Labrosse - Finite Element Analysis

MCG 4102/5108 Finite Element
Analysis
Michel R. Labrosse, Ph.D., P.Eng., Associate Professor
Department of Mechanical Engineering
University of Ottawa
c _M.R. Labrosse, Ottawa, Canada, 2012
Contents
1 Introduction to FEA: Trusses 1
1.1 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Element stiness relationship in local coordinates . . . . . . . 3
1.3 Transformation from local to global coordinates . . . . . . . . 5
1.4 Global element stiness relationship . . . . . . . . . . . . . . . 6
1.5 Assemblage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 Application of loads . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Application of restraints on nodal displacements and solution . 9
1.8 Processing of results . . . . . . . . . . . . . . . . . . . . . . . 11
1.9 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Linear elasticity - Principle of virtual work 13
2.1 Stress at a point . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Equations of static equilibrium . . . . . . . . . . . . . . . . . 14
2.3 Strain at a point . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Strain-displacement relations . . . . . . . . . . . . . . . . . . . 16
2.5 Compatibility equations . . . . . . . . . . . . . . . . . . . . . 17
2.6 A constitutive relationship - Hookes law . . . . . . . . . . . . 17
2.7 Principle of minimum potential energy . . . . . . . . . . . . . 18
2.8 Principle of virtual work . . . . . . . . . . . . . . . . . . . . . 19
3 Finite element for beams 21
3.1 Theory - Beam bending in one plane . . . . . . . . . . . . . . 21
3.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Element stiness and load vectors . . . . . . . . . . . . . . . . 25
3.4 Assemblage and solution . . . . . . . . . . . . . . . . . . . . . 26
3.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.6 Beam element with axial loading . . . . . . . . . . . . . . . . 29
i
CONTENTS ii
3.7 General 3-D beam (or frame) element . . . . . . . . . . . . . . 31
4 Interpolation functions - Integration formulas 33
4.1 Compatibility and completeness requirements . . . . . . . . . 33
4.2 One dimensional (1-D) elements . . . . . . . . . . . . . . . . . 35
4.2.1 2-node C
0
-continuous element . . . . . . . . . . . . . . 35
4.2.2 2-node C
1
-continuous element . . . . . . . . . . . . . . 36
4.3 2-D elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3.1 Triangular C
0
-continuous element . . . . . . . . . . . . 37
4.3.2 Rectangular (quad) C
0
-continuous element . . . . . . . 39
4.3.3 Curved elements . . . . . . . . . . . . . . . . . . . . . 40
4.4 3-D elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.4.1 Four-node, C
0
-continuous tetrahedral element (pyramid) 42
4.4.2 Eight-node, C
0
-continuous brick element . . . . . . . . 42
4.4.3 Axisymmetric elements . . . . . . . . . . . . . . . . . . 43
4.5 Integration formulas . . . . . . . . . . . . . . . . . . . . . . . 43
4.5.1 Direct integration . . . . . . . . . . . . . . . . . . . . . 43
4.5.2 Numerical integration - Gaussian quadrature . . . . . . 44
4.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5 Stress analysis in 2-D with linear triangular element 46
5.1 Plane stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.1.1 The shape function matrix [`] . . . . . . . . . . . . . . 47
5.1.2 The strain-nodal displacement matrix [1] . . . . . . . 47
5.1.3 Constitutive relationship . . . . . . . . . . . . . . . . . 48
5.1.4 Finite element formulation (triangular element) . . . . 48
5.1.5 The element stiness matrix [1
c
] . . . . . . . . . . . . 49
5.1.6 The element nodal force vector ,
c
. . . . . . . . . . 50
5.1.7 Assemblage . . . . . . . . . . . . . . . . . . . . . . . . 54
5.1.8 Prescribed displacements - Solution . . . . . . . . . . . 55
5.1.9 Element resultants . . . . . . . . . . . . . . . . . . . . 55
5.2 Plane strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3 Axisymmetric problems . . . . . . . . . . . . . . . . . . . . . . 57
5.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6 Finite elements for plates and shells 59
6.1 Introduction - Bending of thin plates . . . . . . . . . . . . . . 59
6.1.1 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 59
CONTENTS iii
6.1.2 Kirchho assumptions . . . . . . . . . . . . . . . . . . 60
6.1.3 Constitutive relationships . . . . . . . . . . . . . . . . 61
6.1.4 Equilibrium equations . . . . . . . . . . . . . . . . . . 63
6.2 Finite element formulation (rectangular element) . . . . . . . . 64
6.2.1 Element type . . . . . . . . . . . . . . . . . . . . . . . 64
6.2.2 Displacement function . . . . . . . . . . . . . . . . . . 65
6.2.3 Principle of virtual work . . . . . . . . . . . . . . . . . 66
6.2.4 Element stiness matrix [1
c
] . . . . . . . . . . . . . . 67
6.2.5 Element nodal force vector ,
c
. . . . . . . . . . . . . 67
7 Method of weighted residuals 69
7.1 The method of weighted residuals . . . . . . . . . . . . . . . . 69
7.1.1 General concepts . . . . . . . . . . . . . . . . . . . . . 69
7.1.2 Point collocation . . . . . . . . . . . . . . . . . . . . . 70
7.1.3 Subdomain collocation . . . . . . . . . . . . . . . . . . 71
7.1.4 Least squares . . . . . . . . . . . . . . . . . . . . . . . 72
7.1.5 Galerkin . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.1.6 Comparison with exact solution . . . . . . . . . . . . . 73
7.2 The Galerkin nite element method . . . . . . . . . . . . . . . 74
7.2.1 Formulation . . . . . . . . . . . . . . . . . . . . . . . . 74
7.2.2 Application: 1-D heat transfer in a pin n . . . . . . . 76
8 Steady state thermal analysis 79
8.1 1-D heat conduction . . . . . . . . . . . . . . . . . . . . . . . 79
8.2 Governing equation . . . . . . . . . . . . . . . . . . . . . . . . 79
8.3 FE formulation . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.4 2-D heat conduction . . . . . . . . . . . . . . . . . . . . . . . 83
8.4.1 Governing equation . . . . . . . . . . . . . . . . . . . . 83
8.4.2 Green-Gauss theorem . . . . . . . . . . . . . . . . . . . 83
8.4.3 FE formulation . . . . . . . . . . . . . . . . . . . . . . 85
8.5 Axisymmetric heat conduction . . . . . . . . . . . . . . . . . . 87
9 Fluid ow analysis 89
9.1 2-D potential ow . . . . . . . . . . . . . . . . . . . . . . . . . 89
9.1.1 Velocity potential formulation . . . . . . . . . . . . . . 90
9.1.2 Stream function formulation . . . . . . . . . . . . . . . 92
9.2 2-D incompressible ow . . . . . . . . . . . . . . . . . . . . . . 93
CONTENTS iv
10 Transient and dynamic analyses 96
10.1 Dynamic structural analysis . . . . . . . . . . . . . . . . . . . 96
10.2 Transient thermal analysis . . . . . . . . . . . . . . . . . . . . 97
10.3 Lumped versus consistent matrices . . . . . . . . . . . . . . . 98
10.4 Solution methods . . . . . . . . . . . . . . . . . . . . . . . . . 99
10.4.1 Two-point recurrence scheme . . . . . . . . . . . . . . 99
10.4.2 Three-point recurrence scheme . . . . . . . . . . . . . . 102
10.4.3 Initial conditions . . . . . . . . . . . . . . . . . . . . . 103
10.5 Introduction to modal analysis . . . . . . . . . . . . . . . . . . 104
Introduction
The present notes cover the theoretical material discussed in the MCG 4102
/ 5108 Finite Element Analysis course, which is only introductory in nature.
Although the nite element method is a mathematical approach to solve
many dierent types of partial dierential and integral equations, the pre-
sentation is deeply rooted in mechanical engineering, as it is the background
of most students taking this course. The notes are not intended to replace the
required textbook (N-H Kim and BV Sankar, Introduction to Finite Element
Analysis and Design, Wiley, 2009), but are meant to complement it. How-
ever, for examination purposes, the notations herein prevail. The document
was prepared with the help of several additional textbooks, in particular:
J Fish and T Belytschko, A rst course in nite elements, Wiley, 2007.
FL Stasa, Applied nite element analysis for engineers, Saunders College
Publishing, 1985.
IH Shames and CL Dym, Energy and nite element methods in structural
mechanics, Taylor & Francis, 1985.
DL Logan, A rst course in the nite element method, 4th edition, Thom-
son, 2007.
JN Reddy, An introduction to the nite element method, 3rd edition,
McGraw Hill, 2004.
Questions and comments from the students over the years have also
tremendously contributed to shape the document into its current form. All
cited and anonymous contributors are gratefully acknowledged, and future
comments are welcome.
Finally, the reader should keep in mind that the present document is lim-
ited to linear elasticity. Nonlinear behaviours (related to non-innitesimally
small displacements and/or deformations, material nonlinearities, such as
plasticity, or changing boundary conditions/contact issues) are not covered,
and require the use of yet other textbooks.
v
Chapter 1
Introduction to FEA: Trusses
Trusses: structures composed of straight members connected at joints by
pins (Fig. 1.1). Most or all members in a truss do not experience bending
or torsional moments.
Figure 1.1: Truss bridge example.
Given: external forces, geometry, material properties.
Unknown: displacements at each joint; axial elongation, strain, stress,
force for each member.
1.1 Discretization
Let us consider each member of a truss as an element (Fig. 1.2):
Figure 1.2: Bar element.
1
CHAPTER 1. INTRODUCTION TO FEA: TRUSSES 2
Element c in Fig. 1.2 has two nodes, i and ,. at either end. Forces are
transmitted from one element to the next at nodes. Bar elements can only
take axial forces (beam elements also allow bending moments, as will be seen
later). i and , (lowercase) are the local node numbers within the element.
1 and J (uppercase) denote global node numbers in the whole structure.
Elements are also numbered (Fig. 1.3).
Figure 1.3: Example of discretization of a given truss.
Nodal coordinates:
Node number r coordinate (in) coordinate (in)
1 12 0
2 12 6
3 0 0
4 0 10
Element data (connectivity table):
Element Node i Node , Material ag
1 1 2 1 : 0.5-in (diameter) steel
2 3 1 2 : 0.4-in aluminium
3 3 2 1
4 4 2 2
5 3 4 1
1.2 Element stiness relationship in local co-
ordinates
Let (r
0
.
0
) be the local coordinate system for element c. with r
0
along the
length of element from i to ,, and
0
perpendicular to r
0
.
Figure 1.4: Bar element with degrees of freedom in local coordinate system.
The nodal displacements are noted as n
0
i
and
0
i
in r
0
and
0
, respectively,
at node i (Fig. 1.4). The corresponding forces are noted 1
0
ai
and 1
0
ji
Similarly,
the nodal displacements are noted as n
0
)
and
0
)
in r
0
and
0
, respectively, at
node ,.with corresponding forces 1
0
a)
and 1
0
j)
. From elementary strength of
materials, o =
11
1
. where o is the axial elongation, 1 the member length, 1
the axial force, the cross-sectional area and 1 the elastic modulus. It is
assumed that the elastic range is not exceeded and that is constant. In
other words, writing o = n
0
)
n
0
i
.
1
0
ai
=
1
1
(n
0
i
n
0
)
)
1
0
a)
=
1
1
(n
0
)
n
0
i
)
, where 1
0
ai
= 1
0
a)
for equilibrium.
Because bar elements do not withstand transverse forces, 1
0
ji
= 1
0
j)
= 0.
In matrix form,
1
1
_
_
1 0 1 0
0 0 0 0
1 0 1 0
0 0 0 0
_
_
_
_
n
0
i
0
i
n
0
)
0
)
_
_
=
_
_
1
0
ai
1
0
ji
1
0
a)
1
0
j)
_
_
With global nodal coordinates (r
i
.
i
) and (r
)
.
)
) for nodes i and ,, re-
spectively, the element length can be computed as 1 =
_
(r
)
r
i
)
2
+ (
)
i
)
2
.
The other two properties and 1 can be specied for each element. Con-
cisely,
_
1
c
0
_
l
c
0
_
=
_
1
c
0
_
. where
_
1
c
0
represents the local stiness matrix,

_
l
c
0
_
the local nodal displacement vector, and
_
1
c
0
_
the local nodal force
vector for the element. "c" denotes element, "
0
" denotes local coordinate
system.
Example 1 for Element 3:
_
1
(3)
0
=

(3)
1
(3)
1
(3)
_
_
1 0 1 0
0 0 0 0
1 0 1 0
0 0 0 0
_
_
.
Material ag set to 1: 0.50-in steel:
(3)
=

4
(0.5)
2
= 0.196 in
2
1
(3)
= 30 10
6
psi
1
(3)
=
_
(0 12)
2
+ (0 6)
2
= 13.42 in
.
Then,
_
1
(3)
0
= 10
3
_
_
438 0 438 0
0 0 0 0
438 0 438 0
0 0 0 0
_
_
lbf/in
1.3 Transformation from local to global coor-
dinates
In the global (r. ) coordinate system shown in Fig. 1.5, position vector
: to
an arbitrary point 1 can be written as
: = :
a
i +:
j
, . In the bar (rotated)

coordinate system (r
0
.
0
).

: = :
a
0
i
0
+:
j
0
,
0
.
Figure 1.5: Local and global coordinate systems.
Reminder: Scalar product "." of any two vectors

n and

:
n .
= |
n | |
| cos(
n .
). Therefore,
: .
i = :
a
i .
i +:
j
, .
i = :
a
0
i
0
.
i +:
j
0
,
0
.
i
= :
a
+ 0 = :
a
0 cos(r
0
. r) +:
j
0 cos(
0
. r)
: .
, = :
a
i .
, +:
j
, .
, = :
a
0
i
0
.
, +:
j
0
,
0
.
,
= 0 +:
j
= :
a
0 cos(r
0
. ) +:
j
0 cos(
0
. )
In matrix form,
_
:
a
:
j
_
=
_
cos(r
0
. r) cos(
0
. r)
cos(r
0
. ) cos(
0
. )
_ _
:
a
0
:
j
0
_
or : = [1] :
0
.with
[1] =
_
cos(r
0
. r) cos(
0
. r)
cos(r
0
. ) cos(
0
. )
_
=
_
cos o cos(o +

2
) = sin o
cos(o

2
) = sin o cos o
_
In the following, the notation [1] =
_
:
11
:
12
:
21
:
22
_
is used, where coe-
cients :
11
. :
12
. :
21
and :
22
are called direction cosines.
It can be shown that [1] is orthogonal, i.e. [1]
1
= [1]
T
. therefore
:
0
= [1]
T
:, with [1]
T
=
_
:
11
:
21
:
12
:
22
_
.
1.4 Global element stiness relationship
The same transformation of coordinates as above holds for displacement vec-
tors
_
n
0
i
0
i
_
at node i. and
_
n
0
)
0
)
_
at node ,. We get
_
_
n
0
i
0
i
n
0
)
0
)
_
_
=
_
_
:
11
:
21
0 0
:
12
:
22
0 0
0 0 :
11
:
21
0 0 :
12
:
22
_
_
_
_
n
i
i
n
)
)
_
_
or more concisely,
_
l
c
0
_
= [1] l
c
.where [1] =
_
1
T
0
0 1
T
_
and l
c
=
_
_
n
i
i
n
)
)
_
_
.
Similarly,
_
1
c
0
_
= [1] 1
c
. with 1
c
=
_
_
1
ai
1
ji
1
a)
1
j)
_
_
.
_
1
c
0
_
l
c
0
_
=
_
1
c
0
_
becomes
_
1
c
0
[1] l
c
= [1] 1
c
. or
[1]
T
_
1
c
0
[1] l
c
= 1
c
. since [1]
T
[1] = [1]
1
[1] = [1] .
More simply, [1
c
] l
c
= 1
c
. where [1
c
] = [1]
T
_
1
c
0
[1] is the global

element stiness matrix
[1
c
] =
1
1
_
_
:
2
11
:
21
:
11
:
2
11
:
21
:
11
:
11
:
21
:
2
21
:
11
:
21
:
2
21
:
2
11
:
21
:
11
:
2
11
:
21
:
11
:
11
:
21
:
2
21
:
11
:
21
:
2
21
_
_
with :
11
= cos o
a
=
a
j
a
i
1
:
21
= cos o
j
=
j
j
j
i
1
.
Note that computing :
11
=
a
j
a
i
1
and :
21
=
j
j
j
i
1
is a very robust way to
obtain the direction cosines (Fig. 1.6). It is highly recommended (instead of
using rotation angles).
Figure 1.6: Direction cosines in 2-D.
Example 2 for Element 3:
:
11
=
a
2
a
3
1
(3)
=
120
13.42
= 0.8944
:
21
=
j
2
j
3
1
(3)
=
60
13.42
= 0.4472
_
1
(3)
= 10
3
_
_
351 176 351 176
176 88 176 88
351 176 351 176
176 88 176 88
_
_
lbf/in
1.5 Assemblage
The original structure is put back together from individual elements. This
is done based on the compatibility of nodal displacements: the r- and -
displacements at one node must be identical (in the intended design) to
those of the other nodes from other elements to be merged.
Example 3 for the truss in Fig. 1.7, determine the assemblage stiness
matrix in terms of the 2x2 global stiness submatrices
_
1
c
1,J
.
Figure 1.7: Figure 1.3 repeated here for convenience.
Procedure: let us rst create a null assemblage stiness matrix [1
o
] in-
volving global node numbers 1 to 4.
[1
o
] =
1 2 3 4
_
_
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
0
2a2
_
_
1
2
3
4
From the connectivity table, element 1 has global node numbers 1 and 2,
so
_
1
(1)
=
_
1
(1)
1,1
1
(1)
1,2
1
(1)
2,1
1
(1)
2,2
_
.
Element 2 has global node numbers 3 and 1, so
_
1
(2)
=
_
1
(2)
3,3
1
(2)
3,1
1
(2)
1,3
1
(2)
1,1
_
.
_
1
(3)
=
_
1
(3)
3,3
1
(3)
3,2
1
(3)
2,3
1
(3)
2,2
_
.
_
1
(4)
=
_
1
(4)
4,4
1
(4)
4,2
1
(4)
2,4
1
(4)
2,2
_
.
_
1
(5)
=
_
1
(5)
3,3
1
(5)
3,4
1
(5)
4,3
1
(5)
4,4
_
.
Finally,
[1
o
] =
1 2 3 4
_
_
1
(1)
1,1
+1
(2)
1,1
1
(1)
1,2
1
(2)
1,3
0
2a2
1
(1)
2,1
1
(1)
2,2
+1
(3)
2,2
+1
(4)
2,2
1
(3)
2,3
1
(4)
2,4
1
(2)
3,1
1
(3)
3,2
1
(2)
3,3
+1
(3)
3,3
+1
(5)
3,3
1
(5)
3,4
0
2a2
1
(4)
4,2
1
(5)
4,3
1
(4)
4,4
+1
(5)
4,4
_
_
1
2
3
4
Note that each individual 2x2 element stiness submatrix is symmetric,
and [1
o
] is symmetric. Also note that, depending on the connectivity table,
some submatrices may have to be transposed to t in the assemblage matrix.
1.6 Application of loads
By design, the (known) external loads in a truss can only occur at the joints
(nodes).
Example 4 for the truss in Fig. 1.7, the r- and - components of the ap-
plied load are 1
a
= 2. 000 cos 60
o
= 1. 000 lbf and 1
j
= 2. 000 sin 60
o
=
1. 732 lbf. The load is applied to Node 1. There is an unknown reaction
force in the r-direction at Node 3 (roller) and two unknown reaction forces
in the r- and -directions at Node 4 (pin).
1
o
=
_
_
1. 000
1. 732
0
0
1
a3
0
1
a4
1
j4
_
_
1
a
1
j
1
a
1
j
1
a
1
j
1
a
1
j
Node 1
Node 2
Node 3
Node 4
1.7 Application of restraints on nodal displace-
ments and solution
After assemblage, the system equation is of the form [1
o
] l
o
= 1
o
.
where l
o
is the vector of nodal displacements.
l
o
T
= n
1

1
n
2

2
... n
.

