6

FEM Modeling:
Introduction

6–1
Chapter 6: FEM MODELING: INTRODUCTION 6–2

TABLE OF CONTENTS

Page
§6.1. FEM Terminology 6–3
§6.2. Idealization 6–4
§6.2.1. Models . . . . . . . . . . . . . . . . . . . 6–5
§6.2.2. Mathematical Models . . . . . . . . . . . . . . 6–5
§6.2.3. Implicit vs. Explicit Modeling . . . . . . . . . . . . 6–6
§6.3. Discretization 6–6
§6.3.1. Analytical or Numerical? . . . . . . . . . . . . . 6–6
§6.3.2. Error Sources and Approximation . . . . . . . . . . 6–7
§6.3.3. Other Discretization Methods . . . . . . . . . . . 6–7
§6.4. The Finite Element Method 6–8
§6.5. Element Attributes 6–8
§6.5.1. Dimensionality . . . . . . . . . . . . . . . . . 6–8
§6.5.2. Nodes . . . . . . . . . . . . . . . . . . . 6–9
§6.5.3. Geometry . . . . . . . . . . . . . . . . . . 6–9
§6.5.4. Degrees of Freedom . . . . . . . . . . . . . . . 6–10
§6.5.5. Nodal Forces . . . . . . . . . . . . . . . . . 6–10
§6.5.6. Constitutive Properties . . . . . . . . . . . . . . 6–10
§6.5.7. Fabrication Properties . . . . . . . . . . . . . . 6–10
§6.6. Classification of Mechanical Elements 6–10
§6.6.1. Primitive Structural Elements . . . . . . . . . . . 6–10
§6.6.2. Continuum Elements . . . . . . . . . . . . . . . 6–11
§6.6.3. Special Elements . . . . . . . . . . . . . . . . 6–11
§6.6.4. Macroelements . . . . . . . . . . . . . . . . . 6–11
§6.6.5. Substructures . . . . . . . . . . . . . . . . . 6–12
§6.7. Assembly 6–12
§6.8. Boundary Conditions 6–12
§6.8.1. Essential and Natural B.C. . . . . . . . . . . . . . 6–12
§6.8.2. Boundary Conditions in Structural Problems . . . . . . 6–13
§6. Notes and Bibliography
. . . . . . . . . . . . . . . . . . . . . . 6–13
§6. References. . . . . . . . . . . . . . . . . . . . . . 6–14

6–2
6–3 §6.1 FEM TERMINOLOGY

Chapters 2 through 5 cover material technically known as Matrix Structural Analysis or MSA. This
is a subject that historically preceded the Finite Element Method (FEM), as chronicled in Appendix
H. Here we begin the coverage of the FEM proper. This is distinguished from MSA by two traits:
• The ability to go beyond structures.
• The prominent role of continuum models and variational mathematics.
This Chapter introduces terminology used in FEM modeling, and surveys attributes and types of
finite elements used in structural mechanics. The next Chapter gives more specific rules for defining
meshes and boundary conditions.
§6.1. FEM Terminology
The ubiquitous term “degrees of freedom,” often abbreviated to either freedom or DOF, has figured
prominently in the preceding Chapters. This term, as well as “stiffness matrix” and “force vector,”
originated in structural mechanics, the application for which FEM was invented. These names have
carried over to non-structural applications. This “terminology overspill” is discussed next.
Classical analytical mechanics is that invented by Euler and Lagrange in the XVIII century and
further developed by Hamilton and Jacobi as a systematic formulation of Newtonian mechanics.
Its objects of attention are models of mechanical systems ranging from material particles composed
of sufficiently large number of molecules, through airplanes, to the Solar System.1 The spatial
configuration of any such system is described by its degrees of freedom or DOF. These are also
called generalized coordinates. The terms state variables and primary variables are also used,
particularly in mathematically oriented treatments.
If the number of degrees of freedom is finite, the model is called discrete, and continuous otherwise.
Because FEM is a discretization method, the number of DOF of a FEM model is necessarily finite.
They are collected in a column vector called u. This vector is called the DOF vector or state vector.
The term nodal displacement vector for u is reserved to mechanical applications.
In analytical mechanics, each degree of freedom has a corresponding “conjugate” or “dual” term,
which represents a generalized force.2 In non-mechanical applications, there is a similar set of
conjugate quantities, which for want of a better term are also called forces or forcing terms. They
are the agents of change. These forces are collected in a column vector called f. The inner product
fT u has the meaning of external energy or work.
Just as in the truss problem, the relation between u and f is assumed to be of linear and homogeneous.
The last assumption means that if u vanishes so does f. The relation is then expressed by the master
stiffness equations:
Ku = f. (6.1)
K is universally called the stiffness matrix even in non-structural applications because no consensus
has emerged on different names.
The physical significance of the vectors u and f varies according to the application being modeled,
as illustrated in Table 6.1.
1 For cosmological scales, such as galaxy clusters, the general theory of relativity is necessary. For the atomic and
sub-particle world, quantum mechanics is appropriate.
2 In variational mathematics this is called a duality pairing.

