar flow

of a Newtonian fluid between two long flat plates

 CONVECTION DOMINATED FLOWS 449

sr ffr
pcit(y)— = + (8-225a)
ox dy V
where T is the temperature of the fluid at any point (x,y), » is the .r component of
velocity, p is the fluid density, c is the specific heat (at constant pressure), and k
isthe thermal conductivity. Because of the hydrodynamically, fully developed flow
assumption, the y component of velocity is assumed to be zero and, thus, is not
present in Eq. (8-225a). The velocity u is given by

»(y) = Vm (8-225b)

where 77 is average velocity of the fluid and H is half of the plate separation distance
as shown in Fig. 8-30.

In this formulation, the properties are assumed to be constant. However, we

would proceed in much the same manner if c and k were functions of temperature.
The solution (and the problem), however, would be nonlinear and the direct iteration
method from Sec. 8-2 could be used to obtain the solution for the nodal temperatures.
The density p may not be temperature-dependent because this would not result in
a parabolic velocity profile. In fact, in this case the continuity, Navier-Stokes, and
energy equations would have to be solved simultaneously. Because the formulation
is two-dimensional, Eq. (8-225a) assumes that the two plates are infinitely long
perpendicular to the paper.
It is interesting to note that, except for liquid metals, axial conduction may be
neglected for Peclet numbers greater than 100 [7]. For liquid metals, axial con-
duction may be neglected for Peclet numbers greater than 160. The Peclet number
is denoted as Pe and is defined to be the ratio of energy transport by convection
to energy transport by conduction. If U represents the average velocity in the .v

direction, Dh is the hydraulic diameter, and a is the thermal diffusivity, the Peclet

number is defined by

p e = UHh. (8-226)
a
where the thermal diffusivity is given by

k_
a (8-227)
pc

The hydraulic diameter Dh > s defined by

4 x flow area
Dh = : (8-228)
wetted perimeter

Note that the Peclet number is equal to the product of the Reynolds and Prandtl
numbers, or

Pe = Re Pr (8-229)

where the Reynolds number is defined by

450 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

and the Prandtl number by

where v is the kinematic viscosity, related to the absolute viscosity p. by

In the case of two large flat plates separated by a distance 2 H, the Peclet number
is given by

The Galerkin method will be applied to Eq (8-225a) in the usual manner.

Interestingly it is not even possible to obtain the classical variational principle that
corresponds to Eq. (8-225a). It will be seen that the Galerkin method in this case

yields an unsymmetnc stiffness matrix. The implications of this will be discussed

Try to make an educated guess as to which term
briefly later. in Eq (8-225a) yields

the unsymmetnc stiffness matnx

On an element basis, the Galerkin method requires that we wnte

Nr - =
i-
h£ - £(*£)
0 l8
- 2341

The thickness t is introduced so that each term m Eq (8-234) has dimensions of

energy per unit time.We can take t to be unity for convenience If the Green-Gauss
theorem is applied to the two terms involving second-order denvatives, and if the
boundary integrals are combined by using Eq (8-48), we get

f N r pc» — t dx dy + k' T
f N q n t dC + \ let-— dx dy
J* dx dx dt

dN T dT
+ f
-M'
[
I —kt—dxdy
——kt— dx dy
, , ,
= 0 18-235)
dv
dy dy

in the direction toward

If the heat fluxes q sB on the global boundary are imposed
the fluid, then we have
8 -2361
i. = '

Furthermore, if we represent the temperature T on an element basis and wnte

8 - 237 >
T = Na' <

and note that the vector a' is independent of x and y, Eq (8-235) becomes
 CONVECTION DOMINATED FLOWS
45J

where the composite element stiffness matrix K c

is given by

K = K.« +
‘
K;? + Kue (8-239)

and the element nodal force vector by

V-lVwdC (8-240)

The element stiffness matrices from conduction are identical to the expressions for
Kfr and from Sec. 8-8 [see Eqs. (8-106)]. The element
Kf,, stiffness matrixfrom
the convective transport is given by

Kf, = f
JA '
N rpcw? —dx
dx dy (8-241)

Note that Ku e
is unsymmetric, unlike Kf and Kf
r The element stiffness matrices
t

and nodal force vector may be evaluated for particular two-dimensional elements,
such as the triangular and rectangular elements. This is fairly routine and the details
are omitted here.
The fact that one of the element stiffness matrices is unsymmetric deserves
The assemblage of the element characteristics is done in the usual
special attention.
manner, except that the upper and lower parts of the assemblage stiffness matrix
need to be stored. Recall from Sec. 6-8 that subroutine UACTCL [8] from Appendix
C may be used in this case to obtain the solution for the nodal temperatures (after

the prescribed temperatures have been imposed). In this subroutine, the upper
triangular coefficients are stored in the array A whereas
,
the lower triangular coef-
ficients are stored in the array C. Recall that unity is stored for the diagonal entries
in the C array (the actual diagonal entries are stored in the A array).

Because the assemblage matrix is unsymmetric, the nodal temperatures that

are calculated may or may not be correct. When the results are not correct, it is

usually quite obvious — for example, the temperatures may be oscillatory with

respect to the spacial coordinates. This oscillatory behavior should be expected if

the elements are larger than a certain threshold size. More specifically, the threshold

of oscillatory behavior is given in terms of a particular Peclet number referred to

as the cell Peclet number defined as

Pe (
= —a (8-242)

where L is a characteristic element length. Oscillatory behavior generally occurs

when the cell Peclet number is greater than 2 [9] for one-dimensional models.
Therefore, by reducing the size of the elements (and increasing the number of
elements), we can generally improve the results after the threshold size is reached.
Gresho and Lee [10] contend that the use of higher-order elements (see Chapter 9)
will substantially improve the results. This contention has been verified further by
Srivastiva [11] who did follow-up research on the study by Kundu and Stasa [12].
The Galerkin method is mathematically equivalent to a central finite difference.

Convection-dominated flows generally require a backward (or upwind) difference

452 STEADY-STATE THERMAL AND FLUID FLOW ANALYS IS

to aid the convergence. In the finite element method, upwinding schemes also exist.
References 13 and 14 should be consulted for more information on modeling con-
vection-dominated flow problems. In these references, the use of upwinding schemes
is also discussed. Further discussion of these topics is beyond the intended scope
of this text.

8-13 INCOMPRESSIBLE VISCOUS FLOW

In this section, it will be shown how the Finite element method may be used to
solve for the velocity distribution in convection-dominated flows. Let us consider
the laminar flow of a viscous fluid for the case of moderate Reynolds numbers such
that the convective terms in the Navier-Stokes equations are significant. Typical
examples would include external boundary-layer flows and internal flows. For
example, we may want to determine the two-dimensional velocity field for a fluid
moving through a duct with an obstruction in it as shown in Fig 8-31. The flow
field at the entrance of the duct may be assumed to be either fully developed or
developing Let us indicate how the finite element characteristics may be derived
for such a complicated problem by starting with the two-dimensional form of the
continuity and Navier-Stokes equations for a constant-property, incompressible,
Newtonian fluid.

The continuity equation for an incompressible fluid is given by

3k
(8-243)
dx

where « and x and y components of the fluid velocity at any point

v are the (x,y)

The Navier-Stokes equations in the x and y directions are given by

2 2
3k dp ( U 3 «\
pl u — + + + Fx (8-244a)
dx dy) ~to ^aT 2
dy V

Figure 8-3T Viscous flow through an obstructed duct.

 INCOMPRESSIBLE VISCOUS FLOW 453

dp
P + V + P- (8-244b)
dy

where p is the fluid density, p is the pressure, p. is the absolute viscosity, and F,
and F%
are the body forces per unit volume in the x and v directions, respectively.
In Eqs. (8-244) the convective acceleration terms [on the left-hand side of Eqs.

(8-244)] make the problem to be solved nonlinear. The region to be analyzed may
be discretized in the usual manner with triangular or rectangular elements. The goal
is to determine the nodal values of the pressure and the velocity components, since
the formulation is to be based on the so-called primitive variables.
If the Galerkin method is applied to Eq. (8-243). we get

f N r (— + —\tdxdy = 0 (8-245)
J'V \dx dv)

where the thickness t is introduced so that each term in Eq. (8-245) has dimensions
of volume per unit time (i.e., a volumetric flow rate). For convenience, t can be
taken to be unity'. Let us assume

u = Naf, (8-246a)

and

V = Naf (8-246b)

where af, and af are given by

r
= ["/ "j «J (8-247a)

and
1
= [v 4 v, '’J
(8-247b)

for the triangular element. In Eqs. (8-247), it,, itj , and if* are the .v components of
and Since af and af are independent of ,v and y,
the velocity at nodes /, j, k, etc.

Eq. (8-245) can be written in the form

Kfaf + Kfaf = 0 (8-248)

where

KS
f

JA’
Nr
^
— tdx dv
0N
d.\
(8-249)

and

Kf
r

JA'
Nr
_
—
UN
dv
r dx dv (8-250)

Let us now apply the Galerkin method to Eq. (8-244a) by writing

f
N
T ( du
+
3n\ dp ( dht
-
T ^ "b
d 2u
“3 - F, t d\ dv = 0 (8-251)
i It;; %) s “i Ax ' d\-
4S6 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

3
A|| A'|f A,' Ail Aj§ Ail Aii All K\l r„,“| [7,1
All A?| A? 3 A?i A?? Afl A?:! Ail Ail v, ff
Aii A'?? Ail Kn Kf? K]2 A?; All Ail p, f?
Kit ~Klf Ki\ "^22 ^22 ^22 Ag A]} Kg u, fT
Kl\ K2 i
K\] K\\ KM KM A|| A|| Afi v, = f}
^33 A 33
4!
*1?
4 4
ATI? “22
^32
tf2l
If 32
a 22

If 22
a 2’

^32
All
All
A|I
Aj| A|I “ p,
/T

ki\ aj? aj? a 32 a 32 aif32

23
^33 KM K\l Vk /3
2

3
A, ! A]? ah ^32 ^32 ^32 ^33 ^33 KM_ _pk_
(8-262)

The assemblage of Eq (8-262) may now be done in the usual manner such that

the vector of nodal unknowns is given by

a = «2 v2 u„ v„
T (8-263)
[», v, pi i p2 i ! p„]

where again n nodes are assumed m the discretization of the region being analyzed
Note that this type of assemblage was done routinely in all stress analyses in Chapter
7 and results in a significantly smaller bandwidth than would result otherwise.
Prescribed velocities and pressures would be applied by using either of the two
methods from Sec 3-2. Since the assemblage system equations are unsymmetric,
subroutine UACTCL [8] from Sec 6-8 (see Appendix C) may be used to obtain
the solution for the nodal velocities and pressures.
The above formulation is based on the so-called primitive variables the ve-

locities and pressure Alternate formulations exist that utilize a stream function. As
in the analysis of two-dimensional potential flow problems, the stream function is
defined such that the continuity equation is satisfied exactly The three governing
equations reduce to two partial differential equations, and the order of the highest
derivative present increases from two to three. The interested reader should consult
References 15 and 16 for more details on the stream function formulation, as well
as another approach referred to as the vorticity formulation.
It has been assumed above that the same shape functions may be used for the

pressure and velocity functions This seems to give good results for rather low fluid
velocities For high fluid velocities, the elements (and hence shape functions) used
for the velocity function shouldbe one order higher than those used for pressure.
Several such higher-order elements are presented in Chapter 9. Again the reader is
referred to References 15 and 16 for further details..

8-14 DEVELOPMENT OF A TWO-DIMENSIONAL THERMAL

ANALYSIS PROGRAM

Some helpful hints and comments are given in this section so that the reader can
develop a two-dimensional thermal analysis program with further instructions from
 THERMAL ANALYSIS PROGRAM PROGRAM HEAT

the instructor.* Specific comments are made with respect to the main program,
mesh generation, data storage, variable properties, and point heat sources.

The Main Program

The main program should be kept as brief as possible. An example is the following
seven lines of code:

PROGRAM HEAT
COMMON XL (5500)
MAX = 5500
LCONSL = 3
CALL DRIVER (MAX, LCONSL)
CALL EXIT
END

The variable MAX represents the length of XL array, the contents of which are
described below. This variable should be assigned the value of the dimension of
XL in the unlabeled COMMON. Much larger problems may be solved with the
program by increasing this parameter to the memory limit of the computer being
used. The variable LCONSL (as in the TRUSS program) represents the logical unit
number for the console. Control is then transfered to subroutine DRIVER.

Mesh Generation

The same type of node generator discussed in Sec. 3-6 may be used in this program
except that the nodes no longer need to be equally spaced. This may be accomplished
with the help of two spacing factors ft and/, (FX and FY in the program). These
factors are used as described below. This same approach is used in the stress analysis

program described in Sec. 7-7. The reader who is familiar with the mesh generator
in the stress analysisprogram may wish to skip this section.
Consider the starting node NI and the final node NF shown in Figure 8-32(a).
Additional nodes can be generated between these two end nodes if a nonzero value
of NG is used. The nodes need not be equally spaced, however. Without loss of
generality, let us number the nodes 1,2,. n, as shown in Figure 8-32(b), and
. . ,

refer to the x and y coordinate of node 1 by x , y t etc. Then the spacing factors
t ,

fx and f} may be defined by

*A two-dimensional thermal analysis program and a user s manual are included in the Instructor/
Solution's manual for this text
458 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

(b>

Figure 8-32 (a) Starting node Nl and ending node A'F are used to generate additional
nodes (b) Line along which n nodes are generated (not necessarily equally spaced)

*1 =
f* = x ~ = =
£? £2 . . . -it! £a.-'- (8-264)
2 -T] Xj — X2 X„- 1
— X„-2

and
-
y? = >«
{ = = . . . .
16-265}
’
ft ~ ft ft - ft ft-i - ft-!
Concentrating on the x direction for now, we may write

x2 ~ x, = x2 ~ x, (8-266a)

x3 - x2 = fx(x2 - r,) (8-266h)

2
ft - ft “ /.(ft - ft) = /, (ft - ft) IS-266c)

ft ^ ft-i = /,(ft- 1
- ft— 2 ) - J? 2
(ft - ft)
(S-26Ed)

Adding these results gives

ft - ft = n + f, + ff + - + - ft)
|8 ’ 267)

from which it follows that

ft = ft + - 74' - I 8 2681

1 + Xn
1=1

Since a:, and x„ may be input (as XI and XF, respectively), and since fx may be
input (as FX), x 2 can be found from Eq. (8-268) Note that n is the total number
 THERMAL ANALYSIS PROGRAM 459

of nodes in the generation sequence (including NI and NF). Obviously x3 can be

computed from Eq. (8-266b) or

,t 3 = x2 + /(.r2 - x,) (8-269)

and so forth. It should also be obvious that in the y direction, the same results hold
except that each x, is replaced by y, and/ is replaced by /. Note that if/ (or/,)
r
is greater than unity, then the nodes are spaced farther apart in moving from NI to
NF; if fx (or/) is less than unity (but greater than zero), then the spacing between
two consecutive nodes decreases in moving from NI to NF.

Data Storage

The data for the nodal coordinates, elements, boundary' condition flags, material
boundary condition parameters, and so forth should be stored in the XL
properties,
assemblage stiffness matrix (in column vector form), the
array. In addition, the
assemblage nodal force vector, and the diagonal pointer array (JDIAG from Sec.
6-8) should also be stored in the XL array. The partitions between each of these
sections should float and if fewer nodes are used, for example, more materials may
be present. The total number of storage locations required must not exceed MAX.
If it does, an error message should be displayed on the console and execution should
be terminated.
The numbering of the nodes program does not renumber the
is critical if the
nodes to minimize the bandwidth of the assemblage stiffness matrix. However, the

program should store only the banded portion of this matrix, which reduces the
storage requirements drastically over that of storing the full matrix. Nodes on the
object being analyzed are alwaysnumbered consecutively, from 1 to the maximum
number of nodes. The bandwidth is minimized when the maximum difference
between any two nodes on each element is minimized. Figure 8-33 shows that this
is accomplished quite simply' by numbering the nodes in the direction of fewer

nodes.
Note that in Fig. 8-33(a) the nodes are numbered in the direction of the smaller
number of nodes. In Fig. 8-33(b) the nodes are numbered to the right and in the
direction of the greater number of nodes. The storage requirements for the mesh
higher than for the mesh in the former. In Fig. 8-33(c) the nodes
in the latter are

are numbered in an alternating sweeping fashion which more than doubles the
storage requirements over that required in Fig. 8-33(a). Finally, in Fig. 8-33(d) the
element on the lower left has nodes 1 , 2, and 21; the implication is that the stiffness

matrix is no lonser banded which means higher storage requirements and

in turn

increased execution times. Although the program need not store the zero terms in
the stiffness matrix outside the bandwidth, it should store every'thing within the
bandwidth.
Numbering the elements is not critical but some sort of regular numbering
scheme usually allows the use of the automatic element generation feature. Note
460 STCADY-STATE THERMAL AND FLUID FLOW ANALYSIS

Figure 8-33 (a) Proper node numbering scheme, (b) less desirable node numbering scheme;
(c)nodes should never be numbered in an alternating sweeping fashion, and (d) worst node
numbering scheme (results in an unbanded stiffness matrix)

that in forming the elements shown in the lower right comer of Figure 8-33(a), the
quadrilateral formed by nodes 17, 20, 21. and 18 is divided into two triangles by
using the shorter diagonal (node 17 to node 21) as opposed to the longer diagonal
(node 18 to node 20). This is illustrated in Fig 8-34 Regular triangles generally
give better results than obtuse or needle-shaped triangles.

Variable Property Routine: SUBROUTINE VPROP

It is relatively easy to provide the capability to handle both spactally dependent

and temperature-dependent properties For example, the thermal conductivity, beat
may be functions of either the global coordinates x and
transfer coefficients, etc.

y or the temperature T If any of the properties is a function of temperature, the

problem becomes nonlinear and the direct iteration method from Sec 8-2 is used
to obtain the solution of the nodal temperatures
The method that may be used to accomplish this is now described. If any of
the material or boundary condition properties is a function of x, y, and/or T, then
the user should simply enter a unique negatne integer from —
— 1 to 10 instead
of an actual value for the property. In the discussion below, these unique negative
integers are referred to as the variable-property indicators. For these properties,
the program should be designed to call SUBROUTINE VPROP, an abridged listing
of which is given below:

SUBROUTINE VPROP (PROP)

 )

THERMAL ANALYSIS PROGRAM 46J

Figure 8-34 (a) Preferred way of dividing a quadrilateral into two triangles and (b) less
desirable way of forming two triangles.

COMMON /VPHELP/ X, Y, T
COMMON /CONST / PI, SIGMA, C(30)
IPROP = I ABS PROP (

GOTO (1, 5, 3, 4, 5, fa, 7, fi, B, 10), IPROP

1 PROP = definition of first variable property (may be a function of
X, Y, and/or T)
RETURN
E PROP = definition of second variable property
RETURN

10 PROP = definition of tenth variable property

RETURN
END

The reader may want to examine SUBROUTINE PROPTY in Problem 4-78

in order to see how this may be implemented. The basic idea is that if a negative

value is read for a property, the program (SUBROUTINE PROPTY in particular)

should call SUBROUTINE VPROP when that property is needed and transfers
control to the statementwhose label is equal to the absolute value of the variable-
property indicator.Note that a computed GO TO statement is used. Also note that
the user may define up to 30 constants [C(l), C(2), C(30)] that may be used
. . . ,

in the SUBROUTINE VPROP. These constants should be read by the program in

the first input section. For example, let us say that the thermal conductivity is

known to vary quadratically with temperature as

k = 32 + 10r - 0.5 T2

The user simply needs to decide which variable-property indicator he or she would
64 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

was assumed to be given, and the main purpose of the analysis

velocity profile
i compute the temperature distribution in the flow field. The Galerkin method
to
;seen to provide a relatively straightforward analysis, whereas the variational
hod simply could not be used (why not?). An unsymmetric stiffness matrix was
n to arise for the first time, because of the convective energy term in the governing
'erential equation. It was mentioned that the numerical solution may be Sl-
aved unless relatively small elements (or higher-order elements) are used.
The finite element formulation was provided for the velocity and pressure
:nbution in a viscous fluid flowing in a two-dimensional region under steady-
e conditions. The formulation was based on the Galerkin method. At first glance,
assemblage step appeared to be different from what we have seen so far.

wever, a closer examination revealed that the same type of assemblage procedure
5 used in two- and three-dimensional stress analyses in Chapter 7. Again un-
imetrical stiffness matrices arose, and again it may be necessary to use many
ttively small elements (or higher-order elements) to obtain convergence. The
ivective acceleration terms in the Navier-Stokes equation make the formulation
llinear. The direct iteration method from Sec. 8-2 may be used to obtain the
Jtion for the nodal velocities and pressures Only the formulation with the so-
led primitive variables development Alternate for-
was provided in the formal
lations such as the stream function and vorticity formulations exist, but were
presented in detail here
The two-dimensional, steady-state thermal analysis program was described.

; program is capable of handling variable properties and allows graded meshes

Although substructuring was introduced in Chapter 7 with an application to
ictural and/or stress analysis, it may also be applied to nonstructural problems,

hould be recalled that substructuring makes the use of large finite element models
te practical. In fact, significant problems may be solved on microcomputers if

•Structuring is used. The reader should make an attempt to adapt the material in
7-6 to the analysis of two-dimensional, steady-state heat conduction with the
ite element method

IFERENCES

Arpaci, V S , Conduction Heal Transfer, Addison-Wesley, Reading, Mass , 1966,

p 32
Huebner, K H , The Finite Element Method for Engineers Wiley,
.
New York, 1975,

pp 218-219
Wilson, E L., “Structural Analysis of Axisymmetric Solids,” AIAA Journal, vo! 3 ,
no pp 2267-2274, December 1965
12,
Roberson, J. A., and C. T Crowe, Engineering Fluid Mechanics, 2nd ed , Houghton
Mifflin, New Jersey, 1980, pp 437-449
Shames, I. H Mechanics of Fluids, McGraw-Hill, New York, 1962, pp 173-185.
,
 REFERENCES 465

6. Currie, I. G.. Fundamental Mechanics of Fluids, McGraw-Hill, New York 1974 pp

65-142.
7. Rohsenow, W. M., and J. P. Hartnett, eds.. Handbook of Heat Transfer McGraw-
Hill, New York, 1973.
8. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp. 746-747.
9. Leonard, B. P., “A Survey of Finite Differences of Opinion on Numerical Muddling
of the Incomprehensible Defective Confusion Equation,” Proc. ASME, AMD-vol. 34,
pp. 1-18, New York, December 2-7, 1979.
10. Gresho, P., and R. L. Lee, "Don’t Suppress the Wiggles —They’re Telling You Some-
thing,” Proc. ASME, AMD-vol. 34, pp. 37-62, December 2-7, 1979.
11. Srivastava, A. K., “A Finite Element Method to Solve the Graetz Problem Using
Subparametric Elements,” M.S. Thesis, Florida Institute of Technology, Melbourne,
Fla., July 1982.

12. Kundu, D., and F. L. Stasa, “Finite Element Solution to the Graetz Problem and Its

Extensions,” International Congress on Technology and Technology Exchange, Pitts-

burgh, May 3-6, 1982.

13. Hughes, T. J. R., ed.. Finite Elements Methods for Convection Dominated Flows, Proc.
of the ASME, AMD-vol. 34, December 2-7 1979. ,

14. Chung, T. J., Finite Element Analysis in Fluid Dynamics, McGraw-Hill, New York,
1975.
15. Baker, A. J., Finite Element Computational Fluid Mechanics, McGraw-Hill, New York,
1983, pp. 233-305.
16. Huebner, K. H., The Finite Element Method for Engineers, Wiley, New York, 1975,
pp. 346-361.

PROBLEMS

Note: The properties in Appendix A shouid be used unless indicated otherwise.

8-1 Use the direct iteration method to obtain a solution to the following nonlinear

algebraic equation:

x3 — 3.x
2
+ 6x — 4 = 0

Stop the solution process when three significant digits of accuracy are obtained.

Use an initial guess of

a. x = 0 b. x = 10 c. x = -5

8-2 Use the direct iteration method to obtain a solution to the following nonlinear
algebraic equation:

xi — 5x 3 + 10x 2 + 4x - 24 = 0

Terminate the solution process when three significant digits of accuracy are obtained.

Use an initial guess of

a. x = 0 b. x = 10 c. c = - 1
466 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

8-3 Use the direct iteration method to obtain a solution to the following nonlinear
transcendental equation*

fir cos irt = 1

Use an initial guess of x = 0 and stop the iterations when an accuracy of three
significant digits is obtained It may be verified by direct substitution that x = Vs
is also a solution of this equation. What happens if the direct iteration method is

used for initial guesses of x = 0 33 and x = 0 34? What can be concluded from
this?

8-4 Use the direct iteration method to obtain a solution to the fallowing system of
algebraic equations*

x 2 + xy 2 + 3r - 15 = 0
3x + xy - 7 = 0

Assume an initial guess of x = 3 and >• = 3, and stop the iterations when an
accuracy of three significant digits is obtained

8-5 Use the direct iteration method to obtain a solution to the following system of
algebraic equations

y* + 5x + x1 = 25

x2\ = 10

Assume an initial guess of x = I and y =2. and terminate the iterations when an
accuracy of three significant digits is obtained.

8-6 With the help of the infinitesimal element of length dx shown in Fig P8-6, derive
Eq (8-3) Proceed by performing an energy balance on the infinitesimal element
Assume that the heat flow from conduction in the x direction varies according to a
first-order Taylor expansion as shown in the figure. Use Fourier's law of heat
conduction (a constitutive relationship) to eliminate the heat flux q. Clearly state

all assumptions

hPiT-Tt )dx t
(0P[T* -T*\dx

^QAdx

Figure P8-6
 :

PROBLEMS 467

8-7 Show that the variational principle that corresponds to Eq. (8-3) is given
by
L
c f
/ -rJo
QAT - VihPT1 + hPTJ - VseaPT 5
-
2

+ euPT?T - ZikA dx
\dx
providing that only geometric or natural boundary conditions are allowed. Note that
the one-dimensional body is assumed to have length L .
f

8-8 Determine the more general form of the variational principle in Problem 8-7 for the
case of convection, radiation, and imposed heat fluxes on the boundaries as shown
in Fig. 8-3.

8-9 Show that the element stiffness or conduction matrix, given by Eq. (8-20b), evaluates
to

hPL 2 1
K
6 1 2

if the lineal element from Sec. 6-3 is used. Note that L is the element length. Clearly
state all assumptions made in arriving at this result.

8-1 0 Show that the element stiffness or conduction matrix from boundary radiation, given
by Eq. (8-20e), evaluates to

8 , 071 73 0
K rB
e , ,

0 ZjOAjT]

Use the lineal element from Sec. 6-3. Explain how this result should be used.

8-11 Show that the element nodal force vector from lateral radiation, given by Eq.
(8-21b), evaluates to

_ toPLTj W
2 [l

where L is the element length. Assume the lineal element from Sec. 6-3. Clearly

state all assumptions made in arriving at this result.

8-12 Show that the element nodal force vector from the internal heat source, given by
Eq. (8-21c), evaluates to

« ~
- m 2
1

where L is the element length. Assume the lineal element from Sec. 6-3. Clearly

state all assumptions made in arriving at this result. Does the result seem to be

intuitively correct? Please explain.

8-13 Show that the element nodal force vector from boundary radiation, given by Eq.
(8-2 le), evaluates to

e,A,oT*]
468 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

Assume the lineal element from Sec. 6-3. Clearly explain how this result should
be used

8-14 Show that the element nodal force vector from the imposed boundary heat fluxes,
given by Eq (8-2 If), evaluates to

Assume the lineal element from Sec. 6-3. Clearly explain how this result should
be used Note that the heat fluxes q, and qt are positive if directed toward the body.

8-15 Show that Eq (8-31) holds if the element stiffness matrix from lateral convection
[see Eq (8-20b)] from a one-dimensional body is evaluated under the assumption
that the perimeter P varies linearly from node i to node j over the element Assume
the lineal element from Sec 6-3 and state all assumptions.

8-16 Consider the circular pm fin shown in Fig P8-16. The fin is made of aluminum
and has a length Lf and diameter D The base is held a fixed temperature Tb and
the tip undergoes convection The lateral and boundary heat transfer coefficients
are equal and denoted as h The entire fin is exposed to a fluid at temperature Ta
(including the tip) Discretize the fin into two equal-length elements (i e , with three
nodes) If h = 2500 W/m -°C, Lf = 5 cm, D = t cm, Tb = 75°C, and Ta =
2

15°C, neglect radiation and determine

Figure P8-16

a. The temperatures at the nodal points

b. The heat removal rate from the fin (using the integration method from Sec
4-10)
c. The fin efficiency (see Sec 4-10)

8-17 Consider the circular pin fin shown in Fig P8-16 The fin is made of aluminum
and has a length Lf and diameter D The base is held a fixed temperature Tb and
the tip undergoes convection.The and boundary heat transfer coefficients
lateral
are equal and denoted as h The entire fin is exposed to a fluid at temperature Ta
(including the tip). Discretize the fin into two equal-length elements (i.e., with three

nodes). If A = 100 Btu/hr-ft 2 -°F, Lf = 3 in D = 0.5 m Tb = 150°F, and Ta

, ,

= 8
50 F, neglect radiation and determine,
a. The temperatures at the nodal points
 PROBLEMS 469

b. The heat removal rate from the fin (using the integration method from Sec
4-10)
c. The fin efficiency (see Sec. 4-10)

8-18 Consider the square pin fin of length Lf shown in Fig. P8-18. The fin is made of
copper and has a cross-sectional area A. The base of the fin is held at fixed tem-
perature Tb and the tip undergoes convection. The lateral and boundary heat transfer
coefficients are equal and denoted as h. The entire fin is exposed to a fluid at
temperature Ta (including the tip). Discretize the fin into two equal-length elements
(i.e., with three nodes). If h = 5000 Btu/hr-ft 2 -°F, Lf = 1 in., A = 1.5 in.
2
, Tb
— 185°F, and Ta = 70°F, neglect radiation and determine:

Figure P8-18

a. The temperature at each nodal point

b. The heat removal rate from the fin (see Sec. 4-10)
c. The fin efficiency (see Sec. 4-10)

8-19 Consider the square pin fin of length Lf shown in Fig. P8-18. The fin is made of
copper and has a cross-sectional area A The base of the fin is held a fixed temperature
.

Tb and the tip undergoes convection. The lateral and boundary heat transfer coef-
ficients are equal and denoted as h. The entire fin is exposed to a fluid at temperature

Ta (including the tip). Discretize the fin into two equal-length elements (t.e., with
three nodes). If h — 10,000Lf W/m 2-°C, = 10 cm, A — 9 cm 2 Tb — ,
100°C, and
Ta = 30°C, neglect radiation and determine:
a. The temperature at each nodal point
b. The heat removal rate from the fin (see Sec. 4-10)
c. The fin efficiency (see Sec. 4-10)

8-20 Repeat part (a) of Problem 8-16 if a heat flux of 5 W/cm2 is imposed on the tip of

the fin (toward the fin), and

a. There is no convection from the tip of the fin.

b. There is simultaneous convection from the tip of the fin.

8-21 Repeat part (a) of Problem 8-17 if a heat flux of 500 Btu/hr-ft
2
is imposed on the

tip of the fin (toward the fm), and

a. There is no convection from the tip of the fin.

b. There is simultaneous convection from the tip of the fm

 k
470 STEADY-STATt THERMAL AND FLUID FLOW ANALYSIS

8-22 -7000 2
Repeat part (a) of Problem 8-18 if a heat flux of Btu/hr-ft is imposed on
the tip of the fin (away from the fin), and
a. There is no convection from the tip of the fin
b. There is simultaneous convection from the tip of the fin.

8-23 Repeat part (a) of Problem 8-19 if a heat flux of - 1 0 W/cm 2 is imposed on the
tip of the fin (away from the fin), and
a. There is no convection from the tip of the fin.
b. There is simultaneous convection from the tip of the fin.

8-24 Repeat part (a) of Problem 8-16 if a uniform heat source with a strength of 10
W/cm J exists

8-25 Repeat part (a) of Problem 8-17 if a uniform heat source with a strength of 120
3
Btu/hr-in exists

8-26 Repeat part (a) of Problem 8-18 if a uniform heat source with a strength of 780
3
Btu/hr-in exists

8-27 Repeat part (a) of Problem 8-19 if a uniform heat source with a strength of 0.2
W/cm 3
exists

8-28 Reconsider the onc-dimensional heat conduction model developed in Sec 8-3
Assume that a lateral heat flux is imposed (along the length of the one-dimcn-
sional body) such that a fraction a is absorbed on the surface of the body. Let us
denote the projected area that receives the heat flux as r A
If the heat flux happens

to be from the sun. then the parameter a is referred to as the solar absorptivity
a. Derive the integral expression for the corresponding nodal force vector Use the
Galcrkin method Note that Ap is then interpreted to be the projected area of the
element
b. Evaluate the result for part (a) if the lineal element from Sec 6-3 is used Assume
that a and q lM arc constant over the element

8-29 Consider the annular fin shown in Fig P8-29 Using the general nomenclature from
Sec 8-3 ( is the thermal conductivity, h is the convective heat transfer coefficient,

Figure P8-29
 PROBLEMS 471

etc.), derive the expressions for the element characteristics, in particular, determine
the expressions for the element stiffness matrices from conduction fin the radial
direction), lateral convection, and boundary convection. In addition, determine the
expressions for the element characteristics for the nodal force vectors from lateral
convection, boundary convection, and boundary heat flux. Do not perform the
integrations. Describe the type of element that would be appropriate here (a simple
drawing showing the nodes would suffice). Hint: The governing equation is given
in Problem 4-87 (the boundary conditions, however, are different in this problem).

8-30 Using the annular-shaped element shown in Fig. P8-30. evaluate the integrals that
result in Problem 8-29. Instead of performing exact integrations, evaluate the in-
tegrands at the element centroid and treat the integrands as though they were constant.
Explain how the results from the formulation in Sec. 8-3 could be used directly
without reformulating the problem as required in Problem 8-29. Him: What is the
effective cross-sectional area for heat conduction? What is the effective perimeter?

Figure P8-30

8-31 The governing equation for one-dimensional heat conduction in a sphere with a
volumetric heat source is given by

_1
2
£ 4 -

<? = o
r dr

where T is the temperature, r is the radial coordinate, k is the thermal conductivity,

and Q is the strength of the heat source. What arc the units of Q? Derive the
expressions for the element characteristics for the case of a hollow sphere undergoing
convection with heat transfer coefficients h, and /i„ on the inside and outside surfaces
of the sphere, respectively, to fluids at temperatures T, and T„. Him: When setting

up the integral for the Galcrkin method, integrate the product of N r and the residual
2
with respect to the elemental volume 4-trr dr.

8-32 Explain how the one-dimensional heat conduction model developed in Sec. 8-3

could be used to solve for the temperature in the hollow sphere described in Problem
8-31. Be specific. Him: What is the effective cross-sectional area for heat conduc-
tion? Is there any lateral convection?

8-33 Extend the program developed in Problem 4-78 (or the program furnished by the

instructor) so that it can handle convection and imposed heat fluxes on the boundary
of an arbitrary one-dimensional body as well as a uniform heat source. Boundary
condition flags should be used to indicate the type of boundary conditions that exist
on any given node (generally on the ends of the body). Positive integer flags should
flags
be used to denote prescribed temperature conditions, whereas negative integer
472 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

should be used to denote convection and/or imposed heat fluxes More specifically,
three different input sections should be used in this regard as given below:

BOUNDARY CONDITION FLAG DATA

N ±IBC
(blank line)
CONVECTION AND OR HEAT FLUX DATA
IBC QS H TA
(blank line)
PRESCRIBED TEMPERATURE DATA
IBC TEMP
(blank line)

where N is the node number that has a positive IBC flag if a prescribed temperature
is tobe imposed, and a negative IBC flag if convection exists and/or a heat flux is
to be imposed IBC is zero (by default) if neither condition exists at the node in
question For each positive IBC used, a corresponding prescribed temperature (TEMP)
must be declared in the PRESCRIBED TEMPERATURE DATA input section For
each negative IBC used, a corresponding set of imposed heat fluxes, convective
heat transfer coefficients, and ambient temperatures must be present (QS, H, and
TA, respectively) If only convection is present on a boundary, then the user simply
sets QS to zero in the corresponding input line If only an imposed heat flux exists
on a boundary, then the user simply sets H and TA to zero in a similar fashion
Note that each of the input sections is terminated with a blank line The program
should check if N is zero (1 e , if the line is blank) If N is zero, the next input
section should be read, otherwise, another set of data should be read in the same
input section Let QVOL represent the heat source strength (per unit volume) and
include it in the material property data

The variable property routine (SUBROUTINE VPROP) should be implemented

as described in Problem 4 78 This approach will allow the same program to be
used to solve a wide variety of one-dimenstonal, steady heat conduction problems

8-34 Extend the program in Problem 8-33 (or one furnished by the instructor) to allow

for lateral and boundary radiation as formulated in Sec 8-3 Extend the material

data to include the emissivity (ELAT) and receiver temperature (TRLAT) for the

lateral radiation boundary radiation by using negative boundary condition

Allow for
flags, and modify the next to the last input section (see Problem 8-33) to the
following’

IMPOSED HEAT FLUX, CONVECTION, AND/OR RADIATION DATA

IBC QS H TA E TR
(blank line)

where E and TR are the boundary emissivity and receiver temperatures, respectively
If radiation is not present on a boundary, then the user simply sets E and TR to
zero in the corresponding input line Similar comments hold for the case of ELAT
and TRLAT.
 PROBLEMS 473

8-35 Solve Problems 8-16. 8-20. and 8-24 with the computer program from Problem
8-33 (or with the program furnished by the instructor). For Problem 8-16, use two.
four, and eight elements to show the increased accuracy that results as the number
of elements (and nodes) is increased. For Problems 8-20 and 8-24, use only eieht
elements.

8-36 Solve Problems 8-17, 8-21, and 8-25 with the computer program from Problem
8-33 (or with the program furnished by the instructor). For Problem 8-17, use tw'o,
four, and eight elements to show the increased accuracy that results as the number
of elements (and nodes) is increased. For Problems 8-21 and 8-25. use only eight
elements.

8-37 Solve Problems 8-18. 8-22, and 8-26 with the computer program from Problem
8-33 (or with the program furnished by the instructor). For Problem 8-18. use two.
four, and eight elements to show the increased accuracy that results as the number
of elements (and nodes) is increased. For Problems 8-22 and 8-26. use only eight
elements.

8-38 Solve Problems 8-19. 8-23. and 8-27 with the computer program from Problem
8-33 (or with the program furnished by the instructor). For Problem 8-19, use two,
four, and eight elements to show the increased accuracy that results as the number
of elements (and nodes) is increased. For Problems 8-23 and 8-27. use only eight
elements.

8-39 Use the two-dimensional form of the Green-Gauss theorem [Eqs. (8-42)] to rewrite
below as a sum of two other integrals: one integral around the boundary
the integrals
of the two-dimensional region and the other integral over the area of the region.

f a < df\
a.
U pH
ax
fa— dx dy
j

ox}
’
v

fa.( fa—
hT
b. pH
ay V 3yJ
\
dx dy

8-40 Use the three-dimensional form of the Green-Gauss theorem [Eqs. (8-41 )] to rewrite
the integrals below as a sum of two other integrals: one integral over the surface
of the three-dimensional region and the other integral over the volume of the region.

8-41 unit thickness)

By performing an energy balance on the infinitesimal area element (of
conduction (a constitute e
shown in Fig. P8-41 . and by invoking Fourier’s law of heat
and ?,), deme q.
relationship) to eliminate the heat fluxes from conduction iq,
thickness.
(8-43). Note that a heat source Q is present. Assume a uniform
474 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

9**~sz +

8-42 Extend Eq (8-48) to the three-dimensional case first by inspection and then by
formally extending the development in Sec 8-5

8-43 B> performing an energy balance on the infinitesimal element of thickness t shown
in Fig P8-43. and by invoking Fourier's law of heat conduction (a constitutive

V 4 — **

Figure P8-43

relationship) to eliminate the heat fluxesfrom conduction (<?, and «?,), derive Eq
as a laterally
(8-91) Note that lateral convection and radiation are included, as well
imposed heat flux. In addition, a heat source Q is also included
 4
475

illustrated in (8 ' 9
ExamptT-ToS ° by
tlie natural boundary condition)
^
' he boundar
USI "g 'he
approach

b “" d "» <- ****** or a DrP c TT y

»
,s insulated (, e
specified ol Ite

illustrated in 10 « 9 D by using
Examp^S "o^ ‘a/i (
the approach
radiation, and imposed heat
Z'foStP ° SS,b,1,t V
- of boundary
fluxes convection,

and e,ement s!,ffness matrices

(8- 106b)]. g.ven by Eqs (8-106a)

t
r° rn
a^two-d^mensiona^b^y
given by Eqs.
(8- 106c) and
^^ ™5 ^y^^ v
I

)
'T*'
6 ' '
C ° nvect,on and

£ element s[,ffness
matrices
Nation from
(8-106d)J

bound “> convection and

rad, at, on
matrices given by
Eqs. (8-106e) and
’ * ’ ** e,emem S “ ffneSS
(8- ioho'r™'"'

dimensional
Cl “^ »' «» assumptions 2 C ° n duC tl0n «»

f t h h

1
' he

“ nVeC,lon
-v direction ,n a two-

,f »“ “ *>
“

body'^veTby
“s=d Cent,,
state assnmp’l.o’ 2 m
fTb” I”"'"
“

2, SnSTbuf ta'nnTr tom hm“

J E<? (8 ~ '
6) lf the three-node triangular element
Clearly
C3r ^state alf
State a!l assumptions
is used
made ,n amving at this result

1 n ° daI ^ VeC ‘°r

Sim^tnalhodv glVe " by E<E (8 '" 7) f ,he threey mposed heat flux on 3
node ™ngula, element
fr° m
‘
3 ’ ateraI '
-
'

used. Clearh sS? Jf

[
is
y tate all assumptions made in amving at this result

dimMstnalbbodv"!;
6
" -

8 Ven by Eq
'
^ VeC '° r from an internaI heat source
(8_I l8) ,f the three - n ode triangular
,n a tw °-
element
used and if the h is
neat source is uniformly distributed over the element

,CmCnt n ° dal force vector from bound ai-y

convection from a two-
dimes', nn^h ?,
y 15 8,Ven by Eq (8 - ,2I) ,f ,be tbree-node triangular element
-

used and if th ,s
6 C0nvecll0n occurs
on leg ij What is the form of this result if legs
and kl
1lk 3nd l-i i
ha PP er> to be on 8
the global boundary?

element nodal force vector from boundary

radiation from a two-
y $ 8 Ven by Eq (8 ' l22) ,f the th me-node triangular element
'

ised andlf the J

radlation occurs on leg ij What is the form
,s

md ki happen of this result if legs jk

Inn to be on the
8
global boundary?

° W tbat be
element nodal force vector from an imposed heat
flux on the boundary
p ( wo- |
imensional body is given by Eq (8-123) if the
three-node triangular
476 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

element is used and if the heat flux is imposed on leg y. What is the form of this
result if legs ]k and kt happen to be on the global boundary?

8-57 The expressions for the element stiffness matrices from conduction in a two-
dimensional body are given by Eqs. (8-106a) and (8-106b) Since these results hold
for any two-dimensional element, let us evaluate these stiffness matrices for the
four-node rectangular element presented in Sec. 6-4 by proceeding as follows -

a. With the help of Eqs (6-32), show that

dN _ ldN
dx a dr

dN _ I 3N
dy b ds

where r and s are the serendipity coordinates.

b. Using the results from part (a), rewrite the integrals in terms of derivatives of
the shape functions with respect to the serendipity coordinates r and s Do not
forget to change the limits on the integrations Also note that dx dy = ab dr ds
c. Evaluate the resulting integrands at the element centroid (i e 0 and, at r =*

s = 0) Then treat the integrands as though they are constants and pull them
through the integral Evaluate the remaining trivial integrals Show that the result

for the element stiffness from conduction in the x direction is given by

1 1 -1 -f
1 1-1-1
-1-1 1 1

-1 -1 1 1_

State the assumptions made in arriving at this result

1

d. Derive the corresponding result for What assumptions are made *

8-58 Evaluate the element stiffness matrix from convection from a two-dimensional
lateral

body ]i e , Eq (8- 106c)] if the four-node rectangular element from Sec 6-4 is used
Evaluate the integrals by first evaluating the integrands at the element centroid (l e ,
at r = 0 and s = 0) and then treating the integrands as though they were constants

8-59 Evaluate the element stiffness matrix from boundary convection from a two-
dimensional body (i e , Eq, (8-106e)J if the four-node rectangular element from
Sec. 6 4 is used and if face j happens to be on the global boundary Evaluate the
integrals by first evaluating the integrands at the centroid of face ij (i e , at r = +1
and s = 0) and then treating the integrands as though they were constants

8-GO Evaluate the element stiffness matrix from boundary convection from a two-di-
mensional body [i e Eq (8-1 06e)] if the four-node rectangular element from Sec
,

is used and if face jk. happens to be on the global boundary Evaluate t he

6-4 integrals

by first evaluating the integrands at the centroid of face jk (i.e at r = 0 and ,

s ~ +- 1) and then treating the integrands as though they were constants

8-61 Evaluate the element nodal force vector from lateral convection from a two-dimen-
sional body [i e„ Eq (8- 107a)] if the four-node rectangular element from Sec.
6-4 is used Evaluate the integrals by first evaluating the integrands at the element
 PROBLEMS 477

centroid (i.c., at r = 0 and 5 = 0) and then treating the integrands as though they
were constants.

