You are on page 1of 30

Chapter 16

Finite-element method

16.1 Introduction
The finite-element method is widely used in structural analysis. The method is also used in a wide
range of physical problems 1 including heat transfer, seepage, flow of fluids, and electrical and
magnetic potential. In the finite-element method, a continuum is idealized as an assemblage of
finite elements with specified nodes. The infinite number of degrees of freedom of the continuum
is replaced by specified unknowns at the nodes.
In essence, the analysis of a structure by the finite-element method is an application of
the displacement method. In frames, trusses, and grids, the elements are bars connected
at the nodes; these elements are considered to be one-dimensional. Two-dimensional or
three-dimensional finite elements are used in the analysis of walls, slabs, shells, and mass struc-
tures. The finite elements can have many shapes with nodes at the corners or on the sides
(Figure 16.1). The unknown displacements are nodal translations or rotations or derivatives of
these.
The use of a computer is essential in the finite-element method because of the large number
of degrees of freedom commonly involved. Chapters 21 to 23 discuss the computer analysis of
structures; the approach there applies to structures which are composed of finite elements of
any type. However, the matrices for individual elements, given explicitly in earlier chapters
(e.g. Eqs. 6.6 to 6. 7) and in Chapters 21 to 23, are for one-dimensional bar elements. The
present chapter is mainly concerned with the generation of matrices for finite elements other
than bars. The element matrices required are: the stiffness matrix, relating nodal forces to nodal
displacements; the stress matrix, relating the stress or internal forces at any point within an
element to its nodal displacements; and vectors of restraining nodal forces for use when the
external forces are applied away from the nodes or when the element is subjected to temperature
variation.
For finite elements other than bars, "exact" element matrices cannot be generated. The dis-
placements (e.g. u and v) within an element are expressed in terms of the nodal displacement.
Assumed displacement fields (e.g. a polynomial in x and y) are used. The corresponding strains
are determined by differentiation and the stress by using Hooke's law. Use of the principle of
virtual work or minimization of the total potential energy (Sections 7.5, 7.6 and 9.6) with respect
to the element nodal displacements gives the desired element matrices.
Use of displacement fields as described above is similar to the Rayleigh-Ritz method presented
in Section 10.7, where the deflection of a beam is represented by a series of assumed functions
with indeterminate parameters. The unknown parameters are derived by the principle of virtual
work. They can also be derived by minimization of total potential energy. In the finite-element

1 See Zienkiewicz, O.C. and Taylor, R.L.,The Finite Element Method, 6th ed., McGraw-Hill, London, 2005.
480 Finite-element method

~
A
~ x 2
u c [ l QLC3
_ 4 3 u

t
V
I
y V
t l;=av
ax V

Plate subjected to in-plane forces

0=aw
ay
X

0 =-aw
Y ax
Plate bending

~:
z

z
u

w
w
""u V

w
Solids

Figure 16.1 Examples of finite elements with nodal degrees of freedom shown at a typical node only.

formulation, using assumed displacement fields, the indeterminate parameters are the nodal
displacements. As in the Rayleigh-Ritz method, the solution obtained by the finite-element
method is approximate; however, convergence to the exact solution is achieved as the num-
ber of unknown parameters is increased. In other words, when a finer finite-element mesh is
used, more unknown displacements are involved and a greater accuracy is achieved.
The use of smaller elements for convergence is not required when the true displacement shapes
can be derived and used to generate the element matrices. This is the case with prismatic bar
elements.
This chapter gives the general procedures to derive matrices for finite elements of any type.
This is explained by reference to bar elements and to plate elements subjected to in-plane forces
or to bending. Other element types are discussed in Chapter 17.
Finite-element method 481

It is most common to derive the matrices for finite elements using assumed displacement fields,
and this method will be mainly used in this chapter. It is also possible to generate the element
matrices using assumed fields for displacements and for stresses (or strains): this results in what
are known as hybrid elements. The unknowns will then be the stress (or strain) parameters,
in addition to the nodal displacements. However, all the unknown parameters other than the
nodal displacements are eliminated at the element level; the global system of equations for the
structure involves nodal displacements only. When the same number of equations is solved,
a more accurate analysis is obtained with hybrid elements compared with the analysis using
displacement-based finite elements. Thus, the use of hybrid elements represents a convenient
practical approach. Section 17.13 discusses the generation of stiffness and stress matrices for
hybrid elements.

16.2 Application of the five steps of displacement method


In essence, the analysis of a structure by the finite-element method is an application of the five
steps of the displacement method summarized in Section 5.6. This analysis is explained below by
reference to a plate subjected to in-plane forces (Figure 16.2) which is idealized as an assemblage
of rectangular finite elements. Each element has four nodes with two degrees of freedom per node,
that is, translations u and v in the x and y directions respectively. The purpose of the analysis
is to determine the stress components {a}= {ax, ay, Txy} at 0, the center of each element. The
external loads are nodal forces {Fx, Fy }; at any node i, and body forces with intensities per unit
volume of {Px, Py }m distributed over any element m. Other loadings can also be considered,
such as the effects of temperature variation or of shrinkage (or swelling).

External
F forces at a
I Globa{ Y typical node

-
axes
i
y

l
Fx
-----------
y
V

Typical
I t--
- Opening
degrees of
freedom
at a node
Typical element

Pr
r Px----
;--,._
Body forces
per unit
volume

Stresses at 0

/ '///// '///// ////'///////////////'///

Figure 16.2 Example of a finite-element model for the analysis of stresses in a wall subjected to in-plane
forces.
482 Finite-element method

The five steps in the analysis are as follows:

Step I Define the unknown degrees of freedom by two coordinates u and v at each node. The
actions to be determined for any element mare {A}m = {a}m = {ax,ay, Txy}m-

Step 2 With the loading applied, determine the restraining forces {F} to prevent the displace-
ments at all coordinates. Also, for any element, determine {A,}m = {a,}m, which represents the
values of the actions (the stresses) with the nodal displacements prevented.
The stresses {a,} are produced only when effects of temperature are considered; {ar} due to
body forces is commonly ignored.
The vector {F} is considered equal to the sum of two vectors:

{F} ={Fa}+ {f&} (16.1)

The vector {Fa} is composed of the external nodal forces reversed in sign; {F&} is generated by
assemblage of {Fb}m for individual elements. The vector {Fb}m is composed of forces at the nodes
of element min equilibrium with the external forces on the element body away from the nodes;
in the case of temperature variation, {Fb}m represents a system of nodal forces in equilibrium
producing stresses {a,}. (Eq. 6.43 gives the nodal forces due to temperature variation for a
member of a plane frame; for other finite elements, see Section 16.6.1.)

Step 3 Generate the structure stiffness matrix [S] by assemblage of the stiffness matrices [S]m of
individual elements. Also, generate [Au]m = [au]m, which represents the stress components at 0
in any element due to unit displacement introduced separately at the element nodal coordinates.
For the example considered in Figure 16.2, [au]m will be a 3 x 8 matrix.

Step 4 Solve the equilibrium equations

[S] {D} = - {F} (16.2)

This gives the structure nodal displacements {D}. In the example considered, the number of
elements in {D} is twice the number of nodes.

Step 5 Calculate the required stress components for each element:

(16.3)

or

(16.4)

The values {D}m are the nodal displacements for the element m; in the example considered
(Figure 16.2), {D}m has eight values (subset of the structure displacement vector {D}).