6–3
Chapter 6: FEM MODELING: INTRODUCTION 6–4

Table 6.1. Significance of u and f in Miscellaneous FEM Applications

Application State (DOF) vector u Conjugate vector f

Problem represents represents

Structures and solid mechanics Displacement Mechanical force

Heat conduction Temperature Heat flux
Acoustic fluid Displacement potential Particle velocity
Potential flows Pressure Particle velocity
General flows Velocity Fluxes
Electrostatics Electric potential Charge density
Magnetostatics Magnetic potential Magnetic intensity

If the relation between forces and displacements is linear but not homogeneous, equation (6.1)
generalizes to
Ku = f M + f I . (6.2)

Here f I is the initial node force vector introduced in Chapter 29 for effects such as temperature
changes, and f M is the vector of mechanical forces.
The basic steps of FEM are discussed below in more generality. Although attention is focused on
structural problems, most of the steps translate to other applications problems as noted above. The
role of FEM in numerical simulation is schematized in Figure 6.1, which is a merged simplification
of Figures 1.2 and 1.3. Although this diagram oversimplifies the way FEM is actually used, it serves
to illustrate terminology. The three key simulation steps shown are: idealization, discretization and
solution. Each step is a source of errors. For example, the discretization error is the discrepancy
that appears when the discrete solution is substituted in the mathematical model. The reverse steps:
continuification and realization, are far more difficult and (generally) ill-posed problems.
The idealization and discretization steps, briefly mentioned in Chapter 1, deserve further discussion.
The solution step is dealt with in more detail in Part III of this book.

IDEALIZATION DISCRETIZATION SOLUTION

FEM
Physical Mathematical Discrete Discrete
system model model solution

REALIZATION & CONTINUIFICATION Solution error

IDENTIFICATION
Discretization + solution error
Modeling + discretization + solution error

Figure 6.1. A simplified view of the physical simulation process, reproduced

to illustrate modeling terminology.

6–4
6–5 §6.2 IDEALIZATION

§6.2. Idealization
Idealization passes from the physical system to a mathematical model. This is the most important
step in engineering practice, because it cannot be “canned.” It must be done by a human.

§6.2.1. Models
The word “model” has the traditional meaning of a scaled copy or representation of an object. And
that is precisely how most dictionaries define it. We use here the term in a more modern sense,
which has become increasingly common since the advent of computers:

A model is a symbolic device built to simulate and predict aspects of behavior of a system.
(6.3)
Note the distinction made between behavior and aspects of behavior. To predict everything, in all
physical scales, you must deal with the actual system. A model abstracts aspects of interest to the
modeler. The qualifier symbolic means that a model represents a system in terms of the symbols
and language of another discipline. For example, engineering systems may be (and are) modeled
with the symbols of mathematics and/or computer sciences.3