8-62 Evaluate the element nodal force vector from lateral radiation from a two-dimen-
sional body [i.c., Eq. (8-107b)] if the four-node rectangular element from Sec.
6-4 is used. Evaluate the integrals by first evaluating the integrands at the element
centroid (i.e., at r = 0 and 5 = 0) and then treating the integrands as though they
were constants.

8-63 Evaluate the clement nodal force vector from a laterally imposed heat flux on a
two-dimensional body [i.e., Eq. (8-107c)] if the four-node rectangular element from
Sec. 6-4 is used. Evaluate the integrals by first evaluating the integrands at the
element centroid (i.e., at r = 0 and s = 0) and then treating the integrands as
though they were constants.

8-64 Evaluate the element nodal force vector from an internal heat source in a two-
dimensional bod)' (i.e., Eq. (8-107d)] if the four-node rectangular element from
Sec. 6-4 is used. Evaluate the integrals by first evaluating the integrands at the
element centroid (i.e., at r = 0 and 5 = 0) and then treating the integrands as
though they were constants.

8-65 Evaluate the element nodal force vector from boundary convection from a two-
dimensional body [i.e., Eq. (8-107e)] if the four-node rectangular element from
Sec. 6-4 is used and if face ij happens to be on the global boundary. Evaluate the
integrals by first evaluating the integrands at the centroid of face ij (i.e.. at r = +1
and 5 = 0) and then treating the integrands as though they were constants.

8-66 Evaluate the clement nodal force vector from boundary convection from a two-
dimensional body [i.e., Eq. (8-107e)} if the four-node rectangular element from
Sec. 6-4 is used and if face jk happens to be on the global boundary. Evaluate the
integrals by first evaluating the integrands at the centroid of fac ejk (i.e., at r = 0
and 5 = +1) and then treating the integrands as though they were constants.

8-67 Evaluate the element nodal force vector from boundary radiation from a two-
dimensional body [i.c., Eq. (8-107f)] if the four-node rectangular element from

Sec. 6-4 is used and if face km happens to be on the global boundary. Evaluate the
integrals by first evaluating the integrands at the centroid of face km (i.e. , at r = — 1

and 5 = 0) and then treating the integrands as though they were constants.

8-68 Evaluate the element nodal force vector from a boundary heat flux imposed on a
two-dimensional body [i.e., Eq. (8- 107g)j if the four-node rectangular element from
Sec. 6-4 is used and if be on the global boundary. Evaluate the
face mi happens to

integrals by first evaluating the integrands at the centroid of face mi (i.e., at r

= 0
and 5 = — 1) and then treating the integrands as though they were constants.

8-69 The element resultants (i.e., the heat fluxes from conduction in the .randy directions)
are given by Eqs. (8-125) for the three-node triangular element.
Derive the corre-
rectangular element
sponding expressions for the average heat fluxes if the four-node
the average
is used. Assume that the heat fluxes at the element centroid represent
correct.
heat fluxes in the element. Explain these results seem to be intuitively
why
P8-70. The plate from which
the
8-70 Consider the triangular element shown in Fig.
of 0.5 in. The nodal
element is extracted is made of cast iron and has a thickness
= = xk = 0.6, and y* - 1-8 in.
coordinates are x, = 2.0, y, 1.5,.r
;
= 1.7, y) 3.0,
478 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

a. Determine the element stiffness matrices from conduction in the x and y direc-
tions
b. Determine the element stiffness matrix from lateral convection if the element
convects from both faces to a fluid at 35°F through a convective heat transfer
2
coefficient of 25 Btu/hr ft -°F
c. Determine the element stiffness matrix from boundary convection if leg y happens
to beon the part of the global boundary that undergoes convection to a fluid at

40°F through a convective heat transfer coefficient of 50 Btu/hr-fl 2-°F

8-71 For the element in Problem 8-70, determine the nodal force vectors
a. From lateral convection
b. From a laterally imposed heat flux of 200 Btu/hr-ft 2 on each face of the plate
c. From an internal heat source of 350 Btu/hr ft 3
d. From boundary convection
e. From a boundary heat flux of 425 Btu/hr-ft 1 imposed on leg ij (which is on the
part of the global boundary also undergoing convection)

8-72 Consider the triangular element shown in Fig P8-72 The plate from which the
element is extracted is made of brass and has a thickness of 1 cm The nodal

coordinates are x, = 5, y, = 6, x, = 4, y, = 4, xk = 6, and yk = 4 cm

Figure P8-72
 PROBLEMS 479

a. Determine the element stiffness matrices from conduction in the .v and y direc-
tions.

b. Determine the element stiffness matrix from lateral convection if the element
convects from both faces to a fluid at 20°C through a convective heat transfer
coefficient of 125 \V/cirf-°C
c. Determine the element stiffness matrix from boundary convection if leg ki hap-
pens to be on the part of the global boundary that undergoes convection to a
fluid at 25°C through a convective heat transfer coefficient of 250 W/cm 2 -°C.

8*73 For the element in Problem 8-72, determine the nodal force vectors
a. From lateral convection
b. From a laterally imposed heat flux of 300 W/cm 2 on each face of the plate
c. From an internal heat source of 160 W/cm 3
d. From boundary convection
e. From a boundary heat flux of 235 W/cm 2 imposed on leg ki (which is on the
part of the global boundary also undergoing convection)

8-74 Consider the triangular clement shown in Fig. P8-74. The element is extracted from
a thin plate of thickness 0.5 cm. The material is hot rolled, low carbon steel. The
nodal coordinates are x, = 0,y, = 0,.v
y
= O.y, = -l,:c t = 2,andyt = —1 cm.

Figure P8-74

a. Determine the element stiffness matrices from conduction in the x and y direc-

tions.

b. Determine the element stiffness matrix from lateral convection if the element

convects from both faces to a fluid at 40°C through convective heat transfer

coefficients of 250 and 300 W/cm 2 -°C.

c. Determine the element stiffness matrix from boundary convection if leg jk hap-
pens to be on the part of the global boundary that undergoes convection to a

fluid at 45°C through a convective heat transfer coefficient of 35 W/cm z-°C.

8-75 For the element in Problem 8-74, determine the nodal force vectors
a. From lateral convection
2
b. From a laterally imposed heat flux of 225 W/cm on one face of the plate

c. From an internal heat source of 10 W/cm 3

d. From boundary convection

480 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

e. W
From a boundary beat flux of 35 Jem- imposed on leg ji (u hich is on the part
of the global boundary also undergoing connection)

8-76 Consider the triangular element shown in Rg. P8'76. The element is extracted from
a thin plate of thickness 0.75 in The material is pure copper. The nodal coordinates
arex, = 0,v, ~ 0, Xj — l,y = 2 ,xt~ —I, andji = 2 in.
f

a. Determine the element stiffness matrices from conduction m the x and .>
direc-

tions
b. Determine the element stiffness matrix from lateral convection if the element
convects from both faces to a fluid at 50°F through a convective heat transfer
2
coefficient of 135 Btu.'hr-ft -°F.
c. Determine the element stiffness matrix from boundary convection if leg ki hap-
pens to be on the part of the global boundary that undergoes convection to a
2
fluid at 40°F through a convective heat transfer coefficient of 50 Btu/hr-ft -°F

8-77 For the element in Problem 8-76, determine the nodal force vectors
a. From lateral convection
b. From a laterally imposed heat flux of 150 Btu/hr-ft2 on each face of the plate
c. From an internal heat source of 215 Btu/hr-ft 3
d. From boundary convection
e. From a boundary heat flux of 450 Btu/hr ft' imposed on 1eg h (which is on the
part of the global boundary also undergoing convection)

8-78 Consider the rectangular element shown in Rg. P8-78. The element is extracted
from an aluminum plate with a thickness of 1.25 cm. The coordinates of the nodes

Rgure P8-78
 PROBLEMS 481

are - 4, v, = 2, .y, = 4, v, = 3, x k = 2,
yt = 3, xm = 2, and ym = 2 cm.
.r,

Perform the necessary integrations by following the approaches mentioned in Prob-

lems 8-57 to 8-60, and
a. Determine the element stiffness matrices from conduction in the x and y direc-
tions.

b. Determine the element stiffness matrix from lateral convection if the element
convects from both faces to a fluid at 15°C through convective heat transfer
coefficients of 10 and 12 W/cm 2-°C.
c. Determine the element stiffness matrix from boundary convection if face jk
happens to be on the part of the global boundary that undergoes convection to
a fluid at 20’C through a convective heat transfer coefficient of 35 W/cm 2-°C.
8-79 Consider the element in Problem 8-78. Perform the necessary integrations by fol-
lowing the approaches indicated in Problems 8-61 to 8-68, and determine the nodal
force vectors
a. From lateral convection
b. From a laterally imposed heat flux of 125 W/crrr on each face of the plate
c. From an internal heat source of 15 W/cm
3

d. From boundary convection

2
e. From a boundary heat flux of 350 \V/cm imposed on face jk (which is on the
part of the global boundary' also undergoing convection)

8-80 Consider the rectangular clement shown in Fig. P8-80. The element is extracted
from a brass plate with a thickness of 0.375 in. The coordinates of the nodes are
x, = 5,y, = l.Xj — 5,\j = 4. ,v ( = 2,y1 = 4,.trt = 2, andy„, = 2 in. Perform
the necessary integrations by following the approaches mentioned in Problems
8-57 to 8-60, and

Figure P8-80

a. Determine the element stiffness matrices from conduction in the x and y direc-

tions.
lateral convection if the element
b. Determine the element stiffness matrix from
heat transfer
both faces to a fluid at 60°F through a convective
convects from
2
coefficient of 235 Btu/hr-ft -°F.
convection if face km
c Determine the element stiffness matrix from boundary
undergoes convection to
happens to be on the part of the global boundary that
-
coefficient of 150 Btu/hr-ft
a fluid at 60°F through a convective heat transfer
°F.
482 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

8*81 Consider the element in Problem 8*80 Perform the necessary integrations by fol-
lowing the approaches indicated in Problems 8-61 to 8-68, and determine the nodal

force vectors
a. From lateral convection
b. From laterally imposed heat fluxes of 350 and 400 Btu/hr-ft 2 on each face of
the plate
3
c. From an internal heat source of 100 Btu/hr-ft
d. From boundary convection
e. From a boundary heat flux of 340 Btu/hr-ft 2 imposed on face km (which is on
the part of (he global boundary also undergoing convection)

8-82 For the element in Problem 8-70, determine the element nodal force vector from a
point heat source with a strength of 100 Btu/hr-m (of thickness) if the source is

located at *o = IS and > 0 = 2.0 in

8-83 For the element in Problem 8-72, determine the element nodal force vector from a
point heat source with a strength of 120 W/cm (of thickness) if the source is located
at x0 = 5 and >0 = 5 cm
8-84 For the element in Problem 8-74, determine the element nodal force vector from a
point heat source with a strength of 150 W/cm (of thickness) if the source is located
a tx0 = 2 and j0 = - cm I

8-85 For the element in Problem 8-76, determine the element nodal force vector from a
point heat source with a strength of 230 Btu/hr in (of thickness) if the source is

located at x0 = 1 and y 0 = 2 in

8-86 For the element in Problem 8-78, determine the element nodal force vector from a
point heat source with a strength of 35 W/cm (of thickness) if the source is located
at *0 = 3 2 and >0 = 2 7 cm
8-87 For the element in Problem 8-78, determine the element nodal force vector from a

point heat source with a strength of 45 W/cm (of thickness) if the source is located

at Je0 = 4 and }0 - 2 cm
8-88 For the element in Problem 8-80, determine the element nodal force vector from a
point heat source with a strength of 50 Btu/hr-in (of thickness) if the source is

located at .r
0 = 4 and >0 = 3 in

8-89 Foe the element in Problem 8-80 determine the element nodal force lector from a
,

point heat source with a strength of 60 Btu/hr-m (of thickness) if the source is

located at x0 = 2, and y0 = 4 in

8-90 Consider the mesh shown in Fig. P8-90 The elements are defined in terms of the
global node numbers as indicated in the figure The 3x3 composite element
stiffness matrix for element 1 , for example, may be denoted symbolically as

r*iy ttV A®
K<» = A® A®
LK& A® A®.
the
Note that the superscript (m parentheses) denotes the element number and
subscripts are determined by the global node numbers associated with the element
 PROBLEMS 483

Figure P8-90

(the order ts significant) With the help of this symbolic notation, give the expression
for the assemblage stiffness matrix for the discretized two-dimensional region in

Fig P8-90 What is the half-bandwidth 9 Can the half-bandwidth be reduced 9 If so,
explain how
8-91 Consider the mesh shown in Fig P8-90 The elements are defined in terms of the
global node numbers as indicated in the figure The 3 x 1 composite element nodal
force vector for element 1 , for example, may be denoted symbolically as

,r
7i
fin —
/f"
_/w_

Note that the superscript (in parentheses) denotes the element number and the
subscripts are determined by the global node numbers associated with the element
(the order is significant) With the help of this symbolic notation, give the expression
for the assemblage nodal force vector for the discretized two-dimensional region in

Fig P8-90.

8-92 Consider the mesh shown in Fig P8-92 The elements are defined in terms of the
global node numbers as indicated in the figure Using the symbolic notation from
Problem 8-90, give the expression for the assemblage stiffness matrix for the dis-

cretized two-dimensional region in Fig P8-92 What is the half-bandwidth 9 Can

the half-bandwidth be reduced 9 If so, explain how

Figure P8-92

8-93 Using the symbolic notation from Problem 8-91, give the expression for the assem-

blage nodal force vector for the discretized two-dimensional region in Fig. P8-92
 .

4S4 STEADY-STATE THERMAL AND FLITD ROtt WALTER

Us elements are defined in terras of the global code cumbers as indicated in the
figure.

8-94 Cooider the mesh shown in Fig. PS-9-t. The elements are defined in terras of the
global cole cumbers as indicated in the figure. Using the symbolic cctadeo from
Problem S-90. give the expression for the assemblage stiffness matrix for the dis-

cretized two-dimensional region in Fig PS-94 What b the half-bardwihfa? Can

1
the half-handwihh he reduced If so. explain how

Figure P8-94

8-95 Uxzng the symbolic eotsnea from Problem S-91 . give fix exprexMOa for tie assem-
blage codal force vector tor the discretized two-dunessioftal region in Fig PS-94.
The elements are defined in terms of the global code cum Sees as indicated uj the
figure

8-9 S Coes tier the me-h down m Fig PS-96 The elements are defined m terms of the
global node numbers as radicated us the figure B\ extending the symbolic rxcmoa
from Freblem S-*X> to the ca-e of the rectangular elecxcl. give the expression for
the assemblage Oittroo- rnatnx for the ducretized twxMiimecMOcal region in Fig.
PS-<>6 What r> the half-hmdw >d:h* Can the half-bandwidth he reduced? If so.
explain bow

© © ®
Ev'-e-* NoJh
S'"
'A'K
1
2
3«2i
*

5
/

6 4 3

© © ©
Figure 1*8-86

8-97 By extending the symbolic notation from Problem S-91. give the expression for the
assemblage ncvial force vector for the <h-eretized tw’o-durKnsicnal region in Fig-

PS-96. The elements are defined in terras of the global code rurrbers as indicated

in the figure.
 PROBLEMS 485

8-98 Consider the mesh shown in Fig. P8-98. The elements are defined in terms of the
global node numbers as indicated in the figure. By extending the symbolic notation
from Problem 8-90 to the case of the rectangular element, give the expression
for
the assemblage stiffness matrix for the discretized two-dimensional region
in Fig.
PS-98. What is the half-bandwidth? Can the half-bandwidth be reduced? If
so,
explain how.

Element Nodes
number / / k m
1 5 4 3 2
2 6 5 2 1

Figure P8-98

8-99 By extending the symbolic notation from Problem 8-91 give the expression for the
.

assemblage nodal force vector for the discretized two-dimensional region in Fig.
PS-98. The elements are defined in terms of the global node numbers as indicated
in the figure.

8-100 Consider the mesh shown in Fig. PS-100. The elements are defined in terms of the
global node numbers as indicated in the figure. By extending the symbolic notation

from Problem 8-90 to the case of the rectangular element, give the expression for

the assemblage stiffness matrix for the discretized two-dimensional region in Fig.
P8-100. Note the two different types of elements. What is the half-bandwidth? Can
the half-bandwidth be reduced? If so. explain how.

Element Nodes
number 1 / * m
i 3 4 2 i

2 5 6 4 3
3 6 5 7

extending the symbolic notation from Problem 8-91 , give the expression
8-101 for the
By
the discretized two-dimensional region in Fig.
assemblage nodal force vector for

P8-I00. The elements are defined in terms of the global node

numbers as indicated

in tire figure. Note the use of the two different types

of elements.
486 STEADY -STATE THERMAL AND FLUID FLOW AN ALYStS

8-102 For the element in Problem 8-70 the nodal temperatures are obtained from the
solution of Ka = f and are T, = 175, Tj = 168. and 7» = I73 F. Determine the
C

average heat flukes from conduction in the t andy directions, respecti\ ely. At what
point in the element are these heat fluxes normally associated?

8-103 For the element in Problem 8-72 the nodal temperatures are obtained from the
solution of Ka = f and are T, — 75, Tt — 82, and Tk = 73'C. Determine the
average heat fluxes from conduction in the x and y directions, respectiv ely. At h hat
point in the element are these heat fluxes normally associated?

8-104 For the clement in Problem 8-74 the nodal temperatures are obtained from the
solution of Ka = f and are T, = 94, Tf — 91, and Tt = 96°C. Determine the
a\ erase heat fluxes from conduction in the x andv directions, respectively. At what
point in the element are these heat fluxes normally associated *

8*105 For the element in Problem 8-76 the nodal temperatures are obtained from the
solution of Ka = f and are T, = 154. Tt — 148. and Tt = 157’F Determine the
average heat fluxes from conduction in the x and \ directions, respectively. At what
point in the element are these heat fluxes normally associated 1

8-106 For the element m Problem 8-78 the nodal temperatures are obtained from the
solution of Ka = f and are T, = 92. T = 96. Tt - 89. and Tm - 9SCC. Determine
}
the heat fluxes at the element centroid from conduction and y directions,
in the x
respectively' These heat fluxes may be regarded as the average conduction heat
fluxes.

8-107 For the element in Problem 8-80 the nodal temperatures are obtained from the
T„ = 196 C
C
solution of Ka = f and are T, = 192. T, = 195. Tt = 189, and
Determine the heat fluxes element centroid from conduction in the x and y
at the

directions, respectiv ely These heat fluxes may be regarded as the av entge conduction
heat fluxes

8-108 The necessary condition for the existence of an extremum of the functional

/ = dx dv

is that the first variation 8 must vanish (i e . equal to zero) provided that

B BF\ _ I d BF \1
{ I *
<Lr \«hrn / 2dy\dM„/J
r dF d Ibf \ _ i d (bf v dC = 0
|_chv, - Bx \Bw„) J

and

„
8k.
'
+ —
1aF
il
2 8w„ v
„
8kvx
1 BF
2 8m„
— _ \
/
_ 0
+ - ' nx 8m, dC — n
I

on the boundary of the two-dimensional region. In the above, w is the field variable,
x and y are the two independent coordinates, m x indicates the first (partial) derivative
indicates the second (partial) denv ativ e of m w ith
respect
of w w ith respect to x.
to x, etc.
 FS0BLEM5 487

2. Show that the corresponding Euler-Lagrange equation is given by

SF (6F \ / EF S :
cF
5 _ \ _ (
\

cv- ex \ (5-.’.y / <n v to- _

/ Ar : '
to _ /

-rk-(^)
ay cv.y
c_t \ /
+ i(i
\5w
ay-
), 0
n ]
fa. What order differentia! equation does the Euler-Lagrange equation from part fa)
correspond to? Please explain.
C. What is the highest order derivative present in the functional F?
d. Try to generalize the results from parts (b) and fc). Explain why the variational
formulation may be referred to as the weak forrr.idanon.

8-109 In Problem 8-108. identify:

a. The geometric boundary conditions
fa. The natural bcur.darv cor.drtions

8-110 The bending of a thin isotropic plate of constant thickness r may be described by
the well-known biharmontc equation

6~v; c'w d'K 12(1 - p3 )

J ~
ef.t.y) = 0
0X dr- by- by Et l

where »• is the deflection of the neutral plane of the plate (in the direction perpen-
dicular to the xy plane), q is lateral distributed load per unit area. p. is Poisson's
ratio,and E is the modulus of elasticity Assume that the
plate is rigidly supported
around the boundary such that the plate has zero deflections and zero slopes on the
boundary. More that the rotations 6. and P about the x and y axes at any point in
the plate are related to the deflection at the same point by-

TO*
ex and e..
lx

With the help of the Euler-Lagrange equation from Problem S-10S. show that the

corresponding variational principle under these assumptions is given by

12(1 - y?)
I = dx dy
Ei s

8-111 Derive the variational principle for Problem 8-1 10 by luting

cV 12(1 -
-
ot.r.y) 5w dx dx = 0
tar ex- ex- Et-

and trv* applvinc the Green-Gauss theorem twice. Hit... The boundary integrals tha.

arise are identicallv zero for the assumptions stated in Problem 8-110. Clearly exp: 2 in
why.

8-112 Explain wbat meant by the method of quadrilateral averages. Why is the method
is

used? IHustrate how the method is applied with a numerical example. Assume
that

elements which the element resultants (i.e.. the

we have two adjacent triangular for

conduction heat fluxes in the .r and v directions) are already

known from the ( finite

element solution).
 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

8-113 How could the method of quadrilateral averages be used in two-dimensional stress
analysis (i e., plane stress or plane strain) to improve the resulting stresses when
the three-node triangular element is used? Be specific and illustrate with a numerical
example.

8-114 How could the method of quadrilateral averages be used in axisymmetnc stress

analysis to improve the resulting stresses when the three-node triangular element is

used? Be specific and illustrate with a numerical example.

8-1 15 In the formal development of two-dimensional, steady-state heat conduction in Sec

8-8, only isotropic materials were considered Let us now extend the development
of that section to the case of anisotropic materials For anisotropic materials, the
thermal conductivity is dependent on direction It is convenient to consider the two
values of thermal conductivity in the so-called principal directions as shown in Fig
P8-II5 Let us denote the principal values of the thermal conductivities as k' and
Id m the local x and y directions, respectively Let us denote these local directions
as x' and y' (similar to the development in Chapter 3 where the transformation
matrices for two and three-dimensional trusses were developed). Note that 6 is the
angle between the x and x' axes (and the y and y' axes) as shown in Fig. P8-1 15.
In this case the governing equation for two-dimensional , steady-state heat conduction
must be modified to reflect the fact that there are different values of thermal con-
ductivity in the x' and v' directions Also, it proves to be much more convenient
to express the governing equation in the local coordinate system Only the two terms
involving conduction m Eq (8-91). i e , the first two terms, must be modified with
the resulting governing equation given by

+
£(«'§)
- w - - *) + * - e -
‘ 0

In general the local x'y' axes are not in the same directions as the global xy axes
as shown in Fig P8 1 15 Therefore, it seems reasonable to apply the Galerkin
method on an element basis where the local x'y' coordinate system is used

Figure P8-115

a. Show that the resulting expressions for the element stiffness matrices from
conduction in the x' and y’ directions are given by

3N'
* . , ,
«•
,
 PROBLEMS 489

and

5N' r ., 5N’
Kf d>
-L'*17
where N” indicates that the shape functions must be written in terms of the local
x'y' system before taking the indicated derivatives,
b. Convince yourself that since the field variable is a scalar in this problem (i.e.,
the temperature) that the procedure for implementing anisotropic material prop-
erties in the finite element formulation is as follows: (1) Transform the coor-
dinates of even- node for a given element from the global ay system to the x'y'
system with the help of Eq. (3-13), (2) compute the element stiffness matrices
from the above expressions for the element stiffnesses (or conductances) using
the coordinates in the local x'y' system, and (3) do the remaining part of the
analysis as described in Sec. 8-8. Note that it is not necessary to transform the
local element stiffness matrices to global element stiffness matrices (such as in
Chapter 3) because the field variable here is a scalar, not a vector. In effect,

the local and global element stiffness matrices are identical in this case.

8-116 Let us apply the development of Problem 8-115 to a specific example. Consider
the case of anisotropic heat conduction in a triangular element. The principal values
of the thermal conductivities are denoted as k'x and k[. in the local .r' and v' directions,

respectively. The angle 6 is defined to be the angle between the .t and x' axes (or
they and v' axes) as shown in Fig. P8-115. The plate from which the element is
extracted has a thickness of 0.5 in. The nodal coordinates are.q = 2.0, y, = 1.5,
Xj = 1.7, \j = 3.0, xt = 0.6, and v, = 1.8 in. with respect to the global xy

coordinate system. If k[ and kx are 120 and 60 Btu/hr-ft-°F, respectively, and

. if 9
is 30°, determine the global element stiffness matrices for this element (from con-
duction only).

8-117 Repeat Problem 8-1 16 for an element whose nodal coordinates are .r, = 5, y,- = 6,
Xj = 4, y = 4, xt - 6, and y, = 4 cm with respect to the global xy coordinate
system. The plate from which the element is extracted has a thickness of 1 cm.
W/m-°C, respectively, and 0 25°.
Assume k'x and k[ are 200 and 1 10 is

8-118 Consider the elemental volume shown in Fig. P8-1 18. Perform a steady-state energy
balance that includes the effects of the heat conduction in the r and z directions as
well as the internal heat source Q. Assume that the conduction heat fluxes vary

accordina to a first-order Taylor expansion as shown on the figure. Use Fourier s

law of heat conduction (a constitutive relationship) to eliminate the conduction heat
fluxes, thereby showing the validity' of Eq. (8-126). Hint: Note that the effective

area for heat conduction in the r direction is not constant.

(8-126) given by
that the variational principle that corresponds to Eq.
8-119 is
Show

r dr dz

+ T - VzhJ 1 + fi R TaS T) r dC
2irJ (<f sB
c
 STEADY-STATE THERMAL A.SD FLLID FLOW ANALYSIS

Figure P8-118

providing that imposed heat fluxes and convection are allowed on the boundary of
theaxisymmetnc body Note that the nomenclature from Sec 8-9 is used and the
imposed heat flux q# is posiine if the heat flux is directed toward the body

8-120 Extend the finite element formulation of the axisymmetnc heat conduction problem
to the case of radiation from the surface of the body to a large enclosure The energy
balance on the boundary (or surface of the body) given by Eq (8- 1 33a) must be
modified to include the heat flux from the radiation which is given by
q,B

9,b = e B o(T* - VrB )

Note that this assumes that the heat flux from radiation is directed away from the
surface Does this mean the formulation applies only if the radiation is from the
body (being analyzed) to the receiver 9 Please explain. Does this boundary radiation
yield another element stiffness matrix, another nodal force vector, or both9 Derive
them

8-121 Show that the element stiffness matrix from conduction in the axial direction (r
direction) is given by Eq (8-152) if the three-node, axisymmetnc tnangular element
is used What assumptions are made in amving at this result 9
 PROBLFMS 491

8-122 Evaluate the element stiffness matrix from boundary convection for the axisymmetric
heat conduction problem (given by Eq. (8-142c)] for the case of a three-node,
axisymmetric triangular clement
a. With leg jk on the global boundary
b. With leg ki on the global boundary

8-123 Show that the element nodal force vector from an internal heat source that is assumed
to be uniform over the element is given by Eq. (8-158) if the three-node, axisym-
metric triangular element is used.

8-124 Show that the element nodal force vector from boundary convection from leg ij of
the a three-node, axisymmetric triangular element is given by Eq. (8-159) State
all assumptions made in arriving at this result.

8-125 Evaluate the element nodal force vector from boundary convection from an axi-
symmetric body (given by Eq. (8-143c)) if the three-node, axisymmetric triangular
element is used, with leg jk on the global boundary.

8-126 Evaluate the element nodal force vector from boundary convection from an axi-
symmetric body [given by Eq. (8- 143c)] if the three-node, axisymmetric tnangular

element is used, with leg ki on the global boundary.

8-127 Show that the element nodal force vector from an imposed heat flux on the boundary
of an axisymmetric body is given by Eq. (8-160) if the three-node, axisymmetric
triangular element is used and leg ij is on the global boundary. State all assumptions
made in arriving at this result.

8-128 Evaluate the element nodal force vector from an imposed heat flux on the boundary
of an axisymmetric body [given by Eq. (8- 143b)] if the three-node, axisymmetric
triangular element is used, with leg jk on the global boundary

8-129 Evaluate the element nodal force vector from an imposed heat flux on the boundary
of an axisymmetric body [given by Eq. (8-1 43b)] if the three-node, axisymmetric
triangular element is used, with leg ki on the global boundary'.

8-130 Consider the axisymmetric heat conduciton problem for the case of a "point heat

source. In order for the problem to be considered axisymmetric. the heat source
must be uniform and is generally specified on a unit
in the circumferential direction

circumference basis. Consequently, more appropriate

it to refer to such a heat
is

source as a circumferential line source. Let Q' be the heat source per unit circum-
nodal
ference and unit time at point (r0 ,z0 ). Determine the corresponding element
tnangular element.
force vector for the case of a three-node, axisymmetric

fluxes from
8-131 Determine the expressions for the element resultants, i.e.. the heat

conduction in the r and z directions if the three-node, axisymmetric triangular element

the average heat
is used. Note that these heat fluxes are usually considered to be
element centroi
fluxes over the element and are generally associated with the

problem in Sec.
8-132 In the formal development of the axisymmetric heat conduction
may be heterogeneous,
8-9, only isotropic materials were considered (the material
axisymmetric
however). Recall that it is not possible to have fully anisotropic,
such that t e pnncipa
materials. Why not? It is possible to have stratified materials
492 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

values of thermal conductivity, k, and k t (see Problems 8-115 to 8-117), occur in

the r and z directions, respectively, as shown in Fig. P8-132.

Figure P8-132

a. By using the approach indicated in Problem 8-118, show that the governing
equation in this case is given by

!A + Q = 0
r dr

b. How must the expressions for the element stiffness matrices given by Eqs.
(8-142a) and (8- 1 42b) be modified in order to account for this stratified material?
Be specific

8-133 The expressions for the element stiffness matrices from conduction in an axisym-
metnc body are given by Eqs (8-142a) and (8-142b) Since these results hold for
any axisymmetnc element, let us evaluate these stiffness matrices for the four-node,
axisymmctric rectangular element by proceeding as follows
a. With the help of Eqs (6-32), show that

dN
= j 3N
dx a dr

SN _ 1 SN
By b ds

b. Using the results from part (a), rewrite the integrals in terms of derivatives of
the shape funcitons with respect to the serendipity coordinates r and s Do not
confuse the radius r that appears in the integral with the serendipity coordinate
r (e g use r' to represent the radial coordinate) Do not forget to change the
,

limits on the integrations Also note that the elemental area dr dz in the rz

coordinate system is related to the elemental area in the rs coordinate system

by dr' dz = ab dr ds.
C. Evaluate the resulting integrands at the element centroid (i.e., at r = 0 and
s = 0) Then treat the integrands as though they are constant and puli them
through the integral. Evaluate the remaining trivial integrals Show that the result
for the element stiffness from conduction m the radial direction is given by
 PROBLEMS 493

1 1-1-1
K' =
Irrrkb 1 1-1-1
4a -1-1 1 1

-1 ~1 1 1_

where r is the radial coordinate at the centroid of the element. State the other
assumptions made in arriving at this result,

d. Derive the corresponding result for KL. What assumptions are made?

8-134 Evaluate the element stiffness matrix from boundary convection from an axisym-
metric body [i.e., Eq. (8-142c)J if the four-node, axisymmetric rectangular element
is used and if face ij happens to be on the global boundary. Evaluate the integrals
by first evaluating the integrands at the centroid of face ij (i.e., at serendipity
coordinates r = + 1 and x = 0) and then treating the integrands as though they
were constant. Do not confuse the radial coordinate r with the serendipity coordinate
r (e.g., use r' to denote the radial coordinate).

8-135 Evaluate the element stiffness matrix from boundary’ convection from an axisym-
metric body [i.e.. Eq. (8-142c)] if the four-node, axisymmetric rectangular element
is used and if face jk happens to be on the global boundary’. Evaluate the integrals
by first evaluating the integrands at the centroid of face jk (ie., at serendipity

coordinates r = 0 and x = + 1 ) and then treating the integrands as though they

were constant. Do not confuse the radial coordinate r with the serendipity coordinate
r (e.g., use r' to denote the radial coordinate).

8-136 Evaluate the element nodal force vector from an internal heat source in an axisym-
metric body [i.e.. Eq. (8- 143a)] if the four-node, axisymmetric rectangular element
is used. Evaluate the integrals by first evaluating the integrands at the element
centroid (i.e., at serendipity coordinates r = 0 and x = 0) and then treating the
integrands as though they were constant. Do not confuse the radial coordinate r
with the serendipity coordinate r (e.g., use r' to denote the radial coordinate).

8-137 Evaluate the element nodal force vector from boundary convection from an axi-
symmetric body [i.e., Eq. (8- 143c)] if the four-node, axisymmetric rectangular
element is used and if face ij happens to be on the global boundary. Evaluate the
integrals by first evaluating the integrands at the centroid of face ij (i.e., at serendipity

coordinates r = + and - 0) and then treating the integrands as though they

1 .r

were constant. Do not confuse the radial coordinate r with the serendipity coordinate
r (e.g., use r' to denote the radial coordinate).

8-138 Evaluate the element nodal force vector from boundary convection from an axi-
symmetric body [i.e., Eq. (8-143c)] if the four-node, axisymmetric rectangular
element is if face jk happens to be on the global
used and boundary’. Evaluate the

integrals by first evaluating the integrands at the centroid of fac t jk (i.e., at ser-
endipity coordinates r = 0 and s — + 1 ) and then treating the integrands
as though

coordinate r with the serendipity

they were constant. Do not confuse the radial

coordinate r (e.g., use r' to denote the radial coordinate).

imposed on an
8-139 Evaluate the element nodal force vector from a boundary heat flux
rectangular
axisymmetric body [i.e.. Eq. (8- 143b)] if the four-node, axisymmetric
element is used and if face mi happens to be on the global boundary. Evaluate the
integrals by first evaluating the integrands at the centroid of face mi (i.e., at ser-
494 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

cndtpity coordinates r = 0 and s = — 1) and then treating the integrands as though

they were constant. Do not confuse the radial coordinate r with the serendipity
coordinate r (e g., use r' to denote the radial coordinate)

8-140 Derive the expressions for the average heat fluxes within an element if the four-
node, axisymmetnc rectangular element is used Hint Evaluate the expressions for
Fourier's law of heat conduction at the element centroid and interpret these results
to be the average heat fluxes from conduction m the radial and axial directions

8-141 Consider the axisymmetnc, triangular element shown in Fig. P8-I41. The body
from which the element is extracted is made of cast iron The nodal coordinates
are r, =20,?, = 1 5, r, = 1 7, z, = 3 0, r* = 0 6, and z* = 1.8 in.

a. Determine the element stiffness matrices from conduction in the r and z direc-

tions
b. Determine the element stiffness matrix from boundary convection if leg y happens
to be on the part of the global boundary that undergoes convection to a fluid at

58°F through a convective heat transfer coefficient of 75 Btu/hr-ft 2 -°F

8-142 For the element in Problem 8-141, determine the nodal force vectors
a. From an internal heat source of 40 Btu/hr-ft 3
b. From boundary convection
C. From a boundary heat flux of 325 Btu/hr-ft 2 imposed on leg ij (which is on the

part of the global boundary also undergoing convection )

8-143 Consider the axisymmetnc, triangular element shown in Fig P8 143 the body from
which the element is extracted is made of brass The nodal coordinates are r, = 5,

z, = 6, ry = 4, Zj = 4, rk = 6, and z t = 4 cm
a. Determine the element stiffness matrices from conduction in the r and z direc-
tions
b. Determine the element stiffness matrix from boundary convection if leg ki hap-
pens to be on the part of the global boundary that undergoes convection to a
2-0C.
fluid at 55°C through a convective heat transfer coefficient of 4 W/cm
 PROBLEMS 495

8-144 For the element in Problem 8-143. determine the nodal force vectors
a. From an internal heat source of 80 \\7cm ’

b. From boundary convection

c. From a boundary heat flux of 35 W/crn- imposed on leg ki (which is on the part
of the global boundary' also undergoing convection)

8-145 Consider the axisymmetric, triangular element shown in Fig. PS-145. The element
is extracted from an axisymmetric body that is fabricated from hot rolled, low carbon
steel. The nodal coordinates are r, = 0, z, = 0, rt = 0, zj = - 1, rk = 2, and
= — 1 cm.

a. Determine the element stiffness matrices from conduction in the r and z direc-
tions.

b. Determine the element stiffness matrix from boundary convection if leg jk hap-

pens to be on the part of the global boundary that undergoes convection to a

fluid at 75°C through a convective heat transfer coefficient of 5
\V/cm 2 -°C.

8-146 For the element Problem 8-145, determine the nodal force vectors
in
a. From an internal heat source of 15 /cm
3
W
b. From boundary convection
496 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

c. From a boundary heat flux of 50 W/cm 2 imposed on leg jk (which is on the part
of the global boundary also undergoing convection)

8-147 Consider the axisymmetnc, triangular element shown in Fig. P8-147. The element
is extracted from an axisymmetric body that is fabricated from pure copper. The
nodal coordinates are r,
~ 10, z, = 10, r, = 1 1, z, = 12, rt = 9, and zk — 12 in.
o

a. Determine the clement stiffness matrices from conduction in the r and z direc-
tions
b. Determine the element stiffness matrix from boundary convection if leg fci hap-
pens to be on the part of the global boundary that undergoes convection to a
2
fluid at 56°F through a convective heat transfer coefficient of 63 Btu/hr-ft -°F

8-148 For the element in Problem 8-147, determine the nodal force vectors
a. From an internal heat source of 25 Btu/hr-ft 3
b. From boundary convection
c. From a boundary heat flux of 50 Btu/hr ft 2 imposed on leg ki (which is on the

part of the global boundary also undergoing convection)

8-149 Consider the axisymmetnc, rectangular element shown in Fig P8-149 The element
is extracted from an axisymmetnc body that is made of aluminum The coordinates

of the nodes are r, = 4, z, = 2, r = 4. zt = 3, rk = 2 , z* = 3, rm = 2, and

t

z„ = 2 Perform the necessary integrations by following the approaches men-

cm
tioned in Problems 8 133 to 8-135 and

*,

Figure P8-149
 PROBLEMS 497

a. Determine the element stiffness matrices from conduction in the r and r direc-
tions.

b. Determine the element stiffness matrix from boundan convection if face jk

happens to be on the part of the global boundan,’ that undergoes convection to
a fluid at IO'C through a connective heat transfer coefficient of 5 W/cm 2
-°C.

8-150 Consider the element in Problem 8-149. Perform the necessary integrations by
following the approaches indicated in Problems 8-136 to 8-139. and determine the
nodal force vectors
a. From an interna! heat source of 5 VV cm 3
b. From boundary convection
c. From a boundary heat flux of 30 W/cm- imposed on face jk (which is on the
part of the global boundary also undergoing convection)

8-151 Consider the rectangular element shown in Fig. PS-151. The element is extracted
from an axisymmetric body that is made of brass. The coordinates of the nodes are
r, = 5, — 2, r, = 5. - = 4. rf = 2, :t = 4, rn = 2, and r_ = 2 in. Perform
the necessary integrations by following the approaches mentioned in Problems
S-133 to 8-135 and

®
*? |
r /

^tS

Figure P8-151

a. Determine the element stiffness matrices from conduction in the r and r direc-
tions.
b. Determine the element stiffness matrix from boundan' convection if face km
happens to be on the part of the global boundary that undergoes convection to
2-
a fluid at 720~F through a convective heat transfer coefficient of 75 Btu/hr-ft
C
F.

8-152 Consider the element in Problem 8-151. Perform the necessary integrations by

following the approaches indicated in Problems 8-136 to S-139. and determine the
nodal force vectors
3
a. From an internal heat source of 10,000 Btu/hr-ft
b. From boundary convection
c. From a boundary heat flux of 420 Btu/hr- ft“ imposed on face km (which is on

the part of the global boundary also undergoing convection)

8-153 For the element in Problem 8-141, determine the element

nodal force vector from
of circumference
a circumferential line source with a strength of 50 Btu/hr per inch
located at r = 1,5 and _ 0 — 2.0 in.
at the location of the source. The source is 0
498 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

8-154 For (he element in Problem 8-143, determine the element nodal force vector from
a circumferential line source with a strength of 65 W per centimeter of circumference
at the location of the source The source is located at r0 = 5 and Zo = 5 cm
8-155 For the element Problem 8-145, determine the element nodal force vector from
in

a circumferential line source with a strength of 40 W

per centimeter of circumference
at the location of the source The source is located at r0 = 2 and z0 = — cm 1

8-156 For the element in Problem 8 147, determine the element nodal force vector from
a circumferential line source with a strength of 130 Btu/hrper inch of circumference
at the location of the source The source is located at r0 = 11 and z0 = 12 in

8-157 For the element in Problem 8-149, determine the element nodal force vector from
a circumferential line source with a strength of 30 W per centimeter of circumference
at the location of the source The source is located at r0 — 3 0 and zQ - 2 5 cm

8-158 For the element tn Problem 8-151, determine the element nodal force vector from
a circumferential line source with strength of 30 Btu/hr per inch of circumference
at the location of (he source The source is located at r0 = 4 and z0 — 3 in

8-159 For the element Problem 8-141 the nodal temperatures are obtained from the
in

solution of Ka = f and are T, = 175, 7} = 168, and Tk = 173°F Determine the

average heat fluxes from conduction in the r and z directions, respectively With
what point in the element are these heat (luxes normally associated 1

8-160 For the element in Problem 8-143 the nodal temperatures are obtained from the
solution ofKa = f and are T, - 75, T, = 82, and Tk = 73°C Determine the
average heat fluxes from conduction in the r and z directions, respectively With
what point in the element are these heat fluxes normally associated 1

8-161 For the element in Problem 8-149 the nodal temperatures are obtained from the
solution of Ka = = 92. T = 95, and Tk - 89, Tm = 96°C Determine
f and are T,
}
the heat fluxes at the element centriod from conduction in the r and z directions,
respectively These heat fluxes may be regarded as the average conduction heat
fluxes

8-162 For the element in Problem 8 151 the nodal temperatures are obtained from the
Ka - f and are T, = 192, T, = 195, Tk = 189, and T„ = 196 F.
0
solution of
Determine the heat fluxes at the element centroid from conduction in the r and z
directions, respectively These heat fluxes may be regarded as the average conduction
heat fluxes

8-163 Consider the elemental volume dx dy dz shown in Fig. P8-163 Note that the
conduction heat fluxes in the x, y, and z directions are assumed to vary according
to a first-order Taylor expansion In addition, an internal heat source Q is present
By performing a steady-state energy balance and by invoking Fourier's law of heat
conduction, given by Eq (8-161), show that the governing equation for steady-
state heat conduction tn a heterogeneous, isotropic, three-dimensional body is given
by Eq (8-162)

8-164 Show that the variational principle that corresponds to Eq (8-162) for three-
dimensional, steady-state heat conduction in a heterogeneous, isotropic body is given
by
 PROBLEMS 499

Figure P8-163

l = 'hk dx dy dz

+ VdiB r- + h B TaB T) dS

if imposed heat fluxes (

q sB ) and convection from the boundary (i.e., the surface)
of the three-dimensional body are taken into account. Note that the nomenclature
from Sec. 8-10 is used.

8-165 Extend the formulation for three-dimensional heat conduction in Sec. 8-10 to the
case of radiation from (or to) the surface of the body to (or from) a large enclosure
at temperature TrB . In particular, derive the expressions for the corresponding element
stiffness matrix and element nodal force vector. Note that the governing equation
[given by Eq. (8-162)] does not need to be modified. Why not?

8-166 Show that the element stiffness matrix from conduction in the .v direction in a three-
dimensional body is given by Eq. (8-177) if the four-node tetrahedral element is

used. What assumptions are made in arriving at this result?

8-167 Show that the element stiffness matrix from conduction in they direction in a three-

dimensional body is given by Eq. (8-178) if the four-node tetrahedral element is

used. What assumptions are made in arriving at this result?

8-168 Show that the element stiffness matrix from conduction in the z direction in a three-
dimensional body is given by Eq. (8-179) if the four-node tetrahedral element is

used. What assumptions are made in arriving at this result?