.
T
. where ` is the maximum
number of nodes and n
1
.
1
are the r- and - displacements at global node
number 1. For now, no restraint on the nodal displacements have been con-
sidered, and the whole structure can y into space!!! This is a rigid body
motion, and l
o
cannot be determined because [1
o
] cannot be inverted
(singular matrix, i.e. its determinant is zero). Restraints MUST be applied.
After the restraints corresponding to the boundary conditions are en-
forced (n
3
= 0. and n
4
=
4
= 0 for the truss in Fig. 1.7), the system equation
becomes [1] l = 1 . where [1] is non-singular, therefore [1]
1
exists,
which implies a unique solution for l = [1]
1
1 .
There are dierent methods for enforcing displacement restraints:
Method 1 (interesting for a technique called sub-structuring, but not
used otherwise)
Consider the system /
11
n
1
+/
12
n
2
+/
13
n
3
= ,
1
/
21
n
1
+/
22
n
2
+/
23
n
3
= ,
2
/
31
n
1
+/
32
n
2
+/
33
n
3
= ,
3
.
with /
i)
= /
)i
and i. , = 1...3.
Let us impose n
2
= n
known
. therefore the set of equations above is equiv-
alent to e.g.
/
11
n
1
+/
12
n
2
+/
13
n
3
= ,
1
n
2
= n
known
/
31
n
1
+/
32
n
2
+/
33
n
3
= ,
3
The symmetry has been destroyed but can be restored by transposing
terms involving n
2
to the right-hand side, and with n
known
replacing n
2
:
_
_
/
11
0 /
13
0 1 0
/
31
0 /
33
_
_
_
_
_
n
1
n
2
n
3
_
_
_
=
_
_
_
,
1
/
12
n
known
n
known
,
3
/
32
n
known
_
_
_
. Note that ,
2
disappears:
a known displacement leads to an unknown reaction force.
Method 2 (interesting for programming of the nite element method)
Consider the same initial system as above. Let us select a large number
, (6 to 12 orders of magnitude larger than the largest coecient /
i)
). , is
added to /
ii
if n
i
is prescribed, and the right-hand side of the ith equation is
changed to , times the prescribed value. In matrix form, the system becomes
_
_
/
11
/
12
/
13
/
21
(/
22
+,) /
23
/
31
/
32
/
33
_
_
_
_
_
n
1
n
2
n
3
_
_
_
=
_
_
_
,
1
, n
known
,
3
_
_
_
.
Symmetry is preserved, and this method is very easy to implement in a
computer program. For practical purposes, n
2
= n
known
.
Method 3 (the only one to use when doing hand calculations)
When the boundary conditions are enforced, the initial system equation
[1
o
] l
o
= 1
o
is rewritten into [1] l = 1 by simply removing the
rows corresponding to the unknown reactions, and the columns corresponding
to the known (zero) displacements.
1.8 Processing of results
Assemblage displacement vector
Determine l
o
from l and the boundary conditions.
Reaction forces
Determine the unknown reaction forces (if needed) from 1
o
= [1
o
] l
o
.
Element resultants
Element resultants include axial elongation, strains, stresses and forces in
each element, all of which need to be checked against failure criteria i.e.
maximum allowable displacement, maximum allowable strain, yield stress
and buckling (if compression is present). The element resultants are com-
puted from the nodal displacements. For element c. going back to the de-
nition of the axial elongation, o = n
0
)
n
0
i
. where n
0
)
= :
11
n
)
+ :
21
)
and
n
0
i
= :
11
n
i
+:
21
i
.
At this stage, n
i
.
i
. n
)
.
)
and :
11
. :
21
are known therefore o can be com-
puted, and so can the axial strain =
c
1
. The material is assumed to be
elastic, therefore, the axial (also called normal) stress o = 1 can be de-
termined, and so can the axial force 1 = o. Note that for o. . o and
1. positive values denote tension, while negative values denote compression.
Make sure to check that these signs make physical sense.
Internal forces
Internal forces can be determined as described above, i.e. from the element
strains. Another, sometimes more direct, method consists in writing the
equilibrium of one element of interest and using the known displacements at
its nodes. Of course, all methods must yield the same results!
Example 5 determine the internal forces in Element 3 of the truss in Fig.
1.7, from
_
1
(3)
_
n
3
3
n
2
2
_
_
=
_
_
1
(3)
a3
1
(3)
j3
1
(3)
a2
1
(3)
j2
_
_
(in the global coordinate system)
or from
_
1
(3)0
_
n
0
3
0
3
n
0
2
0
2
_
_
=
_
_
1
(3)0
a3
1
(3)0
j3
1
(3)0
a2
1
(3)0
j2
_
_
(in the local coordinate system).
1.9 Examples
Assignment 1.
For additional problems to practice with, use the results in the statement
of Assignment 1, and nd yourself
_
1
(2)
.
_
1
(4)
and
_
1
(5)
.
Chapter 2
Linear elasticity - Principle of
virtual work
This is a short review of linear elasticity in statics and equilibrium principles,
including the principle of virtual work for use in nite element formulation.
2.1 Stress at a point
Consider a deformable body in static equilibrium, loaded as shown in Fig.
2.1:
Figure 2.1: Static equilibrium of a body.
13
CHAPTER2. LINEARELASTICITY- PRINCIPLEOF VIRTUAL WORK14
The external forces

1
1
.

1
2
. ... are transmitted through the deformable
body in a complex manner. When the body is cut along some plane, a force
1 is required to maintain static equilibrium.

1 has a normal component 1
a
and two tangential orthogonal components 1
t1
and 1
t2
. Consider small area
instead of . Then, 1
a
. 1
t1
. 1
t2
act on . The normal stress
o
a
is dened as o
a
= lim
!0
1n
and the two shear (tangential) stresses as

o
t1
= lim
!0
1
t1
and o
t2
= lim
!0
1
t2
. For an innitesimal volume element

of a body positioned at a point in a global (r. . .) coordinate system (Fig.
2.2):
Figure 2.2: Some of the stresses acting on an innitesimal volume element.
(1
ct
subscript: facet normal; 2
ao
subscript: direction)
2.2 Equations of static equilibrium
Consider an innitesimal 2-D volume element of thickness t (Fig. 2.3). It
is assumed that the normal and shear stresses vary from point to point in
the body in some continuous manner. Therefore, a rst order Taylor series
expansion is used.
Figure 2.3: Stresses acting on a 2-D cross-section.
/
a
. /
j
: components of body force per unit volume.
Equilibrium in r-direction:
(o
aa
+
0oxx
0a
dr)d o
aa
d + (o
ja
+
0oyx
0j
d)dr o
ja
dr +/
a
drd = 0
Equilibrium in -direction:
(o
jj
+
0oyy
0j
d)dr o
jj
dr + (o
aj
+
0oxy
0a
dr)d o
aj
d +/
j
drd = 0
Finally, in 2-D,
0oxx
0a
+
0oyx
0j
+/
a
= 0
0oxy
0a
+
0oyy
0j
+/
j
= 0
and o
aj
= o
ja
for moment equi-
librium.
Similarly in 3-D,
0oxx
0a
+
0oyx
0j
+
0ozx
0:
+/
a
= 0
0oxy
0a
+
0oyy
0j
+
0ozy
0:
+/
j
= 0
0oxz
0a
+
0oyz
0j
+
0ozz
0:
+/
:
= 0
and o
aj
= o
ja
o
a:
= o
:a
o
:j
= o
j:
for mo-
ment equilibrium.
2.3 Strain at a point
Consider a square element whose sides are of unit length (Fig. 2.4). Under
the action of the external loading, the element will deform such that the sides
of the square are no longer perpendicular.
Figure 2.4: Strains in 2-D.
By denition,
aj
=
1
+
2
0 when angle 1
0
C
0
becomes smaller than
2
. Strain is dimensionless.
2.4 Strain-displacement relations
For small deformations strains and displacements are related as follows:
In 2-D,
aa
=
0&
0a

jj
=
0
0j

aj
=
0&
0j
+
0
0a
where n. are the displacements of a point in the r and directions.
In 3-D,
aa
=
0&
0a
aj
=
0&
0j
+
0
0a
jj
=
0
0j
j:
=
0
0:
+
0&
0j
::
=
0&
0:
:a
=
0&
0a
+
0&
0:
where n. . n are the displacements of a point in the r. . . directions.
In matrix form, = [1] l . where,
in 2-D,
T
= [
aa
.
jj
.
aj
] . [1] =
_
_
0
0a
0
0
0
0j
0
0j
0
0a
_
_
and l
T
= [n. ]
in 3-D,
T
= [
aa
.
jj
.
::
.
aj
.
j:
.
:a
] . [1] =
_
_
0
0a
0 0
0
0
0j
0
0 0
0
0:
0
0j
0
0a
0
0
0
0:
0
0j
0
0:
0
0
0a
_
_
and
l
T
= [n. . n]
2.5 Compatibility equations
In 2-D, since
aj
=
0&
0j
+
0
0a
.
aa
=
0&
0a
and
jj
=
0
0j
.
0
2
.xy
0a0j
=
0
2
0&
0a0j
2
+
0
2
0
0a
2
0j
=
0
2
.xx
0j
2
+
0
2
.yy
0a
2
. hence an extra relationship, called
compatibility equation, between three strains and two displacements.
Similar compatibility equations are obtained in 3-D. Compatibility equa-
tions are automatically satised in the stiness approach used in the displacement-
based nite element method described in this course.
2.6 A constitutive relationship - Hookes law
The uniaxial Hookes law o = 1. where 1 is the elastic modulus, can be
generalized, and is expressed in 3-D as o = [1] (
0
) +o
0
. where
[1] is the material property matrix, o the stress vector, the strain
vector,
0
the initial strain vector, and o
0
the initial (or residual) stress
vector.
o
T
= [o
aa
. o
jj
. o
::
. o
aj
. o
j:
. o
:a
]
T
= [
aa
.
jj
.
::
.
aj
.
j:
.
:a
]
[1] =
1
(1+i)(12i)
_
_
1 i i i 0 0 0
i 1 i i 0 0 0
i i 1 i 0 0 0
0 0 0
12i
2
0 0
0 0 0 0
12i
2
0
0 0 0 0 0
12i
2
_
_
1 : elastic modulus, i : Poissons ratio
Other specic [1] matrices will be found for plane stress, plane strain, or
axisymmetric analyses. [1] is symmetric for both isotropic and anisotropic
materials. In the isotropic case, only 1 and i are needed. o
0
represents
stresses that are known to exist in a material before it is loaded. They
must be specied by the analyst.
0
may be the result of crystal growth,
shrinkage, or temperature changes.
Note that in dierent textbooks, shear stresses are often noted with t
(e.g. o
aj
is noted t
aj
). and shear strains are often noted with (e.g.
aj
is
noted
aj
).
2.7 Principle of minimum potential energy
How to nd the mechanical equilibrium of a structure other than by writing
the equilibrium equations for each component of the structure, and hoping
to be able to solve for all of them? The principle of minimum potential en-
ergy (presented in this section, and which youve encountered many times
in physics) and the principle of virtual displacements (presented in next sec-
tion) are powerful tools to look for the equilibrium of a whole structure at
once.
The principle of minimum potential energy (PMPE) states that: "out
of all the possible displacements elds that satisfy the geometric boundary
conditions (i.e. prescribed displacements), the one that also satises the
equations of static equilibrium results in the minimum of total potential
energy for the structure (or body)."
The total potential energy is dened as the sum of the strain energy
(internal potential energy l
i
) and the external potential energy l
c
from the
external forces. = l
i
+ l
c
. For conservative systems, the loss of external
potential energy during the loading process must be equal to the work \
c
done on the system by the external forces, or l
c
= \
c
. and therefore
= l
i
\
c
. is a function of functions (strains and displacements) and is
called a functional. Minimizing is called a variational problem. The rst
variation of the total potential energy o = ol
i
o\
c
must be zero, i.e.
ol
i
= o\
c
.
In a global Cartesian (r. . .) coordinate system,
ol
i
=
_
\
(o
aa
o
aa
+o
jj
o
jj
+o
::
o
::
+o
aj
o
aj
+o
j:
o
j:
+o
:a
o
:a
) d\.
Using matrix notation, ol
i
=
_
\
o
T
o d\.
For the work of external forces, considering a body force / (per unit
volume), a surface traction : (per unit area) and ` point loads ,
j
. then
o\
c
=
_
\
(/
a
on +/
j
o +/
:
on) d\ +
_
(:
a
on +:
j
o +:
:
on) d
+
j=.
j=1
(,
ja
on +,
jj
o +,
j:
on).
Using matrix notation, o\
c
=
_
\
ol
T
/ d\ +
_
ol
T
: d
+
j=.
j=1
ol
T
,
j
. where
/
T
= [/
a
. /
j
. /
:
] : body force vector
:
T
= [:
a
. :
j
. :
:
] : surface traction vector
,
j
T
= [,
ja
. ,
jj
. ,
j:
] : point load vector
ol
T
= [on. o. on] : rst variation of displacement vector
Eliminating o using the linear elastic stress-strain relationship, the
PMPE becomes
_
\
o
T
[1] d\ =
_
\
o
T
[1]
0
d\
_
\
o
T
o
0
d\
+
_
\
ol
T
/ d\ +
_
ol
T
: d +
j=.
j=1
ol
T
,
j
.
Going back to the total potential energy, it is given by
=
1
2
_
\

T
[1] d\
_
\

T
[1]
0
d\ +
_
\

T
o
0
d\
_
\
l
T
/ d\
_
l
T
: d
j=.
j=1
l
T
,
j
Example 6 consider the elongation and an axial force 1 in a uniaxial

stress member (bar) of uniform cross-sectional area . length 1 and modulus
of elasticity 1 (Fig. 2.5). One end of the bar is xed.
Figure 2.5: Extension of a bar.
0
= 0. o
0
= 0. / = 0 and : = 0.
=
1
2
1
2
1 1 with =

1
. therefore =
1
2
1(
1
)
2
1 1.
We want for equilibrium, i.e. for minimum .
c
c
= 1

1
2
1 1 = 0 == =
11
1
: this is well known!
Note that
c
2
c
2
=
1
1
0. therefore, is really at a minimum of .
2.8 Principle of virtual work
This principle, also known as the principle of virtual displacements, will be
very convenient and useful for nite element formulation of complex prob-
lems. Work is the product of a displacement and the component of the force
in the direction of the displacement. Virtual work is imagined to occur when
the forces are real and the displacements are virtual (imagined), or vice-versa,
but this not used herein.
Statement of the principle of virtual work (PVW): if the work done by
the external forces on the structure is equal to the increase in strain energy
for any set of admissible virtual displacements (i.e. satisfying the prescribed
displacements), then the system is in equilibrium.
Let us denote the virtual displacements in r. . . directions as on. o.
on (not variations!). The virtual displacements will cause virtual strains
o
aa
. o
jj
. o
::
. o
aj
. o
j:
. o
:a
.
In matrix form, the PVW becomes
_
\
o
T
o d\ =
_
\
ol
T
/ d\ +
_
ol
T
: d+
j=.
j=1
ol
T
,
j
.
\ol . o .
with notations as in Section 2.7.
This is known as a weak form of the equilibrium equations because this
equation only contains rst derivatives of displacements whereas the original
equilibrium equations (see Section 2.2 + Hookes law) contain second order
derivatives of the displacements.
Chapter 3
Finite element for beams
3.1 Theory - Beam bending in one plane
The deection of the neutral axis of a beamat any location r is represented by
(r) (bending in a plane). The deections are supposed to be small compared
to the length of the beam (typically less than 3% of length if this is not the
case, a more advanced theory must be used). The material is also supposed
to be linearly elastic. Finally, it is assumed that the beam cross-section has
an axis of symmetry in the plane of bending, and that planar cross-sections
remain planar during deformation. Lets take point 1 on the beam neutral
axis, and point Q at distance from the neutral axis (Fig. 3.1, left). After
deformation (Fig. 3.1, right, in which the deection (r) and rotation o(r)
of the cross-section are grossly exaggerated), relationships between dierent
variables can be established as follows. Lets introduce n(r) the longitudinal
displacement of Q due to the deformation.
21
CHAPTER 3. FINITE ELEMENT FOR BEAMS 22
Figure 3.1: Beam deformation.
: local deection of the beam
o
oa
: local slope of the beam, physically interpreted as the rotation of the
local cross-section of the beam, therefore
o
oa
= o.
From trigonometry, tan o =
&
j
. and because o is very small, o tan o.
Therefore, one can write
o
oa
= n(r) : longitudinal displacement due to (r).
Recalling that a longitudinal strain is obtained by taking the rst deriv-
ative of the longitudinal displacement, one obtains
o
2
oa
2
=
a
: longitudinal strain due to (r).
Because the material of the beam is assumed linearly elastic, the local
constitutive equation is such that the longitudinal (i.e. normal) stress in the
beam at point Q is o
a
= 1
a
, where 1 is the materials elastic modulus. In
other words, o
a
= 1
o
2
oa
2
.
In beam cross-section , the external normal force is `(r) =
_
o
a
d.
Because o
a
varies linearly in the -direction, `(r) = 0.
In beam cross-section . the external bending moment is
`(r) =
_
o
a
d = 1
o
2
oa
2
_
2
d. Recalling
_
2
d = 1 : second mo-
ment of area, one obtains, `(r) = 11
o
2
oa
2
. This is the global constitutive
equation for the beam, i.e. it expresses the relationship between the deec-
tion and the global external loads applied to the beam.
Note that beam deection analysis also often aims at determining the
bending stresses in the beam. To do so, one simply applies
o
a
= 1
o
2
oa
2
=
A(a)j
1
for each location of interest in the beam.
Considering the free-body diagram of an elemental beam segment of
length dr (Fig.3.2), one can get additional relationships as follows.
Figure 3.2: Free-body diagram of an elemental segment of beam.
Lets introduce (r) the transverse distributed load (force per unit length
not necessarily constant).
` = 0 : \ dr +d` = 0. therefore \ (r) =

oA(a)
oa
.
1 = 0 : d\ +dr = 0. therefore
o\ (a)
oa
= (r).
In the absence of distributed load, obviously, (r) = 0.
Finally, the governing equations for the beam are:
11
o
2
oa
2
= ` bending moment
11
o
3
oa
3
=
oA
oa
= \ transverse shear force
11
o
4
oa
4
=
o\
oa
= transverse distributed load
3.2 Discretization
Consider the beam element shown in Fig. 3.3:
Figure 3.3: Beam element and its degrees of freedom.
In the absence of distributed lads, the equilibrium equation for the ele-
ment is
o
4
oa
4
= 0. for which the general solution is a third-order polynomial
(r) = c
1
+c
2
r +c
3
r
2
+c
4
r
3
.
The elements end conditions can be expressed in terms of the following
nodal values:
Node 1: (0) =
1
= c
1
o(0)
oa
= o
1
= c
2
Node 2: (1) =
2
= c
1
+c
2
1 +c
3
1
2
+c
4
1
3
o(1)
oa
= o
2
= c
2
+ 2c
3
1 + 3c
4
1
2
In matrix form, the same relationships can be written as:
_
1
o
1
2
o
2
_
_
=
_
_
1 0 0 0
0 1 0 0
1 1 1
2
1
3
0 1 21 31
2
_
_
_
_
c
1
c
2
c
3
c
4
_
_
or l
c
= [] Q . where [] is
called coecient matrix.
By inverting [] . we can solve for Q in terms of l
c
.
[]
1
=
_
_
1 0 0 0
0 1 0 0
3
1
2