§6.2.2. Mathematical Models

Mathematical modeling, or idealization, is a process by which an engineer or scientist passes from
the actual physical system under study, to a mathematical model of the system, where the term
model is understood in the sense of (6.3).
The process is called idealization because the mathematical model is necessarily an abstraction of
the physical reality — note the phrase aspects of behavior in (6.3). The analytical or numerical
results produced by the mathematical model are physically re-interpreted only for those aspects.4
To give an example of the choices that an engineer may face, suppose that the structure is a flat plate
structure subjected to transverse loading. Here is a non-exhaustive list of four possible mathematical
models:
1. A very thin plate model based on Von Karman’s coupled membrane-bending theory.
2. A thin plate model, such as the classical Kirchhoff’s plate theory.
3. A moderately thick plate model, for example Mindlin-Reissner plate theory.
4. A very thick plate model based on three-dimensional elasticity.
The person responsible for this kind of decision is supposed to be familiar with the advantages,
disadvantages, and range of applicability of each model. Furthermore the decision may be different
in static analysis than in dynamics.
Why is the mathematical model an abstraction of reality? Engineering systems, particularly in
Aerospace and Mechanical, tend to be highly complex. For simulation it is necessary to reduce that

3 A problem-definition input file, a digitized earthquake record, or a stress plot are examples of the latter.
4 Whereas idealization can be reasonably taught in advanced design courses, the converse process of “realization” or
“identification” — see Figure 6.1 — generally requires considerable physical understanding and maturity that can only
be gained through professional experience.

6–5
Chapter 6: FEM MODELING: INTRODUCTION 6–6

complexity to manageable proportions. Mathematical modeling is an abstraction tool by which

complexity can be controlled.
This is achieved by “filtering out” physical details
that are not relevant to the analysis process. For ex-
ample, a continuum material model filters out the T
EN
aggregate, crystal, molecular and atomic levels of M PON
al O
C EVE L
atic L
matter. Engineers are typically interested in a few hem
Matmodel
ent
integrated quantities, such as the maximum deflec- ary Com
ponns
atio
Libr t equ
tion of a bridge or the fundamental periods of an FEM onen
omp e
C iscret TEM
d del SYS EL
airplane. Although to a physicist this is the result of mo
LEV e
plet
the interaction of billions and billions of molecules, Comution
sol
em
such details are weeded out by the modeling pro- Systrete
disc del
mo
cess. Consequently, picking a mathematical model sica
Phy tem
l
sys
is equivalent to choosing an information filter.

§6.2.3. Implicit vs. Explicit Modeling

As noted the diagram of Figure 6.1 is an oversimpli- Figure 6.2. A reproduction of Figure 1.5 with
fication of engineering practice. The more common some relabeling. Illustrates implicit modeling:
scenario is that pictured in Figures 1.2, 1.4 and 1.5. picking elements from an existing FEM code
consents to an idealization.
The latter is reproduced in Figure 6.2 for conve-
nience.
A common scenario in industry is: you have to analyze a structure or a substructure, and at your
disposal is a “black box” general-purpose finite element program. Those programs offer a catalog
of element types; for example, bars, beams, plates, shells, axisymmetric solids, general 3D solids,
and so on. The moment you choose specific elements from the catalog you automatically accept
the mathematical models on which the elements are based. This is implicit modeling. Ideally you
should be fully aware of the implications of your choice. Providing such “finite element literacy” is
one of the objective of this book. Unfortunately many users of commercial programs are unaware
of the implied-consent aspect of implicit modeling and their legal implications.
The other extreme happens when you select a mathematical model of the physical problem with
your eyes wide open and then either shop around for a finite element program that implements
that model, or write the program yourself. This is explicit modeling. It requires far more technical
expertise, resources, experience and maturity than implicit modeling. But for problems that fall out
of the ordinary it could be the right thing to do.
In practice a combination of implicit and explicit modeling is common. The physical problem
to be simulated is broken down into subproblems. Those subproblems that are conventional and
fit available programs may be treated with implicit modeling, whereas those that require special
handling may only submit to explicit modeling.

§6.3. Discretization
Mathematical modeling is a simplifying step. But models of physical systems are not necessarily
simple to solve. They often involve coupled partial differential equations in space and time subject to
boundary and/or interface conditions. Such models have an infinite number of degrees of freedom.