500 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

8-1 69 Evaluate the element stiffness matrix and element nodal force sector from boundary
convection from a three-dimensional body {see Eqs. (8-175d) and 8-176c)] if the

four-node tetrahedral element is used, and if face jkm of the tetrahedron happens
to be on the part of the global boundary that undergoes convection

8-170 Evaluate the element stiffness matrix and element nodal force vector from boundary
convection from a three-dimensional body [see Eqs. (8-175d) and (8-l76c)J if the
four-node tetrahedral element is used, and if face ikm of the tetrahedron happens
to be on the part of the global boundary that undergoes com ection.
8-171 Evaluate the element stiffness matrix and element nodal force vector from boundary
convection from a three-dimensional body [see Eq (8-175d) and (8-176c)J if the
four-node tetrahedral element is used, and if face ijm of the tetrahedron happens to
be on the part of the global boundary’ that undergoes convection.

8-172 Evaluate the element nodal force vector from an imposed boundary heat flux on
the surface of a three-dimensional body [see Eq (8-176b)J if the four-node tetra-
hedral element is used Assume that face jkm of the tetrahedron happens to be on
the part of the global boundary on which a heat flux is imposed

8-173 Evaluate the element nodal force vector from an imposed boundary heat flux on
the surface of a three-dimensional body [see Eq (8- 176b)] if the four-node tetra-
hedral element is used Assume that face ijm of the tetrahedron happens to be on
the part of the global boundary on which a heat flux is imposed

8-174 Evaluate the element nodal force vector from an imposed boundary heat flux on
the surface of a three-dimensional body [see Eq <8- 176b)] if the four node tetra-

hedral element is used Assume that face dm of the tetrahedron happens to be on

the part of the global boundary on which a heat flux is imposed

8-175 Show that the element nodal force vector from an internal heat source in a three-

dimensional body is given by Eq (8-182) if the four-node tetrahedral element is

used What assumptions are made id arriving at this result 7

8-176 in a three-dimensional body Let us derive

Consider the case of a point heat source
an expression for the corresponding nodal force vector for such a point source
Assume the strength of the source is Q0 (in units of Watts or Btu/hr) and the location
is Xq, v 0 . and z„. Use the three-dimensional form of the delta-function (see Sec. 8-
8 for the two-dimensional form of the delta-function) to represent the internal heat
source in Eq (8-l"?6a) and evaluate the result to get

A’,fxo.vo,zo)

~ Qo
^*t*o.>0" 2o)

/ '«( r0'>0--c3_
p
,

Explain the physical significance of this result

8-177 Develop a procedure, formula, or algorithm that could be used to determine the
area of a typical face of the tetrahedral element, given the coordinates of the nodes
of the element. Note that this is needed in order to evaluate A,^ in Eqs (8-181).
(8-183), and (8-184).
 PROBLEMS 501

8-178 The solution of Ka - f for the three-dimensional heat conduction

problem yields
the nodal temperatures. Assuming these temperatures are known, show how the
average heat fluxes from conduction in the v. and
\, r directions may be calculated
if the four-node tetrahedral element is used

8-179 The expression for the dement stiffness matrix from conduction in the x direction
in a three-dimensional body, given by Eq (8- 175a), is quite general and may be
applied to any three-dimensional element Evaluate the integral for the case of an
eight-node brick element by evaluating the integrand at the element centroid (i.e.,
at serendipity coordinates r = 0, s ~ 0, and t = 0) and then treating the integrand

as though it were constant.

8-180 Repeat Problem 8-179 for the element stiffness matrix from conduction in the
y
direction in a three-dimensional body, given by Eq. (S-175b).

8-181 Repeat Problem 8-179 for the element stiffness matrix from conduction m the z
direction in a three-dimensional body, given by Eq (8-175c).

8-182 The expression for the element stiffness matrix from convection from (or to) the
surface of a three-dimensional body, given by Eq.(8-175d), is quite general and
may be applied to any three-dimensional element. Evaluate the integral for the case
of an eight-node brick element with face 1 -2-3-4 on the global boundary by eval-
uating the integrand at the centroid of face 1 -2-3-4 (what are the values of serendipity
coordinates at this point?) and then treating the integrand as though it were constant.
Determine the corresponding element nodal force vector in this case.

8-183 Repeat Problem 8-182 if face 1 -2-5-6 of the brick element is on the part of the
global boundary undergoing convection.

8-184 Repeat Problem 8-182 if face 3-4-7-8 of the brick element is on the part of the
global boundary undergoing convection.

8-185 The expression for the element nodal force vector from a distributed internal heat

source in a three-dimensional body, given by Eq. (8-176a), is quite general and

may be applied to any three-dimensional element Evaluate the integral for the case
of an eight-node brick element by evaluating the integrand at the centroid of the

element (what are the values of serendipity coordinates at this point?) and then
treating the integrand as though it were constant.

8-186 The expression for the element nodal force vector from an imposed heat flux on
the surface of a three-dimensional body, given by Eq. (8-176b), is quite general
and may be applied to any three-dimensional element. Evaluate the integral for the
case of an eight-node brick element with face 1 -2-3-4 on the global boundary’ by
evaluating the integrand at the centroid of face 1 -2-3-4 (what are the values of
serendipity coordinates at this point) and then treating the integrand as though
it

were constant.

8-187 Repeat Problem 8-186 if face 1-2-5-6 of the brick element is on the part of the

global boundary over which a heat flux is imposed.

8-188 Repeat Problem 8-186 if face 3-4-7-8 of the brick element is on the part of the

global boundary over which a heat flux is imposed.

502 STEADY-STATE THERMAL AND FLUID FLOW ANALYSIS

8-189 The body from which a tetrahedral element is extracted is made of pure copper.
The nodal coordinates of the element are x, = 5, yt = 6, z, = 0, x} = 4, y} ~ 4,
Zj = 0, xk = 5, >* = 5, tk = 4, x„ = 6, y„ = 4. and zm = 0 cm. Determine the

element stiffness matrices from conduction in

a. The x direction
b. The y direction
c. The z direction

8-190 The body from which a tetrahedral element is extracted is made of aluminum. The
nodal coordinates of the element area:, = 2 0, y, = 1 5, z, = 0 0, Xj = 1.7, y,
-
3.0, Zj = -0 2, xk = 1 5, yk - 2.0, zk = 1.7, = 0.6, yn = 1.8, and z ffl
=
0 1 in Determine the element stiffness matrices from conduction in
a. The x direction
b. They direction
c. The z direction

8-191 For the element in Problem 8-189. determine the element stiffness matrix and ele-
ment nodal force vector from convection if face ijk of the tetrahedron is on the part
of the global boundary that is undergoing convection The convective heat transfer
coefficient is 100 W/m 2 -°C and the ambient temepratuie is 45°C

8-192 For the element Problem 8-190, determine the element stiffness matrix and ele-
in

ment nodal force vector from convection if face ikm of the tetrahedron is on the
part of the global boundary that is undergoing convection The convective heat
0
transfer coefficient is 50 Btu/hT-ft 2 - F and the ambient temperature is 72°F

8-193 For the element in Problem 8-189, determine the element nodal force vector from
a uniform internal heat source of 50 W/cm 3
8-194 For the element in Problem 8-190, determine the element nodal force vector from
3
a uniform internal heat source of 10,000 Btu/hr ft

8-195 For the element Problem 8-189. determine the element nodal force vector from
in

a boundary heat flux of 200 W/cm 2 if face tjk is receiving the imposed flux

8-196 For the element in Problem 8 190, determine the element nodal force vector from
a boundary heat flux of 1000 Btu/hr-ft 2 if face ikm is receiving the imposed flux

8-197 For the element m Problem 8-189, determine the element nodal force vector from
a point heat source of 75 Wat x0 = 5,
y0
= 5, and r0 = 0 cm
8-198 For the element Problem 8-190, determine the element nodal force vector from
in
a point heat source of 185 Btu/hr at Xq = 1 6,
y0 = 2 1 , and z0 = 1 1 in
8-199 Consider the velocity potential formulation of the two-dimensional potential flow
problem in Sec. 8-11.
a. Show that the velocity components defined by Eqs (8-187) satisfy the motational
flow condition exactly.
b. Show that Laplace’s equation (8-188) results if these expressions for the velocity
components are substituted into the two-dimensional continuity equation for an
incompressible fluid

8-200 Extend the two dimensional velocity potential formulation in Sec. 8-1 1 to the three-

dimensional case In particular, determine the equations that correspond to

Eqs.
 PROBLEMS 503

(8-187) to (8-197). Use u, v, and n' to denote the velocity components

in the x, y,
and r directions, respectively. Do not evaluate the integrals that
appear in the
expressions for the element characteristics.

8-201 Under what conditions do Eqs. (8-202) and (8-203) hold? Derive the analogous
equations that would need to be used if leg jk were on the global boundary
and if
the velocity components u and v are assumed to vary linearly over the leg.
At node
j the velocity components are u} and vr and at node k the velocity components are
"i and v*.

8-202 Under what conditions do Eqs. (8-202) and (8-203) hold? Derive the analogous
equations that would need to be used if leg ki is on the global boundary and if the
velocity components u and v are assumed to vary linearly over the leg. At node i
the velocity components are it, and >’,. and at node k the velocity components are
and vL .

8-203 The direction cosines (;i

r
and «,) of the outward normal unit vector on leg ij of the
triangular element may be computed with the help of Eqs. (8-208) and (8-209).
Extend these equations to the case when leg jk is on the global boundary.

8-204 The direction cosines (n, and «,) of the outward normal unit vector on leg ij of the
triangular element may be computed with the help of Eqs. (8-208) and (8-209).
Derive the corresponding equations for the case of leg ki on the global boundary.

8-205 Determine the direction cosines nx and on leg jk for the element in Example
8-22 (see Fig. 8-26).

8-206 Determine the direction cosines n, and n, on leg ki for the clement in Example
8-22 (see Fig. 8-26).

8-207 Consider the stream function formulation of the two-dimensional potential flow'

problem in Sec. 8-11.

a. Show that the velocity components defined by Eqs. (8-214) satisfy the two-
dimensional continuity equation exactly (for an incompressible fluid).

b. Show that Laplace’s equation (8-215) results if these expressions for the velocity

components are substituted into the irrotational flow condition.

8-208 By applying the Galerkin method (on a element basis) to the governing equation
given by Eq. (8-215) for two-dimensional potential flow of an incompressible fluid
[see Eq. (8-216)], show that Eqs. (8-217) to (8-221) hold if the stream function
formulation is used.

8-209 By performing an energy balance on the elemental area dx dy shoum in Fig.

P8-209. derive the governing equation given by Eq. (8-225a). Assume that the
conduction heat fluxes vary according to a first-order Taylor expansion as shown
on the figure. Remember to include the effect of the fluid
motion in the x direction
state the conditions
(i.e., the energy transport by fluid motion or convection). Clearly
under which Eq. (8-225a) holds.

unsymmetric. Recall
8-210 Show element stiffness matrix given by Eq. (8-241) is
that the
convective energy transport term in the governing
that this matrix results from the
(8-241) for the case of
equation [i.e., Eq. (8-225a)]. Evaluate the integral in Eq.
the three-node triangular element.
504 STEADY STATE THERMAL AND FLUID FLOW ANALYSIS

8-211 Show that the element stiffness matrix given by Eq (8-241) is unsymmetric. Recall
that this matrix results from the convective energy transport term in the governing
equation (i e . Eq (8-225a)J Evaluate the integral m Eq (8-241) for the case of
the four-node rectangular element by evaluating the integrand at the element centroid
and then treating the integrand as though it were constant

8-212 Try to obtain the variational principle that would correspond to Eq (8-225a), and
clearly explain the difficulties that are encountered

8-213 Consider the case of a fully developed laminar flow of an incompressible Newtonian
fluid m a duct of circular cross section On the walls of the duct, either the tem-
perature is prescribed or a heat flux is imposed (these two boundary conditions may
occur on different parts of the boundary) It is assumed further that the boundary
conditions are axisymmetnc It is desired to obtain the governing equation in this
case Recall from elementary fluid mechanics that if the viscosity of the fluid is

assumed to be constant, the velocity profile is given by

“ <r) = 25 - (s)
['
]
where a is the average fluid velocity in the duct and R is the radius of the duct. By
performing an energy balance on an annular-shaped elemental volume 2irr dr dz,
show that the governing equation is given by

if axial conduction is not neglected Assume a first-order Taylor expansion for the

conduction heat fluxes. Include the effect of the convective energy transport in the
derivation. Which term in the governing equation abov e results from the convective
energy transport?

8-214 Obtain the expressions for the element characteristics for the situation described in
Problem 8-213 Do not evaluate the resulting integrals Identify the element stiffness
 PROBLEMS 505

matrix that is not symmetric. What is the implication of this in the solution of
Ka = f for the nodal temperatures in the vector a?

8-215 Consider the case of laminar flow of a Newtonian fluid as described in Sec. 8-13.
By beginning with Eq. (S-251) and making use of Eqs. (8-246) and (8-252), show
that Eqs. (8-253) and (8-254) hold.

8-216 Identify the element stiffness matrices in Sec. 8-13 that are unsymmetric. What is
the implication of this during the assemblage step? What effect do the unsymmetric
stiffness matrices have on the solution for the nodal velocities and pressures? How
can the results be improved?

8-217 Clearly explain the differences between the two different assemblage procedures
described in Sec. 8-1 3 when each node has more than one degree of freedom. Which
procedure results in a larger bandwidth? Please explain. Which procedure is anal-
ogous to the assemblage procedure used in stress analysis (see Chapter 7)? Please
explain.

8-218 In a corrosion study to be performed on aluminum, a long specimen with rectangular

cross section is to be held between two isothermal surfaces at 'f, and T2 , as shown
in Fig. P8-218. The height and width of the specimen are H and VP, respectively.

Figure P8-218

a cold corrosive environment at a temperature Tc with

The specimen is exposed to
h c The other side is exposed to relatively
a convective heat transfer coefficient .

convective heat transfer coefficient

warm stagnant water at a temperature T„ with a
 STEADY -STATE THERMAL AND FLUID FLOW ANALYSIS

hK A current js passed through the specimen such that a distributed internal heat
source Q results at a rate given by

where x is measured in the same units as W

and H Note that the heat source is
zero on the surfaces of the specimen and reaches a maximum value at x = 1172.
The global coordinate system is shown in Fig P8-218. The spcciment is isotropic

with a thermal conductivity k A finite element solution for the steady-state tem-
perature distribution is sought with 5 nodes in the x direction and 1 1 nodes in the
-
y direction Determine the temperature distribution for the following parameters
W = 10 cm, H = 25 cm, T, = 125°C, T2 = 40°C. Tc = 5°C. T„ = 25°C, hc =
1000 W/m 2-°C, K
= 75 W/m2-°C, and C, = 2000 W/m\
21 9 Repeat Problem 8-218 for the following parameters IV = 0 32 ft, H = 0 75 ft,

T, = 250°F, T2 = 110°F, Tc = 35°F, T% = 75°F, h e = 0

250 Btu/hr-ft 2 - F, =
15 Btu/hr-ft 2-°F, and C| = 60,000 Btu/hr ft
3

220 Half of the outside surface of a long thick-walled boiler tube receives a uniform
heat flux qsB while the other half is insulated The inside is cooled convectively
with heat transfer coefficient h B to a fluid at temperature ToB The tube is fabricated

from stainless steel and copper as shown in Fig P8-220 The inner and outer radii

of the tube are R, and R0 ,and the interface between the stainless steel and copper
the
is located at a radius Rm Determine the steady-state temperature distribution in
tube for the following parameters qsB = 20,000 Btu/hr-ft 2 , h 8 = 10,000 Btu/hr-
 /

PROBLEMS 507

ft--°F, T = 500 'F. R, = 1.0 in . Rn = 1.25 in and R g = 1.5 in. Note that .

because of the symmetry. only one-half of the tube needs to be analyzed Use
approximately 100 three-node triangular elements

8-221 Repeat Problem 8-220 for the following parameters qsB = 1000 W/cm 2 h B = ,

c
50.000 \V/m 2 - C. = 250"C. R, = 2 5 cm, = 3 5 cm. and /?„ = 4.5 cm.
8-222 It is desired to obtain the two-dimensional stead) -state temperature distribution in
the thin tapered bronze fm of length L. shown in Fig. P8-222. The thickness of the
fin is l, at the base and tapers linearly to t2 at the tip. The height of the fin at the
base and tip are H t
and H 2 . respectively The base is held at a temperature of Tb .

while all exposed surfaces of the fin convect to a fluid a temperature 7} with
at

convective heat transfer coefficients of hi. I’.

: and !h on the lateral faces and tip,
on the bottom edge, and on the top edge as shown in the figure. Use at least 80
three-node triangular elements to determine the two-dimensional temperature dis-
tribution in the fin for the following parameters L; - 2.0 cm. r, = 1 cm, t2 =
0.5 cm. Hi = 1.5 cm. H 2 = 0.75 cm, Tb = 200°C. T{ = 100°C, /t, = 110o" W
c
rrf-'C. h z = 600 \V/m :-cC. and li, = 1500 \V/m 2 - C

8-223 Repeat Problem 8-222 for the following parameters L, =

1.0 in t, ,
-04 in.,
m T = - 212“F, -
r- = 0.25 in.. // = 0 75 in . H2
t
= 0 375 b 390°F, T,
c
.
ft,

2-c
2
and h = 300 Btu'hr-ft-- F
200 Btu/hr-ft F. h = 120 Btuhr-ft
z
-°F. 3
 Higher-Order Isoparametric
Elements and Quadrature

9-1 INTRODUCTION

In Chapter 6 shape functions of the lowest possible order were derived for one-,
two-, and three-dimensional elements. Recall that it was convenient to introduce

local normalized coordinates, such as length, area, volume, and serendipity coor-
dinates. More specifically, length, area, and volume coordinates are used with
lineal, triangular, and tetrahedral elements, respectively. Serendipity coordinates
are used with lineal, rectangular, and brick elements. The use of these types of
coordinates is even more important when higher-order elements are used.
In this chapter several higher-order elements are described, and the shape
functions are given. Numerical integration methods are also presented. Recall further
from Chapter 6 that shape functions generally must meet the compatibility and
completenesscriteria. All the shape functions presented in this section meet these
requirements. In addition, these shape functions are continuous, but their derivatives
are not necessarily continuous from one element to the next at the element bound-
aries. In other words, the shape functions presented here have C°-continuity only.
In this book, the only shape functions with C’-continuity were those derived
for thebeam element in Sec. 7-5. When analyzing plates and shells, C'-continuity
isrequired because both deflection and slope continuity must be assured. The study
of higher-order shape functions with C'-continuity is beyond the scope of this text.
The interested reader may want to consult the book by Zienkiewicz [I].

The letter designations for the local node numbers (i.e., i, j, k, etc.) prove to
be cumbersome for the higher-order elements. Therefore, we will designate the
local node numbers as 1,2, 3, etc., and the length, area, and volume coordinates
will be denoted L h L 2 etc.
,
510 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

9-2 ONE-DIMENSIONAL ELEMENTS

Two higher-order one-dimensional elements can be created by adding one or two

nodes to the interior of the element as shown in Fig. 9-1. Recall that for the two-
node lineal element the shape functions were linear. This is reasonable because the
shape functions provide a convenient interpolation polynomial and a unique straight
line may be drawn through two points. We may refer to this element as the linear-
order lineal element. If we add one node to the lineal element such that it lies

halfway between the two original nodes, we can obtain a quadratic-order lineal
element. Note that either length or serendipity coordinates may be used Similarly
adding two nodes such that the four nodes are equidistant yields a cubic-order lineal
element The shape functions are given below in terms of the length coordinates
Lj and L2 This is followed by the shape functions in terms of the serendipity
coordinate r

Length Coordinates

Linear order (two nodes)

N\ = L,

N2 = L2 (9-11

1 3 2 1 3 2

oj -L 7 L\ * |o 0
- r
1

(b>

-i
iH I— i— |— i—|— rH
1

I* 'I

Figure 9-1 One-dimensional elements in terms of length coordinates L t and Ij. and
sereodipitj coordinate r (a) Linear order, (b) quadratic order, and (c) cubic order
 CEMENTS
Quadratic order
(three nodes):

/Vi = L ^L< - J}

N 2 = L2 (2L -
2 i)

/V, = 4/ /
Cubic-order (four
nodes): 02 -
)

' V| = VlL (3L ~

'
i 0 ( 3L, - 2)
= 0L
2 (3£, _ i )( 3£ 2 _ 2)

^ ^ L,L2 (3L, - ])

A = KLiLiOLn
-'
- |)
(9-3)

Serendipity
Coordinate

Linear-order (two
nodes):

A 'i
= 0(1 - ,-)

A': = 0(1 + r)
Quadratic-order (three (9-4)
nodes):

A'i = -
0 / (1 - r)

A2 = 0/(1 + r)

= (1 + r)(l - r)
Cubic-order (four (9-5)
nodes):

A'i = 9
/u4 (r - 1)( 3/- + I)(3r _ ,
}

A^2 = 9/h +
4 (3/- I)( 3r _ 1)(/. +
N * = % (/• + l)(r - l)( 3/ - _ I}

K “ -%(/•+ l)(r- l„3r +

„
nr .

L2 "“ S S^aHy
T^
- Note how ,h”sh”“ i ' n,red
Bg
tndl"™' i„
” d (our
to tthapa
nodes all0 „T„ear
c
1“ ‘ ,hdt '™.
f„„ ctio „ s inte 1>ol»-ions.
themselves Fif 5(a) L't Indeed,
,d ™ b " in fi80- 5-2(b) and (c),
?„ a
‘ "1 '> are
'

’“ d "’''o
f
ipecive,”
512 HIGHER -ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

(a) (b) {cl

Figure 9-2 Plot of the shape functions for the (a) linear order, (b) quadratic order, and
(c) cubic order one-dimensional elements

9-3 TWO-DIMENSIONAL ELEMENTS

In this section, the triangular and rectangular elements from Chapter 6 are modified
so that some of the higher-order terms are represented in the corresponding parameter
functions. As in the one-dimensional higher-order element, the shape functions will
be seen to be higher-order as well.

The Triangular Element

The triangular element is in the unique position of being able to include complete
polynomials in the parameter function as the Pascal triangle in Fig 9-3 shows.
Recall from Chapter 6 that for the linear-order or three-node triangular element we
assumed a polynomial that involves the first three terms in Fig 9-3 A typical
parameter function c{> is represented by

<|) = Oj + CyX + Ctf {9-7)

where c2 and c 3 are constants The position of the terms m the Pascal tnangle
c,, ,

corresponds to the position of the nodes on the triangular element. The three-node
or linear-order triangular element is shown in Fig. 9-4(a).

The quadratic-order triangular element requires that the parameter function

contain the first six terms from the Pascal tnangle with the nodes on the element

Figure 9-3 The Pascal tnangle

 TWO-DIMENSIONAL ELEMENTS 513

arranged as shown in Fig. 9-4(b). Note that nodes 4, 5, and 6 are on the midsides
of legs 1-2, 2-3, and 3-1, respectively. A typical parameter function in this case
is represented by

4> = c, + c 2.v + cjy + c4x 2 + c$xy + c (j 2 (9-8)

There are six constants in the parameter function and six nodes in the element. The
quadratic-order triangular element may also be referred to as the six-node triangular
element.
If Fig. 9-3 is examined carefully, it is seen that 10 terms are needed to represent
a cubic-order parameter function, or

r< — C| 4* CtT + cyv + c-.r

2
+ 05.13' + c fy 2

+ c7.v 3 + c sx
2 2
y + Cgxy + c 1Qy
3
(9-9)

Note the position of the xy term in the Pascal triangle. Since it is in the interior of
the triangle, a node must be placed in the interior of the triangular element as shown
in Fig. 9-4(c). In fact, this node must be placed at the centroid of the triangle. The
remaining nodes are placed on the legs of the triangle such that each leg is divided
into three equal parts. Note that nodes 4 and 5 are located on leg 1-2, with node
4 closer to node 1. Similar observations may be made about the nodes on the
remaining legs. Node 10 is the interior node.
The shape functions below assume the relative positions of the nodes are as

shown in Fig. 9-4. In Sec. 9-6, these restrictions are relaxed.

Linear-order (three nodes):

N = {
L,

N = 2 L2
N3 = L3 0-10)

order, and (c) cubic

Figure 9-4 The triangular element: (a) linear order, (b) quadratic
order. Note the internal node (node 10) on the cubic-order element.
514 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Quadratic-order (six nodes):

AT, = L,(2L, - 1) N2 = L 2 {2L 2 - 1)

N3 = L3(2L3 - 1) Na = 4L\L2

N5 - 4LjL 3 N6 = 4 L X L3 (9-11)

Cubic-order (ten nodes).

N « AL
t X
(3L, - 1)(3jL, - 2) N2 = ViL 2 (3 L 2 - 1)(3Z. 2 - 2)

N3 = A L3 (3L3 - !)(3L, - 2) N4 = Vi L\L2 (3L, - 1)

Ns » % L,L2 (3L2 - 1) N6 = % L 2L 3 (3 L2 - 1)

N7 = Vi L 2L 3 (3 L 3 - 1) N8 - % L X
L3 (3 L 3 - 1) (9-12)

N9 = Vi L L3 QL -
x x
1) No =
x
27L L2 L 3
{

The interior node (node 10) in the cubic-order element is not shared with any
other element. Therefore, before this element is assembled in the assemblage matnx

K", the node should be condensed by using the substructuring technique described
in Sec. 7-6. Node 10 is treated simply as an interior node, and the element without

this node becomes the superelement The nodal value of the parameter function at

this (and other) interior nodes could be recovered as described in Sec. 7-6

The Rectangular Element

Several higher-order rectangular elements are shown in Fig 9-5. Note that addi-

tional nodes are added to the sides of the rectangle For the quadratic-order element,
the midside nodes must be located midway between the comer nodes For the cubic-
order element, the midside nodes must be located such that the nodes on each side
are equally spaced In addition, the global x and y axes must be parallel to the local

r and s axes, respectively In Sec, 9-6, all these restrictions are relaxed. The shape
functions are given below in terms of the serendipity coordinates r and s:

(b) Id

Figure 9-5 The rectangular (serendipity) element, (a) linear order, (b) quadratic order,

and (c) cubic order.

 niRCIZ-DIMCN’SIONAL ELEMENTS 515

Linear-order (four nodes):

A = l/j
(' + f KI - AN = i/
+
'i s) 4 (l r)(l + s) la ,,,

AN = 'M \ - r)(l S) AN = '/4(1 - r)(l - 5)

Quadratic-order (eight nodes):

A',= !4(I +r)(l - *)(/- 5 - I) AN = !/i( 1 +r)(l + 5X/- + 5 - I)

AN = fcO-rKl^sH-r + i-l) AN = '/,(! - r)( -s)(-r-s-

1
1)
(g . 14)
AN = >/;(l +r)(I -,s 2
) AN = '/’( 1 —r 2
)( 1 + 5)
AN = '/:(1 — r)(i -,v :
) AN = '/:(! - 1
- s)
Cubic-order (12 nodes):

AN = '/>:( 1 4- r)(l - .r)I-I0 + 9(r 2 + s )]

2

AN = '/.:(! + r)(I + 5)j- 10 + 9 (r 2 + 5

2
)]

AN = 'Ail 1
- r)(l + 5)[-10 + 9(r 2 + 5
2
)]

AN = '/<:( 1 - /•)(! - 5)1-10 + 9(/- 2 + 5 2 )]

AN = 9
/<:(l + r)(l - 5
2
)( 1
- 35)

AN = %:( + - +
2
1 r)( I 5 )(1 35)
(9 15)

AN = %:(1 + 5)(1 - 2
r )(l + 3 r)

AN = %:(l + 5)(I - /- 2
)(1 - 3r)

A9 =
r 9
A-(I — r)(I - 2
5 )(i + 3i)

N| 0 = % (\ 2 - r)( 1
- 5 2 )(1 - 35)

AN, = 9
/u(l - 5)(l - r 2 )( 1
- 3r)

Na = 9
/'2(l - 5)d - 2
r )(l + 3 r)

9-4 THREE-DIMENSIONAL ELEMENTS

In Chapter 6, the tetrahedral and brick elements were presented. In this section,

some of the higher-order terms are represented

these elements are modified so that
in the corresponding parameter functions. The shape functions are presented for
these elements in terms of volume coordinates for the tetrahedral element and
serendipity coordinates for the brick element.
S16 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

The Tetrahedral Element

The tetrahedral element is similar to the triangular element in that it is also able to
include the complete polynomial in the parameter function. In Chapter 6, a typical
parameter function <f> is represented by

<{» = c, + CjX + cjy + c4 z {9-16J

for the four-node or linear-order tetrahedral dement shown in Fig. 9-6(a). The
quadratic-order tetrahedral element requires that the parameter function be of the
form

<{> =* Ci + c^x + Cjy + c4z + + c6xy

+ c7y 2 + cgyz + c9 z 2 + c l()zx (9-17)

Note 10 terms are needed, and hence 10 nodes are present in the corresponding
that
dement as shown in Fig. 9-6{b). In a similar fashion, the cubic-order tetrahedral
element requires a parameter function with 20 terms, and hence the element has
20 nodes
In the case of the quadratic-order element, nodes 5 to 10 are located midway
between the respective comer nodes as shown in Fig 9-6(b) For the cubic-order
element, nodes 5 to 16 are located as shown in Fig. 9-6(c), where nodes 5 and 8
are placed such that leg 1-2 is divided into three equal segments. Similar comments
hold about the remaining midside nodes Nodes 17 to 20 are located at the centroids
of faces 1-2-3, 1-3-4, etc as shown in Fig 9-6(c) Most of these restrictions are
relaxed in Sec. 9-7, where the isoparametric tetrahedral element is presented The
shape functions are given below in terms of the volume coordinates for the tetra-

hedral elements.

Linear-order (four nodes):

Figure 9-6 The tetrahedral element, (a) linear order, (b) quadratic order, and (c) cubic

that for the cubic-order element, nodes 17, 18, 19, and 20 are located at the
order. Note
centroid of the respective faces of the tetrahedron
 L L
three-dimensional
elements 517
Quadratic-order (ten
nodes):

N >
= A (2L t
i '
I} A': = L2 (2L 2 - ])
N 3 = L3 (2L 3 I} A'4 = L4 {2L 4 - i)
= 4L,L 2
A6 = 4Z..2L,
n = i 4L L4 (9-19)
}
A 's = 4Z. 2 i
3
A9 = 42^ A'io = 4ZoZ.
4
Cubic-order (twenty
nodes):

^ = -
(31, 1)(3L, - 2) A7 = '/2i2 (31, - 1)(3L, _ 2)
^3 = !4 Z. 3 (3 L3 - i ) ( 3i 3 _ 2) A4 - Vi L 4 (3L 4 - I )(3Z, 4 - 2)
= % L,L 2 (3 L, - l)
^ = 9/2 L L3
x (3L, - i)
A? = % Z.,Z.4 -
(3Z.j i)
As = 9
/2Z.,Z. 2 (3Z., - 1)
N9 = 9
/2 L L3 X (3L3 - i)
^o = % L,L4 (3Z. 4 - l)
/V " = 9/2
^L 3 (9-20)
(3L 2 - i)
^'2 = (37-3 - J)
Nn = % Lj£4 (3Z-3 - i)
N = I9 9/2
V-m (3Z. 4 - i)
= % L2^4 (3L4 - l)
A'l6 = 9
/2 L 2L4 (3 -
2 1)
^>7 = 27L L2L
3 x
W.s = 27L L3L4
N =
i9 27 L2L4
X

X
M20 = 27L2L 3L 4

The Brick Element

Ss r
segments. As mentioned
t:i;rbt
d

in
cubi
r de brick e,emems are
;
it ^ ???“ « ^
sh ° wn jn

^ di d **>
^5 « S
global coordinate ° k mUSt ,ine U with the
functions for
system In Sec 9 7
•’
hese f
restnctIor| s are relaxed.
P
The shape
it
these elpmpnf
56 e,ememS I’
r,
* and ® IVen beIow in terms of the serendipity
coordinates

Linear-order (eight
nodes):

Nl = 1/8
d + -
r)( 1 S)( 1 + ,) N - 2 Vs (1 + r )( 1 + S)(l + t)

a,3 = Vs (1 - +
r)( l s)(l + t) N = 4 !/«(]- r)( 1 - S )( 1 +
NS = 'A (1 + _ _
0(g.2i)
r)(l J)(I {) N6 = Vs (1 + r)(l + s)( 1 - 0
Nl = Vs (1 - + _
r)(1 s)(l t) N s = !/8 (1 - r)(l - s)( 1 - t)
518 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Figure 9-7 The brick element (a) linear order, (b) quadratic order, and (c) cubic order

Quadratic -order (20 nodes)

Ni = Ml+r)(l-s)(l+t)(r-s+t~2)
l

N2 = '/fc(l + r)(l + s)(l + t)(r + j+ r~2)

N} = */&( 1 - r)( + s)( + r)( - r + j + — 2)
1 I f

W 4 = A(l—r)(l — j)(I +/)(— r— J + r- 2)

l

N5 = MI+i-)(l-i)(l-0(r-J-l~2)
,

iV6 = 'M l+r)(l+s)(l~t)(r+s~/~ 2)

V7
] = W(l-r)(l+iXI-0(-r+i-r-2)

fi/8 = m-r)(l~s)(l-t)(~r-s-t~2)
Nr. = Wl-r^n+rin-sl
 t r

THREE-DIMENSIONAL ELEMENTS 519

iV, n = '/i(I —/--){! — s){ — /)I

(9-22)
Au = — r>(
r 2
14(1 -r )(l 1 -s)

A'i’ = '/-( I — r')( 1 -5)(1

A’u = ‘--(1 ~ S 2 )(l — r)(I -rt)

Au =r
'/4I -s : )(l + r)(l — f)

A’is
= '/-(l - .s
2
)(l — r)( 1 -r)

|V, 6 = !4( 1 — s*)( — r){ 1 I -r f)

iV,7 = !
/4I — /~)f — r)(l 1 rj)

iVjg = !
/-(l — 2
)(l — 5)(i — /)

N, 9 = ’/-(l -f 2 )fl — r)(l fi)

iV:o = *4(1 -r2 )(l 4- s)(l -rt)

Cubic-order (32 nodes):

A r
,
= 'Ml -r)(l -s)fl -r/)t9(r 2 -f-j- 2 -i-/ 2 )-I9]

A; =r
'/frs(l-i-r)(l-rf)(l^r)[9(r T-j -rf
2 2 2
)-19]

A3 ='
’/y( 1 - r)( + s){ + rl^r2 + s 2 + 1 1) - 19]
1 1

Ab = 9U( 1 — r)( — s)( 1 I -f r)[9(r

3
-E s- 4- r
2
)
- 1 9J

2
Ii4l4-r)(l-s)(l-/)[9(r 2 4-s -ff )-19]
2
A5 =r

A6 =r
9ca( 1 -fr)(I -rs)(l -f)[9(rtj 2 +l 2)- 19]

A 7 = 'M - r)( - s)( - /)[9(r 2 4- j 2 + r) -

I 1 1 1
9]

2 2 I
A s = VMl-r)(l-.r)(l-r)[9(r +f +i )-i9I
r

i\9 = %4 -r 2 1 )( l -r 3r)( 1 - xK H-r)

A’ I0 = %4 — / 2 )( — 3f)( — s)( — r)
1 I I 1

A'[, = %4 — r 2 1 )( 1 — 3r)( 1 — s)( — r) 1

A' !2 = %*(1 — r 2 )(l — 3r)( — s)(l — r) 1

A' 13 = °M - 12 )( - 3r)( - s)( - r)

1 1 1 1

Nu = M l-
9 2
)(l+ 3 r)( 1 - s)( - r)
I

A i5
r
= Q
/6u(l-7^)(l-3r)(l-s)(l-rr)
520 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

N„ = %.( 1 - r ! )( + 3r)( - S)( + I)

I 1 1
(923|
N„ = %.(l-i 2 )(l-3j)(I+r)(l+l)
N„ = %i(l-j ! )(l+3j)(l+r)(l + l)
N„ = %.<l-s J )(l-3S )(l+r)(l-fl

Nx = 'Ml -s*X 1 +3a)(I + <-)( - 1)

1

N„ = 'Ml — r 2)(l — 3s)(l — r)(i — 1)

= %.(l-i ! )(l+3i)(l-r)(l-0

N2S = %j( 1 -r 2 J(l -3j)(l -r)(l 4-/)

W 2,
= 'M l — j 2)(I +3j)(l -r)([ +()

Njj = %<(l-l 3 )(l+30(l+J)(l+r)

Nk = %j( 1 - r 2 )( 1 - 30(1 + a)(l + r)

N21 = Vw(l — r J )(l + 3r)< 1 + s)(l —I)

Nn = %4(l-r ! )(l-3r)<l+j)(l-/)

N, o = %.(l-l ! )(l-3l)(l+J)(l-r)

W, 0 = !
/«( l - 2
)< 1 + 3l)( + s)< - r)
I 1

N» = 'Ml~i2 )ll -3r)(I + s)(l + ()

N„ = % (l-rJ )(I + 3r)(l+A)(H-l)
t

II should be noted that when i = ± I , the above shape functions reduce to the
corresponding two-dimensional shape functions Therefore, these particular ele-
ments may join a similar-order rectangular or lineal element as shown in Fig.
9-8 for the quadratic-order element

9-5 SUBPARAMETRIC, ISOPARAMETRIC, AND SUPERPARAMETRIC

ELEMENTS

As discussed below, it is relatively easy to distort elements. In particular, the

distorted rectangular and brief elements can be used to accommodate practically

any geometry. For example, let us take the quadratic-order rectangular element and
modify it to a distorted quadrilateral element as shown in Fig. 9-9.Note that the
nodes can be used for two purposes' One is to specify the locations where the
parameter function is sought (i.e , the nodal displacements, temperatures, etc-).
 SUBPARAMETRIC, ISOPARAMETRIC, AND SUPERPARAMETRIC ELEMENTS 521

Figure 9-8 A typical face of the (a) brick element interfaces readily with the two-dimen-
sional rectangular element in (b), which in turn interfaces with the (c) one-dimensional
element.

and the other purpose is to define the geometry of the element. Note in Fig. 9-9
that a quadratic-order curve may be passed through three points (i.e., the nodes on
any one side of the quadrilateral). Thus, it appears to be possible to include curved
boundaries explicitly in the FEM formulations. The geometry of the boundaries is

approximated by polynomials of finite order, and so the boundaries are not exactly
represented in general. However, by using more nodes (or smaller elements), we
can approach the actual curved boundary' to a high degree of accuracy.
The isoparametric element is now defined as follows with the help of Fig.
9-1 0(a).When the same nodes are used to define the element geometry and the
locations where the parameter function is sought, the element is said to be isopar-
ametric.
When the number of nodes used to define the geometry' is less than the number
used to represent the parameter function as shown in Fig. 9- 10(b), the element is

said to be subparametric. Finally, when the number of nodes used to define the

Figure 9-9 The quadratic-order rectangular element shown in (a) may be distorted into

the quadratic-order quadrilateral element shown in (b). Note the curved element boundaries
in (b).
522 HIGHER-ORDER ISOPARAMETRIC ELEMENTS ASD QUADRATURE

geometry is greater than the number used to represent the parameter function as
shown in Fig. 9- 10(c), the element is said to be superparametrie Isoparametric
.

and subparametric elements are used quite frequently in finite element analysis,
whereas superparametrie elements are rarely used.
Let us illustrate how we can distort an element into a more useful shape. For
this purpose let us concentrate on the linear-order element shown in Fig. 9-11. Note
that by distorting the rectangle in Fig 9-1 1(a) into a quadrilateral, we may now

(cl

Figure 9-10 (a) Isoparametric elements, (b) subparametric elements, and (c) superpara-

metric elements. Note that • specifies the locations w here the values of the parameter function
are sought and that O specifies the geometry
 SUBPARAMETRIC. ISOPARAMETRIC, AND SUPERPARAMETRIC ELEMENTS
523

Figure 9-11 The linear-order rectangular element shown in (a) may be distorted into the
linear-order quadrilateral element shown in (b).

place the nodes at more convenient locations. Thus, irregular geometries may be
accommodated. Actually there are some restrictions on the placement of these nodes.
These restrictions are delineated in subsequent sections.
We should be quite familiar by now with representing the parameter functions
in terms of the nodal values with the help of the shape functions. For example, let

us represent the temperature T within the element shown in Fig. 9-1 1(b) as

T = jY,7’| + N2 T2 + N3 T3 + N4 T4 (9-24)

Let us also represent the two global coordinates (.\ ,y) as follows:

.v = A'J.v, + N'^ + Nyx3 + K\ 4 (9-25a)

and

y = 7VJV| + N'2 y2 + N'3y3 + N4y4 (9-25b)

where N[, N2 , etc. represent the shape functions used to define the geometry and
X\, j], etc. are the nodal coordinates. Since the same nodes in Fig. 9-1 1(b) are

being used to define the geometry and the parameter function nodal points, we
have an isoparametric element. For this and all other isoparametric elements, the
shape functions N, and IV,' are equal. In other words, the same shape functions are
used to define the geometry and the parameter function. Therefore,
7 we have

N, = N', for all nodal points i

It is of interest to note that if Eqs. (9-25) are evaluated at a particular node,

for example, node 2, we get

.v = x2

and

v = 3’2
524 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

because N2(x2 ,y2 =

) 1, whereas N (x2 ,y2)
l
= N
3 (x2,y£ = A^fo.yj)
— 0. Clearly
this must be the result if the equations given by Eqs. (9-25) are to be meaningful.
Equations (9-25) can be generalized for all isoparametric elements by writing

X = S N* (9-26a)
1=1

y = 2 Nj, (9-26b)

where the N,' s are the shape functions, (*,,y ) denotes the coordinates of node j,
l

and n is the number of nodes associated with the element.

Equations (9-26) are very powerful in that they allow us to map any point in
the (local) coordinate system to a point in the (jr.y) coordinate system. Therefore,
Eqs. (9-26) are said to provide an isoparametric mapping from the undistorted
element to the distorted element. Undistorted elements may also be referred to as
parent elements. In subsequent sections, the conditions will be given such that we
are assured of a unique or one-to-one mapping This is to say that a point in the
parent element should map into one and only one point in the distorted element
(and vice versa) This will enable us to take a distorted element and map it into an
undistorted element for the purpose of evaluating the integrals that naturally arise.
In Sec. 9-6, the two-dimensional isoparametric elements are developed further.
In Sec 9-7, the three-dimensional isoparametric elements are presented In each
case, the explicit form of the mapping is given

Example 9-1.

Consider the four-node quadrilateral element defined by the following nodal co-
ordinates. *i = 5 y,
= 7 Jt 2 = l y 2 = 4., x 3 = 2 , y3 = 1 x4 — 8., and
, , , ,

y4 = 4. Determine the global coordinates that correspond to r = +10 and s =

+0 75 on the parent element

Solution.

From Eqs (9-26) and (9- 13), we have

* = >/, (1 + /•XI - + Ml + /)(1 + s)-r2

+ Ml - r)(l + s)x3 + 14 (I
- r)( 1 - 4*4
and

y « */4 (1 + r)(l - s)y, + Ml + r)(l + s)y2

+ /* (1
- r)(l + s)y 3 + Ml - r)( I
- s)y4

Substituting the values of the nodal coordinates gives

:

* = V* (I + 'C'
1 + Ml + r)( 1 + s)

+ m 1 - r)(l + s) + Ml - r)(l - S)
 TWO-DIMENSIONAL ISOPARAMETRIC FORMULATIONS 525

and

y = % (1 + r)(l - s) + 4A (1 + r)(l + s)

+ /a (1 -r) (1 + s)+ 4
/4 (1 - r)(l - s)

Now we are in a position to perform the actual mapping:

.v = 5
/4(l + 1)(] - 0.75) + '/4
(1 + 1)(1 + 0.75)

+ %(1 - i)(i + 0.75) + s/4

(1 - 1)(1 - 0.75)

and

3’ = % (1 + 1)(1 - 0.75) + 4
/4 (1 + 1)(1 + 0.75)

+ 'A (1 - 1)(1 + 0.75) + 4

/4 (1 - 1)(1 - 0.75)

or

* = 1.500 and y = 4.375

The parent element should be plotted on a piece of graph paper in order to verify
that (1.5, 4.375) is on the element boundary. More specifically, this point is located
between nodes 1 and 2. Why does this seem reasonable? S9

9-6 TWO-DIMENSIONAL ISOPARAMETRIC FORMULATIONS

In Sec. 9-5 we have seen how it is possible to distort an element in order to give
it a more arbitrary shape. In this section we want to show how the element stiffness
matrix and element nodal force vectors are transformed for the two-dimensional
isoparametric elements into those for the undistorted element. The reason for this
transformation is to simplify the resulting integrations. If the integrands contain the
shape functions directly (i.e., not the derivatives), then we simply represent the
shape functions in terms of the appropriate normalized coordinates. No additional
transformation is necessary in order to evaluate the integrals in this case (see Sec.
9-8).
Both the triangular and quadrilateral element are considered. The quadrilateral
element is considered first because it is described by two independent coordinates

r and s-. It should be recalled that the three area coordinates used to describe the
triangular element are not all independent; this will require some additional special
attention.

