09 - Num Resp (SDOF) - UofP

Leccture Notes: STRUCTURA
S AL DYNAMICS / FALL 2011

2 / Pagge: 1
Lectu
urer: Prof. AP
POSTOLOS S. PAPAGEOR
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
PART (09): NUMERICAL
N E
EVALUATIONN OF DYNAM
MIC RESPONS SE
N
NUMER
RICAL EV
VALUAT
TION OF
F DYNA
AMIC RE
ESPONSE
E
TIME-STEPPING
G (or DIFF
FERENCE
E) METHO
ODS:
Equatio
on of Motion
n of an inelastic system
m:
,
Initiaal Conditiions I. C. ′s : 0 & 0
Importaant requirem
ments for a numerical procedure:
(1) Convergence:
As the time
t step deecreases, th
he numeric cal solutio
on should
approa ach the ex xact solutio on.
(2) Stability:
The nummerical soluution shou uld be stable in the presence
p o
of
numer rical round-off erro ors.
(3) Accuracy::
The nummerical proocedure sho ould proviide resultss that are close
c
enough h to the exxact solutiion.
Types of time-sttepping pr
rocedures::
(1) Meth
hods based on interpo
olation off excitation
n function
n.
(2) Metthods based

d on finite difference
d e expressio
ons of velo
ocity & acc
celeration
n.
(3) Metthods based

d on assum
med variatiion of acce
eleration.
Lecture Notes: STRUCTURAL DYNAMICS / FALL 2011 / Page: 2
Lecturer: Prof. APOSTOLOS S. PAPAGEORGIOU
SEOUL NATIONAL UNIVERSITY
PART (09): NUMERICAL EVALUATION OF DYNAMIC RESPONSE
METHODS BASED ON INTERPOLATION OF EXCITATION (LINEAR SYSTEMS)
Interpolation of the excitation over each time interval
For sufficiently short time intervals, linear interpolation:
Response for ∆ :
(1) Free vibration due to I.C.’s & at
+
(2) Response to step force with zero initial conditions
+
∆
(3) Response to ramp force with zero initial conditions
∆
Each of the above responses is available in closed form for linear systems. Therefore:
Expressions for , , , , , , , are given in TABLE 5.2.1 of CHOPRA (1995).
The above described procedure is used to compute elastic response spectra.

2 / Pagge: 3
Lectu
urer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
METH
HODS BASE
ED ON DIIFFERENC
CE EXPRE
ESSIONS
Centra
al Differen
nce Metho
od
Approxximate & by differen

nce equatio
ons:
2
;
2 ∆ ∆
One waay of obtainiing these formulae is by

b Taylor expansion
e n:
∆ ∆
∆ ∆ (1)
2! 3!
∆ ∆
∆ ∆ (2)
2! 3!
Subtraccting (2) fro

om (1), we obtain:
∆
∆ ∆ 2 ∆
3!
∆ ∆ ∆
⟹
2 ∆ 6
Add (2)) to (1) afterr expanding

g up to 4th po
ower:
∆
∆ ∆ 2 ∆
4!
∆ 2 ∆ ∆
⟹
∆ 1
12
2 / Pagge: 4
Lectu
urer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
Equatio
on of Motion
n at step :
Substitu pproximate expression for

uting the ap & :
2
∆ 2 ∆
Transfeerring the un
nknown quantities to the de, we obtain:
t right sid
̂
̂ ⟺
∆ 2 ∆
where: 2
̂
∆ 2 ∆ ∆
Exp
plicit Meth
hod
To initialize the process:
2
;
2 ∆ ∆
∆
After eliminating , we solvee for : ∆
& are given initial cond

ditions
Equatiion of Mottion: ⟹
Stabiliity require
ement:
∆
∆
Typica
ally, . , ( = natural per
riod of the
e SDOF sys
stem) to define
d the
respon
nse adequately.
2 / Pagge: 5
Lectu
urer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
NEWM
MARK’s M
METHOD
D
‘Newmmark’s Metthod’ comp