Ignoring {a,} caused by body forces (steps 2 and 5) produces an error which diminishes as the
size of the finite elements is reduced. However, when the elements are bars (in framed structures),
{a,}= {A,), and the other matrices for individual members can be determined exactly; for this
reason, {a,} is commonly not ignored and the exact answers can be obtained without the need
to reduce the size of the elements for convergence.
Assemblage of the structure load vector and of the stiffness matrix may be done by Eqs. 21.34
and 21.31. The nonzero elements of [S] are generally limited to a band adjacent to the diagonal.
Finite-element method 483

This property, combined with the symmetry of [S], is used to conserve computer storage and to
reduce the number of computations. These topics and the methods of solution of Eq. 16.2 to
satisfy displacement constraints are discussed in Chapters 21 and 22. Examples of displacement
constraints are a zero or a prescribed value for the displacement at a support.

16.3 Basic equations of elasticity


The stresses and strains in an elastic body are related by Hooke's law, which can be written in
the generalized form

{a}= (d] {E} (16.5)

where {a} and {E} are generalized stress and strain vectors respectively, and [d] is a square
symmetrical matrix referred to as the elasticity matrix.
The strain components are defined as derivatives of the displacement component by the
generalized equation

kl= [aJ {f} (16.6)

where [a] is a matrix of the differential operator, and {/} is a vector of functions describing the
displacement field.
The symbols {a} and {E} will be used to represent stress or strain components in one-, two-, or
three-dimensional bodies. The displacement field {/} will have one, two, and three components:
u, v, and win the direction of orthogonal axes x, y, and z. The differential operator matrix [a]
will represent derivatives with respect to one, two, or three of the variables x, y, and z.
In a bar subjected to an axial force (Figure 7.5a), each of {a) and {E} has one component and [d]
has one element equal to E, the modulus of elasticity. The strain Eis equal to du/ dx, where u is dis-
placement along the beam axis and xis distance measured in the same direction. Thus, we can use
Eqs. 16.5 and 16.6 for the uniaxial stress state, with the symbols having the following meanings:

(16.7)

We shall also use the symbol {a} to represent a vector of stress resultants. For the bar considered
above, we can take {a}= N, the axial force on the bar cross section, and [d] = Ea, where a is
the cross-sectional area. Again, Eqs. 16.5 and 16.6 apply, with the symbols having the following
meanmgs:

(16.8)

The generalized Eqs. 16.5 and 16.6 apply to a bar in bending (Figure 6.5b), with the symbols
having the following meanings:

(16.9)

where M is the bending moment, ifr is the curvature, I is the second moment of area about the
centroidal axis, and v is the displacement in the y direction.
The product {a)T {E} integrated over the volume of an element appears in the strain energy
Eq. 7.14 and in the virtual work Eq. 7.46, both of which will be frequently used. When {a}
represents stress resultants over a cross section of a bar, the integral over the volume has to be
replaced by an integral over the length (see Eq. 7.32). For plates in bending, we shall use {a} to
represent bending and twisting moments {Mx,My,Mxy} and the integral will be over the area.
484 Finite-element method

In the following subsections we shall apply the generalized Eqs. 16.5 and 16.6 in three stress
states.

16.3. I Plane stress and plane strain


Consider a plate subjected to in-plane forces (Figure 16.13). At any point, the stress, strain, and
displacement components are

{a}={ax,ay,Txy} {E}={Ex,Ey,Yxy} {f}={u,v} (16.10)

The strains are defined as derivatives of {f} by the generalized Eq. 16.6, with the differential
operator matrix

a;ax
[B] = 0 (16.11)
[ a;ay

The stress and strain vectors are related by generalized Hooke's law (Eq. 16.5) with the elasticity
matrix [d] given by one of the Eqs. 16.12 or 16.13.
When strain in the z direction is free to occur, O"z = 0 and we have the state of plane stress.
Deep beams and shear walls are examples of structures in a state of plane stress. When strain in
the z direction cannot occur, Ez = 0 and we have the state of plane strain. The state of plane strain
occurs in structures which have a constant cross section perpendicular to the z direction and also
have the dimension in the z direction much larger than those in the x and y directions. Concrete
gravity dams and earth embankments are examples of structures in this category. The analysis
of these structures may be performed for a slice of unit thickness in a state of plane strain.
For an isotropic material, the elasticity matrix in a plane-stress state is

[d] =
E
1-vl
[1
~ 1
V
(16.12)
0

where Eis the modulus of elasticity in tension or in compression, and vis Poisson's ratio.
The elasticity matrix for a plane-strain state is

1 v/(1-v)
d - E(l - v) v 1- v 1 (16.13)
[ ]-(1+v)(1-2v) [ /(O )
0

Equation 16.12 can be derived from Eq. 6.6 or by inversion of the square matrix in Eq. 15.53.
Equation 16.13 can also be derived from Eq. 7.8 by setting Ez = 0 (in addition to 'xz = Tyz = 0).
The same equations give the normal stress in the z direction in the plane-strain state

(16.14)

16.3.2 Bending of plates


For a plate in bending (Figure 15.14), the generalized stress and strain vectors are defined as

{er}= {Mx,My,Mxy} (16.15)


Finite-element method 485

and

(16.16)

One component of body force and one component of displacement exist:

{P}={q} (16.17)
{f} = {w} (16.18)

where q is force in the z direction per unit area and w is deflection in the same direction. The
generalized Eqs. 16.5 and 16.6 apply to a plate in bending, with

-a 2 ;ax2 )
[aJ
I -a 2 ;ay2
2a 2 / (axay)

For an orthotropic plate in bending, the elasticity matrix is


(16.19)

vxEy/ (1-vxvy)
Ey/ (1-vxvy) (16.20)
0

where Ex, Ey are moduli of elasticity in tension or in compression in the x and y directions; vx
and Vy are Poisson's ratios; G is shear modulus of elasticity; and h is the plate thickness. The
shear modulus of elasticity

G=/ExEy/[2(1 + ~ ) ]

When the plate is isotropic, we set E =Ex= Ey and v = Vx = Vy in Eq. 16.20:

d -
Eh3
[ ] - 12 (1 - v2)
[1
V
v
l
O
(16.21)
0

It can be noted that, for a plate in bending, the generalized Eq. 16.5 is simply a condensed form
of Eq. 15.64.

16.3.3 Three-dimensional solid


For a three-dimensional body, the generalized Eqs. 16.5 and 16.6 apply again, with the vectors
{p} and {f} having the following meaning:

{P} = {Px,Py,Pz} (16.22)


{f} = {u, v, w} (16.23)
486 Finite-element method

where {Px,Py,Pz} are body forces per unit volume, and {u, v, w) are translations in the x, y, and
z directions (Figure 7.4). The stress and strain vectors {u} and {E} are defined by Eqs. 7.9 and
7.10, and the [d] matrix is given by Eq. 7.13. The differential operator is

0 a;ay a;az
0
a;ay o a;ax 0 (16.24)
o a;az 0 a;ax

16.4 Displacement interpolation


In the derivation of element matrices, interpolation functions are required to define the deformed
shape of the element. For a finite element of any type, the displacement at any point within the
element can be related to the nodal displacement by the equation

{f}=[L]{D*} (16.25)

where {fl is a vector of displacement components at any point; {D*} is a vector of nodal dis-
placements; and [L] is a matrix of functions of coordinates defining the position of the point
considered within the element (e.g. x and y or~ and 1/ in two-dimensional finite elements).
The interpolation functions [L] are also called shape functions; they describe the deformed
shape of the element due to unit displacements introduced separately at each coordinate. Any
interpolation function L; represents the deformed shape when D'! = 1 while the other nodal
displacements are zero.
The accuracy of a finite element depends upon the choice of the shape functions. These should
satisfy conditions which will ensure convergence to correct answers when a finer finite-element
mesh is used. The derivation of the shape functions and the conditions which should be satisfied
are discussed in Sections 16.8 and 16.9. Examples of shape functions with various types of
elements are given below.