The Quadrilateral Element

Figure 9-12 shows the isoparametric forms of the rectangular element. Both the
parent (rectangular) and distorted quadrilateral elements are shown. Note that for
the linear-order distorted element both the parameter function and the sides of the
526 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Order Parent element Distorted element

Quadratic

Cubic

Figure 9-12 The isoparametric quadrilateral elements

quadrilateral are linear, for the quadratic-order distorted element both the parameter
function and the sides of the element are of quadratic order, for the cubic-order
distorted element both the parameter function and the sides of the element are cubic
order.
It should be recalled from Chapters 5 and 8 that typical, two-dimensional

element stiffness matrices are gi\en by Eqs (5-87) and (8-106) for problems in
stress analysis and heat transfer, respectively. Recall that the derivatives of the
shape functions frequently appear in the integrands If we wish to perform the
integrations over the undistorted element, the integrals must be transformed into
ones that contain only r and s (instead of* and v) This is accomplished as described
below
Let us first note that we have the shape functions m
terms of the serendipity
coordinates r and s for the parent elements as given in Sec 9-3. Let us work with
a typical shape function, for example, the /th, and note that

r = rix.y)

j = s(t.v)

and
N, = N,{r,s)
 1

TWO-DIMENSIONAL ISOPARAMETRIC FORMULATIONS 527

Therefore, we may write the total derivatives of JV, as

dN, _ dN, dx dN, dy

(9-27a)
dr dx dr dy dr

and

dN, _ dN, av BN, dy

(9-27b)
ds dx ds dy ds

Let us write these last two equations in matrix form as

dN, dx dy dN,
dr dr dr dx
(9-28)
dN, dx dy dN,

_ ds _ _ds ds_

The 2x2 matrix on the right-hand side is known as the Jacobian matrix and is

denoted by J, or

dx dy
dr dr
(9-29)
dx dy
ds ds

The reader may recall from calculus [2] that an infinitesimal area element dx dy is

related to an infinestimal area element in the (r,.v) coordinate system by

dx dy = jdet J| dr ds (9-30)

In Eq. (9-30), the determinant of the Jacobian matrix is indicated. This determinant
is referred to simply as the Jacobian. If Eq. (9-28) is premultiplied by J
1

,
we get
the desired result
-
dN, dx df ~BN,
dx dr dr dr
dN, dx dy dN,
_dy_ _ds ds_ _ ds _

Let us examine the Jacobian matrix more carefully. Since we have an isopar-
ametric element, we may write [see Eqs. (9-26)]:

x = 2N jXj (9-32a)

and

3- = 5>,» (9-32b)

where the summations are made over the total number of nodes present in the
element, and x} and y, are the coordinates of the nodes. Therefore, the Jacobian
matrix becomes
528 HIGHER-ORDER ISOPARAMETRIC CLEMENTS AND QUADRATURE

ON, -r? dN,

(9-33)

V dN. *' V’ 3Nj

2 17 2 **
Note that for the linear-, quadratic-, and cubic-order elements, the summations
involve four, eight, and twelve terms, respectively. With the help of Eqs. (9-30),
(9-31) and (9-33), every integral over the element area A* may be transformed into

an integral of the form

K' = H(r.s) dr ds 0-34a)

or

f- f*’ f*‘
g(r,s)drds (9-34M

Note that the integration limits also change. In Sec 9-8, we will see that integrals

in these forms may be evaluated numerically

It should be noted that the size of the element stiffness matrix is directly related

to the order of the element as shown in Table 9- 1 . Note that a two-dimensional

stress analysis problem analyzed with the cubic -order isoparametric, quadrilateral
element has an element stiffness matrix that is of size 24 x 24 In this case, each

node has 2 degreecs of freedom and there are 12 nodes, thus the stiffness matrix
is 24 x 24. Since the corresponding heat transfer problem has I degree of freedom

per node, the element stiffness matrix is of size 12x12 for the cubic-order element.

Recall that integrations around the element boundary result in integrals of the
form

Nr s dS and N r*N! dC
J
The first is recognized as the element nodal force vector from surface tractions,
and the second is recognized as the element stiffness matrix from convection from

the boundary of a two-dimensional body. In the case of the integral for the traction,
the elemental area dS around the boundary may be expressed as

Table 9-1 Size of the Element Stiffness Matrices for the Quadrilateral Elements

Structural Thermal
Order of clement (2 DOF per node) (1 DOF per node)

Linear 8x8 4x4

Quadratic (6 x 16 8x8
Cubic 24 x 24 12 X 12
 TWO-DIMENSIONAL ISOPARAMETRIC FORMULATIONS 529

(IS = t dC (9-35)

where r is tiie clement thickness and dC is the infinitesimal arc length. From calculus,
dC is given by

dC = VuLx) 1 + (dy)
1
(9-36)

since we have ,v = .\(r,s) and y = y(r,s), we may write the differentials dx and dy
as:

dx dx
dx = — dr ,
+ — ds
,

(9-37a)
dr ds

and

dy dv
— dr
,
+ — ds
,

(9-37b)
dr ds

where r and .v are the serendipity coordinates. The boundary integrals need to be
evaluated around the element boundary.
Let us derive the appropriate form of the expression for dC by restricting the

development to the legs of the element over which the serendipity coordinate s is

constant; i.e., s = ± 1 on these faces. Therefore with the help of Eqs. (9-32), we
may write Eqs. (9-37) as

dx = — dr - 2) *j
1
dr (9-38a)
dr dr

and

dy
J = ^dr =
dr
2T
^ dr
i
yj' dr (9-38M

on faces where s = ± 1 . A typical side (side 2-3) of a quadrilateral element for

which Eqs. (9-38) apply is shown in Fig. 9-13. The elemental surface area dS or
t dC on sides of constant s becomes

(IS = tdC = (9-39)

global boundary. Note that

Figure 9-13 Typical quadrilateral element with leg 2-3 on the
s = + 1 on this leg.
530 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

In a similar fashion, it can be shown that on faces of constant r (i.e., faces 1-2
and 3-4), the elemental surface area dS is given

dS = tdC — t (9-40}

The serendipity coordinate s (a scalar) should not be confused with the surface
traction s (a vector). Therefore, when these expressions for dS or t dC are substituted
into typical boundary integrals, we get integrals of the form

f' = J^N T srfS = J ]

g(r,s) dr (9-41)
|

or

K' = N r(iNl</C = J 1
H(r,s) I dr 19-42)

on faces of constant serendipity coordinate s Similar integrals result when faces

of constant r are on the global boundary except that the integrands are evaluated
at r = ±1 before integrating with respect to s
The integrals in Eqs (9-34), (9-41), and (9-42) look formidable. However,
they may be evaluated in a relatively straightforward manner by using Gauss-
Legendre quadrature as described in Sec 9-8
The degree of distortion that is possible before the mapping breaks down (and
is no longer one-to-one) is now given For a linear-order quadrilateral element, a
one-to-one mapping is assured if the maximum angle formed by any two sides of

the quadrilateral is less than 180° as shown in Fig. 9- 14(a) For a quadratic-order
quadrilateral element, not only must the same angle condition be satisfied but also
the midside nodes must be in the middle one-third of the distorted sides This
condition is illustrated in Fig. 9- 14(b) For cubic and other higher-order elements,
no such simple conditions apparently exist and the necessary and sufficient condition
for a one-to-one mapping is stated as follows for a one-to-one mapping, the sign
of the Iacobum (the. detetmmarn of the lacoh’.ao mates*) must reroam the same for
all points in the domain mapped [3] For cubic and higher-order elements, it is not

practical to check the sign of the Jacobian at every point within the element Instead,
a few points are checked as explained later

The Triangular Element

The isoparametnc forms of the triangular element are shown in Fig 9-15. Note
that both the parent and distorted elements are shown As in the case of the rec-
tangular element, the number of nodes on each leg of the triangle determines the
order of element (two, three, and four nodes on a leg for linear-, quadratic-, and
cubic-order, respectively) Note how a quadratic-order curve may be drawn through
three points (i e , the nodes), and a cubic-order curve through four points.
 TWO-DIMENSIONAL ISOPARAMETRIC FORMULATIONS 531

Figure 9-14 For the linear-order quadrilateral element in (a) each internal angle must be
less than ISO" for a unique mapping. For the quadratic-order quadrilateral element in (b)

each internal angle must be less than 180 ° and the midsidc nodes must be within the middle
one-third of each leg.

Order Parent element Distorted element

Linear
Same as the
parent element

Quadratic

Cubic

Figure 9-15 The isoparametric triangular elements.

532 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

The development in this section is very similar to that in the previous section
for the quadrilateral element, except that the local, normalized coordinates are area
coordinates, not serendipity coordinates, ft should be recalled that the three area
coordinates are not all independent because they must sum to unity [see Eq.
(6-23)]. This may be taken into account by taking L, and Ln to be the independent
coordinates. The fact that L3 is dependent (on L, and L2) is explicitly taken into
consideration later. Therefore, we may write

L, « Ux,y)
L2 = L2 { x,y)
and

N, - N,(L„L2 )

from which it follows that

BN, _ df^dx^ dN,dy_

(9-43a)
dL, Ax dL, dy dL,

and

<W, dN,dx_ BN,dy_

= + (9-43b)
BL 2 ~ dx dL 2 dy dL2

In matnx form, we have

I
~BN~ "at dy re>l
gfl -

dLt dLi dL, at

(9-44)
3iV, dx dy dN,
JL 2_ JL 2 dL 2 _ ,dy_
Now the Jacobian matnx J is defined by

"at d\~
dLi dL,
J = (9-45)
at
JL 2 dL 2 _

from which it follows that

-
"aiv," dx dy 1
~BN,
dLi dL, dL,
(9-46)
BN, dx ay a-v,

-d>'- sl 2 _ JL 2 _
Note that the partial denvatives of the shape functions N, are needed in the
column vector on the nght-hand side of Eq. (9-46) Recall that we have assumed
N, to be a function of L, and L 2 only, since L } is then dependent on L, and L2 by
Eq (6-23). This implies that in the computations of dN/dL, and 8N/dL2 , »e must
substitute
 i

TWO-DIMENSIONAL ISOPARAMETRIC FORMULATIONS 533

L, = I - L, - L2 (9-47)

into the sliape functions N, before computing these partial derivatives. For higher-
order elements tiiis approacli is very tedious, fortunately, there is an easier way.
Let us denote the partial derivative of A', with respect to L t
as

where the shape function A', is now a function L x ,

L 2 and L3
, , but L 2 and L3 are
held constant in performing the differentiation. The notation

implies N, is a function of L and L 2 and L 2

x
, is held constant [the L2 is not written
since it is not written in Eq. (9-46)[ . It follows that

^=
dL\ dL[
+
dL x \<5L2/l,.lj
+
r
^-t \^z)
(9 . 48)

and since

rJL,
_ SL2
~ 1 = 0 = -1
BL, dL x

we have

dN,
=
(dN,)
(dN,'
_ fdNA
dL,
(1L, \9L
\dL,/
tl flL
\dL 3 ) lux
— '
,

Here A, is taken to be a
function of L, and Z., as
]{ cre
,
^ 1S j a Lcn to be a function of L,. L2 .

,
<Ulu L\
required by Eq. (9-46)

In a completely analogous manner it follows that

W „ te) - (&) 19-511

dL 2 \dL 2 / Ll \dL 3J ui2

Equations (9-50) and (9-51) make unnecessary to get N, as a function of only L,

it

and L2 before computing the partial derivatives on the right-hand side of Eq.
(9-46)?
The Jacobian matrix may be cast into a more usable form by writing

^ ;,L,
' z :IL,
'

dN f
sr* dNj
>

7T-xj m ,
y>
534 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

where the partial derivatives can be computed with help of Eqs. (9-50) and (9-51),
Note that for the linear-, quadratic-, and cubic-order triangular elements, the sum-
mations are made over three, six, and ten terms, respectively. From elementary
calculus it follows that

dx dy = {del J| dL dL2
x (9-53)

It also follows that typical element stiffness matrices and nodal force vectors may
be cast into the form

1" 11
K' = I'
dL dLz
i 19-54)

and
L
(' = s (L,.L 2 .L,) dL, dLl 19-551
JP [ ’

These integrals are in a form that may be integrated relatively easily as shown
in Sec. 9-9 The resulting sizes of the element stiffness matrices for linear-, quad-
ratic-, and cubic-order elements are given in Table 9-2. Note that the element
stiffness matrices for stress analysisand heat transfer are 20 x 20 and 10 x 10,
respectively, if the cubic-order element is used
The transformation of boundary integrals (i.e., over S ' or Ce) is left to the

exercises. As in the case of the quadrilateral element, the necessary and sufficient
condition for a one-to-one mapping is that the sign of the Jacobian remain the same
for each point in the domain mapped

9-7 THREE-DIMENSIONAL ISOPARAMETRIC FORMULATIONS

In this section the brick and tetrahedral elements are distorted into more general
shapes However, the resulting integrations are to be done in the undistorted region
The brick is considered first because the three local normalized coordinates r, s,
and f associated with such an element are all independent In contrast, the four
volume coordinates (now Z.,, L^, Z- 3 and L 4 ) associated with the tetrahedral element
,

are not all independent. As in the two-dimensional isoparametric formulation, this

will require a little more attention.

Table 9-2 Size of the Element Stiffness Matrices for the Triangular Elements

Structural Thermal
Order of element (2 DOF per node) (1 DOF per node)

Linear 6x6 3x3

Quadratic 12 x 12 6X6
Cubic 20 x 20 10 X 10
 THRi.r.-DlMKNSION \1. ISOPARAMETRIC FORMULATION'S 535

The Brick Element

Tlic isoparametric forms of the brick element arc shown in Fig. 9-16. For the linear-
order distorted brick, the four points used to define a face must lie in a plane. Note
that the linear-, quadratic-, and cubic-order elements have two, three, and four
nodes, respectively, along each of the edges.
As in the two-dimensional case, it is necessary to evaluate the element stiffness
matrices and nodal force vectors for these elements. However, it is very desirable
to perform the integrations in the parent dement, as explained in Sec. 9-8. Let us

show how the integrals for the element characteristics may be transformed into ones
over the undistorted or parent regions.
Recall from Sec. 9-4 that the appropriate shape functions for the parent elements
are given by Eqs. (9-21) to (9-23) in terms of the serendipity coordinates r. s. and
Let us take a typical shape function, e.g. the /' th, and denote its dependence on
r. j. and t by writing

A', = N,(r.s.t)

But we also have

r - rlx.y.z)

s = s(x.y.z)

i = i(A.v.r)

Figure 9-16 The isoparametric brick elements.

 2
536 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

where .r, y, and z are the global coordinates The total derivatives of a typical shape
function N, are given by

dN, BN, dx dN, dy dN, dz

(9-56a)
dr dx dr dy dr dz dr

dN, SN, dx df^dy dN,dz

_ (9-56b)
ds dx ds dy ds dz ds

dN, __
dN,dx dN,dy BN, dz
19-56c)
dt dx dt dy dt dz dt

Rewriting Eqs. (9-56) in matrix form gives

dN, dx dy dz dN,
dr dr dr dr dx
dN, dx dy dz dN,
(9-57)
ds ds ds ds dy
dN. dx dy dz dN,
L s< J _ dt dt dt _ L dz J
Now a 3 x 3 Jacobian matrix arises and with the help of

* = 2 N,x, y = 2N ,y, 1 = 2 Nj 2 ,

we may write the Jacobian matrix as

~ dN dNj dN.
'
^ ar
^ 2 IT*
„ dN. „ dN. _ dN.
2 ^7
2 '
(9-58)

„ dN. aty „
^ y S 2 dN.

? !t Xj
*T '
IT '.

where yy , and zy denote the global coordinates of the jth node. The summations
jc
y,
in Eq (9-58) are made over the 8, 20, and 32 nodes for the linear-, quadratic-,
and cubic -order elements, respectively

If Eq (9-57) is premultiplied by the inverse of the Jacobian matrix, we get (he

desired result

~dN,~ Hr, dN, ^ dN. „ av, - 1

"awr
~dx Sir*- S 77 *
dN, ^dN. ~dN. dN,
- (9-59)
¥ Srr*' Sir 2 '
ds

dN,
v aN, sjdNj „SN. dN,

_ dz _
2
L ir*' Sir* 2 IT*. _dt_
 THREE-DIMENSIONAL ISOCARAMETRIC FORMULATIONS 537

Equation (9-59) is useful because it may be used to replace

derivatives of the form
a.V/cx. cX, 'by. and e.V/o: with expressions involving hS/br. bXjbs.
and bX.jbt.
The elemental volume dx dy dz may be written in terms of dr ds dt by noting that

tlx d\ dz = jdet J dr ds dt (9-60)

Therefore, the element stiffness matrices and nodal force vectors may be written
in the form

K'' — j | j
Htr.s.r) dr dsd: (9-61)

and

f' = gtr.s.n dr ds dt (9-62)

j
i
J t

The simple integration limits should be noted. These integrals may be evaluated
by the numerical integration technique described in Sec. 9-S.
The element stiffness matrices can become quite iarge for these elements as
Table 9-3 shows. Note that for the cubic-order element, the element stiffness
matrices are 96 x 95 and 32 x 32 for problems in stress analysis and thermal
analysis, respectively. The numerical integration technique described in Sec. 9-8
makes the use of these elements practical.

Table 9-3 Size of the Element Stiffness Matrices for the Brick Elements

Structural Thermal
Order of deman: (3 DOF per node) ( i DOF per node)
Linear 2- x 2-t 8 x S
Quadratic 60 x 60 20 x 20
Cubic 96 x 96 32 x 32

f
It will now be shown how the integrals aver the element surfaces S may be
transformed into ones over the faces on the parent element. The integrands may be
s. and t in a straightforward manner, since the
converted directlv to functions of r.
shape functions are already known as a function r. s. and t. It should be noted,
however, that on any given face, one of the serendipity coordinates is plus or minus
unity. For example, on face l -5-6-2. we have r —
— 1. whereas on face 5-6-7-S.
we have r = - 1. and so forth. Therefore, on face I -5-6-2. the integrations are

performed over s and only (since r = — I on this face).

r
e
The only part of the integrals over S that requires special attention is the
elemental surface area dS. Let d.\ represent the outward normal area vector to a
surface of constant serendipity coordinate, i.e.. on one of the faces ot the brick.
For example. let us assume that face I -2-3-4 is on the global boundary. On this
face r = — 1 and it can be shown that dA. in this case, is given by [4]
538 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

rfA = (— +
\dr
I —
dr
j +
dr
X
\ds
i + j-j
dr
+ ^k)
ds
drds (9-63)
) )

where the vector cross product is used and may be evaluated by writing it in

determinant form as follows

i
j k

dr dy dz
(9-64)
dr dr Sr

dx ay dz
ds ds ds

The magnitude of d\ is actually dS, and so we have

Clearly dS may be written as a function of only r, s, and t and the nodal coordinates
if Eqs (9-65) and (9-66) are used Note that Eq (9-65) holds on surfaces of constant
t (such as t = lorf = —1) If integrations over surfaces of constant s (or constant
r) are desired, then an expression analogous to Eq (9-65) may be written by
inspection or easily derived Therefore, integrals over a face of the element may
be converted to integrals of the form

o: H(r,s,t) drds (9-67)

f g(r,s.O * * 9 ' 68 *
I

if the face over which / = + 1 is on the global boundary. Similar integrals result
if the other faces are on the global boundary Again, numerical integration of these
integrals is necessary as explained in Sec 9-8.
 THREE-DIMENSION AL ISOPARAMETRIC FORMULATION'S 539

As in the two-dimensional case, the necessary and sufficient condition for a

one-to-one mapping is that the sign of the Jacobian must remain the same for all

points in the domain that is mapped. Obviously it is not practical to check every'
point. Instead, a few select points are used as explained later

The Tetrahedral Element

Figure 9-17 shows the isoparametric forms of the tetrahedral element. Note that
the linear-, quadratic-, and cubic-order elements haxe 2. 3, and 4 nodes, respec-
tively. on each edge of the tetrahedral element. Again we must convert integrals
over a complicated distorted element to integrals o\ er an undistorted parent element.
This is easily accomplished if the integrals in terms of the global coordinates (.v.y.r)
are transformed into integrals in terms of L x . Lz . Z. } . and L x (the volume coordinates).
The development is similar to that for the triangular element, except that an ad-
ditional coordinate ( L x needs
) to be considered. Now L _• is taken to be the dependent
coordinate related to the others by

L. = 1
- L, - L2 - Lx 0-69}

Qrde- Parent e's'-en: O.sto-ted element

Same as the
pe-ent element

Figure 9-17 The isoparametric tetrahedral elements.

540 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Therefore, a typical shape function, e.g., the ith, may be written as a function of
L], L 2 and
, Z. 3 , or

N, = (9*70)

But we also have

L\ - L x
{x,y,z) 19-71 al

L2 = L2(x,y,z (9-71b)

and

L 3 = L 3 (x,y,z) (9-71 C)

Therefore, the total derivatives are given by

dN, dN, dr dN, 8y dN, dz

(9-72a)
dL x dx dL ,
dy dL] dz dL {

dN, dN, dx dN, dy dN, dz

(9-72b)
ql 2 dx dL2 dy dL 2 dz al2

dN, dN, dx dN, dy dN, dz

(9-72cl
dL 3 dx dL 3 dy dL 3 dz dL3

These three equations may be written in matrix form as

~
dN dx dy dz ~dN,~
dL x dL x
dL x
dL x
dx

d/V, dx dy dz dN,
dL 2 dL 2 dL 2 dL 2 dy

dN, dx dy dz dN,
_dL 3 dL 3 dL 3 dL 3 dz
_
Again a 3 x 3 Jacobian matrix arises and with the help of

x = 1L NjXj y = ^tyyj z = 'ZfyZj

the Jacobian matrix is given by

dNj
y dL
^ x

J =
^dN,
2^ y
^
dNj

di2
(9-74)

Z'SL, 2^ y dNj
^dL 3
summations
where Xj, yJt and 2, denote the global coordinates of the jth node. The
in Eq (9-74) are made over the 4, 10, and 20 nodes for the linear-, quadratic-,
and cubic-order elements, respectively.
 THREE-DIMENSIONAL ISOPARAMETRIC FORMULATIONS
541

If Eq. (9-73) is premultiphed by the inverse of the Jacobian matrix, we get the
following useful result:
- 1 ~
dN,
V 2—y a/v,

dx ^dL* 1
^dL? ^ 1
9L,

dN, aw,
=
^ dL Y^j ,

dy 2
J
^dl/ J
dL 2

dN,

dz
Y
^ 9L
diV,
Ay 2 —
^dL, 1
Y
z
rW
ai 3
/
iy
dN,

3
_
Again it is seen that derivatives of the shape functions must be obtained. Because
we have assumed L u L2 and L 3 , to be independent, and L 4 to be dependent, we
would have to use Eq. (9-69) to eliminate LA in the shape functions before taking
the derivatives. This approach is not very practical and an alternate method is now
presented. Let us denote the partial derivative of N, with respect to L x
as

where the shape function N, is now a function of

W
I 2 L 3 and L A (but L 2 L 3
l, u
, , , ,

and Z. 4 are held constant in performing the differentiation) The notation

dN,

dL\

implies N, is a function of L L2
t , , and L3 only, and L 2 and L 3 are held constant \L 2

and L3 are not written since they are not written in Eq. (9-75)]. It follows that

dN, dL ( dN,\
t
dLj (<M,\
+
X*(»L)
+
dL] dL, \3L t / Ci.L^.U &L\ V^2/ Li Li U ^1 \^*3 / L\ Lz.U

cJL 4 / ON, \
( 9 - 76 )
dL t
\<)L A j i A i 2

But

gL|
1
^2 = 0
^2 = 0 — ^ — — 1 0 77 -
)

dL] dL\ dL\ dL]

and so we have
dN, _ ( rW,\ _ /

dL] \dL\) n.iu \^t/L|LzL^

w
* " ( 9 - 78 )

taken Here A', is taken to be

Here A', is

to be a function a function of L t , L2 .

of only t|, L2 . L\. and Aj

and A, as required
by Eq (9-75)
 A
542 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

It can similarly be shown that

5W, __
/awA _ /awA (9-79)
dlq \dL2 ) I, £j u \^4/ Lite Li
and
dN, _ /awA _ /aw (9-80)
aL 3 \dL3 J Ll L2 .u V^4/ L)U-2 L3
Finally, the element volume dx dy dz may be written as

dx dy dz = |det J| dL\ dL2 dLy (9-80)

Therefore, each integral over the element volume may be transformed into an
integral of the form
U 1
K =
'
IT 1 mL„LMdL,dL2<lL s (9-81)

or
i-i 3 ri-i2-i3

n Jq
ga,,L2 X 3) dL dl2 dL 3
x
(9-82)

These integrals are in forms that are suitable for numerical integration as explained
in Sec 9-9 The resulting sizes of the element stiffness matrices are given in Table
9-4. As in all other previous cases, the necessary and sufficient condition for a
one-to-one mapping is that the sign of the Jacobian be the same for all points
mapped

Table 9-4 Size of the Element Stiffness Matrices for the Tetrahedral Elements

Structural Thermal
Order of clement (3 DOF per node) (1 DOF per node)

Linear x 12
12 4X4
Quadratic 30 x 30 10 X 10
Cubic 60 x 60 20 X 20

9-8 NUMERICAL INTEGRATION: RECTANGULAR AND BRICK

ELEMENTS

As mentioned in Secs. 9-6 and 9-7, the use of numerical integration techniques
makes the isoparametric element practical Integrals such as

/ =
f/W * (9-83)

may be evaluated approximately by writing

 NUMERICAL INTEGRATION RECTANGULAR AND BRICK ELEMENTS 543

fb «

/ = /(a) d\ = 2 u’,/( a,) (9-84)

where the ir, are reterred to as the weights and the .v, as the sampling points.
If the sampling points are chosen such that the interval a < .v < b is divided
into n — 1 equal-length segments, the integration is referred to as Newton-Cotes
quadrature. Quadrature is another name for numerical integration. Familiar ex-
amples of this type of numerical integration are the trapezoidal and Simpson’s rules.
In effect, a polynomial is used on a piecewise basis to represent /(.v) over the

interval. The trapezoidal rule will integrate a linear function exactly, whereas Simp-
son’s rule will integrate a quadratic function exactly. If // sampling points are used,
the integration is exact if/(.v) is a polynomial of order n — 1 or less. This method
is frequently used when the data to be integrated is in tabular form and equally
spaced.
Integrals that arise in the finite element method have integrands that are explicit
functions of the global coordinates. It was shown in Secs. 9-6 and 9-7 how these
integrals could be transformed into ones in the serendipity domains for the rectan-
gular and brick isoparametric elements. Recall, for example, Eqs. (9-34), (9-41),
(9-42), etc. Note that the integration limits are also changed to reflect the fact that

-1 </•< +1, -1 < ,r < +1. etc. Therefore, we need to integrate functions

such as

'f(r)dr (9-85a)
f

f(r,s) dr ds (9-85b)
j J ^

r+] r+i r-i

ils ^ (9-85c)
J j i i i

In Eqs. (9-85), a scalar integrand is implied because the integral of a matrix is

simply the matrix of integrals.

Gauss quadrature a numerical integration method that allows the sampling
is

points to be chosen such that the best possible accuracy may be obtained. If the
sampling points and weights in Eq. (9-84) are based on Legendre polynomials,
then the numerical integration is referred to as Gauss-Legendre quadrature. The
derivation of these weights and sampling points is beyond the scope of this book.
The interested reader may wish to consult reference 5. This method will integrate
polynomials of order In — 1 exactly, where n is the number of sampling points.

Gauss-Legendre quadrature requires the integral to be in the form of Eqs. (9-85),

and in one dimension we have

= = 0-86)
/ f fir) dr Yj wjir,)
J - 1 »=i

where the weights w, and sampling points r ,

are given in Table 9-5 for up to six

sampling points (or n = 6).

 1

544 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Table 9-5 Sampling Point Values and Weights for Gauss-Legendre Quadrature

1" nr) dr =
1=1

r, ip,

n = 1

0 000000000000000 2 000000000000000

n = 2
±0 577350269189626 1 000000000000000

n = 3
0 000000000000000 0.888888888888889
±0 774596669241483 0 555555555555556
n = 4
± 0 339981043584856 0 652145154862546
±0 861136311594053 0 347854845137454
n = 5
0 000000000(KX)000 0 568888888888889
±0 538469310105683 0 478628670499366
±0 906179845938664 0 236926885056189

rt = 6
±0 238619186083197 0 4679 1393457269
±0.661209386466265 0 360761573048139
±0 932469514203152 0 171324492379170

Example 9-2
Evaluate the integral

4
/ = J (r + r) dr
(

by using Gauss-Legendre quadrature Perform an integration of such an order that

the integral is evaluated exactly Compare the result from the numerical evaluation
with that from the analytical integration

Solution

First, we note that the integral is in the proper form for evaluation by Gauss-
Legendre quadrature Second, since the integrand is a polynomial, we can determine
the order of the quadrature that will result in an exact evaluation from

2n - 1 = 4

where the order of the polynomial is 4. Solving for n yields n = Vi, which must
be rounded to give n = 3. Therefore, we must take three sampling points in Table
546 HIGHER ORDER ISOPARAMETRIC CLEMENTS AND QUADRATURE

f =
f t f f i ,
f(r * s >0 dr & dl
p m n

= 2 2 X Wj'Vkf(r„Sj,t
t=ly=li=l
k) (9-93)

where p sampling points are assumed in the t direction. The use of these equations
is demonstrated in Examples 9-3 and 9-4

Example 9-3
Evaluate the integral

4
/ = J (r
2
+ rs)s dr ds
f J )

by using Gauss-Legendre quadrature Compare the result with that from the exact
evaluation

Solution

Because the integrand is a polynomial, we could determine the required number of

sampling points (or Gauss points) in each direction in order to get the exact value
for the integral Note that the integral is of order 2 in r and of order 5 tn s. Therefore,
in the r direction we have 2n - I = 2 or n = Vi and in the s direction 2m - 1

« 5 or m = 3 Hence we take n = 2 3
( /j rounds to 2) and m = 3, or two Gauss
points in the r direction, and three points in the s direction It is instructive to show
the six sampling points (i e ,
2 x 3) on a typical element as shown in Fig 9-18.

0 57735 0 57735

Figure 9-18 Quadrilateral element in Example 9-3 Note that denotes a Gauss (or
sampling) point
 NUMERICAL INTEGRATION RECTANGULAR AND BRICK ELEMENTS 547

Let us denote the integrand by f(r,s), where

4
f(r,s) = (r
2
+ rs)s

The calculations are summarized (to five significant digits) in the following table:

Point r, sj w ,
/O',, 5j)

A 0.57735 -0.77460 1 00000 0 55556 -0.04100 -0.02278

B 0.57735 0.00000 1.00000 0.88889 0 00000 0.00000
C 0.57735 + 0.77460 1 00000 0.55556 0 21800 0.15611
D -0.57735 + 0.77460 1.00000 0.55556 -0 04100 -0.02278
E -0.57735 0.00000 1.00000 0.88889 0.00000 0.00000
F -0.57735 -0.77460 1.00000 0 55556 0.28100 0.15611

2 = 0 26666

This compares favorably with the exact evaluation of / = Vis or / = 0.26667 to

five significant digits. The exact result is approached as the number of significant
digits used is increased. D
Example 9-4
Evaluate the element stiffness (or conductance) matrix for heat conduction in the
x direction for the element shown in Fig. 9-19, where the nodal coordinates are

shown in inches. Assume a thermal conductivity k of 1 Btu/hr-in-°F and an element

thickness f of 1 in.

Solution

From Eq. (8-106a), the expression for the stiffness matrix from heat conduction in
the x direction is given by

K" =f
f —kt — dxdx (8-1 06a)
J.v 0a- dx

15 2.4 3 )

Example 9-4.
Figure 9-19 Quadrilateral element in
548 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

For the element under consideration > the shape function matrix N is given by

Ml + r)(I - s)

'/.(I + r)(l + s)
'A(l - r)(l + s)
K(1 - r)(l - s)

Note that the four nodes are being used for two purposes; (I) to define the element
geometry, and (2) to specify the locations where the values of the parameter function
(in this case the temperature) are sought Hence, the isoparametric formulation from
Sec. 9-6 is directly applicable here. Recall that the Jacobian matrix J is given by
Eq (9-33) for this element or

Note that four terms are included in the summations because there are four nodes
-1
(per element) For convenience, let us denote the entries comprising J as An ,

^ 2 ), and /\ 2 j, or

The Ay’s are functions of r and s From Eq. (9-31) we may write

BN, BN, BN,

= A
97 "l?
+ A{I
H
If all four shape functions are considered at one time, we have

5N aN aN
^ii + A 12
dx dr as

from which it also follows that

r
9N\ 9N r arf
~ An + A
,

' 2
at ~dr ds

Finally, we also have

dx dy ~ |det J| di ds

and may write

k;
 NUMERICAL INTEGRATION RECTANGULAR AND BRICK ELEMENTS 549

Note that the integrand is a function of r and s only. In other words, the original
quadrilateral element has been mapped into the corresponding undistorted (parent)
element. The global coordinates (.v.>) have been eliminated m favor of the seren-
dipity coordinates (r,s).
Let us denote the integrand by H(r,s) or

H(r,s) = T
c ktc |det J|

where

dN dN
c ~ ^11 -
+ ^12 ,
dr ds

Clearly, we need the derivatives of the shape functions with respect to r and s, or

T
'dN,'
dr
dN2 54(1
dN dr 14 (

dr dN2 54 (

dr 14( 1

dN4
. dr _

and
~ dN,~ T

ds
dN: ”-54(1
dN ds 14(1

ds d/V3 54(1

ds -54(1
d Na
ds

The Jacobian matrix J has four entries denoted J\\, J i 2> ^2 i> anc * ^22 From Eq
(9-33), these are given by

j -y 3 3
K
1
"fit, Dr

- - + - '
/j(l ~
= 54(1 s)x J + 14(1 + s)x2 54(1 s)*3

_
Jn — V }’j
a
dr
]=\

= 54(1 - s)y, + 54(1 + s)y2 - 'Ml + J )>’3

_ W ~ s ^’*

= i
j~ J
-
 J J J
552 HIGHFR ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

det J = 0 01197169

7.47043843 - 5.05912314"
4.17652052 8.35304105
r
0 10566243' "- 0.1 0566243'
dN 0 39433757 dN _ 0.10566243
dr -0 39433757 ds 0 39433757
_-0. 10566243, _ -0.39433757,

1 32390396

dN dN~ 2 41131526
c +
dr -4.94087683
1 20565761.

H(r,,s2) = crtoc|detj|

0 02098304 0 03821782 -0.07830976 0.01910891

0 06960868 -0 14263083 0 03480434
0 29225601 -0.07131541
(Symmetric) 0 01740217

Gauss point D

r, = -0 57735027
s, = -0 57735027
0 10000000 0 08943376]
- 0 05000000 0 08943376
det J = 0 01341506

6 66666667 - 6 66666667
.

]
72715342 7
[3 45430684
r
0.39433757“ 0.10566243
dN 0. 10566243 dN 0.10566243
dr -0 10566243 ds 0.39433757
L-0 39433757 L- 0.39433757
3 33333333
, _ l\ 3N + dNl
=
0 00000000
'

“
h Sr
1 A|2
'i7J -3.33333333
0.00000000,
 NUMERICAL INTEGRATION. TRIANGULAR AND TETRAHEDRAL
ELEMENTS 553

H(r,,s,) = r
c Avcjdet Jj

0.14905627 0.00000000 -0.14905627 0.00000000'

0.00000000 0.00000000 0.00000000
0.14905627 0.00000000
(Symmetric) 0.00000000
These results may now be used to obtain the stiffness matrix for this element
by noting that

dNT 4N + +
K “ _~
'

f J f 1
f
i h ,
dX (y ~ J-J-, H(r ’ s) dr ds
dx
2 2

= 22
<=i =i y
H(r,,.v )

where each weight (i.e., w, and wy ) has a value of unity (see Table 9-5 for n = 2).
The final result is given by

0.60330578 -0.12396694 -0.41735537 -0.06198347

0.14876033 -0.09917356 0.07438017
Kf
0.56611571 -0.04958677
(Symmetric) 0.03719008

It is not possible to compare this result with that from an exact, analytical integration.
Why not?

9-9 NUMERICAL INTEGRATION: TRIANGULAR AND TETRAHEDRAL

ELEMENTS

In Sec. 6-7 three special integration formulas were presented that could be used to
evaluate integrals for the lineal, triangular, and tetrahedral elements. These for-
mulas, given by Eqs. (6-48) to (6-50) were illustrated many times in Chapters 7
and 8. Although these formulas are applicable to the higher-order lineal, triangular,
and tetrahedral elements, they are not very practical in these situations. Instead,

we resort to numerical integration.

Recall from Secs. 9-6 and 9-7 that integrals with rather complicated integrands
arose when higher-order triangular and tetrahedral elements were considered. For
example, typical element stiffness matrices and nodal force vectors are of the forms
given by Eqs. (9-54), (9-55), (9-81), and (9-82). Analytical evaluation of such
integrals is impossible, and so we must resort to a numerical scheme. Formulas
derived by Hammer, Marlowe, and Stroud [6] are particularly useful and easy to

apply. The use of these formulas for the case of the triangle is based on the
approximation given by
554 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

U
K' = JT£ H dL, dL2 = 2 (9-94)

where iv, denotes the ith weight and L u Z^„ and Z. 3( denote the coordinates of the
,

ith sampling points. Note that n sampling points are assumed. A similar formula

holds for the tetrahedral element, or

n ri-ij n-Lz-Li
K' =
Jo J Q jo
H dl x
dL 2 dL 3

= 2 Kw,Hf.L,„L1 „Li,M 19-951

Note that the dependent area coordinate Z. 3 and the dependent volume coordinate
La were reintroduced into the integrands in Eqs. (9-94) and (9-95). The sampling
points and weights are given in Figs. 9-21 and 9-22 for the triangular and tetrahedral
elements, respectively The use of these formulas is illustrated in Examples 9-5
and 9-6

Example 9-5
By using Eq (9-94) and Fig 9-21 , evaluate the following element stiffness matrix
for the three-node triangular element

K‘„ = j^AN dx dy
Perform a linear-order integration Recall that this element stiffness matrix arose
in Sec 8-8 and resulted from convection from the lateral surface(s) of a plate [see

Eq. (8- 106c)] The exact integration is given in Example 8-13 and will serve as a
check on the numerical integrations

Solution

In terms of the area coordinates, the shape function matnx N for this element is
given by

N = [Z., L2 L3 ]

Therefore, the integrand H (Z^Xj.Z*) is given by

L x

H(Z.„Z.2 ,Z. 3 ) li h[L x

L2 Z.3 ] |det J|

UaJ
The determinant of From Eq. (9-50)
the Jacobian matnx J needs to be computed.
we get
 NUMERICAL INTEGRATION- TRIANGULAR AND TETRAHEDRAL ELEMENTS
555

Area
Order Points coordinates Weights
L, L7 L3 w,

1/3

- 27/48

25/48

0.225000000

0.132394153

0 125939180

Figure 9-21 Numerical integration formulas for triangles.

-1
556 HIGHER-ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

1/4 1/4 1/4 1/4

Figure 9-22 Numerical integration formulas for tetrahedra

and from Eq. (9-51), we get

,

aw, /a<v,\ /a 11
o o ©
dL 2 '
L3/L,J,2
aw2 (m \2 fiHi) -i-o.i
8L 2 Wi/lu, V
m 3
= (^) Ji ^ =o-i = -
dL 2 ^>3/ L, J.2
 NUMERICAL INTEGRATION. TRIANGULAR AND TETRAHEDRAL ELEMENTS
557

Let us denote the nodal coordinates as Cr,.?,). (.t and (x3 ,y 3 ). Therefore, from
2 ,y 2 ),
the above results and Eq. (9-52), we have

.r, - x3 -
J = 3’ I >3
.v 2 - x3 3’2 ~ >’3

and

det J — (jt, x3 )( v2 v3 ) (x2 — x3 )(yi — y3 )

= -HJ ’2 ~ J3 3'2 “ ~ + x3 y + = 2A
-hi’3 *>J’i ,
.t
2 y3
where A is the area of the triangle. If the reader is not convinced the above expression

is 2A, the expression should be compared with Eq. (6-21e). Therefore, the integrand
H is given by

u
H(L,,L2,L 3 ) = 2/zA l2
.cj
L] l,l2 L,L
= 2/zA L2 L i
Ll LiL
L3L2 L\

For the linear-order integration, only one sampling point and one weight are given
in Fig. 9-21. Note that the sampling point in this case corresponds to the centroid
of the triangle and the weight is unity. Therefore, we have
i

K% = X
<= i
'/2H’H Z*,, L 3i )

2 (’/3)('/3)
C/3) (>/3)(>/3)

= Vi2hA C/3)( /3)

!
C/3)
2
(’/3)(>/3)
2
('/3)('/3) C/3)('/3) C/3)

1 1 1
hA
1 1 1

9
1 1 1

This result differs somewhat from that obtained in Example 8-13 from an exact
integration. However, the error on the values of the nodal unknowns (i.e.. the
vector a in Ka = f) can be reduced by increasing the number of elements used.

Example 9-6
Redo Example 9-5 by performing a quadratic-order quadrature. Compare the re-

sulting stiffness matrix with that from Example 8-13, where an exact integration
was performed.
558 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Solution

The first part of Example 9-5 is applicable here, and so we may begin with
'
L] L\L 2 L]Z-3
H = 2hA LqLj l2 l2 l2
L 2L {
L\ _

Note that the quadratic-order integration requires three sampling points (and three
weights). If the sampling points and weights in Fig 9-21 are applied to the problem
at hand, the result is

‘
m 2
mm) c/ml
WA)ahA) (W/S)
»)('/,)
mm :
i

(0) J
2
e/o (Vi)(0) m>m>
+ W(V5)(2 hA) (0)(H) (0)’- (OKU)
_mm c/i)<0) mf .

r 2
(0) tom) (0)0-4)'
+ H(H)(2 hA) (H)(0) m> 2 mm)
.(H)(0) mm IV,)
2
.

This last result is the same as that in Example 8-13, where an exact integration

was performed Note that the integrand was of quadratic order and hence a quadratic
order numerical integration gave the exact result. B

It should be noted that the Jacobian matrix is a matrix composed solely of

constants for the linear-order (or three- node) triangular element. For higher-order
elements this will not necessarily be the case In other words, the Jacobian matrix
in general will be a function of the three area coordinates, in addition to the nodal
coordinates This presents no problem because the quadrature method presented
here can easily accommodate this

9-10 NUMERICAL INTEGRATION: REQUIRED ORDER

The question as to what order of integration we should use naturally arises. Zien-
kiewicz 17] argues that the linear-order integrations will always be convergent, but
not always practical. The general guidelines may be summarized as follows. For
linear-order triangles and quadrilaterals, single-point integration suffices. For quad-
 PROBLEMS 559

integrations are adequate. For quadratic-order triangular and tetrahedral elements,

we should use three-point and four-point formulas from Figs. 9-21 and 9-22 (i.e.,
quadratic-order integrations). Cubic-order integrations should generally be per-
formed on cubic-order elements. This implies and 3x3
Gauss point 3x3x3
integrations from Table 9-5 for the cubic-order quadrilateral and brick elements,
and four-point and five-point formulas from Figs. 9-21 and 9-22 for the cubic-order
triangular and tetrahedral elements. Zienkiewicz’s book [7] should be consulted for
more information on this important aspect of Finite element analysis.

REFERENCES

1. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp. 253-261.
2. Kaplan, W., Advanced Calculus , 2nd ed., Addison-Wesley, Reading, Mass., 1973,
pp. 269-274.
3. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill, (UK) London, 1977,
p. 186.
4. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill, (UK) London, 1977,
p. 192.
5. Carnahan, B., H. A. Luther, and J. O. Wilkes, Applied Numerical Methods, Wiley,
NewYork, 1969, pp. 100-116.
6. Hammer, P. C., O. P. Marlowe, and A. H. Stroud, “Numerical Integration over Sim-
plexes and Cones,” Math. Tables Aids Comp., Vol. 10., pp. 130-137, 1956.
7. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp. 201-204.

PROBLEMS

9-1 Show that the shape function A, evaluates to unity at node 1 and to zero at all other

nodes for the linear-, quadratic-, and cubic-order lineal element if the shape functions

are given in terms of

a. Length coordinates L andx

Lz
b. Serendipity coordinate r

9-2 Show that the shape function N z evaluates to unity at node 2 and to zero at all other

nodes for the linear-, quadratic-, and cubic-order lineal element if the shape functions
are given in terms of

a. Length coordinates L and

x
C,
b. Serendipity coordinate r

9-3 Show that the shape function iV 3 evaluates to unity at node 3 and to zero at all other

nodes for the quadratic- and cubic-order lineal element if the shape functions are
given in terms of

a. Length coordinates L and Lz

{

b. Serendipity coordinate r
S60 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

9-4 Show that the shape function evaluates to unity at node 4 and to zero at all other
nodes for the cubic-order lineal element if the shape functions are given in terms of

a. Length coordinates L and L 2

l

b. Serendipity coordinate r

9-5 Plot the shape functions for the lineal element on a typical element if the clement is

of
a. Linear order
b. Quadratic order
c. Cubic order

9-6 For the quadratic-order triangular element, show that N 2 evaluates to unity at node
2 Also show that N 2 evaluates to zero if evaluated at each of the five other nodes

9-7 For the quadratic-order triangular element, show that N b evaluates to unity at node
6 Also show that evaluates to zero if evaluated at each of the five other nodes.

