Professional Documents
Culture Documents
Paul M. Kurowski - Finite Element Analysis For Design Engineers-SAE International (2023)
Paul M. Kurowski - Finite Element Analysis For Design Engineers-SAE International (2023)
Paul M. Kurowski - Finite Element Analysis For Design Engineers-SAE International (2023)
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
PAUL M. KUROWSKI
Disclaimer
Every effort has been made to provide an accurate text. The author and the manufacturers
shall not be held liable for any parts developed with this book or held responsible for any
inaccuracies or errors that appear in the book.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form Publisher
or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior Sherry Dickinson Nigam
written permission of SAE International. For permission and licensing requests, contact SAE Permissions,
400 Commonwealth Drive, Warrendale, PA 15096-0001 USA; e-mail: copyright@sae.org; phone: Product Manager
724-772-4028 Amanda Zeidan
ISBN-Print 978-1-4686-0535-8
ISBN-PDF 978-1-4686-0536-5
ISBN-epub 978-1-4686-0537-2
E-mail: CustomerService@sae.org
Phone: 877-606-7323 (inside USA and Canada)
724-776-4970 (outside USA)
Fax: 724-776-0790
Contents
Preface xix
CHAPTER 1
Introduction 1
1.1. What Is Finite Element Analysis? 1
1.2. What Is the Place of FEA among Other Tools of
Computer-Aided Engineering? 2
1.3. Fields of Application of FEA and Mechanism Analysis 2
1.4. Fields of Application of FEA and CFD 4
1.5. What Is “FEA for Design Engineers”? 4
1.6. Note on Hands-On Exercises 5
CHAPTER 2
viii Contents
CHAPTER 3
CHAPTER 4
Contents ix
4.5.3. WEDGE 61
Description 61
Objective 61
Procedure 61
CHAPTER 5
CHAPTER 6
Modeling Process 83
6.1. Modeling Steps 84
6.1.1. Definition of the Objective of Analysis 84
6.1.2. Selection of the Units of Measurement 84
6.1.3. Geometry Preparation 85
6.1.4. Defining Material Properties 85
6.1.5. Defining Boundary Conditions 86
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
x Contents
CHAPTER 7
Contents xi
CHAPTER 8
xii Contents
Procedure 135
8.4.2. L BRACKET02 135
Description 135
Objective 135
Procedure 135
CHAPTER 9
Contents xiii
CHAPTER 10
xiv Contents
CHAPTER 11
Contents xv
CHAPTER 12
xvi Contents
CHAPTER 13
CHAPTER 14
Summary 235
14.1. FEA Quiz 235
14.2. Frequently Asked Questions 238
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Contents xvii
CHAPTER 15
CHAPTER 16
Glossary 253
CHAPTER 17
Preface
During the 60+ years of its development, Finite Element Analysis (FEA) evolved from an
exotic analysis method accessible only to specialized analysts into a mainstream engineering
tool. Progress in computer hardware and operating systems combined with progress in
Computer-Aided Design made FEA available to design engineers for use during product
design process.
Many books have been written about FEA. At one end of the spectrum, we find books
going deep into theory, and at the other end of the spectrum, software manuals explaining
how to use a certain program. There is little FEA literature taking a “middle ground”
approach and specifically addressing the needs of design engineers who use FEA as an
everyday design tool. This book attempts to fill this void by focusing on understanding FEA
fundamentals, which are explained by simple yet meaningful examples. “Finite Element
Analysis for Design Engineers” takes a practical approach, characteristic of the attitudes
of design engineers, and offers the readers an opportunity to try out all discussed topics by
solving downloadable exercises using their own FEA program.
Finite Element Analysis is a very broad field of knowledge. Repetitions in discussing
concepts, methods, and techniques cannot be avoided. Some topics are discussed more
than once, taking advantage of a growing body of knowledge as the reader progresses
through the book.
1
Introduction
1.1. W
hat Is Finite Element Analysis?
Finite Element Analysis, commonly referred to as FEA, is a numerical method used for
analysis of structural and thermal problems encountered by mechanical engineers during
design is process. It is appropriate to start our discussion with definition of what design
analysis and how it relates to FEA. Design analysis is a process of investigating certain
properties of parts or assemblies. Design analysis can be conducted on a real object or on
models that represent certain aspects of the real object. If models are used instead of a real
object, the analysis can be conducted earlier in the design process before physical prototypes
are built. The models can be physical models (scale-down models, mock-ups, photoelastic
models, etc.) or mathematical models where a certain behavior of the design in progress is
captured and described by a mathematical apparatus. Simple mathematical models can
be solved analytically. More complex models require the use of numerical methods. FEA
is one of those numerical methods used to solve complex mathematical models. The FEA
has numerous uses in science and engineering, but the focus of this book is on structural
and thermal analysis. We will alternate between two terms and two acronyms that became
synonymous in the engineering practice: the Finite Element Analysis (FEA) and the Finite
Element Method (FEM).
The FEA is a powerful but demanding tool of engineering analysis. The expertise
expected of FEA users depends on the complexity of analysis but always requires familiarity
with Mechanics of Materials, Kinematics and Dynamics, Vibration, Heat Transfer,
Engineering Design, and other topics found in undergraduate mechanical engineering
curriculum. For this reason, many introductory FEA books offer the readers a quick review
of those engineering fundamentals. Rather than duplicating the efforts of other authors,
Chapter 15 refers readers to some of those books.
1.2. W
hat Is the Place of FEA among
Other Tools of Computer-Aided
Engineering?
FEA is one of many tools of Computer-Aided Engineering (CAE) used in mechanical design
process. Other CAE tools include Fluid Flow Analysis commonly called Computational
Fluid Dynamics (CFD) and Mechanism Analysis. These three major CAE tools, FEA, CFD,
and Motion Analysis, are integrated with Computer-Aided Design (CAD), which is the hub
for all CAE applications. Geometry, material properties, and certain boundary conditions
can be exchanged between CAD and add-ins and among add-ins themselves (Figure 1.1).
FEA, CFD, and Mechanism Analysis have been developed independently and are based
on different numerical techniques; the integration shown in Figure 1.1 is a relatively new
development. However, even if CAE tools are stand-alone programs and not add-ins to
CAD, they can still be interfaced with CAD.
FIGURE 1.1 CAE applications such as FEA, CFD, and Motion Analysis are add-ins to CAD.
They can exchange data with CAD and among themselves.
FEA
Motion
Analysis CFD
© SAE International.
CAD
1.3. F
ields of Application of FEA and
Mechanism Analysis
The main difference between FEA and Motion Analysis is the field of application. FEA is
used for analysis of deformable objects subjected to loads, and Motion Analysis is used for
analysis of motion of mechanisms.
A mechanism is not firmly supported and can move without deformation; components
of a mechanism can move as rigid bodies. On the contrary, any motion of a structure must
involve deformation of its components because a structure is, by definition, firmly supported.
The motion of a structure may take the form of a one-time deformation when a static load
is applied or oscillations about the position of equilibrium when a time-dependent load is
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
CHAPTER 1 Introduction 3
present. In short, a mechanism may move without having to deform its components, while
any motion of a structure must be accompanied by deformation. If an object cannot move
without experiencing deformation, then it can be classified as a structure. Examples of a
mechanism and a structure are shown in Figure 1.2.
Depending on the objective of analysis, an object or its components may be treated
either as a mechanism or as a structure. A helicopter rotor is a mechanism; it spins relative
to the hull. A rotor may be treated as a rigid body, assembly of rigid bodies, or assembly of
elastic bodies. An individual blade may be treated as a rigid body or as a structure if its
vibration characteristics need to be analyzed (Figure 1.3).
A rigid body cannot deform under load; it is a mathematical abstract convenient in
mechanism analysis but not applicable to structural analysis.
1.4. F
ields of Application of FEA
and CFD
In this book we discuss the FEA as a tool of structural and thermal analysis of solid bodies.
Heat transfer analysis clearly differentiates FEA from CFD. Heat transfer in solid bodies is
analyzed with FEA while heat transfer in fluids requires CFD.
1.5. W
hat Is “FEA for Design Engineers”?
What distinguishes “FEA for Design Engineers” from FEA performed by an analyst? To
set tone for the rest of this book, we will highlight the most essential characteristics of FEA
performed by design engineers.
•• FEA is one of many design tools: For design engineers FEA is one of many design
tools and is used along CAD, spreadsheets, catalogs, data bases, hand calculations,
textbooks, etc.
•• FEA is based on CAD models: A CAD model is the starting point for FEA.
•• FEA is concurrent with the design process: Since FEA is a design tool, it should
be used concurrently with the design process. It should keep up or, better, drive the
design process. Analysis iterations must be performed fast, and since results are used
to make design decisions, the results must be reliable even though not enough input
data may be available for analysis conducted early in the design process.
•• Limitations of “FEA for Design Engineers”: As we can see, the FEA used in the
design environment should meet quite high requirements. It must be executed
fast and accurately, even though it is in the hands of design engineers and not
FEA specialists. An obvious question is would it be better to have a dedicated
specialist perform FEA and let design engineers do what they do best: designing
new products? The answer depends on the size of organization, type of products,
company organization and culture, and many other tangible and non-tangible
factors. A consensus is that design engineers should handle relatively simple types
of analysis in support of design process. More complex types of analyses are usually
better handled by a dedicated analyst.
•• Objective of “FEA for Design Engineers”: The ultimate objective of using the FEA
as a design tool is to change the design process from iterative cycles of “design,
prototype, test” into a streamlined process where prototypes are used only for the
final design validation. With the use of FEA, design iterations are moved from the
physical space of prototyping and testing into the virtual space of computer-based
simulations (Figure 1.4). FEA is not the only tool of computer-aided simulation used
in the design process. There are others like CFD, Motion Analysis, etc. jointly called
the tools of Computer Aided Engineering (CAE).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
CHAPTER 1 Introduction 5
FIGURE 1.4 Traditional product development needs prototypes to support the design
process. A CAE-driven product development process uses numerical models, not physical
prototypes, to drive the development process.
PROTOTYPING PROTOTYPING
TESTING TESTING
© SAE International.
PRODUCTION PRODUCTION
1.6. N
ote on Hands-On Exercises
To assure an effective learning, it is necessary to complete hands-on exercises. Therefore,
topics discussed in this book are accompanied by simple, yet informative, examples
presented at the end of related chapters. The exercises are not specific to any software and
can be solved using any commercial FEA program. Completing exercises is essential in the
learning process. The list of exercises and downloading instructions are in Chapter 17.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
2
From CAD Model to
Results of FEA
2.1. F
ormulation of the Mathematical
Model
To underscore the importance of a mathematical model in the analysis process, it is impor-
tant to describe what a mathematical model is, where it fits in the design analysis process,
and how different it is from the CAD model and the Finite Element (FE) model.
Suppose we need to find displacements and stresses in a pulley. The CAD model defines
a volume which is the solution domain (Figure 2.1). The volume has material properties
assigned to it and certain conditions are defined on external faces.
The conditions defined on the external faces of a model are called boundary conditions.
The boundary conditions can be defined in terms of displacements and loads. Displacement
boundary conditions are also called essential boundary conditions, and load boundary
conditions are also called natural boundary conditions.
In our example, the displacement boundary conditions are defined on the inner cylin-
drical face as zero displacement to represent the support provided by a bearing; the bearing
itself is not modeled. Load boundary conditions are applied to a section of the outer cylin-
drical face as pressure and represent the belt load. The above displacement and load
boundary conditions are defined explicitly. All remaining surfaces have implicit load
boundary conditions: zero stress in the direction normal to the outside faces. This reflects
the fact that normal stresses must not exist on an unloaded surface.
FIGURE 2.1 A CAD model of an idler pulley presented as a volume (solid geometry).
Representing the model as a volume, and not as surfaces, affords the inclusion of modeling
details such as small rounds.
© SAE International.
The geometry illustrated in Figure 2.1 is not the only possible representation of the
analyzed pulley. Figure 2.2 shows another possibility where a pulley geometry is represented
by surfaces. This approach does not allow to model rounds.
In many analysis problems, more than one geometry representation is possible. A
pulley geometry lends itself to being represented with either volumes or surfaces. Modeling
a beam offers more choices as the beam can be represented by volumes, surfaces, or curves
as shown in Figures 2.3, 2.4, and 2.5.
The I beam shown in Figures 2.3, 2.4, and 2.5 illustrate the progressive idealization of
a three-dimensional (3D) geometry: from volume (solid geometry) fully representing all
three dimensions to surface geometry missing one dimension (thickness) and to curve
geometry missing two dimensions.
FIGURE 2.2 The idler pulley geometry is represented by surfaces. The surface geometry
is missing one dimension: thickness, which cannot be derived from geometry and must
be defined as a number.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 2.3 Beam represented by a volume (solid geometry). This affords the inclusion of
all details of geometry.
© SAE International.
FIGURE 2.4 Beam represented by surfaces cannot include details such as rounds.
Representing geometry by surfaces misses one dimension: thickness.
© SAE International.
FIGURE 2.5 Beam represented by a curve, shown here as a dotted line. In this idealized
model the beam cross-section properties are assigned to curves. The model contains no
details and can only represent the global beam behavior. Representing geometry by curves
misses two dimensions. Beam cross-section and the second moments of inertia cannot
be derived from geometry and must be defined as numbers.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Geometry alone does not fully define the mathematical model. Loads, restraints, and
material properties are also building blocks of the mathematical model and may be defined
in many ways to differentiate between different types of analyses such as static and dynamic
analyses, linear and nonlinear material, etc.
If the same problem can be described by more than one mathematical model, can
we say which model is the best? The best mathematical model is the one that adequately
represents the aspects of real design that are of interest to us, and does that at the lowest
cost. Consequently, a mathematical model must be constructed keeping in mind the objec-
tive of analysis. Creating a mathematical model that properly represents the data of interest
is the most important step in the modeling process. Yet its importance is often overlooked
in practice or worse, the distinctions among the CAD model, mathematical model, and FE
model are not recognized.
2.2. S
electing the Numerical Method to
Solve the Mathematical Model
Completion of a mathematical model means we have formulated a boundary value problem
that can be solved with different numerical methods. We will briefly describe selected
numerical methods.
FIGURE 2.6 Different methods and approaches used for solving engineering
analysis problems.
DESIGN
ANALYSIS
REAL
MODELS
OBJECTS
PHYSICAL MATHEMATICAL
MODELS MODELS
ANALYTICAL NUMERICAL
SOLUTION SOLUTION
Different methods and approaches used for solving engineering design analysis
problems are schematically presented in Figure 2.6.
© SAE International.
FIGURE 2.8 FE model of the support bracket; the 2D solution domain and boundary
conditions have been discretized. Red dots indicate nodes where the equivalent nodal load is
applied; blue dots indicate nodes where restraints are applied. Originally, continuous loads
and restraints are now represented by loads and restraints applied to element nodes.
© SAE International.
of degrees of freedom (DOFs) is approximated after discretization by a discrete FE model
with a finite number of DOFs.
K d F (2.1)
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
where
[K] is the known stiffness matrix
[F] is the known vector of nodal loads
[d] is the unknown vector of nodal displacements
The stiffness matrix [K] is defined by the model geometry, material properties, and
displacement boundary conditions. Loads are expressed by the load vector [F]. The
unknowns are the nodal displacements [d], and nodal displacements (or temperatures in
thermal problems) are the primary unknowns in the FEM. Strains and stresses are secondary
unknowns, they are calculated once the nodal displacements have been found.
FEM equations (2.1) take the form of linear algebraic equations and can be solved with
various numerical techniques.
FIGURE 2.9 Steps in an FEA project: (1) Converting CAD geometry to FEA geometry, (2)
Definition of the mathematical model, (3) Meshing, (4) Solution.
Loads Restraints
MATHEMATICAL
MODEL
FEA FEA
geometry Material Type of model
properties analysis
© SAE International.
CAD
Results
geometry
edges. Simplified geometry, called FEA geometry, is the first building block of a
mathematical model.
•• Step 2—Creating a mathematical model: Load and restraints, material properties
and the type of analysis are defined. This way a mathematical model is now ready.
Steps 1 and 2 introduce the modeling errors.
•• Step 3—Discretization: The mathematical model is discretized into finite elements.
The discretization is done on geometry, mass, and boundary conditions. There is
nothing continuous left in the FEA model. FE equations (2.1) are created in this
step. Step 3 introduces the discretization errors.
•• Step 4—Solution: FE equations are solved, and results are produced. This introduces
a solution error, also called numerical error. Results are now used to make a design
decision. If results are misunderstood, then the error of interpretation of results is
introduced after solution is completed.
Modeling errors, discretization errors, and solution errors are unavoidable; the error
of interpretation of results is fully preventable. While all errors affect results, only discreti-
zation errors are specific to the FEA. Therefore, only discretization errors can be controlled
using FEA techniques.
2.4. V
erification and Validation of FEA
Results
Verification and validation as they apply to design analysis with FEA are defined as follows:
•• Verification checks if the mathematical model has been correctly discretized and
solved.
•• Validation determines if the solution correctly represents the reality from the
perspective of the intended use of the results. It checks if the results correctly describe
the real-life behavior of the analyzed system.
Verification and validation in the FEA process are pictured in Figure 2.10.
A model with meshing errors would not pass a verification test. For example, using
too large elements, the solution would not be correct. Verification fails if discretization and/
or solution errors invalidate results. Convergence analysis will reveal problems causing
verification failure. A model with incorrect load definitions would pass the verification test
because verification only concerns itself with the correctness of the solution of the math-
ematical model, not if the mathematical model itself is correct.
Establishing the correctness of a mathematical model along with the correctness of its
solution is the process of validation. Validation will fail if conceptual errors are made in
the definition of the mathematical model. The conceptual errors are much more dangerous
than the errors of discretization. They may escape the modeler’s attention, especially because
there is no well-defined process to reveal conceptual errors. The only protection from
validation failure is the correct definition of the analyzed problem.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 2.10 Verification and validation of FEA results. Verification checks the numerical
solution of a mathematical model. Validation checks how well the results apply to reality. The
flowchart also shows errors introduced at each step of the FEA project.
REALITY
Modeling error
MATHEMATICAL MODEL
Discretization error
Validation
Solution error
© SAE International.
RESULTS
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
3
Fundamental
Concepts of FEA
3.1. F
ormulation of a Finite Element
3.1.1. Closer Look at Finite Element
Later in this chapter we find displacements in a thin flat plate that can be simplified to a
2D problem using triangular plane stress elements. The displacement field inside each
element and along the edges is described by certain polynomial functions called displace-
ment interpolation functions. The order of the displacement interpolation function used
in the element defines the order of the element. If the element uses the first order (linear)
displacement interpolation functions, it is called the first order element. If the element uses
the second order displacement interpolation function, it is called the second order
element, etc.
Arguments of displacement interpolation functions are nodal displacements. Once
nodal displacements are found, displacements anywhere in the element can be calculated
using the nodal displacements. Under a load the element deforms and assumes a new shape,
and nodes move from their original locations to the new ones. In the case of a 2D plane
stress element, we need to know the x and y displacement components of all nodes to
describe element transformation from the old to the new shape. This is because, in a 2D
element, nodal displacements are fully defined by only two in-plane displacement compo-
nents. The ability to perform a given displacement is called a Degree of Freedom (DOF).
FIGURE 3.1 A three nodded, first order 2D element has two degrees of freedom per
node: x and y translations. The total number of degrees of freedom is 6 meaning that 6
independent variables fully describe the element transformation from the undeformed to the
deformed shape. A 6 nodded second order 2D element also has two degrees of freedom per
node, but the total number of degrees of freedom for the element is 12. Upon deformation,
the edges of the second order element may assume a curved shape. Edges of the second
order element may be curved even before deformation if the element is mapped to a
curvilinear geometry. Rotational DOFs are not required to describe the deformation of
these elements.
First order
© SAE International.
Second order
The first order three nod element in Figure 3.1 (top) has 2 DOFs per node; therefore, the total
number of DOFs in the element is 6. Edges of the first order elements are straight before
and after the deformation. In the second order elements, the second order polynomials are
used for the displacement interpolation functions, and element edges do not have to remain
straight; they may “bend” assuming a curvilinear shape. To describe this, we need to add
mid-side nodes as shown in Figure 3.1 (bottom). Mid-side nodes differentiate the first order
element from the second order element. This applies to 2D as well as to 3D elements as
shown in Figures 3.1 and 3.2.
Strains are calculated by differentiating displacements. Therefore, if displacements are
described by the first order polynomial functions (linear functions), then strains and stresses
are constant within the element. If displacements are described by the second order poly-
nomial functions, then strains and stresses are distributed linearly within the element
(Figure 3.3).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 3.2 A four-noded, first order 3D solid tetrahedral element (top) has three
degrees of freedom per node: x, y, z translations. In this element linear shape functions model
linear displacements along edges, faces, and in the volume of the element. The element has a
total of 12 DOFs. A 10-noded second order 3D solid tetrahedral element (bottom) models
second order displacements and requires mid-side nodes. The element has a total of 30
DOFs. Edges and faces of the second order element may be curved even before deformation
if the element is mapped to a curvilinear geometry. Rotational DOFs are not required to
describe element deformation.
First order
Second order
© SAE International.
FIGURE 3.3 The first order element models linear displacements and constant stress; the
second order element models the second order displacements and linear stress. This is shown
in the graphs where the ordinate schematically denotes a location in the element. Solid
tetrahedral elements are shown but the same applies to other elements.
d
First order
x x
d
© SAE International.
Second order
x x
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
3.1.2. R
equirements to be Satisfied by Displacement
Interpolation Function
The reason why polynomials are used as displacement interpolation functions is their
versatility. Programmers have considerable freedom in selecting the form of polynomials
to build interpolation functions. For example, incomplete polynomials or different order
polynomials along different edges and faces may be used. However, certain requirements
must be always met to assure that the displacement field inside the element as well as in the
entire mesh is continuous.
•• Internal compatibility: Displacement interpolation functions must be continuous
over the entire element.
•• Inter-element compatibility: Displacement along the common edge and common
face between two elements must be described by the same function to make sure that
displacements are continuous across the element boundaries and that no “cracks”
or “overlaps” form between elements.
•• Rigid body motion (RBM): If the element is displaced as a rigid body (without
deformation), the element must show zero strain. This is called a patch test.
•• Constant strain: The element must be able to model the constant strain cases.
FIGURE 3.4 A thin, flat hollow plate is restrained along the left vertical edge and loaded
with a tensile load in the horizontal direction applied to the right vertical edge. This is a 2D
plane stress problem.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 3.5 Model meshed with the first order triangular elements. The round hole is
represented only approximately because the circle is replaced by straight
element edges.
© SAE International.
FIGURE 3.6 Resultant displacement results in the plate. The linear displacement field in
each element is evidenced by straight lines demarcating colors. The maximum displacement
is 0.112 mm.
0.112
0.102
0.093
0.084
0.074
0.065
0.056
0.046
0.037
0.028
© SAE International.
0.019
0.009
0.000
FIGURE 3.7 Von Mises stress results in the plate; results show constant stress in each
element. The maximum von Mises stress is 216 MPa.
216.0
201.5
187.1
172.6
158.2
143.7
129.3
114.8
100.4
85.9
© SAE International.
71.5
57.0
42.6
3.2. C
hoices of Discretization
Theoretically, there is an infinite number of ways in which a mathematical model can
be turned into an FE model by meshing. The chosen mesh is a realization of a certain choice
of discretization. Three major factors define the choice of discretization:
•• Element size: The element size is defined by its characteristic dimension. What
matters is the relative size of the element in relation to the size of the discretized
features. For a 2D triangular element, the element size may be defined as the diameter
of the circle circumscribed on the element. Element size is commonly denoted as
h (Figure 3.8).
•• Element order: Element order is defined by the order of the displacement
interpolation functions describing the displacement field inside the element, along
element edges, and element faces.
•• Element mapping: When an element is mapped to assume the shape in the FE
mesh, the element shape distorts (Figure 3.9). Excessive distortion causes elements
degeneration.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 3.8 The size of a 2D triangular element is here the diameter of the circumscribed
circle. The size of a 3D tetrahedral element is here the diameter of the circumscribed sphere.
h h
© SAE International.
FIGURE 3.9 Mapping transforms the original element shape (left) into an element in the
mesh (right) is illustrated with a 2D triangular element and 3D tetrahedral element.
3.3. T
ypes of Finite Elements
Commercial FEA programs use many different types of elements. There are many ways to
organize finite elements; we will present only commonly used element groups.
3.3.1. E
lement Dimensionality
Classification according to how the element represents three dimensions distinguishes
among 3D elements (solid, shell, membrane, beam), 2D elements (plane stress, plane strain,
axisymmetric), and one-dimensional (1D) elements.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• 3D Solid element: Solid element fully represents all three dimensions. The
displacement field in a solid element is 3D and each displacement component is
approximated by polynomials of the same degree. Nodes of solid elements have
three degrees of freedom which are translations. Three translations are needed
to define the element transformation from an undeformed to a deformed shape.
Therefore, nodes of commonly used solid elements have 3 DOFs (Figure 3.10).
•• 3D Shell element: Shell element has one dimension collapsed. That collapsed
dimension is the thickness of the shell which is considered small relative to
the other dimensions. Assumptions must be made about stress distribution
along that missing dimension. Stresses normal to the shell cross-section (also
called in-plane stresses) are assumed to have linear distribution across the
thickness; therefore, the shell element can model bending. Transverse shear
stress is constant in thin shell element formulation and parabolic in thick shell
element formulation (Figure 3.11).
© SAE International.
FIGURE 3.11 The arc represented by a surface (left) may be meshed either with shell or
with membrane elements; shell element and membrane element have the same appearance.
Flat shell elements (right) look the same as 2D plate elements.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
The elements shown in these illustrations would be too large for analysis.
•• 3D Membrane Element: The membrane element looks the same as a shell element
but stresses normal to the membrane cross-section (also called in-plane stresses)
are constant and transverse shear stresses are not modeled. The membrane element
cannot model bending. Shell and membrane elements have 6 DOFs per node: three
translations and three rotations. The difference in how stresses normal to a cross-
section is modeled in shell and membrane elements is shown in Figure 3.12.
•• 3D Beam Element: The beam element has two dimensions collapsed. It is assumed
that the width and height of the cross-section are small in comparison to the length.
Linear variations of stress in two directions perpendicular to the normal cross-section
to the beam are modeled. The beam element can be seen as a curve with cross-section
properties: the area and the second moments of inertia. Beam elements have 6 DOFs
per node: three translations and three rotations, the same as shell elements. Figure 3.13
shows how beam elements may be depicted in different FEA programs.
•• 2D Elements: 2D elements are used when a structure response to load can be fully
described in two dimensions. 2D elements fall into three categories: plane stress,
plane strain, and axisymmetric elements. Plane stress elements are used for the
analysis of thin planar structures loaded in-plane, where out-of-plane stress is
assumed to be equal to zero. Plane strain elements are used for the analysis of
thick prismatic structures loaded in-plane, where out-of-plane strain is assumed
to be equal to zero. Axisymmetric elements are intended for the analysis of
axisymmetric structures under axisymmetric loads and restraints. In all these
cases, the structure displacements can be fully described using elements with 2 per
node. For the plane stress and plane strain elements these are two components of in-
plane translation. For axisymmetric elements, these are radial and axial translations.
FIGURE 3.12 A shell element models the linear distribution of in-plane stresses. The in-plane stresses may
be separated into the bending stress component (the same magnitude but in the opposite direction on the top and the
bottom) and the membrane stress component where stress is constant across the element thickness. A membrane
element (bottom) can only model the constant distribution of in-plane stress across the thickness. A shell element has
bending stiffness and membrane stiffness while a membrane element has only membrane stiffness. The illustration shows
element thickness even though the thickness is not represented as a physical dimension.
Thickness
Shell element
Thickness
Membrane element
© SAE International.
A 2D plane stress model is shown in Figure 3.14, a 2D plane strain model is shown in
Figure 3.15, and a 2D axisymmetric model is shown in Figure 3.16. All these figures
show large elements for the clarity of illustrations.
•• 1D Elements: 1D elements have no practical applications but because of their
simplicity are used extensively in introductory FEA courses. A 1D element has 1
DOF per node. 1D elements are called spring or bar elements An example of a 1D
element mesh is shown in Figure 3.17.