16.4. I Straight bar element


For the axial deformation of a bar (Figure 16.3), {fl= {u) is the translation in the x direction
at any section, and {D*} = {u1, u2} is the translation at the two ends. The matrix [L] may be
composed of two linear interpolation functions:

(16.26)

where ~ = x / l, and ! is the bar length.

_____..
I: X
·I ·1 2*

1* ------and
X
u
~=xii

~ l

Figure 16.3 Linear interpolation functions. Shape functions for a bar subjected to an axial force.
Finite-element method 487

~
2 *y( ·;-1-*-------+-+--
--
·+1~4*
x and ~ = xi I

and I 3*
V

1c==:-----=-

L3= e(3-2~)

L4= te(~-1)

Figure 16.4 Shape functions for deflection of a bar in bending.

The shape functions which can be used for a bar in bending are given in Figure 16.4. For
this element, ((} = {v} is the transverse deflection in the y direction at any section; the nodal
displacements are defined as

{D*} = {Vx=O, ( ~:) x=O, Vx=l, ( ~:) x=l }


(16.27)

The shape functions [L] can be the four cubic polynomials

(16.28)

Each shape function Li in Eqs. 16.26 and 16.28 satisfies the requirement that D7 = 1 with other
displacements zero. This requirement is sufficient to derive the functions (see Section 16.8).
The shape functions in Eqs. 16.26 and 16.28 correspond to the true deformed shapes of a
prismatic bar (with shear deformation ignored). The same shape functions may be used to derive
matrices for nonprismatic bars (see Examples 16.4 and 16.5).
We can recognize that the shape functions in Figures 16.3 and 16.4 are the same as the influence
lines for the nodal forces, reversed in sign. (Compare, for example, L2 and L 4 in Figure 16.4
with the influence lines of the member end-moments in Figure 12.7d.)
The shape functions L1 in Figure 16.4 can be considered to be equal to the sum of the straight
lines 1 - t and (L2 + L4)//; the latter term represents the deflected shape corresponding to
clockwise rotations, each equal to 1/l at the two ends. By similar reasoning, we can verify that
L3 =t-(L2 +L4)/l.

16.4.2 Quadrilateral element subjected to in-plane forces


Figure 16.5 shows a quadrilateral element with corner nodes and two degrees of freedom per
node. The element may be used in plane-stress and plane-strain analyses. In this case, the symbols
in the generalized Eq. 16.25 have the following meaning:

[f] = {u, v} (16.29)


488 Finite-element method

1 L; = l/4(1+EE;) (l+ryr1;)
i=l,2,3,4
y 1

4
V

Nodal 1
{ displacements
at a node 2

I
I
~=112 3
Shape function L 2

Figure 16.5 Plane-stress or plane-strain quadrilateral element. Natural coordinates ~ and 7/ define the
location of any point. Pictorial view of a shape function (hyperbolic paraboloid).

where u and v are translations in the x and y directions, and

(16.30)

where u; and v; are translations at node i in the x and y directions. The [L] matrix for the element
in Figure 16.5 is

[L]-[
- L1
0 (16.31)

The function L; is a sum of bilinear functions in the natural coordinates ~ and 11, defined in
Figure 16.5. The value of L; is unity at node i and zero at the other three nodes; any of the four
shape functions may be expressed as

1
4 (1 + ~~;) (1 + 1111;)
L; = with i = l,2, 3,4 (16.32)

If the value of L; is plotted perpendicular to the surface of the element, a hypersurface is obtained.
Along the lines~= constant or 11 = constant, the surface follows straight lines. The shape function
L2 is plotted in pictorial view in Figure 16.5; at node 2, ~i = 1 and 11i = -l, and the function L2
defining the hypersurface is obtained by substitution of the two values in Eq. 16.32.
Equation 16.32 represents one of a family of functions used for interpolation in the isopara-
metric elements, discussed in Section 17.2. We should note that the sum of the values of the four
L; functions at any point is unity.

16.4.3 Rectangular plate-bending element


The rectangular element in Figure 16.6, used in the analysis of plates in bending, has twelve
degrees of freedom (three at each corner), defined as

(16.33)
Finite-element method 489

b
~=xlb T/ = y/c
z (down) 1
~, x,~
-~------.-.
1 2

(dow~)r0x=~; 4
C

0 =-aw
<
y dX
4 3
---- Typical nodal Shape function L4
displacements

1 1

3
Shape function Ls Shape function L 6

Figure 16.6 Plate-bending element.

Here, the displacement f =w is the deflection at any point. The rotations 0 are treated as
derivatives of w:

0x = aw/ay 0y = -aw;ax (16.34)

The shape functions for the rectangular bending element (Figure 16.6) are

[L] = [ (1-~) (1-rJ)- b


1 (L3 1 (L2 +L11)
+L6) + ~ 1
er, (rJ - 1) 2 (1 - ~) 2
-b~ (~ - 1) 2 (1 - rJ) 3
1 1
~ (1 - r,) + b (L3 + L6) +~(Ls+ Ls) 4
CT] (rJ - 1)2 ~ 5
-b~ 2 (~ - 1) (1 - rJ) 6
1 1 (16.35)
~rJ + b (L9 + L12) - ~(Ls+ Ls) 7
CrJ 2 (rJ - 1) ~ 8
-b~ 2 (~ - 1) rJ 9
1 1
(1 - n rJ - b (L9 + L12) - ~ (L2+L11) 10
CrJ 2 (rJ - 1) (1 - ~) 11
2 12
-bn~ - 1) rJ]
'l
490 Finite-element method

where b and c are lengths of element sides; I; = x / b and 17 = y / c. The functions L2, L3, Ls, L6,
Ls, L9, L11, L12 are shape functions given explicitly in Eq. 16.35 on lines 2, 3, 5, 6, 8, 9, 11,
and 12 respectively; they correspond to unit rotations. Pictorial views of three deflected shapes,
L4, Ls, and L6, corresponding to unit displacements at node 2, are included in Figure 16.6. It
can be seen that L 6 is zero along three edges, while along the fourth edge (1-2) the function is
the same as the shape function for a beam (compare with L4 in Figure 16.4, reversed in sign).
Along any line I;= constant, L6 varies linearly, as shown in Figure 16.6. Similarly, Ls has the
same shape as a deflected bar along the edge 3-2 and varies linearly along any line 17 = constant.
The shape functions L1, L4, L7, L10, corresponding to w = l at the corners, are expressed
as the sum of bilinear shape functions (similar to the one shown in Figure 16.5) and the shape
functions corresponding to rotations at the nodes equal to ±(1/b) or ±(1/c). Each of the four
functions has values along two edges which are the same as the shape functions of a bar (L1 or
L3 in Figure 16.4 ). At any point, the sum L1 + L4 + L7 + L10 is equal to unity.
We should note that the deflected surface defined by any of the twelve shape functions does
not generally have a zero slope normal to the element edges. This can produce incompatibility of
slopes in adjacent elements; the effects of these incompatibilities will be discussed in Section 16.9.