prises a fam
mily of time--stepping methods
m baseed on the fo
ollowing
equatio
ons:
1 ∆ ∆
∆ ∆ ∆
The parrameters & :
 ration overr ∆
define the variation of acceler
 determine the
t stabilitty & accur
racy charactteristics
Method
d is impliciit ⟹ iteration
For lineear systems, the metho

od may be modified
m to become
b exp
plicit.
Typica
al selection
n of param
meters:
for sattisfactory performan

p nce, inclu
uding accu
uracy.
Specia
al Cases:
(1) : Constant acceleratiion metho

od
(2) : Average acceleratio

a on method
d
(3) : Linear acceleration

n method
Newmaark’s method
d is stable if:
∆ 1 1
√2 2
∆
For: ⟹ ∞
∆
For: ⟹ .
Constant-Acceleration Method: ( )
⟹ ∆
2
∆
⟹ ∆
2
The above two equations provide two of the three equations for time integration, i.e. to
obtain , , from , , .
The third equation is the Equation of Motion:
Then:
1 ∆
∆ ∆
2
To begin the time integration, we need to know the values of , & , i.e. the values of
displacement, velocity and acceleration at .
Two of them must be specified; the third is obtained by using the Equation of Motion at
.
Average Acceleration Method: ( )
1
2
⟹ ∆
2
2 2
∆
⟹ ∆
4
Linear Acceleration: ( )
2 ∆
⟹ ∆
2
2 6 ∆
1 1
⟹ ∆ ∆
6 3
2 / Pagge: 8
Lectu
urer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
Non-iiterative Formula
F tion of Newmark
N ’s Meth
hod (Line
ear Systeems):
Let:
∆ ≝
∆ ≝
∆ ≝ ∆
Newmaark’s time-s
stepping equations
e :
∆ ∆ ∆ ∆ (1a)
∆
∆ ∆ ∆ ∆ (1b)
2
(1b) ⟹ ∆ ∆ (2)
∆ ∆
(2)) + (1a) ⟹ ∆ ∆ ∆ 1 (3)
∆
Increm
mental Equ
uation of Motion:
∆ ∆ ∆ ∆ (4)
2 3 4 ⟹ ∆ ∆ ̂
1
wherre:
∆ ∆
1 1
an
nd ∆ ̂ ∆ ∆ 1
∆ 2 2
∆
∆
Therefo
ore: ⟹∆ ⟹
Equation 3
Im
mplicit Method
d( )
2 / Pagge: 9
Lectu
urer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
ERROR
RS INVOL
LVED IN NUMERICA
N AL INTEG
GRATION
Types o
of errors:
(1) Round-off
ff errors du
ue to precission of the flloating poin
nt arithmetiic of compu
uters.
(2) Truncatio on errors involved

i in representin ng or by a finite number of
terms in the Taylor serries expansiion. This errror is repreesented anallytically by the
term ∆ .
A very iimportant aspect

a in thee error anallysis of num
merical meth
hods is the growth
g or
accummulation off errors as computatio on progressses (error propagati
p ion).
This con
nsideration
n is related to
t the stab
bility of thee numerical scheme.
The perrformance of
o a numericcal scheme is evaluated
d by examin
ning two im
mportant
charactteristics:
∆
(1) Amplitud
de decay (A
AD) as a fun
nction of
∆
(2) Period Elo
ongation (PE)
( as a fu
unction of
The Cenntral Diffe

ference, Liinear Acce eleration & Average
e Accelera
ation Meth
hods
introdu
uce no artifficial dam
mping, i.e. .
Lecture Notes: STRUCTURAL
S L DYNAMICS
S / FALL 2011
2 / Pagee: 10
Lectuurer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
ANALY
YSIS OF NONLINEA
N AR RESPO
ONSE
Equatio
on of Motion
n: ,
___
________
_________
____
or ∆ ∆ ∆ ∆
The inc
crementall resisting force:
∆ ∙∆
where: = secant stiffness
s (not know
wn)
or apprroximately:
∆ ≅ ∙∆
where: = tangent stiffness
s
Applicaation of New
wmark’s Method,
M wh
hich is the most
m popula
ar method because
b of its
accuraccy:
∙∆ ∆ ̂
1
∆ ∆
wherre:
1 1
∆ ̂ ∆ ∆ 1
∆ 2 2
i.e., sam
me formullae as in th
he linear case
c exceptt that has been replacced by .
Significant error arises for two reasons:
(1) The tangent stiffness is used instead of the secant stiffness;
(2) Use of a constant time step delays detection of the transition in the force
deformation relation.
Error (2) is eliminated by using an iterative process in which integration is resumed

