P. 1
MIT K.J. Bathe's Courses

# MIT K.J. Bathe's Courses

|Views: 25|Likes:

Published by: Bubu Muncus on Dec 27, 2011

### Availability:

Read on Scribd mobile: iPhone, iPad and Android.
See more
See less

11/05/2012

pdf

text

original

# _____t t ~

- - -
Massachusetts Institute of Technology
MIT VideoCOurse
Video Course Study Guide
Finite Element
Procedures for Solids
and Structures­
Linear Analysis
Klaus-JOrgen Bathe
Professor of Mechanical Engineering, MIT
Reorder No 672-2100
PREFACE
The analysis of complexstatic and dynamic problems in­
volves in essence three stages: selectionof a mathematical
model, analysis of the model, andinterpretation of the results.
During recent years the finite element method implementedon
the digital computer has been usedsuccessfullyin modeling
verycomplexproblems invarious areas of engineeringand
has significantlyincreasedthe possibilities for safeand cost­
effectivedesign. However, the efficient use of the method is
onlypossible if the basic assumptions of the procedures
employed areknown, andthe method canbe exercised
confidentlyon the computer.
The objectivein this course is tosummarize modern and
effective finite element procedures for the linear analyses of
static and dynamic problems. The material discussed in the
lectures includes the basic finite element formulations em­
ployed, the effectiveimplementationof these formulations in
computer programs, and recommendations onthe actual use
of the methods in engineering practice. The course is intended
for practicing engineers and scientists who want to solve prob­
lems using modemand efficient finite element methods.
Finite element procedures for the nonlinear analysis of
structures are presentedinthe follow-up course, Finite Element
Procedures for Solids and Structures - Nonlinear Analysis.
Inthis study guideshort descriptions of the lectures and
the viewgraphs used inthe lecture presentations are given.
Belowthe short description of eachlecture, reference is made
to the accompanying textbookfor the course: Finite Element
Procedures inEngineering Analysis, by K.J. Bathe, Prentice­
Hall, Inc., 1982.
The textbook sections and examples, listedbelowthe
short description of eachlecture, provide important reading
andstudy material to the course.
Contents
Lectures
l. Some basic concepts of engineering analysis 1-1
2. Analysis of continuous systems; differential and
variational formulations 2-1
3. Formulation of the displacement-based finite element method_ 3-1
4. Generalized coordinate finite element models 4-1
5. Implementation of methods in computer programs;
examples SAp, ADINA 5-1
6. Formulation and calculation of isoparametric models 6-1
7. Formulation of structural elements 7-1
8. Numerical integrations, modeling considerations 8-1
9. Solution of finite element equilibrium
equations in static analysis 9-1
10. Solution of finite element equilibrium
equations in dynamic analysis 10-1
1l. Mode superposition analysis; time history 11-1
12. Solution methods for calculations of
frequencies and mode shapes 12-1
SOME BASIC
CONCEPTS OF
ENGINEERING
ANALYSIS
LECTURE 1
46 MINUTES
I-I
SolIe basic ccnacepls of eugiDeeriDg ualysis
LECTURE 1 Introduction to the course. objective of lectures
Some basic concepts of engineering analysis.
discrete and continuous systems. problem
types: steady-state. propagation and eigen­
value problems
Analysis of discrete systems: example analysis of
a spring system
Basic solution requirements
Use and explanation of the modern direct stiff­
ness method
Variational formulation
TEXTBOOK: Sections: 3.1 and 3.2.1. 3.2.2. 3.2.3. 3.2.4
Examples: 3.1. 3.2. 3.3. 3.4. 3.5. 3.6. 3.7. 3.8. 3.9.
3.10. 3.11. 3.12. 3.13. 3.14
1-2
Some basic concepts 01 engineering aulysis
INTRODUCTION TO LINEAR
ANALYSIS OF SOLIDS AND STRUCTURES
• The finite element method is now
widely used for analysis of structural
engineering problems.
• 'n civil, aeronautical, mechanical,
ocean, mining, nuclear, biomechani­
cal,... engineering
• Since the first applications two
- we now see applications
in linear, nonlinear, static
and dynamic analysis.
- various computer programs
are available and in significant
use
My objective in this set of
lectures is:
• to introduce to you finite
element methods for the
linear analysis of solids
and structures.
["Iinear" meaning infinitesi­
mally small displacements and
linear elastic material proeer­
ties (Hooke's law applies)j
• to consider
- the formulation of the finite
element equilibrium equations
- the calculation of finite
element matrices
- methods for solution of the
governing equations
- computer implementations
.to discuss modern and effective
techniques, and their practical
usage.
1·3
Some basic concepts of engineering analysis
REMARKS
• Emphasis is given to physical
explanations rather than mathe­
matical derivations
• Techniques discussed are those
employed in the computer pro­
grams
SAP== Structural Analysis Program
Incremental Nonlinear Analysis
• These few lectures represent a very
brief and compact introduction to
the field of finite element analysis
• We shall follow quite closely
certain sections in the book
Finite Element Procedures
in Engineering Analysis,
Prentice-Hall, Inc.
(by K.J. Bathe).
Finite Element Solution Process
Physical problem
Establish finite element
... - - ~ model of physical
problem
1
I
: I,-__S_ol_v_e_th_e_m_o_d_el__
I ~
~ - - - iL-_I_n_te_r.;..p_re_t_t_h_e_re_s_u_lt_S_....J
Revise (refine)
the model?
1-4
SolIe basic concepts of engiDeering analysis
10 ft
15 ft
I 12 at 15°
,.
Analysis of cooling tower.

\\(no restraint assumed)
Altered' grit E= toE
c
.,
Analysis of dam.
1·5
Some basic concepts of engineering analysis
B
.

W
o
E ~ ~ ; ; C = - - - - - - - _ ........
F
Finite element mesh for tire
inflation analysis.
1·6
SolDe basic concepts of engineering analysis
Segment of a spherical cover of a
laser vacuum target chamber.
l,W
p
p
PINCHED CYLINDRICAL
SHELL

EtW-50
P-
100
-150 • 16x 16 MESH
-200 -
DISPLACEMENT DISTRIBUTION ALONG DC OF
PINCHED CYLINDRICAL SHELL
• 16x 16 MESH
-0.2
Mil
""= 0.1
C
BENDING MOMENT DISTRIBUTION ALONG DC OF
PINCHED CYLINDRICAL SHELL
1-7
SoBle basic concepts 01 engineering analysis
I
Finite element idealization of wind
tunnel for dynamic analysis
SOME BASIC CONCEPTS
OF ENGINEERING
ANALYSIS
The analysis of an engineering
system requires:
- idealization of system
- formulation of equili­
brium equations
- solution of equations
- interpretation of results
1·8
SYSTEMS
Some basic concepts of engineering analysis
DISCRETE
response is
described by
variables at a
finite number
of points
set of alge­
braic --
equations
CONTINUOUS
response is
described by
variables at
an infinite
number of
points
set of differ­
ential
equations
PROBLEM TYPES ARE
• STEADY -STATE (statics)
• PROPAGATION (dynamics)
• EIGENVALUE
For discrete and continuous
systems
Analysis of complex continu­
ous system requires solution of
differential equations using
numerical procedures
reduction of continuous
system to discrete form
powerful mechanism:
the finite element methods,
implemented on digital
computers
ANALYSIS OF DISCRETE
SYSTEMS
Steps involved:
- system idealization
into elements
- evaluation of element
equilibrium requirements
- element assemblage
- solution of response
1·9
Some basic concepts of engineering analysis
Example:
steady - state analysis of
system of rigid carts
interconnected by springs
Physical layout
ELEMENTS
U1
U
3
I
~
: ~ \ l ) ..
F(4)
..
3
1
F(4)
k, u1
- F(' ) 1
-1]["1] . [F1
4'
]
- ,
'4 [1
u2 -1
1 U F(4)
3 3
-
F(2 )
F(2)
--- 2
,
'2 [ 1
-1]["I]fF}]
1 u F(
2
)
-
-1
2 2
F(S)
F(S)
2
3
u, u
2
-t] [F(5l]
'5 [1
k
3 F(3)
1 u
2
= F1S )
-1
---
-- 2
3 3
F(3)
,
-r
1
]f
P
l]
'3 [ ]
-1
1 u F(3)
2 2
1·10
SolIe basic cOIcepls of engineering analysis
Element interconnection
requirements :
F(4) +F(S) = R
3 3 3
These equations can be
written in the form
K U =E.
Equilibrium equations
K U = R (a)
+k
4
k
1
+ k
2
+ k
3
~ -k
2
- k
3
U
T
= [u
- 1
R
T
= [R
- 1
·
·
·
·
·
: -k
4
... .'" ............................ ...
· .
· .
· .
K = -k
2
- k
3
~ k
2
+ k
3
+k
S
~ -k
S
· .
... ...............•................•.....•..... ...
· .
· .
1·11
Some basic concepts of engineering analysis
and we note that
~ = t ~ ( i )
i =1
where
::]
o 0
etc ...
This assemblage process is
called the direct stiffness
method
The steady- state analysis is
completed by solving the
equations in (a)
1·12
Some basic concepls 01 engineering analysis
u,
·
.................: :.............. .
· .
· .
... .
· .
· .
· .
· .
· .
· .
· .
u,
K
1
.
... : :.............. .
· .
K=
· .
... .
· .
· .
· :
U1
............................... : .
K ••• ••••••••• ••••••••••• :...............
1·13
SOlDe basic concepts of engineering analysis
· .
... : : .
· .
· .
u,
· . • 'O ••••••••••••••••••••••••••••••••••••••••••••••
· .
· .
· .
K =
u,
+ K
4
;
K
1
+K
2
+ K
3
;-K
2
-K
3
· .
'O'O'O 'O'O'O'O'O'O'O'O'O'O'O'O'O'O:'O'O'O'O'O'O'O'O'O'O'O'O'O'O:'O'O'O'O'O'O'O'O'O'O'O'O'O'O •
· .
· .
K=
· . 'O'O ••••••••••••••••••••••••••••••••••••••••••••••
· .
· .
· .
.
+ K
4
;
K
1
+K
2
+ K
3
~ - K 2 -K
3
-K
4
'O'O •••••••••••••••••••••••••••••••••••••••••••••
K = -K
2
-K
3
~ K 2 + K
3
+ K
5
-K
5
'O'O ••••••••••••••• : •••••••••••••••••••••••••••••
u,
I
K
1·14
Some basic concepts of engineering analysis
In this example we used
the direct approach; alternatively
we could have used a variational
approach.
In the variational approach we
operate on an extremum
formu lation:
u = strain energy of system
W = total potential of the
Equilibrium equations are obtained
from
an - 0 (b)
~ -
1
In the above analysis we have
U = ~ U T ! ! !
W = U
T
R
Invoking (b) we obtain
K U = R
Note: to obtain U and W we
again add the contributions from
all elements
1·15
SOlDe basic concepts of engineering analysis
PROPAGATION PROBLEMS
main characteristic: the response
changes with time ~ need to
include the d'Alembert forces:
For the example:
m,
a a
M = a m
2
a
a a m
3
EIGENVALUE PROBLEMS
we are concerned with the
general ized eigenvalue problem
(EVP)
Av = AB v
!l , .!l are symmetric matrices
of order n
v is a vector of order n
A is a scalar
EVPs arise in dynamic and
buckling analysis
1·16
Some basic concepts of engineering analysis
Example: system of rigid carts

Let
U = <p sin W(t-T)
Then we obtain
_w
2
sin W(t-T)
+ K <p sin W(t-T)= 0
- - -
Hence we obtain the equation
There are 3 solutions
w,
(l)2 ' eigenpairs
w
3
'
In general we have n solutions
1·17
ANALYSIS OF
CONTINUOUS SYSTEMS;
DIFFEBENTIAL AND
VABIATIONAL
FOBMULATIONS
LECTURE 2
59 MINUTES
2-1
Analysis 01 continnous systems; differential and variational lonnDlations
LECTURE 2 Basic concepts in the analysis of continuous
systems
Differential and variational formulations
Essential and natural boundary conditions
Definition of em-I variational problem
Principle of virtual displacements
Relation between stationarity of total potential, the
principle of virtual displacements, and the differ­
ential formulation
Weighted residual methods, Galerkin, least
squares methods
Ritz analysis method
Properties of the weighted residual and Ritz
methods
Example analysis of a nonuniform bar, solution
accuracy, introduction to the finite element
method
TEXTBOOK: Sections: 3.3.1, 3.3.2, 3.3.3
Examples: 3.15, 3.16, 3.17, 3.18, 3.19, 3.20, 3.21,
3.22, 3.23, 3.24, 3.25
2·2
Analysis of continuous systems; differential and variational formulations
BASIC CONCEPTS
OF FINITE
ELEMENT ANALYSIS ­
CONTINUOUS SYSTEMS
• We discussed some
basic concepts of
analysis of discrete
systems
basic concepts are
used in analysis of
continuous systems
CONTINUOUS SYSTEMS
differential
formulation
t
Weighted residual
methods
Galerkin _.. _ - - - - - 4 1 ~ _
least squares
variational
formulation
Ritz Method
....
finite element method
2·3
Analysis of continuous systeDlS; differential and ,arialionalIOl'llulali.
Example - Differential formulation
aAI + A I dx - aAIx
x oX X
/
Young's modulus, E
) mass density,
cross-sectional area, A
R..
The problem governing differential
equation is
Derivation of differential equation
The element force equilibrium require­
ment of a typical differential element
is using d'Alembert's principle

dx

Area A, mass density p
2
= p A a u

The constitutive relation is
au
a = E ­
ax
Combining the two equations above
we obtain
2·4
baIysis 01 COitiDlOU systems; differatial aDd variationaliOl'lDDlatiODS
The boundary conditions are
u(O,t} =°
EA ~ ~ (L,t) = R
O
with initial conditions
u(x,O} = °
~ (x O) =°
at '
9 essential (displ.) B.C.
9 natural (force) B.C.
In general, we have
highest order of (spatial) deriva­
tives in problem-governing dif­
ferential equation is 2m.
highest order of (spatial) deriva­
tives in essential b.c. is (m-1)
highest order of spatial deriva­
tives in natural b.c. is (2m-1)
Definition:
We call this problem a C
m
-
1
variational problem.
2·5
Analysis 01 continuous systems; differential and variatioD,a1 fOl'llolatiODS
Example - Variational formulation
We have in general
II=U-W
For the rod
fL
II = J }EA
o
and
i
L
au 2 B
(--) dx - u f dx - u R
ax L
o
u = 0
o
and we have 0 II = 0
The stationary condition 6II = 0 gives
r
L
au au rL.B
JO(EA ax)(6 ax) dx -)0 6u t- dx
- 6u
L
R = 0
This is the principle of virtual
displacements governing the
problem. In general, we write
this principle as
or
2·6
lIiIysis of ..IiDIGUS systems; differential and variatiooallormulatioDS
However, we can now derive the
differential equation of equilibrium
and the b.c. at x =l .
Writing a8u for 8au , re-
ax ax
calling that EA is constant and
using integration by parts yields
dx + [EA ~ I
ax x=L
- EA ~ \
dX
x=o
Since QUO is zero but QU is
arbitrary at all other points, we
must have
and
au I
EAax- x=L=R
B a
2
u
Also f = -A p - and
, at
2
hence we have
2·7
Analysis of cODtiDaoas syst_ diIIereatial and variatioul fOlllalatiODS
The important point is that invoking
o IT = 0 and using the essential
b.c. only we generate
• the principle of virtual
displacements
• the problem-governing differ­
ential aquatio!)
• the natural b.c. (these are in
essence "contained in" IT ,
i.e., inW).
In the derivation of the problem­
governing differential equation we
used integration by parts
• the highest spatial derivative
in IT is of order m .
• We use integration by parts
m-times.
Total Potential IT
I
Use oIT = 0 and essential "b.c.
~
2·8
Principle of Virtual
Displacements
I
Integration by parts
~
Differential Equation
of Equilibrium
and natural b.c.
_ solve
problem
_solve
problem
balysis of aDa. syst-: diBerential and variatiouallnaiatiOlS
Weighted Residual Methods
Consider the steady-state problem
(3.6)
with the B.C.
B.[</>] = q., i =1,2, •••
1 1
at boundary (3.7)
The basic step in the weighted
residual (and the Ritz analysis)
is to assume a solution of the
form
(3.10)
where the f
i
are linearly indepen­
dent trial functions and the ai
are multipliers that are deter­
mined in the analysis.
Using the weighted residual methods,
we choose the functions f
i
in (3.10)
so as to satisfy all boundary conditions
in (3.7) and we then calculate the
residual,
n
R = r - L2mCL a· f.] (3.11 )
1=1 1 1
The various weighted residual methods
differ in the criterion that they employ
to calculate the ai such that R is small.
In all techniques we determine the ai
so as to make a weighted average of
R vanish.
2·9
Analysis 01 C.tinnoDS systems; differential and variational 10000nlations
Galerkin method
In this technique the parameters ai are
determined from the n equations
f f. R dD=O ;=1,2, ••• ,n
D 1
Least squares method
(3.12)
In this technique the integral of the
square of the residual is minimized with
respect to the parameters ai '
a
aa.
1
;=1,2, ••• ,n
[The methods can be extended to
operate also on the natural boundary
conditions, if these are not satisfied
by the trial functions.]
RITZ ANALYSIS METHOD
Let n be the functional of the
e
m
-
1
variational problem that is
equivalent to the differential
formulation given in (3.6) and (3.7).
In the Ritz method we substitute the
trial functions <p given in (3.10)
into n and generate n simul­
taneous equations for the para­
meters ai using the stationary
condition on n ,
2·10
an 0
aa. =
1
;=1,2, ••• ,n (3.14)
Analysis of continuous systems; differential and variational formulations
Properties
• The trial functions used in the
Ritz analysis need only satisfy the
essential b.c.
• Since the application of oIl = 0
generates the principle of virtual
displacements, we in effect use
this principle in the Ritz analysis.
• By invoking 0 II = 0 we minimize
the violation of the internal equilibrium
requirements and the violation of
the natural b.c.
• A symmetric coefficient matrix
is generated, of form
K U = R
Example
R=100 N
2
Area = 1 em
(
........_- x,u ---- - - - --- ~ - - - r ; ; ; - = = - e -
.F- --.;;B;",.,. C
I-... - - ~ ~ - - - · - I - .. --------·-I
100 em 80 em
Fig. 3.19. Bar subjected to
concentrated end force.
2·11
Analysis of COitiDlOIS systems; differeatial ad ,ariali" fOllDaialiODS
Here we have
1
180
IT = 1 E A ( ~ ) 2 dx
2 ax
o
- 100 uIx = 180
and the essential boundary condition
is uIx=O = 0
Let us assume the displacements
Case 1
u =a
1
x + a
2
i
Case 2
~
u =I1JO
0< x < 100
100 < x < 180
We note that invoking oIT = 0
we obtain
1
180
oIT = (EA ~ ~ ) o ( ~ ~ ) dx - 100 OU Ix=180
o = 0
or the principle of virtual
displacements
£
180
( ~ ~ u ) ( EA ~ ~ ) dx = 100 OU Ix=180
o
J
ET T dV= IT. F.
- - 1 1
V
2·12
Analysis of continuous systems; differential and variational formulations
Exact Solution
Using integration by parts we
obtain
~ (EA ~ ) = 0
ax ax
EA ~ = 100
ax
x=180
The solution is
u = 1~ O x ; 0 < x < 100
100 < x < 180
The stresses in the bar are
a = 100; 0 < x < 100
a =
100 ; 100 < x < 180
(l+x-l00)2
40
2·13
Analysis of continuous systems; differential and variational formulations
Performing now the Ritz analysis:
Case 1
f
180
dx+ I (1+
x
-l00)2
2 40
100
Invoking that orr = 0 we obtain
E [0.4467
116
and
116
34076
128.6
a1 = ---=E=---
a
- 0.341
2 - - E
Hence, we have the approximate
solution
u =
12C.6 0.341
E x - E
2
x
2·14
a = 128.6 - 0.682 x
Analysis of continuous systems; differential and variational formulations
Case 2
Here we have
100
E J 1 2
n=2 (100
u
B)
a
f
180
dx+ I (1+
x
-l00)2
2 40
100
Invoking again
on =0 we obtain
E
[15.4
-13]
[ ~ : ] = [ ~ o o ]
240
-13 13
Hence, we now have
10000 11846.2
U =
E
U
c E B
and
o = 100 0< x < 100
1846.2
= 23.08 x> 100 o =
80
2-15
Aulysis of COilinDmas systems; diUerenliai and varialiOlla1I01'1BDlaIiGlS
u
EXACT

-- --- -::.:--

" Sol ution 2
..,-__--r-__--.,r---
15000
E
10000
E
5000
-E-
100 180
CALCULATED DISPLACEMENTS
(J
50
100-I=:::==-==_==_:=os:=_=_=,==_=_==
"" EXACT

SOLUTION 1
I -< ,JSOLUTION 2
L._._.
-+
100 180
CALCULATED STRESSES
2·18
balysis of coatiDloas systms; diBerenlial ud variational fonnllatioas
We note that in this last analysis
e we used trial functions that do
not satisfy the natural b.c.
e the trial functions themselves
are continuous, but the deriva­
tives are discontinuous at point
B. 1
for a e
m
- variational problem
we only need continuity in the
(m-1)st derivatives of the func­
tions; in this problem m =1 .
edomains A- Band B- e are
finite elements and
WE PERFORMED A
FINITE ELEMENT
ANALYSIS.
2·17
FORMULATION OF THE
DISPLACEMENT-BASED
FINITE ELEMENT
METHOD
LECTURE 3
58 MINUTES
3·1
Formulation of the displacement-based finite element method
LECTURE 3 General effective formulation of the displace­
ment-based finite element method
Principle of virtual displacements
Discussion of various interpolation and element
matrices
Physical explanation of derivations and equa­
tions
Direct stiffness method
Static and dynamic conditions
Imposition of boundary conditions
Example analysis of a nonuniform bar. detailed
discussion of element matrices
TEXTBOOK: Sections: 4.1. 4.2.1. 4.2.2
Examples: 4.1. 4.2. 4.3. 4.4
3·2
Formulation of the displacement-based finite element method
FORMULATION OF
THE DISPLACEMENT ­
BASED FINITE
ELEMENT METHOD
- A very general
formu lation
-Provides the basis of
almost all finite ele­
ment analyses per­
formed in practice
-The formulation is
really a modern appli ­
cation of the Ritz/
Gelerkin procedures
discussed in lecture 2
-Consider static and
dynamic conditions, but
linear analysis
Fig. 4.2. General three-dimensional body.
3·3
FOl'Dlulation of the displaceDlent·based finite e1mnent lDethod
The external forces are
fB
f ~
F
i
X X
fB = fB fS = fS
F
i
= F
i
(4.1)
y y y
fB fS
F
i
Z Z Z
The displacements of the body from
the unloaded configuration are
denoted by U, where
u
T
=[u V w]
The strains corresponding to U are,
(4.2)
~ T = [E
XX
E
yy
E
ZZ
YXy YyZ YZX] (4.3)
and the stresses corresponding to €
are
3·4
Formulation of the displacement-based finite element method
Principle of virtual displacements
where
ITT = [IT If w] (4.6 )
Fig. 4.2. General three-dimensional body.
3·5
Formulation of the displaceaenl-based filile e1eDlenl .ethod
x,u
,
,
"
"
Finite element
For element (m) we use:
!!(m) (x, y, z) =!:!.(m) (x, y, z) 0 (4.8)
"T
!! =[U, V, W, U
2
V
2
W
2
••• UNVNW
N
]
"T
!! =[U,U
2
U
3
... Un] (4.9)
§.(m) (x, y, z) =~ ( m ) (x, y, z) !! (4.'0)
!.(m) =f ( m ) ~ ( m ) + -rI(m) (4.'1)
3·&
'OI'IIalation of the displaceDlenl-based filile eleDlenl method
Rewrite (4.5) as a sum of integrations
over the elements
(4.12)
Substitute into (4.12) for the element
displacements, strains, and stresses,
using (4.8), to (4.10),
____ -(m) T
j- I --£
'iTl 1 B(m) Tc(m)B(m)dv(m)j U=
If v(m) - l- - £1 ---- = f.(m)
j
[I
T (m) j (-£)(m) = B(m) l..u·)
T L l(m) !!(m) 1.B dV(m) - --
I
1 m V I ( )T
_" ,.
:,. L f.
m
) !!sCm)Ti
m
)dScmlj y:(m) =!!(m)
(m) T
El B(m)TTI(m) dv(m)j -US
m;rm) - - -(m) T
-___.r__.........1 ------....
"<I:::
(4.13)
3·7
Formulation of the displacement-based finite element .ethod
We obtain
K U = R
where
(4.14)
R=.Ba + Rs - R
1
+ (4. 15)
K= B(m)Tc(m)B(m)dV(m)
- m
J
V(m)- - - - (4.16)
R = "'1. H(m)TfB(m)dV(m) (4.17)
lm) - -
R ="'1 H
S
(m)Tfs(m)dS(m) (4.18)
-S - -
R ="'1 B(m)TT1(m)dV(m) (4.19)
-1 V(m) - -
R =F
-
In dynamic analysis we have
f
(m)T -B(m)
= V(m) .!:!. [1.
_
MD+KU= R
(4.21 )
(4.22)
(4.20)
B(m) -B(m) •• (m)
1. = 1. - p!!
3·8
Formulation of the displacement-based finite element method
To impose the boundary conditions,
we use

+
t!t>b

= (4.38)
.. ..