9-8 For the cubic-order triangular element, show that N 2 evaluates to unity at node 3
Also show that N2 evaluates to zero if evaluated at each of the nine other nodes

9-9 For the cubic-order triangular element, show that N-, evaluates to unity at node 7.

Also show that N 2 evaluates to zero if evaluated at each of the nine other nodes

9-10 For the cubic order triangular element, show that N i0 evaluates to unity at node 10.
Also show that N )0 evaluates to zero if evaluated at each of the nine other nodes

9-11 For the quadratic-order rectangular element, show that N 2 evaluates to unity at node
3 Also show that /V, evaluates to zero if evaluated at each of the seven other nodes

9-12 For the quadratic-order rectangular element, show that Nt evaluates to unity at node
6 Also show that Nb evaluates to zero if evaluated at each of the seven other nodes

9-13 For the cubic-order rectangular element, show that N 2 evaluates to unity at node 2.
Also show that N 2 evaluates to zero if evaluated at each of the 1 1 other nodes

9-14 For the cubic-order rectangular element, show that N l0 evaluates to unity at node 10
Also show that N lQ evaluates to zero if evaluated at each of the 1 1 other nodes

9-15 For the quadratic-order tetrahedral element, show that N 2 evaluates to unity at node
2 Also show that N2 evaluates to zero if evaluated at each of the nine other nodes

9-16 For the quadratic-order tetrahedral element, show that Ng evaluates to unity at node
8 Also show that Nt evaluates to zero if evaluated at each of the nine other nodes.

9-17 N
For the cubic-order tetrahedral element, show that u evaluates to unity at node 1

Also show that N\] evaluates to zero if evaluated at each of the 19 other nodes

9-18 For the cubic-order tetrahedral clement, show that N i9 evaluates to unity at node 19
Also show that N i9 evaluates to zero if evaluated at each of the 19 other nodes

9-19 For the quadratic-order brick element, show that Ns evaluates to unity at node 5.

Also show that N s evaluates to zero if evaluated at each of the 19 other nodes

9-20 For the quadratic-order brick element, show that Nu evaluates to unity at node 14.

Also show that iV ]4 evaluates to zero if evaluated at each of the 19 other nodes.
 .

PROBLEMS 561

9-21 For the cubic-order brick eiement, show that evaluates to unity at node 4, Also
show that A'. evaluates to zero if evaluated at each of the 31 other nodes.

9-22 For the cubic-order brick element, show that N29 evaluates to unity at node 29. Also
show that A 29
'
evaluates to zero if evaluated at each of the 31 other nodes.

9-23 Consider the linear-order, isoparametric quadrilateral element shown in Fig. P9-23.
The nodal coordinates are shown on the figure. Map the point r = + 0.50 and s =
— 1.0 on the parent element to the proper point on the distorted element (i.e., the
quadrilateral). Show the distorted clement on graph paper in order to see the mapping.

9-24 For the element in Problem 9-23, map the point r = +0.5 and s — —0.75 on the
parent element to the proper point on the distorted element (i.e., the quadrilateral).

Show the distorted element on graph paper in order to see the mapping.

9-25 Consider the quadratic-order, isoparametric quadrilateral element shown in Fig.

P9-25. The nodal coordinates are shown on the figure. Map the point r = —0.35
and s = + 1 .0 on the parent element to the proper point on the distorted element.
Show the distorted element on graph paper in order to see the mapping.

9-26 For the element in Problem 9-25, map the point r = +0.75 and s — —0.25 on the
parent element to the proper point on the distorted element. Show the distorted element
on graph paper in order to see the mapping.

9-27 Consider the cubic-order, isoparametric quadrilateral element shown in Fig. P9-21
The nodal coordinates are shown on the figure. Map the point r = +0.50 and s =
— .0 on the parent element to the proper point on the distorted element Show the
1

distorted element on graph paper in order to see the mapping.

562 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Figure P9-27

9-28 For the element in Problem 9-27, map the point r - -0 45 and s = +0 75 on the
parent element to the proper point on the distorted element Show the distorted element
on graph paper in order to see the mapping

9-29 Consider the quadratic order, isoparametric triangular element shown in Fig P9-29
~
The nodal coordinates are shown on the figure Map the point L\ ~ 0 5 and Li
0 5 on the parent element to the proper point on the distorted element Show the
 PROBLEMS 563

Figure P9-29

distorted element on graph paper in order to see the mapping. What is the value of
Please explain.

9-30 For the element in Problem 9-29, map the point L* = 0.25 and L } - 0 35 on the
parent element to the proper point on the distorted element. Show the distorted element
on graph paper in order to see the mapping. What is the value of L2 ? Please explain.

9-31 Consider the cubic-order, isoparametric triangular element shown in Fig. P9-31 The
.

nodal coordinates are shown on the figure. Map the point L2 = 0.5 and L3 = 0.5
on the parent element to the proper point on the distorted element. Show the distorted
element on graph paper in order to see the mapping. What is the value of L,? Please
explain.

Figure P9-31
S64 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

9-32 For the element in Problem 9-31, map the point L> ~ Vi and Lj — Vi on the parent
element to the proper point on the distorted element Show the distorted element on
graph paper in order to see the mapping. What is the value of L{1 Please explain

9-33 For the isoparametric, quadrilateral element in Fig. P9-33, determine the Jacobian
matnx

Figure P9-33

9-34 Determine the Jacobian matrix for the isoparametric, quadrilateral element shown in

Fig P9-34

Figure P9-34

heat
9-35 Recall from Sec 8-3 that the element stiffness matrix from one-dimensional
conduction is given by
 PROBLEMS 565

«- L
— dS T
dx
IA
kA —
dN
dx
,
dx

where k is the thermal conductivity, A is the cross-sectional area (of the one-dimen-
sional body in the direction x), N is the shape function matrix, and l
e
denotes that
the integration is to be performed over the length of the element. Let us consider a
quadratic-order, lineal element where the nodes are not necessarily equally spaced.
In order to evaluate Kf for this element, we must first perform an isoparametric
mapping.

a. Show that the Jacobian matrix in this case is a scalar and is given by

dx dNj
= =
J
Jr ^ rfr
X’

How many terms are included in the summation? Please explain.

b. Show that Kf may be written

</N r kA dN
K; - dr
r: dr v dN ,
dr
2 ~P X ‘

c. How are the results from parts (a) and (b) affected if the cubic-order, lineal element
is used?

9-36 Consider a one-dimensional body in a two-dimensional space such as the one shown
in Fig. P9-36(a). The body may be discretized as shown in Fig. P9-36(b). Note that
lineal elments of any orderbe used [although Fig. P9-36(b) shows only linear-
may
order elements]. Note further that two global coordinates must be used to define the
location of each node. Let us denote the global coordinates of the fth node as (x,,y,).
Let us also use the coordinate / as shown in Fig. P9-36(c) to represent the direction
that is always tangential to the one-dimensional body. In other words, the coordinate
/ is measured along the body. It then follows that

N, = N,(r)

r = r(l)

and
/ = Kx.y)

where r is the serendipity coordinate, and N, is the shape function for node i.

a. Show that the element stiffness matrix from heat conduction in the / direction

(i.e., from one-dimensional heat conduction along the length of the body) is given
by

K = l —- kAA
'
d
dNT
/
1 —
dN
d/
HI
dt

where k is the thermal conductivity, A is the cross-sectional area (in the l direction),

and l
e
denotes that the integration is to be performed over the length of the element.
 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

fc)

Figure P9-36

b. Show that the Jacobian matrix is a scalar in this case and is given by

c. Show that Kf becomes

d. Extend the results from parts (b) and (c) to the case of a one-dimensional body
in a three-dimensional space

9-37 For the isoparametric, triangular element in Fig P9-37, determine the Jacobian ma-
trix.

9-38 Determine the Jacobian matrix for the isoparametric, triangular element shown in

Fig. P9-38

9-39 Derive the expressions that correspond to Eqs (9-39) and (9-40) for the isoparametric,
triangular element In other words, derive the expressions to be used in the trans-
 /

PROBLEMS 567

Figure P9-37

Figure P9-38

formation of dS (the elemental surface area around the boundary of the two-dimen-
sional element) to the local coordinate system (in terms of the area coordinates). Note
that the area coordinates degenerate to length coordinates in this case. Why?
9-40 Compute the following derivatives for the quadratic-order, isoparametric triangular
element:

BN2 BN s
3. o.
6L, ac,

9-41 Compute the following derivatives for the quadratic-order, isoparametric triangular
element:

a. —3Lj
fa.
^
BLn
568 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

9-42 Compute the following derivatives for the cubic-order, isoparametric triangular ele-
ment.

a. 77" b.
dLj

9-43 Derive the expressions that correspond to Eq (9-65) for the isoparametric brick
element if faces of constant r (i e , r = ±1) happen to be on the global boundary.
In other words, derive the expression to be used in the transformation of dS on faces
of constant r (the elemental surface area on the boundary of the three-dimensional
element) to the local coordinate system (in terms of the serendipity coordinates s and
o.

9-44 Derive the expressions that correspond lo Eq (9-65) for the isoparametric brick
element if faces of constant s (i e , s = ± 1) happen to be on the global boundary.
In other words, derive the expression to be used m the transformation of dS on faces
of constant s (the elemental surface area on the boundary of the three-dimensional
element) to the local coordinate system (in terms of the serendipity coordinates r and
t)

9-45 Derive the expressions that correspond to Eq (9-65) for the isoparametric tetrahedral
element In other words, derive the expression to be used in the transformation of
dS (the elemental surface area on the boundary of the three-dimensional element) to
the local coordinate system (in terms of the volume coordinates) Note that the volume
coordinates degenerate to area coordinates in this case Why 7
9-46 Compute the following derivatives for the quadratic-order, isoparametric tetrahedral
element

cWg
b.
dL 2

9-47 Compute the following derivatives for the cubic order, isoparametric tetrahedral ele-
ment

a. —
aw42 .
b. —
dN, 7
di, ah
9-48 Evaluate the integral given below by using Gauss-Legendre quadrature of such an
order that the integral is evaluated exactly. Compare this result with that from the
analytical evaluation

3
/ = j
'

(r + 3r ) dr
i

9-49 Evaluate the integral given below by using Gauss-Legendre quadrature of such an
order that the integral is evaluated exactly Compare this result with that from the
analytical evaluation

'

3
/ = | (2 + 3r 4 ) dr
 PROBLEMS 569

9-50 Evaluate the integral given below by using two-point

Guass-Legendre quadrature.
Compare this result with that from the analytical evaluation.

+ J

f
/ = I sin
2
nr dr

9-51 Repeat Problem 9-50 by performing a three-point Gauss-Legendre quadrature.

9-52 Evaluate the integral given below by using two-point Gauss-Legendre quadrature.
Compare this result with that from the analytical evaluation.

/ = j cos 2 nr dr
^

9-53 Repeat Problem 9-52 by performing a three-point Gauss-Legendre quadrature.

9-54 Evaluate the integral given below by using Gauss-Legendre quadrature of such an
order that the integral is evaluated exactly. Compare this result with that from the
analytical evaluation.

f = (2rs + 3r V) dr ds
f : f !

9-55 Evaluate the integral given below by using Gauss-Legendre quadrature of such an
order that the integral is evaluated exactly. Compare this result with that from the
analytical evaluation.

3 2
/ = f (2 s + 5rs 3 ) dr ds
j J t

9-56 Evaluate the element nodal force vector from a uniform body force b acting on a
two-dimensional body if the four-node rectangular element with length 2a and height
2b used [see Eq. (7-28)J. Perform a 2 x 2 Gauss-Legendre quadrature. Give the
is

result in terms of the two components of the body force (per unit volume) bx and h,,

the element thickness /, and the element dimensions a and b.

9-57 Evaluate the element nodal force vector from a uniform surface traction s acting on
leg 2-3 of the four-node rectangular element with length 2 a and height 2b [see Eq.
(7-31). Use only two Gauss points. Give the result in terms of the two components
of the surface traction s x and i„ the element thickness t, and the element dimensions
a and b.

9-58 Evaluate the element stiffness matrix from conduction in the ,r direction [given by
Eq. (8-106a)[ for the four-node rectangular element with length 2 a and height 2b by
performing 2x2 Gaussian quadrature. Give the result in terms of the element
dimensions a and b, the thermal conductivity k, and the element thickness 1. Hint:

See Problem 8-57 [parts (a) and (b)].

9-59 Evaluate the element stiffness matrix from conduction the y direction [given by m
Eq. (8-106b)[ for the four-node rectangular element with length 2a and height 2b by
performing 2x2 Gaussian quadrature. Give the result in terms of the element
570 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

dimensions a and b, the thermal conductivity k, and the element thickness t. Hint'
Sec Problem 8-57 [parts (a) and (b)J

9-60 Evaluate the clement stiffness matrix from lateral convection given by Eq (8-I06c)
for the four-node rectangular element with length 2a and height 2b by performing
2x2 Gaussian quadrature Give the result in terms of the element dimensions a
and b, and the convective heat transfer coefficient h

9-61 Evaluate the element stiffness matrix from boundary convection given by Eq.
(8-I06e) for the four-node rectangular clement with length 2a and height 2b by
assuming two Gauss points Assume also that leg 1-2 is on the global boundary (and
undergoes convection) Give the results in terms of the element dimensions a and b,
the convective heat transfer coefficient h B and the element thickness l.
,

9-62 Evaluate the clement stiffness matrix from boundary convection given by Eq
(8-106c) for the four-node rectangular element with length 2 a and height 2b by
assuming two Gauss points Assume also that leg 4-1 is on the global boundary (and
undergoes convection) Give the results in terms of the element dimensions a and b,
the convective heat transfer coefficient b B , and the element thickness t

9-63 Evaluate the element nodal force vector from lateral convection given by Eq
(8- 107a) for the four-node rectangular element with length 2 a and height 2b by
performing 2x2 Gaussian quadrature Give the result in terms of the element
dimensions a and b, the convective heat transfer coefficient h, and the ambient
temperature Ta
9-64 Evaluate the element nodal force vector from a lateral heat flux given by Eq
(8- 107c) for the four-node rectangular clement with length 2a and height 2b by
performing 2x2 Gaussian quadrature Give the result in terms of the element
dimensions a and b, and the imposed heat flux qs

9-65 Evaluate the element nodal force vector from a distributed heat source given by
Eq (8-107d) for the four-node rectangular element with length 2a and height 2b by
performing 2x2 Gaussian quadrature Give the result in terms of the element
dimensions a and b, the heat source strength Q . and the element thickness I.

9-66 Evaluate the clement nodal force vector from boundary convection given by Eq
(8-107e) for the four-node rectangular element with length 2 a and height 2b by
assuming two Gauss points. Also assume that leg 2-3 is the global boundary (and
undergoes convection) Give the result in terms of the element dimensions a and b,
the convective heat transfer coefficient hB , the ambient temperature T^, and the

clement thickness r

9-67 Evaluate the element nodal force vector from a heat flux imposed on leg 3-4 of the
four-node rectangular element with length 2a and height 2b by assuming two Gauss
points See Eq (8- 1 07g) Give the result in terms of the element dimensions a and
b, the heat flux q,B , and the clement thickness t.

9-68 Evaluate the clement nodal force vector from a uniform body force b acting on a
two-dimensional body if the three-node triangular element is used [see Eq (7-28)]

Perform a quadratic order quadrature Give the result in terms of the two components
of the body force (per unit volume) bf and br the dement thickness t, and the element
,

area A.
 PROBLEMS 571

9-69 Evaluate the element nodal force vector from a uniform surface traction s acting on
leg 2-3 of the three-node triangular element. Use only two Gauss points. Give
the
result in terms of the two components of the surface traction ,v and s x . the element
t

thicknessr, and the length of leg 2-3. Hint. Note that h\ is zero on leg 2-3. Replace
N and AS on leg 2-3 with the serendipity form of the shape functions for the
z linear
element and then use Gauss-Legendre quadrature

9-70 Evaluate the clement stiffness matrix from lateral convection given by Eq. (8- 1 06c)
for the three-node triangular element by performing a quadratic order numerical
integration (quadrature). Give the result in terms of the element area A and the
convective heat transfer coefficient It.

9-71 Evaluate the element nodal force vector from lateral convection given by Eq.

(8- 107a) for the three-node triangular element by performing a quadratic-order nu-
merical integration (quadrature). Give the result in terms of the element area A. the
convective heat transfer coefficient h, and the ambient temperature Ta .

9-72 Evaluate the element nodal force vector from a lateral heat flux given by Eq.
(8-107c) for the thTec-node triangular element by performing a quadratic-older nu-
merical integration (quadrature). Give the result m terms of the element area A and
the imposed heat flux q s .

9-73 Evaluate the element nodal force vector from a distributed heat source given by
Eq. (8-I07d) for the three-node triangular element by performing a quadratic-order
numerical integration (quadrature). Give the result in terms of the element area A ,

the heat source strength Q, and the element thickness t.

9-74 Reconsider the element in Example 9-4. Determine the element stiffness (or con-

ductance) matrix from conduction in the v direction [see Eq. (8-106b)]. Assume
2x2 Gaussian quadrature. Take the thermal conductivity k to be 1 Btu/hr-in-°F and
the element thickness to be 1 in.

9-75 Reconsider the element in Example 9-4. Determine the element nodal force vector

from lateral convection [see Eq. (8- 107a)]. Assume 2x2 Gaussian quadrature. Take
2
the convective heat transfer coefficient h to be 4 Btu/hr-in -°F on each side of the
plate and the ambient temperature to be 70°F (the same on both sides).

9-76 Reconsider the element in Example 9-4. Determine the element nodal force vector

from a imposed heat flux [see Eq. (8-1 07c)]. Assume

laterally Gaussian 2x2
2
quadrature. Take the heat flux q s to be 100 Btu/hr-in .

9-77 Reconsider the element in Example 9-4. Determine the element nodal force vector

from an imposed heat flux q s $ acting on leg 2-3 of the element [see Eq. (8-107g)].
2
Assume two Gauss points and a heat flux of 75 Btu/hr-in .

9-78 Consider the element shown in Fig. P9-78 where the nodal coordinates are shown
in centimeters. Determine the element stiffness (or conductance) matrix from con-
duction in the x direction [see Eq. (8- 106a)]. Assume 2x2
Gaussian quadrature.

Take the thermal conductivity k to be 150 W/m-°C and the element thickness to be
2 cm.

9-79 Consider the element from Problem 9-78. Determine the element stiffness (or con-
(8- 1 06b)] Assume
ductance) matrix from conduction in the y direction [see Eq. .
 &
572 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

Figure P9-78

2x2 Gaussian quadrature Take the thermal conductivity k to be 150 W/m-°C and
the element thickness to be 2 cm

9-80 Consider the element from Problem 9-78 Determine the clement nodal force vector
from lateral convection | see Eq (8- 107a)] Assume 2 x 2 Gaussian quadrature Take
the convective heat transfer coefficient h to be 2000 W/nr-°C on each side of the
plate and the ambient temperature to be 48°C (the same on both sides).

9-81 Consider the clement from Problem 9-78 Determine the element nodal force vector
from a laterally imposed heat flux |see Eq (8-l07c)] Assume 2x2 Gaussian
quadrature Take the heat flux qi to be 100 W/cm 2
9-82 Consider the element from Problem 9-78 Determine the element nodal force vector
from an imposed heat flux q of 125 W/cnv acting on leg 3-4 of the element (see
Eq (8-107g)J Assume two Gauss points

9-83 Consider the problem posed in Problem 9-36 An alternate, but completely equivalent,
formulation to such a problem is developed here The main reason for the alternate
formulation given below is that it may be more readily extended to the case of a
two-dimensional body in a three-dimensional space (see Problem 9-84)
Consider a one-dimensional body in a two-dimensional space such as the one
shown in Fig P9-83(a) discretized as shown in Fig. P9-83(b)
The body may be
Note any order may be used (although Fig. P9-83(b) shows
that lineal elements of

only linear-order elements] Note further that two global coordinates must be used
to define the location of each node Let us denote the global coordinates of thejth
node as (t,a,). Let us also use the coordinate*' as shown in Fig P9-83(c) to represent
the direction that is always tangential to the one-dimensional body. In other words,
the coordinate x‘ is measured along the body It then follows that

N, = N,(r)

r = r(x')
 problems 573

Figure P9-83

and

x' - x'(x,y)

where r is the serendipity coordinate, and N, is the shape function for node i.

a. Show that the element stiffness matrix from heat conduction in the x' direction
(i.e., from one-dimensiona! heat conduction along the length of the body) is given
by

dN r ,
dN , ,
K'
-L 77
dxI
'/<
77
dx
ax

where k is the thermal conductivity, A is the cross-sectional area (in the x' di-
e
rection), and l denotes that the integration is to be performed over the length of
the element.

b. Show that the Jacobian matrix is a scalar in this case and is now given by

dx’ ^ dN. ^ dN.

574 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

where rin is the cosine of the angle between the x and x' axes, and n 2 j is the
cosine of the angle between the > and the x' axes 0 e.» « n and « 2 | are the two
direction cosines)
c. Show that K' becomes
+,
f dlf kA dN
J-i dr — Wj dr
"'2 +
*x '

Note that the integral may be evaluated numerically with the help of Gauss-
Legendre quadrature
d. Extend the results from parts (b) and (c) to the case of a one-dimensional body
in a //iree-dimcnsional space
e. Show that formulation above is equivalent to that in Problem 9-36 by showing
that the Jacobian J from Problem 9-36 is equivalent to the expression for J given
above

9-84 Let us extend the formulation presented in Problem 9-83 to the case of a two-
dimensional body in a three-dimensional space as shown in Fig P9-84(a) The body
may be discretized into quadrilateral elements as shown m Fig P9-84(b). Note that
quadrilateral elements of any order may be used [although Fig P9-84(b) shows only
linear-order elements] Note further that three global coordinates must be used to
define the location of each node Let us denote the global coordinates of the jth node
as (Xj,yj, 2j ) Let us also use the coordinates x' and y' as shown in Fig P9-84 to

represent the local coordinate system that is in the plane of the two-dimensional body
Note that x' and j' are always perpendicular to each other (in other words, the local
coordinate system and the global coordinate system both form orthogonal coordinate
systems) It then follows that

N, = N,(r,s)

r = r(x',y')

s = s(x\>’)

where

x‘ = x'(x.y,2)

and

/ = y'(*,y.?)

Note that r and s are the serendipity coordinates, and N, is the shape function for
node i It follows that the element stiffness matrices for conduction in the x' and y
directions are given by

f dtp. aN
K" “
jm -T-r^TT
dy dv
^ dy
 PROBLEMS 575

Figure P9-84

where k is the thermal conductivity, t is the thickness of the element, and Ae denotes
that the integration is to be performed over the area of the element,
a. Show that the Jacobian matrix is now given by

At' dy'

dr dr

df
ds ds

where the entries in the Jacobian matrix are given by

dx' dhJj „ dN, ^ dNj

dr dr •
dr

7, 2 = -=
dy '
^ dN. ^ BN,
+ ''*
v
2,^,
dNj
576 HIGHER ORDER ISOPARAMETRIC ELEMENTS AND QUADRATURE

ax ~<«V. ~dN, T-dN.

77 = "11 2# -57*/ + n2iZ-r-yy + «ji 2,-772,

dy' ^dN, ^BNj ^dNj

~ + +
to

and where n ( | is the cosine of the angle between the x and x' axes, and /i2) is the
cosine of the angle between the y and the x axes, etc (i.e., n lt , n 21 , etc are the
direction cosines)
b. By denoting the entnes in the inverted Jacobian matrix as j4 m , d ]2 , A^, and /i 22 ,

determine the explicit form of the element stiffness (or conductance) matrices for
conduction in the and y directions in terms of the local, serendipity coordinates
r

r and s [and the nodal coordinates Note the integration is performed

over the parent element and, thus, the integrals may be evaluated numerically
with Gauss-Lcgcndrc quadrature Do not attempt to perform the quadrature, how-
ever

9-85 Recall that the condition for a one-to-one mapping is that the sign of the Jacobian
remain the same for all points in the domain mapped For the quadratic-order, lineal

element showthat this implies that the interior node must be in the middle one-half

zone Hint Problem 9 35 gives the expression for the Jacobian

 Transient and Dynamic Analyses

10-1 INTRODUCTION

Up to now we have only considered steady-state nonstructural applications and

static stress analysis applications. In this chapter the finite element method is

extended to transient and dynamic analyses. The reader may recall from Chapter
4 that transient (nonstructural) analyses are frequently referred to as unsteady, time-
dependent, or propagation problems. In structural and stress analyses, time-inde-
pendent problems are referred to as static or equilibrium problems, whereas time-
dependent problems are almost exclusively referred to as dynamic.
The chapter begins with a new notion referred to as partial discretization. In
effect, this will allow us to discretize the time domain. The reader will recall that
up to now we have only had to perform discretization in space.

The introduction to partial discretization is followed by a presentation of the

governing equations for dynamic structural analysis and transient thermal analysis
in Secs. 10-3 and 10-4, respectively. No longer will we get the familiar Ka = f,

but rather several additional terms will arise. The concept of lumped and consistent
mass and capacitance matrices is then introduced in Sec. 10-5 with the implications
of each briefly discussed.
Solution methods are presented for transient thermal analysis problems in Secs.
10-7 and 10-8, following a brief introduction to this subject in Sec. 10-6. The
development in Sec. 10-7 is based on the finite difference method, whereas that in

Sec. 10-8 is based on the finite element method. It will be seen that both methods
give the same two-point recurrence formula which will allow us to compute the
nodal temperatures as a function of time. Indeed, it will be seen that the results

577
578 TRANSIENT AND DYNAMIC ANALYSES

from the finite difference method are actually special cases of the more general
finite element method. It is emphasized that in all cases the goal is to compute the
temperatures at the nodal points as a function of time.
Finally, in Secs. 10-9 dynamic structural anal-
and 10-10, solution methods to

ysts problems are presented. In Sec. 10-9 a three-point recurrence scheme is derived
by using suitable finite elements in time. This will allow us to compute the nodal
displacements within a structure as a function of time. Both the inertia and damping
within the structure will be explicitly included in the analysis. In Sec. 10-10 an
introduction to modal analysis is given which will allow us to compute the undamped
natural frequencies and the so-called mode shapes of a structure.

10-2 PARTIAL DISCRETIZATION

In the previous chapters, the unknown parameter function <{> on an element basis
was assumed to be of the form

<}> = Na' M0-1J

where <J> is a vector or a scalar and may represent the displacement, temperature,
velocity, and/or pressure fields Recall that the shape function matrix N is a function
of the particular coordinates being used (i e ,
global or local) and the vector a' is

composed strictly These constants are the values of the parameter

of constants
functions at the locations of the nodes For example, in two-dimensional stress
analysis, we have assumed the displacement u on an element basis to be given by

u = NCx,y)a' (10-2)

where u itself is given by Eq (5-58) Similarly in two-dimensional thermal analysis,

we have assumed the temperature T on an element basis to be given by

T = N(x,y)a' M0-3)

emphasized that N in both cases is a function of the global (or local) coordinates
It is

whereas a e is composed of constants (albeit unknown constants).

Let us now consider the case of a dynamic structural problem and unsteady
heat transfer problem. Unsteady problems may also be referred to as transient or
time-dependent problems. Since the time variable t enters into such problems, Eqs.
(10-1) to (10-3) could be modified such that the shape function matnx could include
the time variable in addition to the spatial coordinates. For example, Eq. (10-1)
could be written as

<{> = N(x,y,T)a' l 10 ’ 41

where a new shape function matnx N could be derived The principal disadvantage
of this approachis that it increases the number of dimensions by one. For
example,

a one-dimensional problem becomes two-dimensional, a two-dimensional problem

becomes three-dimensional, and so forth
 DYNAMIC STRUCTURAL ANALYSIS 579

This additional complexity may be avoided if we allow the vector of nodal

unknow ns a itself to be a function of time. This is referred to as partial discretization
111- Equation (10-1) could then be written as

6 = ISV(t) (10-5)

where the explicit dependence of a on t e

is show'n. The implication is that no longer
will we get the familiar equation

K'a f = f
f
(10-6)

Instead, two additional terms will arise in dynamic structural analysis problems,
and one additional term will arise in transient thermal analysis problems. The explicit
forms for these equations are given in Secs. 10-3 and 10-4 for dynamic structural
and transient heat transfer problems, respectively.

10-3 DYNAMIC STRUCTURAL ANALYSIS

Recall from Chapter 5 that the principle of minimum potential energy or principle
of virtual work may be used to derive Eq. (5-46). Both approaches gave virtually
the same result, which is repeated here for easy reference:

rs
(8e) r
r
f
a dV = | (8u) b dV +
f
(8u) dS + 2 (8u)% (5-46)

Note that the right-hand side of Eq. (5-46) is composed of three terms, each of
which represents an external force acting on the structure, and hence the element.
By applying D’Alembert’s principle [2], we may introduce additional terms to the
right-hand side from the inertia and the damping as shown below.
The inertia results from the mass of the structure, whereas the damping results
from energy dissipation or friction within the structure. The inertial force per unit
volume given by the product of the mass density p and the acceleration. The
is

damping force per unit volume is generally assumed to be the product of a viscous
matrix {i and the velocity vector. Let us denote the first derivative of the displace-

ment field vector u (i.e., the parameter function) as u, and the second derivative

as u. Note that u and u represent the velocity' and acceleration fields, respectively.

It follows from D’Alembert’s principle that the elemental inertial and damping
forces di, and diD are given by

di, = — pii dV (10-7)

>1 II — piu dV (10-8)

Recall that the minus signs are necessary- if these additional terms are to be regarded
as stemming from external forces. It then follows from Chapter 5 that Eq. (5-46)
may- be w-ritten as
580 TRANSIENT AND DYNAMIC ANALYSES

(5e) r tr dV = ] (8u) r b dV + (Su) s

r
dS + £ (5u) rf
p
f J

r dV — ' (8u) rnu dV

— JV '
(8u) pu
jV
(10-9)

—— = N — = Na
rf(Na

AT
f
) da e
dT
f
(T)

u = Na'(T)

where, for example, a'OO is given by

a* = Iu,(t) v,(t) '

m2 (t) v2 (t) i
i
«„(t) v„(T)]
r (10-1 1b)

for two-dimensional problems. In Eq (10-1 lb), m,(t) and v,(t) are the x and y
components of displacement of node i at time t, and n is the number of nodes used
to define element e
Following the procedure used in Sec 5-7 for static stress analysis formulations,

we may write Eq (10-9) in the form

M'a' + D'a' + K'a' = f' (10-12)

The element stiffness matnx K' and element nodal force vector f' remain the same
as in Sec 5-7 (see Eqs (5-86) to (5-92)] However, two additional terms and
matrices arise. The matrix M' maybe referred to as the element mass matrix,
whereas D" may be referred to as the element damping matnx, both defined by

D' = NVN a
The element damping matrix D f should not be confused with the material property
matnx D from previous chapters because the context makes the meaning clear (also
the matenal property matrix D is never written with a superscript). Note that for
each element, the nodal displacements, velocities, and accelerations are given by
a', a f and a', respectively Equation (10-12) reduces to the more familiar
,

K'a r = f' when the velocities and accelerations are zero or negligible.
The assemblage of the element mass matrices f
to form the assemblage mass M
matnx M“ follows the procedure used to obtain the assemblage stiffness matrix K"
from the element stiffness matrices K'. An explicit form for the assemblage damping
matrix D* is not usually known because the viscous matrix |i is not usually known.
Therefore, the assemblage damping matnx D is generally assumed to be given by
a linear combination of the assemblage mass and stiffness matrices, or
 transient thermal analysis 581

D = aM + pK (10-15)

where a and p
are experimentally determined constants [3], The damping implied
by Eq. (10-15) is referred to as Rayleigh damping. In any event, the assemblage
system equation may be written in the form

Ma + Da + Ka = f (10-16)
a
Note that the superscript ( ) that is normally used to denote assemblage is no longer
written. The solution to Eq. (10-16) requires the specification of initial conditions
on the nodal displacements andvelocities. In addition, the boundary conditions on
the nodal displacements must be imposed. These matters are dealt with in Sec.
10-9 where solution methods for Eq. (10-16) are presented.

Example 10-1

Determine the element mass matrix for one-dimensional, dynamic structural analysis
problems. Assume the two-node, lineal element.

Solution

If length coordinates are used, we may write Eq. (10-13) as

f L\ L,L2
M' P£. L2 A
\
dx pA dx
A* L2 L, L\

With the help of Eq. (6-48) each entry in the matrix may be integrated to give

pAL 2
M f = ~6~
1
1

2
(10-17)

where L, the length of the element, should not be confused with the length coor-
dinates (which are always written with subscripts). Note that the total mass ot the
element is given by pAL. ®

10-4 TRANSIENT THERMAL ANALYSIS

In Chapter 8 steady-state heat transfer problems were formulated. Recall that the
formulations were quite general, but no mechanism for energy storage was included.
It can be shown from the first law of thermodynamics that an additional term arises

in each of the governing equations presented in Chapter 8. For example, for one-
dimensional problems, the governing equation given by Eq. (8-3) becomes

pcA ~ = - (kA - hP(T - Ta ) - s mP(74 - 7?) + QA (10-18)

St dx \ dx)

where p and c are the mass density and specific heat, respectively. Let us limit the

present discussion to heat conduction in solids (and quiescent liquids) so that the

specific heats at constant volume and constant pressure are virtually the same.
582 TRANSIENT AND DYNAM 1C ANALYSES

Therefore, there is no need to distinguish between these two specific heats and,
hence, no subscript on c is needed. In Eq. (10-18), the variable A represents the
cross-sectional area of the object being analyzed and may be a function of x.
and three-dimensional heat conduction problems, the gov-
Similarly, for two-
erning equations are given by

pa
dT d /
Ur —
bt\
+ —d ,
(,
Ur —
dT\
s dx V
1

9x) dy V
1

dy/
- «,(T 4 - T?) + q, + Qt (10-191

and

d (df\
+ —d T\
( 0-
U- + —d (
k—
dT\
( 10 - 20 )
dx V **/ \ siy /
\ to)

In Eq. (10-19), the variable t is the thickness of the two-dimensional region (and
hence the element thickness) Note that the variable t (not r) is used to represent
time The governing equation for axisymmetnc heat conduction is similarly mod-
ified, and the result is

d (df\
-i -d l
\rk —
ef\
+ k— T \
( 10 - 21 )
r dr \ °r / dz \ d2 J

Since a transient formulation is desired, we may employ partial discretization and

write the temperature T within a typical element e as

T = Na'(T)

The shape function matrix N is unchanged from Chapter 8 and a'CO is given by

a'(r) = [r,(r) T2 (t) •••

T„(t))
t (10-22)

where T,(t) is the temperature of node i at time t, and n is the number of nodes
used to define element e
The Galerkin weighted-residual method may be used to derive the finite element
characteristics providing the energy storage term is represented in the governing
equation and hence the residual The energy storage term is on the left-hand side
of Eqs. (10-18) to (10-21). For example, in two-dimensional problems, we begin
by writing the weighted residual equation as

i-
Nr
Uf -
i H) -
i K) ] ** + ° "°- 23 ’

The Green-Gauss theorem is applied in the usual manner on the terms involving
second-order derivatives Except for the energy storage term, the formulation is
identical to that presented in Sec. 8-8 Because of the presence of the energy storage
terms, we no longer get K*a* = f* but rather

(10-24)
C'a r + K r a' = f*
 transient thermal analysis 583

The matrix Cf is referred to as the element capacitance matrix and is defined by

Ce = J
S r pc/.N dx dy (10-25)

for two-dimensional problems. In a similar fashion it can be shown that the element
capacitance matrix in one-dimensional, transient heat conduction is given bv

C e = S T pcAN dx (10-26)
j
whereas in three-dimensional problems it is given by

Ce = L prN dx dy dz (10-27)

Finally, for axisymmetric problems, we have

Ce = lot's
7
pcrS dr dz (10-28)

Example 10-2
Show that the element capacitance matrix for one-dimensional heat conduction
problems is given by £q. (10-26).

Solution

From the weighted residual equation

it follows that an additional term, nameiv. C'a'’. arises because we have

The expression in the brackets is recognized as the element capacitance matrix

defined by Eq. (10-26).
S

Example 10-3
used
Evaluate the element capacitance matrix for the linear-order triangular element
in the discretization of a two-dimensional region
584 TRANSIENT AND DYNAMIC ANALYSES

Solution

The element capacitance matrix in this case is given by Eq. (10-25). If the shape
functions are written in terms of the area coordinates, we get

C' = . L2 pet [L t L2 Lt\ dx dy

jA

r ^*1 ^- 1^2 L L3 x

W-i L\ pet dx dy
\A ,
Ll J

With the help of Eq. (6-49), this matrix evaluates to

2 1 1
p ctA
C' 1 2 1 (10-29)
12
1 1 2

where A is the area of the triangle Note that the total capacitance of the element
(i.e., pc/A) is distributed as given by Eq. (10-29)
The assemblage of the element capacitance matrices into the assemblage ca-
pacitance matnx is done in precisely the same manner as the assemblage of the
element stiffness (or conductance) matrices to form the assemblage stiffness (or
conductance) matrix The assemblage system equation then takes the form

Ca + Ka = f (10-30)

Before Eq (10-30) can be solved for the nodal temperatures, the initial and boundary
conditions must be imposed as shown in Sec 10-7 Note that Eq. (10-30) is not
written with the superscript (°) which is generally used to indicate the assemblage
matrices before application of the geometric boundary conditions This superscript
is dropped because it will unnecessarily clutter the notation in Sections 10-7 and
10 - 8 .

10-5 LUMPED VERSUS CONSISTENT MATRICES

Lumped mass and capacitance matrices are always diagonal matrices, whereas
consistent mass or capacitance matrices are not necessarily diagonal A diagonal
matrix is defined as a square matrix whose entries are zero everywhere but on the
principal diagonal. Recall from Chapter 2 that the principal diagonal always runs
from the upper left comer to the lower right comer of a matnx The identity matnx
is an example of a diagonal matnx

Consistent Matrices

Recall that the element mass matnx in all dynamic structural analysis problems
may be determined from Eq. (10-13), where dV must be taken to be A dr, t dx d).
 LUMPED VERSUS CONSISTENT MATRICES 585

or dx dy dz for one-, two-, and three-dimensional problems, respectively. For dy-

namic axisymmetric stress analysis, we simply take dV to be 2-irr dr dz. When the
shape functions from the previous chapters are used, the resulting element mass
matrices are referred to as consistent element mass matrices.
In a similar fashion, the element capacitance matrices for one-, two-, and three-
dimensional transient thermal analysis problems are given by Eqs. (10-25) to
(10-27). Moreover, the element capacitance matrix for axisymmetric, transient
thermal analysis is given by Eq. (10-28). When the shape functions from previous
chapters are used, the resulting element capacitance matrices are referred to as
consistent element capacitance matrices.
Consistent element mass and capacitance matrices are discussed in more detail
below.

Consistent mass matrices

Recall from Example 10-1 that the element mass matrix for one-dimensional dy-
namic, structural analysis problems is given by

= pAE
[~

M c 2 j
1

2
( 10 - 17 }
6 Li I

providing the linear-order, lineal element is used. Note that this matrix is not
diagonal. Note further that this result was derived in a consistent manner by using
the appropriate shape functions and Eq. (10-13) directly. It is for this reason that
mass matrix. It can be shown that for
this result is referred to as a consistent the

linear-order, triangular element the consistent mass matrix is given by

2 0 1 0 1 0
0 2 0 1 0 1

1 0 2 0 1 0
( 10 - 31 )
0 I 0 2 0 1

1 0 1 0 2 0
0 1 0 1 0 2

Again it is seen that the consistent mass matrix is not a diagonal matrix. This implies
that the assemblage mass matrix also will not be a diagonal matrix. Further im-
plications of this are discussed later in this section. These results may be generalized
to three-dimensional and axisymmetric problems as follows: If the element mass
matrix is evaluated from Eq ( 1 0- 1 3) a consistent mass matrix
. ,
is obtained (assuming

that the shape functions from Chapters 6 and 9 are used).

Consistent capacitance matrices

Let us now consider the result from Example 10-3 where the element capacitance
matrix for two-dimensional, transient heat conduction problems is given by Eq.
not diagonal
(10-29) for the linear-order, triangular element. Note that this matrix is
was derived in a consistent manner by using the appropriate
and that this result
586 TRANSIENT AND DYNAMIC ANALYSES

shape functions and Eq. (10-25) directly. Therefore, the element capacitance matrix
given by Eq. (10-29) is referred to as a consistent capacitance matrix. This implies
that the assemblage capacitance matrix also will not be diagonal. The implications
of this in the solution step are discussed below. These results may be generalized
to other heat conduction problems as follows: If the element capacitance matrix is

evaluated from Eqs. (10-25) to (10-28), a consistent capacitance matrix is obtained

(assuming that the shape functions from Chapters 6 and 9 are used).

Lumped Mass and Capacitance Matrices

In the solution for the nodal displacements and temperatures, the initial part of the
solution may tend to be oscillatory about the true solution if the consistent mass or
capacitance matrixis used. These oscillations generally do not occur if the so-called

lumped mass or capacitance matrix is used Recall that a lumped matrix is a diagonal
matrix. In order to illustrate these trends, consider the temperature versus time
curve for a typical node in a thermal analysis shown in Fig. 10-1. Note that the

solution with the consistent capacitance matrix oscillates about the solution for the
lumped matrix. This does not imply that the solution with the lumped matnx is the
exact solution Some researchers argue that the wiggles that arise when the consistent
matrix is used are a signal to the analyst that smaller time steps should be used in

the vicinity of the wiggles. Others such as Gresho and Lee [4] contend that the
results from the lumped matrix are no more accurate, but since there are no wiggles,
these results are erroneously accepted as correct.

Figure 10-1 Temperature of a typical node versus time for a transient thermal analysis
performed with a consistent capacitance matnx and a lumped capacitance matnx
 ;

LUMPED VERSUS CONSISTENT MATRICES 587

A rule of thumb will now be given that can be used to obtain the lumped form
of the mass and capacitance matrices from the consistent matrices. The
lumped
mass or capacitance matrix is obtained by scaling the diagonal entries in the con-
sistent mass or capacitance matrix such that the total mass or capacitance is preserved
[5]. In general, any lumping that preserves the total mass or capacitance will lead
to convergent results. If this rule is applied to the consistent mass matrices given
by Eqs. (10-17) and (10-31), the corresponding lumped mass matrices are given
by

pAL i ; 0
( 10 - 32 )
2 o 1

and

i 0 0 0 0 0
0 1 0 0 0 0

pfA 0 0 0 0 0
M e

T 0 0
1

0 1 0 0
( 10 - 33 )

"o' 0 0 0 1 0
0 0 0 0 0 1

respectively. Note that in both cases the total mass for the element is equally
allocated to the nodes (and to each degree of freedom). Similarly, the lumped form
of the capacitance matrix that corresponds to Eq. (10-29) is given by

I 0 0
p ctA
CT 0 0 10 - 34 )
~r 0
1

0 1
(

The reader may recall that the finite difference method always yields the lumped
matrices directly. One advantage of the lumped form of the capacitance matrix is
that the solution for the nodal unknowns (i.e., the nodal temperatures) may be
obtained in a more straightforward manner as explained in Sec. 10-7 (i.e., a so-
called explicit solution results).
The above rule of thumb may also be applied to the higher-order elements
from Chapter For example, for the quadratic-order, rectangular element in the
9.
serendipity family, it can be shown that the consistent capacitance matrix is given
by

6 2 3 2 -6 -8 -8 -6
2 6 2 3 -6 -6 -8 -8
3 2 6 2 -8 -6 -6 -8
2 2 6 -8 -8 -6 -6
p ctA 3 10 - 35 )
Ce ~180 -6 -6 -8 -8 32 20 16 20
(

-8 -6 -6 -8 20 32 20 16

—8 -8 -6 -6 16 20 32 20
-6 -8 -8 -6 20 16 20 32
588 TRANSIENT AND DYNAMIC ANALYSES

Note that the nodes must be numbered as shown in Fig. 9-5(b). If the diagonal

entries in Eq. (10-35) are scaled such that the total capacitance is preserved, the
lumped capacitance matrix is given by

-1 0
0 -1
0 0
pctA 0 0
12 0 0
0 0
0 0
0 0

This matrix was obtained by summing the entries m a given row in the consistent

matrix, dividing the result by the total capacitance, and allocating this result to the
diagonal entry of the row under consideration. Interestingly, the capacitance allo-

cated to the comer nodes as given by Eq. (10~36a) is negative. Although this is

known to give good results, this form of the lumped matrix is numerically incon-
venient and is seldom used Instead, the following form of the lumped capacitance
matrix is frequently used

1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
(10-36b)
0 0 0 0 8 0 0 0
0 0 0 0 0 8 0 0
0 0 0 0 0 0 8 0
0 0 0 0 0 0 0 8

Equation (10-36b) gives excellent results and is the recommended form of the

lumped capacitance matrix in this case.