FIGURE 3.13 An appearance of a beam element may differ depending on the FEA
program and the settings used. The cross-section (here, an I beam) may be fully rendered, it
may be depicted only schematically (here, as a circle) or not shown at all.
© SAE International.
Fully rendered cross-section Schematic cross-section No cross-section
© SAE International.
Figures 3.14, 3.15, and 3.16 show large elements for clarity of illustrations. These elements
would be too large for analysis.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 3.17 Mesh of three 1D elements shown as springs; nodes 2 and 3 are “spread out”
vertically for clarity of this illustration. Nodes 1 and 3 are fixed and node 2 can move in the
direction of the applied force.
2 3
1 2
© SAE International.
2 3
FIGURE 3.18 Both h elements and p elements have shapes limited to simple primitives:
tetrahedron, hexahedron, triangular prism wedge. p elements are allowed a much higher
distortion from the ideal shape when mapped to model geometry.
© SAE International.
h elements p elements
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 3.19 The same solid model meshed with solid tetrahedral h elements and solid
tetrahedral p elements. Element shrinkage has been applied to the p element mesh.
© SAE International.
FIGURE 3.20 The same surface model meshed with triangular shell h elements and
triangular shell p elements.
© SAE International.
h element mesh p element mesh
h elements p elements
Element shape Tetrahedron, prism, hexahedron Tetrahedron, prism, hexahedron
Mapping Low deviation from the original Higher deviation from the
shape is allowed original shape is allowed, but this
may introduce errors on highly
curved edges and surfaces
Displacement/temperature field Described by lower-order Described by higher-order
polynomials; usually second polynomials
order
Convergence process Element order does not change Element size does not change
during the convergence process during the convergence process
FIGURE 3.21 Commonly used h elements are categorized according to the dimensionality
of the discretized geometry and the element order. A beam element is shown as an I beam,
but a beam element cross-section may be of any shape.
3D elements 2D elements
Element order Plane stress
Solid Shell Beam Plane strain
Axisymmetric
First order
© SAE International.
Second order
FIGURE 3.22 Four spring elements connect the blue base to the red cube. The point mass
element is located above the top face of the cube; it is attached to the top face with rigid
link elements. The blue base and the red cube are meshed with solid tetrahedral elements.
Mass element
Links element
Springs
© SAE International.
element
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
4
Controlling
Discretization Errors
M
odeling errors, discretization errors, solution errors, and errors of interpretation
results (Figure 2.10) all affect FEM results. However, only discretization errors are
specific to the FEM, and only the discretization errors can be controlled using
FEM tools. For this reason, we discuss them first, discussion of other errors will follow in
later chapters.
The objective of control of the discretization error is to find how the data of interest
(displacement, stress, temperature, etc.) are dependent on the choice of discretization (mesh).
The objective is not to minimize the discretization error but to obtain a solution where the
data of interest do not significantly depend on the choice of discretization. An analysis is
not complete unless we have a reliable estimate of the discretization error. The analysis of
a discretization error is done in a convergence process. An example of different choices of
discretization is shown in Figure 4.1 where the same model is meshed with four meshes,
each one with a different element size.
FIGURE 4.1 The same model meshed with triangular shell elements of different sizes.
This illustrates the global mesh refinement.
Mesh 1
Element size 20 mm
Mesh 2
Element size 10 mm
Mesh 3
Element size 5 mm
© SAE International.
Mesh 4
Element size 2.5 mm
4.1. P
resenting Stress Results
Before discussing the convergence process, we need to look at different ways of presenting
stress results. As we already know, nodal displacements are computed first, and then strains
and stresses are calculated based on the displacement results. Stresses are first calculated
inside the element at certain locations, called Gauss points. Gauss points are locations used
in the numerical generation of the element stiffness matrix. The location of Gauss points
in a triangular second order element is schematically shown in Figure 4.2 . Next, stress
results are extrapolated to nodes. Stress results from all the elements sharing the given node
are averaged, and one value, called nodal stress, is reported for each node. A stress plot is
prepared using nodal stresses. The nodal stress is also called the averaged stress.
An alternative procedure to prepare a stress plot still uses stresses at Gauss points but
averages them among themselves within the element. This method calculates one stress
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.2 Location of Gauss points (black dots) in the second order triangular element.
© SAE International.
value for the entire element. The stress value is called the element stress or the non-averaged
stress because stresses are not averaged between neighboring elements.
Nodal stresses are used more often because they produce continuous stress plots.
However, an examination of element stresses provides important feedback on the quality
of the results. If element stresses in two adjacent elements differ too much, it indicates that
the element size at this location is too large to properly model the stresses.
Nodal and element stresses in Mesh 1 in Figure 4.1 are shown in Figure 4.3. The model is
restrained along the left vertical edge and subjected to a uniform tension applied to the
right vertical edge. A mesh with large elements is used for clarity in this illustration.
Nodal stresses are higher than element stresses because of the extrapolation of the nodal
stresses to the surface.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.3 A mesh of the second order triangular elements represents a hollow plate in
tension. The nodal stress plot shows continuous stress distribution. The element stress plot
shows piecewise constant stresses.
308.8
283.9
259.0
234.1
209.2
184.4
159.5
134.6
109.7
84.8 Nodal stresses
59.9
216.5
200.5
184.5
168.4
152.4
136.4
© SAE International.
120.3
104.3
88.3
Element stresses
72.3
56.2
4.2. T
ypes of Convergence Analysis
Discretization errors are found in the process of making systematic changes to the choice
of discretization. The process is called convergence process; there are several variations of
the convergence process.
FIGURE 4.4 Displacement results produced by four different meshes representing the
same problem; this illustrates h convergence process by global mesh refinement.
0.11723
0.10551
0.09378
0.08206
0.07034
0.05861
0.04689 Mesh 1
0.03517
Element size 20 mm
0.02345
Max. displ. 0.11723 mm
0.01172
0.00000
0.11792
0.10613
0.09434
0.08255
0.07075
0.05896
0.04717 Mesh 2
0.03538 Element size 10 mm
0.02358 Max. displ. 0.11792 mm
0.01179
0.00000
0.11821
0.10639
0.09457
0.08275
0.07093
0.05911
0.04728 Mesh 3
0.03546 Element size 5 mm
0.02364 Max. displ. 0.11821 mm
0.01182
0.00000
0.11824
0.10642
0.09460
0.08277
0.07095
0.05912
Mesh 4
© SAE International.
0.04730
0.03547 Element size 5 mm
0.02365 Max. displ. 0.11824 mm
0.01182
0.00000
other related measure. The graph in Figure 4.6 shows the maximum displacement in the
direction of load as a function of the inverse of the element size. The graph in Figure 4.7
shows the maximum von Mises stress as a function of the inverse of the element size.
In a discretized model, displacements must follow a displacement pattern modeled by
the elements used for meshing. This adds some (very low) stiffness, which we call artificial
stiffness. Every mesh refinement reduces the artificial stiffness. This explains why displace-
ment increases with every iteration as shown in Figure 4.6. Strain is a derivative of displace-
ment, and stress is calculated once strain has been found; therefore, stress also increases
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.5 Von Mises stress results produced by four different meshes representing the same problem; this
illustrates h convergence process by global mesh refinement.
308.8
283.9
259.0
234.1
209.2
184.4
159.5 Mesh 1
134.6
109.7
Element size 20 mm
84.8 Max. von Mises stress 308.8 MPa
59.9
297.1
270.4
243.7
217.0
190.3
163.6
136.9 Mesh 2
110.2 Element size 10 mm
83.5 Max. von Mises stress 297.1 MPa
56.8
30.1
337.9
306.9
275.9
245.0
214.0
183.0
152.1 Mesh 3
121.1 Element size 5 mm
90.1 Max. von Mises stress 337.9 MPa
59.2
28.2
364.3
329.6
294.9
260.2
225.5
190.8
© SAE International.
156.2
Mesh 4
121.5
86.8 Element size 5 mm
52.1 Max. von Mises stress 364.3 MPa
17.4
TABLE 4.1 Results of four steps of h convergence process by global mesh refinement.
0.1184
0.1182
0.1180
Displacement [mm]
0.1178
0.1176
0.1174
© SAE International.
0.1172
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
1/h [mm–1]
FIGURE 4.7 Global mesh refinement; convergence of maximum von Mises stress. The
maximum stress is a function of 1/h, where h is the element size.
400
360
Stress [MPa]
320
Nodal stress
280 Element stress
240
© SAE International.
200
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
1/h [mm–1]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
with mesh refinement. Note the “dip” in the nodal stresses graph shown in Figure 4.7 caused
by error of stress averaging, which is a part of nodal stress calculations.
Having completed four steps of the h convergence process by global mesh refinement,
we notice that displacements converge faster than stresses. To explain this, we need to
differentiate between global and local results. The maximum displacement is a global result;
the stiffness of the entire model contributes to the maximum displacement results. Global
results converge faster with mesh refinement. The maximum stress (here, von Mises stress)
is a local result and is modeled only by a few elements; for faster convergence of the maximum
stress, we would need a more aggressive refinement around stress concentration.
FIGURE 4.8 The same model meshed with shell elements of the same global size but
different sizes along the controlled entity: the edge of the hole. Mesh bias expressed in
millimeters characterizes the element size on the entity where the bias has been applied.
Mesh 1
Global element size 20 mm
No mesh bias
Mesh 2
Global element size 20 mm
Mesh bias 10 mm
Mesh 3
Global element size 20 mm
Mesh bias 5 mm
Mesh 4
Globak element size 20 mm
© SAE International.
Local mesh refinement adds fewer DOFs to the model as compared to the global
mesh refinement, and models solve faster. h convergence process by local mesh refine-
ment is easier to execute than a global mesh refinement, but the h convergence process
by the local mesh refinement requires prerequisite knowledge of stress pattern in the
analyzed model; we need to know where to apply the mesh bias. There is a risk of not
finding stress concentrations if mesh refinement is applied in an incorrect
location.
Von Mises stress results produced by four meshes used in the h convergence process
by the local mesh refinement are shown in Figure 4.9; a convergence graph of the maximum
von Mises stress (nodal) is shown in Figure 4.10.
FIGURE 4.9 Von Mises stress results produced by four different meshes representing the
same problem. This illustrates h convergence process by local mesh refinement.
308.8
283.9
259.0
234.1
209.2
184.4 Mesh 1
159.5
Element size 20 mm
134.6
109.7
No mesh bias
84.8 Max. von Mises stress 308.8 MPa
59.9
307.8
282.2
256.6
231.0
205.5
179.9 Mesh 2
154.3 Global element size 20 mm
128.7 Mesh bias 10 mm
103.1 Max. von Mises stress 307.8 MPa
77.6
52.0
367.9
334.8
301.6
268.5
235.4
202.2 Mesh 3
169.1
136.0
Global element size 20 mm
102.9 Mesh bias 5 mm
69.7 Max. von Mises stress 367.9 MPa
36.6
375.4
340.4
305.4
270.4
235.5
200.5 Mesh 4
© SAE International.
FIGURE 4.10 Convergence of maximum von Mises stress during h convergence by local
mesh refinement. The maximum stress is a function of 1/h, where h is the element size. “Dip”
in stress in the second iteration is caused by an error of stress averaging.
380
360
Stress [MPa]
340
320
© SAE International.
300
0 0.1 0.2 0.3 0.4
1/h [mm–1]
4.2.3. A
daptive h Convergence
h convergence analysis by global mesh refinement is time consuming because the number
of DOFs increases while the element size is reduced, and that increases solution time. h
convergence analysis by local mesh refinement is faster because local mesh refinement has
less effect on the increase of the number of DOFs in the model, but it requires the user’s
decision on where to refine the mesh. This disadvantage of the local h convergence process
is avoided when an adaptive h convergence solution is used.
In the adaptive h convergence process, the mesh is refined automatically during an
iterative solution. The initial mesh is defined by the user and a solution is obtained; let us
call it Iteration #1. Based on the errors found in Iteration #1, the mesh is automatically
refined in the locations characterized by high errors, and a new solution is obtained; this
is Iteration #2. Iterations continue until the user-specified accuracy has been satisfied or
the number of allowed iterations has been reached. The user does not have direct control
over how the final mesh will look like.
The name “adaptive” derives from the fact that mesh refinement is adapted to locations
where errors are found. The measure of error is related to the difference between nodal
stress and element stresses. Consecutive mesh refinements during the h adaptive conver-
gence process are illustrated in Figure 4.11.
A hollow plate is subjected to a uniform tension causing stress concentrations on the
cylindrical surface of the hole, and the mesh is automatically refined there. There are
important differences between this problem and that shown in Figures 4.5 and 4.9. First,
the model is subjected to tension on both end faces and stresses are perfectly symmetric.
Second, because of its larger thickness, the problem requires 3D solid elements rather than
3D shell elements or 2D plate elements. Lower stresses are reported because the thickness
increased while the load remains the same. The summary of results of the h adaptive
convergence process is shown in Figure 4.12.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Iteration # 1
Initial mesh defined by user
von Mises stress 106.1 MPa
Iteration # 3
Mesh after the second refinement
von Mises stress 125.6 MPa
© SAE International.
Iteration # 5
Mesh after the fourth refinement
von Mises stress 126.5 MPa
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.12 Convergence of the maximum von Mises stress in the h adaptive process;
the maximum von Mises stress is shown as a function of the iteration number.
130
125
120
Stress [MPa]
115
110
© SAE International.
105
1 2 3 4 5
Iteration number
FIGURE 4.13 The p element mesh (top), the final edge element order after seven
iterations (middle), and von Mises stress results after seven iterations in a p adaptive
solution process.
p element mesh
with six elements
p=9
p=8
p=7
p=6
p=4 Edge order after seven iterations
p=3
the highest order p = 7
p=2
p=1
© SAE International.
When a dedicated p element mesher is used, the mesh looks very different from the h
element mesh as shown in Figure 4.13.
The mesh in Figure 4.13 is a result of the discretization of a surface; it consists of only
six shell elements: two triangular elements and four quadrilateral elements. The model is
subjected to a uniform tensile load applied to both vertical edges. The model is solved with
a program that can use element order anywhere between p = 1 and p = 9. In this case the
numbers between 1 and 9 refer to the order of the stress interpolation function, not the
displacement interpolation function. The final element order is found after several iterations
during which the edge order is increased selectively based on the errors found in the previous
iteration. This happens in direct analogy to h adaptive convergence process. The results of
the p convergence process are summarized in Figure 4.14.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.14 Convergence of the maximum von Mises stress in the p adaptive process;
the maximum von Mises stress is shown as a function of the iteration number. The accuracy
requirements have been satisfied at p = 7 and further element order upgrade was
not required.
450
400
350
Stress [MPa]
300
250
200
© SAE International.
150
1 2 3 4 5 6 7
Iteration number
FIGURE 4.15 A schematic convergence curve is used to define convergence error and
solution error. The convergence curve tends to the asymptote shown as a dashed line.
Solution error
Convergence error
Data of interest
© SAE International.
1 2 3
Iteration number
result n result n 1
Convergence error (4.1)
result n
Figure 4.15 shows the convergence error of the last performed iteration which is the
third iteration. Using Equation 4.1, the convergence error can be calculated for all steps of
the convergence process except Step 1. The convergence error for Step 1 is unknown because
no prior results exist. We can rephrase this important observation by saying that a single
run produces results with unknown discretization error.
Since the asymptotic solution is not known, we can only estimate the solution error,
but we cannot calculate it exactly.
4.4. P
roblems with Convergence
The FE model solution converges to the exact solution of the mathematical model on which
the model is based. Both h and p convergence processes demonstrated so far demonstrated
that the data of interest converge to finite values. But this is not always the case.
FIGURE 4.16 A thin L bracket is represented by a flat surface and discretized with a
coarse mesh of 2D plane stress elements.
© SAE International.
50 N
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.17 The maximum von Mises stress as a function of 1/h, where h is the
element size.
80.0
70.0
60.0
Stress [MPa]
50.0
40.0
30.0
20.0
© SAE International.
10.0
0 1 2 3 4
1/h [mm ] –1
FIGURE 4.18 Von Mises stress plot produced by mesh with an element size of 0.25 mm.
73.7
67.5
61.4
55.3
49.1
43.0
36.8
30.7
24.6
18.4
12.3
© SAE International.
6.1
0.0
To find the accuracy of this result, we need to examine if stress converges to a finite
value. However, the graph in Figure 4.17 shows no sign of converging to a finite value; instead,
each iteration brings a higher stress result; the maximum stress diverges to infinity. This
happens because the mathematical model does not offer a solution for maximum stress.
The sharp reentrant edge (or corner in 2D model) creates a singularity where the stress is
infinite. Since the objective of the analysis is to find the maximum stress, we conclude that
the FE model is based on the wrong mathematical model. The mathematical model with
stress singularity that coincides with the location of the maximum stress cannot be used
as a basis for the FE model if the objective is to find the maximum stress.
Why did the FE model produce high, but finite, stress instead of infinite stress as
predicted by the mathematical model? This is because the modeling error (using the wrong
mathematical model) is masked by the discretization error. As a result of that, finite stress
is produced where the mathematical model predicts infinite stress.
The stress singularity in the model in Figure 4.16 is caused by a 270° sharp reentrant
corner, but any re-entrant corner in a 2D model or a reentrant edge in a 3D model produces
stress singularity.
The strength of a singularity increases with the angle of the sharp reentrant edge.
Singularities manifest themselves as “hot spots.” To visualize singular stresses caused by the
225° edge, a more aggressive mesh refinement is required as compared to the stress singularity
caused by the 270° edge. Comparing the three models (in Figure 4.19), the singularity in the
model with a 225° edge is the weakest and the singularity in the model with a 340° edge is
the strongest. The strength of singularity indicates how fast stresses diverge to infinity.
By showing the divergence of the data of interest (here the maximum von Mises stress),
the convergence process revealed the modeling error. We can remedy this situation by using
a different mathematical model; one that does not have a stress singularity. The most obvious
way is to model a fillet, which is always present in a real part, even if the edge is very sharp.
The result may show very high stress, but that result will be bounded; it will converge to a
finite value. Another way of eliminating stress singularity is to use a material model capable
of modeling plasticity. To illustrate this, we will use an elastic-perfectly plastic model
characterized by a strain-stress curve in Figure 4.20.
The elastic-perfectly plastic material model limits the maximum von Mises stress to
the plasticity limit; this eliminates stress singularity even though strain in the sharp reen-
trant edge continues diverging with mesh refinement. Von Mises stress results using elastic-
perfectly plastic material and element size of 0.5 mm are shown in Figure 4.21. The solution
shown in Figure 4.21 has been reached in 14 steps during which the load was gradually
increased to the maximum of 50 N. The maximum von Mises stress for each step is shown
in Figure 4.22.
Stresses in the location close to singularity are entirely dependent on the choice of
discretization and, therefore, are meaningless. By manipulating element size, element order,
or both, we can produce any result we want. Using geometry with sharp reentrant corners
in 2D models or sharp reentrant edges in 3D models, while the objective is finding stress
in that location, is a severe yet common modeling error. An erroneous model is shown in
Figure 4.23. This model still can be used for the analysis of displacement because a sharp
reentrant edge does not cause displacement singularities. It can also be used for stress
analysis in a location distant from sharp reentrant edges. It is important to mention that
fillets cannot be ignored, even if stresses along the edge are not of interest if the removal of
fillets changes the model stiffness significantly. If stresses along the edge are of interest,
then fillets, no matter how small, must be modeled (Figure 4.24). Stresses in the fillet may
be very high but, during a convergence process, they will converge to a finite limit.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.19 All sharp reentrant edges shown here produce stress singularity. The closer
the edge to 360° the stronger the singularity.
340°
270°
© SAE International.
225°
FIGURE 4.20 Elastic-perfectly plastic material, von Mises type; this illustration shows the
strain-stress curve of 1060 Aluminum Alloy. The material behaves linearly until the stress
reaches 27.6 MPa, and then the modulus of elasticity becomes zero. Von Mises stress is used
as a measure to control the switch of the modulus of elasticity to zero.
30
20
Stress [MPa]
10
© SAE International.
0
0.0000 0.0005 0.0010 0.0015
Strain [1]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.21 Von Mises stress solution using elastic-perfectly plastic material and
element size of 0.5 mm. This illustration shows the results of Step 14. This is the last step in
the iterative mesh refinement process.
27.6
25.3
23.0
20.7
18.4
16.1
13.8
11.5
9.2
6.9
4.6
2.3
© SAE International.
0.0
FIGURE 4.22 Changes in the maximum von Mises stress during 14 steps of solution using
elastic-perfectly plastic material.
30
20
Stress [MPa]
10
© SAE International.
0
0% 20% 40% 60% 80% 100%
Load
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.23 This model cannot be used for analysis of the maximum stress because the
maximum stresses (most likely) will coincide with the stress singularity caused by the sharp
reentrant edge.
© SAE International.
FIGURE 4.24 This model has a fillet (red) added in place of the sharp reentrant edge. It
can be used for analysis of the maximum stress. Notice a mesh control defined on the fillet to
produce correctly sized and correctly shaped elements.
© SAE International.
FIGURE 4.25 A wedge is loaded at the top and supported along the bottom edge. Side
faces are restrained by rollers.
Rollers
© SAE International.
Fixed support
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.26 The maximum stress diverges because of stress singularity caused by the
edge support. Element size along the supported edge is shown. Refinement is done using
mesh bias. Undeformed plots are shown.
61.9
55.8
49.6
43.5
37.4
31.3
25.1
10 mm 19.0
12.9
6.8
0.6
310.4
279.4
248.4
217.5
186.5
155.5
124.5
2 mm 93.6
62.6
31.6
0.6
545.2
490.7
436.2
381.7
327.3
272.8
218.3
1 mm 163.8
© SAE International.
109.3
54.8
0.4
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.27 The maximum von Mises stress in the three models is shown in Figure 4.26.
Stress is plotted as a function of 1/h, where h is the element size along the supported edge.
600
500
400
Stress [MPa]
300
200
100
© SAE International.
0
0 0.2 0.4 0.6 0.8 1
1/h [mm ]
–1
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
0.0024
0.0021
0.0019
0.0017
0.0014
0.0012
10 mm 0.0010
0.0007
0.0005
0.0002
0.0000
0.0027
0.0024
0.0021
0.0019
0.0016
0.0013
0.0011
2 mm 0.0008
0.0005
0.0003
0.0000
0.0030
0.0027
0.0024
0.0021
0.0018
0.0015
0.0012
1 mm 0.0009
© SAE International.
0.0006
0.0003
0.0000
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.29 The displacements in the three models are shown in Figure 4.28.
Displacement is plotted as a function of 1/h, where h is the element size along the supported
edge. Displacement results diverge.
0.0031
Displacement [mm]
0.0028
0.0026
© SAE International.
0.0023
0 0.2 0.4 0.6 0.8 1
1/h [mm–1]
Type 1 Type 2
Stress Infinite Infinite
Strain energy Finite Infinite
Displacement Finite Infinite
Examples Sharp reentrant edge Point or edge support
Objective
This exercise illustrates a convergence process and demonstrates how the data of interest
(the maximum displacement in the direction of load and the maximum von Mises stress)
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 4.30 HOLLOW PLATE model. Tensile load (magenta arrows) is applied to left
vertical face. Restraints defined on the right vertical face (red arrows) in the direction normal
to face generate reaction. Restraints defined on two corners (blue and green arrows)
eliminate Rigid Body Motions without interfering with plate deformation.
© SAE International.
change while the element size is reduced during several steps of mesh refinement (h conver-
gence) or during several steps of element order upgrade (p convergence).
Procedure
•• Delete all restraints and define a fixed restraint applied on the face opposite to the
loaded face. Notice that stress singularities are now present in the corners of the
supported face. These singularities are difficult to spot when global mesh refinement
is used because stress concentration becomes visible only when a very fine mesh
is used.
4.5.2. L BRACKET01
Description
The L bracket is represented by a surface and meshed with 2D plane stress elements. This
model is not intended for analysis of buckling even though this may be the primary mode
of failure. Fixed restraints are applied to the top edge (blue line). The bending load (red
arrow) is evenly distributed over the right vertical edge (Figure 4.31).
Objective
This exercise illustrates the lack of convergence of stress due to the stress singularity in the
sharp reentrant corner. The stress singularity makes the model useless if the objective is to
find the maximum stress because the location of the maximum stress is coincident with
the stress singularity.
Procedure
FIGURE 4.31 L BRACKET01 model. Fixed restraint is applied to the top edge (blue line).
Bending load is evenly distributed over the right vertical edge. This model is not intended for
analysis of buckling even though this may be the primary mode of failure.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
4.5.3. W
EDGE
Description
The wedge is held between rollers acting on two side faces to prevent tilting sideways. Fixed
restraints are applied to the sharp edge. A pressure of 1.25 MPa is applied to the top face
(Figure 4.32).
Objective
Demonstrate divergence of displacements and stresses in the model with displacement and
stress singularities caused by restraints applied to the entity with zero area (edge).
Procedure
Support
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• Apply restraints eliminating displacements in the direction normal to the two side
faces to prevent tilting.
•• Apply a pressure of 1.25 MPa to the top face.
•• Mesh with solid elements of global size 10 mm and solve. Then apply mesh bias of
2 mm and 1 mm along the sharp edge (three runs).
•• Plot the displacement and stress result as a function of the inverse of element size
along the sharp edge. Each curve will consist of three points corresponding to three
steps in the mesh refinement process.
•• Repeat the exercise using h adaptive solution; only one mesh will be required.
•• Repeat the exercise using p adaptive solution; only one mesh will be required.
•• Observe the divergence of displacements and stresses as demonstrated by the h
convergence process with mesh bias, h adaptive convergence process, and p adaptive
convergence process.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
5
Finite Element Mesh
W
e have already introduced several types of finite elements when discussing the
convergence process. Now we focus on how the elements are assembled into
a mesh. We will review different methods of creating FE mesh as well as the
issues of mesh compatibility, mesh quality, and mesh adequacy.
FIGURE 5.1 A sprocket model meshed with shell elements using a mapped
meshing technique.
© SAE International.
5.1.2. Semi-automatic Meshing
Semi-automatic meshing techniques can be explained with concepts borrowed from CAD.
Just like a solid feature can be created in CAD by extruding or revolving a flat sketch, a
mesh can be created by extruding or revolving a planar surface. That surface needs to
be meshed first with 2D elements and extrusion (Figure 5.2) or revolution (Figure 5.3) is
done in several steps depending on how many layers of elements we need to create. Semi-
automatic meshing is a powerful technique but, by its underlying principle, is limited to a
narrow class of shapes. Just like manual meshing, it is now of historical importance only.
FIGURE 5.2 A mesh created with p elements by extruding a meshed flat surface in ten
steps to create ten layers of elements. Mesh is shown with element shrinkage applied.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.3 A mesh created with p elements by revolving the meshed radial cross-
section in twelve steps along a 360° arc. Mesh is shown with element shrinkage applied.
© SAE International.
FIGURE 5.4 Model meshed with large and small elements. This is what we used in the h
convergence process by global mesh refinement. Second order elements are used, they map
precisely to cylindrical faces of all holes in both coarse and fine meshes.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.5 Mesh control (mesh bias) applied to selected edges produces smaller
elements along these edges.
© SAE International.
FIGURE 5.6 Mesh controls (mesh bias) applied to selected faces (two cylindrical holes)
produce smaller elements on these faces.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.7 Mesh control applied to a vertex produces smaller elements around the
selected corner.
© SAE International.
Hexahedral (brick) solid elements are difficult for meshers; the same applies, to a
lesser degree, to quadrilateral shell elements. For this reason, meshers found in commercial
FEA software are often limited to creating tetrahedral solid elements and triangular
shell elements.
FIGURE 5.8 Two elements joined along an edge are shown as if they shared one edge,
but these are two coincident edges. For two elements to be compatible, the displacement
interpolation functions describing displacements of the edge belonging to element 1 and the
displacement interpolation function describing the displacement of the edge belonging to
element 2 must be the same. This requirement is automatically satisfied if the elements are of
the same type and order as the above two first order triangular shell elements.