2
1
3
1
2

1
1
2
1
3
1
1
2

2
1
3
1
1
2
_
_
Finally, (r) can be written in matrix form as:
(r) =
_
1 r r
2
r
3

Q =
_
1 r r
2
r
3

[]
1
l
c
.
or (r) = [`] l
c
with [`] =
_
1 r r
2
r
3

[]
1
[`] =
_
`
1
(r) `
2
(r) `
3
(r) `
4
(r)

. or more explicitly,
[`] =
_
1
3a
2
1
2
+
2a
3
1
3
r
2a
2
1
+
a
3
1
2
3a
2
1
2

2a
3
1
3

a
2
1
+
a
3
1
2
.
We note that [`] [
a=0
=
_
1 0 0 0

[`] [
a=1
=
_
0 0 1 0

_
o.
oa
[
a=0
=
_
0 1 0 0

_
o.
oa
[
a=1
=
_
0 0 0 1

3.3 Element stiness and load vectors
Let us assume that at Node 1 of the beam element, nodal force 1
1
works
in displacement
1
. nodal moment `
1
works in rotation o
1
. and similarly at
Node 2, 1
2
works in displacement
2
. nodal moment `
2
works in rotation
o
2
.
The principle of virtual work for the element is:
_
\
o
T
o d\ = ol
c
T
_
_
1
1
`
1
1
2
`
2
_
_
. \ol
c
. \o .
Here, o
T
= o
a
. and o = o
a
. as the other components are zero.
Then,
_
\
o
a
o
a
d\ =
_
1
0
__
((o
o
2
(a)
oa
2
)(1
o
2
(a)
oa
2
)d
_
dr
=
_
1
0
11(o
o
2
(a)
oa
2
)(
o
2
(a)
oa
2
)dr since
_
2
d = 1.
From (r) = [`] l
c
, one can derive
o
2
(a)
oa
2
=
o
2
oa
2
[`] l
c
= [1] l
c
.
and o
o
2
(a)
oa
2
= [1] ol
c
.
where [1] =
_

6
1
2
+
12a
1
3
4
1
+
6a
1
2
6
1
2

12a
1
3

2
1
+
6a
1
2
.
Then,
_
1
0
11(o
o
2
(a)
oa
2
)(
o
2
(a)
oa
2
)dr =
_
1
0
11 ol
c
T
[1]
T
[1] l
c
dr
= ol
c
T
_
1
0
[1]
T
11 [1] drl
c
.
Finally, the PVW yields:
ol
c
T
_
1
0
[1]
T
11 [1] drl
c
= ol
c
T
_
_
1
1
`
1
1
2
`
2
_
_
. \ol
c
.
In other words,
_
1
0
[1]
T
11 [1] drl
c
=
_
_
1
1
`
1
1
2
`
2
_
_
. or [1
c
] l
c
= 1
c
.
with [1
c
] =
11
1
3
_
_
12 61 12 61
61 41
2
61 21
2
12 61 12 61
61 21
2
61 41
2
_
_
.
So far, we only considered point forces and moments acting in the dis-
placements and rotations, respectively. For a distributed load n(r). the
virtual work of external forces is:
_
1
0
o(r)n(r)dr = ol
c
T
_
1
0
[`]
T
n(r)dr = ol
c
T
1
&
.where 1
&
is the nodal force vector representing a distributed load on the basis of work
equivalence.
For n(r) = (constant),
1
&
=
_
_
_
1
0
`
1
(r)n(r)dr
_
1
0
`
2
(r)n(r)dr
_
1
0
`
3
(r)n(r)dr
_
1
0
`
4
(r)n(r)dr
_
_
=
_
_
q1
2
q1
2
12
q1
2
q1
2
12
_
_
(Fig. 3.4).
Figure 3.4: Constant distributed load and its nodal equivalents.
3.4 Assemblage and solution
Consider the example shown in Fig. 3.5:
Figure 3.6: Example of beam problem.
Elements 1 and 2 have stiness matrices
_
1
(1)
=
11
1
3
_
_
12 61 12 61
61 41
2
61 21
2
12 61 12 61
61 21
2
61 41
2
_
_
and
_
1
(2)
=
_
1
(1)
. This is
with respect to the degrees of freedom described in Fig. 3.7:
Figure 3.7: Discretization and degrees of freedom for problem in Fig. 3.6.
At this stage, the elements "ignore" each other. Let us nd the assemblage
stiness matrix for the system.
The assemblage must enforce that
2
=
3
and o
2
= o
3
(compatibility
of displacements and rotations between elements). This is done by simply
adding the corresponding matrix entries, and overlapping the element sti-
ness matrices (see Chapter 1):
[1
o
] =
11
1
3
_
_
12 61 12 61 0 0
61 41
2
61 21
2
0 0
12 61 24 0 12 61
61 21
2
0 81
2
61 21
2
0 0 12 61 12 61
0 0 61 21
2
61 41
2
_
1
o
1
C
o
C
works in
1
1
1
`
1
1
C
`
C
The assemblage nodal displacement vector is
l
T
=
_

1
o
1

C
o
C
.
Let us now determine the assemblage nodal force vector, or assemblage
load vector. Can we identify some components in the generic load vector
1
T
=
_
1
1
1
`
1
1
C
`
C
?
Consistent with what was done with the stiness matrix entries,
1
2
+1
3
= 1
1
`
2
+`
3
= `
1
Furthermore, a free-body diagramof the system(Fig. 3.8) gives 1
= 1
.
1
1
= 1. `
1
= `. 1
C
= 1
C
and `
= 0 and `
C
= 0 (obviously, H
= 0).
Figure 3.8: Free-body diagram of beam in Fig. 3.6.
So far, we have 6 equations and 8 unknowns: 1
. 1
C
. and
. o
.
1
. o
1
.
C
. o
C
.
However, the boundary conditions are such that
=
C
= 0. therefore the
actual number of unknowns is 6, and the problem can be solved.
Because
=
C
= 0. we eliminate the corresponding rows and columns
to solve for the unknowns. Then,
11
1
3
_
_
41
2
61 21
2
0
61 24 0 61
21
2
0 81
2
21
2
0 61 21
2
41
2
_
_
_
_
o
1
o
1
o
C
_
_
=
_
_
0
1
`
0
_
_
For the case where 1 = 1. 000 lbf and ` = 0.
o
=
2501
2
11
.
1
=
1671
3
11
. o
1
= 0. o
C
=
2501
2
11
.
From these solutions, and extracting two rows of the assemblage system,
one can determine the reaction forces 1
and 1
C
corresponding to
C
= 0 as
11
1
3
_
12 61 12 61 0 0
0 0 12 61 12 61
_
_
1
o
1
C
o
C
_
_
=
_
1
1
C
_
.
Finally, 1
= 504 lbf and 1

C
= 504 lbf. Note that a value of 504 is ob-
tained instead of the expected 500 due to round-o errors during calculation.
Once the displacements and rotations are known along with all the shear
forces and moments, the bending stresses can be calculated at critical lo-
cations, using the same equations as in strength of materials or machine
design.
3.5 Example
Assignment 2.
3.6 Beam element with axial loading
Outside of buckling and stress stiening (e.g. a taut guitar string), which are
nonlinear cases, we can linearly superimpose the results from bar and beam
elements, because displacements and strains are assumed to be small. Note
that in many textbooks and software packages, such combination describes
a frame element (Fig. 3.9).
Figure 3.9: Combined bar and beam element, or frame element.
[/
c
] =
_
[/
c
Bar
] [0]
[0] [/
c
Beam
]
_
with n
c
T
=
_
n
1
n
2

1
o
1

2
o
2
.
With more convenient n
c
T
=
_
n
1

1
o
1
n
2

2
o
2
, [/
c
] can be
re-organized into:
[/
c
] =
_
_
1
1
0 0
1
1
0 0
0
1211
1
3
611
1
2
0
1211
1
3
611
1
2
0
611
1
2
411
1
0
611
1
2
211
1
1
1
0 0
1
1
0 0
0
1211
1
3

611
1
2
0
1211
1
3

611
1
2
0
611
1
2
211
1
0
611
1
2
411
1
_
_
If the element is oriented at an arbitrary angle from the A axis of
the global reference frame, we have (Fig. 3.10)
Figure 3.10: Local and global degrees of freedom for a 2-D frame element.
n
1
= l
1
cos +l
2
sin
1
= l
1
sin +l
2
cos
o
1
= l
3
n
2
= l
4
cos +l
5
sin
2
= l
4
sin +l
5
cos
o
2
= l
6
and in matrix form,
_
_
n
1
1
o
1
n
2
2
o
2
_
_
=
_
_
cos sin 0 0 0 0
sin cos 0 0 0 0
0 0 1 0 0 0
0 0 0 cos sin 0
0 0 0 sin cos 0
0 0 0 0 0 1
_
_
_
_
l
1
l
2
l
3
l
4
l
5
l
6
_
_
.
or n
c
= [1] l
c
. Then, the element stiness matrix in the global
system is [1
c
] = [1]
T
[/
c
] [1] .
3.7 General 3-D beam (or frame) element
We want to include
axial behaviour (bar)
bending behaviour in rplane (beam)
bending behaviour in r.plane (beam)
torsional behaviour
Items 1 and 2 have been taken care of. Bending in r.plane needs
attention because it is similar to bending in rplane, but instead of o
:
=
o
oa
.
we have o
j
=
o&
oa
. and instead of `
:
= 11
:
o
2
oa
2
. we have `
j
= 11
j
o
2
&
oa
2
(Fig. 3.11).
Figure 3.11: Degrees of freedom for a general 3-D beam element.
Therefore, [/
c
]
a:
=
11
1
3
_
_
12 61 12 61
61 41
2
61 21
2
12 61 12 61
61 21
2
61 41
2
_
_
.
Torsion is represented by stiness matrix [/
c
Torsion
] =
JG
1
_
1 1
1 1
_
.
nodal displacement vector
_
o
a1
o
a2
_
and load vector
_
`
a1
`
a2
_
. Finally, for
the general 3-D beam element:
_
_
[/
c
Bar
] [0] [0] [0]
[0] [/
c
Beam
]
aj
[0] [0]
[0] [0] [/
c
Beam
]
a:
[0]
[0] [0] [0] [/
c
Torsion
]
_
_
_
_
n
1
n
2
1
o
:1
2
o
:2
n
1
o
j1
n
2
o
j2
o
a1
o
a2
_
_
=
_
_
1
a1
1
a2
1
j1
`
:1
1
j2
`
:2
1
:1
`
j1
1
:2
`
j2
`
a1
`
a2
_
_
.
Chapter 4
Interpolation functions -
Integration formulas
From the principle of virtual work, it is evident that in stress analysis prob-
lems, the variables of interest are the displacements (vector). In thermal
analysis, it would be the temperature (scalar). In uid ow problems, it
would be the uid velocities (vector) and the pressure (scalar). Below are
ideas and results regarding interpolation functions for these variables and
integration over element length, surface or volume that can be used for yet
other studies (electromagnetism, mass transfer, etc...).
4.1 Compatibility and completeness require-
ments
Compatibility
For C
0
-continuous problems, the interpolation function must be continuous
along the boundaries of the element. For C
1
-continuous problems, the func-
tion and its rst derivative must be continuous (Fig. 4.1).
33
CHAPTER4. INTERPOLATIONFUNCTIONS - INTEGRATIONFORMULAS34
Figure 4.1: C
0
-continuous function (left), and C
1
-continuous function
(right).
Many problems are C
0
-continuous once formulated using the weak form
(2-D stress, strain, axisymmetric stress, 3-D stress analyses), but some are
C
1
-continuous or higher. Elements that obey the compatibility requirement
are said to be conforming (vs. non-conforming). Some non-conforming ele-
ments are useful, but they should be used with extra caution.
Completeness
For C
a
-continuous problems, the interpolation function must be capable of
representing a constant value of the variable as well as partial derivatives of
up to order :+1 as the element size decreases to a point. Example in uniaxial
stress analysis: let us assume n = c
1
+c
2
r. If c
2
= 0. n = c
1
= constant : rigid
body mode (displacement of the whole body without straining). Also,
o&
oa
=
c
2
= constant : this allows for a constant strain in the element. Therefore,
the interpolation function n = c
1
+ c
2
r can be used for a C
0
-continuous
problem.
Mesh sensitivity analysis
With both compatibility and completeness requirements satised, conver-
gence of the solution during mesh renement can be achieved (Fig. 4.2).
Regardless of the expectations, convergence of the solution during mesh re-
nement is a verication that MUST be done for every analysis (except for
bar elements) when using a nite element code.
Figure 4.2: Convergence of solution upon mesh renement.
4.2 One dimensional (1-D) elements
Interpolation functions are based on polynomials or rational functions. The
functions can be developed with global, natural (or serendipity) or length
coordinates, depending on the base of integration. Many dierent types of
elements can be created, but not all will be useful.
4.2.1 2-node C
0
-continuous element
Global coordinates (Fig. 4.3)
Figure 4.3: Global coordinates.
For variable c. we pose c(r) = c
1
+c
2
r.
or c(r) = [1 r]
_
c
1
c
2
_
= [1 r] Q .
At nodes i. ,. we require c(r
i
) = c
i
= c
1
+c
2
r
i
c(r
)
) = c
)
= c
1
+c
2
r
)
.
In matrix form,
_
c
i
c
)
_
=
_
1 r
i
1 r
)
_ _
c
1
c
2
_
or c
c
= [] Q .
Solving for the vector of constants yields
_
c
1
c
2
_
=
_
1 r
i
1 r
)
_
1
_
c
i
c
)
_
.
Then, c(r) = [1 r]
_
1 r
i
1 r
)
_
1
_
c
i
c
)
_
or
c(r) = [1 r] []
1
c
c
= [`] c
c
with [`] = [
a
j
a
a
j
a
i
aa
i
a
j
a
i
].
Natural coordinates (Fig. 4.4)
Figure 4.4: Natural coordinates.
Let us dene a local, normalized coordinate : dened relative to the global
coordinate r.
: = 0 at r = r =
1
2
(r
i
+r
)
). and 1 _ : _ +1 with : =
2(aa)
a
j
a
i
.
With these coordinates, [`] = [
1
2
(1 :)
1
2
(1 +:)].
Length coordinates (Fig.4.5)
Figure 4.5: Length coordinates.
Let us consider an internal point j (not a node) in element c. and dene
1
i
=
length j)
length i)
and 1
)
=
length ij
length i)
.
Obviously, 0 _ 1
i
_ 1. 0 _ 1
)
_ 1. and 1
i
+1
)
= 1.
Then, [`] = [1
i
1
)
].
4.2.2 2-node C
1
-continuous element
See development of beam element in global coordinates in Chapter 3 (Section
3.2, Discretization). Note that this element is C
2
-continuous in n. but C
1
-
continuous in o.
4.3 2-D elements
4.3.1 Triangular C
0
-continuous element
Global coordinates (Fig. 4.6)
Figure 4.6: Global 2-D coordinates.
We pose c(r. ) = c
1
+c
2
r +c
3
or
c(r. ) = [1 r ]
_
_
_
c
1
c
2
c
3
_
_
_
= [1 r ] Q .
At nodes i. ,. /. we require c(r
|
.
|
) = c
|
= c
1
+ c
2
r
|
+ c
3
|
for | = i. , or
/.
In matrix form,
_
_
_
c
i
c
)
c
I
_
_
_
=
_
_
1 r
i

i
1 r
)

)
1 r
I

I
_
_
_
_
_
c
1
c
2
c
3
_
_
_
or c
c
= [] Q .
Solving for the vector of constants yields Q = []
1
c
c
.
Then, c(r. ) = [1 r ] []
1
c
c
= [`] c
c
with [`] = [`
i
`
)
`
I
]
and `
i
(r. ) = :
11
+:
21
r +:
31
`
)
(r. ) = :
12
+:
22
r +:
32
`
I
(r. ) = :
13
+:
23
r +:
33
where :
11
= (r
)
I
r
I
)
),2 :
21
= (
)
I
),2 :
31
= (r
I
r
)
),2
:
12
= (r
I
i
r
i
I
),2 :
22
= (
I
i
),2 :
32
= (r
i
r
I
),2
:
13
= (r
i
)
r
)
i
),2 :
23
= (
i
)
),2 :
33
= (r
)
r
i
),2
and =
1
2
det
_
_
1 r
i

i
1 r
)