6–6
6–7 §6.3 DISCRETIZATION

§6.3.1. Analytical or Numerical?

At this point one faces the choice of going for analytical or numerical solutions. Analytical solutions,
also called “closed form solutions,” are more intellectually satisfying, particularly if they apply to
a wide class of problems, so that particular instances may be obtained by substituting the values of
free parameters. Unfortunately they tend to be restricted to regular geometries and simple boundary
conditions. Moreover some closed-form solutions, expressed for example as inverses of integral
transforms, may have to be anyway numerically evaluated to be useful.
Most problems faced by the engineer either do not yield to analytical treatment or doing so would
require a disproportionate amount of effort.5 The practical way out is numerical simulation. Here
is where finite element methods enter the scene.
To make numerical simulations practical it is necessary to reduce the number of degrees of freedom
to a finite number. The reduction is called discretization. The product of the discretization process
is the discrete model. As discussed in Chapter 1, for complex engineering systems this model is
the product of a multilevel decomposition.
Discretization can proceed in space dimensions as well as in the time dimension. Because the
present book deals primarily (except in Part IV) with static problems, we need not consider the time
dimension and are free to concentrate on spatial discretization.

§6.3.2. Error Sources and Approximation

Figure 6.1 tries to convey graphically that each simulation step introduces a source of error. In
engineering practice modeling errors are by far the most important. But they are difficult and ex-
pensive to evaluate, because model validation requires access to and comparison with experimental
results. These may be either scarce, or unavailable in the case of a new product in the design stage.
Next in order of importance is the discretization error. Even if solution errors are ignored — and
usually they can — the computed solution of the discrete model is in general only an approximation
in some sense to the exact solution of the mathematical model. A quantitative measurement of
this discrepancy is called the discretization error. The characterization and study of this error is
addressed by a branch of numerical mathematics called approximation theory.
Intuitively one might suspect that the accuracy of the discrete model solution would improve as
the number of degrees of freedom is increased, and that the discretization error goes to zero as that
number goes to infinity. This loosely worded statement describes the convergence requirement of
discrete approximations. One of the key goals of approximation theory is to make the statement as
precise as it can be expected from a branch of mathematics.6

5 This statement has to be tempered in two respects. First, the wider availability and growing power of computer algebra
systems, outlined in Chapter 4, has widened the realm of analytical solutions than can be obtained within a practical time
frame. Second, a combination of analytical and numerical techniques is often effective to reduce the dimensionality of
the problem and to facilitate parameter studies. Important examples are provided by Fourier analysis, perturbation and
boundary-element methods.
6 The discretization error is often overhyped in the FEM literature, since it provides an inexhaustible source of publishable
poppycock. If the mathematical model is way off, reducing the discretization error buys nothing; just a more accurate
answer to the wrong problem.