(4.39)

(4.40)
ransformed
egrees of
\eedom
i
A
!

V
T
I d
!
-
f
Global degrees
of freedom
;-
V

I
(restrained\

rl'f
u
L
[
COS a
T =
sin a
-sin a]
cos a
'/.. U = T IT
Fig. 4.10. Transformation to skew
boundary conditions
3·9
Formulation of the displacement-based finite element method
For the transformation on the
total degrees of freedom we use
so that
..
Mu+Ku=R
where
.th .th
1
J
column
!
1. •• j
(4.41 ) i
th
row 1
cos a. -s ina.
T =
}h
1
(4.42)
sin a. cos a.
1
L
Fig. 4.11. Skew boundary condition
imposed using spring element.
We can now also use this procedure
(penalty method)
Say U
i
=b, then the constraint
equation is
___ 3·10
k U. = k b
,
where
k » k ..
"
(4.44)
FormDlation of the displacement·based finite element method
Example analysis
80
x
z
y
100
Finite elements
area = 1
element ®
100
area = 9
J ~
I"
100
-I
80
~ I
3·11
Formulation of the displacement·based finite element method
Element
interpolation functions
1.0
I ...
L
--I
Displacement and strain
interpolation matrices:
H(l} = [(l-L)
y
a ]
- 100 100
v(m} = H(m}U
!:!.(2} = [
a
(1- L)
:0]
80
!!(l)=[
1 1
a ]
100 100
av = B(m}U
!!(2) = [
1 1
ay - -
a
80
80]
3·12
FOI'IDDlation of the displacement·based finite element method
stiffness matrix
- 1
100
5.= (1 HEllO l ~ O [ - l ~ O l ~ O o}Y
a
a
U
1
- 80
1
80
Hence
E [ 2.4 -2.4
=240 -2.4 15.4
a -13
Similarly for M '.!!B ' and so on.
Boundary conditions must still be
imposed.
3·13
GENERALIZED
COORDINATE FINITE
ELEMENT MODELS
LECTURE 4
57 MINUTES
4·1
Generalized coordinate finite element models
LECTURE 4 Classification of problems: truss, plane stress, plane
strain, axisymmetric, beam, plate and shell con­
ditions: corresponding displacement, strain, and
stress variables
Derivation of generalized coordinate models
One-, two-, three- dimensional elements, plate
and shell elements
Example analysis of a cantilever plate, detailed
derivation of element matrices
Example results
Summary of the finite element solution process
Solution errors
Convergence requirements, physical explana­
tions, the patch test
TEXTBOOK: Sections: 4.2.3, 4.2.4, 4.2.5, 4.2.6
Examples: 4.5, 4.6, 4.7, 4.8, 4.11, 4.12, 4.13, 4.14,
4.15, 4.16, 4.17, 4.18
4-2
Generalized coordinate finite eleDlent models
DERIVATION OF SPECIFIC
FINITE ELEMENTS
• Generalized coordinate
finite element models
~ ( m ) = i B(m)T C(m) B(m) dV (m)
V(m)
aW) = JH(m)T LB(m) dV (m)
V(m)
R(m) = f HS(m)T f S(m) dS (m)
!!S (m) - -
S
etc.
In essence, we need
H(m) B(m) C (m)
- ,- '-
• Convergence of
analysis results
A
Across section A-A:
T
XX
is uniform.
All other stress components
are zero.
Fig. 4.14. Various stress and strain
conditions with illustrative examples.
(a) Uniaxial stress condition: frame
4·3
Ge.raJized coordiDale finite elementlDOIIeIs
Hale
\
I
6
I
\
\
-\
-
1ZI
\
T
XX
' T
yy
, T
XY
are uniform
across the thickness.
All other stress components
are zero.
Fig. 4.14. (b) Plane stress conditions:
membrane and beam under in-plane
actions.
u(x,y), v(x,y)
are non-zero
w= 0 , E
zz
= 0
Fig. 4.14. (e) Plane strain condition:
long dam subjected to water pressure.
4·4
Generalized coordinate finite element models
are axisymmetric.
j(
I
I
I
I
,
II
\--
All other stress components
are non-zero.
Fig. 4.14. (d) Axisymmetric condition:
cylinder under internal pressure.
(before deformation)
(after deformation)
/
SHELL
Fig. 4.14. (e) Plate and shell structures.
4·5
Generalized coordinate finite element models
Problem
Bar
Beam
Plane stress
Plane strain
Axisymmetric
Three-dimensional
Plate Bending
Displacement
Components
u
w
u, v
u, v
u,v
u,v, w
w
Table 4.2 (a) Corresponding Kine­
matic and Static Variables in Various
Problems.
Problem
Bar
Beam
Plane stress
Plane strain
Axisymmetric
Three-dimensional
Plate Bending
Strain Vector ~ T
-
(E"...,)
[IC...,]
(E"..., El'l' )' "7)
(E..., EJ"7 )'..7)
[E..., E"77 )'''7 Eu )
[E..., E"77 Eu )'''7 )'76 )'...,)
(IC..., 1(77 1("7)
. au au au au
Nolallon: E.. = ax' £7 = a/ )'''7 = ay +ax'
a
1
w a
1
w a
1
w
••• , IC..., = -dx
Z
' IC77 = - OyZ,IC.., = 2
0x
oy
Table 4.2 (b) Corresponding Kine­
matic and Static Variables in Various
Problems.
4·&
Problem
Bar
Beam
Plane stress
Plane strain
Axisymmetric
Three-dimensional
Plate Bending
Generalized coordinate finite element models
Stress Vector 1:
T
[T;u,]
[M
n
]
[Tn TJIJI T"'JI]
[Tn TJIJI T"'JI]
[Tn TJIJI T"'JI Tn]
[Tn TYJI Tn T"'JI TJI' Tu ]
[Mn MJIJI M"'JI]
Table 4.2 (e) Corresponding Kine­
matic and Static Variables in Various
Problems.
Problem Material Matrix.£
Bar
Beam
Plane Stress
E
El
[
1 v
E v 1
1-1':&
o 0
1 ~ . ]
Table 4.3 Generalized Stress-Strain
Matrices for Isotropic Materials
and the Problems in Table 4.2.
4·7
Generalized coordinate finite element models
ELEMENT DISPLACEMENT EXPANSIONS:
For one-dimensional bar elements
For two-dimensional elements
(4.47)
For plate bending elements
2
w(x,y) =Y, + Y2
x
+ Y3Y+ Y4xy + Y5x + •..
(4.48)
For three-dimensional solid elements
u(x,y,z) =a, + Ozx + ~ Y + Ci
4
Z + ~ x y + ...
w(x,y,z) =Y, +y
2
x+y
3
y+y
4
z+y
5
xy+ ...
(4.49)
4·8
Hence, in general
u = ~ ex
Generalized coordinate finite element models
(4.50)
(4.51/52)
(4.53/54)
Example
(4.55)
Y.V
X.V
la) Cantilever plate
r
Nodal point 6
lp
9
Element 0
0
5
8
CD
@
Y.V
V
7
1 4 7
X.V V
7
(bl Finite element idealization
Fig. 4.5. Finite element plane
stress analysis; i.e. T
ZZ
=T
Zy
=T
ZX
=0
4·9
Generalized coordinate finite element models
2
LJ2.= US --II--.......---------....~
element ®
Element nodal point no. 4
=structure nodal point
no. 5 .
Fig. 4.6. Typical two-dimensional
four-node element defined in local
coordinate system.
For element 2 we have
[
U{X,y)] (2)
=H(2) u
v{x,y) --
where
u
T
= [U
- 1
4·10
Generalized coordinate linite element models
To establish H (2) we use:
or
[
U(X,y)] = _ ~ l ! .
v(x,y)
where
! =[ ~ ~ } ! = [1 x y xy]
and
Defining
we have
Q= Aa.
Hence
H=iPA-
1
4·11
Generalized coordinate finite element models
Hence
H =fl
- l4
and
(1+x ) ( Hy) : : a
I ••• I I
a : : (1 +x )( 1+y) :
H'ZJ = [0
- 0
Ull
:HII
: HZI
U J VJ U z t': u. v.
U
2
U
3
U
4
Us U
6
U
7
Us U
9
U
1a
I 0 : HIJ H 17 : HI. H 16 : 0 0: HI. H
u
:
o :H ZJ H 21 : H:: H: 6 : 0 0: H.. H
a
:
VI -element degrees of freedom
U12 U13 U14 UIS-assemblage degrees
HIs: 0 0 zeros OJ offreedom
H
zs
: 0 0 zeros O
2x18
(a) Element layout
(b) Local-global degrees of freedom
element (m)
4·12
Generalized coordinate finite element models
In plane-stress conditions the
element strains are
where
E - au . E _ av. _ au + av
xx - ax' yy - ay , Yxy - ay ax
Hence
where
I = [ ~
1 0
I
y'O
I
0 0 0
1
0
I
0 1
I
X 10
I
4·13
Generalized coordinate finite element models
ACTUAL PHYSICAL PROBLEM
GEOMETRIC DOMAIN
MATERIAL
BOUNDARY CONDITIONS
1
MECHANICAL IDEALIZATION
KINEMATICS, e.g. truss
plane stress
three-dimensional
Kirchhoff plate
etc.
MATERIAL, e.g. isotropic linear
elastic
Mooney-Rivlin rubber
etc.
centrifugal
etc.
BOUNDARY CONDITIONS, e.g. prescribed
1
displacements
etc.
FINITE ELEMENT SOLUTION
CHOICE OF ELEMENTS AND
SOLUTION PROCEDURES
YIELDS:
GOVERNING DIFFERENTIAL
EQUATIONS OF MOTION
e.g.
..!.. (EA .!!!) = - p(x)
ax ax
YIELDS:
APPROXIMATE RESPONSE
SOLUTION OF MECHANICAL
IDEALIZATION
Fig. 4.23. Finite Element Solution
Process
4·14
Generalized coordinate finite element models
SECTION
ERROR ERROR OCCURRENCE IN discussing
error
DISCRETIZATION use of finite element 4.2.5
interpolations
NUMERICAL evaluation of finite 5.8. 1
INTEGRATION element matrices using 6.5.3
IN SPACE numerical integration
EVALUATION OF use of nonlinear material 6.4.2
CONSTITUTIVE models
RELATIONS
SOLUTION OF direct time integration, 9.2
DYNAMIC EQUILI-. mode superposition 9.4
BRIUM EQUATIONS
SOLUTION OF Gauss-Seidel, Newton- 8.4
FINITE ELEr1ENT Raphson, Quasi-Newton 8.6
EQUATIONS BY methods, eigenso1utions 9.5
ITERATION 10.4
ROUND-OFF setting-up equations and 8.5
their solution
Table 4.4 Finite Element
Solution Errors
4·15
Generalized coordinate finite element models
CONVERGENCE
Assume a compatible
element layout is used,
then we have monotonic
convergence to the
solution of the problem­
governing differential
equation, provided the
elements contain:
1) all required rigid
body modes
2) all required constant
strain states
~ compatible
LW layout
CD
incompatible
layout
~
t:=
no. of elements
If an incompatible element
layout is used, then in addition
every patch of elements must
be able to represent the constant
strain states. Then we have
convergence but non-monotonic
convergence.
4·16
Geuralized coordinate finite e1eJDeDt models
7 "
/ "
'r-
>
,;
/
(
"
"
1-- - --
,
I
I
I
I
I
I
I
i
I
(a) Rigid body modes of a plane
stress element
......~ _ Q
I
I
I
I
(b) Analysis to illustrate the rigid
body mode condition
Rigid body
translation
and rotation;
element must
be stress­
free.
Fig. 4.24. Use of plane stress element
in analysis of cantilever
4·17
Generalized coordinate filite elellent .adels
-------
Rigid body mode A
2
= 0
Poisson's
ratio" 0.30
Young's r------,
modulus = 1.0 I
I
10
-l
I
I
I
I
I
_1
Rigid body mode Al = 0
I
t
I
I
I
01
I
I
I

I
1.
. . . - - , . " . - ~ \
-- \
\
\ \
\ \
\ \
\ \
\ .J
\ -­
--
('
\
---
\
\
\,.
.....
.....
_-I
-- I
I
I
I
I
I
f
Rigid body mode A
3
=0
..... I
'J
Flexural mode A
4
=0.57692
Fig. 4.25 (a) Eigenvectors and
eigenvalues of four-node plane
stress element
~ -
\
...... "
\
\ \
\
\ . . . ~
-, \
.... \
.... ~
\
\
\
\
\
\
.----"'"="""- \
- - ~
I
'-
Flexural mode As =0.57692 Shear mode A. =0.76923
r--------1
I I
I I
I I
I :
I I
I I
I I
L .J
,-----,
I
I
I
I
I
I
I
I I
I I
I I
I I
I I
Stretching mode A
7
=0.76923 Uniform extension mode As =1.92308
Fig. 4.25 (b) Eigenvectors and
eigenvalues of four-node plane
stress element
4·18
(0
®
G)
® ®-.
®
/
@)
@
@
Generalized coordiDate finite element lDodels
·11
~
17
'c.
IT
,I>
~
.f:
20
IS
a) compatible element mesh; 2
constant stress a = 1000 N/cm
in each element. YY
b) incompatible element mesh;
node 17 belongs to element 4,
nodes 19 and 20 belong to
element 5, and node 18 belongs
to element 6.
Fig. 4.30 (a) Effect of displacement
incompatibility in stress prediction
0yy stress predicted by the
incompatible element mesh:
Point
Oyy(N/m
2
)
A 1066
B 716
C 359
D 1303
E 1303
Fig. 4.30 (b) Effect of displacement
incompatibility in stress prediction
4·19
IMPLEMENTATION or
METHODS IN
COMPUTER PROGRAMS;
LECTURE 5
56 MINUTES
5·1
"pi_entation of metllods in computer prograDlS; examples SIP, ADlRA
LECTURE 5 Implementation of the finite element method
The computer programs SAP and ADINA
Details of allocation of nodal point degrees of
freedom. calculation of matrices. the assem­
blage process
Example analysis of a cantilever plate
Out-of-core solution
Eff&ctive nodal-point numbering
Flow chart of total solution process
Introduction to different effective finite elements
used in one. two. three-dimensional. beam.
plate and shell analyses
TEXTBOOK: Appendix A, Sections: 1.3. 8.2.3
Examples: A.I. A.2. A.3. A.4. Example Program
STAP
5·2
l_pIg_talioa of _ethods in CODIpDter program; mDlples SAP, ADINA
N = no. of d.o.f.
of total structure
T
K(m) = 1. B(m) C(m)B(m) dV(m)
- V(m)- --
R(m) = 1. H(m)T fB(m) dV (m)
-B v(m) - -
H(m) B(m)
- -
kxN .hN
IMPLEMENTATION OF
THE FINITE ELEMENT
METHOD
We derived the equi­
librium equations
where
In practice, we calculate compacted
element matrices.
K = ~ K(m) ; R = ~ R ( m)
- m- -B m!..!B
~ , ~ B '
nxn nxl
n = no. of
element d.o.f.
tl ~
kxn R,xn
The stress analysis process can be
understood to consist of essentially
three phases:
1. Calculation of structure matrices
K , M, C , and R, whichever are
applicable.
2. Solution of equilibrium equations.
3. Evaluation of element stresses.
5·3
IDlpl_81taliol of Dlethods in toDlpuler progrw; mDlples SAP, ADIlI
The calculation of the structure
matrices is performed as follows:
1. The nodal point and element in­
formation are read and/or generated.
2. The element stiffness matrices,
mass and damping matrices, and
equivalent nodal loads are calculated.
3. The structure matrices K, M,
C , and R, whichever are
applicable, are assembled.
l Sz :: 6
t W:: 3
Z
x
/U:: 1
/Sx:: 4
r-----
y V::2 Sy:: 5
Fig. A.1. Possible degrees of
freedom at a nodal point.
I
- nodal point
_.
-
...
-
....i
ID(I,J) =
Degree of
freedom
5·4
....taIiOi of IIeIWs in coapler P.... UUlples SIP, ABilA
Temperature at top face a l00"C
aOem
<D ®
E = l(Jl1 N/cm
2
2 E = 2 x l(Jl1 NIcm
2
t
., - 0.15 t"=0.20 8
4 _1 7 -7
~
\
Temperature at Degree of
bottom face = 70'C freedom
number
a t 6 5 9 t ~ l 1
@ 4 Element
E. 2 x l(Jl1 Nlem2 number
E· lOS N/em
2
t
4
II'" 0.20 t10
.,-0.15 3 8
2 5 __ -9
Node
Fig. A.2. Finite element cantilever
idealization.
1n this case the 10 array is
given by
1 1 1 0 0 0 0 0 0
1 1 1 0 0 0 0 0 0
10 =
1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1
]
1
5·5
0.0 40.0 80.0]
0.0 0.0 0.0]
70.0 85.0 100.0]
IJDpleDIeDtatiOD of methods in CODIpater.programs; examples SAP, ADIIA
and then
0 0 0 1 3 5 7 9 11
0 0 0 2 4 6 8 10 12
10=
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
Also
XT= [ 0.0 0.0 0. 0 60.0 60.0 60.0 120.0 120.0 120.0]
T
Y = [ 0.0 40.0 80.0 0.0 40.0 80.0
T
Z =[0.0 0.0 0.0 0.0 0.0 0.0
TT =[70.0 85.0 100.0 70.0 85.0 100.0
5·6
Implementation of methods in computer programs; examples SAP, ADINA
For the elements we have
Element 1: node numbers: 5,2,1,4;
material property set: 1
Element 2: node numbers: 6325·
I , , ,
material property set: 1
Element 3: node numbers: 8547·
, , , ,
material property set: 2
Element 4: node numbers: 9658-
, , , ,
material property set: 2
CORRESPONDING COLUMN AND ROW NUMBERS
For compacted
I
matrix 1 2 3 4 5 6 7 8
For !1
..,
4 0 0 0 0 1 2 oJ
LM
T
= [3 4 0 0 0 0 1 2]
5·7
Implementation of methods in computer programs; examples SAP, ADINA
Similarly, we can obtain the LM
arrays that correspond to the
elements 2,3, and 4. We have for
element 2,
L MT = [5 6 0 0 0 0 3 4]
for element 3,
L M
T
= [9 10 3 4 1 2 7 8]
and for element 4,
LM
T
= [11 12 5 6 3 4 9 10]
J
SkYline
. ~ 0 0 0
"
o 0 0 0
"
------m =3
o k
36
'0 0 6
'-
k 45 k
46
0"0
(a) Actual stiffness matrix
k
ss
k
S6
k
66
1
2
4
6
10
12
16
18
22
A(21) stores k
S8
Fig. A.3. Storage scheme used for a
typical stiffness matrix.
A(17)
A(16)
A(15t
A(14)
A(13)
A(12)
A(91
A(8)
A(7)
A(6) A(lll
A(lO)
Symmetric
(b) Array A storing elements
of K.
A(l) A(3)
A(2) A(S)
A(4)
,.
m
K
=3
·1
"
k
ll
k
12
0 k
14
k
n
k
23
0
k
33
k
34
K=
k
44
A=
5·8
_pl••taiioo of lDeIJaods in COIDpater prograJDS; eDIIIples SAP, ADINA
x=NONZERO ELEMENT
0= ZERO ELEMENT
.--, COLUMN HEIGHTS
I I I
X 0 0 0 10 0
1
0
o 0 0 0:0 0:0
xix x 010 0 x
XIX 0 010 0 0
XIX 0 0 X 0 0
X 0 X 10 0 0
I xxlxXIO
xix XiX
SYMMETRIC I X X lX
XIX
IX
ELEMENTS IN ORIGINAL STIFFNESS MATRIX
Fig. 10. Typical element pattern in
a stiffness matrix using block storage.
BLOCK 1
BLOCK 2 ~ - - - ~
I
X 0
X 0
XIX
XiX
XiX
I
X
~ _ B L O C K 4
ELEMENTS IN DECOMPOSED STIFFNESS MATRIX
Fig. 10. Typical element pattern in
a stiffness matrix using block storage.
5·9
IIIlpl••tation of methods in computer programs; examples SAP, ABilA
20
3 ~ 1 2 3 4 5 6 7 8 9 10 11 12 1
,
14 15 16 17 18 19
~ 21 22 23 24 25 26 27 28 29 30 31 32 33
32
2283033
(b) Good nodal point numbering,
mk + 1 = 16.
Fig. A.4. Bad and good nodal point
numbering for finite element
assemblage.
(a) Bad nodal point numbering,
mk + 1 = 46.
9 11 14 16 19 21 24 26 29 31 6 4
5
1
2 7 12 17 22 27
~ ,
, 3
8 10 13 15 18 20 23 5
5·10
"pI••tation of Ilethods in cOIlpuler program; exallples SAP, ADINA
START
READ NEXT DATA CASE
Read nodal point data
(coordinates, boundary
conditions) and establish
equation numbers in the
10 array.
Calculate and store load
vecton for all load cases.
Read. generate. and store
element data. Loop over all
element groups.
Read element group data, and
assemble global structure
stiffness matrix. Loop over
all element groups.
Calculate .b..Q..!:.T factorization
of global stiffness matrix(·)
nodal point displacements. ~ - - - 1
Read element group data and
calculate element stresses.
Loop over all element groups.
END
Fig. A.5. Flow chart of program
STAP. *See Section 8.2.2.
5-11
Implementation of methods in computer programs; examples SAP, ADINA
z
ONE - DIMENSIONAL
ELEMENT
I
-'-------4
! RING ELEMENT
x
;--.-------------... y
Fig. 12. Truss element
p. A.42.
5-12
z
2
3
Fig. 13. Two-dimensional plane
stress, plane strain and axisymmetric
elements.
p.. A.43.
y
y
Implementation of metbods in computer programs; examples SAP, ADINA
2
---5
x
~ - - - - - - Fig. 14. Three-dimensional solid -------.... ~
and thick shell element
p. A.44.
z
y
Fig. 15. Three-dimensional beam
element
p A.45.
5·13
Implementation of methods in computer programs; examples SAP, ADINA
3-16 NODES
TRANSITION
ELEMENT

-- --. __e_
- - - L ~ - - -
--

y
x
Fig. 16. Thin shell element
(variable-number-nodes)
p. A.46.
5·14
FOBMULATION AND
CALCULATION OF
ISOPABAMETBIC
MODELS
LECTURE 6
57 MINUTES
6·1
FOl'Dlolation and calculation of isoparmetric models
LECTURE 6 Formulation and calculation of isoparametric
continuum elements
Truss. plane-stress. plane-strain. axisymmetric
and three-dimensional elements
Variable-number-nodes elements. curved ele­
ments
Derivation of interpolations. displacement and
strain interpolation matrices. the Jacobian
transformation
Various examples: shifting of internal nodes to
achieve stress singularities for fracture me­
chanics analysis
TEXTBOOK: Sections: 5.1. 5.2. 5.3.1. 5.3.3. 5.5.1
Examples: 5.1. 5.2. 5.3. 5.4. 5.5. 5.6. 5.7. 5.8. 5.9.
5.10. 5.11, 5.12. 5.13. 5.14. 5.15. 5.16. 5.17
6·2
FOI'DlDlatiOl ud calculation of isopariUHbic models
FORMULATION AND
CALCULATION OF ISO­
PARAMETRIC FINITE
ELEMENTS
interpolation matrices
and element matrices
-We considered earlier
(lecture 4) generalized
coordinate finite
element models
-We now want to discuss
a more general approach
to deriving the required
isoparametric
elements
lsoparametric Elements
Basic Concept: (Continuum Elements)
Interpolate Geometry
N
x=L
i=l
h. x. ;
I I
N
y= L
i =1
h. y. ;
I I
N
z=L
i=l
h. z.
I I
Interpolate Displacements
N
u= 1:
i =1
h. u.
I I
N
v= L
i == 1
h . v .
I I
N
w= L
i =1
h.w.
I I
N=number of nodes
&·3
Formulation and calculation of isoparametric models
1/0 Element Truss
2/0 Elements Plane stress Continuum
Plane strain Elements
Axisymmetric Analysis
3/0 Elements Three-dimensional
Thick Shell
(a) Truss and cable elements
(b) Two-dimensional elements
Fig. 5.2. Some typical continuum
elements
6·4
FOI'Ilalation and calcalatioD 01 isoparametric models
(c) Three-dimensional elements
Fig. 5.2. Some typical continuum
elements
Consider special geometries first:
~ ~ = = - = l = = = = = = ~ I = - = = r = = = = = = r ==~ 1
Truss, 2 units long
6·5
F..utioa and calculation of isoparalDebic lDodels
S
Ill(
1 1
~ J
~
ll(
-
1
1
-
r
1 - D Element
2 Nodes:
2/D element, 2x2 units
Similarly 3/D element 2x2x2 units
(r-s-taxes)
-11.0
~ ~ -+- .. _ h
1
= %(1 +r)
2 -r 1
-r
Formulation ud calculation 01 isoparUletric lIodeis
1.0
-
-
-
-...
-
e_----......:::::...::::=-- -..:...::-:::;. h
2
= Y.z(1- r) - Y.z(1- r
2
)
2 3 1
2 - 0 Element
4 Nodes:
3
Similarly
h
2
=%(1- r)( 1 + 5)
h
3
=%(1- r)(1- 5)
h
4
=%(1 + r)(1-s)
/ - r - r - - - - + ~ ~ - r
h
1
=~ ( 1 + r)(1 + 5)
4
6-7
Formulation and calculation of isoparametric models
6·8
3
3
Construction of S node element
(2 dimensional)
first obtain h
S
:
..... 1
-+--+-I--I--I---I--------I--.._r
Then obtain h1 and h
2
:
_------..1L...4. !1.0
1
h
1
=%(1 + r)(1 + s)
-%h
S
Sim. h
2
= %(1- r)(1 + s)
-%h
S
4
Formulation and calculation of isoparametric models
r = +1
y
6
3
\
\
r =-1
\
\
---;q-----
\
\
r =0
s=o
---..
8 r
x
(a) Four to 9 variable-number-nodes
two-dimensional element
Fig. 5.5. Interpolation functions of
four to nine variable-number-nodes
two-dimensional element.
·1· . -
-;h
6
...
Include only if node i is defined
h, =
h
2
=
h
3
= (1 -s)
h. = +r) (l-s)
h
s
= -r
2
) (1 +s)
'h
6
= (1 - S2) (1 - r)
h
7
= -r
2
) (1 -s)
h
s
= i (1 - s2) (1 + r)