Element 2
First order
Element 1
First order
© SAE International.
5.2.2. Incompatible Elements
If connecting elements are not of the same type, compatibility conditions are not auto-
matically satisfied, and compatibility must be enforced by imposed compatibility
called links.
The elements shown in Figure 5.9 are not compatible. The displacement interpolation
function along the edge of element 1 is of the second order while the displacement inter-
polation function along the edge of element 2 is of the first order. There is nothing “telling”
the mid-side node on the edge belonging to element 1 to follow the displacements of edge
2. As a result, a gap or an overlap will form between the two elements when they deform
under load. Similarly, there is an incompatibility in the mesh in Figure 5.10 even though
the mesh consists of only one element type. A closer look at the transition between small
and large elements reveals that the refined part of the mesh is connected to the coarse part
of the mesh only by two nodes. This results in a gap forming under a tensile load.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.9 Since the mid-side node on edge of element 1 is not attached to the edge of
element 2, a gap or overlap will form when load is applied. Mesh incompatibility occurs due to
the different order of the connected elements.
Element 2
First order
Element 1
Second order
© SAE International.
FIGURE 5.10 The left and right sides of the mesh share only two nodes (red). This is not
evident if the mesh is shown in the undeformed shape (top), but clearly shows under the
tensile load (bottom). This illustration was created in an old FEA program. Using a modern
mesher, it would take some effort to create intentionally this incorrect mesh.
Before deformation
© SAE International.
After deformation
FIGURE 5.11 The spokes, meshed with shell elements, are connected to the hub and to
the rim meshed with solid elements. The mesh is incompatible because nodes of shell
elements have six DOFs while nodes of solid elements have three DOFs.
© SAE International.
FIGURE 5.12 The flat part of the model is meshed with solid hexahedral p elements, and
the curved part is meshed with solid tetrahedral p elements. The edges of tetrahedral
elements running diagonally across the faces of solid elements are not connected to the faces
of solid elements. To eliminate incompatibility, links are required between the corresponding
faces of tetrahedral and hexahedral elements. The location of links is indicated by a
dashed line.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
The rotational DOFs along the edges of the shell elements that connect to the edges of
solid elements are not restrained, and unintentional hinges are created. Links must
be defined to constrain translations and rotations of the edges of the shell element to trans-
lations of the corresponding edges of solid elements. Edges of solid elements become
“masters” and edges of shell elements become “slaves.” Links suppress rotations of edges of
shell elements. The rim and the hub are meshed with large elements for clarity in this
illustration. An analysis would require much smaller elements.
Links assure a continuous displacement field, but should not be used in the location
where accurate stress results are required.
5.3. C
ommon Meshing Problems
New FEA users often expect that meshing should be a fully automated process requiring
little, in any, input from the user. With experience comes a realization that meshing is not
a hands-off task. Instead, it is often a difficult and time-consuming process that requires
user’s input. Leaving the mesher on the default settings may lead to severe meshing errors,
therefore most FEA programs provide users with meshing controls.
FIGURE 5.13 Aspect ratio. A regular tetrahedron and two distorted tetrahedrons with
excessive aspect ratios.
© SAE International.
FIGURE 5.14 Curvature distortion. Elements with high curvature distortion are indicated
by the arrows. The curvature distortion applies only to the second order and higher because
the first order elements cannot have curved edges.
© SAE International.
Turn angle 45°
FIGURE 5.15 Tangent edges. Elements with tangent edges are indicated by the arrows.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.16 Warpage. Warped surface shown with zebra stripes and warped shell
element mesh.
© SAE International.
While element distortion is easy to detect, it may be difficult to control. Recall that
meshing is basically a process of filling up a given volume or surface with certain geometric
shapes; most often with triangles or tetrahedrons. The mesher often finds it impossible to
accomplish its task while keeping the elements within the allowable range of distortion.
FIGURE 5.17 Beam subjected to pure bending is meshed with one layer of first order
element across the thickness. Arrows indicate the moment applied to two end faces; the
visible end face is shown in green color.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.18 Pure bending produces linear stress distribution of in-plane stresses across
the beam thickness.
Correct
Bending Linear stress distribution across
Second order element
Moment the element thickness is modeled
with one layer of second order elements
© SAE International.
Wrong
Bending Constant stress distribution across
First order element
Moment the element thickness is modeled
with one layer of first order elements
FIGURE 5.19 Beam in pure bending is correctly modeled with one layer of second order
elements (top). It can be also modeled with shell elements (middle) because shell elements
model linear distribution of in-plane (bending) stresses. A more complex state of stress
requires two, or more, layers of second order elements (bottom).
© SAE International.
of second order elements
FIGURE 5.20 Large first order elements are unable to map to geometry. Fillets are
changed into chamfers and the round hole into a polygon. The inability of the first order
elements to map to curvilinear geometry combined with modeling constant stress makes the
first order elements practically useless.
© SAE International.
FIGURE 5.21 A second order solid element mesh is required to model this bracket; the
mesh is locally refined in the fillet area.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 5.22 The same model meshed with over twelve thousand second order solid h
elements can be meshed with only 60 p elements. Fewer p elements are required because p
elements are designed to reach higher order during solution; no local mesh refinement is
required in p element mesh. p element mesh is shown with element shrinkage applied.
© SAE International.
h element mesh p element mesh
FIGURE 5.23 Automated converting of thin solid geometry into surface geometry may
produce geometry with disconnected surfaces.
© SAE International.
Objective
Demonstrate the effect of global mesh refinement, h adaptive solution, and p adaptive
solution on stress results. Demonstrate the effect of defeaturing on the complexity of mesh.
Procedure
•• Apply the material properties (steel).
•• Apply the restraint to the back face.
•• Apply a pressure to the blue face.
•• Mesh and solve using second order tetrahedral solid elements mesh with coarse,
medium, and dense meshes.
•• Plot the maximum von Mises stress as a function of the number of DOFs in the
model or as a function of 1/h, where h is the element size.
•• Observe the change in the location of stress concentrations.
•• Repeat the exercise using h adaptive solution; mesh refinement process is not
necessary when h adaptive solution is used.
•• Repeat the exercise using p adaptive solution; mesh refinement process is not
necessary when p adaptive solution is used.
•• Repeat the exercise after removing small outside rounds which have no structural
significance; compare mesh complexity before and after removal of rounds. Do not
remove the large fillets because they are structurally important. The removal would
produce stress singularity.
5.4.2. CANTILEVER
Description
The cantilever beam has fixed restraints applied to the back face. It is loaded with 1000 N
force uniformly distributed to the blue face (Figure 5.25).
Objective
Demonstrate the effect of an incorrect mesh on displacement and stress results.
Procedure
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
6
Modeling Process
I
n any FEA project, most effort is spent creating the mathematical model which is the basis
for the finite element model. When the mathematical model is ready, it is discretized to
turn it into a finite element model.
We discuss issues related to the mathematical model after discussing mesh-related
issues. This order is reversed as compared to how these steps appear in an FEA project.
We do that because we are now better prepared to appreciate the differences between the
CAD model, the mathematical model, and the finite element model. We now realize that
most topics discussed in FEA training courses are not specific to the FEA. Instead, they
deal with the mechanics of materials and with techniques used for the creation of mathe-
matical models. Decisions on the type of analysis (linear or nonlinear?), the dimensionality
of analysis (2D or 3D?), idealization and defeaturing of CAD geometry, applying loads,
restraints, and so forth all belong to the process of creating a mathematical model and not
an finite element model. Still, in everyday FEA practice, the tasks of creating a mathematical
model and the task of creating a finite element model are tightly knitted together. This
justifies the emphasis on mathematical models and modeling techniques used in
their creation.
No matter how much time and effort we spend on modeling, a mathematical model
can never be accepted as a final and true description of the system. Rather, it can at best
be regarded as a good enough description of certain aspects of the system that are of
particular interest to us. Since the ultimate objective is to make a design decision, the
mathematical model should only be good enough to allow making that decision with
reasonable confidence. From this point of view, the best mathematical model is the simplest
one that still provides reliable results. We will review typical steps in creating a mathematical
model and offer tips helpful in creating models that provide the data of interest without
unnecessary complexity.
SI
Unit of length m
Unit of mass kg
Unit of mass density kg/m3
Density of aluminum 2794 kg/m3
System derived from SI
Unit of length mm
Unit of mass tonne
Unit of mass density tonne/mm3
Density of aluminum 2.794 × 10−9 tonne/mm3
Imperial (IPS)
Unit of length in
Unit of mass slug/12
Unit of mass density slug/12/in3
Density of aluminum 2.614 × 10−4 slug/12/in3
6.1.3. G
eometry Preparation
“Geometry preparation” means converting CAD-specific geometry into FEA-specific
geometry, the one that captures all important model features but avoids unnecessary
complexity and, at the same time, meshes correctly.
A common mistake is an attempt to use fully featured, manufacturing-ready CAD
geometry for analysis. Meshing complex CAD geometry is often impossible or the created
mesh is too large to solve or, if it is not too large, it does not have enough elements to capture
essential behavior or has excessively distorted elements. Even if you can create a correct
mesh based on a fully featured CAD geometry model, a very detailed representation of
geometry results in an expensive FE model with little, if any, benefits to results.
When preparing analysis-specific geometry, use Saint-Venant’s principle and concen-
trate modeling details in the regions of interest. Solid elements should not be necessarily
our first choice. Consider using shells, beams, or 2D representation.
The distinction between manufacturing specific CAD geometry and analysis specific
FEA geometry is one of the challenges facing design engineers. We will further discuss it
in Chapter 13.
© SAE International.
6.2. S
elected Modeling Techniques
6.2.1. M
irror Symmetry and Anti-symmetry
Boundary Conditions
The bracket shown in Figure 6.2 is subjected to a load uniformly distributed over the cylin-
drical face; the load is pointing down. The back face is fixed.
The model is characterized by double mirror symmetry. Consider how the bracket will
deform when loaded. Due to the symmetry of geometry as well as the symmetry of loads
and supports about the blue plane, the imaginary cross-section along the plane of symmetry
remains flat in the deformed model. Therefore, the model may be simplified by cutting it
in half along the blue plane removing one-half of the model. Either the left or right half of
the model may be analyzed. One-half of the original load needs to be applied to the
remaining half of the model. Notice that the load needs to be cut in half only if it is defined
as a total load; load defined as force per area (pressure) remains the same. To make the
remaining half of the model behave as if the other half was still there, we must define mirror
symmetry boundary conditions for blue faces (Figure 6.3).
The boundary conditions are necessary to keep the faces flat and coincident with the
plane of symmetry while the model deforms. These conditions are mirror symmetry
boundary conditions, often simply called symmetry boundary conditions. Working with
FIGURE 6.2 A bracket has double symmetry. The symmetry about the blue plane will
be used to define mirror symmetry boundary conditions. The symmetry about the green
plane will be used to define anti-symmetry boundary conditions. The bracket is loaded with
load pointing down (perpendicular to the green plane) evenly distributed over the cylindrical
face. Load and restraints symbols are not shown.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 6.3 To make half of the original geometry represent the whole model, mirror
symmetry boundary conditions must be applied to blue faces created by a cut along the
vertical plane. One-half of the total load is applied to this model. Either the right or left half of
the model may be analyzed.
© SAE International.
a model intended for meshing solid elements with 3 DOFs per node, the symmetry boundary
conditions need to be defined in terms of translations only. Translations in the direction
normal to the blue face are restrained, and translations in in-plane directions are permitted.
Working with a model intended for meshing with shell with 6 DOFs per node, the symmetry
boundary conditions must be also defined in terms of rotations. Rotations in the direction
normal to the plane of symmetry are permitted, and both in-plane rotations are eliminated.
The model shown in Figure 6.2 can be also simplified by taking advantage of the anti-
symmetry present in this problem (Figure 6.4).
The anti-symmetry boundary conditions apply when geometry and restraints are
symmetric about a plane (here the green horizontal plane) and loads are anti-symmetric
as shown in Figure 6.5. The anti-symmetry boundary conditions are exactly opposite to the
symmetry boundary conditions.
Tables 6.2 and 6.3 explain how mirror symmetry and anti-symmetry boundary condi-
tions are defined in each of the three principal planes in the orthogonal coordinate system.
The comparison shows that anti-symmetry boundary conditions can be seen as exactly
opposite to symmetry boundary conditions. Therefore, anti-symmetry boundary conditions
can be defined by reversing symmetry boundary conditions. Displacement components
allowed in symmetry boundary conditions are restrained in anti-symmetry boundary
conditions and the other way around.
Using both symmetry and anti-symmetry boundary conditions leads to further simpli-
fication because now only one-quarter of the bracket geometry needs to be modeled. Using
one-quarter of the bracket requires that one-quarter of the load be applied; this is because
the load is defined as a total load, not as load per unit of area. Different model simplifica-
tions are shown in Figure 6.6.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 6.4 To make this half of the original geometry represent the whole model, the
anti-symmetry boundary conditions must be applied to the green face created by the
symmetry cut along the horizontal plane. One-half of the total load is applied to this model.
Either the top or bottom half of the model may be analyzed.
© SAE International.
FIGURE 6.5 The anti-symmetry boundary conditions require symmetry of geometry and anti-symmetry of loads.
This illustration shows the cross-section along the vertical plane of symmetry only to show the hole. The model intended
for analysis with anti-symmetry boundary conditions is obtained by cutting the original model along the plane of anti-
symmetry only, this is the horizontal plane. Two halves of the model shown on the right serve to visualize the anti-symmetry
of loads.
TABLE 6.2 Definition of symmetry boundary conditions in the three planes of an orthogonal
coordinate system.
Plane of symmetry xy yz xz
Translation x Free Restrained Free
Translation y Free Free Restrained
Translation z Restrained Free Free
Rotation x Restrained Free Restrained
Rotation y Restrained Restrained Free
Rotation z Free Restrained Restrained
FIGURE 6.6 Faces with mirror symmetry boundary conditions are colored blue; faces
with anti-symmetry boundary conditions are colored green. Either one of the above four
models can be used for analysis.
FIGURE 6.7 The use of anti-symmetry boundary conditions allows to model one-half of
an automotive body subjected to torsion along the longitudinal axis. Anti-symmetry
boundary conditions are applied to edges in the plane of symmetry.
© SAE International.
FIGURE 6.8 Anti-symmetry boundary conditions are applied to a shaft subjected to torsion.
© SAE International.
Anti-symmetry
boundary conditions
FIGURE 6.9 An axisymmetric tank, loaded with pressure distributed over a 360° angle
can be represented by a 2D radial section. The section is meshed with 2D axisymmetric
elements. The load applied to a 2D axisymmetric model is interpreted as a load applied over a
360° angle.
© SAE International.
6.2.3. Cyclic Symmetry
Cyclic symmetry is also called repetitive symmetry. Cyclic symmetry may be of angular
or linear type. Angular cyclic symmetry, often called circular symmetry, is applicable to
shapes which may be created by an angular pattern of one segment. The repeatability must
include not only geometry but also loads and restraints. Circular symmetry enforces the
same displacements on the corresponding faces shown in Figure 6.10.
A linear cyclic symmetry works with repeatable shapes that can be generated with
linear patterns; an example is shown in Figure 6.11.
Combinations of different forms of symmetry can also occur. For example, Figure 6.6
demonstrates the use of mirror symmetry and anti-symmetry in one model. The blower
model in Figure 6.10 could be further simplified by taking advantage of mirror symmetry
about the midplane in between the endplates.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 6.10 An impeller exhibits circular symmetry. One section may be copied in a
circular pattern about the axis of the impeller to create the complete model. The impeller is
loaded with a centrifugal load. The load is also repetitive in each section. Faces “a” and “b” of
the section must experience the same displacements, and this is enforced by the circular
symmetry boundary conditions applied to one segment.
Face b2
Face a2
Face b1
© SAE International.
Face a1
FIGURE 6.11 A rack exhibits linear cyclic symmetry. One section may be copied in a linear
pattern to create the complete model. If loads and restraints are also linearly repetitive, then
faces “a” and “b” must experience the same displacements. This is enforced by the cyclic
symmetry boundary conditions applied to one segment.
Face b
© SAE International.
Face a
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 6.12 DOFs of nodes on the face of the hole are aligned with directions of the
cylindrical coordinate system associated with the hole; these are radial, tangential, and axial
translations. Restraining radial translations (red arrows) simulate a pin support. Restraining axial
directions (yellow arrows) prevents the model from sliding along the axis of the hole. There are
no restraints in the circumferential direction allowing the tube to rotate about the axis.
© SAE International.
FIGURE 6.13 The shaft is supported by two spherical bearings that allow for some rotations
in the plane of bending. To model a spherical bearing support, the DOFs on the spherical faces
are aligned with respective directions of the spherical coordinate system: The support offered
by a spherical bearing is modeled by restraining translations in the radial direction (red arrows).
The spherical face of the left bearing is also restrained in circumferential direction (yellow
arrows) to prevent the shaft from spinning about the axis.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
6.2.5. U
sing Point Restraints to Eliminate Rigid
Body Motions
Restraints applied to entities with zero area (point or line) can be used if those restraints
do not generate reactions and are used only to eliminate Rigid Body Motions (RBMs).
Otherwise, restraints applied to entities with zero area create displacements and
stress singularities.
Figure 6.14 illustrates eliminating RBMs in a 3D model subjected to internally balanced
loads. The restraints are necessary to eliminate RBMs.
Restraints 1, 2, and 3 in Figure 6.14 eliminate all RBMs, but do not serve to provide
reactions to the applied loads and, therefore do not interfere with model deformation.
Therefore, these restraints do not introduce displacement or stress singularities.
FIGURE 6.14 Opposite and equal tensile load (magenta arrows) is applied to both
vertical faces. Restraint 1 (red arrows) is applied in three orthogonal directions to one
corner. Restraint 1 eliminates translational RBMs but leaves the model with three
rotational RBMs. Restraint 2 (green arrows) eliminates rotation about the y and z axes
leaving the model with one rotational RBM. Restraint 3 (blue arrow) eliminates rotation
about the x axis.
Y
© SAE International.
Z X
1
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Objective
Demonstrate the use of mirror symmetry boundary conditions.
Procedure
•• If you use Parasolid geometry, cut the model in half as shown in Figure 6.15; if you use
SOLIDWORKS geometry, switch to configuration 02 half sym.
•• Apply the material properties (steel).
•• Apply a fixed restraint to the back face and apply mirror symmetry boundary
conditions to two faces located in the plane of the cut; these are blue faces shown
in Figure 6.15.
•• Apply half of the total load to half of the cylindrical face; refer to Figure 6.3 for load
direction.
•• Mesh and solve.
•• Analyze the displacement and stress results.
FIGURE 6.15 Complete BRACKET02 model and one-half of the model used with
symmetry boundary conditions.
© SAE International.
6.3.2. BRACKET02—Anti-symmetry BC
Description
The bracket is supported along the back side and loaded with 10000 N bending load. Only
one-half is modeled and the other half is simulated with the anti-symmetry boundary
conditions (Figure 6.16). Displacements are small; therefore, non-following anti-symmetry
boundary conditions may be used.
Objective
Demonstrate the use of anti-symmetry boundary conditions.
Procedure
•• If you use Parasolid geometry, cut the model in half as shown in Figure 6.16; if you use
SOLIDWORKS geometry, switch to configuration 03 half anti sym.
•• Apply the material properties (steel).
•• Apply a fixed restraint to the back face and define anti-symmetry boundary
conditions (in-plane translations are eliminated, out-of-plane translations are
allowed).
•• Apply half of the total load to the cylindrical face.
•• Mesh and solve.
•• Analyze the animated displacement plots and stress plots; compare with
Exercise 6.3.1.
FIGURE 6.16 Complete BRACKET02 model and one-half of the model used with anti-
symmetry boundary conditions.
© SAE International.
6.3.3. B
RACKET02—Mirror Symmetry and
Anti-symmetry BC
Description
This exercise combines modeling techniques practiced in Exercises 6.3.1 and 6.3.2. If you use
Parasolid geometry, cut the model twice to quarter as shown in Figure 6.17; if you use
SOLIDWORKS geometry, switch to configuration 04 quarter. Displacements are small;
therefore, non-following anti-symmetry boundary conditions may be used.
Objective
Demonstrate the use of symmetry and anti-symmetry boundary conditions to model bending.
Procedure
•• Apply one-quarter of the total load to the cylindrical face.
•• Apply restraints: mirror symmetry and anti-symmetry boundary conditions.
•• Mesh and solve.
•• Compare the displacement and stress results produced in Exercises 6.3.1, 6.3.2,
and 6.3.3. Results will be slightly different because of different discretization errors
in each solution. Once convergence analysis is conducted, the results of all three
models will converge.
FIGURE 6.17 Complete BRACKET02 model and one-quarter of BRACKET02 model used
with mirror symmetry and anti-symmetry boundary conditions.
© SAE International.
6.3.4. SHAFT01
Description
The shaft is supported at the back face and loaded with a couple of 1000 N forces applied
to two square holes. Only one-half of the shaft needs to be analyzed if anti-symmetry
boundary conditions are used. Load is applied to holes in the direction of arrows; it is
uniformly distributed over a flat face that forms a part of the hole. Displacements are small;
therefore, non-following anti-symmetry boundary conditions may be used.
Objective
Demonstrate the use of anti-symmetry boundary conditions to model torsion.
Procedure
•• If you use Parasolid geometry, cut the model in half as shown in Figure 6.18; if you use
SOLIDWORKS geometry, switch to configuration 02 half.
•• Apply the material properties (steel).
•• Apply a fixed restraint to the end face.
•• Apply anti-symmetry boundary conditions (in-plane translations are eliminated,
out-of-plane translations are allowed).
•• Apply a 1000 N load as shown in Figure 6.18.
•• Mesh and apply local mesh refinement to the fillet.
•• Solve.
•• Analyze the animated displacement plots and stress plots.
•• Repeat the analysis using full geometry; compare displacement and stress results.
FIGURE 6.18 Complete SHAFT01 model and one-half of SHAFT01 model used with anti-
symmetry boundary conditions.
© SAE International.
Objective
Demonstrate the use of symmetry boundary conditions when shell elements are used.
Procedure
•• If you use Parasolid geometry, cut the model in half to work with one-half of the
geometry; if you use SOLIDWORKS geometry, switch to configuration 02 half.
•• Define the material properties (steel).
•• Define the wall thickness (2 mm).
•• Define the fixed restraints to the flat bottom.
•• Define the symmetry boundary conditions along the edges in the plane of symmetry.
Translation in the direction normal to the plane of symmetry and two in-plane
rotations are restrained; consult Table 6.2.
© SAE International.
•• Define the pressure load as 1 MPa; since this is a pressure and not the total load, the
pressure magnitude does not change when working with half of the model.
•• Mesh and solve.
•• Analyze the animated displacement plots and stress plots.
•• Repeat the analysis using the full model.
•• Compare the results of analyses of the half model and the complete model.
6.3.6. R
ING
Description
An axisymmetric ring is subjected to 100 MPa pressure applied to the toroidal face (blue).
Sliding support is applied to the bottom face. Taking advantage of the axial symmetry, the
analysis may be conducted on the radial section shown in Figure 6.20.
Objective
Demonstrate the use of 2D axisymmetric analysis.
Procedure
6.3.7. LINK01
Description
A pin-supported link is loaded with 100 N load uniformly distributed over the top face.
The left pin is fixed and the right one has a floating support shown in Figure 6.21 by rollers.
If contact stresses between the pins and holes are not of interest, then the analysis may
be conducted on the part, rather than on the assembly.
If a full model is analyzed, then restraints must be defined in the cylindrical coor-
dinate systems associated with respective holes. Taking advantage of the mirror
symmetry, the analysis may be conducted on one-half of the model with the mirror
symmetry boundary conditions defined and half of the load applied to the top face.
We assume that one of the hinge supports is “floating,” meaning that the distance between
hinges may change. This prevents the link form developing membrane stresses; analysis
of membrane stresses would require a nonlinear geometry analysis. This will be discussed
in Chapter 7.
Objective
Demonstrate the use of symmetry boundary conditions in a solid element model.
Demonstrate how realignment of the DOFs allows for modeling a hinge support.
100 N
Fixed hinge
50 N
Floating hinge
© SAE International.
Hinge support
Procedure
•• If you use Parasolid geometry, cut the model in half. If you use SOLIDWORKS
geometry, switch to configuration 02 half.
•• Apply the material properties: Nylon 6/10 (E = 8000 MPa, ν = 0.28). Align the DOFs
of the cylindrical faces with the respective cylindrical coordinate system. After
realignment the directions of translational DOFs are radial, circumferential, and
axial. There are no rotational DOFs because the model uses solid elements.
•• Restrain the radial and axial DOFs.
•• Apply the symmetry boundary conditions (restrain translations in the direction
normal to the face in the plane of symmetry).
•• Apply one-half load (50 N) to the top face as shown.
•• Mesh and solve.
•• Analyze the animated displacement plots and stress plots.
•• Repeat the analysis using the complete model and compare results.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
7
Nonlinear Geometry
Analysis
7.1. C
lassification of Different Types of
Nonlinearities
Before we discuss different types of nonlinear analysis, we need to review some terms.
Structural analysis deals with deformable bodies; it finds displacements, strains, and
stresses. The opposite of a deformable body is a rigid body that does not deform under the
applied load. Rigid bodies are universally used in mechanism analysis but cannot be analyzed
using tools of structural analysis. Elastic bodies are deformable bodies that return to the
original shape when the load is removed. Linear structural analysis deals with elastic bodies
which stiffness does not change during the process of load application. Static analysis is a
structural analysis with slowly changing loads so that the inertial effect can be ignored. In
this chapter we discuss nonlinear static structural analysis with stiffness changing during
the process of load application.
Recall the fundamental FEA equation:
K d F (7.1)
where [K] is the stiffness matrix, [d] is the unknown vector of nodal displacements, and [F]
is the vector of nodal loads. Until now we have always assumed that the stiffness matrix
[K] is constant, meaning it is not a function of displacements. The only exception was the
introduction of elastic-perfectly plastic material with a strain-stress curve shown in
Figure 4.20. That material was used to discuss stress singularities.
The assumption that stiffness does not change during the process of load application
defines linear analysis. If stiffness does not change during the loading process, then stiffness
matrix [K] is calculated only once and no updates are necessary. If stiffness does change
during the loading process, then analysis becomes nonlinear, and it is necessary to update
the stiffness matrix [K] while loads are being applied. It is also necessary to introduce a
load time history that defines load as a function of time.
Nonlinear analysis may be classified based on the nature of nonlinear behavior. In this
chapter we will review nonlinearity caused by changes in model geometry and contact. In
Chapter 8 we will discuss nonlinear material and in Chapter 10 buckling.
FIGURE 7.1 Cantilever beam subjected to pressure. The left end is fixed; the right end is
free to move. The model is shown in an undeformed shape.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.2 Time history of pressure applied to the cantilever beam in Figure 7.1.
0.030
0.025
0.020
Pressure [MPa]
0.015
0.010
0.005
© SAE International.
0.000
0 0.2 0.4 0.6 0.8 1
Time [s]
Step 4
0.0045 MPa
Step 7
0.0135 MPa
Step 13
0.03 MPa
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.4 The following load keeps its orientation relative to the model. Here, the
pressure load remains normal to the top face. The following load is also called non-
conservative load because it does not retain its original direction.
© SAE International.
x
Another possibility is a non-following load that retains its original direction relative
to the global coordinate system (Figure 7.5). The non-following load could be produced by
its own weight.
A load retaining its direction relative to the deforming model is called a non-conser-
vative load or following load. A load retaining its direction relative to the global coordinate
system is called conservative or non-following load.
FIGURE 7.5 Non-following load retains its original orientation relative to the global
coordinate system. Non-following load is also called conservative load because it retains its
original direction relative to the global coordinate system.
y
© SAE International.
x
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.6 Model deformation found in linear analysis. The deformed model shows
apparent stretching. These results are incorrect.
Step 4
0.0045 MPa
Step 7
0.0135 MPa
Step 13
0.03 MPa
© SAE International.
FIGURE 7.7 A round shaft loaded with a couple of forces. The forces retain their direction
relative to the deforming model. The far end of the shaft is fixed.