6–7
Chapter 6: FEM MODELING: INTRODUCTION 6–8

§6.3.3. Other Discretization Methods

It was stated in Chapter 1 that the most popular discretization techniques in structural mechanics are
finite element methods and boundary element methods. The finite element method (FEM) is by far
the most widely used. The boundary element method (BEM) has gained in popularity for special
types of problems, particularly those involving infinite domains, but remains a distant second, and
seems to have reached its natural limits.
In non-structural application areas such as fluid mechanics and electromagnetics, the finite element
method is gradually making up ground but faces stiff competition from both the classical and energy-
based finite difference methods. Finite difference and finite volume methods are particularly well
entrenched in computational fluid dynamics spanning moderate to high Reynolds numbers.
§6.4. The Finite Element Method
The finite element method (FEM) is the dominant discretization technique in structural mechanics.
As discussed in Chapter 1, the FEM can be interpreted from either a physical or mathematical
standpoint. The treatment in Chapters 1–10 emphasizes the former.
The basic concept in the physical FEM is the subdivision of the mathematical model into disjoint
(non-overlapping) components of simple geometry called finite elements or elements for short. The
response of each element is expressed in terms of a finite number of degrees of freedom characterized
as the value of an unknown function, or functions, at a set of nodal points. The response of the
mathematical model is then considered to be approximated by that of the discrete model obtained
by connecting or assembling the collection of all elements.
The disconnection-assembly concept occurs naturally when examining many artificial and natural
systems. For example, it is easy to visualize an engine, bridge, building, airplane, or skeleton as
fabricated from simpler components.
Unlike finite difference models, finite elements do not overlap in space. In the mathematical
interpretation of the FEM, this property goes by the name disjoint support or local support.
§6.5. Element Attributes
Just like members in the truss example, one can take finite elements of any kind one at a time. Their
local properties can be developed by considering them in isolation, as individual entities. This is
the key to the modular programming of element libraries.
In the Direct Stiffness Method, elements are isolated by the disconnection and localization steps,
which were described for the truss example in Chapter 2. The procedure involves the separation of
elements from their neighbors by disconnecting the nodes, followed by referral of the element to a
convenient local coordinate system.7 After that we can consider generic elements: a bar element,
a beam element, and so on. From the standpoint of the computer implementation, it means that
you can write one subroutine or module that constructs, by suitable parametrization, all elements
of one type, instead of writing a new one for each element instance.
Following is a summary of the data associated with an individual finite element. This data is used
in finite element programs to carry out element level calculations.
7 Both steps are only carried out in the modeler’s mind. They are placed as part of the DSM for instructional convenience.
In practice, processing begins directly at the element level.

6–8
6–9 §6.5 ELEMENT ATTRIBUTES

1D

2D

2D

3D

Figure 6.3. Typical finite element geometries in one through three dimensions.

§6.5.1. Dimensionality
Elements can have intrinsic dimensionality of one, two or three space dimensions.8 There are also
special elements with zero dimensionality, such as lumped springs or point masses. The intrinsic
dimensionality can be expanded as necessary by use of kinematic transformations. For example a
1D element such as a bar, spar or beam may be used to build a model in 2D or 3D space.
§6.5.2. Nodes
Each element possesses a set of distinguishing points called nodal points or nodes for short. Nodes
serve a dual purpose: definition of element geometry, and home for degrees of freedom. When a
distinction is necessary we call the former geometric nodes and the latter connection nodes. For
most elements studied here, geometric and connector nodes coalesce.
Nodes are usually located at the corners or end points of elements, as illustrated in Figure 6.3. In
the so-called refined or higher-order elements nodes are also placed on sides or faces, as well as
possibly the interior of the element.
Remark 6.1. In some elements geometric and coinnection nodes may be at different locations. This is
illustrated by the Veubeke equilibrium triangle described in Chapter 15. Some elements have purely geometric
nodes, also called orientation nodes to complete the definition of certain geometric attributes. An example is
the spar element in 3D shown in Figure E5.2, in which a third geometric node is used to define a local plane.

§6.5.3. Geometry
The geometry of the element is defined by the placement of the geometric nodal points. Most
elements used in practice have fairly simple geometries. In one-dimension, elements are usually
straight lines or curved segments. In two dimensions they are of triangular or quadrilateral shape.

8 In dynamic analysis, time may appear as an additional dimension.

6–9
Chapter 6: FEM MODELING: INTRODUCTION 6–10

In three dimensions the most common shapes are tetrahedra, pentahedra (also called wedges or
prisms), and hexahedra (also called cuboids or “bricks”). See Figure 6.3.

§6.5.4. Degrees of Freedom

The element degrees of freedom (DOF) specify the state of the element. They also function as
“handles” through which adjacent elements are connected. DOFs are defined as the values (and
possibly derivatives) of a primary field variable at connector node points. The actual selection
depends on criteria studied at length in Part II. Here we simply note that the key factor is the way
in which the primary variable appears in the mathematical model. For mechanical elements, the
primary variable is the displacement field and the DOF for many (but not all) elements are the
displacement components at the nodes.

§6.5.5. Nodal Forces

There is always a set of nodal forces in a one-to-one correspondence with degrees of freedom. In
mechanical elements the correspondence is established through energy arguments.