( 1- r"") (1- S'")
i = 5 i = 6 i = 7 i = 8 I:: r
her
h<j
-ih
q

-ihq
-1 h<j
-th"
-th<f
(b) Interpolation functions
Fig. 5.5. Interpolation functions of
four to nine variable-number-nodes
two-dimensional element:
6·9
Fonnulation and calculation of isoparametric models
Having obtained the hi we can
construct the matrices Hand !!:
- The elements of H are the· h·
- I
(or zero)
- The elements of B are the
derivatives of thehi (or zero),
Because for the 2x2x2 elements
~
we can use 1:;'= ~
x==r
y == s
z == t
EXAMPLE 4 node 2 dim. element
6·10
Formulation and calculation of isoparametric models
ah
1
0 ah
4
0
ar ar
u
1
[
E
r
l
ah
1
ah
4
v
1
0 0 E
SS as as u
2
Y
rs ah
1
ah
1
3h
4
ah
4
as at' as ar
v
4
\.. I
v-
B
We note again r==x
s=y
GENERAL ELEMENTS
Y,v
s r = +1
s =+1
r---t---4_
r

6·11
Formulation and calculation of isoparametric models
Displacement and geometry inter-
polation as before, but
[:] = [::
: ] l ~ ] Aside:
as as as ay
cannot use
or
a a ar
---
ax + ...
ax ar
a a
- = J
ax
(in general) ar -
a _
J-
1 a
(5.25)
a-x-
ar
Using (5.25) we can find the matrix
.!!. of general elements
The !:! and J! matrices are a
function of r, s , t ; for the
integration thus use
dv =det J dr ds dt
6·12
FOI'Dlalation ud calculation of isoparmebic .odels
Fig. 5.9. Some two-dimensional elements
Element 1
z.._----+----......-"·r
+-
3, '4-
"""1--1-----------t.. ~ 1
6 em.
X
Element 2
2.
...
J =
+--_1< '0' I I=>
1+ 3
.....-------.... .;
cDG-W\
o
1 1
213 2
&-13
Formulation and calculation of isoparametric models
Element 3
\c.1V\
(1 +5)]
(3+r)
2 c l"l1
2.
I •
- + - - - ~ ~ - -,c
: ' 3 , . ~
't' .....,..
.I.
...L...
3
-"'
1
------14-
-,
1c.W'I
3
r=-I
Natural space
3

I I
,-. -I'
L/4
Actual physical space
Fig. 5.23. Quarter-point one­
dimensional element.
6·14
Formulation and calculation of isoparametric models
Here we have
3
x=L:
i =1
hence
L 2
h. x. 9 x =-4(1+r )
1 1
J = [!:.. + !'- LJ
- 2 2
and
or
Since
r = 2.Jf- 1
We note
1 singularity at X = 0 !
/x
6·15
Formulation and calculation of isoparaDlebic Dlodels
Numerical Integration
Gauss Integration
Newton-Cotes Formulas
K = '"a··k F··
k
- !:J IJ -IJ
I,J,k
x
x
6·16
FORMULATION OF
STRUCTURAL
ELEMENTS
LECTURE 7
52 MINUTES
7·1
Formulation of structural elements
LECTURE 7 Formulation and calculation of isoparametric
structural elements
Beam, plate and shell elements
Formulation using Mindlin plate theory and uni­
fied geneJ,"al continuum formulation
Assumptions used including shear deformations
Demonstrative examples: two-dimensional beam,
plate elements
Discussion of general variable-number-nodes
elements
Transition elements between structural and con­
tinuum elements
Low- versus high-order elements
TEXTBOOK: Sections: 5.4.1, 5.4.2, 5.5.2, 5.6.1
Examples: 5.20, 5.21, 5.22, 5.23, 5.24, 5.25, 5.26, 5.27
7·2
FORMULATION OF
STRUCTURAL
ELEMENTS
• beam, plate and
shell elements
• isoparametric
approach for
interpolations
Continuum
Approach
FOI'IIDlati.... slnclDrai e1U11DIs
Strength of Materials
Approach
• straight beam
elements
use beam theory
including shear
effects
• plate elements
use plate theory
including shear
effects
(ReissnerIMindlin)
" particles remain on
a straight line during
deformation"
Use the general
principle of virtlial
displacements, but
-- exclude the stress
components not
applicable
-- use kinematic
constraints for
particles on
sections originallv
normal to the mid­
surface
e.g.
beam
e.g.
shell
7·3
Formulation of structural elements
..
x
Neutral
axis
Beam
section
Boundary conditions between
beam elements
Deformation of cross-section
wi = wi ;
-0 +0
x x
dw _ dw
dx -0 - dx +0
x x
a) Beam deformations excluding
shear effect
Fig. 5.29. Beam deformation
mechanisms
Neutral
axis
Beam
section
Deformation of cross-section
WI - Wi
x-
O
x+
O
Boundary conditions between
beam elements
./
b) Beam deformations including
shear effect
Fig. 5.29. Beam deformation
mechanisms
7·4
Formulation of structural elements
We use
dw
S=--y
dx
(5.48)
(5.49)
_
(L
J pw dx
o
L
-Lm S dx
o
(5.50)
L
+ GAkJ ( ~ ~ - S) o ( ~ ~ -S) dx
o
L
-i p oW dx
o
L
-i m oS dx = 0
o
(5.51)
7·5
Formulation of structural elements
E = Young's modulus, G = shear modulus
3
k
=§.. A = ab I = ab
6 ' , 12
Fig. 5.30. Formulation of two­
dimensional beam element
(b) Two, three- and four-node models;
0i ={3i ' i=1, ... ,q (Interpolation
functions are given in Fig. 5.4)
Fig. 5.30. Formulation of two­
dimensional beam element
7·6
Formulation of structural elements
The interpolations are now
q
W = ~ h.w.
L..J 1 1
i =,
q
B = ~ h.e.
L..J 1 1
i =,
(5.52)
w = H U' B = H U
1-/-' .:...:.s-
dW = BU' ~ = B U
dX 1-/ -' dX ~ -
Where
T
Q. =[w, W
q
8, 8
q
J
~ = [h, h
q
0 OJ
~ = [0 0 h, hqJ
(5.53)
(5.54)
and
!!w = J-
1
[ : ~ l ... :> 0... 0]
__, f, dh, dh
q
]
~ - J L
O
... a dr' ... ar (5.55)
7·7
Formulation of structural elements
So that
K=E1
f
1 T
det J dr
-1
and
+ GAk
t
-1
T
J dr
(5.56)
R=
f
p det J
dr
-1
+/
mdet J dr
(5.57)
-1
Considering the order of inter­
polations required, we study
GAk (5.60)
ex. = IT
Hence
- use parabolic (or higher-order)
elements
. discrete Kirchhoff theory
- reduced numerical integration
7-8
Formulation of structural elements
Fig. 5.33. Three-dimensional more
general beam element
Here we use
(5.61)
q
Q,z(r,s,t) = L
k=l
q
+ ~ 'b h Q,V
k
2 L.- k k sx
k=l
q q
Q,y(r,s,t) =L h
k
Q,Yk +i L akh
k
Q , V ~ y
k=l k=l
q
+~ '" b h Q,V
k
2 LJ k k sy
k=l
q
h
k
Q,Zk +i L a
k
h
k
Q , V ~ Z
k=l
q
+
~ 2 '" b h £V
k
LJ k k sz
k=l
7·9
Formulation of structural elements
So that
1 0
u (r,s,t) = x- x
v (r,s,t) = ly _ 0y
(5.62)
1 0
w (r,s,t) = z- z
q
v(r,s,t)=L:
k=l
and
q t q k
u(r,s, t) = L: hku
k
+"2 L: akh
k
V
tx
k=l k=l
q
+t .E bkh
k
V ~ x
k=l
t q
hkv
k
+2 L
k=l
q
+tL:
k=l
q
w(r,s,t)=L:
k=l
(5.63)
7·10
Formulation of structural elements
Finally, we express the vectors V ~
and V ~ in terms of rotations about
the Cartesian axes x, y , z ,
k
a k
v = e x
'is ...:..s ~
where
e
k
x
e = e
k
~ y
e
k
z
(5.65)
(5.66)
We can now find
£nn
q
Yni;
=
~ ! 4 ~
(5.67)
k=l
Ynl;;
where
u
T
=
[Uk v
k
w
k
e
k
e
k
e
k
] (5.68)
~ x y z
and then also have
T
nn
E a a £
nn
T n ~
= a Gk a
Y n ~
TnI';;
0 a Gk
Ynl;;
(5.77)
7-11
Formulation of structural elements
and w=w(x,y)
.... --
----
(5.78)
Hence
Fig. 5.36. Deformation mechanisms
in analysis of plate including shear
deformations
E
XX
d
l
\
dX
dS
E
yy
= z
_-.1.
(5.79)
dy
Y
xy
dS
X
_ dS
y
dy dX
dW
Sy Y
yz
dy -
= (5.80)
dW
Y
zx
-+ S
dX x
7·12
Formulation of structural elements
and
LXX
1 v a
L
yy
=
z_E_
v 1 a
2
l-v
a a
l-v
L
xy
2
(5.81)
aw
L
yz
ay - By
E
(5.82) =
2(1+v)
L
ZX
aw + B
ax x
The total potential for the
element is:
1
II=-
2
L
xy
dz dA
+ ~
2
f fh/\yyZ Yzx] ~ y z J dx dA
A -h/2 ~ z x
-fw P dA
A
(5.83)
7·13
Formulation of structural elements
or performing the integration
through the thickness
IT =t iT .<q, .<dA +t // f,; y dA
A A
-I: P dA (5.84)
A
where
K =
as
_ .-J.-
ay
as
x
_ ~
ay ax
; y =
aw + s
ax x
(5.86)
1 v 0
Eh
3
1 0 C =. v
~ 12(l-v
2
)
1-v
0 0
2
7·14
[
1
Ehk
f.s = 2{1+v) 0
(5.87)
Formulation of structural elements
Using the condition c5TI= 0 we
obtain the principle of virtual
displacements for the plate
element.
-fw p dA = 0
A (5.88)
We use the interpolations
q
w = ~ h . w .
LJ 1 1
i=l
q
S =~ h. ei
y LJ 1 x
i=l
and
q
x =~ h . x .
LJ 1 1
i=l
(5.89)
q
Y = ~ h . y .
LJ 1 1
;=1
7·15
Formulation of structural elements
s
Mid-surface
r
\ . . . . - ~ - - - - - t ~
Fig. 5.38. 9 - node shell element
For shell elements we proceed as in
the formulation of the general beam
elements,
(5.90)
7·16
Formulation of structural elements
Therefore,
where
To express Y ~ in terms of
rotations at the nodal- point k
we define
(5.91)
(5.92)
°V
1
k
= (e x Ov
k
) / Ie x °Vkl (5.93a)
- -y -n -y-n
then
V
k
°Vk °V
k
S
..:...n = - ~ O',k + -1 k
(5.94)
7·17
Finally, we need to recognize the
use of the following stress-strain
law
l = ~ h ~
(5.100)
1 v a a a a
1 a a a a
Jl
a a a
T
( 1 _ ~ 2 )
!2sh
~ h = ~ h 1-v
a a
-2-
1-v
a
-2-
symmetric
1-v
2
(5.101)
16· node parent element with cubic interpolation
I-
2
-I
5
• •
2
• •
Some derived elements:
64£>-[>
000
o \'.' .\
Variable - number - nodes shell element
7·18
Formnlalion of structural elements
a) Shell intersections

b) Solid to shell intersection
Fig. 5.39. Use of shell transition
elements
7·19
NUMERICAL
INTEGRATIONS,
MODELING
CONSIDERATIONS
LECTURE 8
47 MINUTES
8·1
Numerical integrations, modeling considerations
LECTURE 8 Evaluation of isoparametric element matrices
Numercial integrations. Gauss. Newton-Cotes
formulas
Basic concepts used and actual numerical opera­
tions performed
Practical considerations
Required order of integration. simple examples
Calculation of stresses
Recommended elements and integration orders
for one-, two-. three-dimensional analysis. and
plate and shell structures
Modeling considerations using the elements.
TEXTBOOK: Sections: 5.7.1. 5.7.2. 5.7.3. 5.7.4. 5.8.1. 5.8.2. 5.8.3
Examples: 5.28. 5.29. 5.30. 5.31. 5.32. 5.33. 5.34.
5.35. 5.36. 5.37. 5.38. 5.39
8·2
Numerical integrations. modeling considerations
NUMERICAL INTEGRATION.
SOME MODELING CONSIDERATIONS
• Newton-Cotes formulas
• Gauss integration
• Practical considerations
• Choice of elements
K = f BT C B dV (4.29)
- V - --
M = J p HT H dV (4.30)
- V --
R = f HT fB dV (4.31 )
~ V - -
T
R = f H
S
fS dS (4.32)
-s S - -
R
r
= f ~ T !.r dV (4.. 33)
V
8·3
Numerical integrations, modeling considerations
In isoparametric finite element
analysis we have:
-the displacement interpolation
matrix t:! (r,s,t)
-the strain-displacement
interpolation matrix ~ (r,s,t)
Where r,s,t vary from -1 to +1.
Hence we need to use:
dV =det.4 dr ds dt
Hence, we now have, for example in
two-dimensional analysis:
+1 +1
!\$ =f f ~ T ~ ~ det Adr ds
-1 -1
+1 +1
M=f f p tlT tt det J dr ds
-1 -1
etc...
8·4
Numerical integrations, modeling considerations
The evaluation of the integrals
is carried out effectively using
numerical integration, e.g.:
..
- . 4J lJ -lJ
1 J
where
a. ..
IJ
F··
-IJ
i, j denote the integration points
= weight coefficients
= B··
T
C B·· detJ··
-IJ - -IJ
-
r
-
r = ±O.577
5 = ±O.577
r =±O.775 5 =± 0.775
r= 0 5=0
,
\
\
2x2 - point integration
8·5
Numerical integrations. modeling coDSideratiODS
z
L.-- - - - . ~ Y 3x3 - point integration
Consider one-dimensional integration
and the concept of an interpolating
polynomial.
1
st
order interpolating
---"'--polynomial in x.
.-8
a
I
I
a+b
-2-
x
b
Numerical integrations, modeling considerations
I actual function F
2
nd
order interpolating
in x .
a a+b
2
b
etc....
In Newton - Cotes integration we use
sampling points at equal distances,
and
b n
{
J LJ 1 1 n
a ;=0
(5.123)
n = number of intervals
Ci
n
=Newton - Cotes constants
interpolating polynomial is of
order n.
8·7
Numerical integrations, modeling considerations
Upper Bound on
Error R. as
Number of a Function of
Intervals n q q Cn cn
q C·
Cn
the Derivative of F
2 3 5 6
1 1
10-I(b-a}lF"(r)
"2 T
2
1 4 1
10-3(b-a)5PV(r)
6" 6" 6"
3
1 3 3 1
1O-3(b-a)5F'V(r)
"8 "8 "8 "8
4
7 32 12 32 7
10-6(b-a)7FVI(r)
90 90 90 90 90
5
19 75 50 50 75 19
10-6(b-a)7Fv'(r)
288 288 US 288 ill 288
6
41 216 27 272 27 216 41
lO-'(b-a)'FVIU(r)
840 840 840 840 840 840 840
Table 5.1. Newton-Cotes numbers
and error estimates.
In Gauss numerical integration we
use
b
fF(r)dr" U
1
F(r
1
) + u
2
F(r
2
) + ••.
a
+0. F(r )+R
n n n
(5.124)
where both the weights a
1
•... •a
n
and the sampling points r1 •...• ~
are variables.
The interp(llating polynomial is now
of order 2n -1 .
8·8
Numerical integrations, modeling considerations
n rj /X,
1 O. (I5 zeros) 2. (I5 zeros)
2 ±0.57735 02691 89626 1 . 0 ס ס o o 0 ס ס o o 0 ס ס o o
3 ±0.77459 66692 41483 0.55555 55555 55556
0 . 0 ס ס o o 0 ס ס o o 0 ס ס o o 0.88888 88888 88889
4 ±0.86113 63115 94053 0.34785 48451 37454
±0.33998 10435 84856 0.65214 51548 62546
5 ±0.90617 98459 38664 0.23692 68850 56189
±0.53846 93101 05683 0.47862 86704 99366
0 . 0 ס ס o o 0 ס ס o o 0 ס ס o o 0.56888 88888 88889
6 ±0.93246 95142 03152 0.17132 44923 79170
±0.66120 93864 66265 0.36076 15730 48139
±0.23861 91860 83197 0.46791 39345 72691
Table 5.2. Sampling points and
weights in Gauss-Legendre numeri-
cal integration.
Now let,
ri be a sampling point and
eli be the corresponding weight
for the interval -1 to +1.
Then the actual sampling
point and weight for the
interval a to bare
a + b + b - a r. and b - a el.
-2- 2 1 2 I
and the ri and eli can be
tabulated as in Table 5.2.
8·9
Numerical integrations, modeling considerations
In two- and three-dimensional analysis
we use
+1 +1
f f F(r,s) dr ds =I: "1
-1 -1 1
or
+1
f F(ri's) ds
-1
(5.131)
+1 +1
f f F(r,s)drds= I: ,,;,,/(ri'sj)
-1 -1 i ,j
(5.132 )
and corresponding to (5.113),

IJ
• = a. a. , where a. and a.
I J I J
are the integration weights for
one-dimensional integration.
Similarly,
+1 +1 +1
f f 1F(r,s,t}drdsdt
-1 -1 -1
=
LJ 1 J 1 J
i,j,k
(5.133 )
and a··k = a. Q. Q
k
.
IJ I J
8·10
Numerical integrations, modeling considerations
Practical use of numerical integration
.The integration order required to
evaluate a specific element matrix
exactly can be evaluated by study­
ing the function f to be integrated.
• In practice, the integration is
frequently not performed exactly,
but the· integration order must be
high enough.
Considering the evaluation of the
element matrices, we note the
following requirements:
a) stiffness matrix evaluation:
(1) the element matrix does
not contain any spurious zero
energy modes (i.e., the rank of
the element stiffness matrix is
not smaller than evaluated
exactly) ; and
(2) the element contains the
required constant strain states.
b) mass matrix evaluation:
the total element mass must be
included.
c) force vector evaluations:
the total loads must be in­
cluded.
8·11
Numerical integrations, modeling considerations
Demonstrative example
2x2 Gauss integration
"absurd" results
3x3 Gauss integration
correct results
Fig. 5.46. 8 - node plane stress
element supported at B by a
spring.
Stress calculations
(5.136)
• stresses can be calculated at
any point of the element.
• stresses are, in general, discon­
tinuous across element
boundaries.
8-12
Numerical integrations. modelingconsiderations
thickness
= 1 cm
A
-p
3xl0
7 2
1 ~ [
E = N/cm
<3>
e.
CD \) = 0.3
I
1>
p
300 N =

:... ..,- -of
3c.m. 3 Coft'1.
A
8 ...
'100 N!Crrt'l.
/
(a) Cantilever subjected to bending
moment and finite element solutions.
Fig.5.47. Predicted longitudinal
stress distributions in analysis of
cantilever.
= a .
8·13
Numerical integratiODS. modeling coDSideratioDS
'?
,
A
,
~
@ B <D
4l ,
,
C.
I
, s_ a ~
"?
v = 0.3
P = lOON
" "
8
&
<D
Co
174+ /lA/e-t'-
Co
A A
®
B 8
<D
Co
c
~ I ".00"'Ie-."
(b) Cantilever subjected to tip-shear
force and finite element solutions
Fig. 5.47. Predicted longitudinal
stress distributions in analysis of
cantilever.
Some modeling considerations
We need
• a qualitative knowledge of the
response to be predicted
• a thorough knowledge of the
principles of mechanics and
the finite element procedures
available
• parabolic/undistorted elements
usually most effective
8-14
Numerical integrations, modeling considerations
Table 5.6 Elements usually effective
in analysis.
TYPE OF PROBLEM
TRUSS OR CABLE
TWO-DIMENSIONAL
PLANE STRESS
PLANE STRAIN
AXISYMMETRIC
THREE-DIMENSIONAL
ELEMENT
2-node
8-node or
9-node
20-node
D
D
3-D BEAM
-=
~
3-node or
4-node
-/
.....
PLATE
SHELL
9-node
9-node or
16-node
L7
~ ~
8·15
Numerical integrations, modeling considerations
4/'1ode
I
elEJmerrt
1
S node
g I'\oole
..

e1er1l(1'It.
i
J
I
I
a) 4 - node to 8 - node element
transition region
8
4- I\oc:(e
4 node

eIemtnt""
A
4- node
el ....
c
119
B U.s
VA
A
'Ve- llA

C U,
Constraint uA = (uC +uB)/2
equations:
vA = (v
C
+vB)/2
b) 4 - node to 4 - node element
transition
/.
!
c) 8 - node to finer 8 - node element
layout transition region
Fig. 5.49. Some transitions with
compatible element layouts
8·16
SOLUTION OF
FINITE ELEMENT
EQUILIBRIUM
EQUATIONS
IN STATIC ANALYSIS
LECTURE 9
60 MINUTES
9·1
Solution of IiDile e1eDleul equilihrilll equations iu slatic aaalysis
LECTURE 9 Solution of finite element equations in static
analysis
Basic Gauss elimination
Static condensation
Substructuring
Multi-level substructuring
Frontal solution
t l> t
T
- factorization (column reduction scheme)
as used in SAP and ADINA
Cholesky factorization
Out-of-core solution of large systems
Demonstration of basic techniques using simple
examples
Physical interpretation of the basic operations
used
TEXTBOOK: Sections: 8.1. 8.2.1. 8.2.2. 8.2.3. 8.2.4.
Examples: 8.1. 8.2. 8.3. 8.4. 8.5. 8.6. 8.7. 8.8. 8.9. 8.10
9·2
SoJutiOD of filile e1emenl equilihrillD equations in slatic analysis
SOLUTION OF
EQUILIBRIUM
EQUATIONS IN
STATIC ANALYSIS
• Iterative methods,
e.g. Gauss-8eidel
• Direet methods
these are basically
variations of
Gauss elimination
- static condensation
- substructuring
- frontal solution
- .L Q. .bT factorization
- Cholesky decomposition
- Crout
- column reduction
(skyline) solver
THE BASIC GAUSS ELIMINATION PROCEDURE
Consider the Gauss elimination
solution of
5 -4
,
0
U,
0
-4 6 -4
,
U
2
,
= (8.2)
,
-4 6 -4 U
3
0
0
,
-4
5 U
4
0
9·3
Solation of finite element eqailihriUl equations in static analysis
STEP 1: Subtract a multiple of
equation 1 from equations 2 and
3 to obtain zero elements in the
first column of K.
r------------
ol l! 16
I 5 -5
I
I
oI _ ~ 29
: 5 5
I
o: -4
5 -4 1 o
1
-4
5
(8.3)
5 -4 o o
9·4
o
o
o
14 16
5-5
r--------
0: ~ _20
I 7 7
I
0: _ 20 65
I 7 14
I
= (8.4)
Solation of finite element eqailillriUl equations in static analysis
STEP 3:
5 -4 1 0 U
1
0
0
14 16
1 U
2
1
S
-s
15 20
.-
8
(8.5)
0 0
7
-T
U
3 "7
r---
7
0 0 0
I 5
U
4
I
"6 "6
I
I
Now solve for the unknowns u
4
,
U
3
' U
2
and U, :
12
=5
(8.6)
1 - (- 1
5
6) U3 - (1) U4 _ 13
U =--------:;-;;-----
2 14 -S
5
19 36 7
o - (-4) 35 - (1)15 - (
0
)"5 _ 8
U = - - - - ~ - - - -
1 5 - "5
9·5
Solution of finite element eqDilihriDlD equations in slatic analysis
STATIC CONDENSATION
Partition matrices into

[!!a] [Ba]
.!Sea c !!c = Be
Hence
and
(8.28)
(
-1) -1
- .!Sec.!Sea !!a = Ba - .!Sec
---------­
K
aa
Example
tee

I
U
1
0 5 I -4 1 0
I
---+------------
-4 6 -4 1 U
2
1
=
1 -4 6 -4 U
3
0
0 1 -4 5 U
4
0

'---y----'

Hence (8.30) gives

-
,- -
6 -4 -4 [1/5] [-4 1
Kaa
= -4 6 -4 1
1 -4 5 0
'--
-
1....-
9·8
so that
14 16
1
5
-5
K =
16 29
-4
-a.a
-5
5
0] 1 -4 5