© SAE International.
FIGURE 7.8 Displacement plots produced by linear analysis. The holes travel along
straight lines and the shaft “swells” seemingly increasing its diameter. This is because, in
linear analysis, the model shape is not updated and displacements are superimposed on the
undeformed model. There is no intermediate shape between the undeformed model and the
deformed model.
© SAE International.
FIGURE 7.9 The green circle is the diameter of the unloaded shaft. The red circle is the
diameter of the shaft loaded with a torque. Since the analysis is linear, trajectories of points
traveling from the undeformed to the deformed position are straight lines; point A displaces
along a straight line to position A′. This, when repeated for all points on the circumference of
the green circle, produces an apparent increase in the shaft diameter.
A A′
D′
D
B
B′
© SAE International.
C′ C
FIGURE 7.10 Displacement results of nonlinear geometry analysis. The holes travel along
an arc as they should, and the shaft diameter does not increase.
© SAE International.
FIGURE 7.11 Model of a shaft under torsion before and after deformation using the true
scale of deformation. The linear analysis produces an incorrect “swollen” shape. The
nonlinear, large displacement analysis produces the correct shape.
© SAE International.
Undeformed model Results of linear analysis Results of nonlinear analysis
7.3. M
embrane Stress Stiffening
Beam in bending and shaft in torsion problems required nonlinear geometry analysis
because large displacements were changing stiffness significantly. However, there are
problems where nonlinear geometry analysis is required, even though displacements are
small relative to the model size. An example is a thin flat plate under pressure (Figure 7.12).
Initially, when the load is applied to the flat plate, the only mechanism available to resist
the load is bending stiffness. During the process of deformation, the plate acquires membrane
stiffness in addition to the original bending stiffness (Figure 7.13).
Neglecting nonlinear effects in the Round Plate model causes a 55% error in displace-
ment results. This high error occurs even though displacements are small. The maximum
plate displacement is 5 mm when accounting for the membrane stiffness effect and 7.7 mm
when linear analysis is used, and membrane effects are erroneously ignored (Figure 7.14).
FIGURE 7.12 A thin flat plate with a diameter of 500 mm and thickness of 5 mm is rigidly
supported along the circumference. It is subjected to pressure slowly changing from 0 to
0.3 MPa. The plate material is Alloy Steel.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.13 The flat plate has a bending stiffness only, and it responds to pressure with
bending stresses. The deformed plate responds to pressure with bending stresses and tensile
(membrane) stresses; it has bending and membrane stiffness.
Bending stresses
A
© SAE International.
FIGURE 7.14 Displacement results of linear and nonlinear analyses. Using the true scale of
deformation, deformation is barely recognizable. Neither displacement of 7.7 mm nor 5.0 mm
are “large displacement;” still the problem requires nonlinear geometry analysis.
7.7 5.0
7.0 4.5
6.4 4.1
5.7 3.7
5.1 3.3
4.5 2.9
3.8 2.5
3.2 2.1
2.6 1.7
© SAE International.
1.9 1.2
1.3 0.8
0.6 0.4
0.0 0.0
Linear solution, max. displacement 7.7 mm Nonlinear solution, max. displacement 5.0 mm
FIGURE 7.15 Load-displacement curve in the linear solution is a straight line. Plate
stiffness is calculated for the original, flat shape. The stiffness is not updated, and the load is
applied in one step. The load-displacement curve in the nonlinear solution is a curve built on
13 points. The solution was completed in 13 load increments; stiffness was updated 13 times.
The shape of the load-displacement curve demonstrates that plate stiffness increases while
it deforms.
2.00
© SAE International.
0.00
0 0.1 0.2 0.3
Pressure [MPa]
Consider a link supported by two hinges in Figure 7.16. When a linear analysis is used,
it is assumed that one of the hinges is floating as if it was supported by rollers; only bending
is present. But if both hinges are in the fixed position, then the load produces both bending
and elongation, and linear analysis cannot model that. A linear analysis is based on the
stiffness of the undeformed link. That stiffness is based on bending stresses only; membrane
stiffness produced by tensile stresses is not accounted for (Figure 7.13). Linear analysis cannot
distinguish between fixed and floating supports (Figure 7.16). When analyzed with linear
analysis, both configurations shown in Figure 7.16 produce the same displacements in the
direction of load and the floating hinge does not move at all. The analysis of configuration
with both hinges fixed requires nonlinear geometry analysis even if displacements are small.
Configuration with one floating hinge may be solved with linear analysis if displace-
ments are small and horizontal displacement of the floating hinge is not of interest. However,
if the horizontal displacement of the floating hinge is of interest, then nonlinear geometry
analysis is required. This is because the hinge translation is caused by the model deforma-
tion that happens during the process of load application.
Both linear and nonlinear analyses of the assembly shown in Figure 7.16 may be simpli-
fied to the analysis of one part, as shown in Figure 7.17.
Consider the beam shown in Figure 7.18. The wide end is held in a fixed position; the
narrow end (shown in blue) can slide in the horizontal direction. A pressure is applied to
the top face. We are looking for displacement in the direction of load and for horizontal
displacement of the sliding end. Let us review the results of linear and nonlinear
geometry analyses.
Displacement plots in the true scale of deformations found in linear and nonlinear
analyses are shown in Figure 7.19.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.16 Configuration with a floating hinge (top) may be analyzed with linear
analysis if displacements are small and the horizontal translation of the floating hinge is not of
interest. If the horizontal translation of the floating hinge is of interest, then nonlinear
geometry analysis is required. Analysis of configuration with both hinges in fixed position
always requires nonlinear geometry analysis.
Floating hinge
Fixed hinge
Fixed hinge
© SAE International.
Fixed hinge
FIGURE 7.17 Hinge supports are applied to red cylindrical faces. If the link is analyzed
with linear analysis, then the results are valid for configuration with one floating hinge. If it is
analyzed with a nonlinear geometry analysis, then the results are valid for configurations with
both hinges in fixed positions.
Hinge support
© SAE International.
Hinge support
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.18 A thin beam under pressure has one end fixed; the other end may slide in
the horizontal direction.
© SAE International.
FIGURE 7.19 Linear displacement analysis (top) and nonlinear geometry analysis
(bottom) both produce almost the same resultant displacement.
2.86 2.85
2.62 2.61
2.38 2.38
2.14 2.14
1.90 1.90
1.67 Deformed model in linear analysis 1.66
1.43 1.43
1.19 1.19
0.95 0.95
0.71 0.71
0.48 Deformed model in nonlinear geometry analysis 0.48
0.24
© SAE International.
0.24
0.00 0.00
Displacements Displacements
in linear analysis in nonlinear
geometry analysis
The maximum displacement that takes place in the direction of the load is practically
the same in both analyses. If the maximum displacement is all we are interested in, then
the linear analysis suffices.
Figure 7.20 shows displacement plots on a 10:1 scale. The exaggerated scale of deforma-
tion reveals that, in the linear analysis, the sliding end does not slide at all. In the nonlinear
geometry analysis, the sliding is modeled. Therefore, if displacements of the sliding end are
of interest, the problem requires nonlinear geometry analysis.
It is difficult to establish a “rule of thumb” that would say when it is OK to run a linear
analysis and when a nonlinear geometry analysis is required. The magnitude of displace-
ment does not differentiate between linear geometry and nonlinear geometry analysis. The
differentiating factor is whether displacements cause a significant change in stiffness.
In summary, we need to differentiate between the terms “Large displacement analysis”
and “Nonlinear geometry analysis.” These two terms are often confused because, in some
commercial FEA programs, the nonlinear geometry analysis is called “Large displacement
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.20 Undeformed beam (top), deformed shape in linear analysis (middle), and
deformed shape in nonlinear analysis (bottom). The linear analysis does not show any
displacements of the sliding end. The nonlinear analysis does show displacement of the
sliding end. The scale of deformation is 10:1.
Undeformed model
Deformed shape
in linear analysis
Sliding end
does not move
geometry analysis
Sliding end moves
7.4. Contact
Analysis of contact is another type of nonlinear geometry analysis. The source of nonlinear
behavior may be closing the gap or a change in the size of the contact area during the process
of load application. Even though displacements may be very small in comparison to the
overall model size, the change of stiffness is significant and, therefore, requires a
nonlinear analysis.
Analysis of contact between two or more bodies may be conducted to find contact
stresses. Proper meshing is a very important modeling consideration in contact stress
analysis, but often it is difficult to construct a correct mesh. For correct results of contact
stress analysis, elements in and around the contact zone must be sized accordingly to the
expected size of the contact area. However, automatic meshing selects the size of the mesh
based on the model geometry and not based on the size of the contact area that is unknown
prior to the analysis. Automatically generated mesh is rarely adequate for analysis of the
contact stresses. Mesh controls must be used to make elements small enough to model
contact stress. We will illustrate this with an example of an analysis of contact stresses
between two cylindrical faces of a clamp shown in Figure 7.21. Initially, there is a small gap
between the two faces; it closes under the applied load.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.21 Load applied to the top face closes the gap between two cylindrical faces.
The bottom face of the clamp is fixed.
© SAE International.
FIGURE 7.22 Mesh with elements this large cannot be used to model contact stresses.
With elements this large, contact is modeled as a line contact between two rows of nodes.
© SAE International.
A mesh with the default element size does not come even close to the correct modeling
of contact stress. As shown in Figure 7.22, contact is modeled as a line with several nodes
along its length.
A very fine mesh is required to model contact stresses correctly. We demonstrate it
using a 2D plane strain representation (Figure 7.23).
The 2D mesh is shown in Figure 7.24 and stress results are shown in Figure 7.25.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.23 A 2D slice in the plane of symmetry of the model will be meshed with 2D
plane strain elements. 2D representations make it practical to use small elements in the
contact area.
© SAE International.
FIGURE 7.24 A 2D mesh showing very small elements in the contact area.
© SAE International.
The model in Figure 7.21 has a very small gap and the upper arm does not have to move
a lot to close the gap. Therefore, a large displacement analysis is not required. However, the
model in Figure 7.26 does require a large displacement analysis or else the gap closes incor-
rectly (Figure 7.27).
Another category of contact problem is shown in Figure 7.28: an oversized ring with a
gap is assembled into a cylinder. The ring must shrink to fit into the cylinder.
Figure 7.29 shows the model geometry before and after solution. The initial interference
between the assembly components is eliminated. Inserting the ring into the cylinder expands
the cylinder and shrinks the ring.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.25 Von Mises stress results produced by the 2D plane strain model. Correct
modeling of contact stresses in a very small region of the model requires very small elements
as shown in Figure 7.24
© SAE International.
FIGURE 7.26 CLAMP02 loaded with pressure 0.03MPa
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 7.27 Incorrect results of linear analysis and correct results of nonlinear analysis.
Undeformed shapes are superimposed on the deformed plots. To model the closing of a gap
this large requires a large displacement analysis. This problem combines two types of
nonlinearities: large displacements and contact.
© SAE International.
Incorrect gap closure in linear analysis Correct gap closure in nonlinear analysis
FIGURE 7.28 An oversized ring with a gap is installed in a cylinder. The shrinkage
develops contact stresses between the touching faces of the ring and the cylinder.
© SAE International.
FIGURE 7.29 Model before and after solution. The original interference between the ring
and the cylinder has been eliminated in the deformed shape.
© SAE International.
Before solution After solution
FIGURE 7.30 Contact pressure plot on the faces of the ring and the cylinder. The
assembly is shown in two different exploded views to show the faces in contact.
© SAE International.
The ring is held in a deformed shape by the contact pressure between the contacting
faces (Figure 7.30).
Objective
Demonstrate the differences between an incorrect linear solution and two correct nonlinear
geometry solutions: with the following load and with the non-following load.
Procedure
7.5.2. SHAFT02
Description
A long round shaft is subjected to a torque load produced by a couple of forces acting on
two square holes in the lug (Figure 7.7). Flat faces in the square holes make it easy to apply
the following load. Examine the CAD model and notice that the shaft is longer than what
is shown Figure 7.7. Long shaft has low stiffness in torsion resulting in large displacements.
Objective
Demonstrate differences between an incorrect linear solution and the correct nonlinear
geometry solution with the following load.
Procedure
•• Assign the material properties: Steel.
•• Apply restraints: fixed to the end of the shaft.
•• Apply a load of 3500 N to each square hole as shown in Figure 7.7.
•• Mesh with second order solid elements, at least four elements across the shaft
diameter; use mesh control on the fillet at the back side of the lug to control the
element turn angle.
•• Find the linear solution.
•• Find the nonlinear large displacement solution with the following load.
•• Compare displacements and stress results from the above two solutions.
Procedure
The following assumes that the complete model is analyzed with solid elements.
•• Assign the material properties: Steel.
•• Apply fixed restraints to the cylindrical face along the perimeter.
•• Apply a load of 0.5 MPa to the top face.
•• Mesh with the second order solid elements, at least one layer of correctly shaped
elements across the plate thickness.
•• Find a linear solution.
•• Find a nonlinear large displacement solution with the following load.
•• Compare displacements and stress results from the linear and the nonlinear
solutions.
7.5.4. LINK02
Description
A link is supported by two hinges in the fixed positions as shown in Figure 7.16, bottom. A
uniformly distributed pressure is applied to the top face. We are looking for displacements
and stresses in the link, but not for contact stresses at hinge supports (Figure 7.17). This
problem could be analyzed using one-half of the link with symmetry boundary conditions.
Objective
Demonstrate that the linear analysis does not model tensile stresses developing during the
process of load application. As a result, a linear analysis produces incorrect displacement
and stress results.
Procedure
Objective
Demonstrate that a linear analysis does not model displacement of the sliding end. You will
see that the results of the linear analysis are the same for the sliding end and the fixed end.
A 2D plane stress or 2D plane strain representation could be used to demonstrate that the
sliding end does not slide in the linear analysis.
Procedure
•• Assign the material properties: Steel.
•• Apply restraints: fixed support to the thick end; sliding support to the thin end.
•• Apply a load of 0.1 MPa uniformly distributed pressure to the top face.
•• Mesh with the second order solid elements, one layer of correctly shaped elements
across the beam thickness.
•• Find the linear displacement solution; notice that the sliding end does not move.
•• Find the nonlinear large displacement solution with the following load and observe
the translation of the sliding end.
7.5.6. C
LAMP01
Description
A load applied to the top arm closes the gap between two cylindrical faces (Figure 7.21).
We are looking for contact stresses that develop when the gap closes. A 2D plane strain
model is used because it offers a significant reduction in the time required for meshing
and solving. There is no need for a large displacement solution. This problem can
be solved using sufficiently small solid elements if long meshing and solution times
are acceptable.
Objective
Analyze the gap closure and contact stresses. The importance of using small elements in
the contact area.
Procedure
•• Define the mesh control along two edges that will come in contact: use element
size of 0.05 mm.
•• Mesh with 2D plane strain elements.
•• Find the nonlinear displacement solution and analyze contact stresses.
7.5.7. CLAMP02
Description
A plastic clamp is closed by pressure applied to the top arm and fixed restraint is applied
to the corresponding bottom face (Figure 7.26). Gap closing is demonstrated with 2D plane
strain analysis. High stresses develop in the arch, not in the contact area.
Objective
Demonstrate the problem that combines two types of nonlinearities: large displacements
and contact. Demonstrate incorrect gap closing in linear analysis.
Procedure
8
Nonlinear Material
Analysis
8.1. R
eview of Nonlinear Material Models
A linear material model is defined by the modulus of elasticity and Poisson’s ratio. The
linear relationship between strain and stress demonstrates that the modulus of elasticity
does not change with strain (Figure 8.1).
A nonlinear material model is one that does not follow the linear relationship between
strain and stress and, therefore, its modulus of elasticity changes with strain. Using such
material, model stiffness changes during the loading process and the stiffness matrix must
be recalculated during the iterative solution.
A simple nonlinear material model is the elastic-perfectly plastic material. It
assumes a linear relationship between strain and stress up to a certain level of stress
called yield strength. Once the yield strength is reached the stress remains constant
regardless of the strain. As compared to the linear material model, the only additional
piece of information required is the yield strength, usually defined using von Mises
stress. There are many other material models such as nonlinear material with soft-
ening (Figure 8.2), bilinear material with hardening ( Figure 8.2), and many more.
We limit the review of nonlinear material models to the elastic-perfectly plastic
material model.
A bracket made from aluminum with a yield strength of 27.6 MPa is shown in Figure 8.3.
A linear analysis is run first; a plot of von Mises stresses is shown in Figure 8.4. Large
portions of the model show stress above the yield strength.
The next analysis uses the elastic-perfectly plastic material model and provides results
shown in Figures 8.5 and 8.6.
E= = tan
© SAE International.
s s s
© SAE International.
e e e
(a) Elastic-perfectly plastic (b) Nonlinear with softening (c) Bi-linear with hardening
FIGURE 8.3 A bracket loaded with a load uniformly distributed to the blue face and
supported at the back face.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 8.4 Von Mises stress plot produced with a linear material model. The red arrow
on the color legend indicates the yield strength of the material. Large portions of the model
show stress magnitude above the yield strength of 27.6 MPa.
41.3
.
37.8
34.4
31.0
27.5
24.1
20.6
17.2
13.8
10.3
6.9
© SAE International.
3.4
0.0
Yield strength: 27.6
FIGURE 8.5 Von Mises stress plot produced using the elastic-perfectly plastic material
model. The maximum von Mises stress is 27.6 MPa, which corresponds to the flat portion of
the strain-stress curve in the material model.
27.6
25.3
23.0
20.7
18.4
16.1
13.8
11.5
9.2
6.9
© SAE International.
4.6
2.3
0.0
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 8.6 Section clipping applied to the plot in Figure 8.5 shows that almost the entire
cross-section of the cantilever portion of the bracket has reached its yield stress. The bracket
is very close to collapse.
27.6
25.3
23.0
20.7
18.4
16.1
13.8
11.5
9.2
6.9
© SAE International.
4.6
2.3
0.0
The results based on the elastic-perfectly plastic material model show the maximum
stress equal to the yield strength of the material. Plasticized material now occupies almost
the entire cross-section of the cantilever portion of the bracket. Based on this finding
we conclude that the applied load is very close to the maximum load that the bracket can
take. A further increase in the load magnitude will cause plastic bending of the cantilever
portion of the bracket. This applies only to analysis with elastic-perfectly plastic material.
When a linear material model is used, an increase in the load magnitude produces larger
displacements, larger strain, and larger stress, all with no bounds.
8.2. U
se of Nonlinear Material to Control
Stress Singularity
The use of elastic-perfectly plastic material can eliminate stress singularities like those
accompanying sharp reentrant edges. Using elastic-perfectly plastic material places the
upper bound on the stress level; the highest stress equals the yield strength as defined in
the material model. It is important to point out that while the stress magnitude is capped
at the yield strength of the model material, strain diverges with mesh refinement.
Consider a bracket shown in Figure 8.7. If it is analyzed with a linear material model,
the maximum stress diverges with mesh refinement because of stress singularity caused
by the sharp reentrant edge. Stress results in Figure 8.8 show the maximum von Mises stress
of 556 MPa. This maximum stress coincides with the stress singularity and, therefore,
is meaningless.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 8.7 A bracket loaded with a uniformly distributed pressure and support applied
to the back face. A sharp reentrant edge is a part of the model geometry.
Fixed support
to back face
FIGURE 8.8 Von Mises stress plot produced using a linear material model. Stress results
along the sharp reentrant edge are meaningless because of the stress singularity.
556.2
509.9
463.5
417.2
370.8
324.5
278.1
231.8
185.4
139.1
92.7
© SAE International.
46.4
0.0
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 8.9 Section plots of von Mises stress results produced by the linear analysis and
the nonlinear analysis with elastic-perfectly plastic material. The same stress legend applies
to both plots. In the linear analysis, the upper range of the color legend is truncated at 206.8
MPa, and everything above that number shows as red. In the nonlinear analysis, there are no
locations where stress would be higher than 206.8 MPa.
206.8
206.8
189.6
189.6
172.3
172.3
155.1
155.1
137.9
137.9
120.6
120.6
103.4
103.4
86.2
86.2
68.9
68.9
51.7
51.7
34.5
© SAE International.
34.5
17.2
17.2
0.0
0.0
“Hiding” stress singularity by truncating the stress magnitude at the yield strength
leads to erroneous results, as shown in Figure 8.9 (left). The same model analyzed using the
elastic-perfectly plastic material does not show stress singularities. Mesh refinement does
not produce divergent stresses because the maximum stress is capped at 206.8 MPa, which
is the yield strength of the elastic-perfectly material as shown in Figure 8.9 (right).
Differences between incorrect results of the linear analysis and the correct results of
the nonlinear analysis with the elastic-perfectly plastic material (Figure 8.9) demonstrate
that stress singularities cannot be “hidden” by truncating the upper range of the color
legend. Also, the extent of yielding cannot be modeled by truncating the maximum stress
magnitude. A nonlinear material model is required for correct modeling of yielding.
8.3. O
ther Types of Nonlinearities
Geometric nonlinearities were discussed in Chapter 7, and material nonlinearities are
discussed in this chapter. Another important type of nonlinearity, nonlinear buckling, is
presented in Chapter 10.
For a better understanding of nonlinear buckling analysis, we need to review the modal
analysis presented in Chapter 9.
double the load (10000 N) and run it again. In both analyses the load will be applied in
increments as is always the case with nonlinear analysis. However, in the second analysis,
the load will not reach the magnitude of 10000 N because the solver will crash. The last
successfully performed step corresponds to the load that does not yet cause complete plas-
tification of the cantilever portion of the model. The next load increment completes the
plastification of the cross-section causing a plastic hinge which, in turn, causes the solution
to crash.
Objective
Conduct analysis with the elastic-perfectly plastic material model; study the structural
collapse caused by material yielding.
Procedure
8.4.2. L BRACKET02
Description
A bracket with stress singularity caused by the sharp reentrant edge is loaded in bending
(Figure 8.7). Very different stress results are produced using linear and nonlinear materials.
Having completed this exercise, eliminate the stress singularity by adding a fillet. Next,
run the analysis with elastic-perfectly plastic material in the model with fillet and compare
results to those produced using the model with a sharp reentrant edge.
Objective
Demonstrate the use of elastic-perfectly plastic material in a model with stress singularity.
Demonstrate differences between the results of linear analysis and nonlinear analysis used
to study the extent of yielding.
Procedure
•• Assign the material properties: elastic-perfectly plastic material, Steel with a yield
strength of 206.8 MPa.
•• Apply fixed restraint to the top face (Figure 8.7).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
9
Modal Analysis
9.1. D
ifferences between Modal and
Static Analysis
Modal analysis, also called frequency analysis, finds the modes of vibration of a structure.
The mode of vibration is defined by the frequency and the shape of vibration. Each mode
of vibration, with its frequency and the associated shape, corresponds to the situation where
stiffness forces cancel out with inertial forces. Mode of vibration describes the body vibrating
in the absence of excitation force and damping. Even though a real-life structure has an
infinite number of degrees of freedom it still has discrete modes of vibration.
We recall the fundamental FEM equation, applicable to static analysis:
K d F (9.1)
To consider dynamic effects, Equation (9.1) needs to be extended to account for inertial
and damping effects and for the fact that load is a function of time.
where
[M] is the mass matrix, which is known
[C] is the damping matrix, which is known
[K] is the stiffness matrix, which is known
[F] is the vector of nodal loads, which is known
[d] is the unknown vector of nodal displacements
Modal analysis deals with free and undamped vibrations, where F(t) = 0 (no excitation
force) and [C] = 0 (no damping). Therefore, Equation (9.1) is simplified to
Md K d 0 (9.3)
Equation (9.3) has n solutions, where n is the number of DOFs in the system, ωi2 is the
eigenvalue, and the corresponding vector {φ}i is the eigenvector. The relationship between
eigenvalue and frequency expressed in Hertz [Hz] is
fi i /2 (9.5)
FIGURE 9.1 Plot of resultant displacement in the first mode of vibration of a box. There
are no units on the color legend. The numerical values are normalized. Results may be used
only to find the ratio between displacements in different portions of the model. In this case
the ratio between the displacement in two selected locations is 1.25.
0.8 1.0
1.00
0.92
0.83
0.75
0.67
0.58
0.50
0.42
0.33
0.25
© SAE International.
0.17
0.08
0.00
9.3. M
odal Analysis with Rigid Body
Motions
Rigid Body Motions (RBMs) are allowed in modal analysis. Modal analysis can
be executed on a partially supported model or on a model with no supports at all. Rigid
Body Motion is displacement not associated with deformation; hence, the name “rigid.”
Modes corresponding to RBMs are called Rigid Body Modes and are assigned the
frequency of zero. An unsupported 3D model has six Rigid Body Modes with zero
frequency corresponding to six RBMs, which are translations along the three directions
and rotations about the three directions (Figure 9.2). The first mode of vibration that
is associated with deformation is mode number 7. The first two elastic modes of the
model airplane are shown in Figure 9.3.
An analysis of a partially supported model returns as many zero frequency modes as
many RBMs are found in the model. A ball joint assembly is shown in Figure 9.4. Assuming
that the base is fully restrained, the ball has three RBMs, which are rotations about the
three orthogonal axes. The base is included in the assembly only for show because a ball
support joint may be modeled by restraining the spherical surface of the ball in a spherical
coordinate system. This will be done in one of the hands-on exercises in this chapter.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 9.2 Model airplane with no supports has six RBMs: three translations and three
rotations. The model in the original position is shown in gray color. In this example, directions
of translations and rotations are aligned with the global coordinate system, but this is not
always the case.
y
Rigid Body Mode 1— Rigid Body Mode 2— Rigid Body Mode 3—
translation along x translation along y translation along z
z x
© SAE International.
Rigid Body Mode 4— Rigid Body Mode 5— Rigid Body Mode 6—
rotation about x rotation about y rotation about z
FIGURE 9.3 The first two elastic modes of vibration of an unsupported airplane.
© SAE International.
First elastic mode Second elastic mode
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 9.4 A ball joint has three RBMs; these are presented as Rigid Body Modes. The
undeformed shape (gray) is superimposed on the deformed part. The analysis is conducted
on one part and the ball joint socket (magenta) is for show only. All plots are shown in the
same position relative to the global coordinate system.
Y
© SAE International.
Z X
RBM 1 RBM 2 RBM 3
Rotation about x Rotation about y Rotation about z First Elastic Mode
FIGURE 9.5 The first elastic mode of vibration of unsupported (top) and hinge-supported
links (bottom). The undeformed shape is overlaid on the deformed shape. In the unsupported
link, the vibration takes place about the center of mass. In the hinge-supported link the center
of mass participates in vibration.
Hinge Hinge
© SAE International.
First elastic mode of the supported link
Frequency 564 Hz
9.5. A
pplications of Modal Analysis
9.5.1. F
inding Modal Frequencies and Associated
Shapes of Vibration
The frequency and the associated shape of vibration together define the mode of vibration.
The mode of vibration is the structural property of an object. Most often only the first few
modes of vibration are of interest. Knowing natural frequencies and the associated shapes
is important in preventing unwanted vibration.
To demonstrate a sample of results provided by modal analysis, we use a tuning fork
designed to vibrate with a frequency of 440 Hz. This mode is shown in Figure 9.6.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 9.6 A tuning fork vibrating with a frequency of 440 Hz. The undeformed shape is
superimposed on the deformed plot. The color legend is normalized to 1. The numbers can
be used only to find the ratio between displacements in different locations in the model.
Absolute values are meaningless.
1.00
0.92
0.83
0.75
0.67
0.58
0.50
0.42
0.33
0.25
© SAE International.
0.17
0.08
0.00
FIGURE 9.7 The first two modes of vibration of an unsupported box. Stress
concentrations highlight potential “weak spots” indicated by arrows. In this case the first two
modes highlight the same location. Von Mises stress is used in these plots.
© SAE International.
Mode 1 Mode 2
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
9.5.3. M
odal Analysis Provides Input to Vibration
Analysis
In addition to providing very important results on its own, modal analysis is a prerequisite
to vibration analysis based on the Modal Superposition Method. Vibration analysis based
on the Modal Superposition Method assumes that the vibration response of a model can
be represented by superposition of responses of several Single Degree of Freedom (SDOF)
systems corresponding to the modes of vibrations considered for analysis. This will
be discussed in Chapter 11.