16.5 Stiffness and stress matrices for displacement-based


elements
Equation 16.25 defines the element displacement field in terms of the nodal displacement. By
appropriate differentiation, we can derive the strain (Eq. 16.6):

{f} = [L]{D*} (16.36)


{E} = [a] [L] {D*} (16.37)

Thus, the strain at any point in a displacement-based element is

{E} = [Bl{D*} (16.38)

where

[B] = [a][L] (16.39)

The matrix [BJ may be referred to as the nodal displacement-strain transformation matrix.
Any column j of [B] represents the strain components {Euj} due to u;
= 1. Use of Hooke's law
(Eq. 16.5) gives the stress at any point in a displacement-based element:

{a}=[d)[B]{D*} (16.40)

or

{a}= [a]u {D*} (16.41)

where [au] is the stress matrix for the element:

[au]= [d] [B] (16.42)

The elements of any column j of [au] are the stress components at any point due to D1= 1.
Finite-element method 491

An element S7; of the stiffness matrix is the force at coordinate i corresponding to unit
displacement at j; S7; can be determined by the unit-displacement theorem (Eq. 7.47):

(16.43)

where {eruil represents the "actual" stresses at any point due to unit displacement at j; {Eu;}
represents the strains at the same point corresponding to unit virtual displacement at i; and
dv is an elemental volume. 2 The integral over the volume is replaced by an integral over the
length in the case of a bar and over the area in the case of a plate. For this purpose, the symbols
{er} and {E} in Eq. 16.43 represent generalized stress and strain respectively. For example, in a
bar, {er} represents internal forces at a section; in a plate in bending, {E} represents curvatures
(Eq. 16.16).
Using the shape functions [L] to determine the actual stresses and the virtual strains via
Eqs. 16.39 and 16.42, substitution in Eq. 16.43 gives any element of the stiffness matrix:

S7; = 1{B}f [d] {B}; v (16.44)

where {B}; and {B}i are the ith and jth columns of [B]. The stiffness matrix of a finite element is
given by

[S*] 1 [Bf [d] [B]dv (16.45)

In Eqs. 16.43 and 16.44 we are accepting an assumed displacement field, namely Li, as actual.
However, in general, the assumed shape is different from the actual. What we are doing then is
tantamount to imposing the assumed configuration by the application of small distributed forces
on the element body in addition to the nodal forces. The distributed forces have the effect of
changing the actual configuration to the assumed shape; these forces, not accounted for, cause
the stiffness calculated by Eq. 16.45 to be an overestimate. In other words, a finite-element
analysis in which the element stiffness matrices are derived by the above procedure is expected
to give smaller displacements than the actual ones.

16.6 Element load vectors


The vector of restraining forces, to be used in the equilibrium Eq. 16.2, includes a component
{F&} representing the forces in equilibrium with the external loads applied on the body of the
elements away from the nodes (Eq. 16.1). Considering a single element, the equilibrant at node
j to the body forces can be determined by

(16.46)

where {p} represents the magnitudes per unit volume of forces applied in the same directions as
the displacements {(}.
Equation 16.46 can be explained by the principle of virtual work (Eq. 7.46) and also, suc-
cinctly, by Betti's theorem (Section 9.2). Here, the body forces and the nodal equilibrants form

2 The symbol dv representing elemental volume should not be confused with v, which represents a
translational displacement.
492 Finite-element method

one system; the element subjected only to those nodal forces which produce the displacement
configuration L; represents the second system. According to Betti's theorem, the work of the
forces of the first system during displacements by the second system is equal to the work of
the second system during displacements by the first system. Now, the second quantity is zero
because the second system has forces at the nodes only, and the nodal displacements in the first
system are all zero.
By the use of Eq. 16.46, we are treating the shape function L; as the influence line (or influence
surface) of the nodal force at j, reversed in sign (see Section 12.3 ). We should remember that an
approximation is involved in Eq. 16.46 by the acceptance of an assumed deflected shape as the
actual displacement field.
The vector of nodal forces in equilibrium with the forces applied on the element away from
the nodes is

{Fi;}= -1 [Lf {p}dv (16.47)

This vector is referred to as the element consistent load vector because the same shape functions
[L] are used to generate [S*] and {Fb). The superscript* is used here to refer to local coordinates
of an individual element.
When the external forces are applied to the surface of the element, the integral in Eq. 16.47
should be taken over the area of the element. When concentrated forces act, the integ-
ral is replaced by a summation of the forces multiplied by the values of [L]T at the load
positions.

16.6. I Analysis of effects of temperature variation


When an element is subjected to temperature variation (or to shrinkage), with the displacements
restrained, the stresses at any point are given by (Eq. 16.5)

{a,}= -[d] ko} (16.48)

where {Eo} represents the strains which would exist if the change in volume were free to occur. In
a two-dimensional plane-stress or plane-strain state, a rise of temperature of T degrees produces
the free strain

(16.49)

where a is the coefficient of thermal expansion.


For an element subjected to volume change, the consistent vector of restraining forces is
given by

(16.50)

Again, the unit-displacement theory may be used to derive Eq. 16.50. With the actual stress
being {a,}= -[d] {Eo} and the virtual strain being {Eu;}= {B};, Eq. 7.50 gives the jth element of
the consistent load vector.
In most cases, the integrals involved in generating the stiffness matrix and the load vectors for
individual elements are evaluated numerically using Gaussian quadrature (see Section 17.9).
Finite-element method 493

16.7 Derivation of element matrices by minimization of total


potential energy
The element stiffness matrix [S*] and the consistent load vector {Fb} (Eqs. 16.45 and 16.47)
can be derived by the principle of total potential energy (Section 9.6). Consider a finite element
subjected to body forces {p} and nodal forces {Q* }. The total potential energy is defined as the
sum of potential energy and strain energy (Eq. 9.27):

(16.51)

where {f} are displacement components at any point; {p} are body forces per unit volume applied
in the same directions as {f}; {er} are stresses; {E} are strains; and {D*} are nodal displacements.
Substitution of Eqs. 16.36, 16.38, and 16.40 in the above equation gives

<I>= - {D*}T {Q*} - f {D*}T [L]T {p} dv

+!
2
f
V
{D*}T [B]T [d] [B] {D*} dv (16.52)

The principle of minimum total potential energy can be expressed as (Eq. 9.30)

(16.53)

where the subscript i refers to any of the nodal displacements. Partial differentiation 3 with
respect to each nodal displacement gives

~ =-{Q*}-/ [L]T {p} dv


a{D } "
+f [B]T [d] [B] {D*} dv=O
"
(16.54)

Equation 16.54 can be rewritten in the form

[S*) {D*} = - {F*} (16.55)

where {F*} are nodal forces which would prevent the nodal displacements. The restraining forces
are the sum of the nodal forces {Q*} in a reversed direction and of the nodal equilibrants of the
body forces. Thus,

{F*} = {F;} + {h} (16.56)

where {F;} = -{Q*}, and {Fb} is the element consistent load vector. The matrix [S*] is the element
stiffness matrix. Combining Eqs. 16.55 and 16.56 and comparing with Eq. 16.54, we obtain,
by analogy, Eqs. 16.47 and 16.45.

3 It can be shown that if a scalar quantity y is expressed as a sum of products of matrices y= (1/2){x)T[a] {x) +
{x)T{b), where [a] and {b) are constants and [a] is symmetrical, differentiation of y with respect to x, for
i = l,2, ... gives ay;a{x) = {ay;ax1, ay;ax2, .. . ) = [a]{x) + {b).
494 Finite-element method

16.8 Derivation of shape functions


The displacement field {f} may be expressed as polynomials of the coordinates x and y (or
~ and rJ) defining the position of any point. For example, the deflection w in a plate-bending
element or the translations u and v in a plane-stress or a plane-strain element may be expressed as

f (x,y) = [ 1,x,y,x2 , .. . ] {A}= [P] {A} (16.57)

where {A} is a vector of constants, yet to be determined; and [P] is a matrix of polynomial
terms, the number of which equals the number of nodal degrees of freedom. Pascal's triangle
(Figure 16.7) can be used to select the polynomial terms to be included in [P]. In general, the
lower-degree terms are used. Examples of the polynomial terms used in several elements are
given later in this section.
The nodal displacements {D*} can be related to the constants {A} by substituting x and y (or
~ and ri) values at the nodes in Eq. 16.57 (or its derivatives). This gives

{D*} = [C]{A} (16.58)

The elements of [C] are known values depending upon (x;,y;), with i = 1,2, ... referring to
the node numbers. The undetermined constants {A} can now be expressed in terms of {D*} by
mvers1on:

{A}= [C]- 1 {D*} (16.59)

Substituting Eq. 16.59 into Eq. 16.57, and by analogy of the resulting equation with Eq. 16.36,
we obtain the shape functions

[L] = [P][C]- 1 (16.60)

As an example of polynomial selection, let us consider the bar element shown in Figure 16.3,
which is subjected to an axial force. With two degrees of freedom, [P] has only two terms:

[P] = [1 ~] (16.61)

For a bar in bending (Figure 16.4),

(16.62)

X y

x2 xy y2

x3 x2y xy2 y3

x4 x3y x2y2 xy3 y4

Figure 16. 7 Pascal's triangle.