)
1 r
I

I
_
_
= area of triangle i,/.
Area coordinates (Fig. 4.7)
Figure 4.7: Area coordinates.
Let us consider an internal point j (not a node) in element c. and dene
1
i
=
area j)I
area i)I
. 1
)
=
area ijI
area i)I
. and 1
I
=
area i)j
area i)I
Obviously, 0 _ 1
|
_ 1 for | = i. , or /. and 1
i
+1
)
+1
I
= 1.
Also, 1
i
(r
i
.
i
) = 1 1
i
(r
)
.
)
) = 0 1
i
(r
I
.
I
) = 0
1
)
(r
i
.
i
) = 0 1
)
(r
)
.
)
) = 1 1
)
(r
I
.
I
) = 0
1
I
(r
i
.
i
) = 0 1
I
(r
)
.
)
) = 0 1
I
(r
I
.
I
) = 1
Figure 4.8: Area coordinates with 1
I
= constant.
As can be seen in Fig. 4.8, triangles j
1
i, and j
2
i, have the same area
(they have the same base and height), therefore 1
I
= constant is a line
parallel to the opposite leg i,. 1
I
varies linearly between 0 and 1. It can be
shown that [`] = [1
i
1
)
1
I
].
Example 7 show that 1
i
= `
i
.
1
i
=
area j)I
area i)I
=
1
2
det
2
6
6
6
4
1 r
1 r
)

)
1 r
I

I
3
7
7
7
5
1
i
=
(a
j
j
k
a
k
j
j
)+(j
j
j
k
)a+(a
k
a
j
)j
2
1
i
= :
11
+:
21
r +:
31
= `
i
QED.
4.3.2 Rectangular (quad) C
0
-continuous element
Natural coordinates (Fig. 4.9)
Figure 4.9: Natural 2-D coordinates.
We have 1 _ : _ +1 and 1 _ : _ +1 with : =
aa
o
and : =
jj
b
.
Then, c(r. ) = [`
i
`
)
`
I
`
|
] c
c
with `
i
=
1
4
(1 +:)(1 :)
`
)
=
1
4
(1 +:)(1 + :)
`
I
=
1
4
(1 :)(1 + :)
`
|
=
1
4
(1 :)(1 :)
and c
c
=
_
_
c
i
c
)
c
I
c
|
_
_
.
Three rules
1. `
n
= 1 at node :. and 0 at other nodes, for : = i. ,. / or |.
2. At node :. only `
n
= 1. the others are 0.
3. `
i
+`
)
+`
I
+`
|
= 1 \:. :
4.3.3 Curved elements
So far, all the elements considered have had straight edges. Curved ele-
ments (with more nodes) may be needed to describe curved boundaries more
accurately (or with fewer elements). A mapping can be used between the
straight-edged parent element and the curved element (Fig. 4.10). In an
isoparametric mapping, the same interpolation functions are used both for
the variable of interest (c) and the description of the geometry.
Figure 4.10: Parent element (left), and curved element (right).
For simplicity of presentation, let us stick with linear elements (straight
edges), as in Fig. 4.11:
Figure 4.11: Mapping between parent and real geometries.
To map the geometry, we want r = [G
i
G
)
G
I
G
|
]
_
_
r
i
r
)
r
I
r
|
_
_
and
= [G
i
G
)
G
I
G
|
]
_
|
_
_
.
We may want to use the interpolation functions dened previously, such
that `
n
(:. :) = G
n
. for : = i. ,. / or |.
With c(r. ) = c(:. :) = [`
i
`
)
`
I
`
|
] c
c
for the variable of in-
terest, this clearly leads to an isoparametric element.
4.4 3-D elements
4.4.1 Four-node, C
0
-continuous tetrahedral element (pyra-
mid)
Figure 4.12: Tetrahedral element.
c(r. . .) = c
1
+c
2
r +c
3
+c
4
.
4.4.2 Eight-node, C
0
-continuous brick element
Figure 4.13: Brick element.
c(r. . .) = c
1
+c
2
r +c
3
+c
4
. +c
5
r +c
6
. +c
7
.r +c
8
r.
4.4.3 Axisymmetric elements
A problem is axisymmetric if the body of interest is a body of revolution AND
if the material properties, boundary conditions and loads do not change with
o in a global cylindrical coordinate system attached to the body (Fig. 4.14).
Figure 4.14: Cylindrical coordinate system for axisymmetric problem.
In an axisymmetric problem, the variable of interest is a function of :
and . only, therefore the problem is actually two-dimensional. For volume
integrations, d\ = :d:dod. = 2::d:d. = 2::d.
4.5 Integration formulas
Building a stiness matrix or a force vector calls for many integrations of the
interpolation functions and their derivatives over the element length, surface
or volume.
4.5.1 Direct integration
Length coordinates
_
|
1
c
i
1
o
)
d| =
c!o!
(c+o+1)!
| where d| is an elemental length between nodes i and
,. and | is the length of line between nodes i and ,. Exponents c and , must
be positive integers. Recall that :! = : (: 1) 1.
Area coordinates
_
1
c
i
1
o
)
1
I
d =
c!o!!
(c+o++2)!
2 where d is an elemental area of the ele-
ment, and is the area of the triangle formed by nodes i. , and /. Exponents
c. , and must be positive integers.
4.5.2 Numerical integration - Gaussian quadrature
1-D formulas
Consider 1 =
_
a
j
a
i
,(r)dr. First we need to use natural coordinates, such
that, when r = r
i
. : = 1 and when r = r
)
. : = +1.
Take r = r
i
+
1
2
(r
)
r
i
)(1 + :). Then, dr =
oa
ov
d: = Jd:. where J is
the Jacobian of the transformation (Note: the notation is not useful in 1-D,
but becomes very convenient in 2-D and 3-D). Here, J =
1
2
(r
)
r
i
). Finally,
according to the Gauss-Legendre quadrature,
1 =
_
a
j
a
i
,(r)dr =
_
+1
1
,(r(:))Jd: =
_
+1
1
,(:)Jd: J
a
i=1
,(r(:
i
))n
i
(index i has nothing to do with the node index).
where n
i
are the weight factors, :
i
the base points, and : the number of
Gauss points (see numerical methods textbook for values of n
i
and :
i
). A
polynomial of degree j is integrated exactly by employing : =
1
2
(j+1) Gauss
points or nearest larger integer. Note that the location of Gauss points has
nothing to do with that of the element nodes.
2-D and 3-D formulas:
Similarly in 2-D,
1 =
_
+1
1
_
+1
1
,(:. :)d:d:
n
)=1
a
i=1
,(:
i
. :
)
)n
v
i
n
c
)
and in 3-D,
1 =
_
+1
1
_
+1
1
_
+1
1
,(:. :. t)d:d:dt
|
I=1
n
)=1
a
i=1
,(:
i
. :
)
. t
I
)n
v
i
n
c
)
n
t
I
.
Therefore, in 2-D, with drd = det [J] d:d:.
[1
c
] =
__
[1]
T
[1] [1] drd =
_
+1
1
_
+1
1
[1]
T
[1] [1] det [J] d:d:.
and in 3-D, with drdd. = det [J] d:d:dt.
[1
c
] =
___
\
[1]
T
[1] [1] drdd. =
_
+1
1
_
+1
1
_
+1
1
[1]
T
[1] [1] det [J] d:d:dt
where [J] is the Jacobian matrix of the transformation.
[J] shows up in many other instances, anytime coordinate changes are
needed. For example, for each interpolation
`
i
.
0.
i
0v
=
0.
i
0a
0a
0v
+
0.
i
0j
0j
0v
+
0.
i
0:
0:
0v
or, in matrix form,
_
_
_
0.
i
0v
0.
i
0c
0.
i
0t
_
_
_
=
_
_
0a
0v
0j
0v
0:
0v
0a
0c
0j
0c
0:
0c
0a
0t
0j
0t
0:
0t
_
_
_
_
_
0.
i
0a
0.
i
0j
0.
i
0:
_
_
_
= [J]
_
_
_
0.
i
0a
0.
i
0j
0.
i
0:
_
_
_
.
Thus, we have
_
_
_
0.
i
0a
0.
i
0j
0.
i
0:
_
_
_
= [J]
1
_
_
_
0.
i
0v
0.
i
0c
0.
i
0t
_
_
_
.
Such formulas will be used when we build 2-D and 3-D elements in the
following chapters.
4.6 Example
Assignment 3.
Chapter 5
Stress analysis in 2-D with
linear triangular element
5.1 Plane stress
The plane stress model is appropriate for a thin plate loaded uniformly across
its thickness t in a direction parallel to the mid-plate plane (Fig. 5.1). The
thickness need not be constant. Plane stress state implies: o
::
= 0 and
o
a:
= o
j:
= 0.
Figure 5.1: Plane stress problem (top view, left; side view, right).
46
CHAPTER5. STRESS ANALYSIS IN2-DWITHLINEARTRIANGULARELEMENT47
5.1.1 The shape function matrix [`]
In plane stress, element nodes have 2 degrees of freedom: the r and com-
ponents of the displacements. For a triangular element (see Chapter 4)
_
n(r. )
(r. )
_
=
_
`
i
0 `
)
0 `
I
0
0 `
i
0 `
)
0 `
I
_
_
_
n
i
i
n
)
)
n
I
I
_
_
or l = [`] l
c
.
5.1.2 The strain-nodal displacement matrix [1]
Recalling that for 2-D small deformations,
aa
=
0&
0a
.
jj
=
0
0j
and
aj
=
0&
0j
+
0
0a
. In matrix form,
_
_
_
aa
jj
aj
_
_
_
=
_
_
0
0a
0
0
0
0j
0
0j
0
0a
_
_
_
n
_
or = [1] l .
Finally, = [1] [`] l
c
= [1] l
c
.
with [1] = [1] [`] =
_
_
0.
i
0a
0
0.
j
0a
0
0.
k
0a
0
0
0.
i
0j
0
0.
j
0j
0
0.
k
0j
0.
i
0j
0.
i
0a
0.
j
0j
0.
j
0a
0.
k
0j
0.
k
0a
_
_
.
If, as we found in Chapter 4 for a triangular element,
`
i
(r. ) = :
11
+:
21
r +:
31
`
)
(r. ) = :
12
+:
22
r +:
32
`
I
(r. ) = :
13
+:
23
r +:
33
then [1] =
_
_
:
21
0 :
22
0 :
23
0
0 :
31
0 :
32
0 :
33
:
31
:
21
:
32
:
22
:
33
:
23
_
_
.
Since the :
i)
s are known functions of the nodal coordinates, [1] is known.
Note that in this case, [1] is composed of constant entries. In other words,
in a 2-D triangular linear element, strains are constant over an element, and
so are stresses. If a better resolution is needed, more elements or dierent
elements (of dierent shape and/or higher order) are required.
5.1.3 Constitutive relationship
For plane stress, the constitutive relationship for a linear elastic material
o = [1] (
0
) +o
0
is written with o =
_
_
_
o
aa
o
jj
o
aj
_
_
_
.
o
0
=
_
_
_
o
aa0
o
jj0
o
aj0
_
_
_
.
0
=
_
_
_
aa0
jj0
aj0
_
_
_
and [1] =
1
1i
2
_
_
1 i 0
i 1 0
0 0
1i
2
_
_
.
If the self-strain vector
0
is a result of temperature change 1. then
0
is given by
0
=
_
_
_
c
t
1
c
t
1
0
_
_
_
(no shear in isotropic materials) where
c
t
is the coecient of thermal expansion. Note that
::
is not zero in general:
::
=
i
1
(o
aa
+o
jj
) +
::0
.
5.1.4 Finite element formulation (triangular element)
Recalling the principle of virtual work written for one element:
_
\
e
o
T
o d\ =
_
\
e
ol
T
/ d\ +
_
e
ol
T
: d+
j=.
j=1
ol
T
,
j
.
\ol . o .
Using = [1] l
c
. then o = o([1] l
c
) = [1] ol
c
because [1]
is independent of l
c
. and o
T
= ol
c
T
[1]
T
.
Similarly, l = [`] l
c
. leading to ol
T
= ol
c
T
[`]
T
. Therefore,
the PVW is equivalent to
_
\
e
[1]
T
o d\
_
\
e
[`]
T
/ d\
e
[`]
T
: d
j=.
j=1
[`]
T
,
j
= 0 .
but since o = [1] (
0
) +o
0
and = [1] l
c
.
(
_
\
e
[1]
T
[1] [1] d\ ) l
c

_
\
e
[1]
T
[1]
0
d\
+
_
\
e
[1]
T
o
0
d\
_
\
e
[`]
T
/ d\
e
[`]
T
: d
j=.
j=1
[`]
T
,
j
= 0 .
Finally, [1
c
] l
c
= ,
c
where
[1
c
] =
_
\
e
[1]
T
[1] [1] d\ is the element stiness matrix and ,
c
the
nodal force vector is made of ve components:
,
c
=
_
\
e
[1]
T
[1]
0
d\
_
\
e
[1]
T
o
0
d\
+
_
\
e
[`]
T
/ d\ +
_
e
[`]
T
: d +
j=.
j=1
[`]
T
,
j
. or
,
c
= ,
c
.0
,
c
o0
+,
c
b
+,
c
c
+
_
,
c
j
_
.
Note that the previous derivation is general and not limited to 2-D cases.
It is done here for convenience.
5.1.5 The element stiness matrix [1
e
]
From above, [1
c
] =
_
\
e
[1]
T
[1] [1] d\ where both [1] and [1] (in our case)
are composed of constants, and d\ = tdrd.
Therefore, [1
c
] = [1]
T
[1] [1]
_
e
tdrd. For reasonably small elements, t
may be taken as a constant average value. Then, [1
c
] = [1]
T
[1] [1] t. Note
that in our case, [1
c
] is a 6x6 matrix because [1]
T
is 6x3, [1] is 3x3 and [1]
is 3x6. This is consistent with the fact that there are 6 nodal displacements
per triangular element.
Example 8 determine [1
c
] in the case shown in Fig. 5.2:
Figure 5.2: Triangular element.
Steel plate: 1 = 30 10
6
psi
i = 0.3
t = 0.25 in (constant)
=
1
2
det
_
_
1 r
i

i
1 r
)

)
1 r
I

I
_
_
=
1
2
det
_
_
1 4 2
1 0 2
1 0 0
_
_
= 4 in
2
:
21
= (
)
I
),2 =
20
24
= 0.25 in
1
:
31
= (r
I
r
)
),2 =
00
24
= 0
:
22
= (
I
i
),2 =
02
24
= 0.25 in
1
:
32
= (r
i
r
I
),2 = 0.50 in
1
:
23
= (
i
)
),2 = 0 :
33
= (r
)
r
i
),2 = 0.50 in
1
Then, [1] =
_
_
0.25 0 0.25 0 0 0
0 0 0 0.50 0 0.50
0 0.25 0.50 0.25 0.50 0
_
_
in
1
.
[1] =
1
1i
2
_
_
1 i 0
i 1 0
0 0
1i
2
_
_
=
_
_
33.0 9.89 0
9.89 33.0 0
0 0 11.5
_
_
10
6
psi (lbf/in
2
)
Finally,
[1
c
] =
_
_
2.06 0 2.06 1.24 0 1.24
0.72 1.44 0.72 1.44 0
4.94 2.68 2.88 1.24
8.97 1.44 8.25
2.88 0
Sym 8.25
_
_
10
6
lbf/in.
5.1.6 The element nodal force vector ,
e
Self-strain
,
c
.0
=
_
\
e
[1]
T
[1]
0
d\ =
_
e
[1]
T
[1]
0
tdrd.
Considering thermal strains only, with c
t
and 1 constant,
then ,
c
.0
= [1]
T
[1]
0
t.
,
c
.0
is 6x1 because [1]
T
is 6x3, [1] is 3x3 and
0
is 3x1.
Example 9 (continued from above): determine ,
c
.0
if c
t
= 6.0 10
6
in/(in.F) and the temperature increases by 150 F.
0
=
_
_
_
c
t
1
c
t
1
0
_
_
_
=
_
_
_
900
900
0
_
_
_
10
6
in/in.
,
c
.0
=
_
_
9. 650
0
9. 650
19. 300
0
19. 300
_
_
lbf.
Prestresses
,
c
o0
=
_
\
e
[1]
T
o
0
d\ =
_
e
[1]
T
o
0
tdrd = [1]
T
o
0
t.
,
c
o0
is 6x1, and o
0
=
_
_
_
o
aa0
o
jj0
o
aj0
_
_
_
.
Body forces
,
c
b
=
_
\
e
[`]
T
/ d\ =
_
e
_
_
`
i
0
0 `
i
`
)
0
0 `
)
`
I
0
0 `
I
_
_
_
/
a
/
j
_
tdrd.
Recalling that for a triangular element, `
|
= 1
|
for | = i. , or /. where
1
|
is an area coordinate,
,
c
b
=
_
_
_
e
1
i
/
a
tdrd
_
e
1
i
/
j
tdrd
_
e
1
)
/
a
tdrd
_
e
1
)
/
j
tdrd
_
e
1
I
/
a
tdrd
_
e
1
I
/
j
tdrd
_
_
with
_
e
1
i
/
a
tdrd =
1!0!0!
(1+0+0+2)!
/
a
t2 =
1
3
/
a
t.
Finally, ,
c
b
=
t
3
_
_
/
a
/
j
/
a
/
j
/
a
/
j
_
_
(if /
a
and /
j
are constant).
Surface tractions
,
c
c
=
_
e
[`]
T
: d =
_
e
_
_
`
i
0
0 `
i
`
)
0
0 `
)
`
I
0
0 `
I
_
_
_
:
a
:
j
_
d.
Surface tractions are only present at the surface of the structure. Let us
consider an element c with nodes i and , (but not /) on the global boundary
(Fig. 5.3).
Figure 5.3: Surface traction.
A surface traction is assumed to act on leg i,. Then, d = td|. and
`
I
= 1
I
= 0 on leg i,.
,
c
c
=
_
_
_
|
ij
1
i
:
a
td|
_
|
ij
1
i
:
j
td|
_
|
ij
1
)
:
a
td|
_
|
ij
1
)
:
j
td|
0
0
_
_
.
If :
a
and :
j
are assumed constant, then ,
c
c
=
t|
ij
2
_
_
:
a
:
j
:
a
:
j
0
0
_
_
for leg i, on
the global boundary.
Example 10 determine ,
c
c
(dimensions from Fig. 5.2)
Figure 5.4: Surface traction on triangular element.
:
j
= 0
:
a
is not constant over leg ,/. Eective :
a
=
1,600+2,000
2
= 1. 800 psi.
,
c
c
=
t|
jk
2
_
_
0
0
:
a
:
j
:
a
:
j
_
_
=
0.252
2
_
_
0
0
1. 800
0
1. 800
0
_
_
=
_
_
0
0
450
0
450
0
_
_
lbf.
(This is with respect to the local node ordering i. ,. /)
Point loads
_
,
c
j
_
=

j=.
j=1
[`]
T
,
j
=

j=.
j=1
_
_
`
i
0
0 `
i
`
)
0
0 `
)
`
I
0
0 `
I
_
_
_
,
ja
,
jj
_
_
,
c
j
_
=

j=.
j=1
_
_
`
i
(r
j
.
j
),
ja
`
i
(r
j
.
j
),
jj
`
)
(r
j
.
j
),
ja
`
)
(r
j
.
j
),
jj
`
I
(r
j
.
j
),
ja
`
I
(r
j
.
j
),
jj
_
_
.
Example 11 determine
_
,
c
j
_
for a point load acting at coordinates (0.6. 1.6)
for the element above, with ,
ja
= 1. 500 lbf and ,
jj
= 2. 300 lbf.
Since
`
i
(r. ) = :
11
+:
21
r +:
31
`
)
(r. ) = :
12
+:
22
r +:
32
`
I
(r. ) = :
13
+:
23
r +:
33
. calculation of all :
i)
s is needed.
The only ones that have not been calculated yet are:
:
11
= (r
)
I
r
I
)
),2 = 0
:
12
= (r
I
i
r
i
I
),2 = 0
:
13
= (r
i
)
r
)
i
),2 = 1
Therefore, at r = 0.6 and = 1.6.
`
i
= 0 + 0.25 0.6 + 0 1.6 = 0.15
`
)
= 0 0.25 0.6 + 0.5 1.6 = 0.65
`
I
= 1 + 0 0.6 0.25 1.6 = 0.20
_
,
c
j
_
=
_
_
225
345
975
1. 495
300
460
_
_
lbf.
5.1.7 Assemblage
Example 12 see Fig. 5.5
Figure 5.5: Assemblage of two triangular elements.
Connectivity table: Element # Node i Node , Node /
1 2 1 3
2 3 4 2
For Element 1,
_
1
(1)
=
_
_
1
(1)
22
1
(1)
21
1
(1)
23
1
(1)
12
1
(1)
11
1
(1)
13
1
(1)
32
1
(1)
31
1
(1)
33
_
_
n
2
2
n
1
1
n
3
3
For Element 2,
_
1
(2)
=
_
_
1
(2)
33
1
(2)
34
1
(2)
32
1
(2)
43
1
(2)
44
1
(2)
42
1
(2)
23
1
(2)
24
1
(2)
22
_
_
n
3
3
n
4
4
n
2
2
By assemblage, with l
o
T
= [n
1