FIGURE 9.8 The frequency of vibration in the first mode of a helicopter blade is strongly
influenced by tensile stresses produced by the centrifugal force. There are slight differences
between frequencies of the four blades caused by a discretization error associated with
meshing. Therefore, the first four modes show modal shapes of one blade at a time. The
analysis may be done using only one blade.
© SAE International.
FIGURE 9.9 Frequency of the first mode of vibration of the column is strongly influenced
by the applied compressive load producing compressive stresses.
© SAE International.
FIGURE 9.10 The first natural frequency of a column as a function of the magnitude of
compressive load. The graph shows that compressive load reduces the first natural
frequency. The magnitude of compressive load that reduces the first natural frequency to
zero (red marker) is the buckling load.
8.0
6.0
Frequency [Hz]
4.0
2.0
© SAE International.
0.0
0 5000 10000 15000 20000 25000
9.7. S
ymmetry and Anti-symmetry
Boundary Conditions in Modal
Analysis
Modes of vibration of a structure with mirror symmetry are either symmetric or anti-symmetric.
This makes it possible to conduct modal analysis on one-half of a symmetric model, but that has
to be done twice: first with mirror symmetry boundary conditions and then with anti-symmetry
boundary conditions. Mirror symmetry boundary conditions eliminate anti-symmetric modes,
and anti-symmetry boundary conditions eliminate symmetric modes. One-half of a symmetric
model can be used to extract all modes of vibration if the results of the modal analysis with
symmetry boundary condition are combined with the results of the modal analysis of the same
model with anti-symmetry boundary conditions; this is illustrated in Figure 9.11 and Table 9.1.
The overhead required to manage two sets of results is quite high and is worth the effort only in
the case of a large model that could not be solved without cutting it in half.
FIGURE 9.11 One-half of a model with a mirror symmetry model can be used to extract
all modes of vibration. Results of two modal analyses: with mirror symmetry boundary
conditions and with anti-symmetry boundary conditions must be combined.
9.8. C
onvergence of Modal Frequencies
In Chapter 4 we explained that during any convergence process, DOFs are added to the
model and, in effect, “artificial stiffness” resulting from approximations imposed by meshing
are reduced. That makes the FE model softer. The same mechanism works when the conver-
gence process is conducted in modal analysis. As a result of the model becoming softer,
modal frequencies converge from above. Therefore, the results of modal analysis overesti-
mate modal frequencies (Figure 9.12).
FIGURE 9.12 Frequency of the first mode of vibration of the bracket meshed with three
meshes. The frequency converges from above with mesh refinement. A default mesh is
usually sufficient for modal analysis.
1200
1000
Frequency [Hz]
800
Coarse mesh
600
Default mesh
400
Fine mesh
200
© SAE International.
0
1 2 3
Mode number
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
9.9. M
eshing Consideration for Modal
Analysis
As shown in Figure 9.12, the results of modal analysis are not strongly dependent on the
element size. Generally, modal analysis requires less refined mesh than structural stress
analysis of the same model. This is similar to displacement analysis requiring less refined
mesh than stress analysis. Small details, important for stress analysis, can be safely removed
from the model as long the analysis objective is to find the modes of vibration only. However,
make sure that defeaturing does not change the model stiffness or mass significantly.
If modal analysis needs to provide data for the subsequent vibration analysis, and the
objective of vibration analysis is to find stresses, then small details must be included in the
model just like in static stress analysis.
Objective
Study the results of modal analysis with and without supports.
Procedure
Comments
Observe that solution with no supports returns six modes with zero frequency corre-
sponding to six rigid body modes. The first non-zero frequency mode (or the first elastic
mode) is the fundamental frequency of the tuning fork: 440 Hz. The modal solution with
supports returns three modes with a frequency lower than 440 Hz, but those modes require
support to exist. Therefore, they are damped through the support, allowing the supported
tuning fork to vibrate at 440 Hz.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
9.10.2. BOX
Description
Review displacement results in the first elastic mode and probe displacements in the selected
locations to find the ratio between displacements in the selected locations (Figure 9.7).
Observe that numerical values of displacement are meaningless. If your program supports
this option, analyze stress results for each mode; locate “weak spots” corresponding to each
mode. Perform several mesh refinements and analyze the convergence of modal frequen-
cies, noting that they converge from above.
Objective
Demonstrate modal analysis used to locate “weak spots.”
Procedure
•• Apply the material properties of ABS Plastic (E = 2410 MPa, ν = 0.39, ρ = 1070 kg/m3);
notice that material density is required in modal analysis.
•• Specify at least 10 modes to be calculated.
•• Mesh and obtain the modal solutions with and without support.
Comments
Displacements and stress results are meaningless; still those results are assigned numbers
which are results of internal normalizing of results.
9.10.3. AIRPLANE
Description
An airplane model has no supports, it has six RBMs; there is no deformation associated
with these modes (Figure 9.2). The first elastic mode is mode number 7, this mode is associ-
ated with deformation of the model.
Objective
Review Rigid Body Modes and Elastic Modes of an unsupported model.
Procedure
•• Apply the material properties: Nylon 101 (E = 1000 MPa, ν = 0.3, ρ = 1150 kg/m3).
•• Do not define any restraints.
•• Specify at least 12 modes to be calculated.
•• Mesh and solve.
•• Animate six Rigid Body Modes and notice that they show three translations and
three rotations of the model as a rigid body. In this exercise translations and rotations
are aligned with the global coordinate system.
•• Animate six elastic modes and observe that modal shapes are either symmetric or
antisymmetric; this happens because the model geometry is symmetric.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Comments
The model with no restraints is used in this example, but symmetric and anti-symmetric
modes may be also observed in any restrained model if geometry and restraints are symmetric.
9.10.4. BALL
Description
A ball joint (a spherical kinematic pair) has three RBMs, these are three rotations: about
the x, y, and z axes (Figure 9.4). The modal analysis finds these RBMs and models them as
Rigid Body Modes frequency of 0 Hz. The first elastic mode is mode number 4. This exercise
is conducted using one part; the assembly shown in Figure 9.4 is not used.
Objective
Review Modes of a partially supported model with Rigid Body Modes associated with rotation.
Procedure
Comments
The first three modes are Rigid Body Modes corresponding to three rotations; the frequency
reported for these modes is 0 Hz. Mode number 4 and higher are elastic modes associated
with deformation of the model.
9.10.5. LINK03
Description
A link is supported by two hinges restraining radial and axial translations on the faces
of two holes (Figure 9.5). Next, the supports are removed leaving the link with six
RBMs. We need to find the first natural frequency of the fully supported and
unsupported link.
Objective
Demonstrate the effect of mass participation on the natural frequencies.
Procedure
•• Define two hinge supports by restraining the radial and the axial translation on both
cylindrical faces. Each hinge support must be defined in its own local cylindrical
coordinate system.
•• Obtain the modal solution with hinge supports.
•• Delete hinge supports.
•• Obtain the modal solution without supports; expect six Rigid Body Modes.
•• Compare the frequency and the shape of the first mode in the fully supported and
unsupported model.
Comments
Observe that corresponding modal frequencies are lower for the unsupported link because
the unsupported shaft vibrates about its center of mass and less mass participates in vibration.
9.10.6. H
ELICOPTER BLADE
Description
A simplified model of a helicopter rotor has four blades (Figure 9.8). The natural frequencies
of each blade differ slightly due to the discretization error.
Objective
Demonstrate the effect of pre-stress produced by the centrifugal forces on the natural first
frequency of the blade.
Procedure
Comments
Observe that the applied angular velocity generates a centrifugal load that produces tensile
stresses in the blades. Due to the stress stiffening effect, tensile stresses increase the blade
stiffness. Consequently, the natural frequencies of the spinning blade are higher than the
frequencies of the blade that is not moving. The pre-load may also change the shapes of
vibration. In this exercise this effect is visible in higher modes.
The modal analysis with pre-stress is executed in three steps. The first step is calculating
stresses in linear static analysis. The second step is using the stress solution to modify the
model stiffness. Finally, the modal analysis is run on the model with the modified stiffness
(Figure 9.13). These steps are transparent to the user.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 9.13 Frequencies of the first 16 modes of the helicopter rotor without and with
the effect of stress stiffening due to the centrifugal force. The series 1-4, 5-8, 9-12, and 13-16
correspond to different blades vibrating in the same mode. Minute differences in frequencies
within each series are caused by discretization errors. These differences are too small to show
in this graph.
120
100
80
Frequency [Hz]
60
0 RPM
180 RPM
40
20
© SAE International.
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Mode number
9.10.7. COLUMN
Description
The effect of compressive stresses on the natural frequency is illustrated here (Figure 9.9).
The analysis requires several runs with gradually increasing load magnitude until the solver
produces an error message. The exercise requires linear buckling analysis to find the
buckling force. Therefore, you may skip this exercise for now and return to it after completing
Chapter 10. You may also calculate the buckling force using Euler’s formula. Use k = 2 as
the end conditions factor used when the loaded end is free.
2 EI
Pcritical (9.6)
kL
2
Objective
Demonstrate the effect of compressive load on modal frequencies.
Procedure
•• Define the fixed support to the bottom end face of the channel.
•• Specify one or more buckling modes to be found, buckling analysis is needed to
provide the load magnitude for modal analysis with pre-load.
•• Mesh and solve to find the Buckling Load Factor (BLF) or find the buckling force
using Euler’s formula.
•• Run six modal analyses with pre-stress using compressive load with increasing
magnitude: 0%, 20%, 40%, 60%, 85%, 95% of the buckling load.
•• Summarize the results in a graph as shown in Figure 9.10.
Comments
The last modal analysis with pre-stress will report a low frequency because the pre-load is
close to the buckling load. To extend this exercise, try running the analysis again with 100%
of buckling load and observe that the solver produces an error message because the model
has zero stiffness in the direction of the first mode of vibration. Now run the analysis with
the pre-load equal to 99.9% of buckling load: the natural frequency will be very low; the
model has almost no stiffness and that stiffness comes mostly from the artificial stiffness
added to the model by discretization. You now have a tool to demonstrate the presence of
the artificial stiffness. Run the modal analysis again with a pre-load of 99.9% but with a
more refined mesh. The solver will crash because the artificial stiffness has decreased, and
the model has lost the stiffness in the direction of the first mode of vibration. Finally, run
the modal analysis once again with a pre-load of 99.9% but with a coarse mesh. Observe
an increase in the natural frequency. A strong dependence of the results on the meshing
choice makes these results meaningless.
9.10.8. B
RACKET03
Description
Symmetry in the BRACKET03 model includes the mirror symmetry of geometry and
restraints (Figure 9.11). Taking advantage of that symmetry, the modal analysis may
be completed on one-half of the model with symmetry and then with anti-symmetry
boundary conditions. Analysis of the model with the mirror symmetry boundary condi-
tions finds symmetric modes. Analysis of the model with anti-symmetry finds anti-
symmetric modes. This way modes with symmetric shapes and anti-symmetric shapes
are separated. Combining the results of two analyses provides the full set of modes
of vibration.
Objective
Perform modes separation using symmetry and anti-symmetry boundary conditions.
Procedure
•• Apply the material properties: ABS Plastic (E = 2000 MPa, ν = 0.394, ρ = 1020 kg/m3).
•• Apply a fixed support to the back face.
•• Run the modal analysis to find the six modes of vibration.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• If you use Parasolid geometry, cut the model in half; if you use SOLIDWORKS
geometry, switch to configuration 02 half.
•• Obtain the modal solution (six modes) for the half model with mirror symmetry
boundary conditions.
•• Obtain the modal solution (six modes) for the half model with anti-symmetry
boundary conditions.
•• Combine the solutions obtained with mirror symmetry and with anti-symmetry
boundary conditions as shown in Table 9.1.
Comments
Observe that combining solutions produced by the half model with mirror symmetry and
then with anti-symmetry boundary conditions produces the full set of results. Assemble
the symmetric and anti-symmetric solutions and sort results by the ascending order of
modal frequencies.
TABLE 9.1 Modal results from the full model, half model with mirror symmetry boundary
conditions, and half model with anti-symmetry boundary conditions.
Half-model Half-model
symmetric modes anti-symmetric
Mode number Mode description Full model [Hz] [Hz] modes [Hz]
1 Symmetric 412 412
2 Anti-symmetric 720 720
3 Anti-symmetric 1034 1034
4 Symmetric 1808 1814
5 Symmetric 2359 2365
6 Symmetric 2533 2538
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
10
Buckling Analysis
T
he phenomenon of buckling is best presented in the context of two other failure
modes: excessive displacements and yielding as summarized in Table 10.1. Buckling
is often overlooked as a mode of failure; therefore, understanding buckling analysis
with FEA is particularly important.
Buckling analysis is available in commercial FEA programs under the name Buckling
Analysis. It is also available as a part of Nonlinear Analysis. This fragmentized approach
may lead to a misunderstanding of what Buckling Analysis really is. The following discus-
sion should help understand the buckling analysis performed with FEA as well as the
buckling phenomenon itself.
10.1. L
inear Buckling Analysis
Linear buckling analysis is also referred to as Eigenvalue Buckling analysis or Euler’s
Buckling. Linear buckling analysis is in many aspects similar to modal analysis (Table 10.2).
Linear buckling analysis is easy to execute but has important limitations.
The linear buckling analysis calculates buckling loads and the associated buckling
shapes. Buckling takes place when, because of subtracting stress stiffness induced by
compressive load from elastic stiffness, the resultant structure stiffness drops to zero. It is
important to remember that the loss of stiffness in buckling happens in the direction normal
to the applied compressive load, it happens in the direction of the first mode of vibration.
This remains in close analogy to modal analysis with compressive pre-load where the stress
stiffness is subtracted from the elastic stiffness also producing zero resultant stiffness.
FEA programs offer a choice of how many buckling modes should be found.
Theoretically, as many buckling modes as the number of degrees of freedom in the FEA
model could be calculated but in practice only the first buckling mode with positive buckling
load factor is important.
Results of linear buckling analysis do not report the buckling load directly. Instead, the
results present the Buckling Load Factor (BLF). The BLF is a number by which the applied
load must be multiplied by (or divided—depending on the FEA program) to obtain the
magnitude of the buckling load. The buckling mode presents the shape that the structure
assumes when it buckles but says nothing about the numerical values of the displacements
or stresses. The numerical values can be displayed but are meaningless. This is in close analogy
to modal analysis which calculates the natural frequency and provides qualitative information
on the modes of vibration (modal shapes), but not on the actual magnitude of displacements.
If more than one buckling mode is requested, the buckling modes are reported in
ascending order according to their numerical values. A buckling mode with a negative BLF
TABLE 10.2 Analogies and terminology used in buckling and modal analysis.
means that the load direction must be reversed (in addition to multiplying by the BLF) for
the buckling to happen. The first positive BLF marks the mode of interest.
To illustrate the linear buckling analysis, we use a model of a column in compression; the
column material is 1060 Alloy. We study this model in two configurations: with a free end and
a sliding end. Before proceeding, notice a small notch and try to predict which way the column
will buckle in each configuration. The importance of the notch will soon become clear.
Figure 10.1 shows the results of the linear buckling analysis of a column with the free end.
The BLF is 566, therefore considering that 1 N load has been applied, the buckling happens at
566 N. The column has buckled toward the side opposite to the notch. This result is purely coin-
cidental; linear buckling analysis can predict the buckled shape, but not the direction of buckling.
Figure 10.2 shows the result of the analysis of the column with the sliding end. The BLF
is 9622; this is much higher than before because the sliding support makes the column
much stiffer. The direction of buckling shown in Figure 10.2 is purely coincidental. The color
legends with displacement magnitudes in Figures 10.1 and 10.2 are shown only to point out
their meaninglessness in terms of absolute numbers.
Linear buckling analysis cannot provide any quantitative results for displacements or
stresses. It does not provide an answer to what happens to the structure after buckling. Will
it collapse, or will it retain the load-bearing ability in the buckled shape? How much will
it deform when it buckles?
1.00
Column under compression 0.92
0.83
0.75
0.67
0.58
0.50
0.42
0.27 0.33
0.25
© SAE International.
0.17
0.08
Deformed shape in the first buckling mode 1.00 0.00
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
1.00
Column under compression 0.92
0.83
0.75
0.67
0.58
0.50
0.42
0.33
0.25
© SAE International.
0.17
0.08
Deformed shape in the first buckling mode 0.00
FEA software can find any number of buckling shapes and associated buckling loads.
However, usually only the first mode (load and the associated shape) needs to be found
because higher buckling modes have no chance of appearing. Buckling most often causes
catastrophic failure or renders the structure unusable even if the structure may still be able
to hold the load. The first four buckling modes of the column with free end, loaded with
1 N compressive load, are shown in Figure 10.3.
If we are looking for answers beyond the Buckling Load Factor and the buckled shape,
we need to use the nonlinear buckling analysis.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.3 The first four buckling modes of the notched column. The Buckling Load
Factor (BLF) relates to a compressive load of 1 N. Only mode 1 is of practical importance; the
other three modes have no chance to appear.
Mode 1
BLF = 566
Mode 2
BLF = 5094
Mode 3
BLF = 9020
© SAE International.
Mode 4
BLF = 1404
10.2. C
onvergence of Results in Linear
Buckling Analysis
Just like modal frequencies in modal analysis, buckling load factors converge from above.
Therefore, buckling analysis overestimates buckling load and because of discretization error
provides non-conservative results. However, the results of buckling analysis are also over-
estimated due to modeling simplifications. Finite element model represents a geometry
with no imperfections, loads and supports are applied with perfect accuracy with no offsets,
etc. In reality the load is always applied with an offset, walls are never perfectly flat, supports
are never rigid, etc. Considering the combined effect of both discretization error and
modeling error, the results of the buckling analysis must be used with caution.
10.3. N
onlinear Buckling Analysis
Buckling is inherently a nonlinear phenomenon. Compressive loads produce compressive
stresses which, in turn, generate negative stress stiffness. If compressive loads are high
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
enough, then the negative stress stiffness cancels with elastic stiffness making the resultant
stiffness zero in the direction of the first mode of vibration. While we may expect the column
under a compressive load to respond by displacement in the direction of load, the column
“slips out sideways from under the load.” This is what we call buckling.
Linear buckling analysis can only find combinations of load and deformed shape that
produce the cancellation of stiffness. These results, while useful, do not describe the buckling
fully. An in-depth analysis of buckling requires a nonlinear approach. As with any other
nonlinear analysis, the nonlinear buckling analysis requires that load be applied gradually
in multiple steps rather than in one step as it is done in a linear analysis. Each load incre-
ment changes the structure’s shape and stiffness.
The gradual load application, always necessary in a nonlinear analysis, can
be conducted in different ways. The process illustrated in Figure 10.4 is called the Load
Control method and is used in most types of nonlinear analyses, but not in the
buckling analysis.
When buckling happens, the structure experiences a momentary loss of stiffness, and
the load control method would result in numerical instabilities. For this reason, the
nonlinear buckling analysis requires another way of controlling the load application process
which is called the Arc-Length Control method, shown in Figure 10.5. Using the Arc-Length
Control method points corresponding to consecutive load increments are evenly spaced
out along the load-displacement curve.
The nonlinear buckling analysis does calculate displacements and stresses. To under-
stand the inner workings of a nonlinear buckling analysis, we must first consider what
happens if we run a nonlinear buckling analysis on an idealized structure. Imagine a
perfectly straight column under a perfectly aligned compressive load. Theoretically, buckling
will never happen, but in reality, the column will buckle because of imperfections always
FIGURE 10.4 Force Control Method. Load steps are defined either automatically or by the
user so that stiffness change between consecutive steps is not too large.
Displacement
© SAE International.
Load
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.5 Arc-Length Control method. The load is increased in such a way that
consecutive steps are evenly spaced out along the load-displacement curve.
Displacement
© SAE International.
Load
present in the geometry, loads, and supports. If imperfections are absent in the FEA model,
buckling will still happen, but it will be initiated by imperfections introduced by meshing.
Therefore, the nonlinear buckling requires a model with an intentional imperfection. If no
such imperfections exist, they must be added to control the onset of buckling. In our case
the imperfection is the notch shown in Figures 10.1 and 10.2. Another approach is to run
the analysis on a slightly deformed model in shape based on the scaled result of linear
buckling analysis.
Figure 10.6 shows the buckled shape and numerical values of displacement in the Y
direction of the loaded end. The compressive force acts in the negative x direction and does
not change its direction. The column buckles toward the notched side as it should. The
buckling takes place at about 560 N, the load magnitude is close to what the linear buckling
analysis predicted.
Once buckling takes place, the displacement grows at an almost constant force. The
graph stops at 40 mm, but the analysis could have been run further. That implies that the
column would be capable of supporting 550 N acting at an increasing offset, meaning that
it can withstand a large bending moment even though that cannot happen because of
material yielding.
Figure 10.6 shows the results of a nonlinear analysis where the only source of nonlinear
behavior is buckling. The material is linear elastic, and yielding is ignored. This is unrealistic;
experiments demonstrate that buckling initiates a chain of events shown in Figure 10.7.
To model a buckling event, we must account for yielding which is another source of
nonlinearity. We will do so by using the elastic-perfectly plastic material discussed in
Chapter 8. The deformed shape and the load-displacement curve produced by the model
with elastic-perfectly plastic material are shown in Figure 10.8.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.6 Y displacement component in the notched column with a free end. A linear
material model is used.
47.2
43.3
39.4
y 35.4
31.5
Notch
27.6
x
23.6
19.7
15.7
11.8
45
7.9
40
3.9
35 0.0
30
25
Uy [mm]
20
15
10
© SAE International.
0
0 100 200 300 400 500 600
Force [N]
The buckling still happens at 560 N indicating that the onset of buckling takes place
while the material is still in the elastic range. The steep climbing portion of the curve leading
to the maximum load corresponds to the elastic buckling. Once stresses reach the yield,
the load rapidly drops because, to maintain equilibrium, the applied force must be reduced.
If the force stays the same, as is most often the case, the end of the vertical portion of the
curve (red marker) indicates the structural collapse. At this point, the column is totally
plasticized and can no longer support the offset load. Examine the deformed shape and
notice that most of the deformation takes place at the support where the plastic hinge develops.
We will now conduct two analyses of the same column but with a sliding restraint on
the loaded end. The first analysis will use a linear material model, the second an elastic-
perfectly plastic material model. The linear material model (Figure 10.9) predicts buckling
at about 9500 N, which is close to the results of linear buckling analysis.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
BUCKLING
LARGE DISPLACEMENTS
YIELD
© SAE International.
STRUCTURAL COLLAPSE
However, a load this high causes yielding of the entire cross-section in the notched
area, and that cannot be modeled with a linear elastic material model. Using the elastic-
perfectly plastic model (Figure 10.10), we see that buckling takes place at 2500 N when the
notched cross-section is completely plastic.
Buckling does not have to lead to a structural collapse. To illustrate this, we review a
model of a thin, round, spherical steel plate loaded and supported as shown in Figure 10.11.
Linear buckling analysis predicts that buckling takes place at 0.8 MPa (Figure 10.12).
The linear buckling analysis is followed by two nonlinear analyses: first with a linear
material model and then with an elastic-perfectly plastic material model. Both solutions
use the Arc-Length Control of load. The load increments continue until the resultant
displacement exceeds 0.5 mm. The results of the linear analysis are shown in Figure 10.13.
The results of the nonlinear analysis are shown in Figure 10.14.
The load-displacement curves in both analyses are shown in Figure 10.15. Both curves
are identical well past buckling meaning that after buckling the plate is still in the elastic state.
The graph in Figure 10.15 shows a stiffening effect in the post-buckling stage meaning
that the plate regains stiffness after the buckling has taken place and while it remains in
the elastic stage. This is a “snap-through” phenomenon sometimes called an oil-canning effect.
Results of linear buckling analysis and nonlinear buckling analysis are compared in
Table 10.3.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.8 Y displacement component in the notched column with a free end. An elastic-
perfectly plastic material model is used.
40.2
36.8
33.5
30.1
y
26.8
Notch
23.4
x
20.1
16.7
13.4
45 10.0
6.7
40
3.3
35
0.0
30
25
Uy [mm]
20
15
10
© SAE International.
0
0 100 200 300 400 500 600
Force [N]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.9 Y displacement component in the notched column with the sliding end. A linear
material model is used.
0.0
–2.8
y
–5.5
–8.3
x
–11.1
–13.9
Notch –16.6
–19.4
–22.2
–25.0
0.0
–27.7
0 2000 4000 6000 8000 10000 12000
–30.5
–5.0
–33.3
–10.0
–15.0
Uy [mm]
–20.0
–25.0
–30.0
© SAE International.
–35.0
Force [N]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.10 Y displacement component in the notched column with a sliding end.
An elastic-perfectly plastic material model is used.
0.0
–0.4
y
–0.8
–1.2
x Notch
–1.6
–2.0
–2.4
–2.8
–3.2
0.0
0 500 1000 1500 2000 2500 3000 –3.7
–4.1
–1.0
–4.5
–4.9
–2.0
Uy [mm]
–3.0
–4.0
–5.0
© SAE International.
–6.0
Force [N]
FIGURE 10.11 A thin, round plate is subjected to pressure applied to the blue patch. Fixed
restraint is applied along the perimeter. The section view makes it easier to observe a slight
spherical curvature. The irregular shape of the patch introduces imperfections.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.12 Shape in the first buckling mode predicted by the linear buckling analysis.
The pressure causing buckling is 0.8 MPa. Displacements are normalized to 1.
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
© SAE International.
0.20
0.10
0.00
FIGURE 10.13 Results of nonlinear buckling analysis using a linear material model.
Section plots are used to show the curvature of the deformed plate. The maximum resultant
displacement is 0.55 mm. This value triggers the stop of the solution. The maximum von
Mises stress is 1032 MPa, it exceeds the yield strength of the material which is 620 MPa.
0.55
0.49
0.44
0.38
0.33
0.27
0.22
0.16
0.11
0.05
Resultant displacements [mm] 0.00
1032.8
931.8
830.8
729.8
628.8
527.8
426.8
325.8
© SAE International.
224.9
123.9
von Mises stress [MPa]
22.9
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
0.55
0.49
0.44
0.38
0.33
0.27
0.22
0.16
0.11
0.05
Resultant displacements [mm] 0.00
620.4
559.0
497.7
436.3
374.9
313.5
252.1
190.8
© SAE International.
129.4
68.0
von Mises stress [MPa]
6.6
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.15 Load-displacement curves of two models of plate: with linear material and
with elastic-perfectly plastic material. Buckling takes place when the pressure reaches 0.26
MPa, and then deformation progresses with almost no increase in pressure until displacement
reaches 0.27 mm. After that, the plate regains stiffness in the post-buckling shape. Both
curves are identical until pressure reaches 1.5 MPa. The sensor used to construct the load-
displacement curves in both models is in the center of the round plate.
0.6
0.5
0.4
Displacement [mm]
0.1
© SAE International.
0
0 0.5 1 1.5 2
Pressure [MPa]
10.4. C
ontrolling an Onset of Buckling in
Nonlinear Buckling Analysis
In the nonlinear buckling analysis, the onset of buckling must be initiated in a controlled
way. If it is not, the model will still buckle but buckling will happen because of imperfec-
tions introduced by mesh. In the notched column exercise we controlled the onset of
buckling with a notch. In the plate exercise, we applied pressure to an irregular patch. Yet
another approach might be using a slightly curved column shaped after a scaled buckling
shape found in the linear buckling analysis (Figure 10.16).
Nonlinear buckling analysis of a stand (Figure 10.17) uses the load applied with an offset;
this is a similar approach to the plate example. Buckling analysis of a curved sheet (Figure
10.18) does not require any purposely introduced imperfection. The onset of buckling is
controlled by the sheet curvature.
© SAE International.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.18 A curved sheet under a compressive load (magenta arrows) uniformly
distributed along the top edge. Translations of vertical edges (black arrows) in the direction
normal to Plane 1 are restrained. The bottom edge is fully supported; restraint symbols are
not shown.
Plane
1
© SAE International.