Finite-element method 495

For a plate element subjected to in-plane forces (Figure 16.5), each of the displacements u and
v is associated with four nodal displacements. A polynomial with four terms is used for each
of u and v:

(16.63)

For the plate-bending element in Figure 16.6, with twelve degrees of freedom, we express the
deflection as w = [P] {A}, with

(16.64)

For elements with two or more variables, such as x and y (or x, y, and z), the terms included
in [P] should be invariant if the reference axes x and y (or x, y, and z) are interchanged. Thus,
in Eq. 16.63 we should not replace the terms ~T/ by t 2 or by T/ 2 • Similarly, in Eq. 16.64 we
should not replace ~3 17 or tTJ 3 by ~2 TJ 2 • In other words, [P] includes symmetrical terms from
Pascal's triangle (Figure 16.7). When this requirement is satisfied, the element does not have a
"preferred" direction. In consequence, the use of such an element in the analysis of the structure
shown in Figure 16.2 will give the same answers regardless of whether the global axes x and y
are as shown or are rotated through 90° so that x becomes vertical.
The invariance requirement is relaxed for the rectangular plate element subjected to in-plane
forces included in Figure 16.1. The element has four nodes with three nodal displacements per
node: u, v, av/ax. The inclusion of av/ax (but not of other derivatives of u and v) makes the
element behave differently in the x and y directions. The polynomials used for u and for v
are different:

u(x,y) =[l x y xy]lAJ 11 (16.65)

v(x,y)=[] x y x 1 xy x 3 x 2y x 3 y]lAlv (16.66)

The variation of v in the x direction is cubic, while a linear variation is used for v in the y
direction and for u in both the x and y directions. The shape functions for the element are given
in Eq. 16.67 (see Example 16.3).
This element4 gives excellent accuracy when used for structures which have beam-like beha-
vior, e.g. folded plates and box girders. For this use, the element local x axis must be in the
direction of the "beam". For comparison of accuracy of results of this element with other
elements, see Prob. 17.18.

Example 16. I: Beam in flexure


Derive the shape functions for the bar element in Figure 16.4. The de£lection vis expressed
as a cubic polynomial oft, where ~=x/1 (Eqs. 16.57 and 16.62).

v =[ 1 ~ ~2 ~
3
] {A}= [P] IA}

4 The rectangular element considered here is a special case of the quadrilateral element referred to as
QLC3. See Sisodiya, R., Cheung, Y. K. and Ghali, A., "New Finite Element with Application ro Box
Girder Bridges," Proceedings, The Institution of Civil Engineers, London, Supplement 1972, Paper 7495,
pp. 207-225. The combination of QLC3 with the rectangular bending element shown io Figure J 6.6
gives a good shell element with three translations and three rotations per node for tJ,e analysis of spatial
structures (See 17.10.1).
496 Finite-element method

The nodal displacements defined by Eq. 16.27 are substituted in the above equations to
give

; ] {A)

3/I

Inversion of the square matrix in the above equation gives

1 O O OJ
[C]_ 1 = 0 I O 0
[ -3 -2/ 3 -I
2 / -2 /

The product [P] [C]- 1 gives the shape functions [L] (Eq. t6.28).

Example I6.2: Quadrilateral plate subjected to in..plane forces


Derive the shape functions for the plane-stress or plane-strain quadrilateral element shown
in Figure 16.5.
We have u or v = [Pl{A}, with [P] given in Eq. 16.63. Substituting for~ and TJ by their
values at the four corners, we write for u:

:I l
U4
= [ : -:
1 -1
=: -: ]
1 -1
{A}

lnversion of [C], which is the square matrix- in this equation, gives

1 1
-1 l ll
-11 ]
[CJ- 1 = (1/4)
-1 -1 1 1
[
1 -1 1 -1

Substitution in Eq. 16.60 gives

(L1,L2,L3,L4)=¼[1-~-TJ+~TJ, l+~-TJ-$T/, l+$+11+~1J,


l-$+11-~11]

which is the same as Eq. 16.32. The same shape functions apply to v.
Finite-element method 497

Example 16.3: Rectangular element QLC3


Using the polynomials in Eq. 16.66, derive the shape functions for the displacement v in a
rectangular plate element QLC3 subjected to in-plane forces (Figure 16.1).
The nodal displacements are defined as

Using the symbols; =x/b and TJ =y/c, Eq. 16.66 can be written as 11 = [PJ{A}, with

The nodal displacements associated with v are

Substituting for~ and '7 by their values at the nodes, we can write (D:} = [CJ {A}, with

2 1 0 0 0 0 0 0 0
3 0 1/b O 0 0 0 0 0
5 1 1 0 1 0 1 0 0
6 0 1/b O 2/b O 3/b 0 0
(CJ=
8 1 l l 1 1 I 1 1
9 0 1/b O 2/b 1/b 3/b 2/b 3/b
11 1 0 1 0 0 0 0 0
12 0 1/b O O 1/b 0 0 0

Inversion of [C] and substitution in Eq. 16.60 gives the eight shape functions associated
with v: L2, L3, Ls, L6, Ls, L9, L11, L12. For reference, we give the complete shape functions
for the element as follows:

L4 =Hl -,,) Ls= ~2 (3 - 2~)(1 - 7/) L6 = b~ 2 (s -1)(1- T/)


L1=~'1 Ls =H3 - 2~) 1J L9 =hs 2 (~ -1) TJ
L10 = (1-~),, L11 = (1 - 3s2 + 2~ 3 ) 1J L 12 = b; (~ - 1)2 IJ
{16.67)

These functions can be used to express u and II by the equation (u, v} = [LJ(D•I (see
Eq. 16.25), with

L, L10 0 0 ]
0
(16.68)
0 L11 L12
498 Finite-element method

16.9 Convergence conditions


In principle, it is possible to use any continuous shape function for the displacement field of
a finite element; however, polynomials are commonly used. For monotonic convergence to the
correct answer as smaller elements are used, the shape functions should satisfy the following
three requirements:

1. The displacements of adjacent elements along a common boundary must be identical.

It can be seen that this condition is satisfied in the plane-stress or plane-strain element in
Figure 16.5 and in the plate-bending element in Figure 16.6. Along any side 1-2, the translations
u and v in Figure 16.5 or the deflection win Figure 16.6 are functions of the nodal displacements
at 1 and 2 only. It thus follows that two adjacent elements sharing nodes 1 and 2 will have the
same nodal displacements at the two nodes and the same u and v or w along the line 1-2.
In some cases, the first partial derivatives of the element should also be compatible. This
condition needs to be satisfied in plate-bending elements but not in plane-stress or plane-strain
elements.
Two adjacent elements of the type shown in Figure 16.6 have the same deflection and hence
the same slope along the common edge. However, normal to the common edge, the tangents
of the deflected surfaces of the two elements have the same slope only at the nodes. Away from
the nodes, the tangents normal to a common edge can have slopes differing by an angle a. The
quantity (1/2) f Mna ds represents, for the assembled structure, strain energy not accounted for
in the process of minimizing the total potential energy (see Eq. 16.52); here, Mn is the resultant
of stresses normal to the common edge and ds is an elemental length of the edge.
The rule to ensure convergence is that the compatibility be satisfied for [L] and its deriv-
atives of order one less than the derivatives included in [B]. For a plate-bending element, [B]
includes second derivatives (Eqs. 16.39 and 16.19): -a 2 w;ax2, -a 2 w;ay2 , 2a 2 w;axay. Thus,
compatibility is required for aw;ax and aw;ay in addition tow.
Several elements, said to be incompatible or nonconforming, such as the element in
Figure 16.6, do not satisfy the requirement of compatibility of displacement derivatives, and
yet some of these elements give excellent results. An explanation of this behavior is that the
excess stiffness, which is a characteristic of displacement-based finite elements, is compensated
by the increase in flexibility resulting from a lack of compatibility of slopes.
Nonconforming elements converge towards the correct answer when the incompatibilities
disappear as the mesh becomes finer and the strains within the element tend to be constants.

2. When the nodal displacements {D*} correspond to rigid-body motion, the strains [B] {D*}
must be equal to zero.

This condition is easily satisfied when the polynomial matrix [P] in Eq. 16.57 includes the
lower-order terms of Pascal's triangle. For example, for the plate-bending element (Figure 16.6),
inclusion of the terms 1, x, y would allow w to be represented by the equation of an inclined plane.
We can verify that the shape functions in Eq. 16.35 for the element shown in Figure 16.6
allow translation as a rigid body by setting w = 1 at the four corners and 0x = 0y = 0 at all
nodes; Eq. 16.25 will then give w = 1, representing unit downward translation. (This is because
L1 + L4 + L7 + L10 = 1, as noted earlier.) To check that the shape functions allow rigid-body
rotation, let w = 1 at nodes 1 and 2, and 0x = -1 / c at the four nodes of the element, while all
other nodal displacements are zero. It can be seen that substitution of these nodal displacements
and of Eq. 16.35 in Eq. 16.25 gives w = 1 - r,, which is the equation of the plane obtained by
rotation of the element through an angle 0x = -1 / c about the edge 4-3. In a similar way, we can
verify that the shape functions allow a rigid-body rotation 0y = constant.
Finite-element method 499

3. The shape functions must allow the element to be in a state of constant strain.
This is required because, as the elements become smaller, the strains within individual elements
tend to constants. Thus, a smooth curve (or surface) representing the strain variation can be
approximated by step variation.
This requirement will be satisfied when the polynomial [P] in Eq. 16.57 includes the lower
terms which contribute to the strain. For example, for the rectangular bending element shown in
Figure 16.6, the strains are {E} = {-a 2 w/ax 2 ,-a 2 w/ay 2 , 2a 2 w/(axay)}; the terms x 2 , xy, and
y2 of Pascal's triangle must be included in [P].
We can verify that the shape functions in Eq. 16.35 allow a constant curvature, i.e.
-a 2 w / ax2 = constant, by setting 0y = l at nodes 1 and 4, and 0y = - l at nodes 2 and 3, while the
remaining nodal displacements are equal to zero. Substitution in Eq. 16.36 gives w = b?(l - ?),
which represents the surface of a cylinder with a constant curvature of 2/b.
For the same element, 2a 2 w(axay) will be constant(= -4/bc) when the element is twisted so
that w = l at nodes 2 and 4, with the edges remaining straight. Thus, the nodal displacements
will be: w=O at 1 and 3; w= lat 2 and 4; 0x = l/c at 1 and 4; 0x = -l/c at 2 and 3; 0y =-l/b
at 1 and 2; and 0y = l/b at 3 and 4.
Requirement 2 can be considered to be a special case of requirement 3 when the constant strain
is zero. The "patch test" (Section 16.10) is a numerical method for nonconforming elements to
verify that an assemblage of elements can assume a constant-strain state.

16.10 The patch test for convergence


It is advisable to perform a patch test before a new computer program or a new element is adop-
ted. 5 The test is conducted on a patch of several elements (Figure 16.8). Prescribed displacements
are introduced at boundary nodes which correspond to constant-strain conditions. If the strains
determined by the analysis are constant, the patch test is passed. If the test is not passed, con-
vergence with an arbitrary fine mesh is not ensured. Passing the patch test for conforming or
nonconforming elements means that convergence requirements 2 and 3 in the preceding section
are satisfied.
In the patch of plane-stress or plane-strain elements shown in Figure 16.8, a constant-strain
state should exist for all elements of the patch when the displacement at the boundary nodes are
prescribed by

I· b •I• b + b •I
bt/2 bt bt bt/2

t = plate thickness
ay= -1

Figure 16.8 Model for the patch test for the plane-stress or plane-strain element of Figure 16.5.
Consistent nodal forces correspond to a uniform stress ay = -1 on the top edge of the
patch.
5 The importance of the patch test, what it achieves and the philosophy behind it are the subject of a chapter
in Irons, B. M. and Ahmad, S., Techniques of Finite Elements, Ellis Horwood, Chichester, England, and
Halsted Press (Wiley), New York, 1980 (see pp. 149-162).
500 Finite-element method

where 01 to a6 are arbitrary constants. The inner nodes should be left free. The expected strains
are (Eq. 16.6): !Ex, Ey, Yxy) = (a1, as, (a2 + a4)). When che patch test is passed, the computed
strains agree with the exact values co the limit of computer accuracy.
Alternatively, appropriate supports and consistent nodal forces (calculated by Eq. 16.47) may
be introduced to represent the state of constant stress. An example of chis is shown in Figure 16.8,
in which is represented a constant stress Uy= - 1. Other constant-stress (or constant-strain) states
and more than one geometry should be tested to guarantee convergence.

Example 16.4: Axial forces and displacements of a bar of variable cross section
Generate the stiffness matrix and the vector of restraining forces due co a uniform rise in
temperature of T degrees for the bar shown in Figure 16.3. Assume that the cross section
varies as a =ao(2-~), where ao is constant. The coefficient of thermal expansion is a and
rhe modulus of elasticity is E.
Using the shape functions of Eq. 16.26, the [Bl matrix is (Eq. 16.39)

d d 1
IBl= -
dx
[1 eJ = Id~ ll -~ ;] =-
I
l-1 lj

The elasticity matrix in this case bas one element [di=[£). Substitution in Eq. 16.45 gives
the stiffness matrix:

f' t{ 1}[E)7[-1,phantom-l]ao(2-e)de
[s·]=IJo 7 -l1
or

[S"] = Eao
/
[ 1
-1
-1 ]
I
[i~ _e 2

2
]
1

0
= 1.SEao [
/
1
-1
-1 ]
I

If the effect of the temperature change is not restrained, the strain is Eo = aT. The nodal
forces co restrain nodal displacements (Eq. 16.50) are then

The same results would be obtained if the member were treated as a prismatic bar with
a constant cross-sectional area equal to the average of the values at the two ends. The
exact answer for the stiffness matrix is the same as above with the constant 1.5 replaced
by (1/ ln2) = 1.443 (obtained by considering the true deformed shape). As expected, the
use of assumed shape functions resulted in an overestimate of stiffness.

