DESAI

VARIOUS ELEMENTS SHAPES

The process of discretizing or subdividing a continuum is an exercise of engineering
judgment. The factors to be considered in this process will be discussed in chapter 6.
However, the first decision the engineer must take is to select the shape or configuration
of the basic element to be used in the analysis. This choice depends upon the geometry of
the body or structure and upon the number of dependent space coordinates (e.g. x, y or z)
necessary to describe the problem. As we shall see, a finite element usually has a simple
one-, two-, or three-dimensional configuration. Te boundaries of elements are often
straight lines, although for problem that can be best represented in curvilinear
coordinates, it is advantageous for the element shapes to be similarly defined. In this
section we shall emphasis element with straight-line boundaries; in all cases, however,
analogous element with curve-line boundaries are possible. Curved elements will be
treated later in this chapter.

One-Dimensional Elements
When the geometry, material properties, and such dependent variables as a displacement
can all be expressed in term of one independent space coordinate, a one-dimensional
element is appropriate. This coordinates is measured along the axis of the element. This
type of element is used for structures that can be idealized by line drawing, such as the
NODE 3

. . .
NODE 1 NODE 2
Figure 5-6 one-dimensional element

truss and frame show in figure 5-1. In a strict sense, therefore we are speaking of frame
analysis when we speak of one-dimensional elements, because a distinguishing
characteristic of the finite element method is that the elements themselves are continuous
two- or three-dimensional bodies. However, a finite elements approach to the analysis of
beams and framed structures produced some useful insights, particularly when the
geometry and properties of the structure very continuously or discontinuously with the
axis coordinates. In addition one-dimensional problem are a simple but useful tool for
developing our understanding of the finite element method. This utility will be
implemented in Sample Problem 1 (Fig. 5-4).
A one-dimensional element may be represented by a straight line whose ends are nodal
points, Figure 5-6. These nodal points, numbered 1 and 2 in the figure, are called external
nodes because they represent connecting points to the adjacent elements. Some
applications may require additional nodal points, such as node 3 in Figure 5-6. Because
no connection to other elements occurs at this intermediate node, it is called an internal
node.

Two-Dimensional Elements
In Chapter 3 we saw that many problems in solid machines could be approximated by a
two-dimensional formulation are plane strain, plane stress, and plate bending. The
simplest element for two-dimensional problems is such a triangle as shown in Figure 5-5
and 5-7(a). There are two possible types o external nodes for triangular elements. The
corner nodes indicated by 1, 2,and 3 in Figure 5-7(a) are called primary external nodes.
When additional nod4s occur on the sides of the element, like nodes 4, 5, and 6 in Figure
5-7(a), we shall refer to them as secondary external nodes. This disinction is necessary
because the secondary node may have fewer displacements of interest then the corner
nodes. Finally, internal nodes, such as node 7 in Figure 5-7(a), are sometimes also used
in triangular elements.
Other common types of two-dimensional elements are the rectangular and quadrilateral
shapes, Figure 5-7 (b) and 5-7(c). The former can be considered a specialization of the
latter. Although practically any two-dimensional continuum can be represented by an
assemblage of triangles, there are certain problems in which quadrilateral elements are
advantageous. Instead of directly using quadrilateral elements, it is possible to construct
such shapes from two or four triangular elements, as shown in Figures 5-7(d) and 5-7(e).
IN addition to the primary external nodes shown in Figure 5-7(b) through 5-7(e), each of
the quadrilateral elements may also have secondary external nodes and one or more
internal nodes.

(a) Triangular element (b) Rectangular element

(c) Quadrilateral element (d) Quadrilateral formed by two triangles

Figure 5-7 Two-dimensional elements

Axisymmetric Elements

Another type of three-dimensional problem which can be represented completely in terms

of one or two independent variables is the axisymmetric problem. Here a cylindrical
coordinate system, r, z, and , is used; and all properties and variables are independent of
the  coordinate. The finite elements used for these problems are circular rings, called
toroidal elements, with cross-section in the r-z plane similar to the shapes in Figures 5-6
and 5-7. Thus a one-dimensional axisymmetric element is a ring with a triangular [Fig. 5-
7(f)] or quadrilateral cross-section.