10.5. Summary
Linear buckling analysis can only provide limited information about buckling. Nonlinear
buckling analysis is often required, and one should always consider using the nonlinear
material model in nonlinear buckling analysis. What makes buckling counterintuitive is
that the onset of buckling is associated with a loss of stiffness in the direction normal to
load. If the analysis is continued past buckling, the structure can in some cases regain
stiffness in the post-buckling state. That momentary loss of stiffness requires Arc-Length
Control of load shown in Figure 10.5. The change of structure stiffness during buckling is
characterized by a load-displacement curve (Figure 10.19).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 10.19 Solution of a problem with membrane stress stiffening (Round Plate in
Chapter 7) produces a smooth load-displacement curve; at no point stiffness becomes zero.
Force Control solution method may be used (left). The solution to the buckling problem
produces a more complicated load-displacement curve (right). A momentary loss of stiffness
requires the Arc-Length control solution (right).
Displacement
Displacement
© SAE International.
Load Load
10.6.1. N
OTCHED COLUMN - Free End
Description
A compressive, non-following force is applied to a notched column (Figure 10.1). The force
produces buckling. The results strongly depend on the type of material model used: linear
elastic or elastic-perfectly plastic.
Objective
Perform three analyses: linear buckling, nonlinear buckling with a linear material model,
and nonlinear buckling analysis with an elastic-perfectly plastic material model.
Procedure
•• Define the linear material model: 1060 Alloy (E = 69000 MPa, ν = 0.33, σyield = 27.6
MPa).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• Apply a fixed support to the end face opposite to the loaded face.
•• Specify the four buckling modes to be found in linear buckling analysis.
•• Mesh and solve the linear model.
•• Define the displacement sensor at the loaded end.
•• Perform nonlinear buckling analysis; use the Arc-Length Control.
•• Construct a load-displacement curve.
•• Compare the results of linear buckling analysis and nonlinear buckling analysis.
•• Repeat Steps 1-6 using an elastic-perfectly plastic material model.
Comments
Try using different meshes and different solvers (if available) in linear buckling analysis;
you will see that the direction of buckling changes.
10.6.2. N
OTCHED COLUMN - Sliding End
Procedure
Apply restraints to make the loaded end slide in the direction of the load (Figure 10.2).
Follow the same steps as in Exercise 10.6.1. Notice that buckling in the nonlinear analysis
buckling happens when the notched cross-section is plasticized.
10.6.3. R
OUND PLATE02
Description
A round plate with a slight spherical curvature experiences a “snap-through” deformation
when subjected to pressure (Figure 10.11). Modeling this phenomenon requires a nonlinear
buckling analysis with Arc-Length Control.
Objective
Analyze the “snap-through” effect using nonlinear buckling analysis using linear and
nonlinear material models. Compare the results of two nonlinear buckling analyses and a
linear buckling analysis.
Procedure
•• Define the linear material properties: Alloy Steel (E = 210000 MPa, ν = 0.28, σyield
= 620 MPa).
•• Apply a pressure load to the patch on the top face.
•• Define the sensor in the center of the top face where the pressure load is applied.
•• Mesh and solve the linear buckling analysis.
•• Perform the nonlinear buckling analysis with a linear material model.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Comments
After the plate “snaps through,” it does not yield. It remains in the elastic state after buckling
has taken place.
10.6.4. C
URVED COLUMN
Objective
Repeat Exercises 10.6.1 using the curved column and nonlinear material. The onset of
buckling is now controlled by a curvature of the column rather than by a notch (Figure
10.16). Use compressive, non-following load perpendicular to the supported face rather than
load normal to the loaded face.
10.6.5. S
TAND
Description
A steel weldment is modeled as one part for the convenience of model saving (Figure 10.17).
Compressive load is applied to the yellow patch. Fixed support is applied to the bottom
face. The stand buckles sideways causing material yielding.
Objective
Perform a nonlinear buckling analysis with elastic-perfectly plastic material on a model
where the onset of buckling is controlled by an off-center load.
Procedure
•• Define an elastic-perfectly plastic material used in the previous exercises (Aluminum
or Steel).
•• Apply a fixed support to the bottom face of the base plate.
•• Apply a uniformly distributed, normal load over the yellow face.
•• Mesh and solve the linear buckling analysis.
•• Define the sensor in one of the top corners.
•• Solve nonlinear buckling analysis with the elastic-perfectly plastic material model.
•• Construct a load-displacement curve showing the displacement component in the
direction of buckling.
•• Compare the results of linear buckling analysis and nonlinear buckling analysis.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Comments
Stress singularities along sharp reentrant edges are eliminated by elastic-perfectly plastic
material. The maximum stress is limited to the yield strength of the material. Once the
stand begins to deform sideways and the moment arm increases, the load magnitude lowers
to keep the equilibrium.
10.6.6. C
URVED SHEET
Description
A slightly curved steel sheet has simple supports along the vertical edges and fixed support
along the bottom edge (Figure 10.18). Compressive load is applied to the top edge. Material
is elastic-perfectly plastic. The load causes buckling in the direction toward the concave
side of the curvature.
Objective
Perform a nonlinear buckling analysis on a model where the onset of buckling is controlled
by a curvature.
Procedure
•• Define the elastic-perfectly plastic material properties: Alloy Steel (E = 210000 MPa,
ν = 0.28, σyield = 620 MPa).
•• Define a shell thickness of 2 mm.
•• Define the restraints.
•• Apply a vertical load uniformly distributed over the top edge.
•• Mesh and run the linear buckling analysis.
•• Define the sensor in the middle of the loaded edge.
•• Perform a nonlinear buckling analysis with the elastic-perfectly plastic material.
•• Construct a load-displacement curve showing the displacement component in the
direction of buckling.
•• Compare the results of the linear buckling analysis and the nonlinear buckling
analysis with nonlinear material.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
11
Vibration Analysis
A
ll types of analyses that we discussed so far assumed that loads are static, and
inertial effects are ignored. We will now lift the restriction of static load to discuss
common types of vibration analysis, but first, we will clarify an important termi-
nology issue. Many commercial FEA programs use the term “Dynamic Analysis” to
describe what we call “Vibration Analysis” in this book. The term “Dynamic Analysis”
is more general than “Vibration Analysis.” It applies to the analysis of any motion, be it
motion of a rigid body, a deformable body, or an assembly composed of rigid and deform-
able bodies. “Dynamic Analysis” within the scope of FEA deals with the vibration of
deformable bodies about the position of equilibrium; hence, the term “Vibration Analysis”
is more precise.
Vibration Analysis is a very broad topic and cannot be given an in-depth treatment in
this introductory textbook. For this reason, we will discuss only two commonly performed
types of linear vibration analysis: Time Response and Frequency Response. Next, we will
introduce the basic concepts of nonlinear vibration analysis.
11.1. M
odal Superposition Method
The review of linear vibration analysis needs to be preceded by a discussion of the Modal
Superposition Method used in Time Response and Frequency Response analyses.
Each mode of vibration is characterized by the frequency and the shape of vibration.
The shape of vibration is related to the mass participating in the vibration. Furthermore,
the modal shape is associated with a certain stiffness. Collecting these facts, we may
represent an object vibrating in each mode of vibration by a linear Single Degree of Freedom
(SDOF) oscillator with mass, stiffness, and direction of oscillations. All data may be derived
from the properties of the mode of vibration under consideration. Next, we decide how
many modes are important in the vibration response of the analyzed system. For example,
if only three modes are important, the system can be represented by three SDOFs, each
one corresponding to one mode. This way the number of DOFs of the vibrating system is
reduced to just three. Instead of studying the original model with tens or hundreds of
thousands of DOFs, we may study a simple three DOFs system, where each mode is repre-
sented by one SDOF and the response of the system is found as the superposition of indi-
vidual responses of SDOFs. The concept of Modal Superposition Method is schematically
illustrated in Figure 11.1.
When vibration analysis is based on the Modal Superposition Method, the number of
SDOFs contributing to vibration response is equal to the number of modes calculated by
the prerequisite modal analysis. How many modes should be used to represent vibration
response using the Modal Superposition Method? The first few modes are always the most
important. The number of required modes depends on the range of the frequency of excita-
tion, type of analysis, code requirements to be met, and other factors. One should demon-
strate in a convergence process that increasing the number of modes past a certain number
no longer significantly affects the results.
The Modal Superposition Method simplifies vibration analysis, but the vibration
problem must be linear. The Modal Superposition Method cannot be used for nonlinear
problems. Other methods, like Direct Integration, are used in nonlinear problems.
FIGURE 11.1 A vibration system represented by three linear SDOF oscillators. Each
oscillator is characterized by its mass, stiffness, and direction of linear oscillations. It is
assumed that the three modes of vibration are sufficient to model the vibration response.
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
where
[M] is the mass matrix, which is known
[C] is the damping matrix, which is known
[K] is the stiffness matrix, which is known
[F] is the vector of nodal loads, which is known
[d] is the vector of nodal displacements, which is unknown
Time Response Analysis requires the definition of damping, most often expressed as
a damping coefficient that specifies the percentage of the critical damping. Damping must
be defined for all modes used in the Modal Superposition Method. Refer to (1) in Chapter
15 for values of damping coefficients. Load excitation or base excitation may be used.
A typical example of Time Response Analysis is a response to an impact load of short
duration. We illustrate this with a hammer model shown in Figure 11.2. The hammer head
is subjected to an impact force with time history, as shown in Figure 11.3. Vibration time
response in the form of head displacement as a function of time is shown in Figure 11.4.
The second example presents Time Response Analysis with excitation of longer
duration. The same model of the hammer is subjected to a harmonic load shown in Figure
11.5; there is no damping in the system. The frequency of the excitation is 19 Hz, this is close
to the first natural frequency of 21 Hz of the hammer. The displacement response of the
hammer head is shown in Figure 11.6. It demonstrates the effect of beating, showing periodic
amplification and cancellation of displacement amplitude.
Fixed restraint
10000 N
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 11.3 Load time history: the load follows a sine curve; after 0.05 s it reaches the
magnitude of 10000N and disappears after 0.1 s.
10000
8000
6000
Force [N]
4000
2000
© SAE International.
0
0 0.02 0.04 0.06 0.08 0.1
Time [s]
FIGURE 11.4 Displacement time history of the hammer head subjected to the impact
load. When the load disappears after 0.1 s, the hammer starts a free damped vibration.
0.6
0.5
0.4
0.3
Displacement [mm]
0.2
0.1
0.0
0 0.25 0.5 0.75 1
–0.1
–0.2
© SAE International.
–0.3
Time [s]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 11.5 Time history of load excitation is a harmonic function with a frequency of 19
Hz; the maximum load magnitude is 10000 N.
10000
8000
6000
4000
2000
Force [N]
0
0 0.2 0.4 0.6 0.8 1
–2000
–4000
–6000
© SAE International.
–8000
–10000
Time [s]
FIGURE 11.6 Displacement time history of the hammer head subjected to a harmonic
load excitation. There is no damping in the system. Periodic amplification and cancellation of
displacement magnitude is called “beating.”
2
Displacement [mm]
0
0 0.2 0.4 0.6 0.8 1
–2
–4
© SAE International.
–6
Time [s]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 11.7 The elbow pipe is installed on a shaker table. The table performs harmonic
oscillations relative to the base. The frequency of oscillations changes slowly from 0 Hz to
800 Hz.
Location of sensor
Shaker table
Base
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 11.8 Four modes of vibration found in the frequency range 0-800 Hz.
FIGURE 11.9 Result of Frequency Response Analysis; mode 1 and mode 3 are excited by
the oscillating base.
40
Mode 1: 123 Hz
30
Amplitude of displacement [mm]
20
10
Mode 3: 428 Hz
© SAE International.
0
0 200 400 600 800
Excitation frequency [Hz]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
As shown in Figure 11.9, only two modes are excited by the vertical oscillations of the
table: mode 1 and mode 3. Mode 2 has the shape orthogonal to the direction of excitation;
mode 4 is a torsional mode; therefore, mode 2 and mode 4 are not excited by the vertical
oscillation of the shaker table.
When the excitation frequency passes through the natural frequency, the amplitude of
vibration reaches the local maximum. Vibration at the frequency of excitation equal to the natural
frequency is called resonance. The amplitude of vibration is then controlled only by damping.
The second example illustrates Frequency Response Analysis with a variable magnitude
of excitation. This is an analysis of a centrifuge where the magnitude of excitation force is
a square function of the excitation frequency. Such an excitation is called Omega Square
Excitation; it is a common type of excitation in rotating machinery.
Figure 11.10 shows a schematic model of an industrial centrifuge. The magnitude of the
out-of-balance centrifugal load is proportional to ω2, where ω is the angular velocity of the
drum in radians per second (rad/s). During the operation, the angular velocity changes in
the range 0–200 rad/s (0–35 Hz) and the force amplitude changes in the range 0–25000 N
(Figure 11.11). The Frequency Response Analysis is used to find the amplitude of displacement
of the centrifuge body as a function of the angular velocity of the drum.
The displacement response graph reveals one resonant frequency of 9.6 Hz corre-
sponding to the only mode of vibration in the excitation range of 0–35 Hz (Figure 11.12).
FIGURE 11.10 A centrifuge is subjected to a vertical oscillating load; the load magnitude is
proportional to ω2, where ω is the excitation frequency in rad/s. Guiderails, not shown in this
illustration, allow only for vertical movement of the centrifuge body. The body is supported
by four rubber legs. The body is very stiff compared to the legs. For simplicity, the excitation
force is applied to the top face (yellow) of the centrifuge body.
F = mew2 sin(wt)
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 11.11 The magnitude of the excitation force is a square function of the excitation
frequency. The excitation frequency is the angular velocity of the drum. The curve equation is
F = meω2 where m = 5 kg is the out-of-balance mass; e = 0.1 m is the eccentricity and ω is the
angular velocity in rad/s.
25000
20000
Force amplitude [N]
15000
10000
5000
© SAE International.
0
0 5 10 15 20 25 30 35
Excitation frequency [Hz]
FIGURE 11.12 The magnitude of displacement of the centrifuge drum due to ω2 force
excitation in the range of the operating frequencies of 0-35 Hz.
9 Mode 1: 9.6 Hz
8
Displacement amplitude [mm]
1
© SAE International.
0
0 5 10 15 20 25 30 35
Excitation frequency [Hz]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
11.4. N
onlinear Vibration Analysis
Previously discussed Time Response and Frequency Response analyses belong to the class
of linear vibration analysis where the structure stiffness does not change with time (Time
Response) or with the frequency of excitation (Frequency Response). Being linear, these
analyses were based on the Modal Superposition Method leading to a very significant
reduction in the numerical effort required to obtain a solution. Most vibration problems
are linear, this is because displacements are small, and the material remains in the
linear range.
Analysis of nonlinear vibration problems where stiffness does change with time faces
two major challenges and both originate in the inability to use the Modal Superposition
Method. First, the equations of motion must be based on all DOFs in the model. Another
consequence of the inability to use the Modal Superposition Method is that modal damping
cannot be used. Damping in nonlinear problems is defined as Rayleigh damping. Rayleigh
damping makes an arbitrary assumption that the damping matrix is a linear combination
of the mass and stiffness matrices. This assumption is a mathematical convenience for the
purpose of simplification since there is no physical justification for this. Rayleigh damping
is specified by two damping constants: α and β, which are used as multipliers of the mass
matrix M and the stiffness matrix K when calculating damping matrix C:
C M K (11.3)
Coefficients α and β are related to the frequency of excitation ω and the modal damping
coefficient ζ as shown in Equation 11.4.
(11.4)
2 2
FIGURE 11.13 Beam held between two hinges and subjected to pressure is a classic
example of a nonlinear problem where nonlinearity is caused by the stress stiffening effect.
© SAE International.
FIGURE 11.14 Time history of pressure applied to the beam. Pressure follows a sine curve
reaching 0.5 MPa and then drops down to zero after 0.01 s.
0.6
0.5
0.4
Pressure [MPa]
0.3
0.2
0.1
© SAE International.
0
0.000 0.005 0.010 0.015 0.020
Time [s]
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 11.15 Displacement response in the mid-span of the beam in the correct
nonlinear solutions and an incorrect linear solution.
40
30
20
Displacement [mm]
10
0
0 0.04 0.08 0.12 0.16 0.2
–10
–20
–30
–40
© SAE International.
Time [s]
Objective
Perform a Time Response Analysis and construct a displacement time-history graph of the
free end.
Procedure
Comments
Definition of Time Response Analysis includes specifying the time step; make sure it is
short enough to model response due to all modes in the 0–800 Hz range. Modal damping
is usually higher for higher modes. Therefore, applying the same modal damping to all
modes is a conservative approach resulting in a higher amplitude of displacement.
Objective
Perform a Time Response Analysis of a model subjected to a harmonic load with the
frequency of excitation close to the first natural frequency. This demonstrates the phenom-
enon of beating.
Procedure
•• Define the material properties and restraints the same as in Exercise 11.5.1.
•• Mesh and run to find the first mode of vibration; verify that the frequency is 21 Hz.
•• Apply a normal load uniformly distributed over the top face of the hammer head
(Figure 11.2) with time history shown in Figure 11.5; the frequency of excitation is 19
Hz, and the amplitude is 10000 N.
•• Specify only one mode (the first mode of vibration) to be used in the Modal
Superposition Method.
•• Define a 0% modal damping (no damping).
•• Specify a small time step; 1/50 of the vibration period in the first mode.
•• Run a Time Response Analysis for 1 s.
•• Construct a displacement time history in the sensor location for the duration of
analysis (Figure 11.6).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Comments
Run the analysis again using excitation with a different frequency. Observe that when the
excitation frequency approaches the natural frequency, the beating period increases.
Objective
Perform a Frequency Response Analysis of a model subjected to a harmonic base excitation
with frequency in the range 0–800 Hz.
Procedure
Comments
The analysis uses part ELBOW PIPE, not the assembly model shown in Figure 11.7.
11.5.4. CENTRIFUGE
Description
The exercise simulates excitation caused by an out-of-balance load in the rotating drum.
The magnitude of excitation is proportional to ω2, where ω is the angular velocity in rad/s
(Figure 11.10).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Objective
Perform a Frequency Response Analysis of a model subjected to ω2 square force excitation.
Procedure
•• Apply the material properties of Plain Carbon Steel to the centrifuge body.
•• Apply the material properties of Rubber to the four legs (E = 6.1 MPa, ν = 0.49,
ρ = 1000 kg/m3).
•• Define the sensor anywhere on the centrifuge body.
•• Define the fixed restraints to the bottom faces of the four rubber legs.
•• Define the slider restraints to the side faces of the centrifuge body to limit the
centrifuge movement to oscillations in the vertical direction.
•• Specify one mode to be used in the Modal Superposition Method.
•• Apply load to the centrifuge body in the vertical direction; the load is a function of
angular velocity as shown in Figure 11.11. You may use the table in the spreadsheet
CENTRIFUGE.xlsx.
•• Define a 5% modal damping.
•• Mesh and run a Frequency Response Analysis in the range 0–35 Hz.
•• Construct a displacement response graph as shown in Figure 11.12.
Comments
The rubber legs are small in comparison to the centrifuge body. Define mesh controls on
the legs to make sure that the elements are correctly shaped. Highly distorted elements will
model rubber leg stiffness incorrectly. The weight of centrifuge would deform the rubber
legs, but the effect of gravity is ignored in this exercise.
11.5.5. PLANK
Description
The plank is supported by two hinges, the distance between hinges does not change there-
fore, the beam develops membrane stresses while it deforms under the load (Figure 11.13).
Plank stiffness changes with deformation requiring a nonlinear geometry analysis.
Objective
Perform a nonlinear Time Response Analysis of a model subjected to an impact load.
Procedure
Comments
The analysis uses part PLANK, not the assembly shown in Figure 11.13. Use a time step of
0.001 s, both in linear and nonlinear analyses.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
12
Thermal Analysis
T
hermal analysis with FEA deals with heat transfer in solid bodies. The primary
unknown in thermal analysis is temperature, which is a scalar entity. Therefore, only
one degree of freedom needs to be assigned to nodes, be it 3D, 2D, or 1D analysis.
This makes thermal analysis much simpler than structural analysis in terms of computa-
tional effort.
An important conceptual difference between structural and thermal analysis is that
while structural static analysis deals with the state of equilibrium under the applied load,
an analogous steady state thermal analysis does not describe a state of equilibrium. Instead,
it models a steady state condition where heat continues to flow, but temperatures do not
change in time. Therefore, the thermal analogy of a structural static analysis is a steady
state thermal analysis, and an analogy of a vibration (dynamic) structural analysis is a
transient thermal analysis. The temperature in thermal analysis is analogous to displace-
ment in structural analysis; selected analogies are shown in Table 12.1.
In every heat transfer problem, the mechanism of heat flow must be fully defined to
produce a unique temperature solution. An attempt to run thermal analysis without fully
defined heat flow conditions results in an error analogous to the absence of supports in
structural analysis.
12.1. H
eat Flow by Conduction
Consider a model with different temperatures defined on two faces as shown in
Figure 12.1 .
Just like stress may be induced by prescribed displacement, heat flow may be induced
by a prescribed temperature. Heat flow in the bracket (Figure 12.1) is induced by the tempera-
ture gradient. No heat is exchanged through faces other than the two faces with prescribed
temperatures. The model is insulated, except for the two faces where prescribed temperatures
Temp (Kelvin)
393K
393.0
384.7
376.3
368.0
359.7
351.3
343.0
334.7
326.3
318.0
309.7
© SAE International.
301.3
293.0
293K
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
are defined. The only mechanism of heat transfer in the problem illustrated in Figure 12.1
is conduction in the solid body.
12.2. H
eat Flow by Convection
Convection is modeled in FEA as a boundary condition. Convection boundary conditions
require the definition of the convection coefficient [W/m 2/K], and the temperature of the
surrounding fluid called the ambient temperature [K]. Heat transfer by convection takes
place between the face of the solid and the fluid in contact with the face where convection
boundary conditions are defined. Heat transfer by convection requires a moving fluid, it
cannot happen in a vacuum or when fluid is not moving. Fluid motion may be induced
by changes in the specific gravity of hot and cold fluid, this is called natural convection
and requires gravity. It may be also induced by the forced motion of fluid (e.g., a cooling
fan or a pump), this is called forced convection. Convection boundary conditions must
be defined on all faces participating in heat exchange between the solid and the
surrounding fluid.
Figure 12.2 presents a problem where heat is generated in a volume of a solid, this is
called heat power. Heat power may be defined as total heat power [W] or as heat power per
volume [W/m3]. Heat is removed from the model to the surrounding air by convection.
The steady state temperature in the microchip assembly model is shown in Figure 12.3.
Since this is a steady state thermal analysis, we do not know how long it has taken to reach
the steady state.
Figure 12.4 shows a problem that can be solved using a 2D representation, assuming
there is no temperature gradient along the length of the channel.
Temperature and heat flux in the channel are shown in Figure 12.5.
FIGURE 12.2 The bottom face of the microchip (blue) and the bottom face of the radiator
(green) are insulated. Heat generated in the microchip travels by conduction to the radiator,
and then it is dissipated by convection to the surrounding air. It is assumed that heat crosses
the border between the microchip and the radiator without encountering any resistance. Heat
transfer by conduction and by convection are present in this problem.
Radiator
Microchip
© SAE International.
FIGURE 12.3 Temperature in the assembly shows a hot spot in the microchip. The
temperature in the radiator shows a very low temperature gradient of 0.7 K.
Temp.K Temp.K
320.9 305.5
319.3 305.4
317.7 305.3
316.1 305.3
314.4 305.2
312.8 305.1
311.2 305.1
309.6 305.0
308.0 304.9
© SAE International.
306.4 304.9
304.8 304.8
Temperature in assembly Temperature in radiator
FIGURE 12.4 A long channel is heated at the bottom face. This is where the heat flux is
defined. Heat enters the model through the bottom face, travels through the solid body by
conduction, and escapes to the environment through faces where convection boundary
conditions are defined. Assuming that end faces on both sides are insulated, there is no
temperature gradient along the length of the channel. Therefore, the problem may be solved
using 2D representation.
© SAE International.
2D slice
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 12.5 Heat flux plot using vector display. Arrows “coming out” of the walls
visualize heat escaping through the walls. This is enabled by the convection
boundary conditions.
369 7467
369 5432
368 3398
Temperature Heat flux
12.3. H
eat Transfer by Radiation
Heat transfer by radiation is modeled in FEA as a boundary condition. Figure 12.6 shows a space
heater where three mechanisms of heat transfer are present: conduction, convection, and radiation.
Heat power is generated in the volume of the bulb. The heat is dissipated by radiation
to the ambient space, and some of it reaches the concave face of the reflector. A small portion
of the heat enters the reflector through the face touching the reflector. The emissivity of the
FIGURE 12.6 The outside faces of the bulb (red) and the concave side of the reflector
(magenta) are in a vacuum; they radiate heat out to the ambient space and exchange some
heat by radiation between themselves. The back face of the reflector and the rim (gray) are
exposed to air and dissipate heat by convection. We assume no radiation on the air side.
© SAE International.
FIGURE 12.7 A section view of the steady state temperature in the space heater. Notice
the high temperature gradient in the bulb and low temperature gradient in the reflector.
Temp. (K)
885.6
837.5
789.4
741.3
693.2
645.1
597.0
548.9
500.8
452.7
404.6
© SAE International.
356.5
308.4
concave face of the reflector is low, but not zero; therefore, some heat, rather than being
reflected, enters the body of the radiator. Then it travels across the solid body and escapes
by convection to air on the convex side of the reflector. Heat exchange between two faces
with a radiation boundary condition depends on how those faces “see each other” and is
described by view factors.
To summarize, only heat transfer by conduction is modeled directly, radiation and
convection are modeled as boundary conditions. A steady state temperature plot in the
space heater is shown in Figure 12.7.
12.4. M
odeling Considerations in Thermal
Analysis
Mirror symmetry boundary conditions can be used in thermal analysis with convection,
based on the observation that if symmetry exists in both geometry and boundary condi-
tions, then there is no heat flowing through the plane of symmetry. Therefore, after cutting
the model in half (Figure 12.8), nothing needs to be done to the faces created by the cuts.
No convection boundary conditions defined on the faces created by the cut automatically
enforce the symmetry of the heat flow.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 12.8 The heat sink and microchip assembly may be cut in half. Do not define any
convection boundary conditions on the faces created by the cut, and the symmetry of heat
flow will be conserved.
© SAE International.
FIGURE 12.9 Model without fillets is not suitable for analysis of heat flux because sharp
reentrant edges produce heat flux singularity. Analysis of heat flux requires a detailed
modeling of geometry to include fillets even if they are very small.
© SAE International.
The model shown in Figure 12.9 (left) has sharp reentrant edges. This model is suitable
for analysis of temperature but, because of sharp reentrant edges, it is not suitable for
analysis of heat flux in the vicinity of the sharp reentrant edges because of heat flux
singularity. This is in direct analogy to sharp reentrant edges causing stress singularities
in structural models.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
12.5. C
hallenges in Thermal Analysis
Analogies between structural and thermal analyses make thermal analysis easy to learn;
one DOF per node makes thermal problems fast to solve. Still, it may be difficult to produce
meaningful results.
Thermal analysis with FEA is capable of modeling heat transfer in solid bodies only.
However, in the presence of fluids, heat transfer in solid bodies is accompanied by heat
exchange between the faces of solid bodies and the surrounding fluid. Modeling convective
heat transfer is a major problem because FEA cannot model fluid flow around a solid model,
and convection must be modeled as a boundary condition. The only difference between
natural and forced convection or between, for example, air and water surrounding a solid
body is a different value of convection coefficient and ambient temperature.
Meaningful results of thermal analysis with FEA require a correct definition of convec-
tion boundary conditions, but that is often difficult to find. An alternative to defining
convection as a boundary condition is solving a conjugate heat transfer problem that
includes fluid flow around the solid body. The conjugate heat transfer problem finds convec-
tion boundary conditions using flow simulation and transfers them to FEA. Many CAE
programs offer both FEA and Computational Fluid Dynamic (CFD) modules. CFD and
FEA may be interfaced as shown in Figure 12.10.