1
n
2

2
n
3

3
n
4

4
] .
[1
o
] is rst zeroed out, and then entries are added.
[1
o
] =
_
_
1
(1)
11
1
(1)
12
1
(1)
13
0
1
(1)
21
1
(1)
22
+1
(2)
22
1
(1)
23
+1
(2)
23
1
(2)
24
1
(1)
31
1
(1)
32
+1
(2)
32
1
(1)
33
+1
(2)
33
1
(2)
34
0 1
(2)
42
1
(2)
43
1
(2)
44
_
_
. [1
o
] is symmetric.
,
o
=
_
_
,
(1)
1
,
(1)
2
+,
(2)
2
,
(1)
3
+,
(2)
3
,
(2)
4
_
_
.
5.1.8 Prescribed displacements - Solution
[1
o
] is singular, therefore prescribed displacements must be included for
solution. Finally, the system to be solved is [1] l = , . from which
l = [1]
1
, .
5.1.9 Element resultants
From the computed displacements l
c
. one can determine the strains and
stresses in the element.
= [1] l
c
. For a triangular element, the average strains across the
element are:

aa
= :
21
n
i
+:
22
n
)
+:
23
n
I
jj
= :
31
i
+:
32
)
+:
33
aj
= :
31
n
i
+:
21
i
+:
32
n
)
+:
22
)
+:
33
n
I
+:
23
I
.
The average stresses across the element are:
o = [1] ([1] l
c

0
) +o
0
.
5.2 Plane strain
The plane strain model is appropriate when a long prismatic member of
constant cross section is held between two xed rigid planes (Fig. 5.6). Plane
strain state implies:
::
= 0 and
a:
=
j:
= 0.
Figure 5.6: Plane strain problem (side view, left; end view, right).
All the results for plane stress apply, except that the constitutive rela-
tionship is changed: o = [1] (
0
) +o
0
with [1] =
1
(1+i)(12i)
_
_
1 i i 0
i 1 i 0
0 0
12i
2
_
_
.
For a temperature change 1.
0
=
_
_
_
(1 +i)c
t
1
(1 +i)c
t
1
0
_
_
_
.
Note that o
::
,= 0 = i(o
aa
+ o
jj
) + o
::0
1
::0
with
::0
= c
t
1 for
thermal strains.
5.3 Axisymmetric problems
Figure 5.7: Axisymmetric problem.
Strains considered (Fig. 5.7)
T
= [
vv

00

::

v:
] .
stresses considered: o
T
= [o
vv
o
00
o
::
o
v:
] .
with
vv
=
0&
0v
00
=
&
v
::
=
0
0:
v:
=
0&
0:
+
0
0v
.
where n and are the radial and axial displacements, respectively.
In matrix form, = [1] l with l =
_
n
_
and [1] =
_
_
0
0v
0
1
v
0
0
0
0:
0
0:
0
0v
_
_
.
With the same shape function matrix [`] and the same nodal displace-
ments vector l
c
as in plane stress, = [1] l
c
with [1] = [1] [`] .
[1] =
_
_
:
21
0 :
22
0 :
23
0
.
i
v
0
.
j
v
0
.
k
v
0
0 :
31
0 :
32
0 :
33
:
31
:
21
:
32
:
22
:
33
:
23
_
_
.
Note that
.
l
v
is not a constant, but a function of : and .. The constitutive
relationship is o = [1] (
0
) +o
0

with [1] =
1
(1+i)(12i)
_
_
1 i i i 0
i 1 i i 0
i i 1 i 0
0 0 0
12i
2
_
_
.
For a temperature change 1.
0
=
_
_
c
t
1
c
t
1
c
t
1
0
_
_
. Determination of
the element stiness matrix, load vectors, etc... is done using numerical
integration, because the integrands are now functions of : and ..
Example 13 [1
c
] =
_
\
e
[1]
T
[1] [1] d\ = 2:
_
e
[1]
T
[1] [1] :d:d..
5.4 Example
Assignment 4.
Chapter 6
Finite elements for plates and
shells
Plates are at; shells are curved plates.
6.1 Introduction - Bending of thin plates
6.1.1 Geometry
As shown in Fig. 6.1, the plate surfaces are at . = t,2 and its midsurface
at . = 0. The assumed basic geometry of the plate is as follows: t / and
t c (if t 0.2/ or 0.2c. thick plate). The deection n due to is assumed
to be much less than the thickness: n,t 1.
59
CHAPTER 6. FINITE ELEMENTS FOR PLATES AND SHELLS 60
Figure 6.1: Thin plate geometry.
6.1.2 Kirchho assumptions
They generalize ideas taken from beam bending. Consider a dierential slice
of plate (Fig. 6.2). Loading causes the plate to deform laterally in the
.-direction, and the deection n of point 1 is assumed to be a function of
r and only. That is, n = n(r. ). and the plate does not stretch in the
.-direction.
Figure 6.2: Dierential slice of plate, before loading (left), and after loading
(right). Similar displacements in . plane.
The Kirchho assumptions are as follows:
1. Normals remain normals. This implies that shear strains
j:
and
a:
are zero. However,
aj
,= 0 : the plate may twist in its plane.
2. Thickness changes can be neglected, and normals undergo no extension,
i.e.
::
= 0.
3. Normal stress o
::
has no eect on in-plane strains
aa
and
jj
. and is
considered negligible.
4. Membrane or in-plane forces are neglected here (they can be super-
imposed at a later stage). Therefore, the in-plane deformations in
the r- and -directions at the midsurface are assumed to be zero, i.e.
n(r. ) = (r. ) = 0.
According to the above assumptions, any point 1 has displacements
n = .c = .(
0&
0a
) in the r-direction
= ., = .(
0&
0j
) in the -direction
where
0&
0a
and
0&
0j
are the slopes at the midsurface.
aa
=
0&
0a
= .
0
2
&
0a
2
.
jj
=
0
0j
= .
0
2
&
0j
2
.
::
=
0&
0:
= 0
a:
=
0&
0a
+
0&
0:
= 0.
j:
=
0&
0j
+
0
0:
= 0.
aj
=
0&
0j
+
0
0a
= 2.
0
2
&
0a0j
Because in small deection theory, the square of a slope may be regarded
as negligible (recall that in 2-D,
1
vx
=
d
2
v
dx
2
[1+(
dv
dx
)
2
]
3=2

o
2
oa
2
, where :
a
is the radius
of curvature), the curvatures (
1
v
) of the plate are given simply as:
i
aa
=
0
2
&
0a
2
. i
jj
=
0
2
&
0j
2
and i
aj
= 2
0
2
&
0a0j
.
Therefore we have:
aa
= . i
aa
jj
= . i
jj
aj
= . i
aj
.
6.1.3 Constitutive relationships
Plane stress state can be used (Assumption 3), and then:
_
_
_
o
aa
o
jj
o
aj
_
_
_
=
1
1i
2
_
_
1 i 0
i 1 0
0 0
1i
2
_
_
_
_
_
. i
aa
. i
jj
. i
aj
_
_
_
.
These expressions clearly show that the stresses vanish at the midsurface,
and vary linearly over the plate thickness (Fig. 6.3). The stresses distrib-
uted over the side surfaces of the plate, while producing no net force, result
in bending and twisting moments. These moment resultants per unit length
(N.m/m=N) are denoted `
aa
. `
jj
and `
aj
with `
aa
=
_
t2
t2
. o
aa
d.
`
jj
=
_
t2
t2
. o
jj
d.
`
aj
=
_
t2
t2
. o
aj
d.
`
aj
= `
ja
because o
aj
= o
ja
.
Finally,
_
_
_
`
aa
`
jj
`
aj
_
_
_
= 1
_
_
1 i 0
i 1 0
0 0
1i
2
_
_
_
_
_
i
aa
i
jj
i
aj
_
_
_
with 1 =
1t
3
12(1i
2
)
:
bending rigidity of the plate.
The maximum stresses on each edge of the plate are located at the top
or bottom at . = t,2 :
o
aamax
=
6Axx
t
2
. o
jj max
=
6Ayy
t
2
. and o
aj max
=
6Axy
t
2
.
Note that these formulas are similar to the formula for beams, where
o
aa
=
Axx c
1
when applied to a unit width of plate , with c = t,2.
Introducing the shearing forces (or transverse shear line loads, Fig. 6.4),
Q
a
=
_
t2
t2
o
a:
d.
Q
j
=
_
t2
t2
o
j:
d.
Figure 6.3: Dierential plate element with stresses shown on edges.
Figure 6.4: Dierential plate element with dierential moments and forces.
6.1.4 Equilibrium equations
Note that Q and ` follow the same rules as stresses: a positive stress acts
on a positive face in the positive direction (Fig. 6.5).
Figure 6.5: Positive normal and shear stresses.
From Fig. 6.6, the equilibrium in the .-direction yields:
0Qx
0a
drd +
0Qy
0j
drd +drd = 0
Equilibrium of moments about the r-axis yields:
(
0Axy
0a
dr)d (
0Ayy
0j
d)dr +Q
j
drd = 0
Equilibrium of moments about the -axis yields:
(
0Ayx
0j
d)dr + (
0Axx
0a
dr)d Q
a
drd = 0
Fig. 6.8: Equilibirum of the elemental plate.
Then,
0Qx
0a
+
0Qy
0j
+ = 0
0Axx
0a
+
0Axy
0j
Q
a
= 0
0Ayy
0j
+
0Axy
0a
Q
j
= 0
. therefore Q
a
= 1
0
0a
(
0
2
&
0a
2
+
0
2
&
0j
2
)
Q
j
= 1
0
0j
(
0
2
&
0a
2
+
0
2
&
0j
2
)
.
and nally, 1(
0
4
&
0a
4
+ 2
0
4
&
0a
2
0j
2
+
0
4
&
0j
4
) = .
In concise form, \
4
n = ,1. These are the governing equations of plate
deection.
Note: the theory of thin plates neglects the eects on bending of o
::
. as
well as
a:
=
oxz
G
and
j:
=
oyz
G
. However, the shearing forces Q
a
and Q
j
resulting from o
a:
and o
j:
are not negligible. In fact, they are of the same
order of magnitude as the lateral loads and moments.
6.2 Finite element formulation (rectangular
element)
6.2.1 Element type
We will consider the 12-degree of freedom (d.o.f) at-plate bending element
shown in Fig. 6.7:
Figure 6.7: Rectangular shell element.
Each node has 3 d.o.f.: a transverse displacement n in the .-direction, a
rotation o
a
about the r-axis and a rotation o
j
about the -axis. The rotations
are related to the transverse displacement by o
a
=
0&
0j
and o
j
=
0&
0a
(a
negative n displacement is required to produce a positive rotation about the
-axis). The displacement vector for the element is:
l
c
T
= [n
i
o
ai
o
ji
n
)
o
a)
o
j)
n
I
o
aI
o
jI
n
|
o
a|
o
j|
] .
6.2.2 Displacement function
Because there are 12 d.o.f., we select a 12-term polynomial in r and as
follows:
n = c
1
+c
2
r +c
3
+c
4
r
2
+c
5
r +c
6
2
+c
7
r
3
+c
8
r
2
+c
9
r
2
+c
10
3
+
c
11
r
3
+c
12
r
3
.
The function allows for rigid-body motion and constant strain, but con-
tinuity of slopes between elements is not ensured. Along side i , (r-axis),
n = c
1
+c
2
r +c
4
r
2
+c
7
r
3
0&
0a
= c
2
+ 2c
4
r + 3c
7
r
2
0&
0j
= c
3
+c
5
r +c
8
r
2
+c
11
r
3
.
The displacement n is a cubic as used for the beam element, and
0&
0a
is
the same as in beam bending. The four constants c
1
. c
2
. c
4
. c
7
can be dened
by invoking the end point conditions of (n
i
. n
)
. o
ji
. o
j)
). Therefore, n and
0&
0a
are completely dened along side i ,. On the other hand, the normal
slope
0&
0j
is a cubic in r. with four constants, and only two d.of. left (o
ai
. o
a)
).
Therefore, this slope is not uniquely dened, and function n is said to be
nonconforming. However, this element has proven to give acceptable results,
and to converge.
Constants c
1
to c
12
can be determined by expressing 12 simultaneous
equations linking them to the values of n and its slopes at the nodes:
_
_
_
n
0&
0j
0&
0a
_
_
_
=
_
_
1 r r
2
r
2
r
3
r
2
r
2
3
r
3
r
3
0 0 1 0 r 2 0 r
2
2r 3
2
r
3
3r
2
0 1 0 2r 0 3r
2
2r
2
0 3r
2

3
_
_
_
_
c
1
.
.
.
c
12
_
_
_
_
_
n
0&
0j
0&
0a
_
_
_
= [1] Q .
At nodes i. ,. /. |. we require:
_
_
n
i
o
ai
o
ji
.
.
.
n
|
o
a|
o
j|
_
_
=
_
_
1 r
i

i
r
2
i
r
i
i

2
i
r
3
i
r
2
i
i
r
i
2
i

3
i
r
3
i
i
r
i
3
i
0 0 1 0 r
i
2
i
0 r
2
i
2r
i
i
3
2
i
r
3
i
3r
i
2
i
0 1 0 2r
i

i
0 3r
2
i
2r
i
i

2
i
0 3r
2
i
i

3
i
.
.
.
.
.
.
.
.
.
1 r
|

|
r
2
|
r
|
|

2
|
r
3
|
r
2
|
|
r
|
2
|

3
|
r
3
|
|
r
|
3
|
0 0 1 0 r
|
2
|
0 r
2
|
2r
|
|
3
2
|
r
3
|
3r
|
2
|
0 1 0 2r
|

|
0 3r
2
|
2r
|
|

2
|
0 3r
2
|
|

3
|
_
_
_
_
c
1
.
.
.
c
12
_
_
In compact matrix form, l
c
= [] Q and Q = []
1
l
c
.
Finally,
_
_
_
n
0&
0j
0&
0a
_
_
_
= [1] []
1
l
c
= [`] l
c
.
The specic form of [`] can be found in the documentation of nite
element packages.
6.2.3 Principle of virtual work
For an element,
_
\
e
o
T
o d\ =
_
\
e
ol
T
/ d\ +
_
e
ol
T
: d+
j=.
j=1
ol
T
,
j
.
\ol . o .
Focusing on the left-hand side,
_
\
e
o
T
o d\ =
_
\
e
_
_
_
o
aa
o
jj
o
aj
_
_
_
T
_
_
_
o
aa
o
jj
o
aj
_
_
_
d\
_
\
e
o
T
o d\ =
_
e
(
_
t2
t2
.
_
_
_
oi
aa
oi
jj
oi
aj
_
_
_
T
_
_
_
o
aa
o
jj
o
aj
_
_
_
d.)d
_
\
e
o
T
o d\ =
_
e
_
_
_
oi
aa
oi
jj
oi
aj
_
_
_
T
(
_
t2
t2
.
_
_
_
o
aa
o
jj
o
aj
_
_
_
d.)d
_
\
e
o
T
o d\ =
_
e
_
_
_
oi
aa
oi
jj
oi
aj
_
_
_
T
_
_
_
`
aa
`
jj
`
aj
_
_
_
d
_
\
e
o
T
o d\ =
_
e
_
_
_
oi
aa
oi
jj
oi
aj
_
_
_
T
[1]
_
_
_
i
aa
i
jj
i
aj
_
_
_
d
with [1] = 1
_
_
1 i 0
i 1 0
0 0
1i
2
_
_
.
From i =
_
_
_
i
aa
i
jj
i
aj
_
_
_
=
_
_
_
2c
4
6c
7
r 2c
8
6c
11
r
2c
6
2c
9
r 6c
10
6c
12
r
2c
5
4c
8
r 4c
9
6c
11
r
2
6c
12
2
_
_
_
or i = [1] Q . with Q = []
1
l
c
. one can write i = [1] l
c
.
where [1] = [1] []
1
.
Finally,
_
\
e
o
T
o d\ = ol
c
T
_
e
[1]
T
[1] [1] dl
c
.
6.2.4 Element stiness matrix [1
e
]
From above, [1
c
] =
_
e
[1]
T
[1] [1] d.
[1
c
] is a 12x12 matrix. It can be found in the documentation of nite
element packages.
6.2.5 Element nodal force vector ,
e
Body forces:
,
c
b
=
_
\
e
[`]
T
/ d\
Surface tractions:
,
c
c
=
_
e
[`]
T
d
For a uniform load acting over the surface of an element of dimensions
2/ 2c, the force and moments at node i are (similar expressions for ,. /. |):
_
_
_
,
&
i
,
0
xi
,
0
yi
_
_
_
= 4/c
_
_
_
1,4
c,12
/,12
_
_
_
.
Finally, [1
c
] l
c
= ,
c
. Apply boundary conditions and solve.
Chapter 7
Method of weighted residuals
So far, we have always used the principle of virtual displacements (or, equiva-
lently, the minimumof potential energy) towards nite element (FE) formula-
tion. What happens when such principles are not available? FE formulation
can in fact be generalized to virtually all problems describable by ordinary
or partial dierential equations (ODE or PDE). For simplicity, let us limit
ourselves here to linear statics, with [1]l = , being the end point of
the formulation.
7.1 The method of weighted residuals
7.1.1 General concepts
Consider a governing equation involving one independent variable:
, [1(r)] = 0 in domain
where 1 represents the function sought (e.g. temperature), which is a
function of r only. In addition, let us specify the boundary conditions (BC):
q
1
[1(r)] = 0 on
1
q
2
[1(r)] = 0 on
2
where
1
and
2
include only those parts of that are on the bound-
ary. Let us approximate the solution with an approximate function