Three-Dimensional Elements
corresponding to the triangle, the tetrahedron shown in Figure 5-8(a) is the basic finite
element for three-dimensional elements with eight primary external nodes are either in
the form of a general hexahedron, Figure 5-8(c), or a rectangular prism, Figure 5-8(b),
which is the specialization of a hexahedron. If necessary we can introduce secondary
external nodes or internal nodes for each of these elements. By analogy with the two-
dimensional elements, a hexahedron can be constructed form several tetrahedra. A
hexahedron with nine edges is formed from two tetrahedra. We need a larger number of
tetrahedral, such as five, six, twelve, or twenty-four, to construct a general twelve-edge
hexahedron. One such element composed of five tetrahedral is shown in Figure 5-8(d)

(a) Tetrahedron (b) Rectangular prism

Figure 5.8 three-dimensional element

 (d) a hexahedron (cubic) composed of five tetrahedra
Figure 5-8 continued
Mixed Assemblages
The reader can probably which cannot be represented by an assemblage of only of the
types of elements mentioned above. Often two or more types of elements are needed to
approximate a particular problem. One example of this would be the bending of the plate
which is stiffened by edge beams. Here the plate itself would be represented by two-
dimensional elements such as rectangles, while the edge beams would be approximated
by one-dimensional elements appropriately connected to the plate element.

DISPLACEMENT MODEL

The basic philosophy of the finite element method is piecewise approximation. That is,
we approximate a solution to a complicated problem by sub-dividing the region of
interest and representing the solution within each sub-division by a relatively simple
function. Specifically, we have spoken of the displacement method of structural analysis,
in which the structure or body is divided into finite elements. We now consider the
simple functions which are assumed to approximate the displacements for each element.
These functions are called displacement method, displacement function, displacement
fields, or displacement pattern.

Generalized Coordinate Form of Displacement Model

In Chapter 1 we stated that a polynomial as the most common form of displacement
model. There are two principal reasons why this form is so widely used. First, it is easy to
handle the mathematics of polynomials in formulating the desired equations for various
elements and in performing digital computation. In particular, the use of polynomials
permits us to differentiate and integrate with relative ease. Second, a polynomial f
arbitrary order permits a recognizable approximation to the true solution. We can see that
a polynomial of infinite order corresponds to an exact solution. By transecting an infinite
polynomial at different orders, we clearly vary the degree of approximation, as shown for
the simple one-dimensional case in Figure 5-9. In this figure, an “exact” solution for the
displacement u(x) is approximated by various degree polynomials of the general form.

u(x) = 1 + 2x + 3x2 + … + n+1xn (5.1 a)

The greater the number of terms included in the approximation, the more closely the
exact solution s represented. (Obviously, this last statement does not apply to the case
wherein the exact solution is a polynomial of some finite order m. here terms in excess of
m do not improve the representation.)
In equation (5.1 a), the coefficient of the polynomial, the ’s, are known as generalized
coordinates or generalized displacement amplitudes. The number of terms retained in the
polynomial determines the shape of the displacement model, whereas the magnitudes of
the generalized coordinates govern the amplitude. These amplitudes are called
“generalized” because they are not necessarily identified with the physical displacements
of the element on a one-to-one basis; rather, they are linear combinations of some of the
nodal displacements and perhaps of some of the derivatives of displacements at the nodes
as well. The generalized coordinates represent the minimum number of parameters
necessary to specify the polynomial amplitude. Their physical significance (5.1 a) can be
expressed in matrix form as

u(x) = {}T{}
{}T = [1 x x2 … xn] (5.1 b)
{}T = [1 2 3 .. n+1]

The general polynomial form of a two-dimensional displacement model is

u(x, y) = 1 = 2 x + 3 y + 4 x2 + 5 xy + 6 y2 + …. + myn (5.2 a)

v(x, y) = m+1 + m+2 x + m+3y + m+4x2 + m+5 xy + m+6 y2
+ ….. + V2myn
where

and where u and v denote the components of displacements in the x and y directions,
respectively (Fig. 5-5). Equation (5.2 a) can also be expressed in matrix form.