§6.5.6. Constitutive Properties

Physical Mathematical Finite Element
For a mechanical element these are relations Structural Model Name Discretization
that specify the material behavior. For Component
example, in a linear elastic bar element it
bar
is sufficient to specify the elastic modulus E
and the thermal coefficient of expansion α.

§6.5.7. Fabrication Properties beam

For mechanical elements these are fabrica-

tion properties which have been integrated
tube, pipe
out from the element dimensionality. Exam-
ples are cross sectional properties of MoM
elements such as bars, beams and shafts, spar (web)
as well as the thickness of a plate or shell
element.
For computer implementation the foregoing shear panel
(2D version of above)
data sets are organized into appropriate data
structures. These are used by element
generation modules to compute element Figure 6.4. Examples of primitive
structural elements.
stiffness relations in the local system.

§6.6. Classification of Mechanical Elements

The following classification of finite elements in structural mechanics is loosely based on the
“closeness” of the element with respect to the original physical structure. It is given here because
it clarifies points that recur in subsequent sections, as well as providing insight into advanced
modeling techniques such as hierarchical breakdown and global-local analysis.

6–10
6–11 §6.6 CLASSIFICATION OF MECHANICAL ELEMENTS

Physical Finite element Physical Finite element

idealization idealization

plates 3D solids

Figure 6.5. Continuum element examples.

§6.6.1. Primitive Structural Elements

These resemble fabricated structural components. They are often drawn as such; see Figure 6.4.
The qualifier primitive distinguishes them from macroelements, which is another element class
described below. Primitive means that they are not decomposable into simpler elements.
These elements are usually derived from Mechanics-of-Materials simplified theories and are better
understood from a physical, rather than mathematical, standpoint. Examples are the elements
discussed in Chapter 5: bars, cables, beams, shafts, spars.

§6.6.2. Continuum Elements

These do not resemble fabricated structural components at all. They result from the subdivision of
“blobs” of continua, or of structural components viewed as continua.
Unlike structural elements, continuum elements are better understood in terms of their mathematical
interpretation. Examples: plates, slices, shells, axisymmetric solids, general solids. See Figure 6.5.

§6.6.3. Special Elements

Special elements partake of the characteristics of structural and continuum elements. They are
derived from a continuum mechanics standpoint but include features closely related to the physics
of the problem. Examples: crack elements for fracture mechanics applications, shear panels, infinite
and semi-infinite elements, contact and penalty elements, rigid-body elements. See Figure 6.6.

Infinity
double node

Crack Infinite Honeycomb

element element panel

Figure 6.6. Special element examples.

6–11
Chapter 6: FEM MODELING: INTRODUCTION 6–12

§6.6.4. Macroelements

Macroelements are also called mesh units and superelements, although the latter term overlaps with
substructures (defined below). These often resemble structural components, but are fabricated with
simpler elements. See Figure 6.7.
The main reason for introducing macroelements is to
simplify preprocessing tasks. For example, it may be
simpler to define a regular 2D mesh using quadrilat-
erals rather than triangles. The fact that, behind the
scene, the quadrilateral is actually a macroelement
may not be important to most users.
Similarly a box macroelement can save modeling
times for structures that are built by such components;
Figure 6.7. Macroelement examples.
for example box-girder bridges

§6.6.5. Substructures

Also called structural modules and superelements. These are sets of elements with a well defined
structural function, typically obtained by cutting the complete structure into functional components.
Examples: the wings and fuselage in an airplane; the towers, deck and cables in a suspension bridge.
The distinction between substructures and macroelements is not clear-cut. The main conceptual
distinction is that substructures are defined “top down” as parts of a complete structure, whereas
macroelements are built “bottom up” from primitive elements. The term superelement is often used
in a collective sense to embrace element groupings. This topic is further covered in Chapter 10.

§6.7. Assembly

The assembly procedure of the Direct Stiffness Method for a general finite element model follows
rules identical in principle to those discussed for the truss example. As in that case the processs
involves two basic steps:
Globalization. If necessary, the element equations are transformed to a common global coordinate
system.
Merge. The element stiffness equations are merged into the master stiffness equations by appropriate
indexing and matrix-entry addition.
The hand calculations for the example truss conceal the implementation complexity. The master
stiffness equations in practical cases may involve thousands or even millions of freedoms, and
programming can become involved. The topic is elaborated upon in Chapter 25.