Example 16.5: Stiffness matrix of a beam In flexure with variable cross section
Determine element Su
of the stiffness matrix for the bar shown in Figure 16.4, assuming the
second moment of the cross section to vary as J = lo(l +~),where lo is constant. Consider
bending deformations only; E =constant. Also, generate the vector of noda l equilibranrs
of a uniform load q per unit length covering the entire length.
Finite-element method SO I

For a beam in bencling, [dl = [EI], {er}= {M}, and {€} = -d2 v/dx2. Using the shape
functions in Figure 16.4, the [BJ matrix is (Eq. 16.39)

CBJ=-:: 2
2
11 -3s 2 +2s 3 LHs- 1> s 2 <3-2s> ts 2 c~-1>J
1
CBl=-r r-6+12s 1c6~- 4) 6-12s 1(16~-2>]

The required element of the stiffness matrix (Eq. 16.44) is

The exact answer can be calculated by Eq. 11.15, giving Sh= 7.72 Elo/12 . As expected,
the stiffness is overestimated by the use of the assumed shape function L2 instead of the
true deflected shape due to D 2= 1.
The entire stiffness [S*] derived by Eq. 16.45 may be compared with the exact stiffness
matrix (by Eq. 11.15) in the answers to Prob. 16.1
The nodal forces in equilibrium with the uniform load q, with nodal displacements
prevented, are (Eq. 16.47)
2 3

!hi,,,= - fa
1

l1-3s +2s2
1Hs - 1)
~2 (3 - 2s)
1; 2 (~ - 1)
] r -qi
-ql/2
qi ds = -q1;2
I 12 )
-ql2/12
These are the same forces as for a prismatic beam; again, an approximation is involved in
accepting the deflected shapes of a prismatic bar for a bar with a variable J.

Example 16.6: Stiffness matrix of a rectangular plate subjected to in-plane forces


Determine element Sii of the stiffness matrix for a rectangular plane-stress element of
constant thickness h (Figure 16.1). The shape functions for this element are derived in
Example 16.3. Determine also the stresses ac any point due to D 2= 1.
Due to D2= 1, the displacements at any point are given by (Eqs. 16.67 and 16.68) as

T he strain at any point (Eqs. 16.39 and 16.11) is

[B}i= ! (1/c){-l:3s 2 -2s 3 )


(1/b) {-6s +6s2 ) c1 - ,,>
l
The required element of the stiffness matrix is given by Eq. 16.44 as
502 Finite-element method

Substituting for [dl from Eq. 16.12 and performing the integral gives

* = 1Eh
S22 -- -
1 - v2 35
[13(b) +-51(')-b .
-
C
]
(1-v)

The stress at any point due to Di= l is (Eq. 16.42)


(a11h = (d] {Bh

= -E-
1-v 2
{ -v ( -1+3s
C
2-2s 3) ,
cl ( -1 + 3s2- 2~ 3) , 3 (1-
b v) (1 - (
I)) -~ + ~ 2)}

16.11 Constant-strain triang le


The triangular element shown in Figure 16.9 may be used in a plane-stress or plane-strain
analysis. The element has three nodes at the corners, with two nodal displacements u and v. The

(c)
k ~~ ~
k --- ----- --,
k
1: :I
1: :1
I

1:I :I ~
1:I ____ _;_ _ _ _ :I ~

Figure 16.9 Triangular plane-stress or plane-strain element. (a) Nodal coordinates. (b) and (c) Stresses
on element edges lumped in equivalent forces at the nodes.
Finite-element method 503

element is called a constant-strain triangle because the strain, and hence the stress, within the
element is constant.
Each of u and v is associated with three of the six nodal displacements. The same polynomial
may therefore be used for the two variables:

[P] = [1, x, y] (16.69)

We shall now derive the shape functions associated with u, which are the same as the shape
functions associated with v. At the three nodes we have

l X; y ]
1 Xj y; {A) (16.70)
l Xk Yk

or

{u} = [C] {A} (16.71)

Inversion of [C], using Cramer's rule (Appendix A, Section A.9) or other methods, gives

(16.72)

where 2!'. is the determinant of [CJ = 2 x area of triangle. Here,

(16.73)

By cyclic permutation of the subscripts i, j, and k, similar equations can be written for {aj, bi, Cj)
and for {ak, bk, q}.
Substitution in Eq. 16.60 gives the shape functions:

(16.74)

The displacements {u, v) at any point may be expressed in terms of the nodal displacements
(Eq. 16.25):

{ ~ } = [ Loi Lo1 L2
V
0 L3 0 ]
O L2 0 L3 {D* l (16.75)

where

The matrix [B], relating strains {E} to {D*} is (Eqs. 16.39 and 16.11)

1 [ b;
[B]= ,,... 0 (16.76)
2
C;
504 Finite-element method

The product [d] [B] gives the stresses due to unit nodal displacements (Eq. 16.42):

(16.77)

where d;; are elements of the elasticity matrix [d], given by Eqs. 16.12 and 16.13 for the states
of plane stress and plane strain respectively. It can be seen that all the elements of [B] and [au]
are constants, indicating constant strain and stress. If the thickness h is constant, the stiffness
matrix of the element is (Eq. 16.45)

[S*] = ht:,. [B]T [ d] [B] = ht:. [Bf [au] (16.78)

that is,
d11bf +d33cf symmetrical
d21 b;c; + d33b;c; d22cf +d33bf
+ d33C;Cj
d11 b;b; d21 c;b;+ d33c;b; d11bf +d33cf
[S*]=~
4t:,. d21 b;c; + d33b;c; d22c;c; + d33b;b; d21 b;c; + d33 c;b;

d11 b;bk + d33C;(k d21 c;bk + d33b;ck d11 b;bk + d33C;Ck


(16.79)

d22cf +d33bf
d21c;bk +d33b;ck d11b~ +d33c~
d22c;Ck + d33b;bk d21bkck + d33ckck d22c~ + d33b~
If the element is subjected to uniform body forces with intensities per unit volume of {p} = {qx,qy}
the equilibrants at the nodes, when the nodal displacements are prevented, are (Eq. 16.47)

{F*}=-h// [
b
L1O L10 L2
0
0
L2
L3
0
0
L3
]T{ qx }dxdy
qy
(16.80)

Evaluation of the integrals is simplified by noting that J J dx dy = t:.; J Jxdx dy is the first
moment of area about they axis, so that J J ydx dy = (t:./3)(y; +Yi+ Yk). The consistent vector
of forces in equilibrium with the body forces thus becomes

(16.81)

This means that one-third of the load on the triangle is assigned to each node. The same distri-
bution could have been suggested intuitively. (However, it is not always possible to determine
{Fb} intuitively; see Figure 17. 7.)
The consistent vector of restraining forces when the element is subjected to a rise in
temperature of T degrees is obtained by substitution of Eqs. 16.49 and 16.76 into Eq. 16.50:
aTh
{F&}m = -
2 (d11 + d21) {b;, c;, b;, Cj, bk, Ck} (16.82)

Here, we have assumed that the material is isotropic (d22 = d11).


Finite-element method 505

16.12 Interpretation of nodal forces


In the derivation of the stiffness matrix of individual elements, the forces distributed along the
edges of the elements are replaced by equivalent forces lumped at the nodes. The equivalent
forces are determined by the use of the principle of virtual work.
This approach can be seen in the example of the constant-strain triangular element shown in
Figure 16.9. For a unit nodal displacement, say, Di= 1, while other nodal displacements are
zero, the stresses are (Eq. 16.77, first column of [o-u])

(16.83)

where d;1 are elements of the elasticity matrix [d] given in Eqs. 16.12 and 16.13 for plane-stress
or plane-strain states respectively; b; = Yi - Yk; c; = Xk - x1; and D. = area of the triangle.
The stresses are represented by uniform forces on the edges of the element in Figure 16.9c.
The distributed forces in Figure 16.9c are lumped at the nodes in Figure 16.96. The force in
the x or y direction at a node is equal to the sum of one-half of the distributed load on each of
the two sides connected to the node. For example, the horizontal force at node k is one-half of
the load on edges ki and kj; thus,

(16.84)

We can now verify that this is the same as element S51 of the stiffness matrix in Eq. 16.79. Any
other element of the stiffness matrix can be verified in a similar way.
Equation 16.84 means that the virtual work of force S51 during nodal displacement Di =
1 is the same as the work done by the forces distributed over edges ki and kj during their
corresponding virtual displacements (which, in this example, vary linearly).
The consistent load vector also represents the distributed forces lumped at the nodes. If the
element shown in Figure 16.96 is subjected to a rise in temperature of T degrees and the element
expansion is restrained, the stresses in an isotropic material will be (Eqs. 16.48 and 16.49)

(16.85)

where a is the coefficient of thermal expansion.