from with a smaller step whose size is progressively adjusted so that at the end of
such an adjusted time step, the velocity is close to zero.
Lecture Notes: STRUCTURAL
S L DYNAMICS
S / FALL 2011
2 / Pagee: 12
Lectuurer: Prof. AP
RGIOU
SEOUL NA ATIONAL UN
NIVERSITY
P
N E
MIC RESPONS SE
Error (1)
( is elimin
nated by an
n iteration process ( ) that is kno
own as
Newtoon-Raphso on Method d.
Initializze data:
∆ ∆ ̂
Calculaation for ea
ach iteratio
on: 1,2,3, ⋯
∆ ∆ ⟹∆
∆
1
∆ ∆
∆ ∆
∆ ∆ ∆
LINEAR MULTISTEP (LMS) METHODS
LMS Methods for 1st-order Equations:

Consider a system of 1st-order Ordinary Differential Equations (ODE’s):
, (1)
A -step linear multistep method for equation (1) is defined by the following
expression:
∆ , (2)
The & are parameters that define the method.

Note that the word ‘linear’, in ‘linear multistep method’, has nothing to do with the
linearity of equation (1). Indeed, equation (2) is perfectly well defined for a nonlinear
function , .
___________________
Excellent references for LMS methods of this type are:

GEAR, C.W. (1971), Numerical Initial Value Problems in Ordinary Differential Equations, Prentice-
Hall, Englewood Cliffs, NJ.
ISERLES, A. (1996), A First Course in the Numerical Analysis of Differential Equations, Cambridge
University Press.
___________________
EXAMPLE:
The equation of motion of the linear, viscously damped, SDOF system can be cast in 1st-order
form:
1
⟹ 2
Let:
Then: ⟹ 2
Therefore:
0
0 1 1 ,
2 ,
i.e. ,
Multistep Methods:
Runge-Kutta (one step method)
Adams-Bashford
Adams-Multon
Predictor-Corrector Methods

09 - Num Resp (SDOF) - UofP

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

09 - Num Resp (SDOF) - UofP

Uploaded by

Copyright:

Available Formats

Leccture Notes: STRUCTURA

S AL DYNAMICS / FALL 2011

(2) Metthods based

(3) Metthods based

METHODS BASED ON INTERPOLATION OF EXCITATION (LINEAR SYSTEMS)

Interpolation of the excitation over each time interval

For sufficiently short time intervals, linear interpolation:

(1) Free vibration due to I.C.’s & at

Expressions for , , , , , , , are given in TABLE 5.2.1 of CHOPRA (1995).

The above described procedure is used to compute elastic response spectra.

Approxximate & by differen

One waay of obtainiing these formulae is by

Subtraccting (2) fro

Add (2)) to (1) afterr expanding

Substitu pproximate expression for

To initialize the process:

& are given initial cond

‘Newmmark’s Metthod’ comp

The parrameters & :

For lineear systems, the metho

for sattisfactory performan

(1) : Constant acceleratiion metho

(2) : Average acceleratio

(3) : Linear acceleration

The third equation is the Equation of Motion:

Average Acceleration Method: ( )

(2) Truncatio on errors involved

A very iimportant aspect

The Cenntral Diffe

Significant error arises for two reasons:

(1) The tangent stiffness is used instead of the secant stiffness;

Error (2) is eliminated by using an iterative process in which integration is resumed

LINEAR MULTISTEP (LMS) METHODS

LMS Methods for 1st-order Equations:

The & are parameters that define the method.

Excellent references for LMS methods of this type are:

Runge-Kutta (one step method)

You might also like