In each case, the consistent and lumped form of the matrices appear to be quite
different. However, except for the \ery early portion of the solution, both giie
virtually the same results The reader may want to consult Zienktew icz’s book 16]
for more information on the subject of lumping.

10-6 SOLUTION METHODS

Equations (10-16) and (10-30) need to be solved for the nodal displacements and
temperatures, respectively, as a function of time. The boundary and initial conditions
also need to be imposed. The two main methods of solution that can be used are
based on the finite difference element methods. Both of these methods
and finite

result in a two-point recurrence scheme for the solution of Eq (10-30), whereas a

three-point recurrence scheme results for the solution of Eq. (10-16). In Sec.
10-7, Eq. (10-30) is solved by using three different types of finite difference
 TWO-POINT RECURRENCE SCHEMES THE FINITE DIFFERENCE METHOD 589

schemes. In Sec. 10-8, it is shown that these three schemes are really a special
case of the more general finite element solution in time. The three-point recurrence
scheme for the solution of Eq. (10-16) is derived in Sec. 10-9. The
subject of
numerical stability of the various solution methods is a very important one. Although
a very brief introduction to this is given in Sec. 10-7, the reader should consult
references 7 to 9 for more complete discussions. Recurrence schemes may also be
referred to as recursion formulas. In this text, both designations are used inter-
changeably.

10-7 TWO-POINT RECURRENCE SCHEMES: THE FINITE DIFFERENCE

METHOD
Recall that all transient thermal analyses result in the equation

Ca + Ka = f (10-30)

In this section three different solution methods are derived for the solution of this
equation by using three different types of finite differences. In particular, forward,
backward, and central differences are used. It will be seen that the resulting re-

currence schemes can be summarized by one all-encompassing equation. Boundary

and initial conditions will be considered at the appropriate point in the development.

Forward Difference Scheme

By definition, the derivative of the a with respect to the time t is given by

da a(x + At) - a(T)

a lim (10-37)
dr a-r—o At

Recall from elementary differential calculus that a(T) and a(T + At) denote the
at times t and t + At, respectively. However, the notion
values of the vector a
of a finite difference implies that we do not require At to approach zero. Instead,

a small but nonzero At is used, and we approximate a by

a(T + At) - 3(t)

a = (10-38)
At

The change in time, or At, is referred to as the time step. If Eq. (10-38) is used

to eliminate a in Eq. (10-30), we get

a(T + At) - a(r)

+ Ka = f ( 10 -39)
At

Since we desire a forward difference, we must evaluate the remaining terms in Eq.

(10-39) at time t, or
590 TRANSIENT AND DYNAMIC ANALYSES

a(T + At) - apr)

C + Ko(t) = f(T) ( 10 -40 }
At

If Eq. (10-40) is multiplied by At and written such that only the term containing
a(T + At) appears on the left-hand side, we get

Ca(T + At) = [C - KAt]3(t) + f(T) At (10-41}

Let us denote the vectors a(T) and a(T + At) as a, and a, +1 , respectively, and f(T)
and f(T + At) as f, and f1+s . Therefore, we may write Eq. (10-41) as

Ca,+ |
= [C - K AT]a (
+ f, At (10-42)

Equation (10-42) is referred to as a two-point recurrence scheme because the right-

hand side is completely known at any time t, including t ~ 0 for which the initial

conditions apply Therefore, Eq (10-42) may be applied recursively to obtain the

nodal temperatures for a subsequent time given the temperatures for the preceding
time. Boundary conditions may be imposed as described later in this section. The
forward difference method is also known as Euler’s method

Note that if the lumped form of the capacitance matrix is used, the assemblage
capacitance matrix C is diagonal In this case the solution for the ;th nodal tem-
perature is given explicitly by dividing the j\h row on the right-hand side of Eq.
(10-42) by the j\h diagonal entry in the C matrix. Hence, the solution in this case
is frequently referred to as an explicit solution This is the principal advantage of
the forward difference (or Euler’s) method Conversely, if the consistent capacitance
matrix is used, an implicit solution for a + is C is no longer
1 ,
required because
diagonal In this case, the matrix inversion method or the active zone equation
solver (i e , subroutine ACTCOL { 10)) may be used to determine the nodal tem-
peratures at time t + At
Euler's method is also convenient from the standpoint of nonlinear analyses
Recall that a thermal analysis becomes nonlinear if any of the properties is tem-
perature-dependent or if thermal radiation is present. Since the terms in Eq.
(10-30) must be evaluated at time t and since the temperatures are always known
at this time, it is not necessary to iterate to obtain the solution for time t + At
for such problems
The principal disadvantage of this method will be illustrated numerically in
Sec. 10-8 where an application is presented Suffice it to say here that this method
requires a relatively small time step for both stability and accuracy. In other words,
if the time step exceeds a certain critical value for the mesh used, the solution

becomes oscillatory and blows up as the time increases. Even though the time step
may be below this critical value, the results may still be quite inaccurate (but stable)
The accuracy of this method generally improves for successively smaller time steps.
In fact, one somewhat practical way to assess the accuracy of the solution is to
compare the results for two different time steps, e.g., At and 2 At. If the results
for the two different time steps are within some acceptable tolerance, a good
approximation to the true solution has been obtained.
 TVVO-POINT RECURRENCE SCHEMES THE FINITE DIFFERENCE
METHOD 591

Backward Difference Scheme

If the derivative of a with respect to the time t written in the

is backward direction
(with respect to the time) and if we do not require the time step At to be zero, we
may write

a(T) - a(x - At)

a s (10-43)
At

Recognizing that the remaining terms in Eq. (10-30) should be evaluated at time
t, we get

apr) - a(T - At)

+ Ka(T) = f(T) (10-44)
At
After multiplying each term by At and rearranging such that the terms involving
a(T) appear on the left-hand side, we have
[C + KAT]a(T) = Ca(T — At) + f(r) At (10-45)

Note that the arguments t and t - At on a and f are relative; in other words,
nothing is changed if we substitute t + At for t (and t for t - At) to give

[C + KAT]a(T + At) = Ca(r) + f(T + At) At (10-46)

Let us again use the subscripts i and i + 1 to denote times t and t + At, re-

spectively, so that the two-point recurrence scheme from the backward difference
method is given by

[C + K At] a + J I
= Ca, + f,
+ I
At (10-47)

As in the case of the recurrence formula from the forward difference method, the
right-hand side is completely known at time t including f, + because it will be 1

recalled that the vector f represents the forcing function for the analysis. Therefore,
Eq. (10-47) may be applied recursively to obtain the nodal temperatures for a
subsequent time given the temperatures for the previous time.
Note that even if the capacitance matrix C is diagonal, an implicit solution for
the vector a i +must be obtained because the stiffness (or conductance) matrix K
1

is never a diagonal matrix. The backward difference method is stable for all At,
but the accuracy deteriorates as the time step is increased. Again the accuracy may
be assessed by comparing the results for two different time steps.

Central Difference Scheme

Let us again represent the time derivative of a by Eq. (10-38). However instead of
evaluating the other terms in Eq. (10-30) at time t or time t + At, let us use the
average values. In other words, let us take
592 TRANSIENT AND DYNAMIC ANALYSES

a(T + At) + b(t)

a ” 2

and

=
f(T + At) + f( T)
f
2

This is equivalent to taking a central difference and we have

apt + At) — a(x) a(T + At) + a(T)
_ f(T -f- At) + f(r)
+ ^ ~
At 2 2

Multiplying through by At and isolating the terms containing a(T + At) gives

a(T + At)

= C -^) a ,T) +
» fW + f ^t^l^ J1MM
(

In terms of the subscript notation used above, Eq. (10-48) becomes

(c + KfT)a„, = (c
- —~) a, + (.0,3,

As backward difference scheme, the recursion formula given by Eq.

in the
(10-49) requires an implicit solution for the nodal temperatures at time t + At
given those at time t This method results in an oscillatory solution if the critical
time step for stability is exceeded For time steps smaller than this critical value,

the accuracy of the solution improves as the time step is decreased. Not surprisingly,
the central difference method is more accurate than both the forward and backward
difference schemes because the central difference favors neither the temperatures
at time t nor those at time t + At The central difference method is also referred
to as the Crank-Nicolson method.

Summary and Application of the Prescribed Temperatures

The three recurrence formulas given by Eqs. (10-42), (10-47), and (10-49) may be
summarized m one convenient equation as

[C + 0K AT]a + l l
= [C - (1 - 0)K Ar]a (
+ [(I - 0)f, + 9f, + il At (10-50)

where the parameter 0 takes on values 0, 'A and 1 for the forward, central, and
backward difference schemes, respectively
required
Strictly speaking, for values of 0 other than 0, an iterative solution is
temperature dependent or if thermal
for each time step if any of the properties is

radiation is included in the model In other words, the matrices K and C should
 1

TWO-POINT RECURRENCE SCHEMES THE FINITE DIFFERENCE METHOD 593

be evaluated with the nodal temperatures that correspond to the value of 0

really
being used.The temperatures for the (j+ l)st iteration (denoted by a-' +1 ) may be
computed from the nodal temperatures from the yth iteration with the help of the
following equation:

a ;+ = (1 - 9)a/ + 0a/ +1 {10-51}

where a/ and a/+ ,

denote the nodal temperatures at times t and r + At, respec-
tively, for the yth iteration. Note that for the forward difference or Euler’s method
(for which 0 is zero) it is not necessary to iterate (why not?).
Recall that the initial conditions or a 0 = a(0) are automatically incorporated
in the solution process because these values of a (at t = 0) are used to start the
recursion process. This is to say that a*, is taken as a(0) from which a, may be
found from Eq. (10-50). The recursion process is illustrated schematically in Fig.

10 - 2 .

Only prescribed temperature boundary conditions need to be considered at this

point because all other boundary conditions (prescribed heat flux, convection, etc.)
are automatically included in the Finite element formulation. Prescribed temperatures
may be imposed by either of the two methods from Sec. 3-2 because Eq. (10-50)
is of the form K eff a = fef r, where K cff is really C + 0K At from Eq. (10-50) and
feff is given by the entire right-hand side of Eq. (10-50). As in the steady-state case,
it is desirable to preserve symmetry because this substantially reduces the storage
requirements in large problems. In summary, application of the prescribed tem-
perature boundary conditions for transient thermal analyses is really no different
from that in steady-state problems, if the prescribed temperatures are imposed on
the vector a, + 1
in Eq. (10-50). Note that the prescribed temperatures may possibly

be a function of time.
From a practical point of view, if the numerical solution appears to oscillate
when the physics of the problem would preclude this, the time step is too large for
the mesh used. Reducing the time step to below that for which the instability or
oscillation disappears is a very' practical remedy. These comments apply in particular
to the forward and central difference schemes. Accuracy may be assessed by com-

Figure 10-2 Schematic diagram of the two-point recursion process.

594 TRANSIENT AND DYNAMIC ANALYSES

paring the results for two cases with different time steps: If there is a significant
difference between the results, then the time step should be reduced further and
the analysis repeated.

10-8 TWO-POINT RECURRENCE SCHEMES: THE FINITE ELEMENT

METHOD

The finiteelement method itself may be applied to Eq. (10-30) as shown in this
section Because the time domain is to be discretized, the elements used may be
referred to as temporal elements to distinguish them from the elements used to
discretize the body. It will be seen that the three recursion formulas from Sec.
10-7 are really a special case of the more general finite element method. In particular
Eq. (10-50) will be derived by discretizing the time domain from time t to time
t + At with a suitable first-order element. One such element is shown in Fig.
10-3 where times t and t + At are denoted as t, and t1+i » respectively. Note the
shape functions N, and Nl+
shown These shape functions are now a
,
are also
function of time However, mathematically they are no different than those presented
in previous chapters except that t replaces .r. If the weighted-residual method is

used to solve Eq (10-30), we may wnte

J’“'lV[Ca + Ka - f] = 0 [10-521

where the weighting function VV is a scalar in this case Different choices for VV
will yield different two-point recurrence schemes.
With the help of the shape functions, we may represent the vector a in Eq.
(10-52) as

a = W,a, + NI+ ,a + 1 ,
(10-531

because the shape functions are linear with time and evaluation of Eq. (10-53) at

times t, and t + yields a, and a +

1 1
respectively, as desired
I l ,

Figure 10-3 The tv. o- node temporal element and its associated shape functions
 TWO-POINT RECURRENCE SCHEMES HIE ITNITE ELEMENT METHOD 595

For convenience let us define a local normalized coordinate

£ as follows:

£ =
T +
= — AT
(10-54)
/ I D

such that for t = t, and r = t, + 1 ,

we have £ = 0 and £ — 1 , respectively. It

follows that the shape functions N, and N, + ,

are given by

N, = I - £ (10-55a)

and

Nl+l = £ (10-55b)

Figure 10-3 shows the behavior of the shape functions over the time interval from
r = t,- to t = t + i 1
. As mentioned above, these shape functions are no different
from those for the linear-order, lineal element in Chapters 6 and 9.

With the help of Eqs. (10-55), we may write Eq. (10-53) as

a = (1 - £)a, + £a, + 1
(10-56)

from which it follows that

a
da _ — £)a, + £a l + 1 ]
rf£
(10-56)
ch di ch di dr

or

1 1

a = -—-a, + “a, + |
(10-57)
At At

Similarly, f may be represented as

f = NX + JV f+l f( + l
(10-58)

or

f = d - £)f, + (10-59)

From Eq.' (10-54), it follows that d~ is related to d£ by

d- = At di (10-60)

Therefore, Eq. (10-52) may be written as

a, + + Kl(l - £)a, + £a,, ,]

At

- [(1 - £)f, + £f, At r/£ = 0 (10-61)

, 1

he put
where the weighting function IF is still unspecified. Equation (10-61) ma\
into the following more convenient form:
S96 TRANSIENT AND DYNAMIC ANALYSES

j^cJVrfg + KAtJV^ </|ja 1 + l

= - KAtJV( 1 - Qrfgja,

+ [^At JT* W(1 - 0rf?jf,

+ ^Ar/V^f.,, (10-621

Dividing both sides of Eq. (10-62) by IV (/fj gives the desired result:

[C + 0KAT]a(+ = ,
[C - (1 - 0)KAr]a + (
[(1 - 0)f, + 6f +
( 1 ] At (10-63)

where the parameter 0 is defined by

e = -Ji
(10-64)
Wd£
j

Note that Eqs. (10-50) and (10-63) are identical. It is emphasized that Eq. (10-50)
was derived and generalized from three different finite difference schemes, whereas
Eq. (10-63) was derived directly via the finite element method Different choices
for the weighting function W will yield different values for the parameter 0. In

particular the point collocation, subdomain collocation, and Galerkin methods

from Chapter 4 will be used to determine the corresponding value of 0 from
Eq (10-64)

Point Collocation

Recall from Sec 4-6 that the weighting functions for the point collocation method
are given by 8(£ - fj,) such that

8(6 - %)<!% = 1 for * = ij

(10-66a)

Vt, - 6,^6 =0 for (tQ-66bl

where is known as a collocation point. For point collocation at time t, (i e.,

t, = where £ = 0, it can be shown that 0 = 0. Similarly for point collocation

Tj)

at times (t, + ti+ ])/2 and ti+( , where £ = 16 and ij = 1, it can be shown that
0 = Zi and 0=1, respectively. Thus it is seen that point collocations at times t„
(t, + ti+1 )/2, and t)+1 correspond to the forward, central, and backward difference

schemes, respectively

Example 10-4

Show that point collocation at time (t, + ti+i )/2 gives 0 =!6
 TWO-POINT RECURRENCE SCHEMES THE FINITE ELEMENT METHOD
597

Solution

At time (t + t, +] )/ 2 we have
; £ = Vz and so the collocation point must be Vz
Thus the weighting function U' is given by

W = §(£ - Vz)

and Eq. (10-64) gives

1/2
6 =
/:
m- 1
/2) dt

Note that the numerator in the above expression for 0 is zero except at the collocation
point where £ = Vz. [2

Subdomain Collocation

It should be recalled from Sec. 4-6 that the weighting functions for the subdomain
collocation method are unity over a particular subdomain and are zero elsewhere.
Since only one unknown, namely, a, + ) , is to be found, we must take only one
weighting function or W — 1 over the time interval from t to t + At (i.e., from
t toi
t, + i ). The value of 0 for this case may be determined as follows:
1

1 !
^2

«« 2 . „
® “ ri
— n ~ i
(/-

l wd i
L
wd t t
0

Therefore, the subdomain collocation method is analogous to the central difference

scheme or Crank-Nicolson method from Sec. 10-7.

Galerkin

The weighting functions for the Galerkin method are taken to be the shape functions
themselves. In other words, we may take either

W = jV, = 1 - 5 (10-67a)

or

IT = Nl+i — £ (10-67b)

The shape functions are shown graphically in Fig. 10-3. It can be shown that if

= V\ whereas Eq. (10-67b) used, we get 0 =

Eq. (10-67a) is used, we get 0 if is

finite difierence
2
/j. Neither of these values corresponds to any of the results from the
598 TRANSIENT AND DYNAMIC ANALYSES

method. However, 0 = %
is particularly useful because it is more accurate than

the backward difference scheme (0 = 1) and more stable than the central difference
scheme (0 = Vi).

Example 10-5

Show that the parameter 0 takes on a value of Yj if the weighting function is given
by Eq. (10-67b).

Solution

By definition of 0, we have

n n

9 “
vw* ~0i* ^ ~ 2
3 L
~ %
fi e2|‘

I®* | lo

which is the desired result

Example 10-6
Reconsider the circular pin fin from Example 4-11. For convenience, the fin is

shown in Fig 10-4(a) Recall that the fin is made of pure copper with a thermal
conductivity k of 400 W/m-°C The base is held at a temperature Tb of 85°C and

tw

Figure 10-4 Circular pin fin (a) analyzed in Example 10 6 and (b) discretized into two
elements and three nodes.
 ,

TVVO-PO!NT RECURRENCE SCHEMES THE FINITE ELEMENT METHOD 599

:
the ambient temperature Tc is maintained at 25 C. The fin length L is 2 cm, and
the diameter D is 0.4 cm. The tip of the fin at .x — Lj is insulated. Determine the
transient temperature distribution within the fin if the entire fin is initially at 25°C.
The density p and specific heat c are 8900 kg'nf and 375 J/kg-°C. respectively.
Assume the consistent form of the capacitance matrix, a time step of 0. 1 sec. and
6 = i Use only two elements of equal length as shown in Fig. I0-4(b).
/;
.

Solution

Equation (10-50) or (10-63) provides the basis for the transient solution. Because
the same fin was analyzed in Example 4-11. we max use most of the results from
that example. In particular, the element stiffness matrix and nodal force vectors are
unchanged. It follows that the assemblage stiffness matrix and nodal force vector
are also unchanged. Therefore, from Example 4-11. we have

0.50893 -0.49951 0
K = -0.49951 1.01786 -0A9951
0 —0.-9951 0.50893

and since the nodal force %-ector is not time-varying, we have

0.23561
f, - flM = f = 0.47122
0.23561

We now need to compute the element capacitance matrix. It can be shown for
the linear-order lineal element that Eq. (10-26) evaluates to

pcAL 2 1

C' ( 10 - 68 )

6 I 2

where A and L are the cross-sectional area and length of the element, respectively.

Equation (10-68) is an expression for the consistent capacitance matrix. Let us now
evaluate this matrix for the problem at hand. Noting that the two elements are
identical, we may write
s
(8900)(375)( 1.2566 x 10 )(0.01) 1

C(D = c a, = ~>

or

0.13980 0.06990
C" = C 12
1 '

0.06990 0.13980

The assemblage of the element capacitance matrices is done in precisely the same
manner as the assemblage of the element stiffness matrices. The result is

0.13980 0.06990 0
C = 0.06990 0.27960 0.06990
0 0.06990 0.I39S0.
600 TRANSIENT AND DYNAMIC ANALYSES

Note that C is not diagonal because the consistent capacitance matrix has been
used. The matrix K efr may now be computed as

0.13980 0.06990 0
K cf f
C + 0K At = 0.06990 0.27960 0.06990
0 0 06990 0.13980

0 50893 -0.49951 0
+ Vi(0 1) -0.49951 1.01786 -0.49951
0 -0.49951 0.50893

or

0.17373 0 03660 0
0.03660 0 34746 0.03660
0 0 03660 0 17373

Similarly the vector f«rr is given by the nght-hand side of Eq. (10-50) [or Eq.
(10-63)], or

fe ff = [C - (1 - 6)KAT]a l
+ 1(1 - 6)f, + 0f + 1 ]
(
At

Since the force vector f does not change with time in this application, we should
note that

[(1 - 0 )f, + 0 fl+ ,] At = [(1 - 6 )f + 0 f] At = f At

Therefore, from

feff = [C - (1 - 0 )K AtJs, + f At
and the fact that a, is the vector of nodal temperatures at time t =
0 at this point
in the solution, we get

0 13980 0 06990 0
fell
= 0.06990 0 27960 0 06990
0 0 06990 0 13980.
50893 -0 49951 0
- - %) 0 - -0
{\

0.23561
it 49951
0 -0
1 017&6
49951
49951
0.50893 IS
0.47122 (0 . 1 )

0.23561

'o 12284 0 08655 1|~25" '0 02356

0
0.08655 0 24567 0.08655 25. + 0.04712
0 0 08655 0 1 2284 J [25 _ 0 02356
 TWO-POINT RECURRENCE SCHEMES THE FINITE ELEMENT
METHOD 601

The temperatures at time t - 0.1 sec are then computed from the solution of
Keff a = feff , or

"0.17373 '
0.03660 0 >r 5.2582'
0.03660 0.34746 0.03660 T, 10.5165
0 0.03660 0.17373 J 3_ _ 5.2582__

However, the prescribed temperature of 85°C must be imposed on node 1 . Using

Method 1 from Sec. 3-2 (and preserving the symmetry), vve get
"
1.00000 0 0 >r '85.0000'
0 0.34746 0.03660 73 = 7.4055
0 0.03660 0.17373 t3 _ _ 5.2582

Solving this system of linear algebraic equations gives

r, = 85.0°C 73 = 18.5°C r3 = 26.4°C

for the nodal temperatures at the end of the first time step, i.e., at t = 0.1 sec.
Note that the temperature of node 2 (i.e., T2 ) is not physically realizable. This is

a consequence of using the consistent capacitance matrix and is discussed further

below.
The nodal temperatures at subsequent time steps may be computed from the
two-point recurrence relation:

0.17373 0.03660 0 T,(t + At)

0.03660 0.34746 0.03660 73(t + At)
0 0.03660 0.17373 _U T + At)_

0.12284 0.08655 0 r,( t) '0.02356'

= 0.08655 0.24567 0.08655 73( t) + 0.04712
0 0.08655 0.12284 7\(t) 0.02356

The results are summarized in Table 10-1 for t up to 2 sec and in Fig. 10-5(a) for
t up to 6 sec, the time at which steady-state appears to have been attained. In order

to see the initial oscillations in the nodal temperatures, the results for the first five
time steps are shown in Fig. I0-5(b). Recall that these initial oscillations are

characteristic of the solution when the consistent capacitance matrix is used. HB

With the help of the problem posed Example 10-6. let us discuss the accuracy
in

and stability issues further. Figure 10-6 shows the solution for the temperature of
the tip of the fin from Example 10-6 in comparison with two other cases. Both of
these other transients were obtained with a 0.5-sec time step. The first case cor-
responds to Euler’s method (0 = 0); the second case to the Crank -Nicolson method
(0 = Vi). Note that the critical time step for stability has been exceeded in the case
of Euler’s method. This is evidenced by the oscillator}' temperature as a function
of time. Note further that the amplitudes of the oscillations are growing with time
(recall that this solution can be made stable if a small enough time step is used).
In direct contrast the Crank-Nicolson method happens to be stable for a 0.5-sec
time step. Although it is not shown here, the Crank-Nicolson method tends to give
 c
602 TRANSIENT AND DYNAMIC ANALYSES

Table 10-1 Resulting Nodal Temperatures for Example 10-6

e
Time, sec r„ °c T2 , C t3 ,°

0 1 85 0 18.5 26 4
02 85 0 29.7 21.7
03 85 0 36 4 22.7
04 85 0 41.0 25.6
05 85.0 44 7 29 3
06 85 0 47.8 33.1
07 85 0 50.5 36 7
08 85 0 53.0 40 1
09 85 0 55.2 43.2
1 0 85 0 57.3 46.1
1 1 85 0 59 2 48.8
1 2 85 0 61 0 51 3
1 3 85.0 62 6 53 6
1 4 85 0 64 1 55 7
1 5 85 0 65 5 57.7
1.6 85 0 66.8 59 5
1 7 85 0 67 9 61 1

1 8 85 0 69 0 62 7
1 9 85 0 70 0 64 1
20 85 0 70 9 65.4

oscillatory (but stable) results if an excessively large time step is used and the
accuracy of the solution deteriorates as well Recall that the backward difference
scheme is stable for all becomes increasingly less accurate as the
time steps, but it

time step increases Hence, a good compromise is to carry out the solution with a
value of 6 of V)

Example 10-7

Repeat Example 10-6 with a lumped capacitance matrix in order to illustrate the
lack of oscillations in the early portion of the transient solution for the nodal
temperatures

Solution

The lumped form of the capacitance matnx for the two-node. lineal element is

given by

1 o]
“ 0 1
L J
Therefore, numerically the element capacitance matrices are given by

01) = rm = [0 20970 0 ]
L
0 0.2097UJ
 TWO-POINT RECURRENCE SCHEMES THE FINITE
ELEMENT METHOD 603

Figure 10-5(a) Temperature as a function of time for the node at the tip of the fin in
Example 10-6. (b) Temperatures as a function of time of nodes 2 and 3 in Example
10-6. Note the oscillations in the early portion of the transient.

It is left as an exercise to show that the two-point recurrence relation in this case
becomes

0.24363 -0.03330 0 TiCr + At)

-0.03330 0.48726 -0.03330 T2 ( t + At)
0 -0.03330 0.24363 T3 (t + At)_

0. 19274 0.01665 0 T,(t) '0.02356

0.01665 0.38547 0.01665 T2 ( t) + 0.04712
0 0.01665 0.19274_ J 3 (t)_ 0.02356.
04 TRANSIENT AND D\ N KMtC AN ALVSES

Ngure 10-6 Companion of the tip temperatures for the fin in Example 10-6 for different
.alues of the parameter 0

Table 10-2 show s the nodal temperatures for the first 2 sec of the transient. Note
[he lack of oscillations in the nodal temperatures during the first few time steps
Note also that the results
(unlike the results with the consistent capacitance matrix)
from Example 10-6 with a consistent capacitance matrix and the results from this
example are within 1
C
C of each other after about one second B

10-9 THREE-POINT RECURRENCE SCHEMES

In this section we will dexelop a three-point recursion formula for the solution of
Eq (10-16). A suitable temporal element is used as shown below In particular,
he time domain be discretized with second-order, lineal elements. One such
will
rlement is shown in Fig 10-7 where the shape functions are also show n. Note that
he element runs from time t - At to timer + At and the center node corresponds
o time t. Times t — At, t. and t + At will be denoted by t,_i, t,, and rf+ |,
cspectiveh The shape functions for nodes r - 1, /, and i + 1 are denoted as
V,_i, N„ and A',+ j and are gi\en by Eq (9-5) in terms of the serendipity coordinate
Table 10-2

Time, sec

0.1
0.2
0.3
0A
0.5
7,,

85.0
85.0
85.0
85.0
85.0
m
THREE-POINT RECURRENCE SCHEMES

Resulting Nodal Temperatures for Example 10-7

=
C
29.1
34.6
39.1
42.9
46.2
r,.

25.6
27.0
29.2
31.7
34.4
C
C
605

0.6 85.0 49.1 37.2

0.7 85.0 51.6 39.9
0.8 85.0 53.9 42.6
0.9 85.0 56.0 45.1
1.0 85.0 57.9 47.5
1.1 85.0 59.7 49.8

1.2 85.0 61.3 52.0

1.3 85.0 62.7 54.0
1.4 85.0 64.1 55.8
1.5 85.0 65.4 57.6

1.6 85.0 66.5 59.2

1.7 85.0 67.6 60.7

1.8 85.0 68.6 62.1

1.9 85.0 69.6 63.5

2.0 85.0 70.4 64.7

weighted-residual method requires that we weight the residual

with a
The
over the interval from
weighting function, integrate the result with respect to time
7,.,to',xi. and set the result to zero, or

Da + Ka — f} d~ = 0 (10-69)

associated shape functions.

Figure 10-7 The three-node temporal element and its
606 TRANSIENT AND DYNAMIC ANALYSES

As in the case of the two-point recurrence formula, different choices for the scalar
weighting function W will yield different three-point recurrence schemes.
With the help of the shape functions, the vector a in Eq. (10-69) may be written
as

a = + AT, a, + Nl+] a l + l
(10-70)

Note that the vector a is assumed to vary parabolical ly with time such that at time

t,_j, we have a,_ j5 etc. It is emphasized that the shape functions are functions of

time and a _,, a,, and a + are constant vectors, where a + denotes the nodal
( l 1 1 1

displacements at time r + At, etc. The second derivative of a [the a in Eq.

(10-69)] requires that at least second-order shape functions be used; otherwise, a
would be identically zero For convenience the shape functions are given below.

N, = — >/2 r( 1 - r)

N2 = 'Ar(l + r) (9-5)

AT, = (1 + r)(l - r)

Recall that the midside node (at time t ( ) must be midway between the two end
nodes Note that the serendipity coordinate r is related to the time t by

r = — -— = —r— 1
for t,-i - t ^ t,+ i
(10-71)
ti + , t, At

and dr is related to dr by

dr - —
At
dr (10-72)

Therefore, we may write Eq (10-70) as

a = -'Aril - r)a,_, + (1 - /-)(! + r) a, + Vir{\ + r)a, +1 (10-73)

For a, we get

a = —
da
= da dr
= 1 da
— (10-74a)
dr dr dr At dr
or

a = T-[(-'/J + r) a,., - 2ra, + V/i + r)a,.,] I10-74bl

For a, we have

_ d
<
/ di - d (
1 rfa'l \dr (10-75a)
dr \ di ) dr'(At dr) dr '
 THREE-POINT RECURRENCE SCHEMES 607

or

1 d2a
a — - - - ’ — 1

t a <-i ~ 2a, + a, T ,] (10-75b)

2
dr 2
-i '

(At ) (At) 2

Similarly, the vector f may be represented as

f = + NX + A', - if, -i (10-76a)

or

f = - !Ar(l - r)f,_, + (1 - r)(l + r)f, + '/zr(l + r)f, + 1

(10-76b)

Therefore. Eq. (10-69) may be written as

r / \ 2

~
/_ ,
H [a '-i 2a -
+ a '+^

+ D — [(->/’ + r)a,_, - 2ra, + ('/: + r)a /+I ]

+ K[-'/’r(l - r)a,_i + (1 - r)(l + r)a,

+ Vzril + r)a, + I ]
- [~Vir(\ - /-)f,_,

+ (1 - r)(l + r)f, -r Vir(l + r)f, + I J| At dr = 0 (10-77)

It should be emphasized that the weighting function IV is as yet unspecified. Let

us rewrite this last equation in amore convenient form by first multiplying every
term by At and then rearranging to get

'

MI IV dr + D At/ ’
1V( A + r) dr

+ K(At) 2 ValVrd + r) dr a + I

i i
J

- 2M '
IV dr - 2D At IVr dr
J

+ K(At ) 2 'w(l - r)( 1 + r) dr

f

Ml :

j
'
w dr + D AtJ V(-'/2 + r) dr

- K(At ) 2 'a Wr(l - r) dr a /- 1

'xiWril - r)
2
dr (AT) f,_,
J
 ,

608 TRANSIENT AND DYNAMIC ANALYSES

+ - r)(l + r) dr (Ar)
2
f,
j

+ + r)<frj(iT) ; f, tl (10-78)

After dividing both sides by W dr and after some manipulation, we get [II)
[M + -yD At + \K(Ar) 2 Ja, +l

= [2M - (I - 2y)D At - ('A - 2X + -y)K(AT) 2 ]a (

+ - M + ( - y)D At -
[
I ( 'A + X - y)K( A T ) 2 ]a, _

- (S<$ + X - -Y)(AT) f,_

2
t

- (A - 2X + y)(AT) 2 f, - X(A T ) 2 fl+l (10-79}

where X and y are defined by

'

{
l
/2 VVr< 1 + r) dr
X = (10-80)
r ,

J ,
Wdr

and

I \v(V2 + r) dr
y = (10-81)

Wdr
j

Special Cases

Equation (10-79) is a three-point recursion formula for the solution of Eq (10-16).

Various schemes arise depending on the choice of the weighting function W. Table
10-3 shows the values of X and that correspond to the point collocation, subdomain
y
collocation, and Galerkin weighted-residual methods. Note that X and y are not
necessarily positive

Point collocation
by
Recall that the weighting functions for the point collocation method are given
8(r — ry ), such that

(10-82a)
/*V- 1 for r = r}
 h

THREE-POINT RECURRENCE SCHEMES 609

Table 10-3 Summary of the Values of the Parameters X and y for Use in the Three-Point
Recursion Formula

Weighted Residual Method X 7

Point collocation at t.-j 0 -!

Point collocation at t, 0 Z2
Point collocation at t, + )
1
A
3

Subdomain collocation 14 •A

Galerkin based on A',_ -

1 Vs -Vi
Galerkin based on N, Zw Zi

Galerkin based on V l+! % Zi

f S(r - rj) dr - 0 for r t*

^ (10-82bl
(

where r is known as the collocation point. For point collocation at time t,_i where
}
r = - 1 , it can be shown that X = 0and7 = - Vi. Similarly, for point collocation

at times t, and t, + ) ,
where r = 0 and r — + 1, it can be shown that X = 0 and

7 = Vz, and X = 1 and 7 = Vz, respectively. These results are summarized in

Table 10-3.

Example 10-8

Show that X = 0 and 7 = -Vi in Eq. (10-79) for point collocation at time t,_,.

Solution

At time we have r = - and the collocation point r, must be - 1 . Thus the

t, _ ,,
1

weighting function W is given by

W = 5(r + 1)

and Eqs. (10-80) and (10-81) give

+
'

f 'A S(r + I )r( 1 + r) dr

K = °/< = 0
S(r + 1) dr
J — 1

and

+ l)(Vz + r) dr

7 =
+ 1) dr
 )

610 TRANSIENT AND NAM1C ANALYSES

Subdomain collocation

The weighting function for the subdomain collocation method is given by IV’ = 1

over the mterval t _j to t + ( j j

. It is shown in Example 10-9 that X = 14 and 7 =
V2 in this case.

Example 10-9

Show that X = Vt and y = '/> in Eq (10-79) for subdomain collocation over the
interval from T,_ t
to ti+( .

Solution

Using W = 1 in Eqs (10-80) and (10-81) we get

J''w)r(l + r) dr j\'(r/2 + r*/2) dr < r2/4 + r5/ ®

/*,'( Ddr f''dr r

|*‘

and

1)(14 + r)dr r/2 + r !/2)

f ( ( ,

7 7+1
“ pn = V,

('><*
J, r

Galerkin

Since the weighting functions for the Galerkin method are the shape function
themselves, we may take

IV = N,- t
= - r) (10-83a)

W = Nj ~ (1 - r)(l + r (10-83b)

or

IV = JV, + |
= ’/2 r(I + r)
(10-83C}

The shape functions are shown graphically in Fig 10-7. It can be shown that if

used, we get X = -Vs andy = -Vt. Also, if Eq. (IO-83b)

is
Eq. (10-83a) is

used, we get X = Vio and y = Vi Finally, if Eq (10-83c) is used, we get

X = Vs and y = Vi.
 2 1 r

three-point recurrence schemes 611

Example 10-10

Show that X = - 'A and 7 = - Vi in Eq. (10-79) for Galerkin weighting corre-
sponding to Eq. ( 10-83a).

Solution

From the definition X, we have

’

/2 [-‘/2/-(1 -
i

/)]/•( 1 + r) dr 2 - 4
- /
Vt ( r dr
J )
a
X.
r+i

J ^
— Vir(\ — r) dr
I.
:>->*
or

+ 1

/2 (rV 3
l
- 5
r / 5)

A " ~ /5
2
+1
-«
(r / - r
3
l 3)

- 1

and from the definition of 7 we have ,

f l~ ^r(l - r)](Zi + r) dr ( 1 hr + '/2 r

2 — r
3
dr
! j *
)

1 ~ + l

f
- Ar(l
l
- r) dr J*'(r
~ r
2
) dr
J i

or
+ 1

2
(r /4 + r
3
/ 6 - 4
r /4);

7 - +1
- “
_ JL _
~ u
^
2
(r / 2 - r 3/ 3)
-I

Initial and Boundary Conditions

We have seen how the second-order, vector-differential equation given by Eq.

(10-16) may be solved by the three-point recursion formula given by Eq. (10-79).
The values of X and 7 used in the recursive solution are summarized in Table
10-3. In order to start the solution process, the values of a, and a,_i are needed
for / = 1 in order to determine a, + . It will now be shown how the initial conditions
on the nodal displacements and velocities can be used to provide the starting vectors
ao and a Two different methods are used; the first method is simpler but less
(
.

accurate than the second. Both methods require that Eq. (10-16) be written as a
set of two, first-order vector differential equations, or
612 TRANSIENT AND DYNAMIC ANALYSES

Mb + Db + Ka = f (10-84)

a = b (10-C5)

It should be noted that since the vector a represents the nodal displacements, the
vector b (defined to be a) must represent the nodal velocities. The initial displace-

ments and velocities may be denoted as ao and b 0 respectively.

,

Euler starting method

The starting values for a,_ j
are simply given by the initial nodal displacements a 0
We seek the starting values for the vector a,. The simplest approach is to apply the

forward difference scheme or Euler’s method to Eq. (10-85) to get

a(t 4- At) - a(T) _

or

a, - a _,
At
(

= ,b -
Solving for a, gives

a, = a,_, + b,_ ,
At (10-86)

For i = 1, Eq (10-86) becomes

at = a0 + b 0 At (10-87)

Since ao and b 0 are provided by the initial conditions on the nodal displacements
and velocities, respectively, Eq. (10-87) provides a means of obtaining a With t
.

ao and a, known, the three-point recursion formula given by Eq (10-79) could be

used to determine a 2 and so forth The recurrence scheme is shown schematically
,

in Fig. 10-8

figure 10-8 Schematic diagram of the three-point recursion process.

 ,

THREE-POINT RECURRENCE SCHEMES 613

Crank-Nicolson starting method

While the Euler starting method is simple, it is not very accurate and a better
approach results if Eqs. (10-84) and (10-85) are solved using the central difference
or Crank-Nicolson method. In this case Eq. (10-84) becomes

b, b, + + +
M At
+ D
b.
4- K-
a,_ i
f, f,_
( 10 - 88 )

In a similar fashion, we may write Eq. (10-85) as

a, ~ a,_, b, + b,_.
_ (10-89)
At 2

The vector b, contains the nodal velocities at the end of one time step At and is
unknown at the start of the solution. Therefore, if b, is eliminated between Eqs.
(10-88) and (10-89), we get the following convenient result:

2M K At
+ D +
At 2

2M K At f, At
+ 4_
f,
— At
+ D
j

a,-i + 2Mb, - 1
(10-90)
At 2 2 2

Equation (10-90) is useful because it provides the starting values of a, when i = I

for which we also have

a,-, = a0 = a(0) (10-91)

b,_, = b0 = b(0) (10-92)

f,_, = f(0) (10-93)

f, - f(Ar) (10-94)

Equations (10-91) and (10-92) represent the initial conditions on the nodal dis-
placements and velocities, respectively. The vectors f(0) and f(AT) in Eqs. (10-93)
and (10-94) denote the values of ffr) at t = 0 and t = At, respectively.

Boundary conditions
The three-point recurrence scheme given by Eq. (10-79) is of the form K eff a =
where K eff is
fefr , given by M 2
+ yDAr + \K(At) and feff is given by the right-
hand side of Eq. (10-79). As in the case of the two-point recursion formula, we
may use either Method 1 or Method 2 from Sec. 3-2 to impose the prescribed
displacement boundary conditions. Since K C ff is symmetric, it is prudent to preserve
symmetry if Method is used. Note that an implicit solution for
1
the nodal dis-

placements results even if the lumped form of the mass matrix is used because

Table 10-3 shows that \ and y are never both zero.

614 TRANSIENT AND DYNAMIC ANALYSES

10-10 INTRODUCTION TO MODAL ANALYSIS

In Chapter 7 the finite element method was applied to several different problems
in static stress analysts. The resulting nodal displacements were then used to cal-
culate the element strains, stresses, etc. In this chapter several sections were devoted
to dynamic structural analysis In this case, the nodal displacements are computed
some set of initial conditions The element resultants could
as a function of time for
be computed vra the techniques from Chapter 7. In both cases, the prescribed
displacements could be imposed in a routine manner
Many times in dynamic structural analysts, the nodal displacements as a func-
tion of time are not really needed Instead, the natural frequencies of the sustained
vibrations are needed For example, consider the case of the mass and spring system
shown in Fig. 10*9(a) If the spring is assumed to be perfectly elastic and linear

(i.e , linear-elastic) and massless, then the governing differential equation is given

by

Mx + Kx = /(/) (10-95)

where x(t) is the displacement of the mass Af from the equilibrium position, K is

the spring constant (i e , the stiffness), and /(/) denotes the forcing function. If
damping is present as shown in Fig 10-9(b), then the governing equation is given
by

Mx + Ox + Kx = f{t) (10-96)

where D is the damping x and x represent the velocity and

coefficient Note that
acceleration of the mass, respectively Note also that Eq. (10-96) reduces to Eq
(10-95) if the damping is negligible.
The natural frequency of a system is defined to be the frequency at which the
system oscillates if the forcing function is identically zero Nonzero initial conditions
give rise to the oscillations in this casesystem has only one discrete mass
If the

such as that in Fig 10-9(a), then only one natural frequency results However, if

Wo

IT "TTuT
:

T
Figure 10-9 (a) A mass-spring system, and (b) a mass-spnng-damper system.
 INTRODUCTION' TO MODAL ANALYSIS 615

the system is distributed such as a vibrating string, then an infinite number of natural
frequencies exist. Because the element method approximates a continuous
finite

system by discrete elements (and hence masses), the stiffness-based finite element
method is only capable of yielding the natural frequencies on the low end of the
spectrum. In contrast, the force matrix method (not covered in this book) yields
the natural frequencieson the high end of the spectrum. In both cases, the number
and accuracy of the computed natural frequencies increase as the number of elements
is increased.
Let us review how we obtain the natural frequencies by using Eq. (10-95) as
an example. Since we desire the oscillator) or sinusoidal variations of x, we may
represent x(t) in complex polar form as

T
A(i) = ,vc‘“ (10-97)

where from Euler's identity we have

,WT
e = cos cot + i sin cot ( 10-981

Here i denotes the imaginary number defined by

I = \^=T (10-99)

and to is the frequency of the vibration. In Eq. (10-97), x denotes the amplitude of
the vibration corresponding to the frequency co. Recall that the natural frequency

is obtained by assuming a zero forcing function. It follows that

Mx + Kx = 0 ( 10 - 100 )

and co then represents the undamped natural frequency. From Eq. (10-97), we get

and

,wt
(m) 2xe‘°* = -co ac
2
i
^= (10-101)

(10-100) as follows:
Using Eqs. (10-97) and (10-101), we may write Eq.