(5.2 b)

Finally, a three-dimensional displacement model of nth order is given by

where

and where u, v, and w are the components of displacement. Equation (5.3a) can be
written in matrix form as

(5.3 b)

Each of these general polynomials forms of displacement models in equations (5.1),

(5.2).and (5.3) can be truncated at any desired degree to give constant, linear, quadratic,
cubic, or higher order pattern. For example, in the two-dimensional case, a linear model
is

(5.2 c)

and a quadratic model is

(5.2 d)

Because the displacement model in equations (5.1) through(5.3) are expressed in terms of
generalized coordinates, {}, they are referred to as generalized coordinate displacement
models. This is the elementary form of model for the finite element method. In section 5-
4 we shall consider an alternative representation of polynomial displacement fields that
facilitates the formulation of the basic equation for the element. However, for the
remainder for this section we shall continue to discuss the characteristics of displacement
models in the elementary form.

Convergence Requirements
In any acceptable numerical formulation, the numerical solution must converge or tend to
the exact solution of the problem. For the finite element method, it has been shown that
under certain circumstances the displacement formulation provides an upper bound of the
true stiffness of the structures.
In other words, the stiffness coefficients for a given displacements model have
magnitudes higher than those for the exact solution. Thus, under a given load, the
simulation structure deforms less then the actual structure. It follows that, as the finite
element subdivision of the structure is made finer, the approximate displacement solution
will be bound the exact solution from below, that is, we obtain lower bounds to the true
solution. In order for this roundedness to be rigorously assured, three conditions must be
met.

1. The displacement model must be continuous within the elements, and the
displacement must be compatible between adjacent elements. The first part of this
requirement is readily met by choosing polynomial models, which are inherently
continuous. The sound part implies that adjacent element must deform without
causing openings, overlaps, or discontinuities between the elements this will be
discussed further in the next chapter. Inter-element compatibility cannot be
satisfied unless the displacements along the side of an element depend only upon
the displacement of the nodes occurring on that side.
2. The displacement models must include the rigid body displacement of the element.
A rigid body displacement is the most elementary deformation that an element
may undergo. Basically, this condition states that there should exist combination
of values of the generalized coordinates that cause all points on the element to
experience the same displacement. One such combination should occur for each
of the possible rigid body translation and rotation. For instance, in te
displacement model of equation(5.1) the constant term 1 provides for a rigid
body displacement.
3. The displacement model must include the constant strain states of the element.
This requirement can be stated in terms similar to those of the second condition.
There should exist combination of values of the generalized coordinates that
cause all points on the element to experience the same strain. One such
combination should occur of each possible strain. The necessity of this
requirement can be understood physically if we imagine the subdivision of a body
into smaller and smaller elements. As these elements approach infinitesimal size,
the strains in each element approach constant values. Unless our approximation
includes these constant strains, we cannot hope to converge to a correct solution.
The terms associated with 2 and m+3 in equation (5.2 a) provide for uniform
strains x and y in an element.
In the finite element literature, formulations which satisfy the first condition above
are called compatible or conforming. Elements which meet both the second and third
requirements are known as complete. Later we shall see how vocational principles are
used in the finite element formulation. To use such principle meaningfully, we must
ensure that the displacement model will allow continuous nonzero derivatives of the
highest order appearing in the potential energy functional. This is a more general
statement of the completeness criteria.
Displacement models of the types given in equation (5.1), (5.2), and (5.3), when
applied to axial tension, plane strain, and three-dimensional elasticity, respectively,
satisfy the three conditions given above if at least the constant and linear terms are
retained in the polynomials. In other applications, such as the analysis of the beam,
plate, and shell structures, it is more difficult to satisfy the first two requirements.
However, although all three conditions are sufficient to prove convergence in the
 general case, practical results for elements that satisfy only the third requirement
appear to converge acceptably. In particular, elements which relax the first criterion,
that is , which are complete but nonconforming, have been widely and successfully
used. The principal advantage of nonconforming elements is that no longer know in
advance that the stiffness will be an upper bound. On other and, nonconforming
formulations are usually less stiff (In other words, more flexible) tan compatible
ones, so the former may converge more quickly than the latter. Some of the possible
convergence curves for incompatible elements are qualitatively compared to a
compatible element in Figure 5-10.

Figure 5-10 Schematic convergence of three hypothetical incompatible elements in

comparison to a compatible element.