-
and we have obtained the 3x3
unreduced matrix in (8.3)
SoIltiOl of finite elemelt eqlilihrilDl equations in static aualysis
5 -4 0 VI
:1
-4 6 -4 U
2
1 -4 6 -4 U
3
:1 0 1 -4 5 U
4
14
-!§
U
2
"5 5
-!§
29
-4 U
3
0
5
-5
-4 5 V
4
0
Fig. 8.1 Physical systems
considered in the Gauss elimination
solution of the simply supported beam.
9-7
Solutiou of finite element eqDilihriom equations in static analysis
SUBSTRUCTURING
• We use static condensation on the
internal degrees of freedom of a
substructure
• the result is a new stiffness matrix
of the substructure involving
boundary degrees of freedom only
-?-?
- ~ - o - - o
e--c>---n
l
-6
50x50
Example
......--.- L
32x32
Fig. 8.3. Truss element with
linearly varying area.
We have for the element.
9·8
[
17
~ ~ -20
6L
3
-20
48
-28
SoIali. oIliDile e1emeal eqailihrilDl eqaaliODS ia stalic aaalysis
First rearrange the equations
EA, [ '7
6"L
3
-20
Static condensation of U2 gives
EA, Ir7
6L 3
3] [-20]
- [lJ[-20
25 -28 48
or
ll. EA, [ 1
9 L -1
and
9·9
Solution of fiDile elemeul equilibrilll equati. in slatic aDalysis
Multi-level Substructuring
I' L 'I' L , L ,I. L .1
A 2A 4A, I SA, I I 16A,
, , \
· - U Ur;, U6 U
7
Us Ug
I U2 U
3
u. R
s
Bar with linearly varying area
-I I 1-
U,
-
u
3
u
2
---I

.-
U, u
3
(a) First-level substructure
---I I I I 1-
U,
-
Us
u
3
_I I

I 1-
U,
Us
(b) Second-level substructure
_I I I I I I I I 1-
U,
-
ug
Us.Rr;,
-.
I I I I I I I 1-
U, u
g
(c) Third-level substructure and
actual structure.
Fig. 8.5. Analysis of bar using
substructuring.
'-10
Solution of fiDile e1eDleul equilihrio equti. ill static analysis
Frontal Solution
Elementq Element q + 1 Elementq + 2 Elementq + 3
-----
---
m
m+3
~ I N:"
Element 1 Element 4
4
Wave front Wave front
for node 1 for node 2
Fig.8.6. Frontal solution of plane
stress finite element idealization.
• The frontal solution consists of
successive static condensation of
nodal degrees of freedom.
• Solution is performed in the
order of the element numbering .
• Same number of operations are
performed in the frontal solution
as in the skyline solution, if the
element numbering in the wave
front solution corresponds to
the nodal point numbering in the
skyline solution.
9·11
Solution of finite element equilibrium equations in static analysis
L D L
T
FACTORIZATION
- is the basis of the skyline solu-
tion (column reduction scheme)
- Basic Step
L-
1
K = K
--1 - -1
Example:
5 -4 a 5 -4 a
4
-4 6 -4 a
~ 4 16
5
5 5
=
1
a -4 6 -4
a _16
29
-4
-5 5 5
a a a a -4 5 a -4 5
We note
4 4
-1-
5
-5
L =
1
~ 1
1 -1
a a
5 S-
o a a a a a
9·12
Solution of finite element equilibrium equations in static analysis
Proceeding in the same way
-1 -1
1.
2
1.
1
K:= S
x x x x x
x x x x
S
x ....... x
upper
:=
triangular
x x
matrix
x
x
Hence
or
Also, because ~ is symmetric
where
0:= di agona1 rna t r i x
d.. := s ..
11 11
9·13
Solution of finite eleJDent equilihriDII equations in static analysis
I n the Cholesky factorization, we use
where
t = L D ~
SOLUTION OF EQUATIONS
Using
9·14
K = L 0 L
T
we have
L V= R
o L
T
U = V
where
-I
V := L
- -n-l
and
(8.16)
(8.17)
(8.18)
(8.19)
(8.20)
Solution of finite element equilibrimn equations in static analysis
COLUMN REDUCTION SCHEME
5 -4 1
6 -4 1
6 -4
5
~
4 5
4
5
-5
5
14
-4
14
-4
-
5
5
6 -4
6 -4
5
5
~
5
4 1
5
4 1
-5
5
-5
5
14 8
1
14 8
5 7
5
-7
15
-4
15
-4
T
T
5 5
9·15
Solation of finite element eqailihriam eqaati. in static analysis
X=NONZERO ELEMENT
0= ZERO ELEMENT
_ ~ COLUMN HEIGHTS
SYMMETRIC
o 0 000
o 0 000
'-----,
X 000 X
o 0 000
o 0 x 0 0
o X 000
X X X X 0
X X X X
X XX
X
X
ELEMENTS IN ORIGINAL STIFFNESS MATRIX
Typical element pattern in
a stiffness matrix
SKYLINE
o 0 000
o 0 000
L...-_
X 0 0 0 X
X 0 0 0 X
X 0 X 0 X
X X X 0 X
X X X X X
X X X X
X X X
X X
X
ELEMENTS IN DECOMPOSED STIFFNESS MATRIX
Typical element pattern in
a stiffness matrix
9-16
SYMMETRIC
Solution of finite element equilibrium equations in static analysis
x = NONZERO ELEMENT
0= ZERO ELEMENT
COLUMN HEIGHTS
I I I
-x 0 0 0
1
0 0:0
o 0 0 0:0 010
xix x 010 0 x
XlX 0 010 0 0
xIx 0 0 x 0 0
x 0 X\O 0 0
x xix XIO
xix xix
Ix XlX
xIx
Ix
ELEMENTS IN ORIGINAL STIFFNESS MATRIX
Typical element pattern in
a stiffness matrix using block
storage.
9·17
SOLUTION OF
FINITE ELEMENT
EQUILIBRIUM
EQUATIONS
IN DYNAMIC ANALYSIS
LECTURE 10
56 MINUTES
10·1
Solotion of finite e1mnent eqoiIihrio equations in dynaDlic analysis
LECTURE 10 Solution of dynamic response by direct
integration
Basic concepts used
Explicit and implicit techniques
Implementation of methods
Detailed discussion of central difference and
Newmark methods
Stability and accuracy considerations
Integration errors
Modeling of structural vibration and wave propa­
gation problems
Selection of element and time step sizes
I
Recommendations on the use of the methods in
practice
TEXTBOOK: Sections: 9.1. 9.2.1. 9.2.2. 9.2.3. 9.2.4. 9.2.5. 9.4.1.
9.4.2. 9.4.3. 9.4.4
Examples: 9.1. 9.2. 9.3. 9.4. 9.5. 9.12
10·2
Solution of finite element equilihriDl equations in dyDalDic ualysis
DIRECT INTEGRATION
SOLUTION OF EQUILIBRIUM
EQUATIONS IN DYNAMIC
ANALYSIS
MU+CU+KU=R
-- -- -- -
• explicit, implicit
integration
• computational
considerations
• selection of solution
time step (b. t)
• some modeling
considerations
Equilibrium equations in dynamic
analysis
MU+ C U+ KU= R (9.1)
or
10·3
Solution of finite elelleul equilihrilll equatiolS in dynaJDic analysis
time
time
--
Fig. 1. Evaluation of externally
applied nodal point load vector
tR at time t.
THE CENTRAL DIFFERENCE METHOD (COM)
to = _l_(_t-tlt
u
+ t+tlt
U
) (9.4)
- 2tlt - -
an explicit integration scheme
10·4
Solation of finite eleDlent eqailibrimn equations in dynanaic analysis
Combining (9.3) to (9.5) we obtain
(
-'-M + -'- = t R_ __2_ M)t
u
2 - t - - - - 2 - -

-(-'- M_-'-
2 - - -

(9.6)
where we note
! t!!
= (l5-(m) t
lL
) =
Computational considerations
• to start the solution. use
(9.7)
• in practice. mostly used with
lumped mass matrix and low-order
elements.
10·5
Solution of finite element equilibrium equations in dynamic analysis
Stability and Accuracy of COM
-l'I t must be smaller than l'I t er
Tn
l'It
er
= TI ; Tn = smallest natural
period in the system
hence method is conditionally stable
_ in practice, use for continuum elements,
l'It < l'IL
- e
e = ~
for lower-order elements
L'lL = smallest distance between
nodes
for high-order elements
l'IL = (smallest distance between
nodes)/(rel. stiffness factor)
• method used mainly for wave
propagation analysis
• number of operations
ex no. of elements and no. of
time steps
10·6
Solution of finite elelDent eqoiIibriDII eqoatiou in dynandc analysis
THE NEWMARK METHOD
(9.28)
{9.29J
an implicit integration scheme solution
is obtained using
.In practice, we use mostly
a. = la , 0 = ~
which is the
constant-average-acceleration
method
(Newmark's method)
• method is unconditionally stable
• method is used primarily for analysis
of structural dynamics problems
• number of operations
== ~ n m
2
+ 2 nmt
10·7
Solution of finite element equilibriDII equations in dynmic analysis
Accuracy considerations
• time step !'1t is chosen based
on accuracy considerations only
• Consider the equations

and
where
K¢.
--1
Using
¢"1 K ¢ = 0.
2
2
:: w· <p.
1 --1
where
we obtain n equations from which
to solve for xi(t) (see lecture 11)
10·8
.. 2 T
x. + w. x. = R
1 1 1
i=l, ... ,n
Solution 01 finite eleDlent equilibriDll equations in dynaDlic analysis
Hence, the direct step-by-step
solution of

corresponds to the direct step-by­
step solution of
.. 2
x· + w. x·
1 1 1
with
i=l, ... ,n
n
U =
- 1
i =1
Therefore, to study the accuracy of
the Newmark method, we can study
the solution of the single degree of
freedom equation
.. 2
x+w x=r
Consider the case
.. 2
x + w x = a
o· a
x=
0·· 2
x = -w
10·9
Solotion of finite element eqoiIihriDl equations in dynandc analysis
19.0
19.0
15.0
Houbolt
15.0
method
§
11.0 11.0 ..
le
....
5
-
w
E!:.
..
C
le
.g
7.0 0 7.0
'"

C/I
C
0
>
"iii
'" u
"8
5.0 '"
5.0
"0
.;:
'"
'"
"0
Co
'"

C/I
Q.
:!
3.0
E
3.0 c
'"
'"
8.
l:!
tf :!
c

'"

l:!
'"
Q"
1.0 1.0
PE
0.06 0.10 0.14 0.18 0.06 0.10 0.14 0.18
Fig. 9.8 (a) Percentage period elonga-
Fig. 9.8 (b) Percentage period elonga-
tions and amplitude decays.
tions and amplitude decays.
4t-----r----:--r--r----r-----,...-----,-----,
equation
.. 2 . 2 . t
x + + w x = S1n p
static
response
2
1
31----+--f-+-+----t-----t-----.,t----'-1
...
o
'0

"t:J
CtI
o
CJ
'E
CtI
c::

o
1 2 3
Fig. 9.4. The dynamic load factor
10·10
SoIIIi. of filile 81••1eqailihrillD eqaaliOlS in dJllillic analysis
D:.r • 1.05
- nYNAMIC RESPONSE
_ .. - STATIC RESPONSE
=0.05
g
7T
+
gi
r.it
!2C \ '
1
- .j.
'" ,
fs!
,;1
• 1
!
T
81
74- ._--+-- -t-" - -- .... __..--t- ---+-._--+_.. - ........ -... _-.-1
'c.oe o.. fl. 'JO n. I. 00
I I
Response of a single degree
of freedom system.
DLF .. 0.50
- DYNAMIC RESPONSE
--- STATIc.: RESPONSE
.f... = 3.0
w
....
.... , ------ - ./ /'
__
/"7--_____ ......
+
g,
::i- " .. -------t-----+---+I--t- __---+--+1 ::-:---+----,+1:::---+----,+1::-:---+------::+-'::-:---+-----:<'
c.':;: C.25 :."L :.00 : . I.SO 1.75 2.00 2.25 2.50 2.75 3.00
TIllE
Response of a single degree
of freedom system.
10·11
Solution of finite element equilibrium equations in dynamic analysis
Modeling of a structural vibration
problem
1) Identify the frequencies con­
Fourier analysis if necessary.
2) Choose a finite element mesh
that accurately represents all
frequencies up to about four
times the highest frequency
u
3) Perform the direct integration
analysis. The time step /':, t for
this solution should equal about
1
20 Tu,where T
u
= 2n/w
u
'
or be smaller for stability reasons.
Modeling of a wave propagation
problem
If we assume that the wave length
is L
w
' the total time for the
wave to travel past a point is
(9.100)
where c is the wave speed. Assuming
that n time steps are necessary to
represent the wave, we use
(9.101 )
and the "effective length" of a
finite element should be
10·12
c /':,t (9. 102)
SoIaliOi .. filile 81••1eqailihriDl eqaali_ in dJUlDic ualysis
SUMMARY OF STEP-BY-STEP INTEGRATIONS
-INITIAL CALCULATIONS ---
1. Form linear stiffness matrix K,
mass matrix M and damping
matrix ~ , whichever appl icable;
Calculate the following constants:
Newmark method: 0 > 0.50, ex. 2:. 0.25(0.5+0)2
2
a
O
= , / (aAt )
a
4
= 0/ ex.- ,
as =-a
3
a,=O/(aAt)
as = I1t(O/ex.- 2)/2
a
g
= I1t(' - 0)
a3 = , / (2ex. )- ,
a
7
=-a
2
Central difference method:
a, = '/2I1t
... 0 O· 0··
2. Inltlahze !!., !!., !!. ;
For central difference method
only, calculate I1t
u
from
initial conditions: -
3. Form effective linear coefficient
matrix;
in implicit time integration:
in explicit time integration:
M= a ~ + a,f.
10·13
Solution of finite element equilibrium equations in dynamic analysis
4. In dynamic analysis using
implicit time integration
triangularize R:.
--- FOR EACH STEP ---
(j) Form effective load vector;
in implicit time integration:
in explicit time integration:
(ii) Solve for displacement
increments;
in implicit time integration:
in explicit time integration:
10·14
SoI.ti. of filile elOl.1 equilihriDl equations in dynamic analysis
Newmark Method:
Central Difference Method:
10·15
MODE SUPERPOSITION
ANALYSIS; TIME
BISTORY
LECTURE 11
48 MINUTES
11·1
Mode slperpClilion analysis; lillie bistory
LECTURE 11 Solution of dynamic response by mode
superposition
The basic idea of mode superposition
Derivation of decoupled equations
Solution with and without damping
Caughey and Rayleigh damping
Calculation of damping matrix for given
damping ratios
Selection of number of modal coordinates
Errors and use of static correction
Practical considerations
TEXTBOOK: Sections: 9.3.1. 9.3.2. 9.3.3
Examples: 9.6. 9.7. 9.8. 9.9. 9.10. 9.11
11·2
Mode superposition analysis; time history
Mode Superposition Analysis
Basic idea is:
transform dynamic equilibrium
equations into a more effective
form for solution,
using
!L = 1:. !(t)
nxl nxn nxl
P = transformation matrix
! (t ) =general ized displacements
Using
!L(t) = 1:. !(t)
on
MU+ c 0 + KU= R
we obtain
(9.30)
(9.1)
~ R(t) + f i(t) + R!(t) ~ ( t )
(9.31)
where
C fT ~ f ;
R= PT R
(9.32)
11·3
(9.34)
Mode sDperJMlilion ualysis; tiDle history
An effective transformation matrix f
is established using the displacement
solutions of the free vibration equili­
brium equations with damping
neglected,
M 0 + K U = 0
Using
we obtain the generalized eigenproblem,
(9.36)
with the n eigensolutions ( w ~ , p..,) ,
2 2
(
ul
2 ' ~ ) , ... , (w
n
' .P.n) , and
11·4
T 1== 0'
<P
1
" M'" " - _.:t:..
J
i = j
i ., j
2
< W
- n
(9.37)
(9.38)
Mode superposition analysis; time history
Defining
(9.39)
we can write
and have
(9.40)
Now using
!L(t) = ! ~ J t )
¢T M¢ = I (9.41)
(9.42)
we obtain equilibrium equations
that correspond to the modal
generalized displacements
!(t) + !T ~ ! !(t) + r;i ~ ( t ) = !T!S.(t)
(9.43)
The initial conditions on ~ ( t ) are
obtained using (9.42) and the
M - orthonormality of ¢; i.e.,
at time 0 we have
(9.44)
11·5
Mode SUperpClitiOD aualysis; tilDe bislory
Analysis with Damping Neglected
(9.45)
i.e., n individual equations of
the form
2
.x .(t) + w. x. (t) = r. (t )
1 1 1 1
where
with
T a
X'I = lj). M U
1 -1 - -
t=O
• .T O'
X'I =-'-- cp.M U
1 -1 - -
t=O
i = ',2, ... ,n
(9.46)
(9.47)
Using the Duhamel integral we have
=-' jtr1·(T) sinw.(t-T)dT
w. 1
1 0 (9.48)
+ a.. sin w.t + 8. cos w·t
111 1
where a.i and 8i are determined
from the initial conditions in (9.47).
And then
11-&
(9.49)
Mode sDperp.ition analysis; time history

equation
•• 2 . 2 .
x+ E;,wx + W X = S1 n P t
static
response
\-0
31-__-+__+--+-+-__+-__-+ -+-__--.,
0
2
....
0
.....
u
CtI
.....
-0
CtI
0
u
E
CtI
r:::::
>-
0
2 3
Fig. 9.4. The dynamic load factor
Hence we use
uP = ¢. x· (t)
-- 1
i =1
where
uP - U
The error can be measured using
(9.50)
11·7
Mode superposition analysis; time history
Static correction
Assume that we used p
modes to obtain , then let
n

i =1
Hence
T
r. = ¢. R
1 -1-
Then
and
K flU fiR
Analysis with Damping Included
Recall, we have
!(t) + !T f!i(t) + fi !(t) = !T
(9.43)
If the damping is proportional
T
¢. C(po = 2w. E;,. cS. .
-1 ---J 1 1 1J
and we have
(9.51)
x.(t) + 2w. E;,. x.(t) + x.(t) = r
1
·(t)
1 1 1 1 1 1
i=l, ... ,n
(9.52)
11·8
Mode superposition analysis; time history
A damping matrix that satisfies the
relation in (9.51) is obtained using
the Caughey series,
(9.56)
where the coefficients ak ' k = , , ••• , p ,
are calculated from the p simultane-
ous equations
A special case is Rayleigh damping,
C = a ~ 1 + B K
- -- --
example:
Assume ~ , = 0.02
w, = 2
calculate a and B
We use
(9.55)
or
'/
a + Bw:- 2w. ~ .
- - 1 1 1
2w. ~ .
1 1
11·9
Mode superposition analysis; time history
Using this relation for wl ' [,1 and
w2 ' [,2 ' we obtain two equations
for a and 13:
a + 4ii = 0.08
a + 913 = 0.60
The solution is a = -0.336
and 13 = O. 104 . Thus the
damping matrix to be used is
C = -0.336 M+ 0.104 K
Note that since
2
a + 13 w. = 2w. [, .
1 1 1
for any i, we have, once a and
13 have been established,
E,. =
1
2
a + SW.
1
2w.
1
a 13
= - + - w
2w. 2 i
1
11·10
Mode sDperp.ition analysis; time history
Response solution
As in the case of no damping.
we solve P equations
x. + 2w. E,. x· + w ~ x. = r.
1 111111
with

1
I
TO
xi t =0 "--. !i!i .!:L
• ITO'
xi t = 0 = !i f1 .!:L
and then
P
uP ~ ¢ . x. (t)
LJ-1 1
i =1
Practical considerations
mode superposition analysis
is effective
- when the response lies in a
few modes only, P« n
- when the response is to be
obtained over many time in­
tervals (or the modal response
can be obtained in closed form).
e.g. earthquake engineering
vibration excitation
- it may be important to
calculate E
p
(t) or the
static correction.
11·11
SOLUTION METHODS
FOR CALCULATIONS
OF FREQUENCIES
AND MODE SBAPES
LECTURE 12
58 MINUTES
12·1
Solution methods for calculations of frequencies and mode shapes
LECTURE 12 Solution methods for finite element
eigenproblems
Standard and generalized eigenproblems
Basic concepts of vector iteration methods.
polynomial iteration techniques. Sturm
sequence methods.' transformation methods
Large eigenproblems
Details of the determinant search and subspace
iteration methods
Selection of appropriate technique. practical
considerations
TEXTBOOK: Sections: 12.1. 12.2.1. 12.2.2. 12.2.3. 12.3.1. 12.3.2.
12.3.3. 12.3.4. 12.3.6 (the material in Chapter 11
is also referred to)
Examples: 12.1. 12.2. 12.3. 12.4
12·2
Solatiu methods for calculations of frequencies and mode shapes
SOLUTION METHODS FOR
EIGENPROBLEMS
Standard EVP:
r! = \ !
nxn
Generalized EVP:
!sP.-=\!i! - (\=w
2
)
Most emphasis on the generalized
EVP e.g. earthquake engineering
"Large EVP" n> 500
m> 60
1
p=l, ... ,3"n
In dynamic analysis, proportional
damping
r sP.- = w
2
!i!
If zero freq. are present we can
use the following procedure
r sP.- + )1 Ii sP.- = (w
2
+ ~ r ) ! i sP.-
or
(r+)1 !i)sP.- = \ !i sP.­
\ = w
2
+ )1
or
2
W =\-)1
12·3
Solation lIethods lor calcalatiou oIlreqa.cies and lIode shapes
p(A)
p(A) = det(K- A ~ )
In buckling analysis
. ! \$ . ! = A ~ !
where
p(A) = det ( ~ - A ~ )
p(A)
12·4
Solution methods for calculations of frequencies and mode shapes
Rewrite problem as:
and solve for largest K:
..
..
--
-
~
( ~ - ~ ~ ) ! = n .K 2£.
form the generalized EVP or
quadratic EVP into a stand­
ard form, then solve using
one of the many techniques
available
e.g.
.Ki=;\!ii
M=I::I::T i=hTjJ
hence
~ :t = ;\ i ; K= 1::-
1
K [-T
or
M= W0
2
W
T
etc ...
12·5
SolotiOl .elhods lor calcolations oIlreqoeacies ud .ode sllapes
Direct solution is more effective.
Consider the Gen. EVP ! ! = AM!
with
1.
3
. .. 1
n
eigenpairs ( Ai' 1.i)
are required or
i=l, ,p
i=r, ,s
The solution procedures in use
operate on the basic equations
that have to be satisfied.
1) VECTOR ITERATION TECHNIQUES
Equation:
e.g. Inverse It.
~ P _ = A ~ ~
! ~ + l = M~
~ + l
• Forward Iteration
• Rayleigh Quotient Iteration
can be employed to cal­
culate one eigenvalue
and vector, deflate then
eigenpair
Convergence to "an eigenpair",
which one is not guaranteed
(convergence may also be slow)
12·6
Solution methods for calculations of frequencies and mode shapes
2) POLYNOMIAL ITERATION METHODS
! ~ = A ~ ~ ~ (K - A M) ¢ 0
Hence
p(A) det ( ~ - A!:1) = 0
,
,
,
Newton Iteration
p(A)
2 n
a
O
+ alA + a
2
A + ... + anA
b
O
(A-Al) (A-A2) '" (A-An)
Implicit polynomial iteration:
Explicit polynomial iteration:
eExpand the polynomial and
iterate for zeros.
eTechnique not suitable for
larger problems
- much work to obtain ai's
- unstable process
p (Pi) = det (IS. - Pi !y!)
= det L D LT = II d ..
-- - . II
I
e accurate, provided we do not
encounter large multipliers
e we directly solve for Al, ...
e use SECANT ITERATION:
Pi+l = Pi -
e deflate polynomial after
convergence to A1
12-7
Solution methods for calculations of frequencies and mode shapes
]J. 1
1-
p (A) / (A-A,)
I
I
II
I
Convergence guaranteed to A1 ' then
A2 , etc. but can be slow when we
calculate multiple roots.
Care need be taken in L D LT factor­
ization.
12·8
SaI.liOi .6Jds for calculali. of freql8cies iIIld ole shapes
3) STURM SEQUENCE METHODS
1 2 3 4
t
:::}
· . .
. ..
! <p = A!11 9· ~ ; ; .
· . .
. . -. .. .
· . .
· .. .
· . .
Number of negative elements in
D is equal to the number of
eigenvalues smaller than J.1 S .
3rd associated
constraint problem
2nd associated
constraint problem
1st associated
constraint problem
12·9
Solution lDethods lor calculations 01 frequencies ud lDode shapes
3) STURM SEQUENCE METHODS
T
Calculate - ].lS. =h Qh
,
Count number of negative elements
in Q and use a strategy to isolate
eigenvalue(s) .
interval
,
,
,
,/
].ls
1
].lS2
T f ..
• Need to take care in L D L aetonzatlon
---
• Convergence can be very slow
4) TRANSFORMATION METHODS
j
<PTK<P=A
­
<P M<P = I
- - -
Construct <P iteratively:
_
n =[Al ... 'n]
<P = J; H A
--- --..
....
"
12·10
5oI1tiOi .elhods for calculations 01 frequencies ad .ode shapes
T T T
~ ... ~ ~ l ff
1
~ ... ~ - ~
T T T
~ ... ~ f
1
!i f
1
~ ... ~ - l
e.g. generalized Jacobi
method
• Here we calculate all eigenpairs
simultaneously
• Expensive and ineffective
(impossible) or large problems.
For large eigenproblems it is best
to use combinations of the above
basic techniques:
• Determinant search
to get near a root
• Vector iteration to obtain
eigenvector and eigenvalue
• Transformation method for
orthogonalization of itera­
tion vectors.
• Sturm sequence method to ensure
that required eigenvalue(s) has
(or have) been calculated
12·11
Solution methods for calCl1atiou of frequencies and mode sJlapes
THE DETERMINANT SEARCH METHOD
p(A)
A
1) Iterate on polynomial to obtain
shifts close to A1
P(l1;) =det ( ~ - 11; ~ )
T
=det L D L = n d ..
--- ; 11
11;+1 = ].1; - n P(l1;) - P(11;_1)
11;-11;_1
n is normally =1.0
n=2. , 4. , 8. ,... when convergence
is slow
Same procedure can be employed to
obtain shift near A; , provided
P( A) is deflated of A1' . . . ,A; _1
2) Use Sturm sequence property to
check whether 11 ; +1 is larger
than an unknown eigenvalue.
12·12
~ - ..
/ . ..
....
....
....
Solution lOethods for calculations of freqoBcies ud lOode shapes
3) Once lJ i +1 is larger than an
unknown eigenvalue, use inverse
iteration to calculate the eigenvector
and eigenvalue
lJi+1
k =1,2, ...