Under the conditions of restraint, uniform forces act on the edges so as to produce o-x = o-y =
-aT(d11 + d21). Lumping one-half of the distributed load on any edge at the two end nodes
gives the nodal forces of Eq. 16.82.

16.13 General
The displacement method of analysis is applicable to structures composed of finite elements
which may be one-, two-, or three-dimensional. The analysis requires the generation of stiffness
and stress matrices and of load vectors for individual elements. Exact element matrices can be
generated only for bars. Procedures to generate approximate matrices for elements of any type
have been presented in this chapter. Convergence to the exact solution can be ensured as a finer
finite-element mesh is used. The general procedures have been applied to bar elements and to
plate elements subjected to in-plane forces and to bending. Other finite elements are discussed
in Chapter 17.
506 Finite-element method

Assemblage of stiffness matrices and load vectors of individual elements, generation of struc-
ture equilibrium equations and their solutions, and the use of computers are discussed in
Chapters 21 and 22. This is done mainly for structures composed of bar elements (framed
structures). However, the same techniques apply when any type of finite element is used.

Problems
16.1 Determine any element S7; for the bar of Example 16.5, using the shape functions in
Figure 16.4. Compare the answer with an exact value as given by Eq. 11.15.
16.2 Consider the member of Prob. 1 as a cantilever fixed at the end x = I and subjected to a
transverse concentrated load Pat the free end. Find the deflection at the tip of the cantilever,
using the two matrices given in the answers to Prob. 1.
16.3 For the prismatic bar shown, generate the stiffness matrix corresponding to the three
coordinates indicated. Use the following shape functions: L1 = -(1/2)~(1 - ~); L2 =
(1/2)~(1 +~);L 3 = 1- ~2 (Lagrange polynomials, Figure 17.3). Condense the stiffness
matrix by elimination of node 3.

~=1
12
____.:::;::~----_-_-_-_____________________~+--

I. 2
.I. 2
.I
Prob. 16.3

16.4 Assume that the displacement at coordinate 1 of the bar element of Prob. 16.3 is prevented.
A spring support is provided at coordinate 2 so that F2 = - Ku2, where F2 and u2 are,
respectively, the force and the displacement at coordinate 2, and K is the spring constant
equal to Ea/(3/). What will be the displacements at coordinates 2 and 3 and the stress in
the bar when it is subjected to a rise in temperature of T = (l/2)To(l + ~) where To is a
constant? Use the shape functions given in Prob. 16.3. Do you expect exact answers?
16.5 Determines; 1 , S21 , S31 , s: 1 , and s; 1 of the stiffness matrix of the plane-stress finite element
in Figure 16.5. Assume that the element is rectangular with sides band c parallel to the x
and y axes respectively, and is made of an isotropic material with v = 0.2. Element thickness
is constant and equal to h. The nodal displacement vector and shape functions are defined
by Eqs. 16.30 to 16.32.
16.6 Determine the consistent vector of restraining forces {Fb} for a rectangular plane-stress
element (Example 16.2) subjected to a uniform temperature rise of T degrees. Assume
an isotropic material with v = 0.2. Only the first two elements of the vector need to be
determined by Eq. 16.50; the remaining elements can be generated by consideration of
equilibrium and symmetry.
16.7 Considering symmetry and equilibrium, use the results of Prob. 16.5 to generate [S*] for
the element when b = c. For a comparison of the accuracy of this element with other
elements, see Prob. 17.18.
16.8 Use the results of Prob. 16.5 to calculate the deflection at the middle of a beam idealized
by two elements as shown. Take the beam width ash and Poisson's ratio as 0.2. Compare
the result with that obtained by beam theory, considering bending and shear deformations.
Note that the deflection calculated by the finite-element method is smaller than the more
accurate value obtained by beam theory and that the percentage error increases with an
increase in the ratio b / c.
Finite-element method 507

P/2

Tc
P/2
b b
I· ·1· ·1
Prob. 16.8

16.9 Use the displacement determined in Prob. 16.8 to calculate the stresses in the top fiber at
the fixed end.
16.10 Determine the consistent vector of restraining forces {Fb} for the rectangular plane-stress
element considered in Example 16.3 when it is subjected to a uniform temperature rise of
T degrees. Assume an isotropic material. Only the first three elements of the vector need
to be determined by Eq. 16.50; the remaining elements can be obtained by considering
equilibrium and symmetry.
16.11 Determine Sf 1 for the rectangular plate-bending element shown in Figure 16.6. The nodal
displacement vector and the shape functions are defined in Eqs. 16.33 to 16.35. Use the
result to calculate the central deflection of a rectangular plate 2b x 2c with built-in edges
and subjected to a concentrated load P at mid-point. Consider an isotropic material with
v = 0.3. Idealize the plate by four elements and perform the analysis for one element only,
taking advantage of symmetry. Use Figure 16.6 to represent the element analyzed with
node 1 at the center of the plate. Determine also Mx at node 2.
16.12 The answers to this problem are given in terms of the ratio b/c. For comparison, we give
here the exact6 answers for a square plate (b/c = 1): deflection = 0.244 Pb 2 /Eh3; Mx at
node 2 = -0.126 P. The deflection is smaller than the finite-element solution: why?
16.13 The figure shows identical plane-stress elements with three coordinate systems. For the
system in (a), calculate Sfi, S21 , S3i, S33 , S34 , and S44 , and, by considering equilibrium
and symmetry, generate the remaining elements of [S*]. What are the transformation
matrix [T] and the equation to be used to transform [S*] into [SJ for the coordinate

(a) (b) (c) 4


3
1

--------~---- 5

4''
1
2
i
6
4

Prob. 16.13

6 From Timoshenko, S. and Goodier, J. N., Theory of Elasticity, 2nd ed., McGraw-Hill, New York, 1951.
508 Finite-element method

system in (b)? What is the value of the angle a to be substituted in the equations to give
[S] corresponding to the coordinates in (c)? Consider an isotropic material with Poisson's
ratio v = 0. Element thickness is h; element sides are b, b, b,J2.
16.14 Using the answers of Prob. 16.2 and taking advantage of symmetry, determine, for the
beam shown, the nodal displacements and the stress in element 2-6-3. The beam is of
isotropic material; v = 0; beam width is h. The structure idealization is composed of
isosceles right-angle triangles of the same size. The equilibrium equations can be checked
by substitution of {D} given in the answers.

x,u I~
pt
I
y, V

p
9
1
Prob. 16.14

16.15 Generate columns 1, 2 and 3 of the [B] matrix for element QLC3 (Figure 16.1).
16.16 Calculate any stiffness coefficient Sii for a rectangular plane-stress element QLC3
(Figure 16.1 and Example 16.3). Consider an isotropic square element of side b, con-
stant thickness hand Poisson's ratio v = 0.2. The stiffness matrix is given in the answers
and is used in Prob. 17.18 to compare the accuracy of results when using this approach
and two other types of element (the element in Figure 16.5 and the hybrid element of
Example 17.6).