(-urM + K)xe""
r
= 0 ( 10-1021

MT we want to exclude ,v = 0 as a solution because

But e' is not generally zero, and
this would yield the trivial solution x(t)
= 0. Therefore, the expression in the

parenthesis must be zero, and we get

w - (10-103)

frequency
The reader may recall this result from elementary vibrations. The natural
eigenvalue. The nontrivial vector a is referred to
co is frequently referred to as an
61 6 TRANSIENT AND DYNAMIC ANALYSES

as the eigenvector that corresponds to the frequency a). A geometric interpretation

of the eigenvector is given later in this section. The natural frequency of the mass-
spring-damper system in Fig IO-9(b) may be obtained in a similar manner by
beginning with Eq (10-96). However, the eigenvalues in this case are now complex.
This implies that the oscillations decay with time.
Let us now return to Eq (10-16) and extend this development to the finite
element method. Only the undamped natural frequencies are to be determined here.
In this case, Eq (10-16) reduces to

Ma + Ka = 0 (10-104)

where the assemblage nodal force vector (the forcing function) has been set to zero
Now we represent the vector of nodal displacements as

a = ae'
0
" (10-105)

from which it follows that

a = (m) 7 2e ,'° 7 = -<a 7 Se

,UJ
(10-1061

where 5 is now referred to as the eigenvector corresponding to the natural frequency

<i> Equation (10-104) becomes

( - to 2 M + K)5e'“ T = 0 (10-1071

Since the nontrival solutions for a are sought, the determinant of the expression in

the parenthesis must be zero, or

det( -w 2
M + K) = 0 (10-108)

because this guarantees that the matrix -a> 2 M 4- K is singular, thus admitting
nontrivial solutions for the vector a Consequently, Eq (10-108) may be solved
for the natural frequencies w of the structure
Note that if N nodes are used in the discretization and each node has m degrees
of freedom, then Eq (10-108) results in a polynomial of order mN and hence yields
mN natural frequencies. Recall that the original (continuous) structure really has
an infinite number of natural frequencies It follows that the natural frequencies on
the high end of the spectrum (i e , for large <*>’s) are not accurately
computed.
Fortunately, the frequencies on the low end of the spectrum are quite accurate if
enough elements are used Standard library subroutines may be used
FORTRAN
to determine the mN values of to
2
(and hence to) that satisfy Eq (10-108) The
roots may be shown to be positive real numbers because the matrices K and M are
always positive definite The w’s are recognized to be eigenvalues.
Recall that the vector a Eq (10-105) may be referred to as the eigenvector
m
that corresponds to the frequency w. In particular, let us define a to be the eigen- ;

vector that corresponds to the natural frequency <o. It follows from Eq. (10-107)
that the eigenvector a, must satisfy

10 - 103 )
(— <o 2 M + K)a,e'“
T = 0 (
 INTRODUCTION TO MODAL ANALYSIS 6

T
But e'“ is not zero, so we must have

(«,
2
M - K)a, = 0 (10-11

Equation (10-110) has a nontrivial solution for the eigenvector a because the i -

are computed such that Eq. (10-108) holds; otherwise only the trivial solution f
aj exists. It should be recalled that the system of equations in Eq. (10-110)
singular. Therefore, the solution for the eigenvector is not unique and one of tl

entries in a, is arbitrary. Because the eigenvectors are not unique, they are general
normalized such that

afMSj — I

Standard FORTRAN library' subroutines may be used to compute the eigenvector;

The leader is refened to the book by Zienkiewicz ) 12} for a more detailed discussior
The eigenvectors may be interpreted geometrically as follows. Consider tf
simply supported beam shown in Fig. 10-l0(a). The eigevector a, that corresponc
to the lowest natural frequency w, results in the mode shape shown in Fig
10-10(b). The eigenvector a 2 that corresponds to the second lowest natural frequenc
ti>
2 results in the mode shape shown in Fig.
10- 10(c). Because the beam is contir

uous, there are really an infinite number of natural frequencies and eigenvectors c
mode shapes. The human eye sees the superposition of all mode shapes during th
vibration of a structure. If a system with negligible damping is excited at one o
the natural frequencies of the structure, resonance will ocur. The result is usual!

catastrophic.
it should be mentioned that the lumped form of the mass matrix ma;
Finally
be used to determine the natural frequencies and mode shapes of the structure
However, the results from the consistent mass matrix will generally be more accurati
[13].

to

Figure 10-10 (a) A simply supported beam with (b) its first mode shape and (c) its second

mode shape.
618 TRANSIENT AND DYNAMIC ANALYSES

REFERENCES

1. Zienkiewicz, O C , The Finite Element Method, McGraw-Hill (UK), London, 1977,

pp 60-62.
2. Hibbcler, R C , Engineering Mechanics- Dynamics, Macmillan, New York, 1983,
p 80.
3. Clough, R. W., and J Pcnzien, Dynamics of Structures, McGraw-Hill, New York,
1975.
4. Gresho, P and R L
, Lee, “Don’t Suppress the Wiggles —They're Telling You Some-
Com ection Dominated Flows, AMD-Vol 34, pp
thing,” Finite Element Methods for
37-61, Winter Annual Meeting of the ASME, New York, Dec. 2-7, 1979
5. Hinton, E., A. Rock, and O C Zienkicwicz, “A Note on Mass Lumping in Related
Process in the Finite Element Method," Int J Earthquake Eng. Struct Dynam., Vol.
4 pp 245-249. 1976
,

6. Zienkiewicz, O C The , Finite Element Method, McGraw-Hill (UK), London, 1977,

pp 535-539
7. Zienkiewicz, O C , The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp 576-588
8. Bathe, K 3 , Finite Element Procedures in Engineering Analysis, Prentice-Hall, En-
glewood Cliffs, N J , 1982, pp 553-554
9. Reddy, J N An , Introduction to the Finite Element Method McGraw-Hill, New York,
1984, pp 50-56
10. Zienkiewicz, O C , The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp 740-741
11. Zienkiewicz, O C , The Finite Element Method, McGraw-Hill (UK), London, 1977,
p 582
12. Zienkiewicz, O C . The Finite Element Method, McGraw-Hill (UK), London, 1977,
pp. 546-558
13. Reddy, J N An Introduction
. to the Finite Element Method, McGraw-Hill, New York,
1984, p 406

PROBLEMS

10-1 The element mass matrix, applicable

is given by Eq (10-13). Since this expression is

to any element, evaluate it for the case of the three-node lineal element by using
length coordinates Use Eq (6-48) to evaluate the integrals

10-2 The element mass matrix is given by Eq (10-13) Since this expression is applicable
using
to any element, evaluate it for the case of the three -node lineal element by
order
the screndtpity coordinate r Use Gauss-Legendre quadrature of such an
Hmf
that the integrals are evaluated exactly.

Since
10-3 Consider the expression for the element mass matrix given by Eq (10-13)
the t rec-
this expression is applicable to any element, evaluate it for the case of
of area coordinates
node triangular element by writing the shape functions in terms
an or cr t
Hint Use the numerical integration formula from Section 9-9 of such
the integrals are evaluated exactly
 PROBLEMS 619

10-4 Consider the expression for the element mass matrix given by Eq. (10-13). Since
this expression is applicable to any element, evaluate it for the case of the four-

node rectangular element by writing the shape functions in terms of the serendipity
coordinates r and Hint. Use Gauss-Legendre quadrature of such an order that the
.r

integrals are evaluated exactly.

10-5 The element capacitance matrix is given by Eq. (10-26) for all one-dimensional
elements. Evaluate the integral for the case of the three-node lineal element by using
length coordinates. Use Eq. (6-48) to evaluate the integrals.

10-6 One form of the element capacitance matrix is given by Eq. (10-26). Since this
expression is applicable to any one-dimensional thermal analysis problem, evaluate
it for the case of the three-node lineal element by using the serendipity' coordinate
r. Hint: Use Gauss-Legendre quadrature of such an order that the integrals are
evaluated exactly.

10-7 Consider the expression for the element capacitance matrix given by Eq. (10-25).
Since this expression is applicable to any two-dimensional thermal analysis problem,
evaluate it for the case of the four-node rectangular element by writing the shape
functions in terms of the serendipity coordinates r and s. Hint: Use Gauss-Legendre
quadrature of such an order that the integrals are evaluated exactly.

10-8 Consider the expression for the element capacitance matrix given by Eq. (10-25).
Since this expression is applicable to any two-dimensional thermal analysis problem,
evaluate it for the case of the eight-node rectangular element by writing the shape
functions in terms of the serendipity coordinates r and s. In other words, show that

Eq. (10-35) holds in this case. Hint: Use Gauss-Legendre quadrature of such an
order that the integrals are evaluated exactly. It may be more convenient to write
a short FORTRAN program.

10-9 Determine the lumped form of the mass matrix for Problem 10-1. Clearly explain
or indicate how the lumped form is obtained from the consistent form.

10-10 Determine the lumped form of the mass matrix for Problem 10-2. Clearly explain
or indicate how the lumped form is obtained from the consistent form.

10-11 Determine the lumped form of the mass matrix for Problem 10-3. Clearly explain
how the lumped form is obtained from the consistent form.
or indicate

10-12 Determine the lumped form of the mass matrix for Problem 10-4. Clearly explain
how the lumped form is obtained from the consistent form.
or indicate

10-13 Determine the lumped form of the capacitance matrix for Problem 10-5. Clearly
how the lumped form is obtained from the consistent form.
explain or indicate

10-14 Determine the lumped form of the capacitance matrix for Problem 10-6. Clearly
explain or indicate how the lumped form is obtained from the consistent form.

10-15 Determine the lumped form of the capacitance matrix for Problem 10-7. Clearly
how the lumped form is obtained from the consistent form.
explain or indicate

10-16 Determine the lumped form of the capacitance matrix for Problem 10-8. Clearly
is obtained from the consistent form.
explain or indicate how' the lumped form
 J

620 TRANSIENT AND DYNAMIC ANALYSES

1 0-17 Comsder the following assemblage capacitance and stiffness (or conductance) mat-
rices

and the following assemblage nodal force vector:

f = [200]
[200
Using the forward-difference (or Euler's) method, determine the nodal temperatures
0 25-sec time step is used Assume that there are
for the first four time steps if a
no prescribed temperatures and both nodes are initially at 50°F. Explain what might
happen if the time step is increased too much

10-18 Consider the following assemblage capacitance and stiffness (or conductance) mat-
rices

and the following assemblage nodal force vector

Using the forward-difference (or Euler’s) method, determine the nodal temperatures
for the first five time steps if a 0 2-sec time step is used Assume that there arc no
prescribed temperatures and both nodes are initially at 25°C Explain what might
happen if the time step is increased too much

10-19 Consider the following assemblage capacitance and stiffness (or conductance) mat-
rices

"500 250 0 400 -400 0

250 1000 250 and K = -400 800 -400
0 250 500 L o -400 400

and the following assemblage nodal force vector

Using the forward-difference (or Euler’s) method, determine the nodal temperatures
for the first 20 time steps if a 0 1-sec time step is used Assume that node 1 is to
be prescribed at 100°C and that all nodes are initially at 10°C Explain what might
happen if the time step is increased too much
1 0-20 Consider the following assemblage capacitance and stiffness (or conductance) mat-
rices

10 5 o" -100 o’
88
c = 5 20 5 and K = 200 -100
.
0 5 10 o -100 100
 pRoni fms 621

and the following assemblage nodal force sector

8
f = 16
8

Using the forward-difference (or Euler's) method, determine the nodal temperatures
for the first five time steps if a 0 05-sec time step is used Assume
that node is I

to be prescribed at 150°F and that all nodes are initially at 75°F Explain what might

happen if the time step is increased too much

10-21 Repeat Problem 10-17 with the backward difference scheme and a 0 5-sec time
step Explain what happens to the results as the time step is increased

10-22 Repeat Problem 10-18 with the backward difference scheme and a 0 5-sec time
step. Explain what happens to the results as the time step is increased

10-23 Repeat Problem 10-19 for 10 time steps with the backward difference scheme and
a 0.20-sec time step Explain w’hat happens to the results as the time step is increased

10-24 Repeat Problem 10-20 with the backward difference scheme and a 0 05-sec time
step Explain what happens to the results as the time step is increased

10-25 Repeat Problem 10-17 with the central difference scheme and a 0 5-scc time step
Explain what happens to the results as the time step is increased

10-26 Repeat Problem 10-18 with the centra! difference scheme and a 0 5-sec time step
Explain what happens to the results as the time step is increased

10-27 Repeat Problem 10-19 with the centra! difference scheme and a 0 20-scc time step
Explain what happens to the results as the time step is increased

10-28 Repeat Problem 10-20 with the central difference scheme and a 0 05-scc lime step
Explain what happens to the results as the time step is increased

10-29 Show that point collocation at time 7, gives 8 = 0 Which of the finite difference
1

schemes does this situation correspond to ’ Please explain

10-30 Show that point collocation at time t, + t

gives 0 = 1 Which of the finite difference

schemes does this situation correspond to 9 Please explain

10-31 Show that the parameter 8 takes on a value of !A if the weighting function is given

by Eq (10-67a) Is this value of 0 likely to result in a stable solution for any time
step At 9 Pleast explain the plausibility of your conclusion

10-32 Write a small FORTRAN program that can be used to obtain the transient temperature
distributions in a fin It is recommended that the program from Problem 4-78 be
used as the starting point Allow for different values of 0 In addition, allow for

lumped or consistent capacitance matrix (via an input parameter ILUMP.

cither the
where ILUMP is 0 and for a consistent and lumped capacitance maim, respec-
I

tively)

10-33 With computer program from Problem 10-52 (or one furnished by
the help of the
(lie Example 10 6 for 0 = 0 and for both a consistent and lumped
instructor) repeat
steady
capacitance matrix Use a 0 I-scc time step and carry out the solution until
622 TRANSIENT AND DYNAMIC ANALYSES

slate is achieved (approximately 6 sec). Is the solution stable? How do the results
compare with those from Example 10 6 (with 0 = %)?

10-34 With the help of the computer program from Problem 10-32 (or one furnished by
the instructor) repeat Example 10-6 for 0 = Vi and for both a consistent and lumped
capacitance matrix. Use a 0 1-sec time step and carry out the solution until steady
achieved (approximately 6 sec) Is the solution stable?
state is How do the results
compare with those from Example 10-6 (with 0 = 2/s)?

10-35 With the help of the computer program from Problem 10-32 (or one furnished by
the instructor) repeat Example 10-6 for 0 = 1 and for both a consistent and lumped
capacitance matrix Use a 0 1-sec time step and carry out the solution until steady

state is achieved (approximately 6 sec). How do the results compare with those
from Example 10-6 (with 0 = Vsp
10-36 With the help of the computer program from Problem 10-32 (or one furnished by
the instructor) repeat Example 10-6 for 0 = Vs and for both a consistent and lumped
capacitance matnx Use a 0 1 see time step and carry out the solution until steady
state is achieved (approximately 6 see) Is the solution stable ^ How do the results
compare with those from Example 10-6 (with 0 = %)?

10-37 Show that X = 0 and -y = Vi for point collocation at t,

10-38 Show that X = 1 and -y = Vi for point collocation at t, + 1

10-39 Show that X = l

/io and -y = '/: for Galerkin weighting corresponding to Eq
(10-83b)

10-40 Show that X = *A and -y = Vi for GalerKm weighting corresponding to Eq

(10-83c)
 Structural and Thermal Properties

A-1 Structural Properties

English Units

Modulus of Coefficient of Poisson's Weight

elasticity thermal expansion ratio density
6 6 c
10 psi lO^ m in- F Ibf/in
’

Aluminum. 6061 alloy 10. 12 8 0.33 0 100

Brass 15. 10 5 0 33 0 300
Bronze 15. 10 0 0 33 0 300
Cast iron 15. 60 0 27 0 256
Copper, hard drawn 17. 9 4 0 33 0.340
Steel, hot rolled low carbon 30. 6 5 0 30 0 283
Steel, hot rolled high carbon 30. 7 0 0 30 0 283
624 APPENDIX A STRUCTURAL AND THERMAL PROPERTIES

SI Units

Modulus of Coefficient of Poisson’s Weight

elasticity thermal expansion ratio density
lo 6
10 N/m 2
iO' nVm-°C N/cm 3

Aluminum, 6061 alloy 690 23 0 0.33 0 0271

Brass 10 3 18 9 0 33 0 0814
Bronze 10 3 18 0 0 33 0 0814
Cast iron 10 3 10 8 0 27 0 0695
Copper, hard drawn 1J 7 16 9 0 33 0 0923
Steel, hot rolled low carbon 20 7 11 7 0 30 0 0758
Steel, hot rolled high carbon 20 7 12 6 0 30 0 0758

A-2 Thermal Properties

English Units

Thermal Mass Specific

conductivity density heat
e 3
Btu/hr-ft- F lbm/ft Btu/lbm-°F

Aluminum (pure) 117 169 0 208

Brass 64 530 0 089
Bronze 15 540 0 079
Cast iron 30. 454 0 100
Copper (pure) 224. 558 0 092
Stainless steel 92 488 0 107
Steel, hot rolled low carbon 31 1 488 0 108
Steel, hot rolled high carbon 21 7 483 0 113
 Thermal
Mass
conductivity Specific
density
heat
W/m-°C kg/m 1

Aluminum (pure)
J/kc-T
Brass 203.
2720.
872.
Bronze III.
S520.
373.
Cast iron 26.
8670.
331.
Copper (pure) 52.
7305.
419.
Stainless steel 389.
8980 3S6.
Steel, hot rolled 16.
low carbon 7820.
449.
Steel, hot rolled 54.
hi.eh carbon 7830.
452.
38.
7750.
473.
 < , , ,

Program TRUSS

B-1 Listing of Source Code for

Program TRUSS*
program truss

(2
Sot 5?2? fficx^ (20?
R (30) ' ronK30) ' matelg ooj

O0MMOT1MODES,
TITLE (20)
iSS; SSf
NMAXLS, NPDIS, NPLDS, ^nc'
flNFV(40
'

LCCJT,
^ ^(4,4)
LPRINT, LIN,
C

6
8
S
l
^3 ™ SCREm DEVICE NUf-BER
'®ITE (LCCWSL,8)
FORMAT (///// 2x «TMEnm
PEAD (LCCNSL,20)
IF (LIN *B3-
"lin
* ^ NUMBER 0F THE INFOT FILE (6-10): ')

0) GO TO 9999

PfflD AN
80-COLUMN TITLE "Card"
PEAD (TrIN,10)
10 (TITT F C T) ' Jt =
TLE(J)
FORMAT (20A4) 1, 20)

S
C... PEAD THE NUMBER
C OF NOnpc: ELEMMS,
DISPLACEMENTS, TOIOT ? MATERIALS, PRESCRIBED
HEAD (Lm,20) CUTFOT UNIT
20 DES ' PEEffl/ NMAHS,
FORMAT (618) NPDIS, NPLDS, LOOT

w
* "* subrout,nes ln «* order ca Section B-2 describes
• The ln P u t is also
summanzcd
text in Table in the mam part of the

627
628 APPENDIX B PROGRAM TRUSS

C. ... ZERO COT THE "BOUNDARY CONDITION" FUGS

DO 1000 I = 1, 20
NBCX(I) = 0
NBCY(I) = 0
1000 CCNITMJE

C... . READ MESH, MATERIAL, AND BOUNDARY CCNDITION DATA

CALL NODGEN (XCOCR, YCOCR)
CALL ELEGEN (NODI, NODJ, MATFIG)
CALL MATERL CDAIWfil)
CALL BCCND (NBCX, NBCY, FORCE, DISP)

C HUNT SUMMARY OF INPUT DATA

CALL SUMMRY (XCCCR, YCOCR, NDDI, NODJ, MATFD3, DAIMAT, NBCX,
1 NBCY, FORCE, DISP)

C.... ZERO COT THE ASSEMBLAGE NODAL FORCE VECTOR AND STIFFNESS MATRIX

DO 2000 I « 1, 40
ANFV(I) = 0.0
DO 2000 J = 1, 40
ASM(I,J) * 0.0
2000 CCNTINJE

C GENERATE THE GLOBAL ELEMENT AND ASSEMBLAGE STIFFNESS MATRIX

DO 2500 L * 1, NELEM
CALL COORDS (L, NDDI, NODJ, XCOOR, YCCOR, XX, YY)
CALL LEXGTH (XX, YY, ELENTH)
CALL PROPTY (L, MATEIG, DATMAT, AREA, ELEDD)
CALL TRANSF (L, XX, YY, ELENTH, DIRCOS)
CALL STIFF (AREA, ELMDD, ELENTH, DIRCOS, ESM)
CALL ASSEMK (L, NODI, NDDJ, ESM, ASM)
2500 CCNTIMJE

C GENERATE ASSEMBLAGE NODAL FORCE VECTOR

CALL ANEVEC (NBCX, NBCY, FORCE, ANFV)
C IMPOSE RESTRAINTS ON THE NODAL DISPLACEMENTS
CALL PEBC (NBCX, NBCY, DISP, ASM, ANFV)
N = 2*NNODES

C SOLVE THE EQUATIONS, PRINT THE NODAL DISPLACEMENTS, AND PRINT

C THE ELEMENT RESULTANTS
CALL EQSOLV (N, ASM, ANEV, SOLN)
CALL PRINTN (SOLN)
CALL POSTER (NODI, NOOJ, XCOOR, YC03R, MATEIG, DATMAT, SOLN)
GO TO 6

9999 CCNTINUE
STOP
END

SUBROUTINE NODGEN (XCOOR, YCCOR)

READS NXW. COORDINATE DATA AND GENERATES NODES
DIMENSION XCCOR(20) , YCOOR<20)
CCMMDN NNODES, EELEM, NMATIS, NEDIS, NPIDS, IGUT, LPRINT, LIN,
TITLE (20)

READ (LIN,20) SUBT

20 FORMAT (A4)
 APPENDIX B PROGRAM TRUSS 629

25 CONTINUE
C.... READ STARTIN'! NODE NUMBER, INCREMENT, AND FINAL NODE NUMBER
READ (LIN, 30) NI, NO, NF
30 FORMAT (318)

C IF NI IS ZERO, RETURN SINCE NODE SPECIFICATION AND GENERATION

C IS COMPLETE
IF (NI ,B3. 0) GO TO 9999

C IF NG IS ZERO, NO GENERATION IS DESIRED. NODES ARE SPECIFIED

C CN A ONE BY ONE BASIS.
IF (NO .NE. 0) GO TO 40
READ (LIN, 35) XCOGR(NI) , YCOOR(NI)
35 FORMAT (4F8.0)
GO TO 25

40 CONTINUE
C NODAL GENERATION—XI, YI ARE THE STARTING COORDINATE PAIR AND
C XF,YF ARE THE ENDIN3 COORDINATE PAIR
READ (LIN, 35) XI, YI, XF, YF
DIV = (NF - NI) / N3
DX = (XF - XI) / DIV
DY = (YF - YI) / DIV
NDIV = DIV - 1
NN = NI + NG
Nil = NI
XCOOR(NI) = XI
YCOOR(NI) = YI

DO 1000 1=1, NDIV

XCOOR(NN) = XCGOR(NII) + DX
YCOOR(NN) = YCOOR(NII) + DY
Nil = NN
NN = NN + NO
1000 CONTINUE

C.... FINAL COORDINATES DEFINED DIRECTLY TO AVOID ROUNDOFF ERROR

XCOOR(NF) = XF
YCOOR(NF) = YF
GO TO 25

9999 CONTINUE
RETURN
END

SUBROUTINE ELEGEN (NODI, N9DJ, MATFIG)

C. . . . READS ELEMENT DATA AND GENERATES THE ELEMENTS
DIMENSION NODI (30) , N0DJ(30), MATFIG (30)
COMMON NNODES, NELEM, EMAILS, NPDIS, NPLDS, LOUT, LPRINT, LIN,
1 TITLE (20)

READ (LIN, 20) SUBT

20 FORMAT (A4)

25 CONTINUE _
C.... READS THE STARTING ELEMENT NUMBER, MATERIAL SET FLAG, ELEMENT
C INCREMENT, FINAL ELEMENT NUMBER, AND NODAL INCREMENT
630 APPENDIX B PROGRAM TRUSS

REM) (LIN, 30) LI, MS, IG, LF, N3

30 FORMAT (518)

C., .. IF LI IS EQUAL TO ZERO, TEEN ELEMENT DATA IS COMPLETED

IF (LI .E£. 0) GO TO 9999

C. . . . READ TEE GLOBAL NODE NUMBERS OF ELEMENT LI

READ (LIN, 30) NI, MI
NODI (LI) » NI
NODI (LI) = NI
MATFLG(LI) = MS

C.... IF US IS ZERO, INPUT ELEMENT DATA ON AN ELEMENT-BY-ELEMENT BASIS

C CTHE3WISE, GENERATE OTHER ELEMENT DATA
IF (LG .03. 0) GO TO 25
50 CONTINUE

LI » LI + LG
NI ° NI + N3
NI NT + N3

IF (LI .GT. LF) GO TO 25

NODI (LI) - NI
NODJ(LI) = NI
MATFLG(LI) = MS
GO TO SO

9999 CONTINUE
RETURN
END

SUBROUTINE MATERL (DATMAT)

C READS IN THE MATERIAL PROPERTY DATA
DIMENSION DATMAT (5,2)
COMMON NNODES, NELEM, NMATLS, NPDIS, NPLDS, LOUT, LPRINT, LIN,
1 TITLE (20)

READ (LIN, 20) SUBT

20 FORMAT (A4)

C READ FOR MATERIAL SET MSNO: CROSS-SECTIONAL AREA [DATMAT (MSNO,l) 3

C AND MODULUS OF ELASTICITY [DATMAT (MSNO, 2)1

30 CONTINUE
READ (LIN, 40) MSNO, AREA, ELMOD
40 FORMAT (18, 2F8.0)
IF (MSNO .EQ. 0 .CR. MSNO .GT. 5) GO TO 9999
DATMAT (MSNO, 1) = AREA
DATMAT (MSNO, 2 ) = ELMOD
GO TO 30
9999 CCNTTHJE
RETURN
EtT)

StBRCUTINE BCCND (NBCX, NBCY, FORCE, DISP)

READS AND GENERATES BOUNDARY CONDITION FLAGS, NODAL FORCES, AND
NODAL DISPLACEMENTS
DIMENSION NECXC20), NBCY(20), FORCE (15) , DISP(5)
 APPENDIX B PROGRAM TRUSS £31

COMMON nngdes, nelem, nmatls, npdis, nplds, loot, lprint, LIN,

1 TITLE (20)

C. ... ZERO CUT THE N3CX AND ffiCY ARRAYS

DO 10 I = 1, 20
NBCX(I) = 0
NBCY(I) = 0
10 CCNTIHJE

READ (LIN, 20) SOBT

20 FORMAT (A4)

C READS THE STARTING NODE NUMBER, EC FLAGS IN X- AND

C Y-DIRECHCNS, NODAL INCREMENT, AND ENDING NODE MTBER
25 READ (LIN, 30) NI, IBCX, IBCY, N3, NF
30 FORMAT (518)

C IF NI IS ZERO, EC FLAG INPUT IS COMPLETE, READ NODAL FORCES

IF (NI .EQ. 0) GO TO 200

C IF NG IS ZERO, NO GENERATION OF BC'S IS DESIRED. EC INFO IS

C SPECIFIED ON A NODE-BY-NODE BASIS
IF (NG .NE. 0) GO TO 100
NBCX(NI) = IBCX
NBCY(NI) = IBCY
GO TO 25

C.... GENERATION PORTION OF BC'S FLAGS

100 CONTINUE
NN = NI
120 CCNTINUE
NBCX(KN) = IBCX
KBCY(NN) = IBCY
NN = NN + MG
IF (NN .GT. NF) GO TO 25
GO TO 120

200 CCNTINUE
C READ APPLIED NODAL FORCES
READ (LIN, 20) SUBT

C READ FORCE FLAG AND FORCE

220 CCNTINUE
READ (LIN, 230) NFQRCE, FORC
230 FORMAT (18, F8.0)
IF (NFQRCE .EQ. 0 .OR. NFQRCE .GT. 15) GO TO 300
FORCE (NFGRCE) = FCRC
GO TO 220
300 CCNTINUE
C.... READ PRESCRIBED NODAL DISPLACEMENTS
READ (LIN, 20) SUBT

C ?EAD DISPLACEMENT FLAG AND DISPLACEMENT

320 GCNTTNDE
(EAD (LIN,230) NDISP, DISPL
[F (NDISP .EQ. 0 .OR. NDISP .GT. 5) GO TO 9999
DISP (NDISP) = DISEL
GO TO 320
 , , ,) ,

632 APPENDIX B PROGRAM TRUSS

9999 CONTINUE
‘RETURN
EM)

SUBROUTINE SUMMRY (XCCOR, YCCOR, NODI, NOQJ, MATFLG, DA1MAT,

1 NBCX, NBCY, FORCE, DISP)
DIMENSION XCCOR (20) , YCOOR(20) , NODI (30), NODJ(30), MATFLG (30)
1 DATMAT(5,2>, NBCX(20), N3CY(20>, DISP (5), FORCE Q5)
COMMON NNODES, NELEM, NMATLS, NEOIS, NFLDS, LCUT, LPRINT, LIN,
1 TITLE (20)

WRITE (LOUT, 100) TITLE

100 FCRMAT (2X, 20A4, /)

WRITE (LOUT, 110) NNODES, NELEM, NMATLS, NH)IS, NFLDS, IOJT

FORMAT (8X, ’NUMBER OF MODES: ', 13, /,
8X, 'NUMBER OF ELEMENTS: *, 13, /,
8X, 'NUMBER OF MATERIALS: 13, /,
8X, 'NUMBER OF PRES DISP: 13, /,
8X, ’NUMBER OF FT LOADS: *, 13, /,
8X, 'OUTPUT UNIT NUMBER: ', 13)

200 WRITE (LCUT, 230)

230 FORMAT (/, 2X, ’NODE NO.', 4X, 'IBCX', 2X, 'IBCY', 5X,
1 'X-COORD', 6X, 'y-COCRD')
DO 250 I * 1, NM3DES
WRITE (LCUT, 240) I, NBCX(I), NBCY(I), XCCOR(I) , XCCOR(I)
240 FORMAT <2X, 14, 7X, 14, 2X, 14, 4X, 2G13.5)
250 CONTINUE

300 WRITE (LCUT, 320)

320 FORMAT (/, 2X, 'ELEMENT NO.', 4X, ’NOTE I’, 3X, 'NODE J'
1 8X, 'MAT SET FLAG')
DO 350 I = 1, NELEM
WRITE (LOUT, 340) I, NDDI(I), NOOJ(I), MATFIG(I)
340 FCRMAT (16, 10X, 14, 5X, 14, 13X, 14)
350 CONTINUE

WRITE (LCUT, 380)

380 FORMAT (/, 2X, 'MATERIAL', 7X, 'AREA', 7X 'ELASTIC MODULUS'
400 DO 450 1 = 1, NMATLS
TOUTCE UX57S, 420) DKSHKm,2>
420 FCRMAT (4X, 12, BX, G11.4, 4X, Gil. 4)
450 CONTINUE

WRITE (LCUT, 510)

510 FORMAT (/, 2X, 'SUMMARY OF DIFFERENT EXTERNAL LOADS' ,/
1 10X, 'NUMBER', 4X, 'NDDAL FORCE')
DO 550 I = 1, NPLOS
WRITE (LCUT, 520) I, FCRCE(I)
520 FORMAT (12X, 12, 5X, G13.4)
550 CONTINUE

WRITE (LCUT, 610)

610 FORMAT (/, 2X, 'SUMMARY OF PRESCRIBED NODAL DISPLACEMENTS'
1 /, 10X, * NUMBER* , 4X, ’IDDAL DISP.')
DO 650 I = 1, NH3IS
WRITE (LOUT, 520) I, DISP (I)
 APPENDIX B PROGRAM TRUSS 633

650 CONTINUE

REIURN
END

SUBROUTINE COORDS (L, NODI, NODJ, XCCOR, YCCOR, XX, YY)

C. . . . DETERMINES X- AND Y-COCRDINATES FOR NODES I AND J OF ELEMENT L
DIMENSION NODI (30), NODI (30), XCOOR(20), YCCOR 20 ) , XX (2),
(

1 YY(2)

NDI = NODI (L)

NW = NODJ(L)
XX (1) = XCOOR(NDI)
XX (2) = XCOOR(NDJ)
YY(1) = YCOOR(NDI)
YY (2) = YCOOR(NW)
RETURN
END

SUBROUTINE LENGTH (XX, YY, ELENTH)

C. . . . CALCULATES THE DISTANCE BETWEEN TWO POINTS (ELEMENT LENGTH)
DIMENSION XX (2) , YY(2)
ELENTH = SQRT ((XX(1) - XX(2))**2 + (YY(1) - YY(2))**2)
RETURN
END

SUBROUTINE PROFIT (L, MATE1G, DATMAT, AREA, ELM3D)

C DEFINES THE CROSS-SECTIONAL AREA AND MODULUS OF ELASTICITY FOR
C ELEMENT L
DIMENSION MATFIG (30) , DATMAT(5,2)

NFLAG = MPTFLG(L)
AREA = DATT-1AT (NFLPG,1)
ELMOD = DATMAT (NFLAG, 2)
RETURN
END

SUBROUTINE TRANSF (L, XX, YY, ELENTH, DIRCOS)

C COMPUTES THE TWO DIRECTION COSINES
DIMENSION XX(2), YY(2), DIRCOS(2)
DIRCOS (1) = (XX (2) - XX (D) / ELENTH
DIRCOS (2) = (YY(2) - YY(1)> / ELENTH
RETURN
END

SUBROUTINE STIFF (AREA, ELM3D, ELENTH, DIRCOS, ESM)

C.... COMPUTES THE GLOBAL ELEMENT STIFFNESS MATRIX
DIMENSION DIROOSC 2), ESM (4, 4)

COEF = AREA * ELMOD / ELENTH

ESM(1,1) = GOEF * DIRCOS (1) * DIRCOS (1)

ESM (1,2) = COEF * DIRCOS (1) * DIRCOS (2)
ESM(1,3) = —ESM(lrl)
ESM(1,4) = -ESM(1,2)
634 APPENDIX B PROGRAM TRUSS

ESM(2 f l) = ESK (1,2)

ESM(2,2) = COEF * DIRCCS(2) * DIRCOSC2)
ESM(2,3) = -ESM(2,1)
ESM(2,4) = -ESM{2,2)

E5M(3,1) = ESM(1,3)
ESM(3,2) = ESM (2,3)
ESM(3,3) = -ESM(3,1)
ESM<3,4) - -ESM (3, 2)

ESM (4 r l) = ESM(1,4)
ESM(4,2) = ESM (2,4)
ESM (4 f 3) = ESM<3,4>
ESM(4,4) = ESM(2,2)

RETURN
END

SUBROUTINE ASSEMK (L, KDDI, NOW, ESM, ASM)

C.... ASSEMBLES THE ASSEMBLAGE STIFFNESS MATRIX
DIMENSION NODI (30) , N0QJ(30>, ESM(4,4), ASM(40,40)

NI - NODI (L)
NJ = NODJ(L)

ASM(2*NI-1,2 *NI-1 ) = ASM(2*NI-1,2*NI-1) + ESM(1,1)

ASM(2*NI-1,2*NI) = ASM ( 2*NI-1 , 2*NI ) + ESM(1,2)
ASM(2*NI,2*NI-1) = ASM(2*NI,2*NI-1) + ESM(2,1)
ASM(2*NI,2*NI) = ASM(2*NI,2*NI> + ESM(2,2)

ASM(2*NI-1,2*UJ-1) «= ASM(2*NI-1,2*NJ-1) + ESM(1,3)

ASM(2*NI-1,2*NJ) = ASM(2*NI-1,2*NJ) + ESM (1,4)
ASM(2*NI,2*NJ-1) = ASM(2*NI,2*NJ-1) + ESM(2,3)
ASM(2*NI,2*NJ) = ASM(2*NI,2*NI) + ESM(2,4)

ASM(2*NJ-1,2*NI-1) = ASM(2*NJ— 1,2*NI-1) + ESM (3,1)

ASM(2*NJ-1,2*NI) = ASM(2*NJ-1,2*NI) + ESM(3,2)
ASM(2*NJ,2*NI-1) = ASM(2*NJ,2*NI-1) + ESM(4,1)
ASM(2*NJ,2*NI) = ASM(2*NJ,2*NI) + ESM(4,2)

ASM ( 2*NJ-1 , 2*NJ-1) = ASM(2*NJ-1,2*NJ-1) + ESM(3,3)

ASM(2*NJ-l r 2*NJ) = + ESM(3,4)
ASM ( 2*NJ-1 , 2 *MJ )
ASM(2*NJ,2*NJ-1) = ASM (2*NJ,2*NJ-1) + ESM (4, 3)
ASM(2*NJ,2*NJ) = ASM(2*NJ,2*NJ) + ESM (4,4)

RETURN
END

SUBROUTINE ANFVEC (NBCX, NBCY, FORCE, ANFV)

C GENERATE ASSEMBLAGE NODAL FCRCE VECTOR DIRECTLY
DIMENSION NBCX (20) , NBCY(20), FCRCE (15) , ANFV(40)
COMMON NNODES, NELEM, NMATLS, NFDIS, NPLDS, LCDT, LPRINT, LIN,
1 TITLE (20)

DO 100 I = 1, NNODES
IFLAGX = NBCX (I)
IFLAGY = NBCY(I)
 APPENDIX B PROGRAM TRUSS 635

IF (IFLAGX .GE. 0) GO TO 50
IFLAG = -IFLAGX
ANFV (2*1-1) = FORCE (IFLAG)
50 CONTINUE
IF (IFLAGY .GE. 0) GO TO 100
IFLAG = -IFLAGY
ANFV(2*I) = FORCE (IFLAG)
100 CONTINUE

RETURN
END

SUBROUTINE PDBC (NBCX, NBCY, DISP, ASM, ANFV)

C APPLY RESTRAINTS ON THE NODAL DISPLACEMENTS
DIMENSION NBCX (20), NBCY(20), DISP (5), ASM(40,40), ANFV (40)
OOMMCN NNODES, NELEN, NMATLS, NPDIS, NPLDS, LOUT, LPRINT, LIN,
1 TITLE (20)

N2 = 2*NN0DES
DO 1000 1=1, NNODES
IFLAGX = NBCX (I)
IFLAGY = NBCY(I)
IF (IFLAGX .LE. 0) GO TO 500
DO 200 J = 1, N2
ASM(2*I-1,J) = 0.
200 CONTINUE
DO 300 J = 1, N2
ANFV(J) = ANFV(J) - ASM (J, 2*1-1) *DISP(IFLAGX)
ASM (J, 2*1-1) =0.0
300 CONTINUE
ANFV( 2*1-1) = LiSP (IFLAGX)
ASM(2*I-1, 2*1-1) = 1.
500 CONTINUE
IF (IFLAGY .LE. 0) GO TO 1000
DO 700 J = 1, N2
ASM (2*1, J) = 0.
700 CONTINUE
DO 800 J = 1, N2
ANEV(J) = ANFV(J) - ASM(J,2*I)*DISP(IFLASY)
ASM (J, 2*1) = 0.0
800 CONTINUE
ANFV(2*I) = DISP (IFLAGY)
ASM(2*I,2*I) = 1.
1000 CONTINUE

RETURN
END

SUBROUTINE EQSOLV (N, A, B, X)

C EQUATION SOLVER BY MATRIX INVERSION
DIMENSION A (40, 40), B(40) , X(40)
CALL INVDET (A, N, DDJRN, DEEM)
CALL MATVEC (N, A, B, X)
RETURN
END
636 APPENDIX B PROGRAM TRUSS

SUBROUTINE INVDET (C, H, DTNRM, DETM)

C.... MATRIX INVERSION SUBROUTINE—USED BY PERMISSION OF PRENTICE-
C HALL, INC., ENGLENOCD CLIFFS, N.J. FROM ROBERT W. HOKNBEICK,
C NUMERICAL METHODS, 1975, P. 295.

C INVERTS AN N BY N MATRIX C AND RETURNS THE INVERTED MATRIX BACK

C TO THE MATRIX C; DINRM IS THE DETERMINANT OF THE MATRIX DIVIDED
C BY THE EUCLIDEAN NORM; DETM IS SIMPLY THE DETERMINANT. THE
C DIMENSIONS OF J MUST BE AT LEAST 21 GREATER THAN THE RON OR
C COLUMN DIMENSION OF C. THE ROUTINE EMPLOYS GAUSS-JORDAN ELDI-
C INATICN WITH COLUMN SUITING TO MAXIMIZE THE PIVOT ELEMENTS.

DIMENSION C(40,40) , J(80)

PD = 1.
DO 124 L = I, N
ID = 0.
DO 123 K - 1, N
123 ED = ID + C(L,K)*C(L,K)
DD = SQRT (DD)
124 PD - PD*ED
DETM = 1.
DO 125 L - 1, N
125 J (Lf 20) = L
DO 144 L = 1, N
CC - 0.
M = L
DO 135 K = L, N
IF ((ABSCCC) - ABS(C(L,K))) .GE. 0.) GO TO 135
126 M = K
CC « C(L,K)
135 CONTINUE
127 IF (L .EQ. M) GO TO 138
128 K = J (M+20)
J(M+20) = J(Lf20)
J(L+-20) = K
DO 137 K = 1, N
S = C (K,L)
C(K,L) = C(K,M)
137 C(K,M) = S
138 C(L,L) = 1.
DETM = DETM* OS
DO 139 M = 1, N
139 C(L,M) = C(L,M) / CC
DO 142 M = 1, N
IF (L .EQ. M) GO TO 142
129 CC = C (M, L)
IF (ABSCCC) .LE. l.E-10) GO TO 142
130 C(M,L) = 0.
DO 141 K = 1, N
141 C(M,K) = C(M,K) - 02*C(L,K)
142 CONTINUE
144 CCNTINJE
DO 143 L = 1, N
IF (J(IrP20) .EQ. L) GO TO 143
131 M = L
132 M = M + 1
IF (J(ttf20) .EQ. L) GO TO 133
136 IF (N .GT. M) GO TO 132
 APPENDIX B PROGRAM TRUSS 637

133 JO-B-20) = J(Irf-20)

DO 163 K = 1, N
CC = C(L,K)
C(L,K) = C (M,K)
163 C (M,K) = CS3
J (L+20) = L
143 CONTINUE
DETM = ABS (DETM)
ETNRM = DETM / PD
RETURN
END

SUBROUTINE MA3VEC(N, A, B, X)
C PAGE 217 "ENTRY MATVEC" OF CARNAHAN, LUTHER, AND WILKES
DIMENSION A(40, 40) , B(40), X(40)
DO 100 I = 1, N
100 X CD = 0.
DO 200 I = 1, N
DO 200 J = 1, N
200 X(I) = A(I, J) * B(J) + X(I)
RETURN
END

SUBROUTINE RRINTN (ARRAY)

C PRINTS THE NODAL DISPLACEMENTS IN THE OUTPUT
DIMENSION ARRAY (40)
COMMON NNODES, NELEH, NMAMS, NPDIS, NPLDS, LOUT, LPRINT, LIN,
1 TITLE (20)

WRITE (IjQCJT 25)

25 FORMAT (/, 2X, 'SUMMARY OF NODAL DISPLACEMENTS' ,/,
1 5X, 'NODE NO.', 3X, 'X-COMPONENT' , 5X, ’Y-COMPONENT')

DO 100 J = 1, NNODES
WRITE (LOUT, 50) J, ARRAY (2*J-1) , ARRAY (2*J)
50 FORMAT (4X, 15, 6X, G12.5, 4X, G12.5)
100 CONTINUE

RETURN
END

SUBROUTINE POSTER (NODI, NODI, XCCOR, YCOOR, MATFIG,

1 DATMAI, SOLN)
C
C
—
POSTPROCESSOR CALCULATES AXIAL ELONGATIONS, STRAINS, STRESSES,
AND FORCES
DIMENSION NODI (30) , NOEU(30), XCOOR(20)., YCOQR(20),
1 MATFIG(30), DATMAT(5,2), SOLN(40), DIRCOS(2),
2 XX (2) , YY (2)
COMMON NNODES, NELEM, NMATLS, NIDIS, NPLDS, IGUT, LPRINT, LIN,
1 TITLE (20)

WRITE (LOUT, 50)

50 FORMAT (/, 2X, 'SUMMARY OF ELEMENT RESULTANTS’,/,
1 3X, 'ELEMENT NO.’, 4X, 'ELONGATION', 5X, 'STRAIN',
2 5X, 'STRESS', 6X, 'FORCE')

DO 200 L = 1, NELEM
638 APPENDIX B PROGRAM TRUSS

C. . . . PRELIMINARY CALCULATIONS
CALL COORDS (L, NODI, NOOJ, XCOQR, YCCOR, XX, YY)
CALL LETCTH (XX, YY, ELENTH)
CALL PROPTY (L, MATFLG, DAIMAT, AREA, ELMDD)
CALL TRANSF (L, XX, YY, ELENTH, DIRCOS)

C. . . . CALCULATION OF ELEMENT ELONGATION

NI = NODI (LJ
NJ = NODJ(L)

UI - SOIM(2*NI - 1)
VI = SCLN(2*NI>
UJ = SOI2J(2*NJ - 1)
VJ * SOLN(2*NJ>

UIPR = DIRCOS (1)*UI + DIRCOS (2) *VI

UJPR = DIRCOS (1)*UJ + DIRCOS (2) *VJ
DELTA * UJPR - UIPR

C.... CALCULATION OF ELEMENT STRAIN

STRAIN = DELTA / EXENTH

C.... CALCULATION OF ELEMENT STRESS

STRESS * EL MOD * STRAIN

C CALCULATION OF ELEMENT FORCE

FORCE = STRESS • AREA

C.... PRINT THE ELEMENT RESULTANTS

WRITE (LOUT, 100) L, DELTA, STRAIN, STRESS, FORCE
100 FORMAT ( 5X, 14, 7X, G12.5, IX. G12.5, IX, 2012.5)

200 CONTINUE
RETURN
END

B-2 Description of Input to Program TRUSS

The purpose of this part of the appendix is to give the details on the input parameters
needed to run the TRUSS
program described in Sec. 3-6 Any consistent set of
units may convenient to think of the input as being divided into seven
be used It is

different sections The information provided to the program in each of these sections
is summarized below A ready reference is provided in Table 3-1 in the main text.