~ + l
~ + l
=
- T - ~
( ~ + l !i ~ + l )
- T
p ( ~ + l )
=
~ + l !i ~ k
- T ~
~ + l !i ~ + l
4) Iteration vector must be deflated
of the previously calculated
eigenvectors using, e.g. Gram­
Schmidt orthogonalization.
If convergence is slow use Rayleigh
quotient iteration
12·13
Solution methods lor calculations oIlrequencies ud mode shapes
Calculates only eigenpairs actually
required; no prior transformation
of eigenproblem
Many triangular factorizations
• Effective only for small banded systems
We need an algorithm with less
factorizations and more vector iterations
when the bandwidth of the system is large.
SUBSPACE ITERATION METHOD
Iterate with q vectors wher:' the
lowest p eigenvalues and eigen­
vectors are required.
inverse {K
4+1
= ',1
4
k=1,2, ...
iteration --
~ + 1
-T
K
- ~ + 1
=
4+1
~ + 1
-T
~ 1
4+1
=
4+1
~ + 1 ~ + 1 = ~ + 1 ~ + 1 ~ + 1
4+1 = ~ + 1 ~ + 1
12·14
Solution methods for calculations of frequencies and mode shapes
"Under conditions" we have
CONDITION:
starting subspace spanned
by X, must not be orth­
ogonal to least dominant
subspace required.
Use Sturm sequence check
eigenvalue
p eigenvalues
T
!5. - flS t1 = ~ Q~
no. of -ve elements in D must
be equal to p.
Convergence rate:
flS
convergence reached
when
< tal
12·15
Solution methods for calculations of frequencies and mode shapes
Starting Vectors
Two choices
1)
x. = e.,

j=2, ... ,q-l
2.
x = random vector
4
2) Lanczos method
Here we need to use q much
larger than p.
Checks on eigenpairs
1. Sturm sequence checks

E:.=
1 [I K II
- -1 2
important in!!!. solutions.
Reference: An Accelerated Subspace
Iteration Method, J. Computer
Methods in Applied Mechanics
and Engineering, Vol. 23,
pp. 313 - 331,1980.
12·16
MIT OpenCourseWare
http://ocw.mit.edu
Resource: Finite Element Procedures for Solids and Structures
Klaus-Jürgen Bathe
The following may not correspond to a particular course on MIT OpenCourseWare, but has been
provided by the author as an individual learning resource.

PREFACE
The analysis of complex static and dynamic problems in­ volves in essence three stages: selection of a mathematical model, analysis of the model, and interpretation of the results. During recent years the finite element method implemented on the digital computer has been used successfully in modeling very complex problems in various areas of engineering and has significantly increased the possibilities for safe and cost­ effective design. However, the efficient use of the method is only possible if the basic assumptions of the procedures employed are known, and the method can be exercised confidently on the computer. The objective in this course is to summarize modern and effective finite element procedures for the linear analyses of static and dynamic problems. The material discussed in the lectures includes the basic finite element formulations em­ ployed, the effective implementation of these formulations in computer programs, and recommendations on the actual use of the methods in engineering practice. The course is intended for practicing engineers and scientists who want to solve prob­ lems using modem and efficient finite element methods. Finite element procedures for the nonlinear analysis of structures are presented in the follow-up course, Finite Element Procedures for Solids and Structures - Nonlinear Analysis. In this study guide short descriptions of the lectures and the viewgraphs used in the lecture presentations are given. Below the short description of each lecture, reference is made to the accompanying textbook for the course: Finite Element Procedures in Engineering Analysis, by K.J. Bathe, Prentice­ Hall, Inc., 1982. The textbook sections and examples, listed below the short description of each lecture, provide important reading and study material to the course.

Contents

Lectures
l. 2.
Some basic concepts of engineering analysis Analysis of continuous systems; differential and variational formulations Formulation of the displacement-based finite element method_ Generalized coordinate finite element models Implementation of methods in computer programs; examples SAp, ADINA Formulation and calculation of isoparametric models Formulation of structural elements Numerical integrations, modeling considerations Solution of finite element equilibrium equations in static analysis Solution of finite element equilibrium equations in dynamic analysis Mode superposition analysis; time history Solution methods for calculations of frequencies and mode shapes

1-1

2-1 3-1 4-1

3. 4. 5.

5-1 6-1 7-1 8-1

6. 7. 8. 9.

9-1

10.

10-1 11-1

1l. 12.

12-1

SOME BASIC CONCEPTS OF ENGINEERING ANALYSIS LECTURE 1 46 MINUTES I-I .

3. 3.2.10. 3.2.9.7.2.SolIe basic ccnacepls of eugiDeeriDg ualysis LECTURE 1 Introduction to the course. 3.11. 3.3.8.2. 3. 3.14 1-2 . 3. 3.3. 3. discrete and continuous systems.5. 3. 3.1.2. 3.2. 3.4 Examples: 3.12. objective of lectures Some basic concepts of engineering analysis. problem types: steady-state.4.6.1 and 3. propagation and eigen­ value problems Analysis of discrete systems: example analysis of a spring system Basic solution requirements Use and explanation of the modern direct stiff­ ness method Variational formulation TEXTBOOK: Sections: 3. 3.1.13. 3.

nuclear.methods for solution of the governing equations .we now see applications in linear.various computer programs are available and in significant use My objective in this set of lectures is: • to introduce to you finite element methods for the linear analysis of solids and structures. • 'n civil. engineering • Since the first applications two decades ago. static and dynamic analysis. and their practical usage. ocean.the formulation of the finite element equilibrium equations .. ["Iinear" meaning infinitesi­ mally small displacements and linear elastic material proeer­ ties (Hooke's law applies)j • to consider .computer implementations . 1·3 .. .to discuss modern and effective techniques.. .Some basic concepts 01 engineering aulysis INTRODUCTION TO LINEAR ANALYSIS OF SOLIDS AND STRUCTURES • The finite element method is now widely used for analysis of structural engineering problems. mechanical. nonlinear. biomechani­ cal. mining.the calculation of finite element matrices . aeronautical.

iL-_I_n_te_r.- ~ Establish finite element model of physical problem Revise (refine) the model? I : I I..J.-__ S_ol_v_e_th_e_m_o_d_el_ _ ~ ... Bathe). Inc. (by K. Finite Element Solution Process Physical problem .. .Some basic concepts of engineering analysis REMARKS • Emphasis is given to physical explanations rather than mathe­ matical derivations • Techniques discussed are those employed in the computer pro­ grams SAP and ADINA SAP == Structural Analysis Program ADINA =Automatic Dynamic Incremental Nonlinear Analysis • These few lectures represent a very brief and compact introduction to the field of finite element analysis • We shall follow quite closely certain sections in the book Finite Element Procedures in Engineering Analysis.. Prentice-Hall...p_re_t_t_h_e_re_s_u_lt_S_..J 1 ~- 1-4 ..

1·5 .-Fault \\(no restraint assumed) Altered' grit E= toEc . 12 at 15° Analysis of cooling tower. Analysis of dam.SolIe basic concepts of engiDeering analysis 10 ft 15 ft I .. K~~~~~-~..

Some basic concepts of engineering analysis

B

.

W

o

E~~;;C=-------_

F

....

Finite element mesh for tire inflation analysis.

1·6

SolDe basic concepts of engineering analysis

Segment of a spherical cover of a laser vacuum target chamber.

p

l,W

p
-0.2
Mil
• 16x 16 MESH

PINCHED CYLINDRICAL SHELL

OD;:...,...----.---~~~~~~C
EtW -50 P- 100
-150 • 16x 16 MESH

""= 0.1
~
C
BENDING MOMENT DISTRIBUTION ALONG DC OF PINCHED CYLINDRICAL SHELL

-200 DISPLACEMENT DISTRIBUTION ALONG DC OF PINCHED CYLINDRICAL SHELL

1-7

SoBle basic concepts 01 engineering analysis

IFinite element idealization of wind
tunnel for dynamic analysis

SOME BASIC CONCEPTS OF ENGINEERING ANALYSIS

The analysis of an engineering system requires: - idealization of system - formulation of equili­ brium equations - solution of equations - interpretation of results

1·8

implemented on digital computers ANALYSIS OF DISCRETE SYSTEMS Steps involved: .equations Analysis of complex continu­ ous system requires solution of differential equations using numerical procedures reduction of continuous system to discrete form powerful mechanism: the finite element methods.system idealization into elements .solution of response 1·9 .element assemblage .Some basic concepts of engineering analysis SYSTEMS DISCRETE response is described by variables at a finite number of points CONTINUOUS response is described by variables at an infinite number of points set of differ­ ential equations PROBLEM TYPES ARE • STEADY -STATE (statics) • PROPAGATION (dynamics) • EIGENVALUE For discrete and continuous systems set of alge­ braic .evaluation of element equilibrium requirements .

. F(3) k3 '3 [ ] -1 -r ]f 1 2 F(3) -. [F1 4' ] 1 U F(4) '2 [ u..Some basic concepts of engineering analysis Example: steady . .2 2 '5 [-11-t] = [F(5l] 1 u F1S ) 3 3 F(S) 3 1 F(3)l] P u 2 1·10 . F(2) ~ : ~\l) 1 k. 3 3 F(4) 3 u2 F(2 ) --.2 '4 [-11-1]["1] .state analysis of system of rigid carts interconnected by springs Physical layout ELEMENTS U1 U 3 I -. -1 1-1]["I]fF}] 1 u F( 2 ) 2 2 u2 - F(S) 2 --. F(4) 1 . u1 .F(' ) .

. · ...... Equilibrium equations KU= R T U = [u 1 (a) R - T = [R 1 K = · · · +k4 · · -k k1 + k2 + k3 ~ -k 2 .. · ............·. ... · ...........SolIe basic cOIcepls of engineering analysis Element interconnection requirements : F(4) + F(S) = R 3 3 3 These equations can be written in the form KU = E. ...........•. ..k3 : 4 .... · · . -k 2 ...•............. ........ ...k3 ~ k2 + k3 + k S~ -k S 1·11 .... '" . ...•.

This assemblage process is called the direct stiffness method The steady..state analysis is completed by solving the equations in (a) 1·12 ..Some basic concepts of engineering analysis and we note that ~= i =1 t ~(i) where : :] o 0 etc .

. .... U1 .... ........ ~~ ~1---...... . . ...... ...... .. · · · · · :.. .l\fl--r/A~.... ... ....... ..... K ••• ~ ~~....JI..... : . ..1\1\1\~~r/A 1·13 ... . ..Some basic concepls 01 engineering analysis · .... · · · ...... u... ...... ....... .... .. : . : · .... K= K1 : · · · · · :. ~ u.......~ ••••••••• ••••••••••• :..

. .-K 2 -K 3 'O'O'O 'O'O'O'O'O'O'O'O'O'O'O'O'O'O:'O'O'O'O'O'O'O'O'O'O'O'O'O'O:'O'O'O'O'O'O'O'O'O'O'O'O'O'O · · · K= . K 1 +K 2 + K 3 . + K4 . .SOlDe basic concepts of engineering analysis u. u. -K 4 K 1 +K 2 + K 3 ~-K2 -K 3 'O'O • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • K= -K 2 -K 3 ~K2+ K3 + K5 -K 5 'O'O • • • • • • • • • • • • • • • : ••••••••••••••••••••••••••••• I K 1·14 . .. . . + K4 . u. . . . . . . : . . K= • 'O • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • · : · · · · · · . . • 'O'O • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • · · · · .

alternatively we could have used a variational approach.Some basic concepts of engineering analysis In this example we used the direct approach. In the variational approach we operate on an extremum formu lation: u = strain energy of system total potential of the loads W = Equilibrium equations are obtained from ~- an 1 0 (b) In the above analysis we have U=~UT!!! W = U R Invoking (b) we obtain KU = R T Note: to obtain U and W we again add the contributions from all elements 1·15 .

M= a a a a m a 2 a m 3 EIGENVALUE PROBLEMS we are concerned with the general ized eigenvalue problem (EVP) Av = ABv !l .SOlDe basic concepts of engineering analysis PROPAGATION PROBLEMS main characteristic: the response changes with time ~ need to include the d'Alembert forces: For the example: m. .!l are symmetric matrices of order n v is a vector of order n A is a scalar EVPs arise in dynamic and buckling analysis 1·16 .

Some basic concepts of engineering analysis Example: system of rigid carts ~lU+KU=O Let U = <p sin W(t-T) Then we obtain _w 2 ~~ sin W(t-T) + -K - <p sin W(t-T)= -0 Hence we obtain the equation There are 3 solutions w.~. (l)2 ' ~2 w3 ' ~3 eigenpairs In general we have n solutions 1·17 . .

ANALYSIS OF CONTINUOUS SYSTEMS. DIFFEBENTIAL AND VABIATIONAL FOBMULATIONS LECTURE 2 59 MINUTES 2-1 .

23.3.2.25 2·2 .17.19.20. least squares methods Ritz analysis method Properties of the weighted residual and Ritz methods Example analysis of a nonuniform bar. 3. 3.21. and the differ­ ential formulation Weighted residual methods.18. 3. solution accuracy. the principle of virtual displacements. 3.22. differential and variational lonnDlations LECTURE 2 Basic concepts in the analysis of continuous systems Differential and variational formulations Essential and natural boundary conditions Definition of em-I variational problem Principle of virtual displacements Relation between stationarity of total potential. introduction to the finite element method TEXTBOOK: Sections: 3. 3.1.16.15.3 Examples: 3. 3. Galerkin.3. 3.Analysis 01 continnous systems.3. 3. 3. 3. 3.24. 3.

..Analysis of continuous systems.. -----41~_ least squares t Ritz Method . differential and variational formulations BASIC CONCEPTS OF FINITE ELEMENT ANALYSIS ­ CONTINUOUS SYSTEMS • Some additional basic concepts are used in analysis of continuous systems • We discussed some basic concepts of analysis of discrete systems CONTINUOUS SYSTEMS differential formulation variational formulation Weighted residual methods _ Galerkin _.. finite element method 2·3 .

+~~ dx I~ p Area A. E mass density. mass density I = p A au ~ 2 au = E ­ax Combining the two equations above we obtain 2·4 . cross-sectional area. A R.. differential and .aA Ix oX The constitutive relation is a r ~ . ~------- The problem governing differential equation is Derivation of differential equation The element force equilibrium require­ ment of a typical differential element is using d'Alembert's principle aA Ix + A ~a Xdx .Differential formulation / ~Lt: ) Young's modulus. Example .arialionalIOl'llulali.Analysis of continuous systeDlS.-.

t} =° 9 essential (displ.C. differatial aDd variationaliOl'lDDlatiODS The boundary conditions are u(O.t) = R O with initial conditions u(x. 2·5 .c.O} = ° ° ~ (x ' O) = at In general.baIysis 01 COitiDlOU systems. highest order of (spatial) deriva­ tives in essential b. 9 natural (force) B.C.) B.c. is (m-1) highest order of spatial deriva­ tives in natural b. we have highest order of (spatial) deriva­ tives in problem-governing dif­ ferential equation is 2m. EA ~~ (L. is (2m-1) Definition: We call this problem a Cm-1 variational problem.

differential and variatioD.u o LR u o=0 =0 and we have 0 II The stationary condition 6II = 0 gives L au au rL.dx r .Analysis 01 continuous systems. In general.6u L R = 0 This is the principle of virtual displacements governing the problem.Variational formulation We have in general II=U-W For the rod fL II = J } EA o and au 2 (--) ax dx - i L u f B dx .a1 fOl'llolatiODS Example .B JO(EA ax)(6 ax) dx -)0 6u t. we write this principle as or (see also Lecture 3) 2·6 .

differential and variatiooallormulatioDS However. Writing ax a8u for 8 au ax . we must have and EAaxAlso au I x=L=R f B .EA ~\ dX ~I ax x=L x=o Since QUO is zero but QU is arbitrary at all other points. at x = l .c. we can now derive the differential equation of equilibrium and the b..lIiIysis of . re- calling that EA is constant and using integration by parts yields dx + [EA .IiDIGUS systems. = -A p - a2u at 2 and hence we have 2·7 .

I ~ Principle of Virtual Displacements Integration by parts _ solve problem I ~ Differential Equation of Equilibrium and natural b. only we generate • the principle of virtual displacements • the problem-governing differ­ ential aquatio!) • the natural b.. (these are in essence "contained in" IT . inW).c.c.Analysis of cODtiDaoas syst_ diIIereatial and variatioul fOlllalatiODS The important point is that invoking o IT = 0 and using the essential b.c.c. i.e. Total Potential IT Use oIT = 0 and essential "b. _solve problem 2·8 . • We use integration by parts m-times. In the derivation of the problem­ governing differential equation we used integration by parts • the highest spatial derivative in IT is of order m .

[</>] = 1 q.. In all techniques we determine the ai so as to make a weighted average of R vanish.11 ) The various weighted residual methods differ in the criterion that they employ to calculate the ai such that R is small.L2mCL a· f.6) with the B.balysis of aDa. n R = r . Using the weighted residual methods. we choose the functions f i in (3.] 1 =1 1 1 (3.7) and we then calculate the residual.2. B. 1 = 1 .7) i The basic step in the weighted residual (and the Ritz analysis) is to assume a solution of the form (3. 2·9 . syst-: diBerential and variatiouallnaiatiOlS Weighted Residual Methods Consider the steady-state problem (3. ••• at boundary (3.10) where the f i are linearly indepen­ dent trial functions and the ai are multipliers that are deter­ mined in the analysis.10) so as to satisfy all boundary conditions in (3.C.

differential and variational 10000nlations Galerkin method In this technique the parameters ai are determ ined from the n equations fD f.=1.=1. In the Ritz method we substitute the trial functions <p given in (3.] RITZ ANALYSIS METHOD Let n be the functional of the em-1 variational problem that is equivalent to the differential formulation given in (3.7).=1. = 0 .10) into n and generate n simul­ taneous equations for the para­ meters ai using the stationary condition on n .n [The methods can be extended to operate also on the natural boundary conditions.tinnoDS systems. ••• . R dD=O . an aa.2.n (3.14) 1 2·10 .n (3. if these are not satisfied by the trial functions.Analysis 01 C. ••• .2. 1 .12) 1 Least squares method In this technique the integral of the square of the residual is minimized with respect to the parameters ai ' a aa.2. ••• .6) and (3.

F--. 2·11 ..".19.--... • A symmetric coefficient matrix is generated.80 em 100 em Fig...Analysis of continuous systems. • Since the application of oIl = 0 generates the principle of virtual displacements. --... -------·-I ._.-==-e- R = 100 N C I-.. -~~---·-I...u . 3...B..c. • By invoking 0 II = 0 we minimize the violation of the internal equilibrium requirements and the violation of the natural b.c..... differential and variational formulations Properties • The trial functions used in the Ritz analysis need only satisfy the essential b.~---r.x. we in effect use this principle in the Ritz analysis. Bar subjected to concentrated end force... of form KU =R Example Area = 1 em 2 ( .

F.100 OU Ix=180 =0 or the principle of virtual displacements £ o - 180 (~~u)( EA ~~) dV = IT.ariali" fOllDaialiODS Here we have IT = 1 o 180 1 EA(~)2 2 ax dx - 100 u Ix = 180 and the essential boundary condition is u Ix=O = 0 Let us assume the displacements Case 1 u = a1x + a 2 i 0< x < 100 Case 2 u ~ = I1JO 100 < x < 180 We note that invoking we obtain oIT = 0 oIT = 1 o 180 (EA ~~) o(~~) dx . differeatial ad . 1 dx = 100 OU x=180 I J V ET T - 1 2·12 .Analysis of COitiDlOIS systems.

100 < x < 180 (l+x-l00)2 40 2·13 . a = 0 < x < 100 100 . differential and variational formulations Exact Solution Using integration by parts we obtain ~ ax EA (EA ~) ax x=180 = 0 ~ ax = 100 The solution is u = 1~O x .Analysis of continuous systems. 0 < x < 100 100 < x < 180 The stresses in the bar are a = 100.

0.. differential and variational formulations Performing now the Ritz analysis: Case 1 dx+ I 2 f 180 (1+ x-l00)2 40 100 I nvoking that orr = 0 116 we obtain E [0.6 .341 E x 2 128. we have the approximate solution u a = = 12C.6 E x - 0.4467 116 and 34076 a1 = ---=E=--- 128.341 E Hence.6 a 2 .682 x 2·14 .0.Analysis of continuous systems.

4 -13] -13 13 [~:] [~oo] = Hence.Analysis of continuous systems.08 80 2-15 . we now have U B = 10000 E U c 11846.2 = 23. differential and variational formulations Case 2 Here we have E n=2 J a 100 1 2 (100 u B) dx+ I 2 100 f 180 (1+ x -l00)2 40 Invoking again on = 0 we obtain E 240 [15.2 E and o = 100 0< x < 100 x> 100 o = 1846.

diUerenliai and varialiOlla1I01'1BDlaIiGlS u 15000 E 10000 E -E---..-.--.. -< .-.==_=_== "" "I~ ~ EXACT SOLUTION 1 SOLUTION 2 50 I L._.~--------r-------~X 100 CALCULATED STRESSES 2·18 .:--~ Sol ution 2 5000 ._.-_ _--r-_ _--._._ 180 -+ ---.~ " EXACT -::. -.Aulysis of COilinDmas systems..J ~..r--- ~X 100 CALCULATED DISPLACEMENTS 180 (J 100 -I=:::==-==_==_:=os:=_=_=.I~ ~~.

c.balysis of coatiDloas systms. diBerenlial ud variational fonnllatioas We note that in this last analysis e we used trial functions that do not satisfy the natural b. 2·17 . edomains A . 1 for a em. but the deriva­ tives are discontinuous at point B.e are finite elements and WE PERFORMED A FINITE ELEMENT ANALYSIS.Band B . in this problem m = 1 .variational problem we only need continuity in the (m-1)st derivatives of the func­ tions. e the trial functions themselves are continuous.

FORMULATION OF THE DISPLACEMENT-BASED FINITE ELEMENT METHOD LECTURE 3 58 MINUTES 3·1 .

4.2.2.1.3.2.1.1.2 Examples: 4. 4. detailed discussion of element matrices TEXTBOOK: Sections: 4.Formulation of the displacement-based finite element method LECTURE 3 General effective formulation of the displace­ ment-based finite element method Principle of virtual displacements Discussion of various interpolation and element matrices Physical explanation of derivations and equa­ tions Direct stiffness method Static and dynamic conditions Imposition of boundary conditions Example analysis of a nonuniform bar. 4.4 3·2 . 4. 4.

A very general formu lation -Provides the basis of almost all finite ele­ ment analyses per­ formed in practice -The formulation is really a modern appli ­ cation of the Ritz/ Gelerkin procedures discussed in lecture 2 -Consider static and dynamic conditions. but linear analysis Fig. 3·3 . 4.2.Formulation of the displacement-based finite element method FORMULATION OF THE DISPLACEMENT ­ BASED FINITE ELEMENT METHOD . General three-dimensional body.

2) The strains corresponding to U are. where uT = [u V w] (4. ~T = [E XX Eyy EZZ YXy YyZ YZX] and the stresses corresponding to are € (4.3) 3·4 .1) = fB y fB Z = fS y fS Z The displacements of the body from the unloaded configuration are denoted by U.FOl'Dlulation of the displaceDlent·based finite e1mnent lDethod The external forces are fB fB X f~ fS i FX i Fi = Fy i FZ (4.

3·5 .Formulation of the displacement-based finite element method Principle of virtual displacements where ITT = [IT If w] (4.6 ) Fig. General three-dimensional body. 4.2.

.Formulation of the displaceaenl-based filile e1eDlenl .ethod . y. z) U2V2W ••• UNVNW N] 2 . W. "T !! = [U. y.(m) (x.9) z) =~(m) (x. + -rI(m) !! (4.'1) !. x.'0) (4. y. z) =!:!.. z) 0 (4.u "" Finite element For element (m) we use: !!(m) (x.. y. V.(m) =f(m)~(m) 3·& .U 2U3 §.(m) (x.8) "T !! = [U. Un] (4.