§6.8. Boundary Conditions

A key strength of the FEM is the ease and elegance with which it handles arbitrary boundary and
interface conditions. This power, however, has a down side. A big hurdle faced by FEM newcomers
is the understanding and proper handling of boundary conditions. Below is a simple recipe for
treating boundary conditions. The following Chapter provides specific rules and examples.

6–12
6–13 §6. Notes and Bibliography

§6.8.1. Essential and Natural B.C.

The key thing to remember is that boundary conditions (BCs) come in two basic flavors: essential
and natural.
Essential BCs directly affect DOFs, and are imposed on the left-hand side vector u.
Natural BCs do not directly affect DOFs and are imposed on the right-hand side vector f.
The mathematical justification for this distinction requires use of concepts from variational calculus,
and is consequently relegated to Part II. For the moment, the basic recipe is:

1. If a boundary condition involves one or more degrees of freedom in a direct way, it

is essential. An example is a prescribed node displacement.
2. Otherwise it is natural.

The term “direct” is meant to exclude derivatives of the primary function, unless those derivatives
also appear as degrees of freedom, such as rotations in beams and plates.

§6.8.2. Boundary Conditions in Structural Problems

Essential boundary conditions in mechanical problems involve displacements (but not strain-type
displacement derivatives). Support conditions for a building or bridge problem furnish a particularly
simple example. But there are more general boundary conditions that occur in practice. A structural
engineer must be familiar with displacement B.C. of the following types.

Ground or support constraints. Directly restraint the structure against rigid body motions.
Symmetry conditions. To impose symmetry or antisymmetry restraints at certain points, lines or
planes of structural symmetry. This allows the discretization to proceed only over part of the
structure with a consequent savings in modeling effort and number of equations to be solved.
Ignorable freedoms. To suppress displacements that are irrelevant to the problem.9 Even experi-
enced users of finite element programs are sometimes baffled by this kind. An example are rotational
degrees of freedom normal to smooth shell surfaces.
Connection constraints. To provide connectivity to adjoining structures or substructures, or to
specify relations between degrees of freedom. Many conditions of this type can be subsumed under
the label multipoint constraints or multifreedom constraints. These can be notoriously difficult to
handle from a numerical standpoint, and are covered in Chapters 8–9.

9 In classical dynamics these are called ignorable coordinates.

6–13
Chapter 6: FEM MODELING: INTRODUCTION 6–14

Notes and Bibliography

Most FEM textbooks do not provide a systematic treatment of modeling. This is no accident: few academic
authors have experience with complex engineering systems. Good engineers are too busy (and in demand) to
have time for writing books. This gap has been particularly acute since FEM came on the scene because of
generational gaps: “real engineers” tend to mistrust the computer, and often for good reason. The notion of
explicit versus implicit modeling, which has legal and professional implications, is rarely mentioned.
FEM terminology is by now standard, and so is a majority of the notation. But that is not so in early publications.
E.g. K is universally used10 for stiffness matrix in virtually all post-1960 books. There are a few exceptions:
the often cited Przemieniecki [200] uses S. There is less unanimity on u and f for node displacement and force
vectors, respectively; some books such as Zienkiewicz and Taylor [274] still use different symbols.
The element classification given here attempts to systematize dispersed references. In particular, the distinction
between macroelements, substructures and superelements is an ongoing source of confusion, particularly
since massively parallel computation popularized the notion of “domain decomposition” in the computer
science community. The all-encompassing term “superelement” emerged in Norway by 1968 as part of the
implementation of the computer program SESAM. Additional historical details are provided in Chapter 10.
The topic of BC classification and handling is a crucial one in practice. More modeling mistakes are done in
this aspect of FEM application than anywhere else.

References
Referenced items have been moved to Appendix R.

10 A symbol derived from the “spring constant” k that measures the stiffness of a mechanical spring.

6–14