1 given
by

1 =

1(r; c
1
. . c
a
) =

a
i=1
c
i
`
i
(r) which has : unknown constant
parameters c
i
and satises the boundary conditions exactly. The `
i
(r) are
referred to as trial functions. If the approximate solution

1 is substituted in
the governing equation, some residual error 1(r; c
1
. . c
a
) appears:
69
CHAPTER 7. METHOD OF WEIGHTED RESIDUALS 70
, [
1(r; c
1
. . c
a
)] = 1(r; c
1
. . c
a
).
The method of weighted residuals requires that parameters c
i
be deter-
mined by satisfying:
_
n
i
(r) 1(r; c
1
. . c
a
)dr = 0 . 1 _ i _ :
where the n
i
(r) are : arbitrary weighting functions. Many functions can
be used; the most popular methods are described below.
7.1.2 Point collocation
n
i
(r) = o(rr
i
) (Dirac function) such that
_
b
o
o(r r
i
)dr = 1 for r = r
i
_
b
o
o(r r
i
)dr = 0 for r ,= r
i
.
The r
i
are the collocation points and are selected arbitrarily by the ana-
lyst.
If
_
o(rr
i
) 1(r; c
1
. . c
a
)dr = 0 is evaluated at : collocation points,
: algebraic equations in : unknowns (c
i
) result:
1(r
1
; c
1
. . c
a
) = 0
1(r
2
; c
1
. . c
a
) = 0
.
.
.
1(r
a
; c
1
. . c
a
) = 0
Once the c
i
are determined, the approximate solution

1 is found, and the
problem is solved.
Example 14 solve the ODE:
o
2
T
oa
2
+ 1000r
2
= 0 for 0 _ r _ 1 with 1(0) =
1(1) = 0. using `
1
(r) = r(1 r
2
) (satises BC)
1 = c
1
`
1
(r) = c
1
r(1 r
2
)
1(r; c
1
) =
o
2 b
T
oa
2
+ 1000r
2
= 6c
1
r + 1000r
2
Collocation point (chosen): r
1
=
1
2
_
1
0
o(r
1
2
) 1(r; c
1
)dr = 0 =
_
1
0
o(r
1
2
) (6c
1
r + 1000r
2
)dr = 0
=1(r
1
; c
1
) = 0
=6c
1
(
1
2
) + 1000(
1
2
)
2
= 0
=c
1
=
1000
12
Finally,

1(r) =
1000
12
r(1 r
2
) (dashed line in gure below).
7.1.3 Subdomain collocation
The domain is subdivided into : subdomains over which n
i
(r) is unity
and 0 elsewhere.
n
1
(r) =
1 for r in
1
0 for r not in
1
.
.
.
n
a
(r) =
1 for r in
a
0 for r not in
a
Figure 7.1: Subdomain collocation.
If
_
n
i
(r) 1(r; c
1
. . c
a
)dr = 0 is evaluated for 1 _ i _ :. : integral
equations result:
_
1
1(r; c
1
. . c
a
)dr = 0
.
.
.
_
n
1(r; c
1
. . c
a
)dr = 0
These equations can be solved for c
i
. and

1 determined.
Example 15 same problem as above: solve the ODE:
o
2
T
oa
2
+ 1000r
2
= 0 for
0 _ r _ 1 with 1(0) = 1(1) = 0. using `
1
(r) = r(1 r
2
)
1 = c
1
`
1
(r) = c
1
r(1 r
2
)
Only one integral of the residual needs to be evaluated (because one
parameter, c
1
).
1(r; c
1
) =
o
2 b
T
oa
2
+ 1000r
2
= 6c
1
r + 1000r
2
_
1
0
1(r; c
1
)dr = 0 =c
1
=
1000
9
Finally,

1(r) =
1000
9
r(1 r
2
) (thin solid line in Fig. 7.2).
Note: this method is equivalent to the Ritz (variational) method.
7.1.4 Least squares
The method of least squares requires that the integral 1 of the square of the
residual 1 be minimized:
1 =
_
[1(r; c
1
. . c
a
)]
2
dr.
In other words, parameters c
i
need to be determined so that 1 is mini-
mized. This is done by writing:
01
0o
1
= 0 =
0
0o
1
_
[1(r; c
1
. . c
a
)]
2
dr = 0
.
.
.
01
0on
= 0 =
0
0on
_
[1(r; c
1
. . c
a
)]
2
dr = 0
Because the limits on the integral are not functions of c
i
. the order of
integration and dierentiation may be interchanged to give:
_
0
0o
1
[1(r; c
1
. . c
a
)]
2
dr = 0 etc... , or
_
1
01
0o
1
dr = 0 etc...
Finally, we observe that in this method, n
i
(r) =
01
0o
i
. 1 _ i _ :.
The : equations obtained will be used to solve for the : parameters c
i
.
Example 16 from above: 1(r; c
1
) =
o
2 b
T
oa
2
+ 1000r
2
= 6c
1
r + 1000r
2
01
0o
1
= 6r
_
1
0
1
01
0o
1
dr = 0 =c
1
=
1000
8
Finally,

1(r) =
1000
8
r(1 r
2
) (dotted line in Fig. 7.2).
7.1.5 Galerkin
In the Galerkin method, the trial functions `
i
(r) are used as weighting func-
tions, that is n
i
(r) = `
i
(r).
If
_
`
i
(r) 1(r; c
1
. . c
a
)dr = 0 is evaluated for 1 _ i _ :. : integral
equations result and can be solved for : parameters c
i
.
Example 17 `
1
(r) = r(1 r
2
) from above
_
1
0
r(1 r
2
)(6c
1
r + 1000r
2
)dr = 0 =c
1
=
5000
48
Finally,

1(r) =
5000
48
r(1 r
2
) (medium solid line in Fig. 7.2).
Note: this method is equivalent to the Rayleigh-Ritz (variational) method.
7.1.6 Comparison with exact solution
The exact solution to the above problem is 1(r) =
1000
12
r(1 r
3
) (thick solid
line in Fig. 7.2), while our approximate solution was assumed to be of the
form

1(r) = c
1
r(1 r
2
). None of the methods seems to do an outstanding
job, but the trial functions have not been worked on. In what follows, we
will use the Galerkin method in combination with trial functions that are no
longer applied to the entire domain, but applied locally over each element.
This is a very successful method.
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
0
10
20
30
40
50
x
y
Figure 7.2: Comparison between solutions.
7.2 The Galerkin nite element method
7.2.1 Formulation
Piecewise continuous trial functions (or shape functions) will be employed
for each element. For a 1-D problem, as in Fig. 7.3:
Figure 7.3: Piecewise formulation.
The problem domain c _ r _ / is divided into ` elements. A typical
element c has two nodes , and / with coordinates r
)
and r
I
. The eld variable
at nodes , and / has values
)
and
I
. Since each element has two nodes
and since a unique straight line may be drawn between two points (Fig. 7.4),
the interpolation polynomial to be used must be of the form
c
= :r +/. In
other words, we wish to express
c
(r) in the form
c
(r) =
)
`
)
(r) +
I
`
I
(r)
with `
)
(r) =
a
k
a
a
k
a
j
and `
I
(r) =
aa
j
a
k
a
j
. Clearly,
c
(r
)
) =
)
and
c
(r
I
) =
I
.
Figure 7.4: Linear interpolation over one element.
Let us now resolve the problem in Section 1 by the Galerkin FE method
using six equally spaced nodes.
Over one element, 1
c
= `
)
(r)1
)
+`
I
(r)1
I
. where 1
)
and 1
I
are analogous
to the previous unknown parameters c
1
and c
2
. For all the elements, from 1
to `.
A
c=1
_
a
k
a
j
`
)
(r)1
c
(r; 1
)
. 1
I
)dr = 0
A
c=1
_
a
k
a
j
`
I
(r)1
c
(r; 1
)
. 1
I
)dr = 0
In matrix notation, with [`]
T
=
_
`
)
(r)
`
I
(r)
_
. the previous equations
become:
A
c=1
_
a
k
a
j
[`]
T
1
c
(r; 1
)
. 1
I
)dr = 0.
The residual 1
c
for a typical element for the present problem is
1
c
=
o
2
T
e
oa
2
+ 1000r
2
.
Therefore,

A
c=1
_
a
k
a
j
[`]
T
(
o
2
T
e
oa
2
+1000r
2
)dr = 0. For simplicity, let us
drop the summation sign (corresponding to a routine assemblage).
_
a
k
a
j
[`]
T
(
o
2
T
e
oa
2
+ 1000r
2
)dr = 0
=
_
a
k
a
j
[`]
T o
2
T
e
oa
2
dr +
_
a
k
a
j
[`]
T
1000r
2
dr = 0
Recalling the integration by parts rule:
_
b
o
n
o
oa
dr = [n]
b
o
_
b
o

o&
oa
dr.
_
a
k
a
j
[`]
T o
2
T
e
oa
2
dr =
_
[`]
T oT
e
oa
a
k
a
j

_
a
k
a
j
o
oa
[`]
T oT
e
oa
dr. and then,
_
[`]
T oT
e
oa
a
k
a
j

_
a
k
a
j
o
oa
[`]
T oT
e
oa
dr +
_
a
k
a
j
[`]
T
1000r
2
dr = 0.
Since we already know that 1
c
= [`
)
`
I
]
_
1
)
1
I
_
= [`]l
c
.
_
[`]
T oT
e
oa
a
k
a
j

_
a
k
a
j
o
oa
[`]
T o
oa
[`]l
c
dr +
_
a
k
a
j
[`]
T
1000r
2
dr = 0.
Note that not using the interpolation to calculate
_
[`]
T oT
e
oa
a
k
a
j
allows for
imposing gradient boundary conditions, and is convenient.
Finally, [1
c
]l
c
= ,
c
where
[1
c
] =
_
a
k
a
j
o
oa
[`]
T o
oa
[`]dr (element stiness matrix) and
,
c
=
_
[`]
T oT
e
oa
a
k
a
j
+
_
a
k
a
j
[`]
T
1000r
2
dr (nodal force vector).
After assemblage and imposition of boundary conditions, the system
[1]l = , can be solved. The solution matches the exact solution.
Two linear shape functions applied locally give much superior results than a
cubic function applied globally (as done in Section 1 of this chapter).
7.2.2 Application: 1-D heat transfer in a pin n
Let us derive the governing equation for the temperature 1 (r) in a circular
pin n of varying cross sectional area and perimeter 1 (Fig. 7.5):
Figure 7.5: Pin n.
Conduction (through n) and convection (with surrounding uid) are
present.
a
(
a
+
o
oa
(
a
)dr) /1(1 1
o
)dr = 0
conduction conduction convection
in out out
a
: heat ux
/ : convective heat transfer coecient
1
o
: ambient uid temperature far removed from n
Finally, the steady state energy balance on an innitesimal slice yields:
o
oa
(
a
) /1(1 1
o
) = 0.
Applying Fouriers law of heat conduction (this is a constitutive relation-
ship, like o = 1) whereby
a
= /
oT
oa
. where / is the thermal conductivity,
and the minus sign is necessary because
a
is assumed positive in the direc-
tion of decreasing temperature, the governing equation for 1-D heat transfer
is:
o
oa
(/
oT
oa
) /1(1 1
o
) = 0 for 0 _ r _ 1
)
. subject to two boundary
conditions 1(0) = 1
b
oT
oa
(1
)
) = 0
.
Let us now derive equations to obtain an approximate solution using the
Galerkin FE method.
Element characteristics:
Figure 7.6: 1-D element with nodes , and /.
For a typical element (Fig. 7.6), the Galerkin formulation yields:
_
a
k
a
j
[`]
T
_
o
oa
(/
oT
oa
) /1(1 1
o
)
dr = 0.
Integrating
_
a
k
a
j
[`]
T o
oa
(/
oT
oa
)dr by parts,
_
a
k
a
j
[`]
T o
oa
(/
oT
oa
)dr =
_
[`]
T
/
oT
oa
a
k
a
j
These terms will cancel out upon assemblage,
and at r = 1
)
.
oT
oa
= 0
_
a
k
a
j
o
oa
[`]
T
/
oT
oa
dr.
Therefore,
_
a
k
a
j
o
oa
[`]
T
/
oT
oa
dr
_
a
k
a
j
[`]
T
/11dr +
_
a
k
a
j
[`]
T
/11
o
dr = 0.
By using 1 = [`]l
c
and noting that [`] is only a function of r. we
get [1
c
]l
c
= ,
c
where
[1
c
] = [1
c
a
] + [1
c
c
] and
[1
c
a
] =
_
a
k
a
j
o
oa
[`]
T
/
o
oa
[`]dr.
[1
c
c
] =
_
a
k
a
j
[`]
T
/1[`]dr. and
,
c
=
_
a
k
a
j
[`]
T
/11
o
dr.
The vector of nodal unknowns l
c
is l
c
T
= [1
)
1
I
] . and the shape
function matrix [`] is [`] = [`
)
`
I
].
Let us consider parameters /. /. . 1 and 1
o
to be constant in any given
element by using the values at the midpoint of the element, where r = r.
Then,
[1
c
a
] =
I
1
_
1 1
1 1
_
.
[1
c
c
] =
I11
6
_
2 1
1 2
_
. and
,
c
=
I11Ta
2
_
1
1
_
. where 1 = r
I
r
)
.
The assemblage of these 2x2 matrices is performed as usual to give a sys-
tem of ` equations in the ` unknown nodal temperatures, where ` = `1
is the number of elements. Overall, [1
o
]l
o
= ,
o
. and the prescribed
boundary temperature is applied, leading to [1]l = , which may be
solved for l.
Note: the Galerkin formulation leads to what is called a weak formula-
tion: because of the integration by parts, the order of the derivative in the
governing equation (strong formulation) is reduced in the working equations.
Chapter 8
Steady state thermal analysis
In the illustrative heat conduction problem that was considered in the pre-
vious chapter, thermal radiation and heat generation were not included. In
addition, we need to look into 2-D and axisymmetric problems. Let us start
with a 1-D problem.
8.1 1-D heat conduction
8.2 Governing equation
The governing equation for temperature 1 is given by (Fig. 8.1):
o
oa
(/
oT
oa
) /1(1 1
o
) o1(1
4
1
4
v
) +Q = 0 where
/ : thermal conductivity of the material (W/m
o
C)
/ : conductive heat transfer coecient (W/m
2o
C)
: emissivity of the body surface (-)
1 : perimeter
: cross-sectional area of body
1
o
: ambient uid temperature far removed from body
1
v
: receiver temperature for radiation
Q : heat generation rate per unit volume (W/m
3
)
o : Stefan-Boltzmann constant (5.67010
8
W/m
2
K
4
)
79
CHAPTER 8. STEADY STATE THERMAL ANALYSIS 80
Figure 8.1: 1-D heat conduction problem.
Note that:
/. /. may be temperature dependent
absolute temperature (K) must be used if radiation is present
8.3 FE formulation
Using the Galerkin method for ` elements such as in Fig. 8.2,
Figure 8.2: 1-D heat tranfer element.
A
c=1
_
|
e
[`]
T
1
c
dr = 0 where [`] is the shape matrix function. In
what follows, we drop the summation sign representing the routine assem-
blage process. Therefore,
_
|
e
[`]
T
1
c
dr =
_
|
e
[`]
T
_
o
oa
(/
oT
oa
) /1(1 1
o
) o1(1
4
1
4
v
) +Q
dr = 0.
Integrating the rst term by parts,
_
[`]
T
/
oT
oa
a
j
a
i

_
|
e
o
oa
[`]
T
/
oT
oa
dr
_
|
e
[`]
T
/11dr +
_
|
e
[`]
T
/11
o
dr
_
|
e
[`]
T
o11
4
dr
+
_
|
e
[`]
T
o11
4
v
dr +
_
|
e
[`]
T
Qdr = 0.
The rst term is related to the heat transfer rate per unit area in the
r-direction:
a
= /
oT
oa
. It will contribute to the assemblage equations for
only the two elements at each end of the body. Therefore, this term needs
to be evaluated for elements 1 and ` only.
Boundary conditions
Heat transfer to and from the ends can be a combination of convection
c
.
radiation
v
. and prescribed heat uxes
c
. (Think: into end:
a
.
c
; out of
end:
c
.
v
, Fig. 8.3).
Figure 8.3: Equilibrium of one end of the body.
An energy balance on the end of the body gives:
a
+
c
=
c
+
v
. or
a
=
c
+
v
c
.
Note that
c
is positive if the heat ux is imposed toward the surface.
At node ,.
c
= /
)
(1 1
o)
)
v
=
)
o(1
4
1
4
v)
)
c
=
)
.
Therefore, for the integrated term at r = r
)
.
[`]
T
/
oT
oa
[
a=a
j
= [`]
T
(
a
)[
a=a
j
= [`]
T
)
(
c
v
+
c
)[
a=a
j
= ([`]
T
/
)
)
1 + [`]
T
/
)
)
1
o)
[`]
T
)
o1
4
+ [`]
T
)
o1
4
v)
+ [`]
T
)
)[
a=a
j
Similarly at node i. with
a
=
c

v
+
c
(Fig. 8.4)
Figure 8.4: Equilibrium of other end of the body.
[`]
T
/
oT
oa
[
a=a
i
= [`]
T
i
(
c
v
+
c
)[
a=a
i
Finally, the integrated term is given by the SUM of both right-hand sides.
Element characteristics
As usual, 1 = [`]l
c
.
For convenience, we write 1
4
= 1
3
1 = ([`]l
c
)
3
[`]l
c
.
From the Galerkin method, we can write [1
c
]l
c
= ,
c
with
[1
c
] = [1
c
a
] + [1
c
c
] + [1
c
v
] + [1
c
c1
] + [1
c
v1
] and
,
c
= ,
c
c
+,
c
v
+,
c
Q
+,
c
c1
+,
c
v1
+,
c
q1
where subscript 1
denotes those terms arising from boundary conditions. Furthermore,
[1
c
a
] =
_
|
e
o
oa
[`]
T
/
o
oa
[`]dr.
[1
c
c
] =
_
|
e
[`]
T
/1[`]dr.
[1
c
v
] =
_
|
e
[`]
T
o1([`]l
c
)
3
[`]dr.
[1
c
c1
] = [`]
T
/
i
i
[`][
a=a
i
+ [`]
T
/
)
)
[`][
a=a
j
.
[1
c
v1
] = [`]
T
i
o([`]l
c
)
3
[`][
a=a
i
+ [`]
T
)
o([`]l
c
)
3
[`][
a=a
j
.
,
c
c
=
_
|
e
[`]
T
/11
o
dr.
,
c
v
=
_
|
e
[`]
T
o11
4
v
dr.
,
c
Q
=
_
|
e
[`]
T
Qdr.
,
c
c1
= [`]
T
/
i
i
1
oi
[
a=a
i
+ [`]
T
/
)
)
1
o)
[
a=a
j
.
,
c
v1
= [`]
T
i
o1
4
vi
[
a=a
i
+ [`]
T
)
o1
4
v)
[
a=a
j
. and
,
c
q1
= [`]
T
i
[
a=a
i
+ [`]
T
)
[
a=a
j
.
8.4 2-D heat conduction
Consider a plate of variable thickness (Fig. 8.5).
Figure 8.5: 2-D heat transfer problem (top view, left; side view, right).
8.4.1 Governing equation
Energy balance on volume tdrd yields:
0
0a
(/t
0T
0a
) +
0
0j
(/t
0T
0j
) /(1 1
o
) o(1
4
1
4
v
) +
c
+ Qt = 0, where
/. .
c
are the sums of their contributions on two lateral faces. Note that
c
is positive if directed as shown above. Using the governing equation to derive
the FE formulation will require use of the Green-Gauss theorem.
8.4.2 Green-Gauss theorem
It can be seen as a multidimensional version of integration by parts. Starting
from the divergence theorem:
in 3-D:
_
\
\
d\ =
_
S
: do, where \
=
0qx
0a
+
0qy
0j
+
0qz
0:
.
is any vector, d\ a volume element, do a surface element bounding \.