...5) as a sum of integrations over the elements (4..u·) -- m V . strains.12) for the element displacements...(m) ~(m) (-£ )(m) - = B(m) l.12) Substitute into (4.. to (4.. _~m ( )T :.8).. j- ____---....B dV(m) j ---.. I .r__... ) !!sCm)Ti m)dScmlj m _m_JV.ll.<. and stresses.==~I______ m -___. L f...13) 3·7 .rm) "<I::: B(m)TTI(m) dv(m)j -US ------.10)...~(m) = f.c=~~-----I ~ ~ If 'iTl 1 --£ -(m) T B(m) Tc(m)B(m)dv(m)j U = v(m) l- £1 I 1 _" j [I T L l(m) T (m) !!(m) 1.. ~ . (4. using (4..:.'OI'IIalation of the displaceDlenl-based filile eleDlenl method Rewrite (4..1 y:(m) (m) T -(m) T =!!(m) ~ El.

Formulation of the displacement-based finite element . (4.!:!.p!! •• (m) _ p(m). B(m) = 1.(m)~]dV(m) MD+KU= R 3·8 .R + ~ 1 K= ( 4.!:!. -B(m) .14) R=.Ba + Rs . [1.20) ~ R =-F In dynam ic analysis we have ~B = ~ f (m)T -B(m) V(m) .- R = "'1.19) (4.18) ~ ~m) B(m)TT1(m)dV(m) ~ V(m) - R (4. ~ lm) B(m)Tc(m)B(m)dV(m) (4.22) 1.21 ) (4.16) H(m)TfB(m)dV(m) (4.17) - "'1 = "'1 -1 = -S HS (m)Tfs(m)dS(m) (4.ethod We obtain KU where =R (4. 15) ~ R ~f J m V(m).

~ rl'f \eedom u Transformed degrees of f V (restrained\ C.38) ... 4. (4.40) i I ! ! • I ~ A V Global degrees of freedom . we use ~a ~b + ~a ~b -~ ~a t!t>b = . ~a~+~a~=~-~b~-~b~ (4.10.Formulation of the displacement-based finite element method To impose the boundary conditions.39) ~=~a~+~b~+~a~+~b~ (4. Transformation to skew boundary conditions 3·9 . U= T IT Fig..-:~:e) a -sin L COS T= [ a] sin a cos a '/.

44) k » k . (4.11. 1 cos a.42) }h sin a.th 1. = k b where .Formulation of the displacement-based finite element method . 4. then the constraint equation is k U.th 1 column For the transformation on the total degrees of freedom we use J .41 ) so that i th row 1 cos a. -s ina. " ___ 3·10 . We can now also use this procedure (penalty method) Say Ui = b. T= (4.. 1 ! Mu+Ku=R where .. L Fig. •• j (4. Skew boundary condition imposed using spring element.

FormDlation of the displacement·based finite element method Example analysis z area =1 100 x y area 100 =9 80 element ® Finite elements J~ I" 100 -I 80 ~I 3·11 .

Formulation of the displacement·based finite element method Element interpolation functions 1..L) :0] 1 100 1 80 a] 1 80] av = B(m}U ay - a 3·12 . L --I Displacement and strain interpolation matrices: - H(l} = [(l-L) y 100 100 80 a] v(m} = H(m}U !:!.(2} = [ !!(l)=[ !!(2) = [ a 1 100 (1..0 I .

80 o}Y 1 1 80 Hence E [ 2.4 15.4 =240 a -13 Similarly for M '.1 100 5.4 -2. Boundary conditions must still be imposed.4 -2.!!B ' and so on.FOI'IDDlation of the displacement·based finite element method stiffness matrix . 3·13 .= (1 HEllO a l~O [-l~O l~O a U .

GENERALIZED COORDINATE FINITE ELEMENT MODELS LECTURE 4 57 MINUTES 4·1 .

4.2.5.8. the patch test TEXTBOOK: Sections: 4.2.Generalized coordinate finite element models LECTURE 4 Classification of problems: truss. and stress variables Derivation of generalized coordinate models One-. plate and shell con­ ditions: corresponding displacement.4. three.6.12.dimensional elements. 4.13. 4.17.6 Examples: 4. two-.7. detailed derivation of element matrices Lumped and consistent loading Example results Summary of the finite element solution process Solu tion errors Convergence requirements.14.18 4-2 .15. 4. 4. strain.3. plate and shell elements Example analysis of a cantilever plate.2. 4. 4.11. 4. 4. 4. 4. 4.5. 4.16. plane stress. plane strain. beam. 4. physical explana­ tions. axisymmetric.2.

14.Generalized coordinate finite eleDlent models DERIVATION OF SPECIFIC FINITE ELEMENTS • Generalized coordinate finite element models ~(m) = V(m) i In essence. we need B(m)T C(m) B(m) dV (m) H(m)T LB(m) dV (m) H(m) B (m) C (m) . • Convergence of analysis results A Across section A-A: T XX is uniform. 4. Various stress and strain conditions with illustrative examples. 4·3 . All other stress components are zero. Fig. (a) Uniaxial stress condition: frame under concentrated loads.- '- aW) = R(m) V(m) J !!S = f S (m) - HS(m)T f S(m) dS (m) etc.

v(x. 4·4 . u(x. (b) Plane stress conditions: membrane and beam under in-plane actions. 4. E zz = 0 Fig.y) are non-zero w= 0 . All other stress components are zero.Ge. TXX Fig. (e) Plane strain condition: long dam subjected to water pressure.14.raJized coordiDale finite elementlDOIIeIs Hale \ I \ - 6 I 1ZI -\ \ \ ' Tyy . TXY are uniform across the thickness.y).14. 4.

Fig. 4. j( I I I \-I I .Generalized coordinate finite element models Structure and loading are axisymmetric.14.14. (d) Axisymmetric condition: cylinder under internal pressure. I All other stress components are non-zero. (e) Plate and shell structures. 4·5 . / (before deformation) (after deformation) SHELL Fig. 4.

.. au )'''7 = ay + au au ax' a/ aw aw aoy w ••• ..7) Axisymmetric [E. = 2 E. = ax' £7 = 1 au 1 1 Z 77 0x Table 4...2 (a) Corresponding Kine­ matic and Static Variables in Various Problems..IC..Generalized coordinate finite element models Problem Bar Beam Plane stress Plane strain Axisymmetric Three-dimensional Plate Bending Displacement Components u w u.. El'l' )' "7) Plane strain (E.. 4·& . IC.OyZ. = -dx ' IC = ... E"77 )'''7 Eu ) Three-dimensional [E. v u...v u.2 (b) Corresponding Kine­ matic and Static Variables in Various Problems.... 1(77 1("7) - )'.v. E"77 Eu )'''7 )'76 Plate Bending (IC.. Problem Strain Vector ~T (E".] Beam Plane stress (E".. EJ"7 )'..) Nolallon: ...) Bar [IC.......... w w Table 4.. v u.

3 Generalized Stress-Strain Matrices for Isotropic Materials and the Problems in Table 4.2 (e) Corresponding Kine­ matic and Static Variables in Various Problems.] [Mn ] [Tn TJIJI T"'JI] [Tn TJIJI T"'JI] [Tn TJIJI T"'JI Tn] [Tn TYJI Tn T"'JI TJI' Tu ] Bar Beam Plane stress Plane strain Axisymmetric Three-dimensional Plate Bending [Mn MJIJI M"'JI] Table 4.Generalized coordinate finite element models Problem Stress Vector 1:T [T. Problem Bar Beam Plane Stress Material Matrix.£ E El 1 v v 1 E 1-1':& [o 0 1 ~. 4·7 .u.] Table 4.2.

Generalized coordinate finite element models ELEMENT DISPLACEMENT EXPANSIONS: For one-dimensional bar elements For two-dimensional elements (4. 3 4 5 (4. +y 2x+y y+y z+y xy+ .49) 4·8 ..z) =Y.z) = a.. + Y2 x + Y3Y+ Y4xy + Y5x + •..47) For plate bending elements w(x.y. w(x.y) = Y. + Ozx + ~Y + Ci4Z + ~xy + ...48) 2 For three-dimensional solid elements u (x.y. (4.

50) (4.V (bl Finite element idealization V7 Fig. T = T = T ZZ Zy ZX =0 4·9 .55) Example r Y.e.51/52) (4. 4. i.V 1 Nodal point 6 Element lp 9 0 5 0 8 CD 4 @ V7 7 Y. in general u = ~ ex (4.V la) Cantilever plate X.53/54) (4.V X. Finite element plane stress analysis.5.Generalized coordinate finite element models Hence.

. For element 2 we have U{X.. 4.~ = structure element Element nodal point no.6. 5 ...-..y) where = H(2) u -- - uT = [U 1 4·10 .Generalized coordinate finite element models LJ2.y)] (2) [ v{x... ® Fig........ 4 nodal point no... Typical two-dimensional four-node element defined in local coordinate system..= US 2 --II--.

y)] [ v(x.Generalized coordinate linite element models To establish H (2) we use: or U(X. ! and = [~ ~}!= [1 x y xy] Defining we have Q = Aa.y) where =_~l!. Hence H=iPA.1 4·11 .

HIs: 0 H zs : 0 0 : H IJ 12 1 14 Ull U U 3 U U -assemblage degrees IS zeros zeros OJ offreedom O 2x18 :H II : H ZI 0 0 (a) Element layout (b) Local-global degrees of freedom Fig. Hu : Ha : - H'ZJ = [0 U 2 I U U Us 4 3 H 17 : U 6 H 16 : U Us 7 U U 1a 9 0 0 0: HI. o : H ZJ H 21 : H:: H: 6 : 0 0: H.Generalized coordinate finite element models Hence - H = fl l4 (1+x ) ( Hy) : : a I a UJ : I ••• I : (1 +x )( 1+y) : U and VJ z t': u.7. 4. v. Pressure loading on element (m) 4·12 .. VI -element degrees of freedom HI.

ax' yy .Generalized coordinate finite element models In plane-stress conditions the element strains are where E .au .ay ax Hence where I = [~ 1 0 y'O I 0 0 0 I 0 10 I X 1 10 I I 4·13 . E _ av. Yxy .ay . _ au + av xx .

e. YIELDS: GOVERNING DIFFERENTIAL EQUATIONS OF MOTION e. Finite Element Solution Process 4·14 .Generalized coordinate finite element models ACTUAL PHYSICAL PROBLEM GEOMETRIC DOMAIN MATERIAL LOADING BOUNDARY CONDITIONS 1 MECHANICAL IDEALIZATION KINEMATICS. MATERIAL. LOADING. e. isotropic linear elastic Mooney-Rivlin rubber etc. BOUNDARY CONDITIONS. e. 4.g.!.23. prescribed displacements etc. e.g.. concentrated centrifugal etc.!!!) = .p(x) 1 FINITE ELEMENT SOLUTION CHOICE OF ELEMENTS AND SOLUTION PROCEDURES YIELDS: APPROXIMATE RESPONSE SOLUTION OF MECHANICAL IDEALIZATION Fig. truss plane stress three-dimensional Kirchhoff plate etc. (EA ax ax .. .g.g.g.

4 Finite Element Solution Errors 4·15 .2 direct time integration.4 Gauss-Seidel.4.5 10.5.5 NUMERICAL INTEGRATION IN SPACE EVALUATION OF CONSTITUTIVE RELATIONS SOLUTION OF DYNAMIC EQUILI-.4 8.5 setting-up equations and their solution Table 4.Generalized coordinate finite element models ERROR DISCRETIZATION ERROR OCCURRENCE IN use of finite element interpolations evaluation of finite element matrices using numerical integration use of nonlinear material models SECTION discussing error 4.6 9.2 9.2. BRIUM EQUATIONS SOLUTION OF FINITE ELEr1ENT EQUATIONS BY ITERATION ROUND-OFF 5.8. mode superposition 9.4 8. eigenso1utions 8. Quasi-Newton methods. 1 6. NewtonRaphson.3 6.

of elements If an incompatible element layout is used.Generalized coordinate finite element models CONVERGENCE Assume a compatible element layout is used. provided the elements contain: 1) all required rigid body modes 2) all required constant strain states ~ compatible LW layout incompatible layout ~ CD t:= no. 4·16 . then in addition every patch of elements must be able to represent the constant strain states. Then we have convergence but non-monotonic convergence. then we have monotonic convergence to the solution of the problem­ governing differential equation.

24. element must be stress­ free..Geuralized coordinate finite e1eJDeDt models I . I " > " / (a) Rigid body modes of a plane stress element .. (b) Analysis to illustrate the rigid body mode condition Fig. 4. I I I 7 " I I / i I ( 1-.-- " 'r... Use of plane stress element in analysis of cantilever 4·17 ..~_Q I I I I Rigid body translation and rotation..

" \ \ \ \ \ \ \ \ '- .. =0...J .... --~ . _-I -I I I I \ \ \ -.-~\ \ \ \ \ \ \ \ \ (' \ \ --\ \. .25 (b) Eigenvectors and eigenvalues of four-node plane stress element 4·18 ..-----..--. I I I f 'J Rigid body mode A3 =0 Flexural mode A4 =0."... ...Generalized coordinate filite elellent . I I 1.. \~ Flexural mode As = 0.76923 .. 4.1 I I I I I I I I I I I I I I I I I I I I L I I I : I ..JI I I I I Stretching mode A7 =0..\ \....adels 10 I I Young's modulus = 1. t • -l I I I I I I I Poisson's ratio" 0....25 (a) Eigenvectors and eigenvalues of four-node plane stress element \ \ \ I \ ~- .57692 Shear mode A.----"'"=""".0 r------... I r-------.30 01 I I I ------Rigid body mode A2 = 0 _1 I Rigid body mode Al = 0 -\ .57692 Fig. 4.-­ .76923 Uniform extension mode As = 1..92308 Fig.. .~ -....

30 (a) Effect of displacement incompatibility in stress prediction 0yy stress predicted by the incompatible element mesh: Point A 2 Oyy(N/m ) 1066 716 359 1303 1303 B C D E Fig. and node 18 belongs to element 6. Fig. 4. nodes 19 and 20 belong to element 5.Generalized coordiDate finite element lDodels (0 ® ·11 G) ~ 17 IT 'c. 2 constant stress a = 1000 N/cm in each element.f: / @ ~ @ @) IS a) compatible element mesh. node 17 belongs to element 4.I> ® ®-. YY b) incompatible element mesh. 4. ® 20 .30 (b) Effect of displacement incompatibility in stress prediction 4·19 . .

EXAMPLES SAP.IMPLEMENTATION or METHODS IN COMPUTER PROGRAMS. ADINA LECTURE 5 56 MINUTES 5·1 .

plate and shell analyses TEXTBOOK: Appendix A. two.3. A. 8. Sections: 1. examples SIP."pi_entation of metllods in computer prograDlS.3.2. beam.4. Example Program STAP 5·2 . the assem­ blage process Example analysis of a cantilever plate Out-of-core solution Eff&ctive nodal-point numbering Flow chart of total solution process Introduction to different effective finite elements used in one. A.I. three-dimensional. A.2.3 Examples: A. calculation of matrices. ADlRA LECTURE 5 Implementation of the finite element method The computer programs SAP and ADINA Details of allocation of nodal point degrees of freedom.

whichever are applicable. and R.o. ADINA IMPLEMENTATION OF THE FINITE ELEMENT METHOD K(m) = 1. tl kxn ~ R. B(m) C(m)B(m) dV(m) V(m)-R(m) = 1.l_pIg_talioa of _ethods in CODIpDter program. of total structure where In practice.!B ~ . 3.f. R = ~ R ( m) m-B m!. Calculation of structure matrices K . of d. M .hN N = no. 2.o. we calculate compacted element matrices. C .f. H(m)T fB(m) dV (m) -B v(m) H(m) kxN T We derived the equi­ librium equations - - B(m) . mDlples SAP. ~B' nxn nxl n = no. K = ~ K(m) ..xn The stress analysis process can be understood to consist of essentially three phases: 1. of element d. Solution of equilibrium equations. 5·3 . Evaluation of element stresses.

The structure matrices K. whichever are applicable. ADIlI The calculation of the structure matrices is performed as follows: 1.. A. mass and damping matrices.. - nodal point . and R..IDlpl_81taliol of Dlethods in toDlpuler progrw. mDlples SAP. M . C . _. - I 5·4 - .. are assembled. Possible degrees of freedom at a nodal point. l Sz :: 6 t W :: 3 Z r----y V::2 Sy:: 5 x /U:: 1 /Sx:: 4 ID(I.J) = Fig. The nodal point and element in­ formation are read and/or generated.i Degree of freedom . The element stiffness matrices.. 2. 3.1. and equivalent nodal loads are calculated.

20 ® t" t 4 _1 7 8 -7 ~ Node Temperature at bottom face = 70'C Degree of freedom number \ Fig.20 10 4 3 8 -9 5 __ t t E = l(Jl1 N/cm 2 <D .-0.. UUlples SIP..taIiOi of IIeIWs in coapler P.0. 1n this case the 10 array is given by 1 1 10 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 1 1 1 1 0 0 1 1 1 1 0 0 1 1 1 1 0 0 1 1 1 ] 0 0 1 1 1 1 0 1 1 1 1 = 1 1 1 1 5·5 . 2 x l(Jl1 N lem2 number II'" 0.. . A. ..2. . ...15 2 5 4 9 t~l1 Element E. Finite element cantilever idealization. ABilA Temperature at top face a l00"C aOem a t6 @ E· lOS N/em 2 .15 2 E = 2 x l(Jl1 N Icm2 =0.

ADIIA and then 0 0 0 0 0 0 0 1 0 3 0 5 0 7 9 11 0 0 0 0 0 10= 2 4 6 8 10 12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Also X = [ 0.0 60.0 70.0 80.0 120.0 80.0] 5·6 . 0 60.0 40.0 0.0 40.0] 80.0 120.0 100.programs.0 ZT = [0.0] 0.IJDpleDIeDtatiOD of methods in CODIpater.0 60.0 120.0 0.0 0 .0 100.0 0.0 85.0 0.0 85.0 100.0 40.0 85.0 T 0.0 0.0 70.0 0.0] T Y = [ 0. examples SAP.0 0.0 0.0 = [70.0 TT 0.

. ..1. . examples SAP. . oJ I 4 0 0 0 0 1 2 = [3 4 0 0 0 0 1 2] 5·7 .Implementation of methods in computer programs. ..2. material property set: 1 Element 2: node numbers: 6325· material property set: 1 I . material property set: 2 Element 4: node numbers: 9658.4. . . . material property set: 2 CORRESPONDING COLUMN AND ROW NUMBERS For compacted matrix For !1 LMT 1 2 3 4 5 6 7 8 . ADINA For the elements we have Element 1: node numbers: 5. Element 3: node numbers: 8547· .

Storage scheme used for a typical stiffness matrix. We have for element 2.Implementation of methods in computer programs. k l l k 12 mK =3 0 k 23 k 33 " ·1 k 14 SkYline . L M = [5 T 6 0 0 0 0 3 4] for element 3.3. A. we can obtain the LM arrays that correspond to the elements 2.. (a) Actual stiffness matrix A(21) stores k S8 A(l) A(3) A(91 A(S) A(8) A(4) A(7) 1 A(2) 2 A(15t 4 A= A(6) A(lll A(14) 6 10 12 16 18 A(16) A(lO) A(13) A(12) A(17) (b) Array A storing elements of K.~ kn 0 k 34 k 44 o" o k 45 k ss J " 0 0 0 0 0 0 k 36 k 46 ------m6 =3 '0 0 '0" 0 K= k S6 Symmetric k 66 Fig.3. ADINA Similarly. examples SAP. T L M = [9 10 3 4 1 2 7 8] and for element 4. and 4. LM T = [11 12 5 6 3 4 9 10] . 22 5·8 .

10. I X0 X0 XIX XiX XiX BLOCK 1 BLOCK 2 ~---~ I X ~_BLOCK 4 ELEMENTS IN DECOMPOSED STIFFNESS MATRIX Fig.. 5·9 .. Typical element pattern in a stiffness matrix using block storage. eDIIIples SAP. ._pl• •taiioo of lDeIJaods in COIDpater prograJDS. COLUMN HEIGHTS I ELEMENT 0= ZERO ELEMENT x = NONZERO X 0 o 0 xix XIX XIX X I SYMMETRIC I 10 0 0 10 0 0 0:0 0:0 x 010 0 x 0 010 0 0 0 0 X 0 0 0 X 10 0 0 xxlxXIO I xix IX XiX X X l XIX IX ELEMENTS IN ORIGINAL STIFFNESS MATRIX Fig. 10. Typical element pattern in a stiffness matrix using block storage. ADINA .

1 2 ~. examples SAP. A. mk + 1 = 16. Fig. mk + 1 = 46. 1 14 2 3 15 4 5 16 6 7 17 8 9 18 10 11 19 12 13 20 ~ 21 22 23 24 25 26 27 28 29 30 31 32 33 (a) Bad nodal point numbering. 5·10 . ABilA ~ . Bad and good nodal point numbering for finite element assemblage. 4 6 9 7 11 12 14 16 17 19 21 22 24 26 27 29 31 32 . 3 5 8 10 13 15 18 20 23 5 2283033 (b) Good nodal point numbering.4.IIIlpl• •tation of methods in computer programs.

DIMENSIONAL ELEMENT ! RING ELEMENT . p .... . A... examples SAP..Implementation of methods in computer programs. x z 2 3 y Fig..43. Two-dimensional plane stress. ADINA z -'-------4 I ONE .. 5-12 .42..-. A. Truss element p.... 13.... y Fig. 12.. . plane strain and axisymmetric elements.

~ x and thick shell element p.. A. Three-dimensional solid .Fig. examples SAP. 5·13 . 15. ADINA y 2 ---5 ~-----.. Three-dimensional beam element p A..-. 14.Implementation of metbods in computer programs.... z y Fig.44...45.

16. __e_ ---L~--- • -- • • 3-16 NODES y x TRANSITION ELEMENT Fig. ADINA -.46. 5·14 .Implementation of methods in computer programs. A. Thin shell element (variable-number-nodes) p. examples SAP.--.

FOBMULATION AND CALCULATION OF ISOPABAMETBIC MODELS LECTURE 6 57 MINUTES 6·1 .

2.1.16. plane-stress. 5.8. 5. 5. displacement and strain interpolation matrices.1 Examples: 5. 5. 5.3. 5.12. 5.3.7. 5. the Jacobian transformation Various examples: shifting of internal nodes to achieve stress singularities for fracture me­ chanics analysis TEXTBOOK: Sections: 5. 5.1.6. 5.14. 5.1. 5. axisymmetric and three-dimensional elements Variable-number-nodes elements.10. 5. 5.4.11.5. 5.3. 5. plane-strain.3.5. 5.9.15. 5.2. curved ele­ ments Derivation of interpolations. 5.13. 5.FOl'Dlolation and calculation of isoparmetric models LECTURE 6 Formulation and calculation of isoparametric continuum elements Truss.17 6·2 .

u. I I i=l Interpolate Displacements u= i 1: N =1 h. I I v= L i == 1 N h. . I I N = number of nodes &·3 . y= I I x=L L i N h. I I i=l =1 z=L N h. I I w= L i N =1 h.v.w.FOI'DlDlatiOl ud calculation of isopariUHbic models FORMULATION AND CALCULATION OF ISO­ PARAMETRIC FINITE ELEMENTS interpolation matrices and element matrices -We considered earlier (lecture 4) generalized coordinate finite element models -We now want to discuss a more general approach to deriving the required isoparametric elements lsoparametric Elements Basic Concept: (Continuum Elements) Interpolate Geometry N h. x. y. z. .

5.2. Some typical continuum elements 6·4 .Formulation and calculation of isoparametric models 1/0 Element 2/0 Elements Truss Plane stress Continuum Plane strain Elements Axisymmetric Analysis Three-dimensional Thick Shell 3/0 Elements (a) Truss and cable elements (b) Two-dimensional elements Fig.

2 units long 6·5 .2. 5.FOI'Ilalation and calcalatioD 01 isoparametric models (c) Three-dimensional elements Fig. Some typical continuum elements Consider special geometries first: ~~==-=l======~I=-==r======r ~1 == Truss.

S Ill( 1 ~ ll( 1 ~ J 1 r 1 2/D element.utioa and calculation of isoparalDebic lDodels . _ 1 h1 = %(1 + r) 2 -r . 2x2 units Similarly 3/D element 2x2x2 units (r-s-taxes) 1 ..0 ~~ -+-r .D Element 2 Nodes: -11.F .

r)(1...z(1.:.Y.r)( 1 + 5) 2 h3 =%(1...r) .Formulation ud calculation 01 isoparUletric lIodeis 1..::-:::.. h2 = Y.0 Element 4 Nodes: /-r-r----+~~-r h1 =~(1 + r)(1 + 5) 3 4 Similarly h =%(1..-... -..z(1.5) h4 = %(1 + r)(1-s) 6-7 ...0 2 e_----.:::::..::::=-3 -..r 2 ) 1 2 .

..4..--+-+--+~_ ._r 3 Then obtain h 1 and h tfF--...r)(1 + s) 2 -%h S 3 4 6·8 ...-~-_-_-~-r~~_- 2 : -----.. 1 !1..Formulation and calculation of isoparametric models Construction of S node element (2 dimensional) first obtain h S: . 1 -+--+-I--I--I---I--------I--..1L. h = %(1.0 h 1 = %(1 + r)(1 + s) -%h S Sim..