Section Description of Input

title (printed in
1 Contains two lines of input, the first of which is an 80 column

the output) and the second contains the “master control data" which define the
number of nodes, the number of elements, etc

2 Contains nodal coordinate information, the nodes could be defined both with
and

without nodal coordinate generation

 AmNDSXIS PROGRAM TRUSS 639

Section Description of Input

3 Contains the element data including the nodal connects lty as well as the materia!
set specification for eacli element, the element data could be defined both w ith

and without element generation

4 Contains the two “material properties” tcross-sectional area and elastic modulus)
for each “material” present.

5 Contains the “boundary condition” flags, which are defined in detail later, again
the flags could be generated or specified on a node-by-node basis

6 Contains the applied nodal forces.

7 Contains the imposed nodal displacements

Section 1 Input
Line — Format 20A4
1

TITLE
Line 2 — Format 618
NNODES NELEM NMATLS NPDIS NPLDS LOUT
where

TITLE = 80-column title

NNODES = total number of nodes

NELEM = total number of elements

NMATLS = number of different materials

(usually only 1)
NPDIS = number of different prescribed displacements
NPLDS = number of different point loads
Plus microcomputer,
LOUT = output unit (for example, on the Apple II

the console is 3)

Notes:
NELEM £ NMATLS < NPDIS s
1. Restrictions: NNODES s 20, 30, 5.

5, and NPLDS <15 ,

the only input section which is nor terminated with a blank line
2. This is

Section 2 Input
Line — Format 20 A4
1

SUBT
Line — Format
2, 4, 6, etc. 318

NI NG NF
Line — Format 4F8.0
3, 5. 7, etc.
XI Yl XF YF
where

appropriate identifier or subtitle

like “NODAL COORDINA'I L
SUBT =
DATA”
640 APPENDIX B PROGRAM TRUSS

NI = starting node number in the generation sequence (see note 2 below)

NG = nodal increment
NF = number of the final node to be generated in this sequence
XI = x coordinate of node NI,
YI = y coordinate of node NI
XF = x coordinate of node NF
YF = y coordinate of node NF
Notes:
1. Restrictions: NI s 20, NF ^ 20
2. If no generation is desired, then NG, NF, XF and YF need not be input
(or zero will do).
3 A mandatory blank line must end this input section.

4. See Sec. 3-6 for more details.

Section 3 Input
Line 1 —Format 20A4
SUBT
Line 2, 4, 6, etc —
Format 518
LI MS LG LF NG
Line 3, 5, 7, etc —Format 218
NI NJ

where

SUBT = appropriate identifier or subtitle like “ELEMENT DATA”

LI = starting element number in the element generation sequence (see note
2 below)
MS — material set flag to be set for all elements in this generation sequence
LG = element number increment
LF = number of the final element to be generated in the sequence
NG = nodal increment
“ ”
NI = global node number corresponding to node 1on element LI
NJ = global node number corresponding to node “j” on element LI

Notes*
1 Restrictions LI ^ 30, MS < 5, LF < 30, NI < 20, NJ < 20
2 If no generation is desired, then LG, LF and NG need not be input (or zero
will do).
3 A mandatory blank line must end this input section
4 See Sec. 3-6 for more details.

Section 4 Input
Line I —Format 20A4
SUBT
Line 2, 3, 4, etc. —Format 18, 2F8 0
MSNO AREA ELMOD
 APPl \D!X D PROGRAM TRLSS 641

where

SUBT = appropriate identifier or subtitle such as

“MATERIAL PROP
ERTY DATA"
MSNO = unique material set flag
AREA = corresponding cross-sectional area
ELMOD = corresponding elastic modulus

Notes
1 Restriction MSNO s 5
2 A mandatory blank line must end tins input section
3 Each material set may be defined only once

Section 5 Input
Line — Format 20A4
1

SUBT
Lines — Format
2, 3, 4, etc 518
NI IBCX IBCY NG NF
where

SUBT = appropriate identifier or subtitle like “BOUNDARY CONDITION

FLAG DATA”
NI = number of the starting node in this generation sequence (see note 2
below),
IBCX = boundary condition flag on the x degree of freedom which will be
assigned to all nodes in this generation sequence, the meaning of
each of the possible values of IBCX is given below

IBCX Meaning

-N The A'th force (see Section 6 Input) is applied in the v direction

0 The node (or nodes) is neither restrained nor “loaded" in the
v direction

N The A'th displacement (see Section 7 Input) is imposed m the

r direction

IBCY = boundary condition flag on the v degree of freedom which will be

assigned to all nodes in this generation sequence, the meaning of
each of the possible values of IBCY is identical to those for IBCX.
except the v direction is affected
NG = increment to be added to NI to get the next node number with the
same boundary' condition flags, which is in tum incremented again
to get the next node number and so forth (sec note 2 below)
NF = final node number whose boundary condition flags are set to IBCX
and IBCY m the generation sequence
642 APPENDIX B PROGRAM TRUSS

Notes:
1. Restrictions: NI < 20, NF < 20,- 15 < 1BCX <5, - 15 < IBCY < 5
2. If no generation is desired, then NG and NF need not be input (or zero will
do).
3. A mandatory blank line must end this input section

4. If the flags are not specified for one or more nodes, both flags are auto-
matically taken to be zero by the program

Section 6 Input
Line 1 —Format 20A4
SUBT
Line 2, 3, 4, etc —Format 18, F8 0
NFORCE FORCE
where

SUBT = appropnate identifier or subtitle like “NODAL LOADS”

NFORCE = load identification number
FORCE = corresponding force

Notes.
1 Restrictions 1 < NFORCE s 15
2. A mandatory blank line must end this input section

Section 7 Input
Line 1
— Format 20A4
SUBT
Line 2, — Format
3, 4. etc 18, F8 0
ND1SP DISP

where

SUBT = appropnate identifier or subtitle like “PRESCRIBED DISPLACE-

MENTS"
ND1SP = displacement identification number
DISP = corresponding displacement

Notes.
1 Restrictions 1 < NDISP £ 5
2 A mandatory blank line must end this input section.
 Active Zone Equation Solvers

SUBROUTINES ACTCOL and UACTCL were adapted from the subroutines of the
same name in O. C. Zienkiewicz’s book, The Finite Element Method, published
by McGraw-Hill Book Company (UK), 1977. These two subroutines are listed
below and require the dot-product function, FUNCTION DOT (also listed below).
These subprograms are used with the written permission of McGraw-Hill.
SUBROUTINES ACTCOL and UACTCL should be used when the assemblage
stiffness matrix is symmetric and unsymmetric, respectively. Recall from Sec.
6-8 in the text that the lower triangular coefficients of an unsymmetric stiffness
matrix are stored in the C array, with the diagonal entries set to unity (the actual
diagonal entries are stored in the A array). See Eq. (6-72) and Fig. 6-19 for an
example of the A and JDIAG arrays, and refer toTable 6-3 on page 280 for the
definition of the parameters A, B, C, JDIAG, NEQ, AFAC, and BACK.

SUBROUTINE ACTCOL (A, B, JDIAG, NEQ, AFAC, BACK)

LOGICAL AFAC, BACK
REAL JDIAG(I)
DIMENSION A(1), B(1)
C
C FACTOR A TO UT«D*U, REDUCE B
C
AENGY =0.0
JR = 0
DO 600 J = 1, NEQ
JD = JDIAG(J)
JH = JD - JR
IS = J - JH + 2
IF (JH - 2) 600, 300, 100
643
644 APPENDIX C ACTIVE ZONE EQUATION SOLVERS

100 IF (.NOT. AFAC) GO TO 500

IE = J - 1
K = JR + 2
ID = JDIAG(IS - 1)
C
C REDUCE ALL EQUATIONS EXCEPT DIAGONAL
C
DO 200 I = IS, IE
IR = ID
ID s JDIAG(I)
IH = MINO (ID - IR - 1, I - IS + 1)
IF (IH .LE. 0) GO TO 200
KKK = K - IH
KKL = ID - IH
A(K) = A(K) - DOT(A(KKK), A(KKL ) IH)
200 K = K + 1
C
C REDUCE DIAGONAL TERM
C
300 IF (.NOT. AFAC) GO TO 500
IR = JR + 1
IE r JD - 1
K = J - JD
DO 400 I = IR, IE
KKK = K + I
ID * JDIAG(KKK)
IF (A(ID) .EQ. 0.0) GO TO 400
D = A(I)
A(I) = A(I)/A(ID)
A( JD) = A( JD) - D*A(I)
400 CONTINUE
C
C REDUCE RIGHT-HAND SIDE
C
500 IF (BACK) B(J) = B(J) -DOT (A(JR 1),B(IS - 1), JH - 1)
600 JR = JD
IF {.NOT. BACK) RETURN
C
C DIVIDE BY DIAGONAL PIVOTS
C
DO 700 1=1, NEQ
ID = JDIAG(I)
IF (A(ID) .HE. 0.0) B(I) = B(I)/A(ID)
700 AENGY = AENGY + B(I)«B(I)»A(ID)
C
C BACK SUBSTITUTE
C
J = NEQ
JD = JDIAG (J)
800 D = B( J)
J = J - 1
IF (J .LE. 0) RETURN
 )
APPENDIX C ACTIVE ZONE EQUATION SOLVERS
645

JR = JDIAG (J)
IF ((JD - JR) .LE. 1) GO TO 1000
IS = J - JD + JR + 2
K = JR - IS + 1
DO 900 I r IS, J
KKK = I + K
900 B(I) = B(I) - A(KKK) C D
1000 JD = JR
GO TO 800
END
SUBROUTINE UACTCL (A, C, B, JDIAG, NEQ, AFAC, BACK)
REAL A(1), B(1), C(1), JDIAG(I)
LOGICAL AFAC, BACK
O
O FACTOR A TO UT«D*U, REDUCE B TO Y
O
JR = O
DO 300 J = 1 NEQ
,

JD = JDIAG(J)
JH = JD - JR
IF (JH .LE. 1) GO TO 300
IS = J + 1 - JH
IE = J - 1
IF (.NOT. AFAC) GO TO 250
K = JR + 1
ID = 0

OOO REDUCE ALL EQUATIONS EXCEPT DIAGONAL

DO 200 I = IS, IE
IR = ID
ID r JDIAG(I)
IH = MIN (ID - IR - 1, I - IS)
IF (IH .EQ. 0) GO TO 150
A(K) = A(K) - DOT (A(K - IH), C(ID - IH), IH)
C(K) = C(K) - DOT (C(K - IH), A(ID - IH), IH)
150 IF (A(ID) .NE. 0.0) C(K) = C(K)/A(ID)
K = K + 1

REDUCE DIAGONAL TERM

OOCJOOO
A( JD) = A( JD) - DOT (A( JR + 1), C(JR + 1), JH - 1)

FORWARD REDUCE THE R.H.S.

IF (BACK) B( J) = B( J) - DOT (C(JR +1), B(IS), JH - 1

JR = JD
IF (.NOT. BACK) RETURN

OOCJ
BACK SUBSTITUTION

J = NEQ
646 APPENDIX C ACTIVE ZONE EQUATION SOLVERS

JD = JDIAG(J)
500 IF (A{JD) .NE. 0.0) B(J) = B<J)/A(JD)
D = B(J)
J = J - 1
IF {J .LE. 0) RETURN
JR = JDIAG(J)
IF ((JD - JR) .LE. 1) GO TO 700
IS = J - JD + JR + 2
K = JR - IS + 1
DO 600 I s IS, J
600 B(I) = B(I) - A(I + K)*D
700 JD = JR
GO TO 500
C
END

FUNCTION DOT (A, B, N)

DIMENSION A(1), B(1)
DOT = 0.0
DO 100 I = 1, N
DOT = DOT + A(I)«B(I)
100 CONTINUE
RETURN
END
 1

INDEX
Absolute v iscoskv . 450 Assemblage step
ACTCOL. subroutine two dimensional heat conduction, 413—419
callinu parameters, two dimensional incompressible viscous
definitions (table). 2.50 flow, 455—456
list. 279 two dimensional stress analvsis. 302-304
source code (listing). 643-645 Associative properties, 24
use of. Average strains
substructuring, 339 axisjmmctnc stress ana!) sis. 315. 3)6
transient thermal anilssis, 500 plane stress and plane strain, 305
Active zone equation solver (see fqu.it ion sol three-dimensional stress analysis. 322
ver, active zone) Average stresses
Adjoint. 27 axisvmmetric stress anal) sis, 315, 316
AfAC. 279. 250 plane stress and plane strain, 305
Allowable distortion three dimensional stress anal>sis, 322
brief clement, 539 Axisvmmetric elements
quadrilateral clement. 534 shape functions, 265, 266
tetrahedral clement. 542 Axisvmmetric heal conduction. 421-430
triangular element. 534 boundar) conditions, 424, 425
Anisotropic heat conduction FEM formulation,
two-dimensional heat conduction program, examples, 426-430
462. 467 general, 422-426
axisvmmetric (stratified), 491. 492 governing equation, 422
problems, 4S9 A\is)mmetnc stress anal) sis, 306-316
theorj. 4SS. 4S9 circumferential strain, 306. 307
Anisotropic materials, 1 circumferential stress. 309
Annular fin. 180. 181. 470. 471 constitutive relationship, 309, 310
Application of prescribed displacements, definition. 306
two-dimensional truss. 72 dement nodal force vectors. 31 1-315
Approximate solution methods [see Integral bod) forces. 312, 313
formulations) line loads, 314, 315
Area coordinates prestresses (residual stresses), 312
definition, 251 surface tractions, 313, 314
integration formula. 268 thermal strains (self-strains), 31 1, 312
Area moment of inertia [see Moment of in- clement resultants. 315, 316
ertia) element stiffness matrix, 310, 311
Assemblage, 72 linear operator matrix. 308

647
648 INDEX

Axisymmetnc stress analysis (continued) example, 602-604

material properly matrix, 310 one-dimensional,
shape function matrix, 308 general, 583
strain-displacement relationships, 306-308 lineal element, 599
strain-nodal displacement matnx, 309 three-dimensional, 583
strain vector, 307, 308 two-dimensional,
general, 583
BACK, 279, 280 clement, 583, 384
Baltimore truss, 44 Cell Peclet number, 451
Banded matrix, 57 Circulation, 437
Bandwidth, 57 Circumferential strain, 306, 307
Bar elements, 45 Circumferential stress, 309
Bascule (truss), 44 Classification of matrices, 276
Beam elements, 45, 325-328 Coefficient of thermal expansion, 292, 293
Cofactors, 26
analysis of, 322-337 Column matrix, 29-30
elementary theory of, 322-325 Column vector, 20
element nodal force vectors, 331, 332 Commutative properties
element resultants, 332, 333 matrices, 24
element stiffness matrix, 331 variational calculus, 114
example, 333-337 Compatibility equation, 197, 198
Galerkin method, 329-332 Complete elements, 243
maximum bending stress, 333 Complex polar form, 615
maximum shear stress. 333 Composite element stiffness matrix
shape functions, 325-328 incompressible viscous flow, 453-455
Bending moment, 332 potential flow,
beam analysis, 324, 325 stream function formulation, 446
B matnx (see Strain-nodal displacement velocity potential formulation, 439
matrix) thermal analysis,
Body forces axisymmetnc heat conduction, 426
axisymmetnc stress analysis 312, 313 convection dominated flow, 451
plane stress and plane strain. 298 one-dimensional heat conduction, 378,
three dimensional stress analysis, 320 379
Body of revolution three-dimensional heat conduction, 433,
general discussion, 265 434
stress analysis,307 two dimensional heat conduction, 402,
thermal analysis, 421 403
Boundary conditions Composite nodal force vector
dynamic structural analysis, 613 incompressible viscous flow, 453-455
steady-stale heat conduction, potential flow.
axisymmetnc 424, 425 stream function formulation, 446
one-dimensional, 377, 378 velocity potential formulation. 439
three-dimensional, 432, 433 stress analysis, 214, 215
two dimensional, 401 plane stress and plane strain. 302
transient thermal anal) sis, 593 thermal analysis,
two-dimensional potential flow, 440. 441 axisymmetnc heat conduction. 426
Boundary convection convection dominated Bow. 45J
axisymmetnc hear conduction, 426 one-dimensional heat conduction, 378,
one-dimensional heat conduction, 379 379
three-dimensional heat conduction, 433, 434 three-dimensional heat conduction, 433.
two-dimensional heat conduction 402, 403 434
Boundary radiation two-dimensional heat conduction, 402.
axisymmetnc heat conduction, 490 403
onc-dimensiona! heat conduction, 379 Computational limes
three-dimensional heat conduction, 499 two-dimensional vs three-dimensional stress
two-dimensional heat conduction, 402, 403 analysis, 316
Brick elements (see also Elements, types of) Condensation, 339
numbenng convention, 263 Conduction heal (see Heat conduction)
Bridge trusses. 44 C -cominuous shape functions, 325-328
l

Conformable elements (see Conforming ele-

Cantilever truss, 44 ments)
Capacitance matnx Conforming elements, 242
assemblage of, 584 Consistent matnees, 584-586
axisymmetnc. 583 capacitance, 585. 586
consistent, 585, 586 mass, 585
example, 598-602 Constant strain triangle. 306
lumped, 586-588 Constitutive relationships
 11 ) 1

INDEX 549

Constitutive relationships continued Dynamic

( structural analysis, 577-581, 584-
Fourier’s law, 586, 604-617
axisymmetric form, 489 solution methods, 604-608
one-dimensional form, 173 boundary conditions, application of, 613
three-dimensional form, 430 initial conditions, use of,
two-dimensional form, 390 Crank-Nicolson starting method, 613
Hooke’s law, Euler starting method, 612
generalized, 199 special cases,
axisymmetric stress, 309, 310 Galerkin, 610, 61
plane stress and plane strain, 291-294 point collocation, 608, 609
three-dimensional stress, 319 subdomain collocation, 610
uniaxial stress, 198
Continuity equation: Eigenvalue, 614, 615
incompressible visous flow. 452 Eigenvector, 616, 617
two-dimensional potential flow, 437 Elastic modulus, 198
Convection dominated flow, 448-452 Elementary theory of beams, 323-325
UACTCL, application of, 45 Element capacitance matrix, 582, 583 (see
Convergence, 5 also Capacitance matrix)
Crank-Nicolson method, 592 Element characteristics:
Crank-Nicolson starting method, 612, 613 global, 71
Crystal growth, 199 local, 71
Cylindrical coordinate system, 265 Element conductance matrix (see Element stiff-
C’-continuous shape functions (see also Shape ness matrix)
functions) Element damping matrix, 580 (see also Damp-
properties (lowest order), 21 ing matrix)
Element generation:
D (see Material property matrix), 199 two-dimensional heat conduction, 457-459
Damping matrix, 580 two-dimensional stress analysis, 340, 341
Data storage: Element mass matrix, 580 (see also Mass
two-dimensional heat conduction, 459 matrix)
two-dimensional stress analysis, 342, 343 Element nodal force vectors:
Deflection function (see Parameter functions, axisymmetric stress analysis,
beam analysis) body forces, 312, 313
Degrees of freedom: line loads, 314, 315
two-dimensional truss, 57 prestresses (residual stresses), 312
Derivative: surface tractions, 313, 314
of a matrix, 131 thermal strain (self-strain), 311, 312
of a scalar with respect to a vector, 131, plane stress and plane strain,
132 body forces, 298
Determinant, 25-27 point loads, 301
Differential of a function, 1 13 prestresses (residual stresses), 297
Direct approach: surface tractions, 299, 300
three-dimensional truss, 73-78 thermal strain (self-strain), 296, 297
two-dimensional truss, 45-62 three-dimensional stress analysis, 320-322
Direction cosines, 33-35 Element numbering:
Direct iteration method, 371-374 two-dimensional heat conduction, 459, 460
examples, 373, 374 two-dimensional stress analysis, 342, 343
flow chart, 372 Element resultants, 72
Discretization, 5, 70, 71 potential flow, two-dimensional, 443, 444
two-dimensional truss, 45—47 stress analysis,
value of tt, 6-8 axisymmetric stress analysis, 315, 316
Displacement functions, 186, 240 beams, 332, 333
Displacement method (see Stiffness method) one-dimensional stress analysis, 219
Displacement vector, definition of, 197 plane stress and plane strain, 305
Distorted elements: three-dimensional stress analysis, 322
brick, 535 thermal analysis,
quadrilateral, 526 axisymmetric heat conduction, 491
tetrahedral, 539 one-dimensional heat conduction. 149-
triangular, 531 15!
Distributed loads (see also Surface tractions) two-dimensional heat conduction, 409
beam analysis, 324, 325 Element stiffness matrix,
Distributive properties, 24 axisymmetric heat conduction, 426
Divergence theorem, 385 axisymmetric stress analysis, 310, 311
Dot product, 32-33 convection dominated flow, 451
function DOT, 279 incompressible viscous flow, 453-455
Dynamic stress analysis (see Dynamic structu- plane stress and plane strain, 294
ral analysis) size of.
650 INDEX

Element stiffness matrix ( continued) cubic-order (10 nodes).

structural analysis, area cooid mates, 514
bnck elements, 537 linear-order (3 nodes). 249-253
quadrilateral elements, 528 area coordinates, 251-253, 513
tetrahedral elements, 542 global coordinates, 250, 251
triangular elements. 534 quadratic-order (6 nodes),
thermal analysis, area coordinates, 514
bnck elements, 537 Equation solver, active zone, 270-280
quadrilateral elements, 528 backward substitution, 273-275
tetrahedral elements, 542 forward elimination, 273, 274
triangular elements, 534 subroutine ACTCOL, 279, 280
three-dimensional heat conduction, 433, 434 subroutine UACTCL, 279, 280
three-dimensional stress analysis, 319 triangular decomposition, 270-273
three-dimensional truss, Equations of static equilibrium, 189-191
global, 77 Equilibrium equations, 189-191
local, 76 three-dimensional form, 190, 191
transformation, 73-77 two-dimensional form, 190
two-dimensional heat conduction, Equilibrium problems, 101
general model, 402, 403 Euler equation (see Euler-Lagrangc equation)
simple model, Eulenan approach, 240
Galerkin method, 398 Euler-Lagrange equation
variational, 396, 397 ordinary differential equations (ODE),
two-dimensional potential flow, fourth-order, 120, 121
stream function formulation, 446 second-order, 115, 116
velocity potential formulation. 439 partial differentia! equations (PDE),
two-dimensional truss, fourth-order, 486, 487
assemblage, 55-57 second order, 392
global, 51-53 Euler's identity, 615
local. 47-49 Euler's method, 590
transformation, 49-51 Euler starting method, 612
Elements, types of Explicit solutions, 590
bnck element, Extended surface (see One-dimensional heat
cubic-order (32 nodes). conduction, see also Fin, pin)
serendipity coordinates. 519, 520 Extrcmization, 107
linear-order (8 nodes).
serendipity coordinates. 264 265. 517 Fin, pm
quadratic-order (20 nodes). boundary conditions, 146
serendipity coordinates 518,519 element characteristics, 147-148
lineal element, element nodal force vector, 148
cubic-order (4 nodes), element stiffness (conductance) matrices,
length coordinates. 511 148
serendipity coordinates, 511 example, 151-155
linear-order (2 nodes), application of prescribed base tempera-
global coordinates, 244, 245 ture, 153. 154
length coordinates, 248. 249, 510 assemblage stiffness (conductance) mat-
serendipity coordinates, 247, 248 rix. 153
quadratic-order (3 nodes), element nodal force vectors, 152, 153
length coordinates, 511 element stiffness matrices, 152, 153
serendipity coordinates, 511 exact solution, 154. 155
rectangular, fin efficiency, 154
cubic-order (12 nodes), heat removal rate, 154
serendipity coordinates. 515 solution for nodal temperatures, 154
linear-order (4 nodes), table of results, 155
global coordinates, 254, 255 fin efficiency, 150. 151
serendipity coordinates, 255-257, 514, governing equations. 146
515 derivation of, 172, 173
quadrat ic -order (8 nodes), heat removal rate. 149-150
serendipity coordinates, 515 by differentiation, 149
tetrahedral, by integration, 150
cubic-order (20 nodes), simple fin model, 146-155
volume coordinates, 517 variational,
linear-order (4 nodes), 258-263 FEM, 173-175
global coordinates, 260, 261 global, 173
volume coordinates, 261-263, 516 method
Finite-difference
quadratic-order (10 nodes), comparison with FEM, 8-10
57/-
volume coordinates, 517 transient thermal analysis, applied to.
triangular. 579, 581-601
 1 1

INDEX 651

Finite element. 3-5 dimensional thermal analysis, 459-

tsso
Fink truss. 44 461
nrst-order Ritz method. 104-106 Green-Gauss theorem
Hrst-order Taj lor senes. 113 applications,
Flexibility method (see force mitnx method! axtsymnietnc heat conduction, 424
flexural rigidity, 325 consection dominated floss, 450
Fluid floss, viscous (see Incompressible fourth-order PDE, 486. 487
siscous floss) three-dimensional heat conduction, 432
Force matrix method. 15. 89 tsso-dimensional heat conduction,
Fourier's lass Galerkm method. 397, 400
axisymmetne analysts. 489 sanational, 392
one-dimensional, j 73 tsso-dimensional potential floss, 438
three-dimensional. 430 tss o dimensional potential floss , stream
tsxo-dimcnsional. 390 function. 446
Frame elements. 45 theory. 385-388
Functional. 107
Half-bandssidth, 57
Half-plane. 266
Heat conduction
Galerkm method axissmmetnc. 421-430
dynamic structural analssis. use in. 610 onc-dimensional,
example. 61 general model, 374-385
global basis, 128 simple model. 146— 1 55
example. 12S. 129 three-dimensional. 430-436
59S
transient analssis. use tn. 597. tsso dimensional
example, 59S general model, 398-421
Galerkm method (ITiM). 14. 15 simple model.
axissmmetnc heat conduction. 421-430 Galcrkin. 388-391. 397-398
beam analssis. 322-337 sanational. 388-397
consection dominated floss, 448-452 Heterogeneous materials ( see
dsnamic structural analssis, 610, 611 Nonhomogeneous materials)
example, 143-145 Higher-order elements, 509-520
element nodal force sector. 145 Htgher-order elements
clement stiffness matrix. 145 axisymmetric.
incompressible siscous floss. 452-456 rectangular clement. 514. 515
interpolation polynomials. 142, 143 triangular element. 5 1 2-5 14
introduction, 142. 143 one-dimensional,
onc-dimensional heat conduction, length coordinates, 510. 511
fin. 146-155 serendipity coordinates, 511,512
general model. 374-385 three-dimensional,
shape functions. 143 brick clement. 517-520
three-dimensional heat conduction. 430-436 tetrahedral element, 516, 517
transient thermal anal) sis. 582. 597. 598 tsso-dimensional,
tsso-dimensional potential floss. 436-445 rectangular element. 514. 515
tsso-dtmcnsional heat conduction, tnangular element, 512-514
general model. 39S-42I Homogeneous materials, 1

simple model, 397. 398 Hooke's lasv

Galcrkin method genera! stress state, 199

dsnamic structural analysis, use in. 610 uniaxial stress state. 198
example, 61 Hosse truss, 44
global basis, 128 Hydraulic diameter, 449
example, 128, 129
transient analssis, use in, 597, 598 Identity matrix, 20
example, 598 Implicit solutions. 590-592
Gauss-Legendre quadrature (.sec also Quad- Incompressible viscous flow, 452-456
rature) assemblage step, 455, 456
of, 544 continuity equation, 452
sampling points and sseights, table
FEM formulation, 453-455
Geometric boundary conditions
ordinary differential equations (ODE), Navier-Stokes equations, 452. 453
fourth-order, 120, 121 UACTCL. application of. 456
second-order, 117. 1 19 Indefinite matnx, 276
partial differential equations (PDE), initial conditions
dynamic structural analysts,
fourth-order, 486, 487
second-order, 392 Crank-Nicolson starting method, 612, 613
Global node numbers. 45 Euler starting method, 612
transient thermal analysis, 593
proper numbering,
Initial stresses (see Residual stress vector)
tsso-dimensional stress analysis. 342, 343
 1 1

652 INDEX

Inner product (see Dot product) Lateral convection,

Integra! one dimensional model, 148, 379
of a matrix, 13 two-dimensional model, 402, 403
as sum of other integrals, 133 Lateral imposed heat flux
Integral formulations (global), one-dirnensicmal model, 470
general concepts, 103, 104 two-dimensional model, 402, 403
method of weighted residuals, 121-130 Lateral loads (see Distributed loads)
Galerkin method, 128, 129 Lateral radiation
genera! concepts, 121, 122 one -dimensional model, 379
least squares method, 126-128 two-dimensional model, 402, 403
point collocation method, 122-124 Least squares weighted residual method
subdomain collocation method, 124-126 global basis, 126, 127
Rayleigh-Ritz method, 107-109 example, 127, 128
example, 108, 109 Length coordinates
Ritz method, 104-106 definition, 248
example, 105, 106 integration formula, 267
variational method, 107-109 Length of vector, 31
example, 108, 109 Lift, 437
Integral methods (see Integra! formulations) Line loads
Integrated term, 113 axisymmetnc stress analysis, 314, 315
Rayleigh-Ritz FEM, 136 Lineal elements (see Elements, types of)
Integration formulas Linear operator matnx
analytical, 267-270 axisymmetnc stress analysis, 308
area coordinates, 268 plane stress and plane strain, 291
length coordinates, 267 three-dimensional stress analysis, 318
volume coordinates, 269 Linear spring element (see Spring element)
Gauss Legendre quadrature, 542-546 Local normalized coordinates (see Length,
numerical, 542-559 Area, and Volume coordinates and
for rectangles and bricks, 542-546 Serendipity coordinates)
for triangles and tetrahedra, 553-556 Lumped matnees
Integration by parts 113 capacitance, 586-588
Inverse of a matnx. 27-28 mass, 586-588
Inotational flow condition, 437
Isoparametnc elements, 520-525 Mapping, isoparametnc, 522-524
Isoparametnc formulations allowable distortion,
one dimensional, bnck dement, 539
in three dimensional space, 574 quadnlateral element, 534
in two-dimensional space, 564-566 572- tetrahedral element, 542
574 tnangular element, 534
three-dimensional. 534-542 examples. 524, 525, 547-553
bnck element, 535-539 Mass matnx, 580
tetrahedral element. 539-542 assemblage of, 580
two-dimensional 525-534 consistent, 585
quadnlateral element. 525-530 lumped, 586-588
m three-dimensional space, 574-576 Mass-sprmg-damper system, 614
tnangular element, 530-534 Mass-spnng system. 614-616
Isoparametnc mapping (see Mapping, Matenal property matrix, 199
isoparametnc) axisymmetnc stress, 310
Isoparametnc mapping, example. 524 525 plane strain, 293
Isotropic matenals, 1 plane stress, 292
three-dimensional stress, 319
Jacobian matnx Matnees
classification of symmetnc real matrices,
bnck element, 536
quadnlateral element, 527, 528 276
tetrahedral element, 540 Matnx
tnangular element, 532, 533 adjoint, 27
JDIAG array, 278, 279 algebra, 21-24
addition, 22
associative property, 24
Kinematic relationships (see Strain- commutative property, 24
displacement relations) distnbutive property, 24
Kinematic viscosity, 450 multiplication.
Kutta-Joukowski condition, 437 matnx by scalar, 22
two matrices, 23
L matnx (see Linear-operator matnx) subtraction, 22
Lagrangian approach, 240 cofactors, 26
Lame constant, 292, 293 definition, 19
 1 1

INDEX 653

Matrix ( continued) subroutine PROPTY, 178

determinant. 25-27 subroutine VPROP, 178. 179
element, 19 general model, 374-385
equality, 21 examples, 379-3S5
identity, 20 FEM formulation. 376-379
inverse, 27-28 governing equation, 375
minors, 26 miscellaneous fins, 175-177. ISO, 1S1,
null, 20 468-470
orthogonal. 28 pm fin.
partitioning. 28. 29 boundary conditions, 146
transpose. 24 element characteristics. 147-14S
Matrix calculus: element nodal force vector, 148
derivative with respect to a scalar, 131 element stiffness (conductance) matrices,
derivative of a scalar with respect to a rec- 148
tor, 131. 132 example, 151-155
integral of a matrix. 131 application of prescribed base tempera-
integral as a sum of other integrals. 133 ture. 153. 154
Maximum bending stress. 333 assemblage stiffness (conductance) mat-
Maximum shear stress, 333 rix. )53
Mesh. 1 element nodal force vectors, 152. 153
Mesh generation, 46 element stiffness matrices. 152. 153
plane stress and plane strain. 340. 341 exact solution. 154. 155
two-dimensional heat conduction, 457-459 fin efficiency, 154
Mesh, graded. 11 heat removal rate, 154
Method of exhaustion. 5 solution for nodal temperatures, 154
Minimum potential energy principle (see Prin- table of results. 155
ciple of minimum potential energy) fin efficiency, 150, 151
Minors. 26 goxeming equations, 146
Modal analysis, 614-617 derivation of. 172, 173
Mode shapes, 617 heat removal rate. 149-150
Modulus of elasticity. 198, 292 by differentiation. 149
Mohr's circle. 192. 194, 195 by integrauon. 150
Moment of inertia, 325 variational.
finite element solution. 173-175
NASTRAN, 1, 12 global, 173
Natural boundary conditions: simple fin model. 146-155
ordinary differential equations (ODE), One-dimenstona] stress analysis, (see also Un-
fourth-order. 120. 121 iaxial stress member)
second-order. 117, 119 computer program,
partial differential equations (PDE). subroutine PROPTY. 236
fourth-order, 486, 487 subroutine VPROP. 236, 237
second-order. 392 example. 219-223
Natural frequency, 614 FEM formulation. 215-219
Navier-Stokes equations, 452. 453 Orthogonal matrix. 51
Negative definite matrix, 276
Negative faces. 188 Parameter functions. 239-243
Negative semidefinite matrix. 276 beam analysis, 325. 326
N matrix (see Shape function matrix) compatibility, 24). 242
Node generation: completeness, 242. 243
plane stress and plane strain, 340. 341 definition, 240
two-dimensional heat conduction. 457-459 restrictions. 240-243
Node numbering, restrictions on: weak formulations, 241
plane stress and plane strain, 342, 343 Parent element. 524
two-dimensional heat conduction, 459, 460 Partial discretization, 578. 579

Nonconformable elements, 242 Partitioned matrices. 28. 29

Nonconforming elements, 242 Pascal triangle, 512. 513
Nonhomogeneous materials. 1
Peclet number. 449
Normal stress: Piecewise continuous trial functions. 130
definition. 188 Pm fin (see Fin, pin and One-dimensional heat
sign convention. 18S. 189 conduction)
Null matrix, 20 Planar domain (in axisvmmetnc problems),
Numerical integration (see Quadrature) 266
Plane strain:
constitutive relationship. 293. 294
One-dimensional elements (see Elements,
types of) definition. 289
One-dimensional heat conduction: material propertj matrix. 293
computer program, 178. 179 shape function matrix. 2S9. 290
654 INDEX

Plane strain (continued) total potential energy, 200, 202

strain-nodal displacement matrix, 290, 291 Principle of virtual displacements, 185, 204-
thermal strains (self-strains), 292 206
Plane stress example, 204, 205
constitutive relationship, 292, 293 mathematical statement, 205
definition, 288 Principle of virtual forces, 204
material property matrix, 292 Principle of virtual work (see Principle of vir-
shape function matrix, 289, 290 tual displacements, or Principle of
strain-nodal displacement matrix, 290, 291 virtual forces)
thermal strains (self-strams), 292
Point collocation Quadrature
dynamic structural analysis, use m, 608, brick element, 545, 546
609 Gauss-Legendre,
example, 609 sampling points and weights, table of,
global basis, 122, 123 544
example, 123, 124 rectangular element, 542-553
transient analysis, use m, 596 required order, 558, 559
example, 596, 597 tetrahedral element, 554, 556
Pointer array (see JDIAG array) sampling points and weights, summary
Point heat source of, 556
axisymmctnc models, 430 triangular element, 553-555
computer program, implementation in, 463 sampling points and weights, summary
three dimensional models, 500 of, 555
two-dimensional models, 407-408 Quadrilateral averages, method of
Point loads two-dimensional heat conduction, 419-421
beams, 323, 324 two-dimensional stress analysis, 420
plane stress and plane strain, 301
three-dimensional stress. 321, 322 Rayleigh dampling, 581
Position vector, 31 Rayleigh Ritz finite element method
Positive definite matrix, 276 C‘-connnuous shape functions, 136
Positive faces, 188 C°-contmuous shape functions, 136
Positive semidefimte matrix 276 example, 136-142
Potential flow two-dimensional, 436-448 element nodal force vector, 137
circulation, 437 element stiffness matrix, 137
continuity equation, 437 interpolation polynomials, 133, 134
irrotational flow condition. 437 introduction, 133-136
Kutta Joukowski condition, 437 shape functions, 134, 135
lift, 437 Rayleigh-Ritz method, 102, 103 (see also
stream function formulation, Rayleigh Ritz finite element method)
application, 446-448 FEM, 133-136
general, 445, 446 example, 136-142
velocity potential formulation, general. 115-121
application, 444, 445 global. 107-109
element resultants, 443, 444 example, 107-109
examples, 439-443 Rectangular elements (see also Elements, types
general, 437-439 of)
Prandtl number, 450 numbering convention, 253, 254
Pratt truss, 44 Rectangular prismatic element (see Brick ele-
Prescribed displacements, application of, ment)
plane stress and plane strain, 304 Redundant force method, 15
Prescribed temperatures, application of, Residual stresses (see Prestresses)
two dimensional heat conduction, 419 Resolution capability, 276
Pressure functions, 240 Resonance, 617
Prestress (residual stresses) Reynolds number, 449, 450
axisymmetnc stress analysis, 312 Ritz method (global), 102, 104-106
one dimensional stress analysis, 218 example, 105, 106
plane stress and plane strain, 297 Roof misses, 44
fhrce-dimensionat stress analysis, 320 Row matrix, 29-30
Primitive variables, 102, 456 Row vector, 20
Principal stresses, 191-195
three dimensional, 192, 193 Sampling points and weights
two-dimensional, 191, 192 Gauss- Lege ngre quadrature, 544
example, 193-195 serendipity elements, 544
Principle of minimum potential energy, 185, tetrahedral elements, 556
200-204 triangular dements, 555
Scalar product (of two vectors), 32-33
example, 202-204 (see
external potential energy, 200 also Dot product)
internal potential energy, 200 Self-strains (see Thermal strains)
656 INDEX

Tetrahedral element (continued) two-dimensional,

numbering convention, 258-263 application, 62-70
volume of, 261 application of displacement contramts.
Thermal diffusmty, 449 59-61
Thermal strain (self-strains): Method I, 59-60
axisymmetnc stress analysis, 309 Method 2 (penalty function), 60-61
plane strain, 293 application of loads, 58-59
plane stress, 292 assemblage, 55-58
three dimensional stress analysis, 319, 320 degrees of freedom, 57
Thermal stresses, 199 direction cosines, 51-53
Three-dimensional elements (see Elements, discretization, 45-47
types of) element resultants, 61-62.
Three-dimensional heat conduction. 430-436 axial elongation, 61-62
FEM formulation, examples, 435, 436 axial force, 61—62
general, 431-434 axial strain, 61-62
Fourier's law, 430 axial stress, 61-62
governing equation, 431 global element stiffness matnx, 51-54
Three-dimensional stress analysis, 316-322 half bandwidth, 57-58
computational times, 316 localelement stiffness matnx, 47-49
constitutive relationship, 319 transformation matnx, 49-51
displacement vector, 317 types, 44
element nodal force vectors, 320-322 TRUSS program, 79-87
element resultants, 322 Two-dimensional elements (see Elements,
element stiffness matrix, 319 types of)
linear operator matrix, 318 Two-dimensional heat conduction
material property matrix, 319 assemblage step,
shape function matrix, 318 example, 414-419
straw-displacement relationship. 317 computer program,
strain nodal displacement matrix, 318 anisotropic matenal, 462, 463
strain vector, 317 data storage, 459.460
thermal strains (self-strains), 319 main program, 457
Three-dimensional truss ( see Truss, three- mesh generation. 457-459
dimensional) node numbenng schemes, 459, 460
Three-point recurrence schemes. 604-613 (see point heat source, implementation of, 463
also Dynamic structural analysis) vanable property routine, 460-463
Time step, 589 element resultants, example, 419-421
Transient thermal analysis, 577-579, 581-604 general model, 409
ACTCOL, use of, 590 FEM formulation, 400-421
examples. 598-604 application, 410-421
prescribed temperatures, application of examples, 403-410
592-594 general development, 400-403
solution methods, 588-598 governing equation. 399
backward difference scheme, 591 schematic, 399
central difference scheme, 591, 592 quadrilateral averages, 420-421
difference schemes, generalized, 592-594 simple model,
FEM based, 594-598 boundary conditions. 389
Galerkm, 597, 598 Fourier’s law. 390
generalized solution, 596 Galerkin method, 397-398
point collocation. 596, 597 governing equation, 3 88
subdomain collocation, 597 variational formulation, 391-397
forward difference scheme. 589-590 Two-dimensional stress analysis (see also
Transpose of a matnx, 24 Plane stress and plane strain)
Trial functions, 14 computer program,
piecewise continuous, 14 data storage, 342. 343
requirements, 106 main program, 340
Triangular elements (see also Elements, types mesh generation, 340, 341
of) node numbenng schemes, 342, 343
area of, 251 Two node lineal element (see Elements, types
numbering convention, 249, 250 of)
Truss Two-point recurrence schemes, (see also Tran-
definition, 43-44 sient thermal analysis)
three-dimensional, based on FEM, 594-598
degrees of freedom, 75 based on finite differences, 589-594
direction cosines. 77-78 schematic of, 593
element resultants, 78
global element stiffness matnx. 76-77 UACTCL, subroutine
local element stiffness matrix, 75-76 application of, 451
transformation matnx, 73-74 calling parameters.
 INDEX 657

UACTCL, subroutine ( continued second-order, 392

definitions (table), 280 introduction, 110-115
list,279 miscellaneous rules, 114, 115
source code (listing), 645-6-16 natural boundary conditions,
Undamped natural frequency,615 ordinary differential equations (ODE),
Uniaxial stress member, tapered, 215-225 fourth-order, 120, 121
Uniaxial stress member: second-order, 117, 119
cancellation of internal forces, 223, 224 partial differential equations (PDE),
element nodal force vectors. 218 founh-order, 486, 487
element resultants, 21S second-order, 392
element stiffness matrix, 217 odd-order derivatives, 121
example, 218-225 power rule, 115
application of prescribed displacement, product rule, 1 15
222 quotient rule, 1 15
assemblage, 221, 222 variation of a functional, 1 14
discretization, 220 Variational method, 102, 103
element nodal force vectors, 221, 222 example. 108, 109
element resultants, 223 global. 107-109
clement stiffness matrices, 221, 222 Variational principles (see Variational calculus)
solution for nodal displacements. 222 Vector:
table of results, 225 algebra, 32
Galcrkin FEM, 170, 171 addition, 32
governing equation, 166, 167 dot product. 32-33
linear-operator matrix L, 216 subtraction, 32
self-strain(from temperature change), 217 definition, 29-30
shape function matrix, 216 direction cosines, 33-35
strain-nodal displacement matrix B, 216 equality, 31
stress-strain relationship, 217 length, 31
variational FEM, 168-170 position, 31
variational formulation (global), 167, 16S Velocity functions. 240
Unit vectors, 29-30 Virtual displacement principle (see Principle of
Unsteady heat conduction (sec Transient ther- virtual displacements)
mal analysis) Virtual force principle (see Principle of virtual
Unsteady thermal analysis (see Transient ther- forces)
mal analysis) Virtual svork principle (see Principle of virtual
Upwinding schemes, 451, 452 displacements, or Principle of virtual
forces)
Variable property routine: Viscosity:
one-dimensional heat conduction program, absolute, 450
178, 179 kinematic, 450
one-dimensional stress analysis program, Volume coordinates:
235-237 definition, 261, 262
two-dimensional heat conduction program, integration formula, 269
460-462 Vorticity formulation, 102
Variational calculus:
problem, 10-112
classical 1 Warren truss, 44
commutative properties, 1 14 Weak formulation. 108
definition, 102 Weighted-residual methods (global), 102, 103
Euler-Lagrangc equation, Galcrkin, 126-128
ordinary differential equations (ODE), general concepts, 121, 122
fourth-order, 120, 121 least squares, 126-128
second-order, 115, 116 point collocation, 122-124
partial differential equations (PDE), subdomain collocation, 124-126
fourth-order, 486, 487 Weights (see Sampling points and weights)
second-order, 392
functional, 107 XI. arrav:
geometric boundary conditions, two-dimensional heat conduction program,
ordinary differential equations (ODE), 459, 460
fourth-order, 120, 121 two-dimensional stress analysis program
second-order, 117, 119 342, 343
partial differential equations (PDE),
fourth-order, 486, 487 Young’s modulus, 198