and

: a unit vector always normal to closed surface o and always directed
outward (Fig. 8.6).
Figure 8.6: Normal to the surface.
in 2-D:
_
d =
_
C
: dC, where the integration path around path C

must be performed counterclockwise.
Let us now derive the Green-Gauss theorem. Let
= ,
j . where , is a
scalar function of r. and ., and

j a vector. Then,
_
\
\
d\ =
_
\
\ (,
j )d\ =
_
S
,
: do.
With \ (,
j ) = ,\

j + \,
j . where \, denotes the gradient of

scalar function ,. i.e. \, =
0o
0a
i +
0o
0j
, +
0o
0:
/ .
one obtains: _
\
\ (,
j )d\ =
_
\
,\

j d\ +
_
\
\,
j d\ =
_
S
,
: do. from which

we derive
in 3-D:
_
\
,\
j d\ =
_
S
,
: do
_
\
\,
j d\
or
in 2-D:
_
,\
j d =
_
C
,
: dC
_
\,
j d.
Since j
a
is independent of j
j
and j
:
and vice-versa, it follows that
in 3-D:
_
\
,
0jx
0a
d\ =
_
S
,j
a
:
a
do
_
\
0o
0a
j
a
d\
_
\
,
0jy
0j
d\ =
_
S
,j
j
:
j
do
_
\
0o
0j
j
j
d\
_
\
,
0jz
0:
d\ =
_
S
,j
:
:
:
do
_
\
0o
0:
j
:
d\
and
in 2-D:
_
,
0jx
0a
d =
_
C
,j
a
:
a
dC
_
0o
0a
j
a
d
_
,
0jy
0j
d =
_
C
,j
j
:
j
dC
_
0o
0j
j
j
d
8.4.3 FE formulation
Again, the Galerkin method is used, and only one element c is considered.
With shape matrix function [`]. the Galerkin method requires
_
e
[`]
T
_
0
0a
(/t
0T
0a
) +
0
0j
(/t
0T
0j
) /(1 1
o
)
o(1
4
1
4
v
) +
c
+Qt
_
drd = 0.
Applying the Green-Gauss theorem to the two terms containing 2
ao
order
derivatives, we get
_
C
e
[`]
T
/t
0T
0a
:
a
dC
_
e
0
0a
[`]
T
/t
0T
0a
drd+
_
C
e
[`]
T
/t
0T
0j
:
j
dC
_
e
0
0j
[`]
T
/t
0T
0j
drd
e
[`]
T
/1drd +
_
e
[`]
T
/1
o
drd
e
[`]
T
o1
4
drd +
_
e
[`]
T
o1
4
v
drd
+
_
e
[`]
T
c
drd +
_
e
[`]
T
Qtdrd = 0
One can combine
_
C
e
[`]
T
(/
0T
0a
:
a
+ /
0T
0j
:
j
)tdC into
_
C
e
[`]
T
(
a
)tdC
where
a
represents the heat ux from conduction in the direction of the
outward normal of the boundary (Fig. 8.7)
Figure 8.7: Equilibrium of boundary.
a
=
: . with
a
=
a

i +
j
, .
a
= /
0T
0a
:
a
/
0T
0j
:
j
and :
a
=
: .
:
j
=
: .
On the global boundary, an energy balance (on an area basis) gives:
a
+
c1
=
c1
+
v1
imposed convective radiative
heat ux heat ux heat ux
or
a
=
c1
+
v1
c1
.
with
c1
= /
1
(1 1
o1
) and
v1
=
1
o(1
4
1
4
v1
).
Therefore,
_
C
e
[`]
T
(
a
)tdC =
_
C
e
[`]
T
c1
tdC
_
C
e
[`]
T
/
1
1tdC +
_
C
e
[`]
T
/
1
1
o1
tdC
_
C
e
[`]
T
1
o1
4
tdC +
_
C
e
[`]
T
1
o1
4
v1
tdC.
Recalling that 1 = [`]l
c
and 1
4
= ([`]l
c
)
3
[`]l
c
. the Galerkin
method condenses the problem into
[1
c
]l
c
= ,
c
with
[1
c
] = [1
c
aa
] + [1
c
jj
] + [1
c
c
] + [1
c
v
] + [1
c
c1
] + [1
c
v1
] and
,
c
= ,
c
c
+,
c
v
+,
c
q
+,
c
Q
+,
c
c1
+,
c
v1
+,
c
q1
.
The element stiness matrices are given by:
[1
c
aa
] =
_
e
0
0a
[`]
T
/t
0
0a
[`]drd. [1
c
jj
] =
_
e
0
0j
[`]
T
/t
0
0j
[`]drd.
[1
c
c
] =
_
e
[`]
T
/[`]drd.
[1
c
v
] =
_
e
[`]
T
o1([`]l
c
)
3
[`]drd.
[1
c
c1
] =
_
C
e
[`]
T
/
1
t[`]dC.
[1
c
v1
] =
_
C
e
[`]
T
1
ot([`]l
c
)
3
[`]dC.
and the element nodal force vectors by:
,
c
c
=
_
e
[`]
T
/1
o
drd.
,
c
v
=
_
e
[`]
T
o1
4
v
drd.
,
c
q
=
_
e
[`]
T
c
drd
,
c
Q
=
_
e
[`]
T
Qtdrd.
,
c
c1
=
_
C
e
[`]
T
/
1
t1
o1
dC.
,
c
v1
=
_
C
e
[`]
T
1
ot1
4
v1
dC. and
,
c
q1
=
_
C
e
[`]
T
c1
tdC.
8.5 Axisymmetric heat conduction
For simplicity, let us just consider an isotropic, heterogeneous body with
a volumetric heat source. The governing equation for axisymmetric heat
conduction is:
1
v
0
0v
(:/
0T
0v
) +
0
0:
(/
0T
0:
) +Q = 0.
The boundary conditions for convection and imposed heat uxes are given
later during the FE formulation, because it is more convenient then. Using
the Galerkin method on an axisymmetric element basis,
_
\
e
[`]
T
_
1
v
0
0v
(:/
0T
0v
) +
0
0:
(/
0T
0:
) +Q
d\ = 0. where d\ = 2::d:d..
Because : is independent of ..
0
0:
(/
0T
0:
): =
0
0:
(:/
0T
0:
).
Applying the Green-Gauss theorem to the two terms involving 2
ao
order
derivatives,
2:
_
C
e
[`]
T
:/
0T
0v
:
v
dC 2:
_
e
0
0v
[`]
T
:/
0T
0v
d:d.
+2:
_
C
e
[`]
T
:/
0T
0:
:
:
dC 2:
_
e
0
0:
[`]
T
:/
0T
0:
d:d.
+2:
_
e
[`]
T
Q:d:d. = 0.
On the global boundary (Fig. 8.8), an energy balance gives:
Figure 8.8: Equilibrium of boundary.
a
+
c1
=
c1
or
a
=
c1

c1
. with
a
= /
0T
0v
:
v
/
0T
0:
:
:
. and
c1
= /
1
(1 1
o1
).
Therefore,
2:
_
C
e
[`]
T
/
1
:(1 1
o1
)dC + 2:
_
C
e
[`]
T
c1
:dC
2:
_
e
0
0v
[`]
T
:/
0T
0v
d:d. 2:
_
e
0
0:
[`]
T
:/
0T
0:
d:d.
+2:
_
e
[`]
T
Q:d:d. = 0.
Recalling that 1 = [`]l
c
, we obtain [1
c
]l
c
= ,
c
with
[1
c
] = [1
c
vv
] + [1
c
::
] + [1
c
c1
] and
,
c
= ,
c
Q
+,
c
q1
+,
c
c1
.
The element stiness matrices are given by:
[1
c
vv
] = 2:
_
e
0
0v
[`]
T
:/
0
0v
[`]d:d..
[1
c
::
] = 2:
_
e
0
0:
[`]
T
:/
0
0:
[`]d:d..
[1
c
c1
] = 2:
_
C
e
[`]
T
/
1
:[`]dC.
and the element nodal force vectors by:
,
c
Q
= 2:
_
e
[`]
T
Q:d:d..
,
c
q1
= 2:
_
C
e
[`]
T
c1
:dC. and
,
c
c1
= 2:
_
C
e
[`]
T
/
1
:1
o1
dC.
Chapter 9
Fluid ow analysis
Note: please refer to a uid mechanics course for details on the derivation of
the governing equations used in this chapter.
9.1 2-D potential ow
An ideal uid is incompressible and frictionless. The ow of an ideal uid
is called potential ow. No real uid is frictionless, but outside the bound-
ary layer, frictional eects may be neglected. The ow of an ideal uid is
also irrotational. Applications of potential ow include ows around airfoils,
groundwater ow, ows through reservoirs, around corners.
The continuity equation in 2-D is given by
0&
0a
+
0
0j
= 0, and the irrota-
tional ow condition by
0&
0j

0
0a
= 0. where n and represent the r and
components of the uid velocity. The following developments are restricted
to ow around symmetric bodies without lift (Fig. 9.1).
Figure 9.1: 2-D potential ow.
89
CHAPTER 9. FLUID FLOW ANALYSIS 90
9.1.1 Velocity potential formulation
The irrotational ow condition is satised exactly if the velocity potential
c is dened by n =
0
0a
and =
0
0j
. Subbing these equations into the
continuity equation yields Laplaces equation:
0
2
0a
2
+
0
2
0j
2
= 0.
On an element basis, the Galerkin method gives
_
e
[`]
T
_
0
2
0a
2
+
0
2
0j
2
_
drd = 0. where the shape function matrix [`]
depends on the element type used. From the Green-Gauss theorem, we get
_
C
e
[`]
T 0
0a
:
a
dC
_
e
0
0a
[`]
T 0
0a
drd
+
_
C
e
[`]
T 0
0j
:
j
dC
_
e
0
0j
[`]
T 0
0j
drd = 0.
In other words,
_
C
e
[`]
T
n:
a
dC
_
C
e
[`]
T
:
j
dC
e
0
0a
[`]
T 0
0a
drd
_
e
0
0j
[`]
T 0
0j
drd = 0.
Representing the velocity potential c as c = [`]l
c
. where l
c
con-
tains the values of the potential at the nodes of element c. the above equa-
tion can be combined into [1
c
]l
c
= ,
c
with [1
c
] = [1
c
aa
] + [1
c
jj
] and
,
c
= ,
c
&1
+,
c
1
.
The element stiness matrices are dened by
[1
c
aa
] =
_
e
0
0a
[`]
T 0
0a
[`]drd.
[1
c
jj
] =
_
e
0
0j
[`]
T 0
0j
[`]drd
and the element force vectors by
,
c
&1
=
_
C
e
[`]
T
n:
a
dC and ,
c
1
=
_
C
e
[`]
T
:
j
dC.
If a 3-node triangular element is used (see Chapter 4), then
[1
c
aa
] =
_
_
:
2
21
:
21
:
22
:
21
:
23
:
22
:
21
:
2
22
:
22
:
23
:
23
:
21
:
23
:
22
:
2
23
_
_
[1
c
jj
] =
_
_
:
2
31
:
31
:
32
:
31
:
33
:
32
:
31
:
2
32
:
32
:
33
:
33
:
31
:
33
:
32
:
2
33
_
_
The integrals for ,
c
need to be evaluated on the element boundaries
only. However, each direction cosine takes on opposite signs on legs shared
by two elements. Therefore, the contributions from internal legs cancel out
during assemblage, and ,
c
needs to be evaluated only for those elements
with legs on the global boundaries.
Example 18 evaluate ,
c
&1
in the case below (Fig.9.2):
Figure 9.2: Boundary loading.
On leg i,. we have `
I
= 1
I
= 0. thus n = 1
i
n
i
+ 1
)
n
)
(linear variation
as required)
,
c
&1
=
_
|
ij
_
_
_
1
i
1
)
0
_
_
_
(1
i
n
i
+1
)
n
)
):
a
d|, where |
i)
is the length of leg i,.
,
c
&1
=
|
ij
ax
6
_
_
_
2n
i
+n
)
n
i
+ 2n
)
0
_
_
_
.
Similarly, ,
c
1
=
|
ij
ay
6
_
_
_
2
i
+
)
i
+ 2
)
0
_
_
_
.
One method for determining :
a
and :
j
for any leg of the element is now
presented. Let us dene vector

:
i)
running from node i to node , along leg
i, (Fig. 9.3).
Figure 9.3: Normal components determination.
:
i)
= (r
)
r
i
)
i + (
)
i
)
: = :
a
i +:
j
: =
!
v
ij
!
I
[
!
v
ij
!
I [
. where cross product

c
/ = det
_
_
i

,

/
c
a
c
j
c
:
/
a
/
j
/
:
_
_
.
Finally, :
a
=
j
j
j
i
|
ij
. :
j
=
a
j
a
i
|
ij
. and |
i)
=
_
(r
)
r
i
)
2
+ (
)
i
)
2
.
Similar results are obtained for leg ,/ and leg /i.
One may also write \
a
= n:
a
+:
j
. and then
,
c
\ 1
= ,
c
&1
+ ,
c
1
=
_
C
e
[`]
T
\
a
dC, with \
a
=
0
0a
. where :
represents the coordinate in the direction of the outward normal.
The average velocities n and for a typical triangular element c can be
determined from
n =
0
0a
=
0
0a
[`]l
c
= :
21
c
i
:
22
c
)
:
23
c
I
=
0
0j
=
0
0j
[`]l
c
= :
31
c
i
:
32
c
)
:
33
c
I
where the values of c
i
. c
)
. c
I
are known from the solution of [1]l =
,.
9.1.2 Stream function formulation
The continuity equation is satised exactly if the stream function is dened
by n =
0
0j
and =
0
0a
. Subbing these expressions into the irrotational ow
condition yields
0
2
0a
2
+
0
2
0j
2
= 0.
Lines of constant (streamlines) are perpendicular to the lines of constant
c. The FE formulation is very similar to that with the velocity potential and
is not derived here.
9.2 2-D incompressible ow
Let us consider the laminar ow of a viscous uid for the case of moderate
Reynolds numbers (Fig. 9.4). For example, let us determine the 2-D velocity
eld for a uid moving through a duct with an obstruction.
Figure 9.4: 2-D incompressible ow.
The continuity equation for an incompressible uid is given by
0&
0a
+
0
0j
= 0.
The Navier-Stokes equations in the r- and -directions are
j(n
0&
0a
+
0&
0j
) =
0j
0a
+j(
0
2
&
0a
2
+
0
2
&
0j
2
) +/
a
j(n
0
0a
+
0
0j
) =
0j
0j
+j(
0
2
0a
2
+
0
2
0j
2
) +/
j
where j is the uid density, j is the pressure, j the absolute viscosity, /
a
and /
j
body forces per unit volume in the r- and -directions, respectively.
Let us note at once that the left-hand side of the above equations makes
the problem nonlinear. Applying the Galerkin method to the continuity
equation, we get
_
e
[`]
T
(
0&
0a
+
0
0j
)tdrd = 0. where t is the thickness (can be taken as
unity) introduced so that the terms in the integral are volumetric ow rates.
Assuming n = [`]l
c
&
and = [`]l
c
with
l
c
&
T
= [n
i
n
)
n
I
] and l
c
T
= [
i

)

I
] for triangular elements,
the previous equation can be recast in matrix form as
[1
c
&
]l
c
&
+ [1
c
]l
c
= 0. where
[1
c
&
] =
_
e
[`]
T 0
0a
[`]tdrd and [1
c
] =
_
e
[`]
T 0
0j
[`]tdrd.
Applying now the Galerkin method to the rst Navier-Stokes equation
yields
_
e
[`]
T
(j(n
0&
0a
+
0&
0j
) +
0j
0a
j(
0
2
&
0a
2
+
0
2
&
0j
2
) /
a
)tdrd = 0
where t is introduced so that the terms in the integral have units of
force. Applying the Green-Gauss theorem to the terms involving 2
ao
order
derivatives, and representing pressure j over an element by j = [`]l
c
j
. one
obtains
_
[1
c
&&
] + [1
c
&
] + [1
c
&a
] + [1
c
&j
]
_
l
c
&
+ [1
c
ja
]l
c
j
= ,
c
a
, where
[1
c
&&
] =
_
e
[`]
T
j[`]l
c
&
0
0a
[`]tdrd.
[1
c
&
] =
_
e
[`]
T
j[`]l
c
0
0j
[`]tdrd.
[1
c
&a
] =
_
e
0
0a
[`]
T
j
0
0a
[`]tdrd.
[1
c
&j
] =
_
e
0
0j
[`]
T
j
0
0j
[`]tdrd.
[1
c
ja
] =
_
e
[`]
T 0
0a
[`]tdrd. and
,
c
a
=
_
e
[`]
T
/
a
tdrd.
The integrals around the element bounding C
c
need not be considered, be-
cause the velocity components are usually prescribed on the global boundary.
In a similar fashion, if the Galerkin method is applied to the second Navier-
Stokes equation,
_
[1
c
&
] + [1
c

] + [1
c
a
] + [1
c
j
]
_
l
c
+ [1
c
jj
]l
c
j
= ,
c
j
.
where
[1
c
&
] = [1
c
&&
].
[1
c