.q----\ \ \ s=o ---.5.~hs -~ her -~ h<j ~(l-r)(l+s) ~(l-r) (1 -s) ~(1 +r) (l-s) ~(1 . Interpolation functions of four to nine variable-number-nodes two-dimensional element.r 2 ) (1 -s) -1 h<j -th" -th<f hs i (1 - s2) (1 + r) h~:: ( 1. Include only if node i is defined i =5 h. 5.r) ~(1 .h 6 . 8 r 3 \ \ r =-1 r =0 r = +1 x (a) Four to 9 variable-number-nodes two-dimensional element Fig. i =6 i = 7 i = 8 I:: r = = = = = = = ~(l+r)(l+s) -~hs -~hs ·1· .S'") (b) Interpolation functions Fig. h2 h3 h.r"") (1.r 2 ) (1 +s) -..Formulation and calculation of isoparametric models y \ 6 ---.5.. 5. Interpolation functions of four to nine variable-number-nodes two-dimensional element: 6·9 .S2) (1 . -~h6 -~h7 -ih q -~hq hs = -ihq 'h 6 h7 ~ (1 ..

' x==r y == s z == t EXAMPLE 4 node 2 dim.Fonnulation and calculation of isoparametric models Having obtained the hi we can construct the matrices Hand !!: .The elements of B are the derivatives of thehi (or zero).The elements of -H are the· h·I (or zero) . element 6·10 . Because for the 2x2x2 elements ~ we can use =~ 1:.

. as ah at' 1 1 ah 0 3h 4 as ah 4 I 4 v1 u 2 1 as ar vB v4 We note again r==x s=y GENERAL ELEMENTS s r = +1 s = +1 Y.4 _r • 6·11 ..v r ..t ..Formulation and calculation of isoparametric models ah E ar r E SS [Y 1 0 ah ar ah 4 0 u 1 rs l 0 ah as \..

s .!!. Using (5. t ..25) Aside: cannot use --- a a ar + . for the integration thus use dv = det J dr ds dt 6·12 .25) we can find the matrix of general elements The !:! and J! matrices are a function of r. ax ar ax a J . but [ : ] = [:: as or as a ax :]l~] as ay (in general) (5.J.Formulation and calculation of isoparametric models Displacement and geometry interpolation as before..=ar a _ 1 a a-x.ar .

. +--_1< '0' I I=> 3 J ._----+----...-------...FOI'Dlalation ud calculation of isoparmebic .9." · r +- 3... 5. cDG-W\ o 1 2 = 1 213 &-13 ...... X .... Element 2 2.. '4"""1--1-----------t~1 ... Some two-dimensional elements Element 1 z.odels Fig.1+ . 6 em.

c : '3.Formulation and calculation of isoparametric models Element 3 \c.~ 't' ... (1 +5)] (3+r) 3 r=-I Natural space 3 • .L.. I 6·14 .W'I -+---~~. -"'------1431 I• 2 c l"l1 -. 1c....23.I...1V\ 2.-. . Quarter-point one­ dimensional element.. 5... .. L/4 I -I' Actual physical space Fig.-.

9 1 1 L x =-4(1+r ) 2 J - = [!:.Jf.1 We note /x 1 singularity at X = 0 ! 6·15 . x.LJ 2 2 and or Since r = 2. + !'.Formulation and calculation of isoparametric models Here we have x=L: i =1 hence 3 h ..

k x x 6·16 .J.= !:J a··k -IJ k I.Formulation and calculation of isoparaDlebic Dlodels Numerical Integration Gauss Integration Newton-Cotes Formulas K ' " IJ F·· .

FORMULATION OF STRUCTURAL ELEMENTS LECTURE 7 52 MINUTES 7·1 .

23. 5.1 Examples: 5. plate elements Discussion of general variable-number-nodes elements Transition elements between structural and con­ tinuum elements Low.5.26.20.21. 5.4.25.24.6. 5.27 7·2 .2.Formulation of structural elements LECTURE 7 Formulation and calculation of isoparametric structural elements Beam. 5. 5. 5.versus high-order elements TEXTBOOK: Sections: 5. 5. plate and shell elements Formulation using Mindlin plate theory and uni­ fied geneJ. 5.22.4."al continuum formulation Assumptions used including shear deformations Demonstrative examples: two-dimensional beam. 5.2. 5.1.

use kinematic constraints for particles on sections originallv normal to the mid­ surface " particles remain on a straight line during deformation" e. but -. shell 7·3 .. beam e.exclude the stress components not applicable -..g.FOI'IIDlati.g. slnclDrai e1U11DIs FORMULATION OF STRUCTURAL ELEMENTS • beam.. plate and shell elements • isoparametric approach for interpolations Strength of Materials Approach • straight beam elements use beam theory including shear effects • plate elements use plate theory including shear effects (ReissnerIMindlin) Continuum Approach Use the general principle of virtlial displacements.

dw dx x _ dw -0 .O . Beam section Boundary conditions between beam elements Deformation of cross-section wi x-0 = wi x+0 ./ Neutral axis WI Beam section x.29. Beam deformation mechanisms 7·4 .dx x +0 a) Beam deformations excluding shear effect Fig.29.. 5.Formulation of structural elements Neutral axis x .Wi x+ O Deformation of cross-section Boundary conditions between beam elements b) Beam deformations including shear effect Fig. Beam deformation mechanisms . 5.

S) o(~~ .49) _ (L J pw dx o -L o L m S dx (5.48) (5.51) 7·5 .Formulation of structural elements We use dw S=--y dx (5.S) dx -i L p oW dx o -io L m oS dx = 0 (5.50) L + GAkJ o (~~ .

30.30.q (Interpolation functions are given in Fig. G = shear modulus k = §. 5. 5.4) Fig. ' A = ab . 0i ={3i ' i=1.. three.. Formulation of two­ dimensional beam element (b) Two.Formulation of structural elements (a) Beam with applied loading E = Young's modulus. I = ab 6 12 3 Fig... 5. Formulation of two­ dimensional beam element 7·6 .and four-node models..

... ar ] (5. a dr' .. T W q 8..Formulation of structural elements The interpolations are now q W q 1 1 = ~ h.54) !!w = J~. 1 1 (5..53) ~ =B U dX ~- dW dX = 1-/ . OJ hqJ ~ = [0 and (5.O. f..J i h.w.' BU' Where Q. = [w.:. dh q L 7·7 .J i B = =. ~ = [h.52) w = 1-/-' H U' B = .sH U (5. 8qJ hq 0 0 h..e... 0 ] _ _. L.:. :> 0.J 1[:~l . ~ L. =...55) dh.

57) Considering the order of inter­ polations required.reduced numerical integration 7-8 .use parabolic (or higher-order) elements .Formulation of structural elements So that K = E1 -1 f 1 T ~ ~ det J dr T + GAk t -1 p (~-tla) (~-~)det J dr (5. discrete Kirchhoff theory .60) Hence . we study ex.56) and R= f -1 -1 ~ det J dr +/ ~ m det J dr (5. = GAk IT (5.

t) = k=l L hk Q.Zk + iL k=l q q a k hk Q. Three-dimensional more general beam element Here we use q Q.V~y k=l q q +~ ' b h Q.s.Formulation of structural elements Fig. 5..z(r.Yk +i L akh k Q.y(r.t) = L k=l q hk Q.V k 2 LJ k k sy k=l Q.V k 2 L.k k sx k=l q (5.V~Z + ~2 ' " LJ k=l k bkhk £V sz 7·9 .61) + ~ ' " b h Q.33.s.

x v (r.s.s.s.t)=L: k=l (5.s.62) 1z.63) 7·10 .E k=l q bkh k V~x q t q v(r.s.t) = ly _ 0y w (r.0z and u(r.s.t) = (5.t) = 1 0 x. t) = L: hku k +"2 L: akh k Vtx k=l k=l q t q k + t .Formulation of structural elements So that u (r.t)=L: hkv k +2 k=l k=l L q +tL: k=l q w(r.

.. a Gk Ynl. where ~ q ~!4~ k=l (5.:. vk = ~ e .Formulation of structural elements Finally.. we express the vectors V~ and V~ in terms of rotations about the Cartesian axes x. = Ynl.77) 7-11 . y .65) k ex ~ e = k ey k ez (5.67) u T= k k k [Uk vk wk ex ey ez ] (5. z . (5...66) We can now find £nn Yni.68) and then also have T nn Tn~ E a a £ nn Yn~ = a 0 Gk a TnI'..s where x ak 'is (5.

36...Sy (5.. dy dy (5.78) Fig.79) _ Y xy dS X dS y dX Y yz = dW dy . -. Deformation mechanisms in analysis of plate including shear deformations Hence E XX E dl\ dX = yy z dS _-.1.---- and w=w(x.80) Y zx dW .Formulation of structural elements .+ Sx dX 7·12 . 5.y) (5.

82) aw + B ax x The total potential for the element is: II=L 1 2 dz dA xy +~ 2 A -h/2 f fh/\yyZ Yzx] ~yzJ ~zx A dx dA -fw P dA (5.Formulation of structural elements and LXX 1 v a a l-v 2 L yy = z_E_ v 2 1 l-v L xy a a (5.83) 7·13 .81) aw ay .By E = 2(1+v) L ZX L yz (5.

Formulation of structural elements or performing the integration through the thickness IT = t iT A .-J.<dA + t // f.ay as . A P dA y dA -I: A (5.s 7·14 = 2{1+v) [ Ehk 0 (5.86) as x _ ~ ay ax 1 v 1 0 0 1-v C ~ =. . Eh 3 12(l-v 2 ) v 0 0 2 1 f. y = aw + s ax x (5.<q..87) .84) where K = _ .

i=l Y=~h.w. -fw A p dA =0 (5.Formulation of structural elements Using the condition c5TI= 0 we obtain the principle of virtual displacements for the plate element. LJ 1 1 .y. LJ i=l q 1 1 S y and = ~ h.=1 7·15 .89) q x q = LJ 1 1 ~h.x.88) We use the interpolations q w=~h. ei LJ i=l 1 x (5.

5.90) 7·16 .38.. 9 ..node shell element For shell elements we proceed as in the formulation of the general beam elements.Formulation of structural elements s Mid-surface \.-~-----t~ r Fig.. (5.

91) (5...93a) then V .94) 7·17 ..:.point k k °V 1 - = (e x Ov k ) / Ie yx-°Vkl -y -n n (5.k + °V -1 Sk (5.n k = .°Vk ~ k O'.Formulation of structural elements Therefore.92) To express we define Y~ in terms of rotations at the nodal. where (5.

100) 1 v a a Jl a a a 1-v -2- a a a a 1-v -2- a a a a a 1-v 2 1 ~h=~h T ( 1_~2 ) !2sh symmetric (5.101) 16· node parent element with cubic interpolation I- 2 5 -I • • • 2 • Some derived elements: 64£>-[> 000 o \'. we need to recognize the use of the following stress-strain law l = ~h ~ (5.\ Variable .number .' .Finally.nodes shell element 7·18 .

Formnlalion of structural elements a) Shell intersections • b) Solid to shell intersection Fig. Use of shell transition elements 7·19 . 5.39.

NUMERICAL INTEGRATIONS. MODELING CONSIDERATIONS LECTURE 8 47 MINUTES 8·1 .

5.36. modeling considerations LECTURE 8 Evaluation of isoparametric element matrices Numercial integrations. 5. 5. 5.34. two-. Newton-Cotes formulas Basic concepts used and actual numerical opera­ tions performed Practical considerations Required order of integration. 5.37.7.2.35. 5. 5. 5.8.2. Gauss.28. 5. 5.7.38.3 Examples: 5. TEXTBOOK: Sections: 5. 5. and plate and shell structures Modeling considerations using the elements. 5.7.1.4.29.30. three-dimensional analysis.8.1. 5. 5.33.32.Numerical integrations.8. 5. 5.31.3. 5. simple examples Calculation of stresses Recommended elements and integration orders for one-.39 8·2 .7.

HS fS dS dV f -s = S R T Rr= f ~T !.Numerical integrations.30) (4. SOME MODELING CONSIDERATIONS • Newton-Cotes formu las • Gauss integration • Practical considerations • Choice of elements We had K = f BT C B dV V .29) (4.31 ) (4.-M = J p HT H dV V -~ (4. modeling considerations NUMERICAL INTEGRATION. 33) 8·3 .32) R= f HT fB dV V.r V (4..

modeling considerations In isoparametric finite element analysis we have: -the displacement interpolation matrix t:! (r. -1 -1 8·4 .s.s.t) -the strain-displacement interpolation matrix ~ (r . we now have.4 dr ds dt Hence. for example in two-dimensional analysis: !\$ = f f~ +1 +1 T ~ ~ det A dr ds -1 -1 M= ff +1 +1 p tl T tt det J dr ds etc.t) Where r.t vary from -1 to +1..s.Numerical integrations. Hence we need to use: dV = det..

e..577 5 =± 0.Numerical integrations. .g..: K=L~a. .-IJ detJ·· B·· C B·· ~J a.·. r r = ±O. 4J lJ -lJ 1 J where i.F .775 r= 0 2x2 .577 = ±O.point integration 8·5 . = weight coefficients IJ F·· -IJ .775 5=0 5 \ \ r = ±O. modeling considerations The evaluation of the integrals is carried out effectively using numerical integration. j denote the integration points = -IJ T .

modeling coDSideratiODS z L.-- ---.Numerical integrations. I I a a+b -2- b x .~ Y 3x3 .point integration Consider one-dimensional integration and the concept of an interpolating polynomial.-8 . 1st order interpolating ---"'--polynomial in x.

Numerical integrations.+R n LJ a . and { J F(r)dr=(b-a)~C.Cotes constants interpolating polynomial is of order n..123) n = number of intervals Ci n = Newton . In Newton ..=0 1 1 b n (5. a a+b 2 b etc.nF. modeling considerations I actual function F 2 nd order interpolating ~~~~polynomial in x .. 8·7 .Cotes integration we use sampling points at equal distances.

.. Newton-Cotes numbers and error estimates.• ~ are variables. 1 +0.. 8·8 .Numerical integrations. modeling considerations Number of Intervals n q "2 1 q T 1 Cn 2 cn 3 q C· 5 Cn 6 Upper Bound on Error R..124) where both the weights a 1 •.1. •a n and the sampling points r1 •. as a Function of the Derivative of F 10-I(b-a}lF"(r) 2 3 4 5 6 "8 19 288 41 840 1 6" 1 7 4 6" 3 "8 1 6" 3 10-3(b-a)5PV(r) 1 "8 "8 32 90 1O-3(b-a)5F'V(r) 90 32 90 75 288 216 840 12 90 50 US 27 840 50 288 272 840 ill 7 90 75 10-6(b-a)7FVI(r) 19 288 216 840 10-6(b-a)7Fv'(r) 27 840 41 840 lO-'(b-a)'FVIU(r) Table 5. F(r )+R n n n (5. In Gauss numerical integration we use f a b F(r)dr" U F(r 1 ) + u2 F(r2 ) + ••. The interp(llating polynomial is now of order 2n -1 .

Sampling points and weights in Gauss-Legendre numerical integration. (I5 zeros) ±0.77459 66692 41483 0.36076 0.a r.55555 0. -22 1 2 I and the ri and eli can be tabulated as in Table 5.47862 0.66120 93864 66265 ±0.86113 ±0.57735 02691 89626 ±0.2.34785 0.23861 91860 83197 0. (I5 zeros) 0‫סס‬oo 1.2.17132 0.0‫סס‬oo 0‫סס‬oo /X. Now let.65214 0.0‫סס‬oo 63115 94053 10435 84856 98459 38664 93101 05683 0‫סס‬oo 0‫סס‬oo ±0.93246 95142 03152 ±0.56888 0.88888 0.33998 ±0. 2. 8·9 .a el.46791 55555 88888 48451 51548 68850 86704 88888 44923 15730 39345 55556 88889 37454 62546 56189 99366 88889 79170 48139 72691 Table 5.53846 0.23692 0. Then the actual sampling point and weight for the interval a to bare a + b + b .Numerical integrations.0‫סס‬oo 0‫סס‬oo 0‫סס‬oo ±0.90617 ±0. and b . modeling considerations n 1 2 3 4 5 6 rj O. ri be a sampling point and eli be the corresponding weight for the interval -1 to +1.

a·IJ• = a. modeling considerations In two.k F(r.j. where a. J I J are the integration weights for one-dimensional integration.s) dr ds = I: 1 "1 f -1 +1 F(ri's) ds (5. ff1 +1 +1 +1 -1 -1 -1 i. Q. .·a. Q .t}drdsdt = LJ 1 J ~a.j ..s.and three-dimensional analysis we use f -1 or +1 f -1 +1 F(r. and a.133 ) and a··k = a.s.I a.·a. Similarly.131) f f +1 -1 -1 +1 F(r..Numerical integrations.tk) 1 J (5..kF(r.s)drds= I: i . IJ I J k 8·10 ...132 ) and corresponding to (5../(ri'sj) (5.113).

Considering the evaluation of the element matrices. the rank of the element stiffness matrix is not smaller than evaluated exactly) . the integration is frequently not performed exactly. and (2) the element contains the required constant strain states.Numerical integrations.e.The integration order required to evaluate a specific element matrix exactly can be evaluated by study­ ing the function f to be integrated. b) mass matrix evaluation: the total element mass must be included. modeling considerations Practical use of numerical integration . • In practice. c) force vector evaluations: the total loads must be in­ cluded. we note the following requirements: a) stiffness matrix evaluation: (1) the element matrix does not contain any spurious zero energy modes (i. 8·11 .. but the· integration order must be high enough.

136) • stresses can be calculated at any point of the element. modeling considerations Demonstrative example 2x2 Gauss integration "absurd" results 3x3 Gauss integration correct results Fig.46. • stresses are.node plane stress element supported at B by a spring. 5. 8-12 . discon­ tinuous across element boundaries.Numerical integrations. in general. 8 . Stress calculations (5.

5.47.. . -of A 8 .c· -p CD 1> 3 Coft'1. Fig. / = a.m. 8·13 .Numerical integrations. E 2 = 3xl0 7 N/cm \) = 0. modeling considerations thickness = 1 cm A 1~[ I :. <3> e. '100 N!Crrt'l... (a) Cantilever subjected to bending moment and finite element solutions..3 p = 300 N 3c. Predicted longitudinal stress distributions in analysis of cantilever...

5. @ B C. I . modeling coDSideratioDS ." ® B Co <D (b) Cantilever subjected to tip-shear force and finite element solutions Fig.Numerical integratiODS. Some modeling considerations We need • a qualitative knowledge of the response to be predicted • a thorough knowledge of the principles of mechanics and the finite element procedures available • parabolic/undistorted elements usually most effective 8-14 .3 lOON s_ " 8 Co " & <D 174+ /lA/e-t'- Co A A 8 c ~I ".00 "'Ie-.~ . '? A .47. . Predicted longitudinal stress distributions in analysis of cantilever. <D a~ 4l "? v P = = 0.

~ THREE-DIMENSIONAL 20-node 3-D BEAM 3-node or 4-node -/ PLATE 9-node SHELL 9-node or 16-node L7 ~~ 8·15 .6 Elements usually effective in analysis.. modeling considerations Table 5....Numerical integrations. TYPE OF PROBLEM TRUSS OR CABLE TWO-DIMENSIONAL PLANE STRESS PLANE STRAIN AXISYMMETRIC ELEMENT 2-node 8-node or 9-node D D -= .

node to 4 . ! c) 8 . . modeling considerations I 1 4/'1ode elEJmerrt S node e 1er1l(1'It. i I g I'\oole ~kl'7ll'"t ~ J I a) 4 .node element layout transition region Fig.49. U.."~"t.node element transition region 119 8 4..node el .node to 8 . 5. ~I\"t A llA ~ c Constraint equations: C U. uA = (uC + uB)/2 vA = (vC + vB)/2 b) 4 .node to finer 8 . Some transitions with compatible element layouts 8·16 ..I\oc:(e B ele.node element transition /.Numerical integrations.s 4 node e Iem tnt"" VA 'Ve- A 4.

SOLUTION OF FINITE ELEMENT EQUILIBRIUM EQUATIONS IN STATIC ANALYSIS LECTURE 9 60 MINUTES 9·1 .

8.Solution of IiDile e1eDleul equilihrilll equations iu slatic aaalysis LECTURE 9 Solution of finite element equations in static analysis Basic Gauss elimination Static condensation Substructuring Multi-level substructuring Frontal solution t l> t T .7.5. 8.3. 8. 8.4.10 9·2 .1.1.2. 8.8.2.2.9.2. 8.6.3.2.2. Examples: 8.4.1. 8. 8. 8. 8. 8. 8.factorization (column reduction scheme) as used in SAP and ADINA Cholesky factorization Out-of-core solution of large systems Demonstration of basic techniques using simple examples Physical interpretation of the basic operations used TEXTBOOK: Sections: 8. 8.

frontal solution .L Q.substructuring .g.2) 0 0 0 -4 . e. (8.Crout . -4 6 -4 . .bT factorization . 0 .. Gauss-8eidel • Direet methods these are basically variations of Gauss elimination . -4 9·3 .column reduction (skyline) solver THE BASIC GAUSS ELIMINATION PROCEDURE Consider the Gauss elimination solution of 5 -4 6 .Cholesky decomposition . U 2 U 3 U 4 = . -4 5 0 U.SoJutiOD of filile e1emenl equilihrillD equations in slatic analysis SOLUTION OF EQUILIBRIUM EQUATIONS IN STATIC ANALYSIS • Iterative methods.static condensation .

4) 0: _ 20 I 7 I 9·4 . 5 olI l! 5 o I_~ : I I I r------------5 16 -4 1 o 1 (8.Solation of finite element eqailihriUl equations in static analysis STEP 1: Subtract a multiple of equation 1 from equations 2 and 3 to obtain zero elements in the first column of K.3) 5 5 29 -4 5 o: -4 5 -4 o 16 o = o o o 5-5 0: I I 14 r-------~ _20 7 7 65 14 (8.

..-.Solation of finite element eqailillriUl equations in static analysis STEP 3: 5 0 0 0 -4 1 0 U 1 U 2 0 S 14 0 0 -s 15 7 0 I I I I 16 1 1 r--- -T 5 20 .( 0 )"5 _ 8 U1 =----~---. =5 12 1 .(1)15 .----2 14 _ 13 -S (8. : u4 .16) U3 .5) U 3 "7 U 4 7 "6 "6 Now solve for the unknowns U3 ' U2 and U..(1) U4 5 U =--------:.8 (8.(."5 5 o- 9·5 .6) 5 19 36 7 (-4) 35 .

!Sec ~ K Example tee 5 I I I r:~a 1 0 U 1 U 2 U 3 U 4 0 -4 6 -4 1 -4 5 ---+------------4 1 0 6 -4 1 -4 = 1 0 0 14 5 16 -5 29 5 -4 1 -4 5 so that ~c '---y----' ~a -a.~ilC . a Hence (8.!Sec.~c .!Sea -1) -1 !!a = Ba .Solution of finite element eqDilihriDlD equations in slatic analysis STATIC CONDENSATION Partition matrices into ~a ~-ac] [!!a] [ Ba] [ ..- [1/5] [-4 1 0] -4 1 0 1.3) .- - Kaa 9·8 = -4 1 '-- - and we have obtained the 3x3 unreduced matrix in (8..!Sea ~-e c !!c = Be Hence (8.28) and ( ---------­ aa ~a ..30) gives ~ K = 16 -5 1 ~ 6 -4 6 -4 -4 5 .

SoIltiOl of finite elemelt eqlilihrilDl equations in static aualysis 5 -4 0 VI -4 1 6 -4 -4 1 U2 U3 U4 6 -4 -4 5 0 :1 :1 14 "5 -!§ 5 -!§ 5 29 U2 -4 5 -5 U3 V4 0 0 -4 Fig. 8. 9-7 .1 Physical systems considered in the Gauss elimination solution of the simply supported beam.

.Solutiou of finite element eqDilihriom equations in static analysis SUBSTRUCTUR ING • We use static condensation on the internal degrees of freedom of a substructure • the result is a new stiffness matrix of the substructure involving boundary degrees of freedom only .3. Truss element with linearly varying area. 17 -20 48 ~~ 6L -20 [ 3 -28 9·8 .--. We have for the element.. 8..-~-o--o L -?-? -6 l e--c>---n 50x50 32x32 Example Fig...

9 EA. oIliDile e1emeal eqailihrilDl eqaaliODS ia stalic aaalysis First rearrange the equations 6"L EA.SoIali. 6L Ir 7 3 25 3] -[-20] [lJ[-20 -28 48 or ll. [ 1 L -1 and 9·9 . [ '7 3 -20 Static condensation of U2 gives EA.

U. '-10 .. I L .Solution of fiDile elemeul equilibrilll equati. • Us (b) Second-level substructure _I I U. \ U Ur. Fig. ---I I 1- u2 u3 U.-u 3 (a) First-level substructure ---I I U. I I I I I 1- Us.I. ~I . in slatic aDalysis Multi-level Substructuring A I' 'I' L 2A -\&-o=2:E~f' · 'n-~ u. _I I I I 1- u3 Us I 1- I U.. I -. . 8. '~ Us Ug .5.Rr. L 4A. Rs L SA.1 I U2 U3 - Bar with linearly varying area -I U. • . U6 U7 . Analysis of bar using substructuring. I I I ug I I I 1- ug (c) Third-level substructure and actual structure. I 16A.

Frontal solution of plane stress finite element idealization. if the element numbering in the wave front solution corresponds to the nodal point numbering in the skyline solution. 9·11 .8. • Same number of operations are performed in the frontal solution as in the skyline solution. ill static analysis Frontal Solution Elementq Element q + 1 Elementq + 2 Elementq + 3 m ------m+3 Element 1 Element 4 4 Wave front for node 1 Wave front for node 2 ~I N:" Fig. • Solution is performed in the order of the element numbering . • The frontal solution consists of successive static condensation of nodal degrees of freedom.6.Solution of fiDile e1eDleul equilihrio equti.

is the basis of the skyline solution (column reduction scheme) .Solution of finite element equilibrium equations in static analysis L D LT FACTORIZATION .= -1 Example: 5 -4 6 a -4 5 -4 ~4 a 16 4 5 -4 a = -4 5 5 5 5 29 -5 1 a a -4 6 a _16 a 5 -4 -4 5 a a a -4 We note 4 L-1-1 5 = -5 4 1 5 a a a ~1 1 S- a a a o a 9·12 .Basic Step L.1 K K --1 .