On a side note, we should point out an ambiguity in the common use of the terms CFD
and FEA. CFD refers to the field of application: the analysis of fluid motion. FEA refers to
a numerical method. Computational Fluid Dynamics problems are usually solved using
the Finite Volume Method. Heat flow in solids problems are usually solved with Finite
Element Method.
CFD Convection
Boundary
FEA
Conditions
© SAE International.
Finite Volume Finite Element
Method Method
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Objective
Demonstrate thermal analysis with prescribed temperatures where conduction is the only
mechanism of heat transfer.
Procedure
•• Define any material properties.
•• Define prescribed temperatures as shown in Figure 12.1.
•• Mesh and solve.
•• Repeat the analysis for different material properties.
•• Analyze the temperature and heat flux results.
Comments
Notice that no heat escapes or enters the model through the walls other than the faces with
prescribed temperatures because no convection boundary conditions are defined.
12.6.2. H
EAT SINK
Description
A ceramic microchip produces heat power in its volume (Figure 12.2). The heat travels by
conduction to the aluminum radiator, and then it escapes to the surrounding air
by convection.
Objective
Solve a thermal analysis problem with a heat source, conduction, and convection.
Procedure
•• Assign the material properties of aluminum to the radiator and ceramic porcelain
to the microchip. The conductivity of aluminum is 200 W/m/K and conductivity
of ceramic porcelain is 1.5 W/m/K.
•• Define the heat power of 10 W in the volume of the microchip.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• Define the convection boundary conditions on all outside faces of the radiator
except for the bottom face; convection coefficient at 50 W/m2/K and ambient (bulk)
temperature at 293 K.
•• Mesh, avoiding high turn angles in elements meshing round faces, and solve.
•• Analyze the temperature and heat flux results; use fringe plots for temperature and
vector plots for heat flux.
Comments
Repeat the exercise after deleting all convection boundary conditions. Observe that that
solution fails because heat cannot be removed from the model.
12.6.3. CHANNEL
Description
A long channel with no temperature gradient along the length may be reduced to a 2D
representation in close analogy to a 2D plane strain problem (Figure 12.4).
Objective
Demonstrate a 2D thermal analysis with heat power, conduction, and convection.
Procedure
Comments
This exercise may also be completed using a 3D model without 2D representation. Section
plots enable an analysis of heat flow inside the solid body. Section plots are frequently used
in thermal analysis. Vector plots are useful in visualizing heat flux results because heat flux
is a vector. Vector plots are frequently used in thermal analysis.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
12.6.4. S
PACE HEATER
Description
Heat power is produced in the volume of the hollow bulb (red) (Figure 12.6). The heat escapes
to the ambient space, some of it directly, and some first bounce off the face of the reflector.
Yet another portion of the heat enters the reflector body on the concave side and exits by
convection to the air on the convex side. Some heat travels directly to reflector through the
base of the bulb.
Objective
Demonstrate a heat transfer problem with conduction, radiation, and convection.
Procedure
•• Assign the material properties of aluminum to the reflector and glass to the bulb.
Thermal conductivity of aluminum at 200 W/m/K; thermal conductivity of glass
at 0.75 W/m/K.
•• Define the heat power 300 W to the volume of the bulb.
•• Define the convection boundary conditions on the back face of the reflector and on
the rim: convection coefficient at 50 W/m2/K and ambient temperature at 293 K.
•• Define the radiation boundary conditions on the outside faces of the bulb: emissivity
of 0.7, ambient temperature at 293 K.
•• Define the radiation boundary condition on the front (concave) face of the reflector:
emissivity of 0.1, ambient temperature at 293 K.
•• Mesh and solve.
•• Analyze the temperature distribution and heat flux results; use section plots for
temperature and vector plots for heat flux.
Comments
Emissivity compares the radiating surface to a black body whose emissivity is 1 and to a
perfectly reflective body whose emissivity is 0. Ambient temperature in a radiation boundary
condition is also called the temperature of a distant enclosure of a far-field temperature.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
13
Implementation of Finite
Element Analysis in
the Design Process
I
n this chapter we focus on the implementation issues: how to realize the full benefits of
FEA used as a design tool during product development process.
13.1. D
ifferences between CAD and FEA
Geometry
The differences between CAD and FEA geometries are important issues encountered by
design engineers using FEA as a design tool. CAD geometry is a fully detailed geometry
of a part or an assembly; it contains all information necessary for manufacturing. Why is
FEA geometry different, why cannot we use manufacturing CAD geometry “as is” for FEA?
The reason is that geometry must be meshed prior to analysis. Further, the mesh must
be able to model the data of interest properly and must be of reasonable complexity suitable
for the solver. The requirements are summarized below:
•• CAD geometry: Must contain information necessary for manufacturing.
•• FEA geometry: Must be meshable; must allow for the creation of a mesh that will
correctly model the data of interest; must allow for the creation of a mesh solvable
within a reasonable time.
13.1.1. D
efeaturing
CAD geometry contains features necessary for manufacturing. Many of those features are
unimportant for analysis and should be removed prior to meshing (Figure 13.1). Leaving
such features results in an unnecessarily complicated mesh and in long solution times.
Worse, it may prevent the mesher from completing the mesh.
Decision on which features can be removed and which should be included in the finite
element model requires careful engineering judgment. A small size of a feature compared
to the overall size of the model does not automatically justify removing it. For example,
small fillets should be retained if the analysis objective is to find stresses where the fillets
are located. Defeaturing would create stress singularity (Figure 13.2). However, a fillet can
be removed if the objective of analysis are, for example, displacements or the stress in
another part of the structure far from the edges.
Defeaturing is done on solid CAD geometry. After defeaturing, the geometry remains
a solid geometry, eventually leading to a 3D solid element model. Defeaturing does not
change the type of geometry.
Remove
© SAE International.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 13.2 Fillets (rounds) along internal edges are highlighted in red. The fillets cannot
be removed, no matter how small, if the analysis objective is to find stress along these edges.
© SAE International.
13.1.2. Idealization
Idealization modifies CAD geometry more substantially than defeaturing. For example,
idealization may require converting 3D solid geometry into a surface geometry suitable for
meshing with shell elements (Figure 13.3) or into a wireframe geometry suitable for meshing
with beam elements (Figure 13.4). At times idealization leads to an analysis of reduced
dimensionality, as in the case of the vase in Figure 13.5. In all these cases geometry is not just
simplified to facilitate meshing. It is converted into an abstract geometry suitable exclusively
for analysis. Often, solid CAD geometry cannot be converted into an idealized FEA geometry,
and the idealized geometry must be constructed solely for the purpose of FEA.
FIGURE 13.3 A solid model of a stamped steel pulley is converted into a surface
geometry for meshing with shell elements.
© SAE International.
FIGURE 13.4 A solid model of a cage (left) is converted into a wireframe geometry for
meshing with beam elements.
© SAE International.
Solid geometry Wireframe geometry
FIGURE 13.5 A solid model of a vase is used to prepare a 2D slice and mesh it with 2D
axisymmetric elements.
© SAE International.
13.1.3. Cleanup
An ambivalent term “geometry cleanup” refers to geometry quality issues that must be dealt
with to enable meshing. Sometimes, geometry that is satisfactory for manufacturing
purposes contains features that will not mesh or will force the mesher to create an excessively
large number of elements. Examples are very short edges or faces, as illustrated in Figures
13.6 –13.8. Those minuscule features have no importance for analysis and must be removed.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 13.6 A sliver face created by a geometry modeling error. It must be removed to
enable meshing.
© SAE International.
FIGURE 13.7 This geometry requires both defeaturing by removing the chamfer (green)
and cleanup, as shown in Figure 13.8.
© SAE International.
Mesh creation may also fail because of geometry quality issues like coincident entities,
floating solids, and other problems that can be detected by CAD quality control tools.
Meshing places high-quality requirements on CAD geometry, and meshing can double
as a CAD geometry quality inspection. Therefore, mesh creation may be thought of as a
geometry quality test worthwhile performing even if no FEA is intended. Usually, geometry
cleanup intended to facilitate meshing also results in a better manufacturing geometry.
Geometry cleanup may be combined with defeaturing or may precede geometry idealization.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 13.8 The minuscule face (red) should be removed prior to meshing.
© SAE International.
13.2. Common Meshing Problems
The tasks described previously: defeaturing, idealization, and cleanup, have one common
goal: to produce geometry that captures all important features of the analyzed structure
and, at the same time, is meshable into a correct, and preferably simple, finite
element mesh.
Even though meshing is an automated process, it is not a hands-off process. User’s
input is still required to control the element type and order, element size, and mesh bias.
Creating a mesh can be looked at as a process of filling up a volume or surface with
primitives of certain shape. Solid elements can be tetrahedrons, wedges, and hexahedrons.
We will sketch out the mesh quality issues using a tetrahedral element as an example. Since
it is not possible to represent an arbitrary geometry with regular tetrahedrons of the same
size, elements in the mesh are scaled and distorted. The resulting mesh can be thought of
as an assembly of tetrahedrons in various sizes and with various degrees of distortion.
While elements are always distorted in the process of mapping to geometry, excessive
distortion leads to element degeneration as described in Chapter 5. Complex CAD geometry
invariably contains portions that are “difficult” to mesh with properly shaped elements.
Those portions often end up being represented with highly distorted elements. The same
“difficult” portions of model geometry often coincide with areas of high stress, but highly
distorted elements are unable to produce correct results.
Excessive mesh distortion can be prevented by controlling the default element size or
by applying local mesh controls. Most FEA programs offer tools to check mesh quality. The
aspect ratio check is shown in Figure 13.9; curvature distortion, called here the Jacobian
ratio, is shown in Figure 13.10.
Sometimes, geometry makes it impossible to avoid excessively distorted elements
without modifying the geometry (Figure 13.11).
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FIGURE 13.9 Aspect ratio plots. Poorly shaped elements are created in thin portions of
the model; elements there have the highest aspect ratio. The range of the aspect ratio in the
mesh with large elements is 1.1–13.0. The range of aspect ratio in the mesh with small
elements is 1.0–3.8. The aspect ratio for a regular tetrahedron is 1.
3.5 1.6
2.3 1.3
1.1 1.0
Large elements Small elements
Range of aspect ratio 1.1-13.0 Range of aspect ratio 1.0-3.8
FIGURE 13.10 Jacobian ratio plots. Using large elements to mesh curvilinear geometry
leads to excessive curvature distortion. The range of the Jacobian ratio in the mesh with large
elements is 1.0–1.55. The range of the Jacobian ratio in the mesh with small elements is
1.0–1.21. The Jacobian ratio for an element with all straight edges is 1.
1.11 1.04
1.06 1.02
1.00 1.00
Large elements Small elements
Range of Jacobian ratio 1.0-1.55 Range of Jacobian ratio 1.0-1.21
FIGURE 13.11 Because of tangent faces, meshing this geometry produces poorly shaped
elements with high aspect ratio.
Aspect ratio
17.3
15.6
14.0
12.4
10.8
9.2
7.5
© SAE International.
5.9
4.3
2.7
1.0
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Even though we use solid elements to discuss meshing problems, similar meshing
considerations apply to shell and beam elements. In the case of shell elements, element
distortion applies to the element shape and the element warpage.
FIGURE 13.12 All first order elements are correctly shaped but the mesh is incapable of
modeling bending, which is the primary mechanism of load transfer in this structure. Several
layers of the first order elements would be required to model bending stresses properly. The
second, or higher order elements should be used.
Sharp reentrant edges cause stress singularity. This is a modeling error and not related
to the element order or other meshing issues. The correct second order element mesh has
two layers of the second order elements; stress singularities are eliminated by modeling rounds.
© SAE International.
Incorrect mesh with first order elements Correct mesh with second order elements
and local refinement in fillets
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
13.4.3. C
omputer Aided Engineering (CAE)
Programs
Modern CAE programs feature different engineering applications integrated with CAD in
such a way that information can be exchanged both ways between an application and CAD
and between applications themselves. (Figure 13.13). The list of CAE applications integrated
with CAD keeps expanding with the progress in CAE tools.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
FEA
CAD
© SAE International.
MOTION
CFD
ANALYSIS
FIGURE 13.14 Concurrent CAD-FEA design process and FEA-driven design process.
Design Design
Design Design
© SAE International.
ANALYSIS
Design DRAFTING
© SAE International.
Design
Engineer Engineer
Past Present
TABLE 13.1 Differences between FEA performed by a design engineer and by an analyst.
both CAD and FEA programs satisfy the requirements discussed in the next chapter, the
FEA training required to start a productive work is easy to complete.
The key issue in FEA training is providing users with conceptual understanding of
FEA fundamentals: major assumptions, limitations, inherent errors as well as common
mistakes, traps, and misconceptions. The training course should include hands-on exercises
for participants to benefit from the synergy between acquiring software skills and
familiarization with the FEA background at the same time. Meaningful examples illustrating
both FEA theory and software capabilities can be solved after a short introduction to the
FEA background and a quick introduction to the software. Geometry should be prepared
in advance, so no time is wasted on CAD-specific tasks. The progressive familiarity with
software should be worked into examples of increasing complexity. While extensive
theoretical treatment is not required, a simplistic approach should also be avoided. A balance
must be reached between the FEA theory and software-specific functions because too much
focus on “how to run software” may overshadow important issues of FEA fundamentals.
It may also give a false impression that familiarity with the program equals expertise in
FEA. Users who know the fundamentals of FEA can easily learn the software, but skills in
operating the software do not lead to a full understanding of FEA. After basic training,
users should be given an opportunity to try out their skills on simple projects under the
guidance of an experienced FEA user before more advanced training commences.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
The highlights of FEA training for a design engineer are summarized below:
•• Focus on conceptual understanding of FEA.
•• Make it hands-on with examples of progressing complexity.
•• Have examples prepared in the form of CAD models.
•• Combine theory with hands-on examples.
•• Use integrated CAD-FEA software.
•• Stress out that CAD and FEA are different applications.
•• Follow up with a more advanced training.
A basic training should be provided to all personnel who is not involved with FEA
directly but supervise design engineers, request analysis, interpret and implement FEA
results, etc. Training may be conducted in face-to-face format or as a webinar. In the author’s
opinion, none presents significant advantages over the other.
The FEA-specific tasks should not be hidden from users. Therefore, the FEA software
should offer control over meshing, type and order of elements, idealization scheme,
and the desired solution method. While default choices offered by advanced FEA
programs are often acceptable, the user should be able to control FEA-specific tasks,
if such an intervention is required.
•• FEA software should have a good mesher and fast solver.
The quality of FE mesh is essential in producing good quality FEA results, and a fast
solver is important to produce those results in a timely manner. A good mesher and fast
solver reduce the efforts spent on geometry preparation because even large CAD models
mesh and solve fast, providing results quick enough to keep up with the design process.
•• The FEA program should handle common types of analyses.
Analysis problems solved concurrently with the design process include the following
types of analyses:
■■ Linear static analysis
■■ Nonlinear large displacements analysis
■■ Nonlinear material analysis (simple types of nonlinear material models)
■■ Contact stress analysis
■■ Modal analysis
■■ Linear buckling analysis
■■ Steady state thermal analysis
The selected FEA program should support these types of analyses while more advanced
capabilities are optional. Chances are that occasional projects requiring advanced types of
analysis either will be too complex or data will not be available to execute the analysis
concurrently with the design process. It is best to hand over those complex projects to
analysts. To facilitate data exchange between the design engineer and analyst, the selected
FEA software should ideally be a subset of a larger FEA program. If a more advanced analysis
becomes necessary, models prepared by the design engineer can be forwarded to the analyst
without the need for geometry reconstruction or translation.
•• FEA program should have good presentation tools.
The FEA program should incorporate tools to communicate the design intent to
the rest of the organization. Results should be accessible to anybody with standard
office software, without having to use the FEA program. Presentation tools form a
very important part of the FEA program.
•• FEA program expertise should be available to users.
The process of selection of an FEA program, should include the check of the avail-
ability of expertise starting from on-line help, through hotline support, to users’
groups and consultants independent of the program developers and vendors. The
software vendor should not be the sole source of expertise in the chosen FEA program.
•• Cost of FEA program.
Finally, the FEA program should not be costly, either in terms licensing cost, cost
of implementation, or training. However, the cost alone should not be the first
consideration. The cost of buying an inappropriate program far outweighs savings
realized on the purchase price.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
13.5.4. H
ardware Selection
In the 1990s the Windows® Operating System has displaced other operating systems used
in engineering applications. Since then, hardware selection has become a secondary issue
as compared to software selection. Generally, any computer capable of running solid,
parametric, feature-based CAD is also adequate to run FEA programs. Factors to consider
include the amount of RAM memory, capacity of storage devices, and graphics card certi-
fied for use with the CAD program.
TABLE 13.2 A partial list of methods available for validating results of stress analysis
with FEA.
Experimental method Advantages Disadvantages
Strain gauge Real loads, real structure, Provides only point information
reliable stress information
Photoelasticity Provides field information Expensive and time consuming;
similitude rules difficult to satisfy
Mostly of historical importance
Stress coating Real loads, real structure, Good for qualitative information only,
relatively easy to apply quantitative results are unreliable
Brittle coating Easy to use, fast Limited to qualitative analysis
Results of laboratory/field Reliable results Testing requires a prototype that is not
tests yet available
Results of previous similar Quick to assess relative Link to reality is still unknown if the
FEA improvement between two first FEA was not validated
models
13.5.6. R
eturn on Investment
The actual return on investment depends on several factors that we have already discussed:
what CAD and FEA programs are used, how is FEA implemented in the design process,
and how well the FEA users have been trained. Experience indicates that the cost of the
FEA program and training can be recovered in a single implementation of FEA. Based on
the author’s experience in the field of automotive engineering, the cost of prototyping and
testing a simple part like an engine bracket, pulley, or door hardware can easily run some-
where between $10,000 and $50,000, taking several weeks to complete. This compares to
the cost of an FEA program in the range of $4,000–$10,000 plus $2,000 to $4,000 for training.
Even more importantly, results can be produced in a matter of hours as opposed to the days
or weeks required for prototyping and testing. The direct cost comparison is even more in
favor of FEA and against the traditional prototyping and testing approach if more complex
parts are analyzed. For complex parts the combined cost of prototyping and testing of one
design easily runs into hundreds of thousands of dollars and takes several months to
complete. The cost comparison still does not account for savings from reduced warranty
costs and improved overall product quality. The ROI will certainly differ in each individual
case, and it might be easier to talk about losses caused by not using correctly implemented FEA.
13.6.1. B
efore FEA Project Starts
Answering the following questions will help in a successful execution of an FEA project:
•• Do I really need FEA?
This is the most fundamental question to address before any analysis starts. FEA
consumes significant resources, and each use should be justified. Providing answers
to the following questions may help to decide if FEA should be executed.
•• Can I use previous test results or previous FEA results?
•• Is this a standard design so no analysis is necessary?
•• Are loads, restraints, and material properties known well enough to make FEA
worthwhile?
•• Would a simplified analytical model do?
•• Does my customer demand the FEA?
•• What shall I do with the results of the FEA?
•• Should the analysis be done in-house or should it be contracted out?
•• Conducting analysis in-house and using consultants both have advantages and
disadvantages. Consultants may produce results faster while in-house analysis is
conducive to establishing company expertise leading to long-term savings.
•• How fast do I need to produce results?
•• Do I have enough time and resources to do a complete FEA before design decisions
must be made?
13.6.2. E
stablish the Scope of Analysis
The following questions help in defining the scope of analysis:
•• Is this a standard analysis of a new product from an established product line?
•• Is this the last check of a production-ready new design before the final testing?
•• Is this a quick check of a design in progress to assist the designer?
•• Is this an aid to the R&D project (detail of a design, gauge, fixture, etc.)?
•• Is this conceptual analysis to support a design at an early stage of development
(R&D project)?
•• What is an estimated number of model iterations, load cases, etc.?
•• How should I analyze the results? What are the evaluation criteria and safety factors?
•• How will I know whether the results can be trusted?
13.6.3. C
reate the Mathematical Model
Having established the scope of analysis, the modeling begins. The best model is the
simplest one that still provides the required results with acceptable accuracy. Therefore,
the mathematical model should be as simple as possible to minimize project cost and
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
duration, yet it should account for the essential characteristics of the analyzed object.
We need to decide on the acceptable defeaturing and/or idealization. The goal is to produce
a meshable geometry properly representing the analyzed problem. Decisions on using
2D representation may have to be made. Loads and restraints must be formulated in
accordance with the type of performed analysis. For example, vibration analysis requires
loads and/or supports defined as a function of time. Large deformation analysis requires
defining loads as following or non-following, etc. Material properties must be defined in
accordance with the type of analysis.
13.6.4. C
reate the Finite Element Model and Solve It
The Finite Element model is created by discretization of a mathematical model; this is
commonly called meshing. Although the term “meshing” implies that only geometry is
discretized, the discretization also affects loads and restraints. Meshing and solution are
largely automated steps but still require the user’s input. Creating a finite element model
based on the accepted mathematical model includes decisions on:
•• Element type to be used
•• Element size and size tolerance
•• Mesh controls (if any)
•• Type of mesher to be used
•• Type of load control
•• Type of solver to be used
13.6.5. R
eview the Results
FEA results must be critically reviewed prior to using them to make design decisions. This
includes review of assumptions and assessment of the results. This is an iterative step that
may require several loops to debug the FEA model and establish confidence in the results.
Review of results include:
•• Study of animated displacements.
•• Check overall stress levels (order of magnitude) using analytical methods to verify
the applied loads.
•• Check reactions and compare them to reactions found using free body diagrams.
•• Conduct convergence process and review discretization errors.
•• Analyze stress concentrations and the ability of mesh to model them properly.
•• Review results in difficult-to-model locations like thin walls and stress
concentrations.
•• Investigate the impact of element distortions on the data of interest.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
13.6.6. P
resentation of the Results
The exact execution of this step depends on the objective of the analysis. In all cases however,
results should be presented in a way suitable for using them to make a design decision.
Analysis of results includes the following steps:
•• Present displacement results.
•• Present modal frequencies and associated modes of vibration (if applicable).
•• Present stress results and corresponding factors of safety.
•• Propose modifications to the analyzed structure to eliminate excessive stresses to
improve material utilization and manufacturability.
•• Discuss results and repeat iterations until the acceptable solution is found.
Management of the FEA project requires the manager’s involvement during project
execution. The correctness of FEA results cannot be established only by a review of the
results. Involvement during the project progress helps managers to keep in the loop and
improve communication with the person performing the analysis. Several checkpoints are
suggested in Figure 13.16.
FIGURE 13.16 Checkpoints in an FEA project; the project is allowed to proceed after each
step has been approved by the manager.
Geometry OK?
Mesh OK?
© SAE International.
Results OK?
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
13.6.8. P
roject Documentation and Backups
All FEA-related documents should be subjected to applicable document quality standards.
FEA reports and backups can be used for:
•• Audit of the work performed
•• Restart of the work
•• Basis for execution of modified analyses
•• Basis for personnel training
•• Establishing in-house expertise in FEA
•• Legal documents when liability is involved
A comprehensive project documentation, complete with backups and all additional
documentation, is the major building block in the process of accumulating and retaining
in-house FEA expertise. The project documentation should be sufficient to recreate the
results or run a modified analysis without any need for verbal communications. Without
proper project documentation, expertise gained in the project may be lost.
Project definition must be specific; client and consultant should know exactly what
needs to be done. The in-house design engineer analyst should approve each step: type of
analysis, definition of loads, restraints, mesh creation, result analysis, conclusions. It is
advisable to get a warranty in case if results do not correlate with experimental data. Each
FEA project, even if contracted out, should contribute to in-house experience in FEA. The
main steps in contracted-out FEA projects are shown in Figure 13.17. A project that is
contracted out still requires significant involvement of the requestor, and client-consultant
communication must be open at all times.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Finding consultant
Request for quotes Implementation of results
Technical expertise Follow up by consultant
Prevoiusly completed projects In-house follow-up
Time and cost
© SAE International.
Assessment of quotes Accumlating in-house expertise
Understanding the quotes Project documentation
Work scope Leassong learned
Warranty FEA Archive
13.6.10. C
ommon Errors in Management of FEA
Projects
FEA projects are often challenging, and management errors are easily made. Below is a
sample of common errors with short comments. This is not an “all inclusive” list and is
intended only to highlight some frequently encountered problems.
•• “Please do FEA”
Project objectives are not clearly defined, neither the manager nor the person
performing FEA knows exactly what the project objectives and expected benefits are.
•• Too high expectations placed on FEA
FEA results are viewed as an assurance of a sound design.
•• No time limits
The project is allowed to take as much time as the person performing FEA wishes
to take.
•• “We do FEA on everything”
FEA is done on models that could be analyzed with less-expensive methods.
•• Proficiency in software confused with expertise in FEA
Skills in using the FEA program are taken as proof of expertise in FEA.
•• Confusing CAD and FEA
The same measures are applied to CAD and FEA projects and personnel.
•• Not enough time allowed for training in FEA
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
14
Summary
with solid elements. When making meshing decisions, we should always consider
alternatives to solids: shell elements, beam elements, or 2D elements.
4. Better (more expensive) FEA software gives better results: true/false?
False. Correct results depend on the user’s modeling skills. An expensive program
may produce wrong results and vice versa.
5. Automatic meshing is better than manual meshing: true/false?
This is true and false at the same time. Automatic meshing is for sure better because
it takes little time as compared to mapped or truly manual meshing. Automatic
meshing is the only method available in modern FEA programs, but an automatically
produced mesh is not automatically correct. A mesher does not know what the objec-
tive of your analysis is. It just fills up model geometry with elements. It is our respon-
sibility to ensure that the mesh is refined where stress concentrations are expected,
there are enough elements across members in bending, etc. For more difficult
geometry, meshers tend to produce distorted elements and place elements with no
regard for laws of mechanics (i.e., one first order tetrahedral element across the wall
in bending). Meshers are much “safer” when used with second or higher order
elements. Never use first order elements unless you have a very good reason to use
them. Do not rely on the mesher being left on defaults; almost always, the user must
control the element size and mesh bias.
6. High accuracy of FEA results from high processing accuracy of the digital
computer: true/false?
False. First, who said that FEA is highly accurate? It may be if it is used properly and
if that high accuracy is desired. Accuracy of FEA cannot be confused with the
accuracy of solvers. Most FEA programs use a double-precision format for lower
numerical error, and in most cases, the round-off error is small as compared to other
errors: modeling and discretization errors.
7. If your FEA software reports no error, the solution will be correct: true/false?
False. No error means only that the model is correct from the solver’s point of
view. The solver will solve an incorrect model if it does not run into numerical
problems.
8. You do not really need any error estimation, the FEA is always accurate enough:
true/false?
False. One single run provides results with unknown error. The error may be low, but
we still do not know what that error is. Unless we have previous experience with
similar problems, convergence analysis is required to estimate the discretization error.
9. You should always make a very fine mesh, so you do not have to worry about error:
true/false?
False. While very fine mesh produces low discretization error, but we still need more
than one mesh to know what that error is. Also, a fine mesh does nothing to
modeling errors.
10. Higher order elements will give you more accurate results: true/false?
True if “more accurate” means “lower discretization error.” Using higher order
elements in place of lower order elements is equivalent to mesh refinement. Comparing
two otherwise identical models, with the first order elements and with the second
elements, the latter will provide more accurate results in terms of discretization error.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Also, it is much easier to construct a correct mesh using second order elements because
second order elements map to curvilinear geometry and model linear stress distribu-
tion while first order elements cannot map to curvilinear geometry and model
constant stress.
11. First order tetrahedral solid elements are too stiff and should be avoided: true/
false?
Generally true. When using first order tetrahedral elements, we need substantially
more elements for modeling stress gradients and for correct representation of curvi-
linear geometry. Unless you have a good reason, do not use first order elements.
12. Make a coarse mesh first to find stress concentration and then refine it as needed:
true/false?
True in most cases. However, the first coarse mesh still needs to be fine enough to
detect stress concentrations. If the element size is too large in comparison to the size
of a “hot spot,” then stress concentration will not be detected, and you will never
know that you should have refined the mesh in that particular location.
13. If FEA results correlate well with the experiment (e.g., strain-gauge readings), then
all FEA results are OK: true/false?