] = [1
c
&
].
[1
c
a
] = [1
c
&a
].
[1
c
j
] = [1
c
&j
].
[1
c
jj
] =
_
e
[`]
T 0
0j
[`]tdrd. and
,
c
j
=
_
e
[`]
T
/
j
tdrd.
Combining all the equations relative to one element, one gets
_
_
[1
c
&
] [1
c
] [0]
[1
c
&&
] + [1
c
&
] + [1
c
&a
] + [1
c
&j
] [0] [1
c
ja
]
[0] [1
c
&
] + [1
c

] + [1
c
a
] + [1
c
j
] [1
c
jj
]
_
_
_
_
_
l
c
&
l
c
l
c
j
_
_
_
=
_
_
_
0
,
c
a
,
c
j
_
_
_
.
A possible assemblage would produce a vector of nodal unknowns as
l
T
= [n
1
n
a

1

a
j
1
j] . but this would re-
quire increased bandwidth of the assemblage system equations. Instead, lets
denote the above submatrices and subvectors as follows
_
_
[1
11
] [1
12
[1
13
]
[1
21
] [1
22
] [1
23
]
[1
31
] [1
32
] [1
33
]
_
_
_
_
_
l
c
&
l
c
l
c
j
_
_
_
=
_
_
_
,
1
,
2
,
3
_
_
_
and let us now use subscripts to indicate the entries within each subma-
trix, e.g.
[1
c
&
] = [1
11
] =
_
_
[1
11
11
] [1
11
12
] [1
11
13
]
[1
11
21
] [1
11
22
] [1
11
23
]
[1
11
31
] [1
11
32
] [1
11
33
]
_
_
and ,
c
a
= ,
2
=
_
_
_
,
2
1
,
2
2
,
2
3
_
_
_
.
On an element basis, with l
c
T
= [n
i

i
j
i
n
)

)
j
)
n
I

I
j
I
] .
_
_
[1
11
11
] [1
12
11
[1
13
11
]
[1
21
11
] [1
22
11
] [1
23
11
]
[1
31
11
] [1
32
11
] [1
33
11
]
[1
11
12
] [1
12
12
[1
13
12
]
[1
21
12
] [1
22
12
] [1
23
12
]
[1
31
12
] [1
32
12
] [1
33
12
]
[1
11
13
] [1
12
13
[1
13
13
]
[1
21
13
] [1
22
13
] [1
23
13
]
[1
31
13
] [1
32
13
] [1
33
13
]
[1
11
21
] [1
12
21
[1
13
21
]
[1
21
21
] [1
22
21
] [1
23
21
]
[1
31
21
] [1
32
21
] [1
33
21
]
[1
11
22
] [1
12
22
[1
13
22
]
[1
21
22
] [1
22
22
] [1
23
22
]
[1
31
22
] [1
32
22
] [1
33
22
]
[1
11
23
] [1
12
23
[1
13
23
]
[1
21
23
] [1
22
23
] [1
23
23
]
[1
31
23
] [1
32
23
] [1
33
23
]
[1
11
31
] [1
12
31
[1
13
31
]
[1
21
31
] [1
22
31
] [1
23
31
]
[1
31
31
] [1
32
31
] [1
33
31
]
[1
11
22
] [1
12
22
[1
13
22
]
[1
21
22
] [1
22
22
] [1
23
22
]
[1
31
22
] [1
32
22
] [1
33
22
]
[1
11
33
] [1
12
33
[1
13
33
]
[1
21
33
] [1
22
33
] [1
23
33
]
[1
31
33
] [1
32
33
] [1
33
33
]
_
_
_
_
n
i
i
j
i
n
)
)
j
)
n
I
I
j
I
_
_
=
_
_
,
1
1
,
2
1
,
3
1
,
1
2
,
2
2
,
3
2
,
1
3
,
2
3
,
3
3
_
_
.
The assemblage is now possible such that the vector of nodal unknowns
is
l
T
= [n
1

1
j
1
n
a

a
j
a
] .
Note that the matrix is not symmetric, and that the system is nonlinear
(matrix depends on l).
Chapter 10
Transient and dynamic analyses
Transient and dynamic refer both to unsteady, time-dependent or propaga-
tion problems. Transient is mostly used for non structural analyses, while dy-
namic is used to describe structural problems. So far, we have only dealt with
discretizations in space, e.g. c(r. ) = [`(r. )]c
c
. where c
c
is composed
of constants. In time-dependent problems, time t is important. We could
use a space and time discretization such that c(r. . t) = [`(r. . t)]c
c
.
but this increases the number of dimensions by one. This complexity can be
avoided by using partial discretization, where the vector of nodal unknowns
is a function of time: c(r. . t) = [`(r. )]c
c
(t). We will no longer have
[1
c
]c
c
= ,
c
. additional terms will appear.
10.1 Dynamic structural analysis
Using dAlemberts principle, the elemental inertial and damping forces d,
1
and d,
1
are given by
d,
1
= j nd\ and d,
1
= [j] _ nd\. where j is the mass density,
[j] is the viscous matrix, n the acceleration vector and _ n the velocity
vector, and the minus sign describes that these are external forces.
Then, the principle of virtual work in dynamics, for one element, is
_
\
e
o
T
o d\ =
_
\
e
on
T
/ d\ +
_
e
on
T
: d +
j=.
j=1
on
T
,
j
_
\
e
on
T
j nd\
_
\
e
on
T
[j] _ nd\.
\on . o
Note that, with n = [`]l
c
.
96
CHAPTER 10. TRANSIENT AND DYNAMIC ANALYSES 97
_ n =
o
ot
n =
o
ot
([`]l
c
) = [`]
o
ot
l
c
= [`]
_
l
c
. and similarly,
n = [`]
l
c
.
One nally obtains [`
c
]
l
c
+[1
c
]
_
l
c
+[1
c
]l
c
= ,
c
. where [1
c
]
and ,
c
are the same as in statics, and [`
c
] is the element mass matrix,
while [1
c
] is the element damping matrix.
[`
c
] =
_
\
e
[`]
T
j[`]d\
[1
c
] =
_
\
e
[`]
T
[j][`]d\
The viscous matrix [j] is not known in general. After assemblage, [1] is
usually taken to be of the form
[1] = c[`] + ,[1]. where c. , are experimentally determined constants
(Rayleigh damping). In any event, the assemblage system of equations can
be written as
[`]
l+[1]
_
l+[1]l = , once boundary conditions are applied.
Example 19 determine [`
c
] for a 2-node, 1-D element
[`
c
] =
_
\
e
[`]
T
j[`]d\ =
_
|
e
_
1
1
1
2
_
j [1
1
1
2
] dr
[`
c
] =
_
|
e
_
1
2
1
1
1
1
2
1
2
1
1
1
2
2
_
jdr
[`
c
] =
j1
6
_
2 1
1 2
_
10.2 Transient thermal analysis
If an energy storage mechanism is included into the steady state heat transfer
equation, the 1-D governing equation becomes
jc
0T
0t
=
0
0a
(/
0T
0a
) /1(1 1
o
) o1(1
4
1
4
v
) +Q. where
j : mass density
c : specic heat (at constant volume or constant pressure).
In 2-D,
jct
0T
0t
=
0
0a
(/t
0T
0a
) +
0
0j
(/t
0T
0j
) /(1 1
o
) o(1
4
1
4
v
) +
c
+Qt. where
t : thickness of the 2-D region (not to be confused with time t).
In 3-D,
jc
0T
0t
=
0
0a
(/
0T
0a
) +
0
0j
(/
0T
0j
) +
0
0:
(/
0T
0:
) +Q.
For axisymmetric problems,
jc
0T
0t
=
1
v
0
0v
(:/
0T
0v
) +
0
0:
(/
0T
0:
) +Q.
Using partial discretization, 1 = [`]l
c
(t). where the shape function
matrix [`] is unchanged compared to the steady state case,
and l
c
(t)
T
= [1
1
(t) 1
a
(t)] . where 1
i
(t) is the temperature of
node i at time t. and : is the number of nodes dening element c.
Applying the Galerkin method to the above equation in the 2-D case
yields
_
e
[`]
T
_
jct
0T
0t

0
0a
(/t
0T
0a
)
0
0j
(/t
0T
0j
)+
/(1 1
o
) +o(1
4
1
4
v
)
c
Qt
_
drd = 0. from which
is derived
[C
c
] =
_
e
[`]
T
jct[`]drd in 2-D,
[C
c
] =
_
|
e
[`]
T
jc[`]dr in 1-D,
[C
c
] =
_
\
e
[`]
T
jc[`]drdd. in 3-D, and
[C
c
] =
_
e
2:[`]
T
jc:[`]d:d. for axisymmetric problems.
Example 20 determine [C
c
] for a 3-node triangular element in 2-D
[C
c
] =
_
e
[`]
T
jct[`]drd =
[C
c
] =
_
e
_
_
_
1
1
1
2
1
3
_
_
_
jct[1
1
1
2
1
3
]drd
[C
c
] =
_
e
_
_
1
2
1
1
1
1
2
1
1
1
3
1
2
1
1
1
2
2
1
2
1
3
1
3
1
1
1
3
1
2
1
2
3
_
_
jctdrd
[C
c
] =
jct
12
_
_
2 1 1
1 2 1
1 1 2
_
_
. where is the element area.
Assemblage of the matrices is done as usual, and the initial and boundary
conditions must be imposed before solving.
10.3 Lumped versus consistent matrices
Consistent matrices are those obtained directly fromtheir denitions for mass
and capacitance matrices. They are not necessarily diagonal. By contrast,
lumped mass and capacitance matrices are made diagonal for numerical ad-
vantages. There are dierent ways of doing it. A rule of thumb consists in
summing the entries in a given row in the consistent matrix , dividing the
result by the total of all entries, and allocating this result to the diagonal
entry of the row under consideration.
Example 21 [`
c
] =
j1
6
_
2 1
1 2
_
becomes [`
c
] =
j1
2
_
1 0
0 1
_
.
Advantage: initial oscillations of the solution do not occur if lumped
matrices are used (Fig. 10.1).
Figure 10.1: Comparison of transient solutions, with consistent capacitance
matrix versus lumped capacitance matrix.
10.4 Solution methods
10.4.1 Two-point recurrence scheme
Let us consider equation [C]
_
l + [1]l = ,.
Temporal elements can be introduced to discretize the time domain, and
the weighted-residual method can be used from time t to t+t. or t
i
to t
i+1
:
_
t
i+1
t
i
\
_
[C]
_
l + [1]l ,
_
dt = 0. where the weighting func-
tion \ is a scalar.
With the help of the shape functions, we can represent vector l by
l = `
i
l
i
+`
i+1
l
i+1
. where the shape functions are linear in time
(Fig. 10.2).
Note that l = l
i
at t = t
i
and l = l
i+1
at t = t
i+1
.
Figure 10.2: Linear timewise interpolation between time t
i
and t
i+1
.
For convenience, let us write `
i
= 1 and `
i+1
= . where =
tt
i
t
with
t = t
i+1
t
i
. Then, 0 _ _ 1. and
l = (1 )l
i
+l
i+1
. from which it follows that
_
l =
o
ot
l =
o
o
l
o
ot
_
l =
o
o
[(1 )l
i
+l
i+1
]
o
ot
. or
_
l =
1
t
l
i
+
1
t
l
i+1
.
Similarly,
, = (1),
i
+,
i+1
. therefore the weighted residual method yields,
with dt = t d
_
1
0
\
_
_
_
[C]
_
1
t
l
i
+
1
t
l
i+1
+[1] [(1 )l
i
+l
i+1
]
[(1 ),
i
+,
i+1
]
_
_
_
t d = 0. or
_
[C]
_
1
0
\d + [1]t
_
1
0
\d
_
l
i+1
=
_
[C]
_
1
0
\d [1]t
_
1
0
\(1 )d
_
l
i
+
_
t
_
1
0
\(1 )d
_
,
i
+
_
t
_
1
0
\d
_
,
i+1
.
Dividing both sides by
_
1
0
\d gives:
[[C] +o[1]t ] l
i+1
=
[[C] (1 o)[1]t ] l
i
+ [(1 o),
i
+o,
i+1
] t . where
o =
_
1
0
Wo
_
1
0
Wo
.
Depending on the choice of weighting function \. o varies.
Point collocation
at time t
i
: o = 0 (forward dierence, or Eulers method)
at time
t
i
+t
i+1
2
: o =
1
2
(central dierence, or Crank-Nicholsons method)
at time t
i+1
: o = 1 (backward dierence)
If o = 0. [C] l
i+1
= [[C] [1]t ] l
i
+,
i
t
known at any time t. including t = 0
If lumped (diagonal) [C] is used, l
i+1
can be explicitly calculated:
explicit method.
If consistent [C] is used, [C]
1
must be calculated: implicit method.
Potential problems with stability and accuracy of solution.
If o = 1. [[C] + [1]t ] l
i+1
= [C]l
i
+,
i+1
t
known at any time t.
because ,
i+1
is the forcing function
Implicit solution because [1] cannot be lumped.
Method always stable, but accuracy deteriorates if t increases.
If o =
1
2
.
_
[C] + [1]
t
2
l
i+1
=
_
[C] [1]
t
2
l
i
+
t
2
[,
i
+,
i+1
]
known at any time t
Implicit solution.
Stability problems if t above critical value.
Accuracy decreases as t decreases.
Most accurate method of the three.
Subdomain collocation
with \ = 1 from t
i
to t
i+1
: o =
1
2
.
Galerkin method
with \ = `
i
= 1 : o =
1
3
with \ = `
i+1
= : o =
2
3
If o =
2
3
. the solution is more accurate than if o = 1. and more stable
than if o =
1
2
.
Note: if [C] or [1] is temperature dependent, the equations are nonlinear
and require an iterative solution such that l
)+1
= (1o)l
i
)
+ol
i+1
)
.
where , is the iteration number. Exception: when o = 0. no iteration is
necessary, because all is known at t.
10.4.2 Three-point recurrence scheme
Let us consider equation [`]
l + [1]
_
l + [1]l = ,. and let t
i1
. t
i
and t
i+1
respectively represent t t. t. and t + t. The weighted residual
method requires
_
t
i+1
t
i
1
\
_
[`]
l + [1]
_
l
+[1]l ,
_
dt = 0.
Assuming `
i1
=
1
2
:(1 :)
`
i
=
1
2
:(1 +:)
`
i+1
= (1 +:)(1 :)
with : =
tt
i
t
. t = t
i+1
t
i1
.
t
i1
_ t _ t
i+1
. Then, 1 _ : _ 1. and d: =
ot
t
.
Figure 10.3: Quadratic interpolation between t
i1
. t
i
and t
i+1
.
In a fashion similar to what was done with the two-point scheme, we get:
[[`] +[1]t +`[1]t
2
]l
i+1
=
[2[`] (1 2)[1]t (
1
2
2` +)[1]t
2
]l
i
+[[`] + (1 )[1]t (
1
2
+` )[1]t
2
]l
i1
+(
1
2
+` )t
2
],
i1
+ (
1
2
2` +)t
2
],
i
`t
2
],
i+1
.
where ` =
_
+1
1
1
2
Wv(1+v)ov
_
+1
1
1
2
Wov
. =
_
+1
1
W(
1
2
+v)ov
_
+1
1
1
2
Wov
.
This is because
_
l =
1
t
[(
1
2
+:)l
i1
+ (
1
2
+:)l
i
2:l
i+1
].
l =
1
t
2
[l
i1
2l
i
+l
i+1
]. and
, = `
i1
,
i1
+`
i
,
i
+`
i+1
,
i+1
.
\ `
Point collocation at t
i1
at t
i
at t
i+1
0
0
1
1
2
1
2
3
2
Subdomain collocation
1
6
1
2
Galerkin,based on `
i1
on `
i
on `
i+1
1
5
1
10
4
5
1
2
1
2
3
2
10.4.3 Initial conditions
In the three-point scheme, the values of l
i
and l
i1
are needed to de-
termine l
i+1
. even at i = 1.
Let us write [`]
_
\ + [1]\ + [1]l = , with
_
l = \ . The
initial displacements and velocities are noted l
0
and \
0
. We need l
1
.
Euler starting method (less accurate)
fl
i
gfl
i1
g
t
= \
i1
. therefore l
1
= l
0
+\
0
t
Crank-Nicholson starting method (more accurate)
[`]
f\
i
gf\
i1
g
t
+ [1]
f\
i
g+f\
i1
g
2
+ [1]
fl
i
g+fl
i1
g
2
=
1
2
(,
i
+,
i1
). with
fl
i
gfl
i1
g
t
=
f\
i
g+f\
i1
g
2
. therefore
l
1
=
_
2
[A]
t
+ [1] + [1]
t
2
_
1
__
2
[A]
t
+ [1] [1]
t
2
_
l
0
+ 2[`]\
0
+
t
2
,
1
+
t
2
,
0
_
.
10.5 Introduction to modal analysis
Many times in dynamic structural analysis, the nodal displacements are
not really needed. Instead, the natural frequencies of the sustained vi-
brations are of interest. Natural frequencies occur when the forcing func-
tion ,(t) is identically zero, and damping is negligible. Therefore, let us
consider [`]
l + [1]l = 0. and represent l as l = lc

).t
(,
2
= 1), where l is the eigenvector corresponding to natural frequency
.. Then,
l = (,.)
2
lc
).t
= .
2
lc
).t
. The initial equation be-
comes (.
2
[`] + [1])lc
).t
= 0. Since only non-trivial solutions for
l are sought, the determinant det(.
2
[`] + [1]) must be zero, guar-
anteeing that matrix .
2
[`] + [1] is singular. det(.
2
[`] + [1]) = 0
can be solved for the natural frequencies . of the structure. If ` nodes
are used in the discretization of the structure, and each node has : d.o.f.,
we get :` natural frequencies (vs. an innite number for a continuous
structure, i.e. not discrete). The frequencies obtained on the low end of
the spectrum are quite accurate if enough elements are used. Eigenvectors
(i.e. such that (.
2
[`] +[1])l
)
= 0) are usually normalized such that
l
)
T
[`]l
)
= 1. They represent mode shapes. They are also used to con-
struct solutions to problems with damping and/or forcing functions (mode
superposition).

Labrosse - Finite Element Analysis

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Labrosse - Finite Element Analysis

Caricato da

Copyright:

Formati disponibili

MCG 4102/5108 Finite Element

represents the local stiness matrix,

, . In the bar (rotated)

[1] is the global

1 is required to maintain static equilibrium.

and the two shear (tangential) stresses as

. For an innitesimal volume element

Example 6 consider the elongation and an axial force 1 in a uniaxial

` = 0 : \ dr +d` = 0. therefore \ (r) =

= 504 lbf and 1

CHAPTER5. STRESS ANALYSIS IN2-DWITHLINEARTRIANGULARELEMENT58

is any vector, d\ a volume element, do a surface element bounding \.

: dC, where the integration path around path C

j . where \, denotes the gradient of

: do. from which

l + [1]l = 0. and represent l as l = lc

Potrebbero piacerti anche