:= s ...1 K:= S S := x x x x x x x x . x x x x x x x upper triangular matrix Hence or Also....2 1.Solution of finite element equilibrium equations in static analysis Proceeding in the same way -1 -1 1.. 11 11 9·13 .. because ~ is symmetric where 0:= d i ago na1 rna t r i x d ..

17) V o LT where U = (8. we use where t = L D~ SOLUTION OF EQUATIONS Using T K= L 0 L we have (8.19) (8.18) V := L-I -n-l and (8.Solution of finite eleJDent equilihriDII equations in static analysis I n the Cholesky factorization.20) 9·14 .16) L V= R (8.

Solution of finite element equilibrimn equations in static analysis COLUMN REDUCTION SCHEME 5 -4 6 1 -4 6 1 -4 5 5 4 5 14 5 ~ -4 6 5 -5 4 -4 6 5 -4 5 14 -4 5 5 -5 5 14 4 ~ 1 5 8 7 5 1 -4 5 4 -5 1 5 -7 8 5 14 15 T 15 T -4 5 9·15 .

Solation of finite element eqailihriam eqaati. in static analysis
X=NONZERO ELEMENT 0= ZERO ELEMENT
_~

COLUMN HEIGHTS

o 0 o 0 '-----,

SYMMETRIC

000 000 X 000 X o 0 000 o 0 x 0 0 o X 000 X X X X 0 X X X X X XX
X

X
ELEMENTS IN ORIGINAL STIFFNESS MATRIX
Typical element pattern in a stiffness matrix

SKYLINE

L...-_

o o

X X X X

0 0 0 0 0 X

000 000 0 0 X 0 0 X X 0 X X 0 X

X X X X X

X X X X X X X

X X
X
ELEMENTS IN DECOMPOSED STIFFNESS MATRIX
Typical element pattern in a stiffness matrix

9-16

Solution of finite element equilibrium equations in static analysis
x = NONZERO
ELEMENT 0= ZERO ELEMENT COLUMN HEIGHTS

I

-x o

0 0 0 0

xix x
XlX 0

xIx

0 x 0

I 0 10 0:0 010 010 0 x X\O

I

0:0 010 0 x 0 0 0 0 0 0

SYMMETRIC

x xix XIO xix xix Ix XlX
xIx
Ix

ELEMENTS IN ORIGINAL STIFFNESS MATRIX
Typical element pattern in a stiffness matrix using block storage.

9·17

SOLUTION OF FINITE ELEMENT EQUILIBRIUM EQUATIONS IN DYNAMIC ANALYSIS

LECTURE 10
56 MINUTES

10·1

5.2.2.3.2.4. 9. 9.4 Examples: 9.4.4. 9. 9.5. 9.3.Solotion of finite e1mnent eqoiIihrio equations in dynaDlic analysis LECTURE 10 Solution of dynamic response by direct integration Basic concepts used Explicit and implicit techniques Implementation of methods Detailed discussion of central difference and Newmark methods Stability and accuracy considerations Integration errors Modeling of structural vibration and wave propa­ gation problems Selection of element and time step sizes I Recommendations on the use of the methods in practice TEXTBOOK: Sections: 9.1.2. 9.2.2.4. 9.1.2.4.1.12 10·2 . 9. 9.4.3. 9.2. 9.1. 9. 9. 9.

Solution of finite element equilihriDl equations in dyDalDic ualysis

DIRECT INTEGRATION SOLUTION OF EQUILIBRIUM EQUATIONS IN DYNAMIC ANALYSIS MU+CU+KU=R

-- -- --

• selection of solution time step (b. t) • some modeling considerations

• explicit, implicit integration • computational considerations

Equilibrium equations in dynamic analysis

MU + C U + K U = R
or

(9.1)

10·3

Solution of finite elelleul equilihrilll equatiolS in dynaJDic analysis

time

-Fig. 1. Evaluation of externally applied nodal point load vector tR at time t.

time

THE CENTRAL DIFFERENCE METHOD (COM)

to = _l_(_t-tlt u + t+tlt U) 2tlt -

(9.4)

an explicit integration scheme

10·4

Solation of finite eleDlent eqailibrimn equations in dynanaic analysis

Combining (9.3) to (9.5) we obtain

-'-M + -'- c)t+~tu ( ~t 2 - 2~ t -

=

t - _ ~K __2_ - -u M)t R 2
~t

-(-'- - 2~t -c)t-~tu 2 M_-'~t

(9.6)
where we note

! t!! =(~!(mT!!
=

~ (l5-(m) t lL)

=

~t£(m)

Computational considerations • to start the solution. use

(9.7)
• in practice. mostly used with lumped mass matrix and low-order elements.

10·5

of elements and no.e e=~ for lower-order elements L'lL = smallest distance between nodes for high-order elements l'IL = (smallest distance between nodes)/(rel. use for continuum elements. stiffness factor) • method used mainly for wave propagation analysis • number of operations ex no. l'It < l'IL . Tn Tn = smallest natural period in the system hence method is conditionally stable _ in practice.Solution of finite element equilibrium equations in dynamic analysis Stability and Accuracy of COM -l'I t must be smaller than l'It e r l'It er = TI . of time steps 10·6 .

= la .28) {9.Solution of finite elelDent eqoiIibriDII eqoatiou in dynandc analysis THE NEWMARK METHOD (9. we use mostly a. 0 =~ which is the constant-average-acceleration method (Newmark's method) • method is unconditionally stable • method is used primarily for analysis of structural dynamics problems • number of operations == 2 ~n m + 2 n mt 10·7 .In practice.29J an implicit integration scheme solution is obtained using .

:: w· 1 ~1 <p.1 + w. . x.n 10·8 . . 1 1 2 = ~...Solution of finite element equilibriDII equations in dynmic analysis Accuracy considerations • time step !'1 t is chosen based on accuracy considerations only • Consider the equations ~1U+KU=R and where --1 2 K ¢. --1 Using ¢"1 K ¢ = 0.. 2 where we obtain n equations from which to solve for xi(t) (see lecture 11) . T ~1- R i=l. x.

we can study the solution of the single degree of freedom equation x+w x=r .. .. the direct step-by-step solution of r~O+KU=R corresponds to the direct step-by­ step solution of . . 2 Consider the case .. x+ w 2 x = a o· x= a 0·· x= -w 2 10·9 . x· + w.n with n - U= ~<I>.Solution 01 finite eleDlent equilibriDll equations in dynaDlic analysis Hence.x. to study the accuracy of the Newmark method.. ~-l 1 i =1 Therefore. x· 1 1 1 2 i=l..

0 Houbolt method 19... x 31----+--f-+-+----t-----t-----.~ Q. Fig. equation . 9.0 '" E 3. The dynamic load factor 10·10 .14 0.06 0. w C le E!:.0 '.8 (b) Percentage period elongations and amplitude decays.. 4t-----r----:--r--r----r-----. 11.0 0 "iii .. static 1 response 'E CtI o >­ c:: CJ 1 2 3 Fig.01~4t€ ~ 0.14 :! c '" l:! Q" '" 8.18 Fig. 2 ...06 0.0 .-----.8 (a) Percentage period elongations and amplitude decays.0 § . + 2~wx + w x = S 1 n p t .: Co '" '" :! c l:! '" C/I 0 ~ > u 7. 9..4.0 5.g C/I '" C 5 .t----'-1 2 ~ "t:J CtI o '0 o .0 .0 11.0 tf 1. ...0 "8 "0 "0 ~ '" '" 5.Solotion of finite element eqoiIihriDl equations in dynandc analysis 19. '" 1.0 15. 3.10 0..18 PE 0.0 1.-----.~ 0.. le 7. 9.0 15.10 0...

I)C ::i. .... /' _....:..?~ I..SO 1. fl../ /"7--_____ .75 3.. __.:.. of filile 81• •1 eqailihrillD eqaaliOlS in dJllillic analysis g 7T !2C 1 \ ~ '" .r • 1. 10·11 .+1::-:---+------::+-'::-:---+-----:<' :._--+_...00 : ."~ I I ..25 :..--==---/- g. ....f..oe o..00 2. _-.':.STATIC RESPONSE T 74.+-~--+---+--- c...00 TIllE Response of a single degree of freedom system.STATIc.05 D:.::::7'-' 81 r. -------t-----+---+I... ....it gi + ~ =0. 0. fs!..-1 n.-t-" . ..t . DLF ._--+-.50 DYNAMIC RESPONSE --.---~ - -----=~---':.. + " ~. .~.50 2. .+1:::---+----.05 nYNAMIC RESPONSE _ .. 00 Response of a single degree of freedom system..~~._ _---+--+1~--+--+I ::-:---+----.--t'c...-.SoIIIi.: C. 7~ I.."L .0 w ..: RESPONSE .---= __ ==_7'~--. 'JO ---+-. = 3.-. ---..-.75 2.25 2.1 • 1 ! z~~..j. .

Solution of finite element equilibrium equations in dynamic analysis

Modeling of a structural vibration problem 1) Identify the frequencies con­ tained in the loading, using a Fourier analysis if necessary. 2) Choose a finite element mesh that accurately represents all frequencies up to about four times the highest frequency w contained in the loading.
u

3) Perform the direct integration analysis. The time step /':, t for this solution should equal about 1 20 Tu,where Tu = 2n/w u ' or be smaller for stability reasons.

Modeling of a wave propagation problem If we assume that the wave length is Lw ' the total time for the wave to travel past a point is

(9.100)

where c is the wave speed. Assuming that n time steps are necessary to represent the wave, we use

(9.101 )

and the "effective length" of a finite element should be

c /':,t

(9. 102)

10·12

SoIaliOi .. filile 81• •1eqailihriDl eqaali_ in dJUlDic ualysis

SUMMARY OF STEP-BY-STEP INTEGRATIONS -INITIAL CALCULATIONS ---

1. Form linear stiffness matrix K, mass matrix M and damping matrix ~, whichever appl icable; Calculate the following constants: Newmark method: 0 > 0.50, ex. 2:. 0.25(0.5+0)2

a = , / (aAt 2 ) O a = 0/ ex.- , 4

a,=O/(aAt) as = I1t(O/ex.- 2)/2 a g = I1t(' - 0)

a3 = , / (2ex. )- ,

a 7 = -a 2

as = -a 3

Central difference method:

a, = '/2I1t

... 0 O· 0·· 2. Inltlahze !!., !!., !!. ;
For central difference method only, calculate I1t u from initial conditions: -

3. Form effective linear coefficient matrix; in implicit time integration:

in explicit time integration:

M = a~ +

a,f.

10·13

Solution of finite element equilibrium equations in dynamic analysis

4. In dynamic analysis using implicit time integration triangularize R:.
--- FOR EACH STEP --(j) Form effective load vector;

in implicit time integration:

in explicit time integration:

(ii) Solve for displacement

increments; in implicit time integration:

in explicit time integration:

10·14

ti. of filile elOl.SoI.1 equilihriDl equations in dynamic analysis Newmark Method: Central Difference Method: 10·15 .

MODE SUPERPOSITION ANALYSIS. TIME BISTORY LECTURE 11 48 MINUTES 11·1 .

1. 9. 9. 9.7.2.8.3. 9. 9.3.3.3 Examples: 9. lillie bistory LECTURE 11 Solution of dynamic response by mode superposition The basic idea of mode superposition Derivation of decoupled equations Solution with and without damping Caughey and Rayleigh damping Calculation of damping matrix for given damping ratios Selection of number of modal coordinates Errors and use of static correction Practical considerations TEXTBOOK: Sections: 9.11 11·2 . 9.10.Mode slperpClilion analysis.6.9. 9.

R = PT R (9. !(t) nxl nxn nxl P = transformation matrix ! (t ) =general ized displacements Using !L(t) on = 1:.Mode superposition analysis. !(t) (9. time history Mode Superposition Analysis Basic idea is: transform dynamic equilibrium equations into a more effective form for solution.1) we obtain ~ R(t) + where f i(t) + R!(t) ~(t) (9.30) MU + c 0 + KU= R (9. using !L = 1:.32) 11·3 .31) C fT ~ f .

and n .38) 11·4 .) 2 2 ( ul 2 ' ~) .34) Using we obtain the generalized eigenproblem.. tiDle history An effective transformation matrix f is established using the displacement solutions of the free vibration equili­ brium equations with damping neglected. M0+ K U =0 (9..Mode sDperJMlilion ualysis.37) - < W 2 n (9. ...:t:. <P " M'" .1 _.36) with the n eigensolutions (w~. (9. " T J 1== 0' i =j i .. j (9. p.. (w ' .P.n) . .

(t) (9.e.39) we can write (9.42) and the M .i ~(t) = !T !S. i.Mode superposition analysis..40) and have ¢T M¢ = I (9. at time 0 we have (9.43) The initial conditions on ~(t) are obtained using (9. time history Defining (9.41) Now using !L(t) = ! ~Jt) (9.44) 11·5 .42) we obtain equilibrium equations that correspond to the modal generalized displacements !(t) + !T ~! !(t) + r.orthonormality of ¢.

(t ) 1 1 1 i = '... (t) = r. tilDe bislory Analysis with Damping Neglected (9.e. M -1 - T a U - (9.. (9.49) 11-& . sin1 1 cos w·t 1 w.T O' t=O =-'-.t + 8.( where 2 + w.M U - Using the Duhamel integral we have =-' jtr1·(T) sinw.2.47).48) + a.(t-T)dT w. .i and And then 8i are determined from the initial conditions in (9. 1 where a. 1 1 0 (9.46) with X'I t=O = 1 X'I 1 • lj).Mode SUperpClitiOD aualysis.-1 cp. .45) i.. n individual equations of the form .x 1 t) .n (9.47) . x.

..Mode sDperp.wx + W2X = S 1 n P t 31-_ _-+_ _+--+-+-_ _+-_ _-+ -+-_ _--.-----:--.4. 9..U The error can be measured using (9..--r----. x·1 (t) where uP . 0 0 .... x + 2E. u CtI 2 .----~---_r_---.. The dynamic load factor 3 Hence we use -- uP = ~--l i =1 ~ ¢. time history 4f----....ition analysis. ...50) 11·7 ..... -0 CtI static response 0 u CtI E ~= \-0 0 >- r::::: 2 Fig. .. equation •• ...

51) and we have x.(t) + 2w. then let ~_=LriUl~) i =1 Hence n r. time history Static correction Assume that we used p modes to obtain ~p . E.52) 11·8 . we have !(t) + !T f!i(t) + fi !(t) = !T ~(t) (9. 1 1 1J (9.(t) + 1 1 1 1 w~ x.n (9. x...Mode superposition analysis. .... 1 = -1- ¢.43) If the damping is proportional -1 ---J T ¢. . T R Then and K flU fiR Analysis with Damping Included Recall. cS. C (po = 2w..(t) 1 1 = r 1 ·(t) i=l. E. .

. time history A damping matrix that satisfies the relation in (9. C - = -- a ~1 + -BK (9. 1 ~.55) example: Assume ~. 1 11·9 . are calculated from the p simultaneous equations A special case is Rayleigh damping. = 0. 1 or - ~.02 w. 1 a + Bw:- '/ 1 2w. ••• .56) where the coefficients a k ' k = . p . = 2 calculate a and We use B 2w.Mode superposition analysis.51) is obtained using the Caughey series. (9.

Mode superposition analysis. 104 .104 K Note that since 2 a + 13 w. 2 a + SW. = 1 1 2w.336 and 13 = O. = 2w. 2 i 1 11·10 . once a and 13 have been established..w 13 2w.1 and w2 ' [. 1 = - a + .2 ' we obtain two equations for a and 13: a + 4ii = 0. E.08 a + 913 = 0. we have. . [.60 The solution is a = -0. time history Using this relation for wl ' [. Thus the damping matrix to be used is C = -0. 1 1 1 for any i.336 M + 0.

x· + x. P« n .it may be important to calculate E p (t) or the static correction.!:L and then uP LJ-1 i =1 P ~¢.g..when the response lies in a few modes only. = r.when the response is to be obtained over many time in­ tervals (or the modal response can be obtained in closed form). !i!i . E. 11·11 . e. 111111 w~ with r· 1 TO xi I t = 0 "--. x. time history Response solution As in the case of no damping. (t) 1 Practical considerations mode superposition analysis is effective . earthquake engineering vibration excitation . we solve P equations x.ition analysis.Mode sDperp.1 + 2w.!:L • ITO' xi t = 0 = !i f1 .

SOLUTION METHODS FOR CALCULATIONS OF FREQUENCIES AND MODE SBAPES LECTURE 12 58 MINUTES 12·1 .

2. 12. 12.2.3.3. 12.2.' transformation methods Large eigenproblems Details of the determinant search and subspace iteration methods Selection of appropriate technique.Solution methods for calculations of frequencies and mode shapes LECTURE 12 Solution methods for finite element eigenproblems Standard and generalized eigenproblems Basic concepts of vector iteration methods.3.2. 12.3. 12. 12.1.1.2.4.1.3.3. 12. Sturm sequence methods.2. practical considerations TEXTBOOK: Sections: 12.4 12·2 . 12. 12.6 (the material in Chapter 11 is also referred to) Examples: 12.3.3. polynomial iteration techniques.1. 12. 12.

­ \ = w2 + )1 W or 2 =\-)1 12·3 ..- + )1 Ii sP. are present we can use the following procedure r or sP.= w !i! If zero freq.. proportional damping 2 r sP.Solatiu methods for calculations of frequencies and mode shapes SOLUTION METHODS FOR EIGENPROBLEMS Standard EVP: nxn r! = \ ! (\=w 2 ) Generalized EVP: !sP..- (r+)1 !i)sP..-=\!i! Quadratic EVP: Most emphasis on the generalized EVP e.= \ !i sP. earthquake engineering "Large EVP" n> 500 m> 60 p=l. . .= (w 2 + ~r)!i sP.g.3"n 1 In dynamic analysis..

A ~) In buckling analysis .cies and lIode shapes p(A) p(A) = det(K .!\$.Solation lIethods lor calcalatiou oIlreqa.!=A~! where p(A) = det (~ - A ~) p(A) 12·4 .

(~ ..Ki=. then solve using one of the many techniques available e.\ i .~ ~)! = n Traditional Approach: Trans­ form the generalized EVP or quadratic EVP into a stand­ ard form. K = 1::. .g.\!ii M=I::I::T i=hTjJ hence ~ or :t = ..1 etc ... -~ .T T M= W02 W 12·5 .Solution methods for calculations of frequencies and mode shapes Rewrite problem as: and solve for largest K: .K 2£. . K [.

The solution procedures in use operate on the basic equations that have to be satisfied. or i=r. 1n . i) i=l. 1) VECTOR ITERATION TECHNIQUES Equation: e.g. Consider the Gen. .3 eigenpairs ( Ai' are required .SolotiOl . EVP ! ! = with AM! 1. which one is not guaranteed (convergence may also be slow) 12·6 . Inverse It.. ~P_=A~~ ! ~+l = M ~ ~+l • Forward Iteration • Rayleigh Quotient Iteration can be employed to cal­ culate one eigenvalue and vector.elhods lor calcolations oIlreqoeacies ud .s 1. deflate then to calculate additional eigenpair Convergence to " an eigenpair".ode sllapes Direct solution is more effective.p .

unstable process Pi+l = Pi = det L D LT = II d ..I .A!:1) = 0 .Solution methods for calculations of frequencies and mode shapes 2) POLYNOMIAL ITERATION METHODS ! ~ = A ~ ~ ~ (K . eTechnique not suitable for larger problems . Newton Iteration p(A) aO + alA + a2A + . + anA bO (A-Al) (A-A2) '" (A-An) 2 n Implicit polynomial iteration: p (Pi) = det (IS. . .... -.A M) ¢ 0 Hence p(A) det (~.. provided we do not encounter large multipliers e we directly solve for Al. .. II e accurate. e use SECANT ITERATION: e deflate polynom ial after convergence to A1 12-7 .much work to obtain ai's .Pi !y!) Explicit polynomial iteration: eExpand the polynomial and iterate for zeros.

1 1- p (A) / (A-A. 12·8 . Care need be taken in L D LT factor­ ization. etc.Solution methods for calculations of frequencies and mode shapes ]J. but can be slow when we calculate multiple roots.) I I III Convergence guaranteed to A1 ' then A2 .

. . .. ... Number of negative elements in D is equal to the number of eigenvalues smaller than J. of freql8cies iIIld ole shapes 3) STURM SEQUENCE METHODS 3rd associated constraint problem 2nd associated constraint problem 1st associated constraint problem ! <p = A!11 . . -. . ..1 S . . . · · 1 t :::} 2 3 4 . .6Jds for calculali... 12·9 .liOi . .SaI. · 9· ~ · ..

lS2 Tf ..--..].. . interval ..ls 1 ]. ~ J./ " ..<P ­ - = I Construct <P iteratively: _ H n <P = [~.. ]. • Need to take care in L D L aetonzatlon --- • Convergence can be very slow 4) TRANSFORMATION METHODS ~!=A~!--T-- j <PTK<P=A <P M ..Solution lDethods lor calculations 01 frequencies ud lDode shapes 3) STURM SEQUENCE METHODS Calculate ~ . = [Al .lS... A 'n] 12·10 . ~ = h . --.. .. Qh T Count number of negative elements in Q and use a strategy to isolate eigenvalue(s) .

.. generalized Jacobi method • Here we calculate all eigenpairs simultaneously • Expensive and ineffective (impossible) or large problems.... ~-l e.5oI1tiOi . ~ f 1 !i f 1 T T ~ ..g..elhods for calculations 01 frequencies ad .ode shapes T T T ~ . For large eigenproblems it is best to use combinations of the above basic techniques: • Determinant search to get near a root • Vector iteration to obtain eigenvector and eigenvalue • Transformation method for orthogonalization of itera­ tion vectors.. ~ ~l f f1 ~ . ~-~ ~ T . • Sturm sequence method to ensure that required eigenvalue(s) has (or have) been calculated 12·11 .

8. _1 P( A) 2) Use Sturm sequence property to check whether 11 .. ..) = det (~. provided is deflated of A1' .. - n P(l1.+1 = ]._1) 11. 11.Solution methods for calCl1atiou of frequencies and mode sJlapes THE DETERMINANT SEARCH METHOD p(A) / A ~. .11. .) . ..0 n = 2. 12·12 . 1) Iterate on polynomial to obtain shifts close to A1 P(l1. . .P(11.A.. 4.. . ~) = det .-11..1. when convergence is slow Same procedure can be employed to obtain shift near A. +1 is larger than an unknown eigenvalue.-T = n d 11 L DL . . ._1 n is normally = 1. .

. e.. •~+l p = .g.2. .....T (~+l) 4) Iteration vector must be deflated of the previously calculated eigenvectors using. .Solution lOethods for calculations of freqoBcies ud lOode shapes . If convergence is slow use Rayleigh quotient iteration 12·13 . Gram­ Schmidt orthogonalization.....T ~+l !i ~k ~+l !i ~+l ~ . use inverse iteration to calculate the eigenvector and eigenvalue lJi+ 1 k =1... 3) Once lJ i +1 is larger than an unknown eigenvalue.~+l T ~ (~+l !i ~+l) =.

1 iteration -. no prior transformation of eigenproblem Disadvantage: Many triangular factorizations • Effective only for small banded systems We need an algorithm with less factorizations and more vector iterations when the bandwidth of the system is large.. -T ~+1 = 4+1 -T ~+1 = 4+1 -~+1 4+1 ~+1 ~+1 = ~+1 ~+1 ~+1 4+1 = ~+1 ~+1 12·14 . inverse {K = '. .. SUBSPACE ITERATION METHOD Iterate with q vectors wher:' the lowest p eigenvalues and eigen­ vectors are required.2.4+1 4 K ~1 k=1.Solution methods lor calculations oIlrequencies ud mode shapes Advantage: Calculates only eigenpairs actually required.

Convergence rate: convergence reached when < tal 12·15 . Use Sturm sequence check eigenvalue p eigenvalues flS !5.Solution methods for calculations of frequencies and mode shapes "Under conditions" we have CONDITION: starting subspace spanned by X.flS t1 = T ~ Q~ no. . must not be orth­ ogonal to least dominant subspace required. of -ve elements in D must be equal to p.

+l) II 2 important in!!!. . .. Sturm sequence checks 2.1980.Solution methods for calculations of frequencies and mode shapes Starting Vectors Two choices 1) ~l ~ x. pp. J. 23. Vol. 313 . Computer Methods in Applied Mechanics and Engineering. = ~ e.331.+l) ~!~Q.+1)_ A~Q.. j=2. Checks on eigenpairs 1. E:. solutions.q-l 4 x = random vector 2) Lanczos method Here we need to use q much larger than p. Reference: An Accelerated Subspace Iteration Method.+1)[12 1 [I - K -1 ¢~9. 12·16 ..= 11~!~Q.

mit.edu/terms. For information about citing these materials or our Terms of Use. visit: http://ocw.edu Resource: Finite Element Procedures for Solids and Structures Klaus-Jürgen Bathe The following may not correspond to a particular course on MIT OpenCourseWare. .mit. but has been provided by the author as an individual learning resource.MIT OpenCourseWare http://ocw.

scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->