False. A strain gauge may be placed in a spot that is modeled correctly in the finite
element model. However, correlation in one or more locations does not guarantee
that everything is fine with the model. The opposite statement is true: if FEA does
not correlate with the experiment, then something is wrong with the model or with
the experiment or both.
14. All major FEA programs have been extensively tested so you do not need to do any
benchmarking yourself: true/false?
Generally true. Still, it is recommended to run benchmark tests to get the feel of the
FEA program, particularly in the more demanding applications.
15. Highly distorted elements are OK if they are far away from stress concentrations:
true/false?
False. Distorted elements may affect the global model stiffness. In other words, they
“pollute” the model, and that may propagate to the area of interest rendering
erroneous results.
16. If you study displacements and not stresses, then you can use a coarse mesh: true/
false?
True. Nodal displacements converge faster than stresses. Still, using a certain element
size should be justified by the results of the convergence analysis.
17. Modal analysis can use a coarser mesh than stress analysis: true/false?
True. Modal analysis finds natural frequencies and shapes of vibration which are
global measures as opposed to stresses which are local measures. Details may usually
be ignored. However, if modal analysis provides prerequisite results to vibration
analysis which finds stresses, then the mesh used in the modal analysis should satisfy
all meshing requirements applicable to stress analysis.
18. FEA offers a deceiving level of detail: true/false?
True. Results may come with numbers with 10 digits accuracy, and different plots
may be produced and animated. It is easy to forget that all results rest on simplifying
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
K d F (14.1)
where [K] is a stiffness matrix, [d] is an unknown vector of nodal displacements, and
[F ] is a vector of nodal loads.
9. What is a displacement interpolation function?
A displacement interpolation function is a formula used to calculate displacement
in the given location of the element using displacements of nodes. Displacement
interpolation functions are defined on element edges, faces, and element interior.
Displacement interpolation functions must satisfy several requirements:
A. Internal compatibility: displacements must be continuous over the
entire element.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
A p convergence does not affect the element size; the mesh stays the same during
the convergence process, but the element order is upgraded. A p convergence is done
automatically in an iterative solution until the user-specified convergence criterion
is satisfied; the only input required from the user is convergence criteria and the
desired accuracy. A p convergence is an adaptive process and requires elements capable
of upgrading the order of selected edges and faces.
Sometimes the desired accuracy cannot be achieved even with the highest avail-
able element p-order. In that case, the user must refine the mesh manually like in the
h convergence process and then run the p convergence process again. This is called
p-h or h-p convergence.
18. What is the objective of convergence analysis?
FEA gives only an approximate solution. Convergence analysis estimates the discreti-
zation error. The discretization error in the FEA solution does not have to be mini-
mized but should be low enough to satisfy user’s requirements.
Examples of convergence criteria:
•• Strain energy
•• RMS strain energy (local or global)
•• Stress (local or global)
•• Displacement (local or global)
19. What is the difference between verification and validation of FEA results?
Verification checks if the mathematical model, as submitted to be solved with FEA,
has been correctly discretized and solved; verification deals with the solution of the
mathematical model. Validation determines if the results correctly represent reality
from the perspective of the intended use of the model. It checks if the results correctly
apply to reality.
20. What types of errors affect the accuracy of FEA results?
•• Modeling error
Modeling error originates from the fact that FEA does not work on the real
structure but on its idealized mathematical model. The idealizations introduce
modeling errors that can be reduced, but never eliminated, by good modeling
practices. The modeling error cannot be estimated by any FEA technique because
FEA provides an approximate solution to whatever mathematical model was
chosen. Modeling error happens before FEA enters the stage. The study of
modeling errors answers the question “How well does the mathematical model
represent reality?”
•• Discretization error
Discretization error results from using a finite number of DOFs to approximate a
solution characterized by an infinite number of DOFs. Discretization is accom-
plished by meshing. Discretization imposes certain assumptions on the data of
interest, for example, displacements inside the element must be of the second order.
Discretization errors are assessed in the convergence process.
Discretization error can be defined as a convergence error or a solution error.
Convergence error is the difference in the results between two consecutive steps
in the convergence process. This error can be calculated explicitly. Solution error
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
is the difference between FEA results and results that would have been provided
by a continuous mathematical model. Solution error can be only estimated but
not calculated exactly. To estimate the solution error, one must assess the conver-
gence of results and predict changes in the results within the several next iterations
as if they were performed. Analysis of discretization error answers the question
“What is the error of solution of the mathematical model?”
•• Numerical error also called solution error
Numerical error of FEA results is the round-off error accumulated by the solver;
usually it is low.
•• Error of interpretation of results
A typical example would be using von Mises stress to calculate the factor of safety
of a brittle material. The error of interpretation of results should be completely
eliminated by the FEA user’s training.
21. Can you estimate discretization error without convergence analysis?
Strictly speaking, No; a single solution provides results with an unknown discretiza-
tion error. However, there are cases where skipping convergence analysis may be OK.
If you have successfully analyzed and validated similar problems, then analysis of
discretization error may be skipped, if your model is “reasonably” close to the former
one. You can also use global or local error estimators provided in your software. These
error estimators are related to the difference between averaged and non-
averaged stresses.
22. What is stress averaging?
Stress averaging is a technique used to present stress results. Implementation may
vary in different FEA programs. Stress-averaging procedure can be explained as
follows. Stress results are calculated for each element in certain locations called Gauss
points. If more than one Gauss point is present in the element (as is the case with
second and higher order elements), one averaged stress is reported for the element.
Even though those stresses are averaged between Gauss points, they are called non-
averaged stresses or element stresses because averaging is done internally within the
same element. Alternatively, stresses in Gauss points may be extrapolated to element
nodes without prior averaging between themselves. Most often, one node is shared
by several elements, and each element reports different stress at the shared node.
Stress is then averaged on node producing averaged (or nodal) stress results. Both
nodal (averaged) and element (non-averaged) stress should be examined during the
analysis of stress results. The large difference between nodal and element stress indi-
cates the need for mesh refinement.
23. What is the difference between loads and boundary conditions?
Loads and restraints applied to the boundary of the solution domain are both
boundary conditions. Loads are called natural boundary conditions and restraints
are called essential boundary conditions. Volume loads like gravity or inertial load
are not boundary conditions.
24. How many sets of loads and restraints may be defined for one linear static analysis?
If your program permits it, you can define different load cases because the stiffness
matrix remains the same, no matter what load is applied. However, only one set of
restraints may be defined because the stiffness matrix changes with the change
of restraints.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
25. What is the difference between boundary conditions applied to geometry and
FEA model?
A common way to apply loads and restraints is to define them on CAD geometry.
The loads and restraints are converted into nodal loads and nodal restraints during
meshing. Even though meshing is commonly understood as discretization of
geometry, everything else in the mathematical model is also discretized. Loads,
restraints, and mass are discretized and applied to nodes. There are no continuous
entities in an finite element model.
26. How are loads and restraints represented in the FE model?
Real loads acting on a structure—pressure, own weight, inertial forces, etc.—are
continuous. Point loads or line loads do not exist in nature. The same applies to
restraints. Loads in a finite element model are represented as nodal forces and
moments. Restraints are defined on nodes as prescribed displacement. For example,
a fixed restraint defines zero displacement.
27. What are the major types of FEA?
This depends on how FEA is classified. There is structural and thermal FEA, linear
and nonlinear FEA, static and vibration (dynamic) FEA, etc. Also, we can classify
FEA by the objective like Design FEA when it is used as a design tool, forensic analysis
to find out why a structure failed, or an optimization exercise to improve the
existing design.
28. What are the types of nonlinearities?
A nonlinear analysis is the one in which the stiffness of the structure changes during
the process of load application. That change needs to be modeled in the process of
iterative solution that is always required in nonlinear analysis. There are different
sources of nonlinearity; the following is not a complete list:
•• Nonlinear supports
•• Contact stress
•• Tension/compression only links
•• Nonlinear elastic foundation
•• Offset or eccentric loads
•• Buckling
•• Geometric non-linearity
•• Material nonlinearity
29. How to reduce the size of the finite element model.
•• Taking advantage of repetitive geometry
Sometimes geometry and boundary conditions (loads and restraints) are charac-
terized by certain symmetry such as mirror symmetry, anti-symmetry, cyclic
symmetry, or axial symmetry. In these cases, it is enough to model only one portion
of the model and simulate the remaining portion with the applicable boundary
conditions. The extra time spent on modification of geometry is usually well
worth it.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• Defeaturing geometry
Defeaturing of geometry, meaning deleting structurally insignificant details, is
particularly important when working with complex CAD geometry. Manufacturing
geometry produced in CAD is not oriented toward analysis. If used without modi-
fications, it may result in a very complicated model that is too large to solve.
Defeaturing must be done very carefully; features important for the objective of
analysis must not be removed.
•• Idealization: using shell or beam elements in place of solid elements
For a certain class of geometry, meshing with solid elements is out of the question.
For example, sheet metal must be meshed with shell elements and weldments with
beam elements.
•• Using 2D representations
Working with 2D models simplifies model generation, solution, and analysis of
results but is applicable only to a narrow class of problems suitable for plane stress,
plane strain, or axisymmetric representation.
30. What is the bandwidth of stiffness matrix, why is it important, and how can it
be reduced?
Bandwidth is a parameter describing a stiffness matrix. It is an important consider-
ation when solving FEA equations. The stiffness matrix in FEA is symmetric, and in
most cases, all non-zero elements of this matrix are grouped along the main diagonal.
The bandwidth is the largest distance between non-zero coefficients in any row. The
larger the bandwidth, the more numerically demanding is the solution. For low
bandwidth, nodes should be numbered in such a way that the numbers assigned to
nodes belonging to each element are close to each other. Nodes are numbered auto-
matically to minimize the bandwidth.
31. How should a typical FEA report look like?
Typical parts of an FEA report include
•• Executive summary
Explains the objective of the project, part number, project number, presents essen-
tial assumption, results, and conclusions.
•• Introduction
Describes the problem: why did it require FEA; what kind of FEA was that, static,
contact stress, vibration analysis, etc.
•• Description of material model, loads, and restraints
Explains the material model in use; describes restraints and load cases analyzed.
Load diagrams and discussion of simplifying assumptions should be included in
this part.
•• Geometry
Describes the model geometry, defeaturing, and/or idealization, if applicable;
justifies the modeling approach (solids elements? shells elements? 2D?) and the
choice of meshing.
•• Analysis of results
Presents the data of interest; estimates error; calculates the factors of safety; etc.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
•• Conclusions
Makes recommendations to the requester, proposes necessary modifications and
further studies if needed; recommends testing procedure (e.g., strain-gauge test,
fatigue life test) to validate results; makes a recommendation regarding future
similar designs.
•• Project documentation
To build in-house expertise you need very good documentation of the project
besides the report so the project can be easily reused in the future by somebody
else. The documentation should include extensive write-up, design drawings,
model explanations, and computer backups. You should allow a considerable
amount of time to prepare project documentation.
•• Follow-up
After tests are completed, append your report with test results, discuss result
correlation, and re-run the FEA analysis if the correlation was not satisfactory.
32. Where are mistakes often made?
Mistakes can be made just about anywhere during an FEA project, but the most
common are
•• Incorrect mathematical model
•• Improper use of elements in a mesh
For example, one layer of first order elements is placed across the thickness of the
wall in bending or elements incorrectly mapped to geometry.
•• Using excessively distorted elements
Excessively distorted are often produced with an automatic mesh generator
running on default settings; the user’s input into meshing is almost always required.
•• Too few elements in a mesh
If a mesh is too coarse, it will not be able to model the structure properly; stress
concentrations may not be detected.
•• Incorrect restraints
A common mistake is an overly restrained model resulting in an overly stiff struc-
ture. Displacement and, consequently, stresses are then underestimated.
•• Hasty result analysis, no lessons learned
FEA provides a wealth of results which need to be analyzed properly. Each project
should contribute to the accumulated corporate experience.
33. What are alternatives to FEA?
As far as Computer Aided Engineering is concerned, we can use other numerical
tools like the Finite Volume Method or Boundary Element Method. Also, meshless
methods are being developed and look very promising. However, let us not forget
about the more traditional approach like long-hand calculations, computerized tables,
and testing. These methods are less expensive and often more reliable than FEA.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
15
FEA Resources
T
here are many sources of FEA expertise available to users that include, but are not
limited to,
•• Engineering textbooks
•• Software specific manuals
•• Engineering journals
•• Professional development courses that include face-to-face and distant learning
•• FEA users’ groups
•• Government organizations
Engineering literature offers a large selection of reference material on FEA and engi-
neering analysis closely related to FEA; we will list only a few of them in the order of
relevance to topics discussed in this book.
1. Adams, V. and Askenazi, A., Building Better Products with Finite Element Analysis (Santa Fe, NM:
OnWord Press, 1999)
2. Incropera, F., Dewitt, D., Bergman, T., and Lavine, A., Fundamentals of Heat and Mass Transfer
(Hoboken, NJ: John Wiley & Sons, Inc., 2007)
3. Inman, D., Engineering Vibration (Englewood Cliffs, NJ: Prentice Hall, 1994)
4. Logan, D., A First Course in the Finite Element Method (Boston, MA: Brooks/Cole, 2007)
5. Szabo, B. and Babuska, I., Finite Element Analysis (Hoboken, NJ: John Wiley & Sons, Inc., 1991)
Kurowski, P., Engineering Analysis with SOLIDWORKS Simulation 2022 (Mission, KS: SDC
Publications, 2022)
Kurowski, P., Vibration Analysis with SOLIDWORKS Simulation 2022 (Mission, KS: SDC Publications,
2022)
Kurowski, P., Thermal Analysis with SOLIDWORKS Simulation 2022 and Flow Simulation 2022
(Mission, KS: SDC Publications, 2022)
With so many applications for FEA, attempts have been made to create a governing
body overlooking FEA standards and practices. One of the leading organizations in this
field is the National Agency for Finite Element Methods and Standards (NAFEMS).
NAFEMS has published many excellent titles such as:
•• A Finite Element Primer
•• A Finite Element Dynamics Primer
•• Guidelines to Finite Element Practice
•• Background to Benchmarks
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
16
Glossary
Averaged Stresses These stresses are calculated at nodes by averaging stresses at the node
as reported by all elements sharing that node. Nodal stresses are “smoothed out” and, by
virtue of averaging, produce continuous stresses in the model. Average stresses are also
called nodal stresses.
Beam Element Element produced by meshing a line (may be straight or curved).
Black Body A black body is an ideal radiator and an ideal absorber of heat in the form
of electromagnetic waves.
Boundary Element Method A numerical method used for solving field problems; only
the boundary of the solution domain needs to be discretized. The Boundary Element Method
is very efficient for analyzing compact 3D shapes, but difficult to use on more “spread
out” shapes.
CAD Computer Aided Design.
CAD Specific Geometry Geometry with details required for manufacturing.
Cleanup Process of removing and/or repairing geometric features that would prevent
the mesher from creating the mesh or would result in an incorrect mesh.
Conjugate Heat Transfer Heat transfer process in solids and fluids.
Convergence Criteria Convergence criteria are conditions assessed during the conver-
gence process. In an automated convergence process convergence criteria are used to decide
when the process should stop.
Convergence Process A process of systematic changes to the mesh to see how the data
of interest change with the change of mesh and (hopefully) prove that the data of interest
are not significantly dependent on the choice of discretization. A convergence process may
be performed as h convergence or p convergence.
Defeaturing Process of removing (or suppressing) geometric features in CAD geometry
to simplify the finite element mesh.
Degree of Freedom (DOF) Ability of a node to perform displacement in the form of
translation and/or rotation. In structural analysis, a node may have up to 6 DOFs, depending
on the element type. In thermal analysis nodes have 1 DOF, this is temperature.
Discretization The process of dividing a continuous mathematical model into discrete
“pieces” called elements. A visible effect of discretization is a finite element mesh. However,
mass, loads, and restraints are also discretized.
Discretization Error This error affects FEA results because FEA works on an assembly
of discrete elements (mesh) rather than on a continuous structure. The finer the finite
element mesh the lower the discretization error, but the solution takes more time.
Displacement Interpolation Function A polynomial function defined on nodal degrees
of freedom: displacements or temperature. It calculates displacements or temperatures
inside an element (in the element volume, on the element faces, and along the element edges)
based on the known nodal displacements or temperatures.
Distortion of an Element Departure from the element ideal shape. It happens as a result
of element mapping onto the model geometry.
Emissivity Effectiveness in emitting energy as thermal radiation.
FEA Finite Element Analysis.
FEA-Specific Geometry Geometry intended for FEA, usually derived from CAD geometry
by defeaturing, idealization, and cleanup.
Finite Difference Method An alternative to the FEA; the solution domain is discretized
into a grid. The Finite Difference Method is generally less efficient for solving structural
and thermal problems but is often used in fluid dynamics problems.
Finite Element Finite elements are the building blocks of a mesh, defined by the coordi-
nates of their nodes and by functions approximating of distribution of sought-for quantities,
such as displacements or temperatures.
Finite Volumes Method This is an alternative to the FEA method of solving field problems,
similar to the Finite Difference Method and frequently used in fluid dynamics problems.
Following Load Load which retains its orientation relative to the deforming model. The
difference between the following and the non-following load is an important consideration
in nonlinear geometry analysis.
Frequency Response Analysis Type of vibration analysis where excitation is a function
of frequency.
h Element h elements are all elements, for which the order does not change during analysis.
Convergence analysis of a model using h elements is done by refining the mesh and
comparing results (like displacement, stresses, etc.) before and after the refinement.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
The name h comes from the element characteristic dimension h, which is reduced during
consecutive mesh refinements.
Idealization Making simplifying assumptions in the process of defining a mathematical
model. Idealization may involve simplifying geometry, replacing thin walls with midplane
surfaces, using 2D representations, etc.
Idealization Error The idealization error results from the fact that analysis is conducted
on a model and not on a real-life object. Geometry, material properties, loads, and restraints
all are idealized in models submitted to FEA.
Linear Material Material model where stress is linearly proportional to strain.
Mapping Process of changing an element shape from the ideal shape to a shape it assumes
in the mesh; element mapping takes place during meshing.
Membrane Element Element produced by meshing a surface. As opposed to shell element,
membrane element does not have bending stiffness.
Meshing Process of discretization geometry, mass, restraints, and loads. This is an auto-
mated process with user-specified controls.
Mesh Diagnostic Task performed by an FEA program that checks the mesh for element
distortion, element connectivity, etc.
Meshing Process of discretizing the model geometry. As a result of meshing, the originally
continuous geometry is represented by an assembly of finite elements.
Modal Analysis Modal analysis calculates the natural frequencies and the associated
shapes of vibration. Modal analysis does not calculate displacements or stresses.
Modeling Error This type of error results from idealizations that are introduced in the
process of creating a mathematical model. See Idealization Error.
Non-averaged Stresses Stresses in an element without averaging them with stresses
reported by neighboring elements; also called element stresses.
Non-following Load Load which retains its orientation relative to the external coordinate
system and not to the deforming model. The difference between the following and the non-
following load is an important consideration in nonlinear geometry analysis.
Numerical Error Error caused by the accumulated round-off error during solution. The
value of numerical errors is usually very low.
p Adaptive Solution Solution method using p elements. A p adaptive solution provides
results with a narrowly specified accuracy but is more time consuming than the solution
using h elements.
p Element p Elements are elements that do not have a predefined order. The solution of
a p element model requires several iterations while the element order is upgraded until the
difference in the user-specified measures (e.g., global strain energy, RMS stress) becomes
less than the requested accuracy. The name p comes from the order of polynomial functions
that is changed during the solution.
Pre-load Pre-load is a load that modifies the structure stiffness. Pre-load is an important
consideration in modal analysis where it may change natural frequencies. In some cases,
pre-load must be also considered in static analysis.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Transient Thermal Analysis Thermal analysis where temperature and other heat transfer
parameters change with time.
Validation Validation determines if an FEA model correctly represents reality from the
perspective of the intended use of the model. It checks if the results correctly describe the
real-life behavior of the analyzed object.
Verification Verification checks if the mathematical model, as submitted to be solved
with FEA, has been correctly discretized and solved.
Von Mises Stress A stress measure that takes into consideration all six stress components
of the state of stress. Von Mises stress is a scalar, non-negative value. The magnitude of von
Mises stress can be used to determine the safety factors of ductile materials.
Yield Strength This refers to the maximum stress that can be allowed in a model before
plastic deformation takes place. Yield strength is used to calculate the factor of safety in
ductile materials.
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
17
List of Exercises
TABLE 17.1 List of exercise files in Parasolid format and SOLIDWORKS 2022 format.
File are listed in the order of appearance in hands-on exercises.
Chapter Parasolid SOLIDWORKS 2022
4 HOLLOW PLATE.x_t HOLLOW PLATE.sldprt
L BRACKET01.x_t L BRACKET01.sldprt
WEDGE.x_t WEDGE.sldprt
5 BRACKET01.x_t BRACKET01.sldprt
CANTILEVER.x_t CANTILEVER.sldprt
6 BRACKET02.x_t BRACKET02.sldprt
SHAFT01.x_t SHAFT01.sldprt
PRESSURE TANK.x_t PRESSURE TANK.sldprt
RING.x_t RING.sldprt
LINK01.x_t LINK01.sldprt
PLATE.x_t PLATE.sldprt
7 CANTILEVER BEAM.x_t CANTILEVER BEAM.sldprt
SHAFT02.x_t SHAFT02.sldprt
ROUND PLATE01.x_t ROUND PLATE01.sldprt
LINK02.x_t LINK02.sldprt
SLIDING SUPPORT.x_t SLIDING SUPPORT.sldprt
CLAMP01.x_t CLAMP01.sldprt
CLAMP02.x_t CLAMP02.sldprt
8 BRACKET NL.x_t BRACKET NL.sldprt
L BRACKET02.x_t L BRACKET02.sldprt
9 TUNING FORK.x_t TUNING FORK.sldprt
BOX.x_t BOX.sldprt
AIRPLANE.x_t AIRPLANE.sldprt
BALL.x_t BALL.sldprt
LINK03.x_t LINK03.sldprt
HELICOPTER BLADE.x_t HELICOPTER BLADE.sldprt
COLUMN.x_t COLUMN.sldprt
BRACKET03.x_t BRACKET03.sldprt
10 NOTCHED COLUMN.x_t NOTCHED COLUMN.sldprt
ROUND PLATE02.x_t ROUND PLATE02.sldprt
CURVED COLUMN.x_t CURVED COLUMN.sldprt
STAND.x_t STAND.sldprt
CURVED SHEET.x_t CURVED SHEET.sldprt
11 HAMMER.x_t HAMMER.sldprt
ELBOW PIPE.x_t ELBOW PIPE.sldprt
CENTRIFUGE.x_t CENTRIFUGE.sldprt
PLANK.x_t CENTRIFUGE.xlsx
PLANK.sldprt
12 BRACKET04.x_t BRACKET04.sldprt
HEAT SINK.x_t HEAT SINK.sldprt
CHANNEL.x_t CHANNEL.sldprt
SPACE HEATER.x_t SPACE HEATER.sldasm
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Index
262 Index
Index 263
vs. static analysis, 139–140 idealization, 213, 214 nonlinear vibration analysis, 190, 191
stress results, 140 integrated with CAD, 219 results, 112, 113
vibration analysis, 146 management errors, 232–233 thin beam under pressure, 114, 116
vibration shape, 144–145 mathematical model, 227–228 thin flat plate, 112, 123–124
weak spots, 145, 151 meshing Structural analysis, 199–200
Modal Superposition Method, 146, 181–182 aspect ratio plots, 216, 217 T
Modeling error, 13–14, 242 Jacobian ratio, 216, 217 Thermal analysis
Motion Analysis, 2–4 mesh adequacy, 218 challenges, 206
tangent faces, 216, 217
N heat flow
personnel training, 221–223
Nodal stress, 34–36 by conduction, 200–201, 207, 209
positioning activities, 220, 221
Non-averaged stress, 36 by convection, 201–203, 208, 209
program selection, 223–225
Non-compatible elements, 241 by radiation, 203–204, 209
project documentation and backups, 231
Nonlinear buckling analysis heat sink and microchip assembly,
project execution, 227
Arc-Length Control method, 164, 165 204–205, 207–208
reports, 230
buckling mode shape, 167, 171 sharp reentrant edges, 205
return on investment, 226
chain of events, 165, 167 structural analysis and, 199–200
review of results, 228
compressive loads, 163–164 Three-dimensional (3D) geometry, 8–9
scope of analysis, 227
elastic-perfectly plastic material model, Time Response Analysis
stand-alone programs, 219
165, 167, 168, 170 displacement time history, 183–185
vs. linear buckling analysis, 167, 173 R impact force, 183
linear material model, 165, 166, 169 Radiation, 203–204, 209 load time history, 183, 184, 192–194
Load Control method, 164 Repetitive symmetry, 92–93 V
load-displacement curve, 167, 173 Restraints, 10, 54, 59, 86, 88, 95, 143 Validation, 14–15, 242
onset controlling, 174–175, 178–179 Rigid Body Motions (RBMs), 20, 95, 141–143 Verification, 14–15, 242
results, 167, 171, 172 Vibration analysis
“snap-through” effect, 177–178 S
Frequency Response Analysis
stress stiffness, 164 Semi-automatic meshing, 66–67
elbow pipe, 186, 194
structural collapse, 167, 170 Shell element, 24, 25, 34, 36, 72
industrial centrifuge, 188–189, 194–195
Nonlinear material analysis, 134–135 Single Degree of Freedom (SDOF), 146, 182
results, 186, 187
section clipping, 129, 132 “Snap-through” effect, 177–178
vibration modes, 186, 187
stress singularity Solid element, 24, 58, 59, 94, 235
harmonic impulse load, 192–193
elastic-perfectly material, 134 SOLIDWORKS Simulation, 250, 259, 260
Modal Superposition Method, 181–182
maximum von Mises stress, 132, 133 Solution error, 14
nonlinear analysis
sharp reentrant edge, 132, 133, 135, 136 Standards, 250
displacement response, 190, 192
stress-strain curve, 129, 130 Stiffness matrix, 12–13, 105–106, 245
plank stiffness, 190, 191, 195–196
Von Mises stress, 129, 131 Strain, 20, 26, 27, 129–130
time history of pressure, 190, 191
Nonlinear vibration analysis Stress
Time Response Analysis
displacement response, 190, 192 analysis, 84
displacement time history, 183–185
plank stiffness, 190, 191, 195–196 concentration, 237
impact force, 183
time history of pressure, 190, 191 constant, 18, 19
load time history, 183, 184, 192–194
Numerical error, 14, 243 in-plane, 24, 25, 75, 76
Von Mises stress
results, 20–22
O Stress singularity, 58 displacement interpolation function,
Omega Square Excitation, 188 20–22
elastic-perfectly plastic material,
h convergence
P 50–52, 134
adaptive h process, 42–44
Parasolid format, 259, 260 sharp reentrant edge, 50, 51, 53, 132, 133,
global mesh refinement, 36, 38–40
Product development process, 211 135, 136
local mesh refinement, 41–42
analysis of results, 229 2D plane stress problem, 48, 49
linear material model, 132, 133
building confidence, 225–226 Von Mises stress, 48, 49, 132, 133
nonlinear geometry analysis, 118, 120
CAE programs, 219–220 Stress stiffening
nonlinear material analysis, 129, 131
contracted-out project, 231–232 bending stiffness, 112, 113
p adaptive process, 45, 46
defeaturing, 212, 213 buckling analysis, 175, 176
singularity, 48, 49, 54, 56
Finite Element model, 228 floating hinge configuration, 114, 115, 124
geometry cleanup, 214–216 linear displacement analysis, 114, 116, 117 Y
hardware selection, 225 load-displacement curve, 113, 114 Yield strength, 129
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Downloaded from SAE International by SAE International [Sales Team], Tuesday, December 20, 2022
Dr. Paul M. Kurowski obtained his M.Sc. and Ph.D. in Applied Mechanics from Warsaw
University of Technology. He completed postdoctoral work at Kyoto University. Paul
Kurowski is a professor in the Faculty of Engineering at the University of Western Ontario.
His teachings include undergraduate and graduate courses in Product Design and