Selection of the Order of the Polynomial

In selecting the order of a polynomial displacement model, the completeness and
compatibility requirement considerations; but they are not the only factors we must tae
into account. An additional consideration in the selection of the order of the model is that
the pattern should be independent of the orientation of the local coordinate system. This
property of the model is known as geometric isotropy, spatial isotropy, or geometric
invariance. For polynomials of linear order, the isotropy requirement is usually
equivalent to the necessity of including constant strain states. For higher order patterns,
we can see intuitively that it is undesirable to have a preferential coordinate direction, in
other words to have displacements shape which change with a change in local coordinate
system. Experience confirms this intuition. One way to envision the achievement of
isotropy is to consider the Pascal triangle for the variable terms of the two-dimensional
polynomial equation (5.2).

(5.4)

For the two-dimensional approach, we should not include any terms from one side of the
axis of symmetry of the triangle without including its counterpart from the other side. For
example, if we wish to construct a cubic model with eight terms, the following are
geometrically isotropic: (1) all the constant, linear, and quadratic terms plus the x 2y and
xy2 terms.
Immediately below, we shall define element degrees of freedom and discuss their
selection. Meanwhile the final consideration in the section of the displacement
polynomial is that the total number of generalized coordinates for an element must be
equal to or greater than the number of joint or external degrees of freedom of the element.
The usual procedure employs the same number of generalized coordinates at the degree
of freedom. It is possible to utilize an excess of generalized coordinates to improve the
element stiffness matrix, that is, to make it less stiff or more flexible. These excess
coordinates are generally associated with internal nodes and improve the approximation
of equilibrium within the element. However, they do not improve inter element
equilibrium. Therefore, more tan a few of such extra coordinates are rarely justifiable; in
fact, their inclusion may be detrimental on some cases.

Nodal Degrees of Freedom

The nodal displacements, rotation, and/or strains necessary to specify completely the
deformation of the finite element are the degrees of freedom of the element. The degrees
of freedom differ from the generalized coordinates in that each is specifically identified
with the single nodal point and represents a displacement (or rotation or strain) having a
clear physical interpretation. We shall distinguish between degrees of freedom occurring
at external and internal nodes by referring to them as joint or nodal degrees of freedom
and internal degrees of freedom, respectively. In the formulation of individual element
properties, however, we need not make this distinction; it is only during the assembly
process that the difference becomes important.
The minimum number of degrees of freedom (or generalized coordinates) necessary for a
given element is determined by the completeness requirements for convergence, the
requirement of geometric isotropy, and the necessity of an adequate representation of the
terms in the potential energy functional. Additional degrees of freedom beyond the
minimum number may be included for element by adding secondary external nodes or by
specifying as degrees of freedom higher order derivatives of displacements at the primary
nodes. The latter approach is preferred because it leads to more compact numerical
formulation in this assembly process. Elements with additional degrees of freedom are
called higher order elements. The more such additional degrees of freedom are added, the
more flexible becomes the element stiffness for an element of given size. However this
improvement in the upper bound to the stiffness is purchased at the price of an
increasingly complex formulation of the individual element properties. The nature of this
tradeoff will be discussed more fully in the next chapter.
We may now relate the degrees of freedom and the generalized coordinates by employing
the displacement model. We can evaluate the displacements at the nodes by substituting
the nodal coordinates into the model. For example, using a model of the form given by
the first equations in equations (5.1 b), (5.2 b), and (5.3 b), we may write:

(5.5)

(5.6)

Here M is the total number of nodes for the element being considered, {q} is the vector of
nodal displacements, and the notation in parentheses indicates that the dependent variable
are assigned their value at the particular node. We may invert equation (5.6) to give

{} = [A-1]{q} (5.7)

where [A-1] is a displacement transformation matrix. Note that [A] is a square matrix;
hence, the total number of generalizes coordinates equals the total number of joint and
internal degrees of freedom. When equation (5.7) is substitute into equation (5.5), we can
eliminate the generalized coordinates and obtain

{u} = [][A-1]{q} = [N]{q} (5.8)

Equation (5.8) expressed the displacements {u} at any point P (Fig. 5-5) within the
element in terms of the displacement of the nodes, {q}.
One limitation of generalized coordinates displacements models is that it is not always
possible to obtain the inverse of matrix [A]. The interpolation function representation of
displacement models discussed in the next section avoids this difficulty.
We shall now illustrate the material covered in this section by applying it to the two
sample problems.